2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
33 #include <linux/init_task.h>
34 #include <linux/kernel.h>
35 #include <linux/list.h>
37 #include <linux/mutex.h>
38 #include <linux/mount.h>
39 #include <linux/pagemap.h>
40 #include <linux/proc_fs.h>
41 #include <linux/rcupdate.h>
42 #include <linux/sched.h>
43 #include <linux/backing-dev.h>
44 #include <linux/seq_file.h>
45 #include <linux/slab.h>
46 #include <linux/magic.h>
47 #include <linux/spinlock.h>
48 #include <linux/string.h>
49 #include <linux/sort.h>
50 #include <linux/kmod.h>
51 #include <linux/module.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/namei.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/idr.h>
58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
59 #include <linux/eventfd.h>
60 #include <linux/poll.h>
61 #include <linux/flex_array.h> /* used in cgroup_attach_task */
62 #include <linux/kthread.h>
63 #include <linux/file.h>
65 #include <linux/atomic.h>
68 * cgroup_mutex is the master lock. Any modification to cgroup or its
69 * hierarchy must be performed while holding it.
71 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
72 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
73 * release_agent_path and so on. Modifying requires both cgroup_mutex and
74 * cgroup_root_mutex. Readers can acquire either of the two. This is to
75 * break the following locking order cycle.
77 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
78 * B. namespace_sem -> cgroup_mutex
80 * B happens only through cgroup_show_options() and using cgroup_root_mutex
83 #ifdef CONFIG_PROVE_RCU
84 DEFINE_MUTEX(cgroup_mutex
);
85 EXPORT_SYMBOL_GPL(cgroup_mutex
); /* only for lockdep */
87 static DEFINE_MUTEX(cgroup_mutex
);
90 static DEFINE_MUTEX(cgroup_root_mutex
);
93 * Generate an array of cgroup subsystem pointers. At boot time, this is
94 * populated with the built in subsystems, and modular subsystems are
95 * registered after that. The mutable section of this array is protected by
98 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
99 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
100 static struct cgroup_subsys
*cgroup_subsys
[CGROUP_SUBSYS_COUNT
] = {
101 #include <linux/cgroup_subsys.h>
105 * The dummy hierarchy, reserved for the subsystems that are otherwise
106 * unattached - it never has more than a single cgroup, and all tasks are
107 * part of that cgroup.
109 static struct cgroupfs_root cgroup_dummy_root
;
111 /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */
112 static struct cgroup
* const cgroup_dummy_top
= &cgroup_dummy_root
.top_cgroup
;
115 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
118 struct list_head node
;
119 struct dentry
*dentry
;
121 struct cgroup_subsys_state
*css
;
124 struct simple_xattrs xattrs
;
128 * cgroup_event represents events which userspace want to receive.
130 struct cgroup_event
{
132 * css which the event belongs to.
134 struct cgroup_subsys_state
*css
;
136 * Control file which the event associated.
140 * eventfd to signal userspace about the event.
142 struct eventfd_ctx
*eventfd
;
144 * Each of these stored in a list by the cgroup.
146 struct list_head list
;
148 * All fields below needed to unregister event when
149 * userspace closes eventfd.
152 wait_queue_head_t
*wqh
;
154 struct work_struct remove
;
157 /* The list of hierarchy roots */
159 static LIST_HEAD(cgroup_roots
);
160 static int cgroup_root_count
;
163 * Hierarchy ID allocation and mapping. It follows the same exclusion
164 * rules as other root ops - both cgroup_mutex and cgroup_root_mutex for
165 * writes, either for reads.
167 static DEFINE_IDR(cgroup_hierarchy_idr
);
169 static struct cgroup_name root_cgroup_name
= { .name
= "/" };
172 * Assign a monotonically increasing serial number to cgroups. It
173 * guarantees cgroups with bigger numbers are newer than those with smaller
174 * numbers. Also, as cgroups are always appended to the parent's
175 * ->children list, it guarantees that sibling cgroups are always sorted in
176 * the ascending serial number order on the list. Protected by
179 static u64 cgroup_serial_nr_next
= 1;
181 /* This flag indicates whether tasks in the fork and exit paths should
182 * check for fork/exit handlers to call. This avoids us having to do
183 * extra work in the fork/exit path if none of the subsystems need to
186 static int need_forkexit_callback __read_mostly
;
188 static struct cftype cgroup_base_files
[];
190 static void cgroup_destroy_css_killed(struct cgroup
*cgrp
);
191 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
192 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
196 * cgroup_css - obtain a cgroup's css for the specified subsystem
197 * @cgrp: the cgroup of interest
198 * @ss: the subsystem of interest (%NULL returns the dummy_css)
200 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
201 * function must be called either under cgroup_mutex or rcu_read_lock() and
202 * the caller is responsible for pinning the returned css if it wants to
203 * keep accessing it outside the said locks. This function may return
204 * %NULL if @cgrp doesn't have @subsys_id enabled.
206 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
207 struct cgroup_subsys
*ss
)
210 return rcu_dereference_check(cgrp
->subsys
[ss
->subsys_id
],
211 lockdep_is_held(&cgroup_mutex
));
213 return &cgrp
->dummy_css
;
216 /* convenient tests for these bits */
217 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
219 return test_bit(CGRP_DEAD
, &cgrp
->flags
);
223 * cgroup_is_descendant - test ancestry
224 * @cgrp: the cgroup to be tested
225 * @ancestor: possible ancestor of @cgrp
227 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
228 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
229 * and @ancestor are accessible.
231 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
234 if (cgrp
== ancestor
)
240 EXPORT_SYMBOL_GPL(cgroup_is_descendant
);
242 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
245 (1 << CGRP_RELEASABLE
) |
246 (1 << CGRP_NOTIFY_ON_RELEASE
);
247 return (cgrp
->flags
& bits
) == bits
;
250 static int notify_on_release(const struct cgroup
*cgrp
)
252 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
256 * for_each_subsys - iterate all loaded cgroup subsystems
257 * @ss: the iteration cursor
258 * @i: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
260 * Should be called under cgroup_mutex.
262 #define for_each_subsys(ss, i) \
263 for ((i) = 0; (i) < CGROUP_SUBSYS_COUNT; (i)++) \
264 if (({ lockdep_assert_held(&cgroup_mutex); \
265 !((ss) = cgroup_subsys[i]); })) { } \
269 * for_each_builtin_subsys - iterate all built-in cgroup subsystems
270 * @ss: the iteration cursor
271 * @i: the index of @ss, CGROUP_BUILTIN_SUBSYS_COUNT after reaching the end
273 * Bulit-in subsystems are always present and iteration itself doesn't
274 * require any synchronization.
276 #define for_each_builtin_subsys(ss, i) \
277 for ((i) = 0; (i) < CGROUP_BUILTIN_SUBSYS_COUNT && \
278 (((ss) = cgroup_subsys[i]) || true); (i)++)
280 /* iterate each subsystem attached to a hierarchy */
281 #define for_each_root_subsys(root, ss) \
282 list_for_each_entry((ss), &(root)->subsys_list, sibling)
284 /* iterate across the active hierarchies */
285 #define for_each_active_root(root) \
286 list_for_each_entry((root), &cgroup_roots, root_list)
288 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
290 return dentry
->d_fsdata
;
293 static inline struct cfent
*__d_cfe(struct dentry
*dentry
)
295 return dentry
->d_fsdata
;
298 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
300 return __d_cfe(dentry
)->type
;
304 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
305 * @cgrp: the cgroup to be checked for liveness
307 * On success, returns true; the mutex should be later unlocked. On
308 * failure returns false with no lock held.
310 static bool cgroup_lock_live_group(struct cgroup
*cgrp
)
312 mutex_lock(&cgroup_mutex
);
313 if (cgroup_is_dead(cgrp
)) {
314 mutex_unlock(&cgroup_mutex
);
320 /* the list of cgroups eligible for automatic release. Protected by
321 * release_list_lock */
322 static LIST_HEAD(release_list
);
323 static DEFINE_RAW_SPINLOCK(release_list_lock
);
324 static void cgroup_release_agent(struct work_struct
*work
);
325 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
326 static void check_for_release(struct cgroup
*cgrp
);
329 * A cgroup can be associated with multiple css_sets as different tasks may
330 * belong to different cgroups on different hierarchies. In the other
331 * direction, a css_set is naturally associated with multiple cgroups.
332 * This M:N relationship is represented by the following link structure
333 * which exists for each association and allows traversing the associations
336 struct cgrp_cset_link
{
337 /* the cgroup and css_set this link associates */
339 struct css_set
*cset
;
341 /* list of cgrp_cset_links anchored at cgrp->cset_links */
342 struct list_head cset_link
;
344 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
345 struct list_head cgrp_link
;
348 /* The default css_set - used by init and its children prior to any
349 * hierarchies being mounted. It contains a pointer to the root state
350 * for each subsystem. Also used to anchor the list of css_sets. Not
351 * reference-counted, to improve performance when child cgroups
352 * haven't been created.
355 static struct css_set init_css_set
;
356 static struct cgrp_cset_link init_cgrp_cset_link
;
359 * css_set_lock protects the list of css_set objects, and the chain of
360 * tasks off each css_set. Nests outside task->alloc_lock due to
361 * css_task_iter_start().
363 static DEFINE_RWLOCK(css_set_lock
);
364 static int css_set_count
;
367 * hash table for cgroup groups. This improves the performance to find
368 * an existing css_set. This hash doesn't (currently) take into
369 * account cgroups in empty hierarchies.
371 #define CSS_SET_HASH_BITS 7
372 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
374 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
376 unsigned long key
= 0UL;
377 struct cgroup_subsys
*ss
;
380 for_each_subsys(ss
, i
)
381 key
+= (unsigned long)css
[i
];
382 key
= (key
>> 16) ^ key
;
388 * We don't maintain the lists running through each css_set to its task
389 * until after the first call to css_task_iter_start(). This reduces the
390 * fork()/exit() overhead for people who have cgroups compiled into their
391 * kernel but not actually in use.
393 static int use_task_css_set_links __read_mostly
;
395 static void __put_css_set(struct css_set
*cset
, int taskexit
)
397 struct cgrp_cset_link
*link
, *tmp_link
;
400 * Ensure that the refcount doesn't hit zero while any readers
401 * can see it. Similar to atomic_dec_and_lock(), but for an
404 if (atomic_add_unless(&cset
->refcount
, -1, 1))
406 write_lock(&css_set_lock
);
407 if (!atomic_dec_and_test(&cset
->refcount
)) {
408 write_unlock(&css_set_lock
);
412 /* This css_set is dead. unlink it and release cgroup refcounts */
413 hash_del(&cset
->hlist
);
416 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
417 struct cgroup
*cgrp
= link
->cgrp
;
419 list_del(&link
->cset_link
);
420 list_del(&link
->cgrp_link
);
422 /* @cgrp can't go away while we're holding css_set_lock */
423 if (list_empty(&cgrp
->cset_links
) && notify_on_release(cgrp
)) {
425 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
426 check_for_release(cgrp
);
432 write_unlock(&css_set_lock
);
433 kfree_rcu(cset
, rcu_head
);
437 * refcounted get/put for css_set objects
439 static inline void get_css_set(struct css_set
*cset
)
441 atomic_inc(&cset
->refcount
);
444 static inline void put_css_set(struct css_set
*cset
)
446 __put_css_set(cset
, 0);
449 static inline void put_css_set_taskexit(struct css_set
*cset
)
451 __put_css_set(cset
, 1);
455 * compare_css_sets - helper function for find_existing_css_set().
456 * @cset: candidate css_set being tested
457 * @old_cset: existing css_set for a task
458 * @new_cgrp: cgroup that's being entered by the task
459 * @template: desired set of css pointers in css_set (pre-calculated)
461 * Returns true if "cset" matches "old_cset" except for the hierarchy
462 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
464 static bool compare_css_sets(struct css_set
*cset
,
465 struct css_set
*old_cset
,
466 struct cgroup
*new_cgrp
,
467 struct cgroup_subsys_state
*template[])
469 struct list_head
*l1
, *l2
;
471 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
))) {
472 /* Not all subsystems matched */
477 * Compare cgroup pointers in order to distinguish between
478 * different cgroups in heirarchies with no subsystems. We
479 * could get by with just this check alone (and skip the
480 * memcmp above) but on most setups the memcmp check will
481 * avoid the need for this more expensive check on almost all
485 l1
= &cset
->cgrp_links
;
486 l2
= &old_cset
->cgrp_links
;
488 struct cgrp_cset_link
*link1
, *link2
;
489 struct cgroup
*cgrp1
, *cgrp2
;
493 /* See if we reached the end - both lists are equal length. */
494 if (l1
== &cset
->cgrp_links
) {
495 BUG_ON(l2
!= &old_cset
->cgrp_links
);
498 BUG_ON(l2
== &old_cset
->cgrp_links
);
500 /* Locate the cgroups associated with these links. */
501 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
502 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
505 /* Hierarchies should be linked in the same order. */
506 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
509 * If this hierarchy is the hierarchy of the cgroup
510 * that's changing, then we need to check that this
511 * css_set points to the new cgroup; if it's any other
512 * hierarchy, then this css_set should point to the
513 * same cgroup as the old css_set.
515 if (cgrp1
->root
== new_cgrp
->root
) {
516 if (cgrp1
!= new_cgrp
)
527 * find_existing_css_set - init css array and find the matching css_set
528 * @old_cset: the css_set that we're using before the cgroup transition
529 * @cgrp: the cgroup that we're moving into
530 * @template: out param for the new set of csses, should be clear on entry
532 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
534 struct cgroup_subsys_state
*template[])
536 struct cgroupfs_root
*root
= cgrp
->root
;
537 struct cgroup_subsys
*ss
;
538 struct css_set
*cset
;
543 * Build the set of subsystem state objects that we want to see in the
544 * new css_set. while subsystems can change globally, the entries here
545 * won't change, so no need for locking.
547 for_each_subsys(ss
, i
) {
548 if (root
->subsys_mask
& (1UL << i
)) {
549 /* Subsystem is in this hierarchy. So we want
550 * the subsystem state from the new
552 template[i
] = cgroup_css(cgrp
, ss
);
554 /* Subsystem is not in this hierarchy, so we
555 * don't want to change the subsystem state */
556 template[i
] = old_cset
->subsys
[i
];
560 key
= css_set_hash(template);
561 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
562 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
565 /* This css_set matches what we need */
569 /* No existing cgroup group matched */
573 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
575 struct cgrp_cset_link
*link
, *tmp_link
;
577 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
578 list_del(&link
->cset_link
);
584 * allocate_cgrp_cset_links - allocate cgrp_cset_links
585 * @count: the number of links to allocate
586 * @tmp_links: list_head the allocated links are put on
588 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
589 * through ->cset_link. Returns 0 on success or -errno.
591 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
593 struct cgrp_cset_link
*link
;
596 INIT_LIST_HEAD(tmp_links
);
598 for (i
= 0; i
< count
; i
++) {
599 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
601 free_cgrp_cset_links(tmp_links
);
604 list_add(&link
->cset_link
, tmp_links
);
610 * link_css_set - a helper function to link a css_set to a cgroup
611 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
612 * @cset: the css_set to be linked
613 * @cgrp: the destination cgroup
615 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
618 struct cgrp_cset_link
*link
;
620 BUG_ON(list_empty(tmp_links
));
621 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
624 list_move(&link
->cset_link
, &cgrp
->cset_links
);
626 * Always add links to the tail of the list so that the list
627 * is sorted by order of hierarchy creation
629 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
633 * find_css_set - return a new css_set with one cgroup updated
634 * @old_cset: the baseline css_set
635 * @cgrp: the cgroup to be updated
637 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
638 * substituted into the appropriate hierarchy.
640 static struct css_set
*find_css_set(struct css_set
*old_cset
,
643 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
644 struct css_set
*cset
;
645 struct list_head tmp_links
;
646 struct cgrp_cset_link
*link
;
649 lockdep_assert_held(&cgroup_mutex
);
651 /* First see if we already have a cgroup group that matches
653 read_lock(&css_set_lock
);
654 cset
= find_existing_css_set(old_cset
, cgrp
, template);
657 read_unlock(&css_set_lock
);
662 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
666 /* Allocate all the cgrp_cset_link objects that we'll need */
667 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
672 atomic_set(&cset
->refcount
, 1);
673 INIT_LIST_HEAD(&cset
->cgrp_links
);
674 INIT_LIST_HEAD(&cset
->tasks
);
675 INIT_HLIST_NODE(&cset
->hlist
);
677 /* Copy the set of subsystem state objects generated in
678 * find_existing_css_set() */
679 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
681 write_lock(&css_set_lock
);
682 /* Add reference counts and links from the new css_set. */
683 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
684 struct cgroup
*c
= link
->cgrp
;
686 if (c
->root
== cgrp
->root
)
688 link_css_set(&tmp_links
, cset
, c
);
691 BUG_ON(!list_empty(&tmp_links
));
695 /* Add this cgroup group to the hash table */
696 key
= css_set_hash(cset
->subsys
);
697 hash_add(css_set_table
, &cset
->hlist
, key
);
699 write_unlock(&css_set_lock
);
705 * Return the cgroup for "task" from the given hierarchy. Must be
706 * called with cgroup_mutex held.
708 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
709 struct cgroupfs_root
*root
)
711 struct css_set
*cset
;
712 struct cgroup
*res
= NULL
;
714 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
715 read_lock(&css_set_lock
);
717 * No need to lock the task - since we hold cgroup_mutex the
718 * task can't change groups, so the only thing that can happen
719 * is that it exits and its css is set back to init_css_set.
721 cset
= task_css_set(task
);
722 if (cset
== &init_css_set
) {
723 res
= &root
->top_cgroup
;
725 struct cgrp_cset_link
*link
;
727 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
728 struct cgroup
*c
= link
->cgrp
;
730 if (c
->root
== root
) {
736 read_unlock(&css_set_lock
);
742 * There is one global cgroup mutex. We also require taking
743 * task_lock() when dereferencing a task's cgroup subsys pointers.
744 * See "The task_lock() exception", at the end of this comment.
746 * A task must hold cgroup_mutex to modify cgroups.
748 * Any task can increment and decrement the count field without lock.
