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_removed(const struct cgroup
*cgrp
)
231 return test_bit(CGRP_REMOVED
, &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_removed(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
);
318 /* Link structure for associating css_set objects with cgroups */
319 struct cg_cgroup_link
{
321 * List running through cg_cgroup_links associated with a
322 * cgroup, anchored on cgroup->css_sets
324 struct list_head cgrp_link_list
;
327 * List running through cg_cgroup_links pointing at a
328 * single css_set object, anchored on css_set->cg_links
330 struct list_head cg_link_list
;
334 /* The default css_set - used by init and its children prior to any
335 * hierarchies being mounted. It contains a pointer to the root state
336 * for each subsystem. Also used to anchor the list of css_sets. Not
337 * reference-counted, to improve performance when child cgroups
338 * haven't been created.
341 static struct css_set init_css_set
;
342 static struct cg_cgroup_link init_css_set_link
;
344 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
345 struct cgroup_subsys_state
*css
);
347 /* css_set_lock protects the list of css_set objects, and the
348 * chain of tasks off each css_set. Nests outside task->alloc_lock
349 * due to cgroup_iter_start() */
350 static DEFINE_RWLOCK(css_set_lock
);
351 static int css_set_count
;
354 * hash table for cgroup groups. This improves the performance to find
355 * an existing css_set. This hash doesn't (currently) take into
356 * account cgroups in empty hierarchies.
358 #define CSS_SET_HASH_BITS 7
359 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
361 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
364 unsigned long key
= 0UL;
366 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
367 key
+= (unsigned long)css
[i
];
368 key
= (key
>> 16) ^ key
;
373 /* We don't maintain the lists running through each css_set to its
374 * task until after the first call to cgroup_iter_start(). This
375 * reduces the fork()/exit() overhead for people who have cgroups
376 * compiled into their kernel but not actually in use */
377 static int use_task_css_set_links __read_mostly
;
379 static void __put_css_set(struct css_set
*cg
, int taskexit
)
381 struct cg_cgroup_link
*link
;
382 struct cg_cgroup_link
*saved_link
;
384 * Ensure that the refcount doesn't hit zero while any readers
385 * can see it. Similar to atomic_dec_and_lock(), but for an
388 if (atomic_add_unless(&cg
->refcount
, -1, 1))
390 write_lock(&css_set_lock
);
391 if (!atomic_dec_and_test(&cg
->refcount
)) {
392 write_unlock(&css_set_lock
);
396 /* This css_set is dead. unlink it and release cgroup refcounts */
397 hash_del(&cg
->hlist
);
400 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
402 struct cgroup
*cgrp
= link
->cgrp
;
403 list_del(&link
->cg_link_list
);
404 list_del(&link
->cgrp_link_list
);
407 * We may not be holding cgroup_mutex, and if cgrp->count is
408 * dropped to 0 the cgroup can be destroyed at any time, hence
409 * rcu_read_lock is used to keep it alive.
412 if (atomic_dec_and_test(&cgrp
->count
) &&
413 notify_on_release(cgrp
)) {
415 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
416 check_for_release(cgrp
);
423 write_unlock(&css_set_lock
);
424 kfree_rcu(cg
, rcu_head
);
428 * refcounted get/put for css_set objects
430 static inline void get_css_set(struct css_set
*cg
)
432 atomic_inc(&cg
->refcount
);
435 static inline void put_css_set(struct css_set
*cg
)
437 __put_css_set(cg
, 0);
440 static inline void put_css_set_taskexit(struct css_set
*cg
)
442 __put_css_set(cg
, 1);
446 * compare_css_sets - helper function for find_existing_css_set().
447 * @cg: candidate css_set being tested
448 * @old_cg: existing css_set for a task
449 * @new_cgrp: cgroup that's being entered by the task
450 * @template: desired set of css pointers in css_set (pre-calculated)
452 * Returns true if "cg" matches "old_cg" except for the hierarchy
453 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
455 static bool compare_css_sets(struct css_set
*cg
,
456 struct css_set
*old_cg
,
457 struct cgroup
*new_cgrp
,
458 struct cgroup_subsys_state
*template[])
460 struct list_head
*l1
, *l2
;
462 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
463 /* Not all subsystems matched */
468 * Compare cgroup pointers in order to distinguish between
469 * different cgroups in heirarchies with no subsystems. We
470 * could get by with just this check alone (and skip the
471 * memcmp above) but on most setups the memcmp check will
472 * avoid the need for this more expensive check on almost all
477 l2
= &old_cg
->cg_links
;
479 struct cg_cgroup_link
*cgl1
, *cgl2
;
480 struct cgroup
*cg1
, *cg2
;
484 /* See if we reached the end - both lists are equal length. */
485 if (l1
== &cg
->cg_links
) {
486 BUG_ON(l2
!= &old_cg
->cg_links
);
489 BUG_ON(l2
== &old_cg
->cg_links
);
491 /* Locate the cgroups associated with these links. */
492 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
493 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
496 /* Hierarchies should be linked in the same order. */
497 BUG_ON(cg1
->root
!= cg2
->root
);
500 * If this hierarchy is the hierarchy of the cgroup
501 * that's changing, then we need to check that this
502 * css_set points to the new cgroup; if it's any other
503 * hierarchy, then this css_set should point to the
504 * same cgroup as the old css_set.
506 if (cg1
->root
== new_cgrp
->root
) {
518 * find_existing_css_set() is a helper for
519 * find_css_set(), and checks to see whether an existing
520 * css_set is suitable.
522 * oldcg: the cgroup group that we're using before the cgroup
525 * cgrp: the cgroup that we're moving into
527 * template: location in which to build the desired set of subsystem
528 * state objects for the new cgroup group
530 static struct css_set
*find_existing_css_set(
531 struct css_set
*oldcg
,
533 struct cgroup_subsys_state
*template[])
536 struct cgroupfs_root
*root
= cgrp
->root
;
541 * Build the set of subsystem state objects that we want to see in the
542 * new css_set. while subsystems can change globally, the entries here
543 * won't change, so no need for locking.
545 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
546 if (root
->subsys_mask
& (1UL << i
)) {
547 /* Subsystem is in this hierarchy. So we want
548 * the subsystem state from the new
550 template[i
] = cgrp
->subsys
[i
];
552 /* Subsystem is not in this hierarchy, so we
553 * don't want to change the subsystem state */
554 template[i
] = oldcg
->subsys
[i
];
558 key
= css_set_hash(template);
559 hash_for_each_possible(css_set_table
, cg
, hlist
, key
) {
560 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
563 /* This css_set matches what we need */
567 /* No existing cgroup group matched */
571 static void free_cg_links(struct list_head
*tmp
)
573 struct cg_cgroup_link
*link
;
574 struct cg_cgroup_link
*saved_link
;
576 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
577 list_del(&link
->cgrp_link_list
);
583 * allocate_cg_links() allocates "count" cg_cgroup_link structures
584 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
585 * success or a negative error
587 static int allocate_cg_links(int count
, struct list_head
*tmp
)
589 struct cg_cgroup_link
*link
;
592 for (i
= 0; i
< count
; i
++) {
593 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
598 list_add(&link
->cgrp_link_list
, tmp
);
604 * link_css_set - a helper function to link a css_set to a cgroup
605 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
606 * @cg: the css_set to be linked
607 * @cgrp: the destination cgroup
609 static void link_css_set(struct list_head
*tmp_cg_links
,
610 struct css_set
*cg
, struct cgroup
*cgrp
)
612 struct cg_cgroup_link
*link
;
614 BUG_ON(list_empty(tmp_cg_links
));
615 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
619 atomic_inc(&cgrp
->count
);
620 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
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
->cg_link_list
, &cg
->cg_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(
636 struct css_set
*oldcg
, struct cgroup
*cgrp
)
639 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
641 struct list_head tmp_cg_links
;
643 struct cg_cgroup_link
*link
;
646 /* First see if we already have a cgroup group that matches
648 read_lock(&css_set_lock
);
649 res
= find_existing_css_set(oldcg
, cgrp
, template);
652 read_unlock(&css_set_lock
);
657 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
661 /* Allocate all the cg_cgroup_link objects that we'll need */
662 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
667 atomic_set(&res
->refcount
, 1);
668 INIT_LIST_HEAD(&res
->cg_links
);
669 INIT_LIST_HEAD(&res
->tasks
);
670 INIT_HLIST_NODE(&res
->hlist
);
672 /* Copy the set of subsystem state objects generated in
673 * find_existing_css_set() */
674 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
676 write_lock(&css_set_lock
);
677 /* Add reference counts and links from the new css_set. */
678 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
679 struct cgroup
*c
= link
->cgrp
;
680 if (c
->root
== cgrp
->root
)
682 link_css_set(&tmp_cg_links
, res
, c
);
685 BUG_ON(!list_empty(&tmp_cg_links
));
689 /* Add this cgroup group to the hash table */
690 key
= css_set_hash(res
->subsys
);
691 hash_add(css_set_table
, &res
->hlist
, key
);
693 write_unlock(&css_set_lock
);
699 * Return the cgroup for "task" from the given hierarchy. Must be
700 * called with cgroup_mutex held.
702 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
703 struct cgroupfs_root
*root
)
706 struct cgroup
*res
= NULL
;
708 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
709 read_lock(&css_set_lock
);
711 * No need to lock the task - since we hold cgroup_mutex the
712 * task can't change groups, so the only thing that can happen
713 * is that it exits and its css is set back to init_css_set.
716 if (css
== &init_css_set
) {
717 res
= &root
->top_cgroup
;
719 struct cg_cgroup_link
*link
;
720 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
721 struct cgroup
*c
= link
->cgrp
;
722 if (c
->root
== root
) {
728 read_unlock(&css_set_lock
);
734 * There is one global cgroup mutex. We also require taking
735 * task_lock() when dereferencing a task's cgroup subsys pointers.
736 * See "The task_lock() exception", at the end of this comment.
738 * A task must hold cgroup_mutex to modify cgroups.
740 * Any task can increment and decrement the count field without lock.
741 * So in general, code holding cgroup_mutex can't rely on the count
742 * field not changing. However, if the count goes to zero, then only
743 * cgroup_attach_task() can increment it again. Because a count of zero
744 * means that no tasks are currently attached, therefore there is no
745 * way a task attached to that cgroup can fork (the other way to
746 * increment the count). So code holding cgroup_mutex can safely
747 * assume that if the count is zero, it will stay zero. Similarly, if
748 * a task holds cgroup_mutex on a cgroup with zero count, it
749 * knows that the cgroup won't be removed, as cgroup_rmdir()
752 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
753 * (usually) take cgroup_mutex. These are the two most performance
754 * critical pieces of code here. The exception occurs on cgroup_exit(),
755 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
756 * is taken, and if the cgroup count is zero, a usermode call made
757 * to the release agent with the name of the cgroup (path relative to
758 * the root of cgroup file system) as the argument.
