In cgroup_destroy_locked(), move setting of CGRP_DEAD above
invocations of kill_css(). This doesn't make any visible behavior
difference now but will be used to inhibit manipulating controller
enable states of a dying cgroup on the unified hierarchy.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_transfer_tasks() can currently fail in the middle due to memory
allocation failure. When that happens, the function just aborts and
returns error code and there's no way to tell how many actually got
migrated at the point of failure and or to revert the partial
migration.
Update it to use cgroup_migrate{_add_src|prepare_dst|migrate|finish}()
so that the function either succeeds or fails as a whole as long as
->can_attach() doesn't fail.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
For optimization, task_lock() is additionally used to protect
task->cgroups. The optimization is pretty dubious as either
css_set_rwsem is grabbed anyway or PF_EXITING already protects
task->cgroups. It adds only overhead and confusion at this point.
Let's drop task_[un]lock() and update comments accordingly.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
When a new process is forked, cgroup_fork() associates it with the
css_set of its parent but doesn't link it into it. After the new
process is linked to tasklist, cgroup_post_fork() does the linking.
This is problematic for cgroup_transfer_tasks() as there's no way to
tell whether there are tasks which are pointing to a css_set but not
linked yet. It is impossible to implement an operation which transfer
all tasks of a cgroup to another and the current
cgroup_transfer_tasks() can easily be tricked into leaving a newly
forked process behind if it gets called between cgroup_fork() and
cgroup_post_fork().
Let's make association with a css_set and linking atomic by moving it
to cgroup_post_fork(). cgroup_fork() sets child->cgroups to
init_css_set as a placeholder and cgroup_post_fork() is updated to
perform both the association with the parent's cgroup and linking
there. This means that a newly created task will point to
init_css_set without holding a ref to it much like what it does on the
exit path. Empty cg_list is used to indicate that the task isn't
holding a ref to the associated css_set.
This fixes an actual bug with cgroup_transfer_tasks(); however, I'm
not marking it for -stable. The whole thing is broken in multiple
other ways which require invasive updates to fix and I don't think
it's worthwhile to bother with backporting this particular one.
Fortunately, the only user is cpuset and these bugs don't crash the
machine.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, process / task migration is a single operation which may
fail depending on memory pressure or the involved controllers'
->can_attach() callbacks. One problem with this approach is migration
of multiple targets. It's impossible to tell whether a given target
will be successfully migrated beforehand and cgroup core can't keep
track of enough states to roll back after intermediate failure.
This is already an issue with cgroup_transfer_tasks(). Also, we're
gonna need multiple target migration for unified hierarchy.
This patch splits migration into four stages -
cgroup_migrate_add_src(), cgroup_migrate_prepare_dst(),
cgroup_migrate() and cgroup_migrate_finish(), where
cgroup_migrate_prepare_dst() performs all the operations which may
fail due to allocation failure without actually migrating the target.
The four separate stages mean that, disregarding ->can_attach()
failures, the success or failure of multi target migration can be
determined before performing any actual migration. If preparations of
all targets succeed, the whole thing will succeed. If not, the whole
operation can fail without any side-effect.
Since the previous patch to use css_set->mg_tasks to keep track of
migration targets, the only thing which may need memory allocation
during migration is the target css_sets. cgroup_migrate_prepare()
pins all source and target css_sets and link them up. Note that this
can be performed without holding threadgroup_lock even if the target
is a process. As long as cgroup_mutex is held, no new css_set can be
put into play.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, while migrating tasks from one cgroup to another,
cgroup_attach_task() builds a flex array of all target tasks;
unfortunately, this has a couple issues.
* Flex array has size limit. On 64bit, struct task_and_cgroup is
24bytes making the flex element limit around 87k. It is a high
number but not impossible to hit. This means that the current
cgroup implementation can't migrate a process with more than 87k
threads.
* Process migration involves memory allocation whose size is dependent
on the number of threads the process has. This means that cgroup
core can't guarantee success or failure of multi-process migrations
as memory allocation failure can happen in the middle. This is in
part because cgroup can't grab threadgroup locks of multiple
processes at the same time, so when there are multiple processes to
migrate, it is imposible to tell how many tasks are to be migrated
beforehand.
