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>
64 #include <linux/atomic.h>
66 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
67 #define CSS_DEACT_BIAS INT_MIN
70 * cgroup_mutex is the master lock. Any modification to cgroup or its
71 * hierarchy must be performed while holding it.
73 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
74 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
75 * release_agent_path and so on. Modifying requires both cgroup_mutex and
76 * cgroup_root_mutex. Readers can acquire either of the two. This is to
77 * break the following locking order cycle.
79 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
80 * B. namespace_sem -> cgroup_mutex
82 * B happens only through cgroup_show_options() and using cgroup_root_mutex
85 #ifdef CONFIG_PROVE_RCU
86 DEFINE_MUTEX(cgroup_mutex
);
87 EXPORT_SYMBOL_GPL(cgroup_mutex
); /* only for task_subsys_state_check() */
89 static DEFINE_MUTEX(cgroup_mutex
);
92 static DEFINE_MUTEX(cgroup_root_mutex
);
95 * Generate an array of cgroup subsystem pointers. At boot time, this is
96 * populated with the built in subsystems, and modular subsystems are
97 * registered after that. The mutable section of this array is protected by
100 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
101 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
102 static struct cgroup_subsys
*subsys
[CGROUP_SUBSYS_COUNT
] = {
103 #include <linux/cgroup_subsys.h>
107 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
108 * subsystems that are otherwise unattached - it never has more than a
109 * single cgroup, and all tasks are part of that cgroup.
111 static struct cgroupfs_root rootnode
;
114 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
117 struct list_head node
;
118 struct dentry
*dentry
;
122 struct simple_xattrs xattrs
;
126 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
127 * cgroup_subsys->use_id != 0.
129 #define CSS_ID_MAX (65535)
132 * The css to which this ID points. This pointer is set to valid value
133 * after cgroup is populated. If cgroup is removed, this will be NULL.
134 * This pointer is expected to be RCU-safe because destroy()
135 * is called after synchronize_rcu(). But for safe use, css_tryget()
136 * should be used for avoiding race.
138 struct cgroup_subsys_state __rcu
*css
;
144 * Depth in hierarchy which this ID belongs to.
146 unsigned short depth
;
148 * ID is freed by RCU. (and lookup routine is RCU safe.)
150 struct rcu_head rcu_head
;
152 * Hierarchy of CSS ID belongs to.
154 unsigned short stack
[0]; /* Array of Length (depth+1) */
158 * cgroup_event represents events which userspace want to receive.
160 struct cgroup_event
{
162 * Cgroup which the event belongs to.
166 * Control file which the event associated.
170 * eventfd to signal userspace about the event.
172 struct eventfd_ctx
*eventfd
;
174 * Each of these stored in a list by the cgroup.
176 struct list_head list
;
178 * All fields below needed to unregister event when
179 * userspace closes eventfd.
182 wait_queue_head_t
*wqh
;
184 struct work_struct remove
;
187 /* The list of hierarchy roots */
189 static LIST_HEAD(roots
);
190 static int root_count
;
193 * Hierarchy ID allocation and mapping. It follows the same exclusion
194 * rules as other root ops - both cgroup_mutex and cgroup_root_mutex for
195 * writes, either for reads.
197 static DEFINE_IDR(cgroup_hierarchy_idr
);
199 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
200 #define dummytop (&rootnode.top_cgroup)
202 static struct cgroup_name root_cgroup_name
= { .name
= "/" };
204 /* This flag indicates whether tasks in the fork and exit paths should
205 * check for fork/exit handlers to call. This avoids us having to do
206 * extra work in the fork/exit path if none of the subsystems need to
209 static int need_forkexit_callback __read_mostly
;
211 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
212 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
213 struct cftype cfts
[], bool is_add
);
215 static int css_unbias_refcnt(int refcnt
)
217 return refcnt
>= 0 ? refcnt
: refcnt
- CSS_DEACT_BIAS
;
220 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
221 static int css_refcnt(struct cgroup_subsys_state
*css
)
223 int v
= atomic_read(&css
->refcnt
);
225 return css_unbias_refcnt(v
);
228 /* convenient tests for these bits */
229 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
231 return test_bit(CGRP_DEAD
, &cgrp
->flags
);
235 * cgroup_is_descendant - test ancestry
236 * @cgrp: the cgroup to be tested
237 * @ancestor: possible ancestor of @cgrp
239 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
240 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
241 * and @ancestor are accessible.
243 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
246 if (cgrp
== ancestor
)
252 EXPORT_SYMBOL_GPL(cgroup_is_descendant
);
254 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
257 (1 << CGRP_RELEASABLE
) |
258 (1 << CGRP_NOTIFY_ON_RELEASE
);
259 return (cgrp
->flags
& bits
) == bits
;
262 static int notify_on_release(const struct cgroup
*cgrp
)
264 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
268 * for_each_subsys() allows you to iterate on each subsystem attached to
269 * an active hierarchy
271 #define for_each_subsys(_root, _ss) \
272 list_for_each_entry(_ss, &_root->subsys_list, sibling)
274 /* for_each_active_root() allows you to iterate across the active hierarchies */
275 #define for_each_active_root(_root) \
276 list_for_each_entry(_root, &roots, root_list)
278 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
280 return dentry
->d_fsdata
;
283 static inline struct cfent
*__d_cfe(struct dentry
*dentry
)
285 return dentry
->d_fsdata
;
288 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
290 return __d_cfe(dentry
)->type
;
294 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
295 * @cgrp: the cgroup to be checked for liveness
297 * On success, returns true; the mutex should be later unlocked. On
298 * failure returns false with no lock held.
300 static bool cgroup_lock_live_group(struct cgroup
*cgrp
)
302 mutex_lock(&cgroup_mutex
);
303 if (cgroup_is_dead(cgrp
)) {
304 mutex_unlock(&cgroup_mutex
);
310 /* the list of cgroups eligible for automatic release. Protected by
311 * release_list_lock */
312 static LIST_HEAD(release_list
);
313 static DEFINE_RAW_SPINLOCK(release_list_lock
);
314 static void cgroup_release_agent(struct work_struct
*work
);
315 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
316 static void check_for_release(struct cgroup
*cgrp
);
319 * A cgroup can be associated with multiple css_sets as different tasks may
320 * belong to different cgroups on different hierarchies. In the other
321 * direction, a css_set is naturally associated with multiple cgroups.
322 * This M:N relationship is represented by the following link structure
323 * which exists for each association and allows traversing the associations
326 struct cgrp_cset_link
{
327 /* the cgroup and css_set this link associates */
329 struct css_set
*cset
;
331 /* list of cgrp_cset_links anchored at cgrp->cset_links */
332 struct list_head cset_link
;
334 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
335 struct list_head cgrp_link
;
338 /* The default css_set - used by init and its children prior to any
339 * hierarchies being mounted. It contains a pointer to the root state
340 * for each subsystem. Also used to anchor the list of css_sets. Not
341 * reference-counted, to improve performance when child cgroups
342 * haven't been created.
345 static struct css_set init_css_set
;
346 static struct cgrp_cset_link init_cgrp_cset_link
;
348 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
349 struct cgroup_subsys_state
*css
);
351 /* css_set_lock protects the list of css_set objects, and the
352 * chain of tasks off each css_set. Nests outside task->alloc_lock
353 * due to cgroup_iter_start() */
354 static DEFINE_RWLOCK(css_set_lock
);
355 static int css_set_count
;
358 * hash table for cgroup groups. This improves the performance to find
359 * an existing css_set. This hash doesn't (currently) take into
360 * account cgroups in empty hierarchies.
362 #define CSS_SET_HASH_BITS 7
363 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
365 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
368 unsigned long key
= 0UL;
370 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
371 key
+= (unsigned long)css
[i
];
372 key
= (key
>> 16) ^ key
;
377 /* We don't maintain the lists running through each css_set to its
378 * task until after the first call to cgroup_iter_start(). This
379 * reduces the fork()/exit() overhead for people who have cgroups
380 * compiled into their kernel but not actually in use */
381 static int use_task_css_set_links __read_mostly
;
383 static void __put_css_set(struct css_set
*cset
, int taskexit
)
385 struct cgrp_cset_link
*link
, *tmp_link
;
388 * Ensure that the refcount doesn't hit zero while any readers
389 * can see it. Similar to atomic_dec_and_lock(), but for an
392 if (atomic_add_unless(&cset
->refcount
, -1, 1))
394 write_lock(&css_set_lock
);
395 if (!atomic_dec_and_test(&cset
->refcount
)) {
396 write_unlock(&css_set_lock
);
400 /* This css_set is dead. unlink it and release cgroup refcounts */
401 hash_del(&cset
->hlist
);
404 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
405 struct cgroup
*cgrp
= link
->cgrp
;
407 list_del(&link
->cset_link
);
408 list_del(&link
->cgrp_link
);
410 /* @cgrp can't go away while we're holding css_set_lock */
411 if (atomic_dec_and_test(&cgrp
->count
) &&
412 notify_on_release(cgrp
)) {
414 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
415 check_for_release(cgrp
);
421 write_unlock(&css_set_lock
);
422 kfree_rcu(cset
, rcu_head
);
426 * refcounted get/put for css_set objects
428 static inline void get_css_set(struct css_set
*cset
)
430 atomic_inc(&cset
->refcount
);
433 static inline void put_css_set(struct css_set
*cset
)
435 __put_css_set(cset
, 0);
438 static inline void put_css_set_taskexit(struct css_set
*cset
)
440 __put_css_set(cset
, 1);
444 * compare_css_sets - helper function for find_existing_css_set().
445 * @cset: candidate css_set being tested
446 * @old_cset: existing css_set for a task
447 * @new_cgrp: cgroup that's being entered by the task
448 * @template: desired set of css pointers in css_set (pre-calculated)
450 * Returns true if "cg" matches "old_cg" except for the hierarchy
451 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
453 static bool compare_css_sets(struct css_set
*cset
,
454 struct css_set
*old_cset
,
455 struct cgroup
*new_cgrp
,
456 struct cgroup_subsys_state
*template[])
458 struct list_head
*l1
, *l2
;
460 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
))) {
461 /* Not all subsystems matched */
466 * Compare cgroup pointers in order to distinguish between
467 * different cgroups in heirarchies with no subsystems. We
468 * could get by with just this check alone (and skip the
469 * memcmp above) but on most setups the memcmp check will
470 * avoid the need for this more expensive check on almost all
474 l1
= &cset
->cgrp_links
;
475 l2
= &old_cset
->cgrp_links
;
477 struct cgrp_cset_link
*link1
, *link2
;
478 struct cgroup
*cgrp1
, *cgrp2
;
482 /* See if we reached the end - both lists are equal length. */
483 if (l1
== &cset
->cgrp_links
) {
484 BUG_ON(l2
!= &old_cset
->cgrp_links
);
487 BUG_ON(l2
== &old_cset
->cgrp_links
);
489 /* Locate the cgroups associated with these links. */
490 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
491 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
494 /* Hierarchies should be linked in the same order. */
495 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
498 * If this hierarchy is the hierarchy of the cgroup
499 * that's changing, then we need to check that this
500 * css_set points to the new cgroup; if it's any other
501 * hierarchy, then this css_set should point to the
502 * same cgroup as the old css_set.
504 if (cgrp1
->root
== new_cgrp
->root
) {
505 if (cgrp1
!= new_cgrp
)
516 * find_existing_css_set() is a helper for
517 * find_css_set(), and checks to see whether an existing
518 * css_set is suitable.
520 * oldcg: the cgroup group that we're using before the cgroup
523 * cgrp: the cgroup that we're moving into
525 * template: location in which to build the desired set of subsystem
526 * state objects for the new cgroup group
528 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
530 struct cgroup_subsys_state
*template[])
533 struct cgroupfs_root
*root
= cgrp
->root
;
534 struct css_set
*cset
;
538 * Build the set of subsystem state objects that we want to see in the
539 * new css_set. while subsystems can change globally, the entries here
540 * won't change, so no need for locking.
542 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
543 if (root
->subsys_mask
& (1UL << i
)) {
544 /* Subsystem is in this hierarchy. So we want
545 * the subsystem state from the new
547 template[i
] = cgrp
->subsys
[i
];
549 /* Subsystem is not in this hierarchy, so we
550 * don't want to change the subsystem state */
551 template[i
] = old_cset
->subsys
[i
];
555 key
= css_set_hash(template);
556 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
557 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
560 /* This css_set matches what we need */
564 /* No existing cgroup group matched */
568 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
570 struct cgrp_cset_link
*link
, *tmp_link
;
572 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
573 list_del(&link
->cset_link
);
579 * allocate_cgrp_cset_links - allocate cgrp_cset_links
580 * @count: the number of links to allocate
581 * @tmp_links: list_head the allocated links are put on
583 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
584 * through ->cset_link. Returns 0 on success or -errno.
586 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
588 struct cgrp_cset_link
*link
;
591 INIT_LIST_HEAD(tmp_links
);
593 for (i
= 0; i
< count
; i
++) {
594 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
596 free_cgrp_cset_links(tmp_links
);
599 list_add(&link
->cset_link
, tmp_links
);
605 * link_css_set - a helper function to link a css_set to a cgroup
606 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
607 * @cset: the css_set to be linked
608 * @cgrp: the destination cgroup
610 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
613 struct cgrp_cset_link
*link
;
615 BUG_ON(list_empty(tmp_links
));
616 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
619 atomic_inc(&cgrp
->count
);
620 list_move(&link
->cset_link
, &cgrp
->cset_links
);
622 * Always add links to the tail of the list so that the list
623 * is sorted by order of hierarchy creation
625 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
629 * find_css_set() takes an existing cgroup group and a
630 * cgroup object, and returns a css_set object that's
631 * equivalent to the old group, but with the given cgroup
632 * substituted into the appropriate hierarchy. Must be called with
635 static struct css_set
*find_css_set(struct css_set
*old_cset
,
638 struct css_set
*cset
;
639 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
640 struct list_head tmp_links
;
641 struct cgrp_cset_link
*link
;
644 /* First see if we already have a cgroup group that matches
646 read_lock(&css_set_lock
);
647 cset
= find_existing_css_set(old_cset
, cgrp
, template);
650 read_unlock(&css_set_lock
);
655 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
659 /* Allocate all the cgrp_cset_link objects that we'll need */
660 if (allocate_cgrp_cset_links(root_count
, &tmp_links
) < 0) {
665 atomic_set(&cset
->refcount
, 1);
666 INIT_LIST_HEAD(&cset
->cgrp_links
);
667 INIT_LIST_HEAD(&cset
->tasks
);
668 INIT_HLIST_NODE(&cset
->hlist
);
670 /* Copy the set of subsystem state objects generated in
671 * find_existing_css_set() */
672 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
674 write_lock(&css_set_lock
);
675 /* Add reference counts and links from the new css_set. */
676 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
677 struct cgroup
*c
= link
->cgrp
;
679 if (c
->root
== cgrp
->root
)
681 link_css_set(&tmp_links
, cset
, c
);
684 BUG_ON(!list_empty(&tmp_links
));
688 /* Add this cgroup group to the hash table */
689 key
= css_set_hash(cset
->subsys
);
690 hash_add(css_set_table
, &cset
->hlist
, key
);
692 write_unlock(&css_set_lock
);
698 * Return the cgroup for "task" from the given hierarchy. Must be
699 * called with cgroup_mutex held.
701 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
702 struct cgroupfs_root
*root
)
704 struct css_set
*cset
;
705 struct cgroup
*res
= NULL
;
707 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
708 read_lock(&css_set_lock
);
710 * No need to lock the task - since we hold cgroup_mutex the
711 * task can't change groups, so the only thing that can happen
712 * is that it exits and its css is set back to init_css_set.
714 cset
= task
->cgroups
;
715 if (cset
== &init_css_set
) {
716 res
= &root
->top_cgroup
;
718 struct cgrp_cset_link
*link
;
720 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
721 struct cgroup
*c
= link
->cgrp
;
723 if (c
->root
== root
) {
729 read_unlock(&css_set_lock
);
735 * There is one global cgroup mutex. We also require taking
736 * task_lock() when dereferencing a task's cgroup subsys pointers.
737 * See "The task_lock() exception", at the end of this comment.
739 * A task must hold cgroup_mutex to modify cgroups.
741 * Any task can increment and decrement the count field without lock.
742 * So in general, code holding cgroup_mutex can't rely on the count
743 * field not changing. However, if the count goes to zero, then only
744 * cgroup_attach_task() can increment it again. Because a count of zero
745 * means that no tasks are currently attached, therefore there is no
746 * way a task attached to that cgroup can fork (the other way to
747 * increment the count). So code holding cgroup_mutex can safely
748 * assume that if the count is zero, it will stay zero. Similarly, if
749 * a task holds cgroup_mutex on a cgroup with zero count, it
750 * knows that the cgroup won't be removed, as cgroup_rmdir()
753 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
754 * (usually) take cgroup_mutex. These are the two most performance
755 * critical pieces of code here. The exception occurs on cgroup_exit(),
756 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
757 * is taken, and if the cgroup count is zero, a usermode call made
758 * to the release agent with the name of the cgroup (path relative to
759 * the root of cgroup file system) as the argument.
