linux_dsm_epyc7002/net/core/netclassid_cgroup.c

132 lines
3.0 KiB
C
Raw Normal View History

/*
* net/core/netclassid_cgroup.c Classid Cgroupfs Handling
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*
* Authors: Thomas Graf <tgraf@suug.ch>
*/
#include <linux/slab.h>
#include <linux/cgroup.h>
#include <linux/fdtable.h>
#include <linux/sched/task.h>
#include <net/cls_cgroup.h>
#include <net/sock.h>
static inline struct cgroup_cls_state *css_cls_state(struct cgroup_subsys_state *css)
{
return css ? container_of(css, struct cgroup_cls_state, css) : NULL;
}
struct cgroup_cls_state *task_cls_state(struct task_struct *p)
{
cgroup: net_cls: fix false-positive "suspicious RCU usage" In dev_queue_xmit() net_cls protected with rcu-bh. [ 270.730026] =============================== [ 270.730029] [ INFO: suspicious RCU usage. ] [ 270.730033] 4.2.0-rc3+ #2 Not tainted [ 270.730036] ------------------------------- [ 270.730040] include/linux/cgroup.h:353 suspicious rcu_dereference_check() usage! [ 270.730041] other info that might help us debug this: [ 270.730043] rcu_scheduler_active = 1, debug_locks = 1 [ 270.730045] 2 locks held by dhclient/748: [ 270.730046] #0: (rcu_read_lock_bh){......}, at: [<ffffffff81682b70>] __dev_queue_xmit+0x50/0x960 [ 270.730085] #1: (&qdisc_tx_lock){+.....}, at: [<ffffffff81682d60>] __dev_queue_xmit+0x240/0x960 [ 270.730090] stack backtrace: [ 270.730096] CPU: 0 PID: 748 Comm: dhclient Not tainted 4.2.0-rc3+ #2 [ 270.730098] Hardware name: OpenStack Foundation OpenStack Nova, BIOS Bochs 01/01/2011 [ 270.730100] 0000000000000001 ffff8800bafeba58 ffffffff817ad487 0000000000000007 [ 270.730103] ffff880232a0a780 ffff8800bafeba88 ffffffff810ca4f2 ffff88022fb23e00 [ 270.730105] ffff880232a0a780 ffff8800bafebb68 ffff8800bafebb68 ffff8800bafebaa8 [ 270.730108] Call Trace: [ 270.730121] [<ffffffff817ad487>] dump_stack+0x4c/0x65 [ 270.730148] [<ffffffff810ca4f2>] lockdep_rcu_suspicious+0xe2/0x120 [ 270.730153] [<ffffffff816a62d2>] task_cls_state+0x92/0xa0 [ 270.730158] [<ffffffffa00b534f>] cls_cgroup_classify+0x4f/0x120 [cls_cgroup] [ 270.730164] [<ffffffff816aac74>] tc_classify_compat+0x74/0xc0 [ 270.730166] [<ffffffff816ab573>] tc_classify+0x33/0x90 [ 270.730170] [<ffffffffa00bcb0a>] htb_enqueue+0xaa/0x4a0 [sch_htb] [ 270.730172] [<ffffffff81682e26>] __dev_queue_xmit+0x306/0x960 [ 270.730174] [<ffffffff81682b70>] ? __dev_queue_xmit+0x50/0x960 [ 270.730176] [<ffffffff816834a3>] dev_queue_xmit_sk+0x13/0x20 [ 270.730185] [<ffffffff81787770>] dev_queue_xmit+0x10/0x20 [ 270.730187] [<ffffffff8178b91c>] packet_snd.isra.62+0x54c/0x760 [ 270.730190] [<ffffffff8178be25>] packet_sendmsg+0x2f5/0x3f0 [ 270.730203] [<ffffffff81665245>] ? sock_def_readable+0x5/0x190 [ 270.730210] [<ffffffff817b64bb>] ? _raw_spin_unlock+0x2b/0x40 [ 270.730216] [<ffffffff8173bcbc>] ? unix_dgram_sendmsg+0x5cc/0x640 [ 270.730219] [<ffffffff8165f367>] sock_sendmsg+0x47/0x50 [ 270.730221] [<ffffffff8165f42f>] sock_write_iter+0x7f/0xd0 [ 270.730232] [<ffffffff811fd4c7>] __vfs_write+0xa7/0xf0 [ 270.730234] [<ffffffff811fe5b8>] vfs_write+0xb8/0x190 [ 270.730236] [<ffffffff811fe8c2>] SyS_write+0x52/0xb0 [ 270.730239] [<ffffffff817b6bae>] entry_SYSCALL_64_fastpath+0x12/0x76 Signed-off-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-07-22 16:23:20 +07:00
return css_cls_state(task_css_check(p, net_cls_cgrp_id,
rcu_read_lock_bh_held()));
}
EXPORT_SYMBOL_GPL(task_cls_state);
static struct cgroup_subsys_state *
cgrp_css_alloc(struct cgroup_subsys_state *parent_css)
{
struct cgroup_cls_state *cs;
cs = kzalloc(sizeof(*cs), GFP_KERNEL);
if (!cs)
return ERR_PTR(-ENOMEM);
return &cs->css;
}
static int cgrp_css_online(struct cgroup_subsys_state *css)
{
struct cgroup_cls_state *cs = css_cls_state(css);
struct cgroup_cls_state *parent = css_cls_state(css->parent);
if (parent)
cs->classid = parent->classid;
return 0;
}
static void cgrp_css_free(struct cgroup_subsys_state *css)
{
kfree(css_cls_state(css));
}
static int update_classid_sock(const void *v, struct file *file, unsigned n)
{
int err;
struct socket *sock = sock_from_file(file, &err);
sock, cgroup: add sock->sk_cgroup In cgroup v1, dealing with cgroup membership was difficult because the number of membership associations was unbound. As a result, cgroup v1 grew several controllers whose primary purpose is either tagging membership or pull in configuration knobs from other subsystems so that cgroup membership test can be avoided. net_cls and net_prio controllers are examples of the latter. They allow configuring network-specific attributes from cgroup side so that network subsystem can avoid testing cgroup membership; unfortunately, these are not only cumbersome but also problematic. Both net_cls and net_prio aren't properly hierarchical. Both inherit configuration from the parent on creation but there's no interaction afterwards. An ancestor doesn't restrict the behavior in its subtree in anyway and configuration changes aren't propagated downwards. Especially when combined with cgroup delegation, this is problematic because delegatees can mess up whatever network configuration implemented at the system level. net_prio would allow the