Note that this already affects cgroup_transfer_tasks(). cgroup
currently cannot guarantee atomic success or failure of the
operation. It may fail in the middle and after such failure cgroup
doesn't have enough information to roll back properly. It just
aborts with some tasks migrated and others not.
To resolve the situation, this patch updates the migration path to use
task->cg_list to track target tasks. The previous patch already added
css_set->mg_tasks and updated iterations in non-migration paths to
include them during task migration. This patch updates migration path
to actually make use of it.
Instead of putting onto a flex_array, each target task is moved from
its css_set->tasks list to css_set->mg_tasks and the migration path
keeps trace of all the source css_sets and the associated cgroups.
Once all source css_sets are determined, the destination css_set for
each is determined, linked to the matching source css_set and put on a
separate list.
To iterate the target tasks, migration path just needs to iterat
through either the source or target css_sets, depending on whether
migration has been committed or not, and the tasks on their ->mg_tasks
lists. cgroup_taskset is updated to contain the list_heads for source
and target css_sets and the iteration cursor. cgroup_taskset_*() are
accordingly updated to walk through css_sets and their ->mg_tasks.
This resolves the above listed issues with moderate additional
complexity.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, while migrating tasks from one cgroup to another,
cgroup_attach_task() builds a flex array of all target tasks;
unfortunately, this has a couple issues.
* Flex array has size limit. On 64bit, struct task_and_cgroup is
24bytes making the flex element limit around 87k. It is a high
number but not impossible to hit. This means that the current
cgroup implementation can't migrate a process with more than 87k
threads.
* Process migration involves memory allocation whose size is dependent
on the number of threads the process has. This means that cgroup
core can't guarantee success or failure of multi-process migrations
as memory allocation failure can happen in the middle. This is in
part because cgroup can't grab threadgroup locks of multiple
processes at the same time, so when there are multiple processes to
migrate, it is imposible to tell how many tasks are to be migrated
beforehand.
Note that this already affects cgroup_transfer_tasks(). cgroup
currently cannot guarantee atomic success or failure of the
operation. It may fail in the middle and after such failure cgroup
doesn't have enough information to roll back properly. It just
aborts with some tasks migrated and others not.
To resolve the situation, we're going to use task->cg_list during
migration too. Instead of building a separate array, target tasks
will be linked into a dedicated migration list_head on the owning
css_set. Tasks on the migration list are treated the same as tasks on
the usual tasks list; however, being on a separate list allows cgroup
migration code path to keep track of the target tasks by simply
keeping the list of css_sets with tasks being migrated, making
unpredictable dynamic allocation unnecessary.
In prepartion of such migration path update, this patch introduces
css_set->mg_tasks list and updates css_set task iterations so that
they walk both css_set->tasks and ->mg_tasks. Note that ->mg_tasks
isn't used yet.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Pull in for-3.14-fixes to receive 532de3fc72 ("cgroup: update
cgroup_enable_task_cg_lists() to grab siglock") which conflicts with
afeb0f9fd4 ("cgroup: relocate cgroup_enable_task_cg_lists()") and
the following cg_lists updates. This is likely to cause further
conflicts down the line too, so let's merge it early.
As cgroup_enable_task_cg_lists() is relocated in for-3.15, this merge
causes conflict in the original position. It's resolved by applying
siglock changes to the updated version in the new location.
Conflicts:
kernel/cgroup.c
Signed-off-by: Tejun Heo <tj@kernel.org>
Currently, there's nothing preventing cgroup_enable_task_cg_lists()
from missing set PF_EXITING and race against cgroup_exit(). Depending
on the timing, cgroup_exit() may finish with the task still linked on
css_set leading to list corruption. Fix it by grabbing siglock in
cgroup_enable_task_cg_lists() so that PF_EXITING is guaranteed to be
visible.