760 * A cgroup can only be deleted if both its 'count' of using tasks
761 * is zero, and its list of 'children' cgroups is empty. Since all
762 * tasks in the system use _some_ cgroup, and since there is always at
763 * least one task in the system (init, pid == 1), therefore, top_cgroup
764 * always has either children cgroups and/or using tasks. So we don't
765 * need a special hack to ensure that top_cgroup cannot be deleted.
767 * The task_lock() exception
769 * The need for this exception arises from the action of
770 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
771 * another. It does so using cgroup_mutex, however there are
772 * several performance critical places that need to reference
773 * task->cgroup without the expense of grabbing a system global
774 * mutex. Therefore except as noted below, when dereferencing or, as
775 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
776 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
777 * the task_struct routinely used for such matters.
779 * P.S. One more locking exception. RCU is used to guard the
780 * update of a tasks cgroup pointer by cgroup_attach_task()
784 * A couple of forward declarations required, due to cyclic reference loop:
785 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
786 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
790 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
791 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, unsigned int);
792 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
793 static int cgroup_populate_dir(struct cgroup
*cgrp
, bool base_files
,
794 unsigned long subsys_mask
);
795 static const struct inode_operations cgroup_dir_inode_operations
;
796 static const struct file_operations proc_cgroupstats_operations
;
798 static struct backing_dev_info cgroup_backing_dev_info
= {
800 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
803 static int alloc_css_id(struct cgroup_subsys
*ss
,
804 struct cgroup
*parent
, struct cgroup
*child
);
806 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
808 struct inode
*inode
= new_inode(sb
);
811 inode
->i_ino
= get_next_ino();
812 inode
->i_mode
= mode
;
813 inode
->i_uid
= current_fsuid();
814 inode
->i_gid
= current_fsgid();
815 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
816 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
821 static struct cgroup_name
*cgroup_alloc_name(struct dentry
*dentry
)
823 struct cgroup_name
*name
;
825 name
= kmalloc(sizeof(*name
) + dentry
->d_name
.len
+ 1, GFP_KERNEL
);
828 strcpy(name
->name
, dentry
->d_name
.name
);
832 static void cgroup_free_fn(struct work_struct
*work
)
834 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, free_work
);
835 struct cgroup_subsys
*ss
;
837 mutex_lock(&cgroup_mutex
);
839 * Release the subsystem state objects.
841 for_each_subsys(cgrp
->root
, ss
)
844 cgrp
->root
->number_of_cgroups
--;
845 mutex_unlock(&cgroup_mutex
);
848 * We get a ref to the parent's dentry, and put the ref when
849 * this cgroup is being freed, so it's guaranteed that the
850 * parent won't be destroyed before its children.
852 dput(cgrp
->parent
->dentry
);
854 ida_simple_remove(&cgrp
->root
->cgroup_ida
, cgrp
->id
);
857 * Drop the active superblock reference that we took when we
858 * created the cgroup. This will free cgrp->root, if we are
859 * holding the last reference to @sb.
861 deactivate_super(cgrp
->root
->sb
);
864 * if we're getting rid of the cgroup, refcount should ensure
865 * that there are no pidlists left.
867 BUG_ON(!list_empty(&cgrp
->pidlists
));
869 simple_xattrs_free(&cgrp
->xattrs
);
871 kfree(rcu_dereference_raw(cgrp
->name
));
875 static void cgroup_free_rcu(struct rcu_head
*head
)
877 struct cgroup
*cgrp
= container_of(head
, struct cgroup
, rcu_head
);
879 schedule_work(&cgrp
->free_work
);
882 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
884 /* is dentry a directory ? if so, kfree() associated cgroup */
885 if (S_ISDIR(inode
->i_mode
)) {
886 struct cgroup
*cgrp
= dentry
->d_fsdata
;
888 BUG_ON(!(cgroup_is_removed(cgrp
)));
889 call_rcu(&cgrp
->rcu_head
, cgroup_free_rcu
);
891 struct cfent
*cfe
= __d_cfe(dentry
);
892 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
894 WARN_ONCE(!list_empty(&cfe
->node
) &&
895 cgrp
!= &cgrp
->root
->top_cgroup
,
896 "cfe still linked for %s\n", cfe
->type
->name
);
897 simple_xattrs_free(&cfe
->xattrs
);
903 static int cgroup_delete(const struct dentry
*d
)
908 static void remove_dir(struct dentry
*d
)
910 struct dentry
*parent
= dget(d
->d_parent
);
913 simple_rmdir(parent
->d_inode
, d
);
917 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
921 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
922 lockdep_assert_held(&cgroup_mutex
);
925 * If we're doing cleanup due to failure of cgroup_create(),
926 * the corresponding @cfe may not exist.
928 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
929 struct dentry
*d
= cfe
->dentry
;
931 if (cft
&& cfe
->type
!= cft
)
936 simple_unlink(cgrp
->dentry
->d_inode
, d
);
937 list_del_init(&cfe
->node
);
945 * cgroup_clear_directory - selective removal of base and subsystem files
946 * @dir: directory containing the files
947 * @base_files: true if the base files should be removed
948 * @subsys_mask: mask of the subsystem ids whose files should be removed
950 static void cgroup_clear_directory(struct dentry
*dir
, bool base_files
,
951 unsigned long subsys_mask
)
953 struct cgroup
*cgrp
= __d_cgrp(dir
);
954 struct cgroup_subsys
*ss
;
956 for_each_subsys(cgrp
->root
, ss
) {
957 struct cftype_set
*set
;
958 if (!test_bit(ss
->subsys_id
, &subsys_mask
))
960 list_for_each_entry(set
, &ss
->cftsets
, node
)
961 cgroup_addrm_files(cgrp
, NULL
, set
->cfts
, false);
964 while (!list_empty(&cgrp
->files
))
965 cgroup_rm_file(cgrp
, NULL
);
970 * NOTE : the dentry must have been dget()'ed
972 static void cgroup_d_remove_dir(struct dentry
*dentry
)
974 struct dentry
*parent
;
975 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
977 cgroup_clear_directory(dentry
, true, root
->subsys_mask
);
979 parent
= dentry
->d_parent
;
980 spin_lock(&parent
->d_lock
);
981 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
982 list_del_init(&dentry
->d_u
.d_child
);
983 spin_unlock(&dentry
->d_lock
);
984 spin_unlock(&parent
->d_lock
);
989 * Call with cgroup_mutex held. Drops reference counts on modules, including
990 * any duplicate ones that parse_cgroupfs_options took. If this function
991 * returns an error, no reference counts are touched.
993 static int rebind_subsystems(struct cgroupfs_root
*root
,
994 unsigned long final_subsys_mask
)
996 unsigned long added_mask
, removed_mask
;
997 struct cgroup
*cgrp
= &root
->top_cgroup
;
1000 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1001 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1003 removed_mask
= root
->actual_subsys_mask
& ~final_subsys_mask
;
1004 added_mask
= final_subsys_mask
& ~root
->actual_subsys_mask
;
1005 /* Check that any added subsystems are currently free */
1006 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1007 unsigned long bit
= 1UL << i
;
1008 struct cgroup_subsys
*ss
= subsys
[i
];
1009 if (!(bit
& added_mask
))
1012 * Nobody should tell us to do a subsys that doesn't exist:
1013 * parse_cgroupfs_options should catch that case and refcounts
1014 * ensure that subsystems won't disappear once selected.
1017 if (ss
->root
!= &rootnode
) {
1018 /* Subsystem isn't free */
1023 /* Currently we don't handle adding/removing subsystems when
1024 * any child cgroups exist. This is theoretically supportable
1025 * but involves complex error handling, so it's being left until
1027 if (root
->number_of_cgroups
> 1)
1030 /* Process each subsystem */
1031 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1032 struct cgroup_subsys
*ss
= subsys
[i
];
1033 unsigned long bit
= 1UL << i
;
1034 if (bit
& added_mask
) {
1035 /* We're binding this subsystem to this hierarchy */
1037 BUG_ON(cgrp
->subsys
[i
]);
1038 BUG_ON(!dummytop
->subsys
[i
]);
1039 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
1040 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
1041 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1042 list_move(&ss
->sibling
, &root
->subsys_list
);
1046 /* refcount was already taken, and we're keeping it */
1047 } else if (bit
& removed_mask
) {
1048 /* We're removing this subsystem */
1050 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1051 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1054 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1055 cgrp
->subsys
[i
] = NULL
;
1056 subsys
[i
]->root
= &rootnode
;
1057 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1058 /* subsystem is now free - drop reference on module */
1059 module_put(ss
->module
);
1060 } else if (bit
& final_subsys_mask
) {
1061 /* Subsystem state should already exist */
1063 BUG_ON(!cgrp
->subsys
[i
]);
1065 * a refcount was taken, but we already had one, so
1066 * drop the extra reference.
1068 module_put(ss
->module
);
1069 #ifdef CONFIG_MODULE_UNLOAD
1070 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1073 /* Subsystem state shouldn't exist */
1074 BUG_ON(cgrp
->subsys
[i
]);
1077 root
->subsys_mask
= root
->actual_subsys_mask
= final_subsys_mask
;
1082 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1084 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1085 struct cgroup_subsys
*ss
;
1087 mutex_lock(&cgroup_root_mutex
);
1088 for_each_subsys(root
, ss
)
1089 seq_printf(seq
, ",%s", ss
->name
);
1090 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
)
1091 seq_puts(seq
, ",sane_behavior");
1092 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1093 seq_puts(seq
, ",noprefix");
1094 if (root
->flags
& CGRP_ROOT_XATTR
)
1095 seq_puts(seq
, ",xattr");
1096 if (strlen(root
->release_agent_path
))
1097 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1098 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
))
1099 seq_puts(seq
, ",clone_children");
1100 if (strlen(root
->name
))
1101 seq_printf(seq
, ",name=%s", root
->name
);
1102 mutex_unlock(&cgroup_root_mutex
);
1106 struct cgroup_sb_opts
{
1107 unsigned long subsys_mask
;
1108 unsigned long flags
;
1109 char *release_agent
;
1110 bool cpuset_clone_children
;
1112 /* User explicitly requested empty subsystem */
1115 struct cgroupfs_root
*new_root
;
1120 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1121 * with cgroup_mutex held to protect the subsys[] array. This function takes
1122 * refcounts on subsystems to be used, unless it returns error, in which case
1123 * no refcounts are taken.