delegatees to set whatever priority value regardless of CAP_NET_ADMIN and net_cls the same for classid. While it is possible to solve these issues from controller side by implementing hierarchical allowable ranges in both controllers, it would involve quite a bit of complexity in the controllers and further obfuscate network configuration as it becomes even more difficult to tell what's actually being configured looking from the network side. While not much can be done for v1 at this point, as membership handling is sane on cgroup v2, it'd be better to make cgroup matching behave like other network matches and classifiers than introducing further complications. In preparation, this patch updates sock->sk_cgrp_data handling so that it points to the v2 cgroup that sock was created in until either net_prio or net_cls is used. Once either of the two is used, sock->sk_cgrp_data reverts to its previous role of carrying prioidx and classid. This is to avoid adding yet another cgroup related field to struct sock. As the mode switching can happen at most once per boot, the switching mechanism is aimed at lowering hot path overhead. It may leak a finite, likely small, number of cgroup refs and report spurious prioidx or classid on switching; however, dynamic updates of prioidx and classid have always been racy and lossy - socks between creation and fd installation are never updated, config changes don't update existing sockets at all, and prioidx may index with dead and recycled cgroup IDs. Non-critical inaccuracies from small race windows won't make any noticeable difference. This patch doesn't make use of the pointer yet. The following patch will implement netfilter match for cgroup2 membership. v2: Use sock_cgroup_data to avoid inflating struct sock w/ another cgroup specific field. v3: Add comments explaining why sock_data_prioidx() and sock_data_classid() use different fallback values. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Daniel Borkmann <daniel@iogearbox.net> Cc: Daniel Wagner <daniel.wagner@bmw-carit.de> CC: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-12-08 05:38:53 +07:00
if (sock) {
spin_lock(&cgroup_sk_update_lock);
sock_cgroup_set_classid(&sock->sk->sk_cgrp_data,
(unsigned long)v);
sock, cgroup: add sock->sk_cgroup In cgroup v1, dealing with cgroup membership was difficult because the number of membership associations was unbound. As a result, cgroup v1 grew several controllers whose primary purpose is either tagging membership or pull in configuration knobs from other subsystems so that cgroup membership test can be avoided. net_cls and net_prio controllers are examples of the latter. They allow configuring network-specific attributes from cgroup side so that network subsystem can avoid testing cgroup membership; unfortunately, these are not only cumbersome but also problematic. Both net_cls and net_prio aren't properly hierarchical. Both inherit configuration from the parent on creation but there's no interaction afterwards. An ancestor doesn't restrict the behavior in its subtree in anyway and configuration changes aren't propagated downwards. Especially when combined with cgroup delegation, this is problematic because delegatees can mess up whatever network configuration implemented at the system level. net_prio would allow the delegatees to set whatever priority value regardless of CAP_NET_ADMIN and net_cls the same for classid. While it is possible to solve these issues from controller side by implementing hierarchical allowable ranges in both controllers, it would involve quite a bit of complexity in the controllers and further obfuscate network configuration as it becomes even more difficult to tell what's actually being configured looking from the network side. While not much can be done for v1 at this point, as membership handling is sane on cgroup v2, it'd be better to make cgroup matching behave like other network matches and classifiers than introducing further complications. In preparation, this patch updates sock->sk_cgrp_data handling so that it points to the v2 cgroup that sock was created in until either net_prio or net_cls is used. Once either of the two is used, sock->sk_cgrp_data reverts to its previous role of carrying prioidx and classid. This is to avoid adding yet another cgroup related field to struct sock. As the mode switching can happen at most once per boot, the switching mechanism is aimed at lowering hot path overhead. It may leak a finite, likely small, number of cgroup refs and report spurious prioidx or classid on switching; however, dynamic updates of prioidx and classid have always been racy and lossy - socks between creation and fd installation are never updated, config changes don't update existing sockets at all, and prioidx may index with dead and recycled cgroup IDs. Non-critical inaccuracies from small race windows won't make any noticeable difference. This patch doesn't make use of the pointer yet. The following patch will implement netfilter match for cgroup2 membership. v2: Use sock_cgroup_data to avoid inflating struct sock w/ another cgroup specific field. v3: Add comments explaining why sock_data_prioidx() and sock_data_classid() use different fallback values. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Daniel Borkmann <daniel@iogearbox.net> Cc: Daniel Wagner <daniel.wagner@bmw-carit.de> CC: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-12-08 05:38:53 +07:00