This whole on-demand cg_list optimization is extremely fragile and has
ample possibility to lead to bugs which can cause things like
once-a-year oops during boot. I'm wondering whether the better
approach would be just adding "cgroup_disable=all" handling which
disables the whole cgroup rather than tempting fate with this
on-demand craziness.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: stable@vger.kernel.org
Fengguang reported this bug:
BUG: unable to handle kernel NULL pointer dereference at 0000003c
IP: [<cc90b4ad>] cgroup_cfts_commit+0x27/0x1c1
...
Call Trace:
[<cc9d1129>] ? kmem_cache_alloc_trace+0x33f/0x3b7
[<cc90c6fc>] cgroup_add_cftypes+0x8f/0xca
[<cd78b646>] cgroup_init+0x6a/0x26a
[<cd764d7d>] start_kernel+0x4d7/0x57a
[<cd7642ef>] i386_start_kernel+0x92/0x96
This happens in a corner case. If CGROUP_SCHED=y but CFS_BANDWIDTH=n &&
FAIR_GROUP_SCHED=n && RT_GROUP_SCHED=n, we have:
cpu_files[] = {
{ } /* terminate */
}
When we pass cpu_files to cgroup_apply_cftypes(), as cpu_files[0].ss
is NULL, we'll access NULL pointer.
The bug was introduced by commit de00ffa56e
("cgroup: make cgroup_subsys->base_cftypes use cgroup_add_cftypes()").
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
We should free the memory allocated in parse_cgroupfs_options() before
calling this function again.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
css_set_lock has been converted to css_set_rwsem, and rwsem can't nest
inside rcu_read_lock.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
kernel/cgroup.c:2256:1-3: WARNING: PTR_RET can be used
Use PTR_ERR_OR_ZERO rather than if(IS_ERR(...)) + PTR_ERR
Generated by: coccinelle/api/ptr_ret.cocci
Signed-off-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
With module support gone, a lot of functions no longer need to be
exported. Unexport them.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_attach_task() is planned to go through restructuring. Let's
tidy it up a bit in preparation.
* Update cgroup_attach_task() to receive the target task argument in
@leader instead of @tsk.
* Rename @tsk to @task.
* Rename @retval to @ret.
This is purely cosmetic.
v2: get_nr_threads() was using uninitialized @task instead of @leader.
Fixed. Reported by Dan Carpenter.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
The two functions don't have any users left. Remove them along with
cgroup_taskset->cur_cgrp.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Instead of repeatedly locking and unlocking css_set_rwsem inside
cgroup_task_migrate(), update cgroup_attach_task() to grab it outside
of the loop and update cgroup_task_migrate() to use
put_css_set_locked().
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
put_css_set() is performed in two steps - it first tries to put
without grabbing css_set_rwsem if such put wouldn't make the count
zero. If that fails, it puts after write-locking css_set_rwsem. This
patch separates out the second phase into put_css_set_locked() which
should be called with css_set_rwsem locked.
Also, put_css_set_taskexit() is droped and put_css_set() is made to
take @taskexit. There are only a handful users of these functions.
No point in providing different variants.
put_css_locked() will be used by later changes. This patch doesn't
introduce any functional changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently there are two ways to walk tasks of a cgroup -
css_task_iter_start/next/end() and css_scan_tasks(). The latter
builds on the former but allows blocking while iterating.
Unfortunately, the way css_scan_tasks() is implemented is rather
nasty, it uses a priority heap of pointers to extract some number of
tasks in task creation order and loops over them invoking the callback
and repeats that until it reaches the end. It requires either
preallocated heap or may fail under memory pressure, while unlikely to
be problematic, the complexity is O(N^2), and in general just nasty.
We're gonna convert all css_scan_users() to
css_task_iter_start/next/end() and remove css_scan_users(). As
css_scan_tasks() users may block, let's convert css_set_lock to a
rwsem so that tasks can block during css_task_iter_*() is in progress.
While this does increase the chance of possible deadlock scenarios,
given the current usage, the probability is relatively low, and even
if that happens, the right thing to do is updating the iteration in
the similar way to css iterators so that it can handle blocking.
Most conversions are trivial; however, task_cgroup_path() now expects
to be called with css_set_rwsem locked instead of locking itself.