1125 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1127 char *token
, *o
= data
;
1128 bool all_ss
= false, one_ss
= false;
1129 unsigned long mask
= (unsigned long)-1;
1131 bool module_pin_failed
= false;
1133 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1135 #ifdef CONFIG_CPUSETS
1136 mask
= ~(1UL << cpuset_subsys_id
);
1139 memset(opts
, 0, sizeof(*opts
));
1141 while ((token
= strsep(&o
, ",")) != NULL
) {
1144 if (!strcmp(token
, "none")) {
1145 /* Explicitly have no subsystems */
1149 if (!strcmp(token
, "all")) {
1150 /* Mutually exclusive option 'all' + subsystem name */
1156 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1157 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1160 if (!strcmp(token
, "noprefix")) {
1161 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1164 if (!strcmp(token
, "clone_children")) {
1165 opts
->cpuset_clone_children
= true;
1168 if (!strcmp(token
, "xattr")) {
1169 opts
->flags
|= CGRP_ROOT_XATTR
;
1172 if (!strncmp(token
, "release_agent=", 14)) {
1173 /* Specifying two release agents is forbidden */
1174 if (opts
->release_agent
)
1176 opts
->release_agent
=
1177 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1178 if (!opts
->release_agent
)
1182 if (!strncmp(token
, "name=", 5)) {
1183 const char *name
= token
+ 5;
1184 /* Can't specify an empty name */
1187 /* Must match [\w.-]+ */
1188 for (i
= 0; i
< strlen(name
); i
++) {
1192 if ((c
== '.') || (c
== '-') || (c
== '_'))
1196 /* Specifying two names is forbidden */
1199 opts
->name
= kstrndup(name
,
1200 MAX_CGROUP_ROOT_NAMELEN
- 1,
1208 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1209 struct cgroup_subsys
*ss
= subsys
[i
];
1212 if (strcmp(token
, ss
->name
))
1217 /* Mutually exclusive option 'all' + subsystem name */
1220 set_bit(i
, &opts
->subsys_mask
);
1225 if (i
== CGROUP_SUBSYS_COUNT
)
1230 * If the 'all' option was specified select all the subsystems,
1231 * otherwise if 'none', 'name=' and a subsystem name options
1232 * were not specified, let's default to 'all'
1234 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
)) {
1235 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1236 struct cgroup_subsys
*ss
= subsys
[i
];
1241 set_bit(i
, &opts
->subsys_mask
);
1245 /* Consistency checks */
1247 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1248 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1250 if (opts
->flags
& CGRP_ROOT_NOPREFIX
) {
1251 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1255 if (opts
->cpuset_clone_children
) {
1256 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1262 * Option noprefix was introduced just for backward compatibility
1263 * with the old cpuset, so we allow noprefix only if mounting just
1264 * the cpuset subsystem.
1266 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1270 /* Can't specify "none" and some subsystems */
1271 if (opts
->subsys_mask
&& opts
->none
)
1275 * We either have to specify by name or by subsystems. (So all
1276 * empty hierarchies must have a name).
1278 if (!opts
->subsys_mask
&& !opts
->name
)
1282 * Grab references on all the modules we'll need, so the subsystems
1283 * don't dance around before rebind_subsystems attaches them. This may
1284 * take duplicate reference counts on a subsystem that's already used,
1285 * but rebind_subsystems handles this case.
1287 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1288 unsigned long bit
= 1UL << i
;
1290 if (!(bit
& opts
->subsys_mask
))
1292 if (!try_module_get(subsys
[i
]->module
)) {
1293 module_pin_failed
= true;
1297 if (module_pin_failed
) {
1299 * oops, one of the modules was going away. this means that we
1300 * raced with a module_delete call, and to the user this is
1301 * essentially a "subsystem doesn't exist" case.
1303 for (i
--; i
>= 0; i
--) {
1304 /* drop refcounts only on the ones we took */
1305 unsigned long bit
= 1UL << i
;
1307 if (!(bit
& opts
->subsys_mask
))
1309 module_put(subsys
[i
]->module
);
1317 static void drop_parsed_module_refcounts(unsigned long subsys_mask
)
1320 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1321 unsigned long bit
= 1UL << i
;
1323 if (!(bit
& subsys_mask
))
1325 module_put(subsys
[i
]->module
);
1329 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1332 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1333 struct cgroup
*cgrp
= &root
->top_cgroup
;
1334 struct cgroup_sb_opts opts
;
1335 unsigned long added_mask
, removed_mask
;
1337 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1338 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1342 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1343 mutex_lock(&cgroup_mutex
);
1344 mutex_lock(&cgroup_root_mutex
);
1346 /* See what subsystems are wanted */
1347 ret
= parse_cgroupfs_options(data
, &opts
);
1351 if (opts
.subsys_mask
!= root
->actual_subsys_mask
|| opts
.release_agent
)
1352 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1353 task_tgid_nr(current
), current
->comm
);
1355 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1356 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1358 /* Don't allow flags or name to change at remount */
1359 if (opts
.flags
!= root
->flags
||
1360 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1362 drop_parsed_module_refcounts(opts
.subsys_mask
);
1367 * Clear out the files of subsystems that should be removed, do
1368 * this before rebind_subsystems, since rebind_subsystems may
1369 * change this hierarchy's subsys_list.
1371 cgroup_clear_directory(cgrp
->dentry
, false, removed_mask
);
1373 ret
= rebind_subsystems(root
, opts
.subsys_mask
);
1375 /* rebind_subsystems failed, re-populate the removed files */
1376 cgroup_populate_dir(cgrp
, false, removed_mask
);
1377 drop_parsed_module_refcounts(opts
.subsys_mask
);
1381 /* re-populate subsystem files */
1382 cgroup_populate_dir(cgrp
, false, added_mask
);
1384 if (opts
.release_agent
)
1385 strcpy(root
->release_agent_path
, opts
.release_agent
);
1387 kfree(opts
.release_agent
);
1389 mutex_unlock(&cgroup_root_mutex
);
1390 mutex_unlock(&cgroup_mutex
);
1391 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1395 static const struct super_operations cgroup_ops
= {
1396 .statfs
= simple_statfs
,
1397 .drop_inode
= generic_delete_inode
,
1398 .show_options
= cgroup_show_options
,
1399 .remount_fs
= cgroup_remount
,
1402 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1404 INIT_LIST_HEAD(&cgrp
->sibling
);
1405 INIT_LIST_HEAD(&cgrp
->children
);
1406 INIT_LIST_HEAD(&cgrp
->files
);
1407 INIT_LIST_HEAD(&cgrp
->css_sets
);
1408 INIT_LIST_HEAD(&cgrp
->allcg_node
);
1409 INIT_LIST_HEAD(&cgrp
->release_list
);
1410 INIT_LIST_HEAD(&cgrp
->pidlists
);
1411 INIT_WORK(&cgrp
->free_work
, cgroup_free_fn
);
1412 mutex_init(&cgrp
->pidlist_mutex
);
1413 INIT_LIST_HEAD(&cgrp
->event_list
);
1414 spin_lock_init(&cgrp
->event_list_lock
);
1415 simple_xattrs_init(&cgrp
->xattrs
);
1418 static void init_cgroup_root(struct cgroupfs_root
*root
)
1420 struct cgroup
*cgrp
= &root
->top_cgroup
;
1422 INIT_LIST_HEAD(&root
->subsys_list
);
1423 INIT_LIST_HEAD(&root
->root_list
);
1424 INIT_LIST_HEAD(&root
->allcg_list
);
1425 root
->number_of_cgroups
= 1;
1427 cgrp
->name
= &root_cgroup_name
;
1428 init_cgroup_housekeeping(cgrp
);
1429 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
1432 static int cgroup_init_root_id(struct cgroupfs_root
*root
)
1436 lockdep_assert_held(&cgroup_mutex
);
1437 lockdep_assert_held(&cgroup_root_mutex
);
1439 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 2, 0, GFP_KERNEL
);
1443 root
->hierarchy_id
= id
;
1447 static void cgroup_exit_root_id(struct cgroupfs_root
*root
)
1449 lockdep_assert_held(&cgroup_mutex
);
1450 lockdep_assert_held(&cgroup_root_mutex
);
1452 if (root
->hierarchy_id
) {
1453 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1454 root
->hierarchy_id
= 0;
1458 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1460 struct cgroup_sb_opts
*opts
= data
;
1461 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1463 /* If we asked for a name then it must match */
1464 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1468 * If we asked for subsystems (or explicitly for no
1469 * subsystems) then they must match
1471 if ((opts
->subsys_mask
|| opts
->none
)
1472 && (opts
->subsys_mask
!= root
->subsys_mask
))
1478 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1480 struct cgroupfs_root
*root
;
1482 if (!opts
->subsys_mask
&& !opts
->none
)
1485 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1487 return ERR_PTR(-ENOMEM
);
1489 init_cgroup_root(root
);
1491 root
->subsys_mask
= opts
->subsys_mask
;
1492 root
->flags
= opts
->flags
;
1493 ida_init(&root
->cgroup_ida
);
1494 if (opts
->release_agent
)
1495 strcpy(root
->release_agent_path
, opts
->release_agent
);
1497 strcpy(root
->name
, opts
->name
);
1498 if (opts
->cpuset_clone_children
)
1499 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1503 static void cgroup_free_root(struct cgroupfs_root
*root
)
1506 /* hierarhcy ID shoulid already have been released */
1507 WARN_ON_ONCE(root
->hierarchy_id
);
1509 ida_destroy(&root
->cgroup_ida
);
1514 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1517 struct cgroup_sb_opts
*opts
= data
;
1519 /* If we don't have a new root, we can't set up a new sb */
1520 if (!opts
->new_root
)
1523 BUG_ON(!opts
->subsys_mask
&& !opts
->none
);
1525 ret
= set_anon_super(sb
, NULL
);
1529 sb
->s_fs_info
= opts
->new_root
;
1530 opts
->new_root
->sb
= sb
;
1532 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1533 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1534 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1535 sb
->s_op
= &cgroup_ops
;
1540 static int cgroup_get_rootdir(struct super_block
*sb
)
1542 static const struct dentry_operations cgroup_dops
= {
1543 .d_iput
= cgroup_diput
,
1544 .d_delete
= cgroup_delete
,
1547 struct inode
*inode
=
1548 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1553 inode
->i_fop
= &simple_dir_operations
;
1554 inode
->i_op
= &cgroup_dir_inode_operations
;
1555 /* directories start off with i_nlink == 2 (for "." entry) */
1557 sb
->s_root
= d_make_root(inode
);
1560 /* for everything else we want ->d_op set */
1561 sb
->s_d_op
= &cgroup_dops
;
1565 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1566 int flags
, const char *unused_dev_name
,
1569 struct cgroup_sb_opts opts
;
1570 struct cgroupfs_root
*root
;
1572 struct super_block
*sb
;
1573 struct cgroupfs_root
*new_root
;
1574 struct inode
*inode
;
1576 /* First find the desired set of subsystems */
1577 mutex_lock(&cgroup_mutex
);
1578 ret
= parse_cgroupfs_options(data
, &opts
);
1579 mutex_unlock(&cgroup_mutex
);
1584 * Allocate a new cgroup root. We may not need it if we're
1585 * reusing an existing hierarchy.
