linux_dsm_epyc7002/drivers/net/bonding/bond_main.c

4907 lines
136 KiB
C
Raw Normal View History

/*
* originally based on the dummy device.
*
* Copyright 1999, Thomas Davis, tadavis@lbl.gov.
* Licensed under the GPL. Based on dummy.c, and eql.c devices.
*
* bonding.c: an Ethernet Bonding driver
*
* This is useful to talk to a Cisco EtherChannel compatible equipment:
* Cisco 5500
* Sun Trunking (Solaris)
* Alteon AceDirector Trunks
* Linux Bonding
* and probably many L2 switches ...
*
* How it works:
* ifconfig bond0 ipaddress netmask up
* will setup a network device, with an ip address. No mac address
* will be assigned at this time. The hw mac address will come from
* the first slave bonded to the channel. All slaves will then use
* this hw mac address.
*
* ifconfig bond0 down
* will release all slaves, marking them as down.
*
* ifenslave bond0 eth0
* will attach eth0 to bond0 as a slave. eth0 hw mac address will either
* a: be used as initial mac address
* b: if a hw mac address already is there, eth0's hw mac address
* will then be set from bond0.
*
*/
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/types.h>
#include <linux/fcntl.h>
#include <linux/interrupt.h>
#include <linux/ptrace.h>
#include <linux/ioport.h>
#include <linux/in.h>
#include <net/ip.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/udp.h>
#include <linux/slab.h>
#include <linux/string.h>
#include <linux/init.h>
#include <linux/timer.h>
#include <linux/socket.h>
#include <linux/ctype.h>
#include <linux/inet.h>
#include <linux/bitops.h>
#include <linux/io.h>
#include <asm/dma.h>
#include <linux/uaccess.h>
#include <linux/errno.h>
#include <linux/netdevice.h>
#include <linux/inetdevice.h>
#include <linux/igmp.h>
#include <linux/etherdevice.h>
#include <linux/skbuff.h>
#include <net/sock.h>
#include <linux/rtnetlink.h>
#include <linux/smp.h>
#include <linux/if_ether.h>
#include <net/arp.h>
#include <linux/mii.h>
#include <linux/ethtool.h>
#include <linux/if_vlan.h>
#include <linux/if_bonding.h>
#include <linux/jiffies.h>
#include <linux/preempt.h>
#include <net/route.h>
#include <net/net_namespace.h>
#include <net/netns/generic.h>
bonding: Fix corrupted queue_mapping In the transmit path of the bonding driver, skb->cb is used to stash the skb->queue_mapping so that the bonding device can set its own queue mapping. This value becomes corrupted since the skb->cb is also used in __dev_xmit_skb. When transmitting through bonding driver, bond_select_queue is called from dev_queue_xmit. In bond_select_queue the original skb->queue_mapping is copied into skb->cb (via bond_queue_mapping) and skb->queue_mapping is overwritten with the bond driver queue. Subsequently in dev_queue_xmit, __dev_xmit_skb is called which writes the packet length into skb->cb, thereby overwriting the stashed queue mappping. In bond_dev_queue_xmit (called from hard_start_xmit), the queue mapping for the skb is set to the stashed value which is now the skb length and hence is an invalid queue for the slave device. If we want to save skb->queue_mapping into skb->cb[], best place is to add a field in struct qdisc_skb_cb, to make sure it wont conflict with other layers (eg : Qdiscc, Infiniband...) This patchs also makes sure (struct qdisc_skb_cb)->data is aligned on 8 bytes : netem qdisc for example assumes it can store an u64 in it, without misalignment penalty. Note : we only have 20 bytes left in (struct qdisc_skb_cb)->data[]. The largest user is CHOKe and it fills it. Based on a previous patch from Tom Herbert. Signed-off-by: Eric Dumazet <edumazet@google.com> Reported-by: Tom Herbert <therbert@google.com> Cc: John Fastabend <john.r.fastabend@intel.com> Cc: Roland Dreier <roland@kernel.org> Acked-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-06-12 13:03:51 +07:00
#include <net/pkt_sched.h>
bonding: initial RCU conversion This patch does the initial bonding conversion to RCU. After it the following modes are protected by RCU alone: roundrobin, active-backup, broadcast and xor. Modes ALB/TLB and 3ad still acquire bond->lock for reading, and will be dealt with later. curr_active_slave needs to be dereferenced via rcu in the converted modes because the only thing protecting the slave after this patch is rcu_read_lock, so we need the proper barrier for weakly ordered archs and to make sure we don't have stale pointer. It's not tagged with __rcu yet because there's still work to be done to remove the curr_slave_lock, so sparse will complain when rcu_assign_pointer and rcu_dereference are used, but the alternative to use rcu_dereference_protected would've created much bigger code churn which is more difficult to test and review. That will be converted in time. 1. Active-backup mode 1.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.55% in bonding, system spent 0.29% CPU in bonding - new bonding: iperf spent 0.29% in bonding, system spent 0.15% CPU in bonding 1.2. Bandwidth measurements - old bonding: 16.1 gbps consistently - new bonding: 17.5 gbps consistently 2. Round-robin mode 2.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.51% in bonding, system spent 0.24% CPU in bonding - new bonding: iperf spent 0.16% in bonding, system spent 0.11% CPU in bonding 2.2 Bandwidth measurements - old bonding: 8 gbps (variable due to packet reorderings) - new bonding: 10 gbps (variable due to packet reorderings) Of course the latency has improved in all converted modes, and moreover while doing enslave/release (since it doesn't affect tx anymore). Also I've stress tested all modes doing enslave/release in a loop while transmitting traffic. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 21:54:51 +07:00
#include <linux/rculist.h>
#include <net/flow_dissector.h>
#include <net/switchdev.h>
#include <net/bonding.h>
#include <net/bond_3ad.h>
#include <net/bond_alb.h>
#include "bonding_priv.h"
/*---------------------------- Module parameters ----------------------------*/
/* monitor all links that often (in milliseconds). <=0 disables monitoring */
static int max_bonds = BOND_DEFAULT_MAX_BONDS;
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
static int tx_queues = BOND_DEFAULT_TX_QUEUES;
static int num_peer_notif = 1;
static int miimon;
static int updelay;
static int downdelay;
static int use_carrier = 1;
static char *mode;
static char *primary;
static char *primary_reselect;
static char *lacp_rate;
static int min_links;
static char *ad_select;
static char *xmit_hash_policy;
static int arp_interval;
static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
static char *arp_validate;
bonding: add an option to fail when any of arp_ip_target is inaccessible Currently, we fail only when all of the ips in arp_ip_target are gone. However, in some situations we might need to fail if even one host from arp_ip_target becomes unavailable. All situations, obviously, rely on the idea that we need *completely* functional network, with all interfaces/addresses working correctly. One real world example might be: vlans on top on bond (hybrid port). If bond and vlans have ips assigned and we have their peers monitored via arp_ip_target - in case of switch misconfiguration (trunk/access port), slave driver malfunction or tagged/untagged traffic dropped on the way - we will be able to switch to another slave. Though any other configuration needs that if we need to have access to all arp_ip_targets. This patch adds this possibility by adding a new parameter - arp_all_targets (both as a module parameter and as a sysfs knob). It can be set to: 0 or any (the default) - which works exactly as it's working now - the slave is up if any of the arp_ip_targets are up. 1 or all - the slave is up if all of the arp_ip_targets are up. This parameter can be changed on the fly (via sysfs), and requires the mode to be active-backup and arp_validate to be enabled (it obeys the arp_validate config on which slaves to validate). Internally it's done through: 1) Add target_last_arp_rx[BOND_MAX_ARP_TARGETS] array to slave struct. It's an array of jiffies, meaning that slave->target_last_arp_rx[i] is the last time we've received arp from bond->params.arp_targets[i] on this slave. 2) If we successfully validate an arp from bond->params.arp_targets[i] in bond_validate_arp() - update the slave->target_last_arp_rx[i] with the current jiffies value. 3) When getting slave's last_rx via slave_last_rx(), we return the oldest time when we've received an arp from any address in bond->params.arp_targets[]. If the value of arp_all_targets == 0 - we still work the same way as before. Also, update the documentation to reflect the new parameter. v3->v4: Kill the forgotten rtnl_unlock(), rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Also, print a warning if the last arp_ip_target is removed while the arp_interval is on, but not the arp_validate. v2->v3: Use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). On bond_enslave(), use the same initialization value for target_last_arp_rx[] as is used for the default last_arp_rx, to avoid useless interface flaps. Also, instead of failing to remove the last arp_ip_target just print a warning - otherwise it might break existing scripts. v1->v2: Correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes. Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-24 16:49:34 +07:00
static char *arp_all_targets;
static char *fail_over_mac;
static int all_slaves_active;
static struct bond_params bonding_defaults;
static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
static int packets_per_slave = 1;
static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
module_param(max_bonds, int, 0);
MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
module_param(tx_queues, int, 0);
MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
module_param_named(num_grat_arp, num_peer_notif, int, 0644);
MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
"failover event (alias of num_unsol_na)");
module_param_named(num_unsol_na, num_peer_notif, int, 0644);
MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
"failover event (alias of num_grat_arp)");
module_param(miimon, int, 0);
MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
module_param(updelay, int, 0);
MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
module_param(downdelay, int, 0);
MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
"in milliseconds");
module_param(use_carrier, int, 0);
MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
"0 for off, 1 for on (default)");
module_param(mode, charp, 0);
MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
"1 for active-backup, 2 for balance-xor, "
"3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
"6 for balance-alb");
module_param(primary, charp, 0);
MODULE_PARM_DESC(primary, "Primary network device to use");
module_param(primary_reselect, charp, 0);
MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
"once it comes up; "
"0 for always (default), "
"1 for only if speed of primary is "
"better, "
"2 for only on active slave "
"failure");
module_param(lacp_rate, charp, 0);
MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
"0 for slow, 1 for fast");
module_param(ad_select, charp, 0);
MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
"0 for stable (default), 1 for bandwidth, "
"2 for count");
module_param(min_links, int, 0);
MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
module_param(xmit_hash_policy, charp, 0);
MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
"0 for layer 2 (default), 1 for layer 3+4, "
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
"2 for layer 2+3, 3 for encap layer 2+3, "
"4 for encap layer 3+4");
module_param(arp_interval, int, 0);
MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
module_param_array(arp_ip_target, charp, NULL, 0);
MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
module_param(arp_validate, charp, 0);
MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
"0 for none (default), 1 for active, "
"2 for backup, 3 for all");
bonding: add an option to fail when any of arp_ip_target is inaccessible Currently, we fail only when all of the ips in arp_ip_target are gone. However, in some situations we might need to fail if even one host from arp_ip_target becomes unavailable. All situations, obviously, rely on the idea that we need *completely* functional network, with all interfaces/addresses working correctly. One real world example might be: vlans on top on bond (hybrid port). If bond and vlans have ips assigned and we have their peers monitored via arp_ip_target - in case of switch misconfiguration (trunk/access port), slave driver malfunction or tagged/untagged traffic dropped on the way - we will be able to switch to another slave. Though any other configuration needs that if we need to have access to all arp_ip_targets. This patch adds this possibility by adding a new parameter - arp_all_targets (both as a module parameter and as a sysfs knob). It can be set to: 0 or any (the default) - which works exactly as it's working now - the slave is up if any of the arp_ip_targets are up. 1 or all - the slave is up if all of the arp_ip_targets are up. This parameter can be changed on the fly (via sysfs), and requires the mode to be active-backup and arp_validate to be enabled (it obeys the arp_validate config on which slaves to validate). Internally it's done through: 1) Add target_last_arp_rx[BOND_MAX_ARP_TARGETS] array to slave struct. It's an array of jiffies, meaning that slave->target_last_arp_rx[i] is the last time we've received arp from bond->params.arp_targets[i] on this slave. 2) If we successfully validate an arp from bond->params.arp_targets[i] in bond_validate_arp() - update the slave->target_last_arp_rx[i] with the current jiffies value. 3) When getting slave's last_rx via slave_last_rx(), we return the oldest time when we've received an arp from any address in bond->params.arp_targets[]. If the value of arp_all_targets == 0 - we still work the same way as before. Also, update the documentation to reflect the new parameter. v3->v4: Kill the forgotten rtnl_unlock(), rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Also, print a warning if the last arp_ip_target is removed while the arp_interval is on, but not the arp_validate. v2->v3: Use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). On bond_enslave(), use the same initialization value for target_last_arp_rx[] as is used for the default last_arp_rx, to avoid useless interface flaps. Also, instead of failing to remove the last arp_ip_target just print a warning - otherwise it might break existing scripts. v1->v2: Correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes. Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-24 16:49:34 +07:00
module_param(arp_all_targets, charp, 0);
MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
module_param(fail_over_mac, charp, 0);
MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
"the same MAC; 0 for none (default), "
"1 for active, 2 for follow");
module_param(all_slaves_active, int, 0);
MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
"by setting active flag for all slaves; "
"0 for never (default), 1 for always.");
module_param(resend_igmp, int, 0);
MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
"link failure");
module_param(packets_per_slave, int, 0);
MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
"mode; 0 for a random slave, 1 packet per "
"slave (default), >1 packets per slave.");
module_param(lp_interval, uint, 0);
MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
"the bonding driver sends learning packets to "
"each slaves peer switch. The default is 1.");
/*----------------------------- Global variables ----------------------------*/
#ifdef CONFIG_NET_POLL_CONTROLLER
net: Convert netpoll blocking api in bonding driver to be a counter A while back I made some changes to enable netpoll in the bonding driver. Among them was a per-cpu flag that indicated we were in a path that held locks which could cause the netpoll path to block in during tx, and as such the tx path should queue the frame for later use. This appears to have given rise to a regression. If one of those paths on which we hold the per-cpu flag yields the cpu, its possible for us to come back on a different cpu, leading to us clearing a different flag than we set. This results in odd netpoll drops, and BUG backtraces appearing in the log, as we check to make sure that we only clear set bits, and only set clear bits. I had though briefly about changing the offending paths so that they wouldn't sleep, but looking at my origional work more closely, it doesn't appear that a per-cpu flag is warranted. We alrady gate the checking of this flag on IFF_IN_NETPOLL, so we don't hit this in the normal tx case anyway. And practically speaking, the normal use case for netpoll is to only have one client anyway, so we're not going to erroneously queue netpoll frames when its actually safe to do so. As such, lets just convert that per-cpu flag to an atomic counter. It fixes the rescheduling bugs, is equivalent from a performance perspective and actually eliminates some code in the process. Tested by the reporter and myself, successfully Reported-by: Liang Zheng <lzheng@redhat.com> CC: Jay Vosburgh <fubar@us.ibm.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: David S. Miller <davem@davemloft.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-12-06 16:05:50 +07:00
atomic_t netpoll_block_tx = ATOMIC_INIT(0);
#endif
netns: make struct pernet_operations::id unsigned int Make struct pernet_operations::id unsigned. There are 2 reasons to do so: 1) This field is really an index into an zero based array and thus is unsigned entity. Using negative value is out-of-bound access by definition. 2) On x86_64 unsigned 32-bit data which are mixed with pointers via array indexing or offsets added or subtracted to pointers are preffered to signed 32-bit data. "int" being used as an array index needs to be sign-extended to 64-bit before being used. void f(long *p, int i) { g(p[i]); } roughly translates to movsx rsi, esi mov rdi, [rsi+...] call g MOVSX is 3 byte instruction which isn't necessary if the variable is unsigned because x86_64 is zero extending by default. Now, there is net_generic() function which, you guessed it right, uses "int" as an array index: static inline void *net_generic(const struct net *net, int id) { ... ptr = ng->ptr[id - 1]; ... } And this function is used a lot, so those sign extensions add up. Patch snipes ~1730 bytes on allyesconfig kernel (without all junk messing with code generation): add/remove: 0/0 grow/shrink: 70/598 up/down: 396/-2126 (-1730) Unfortunately some functions actually grow bigger. This is a semmingly random artefact of code generation with register allocator being used differently. gcc decides that some variable needs to live in new r8+ registers and every access now requires REX prefix. Or it is shifted into r12, so [r12+0] addressing mode has to be used which is longer than [r8] However, overall balance is in negative direction: add/remove: 0/0 grow/shrink: 70/598 up/down: 396/-2126 (-1730) function old new delta nfsd4_lock 3886 3959 +73 tipc_link_build_proto_msg 1096 1140 +44 mac80211_hwsim_new_radio 2776 2808 +32 tipc_mon_rcv 1032 1058 +26 svcauth_gss_legacy_init 1413 1429 +16 tipc_bcbase_select_primary 379 392 +13 nfsd4_exchange_id 1247 1260 +13 nfsd4_setclientid_confirm 782 793 +11 ... put_client_renew_locked 494 480 -14 ip_set_sockfn_get 730 716 -14 geneve_sock_add 829 813 -16 nfsd4_sequence_done 721 703 -18 nlmclnt_lookup_host 708 686 -22 nfsd4_lockt 1085 1063 -22 nfs_get_client 1077 1050 -27 tcf_bpf_init 1106 1076 -30 nfsd4_encode_fattr 5997 5930 -67 Total: Before=154856051, After=154854321, chg -0.00% Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-11-17 08:58:21 +07:00
unsigned int bond_net_id __read_mostly;
/*-------------------------- Forward declarations ---------------------------*/
static int bond_init(struct net_device *bond_dev);
static void bond_uninit(struct net_device *bond_dev);
static void bond_get_stats(struct net_device *bond_dev,
struct rtnl_link_stats64 *stats);
static void bond_slave_arr_handler(struct work_struct *work);
bonding: Fix ARP monitor validation The current logic in bond_arp_rcv will accept an incoming ARP for validation if (a) the receiving slave is either "active" (which includes the currently active slave, or the current ARP slave) or, (b) there is a currently active slave, and it has received an ARP since it became active. For case (b), the receiving slave isn't the currently active slave, and is receiving the original broadcast ARP request, not an ARP reply from the target. This logic can fail if there is no currently active slave. In this situation, the ARP probe logic cycles through all slaves, assigning each in turn as the "current_arp_slave" for one arp_interval, then setting that one as "active," and sending an ARP probe from that slave. The current logic expects the ARP reply to arrive on the sending current_arp_slave, however, due to switch FDB updating delays, the reply may be directed to another slave. This can arise if the bonding slaves and switch are working, but the ARP target is not responding. When the ARP target recovers, a condition may result wherein the ARP target host replies faster than the switch can update its forwarding table, causing each ARP reply to be sent to the previous current_arp_slave. This will never pass the logic in bond_arp_rcv, as neither of the above conditions (a) or (b) are met. Some experimentation on a LAN shows ARP reply round trips in the 200 usec range, but my available switches never update their FDB in less than 4000 usec. This patch changes the logic in bond_arp_rcv to additionally accept an ARP reply for validation on any slave if there is a current ARP slave and it sent an ARP probe during the previous arp_interval. Fixes: aeea64ac717a ("bonding: don't trust arp requests unless active slave really works") Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <gospo@cumulusnetworks.com> Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-02-03 04:35:56 +07:00
static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
int mod);
bonding: avoid possible dead-lock Syzkaller reported this on a slightly older kernel but it's still applicable to the current kernel - ====================================================== WARNING: possible circular locking dependency detected 4.18.0-next-20180823+ #46 Not tainted ------------------------------------------------------ syz-executor4/26841 is trying to acquire lock: 00000000dd41ef48 ((wq_completion)bond_dev->name){+.+.}, at: flush_workqueue+0x2db/0x1e10 kernel/workqueue.c:2652 but task is already holding lock: 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnl_lock net/core/rtnetlink.c:77 [inline] 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnetlink_rcv_msg+0x412/0xc30 net/core/rtnetlink.c:4708 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #2 (rtnl_mutex){+.+.}: __mutex_lock_common kernel/locking/mutex.c:925 [inline] __mutex_lock+0x171/0x1700 kernel/locking/mutex.c:1073 mutex_lock_nested+0x16/0x20 kernel/locking/mutex.c:1088 rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77 bond_netdev_notify drivers/net/bonding/bond_main.c:1310 [inline] bond_netdev_notify_work+0x44/0xd0 drivers/net/bonding/bond_main.c:1320 process_one_work+0xc73/0x1aa0 kernel/workqueue.c:2153 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #1 ((work_completion)(&(&nnw->work)->work)){+.+.}: process_one_work+0xc0b/0x1aa0 kernel/workqueue.c:2129 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #0 ((wq_completion)bond_dev->name){+.+.}: lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe other info that might help us debug this: Chain exists of: (wq_completion)bond_dev->name --> (work_completion)(&(&nnw->work)->work) --> rtnl_mutex Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(rtnl_mutex); lock((work_completion)(&(&nnw->work)->work)); lock(rtnl_mutex); lock((wq_completion)bond_dev->name); *** DEADLOCK *** 1 lock held by syz-executor4/26841: stack backtrace: CPU: 1 PID: 26841 Comm: syz-executor4 Not tainted 4.18.0-next-20180823+ #46 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113 print_circular_bug.isra.34.cold.55+0x1bd/0x27d kernel/locking/lockdep.c:1222 check_prev_add kernel/locking/lockdep.c:1862 [inline] check_prevs_add kernel/locking/lockdep.c:1975 [inline] validate_chain kernel/locking/lockdep.c:2416 [inline] __lock_acquire+0x3449/0x5020 kernel/locking/lockdep.c:3412 lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x457089 Code: fd b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 00 66 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 cb b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007f2df20a5c78 EFLAGS: 00000246 ORIG_RAX: 000000000000002e RAX: ffffffffffffffda RBX: 00007f2df20a66d4 RCX: 0000000000457089 RDX: 0000000000000000 RSI: 0000000020000180 RDI: 0000000000000003 RBP: 0000000000930140 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 00000000ffffffff R13: 00000000004d40b8 R14: 00000000004c8ad8 R15: 0000000000000001 Signed-off-by: Mahesh Bandewar <maheshb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-25 04:40:11 +07:00
static void bond_netdev_notify_work(struct work_struct *work);
/*---------------------------- General routines -----------------------------*/
const char *bond_mode_name(int mode)
{
static const char *names[] = {
[BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
[BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
[BOND_MODE_XOR] = "load balancing (xor)",
[BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
[BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
[BOND_MODE_TLB] = "transmit load balancing",
[BOND_MODE_ALB] = "adaptive load balancing",
};
if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
return "unknown";
return names[mode];
}
/*---------------------------------- VLAN -----------------------------------*/
/**
* bond_dev_queue_xmit - Prepare skb for xmit.
*
* @bond: bond device that got this skb for tx.
* @skb: hw accel VLAN tagged skb to transmit
* @slave_dev: slave that is supposed to xmit this skbuff
*/
void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
struct net_device *slave_dev)
{
skb->dev = slave_dev;
bonding: Fix corrupted queue_mapping In the transmit path of the bonding driver, skb->cb is used to stash the skb->queue_mapping so that the bonding device can set its own queue mapping. This value becomes corrupted since the skb->cb is also used in __dev_xmit_skb. When transmitting through bonding driver, bond_select_queue is called from dev_queue_xmit. In bond_select_queue the original skb->queue_mapping is copied into skb->cb (via bond_queue_mapping) and skb->queue_mapping is overwritten with the bond driver queue. Subsequently in dev_queue_xmit, __dev_xmit_skb is called which writes the packet length into skb->cb, thereby overwriting the stashed queue mappping. In bond_dev_queue_xmit (called from hard_start_xmit), the queue mapping for the skb is set to the stashed value which is now the skb length and hence is an invalid queue for the slave device. If we want to save skb->queue_mapping into skb->cb[], best place is to add a field in struct qdisc_skb_cb, to make sure it wont conflict with other layers (eg : Qdiscc, Infiniband...) This patchs also makes sure (struct qdisc_skb_cb)->data is aligned on 8 bytes : netem qdisc for example assumes it can store an u64 in it, without misalignment penalty. Note : we only have 20 bytes left in (struct qdisc_skb_cb)->data[]. The largest user is CHOKe and it fills it. Based on a previous patch from Tom Herbert. Signed-off-by: Eric Dumazet <edumazet@google.com> Reported-by: Tom Herbert <therbert@google.com> Cc: John Fastabend <john.r.fastabend@intel.com> Cc: Roland Dreier <roland@kernel.org> Acked-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-06-12 13:03:51 +07:00
BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
if (unlikely(netpoll_tx_running(bond->dev)))
bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
else
dev_queue_xmit(skb);
}
/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
* We don't protect the slave list iteration with a lock because:
* a. This operation is performed in IOCTL context,
* b. The operation is protected by the RTNL semaphore in the 8021q code,
* c. Holding a lock with BH disabled while directly calling a base driver
* entry point is generally a BAD idea.
*
* The design of synchronization/protection for this operation in the 8021q
* module is good for one or more VLAN devices over a single physical device
* and cannot be extended for a teaming solution like bonding, so there is a
* potential race condition here where a net device from the vlan group might
* be referenced (either by a base driver or the 8021q code) while it is being
* removed from the system. However, it turns out we're not making matters
* worse, and if it works for regular VLAN usage it will work here too.
*/
/**
* bond_vlan_rx_add_vid - Propagates adding an id to slaves
* @bond_dev: bonding net device that got called
* @vid: vlan id being added
*/
static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
__be16 proto, u16 vid)
{
struct bonding *bond = netdev_priv(bond_dev);
struct slave *slave, *rollback_slave;
struct list_head *iter;
int res;
bond_for_each_slave(bond, slave, iter) {
res = vlan_vid_add(slave->dev, proto, vid);
if (res)
goto unwind;
}
return 0;
unwind:
/* unwind to the slave that failed */
bond_for_each_slave(bond, rollback_slave, iter) {
if (rollback_slave == slave)
break;
vlan_vid_del(rollback_slave->dev, proto, vid);
}
return res;
}
/**
* bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
* @bond_dev: bonding net device that got called
* @vid: vlan id being removed
*/
static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
__be16 proto, u16 vid)
{
struct bonding *bond = netdev_priv(bond_dev);
struct list_head *iter;
struct slave *slave;
bond_for_each_slave(bond, slave, iter)
vlan_vid_del(slave->dev, proto, vid);
if (bond_is_lb(bond))
bond_alb_clear_vlan(bond, vid);
return 0;
}
/*------------------------------- Link status -------------------------------*/
/* Set the carrier state for the master according to the state of its
* slaves. If any slaves are up, the master is up. In 802.3ad mode,
* do special 802.3ad magic.
*
* Returns zero if carrier state does not change, nonzero if it does.
*/
int bond_set_carrier(struct bonding *bond)
{
struct list_head *iter;
struct slave *slave;
if (!bond_has_slaves(bond))
goto down;
if (BOND_MODE(bond) == BOND_MODE_8023AD)
return bond_3ad_set_carrier(bond);
bond_for_each_slave(bond, slave, iter) {
if (slave->link == BOND_LINK_UP) {
if (!netif_carrier_ok(bond->dev)) {
netif_carrier_on(bond->dev);
return 1;
}
return 0;
}
}
down:
if (netif_carrier_ok(bond->dev)) {
netif_carrier_off(bond->dev);
return 1;
}
return 0;
}
/* Get link speed and duplex from the slave's base driver
* using ethtool. If for some reason the call fails or the
* values are invalid, set speed and duplex to -1,
* and return. Return 1 if speed or duplex settings are
* UNKNOWN; 0 otherwise.
*/
static int bond_update_speed_duplex(struct slave *slave)
{
struct net_device *slave_dev = slave->dev;
struct ethtool_link_ksettings ecmd;
int res;
slave->speed = SPEED_UNKNOWN;
slave->duplex = DUPLEX_UNKNOWN;
res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
if (res < 0)
return 1;
if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
return 1;
switch (ecmd.base.duplex) {
case DUPLEX_FULL:
case DUPLEX_HALF:
break;
default:
return 1;
}
slave->speed = ecmd.base.speed;
slave->duplex = ecmd.base.duplex;
return 0;
}
const char *bond_slave_link_status(s8 link)
{
switch (link) {
case BOND_LINK_UP:
return "up";
case BOND_LINK_FAIL:
return "going down";
case BOND_LINK_DOWN:
return "down";
case BOND_LINK_BACK:
return "going back";
default:
return "unknown";
}
}
/* if <dev> supports MII link status reporting, check its link status.
*
* We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
* depending upon the setting of the use_carrier parameter.
*
* Return either BMSR_LSTATUS, meaning that the link is up (or we
* can't tell and just pretend it is), or 0, meaning that the link is
* down.
*
* If reporting is non-zero, instead of faking link up, return -1 if
* both ETHTOOL and MII ioctls fail (meaning the device does not
* support them). If use_carrier is set, return whatever it says.
* It'd be nice if there was a good way to tell if a driver supports
* netif_carrier, but there really isn't.
*/
static int bond_check_dev_link(struct bonding *bond,
struct net_device *slave_dev, int reporting)
{
const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
int (*ioctl)(struct net_device *, struct ifreq *, int);
struct ifreq ifr;
struct mii_ioctl_data *mii;
if (!reporting && !netif_running(slave_dev))
return 0;
if (bond->params.use_carrier)
return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
/* Try to get link status using Ethtool first. */
if (slave_dev->ethtool_ops->get_link)
return slave_dev->ethtool_ops->get_link(slave_dev) ?
BMSR_LSTATUS : 0;
/* Ethtool can't be used, fallback to MII ioctls. */
ioctl = slave_ops->ndo_do_ioctl;
if (ioctl) {
/* TODO: set pointer to correct ioctl on a per team member
* bases to make this more efficient. that is, once
* we determine the correct ioctl, we will always
* call it and not the others for that team
* member.
*/
/* We cannot assume that SIOCGMIIPHY will also read a
* register; not all network drivers (e.g., e100)
* support that.
*/
/* Yes, the mii is overlaid on the ifreq.ifr_ifru */
strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
mii = if_mii(&ifr);
if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
mii->reg_num = MII_BMSR;
if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
return mii->val_out & BMSR_LSTATUS;
}
}
/* If reporting, report that either there's no dev->do_ioctl,
* or both SIOCGMIIREG and get_link failed (meaning that we
* cannot report link status). If not reporting, pretend
* we're ok.
*/
return reporting ? -1 : BMSR_LSTATUS;
}
/*----------------------------- Multicast list ------------------------------*/
/* Push the promiscuity flag down to appropriate slaves */
static int bond_set_promiscuity(struct bonding *bond, int inc)
{
struct list_head *iter;
int err = 0;
if (bond_uses_primary(bond)) {
struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
if (curr_active)
err = dev_set_promiscuity(curr_active->dev, inc);
} else {
struct slave *slave;
bond_for_each_slave(bond, slave, iter) {
err = dev_set_promiscuity(slave->dev, inc);
if (err)
return err;
}
}
return err;
}
/* Push the allmulti flag down to all slaves */
static int bond_set_allmulti(struct bonding *bond, int inc)
{
struct list_head *iter;
int err = 0;
if (bond_uses_primary(bond)) {
struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
if (curr_active)
err = dev_set_allmulti(curr_active->dev, inc);
} else {
struct slave *slave;
bond_for_each_slave(bond, slave, iter) {
err = dev_set_allmulti(slave->dev, inc);
if (err)
return err;
}
}
return err;
}
/* Retrieve the list of registered multicast addresses for the bonding
* device and retransmit an IGMP JOIN request to the current active
* slave.
*/
static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
{
struct bonding *bond = container_of(work, struct bonding,
mcast_work.work);
if (!rtnl_trylock()) {
queue_delayed_work(bond->wq, &bond->mcast_work, 1);
return;
}
call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
bonding: fix igmp_retrans type and two related races First the type of igmp_retrans (which is the actual counter of igmp_resend parameter) is changed to u8 to be able to store values up to 255 (as per documentation). There are two races that were hidden there and which are easy to trigger after the previous fix, the first is between bond_resend_igmp_join_requests and bond_change_active_slave where igmp_retrans is set and can be altered by the periodic. The second race condition is between multiple running instances of the periodic (upon execution it can be scheduled again for immediate execution which can cause the counter to go < 0 which in the unsigned case leads to unnecessary igmp retransmissions). Since in bond_change_active_slave bond->lock is held for reading and curr_slave_lock for writing, we use curr_slave_lock for mutual exclusion. We can't drop them as there're cases where RTNL is not held when bond_change_active_slave is called. RCU is unlocked in bond_resend_igmp_join_requests before getting curr_slave_lock since we don't need it there and it's pointless to delay. The decrement is moved inside the "if" block because if we decrement unconditionally there's still a possibility for a race condition although it is much more difficult to hit (many changes have to happen in a very short period in order to trigger) which in the case of 3 parallel running instances of this function and igmp_retrans == 1 (with check bond->igmp_retrans-- > 1) is: f1 passes, doesn't re-schedule, but decrements - igmp_retrans = 0 f2 then passes, doesn't re-schedule, but decrements - igmp_retrans = 255 f3 does the unnecessary retransmissions. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-12 05:07:02 +07:00
if (bond->igmp_retrans > 1) {
bond->igmp_retrans--;
queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
bonding: fix igmp_retrans type and two related races First the type of igmp_retrans (which is the actual counter of igmp_resend parameter) is changed to u8 to be able to store values up to 255 (as per documentation). There are two races that were hidden there and which are easy to trigger after the previous fix, the first is between bond_resend_igmp_join_requests and bond_change_active_slave where igmp_retrans is set and can be altered by the periodic. The second race condition is between multiple running instances of the periodic (upon execution it can be scheduled again for immediate execution which can cause the counter to go < 0 which in the unsigned case leads to unnecessary igmp retransmissions). Since in bond_change_active_slave bond->lock is held for reading and curr_slave_lock for writing, we use curr_slave_lock for mutual exclusion. We can't drop them as there're cases where RTNL is not held when bond_change_active_slave is called. RCU is unlocked in bond_resend_igmp_join_requests before getting curr_slave_lock since we don't need it there and it's pointless to delay. The decrement is moved inside the "if" block because if we decrement unconditionally there's still a possibility for a race condition although it is much more difficult to hit (many changes have to happen in a very short period in order to trigger) which in the case of 3 parallel running instances of this function and igmp_retrans == 1 (with check bond->igmp_retrans-- > 1) is: f1 passes, doesn't re-schedule, but decrements - igmp_retrans = 0 f2 then passes, doesn't re-schedule, but decrements - igmp_retrans = 255 f3 does the unnecessary retransmissions. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-12 05:07:02 +07:00
}
rtnl_unlock();
}
/* Flush bond's hardware addresses from slave */
static void bond_hw_addr_flush(struct net_device *bond_dev,
struct net_device *slave_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
dev_uc_unsync(slave_dev, bond_dev);
dev_mc_unsync(slave_dev, bond_dev);
if (BOND_MODE(bond) == BOND_MODE_8023AD) {
/* del lacpdu mc addr from mc list */
u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
dev_mc_del(slave_dev, lacpdu_multicast);
}
}
/*--------------------------- Active slave change ---------------------------*/
/* Update the hardware address list and promisc/allmulti for the new and
* old active slaves (if any). Modes that are not using primary keep all
* slaves up date at all times; only the modes that use primary need to call
* this function to swap these settings during a failover.
*/
static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
struct slave *old_active)
{
if (old_active) {
if (bond->dev->flags & IFF_PROMISC)
dev_set_promiscuity(old_active->dev, -1);
if (bond->dev->flags & IFF_ALLMULTI)
dev_set_allmulti(old_active->dev, -1);
bond_hw_addr_flush(bond->dev, old_active->dev);
}
if (new_active) {
/* FIXME: Signal errors upstream. */
if (bond->dev->flags & IFF_PROMISC)
dev_set_promiscuity(new_active->dev, 1);
if (bond->dev->flags & IFF_ALLMULTI)
dev_set_allmulti(new_active->dev, 1);
netif_addr_lock_bh(bond->dev);
dev_uc_sync(new_active->dev, bond->dev);
dev_mc_sync(new_active->dev, bond->dev);
netif_addr_unlock_bh(bond->dev);
}
}
/**
* bond_set_dev_addr - clone slave's address to bond
* @bond_dev: bond net device
* @slave_dev: slave net device
*
* Should be called with RTNL held.
*/
static void bond_set_dev_addr(struct net_device *bond_dev,
struct net_device *slave_dev)
{
netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
bond_dev->addr_assign_type = NET_ADDR_STOLEN;
call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
}
static struct slave *bond_get_old_active(struct bonding *bond,
struct slave *new_active)
{
struct slave *slave;
struct list_head *iter;
bond_for_each_slave(bond, slave, iter) {
if (slave == new_active)
continue;
if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
return slave;
}
return NULL;
}
/* bond_do_fail_over_mac
*
* Perform special MAC address swapping for fail_over_mac settings
*
* Called with RTNL
*/
static void bond_do_fail_over_mac(struct bonding *bond,
struct slave *new_active,
struct slave *old_active)
{
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
u8 tmp_mac[MAX_ADDR_LEN];
struct sockaddr_storage ss;
int rv;
switch (bond->params.fail_over_mac) {
case BOND_FOM_ACTIVE:
if (new_active)
bond_set_dev_addr(bond->dev, new_active->dev);
break;
case BOND_FOM_FOLLOW:
/* if new_active && old_active, swap them
* if just old_active, do nothing (going to no active slave)
* if just new_active, set new_active to bond's MAC
*/
if (!new_active)
return;
if (!old_active)
old_active = bond_get_old_active(bond, new_active);
if (old_active) {
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
new_active->dev->addr_len);
bond_hw_addr_copy(ss.__data,
old_active->dev->dev_addr,
old_active->dev->addr_len);
ss.ss_family = new_active->dev->type;
} else {
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
bond->dev->addr_len);
ss.ss_family = bond->dev->type;
}
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
rv = dev_set_mac_address(new_active->dev,
(struct sockaddr *)&ss);
if (rv) {
netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
-rv, new_active->dev->name);
goto out;
}
if (!old_active)
goto out;
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
bond_hw_addr_copy(ss.__data, tmp_mac,
new_active->dev->addr_len);
ss.ss_family = old_active->dev->type;
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
rv = dev_set_mac_address(old_active->dev,
(struct sockaddr *)&ss);
if (rv)
netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
-rv, new_active->dev->name);
out:
break;
default:
netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
bond->params.fail_over_mac);
break;
}
}
static struct slave *bond_choose_primary_or_current(struct bonding *bond)
{
struct slave *prim = rtnl_dereference(bond->primary_slave);
struct slave *curr = rtnl_dereference(bond->curr_active_slave);
if (!prim || prim->link != BOND_LINK_UP) {
if (!curr || curr->link != BOND_LINK_UP)
return NULL;
return curr;
}
if (bond->force_primary) {
bond->force_primary = false;
return prim;
}
if (!curr || curr->link != BOND_LINK_UP)
return prim;
/* At this point, prim and curr are both up */
switch (bond->params.primary_reselect) {
case BOND_PRI_RESELECT_ALWAYS:
return prim;
case BOND_PRI_RESELECT_BETTER:
if (prim->speed < curr->speed)
return curr;
if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
return curr;
return prim;
case BOND_PRI_RESELECT_FAILURE:
return curr;
default:
netdev_err(bond->dev, "impossible primary_reselect %d\n",
bond->params.primary_reselect);
return curr;
}
}
/**
* bond_find_best_slave - select the best available slave to be the active one
* @bond: our bonding struct
*/
static struct slave *bond_find_best_slave(struct bonding *bond)
{
struct slave *slave, *bestslave = NULL;
struct list_head *iter;
int mintime = bond->params.updelay;
slave = bond_choose_primary_or_current(bond);
if (slave)
return slave;
bond_for_each_slave(bond, slave, iter) {
if (slave->link == BOND_LINK_UP)
return slave;
if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
slave->delay < mintime) {
mintime = slave->delay;
bestslave = slave;
}
}
return bestslave;
}
static bool bond_should_notify_peers(struct bonding *bond)
{
struct slave *slave;
rcu_read_lock();
slave = rcu_dereference(bond->curr_active_slave);
rcu_read_unlock();
netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
slave ? slave->dev->name : "NULL");
if (!slave || !bond->send_peer_notif ||
!netif_carrier_ok(bond->dev) ||
test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
return false;
return true;
}
/**
* change_active_interface - change the active slave into the specified one
* @bond: our bonding struct
* @new: the new slave to make the active one
*
* Set the new slave to the bond's settings and unset them on the old
* curr_active_slave.
* Setting include flags, mc-list, promiscuity, allmulti, etc.
*
* If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
* because it is apparently the best available slave we have, even though its
* updelay hasn't timed out yet.
*
* Caller must hold RTNL.
*/
void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
{
struct slave *old_active;
ASSERT_RTNL();
old_active = rtnl_dereference(bond->curr_active_slave);
if (old_active == new_active)
return;
if (new_active) {
new_active->last_link_up = jiffies;
if (new_active->link == BOND_LINK_BACK) {
if (bond_uses_primary(bond)) {
netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
new_active->dev->name,
(bond->params.updelay - new_active->delay) * bond->params.miimon);
}
new_active->delay = 0;
bond_set_slave_link_state(new_active, BOND_LINK_UP,
BOND_SLAVE_NOTIFY_NOW);
if (BOND_MODE(bond) == BOND_MODE_8023AD)
bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
if (bond_is_lb(bond))
bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
} else {
if (bond_uses_primary(bond)) {
netdev_info(bond->dev, "making interface %s the new active one\n",
new_active->dev->name);
}
}
}
if (bond_uses_primary(bond))
bond_hw_addr_swap(bond, new_active, old_active);
if (bond_is_lb(bond)) {
bond_alb_handle_active_change(bond, new_active);
if (old_active)
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_inactive_flags(old_active,
BOND_SLAVE_NOTIFY_NOW);
if (new_active)
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_active_flags(new_active,
BOND_SLAVE_NOTIFY_NOW);
} else {
bonding: initial RCU conversion This patch does the initial bonding conversion to RCU. After it the following modes are protected by RCU alone: roundrobin, active-backup, broadcast and xor. Modes ALB/TLB and 3ad still acquire bond->lock for reading, and will be dealt with later. curr_active_slave needs to be dereferenced via rcu in the converted modes because the only thing protecting the slave after this patch is rcu_read_lock, so we need the proper barrier for weakly ordered archs and to make sure we don't have stale pointer. It's not tagged with __rcu yet because there's still work to be done to remove the curr_slave_lock, so sparse will complain when rcu_assign_pointer and rcu_dereference are used, but the alternative to use rcu_dereference_protected would've created much bigger code churn which is more difficult to test and review. That will be converted in time. 1. Active-backup mode 1.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.55% in bonding, system spent 0.29% CPU in bonding - new bonding: iperf spent 0.29% in bonding, system spent 0.15% CPU in bonding 1.2. Bandwidth measurements - old bonding: 16.1 gbps consistently - new bonding: 17.5 gbps consistently 2. Round-robin mode 2.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.51% in bonding, system spent 0.24% CPU in bonding - new bonding: iperf spent 0.16% in bonding, system spent 0.11% CPU in bonding 2.2 Bandwidth measurements - old bonding: 8 gbps (variable due to packet reorderings) - new bonding: 10 gbps (variable due to packet reorderings) Of course the latency has improved in all converted modes, and moreover while doing enslave/release (since it doesn't affect tx anymore). Also I've stress tested all modes doing enslave/release in a loop while transmitting traffic. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 21:54:51 +07:00
rcu_assign_pointer(bond->curr_active_slave, new_active);
}
if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
if (old_active)
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_inactive_flags(old_active,
BOND_SLAVE_NOTIFY_NOW);
if (new_active) {
bool should_notify_peers = false;
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_active_flags(new_active,
BOND_SLAVE_NOTIFY_NOW);
if (bond->params.fail_over_mac)
bond_do_fail_over_mac(bond, new_active,
old_active);
if (netif_running(bond->dev)) {
bond->send_peer_notif =
bond->params.num_peer_notif;
should_notify_peers =
bond_should_notify_peers(bond);
}
call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
if (should_notify_peers)
call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
bond->dev);
}
}
/* resend IGMP joins since active slave has changed or
* all were sent on curr_active_slave.
* resend only if bond is brought up with the affected
* bonding modes and the retransmission is enabled
*/
if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
((bond_uses_primary(bond) && new_active) ||
BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
bond->igmp_retrans = bond->params.resend_igmp;
queue_delayed_work(bond->wq, &bond->mcast_work, 1);
}
}
/**
* bond_select_active_slave - select a new active slave, if needed
* @bond: our bonding struct
*
* This functions should be called when one of the following occurs:
* - The old curr_active_slave has been released or lost its link.
* - The primary_slave has got its link back.
* - A slave has got its link back and there's no old curr_active_slave.
*
* Caller must hold RTNL.