This is because the function is called with RCU read lock held and
rwsem locking should nest outside RCU read lock.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Reimplement cgroup_transfer_tasks() so that it repeatedly fetches the
first task in the cgroup and then tranfers it. This achieves the same
result without using css_scan_tasks() which is scheduled to be
removed.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_task_count() read-locks css_set_lock and walks all tasks to
count them and then returns the result. The only thing all the users
want is determining whether the cgroup is empty or not. This patch
implements cgroup_has_tasks() which tests whether cgroup->cset_links
is empty, replaces all cgroup_task_count() usages and unexports it.
Note that the test isn't synchronized. This is the same as before.
The test has always been racy.
This will help planned css_set locking update.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Move it above so that prototype isn't necessary. Let's also move the
definition of use_task_css_set_links next to it.
This is purely cosmetic.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Tasks are not linked on their css_sets until cgroup task iteration is
actually used. This is to avoid incurring overhead on the fork and
exit paths for systems which have cgroup compiled in but don't use it.
This lazy binding also affects the task migration path. It has to be
careful so that it doesn't link tasks to css_sets when task_cg_lists
linking is not enabled yet. Unfortunately, this conditional linking
in the migration path interferes with planned migration updates.
This patch moves the lazy binding a bit earlier, to the first cgroup
mount. It's a clear indication that cgroup is being used on the
system and task_cg_lists linking is highly likely to be enabled soon
anyway through "tasks" and "cgroup.procs" files.
This allows cgroup_task_migrate() to always link @tsk->cg_list. Note
that it may still race with cgroup_post_fork() but who wins that race
is inconsequential.
While at it, make use_task_css_set_links a bool, add sanity checks in
cgroup_enable_task_cg_lists() and css_task_iter_start(), and update
the former so that it's guaranteed and assumes to run only once.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Before kernfs conversion, due to the way super_block lookup works,
cgroup roots were created and made visible before being fully
initialized. This in turn required a special flag to mark that the
root hasn't been fully initialized so that the destruction path can
tell fully bound ones from half initialized.
That flag is CGRP_ROOT_SUBSYS_BOUND and no longer necessary after the
kernfs conversion as the lookup and creation of new root are atomic
w.r.t. cgroup_mutex. This patch removes the flag and passes the
requests subsystem mask to cgroup_setup_root() so that it can set the
respective mask bits as subsystems are bound.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Disallow more mount options if sane_behavior. Note that xattr used to
generate warning.
While at it, simplify option check in cgroup_mount() and update
sane_behavior comment in cgroup.h.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
This reverts commit ab3f5faa62.
Explanation from Hugh:
It's because more thorough testing, by others here, found that it
wasn't always solving the problem: so I asked Tejun privately to
hold off from sending it in, until we'd worked out why not.
Most of our testing being on a v3,11-based kernel, it was perfectly
possible that the problem was merely our own e.g. missing Tejun's
8a2b753844 ("workqueue: fix ordered workqueues in NUMA setups").
But that turned out not to be enough to fix it either. Then Filipe
pointed out how percpu_ref_kill_and_confirm() uses call_rcu_sched()
before we ever get to put the offline on to the workqueue: by the
time we get to the workqueue, the ordering has already been lost.
So, thanks for the Acks, but I'm afraid that this ordered workqueue
solution is just not good enough: we should simply forget that patch
and provide a different answer."
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Currently, cgroupfs_root and its ->top_cgroup are separated reference
counted and the latter's is ignored. There's no reason to do this
separately. This patch removes cgroupfs_root->refcnt and destroys
cgroupfs_root when the top_cgroup is released.
* cgroup_put() updated to ignore cgroup_is_dead() test for top
cgroups. cgroup_free_fn() updated to handle root destruction when
releasing a top cgroup.
* As root destruction is now bounced through cgroup destruction, it is
asynchronous. Update cgroup_mount() so that it waits for pending
release which is currently implemented using msleep(). Converting
this to proper wait_queue isn't hard but likely unnecessary.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
root->number_of_cgroups is currently an integer protected with
cgroup_mutex. Except for sanity checks and proc reporting, the only
place it's used is to check whether the root has any child during
remount; however, this is a bit flawed as the counter is not
decremented when the cgroup is unlinked but when it's released,
meaning that there could be an extended period where all cgroups are
removed but remount is still not allowed because some internal objects
are lingering. While not perfect either, it'd be better to use
emptiness test on root->top_cgroup.children.