749 * So in general, code holding cgroup_mutex can't rely on the count
750 * field not changing. However, if the count goes to zero, then only
751 * cgroup_attach_task() can increment it again. Because a count of zero
752 * means that no tasks are currently attached, therefore there is no
753 * way a task attached to that cgroup can fork (the other way to
754 * increment the count). So code holding cgroup_mutex can safely
755 * assume that if the count is zero, it will stay zero. Similarly, if
756 * a task holds cgroup_mutex on a cgroup with zero count, it
757 * knows that the cgroup won't be removed, as cgroup_rmdir()
760 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
761 * (usually) take cgroup_mutex. These are the two most performance
762 * critical pieces of code here. The exception occurs on cgroup_exit(),
763 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
764 * is taken, and if the cgroup count is zero, a usermode call made
765 * to the release agent with the name of the cgroup (path relative to
766 * the root of cgroup file system) as the argument.
768 * A cgroup can only be deleted if both its 'count' of using tasks
769 * is zero, and its list of 'children' cgroups is empty. Since all
770 * tasks in the system use _some_ cgroup, and since there is always at
771 * least one task in the system (init, pid == 1), therefore, top_cgroup
772 * always has either children cgroups and/or using tasks. So we don't
773 * need a special hack to ensure that top_cgroup cannot be deleted.
775 * The task_lock() exception
777 * The need for this exception arises from the action of
778 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
779 * another. It does so using cgroup_mutex, however there are
780 * several performance critical places that need to reference
781 * task->cgroup without the expense of grabbing a system global
782 * mutex. Therefore except as noted below, when dereferencing or, as
783 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
784 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
785 * the task_struct routinely used for such matters.
787 * P.S. One more locking exception. RCU is used to guard the
788 * update of a tasks cgroup pointer by cgroup_attach_task()
792 * A couple of forward declarations required, due to cyclic reference loop:
793 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
794 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
798 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
799 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
800 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
);
801 static const struct inode_operations cgroup_dir_inode_operations
;
802 static const struct file_operations proc_cgroupstats_operations
;
804 static struct backing_dev_info cgroup_backing_dev_info
= {
806 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
809 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
811 struct inode
*inode
= new_inode(sb
);
814 inode
->i_ino
= get_next_ino();
815 inode
->i_mode
= mode
;
816 inode
->i_uid
= current_fsuid();
817 inode
->i_gid
= current_fsgid();
818 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
819 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
824 static struct cgroup_name
*cgroup_alloc_name(struct dentry
*dentry
)
826 struct cgroup_name
*name
;
828 name
= kmalloc(sizeof(*name
) + dentry
->d_name
.len
+ 1, GFP_KERNEL
);
831 strcpy(name
->name
, dentry
->d_name
.name
);
835 static void cgroup_free_fn(struct work_struct
*work
)
837 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, destroy_work
);
839 mutex_lock(&cgroup_mutex
);
840 cgrp
->root
->number_of_cgroups
--;
841 mutex_unlock(&cgroup_mutex
);
844 * We get a ref to the parent's dentry, and put the ref when
845 * this cgroup is being freed, so it's guaranteed that the
846 * parent won't be destroyed before its children.
848 dput(cgrp
->parent
->dentry
);
851 * Drop the active superblock reference that we took when we
852 * created the cgroup. This will free cgrp->root, if we are
853 * holding the last reference to @sb.
855 deactivate_super(cgrp
->root
->sb
);
858 * if we're getting rid of the cgroup, refcount should ensure
859 * that there are no pidlists left.
861 BUG_ON(!list_empty(&cgrp
->pidlists
));
863 simple_xattrs_free(&cgrp
->xattrs
);
865 kfree(rcu_dereference_raw(cgrp
->name
));
869 static void cgroup_free_rcu(struct rcu_head
*head
)
871 struct cgroup
*cgrp
= container_of(head
, struct cgroup
, rcu_head
);
873 INIT_WORK(&cgrp
->destroy_work
, cgroup_free_fn
);
874 schedule_work(&cgrp
->destroy_work
);
877 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
879 /* is dentry a directory ? if so, kfree() associated cgroup */
880 if (S_ISDIR(inode
->i_mode
)) {
881 struct cgroup
*cgrp
= dentry
->d_fsdata
;
883 BUG_ON(!(cgroup_is_dead(cgrp
)));
884 call_rcu(&cgrp
->rcu_head
, cgroup_free_rcu
);
886 struct cfent
*cfe
= __d_cfe(dentry
);
887 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
889 WARN_ONCE(!list_empty(&cfe
->node
) &&
890 cgrp
!= &cgrp
->root
->top_cgroup
,
891 "cfe still linked for %s\n", cfe
->type
->name
);
892 simple_xattrs_free(&cfe
->xattrs
);
898 static void remove_dir(struct dentry
*d
)
900 struct dentry
*parent
= dget(d
->d_parent
);
903 simple_rmdir(parent
->d_inode
, d
);
907 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
911 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
912 lockdep_assert_held(&cgroup_mutex
);
915 * If we're doing cleanup due to failure of cgroup_create(),
916 * the corresponding @cfe may not exist.
918 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
919 struct dentry
*d
= cfe
->dentry
;
921 if (cft
&& cfe
->type
!= cft
)
926 simple_unlink(cgrp
->dentry
->d_inode
, d
);
927 list_del_init(&cfe
->node
);
935 * cgroup_clear_dir - remove subsys files in a cgroup directory
936 * @cgrp: target cgroup
937 * @subsys_mask: mask of the subsystem ids whose files should be removed
939 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
941 struct cgroup_subsys
*ss
;
944 for_each_subsys(ss
, i
) {
945 struct cftype_set
*set
;
947 if (!test_bit(i
, &subsys_mask
))
949 list_for_each_entry(set
, &ss
->cftsets
, node
)
950 cgroup_addrm_files(cgrp
, set
->cfts
, false);
955 * NOTE : the dentry must have been dget()'ed
957 static void cgroup_d_remove_dir(struct dentry
*dentry
)
959 struct dentry
*parent
;
961 parent
= dentry
->d_parent
;
962 spin_lock(&parent
->d_lock
);
963 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
964 list_del_init(&dentry
->d_u
.d_child
);
965 spin_unlock(&dentry
->d_lock
);
966 spin_unlock(&parent
->d_lock
);
971 * Call with cgroup_mutex held. Drops reference counts on modules, including
972 * any duplicate ones that parse_cgroupfs_options took. If this function
973 * returns an error, no reference counts are touched.
975 static int rebind_subsystems(struct cgroupfs_root
*root
,
976 unsigned long added_mask
, unsigned removed_mask
)
978 struct cgroup
*cgrp
= &root
->top_cgroup
;
979 struct cgroup_subsys
*ss
;
980 unsigned long pinned
= 0;
983 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
984 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
986 /* Check that any added subsystems are currently free */
987 for_each_subsys(ss
, i
) {
988 if (!(added_mask
& (1 << i
)))
991 /* is the subsystem mounted elsewhere? */
992 if (ss
->root
!= &cgroup_dummy_root
) {
998 if (!try_module_get(ss
->module
)) {
1005 /* subsys could be missing if unloaded between parsing and here */
1006 if (added_mask
!= pinned
) {
1011 ret
= cgroup_populate_dir(cgrp
, added_mask
);
1016 * Nothing can fail from this point on. Remove files for the
1017 * removed subsystems and rebind each subsystem.
1019 cgroup_clear_dir(cgrp
, removed_mask
);
1021 for_each_subsys(ss
, i
) {
1022 unsigned long bit
= 1UL << i
;
1024 if (bit
& added_mask
) {
1025 /* We're binding this subsystem to this hierarchy */
1026 BUG_ON(cgroup_css(cgrp
, ss
));
1027 BUG_ON(!cgroup_css(cgroup_dummy_top
, ss
));
1028 BUG_ON(cgroup_css(cgroup_dummy_top
, ss
)->cgroup
!= cgroup_dummy_top
);
1030 rcu_assign_pointer(cgrp
->subsys
[i
],
1031 cgroup_css(cgroup_dummy_top
, ss
));
1032 cgroup_css(cgrp
, ss
)->cgroup
= cgrp
;
1034 list_move(&ss
->sibling
, &root
->subsys_list
);
1037 ss
->bind(cgroup_css(cgrp
, ss
));
1039 /* refcount was already taken, and we're keeping it */
1040 root
->subsys_mask
|= bit
;
1041 } else if (bit
& removed_mask
) {
1042 /* We're removing this subsystem */
1043 BUG_ON(cgroup_css(cgrp
, ss
) != cgroup_css(cgroup_dummy_top
, ss
));
1044 BUG_ON(cgroup_css(cgrp
, ss
)->cgroup
!= cgrp
);
1047 ss
->bind(cgroup_css(cgroup_dummy_top
, ss
));
1049 cgroup_css(cgroup_dummy_top
, ss
)->cgroup
= cgroup_dummy_top
;
1050 RCU_INIT_POINTER(cgrp
->subsys
[i
], NULL
);
1052 cgroup_subsys
[i
]->root
= &cgroup_dummy_root
;
1053 list_move(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
1055 /* subsystem is now free - drop reference on module */
1056 module_put(ss
->module
);
1057 root
->subsys_mask
&= ~bit
;
1062 * Mark @root has finished binding subsystems. @root->subsys_mask
1063 * now matches the bound subsystems.
1065 root
->flags
|= CGRP_ROOT_SUBSYS_BOUND
;
1070 for_each_subsys(ss
, i
)
1071 if (pinned
& (1 << i
))
1072 module_put(ss
->module
);
1076 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1078 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1079 struct cgroup_subsys
*ss
;
1081 mutex_lock(&cgroup_root_mutex
);
1082 for_each_root_subsys(root
, ss
)
1083 seq_printf(seq
, ",%s", ss
->name
);
1084 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
)
1085 seq_puts(seq
, ",sane_behavior");
1086 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1087 seq_puts(seq
, ",noprefix");
1088 if (root
->flags
& CGRP_ROOT_XATTR
)
1089 seq_puts(seq
, ",xattr");
1090 if (strlen(root
->release_agent_path
))
1091 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1092 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
))
1093 seq_puts(seq
, ",clone_children");
1094 if (strlen(root
->name
))
1095 seq_printf(seq
, ",name=%s", root
->name
);
1096 mutex_unlock(&cgroup_root_mutex
);
1100 struct cgroup_sb_opts
{
1101 unsigned long subsys_mask
;
1102 unsigned long flags
;
1103 char *release_agent
;
1104 bool cpuset_clone_children
;
1106 /* User explicitly requested empty subsystem */
1109 struct cgroupfs_root
*new_root
;
1114 * Convert a hierarchy specifier into a bitmask of subsystems and
1115 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1116 * array. This function takes refcounts on subsystems to be used, unless it
1117 * returns error, in which case no refcounts are taken.
1119 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1121 char *token
, *o
= data
;
1122 bool all_ss
= false, one_ss
= false;
1123 unsigned long mask
= (unsigned long)-1;
1124 struct cgroup_subsys
*ss
;
1127 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1129 #ifdef CONFIG_CPUSETS
1130 mask
= ~(1UL << cpuset_subsys_id
);
1133 memset(opts
, 0, sizeof(*opts
));
1135 while ((token
= strsep(&o
, ",")) != NULL
) {
1138 if (!strcmp(token
, "none")) {
1139 /* Explicitly have no subsystems */
1143 if (!strcmp(token
, "all")) {
1144 /* Mutually exclusive option 'all' + subsystem name */
1150 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1151 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1154 if (!strcmp(token
, "noprefix")) {
1155 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1158 if (!strcmp(token
, "clone_children")) {
1159 opts
->cpuset_clone_children
= true;
1162 if (!strcmp(token
, "xattr")) {
1163 opts
->flags
|= CGRP_ROOT_XATTR
;
1166 if (!strncmp(token
, "release_agent=", 14)) {
1167 /* Specifying two release agents is forbidden */
1168 if (opts
->release_agent
)
1170 opts
->release_agent
=
1171 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1172 if (!opts
->release_agent
)
1176 if (!strncmp(token
, "name=", 5)) {
1177 const char *name
= token
+ 5;
1178 /* Can't specify an empty name */
1181 /* Must match [\w.-]+ */
1182 for (i
= 0; i
< strlen(name
); i
++) {
1186 if ((c
== '.') || (c
== '-') || (c
== '_'))
1190 /* Specifying two names is forbidden */
1193 opts
->name
= kstrndup(name
,
1194 MAX_CGROUP_ROOT_NAMELEN
- 1,
1202 for_each_subsys(ss
, i
) {
1203 if (strcmp(token
, ss
->name
))
1208 /* Mutually exclusive option 'all' + subsystem name */
1211 set_bit(i
, &opts
->subsys_mask
);
1216 if (i
== CGROUP_SUBSYS_COUNT
)
1221 * If the 'all' option was specified select all the subsystems,
1222 * otherwise if 'none', 'name=' and a subsystem name options
1223 * were not specified, let's default to 'all'
1225 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1226 for_each_subsys(ss
, i
)
1228 set_bit(i
, &opts
->subsys_mask
);
1230 /* Consistency checks */
1232 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1233 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1235 if (opts
->flags
& CGRP_ROOT_NOPREFIX
) {
1236 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1240 if (opts
->cpuset_clone_children
) {
1241 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1247 * Option noprefix was introduced just for backward compatibility
1248 * with the old cpuset, so we allow noprefix only if mounting just
1249 * the cpuset subsystem.
1251 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1255 /* Can't specify "none" and some subsystems */
1256 if (opts
->subsys_mask
&& opts
->none
)
1260 * We either have to specify by name or by subsystems. (So all
1261 * empty hierarchies must have a name).
1263 if (!opts
->subsys_mask
&& !opts
->name
)
1269 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1272 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1273 struct cgroup
*cgrp
= &root
->top_cgroup
;
1274 struct cgroup_sb_opts opts
;
1275 unsigned long added_mask
, removed_mask
;
1277 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1278 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1282 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1283 mutex_lock(&cgroup_mutex
);
1284 mutex_lock(&cgroup_root_mutex
);
1286 /* See what subsystems are wanted */
1287 ret
= parse_cgroupfs_options(data
, &opts
);
1291 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1292 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1293 task_tgid_nr(current
), current
->comm
);
1295 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1296 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1298 /* Don't allow flags or name to change at remount */
1299 if (((opts
.flags
^ root
->flags
) & CGRP_ROOT_OPTION_MASK
) ||
1300 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1301 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1302 opts
.flags
& CGRP_ROOT_OPTION_MASK
, opts
.name
?: "",
1303 root
->flags
& CGRP_ROOT_OPTION_MASK
, root
->name
);
1308 /* remounting is not allowed for populated hierarchies */
1309 if (root
->number_of_cgroups
> 1) {
1314 ret
= rebind_subsystems(root
, added_mask
, removed_mask
);
1318 if (opts
.release_agent
)
1319 strcpy(root
->release_agent_path
, opts
.release_agent
);
1321 kfree(opts
.release_agent
);
1323 mutex_unlock(&cgroup_root_mutex
);
1324 mutex_unlock(&cgroup_mutex
);
1325 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1329 static const struct super_operations cgroup_ops
= {
1330 .statfs
= simple_statfs
,
1331 .drop_inode
= generic_delete_inode
,
1332 .show_options
= cgroup_show_options
,
1333 .remount_fs
= cgroup_remount
,
1336 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1338 INIT_LIST_HEAD(&cgrp
->sibling
);
1339 INIT_LIST_HEAD(&cgrp
->children
);
1340 INIT_LIST_HEAD(&cgrp
->files
);
1341 INIT_LIST_HEAD(&cgrp
->cset_links
);
1342 INIT_LIST_HEAD(&cgrp
->release_list
);
1343 INIT_LIST_HEAD(&cgrp
->pidlists
);
1344 mutex_init(&cgrp
->pidlist_mutex
);
1345 cgrp
->dummy_css
.cgroup
= cgrp
;
1346 INIT_LIST_HEAD(&cgrp
->event_list
);
1347 spin_lock_init(&cgrp
->event_list_lock
);
1348 simple_xattrs_init(&cgrp
->xattrs
);
1351 static void init_cgroup_root(struct cgroupfs_root
*root
)
1353 struct cgroup
*cgrp
= &root
->top_cgroup
;
1355 INIT_LIST_HEAD(&root
->subsys_list
);
1356 INIT_LIST_HEAD(&root
->root_list
);
1357 root
->number_of_cgroups
= 1;
1359 RCU_INIT_POINTER(cgrp
->name
, &root_cgroup_name
);
1360 init_cgroup_housekeeping(cgrp
);
1361 idr_init(&root
->cgroup_idr
);
1364 static int cgroup_init_root_id(struct cgroupfs_root
*root
, int start
, int end
)
1368 lockdep_assert_held(&cgroup_mutex
);
1369 lockdep_assert_held(&cgroup_root_mutex
);
1371 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, start
, end
,
1376 root
->hierarchy_id
= id
;
1380 static void cgroup_exit_root_id(struct cgroupfs_root
*root
)
1382 lockdep_assert_held(&cgroup_mutex
);
1383 lockdep_assert_held(&cgroup_root_mutex
);
1385 if (root
->hierarchy_id
) {
1386 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1387 root
->hierarchy_id
= 0;
1391 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1393 struct cgroup_sb_opts
*opts
= data
;
1394 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1396 /* If we asked for a name then it must match */
1397 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1401 * If we asked for subsystems (or explicitly for no
1402 * subsystems) then they must match
1404 if ((opts
->subsys_mask
|| opts
->none
)
1405 && (opts
->subsys_mask
!= root
->subsys_mask
))
1411 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1413 struct cgroupfs_root
*root
;
1415 if (!opts
->subsys_mask
&& !opts
->none
)
1418 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1420 return ERR_PTR(-ENOMEM
);
1422 init_cgroup_root(root
);
1425 * We need to set @root->subsys_mask now so that @root can be
1426 * matched by cgroup_test_super() before it finishes
1427 * initialization; otherwise, competing mounts with the same
1428 * options may try to bind the same subsystems instead of waiting
1429 * for the first one leading to unexpected mount errors.
1430 * SUBSYS_BOUND will be set once actual binding is complete.