761 * A cgroup can only be deleted if both its 'count' of using tasks
762 * is zero, and its list of 'children' cgroups is empty. Since all
763 * tasks in the system use _some_ cgroup, and since there is always at
764 * least one task in the system (init, pid == 1), therefore, top_cgroup
765 * always has either children cgroups and/or using tasks. So we don't
766 * need a special hack to ensure that top_cgroup cannot be deleted.
768 * The task_lock() exception
770 * The need for this exception arises from the action of
771 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
772 * another. It does so using cgroup_mutex, however there are
773 * several performance critical places that need to reference
774 * task->cgroup without the expense of grabbing a system global
775 * mutex. Therefore except as noted below, when dereferencing or, as
776 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
777 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
778 * the task_struct routinely used for such matters.
780 * P.S. One more locking exception. RCU is used to guard the
781 * update of a tasks cgroup pointer by cgroup_attach_task()
785 * A couple of forward declarations required, due to cyclic reference loop:
786 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
787 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
791 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
792 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, unsigned int);
793 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
794 static int cgroup_populate_dir(struct cgroup
*cgrp
, bool base_files
,
795 unsigned long subsys_mask
);
796 static const struct inode_operations cgroup_dir_inode_operations
;
797 static const struct file_operations proc_cgroupstats_operations
;
799 static struct backing_dev_info cgroup_backing_dev_info
= {
801 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
804 static int alloc_css_id(struct cgroup_subsys
*ss
,
805 struct cgroup
*parent
, struct cgroup
*child
);
807 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
809 struct inode
*inode
= new_inode(sb
);
812 inode
->i_ino
= get_next_ino();
813 inode
->i_mode
= mode
;
814 inode
->i_uid
= current_fsuid();
815 inode
->i_gid
= current_fsgid();
816 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
817 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
822 static struct cgroup_name
*cgroup_alloc_name(struct dentry
*dentry
)
824 struct cgroup_name
*name
;
826 name
= kmalloc(sizeof(*name
) + dentry
->d_name
.len
+ 1, GFP_KERNEL
);
829 strcpy(name
->name
, dentry
->d_name
.name
);
833 static void cgroup_free_fn(struct work_struct
*work
)
835 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, free_work
);
836 struct cgroup_subsys
*ss
;
838 mutex_lock(&cgroup_mutex
);
840 * Release the subsystem state objects.
842 for_each_subsys(cgrp
->root
, ss
)
845 cgrp
->root
->number_of_cgroups
--;
846 mutex_unlock(&cgroup_mutex
);
849 * We get a ref to the parent's dentry, and put the ref when
850 * this cgroup is being freed, so it's guaranteed that the
851 * parent won't be destroyed before its children.
853 dput(cgrp
->parent
->dentry
);
855 ida_simple_remove(&cgrp
->root
->cgroup_ida
, cgrp
->id
);
858 * Drop the active superblock reference that we took when we
859 * created the cgroup. This will free cgrp->root, if we are
860 * holding the last reference to @sb.
862 deactivate_super(cgrp
->root
->sb
);
865 * if we're getting rid of the cgroup, refcount should ensure
866 * that there are no pidlists left.
868 BUG_ON(!list_empty(&cgrp
->pidlists
));
870 simple_xattrs_free(&cgrp
->xattrs
);
872 kfree(rcu_dereference_raw(cgrp
->name
));
876 static void cgroup_free_rcu(struct rcu_head
*head
)
878 struct cgroup
*cgrp
= container_of(head
, struct cgroup
, rcu_head
);
880 schedule_work(&cgrp
->free_work
);
883 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
885 /* is dentry a directory ? if so, kfree() associated cgroup */
886 if (S_ISDIR(inode
->i_mode
)) {
887 struct cgroup
*cgrp
= dentry
->d_fsdata
;
889 BUG_ON(!(cgroup_is_dead(cgrp
)));
890 call_rcu(&cgrp
->rcu_head
, cgroup_free_rcu
);
892 struct cfent
*cfe
= __d_cfe(dentry
);
893 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
895 WARN_ONCE(!list_empty(&cfe
->node
) &&
896 cgrp
!= &cgrp
->root
->top_cgroup
,
897 "cfe still linked for %s\n", cfe
->type
->name
);
898 simple_xattrs_free(&cfe
->xattrs
);
904 static int cgroup_delete(const struct dentry
*d
)
909 static void remove_dir(struct dentry
*d
)
911 struct dentry
*parent
= dget(d
->d_parent
);
914 simple_rmdir(parent
->d_inode
, d
);
918 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
922 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
923 lockdep_assert_held(&cgroup_mutex
);
926 * If we're doing cleanup due to failure of cgroup_create(),
927 * the corresponding @cfe may not exist.
929 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
930 struct dentry
*d
= cfe
->dentry
;
932 if (cft
&& cfe
->type
!= cft
)
937 simple_unlink(cgrp
->dentry
->d_inode
, d
);
938 list_del_init(&cfe
->node
);
946 * cgroup_clear_directory - selective removal of base and subsystem files
947 * @dir: directory containing the files
948 * @base_files: true if the base files should be removed
949 * @subsys_mask: mask of the subsystem ids whose files should be removed
951 static void cgroup_clear_directory(struct dentry
*dir
, bool base_files
,
952 unsigned long subsys_mask
)
954 struct cgroup
*cgrp
= __d_cgrp(dir
);
955 struct cgroup_subsys
*ss
;
957 for_each_subsys(cgrp
->root
, ss
) {
958 struct cftype_set
*set
;
959 if (!test_bit(ss
->subsys_id
, &subsys_mask
))
961 list_for_each_entry(set
, &ss
->cftsets
, node
)
962 cgroup_addrm_files(cgrp
, NULL
, set
->cfts
, false);
965 while (!list_empty(&cgrp
->files
))
966 cgroup_rm_file(cgrp
, NULL
);
971 * NOTE : the dentry must have been dget()'ed
973 static void cgroup_d_remove_dir(struct dentry
*dentry
)
975 struct dentry
*parent
;
976 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
978 cgroup_clear_directory(dentry
, true, root
->subsys_mask
);
980 parent
= dentry
->d_parent
;
981 spin_lock(&parent
->d_lock
);
982 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
983 list_del_init(&dentry
->d_u
.d_child
);
984 spin_unlock(&dentry
->d_lock
);
985 spin_unlock(&parent
->d_lock
);
990 * Call with cgroup_mutex held. Drops reference counts on modules, including
991 * any duplicate ones that parse_cgroupfs_options took. If this function
992 * returns an error, no reference counts are touched.
994 static int rebind_subsystems(struct cgroupfs_root
*root
,
995 unsigned long final_subsys_mask
)
997 unsigned long added_mask
, removed_mask
;
998 struct cgroup
*cgrp
= &root
->top_cgroup
;
1001 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1002 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1004 removed_mask
= root
->actual_subsys_mask
& ~final_subsys_mask
;
1005 added_mask
= final_subsys_mask
& ~root
->actual_subsys_mask
;
1006 /* Check that any added subsystems are currently free */
1007 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1008 unsigned long bit
= 1UL << i
;
1009 struct cgroup_subsys
*ss
= subsys
[i
];
1010 if (!(bit
& added_mask
))
1013 * Nobody should tell us to do a subsys that doesn't exist:
1014 * parse_cgroupfs_options should catch that case and refcounts
1015 * ensure that subsystems won't disappear once selected.
1018 if (ss
->root
!= &rootnode
) {
1019 /* Subsystem isn't free */
1024 /* Currently we don't handle adding/removing subsystems when
1025 * any child cgroups exist. This is theoretically supportable
1026 * but involves complex error handling, so it's being left until
1028 if (root
->number_of_cgroups
> 1)
1031 /* Process each subsystem */
1032 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1033 struct cgroup_subsys
*ss
= subsys
[i
];
1034 unsigned long bit
= 1UL << i
;
1035 if (bit
& added_mask
) {
1036 /* We're binding this subsystem to this hierarchy */
1038 BUG_ON(cgrp
->subsys
[i
]);
1039 BUG_ON(!dummytop
->subsys
[i
]);
1040 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
1041 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
1042 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1043 list_move(&ss
->sibling
, &root
->subsys_list
);
1047 /* refcount was already taken, and we're keeping it */
1048 } else if (bit
& removed_mask
) {
1049 /* We're removing this subsystem */
1051 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1052 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1055 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1056 cgrp
->subsys
[i
] = NULL
;
1057 subsys
[i
]->root
= &rootnode
;
1058 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1059 /* subsystem is now free - drop reference on module */
1060 module_put(ss
->module
);
1061 } else if (bit
& final_subsys_mask
) {
1062 /* Subsystem state should already exist */
1064 BUG_ON(!cgrp
->subsys
[i
]);
1066 * a refcount was taken, but we already had one, so
1067 * drop the extra reference.
1069 module_put(ss
->module
);
1070 #ifdef CONFIG_MODULE_UNLOAD
1071 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1074 /* Subsystem state shouldn't exist */
1075 BUG_ON(cgrp
->subsys
[i
]);
1078 root
->subsys_mask
= root
->actual_subsys_mask
= final_subsys_mask
;
1083 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1085 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1086 struct cgroup_subsys
*ss
;
1088 mutex_lock(&cgroup_root_mutex
);
1089 for_each_subsys(root
, ss
)
1090 seq_printf(seq
, ",%s", ss
->name
);
1091 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
)
1092 seq_puts(seq
, ",sane_behavior");
1093 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1094 seq_puts(seq
, ",noprefix");
1095 if (root
->flags
& CGRP_ROOT_XATTR
)
1096 seq_puts(seq
, ",xattr");
1097 if (strlen(root
->release_agent_path
))
1098 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1099 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
))
1100 seq_puts(seq
, ",clone_children");
1101 if (strlen(root
->name
))
1102 seq_printf(seq
, ",name=%s", root
->name
);
1103 mutex_unlock(&cgroup_root_mutex
);
1107 struct cgroup_sb_opts
{
1108 unsigned long subsys_mask
;
1109 unsigned long flags
;
1110 char *release_agent
;
1111 bool cpuset_clone_children
;
1113 /* User explicitly requested empty subsystem */
1116 struct cgroupfs_root
*new_root
;
1121 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1122 * with cgroup_mutex held to protect the subsys[] array. This function takes
1123 * refcounts on subsystems to be used, unless it returns error, in which case
1124 * no refcounts are taken.
1126 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1128 char *token
, *o
= data
;
1129 bool all_ss
= false, one_ss
= false;
1130 unsigned long mask
= (unsigned long)-1;
1132 bool module_pin_failed
= false;
1134 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1136 #ifdef CONFIG_CPUSETS
1137 mask
= ~(1UL << cpuset_subsys_id
);
1140 memset(opts
, 0, sizeof(*opts
));
1142 while ((token
= strsep(&o
, ",")) != NULL
) {
1145 if (!strcmp(token
, "none")) {
1146 /* Explicitly have no subsystems */
1150 if (!strcmp(token
, "all")) {
1151 /* Mutually exclusive option 'all' + subsystem name */
1157 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1158 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1161 if (!strcmp(token
, "noprefix")) {
1162 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1165 if (!strcmp(token
, "clone_children")) {
1166 opts
->cpuset_clone_children
= true;
1169 if (!strcmp(token
, "xattr")) {
1170 opts
->flags
|= CGRP_ROOT_XATTR
;
1173 if (!strncmp(token
, "release_agent=", 14)) {
1174 /* Specifying two release agents is forbidden */
1175 if (opts
->release_agent
)
1177 opts
->release_agent
=
1178 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1179 if (!opts
->release_agent
)
1183 if (!strncmp(token
, "name=", 5)) {
1184 const char *name
= token
+ 5;
1185 /* Can't specify an empty name */
1188 /* Must match [\w.-]+ */
1189 for (i
= 0; i
< strlen(name
); i
++) {
1193 if ((c
== '.') || (c
== '-') || (c
== '_'))
1197 /* Specifying two names is forbidden */
1200 opts
->name
= kstrndup(name
,
1201 MAX_CGROUP_ROOT_NAMELEN
- 1,
1209 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1210 struct cgroup_subsys
*ss
= subsys
[i
];
1213 if (strcmp(token
, ss
->name
))
1218 /* Mutually exclusive option 'all' + subsystem name */
1221 set_bit(i
, &opts
->subsys_mask
);
1226 if (i
== CGROUP_SUBSYS_COUNT
)
1231 * If the 'all' option was specified select all the subsystems,
1232 * otherwise if 'none', 'name=' and a subsystem name options
1233 * were not specified, let's default to 'all'
1235 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
)) {
1236 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1237 struct cgroup_subsys
*ss
= subsys
[i
];
1242 set_bit(i
, &opts
->subsys_mask
);
1246 /* Consistency checks */
1248 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1249 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1251 if (opts
->flags
& CGRP_ROOT_NOPREFIX
) {
1252 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1256 if (opts
->cpuset_clone_children
) {
1257 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1263 * Option noprefix was introduced just for backward compatibility
1264 * with the old cpuset, so we allow noprefix only if mounting just
1265 * the cpuset subsystem.
1267 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1271 /* Can't specify "none" and some subsystems */
1272 if (opts
->subsys_mask
&& opts
->none
)
1276 * We either have to specify by name or by subsystems. (So all
1277 * empty hierarchies must have a name).
1279 if (!opts
->subsys_mask
&& !opts
->name
)
1283 * Grab references on all the modules we'll need, so the subsystems
1284 * don't dance around before rebind_subsystems attaches them. This may
1285 * take duplicate reference counts on a subsystem that's already used,
1286 * but rebind_subsystems handles this case.
1288 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1289 unsigned long bit
= 1UL << i
;
1291 if (!(bit
& opts
->subsys_mask
))
1293 if (!try_module_get(subsys
[i
]->module
)) {
1294 module_pin_failed
= true;
1298 if (module_pin_failed
) {
1300 * oops, one of the modules was going away. this means that we
1301 * raced with a module_delete call, and to the user this is
1302 * essentially a "subsystem doesn't exist" case.
1304 for (i
--; i
>= 0; i
--) {
1305 /* drop refcounts only on the ones we took */
1306 unsigned long bit
= 1UL << i
;
1308 if (!(bit
& opts
->subsys_mask
))
1310 module_put(subsys
[i
]->module
);
1318 static void drop_parsed_module_refcounts(unsigned long subsys_mask
)
1321 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1322 unsigned long bit
= 1UL << i
;
1324 if (!(bit
& subsys_mask
))
1326 module_put(subsys
[i
]->module
);
1330 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1333 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1334 struct cgroup
*cgrp
= &root
->top_cgroup
;
1335 struct cgroup_sb_opts opts
;
1336 unsigned long added_mask
, removed_mask
;
1338 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1339 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1343 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1344 mutex_lock(&cgroup_mutex
);
1345 mutex_lock(&cgroup_root_mutex
);
1347 /* See what subsystems are wanted */
1348 ret
= parse_cgroupfs_options(data
, &opts
);
1352 if (opts
.subsys_mask
!= root
->actual_subsys_mask
|| opts
.release_agent
)
1353 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1354 task_tgid_nr(current
), current
->comm
);
1356 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1357 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1359 /* Don't allow flags or name to change at remount */
1360 if (opts
.flags
!= root
->flags
||
1361 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1363 drop_parsed_module_refcounts(opts
.subsys_mask
);
1368 * Clear out the files of subsystems that should be removed, do
1369 * this before rebind_subsystems, since rebind_subsystems may
1370 * change this hierarchy's subsys_list.
