After running:
# mount -t cgroup cpu xxx /cgroup && mkdir /cgroup/sub && \
rmdir /cgroup/sub && umount /cgroup
I found the cgroup root still existed:
# cat /proc/cgroups
#subsys_name hierarchy num_cgroups enabled
cpuset 0 1 1
cpu 1 1 1
...
It turned out css_has_online_children() is broken.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Sigend-off-by: Tejun Heo <tj@kernel.org>
Pull cgroup updates from Tejun Heo:
"A lot of activities on cgroup side. Heavy restructuring including
locking simplification took place to improve the code base and enable
implementation of the unified hierarchy, which currently exists behind
a __DEVEL__ mount option. The core support is mostly complete but
individual controllers need further work. To explain the design and
rationales of the the unified hierarchy
Documentation/cgroups/unified-hierarchy.txt
is added.
Another notable change is css (cgroup_subsys_state - what each
controller uses to identify and interact with a cgroup) iteration
update. This is part of continuing updates on css object lifetime and
visibility. cgroup started with reference count draining on removal
way back and is now reaching a point where csses behave and are
iterated like normal refcnted objects albeit with some complexities to
allow distinguishing the state where they're being deleted. The css
iteration update isn't taken advantage of yet but is planned to be
used to simplify memcg significantly"
* 'for-3.16' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup: (77 commits)
cgroup: disallow disabled controllers on the default hierarchy
cgroup: don't destroy the default root
cgroup: disallow debug controller on the default hierarchy
cgroup: clean up MAINTAINERS entries
cgroup: implement css_tryget()
device_cgroup: use css_has_online_children() instead of has_children()
cgroup: convert cgroup_has_live_children() into css_has_online_children()
cgroup: use CSS_ONLINE instead of CGRP_DEAD
cgroup: iterate cgroup_subsys_states directly
cgroup: introduce CSS_RELEASED and reduce css iteration fallback window
cgroup: move cgroup->serial_nr into cgroup_subsys_state
cgroup: link all cgroup_subsys_states in their sibling lists
cgroup: move cgroup->sibling and ->children into cgroup_subsys_state
cgroup: remove cgroup->parent
device_cgroup: remove direct access to cgroup->children
memcg: update memcg_has_children() to use css_next_child()
memcg: remove tasks/children test from mem_cgroup_force_empty()
cgroup: remove css_parent()
cgroup: skip refcnting on normal root csses and cgrp_dfl_root self css
cgroup: use cgroup->self.refcnt for cgroup refcnting
...
After booting with cgroup_disable=memory, I still saw memcg files
in the default hierarchy, and I can write to them, though it won't
take effect.
# dmesg
...
Disabling memory control group subsystem
...
# mount -t cgroup -o __DEVEL__sane_behavior xxx /cgroup
# ls /cgroup
...
memory.failcnt memory.move_charge_at_immigrate
memory.force_empty memory.numa_stat
memory.limit_in_bytes memory.oom_control
...
# cat /cgroup/memory.usage_in_bytes
0
tj: Minor comment update.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
The default root is allocated and initialized at boot phase, so we
shouldn't destroy the default root when it's umounted, otherwise
it will lead to disaster.
Just try mount and then umount the default root, and the kernel will
crash immediately.
v2:
- No need to check for CSS_NO_REF in cgroup_get/put(). (Tejun)
- Better call cgroup_put() for the default root in kill_sb(). (Tejun)
- Add a comment.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
There is still one residue of sysfs remaining: the sb_magic
SYSFS_MAGIC. However this should be kernfs user specific,
so this patch moves it out. Kerrnfs user should specify their
magic number while mouting.
Signed-off-by: Jianyu Zhan <nasa4836@gmail.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The debug controller, as its name suggests, exposes cgroup core
internals to userland to aid debugging. Unfortunately, except for the
name, there's no provision to prevent its usage in production
configurations and the controller is widely enabled and mounted
leaking internal details to userland. Like most other debug
information, the information exposed by debug isn't interesting even
for debugging itself once the related parts are working reliably.
This controller has no reason for existing. This patch implements
cgrp_dfl_root_inhibit_ss_mask which can suppress specific subsystems
on the default hierarchy and adds the debug subsystem to it so that it
can be gradually deprecated as usages move towards the unified
hierarchy.
Signed-off-by: Tejun Heo <tj@kernel.org>
Now that cgroup liveliness and css onliness are the same state,
convert cgroup_has_live_children() into css_has_online_children() so
that it can be used for actual csses too. The function now uses
css_for_each_child() for iteration and is published.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Use CSS_ONLINE on the self css to indicate whether a cgroup has been
killed instead of CGRP_DEAD. This will allow re-using css online test
for cgroup liveliness test. This doesn't introduce any functional
change.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, css_next_child() is implemented as finding the next child
cgroup which has the css enabled, which used to be the only way to do
it as only cgroups participated in sibling lists and thus could be
iteratd. This works as long as what's required during iteration is
not missing online csses; however, it turns out that there are use
cases where offlined but not yet released csses need to be iterated.
This is difficult to implement through cgroup iteration the unified
hierarchy as there may be multiple dying csses for the same subsystem
associated with single cgroup.
After the recent changes, the cgroup self and regular csses behave
identically in how they're linked and unlinked from the sibling lists
including assertion of CSS_RELEASED and css_next_child() can simply
switch to iterating csses directly. This both simplifies the logic
and ensures that all visible non-released csses are included in the
iteration whether there are multiple dying csses for a subsystem or
not.
As all other iterators depend on css_next_child() for sibling
iteration, this changes behaviors of all css iterators. Add and
update explanations on the css states which are included in traversal
to all iterators.
As css iteration could always contain offlined csses, this shouldn't
break any of the current users and new usages which need iteration of
all on and offline csses can make use of the new semantics.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
css iterations allow the caller to drop RCU read lock. As long as the
caller keeps the current position accessible, it can simply re-grab
RCU read lock later and continue iteration. This is achieved by using
CGRP_DEAD to detect whether the current positions next pointer is safe
to dereference and if not re-iterate from the beginning to the next
position using ->serial_nr.
CGRP_DEAD is used as the marker to invalidate the next pointer and the
only requirement is that the marker is set before the next sibling
starts its RCU grace period. Because CGRP_DEAD is set at the end of
cgroup_destroy_locked() but the cgroup is unlinked when the reference
count reaches zero, we currently have a rather large window where this
fallback re-iteration logic can be triggered.
This patch introduces CSS_RELEASED which is set when a css is unlinked
from its sibling list. This still keeps the re-iteration logic
working while drastically reducing the window of its activation.
While at it, rewrite the comment in css_next_child() to reflect the
new flag and better explain the synchronization.
This will also enable iterating csses directly instead of through
cgroups.
v2: CSS_RELEASED now assigned to 1 << 2 as 1 << 0 is used by
CSS_NO_REF.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
We're moving towards using cgroup_subsys_states as the fundamental
structural blocks. All csses including the cgroup->self and actual
ones now form trees through css->children and ->sibling which follow
the same rules as what cgroup->children and ->sibling followed. This
patch moves cgroup->serial_nr which is used to implement css iteration
into css.