*/
void bond_select_active_slave(struct bonding *bond)
{
struct slave *best_slave;
int rv;
ASSERT_RTNL();
best_slave = bond_find_best_slave(bond);
if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
bond_change_active_slave(bond, best_slave);
rv = bond_set_carrier(bond);
if (!rv)
return;
if (netif_carrier_ok(bond->dev))
netdev_info(bond->dev, "first active interface up!\n");
else
netdev_info(bond->dev, "now running without any active interface!\n");
}
}
#ifdef CONFIG_NET_POLL_CONTROLLER
static inline int slave_enable_netpoll(struct slave *slave)
{
struct netpoll *np;
int err = 0;
netpoll: Remove gfp parameter from __netpoll_setup The gfp parameter was added in: commit 47be03a28cc6c80e3aa2b3e8ed6d960ff0c5c0af Author: Amerigo Wang <amwang@redhat.com> Date: Fri Aug 10 01:24:37 2012 +0000 netpoll: use GFP_ATOMIC in slave_enable_netpoll() and __netpoll_setup() slave_enable_netpoll() and __netpoll_setup() may be called with read_lock() held, so should use GFP_ATOMIC to allocate memory. Eric suggested to pass gfp flags to __netpoll_setup(). Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: "David S. Miller" <davem@davemloft.net> Reported-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Cong Wang <amwang@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> The reason for the gfp parameter was removed in: commit c4cdef9b7183159c23c7302aaf270d64c549f557 Author: dingtianhong <dingtianhong@huawei.com> Date: Tue Jul 23 15:25:27 2013 +0800 bonding: don't call slave_xxx_netpoll under spinlocks The slave_xxx_netpoll will call synchronize_rcu_bh(), so the function may schedule and sleep, it should't be called under spinlocks. bond_netpoll_setup() and bond_netpoll_cleanup() are always protected by rtnl lock, it is no need to take the read lock, as the slave list couldn't be changed outside rtnl lock. Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net> Nothing else that calls __netpoll_setup or ndo_netpoll_setup requires a gfp paramter, so remove the gfp parameter from both of these functions making the code clearer. Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-03-28 05:36:38 +07:00
np = kzalloc(sizeof(*np), GFP_KERNEL);
err = -ENOMEM;
if (!np)
goto out;
netpoll: Remove gfp parameter from __netpoll_setup The gfp parameter was added in: commit 47be03a28cc6c80e3aa2b3e8ed6d960ff0c5c0af Author: Amerigo Wang <amwang@redhat.com> Date: Fri Aug 10 01:24:37 2012 +0000 netpoll: use GFP_ATOMIC in slave_enable_netpoll() and __netpoll_setup() slave_enable_netpoll() and __netpoll_setup() may be called with read_lock() held, so should use GFP_ATOMIC to allocate memory. Eric suggested to pass gfp flags to __netpoll_setup(). Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: "David S. Miller" <davem@davemloft.net> Reported-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Cong Wang <amwang@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> The reason for the gfp parameter was removed in: commit c4cdef9b7183159c23c7302aaf270d64c549f557 Author: dingtianhong <dingtianhong@huawei.com> Date: Tue Jul 23 15:25:27 2013 +0800 bonding: don't call slave_xxx_netpoll under spinlocks The slave_xxx_netpoll will call synchronize_rcu_bh(), so the function may schedule and sleep, it should't be called under spinlocks. bond_netpoll_setup() and bond_netpoll_cleanup() are always protected by rtnl lock, it is no need to take the read lock, as the slave list couldn't be changed outside rtnl lock. Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net> Nothing else that calls __netpoll_setup or ndo_netpoll_setup requires a gfp paramter, so remove the gfp parameter from both of these functions making the code clearer. Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-03-28 05:36:38 +07:00
err = __netpoll_setup(np, slave->dev);
if (err) {
kfree(np);
goto out;
}
slave->np = np;
out:
return err;
}
static inline void slave_disable_netpoll(struct slave *slave)
{
struct netpoll *np = slave->np;
if (!np)
return;
slave->np = NULL;
__netpoll_free(np);
}
static void bond_poll_controller(struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
struct slave *slave = NULL;
struct list_head *iter;
struct ad_info ad_info;
if (BOND_MODE(bond) == BOND_MODE_8023AD)
if (bond_3ad_get_active_agg_info(bond, &ad_info))
return;
bond_for_each_slave_rcu(bond, slave, iter) {
if (!bond_slave_is_up(slave))
continue;
if (BOND_MODE(bond) == BOND_MODE_8023AD) {
struct aggregator *agg =
SLAVE_AD_INFO(slave)->port.aggregator;
if (agg &&
agg->aggregator_identifier != ad_info.aggregator_id)
continue;
}
netpoll_poll_dev(slave->dev);
}
}
static void bond_netpoll_cleanup(struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
struct list_head *iter;
struct slave *slave;
bond_for_each_slave(bond, slave, iter)
if (bond_slave_is_up(slave))
slave_disable_netpoll(slave);
}
netpoll: Remove gfp parameter from __netpoll_setup The gfp parameter was added in: commit 47be03a28cc6c80e3aa2b3e8ed6d960ff0c5c0af Author: Amerigo Wang <amwang@redhat.com> Date: Fri Aug 10 01:24:37 2012 +0000 netpoll: use GFP_ATOMIC in slave_enable_netpoll() and __netpoll_setup() slave_enable_netpoll() and __netpoll_setup() may be called with read_lock() held, so should use GFP_ATOMIC to allocate memory. Eric suggested to pass gfp flags to __netpoll_setup(). Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: "David S. Miller" <davem@davemloft.net> Reported-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Cong Wang <amwang@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> The reason for the gfp parameter was removed in: commit c4cdef9b7183159c23c7302aaf270d64c549f557 Author: dingtianhong <dingtianhong@huawei.com> Date: Tue Jul 23 15:25:27 2013 +0800 bonding: don't call slave_xxx_netpoll under spinlocks The slave_xxx_netpoll will call synchronize_rcu_bh(), so the function may schedule and sleep, it should't be called under spinlocks. bond_netpoll_setup() and bond_netpoll_cleanup() are always protected by rtnl lock, it is no need to take the read lock, as the slave list couldn't be changed outside rtnl lock. Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net> Nothing else that calls __netpoll_setup or ndo_netpoll_setup requires a gfp paramter, so remove the gfp parameter from both of these functions making the code clearer. Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-03-28 05:36:38 +07:00
static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
{
struct bonding *bond = netdev_priv(dev);
struct list_head *iter;
struct slave *slave;
int err = 0;
bond_for_each_slave(bond, slave, iter) {
err = slave_enable_netpoll(slave);
if (err) {
bond_netpoll_cleanup(dev);
break;
}
}
return err;
}
#else
static inline int slave_enable_netpoll(struct slave *slave)
{
return 0;
}
static inline void slave_disable_netpoll(struct slave *slave)
{
}
static void bond_netpoll_cleanup(struct net_device *bond_dev)
{
}
#endif
/*---------------------------------- IOCTL ----------------------------------*/
static netdev_features_t bond_fix_features(struct net_device *dev,
netdev_features_t features)
{
struct bonding *bond = netdev_priv(dev);
struct list_head *iter;
netdev_features_t mask;
struct slave *slave;
mask = features;
features &= ~NETIF_F_ONE_FOR_ALL;
features |= NETIF_F_ALL_FOR_ALL;
bond_for_each_slave(bond, slave, iter) {
features = netdev_increment_features(features,
slave->dev->features,
mask);
}
features = netdev_add_tso_features(features, mask);
return features;
}
#define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
NETIF_F_HIGHDMA | NETIF_F_LRO)
#define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
static void bond_compute_features(struct bonding *bond)
{
unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
IFF_XMIT_DST_RELEASE_PERM;
netdev_features_t vlan_features = BOND_VLAN_FEATURES;
netdev_features_t enc_features = BOND_ENC_FEATURES;
struct net_device *bond_dev = bond->dev;
struct list_head *iter;
struct slave *slave;
unsigned short max_hard_header_len = ETH_HLEN;
unsigned int gso_max_size = GSO_MAX_SIZE;
u16 gso_max_segs = GSO_MAX_SEGS;
if (!bond_has_slaves(bond))
goto done;
vlan_features &= NETIF_F_ALL_FOR_ALL;
bond_for_each_slave(bond, slave, iter) {
vlan_features = netdev_increment_features(vlan_features,
slave->dev->vlan_features, BOND_VLAN_FEATURES);
enc_features = netdev_increment_features(enc_features,
slave->dev->hw_enc_features,
BOND_ENC_FEATURES);
dst_release_flag &= slave->dev->priv_flags;
if (slave->dev->hard_header_len > max_hard_header_len)
max_hard_header_len = slave->dev->hard_header_len;
gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
}
bonding: avoid defaulting hard_header_len to ETH_HLEN on slave removal On slave list updates, the bonding driver computes its hard_header_len as the maximum of all enslaved devices's hard_header_len. If the slave list is empty, e.g. on last enslaved device removal, ETH_HLEN is used. Since the bonding header_ops are set only when the first enslaved device is attached, the above can lead to header_ops->create() being called with the wrong skb headroom in place. If bond0 is configured on top of ipoib devices, with the following commands: ifup bond0 for slave in $BOND_SLAVES_LIST; do ip link set dev $slave nomaster done ping -c 1 <ip on bond0 subnet> we will obtain a skb_under_panic() with a similar call trace: skb_push+0x3d/0x40 push_pseudo_header+0x17/0x30 [ib_ipoib] ipoib_hard_header+0x4e/0x80 [ib_ipoib] arp_create+0x12f/0x220 arp_send_dst.part.19+0x28/0x50 arp_solicit+0x115/0x290 neigh_probe+0x4d/0x70 __neigh_event_send+0xa7/0x230 neigh_resolve_output+0x12e/0x1c0 ip_finish_output2+0x14b/0x390 ip_finish_output+0x136/0x1e0 ip_output+0x76/0xe0 ip_local_out+0x35/0x40 ip_send_skb+0x19/0x40 ip_push_pending_frames+0x33/0x40 raw_sendmsg+0x7d3/0xb50 inet_sendmsg+0x31/0xb0 sock_sendmsg+0x38/0x50 SYSC_sendto+0x102/0x190 SyS_sendto+0xe/0x10 do_syscall_64+0x67/0x180 entry_SYSCALL64_slow_path+0x25/0x25 This change addresses the issue avoiding updating the bonding device hard_header_len when the slaves list become empty, forbidding to shrink it below the value used by header_ops->create(). The bug is there since commit 54ef31371407 ("[PATCH] bonding: Handle large hard_header_len") but the panic can be triggered only since commit fc791b633515 ("IB/ipoib: move back IB LL address into the hard header"). Reported-by: Norbert P <noe@physik.uzh.ch> Fixes: 54ef31371407 ("[PATCH] bonding: Handle large hard_header_len") Fixes: fc791b633515 ("IB/ipoib: move back IB LL address into the hard header") Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-28 00:29:34 +07:00
bond_dev->hard_header_len = max_hard_header_len;
done:
bond_dev->vlan_features = vlan_features;
bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
NETIF_F_GSO_UDP_L4;
bond_dev->gso_max_segs = gso_max_segs;
netif_set_gso_max_size(bond_dev, gso_max_size);
bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
netdev_change_features(bond_dev);
}
static void bond_setup_by_slave(struct net_device *bond_dev,
struct net_device *slave_dev)
{
bond_dev->header_ops = slave_dev->header_ops;
bond_dev->type = slave_dev->type;
bond_dev->hard_header_len = slave_dev->hard_header_len;
bond_dev->addr_len = slave_dev->addr_len;
memcpy(bond_dev->broadcast, slave_dev->broadcast,
slave_dev->addr_len);
}
/* On bonding slaves other than the currently active slave, suppress
* duplicates except for alb non-mcast/bcast.
*/
static bool bond_should_deliver_exact_match(struct sk_buff *skb,
struct slave *slave,
struct bonding *bond)
{
if (bond_is_slave_inactive(slave)) {
if (BOND_MODE(bond) == BOND_MODE_ALB &&
skb->pkt_type != PACKET_BROADCAST &&
skb->pkt_type != PACKET_MULTICAST)
return false;
return true;
}
return false;
}
static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
{
struct sk_buff *skb = *pskb;
struct slave *slave;
struct bonding *bond;
int (*recv_probe)(const struct sk_buff *, struct bonding *,
struct slave *);
int ret = RX_HANDLER_ANOTHER;
skb = skb_share_check(skb, GFP_ATOMIC);
if (unlikely(!skb))
return RX_HANDLER_CONSUMED;
*pskb = skb;
slave = bond_slave_get_rcu(skb->dev);
bond = slave->bond;
locking/atomics: COCCINELLE/treewide: Convert trivial ACCESS_ONCE() patterns to READ_ONCE()/WRITE_ONCE() Please do not apply this to mainline directly, instead please re-run the coccinelle script shown below and apply its output. For several reasons, it is desirable to use {READ,WRITE}_ONCE() in preference to ACCESS_ONCE(), and new code is expected to use one of the former. So far, there's been no reason to change most existing uses of ACCESS_ONCE(), as these aren't harmful, and changing them results in churn. However, for some features, the read/write distinction is critical to correct operation. To distinguish these cases, separate read/write accessors must be used. This patch migrates (most) remaining ACCESS_ONCE() instances to {READ,WRITE}_ONCE(), using the following coccinelle script: ---- // Convert trivial ACCESS_ONCE() uses to equivalent READ_ONCE() and // WRITE_ONCE() // $ make coccicheck COCCI=/home/mark/once.cocci SPFLAGS="--include-headers" MODE=patch virtual patch @ depends on patch @ expression E1, E2; @@ - ACCESS_ONCE(E1) = E2 + WRITE_ONCE(E1, E2) @ depends on patch @ expression E; @@ - ACCESS_ONCE(E) + READ_ONCE(E) ---- Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: davem@davemloft.net Cc: linux-arch@vger.kernel.org Cc: mpe@ellerman.id.au Cc: shuah@kernel.org Cc: snitzer@redhat.com Cc: thor.thayer@linux.intel.com Cc: tj@kernel.org Cc: viro@zeniv.linux.org.uk Cc: will.deacon@arm.com Link: http://lkml.kernel.org/r/1508792849-3115-19-git-send-email-paulmck@linux.vnet.ibm.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-24 04:07:29 +07:00
recv_probe = READ_ONCE(bond->recv_probe);
if (recv_probe) {
ret = recv_probe(skb, bond, slave);
if (ret == RX_HANDLER_CONSUMED) {
consume_skb(skb);
return ret;
}
}
/* Link-local multicast packets should be passed to the
* stack on the link they arrive as well as pass them to the
* bond-master device. These packets are mostly usable when
* stack receives it with the link on which they arrive
* (e.g. LLDP) they also must be available on master. Some of
* the use cases include (but are not limited to): LLDP agents
* that must be able to operate both on enslaved interfaces as
* well as on bonds themselves; linux bridges that must be able
* to process/pass BPDUs from attached bonds when any kind of
* STP version is enabled on the network.
*/
if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) {
struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
if (nskb) {
nskb->dev = bond->dev;
nskb->queue_mapping = 0;
netif_rx(nskb);
}
return RX_HANDLER_PASS;
}
if (bond_should_deliver_exact_match(skb, slave, bond))
return RX_HANDLER_EXACT;
skb->dev = bond->dev;
if (BOND_MODE(bond) == BOND_MODE_ALB &&
bond->dev->priv_flags & IFF_BRIDGE_PORT &&
skb->pkt_type == PACKET_HOST) {
if (unlikely(skb_cow_head(skb,
skb->data - skb_mac_header(skb)))) {
kfree_skb(skb);
return RX_HANDLER_CONSUMED;
}
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
bond->dev->addr_len);
}
return ret;
}
static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
{
switch (BOND_MODE(bond)) {
case BOND_MODE_ROUNDROBIN:
return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
case BOND_MODE_ACTIVEBACKUP:
return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
case BOND_MODE_BROADCAST:
return NETDEV_LAG_TX_TYPE_BROADCAST;
case BOND_MODE_XOR:
case BOND_MODE_8023AD:
return NETDEV_LAG_TX_TYPE_HASH;
default:
return NETDEV_LAG_TX_TYPE_UNKNOWN;
}
}
static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
enum netdev_lag_tx_type type)
{
if (type != NETDEV_LAG_TX_TYPE_HASH)
return NETDEV_LAG_HASH_NONE;
switch (bond->params.xmit_policy) {
case BOND_XMIT_POLICY_LAYER2:
return NETDEV_LAG_HASH_L2;
case BOND_XMIT_POLICY_LAYER34:
return NETDEV_LAG_HASH_L34;
case BOND_XMIT_POLICY_LAYER23:
return NETDEV_LAG_HASH_L23;
case BOND_XMIT_POLICY_ENCAP23:
return NETDEV_LAG_HASH_E23;
case BOND_XMIT_POLICY_ENCAP34:
return NETDEV_LAG_HASH_E34;
default:
return NETDEV_LAG_HASH_UNKNOWN;
}
}
static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
struct netlink_ext_ack *extack)
{
struct netdev_lag_upper_info lag_upper_info;
enum netdev_lag_tx_type type;
type = bond_lag_tx_type(bond);
lag_upper_info.tx_type = type;
lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
&lag_upper_info, extack);
}
static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
{
netdev_upper_dev_unlink(slave->dev, bond->dev);
slave->dev->flags &= ~IFF_SLAVE;
}
static struct slave *bond_alloc_slave(struct bonding *bond)
{
struct slave *slave = NULL;
slave = kzalloc(sizeof(*slave), GFP_KERNEL);
if (!slave)
return NULL;
if (BOND_MODE(bond) == BOND_MODE_8023AD) {
SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
GFP_KERNEL);
if (!SLAVE_AD_INFO(slave)) {
kfree(slave);
return NULL;
}
}
bonding: avoid possible dead-lock Syzkaller reported this on a slightly older kernel but it's still applicable to the current kernel - ====================================================== WARNING: possible circular locking dependency detected 4.18.0-next-20180823+ #46 Not tainted ------------------------------------------------------ syz-executor4/26841 is trying to acquire lock: 00000000dd41ef48 ((wq_completion)bond_dev->name){+.+.}, at: flush_workqueue+0x2db/0x1e10 kernel/workqueue.c:2652 but task is already holding lock: 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnl_lock net/core/rtnetlink.c:77 [inline] 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnetlink_rcv_msg+0x412/0xc30 net/core/rtnetlink.c:4708 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #2 (rtnl_mutex){+.+.}: __mutex_lock_common kernel/locking/mutex.c:925 [inline] __mutex_lock+0x171/0x1700 kernel/locking/mutex.c:1073 mutex_lock_nested+0x16/0x20 kernel/locking/mutex.c:1088 rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77 bond_netdev_notify drivers/net/bonding/bond_main.c:1310 [inline] bond_netdev_notify_work+0x44/0xd0 drivers/net/bonding/bond_main.c:1320 process_one_work+0xc73/0x1aa0 kernel/workqueue.c:2153 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #1 ((work_completion)(&(&nnw->work)->work)){+.+.}: process_one_work+0xc0b/0x1aa0 kernel/workqueue.c:2129 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #0 ((wq_completion)bond_dev->name){+.+.}: lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe other info that might help us debug this: Chain exists of: (wq_completion)bond_dev->name --> (work_completion)(&(&nnw->work)->work) --> rtnl_mutex Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(rtnl_mutex); lock((work_completion)(&(&nnw->work)->work)); lock(rtnl_mutex); lock((wq_completion)bond_dev->name); *** DEADLOCK *** 1 lock held by syz-executor4/26841: stack backtrace: CPU: 1 PID: 26841 Comm: syz-executor4 Not tainted 4.18.0-next-20180823+ #46 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113 print_circular_bug.isra.34.cold.55+0x1bd/0x27d kernel/locking/lockdep.c:1222 check_prev_add kernel/locking/lockdep.c:1862 [inline] check_prevs_add kernel/locking/lockdep.c:1975 [inline] validate_chain kernel/locking/lockdep.c:2416 [inline] __lock_acquire+0x3449/0x5020 kernel/locking/lockdep.c:3412 lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x457089 Code: fd b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 00 66 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 cb b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007f2df20a5c78 EFLAGS: 00000246 ORIG_RAX: 000000000000002e RAX: ffffffffffffffda RBX: 00007f2df20a66d4 RCX: 0000000000457089 RDX: 0000000000000000 RSI: 0000000020000180 RDI: 0000000000000003 RBP: 0000000000930140 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 00000000ffffffff R13: 00000000004d40b8 R14: 00000000004c8ad8 R15: 0000000000000001 Signed-off-by: Mahesh Bandewar <maheshb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-25 04:40:11 +07:00
INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
return slave;
}
static void bond_free_slave(struct slave *slave)
{
struct bonding *bond = bond_get_bond_by_slave(slave);
bonding: avoid possible dead-lock Syzkaller reported this on a slightly older kernel but it's still applicable to the current kernel - ====================================================== WARNING: possible circular locking dependency detected 4.18.0-next-20180823+ #46 Not tainted ------------------------------------------------------ syz-executor4/26841 is trying to acquire lock: 00000000dd41ef48 ((wq_completion)bond_dev->name){+.+.}, at: flush_workqueue+0x2db/0x1e10 kernel/workqueue.c:2652 but task is already holding lock: 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnl_lock net/core/rtnetlink.c:77 [inline] 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnetlink_rcv_msg+0x412/0xc30 net/core/rtnetlink.c:4708 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #2 (rtnl_mutex){+.+.}: __mutex_lock_common kernel/locking/mutex.c:925 [inline] __mutex_lock+0x171/0x1700 kernel/locking/mutex.c:1073 mutex_lock_nested+0x16/0x20 kernel/locking/mutex.c:1088 rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77 bond_netdev_notify drivers/net/bonding/bond_main.c:1310 [inline] bond_netdev_notify_work+0x44/0xd0 drivers/net/bonding/bond_main.c:1320 process_one_work+0xc73/0x1aa0 kernel/workqueue.c:2153 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #1 ((work_completion)(&(&nnw->work)->work)){+.+.}: process_one_work+0xc0b/0x1aa0 kernel/workqueue.c:2129 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #0 ((wq_completion)bond_dev->name){+.+.}: lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe other info that might help us debug this: Chain exists of: (wq_completion)bond_dev->name --> (work_completion)(&(&nnw->work)->work) --> rtnl_mutex Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(rtnl_mutex); lock((work_completion)(&(&nnw->work)->work)); lock(rtnl_mutex); lock((wq_completion)bond_dev->name); *** DEADLOCK *** 1 lock held by syz-executor4/26841: stack backtrace: CPU: 1 PID: 26841 Comm: syz-executor4 Not tainted 4.18.0-next-20180823+ #46 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113 print_circular_bug.isra.34.cold.55+0x1bd/0x27d kernel/locking/lockdep.c:1222 check_prev_add kernel/locking/lockdep.c:1862 [inline] check_prevs_add kernel/locking/lockdep.c:1975 [inline] validate_chain kernel/locking/lockdep.c:2416 [inline] __lock_acquire+0x3449/0x5020 kernel/locking/lockdep.c:3412 lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x457089 Code: fd b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 00 66 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 cb b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007f2df20a5c78 EFLAGS: 00000246 ORIG_RAX: 000000000000002e RAX: ffffffffffffffda RBX: 00007f2df20a66d4 RCX: 0000000000457089 RDX: 0000000000000000 RSI: 0000000020000180 RDI: 0000000000000003 RBP: 0000000000930140 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 00000000ffffffff R13: 00000000004d40b8 R14: 00000000004c8ad8 R15: 0000000000000001 Signed-off-by: Mahesh Bandewar <maheshb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-25 04:40:11 +07:00
cancel_delayed_work_sync(&slave->notify_work);
if (BOND_MODE(bond) == BOND_MODE_8023AD)
kfree(SLAVE_AD_INFO(slave));
kfree(slave);
}
static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
{
info->bond_mode = BOND_MODE(bond);
info->miimon = bond->params.miimon;
info->num_slaves = bond->slave_cnt;
}
static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
{
strcpy(info->slave_name, slave->dev->name);
info->link = slave->link;
info->state = bond_slave_state(slave);
info->link_failure_count = slave->link_failure_count;
}
static void bond_netdev_notify_work(struct work_struct *_work)
{
bonding: avoid possible dead-lock Syzkaller reported this on a slightly older kernel but it's still applicable to the current kernel - ====================================================== WARNING: possible circular locking dependency detected 4.18.0-next-20180823+ #46 Not tainted ------------------------------------------------------ syz-executor4/26841 is trying to acquire lock: 00000000dd41ef48 ((wq_completion)bond_dev->name){+.+.}, at: flush_workqueue+0x2db/0x1e10 kernel/workqueue.c:2652 but task is already holding lock: 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnl_lock net/core/rtnetlink.c:77 [inline] 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnetlink_rcv_msg+0x412/0xc30 net/core/rtnetlink.c:4708 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #2 (rtnl_mutex){+.+.}: __mutex_lock_common kernel/locking/mutex.c:925 [inline] __mutex_lock+0x171/0x1700 kernel/locking/mutex.c:1073 mutex_lock_nested+0x16/0x20 kernel/locking/mutex.c:1088 rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77 bond_netdev_notify drivers/net/bonding/bond_main.c:1310 [inline] bond_netdev_notify_work+0x44/0xd0 drivers/net/bonding/bond_main.c:1320 process_one_work+0xc73/0x1aa0 kernel/workqueue.c:2153 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #1 ((work_completion)(&(&nnw->work)->work)){+.+.}: process_one_work+0xc0b/0x1aa0 kernel/workqueue.c:2129 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #0 ((wq_completion)bond_dev->name){+.+.}: lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe other info that might help us debug this: Chain exists of: (wq_completion)bond_dev->name --> (work_completion)(&(&nnw->work)->work) --> rtnl_mutex Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(rtnl_mutex); lock((work_completion)(&(&nnw->work)->work)); lock(rtnl_mutex); lock((wq_completion)bond_dev->name); *** DEADLOCK *** 1 lock held by syz-executor4/26841: stack backtrace: CPU: 1 PID: 26841 Comm: syz-executor4 Not tainted 4.18.0-next-20180823+ #46 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113 print_circular_bug.isra.34.cold.55+0x1bd/0x27d kernel/locking/lockdep.c:1222 check_prev_add kernel/locking/lockdep.c:1862 [inline] check_prevs_add kernel/locking/lockdep.c:1975 [inline] validate_chain kernel/locking/lockdep.c:2416 [inline] __lock_acquire+0x3449/0x5020 kernel/locking/lockdep.c:3412 lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x457089 Code: fd b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 00 66 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 cb b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007f2df20a5c78 EFLAGS: 00000246 ORIG_RAX: 000000000000002e RAX: ffffffffffffffda RBX: 00007f2df20a66d4 RCX: 0000000000457089 RDX: 0000000000000000 RSI: 0000000020000180 RDI: 0000000000000003 RBP: 0000000000930140 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 00000000ffffffff R13: 00000000004d40b8 R14: 00000000004c8ad8 R15: 0000000000000001 Signed-off-by: Mahesh Bandewar <maheshb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-25 04:40:11 +07:00
struct slave *slave = container_of(_work, struct slave,
notify_work.work);
if (rtnl_trylock()) {
struct netdev_bonding_info binfo;
bonding: avoid possible dead-lock Syzkaller reported this on a slightly older kernel but it's still applicable to the current kernel - ====================================================== WARNING: possible circular locking dependency detected 4.18.0-next-20180823+ #46 Not tainted ------------------------------------------------------ syz-executor4/26841 is trying to acquire lock: 00000000dd41ef48 ((wq_completion)bond_dev->name){+.+.}, at: flush_workqueue+0x2db/0x1e10 kernel/workqueue.c:2652 but task is already holding lock: 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnl_lock net/core/rtnetlink.c:77 [inline] 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnetlink_rcv_msg+0x412/0xc30 net/core/rtnetlink.c:4708 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #2 (rtnl_mutex){+.+.}: __mutex_lock_common kernel/locking/mutex.c:925 [inline] __mutex_lock+0x171/0x1700 kernel/locking/mutex.c:1073 mutex_lock_nested+0x16/0x20 kernel/locking/mutex.c:1088 rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77 bond_netdev_notify drivers/net/bonding/bond_main.c:1310 [inline] bond_netdev_notify_work+0x44/0xd0 drivers/net/bonding/bond_main.c:1320 process_one_work+0xc73/0x1aa0 kernel/workqueue.c:2153 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #1 ((work_completion)(&(&nnw->work)->work)){+.+.}: process_one_work+0xc0b/0x1aa0 kernel/workqueue.c:2129 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #0 ((wq_completion)bond_dev->name){+.+.}: lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe other info that might help us debug this: Chain exists of: (wq_completion)bond_dev->name --> (work_completion)(&(&nnw->work)->work) --> rtnl_mutex Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(rtnl_mutex); lock((work_completion)(&(&nnw->work)->work)); lock(rtnl_mutex); lock((wq_completion)bond_dev->name); *** DEADLOCK *** 1 lock held by syz-executor4/26841: stack backtrace: CPU: 1 PID: 26841 Comm: syz-executor4 Not tainted 4.18.0-next-20180823+ #46 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113 print_circular_bug.isra.34.cold.55+0x1bd/0x27d kernel/locking/lockdep.c:1222 check_prev_add kernel/locking/lockdep.c:1862 [inline] check_prevs_add kernel/locking/lockdep.c:1975 [inline] validate_chain kernel/locking/lockdep.c:2416 [inline] __lock_acquire+0x3449/0x5020 kernel/locking/lockdep.c:3412 lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x457089 Code: fd b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 00 66 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 cb b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007f2df20a5c78 EFLAGS: 00000246 ORIG_RAX: 000000000000002e RAX: ffffffffffffffda RBX: 00007f2df20a66d4 RCX: 0000000000457089 RDX: 0000000000000000 RSI: 0000000020000180 RDI: 0000000000000003 RBP: 0000000000930140 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 00000000ffffffff R13: 00000000004d40b8 R14: 00000000004c8ad8 R15: 0000000000000001 Signed-off-by: Mahesh Bandewar <maheshb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-25 04:40:11 +07:00
bond_fill_ifslave(slave, &binfo.slave);
bond_fill_ifbond(slave->bond, &binfo.master);
netdev_bonding_info_change(slave->dev, &binfo);
rtnl_unlock();
} else {
queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
}
}
void bond_queue_slave_event(struct slave *slave)
{
bonding: avoid possible dead-lock Syzkaller reported this on a slightly older kernel but it's still applicable to the current kernel - ====================================================== WARNING: possible circular locking dependency detected 4.18.0-next-20180823+ #46 Not tainted ------------------------------------------------------ syz-executor4/26841 is trying to acquire lock: 00000000dd41ef48 ((wq_completion)bond_dev->name){+.+.}, at: flush_workqueue+0x2db/0x1e10 kernel/workqueue.c:2652 but task is already holding lock: 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnl_lock net/core/rtnetlink.c:77 [inline] 00000000768ab431 (rtnl_mutex){+.+.}, at: rtnetlink_rcv_msg+0x412/0xc30 net/core/rtnetlink.c:4708 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #2 (rtnl_mutex){+.+.}: __mutex_lock_common kernel/locking/mutex.c:925 [inline] __mutex_lock+0x171/0x1700 kernel/locking/mutex.c:1073 mutex_lock_nested+0x16/0x20 kernel/locking/mutex.c:1088 rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77 bond_netdev_notify drivers/net/bonding/bond_main.c:1310 [inline] bond_netdev_notify_work+0x44/0xd0 drivers/net/bonding/bond_main.c:1320 process_one_work+0xc73/0x1aa0 kernel/workqueue.c:2153 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #1 ((work_completion)(&(&nnw->work)->work)){+.+.}: process_one_work+0xc0b/0x1aa0 kernel/workqueue.c:2129 worker_thread+0x189/0x13c0 kernel/workqueue.c:2296 kthread+0x35a/0x420 kernel/kthread.c:246 ret_from_fork+0x3a/0x50 arch/x86/entry/entry_64.S:415 -> #0 ((wq_completion)bond_dev->name){+.+.}: lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe other info that might help us debug this: Chain exists of: (wq_completion)bond_dev->name --> (work_completion)(&(&nnw->work)->work) --> rtnl_mutex Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(rtnl_mutex); lock((work_completion)(&(&nnw->work)->work)); lock(rtnl_mutex); lock((wq_completion)bond_dev->name); *** DEADLOCK *** 1 lock held by syz-executor4/26841: stack backtrace: CPU: 1 PID: 26841 Comm: syz-executor4 Not tainted 4.18.0-next-20180823+ #46 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113 print_circular_bug.isra.34.cold.55+0x1bd/0x27d kernel/locking/lockdep.c:1222 check_prev_add kernel/locking/lockdep.c:1862 [inline] check_prevs_add kernel/locking/lockdep.c:1975 [inline] validate_chain kernel/locking/lockdep.c:2416 [inline] __lock_acquire+0x3449/0x5020 kernel/locking/lockdep.c:3412 lock_acquire+0x1e4/0x4f0 kernel/locking/lockdep.c:3901 flush_workqueue+0x30a/0x1e10 kernel/workqueue.c:2655 drain_workqueue+0x2a9/0x640 kernel/workqueue.c:2820 destroy_workqueue+0xc6/0x9d0 kernel/workqueue.c:4155 __alloc_workqueue_key+0xef9/0x1190 kernel/workqueue.c:4138 bond_init+0x269/0x940 drivers/net/bonding/bond_main.c:4734 register_netdevice+0x337/0x1100 net/core/dev.c:8410 bond_newlink+0x49/0xa0 drivers/net/bonding/bond_netlink.c:453 rtnl_newlink+0xef4/0x1d50 net/core/rtnetlink.c:3099 rtnetlink_rcv_msg+0x46e/0xc30 net/core/rtnetlink.c:4711 netlink_rcv_skb+0x172/0x440 net/netlink/af_netlink.c:2454 rtnetlink_rcv+0x1c/0x20 net/core/rtnetlink.c:4729 netlink_unicast_kernel net/netlink/af_netlink.c:1317 [inline] netlink_unicast+0x5a0/0x760 net/netlink/af_netlink.c:1343 netlink_sendmsg+0xa18/0xfc0 net/netlink/af_netlink.c:1908 sock_sendmsg_nosec net/socket.c:622 [inline] sock_sendmsg+0xd5/0x120 net/socket.c:632 ___sys_sendmsg+0x7fd/0x930 net/socket.c:2115 __sys_sendmsg+0x11d/0x290 net/socket.c:2153 __do_sys_sendmsg net/socket.c:2162 [inline] __se_sys_sendmsg net/socket.c:2160 [inline] __x64_sys_sendmsg+0x78/0xb0 net/socket.c:2160 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x457089 Code: fd b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 00 66 90 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 cb b4 fb ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007f2df20a5c78 EFLAGS: 00000246 ORIG_RAX: 000000000000002e RAX: ffffffffffffffda RBX: 00007f2df20a66d4 RCX: 0000000000457089 RDX: 0000000000000000 RSI: 0000000020000180 RDI: 0000000000000003 RBP: 0000000000930140 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 00000000ffffffff R13: 00000000004d40b8 R14: 00000000004c8ad8 R15: 0000000000000001 Signed-off-by: Mahesh Bandewar <maheshb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-09-25 04:40:11 +07:00
queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
}
void bond_lower_state_changed(struct slave *slave)
{
struct netdev_lag_lower_state_info info;
info.link_up = slave->link == BOND_LINK_UP ||
slave->link == BOND_LINK_FAIL;
info.tx_enabled = bond_is_active_slave(slave);
netdev_lower_state_changed(slave->dev, &info);
}
/* enslave device <slave> to bond device <master> */
int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
struct netlink_ext_ack *extack)
{
struct bonding *bond = netdev_priv(bond_dev);
const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
struct slave *new_slave = NULL, *prev_slave;
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
struct sockaddr_storage ss;
int link_reporting;
bonding: add an option to fail when any of arp_ip_target is inaccessible Currently, we fail only when all of the ips in arp_ip_target are gone. However, in some situations we might need to fail if even one host from arp_ip_target becomes unavailable. All situations, obviously, rely on the idea that we need *completely* functional network, with all interfaces/addresses working correctly. One real world example might be: vlans on top on bond (hybrid port). If bond and vlans have ips assigned and we have their peers monitored via arp_ip_target - in case of switch misconfiguration (trunk/access port), slave driver malfunction or tagged/untagged traffic dropped on the way - we will be able to switch to another slave. Though any other configuration needs that if we need to have access to all arp_ip_targets. This patch adds this possibility by adding a new parameter - arp_all_targets (both as a module parameter and as a sysfs knob). It can be set to: 0 or any (the default) - which works exactly as it's working now - the slave is up if any of the arp_ip_targets are up. 1 or all - the slave is up if all of the arp_ip_targets are up. This parameter can be changed on the fly (via sysfs), and requires the mode to be active-backup and arp_validate to be enabled (it obeys the arp_validate config on which slaves to validate). Internally it's done through: 1) Add target_last_arp_rx[BOND_MAX_ARP_TARGETS] array to slave struct. It's an array of jiffies, meaning that slave->target_last_arp_rx[i] is the last time we've received arp from bond->params.arp_targets[i] on this slave. 2) If we successfully validate an arp from bond->params.arp_targets[i] in bond_validate_arp() - update the slave->target_last_arp_rx[i] with the current jiffies value. 3) When getting slave's last_rx via slave_last_rx(), we return the oldest time when we've received an arp from any address in bond->params.arp_targets[]. If the value of arp_all_targets == 0 - we still work the same way as before. Also, update the documentation to reflect the new parameter. v3->v4: Kill the forgotten rtnl_unlock(), rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Also, print a warning if the last arp_ip_target is removed while the arp_interval is on, but not the arp_validate. v2->v3: Use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). On bond_enslave(), use the same initialization value for target_last_arp_rx[] as is used for the default last_arp_rx, to avoid useless interface flaps. Also, instead of failing to remove the last arp_ip_target just print a warning - otherwise it might break existing scripts. v1->v2: Correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes. Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-24 16:49:34 +07:00
int res = 0, i;
if (!bond->params.use_carrier &&
slave_dev->ethtool_ops->get_link == NULL &&
slave_ops->ndo_do_ioctl == NULL) {
netdev_warn(bond_dev, "no link monitoring support for %s\n",
slave_dev->name);
}
/* already in-use? */
if (netdev_is_rx_handler_busy(slave_dev)) {
NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
netdev_err(bond_dev,
"Error: Device is in use and cannot be enslaved\n");
return -EBUSY;
}
if (bond_dev == slave_dev) {
NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
netdev_err(bond_dev, "cannot enslave bond to itself.\n");
return -EPERM;
}
/* vlan challenged mutual exclusion */
/* no need to lock since we're protected by rtnl_lock */
if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
slave_dev->name);
if (vlan_uses_dev(bond_dev)) {
NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
slave_dev->name, bond_dev->name);
return -EPERM;
} else {
netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
slave_dev->name, slave_dev->name,
bond_dev->name);
}
} else {
netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
slave_dev->name);
}
/* Old ifenslave binaries are no longer supported. These can
* be identified with moderate accuracy by the state of the slave:
* the current ifenslave will set the interface down prior to
* enslaving it; the old ifenslave will not.
*/
if (slave_dev->flags & IFF_UP) {
NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
slave_dev->name);
return -EPERM;
}
/* set bonding device ether type by slave - bonding netdevices are
* created with ether_setup, so when the slave type is not ARPHRD_ETHER
* there is a need to override some of the type dependent attribs/funcs.
*
* bond ether type mutual exclusion - don't allow slaves of dissimilar
* ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
*/
if (!bond_has_slaves(bond)) {
bonding: clean muticast addresses when device changes type Bonding device forbids slave device of different types under the same master. However, it is possible for a bonding master to change type during its lifetime. This can be either from ARPHRD_ETHER to ARPHRD_INFINIBAND or the other way arround. The change of type requires device level multicast address cleanup because device level multicast addresses depend on the device type. The patch adds a call to dev_close() before the bonding master changes type and dev_open() just after that. In the example below I enslaved an IPoIB device (ib0) under bond0. Since each bonding master starts as device of type ARPHRD_ETHER by default, a change of type occurs when ib0 is enslaved. This is how /proc/net/dev_mcast looks like without the patch 5 bond0 1 0 00ffffffff12601bffff000000000001ff96ca05 5 bond0 1 0 01005e000116 5 bond0 1 0 01005e7ffffd 5 bond0 1 0 01005e000001 5 bond0 1 0 333300000001 6 ib0 1 0 00ffffffff12601bffff000000000001ff96ca05 6 ib0 1 0 333300000001 6 ib0 1 0 01005e000001 6 ib0 1 0 01005e7ffffd 6 ib0 1 0 01005e000116 6 ib0 1 0 00ffffffff12401bffff00000000000000000001 6 ib0 1 0 00ffffffff12601bffff00000000000000000001 and this is how it looks like after the patch. 5 bond0 1 0 00ffffffff12601bffff000000000001ff96ca05 5 bond0 1 0 00ffffffff12601bffff00000000000000000001 5 bond0 1 0 00ffffffff12401bffff0000000000000ffffffd 5 bond0 1 0 00ffffffff12401bffff00000000000000000116 5 bond0 1 0 00ffffffff12401bffff00000000000000000001 6 ib0 1 0 00ffffffff12601bffff000000000001ff96ca05 6 ib0 1 0 00ffffffff12401bffff00000000000000000116 6 ib0 1 0 00ffffffff12401bffff0000000000000ffffffd 6 ib0 2 0 00ffffffff12401bffff00000000000000000001 6 ib0 2 0 00ffffffff12601bffff00000000000000000001 Signed-off-by: Moni Shoua <monis@voltaire.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2009-07-15 11:56:31 +07:00
if (bond_dev->type != slave_dev->type) {
netdev_dbg(bond_dev, "change device type from %d to %d\n",
bond_dev->type, slave_dev->type);
res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
bond_dev);
res = notifier_to_errno(res);
if (res) {
netdev_err(bond_dev, "refused to change device type\n");
return -EBUSY;
}
/* Flush unicast and multicast addresses */
dev_uc_flush(bond_dev);
dev_mc_flush(bond_dev);
bonding: clean muticast addresses when device changes type Bonding device forbids slave device of different types under the same master. However, it is possible for a bonding master to change type during its lifetime. This can be either from ARPHRD_ETHER to ARPHRD_INFINIBAND or the other way arround. The change of type requires device level multicast address cleanup because device level multicast addresses depend on the device type. The patch adds a call to dev_close() before the bonding master changes type and dev_open() just after that. In the example below I enslaved an IPoIB device (ib0) under bond0. Since each bonding master starts as device of type ARPHRD_ETHER by default, a change of type occurs when ib0 is enslaved. This is how /proc/net/dev_mcast looks like without the patch 5 bond0 1 0 00ffffffff12601bffff000000000001ff96ca05 5 bond0 1 0 01005e000116 5 bond0 1 0 01005e7ffffd 5 bond0 1 0 01005e000001 5 bond0 1 0 333300000001 6 ib0 1 0 00ffffffff12601bffff000000000001ff96ca05 6 ib0 1 0 333300000001 6 ib0 1 0 01005e000001 6 ib0 1 0 01005e7ffffd 6 ib0 1 0 01005e000116 6 ib0 1 0 00ffffffff12401bffff00000000000000000001 6 ib0 1 0 00ffffffff12601bffff00000000000000000001 and this is how it looks like after the patch. 5 bond0 1 0 00ffffffff12601bffff000000000001ff96ca05 5 bond0 1 0 00ffffffff12601bffff00000000000000000001 5 bond0 1 0 00ffffffff12401bffff0000000000000ffffffd 5 bond0 1 0 00ffffffff12401bffff00000000000000000116 5 bond0 1 0 00ffffffff12401bffff00000000000000000001 6 ib0 1 0 00ffffffff12601bffff000000000001ff96ca05 6 ib0 1 0 00ffffffff12401bffff00000000000000000116 6 ib0 1 0 00ffffffff12401bffff0000000000000ffffffd 6 ib0 2 0 00ffffffff12401bffff00000000000000000001 6 ib0 2 0 00ffffffff12601bffff00000000000000000001 Signed-off-by: Moni Shoua <monis@voltaire.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2009-07-15 11:56:31 +07:00
if (slave_dev->type != ARPHRD_ETHER)
bond_setup_by_slave(bond_dev, slave_dev);
else {
bonding: clean muticast addresses when device changes type Bonding device forbids slave device of different types under the same master. However, it is possible for a bonding master to change type during its lifetime. This can be either from ARPHRD_ETHER to ARPHRD_INFINIBAND or the other way arround. The change of type requires device level multicast address cleanup because device level multicast addresses depend on the device type. The patch adds a call to dev_close() before the bonding master changes type and dev_open() just after that. In the example below I enslaved an IPoIB device (ib0) under bond0. Since each bonding master starts as device of type ARPHRD_ETHER by default, a change of type occurs when ib0 is enslaved. This is how /proc/net/dev_mcast looks like without the patch 5 bond0 1 0 00ffffffff12601bffff000000000001ff96ca05 5 bond0 1 0 01005e000116 5 bond0 1 0 01005e7ffffd 5 bond0 1 0 01005e000001 5 bond0 1 0 333300000001 6 ib0 1 0 00ffffffff12601bffff000000000001ff96ca05 6 ib0 1 0 333300000001 6 ib0 1 0 01005e000001 6 ib0 1 0 01005e7ffffd 6 ib0 1 0 01005e000116 6 ib0 1 0 00ffffffff12401bffff00000000000000000001 6 ib0 1 0 00ffffffff12601bffff00000000000000000001 and this is how it looks like after the patch. 5 bond0 1 0 00ffffffff12601bffff000000000001ff96ca05 5 bond0 1 0 00ffffffff12601bffff00000000000000000001 5 bond0 1 0 00ffffffff12401bffff0000000000000ffffffd 5 bond0 1 0 00ffffffff12401bffff00000000000000000116 5 bond0 1 0 00ffffffff12401bffff00000000000000000001 6 ib0 1 0 00ffffffff12601bffff000000000001ff96ca05 6 ib0 1 0 00ffffffff12401bffff00000000000000000116 6 ib0 1 0 00ffffffff12401bffff0000000000000ffffffd 6 ib0 2 0 00ffffffff12401bffff00000000000000000001 6 ib0 2 0 00ffffffff12601bffff00000000000000000001 Signed-off-by: Moni Shoua <monis@voltaire.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2009-07-15 11:56:31 +07:00
ether_setup(bond_dev);
bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
}
call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
bond_dev);
bonding: clean muticast addresses when device changes type Bonding device forbids slave device of different types under the same master. However, it is possible for a bonding master to change type during its lifetime. This can be either from ARPHRD_ETHER to ARPHRD_INFINIBAND or the other way arround. The change of type requires device level multicast address cleanup because device level multicast addresses depend on the device type. The patch adds a call to dev_close() before the bonding master changes type and dev_open() just after that. In the example below I enslaved an IPoIB device (ib0) under bond0. Since each bonding master starts as device of type ARPHRD_ETHER by default, a change of type occurs when ib0 is enslaved. This is how /proc/net/dev_mcast looks like without the patch 5 bond0 1 0 00ffffffff12601bffff000000000001ff96ca05 5 bond0 1 0 01005e000116 5 bond0 1 0 01005e7ffffd 5 bond0 1 0 01005e000001 5 bond0 1 0 333300000001 6 ib0 1 0 00ffffffff12601bffff000000000001ff96ca05 6 ib0 1 0 333300000001 6 ib0 1 0 01005e000001 6 ib0 1 0 01005e7ffffd 6 ib0 1 0 01005e000116 6 ib0 1 0 00ffffffff12401bffff00000000000000000001 6 ib0 1 0 00ffffffff12601bffff00000000000000000001 and this is how it looks like after the patch. 5 bond0 1 0 00ffffffff12601bffff000000000001ff96ca05 5 bond0 1 0 00ffffffff12601bffff00000000000000000001 5 bond0 1 0 00ffffffff12401bffff0000000000000ffffffd 5 bond0 1 0 00ffffffff12401bffff00000000000000000116 5 bond0 1 0 00ffffffff12401bffff00000000000000000001 6 ib0 1 0 00ffffffff12601bffff000000000001ff96ca05 6 ib0 1 0 00ffffffff12401bffff00000000000000000116 6 ib0 1 0 00ffffffff12401bffff0000000000000ffffffd 6 ib0 2 0 00ffffffff12401bffff00000000000000000001 6 ib0 2 0 00ffffffff12601bffff00000000000000000001 Signed-off-by: Moni Shoua <monis@voltaire.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2009-07-15 11:56:31 +07:00
}
} else if (bond_dev->type != slave_dev->type) {
NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
slave_dev->name, slave_dev->type, bond_dev->type);
return -EINVAL;
}
if (slave_dev->type == ARPHRD_INFINIBAND &&
BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
slave_dev->type);
res = -EOPNOTSUPP;
goto err_undo_flags;
}
if (!slave_ops->ndo_set_mac_address ||
slave_dev->type == ARPHRD_INFINIBAND) {
netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
if (!bond_has_slaves(bond)) {
bond->params.fail_over_mac = BOND_FOM_ACTIVE;
netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
} else {
NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
res = -EOPNOTSUPP;
goto err_undo_flags;
}
}
}
call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
/* If this is the first slave, then we need to set the master's hardware
* address to be the same as the slave's.
