Commit 644595f896 ("compat: Handle COMPAT_USE_64BIT_TIME in
net/socket.c") introduced a bug where the helper functions to take
either a 64-bit or compat time[spec|val] got the arguments in the wrong
order, passing the kernel stack pointer off as a user pointer (and vice
versa).
Because of the user address range check, that in turn then causes an
EFAULT due to the user pointer range checking failing for the kernel
address. Incorrectly resuling in a failed system call for 32-bit
processes with a 64-bit kernel.
On odder architectures like HP-PA (with separate user/kernel address
spaces), it can be used read kernel memory.
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The implementation of dev_ifconf() for the compat ioctl interface uses
an intermediate ifc structure allocated in userland for the duration of
the syscall. Though, it fails to initialize the padding bytes inserted
for alignment and that for leaks four bytes of kernel stack. Add an
explicit memset(0) before filling the structure to avoid the info leak.
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of updating the sk_cgrp_prioidx struct field on every send
this only updates the field when a task is moved via cgroup
infrastructure.
This allows sockets that may be used by a kernel worker thread
to be managed. For example in the iscsi case today a user can
put iscsid in a netprio cgroup and control traffic will be sent
with the correct sk_cgrp_prioidx value set but as soon as data
is sent the kernel worker thread isssues a send and sk_cgrp_prioidx
is updated with the kernel worker threads value which is the
default case.
It seems more correct to only update the field when the user
explicitly sets it via control group infrastructure. This allows
the users to manage sockets that may be used with other threads.
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes a crash
tun_chr_close -> netdev_run_todo -> tun_free_netdev -> sk_release_kernel ->
sock_release -> iput(SOCK_INODE(sock))
introduced by commit 1ab5ecb90c
The problem is that this socket is embedded in struct tun_struct, it has
no inode, iput is called on invalid inode, which modifies invalid memory
and optionally causes a crash.
sock_release also decrements sockets_in_use, this causes a bug that
"sockets: used" field in /proc/*/net/sockstat keeps on decreasing when
creating and closing tun devices.
This patch introduces a flag SOCK_EXTERNALLY_ALLOCATED that instructs
sock_release to not free the inode and not decrement sockets_in_use,
fixing both memory corruption and sockets_in_use underflow.
It should be backported to 3.3 an 3.4 stabke.
Signed-off-by: Mikulas Patocka <mikulas@artax.karlin.mff.cuni.cz>
Cc: stable@kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull percpu updates from Tejun Heo:
"Contains Alex Shi's three patches to remove percpu_xxx() which overlap
with this_cpu_xxx(). There shouldn't be any functional change."
* 'for-3.5' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu:
percpu: remove percpu_xxx() functions
x86: replace percpu_xxx funcs with this_cpu_xxx
net: replace percpu_xxx funcs with this_cpu_xxx or __this_cpu_xxx
Standardize the net core ratelimited logging functions.
Coalesce formats, align arguments.
Change a printk then vprintk sequence to use printf extension %pV.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
percpu_xxx funcs are duplicated with this_cpu_xxx funcs, so replace
them for further code clean up.
And in preempt safe scenario, __this_cpu_xxx funcs may has a bit
better performance since __this_cpu_xxx has no redundant
preempt_enable/preempt_disable on some architectures.
Signed-off-by: Alex Shi <alex.shi@intel.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: David S. Miller <davem@davemloft.net>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
iov of more than 8 entries are allocated in sendmsg()/recvmsg() through
sock_kmalloc()
As these allocations are temporary only and small enough, it makes sense
to use plain kmalloc() and avoid sk_omem_alloc atomic overhead.
Slightly changed fast path to be even faster.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Mike Waychison <mikew@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the netfilter code is modified to use register_net_sysctl_table the
kernel fails to boot because the per net sysctl infrasturce is not setup
soon enough. So to avoid races call net_sysctl_init from sock_init().