Note that all csses, regardless of their types, allocate their serial
numbers from the same monotonically increasing counter. This doesn't
affect the ordering needed by css iteration or cause any other
material behavior changes. This will be used to update css iteration.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, while all csses have ->children and ->sibling, only the
self csses of cgroups make use of them. This patch makes all other
csses to link themselves on the sibling lists too. This will be used
to update css iteration.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
We're moving towards using cgroup_subsys_states as the fundamental
structural blocks. Let's move cgroup->sibling and ->children into
cgroup_subsys_state. This is pure move without functional change and
only cgroup->self's fields are actually used. Other csses will make
use of the fields later.
While at it, update init_and_link_css() so that it zeroes the whole
css before initializing it and remove explicit zeroing of ->flags.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup->parent is redundant as cgroup->self.parent can also be used to
determine the parent cgroup and we're moving towards using
cgroup_subsys_states as the fundamental structural blocks. This patch
introduces cgroup_parent() which follows cgroup->self.parent and
removes cgroup->parent.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup in general is moving towards using cgroup_subsys_state as the
fundamental structural component and css_parent() was introduced to
convert from using cgroup->parent to css->parent. It was quite some
time ago and we're moving forward with making css more prominent.
This patch drops the trivial wrapper css_parent() and let the users
dereference css->parent. While at it, explicitly mark fields of css
which are public and immutable.
v2: New usage from device_cgroup.c converted.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: "David S. Miller" <davem@davemloft.net>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Johannes Weiner <hannes@cmpxchg.org>
9395a45004 ("cgroup: enable refcnting for root csses") enabled
reference counting for root csses (cgroup_subsys_states) so that
cgroup's self csses can be used to manage the lifetime of the
containing cgroups.
Unfortunately, this change was incorrect. During early init,
cgrp_dfl_root self css refcnt is used. percpu_ref can't initialized
during early init and its initialization is deferred till
cgroup_init() time. This means that cpu was using percpu_ref which
wasn't properly initialized. Due to the way percpu variables are laid
out on x86, this didn't blow up immediately on x86 but ended up
incrementing and decrementing the percpu variable at offset zero,
whatever it may be; however, on other archs, this caused fault and
early boot failure.
As cgroup self csses for root cgroups of non-dfl hierarchies need
working refcounting, we can't revert 9395a45004. This patch adds
CSS_NO_REF which explicitly inhibits reference counting on the css and
sets it on all normal (non-self) csses and cgroup_dfl_root self css.
v2: cgrp_dfl_root.self is the offending one. Set the flag on it.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Stephen Warren <swarren@nvidia.com>
Tested-by: Stephen Warren <swarren@nvidia.com>
Fixes: 9395a45004 ("cgroup: enable refcnting for root csses")
Currently cgroup implements refcnting separately using atomic_t
cgroup->refcnt. The destruction paths of cgroup and css are rather
complex and bear a lot of similiarities including the use of RCU and
bouncing to a work item.
This patch makes cgroup use the refcnt of self css for refcnting
instead of using its own. This makes cgroup refcnting use css's
percpu refcnt and share the destruction mechanism.
* css_release_work_fn() and css_free_work_fn() are updated to handle
both csses and cgroups. This is a bit messy but should do until we
can make cgroup->self a full css, which currently can't be done
thanks to multiple hierarchies.
* cgroup_destroy_locked() now performs
percpu_ref_kill(&cgrp->self.refcnt) instead of cgroup_put(cgrp).
* Negative refcnt sanity check in cgroup_get() is no longer necessary
as percpu_ref already handles it.
* Similarly, as a cgroup which hasn't been killed will never be
released regardless of its refcnt value and percpu_ref has sanity
check on kill, cgroup_is_dead() sanity check in cgroup_put() is no
longer necessary.
* As whether a refcnt reached zero or not can only be decided after
the reference count is killed, cgroup_root->cgrp's refcnting can no
longer be used to decide whether to kill the root or not. Let's
make cgroup_kill_sb() explicitly initiate destruction if the root
doesn't have any children. This makes sense anyway as unmounted
cgroup hierarchy without any children should be destroyed.
While this is a bit messy, this will allow pushing more bookkeeping
towards cgroup->self and thus handling cgroups and csses in more
uniform way. In the very long term, it should be possible to
introduce a base subsystem and convert the self css to a proper one
making things whole lot simpler and unified.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, css_get(), css_tryget() and css_tryget_online() are noops
for root csses as an optimization; however, we're planning to use css
refcnts to track of cgroup lifetime too and root cgroups also need to
be reference counted. Since css has been converted to percpu_refcnt,
the overhead of refcnting is miniscule and this optimization isn't too
meaningful anymore. Furthermore, controllers which optimize the root
cgroup often never even invoke these functions in their hot paths.
This patch enables refcnting for root csses too. This makes CSS_ROOT
flag unused and removes it.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
css release is planned to do more and would require process context.
Bounce it through css->destroy_work.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_destroy_css_killed() is cgroup destruction stage which happens
after all csses are offlined. After the recent updates, it no longer
does anything other than putting the base reference. This patch
removes the function and makes cgroup_destroy_locked() put the base
ref at the end isntead.
This also makes cgroup->nr_css unnecessary. Removed.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Move cgroup->sibling unlinking from cgroup_destroy_css_killed() to
cgroup_put(). This is later but still before the RCU grace period, so
it doesn't break css_next_child() although there now is a larger
window in which a dead cgroup is visible during css iteration. As css
iteration always could have included offline csses, this doesn't
affect correctness; however, it does make css_next_child() fall back
to reiterting mode more often. This also makes cgroup_put() directly
take cgroup_mutex, which limits where it can be called from. These
are not immediately problematic and will be dealt with later.
This change enables simplification of cgroup destruction path.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, check_for_release() on the parent of a destroyed cgroup is
invoked from cgroup_destroy_css_killed(). This is because this is
where the destroyed cgroup can be removed from the parent's children
list. check_for_release() tests the emptiness of the list directly,
so invoking it before removing the cgroup from the list makes it think
that the parent still has children even when it no longer does.
This patch updates check_for_release() to use
cgroup_has_live_children() instead of directly testing ->children
emptiness and moves check_for_release(parent) earlier to the end of
cgroup_destroy_locked(). As cgroup_has_live_children() ignores
cgroups marked DEAD, check_for_release() functions correctly as long
as it's called after asserting DEAD.
This makes release notification slightly more timely and more
importantly enables further simplification of cgroup destruction path.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup->dummy_css is used as the placeholder css when performing css
oriended operations on the cgroup. We're gonna shift more cgroup
management to this css. Let's rename it to ->self and move it to the
top.
This is pure rename and field relocation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_mount() uses dumb delay-and-retry logic to wait for cgroup_root
which is being destroyed. The retry currently loops inside
cgroup_mount() proper. This patch makes it return with
restart_syscall() instead so that retry travels out to userland
boundary.
This slightly simplifies the logic and more importantly makes the
retry logic behave better when the wait for some reason becomes
lengthy or infinite by allowing the operation to be suspended or
terminated from userland.
v2: The original patch forgot to free memory allocated for @opts.