1432 root
->subsys_mask
= opts
->subsys_mask
;
1433 root
->flags
= opts
->flags
;
1434 if (opts
->release_agent
)
1435 strcpy(root
->release_agent_path
, opts
->release_agent
);
1437 strcpy(root
->name
, opts
->name
);
1438 if (opts
->cpuset_clone_children
)
1439 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1443 static void cgroup_free_root(struct cgroupfs_root
*root
)
1446 /* hierarhcy ID shoulid already have been released */
1447 WARN_ON_ONCE(root
->hierarchy_id
);
1449 idr_destroy(&root
->cgroup_idr
);
1454 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1457 struct cgroup_sb_opts
*opts
= data
;
1459 /* If we don't have a new root, we can't set up a new sb */
1460 if (!opts
->new_root
)
1463 BUG_ON(!opts
->subsys_mask
&& !opts
->none
);
1465 ret
= set_anon_super(sb
, NULL
);
1469 sb
->s_fs_info
= opts
->new_root
;
1470 opts
->new_root
->sb
= sb
;
1472 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1473 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1474 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1475 sb
->s_op
= &cgroup_ops
;
1480 static int cgroup_get_rootdir(struct super_block
*sb
)
1482 static const struct dentry_operations cgroup_dops
= {
1483 .d_iput
= cgroup_diput
,
1484 .d_delete
= always_delete_dentry
,
1487 struct inode
*inode
=
1488 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1493 inode
->i_fop
= &simple_dir_operations
;
1494 inode
->i_op
= &cgroup_dir_inode_operations
;
1495 /* directories start off with i_nlink == 2 (for "." entry) */
1497 sb
->s_root
= d_make_root(inode
);
1500 /* for everything else we want ->d_op set */
1501 sb
->s_d_op
= &cgroup_dops
;
1505 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1506 int flags
, const char *unused_dev_name
,
1509 struct cgroup_sb_opts opts
;
1510 struct cgroupfs_root
*root
;
1512 struct super_block
*sb
;
1513 struct cgroupfs_root
*new_root
;
1514 struct list_head tmp_links
;
1515 struct inode
*inode
;
1516 const struct cred
*cred
;
1518 /* First find the desired set of subsystems */
1519 mutex_lock(&cgroup_mutex
);
1520 ret
= parse_cgroupfs_options(data
, &opts
);
1521 mutex_unlock(&cgroup_mutex
);
1526 * Allocate a new cgroup root. We may not need it if we're
1527 * reusing an existing hierarchy.
1529 new_root
= cgroup_root_from_opts(&opts
);
1530 if (IS_ERR(new_root
)) {
1531 ret
= PTR_ERR(new_root
);
1534 opts
.new_root
= new_root
;
1536 /* Locate an existing or new sb for this hierarchy */
1537 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, 0, &opts
);
1540 cgroup_free_root(opts
.new_root
);
1544 root
= sb
->s_fs_info
;
1546 if (root
== opts
.new_root
) {
1547 /* We used the new root structure, so this is a new hierarchy */
1548 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1549 struct cgroupfs_root
*existing_root
;
1551 struct css_set
*cset
;
1553 BUG_ON(sb
->s_root
!= NULL
);
1555 ret
= cgroup_get_rootdir(sb
);
1557 goto drop_new_super
;
1558 inode
= sb
->s_root
->d_inode
;
1560 mutex_lock(&inode
->i_mutex
);
1561 mutex_lock(&cgroup_mutex
);
1562 mutex_lock(&cgroup_root_mutex
);
1564 root_cgrp
->id
= idr_alloc(&root
->cgroup_idr
, root_cgrp
,
1566 if (root_cgrp
->id
< 0)
1569 /* Check for name clashes with existing mounts */
1571 if (strlen(root
->name
))
1572 for_each_active_root(existing_root
)
1573 if (!strcmp(existing_root
->name
, root
->name
))
1577 * We're accessing css_set_count without locking
1578 * css_set_lock here, but that's OK - it can only be
1579 * increased by someone holding cgroup_lock, and
1580 * that's us. The worst that can happen is that we
1581 * have some link structures left over
1583 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1587 /* ID 0 is reserved for dummy root, 1 for unified hierarchy */
1588 ret
= cgroup_init_root_id(root
, 2, 0);
1592 sb
->s_root
->d_fsdata
= root_cgrp
;
1593 root_cgrp
->dentry
= sb
->s_root
;
1596 * We're inside get_sb() and will call lookup_one_len() to
1597 * create the root files, which doesn't work if SELinux is
1598 * in use. The following cred dancing somehow works around
1599 * it. See 2ce9738ba ("cgroupfs: use init_cred when
1600 * populating new cgroupfs mount") for more details.
1602 cred
= override_creds(&init_cred
);
1604 ret
= cgroup_addrm_files(root_cgrp
, cgroup_base_files
, true);
1608 ret
= rebind_subsystems(root
, root
->subsys_mask
, 0);
1615 * There must be no failure case after here, since rebinding
1616 * takes care of subsystems' refcounts, which are explicitly
1617 * dropped in the failure exit path.
1620 list_add(&root
->root_list
, &cgroup_roots
);
1621 cgroup_root_count
++;
1623 /* Link the top cgroup in this hierarchy into all
1624 * the css_set objects */
1625 write_lock(&css_set_lock
);
1626 hash_for_each(css_set_table
, i
, cset
, hlist
)
1627 link_css_set(&tmp_links
, cset
, root_cgrp
);
1628 write_unlock(&css_set_lock
);
1630 free_cgrp_cset_links(&tmp_links
);
1632 BUG_ON(!list_empty(&root_cgrp
->children
));
1633 BUG_ON(root
->number_of_cgroups
!= 1);
1635 mutex_unlock(&cgroup_root_mutex
);
1636 mutex_unlock(&cgroup_mutex
);
1637 mutex_unlock(&inode
->i_mutex
);
1640 * We re-used an existing hierarchy - the new root (if
1641 * any) is not needed
1643 cgroup_free_root(opts
.new_root
);
1645 if ((root
->flags
^ opts
.flags
) & CGRP_ROOT_OPTION_MASK
) {
1646 if ((root
->flags
| opts
.flags
) & CGRP_ROOT_SANE_BEHAVIOR
) {
1647 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1649 goto drop_new_super
;
1651 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1656 kfree(opts
.release_agent
);
1658 return dget(sb
->s_root
);
1661 free_cgrp_cset_links(&tmp_links
);
1662 cgroup_addrm_files(&root
->top_cgroup
, cgroup_base_files
, false);
1665 cgroup_exit_root_id(root
);
1666 mutex_unlock(&cgroup_root_mutex
);
1667 mutex_unlock(&cgroup_mutex
);
1668 mutex_unlock(&inode
->i_mutex
);
1670 deactivate_locked_super(sb
);
1672 kfree(opts
.release_agent
);
1674 return ERR_PTR(ret
);
1677 static void cgroup_kill_sb(struct super_block
*sb
) {
1678 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1679 struct cgroup
*cgrp
= &root
->top_cgroup
;
1680 struct cgrp_cset_link
*link
, *tmp_link
;
1685 BUG_ON(root
->number_of_cgroups
!= 1);
1686 BUG_ON(!list_empty(&cgrp
->children
));
1688 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1689 mutex_lock(&cgroup_mutex
);
1690 mutex_lock(&cgroup_root_mutex
);
1692 /* Rebind all subsystems back to the default hierarchy */
1693 if (root
->flags
& CGRP_ROOT_SUBSYS_BOUND
) {
1694 ret
= rebind_subsystems(root
, 0, root
->subsys_mask
);
1695 /* Shouldn't be able to fail ... */
1700 * Release all the links from cset_links to this hierarchy's
1703 write_lock(&css_set_lock
);
1705 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1706 list_del(&link
->cset_link
);
1707 list_del(&link
->cgrp_link
);
1710 write_unlock(&css_set_lock
);
1712 if (!list_empty(&root
->root_list
)) {
1713 list_del(&root
->root_list
);
1714 cgroup_root_count
--;
1717 cgroup_exit_root_id(root
);
1719 mutex_unlock(&cgroup_root_mutex
);
1720 mutex_unlock(&cgroup_mutex
);
1721 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1723 simple_xattrs_free(&cgrp
->xattrs
);
1725 kill_litter_super(sb
);
1726 cgroup_free_root(root
);
1729 static struct file_system_type cgroup_fs_type
= {
1731 .mount
= cgroup_mount
,
1732 .kill_sb
= cgroup_kill_sb
,
1735 static struct kobject
*cgroup_kobj
;
1738 * cgroup_path - generate the path of a cgroup
1739 * @cgrp: the cgroup in question
1740 * @buf: the buffer to write the path into
1741 * @buflen: the length of the buffer
1743 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1745 * We can't generate cgroup path using dentry->d_name, as accessing
1746 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1747 * inode's i_mutex, while on the other hand cgroup_path() can be called
1748 * with some irq-safe spinlocks held.
1750 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1752 int ret
= -ENAMETOOLONG
;
1755 if (!cgrp
->parent
) {
1756 if (strlcpy(buf
, "/", buflen
) >= buflen
)
1757 return -ENAMETOOLONG
;
1761 start
= buf
+ buflen
- 1;
1766 const char *name
= cgroup_name(cgrp
);
1770 if ((start
-= len
) < buf
)
1772 memcpy(start
, name
, len
);
1778 cgrp
= cgrp
->parent
;
1779 } while (cgrp
->parent
);
1781 memmove(buf
, start
, buf
+ buflen
- start
);
1786 EXPORT_SYMBOL_GPL(cgroup_path
);
1789 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1790 * @task: target task
1791 * @buf: the buffer to write the path into
1792 * @buflen: the length of the buffer
1794 * Determine @task's cgroup on the first (the one with the lowest non-zero
1795 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1796 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1797 * cgroup controller callbacks.
1799 * Returns 0 on success, fails with -%ENAMETOOLONG if @buflen is too short.
1801 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
1803 struct cgroupfs_root
*root
;
1804 struct cgroup
*cgrp
;
1805 int hierarchy_id
= 1, ret
= 0;
1808 return -ENAMETOOLONG
;
1810 mutex_lock(&cgroup_mutex
);
1812 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
1815 cgrp
= task_cgroup_from_root(task
, root
);
1816 ret
= cgroup_path(cgrp
, buf
, buflen
);
1818 /* if no hierarchy exists, everyone is in "/" */
1819 memcpy(buf
, "/", 2);
1822 mutex_unlock(&cgroup_mutex
);
1825 EXPORT_SYMBOL_GPL(task_cgroup_path
);
1828 * Control Group taskset
1830 struct task_and_cgroup
{
1831 struct task_struct
*task
;
1832 struct cgroup
*cgrp
;
1833 struct css_set
*cset
;
1836 struct cgroup_taskset
{
1837 struct task_and_cgroup single
;
1838 struct flex_array
*tc_array
;
1841 struct cgroup
*cur_cgrp
;
1845 * cgroup_taskset_first - reset taskset and return the first task
1846 * @tset: taskset of interest
1848 * @tset iteration is initialized and the first task is returned.
1850 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1852 if (tset
->tc_array
) {
1854 return cgroup_taskset_next(tset
);
1856 tset
->cur_cgrp
= tset
->single
.cgrp
;
1857 return tset
->single
.task
;
1860 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1863 * cgroup_taskset_next - iterate to the next task in taskset
1864 * @tset: taskset of interest
1866 * Return the next task in @tset. Iteration must have been initialized
1867 * with cgroup_taskset_first().
1869 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1871 struct task_and_cgroup
*tc
;
1873 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1876 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1877 tset
->cur_cgrp
= tc
->cgrp
;
1880 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1883 * cgroup_taskset_cur_css - return the matching css for the current task
1884 * @tset: taskset of interest
1885 * @subsys_id: the ID of the target subsystem
1887 * Return the css for the current (last returned) task of @tset for
1888 * subsystem specified by @subsys_id. This function must be preceded by
1889 * either cgroup_taskset_first() or cgroup_taskset_next().
1891 struct cgroup_subsys_state
*cgroup_taskset_cur_css(struct cgroup_taskset
*tset
,
1894 return cgroup_css(tset
->cur_cgrp
, cgroup_subsys
[subsys_id
]);
1896 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css
);
1899 * cgroup_taskset_size - return the number of tasks in taskset
1900 * @tset: taskset of interest
1902 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1904 return tset
->tc_array
? tset
->tc_array_len
: 1;
1906 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1910 * cgroup_task_migrate - move a task from one cgroup to another.
1912 * Must be called with cgroup_mutex and threadgroup locked.
1914 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
1915 struct task_struct
*tsk
,
1916 struct css_set
*new_cset
)
1918 struct css_set
*old_cset
;
1921 * We are synchronized through threadgroup_lock() against PF_EXITING
1922 * setting such that we can't race against cgroup_exit() changing the
1923 * css_set to init_css_set and dropping the old one.
1925 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1926 old_cset
= task_css_set(tsk
);
1929 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
1932 /* Update the css_set linked lists if we're using them */
1933 write_lock(&css_set_lock
);
1934 if (!list_empty(&tsk
->cg_list
))
1935 list_move(&tsk
->cg_list
, &new_cset
->tasks
);
1936 write_unlock(&css_set_lock
);
1939 * We just gained a reference on old_cset by taking it from the
1940 * task. As trading it for new_cset is protected by cgroup_mutex,
1941 * we're safe to drop it here; it will be freed under RCU.
1943 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
1944 put_css_set(old_cset
);
1948 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1949 * @cgrp: the cgroup to attach to
1950 * @tsk: the task or the leader of the threadgroup to be attached
1951 * @threadgroup: attach the whole threadgroup?
1953 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1954 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1956 static int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
,
1959 int retval
, i
, group_size
;
1960 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1961 struct cgroupfs_root
*root
= cgrp
->root
;
1962 /* threadgroup list cursor and array */
1963 struct task_struct
*leader
= tsk
;
1964 struct task_and_cgroup
*tc
;
1965 struct flex_array
*group
;
1966 struct cgroup_taskset tset
= { };
1969 * step 0: in order to do expensive, possibly blocking operations for
1970 * every thread, we cannot iterate the thread group list, since it needs
1971 * rcu or tasklist locked. instead, build an array of all threads in the
1972 * group - group_rwsem prevents new threads from appearing, and if
1973 * threads exit, this will just be an over-estimate.
1976 group_size
= get_nr_threads(tsk
);
1979 /* flex_array supports very large thread-groups better than kmalloc. */
1980 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
1983 /* pre-allocate to guarantee space while iterating in rcu read-side. */
1984 retval
= flex_array_prealloc(group
, 0, group_size
, GFP_KERNEL
);
1986 goto out_free_group_list
;
1990 * Prevent freeing of tasks while we take a snapshot. Tasks that are
1991 * already PF_EXITING could be freed from underneath us unless we
1992 * take an rcu_read_lock.
1996 struct task_and_cgroup ent
;
1998 /* @tsk either already exited or can't exit until the end */
1999 if (tsk
->flags
& PF_EXITING
)
2002 /* as per above, nr_threads may decrease, but not increase. */
2003 BUG_ON(i
>= group_size
);
2005 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2006 /* nothing to do if this task is already in the cgroup */
2007 if (ent
.cgrp
== cgrp
)
2010 * saying GFP_ATOMIC has no effect here because we did prealloc
2011 * earlier, but it's good form to communicate our expectations.
2013 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2014 BUG_ON(retval
!= 0);
2019 } while_each_thread(leader
, tsk
);
2021 /* remember the number of threads in the array for later. */
2023 tset
.tc_array
= group
;
2024 tset
.tc_array_len
= group_size
;
2026 /* methods shouldn't be called if no task is actually migrating */
2029 goto out_free_group_list
;
2032 * step 1: check that we can legitimately attach to the cgroup.
2034 for_each_root_subsys(root
, ss
) {
2035 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, ss
);
2037 if (ss
->can_attach
) {
2038 retval
= ss
->can_attach(css
, &tset
);
2041 goto out_cancel_attach
;
2047 * step 2: make sure css_sets exist for all threads to be migrated.
2048 * we use find_css_set, which allocates a new one if necessary.
2050 for (i
= 0; i
< group_size
; i
++) {
2051 struct css_set
*old_cset
;
2053 tc
= flex_array_get(group
, i
);
2054 old_cset
= task_css_set(tc
->task
);
2055 tc
->cset
= find_css_set(old_cset
, cgrp
);
2058 goto out_put_css_set_refs
;
2063 * step 3: now that we're guaranteed success wrt the css_sets,
2064 * proceed to move all tasks to the new cgroup. There are no
2065 * failure cases after here, so this is the commit point.
2067 for (i
= 0; i
< group_size
; i
++) {
2068 tc
= flex_array_get(group
, i
);
2069 cgroup_task_migrate(tc
->cgrp
, tc
->task
, tc
->cset
);
2071 /* nothing is sensitive to fork() after this point. */
2074 * step 4: do subsystem attach callbacks.
2076 for_each_root_subsys(root
, ss
) {
2077 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, ss
);
2080 ss
->attach(css
, &tset
);
2084 * step 5: success! and cleanup
2087 out_put_css_set_refs
:
2089 for (i
= 0; i
< group_size
; i
++) {
2090 tc
= flex_array_get(group
, i
);
2093 put_css_set(tc
->cset
);
2098 for_each_root_subsys(root
, ss
) {
2099 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, ss
);
2101 if (ss
== failed_ss
)
2103 if (ss
->cancel_attach
)
2104 ss
->cancel_attach(css
, &tset
);
2107 out_free_group_list
:
2108 flex_array_free(group
);
2113 * Find the task_struct of the task to attach by vpid and pass it along to the
2114 * function to attach either it or all tasks in its threadgroup. Will lock
2115 * cgroup_mutex and threadgroup; may take task_lock of task.
2117 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2119 struct task_struct
*tsk
;
2120 const struct cred
*cred
= current_cred(), *tcred
;
2123 if (!cgroup_lock_live_group(cgrp
))
2129 tsk
= find_task_by_vpid(pid
);
2133 goto out_unlock_cgroup
;
2136 * even if we're attaching all tasks in the thread group, we
2137 * only need to check permissions on one of them.