1372 cgroup_clear_directory(cgrp
->dentry
, false, removed_mask
);
1374 ret
= rebind_subsystems(root
, opts
.subsys_mask
);
1376 /* rebind_subsystems failed, re-populate the removed files */
1377 cgroup_populate_dir(cgrp
, false, removed_mask
);
1378 drop_parsed_module_refcounts(opts
.subsys_mask
);
1382 /* re-populate subsystem files */
1383 cgroup_populate_dir(cgrp
, false, added_mask
);
1385 if (opts
.release_agent
)
1386 strcpy(root
->release_agent_path
, opts
.release_agent
);
1388 kfree(opts
.release_agent
);
1390 mutex_unlock(&cgroup_root_mutex
);
1391 mutex_unlock(&cgroup_mutex
);
1392 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1396 static const struct super_operations cgroup_ops
= {
1397 .statfs
= simple_statfs
,
1398 .drop_inode
= generic_delete_inode
,
1399 .show_options
= cgroup_show_options
,
1400 .remount_fs
= cgroup_remount
,
1403 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1405 INIT_LIST_HEAD(&cgrp
->sibling
);
1406 INIT_LIST_HEAD(&cgrp
->children
);
1407 INIT_LIST_HEAD(&cgrp
->files
);
1408 INIT_LIST_HEAD(&cgrp
->cset_links
);
1409 INIT_LIST_HEAD(&cgrp
->allcg_node
);
1410 INIT_LIST_HEAD(&cgrp
->release_list
);
1411 INIT_LIST_HEAD(&cgrp
->pidlists
);
1412 INIT_WORK(&cgrp
->free_work
, cgroup_free_fn
);
1413 mutex_init(&cgrp
->pidlist_mutex
);
1414 INIT_LIST_HEAD(&cgrp
->event_list
);
1415 spin_lock_init(&cgrp
->event_list_lock
);
1416 simple_xattrs_init(&cgrp
->xattrs
);
1419 static void init_cgroup_root(struct cgroupfs_root
*root
)
1421 struct cgroup
*cgrp
= &root
->top_cgroup
;
1423 INIT_LIST_HEAD(&root
->subsys_list
);
1424 INIT_LIST_HEAD(&root
->root_list
);
1425 INIT_LIST_HEAD(&root
->allcg_list
);
1426 root
->number_of_cgroups
= 1;
1428 cgrp
->name
= &root_cgroup_name
;
1429 init_cgroup_housekeeping(cgrp
);
1430 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
1433 static int cgroup_init_root_id(struct cgroupfs_root
*root
)
1437 lockdep_assert_held(&cgroup_mutex
);
1438 lockdep_assert_held(&cgroup_root_mutex
);
1440 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 2, 0, GFP_KERNEL
);
1444 root
->hierarchy_id
= id
;
1448 static void cgroup_exit_root_id(struct cgroupfs_root
*root
)
1450 lockdep_assert_held(&cgroup_mutex
);
1451 lockdep_assert_held(&cgroup_root_mutex
);
1453 if (root
->hierarchy_id
) {
1454 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1455 root
->hierarchy_id
= 0;
1459 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1461 struct cgroup_sb_opts
*opts
= data
;
1462 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1464 /* If we asked for a name then it must match */
1465 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1469 * If we asked for subsystems (or explicitly for no
1470 * subsystems) then they must match
1472 if ((opts
->subsys_mask
|| opts
->none
)
1473 && (opts
->subsys_mask
!= root
->subsys_mask
))
1479 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1481 struct cgroupfs_root
*root
;
1483 if (!opts
->subsys_mask
&& !opts
->none
)
1486 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1488 return ERR_PTR(-ENOMEM
);
1490 init_cgroup_root(root
);
1492 root
->subsys_mask
= opts
->subsys_mask
;
1493 root
->flags
= opts
->flags
;
1494 ida_init(&root
->cgroup_ida
);
1495 if (opts
->release_agent
)
1496 strcpy(root
->release_agent_path
, opts
->release_agent
);
1498 strcpy(root
->name
, opts
->name
);
1499 if (opts
->cpuset_clone_children
)
1500 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1504 static void cgroup_free_root(struct cgroupfs_root
*root
)
1507 /* hierarhcy ID shoulid already have been released */
1508 WARN_ON_ONCE(root
->hierarchy_id
);
1510 ida_destroy(&root
->cgroup_ida
);
1515 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1518 struct cgroup_sb_opts
*opts
= data
;
1520 /* If we don't have a new root, we can't set up a new sb */
1521 if (!opts
->new_root
)
1524 BUG_ON(!opts
->subsys_mask
&& !opts
->none
);
1526 ret
= set_anon_super(sb
, NULL
);
1530 sb
->s_fs_info
= opts
->new_root
;
1531 opts
->new_root
->sb
= sb
;
1533 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1534 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1535 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1536 sb
->s_op
= &cgroup_ops
;
1541 static int cgroup_get_rootdir(struct super_block
*sb
)
1543 static const struct dentry_operations cgroup_dops
= {
1544 .d_iput
= cgroup_diput
,
1545 .d_delete
= cgroup_delete
,
1548 struct inode
*inode
=
1549 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1554 inode
->i_fop
= &simple_dir_operations
;
1555 inode
->i_op
= &cgroup_dir_inode_operations
;
1556 /* directories start off with i_nlink == 2 (for "." entry) */
1558 sb
->s_root
= d_make_root(inode
);
1561 /* for everything else we want ->d_op set */
1562 sb
->s_d_op
= &cgroup_dops
;
1566 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1567 int flags
, const char *unused_dev_name
,
1570 struct cgroup_sb_opts opts
;
1571 struct cgroupfs_root
*root
;
1573 struct super_block
*sb
;
1574 struct cgroupfs_root
*new_root
;
1575 struct inode
*inode
;
1577 /* First find the desired set of subsystems */
1578 mutex_lock(&cgroup_mutex
);
1579 ret
= parse_cgroupfs_options(data
, &opts
);
1580 mutex_unlock(&cgroup_mutex
);
1585 * Allocate a new cgroup root. We may not need it if we're
1586 * reusing an existing hierarchy.
1588 new_root
= cgroup_root_from_opts(&opts
);
1589 if (IS_ERR(new_root
)) {
1590 ret
= PTR_ERR(new_root
);
1593 opts
.new_root
= new_root
;
1595 /* Locate an existing or new sb for this hierarchy */
1596 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, 0, &opts
);
1599 cgroup_free_root(opts
.new_root
);
1603 root
= sb
->s_fs_info
;
1605 if (root
== opts
.new_root
) {
1606 /* We used the new root structure, so this is a new hierarchy */
1607 struct list_head tmp_links
;
1608 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1609 struct cgroupfs_root
*existing_root
;
1610 const struct cred
*cred
;
1612 struct css_set
*cset
;
1614 BUG_ON(sb
->s_root
!= NULL
);
1616 ret
= cgroup_get_rootdir(sb
);
1618 goto drop_new_super
;
1619 inode
= sb
->s_root
->d_inode
;
1621 mutex_lock(&inode
->i_mutex
);
1622 mutex_lock(&cgroup_mutex
);
1623 mutex_lock(&cgroup_root_mutex
);
1625 /* Check for name clashes with existing mounts */
1627 if (strlen(root
->name
))
1628 for_each_active_root(existing_root
)
1629 if (!strcmp(existing_root
->name
, root
->name
))
1633 * We're accessing css_set_count without locking
1634 * css_set_lock here, but that's OK - it can only be
1635 * increased by someone holding cgroup_lock, and
1636 * that's us. The worst that can happen is that we
1637 * have some link structures left over
1639 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1643 ret
= cgroup_init_root_id(root
);
1647 ret
= rebind_subsystems(root
, root
->subsys_mask
);
1648 if (ret
== -EBUSY
) {
1649 free_cgrp_cset_links(&tmp_links
);
1653 * There must be no failure case after here, since rebinding
1654 * takes care of subsystems' refcounts, which are explicitly
1655 * dropped in the failure exit path.
1658 /* EBUSY should be the only error here */
1661 list_add(&root
->root_list
, &roots
);
1664 sb
->s_root
->d_fsdata
= root_cgrp
;
1665 root
->top_cgroup
.dentry
= sb
->s_root
;
1667 /* Link the top cgroup in this hierarchy into all
1668 * the css_set objects */
1669 write_lock(&css_set_lock
);
1670 hash_for_each(css_set_table
, i
, cset
, hlist
)
1671 link_css_set(&tmp_links
, cset
, root_cgrp
);
1672 write_unlock(&css_set_lock
);
1674 free_cgrp_cset_links(&tmp_links
);
1676 BUG_ON(!list_empty(&root_cgrp
->children
));
1677 BUG_ON(root
->number_of_cgroups
!= 1);
1679 cred
= override_creds(&init_cred
);
1680 cgroup_populate_dir(root_cgrp
, true, root
->subsys_mask
);
1682 mutex_unlock(&cgroup_root_mutex
);
1683 mutex_unlock(&cgroup_mutex
);
1684 mutex_unlock(&inode
->i_mutex
);
1687 * We re-used an existing hierarchy - the new root (if
1688 * any) is not needed
1690 cgroup_free_root(opts
.new_root
);
1692 if (((root
->flags
| opts
.flags
) & CGRP_ROOT_SANE_BEHAVIOR
) &&
1693 root
->flags
!= opts
.flags
) {
1694 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1696 goto drop_new_super
;
1699 /* no subsys rebinding, so refcounts don't change */
1700 drop_parsed_module_refcounts(opts
.subsys_mask
);
1703 kfree(opts
.release_agent
);
1705 return dget(sb
->s_root
);
1708 cgroup_exit_root_id(root
);
1709 mutex_unlock(&cgroup_root_mutex
);
1710 mutex_unlock(&cgroup_mutex
);
1711 mutex_unlock(&inode
->i_mutex
);
1713 deactivate_locked_super(sb
);
1715 drop_parsed_module_refcounts(opts
.subsys_mask
);
1717 kfree(opts
.release_agent
);
1719 return ERR_PTR(ret
);
1722 static void cgroup_kill_sb(struct super_block
*sb
) {
1723 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1724 struct cgroup
*cgrp
= &root
->top_cgroup
;
1725 struct cgrp_cset_link
*link
, *tmp_link
;
1730 BUG_ON(root
->number_of_cgroups
!= 1);
1731 BUG_ON(!list_empty(&cgrp
->children
));
1733 mutex_lock(&cgroup_mutex
);
1734 mutex_lock(&cgroup_root_mutex
);
1736 /* Rebind all subsystems back to the default hierarchy */
1737 ret
= rebind_subsystems(root
, 0);
1738 /* Shouldn't be able to fail ... */
1742 * Release all the links from cset_links to this hierarchy's
1745 write_lock(&css_set_lock
);
1747 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1748 list_del(&link
->cset_link
);
1749 list_del(&link
->cgrp_link
);
1752 write_unlock(&css_set_lock
);
1754 if (!list_empty(&root
->root_list
)) {
1755 list_del(&root
->root_list
);
1759 cgroup_exit_root_id(root
);
1761 mutex_unlock(&cgroup_root_mutex
);
1762 mutex_unlock(&cgroup_mutex
);
1764 simple_xattrs_free(&cgrp
->xattrs
);
1766 kill_litter_super(sb
);
1767 cgroup_free_root(root
);
1770 static struct file_system_type cgroup_fs_type
= {
1772 .mount
= cgroup_mount
,
1773 .kill_sb
= cgroup_kill_sb
,
1776 static struct kobject
*cgroup_kobj
;
1779 * cgroup_path - generate the path of a cgroup
1780 * @cgrp: the cgroup in question
1781 * @buf: the buffer to write the path into
1782 * @buflen: the length of the buffer
1784 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1786 * We can't generate cgroup path using dentry->d_name, as accessing
1787 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1788 * inode's i_mutex, while on the other hand cgroup_path() can be called
1789 * with some irq-safe spinlocks held.
1791 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1793 int ret
= -ENAMETOOLONG
;
1796 if (!cgrp
->parent
) {
1797 if (strlcpy(buf
, "/", buflen
) >= buflen
)
1798 return -ENAMETOOLONG
;
1802 start
= buf
+ buflen
- 1;
1807 const char *name
= cgroup_name(cgrp
);
1811 if ((start
-= len
) < buf
)
1813 memcpy(start
, name
, len
);
1819 cgrp
= cgrp
->parent
;
1820 } while (cgrp
->parent
);
1822 memmove(buf
, start
, buf
+ buflen
- start
);
1827 EXPORT_SYMBOL_GPL(cgroup_path
);
1830 * task_cgroup_path_from_hierarchy - cgroup path of a task on a hierarchy
1831 * @task: target task
1832 * @hierarchy_id: the hierarchy to look up @task's cgroup from
1833 * @buf: the buffer to write the path into
1834 * @buflen: the length of the buffer
1836 * Determine @task's cgroup on the hierarchy specified by @hierarchy_id and
1837 * copy its path into @buf. This function grabs cgroup_mutex and shouldn't
1838 * be used inside locks used by cgroup controller callbacks.
1840 int task_cgroup_path_from_hierarchy(struct task_struct
*task
, int hierarchy_id
,
1841 char *buf
, size_t buflen
)
1843 struct cgroupfs_root
*root
;
1844 struct cgroup
*cgrp
= NULL
;
1847 mutex_lock(&cgroup_mutex
);
1849 root
= idr_find(&cgroup_hierarchy_idr
, hierarchy_id
);
1851 cgrp
= task_cgroup_from_root(task
, root
);
1852 ret
= cgroup_path(cgrp
, buf
, buflen
);
1855 mutex_unlock(&cgroup_mutex
);
1859 EXPORT_SYMBOL_GPL(task_cgroup_path_from_hierarchy
);
1862 * Control Group taskset
1864 struct task_and_cgroup
{
1865 struct task_struct
*task
;
1866 struct cgroup
*cgrp
;
1870 struct cgroup_taskset
{
1871 struct task_and_cgroup single
;
1872 struct flex_array
*tc_array
;
1875 struct cgroup
*cur_cgrp
;
1879 * cgroup_taskset_first - reset taskset and return the first task
1880 * @tset: taskset of interest
1882 * @tset iteration is initialized and the first task is returned.
1884 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1886 if (tset
->tc_array
) {
1888 return cgroup_taskset_next(tset
);
1890 tset
->cur_cgrp
= tset
->single
.cgrp
;
1891 return tset
->single
.task
;
1894 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1897 * cgroup_taskset_next - iterate to the next task in taskset
1898 * @tset: taskset of interest
1900 * Return the next task in @tset. Iteration must have been initialized
1901 * with cgroup_taskset_first().
1903 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1905 struct task_and_cgroup
*tc
;
1907 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1910 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1911 tset
->cur_cgrp
= tc
->cgrp
;
1914 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1917 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1918 * @tset: taskset of interest
1920 * Return the cgroup for the current (last returned) task of @tset. This
1921 * function must be preceded by either cgroup_taskset_first() or
1922 * cgroup_taskset_next().
1924 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1926 return tset
->cur_cgrp
;
1928 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1931 * cgroup_taskset_size - return the number of tasks in taskset
1932 * @tset: taskset of interest
1934 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1936 return tset
->tc_array
? tset
->tc_array_len
: 1;
1938 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1942 * cgroup_task_migrate - move a task from one cgroup to another.
1944 * Must be called with cgroup_mutex and threadgroup locked.
1946 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
1947 struct task_struct
*tsk
,
1948 struct css_set
*new_cset
)
1950 struct css_set
*old_cset
;
1953 * We are synchronized through threadgroup_lock() against PF_EXITING
1954 * setting such that we can't race against cgroup_exit() changing the
1955 * css_set to init_css_set and dropping the old one.
1957 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1958 old_cset
= tsk
->cgroups
;
1961 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
1964 /* Update the css_set linked lists if we're using them */
1965 write_lock(&css_set_lock
);
1966 if (!list_empty(&tsk
->cg_list
))
1967 list_move(&tsk
->cg_list
, &new_cset
->tasks
);
1968 write_unlock(&css_set_lock
);
1971 * We just gained a reference on old_cset by taking it from the
1972 * task. As trading it for new_cset is protected by cgroup_mutex,
1973 * we're safe to drop it here; it will be freed under RCU.
1975 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
1976 put_css_set(old_cset
);
1980 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1981 * @cgrp: the cgroup to attach to
1982 * @tsk: the task or the leader of the threadgroup to be attached
1983 * @threadgroup: attach the whole threadgroup?
1985 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1986 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1988 static int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
,
1991 int retval
, i
, group_size
;
1992 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1993 struct cgroupfs_root
*root
= cgrp
->root
;
1994 /* threadgroup list cursor and array */
1995 struct task_struct
*leader
= tsk
;
1996 struct task_and_cgroup
*tc
;
1997 struct flex_array
*group
;
1998 struct cgroup_taskset tset
= { };
2001 * step 0: in order to do expensive, possibly blocking operations for
2002 * every thread, we cannot iterate the thread group list, since it needs
2003 * rcu or tasklist locked. instead, build an array of all threads in the
2004 * group - group_rwsem prevents new threads from appearing, and if
2005 * threads exit, this will just be an over-estimate.
2008 group_size
= get_nr_threads(tsk
);
2011 /* flex_array supports very large thread-groups better than kmalloc. */
2012 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2015 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2016 retval
= flex_array_prealloc(group
, 0, group_size
, GFP_KERNEL
);
2018 goto out_free_group_list
;
2022 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2023 * already PF_EXITING could be freed from underneath us unless we
2024 * take an rcu_read_lock.
2028 struct task_and_cgroup ent
;
2030 /* @tsk either already exited or can't exit until the end */
2031 if (tsk
->flags
& PF_EXITING
)
2034 /* as per above, nr_threads may decrease, but not increase. */
2035 BUG_ON(i
>= group_size
);
2037 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2038 /* nothing to do if this task is already in the cgroup */
2039 if (ent
.cgrp
== cgrp
)
2042 * saying GFP_ATOMIC has no effect here because we did prealloc
2043 * earlier, but it's good form to communicate our expectations.
2045 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2046 BUG_ON(retval
!= 0);
2051 } while_each_thread(leader
, tsk
);
2053 /* remember the number of threads in the array for later. */
2055 tset
.tc_array
= group
;
2056 tset
.tc_array_len
= group_size
;
2058 /* methods shouldn't be called if no task is actually migrating */
2061 goto out_free_group_list
;
2064 * step 1: check that we can legitimately attach to the cgroup.
2066 for_each_subsys(root
, ss
) {
2067 if (ss
->can_attach
) {
2068 retval
= ss
->can_attach(cgrp
, &tset
);
2071 goto out_cancel_attach
;
2077 * step 2: make sure css_sets exist for all threads to be migrated.