spin_unlock(&cgroup_sk_update_lock);
}
return 0;
}
cgroup, net_cls: iterate the fds of only the tasks which are being migrated The net_cls controller controls the classid field of each socket which is associated with the cgroup. Because the classid is per-socket attribute, when a task migrates to another cgroup or the configured classid of the cgroup changes, the controller needs to walk all sockets and update the classid value, which was implemented by 3b13758f51de ("cgroups: Allow dynamically changing net_classid"). While the approach is not scalable, migrating tasks which have a lot of fds attached to them is rare and the cost is born by the ones initiating the operations. However, for simplicity, both the migration and classid config change paths call update_classid() which scans all fds of all tasks in the target css. This is an overkill for the migration path which only needs to cover a much smaller subset of tasks which are actually getting migrated in. On cgroup v1, this can lead to unexpected scalability issues when one tries to migrate a task or process into a net_cls cgroup which already contains a lot of fds. Even if the migration traget doesn't have many to get scanned, update_classid() ends up scanning all fds in the target cgroup which can be extremely numerous. Unfortunately, on cgroup v2 which doesn't use net_cls, the problem is even worse. Before bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"), cgroup core would call the ->css_attach callback even for controllers which don't see actual migration to a different css. As net_cls is always disabled but still mounted on cgroup v2, whenever a process is migrated on the cgroup v2 hierarchy, net_cls sees identity migration from root to root and cgroup core used to call ->css_attach callback for those. The net_cls ->css_attach ends up calling update_classid() on the root net_cls css to which all processes on the system belong to as the controller isn't used. This makes any cgroup v2 migration O(total_number_of_fds_on_the_system) which is horrible and easily leads to noticeable stalls triggering RCU stall warnings and so on. The worst symptom is already fixed in upstream by bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"); however, backporting that commit is too invasive and we want to avoid other cases too. This patch updates net_cls's cgrp_attach() to iterate fds of only the processes which are actually getting migrated. This removes the surprising migration cost which is dependent on the total number of fds in the target cgroup. As this leaves write_classid() the only user of update_classid(), open-code the helper into write_classid(). Reported-by: David Goode <dgoode@fb.com> Fixes: 3b13758f51de ("cgroups: Allow dynamically changing net_classid") Cc: stable@vger.kernel.org # v4.4+ Cc: Nina Schiff <ninasc@fb.com> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-15 06:25:56 +07:00
static void cgrp_attach(struct cgroup_taskset *tset)
{
cgroup, net_cls: iterate the fds of only the tasks which are being migrated The net_cls controller controls the classid field of each socket which is associated with the cgroup. Because the classid is per-socket attribute, when a task migrates to another cgroup or the configured classid of the cgroup changes, the controller needs to walk all sockets and update the classid value, which was implemented by 3b13758f51de ("cgroups: Allow dynamically changing net_classid"). While the approach is not scalable, migrating tasks which have a lot of fds attached to them is rare and the cost is born by the ones initiating the operations. However, for simplicity, both the migration and classid config change paths call update_classid() which scans all fds of all tasks in the target css. This is an overkill for the migration path which only needs to cover a much smaller subset of tasks which are actually getting migrated in. On cgroup v1, this can lead to unexpected scalability issues when one tries to migrate a task or process into a net_cls cgroup which already contains a lot of fds. Even if the migration traget doesn't have many to get scanned, update_classid() ends up scanning all fds in the target cgroup which can be extremely numerous. Unfortunately, on cgroup v2 which doesn't use net_cls, the problem is even worse. Before bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"), cgroup core would call the ->css_attach callback even for controllers which don't see actual migration to a different css. As net_cls is always disabled but still mounted on cgroup v2, whenever a process is migrated on the cgroup v2 hierarchy, net_cls sees identity migration from root to root and cgroup core used to call ->css_attach callback for those. The net_cls ->css_attach ends up calling update_classid() on the root net_cls css to which all processes on the system belong to as the controller