1587 new_root
= cgroup_root_from_opts(&opts
);
1588 if (IS_ERR(new_root
)) {
1589 ret
= PTR_ERR(new_root
);
1592 opts
.new_root
= new_root
;
1594 /* Locate an existing or new sb for this hierarchy */
1595 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, 0, &opts
);
1598 cgroup_free_root(opts
.new_root
);
1602 root
= sb
->s_fs_info
;
1604 if (root
== opts
.new_root
) {
1605 /* We used the new root structure, so this is a new hierarchy */
1606 struct list_head tmp_cg_links
;
1607 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1608 struct cgroupfs_root
*existing_root
;
1609 const struct cred
*cred
;
1613 BUG_ON(sb
->s_root
!= NULL
);
1615 ret
= cgroup_get_rootdir(sb
);
1617 goto drop_new_super
;
1618 inode
= sb
->s_root
->d_inode
;
1620 mutex_lock(&inode
->i_mutex
);
1621 mutex_lock(&cgroup_mutex
);
1622 mutex_lock(&cgroup_root_mutex
);
1624 /* Check for name clashes with existing mounts */
1626 if (strlen(root
->name
))
1627 for_each_active_root(existing_root
)
1628 if (!strcmp(existing_root
->name
, root
->name
))
1632 * We're accessing css_set_count without locking
1633 * css_set_lock here, but that's OK - it can only be
1634 * increased by someone holding cgroup_lock, and
1635 * that's us. The worst that can happen is that we
1636 * have some link structures left over
1638 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1642 ret
= cgroup_init_root_id(root
);
1646 ret
= rebind_subsystems(root
, root
->subsys_mask
);
1647 if (ret
== -EBUSY
) {
1648 free_cg_links(&tmp_cg_links
);
1652 * There must be no failure case after here, since rebinding
1653 * takes care of subsystems' refcounts, which are explicitly
1654 * dropped in the failure exit path.
1657 /* EBUSY should be the only error here */
1660 list_add(&root
->root_list
, &roots
);
1663 sb
->s_root
->d_fsdata
= root_cgrp
;
1664 root
->top_cgroup
.dentry
= sb
->s_root
;
1666 /* Link the top cgroup in this hierarchy into all
1667 * the css_set objects */
1668 write_lock(&css_set_lock
);
1669 hash_for_each(css_set_table
, i
, cg
, hlist
)
1670 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1671 write_unlock(&css_set_lock
);
1673 free_cg_links(&tmp_cg_links
);
1675 BUG_ON(!list_empty(&root_cgrp
->children
));
1676 BUG_ON(root
->number_of_cgroups
!= 1);
1678 cred
= override_creds(&init_cred
);
1679 cgroup_populate_dir(root_cgrp
, true, root
->subsys_mask
);
1681 mutex_unlock(&cgroup_root_mutex
);
1682 mutex_unlock(&cgroup_mutex
);
1683 mutex_unlock(&inode
->i_mutex
);
1686 * We re-used an existing hierarchy - the new root (if
1687 * any) is not needed
1689 cgroup_free_root(opts
.new_root
);
1691 if (((root
->flags
| opts
.flags
) & CGRP_ROOT_SANE_BEHAVIOR
) &&
1692 root
->flags
!= opts
.flags
) {
1693 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1695 goto drop_new_super
;
1698 /* no subsys rebinding, so refcounts don't change */
1699 drop_parsed_module_refcounts(opts
.subsys_mask
);
1702 kfree(opts
.release_agent
);
1704 return dget(sb
->s_root
);
1707 cgroup_exit_root_id(root
);
1708 mutex_unlock(&cgroup_root_mutex
);
1709 mutex_unlock(&cgroup_mutex
);
1710 mutex_unlock(&inode
->i_mutex
);
1712 deactivate_locked_super(sb
);
1714 drop_parsed_module_refcounts(opts
.subsys_mask
);
1716 kfree(opts
.release_agent
);
1718 return ERR_PTR(ret
);
1721 static void cgroup_kill_sb(struct super_block
*sb
) {
1722 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1723 struct cgroup
*cgrp
= &root
->top_cgroup
;
1725 struct cg_cgroup_link
*link
;
1726 struct cg_cgroup_link
*saved_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 css_sets to this hierarchy's
1745 write_lock(&css_set_lock
);
1747 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1749 list_del(&link
->cg_link_list
);
1750 list_del(&link
->cgrp_link_list
);
1753 write_unlock(&css_set_lock
);
1755 if (!list_empty(&root
->root_list
)) {
1756 list_del(&root
->root_list
);
1760 cgroup_exit_root_id(root
);
1762 mutex_unlock(&cgroup_root_mutex
);
1763 mutex_unlock(&cgroup_mutex
);
1765 simple_xattrs_free(&cgrp
->xattrs
);
1767 kill_litter_super(sb
);
1768 cgroup_free_root(root
);
1771 static struct file_system_type cgroup_fs_type
= {
1773 .mount
= cgroup_mount
,
1774 .kill_sb
= cgroup_kill_sb
,
1777 static struct kobject
*cgroup_kobj
;
1780 * cgroup_path - generate the path of a cgroup
1781 * @cgrp: the cgroup in question
1782 * @buf: the buffer to write the path into
1783 * @buflen: the length of the buffer
1785 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1787 * We can't generate cgroup path using dentry->d_name, as accessing
1788 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1789 * inode's i_mutex, while on the other hand cgroup_path() can be called
1790 * with some irq-safe spinlocks held.
1792 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1794 int ret
= -ENAMETOOLONG
;
1797 if (!cgrp
->parent
) {
1798 if (strlcpy(buf
, "/", buflen
) >= buflen
)
1799 return -ENAMETOOLONG
;
1803 start
= buf
+ buflen
- 1;
1808 const char *name
= cgroup_name(cgrp
);
1812 if ((start
-= len
) < buf
)
1814 memcpy(start
, name
, len
);
1820 cgrp
= cgrp
->parent
;
1821 } while (cgrp
->parent
);
1823 memmove(buf
, start
, buf
+ buflen
- start
);
1828 EXPORT_SYMBOL_GPL(cgroup_path
);
1831 * task_cgroup_path_from_hierarchy - cgroup path of a task on a hierarchy
1832 * @task: target task
1833 * @hierarchy_id: the hierarchy to look up @task's cgroup from
1834 * @buf: the buffer to write the path into
1835 * @buflen: the length of the buffer
1837 * Determine @task's cgroup on the hierarchy specified by @hierarchy_id and
1838 * copy its path into @buf. This function grabs cgroup_mutex and shouldn't
1839 * be used inside locks used by cgroup controller callbacks.
1841 int task_cgroup_path_from_hierarchy(struct task_struct
*task
, int hierarchy_id
,
1842 char *buf
, size_t buflen
)
1844 struct cgroupfs_root
*root
;
1845 struct cgroup
*cgrp
= NULL
;
1848 mutex_lock(&cgroup_mutex
);
1850 root
= idr_find(&cgroup_hierarchy_idr
, hierarchy_id
);
1852 cgrp
= task_cgroup_from_root(task
, root
);
1853 ret
= cgroup_path(cgrp
, buf
, buflen
);
1856 mutex_unlock(&cgroup_mutex
);
1860 EXPORT_SYMBOL_GPL(task_cgroup_path_from_hierarchy
);
1863 * Control Group taskset
1865 struct task_and_cgroup
{
1866 struct task_struct
*task
;
1867 struct cgroup
*cgrp
;
1871 struct cgroup_taskset
{
1872 struct task_and_cgroup single
;
1873 struct flex_array
*tc_array
;
1876 struct cgroup
*cur_cgrp
;
1880 * cgroup_taskset_first - reset taskset and return the first task
1881 * @tset: taskset of interest
1883 * @tset iteration is initialized and the first task is returned.
1885 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1887 if (tset
->tc_array
) {
1889 return cgroup_taskset_next(tset
);
1891 tset
->cur_cgrp
= tset
->single
.cgrp
;
1892 return tset
->single
.task
;
1895 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1898 * cgroup_taskset_next - iterate to the next task in taskset
1899 * @tset: taskset of interest
1901 * Return the next task in @tset. Iteration must have been initialized
1902 * with cgroup_taskset_first().
1904 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1906 struct task_and_cgroup
*tc
;
1908 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1911 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1912 tset
->cur_cgrp
= tc
->cgrp
;
1915 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1918 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1919 * @tset: taskset of interest
1921 * Return the cgroup for the current (last returned) task of @tset. This
1922 * function must be preceded by either cgroup_taskset_first() or
1923 * cgroup_taskset_next().
1925 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1927 return tset
->cur_cgrp
;
1929 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1932 * cgroup_taskset_size - return the number of tasks in taskset
1933 * @tset: taskset of interest
1935 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1937 return tset
->tc_array
? tset
->tc_array_len
: 1;
1939 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1943 * cgroup_task_migrate - move a task from one cgroup to another.
1945 * Must be called with cgroup_mutex and threadgroup locked.
1947 static void cgroup_task_migrate(struct cgroup
*oldcgrp
,
1948 struct task_struct
*tsk
, struct css_set
*newcg
)
1950 struct css_set
*oldcg
;
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 oldcg
= tsk
->cgroups
;
1961 rcu_assign_pointer(tsk
->cgroups
, newcg
);
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
, &newcg
->tasks
);
1968 write_unlock(&css_set_lock
);
1971 * We just gained a reference on oldcg by taking it from the task. As
1972 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1973 * it here; it will be freed under RCU.
1975 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
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_removed(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_removed(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
=
2473 kzalloc(sizeof(*state
), GFP_USER
);
2477 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2478 file
->f_op
= &cgroup_seqfile_operations
;
2479 err
= single_open(file
, cgroup_seqfile_show
, state
);
2482 } else if (cft
->open
)
2483 err
= cft
->open(inode
, file
);
2490 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2492 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2494 return cft
->release(inode
, file
);
2499 * cgroup_rename - Only allow simple rename of directories in place.