2139 tcred
= __task_cred(tsk
);
2140 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2141 !uid_eq(cred
->euid
, tcred
->uid
) &&
2142 !uid_eq(cred
->euid
, tcred
->suid
)) {
2145 goto out_unlock_cgroup
;
2151 tsk
= tsk
->group_leader
;
2154 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2155 * trapped in a cpuset, or RT worker may be born in a cgroup
2156 * with no rt_runtime allocated. Just say no.
2158 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2161 goto out_unlock_cgroup
;
2164 get_task_struct(tsk
);
2167 threadgroup_lock(tsk
);
2169 if (!thread_group_leader(tsk
)) {
2171 * a race with de_thread from another thread's exec()
2172 * may strip us of our leadership, if this happens,
2173 * there is no choice but to throw this task away and
2174 * try again; this is
2175 * "double-double-toil-and-trouble-check locking".
2177 threadgroup_unlock(tsk
);
2178 put_task_struct(tsk
);
2179 goto retry_find_task
;
2183 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2185 threadgroup_unlock(tsk
);
2187 put_task_struct(tsk
);
2189 mutex_unlock(&cgroup_mutex
);
2194 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2195 * @from: attach to all cgroups of a given task
2196 * @tsk: the task to be attached
2198 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2200 struct cgroupfs_root
*root
;
2203 mutex_lock(&cgroup_mutex
);
2204 for_each_active_root(root
) {
2205 struct cgroup
*from_cgrp
= task_cgroup_from_root(from
, root
);
2207 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2211 mutex_unlock(&cgroup_mutex
);
2215 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2217 static int cgroup_tasks_write(struct cgroup_subsys_state
*css
,
2218 struct cftype
*cft
, u64 pid
)
2220 return attach_task_by_pid(css
->cgroup
, pid
, false);
2223 static int cgroup_procs_write(struct cgroup_subsys_state
*css
,
2224 struct cftype
*cft
, u64 tgid
)
2226 return attach_task_by_pid(css
->cgroup
, tgid
, true);
2229 static int cgroup_release_agent_write(struct cgroup_subsys_state
*css
,
2230 struct cftype
*cft
, const char *buffer
)
2232 BUILD_BUG_ON(sizeof(css
->cgroup
->root
->release_agent_path
) < PATH_MAX
);
2233 if (strlen(buffer
) >= PATH_MAX
)
2235 if (!cgroup_lock_live_group(css
->cgroup
))
2237 mutex_lock(&cgroup_root_mutex
);
2238 strcpy(css
->cgroup
->root
->release_agent_path
, buffer
);
2239 mutex_unlock(&cgroup_root_mutex
);
2240 mutex_unlock(&cgroup_mutex
);
2244 static int cgroup_release_agent_show(struct cgroup_subsys_state
*css
,
2245 struct cftype
*cft
, struct seq_file
*seq
)
2247 struct cgroup
*cgrp
= css
->cgroup
;
2249 if (!cgroup_lock_live_group(cgrp
))
2251 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2252 seq_putc(seq
, '\n');
2253 mutex_unlock(&cgroup_mutex
);
2257 static int cgroup_sane_behavior_show(struct cgroup_subsys_state
*css
,
2258 struct cftype
*cft
, struct seq_file
*seq
)
2260 seq_printf(seq
, "%d\n", cgroup_sane_behavior(css
->cgroup
));
2264 /* A buffer size big enough for numbers or short strings */
2265 #define CGROUP_LOCAL_BUFFER_SIZE 64
2267 static ssize_t
cgroup_write_X64(struct cgroup_subsys_state
*css
,
2268 struct cftype
*cft
, struct file
*file
,
2269 const char __user
*userbuf
, size_t nbytes
,
2270 loff_t
*unused_ppos
)
2272 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2278 if (nbytes
>= sizeof(buffer
))
2280 if (copy_from_user(buffer
, userbuf
, nbytes
))
2283 buffer
[nbytes
] = 0; /* nul-terminate */
2284 if (cft
->write_u64
) {
2285 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2288 retval
= cft
->write_u64(css
, cft
, val
);
2290 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2293 retval
= cft
->write_s64(css
, cft
, val
);
2300 static ssize_t
cgroup_write_string(struct cgroup_subsys_state
*css
,
2301 struct cftype
*cft
, struct file
*file
,
2302 const char __user
*userbuf
, size_t nbytes
,
2303 loff_t
*unused_ppos
)
2305 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2307 size_t max_bytes
= cft
->max_write_len
;
2308 char *buffer
= local_buffer
;
2311 max_bytes
= sizeof(local_buffer
) - 1;
2312 if (nbytes
>= max_bytes
)
2314 /* Allocate a dynamic buffer if we need one */
2315 if (nbytes
>= sizeof(local_buffer
)) {
2316 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2320 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2325 buffer
[nbytes
] = 0; /* nul-terminate */
2326 retval
= cft
->write_string(css
, cft
, strstrip(buffer
));
2330 if (buffer
!= local_buffer
)
2335 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2336 size_t nbytes
, loff_t
*ppos
)
2338 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2339 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2340 struct cgroup_subsys_state
*css
= cfe
->css
;
2343 return cft
->write(css
, cft
, file
, buf
, nbytes
, ppos
);
2344 if (cft
->write_u64
|| cft
->write_s64
)
2345 return cgroup_write_X64(css
, cft
, file
, buf
, nbytes
, ppos
);
2346 if (cft
->write_string
)
2347 return cgroup_write_string(css
, cft
, file
, buf
, nbytes
, ppos
);
2349 int ret
= cft
->trigger(css
, (unsigned int)cft
->private);
2350 return ret
? ret
: nbytes
;
2355 static ssize_t
cgroup_read_u64(struct cgroup_subsys_state
*css
,
2356 struct cftype
*cft
, struct file
*file
,
2357 char __user
*buf
, size_t nbytes
, loff_t
*ppos
)
2359 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2360 u64 val
= cft
->read_u64(css
, cft
);
2361 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2363 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2366 static ssize_t
cgroup_read_s64(struct cgroup_subsys_state
*css
,
2367 struct cftype
*cft
, struct file
*file
,
2368 char __user
*buf
, size_t nbytes
, loff_t
*ppos
)
2370 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2371 s64 val
= cft
->read_s64(css
, cft
);
2372 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2374 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2377 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2378 size_t nbytes
, loff_t
*ppos
)
2380 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2381 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2382 struct cgroup_subsys_state
*css
= cfe
->css
;
2385 return cft
->read(css
, cft
, file
, buf
, nbytes
, ppos
);
2387 return cgroup_read_u64(css
, cft
, file
, buf
, nbytes
, ppos
);
2389 return cgroup_read_s64(css
, cft
, file
, buf
, nbytes
, ppos
);
2394 * seqfile ops/methods for returning structured data. Currently just
2395 * supports string->u64 maps, but can be extended in future.
2398 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2400 struct seq_file
*sf
= cb
->state
;
2401 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2404 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2406 struct cfent
*cfe
= m
->private;
2407 struct cftype
*cft
= cfe
->type
;
2408 struct cgroup_subsys_state
*css
= cfe
->css
;
2410 if (cft
->read_map
) {
2411 struct cgroup_map_cb cb
= {
2412 .fill
= cgroup_map_add
,
2415 return cft
->read_map(css
, cft
, &cb
);
2417 return cft
->read_seq_string(css
, cft
, m
);
2420 static const struct file_operations cgroup_seqfile_operations
= {
2422 .write
= cgroup_file_write
,
2423 .llseek
= seq_lseek
,
2424 .release
= single_release
,
2427 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2429 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2430 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2431 struct cgroup
*cgrp
= __d_cgrp(cfe
->dentry
->d_parent
);
2432 struct cgroup_subsys_state
*css
;
2435 err
= generic_file_open(inode
, file
);
2440 * If the file belongs to a subsystem, pin the css. Will be
2441 * unpinned either on open failure or release. This ensures that
2442 * @css stays alive for all file operations.
2445 css
= cgroup_css(cgrp
, cft
->ss
);
2446 if (cft
->ss
&& !css_tryget(css
))
2454 * @cfe->css is used by read/write/close to determine the
2455 * associated css. @file->private_data would be a better place but
2456 * that's already used by seqfile. Multiple accessors may use it
2457 * simultaneously which is okay as the association never changes.
2459 WARN_ON_ONCE(cfe
->css
&& cfe
->css
!= css
);
2462 if (cft
->read_map
|| cft
->read_seq_string
) {
2463 file
->f_op
= &cgroup_seqfile_operations
;
2464 err
= single_open(file
, cgroup_seqfile_show
, cfe
);
2465 } else if (cft
->open
) {
2466 err
= cft
->open(inode
, file
);
2474 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2476 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2477 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2478 struct cgroup_subsys_state
*css
= cfe
->css
;
2482 ret
= cft
->release(inode
, file
);
2489 * cgroup_rename - Only allow simple rename of directories in place.
2491 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2492 struct inode
*new_dir
, struct dentry
*new_dentry
)
2495 struct cgroup_name
*name
, *old_name
;
2496 struct cgroup
*cgrp
;
2499 * It's convinient to use parent dir's i_mutex to protected
2502 lockdep_assert_held(&old_dir
->i_mutex
);
2504 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2506 if (new_dentry
->d_inode
)
2508 if (old_dir
!= new_dir
)
2511 cgrp
= __d_cgrp(old_dentry
);
2514 * This isn't a proper migration and its usefulness is very
2515 * limited. Disallow if sane_behavior.
2517 if (cgroup_sane_behavior(cgrp
))
2520 name
= cgroup_alloc_name(new_dentry
);
2524 ret
= simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2530 old_name
= rcu_dereference_protected(cgrp
->name
, true);
2531 rcu_assign_pointer(cgrp
->name
, name
);
2533 kfree_rcu(old_name
, rcu_head
);
2537 static struct simple_xattrs
*__d_xattrs(struct dentry
*dentry
)
2539 if (S_ISDIR(dentry
->d_inode
->i_mode
))
2540 return &__d_cgrp(dentry
)->xattrs
;
2542 return &__d_cfe(dentry
)->xattrs
;
2545 static inline int xattr_enabled(struct dentry
*dentry
)
2547 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
2548 return root
->flags
& CGRP_ROOT_XATTR
;
2551 static bool is_valid_xattr(const char *name
)
2553 if (!strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) ||
2554 !strncmp(name
, XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
))
2559 static int cgroup_setxattr(struct dentry
*dentry
, const char *name
,
2560 const void *val
, size_t size
, int flags
)
2562 if (!xattr_enabled(dentry
))
2564 if (!is_valid_xattr(name
))
2566 return simple_xattr_set(__d_xattrs(dentry
), name
, val
, size
, flags
);
2569 static int cgroup_removexattr(struct dentry
*dentry
, const char *name
)
2571 if (!xattr_enabled(dentry
))
2573 if (!is_valid_xattr(name
))
2575 return simple_xattr_remove(__d_xattrs(dentry
), name
);
2578 static ssize_t
cgroup_getxattr(struct dentry
*dentry
, const char *name
,
2579 void *buf
, size_t size
)
2581 if (!xattr_enabled(dentry
))
2583 if (!is_valid_xattr(name
))
2585 return simple_xattr_get(__d_xattrs(dentry
), name
, buf
, size
);
2588 static ssize_t
cgroup_listxattr(struct dentry
*dentry
, char *buf
, size_t size
)
2590 if (!xattr_enabled(dentry
))
2592 return simple_xattr_list(__d_xattrs(dentry
), buf
, size
);
2595 static const struct file_operations cgroup_file_operations
= {
2596 .read
= cgroup_file_read
,
2597 .write
= cgroup_file_write
,
2598 .llseek
= generic_file_llseek
,
2599 .open
= cgroup_file_open
,
2600 .release
= cgroup_file_release
,
2603 static const struct inode_operations cgroup_file_inode_operations
= {
2604 .setxattr
= cgroup_setxattr
,
2605 .getxattr
= cgroup_getxattr
,
2606 .listxattr
= cgroup_listxattr
,
2607 .removexattr
= cgroup_removexattr
,
2610 static const struct inode_operations cgroup_dir_inode_operations
= {
2611 .lookup
= simple_lookup
,
2612 .mkdir
= cgroup_mkdir
,
2613 .rmdir
= cgroup_rmdir
,
2614 .rename
= cgroup_rename
,
2615 .setxattr
= cgroup_setxattr
,
2616 .getxattr
= cgroup_getxattr
,
2617 .listxattr
= cgroup_listxattr
,
2618 .removexattr
= cgroup_removexattr
,
2622 * Check if a file is a control file
2624 static inline struct cftype
*__file_cft(struct file
*file
)
2626 if (file_inode(file
)->i_fop
!= &cgroup_file_operations
)
2627 return ERR_PTR(-EINVAL
);
2628 return __d_cft(file
->f_dentry
);
2631 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2632 struct super_block
*sb
)
2634 struct inode
*inode
;
2638 if (dentry
->d_inode
)
2641 inode
= cgroup_new_inode(mode
, sb
);
2645 if (S_ISDIR(mode
)) {
2646 inode
->i_op
= &cgroup_dir_inode_operations
;
2647 inode
->i_fop
= &simple_dir_operations
;
2649 /* start off with i_nlink == 2 (for "." entry) */
2651 inc_nlink(dentry
->d_parent
->d_inode
);
2654 * Control reaches here with cgroup_mutex held.
2655 * @inode->i_mutex should nest outside cgroup_mutex but we
2656 * want to populate it immediately without releasing
2657 * cgroup_mutex. As @inode isn't visible to anyone else
2658 * yet, trylock will always succeed without affecting
2661 WARN_ON_ONCE(!mutex_trylock(&inode
->i_mutex
));
2662 } else if (S_ISREG(mode
)) {
2664 inode
->i_fop
= &cgroup_file_operations
;
2665 inode
->i_op
= &cgroup_file_inode_operations
;
2667 d_instantiate(dentry
, inode
);
2668 dget(dentry
); /* Extra count - pin the dentry in core */
2673 * cgroup_file_mode - deduce file mode of a control file
2674 * @cft: the control file in question
2676 * returns cft->mode if ->mode is not 0
2677 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2678 * returns S_IRUGO if it has only a read handler
2679 * returns S_IWUSR if it has only a write hander
2681 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2688 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2689 cft
->read_map
|| cft
->read_seq_string
)
2692 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2693 cft
->write_string
|| cft
->trigger
)
2699 static int cgroup_add_file(struct cgroup
*cgrp
, struct cftype
*cft
)
2701 struct dentry
*dir
= cgrp
->dentry
;
2702 struct cgroup
*parent
= __d_cgrp(dir
);
2703 struct dentry
*dentry
;
2707 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2709 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
2710 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
2711 strcpy(name
, cft
->ss
->name
);
2714 strcat(name
, cft
->name
);
2716 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2718 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2722 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2723 if (IS_ERR(dentry
)) {
2724 error
= PTR_ERR(dentry
);
2728 cfe
->type
= (void *)cft
;
2729 cfe
->dentry
= dentry
;
2730 dentry
->d_fsdata
= cfe
;
2731 simple_xattrs_init(&cfe
->xattrs
);
2733 mode
= cgroup_file_mode(cft
);
2734 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2736 list_add_tail(&cfe
->node
, &parent
->files
);
2746 * cgroup_addrm_files - add or remove files to a cgroup directory
2747 * @cgrp: the target cgroup
2748 * @cfts: array of cftypes to be added
2749 * @is_add: whether to add or remove
2751 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2752 * For removals, this function never fails. If addition fails, this
2753 * function doesn't remove files already added. The caller is responsible
2756 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
2762 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
2763 lockdep_assert_held(&cgroup_mutex
);
2765 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2766 /* does cft->flags tell us to skip this file on @cgrp? */
2767 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_sane_behavior(cgrp
))
2769 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2771 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2775 ret
= cgroup_add_file(cgrp
, cft
);
2777 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2782 cgroup_rm_file(cgrp
, cft
);
2788 static void cgroup_cfts_prepare(void)
2789 __acquires(&cgroup_mutex
)
2792 * Thanks to the entanglement with vfs inode locking, we can't walk
2793 * the existing cgroups under cgroup_mutex and create files.
2794 * Instead, we use css_for_each_descendant_pre() and drop RCU read
2795 * lock before calling cgroup_addrm_files().
2797 mutex_lock(&cgroup_mutex
);
2800 static int cgroup_cfts_commit(struct cftype
*cfts
, bool is_add
)
2801 __releases(&cgroup_mutex
)
2804 struct cgroup_subsys
*ss
= cfts
[0].ss
;
2805 struct cgroup
*root
= &ss
->root
->top_cgroup
;
2806 struct super_block
*sb
= ss
->root
->sb
;
2807 struct dentry
*prev
= NULL
;
2808 struct inode
*inode
;
2809 struct cgroup_subsys_state
*css
;
2813 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2814 if (!cfts
|| ss
->root
== &cgroup_dummy_root
||
2815 !atomic_inc_not_zero(&sb
->s_active
)) {
2816 mutex_unlock(&cgroup_mutex
);
2821 * All cgroups which are created after we drop cgroup_mutex will
2822 * have the updated set of files, so we only need to update the
2823 * cgroups created before the current @cgroup_serial_nr_next.
2825 update_before
= cgroup_serial_nr_next
;
2827 mutex_unlock(&cgroup_mutex
);
2829 /* add/rm files for all cgroups created before */
2831 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
2832 struct cgroup
*cgrp
= css
->cgroup
;
2834 if (cgroup_is_dead(cgrp
))
2837 inode
= cgrp
->dentry
->d_inode
;
2842 prev
= cgrp
->dentry
;
2844 mutex_lock(&inode
->i_mutex
);
2845 mutex_lock(&cgroup_mutex
);
2846 if (cgrp
->serial_nr
< update_before
&& !cgroup_is_dead(cgrp
))
2847 ret
= cgroup_addrm_files(cgrp
, cfts
, is_add
);
2848 mutex_unlock(&cgroup_mutex
);
2849 mutex_unlock(&inode
->i_mutex
);
2857 deactivate_super(sb
);
2862 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2863 * @ss: target cgroup subsystem
2864 * @cfts: zero-length name terminated array of cftypes
2866 * Register @cfts to @ss. Files described by @cfts are created for all
2867 * existing cgroups to which @ss is attached and all future cgroups will
2868 * have them too. This function can be called anytime whether @ss is
2871 * Returns 0 on successful registration, -errno on failure. Note that this
2872 * function currently returns 0 as long as @cfts registration is successful
2873 * even if some file creation attempts on existing cgroups fail.