isn't used. This makes any cgroup v2 migration O(total_number_of_fds_on_the_system) which is horrible and easily leads to noticeable stalls triggering RCU stall warnings and so on. The worst symptom is already fixed in upstream by bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"); however, backporting that commit is too invasive and we want to avoid other cases too. This patch updates net_cls's cgrp_attach() to iterate fds of only the processes which are actually getting migrated. This removes the surprising migration cost which is dependent on the total number of fds in the target cgroup. As this leaves write_classid() the only user of update_classid(), open-code the helper into write_classid(). Reported-by: David Goode <dgoode@fb.com> Fixes: 3b13758f51de ("cgroups: Allow dynamically changing net_classid") Cc: stable@vger.kernel.org # v4.4+ Cc: Nina Schiff <ninasc@fb.com> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-15 06:25:56 +07:00
struct cgroup_subsys_state *css;
struct task_struct *p;
cgroup, net_cls: iterate the fds of only the tasks which are being migrated The net_cls controller controls the classid field of each socket which is associated with the cgroup. Because the classid is per-socket attribute, when a task migrates to another cgroup or the configured classid of the cgroup changes, the controller needs to walk all sockets and update the classid value, which was implemented by 3b13758f51de ("cgroups: Allow dynamically changing net_classid"). While the approach is not scalable, migrating tasks which have a lot of fds attached to them is rare and the cost is born by the ones initiating the operations. However, for simplicity, both the migration and classid config change paths call update_classid() which scans all fds of all tasks in the target css. This is an overkill for the migration path which only needs to cover a much smaller subset of tasks which are actually getting migrated in. On cgroup v1, this can lead to unexpected scalability issues when one tries to migrate a task or process into a net_cls cgroup which already contains a lot of fds. Even if the migration traget doesn't have many to get scanned, update_classid() ends up scanning all fds in the target cgroup which can be extremely numerous. Unfortunately, on cgroup v2 which doesn't use net_cls, the problem is even worse. Before bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"), cgroup core would call the ->css_attach callback even for controllers which don't see actual migration to a different css. As net_cls is always disabled but still mounted on cgroup v2, whenever a process is migrated on the cgroup v2 hierarchy, net_cls sees identity migration from root to root and cgroup core used to call ->css_attach callback for those. The net_cls ->css_attach ends up calling update_classid() on the root net_cls css to which all processes on the system belong to as the controller isn't used. This makes any cgroup v2 migration O(total_number_of_fds_on_the_system) which is horrible and easily leads to noticeable stalls triggering RCU stall warnings and so on. The worst symptom is already fixed in upstream by bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"); however, backporting that commit is too invasive and we want to avoid other cases too. This patch updates net_cls's cgrp_attach() to iterate fds of only the processes which are actually getting migrated. This removes the surprising migration cost which is dependent on the total number of fds in the target cgroup. As this leaves write_classid() the only user of update_classid(), open-code the helper into write_classid(). Reported-by: David Goode <dgoode@fb.com> Fixes: 3b13758f51de ("cgroups: Allow dynamically changing net_classid") Cc: stable@vger.kernel.org # v4.4+ Cc: Nina Schiff <ninasc@fb.com> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-15 06:25:56 +07:00
cgroup_taskset_for_each(p, css, tset) {
task_lock(p);
cgroup, net_cls: iterate the fds of only the tasks which are being migrated The net_cls controller controls the classid field of each socket which is associated with the cgroup. Because the classid is per-socket attribute, when a task migrates to another cgroup or the configured classid of the cgroup changes, the controller needs to walk all sockets and update the classid value, which was implemented by 3b13758f51de ("cgroups: Allow dynamically changing net_classid"). While the approach is not scalable, migrating tasks which have a lot of fds attached to them is rare and the cost is born by the ones initiating the operations. However, for simplicity, both the migration and classid config change paths call update_classid() which scans all fds of all tasks in the target css. This is an overkill for the migration path which only needs to cover a much smaller subset of tasks which are actually getting migrated in. On cgroup v1, this can lead to unexpected scalability issues when one tries to migrate a task or process into a net_cls cgroup which already contains a lot of fds. Even if the migration traget doesn't have many to get scanned, update_classid() ends up scanning all fds in the target cgroup