*/
if (!bond_has_slaves(bond) &&
bond->dev->addr_assign_type == NET_ADDR_RANDOM)
bond_set_dev_addr(bond->dev, slave_dev);
new_slave = bond_alloc_slave(bond);
if (!new_slave) {
res = -ENOMEM;
goto err_undo_flags;
}
new_slave->bond = bond;
new_slave->dev = slave_dev;
/* Set the new_slave's queue_id to be zero. Queue ID mapping
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
* is set via sysfs or module option if desired.
*/
new_slave->queue_id = 0;
/* Save slave's original mtu and then set it to match the bond */
new_slave->original_mtu = slave_dev->mtu;
res = dev_set_mtu(slave_dev, bond->dev->mtu);
if (res) {
netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
goto err_free;
}
/* Save slave's original ("permanent") mac address for modes
* that need it, and for restoring it upon release, and then
* set it to the master's address
*/
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
slave_dev->addr_len);
if (!bond->params.fail_over_mac ||
BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
/* Set slave to master's mac address. The application already
* set the master's mac address to that of the first slave
*/
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
ss.ss_family = slave_dev->type;
res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
if (res) {
netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
goto err_restore_mtu;
}
}
/* set slave flag before open to prevent IPv6 addrconf */
slave_dev->flags |= IFF_SLAVE;
/* open the slave since the application closed it */
res = dev_open(slave_dev, extack);
if (res) {
netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
goto err_restore_mac;
}
slave_dev->priv_flags |= IFF_BONDING;
/* initialize slave stats */
dev_get_stats(new_slave->dev, &new_slave->slave_stats);
if (bond_is_lb(bond)) {
/* bond_alb_init_slave() must be called before all other stages since
* it might fail and we do not want to have to undo everything
*/
res = bond_alb_init_slave(bond, new_slave);
if (res)
goto err_close;
}
res = vlan_vids_add_by_dev(slave_dev, bond_dev);
if (res) {
netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
slave_dev->name);
bonding: move dev_mc_sync after master_upper_dev_link in bond_enslave Beniamino found a crash when adding vlan as slave of bond which is also the parent link: ip link add bond1 type bond ip link set bond1 up ip link add link bond1 vlan1 type vlan id 80 ip link set vlan1 master bond1 The call trace is as below: [<ffffffffa850842a>] queued_spin_lock_slowpath+0xb/0xf [<ffffffffa8515680>] _raw_spin_lock+0x20/0x30 [<ffffffffa83f6f07>] dev_mc_sync+0x37/0x80 [<ffffffffc08687dc>] vlan_dev_set_rx_mode+0x1c/0x30 [8021q] [<ffffffffa83efd2a>] __dev_set_rx_mode+0x5a/0xa0 [<ffffffffa83f7138>] dev_mc_sync_multiple+0x78/0x80 [<ffffffffc084127c>] bond_enslave+0x67c/0x1190 [bonding] [<ffffffffa8401909>] do_setlink+0x9c9/0xe50 [<ffffffffa8403bf2>] rtnl_newlink+0x522/0x880 [<ffffffffa8403ff7>] rtnetlink_rcv_msg+0xa7/0x260 [<ffffffffa8424ecb>] netlink_rcv_skb+0xab/0xc0 [<ffffffffa83fe498>] rtnetlink_rcv+0x28/0x30 [<ffffffffa8424850>] netlink_unicast+0x170/0x210 [<ffffffffa8424bf8>] netlink_sendmsg+0x308/0x420 [<ffffffffa83cc396>] sock_sendmsg+0xb6/0xf0 This is actually a dead lock caused by sync slave hwaddr from master when the master is the slave's 'slave'. This dead loop check is actually done by netdev_master_upper_dev_link. However, Commit 1f718f0f4f97 ("bonding: populate neighbour's private on enslave") moved it after dev_mc_sync. This patch is to fix it by moving dev_mc_sync after master_upper_dev_link, so that this loop check would be earlier than dev_mc_sync. It also moves if (mode == BOND_MODE_8023AD) into if (!bond_uses_primary) clause as an improvement. Note team driver also has this issue, I will fix it in another patch. Fixes: 1f718f0f4f97 ("bonding: populate neighbour's private on enslave") Reported-by: Beniamino Galvani <bgalvani@redhat.com> Signed-off-by: Xin Long <lucien.xin@gmail.com> Acked-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-03-26 00:16:46 +07:00
goto err_close;
}
prev_slave = bond_last_slave(bond);
new_slave->delay = 0;
new_slave->link_failure_count = 0;
if (bond_update_speed_duplex(new_slave) &&
bond_needs_speed_duplex(bond))
new_slave->link = BOND_LINK_DOWN;
new_slave->last_rx = jiffies -
(msecs_to_jiffies(bond->params.arp_interval) + 1);
bonding: add an option to fail when any of arp_ip_target is inaccessible Currently, we fail only when all of the ips in arp_ip_target are gone. However, in some situations we might need to fail if even one host from arp_ip_target becomes unavailable. All situations, obviously, rely on the idea that we need *completely* functional network, with all interfaces/addresses working correctly. One real world example might be: vlans on top on bond (hybrid port). If bond and vlans have ips assigned and we have their peers monitored via arp_ip_target - in case of switch misconfiguration (trunk/access port), slave driver malfunction or tagged/untagged traffic dropped on the way - we will be able to switch to another slave. Though any other configuration needs that if we need to have access to all arp_ip_targets. This patch adds this possibility by adding a new parameter - arp_all_targets (both as a module parameter and as a sysfs knob). It can be set to: 0 or any (the default) - which works exactly as it's working now - the slave is up if any of the arp_ip_targets are up. 1 or all - the slave is up if all of the arp_ip_targets are up. This parameter can be changed on the fly (via sysfs), and requires the mode to be active-backup and arp_validate to be enabled (it obeys the arp_validate config on which slaves to validate). Internally it's done through: 1) Add target_last_arp_rx[BOND_MAX_ARP_TARGETS] array to slave struct. It's an array of jiffies, meaning that slave->target_last_arp_rx[i] is the last time we've received arp from bond->params.arp_targets[i] on this slave. 2) If we successfully validate an arp from bond->params.arp_targets[i] in bond_validate_arp() - update the slave->target_last_arp_rx[i] with the current jiffies value. 3) When getting slave's last_rx via slave_last_rx(), we return the oldest time when we've received an arp from any address in bond->params.arp_targets[]. If the value of arp_all_targets == 0 - we still work the same way as before. Also, update the documentation to reflect the new parameter. v3->v4: Kill the forgotten rtnl_unlock(), rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Also, print a warning if the last arp_ip_target is removed while the arp_interval is on, but not the arp_validate. v2->v3: Use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). On bond_enslave(), use the same initialization value for target_last_arp_rx[] as is used for the default last_arp_rx, to avoid useless interface flaps. Also, instead of failing to remove the last arp_ip_target just print a warning - otherwise it might break existing scripts. v1->v2: Correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes. Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-24 16:49:34 +07:00
for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
new_slave->target_last_arp_rx[i] = new_slave->last_rx;
if (bond->params.miimon && !bond->params.use_carrier) {
link_reporting = bond_check_dev_link(bond, slave_dev, 1);
if ((link_reporting == -1) && !bond->params.arp_interval) {
/* miimon is set but a bonded network driver
* does not support ETHTOOL/MII and
* arp_interval is not set. Note: if
* use_carrier is enabled, we will never go
* here (because netif_carrier is always
* supported); thus, we don't need to change
* the messages for netif_carrier.
*/
netdev_warn(bond_dev, "MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n",
slave_dev->name);
} else if (link_reporting == -1) {
/* unable get link status using mii/ethtool */
netdev_warn(bond_dev, "can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface\n",
slave_dev->name);
}
}
/* check for initial state */
new_slave->link = BOND_LINK_NOCHANGE;
if (bond->params.miimon) {
if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
if (bond->params.updelay) {
bond_set_slave_link_state(new_slave,
BOND_LINK_BACK,
BOND_SLAVE_NOTIFY_NOW);
new_slave->delay = bond->params.updelay;
} else {
bond_set_slave_link_state(new_slave,
BOND_LINK_UP,
BOND_SLAVE_NOTIFY_NOW);
}
} else {
bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
BOND_SLAVE_NOTIFY_NOW);
}
} else if (bond->params.arp_interval) {
bond_set_slave_link_state(new_slave,
(netif_carrier_ok(slave_dev) ?
BOND_LINK_UP : BOND_LINK_DOWN),
BOND_SLAVE_NOTIFY_NOW);
} else {
bond_set_slave_link_state(new_slave, BOND_LINK_UP,
BOND_SLAVE_NOTIFY_NOW);
}
if (new_slave->link != BOND_LINK_DOWN)
new_slave->last_link_up = jiffies;
netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
new_slave->link == BOND_LINK_DOWN ? "DOWN" :
(new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
if (bond_uses_primary(bond) && bond->params.primary[0]) {
/* if there is a primary slave, remember it */
if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
rcu_assign_pointer(bond->primary_slave, new_slave);
bond->force_primary = true;
}
}
switch (BOND_MODE(bond)) {
case BOND_MODE_ACTIVEBACKUP:
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_inactive_flags(new_slave,
BOND_SLAVE_NOTIFY_NOW);
break;
case BOND_MODE_8023AD:
/* in 802.3ad mode, the internal mechanism
* will activate the slaves in the selected
* aggregator
*/
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
/* if this is the first slave */
if (!prev_slave) {
SLAVE_AD_INFO(new_slave)->id = 1;
/* Initialize AD with the number of times that the AD timer is called in 1 second
* can be called only after the mac address of the bond is set
*/
bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
} else {
SLAVE_AD_INFO(new_slave)->id =
SLAVE_AD_INFO(prev_slave)->id + 1;
}
bond_3ad_bind_slave(new_slave);
break;
case BOND_MODE_TLB:
case BOND_MODE_ALB:
bond_set_active_slave(new_slave);
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
break;
default:
netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
/* always active in trunk mode */
bond_set_active_slave(new_slave);
/* In trunking mode there is little meaning to curr_active_slave
* anyway (it holds no special properties of the bond device),
* so we can change it without calling change_active_interface()
*/
if (!rcu_access_pointer(bond->curr_active_slave) &&
new_slave->link == BOND_LINK_UP)
bonding: initial RCU conversion This patch does the initial bonding conversion to RCU. After it the following modes are protected by RCU alone: roundrobin, active-backup, broadcast and xor. Modes ALB/TLB and 3ad still acquire bond->lock for reading, and will be dealt with later. curr_active_slave needs to be dereferenced via rcu in the converted modes because the only thing protecting the slave after this patch is rcu_read_lock, so we need the proper barrier for weakly ordered archs and to make sure we don't have stale pointer. It's not tagged with __rcu yet because there's still work to be done to remove the curr_slave_lock, so sparse will complain when rcu_assign_pointer and rcu_dereference are used, but the alternative to use rcu_dereference_protected would've created much bigger code churn which is more difficult to test and review. That will be converted in time. 1. Active-backup mode 1.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.55% in bonding, system spent 0.29% CPU in bonding - new bonding: iperf spent 0.29% in bonding, system spent 0.15% CPU in bonding 1.2. Bandwidth measurements - old bonding: 16.1 gbps consistently - new bonding: 17.5 gbps consistently 2. Round-robin mode 2.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.51% in bonding, system spent 0.24% CPU in bonding - new bonding: iperf spent 0.16% in bonding, system spent 0.11% CPU in bonding 2.2 Bandwidth measurements - old bonding: 8 gbps (variable due to packet reorderings) - new bonding: 10 gbps (variable due to packet reorderings) Of course the latency has improved in all converted modes, and moreover while doing enslave/release (since it doesn't affect tx anymore). Also I've stress tested all modes doing enslave/release in a loop while transmitting traffic. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 21:54:51 +07:00
rcu_assign_pointer(bond->curr_active_slave, new_slave);
break;
} /* switch(bond_mode) */
#ifdef CONFIG_NET_POLL_CONTROLLER
if (bond->dev->npinfo) {
if (slave_enable_netpoll(new_slave)) {
netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
res = -EBUSY;
goto err_detach;
}
}
#endif
if (!(bond_dev->features & NETIF_F_LRO))
dev_disable_lro(slave_dev);
res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
new_slave);
if (res) {
netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
goto err_detach;
}
res = bond_master_upper_dev_link(bond, new_slave, extack);
if (res) {
netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
goto err_unregister;
}
res = bond_sysfs_slave_add(new_slave);
if (res) {
netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
goto err_upper_unlink;
}
bonding: avoid lockdep confusion in bond_get_stats() syzbot found that the following sequence produces a LOCKDEP splat [1] ip link add bond10 type bond ip link add bond11 type bond ip link set bond11 master bond10 To fix this, we can use the already provided nest_level. This patch also provides correct nesting for dev->addr_list_lock [1] WARNING: possible recursive locking detected 4.18.0-rc6+ #167 Not tainted -------------------------------------------- syz-executor751/4439 is trying to acquire lock: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 but task is already holding lock: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 other info that might help us debug this: Possible unsafe locking scenario: CPU0 ---- lock(&(&bond->stats_lock)->rlock); lock(&(&bond->stats_lock)->rlock); *** DEADLOCK *** May be due to missing lock nesting notation 3 locks held by syz-executor751/4439: #0: (____ptrval____) (rtnl_mutex){+.+.}, at: rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77 #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 #2: (____ptrval____) (rcu_read_lock){....}, at: bond_get_stats+0x0/0x560 include/linux/compiler.h:215 stack backtrace: CPU: 0 PID: 4439 Comm: syz-executor751 Not tainted 4.18.0-rc6+ #167 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113 print_deadlock_bug kernel/locking/lockdep.c:1765 [inline] check_deadlock kernel/locking/lockdep.c:1809 [inline] validate_chain kernel/locking/lockdep.c:2405 [inline] __lock_acquire.cold.64+0x1fb/0x486 kernel/locking/lockdep.c:3435 lock_acquire+0x1e4/0x540 kernel/locking/lockdep.c:3924 __raw_spin_lock include/linux/spinlock_api_smp.h:142 [inline] _raw_spin_lock+0x2a/0x40 kernel/locking/spinlock.c:144 spin_lock include/linux/spinlock.h:310 [inline] bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 dev_get_stats+0x10f/0x470 net/core/dev.c:8316 bond_get_stats+0x232/0x560 drivers/net/bonding/bond_main.c:3432 dev_get_stats+0x10f/0x470 net/core/dev.c:8316 rtnl_fill_stats+0x4d/0xac0 net/core/rtnetlink.c:1169 rtnl_fill_ifinfo+0x1aa6/0x3fb0 net/core/rtnetlink.c:1611 rtmsg_ifinfo_build_skb+0xc8/0x190 net/core/rtnetlink.c:3268 rtmsg_ifinfo_event.part.30+0x45/0xe0 net/core/rtnetlink.c:3300 rtmsg_ifinfo_event net/core/rtnetlink.c:3297 [inline] rtnetlink_event+0x144/0x170 net/core/rtnetlink.c:4716 notifier_call_chain+0x180/0x390 kernel/notifier.c:93 __raw_notifier_call_chain kernel/notifier.c:394 [inline] raw_notifier_call_chain+0x2d/0x40 kernel/notifier.c:401 call_netdevice_notifiers_info+0x3f/0x90 net/core/dev.c:1735 call_netdevice_notifiers net/core/dev.c:1753 [inline] netdev_features_change net/core/dev.c:1321 [inline] netdev_change_features+0xb3/0x110 net/core/dev.c:7759 bond_compute_features.isra.47+0x585/0xa50 drivers/net/bonding/bond_main.c:1120 bond_enslave+0x1b25/0x5da0 drivers/net/bonding/bond_main.c:1755 bond_do_ioctl+0x7cb/0xae0 drivers/net/bonding/bond_main.c:3528 dev_ifsioc+0x43c/0xb30 net/core/dev_ioctl.c:327 dev_ioctl+0x1b5/0xcc0 net/core/dev_ioctl.c:493 sock_do_ioctl+0x1d3/0x3e0 net/socket.c:992 sock_ioctl+0x30d/0x680 net/socket.c:1093 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:500 [inline] do_vfs_ioctl+0x1de/0x1720 fs/ioctl.c:684 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701 __do_sys_ioctl fs/ioctl.c:708 [inline] __se_sys_ioctl fs/ioctl.c:706 [inline] __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x440859 Code: e8 2c af 02 00 48 83 c4 18 c3 0f 1f 80 00 00 00 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 3b 10 fc ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007ffc51a92878 EFLAGS: 00000213 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 0000000000440859 RDX: 0000000020000040 RSI: 0000000000008990 RDI: 0000000000000003 RBP: 0000000000000000 R08: 00000000004002c8 R09: 00000000004002c8 R10: 00000000022d5880 R11: 0000000000000213 R12: 0000000000007390 R13: 0000000000401db0 R14: 0000000000000000 R15: 0000000000000000 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Jay Vosburgh <j.vosburgh@gmail.com> Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-07-31 20:30:54 +07:00
bond->nest_level = dev_get_nest_level(bond_dev) + 1;
bonding: move dev_mc_sync after master_upper_dev_link in bond_enslave Beniamino found a crash when adding vlan as slave of bond which is also the parent link: ip link add bond1 type bond ip link set bond1 up ip link add link bond1 vlan1 type vlan id 80 ip link set vlan1 master bond1 The call trace is as below: [<ffffffffa850842a>] queued_spin_lock_slowpath+0xb/0xf [<ffffffffa8515680>] _raw_spin_lock+0x20/0x30 [<ffffffffa83f6f07>] dev_mc_sync+0x37/0x80 [<ffffffffc08687dc>] vlan_dev_set_rx_mode+0x1c/0x30 [8021q] [<ffffffffa83efd2a>] __dev_set_rx_mode+0x5a/0xa0 [<ffffffffa83f7138>] dev_mc_sync_multiple+0x78/0x80 [<ffffffffc084127c>] bond_enslave+0x67c/0x1190 [bonding] [<ffffffffa8401909>] do_setlink+0x9c9/0xe50 [<ffffffffa8403bf2>] rtnl_newlink+0x522/0x880 [<ffffffffa8403ff7>] rtnetlink_rcv_msg+0xa7/0x260 [<ffffffffa8424ecb>] netlink_rcv_skb+0xab/0xc0 [<ffffffffa83fe498>] rtnetlink_rcv+0x28/0x30 [<ffffffffa8424850>] netlink_unicast+0x170/0x210 [<ffffffffa8424bf8>] netlink_sendmsg+0x308/0x420 [<ffffffffa83cc396>] sock_sendmsg+0xb6/0xf0 This is actually a dead lock caused by sync slave hwaddr from master when the master is the slave's 'slave'. This dead loop check is actually done by netdev_master_upper_dev_link. However, Commit 1f718f0f4f97 ("bonding: populate neighbour's private on enslave") moved it after dev_mc_sync. This patch is to fix it by moving dev_mc_sync after master_upper_dev_link, so that this loop check would be earlier than dev_mc_sync. It also moves if (mode == BOND_MODE_8023AD) into if (!bond_uses_primary) clause as an improvement. Note team driver also has this issue, I will fix it in another patch. Fixes: 1f718f0f4f97 ("bonding: populate neighbour's private on enslave") Reported-by: Beniamino Galvani <bgalvani@redhat.com> Signed-off-by: Xin Long <lucien.xin@gmail.com> Acked-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-03-26 00:16:46 +07:00
/* If the mode uses primary, then the following is handled by
* bond_change_active_slave().
*/
if (!bond_uses_primary(bond)) {
/* set promiscuity level to new slave */
if (bond_dev->flags & IFF_PROMISC) {
res = dev_set_promiscuity(slave_dev, 1);
if (res)
goto err_sysfs_del;
}
/* set allmulti level to new slave */
if (bond_dev->flags & IFF_ALLMULTI) {
res = dev_set_allmulti(slave_dev, 1);
if (res) {
if (bond_dev->flags & IFF_PROMISC)
dev_set_promiscuity(slave_dev, -1);
bonding: move dev_mc_sync after master_upper_dev_link in bond_enslave Beniamino found a crash when adding vlan as slave of bond which is also the parent link: ip link add bond1 type bond ip link set bond1 up ip link add link bond1 vlan1 type vlan id 80 ip link set vlan1 master bond1 The call trace is as below: [<ffffffffa850842a>] queued_spin_lock_slowpath+0xb/0xf [<ffffffffa8515680>] _raw_spin_lock+0x20/0x30 [<ffffffffa83f6f07>] dev_mc_sync+0x37/0x80 [<ffffffffc08687dc>] vlan_dev_set_rx_mode+0x1c/0x30 [8021q] [<ffffffffa83efd2a>] __dev_set_rx_mode+0x5a/0xa0 [<ffffffffa83f7138>] dev_mc_sync_multiple+0x78/0x80 [<ffffffffc084127c>] bond_enslave+0x67c/0x1190 [bonding] [<ffffffffa8401909>] do_setlink+0x9c9/0xe50 [<ffffffffa8403bf2>] rtnl_newlink+0x522/0x880 [<ffffffffa8403ff7>] rtnetlink_rcv_msg+0xa7/0x260 [<ffffffffa8424ecb>] netlink_rcv_skb+0xab/0xc0 [<ffffffffa83fe498>] rtnetlink_rcv+0x28/0x30 [<ffffffffa8424850>] netlink_unicast+0x170/0x210 [<ffffffffa8424bf8>] netlink_sendmsg+0x308/0x420 [<ffffffffa83cc396>] sock_sendmsg+0xb6/0xf0 This is actually a dead lock caused by sync slave hwaddr from master when the master is the slave's 'slave'. This dead loop check is actually done by netdev_master_upper_dev_link. However, Commit 1f718f0f4f97 ("bonding: populate neighbour's private on enslave") moved it after dev_mc_sync. This patch is to fix it by moving dev_mc_sync after master_upper_dev_link, so that this loop check would be earlier than dev_mc_sync. It also moves if (mode == BOND_MODE_8023AD) into if (!bond_uses_primary) clause as an improvement. Note team driver also has this issue, I will fix it in another patch. Fixes: 1f718f0f4f97 ("bonding: populate neighbour's private on enslave") Reported-by: Beniamino Galvani <bgalvani@redhat.com> Signed-off-by: Xin Long <lucien.xin@gmail.com> Acked-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-03-26 00:16:46 +07:00
goto err_sysfs_del;
}
bonding: move dev_mc_sync after master_upper_dev_link in bond_enslave Beniamino found a crash when adding vlan as slave of bond which is also the parent link: ip link add bond1 type bond ip link set bond1 up ip link add link bond1 vlan1 type vlan id 80 ip link set vlan1 master bond1 The call trace is as below: [<ffffffffa850842a>] queued_spin_lock_slowpath+0xb/0xf [<ffffffffa8515680>] _raw_spin_lock+0x20/0x30 [<ffffffffa83f6f07>] dev_mc_sync+0x37/0x80 [<ffffffffc08687dc>] vlan_dev_set_rx_mode+0x1c/0x30 [8021q] [<ffffffffa83efd2a>] __dev_set_rx_mode+0x5a/0xa0 [<ffffffffa83f7138>] dev_mc_sync_multiple+0x78/0x80 [<ffffffffc084127c>] bond_enslave+0x67c/0x1190 [bonding] [<ffffffffa8401909>] do_setlink+0x9c9/0xe50 [<ffffffffa8403bf2>] rtnl_newlink+0x522/0x880 [<ffffffffa8403ff7>] rtnetlink_rcv_msg+0xa7/0x260 [<ffffffffa8424ecb>] netlink_rcv_skb+0xab/0xc0 [<ffffffffa83fe498>] rtnetlink_rcv+0x28/0x30 [<ffffffffa8424850>] netlink_unicast+0x170/0x210 [<ffffffffa8424bf8>] netlink_sendmsg+0x308/0x420 [<ffffffffa83cc396>] sock_sendmsg+0xb6/0xf0 This is actually a dead lock caused by sync slave hwaddr from master when the master is the slave's 'slave'. This dead loop check is actually done by netdev_master_upper_dev_link. However, Commit 1f718f0f4f97 ("bonding: populate neighbour's private on enslave") moved it after dev_mc_sync. This patch is to fix it by moving dev_mc_sync after master_upper_dev_link, so that this loop check would be earlier than dev_mc_sync. It also moves if (mode == BOND_MODE_8023AD) into if (!bond_uses_primary) clause as an improvement. Note team driver also has this issue, I will fix it in another patch. Fixes: 1f718f0f4f97 ("bonding: populate neighbour's private on enslave") Reported-by: Beniamino Galvani <bgalvani@redhat.com> Signed-off-by: Xin Long <lucien.xin@gmail.com> Acked-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-03-26 00:16:46 +07:00
}
netif_addr_lock_bh(bond_dev);
dev_mc_sync_multiple(slave_dev, bond_dev);
dev_uc_sync_multiple(slave_dev, bond_dev);
netif_addr_unlock_bh(bond_dev);
if (BOND_MODE(bond) == BOND_MODE_8023AD) {
/* add lacpdu mc addr to mc list */
u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
dev_mc_add(slave_dev, lacpdu_multicast);
}
}
bond->slave_cnt++;
bond_compute_features(bond);
bond_set_carrier(bond);
if (bond_uses_primary(bond)) {
block_netpoll_tx();
bond_select_active_slave(bond);
unblock_netpoll_tx();
}
if (bond_mode_can_use_xmit_hash(bond))
bond_update_slave_arr(bond, NULL);
netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
slave_dev->name,
bond_is_active_slave(new_slave) ? "an active" : "a backup",
new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
/* enslave is successful */
bond_queue_slave_event(new_slave);
return 0;
/* Undo stages on error */
bonding: move dev_mc_sync after master_upper_dev_link in bond_enslave Beniamino found a crash when adding vlan as slave of bond which is also the parent link: ip link add bond1 type bond ip link set bond1 up ip link add link bond1 vlan1 type vlan id 80 ip link set vlan1 master bond1 The call trace is as below: [<ffffffffa850842a>] queued_spin_lock_slowpath+0xb/0xf [<ffffffffa8515680>] _raw_spin_lock+0x20/0x30 [<ffffffffa83f6f07>] dev_mc_sync+0x37/0x80 [<ffffffffc08687dc>] vlan_dev_set_rx_mode+0x1c/0x30 [8021q] [<ffffffffa83efd2a>] __dev_set_rx_mode+0x5a/0xa0 [<ffffffffa83f7138>] dev_mc_sync_multiple+0x78/0x80 [<ffffffffc084127c>] bond_enslave+0x67c/0x1190 [bonding] [<ffffffffa8401909>] do_setlink+0x9c9/0xe50 [<ffffffffa8403bf2>] rtnl_newlink+0x522/0x880 [<ffffffffa8403ff7>] rtnetlink_rcv_msg+0xa7/0x260 [<ffffffffa8424ecb>] netlink_rcv_skb+0xab/0xc0 [<ffffffffa83fe498>] rtnetlink_rcv+0x28/0x30 [<ffffffffa8424850>] netlink_unicast+0x170/0x210 [<ffffffffa8424bf8>] netlink_sendmsg+0x308/0x420 [<ffffffffa83cc396>] sock_sendmsg+0xb6/0xf0 This is actually a dead lock caused by sync slave hwaddr from master when the master is the slave's 'slave'. This dead loop check is actually done by netdev_master_upper_dev_link. However, Commit 1f718f0f4f97 ("bonding: populate neighbour's private on enslave") moved it after dev_mc_sync. This patch is to fix it by moving dev_mc_sync after master_upper_dev_link, so that this loop check would be earlier than dev_mc_sync. It also moves if (mode == BOND_MODE_8023AD) into if (!bond_uses_primary) clause as an improvement. Note team driver also has this issue, I will fix it in another patch. Fixes: 1f718f0f4f97 ("bonding: populate neighbour's private on enslave") Reported-by: Beniamino Galvani <bgalvani@redhat.com> Signed-off-by: Xin Long <lucien.xin@gmail.com> Acked-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-03-26 00:16:46 +07:00
err_sysfs_del:
bond_sysfs_slave_del(new_slave);
err_upper_unlink:
bond_upper_dev_unlink(bond, new_slave);
err_unregister:
netdev_rx_handler_unregister(slave_dev);
err_detach:
vlan_vids_del_by_dev(slave_dev, bond_dev);
if (rcu_access_pointer(bond->primary_slave) == new_slave)
RCU_INIT_POINTER(bond->primary_slave, NULL);
if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
block_netpoll_tx();
bond_change_active_slave(bond, NULL);
bond_select_active_slave(bond);
unblock_netpoll_tx();
}
/* either primary_slave or curr_active_slave might've changed */
synchronize_rcu();
slave_disable_netpoll(new_slave);
err_close:
slave_dev->priv_flags &= ~IFF_BONDING;
dev_close(slave_dev);
err_restore_mac:
slave_dev->flags &= ~IFF_SLAVE;
if (!bond->params.fail_over_mac ||
BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
/* XXX TODO - fom follow mode needs to change master's
* MAC if this slave's MAC is in use by the bond, or at
* least print a warning.
*/
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
new_slave->dev->addr_len);
ss.ss_family = slave_dev->type;
dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
}
err_restore_mtu:
dev_set_mtu(slave_dev, new_slave->original_mtu);
err_free:
bond_free_slave(new_slave);
err_undo_flags:
/* Enslave of first slave has failed and we need to fix master's mac */
bonding: correctly handle bonding type change on enslave failure If the bond is enslaving a device with different type it will be setup by it, but if after being setup the enslave fails the bond doesn't switch back its type and also keeps pointers to foreign structures that can be long gone. Thus revert back any type changes if the enslave failed and the bond had to change its type. Example: Before patch: $ echo lo > bond0/bonding/slaves -bash: echo: write error: Cannot assign requested address $ ip l sh bond0 20: bond0: <BROADCAST,MULTICAST,MASTER> mtu 1500 qdisc noop state DOWN mode DEFAULT group default link/loopback 16:54:78:34:bd:41 brd 00:00:00:00:00:00 $ echo +eth1 > bond0/bonding/slaves $ ip l sh bond0 20: bond0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode DEFAULT group default qlen 1000 link/ether 52:54:00:3f:47:69 brd ff:ff:ff:ff:ff:ff (notice the MASTER flag is gone) After patch: $ echo lo > bond0/bonding/slaves -bash: echo: write error: Cannot assign requested address $ ip l sh bond0 21: bond0: <BROADCAST,MULTICAST,MASTER> mtu 1500 qdisc noop state DOWN mode DEFAULT group default qlen 1000 link/ether 6e:66:94:f6:07:fc brd ff:ff:ff:ff:ff:ff $ echo +eth1 > bond0/bonding/slaves $ ip l sh bond0 21: bond0: <BROADCAST,MULTICAST,MASTER> mtu 1500 qdisc noop state DOWN mode DEFAULT group default qlen 1000 link/ether 52:54:00:3f:47:69 brd ff:ff:ff:ff:ff:ff Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com> Fixes: e36b9d16c6a6 ("bonding: clean muticast addresses when device changes type") Signed-off-by: David S. Miller <davem@davemloft.net>
2015-07-16 03:57:01 +07:00
if (!bond_has_slaves(bond)) {
if (ether_addr_equal_64bits(bond_dev->dev_addr,
slave_dev->dev_addr))
eth_hw_addr_random(bond_dev);
if (bond_dev->type != ARPHRD_ETHER) {
dev_close(bond_dev);
bonding: correctly handle bonding type change on enslave failure If the bond is enslaving a device with different type it will be setup by it, but if after being setup the enslave fails the bond doesn't switch back its type and also keeps pointers to foreign structures that can be long gone. Thus revert back any type changes if the enslave failed and the bond had to change its type. Example: Before patch: $ echo lo > bond0/bonding/slaves -bash: echo: write error: Cannot assign requested address $ ip l sh bond0 20: bond0: <BROADCAST,MULTICAST,MASTER> mtu 1500 qdisc noop state DOWN mode DEFAULT group default link/loopback 16:54:78:34:bd:41 brd 00:00:00:00:00:00 $ echo +eth1 > bond0/bonding/slaves $ ip l sh bond0 20: bond0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode DEFAULT group default qlen 1000 link/ether 52:54:00:3f:47:69 brd ff:ff:ff:ff:ff:ff (notice the MASTER flag is gone) After patch: $ echo lo > bond0/bonding/slaves -bash: echo: write error: Cannot assign requested address $ ip l sh bond0 21: bond0: <BROADCAST,MULTICAST,MASTER> mtu 1500 qdisc noop state DOWN mode DEFAULT group default qlen 1000 link/ether 6e:66:94:f6:07:fc brd ff:ff:ff:ff:ff:ff $ echo +eth1 > bond0/bonding/slaves $ ip l sh bond0 21: bond0: <BROADCAST,MULTICAST,MASTER> mtu 1500 qdisc noop state DOWN mode DEFAULT group default qlen 1000 link/ether 52:54:00:3f:47:69 brd ff:ff:ff:ff:ff:ff Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com> Fixes: e36b9d16c6a6 ("bonding: clean muticast addresses when device changes type") Signed-off-by: David S. Miller <davem@davemloft.net>
2015-07-16 03:57:01 +07:00
ether_setup(bond_dev);
bond_dev->flags |= IFF_MASTER;
bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
}
}
return res;
}
/* Try to release the slave device <slave> from the bond device <master>
* It is legal to access curr_active_slave without a lock because all the function
* is RTNL-locked. If "all" is true it means that the function is being called
* while destroying a bond interface and all slaves are being released.
*
* The rules for slave state should be:
* for Active/Backup:
* Active stays on all backups go down
* for Bonded connections:
* The first up interface should be left on and all others downed.
*/
static int __bond_release_one(struct net_device *bond_dev,
struct net_device *slave_dev,
bool all, bool unregister)
{
struct bonding *bond = netdev_priv(bond_dev);
struct slave *slave, *oldcurrent;
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
struct sockaddr_storage ss;
bonding: Fix broken promiscuity reference counting issue Recently grabbed this report: https://bugzilla.redhat.com/show_bug.cgi?id=1005567 Of an issue in which the bonding driver, with an attached vlan encountered the following errors when bond0 was taken down and back up: dummy1: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken. The error occurs because, during __bond_release_one, if we release our last slave, we take on a random mac address and issue a NETDEV_CHANGEADDR notification. With an attached vlan, the vlan may see that the vlan and bond mac address were in sync, but no longer are. This triggers a call to dev_uc_add and dev_set_rx_mode, which enables IFF_PROMISC on the bond device. Then, when we complete __bond_release_one, we use the current state of the bond flags to determine if we should decrement the promiscuity of the releasing slave. But since the bond changed promiscuity state during the release operation, we incorrectly decrement the slave promisc count when it wasn't in promiscuous mode to begin with, causing the above error Fix is pretty simple, just cache the bonding flags at the start of the function and use those when determining the need to set promiscuity. This is also needed for the ALLMULTI flag CC: Jay Vosburgh <fubar@us.ibm.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Mark Wu <wudxw@linux.vnet.ibm.com> CC: "David S. Miller" <davem@davemloft.net> Reported-by: Mark Wu <wudxw@linux.vnet.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-09-27 23:22:15 +07:00
int old_flags = bond_dev->flags;
netdev_features_t old_features = bond_dev->features;
/* slave is not a slave or master is not master of this slave */
if (!(slave_dev->flags & IFF_SLAVE) ||
!netdev_has_upper_dev(slave_dev, bond_dev)) {
netdev_dbg(bond_dev, "cannot release %s\n",
slave_dev->name);
return -EINVAL;
}
block_netpoll_tx();
slave = bond_get_slave_by_dev(bond, slave_dev);
if (!slave) {
/* not a slave of this bond */
netdev_info(bond_dev, "%s not enslaved\n",
slave_dev->name);
unblock_netpoll_tx();
return -EINVAL;
}
bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
bond_sysfs_slave_del(slave);
/* recompute stats just before removing the slave */
bond_get_stats(bond->dev, &bond->bond_stats);
bond_upper_dev_unlink(bond, slave);
/* unregister rx_handler early so bond_handle_frame wouldn't be called
* for this slave anymore.
*/
netdev_rx_handler_unregister(slave_dev);
if (BOND_MODE(bond) == BOND_MODE_8023AD)
bond_3ad_unbind_slave(slave);
if (bond_mode_can_use_xmit_hash(bond))
bond_update_slave_arr(bond, slave);
netdev_info(bond_dev, "Releasing %s interface %s\n",
bond_is_active_slave(slave) ? "active" : "backup",
slave_dev->name);
oldcurrent = rcu_access_pointer(bond->curr_active_slave);
RCU_INIT_POINTER(bond->current_arp_slave, NULL);
if (!all && (!bond->params.fail_over_mac ||
BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
bond_has_slaves(bond))
netdev_warn(bond_dev, "the permanent HWaddr of %s - %pM - is still in use by %s - set the HWaddr of %s to a different address to avoid conflicts\n",
slave_dev->name, slave->perm_hwaddr,
bond_dev->name, slave_dev->name);
}
if (rtnl_dereference(bond->primary_slave) == slave)
RCU_INIT_POINTER(bond->primary_slave, NULL);
if (oldcurrent == slave)
bond_change_active_slave(bond, NULL);
if (bond_is_lb(bond)) {
/* Must be called only after the slave has been
* detached from the list and the curr_active_slave
* has been cleared (if our_slave == old_current),
* but before a new active slave is selected.
*/
bond_alb_deinit_slave(bond, slave);
}
if (all) {
RCU_INIT_POINTER(bond->curr_active_slave, NULL);
} else if (oldcurrent == slave) {
/* Note that we hold RTNL over this sequence, so there
* is no concern that another slave add/remove event
* will interfere.
*/
bond_select_active_slave(bond);
}
if (!bond_has_slaves(bond)) {
bond_set_carrier(bond);
eth_hw_addr_random(bond_dev);
}
unblock_netpoll_tx();
bonding: initial RCU conversion This patch does the initial bonding conversion to RCU. After it the following modes are protected by RCU alone: roundrobin, active-backup, broadcast and xor. Modes ALB/TLB and 3ad still acquire bond->lock for reading, and will be dealt with later. curr_active_slave needs to be dereferenced via rcu in the converted modes because the only thing protecting the slave after this patch is rcu_read_lock, so we need the proper barrier for weakly ordered archs and to make sure we don't have stale pointer. It's not tagged with __rcu yet because there's still work to be done to remove the curr_slave_lock, so sparse will complain when rcu_assign_pointer and rcu_dereference are used, but the alternative to use rcu_dereference_protected would've created much bigger code churn which is more difficult to test and review. That will be converted in time. 1. Active-backup mode 1.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.55% in bonding, system spent 0.29% CPU in bonding - new bonding: iperf spent 0.29% in bonding, system spent 0.15% CPU in bonding 1.2. Bandwidth measurements - old bonding: 16.1 gbps consistently - new bonding: 17.5 gbps consistently 2. Round-robin mode 2.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.51% in bonding, system spent 0.24% CPU in bonding - new bonding: iperf spent 0.16% in bonding, system spent 0.11% CPU in bonding 2.2 Bandwidth measurements - old bonding: 8 gbps (variable due to packet reorderings) - new bonding: 10 gbps (variable due to packet reorderings) Of course the latency has improved in all converted modes, and moreover while doing enslave/release (since it doesn't affect tx anymore). Also I've stress tested all modes doing enslave/release in a loop while transmitting traffic. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 21:54:51 +07:00
synchronize_rcu();
bond->slave_cnt--;
if (!bond_has_slaves(bond)) {
call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
}
bond_compute_features(bond);
if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
(old_features & NETIF_F_VLAN_CHALLENGED))
netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
slave_dev->name, bond_dev->name);
vlan_vids_del_by_dev(slave_dev, bond_dev);
/* If the mode uses primary, then this case was handled above by
* bond_change_active_slave(..., NULL)
*/
if (!bond_uses_primary(bond)) {
bonding: Fix broken promiscuity reference counting issue Recently grabbed this report: https://bugzilla.redhat.com/show_bug.cgi?id=1005567 Of an issue in which the bonding driver, with an attached vlan encountered the following errors when bond0 was taken down and back up: dummy1: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken. The error occurs because, during __bond_release_one, if we release our last slave, we take on a random mac address and issue a NETDEV_CHANGEADDR notification. With an attached vlan, the vlan may see that the vlan and bond mac address were in sync, but no longer are. This triggers a call to dev_uc_add and dev_set_rx_mode, which enables IFF_PROMISC on the bond device. Then, when we complete __bond_release_one, we use the current state of the bond flags to determine if we should decrement the promiscuity of the releasing slave. But since the bond changed promiscuity state during the release operation, we incorrectly decrement the slave promisc count when it wasn't in promiscuous mode to begin with, causing the above error Fix is pretty simple, just cache the bonding flags at the start of the function and use those when determining the need to set promiscuity. This is also needed for the ALLMULTI flag CC: Jay Vosburgh <fubar@us.ibm.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Mark Wu <wudxw@linux.vnet.ibm.com> CC: "David S. Miller" <davem@davemloft.net> Reported-by: Mark Wu <wudxw@linux.vnet.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-09-27 23:22:15 +07:00
/* unset promiscuity level from slave
* NOTE: The NETDEV_CHANGEADDR call above may change the value
* of the IFF_PROMISC flag in the bond_dev, but we need the
* value of that flag before that change, as that was the value
* when this slave was attached, so we cache at the start of the
* function and use it here. Same goes for ALLMULTI below
*/
if (old_flags & IFF_PROMISC)
dev_set_promiscuity(slave_dev, -1);
/* unset allmulti level from slave */
bonding: Fix broken promiscuity reference counting issue Recently grabbed this report: https://bugzilla.redhat.com/show_bug.cgi?id=1005567 Of an issue in which the bonding driver, with an attached vlan encountered the following errors when bond0 was taken down and back up: dummy1: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken. The error occurs because, during __bond_release_one, if we release our last slave, we take on a random mac address and issue a NETDEV_CHANGEADDR notification. With an attached vlan, the vlan may see that the vlan and bond mac address were in sync, but no longer are. This triggers a call to dev_uc_add and dev_set_rx_mode, which enables IFF_PROMISC on the bond device. Then, when we complete __bond_release_one, we use the current state of the bond flags to determine if we should decrement the promiscuity of the releasing slave. But since the bond changed promiscuity state during the release operation, we incorrectly decrement the slave promisc count when it wasn't in promiscuous mode to begin with, causing the above error Fix is pretty simple, just cache the bonding flags at the start of the function and use those when determining the need to set promiscuity. This is also needed for the ALLMULTI flag CC: Jay Vosburgh <fubar@us.ibm.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Mark Wu <wudxw@linux.vnet.ibm.com> CC: "David S. Miller" <davem@davemloft.net> Reported-by: Mark Wu <wudxw@linux.vnet.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-09-27 23:22:15 +07:00
if (old_flags & IFF_ALLMULTI)
dev_set_allmulti(slave_dev, -1);
bond_hw_addr_flush(bond_dev, slave_dev);
}
slave_disable_netpoll(slave);
/* close slave before restoring its mac address */
dev_close(slave_dev);
if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
/* restore original ("permanent") mac address */
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
slave->dev->addr_len);
ss.ss_family = slave_dev->type;
dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
}
if (unregister)
__dev_set_mtu(slave_dev, slave->original_mtu);
else
dev_set_mtu(slave_dev, slave->original_mtu);
slave_dev->priv_flags &= ~IFF_BONDING;
bond_free_slave(slave);
return 0;
}
/* A wrapper used because of ndo_del_link */
int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
{
return __bond_release_one(bond_dev, slave_dev, false, false);
}
/* First release a slave and then destroy the bond if no more slaves are left.
* Must be under rtnl_lock when this function is called.