2875 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2877 struct cftype_set
*set
;
2881 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2885 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++)
2888 cgroup_cfts_prepare();
2890 list_add_tail(&set
->node
, &ss
->cftsets
);
2891 ret
= cgroup_cfts_commit(cfts
, true);
2893 cgroup_rm_cftypes(cfts
);
2896 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2899 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2900 * @cfts: zero-length name terminated array of cftypes
2902 * Unregister @cfts. Files described by @cfts are removed from all
2903 * existing cgroups and all future cgroups won't have them either. This
2904 * function can be called anytime whether @cfts' subsys is attached or not.
2906 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2909 int cgroup_rm_cftypes(struct cftype
*cfts
)
2911 struct cftype_set
*set
;
2913 if (!cfts
|| !cfts
[0].ss
)
2916 cgroup_cfts_prepare();
2918 list_for_each_entry(set
, &cfts
[0].ss
->cftsets
, node
) {
2919 if (set
->cfts
== cfts
) {
2920 list_del(&set
->node
);
2922 cgroup_cfts_commit(cfts
, false);
2927 cgroup_cfts_commit(NULL
, false);
2932 * cgroup_task_count - count the number of tasks in a cgroup.
2933 * @cgrp: the cgroup in question
2935 * Return the number of tasks in the cgroup.
2937 int cgroup_task_count(const struct cgroup
*cgrp
)
2940 struct cgrp_cset_link
*link
;
2942 read_lock(&css_set_lock
);
2943 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
2944 count
+= atomic_read(&link
->cset
->refcount
);
2945 read_unlock(&css_set_lock
);
2950 * To reduce the fork() overhead for systems that are not actually using
2951 * their cgroups capability, we don't maintain the lists running through
2952 * each css_set to its tasks until we see the list actually used - in other
2953 * words after the first call to css_task_iter_start().
2955 static void cgroup_enable_task_cg_lists(void)
2957 struct task_struct
*p
, *g
;
2958 write_lock(&css_set_lock
);
2959 use_task_css_set_links
= 1;
2961 * We need tasklist_lock because RCU is not safe against
2962 * while_each_thread(). Besides, a forking task that has passed
2963 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2964 * is not guaranteed to have its child immediately visible in the
2965 * tasklist if we walk through it with RCU.
2967 read_lock(&tasklist_lock
);
2968 do_each_thread(g
, p
) {
2971 * We should check if the process is exiting, otherwise
2972 * it will race with cgroup_exit() in that the list
2973 * entry won't be deleted though the process has exited.
2975 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2976 list_add(&p
->cg_list
, &task_css_set(p
)->tasks
);
2978 } while_each_thread(g
, p
);
2979 read_unlock(&tasklist_lock
);
2980 write_unlock(&css_set_lock
);
2984 * css_next_child - find the next child of a given css
2985 * @pos_css: the current position (%NULL to initiate traversal)
2986 * @parent_css: css whose children to walk
2988 * This function returns the next child of @parent_css and should be called
2989 * under RCU read lock. The only requirement is that @parent_css and
2990 * @pos_css are accessible. The next sibling is guaranteed to be returned
2991 * regardless of their states.
2993 struct cgroup_subsys_state
*
2994 css_next_child(struct cgroup_subsys_state
*pos_css
,
2995 struct cgroup_subsys_state
*parent_css
)
2997 struct cgroup
*pos
= pos_css
? pos_css
->cgroup
: NULL
;
2998 struct cgroup
*cgrp
= parent_css
->cgroup
;
2999 struct cgroup
*next
;
3001 WARN_ON_ONCE(!rcu_read_lock_held());
3004 * @pos could already have been removed. Once a cgroup is removed,
3005 * its ->sibling.next is no longer updated when its next sibling
3006 * changes. As CGRP_DEAD assertion is serialized and happens
3007 * before the cgroup is taken off the ->sibling list, if we see it
3008 * unasserted, it's guaranteed that the next sibling hasn't
3009 * finished its grace period even if it's already removed, and thus
3010 * safe to dereference from this RCU critical section. If
3011 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
3012 * to be visible as %true here.
3014 * If @pos is dead, its next pointer can't be dereferenced;
3015 * however, as each cgroup is given a monotonically increasing
3016 * unique serial number and always appended to the sibling list,
3017 * the next one can be found by walking the parent's children until
3018 * we see a cgroup with higher serial number than @pos's. While
3019 * this path can be slower, it's taken only when either the current
3020 * cgroup is removed or iteration and removal race.
3023 next
= list_entry_rcu(cgrp
->children
.next
, struct cgroup
, sibling
);
3024 } else if (likely(!cgroup_is_dead(pos
))) {
3025 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
, sibling
);
3027 list_for_each_entry_rcu(next
, &cgrp
->children
, sibling
)
3028 if (next
->serial_nr
> pos
->serial_nr
)
3032 if (&next
->sibling
== &cgrp
->children
)
3035 return cgroup_css(next
, parent_css
->ss
);
3037 EXPORT_SYMBOL_GPL(css_next_child
);
3040 * css_next_descendant_pre - find the next descendant for pre-order walk
3041 * @pos: the current position (%NULL to initiate traversal)
3042 * @root: css whose descendants to walk
3044 * To be used by css_for_each_descendant_pre(). Find the next descendant
3045 * to visit for pre-order traversal of @root's descendants. @root is
3046 * included in the iteration and the first node to be visited.
3048 * While this function requires RCU read locking, it doesn't require the
3049 * whole traversal to be contained in a single RCU critical section. This
3050 * function will return the correct next descendant as long as both @pos
3051 * and @root are accessible and @pos is a descendant of @root.
3053 struct cgroup_subsys_state
*
3054 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3055 struct cgroup_subsys_state
*root
)
3057 struct cgroup_subsys_state
*next
;
3059 WARN_ON_ONCE(!rcu_read_lock_held());
3061 /* if first iteration, visit @root */
3065 /* visit the first child if exists */
3066 next
= css_next_child(NULL
, pos
);
3070 /* no child, visit my or the closest ancestor's next sibling */
3071 while (pos
!= root
) {
3072 next
= css_next_child(pos
, css_parent(pos
));
3075 pos
= css_parent(pos
);
3080 EXPORT_SYMBOL_GPL(css_next_descendant_pre
);
3083 * css_rightmost_descendant - return the rightmost descendant of a css
3084 * @pos: css of interest
3086 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3087 * is returned. This can be used during pre-order traversal to skip
3090 * While this function requires RCU read locking, it doesn't require the
3091 * whole traversal to be contained in a single RCU critical section. This
3092 * function will return the correct rightmost descendant as long as @pos is
3095 struct cgroup_subsys_state
*
3096 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3098 struct cgroup_subsys_state
*last
, *tmp
;
3100 WARN_ON_ONCE(!rcu_read_lock_held());
3104 /* ->prev isn't RCU safe, walk ->next till the end */
3106 css_for_each_child(tmp
, last
)
3112 EXPORT_SYMBOL_GPL(css_rightmost_descendant
);
3114 static struct cgroup_subsys_state
*
3115 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3117 struct cgroup_subsys_state
*last
;
3121 pos
= css_next_child(NULL
, pos
);
3128 * css_next_descendant_post - find the next descendant for post-order walk
3129 * @pos: the current position (%NULL to initiate traversal)
3130 * @root: css whose descendants to walk
3132 * To be used by css_for_each_descendant_post(). Find the next descendant
3133 * to visit for post-order traversal of @root's descendants. @root is
3134 * included in the iteration and the last node to be visited.
3136 * While this function requires RCU read locking, it doesn't require the
3137 * whole traversal to be contained in a single RCU critical section. This
3138 * function will return the correct next descendant as long as both @pos
3139 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3141 struct cgroup_subsys_state
*
3142 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3143 struct cgroup_subsys_state
*root
)
3145 struct cgroup_subsys_state
*next
;
3147 WARN_ON_ONCE(!rcu_read_lock_held());
3149 /* if first iteration, visit leftmost descendant which may be @root */
3151 return css_leftmost_descendant(root
);
3153 /* if we visited @root, we're done */
3157 /* if there's an unvisited sibling, visit its leftmost descendant */
3158 next
= css_next_child(pos
, css_parent(pos
));
3160 return css_leftmost_descendant(next
);
3162 /* no sibling left, visit parent */
3163 return css_parent(pos
);
3165 EXPORT_SYMBOL_GPL(css_next_descendant_post
);
3168 * css_advance_task_iter - advance a task itererator to the next css_set
3169 * @it: the iterator to advance
3171 * Advance @it to the next css_set to walk.
3173 static void css_advance_task_iter(struct css_task_iter
*it
)
3175 struct list_head
*l
= it
->cset_link
;
3176 struct cgrp_cset_link
*link
;
3177 struct css_set
*cset
;
3179 /* Advance to the next non-empty css_set */
3182 if (l
== &it
->origin_css
->cgroup
->cset_links
) {
3183 it
->cset_link
= NULL
;
3186 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3188 } while (list_empty(&cset
->tasks
));
3190 it
->task
= cset
->tasks
.next
;
3194 * css_task_iter_start - initiate task iteration
3195 * @css: the css to walk tasks of
3196 * @it: the task iterator to use
3198 * Initiate iteration through the tasks of @css. The caller can call
3199 * css_task_iter_next() to walk through the tasks until the function
3200 * returns NULL. On completion of iteration, css_task_iter_end() must be
3203 * Note that this function acquires a lock which is released when the
3204 * iteration finishes. The caller can't sleep while iteration is in
3207 void css_task_iter_start(struct cgroup_subsys_state
*css
,
3208 struct css_task_iter
*it
)
3209 __acquires(css_set_lock
)
3212 * The first time anyone tries to iterate across a css, we need to
3213 * enable the list linking each css_set to its tasks, and fix up
3214 * all existing tasks.
3216 if (!use_task_css_set_links
)
3217 cgroup_enable_task_cg_lists();
3219 read_lock(&css_set_lock
);
3221 it
->origin_css
= css
;
3222 it
->cset_link
= &css
->cgroup
->cset_links
;
3224 css_advance_task_iter(it
);
3228 * css_task_iter_next - return the next task for the iterator
3229 * @it: the task iterator being iterated
3231 * The "next" function for task iteration. @it should have been
3232 * initialized via css_task_iter_start(). Returns NULL when the iteration
3235 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
3237 struct task_struct
*res
;
3238 struct list_head
*l
= it
->task
;
3239 struct cgrp_cset_link
*link
;
3241 /* If the iterator cg is NULL, we have no tasks */
3244 res
= list_entry(l
, struct task_struct
, cg_list
);
3245 /* Advance iterator to find next entry */
3247 link
= list_entry(it
->cset_link
, struct cgrp_cset_link
, cset_link
);
3248 if (l
== &link
->cset
->tasks
) {
3250 * We reached the end of this task list - move on to the
3251 * next cgrp_cset_link.
3253 css_advance_task_iter(it
);
3261 * css_task_iter_end - finish task iteration
3262 * @it: the task iterator to finish
3264 * Finish task iteration started by css_task_iter_start().
3266 void css_task_iter_end(struct css_task_iter
*it
)
3267 __releases(css_set_lock
)
3269 read_unlock(&css_set_lock
);
3272 static inline int started_after_time(struct task_struct
*t1
,
3273 struct timespec
*time
,
3274 struct task_struct
*t2
)
3276 int start_diff
= timespec_compare(&t1
->start_time
, time
);
3277 if (start_diff
> 0) {
3279 } else if (start_diff
< 0) {
3283 * Arbitrarily, if two processes started at the same
3284 * time, we'll say that the lower pointer value
3285 * started first. Note that t2 may have exited by now
3286 * so this may not be a valid pointer any longer, but
3287 * that's fine - it still serves to distinguish
3288 * between two tasks started (effectively) simultaneously.
3295 * This function is a callback from heap_insert() and is used to order
3297 * In this case we order the heap in descending task start time.
3299 static inline int started_after(void *p1
, void *p2
)
3301 struct task_struct
*t1
= p1
;
3302 struct task_struct
*t2
= p2
;
3303 return started_after_time(t1
, &t2
->start_time
, t2
);
3307 * css_scan_tasks - iterate though all the tasks in a css
3308 * @css: the css to iterate tasks of
3309 * @test: optional test callback
3310 * @process: process callback
3311 * @data: data passed to @test and @process
3312 * @heap: optional pre-allocated heap used for task iteration
3314 * Iterate through all the tasks in @css, calling @test for each, and if it
3315 * returns %true, call @process for it also.
3317 * @test may be NULL, meaning always true (select all tasks), which
3318 * effectively duplicates css_task_iter_{start,next,end}() but does not
3319 * lock css_set_lock for the call to @process.
3321 * It is guaranteed that @process will act on every task that is a member
3322 * of @css for the duration of this call. This function may or may not
3323 * call @process for tasks that exit or move to a different css during the
3324 * call, or are forked or move into the css during the call.
3326 * Note that @test may be called with locks held, and may in some
3327 * situations be called multiple times for the same task, so it should be
3330 * If @heap is non-NULL, a heap has been pre-allocated and will be used for
3331 * heap operations (and its "gt" member will be overwritten), else a
3332 * temporary heap will be used (allocation of which may cause this function
3335 int css_scan_tasks(struct cgroup_subsys_state
*css
,
3336 bool (*test
)(struct task_struct
*, void *),
3337 void (*process
)(struct task_struct
*, void *),
3338 void *data
, struct ptr_heap
*heap
)
3341 struct css_task_iter it
;
3342 struct task_struct
*p
, *dropped
;
3343 /* Never dereference latest_task, since it's not refcounted */
3344 struct task_struct
*latest_task
= NULL
;
3345 struct ptr_heap tmp_heap
;
3346 struct timespec latest_time
= { 0, 0 };
3349 /* The caller supplied our heap and pre-allocated its memory */
3350 heap
->gt
= &started_after
;
3352 /* We need to allocate our own heap memory */
3354 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3356 /* cannot allocate the heap */
3362 * Scan tasks in the css, using the @test callback to determine
3363 * which are of interest, and invoking @process callback on the
3364 * ones which need an update. Since we don't want to hold any
3365 * locks during the task updates, gather tasks to be processed in a
3366 * heap structure. The heap is sorted by descending task start
3367 * time. If the statically-sized heap fills up, we overflow tasks
3368 * that started later, and in future iterations only consider tasks
3369 * that started after the latest task in the previous pass. This
3370 * guarantees forward progress and that we don't miss any tasks.
3373 css_task_iter_start(css
, &it
);
3374 while ((p
= css_task_iter_next(&it
))) {
3376 * Only affect tasks that qualify per the caller's callback,
3377 * if he provided one
3379 if (test
&& !test(p
, data
))
3382 * Only process tasks that started after the last task
3385 if (!started_after_time(p
, &latest_time
, latest_task
))
3387 dropped
= heap_insert(heap
, p
);
3388 if (dropped
== NULL
) {
3390 * The new task was inserted; the heap wasn't
3394 } else if (dropped
!= p
) {
3396 * The new task was inserted, and pushed out a
3400 put_task_struct(dropped
);
3403 * Else the new task was newer than anything already in
3404 * the heap and wasn't inserted
3407 css_task_iter_end(&it
);
3410 for (i
= 0; i
< heap
->size
; i
++) {
3411 struct task_struct
*q
= heap
->ptrs
[i
];
3413 latest_time
= q
->start_time
;
3416 /* Process the task per the caller's callback */
3421 * If we had to process any tasks at all, scan again
3422 * in case some of them were in the middle of forking
3423 * children that didn't get processed.
3424 * Not the most efficient way to do it, but it avoids
3425 * having to take callback_mutex in the fork path
3429 if (heap
== &tmp_heap
)
3430 heap_free(&tmp_heap
);
3434 static void cgroup_transfer_one_task(struct task_struct
*task
, void *data
)
3436 struct cgroup
*new_cgroup
= data
;
3438 mutex_lock(&cgroup_mutex
);
3439 cgroup_attach_task(new_cgroup
, task
, false);
3440 mutex_unlock(&cgroup_mutex
);
3444 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3445 * @to: cgroup to which the tasks will be moved
3446 * @from: cgroup in which the tasks currently reside
3448 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3450 return css_scan_tasks(&from
->dummy_css
, NULL
, cgroup_transfer_one_task
,
3455 * Stuff for reading the 'tasks'/'procs' files.
3457 * Reading this file can return large amounts of data if a cgroup has
3458 * *lots* of attached tasks. So it may need several calls to read(),
3459 * but we cannot guarantee that the information we produce is correct
3460 * unless we produce it entirely atomically.
3464 /* which pidlist file are we talking about? */
3465 enum cgroup_filetype
{
3471 * A pidlist is a list of pids that virtually represents the contents of one
3472 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3473 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3476 struct cgroup_pidlist
{
3478 * used to find which pidlist is wanted. doesn't change as long as
3479 * this particular list stays in the list.
3481 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3484 /* how many elements the above list has */
3486 /* how many files are using the current array */
3488 /* each of these stored in a list by its cgroup */
3489 struct list_head links
;
3490 /* pointer to the cgroup we belong to, for list removal purposes */
3491 struct cgroup
*owner
;
3492 /* protects the other fields */
3493 struct rw_semaphore rwsem
;
3497 * The following two functions "fix" the issue where there are more pids
3498 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3499 * TODO: replace with a kernel-wide solution to this problem
3501 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3502 static void *pidlist_allocate(int count
)
3504 if (PIDLIST_TOO_LARGE(count
))
3505 return vmalloc(count
* sizeof(pid_t
));
3507 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3509 static void pidlist_free(void *p
)
3511 if (is_vmalloc_addr(p
))
3518 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3519 * Returns the number of unique elements.