2078 * we use find_css_set, which allocates a new one if necessary.
2080 for (i
= 0; i
< group_size
; i
++) {
2081 tc
= flex_array_get(group
, i
);
2082 tc
->cg
= find_css_set(tc
->task
->cgroups
, cgrp
);
2085 goto out_put_css_set_refs
;
2090 * step 3: now that we're guaranteed success wrt the css_sets,
2091 * proceed to move all tasks to the new cgroup. There are no
2092 * failure cases after here, so this is the commit point.
2094 for (i
= 0; i
< group_size
; i
++) {
2095 tc
= flex_array_get(group
, i
);
2096 cgroup_task_migrate(tc
->cgrp
, tc
->task
, tc
->cg
);
2098 /* nothing is sensitive to fork() after this point. */
2101 * step 4: do subsystem attach callbacks.
2103 for_each_subsys(root
, ss
) {
2105 ss
->attach(cgrp
, &tset
);
2109 * step 5: success! and cleanup
2112 out_put_css_set_refs
:
2114 for (i
= 0; i
< group_size
; i
++) {
2115 tc
= flex_array_get(group
, i
);
2118 put_css_set(tc
->cg
);
2123 for_each_subsys(root
, ss
) {
2124 if (ss
== failed_ss
)
2126 if (ss
->cancel_attach
)
2127 ss
->cancel_attach(cgrp
, &tset
);
2130 out_free_group_list
:
2131 flex_array_free(group
);
2136 * Find the task_struct of the task to attach by vpid and pass it along to the
2137 * function to attach either it or all tasks in its threadgroup. Will lock
2138 * cgroup_mutex and threadgroup; may take task_lock of task.
2140 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2142 struct task_struct
*tsk
;
2143 const struct cred
*cred
= current_cred(), *tcred
;
2146 if (!cgroup_lock_live_group(cgrp
))
2152 tsk
= find_task_by_vpid(pid
);
2156 goto out_unlock_cgroup
;
2159 * even if we're attaching all tasks in the thread group, we
2160 * only need to check permissions on one of them.
2162 tcred
= __task_cred(tsk
);
2163 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2164 !uid_eq(cred
->euid
, tcred
->uid
) &&
2165 !uid_eq(cred
->euid
, tcred
->suid
)) {
2168 goto out_unlock_cgroup
;
2174 tsk
= tsk
->group_leader
;
2177 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2178 * trapped in a cpuset, or RT worker may be born in a cgroup
2179 * with no rt_runtime allocated. Just say no.
2181 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2184 goto out_unlock_cgroup
;
2187 get_task_struct(tsk
);
2190 threadgroup_lock(tsk
);
2192 if (!thread_group_leader(tsk
)) {
2194 * a race with de_thread from another thread's exec()
2195 * may strip us of our leadership, if this happens,
2196 * there is no choice but to throw this task away and
2197 * try again; this is
2198 * "double-double-toil-and-trouble-check locking".
2200 threadgroup_unlock(tsk
);
2201 put_task_struct(tsk
);
2202 goto retry_find_task
;
2206 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2208 threadgroup_unlock(tsk
);
2210 put_task_struct(tsk
);
2212 mutex_unlock(&cgroup_mutex
);
2217 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2218 * @from: attach to all cgroups of a given task
2219 * @tsk: the task to be attached
2221 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2223 struct cgroupfs_root
*root
;
2226 mutex_lock(&cgroup_mutex
);
2227 for_each_active_root(root
) {
2228 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
2230 retval
= cgroup_attach_task(from_cg
, tsk
, false);
2234 mutex_unlock(&cgroup_mutex
);
2238 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2240 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2242 return attach_task_by_pid(cgrp
, pid
, false);
2245 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2247 return attach_task_by_pid(cgrp
, tgid
, true);
2250 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2253 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2254 if (strlen(buffer
) >= PATH_MAX
)
2256 if (!cgroup_lock_live_group(cgrp
))
2258 mutex_lock(&cgroup_root_mutex
);
2259 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2260 mutex_unlock(&cgroup_root_mutex
);
2261 mutex_unlock(&cgroup_mutex
);
2265 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2266 struct seq_file
*seq
)
2268 if (!cgroup_lock_live_group(cgrp
))
2270 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2271 seq_putc(seq
, '\n');
2272 mutex_unlock(&cgroup_mutex
);
2276 static int cgroup_sane_behavior_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2277 struct seq_file
*seq
)
2279 seq_printf(seq
, "%d\n", cgroup_sane_behavior(cgrp
));
2283 /* A buffer size big enough for numbers or short strings */
2284 #define CGROUP_LOCAL_BUFFER_SIZE 64
2286 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2288 const char __user
*userbuf
,
2289 size_t nbytes
, loff_t
*unused_ppos
)
2291 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2297 if (nbytes
>= sizeof(buffer
))
2299 if (copy_from_user(buffer
, userbuf
, nbytes
))
2302 buffer
[nbytes
] = 0; /* nul-terminate */
2303 if (cft
->write_u64
) {
2304 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2307 retval
= cft
->write_u64(cgrp
, cft
, val
);
2309 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2312 retval
= cft
->write_s64(cgrp
, cft
, val
);
2319 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2321 const char __user
*userbuf
,
2322 size_t nbytes
, loff_t
*unused_ppos
)
2324 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2326 size_t max_bytes
= cft
->max_write_len
;
2327 char *buffer
= local_buffer
;
2330 max_bytes
= sizeof(local_buffer
) - 1;
2331 if (nbytes
>= max_bytes
)
2333 /* Allocate a dynamic buffer if we need one */
2334 if (nbytes
>= sizeof(local_buffer
)) {
2335 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2339 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2344 buffer
[nbytes
] = 0; /* nul-terminate */
2345 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2349 if (buffer
!= local_buffer
)
2354 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2355 size_t nbytes
, loff_t
*ppos
)
2357 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2358 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2360 if (cgroup_is_dead(cgrp
))
2363 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2364 if (cft
->write_u64
|| cft
->write_s64
)
2365 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2366 if (cft
->write_string
)
2367 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2369 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2370 return ret
? ret
: nbytes
;
2375 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2377 char __user
*buf
, size_t nbytes
,
2380 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2381 u64 val
= cft
->read_u64(cgrp
, cft
);
2382 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2384 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2387 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2389 char __user
*buf
, size_t nbytes
,
2392 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2393 s64 val
= cft
->read_s64(cgrp
, cft
);
2394 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2396 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2399 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2400 size_t nbytes
, loff_t
*ppos
)
2402 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2403 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2405 if (cgroup_is_dead(cgrp
))
2409 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2411 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2413 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2418 * seqfile ops/methods for returning structured data. Currently just
2419 * supports string->u64 maps, but can be extended in future.
2422 struct cgroup_seqfile_state
{
2424 struct cgroup
*cgroup
;
2427 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2429 struct seq_file
*sf
= cb
->state
;
2430 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2433 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2435 struct cgroup_seqfile_state
*state
= m
->private;
2436 struct cftype
*cft
= state
->cft
;
2437 if (cft
->read_map
) {
2438 struct cgroup_map_cb cb
= {
2439 .fill
= cgroup_map_add
,
2442 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2444 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2447 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2449 struct seq_file
*seq
= file
->private_data
;
2450 kfree(seq
->private);
2451 return single_release(inode
, file
);
2454 static const struct file_operations cgroup_seqfile_operations
= {
2456 .write
= cgroup_file_write
,
2457 .llseek
= seq_lseek
,
2458 .release
= cgroup_seqfile_release
,
2461 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2466 err
= generic_file_open(inode
, file
);
2469 cft
= __d_cft(file
->f_dentry
);
2471 if (cft
->read_map
|| cft
->read_seq_string
) {
2472 struct cgroup_seqfile_state
*state
;
2474 state
= kzalloc(sizeof(*state
), GFP_USER
);
2479 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2480 file
->f_op
= &cgroup_seqfile_operations
;
2481 err
= single_open(file
, cgroup_seqfile_show
, state
);
2484 } else if (cft
->open
)
2485 err
= cft
->open(inode
, file
);
2492 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2494 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2496 return cft
->release(inode
, file
);
2501 * cgroup_rename - Only allow simple rename of directories in place.
2503 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2504 struct inode
*new_dir
, struct dentry
*new_dentry
)
2507 struct cgroup_name
*name
, *old_name
;
2508 struct cgroup
*cgrp
;
2511 * It's convinient to use parent dir's i_mutex to protected
2514 lockdep_assert_held(&old_dir
->i_mutex
);
2516 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2518 if (new_dentry
->d_inode
)
2520 if (old_dir
!= new_dir
)
2523 cgrp
= __d_cgrp(old_dentry
);
2525 name
= cgroup_alloc_name(new_dentry
);
2529 ret
= simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2535 old_name
= cgrp
->name
;
2536 rcu_assign_pointer(cgrp
->name
, name
);
2538 kfree_rcu(old_name
, rcu_head
);
2542 static struct simple_xattrs
*__d_xattrs(struct dentry
*dentry
)
2544 if (S_ISDIR(dentry
->d_inode
->i_mode
))
2545 return &__d_cgrp(dentry
)->xattrs
;
2547 return &__d_cfe(dentry
)->xattrs
;
2550 static inline int xattr_enabled(struct dentry
*dentry
)
2552 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
2553 return root
->flags
& CGRP_ROOT_XATTR
;
2556 static bool is_valid_xattr(const char *name
)
2558 if (!strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) ||
2559 !strncmp(name
, XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
))
2564 static int cgroup_setxattr(struct dentry
*dentry
, const char *name
,
2565 const void *val
, size_t size
, int flags
)
2567 if (!xattr_enabled(dentry
))
2569 if (!is_valid_xattr(name
))
2571 return simple_xattr_set(__d_xattrs(dentry
), name
, val
, size
, flags
);
2574 static int cgroup_removexattr(struct dentry
*dentry
, const char *name
)
2576 if (!xattr_enabled(dentry
))
2578 if (!is_valid_xattr(name
))
2580 return simple_xattr_remove(__d_xattrs(dentry
), name
);
2583 static ssize_t
cgroup_getxattr(struct dentry
*dentry
, const char *name
,
2584 void *buf
, size_t size
)
2586 if (!xattr_enabled(dentry
))
2588 if (!is_valid_xattr(name
))
2590 return simple_xattr_get(__d_xattrs(dentry
), name
, buf
, size
);
2593 static ssize_t
cgroup_listxattr(struct dentry
*dentry
, char *buf
, size_t size
)
2595 if (!xattr_enabled(dentry
))
2597 return simple_xattr_list(__d_xattrs(dentry
), buf
, size
);
2600 static const struct file_operations cgroup_file_operations
= {
2601 .read
= cgroup_file_read
,
2602 .write
= cgroup_file_write
,
2603 .llseek
= generic_file_llseek
,
2604 .open
= cgroup_file_open
,
2605 .release
= cgroup_file_release
,
2608 static const struct inode_operations cgroup_file_inode_operations
= {
2609 .setxattr
= cgroup_setxattr
,
2610 .getxattr
= cgroup_getxattr
,
2611 .listxattr
= cgroup_listxattr
,
2612 .removexattr
= cgroup_removexattr
,
2615 static const struct inode_operations cgroup_dir_inode_operations
= {
2616 .lookup
= cgroup_lookup
,
2617 .mkdir
= cgroup_mkdir
,
2618 .rmdir
= cgroup_rmdir
,
2619 .rename
= cgroup_rename
,
2620 .setxattr
= cgroup_setxattr
,
2621 .getxattr
= cgroup_getxattr
,
2622 .listxattr
= cgroup_listxattr
,
2623 .removexattr
= cgroup_removexattr
,
2626 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
2628 if (dentry
->d_name
.len
> NAME_MAX
)
2629 return ERR_PTR(-ENAMETOOLONG
);
2630 d_add(dentry
, NULL
);
2635 * Check if a file is a control file
2637 static inline struct cftype
*__file_cft(struct file
*file
)
2639 if (file_inode(file
)->i_fop
!= &cgroup_file_operations
)
2640 return ERR_PTR(-EINVAL
);
2641 return __d_cft(file
->f_dentry
);
2644 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2645 struct super_block
*sb
)
2647 struct inode
*inode
;
2651 if (dentry
->d_inode
)
2654 inode
= cgroup_new_inode(mode
, sb
);
2658 if (S_ISDIR(mode
)) {
2659 inode
->i_op
= &cgroup_dir_inode_operations
;
2660 inode
->i_fop
= &simple_dir_operations
;
2662 /* start off with i_nlink == 2 (for "." entry) */
2664 inc_nlink(dentry
->d_parent
->d_inode
);
2667 * Control reaches here with cgroup_mutex held.
2668 * @inode->i_mutex should nest outside cgroup_mutex but we
2669 * want to populate it immediately without releasing
2670 * cgroup_mutex. As @inode isn't visible to anyone else
2671 * yet, trylock will always succeed without affecting
2674 WARN_ON_ONCE(!mutex_trylock(&inode
->i_mutex
));
2675 } else if (S_ISREG(mode
)) {
2677 inode
->i_fop
= &cgroup_file_operations
;
2678 inode
->i_op
= &cgroup_file_inode_operations
;
2680 d_instantiate(dentry
, inode
);
2681 dget(dentry
); /* Extra count - pin the dentry in core */
2686 * cgroup_file_mode - deduce file mode of a control file
2687 * @cft: the control file in question
2689 * returns cft->mode if ->mode is not 0
2690 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2691 * returns S_IRUGO if it has only a read handler
2692 * returns S_IWUSR if it has only a write hander
2694 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2701 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2702 cft
->read_map
|| cft
->read_seq_string
)
2705 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2706 cft
->write_string
|| cft
->trigger
)
2712 static int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2715 struct dentry
*dir
= cgrp
->dentry
;
2716 struct cgroup
*parent
= __d_cgrp(dir
);
2717 struct dentry
*dentry
;
2721 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2723 if (subsys
&& !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
2724 strcpy(name
, subsys
->name
);
2727 strcat(name
, cft
->name
);
2729 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2731 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2735 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2736 if (IS_ERR(dentry
)) {
2737 error
= PTR_ERR(dentry
);
2741 cfe
->type
= (void *)cft
;
2742 cfe
->dentry
= dentry
;
2743 dentry
->d_fsdata
= cfe
;
2744 simple_xattrs_init(&cfe
->xattrs
);
2746 mode
= cgroup_file_mode(cft
);
2747 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2749 list_add_tail(&cfe
->node
, &parent
->files
);
2758 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2759 struct cftype cfts
[], bool is_add
)
2764 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2765 /* does cft->flags tell us to skip this file on @cgrp? */
2766 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_sane_behavior(cgrp
))
2768 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2770 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2774 err
= cgroup_add_file(cgrp
, subsys
, cft
);
2776 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2780 cgroup_rm_file(cgrp
, cft
);
2786 static DEFINE_MUTEX(cgroup_cft_mutex
);
2788 static void cgroup_cfts_prepare(void)
2789 __acquires(&cgroup_cft_mutex
) __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 increment reference on all cgroups and build list of
2795 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2796 * exclusive access to the field.
2798 mutex_lock(&cgroup_cft_mutex
);
2799 mutex_lock(&cgroup_mutex
);
2802 static void cgroup_cfts_commit(struct cgroup_subsys
*ss
,
2803 struct cftype
*cfts
, bool is_add
)
2804 __releases(&cgroup_mutex
) __releases(&cgroup_cft_mutex
)
2807 struct cgroup
*cgrp
, *n
;
2809 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2810 if (cfts
&& ss
->root
!= &rootnode
) {
2811 list_for_each_entry(cgrp
, &ss
->root
->allcg_list
, allcg_node
) {
2813 list_add_tail(&cgrp
->cft_q_node
, &pending
);
2817 mutex_unlock(&cgroup_mutex
);
2820 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2821 * files for all cgroups which were created before.
2823 list_for_each_entry_safe(cgrp
, n
, &pending
, cft_q_node
) {
2824 struct inode
*inode
= cgrp
->dentry
->d_inode
;
2826 mutex_lock(&inode
->i_mutex
);
2827 mutex_lock(&cgroup_mutex
);
2828 if (!cgroup_is_dead(cgrp
))
2829 cgroup_addrm_files(cgrp
, ss
, cfts
, is_add
);
2830 mutex_unlock(&cgroup_mutex
);
2831 mutex_unlock(&inode
->i_mutex
);
2833 list_del_init(&cgrp
->cft_q_node
);
2837 mutex_unlock(&cgroup_cft_mutex
);
2841 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2842 * @ss: target cgroup subsystem
2843 * @cfts: zero-length name terminated array of cftypes
2845 * Register @cfts to @ss. Files described by @cfts are created for all
2846 * existing cgroups to which @ss is attached and all future cgroups will
2847 * have them too. This function can be called anytime whether @ss is
2850 * Returns 0 on successful registration, -errno on failure. Note that this
2851 * function currently returns 0 as long as @cfts registration is successful
2852 * even if some file creation attempts on existing cgroups fail.