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use of "unsigned int" is preferred to bare "unsigned" in net tree.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
commit 2f53384424 (tcp: allow splice() to build full TSO packets) added
a regression for splice() calls using SPLICE_F_MORE.
We need to call tcp_flush() at the end of the last page processed in
tcp_sendpages(), or else transmits can be deferred and future sends
stall.
Add a new internal flag, MSG_SENDPAGE_NOTLAST, acting like MSG_MORE, but
with different semantic.
For all sendpage() providers, its a transparent change. Only
sock_sendpage() and tcp_sendpages() can differentiate the two different
flags provided by pipe_to_sendpage()
Reported-by: Tom Herbert <therbert@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Cc: H.K. Jerry Chu <hkchu@google.com>
Cc: Maciej Żenczykowski <maze@google.com>
Cc: Mahesh Bandewar <maheshb@google.com>
Cc: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail>com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull x32 support for x86-64 from Ingo Molnar:
"This tree introduces the X32 binary format and execution mode for x86:
32-bit data space binaries using 64-bit instructions and 64-bit kernel
syscalls.
This allows applications whose working set fits into a 32 bits address
space to make use of 64-bit instructions while using a 32-bit address
space with shorter pointers, more compressed data structures, etc."
Fix up trivial context conflicts in arch/x86/{Kconfig,vdso/vma.c}
* 'x86-x32-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (71 commits)
x32: Fix alignment fail in struct compat_siginfo
x32: Fix stupid ia32/x32 inversion in the siginfo format
x32: Add ptrace for x32
x32: Switch to a 64-bit clock_t
x32: Provide separate is_ia32_task() and is_x32_task() predicates
x86, mtrr: Use explicit sizing and padding for the 64-bit ioctls
x86/x32: Fix the binutils auto-detect
x32: Warn and disable rather than error if binutils too old
x32: Only clear TIF_X32 flag once
x32: Make sure TS_COMPAT is cleared for x32 tasks
fs: Remove missed ->fds_bits from cessation use of fd_set structs internally
fs: Fix close_on_exec pointer in alloc_fdtable
x32: Drop non-__vdso weak symbols from the x32 VDSO
x32: Fix coding style violations in the x32 VDSO code
x32: Add x32 VDSO support
x32: Allow x32 to be configured
x32: If configured, add x32 system calls to system call tables
x32: Handle process creation
x32: Signal-related system calls
x86: Add #ifdef CONFIG_COMPAT to <asm/sys_ia32.h>
...
The following 4 functions:
move_addr_to_kernel
move_addr_to_user
verify_iovec
verify_compat_iovec
are always effectively called with a sockaddr_storage.
Make this explicit by changing their signature.
This removes a large number of casts from sockaddr_storage to sockaddr.
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use helper functions aware of COMPAT_USE_64BIT_TIME to write struct
timeval and struct timespec to userspace in net/socket.c.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
commit a9b3cd7f32 (rcu: convert uses of rcu_assign_pointer(x, NULL) to
RCU_INIT_POINTER) did a lot of incorrect changes, since it did a
complete conversion of rcu_assign_pointer(x, y) to RCU_INIT_POINTER(x,
y).
We miss needed barriers, even on x86, when y is not NULL.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next: (1958 commits)
net: pack skb_shared_info more efficiently
net_sched: red: split red_parms into parms and vars
net_sched: sfq: extend limits
cnic: Improve error recovery on bnx2x devices
cnic: Re-init dev->stats_addr after chip reset
net_sched: Bug in netem reordering
bna: fix sparse warnings/errors
bna: make ethtool_ops and strings const
xgmac: cleanups
net: make ethtool_ops const
vmxnet3" make ethtool ops const
xen-netback: make ops structs const
virtio_net: Pass gfp flags when allocating rx buffers.
ixgbe: FCoE: Add support for ndo_get_fcoe_hbainfo() call
netdev: FCoE: Add new ndo_get_fcoe_hbainfo() call
igb: reset PHY after recovering from PHY power down
igb: add basic runtime PM support
igb: Add support for byte queue limits.
e1000: cleanup CE4100 MDIO registers access
e1000: unmap ce4100_gbe_mdio_base_virt in e1000_remove
...