*/
static int bond_release_and_destroy(struct net_device *bond_dev,
struct net_device *slave_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
int ret;
ret = __bond_release_one(bond_dev, slave_dev, false, true);
if (ret == 0 && !bond_has_slaves(bond)) {
bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
netdev_info(bond_dev, "Destroying bond %s\n",
bond_dev->name);
bonding: fix destruction of bond with devices different from arphrd_ether When the bonding is being unloaded and the netdevice notifier is unregistered it executes NETDEV_UNREGISTER for each device which should remove the bond's proc entry but if the device enslaved is not of ARPHRD_ETHER type and is in front of the bonding, it may execute bond_release_and_destroy() first which would release the last slave and destroy the bond device leaving the proc entry and thus we will get the following error (with dynamic debug on for bond_netdev_event to see the events order): [ 908.963051] eql: event: 9 [ 908.963052] eql: IFF_SLAVE [ 908.963054] eql: event: 2 [ 908.963056] eql: IFF_SLAVE [ 908.963058] eql: event: 6 [ 908.963059] eql: IFF_SLAVE [ 908.963110] bond0: Releasing active interface eql [ 908.976168] bond0: Destroying bond bond0 [ 908.976266] bond0 (unregistering): Released all slaves [ 908.984097] ------------[ cut here ]------------ [ 908.984107] WARNING: CPU: 0 PID: 1787 at fs/proc/generic.c:575 remove_proc_entry+0x112/0x160() [ 908.984110] remove_proc_entry: removing non-empty directory 'net/bonding', leaking at least 'bond0' [ 908.984111] Modules linked in: bonding(-) eql(O) 9p nfsd auth_rpcgss oid_registry nfs_acl nfs lockd grace fscache sunrpc crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel ppdev qxl drm_kms_helper snd_hda_codec_generic aesni_intel ttm aes_x86_64 glue_helper pcspkr lrw gf128mul ablk_helper cryptd snd_hda_intel virtio_console snd_hda_codec psmouse serio_raw snd_hwdep snd_hda_core 9pnet_virtio 9pnet evdev joydev drm virtio_balloon snd_pcm snd_timer snd soundcore i2c_piix4 i2c_core pvpanic acpi_cpufreq parport_pc parport processor thermal_sys button autofs4 ext4 crc16 mbcache jbd2 hid_generic usbhid hid sg sr_mod cdrom ata_generic virtio_blk virtio_net floppy ata_piix e1000 libata ehci_pci virtio_pci scsi_mod uhci_hcd ehci_hcd virtio_ring virtio usbcore usb_common [last unloaded: bonding] [ 908.984168] CPU: 0 PID: 1787 Comm: rmmod Tainted: G W O 4.2.0-rc2+ #8 [ 908.984170] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011 [ 908.984172] 0000000000000000 ffffffff81732d41 ffffffff81525b34 ffff8800358dfda8 [ 908.984175] ffffffff8106c521 ffff88003595af78 ffff88003595af40 ffff88003e3a4280 [ 908.984178] ffffffffa058d040 0000000000000000 ffffffff8106c59a ffffffff8172ebd0 [ 908.984181] Call Trace: [ 908.984188] [<ffffffff81525b34>] ? dump_stack+0x40/0x50 [ 908.984193] [<ffffffff8106c521>] ? warn_slowpath_common+0x81/0xb0 [ 908.984196] [<ffffffff8106c59a>] ? warn_slowpath_fmt+0x4a/0x50 [ 908.984199] [<ffffffff81218352>] ? remove_proc_entry+0x112/0x160 [ 908.984205] [<ffffffffa05850e6>] ? bond_destroy_proc_dir+0x26/0x30 [bonding] [ 908.984208] [<ffffffffa057540e>] ? bond_net_exit+0x8e/0xa0 [bonding] [ 908.984217] [<ffffffff8142f407>] ? ops_exit_list.isra.4+0x37/0x70 [ 908.984225] [<ffffffff8142f52d>] ? unregister_pernet_operations+0x8d/0xd0 [ 908.984228] [<ffffffff8142f58d>] ? unregister_pernet_subsys+0x1d/0x30 [ 908.984232] [<ffffffffa0585269>] ? bonding_exit+0x23/0xdba [bonding] [ 908.984236] [<ffffffff810e28ba>] ? SyS_delete_module+0x18a/0x250 [ 908.984241] [<ffffffff81086f99>] ? task_work_run+0x89/0xc0 [ 908.984244] [<ffffffff8152b732>] ? entry_SYSCALL_64_fastpath+0x16/0x75 [ 908.984247] ---[ end trace 7c006ed4abbef24b ]--- Thus remove the proc entry manually if bond_release_and_destroy() is used. Because of the checks in bond_remove_proc_entry() it's not a problem for a bond device to change namespaces (the bug fixed by the Fixes commit) but since commit f9399814927ad ("bonding: Don't allow bond devices to change network namespaces.") that can't happen anyway. Reported-by: Carol Soto <clsoto@linux.vnet.ibm.com> Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com> Fixes: a64d49c3dd50 ("bonding: Manage /proc/net/bonding/ entries from the netdev events") Tested-by: Carol L Soto <clsoto@linux.vnet.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-07-16 02:52:51 +07:00
bond_remove_proc_entry(bond);
unregister_netdevice(bond_dev);
}
return ret;
}
static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
{
struct bonding *bond = netdev_priv(bond_dev);
bond_fill_ifbond(bond, info);
}
static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
{
struct bonding *bond = netdev_priv(bond_dev);
struct list_head *iter;
int i = 0, res = -ENODEV;
struct slave *slave;
bond_for_each_slave(bond, slave, iter) {
if (i++ == (int)info->slave_id) {
res = 0;
bond_fill_ifslave(slave, info);
break;
}
}
return res;
}
/*-------------------------------- Monitoring -------------------------------*/
/* called with rcu_read_lock() */
static int bond_miimon_inspect(struct bonding *bond)
{
int link_state, commit = 0;
struct list_head *iter;
struct slave *slave;
bool ignore_updelay;
ignore_updelay = !rcu_dereference(bond->curr_active_slave);
bond_for_each_slave_rcu(bond, slave, iter) {
slave->new_link = BOND_LINK_NOCHANGE;
bonding: fix slave stuck in BOND_LINK_FAIL state The bonding miimon logic has a flaw, in that a failure of the rtnl_trylock can cause a slave to become permanently stuck in BOND_LINK_FAIL state. The sequence of events to cause this is as follows: 1) bond_miimon_inspect finds that a slave's link is down, and so calls bond_propose_link_state, setting slave->new_link_state to BOND_LINK_FAIL, then sets slave->new_link to BOND_LINK_DOWN and returns non-zero. 2) In bond_mii_monitor, the rtnl_trylock fails, and the timer is rescheduled. No change is committed. 3) bond_miimon_inspect is called again, but this time the slave from step 1 has recovered. slave->new_link is reset to NOCHANGE, and, as slave->link was never changed, the switch enters the BOND_LINK_UP case, and does nothing. The pending BOND_LINK_FAIL state from step 1 remains pending, as new_link_state is not reset. 4) The state from step 3 persists until another slave changes link state and causes bond_miimon_inspect to return non-zero. At this point, the BOND_LINK_FAIL state change on the slave from steps 1-3 is committed, and the slave will remain stuck in BOND_LINK_FAIL state even though it is actually link up. The remedy for this is to initialize new_link_state on each entry to bond_miimon_inspect, as is already done with new_link. Fixes: fb9eb899a6dc ("bonding: handle link transition from FAIL to UP correctly") Reported-by: Alex Sidorenko <alexandre.sidorenko@hpe.com> Reviewed-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com> Acked-by: Mahesh Bandewar <maheshb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-11-07 17:50:07 +07:00
slave->link_new_state = slave->link;
link_state = bond_check_dev_link(bond, slave->dev, 0);
switch (slave->link) {
case BOND_LINK_UP:
if (link_state)
continue;
bond_propose_link_state(slave, BOND_LINK_FAIL);
commit++;
slave->delay = bond->params.downdelay;
if (slave->delay) {
netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
(BOND_MODE(bond) ==
BOND_MODE_ACTIVEBACKUP) ?
(bond_is_active_slave(slave) ?
"active " : "backup ") : "",
slave->dev->name,
bond->params.downdelay * bond->params.miimon);
}
/*FALLTHRU*/
case BOND_LINK_FAIL:
if (link_state) {
/* recovered before downdelay expired */
bond_propose_link_state(slave, BOND_LINK_UP);
slave->last_link_up = jiffies;
netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
(bond->params.downdelay - slave->delay) *
bond->params.miimon,
slave->dev->name);
commit++;
continue;
}
if (slave->delay <= 0) {
slave->new_link = BOND_LINK_DOWN;
commit++;
continue;
}
slave->delay--;
break;
case BOND_LINK_DOWN:
if (!link_state)
continue;
bond_propose_link_state(slave, BOND_LINK_BACK);
commit++;
slave->delay = bond->params.updelay;
if (slave->delay) {
netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
slave->dev->name,
ignore_updelay ? 0 :
bond->params.updelay *
bond->params.miimon);
}
/*FALLTHRU*/
case BOND_LINK_BACK:
if (!link_state) {
bond_propose_link_state(slave, BOND_LINK_DOWN);
netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
(bond->params.updelay - slave->delay) *
bond->params.miimon,
slave->dev->name);
commit++;
continue;
}
if (ignore_updelay)
slave->delay = 0;
if (slave->delay <= 0) {
slave->new_link = BOND_LINK_UP;
commit++;
ignore_updelay = false;
continue;
}
slave->delay--;
break;
}
}
return commit;
}
static void bond_miimon_link_change(struct bonding *bond,
struct slave *slave,
char link)
{
switch (BOND_MODE(bond)) {
case BOND_MODE_8023AD:
bond_3ad_handle_link_change(slave, link);
break;
case BOND_MODE_TLB:
case BOND_MODE_ALB:
bond_alb_handle_link_change(bond, slave, link);
break;
case BOND_MODE_XOR:
bond_update_slave_arr(bond, NULL);
break;
}
}
static void bond_miimon_commit(struct bonding *bond)
{
struct list_head *iter;
struct slave *slave, *primary;
bond_for_each_slave(bond, slave, iter) {
switch (slave->new_link) {
case BOND_LINK_NOCHANGE:
continue;
case BOND_LINK_UP:
if (bond_update_speed_duplex(slave) &&
bond_needs_speed_duplex(bond)) {
slave->link = BOND_LINK_DOWN;
if (net_ratelimit())
netdev_warn(bond->dev,
"failed to get link speed/duplex for %s\n",
slave->dev->name);
continue;
}
bond_set_slave_link_state(slave, BOND_LINK_UP,
BOND_SLAVE_NOTIFY_NOW);
slave->last_link_up = jiffies;
primary = rtnl_dereference(bond->primary_slave);
if (BOND_MODE(bond) == BOND_MODE_8023AD) {
/* prevent it from being the active one */
bond_set_backup_slave(slave);
} else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
/* make it immediately active */
bond_set_active_slave(slave);
} else if (slave != primary) {
/* prevent it from being the active one */
bond_set_backup_slave(slave);
}
netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
slave->dev->name,
slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
slave->duplex ? "full" : "half");
bond_miimon_link_change(bond, slave, BOND_LINK_UP);
if (!bond->curr_active_slave || slave == primary)
goto do_failover;
continue;
case BOND_LINK_DOWN:
if (slave->link_failure_count < UINT_MAX)
slave->link_failure_count++;
bond_set_slave_link_state(slave, BOND_LINK_DOWN,
BOND_SLAVE_NOTIFY_NOW);
if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
BOND_MODE(bond) == BOND_MODE_8023AD)
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_inactive_flags(slave,
BOND_SLAVE_NOTIFY_NOW);
netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
slave->dev->name);
bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
if (slave == rcu_access_pointer(bond->curr_active_slave))
goto do_failover;
continue;
default:
netdev_err(bond->dev, "invalid new link %d on slave %s\n",
slave->new_link, slave->dev->name);
slave->new_link = BOND_LINK_NOCHANGE;
continue;
}
do_failover:
block_netpoll_tx();
bond_select_active_slave(bond);
unblock_netpoll_tx();
}
bond_set_carrier(bond);
}
/* bond_mii_monitor
*
* Really a wrapper that splits the mii monitor into two phases: an
* inspection, then (if inspection indicates something needs to be done)
* an acquisition of appropriate locks followed by a commit phase to
* implement whatever link state changes are indicated.
*/
static void bond_mii_monitor(struct work_struct *work)
{
struct bonding *bond = container_of(work, struct bonding,
mii_work.work);
bool should_notify_peers = false;
unsigned long delay;
struct slave *slave;
struct list_head *iter;
delay = msecs_to_jiffies(bond->params.miimon);
if (!bond_has_slaves(bond))
goto re_arm;
rcu_read_lock();
should_notify_peers = bond_should_notify_peers(bond);
if (bond_miimon_inspect(bond)) {
rcu_read_unlock();
/* Race avoidance with bond_close cancel of workqueue */
if (!rtnl_trylock()) {
delay = 1;
should_notify_peers = false;
goto re_arm;
}
bond_for_each_slave(bond, slave, iter) {
bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
}
bond_miimon_commit(bond);
rtnl_unlock(); /* might sleep, hold no other locks */
} else
rcu_read_unlock();
re_arm:
bonding: eliminate bond_close race conditions This patch resolves two sets of race conditions. Mitsuo Hayasaka <mitsuo.hayasaka.hu@hitachi.com> reported the first, as follows: The bond_close() calls cancel_delayed_work() to cancel delayed works. It, however, cannot cancel works that were already queued in workqueue. The bond_open() initializes work->data, and proccess_one_work() refers get_work_cwq(work)->wq->flags. The get_work_cwq() returns NULL when work->data has been initialized. Thus, a panic occurs. He included a patch that converted the cancel_delayed_work calls in bond_close to flush_delayed_work_sync, which eliminated the above problem. His patch is incorporated, at least in principle, into this patch. In this patch, we use cancel_delayed_work_sync in place of flush_delayed_work_sync, and also convert bond_uninit in addition to bond_close. This conversion to _sync, however, opens new races between bond_close and three periodically executing workqueue functions: bond_mii_monitor, bond_alb_monitor and bond_activebackup_arp_mon. The race occurs because bond_close and bond_uninit are always called with RTNL held, and these workqueue functions may acquire RTNL to perform failover-related activities. If bond_close or bond_uninit is waiting in cancel_delayed_work_sync, deadlock occurs. These deadlocks are resolved by having the workqueue functions acquire RTNL conditionally. If the rtnl_trylock() fails, the functions reschedule and return immediately. For the cases that are attempting to perform link failover, a delay of 1 is used; for the other cases, the normal interval is used (as those activities are not as time critical). Additionally, the bond_mii_monitor function now stores the delay in a variable (mimicing the structure of activebackup_arp_mon). Lastly, all of the above renders the kill_timers sentinel moot, and therefore it has been removed. Tested-by: Mitsuo Hayasaka <mitsuo.hayasaka.hu@hitachi.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-10-28 22:42:50 +07:00
if (bond->params.miimon)
queue_delayed_work(bond->wq, &bond->mii_work, delay);
if (should_notify_peers) {
if (!rtnl_trylock())
return;
call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
rtnl_unlock();
}
}
static int bond_upper_dev_walk(struct net_device *upper, void *data)
{
__be32 ip = *((__be32 *)data);
return ip == bond_confirm_addr(upper, 0, ip);
}
static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
{
bool ret = false;
bonding: remove entries for master_ip and vlan_ip and query devices instead The following patch aimed to resolve an issue where secondary, tertiary, etc. addresses added to bond interfaces could overwrite the bond->master_ip and vlan_ip values. commit 917fbdb32f37e9a93b00bb12ee83532982982df3 Author: Henrik Saavedra Persson <henrik.e.persson@ericsson.com> Date: Wed Nov 23 23:37:15 2011 +0000 bonding: only use primary address for ARP That patch was good because it prevented bonds using ARP monitoring from sending frames with an invalid source IP address. Unfortunately, it didn't always work as expected. When using an ioctl (like ifconfig does) to set the IP address and netmask, 2 separate ioctls are actually called to set the IP and netmask if the mask chosen doesn't match the standard mask for that class of address. The first ioctl did not have a mask that matched the one in the primary address and would still cause the device address to be overwritten. The second ioctl that was called to set the mask would then detect as secondary and ignored, but the damage was already done. This was not an issue when using an application that used netlink sockets as the setting of IP and netmask came down at once. The inconsistent behavior between those two interfaces was something that needed to be resolved. While I was thinking about how I wanted to resolve this, Ralf Zeidler came with a patch that resolved this on a RHEL kernel by keeping a full shadow of the entries in dev->ifa_list for the bonding device and vlan devices in the bonding driver. I didn't like the duplication of the list as I want to see the 'bonding' struct and code shrink rather than grow, but liked the general idea. As the Subject indicates this patch drops the master_ip and vlan_ip elements from the 'bonding' and 'vlan_entry' structs, respectively. This can be done because a device's address-list is now traversed to determine the optimal source IP address for ARP requests and for checks to see if the bonding device has a particular IP address. This code could have all be contained inside the bonding driver, but it made more sense to me to EXPORT and call inet_confirm_addr since it did exactly what was needed. I tested this and a backported patch and everything works as expected. Ralf also helped with verification of the backported patch. Thanks to Ralf for all his help on this. v2: Whitespace and organizational changes based on suggestions from Jay Vosburgh and Dave Miller. v3: Fixup incorrect usage of rcu_read_unlock based on Dave Miller's suggestion. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> CC: Ralf Zeidler <ralf.zeidler@nsn.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-03-22 23:14:29 +07:00
if (ip == bond_confirm_addr(bond->dev, 0, ip))
return true;
rcu_read_lock();
if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
ret = true;
rcu_read_unlock();
return ret;
}
/* We go to the (large) trouble of VLAN tagging ARP frames because
* switches in VLAN mode (especially if ports are configured as
* "native" to a VLAN) might not pass non-tagged frames.
*/
bonding: support QinQ for bond arp interval The bond send arp request to indicate that the slave is active, and if the bond dev is a vlan dev, it will set the vlan tag in skb to notice the vlan group, but the bond could only send a skb with 802.1q proto, not support for QinQ. So add outer tag for lower vlan tag and inner tag for upper vlan tag to support QinQ, The new skb will be consist of two vlan tag just like this: dst mac | src mac | outer vlan tag | inner vlan tag | data | ..... If We don't need QinQ, the inner vlan tag could be set to 0 and use outer vlan tag as a normal vlan group. Using "ip link" to configure the bond for QinQ and add test log: ip link add link bond0 bond0.20 type vlan proto 802.1ad id 20 ip link add link bond0.20 bond0.20.200 type vlan proto 802.1q id 200 ifconfig bond0.20 11.11.20.36/24 ifconfig bond0.20.200 11.11.200.36/24 echo +11.11.200.37 > /sys/class/net/bond0/bonding/arp_ip_target 90:e2:ba:07:4a:5c (oui Unknown) > Broadcast, ethertype 802.1Q-QinQ (0x88a8),length 50: vlan 20, p 0,ethertype 802.1Q, vlan 200, p 0, ethertype ARP, Ethernet (len 6), IPv4 (len 4), Request who-has 11.11.200.37 tell 11.11.200.36, length 28 90:e2:ba:06:f9:86 (oui Unknown) > 90:e2:ba:07:4a:5c (oui Unknown), ethertype 802.1Q-QinQ (0x88a8), length 50: vlan 20, p 0, ethertype 802.1Q, vlan 200, p 0, ethertype ARP, Ethernet (len 6), IPv4 (len 4), Reply 11.11.200.37 is-at 90:e2:ba:06:f9:86 (oui Unknown), length 28 v1->v2: remove the comment "TODO: QinQ?". Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-03-25 16:44:43 +07:00
static void bond_arp_send(struct net_device *slave_dev, int arp_op,
__be32 dest_ip, __be32 src_ip,
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
struct bond_vlan_tag *tags)
{
struct sk_buff *skb;
struct bond_vlan_tag *outer_tag = tags;
netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
arp_op, slave_dev->name, &dest_ip, &src_ip);
skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
NULL, slave_dev->dev_addr, NULL);
if (!skb) {
net_err_ratelimited("ARP packet allocation failed\n");
return;
}
bonding: support QinQ for bond arp interval The bond send arp request to indicate that the slave is active, and if the bond dev is a vlan dev, it will set the vlan tag in skb to notice the vlan group, but the bond could only send a skb with 802.1q proto, not support for QinQ. So add outer tag for lower vlan tag and inner tag for upper vlan tag to support QinQ, The new skb will be consist of two vlan tag just like this: dst mac | src mac | outer vlan tag | inner vlan tag | data | ..... If We don't need QinQ, the inner vlan tag could be set to 0 and use outer vlan tag as a normal vlan group. Using "ip link" to configure the bond for QinQ and add test log: ip link add link bond0 bond0.20 type vlan proto 802.1ad id 20 ip link add link bond0.20 bond0.20.200 type vlan proto 802.1q id 200 ifconfig bond0.20 11.11.20.36/24 ifconfig bond0.20.200 11.11.200.36/24 echo +11.11.200.37 > /sys/class/net/bond0/bonding/arp_ip_target 90:e2:ba:07:4a:5c (oui Unknown) > Broadcast, ethertype 802.1Q-QinQ (0x88a8),length 50: vlan 20, p 0,ethertype 802.1Q, vlan 200, p 0, ethertype ARP, Ethernet (len 6), IPv4 (len 4), Request who-has 11.11.200.37 tell 11.11.200.36, length 28 90:e2:ba:06:f9:86 (oui Unknown) > 90:e2:ba:07:4a:5c (oui Unknown), ethertype 802.1Q-QinQ (0x88a8), length 50: vlan 20, p 0, ethertype 802.1Q, vlan 200, p 0, ethertype ARP, Ethernet (len 6), IPv4 (len 4), Reply 11.11.200.37 is-at 90:e2:ba:06:f9:86 (oui Unknown), length 28 v1->v2: remove the comment "TODO: QinQ?". Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-03-25 16:44:43 +07:00
if (!tags || tags->vlan_proto == VLAN_N_VID)
goto xmit;
tags++;
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
/* Go through all the tags backwards and add them to the packet */
while (tags->vlan_proto != VLAN_N_VID) {
if (!tags->vlan_id) {
tags++;
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
continue;
}
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
ntohs(outer_tag->vlan_proto), tags->vlan_id);
skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
tags->vlan_id);
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
if (!skb) {
net_err_ratelimited("failed to insert inner VLAN tag\n");
return;
}
tags++;
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
}
/* Set the outer tag */
if (outer_tag->vlan_id) {
netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
__vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
outer_tag->vlan_id);
}
xmit:
arp_xmit(skb);
}
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
/* Validate the device path between the @start_dev and the @end_dev.
* The path is valid if the @end_dev is reachable through device
* stacking.
* When the path is validated, collect any vlan information in the
* path.
*/
struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
struct net_device *end_dev,
int level)
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
{
struct bond_vlan_tag *tags;
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
struct net_device *upper;
struct list_head *iter;
if (start_dev == end_dev) {
treewide: kzalloc() -> kcalloc() The kzalloc() function has a 2-factor argument form, kcalloc(). This patch replaces cases of: kzalloc(a * b, gfp) with: kcalloc(a * b, gfp) as well as handling cases of: kzalloc(a * b * c, gfp) with: kzalloc(array3_size(a, b, c), gfp) as it's slightly less ugly than: kzalloc_array(array_size(a, b), c, gfp) This does, however, attempt to ignore constant size factors like: kzalloc(4 * 1024, gfp) though any constants defined via macros get caught up in the conversion. Any factors with a sizeof() of "unsigned char", "char", and "u8" were dropped, since they're redundant. The Coccinelle script used for this was: // Fix redundant parens around sizeof(). @@ type TYPE; expression THING, E; @@ ( kzalloc( - (sizeof(TYPE)) * E + sizeof(TYPE) * E , ...) | kzalloc( - (sizeof(THING)) * E + sizeof(THING) * E , ...) ) // Drop single-byte sizes and redundant parens. @@ expression COUNT; typedef u8; typedef __u8; @@ ( kzalloc( - sizeof(u8) * (COUNT) + COUNT , ...) | kzalloc( - sizeof(__u8) * (COUNT) + COUNT , ...) | kzalloc( - sizeof(char) * (COUNT) + COUNT , ...) | kzalloc( - sizeof(unsigned char) * (COUNT) + COUNT , ...) | kzalloc( - sizeof(u8) * COUNT + COUNT , ...) | kzalloc( - sizeof(__u8) * COUNT + COUNT , ...) | kzalloc( - sizeof(char) * COUNT + COUNT , ...) | kzalloc( - sizeof(unsigned char) * COUNT + COUNT , ...) ) // 2-factor product with sizeof(type/expression) and identifier or constant. @@ type TYPE; expression THING; identifier COUNT_ID; constant COUNT_CONST; @@ ( - kzalloc + kcalloc ( - sizeof(TYPE) * (COUNT_ID) + COUNT_ID, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(TYPE) * COUNT_ID + COUNT_ID, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(TYPE) * (COUNT_CONST) + COUNT_CONST, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(TYPE) * COUNT_CONST + COUNT_CONST, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * (COUNT_ID) + COUNT_ID, sizeof(THING) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * COUNT_ID + COUNT_ID, sizeof(THING) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * (COUNT_CONST) + COUNT_CONST, sizeof(THING) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * COUNT_CONST + COUNT_CONST, sizeof(THING) , ...) ) // 2-factor product, only identifiers. @@ identifier SIZE, COUNT; @@ - kzalloc + kcalloc ( - SIZE * COUNT + COUNT, SIZE , ...) // 3-factor product with 1 sizeof(type) or sizeof(expression), with // redundant parens removed. @@ expression THING; identifier STRIDE, COUNT; type TYPE; @@ ( kzalloc( - sizeof(TYPE) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kzalloc( - sizeof(TYPE) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kzalloc( - sizeof(TYPE) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kzalloc( - sizeof(TYPE) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(TYPE)) , ...) | kzalloc( - sizeof(THING) * (COUNT) * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | kzalloc( - sizeof(THING) * (COUNT) * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | kzalloc( - sizeof(THING) * COUNT * (STRIDE) + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) | kzalloc( - sizeof(THING) * COUNT * STRIDE + array3_size(COUNT, STRIDE, sizeof(THING)) , ...) ) // 3-factor product with 2 sizeof(variable), with redundant parens removed. @@ expression THING1, THING2; identifier COUNT; type TYPE1, TYPE2; @@ ( kzalloc( - sizeof(TYPE1) * sizeof(TYPE2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | kzalloc( - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2)) , ...) | kzalloc( - sizeof(THING1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | kzalloc( - sizeof(THING1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(THING1), sizeof(THING2)) , ...) | kzalloc( - sizeof(TYPE1) * sizeof(THING2) * COUNT + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) | kzalloc( - sizeof(TYPE1) * sizeof(THING2) * (COUNT) + array3_size(COUNT, sizeof(TYPE1), sizeof(THING2)) , ...) ) // 3-factor product, only identifiers, with redundant parens removed. @@ identifier STRIDE, SIZE, COUNT; @@ ( kzalloc( - (COUNT) * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - COUNT * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - COUNT * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - (COUNT) * (STRIDE) * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - COUNT * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - (COUNT) * STRIDE * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - (COUNT) * (STRIDE) * (SIZE) + array3_size(COUNT, STRIDE, SIZE) , ...) | kzalloc( - COUNT * STRIDE * SIZE + array3_size(COUNT, STRIDE, SIZE) , ...) ) // Any remaining multi-factor products, first at least 3-factor products, // when they're not all constants... @@ expression E1, E2, E3; constant C1, C2, C3; @@ ( kzalloc(C1 * C2 * C3, ...) | kzalloc( - (E1) * E2 * E3 + array3_size(E1, E2, E3) , ...) | kzalloc( - (E1) * (E2) * E3 + array3_size(E1, E2, E3) , ...) | kzalloc( - (E1) * (E2) * (E3) + array3_size(E1, E2, E3) , ...) | kzalloc( - E1 * E2 * E3 + array3_size(E1, E2, E3) , ...) ) // And then all remaining 2 factors products when they're not all constants, // keeping sizeof() as the second factor argument. @@ expression THING, E1, E2; type TYPE; constant C1, C2, C3; @@ ( kzalloc(sizeof(THING) * C2, ...) | kzalloc(sizeof(TYPE) * C2, ...) | kzalloc(C1 * C2 * C3, ...) | kzalloc(C1 * C2, ...) | - kzalloc + kcalloc ( - sizeof(TYPE) * (E2) + E2, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(TYPE) * E2 + E2, sizeof(TYPE) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * (E2) + E2, sizeof(THING) , ...) | - kzalloc + kcalloc ( - sizeof(THING) * E2 + E2, sizeof(THING) , ...) | - kzalloc + kcalloc ( - (E1) * E2 + E1, E2 , ...) | - kzalloc + kcalloc ( - (E1) * (E2) + E1, E2 , ...) | - kzalloc + kcalloc ( - E1 * E2 + E1, E2 , ...) ) Signed-off-by: Kees Cook <keescook@chromium.org>
2018-06-13 04:03:40 +07:00
tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
if (!tags)
return ERR_PTR(-ENOMEM);
tags[level].vlan_proto = VLAN_N_VID;
return tags;
}
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
tags = bond_verify_device_path(upper, end_dev, level + 1);
if (IS_ERR_OR_NULL(tags)) {
if (IS_ERR(tags))
return tags;
continue;
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
}
if (is_vlan_dev(upper)) {
tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
tags[level].vlan_id = vlan_dev_vlan_id(upper);
}
return tags;
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
}
return NULL;
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
}
static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
{
struct rtable *rt;
struct bond_vlan_tag *tags;
__be32 *targets = bond->params.arp_targets, addr;
bonding: support QinQ for bond arp interval The bond send arp request to indicate that the slave is active, and if the bond dev is a vlan dev, it will set the vlan tag in skb to notice the vlan group, but the bond could only send a skb with 802.1q proto, not support for QinQ. So add outer tag for lower vlan tag and inner tag for upper vlan tag to support QinQ, The new skb will be consist of two vlan tag just like this: dst mac | src mac | outer vlan tag | inner vlan tag | data | ..... If We don't need QinQ, the inner vlan tag could be set to 0 and use outer vlan tag as a normal vlan group. Using "ip link" to configure the bond for QinQ and add test log: ip link add link bond0 bond0.20 type vlan proto 802.1ad id 20 ip link add link bond0.20 bond0.20.200 type vlan proto 802.1q id 200 ifconfig bond0.20 11.11.20.36/24 ifconfig bond0.20.200 11.11.200.36/24 echo +11.11.200.37 > /sys/class/net/bond0/bonding/arp_ip_target 90:e2:ba:07:4a:5c (oui Unknown) > Broadcast, ethertype 802.1Q-QinQ (0x88a8),length 50: vlan 20, p 0,ethertype 802.1Q, vlan 200, p 0, ethertype ARP, Ethernet (len 6), IPv4 (len 4), Request who-has 11.11.200.37 tell 11.11.200.36, length 28 90:e2:ba:06:f9:86 (oui Unknown) > 90:e2:ba:07:4a:5c (oui Unknown), ethertype 802.1Q-QinQ (0x88a8), length 50: vlan 20, p 0, ethertype 802.1Q, vlan 200, p 0, ethertype ARP, Ethernet (len 6), IPv4 (len 4), Reply 11.11.200.37 is-at 90:e2:ba:06:f9:86 (oui Unknown), length 28 v1->v2: remove the comment "TODO: QinQ?". Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-03-25 16:44:43 +07:00
int i;
for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
tags = NULL;
/* Find out through which dev should the packet go */
rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
RTO_ONLINK, 0);
if (IS_ERR(rt)) {
/* there's no route to target - try to send arp
* probe to generate any traffic (arp_validate=0)
*/
if (bond->params.arp_validate)
net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
bond->dev->name,
&targets[i]);
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
0, tags);
continue;
}
/* bond device itself */
if (rt->dst.dev == bond->dev)
goto found;
rcu_read_lock();
tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
rcu_read_unlock();
if (!IS_ERR_OR_NULL(tags))
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
goto found;
/* Not our device - skip */
netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
&targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
ip_rt_put(rt);
continue;
found:
addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
ip_rt_put(rt);
bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
bonding: Fix stacked device detection in arp monitoring Prior to commit fbd929f2dce460456807a51e18d623db3db9f077 bonding: support QinQ for bond arp interval the arp monitoring code allowed for proper detection of devices stacked on top of vlans. Since the above commit, the code can still detect a device stacked on top of single vlan, but not a device stacked on top of Q-in-Q configuration. The search will only set the inner vlan tag if the route device is the vlan device. However, this is not always the case, as it is possible to extend the stacked configuration. With this patch it is possible to provision devices on top Q-in-Q vlan configuration that should be used as a source of ARP monitoring information. For example: ip link add link bond0 vlan10 type vlan proto 802.1q id 10 ip link add link vlan10 vlan100 type vlan proto 802.1q id 100 ip link add link vlan100 type macvlan Note: This patch limites the number of stacked VLANs to 2, just like before. The original, however had another issue in that if we had more then 2 levels of VLANs, we would end up generating incorrectly tagged traffic. This is no longer possible. Fixes: fbd929f2dce460456807a51e18d623db3db9f077 (bonding: support QinQ for bond arp interval) CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@redhat.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> CC: Patric McHardy <kaber@trash.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-05-17 04:20:38 +07:00
addr, tags);
kfree(tags);
}
}
static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
{
bonding: add an option to fail when any of arp_ip_target is inaccessible Currently, we fail only when all of the ips in arp_ip_target are gone. However, in some situations we might need to fail if even one host from arp_ip_target becomes unavailable. All situations, obviously, rely on the idea that we need *completely* functional network, with all interfaces/addresses working correctly. One real world example might be: vlans on top on bond (hybrid port). If bond and vlans have ips assigned and we have their peers monitored via arp_ip_target - in case of switch misconfiguration (trunk/access port), slave driver malfunction or tagged/untagged traffic dropped on the way - we will be able to switch to another slave. Though any other configuration needs that if we need to have access to all arp_ip_targets. This patch adds this possibility by adding a new parameter - arp_all_targets (both as a module parameter and as a sysfs knob). It can be set to: 0 or any (the default) - which works exactly as it's working now - the slave is up if any of the arp_ip_targets are up. 1 or all - the slave is up if all of the arp_ip_targets are up. This parameter can be changed on the fly (via sysfs), and requires the mode to be active-backup and arp_validate to be enabled (it obeys the arp_validate config on which slaves to validate). Internally it's done through: 1) Add target_last_arp_rx[BOND_MAX_ARP_TARGETS] array to slave struct. It's an array of jiffies, meaning that slave->target_last_arp_rx[i] is the last time we've received arp from bond->params.arp_targets[i] on this slave. 2) If we successfully validate an arp from bond->params.arp_targets[i] in bond_validate_arp() - update the slave->target_last_arp_rx[i] with the current jiffies value. 3) When getting slave's last_rx via slave_last_rx(), we return the oldest time when we've received an arp from any address in bond->params.arp_targets[]. If the value of arp_all_targets == 0 - we still work the same way as before. Also, update the documentation to reflect the new parameter. v3->v4: Kill the forgotten rtnl_unlock(), rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Also, print a warning if the last arp_ip_target is removed while the arp_interval is on, but not the arp_validate. v2->v3: Use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). On bond_enslave(), use the same initialization value for target_last_arp_rx[] as is used for the default last_arp_rx, to avoid useless interface flaps. Also, instead of failing to remove the last arp_ip_target just print a warning - otherwise it might break existing scripts. v1->v2: Correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes. Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-24 16:49:34 +07:00
int i;
if (!sip || !bond_has_this_ip(bond, tip)) {
netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
&sip, &tip);
return;
}
bonding: add an option to fail when any of arp_ip_target is inaccessible Currently, we fail only when all of the ips in arp_ip_target are gone. However, in some situations we might need to fail if even one host from arp_ip_target becomes unavailable. All situations, obviously, rely on the idea that we need *completely* functional network, with all interfaces/addresses working correctly. One real world example might be: vlans on top on bond (hybrid port). If bond and vlans have ips assigned and we have their peers monitored via arp_ip_target - in case of switch misconfiguration (trunk/access port), slave driver malfunction or tagged/untagged traffic dropped on the way - we will be able to switch to another slave. Though any other configuration needs that if we need to have access to all arp_ip_targets. This patch adds this possibility by adding a new parameter - arp_all_targets (both as a module parameter and as a sysfs knob). It can be set to: 0 or any (the default) - which works exactly as it's working now - the slave is up if any of the arp_ip_targets are up. 1 or all - the slave is up if all of the arp_ip_targets are up. This parameter can be changed on the fly (via sysfs), and requires the mode to be active-backup and arp_validate to be enabled (it obeys the arp_validate config on which slaves to validate). Internally it's done through: 1) Add target_last_arp_rx[BOND_MAX_ARP_TARGETS] array to slave struct. It's an array of jiffies, meaning that slave->target_last_arp_rx[i] is the last time we've received arp from bond->params.arp_targets[i] on this slave. 2) If we successfully validate an arp from bond->params.arp_targets[i] in bond_validate_arp() - update the slave->target_last_arp_rx[i] with the current jiffies value. 3) When getting slave's last_rx via slave_last_rx(), we return the oldest time when we've received an arp from any address in bond->params.arp_targets[]. If the value of arp_all_targets == 0 - we still work the same way as before. Also, update the documentation to reflect the new parameter. v3->v4: Kill the forgotten rtnl_unlock(), rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Also, print a warning if the last arp_ip_target is removed while the arp_interval is on, but not the arp_validate. v2->v3: Use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). On bond_enslave(), use the same initialization value for target_last_arp_rx[] as is used for the default last_arp_rx, to avoid useless interface flaps. Also, instead of failing to remove the last arp_ip_target just print a warning - otherwise it might break existing scripts. v1->v2: Correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes. Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-24 16:49:34 +07:00
i = bond_get_targets_ip(bond->params.arp_targets, sip);
if (i == -1) {
netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
&sip);
return;
}
slave->last_rx = jiffies;
bonding: add an option to fail when any of arp_ip_target is inaccessible Currently, we fail only when all of the ips in arp_ip_target are gone. However, in some situations we might need to fail if even one host from arp_ip_target becomes unavailable. All situations, obviously, rely on the idea that we need *completely* functional network, with all interfaces/addresses working correctly. One real world example might be: vlans on top on bond (hybrid port). If bond and vlans have ips assigned and we have their peers monitored via arp_ip_target - in case of switch misconfiguration (trunk/access port), slave driver malfunction or tagged/untagged traffic dropped on the way - we will be able to switch to another slave. Though any other configuration needs that if we need to have access to all arp_ip_targets. This patch adds this possibility by adding a new parameter - arp_all_targets (both as a module parameter and as a sysfs knob). It can be set to: 0 or any (the default) - which works exactly as it's working now - the slave is up if any of the arp_ip_targets are up. 1 or all - the slave is up if all of the arp_ip_targets are up. This parameter can be changed on the fly (via sysfs), and requires the mode to be active-backup and arp_validate to be enabled (it obeys the arp_validate config on which slaves to validate). Internally it's done through: 1) Add target_last_arp_rx[BOND_MAX_ARP_TARGETS] array to slave struct. It's an array of jiffies, meaning that slave->target_last_arp_rx[i] is the last time we've received arp from bond->params.arp_targets[i] on this slave. 2) If we successfully validate an arp from bond->params.arp_targets[i] in bond_validate_arp() - update the slave->target_last_arp_rx[i] with the current jiffies value. 3) When getting slave's last_rx via slave_last_rx(), we return the oldest time when we've received an arp from any address in bond->params.arp_targets[]. If the value of arp_all_targets == 0 - we still work the same way as before. Also, update the documentation to reflect the new parameter. v3->v4: Kill the forgotten rtnl_unlock(), rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Also, print a warning if the last arp_ip_target is removed while the arp_interval is on, but not the arp_validate. v2->v3: Use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). On bond_enslave(), use the same initialization value for target_last_arp_rx[] as is used for the default last_arp_rx, to avoid useless interface flaps. Also, instead of failing to remove the last arp_ip_target just print a warning - otherwise it might break existing scripts. v1->v2: Correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes. Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-24 16:49:34 +07:00
slave->target_last_arp_rx[i] = jiffies;
}
int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
struct slave *slave)
{
struct arphdr *arp = (struct arphdr *)skb->data;
bonding: Fix ARP monitor validation The current logic in bond_arp_rcv will accept an incoming ARP for validation if (a) the receiving slave is either "active" (which includes the currently active slave, or the current ARP slave) or, (b) there is a currently active slave, and it has received an ARP since it became active. For case (b), the receiving slave isn't the currently active slave, and is receiving the original broadcast ARP request, not an ARP reply from the target. This logic can fail if there is no currently active slave. In this situation, the ARP probe logic cycles through all slaves, assigning each in turn as the "current_arp_slave" for one arp_interval, then setting that one as "active," and sending an ARP probe from that slave. The current logic expects the ARP reply to arrive on the sending current_arp_slave, however, due to switch FDB updating delays, the reply may be directed to another slave. This can arise if the bonding slaves and switch are working, but the ARP target is not responding. When the ARP target recovers, a condition may result wherein the ARP target host replies faster than the switch can update its forwarding table, causing each ARP reply to be sent to the previous current_arp_slave. This will never pass the logic in bond_arp_rcv, as neither of the above conditions (a) or (b) are met. Some experimentation on a LAN shows ARP reply round trips in the 200 usec range, but my available switches never update their FDB in less than 4000 usec. This patch changes the logic in bond_arp_rcv to additionally accept an ARP reply for validation on any slave if there is a current ARP slave and it sent an ARP probe during the previous arp_interval. Fixes: aeea64ac717a ("bonding: don't trust arp requests unless active slave really works") Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <gospo@cumulusnetworks.com> Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-02-03 04:35:56 +07:00
struct slave *curr_active_slave, *curr_arp_slave;
unsigned char *arp_ptr;
__be32 sip, tip;
int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
unsigned int alen;
if (!slave_do_arp_validate(bond, slave)) {
if ((slave_do_arp_validate_only(bond) && is_arp) ||
!slave_do_arp_validate_only(bond))
slave->last_rx = jiffies;
return RX_HANDLER_ANOTHER;
} else if (!is_arp) {
return RX_HANDLER_ANOTHER;
}
alen = arp_hdr_len(bond->dev);
netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
skb->dev->name);
if (alen > skb_headlen(skb)) {
arp = kmalloc(alen, GFP_ATOMIC);
if (!arp)
goto out_unlock;
if (skb_copy_bits(skb, 0, arp, alen) < 0)
goto out_unlock;
}
if (arp->ar_hln != bond->dev->addr_len ||
skb->pkt_type == PACKET_OTHERHOST ||
skb->pkt_type == PACKET_LOOPBACK ||
arp->ar_hrd != htons(ARPHRD_ETHER) ||
arp->ar_pro != htons(ETH_P_IP) ||
arp->ar_pln != 4)
goto out_unlock;
arp_ptr = (unsigned char *)(arp + 1);
arp_ptr += bond->dev->addr_len;
memcpy(&sip, arp_ptr, 4);
arp_ptr += 4 + bond->dev->addr_len;
memcpy(&tip, arp_ptr, 4);
netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
slave->dev->name, bond_slave_state(slave),
bond->params.arp_validate, slave_do_arp_validate(bond, slave),
&sip, &tip);
curr_active_slave = rcu_dereference(bond->curr_active_slave);
bonding: Fix ARP monitor validation The current logic in bond_arp_rcv will accept an incoming ARP for validation if (a) the receiving slave is either "active" (which includes the currently active slave, or the current ARP slave) or, (b) there is a currently active slave, and it has received an ARP since it became active. For case (b), the receiving slave isn't the currently active slave, and is receiving the original broadcast ARP request, not an ARP reply from the target. This logic can fail if there is no currently active slave. In this situation, the ARP probe logic cycles through all slaves, assigning each in turn as the "current_arp_slave" for one arp_interval, then setting that one as "active," and sending an ARP probe from that slave. The current logic expects the ARP reply to arrive on the sending current_arp_slave, however, due to switch FDB updating delays, the reply may be directed to another slave. This can arise if the bonding slaves and switch are working, but the ARP target is not responding. When the ARP target recovers, a condition may result wherein the ARP target host replies faster than the switch can update its forwarding table, causing each ARP reply to be sent to the previous current_arp_slave. This will never pass the logic in bond_arp_rcv, as neither of the above conditions (a) or (b) are met. Some experimentation on a LAN shows ARP reply round trips in the 200 usec range, but my available switches never update their FDB in less than 4000 usec. This patch changes the logic in bond_arp_rcv to additionally accept an ARP reply for validation on any slave if there is a current ARP slave and it sent an ARP probe during the previous arp_interval. Fixes: aeea64ac717a ("bonding: don't trust arp requests unless active slave really works") Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <gospo@cumulusnetworks.com> Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-02-03 04:35:56 +07:00
curr_arp_slave = rcu_dereference(bond->current_arp_slave);
bonding: Fix ARP monitor validation The current logic in bond_arp_rcv will accept an incoming ARP for validation if (a) the receiving slave is either "active" (which includes the currently active slave, or the current ARP slave) or, (b) there is a currently active slave, and it has received an ARP since it became active. For case (b), the receiving slave isn't the currently active slave, and is receiving the original broadcast ARP request, not an ARP reply from the target. This logic can fail if there is no currently active slave. In this situation, the ARP probe logic cycles through all slaves, assigning each in turn as the "current_arp_slave" for one arp_interval, then setting that one as "active," and sending an ARP probe from that slave. The current logic expects the ARP reply to arrive on the sending current_arp_slave, however, due to switch FDB updating delays, the reply may be directed to another slave. This can arise if the bonding slaves and switch are working, but the ARP target is not responding. When the ARP target recovers, a condition may result wherein the ARP target host replies faster than the switch can update its forwarding table, causing each ARP reply to be sent to the previous current_arp_slave. This will never pass the logic in bond_arp_rcv, as neither of the above conditions (a) or (b) are met. Some experimentation on a LAN shows ARP reply round trips in the 200 usec range, but my available switches never update their FDB in less than 4000 usec. This patch changes the logic in bond_arp_rcv to additionally accept an ARP reply for validation on any slave if there is a current ARP slave and it sent an ARP probe during the previous arp_interval. Fixes: aeea64ac717a ("bonding: don't trust arp requests unless active slave really works") Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <gospo@cumulusnetworks.com> Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-02-03 04:35:56 +07:00
/* We 'trust' the received ARP enough to validate it if:
*
bonding: Fix ARP monitor validation The current logic in bond_arp_rcv will accept an incoming ARP for validation if (a) the receiving slave is either "active" (which includes the currently active slave, or the current ARP slave) or, (b) there is a currently active slave, and it has received an ARP since it became active. For case (b), the receiving slave isn't the currently active slave, and is receiving the original broadcast ARP request, not an ARP reply from the target. This logic can fail if there is no currently active slave. In this situation, the ARP probe logic cycles through all slaves, assigning each in turn as the "current_arp_slave" for one arp_interval, then setting that one as "active," and sending an ARP probe from that slave. The current logic expects the ARP reply to arrive on the sending current_arp_slave, however, due to switch FDB updating delays, the reply may be directed to another slave. This can arise if the bonding slaves and switch are working, but the ARP target is not responding. When the ARP target recovers, a condition may result wherein the ARP target host replies faster than the switch can update its forwarding table, causing each ARP reply to be sent to the previous current_arp_slave. This will never pass the logic in bond_arp_rcv, as neither of the above conditions (a) or (b) are met. Some experimentation on a LAN shows ARP reply round trips in the 200 usec range, but my available switches never update their FDB in less than 4000 usec. This patch changes the logic in bond_arp_rcv to additionally accept an ARP reply for validation on any slave if there is a current ARP slave and it sent an ARP probe during the previous arp_interval. Fixes: aeea64ac717a ("bonding: don't trust arp requests unless active slave really works") Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <gospo@cumulusnetworks.com> Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-02-03 04:35:56 +07:00
* (a) the slave receiving the ARP is active (which includes the
* current ARP slave, if any), or
*
* (b) the receiving slave isn't active, but there is a currently
* active slave and it received valid arp reply(s) after it became
* the currently active slave, or
*
* (c) there is an ARP slave that sent an ARP during the prior ARP
* interval, and we receive an ARP reply on any slave. We accept
* these because switch FDB update delays may deliver the ARP
* reply to a slave other than the sender of the ARP request.