Fixed. Caught by Li Zefan.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_tree_mutex was introduced to work around the circular
dependency between cgroup_mutex and kernfs active protection - some
kernfs file and directory operations needed cgroup_mutex putting
cgroup_mutex under active protection but cgroup also needs to be able
to access cgroup hierarchies and cftypes to determine which
kernfs_nodes need to be removed. cgroup_tree_mutex nested above both
cgroup_mutex and kernfs active protection and used to protect the
hierarchy and cftypes. While this worked, it added a lot of double
lockings and was generally cumbersome.
kernfs provides a mechanism to opt out of active protection and cgroup
was already using it for removal and subtree_control. There's no
reason to mix both methods of avoiding circular locking dependency and
the preceding cgroup_kn_lock_live() changes applied it to all relevant
cgroup kernfs operations making it unnecessary to nest cgroup_mutex
under kernfs active protection. The previous patch reversed the
original lock ordering and put cgroup_mutex above kernfs active
protection.
After these changes, all cgroup_tree_mutex usages are now accompanied
by cgroup_mutex making the former completely redundant. This patch
removes cgroup_tree_mutex and all its usages.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
After the recent cgroup_kn_lock_live() changes, cgroup_mutex is no
longer nested below kernfs active protection. The two don't have any
relationship now.
This patch nests kernfs active protection under cgroup_mutex. All
cftype operations now require both cgroup_tree_mutex and cgroup_mutex,
temporary cgroup_mutex releases over kernfs operations are removed,
and cgroup_add/rm_cftypes() grab both mutexes.
This makes cgroup_tree_mutex redundant, which will be removed by the
next patch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Make __cgroup_procs_write() and cgroup_release_agent_write() use
cgroup_kn_lock_live() and cgroup_kn_unlock() instead of
cgroup_lock_live_group(). This puts the operations under both
cgroup_tree_mutex and cgroup_mutex protection without circular
dependency from kernfs active protection. Also, this means that
cgroup_mutex is no longer nested below kernfs active protection.
There is no longer any place where the two locks interact.
This leaves cgroup_lock_live_group() without any user. Removed.
This will help simplifying cgroup locking.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_mkdir(), cgroup_rmdir() and cgroup_subtree_control_write()
share the logic to break active protection so that they can grab
cgroup_tree_mutex which nests above active protection and/or remove
self. Factor out this logic into cgroup_kn_lock_live() and
cgroup_kn_unlock().
This patch doesn't introduce any functional changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
The ->priv field of a cgroup directory kernfs_node points back to the
cgroup. This field is RCU cleared in cgroup_destroy_locked() for
non-kernfs accesses from css_tryget_from_dir() and
cgroupstats_build().
As these are only applicable to cgroups which finished creation
successfully and fully initialized cgroups are always removed by
cgroup_rmdir(), this can be safely moved to the end of cgroup_rmdir().
This will help simplifying cgroup locking and shouldn't introduce any
behavior difference.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Move cgroup_lock_live_group() invocation upwards to right below
cgroup_tree_mutex in cgroup_subtree_control_write(). This is to help
the planned locking simplification.
This doesn't make any userland-visible behavioral changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_mkdir() is the sole user of cgroup_create(). Let's collapse
the latter into the former. This will help simplifying locking.
While at it, remove now stale comment about inode locking.
This patch doesn't introduce any functional changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Reorganize cgroup_create() so that all paths share unlock out path.
* All err_* labels are renamed to out_* as they're now shared by both
success and failure paths.
* @err renamed to @ret for the similar reason as above and so that
it's more consistent with other functions.
* cgroup memory allocation moved after locking so that freeing failed
cgroup happens before unlocking. While this moves more code inside
critical section, memory allocations inside cgroup locking are
already pretty common and this is unlikely to make any noticeable
difference.
* While at it, replace a stray @parent->root dereference with @root.
This reorganization will help simplifying locking.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Now that cgroup_subtree_control_write() has access to the associated
kernfs_open_file and thus the kernfs_node, there's no need to cache it
in cgroup->control_kn on creation. Remove cgroup->control_kn and use
@of->kn directly.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_tasks_write() and cgroup_procs_write() are currently using
cftype->write_u64(). This patch converts them to use cftype->write()
instead. This allows access to the associated kernfs_open_file which
will be necessary to implement the planned kernfs active protection
manipulation for these files.
This shifts buffer parsing to attach_task_by_pid() and makes it return
@nbytes on success. Let's rename it to __cgroup_procs_write() to
clearly indicate that this is a write handler implementation.
This patch doesn't introduce any visible behavior changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cftype->trigger() is pointless. It's trivial to ignore the input
buffer from a regular ->write() operation. Convert all ->trigger()
users to ->write() and remove ->trigger().
This patch doesn't introduce any visible behavior changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Convert all cftype->write_string() users to the new cftype->write()
which maps directly to kernfs write operation and has full access to
kernfs and cgroup contexts. The conversions are mostly mechanical.
* @css and @cft are accessed using of_css() and of_cft() accessors
respectively instead of being specified as arguments.
* Should return @nbytes on success instead of 0.
* @buf is not trimmed automatically. Trim if necessary. Note that
blkcg and netprio don't need this as the parsers already handle
whitespaces.
cftype->write_string() has no user left after the conversions and
removed.
While at it, remove unnecessary local variable @p in
cgroup_subtree_control_write() and stale comment about
CGROUP_LOCAL_BUFFER_SIZE in cgroup_freezer.c.
This patch doesn't introduce any visible behavior changes.
v2: netprio was missing from conversion. Converted.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Aristeu Rozanski <arozansk@redhat.com>
Acked-by: Vivek Goyal <vgoyal@redhat.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Neil Horman <nhorman@tuxdriver.com>
Cc: "David S. Miller" <davem@davemloft.net>
During the recent conversion to kernfs, cftype's seq_file operations
are updated so that they are directly mapped to kernfs operations and
thus can fully access the associated kernfs and cgroup contexts;
however, write path hasn't seen similar updates and none of the
existing write operations has access to, for example, the associated
kernfs_open_file.
Let's introduce a new operation cftype->write() which maps directly to
the kernfs write operation and has access to all the arguments and
contexts. This will replace ->write_string() and ->trigger() and ease
manipulation of kernfs active protection from cgroup file operations.
Two accessors - of_cft() and of_css() - are introduced to enable
accessing the associated cgroup context from cftype->write() which
only takes kernfs_open_file for the context information. The
accessors for seq_file operations - seq_cft() and seq_css() - are
rewritten to wrap the of_ accessors.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Unlike the more usual refcnting, what css_tryget() provides is the
distinction between online and offline csses instead of protection
against upping a refcnt which already reached zero. cgroup is
planning to provide actual tryget which fails if the refcnt already
reached zero. Let's rename the existing trygets so that they clearly
indicate that they're onliness.
I thought about keeping the existing names as-are and introducing new
names for the planned actual tryget; however, given that each
controller participates in the synchronization of the online state, it
seems worthwhile to make it explicit that these functions are about
on/offline state.