2501 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2502 struct inode
*new_dir
, struct dentry
*new_dentry
)
2505 struct cgroup_name
*name
, *old_name
;
2506 struct cgroup
*cgrp
;
2509 * It's convinient to use parent dir's i_mutex to protected
2512 lockdep_assert_held(&old_dir
->i_mutex
);
2514 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2516 if (new_dentry
->d_inode
)
2518 if (old_dir
!= new_dir
)
2521 cgrp
= __d_cgrp(old_dentry
);
2523 name
= cgroup_alloc_name(new_dentry
);
2527 ret
= simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2533 old_name
= cgrp
->name
;
2534 rcu_assign_pointer(cgrp
->name
, name
);
2536 kfree_rcu(old_name
, rcu_head
);
2540 static struct simple_xattrs
*__d_xattrs(struct dentry
*dentry
)
2542 if (S_ISDIR(dentry
->d_inode
->i_mode
))
2543 return &__d_cgrp(dentry
)->xattrs
;
2545 return &__d_cfe(dentry
)->xattrs
;
2548 static inline int xattr_enabled(struct dentry
*dentry
)
2550 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
2551 return root
->flags
& CGRP_ROOT_XATTR
;
2554 static bool is_valid_xattr(const char *name
)
2556 if (!strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) ||
2557 !strncmp(name
, XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
))
2562 static int cgroup_setxattr(struct dentry
*dentry
, const char *name
,
2563 const void *val
, size_t size
, int flags
)
2565 if (!xattr_enabled(dentry
))
2567 if (!is_valid_xattr(name
))
2569 return simple_xattr_set(__d_xattrs(dentry
), name
, val
, size
, flags
);
2572 static int cgroup_removexattr(struct dentry
*dentry
, const char *name
)
2574 if (!xattr_enabled(dentry
))
2576 if (!is_valid_xattr(name
))
2578 return simple_xattr_remove(__d_xattrs(dentry
), name
);
2581 static ssize_t
cgroup_getxattr(struct dentry
*dentry
, const char *name
,
2582 void *buf
, size_t size
)
2584 if (!xattr_enabled(dentry
))
2586 if (!is_valid_xattr(name
))
2588 return simple_xattr_get(__d_xattrs(dentry
), name
, buf
, size
);
2591 static ssize_t
cgroup_listxattr(struct dentry
*dentry
, char *buf
, size_t size
)
2593 if (!xattr_enabled(dentry
))
2595 return simple_xattr_list(__d_xattrs(dentry
), buf
, size
);
2598 static const struct file_operations cgroup_file_operations
= {
2599 .read
= cgroup_file_read
,
2600 .write
= cgroup_file_write
,
2601 .llseek
= generic_file_llseek
,
2602 .open
= cgroup_file_open
,
2603 .release
= cgroup_file_release
,
2606 static const struct inode_operations cgroup_file_inode_operations
= {
2607 .setxattr
= cgroup_setxattr
,
2608 .getxattr
= cgroup_getxattr
,
2609 .listxattr
= cgroup_listxattr
,
2610 .removexattr
= cgroup_removexattr
,
2613 static const struct inode_operations cgroup_dir_inode_operations
= {
2614 .lookup
= cgroup_lookup
,
2615 .mkdir
= cgroup_mkdir
,
2616 .rmdir
= cgroup_rmdir
,
2617 .rename
= cgroup_rename
,
2618 .setxattr
= cgroup_setxattr
,
2619 .getxattr
= cgroup_getxattr
,
2620 .listxattr
= cgroup_listxattr
,
2621 .removexattr
= cgroup_removexattr
,
2624 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
2626 if (dentry
->d_name
.len
> NAME_MAX
)
2627 return ERR_PTR(-ENAMETOOLONG
);
2628 d_add(dentry
, NULL
);
2633 * Check if a file is a control file
2635 static inline struct cftype
*__file_cft(struct file
*file
)
2637 if (file_inode(file
)->i_fop
!= &cgroup_file_operations
)
2638 return ERR_PTR(-EINVAL
);
2639 return __d_cft(file
->f_dentry
);
2642 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2643 struct super_block
*sb
)
2645 struct inode
*inode
;
2649 if (dentry
->d_inode
)
2652 inode
= cgroup_new_inode(mode
, sb
);
2656 if (S_ISDIR(mode
)) {
2657 inode
->i_op
= &cgroup_dir_inode_operations
;
2658 inode
->i_fop
= &simple_dir_operations
;
2660 /* start off with i_nlink == 2 (for "." entry) */
2662 inc_nlink(dentry
->d_parent
->d_inode
);
2665 * Control reaches here with cgroup_mutex held.
2666 * @inode->i_mutex should nest outside cgroup_mutex but we
2667 * want to populate it immediately without releasing
2668 * cgroup_mutex. As @inode isn't visible to anyone else
2669 * yet, trylock will always succeed without affecting
2672 WARN_ON_ONCE(!mutex_trylock(&inode
->i_mutex
));
2673 } else if (S_ISREG(mode
)) {
2675 inode
->i_fop
= &cgroup_file_operations
;
2676 inode
->i_op
= &cgroup_file_inode_operations
;
2678 d_instantiate(dentry
, inode
);
2679 dget(dentry
); /* Extra count - pin the dentry in core */
2684 * cgroup_file_mode - deduce file mode of a control file
2685 * @cft: the control file in question
2687 * returns cft->mode if ->mode is not 0
2688 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2689 * returns S_IRUGO if it has only a read handler
2690 * returns S_IWUSR if it has only a write hander
2692 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2699 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2700 cft
->read_map
|| cft
->read_seq_string
)
2703 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2704 cft
->write_string
|| cft
->trigger
)
2710 static int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2713 struct dentry
*dir
= cgrp
->dentry
;
2714 struct cgroup
*parent
= __d_cgrp(dir
);
2715 struct dentry
*dentry
;
2719 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2721 if (subsys
&& !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
2722 strcpy(name
, subsys
->name
);
2725 strcat(name
, cft
->name
);
2727 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2729 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2733 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2734 if (IS_ERR(dentry
)) {
2735 error
= PTR_ERR(dentry
);
2739 cfe
->type
= (void *)cft
;
2740 cfe
->dentry
= dentry
;
2741 dentry
->d_fsdata
= cfe
;
2742 simple_xattrs_init(&cfe
->xattrs
);
2744 mode
= cgroup_file_mode(cft
);
2745 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2747 list_add_tail(&cfe
->node
, &parent
->files
);
2756 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2757 struct cftype cfts
[], bool is_add
)
2762 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2763 /* does cft->flags tell us to skip this file on @cgrp? */
2764 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_sane_behavior(cgrp
))
2766 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2768 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2772 err
= cgroup_add_file(cgrp
, subsys
, cft
);
2774 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2778 cgroup_rm_file(cgrp
, cft
);
2784 static DEFINE_MUTEX(cgroup_cft_mutex
);
2786 static void cgroup_cfts_prepare(void)
2787 __acquires(&cgroup_cft_mutex
) __acquires(&cgroup_mutex
)
2790 * Thanks to the entanglement with vfs inode locking, we can't walk
2791 * the existing cgroups under cgroup_mutex and create files.
2792 * Instead, we increment reference on all cgroups and build list of
2793 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2794 * exclusive access to the field.
2796 mutex_lock(&cgroup_cft_mutex
);
2797 mutex_lock(&cgroup_mutex
);
2800 static void cgroup_cfts_commit(struct cgroup_subsys
*ss
,
2801 struct cftype
*cfts
, bool is_add
)
2802 __releases(&cgroup_mutex
) __releases(&cgroup_cft_mutex
)
2805 struct cgroup
*cgrp
, *n
;
2807 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2808 if (cfts
&& ss
->root
!= &rootnode
) {
2809 list_for_each_entry(cgrp
, &ss
->root
->allcg_list
, allcg_node
) {
2811 list_add_tail(&cgrp
->cft_q_node
, &pending
);
2815 mutex_unlock(&cgroup_mutex
);
2818 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2819 * files for all cgroups which were created before.
2821 list_for_each_entry_safe(cgrp
, n
, &pending
, cft_q_node
) {
2822 struct inode
*inode
= cgrp
->dentry
->d_inode
;
2824 mutex_lock(&inode
->i_mutex
);
2825 mutex_lock(&cgroup_mutex
);
2826 if (!cgroup_is_removed(cgrp
))
2827 cgroup_addrm_files(cgrp
, ss
, cfts
, is_add
);
2828 mutex_unlock(&cgroup_mutex
);
2829 mutex_unlock(&inode
->i_mutex
);
2831 list_del_init(&cgrp
->cft_q_node
);
2835 mutex_unlock(&cgroup_cft_mutex
);
2839 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2840 * @ss: target cgroup subsystem
2841 * @cfts: zero-length name terminated array of cftypes
2843 * Register @cfts to @ss. Files described by @cfts are created for all
2844 * existing cgroups to which @ss is attached and all future cgroups will
2845 * have them too. This function can be called anytime whether @ss is
2848 * Returns 0 on successful registration, -errno on failure. Note that this
2849 * function currently returns 0 as long as @cfts registration is successful
2850 * even if some file creation attempts on existing cgroups fail.
2852 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2854 struct cftype_set
*set
;
2856 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2860 cgroup_cfts_prepare();
2862 list_add_tail(&set
->node
, &ss
->cftsets
);
2863 cgroup_cfts_commit(ss
, cfts
, true);
2867 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2870 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2871 * @ss: target cgroup subsystem
2872 * @cfts: zero-length name terminated array of cftypes
2874 * Unregister @cfts from @ss. Files described by @cfts are removed from
2875 * all existing cgroups to which @ss is attached and all future cgroups
2876 * won't have them either. This function can be called anytime whether @ss
2877 * is attached or not.
2879 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2880 * registered with @ss.
2882 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2884 struct cftype_set
*set
;
2886 cgroup_cfts_prepare();
2888 list_for_each_entry(set
, &ss
->cftsets
, node
) {
2889 if (set
->cfts
== cfts
) {
2890 list_del_init(&set
->node
);
2891 cgroup_cfts_commit(ss
, cfts
, false);
2896 cgroup_cfts_commit(ss
, NULL
, false);
2901 * cgroup_task_count - count the number of tasks in a cgroup.
2902 * @cgrp: the cgroup in question
2904 * Return the number of tasks in the cgroup.
2906 int cgroup_task_count(const struct cgroup
*cgrp
)
2909 struct cg_cgroup_link
*link
;
2911 read_lock(&css_set_lock
);
2912 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2913 count
+= atomic_read(&link
->cg
->refcount
);
2915 read_unlock(&css_set_lock
);
2920 * Advance a list_head iterator. The iterator should be positioned at
2921 * the start of a css_set
2923 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2924 struct cgroup_iter
*it
)
2926 struct list_head
*l
= it
->cg_link
;
2927 struct cg_cgroup_link
*link
;
2930 /* Advance to the next non-empty css_set */
2933 if (l
== &cgrp
->css_sets
) {
2937 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2939 } while (list_empty(&cg
->tasks
));
2941 it
->task
= cg
->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_REMOVED 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_removed() is guaranteed to be visible as %true here.
3003 if (likely(!cgroup_is_removed(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
->cg_link
= &cgrp
->css_sets
;
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 cg_cgroup_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
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
3179 if (l
== &link
->cg
->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
= kmalloc(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
);
3519 l
->use_count
= 0; /* don't increment here */
3522 list_add(&l
->links
, &cgrp
->pidlists
);
3523 mutex_unlock(&cgrp
->pidlist_mutex
);
3528 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3530 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3531 struct cgroup_pidlist
**lp
)
3535 int pid
, n
= 0; /* used for populating the array */
3536 struct cgroup_iter it
;
3537 struct task_struct
*tsk
;
3538 struct cgroup_pidlist
*l
;
3541 * If cgroup gets more users after we read count, we won't have
3542 * enough space - tough. This race is indistinguishable to the
3543 * caller from the case that the additional cgroup users didn't
3544 * show up until sometime later on.