2854 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2856 struct cftype_set
*set
;
2858 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2862 cgroup_cfts_prepare();
2864 list_add_tail(&set
->node
, &ss
->cftsets
);
2865 cgroup_cfts_commit(ss
, cfts
, true);
2869 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2872 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2873 * @ss: target cgroup subsystem
2874 * @cfts: zero-length name terminated array of cftypes
2876 * Unregister @cfts from @ss. Files described by @cfts are removed from
2877 * all existing cgroups to which @ss is attached and all future cgroups
2878 * won't have them either. This function can be called anytime whether @ss
2879 * is attached or not.
2881 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2882 * registered with @ss.
2884 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2886 struct cftype_set
*set
;
2888 cgroup_cfts_prepare();
2890 list_for_each_entry(set
, &ss
->cftsets
, node
) {
2891 if (set
->cfts
== cfts
) {
2892 list_del_init(&set
->node
);
2893 cgroup_cfts_commit(ss
, cfts
, false);
2898 cgroup_cfts_commit(ss
, NULL
, false);
2903 * cgroup_task_count - count the number of tasks in a cgroup.
2904 * @cgrp: the cgroup in question
2906 * Return the number of tasks in the cgroup.
2908 int cgroup_task_count(const struct cgroup
*cgrp
)
2911 struct cgrp_cset_link
*link
;
2913 read_lock(&css_set_lock
);
2914 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
2915 count
+= atomic_read(&link
->cset
->refcount
);
2916 read_unlock(&css_set_lock
);
2921 * Advance a list_head iterator. The iterator should be positioned at
2922 * the start of a css_set
2924 static void cgroup_advance_iter(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2926 struct list_head
*l
= it
->cset_link
;
2927 struct cgrp_cset_link
*link
;
2928 struct css_set
*cset
;
2930 /* Advance to the next non-empty css_set */
2933 if (l
== &cgrp
->cset_links
) {
2934 it
->cset_link
= NULL
;
2937 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
2939 } while (list_empty(&cset
->tasks
));
2941 it
->task
= cset
->tasks
.next
;
2945 * To reduce the fork() overhead for systems that are not actually
2946 * using their cgroups capability, we don't maintain the lists running
2947 * through each css_set to its tasks until we see the list actually
2948 * used - in other words after the first call to cgroup_iter_start().
2950 static void cgroup_enable_task_cg_lists(void)
2952 struct task_struct
*p
, *g
;
2953 write_lock(&css_set_lock
);
2954 use_task_css_set_links
= 1;
2956 * We need tasklist_lock because RCU is not safe against
2957 * while_each_thread(). Besides, a forking task that has passed
2958 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2959 * is not guaranteed to have its child immediately visible in the
2960 * tasklist if we walk through it with RCU.
2962 read_lock(&tasklist_lock
);
2963 do_each_thread(g
, p
) {
2966 * We should check if the process is exiting, otherwise
2967 * it will race with cgroup_exit() in that the list
2968 * entry won't be deleted though the process has exited.
2970 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2971 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2973 } while_each_thread(g
, p
);
2974 read_unlock(&tasklist_lock
);
2975 write_unlock(&css_set_lock
);
2979 * cgroup_next_sibling - find the next sibling of a given cgroup
2980 * @pos: the current cgroup
2982 * This function returns the next sibling of @pos and should be called
2983 * under RCU read lock. The only requirement is that @pos is accessible.
2984 * The next sibling is guaranteed to be returned regardless of @pos's
2987 struct cgroup
*cgroup_next_sibling(struct cgroup
*pos
)
2989 struct cgroup
*next
;
2991 WARN_ON_ONCE(!rcu_read_lock_held());
2994 * @pos could already have been removed. Once a cgroup is removed,
2995 * its ->sibling.next is no longer updated when its next sibling
2996 * changes. As CGRP_DEAD is set on removal which is fully
2997 * serialized, if we see it unasserted, it's guaranteed that the
2998 * next sibling hasn't finished its grace period even if it's
2999 * already removed, and thus safe to dereference from this RCU
3000 * critical section. If ->sibling.next is inaccessible,
3001 * cgroup_is_dead() is guaranteed to be visible as %true here.
3003 if (likely(!cgroup_is_dead(pos
))) {
3004 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
, sibling
);
3005 if (&next
->sibling
!= &pos
->parent
->children
)
3011 * Can't dereference the next pointer. Each cgroup is given a
3012 * monotonically increasing unique serial number and always
3013 * appended to the sibling list, so the next one can be found by
3014 * walking the parent's children until we see a cgroup with higher
3015 * serial number than @pos's.
3017 * While this path can be slow, it's taken only when either the
3018 * current cgroup is removed or iteration and removal race.
3020 list_for_each_entry_rcu(next
, &pos
->parent
->children
, sibling
)
3021 if (next
->serial_nr
> pos
->serial_nr
)
3025 EXPORT_SYMBOL_GPL(cgroup_next_sibling
);
3028 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
3029 * @pos: the current position (%NULL to initiate traversal)
3030 * @cgroup: cgroup whose descendants to walk
3032 * To be used by cgroup_for_each_descendant_pre(). Find the next
3033 * descendant to visit for pre-order traversal of @cgroup's descendants.
3035 * While this function requires RCU read locking, it doesn't require the
3036 * whole traversal to be contained in a single RCU critical section. This
3037 * function will return the correct next descendant as long as both @pos
3038 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3040 struct cgroup
*cgroup_next_descendant_pre(struct cgroup
*pos
,
3041 struct cgroup
*cgroup
)
3043 struct cgroup
*next
;
3045 WARN_ON_ONCE(!rcu_read_lock_held());
3047 /* if first iteration, pretend we just visited @cgroup */
3051 /* visit the first child if exists */
3052 next
= list_first_or_null_rcu(&pos
->children
, struct cgroup
, sibling
);
3056 /* no child, visit my or the closest ancestor's next sibling */
3057 while (pos
!= cgroup
) {
3058 next
= cgroup_next_sibling(pos
);
3066 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre
);
3069 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
3070 * @pos: cgroup of interest
3072 * Return the rightmost descendant of @pos. If there's no descendant,
3073 * @pos is returned. This can be used during pre-order traversal to skip
3076 * While this function requires RCU read locking, it doesn't require the
3077 * whole traversal to be contained in a single RCU critical section. This
3078 * function will return the correct rightmost descendant as long as @pos is
3081 struct cgroup
*cgroup_rightmost_descendant(struct cgroup
*pos
)
3083 struct cgroup
*last
, *tmp
;
3085 WARN_ON_ONCE(!rcu_read_lock_held());
3089 /* ->prev isn't RCU safe, walk ->next till the end */
3091 list_for_each_entry_rcu(tmp
, &last
->children
, sibling
)
3097 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant
);
3099 static struct cgroup
*cgroup_leftmost_descendant(struct cgroup
*pos
)
3101 struct cgroup
*last
;
3105 pos
= list_first_or_null_rcu(&pos
->children
, struct cgroup
,
3113 * cgroup_next_descendant_post - find the next descendant for post-order walk
3114 * @pos: the current position (%NULL to initiate traversal)
3115 * @cgroup: cgroup whose descendants to walk
3117 * To be used by cgroup_for_each_descendant_post(). Find the next
3118 * descendant to visit for post-order traversal of @cgroup's descendants.
3120 * While this function requires RCU read locking, it doesn't require the
3121 * whole traversal to be contained in a single RCU critical section. This
3122 * function will return the correct next descendant as long as both @pos
3123 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3125 struct cgroup
*cgroup_next_descendant_post(struct cgroup
*pos
,
3126 struct cgroup
*cgroup
)
3128 struct cgroup
*next
;
3130 WARN_ON_ONCE(!rcu_read_lock_held());
3132 /* if first iteration, visit the leftmost descendant */
3134 next
= cgroup_leftmost_descendant(cgroup
);
3135 return next
!= cgroup
? next
: NULL
;
3138 /* if there's an unvisited sibling, visit its leftmost descendant */
3139 next
= cgroup_next_sibling(pos
);
3141 return cgroup_leftmost_descendant(next
);
3143 /* no sibling left, visit parent */
3145 return next
!= cgroup
? next
: NULL
;
3147 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post
);
3149 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3150 __acquires(css_set_lock
)
3153 * The first time anyone tries to iterate across a cgroup,
3154 * we need to enable the list linking each css_set to its
3155 * tasks, and fix up all existing tasks.
3157 if (!use_task_css_set_links
)
3158 cgroup_enable_task_cg_lists();
3160 read_lock(&css_set_lock
);
3161 it
->cset_link
= &cgrp
->cset_links
;
3162 cgroup_advance_iter(cgrp
, it
);
3165 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
3166 struct cgroup_iter
*it
)
3168 struct task_struct
*res
;
3169 struct list_head
*l
= it
->task
;
3170 struct cgrp_cset_link
*link
;
3172 /* If the iterator cg is NULL, we have no tasks */
3175 res
= list_entry(l
, struct task_struct
, cg_list
);
3176 /* Advance iterator to find next entry */
3178 link
= list_entry(it
->cset_link
, struct cgrp_cset_link
, cset_link
);
3179 if (l
== &link
->cset
->tasks
) {
3180 /* We reached the end of this task list - move on to
3181 * the next cg_cgroup_link */
3182 cgroup_advance_iter(cgrp
, it
);
3189 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3190 __releases(css_set_lock
)
3192 read_unlock(&css_set_lock
);
3195 static inline int started_after_time(struct task_struct
*t1
,
3196 struct timespec
*time
,
3197 struct task_struct
*t2
)
3199 int start_diff
= timespec_compare(&t1
->start_time
, time
);
3200 if (start_diff
> 0) {
3202 } else if (start_diff
< 0) {
3206 * Arbitrarily, if two processes started at the same
3207 * time, we'll say that the lower pointer value
3208 * started first. Note that t2 may have exited by now
3209 * so this may not be a valid pointer any longer, but
3210 * that's fine - it still serves to distinguish
3211 * between two tasks started (effectively) simultaneously.
3218 * This function is a callback from heap_insert() and is used to order
3220 * In this case we order the heap in descending task start time.
3222 static inline int started_after(void *p1
, void *p2
)
3224 struct task_struct
*t1
= p1
;
3225 struct task_struct
*t2
= p2
;
3226 return started_after_time(t1
, &t2
->start_time
, t2
);
3230 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3231 * @scan: struct cgroup_scanner containing arguments for the scan
3233 * Arguments include pointers to callback functions test_task() and
3235 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3236 * and if it returns true, call process_task() for it also.
3237 * The test_task pointer may be NULL, meaning always true (select all tasks).
3238 * Effectively duplicates cgroup_iter_{start,next,end}()
3239 * but does not lock css_set_lock for the call to process_task().
3240 * The struct cgroup_scanner may be embedded in any structure of the caller's
3242 * It is guaranteed that process_task() will act on every task that
3243 * is a member of the cgroup for the duration of this call. This
3244 * function may or may not call process_task() for tasks that exit
3245 * or move to a different cgroup during the call, or are forked or
3246 * move into the cgroup during the call.
3248 * Note that test_task() may be called with locks held, and may in some
3249 * situations be called multiple times for the same task, so it should
3251 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3252 * pre-allocated and will be used for heap operations (and its "gt" member will
3253 * be overwritten), else a temporary heap will be used (allocation of which
3254 * may cause this function to fail).
3256 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
3259 struct cgroup_iter it
;
3260 struct task_struct
*p
, *dropped
;
3261 /* Never dereference latest_task, since it's not refcounted */
3262 struct task_struct
*latest_task
= NULL
;
3263 struct ptr_heap tmp_heap
;
3264 struct ptr_heap
*heap
;
3265 struct timespec latest_time
= { 0, 0 };
3268 /* The caller supplied our heap and pre-allocated its memory */
3270 heap
->gt
= &started_after
;
3272 /* We need to allocate our own heap memory */
3274 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3276 /* cannot allocate the heap */
3282 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3283 * to determine which are of interest, and using the scanner's
3284 * "process_task" callback to process any of them that need an update.
3285 * Since we don't want to hold any locks during the task updates,
3286 * gather tasks to be processed in a heap structure.
3287 * The heap is sorted by descending task start time.
3288 * If the statically-sized heap fills up, we overflow tasks that
3289 * started later, and in future iterations only consider tasks that
3290 * started after the latest task in the previous pass. This
3291 * guarantees forward progress and that we don't miss any tasks.
3294 cgroup_iter_start(scan
->cg
, &it
);
3295 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
3297 * Only affect tasks that qualify per the caller's callback,
3298 * if he provided one
3300 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
3303 * Only process tasks that started after the last task
3306 if (!started_after_time(p
, &latest_time
, latest_task
))
3308 dropped
= heap_insert(heap
, p
);
3309 if (dropped
== NULL
) {
3311 * The new task was inserted; the heap wasn't
3315 } else if (dropped
!= p
) {
3317 * The new task was inserted, and pushed out a
3321 put_task_struct(dropped
);
3324 * Else the new task was newer than anything already in
3325 * the heap and wasn't inserted
3328 cgroup_iter_end(scan
->cg
, &it
);
3331 for (i
= 0; i
< heap
->size
; i
++) {
3332 struct task_struct
*q
= heap
->ptrs
[i
];
3334 latest_time
= q
->start_time
;
3337 /* Process the task per the caller's callback */
3338 scan
->process_task(q
, scan
);
3342 * If we had to process any tasks at all, scan again
3343 * in case some of them were in the middle of forking
3344 * children that didn't get processed.
3345 * Not the most efficient way to do it, but it avoids
3346 * having to take callback_mutex in the fork path
3350 if (heap
== &tmp_heap
)
3351 heap_free(&tmp_heap
);
3355 static void cgroup_transfer_one_task(struct task_struct
*task
,
3356 struct cgroup_scanner
*scan
)
3358 struct cgroup
*new_cgroup
= scan
->data
;
3360 mutex_lock(&cgroup_mutex
);
3361 cgroup_attach_task(new_cgroup
, task
, false);
3362 mutex_unlock(&cgroup_mutex
);
3366 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3367 * @to: cgroup to which the tasks will be moved
3368 * @from: cgroup in which the tasks currently reside
3370 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3372 struct cgroup_scanner scan
;
3375 scan
.test_task
= NULL
; /* select all tasks in cgroup */
3376 scan
.process_task
= cgroup_transfer_one_task
;
3380 return cgroup_scan_tasks(&scan
);
3384 * Stuff for reading the 'tasks'/'procs' files.
3386 * Reading this file can return large amounts of data if a cgroup has
3387 * *lots* of attached tasks. So it may need several calls to read(),
3388 * but we cannot guarantee that the information we produce is correct
3389 * unless we produce it entirely atomically.
3393 /* which pidlist file are we talking about? */
3394 enum cgroup_filetype
{
3400 * A pidlist is a list of pids that virtually represents the contents of one
3401 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3402 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3405 struct cgroup_pidlist
{
3407 * used to find which pidlist is wanted. doesn't change as long as
3408 * this particular list stays in the list.
3410 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3413 /* how many elements the above list has */
3415 /* how many files are using the current array */
3417 /* each of these stored in a list by its cgroup */
3418 struct list_head links
;
3419 /* pointer to the cgroup we belong to, for list removal purposes */
3420 struct cgroup
*owner
;
3421 /* protects the other fields */
3422 struct rw_semaphore mutex
;
3426 * The following two functions "fix" the issue where there are more pids
3427 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3428 * TODO: replace with a kernel-wide solution to this problem
3430 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3431 static void *pidlist_allocate(int count
)
3433 if (PIDLIST_TOO_LARGE(count
))
3434 return vmalloc(count
* sizeof(pid_t
));
3436 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3438 static void pidlist_free(void *p
)
3440 if (is_vmalloc_addr(p
))
3447 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3448 * Returns the number of unique elements.
3450 static int pidlist_uniq(pid_t
*list
, int length
)
3455 * we presume the 0th element is unique, so i starts at 1. trivial
3456 * edge cases first; no work needs to be done for either
3458 if (length
== 0 || length
== 1)
3460 /* src and dest walk down the list; dest counts unique elements */
3461 for (src
= 1; src
< length
; src
++) {
3462 /* find next unique element */
3463 while (list
[src
] == list
[src
-1]) {
3468 /* dest always points to where the next unique element goes */
3469 list
[dest
] = list
[src
];
3476 static int cmppid(const void *a
, const void *b
)
3478 return *(pid_t
*)a
- *(pid_t
*)b
;
3482 * find the appropriate pidlist for our purpose (given procs vs tasks)
3483 * returns with the lock on that pidlist already held, and takes care
3484 * of the use count, or returns NULL with no locks held if we're out of
3487 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3488 enum cgroup_filetype type
)
3490 struct cgroup_pidlist
*l
;
3491 /* don't need task_nsproxy() if we're looking at ourself */
3492 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3495 * We can't drop the pidlist_mutex before taking the l->mutex in case
3496 * the last ref-holder is trying to remove l from the list at the same
3497 * time. Holding the pidlist_mutex precludes somebody taking whichever
3498 * list we find out from under us - compare release_pid_array().