Rename css_tryget() to css_tryget_online() and css_tryget_from_dir()
to css_tryget_online_from_dir(). This is pure rename.
v2: cgroup_freezer grew new usages of css_tryget(). Update
accordingly.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Vivek Goyal <vgoyal@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
release_path is now protected by release_agent_path_lock to allow
accessing it without grabbing cgroup_mutex; however,
cgroup_release_agent_show() was still grabbing cgroup_mutex. Let's
convert it to release_agent_path_lock so that we don't have to worry
about this one for the planned locking updates.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
After waiting for a child to finish offline,
cgroup_subtree_control_write() jumps up to retry from after the input
parsing and active protection breaking. This retry makes the
scheduled locking update - removal of cgroup_tree_mutex - more
difficult. Let's simplify it by returning with restart_syscall() for
retries.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
I was confused that strsep() was equivalent to strtok_r() in skipping
over consecutive delimiters. strsep() just splits at the first
occurrence of one of the delimiters which makes the parsing very
inflexible, which makes allowing multiple whitespace chars as
delimters kinda moot. Let's just be consistently strict and require
list of tokens separated by spaces. This is what
Documentation/cgroups/unified-hierarchy.txt describes too.
Also, parsing may access beyond the end of the string if the string
ends with spaces or is zero-length. Make sure it skips zero-length
tokens. Note that this also ensures that the parser doesn't puke on
multiple consecutive spaces.
v2: Add zero-length token skipping.
v3: Added missing space after "==". Spotted by Li.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
c1a71504e9 ("cgroup: don't recycle cgroup id until all csses' have
been destroyed") made cgroup ID persist until a cgroup is released and
add cgroup->subsys[] clearing to css_release() so that css_from_id()
doesn't return a css which has already been released which happens
before cgroup release; however, the right change here was updating
offline_css() to clear cgroup->subsys[] which was done by e329780310
("cgroup: cgroup->subsys[] should be cleared after the css is
offlined") instead of clearing it from css_release().
We're now clearing cgroup->subsys[] twice. This is okay for
traditional hierarchies as a css's lifetime is the same as its
cgroup's; however, this confuses unified hierarchy and turning on and
off a controller repeatedly using "cgroup.subtree_control" can lead to
an oops like the following which happens because cgroup->subsys[] is
incorrectly cleared asynchronously by css_release().
BUG: unable to handle kernel NULL pointer dereference at 00000000000000 08
IP: [<ffffffff81130c11>] kill_css+0x21/0x1c0
PGD 1170d067 PUD f0ab067 PMD 0
Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
Modules linked in:
CPU: 2 PID: 459 Comm: bash Not tainted 3.15.0-rc2-work+ #5
Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
task: ffff880009296710 ti: ffff88000e198000 task.ti: ffff88000e198000
RIP: 0010:[<ffffffff81130c11>] [<ffffffff81130c11>] kill_css+0x21/0x1c0
RSP: 0018:ffff88000e199dc8 EFLAGS: 00010202
RAX: 0000000000000001 RBX: 0000000000000000 RCX: 0000000000000001
RDX: 0000000000000001 RSI: ffffffff8238a968 RDI: ffff880009296f98
RBP: ffff88000e199de0 R08: 0000000000000001 R09: 02b0000000000000
R10: 0000000000000000 R11: ffff880009296fc0 R12: 0000000000000001
R13: ffff88000db6fc58 R14: 0000000000000001 R15: ffff8800139dcc00
FS: 00007ff9160c5740(0000) GS:ffff88001fb00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000008 CR3: 0000000013947000 CR4: 00000000000006e0
Stack:
ffff88000e199de0 ffffffff82389160 0000000000000001 ffff88000e199e80
ffffffff8113537f 0000000000000007 ffff88000e74af00 ffff88000e199e48
ffff880009296710 ffff88000db6fc00 ffffffff8239c100 0000000000000002
Call Trace:
[<ffffffff8113537f>] cgroup_subtree_control_write+0x85f/0xa00
[<ffffffff8112fd18>] cgroup_file_write+0x38/0x1d0
[<ffffffff8126fc97>] kernfs_fop_write+0xe7/0x170
[<ffffffff811f2ae6>] vfs_write+0xb6/0x1c0
[<ffffffff811f35ad>] SyS_write+0x4d/0xc0
[<ffffffff81d0acd2>] system_call_fastpath+0x16/0x1b
Code: 5c 41 5d 41 5e 41 5f 5d c3 90 0f 1f 44 00 00 55 48 89 e5 41 54 53 48 89 fb 48 83 ec 08 8b 05 37 ad 29 01 85 c0 0f 85 df 00 00 00 <48> 8b 43 08 48 8b 3b be 01 00 00 00 8b 48 5c d3 e6 e8 49 ff ff
RIP [<ffffffff81130c11>] kill_css+0x21/0x1c0
RSP <ffff88000e199dc8>
CR2: 0000000000000008
---[ end trace e7aae1f877c4e1b4 ]---
Remove the unnecessary cgroup->subsys[] clearing from css_release().
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_idr_remove() can be invoked from bh leading to lockdep
detecting possible AA deadlock (IN_BH/ON_BH). Make the lock bh-safe.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_subtree_control_write() waits for offline to complete
child-by-child before enabling a controller; however, it has a couple
bugs.
* It doesn't initialize the wait_queue_t. This can lead to infinite
hang on the following schedule() among other things.
* It forgets to pin the child before releasing cgroup_tree_mutex and
performing schedule(). The child may already be gone by the time it
wakes up and invokes finish_wait(). Pin the child being waited on.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Pull to receive e37a06f109 ("cgroup: fix the retry path of
cgroup_mount()") to avoid unnecessary conflicts with planned
cgroup_tree_mutex removal and also to be able to remove the temp fix
added by 36c38fb714 ("blkcg: use trylock on blkcg_pol_mutex in
blkcg_reset_stats()") afterwards.
Signed-off-by: Tejun Heo <tj@kernel.org>
Determining the css of a task usually requires RCU read lock as that's
the only thing which keeps the returned css accessible till its
reference is acquired; however, testing whether a task belongs to the
root can be performed without dereferencing the returned css by
comparing the returned pointer against the root one in init_css_set[]
which never changes.
Implement task_css_is_root() which can be invoked in any context.
This will be used by the scheduled cgroup_freezer change.
v2: cgroup no longer supports modular controllers. No need to export
init_css_set. Pointed out by Li.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Until now, cgroup->id has been used to identify all the associated
csses and css_from_id() takes cgroup ID and returns the matching css
by looking up the cgroup and then dereferencing the css associated
with it; however, now that the lifetimes of cgroup and css are
separate, this is incorrect and breaks on the unified hierarchy when a
controller is disabled and enabled back again before the previous
instance is released.
This patch adds css->id which is a subsystem-unique ID and converts
css_from_id() to look up by the new css->id instead. memcg is the
only user of css_from_id() and also converted to use css->id instead.
For traditional hierarchies, this shouldn't make any functional
difference.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jianyu Zhan <nasa4836@gmail.com>
Acked-by: Li Zefan <lizefan@huawei.com>
init_css() takes the cgroup the new css belongs to as an argument and
initializes the new css's ->cgroup and ->parent pointers but doesn't
acquire the matching reference counts. After the previous patch,
create_css() puts init_css() and reference acquisition right next to
each other. Let's move reference acquistion into init_css() and
rename the function to init_and_link_css(). This makes sense and is
easier to follow. This makes the root csses to hold a reference on
cgrp_dfl_root.cgrp, which is harmless.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, when create_css() fails in the middle, the half-initialized
css is freed by invoking cgroup_subsys->css_free() directly. This
patch updates the function so that it invokes RCU free path instead.