*
* Note: for (b), backup slaves are receiving the broadcast ARP
* request, not a reply. This request passes from the sending
* slave through the L2 switch(es) to the receiving slave. Since
* this is checking the request, sip/tip are swapped for
* validation.
*
* This is done to avoid endless looping when we can't reach the
* arp_ip_target and fool ourselves with our own arp requests.
*/
if (bond_is_active_slave(slave))
bond_validate_arp(bond, slave, sip, tip);
else if (curr_active_slave &&
time_after(slave_last_rx(bond, curr_active_slave),
curr_active_slave->last_link_up))
bond_validate_arp(bond, slave, tip, sip);
bonding: Fix ARP monitor validation The current logic in bond_arp_rcv will accept an incoming ARP for validation if (a) the receiving slave is either "active" (which includes the currently active slave, or the current ARP slave) or, (b) there is a currently active slave, and it has received an ARP since it became active. For case (b), the receiving slave isn't the currently active slave, and is receiving the original broadcast ARP request, not an ARP reply from the target. This logic can fail if there is no currently active slave. In this situation, the ARP probe logic cycles through all slaves, assigning each in turn as the "current_arp_slave" for one arp_interval, then setting that one as "active," and sending an ARP probe from that slave. The current logic expects the ARP reply to arrive on the sending current_arp_slave, however, due to switch FDB updating delays, the reply may be directed to another slave. This can arise if the bonding slaves and switch are working, but the ARP target is not responding. When the ARP target recovers, a condition may result wherein the ARP target host replies faster than the switch can update its forwarding table, causing each ARP reply to be sent to the previous current_arp_slave. This will never pass the logic in bond_arp_rcv, as neither of the above conditions (a) or (b) are met. Some experimentation on a LAN shows ARP reply round trips in the 200 usec range, but my available switches never update their FDB in less than 4000 usec. This patch changes the logic in bond_arp_rcv to additionally accept an ARP reply for validation on any slave if there is a current ARP slave and it sent an ARP probe during the previous arp_interval. Fixes: aeea64ac717a ("bonding: don't trust arp requests unless active slave really works") Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <gospo@cumulusnetworks.com> Signed-off-by: Jay Vosburgh <jay.vosburgh@canonical.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2016-02-03 04:35:56 +07:00
else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
bond_time_in_interval(bond,
dev_trans_start(curr_arp_slave->dev), 1))
bond_validate_arp(bond, slave, sip, tip);
out_unlock:
if (arp != (struct arphdr *)skb->data)
kfree(arp);
return RX_HANDLER_ANOTHER;
}
/* function to verify if we're in the arp_interval timeslice, returns true if
* (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
* arp_interval/2) . the arp_interval/2 is needed for really fast networks.
*/
static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
int mod)
{
int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
return time_in_range(jiffies,
last_act - delta_in_ticks,
last_act + mod * delta_in_ticks + delta_in_ticks/2);
}
/* This function is called regularly to monitor each slave's link
* ensuring that traffic is being sent and received when arp monitoring
* is used in load-balancing mode. if the adapter has been dormant, then an
* arp is transmitted to generate traffic. see activebackup_arp_monitor for
* arp monitoring in active backup mode.
*/
static void bond_loadbalance_arp_mon(struct bonding *bond)
{
struct slave *slave, *oldcurrent;
struct list_head *iter;
int do_failover = 0, slave_state_changed = 0;
if (!bond_has_slaves(bond))
goto re_arm;
rcu_read_lock();
oldcurrent = rcu_dereference(bond->curr_active_slave);
/* see if any of the previous devices are up now (i.e. they have
* xmt and rcv traffic). the curr_active_slave does not come into
* the picture unless it is null. also, slave->last_link_up is not
* needed here because we send an arp on each slave and give a slave
* as long as it needs to get the tx/rx within the delta.
* TODO: what about up/down delay in arp mode? it wasn't here before
* so it can wait
*/
bond_for_each_slave_rcu(bond, slave, iter) {
unsigned long trans_start = dev_trans_start(slave->dev);
bonding: Don't update slave->link until ready to commit In the loadbalance arp monitoring scheme, when a slave link change is detected, the slave->link is immediately updated and slave_state_changed is set. Later down the function, the rtnl_lock is acquired and the changes are committed, updating the bond link state. However, the acquisition of the rtnl_lock can fail. The next time the monitor runs, since slave->link is already updated, it determines that link is unchanged. This results in the bond link state permanently out of sync with the slave link. This patch modifies bond_loadbalance_arp_mon() to handle link changes identical to bond_ab_arp_{inspect/commit}(). The new link state is maintained in slave->new_link until we're ready to commit at which point it's copied into slave->link. NOTE: miimon_{inspect/commit}() has a more complex state machine requiring the use of the bond_{propose,commit}_link_state() functions which maintains the intermediate state in slave->link_new_state. The arp monitors don't require that. Testing: This bug is very easy to reproduce with the following steps. 1. In a loop, toggle a slave link of a bond slave interface. 2. In a separate loop, do ifconfig up/down of an unrelated interface to create contention for rtnl_lock. Within a few iterations, the bond link goes out of sync with the slave link. Signed-off-by: Nithin Nayak Sujir <nsujir@tintri.com> Cc: Mahesh Bandewar <maheshb@google.com> Cc: Jay Vosburgh <jay.vosburgh@canonical.com> Acked-by: Mahesh Bandewar <maheshb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-25 09:45:17 +07:00
slave->new_link = BOND_LINK_NOCHANGE;
if (slave->link != BOND_LINK_UP) {
if (bond_time_in_interval(bond, trans_start, 1) &&
bond_time_in_interval(bond, slave->last_rx, 1)) {
bonding: Don't update slave->link until ready to commit In the loadbalance arp monitoring scheme, when a slave link change is detected, the slave->link is immediately updated and slave_state_changed is set. Later down the function, the rtnl_lock is acquired and the changes are committed, updating the bond link state. However, the acquisition of the rtnl_lock can fail. The next time the monitor runs, since slave->link is already updated, it determines that link is unchanged. This results in the bond link state permanently out of sync with the slave link. This patch modifies bond_loadbalance_arp_mon() to handle link changes identical to bond_ab_arp_{inspect/commit}(). The new link state is maintained in slave->new_link until we're ready to commit at which point it's copied into slave->link. NOTE: miimon_{inspect/commit}() has a more complex state machine requiring the use of the bond_{propose,commit}_link_state() functions which maintains the intermediate state in slave->link_new_state. The arp monitors don't require that. Testing: This bug is very easy to reproduce with the following steps. 1. In a loop, toggle a slave link of a bond slave interface. 2. In a separate loop, do ifconfig up/down of an unrelated interface to create contention for rtnl_lock. Within a few iterations, the bond link goes out of sync with the slave link. Signed-off-by: Nithin Nayak Sujir <nsujir@tintri.com> Cc: Mahesh Bandewar <maheshb@google.com> Cc: Jay Vosburgh <jay.vosburgh@canonical.com> Acked-by: Mahesh Bandewar <maheshb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-25 09:45:17 +07:00
slave->new_link = BOND_LINK_UP;
slave_state_changed = 1;
/* primary_slave has no meaning in round-robin
* mode. the window of a slave being up and
* curr_active_slave being null after enslaving
* is closed.
*/
if (!oldcurrent) {
netdev_info(bond->dev, "link status definitely up for interface %s\n",
slave->dev->name);
do_failover = 1;
} else {
netdev_info(bond->dev, "interface %s is now up\n",
slave->dev->name);
}
}
} else {
/* slave->link == BOND_LINK_UP */
/* not all switches will respond to an arp request
* when the source ip is 0, so don't take the link down
* if we don't know our ip yet
*/
if (!bond_time_in_interval(bond, trans_start, 2) ||
!bond_time_in_interval(bond, slave->last_rx, 2)) {
bonding: Don't update slave->link until ready to commit In the loadbalance arp monitoring scheme, when a slave link change is detected, the slave->link is immediately updated and slave_state_changed is set. Later down the function, the rtnl_lock is acquired and the changes are committed, updating the bond link state. However, the acquisition of the rtnl_lock can fail. The next time the monitor runs, since slave->link is already updated, it determines that link is unchanged. This results in the bond link state permanently out of sync with the slave link. This patch modifies bond_loadbalance_arp_mon() to handle link changes identical to bond_ab_arp_{inspect/commit}(). The new link state is maintained in slave->new_link until we're ready to commit at which point it's copied into slave->link. NOTE: miimon_{inspect/commit}() has a more complex state machine requiring the use of the bond_{propose,commit}_link_state() functions which maintains the intermediate state in slave->link_new_state. The arp monitors don't require that. Testing: This bug is very easy to reproduce with the following steps. 1. In a loop, toggle a slave link of a bond slave interface. 2. In a separate loop, do ifconfig up/down of an unrelated interface to create contention for rtnl_lock. Within a few iterations, the bond link goes out of sync with the slave link. Signed-off-by: Nithin Nayak Sujir <nsujir@tintri.com> Cc: Mahesh Bandewar <maheshb@google.com> Cc: Jay Vosburgh <jay.vosburgh@canonical.com> Acked-by: Mahesh Bandewar <maheshb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-25 09:45:17 +07:00
slave->new_link = BOND_LINK_DOWN;
slave_state_changed = 1;
if (slave->link_failure_count < UINT_MAX)
slave->link_failure_count++;
netdev_info(bond->dev, "interface %s is now down\n",
slave->dev->name);
if (slave == oldcurrent)
do_failover = 1;
}
}
/* note: if switch is in round-robin mode, all links
* must tx arp to ensure all links rx an arp - otherwise
* links may oscillate or not come up at all; if switch is
* in something like xor mode, there is nothing we can
* do - all replies will be rx'ed on same link causing slaves
* to be unstable during low/no traffic periods
*/
if (bond_slave_is_up(slave))
bond_arp_send_all(bond, slave);
}
rcu_read_unlock();
if (do_failover || slave_state_changed) {
if (!rtnl_trylock())
goto re_arm;
bonding: Don't update slave->link until ready to commit In the loadbalance arp monitoring scheme, when a slave link change is detected, the slave->link is immediately updated and slave_state_changed is set. Later down the function, the rtnl_lock is acquired and the changes are committed, updating the bond link state. However, the acquisition of the rtnl_lock can fail. The next time the monitor runs, since slave->link is already updated, it determines that link is unchanged. This results in the bond link state permanently out of sync with the slave link. This patch modifies bond_loadbalance_arp_mon() to handle link changes identical to bond_ab_arp_{inspect/commit}(). The new link state is maintained in slave->new_link until we're ready to commit at which point it's copied into slave->link. NOTE: miimon_{inspect/commit}() has a more complex state machine requiring the use of the bond_{propose,commit}_link_state() functions which maintains the intermediate state in slave->link_new_state. The arp monitors don't require that. Testing: This bug is very easy to reproduce with the following steps. 1. In a loop, toggle a slave link of a bond slave interface. 2. In a separate loop, do ifconfig up/down of an unrelated interface to create contention for rtnl_lock. Within a few iterations, the bond link goes out of sync with the slave link. Signed-off-by: Nithin Nayak Sujir <nsujir@tintri.com> Cc: Mahesh Bandewar <maheshb@google.com> Cc: Jay Vosburgh <jay.vosburgh@canonical.com> Acked-by: Mahesh Bandewar <maheshb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-25 09:45:17 +07:00
bond_for_each_slave(bond, slave, iter) {
if (slave->new_link != BOND_LINK_NOCHANGE)
slave->link = slave->new_link;
}
if (slave_state_changed) {
bond_slave_state_change(bond);
if (BOND_MODE(bond) == BOND_MODE_XOR)
bond_update_slave_arr(bond, NULL);
bonding: fix curr_active_slave/carrier with loadbalance arp monitoring Since commit 6fde8f037e60 ("bonding: fix locking in bond_loadbalance_arp_mon()") we can have a stale bond carrier state and stale curr_active_slave when using arp monitoring in loadbalance modes. The reason is that in bond_loadbalance_arp_mon() we can't have do_failover == true but slave_state_changed == false, whenever do_failover is true then slave_state_changed is also true. Then the following piece from bond_loadbalance_arp_mon(): if (slave_state_changed) { bond_slave_state_change(bond); if (BOND_MODE(bond) == BOND_MODE_XOR) bond_update_slave_arr(bond, NULL); } else if (do_failover) { block_netpoll_tx(); bond_select_active_slave(bond); unblock_netpoll_tx(); } will execute only the first branch, always and regardless of do_failover. Since these two events aren't related in such way, we need to decouple and consider them separately. For example this issue could lead to the following result: Bonding Mode: load balancing (round-robin) *MII Status: down* MII Polling Interval (ms): 0 Up Delay (ms): 0 Down Delay (ms): 0 ARP Polling Interval (ms): 100 ARP IP target/s (n.n.n.n form): 192.168.9.2 Slave Interface: ens12 *MII Status: up* Speed: 10000 Mbps Duplex: full Link Failure Count: 2 Permanent HW addr: 00:0f:53:01:42:2c Slave queue ID: 0 Slave Interface: eth1 *MII Status: up* Speed: Unknown Duplex: Unknown Link Failure Count: 70 Permanent HW addr: 52:54:00:2f:0f:8e Slave queue ID: 0 Since some interfaces are up, then the status of the bond should also be up, but it will never change unless something invokes bond_set_carrier() (i.e. enslave, bond_select_active_slave etc). Now, if I force the calling of bond_select_active_slave via for example changing primary_reselect (it can change in any mode), then the MII status goes to "up" because it calls bond_select_active_slave() which should've been done from bond_loadbalance_arp_mon() itself. CC: Veaceslav Falico <vfalico@gmail.com> CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Ding Tianhong <dingtianhong@huawei.com> Fixes: 6fde8f037e60 ("bonding: fix locking in bond_loadbalance_arp_mon()") Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Veaceslav Falico <vfalico@gmail.com> Acked-by: Andy Gospodarek <gospo@cumulusnetworks.com> Acked-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-11-18 21:14:44 +07:00
}
if (do_failover) {
block_netpoll_tx();
bond_select_active_slave(bond);
unblock_netpoll_tx();
}
rtnl_unlock();
}
re_arm:
bonding: eliminate bond_close race conditions This patch resolves two sets of race conditions. Mitsuo Hayasaka <mitsuo.hayasaka.hu@hitachi.com> reported the first, as follows: The bond_close() calls cancel_delayed_work() to cancel delayed works. It, however, cannot cancel works that were already queued in workqueue. The bond_open() initializes work->data, and proccess_one_work() refers get_work_cwq(work)->wq->flags. The get_work_cwq() returns NULL when work->data has been initialized. Thus, a panic occurs. He included a patch that converted the cancel_delayed_work calls in bond_close to flush_delayed_work_sync, which eliminated the above problem. His patch is incorporated, at least in principle, into this patch. In this patch, we use cancel_delayed_work_sync in place of flush_delayed_work_sync, and also convert bond_uninit in addition to bond_close. This conversion to _sync, however, opens new races between bond_close and three periodically executing workqueue functions: bond_mii_monitor, bond_alb_monitor and bond_activebackup_arp_mon. The race occurs because bond_close and bond_uninit are always called with RTNL held, and these workqueue functions may acquire RTNL to perform failover-related activities. If bond_close or bond_uninit is waiting in cancel_delayed_work_sync, deadlock occurs. These deadlocks are resolved by having the workqueue functions acquire RTNL conditionally. If the rtnl_trylock() fails, the functions reschedule and return immediately. For the cases that are attempting to perform link failover, a delay of 1 is used; for the other cases, the normal interval is used (as those activities are not as time critical). Additionally, the bond_mii_monitor function now stores the delay in a variable (mimicing the structure of activebackup_arp_mon). Lastly, all of the above renders the kill_timers sentinel moot, and therefore it has been removed. Tested-by: Mitsuo Hayasaka <mitsuo.hayasaka.hu@hitachi.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2011-10-28 22:42:50 +07:00
if (bond->params.arp_interval)
queue_delayed_work(bond->wq, &bond->arp_work,
msecs_to_jiffies(bond->params.arp_interval));
}
/* Called to inspect slaves for active-backup mode ARP monitor link state
* changes. Sets new_link in slaves to specify what action should take
* place for the slave. Returns 0 if no changes are found, >0 if changes
* to link states must be committed.
*
* Called with rcu_read_lock held.
*/
static int bond_ab_arp_inspect(struct bonding *bond)
{
unsigned long trans_start, last_rx;
struct list_head *iter;
struct slave *slave;
int commit = 0;
bond_for_each_slave_rcu(bond, slave, iter) {
slave->new_link = BOND_LINK_NOCHANGE;
last_rx = slave_last_rx(bond, slave);
if (slave->link != BOND_LINK_UP) {
if (bond_time_in_interval(bond, last_rx, 1)) {
slave->new_link = BOND_LINK_UP;
commit++;
}
continue;
}
/* Give slaves 2*delta after being enslaved or made
* active. This avoids bouncing, as the last receive
* times need a full ARP monitor cycle to be updated.
*/
if (bond_time_in_interval(bond, slave->last_link_up, 2))
continue;
/* Backup slave is down if:
* - No current_arp_slave AND
* - more than 3*delta since last receive AND
* - the bond has an IP address
*
* Note: a non-null current_arp_slave indicates
* the curr_active_slave went down and we are
* searching for a new one; under this condition
* we only take the curr_active_slave down - this
* gives each slave a chance to tx/rx traffic
* before being taken out
*/
if (!bond_is_active_slave(slave) &&
!rcu_access_pointer(bond->current_arp_slave) &&
!bond_time_in_interval(bond, last_rx, 3)) {
slave->new_link = BOND_LINK_DOWN;
commit++;
}
/* Active slave is down if:
* - more than 2*delta since transmitting OR
* - (more than 2*delta since receive AND
* the bond has an IP address)
*/
trans_start = dev_trans_start(slave->dev);
if (bond_is_active_slave(slave) &&
(!bond_time_in_interval(bond, trans_start, 2) ||
!bond_time_in_interval(bond, last_rx, 2))) {
slave->new_link = BOND_LINK_DOWN;
commit++;
}
}
return commit;
}
/* Called to commit link state changes noted by inspection step of
* active-backup mode ARP monitor.
*
* Called with RTNL hold.
*/
static void bond_ab_arp_commit(struct bonding *bond)
{
unsigned long trans_start;
struct list_head *iter;
struct slave *slave;
bond_for_each_slave(bond, slave, iter) {
switch (slave->new_link) {
case BOND_LINK_NOCHANGE:
continue;
case BOND_LINK_UP:
trans_start = dev_trans_start(slave->dev);
if (rtnl_dereference(bond->curr_active_slave) != slave ||
(!rtnl_dereference(bond->curr_active_slave) &&
bond_time_in_interval(bond, trans_start, 1))) {
struct slave *current_arp_slave;
current_arp_slave = rtnl_dereference(bond->current_arp_slave);
bond_set_slave_link_state(slave, BOND_LINK_UP,
BOND_SLAVE_NOTIFY_NOW);
if (current_arp_slave) {
bond_set_slave_inactive_flags(
current_arp_slave,
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
BOND_SLAVE_NOTIFY_NOW);
RCU_INIT_POINTER(bond->current_arp_slave, NULL);
}
netdev_info(bond->dev, "link status definitely up for interface %s\n",
slave->dev->name);
if (!rtnl_dereference(bond->curr_active_slave) ||
slave == rtnl_dereference(bond->primary_slave))
goto do_failover;
}
continue;
case BOND_LINK_DOWN:
if (slave->link_failure_count < UINT_MAX)
slave->link_failure_count++;
bond_set_slave_link_state(slave, BOND_LINK_DOWN,
BOND_SLAVE_NOTIFY_NOW);
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_inactive_flags(slave,
BOND_SLAVE_NOTIFY_NOW);
netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
slave->dev->name);
if (slave == rtnl_dereference(bond->curr_active_slave)) {
RCU_INIT_POINTER(bond->current_arp_slave, NULL);
goto do_failover;
}
continue;
default:
netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
slave->new_link, slave->dev->name);
continue;
}
do_failover:
block_netpoll_tx();
bond_select_active_slave(bond);
unblock_netpoll_tx();
}
bond_set_carrier(bond);
}
/* Send ARP probes for active-backup mode ARP monitor.
*
* Called with rcu_read_lock held.
*/
static bool bond_ab_arp_probe(struct bonding *bond)
{
struct slave *slave, *before = NULL, *new_slave = NULL,
*curr_arp_slave = rcu_dereference(bond->current_arp_slave),
*curr_active_slave = rcu_dereference(bond->curr_active_slave);
struct list_head *iter;
bool found = false;
bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
if (curr_arp_slave && curr_active_slave)
netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
curr_arp_slave->dev->name,
curr_active_slave->dev->name);
if (curr_active_slave) {
bond_arp_send_all(bond, curr_active_slave);
return should_notify_rtnl;
}
/* if we don't have a curr_active_slave, search for the next available
* backup slave from the current_arp_slave and make it the candidate
* for becoming the curr_active_slave
*/
if (!curr_arp_slave) {
curr_arp_slave = bond_first_slave_rcu(bond);
if (!curr_arp_slave)
return should_notify_rtnl;
}
bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
bond_for_each_slave_rcu(bond, slave, iter) {
if (!found && !before && bond_slave_is_up(slave))
before = slave;
if (found && !new_slave && bond_slave_is_up(slave))
new_slave = slave;
/* if the link state is up at this point, we
* mark it down - this can happen if we have
* simultaneous link failures and
* reselect_active_interface doesn't make this
* one the current slave so it is still marked
* up when it is actually down
*/
if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
bond_set_slave_link_state(slave, BOND_LINK_DOWN,
BOND_SLAVE_NOTIFY_LATER);
if (slave->link_failure_count < UINT_MAX)
slave->link_failure_count++;
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_inactive_flags(slave,
BOND_SLAVE_NOTIFY_LATER);
netdev_info(bond->dev, "backup interface %s is now down\n",
slave->dev->name);
}
if (slave == curr_arp_slave)
found = true;
}
if (!new_slave && before)
new_slave = before;
if (!new_slave)
goto check_state;
bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
BOND_SLAVE_NOTIFY_LATER);
bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
bond_arp_send_all(bond, new_slave);
new_slave->last_link_up = jiffies;
rcu_assign_pointer(bond->current_arp_slave, new_slave);
check_state:
bond_for_each_slave_rcu(bond, slave, iter) {
if (slave->should_notify || slave->should_notify_link) {
should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
break;
}
}
return should_notify_rtnl;
}
static void bond_activebackup_arp_mon(struct bonding *bond)
{
bool should_notify_peers = false;
bool should_notify_rtnl = false;
int delta_in_ticks;
delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
if (!bond_has_slaves(bond))
goto re_arm;
rcu_read_lock();
should_notify_peers = bond_should_notify_peers(bond);
if (bond_ab_arp_inspect(bond)) {
rcu_read_unlock();
/* Race avoidance with bond_close flush of workqueue */
if (!rtnl_trylock()) {
delta_in_ticks = 1;
should_notify_peers = false;
goto re_arm;
}
bond_ab_arp_commit(bond);
rtnl_unlock();
rcu_read_lock();
}
should_notify_rtnl = bond_ab_arp_probe(bond);
rcu_read_unlock();
re_arm:
if (bond->params.arp_interval)
queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
if (should_notify_peers || should_notify_rtnl) {
if (!rtnl_trylock())
return;
if (should_notify_peers)
call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
bond->dev);
if (should_notify_rtnl) {
bond_slave_state_notify(bond);
bond_slave_link_notify(bond);
}
rtnl_unlock();
}
}
static void bond_arp_monitor(struct work_struct *work)
{
struct bonding *bond = container_of(work, struct bonding,
arp_work.work);
if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
bond_activebackup_arp_mon(bond);
else
bond_loadbalance_arp_mon(bond);
}
/*-------------------------- netdev event handling --------------------------*/
/* Change device name */
static int bond_event_changename(struct bonding *bond)
{
bond_remove_proc_entry(bond);
bond_create_proc_entry(bond);
bond_debug_reregister(bond);
return NOTIFY_DONE;
}
static int bond_master_netdev_event(unsigned long event,
struct net_device *bond_dev)
{
struct bonding *event_bond = netdev_priv(bond_dev);
switch (event) {
case NETDEV_CHANGENAME:
return bond_event_changename(event_bond);
case NETDEV_UNREGISTER:
bond_remove_proc_entry(event_bond);
break;
case NETDEV_REGISTER:
bond_create_proc_entry(event_bond);
break;
case NETDEV_NOTIFY_PEERS:
if (event_bond->send_peer_notif)
event_bond->send_peer_notif--;
break;
default:
break;
}
return NOTIFY_DONE;
}
static int bond_slave_netdev_event(unsigned long event,
struct net_device *slave_dev)
{
struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
struct bonding *bond;
struct net_device *bond_dev;
/* A netdev event can be generated while enslaving a device
* before netdev_rx_handler_register is called in which case
* slave will be NULL
*/
if (!slave)
return NOTIFY_DONE;
bond_dev = slave->bond->dev;
bond = slave->bond;
primary = rtnl_dereference(bond->primary_slave);
switch (event) {
case NETDEV_UNREGISTER:
if (bond_dev->type != ARPHRD_ETHER)
bond_release_and_destroy(bond_dev, slave_dev);
else
__bond_release_one(bond_dev, slave_dev, false, true);
break;
case NETDEV_UP:
case NETDEV_CHANGE:
/* For 802.3ad mode only:
* Getting invalid Speed/Duplex values here will put slave
* in weird state. So mark it as link-fail for the time
* being and let link-monitoring (miimon) set it right when
* correct speeds/duplex are available.
*/
if (bond_update_speed_duplex(slave) &&
BOND_MODE(bond) == BOND_MODE_8023AD)
slave->link = BOND_LINK_FAIL;
if (BOND_MODE(bond) == BOND_MODE_8023AD)
bond_3ad_adapter_speed_duplex_changed(slave);
/* Fallthrough */
case NETDEV_DOWN:
/* Refresh slave-array if applicable!
* If the setup does not use miimon or arpmon (mode-specific!),
* then these events will not cause the slave-array to be
* refreshed. This will cause xmit to use a slave that is not
* usable. Avoid such situation by refeshing the array at these
* events. If these (miimon/arpmon) parameters are configured
* then array gets refreshed twice and that should be fine!
*/
if (bond_mode_can_use_xmit_hash(bond))
bond_update_slave_arr(bond, NULL);
break;
case NETDEV_CHANGEMTU:
/* TODO: Should slaves be allowed to
* independently alter their MTU? For
* an active-backup bond, slaves need
* not be the same type of device, so
* MTUs may vary. For other modes,
* slaves arguably should have the
* same MTUs. To do this, we'd need to
* take over the slave's change_mtu
* function for the duration of their
* servitude.
*/
break;
case NETDEV_CHANGENAME:
/* we don't care if we don't have primary set */
if (!bond_uses_primary(bond) ||
!bond->params.primary[0])
break;
if (slave == primary) {
/* slave's name changed - he's no longer primary */
RCU_INIT_POINTER(bond->primary_slave, NULL);
} else if (!strcmp(slave_dev->name, bond->params.primary)) {
/* we have a new primary slave */
rcu_assign_pointer(bond->primary_slave, slave);
} else { /* we didn't change primary - exit */
break;
}
netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
primary ? slave_dev->name : "none");
block_netpoll_tx();
bond_select_active_slave(bond);
unblock_netpoll_tx();
break;
case NETDEV_FEAT_CHANGE:
bond_compute_features(bond);
break;
case NETDEV_RESEND_IGMP:
/* Propagate to master device */
call_netdevice_notifiers(event, slave->bond->dev);
break;
default:
break;
}
return NOTIFY_DONE;
}
/* bond_netdev_event: handle netdev notifier chain events.
*
* This function receives events for the netdev chain. The caller (an
[PATCH] Notifier chain update: API changes The kernel's implementation of notifier chains is unsafe. There is no protection against entries being added to or removed from a chain while the chain is in use. The issues were discussed in this thread: http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2 We noticed that notifier chains in the kernel fall into two basic usage classes: "Blocking" chains are always called from a process context and the callout routines are allowed to sleep; "Atomic" chains can be called from an atomic context and the callout routines are not allowed to sleep. We decided to codify this distinction and make it part of the API. Therefore this set of patches introduces three new, parallel APIs: one for blocking notifiers, one for atomic notifiers, and one for "raw" notifiers (which is really just the old API under a new name). New kinds of data structures are used for the heads of the chains, and new routines are defined for registration, unregistration, and calling a chain. The three APIs are explained in include/linux/notifier.h and their implementation is in kernel/sys.c. With atomic and blocking chains, the implementation guarantees that the chain links will not be corrupted and that chain callers will not get messed up by entries being added or removed. For raw chains the implementation provides no guarantees at all; users of this API must provide their own protections. (The idea was that situations may come up where the assumptions of the atomic and blocking APIs are not appropriate, so it should be possible for users to handle these things in their own way.) There are some limitations, which should not be too hard to live with. For atomic/blocking chains, registration and unregistration must always be done in a process context since the chain is protected by a mutex/rwsem. Also, a callout routine for a non-raw chain must not try to register or unregister entries on its own chain. (This did happen in a couple of places and the code had to be changed to avoid it.) Since atomic chains may be called from within an NMI handler, they cannot use spinlocks for synchronization. Instead we use RCU. The overhead falls almost entirely in the unregister routine, which is okay since unregistration is much less frequent that calling a chain. Here is the list of chains that we adjusted and their classifications. None of them use the raw API, so for the moment it is only a placeholder. ATOMIC CHAINS ------------- arch/i386/kernel/traps.c: i386die_chain arch/ia64/kernel/traps.c: ia64die_chain arch/powerpc/kernel/traps.c: powerpc_die_chain arch/sparc64/kernel/traps.c: sparc64die_chain arch/x86_64/kernel/traps.c: die_chain drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list kernel/panic.c: panic_notifier_list kernel/profile.c: task_free_notifier net/bluetooth/hci_core.c: hci_notifier net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain net/ipv6/addrconf.c: inet6addr_chain net/netfilter/nf_conntrack_core.c: nf_conntrack_chain net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain net/netlink/af_netlink.c: netlink_chain BLOCKING CHAINS --------------- arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain arch/s390/kernel/process.c: idle_chain arch/x86_64/kernel/process.c idle_notifier drivers/base/memory.c: memory_chain drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list drivers/macintosh/adb.c: adb_client_list drivers/macintosh/via-pmu.c sleep_notifier_list drivers/macintosh/via-pmu68k.c sleep_notifier_list drivers/macintosh/windfarm_core.c wf_client_list drivers/usb/core/notify.c usb_notifier_list drivers/video/fbmem.c fb_notifier_list kernel/cpu.c cpu_chain kernel/module.c module_notify_list kernel/profile.c munmap_notifier kernel/profile.c task_exit_notifier kernel/sys.c reboot_notifier_list net/core/dev.c netdev_chain net/decnet/dn_dev.c: dnaddr_chain net/ipv4/devinet.c: inetaddr_chain It's possible that some of these classifications are wrong. If they are, please let us know or submit a patch to fix them. Note that any chain that gets called very frequently should be atomic, because the rwsem read-locking used for blocking chains is very likely to incur cache misses on SMP systems. (However, if the chain's callout routines may sleep then the chain cannot be atomic.) The patch set was written by Alan Stern and Chandra Seetharaman, incorporating material written by Keith Owens and suggestions from Paul McKenney and Andrew Morton. [jes@sgi.com: restructure the notifier chain initialization macros] Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com> Signed-off-by: Jes Sorensen <jes@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 16:16:30 +07:00
* ioctl handler calling blocking_notifier_call_chain) holds the necessary
* locks for us to safely manipulate the slave devices (RTNL lock,
* dev_probe_lock).
*/
static int bond_netdev_event(struct notifier_block *this,
unsigned long event, void *ptr)
{
struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
netdev_dbg(event_dev, "event: %lx\n", event);
if (!(event_dev->priv_flags & IFF_BONDING))
return NOTIFY_DONE;
if (event_dev->flags & IFF_MASTER) {
netdev_dbg(event_dev, "IFF_MASTER\n");
return bond_master_netdev_event(event, event_dev);
}
if (event_dev->flags & IFF_SLAVE) {
netdev_dbg(event_dev, "IFF_SLAVE\n");
return bond_slave_netdev_event(event, event_dev);
}
return NOTIFY_DONE;
}
static struct notifier_block bond_netdev_notifier = {
.notifier_call = bond_netdev_event,
};
/*---------------------------- Hashing Policies -----------------------------*/
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
/* L2 hash helper */
static inline u32 bond_eth_hash(struct sk_buff *skb)
bonding: support for IPv6 transmit hashing Currently the "bonding" driver does not support load balancing outgoing traffic in LACP mode for IPv6 traffic. IPv4 (and TCP or UDP over IPv4) are currently supported; this patch adds transmit hashing for IPv6 (and TCP or UDP over IPv6), bringing IPv6 up to par with IPv4 support in the bonding driver. In addition, bounds checking has been added to all transmit hashing functions. The algorithm chosen (xor'ing the bottom three quads of the source and destination addresses together, then xor'ing each byte of that result into the bottom byte, finally xor'ing with the last bytes of the MAC addresses) was selected after testing almost 400,000 unique IPv6 addresses harvested from server logs. This algorithm had the most even distribution for both big- and little-endian architectures while still using few instructions. Its behavior also attempts to closely match that of the IPv4 algorithm. The IPv6 flow label was intentionally not included in the hash as it appears to be unset in the vast majority of IPv6 traffic sampled, and the current algorithm not using the flow label already offers a very even distribution. Fragmented IPv6 packets are handled the same way as fragmented IPv4 packets, ie, they are not balanced based on layer 4 information. Additionally, IPv6 packets with intermediate headers are not balanced based on layer 4 information. In practice these intermediate headers are not common and this should not cause any problems, and the alternative (a packet-parsing loop and look-up table) seemed slow and complicated for little gain. Tested-by: John Eaglesham <linux@8192.net> Signed-off-by: John Eaglesham <linux@8192.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-08-22 03:43:35 +07:00
{
struct ethhdr *ep, hdr_tmp;
bonding: support for IPv6 transmit hashing Currently the "bonding" driver does not support load balancing outgoing traffic in LACP mode for IPv6 traffic. IPv4 (and TCP or UDP over IPv4) are currently supported; this patch adds transmit hashing for IPv6 (and TCP or UDP over IPv6), bringing IPv6 up to par with IPv4 support in the bonding driver. In addition, bounds checking has been added to all transmit hashing functions. The algorithm chosen (xor'ing the bottom three quads of the source and destination addresses together, then xor'ing each byte of that result into the bottom byte, finally xor'ing with the last bytes of the MAC addresses) was selected after testing almost 400,000 unique IPv6 addresses harvested from server logs. This algorithm had the most even distribution for both big- and little-endian architectures while still using few instructions. Its behavior also attempts to closely match that of the IPv4 algorithm. The IPv6 flow label was intentionally not included in the hash as it appears to be unset in the vast majority of IPv6 traffic sampled, and the current algorithm not using the flow label already offers a very even distribution. Fragmented IPv6 packets are handled the same way as fragmented IPv4 packets, ie, they are not balanced based on layer 4 information. Additionally, IPv6 packets with intermediate headers are not balanced based on layer 4 information. In practice these intermediate headers are not common and this should not cause any problems, and the alternative (a packet-parsing loop and look-up table) seemed slow and complicated for little gain. Tested-by: John Eaglesham <linux@8192.net> Signed-off-by: John Eaglesham <linux@8192.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-08-22 03:43:35 +07:00
ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
if (ep)
return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
bonding: support for IPv6 transmit hashing Currently the "bonding" driver does not support load balancing outgoing traffic in LACP mode for IPv6 traffic. IPv4 (and TCP or UDP over IPv4) are currently supported; this patch adds transmit hashing for IPv6 (and TCP or UDP over IPv6), bringing IPv6 up to par with IPv4 support in the bonding driver. In addition, bounds checking has been added to all transmit hashing functions. The algorithm chosen (xor'ing the bottom three quads of the source and destination addresses together, then xor'ing each byte of that result into the bottom byte, finally xor'ing with the last bytes of the MAC addresses) was selected after testing almost 400,000 unique IPv6 addresses harvested from server logs. This algorithm had the most even distribution for both big- and little-endian architectures while still using few instructions. Its behavior also attempts to closely match that of the IPv4 algorithm. The IPv6 flow label was intentionally not included in the hash as it appears to be unset in the vast majority of IPv6 traffic sampled, and the current algorithm not using the flow label already offers a very even distribution. Fragmented IPv6 packets are handled the same way as fragmented IPv4 packets, ie, they are not balanced based on layer 4 information. Additionally, IPv6 packets with intermediate headers are not balanced based on layer 4 information. In practice these intermediate headers are not common and this should not cause any problems, and the alternative (a packet-parsing loop and look-up table) seemed slow and complicated for little gain. Tested-by: John Eaglesham <linux@8192.net> Signed-off-by: John Eaglesham <linux@8192.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-08-22 03:43:35 +07:00
return 0;
}
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
/* Extract the appropriate headers based on bond's xmit policy */
static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
struct flow_keys *fk)
{
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
const struct ipv6hdr *iph6;
const struct iphdr *iph;
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
int noff, proto = -1;
bonding: support for IPv6 transmit hashing Currently the "bonding" driver does not support load balancing outgoing traffic in LACP mode for IPv6 traffic. IPv4 (and TCP or UDP over IPv4) are currently supported; this patch adds transmit hashing for IPv6 (and TCP or UDP over IPv6), bringing IPv6 up to par with IPv4 support in the bonding driver. In addition, bounds checking has been added to all transmit hashing functions. The algorithm chosen (xor'ing the bottom three quads of the source and destination addresses together, then xor'ing each byte of that result into the bottom byte, finally xor'ing with the last bytes of the MAC addresses) was selected after testing almost 400,000 unique IPv6 addresses harvested from server logs. This algorithm had the most even distribution for both big- and little-endian architectures while still using few instructions. Its behavior also attempts to closely match that of the IPv4 algorithm. The IPv6 flow label was intentionally not included in the hash as it appears to be unset in the vast majority of IPv6 traffic sampled, and the current algorithm not using the flow label already offers a very even distribution. Fragmented IPv6 packets are handled the same way as fragmented IPv4 packets, ie, they are not balanced based on layer 4 information. Additionally, IPv6 packets with intermediate headers are not balanced based on layer 4 information. In practice these intermediate headers are not common and this should not cause any problems, and the alternative (a packet-parsing loop and look-up table) seemed slow and complicated for little gain. Tested-by: John Eaglesham <linux@8192.net> Signed-off-by: John Eaglesham <linux@8192.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-08-22 03:43:35 +07:00
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
return skb_flow_dissect_flow_keys(skb, fk, 0);
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
fk->ports.ports = 0;
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
noff = skb_network_offset(skb);
if (skb->protocol == htons(ETH_P_IP)) {
if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
return false;
bonding: support for IPv6 transmit hashing Currently the "bonding" driver does not support load balancing outgoing traffic in LACP mode for IPv6 traffic. IPv4 (and TCP or UDP over IPv4) are currently supported; this patch adds transmit hashing for IPv6 (and TCP or UDP over IPv6), bringing IPv6 up to par with IPv4 support in the bonding driver. In addition, bounds checking has been added to all transmit hashing functions. The algorithm chosen (xor'ing the bottom three quads of the source and destination addresses together, then xor'ing each byte of that result into the bottom byte, finally xor'ing with the last bytes of the MAC addresses) was selected after testing almost 400,000 unique IPv6 addresses harvested from server logs. This algorithm had the most even distribution for both big- and little-endian architectures while still using few instructions. Its behavior also attempts to closely match that of the IPv4 algorithm. The IPv6 flow label was intentionally not included in the hash as it appears to be unset in the vast majority of IPv6 traffic sampled, and the current algorithm not using the flow label already offers a very even distribution. Fragmented IPv6 packets are handled the same way as fragmented IPv4 packets, ie, they are not balanced based on layer 4 information. Additionally, IPv6 packets with intermediate headers are not balanced based on layer 4 information. In practice these intermediate headers are not common and this should not cause any problems, and the alternative (a packet-parsing loop and look-up table) seemed slow and complicated for little gain. Tested-by: John Eaglesham <linux@8192.net> Signed-off-by: John Eaglesham <linux@8192.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-08-22 03:43:35 +07:00
iph = ip_hdr(skb);
iph_to_flow_copy_v4addrs(fk, iph);
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
noff += iph->ihl << 2;
if (!ip_is_fragment(iph))
proto = iph->protocol;
} else if (skb->protocol == htons(ETH_P_IPV6)) {
if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
return false;
iph6 = ipv6_hdr(skb);
iph_to_flow_copy_v6addrs(fk, iph6);
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
noff += sizeof(*iph6);
proto = iph6->nexthdr;
} else {
return false;
}
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
return true;
}
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
/**
* bond_xmit_hash - generate a hash value based on the xmit policy
* @bond: bonding device
* @skb: buffer to use for headers
*
* This function will extract the necessary headers from the skb buffer and use
* them to generate a hash based on the xmit_policy set in the bonding device
*/
u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
{
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
struct flow_keys flow;
u32 hash;
if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
skb->l4_hash)
return skb->hash;
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
!bond_flow_dissect(bond, skb, &flow))
return bond_eth_hash(skb);
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
hash = bond_eth_hash(skb);
else
hash = (__force u32)flow.ports.ports;
hash ^= (__force u32)flow_get_u32_dst(&flow) ^
(__force u32)flow_get_u32_src(&flow);
bonding: modify the old and add new xmit hash policies This patch adds two new hash policy modes which use skb_flow_dissect: 3 - Encapsulated layer 2+3 4 - Encapsulated layer 3+4 There should be a good improvement for tunnel users in those modes. It also changes the old hash functions to: hash ^= (__force u32)flow.dst ^ (__force u32)flow.src; hash ^= (hash >> 16); hash ^= (hash >> 8); Where hash will be initialized either to L2 hash, that is SRCMAC[5] XOR DSTMAC[5], or to flow->ports which should be extracted from the upper layer. Flow's dst and src are also extracted based on the xmit policy either directly from the buffer or by using skb_flow_dissect, but in both cases if the protocol is IPv6 then dst and src are obtained by ipv6_addr_hash() on the real addresses. In case of a non-dissectable packet, the algorithms fall back to L2 hashing. The bond_set_mode_ops() function is now obsolete and thus deleted because it was used only to set the proper hash policy. Also we trim a pointer from struct bonding because we no longer need to keep the hash function, now there's only a single hash function - bond_xmit_hash that works based on bond->params.xmit_policy. The hash function and skb_flow_dissect were suggested by Eric Dumazet. The layer names were suggested by Andy Gospodarek, because I suck at semantics. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Acked-by: Eric Dumazet <edumazet@google.com> Acked-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-10-02 18:39:25 +07:00
hash ^= (hash >> 16);
hash ^= (hash >> 8);
return hash >> 1;
}
/*-------------------------- Device entry points ----------------------------*/
bonding: fix wq initialization for links created via netlink Earlier patch 4493b81bea ("bonding: initialize work-queues during creation of bond") moved the work-queue initialization from bond_open() to bond_create(). However this caused the link those are created using netlink 'create bond option' (ip link add bondX type bond); create the new trunk without initializing work-queues. Prior to the above mentioned change, ndo_open was in both paths and things worked correctly. The consequence is visible in the report shared by Joe Stringer - I've noticed that this patch breaks bonding within namespaces if you're not careful to perform device cleanup correctly. Here's my repro script, you can run on any net-next with this patch and you'll start seeing some weird behaviour: ip netns add foo ip li add veth0 type veth peer name veth0+ netns foo ip li add veth1 type veth peer name veth1+ netns foo ip netns exec foo ip li add bond0 type bond ip netns exec foo ip li set dev veth0+ master bond0 ip netns exec foo ip li set dev veth1+ master bond0 ip netns exec foo ip addr add dev bond0 192.168.0.1/24 ip netns exec foo ip li set dev bond0 up ip li del dev veth0 ip li del dev veth1 The second to last command segfaults, last command hangs. rtnl is now permanently locked. It's not a problem if you take bond0 down before deleting veths, or delete bond0 before deleting veths. If you delete either end of the veth pair as per above, either inside or outside the namespace, it hits this problem. Here's some kernel logs: [ 1221.801610] bond0: Enslaving veth0+ as an active interface with an up link [ 1224.449581] bond0: Enslaving veth1+ as an active interface with an up link [ 1281.193863] bond0: Releasing backup interface veth0+ [ 1281.193866] bond0: the permanent HWaddr of veth0+ - 16:bf:fb:e0:b8:43 - is still in use by bond0 - set the HWaddr of veth0+ to a different address to avoid conflicts [ 1281.193867] ------------[ cut here ]------------ [ 1281.193873] WARNING: CPU: 0 PID: 2024 at kernel/workqueue.c:1511 __queue_delayed_work+0x13f/0x150 [ 1281.193873] Modules linked in: bonding veth openvswitch nf_nat_ipv6 nf_nat_ipv4 nf_nat autofs4 nfsd auth_rpcgss nfs_acl binfmt_misc nfs lockd grace sunrpc fscache ppdev vmw_balloon coretemp psmouse serio_raw vmwgfx ttm drm_kms_helper vmw_vmci netconsole parport_pc configfs drm i2c_piix4 fb_sys_fops syscopyarea sysfillrect sysimgblt shpchp mac_hid nf_conntrack_ipv6 nf_defrag_ipv6 nf_conntrack_ipv4 nf_defrag_ipv4 nf_conntrack libcrc32c lp parport hid_generic usbhid hid mptspi mptscsih e1000 mptbase ahci libahci [ 1281.193905] CPU: 0 PID: 2024 Comm: ip Tainted: G W 4.10.0-bisect-bond-v0.14 #37 [ 1281.193906] Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 09/30/2014 [ 1281.193906] Call Trace: [ 1281.193912] dump_stack+0x63/0x89 [ 1281.193915] __warn+0xd1/0xf0 [ 1281.193917] warn_slowpath_null+0x1d/0x20 [ 1281.193918] __queue_delayed_work+0x13f/0x150 [ 1281.193920] queue_delayed_work_on+0x27/0x40 [ 1281.193929] bond_change_active_slave+0x25b/0x670 [bonding] [ 1281.193932] ? synchronize_rcu_expedited+0x27/0x30 [ 1281.193935] __bond_release_one+0x489/0x510 [bonding] [ 1281.193939] ? addrconf_notify+0x1b7/0xab0 [ 1281.193942] bond_netdev_event+0x2c5/0x2e0 [bonding] [ 1281.193944] ? netconsole_netdev_event+0x124/0x190 [netconsole] [ 1281.193947] notifier_call_chain+0x49/0x70 [ 1281.193948] raw_notifier_call_chain+0x16/0x20 [ 1281.193950] call_netdevice_notifiers_info+0x35/0x60 [ 1281.193951] rollback_registered_many+0x23b/0x3e0 [ 1281.193953] unregister_netdevice_many+0x24/0xd0 [ 1281.193955] rtnl_delete_link+0x3c/0x50 [ 1281.193956] rtnl_dellink+0x8d/0x1b0 [ 1281.193960] rtnetlink_rcv_msg+0x95/0x220 [ 1281.193962] ? __kmalloc_node_track_caller+0x35/0x280 [ 1281.193964] ? __netlink_lookup+0xf1/0x110 [ 1281.193966] ? rtnl_newlink+0x830/0x830 [ 1281.193967] netlink_rcv_skb+0xa7/0xc0 [ 1281.193969] rtnetlink_rcv+0x28/0x30 [ 1281.193970] netlink_unicast+0x15b/0x210 [ 1281.193971] netlink_sendmsg+0x319/0x390 [ 1281.193974] sock_sendmsg+0x38/0x50 [ 1281.193975] ___sys_sendmsg+0x25c/0x270 [ 1281.193978] ? mem_cgroup_commit_charge+0x76/0xf0 [ 1281.193981] ? page_add_new_anon_rmap+0x89/0xc0 [ 1281.193984] ? lru_cache_add_active_or_unevictable+0x35/0xb0 [ 1281.193985] ? __handle_mm_fault+0x4e9/0x1170 [ 1281.193987] __sys_sendmsg+0x45/0x80 [ 1281.193989] SyS_sendmsg+0x12/0x20 [ 1281.193991] do_syscall_64+0x6e/0x180 [ 1281.193993] entry_SYSCALL64_slow_path+0x25/0x25 [ 1281.193995] RIP: 0033:0x7f6ec122f5a0 [ 1281.193995] RSP: 002b:00007ffe69e89c48 EFLAGS: 00000246 ORIG_RAX: 000000000000002e [ 1281.193997] RAX: ffffffffffffffda RBX: 00007ffe69e8dd60 RCX: 00007f6ec122f5a0 [ 1281.193997] RDX: 0000000000000000 RSI: 00007ffe69e89c90 RDI: 0000000000000003 [ 1281.193998] RBP: 00007ffe69e89c90 R08: 0000000000000000 R09: 0000000000000003 [ 1281.193999] R10: 00007ffe69e89a10 R11: 0000000000000246 R12: 0000000058f14b9f [ 1281.193999] R13: 0000000000000000 R14: 00000000006473a0 R15: 00007ffe69e8e450 [ 1281.194001] ---[ end trace 713a77486cbfbfa3 ]--- Fixes: 4493b81bea ("bonding: initialize work-queues during creation of bond") Reported-by: Joe Stringer <joe@ovn.org> Tested-by: Joe Stringer <joe@ovn.org> Signed-off-by: Mahesh Bandewar <maheshb@google.com> Acked-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-21 02:49:24 +07:00