3500 mutex_lock(&cgrp
->pidlist_mutex
);
3501 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3502 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3503 /* make sure l doesn't vanish out from under us */
3504 down_write(&l
->mutex
);
3505 mutex_unlock(&cgrp
->pidlist_mutex
);
3509 /* entry not found; create a new one */
3510 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3512 mutex_unlock(&cgrp
->pidlist_mutex
);
3515 init_rwsem(&l
->mutex
);
3516 down_write(&l
->mutex
);
3518 l
->key
.ns
= get_pid_ns(ns
);
3520 list_add(&l
->links
, &cgrp
->pidlists
);
3521 mutex_unlock(&cgrp
->pidlist_mutex
);
3526 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3528 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3529 struct cgroup_pidlist
**lp
)
3533 int pid
, n
= 0; /* used for populating the array */
3534 struct cgroup_iter it
;
3535 struct task_struct
*tsk
;
3536 struct cgroup_pidlist
*l
;
3539 * If cgroup gets more users after we read count, we won't have
3540 * enough space - tough. This race is indistinguishable to the
3541 * caller from the case that the additional cgroup users didn't
3542 * show up until sometime later on.
3544 length
= cgroup_task_count(cgrp
);
3545 array
= pidlist_allocate(length
);
3548 /* now, populate the array */
3549 cgroup_iter_start(cgrp
, &it
);
3550 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3551 if (unlikely(n
== length
))
3553 /* get tgid or pid for procs or tasks file respectively */
3554 if (type
== CGROUP_FILE_PROCS
)
3555 pid
= task_tgid_vnr(tsk
);
3557 pid
= task_pid_vnr(tsk
);
3558 if (pid
> 0) /* make sure to only use valid results */
3561 cgroup_iter_end(cgrp
, &it
);
3563 /* now sort & (if procs) strip out duplicates */
3564 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3565 if (type
== CGROUP_FILE_PROCS
)
3566 length
= pidlist_uniq(array
, length
);
3567 l
= cgroup_pidlist_find(cgrp
, type
);
3569 pidlist_free(array
);
3572 /* store array, freeing old if necessary - lock already held */
3573 pidlist_free(l
->list
);
3577 up_write(&l
->mutex
);
3583 * cgroupstats_build - build and fill cgroupstats
3584 * @stats: cgroupstats to fill information into
3585 * @dentry: A dentry entry belonging to the cgroup for which stats have
3588 * Build and fill cgroupstats so that taskstats can export it to user
3591 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3594 struct cgroup
*cgrp
;
3595 struct cgroup_iter it
;
3596 struct task_struct
*tsk
;
3599 * Validate dentry by checking the superblock operations,
3600 * and make sure it's a directory.
3602 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3603 !S_ISDIR(dentry
->d_inode
->i_mode
))
3607 cgrp
= dentry
->d_fsdata
;
3609 cgroup_iter_start(cgrp
, &it
);
3610 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3611 switch (tsk
->state
) {
3613 stats
->nr_running
++;
3615 case TASK_INTERRUPTIBLE
:
3616 stats
->nr_sleeping
++;
3618 case TASK_UNINTERRUPTIBLE
:
3619 stats
->nr_uninterruptible
++;
3622 stats
->nr_stopped
++;
3625 if (delayacct_is_task_waiting_on_io(tsk
))
3626 stats
->nr_io_wait
++;
3630 cgroup_iter_end(cgrp
, &it
);
3638 * seq_file methods for the tasks/procs files. The seq_file position is the
3639 * next pid to display; the seq_file iterator is a pointer to the pid
3640 * in the cgroup->l->list array.
3643 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3646 * Initially we receive a position value that corresponds to
3647 * one more than the last pid shown (or 0 on the first call or
3648 * after a seek to the start). Use a binary-search to find the
3649 * next pid to display, if any
3651 struct cgroup_pidlist
*l
= s
->private;
3652 int index
= 0, pid
= *pos
;
3655 down_read(&l
->mutex
);
3657 int end
= l
->length
;
3659 while (index
< end
) {
3660 int mid
= (index
+ end
) / 2;
3661 if (l
->list
[mid
] == pid
) {
3664 } else if (l
->list
[mid
] <= pid
)
3670 /* If we're off the end of the array, we're done */
3671 if (index
>= l
->length
)
3673 /* Update the abstract position to be the actual pid that we found */
3674 iter
= l
->list
+ index
;
3679 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3681 struct cgroup_pidlist
*l
= s
->private;
3685 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3687 struct cgroup_pidlist
*l
= s
->private;
3689 pid_t
*end
= l
->list
+ l
->length
;
3691 * Advance to the next pid in the array. If this goes off the
3703 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3705 return seq_printf(s
, "%d\n", *(int *)v
);
3709 * seq_operations functions for iterating on pidlists through seq_file -
3710 * independent of whether it's tasks or procs
3712 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3713 .start
= cgroup_pidlist_start
,
3714 .stop
= cgroup_pidlist_stop
,
3715 .next
= cgroup_pidlist_next
,
3716 .show
= cgroup_pidlist_show
,
3719 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3722 * the case where we're the last user of this particular pidlist will
3723 * have us remove it from the cgroup's list, which entails taking the
3724 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3725 * pidlist_mutex, we have to take pidlist_mutex first.
3727 mutex_lock(&l
->owner
->pidlist_mutex
);
3728 down_write(&l
->mutex
);
3729 BUG_ON(!l
->use_count
);
3730 if (!--l
->use_count
) {
3731 /* we're the last user if refcount is 0; remove and free */
3732 list_del(&l
->links
);
3733 mutex_unlock(&l
->owner
->pidlist_mutex
);
3734 pidlist_free(l
->list
);
3735 put_pid_ns(l
->key
.ns
);
3736 up_write(&l
->mutex
);
3740 mutex_unlock(&l
->owner
->pidlist_mutex
);
3741 up_write(&l
->mutex
);
3744 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3746 struct cgroup_pidlist
*l
;
3747 if (!(file
->f_mode
& FMODE_READ
))
3750 * the seq_file will only be initialized if the file was opened for
3751 * reading; hence we check if it's not null only in that case.
3753 l
= ((struct seq_file
*)file
->private_data
)->private;
3754 cgroup_release_pid_array(l
);
3755 return seq_release(inode
, file
);
3758 static const struct file_operations cgroup_pidlist_operations
= {
3760 .llseek
= seq_lseek
,
3761 .write
= cgroup_file_write
,
3762 .release
= cgroup_pidlist_release
,
3766 * The following functions handle opens on a file that displays a pidlist
3767 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3770 /* helper function for the two below it */
3771 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3773 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3774 struct cgroup_pidlist
*l
;
3777 /* Nothing to do for write-only files */
3778 if (!(file
->f_mode
& FMODE_READ
))
3781 /* have the array populated */
3782 retval
= pidlist_array_load(cgrp
, type
, &l
);
3785 /* configure file information */
3786 file
->f_op
= &cgroup_pidlist_operations
;
3788 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3790 cgroup_release_pid_array(l
);
3793 ((struct seq_file
*)file
->private_data
)->private = l
;
3796 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3798 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3800 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3802 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3805 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3808 return notify_on_release(cgrp
);
3811 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3815 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3817 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3819 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3824 * Unregister event and free resources.
3826 * Gets called from workqueue.
3828 static void cgroup_event_remove(struct work_struct
*work
)
3830 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3832 struct cgroup
*cgrp
= event
->cgrp
;
3834 remove_wait_queue(event
->wqh
, &event
->wait
);
3836 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3838 /* Notify userspace the event is going away. */
3839 eventfd_signal(event
->eventfd
, 1);
3841 eventfd_ctx_put(event
->eventfd
);
3847 * Gets called on POLLHUP on eventfd when user closes it.
3849 * Called with wqh->lock held and interrupts disabled.
3851 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3852 int sync
, void *key
)
3854 struct cgroup_event
*event
= container_of(wait
,
3855 struct cgroup_event
, wait
);
3856 struct cgroup
*cgrp
= event
->cgrp
;
3857 unsigned long flags
= (unsigned long)key
;
3859 if (flags
& POLLHUP
) {
3861 * If the event has been detached at cgroup removal, we
3862 * can simply return knowing the other side will cleanup
3865 * We can't race against event freeing since the other
3866 * side will require wqh->lock via remove_wait_queue(),
3869 spin_lock(&cgrp
->event_list_lock
);
3870 if (!list_empty(&event
->list
)) {
3871 list_del_init(&event
->list
);
3873 * We are in atomic context, but cgroup_event_remove()
3874 * may sleep, so we have to call it in workqueue.
3876 schedule_work(&event
->remove
);
3878 spin_unlock(&cgrp
->event_list_lock
);
3884 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3885 wait_queue_head_t
*wqh
, poll_table
*pt
)
3887 struct cgroup_event
*event
= container_of(pt
,
3888 struct cgroup_event
, pt
);
3891 add_wait_queue(wqh
, &event
->wait
);
3895 * Parse input and register new cgroup event handler.
3897 * Input must be in format '<event_fd> <control_fd> <args>'.
3898 * Interpretation of args is defined by control file implementation.
3900 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3903 struct cgroup_event
*event
= NULL
;
3904 struct cgroup
*cgrp_cfile
;
3905 unsigned int efd
, cfd
;
3906 struct file
*efile
= NULL
;
3907 struct file
*cfile
= NULL
;
3911 efd
= simple_strtoul(buffer
, &endp
, 10);
3916 cfd
= simple_strtoul(buffer
, &endp
, 10);
3917 if ((*endp
!= ' ') && (*endp
!= '\0'))
3921 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3925 INIT_LIST_HEAD(&event
->list
);
3926 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3927 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3928 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3930 efile
= eventfd_fget(efd
);
3931 if (IS_ERR(efile
)) {
3932 ret
= PTR_ERR(efile
);
3936 event
->eventfd
= eventfd_ctx_fileget(efile
);
3937 if (IS_ERR(event
->eventfd
)) {
3938 ret
= PTR_ERR(event
->eventfd
);
3948 /* the process need read permission on control file */
3949 /* AV: shouldn't we check that it's been opened for read instead? */
3950 ret
= inode_permission(file_inode(cfile
), MAY_READ
);
3954 event
->cft
= __file_cft(cfile
);
3955 if (IS_ERR(event
->cft
)) {
3956 ret
= PTR_ERR(event
->cft
);
3961 * The file to be monitored must be in the same cgroup as
3962 * cgroup.event_control is.
3964 cgrp_cfile
= __d_cgrp(cfile
->f_dentry
->d_parent
);
3965 if (cgrp_cfile
!= cgrp
) {
3970 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3975 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3976 event
->eventfd
, buffer
);
3980 efile
->f_op
->poll(efile
, &event
->pt
);
3983 * Events should be removed after rmdir of cgroup directory, but before
3984 * destroying subsystem state objects. Let's take reference to cgroup
3985 * directory dentry to do that.
3989 spin_lock(&cgrp
->event_list_lock
);
3990 list_add(&event
->list
, &cgrp
->event_list
);
3991 spin_unlock(&cgrp
->event_list_lock
);
4002 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
4003 eventfd_ctx_put(event
->eventfd
);
4005 if (!IS_ERR_OR_NULL(efile
))
4013 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
4016 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4019 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
4024 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4026 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4030 static struct cftype cgroup_base_files
[] = {
4032 .name
= "cgroup.procs",
4033 .open
= cgroup_procs_open
,
4034 .write_u64
= cgroup_procs_write
,
4035 .release
= cgroup_pidlist_release
,
4036 .mode
= S_IRUGO
| S_IWUSR
,
4039 .name
= "cgroup.event_control",
4040 .write_string
= cgroup_write_event_control
,
4044 .name
= "cgroup.clone_children",
4045 .flags
= CFTYPE_INSANE
,
4046 .read_u64
= cgroup_clone_children_read
,
4047 .write_u64
= cgroup_clone_children_write
,
4050 .name
= "cgroup.sane_behavior",
4051 .flags
= CFTYPE_ONLY_ON_ROOT
,
4052 .read_seq_string
= cgroup_sane_behavior_show
,
4056 * Historical crazy stuff. These don't have "cgroup." prefix and
4057 * don't exist if sane_behavior. If you're depending on these, be
4058 * prepared to be burned.
4062 .flags
= CFTYPE_INSANE
, /* use "procs" instead */
4063 .open
= cgroup_tasks_open
,
4064 .write_u64
= cgroup_tasks_write
,
4065 .release
= cgroup_pidlist_release
,
4066 .mode
= S_IRUGO
| S_IWUSR
,
4069 .name
= "notify_on_release",
4070 .flags
= CFTYPE_INSANE
,
4071 .read_u64
= cgroup_read_notify_on_release
,
4072 .write_u64
= cgroup_write_notify_on_release
,
4075 .name
= "release_agent",
4076 .flags
= CFTYPE_INSANE
| CFTYPE_ONLY_ON_ROOT
,
4077 .read_seq_string
= cgroup_release_agent_show
,
4078 .write_string
= cgroup_release_agent_write
,
4079 .max_write_len
= PATH_MAX
,
4085 * cgroup_populate_dir - selectively creation of files in a directory
4086 * @cgrp: target cgroup
4087 * @base_files: true if the base files should be added
4088 * @subsys_mask: mask of the subsystem ids whose files should be added
4090 static int cgroup_populate_dir(struct cgroup
*cgrp
, bool base_files
,
4091 unsigned long subsys_mask
)
4094 struct cgroup_subsys
*ss
;
4097 err
= cgroup_addrm_files(cgrp
, NULL
, cgroup_base_files
, true);
4102 /* process cftsets of each subsystem */
4103 for_each_subsys(cgrp
->root
, ss
) {
4104 struct cftype_set
*set
;
4105 if (!test_bit(ss
->subsys_id
, &subsys_mask
))
4108 list_for_each_entry(set
, &ss
->cftsets
, node
)
4109 cgroup_addrm_files(cgrp
, ss
, set
->cfts
, true);
4112 /* This cgroup is ready now */
4113 for_each_subsys(cgrp
->root
, ss
) {
4114 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4116 * Update id->css pointer and make this css visible from
4117 * CSS ID functions. This pointer will be dereferened
4118 * from RCU-read-side without locks.
4121 rcu_assign_pointer(css
->id
->css
, css
);
4127 static void css_dput_fn(struct work_struct
*work
)
4129 struct cgroup_subsys_state
*css
=
4130 container_of(work
, struct cgroup_subsys_state
, dput_work
);
4131 struct dentry
*dentry
= css
->cgroup
->dentry
;
4132 struct super_block
*sb
= dentry
->d_sb
;
4134 atomic_inc(&sb
->s_active
);
4136 deactivate_super(sb
);
4139 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
4140 struct cgroup_subsys
*ss
,
4141 struct cgroup
*cgrp
)
4144 atomic_set(&css
->refcnt
, 1);
4147 if (cgrp
== dummytop
)
4148 css
->flags
|= CSS_ROOT
;
4149 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
4150 cgrp
->subsys
[ss
->subsys_id
] = css
;
4153 * css holds an extra ref to @cgrp->dentry which is put on the last
4154 * css_put(). dput() requires process context, which css_put() may
4155 * be called without. @css->dput_work will be used to invoke
4156 * dput() asynchronously from css_put().
4158 INIT_WORK(&css
->dput_work
, css_dput_fn
);
4161 /* invoke ->post_create() on a new CSS and mark it online if successful */
4162 static int online_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4166 lockdep_assert_held(&cgroup_mutex
);
4169 ret
= ss
->css_online(cgrp
);
4171 cgrp
->subsys
[ss
->subsys_id
]->flags
|= CSS_ONLINE
;
4175 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4176 static void offline_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4177 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4179 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4181 lockdep_assert_held(&cgroup_mutex
);
4183 if (!(css
->flags
& CSS_ONLINE
))
4186 if (ss
->css_offline
)
4187 ss
->css_offline(cgrp
);
4189 cgrp
->subsys
[ss
->subsys_id
]->flags
&= ~CSS_ONLINE
;
4193 * cgroup_create - create a cgroup
4194 * @parent: cgroup that will be parent of the new cgroup
4195 * @dentry: dentry of the new cgroup
4196 * @mode: mode to set on new inode
4198 * Must be called with the mutex on the parent inode held
4200 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
4203 static atomic64_t serial_nr_cursor
= ATOMIC64_INIT(0);
4204 struct cgroup
*cgrp
;
4205 struct cgroup_name
*name
;
4206 struct cgroupfs_root
*root
= parent
->root
;
4208 struct cgroup_subsys
*ss
;
4209 struct super_block
*sb
= root
->sb
;
4211 /* allocate the cgroup and its ID, 0 is reserved for the root */
4212 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4216 name
= cgroup_alloc_name(dentry
);
4219 rcu_assign_pointer(cgrp
->name
, name
);
4221 cgrp
->id
= ida_simple_get(&root
->cgroup_ida
, 1, 0, GFP_KERNEL
);
4226 * Only live parents can have children. Note that the liveliness
4227 * check isn't strictly necessary because cgroup_mkdir() and
4228 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4229 * anyway so that locking is contained inside cgroup proper and we
4230 * don't get nasty surprises if we ever grow another caller.