We're doing some odd things there, which already messes up various users
(see the net/socket.c code that this removes), and it was going to add
yet more crud to the block layer because of the incorrect error code
translation.
ENOIOCTLCMD is not an error return that should be returned to user mode
from the "ioctl()" system call, but it should *not* be translated as
EINVAL ("Invalid argument"). It should be translated as ENOTTY
("Inappropriate ioctl for device").
That EINVAL confusion has apparently so permeated some code that the
block layer actually checks for it, which is sad. We continue to do so
for now, but add a big comment about how wrong that is, and we should
remove it entirely eventually. In the meantime, this tries to keep the
changes localized to just the EINVAL -> ENOTTY fix, and removing code
that makes it harder to do the right thing.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Define special location values for RX NFC that request the driver to
select the actual rule location. This allows for implementation on
devices that use hash-based filter lookup, whereas currently the API is
more suited to devices with TCAM lookup or linear search.
In ethtool_set_rxnfc() and the compat wrapper ethtool_ioctl(), copy
the structure back to user-space after insertion so that the actual
location is returned.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds in the infrastructure code to create the network priority
cgroup. The cgroup, in addition to the standard processes file creates two
control files:
1) prioidx - This is a read-only file that exports the index of this cgroup.
This is a value that is both arbitrary and unique to a cgroup in this subsystem,
and is used to index the per-device priority map
2) priomap - This is a writeable file. On read it reports a table of 2-tuples
<name:priority> where name is the name of a network interface and priority is
indicates the priority assigned to frames egresessing on the named interface and
originating from a pid in this cgroup
This cgroup allows for skb priority to be set prior to a root qdisc getting
selected. This is benenficial for DCB enabled systems, in that it allows for any
application to use dcb configured priorities so without application modification
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
CC: Robert Love <robert.w.love@intel.com>
CC: "David S. Miller" <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
The 802.1X EAPOL handshake hostapd does requires
knowing whether the frame was ack'ed by the peer.
Currently, we fudge this pretty badly by not even
transmitting the frame as a normal data frame but
injecting it with radiotap and getting the status
out of radiotap monitor as well. This is rather
complex, confuses users (mon.wlan0 presence) and
doesn't work with all hardware.
To get rid of that hack, introduce a real wifi TX
status option for data frame transmissions.
This works similar to the existing TX timestamping
in that it reflects the SKB back to the socket's
error queue with a SCM_WIFI_STATUS cmsg that has
an int indicating ACK status (0/1).
Since it is possible that at some point we will
want to have TX timestamping and wifi status in a
single errqueue SKB (there's little point in not
doing that), redefine SO_EE_ORIGIN_TIMESTAMPING
to SO_EE_ORIGIN_TXSTATUS which can collect more
than just the timestamp; keep the old constant
as an alias of course. Currently the internal APIs
don't make that possible, but it wouldn't be hard
to split them up in a way that makes it possible.
Thanks to Neil Horman for helping me figure out
the functions that add the control messages.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Dereferencing a user pointer directly from kernel-space without going
through the copy_from_user family of functions is a bad idea. Two of
such usages can be found in the sendmsg code path called from sendmmsg,
added by
commit c71d8ebe7a upstream.
commit 5b47b8038f183b44d2d8ff1c7d11a5c1be706b34 in the 3.0-stable tree.
Usages are performed through memcmp() and memcpy() directly. Fix those
by using the already copied msg_sys structure instead of the __user *msg
structure. Note that msg_sys can be set to NULL by verify_compat_iovec()
or verify_iovec(), which requires additional NULL pointer checks.