This patch updates cgroup_remount() to test top_cgroup's children
instead, which makes number_of_cgroups only actual usage statistics
printing in proc implemented in proc_cgroupstats_show(). Let's
shorten its name and make it an atomic_t so that we don't have to
worry about its synchronization. It's purely auxiliary at this point.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup->name handling became quite complicated over time involving
dedicated struct cgroup_name for RCU protection. Now that cgroup is
on kernfs, we can drop all of it and simply use kernfs_name/path() and
friends. Replace cgroup->name and all related code with kernfs
name/path constructs.
* Reimplement cgroup_name() and cgroup_path() as thin wrappers on top
of kernfs counterparts, which involves semantic changes.
pr_cont_cgroup_name() and pr_cont_cgroup_path() added.
* cgroup->name handling dropped from cgroup_rename().
* All users of cgroup_name/path() updated to the new semantics. Users
which were formatting the string just to printk them are converted
to use pr_cont_cgroup_name/path() instead, which simplifies things
quite a bit. As cgroup_name() no longer requires RCU read lock
around it, RCU lockings which were protecting only cgroup_name() are
removed.
v2: Comment above oom_info_lock updated as suggested by Michal.
v3: dummy_top doesn't have a kn associated and
pr_cont_cgroup_name/path() ended up calling the matching kernfs
functions with NULL kn leading to oops. Test for NULL kn and
print "/" if so. This issue was reported by Fengguang Wu.
v4: Rebased on top of 0ab02ca8f8 ("cgroup: protect modifications to
cgroup_idr with cgroup_mutex").
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Fengguang Wu <fengguang.wu@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
cgroup currently releases its kernfs_node when it gets removed. While
not buggy, this makes cgroup->kn access rules complicated than
necessary and leads to things like get/put protection around
kernfs_remove() in cgroup_destroy_locked(). In addition, we want to
use kernfs_name/path() and friends but also want to be able to
determine a cgroup's name between removal and release.
This patch makes cgroup hold onto its kernfs_node until freed so that
cgroup->kn is always accessible.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Dynamic cftype addition and removal using cgroup_add/rm_cftypes()
respectively has been quite hairy due to vfs i_mutex. As i_mutex
nests outside cgroup_mutex, cgroup_mutex has to be released and
regrabbed on each iteration through the hierarchy complicating the
process. Now that i_mutex is no longer in play, it can be simplified.
* Just holding cgroup_tree_mutex is enough. No need to meddle with
cgroup_mutex.
* No reason to play the unlock - relock - check serial_nr dancing.
Everything can be atomically while holding cgroup_tree_mutex.
* cgroup_cfts_prepare() is replaced with direct locking of
cgroup_tree_mutex.
* cgroup_cfts_commit() no longer fiddles with locking. It just
applies the cftypes change to the existing cgroups in the hierarchy.
Renamed to cgroup_cfts_apply().
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cftype_set was added primarily to allow registering the same cftype
array more than once for different subsystems. Nobody uses or needs
such thing and it's already broken because each cftype has ->ss
pointer which is initialized during registration.
Let's add list_head ->node to cftype and use the first cftype entry in
the array to link them instead of allocating separate cftype_set.
While at it, trigger WARN if cft seems previously initialized during
registration.
This simplifies cftype handling a bit.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cftype handling is about to be revamped. Relocate cgroup_rm_cftypes()
above cgroup_add_cftypes() in preparation. This is pure relocation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Mount option "xattr" is no longer necessary as it's enabled by default
on kernfs. Warn if "xattr" is specified with "sane_behavior" so that
the option can be removed in the future.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup filesystem code was derived from the original sysfs
implementation which was heavily intertwined with vfs objects and
locking with the goal of re-using the existing vfs infrastructure.
That experiment turned out rather disastrous and sysfs switched, a
long time ago, to distributed filesystem model where a separate
representation is maintained which is queried by vfs. Unfortunately,
cgroup stuck with the failed experiment all these years and
accumulated even more problems over time.