which can be extremely numerous. Unfortunately, on cgroup v2 which doesn't use net_cls, the problem is even worse. Before bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"), cgroup core would call the ->css_attach callback even for controllers which don't see actual migration to a different css. As net_cls is always disabled but still mounted on cgroup v2, whenever a process is migrated on the cgroup v2 hierarchy, net_cls sees identity migration from root to root and cgroup core used to call ->css_attach callback for those. The net_cls ->css_attach ends up calling update_classid() on the root net_cls css to which all processes on the system belong to as the controller isn't used. This makes any cgroup v2 migration O(total_number_of_fds_on_the_system) which is horrible and easily leads to noticeable stalls triggering RCU stall warnings and so on. The worst symptom is already fixed in upstream by bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"); however, backporting that commit is too invasive and we want to avoid other cases too. This patch updates net_cls's cgrp_attach() to iterate fds of only the processes which are actually getting migrated. This removes the surprising migration cost which is dependent on the total number of fds in the target cgroup. As this leaves write_classid() the only user of update_classid(), open-code the helper into write_classid(). Reported-by: David Goode <dgoode@fb.com> Fixes: 3b13758f51de ("cgroups: Allow dynamically changing net_classid") Cc: stable@vger.kernel.org # v4.4+ Cc: Nina Schiff <ninasc@fb.com> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-15 06:25:56 +07:00
iterate_fd(p->files, 0, update_classid_sock,
(void *)(unsigned long)css_cls_state(css)->classid);
task_unlock(p);
}
}
static u64 read_classid(struct cgroup_subsys_state *css, struct cftype *cft)
{
return css_cls_state(css)->classid;
}
static int write_classid(struct cgroup_subsys_state *css, struct cftype *cft,
u64 value)
{
struct cgroup_cls_state *cs = css_cls_state(css);
cgroup, net_cls: iterate the fds of only the tasks which are being migrated The net_cls controller controls the classid field of each socket which is associated with the cgroup. Because the classid is per-socket attribute, when a task migrates to another cgroup or the configured classid of the cgroup changes, the controller needs to walk all sockets and update the classid value, which was implemented by 3b13758f51de ("cgroups: Allow dynamically changing net_classid"). While the approach is not scalable, migrating tasks which have a lot of fds attached to them is rare and the cost is born by the ones initiating the operations. However, for simplicity, both the migration and classid config change paths call update_classid() which scans all fds of all tasks in the target css. This is an overkill for the migration path which only needs to cover a much smaller subset of tasks which are actually getting migrated in. On cgroup v1, this can lead to unexpected scalability issues when one tries to migrate a task or process into a net_cls cgroup which already contains a lot of fds. Even if the migration traget doesn't have many to get scanned, update_classid() ends up scanning all fds in the target cgroup which can be extremely numerous. Unfortunately, on cgroup v2 which doesn't use net_cls, the problem is even worse. Before bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"), cgroup core would call the ->css_attach callback even for controllers which don't see actual migration to a different css. As net_cls is always disabled but still mounted on cgroup v2, whenever a process is migrated on the cgroup v2 hierarchy, net_cls sees identity migration from root to root and cgroup core used to call ->css_attach callback for those. The net_cls ->css_attach ends up calling update_classid() on the root net_cls css to which all processes on the system belong to as the controller isn't used. This makes any cgroup v2 migration O(total_number_of_fds_on_the_system) which is horrible and easily leads to noticeable stalls triggering RCU stall warnings and so on. The worst symptom is already fixed in upstream by bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"); however, backporting that commit is too invasive and we want to avoid other cases too. This patch updates net_cls's cgrp_attach() to iterate fds of only the processes which are actually getting migrated. This removes the surprising migration cost which is dependent on the total number of fds in the target cgroup. As this leaves write_classid() the only user of update_classid(), open-code the helper into write_classid(). Reported-by: David Goode <dgoode@fb.com> Fixes: 3b13758f51de ("cgroups: Allow dynamically changing net_classid") Cc: stable@vger.kernel.org # v4.4+ Cc: Nina Schiff <ninasc@fb.com> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-15 06:25:56 +07:00
struct css_task_iter it;
struct task_struct *p;