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: David Goulet <dgoulet@ev0ke.net>
CC: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
CC: Anton Blanchard <anton@samba.org>
CC: David S. Miller <davem@davemloft.net>
CC: stable <stable@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The sendmmsg() introduced by commit 228e548e "net: Add sendmmsg socket system
call" is capable of sending to multiple different destination addresses.
SMACK is using destination's address for checking sendmsg() permission.
However, security_socket_sendmsg() is called for only once even if multiple
different destination addresses are passed to sendmmsg().
Therefore, we need to call security_socket_sendmsg() for each destination
address rather than only the first destination address.
Since calling security_socket_sendmsg() every time when only single destination
address was passed to sendmmsg() is a waste of time, omit calling
security_socket_sendmsg() unless destination address of previous datagram and
that of current datagram differs.
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Acked-by: Anton Blanchard <anton@samba.org>
Cc: stable <stable@kernel.org> [3.0+]
Signed-off-by: David S. Miller <davem@davemloft.net>
To limit the amount of time we can spend in sendmmsg, cap the
number of elements to UIO_MAXIOV (currently 1024).
For error handling an application using sendmmsg needs to retry at
the first unsent message, so capping is simpler and requires less
application logic than returning EINVAL.
Signed-off-by: Anton Blanchard <anton@samba.org>
Cc: stable <stable@kernel.org> [3.0+]
Signed-off-by: David S. Miller <davem@davemloft.net>
sendmmsg uses a similar error return strategy as recvmmsg but it
turns out to be a confusing way to communicate errors.
The current code stores the error code away and returns it on the next
sendmmsg call. This means a call with completely valid arguments could
get an error from a previous call.
Change things so we only return an error if no datagrams could be sent.
If less than the requested number of messages were sent, the application
must retry starting at the first failed one and if the problem is
persistent the error will be returned.
This matches the behaviour of other syscalls like read/write - it
is not an error if less than the requested number of elements are sent.
Signed-off-by: Anton Blanchard <anton@samba.org>
Cc: stable <stable@kernel.org> [3.0+]
Signed-off-by: David S. Miller <davem@davemloft.net>
When assigning a NULL value to an RCU protected pointer, no barrier
is needed. The rcu_assign_pointer, used to handle that but will soon
change to not handle the special case.
Convert all rcu_assign_pointer of NULL value.
//smpl
@@ expression P; @@
- rcu_assign_pointer(P, NULL)
+ RCU_INIT_POINTER(P, NULL)
// </smpl>
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (32 commits)
tg3: Remove 5719 jumbo frames and TSO blocks
tg3: Break larger frags into 4k chunks for 5719
tg3: Add tx BD budgeting code
tg3: Consolidate code that calls tg3_tx_set_bd()
tg3: Add partial fragment unmapping code
tg3: Generalize tg3_skb_error_unmap()
tg3: Remove short DMA check for 1st fragment
tg3: Simplify tx bd assignments
tg3: Reintroduce tg3_tx_ring_info
ASIX: Use only 11 bits of header for data size
ASIX: Simplify condition in rx_fixup()
Fix cdc-phonet build
bonding: reduce noise during init
bonding: fix string comparison errors
net: Audit drivers to identify those needing IFF_TX_SKB_SHARING cleared
net: add IFF_SKB_TX_SHARED flag to priv_flags
net: sock_sendmsg_nosec() is static
forcedeth: fix vlans
gianfar: fix bug caused by 87c288c6e9
gro: Only reset frag0 when skb can be pulled
...