4232 if (!cgroup_lock_live_group(parent
)) {
4237 /* Grab a reference on the superblock so the hierarchy doesn't
4238 * get deleted on unmount if there are child cgroups. This
4239 * can be done outside cgroup_mutex, since the sb can't
4240 * disappear while someone has an open control file on the
4242 atomic_inc(&sb
->s_active
);
4244 init_cgroup_housekeeping(cgrp
);
4246 dentry
->d_fsdata
= cgrp
;
4247 cgrp
->dentry
= dentry
;
4249 cgrp
->parent
= parent
;
4250 cgrp
->root
= parent
->root
;
4252 if (notify_on_release(parent
))
4253 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4255 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4256 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4258 for_each_subsys(root
, ss
) {
4259 struct cgroup_subsys_state
*css
;
4261 css
= ss
->css_alloc(cgrp
);
4266 init_cgroup_css(css
, ss
, cgrp
);
4268 err
= alloc_css_id(ss
, parent
, cgrp
);
4275 * Create directory. cgroup_create_file() returns with the new
4276 * directory locked on success so that it can be populated without
4277 * dropping cgroup_mutex.
4279 err
= cgroup_create_file(dentry
, S_IFDIR
| mode
, sb
);
4282 lockdep_assert_held(&dentry
->d_inode
->i_mutex
);
4285 * Assign a monotonically increasing serial number. With the list
4286 * appending below, it guarantees that sibling cgroups are always
4287 * sorted in the ascending serial number order on the parent's
4290 cgrp
->serial_nr
= atomic64_inc_return(&serial_nr_cursor
);
4292 /* allocation complete, commit to creation */
4293 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
4294 list_add_tail_rcu(&cgrp
->sibling
, &cgrp
->parent
->children
);
4295 root
->number_of_cgroups
++;
4297 /* each css holds a ref to the cgroup's dentry */
4298 for_each_subsys(root
, ss
)
4301 /* hold a ref to the parent's dentry */
4302 dget(parent
->dentry
);
4304 /* creation succeeded, notify subsystems */
4305 for_each_subsys(root
, ss
) {
4306 err
= online_css(ss
, cgrp
);
4310 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4312 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",
4313 current
->comm
, current
->pid
, ss
->name
);
4314 if (!strcmp(ss
->name
, "memory"))
4315 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4316 ss
->warned_broken_hierarchy
= true;
4320 err
= cgroup_populate_dir(cgrp
, true, root
->subsys_mask
);
4324 mutex_unlock(&cgroup_mutex
);
4325 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4330 for_each_subsys(root
, ss
) {
4331 if (cgrp
->subsys
[ss
->subsys_id
])
4334 mutex_unlock(&cgroup_mutex
);
4335 /* Release the reference count that we took on the superblock */
4336 deactivate_super(sb
);
4338 ida_simple_remove(&root
->cgroup_ida
, cgrp
->id
);
4340 kfree(rcu_dereference_raw(cgrp
->name
));
4346 cgroup_destroy_locked(cgrp
);
4347 mutex_unlock(&cgroup_mutex
);
4348 mutex_unlock(&dentry
->d_inode
->i_mutex
);
4352 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4354 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4356 /* the vfs holds inode->i_mutex already */
4357 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4360 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4361 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4363 struct dentry
*d
= cgrp
->dentry
;
4364 struct cgroup
*parent
= cgrp
->parent
;
4365 struct cgroup_event
*event
, *tmp
;
4366 struct cgroup_subsys
*ss
;
4369 lockdep_assert_held(&d
->d_inode
->i_mutex
);
4370 lockdep_assert_held(&cgroup_mutex
);
4373 * css_set_lock prevents @cgrp from being removed while
4374 * __put_css_set() is in progress.
4376 read_lock(&css_set_lock
);
4377 empty
= !atomic_read(&cgrp
->count
) && list_empty(&cgrp
->children
);
4378 read_unlock(&css_set_lock
);
4383 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4384 * removed. This makes future css_tryget() and child creation
4385 * attempts fail thus maintaining the removal conditions verified
4388 * Note that CGRP_DEAD assertion is depended upon by
4389 * cgroup_next_sibling() to resume iteration after dropping RCU
4390 * read lock. See cgroup_next_sibling() for details.
4392 for_each_subsys(cgrp
->root
, ss
) {
4393 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4395 WARN_ON(atomic_read(&css
->refcnt
) < 0);
4396 atomic_add(CSS_DEACT_BIAS
, &css
->refcnt
);
4398 set_bit(CGRP_DEAD
, &cgrp
->flags
);
4400 /* tell subsystems to initate destruction */
4401 for_each_subsys(cgrp
->root
, ss
)
4402 offline_css(ss
, cgrp
);
4405 * Put all the base refs. Each css holds an extra reference to the
4406 * cgroup's dentry and cgroup removal proceeds regardless of css
4407 * refs. On the last put of each css, whenever that may be, the
4408 * extra dentry ref is put so that dentry destruction happens only
4409 * after all css's are released.
4411 for_each_subsys(cgrp
->root
, ss
)
4412 css_put(cgrp
->subsys
[ss
->subsys_id
]);
4414 raw_spin_lock(&release_list_lock
);
4415 if (!list_empty(&cgrp
->release_list
))
4416 list_del_init(&cgrp
->release_list
);
4417 raw_spin_unlock(&release_list_lock
);
4419 /* delete this cgroup from parent->children */
4420 list_del_rcu(&cgrp
->sibling
);
4421 list_del_init(&cgrp
->allcg_node
);
4424 cgroup_d_remove_dir(d
);
4427 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4428 check_for_release(parent
);
4431 * Unregister events and notify userspace.
4432 * Notify userspace about cgroup removing only after rmdir of cgroup
4433 * directory to avoid race between userspace and kernelspace.
4435 spin_lock(&cgrp
->event_list_lock
);
4436 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4437 list_del_init(&event
->list
);
4438 schedule_work(&event
->remove
);
4440 spin_unlock(&cgrp
->event_list_lock
);
4445 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4449 mutex_lock(&cgroup_mutex
);
4450 ret
= cgroup_destroy_locked(dentry
->d_fsdata
);
4451 mutex_unlock(&cgroup_mutex
);
4456 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4458 INIT_LIST_HEAD(&ss
->cftsets
);
4461 * base_cftset is embedded in subsys itself, no need to worry about
4464 if (ss
->base_cftypes
) {
4465 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4466 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4470 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4472 struct cgroup_subsys_state
*css
;
4474 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4476 mutex_lock(&cgroup_mutex
);
4478 /* init base cftset */
4479 cgroup_init_cftsets(ss
);
4481 /* Create the top cgroup state for this subsystem */
4482 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4483 ss
->root
= &rootnode
;
4484 css
= ss
->css_alloc(dummytop
);
4485 /* We don't handle early failures gracefully */
4486 BUG_ON(IS_ERR(css
));
4487 init_cgroup_css(css
, ss
, dummytop
);
4489 /* Update the init_css_set to contain a subsys
4490 * pointer to this state - since the subsystem is
4491 * newly registered, all tasks and hence the
4492 * init_css_set is in the subsystem's top cgroup. */
4493 init_css_set
.subsys
[ss
->subsys_id
] = css
;
4495 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4497 /* At system boot, before all subsystems have been
4498 * registered, no tasks have been forked, so we don't
4499 * need to invoke fork callbacks here. */
4500 BUG_ON(!list_empty(&init_task
.tasks
));
4502 BUG_ON(online_css(ss
, dummytop
));
4504 mutex_unlock(&cgroup_mutex
);
4506 /* this function shouldn't be used with modular subsystems, since they
4507 * need to register a subsys_id, among other things */
4512 * cgroup_load_subsys: load and register a modular subsystem at runtime
4513 * @ss: the subsystem to load
4515 * This function should be called in a modular subsystem's initcall. If the
4516 * subsystem is built as a module, it will be assigned a new subsys_id and set
4517 * up for use. If the subsystem is built-in anyway, work is delegated to the
4518 * simpler cgroup_init_subsys.
4520 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4522 struct cgroup_subsys_state
*css
;
4524 struct hlist_node
*tmp
;
4525 struct css_set
*cset
;
4528 /* check name and function validity */
4529 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4530 ss
->css_alloc
== NULL
|| ss
->css_free
== NULL
)
4534 * we don't support callbacks in modular subsystems. this check is
4535 * before the ss->module check for consistency; a subsystem that could
4536 * be a module should still have no callbacks even if the user isn't
4537 * compiling it as one.
4539 if (ss
->fork
|| ss
->exit
)
4543 * an optionally modular subsystem is built-in: we want to do nothing,
4544 * since cgroup_init_subsys will have already taken care of it.
4546 if (ss
->module
== NULL
) {
4547 /* a sanity check */
4548 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4552 /* init base cftset */
4553 cgroup_init_cftsets(ss
);
4555 mutex_lock(&cgroup_mutex
);
4556 subsys
[ss
->subsys_id
] = ss
;
4559 * no ss->css_alloc seems to need anything important in the ss
4560 * struct, so this can happen first (i.e. before the rootnode
4563 css
= ss
->css_alloc(dummytop
);
4565 /* failure case - need to deassign the subsys[] slot. */
4566 subsys
[ss
->subsys_id
] = NULL
;
4567 mutex_unlock(&cgroup_mutex
);
4568 return PTR_ERR(css
);
4571 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4572 ss
->root
= &rootnode
;
4574 /* our new subsystem will be attached to the dummy hierarchy. */
4575 init_cgroup_css(css
, ss
, dummytop
);
4576 /* init_idr must be after init_cgroup_css because it sets css->id. */
4578 ret
= cgroup_init_idr(ss
, css
);
4584 * Now we need to entangle the css into the existing css_sets. unlike
4585 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4586 * will need a new pointer to it; done by iterating the css_set_table.
4587 * furthermore, modifying the existing css_sets will corrupt the hash
4588 * table state, so each changed css_set will need its hash recomputed.
4589 * this is all done under the css_set_lock.
4591 write_lock(&css_set_lock
);
4592 hash_for_each_safe(css_set_table
, i
, tmp
, cset
, hlist
) {
4593 /* skip entries that we already rehashed */
4594 if (cset
->subsys
[ss
->subsys_id
])
4596 /* remove existing entry */
4597 hash_del(&cset
->hlist
);
4599 cset
->subsys
[ss
->subsys_id
] = css
;
4600 /* recompute hash and restore entry */
4601 key
= css_set_hash(cset
->subsys
);
4602 hash_add(css_set_table
, &cset
->hlist
, key
);
4604 write_unlock(&css_set_lock
);
4606 ret
= online_css(ss
, dummytop
);
4611 mutex_unlock(&cgroup_mutex
);
4615 mutex_unlock(&cgroup_mutex
);
4616 /* @ss can't be mounted here as try_module_get() would fail */
4617 cgroup_unload_subsys(ss
);
4620 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4623 * cgroup_unload_subsys: unload a modular subsystem
4624 * @ss: the subsystem to unload
4626 * This function should be called in a modular subsystem's exitcall. When this
4627 * function is invoked, the refcount on the subsystem's module will be 0, so
4628 * the subsystem will not be attached to any hierarchy.
4630 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4632 struct cgrp_cset_link
*link
;
4634 BUG_ON(ss
->module
== NULL
);
4637 * we shouldn't be called if the subsystem is in use, and the use of
4638 * try_module_get in parse_cgroupfs_options should ensure that it
4639 * doesn't start being used while we're killing it off.
4641 BUG_ON(ss
->root
!= &rootnode
);
4643 mutex_lock(&cgroup_mutex
);
4645 offline_css(ss
, dummytop
);
4648 idr_destroy(&ss
->idr
);
4650 /* deassign the subsys_id */
4651 subsys
[ss
->subsys_id
] = NULL
;
4653 /* remove subsystem from rootnode's list of subsystems */
4654 list_del_init(&ss
->sibling
);
4657 * disentangle the css from all css_sets attached to the dummytop. as
4658 * in loading, we need to pay our respects to the hashtable gods.
4660 write_lock(&css_set_lock
);
4661 list_for_each_entry(link
, &dummytop
->cset_links
, cset_link
) {
4662 struct css_set
*cset
= link
->cset
;
4665 hash_del(&cset
->hlist
);
4666 cset
->subsys
[ss
->subsys_id
] = NULL
;
4667 key
= css_set_hash(cset
->subsys
);
4668 hash_add(css_set_table
, &cset
->hlist
, key
);
4670 write_unlock(&css_set_lock
);
4673 * remove subsystem's css from the dummytop and free it - need to
4674 * free before marking as null because ss->css_free needs the
4675 * cgrp->subsys pointer to find their state. note that this also
4676 * takes care of freeing the css_id.
4678 ss
->css_free(dummytop
);
4679 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4681 mutex_unlock(&cgroup_mutex
);
4683 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4686 * cgroup_init_early - cgroup initialization at system boot
4688 * Initialize cgroups at system boot, and initialize any
4689 * subsystems that request early init.
4691 int __init
cgroup_init_early(void)
4694 atomic_set(&init_css_set
.refcount
, 1);
4695 INIT_LIST_HEAD(&init_css_set
.cgrp_links
);
4696 INIT_LIST_HEAD(&init_css_set
.tasks
);
4697 INIT_HLIST_NODE(&init_css_set
.hlist
);
4699 init_cgroup_root(&rootnode
);
4701 init_task
.cgroups
= &init_css_set
;
4703 init_cgrp_cset_link
.cset
= &init_css_set
;
4704 init_cgrp_cset_link
.cgrp
= dummytop
;
4705 list_add(&init_cgrp_cset_link
.cset_link
, &rootnode
.top_cgroup
.cset_links
);
4706 list_add(&init_cgrp_cset_link
.cgrp_link
, &init_css_set
.cgrp_links
);
4708 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4709 struct cgroup_subsys
*ss
= subsys
[i
];
4711 /* at bootup time, we don't worry about modular subsystems */
4712 if (!ss
|| ss
->module
)
4716 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4717 BUG_ON(!ss
->css_alloc
);
4718 BUG_ON(!ss
->css_free
);
4719 if (ss
->subsys_id
!= i
) {
4720 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4721 ss
->name
, ss
->subsys_id
);
4726 cgroup_init_subsys(ss
);
4732 * cgroup_init - cgroup initialization
4734 * Register cgroup filesystem and /proc file, and initialize
4735 * any subsystems that didn't request early init.
4737 int __init
cgroup_init(void)
4743 err
= bdi_init(&cgroup_backing_dev_info
);
4747 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4748 struct cgroup_subsys
*ss
= subsys
[i
];
4750 /* at bootup time, we don't worry about modular subsystems */
4751 if (!ss
|| ss
->module
)
4753 if (!ss
->early_init
)
4754 cgroup_init_subsys(ss
);
4756 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4759 /* Add init_css_set to the hash table */
4760 key
= css_set_hash(init_css_set
.subsys
);
4761 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4763 /* allocate id for the dummy hierarchy */
4764 mutex_lock(&cgroup_mutex
);
4765 mutex_lock(&cgroup_root_mutex
);
4767 BUG_ON(cgroup_init_root_id(&rootnode
));
4769 mutex_unlock(&cgroup_root_mutex
);
4770 mutex_unlock(&cgroup_mutex
);
4772 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4778 err
= register_filesystem(&cgroup_fs_type
);
4780 kobject_put(cgroup_kobj
);
4784 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4788 bdi_destroy(&cgroup_backing_dev_info
);
4794 * proc_cgroup_show()
4795 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4796 * - Used for /proc/<pid>/cgroup.
4797 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4798 * doesn't really matter if tsk->cgroup changes after we read it,
4799 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4800 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4801 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4802 * cgroup to top_cgroup.
4805 /* TODO: Use a proper seq_file iterator */
4806 int proc_cgroup_show(struct seq_file
*m
, void *v
)
4809 struct task_struct
*tsk
;
4812 struct cgroupfs_root
*root
;
4815 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4821 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4827 mutex_lock(&cgroup_mutex
);
4829 for_each_active_root(root
) {
4830 struct cgroup_subsys
*ss
;
4831 struct cgroup
*cgrp
;
4834 seq_printf(m
, "%d:", root
->hierarchy_id
);
4835 for_each_subsys(root
, ss
)
4836 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4837 if (strlen(root
->name
))
4838 seq_printf(m
, "%sname=%s", count
? "," : "",
4841 cgrp
= task_cgroup_from_root(tsk
, root
);
4842 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4850 mutex_unlock(&cgroup_mutex
);
4851 put_task_struct(tsk
);
4858 /* Display information about each subsystem and each hierarchy */
4859 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4863 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4865 * ideally we don't want subsystems moving around while we do this.
4866 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4867 * subsys/hierarchy state.
4869 mutex_lock(&cgroup_mutex
);
4870 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4871 struct cgroup_subsys
*ss
= subsys
[i
];
4874 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4875 ss
->name
, ss
->root
->hierarchy_id
,
4876 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4878 mutex_unlock(&cgroup_mutex
);
4882 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4884 return single_open(file
, proc_cgroupstats_show
, NULL
);
4887 static const struct file_operations proc_cgroupstats_operations
= {
4888 .open
= cgroupstats_open
,
4890 .llseek
= seq_lseek
,
4891 .release
= single_release
,
4895 * cgroup_fork - attach newly forked task to its parents cgroup.