3521 static int pidlist_uniq(pid_t
*list
, int length
)
3526 * we presume the 0th element is unique, so i starts at 1. trivial
3527 * edge cases first; no work needs to be done for either
3529 if (length
== 0 || length
== 1)
3531 /* src and dest walk down the list; dest counts unique elements */
3532 for (src
= 1; src
< length
; src
++) {
3533 /* find next unique element */
3534 while (list
[src
] == list
[src
-1]) {
3539 /* dest always points to where the next unique element goes */
3540 list
[dest
] = list
[src
];
3547 static int cmppid(const void *a
, const void *b
)
3549 return *(pid_t
*)a
- *(pid_t
*)b
;
3553 * find the appropriate pidlist for our purpose (given procs vs tasks)
3554 * returns with the lock on that pidlist already held, and takes care
3555 * of the use count, or returns NULL with no locks held if we're out of
3558 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3559 enum cgroup_filetype type
)
3561 struct cgroup_pidlist
*l
;
3562 /* don't need task_nsproxy() if we're looking at ourself */
3563 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3566 * We can't drop the pidlist_mutex before taking the l->rwsem in case
3567 * the last ref-holder is trying to remove l from the list at the same
3568 * time. Holding the pidlist_mutex precludes somebody taking whichever
3569 * list we find out from under us - compare release_pid_array().
3571 mutex_lock(&cgrp
->pidlist_mutex
);
3572 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3573 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3574 /* make sure l doesn't vanish out from under us */
3575 down_write(&l
->rwsem
);
3576 mutex_unlock(&cgrp
->pidlist_mutex
);
3580 /* entry not found; create a new one */
3581 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3583 mutex_unlock(&cgrp
->pidlist_mutex
);
3586 init_rwsem(&l
->rwsem
);
3587 down_write(&l
->rwsem
);
3589 l
->key
.ns
= get_pid_ns(ns
);
3591 list_add(&l
->links
, &cgrp
->pidlists
);
3592 mutex_unlock(&cgrp
->pidlist_mutex
);
3597 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3599 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3600 struct cgroup_pidlist
**lp
)
3604 int pid
, n
= 0; /* used for populating the array */
3605 struct css_task_iter it
;
3606 struct task_struct
*tsk
;
3607 struct cgroup_pidlist
*l
;
3610 * If cgroup gets more users after we read count, we won't have
3611 * enough space - tough. This race is indistinguishable to the
3612 * caller from the case that the additional cgroup users didn't
3613 * show up until sometime later on.
3615 length
= cgroup_task_count(cgrp
);
3616 array
= pidlist_allocate(length
);
3619 /* now, populate the array */
3620 css_task_iter_start(&cgrp
->dummy_css
, &it
);
3621 while ((tsk
= css_task_iter_next(&it
))) {
3622 if (unlikely(n
== length
))
3624 /* get tgid or pid for procs or tasks file respectively */
3625 if (type
== CGROUP_FILE_PROCS
)
3626 pid
= task_tgid_vnr(tsk
);
3628 pid
= task_pid_vnr(tsk
);
3629 if (pid
> 0) /* make sure to only use valid results */
3632 css_task_iter_end(&it
);
3634 /* now sort & (if procs) strip out duplicates */
3635 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3636 if (type
== CGROUP_FILE_PROCS
)
3637 length
= pidlist_uniq(array
, length
);
3638 l
= cgroup_pidlist_find(cgrp
, type
);
3640 pidlist_free(array
);
3643 /* store array, freeing old if necessary - lock already held */
3644 pidlist_free(l
->list
);
3648 up_write(&l
->rwsem
);
3654 * cgroupstats_build - build and fill cgroupstats
3655 * @stats: cgroupstats to fill information into
3656 * @dentry: A dentry entry belonging to the cgroup for which stats have
3659 * Build and fill cgroupstats so that taskstats can export it to user
3662 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3665 struct cgroup
*cgrp
;
3666 struct css_task_iter it
;
3667 struct task_struct
*tsk
;
3670 * Validate dentry by checking the superblock operations,
3671 * and make sure it's a directory.
3673 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3674 !S_ISDIR(dentry
->d_inode
->i_mode
))
3678 cgrp
= dentry
->d_fsdata
;
3680 css_task_iter_start(&cgrp
->dummy_css
, &it
);
3681 while ((tsk
= css_task_iter_next(&it
))) {
3682 switch (tsk
->state
) {
3684 stats
->nr_running
++;
3686 case TASK_INTERRUPTIBLE
:
3687 stats
->nr_sleeping
++;
3689 case TASK_UNINTERRUPTIBLE
:
3690 stats
->nr_uninterruptible
++;
3693 stats
->nr_stopped
++;
3696 if (delayacct_is_task_waiting_on_io(tsk
))
3697 stats
->nr_io_wait
++;
3701 css_task_iter_end(&it
);
3709 * seq_file methods for the tasks/procs files. The seq_file position is the
3710 * next pid to display; the seq_file iterator is a pointer to the pid
3711 * in the cgroup->l->list array.
3714 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3717 * Initially we receive a position value that corresponds to
3718 * one more than the last pid shown (or 0 on the first call or
3719 * after a seek to the start). Use a binary-search to find the
3720 * next pid to display, if any
3722 struct cgroup_pidlist
*l
= s
->private;
3723 int index
= 0, pid
= *pos
;
3726 down_read(&l
->rwsem
);
3728 int end
= l
->length
;
3730 while (index
< end
) {
3731 int mid
= (index
+ end
) / 2;
3732 if (l
->list
[mid
] == pid
) {
3735 } else if (l
->list
[mid
] <= pid
)
3741 /* If we're off the end of the array, we're done */
3742 if (index
>= l
->length
)
3744 /* Update the abstract position to be the actual pid that we found */
3745 iter
= l
->list
+ index
;
3750 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3752 struct cgroup_pidlist
*l
= s
->private;
3756 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3758 struct cgroup_pidlist
*l
= s
->private;
3760 pid_t
*end
= l
->list
+ l
->length
;
3762 * Advance to the next pid in the array. If this goes off the
3774 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3776 return seq_printf(s
, "%d\n", *(int *)v
);
3780 * seq_operations functions for iterating on pidlists through seq_file -
3781 * independent of whether it's tasks or procs
3783 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3784 .start
= cgroup_pidlist_start
,
3785 .stop
= cgroup_pidlist_stop
,
3786 .next
= cgroup_pidlist_next
,
3787 .show
= cgroup_pidlist_show
,
3790 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3793 * the case where we're the last user of this particular pidlist will
3794 * have us remove it from the cgroup's list, which entails taking the
3795 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3796 * pidlist_mutex, we have to take pidlist_mutex first.
3798 mutex_lock(&l
->owner
->pidlist_mutex
);
3799 down_write(&l
->rwsem
);
3800 BUG_ON(!l
->use_count
);
3801 if (!--l
->use_count
) {
3802 /* we're the last user if refcount is 0; remove and free */
3803 list_del(&l
->links
);
3804 mutex_unlock(&l
->owner
->pidlist_mutex
);
3805 pidlist_free(l
->list
);
3806 put_pid_ns(l
->key
.ns
);
3807 up_write(&l
->rwsem
);
3811 mutex_unlock(&l
->owner
->pidlist_mutex
);
3812 up_write(&l
->rwsem
);
3815 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3817 struct cgroup_pidlist
*l
;
3818 if (!(file
->f_mode
& FMODE_READ
))
3821 * the seq_file will only be initialized if the file was opened for
3822 * reading; hence we check if it's not null only in that case.
3824 l
= ((struct seq_file
*)file
->private_data
)->private;
3825 cgroup_release_pid_array(l
);
3826 return seq_release(inode
, file
);
3829 static const struct file_operations cgroup_pidlist_operations
= {
3831 .llseek
= seq_lseek
,
3832 .write
= cgroup_file_write
,
3833 .release
= cgroup_pidlist_release
,
3837 * The following functions handle opens on a file that displays a pidlist
3838 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3841 /* helper function for the two below it */
3842 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3844 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3845 struct cgroup_pidlist
*l
;
3848 /* Nothing to do for write-only files */
3849 if (!(file
->f_mode
& FMODE_READ
))
3852 /* have the array populated */
3853 retval
= pidlist_array_load(cgrp
, type
, &l
);
3856 /* configure file information */
3857 file
->f_op
= &cgroup_pidlist_operations
;
3859 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3861 cgroup_release_pid_array(l
);
3864 ((struct seq_file
*)file
->private_data
)->private = l
;
3867 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3869 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3871 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3873 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3876 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
3879 return notify_on_release(css
->cgroup
);
3882 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
3883 struct cftype
*cft
, u64 val
)
3885 clear_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
3887 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
3889 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
3894 * When dput() is called asynchronously, if umount has been done and
3895 * then deactivate_super() in cgroup_free_fn() kills the superblock,
3896 * there's a small window that vfs will see the root dentry with non-zero
3897 * refcnt and trigger BUG().
3899 * That's why we hold a reference before dput() and drop it right after.
3901 static void cgroup_dput(struct cgroup
*cgrp
)
3903 struct super_block
*sb
= cgrp
->root
->sb
;
3905 atomic_inc(&sb
->s_active
);
3907 deactivate_super(sb
);
3911 * Unregister event and free resources.
3913 * Gets called from workqueue.
3915 static void cgroup_event_remove(struct work_struct
*work
)
3917 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3919 struct cgroup_subsys_state
*css
= event
->css
;
3921 remove_wait_queue(event
->wqh
, &event
->wait
);
3923 event
->cft
->unregister_event(css
, event
->cft
, event
->eventfd
);
3925 /* Notify userspace the event is going away. */
3926 eventfd_signal(event
->eventfd
, 1);
3928 eventfd_ctx_put(event
->eventfd
);
3934 * Gets called on POLLHUP on eventfd when user closes it.
3936 * Called with wqh->lock held and interrupts disabled.
3938 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3939 int sync
, void *key
)
3941 struct cgroup_event
*event
= container_of(wait
,
3942 struct cgroup_event
, wait
);
3943 struct cgroup
*cgrp
= event
->css
->cgroup
;
3944 unsigned long flags
= (unsigned long)key
;
3946 if (flags
& POLLHUP
) {
3948 * If the event has been detached at cgroup removal, we
3949 * can simply return knowing the other side will cleanup
3952 * We can't race against event freeing since the other
3953 * side will require wqh->lock via remove_wait_queue(),
3956 spin_lock(&cgrp
->event_list_lock
);
3957 if (!list_empty(&event
->list
)) {
3958 list_del_init(&event
->list
);
3960 * We are in atomic context, but cgroup_event_remove()
3961 * may sleep, so we have to call it in workqueue.
3963 schedule_work(&event
->remove
);
3965 spin_unlock(&cgrp
->event_list_lock
);
3971 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3972 wait_queue_head_t
*wqh
, poll_table
*pt
)
3974 struct cgroup_event
*event
= container_of(pt
,
3975 struct cgroup_event
, pt
);
3978 add_wait_queue(wqh
, &event
->wait
);
3982 * Parse input and register new cgroup event handler.
3984 * Input must be in format '<event_fd> <control_fd> <args>'.
3985 * Interpretation of args is defined by control file implementation.
3987 static int cgroup_write_event_control(struct cgroup_subsys_state
*dummy_css
,
3988 struct cftype
*cft
, const char *buffer
)
3990 struct cgroup
*cgrp
= dummy_css
->cgroup
;
3991 struct cgroup_event
*event
;
3992 struct cgroup_subsys_state
*cfile_css
;
3993 unsigned int efd
, cfd
;
3999 efd
= simple_strtoul(buffer
, &endp
, 10);
4004 cfd
= simple_strtoul(buffer
, &endp
, 10);
4005 if ((*endp
!= ' ') && (*endp
!= '\0'))
4009 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
4013 INIT_LIST_HEAD(&event
->list
);
4014 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
4015 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
4016 INIT_WORK(&event
->remove
, cgroup_event_remove
);
4024 event
->eventfd
= eventfd_ctx_fileget(efile
.file
);
4025 if (IS_ERR(event
->eventfd
)) {
4026 ret
= PTR_ERR(event
->eventfd
);
4033 goto out_put_eventfd
;
4036 /* the process need read permission on control file */
4037 /* AV: shouldn't we check that it's been opened for read instead? */
4038 ret
= inode_permission(file_inode(cfile
.file
), MAY_READ
);
4042 event
->cft
= __file_cft(cfile
.file
);
4043 if (IS_ERR(event
->cft
)) {
4044 ret
= PTR_ERR(event
->cft
);
4048 if (!event
->cft
->ss
) {
4054 * Determine the css of @cfile, verify it belongs to the same
4055 * cgroup as cgroup.event_control, and associate @event with it.
4056 * Remaining events are automatically removed on cgroup destruction
4057 * but the removal is asynchronous, so take an extra ref.
4062 event
->css
= cgroup_css(cgrp
, event
->cft
->ss
);
4063 cfile_css
= css_from_dir(cfile
.file
->f_dentry
->d_parent
, event
->cft
->ss
);
4064 if (event
->css
&& event
->css
== cfile_css
&& css_tryget(event
->css
))
4071 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
4076 ret
= event
->cft
->register_event(event
->css
, event
->cft
,
4077 event
->eventfd
, buffer
);
4081 efile
.file
->f_op
->poll(efile
.file
, &event
->pt
);
4083 spin_lock(&cgrp
->event_list_lock
);
4084 list_add(&event
->list
, &cgrp
->event_list
);
4085 spin_unlock(&cgrp
->event_list_lock
);
4093 css_put(event
->css
);
4097 eventfd_ctx_put(event
->eventfd
);
4106 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
4109 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4112 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
4113 struct cftype
*cft
, u64 val
)
4116 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4118 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4122 static struct cftype cgroup_base_files
[] = {
4124 .name
= "cgroup.procs",
4125 .open
= cgroup_procs_open
,
4126 .write_u64
= cgroup_procs_write
,
4127 .release
= cgroup_pidlist_release
,
4128 .mode
= S_IRUGO
| S_IWUSR
,
4131 .name
= "cgroup.event_control",
4132 .write_string
= cgroup_write_event_control
,
4136 .name
= "cgroup.clone_children",
4137 .flags
= CFTYPE_INSANE
,
4138 .read_u64
= cgroup_clone_children_read
,
4139 .write_u64
= cgroup_clone_children_write
,
4142 .name
= "cgroup.sane_behavior",
4143 .flags
= CFTYPE_ONLY_ON_ROOT
,
4144 .read_seq_string
= cgroup_sane_behavior_show
,
4148 * Historical crazy stuff. These don't have "cgroup." prefix and
4149 * don't exist if sane_behavior. If you're depending on these, be
4150 * prepared to be burned.
4154 .flags
= CFTYPE_INSANE
, /* use "procs" instead */
4155 .open
= cgroup_tasks_open
,
4156 .write_u64
= cgroup_tasks_write
,
4157 .release
= cgroup_pidlist_release
,
4158 .mode
= S_IRUGO
| S_IWUSR
,
4161 .name
= "notify_on_release",
4162 .flags
= CFTYPE_INSANE
,
4163 .read_u64
= cgroup_read_notify_on_release
,
4164 .write_u64
= cgroup_write_notify_on_release
,
4167 .name
= "release_agent",
4168 .flags
= CFTYPE_INSANE
| CFTYPE_ONLY_ON_ROOT
,
4169 .read_seq_string
= cgroup_release_agent_show
,
4170 .write_string
= cgroup_release_agent_write
,
4171 .max_write_len
= PATH_MAX
,
4177 * cgroup_populate_dir - create subsys files in a cgroup directory
4178 * @cgrp: target cgroup
4179 * @subsys_mask: mask of the subsystem ids whose files should be added
4181 * On failure, no file is added.
4183 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
4185 struct cgroup_subsys
*ss
;
4188 /* process cftsets of each subsystem */
4189 for_each_subsys(ss
, i
) {
4190 struct cftype_set
*set
;
4192 if (!test_bit(i
, &subsys_mask
))
4195 list_for_each_entry(set
, &ss
->cftsets
, node
) {
4196 ret
= cgroup_addrm_files(cgrp
, set
->cfts
, true);
4203 cgroup_clear_dir(cgrp
, subsys_mask
);
4208 * css destruction is four-stage process.
4210 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4211 * Implemented in kill_css().
4213 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4214 * and thus css_tryget() is guaranteed to fail, the css can be offlined
4215 * by invoking offline_css(). After offlining, the base ref is put.
4216 * Implemented in css_killed_work_fn().
4218 * 3. When the percpu_ref reaches zero, the only possible remaining
4219 * accessors are inside RCU read sections. css_release() schedules the
4222 * 4. After the grace period, the css can be freed. Implemented in
4223 * css_free_work_fn().
4225 * It is actually hairier because both step 2 and 4 require process context
4226 * and thus involve punting to css->destroy_work adding two additional
4227 * steps to the already complex sequence.
4229 static void css_free_work_fn(struct work_struct
*work
)
4231 struct cgroup_subsys_state
*css
=
4232 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4233 struct cgroup
*cgrp
= css
->cgroup
;
4236 css_put(css
->parent
);
4238 css
->ss
->css_free(css
);
4242 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
4244 struct cgroup_subsys_state
*css
=
4245 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
4248 * css holds an extra ref to @cgrp->dentry which is put on the last
4249 * css_put(). dput() requires process context which we don't have.