Workloads using pipes and sockets hit inode_sb_list_lock contention.
superblock s_inodes list is needed for quota, dirty, pagecache and
fsnotify management. pipe/anon/socket fs are clearly not candidates for
these.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next-2.6: (1446 commits)
macvlan: fix panic if lowerdev in a bond
tg3: Add braces around 5906 workaround.
tg3: Fix NETIF_F_LOOPBACK error
macvlan: remove one synchronize_rcu() call
networking: NET_CLS_ROUTE4 depends on INET
irda: Fix error propagation in ircomm_lmp_connect_response()
irda: Kill set but unused variable 'bytes' in irlan_check_command_param()
irda: Kill set but unused variable 'clen' in ircomm_connect_indication()
rxrpc: Fix set but unused variable 'usage' in rxrpc_get_transport()
be2net: Kill set but unused variable 'req' in lancer_fw_download()
irda: Kill set but unused vars 'saddr' and 'daddr' in irlan_provider_connect_indication()
atl1c: atl1c_resume() is only used when CONFIG_PM_SLEEP is defined.
rxrpc: Fix set but unused variable 'usage' in rxrpc_get_peer().
rxrpc: Kill set but unused variable 'local' in rxrpc_UDP_error_handler()
rxrpc: Kill set but unused variable 'sp' in rxrpc_process_connection()
rxrpc: Kill set but unused variable 'sp' in rxrpc_rotate_tx_window()
pkt_sched: Kill set but unused variable 'protocol' in tc_classify()
isdn: capi: Use pr_debug() instead of ifdefs.
tg3: Update version to 3.119
tg3: Apply rx_discards fix to 5719/5720
...
Fix up trivial conflicts in arch/x86/Kconfig and net/mac80211/agg-tx.c
as per Davem.
recvmmsg fails on a raw socket with EINVAL. The reason for this is
packet_recvmsg checks the incoming flags:
err = -EINVAL;
if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
goto out;
This patch strips out MSG_WAITFORONE when calling recvmmsg which
fixes the issue.
Signed-off-by: Anton Blanchard <anton@samba.org>
Cc: stable@kernel.org [2.6.34+]
Signed-off-by: David S. Miller <davem@davemloft.net>
The rcu callback wq_free_rcu() just calls a kfree(),
so we use kfree_rcu() instead of the call_rcu(wq_free_rcu).
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
This patch adds a multiple message send syscall and is the send
version of the existing recvmmsg syscall. This is heavily
based on the patch by Arnaldo that added recvmmsg.
I wrote a microbenchmark to test the performance gains of using
this new syscall:
http://ozlabs.org/~anton/junkcode/sendmmsg_test.c
The test was run on a ppc64 box with a 10 Gbit network card. The
benchmark can send both UDP and RAW ethernet packets.
64B UDP
batch pkts/sec
1 804570
2 872800 (+ 8 %)
4 916556 (+14 %)
8 939712 (+17 %)
16 952688 (+18 %)
32 956448 (+19 %)
64 964800 (+20 %)
64B raw socket
batch pkts/sec
1 1201449
2 1350028 (+12 %)
4 1461416 (+22 %)
8 1513080 (+26 %)
16 1541216 (+28 %)
32 1553440 (+29 %)
64 1557888 (+30 %)
We see a 20% improvement in throughput on UDP send and 30%
on raw socket send.
[ Add sparc syscall entries. -DaveM ]
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This change is meant to add an ntuple data extensions to the rx network flow
classification specifiers. The idea is to allow ntuple to be displayed via
the network flow classification interface.
The first patch had some left over stuff from the original flow extension
flags I had added. That bit is removed in this patch.
The second had some left over comments that stated we ignored bits in the
masks when we actually match them.
This work is based on input from Ben Hutchings.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Reviewed-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This structure was accidentally defined such that its layout can
differ between 32-bit and 64-bit processes. Add compat structure
definitions and an ioctl wrapper function.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Acked-by: Alexander Duyck <alexander.h.duyck@intel.com>
Cc: stable@kernel.org [2.6.30+]
Signed-off-by: David S. Miller <davem@davemloft.net>
Use __force to quiet sparse warnings for cases where the code
is simulating user space pointers.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add proper RCU annotations/verbs to sk_wq and wq members
Fix __sctp_write_space() sk_sleep() abuse (and sock->wq access)
Fix sunrpc sk_sleep() abuse too
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The problem that this patch aims to fix is vfsmount refcounting scalability.