void bond_work_init_all(struct bonding *bond)
bonding: fix miimon and arp_interval delayed work race conditions First I would give three observations which will be used later. Observation 1: if (delayed_work_pending(wq)) cancel_delayed_work(wq) This usage is wrong because the pending bit is cleared just before the work's fn is executed and if the function re-arms itself we might end up with the work still running. It's safe to call cancel_delayed_work_sync() even if the work is not queued at all. Observation 2: Use of INIT_DELAYED_WORK() Work needs to be initialized only once prior to (de/en)queueing. Observation 3: IFF_UP is set only after ndo_open is called Related race conditions: 1. Race between bonding_store_miimon() and bonding_store_arp_interval() Because of Obs.1 we can end up having both works enqueued. 2. Multiple races with INIT_DELAYED_WORK() Since the works are not protected by anything between INIT_DELAYED_WORK() and calls to (en/de)queue it is possible for races between the following functions: (races are also possible between the calls to INIT_DELAYED_WORK() and workqueue code) bonding_store_miimon() - bonding_store_arp_interval(), bond_close(), bond_open(), enqueued functions bonding_store_arp_interval() - bonding_store_miimon(), bond_close(), bond_open(), enqueued functions 3. By Obs.1 we need to change bond_cancel_all() Bugs 1 and 2 are fixed by moving all work initializations in bond_open which by Obs. 2 and Obs. 3 and the fact that we make sure that all works are cancelled in bond_close(), is guaranteed not to have any work enqueued. Also RTNL lock is now acquired in bonding_store_miimon/arp_interval so they can't race with bond_close and bond_open. The opposing work is cancelled only if the IFF_UP flag is set and it is cancelled unconditionally. The opposing work is already cancelled if the interface is down so no need to cancel it again. This way we don't need new synchronizations for the bonding workqueue. These bugs (and fixes) are tied together and belong in the same patch. Note: I have left 1 line intentionally over 80 characters (84) because I didn't like how it looks broken down. If you'd prefer it otherwise, then simply break it. v2: Make description text < 75 columns Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-11-29 08:31:31 +07:00
{
INIT_DELAYED_WORK(&bond->mcast_work,
bond_resend_igmp_join_requests_delayed);
INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
bonding: fix miimon and arp_interval delayed work race conditions First I would give three observations which will be used later. Observation 1: if (delayed_work_pending(wq)) cancel_delayed_work(wq) This usage is wrong because the pending bit is cleared just before the work's fn is executed and if the function re-arms itself we might end up with the work still running. It's safe to call cancel_delayed_work_sync() even if the work is not queued at all. Observation 2: Use of INIT_DELAYED_WORK() Work needs to be initialized only once prior to (de/en)queueing. Observation 3: IFF_UP is set only after ndo_open is called Related race conditions: 1. Race between bonding_store_miimon() and bonding_store_arp_interval() Because of Obs.1 we can end up having both works enqueued. 2. Multiple races with INIT_DELAYED_WORK() Since the works are not protected by anything between INIT_DELAYED_WORK() and calls to (en/de)queue it is possible for races between the following functions: (races are also possible between the calls to INIT_DELAYED_WORK() and workqueue code) bonding_store_miimon() - bonding_store_arp_interval(), bond_close(), bond_open(), enqueued functions bonding_store_arp_interval() - bonding_store_miimon(), bond_close(), bond_open(), enqueued functions 3. By Obs.1 we need to change bond_cancel_all() Bugs 1 and 2 are fixed by moving all work initializations in bond_open which by Obs. 2 and Obs. 3 and the fact that we make sure that all works are cancelled in bond_close(), is guaranteed not to have any work enqueued. Also RTNL lock is now acquired in bonding_store_miimon/arp_interval so they can't race with bond_close and bond_open. The opposing work is cancelled only if the IFF_UP flag is set and it is cancelled unconditionally. The opposing work is already cancelled if the interface is down so no need to cancel it again. This way we don't need new synchronizations for the bonding workqueue. These bugs (and fixes) are tied together and belong in the same patch. Note: I have left 1 line intentionally over 80 characters (84) because I didn't like how it looks broken down. If you'd prefer it otherwise, then simply break it. v2: Make description text < 75 columns Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-11-29 08:31:31 +07:00
INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
bonding: fix miimon and arp_interval delayed work race conditions First I would give three observations which will be used later. Observation 1: if (delayed_work_pending(wq)) cancel_delayed_work(wq) This usage is wrong because the pending bit is cleared just before the work's fn is executed and if the function re-arms itself we might end up with the work still running. It's safe to call cancel_delayed_work_sync() even if the work is not queued at all. Observation 2: Use of INIT_DELAYED_WORK() Work needs to be initialized only once prior to (de/en)queueing. Observation 3: IFF_UP is set only after ndo_open is called Related race conditions: 1. Race between bonding_store_miimon() and bonding_store_arp_interval() Because of Obs.1 we can end up having both works enqueued. 2. Multiple races with INIT_DELAYED_WORK() Since the works are not protected by anything between INIT_DELAYED_WORK() and calls to (en/de)queue it is possible for races between the following functions: (races are also possible between the calls to INIT_DELAYED_WORK() and workqueue code) bonding_store_miimon() - bonding_store_arp_interval(), bond_close(), bond_open(), enqueued functions bonding_store_arp_interval() - bonding_store_miimon(), bond_close(), bond_open(), enqueued functions 3. By Obs.1 we need to change bond_cancel_all() Bugs 1 and 2 are fixed by moving all work initializations in bond_open which by Obs. 2 and Obs. 3 and the fact that we make sure that all works are cancelled in bond_close(), is guaranteed not to have any work enqueued. Also RTNL lock is now acquired in bonding_store_miimon/arp_interval so they can't race with bond_close and bond_open. The opposing work is cancelled only if the IFF_UP flag is set and it is cancelled unconditionally. The opposing work is already cancelled if the interface is down so no need to cancel it again. This way we don't need new synchronizations for the bonding workqueue. These bugs (and fixes) are tied together and belong in the same patch. Note: I have left 1 line intentionally over 80 characters (84) because I didn't like how it looks broken down. If you'd prefer it otherwise, then simply break it. v2: Make description text < 75 columns Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-11-29 08:31:31 +07:00
}
static void bond_work_cancel_all(struct bonding *bond)
{
cancel_delayed_work_sync(&bond->mii_work);
cancel_delayed_work_sync(&bond->arp_work);
cancel_delayed_work_sync(&bond->alb_work);
cancel_delayed_work_sync(&bond->ad_work);
cancel_delayed_work_sync(&bond->mcast_work);
cancel_delayed_work_sync(&bond->slave_arr_work);
bonding: fix miimon and arp_interval delayed work race conditions First I would give three observations which will be used later. Observation 1: if (delayed_work_pending(wq)) cancel_delayed_work(wq) This usage is wrong because the pending bit is cleared just before the work's fn is executed and if the function re-arms itself we might end up with the work still running. It's safe to call cancel_delayed_work_sync() even if the work is not queued at all. Observation 2: Use of INIT_DELAYED_WORK() Work needs to be initialized only once prior to (de/en)queueing. Observation 3: IFF_UP is set only after ndo_open is called Related race conditions: 1. Race between bonding_store_miimon() and bonding_store_arp_interval() Because of Obs.1 we can end up having both works enqueued. 2. Multiple races with INIT_DELAYED_WORK() Since the works are not protected by anything between INIT_DELAYED_WORK() and calls to (en/de)queue it is possible for races between the following functions: (races are also possible between the calls to INIT_DELAYED_WORK() and workqueue code) bonding_store_miimon() - bonding_store_arp_interval(), bond_close(), bond_open(), enqueued functions bonding_store_arp_interval() - bonding_store_miimon(), bond_close(), bond_open(), enqueued functions 3. By Obs.1 we need to change bond_cancel_all() Bugs 1 and 2 are fixed by moving all work initializations in bond_open which by Obs. 2 and Obs. 3 and the fact that we make sure that all works are cancelled in bond_close(), is guaranteed not to have any work enqueued. Also RTNL lock is now acquired in bonding_store_miimon/arp_interval so they can't race with bond_close and bond_open. The opposing work is cancelled only if the IFF_UP flag is set and it is cancelled unconditionally. The opposing work is already cancelled if the interface is down so no need to cancel it again. This way we don't need new synchronizations for the bonding workqueue. These bugs (and fixes) are tied together and belong in the same patch. Note: I have left 1 line intentionally over 80 characters (84) because I didn't like how it looks broken down. If you'd prefer it otherwise, then simply break it. v2: Make description text < 75 columns Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-11-29 08:31:31 +07:00
}
static int bond_open(struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
struct list_head *iter;
struct slave *slave;
/* reset slave->backup and slave->inactive */
if (bond_has_slaves(bond)) {
bond_for_each_slave(bond, slave, iter) {
if (bond_uses_primary(bond) &&
slave != rcu_access_pointer(bond->curr_active_slave)) {
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_inactive_flags(slave,
BOND_SLAVE_NOTIFY_NOW);
} else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
bonding: Fix RTNL: assertion failed at net/core/rtnetlink.c for 802.3ad mode The problem was introduced by the commit 1d3ee88ae0d (bonding: add netlink attributes to slave link dev). The bond_set_active_slave() and bond_set_backup_slave() will use rtmsg_ifinfo to send slave's states, so these two functions should be called in RTNL. In 802.3ad mode, acquiring RTNL for the __enable_port and __disable_port cases is difficult, as those calls generally already hold the state machine lock, and cannot unconditionally call rtnl_lock because either they already hold RTNL (for calls via bond_3ad_unbind_slave) or due to the potential for deadlock with bond_3ad_adapter_speed_changed, bond_3ad_adapter_duplex_changed, bond_3ad_link_change, or bond_3ad_update_lacp_rate. All four of those are called with RTNL held, and acquire the state machine lock second. The calling contexts for __enable_port and __disable_port already hold the state machine lock, and may or may not need RTNL. According to the Jay's opinion, I don't think it is a problem that the slave don't send notify message synchronously when the status changed, normally the state machine is running every 100 ms, send the notify message at the end of the state machine if the slave's state changed should be better. I fix the problem through these steps: 1). add a new function bond_set_slave_state() which could change the slave's state and call rtmsg_ifinfo() according to the input parameters called notify. 2). Add a new slave parameter which called should_notify, if the slave's state changed and don't notify yet, the parameter will be set to 1, and then if the slave's state changed again, the param will be set to 0, it indicate that the slave's state has been restored, no need to notify any one. 3). the __enable_port and __disable_port should not call rtmsg_ifinfo in the state machine lock, any change in the state of slave could set a flag in the slave, it will indicated that an rtmsg_ifinfo should be called at the end of the state machine. Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-02-26 10:05:22 +07:00
bond_set_slave_active_flags(slave,
BOND_SLAVE_NOTIFY_NOW);
}
}
}
if (bond_is_lb(bond)) {
/* bond_alb_initialize must be called before the timer
* is started.
*/
if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
return -ENOMEM;
if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
queue_delayed_work(bond->wq, &bond->alb_work, 0);
}
bonding: fix miimon and arp_interval delayed work race conditions First I would give three observations which will be used later. Observation 1: if (delayed_work_pending(wq)) cancel_delayed_work(wq) This usage is wrong because the pending bit is cleared just before the work's fn is executed and if the function re-arms itself we might end up with the work still running. It's safe to call cancel_delayed_work_sync() even if the work is not queued at all. Observation 2: Use of INIT_DELAYED_WORK() Work needs to be initialized only once prior to (de/en)queueing. Observation 3: IFF_UP is set only after ndo_open is called Related race conditions: 1. Race between bonding_store_miimon() and bonding_store_arp_interval() Because of Obs.1 we can end up having both works enqueued. 2. Multiple races with INIT_DELAYED_WORK() Since the works are not protected by anything between INIT_DELAYED_WORK() and calls to (en/de)queue it is possible for races between the following functions: (races are also possible between the calls to INIT_DELAYED_WORK() and workqueue code) bonding_store_miimon() - bonding_store_arp_interval(), bond_close(), bond_open(), enqueued functions bonding_store_arp_interval() - bonding_store_miimon(), bond_close(), bond_open(), enqueued functions 3. By Obs.1 we need to change bond_cancel_all() Bugs 1 and 2 are fixed by moving all work initializations in bond_open which by Obs. 2 and Obs. 3 and the fact that we make sure that all works are cancelled in bond_close(), is guaranteed not to have any work enqueued. Also RTNL lock is now acquired in bonding_store_miimon/arp_interval so they can't race with bond_close and bond_open. The opposing work is cancelled only if the IFF_UP flag is set and it is cancelled unconditionally. The opposing work is already cancelled if the interface is down so no need to cancel it again. This way we don't need new synchronizations for the bonding workqueue. These bugs (and fixes) are tied together and belong in the same patch. Note: I have left 1 line intentionally over 80 characters (84) because I didn't like how it looks broken down. If you'd prefer it otherwise, then simply break it. v2: Make description text < 75 columns Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-11-29 08:31:31 +07:00
if (bond->params.miimon) /* link check interval, in milliseconds. */
queue_delayed_work(bond->wq, &bond->mii_work, 0);
if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
queue_delayed_work(bond->wq, &bond->arp_work, 0);
bond->recv_probe = bond_arp_rcv;
}
if (BOND_MODE(bond) == BOND_MODE_8023AD) {
queue_delayed_work(bond->wq, &bond->ad_work, 0);
/* register to receive LACPDUs */
bond->recv_probe = bond_3ad_lacpdu_recv;
bond_3ad_initiate_agg_selection(bond, 1);
}
if (bond_mode_can_use_xmit_hash(bond))
bond_update_slave_arr(bond, NULL);
return 0;
}
static int bond_close(struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
bonding: fix miimon and arp_interval delayed work race conditions First I would give three observations which will be used later. Observation 1: if (delayed_work_pending(wq)) cancel_delayed_work(wq) This usage is wrong because the pending bit is cleared just before the work's fn is executed and if the function re-arms itself we might end up with the work still running. It's safe to call cancel_delayed_work_sync() even if the work is not queued at all. Observation 2: Use of INIT_DELAYED_WORK() Work needs to be initialized only once prior to (de/en)queueing. Observation 3: IFF_UP is set only after ndo_open is called Related race conditions: 1. Race between bonding_store_miimon() and bonding_store_arp_interval() Because of Obs.1 we can end up having both works enqueued. 2. Multiple races with INIT_DELAYED_WORK() Since the works are not protected by anything between INIT_DELAYED_WORK() and calls to (en/de)queue it is possible for races between the following functions: (races are also possible between the calls to INIT_DELAYED_WORK() and workqueue code) bonding_store_miimon() - bonding_store_arp_interval(), bond_close(), bond_open(), enqueued functions bonding_store_arp_interval() - bonding_store_miimon(), bond_close(), bond_open(), enqueued functions 3. By Obs.1 we need to change bond_cancel_all() Bugs 1 and 2 are fixed by moving all work initializations in bond_open which by Obs. 2 and Obs. 3 and the fact that we make sure that all works are cancelled in bond_close(), is guaranteed not to have any work enqueued. Also RTNL lock is now acquired in bonding_store_miimon/arp_interval so they can't race with bond_close and bond_open. The opposing work is cancelled only if the IFF_UP flag is set and it is cancelled unconditionally. The opposing work is already cancelled if the interface is down so no need to cancel it again. This way we don't need new synchronizations for the bonding workqueue. These bugs (and fixes) are tied together and belong in the same patch. Note: I have left 1 line intentionally over 80 characters (84) because I didn't like how it looks broken down. If you'd prefer it otherwise, then simply break it. v2: Make description text < 75 columns Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: Jay Vosburgh <fubar@us.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2012-11-29 08:31:31 +07:00
bond_work_cancel_all(bond);
bond->send_peer_notif = 0;
if (bond_is_lb(bond))
bond_alb_deinitialize(bond);
bond->recv_probe = NULL;
return 0;
}
/* fold stats, assuming all rtnl_link_stats64 fields are u64, but
* that some drivers can provide 32bit values only.
*/
static void bond_fold_stats(struct rtnl_link_stats64 *_res,
const struct rtnl_link_stats64 *_new,
const struct rtnl_link_stats64 *_old)
{
const u64 *new = (const u64 *)_new;
const u64 *old = (const u64 *)_old;
u64 *res = (u64 *)_res;
int i;
for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
u64 nv = new[i];
u64 ov = old[i];
s64 delta = nv - ov;
/* detects if this particular field is 32bit only */
if (((nv | ov) >> 32) == 0)
delta = (s64)(s32)((u32)nv - (u32)ov);
/* filter anomalies, some drivers reset their stats
* at down/up events.
*/
if (delta > 0)
res[i] += delta;
}
}
bonding: avoid lockdep confusion in bond_get_stats() syzbot found that the following sequence produces a LOCKDEP splat [1] ip link add bond10 type bond ip link add bond11 type bond ip link set bond11 master bond10 To fix this, we can use the already provided nest_level. This patch also provides correct nesting for dev->addr_list_lock [1] WARNING: possible recursive locking detected 4.18.0-rc6+ #167 Not tainted -------------------------------------------- syz-executor751/4439 is trying to acquire lock: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 but task is already holding lock: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 other info that might help us debug this: Possible unsafe locking scenario: CPU0 ---- lock(&(&bond->stats_lock)->rlock); lock(&(&bond->stats_lock)->rlock); *** DEADLOCK *** May be due to missing lock nesting notation 3 locks held by syz-executor751/4439: #0: (____ptrval____) (rtnl_mutex){+.+.}, at: rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77 #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 #2: (____ptrval____) (rcu_read_lock){....}, at: bond_get_stats+0x0/0x560 include/linux/compiler.h:215 stack backtrace: CPU: 0 PID: 4439 Comm: syz-executor751 Not tainted 4.18.0-rc6+ #167 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113 print_deadlock_bug kernel/locking/lockdep.c:1765 [inline] check_deadlock kernel/locking/lockdep.c:1809 [inline] validate_chain kernel/locking/lockdep.c:2405 [inline] __lock_acquire.cold.64+0x1fb/0x486 kernel/locking/lockdep.c:3435 lock_acquire+0x1e4/0x540 kernel/locking/lockdep.c:3924 __raw_spin_lock include/linux/spinlock_api_smp.h:142 [inline] _raw_spin_lock+0x2a/0x40 kernel/locking/spinlock.c:144 spin_lock include/linux/spinlock.h:310 [inline] bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 dev_get_stats+0x10f/0x470 net/core/dev.c:8316 bond_get_stats+0x232/0x560 drivers/net/bonding/bond_main.c:3432 dev_get_stats+0x10f/0x470 net/core/dev.c:8316 rtnl_fill_stats+0x4d/0xac0 net/core/rtnetlink.c:1169 rtnl_fill_ifinfo+0x1aa6/0x3fb0 net/core/rtnetlink.c:1611 rtmsg_ifinfo_build_skb+0xc8/0x190 net/core/rtnetlink.c:3268 rtmsg_ifinfo_event.part.30+0x45/0xe0 net/core/rtnetlink.c:3300 rtmsg_ifinfo_event net/core/rtnetlink.c:3297 [inline] rtnetlink_event+0x144/0x170 net/core/rtnetlink.c:4716 notifier_call_chain+0x180/0x390 kernel/notifier.c:93 __raw_notifier_call_chain kernel/notifier.c:394 [inline] raw_notifier_call_chain+0x2d/0x40 kernel/notifier.c:401 call_netdevice_notifiers_info+0x3f/0x90 net/core/dev.c:1735 call_netdevice_notifiers net/core/dev.c:1753 [inline] netdev_features_change net/core/dev.c:1321 [inline] netdev_change_features+0xb3/0x110 net/core/dev.c:7759 bond_compute_features.isra.47+0x585/0xa50 drivers/net/bonding/bond_main.c:1120 bond_enslave+0x1b25/0x5da0 drivers/net/bonding/bond_main.c:1755 bond_do_ioctl+0x7cb/0xae0 drivers/net/bonding/bond_main.c:3528 dev_ifsioc+0x43c/0xb30 net/core/dev_ioctl.c:327 dev_ioctl+0x1b5/0xcc0 net/core/dev_ioctl.c:493 sock_do_ioctl+0x1d3/0x3e0 net/socket.c:992 sock_ioctl+0x30d/0x680 net/socket.c:1093 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:500 [inline] do_vfs_ioctl+0x1de/0x1720 fs/ioctl.c:684 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701 __do_sys_ioctl fs/ioctl.c:708 [inline] __se_sys_ioctl fs/ioctl.c:706 [inline] __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x440859 Code: e8 2c af 02 00 48 83 c4 18 c3 0f 1f 80 00 00 00 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 3b 10 fc ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007ffc51a92878 EFLAGS: 00000213 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 0000000000440859 RDX: 0000000020000040 RSI: 0000000000008990 RDI: 0000000000000003 RBP: 0000000000000000 R08: 00000000004002c8 R09: 00000000004002c8 R10: 00000000022d5880 R11: 0000000000000213 R12: 0000000000007390 R13: 0000000000401db0 R14: 0000000000000000 R15: 0000000000000000 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Jay Vosburgh <j.vosburgh@gmail.com> Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-07-31 20:30:54 +07:00
static int bond_get_nest_level(struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
return bond->nest_level;
}
static void bond_get_stats(struct net_device *bond_dev,
struct rtnl_link_stats64 *stats)
{
struct bonding *bond = netdev_priv(bond_dev);
struct rtnl_link_stats64 temp;
struct list_head *iter;
struct slave *slave;
bonding: avoid lockdep confusion in bond_get_stats() syzbot found that the following sequence produces a LOCKDEP splat [1] ip link add bond10 type bond ip link add bond11 type bond ip link set bond11 master bond10 To fix this, we can use the already provided nest_level. This patch also provides correct nesting for dev->addr_list_lock [1] WARNING: possible recursive locking detected 4.18.0-rc6+ #167 Not tainted -------------------------------------------- syz-executor751/4439 is trying to acquire lock: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 but task is already holding lock: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 other info that might help us debug this: Possible unsafe locking scenario: CPU0 ---- lock(&(&bond->stats_lock)->rlock); lock(&(&bond->stats_lock)->rlock); *** DEADLOCK *** May be due to missing lock nesting notation 3 locks held by syz-executor751/4439: #0: (____ptrval____) (rtnl_mutex){+.+.}, at: rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77 #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 #2: (____ptrval____) (rcu_read_lock){....}, at: bond_get_stats+0x0/0x560 include/linux/compiler.h:215 stack backtrace: CPU: 0 PID: 4439 Comm: syz-executor751 Not tainted 4.18.0-rc6+ #167 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113 print_deadlock_bug kernel/locking/lockdep.c:1765 [inline] check_deadlock kernel/locking/lockdep.c:1809 [inline] validate_chain kernel/locking/lockdep.c:2405 [inline] __lock_acquire.cold.64+0x1fb/0x486 kernel/locking/lockdep.c:3435 lock_acquire+0x1e4/0x540 kernel/locking/lockdep.c:3924 __raw_spin_lock include/linux/spinlock_api_smp.h:142 [inline] _raw_spin_lock+0x2a/0x40 kernel/locking/spinlock.c:144 spin_lock include/linux/spinlock.h:310 [inline] bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 dev_get_stats+0x10f/0x470 net/core/dev.c:8316 bond_get_stats+0x232/0x560 drivers/net/bonding/bond_main.c:3432 dev_get_stats+0x10f/0x470 net/core/dev.c:8316 rtnl_fill_stats+0x4d/0xac0 net/core/rtnetlink.c:1169 rtnl_fill_ifinfo+0x1aa6/0x3fb0 net/core/rtnetlink.c:1611 rtmsg_ifinfo_build_skb+0xc8/0x190 net/core/rtnetlink.c:3268 rtmsg_ifinfo_event.part.30+0x45/0xe0 net/core/rtnetlink.c:3300 rtmsg_ifinfo_event net/core/rtnetlink.c:3297 [inline] rtnetlink_event+0x144/0x170 net/core/rtnetlink.c:4716 notifier_call_chain+0x180/0x390 kernel/notifier.c:93 __raw_notifier_call_chain kernel/notifier.c:394 [inline] raw_notifier_call_chain+0x2d/0x40 kernel/notifier.c:401 call_netdevice_notifiers_info+0x3f/0x90 net/core/dev.c:1735 call_netdevice_notifiers net/core/dev.c:1753 [inline] netdev_features_change net/core/dev.c:1321 [inline] netdev_change_features+0xb3/0x110 net/core/dev.c:7759 bond_compute_features.isra.47+0x585/0xa50 drivers/net/bonding/bond_main.c:1120 bond_enslave+0x1b25/0x5da0 drivers/net/bonding/bond_main.c:1755 bond_do_ioctl+0x7cb/0xae0 drivers/net/bonding/bond_main.c:3528 dev_ifsioc+0x43c/0xb30 net/core/dev_ioctl.c:327 dev_ioctl+0x1b5/0xcc0 net/core/dev_ioctl.c:493 sock_do_ioctl+0x1d3/0x3e0 net/socket.c:992 sock_ioctl+0x30d/0x680 net/socket.c:1093 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:500 [inline] do_vfs_ioctl+0x1de/0x1720 fs/ioctl.c:684 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701 __do_sys_ioctl fs/ioctl.c:708 [inline] __se_sys_ioctl fs/ioctl.c:706 [inline] __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x440859 Code: e8 2c af 02 00 48 83 c4 18 c3 0f 1f 80 00 00 00 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 3b 10 fc ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007ffc51a92878 EFLAGS: 00000213 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 0000000000440859 RDX: 0000000020000040 RSI: 0000000000008990 RDI: 0000000000000003 RBP: 0000000000000000 R08: 00000000004002c8 R09: 00000000004002c8 R10: 00000000022d5880 R11: 0000000000000213 R12: 0000000000007390 R13: 0000000000401db0 R14: 0000000000000000 R15: 0000000000000000 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Jay Vosburgh <j.vosburgh@gmail.com> Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-07-31 20:30:54 +07:00
spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
memcpy(stats, &bond->bond_stats, sizeof(*stats));
rcu_read_lock();
bond_for_each_slave_rcu(bond, slave, iter) {
const struct rtnl_link_stats64 *new =
dev_get_stats(slave->dev, &temp);
bond_fold_stats(stats, new, &slave->slave_stats);
/* save off the slave stats for the next run */
memcpy(&slave->slave_stats, new, sizeof(*new));
}
rcu_read_unlock();
memcpy(&bond->bond_stats, stats, sizeof(*stats));
spin_unlock(&bond->stats_lock);
}
static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
{
struct bonding *bond = netdev_priv(bond_dev);
struct net_device *slave_dev = NULL;
struct ifbond k_binfo;
struct ifbond __user *u_binfo = NULL;
struct ifslave k_sinfo;
struct ifslave __user *u_sinfo = NULL;
struct mii_ioctl_data *mii = NULL;
struct bond_opt_value newval;
struct net *net;
int res = 0;
netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
switch (cmd) {
case SIOCGMIIPHY:
mii = if_mii(ifr);
if (!mii)
return -EINVAL;
mii->phy_id = 0;
/* Fall Through */
case SIOCGMIIREG:
/* We do this again just in case we were called by SIOCGMIIREG
* instead of SIOCGMIIPHY.
*/
mii = if_mii(ifr);
if (!mii)
return -EINVAL;
if (mii->reg_num == 1) {
mii->val_out = 0;
if (netif_carrier_ok(bond->dev))
mii->val_out = BMSR_LSTATUS;
}
return 0;
case BOND_INFO_QUERY_OLD:
case SIOCBONDINFOQUERY:
u_binfo = (struct ifbond __user *)ifr->ifr_data;
if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
return -EFAULT;
bond_info_query(bond_dev, &k_binfo);
if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
return -EFAULT;
return 0;
case BOND_SLAVE_INFO_QUERY_OLD:
case SIOCBONDSLAVEINFOQUERY:
u_sinfo = (struct ifslave __user *)ifr->ifr_data;
if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
return -EFAULT;
res = bond_slave_info_query(bond_dev, &k_sinfo);
if (res == 0 &&
copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
return -EFAULT;
return res;
default:
break;
}
net = dev_net(bond_dev);
if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
return -EPERM;
slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
if (!slave_dev)
return -ENODEV;
netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
switch (cmd) {
case BOND_ENSLAVE_OLD:
case SIOCBONDENSLAVE:
res = bond_enslave(bond_dev, slave_dev, NULL);
break;
case BOND_RELEASE_OLD:
case SIOCBONDRELEASE:
res = bond_release(bond_dev, slave_dev);
break;
case BOND_SETHWADDR_OLD:
case SIOCBONDSETHWADDR:
bond_set_dev_addr(bond_dev, slave_dev);
res = 0;
break;
case BOND_CHANGE_ACTIVE_OLD:
case SIOCBONDCHANGEACTIVE:
bond_opt_initstr(&newval, slave_dev->name);
res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
&newval);
break;
default:
res = -EOPNOTSUPP;
}
return res;
}
static void bond_change_rx_flags(struct net_device *bond_dev, int change)
{
struct bonding *bond = netdev_priv(bond_dev);
if (change & IFF_PROMISC)
bond_set_promiscuity(bond,
bond_dev->flags & IFF_PROMISC ? 1 : -1);
if (change & IFF_ALLMULTI)
bond_set_allmulti(bond,
bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
}
static void bond_set_rx_mode(struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
struct list_head *iter;
struct slave *slave;
rcu_read_lock();
if (bond_uses_primary(bond)) {
slave = rcu_dereference(bond->curr_active_slave);
if (slave) {
dev_uc_sync(slave->dev, bond_dev);
dev_mc_sync(slave->dev, bond_dev);
}
} else {
bond_for_each_slave_rcu(bond, slave, iter) {
dev_uc_sync_multiple(slave->dev, bond_dev);
dev_mc_sync_multiple(slave->dev, bond_dev);
}
}
rcu_read_unlock();
}
static int bond_neigh_init(struct neighbour *n)
{
struct bonding *bond = netdev_priv(n->dev);
const struct net_device_ops *slave_ops;
struct neigh_parms parms;
struct slave *slave;
int ret;
slave = bond_first_slave(bond);
if (!slave)
return 0;
slave_ops = slave->dev->netdev_ops;
if (!slave_ops->ndo_neigh_setup)
return 0;
parms.neigh_setup = NULL;
parms.neigh_cleanup = NULL;
ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
if (ret)
return ret;
/* Assign slave's neigh_cleanup to neighbour in case cleanup is called
* after the last slave has been detached. Assumes that all slaves
* utilize the same neigh_cleanup (true at this writing as only user
* is ipoib).
*/
n->parms->neigh_cleanup = parms.neigh_cleanup;
if (!parms.neigh_setup)
return 0;
return parms.neigh_setup(n);
}
/* The bonding ndo_neigh_setup is called at init time beofre any
* slave exists. So we must declare proxy setup function which will
* be used at run time to resolve the actual slave neigh param setup.
*
* It's also called by master devices (such as vlans) to setup their
* underlying devices. In that case - do nothing, we're already set up from
* our init.
*/
static int bond_neigh_setup(struct net_device *dev,
struct neigh_parms *parms)
{
/* modify only our neigh_parms */
if (parms->dev == dev)
parms->neigh_setup = bond_neigh_init;
return 0;
}
/* Change the MTU of all of a master's slaves to match the master */
static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
{
struct bonding *bond = netdev_priv(bond_dev);
struct slave *slave, *rollback_slave;
struct list_head *iter;
int res = 0;
netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
bond_for_each_slave(bond, slave, iter) {
netdev_dbg(bond_dev, "s %p c_m %p\n",
slave, slave->dev->netdev_ops->ndo_change_mtu);
res = dev_set_mtu(slave->dev, new_mtu);
if (res) {
/* If we failed to set the slave's mtu to the new value
* we must abort the operation even in ACTIVE_BACKUP
* mode, because if we allow the backup slaves to have
* different mtu values than the active slave we'll
* need to change their mtu when doing a failover. That
* means changing their mtu from timer context, which
* is probably not a good idea.
*/
netdev_dbg(bond_dev, "err %d %s\n", res,
slave->dev->name);
goto unwind;
}
}
bond_dev->mtu = new_mtu;
return 0;
unwind:
/* unwind from head to the slave that failed */
bond_for_each_slave(bond, rollback_slave, iter) {
int tmp_res;
if (rollback_slave == slave)
break;
tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
if (tmp_res) {
netdev_dbg(bond_dev, "unwind err %d dev %s\n",
tmp_res, rollback_slave->dev->name);
}
}
return res;
}
/* Change HW address
*
* Note that many devices must be down to change the HW address, and
* downing the master releases all slaves. We can make bonds full of
* bonding devices to test this, however.
*/
static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
{
struct bonding *bond = netdev_priv(bond_dev);
struct slave *slave, *rollback_slave;
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
struct sockaddr_storage *ss = addr, tmp_ss;
struct list_head *iter;
int res = 0;
if (BOND_MODE(bond) == BOND_MODE_ALB)
return bond_alb_set_mac_address(bond_dev, addr);
netdev_dbg(bond_dev, "bond=%p\n", bond);
/* If fail_over_mac is enabled, do nothing and return success.
* Returning an error causes ifenslave to fail.
*/
if (bond->params.fail_over_mac &&
BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
return 0;
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
if (!is_valid_ether_addr(ss->__data))
return -EADDRNOTAVAIL;
bond_for_each_slave(bond, slave, iter) {
netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
res = dev_set_mac_address(slave->dev, addr);
if (res) {
/* TODO: consider downing the slave
* and retry ?
* User should expect communications
* breakage anyway until ARP finish
* updating, so...
*/
netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
goto unwind;
}
}
/* success */
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
return 0;
unwind:
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
tmp_ss.ss_family = bond_dev->type;
/* unwind from head to the slave that failed */
bond_for_each_slave(bond, rollback_slave, iter) {
int tmp_res;
if (rollback_slave == slave)
break;
bonding: attempt to better support longer hw addresses People are using bonding over Infiniband IPoIB connections, and who knows what else. Infiniband has a hardware address length of 20 octets (INFINIBAND_ALEN), and the network core defines a MAX_ADDR_LEN of 32. Various places in the bonding code are currently hard-wired to 6 octets (ETH_ALEN), such as the 3ad code, which I've left untouched here. Besides, only alb is currently possible on Infiniband links right now anyway, due to commit 1533e7731522, so the alb code is where most of the changes are. One major component of this change is the addition of a bond_hw_addr_copy function that takes a length argument, instead of using ether_addr_copy everywhere that hardware addresses need to be copied about. The other major component of this change is converting the bonding code from using struct sockaddr for address storage to struct sockaddr_storage, as the former has an address storage space of only 14, while the latter is 128 minus a few, which is necessary to support bonding over device with up to MAX_ADDR_LEN octet hardware addresses. Additionally, this probably fixes up some memory corruption issues with the current code, where it's possible to write an infiniband hardware address into a sockaddr declared on the stack. Lightly tested on a dual mlx4 IPoIB setup, which properly shows a 20-octet hardware address now: $ cat /proc/net/bonding/bond0 Ethernet Channel Bonding Driver: v3.7.1 (April 27, 2011) Bonding Mode: fault-tolerance (active-backup) (fail_over_mac active) Primary Slave: mlx4_ib0 (primary_reselect always) Currently Active Slave: mlx4_ib0 MII Status: up MII Polling Interval (ms): 100 Up Delay (ms): 100 Down Delay (ms): 100 Slave Interface: mlx4_ib0 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:08:fe:80:00:00:00:00:00:00:e4:1d:2d:03:00:1d:67:01 Slave queue ID: 0 Slave Interface: mlx4_ib1 MII Status: up Speed: Unknown Duplex: Unknown Link Failure Count: 0 Permanent HW addr: 80:00:02:09:fe:80:00:00:00:00:00:01:e4:1d:2d:03:00:1d:67:02 Slave queue ID: 0 Also tested with a standard 1Gbps NIC bonding setup (with a mix of e1000 and e1000e cards), running LNST's bonding tests. CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-04-05 04:32:42 +07:00
tmp_res = dev_set_mac_address(rollback_slave->dev,
(struct sockaddr *)&tmp_ss);
if (tmp_res) {
netdev_dbg(bond_dev, "unwind err %d dev %s\n",
tmp_res, rollback_slave->dev->name);
}
}
return res;
}
/**
* bond_xmit_slave_id - transmit skb through slave with slave_id
* @bond: bonding device that is transmitting
* @skb: buffer to transmit
* @slave_id: slave id up to slave_cnt-1 through which to transmit
*
* This function tries to transmit through slave with slave_id but in case
* it fails, it tries to find the first available slave for transmission.
* The skb is consumed in all cases, thus the function is void.
*/
static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
{
struct list_head *iter;
struct slave *slave;
int i = slave_id;
/* Here we start from the slave with slave_id */
bond_for_each_slave_rcu(bond, slave, iter) {
if (--i < 0) {
if (bond_slave_can_tx(slave)) {
bond_dev_queue_xmit(bond, skb, slave->dev);
return;
}
}
}
/* Here we start from the first slave up to slave_id */
i = slave_id;
bond_for_each_slave_rcu(bond, slave, iter) {
if (--i < 0)
break;
if (bond_slave_can_tx(slave)) {
bond_dev_queue_xmit(bond, skb, slave->dev);
return;
}
}
/* no slave that can tx has been found */
bond_tx_drop(bond->dev, skb);
}
/**
* bond_rr_gen_slave_id - generate slave id based on packets_per_slave
* @bond: bonding device to use
*
* Based on the value of the bonding device's packets_per_slave parameter
* this function generates a slave id, which is usually used as the next
* slave to transmit through.
*/
static u32 bond_rr_gen_slave_id(struct bonding *bond)
{
u32 slave_id;
reciprocal_divide: update/correction of the algorithm Jakub Zawadzki noticed that some divisions by reciprocal_divide() were not correct [1][2], which he could also show with BPF code after divisions are transformed into reciprocal_value() for runtime invariance which can be passed to reciprocal_divide() later on; reverse in BPF dump ended up with a different, off-by-one K in some situations. This has been fixed by Eric Dumazet in commit aee636c4809fa5 ("bpf: do not use reciprocal divide"). This follow-up patch improves reciprocal_value() and reciprocal_divide() to work in all cases by using Granlund and Montgomery method, so that also future use is safe and without any non-obvious side-effects. Known problems with the old implementation were that division by 1 always returned 0 and some off-by-ones when the dividend and divisor where very large. This seemed to not be problematic with its current users, as far as we can tell. Eric Dumazet checked for the slab usage, we cannot surely say so in the case of flex_array. Still, in order to fix that, we propose an extension from the original implementation from commit 6a2d7a955d8d resp. [3][4], by using the algorithm proposed in "Division by Invariant Integers Using Multiplication" [5], Torbjörn Granlund and Peter L. Montgomery, that is, pseudocode for q = n/d where q, n, d is in u32 universe: 1) Initialization: int l = ceil(log_2 d) uword m' = floor((1<<32)*((1<<l)-d)/d)+1 int sh_1 = min(l,1) int sh_2 = max(l-1,0) 2) For q = n/d, all uword: uword t = (n*m')>>32 q = (t+((n-t)>>sh_1))>>sh_2 The assembler implementation from Agner Fog [6] also helped a lot while implementing. We have tested the implementation on x86_64, ppc64, i686, s390x; on x86_64/haswell we're still half the latency compared to normal divide. Joint work with Daniel Borkmann. [1] http://www.wireshark.org/~darkjames/reciprocal-buggy.c [2] http://www.wireshark.org/~darkjames/set-and-dump-filter-k-bug.c [3] https://gmplib.org/~tege/division-paper.pdf [4] http://homepage.cs.uiowa.edu/~jones/bcd/divide.html [5] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.1.2556 [6] http://www.agner.org/optimize/asmlib.zip Reported-by: Jakub Zawadzki <darkjames-ws@darkjames.pl> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Austin S Hemmelgarn <ahferroin7@gmail.com> Cc: linux-kernel@vger.kernel.org Cc: Jesse Gross <jesse@nicira.com> Cc: Jamal Hadi Salim <jhs@mojatatu.com> Cc: Stephen Hemminger <stephen@networkplumber.org> Cc: Matt Mackall <mpm@selenic.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Andy Gospodarek <andy@greyhouse.net> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Jakub Zawadzki <darkjames-ws@darkjames.pl> Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-22 08:29:41 +07:00
struct reciprocal_value reciprocal_packets_per_slave;
int packets_per_slave = bond->params.packets_per_slave;
switch (packets_per_slave) {
case 0:
slave_id = prandom_u32();
break;
case 1:
slave_id = bond->rr_tx_counter;
break;
default:
reciprocal_divide: update/correction of the algorithm Jakub Zawadzki noticed that some divisions by reciprocal_divide() were not correct [1][2], which he could also show with BPF code after divisions are transformed into reciprocal_value() for runtime invariance which can be passed to reciprocal_divide() later on; reverse in BPF dump ended up with a different, off-by-one K in some situations. This has been fixed by Eric Dumazet in commit aee636c4809fa5 ("bpf: do not use reciprocal divide"). This follow-up patch improves reciprocal_value() and reciprocal_divide() to work in all cases by using Granlund and Montgomery method, so that also future use is safe and without any non-obvious side-effects. Known problems with the old implementation were that division by 1 always returned 0 and some off-by-ones when the dividend and divisor where very large. This seemed to not be problematic with its current users, as far as we can tell. Eric Dumazet checked for the slab usage, we cannot surely say so in the case of flex_array. Still, in order to fix that, we propose an extension from the original implementation from commit 6a2d7a955d8d resp. [3][4], by using the algorithm proposed in "Division by Invariant Integers Using Multiplication" [5], Torbjörn Granlund and Peter L. Montgomery, that is, pseudocode for q = n/d where q, n, d is in u32 universe: 1) Initialization: int l = ceil(log_2 d) uword m' = floor((1<<32)*((1<<l)-d)/d)+1 int sh_1 = min(l,1) int sh_2 = max(l-1,0) 2) For q = n/d, all uword: uword t = (n*m')>>32 q = (t+((n-t)>>sh_1))>>sh_2 The assembler implementation from Agner Fog [6] also helped a lot while implementing. We have tested the implementation on x86_64, ppc64, i686, s390x; on x86_64/haswell we're still half the latency compared to normal divide. Joint work with Daniel Borkmann. [1] http://www.wireshark.org/~darkjames/reciprocal-buggy.c [2] http://www.wireshark.org/~darkjames/set-and-dump-filter-k-bug.c [3] https://gmplib.org/~tege/division-paper.pdf [4] http://homepage.cs.uiowa.edu/~jones/bcd/divide.html [5] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.1.2556 [6] http://www.agner.org/optimize/asmlib.zip Reported-by: Jakub Zawadzki <darkjames-ws@darkjames.pl> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Austin S Hemmelgarn <ahferroin7@gmail.com> Cc: linux-kernel@vger.kernel.org Cc: Jesse Gross <jesse@nicira.com> Cc: Jamal Hadi Salim <jhs@mojatatu.com> Cc: Stephen Hemminger <stephen@networkplumber.org> Cc: Matt Mackall <mpm@selenic.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Andy Gospodarek <andy@greyhouse.net> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Jakub Zawadzki <darkjames-ws@darkjames.pl> Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-22 08:29:41 +07:00
reciprocal_packets_per_slave =
bond->params.reciprocal_packets_per_slave;
slave_id = reciprocal_divide(bond->rr_tx_counter,
reciprocal_divide: update/correction of the algorithm Jakub Zawadzki noticed that some divisions by reciprocal_divide() were not correct [1][2], which he could also show with BPF code after divisions are transformed into reciprocal_value() for runtime invariance which can be passed to reciprocal_divide() later on; reverse in BPF dump ended up with a different, off-by-one K in some situations. This has been fixed by Eric Dumazet in commit aee636c4809fa5 ("bpf: do not use reciprocal divide"). This follow-up patch improves reciprocal_value() and reciprocal_divide() to work in all cases by using Granlund and Montgomery method, so that also future use is safe and without any non-obvious side-effects. Known problems with the old implementation were that division by 1 always returned 0 and some off-by-ones when the dividend and divisor where very large. This seemed to not be problematic with its current users, as far as we can tell. Eric Dumazet checked for the slab usage, we cannot surely say so in the case of flex_array. Still, in order to fix that, we propose an extension from the original implementation from commit 6a2d7a955d8d resp. [3][4], by using the algorithm proposed in "Division by Invariant Integers Using Multiplication" [5], Torbjörn Granlund and Peter L. Montgomery, that is, pseudocode for q = n/d where q, n, d is in u32 universe: 1) Initialization: int l = ceil(log_2 d) uword m' = floor((1<<32)*((1<<l)-d)/d)+1 int sh_1 = min(l,1) int sh_2 = max(l-1,0) 2) For q = n/d, all uword: uword t = (n*m')>>32 q = (t+((n-t)>>sh_1))>>sh_2 The assembler implementation from Agner Fog [6] also helped a lot while implementing. We have tested the implementation on x86_64, ppc64, i686, s390x; on x86_64/haswell we're still half the latency compared to normal divide. Joint work with Daniel Borkmann. [1] http://www.wireshark.org/~darkjames/reciprocal-buggy.c [2] http://www.wireshark.org/~darkjames/set-and-dump-filter-k-bug.c [3] https://gmplib.org/~tege/division-paper.pdf [4] http://homepage.cs.uiowa.edu/~jones/bcd/divide.html [5] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.1.2556 [6] http://www.agner.org/optimize/asmlib.zip Reported-by: Jakub Zawadzki <darkjames-ws@darkjames.pl> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Austin S Hemmelgarn <ahferroin7@gmail.com> Cc: linux-kernel@vger.kernel.org Cc: Jesse Gross <jesse@nicira.com> Cc: Jamal Hadi Salim <jhs@mojatatu.com> Cc: Stephen Hemminger <stephen@networkplumber.org> Cc: Matt Mackall <mpm@selenic.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Andy Gospodarek <andy@greyhouse.net> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Jakub Zawadzki <darkjames-ws@darkjames.pl> Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-22 08:29:41 +07:00
reciprocal_packets_per_slave);
break;
}
bond->rr_tx_counter++;
return slave_id;
}
static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
struct iphdr *iph = ip_hdr(skb);
struct slave *slave;
u32 slave_id;
/* Start with the curr_active_slave that joined the bond as the
* default for sending IGMP traffic. For failover purposes one
* needs to maintain some consistency for the interface that will
* send the join/membership reports. The curr_active_slave found
* will send all of this type of traffic.