sock, cgroup: add sock->sk_cgroup In cgroup v1, dealing with cgroup membership was difficult because the number of membership associations was unbound. As a result, cgroup v1 grew several controllers whose primary purpose is either tagging membership or pull in configuration knobs from other subsystems so that cgroup membership test can be avoided. net_cls and net_prio controllers are examples of the latter. They allow configuring network-specific attributes from cgroup side so that network subsystem can avoid testing cgroup membership; unfortunately, these are not only cumbersome but also problematic. Both net_cls and net_prio aren't properly hierarchical. Both inherit configuration from the parent on creation but there's no interaction afterwards. An ancestor doesn't restrict the behavior in its subtree in anyway and configuration changes aren't propagated downwards. Especially when combined with cgroup delegation, this is problematic because delegatees can mess up whatever network configuration implemented at the system level. net_prio would allow the delegatees to set whatever priority value regardless of CAP_NET_ADMIN and net_cls the same for classid. While it is possible to solve these issues from controller side by implementing hierarchical allowable ranges in both controllers, it would involve quite a bit of complexity in the controllers and further obfuscate network configuration as it becomes even more difficult to tell what's actually being configured looking from the network side. While not much can be done for v1 at this point, as membership handling is sane on cgroup v2, it'd be better to make cgroup matching behave like other network matches and classifiers than introducing further complications. In preparation, this patch updates sock->sk_cgrp_data handling so that it points to the v2 cgroup that sock was created in until either net_prio or net_cls is used. Once either of the two is used, sock->sk_cgrp_data reverts to its previous role of carrying prioidx and classid. This is to avoid adding yet another cgroup related field to struct sock. As the mode switching can happen at most once per boot, the switching mechanism is aimed at lowering hot path overhead. It may leak a finite, likely small, number of cgroup refs and report spurious prioidx or classid on switching; however, dynamic updates of prioidx and classid have always been racy and lossy - socks between creation and fd installation are never updated, config changes don't update existing sockets at all, and prioidx may index with dead and recycled cgroup IDs. Non-critical inaccuracies from small race windows won't make any noticeable difference. This patch doesn't make use of the pointer yet. The following patch will implement netfilter match for cgroup2 membership. v2: Use sock_cgroup_data to avoid inflating struct sock w/ another cgroup specific field. v3: Add comments explaining why sock_data_prioidx() and sock_data_classid() use different fallback values. Signed-off-by: Tejun Heo <tj@kernel.org> Cc: Daniel Borkmann <daniel@iogearbox.net> Cc: Daniel Wagner <daniel.wagner@bmw-carit.de> CC: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2015-12-08 05:38:53 +07:00
cgroup_sk_alloc_disable();
cs->classid = (u32)value;
css_task_iter_start(css, 0, &it);
cgroup, net_cls: iterate the fds of only the tasks which are being migrated The net_cls controller controls the classid field of each socket which is associated with the cgroup. Because the classid is per-socket attribute, when a task migrates to another cgroup or the configured classid of the cgroup changes, the controller needs to walk all sockets and update the classid value, which was implemented by 3b13758f51de ("cgroups: Allow dynamically changing net_classid"). While the approach is not scalable, migrating tasks which have a lot of fds attached to them is rare and the cost is born by the ones initiating the operations. However, for simplicity, both the migration and classid config change paths call update_classid() which scans all fds of all tasks in the target css. This is an overkill for the migration path which only needs to cover a much smaller subset of tasks which are actually getting migrated in. On cgroup v1, this can lead to unexpected scalability issues when one tries to migrate a task or process into a net_cls cgroup which already contains a lot of fds. Even if the migration traget doesn't have many to get scanned, update_classid() ends up scanning all fds in the target cgroup which can be extremely numerous. Unfortunately, on cgroup v2 which doesn't use net_cls, the problem is even worse. Before bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"), cgroup core would call the ->css_attach callback even for controllers which don't see actual migration to a different css. As net_cls is always disabled but still mounted on cgroup v2, whenever a process is migrated on the cgroup v2 hierarchy, net_cls sees identity migration from root to root and cgroup core used to call ->css_attach callback for those. The net_cls ->css_attach ends up calling update_classid() on the root net_cls css to which all processes on the system belong to as the controller isn't used. This makes any cgroup v2 migration O(total_number_of_fds_on_the_system) which is horrible and easily leads to noticeable stalls triggering RCU stall warnings and so on. The worst symptom is already fixed in upstream by bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"); however, backporting that commit is too invasive and we want to avoid other cases too. This patch updates net_cls's cgrp_attach() to iterate fds of only the processes which are actually getting migrated. This removes the surprising migration cost which is dependent on the total number of fds in the target cgroup. As this leaves write_classid() the only user of update_classid(), open-code the helper into write_classid(). Reported-by: David Goode <dgoode@fb.com> Fixes: 3b13758f51de ("cgroups: Allow dynamically changing net_classid") Cc: stable@vger.kernel.org # v4.4+ Cc: Nina Schiff <ninasc@fb.com> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-15 06:25:56 +07:00