4251 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
4252 schedule_work(&css
->destroy_work
);
4255 static void css_release(struct percpu_ref
*ref
)
4257 struct cgroup_subsys_state
*css
=
4258 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4260 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4263 static void init_css(struct cgroup_subsys_state
*css
, struct cgroup_subsys
*ss
,
4264 struct cgroup
*cgrp
)
4271 css
->parent
= cgroup_css(cgrp
->parent
, ss
);
4273 css
->flags
|= CSS_ROOT
;
4275 BUG_ON(cgroup_css(cgrp
, ss
));
4278 /* invoke ->css_online() on a new CSS and mark it online if successful */
4279 static int online_css(struct cgroup_subsys_state
*css
)
4281 struct cgroup_subsys
*ss
= css
->ss
;
4284 lockdep_assert_held(&cgroup_mutex
);
4287 ret
= ss
->css_online(css
);
4289 css
->flags
|= CSS_ONLINE
;
4290 css
->cgroup
->nr_css
++;
4291 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->subsys_id
], css
);
4296 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4297 static void offline_css(struct cgroup_subsys_state
*css
)
4299 struct cgroup_subsys
*ss
= css
->ss
;
4301 lockdep_assert_held(&cgroup_mutex
);
4303 if (!(css
->flags
& CSS_ONLINE
))
4306 if (ss
->css_offline
)
4307 ss
->css_offline(css
);
4309 css
->flags
&= ~CSS_ONLINE
;
4310 css
->cgroup
->nr_css
--;
4311 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->subsys_id
], css
);
4315 * cgroup_create - create a cgroup
4316 * @parent: cgroup that will be parent of the new cgroup
4317 * @dentry: dentry of the new cgroup
4318 * @mode: mode to set on new inode
4320 * Must be called with the mutex on the parent inode held
4322 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
4325 struct cgroup_subsys_state
*css_ar
[CGROUP_SUBSYS_COUNT
] = { };
4326 struct cgroup
*cgrp
;
4327 struct cgroup_name
*name
;
4328 struct cgroupfs_root
*root
= parent
->root
;
4330 struct cgroup_subsys
*ss
;
4331 struct super_block
*sb
= root
->sb
;
4333 /* allocate the cgroup and its ID, 0 is reserved for the root */
4334 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4338 name
= cgroup_alloc_name(dentry
);
4341 rcu_assign_pointer(cgrp
->name
, name
);
4344 * Temporarily set the pointer to NULL, so idr_find() won't return
4345 * a half-baked cgroup.
4347 cgrp
->id
= idr_alloc(&root
->cgroup_idr
, NULL
, 1, 0, GFP_KERNEL
);
4352 * Only live parents can have children. Note that the liveliness
4353 * check isn't strictly necessary because cgroup_mkdir() and
4354 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4355 * anyway so that locking is contained inside cgroup proper and we
4356 * don't get nasty surprises if we ever grow another caller.
4358 if (!cgroup_lock_live_group(parent
)) {
4363 /* Grab a reference on the superblock so the hierarchy doesn't
4364 * get deleted on unmount if there are child cgroups. This
4365 * can be done outside cgroup_mutex, since the sb can't
4366 * disappear while someone has an open control file on the
4368 atomic_inc(&sb
->s_active
);
4370 init_cgroup_housekeeping(cgrp
);
4372 dentry
->d_fsdata
= cgrp
;
4373 cgrp
->dentry
= dentry
;
4375 cgrp
->parent
= parent
;
4376 cgrp
->dummy_css
.parent
= &parent
->dummy_css
;
4377 cgrp
->root
= parent
->root
;
4379 if (notify_on_release(parent
))
4380 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4382 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4383 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4385 for_each_root_subsys(root
, ss
) {
4386 struct cgroup_subsys_state
*css
;
4388 css
= ss
->css_alloc(cgroup_css(parent
, ss
));
4393 css_ar
[ss
->subsys_id
] = css
;
4395 err
= percpu_ref_init(&css
->refcnt
, css_release
);
4399 init_css(css
, ss
, cgrp
);
4403 * Create directory. cgroup_create_file() returns with the new
4404 * directory locked on success so that it can be populated without
4405 * dropping cgroup_mutex.
4407 err
= cgroup_create_file(dentry
, S_IFDIR
| mode
, sb
);
4410 lockdep_assert_held(&dentry
->d_inode
->i_mutex
);
4412 cgrp
->serial_nr
= cgroup_serial_nr_next
++;
4414 /* allocation complete, commit to creation */
4415 list_add_tail_rcu(&cgrp
->sibling
, &cgrp
->parent
->children
);
4416 root
->number_of_cgroups
++;
4418 /* each css holds a ref to the cgroup's dentry and the parent css */
4419 for_each_root_subsys(root
, ss
) {
4420 struct cgroup_subsys_state
*css
= css_ar
[ss
->subsys_id
];
4423 css_get(css
->parent
);
4426 /* hold a ref to the parent's dentry */
4427 dget(parent
->dentry
);
4429 /* creation succeeded, notify subsystems */
4430 for_each_root_subsys(root
, ss
) {
4431 struct cgroup_subsys_state
*css
= css_ar
[ss
->subsys_id
];
4433 err
= online_css(css
);
4437 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4439 pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4440 current
->comm
, current
->pid
, ss
->name
);
4441 if (!strcmp(ss
->name
, "memory"))
4442 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4443 ss
->warned_broken_hierarchy
= true;
4447 idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4449 err
= cgroup_addrm_files(cgrp
, cgroup_base_files
, true);
4453 err
= cgroup_populate_dir(cgrp
, root
->subsys_mask
);
4457 mutex_unlock(&cgroup_mutex
);
4458 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4463 for_each_root_subsys(root
, ss
) {
4464 struct cgroup_subsys_state
*css
= css_ar
[ss
->subsys_id
];
4467 percpu_ref_cancel_init(&css
->refcnt
);
4471 mutex_unlock(&cgroup_mutex
);
4472 /* Release the reference count that we took on the superblock */
4473 deactivate_super(sb
);
4475 idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4477 kfree(rcu_dereference_raw(cgrp
->name
));
4483 cgroup_destroy_locked(cgrp
);
4484 mutex_unlock(&cgroup_mutex
);
4485 mutex_unlock(&dentry
->d_inode
->i_mutex
);
4489 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4491 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4493 /* the vfs holds inode->i_mutex already */
4494 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4498 * This is called when the refcnt of a css is confirmed to be killed.
4499 * css_tryget() is now guaranteed to fail.
4501 static void css_killed_work_fn(struct work_struct
*work
)
4503 struct cgroup_subsys_state
*css
=
4504 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4505 struct cgroup
*cgrp
= css
->cgroup
;
4507 mutex_lock(&cgroup_mutex
);
4510 * css_tryget() is guaranteed to fail now. Tell subsystems to
4511 * initate destruction.
4516 * If @cgrp is marked dead, it's waiting for refs of all css's to
4517 * be disabled before proceeding to the second phase of cgroup
4518 * destruction. If we are the last one, kick it off.
4520 if (!cgrp
->nr_css
&& cgroup_is_dead(cgrp
))
4521 cgroup_destroy_css_killed(cgrp
);
4523 mutex_unlock(&cgroup_mutex
);
4526 * Put the css refs from kill_css(). Each css holds an extra
4527 * reference to the cgroup's dentry and cgroup removal proceeds
4528 * regardless of css refs. On the last put of each css, whenever
4529 * that may be, the extra dentry ref is put so that dentry
4530 * destruction happens only after all css's are released.
4535 /* css kill confirmation processing requires process context, bounce */
4536 static void css_killed_ref_fn(struct percpu_ref
*ref
)
4538 struct cgroup_subsys_state
*css
=
4539 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4541 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
4542 schedule_work(&css
->destroy_work
);
4546 * kill_css - destroy a css
4547 * @css: css to destroy
4549 * This function initiates destruction of @css by removing cgroup interface
4550 * files and putting its base reference. ->css_offline() will be invoked
4551 * asynchronously once css_tryget() is guaranteed to fail and when the
4552 * reference count reaches zero, @css will be released.
4554 static void kill_css(struct cgroup_subsys_state
*css
)
4556 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->subsys_id
);
4559 * Killing would put the base ref, but we need to keep it alive
4560 * until after ->css_offline().
4565 * cgroup core guarantees that, by the time ->css_offline() is
4566 * invoked, no new css reference will be given out via
4567 * css_tryget(). We can't simply call percpu_ref_kill() and
4568 * proceed to offlining css's because percpu_ref_kill() doesn't
4569 * guarantee that the ref is seen as killed on all CPUs on return.
4571 * Use percpu_ref_kill_and_confirm() to get notifications as each
4572 * css is confirmed to be seen as killed on all CPUs.
4574 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
4578 * cgroup_destroy_locked - the first stage of cgroup destruction
4579 * @cgrp: cgroup to be destroyed
4581 * css's make use of percpu refcnts whose killing latency shouldn't be
4582 * exposed to userland and are RCU protected. Also, cgroup core needs to
4583 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
4584 * invoked. To satisfy all the requirements, destruction is implemented in
4585 * the following two steps.
4587 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4588 * userland visible parts and start killing the percpu refcnts of
4589 * css's. Set up so that the next stage will be kicked off once all
4590 * the percpu refcnts are confirmed to be killed.
4592 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4593 * rest of destruction. Once all cgroup references are gone, the
4594 * cgroup is RCU-freed.
4596 * This function implements s1. After this step, @cgrp is gone as far as
4597 * the userland is concerned and a new cgroup with the same name may be
4598 * created. As cgroup doesn't care about the names internally, this
4599 * doesn't cause any problem.
4601 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4602 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4604 struct dentry
*d
= cgrp
->dentry
;
4605 struct cgroup_event
*event
, *tmp
;
4606 struct cgroup_subsys
*ss
;
4607 struct cgroup
*child
;
4610 lockdep_assert_held(&d
->d_inode
->i_mutex
);
4611 lockdep_assert_held(&cgroup_mutex
);
4614 * css_set_lock synchronizes access to ->cset_links and prevents
4615 * @cgrp from being removed while __put_css_set() is in progress.
4617 read_lock(&css_set_lock
);
4618 empty
= list_empty(&cgrp
->cset_links
);
4619 read_unlock(&css_set_lock
);
4624 * Make sure there's no live children. We can't test ->children
4625 * emptiness as dead children linger on it while being destroyed;
4626 * otherwise, "rmdir parent/child parent" may fail with -EBUSY.
4630 list_for_each_entry_rcu(child
, &cgrp
->children
, sibling
) {
4631 empty
= cgroup_is_dead(child
);
4640 * Initiate massacre of all css's. cgroup_destroy_css_killed()
4641 * will be invoked to perform the rest of destruction once the
4642 * percpu refs of all css's are confirmed to be killed.
4644 for_each_root_subsys(cgrp
->root
, ss
)
4645 kill_css(cgroup_css(cgrp
, ss
));
4648 * Mark @cgrp dead. This prevents further task migration and child
4649 * creation by disabling cgroup_lock_live_group(). Note that
4650 * CGRP_DEAD assertion is depended upon by css_next_child() to
4651 * resume iteration after dropping RCU read lock. See
4652 * css_next_child() for details.
4654 set_bit(CGRP_DEAD
, &cgrp
->flags
);
4656 /* CGRP_DEAD is set, remove from ->release_list for the last time */
4657 raw_spin_lock(&release_list_lock
);
4658 if (!list_empty(&cgrp
->release_list
))
4659 list_del_init(&cgrp
->release_list
);
4660 raw_spin_unlock(&release_list_lock
);
4663 * If @cgrp has css's attached, the second stage of cgroup
4664 * destruction is kicked off from css_killed_work_fn() after the
4665 * refs of all attached css's are killed. If @cgrp doesn't have
4666 * any css, we kick it off here.
4669 cgroup_destroy_css_killed(cgrp
);
4672 * Clear the base files and remove @cgrp directory. The removal
4673 * puts the base ref but we aren't quite done with @cgrp yet, so
4676 cgroup_addrm_files(cgrp
, cgroup_base_files
, false);
4678 cgroup_d_remove_dir(d
);
4681 * Unregister events and notify userspace.
4682 * Notify userspace about cgroup removing only after rmdir of cgroup
4683 * directory to avoid race between userspace and kernelspace.
4685 spin_lock(&cgrp
->event_list_lock
);
4686 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4687 list_del_init(&event
->list
);
4688 schedule_work(&event
->remove
);
4690 spin_unlock(&cgrp
->event_list_lock
);
4696 * cgroup_destroy_css_killed - the second step of cgroup destruction
4697 * @work: cgroup->destroy_free_work
4699 * This function is invoked from a work item for a cgroup which is being
4700 * destroyed after all css's are offlined and performs the rest of
4701 * destruction. This is the second step of destruction described in the
4702 * comment above cgroup_destroy_locked().
4704 static void cgroup_destroy_css_killed(struct cgroup
*cgrp
)
4706 struct cgroup
*parent
= cgrp
->parent
;
4707 struct dentry
*d
= cgrp
->dentry
;
4709 lockdep_assert_held(&cgroup_mutex
);
4711 /* delete this cgroup from parent->children */
4712 list_del_rcu(&cgrp
->sibling
);
4715 * We should remove the cgroup object from idr before its grace
4716 * period starts, so we won't be looking up a cgroup while the
4717 * cgroup is being freed.
4719 idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4724 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4725 check_for_release(parent
);
4728 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4732 mutex_lock(&cgroup_mutex
);
4733 ret
= cgroup_destroy_locked(dentry
->d_fsdata
);
4734 mutex_unlock(&cgroup_mutex
);
4739 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4741 INIT_LIST_HEAD(&ss
->cftsets
);
4744 * base_cftset is embedded in subsys itself, no need to worry about
4747 if (ss
->base_cftypes
) {
4750 for (cft
= ss
->base_cftypes
; cft
->name
[0] != '\0'; cft
++)
4753 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4754 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4758 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4760 struct cgroup_subsys_state
*css
;
4762 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4764 mutex_lock(&cgroup_mutex
);
4766 /* init base cftset */
4767 cgroup_init_cftsets(ss
);
4769 /* Create the top cgroup state for this subsystem */
4770 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4771 ss
->root
= &cgroup_dummy_root
;
4772 css
= ss
->css_alloc(cgroup_css(cgroup_dummy_top
, ss
));
4773 /* We don't handle early failures gracefully */
4774 BUG_ON(IS_ERR(css
));
4775 init_css(css
, ss
, cgroup_dummy_top
);
4777 /* Update the init_css_set to contain a subsys
4778 * pointer to this state - since the subsystem is
4779 * newly registered, all tasks and hence the
4780 * init_css_set is in the subsystem's top cgroup. */
4781 init_css_set
.subsys
[ss
->subsys_id
] = css
;
4783 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4785 /* At system boot, before all subsystems have been
4786 * registered, no tasks have been forked, so we don't
4787 * need to invoke fork callbacks here. */
4788 BUG_ON(!list_empty(&init_task
.tasks
));
4790 BUG_ON(online_css(css
));
4792 mutex_unlock(&cgroup_mutex
);
4794 /* this function shouldn't be used with modular subsystems, since they
4795 * need to register a subsys_id, among other things */
4800 * cgroup_load_subsys: load and register a modular subsystem at runtime
4801 * @ss: the subsystem to load
4803 * This function should be called in a modular subsystem's initcall. If the
4804 * subsystem is built as a module, it will be assigned a new subsys_id and set
4805 * up for use. If the subsystem is built-in anyway, work is delegated to the
4806 * simpler cgroup_init_subsys.
4808 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4810 struct cgroup_subsys_state
*css
;
4812 struct hlist_node
*tmp
;
4813 struct css_set
*cset
;
4816 /* check name and function validity */
4817 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4818 ss
->css_alloc
== NULL
|| ss
->css_free
== NULL
)
4822 * we don't support callbacks in modular subsystems. this check is
4823 * before the ss->module check for consistency; a subsystem that could
4824 * be a module should still have no callbacks even if the user isn't
4825 * compiling it as one.
4827 if (ss
->fork
|| ss
->exit
)
4831 * an optionally modular subsystem is built-in: we want to do nothing,
4832 * since cgroup_init_subsys will have already taken care of it.
4834 if (ss
->module
== NULL
) {
4835 /* a sanity check */
4836 BUG_ON(cgroup_subsys
[ss
->subsys_id
] != ss
);
4840 /* init base cftset */
4841 cgroup_init_cftsets(ss
);
4843 mutex_lock(&cgroup_mutex
);
4844 cgroup_subsys
[ss
->subsys_id
] = ss
;
4847 * no ss->css_alloc seems to need anything important in the ss
4848 * struct, so this can happen first (i.e. before the dummy root
4851 css
= ss
->css_alloc(cgroup_css(cgroup_dummy_top
, ss
));
4853 /* failure case - need to deassign the cgroup_subsys[] slot. */
4854 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4855 mutex_unlock(&cgroup_mutex
);
4856 return PTR_ERR(css
);
4859 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4860 ss
->root
= &cgroup_dummy_root
;
4862 /* our new subsystem will be attached to the dummy hierarchy. */
4863 init_css(css
, ss
, cgroup_dummy_top
);
4866 * Now we need to entangle the css into the existing css_sets. unlike
4867 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4868 * will need a new pointer to it; done by iterating the css_set_table.
4869 * furthermore, modifying the existing css_sets will corrupt the hash
4870 * table state, so each changed css_set will need its hash recomputed.
4871 * this is all done under the css_set_lock.
4873 write_lock(&css_set_lock
);
4874 hash_for_each_safe(css_set_table
, i
, tmp
, cset
, hlist
) {
4875 /* skip entries that we already rehashed */
4876 if (cset
->subsys
[ss
->subsys_id
])
4878 /* remove existing entry */
4879 hash_del(&cset
->hlist
);
4881 cset
->subsys
[ss
->subsys_id
] = css
;
4882 /* recompute hash and restore entry */
4883 key
= css_set_hash(cset
->subsys
);
4884 hash_add(css_set_table
, &cset
->hlist
, key
);
4886 write_unlock(&css_set_lock
);
4888 ret
= online_css(css
);
4893 mutex_unlock(&cgroup_mutex
);
4897 mutex_unlock(&cgroup_mutex
);
4898 /* @ss can't be mounted here as try_module_get() would fail */
4899 cgroup_unload_subsys(ss
);
4902 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4905 * cgroup_unload_subsys: unload a modular subsystem
4906 * @ss: the subsystem to unload
4908 * This function should be called in a modular subsystem's exitcall. When this
4909 * function is invoked, the refcount on the subsystem's module will be 0, so
4910 * the subsystem will not be attached to any hierarchy.
4912 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4914 struct cgrp_cset_link
*link
;
4916 BUG_ON(ss
->module
== NULL
);
4919 * we shouldn't be called if the subsystem is in use, and the use of
4920 * try_module_get() in rebind_subsystems() should ensure that it
4921 * doesn't start being used while we're killing it off.