*/
if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
bonding: initial RCU conversion This patch does the initial bonding conversion to RCU. After it the following modes are protected by RCU alone: roundrobin, active-backup, broadcast and xor. Modes ALB/TLB and 3ad still acquire bond->lock for reading, and will be dealt with later. curr_active_slave needs to be dereferenced via rcu in the converted modes because the only thing protecting the slave after this patch is rcu_read_lock, so we need the proper barrier for weakly ordered archs and to make sure we don't have stale pointer. It's not tagged with __rcu yet because there's still work to be done to remove the curr_slave_lock, so sparse will complain when rcu_assign_pointer and rcu_dereference are used, but the alternative to use rcu_dereference_protected would've created much bigger code churn which is more difficult to test and review. That will be converted in time. 1. Active-backup mode 1.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.55% in bonding, system spent 0.29% CPU in bonding - new bonding: iperf spent 0.29% in bonding, system spent 0.15% CPU in bonding 1.2. Bandwidth measurements - old bonding: 16.1 gbps consistently - new bonding: 17.5 gbps consistently 2. Round-robin mode 2.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.51% in bonding, system spent 0.24% CPU in bonding - new bonding: iperf spent 0.16% in bonding, system spent 0.11% CPU in bonding 2.2 Bandwidth measurements - old bonding: 8 gbps (variable due to packet reorderings) - new bonding: 10 gbps (variable due to packet reorderings) Of course the latency has improved in all converted modes, and moreover while doing enslave/release (since it doesn't affect tx anymore). Also I've stress tested all modes doing enslave/release in a loop while transmitting traffic. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 21:54:51 +07:00
slave = rcu_dereference(bond->curr_active_slave);
if (slave)
bond_dev_queue_xmit(bond, skb, slave->dev);
else
bond_xmit_slave_id(bond, skb, 0);
} else {
locking/atomics: COCCINELLE/treewide: Convert trivial ACCESS_ONCE() patterns to READ_ONCE()/WRITE_ONCE() Please do not apply this to mainline directly, instead please re-run the coccinelle script shown below and apply its output. For several reasons, it is desirable to use {READ,WRITE}_ONCE() in preference to ACCESS_ONCE(), and new code is expected to use one of the former. So far, there's been no reason to change most existing uses of ACCESS_ONCE(), as these aren't harmful, and changing them results in churn. However, for some features, the read/write distinction is critical to correct operation. To distinguish these cases, separate read/write accessors must be used. This patch migrates (most) remaining ACCESS_ONCE() instances to {READ,WRITE}_ONCE(), using the following coccinelle script: ---- // Convert trivial ACCESS_ONCE() uses to equivalent READ_ONCE() and // WRITE_ONCE() // $ make coccicheck COCCI=/home/mark/once.cocci SPFLAGS="--include-headers" MODE=patch virtual patch @ depends on patch @ expression E1, E2; @@ - ACCESS_ONCE(E1) = E2 + WRITE_ONCE(E1, E2) @ depends on patch @ expression E; @@ - ACCESS_ONCE(E) + READ_ONCE(E) ---- Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: davem@davemloft.net Cc: linux-arch@vger.kernel.org Cc: mpe@ellerman.id.au Cc: shuah@kernel.org Cc: snitzer@redhat.com Cc: thor.thayer@linux.intel.com Cc: tj@kernel.org Cc: viro@zeniv.linux.org.uk Cc: will.deacon@arm.com Link: http://lkml.kernel.org/r/1508792849-3115-19-git-send-email-paulmck@linux.vnet.ibm.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-24 04:07:29 +07:00
int slave_cnt = READ_ONCE(bond->slave_cnt);
bonding: fix div by zero while enslaving and transmitting The problem is that the slave is first linked and slave_cnt is incremented afterwards leading to a div by zero in the modes that use it as a modulus. What happens is that in bond_start_xmit() bond_has_slaves() is used to evaluate further transmission and it becomes true after the slave is linked in, but when slave_cnt is used in the xmit path it is still 0, so fetch it once and transmit based on that. Since it is used only in round-robin and XOR modes, the fix is only for them. Thanks to Eric Dumazet for pointing out the fault in my first try to fix this. Call trace (took it out of net-next kernel, but it's the same with net): [46934.330038] divide error: 0000 [#1] SMP [46934.330041] Modules linked in: bonding(O) 9p fscache snd_hda_codec_generic crct10dif_pclmul [46934.330041] bond0: Enslaving eth1 as an active interface with an up link [46934.330051] ppdev joydev crc32_pclmul crc32c_intel 9pnet_virtio ghash_clmulni_intel snd_hda_intel 9pnet snd_hda_controller parport_pc serio_raw pcspkr snd_hda_codec parport virtio_balloon virtio_console snd_hwdep snd_pcm pvpanic i2c_piix4 snd_timer i2ccore snd soundcore virtio_blk virtio_net virtio_pci virtio_ring virtio ata_generic pata_acpi floppy [last unloaded: bonding] [46934.330053] CPU: 1 PID: 3382 Comm: ping Tainted: G O 3.17.0-rc4+ #27 [46934.330053] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011 [46934.330054] task: ffff88005aebf2c0 ti: ffff88005b728000 task.ti: ffff88005b728000 [46934.330059] RIP: 0010:[<ffffffffa0198c33>] [<ffffffffa0198c33>] bond_start_xmit+0x1c3/0x450 [bonding] [46934.330060] RSP: 0018:ffff88005b72b7f8 EFLAGS: 00010246 [46934.330060] RAX: 0000000000000679 RBX: ffff88004b077000 RCX: 000000000000002a [46934.330061] RDX: 0000000000000000 RSI: ffff88004b3f0500 RDI: ffff88004b077940 [46934.330061] RBP: ffff88005b72b830 R08: 00000000000000c0 R09: ffff88004a83e000 [46934.330062] R10: 000000000000ffff R11: ffff88004b1f12c0 R12: ffff88004b3f0500 [46934.330062] R13: ffff88004b3f0500 R14: 000000000000002a R15: ffff88004b077940 [46934.330063] FS: 00007fbd91a4c740(0000) GS:ffff88005f080000(0000) knlGS:0000000000000000 [46934.330064] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [46934.330064] CR2: 00007f803a8bb000 CR3: 000000004b2c9000 CR4: 00000000000406e0 [46934.330069] Stack: [46934.330071] ffffffff811e6169 00000000e772fa05 ffff88004b077000 ffff88004b3f0500 [46934.330072] ffffffff81d17d18 000000000000002a 0000000000000000 ffff88005b72b8a0 [46934.330073] ffffffff81620108 ffffffff8161fe0e ffff88005b72b8c4 ffff88005b302000 [46934.330073] Call Trace: [46934.330077] [<ffffffff811e6169>] ? __kmalloc_node_track_caller+0x119/0x300 [46934.330084] [<ffffffff81620108>] dev_hard_start_xmit+0x188/0x410 [46934.330086] [<ffffffff8161fe0e>] ? harmonize_features+0x2e/0x90 [46934.330088] [<ffffffff81620b06>] __dev_queue_xmit+0x456/0x590 [46934.330089] [<ffffffff81620c50>] dev_queue_xmit+0x10/0x20 [46934.330090] [<ffffffff8168f022>] arp_xmit+0x22/0x60 [46934.330091] [<ffffffff8168f090>] arp_send.part.16+0x30/0x40 [46934.330092] [<ffffffff8168f1e5>] arp_solicit+0x115/0x2b0 [46934.330094] [<ffffffff8160b5d7>] ? copy_skb_header+0x17/0xa0 [46934.330096] [<ffffffff8162875a>] neigh_probe+0x4a/0x70 [46934.330097] [<ffffffff8162979c>] __neigh_event_send+0xac/0x230 [46934.330098] [<ffffffff8162a00b>] neigh_resolve_output+0x13b/0x220 [46934.330100] [<ffffffff8165f120>] ? ip_forward_options+0x1c0/0x1c0 [46934.330101] [<ffffffff81660478>] ip_finish_output+0x1f8/0x860 [46934.330102] [<ffffffff81661f08>] ip_output+0x58/0x90 [46934.330103] [<ffffffff81661602>] ? __ip_local_out+0xa2/0xb0 [46934.330104] [<ffffffff81661640>] ip_local_out_sk+0x30/0x40 [46934.330105] [<ffffffff81662a66>] ip_send_skb+0x16/0x50 [46934.330106] [<ffffffff81662ad3>] ip_push_pending_frames+0x33/0x40 [46934.330107] [<ffffffff8168854c>] raw_sendmsg+0x88c/0xa30 [46934.330110] [<ffffffff81612b31>] ? skb_recv_datagram+0x41/0x60 [46934.330111] [<ffffffff816875a9>] ? raw_recvmsg+0xa9/0x1f0 [46934.330113] [<ffffffff816978d4>] inet_sendmsg+0x74/0xc0 [46934.330114] [<ffffffff81697a9b>] ? inet_recvmsg+0x8b/0xb0 [46934.330115] bond0: Adding slave eth2 [46934.330116] [<ffffffff8160357c>] sock_sendmsg+0x9c/0xe0 [46934.330118] [<ffffffff81603248>] ? move_addr_to_kernel.part.20+0x28/0x80 [46934.330121] [<ffffffff811b4477>] ? might_fault+0x47/0x50 [46934.330122] [<ffffffff816039b9>] ___sys_sendmsg+0x3a9/0x3c0 [46934.330125] [<ffffffff8144a14a>] ? n_tty_write+0x3aa/0x530 [46934.330127] [<ffffffff810d1ae4>] ? __wake_up+0x44/0x50 [46934.330129] [<ffffffff81242b38>] ? fsnotify+0x238/0x310 [46934.330130] [<ffffffff816048a1>] __sys_sendmsg+0x51/0x90 [46934.330131] [<ffffffff816048f2>] SyS_sendmsg+0x12/0x20 [46934.330134] [<ffffffff81738b29>] system_call_fastpath+0x16/0x1b [46934.330144] Code: 48 8b 10 4c 89 ee 4c 89 ff e8 aa bc ff ff 31 c0 e9 1a ff ff ff 0f 1f 00 4c 89 ee 4c 89 ff e8 65 fb ff ff 31 d2 4c 89 ee 4c 89 ff <f7> b3 64 09 00 00 e8 02 bd ff ff 31 c0 e9 f2 fe ff ff 0f 1f 00 [46934.330146] RIP [<ffffffffa0198c33>] bond_start_xmit+0x1c3/0x450 [bonding] [46934.330146] RSP <ffff88005b72b7f8> CC: Eric Dumazet <eric.dumazet@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> Fixes: 278b208375 ("bonding: initial RCU conversion") Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-12 22:38:18 +07:00
if (likely(slave_cnt)) {
slave_id = bond_rr_gen_slave_id(bond);
bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
} else {
bond_tx_drop(bond_dev, skb);
bonding: fix div by zero while enslaving and transmitting The problem is that the slave is first linked and slave_cnt is incremented afterwards leading to a div by zero in the modes that use it as a modulus. What happens is that in bond_start_xmit() bond_has_slaves() is used to evaluate further transmission and it becomes true after the slave is linked in, but when slave_cnt is used in the xmit path it is still 0, so fetch it once and transmit based on that. Since it is used only in round-robin and XOR modes, the fix is only for them. Thanks to Eric Dumazet for pointing out the fault in my first try to fix this. Call trace (took it out of net-next kernel, but it's the same with net): [46934.330038] divide error: 0000 [#1] SMP [46934.330041] Modules linked in: bonding(O) 9p fscache snd_hda_codec_generic crct10dif_pclmul [46934.330041] bond0: Enslaving eth1 as an active interface with an up link [46934.330051] ppdev joydev crc32_pclmul crc32c_intel 9pnet_virtio ghash_clmulni_intel snd_hda_intel 9pnet snd_hda_controller parport_pc serio_raw pcspkr snd_hda_codec parport virtio_balloon virtio_console snd_hwdep snd_pcm pvpanic i2c_piix4 snd_timer i2ccore snd soundcore virtio_blk virtio_net virtio_pci virtio_ring virtio ata_generic pata_acpi floppy [last unloaded: bonding] [46934.330053] CPU: 1 PID: 3382 Comm: ping Tainted: G O 3.17.0-rc4+ #27 [46934.330053] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011 [46934.330054] task: ffff88005aebf2c0 ti: ffff88005b728000 task.ti: ffff88005b728000 [46934.330059] RIP: 0010:[<ffffffffa0198c33>] [<ffffffffa0198c33>] bond_start_xmit+0x1c3/0x450 [bonding] [46934.330060] RSP: 0018:ffff88005b72b7f8 EFLAGS: 00010246 [46934.330060] RAX: 0000000000000679 RBX: ffff88004b077000 RCX: 000000000000002a [46934.330061] RDX: 0000000000000000 RSI: ffff88004b3f0500 RDI: ffff88004b077940 [46934.330061] RBP: ffff88005b72b830 R08: 00000000000000c0 R09: ffff88004a83e000 [46934.330062] R10: 000000000000ffff R11: ffff88004b1f12c0 R12: ffff88004b3f0500 [46934.330062] R13: ffff88004b3f0500 R14: 000000000000002a R15: ffff88004b077940 [46934.330063] FS: 00007fbd91a4c740(0000) GS:ffff88005f080000(0000) knlGS:0000000000000000 [46934.330064] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [46934.330064] CR2: 00007f803a8bb000 CR3: 000000004b2c9000 CR4: 00000000000406e0 [46934.330069] Stack: [46934.330071] ffffffff811e6169 00000000e772fa05 ffff88004b077000 ffff88004b3f0500 [46934.330072] ffffffff81d17d18 000000000000002a 0000000000000000 ffff88005b72b8a0 [46934.330073] ffffffff81620108 ffffffff8161fe0e ffff88005b72b8c4 ffff88005b302000 [46934.330073] Call Trace: [46934.330077] [<ffffffff811e6169>] ? __kmalloc_node_track_caller+0x119/0x300 [46934.330084] [<ffffffff81620108>] dev_hard_start_xmit+0x188/0x410 [46934.330086] [<ffffffff8161fe0e>] ? harmonize_features+0x2e/0x90 [46934.330088] [<ffffffff81620b06>] __dev_queue_xmit+0x456/0x590 [46934.330089] [<ffffffff81620c50>] dev_queue_xmit+0x10/0x20 [46934.330090] [<ffffffff8168f022>] arp_xmit+0x22/0x60 [46934.330091] [<ffffffff8168f090>] arp_send.part.16+0x30/0x40 [46934.330092] [<ffffffff8168f1e5>] arp_solicit+0x115/0x2b0 [46934.330094] [<ffffffff8160b5d7>] ? copy_skb_header+0x17/0xa0 [46934.330096] [<ffffffff8162875a>] neigh_probe+0x4a/0x70 [46934.330097] [<ffffffff8162979c>] __neigh_event_send+0xac/0x230 [46934.330098] [<ffffffff8162a00b>] neigh_resolve_output+0x13b/0x220 [46934.330100] [<ffffffff8165f120>] ? ip_forward_options+0x1c0/0x1c0 [46934.330101] [<ffffffff81660478>] ip_finish_output+0x1f8/0x860 [46934.330102] [<ffffffff81661f08>] ip_output+0x58/0x90 [46934.330103] [<ffffffff81661602>] ? __ip_local_out+0xa2/0xb0 [46934.330104] [<ffffffff81661640>] ip_local_out_sk+0x30/0x40 [46934.330105] [<ffffffff81662a66>] ip_send_skb+0x16/0x50 [46934.330106] [<ffffffff81662ad3>] ip_push_pending_frames+0x33/0x40 [46934.330107] [<ffffffff8168854c>] raw_sendmsg+0x88c/0xa30 [46934.330110] [<ffffffff81612b31>] ? skb_recv_datagram+0x41/0x60 [46934.330111] [<ffffffff816875a9>] ? raw_recvmsg+0xa9/0x1f0 [46934.330113] [<ffffffff816978d4>] inet_sendmsg+0x74/0xc0 [46934.330114] [<ffffffff81697a9b>] ? inet_recvmsg+0x8b/0xb0 [46934.330115] bond0: Adding slave eth2 [46934.330116] [<ffffffff8160357c>] sock_sendmsg+0x9c/0xe0 [46934.330118] [<ffffffff81603248>] ? move_addr_to_kernel.part.20+0x28/0x80 [46934.330121] [<ffffffff811b4477>] ? might_fault+0x47/0x50 [46934.330122] [<ffffffff816039b9>] ___sys_sendmsg+0x3a9/0x3c0 [46934.330125] [<ffffffff8144a14a>] ? n_tty_write+0x3aa/0x530 [46934.330127] [<ffffffff810d1ae4>] ? __wake_up+0x44/0x50 [46934.330129] [<ffffffff81242b38>] ? fsnotify+0x238/0x310 [46934.330130] [<ffffffff816048a1>] __sys_sendmsg+0x51/0x90 [46934.330131] [<ffffffff816048f2>] SyS_sendmsg+0x12/0x20 [46934.330134] [<ffffffff81738b29>] system_call_fastpath+0x16/0x1b [46934.330144] Code: 48 8b 10 4c 89 ee 4c 89 ff e8 aa bc ff ff 31 c0 e9 1a ff ff ff 0f 1f 00 4c 89 ee 4c 89 ff e8 65 fb ff ff 31 d2 4c 89 ee 4c 89 ff <f7> b3 64 09 00 00 e8 02 bd ff ff 31 c0 e9 f2 fe ff ff 0f 1f 00 [46934.330146] RIP [<ffffffffa0198c33>] bond_start_xmit+0x1c3/0x450 [bonding] [46934.330146] RSP <ffff88005b72b7f8> CC: Eric Dumazet <eric.dumazet@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> Fixes: 278b208375 ("bonding: initial RCU conversion") Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-09-12 22:38:18 +07:00
}
}
return NETDEV_TX_OK;
}
/* In active-backup mode, we know that bond->curr_active_slave is always valid if
* the bond has a usable interface.
*/
static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
struct slave *slave;
bonding: initial RCU conversion This patch does the initial bonding conversion to RCU. After it the following modes are protected by RCU alone: roundrobin, active-backup, broadcast and xor. Modes ALB/TLB and 3ad still acquire bond->lock for reading, and will be dealt with later. curr_active_slave needs to be dereferenced via rcu in the converted modes because the only thing protecting the slave after this patch is rcu_read_lock, so we need the proper barrier for weakly ordered archs and to make sure we don't have stale pointer. It's not tagged with __rcu yet because there's still work to be done to remove the curr_slave_lock, so sparse will complain when rcu_assign_pointer and rcu_dereference are used, but the alternative to use rcu_dereference_protected would've created much bigger code churn which is more difficult to test and review. That will be converted in time. 1. Active-backup mode 1.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.55% in bonding, system spent 0.29% CPU in bonding - new bonding: iperf spent 0.29% in bonding, system spent 0.15% CPU in bonding 1.2. Bandwidth measurements - old bonding: 16.1 gbps consistently - new bonding: 17.5 gbps consistently 2. Round-robin mode 2.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.51% in bonding, system spent 0.24% CPU in bonding - new bonding: iperf spent 0.16% in bonding, system spent 0.11% CPU in bonding 2.2 Bandwidth measurements - old bonding: 8 gbps (variable due to packet reorderings) - new bonding: 10 gbps (variable due to packet reorderings) Of course the latency has improved in all converted modes, and moreover while doing enslave/release (since it doesn't affect tx anymore). Also I've stress tested all modes doing enslave/release in a loop while transmitting traffic. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 21:54:51 +07:00
slave = rcu_dereference(bond->curr_active_slave);
if (slave)
bond_dev_queue_xmit(bond, skb, slave->dev);
else
bond_tx_drop(bond_dev, skb);
return NETDEV_TX_OK;
}
/* Use this to update slave_array when (a) it's not appropriate to update
* slave_array right away (note that update_slave_array() may sleep)
* and / or (b) RTNL is not held.
*/
void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
{
queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
}
/* Slave array work handler. Holds only RTNL */
static void bond_slave_arr_handler(struct work_struct *work)
{
struct bonding *bond = container_of(work, struct bonding,
slave_arr_work.work);
int ret;
if (!rtnl_trylock())
goto err;
ret = bond_update_slave_arr(bond, NULL);
rtnl_unlock();
if (ret) {
pr_warn_ratelimited("Failed to update slave array from WT\n");
goto err;
}
return;
err:
bond_slave_arr_work_rearm(bond, 1);
}
/* Build the usable slaves array in control path for modes that use xmit-hash
* to determine the slave interface -
* (a) BOND_MODE_8023AD
* (b) BOND_MODE_XOR
* (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
*
* The caller is expected to hold RTNL only and NO other lock!
*/
int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
{
struct slave *slave;
struct list_head *iter;
struct bond_up_slave *new_arr, *old_arr;
int agg_id = 0;
int ret = 0;
#ifdef CONFIG_LOCKDEP
WARN_ON(lockdep_is_held(&bond->mode_lock));
#endif
new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
GFP_KERNEL);
if (!new_arr) {
ret = -ENOMEM;
pr_err("Failed to build slave-array.\n");
goto out;
}
if (BOND_MODE(bond) == BOND_MODE_8023AD) {
struct ad_info ad_info;
if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
pr_debug("bond_3ad_get_active_agg_info failed\n");
kfree_rcu(new_arr, rcu);
/* No active aggragator means it's not safe to use
* the previous array.
*/
old_arr = rtnl_dereference(bond->slave_arr);
if (old_arr) {
RCU_INIT_POINTER(bond->slave_arr, NULL);
kfree_rcu(old_arr, rcu);
}
goto out;
}
agg_id = ad_info.aggregator_id;
}
bond_for_each_slave(bond, slave, iter) {
if (BOND_MODE(bond) == BOND_MODE_8023AD) {
struct aggregator *agg;
agg = SLAVE_AD_INFO(slave)->port.aggregator;
if (!agg || agg->aggregator_identifier != agg_id)
continue;
}
if (!bond_slave_can_tx(slave))
continue;
if (skipslave == slave)
continue;
netdev_dbg(bond->dev,
"Adding slave dev %s to tx hash array[%d]\n",
slave->dev->name, new_arr->count);
new_arr->arr[new_arr->count++] = slave;
}
old_arr = rtnl_dereference(bond->slave_arr);
rcu_assign_pointer(bond->slave_arr, new_arr);
if (old_arr)
kfree_rcu(old_arr, rcu);
out:
if (ret != 0 && skipslave) {
int idx;
/* Rare situation where caller has asked to skip a specific
* slave but allocation failed (most likely!). BTW this is
* only possible when the call is initiated from
* __bond_release_one(). In this situation; overwrite the
* skipslave entry in the array with the last entry from the
* array to avoid a situation where the xmit path may choose
* this to-be-skipped slave to send a packet out.
*/
old_arr = rtnl_dereference(bond->slave_arr);
for (idx = 0; idx < old_arr->count; idx++) {
if (skipslave == old_arr->arr[idx]) {
old_arr->arr[idx] =
old_arr->arr[old_arr->count-1];
old_arr->count--;
break;
}
}
}
return ret;
}
/* Use this Xmit function for 3AD as well as XOR modes. The current
* usable slave array is formed in the control path. The xmit function
* just calculates hash and sends the packet out.
*/
static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
struct net_device *dev)
{
struct bonding *bond = netdev_priv(dev);
struct slave *slave;
struct bond_up_slave *slaves;
unsigned int count;
slaves = rcu_dereference(bond->slave_arr);
locking/atomics: COCCINELLE/treewide: Convert trivial ACCESS_ONCE() patterns to READ_ONCE()/WRITE_ONCE() Please do not apply this to mainline directly, instead please re-run the coccinelle script shown below and apply its output. For several reasons, it is desirable to use {READ,WRITE}_ONCE() in preference to ACCESS_ONCE(), and new code is expected to use one of the former. So far, there's been no reason to change most existing uses of ACCESS_ONCE(), as these aren't harmful, and changing them results in churn. However, for some features, the read/write distinction is critical to correct operation. To distinguish these cases, separate read/write accessors must be used. This patch migrates (most) remaining ACCESS_ONCE() instances to {READ,WRITE}_ONCE(), using the following coccinelle script: ---- // Convert trivial ACCESS_ONCE() uses to equivalent READ_ONCE() and // WRITE_ONCE() // $ make coccicheck COCCI=/home/mark/once.cocci SPFLAGS="--include-headers" MODE=patch virtual patch @ depends on patch @ expression E1, E2; @@ - ACCESS_ONCE(E1) = E2 + WRITE_ONCE(E1, E2) @ depends on patch @ expression E; @@ - ACCESS_ONCE(E) + READ_ONCE(E) ---- Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: davem@davemloft.net Cc: linux-arch@vger.kernel.org Cc: mpe@ellerman.id.au Cc: shuah@kernel.org Cc: snitzer@redhat.com Cc: thor.thayer@linux.intel.com Cc: tj@kernel.org Cc: viro@zeniv.linux.org.uk Cc: will.deacon@arm.com Link: http://lkml.kernel.org/r/1508792849-3115-19-git-send-email-paulmck@linux.vnet.ibm.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-24 04:07:29 +07:00
count = slaves ? READ_ONCE(slaves->count) : 0;
if (likely(count)) {
slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
bond_dev_queue_xmit(bond, skb, slave->dev);
} else {
bond_tx_drop(dev, skb);
}
return NETDEV_TX_OK;
}
/* in broadcast mode, we send everything to all usable interfaces. */
static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
struct slave *slave = NULL;
struct list_head *iter;
bond_for_each_slave_rcu(bond, slave, iter) {
if (bond_is_last_slave(bond, slave))
break;
if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
if (!skb2) {
net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
bond_dev->name, __func__);
continue;
}
bond_dev_queue_xmit(bond, skb2, slave->dev);
}
}
if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
bond_dev_queue_xmit(bond, skb, slave->dev);
else
bond_tx_drop(bond_dev, skb);
return NETDEV_TX_OK;
}
/*------------------------- Device initialization ---------------------------*/
/* Lookup the slave that corresponds to a qid */
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
static inline int bond_slave_override(struct bonding *bond,
struct sk_buff *skb)
{
struct slave *slave = NULL;
struct list_head *iter;
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
if (!skb_rx_queue_recorded(skb))
return 1;
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
/* Find out if any slaves have the same mapping as this skb. */
bond_for_each_slave_rcu(bond, slave, iter) {
if (slave->queue_id == skb_get_queue_mapping(skb)) {
if (bond_slave_is_up(slave) &&
slave->link == BOND_LINK_UP) {
bond_dev_queue_xmit(bond, skb, slave->dev);
return 0;
}
/* If the slave isn't UP, use default transmit policy. */
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
break;
}
}
return 1;
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
}
net: core: explicitly select a txq before doing l2 forwarding Currently, the tx queue were selected implicitly in ndo_dfwd_start_xmit(). The will cause several issues: - NETIF_F_LLTX were removed for macvlan, so txq lock were done for macvlan instead of lower device which misses the necessary txq synchronization for lower device such as txq stopping or frozen required by dev watchdog or control path. - dev_hard_start_xmit() was called with NULL txq which bypasses the net device watchdog. - dev_hard_start_xmit() does not check txq everywhere which will lead a crash when tso is disabled for lower device. Fix this by explicitly introducing a new param for .ndo_select_queue() for just selecting queues in the case of l2 forwarding offload. netdev_pick_tx() was also extended to accept this parameter and dev_queue_xmit_accel() was used to do l2 forwarding transmission. With this fixes, NETIF_F_LLTX could be preserved for macvlan and there's no need to check txq against NULL in dev_hard_start_xmit(). Also there's no need to keep a dedicated ndo_dfwd_start_xmit() and we can just reuse the code of dev_queue_xmit() to do the transmission. In the future, it was also required for macvtap l2 forwarding support since it provides a necessary synchronization method. Cc: John Fastabend <john.r.fastabend@intel.com> Cc: Neil Horman <nhorman@tuxdriver.com> Cc: e1000-devel@lists.sourceforge.net Signed-off-by: Jason Wang <jasowang@redhat.com> Acked-by: Neil Horman <nhorman@tuxdriver.com> Acked-by: John Fastabend <john.r.fastabend@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-10 15:18:26 +07:00
static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
struct net_device *sb_dev,
select_queue_fallback_t fallback)
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
{
/* This helper function exists to help dev_pick_tx get the correct
* destination queue. Using a helper function skips a call to
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
* skb_tx_hash and will put the skbs in the queue we expect on their
* way down to the bonding driver.
*/
u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
/* Save the original txq to restore before passing to the driver */
qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
if (unlikely(txq >= dev->real_num_tx_queues)) {
do {
txq -= dev->real_num_tx_queues;
} while (txq >= dev->real_num_tx_queues);
}
return txq;
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
}
static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
{
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
struct bonding *bond = netdev_priv(dev);
if (bond_should_override_tx_queue(bond) &&
!bond_slave_override(bond, skb))
return NETDEV_TX_OK;
switch (BOND_MODE(bond)) {
case BOND_MODE_ROUNDROBIN:
return bond_xmit_roundrobin(skb, dev);
case BOND_MODE_ACTIVEBACKUP:
return bond_xmit_activebackup(skb, dev);
case BOND_MODE_8023AD:
case BOND_MODE_XOR:
return bond_3ad_xor_xmit(skb, dev);
case BOND_MODE_BROADCAST:
return bond_xmit_broadcast(skb, dev);
case BOND_MODE_ALB:
return bond_alb_xmit(skb, dev);
case BOND_MODE_TLB:
return bond_tlb_xmit(skb, dev);
default:
/* Should never happen, mode already checked */
netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
WARN_ON_ONCE(1);
bond_tx_drop(dev, skb);
return NETDEV_TX_OK;
}
}
static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
{
struct bonding *bond = netdev_priv(dev);
netdev_tx_t ret = NETDEV_TX_OK;
/* If we risk deadlock from transmitting this in the
* netpoll path, tell netpoll to queue the frame for later tx
*/
if (unlikely(is_netpoll_tx_blocked(dev)))
return NETDEV_TX_BUSY;
bonding: initial RCU conversion This patch does the initial bonding conversion to RCU. After it the following modes are protected by RCU alone: roundrobin, active-backup, broadcast and xor. Modes ALB/TLB and 3ad still acquire bond->lock for reading, and will be dealt with later. curr_active_slave needs to be dereferenced via rcu in the converted modes because the only thing protecting the slave after this patch is rcu_read_lock, so we need the proper barrier for weakly ordered archs and to make sure we don't have stale pointer. It's not tagged with __rcu yet because there's still work to be done to remove the curr_slave_lock, so sparse will complain when rcu_assign_pointer and rcu_dereference are used, but the alternative to use rcu_dereference_protected would've created much bigger code churn which is more difficult to test and review. That will be converted in time. 1. Active-backup mode 1.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.55% in bonding, system spent 0.29% CPU in bonding - new bonding: iperf spent 0.29% in bonding, system spent 0.15% CPU in bonding 1.2. Bandwidth measurements - old bonding: 16.1 gbps consistently - new bonding: 17.5 gbps consistently 2. Round-robin mode 2.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.51% in bonding, system spent 0.24% CPU in bonding - new bonding: iperf spent 0.16% in bonding, system spent 0.11% CPU in bonding 2.2 Bandwidth measurements - old bonding: 8 gbps (variable due to packet reorderings) - new bonding: 10 gbps (variable due to packet reorderings) Of course the latency has improved in all converted modes, and moreover while doing enslave/release (since it doesn't affect tx anymore). Also I've stress tested all modes doing enslave/release in a loop while transmitting traffic. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 21:54:51 +07:00
rcu_read_lock();
if (bond_has_slaves(bond))
ret = __bond_start_xmit(skb, dev);
else
bond_tx_drop(dev, skb);
bonding: initial RCU conversion This patch does the initial bonding conversion to RCU. After it the following modes are protected by RCU alone: roundrobin, active-backup, broadcast and xor. Modes ALB/TLB and 3ad still acquire bond->lock for reading, and will be dealt with later. curr_active_slave needs to be dereferenced via rcu in the converted modes because the only thing protecting the slave after this patch is rcu_read_lock, so we need the proper barrier for weakly ordered archs and to make sure we don't have stale pointer. It's not tagged with __rcu yet because there's still work to be done to remove the curr_slave_lock, so sparse will complain when rcu_assign_pointer and rcu_dereference are used, but the alternative to use rcu_dereference_protected would've created much bigger code churn which is more difficult to test and review. That will be converted in time. 1. Active-backup mode 1.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.55% in bonding, system spent 0.29% CPU in bonding - new bonding: iperf spent 0.29% in bonding, system spent 0.15% CPU in bonding 1.2. Bandwidth measurements - old bonding: 16.1 gbps consistently - new bonding: 17.5 gbps consistently 2. Round-robin mode 2.1 Perf recording while doing iperf -P 4 - old bonding: iperf spent 0.51% in bonding, system spent 0.24% CPU in bonding - new bonding: iperf spent 0.16% in bonding, system spent 0.11% CPU in bonding 2.2 Bandwidth measurements - old bonding: 8 gbps (variable due to packet reorderings) - new bonding: 10 gbps (variable due to packet reorderings) Of course the latency has improved in all converted modes, and moreover while doing enslave/release (since it doesn't affect tx anymore). Also I've stress tested all modes doing enslave/release in a loop while transmitting traffic. Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-08-01 21:54:51 +07:00
rcu_read_unlock();
return ret;
}
static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
struct ethtool_link_ksettings *cmd)
{
struct bonding *bond = netdev_priv(bond_dev);
unsigned long speed = 0;
struct list_head *iter;
struct slave *slave;
cmd->base.duplex = DUPLEX_UNKNOWN;
cmd->base.port = PORT_OTHER;
/* Since bond_slave_can_tx returns false for all inactive or down slaves, we
* do not need to check mode. Though link speed might not represent
* the true receive or transmit bandwidth (not all modes are symmetric)
* this is an accurate maximum.
*/
bond_for_each_slave(bond, slave, iter) {
if (bond_slave_can_tx(slave)) {
if (slave->speed != SPEED_UNKNOWN)
speed += slave->speed;
if (cmd->base.duplex == DUPLEX_UNKNOWN &&
slave->duplex != DUPLEX_UNKNOWN)
cmd->base.duplex = slave->duplex;
}
}
cmd->base.speed = speed ? : SPEED_UNKNOWN;
return 0;
}
static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
struct ethtool_drvinfo *drvinfo)
{
strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
BOND_ABI_VERSION);
}
static const struct ethtool_ops bond_ethtool_ops = {
.get_drvinfo = bond_ethtool_get_drvinfo,
.get_link = ethtool_op_get_link,
.get_link_ksettings = bond_ethtool_get_link_ksettings,
};
static const struct net_device_ops bond_netdev_ops = {
.ndo_init = bond_init,
.ndo_uninit = bond_uninit,
.ndo_open = bond_open,
.ndo_stop = bond_close,
.ndo_start_xmit = bond_start_xmit,
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
.ndo_select_queue = bond_select_queue,
.ndo_get_stats64 = bond_get_stats,
.ndo_do_ioctl = bond_do_ioctl,
.ndo_change_rx_flags = bond_change_rx_flags,
.ndo_set_rx_mode = bond_set_rx_mode,
.ndo_change_mtu = bond_change_mtu,
.ndo_set_mac_address = bond_set_mac_address,
.ndo_neigh_setup = bond_neigh_setup,
.ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
.ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
bonding: avoid lockdep confusion in bond_get_stats() syzbot found that the following sequence produces a LOCKDEP splat [1] ip link add bond10 type bond ip link add bond11 type bond ip link set bond11 master bond10 To fix this, we can use the already provided nest_level. This patch also provides correct nesting for dev->addr_list_lock [1] WARNING: possible recursive locking detected 4.18.0-rc6+ #167 Not tainted -------------------------------------------- syz-executor751/4439 is trying to acquire lock: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 but task is already holding lock: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 other info that might help us debug this: Possible unsafe locking scenario: CPU0 ---- lock(&(&bond->stats_lock)->rlock); lock(&(&bond->stats_lock)->rlock); *** DEADLOCK *** May be due to missing lock nesting notation 3 locks held by syz-executor751/4439: #0: (____ptrval____) (rtnl_mutex){+.+.}, at: rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77 #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 #2: (____ptrval____) (rcu_read_lock){....}, at: bond_get_stats+0x0/0x560 include/linux/compiler.h:215 stack backtrace: CPU: 0 PID: 4439 Comm: syz-executor751 Not tainted 4.18.0-rc6+ #167 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113 print_deadlock_bug kernel/locking/lockdep.c:1765 [inline] check_deadlock kernel/locking/lockdep.c:1809 [inline] validate_chain kernel/locking/lockdep.c:2405 [inline] __lock_acquire.cold.64+0x1fb/0x486 kernel/locking/lockdep.c:3435 lock_acquire+0x1e4/0x540 kernel/locking/lockdep.c:3924 __raw_spin_lock include/linux/spinlock_api_smp.h:142 [inline] _raw_spin_lock+0x2a/0x40 kernel/locking/spinlock.c:144 spin_lock include/linux/spinlock.h:310 [inline] bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 dev_get_stats+0x10f/0x470 net/core/dev.c:8316 bond_get_stats+0x232/0x560 drivers/net/bonding/bond_main.c:3432 dev_get_stats+0x10f/0x470 net/core/dev.c:8316 rtnl_fill_stats+0x4d/0xac0 net/core/rtnetlink.c:1169 rtnl_fill_ifinfo+0x1aa6/0x3fb0 net/core/rtnetlink.c:1611 rtmsg_ifinfo_build_skb+0xc8/0x190 net/core/rtnetlink.c:3268 rtmsg_ifinfo_event.part.30+0x45/0xe0 net/core/rtnetlink.c:3300 rtmsg_ifinfo_event net/core/rtnetlink.c:3297 [inline] rtnetlink_event+0x144/0x170 net/core/rtnetlink.c:4716 notifier_call_chain+0x180/0x390 kernel/notifier.c:93 __raw_notifier_call_chain kernel/notifier.c:394 [inline] raw_notifier_call_chain+0x2d/0x40 kernel/notifier.c:401 call_netdevice_notifiers_info+0x3f/0x90 net/core/dev.c:1735 call_netdevice_notifiers net/core/dev.c:1753 [inline] netdev_features_change net/core/dev.c:1321 [inline] netdev_change_features+0xb3/0x110 net/core/dev.c:7759 bond_compute_features.isra.47+0x585/0xa50 drivers/net/bonding/bond_main.c:1120 bond_enslave+0x1b25/0x5da0 drivers/net/bonding/bond_main.c:1755 bond_do_ioctl+0x7cb/0xae0 drivers/net/bonding/bond_main.c:3528 dev_ifsioc+0x43c/0xb30 net/core/dev_ioctl.c:327 dev_ioctl+0x1b5/0xcc0 net/core/dev_ioctl.c:493 sock_do_ioctl+0x1d3/0x3e0 net/socket.c:992 sock_ioctl+0x30d/0x680 net/socket.c:1093 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:500 [inline] do_vfs_ioctl+0x1de/0x1720 fs/ioctl.c:684 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701 __do_sys_ioctl fs/ioctl.c:708 [inline] __se_sys_ioctl fs/ioctl.c:706 [inline] __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x440859 Code: e8 2c af 02 00 48 83 c4 18 c3 0f 1f 80 00 00 00 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 3b 10 fc ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007ffc51a92878 EFLAGS: 00000213 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 0000000000440859 RDX: 0000000020000040 RSI: 0000000000008990 RDI: 0000000000000003 RBP: 0000000000000000 R08: 00000000004002c8 R09: 00000000004002c8 R10: 00000000022d5880 R11: 0000000000000213 R12: 0000000000007390 R13: 0000000000401db0 R14: 0000000000000000 R15: 0000000000000000 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Jay Vosburgh <j.vosburgh@gmail.com> Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-07-31 20:30:54 +07:00
.ndo_get_lock_subclass = bond_get_nest_level,
#ifdef CONFIG_NET_POLL_CONTROLLER
.ndo_netpoll_setup = bond_netpoll_setup,
.ndo_netpoll_cleanup = bond_netpoll_cleanup,
.ndo_poll_controller = bond_poll_controller,
#endif
.ndo_add_slave = bond_enslave,
.ndo_del_slave = bond_release,
.ndo_fix_features = bond_fix_features,
.ndo_features_check = passthru_features_check,
};
static const struct device_type bond_type = {
.name = "bond",
};
static void bond_destructor(struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
if (bond->wq)
destroy_workqueue(bond->wq);
}
void bond_setup(struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
spin_lock_init(&bond->mode_lock);
spin_lock_init(&bond->stats_lock);
bond->params = bonding_defaults;
/* Initialize pointers */
bond->dev = bond_dev;
/* Initialize the device entry points */
ether_setup(bond_dev);
bond_dev->max_mtu = ETH_MAX_MTU;
bond_dev->netdev_ops = &bond_netdev_ops;
bond_dev->ethtool_ops = &bond_ethtool_ops;
net: Fix inconsistent teardown and release of private netdev state. Network devices can allocate reasources and private memory using netdev_ops->ndo_init(). However, the release of these resources can occur in one of two different places. Either netdev_ops->ndo_uninit() or netdev->destructor(). The decision of which operation frees the resources depends upon whether it is necessary for all netdev refs to be released before it is safe to perform the freeing. netdev_ops->ndo_uninit() presumably can occur right after the NETDEV_UNREGISTER notifier completes and the unicast and multicast address lists are flushed. netdev->destructor(), on the other hand, does not run until the netdev references all go away. Further complicating the situation is that netdev->destructor() almost universally does also a free_netdev(). This creates a problem for the logic in register_netdevice(). Because all callers of register_netdevice() manage the freeing of the netdev, and invoke free_netdev(dev) if register_netdevice() fails. If netdev_ops->ndo_init() succeeds, but something else fails inside of register_netdevice(), it does call ndo_ops->ndo_uninit(). But it is not able to invoke netdev->destructor(). This is because netdev->destructor() will do a free_netdev() and then the caller of register_netdevice() will do the same. However, this means that the resources that would normally be released by netdev->destructor() will not be. Over the years drivers have added local hacks to deal with this, by invoking their destructor parts by hand when register_netdevice() fails. Many drivers do not try to deal with this, and instead we have leaks. Let's close this hole by formalizing the distinction between what private things need to be freed up by netdev->destructor() and whether the driver needs unregister_netdevice() to perform the free_netdev(). netdev->priv_destructor() performs all actions to free up the private resources that used to be freed by netdev->destructor(), except for free_netdev(). netdev->needs_free_netdev is a boolean that indicates whether free_netdev() should be done at the end of unregister_netdevice(). Now, register_netdevice() can sanely release all resources after ndo_ops->ndo_init() succeeds, by invoking both ndo_ops->ndo_uninit() and netdev->priv_destructor(). And at the end of unregister_netdevice(), we invoke netdev->priv_destructor() and optionally call free_netdev(). Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-08 23:52:56 +07:00
bond_dev->needs_free_netdev = true;
bond_dev->priv_destructor = bond_destructor;
SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
/* Initialize the device options */
bond_dev->flags |= IFF_MASTER;
bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
/* don't acquire bond device's netif_tx_lock when transmitting */
bond_dev->features |= NETIF_F_LLTX;
/* By default, we declare the bond to be fully
* VLAN hardware accelerated capable. Special
* care is taken in the various xmit functions
* when there are slaves that are not hw accel
* capable
*/
/* Don't allow bond devices to change network namespaces. */
bond_dev->features |= NETIF_F_NETNS_LOCAL;
bond_dev->hw_features = BOND_VLAN_FEATURES |
NETIF_F_HW_VLAN_CTAG_TX |
NETIF_F_HW_VLAN_CTAG_RX |
NETIF_F_HW_VLAN_CTAG_FILTER;
bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
bond_dev->features |= bond_dev->hw_features;
}
/* Destroy a bonding device.