4896 * @child: pointer to task_struct of forking parent process.
4898 * Description: A task inherits its parent's cgroup at fork().
4900 * A pointer to the shared css_set was automatically copied in
4901 * fork.c by dup_task_struct(). However, we ignore that copy, since
4902 * it was not made under the protection of RCU or cgroup_mutex, so
4903 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4904 * have already changed current->cgroups, allowing the previously
4905 * referenced cgroup group to be removed and freed.
4907 * At the point that cgroup_fork() is called, 'current' is the parent
4908 * task, and the passed argument 'child' points to the child task.
4910 void cgroup_fork(struct task_struct
*child
)
4913 child
->cgroups
= current
->cgroups
;
4914 get_css_set(child
->cgroups
);
4915 task_unlock(current
);
4916 INIT_LIST_HEAD(&child
->cg_list
);
4920 * cgroup_post_fork - called on a new task after adding it to the task list
4921 * @child: the task in question
4923 * Adds the task to the list running through its css_set if necessary and
4924 * call the subsystem fork() callbacks. Has to be after the task is
4925 * visible on the task list in case we race with the first call to
4926 * cgroup_iter_start() - to guarantee that the new task ends up on its
4929 void cgroup_post_fork(struct task_struct
*child
)
4934 * use_task_css_set_links is set to 1 before we walk the tasklist
4935 * under the tasklist_lock and we read it here after we added the child
4936 * to the tasklist under the tasklist_lock as well. If the child wasn't
4937 * yet in the tasklist when we walked through it from
4938 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4939 * should be visible now due to the paired locking and barriers implied
4940 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4941 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4944 if (use_task_css_set_links
) {
4945 write_lock(&css_set_lock
);
4947 if (list_empty(&child
->cg_list
))
4948 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4950 write_unlock(&css_set_lock
);
4954 * Call ss->fork(). This must happen after @child is linked on
4955 * css_set; otherwise, @child might change state between ->fork()
4956 * and addition to css_set.
4958 if (need_forkexit_callback
) {
4960 * fork/exit callbacks are supported only for builtin
4961 * subsystems, and the builtin section of the subsys
4962 * array is immutable, so we don't need to lock the
4963 * subsys array here. On the other hand, modular section
4964 * of the array can be freed at module unload, so we
4967 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4968 struct cgroup_subsys
*ss
= subsys
[i
];
4977 * cgroup_exit - detach cgroup from exiting task
4978 * @tsk: pointer to task_struct of exiting process
4979 * @run_callback: run exit callbacks?
4981 * Description: Detach cgroup from @tsk and release it.
4983 * Note that cgroups marked notify_on_release force every task in
4984 * them to take the global cgroup_mutex mutex when exiting.
4985 * This could impact scaling on very large systems. Be reluctant to
4986 * use notify_on_release cgroups where very high task exit scaling
4987 * is required on large systems.
4989 * the_top_cgroup_hack:
4991 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4993 * We call cgroup_exit() while the task is still competent to
4994 * handle notify_on_release(), then leave the task attached to the
4995 * root cgroup in each hierarchy for the remainder of its exit.
4997 * To do this properly, we would increment the reference count on
4998 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4999 * code we would add a second cgroup function call, to drop that
5000 * reference. This would just create an unnecessary hot spot on
5001 * the top_cgroup reference count, to no avail.
5003 * Normally, holding a reference to a cgroup without bumping its
5004 * count is unsafe. The cgroup could go away, or someone could
5005 * attach us to a different cgroup, decrementing the count on
5006 * the first cgroup that we never incremented. But in this case,
5007 * top_cgroup isn't going away, and either task has PF_EXITING set,
5008 * which wards off any cgroup_attach_task() attempts, or task is a failed
5009 * fork, never visible to cgroup_attach_task.
5011 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
5013 struct css_set
*cset
;
5017 * Unlink from the css_set task list if necessary.
5018 * Optimistically check cg_list before taking
5021 if (!list_empty(&tsk
->cg_list
)) {
5022 write_lock(&css_set_lock
);
5023 if (!list_empty(&tsk
->cg_list
))
5024 list_del_init(&tsk
->cg_list
);
5025 write_unlock(&css_set_lock
);
5028 /* Reassign the task to the init_css_set. */
5030 cset
= tsk
->cgroups
;
5031 tsk
->cgroups
= &init_css_set
;
5033 if (run_callbacks
&& need_forkexit_callback
) {
5035 * fork/exit callbacks are supported only for builtin
5036 * subsystems, see cgroup_post_fork() for details.
5038 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
5039 struct cgroup_subsys
*ss
= subsys
[i
];
5042 struct cgroup
*old_cgrp
=
5043 rcu_dereference_raw(cset
->subsys
[i
])->cgroup
;
5044 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
5045 ss
->exit(cgrp
, old_cgrp
, tsk
);
5051 put_css_set_taskexit(cset
);
5054 static void check_for_release(struct cgroup
*cgrp
)
5056 if (cgroup_is_releasable(cgrp
) &&
5057 !atomic_read(&cgrp
->count
) && list_empty(&cgrp
->children
)) {
5059 * Control Group is currently removeable. If it's not
5060 * already queued for a userspace notification, queue
5063 int need_schedule_work
= 0;
5065 raw_spin_lock(&release_list_lock
);
5066 if (!cgroup_is_dead(cgrp
) &&
5067 list_empty(&cgrp
->release_list
)) {
5068 list_add(&cgrp
->release_list
, &release_list
);
5069 need_schedule_work
= 1;
5071 raw_spin_unlock(&release_list_lock
);
5072 if (need_schedule_work
)
5073 schedule_work(&release_agent_work
);
5077 /* Caller must verify that the css is not for root cgroup */
5078 bool __css_tryget(struct cgroup_subsys_state
*css
)
5083 v
= css_refcnt(css
);
5084 t
= atomic_cmpxchg(&css
->refcnt
, v
, v
+ 1);
5092 EXPORT_SYMBOL_GPL(__css_tryget
);
5094 /* Caller must verify that the css is not for root cgroup */
5095 void __css_put(struct cgroup_subsys_state
*css
)
5099 v
= css_unbias_refcnt(atomic_dec_return(&css
->refcnt
));
5101 schedule_work(&css
->dput_work
);
5103 EXPORT_SYMBOL_GPL(__css_put
);
5106 * Notify userspace when a cgroup is released, by running the
5107 * configured release agent with the name of the cgroup (path
5108 * relative to the root of cgroup file system) as the argument.
5110 * Most likely, this user command will try to rmdir this cgroup.
5112 * This races with the possibility that some other task will be
5113 * attached to this cgroup before it is removed, or that some other
5114 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5115 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5116 * unused, and this cgroup will be reprieved from its death sentence,
5117 * to continue to serve a useful existence. Next time it's released,
5118 * we will get notified again, if it still has 'notify_on_release' set.
5120 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5121 * means only wait until the task is successfully execve()'d. The
5122 * separate release agent task is forked by call_usermodehelper(),
5123 * then control in this thread returns here, without waiting for the
5124 * release agent task. We don't bother to wait because the caller of
5125 * this routine has no use for the exit status of the release agent
5126 * task, so no sense holding our caller up for that.
5128 static void cgroup_release_agent(struct work_struct
*work
)
5130 BUG_ON(work
!= &release_agent_work
);
5131 mutex_lock(&cgroup_mutex
);
5132 raw_spin_lock(&release_list_lock
);
5133 while (!list_empty(&release_list
)) {
5134 char *argv
[3], *envp
[3];
5136 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5137 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5140 list_del_init(&cgrp
->release_list
);
5141 raw_spin_unlock(&release_list_lock
);
5142 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5145 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5147 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5152 argv
[i
++] = agentbuf
;
5153 argv
[i
++] = pathbuf
;
5157 /* minimal command environment */
5158 envp
[i
++] = "HOME=/";
5159 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5162 /* Drop the lock while we invoke the usermode helper,
5163 * since the exec could involve hitting disk and hence
5164 * be a slow process */
5165 mutex_unlock(&cgroup_mutex
);
5166 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5167 mutex_lock(&cgroup_mutex
);
5171 raw_spin_lock(&release_list_lock
);
5173 raw_spin_unlock(&release_list_lock
);
5174 mutex_unlock(&cgroup_mutex
);
5177 static int __init
cgroup_disable(char *str
)
5182 while ((token
= strsep(&str
, ",")) != NULL
) {
5185 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
5186 struct cgroup_subsys
*ss
= subsys
[i
];
5189 * cgroup_disable, being at boot time, can't
5190 * know about module subsystems, so we don't
5193 if (!ss
|| ss
->module
)
5196 if (!strcmp(token
, ss
->name
)) {
5198 printk(KERN_INFO
"Disabling %s control group"
5199 " subsystem\n", ss
->name
);
5206 __setup("cgroup_disable=", cgroup_disable
);
5209 * Functons for CSS ID.
5212 /* to get ID other than 0, this should be called when !cgroup_is_dead() */
5213 unsigned short css_id(struct cgroup_subsys_state
*css
)
5215 struct css_id
*cssid
;
5218 * This css_id() can return correct value when somone has refcnt
5219 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5220 * it's unchanged until freed.
5222 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5228 EXPORT_SYMBOL_GPL(css_id
);
5231 * css_is_ancestor - test "root" css is an ancestor of "child"
5232 * @child: the css to be tested.
5233 * @root: the css supporsed to be an ancestor of the child.
5235 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5236 * this function reads css->id, the caller must hold rcu_read_lock().
5237 * But, considering usual usage, the csses should be valid objects after test.
5238 * Assuming that the caller will do some action to the child if this returns
5239 * returns true, the caller must take "child";s reference count.
5240 * If "child" is valid object and this returns true, "root" is valid, too.
5243 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5244 const struct cgroup_subsys_state
*root
)
5246 struct css_id
*child_id
;
5247 struct css_id
*root_id
;
5249 child_id
= rcu_dereference(child
->id
);
5252 root_id
= rcu_dereference(root
->id
);
5255 if (child_id
->depth
< root_id
->depth
)
5257 if (child_id
->stack
[root_id
->depth
] != root_id
->id
)
5262 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5264 struct css_id
*id
= css
->id
;
5265 /* When this is called before css_id initialization, id can be NULL */
5269 BUG_ON(!ss
->use_id
);
5271 rcu_assign_pointer(id
->css
, NULL
);
5272 rcu_assign_pointer(css
->id
, NULL
);
5273 spin_lock(&ss
->id_lock
);
5274 idr_remove(&ss
->idr
, id
->id
);
5275 spin_unlock(&ss
->id_lock
);
5276 kfree_rcu(id
, rcu_head
);
5278 EXPORT_SYMBOL_GPL(free_css_id
);
5281 * This is called by init or create(). Then, calls to this function are
5282 * always serialized (By cgroup_mutex() at create()).
5285 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5287 struct css_id
*newid
;
5290 BUG_ON(!ss
->use_id
);
5292 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5293 newid
= kzalloc(size
, GFP_KERNEL
);
5295 return ERR_PTR(-ENOMEM
);
5297 idr_preload(GFP_KERNEL
);
5298 spin_lock(&ss
->id_lock
);
5299 /* Don't use 0. allocates an ID of 1-65535 */
5300 ret
= idr_alloc(&ss
->idr
, newid
, 1, CSS_ID_MAX
+ 1, GFP_NOWAIT
);
5301 spin_unlock(&ss
->id_lock
);
5304 /* Returns error when there are no free spaces for new ID.*/
5309 newid
->depth
= depth
;
5313 return ERR_PTR(ret
);
5317 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5318 struct cgroup_subsys_state
*rootcss
)
5320 struct css_id
*newid
;
5322 spin_lock_init(&ss
->id_lock
);
5325 newid
= get_new_cssid(ss
, 0);
5327 return PTR_ERR(newid
);
5329 newid
->stack
[0] = newid
->id
;
5330 newid
->css
= rootcss
;
5331 rootcss
->id
= newid
;
5335 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5336 struct cgroup
*child
)
5338 int subsys_id
, i
, depth
= 0;
5339 struct cgroup_subsys_state
*parent_css
, *child_css
;
5340 struct css_id
*child_id
, *parent_id
;
5342 subsys_id
= ss
->subsys_id
;
5343 parent_css
= parent
->subsys
[subsys_id
];
5344 child_css
= child
->subsys
[subsys_id
];
5345 parent_id
= parent_css
->id
;
5346 depth
= parent_id
->depth
+ 1;
5348 child_id
= get_new_cssid(ss
, depth
);
5349 if (IS_ERR(child_id
))
5350 return PTR_ERR(child_id
);
5352 for (i
= 0; i
< depth
; i
++)
5353 child_id
->stack
[i
] = parent_id
->stack
[i
];
5354 child_id
->stack
[depth
] = child_id
->id
;
5356 * child_id->css pointer will be set after this cgroup is available
5357 * see cgroup_populate_dir()
5359 rcu_assign_pointer(child_css
->id
, child_id
);
5365 * css_lookup - lookup css by id
5366 * @ss: cgroup subsys to be looked into.
5369 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5370 * NULL if not. Should be called under rcu_read_lock()
5372 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5374 struct css_id
*cssid
= NULL
;
5376 BUG_ON(!ss
->use_id
);
5377 cssid
= idr_find(&ss
->idr
, id
);
5379 if (unlikely(!cssid
))
5382 return rcu_dereference(cssid
->css
);
5384 EXPORT_SYMBOL_GPL(css_lookup
);
5387 * get corresponding css from file open on cgroupfs directory
5389 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5391 struct cgroup
*cgrp
;
5392 struct inode
*inode
;
5393 struct cgroup_subsys_state
*css
;
5395 inode
= file_inode(f
);
5396 /* check in cgroup filesystem dir */
5397 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5398 return ERR_PTR(-EBADF
);
5400 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5401 return ERR_PTR(-EINVAL
);
5404 cgrp
= __d_cgrp(f
->f_dentry
);
5405 css
= cgrp
->subsys
[id
];
5406 return css
? css
: ERR_PTR(-ENOENT
);
5409 #ifdef CONFIG_CGROUP_DEBUG
5410 static struct cgroup_subsys_state
*debug_css_alloc(struct cgroup
*cont
)
5412 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5415 return ERR_PTR(-ENOMEM
);
5420 static void debug_css_free(struct cgroup
*cont
)
5422 kfree(cont
->subsys
[debug_subsys_id
]);
5425 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5427 return atomic_read(&cont
->count
);
5430 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5432 return cgroup_task_count(cont
);
5435 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5437 return (u64
)(unsigned long)current
->cgroups
;
5440 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5446 count
= atomic_read(¤t
->cgroups
->refcount
);
5451 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5453 struct seq_file
*seq
)
5455 struct cgrp_cset_link
*link
;
5456 struct css_set
*cset
;
5458 read_lock(&css_set_lock
);
5460 cset
= rcu_dereference(current
->cgroups
);
5461 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5462 struct cgroup
*c
= link
->cgrp
;
5466 name
= c
->dentry
->d_name
.name
;
5469 seq_printf(seq
, "Root %d group %s\n",
5470 c
->root
->hierarchy_id
, name
);
5473 read_unlock(&css_set_lock
);
5477 #define MAX_TASKS_SHOWN_PER_CSS 25
5478 static int cgroup_css_links_read(struct cgroup
*cont
,
5480 struct seq_file
*seq
)
5482 struct cgrp_cset_link
*link
;
5484 read_lock(&css_set_lock
);
5485 list_for_each_entry(link
, &cont
->cset_links
, cset_link
) {
5486 struct css_set
*cset
= link
->cset
;
5487 struct task_struct
*task
;
5489 seq_printf(seq
, "css_set %p\n", cset
);
5490 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5491 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5492 seq_puts(seq
, " ...\n");
5495 seq_printf(seq
, " task %d\n",
5496 task_pid_vnr(task
));
5500 read_unlock(&css_set_lock
);
5504 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5506 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5509 static struct cftype debug_files
[] = {
5511 .name
= "cgroup_refcount",
5512 .read_u64
= cgroup_refcount_read
,
5515 .name
= "taskcount",
5516 .read_u64
= debug_taskcount_read
,
5520 .name
= "current_css_set",
5521 .read_u64
= current_css_set_read
,
5525 .name
= "current_css_set_refcount",
5526 .read_u64
= current_css_set_refcount_read
,
5530 .name
= "current_css_set_cg_links",
5531 .read_seq_string
= current_css_set_cg_links_read
,
5535 .name
= "cgroup_css_links",
5536 .read_seq_string
= cgroup_css_links_read
,
5540 .name
= "releasable",
5541 .read_u64
= releasable_read
,
5547 struct cgroup_subsys debug_subsys
= {
5549 .css_alloc
= debug_css_alloc
,
5550 .css_free
= debug_css_free
,
5551 .subsys_id
= debug_subsys_id
,
5552 .base_cftypes
= debug_files
,
5554 #endif /* CONFIG_CGROUP_DEBUG */