3546 length
= cgroup_task_count(cgrp
);
3547 array
= pidlist_allocate(length
);
3550 /* now, populate the array */
3551 cgroup_iter_start(cgrp
, &it
);
3552 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3553 if (unlikely(n
== length
))
3555 /* get tgid or pid for procs or tasks file respectively */
3556 if (type
== CGROUP_FILE_PROCS
)
3557 pid
= task_tgid_vnr(tsk
);
3559 pid
= task_pid_vnr(tsk
);
3560 if (pid
> 0) /* make sure to only use valid results */
3563 cgroup_iter_end(cgrp
, &it
);
3565 /* now sort & (if procs) strip out duplicates */
3566 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3567 if (type
== CGROUP_FILE_PROCS
)
3568 length
= pidlist_uniq(array
, length
);
3569 l
= cgroup_pidlist_find(cgrp
, type
);
3571 pidlist_free(array
);
3574 /* store array, freeing old if necessary - lock already held */
3575 pidlist_free(l
->list
);
3579 up_write(&l
->mutex
);
3585 * cgroupstats_build - build and fill cgroupstats
3586 * @stats: cgroupstats to fill information into
3587 * @dentry: A dentry entry belonging to the cgroup for which stats have
3590 * Build and fill cgroupstats so that taskstats can export it to user
3593 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3596 struct cgroup
*cgrp
;
3597 struct cgroup_iter it
;
3598 struct task_struct
*tsk
;
3601 * Validate dentry by checking the superblock operations,
3602 * and make sure it's a directory.
3604 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3605 !S_ISDIR(dentry
->d_inode
->i_mode
))
3609 cgrp
= dentry
->d_fsdata
;
3611 cgroup_iter_start(cgrp
, &it
);
3612 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3613 switch (tsk
->state
) {
3615 stats
->nr_running
++;
3617 case TASK_INTERRUPTIBLE
:
3618 stats
->nr_sleeping
++;
3620 case TASK_UNINTERRUPTIBLE
:
3621 stats
->nr_uninterruptible
++;
3624 stats
->nr_stopped
++;
3627 if (delayacct_is_task_waiting_on_io(tsk
))
3628 stats
->nr_io_wait
++;
3632 cgroup_iter_end(cgrp
, &it
);
3640 * seq_file methods for the tasks/procs files. The seq_file position is the
3641 * next pid to display; the seq_file iterator is a pointer to the pid
3642 * in the cgroup->l->list array.
3645 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3648 * Initially we receive a position value that corresponds to
3649 * one more than the last pid shown (or 0 on the first call or
3650 * after a seek to the start). Use a binary-search to find the
3651 * next pid to display, if any
3653 struct cgroup_pidlist
*l
= s
->private;
3654 int index
= 0, pid
= *pos
;
3657 down_read(&l
->mutex
);
3659 int end
= l
->length
;
3661 while (index
< end
) {
3662 int mid
= (index
+ end
) / 2;
3663 if (l
->list
[mid
] == pid
) {
3666 } else if (l
->list
[mid
] <= pid
)
3672 /* If we're off the end of the array, we're done */
3673 if (index
>= l
->length
)
3675 /* Update the abstract position to be the actual pid that we found */
3676 iter
= l
->list
+ index
;
3681 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3683 struct cgroup_pidlist
*l
= s
->private;
3687 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3689 struct cgroup_pidlist
*l
= s
->private;
3691 pid_t
*end
= l
->list
+ l
->length
;
3693 * Advance to the next pid in the array. If this goes off the
3705 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3707 return seq_printf(s
, "%d\n", *(int *)v
);
3711 * seq_operations functions for iterating on pidlists through seq_file -
3712 * independent of whether it's tasks or procs
3714 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3715 .start
= cgroup_pidlist_start
,
3716 .stop
= cgroup_pidlist_stop
,
3717 .next
= cgroup_pidlist_next
,
3718 .show
= cgroup_pidlist_show
,
3721 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3724 * the case where we're the last user of this particular pidlist will
3725 * have us remove it from the cgroup's list, which entails taking the
3726 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3727 * pidlist_mutex, we have to take pidlist_mutex first.
3729 mutex_lock(&l
->owner
->pidlist_mutex
);
3730 down_write(&l
->mutex
);
3731 BUG_ON(!l
->use_count
);
3732 if (!--l
->use_count
) {
3733 /* we're the last user if refcount is 0; remove and free */
3734 list_del(&l
->links
);
3735 mutex_unlock(&l
->owner
->pidlist_mutex
);
3736 pidlist_free(l
->list
);
3737 put_pid_ns(l
->key
.ns
);
3738 up_write(&l
->mutex
);
3742 mutex_unlock(&l
->owner
->pidlist_mutex
);
3743 up_write(&l
->mutex
);
3746 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3748 struct cgroup_pidlist
*l
;
3749 if (!(file
->f_mode
& FMODE_READ
))
3752 * the seq_file will only be initialized if the file was opened for
3753 * reading; hence we check if it's not null only in that case.
3755 l
= ((struct seq_file
*)file
->private_data
)->private;
3756 cgroup_release_pid_array(l
);
3757 return seq_release(inode
, file
);
3760 static const struct file_operations cgroup_pidlist_operations
= {
3762 .llseek
= seq_lseek
,
3763 .write
= cgroup_file_write
,
3764 .release
= cgroup_pidlist_release
,
3768 * The following functions handle opens on a file that displays a pidlist
3769 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3772 /* helper function for the two below it */
3773 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3775 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3776 struct cgroup_pidlist
*l
;
3779 /* Nothing to do for write-only files */
3780 if (!(file
->f_mode
& FMODE_READ
))
3783 /* have the array populated */
3784 retval
= pidlist_array_load(cgrp
, type
, &l
);
3787 /* configure file information */
3788 file
->f_op
= &cgroup_pidlist_operations
;
3790 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3792 cgroup_release_pid_array(l
);
3795 ((struct seq_file
*)file
->private_data
)->private = l
;
3798 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3800 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3802 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3804 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3807 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3810 return notify_on_release(cgrp
);
3813 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3817 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3819 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3821 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3826 * Unregister event and free resources.
3828 * Gets called from workqueue.
3830 static void cgroup_event_remove(struct work_struct
*work
)
3832 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3834 struct cgroup
*cgrp
= event
->cgrp
;
3836 remove_wait_queue(event
->wqh
, &event
->wait
);
3838 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3840 /* Notify userspace the event is going away. */
3841 eventfd_signal(event
->eventfd
, 1);
3843 eventfd_ctx_put(event
->eventfd
);
3849 * Gets called on POLLHUP on eventfd when user closes it.
3851 * Called with wqh->lock held and interrupts disabled.
3853 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3854 int sync
, void *key
)
3856 struct cgroup_event
*event
= container_of(wait
,
3857 struct cgroup_event
, wait
);
3858 struct cgroup
*cgrp
= event
->cgrp
;
3859 unsigned long flags
= (unsigned long)key
;
3861 if (flags
& POLLHUP
) {
3863 * If the event has been detached at cgroup removal, we
3864 * can simply return knowing the other side will cleanup
3867 * We can't race against event freeing since the other
3868 * side will require wqh->lock via remove_wait_queue(),
3871 spin_lock(&cgrp
->event_list_lock
);
3872 if (!list_empty(&event
->list
)) {
3873 list_del_init(&event
->list
);
3875 * We are in atomic context, but cgroup_event_remove()
3876 * may sleep, so we have to call it in workqueue.
3878 schedule_work(&event
->remove
);
3880 spin_unlock(&cgrp
->event_list_lock
);
3886 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3887 wait_queue_head_t
*wqh
, poll_table
*pt
)
3889 struct cgroup_event
*event
= container_of(pt
,
3890 struct cgroup_event
, pt
);
3893 add_wait_queue(wqh
, &event
->wait
);
3897 * Parse input and register new cgroup event handler.
3899 * Input must be in format '<event_fd> <control_fd> <args>'.
3900 * Interpretation of args is defined by control file implementation.
3902 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3905 struct cgroup_event
*event
= NULL
;
3906 struct cgroup
*cgrp_cfile
;
3907 unsigned int efd
, cfd
;
3908 struct file
*efile
= NULL
;
3909 struct file
*cfile
= NULL
;
3913 efd
= simple_strtoul(buffer
, &endp
, 10);
3918 cfd
= simple_strtoul(buffer
, &endp
, 10);
3919 if ((*endp
!= ' ') && (*endp
!= '\0'))
3923 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3927 INIT_LIST_HEAD(&event
->list
);
3928 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3929 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3930 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3932 efile
= eventfd_fget(efd
);
3933 if (IS_ERR(efile
)) {
3934 ret
= PTR_ERR(efile
);
3938 event
->eventfd
= eventfd_ctx_fileget(efile
);
3939 if (IS_ERR(event
->eventfd
)) {
3940 ret
= PTR_ERR(event
->eventfd
);
3950 /* the process need read permission on control file */
3951 /* AV: shouldn't we check that it's been opened for read instead? */
3952 ret
= inode_permission(file_inode(cfile
), MAY_READ
);
3956 event
->cft
= __file_cft(cfile
);
3957 if (IS_ERR(event
->cft
)) {
3958 ret
= PTR_ERR(event
->cft
);
3963 * The file to be monitored must be in the same cgroup as
3964 * cgroup.event_control is.
3966 cgrp_cfile
= __d_cgrp(cfile
->f_dentry
->d_parent
);
3967 if (cgrp_cfile
!= cgrp
) {
3972 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3977 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3978 event
->eventfd
, buffer
);
3982 efile
->f_op
->poll(efile
, &event
->pt
);
3985 * Events should be removed after rmdir of cgroup directory, but before
3986 * destroying subsystem state objects. Let's take reference to cgroup
3987 * directory dentry to do that.