Locking and object lifetime management being entangled with vfs is
probably the most egregious. vfs is never designed to be misused like
this and cgroup ends up jumping through various convoluted dancing to
make things work. Even then, operations across multiple cgroups can't
be done safely as it'll deadlock with rename locking.
Recently, kernfs is separated out from sysfs so that it can be used by
users other than sysfs. This patch converts cgroup to use kernfs,
which will bring the following benefits.
* Separation from vfs internals. Locking and object lifetime
management is contained in cgroup proper making things a lot
simpler. This removes significant amount of locking convolutions,
hairy object lifetime rules and the restriction on multi-cgroup
operations.
* Can drop a lot of code to implement filesystem interface as most are
provided by kernfs.
* Proper "severing" semantics, which allows controllers to not worry
about lingering file accesses after offline.
While the preceding patches did as much as possible to make the
transition less painful, large part of the conversion has to be one
discrete step making this patch rather large. The rest of the commit
message lists notable changes in different areas.
Overall
-------
* vfs constructs replaced with kernfs ones. cgroup->dentry w/ ->kn,
cgroupfs_root->sb w/ ->kf_root.
* All dentry accessors are removed. Helpers to map from kernfs
constructs are added.
* All vfs plumbing around dentry, inode and bdi removed.
* cgroup_mount() now directly looks for matching root and then
proceeds to create a new one if not found.
Synchronization and object lifetime
-----------------------------------
* vfs inode locking removed. Among other things, this removes the
need for the convolution in cgroup_cfts_commit(). Future patches
will further simplify it.
* vfs refcnting replaced with cgroup internal ones. cgroup->refcnt,
cgroupfs_root->refcnt added. cgroup_put_root() now directly puts
root->refcnt and when it reaches zero proceeds to destroy it thus
merging cgroup_put_root() and the former cgroup_kill_sb().
Simliarly, cgroup_put() now directly schedules cgroup_free_rcu()
when refcnt reaches zero.
* Unlike before, kernfs objects don't hold onto cgroup objects. When
cgroup destroys a kernfs node, all existing operations are drained
and the association is broken immediately. The same for
cgroupfs_roots and mounts.
* All operations which come through kernfs guarantee that the
associated cgroup is and stays valid for the duration of operation;
however, there are two paths which need to find out the associated
cgroup from dentry without going through kernfs -
css_tryget_from_dir() and cgroupstats_build(). For these two,
kernfs_node->priv is RCU managed so that they can dereference it
under RCU read lock.
File and directory handling
---------------------------
* File and directory operations converted to kernfs_ops and
kernfs_syscall_ops.
* xattrs is implicitly supported by kernfs. No need to worry about it
from cgroup. This means that "xattr" mount option is no longer
necessary. A future patch will add a deprecated warning message
when sane_behavior.
* When cftype->max_write_len > PAGE_SIZE, it's necessary to make a
private copy of one of the kernfs_ops to set its atomic_write_len.
cftype->kf_ops is added and cgroup_init/exit_cftypes() are updated
to handle it.
* cftype->lockdep_key added so that kernfs lockdep annotation can be
per cftype.
* Inidividual file entries and open states are now managed by kernfs.
No need to worry about them from cgroup. cfent, cgroup_open_file
and their friends are removed.
* kernfs_nodes are created deactivated and kernfs_activate()
invocations added to places where creation of new nodes are
committed.
* cgroup_rmdir() uses kernfs_[un]break_active_protection() for
self-removal.
v2: - Li pointed out in an earlier patch that specifying "name="
during mount without subsystem specification should succeed if
there's an existing hierarchy with a matching name although it
should fail with -EINVAL if a new hierarchy should be created.
Prior to the conversion, this used by handled by deferring
failure from NULL return from cgroup_root_from_opts(), which was
necessary because root was being created before checking for
existing ones. Note that cgroup_root_from_opts() returned an
ERR_PTR() value for error conditions which require immediate
mount failure.
As we now have separate search and creation steps, deferring
failure from cgroup_root_from_opts() is no longer necessary.
cgroup_root_from_opts() is updated to always return ERR_PTR()
value on failure.