while ((p = css_task_iter_next(&it))) {
task_lock(p);
iterate_fd(p->files, 0, update_classid_sock,
(void *)(unsigned long)cs->classid);
task_unlock(p);
cgroup, netclassid: add a preemption point to write_classid We have seen a customer complaining about soft lockups on !PREEMPT kernel config with 4.4 based kernel [1072141.435366] NMI watchdog: BUG: soft lockup - CPU#21 stuck for 22s! [systemd:1] [1072141.444090] Modules linked in: mpt3sas raid_class binfmt_misc af_packet 8021q garp mrp stp llc xfs libcrc32c bonding iscsi_ibft iscsi_boot_sysfs msr ext4 crc16 jbd2 mbcache cdc_ether usbnet mii joydev hid_generic usbhid intel_rapl x86_pkg_temp_thermal intel_powerclamp coretemp crct10dif_pclmul crc32_pclmul ghash_clmulni_intel ipmi_ssif mgag200 i2c_algo_bit ttm ipmi_devintf drbg ixgbe drm_kms_helper vxlan ansi_cprng ip6_udp_tunnel drm aesni_intel udp_tunnel aes_x86_64 iTCO_wdt syscopyarea ptp xhci_pci lrw iTCO_vendor_support pps_core gf128mul ehci_pci glue_helper sysfillrect mdio pcspkr sb_edac ablk_helper cryptd ehci_hcd sysimgblt xhci_hcd fb_sys_fops edac_core mei_me lpc_ich ses usbcore enclosure dca mfd_core ipmi_si mei i2c_i801 scsi_transport_sas usb_common ipmi_msghandler shpchp fjes wmi processor button acpi_pad btrfs xor raid6_pq sd_mod crc32c_intel megaraid_sas sg dm_multipath dm_mod scsi_dh_rdac scsi_dh_emc scsi_dh_alua scsi_mod md_mod autofs4 [1072141.444146] Supported: Yes [1072141.444149] CPU: 21 PID: 1 Comm: systemd Not tainted 4.4.121-92.80-default #1 [1072141.444150] Hardware name: LENOVO Lenovo System x3650 M5 -[5462P4U]- -[5462P4U]-/01GR451, BIOS -[TCE136H-2.70]- 06/13/2018 [1072141.444151] task: ffff880191bd0040 ti: ffff880191bd4000 task.ti: ffff880191bd4000 [1072141.444153] RIP: 0010:[<ffffffff815229f9>] [<ffffffff815229f9>] update_classid_sock+0x29/0x40 [1072141.444157] RSP: 0018:ffff880191bd7d58 EFLAGS: 00000286 [1072141.444158] RAX: ffff883b177cb7c0 RBX: 0000000000000000 RCX: 0000000000000000 [1072141.444159] RDX: 00000000000009c7 RSI: ffff880191bd7d5c RDI: ffff8822e29bb200 [1072141.444160] RBP: ffff883a72230980 R08: 0000000000000101 R09: 0000000000000000 [1072141.444161] R10: 0000000000000008 R11: f000000000000000 R12: ffffffff815229d0 [1072141.444162] R13: 0000000000000000 R14: ffff881fd0a47ac0 R15: ffff880191bd7f28 [1072141.444163] FS: 00007f3e2f1eb8c0(0000) GS:ffff882000340000(0000) knlGS:0000000000000000 [1072141.444164] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [1072141.444165] CR2: 00007f3e2f200000 CR3: 0000001ffea4e000 CR4: 00000000001606f0 [1072141.444166] Stack: [1072141.444166] ffffffa800000246 00000000000009c7 ffffffff8121d583 ffff8818312a05c0 [1072141.444168] ffff8818312a1100 ffff880197c3b280 ffff881861422858 ffffffffffffffea [1072141.444170] ffffffff81522b1c ffffffff81d0ca20 ffff8817fa17b950 ffff883fdd8121e0 [1072141.444171] Call Trace: [1072141.444179] [<ffffffff8121d583>] iterate_fd+0x53/0x80 [1072141.444182] [<ffffffff81522b1c>] write_classid+0x4c/0x80 [1072141.444187] [<ffffffff8111328b>] cgroup_file_write+0x9b/0x100 [1072141.444193] [<ffffffff81278bcb>] kernfs_fop_write+0x11b/0x150 [1072141.444198] [<ffffffff81201566>] __vfs_write+0x26/0x100 [1072141.444201] [<ffffffff81201bed>] vfs_write+0x9d/0x190 [1072141.444203] [<ffffffff812028c2>] SyS_write+0x42/0xa0 [1072141.444207] [<ffffffff815f58c3>] entry_SYSCALL_64_fastpath+0x1e/0xca [1072141.445490] DWARF2 unwinder stuck at entry_SYSCALL_64_fastpath+0x1e/0xca If a cgroup has many tasks with many open file descriptors then we would end up in a large loop without any rescheduling point throught the operation. Add cond_resched once per task. Signed-off-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Tejun Heo <tj@kernel.org>
2018-10-18 15:56:17 +07:00
cond_resched();