3991 spin_lock(&cgrp
->event_list_lock
);
3992 list_add(&event
->list
, &cgrp
->event_list
);
3993 spin_unlock(&cgrp
->event_list_lock
);
4004 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
4005 eventfd_ctx_put(event
->eventfd
);
4007 if (!IS_ERR_OR_NULL(efile
))
4015 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
4018 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4021 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
4026 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4028 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4032 static struct cftype cgroup_base_files
[] = {
4034 .name
= "cgroup.procs",
4035 .open
= cgroup_procs_open
,
4036 .write_u64
= cgroup_procs_write
,
4037 .release
= cgroup_pidlist_release
,
4038 .mode
= S_IRUGO
| S_IWUSR
,
4041 .name
= "cgroup.event_control",
4042 .write_string
= cgroup_write_event_control
,
4046 .name
= "cgroup.clone_children",
4047 .flags
= CFTYPE_INSANE
,
4048 .read_u64
= cgroup_clone_children_read
,
4049 .write_u64
= cgroup_clone_children_write
,
4052 .name
= "cgroup.sane_behavior",
4053 .flags
= CFTYPE_ONLY_ON_ROOT
,
4054 .read_seq_string
= cgroup_sane_behavior_show
,
4058 * Historical crazy stuff. These don't have "cgroup." prefix and
4059 * don't exist if sane_behavior. If you're depending on these, be
4060 * prepared to be burned.
4064 .flags
= CFTYPE_INSANE
, /* use "procs" instead */
4065 .open
= cgroup_tasks_open
,
4066 .write_u64
= cgroup_tasks_write
,
4067 .release
= cgroup_pidlist_release
,
4068 .mode
= S_IRUGO
| S_IWUSR
,
4071 .name
= "notify_on_release",
4072 .flags
= CFTYPE_INSANE
,
4073 .read_u64
= cgroup_read_notify_on_release
,
4074 .write_u64
= cgroup_write_notify_on_release
,
4077 .name
= "release_agent",
4078 .flags
= CFTYPE_INSANE
| CFTYPE_ONLY_ON_ROOT
,
4079 .read_seq_string
= cgroup_release_agent_show
,
4080 .write_string
= cgroup_release_agent_write
,
4081 .max_write_len
= PATH_MAX
,
4087 * cgroup_populate_dir - selectively creation of files in a directory
4088 * @cgrp: target cgroup
4089 * @base_files: true if the base files should be added
4090 * @subsys_mask: mask of the subsystem ids whose files should be added
4092 static int cgroup_populate_dir(struct cgroup
*cgrp
, bool base_files
,
4093 unsigned long subsys_mask
)
4096 struct cgroup_subsys
*ss
;
4099 err
= cgroup_addrm_files(cgrp
, NULL
, cgroup_base_files
, true);
4104 /* process cftsets of each subsystem */
4105 for_each_subsys(cgrp
->root
, ss
) {
4106 struct cftype_set
*set
;
4107 if (!test_bit(ss
->subsys_id
, &subsys_mask
))
4110 list_for_each_entry(set
, &ss
->cftsets
, node
)
4111 cgroup_addrm_files(cgrp
, ss
, set
->cfts
, true);
4114 /* This cgroup is ready now */
4115 for_each_subsys(cgrp
->root
, ss
) {
4116 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4118 * Update id->css pointer and make this css visible from
4119 * CSS ID functions. This pointer will be dereferened
4120 * from RCU-read-side without locks.
4123 rcu_assign_pointer(css
->id
->css
, css
);
4129 static void css_dput_fn(struct work_struct
*work
)
4131 struct cgroup_subsys_state
*css
=
4132 container_of(work
, struct cgroup_subsys_state
, dput_work
);
4133 struct dentry
*dentry
= css
->cgroup
->dentry
;
4134 struct super_block
*sb
= dentry
->d_sb
;
4136 atomic_inc(&sb
->s_active
);
4138 deactivate_super(sb
);
4141 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
4142 struct cgroup_subsys
*ss
,
4143 struct cgroup
*cgrp
)
4146 atomic_set(&css
->refcnt
, 1);
4149 if (cgrp
== dummytop
)
4150 css
->flags
|= CSS_ROOT
;
4151 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
4152 cgrp
->subsys
[ss
->subsys_id
] = css
;
4155 * css holds an extra ref to @cgrp->dentry which is put on the last
4156 * css_put(). dput() requires process context, which css_put() may
4157 * be called without. @css->dput_work will be used to invoke
4158 * dput() asynchronously from css_put().
4160 INIT_WORK(&css
->dput_work
, css_dput_fn
);
4163 /* invoke ->post_create() on a new CSS and mark it online if successful */
4164 static int online_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4168 lockdep_assert_held(&cgroup_mutex
);
4171 ret
= ss
->css_online(cgrp
);
4173 cgrp
->subsys
[ss
->subsys_id
]->flags
|= CSS_ONLINE
;
4177 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4178 static void offline_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4179 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4181 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4183 lockdep_assert_held(&cgroup_mutex
);
4185 if (!(css
->flags
& CSS_ONLINE
))
4188 if (ss
->css_offline
)
4189 ss
->css_offline(cgrp
);
4191 cgrp
->subsys
[ss
->subsys_id
]->flags
&= ~CSS_ONLINE
;
4195 * cgroup_create - create a cgroup
4196 * @parent: cgroup that will be parent of the new cgroup
4197 * @dentry: dentry of the new cgroup
4198 * @mode: mode to set on new inode
4200 * Must be called with the mutex on the parent inode held
4202 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
4205 static atomic64_t serial_nr_cursor
= ATOMIC64_INIT(0);
4206 struct cgroup
*cgrp
;
4207 struct cgroup_name
*name
;
4208 struct cgroupfs_root
*root
= parent
->root
;
4210 struct cgroup_subsys
*ss
;
4211 struct super_block
*sb
= root
->sb
;
4213 /* allocate the cgroup and its ID, 0 is reserved for the root */
4214 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4218 name
= cgroup_alloc_name(dentry
);
4221 rcu_assign_pointer(cgrp
->name
, name
);
4223 cgrp
->id
= ida_simple_get(&root
->cgroup_ida
, 1, 0, GFP_KERNEL
);
4228 * Only live parents can have children. Note that the liveliness
4229 * check isn't strictly necessary because cgroup_mkdir() and
4230 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4231 * anyway so that locking is contained inside cgroup proper and we
4232 * don't get nasty surprises if we ever grow another caller.
4234 if (!cgroup_lock_live_group(parent
)) {
4239 /* Grab a reference on the superblock so the hierarchy doesn't
4240 * get deleted on unmount if there are child cgroups. This
4241 * can be done outside cgroup_mutex, since the sb can't
4242 * disappear while someone has an open control file on the
4244 atomic_inc(&sb
->s_active
);
4246 init_cgroup_housekeeping(cgrp
);
4248 dentry
->d_fsdata
= cgrp
;
4249 cgrp
->dentry
= dentry
;
4251 cgrp
->parent
= parent
;
4252 cgrp
->root
= parent
->root
;
4254 if (notify_on_release(parent
))
4255 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4257 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4258 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4260 for_each_subsys(root
, ss
) {
4261 struct cgroup_subsys_state
*css
;
4263 css
= ss
->css_alloc(cgrp
);
4268 init_cgroup_css(css
, ss
, cgrp
);
4270 err
= alloc_css_id(ss
, parent
, cgrp
);
4277 * Create directory. cgroup_create_file() returns with the new
4278 * directory locked on success so that it can be populated without
4279 * dropping cgroup_mutex.
4281 err
= cgroup_create_file(dentry
, S_IFDIR
| mode
, sb
);
4284 lockdep_assert_held(&dentry
->d_inode
->i_mutex
);
4287 * Assign a monotonically increasing serial number. With the list
4288 * appending below, it guarantees that sibling cgroups are always
4289 * sorted in the ascending serial number order on the parent's
4292 cgrp
->serial_nr
= atomic64_inc_return(&serial_nr_cursor
);
4294 /* allocation complete, commit to creation */
4295 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
4296 list_add_tail_rcu(&cgrp
->sibling
, &cgrp
->parent
->children
);
4297 root
->number_of_cgroups
++;
4299 /* each css holds a ref to the cgroup's dentry */
4300 for_each_subsys(root
, ss
)
4303 /* hold a ref to the parent's dentry */
4304 dget(parent
->dentry
);
4306 /* creation succeeded, notify subsystems */
4307 for_each_subsys(root
, ss
) {
4308 err
= online_css(ss
, cgrp
);
4312 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4314 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",
4315 current
->comm
, current
->pid
, ss
->name
);
4316 if (!strcmp(ss
->name
, "memory"))
4317 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4318 ss
->warned_broken_hierarchy
= true;
4322 err
= cgroup_populate_dir(cgrp
, true, root
->subsys_mask
);
4326 mutex_unlock(&cgroup_mutex
);
4327 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4332 for_each_subsys(root
, ss
) {
4333 if (cgrp
->subsys
[ss
->subsys_id
])
4336 mutex_unlock(&cgroup_mutex
);
4337 /* Release the reference count that we took on the superblock */
4338 deactivate_super(sb
);
4340 ida_simple_remove(&root
->cgroup_ida
, cgrp
->id
);
4342 kfree(rcu_dereference_raw(cgrp
->name
));
4348 cgroup_destroy_locked(cgrp
);
4349 mutex_unlock(&cgroup_mutex
);
4350 mutex_unlock(&dentry
->d_inode
->i_mutex
);
4354 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4356 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4358 /* the vfs holds inode->i_mutex already */
4359 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4362 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4363 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4365 struct dentry
*d
= cgrp
->dentry
;
4366 struct cgroup
*parent
= cgrp
->parent
;
4367 struct cgroup_event
*event
, *tmp
;
4368 struct cgroup_subsys
*ss
;
4370 lockdep_assert_held(&d
->d_inode
->i_mutex
);
4371 lockdep_assert_held(&cgroup_mutex
);
4373 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
))
4377 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4378 * removed. This makes future css_tryget() and child creation
4379 * attempts fail thus maintaining the removal conditions verified
4382 * Note that CGRP_REMVOED clearing is depended upon by
4383 * cgroup_next_sibling() to resume iteration after dropping RCU
4384 * read lock. See cgroup_next_sibling() for details.
4386 for_each_subsys(cgrp
->root
, ss
) {
4387 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4389 WARN_ON(atomic_read(&css
->refcnt
) < 0);
4390 atomic_add(CSS_DEACT_BIAS
, &css
->refcnt
);
4392 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
4394 /* tell subsystems to initate destruction */
4395 for_each_subsys(cgrp
->root
, ss
)
4396 offline_css(ss
, cgrp
);
4399 * Put all the base refs. Each css holds an extra reference to the
4400 * cgroup's dentry and cgroup removal proceeds regardless of css
4401 * refs. On the last put of each css, whenever that may be, the
4402 * extra dentry ref is put so that dentry destruction happens only
4403 * after all css's are released.
4405 for_each_subsys(cgrp
->root
, ss
)
4406 css_put(cgrp
->subsys
[ss
->subsys_id
]);
4408 raw_spin_lock(&release_list_lock
);
4409 if (!list_empty(&cgrp
->release_list
))
4410 list_del_init(&cgrp
->release_list
);
4411 raw_spin_unlock(&release_list_lock
);
4413 /* delete this cgroup from parent->children */
4414 list_del_rcu(&cgrp
->sibling
);
4415 list_del_init(&cgrp
->allcg_node
);
4418 cgroup_d_remove_dir(d
);
4421 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4422 check_for_release(parent
);
4425 * Unregister events and notify userspace.