We need to take a reference on the vfsmount for every successful path lookup,
which often go to the same mount point.
The fundamental difficulty is that a "simple" reference count can never be made
scalable, because any time a reference is dropped, we must check whether that
was the last reference. To do that requires communication with all other CPUs
that may have taken a reference count.
We can make refcounts more scalable in a couple of ways, involving keeping
distributed counters, and checking for the global-zero condition less
frequently.
- check the global sum once every interval (this will delay zero detection
for some interval, so it's probably a showstopper for vfsmounts).
- keep a local count and only taking the global sum when local reaches 0 (this
is difficult for vfsmounts, because we can't hold preempt off for the life of
a reference, so a counter would need to be per-thread or tied strongly to a
particular CPU which requires more locking).
- keep a local difference of increments and decrements, which allows us to sum
the total difference and hence find the refcount when summing all CPUs. Then,
keep a single integer "long" refcount for slow and long lasting references,
and only take the global sum of local counters when the long refcount is 0.
This last scheme is what I implemented here. Attached mounts and process root
and working directory references are "long" references, and everything else is
a short reference.
This allows scalable vfsmount references during path walking over mounted
subtrees and unattached (lazy umounted) mounts with processes still running
in them.
This results in one fewer atomic op in the fastpath: mntget is now just a
per-CPU inc, rather than an atomic inc; and mntput just requires a spinlock
and non-atomic decrement in the common case. However code is otherwise bigger
and heavier, so single threaded performance is basically a wash.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Regardless of how much we possibly try to scale dcache, there is likely
always going to be some fundamental contention when adding or removing children
under the same parent. Pseudo filesystems do not seem need to have connected
dentries because by definition they are disconnected.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Reduce some branches and memory accesses in dcache lookup by adding dentry
flags to indicate common d_ops are set, rather than having to check them.
This saves a pointer memory access (dentry->d_op) in common path lookup
situations, and saves another pointer load and branch in cases where we
have d_op but not the particular operation.
Patched with:
git grep -E '[.>]([[:space:]])*d_op([[:space:]])*=' | xargs sed -e 's/\([^\t ]*\)->d_op = \(.*\);/d_set_d_op(\1, \2);/' -e 's/\([^\t ]*\)\.d_op = \(.*\);/d_set_d_op(\&\1, \2);/' -i
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Pseudo filesystems that don't put inode on RCU list or reachable by
rcu-walk dentries do not need to RCU free their inodes.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
RCU free the struct inode. This will allow:
- Subsequent store-free path walking patch. The inode must be consulted for
permissions when walking, so an RCU inode reference is a must.
- sb_inode_list_lock to be moved inside i_lock because sb list walkers who want
to take i_lock no longer need to take sb_inode_list_lock to walk the list in
the first place. This will simplify and optimize locking.
- Could remove some nested trylock loops in dcache code
- Could potentially simplify things a bit in VM land. Do not need to take the
page lock to follow page->mapping.
The downsides of this is the performance cost of using RCU. In a simple
creat/unlink microbenchmark, performance drops by about 10% due to inability to
reuse cache-hot slab objects. As iterations increase and RCU freeing starts
kicking over, this increases to about 20%.
In cases where inode lifetimes are longer (ie. many inodes may be allocated
during the average life span of a single inode), a lot of this cache reuse is
not applicable, so the regression caused by this patch is smaller.
The cache-hot regression could largely be avoided by using SLAB_DESTROY_BY_RCU,
however this adds some complexity to list walking and store-free path walking,
so I prefer to implement this at a later date, if it is shown to be a win in
real situations. I haven't found a regression in any non-micro benchmark so I
doubt it will be a problem.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