As the RCU free path puts the parent css and owning cgroup, their
references are now acquired right after a new css is successfully
allocated.
This doesn't make any visible difference now but is to enable
implementing css->id and RCU protected lookup by such IDs.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, cgroup_root->cgroup_idr is protected by cgroup_mutex, which
ends up requiring cgroup_put() to be invoked under sleepable context.
This is okay for now but is an unusual requirement and we'll soon add
css->id which will have the same problem but won't be able to simply
grab cgroup_mutex as removal will have to happen from css_release()
which can't sleep.
Introduce cgroup_idr_lock and idr_alloc/replace/remove() wrappers
which protects the idr operations with the lock and use them for
cgroup_root->cgroup_idr. cgroup_put() no longer needs to grab
cgroup_mutex and css_from_id() is updated to always require RCU read
lock instead of either RCU read lock or cgroup_mutex, which doesn't
affect the existing users.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, cgroup->id is allocated from 0, which is always assigned to
the root cgroup; unfortunately, memcg wants to use ID 0 to indicate
invalid IDs and ends up incrementing all IDs by one.
It's reasonable to reserve 0 for special purposes. This patch updates
cgroup core so that ID 0 is not used and the root cgroups get ID 1.
The ID incrementing is removed form memcg.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Li Zefan <lizefan@huawei.com>
There's no reason to use atomic bitops for cgroup_subsys_state->flags,
cgroup_root->flags and various subsys_masks. This patch updates those
to use bitwise and/or operations instead and converts them form
unsigned long to unsigned int.
This makes the fields occupy (marginally) smaller space and makes it
clear that they don't require atomicity.
This patch doesn't cause any behavior difference.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Use pr_fmt and remove embedded prefixes.
Realign modified multi-line statements to open parenthesis.
Convert embedded function name to "%s: ", __func__
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
As suggested by scripts/checkpatch.pl, substitude all pr_warning()
with pr_warn().
No functional change.
Signed-off-by: Jianyu Zhan <nasa4836@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
6612f05b88 ("cgroup: unify pidlist and other file handling")
has removed the only user of cgroup_pidlist_seq_operations :
cgroup_pidlist_open().
This patch removes it.
Signed-off-by: Jianyu Zhan <nasa4836@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
1d5be6b287 ("cgroup: move module ref handling into
rebind_subsystems()") makes parse_cgroupfs_options() no longer takes
refcounts on subsystems.
And unified hierachy makes parse_cgroupfs_options not need to call
with cgroup_mutex held to protect the cgroup_subsys[].
So this patch removes BUG_ON() and the comment. As the comment
doesn't contain useful information afterwards, the whole comment is
removed.
Signed-off-by: Jianyu Zhan <nasa4836@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
cgroup users often need a way to determine when a cgroup's
subhierarchy becomes empty so that it can be cleaned up. cgroup
currently provides release_agent for it; unfortunately, this mechanism
is riddled with issues.
* It delivers events by forking and execing a userland binary
specified as the release_agent. This is a long deprecated method of
notification delivery. It's extremely heavy, slow and cumbersome to
integrate with larger infrastructure.
* There is single monitoring point at the root. There's no way to
delegate management of a subtree.
* The event isn't recursive. It triggers when a cgroup doesn't have
any tasks or child cgroups. Events for internal nodes trigger only
after all children are removed. This again makes it impossible to
delegate management of a subtree.
* Events are filtered from the kernel side. "notify_on_release" file
is used to subscribe to or suppress release event. This is
unnecessarily complicated and probably done this way because event
delivery itself was expensive.
This patch implements interface file "cgroup.populated" which can be
used to monitor whether the cgroup's subhierarchy has tasks in it or
not. Its value is 0 if there is no task in the cgroup and its
descendants; otherwise, 1, and kernfs_notify() notificaiton is
triggers when the value changes, which can be monitored through poll
and [di]notify.
This is a lot ligther and simpler and trivially allows delegating
management of subhierarchy - subhierarchy monitoring can block further
propgation simply by putting itself or another process in the root of
the subhierarchy and monitor events that it's interested in from there
without interfering with monitoring higher in the tree.
v2: Patch description updated as per Serge.
v3: "cgroup.subtree_populated" renamed to "cgroup.populated". The
subtree_ prefix was a bit confusing because
"cgroup.subtree_control" uses it to denote the tree rooted at the
cgroup sans the cgroup itself while the populated state includes
the cgroup itself.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Serge Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Lennart Poettering <lennart@poettering.net>
cgroup is switching away from multiple hierarchies and will use one
unified default hierarchy where controllers can be dynamically enabled
and disabled per subtree. The default hierarchy will serve as the
unified hierarchy to which all controllers are attached and a css on
the default hierarchy would need to also serve the tasks of descendant
cgroups which don't have the controller enabled - ie. the tree may be
collapsed from leaf towards root when viewed from specific
controllers. This has been implemented through effective css in the
previous patches.
This patch finally implements dynamic subtree controller
enable/disable on the default hierarchy via a new knob -
"cgroup.subtree_control" which controls which controllers are enabled
on the child cgroups. Let's assume a hierarchy like the following.
root - A - B - C
\ D
root's "cgroup.subtree_control" determines which controllers are
enabled on A. A's on B. B's on C and D. This coincides with the
fact that controllers on the immediate sub-level are used to
distribute the resources of the parent. In fact, it's natural to
assume that resource control knobs of a child belong to its parent.
Enabling a controller in "cgroup.subtree_control" declares that
distribution of the respective resources of the cgroup will be
controlled. Note that this means that controller enable states are
shared among siblings.
The default hierarchy has an extra restriction - only cgroups which
don't contain any task may have controllers enabled in
"cgroup.subtree_control". Combined with the other properties of the
default hierarchy, this guarantees that, from the view point of
controllers, tasks are only on the leaf cgroups. In other words, only
leaf csses may contain tasks. This rules out situations where child
cgroups compete against internal tasks of the parent, which is a
competition between two different types of entities without any clear
way to determine resource distribution between the two. Different
controllers handle it differently and all the implemented behaviors
are ambiguous, ad-hoc, cumbersome and/or just wrong. Having this
structural constraints imposed from cgroup core removes the burden
from controller implementations and enables showing one consistent
behavior across all controllers.
When a controller is enabled or disabled, css associations for the
controller in the subtrees of each child should be updated. After
enabling, the whole subtree of a child should point to the new css of
the child. After disabling, the whole subtree of a child should point
to the cgroup's css. This is implemented by first updating cgroup
states such that cgroup_e_css() result points to the appropriate css
and then invoking cgroup_update_dfl_csses() which migrates all tasks
in the affected subtrees to the self cgroup on the default hierarchy.
* When read, "cgroup.subtree_control" lists all the currently enabled
controllers on the children of the cgroup.