4923 BUG_ON(ss
->root
!= &cgroup_dummy_root
);
4925 mutex_lock(&cgroup_mutex
);
4927 offline_css(cgroup_css(cgroup_dummy_top
, ss
));
4929 /* deassign the subsys_id */
4930 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4932 /* remove subsystem from the dummy root's list of subsystems */
4933 list_del_init(&ss
->sibling
);
4936 * disentangle the css from all css_sets attached to the dummy
4937 * top. as in loading, we need to pay our respects to the hashtable
4940 write_lock(&css_set_lock
);
4941 list_for_each_entry(link
, &cgroup_dummy_top
->cset_links
, cset_link
) {
4942 struct css_set
*cset
= link
->cset
;
4945 hash_del(&cset
->hlist
);
4946 cset
->subsys
[ss
->subsys_id
] = NULL
;
4947 key
= css_set_hash(cset
->subsys
);
4948 hash_add(css_set_table
, &cset
->hlist
, key
);
4950 write_unlock(&css_set_lock
);
4953 * remove subsystem's css from the cgroup_dummy_top and free it -
4954 * need to free before marking as null because ss->css_free needs
4955 * the cgrp->subsys pointer to find their state.
4957 ss
->css_free(cgroup_css(cgroup_dummy_top
, ss
));
4958 RCU_INIT_POINTER(cgroup_dummy_top
->subsys
[ss
->subsys_id
], NULL
);
4960 mutex_unlock(&cgroup_mutex
);
4962 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4965 * cgroup_init_early - cgroup initialization at system boot
4967 * Initialize cgroups at system boot, and initialize any
4968 * subsystems that request early init.
4970 int __init
cgroup_init_early(void)
4972 struct cgroup_subsys
*ss
;
4975 atomic_set(&init_css_set
.refcount
, 1);
4976 INIT_LIST_HEAD(&init_css_set
.cgrp_links
);
4977 INIT_LIST_HEAD(&init_css_set
.tasks
);
4978 INIT_HLIST_NODE(&init_css_set
.hlist
);
4980 init_cgroup_root(&cgroup_dummy_root
);
4981 cgroup_root_count
= 1;
4982 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
4984 init_cgrp_cset_link
.cset
= &init_css_set
;
4985 init_cgrp_cset_link
.cgrp
= cgroup_dummy_top
;
4986 list_add(&init_cgrp_cset_link
.cset_link
, &cgroup_dummy_top
->cset_links
);
4987 list_add(&init_cgrp_cset_link
.cgrp_link
, &init_css_set
.cgrp_links
);
4989 /* at bootup time, we don't worry about modular subsystems */
4990 for_each_builtin_subsys(ss
, i
) {
4992 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4993 BUG_ON(!ss
->css_alloc
);
4994 BUG_ON(!ss
->css_free
);
4995 if (ss
->subsys_id
!= i
) {
4996 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4997 ss
->name
, ss
->subsys_id
);
5002 cgroup_init_subsys(ss
);
5008 * cgroup_init - cgroup initialization
5010 * Register cgroup filesystem and /proc file, and initialize
5011 * any subsystems that didn't request early init.
5013 int __init
cgroup_init(void)
5015 struct cgroup_subsys
*ss
;
5019 err
= bdi_init(&cgroup_backing_dev_info
);
5023 for_each_builtin_subsys(ss
, i
) {
5024 if (!ss
->early_init
)
5025 cgroup_init_subsys(ss
);
5028 /* allocate id for the dummy hierarchy */
5029 mutex_lock(&cgroup_mutex
);
5030 mutex_lock(&cgroup_root_mutex
);
5032 /* Add init_css_set to the hash table */
5033 key
= css_set_hash(init_css_set
.subsys
);
5034 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
5036 BUG_ON(cgroup_init_root_id(&cgroup_dummy_root
, 0, 1));
5038 err
= idr_alloc(&cgroup_dummy_root
.cgroup_idr
, cgroup_dummy_top
,
5042 mutex_unlock(&cgroup_root_mutex
);
5043 mutex_unlock(&cgroup_mutex
);
5045 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
5051 err
= register_filesystem(&cgroup_fs_type
);
5053 kobject_put(cgroup_kobj
);
5057 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
5061 bdi_destroy(&cgroup_backing_dev_info
);
5067 * proc_cgroup_show()
5068 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5069 * - Used for /proc/<pid>/cgroup.
5070 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
5071 * doesn't really matter if tsk->cgroup changes after we read it,
5072 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
5073 * anyway. No need to check that tsk->cgroup != NULL, thanks to
5074 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
5075 * cgroup to top_cgroup.
5078 /* TODO: Use a proper seq_file iterator */
5079 int proc_cgroup_show(struct seq_file
*m
, void *v
)
5082 struct task_struct
*tsk
;
5085 struct cgroupfs_root
*root
;
5088 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5094 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
5100 mutex_lock(&cgroup_mutex
);
5102 for_each_active_root(root
) {
5103 struct cgroup_subsys
*ss
;
5104 struct cgroup
*cgrp
;
5107 seq_printf(m
, "%d:", root
->hierarchy_id
);
5108 for_each_root_subsys(root
, ss
)
5109 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
5110 if (strlen(root
->name
))
5111 seq_printf(m
, "%sname=%s", count
? "," : "",
5114 cgrp
= task_cgroup_from_root(tsk
, root
);
5115 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
5123 mutex_unlock(&cgroup_mutex
);
5124 put_task_struct(tsk
);
5131 /* Display information about each subsystem and each hierarchy */
5132 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5134 struct cgroup_subsys
*ss
;
5137 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5139 * ideally we don't want subsystems moving around while we do this.
5140 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5141 * subsys/hierarchy state.
5143 mutex_lock(&cgroup_mutex
);
5145 for_each_subsys(ss
, i
)
5146 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5147 ss
->name
, ss
->root
->hierarchy_id
,
5148 ss
->root
->number_of_cgroups
, !ss
->disabled
);
5150 mutex_unlock(&cgroup_mutex
);
5154 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5156 return single_open(file
, proc_cgroupstats_show
, NULL
);
5159 static const struct file_operations proc_cgroupstats_operations
= {
5160 .open
= cgroupstats_open
,
5162 .llseek
= seq_lseek
,
5163 .release
= single_release
,
5167 * cgroup_fork - attach newly forked task to its parents cgroup.
5168 * @child: pointer to task_struct of forking parent process.
5170 * Description: A task inherits its parent's cgroup at fork().
5172 * A pointer to the shared css_set was automatically copied in
5173 * fork.c by dup_task_struct(). However, we ignore that copy, since
5174 * it was not made under the protection of RCU or cgroup_mutex, so
5175 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
5176 * have already changed current->cgroups, allowing the previously
5177 * referenced cgroup group to be removed and freed.
5179 * At the point that cgroup_fork() is called, 'current' is the parent
5180 * task, and the passed argument 'child' points to the child task.
5182 void cgroup_fork(struct task_struct
*child
)
5185 get_css_set(task_css_set(current
));
5186 child
->cgroups
= current
->cgroups
;
5187 task_unlock(current
);
5188 INIT_LIST_HEAD(&child
->cg_list
);
5192 * cgroup_post_fork - called on a new task after adding it to the task list
5193 * @child: the task in question
5195 * Adds the task to the list running through its css_set if necessary and
5196 * call the subsystem fork() callbacks. Has to be after the task is
5197 * visible on the task list in case we race with the first call to
5198 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5201 void cgroup_post_fork(struct task_struct
*child
)
5203 struct cgroup_subsys
*ss
;
5207 * use_task_css_set_links is set to 1 before we walk the tasklist
5208 * under the tasklist_lock and we read it here after we added the child
5209 * to the tasklist under the tasklist_lock as well. If the child wasn't
5210 * yet in the tasklist when we walked through it from
5211 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
5212 * should be visible now due to the paired locking and barriers implied
5213 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
5214 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
5217 if (use_task_css_set_links
) {
5218 write_lock(&css_set_lock
);
5220 if (list_empty(&child
->cg_list
))
5221 list_add(&child
->cg_list
, &task_css_set(child
)->tasks
);
5223 write_unlock(&css_set_lock
);
5227 * Call ss->fork(). This must happen after @child is linked on
5228 * css_set; otherwise, @child might change state between ->fork()
5229 * and addition to css_set.
5231 if (need_forkexit_callback
) {
5233 * fork/exit callbacks are supported only for builtin
5234 * subsystems, and the builtin section of the subsys
5235 * array is immutable, so we don't need to lock the
5236 * subsys array here. On the other hand, modular section
5237 * of the array can be freed at module unload, so we
5240 for_each_builtin_subsys(ss
, i
)
5247 * cgroup_exit - detach cgroup from exiting task
5248 * @tsk: pointer to task_struct of exiting process
5249 * @run_callback: run exit callbacks?
5251 * Description: Detach cgroup from @tsk and release it.
5253 * Note that cgroups marked notify_on_release force every task in
5254 * them to take the global cgroup_mutex mutex when exiting.
5255 * This could impact scaling on very large systems. Be reluctant to
5256 * use notify_on_release cgroups where very high task exit scaling
5257 * is required on large systems.
5259 * the_top_cgroup_hack:
5261 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
5263 * We call cgroup_exit() while the task is still competent to
5264 * handle notify_on_release(), then leave the task attached to the
5265 * root cgroup in each hierarchy for the remainder of its exit.
5267 * To do this properly, we would increment the reference count on
5268 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
5269 * code we would add a second cgroup function call, to drop that
5270 * reference. This would just create an unnecessary hot spot on
5271 * the top_cgroup reference count, to no avail.
5273 * Normally, holding a reference to a cgroup without bumping its
5274 * count is unsafe. The cgroup could go away, or someone could
5275 * attach us to a different cgroup, decrementing the count on
5276 * the first cgroup that we never incremented. But in this case,
5277 * top_cgroup isn't going away, and either task has PF_EXITING set,
5278 * which wards off any cgroup_attach_task() attempts, or task is a failed
5279 * fork, never visible to cgroup_attach_task.
5281 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
5283 struct cgroup_subsys
*ss
;
5284 struct css_set
*cset
;
5288 * Unlink from the css_set task list if necessary.
5289 * Optimistically check cg_list before taking
5292 if (!list_empty(&tsk
->cg_list
)) {
5293 write_lock(&css_set_lock
);
5294 if (!list_empty(&tsk
->cg_list
))
5295 list_del_init(&tsk
->cg_list
);
5296 write_unlock(&css_set_lock
);
5299 /* Reassign the task to the init_css_set. */
5301 cset
= task_css_set(tsk
);
5302 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5304 if (run_callbacks
&& need_forkexit_callback
) {
5306 * fork/exit callbacks are supported only for builtin
5307 * subsystems, see cgroup_post_fork() for details.
5309 for_each_builtin_subsys(ss
, i
) {
5311 struct cgroup_subsys_state
*old_css
= cset
->subsys
[i
];
5312 struct cgroup_subsys_state
*css
= task_css(tsk
, i
);
5314 ss
->exit(css
, old_css
, tsk
);
5320 put_css_set_taskexit(cset
);
5323 static void check_for_release(struct cgroup
*cgrp
)
5325 if (cgroup_is_releasable(cgrp
) &&
5326 list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
)) {
5328 * Control Group is currently removeable. If it's not
5329 * already queued for a userspace notification, queue
5332 int need_schedule_work
= 0;
5334 raw_spin_lock(&release_list_lock
);
5335 if (!cgroup_is_dead(cgrp
) &&
5336 list_empty(&cgrp
->release_list
)) {
5337 list_add(&cgrp
->release_list
, &release_list
);
5338 need_schedule_work
= 1;
5340 raw_spin_unlock(&release_list_lock
);
5341 if (need_schedule_work
)
5342 schedule_work(&release_agent_work
);
5347 * Notify userspace when a cgroup is released, by running the
5348 * configured release agent with the name of the cgroup (path
5349 * relative to the root of cgroup file system) as the argument.
5351 * Most likely, this user command will try to rmdir this cgroup.
5353 * This races with the possibility that some other task will be
5354 * attached to this cgroup before it is removed, or that some other
5355 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5356 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5357 * unused, and this cgroup will be reprieved from its death sentence,
5358 * to continue to serve a useful existence. Next time it's released,
5359 * we will get notified again, if it still has 'notify_on_release' set.
5361 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5362 * means only wait until the task is successfully execve()'d. The
5363 * separate release agent task is forked by call_usermodehelper(),
5364 * then control in this thread returns here, without waiting for the
5365 * release agent task. We don't bother to wait because the caller of
5366 * this routine has no use for the exit status of the release agent
5367 * task, so no sense holding our caller up for that.
5369 static void cgroup_release_agent(struct work_struct
*work
)
5371 BUG_ON(work
!= &release_agent_work
);
5372 mutex_lock(&cgroup_mutex
);
5373 raw_spin_lock(&release_list_lock
);
5374 while (!list_empty(&release_list
)) {
5375 char *argv
[3], *envp
[3];
5377 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5378 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5381 list_del_init(&cgrp
->release_list
);
5382 raw_spin_unlock(&release_list_lock
);
5383 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5386 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5388 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5393 argv
[i
++] = agentbuf
;
5394 argv
[i
++] = pathbuf
;
5398 /* minimal command environment */
5399 envp
[i
++] = "HOME=/";
5400 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5403 /* Drop the lock while we invoke the usermode helper,
5404 * since the exec could involve hitting disk and hence
5405 * be a slow process */
5406 mutex_unlock(&cgroup_mutex
);
5407 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5408 mutex_lock(&cgroup_mutex
);
5412 raw_spin_lock(&release_list_lock
);
5414 raw_spin_unlock(&release_list_lock
);
5415 mutex_unlock(&cgroup_mutex
);
5418 static int __init
cgroup_disable(char *str
)
5420 struct cgroup_subsys
*ss
;
5424 while ((token
= strsep(&str
, ",")) != NULL
) {
5429 * cgroup_disable, being at boot time, can't know about
5430 * module subsystems, so we don't worry about them.
5432 for_each_builtin_subsys(ss
, i
) {
5433 if (!strcmp(token
, ss
->name
)) {
5435 printk(KERN_INFO
"Disabling %s control group"
5436 " subsystem\n", ss
->name
);
5443 __setup("cgroup_disable=", cgroup_disable
);
5446 * css_from_dir - get corresponding css from the dentry of a cgroup dir
5447 * @dentry: directory dentry of interest
5448 * @ss: subsystem of interest
5450 * Must be called under RCU read lock. The caller is responsible for
5451 * pinning the returned css if it needs to be accessed outside the RCU
5454 struct cgroup_subsys_state
*css_from_dir(struct dentry
*dentry
,
5455 struct cgroup_subsys
*ss
)
5457 struct cgroup
*cgrp
;
5459 WARN_ON_ONCE(!rcu_read_lock_held());
5461 /* is @dentry a cgroup dir? */
5462 if (!dentry
->d_inode
||
5463 dentry
->d_inode
->i_op
!= &cgroup_dir_inode_operations
)
5464 return ERR_PTR(-EBADF
);
5466 cgrp
= __d_cgrp(dentry
);
5467 return cgroup_css(cgrp
, ss
) ?: ERR_PTR(-ENOENT
);
5471 * css_from_id - lookup css by id
5472 * @id: the cgroup id
5473 * @ss: cgroup subsys to be looked into
5475 * Returns the css if there's valid one with @id, otherwise returns NULL.
5476 * Should be called under rcu_read_lock().
5478 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5480 struct cgroup
*cgrp
;
5482 rcu_lockdep_assert(rcu_read_lock_held() ||
5483 lockdep_is_held(&cgroup_mutex
),
5484 "css_from_id() needs proper protection");
5486 cgrp
= idr_find(&ss
->root
->cgroup_idr
, id
);
5488 return cgroup_css(cgrp
, ss
);
5492 #ifdef CONFIG_CGROUP_DEBUG
5493 static struct cgroup_subsys_state
*
5494 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
5496 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5499 return ERR_PTR(-ENOMEM
);
5504 static void debug_css_free(struct cgroup_subsys_state
*css
)
5509 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
5512 return cgroup_task_count(css
->cgroup
);
5515 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
5518 return (u64
)(unsigned long)current
->cgroups
;
5521 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
5527 count
= atomic_read(&task_css_set(current
)->refcount
);
5532 static int current_css_set_cg_links_read(struct cgroup_subsys_state
*css
,
5534 struct seq_file
*seq
)
5536 struct cgrp_cset_link
*link
;
5537 struct css_set
*cset
;
5539 read_lock(&css_set_lock
);
5541 cset
= rcu_dereference(current
->cgroups
);
5542 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5543 struct cgroup
*c
= link
->cgrp
;
5547 name
= c
->dentry
->d_name
.name
;
5550 seq_printf(seq
, "Root %d group %s\n",
5551 c
->root
->hierarchy_id
, name
);
5554 read_unlock(&css_set_lock
);
5558 #define MAX_TASKS_SHOWN_PER_CSS 25
5559 static int cgroup_css_links_read(struct cgroup_subsys_state
*css
,
5560 struct cftype
*cft
, struct seq_file
*seq
)
5562 struct cgrp_cset_link
*link
;
5564 read_lock(&css_set_lock
);
5565 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
5566 struct css_set
*cset
= link
->cset
;
5567 struct task_struct
*task
;
5569 seq_printf(seq
, "css_set %p\n", cset
);
5570 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5571 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5572 seq_puts(seq
, " ...\n");
5575 seq_printf(seq
, " task %d\n",
5576 task_pid_vnr(task
));
5580 read_unlock(&css_set_lock
);
5584 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
5586 return test_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
5589 static struct cftype debug_files
[] = {
5591 .name
= "taskcount",
5592 .read_u64
= debug_taskcount_read
,
5596 .name
= "current_css_set",
5597 .read_u64
= current_css_set_read
,
5601 .name
= "current_css_set_refcount",
5602 .read_u64
= current_css_set_refcount_read
,
5606 .name
= "current_css_set_cg_links",
5607 .read_seq_string
= current_css_set_cg_links_read
,
5611 .name
= "cgroup_css_links",
5612 .read_seq_string
= cgroup_css_links_read
,
5616 .name
= "releasable",
5617 .read_u64
= releasable_read
,
5623 struct cgroup_subsys debug_subsys
= {
5625 .css_alloc
= debug_css_alloc
,
5626 .css_free
= debug_css_free
,
5627 .subsys_id
= debug_subsys_id
,
5628 .base_cftypes
= debug_files
,
5630 #endif /* CONFIG_CGROUP_DEBUG */