cgroup, net_cls: iterate the fds of only the tasks which are being migrated The net_cls controller controls the classid field of each socket which is associated with the cgroup. Because the classid is per-socket attribute, when a task migrates to another cgroup or the configured classid of the cgroup changes, the controller needs to walk all sockets and update the classid value, which was implemented by 3b13758f51de ("cgroups: Allow dynamically changing net_classid"). While the approach is not scalable, migrating tasks which have a lot of fds attached to them is rare and the cost is born by the ones initiating the operations. However, for simplicity, both the migration and classid config change paths call update_classid() which scans all fds of all tasks in the target css. This is an overkill for the migration path which only needs to cover a much smaller subset of tasks which are actually getting migrated in. On cgroup v1, this can lead to unexpected scalability issues when one tries to migrate a task or process into a net_cls cgroup which already contains a lot of fds. Even if the migration traget doesn't have many to get scanned, update_classid() ends up scanning all fds in the target cgroup which can be extremely numerous. Unfortunately, on cgroup v2 which doesn't use net_cls, the problem is even worse. Before bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"), cgroup core would call the ->css_attach callback even for controllers which don't see actual migration to a different css. As net_cls is always disabled but still mounted on cgroup v2, whenever a process is migrated on the cgroup v2 hierarchy, net_cls sees identity migration from root to root and cgroup core used to call ->css_attach callback for those. The net_cls ->css_attach ends up calling update_classid() on the root net_cls css to which all processes on the system belong to as the controller isn't used. This makes any cgroup v2 migration O(total_number_of_fds_on_the_system) which is horrible and easily leads to noticeable stalls triggering RCU stall warnings and so on. The worst symptom is already fixed in upstream by bfc2cf6f61fc ("cgroup: call subsys->*attach() only for subsystems which are actually affected by migration"); however, backporting that commit is too invasive and we want to avoid other cases too. This patch updates net_cls's cgrp_attach() to iterate fds of only the processes which are actually getting migrated. This removes the surprising migration cost which is dependent on the total number of fds in the target cgroup. As this leaves write_classid() the only user of update_classid(), open-code the helper into write_classid(). Reported-by: David Goode <dgoode@fb.com> Fixes: 3b13758f51de ("cgroups: Allow dynamically changing net_classid") Cc: stable@vger.kernel.org # v4.4+ Cc: Nina Schiff <ninasc@fb.com> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Tejun Heo <tj@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-15 06:25:56 +07:00
}
css_task_iter_end(&it);
return 0;
}
static struct cftype ss_files[] = {
{
.name = "classid",
.read_u64 = read_classid,
.write_u64 = write_classid,
},
{ } /* terminate */
};
cgroup: clean up cgroup_subsys names and initialization cgroup_subsys is a bit messier than it needs to be. * The name of a subsys can be different from its internal identifier defined in cgroup_subsys.h. Most subsystems use the matching name but three - cpu, memory and perf_event - use different ones. * cgroup_subsys_id enums are postfixed with _subsys_id and each cgroup_subsys is postfixed with _subsys. cgroup.h is widely included throughout various subsystems, it doesn't and shouldn't have claim on such generic names which don't have any qualifier indicating that they belong to cgroup. * cgroup_subsys->subsys_id should always equal the matching cgroup_subsys_id enum; however, we require each controller to initialize it and then BUG if they don't match, which is a bit silly. This patch cleans up cgroup_subsys names and initialization by doing the followings. * cgroup_subsys_id enums are now postfixed with _cgrp_id, and each cgroup_subsys with _cgrp_subsys. * With the above, renaming subsys identifiers to match the userland visible names doesn't cause any naming conflicts. All non-matching identifiers are renamed to match the official names. cpu_cgroup -> cpu mem_cgroup -> memory perf -> perf_event * controllers no longer need to initialize ->subsys_id and ->name. They're generated in cgroup core and set automatically during boot. * Redundant cgroup_subsys declarations removed. * While updating BUG_ON()s in cgroup_init_early(), convert them to WARN()s. BUGging that early during boot is stupid - the kernel can't print anything, even through serial console and the trap handler doesn't even link stack frame properly for back-tracing. This patch doesn't introduce any behavior changes. v2: Rebased on top of fe1217c4f3f7 ("net: net_cls: move cgroupfs classid handling into core"). Signed-off-by: Tejun Heo <tj@kernel.org> Acked-by: Neil Horman <nhorman@tuxdriver.com> Acked-by: "David S. Miller" <davem@davemloft.net> Acked-by: "Rafael J. Wysocki" <rjw@rjwysocki.net> Acked-by: Michal Hocko <mhocko@suse.cz> Acked-by: Peter Zijlstra <peterz@infradead.org> Acked-by: Aristeu Rozanski <aris@redhat.com> Acked-by: Ingo Molnar <mingo@redhat.com> Acked-by: Li Zefan <lizefan@huawei.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Balbir Singh <bsingharora@gmail.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Serge E. Hallyn <serue@us.ibm.com> Cc: Vivek Goyal <vgoyal@redhat.com> Cc: Thomas Graf <tgraf@suug.ch>
2014-02-08 22:36:58 +07:00
struct cgroup_subsys net_cls_cgrp_subsys = {
.css_alloc = cgrp_css_alloc,
.css_online = cgrp_css_online,
.css_free = cgrp_css_free,
.attach = cgrp_attach,
.legacy_cftypes = ss_files,
};