* Must be under rtnl_lock when this function is called.
*/
static void bond_uninit(struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
struct list_head *iter;
struct slave *slave;
struct bond_up_slave *arr;
bond_netpoll_cleanup(bond_dev);
/* Release the bonded slaves */
bond_for_each_slave(bond, slave, iter)
__bond_release_one(bond_dev, slave->dev, true, true);
netdev_info(bond_dev, "Released all slaves\n");
arr = rtnl_dereference(bond->slave_arr);
if (arr) {
RCU_INIT_POINTER(bond->slave_arr, NULL);
kfree_rcu(arr, rcu);
}
list_del(&bond->bond_list);
bond_debug_unregister(bond);
}
/*------------------------- Module initialization ---------------------------*/
static int bond_check_params(struct bond_params *params)
{
int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
struct bond_opt_value newval;
const struct bond_opt_value *valptr;
bonding: fix randomly populated arp target array In commit dc9c4d0fe023, the arp_target array moved from a static global to a local variable. By the nature of static globals, the array used to be initialized to all 0. At present, it's full of random data, which that gets interpreted as arp_target values, when none have actually been specified. Systems end up booting with spew along these lines: [ 32.161783] IPv6: ADDRCONF(NETDEV_UP): lacp0: link is not ready [ 32.168475] IPv6: ADDRCONF(NETDEV_UP): lacp0: link is not ready [ 32.175089] 8021q: adding VLAN 0 to HW filter on device lacp0 [ 32.193091] IPv6: ADDRCONF(NETDEV_UP): lacp0: link is not ready [ 32.204892] lacp0: Setting MII monitoring interval to 100 [ 32.211071] lacp0: Removing ARP target 216.124.228.17 [ 32.216824] lacp0: Removing ARP target 218.160.255.255 [ 32.222646] lacp0: Removing ARP target 185.170.136.184 [ 32.228496] lacp0: invalid ARP target 255.255.255.255 specified for removal [ 32.236294] lacp0: option arp_ip_target: invalid value (-255.255.255.255) [ 32.243987] lacp0: Removing ARP target 56.125.228.17 [ 32.249625] lacp0: Removing ARP target 218.160.255.255 [ 32.255432] lacp0: Removing ARP target 15.157.233.184 [ 32.261165] lacp0: invalid ARP target 255.255.255.255 specified for removal [ 32.268939] lacp0: option arp_ip_target: invalid value (-255.255.255.255) [ 32.276632] lacp0: Removing ARP target 16.0.0.0 [ 32.281755] lacp0: Removing ARP target 218.160.255.255 [ 32.287567] lacp0: Removing ARP target 72.125.228.17 [ 32.293165] lacp0: Removing ARP target 218.160.255.255 [ 32.298970] lacp0: Removing ARP target 8.125.228.17 [ 32.304458] lacp0: Removing ARP target 218.160.255.255 None of these were actually specified as ARP targets, and the driver does seem to clean up the mess okay, but it's rather noisy and confusing, leaks values to userspace, and the 255.255.255.255 spew shows up even when debug prints are disabled. The fix: just zero out arp_target at init time. While we're in here, init arp_all_targets_value in the right place. Fixes: dc9c4d0fe023 ("bonding: reduce scope of some global variables") CC: Mahesh Bandewar <maheshb@google.com> CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org CC: stable@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Acked-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-20 01:46:46 +07:00
int arp_all_targets_value = 0;
u16 ad_actor_sys_prio = 0;
u16 ad_user_port_key = 0;
bonding: fix randomly populated arp target array In commit dc9c4d0fe023, the arp_target array moved from a static global to a local variable. By the nature of static globals, the array used to be initialized to all 0. At present, it's full of random data, which that gets interpreted as arp_target values, when none have actually been specified. Systems end up booting with spew along these lines: [ 32.161783] IPv6: ADDRCONF(NETDEV_UP): lacp0: link is not ready [ 32.168475] IPv6: ADDRCONF(NETDEV_UP): lacp0: link is not ready [ 32.175089] 8021q: adding VLAN 0 to HW filter on device lacp0 [ 32.193091] IPv6: ADDRCONF(NETDEV_UP): lacp0: link is not ready [ 32.204892] lacp0: Setting MII monitoring interval to 100 [ 32.211071] lacp0: Removing ARP target 216.124.228.17 [ 32.216824] lacp0: Removing ARP target 218.160.255.255 [ 32.222646] lacp0: Removing ARP target 185.170.136.184 [ 32.228496] lacp0: invalid ARP target 255.255.255.255 specified for removal [ 32.236294] lacp0: option arp_ip_target: invalid value (-255.255.255.255) [ 32.243987] lacp0: Removing ARP target 56.125.228.17 [ 32.249625] lacp0: Removing ARP target 218.160.255.255 [ 32.255432] lacp0: Removing ARP target 15.157.233.184 [ 32.261165] lacp0: invalid ARP target 255.255.255.255 specified for removal [ 32.268939] lacp0: option arp_ip_target: invalid value (-255.255.255.255) [ 32.276632] lacp0: Removing ARP target 16.0.0.0 [ 32.281755] lacp0: Removing ARP target 218.160.255.255 [ 32.287567] lacp0: Removing ARP target 72.125.228.17 [ 32.293165] lacp0: Removing ARP target 218.160.255.255 [ 32.298970] lacp0: Removing ARP target 8.125.228.17 [ 32.304458] lacp0: Removing ARP target 218.160.255.255 None of these were actually specified as ARP targets, and the driver does seem to clean up the mess okay, but it's rather noisy and confusing, leaks values to userspace, and the 255.255.255.255 spew shows up even when debug prints are disabled. The fix: just zero out arp_target at init time. While we're in here, init arp_all_targets_value in the right place. Fixes: dc9c4d0fe023 ("bonding: reduce scope of some global variables") CC: Mahesh Bandewar <maheshb@google.com> CC: Jay Vosburgh <j.vosburgh@gmail.com> CC: Veaceslav Falico <vfalico@gmail.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: netdev@vger.kernel.org CC: stable@vger.kernel.org Signed-off-by: Jarod Wilson <jarod@redhat.com> Acked-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-20 01:46:46 +07:00
__be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
int arp_ip_count;
int bond_mode = BOND_MODE_ROUNDROBIN;
int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
int lacp_fast = 0;
int tlb_dynamic_lb;
/* Convert string parameters. */
if (mode) {
bond_opt_initstr(&newval, mode);
valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
if (!valptr) {
pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
return -EINVAL;
}
bond_mode = valptr->value;
}
if (xmit_hash_policy) {
if (bond_mode == BOND_MODE_ROUNDROBIN ||
bond_mode == BOND_MODE_ACTIVEBACKUP ||
bond_mode == BOND_MODE_BROADCAST) {
pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
bond_mode_name(bond_mode));
} else {
bond_opt_initstr(&newval, xmit_hash_policy);
valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
&newval);
if (!valptr) {
pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
xmit_hash_policy);
return -EINVAL;
}
xmit_hashtype = valptr->value;
}
}
if (lacp_rate) {
if (bond_mode != BOND_MODE_8023AD) {
pr_info("lacp_rate param is irrelevant in mode %s\n",
bond_mode_name(bond_mode));
} else {
bond_opt_initstr(&newval, lacp_rate);
valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
&newval);
if (!valptr) {
pr_err("Error: Invalid lacp rate \"%s\"\n",
lacp_rate);
return -EINVAL;
}
lacp_fast = valptr->value;
}
}
if (ad_select) {
bond_opt_initstr(&newval, ad_select);
valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
&newval);
if (!valptr) {
pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
return -EINVAL;
}
params->ad_select = valptr->value;
if (bond_mode != BOND_MODE_8023AD)
pr_warn("ad_select param only affects 802.3ad mode\n");
} else {
params->ad_select = BOND_AD_STABLE;
}
if (max_bonds < 0) {
pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
max_bonds = BOND_DEFAULT_MAX_BONDS;
}
if (miimon < 0) {
pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
miimon, INT_MAX);
miimon = 0;
}
if (updelay < 0) {
pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
updelay, INT_MAX);
updelay = 0;
}
if (downdelay < 0) {
pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
downdelay, INT_MAX);
downdelay = 0;
}
if ((use_carrier != 0) && (use_carrier != 1)) {
pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
use_carrier);
use_carrier = 1;
}
if (num_peer_notif < 0 || num_peer_notif > 255) {
pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
num_peer_notif);
num_peer_notif = 1;
}
/* reset values for 802.3ad/TLB/ALB */
if (!bond_mode_uses_arp(bond_mode)) {
if (!miimon) {
pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
pr_warn("Forcing miimon to 100msec\n");
miimon = BOND_DEFAULT_MIIMON;
}
}
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
if (tx_queues < 1 || tx_queues > 255) {
pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
tx_queues, BOND_DEFAULT_TX_QUEUES);
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
tx_queues = BOND_DEFAULT_TX_QUEUES;
}
if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
all_slaves_active);
all_slaves_active = 0;
}
if (resend_igmp < 0 || resend_igmp > 255) {
pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
resend_igmp, BOND_DEFAULT_RESEND_IGMP);
resend_igmp = BOND_DEFAULT_RESEND_IGMP;
}
bond_opt_initval(&newval, packets_per_slave);
if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
packets_per_slave, USHRT_MAX);
packets_per_slave = 1;
}
if (bond_mode == BOND_MODE_ALB) {
pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
updelay);
}
if (!miimon) {
if (updelay || downdelay) {
/* just warn the user the up/down delay will have
* no effect since miimon is zero...
*/
pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
updelay, downdelay);
}
} else {
/* don't allow arp monitoring */
if (arp_interval) {
pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
miimon, arp_interval);
arp_interval = 0;
}
if ((updelay % miimon) != 0) {
pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
updelay, miimon, (updelay / miimon) * miimon);
}
updelay /= miimon;
if ((downdelay % miimon) != 0) {
pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
downdelay, miimon,
(downdelay / miimon) * miimon);
}
downdelay /= miimon;
}
if (arp_interval < 0) {
pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
arp_interval, INT_MAX);
arp_interval = 0;
}
for (arp_ip_count = 0, i = 0;
(arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
__be32 ip;
/* not a complete check, but good enough to catch mistakes */
if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
!bond_is_ip_target_ok(ip)) {
pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
arp_ip_target[i]);
arp_interval = 0;
} else {
if (bond_get_targets_ip(arp_target, ip) == -1)
arp_target[arp_ip_count++] = ip;
else
pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
&ip);
}
}
if (arp_interval && !arp_ip_count) {
/* don't allow arping if no arp_ip_target given... */
pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
arp_interval);
arp_interval = 0;
}
if (arp_validate) {
if (!arp_interval) {
pr_err("arp_validate requires arp_interval\n");
return -EINVAL;
}
bond_opt_initstr(&newval, arp_validate);
valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
&newval);
if (!valptr) {
pr_err("Error: invalid arp_validate \"%s\"\n",
arp_validate);
return -EINVAL;
}
arp_validate_value = valptr->value;
} else {
arp_validate_value = 0;
}
bonding: add an option to fail when any of arp_ip_target is inaccessible Currently, we fail only when all of the ips in arp_ip_target are gone. However, in some situations we might need to fail if even one host from arp_ip_target becomes unavailable. All situations, obviously, rely on the idea that we need *completely* functional network, with all interfaces/addresses working correctly. One real world example might be: vlans on top on bond (hybrid port). If bond and vlans have ips assigned and we have their peers monitored via arp_ip_target - in case of switch misconfiguration (trunk/access port), slave driver malfunction or tagged/untagged traffic dropped on the way - we will be able to switch to another slave. Though any other configuration needs that if we need to have access to all arp_ip_targets. This patch adds this possibility by adding a new parameter - arp_all_targets (both as a module parameter and as a sysfs knob). It can be set to: 0 or any (the default) - which works exactly as it's working now - the slave is up if any of the arp_ip_targets are up. 1 or all - the slave is up if all of the arp_ip_targets are up. This parameter can be changed on the fly (via sysfs), and requires the mode to be active-backup and arp_validate to be enabled (it obeys the arp_validate config on which slaves to validate). Internally it's done through: 1) Add target_last_arp_rx[BOND_MAX_ARP_TARGETS] array to slave struct. It's an array of jiffies, meaning that slave->target_last_arp_rx[i] is the last time we've received arp from bond->params.arp_targets[i] on this slave. 2) If we successfully validate an arp from bond->params.arp_targets[i] in bond_validate_arp() - update the slave->target_last_arp_rx[i] with the current jiffies value. 3) When getting slave's last_rx via slave_last_rx(), we return the oldest time when we've received an arp from any address in bond->params.arp_targets[]. If the value of arp_all_targets == 0 - we still work the same way as before. Also, update the documentation to reflect the new parameter. v3->v4: Kill the forgotten rtnl_unlock(), rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Also, print a warning if the last arp_ip_target is removed while the arp_interval is on, but not the arp_validate. v2->v3: Use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). On bond_enslave(), use the same initialization value for target_last_arp_rx[] as is used for the default last_arp_rx, to avoid useless interface flaps. Also, instead of failing to remove the last arp_ip_target just print a warning - otherwise it might break existing scripts. v1->v2: Correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes. Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-24 16:49:34 +07:00
if (arp_all_targets) {
bond_opt_initstr(&newval, arp_all_targets);
valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
&newval);
if (!valptr) {
bonding: add an option to fail when any of arp_ip_target is inaccessible Currently, we fail only when all of the ips in arp_ip_target are gone. However, in some situations we might need to fail if even one host from arp_ip_target becomes unavailable. All situations, obviously, rely on the idea that we need *completely* functional network, with all interfaces/addresses working correctly. One real world example might be: vlans on top on bond (hybrid port). If bond and vlans have ips assigned and we have their peers monitored via arp_ip_target - in case of switch misconfiguration (trunk/access port), slave driver malfunction or tagged/untagged traffic dropped on the way - we will be able to switch to another slave. Though any other configuration needs that if we need to have access to all arp_ip_targets. This patch adds this possibility by adding a new parameter - arp_all_targets (both as a module parameter and as a sysfs knob). It can be set to: 0 or any (the default) - which works exactly as it's working now - the slave is up if any of the arp_ip_targets are up. 1 or all - the slave is up if all of the arp_ip_targets are up. This parameter can be changed on the fly (via sysfs), and requires the mode to be active-backup and arp_validate to be enabled (it obeys the arp_validate config on which slaves to validate). Internally it's done through: 1) Add target_last_arp_rx[BOND_MAX_ARP_TARGETS] array to slave struct. It's an array of jiffies, meaning that slave->target_last_arp_rx[i] is the last time we've received arp from bond->params.arp_targets[i] on this slave. 2) If we successfully validate an arp from bond->params.arp_targets[i] in bond_validate_arp() - update the slave->target_last_arp_rx[i] with the current jiffies value. 3) When getting slave's last_rx via slave_last_rx(), we return the oldest time when we've received an arp from any address in bond->params.arp_targets[]. If the value of arp_all_targets == 0 - we still work the same way as before. Also, update the documentation to reflect the new parameter. v3->v4: Kill the forgotten rtnl_unlock(), rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Also, print a warning if the last arp_ip_target is removed while the arp_interval is on, but not the arp_validate. v2->v3: Use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). On bond_enslave(), use the same initialization value for target_last_arp_rx[] as is used for the default last_arp_rx, to avoid useless interface flaps. Also, instead of failing to remove the last arp_ip_target just print a warning - otherwise it might break existing scripts. v1->v2: Correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes. Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-24 16:49:34 +07:00
pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
arp_all_targets);
arp_all_targets_value = 0;
} else {
arp_all_targets_value = valptr->value;
bonding: add an option to fail when any of arp_ip_target is inaccessible Currently, we fail only when all of the ips in arp_ip_target are gone. However, in some situations we might need to fail if even one host from arp_ip_target becomes unavailable. All situations, obviously, rely on the idea that we need *completely* functional network, with all interfaces/addresses working correctly. One real world example might be: vlans on top on bond (hybrid port). If bond and vlans have ips assigned and we have their peers monitored via arp_ip_target - in case of switch misconfiguration (trunk/access port), slave driver malfunction or tagged/untagged traffic dropped on the way - we will be able to switch to another slave. Though any other configuration needs that if we need to have access to all arp_ip_targets. This patch adds this possibility by adding a new parameter - arp_all_targets (both as a module parameter and as a sysfs knob). It can be set to: 0 or any (the default) - which works exactly as it's working now - the slave is up if any of the arp_ip_targets are up. 1 or all - the slave is up if all of the arp_ip_targets are up. This parameter can be changed on the fly (via sysfs), and requires the mode to be active-backup and arp_validate to be enabled (it obeys the arp_validate config on which slaves to validate). Internally it's done through: 1) Add target_last_arp_rx[BOND_MAX_ARP_TARGETS] array to slave struct. It's an array of jiffies, meaning that slave->target_last_arp_rx[i] is the last time we've received arp from bond->params.arp_targets[i] on this slave. 2) If we successfully validate an arp from bond->params.arp_targets[i] in bond_validate_arp() - update the slave->target_last_arp_rx[i] with the current jiffies value. 3) When getting slave's last_rx via slave_last_rx(), we return the oldest time when we've received an arp from any address in bond->params.arp_targets[]. If the value of arp_all_targets == 0 - we still work the same way as before. Also, update the documentation to reflect the new parameter. v3->v4: Kill the forgotten rtnl_unlock(), rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Also, print a warning if the last arp_ip_target is removed while the arp_interval is on, but not the arp_validate. v2->v3: Use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). On bond_enslave(), use the same initialization value for target_last_arp_rx[] as is used for the default last_arp_rx, to avoid useless interface flaps. Also, instead of failing to remove the last arp_ip_target just print a warning - otherwise it might break existing scripts. v1->v2: Correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes. Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-24 16:49:34 +07:00
}
}
if (miimon) {
pr_info("MII link monitoring set to %d ms\n", miimon);
} else if (arp_interval) {
valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
arp_validate_value);
pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
arp_interval, valptr->string, arp_ip_count);
for (i = 0; i < arp_ip_count; i++)
pr_cont(" %s", arp_ip_target[i]);
pr_cont("\n");
} else if (max_bonds) {
/* miimon and arp_interval not set, we need one so things
* work as expected, see bonding.txt for details
*/
pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details\n");
}
if (primary && !bond_mode_uses_primary(bond_mode)) {
/* currently, using a primary only makes sense
* in active backup, TLB or ALB modes
*/
pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
primary, bond_mode_name(bond_mode));
primary = NULL;
}
if (primary && primary_reselect) {
bond_opt_initstr(&newval, primary_reselect);
valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
&newval);
if (!valptr) {
pr_err("Error: Invalid primary_reselect \"%s\"\n",
primary_reselect);
return -EINVAL;
}
primary_reselect_value = valptr->value;
} else {
primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
}
if (fail_over_mac) {
bond_opt_initstr(&newval, fail_over_mac);
valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
&newval);
if (!valptr) {
pr_err("Error: invalid fail_over_mac \"%s\"\n",
fail_over_mac);
return -EINVAL;
}
fail_over_mac_value = valptr->value;
if (bond_mode != BOND_MODE_ACTIVEBACKUP)
pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
} else {
fail_over_mac_value = BOND_FOM_NONE;
}
bond_opt_initstr(&newval, "default");
valptr = bond_opt_parse(
bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
&newval);
if (!valptr) {
pr_err("Error: No ad_actor_sys_prio default value");
return -EINVAL;
}
ad_actor_sys_prio = valptr->value;
valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
&newval);
if (!valptr) {
pr_err("Error: No ad_user_port_key default value");
return -EINVAL;
}
ad_user_port_key = valptr->value;
bond_opt_initstr(&newval, "default");
valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
if (!valptr) {
pr_err("Error: No tlb_dynamic_lb default value");
return -EINVAL;
}
tlb_dynamic_lb = valptr->value;
if (lp_interval == 0) {
pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
}
/* fill params struct with the proper values */
params->mode = bond_mode;
params->xmit_policy = xmit_hashtype;
params->miimon = miimon;
params->num_peer_notif = num_peer_notif;
params->arp_interval = arp_interval;
params->arp_validate = arp_validate_value;
bonding: add an option to fail when any of arp_ip_target is inaccessible Currently, we fail only when all of the ips in arp_ip_target are gone. However, in some situations we might need to fail if even one host from arp_ip_target becomes unavailable. All situations, obviously, rely on the idea that we need *completely* functional network, with all interfaces/addresses working correctly. One real world example might be: vlans on top on bond (hybrid port). If bond and vlans have ips assigned and we have their peers monitored via arp_ip_target - in case of switch misconfiguration (trunk/access port), slave driver malfunction or tagged/untagged traffic dropped on the way - we will be able to switch to another slave. Though any other configuration needs that if we need to have access to all arp_ip_targets. This patch adds this possibility by adding a new parameter - arp_all_targets (both as a module parameter and as a sysfs knob). It can be set to: 0 or any (the default) - which works exactly as it's working now - the slave is up if any of the arp_ip_targets are up. 1 or all - the slave is up if all of the arp_ip_targets are up. This parameter can be changed on the fly (via sysfs), and requires the mode to be active-backup and arp_validate to be enabled (it obeys the arp_validate config on which slaves to validate). Internally it's done through: 1) Add target_last_arp_rx[BOND_MAX_ARP_TARGETS] array to slave struct. It's an array of jiffies, meaning that slave->target_last_arp_rx[i] is the last time we've received arp from bond->params.arp_targets[i] on this slave. 2) If we successfully validate an arp from bond->params.arp_targets[i] in bond_validate_arp() - update the slave->target_last_arp_rx[i] with the current jiffies value. 3) When getting slave's last_rx via slave_last_rx(), we return the oldest time when we've received an arp from any address in bond->params.arp_targets[]. If the value of arp_all_targets == 0 - we still work the same way as before. Also, update the documentation to reflect the new parameter. v3->v4: Kill the forgotten rtnl_unlock(), rephrase the documentation part to be more clear, don't fail setting arp_all_targets if arp_validate is not set - it has no effect anyway but can be easier to set up. Also, print a warning if the last arp_ip_target is removed while the arp_interval is on, but not the arp_validate. v2->v3: Use _bh spinlock, remove useless rtnl_lock() and use jiffies for new arp_ip_target last arp, instead of slave_last_rx(). On bond_enslave(), use the same initialization value for target_last_arp_rx[] as is used for the default last_arp_rx, to avoid useless interface flaps. Also, instead of failing to remove the last arp_ip_target just print a warning - otherwise it might break existing scripts. v1->v2: Correctly handle adding/removing hosts in arp_ip_target - we need to shift/initialize all slave's target_last_arp_rx. Also, don't fail module loading on arp_all_targets misconfiguration, just disable it, and some minor style fixes. Signed-off-by: Veaceslav Falico <vfalico@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-06-24 16:49:34 +07:00
params->arp_all_targets = arp_all_targets_value;
params->updelay = updelay;
params->downdelay = downdelay;
params->use_carrier = use_carrier;
params->lacp_fast = lacp_fast;
params->primary[0] = 0;
params->primary_reselect = primary_reselect_value;
params->fail_over_mac = fail_over_mac_value;
bonding: allow user-controlled output slave selection v2: changed bonding module version, modified to apply on top of changes from previous patch in series, and updated documentation to elaborate on multiqueue awareness that now exists in bonding driver. This patch give the user the ability to control the output slave for round-robin and active-backup bonding. Similar functionality was discussed in the past, but Jay Vosburgh indicated he would rather see a feature like this added to existing modes rather than creating a completely new mode. Jay's thoughts as well as Neil's input surrounding some of the issues with the first implementation pushed us toward a design that relied on the queue_mapping rather than skb marks. Round-robin and active-backup modes were chosen as the first users of this slave selection as they seemed like the most logical choices when considering a multi-switch environment. Round-robin mode works without any modification, but active-backup does require inclusion of the first patch in this series and setting the 'all_slaves_active' flag. This will allow reception of unicast traffic on any of the backup interfaces. This was tested with IPv4-based filters as well as VLAN-based filters with good results. More information as well as a configuration example is available in the patch to Documentation/networking/bonding.txt. Signed-off-by: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-06-02 15:40:18 +07:00
params->tx_queues = tx_queues;
params->all_slaves_active = all_slaves_active;
params->resend_igmp = resend_igmp;
params->min_links = min_links;
params->lp_interval = lp_interval;
reciprocal_divide: update/correction of the algorithm Jakub Zawadzki noticed that some divisions by reciprocal_divide() were not correct [1][2], which he could also show with BPF code after divisions are transformed into reciprocal_value() for runtime invariance which can be passed to reciprocal_divide() later on; reverse in BPF dump ended up with a different, off-by-one K in some situations. This has been fixed by Eric Dumazet in commit aee636c4809fa5 ("bpf: do not use reciprocal divide"). This follow-up patch improves reciprocal_value() and reciprocal_divide() to work in all cases by using Granlund and Montgomery method, so that also future use is safe and without any non-obvious side-effects. Known problems with the old implementation were that division by 1 always returned 0 and some off-by-ones when the dividend and divisor where very large. This seemed to not be problematic with its current users, as far as we can tell. Eric Dumazet checked for the slab usage, we cannot surely say so in the case of flex_array. Still, in order to fix that, we propose an extension from the original implementation from commit 6a2d7a955d8d resp. [3][4], by using the algorithm proposed in "Division by Invariant Integers Using Multiplication" [5], Torbjörn Granlund and Peter L. Montgomery, that is, pseudocode for q = n/d where q, n, d is in u32 universe: 1) Initialization: int l = ceil(log_2 d) uword m' = floor((1<<32)*((1<<l)-d)/d)+1 int sh_1 = min(l,1) int sh_2 = max(l-1,0) 2) For q = n/d, all uword: uword t = (n*m')>>32 q = (t+((n-t)>>sh_1))>>sh_2 The assembler implementation from Agner Fog [6] also helped a lot while implementing. We have tested the implementation on x86_64, ppc64, i686, s390x; on x86_64/haswell we're still half the latency compared to normal divide. Joint work with Daniel Borkmann. [1] http://www.wireshark.org/~darkjames/reciprocal-buggy.c [2] http://www.wireshark.org/~darkjames/set-and-dump-filter-k-bug.c [3] https://gmplib.org/~tege/division-paper.pdf [4] http://homepage.cs.uiowa.edu/~jones/bcd/divide.html [5] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.1.2556 [6] http://www.agner.org/optimize/asmlib.zip Reported-by: Jakub Zawadzki <darkjames-ws@darkjames.pl> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Austin S Hemmelgarn <ahferroin7@gmail.com> Cc: linux-kernel@vger.kernel.org Cc: Jesse Gross <jesse@nicira.com> Cc: Jamal Hadi Salim <jhs@mojatatu.com> Cc: Stephen Hemminger <stephen@networkplumber.org> Cc: Matt Mackall <mpm@selenic.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Andy Gospodarek <andy@greyhouse.net> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Jakub Zawadzki <darkjames-ws@darkjames.pl> Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-22 08:29:41 +07:00
params->packets_per_slave = packets_per_slave;
params->tlb_dynamic_lb = tlb_dynamic_lb;
params->ad_actor_sys_prio = ad_actor_sys_prio;
bonding: Allow userspace to set actors' macaddr in an AD-system. In an AD system, the communication between actor and partner is the business between these two entities. In the current setup anyone on the same L2 can "guess" the LACPDU contents and then possibly send the spoofed LACPDUs and trick the partner causing connectivity issues for the AD system. This patch allows to use a random mac-address obscuring it's identity making it harder for someone in the L2 is do the same thing. This patch allows user-space to choose the mac-address for the AD-system. This mac-address can not be NULL or a Multicast. If the mac-address is set from user-space; kernel will honor it and will not overwrite it. In the absence (value from user space); the logic will default to using the masters' mac as the mac-address for the AD-system. It can be set using example code below - # modprobe bonding mode=4 # sys_mac_addr=$(printf '%02x:%02x:%02x:%02x:%02x:%02x' \ $(( (RANDOM & 0xFE) | 0x02 )) \ $(( RANDOM & 0xFF )) \ $(( RANDOM & 0xFF )) \ $(( RANDOM & 0xFF )) \ $(( RANDOM & 0xFF )) \ $(( RANDOM & 0xFF ))) # echo $sys_mac_addr > /sys/class/net/bond0/bonding/ad_actor_system # echo +eth1 > /sys/class/net/bond0/bonding/slaves ... # ip link set bond0 up Signed-off-by: Mahesh Bandewar <maheshb@google.com> Reviewed-by: Nikolay Aleksandrov <nikolay@redhat.com> [jt: fixed up style issues reported by checkpatch] Signed-off-by: Jonathan Toppins <jtoppins@cumulusnetworks.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-05-09 14:01:56 +07:00
eth_zero_addr(params->ad_actor_system);
params->ad_user_port_key = ad_user_port_key;
reciprocal_divide: update/correction of the algorithm Jakub Zawadzki noticed that some divisions by reciprocal_divide() were not correct [1][2], which he could also show with BPF code after divisions are transformed into reciprocal_value() for runtime invariance which can be passed to reciprocal_divide() later on; reverse in BPF dump ended up with a different, off-by-one K in some situations. This has been fixed by Eric Dumazet in commit aee636c4809fa5 ("bpf: do not use reciprocal divide"). This follow-up patch improves reciprocal_value() and reciprocal_divide() to work in all cases by using Granlund and Montgomery method, so that also future use is safe and without any non-obvious side-effects. Known problems with the old implementation were that division by 1 always returned 0 and some off-by-ones when the dividend and divisor where very large. This seemed to not be problematic with its current users, as far as we can tell. Eric Dumazet checked for the slab usage, we cannot surely say so in the case of flex_array. Still, in order to fix that, we propose an extension from the original implementation from commit 6a2d7a955d8d resp. [3][4], by using the algorithm proposed in "Division by Invariant Integers Using Multiplication" [5], Torbjörn Granlund and Peter L. Montgomery, that is, pseudocode for q = n/d where q, n, d is in u32 universe: 1) Initialization: int l = ceil(log_2 d) uword m' = floor((1<<32)*((1<<l)-d)/d)+1 int sh_1 = min(l,1) int sh_2 = max(l-1,0) 2) For q = n/d, all uword: uword t = (n*m')>>32 q = (t+((n-t)>>sh_1))>>sh_2 The assembler implementation from Agner Fog [6] also helped a lot while implementing. We have tested the implementation on x86_64, ppc64, i686, s390x; on x86_64/haswell we're still half the latency compared to normal divide. Joint work with Daniel Borkmann. [1] http://www.wireshark.org/~darkjames/reciprocal-buggy.c [2] http://www.wireshark.org/~darkjames/set-and-dump-filter-k-bug.c [3] https://gmplib.org/~tege/division-paper.pdf [4] http://homepage.cs.uiowa.edu/~jones/bcd/divide.html [5] http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.1.2556 [6] http://www.agner.org/optimize/asmlib.zip Reported-by: Jakub Zawadzki <darkjames-ws@darkjames.pl> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Austin S Hemmelgarn <ahferroin7@gmail.com> Cc: linux-kernel@vger.kernel.org Cc: Jesse Gross <jesse@nicira.com> Cc: Jamal Hadi Salim <jhs@mojatatu.com> Cc: Stephen Hemminger <stephen@networkplumber.org> Cc: Matt Mackall <mpm@selenic.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Andy Gospodarek <andy@greyhouse.net> Cc: Veaceslav Falico <vfalico@redhat.com> Cc: Jay Vosburgh <fubar@us.ibm.com> Cc: Jakub Zawadzki <darkjames-ws@darkjames.pl> Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2014-01-22 08:29:41 +07:00
if (packets_per_slave > 0) {
params->reciprocal_packets_per_slave =
reciprocal_value(packets_per_slave);
} else {
/* reciprocal_packets_per_slave is unused if
* packets_per_slave is 0 or 1, just initialize it
*/
params->reciprocal_packets_per_slave =
(struct reciprocal_value) { 0 };
}
if (primary) {
strncpy(params->primary, primary, IFNAMSIZ);
params->primary[IFNAMSIZ - 1] = 0;
}
memcpy(params->arp_targets, arp_target, sizeof(arp_target));
return 0;
}
/* Called from registration process */
static int bond_init(struct net_device *bond_dev)
{
struct bonding *bond = netdev_priv(bond_dev);
struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
netdev_dbg(bond_dev, "Begin bond_init\n");
bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
if (!bond->wq)
return -ENOMEM;
bonding: avoid lockdep confusion in bond_get_stats() syzbot found that the following sequence produces a LOCKDEP splat [1] ip link add bond10 type bond ip link add bond11 type bond ip link set bond11 master bond10 To fix this, we can use the already provided nest_level. This patch also provides correct nesting for dev->addr_list_lock [1] WARNING: possible recursive locking detected 4.18.0-rc6+ #167 Not tainted -------------------------------------------- syz-executor751/4439 is trying to acquire lock: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 but task is already holding lock: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 other info that might help us debug this: Possible unsafe locking scenario: CPU0 ---- lock(&(&bond->stats_lock)->rlock); lock(&(&bond->stats_lock)->rlock); *** DEADLOCK *** May be due to missing lock nesting notation 3 locks held by syz-executor751/4439: #0: (____ptrval____) (rtnl_mutex){+.+.}, at: rtnl_lock+0x17/0x20 net/core/rtnetlink.c:77 #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: spin_lock include/linux/spinlock.h:310 [inline] #1: (____ptrval____) (&(&bond->stats_lock)->rlock){+.+.}, at: bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 #2: (____ptrval____) (rcu_read_lock){....}, at: bond_get_stats+0x0/0x560 include/linux/compiler.h:215 stack backtrace: CPU: 0 PID: 4439 Comm: syz-executor751 Not tainted 4.18.0-rc6+ #167 Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011 Call Trace: __dump_stack lib/dump_stack.c:77 [inline] dump_stack+0x1c9/0x2b4 lib/dump_stack.c:113 print_deadlock_bug kernel/locking/lockdep.c:1765 [inline] check_deadlock kernel/locking/lockdep.c:1809 [inline] validate_chain kernel/locking/lockdep.c:2405 [inline] __lock_acquire.cold.64+0x1fb/0x486 kernel/locking/lockdep.c:3435 lock_acquire+0x1e4/0x540 kernel/locking/lockdep.c:3924 __raw_spin_lock include/linux/spinlock_api_smp.h:142 [inline] _raw_spin_lock+0x2a/0x40 kernel/locking/spinlock.c:144 spin_lock include/linux/spinlock.h:310 [inline] bond_get_stats+0xb4/0x560 drivers/net/bonding/bond_main.c:3426 dev_get_stats+0x10f/0x470 net/core/dev.c:8316 bond_get_stats+0x232/0x560 drivers/net/bonding/bond_main.c:3432 dev_get_stats+0x10f/0x470 net/core/dev.c:8316 rtnl_fill_stats+0x4d/0xac0 net/core/rtnetlink.c:1169 rtnl_fill_ifinfo+0x1aa6/0x3fb0 net/core/rtnetlink.c:1611 rtmsg_ifinfo_build_skb+0xc8/0x190 net/core/rtnetlink.c:3268 rtmsg_ifinfo_event.part.30+0x45/0xe0 net/core/rtnetlink.c:3300 rtmsg_ifinfo_event net/core/rtnetlink.c:3297 [inline] rtnetlink_event+0x144/0x170 net/core/rtnetlink.c:4716 notifier_call_chain+0x180/0x390 kernel/notifier.c:93 __raw_notifier_call_chain kernel/notifier.c:394 [inline] raw_notifier_call_chain+0x2d/0x40 kernel/notifier.c:401 call_netdevice_notifiers_info+0x3f/0x90 net/core/dev.c:1735 call_netdevice_notifiers net/core/dev.c:1753 [inline] netdev_features_change net/core/dev.c:1321 [inline] netdev_change_features+0xb3/0x110 net/core/dev.c:7759 bond_compute_features.isra.47+0x585/0xa50 drivers/net/bonding/bond_main.c:1120 bond_enslave+0x1b25/0x5da0 drivers/net/bonding/bond_main.c:1755 bond_do_ioctl+0x7cb/0xae0 drivers/net/bonding/bond_main.c:3528 dev_ifsioc+0x43c/0xb30 net/core/dev_ioctl.c:327 dev_ioctl+0x1b5/0xcc0 net/core/dev_ioctl.c:493 sock_do_ioctl+0x1d3/0x3e0 net/socket.c:992 sock_ioctl+0x30d/0x680 net/socket.c:1093 vfs_ioctl fs/ioctl.c:46 [inline] file_ioctl fs/ioctl.c:500 [inline] do_vfs_ioctl+0x1de/0x1720 fs/ioctl.c:684 ksys_ioctl+0xa9/0xd0 fs/ioctl.c:701 __do_sys_ioctl fs/ioctl.c:708 [inline] __se_sys_ioctl fs/ioctl.c:706 [inline] __x64_sys_ioctl+0x73/0xb0 fs/ioctl.c:706 do_syscall_64+0x1b9/0x820 arch/x86/entry/common.c:290 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x440859 Code: e8 2c af 02 00 48 83 c4 18 c3 0f 1f 80 00 00 00 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 0f 83 3b 10 fc ff c3 66 2e 0f 1f 84 00 00 00 00 RSP: 002b:00007ffc51a92878 EFLAGS: 00000213 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 0000000000440859 RDX: 0000000020000040 RSI: 0000000000008990 RDI: 0000000000000003 RBP: 0000000000000000 R08: 00000000004002c8 R09: 00000000004002c8 R10: 00000000022d5880 R11: 0000000000000213 R12: 0000000000007390 R13: 0000000000401db0 R14: 0000000000000000 R15: 0000000000000000 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Jay Vosburgh <j.vosburgh@gmail.com> Cc: Veaceslav Falico <vfalico@gmail.com> Cc: Andy Gospodarek <andy@greyhouse.net> Signed-off-by: David S. Miller <davem@davemloft.net>
2018-07-31 20:30:54 +07:00
bond->nest_level = SINGLE_DEPTH_NESTING;
netdev_lockdep_set_classes(bond_dev);
list_add_tail(&bond->bond_list, &bn->dev_list);
bond_prepare_sysfs_group(bond);
bond_debug_register(bond);
/* Ensure valid dev_addr */
if (is_zero_ether_addr(bond_dev->dev_addr) &&
bond_dev->addr_assign_type == NET_ADDR_PERM)
eth_hw_addr_random(bond_dev);
return 0;
}
unsigned int bond_get_num_tx_queues(void)
{
return tx_queues;
}
/* Create a new bond based on the specified name and bonding parameters.
* If name is NULL, obtain a suitable "bond%d" name for us.
* Caller must NOT hold rtnl_lock; we need to release it here before we
* set up our sysfs entries.
*/
int bond_create(struct net *net, const char *name)
{
struct net_device *bond_dev;
struct bonding *bond;
struct alb_bond_info *bond_info;
int res;
rtnl_lock();
bond_dev = alloc_netdev_mq(sizeof(struct bonding),
name ? name : "bond%d", NET_NAME_UNKNOWN,
bond_setup, tx_queues);
if (!bond_dev) {
pr_err("%s: eek! can't alloc netdev!\n", name);
rtnl_unlock();
return -ENOMEM;
}
/*
* Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
* It is set to 0 by default which is wrong.
*/
bond = netdev_priv(bond_dev);
bond_info = &(BOND_ALB_INFO(bond));
bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
dev_net_set(bond_dev, net);
bond_dev->rtnl_link_ops = &bond_link_ops;
res = register_netdevice(bond_dev);
netif_carrier_off(bond_dev);
bond_work_init_all(bond);
rtnl_unlock();
if (res < 0)
net: Fix inconsistent teardown and release of private netdev state. Network devices can allocate reasources and private memory using netdev_ops->ndo_init(). However, the release of these resources can occur in one of two different places. Either netdev_ops->ndo_uninit() or netdev->destructor(). The decision of which operation frees the resources depends upon whether it is necessary for all netdev refs to be released before it is safe to perform the freeing. netdev_ops->ndo_uninit() presumably can occur right after the NETDEV_UNREGISTER notifier completes and the unicast and multicast address lists are flushed. netdev->destructor(), on the other hand, does not run until the netdev references all go away. Further complicating the situation is that netdev->destructor() almost universally does also a free_netdev(). This creates a problem for the logic in register_netdevice(). Because all callers of register_netdevice() manage the freeing of the netdev, and invoke free_netdev(dev) if register_netdevice() fails. If netdev_ops->ndo_init() succeeds, but something else fails inside of register_netdevice(), it does call ndo_ops->ndo_uninit(). But it is not able to invoke netdev->destructor(). This is because netdev->destructor() will do a free_netdev() and then the caller of register_netdevice() will do the same. However, this means that the resources that would normally be released by netdev->destructor() will not be. Over the years drivers have added local hacks to deal with this, by invoking their destructor parts by hand when register_netdevice() fails. Many drivers do not try to deal with this, and instead we have leaks. Let's close this hole by formalizing the distinction between what private things need to be freed up by netdev->destructor() and whether the driver needs unregister_netdevice() to perform the free_netdev(). netdev->priv_destructor() performs all actions to free up the private resources that used to be freed by netdev->destructor(), except for free_netdev(). netdev->needs_free_netdev is a boolean that indicates whether free_netdev() should be done at the end of unregister_netdevice(). Now, register_netdevice() can sanely release all resources after ndo_ops->ndo_init() succeeds, by invoking both ndo_ops->ndo_uninit() and netdev->priv_destructor(). And at the end of unregister_netdevice(), we invoke netdev->priv_destructor() and optionally call free_netdev(). Signed-off-by: David S. Miller <davem@davemloft.net>
2017-05-08 23:52:56 +07:00
free_netdev(bond_dev);
return res;
}
static int __net_init bond_net_init(struct net *net)
{
struct bond_net *bn = net_generic(net, bond_net_id);
bn->net = net;
INIT_LIST_HEAD(&bn->dev_list);
bond_create_proc_dir(bn);
bond_create_sysfs(bn);
return 0;
}
static void __net_exit bond_net_exit(struct net *net)
{
struct bond_net *bn = net_generic(net, bond_net_id);
struct bonding *bond, *tmp_bond;
LIST_HEAD(list);
bond_destroy_sysfs(bn);
/* Kill off any bonds created after unregistering bond rtnl ops */
rtnl_lock();
list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
unregister_netdevice_queue(bond->dev, &list);
unregister_netdevice_many(&list);
rtnl_unlock();
bond_destroy_proc_dir(bn);
}
static struct pernet_operations bond_net_ops = {
.init = bond_net_init,
.exit = bond_net_exit,
.id = &bond_net_id,
.size = sizeof(struct bond_net),
};
static int __init bonding_init(void)
{
int i;
int res;
pr_info("%s", bond_version);
res = bond_check_params(&bonding_defaults);
if (res)
goto out;
res = register_pernet_subsys(&bond_net_ops);
if (res)
goto out;
res = bond_netlink_init();
if (res)
goto err_link;
bond_create_debugfs();
for (i = 0; i < max_bonds; i++) {
res = bond_create(&init_net, NULL);
if (res)
goto err;
}
register_netdevice_notifier(&bond_netdev_notifier);
out:
return res;
err:
bond_destroy_debugfs();
bond_netlink_fini();
err_link:
unregister_pernet_subsys(&bond_net_ops);
goto out;
}
static void __exit bonding_exit(void)
{
unregister_netdevice_notifier(&bond_netdev_notifier);
bond_destroy_debugfs();
bond_netlink_fini();
unregister_pernet_subsys(&bond_net_ops);
#ifdef CONFIG_NET_POLL_CONTROLLER
/* Make sure we don't have an imbalance on our netpoll blocking */
net: Convert netpoll blocking api in bonding driver to be a counter A while back I made some changes to enable netpoll in the bonding driver. Among them was a per-cpu flag that indicated we were in a path that held locks which could cause the netpoll path to block in during tx, and as such the tx path should queue the frame for later use. This appears to have given rise to a regression. If one of those paths on which we hold the per-cpu flag yields the cpu, its possible for us to come back on a different cpu, leading to us clearing a different flag than we set. This results in odd netpoll drops, and BUG backtraces appearing in the log, as we check to make sure that we only clear set bits, and only set clear bits. I had though briefly about changing the offending paths so that they wouldn't sleep, but looking at my origional work more closely, it doesn't appear that a per-cpu flag is warranted. We alrady gate the checking of this flag on IFF_IN_NETPOLL, so we don't hit this in the normal tx case anyway. And practically speaking, the normal use case for netpoll is to only have one client anyway, so we're not going to erroneously queue netpoll frames when its actually safe to do so. As such, lets just convert that per-cpu flag to an atomic counter. It fixes the rescheduling bugs, is equivalent from a performance perspective and actually eliminates some code in the process. Tested by the reporter and myself, successfully Reported-by: Liang Zheng <lzheng@redhat.com> CC: Jay Vosburgh <fubar@us.ibm.com> CC: Andy Gospodarek <andy@greyhouse.net> CC: David S. Miller <davem@davemloft.net> Signed-off-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2010-12-06 16:05:50 +07:00
WARN_ON(atomic_read(&netpoll_block_tx));
#endif
}
module_init(bonding_init);
module_exit(bonding_exit);
MODULE_LICENSE("GPL");
MODULE_VERSION(DRV_VERSION);
MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");