2013-12-30 01:27:10 +08:00
|
|
|
/*
|
|
|
|
* 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>
|
2017-02-06 17:57:33 +08:00
|
|
|
#include <linux/sched/task.h>
|
|
|
|
|
2013-12-30 01:27:10 +08:00
|
|
|
#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)
|
|
|
|
{
|
2015-07-22 17:23:20 +08:00
|
|
|
return css_cls_state(task_css_check(p, net_cls_cgrp_id,
|
|
|
|
rcu_read_lock_bh_held()));
|
2013-12-30 01:27:10 +08:00
|
|
|
}
|
|
|
|
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);
|
2014-05-17 01:22:48 +08:00
|
|
|
struct cgroup_cls_state *parent = css_cls_state(css->parent);
|
2013-12-30 01:27:10 +08:00
|
|
|
|
|
|
|
if (parent)
|
|
|
|
cs->classid = parent->classid;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void cgrp_css_free(struct cgroup_subsys_state *css)
|
|
|
|
{
|
|
|
|
kfree(css_cls_state(css));
|
|
|
|
}
|
|
|
|
|
2015-11-21 04:31:39 +08:00
|
|
|
static int update_classid_sock(const void *v, struct file *file, unsigned n)
|
2013-12-30 01:27:10 +08:00
|
|
|
{
|
|
|
|
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 06:38:53 +08:00
|
|
|
if (sock) {
|
|
|
|
spin_lock(&cgroup_sk_update_lock);
|
2015-12-08 06:38:52 +08:00
|
|
|
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 06:38:53 +08:00
|
|
|
spin_unlock(&cgroup_sk_update_lock);
|
|
|
|
}
|
2013-12-30 01:27:10 +08:00
|
|
|
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 07:25:56 +08:00
|
|
|
static void cgrp_attach(struct cgroup_taskset *tset)
|
2013-12-30 01:27:10 +08:00
|
|
|
{
|
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 07:25:56 +08:00
|
|
|
struct cgroup_subsys_state *css;
|
2013-12-30 01:27:10 +08:00
|
|
|
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 07:25:56 +08:00
|
|
|
cgroup_taskset_for_each(p, css, tset) {
|
2013-12-30 01:27:10 +08:00
|
|
|
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 07:25:56 +08:00
|
|
|
iterate_fd(p->files, 0, update_classid_sock,
|
|
|
|
(void *)(unsigned long)css_cls_state(css)->classid);
|
2013-12-30 01:27:10 +08:00
|
|
|
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)
|
|
|
|
{
|
2015-11-21 04:31:39 +08:00
|
|
|
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 07:25:56 +08:00
|
|
|
struct css_task_iter it;
|
|
|
|
struct task_struct *p;
|
2015-11-21 04:31:39 +08:00
|
|
|
|
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 06:38:53 +08:00
|
|
|
cgroup_sk_alloc_disable();
|
|
|
|
|
2015-11-21 04:31:39 +08:00
|
|
|
cs->classid = (u32)value;
|
2013-12-30 01:27:10 +08:00
|
|
|
|
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 07:25:56 +08:00
|
|
|
css_task_iter_start(css, &it);
|
|
|
|
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);
|
|
|
|
}
|
|
|
|
css_task_iter_end(&it);
|
|
|
|
|
2013-12-30 01:27:10 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct cftype ss_files[] = {
|
|
|
|
{
|
|
|
|
.name = "classid",
|
|
|
|
.read_u64 = read_classid,
|
|
|
|
.write_u64 = write_classid,
|
|
|
|
},
|
|
|
|
{ } /* terminate */
|
|
|
|
};
|
|
|
|
|
2014-02-08 23:36:58 +08:00
|
|
|
struct cgroup_subsys net_cls_cgrp_subsys = {
|
2013-12-30 01:27:10 +08:00
|
|
|
.css_alloc = cgrp_css_alloc,
|
|
|
|
.css_online = cgrp_css_online,
|
|
|
|
.css_free = cgrp_css_free,
|
|
|
|
.attach = cgrp_attach,
|
2014-07-15 23:05:09 +08:00
|
|
|
.legacy_cftypes = ss_files,
|
2013-12-30 01:27:10 +08:00
|
|
|
};
|