- The logic to match existing roots is updated so that a mount
attempt with a matching name but different subsys_mask are
rejected. This was handled by a separate matching loop under
the comment "Check for name clashes with existing mounts" but
got lost during conversion. Merge the check into the main
search loop.
- Add __rcu __force casting in RCU_INIT_POINTER() in
cgroup_destroy_locked() to avoid the sparse address space
warning reported by kbuild test bot. Maybe we want an explicit
interface to use kn->priv as RCU protected pointer?
v3: Make CONFIG_CGROUPS select CONFIG_KERNFS.
v4: Rebased on top of 0ab02ca8f8 ("cgroup: protect modifications to
cgroup_idr with cgroup_mutex").
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: kbuild test robot fengguang.wu@intel.com>
Relocate cgroup_init/exit_root_id(), cgroup_free_root(),
cgroup_kill_sb() and cgroup_file_name() in preparation of kernfs
conversion.
These are pure relocations to make kernfs conversion easier to follow.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
* Un-inline seq_css(). After kernfs conversion, the function will
need to dereference internal data structures.
* Add cgroup_get/put_root() and replace direct super_block->s_active
manipulatinos with them. These will be converted to kernfs_root
refcnting.
* Add cgroup_get/put() and replace dget/put() on cgrp->dentry with
them. These will be converted to kernfs refcnting.
* Update current_css_set_cg_links_read() to use cgroup_name() instead
of reaching into the dentry name. The end result is the same.
These changes don't make functional differences but will make
transition to kernfs easier.
v2: Rebased on top of 0ab02ca8f8 ("cgroup: protect modifications to
cgroup_idr with cgroup_mutex").
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
mm/memory-failure.c::hwpoison_filter_task() has been reaching into
cgroup to extract the associated ino to be used as a filtering
criterion. This is an implementation detail which shouldn't be
depended upon from outside cgroup proper and is about to change with
the scheduled kernfs conversion.
This patch introduces a proper interface to determine the associated
ino, cgroup_ino(), and updates hwpoison_filter_task() to use it
instead of reaching directly into cgroup.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Factor out cft->ss initialization into cgroup_init_cftypes() from
cgroup_add_cftypes() and add cft->ss clearing to cgroup_rm_cftypes()
through cgroup_exit_cftypes().
This doesn't make any meaningful difference now but the two new
functions will be expanded during kernfs transition.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cftype->max_write_len is used to extend the maximum size of writes.
It's interpreted in such a way that the actual maximum size is one
less than the specified value. The default size is defined by
CGROUP_LOCAL_BUFFER_SIZE. Its interpretation is quite confusing - its
value is decremented by 1 and then compared for equality with max
size, which means that the actual default size is
CGROUP_LOCAL_BUFFER_SIZE - 2, which is 62 chars.
There's no point in having a limit that low. Update its definition so
that it means the actual string length sans termination and anything
below PAGE_SIZE-1 is treated as PAGE_SIZE-1.
.max_write_len for "release_agent" is updated to PATH_MAX-1 and
cgroup_release_agent_write() is updated so that the redundant strlen()
check is removed and it uses strlcpy() instead of strcpy().
.max_write_len initializations in blk-throttle.c and cfq-iosched.c are
no longer necessary and removed. The one in cpuset is kept unchanged
as it's an approximated value to begin with.
This will also make transition to kernfs smoother.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, cgroup_subsys->base_cftypes registration is different from
dynamic cftypes registartion. Instead of going through
cgroup_add_cftypes(), cgroup_init_subsys() invokes
cgroup_init_cftsets() which makes use of cgroup_subsys->base_cftset
which doesn't involve dynamic allocation.
While avoiding dynamic allocation is somewhat nice, having two
separate paths for cftypes registration is nasty, especially as we're
planning to add more operations during cftypes registration.
This patch drops cgroup_init_cftsets() and cgroup_subsys->base_cftset
and registers base_cftypes using cgroup_add_cftypes(). This is done
as a separate step in cgroup_init() instead of a part of
cgroup_init_subsys(). This is because cgroup_init_subsys() can be
called very early during boot when kmalloc() isn't available yet.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Straightforward updates to cgroup name handling in preparation of
kernfs conversion.