* White-space separated list of controller names prefixed with either
'+' or '-' can be written to "cgroup.subtree_control". The ones
prefixed with '+' are enabled on the controller and '-' disabled.
* A controller can be enabled iff the parent's
"cgroup.subtree_control" enables it and disabled iff no child's
"cgroup.subtree_control" has it enabled.
* If a cgroup has tasks, no controller can be enabled via
"cgroup.subtree_control". Likewise, if "cgroup.subtree_control" has
some controllers enabled, tasks can't be migrated into the cgroup.
* All controllers which aren't bound on other hierarchies are
automatically associated with the root cgroup of the default
hierarchy. All the controllers which are bound to the default
hierarchy are listed in the read-only file "cgroup.controllers" in
the root directory.
* "cgroup.controllers" in all non-root cgroups is read-only file whose
content is equal to that of "cgroup.subtree_control" of the parent.
This indicates which controllers can be used in the cgroup's
"cgroup.subtree_control".
This is still experimental and there are some holes, one of which is
that ->can_attach() failure during cgroup_update_dfl_csses() may leave
the cgroups in an undefined state. The issues will be addressed by
future patches.
v2: Non-root cgroups now also have "cgroup.controllers".
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Unified hierarchy implementation would require re-migrating tasks onto
the same cgroup on the default hierarchy to reflect updated effective
csses. Update cgroup_migrate_prepare_dst() so that it accepts NULL as
the destination cgrp. When NULL is specified, the destination is
considered to be the cgroup on the default hierarchy associated with
each css_set.
After this change, the identity check in cgroup_migrate_add_src()
isn't sufficient for noop detection as the associated csses may change
without any cgroup association changing. The only way to tell whether
a migration is noop or not is testing whether the source and
destination csets are identical. The noop check in
cgroup_migrate_add_src() is removed and cset identity test is added to
cgroup_migreate_prepare_dst(). If it's detected that source and
destination csets are identical, the cset is removed removed from
@preloaded_csets and all the migration nodes are cleared which makes
cgroup_migrate() ignore the cset.
Also, make the function append the destination css_sets to
@preloaded_list so that destination css_sets always come after source
css_sets.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Because the default root couldn't have any non-root csses attached to
it, rebinding away from it was always allowed; however, the default
hierarchy will soon host the unified hierarchy and have non-root csses
so the rebind restrictions need to be updated accordingly.
Instead of special casing rebinding from the default hierarchy and
then checking whether the source hierarchy has children cgroups, which
implies non-root csses for !dfl hierarchies, simply check whether the
source hierarchy has non-root csses for the subsystem using
css_next_child().
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
To implement the unified hierarchy behavior, we'll need to be able to
determine the associated cgroup on the default hierarchy from css_set.
Let's add css_set->dfl_cgrp so that it can be accessed conveniently
and efficiently.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Now that effective css handling has been added and iterators updated
accordingly, it's safe to allow cgroup creation in the default
hierarchy. Unblock cgroup creation in the default hierarchy.
As the default hierarchy will implement explicit enabling and
disabling of controllers on each cgroup, suppress automatic css
enabling on cgroup creation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
After a css finishes offlining, offline_css() mistakenly performs
RCU_INIT_POINTER(css->cgroup->subsys[ss->id], css) which just sets the
cgroup->subsys[] pointer to the current value. The intention was to
clear it after offline is complete, not reassign the same value.
Update it to assign NULL instead of the current value. This makes
cgroup_css() to return NULL once offline is complete. All the
existing users of the function either can handle NULL return already
or guarantee that the css doesn't get offlined.
While this is a bugfix, as css lifetime is currently tied to the
cgroup it belongs to, this bug doesn't cause any actual problems.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, css_task_iter iterates tasks associated with a css by
visiting each css_set associated with the owning cgroup and walking
tasks of each of them. This works fine for !unified hierarchies as
each cgroup has its own css for each associated subsystem on the
hierarchy; however, on the planned unified hierarchy, a cgroup may not
have csses associated and its tasks would be considered associated
with the matching css of the nearest ancestor which has the subsystem
enabled.
This means that on the default unified hierarchy, just walking all
tasks associated with a cgroup isn't enough to walk all tasks which
are associated with the specified css. If any of its children doesn't
have the matching css enabled, task iteration should also include all
tasks from the subtree. We already added cgroup->e_csets[] to list
all css_sets effectively associated with a given css and walk css_sets
on that list instead to achieve such iteration.
This patch updates css_task_iter iteration such that it walks css_sets
on cgroup->e_csets[] instead of cgroup->cset_links if iteration is
requested on an non-dummy css. Thanks to the previous iteration
update, this change can be achieved with the addition of
css_task_iter->ss and minimal updates to css_advance_task_iter() and
css_task_iter_start().
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
This patch reorganizes css_task_iter so that adding effective css
support is easier.
* s/->cset_link/->cset_pos/ and s/->task/->task_pos/ for consistency
* ->origin_css is used to determine whether the iteration reached the
last css_set. Replace it with explicit ->cset_head so that
css_advance_task_iter() doesn't have to know the termination
condition directly.
* css_task_iter_next() currently assumes that it's walking list of
cgrp_cset_link and reaches into the current cset through the current
link to determine the termination conditions for task walking. As
this won't always be true for effective css walking, add
->tasks_head and ->mg_tasks_head and use them to control task
walking so that css_task_iter_next() doesn't have to know how
css_sets are being walked.
This patch doesn't make any behavior changes. The iteration logic
stays unchanged after the patch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
css_next_child() walks the children of the specified css. It does
this by finding the next cgroup and then returning the requested css.
On the default unified hierarchy, a cgroup may not have a css
associated with it even if the hierarchy has the subsystem enabled.
This patch updates css_next_child() so that it skips children without
the requested css associated.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
On the default unified hierarchy, a cgroup may be associated with
csses of its ancestors, which means that a css of a given cgroup may
be associated with css_sets of descendant cgroups. This means that we
can't walk all tasks associated with a css by iterating the css_sets
associated with the cgroup as there are css_sets which are pointing to
the css but linked on the descendants.
This patch adds per-subsystem list heads cgroup->e_csets[]. Any
css_set which is pointing to a css is linked to
css->cgroup->e_csets[$SUBSYS_ID] through
css_set->e_cset_node[$SUBSYS_ID]. The lists are protected by
css_set_rwsem and will allow us to walk all css_sets associated with a
given css so that we can find out all associated tasks.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
In the planned default unified hierarchy, controllers may get
dynamically attached to and detached from a cgroup and a cgroup may
not have csses for all the controllers associated with the hierarchy.
When a cgroup doesn't have its own css for a given controller, the css
of the nearest ancestor with the controller enabled will be used,
which is called the effective css. This patch introduces
cgroup_e_css() and for_each_e_css() to access the effective csses and
convert compare_css_sets(), find_existing_css_set() and
cgroup_migrate() to use the effective csses so that they can handle
cgroups with partial csses correctly.
This means that for two css_sets to be considered identical, they
should have both matching csses and cgroups. compare_css_sets()
already compares both, not for correctness but for optimization. As
this now becomes a matter of correctness, update the comments
accordingly.
For all !default hierarchies, cgroup_e_css() always equals
cgroup_css(), so this patch doesn't change behavior.
While at it, fix incorrect locking comment for for_each_css().
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
944196278d ("cgroup: move ->subsys_mask from cgroupfs_root to
cgroup") moved ->subsys_mask from cgroup_root to cgroup to prepare for
the unified hierarhcy; however, it turns out that carrying the
subsys_mask of the children in the parent, instead of itself, is a lot
more natural. This patch restores cgroup_root->subsys_mask and morphs
cgroup->subsys_mask into cgroup->child_subsys_mask.
* Uses of root->cgrp.subsys_mask are restored to root->subsys_mask.
* Remove automatic setting and clearing of cgrp->subsys_mask and
instead just inherit ->child_subsys_mask from the parent during
cgroup creation. Note that this doesn't affect any current
behaviors.
* Undo __kill_css() separation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_apply_cftypes() skip creating or removing files if the
subsystem is attached to the default hierarchy, which led to missing
files in the root of the default hierarchy.
Skipping made sense when the default hierarchy was dummy; however, now
that the default hierarchy is full functional and planned to be used
as the unified hierarchy, it shouldn't be skipped over.
Reported-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
If we hit the retry path, we'll call parse_cgroupfs_options() again,
but the string we pass to it has been modified by the previous call
to this function.
This bug can be observed by:
# mount -t cgroup -o name=foo,cpuset xxx /mnt && umount /mnt && \
mount -t cgroup -o name=foo,cpuset xxx /mnt
mount: wrong fs type, bad option, bad superblock on xxx,
missing codepage or helper program, or other error
...
The second mount passed "name=foo,cpuset" to the parser, and then it
hit the retry path and call the parser again, but this time the string
passed to the parser is "name=foo".
To fix this, we avoid calling parse_cgroupfs_options() again in this
case.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
While converting cgroup to kernfs, 2bd59d48eb ("cgroup: convert to
kernfs") accidentally dropped the logic which makes newly created
cgroup dirs and files owned by the current uid / gid. This broke
cases where cgroup subtree management is delegated to !root as the sub
manager wouldn't be able to create more than single level of hierarchy
or put tasks into child cgroups it created.
Among other things, this breaks user session management in systemd and
one of the symptoms was 90s hang during shutdown. User session
systemd running as the user creates a sub-service to initiate shutdown
and tries to put kill(1) into it but fails because cgroup.procs is
owned by root. This leads to 90s hang during shutdown.
Implement cgroup_kn_set_ugid() which sets a kn's uid and gid to those
of the caller and use it from file and dir creation paths.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
As mount() and kill_sb() is not a one-to-one match, If we mount the same
cgroupfs in serveral mount points, and then umount all of them, kill_sb()
will be called only once.
Try:
# mount -t cgroup -o cpuacct xxx /cgroup
# mount -t cgroup -o cpuacct xxx /cgroup2
# cat /proc/cgroups | grep cpuacct
cpuacct 2 1 1
# umount /cgroup
# umount /cgroup2
# cat /proc/cgroups | grep cpuacct
cpuacct 2 1 1
You'll see cgroupfs will never be freed.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Pull cgroup updates from Tejun Heo:
"A lot updates for cgroup:
- The biggest one is cgroup's conversion to kernfs. cgroup took
after the long abandoned vfs-entangled sysfs implementation and
made it even more convoluted over time. cgroup's internal objects
were fused with vfs objects which also brought in vfs locking and
object lifetime rules. Naturally, there are places where vfs rules
don't fit and nasty hacks, such as credential switching or lock
dance interleaving inode mutex and cgroup_mutex with object serial
number comparison thrown in to decide whether the operation is
actually necessary, needed to be employed.
After conversion to kernfs, internal object lifetime and locking
rules are mostly isolated from vfs interactions allowing shedding
of several nasty hacks and overall simplification. This will also
allow implmentation of operations which may affect multiple cgroups
which weren't possible before as it would have required nesting
i_mutexes.
- Various simplifications including dropping of module support,
easier cgroup name/path handling, simplified cgroup file type
handling and task_cg_lists optimization.
- Prepatory changes for the planned unified hierarchy, which is still
a patchset away from being actually operational. The dummy
hierarchy is updated to serve as the default unified hierarchy.
Controllers which aren't claimed by other hierarchies are
associated with it, which BTW was what the dummy hierarchy was for
anyway.
- Various fixes from Li and others. This pull request includes some
patches to add missing slab.h to various subsystems. This was
triggered xattr.h include removal from cgroup.h. cgroup.h
indirectly got included a lot of files which brought in xattr.h
which brought in slab.h.
There are several merge commits - one to pull in kernfs updates
necessary for converting cgroup (already in upstream through
driver-core), others for interfering changes in the fixes branch"
* 'for-3.15' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup: (74 commits)
cgroup: remove useless argument from cgroup_exit()
cgroup: fix spurious lockdep warning in cgroup_exit()
cgroup: Use RCU_INIT_POINTER(x, NULL) in cgroup.c
cgroup: break kernfs active_ref protection in cgroup directory operations
cgroup: fix cgroup_taskset walking order
cgroup: implement CFTYPE_ONLY_ON_DFL
cgroup: make cgrp_dfl_root mountable
cgroup: drop const from @buffer of cftype->write_string()
cgroup: rename cgroup_dummy_root and related names
cgroup: move ->subsys_mask from cgroupfs_root to cgroup
cgroup: treat cgroup_dummy_root as an equivalent hierarchy during rebinding
cgroup: remove NULL checks from [pr_cont_]cgroup_{name|path}()
cgroup: use cgroup_setup_root() to initialize cgroup_dummy_root
cgroup: reorganize cgroup bootstrapping
cgroup: relocate setting of CGRP_DEAD
cpuset: use rcu_read_lock() to protect task_cs()
cgroup_freezer: document freezer_fork() subtleties
cgroup: update cgroup_transfer_tasks() to either succeed or fail
cgroup: drop task_lock() protection around task->cgroups
cgroup: update how a newly forked task gets associated with css_set
...
cgroup_exit() is called in fork and exit path. If it's called in the
failure path during fork, PF_EXITING isn't set, and then lockdep will
complain.
Fix this by removing cgroup_exit() in that failure path. cgroup_fork()
does nothing that needs cleanup.
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
This patch replaces rcu_assign_pointer(x, NULL) with
RCU_INIT_POINTER(x, NULL)
The rcu_assign_pointer() ensures that the initialization of a
structure is carried out before storing a pointer to that structure.
And in the case of the NULL pointer, there is no structure to
initialize. So, rcu_assign_pointer(p, NULL) can be safely converted
to RCU_INIT_POINTER(p, NULL)
Signed-off-by: Monam Agarwal <monamagarwal123@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
cgroup_tree_mutex should nest above the kernfs active_ref protection;
however, cgroup_create() and cgroup_rename() were grabbing
cgroup_tree_mutex while under kernfs active_ref protection. This has
actualy possibility to lead to deadlocks in case these operations race
against cgroup_rmdir() which invokes kernfs_remove() on directory
kernfs_node while holding cgroup_tree_mutex.