* cgroup_alloc_name() is updated to take const char * isntead of
dentry * for name source.
* cgroup name formatting is separated out into cgroup_file_name().
While at it, buffer length protection is added.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Factor out new root initialization into cgroup_setup_root() from
cgroup_mount(). This makes it easier to follow and will ease kernfs
conversion.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup is scheduled to be converted to kernfs. After conversion,
cgroup_mount() won't use the sget() machinery for finding out existing
super_blocks but instead would do that directly. It'll search the
existing cgroupfs_roots for a matching one and create a new one iff a
match doesn't exist. To ease such conversion, this patch restructures
locking and error handling of the function.
cgroup_tree_mutex and cgroup_mutex are grabbed from the get-go and
held until return. For now, due to the way vfs locks nest outside
cgroup mutexes, the two cgroup mutexes are temporarily dropped across
sget() and inode mutex locking, which looks quite ridiculous; however,
these will be removed through kernfs conversion and structuring the
code this way makes the conversion less painful.
The error goto labels are consolidated to two. This looks unwieldy
now but the next patch will factor out creation of new root into a
separate function with accompanying error handling and it'll look a
lot better.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Now that cftypes and all tree modification operations are protected by
cgroup_tree_mutex, we can drop cgroup_mutex while deleting files and
directories. Drop cgroup_mutex over removals.
This doesn't make any noticeable difference now but is to help kernfs
conversion. In kernfs, removals are sync points which drain in-flight
operations as those operations would grab cgroup_mutex, trying to
delete under cgroup_mutex would deadlock. This can be resolved by
just holding the outer cgroup_tree_mutex which nests outside both
kernfs active reference and cgroup_mutex.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently cgroup uses combination of inode->i_mutex'es and
cgroup_mutex for synchronization. With the scheduled kernfs
conversion, i_mutex'es will be removed. Unfortunately, just using
cgroup_mutex isn't possible. All kernfs file and syscall operations,
most of which require grabbing cgroup_mutex, will be called with
kernfs active ref held and, if we try to perform kernfs removals under
cgroup_mutex, it can deadlock as kernfs_remove() tries to drain the
target node.
Let's introduce a new outer mutex, cgroup_tree_mutex, which protects
stuff used during hierarchy changing operations - cftypes and all the
operations which may affect the cgroupfs. It also covers css
association and iteration. This allows cgroup_css(), for_each_css()
and other css iterators to be called under cgroup_tree_mutex. The new
mutex will nest above both kernfs's active ref protection and
cgroup_mutex. By protecting tree modifications with a separate outer
mutex, we can get rid of the forementioned deadlock condition.
Actual file additions and removals now require cgroup_tree_mutex
instead of cgroup_mutex. Currently, cgroup_tree_mutex is never used
without cgroup_mutex; however, we'll soon add hierarchy modification
sections which are only protected by cgroup_tree_mutex. In the
future, we might want to make the locking more granular by better
splitting the coverages of the two mutexes. For now, this should do.
v2: Rebased on top of 0ab02ca8f8 ("cgroup: protect modifications to
cgroup_idr with cgroup_mutex").
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
css_from_dir() returns the matching css (cgroup_subsys_state) given a
dentry and subsystem. The function doesn't pin the css before
returning and requires the caller to be holding RCU read lock or
cgroup_mutex and handling pinning on the caller side.
Given that users of the function are likely to want to pin the
returned css (both existing users do) and that getting and putting
css's are very cheap, there's no reason for the interface to be tricky
like this.
Rename css_from_dir() to css_tryget_from_dir() and make it try to pin
the found css and return it only if pinning succeeded. The callers
are updated so that they no longer do RCU locking and pinning around
the function and just use the returned css.
This will also ease converting cgroup to kernfs.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Balbir Singh <bsingharora@gmail.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Pull for-3.14-fixes to receive 0ab02ca8f8 ("cgroup: protect
modifications to cgroup_idr with cgroup_mutex") prior to kernfs
conversion series to avoid non-trivial conflicts.
Signed-off-by: Tejun Heo <tj@kernel.org>