Neither cgroup_create() or cgroup_rename() requires active_ref
protection. The former already has enough synchronization through
cgroup_lock_live_group() and the latter doesn't care, so this can be
fixed by updating both functions to break all active_ref protections
before grabbing cgroup_tree_mutex.
While this patch fixes the immediate issue, it probably needs further
work in the long term - kernfs directories should enable lockdep
annotations and maybe the better way to handle this is marking
directory nodes as not needing active_ref protection rather than
breaking it in each operation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
cgroup_taskset is used to track and iterate target tasks while
migrating a task or process and should guarantee that the first task
iterated is the task group leader if a process is being migrated.
b3dc094e93 ("cgroup: use css_set->mg_tasks to track target tasks
during migration") replaced flex array cgroup_taskset->tc_array with
css_set->mg_tasks list to remove process size limit and dynamic
allocation during migration; unfortunately, it incorrectly used list
operations which don't preserve order breaking the guarantee that
cgroup_taskset_first() returns the leader for a process target.
Fix it by using order preserving list operations. Note that as
multiple src_csets may map to a single dst_cset, the iteration order
may change across cgroup_task_migrate(); however, the leader is still
guaranteed to be the first entry.
The switch to list_splice_tail_init() at the end of cgroup_migrate()
isn't strictly necessary. Let's still do it for consistency.
Signed-off-by: Tejun Heo <tj@kernel.org>
This cftype flag makes the file only appear on the default hierarchy.
This will later be used for cgroup.controllers file.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgrp_dfl_root will be used as the default unified hierarchy. This
patch makes cgrp_dfl_root mountable by making the following changes.
* cgroup_init_early() now initializes cgrp_dfl_root w/
CGRP_ROOT_SANE_BEHAVIOR. The default hierarchy is always sane.
* parse_cgroupfs_options() and cgroup_mount() are updated such that
cgrp_dfl_root is mounted if sane_behavior is specified w/o any
subsystems.
* rebind_subsystems() now populates the root directory of
cgrp_dfl_root. Note that the function still guarantees success of
rebinding subsystems to cgrp_dfl_root. If populating fails while
rebinding to cgrp_dfl_root, it whines but ignores the error.
* For backward compatibility, the default hierarchy shows up in
/proc/$PID/cgroup only after it's explicitly mounted so that
userland which doesn't make use of it doesn't see any change.
* "current_css_set_cg_links" file of debug cgroup now treats the
default hierarchy the same as other hierarchies. This is visible to
userland. Given that it's for debug controller, this should be
fine.
* While at it, implement cgroup_on_dfl() which tests whether a give
cgroup is on the default hierarchy or not.
The above changes make cgrp_dfl_root mostly equivalent to other
controllers but the actual unified hierarchy behaviors are not
implemented yet. Let's plug child cgroup creation in cgrp_dfl_root
from create_cgroup() for now.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cftype->write_string() just passes on the writeable buffer from kernfs
and there's no reason to add const restriction on the buffer. The
only thing const achieves is unnecessarily complicating parsing of the
buffer. Drop const from @buffer.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Daniel Borkmann <dborkman@redhat.com>
Cc: 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>
The dummy root will be repurposed to serve as the default unified
hierarchy. Let's rename things in preparation.
* s/cgroup_dummy_root/cgrp_dfl_root/
* s/cgroupfs_root/cgroup_root/ as we don't do fs part directly anymore
* s/cgroup_root->top_cgroup/cgroup_root->cgrp/ for brevity
This is pure rename.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroupfs_root->subsys_mask represents the controllers attached to the
hierarchy. This patch moves the field to cgroup. Subsystem
initialization and rebinding updates the top cgroup's subsys_mask.
For !root cgroups, the subsys_mask bits are set from create_css() and
cleared from kill_css(), which effectively means that all cgroups will
have the same subsys_mask as the top cgroup.
While this doesn't make any difference now, this will help
implementation of the default unified hierarchy where !root cgroups
may have subsets of the top_cgroup's subsys_mask.
While at it, __kill_css() is split out of kill_css(). The former
doesn't care about the subsys_mask while the latter becomes noop if
the controller is already killed and clears the matching bit if not
before proceeding to killing the css. This will be used later by the
default unified hierarchy implementation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
Currently, while rebinding, cgroup_dummy_root serves as the anchor
point. In addition to the target root, rebind_subsystems() takes
@added_mask and @removed_mask. The subsystems specified in the former
are expected to be on the dummy root and then moved to the target
root. The ones in the latter are moved from non-dummy root to dummy.
Now that the dummy root is a fully functional one and we're planning
to use it for the default unified hierarchy, this level of distinction
between dummy and non-dummy roots is quite awkward.
This patch updates rebind_subsystems() to take the target root and one
subsystem mask and move the specified subsystmes to the target root
which may or may not be the dummy root. IOW, unbinding now becomes
moving the subsystems to the dummy root and binding to non-dummy root.
This makes the dummy root mostly equivalent to other hierarchies in
terms of the mechanism of moving subsystems around; however, we still
retain all the semantical restrictions so that this patch doesn't
introduce any visible behavior differences. Another noteworthy detail
is that rebind_subsystems() guarantees that moving a subsystem to the
dummy root never fails so that valid unmounting attempts always
succeed.
This unifies binding and unbinding of subsystems. The invocation
points of ->bind() were inconsistent between the two and now moved
after whole rebinding is complete. This doesn't break the current
users and generally makes more sense.
All rebind_subsystems() users are converted accordingly. Note that
cgroup_remount() now makes two calls to rebind_subsystems() to bind
and then unbind the requested subsystems.
This will allow repurposing of the dummy hierarchy as the default
unified hierarchy and shouldn't make any userland visible behavior
difference.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
cgroup_dummy_root is used to host controllers which aren't attached to
any other hierarchy. The root is minimally set up during kernfs
bootstrap and didn't go through full hierarchy initialization. We're
planning to use cgroup_dummy_root for the default unified hierarchy
and thus want it to be fully functional.
Replace the special initialization, which was collected into
cgroup_init() by the previous patch, with an invocation of
cgroup_setup_root(). This simplifies the init path and makes
cgroup_dummy_root a full hierarchy with its own kernfs_root and all.
As this puts the dummy hierarchy on the cgroup_roots list, rename
for_each_active_root() to for_each_root() and update its users to skip
the dummy root for now.
This patch doesn't cause any userland visible behavior changes at this
point.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
* Fields of init_css_set and css_set_count are now set using
initializer instead of programmatically from cgroup_init_early().
* init_cgroup_root() now also takes @opts and performs the optional
part of initialization too. The leftover part of
cgroup_root_from_opts() is collapsed into its only caller -
cgroup_mount().
* Initialization of cgroup_root_count and linking of init_css_set are
moved from cgroup_init_early() to to cgroup_init(). None of the
early_init users depends on init_css_set being linked.
* Subsystem initializations are moved after dummy hierarchy init and
init_css_set linking.
These changes reorganize the bootstrap logic so that the dummy
hierarchy can share the usual hierarchy init path and be made more
normal. These changes don't make noticeable behavior changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
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>
If online_css() fails, we should remove cgroup files belonging
to css->ss.
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
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