4426 * Notify userspace about cgroup removing only after rmdir of cgroup
4427 * directory to avoid race between userspace and kernelspace.
4429 spin_lock(&cgrp
->event_list_lock
);
4430 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4431 list_del_init(&event
->list
);
4432 schedule_work(&event
->remove
);
4434 spin_unlock(&cgrp
->event_list_lock
);
4439 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4443 mutex_lock(&cgroup_mutex
);
4444 ret
= cgroup_destroy_locked(dentry
->d_fsdata
);
4445 mutex_unlock(&cgroup_mutex
);
4450 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4452 INIT_LIST_HEAD(&ss
->cftsets
);
4455 * base_cftset is embedded in subsys itself, no need to worry about
4458 if (ss
->base_cftypes
) {
4459 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4460 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4464 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4466 struct cgroup_subsys_state
*css
;
4468 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4470 mutex_lock(&cgroup_mutex
);
4472 /* init base cftset */
4473 cgroup_init_cftsets(ss
);
4475 /* Create the top cgroup state for this subsystem */
4476 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4477 ss
->root
= &rootnode
;
4478 css
= ss
->css_alloc(dummytop
);
4479 /* We don't handle early failures gracefully */
4480 BUG_ON(IS_ERR(css
));
4481 init_cgroup_css(css
, ss
, dummytop
);
4483 /* Update the init_css_set to contain a subsys
4484 * pointer to this state - since the subsystem is
4485 * newly registered, all tasks and hence the
4486 * init_css_set is in the subsystem's top cgroup. */
4487 init_css_set
.subsys
[ss
->subsys_id
] = css
;
4489 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4491 /* At system boot, before all subsystems have been
4492 * registered, no tasks have been forked, so we don't
4493 * need to invoke fork callbacks here. */
4494 BUG_ON(!list_empty(&init_task
.tasks
));
4496 BUG_ON(online_css(ss
, dummytop
));
4498 mutex_unlock(&cgroup_mutex
);
4500 /* this function shouldn't be used with modular subsystems, since they
4501 * need to register a subsys_id, among other things */
4506 * cgroup_load_subsys: load and register a modular subsystem at runtime
4507 * @ss: the subsystem to load
4509 * This function should be called in a modular subsystem's initcall. If the
4510 * subsystem is built as a module, it will be assigned a new subsys_id and set
4511 * up for use. If the subsystem is built-in anyway, work is delegated to the
4512 * simpler cgroup_init_subsys.
4514 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4516 struct cgroup_subsys_state
*css
;
4518 struct hlist_node
*tmp
;
4522 /* check name and function validity */
4523 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4524 ss
->css_alloc
== NULL
|| ss
->css_free
== NULL
)
4528 * we don't support callbacks in modular subsystems. this check is
4529 * before the ss->module check for consistency; a subsystem that could
4530 * be a module should still have no callbacks even if the user isn't
4531 * compiling it as one.
4533 if (ss
->fork
|| ss
->exit
)
4537 * an optionally modular subsystem is built-in: we want to do nothing,
4538 * since cgroup_init_subsys will have already taken care of it.
4540 if (ss
->module
== NULL
) {
4541 /* a sanity check */
4542 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4546 /* init base cftset */
4547 cgroup_init_cftsets(ss
);
4549 mutex_lock(&cgroup_mutex
);
4550 subsys
[ss
->subsys_id
] = ss
;
4553 * no ss->css_alloc seems to need anything important in the ss
4554 * struct, so this can happen first (i.e. before the rootnode
4557 css
= ss
->css_alloc(dummytop
);
4559 /* failure case - need to deassign the subsys[] slot. */
4560 subsys
[ss
->subsys_id
] = NULL
;
4561 mutex_unlock(&cgroup_mutex
);
4562 return PTR_ERR(css
);
4565 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4566 ss
->root
= &rootnode
;
4568 /* our new subsystem will be attached to the dummy hierarchy. */
4569 init_cgroup_css(css
, ss
, dummytop
);
4570 /* init_idr must be after init_cgroup_css because it sets css->id. */
4572 ret
= cgroup_init_idr(ss
, css
);
4578 * Now we need to entangle the css into the existing css_sets. unlike
4579 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4580 * will need a new pointer to it; done by iterating the css_set_table.
4581 * furthermore, modifying the existing css_sets will corrupt the hash
4582 * table state, so each changed css_set will need its hash recomputed.
4583 * this is all done under the css_set_lock.
4585 write_lock(&css_set_lock
);
4586 hash_for_each_safe(css_set_table
, i
, tmp
, cg
, hlist
) {
4587 /* skip entries that we already rehashed */
4588 if (cg
->subsys
[ss
->subsys_id
])
4590 /* remove existing entry */
4591 hash_del(&cg
->hlist
);
4593 cg
->subsys
[ss
->subsys_id
] = css
;
4594 /* recompute hash and restore entry */
4595 key
= css_set_hash(cg
->subsys
);
4596 hash_add(css_set_table
, &cg
->hlist
, key
);
4598 write_unlock(&css_set_lock
);
4600 ret
= online_css(ss
, dummytop
);
4605 mutex_unlock(&cgroup_mutex
);
4609 mutex_unlock(&cgroup_mutex
);
4610 /* @ss can't be mounted here as try_module_get() would fail */
4611 cgroup_unload_subsys(ss
);
4614 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4617 * cgroup_unload_subsys: unload a modular subsystem
4618 * @ss: the subsystem to unload
4620 * This function should be called in a modular subsystem's exitcall. When this
4621 * function is invoked, the refcount on the subsystem's module will be 0, so
4622 * the subsystem will not be attached to any hierarchy.
4624 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4626 struct cg_cgroup_link
*link
;
4628 BUG_ON(ss
->module
== NULL
);
4631 * we shouldn't be called if the subsystem is in use, and the use of
4632 * try_module_get in parse_cgroupfs_options should ensure that it
4633 * doesn't start being used while we're killing it off.
4635 BUG_ON(ss
->root
!= &rootnode
);
4637 mutex_lock(&cgroup_mutex
);
4639 offline_css(ss
, dummytop
);
4642 idr_destroy(&ss
->idr
);
4644 /* deassign the subsys_id */
4645 subsys
[ss
->subsys_id
] = NULL
;
4647 /* remove subsystem from rootnode's list of subsystems */
4648 list_del_init(&ss
->sibling
);
4651 * disentangle the css from all css_sets attached to the dummytop. as
4652 * in loading, we need to pay our respects to the hashtable gods.
4654 write_lock(&css_set_lock
);
4655 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4656 struct css_set
*cg
= link
->cg
;
4659 hash_del(&cg
->hlist
);
4660 cg
->subsys
[ss
->subsys_id
] = NULL
;
4661 key
= css_set_hash(cg
->subsys
);
4662 hash_add(css_set_table
, &cg
->hlist
, key
);
4664 write_unlock(&css_set_lock
);
4667 * remove subsystem's css from the dummytop and free it - need to
4668 * free before marking as null because ss->css_free needs the
4669 * cgrp->subsys pointer to find their state. note that this also
4670 * takes care of freeing the css_id.
4672 ss
->css_free(dummytop
);
4673 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4675 mutex_unlock(&cgroup_mutex
);
4677 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4680 * cgroup_init_early - cgroup initialization at system boot
4682 * Initialize cgroups at system boot, and initialize any
4683 * subsystems that request early init.
4685 int __init
cgroup_init_early(void)
4688 atomic_set(&init_css_set
.refcount
, 1);
4689 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4690 INIT_LIST_HEAD(&init_css_set
.tasks
);
4691 INIT_HLIST_NODE(&init_css_set
.hlist
);
4693 init_cgroup_root(&rootnode
);
4695 init_task
.cgroups
= &init_css_set
;
4697 init_css_set_link
.cg
= &init_css_set
;
4698 init_css_set_link
.cgrp
= dummytop
;
4699 list_add(&init_css_set_link
.cgrp_link_list
,
4700 &rootnode
.top_cgroup
.css_sets
);
4701 list_add(&init_css_set_link
.cg_link_list
,
4702 &init_css_set
.cg_links
);
4704 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4705 struct cgroup_subsys
*ss
= subsys
[i
];
4707 /* at bootup time, we don't worry about modular subsystems */
4708 if (!ss
|| ss
->module
)
4712 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4713 BUG_ON(!ss
->css_alloc
);
4714 BUG_ON(!ss
->css_free
);
4715 if (ss
->subsys_id
!= i
) {
4716 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4717 ss
->name
, ss
->subsys_id
);
4722 cgroup_init_subsys(ss
);
4728 * cgroup_init - cgroup initialization
4730 * Register cgroup filesystem and /proc file, and initialize
4731 * any subsystems that didn't request early init.
4733 int __init
cgroup_init(void)
4739 err
= bdi_init(&cgroup_backing_dev_info
);
4743 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4744 struct cgroup_subsys
*ss
= subsys
[i
];
4746 /* at bootup time, we don't worry about modular subsystems */
4747 if (!ss
|| ss
->module
)
4749 if (!ss
->early_init
)
4750 cgroup_init_subsys(ss
);
4752 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4755 /* Add init_css_set to the hash table */
4756 key
= css_set_hash(init_css_set
.subsys
);
4757 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4759 /* allocate id for the dummy hierarchy */
4760 mutex_lock(&cgroup_mutex
);
4761 mutex_lock(&cgroup_root_mutex
);
4763 BUG_ON(cgroup_init_root_id(&rootnode
));
4765 mutex_unlock(&cgroup_root_mutex
);
4766 mutex_unlock(&cgroup_mutex
);
4768 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4774 err
= register_filesystem(&cgroup_fs_type
);
4776 kobject_put(cgroup_kobj
);
4780 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4784 bdi_destroy(&cgroup_backing_dev_info
);
4790 * proc_cgroup_show()
4791 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4792 * - Used for /proc/<pid>/cgroup.
4793 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4794 * doesn't really matter if tsk->cgroup changes after we read it,
4795 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4796 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4797 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4798 * cgroup to top_cgroup.
4801 /* TODO: Use a proper seq_file iterator */
4802 int proc_cgroup_show(struct seq_file
*m
, void *v
)
4805 struct task_struct
*tsk
;
4808 struct cgroupfs_root
*root
;
4811 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4817 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4823 mutex_lock(&cgroup_mutex
);
4825 for_each_active_root(root
) {
4826 struct cgroup_subsys
*ss
;
4827 struct cgroup
*cgrp
;
4830 seq_printf(m
, "%d:", root
->hierarchy_id
);
4831 for_each_subsys(root
, ss
)
4832 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4833 if (strlen(root
->name
))
4834 seq_printf(m
, "%sname=%s", count
? "," : "",
4837 cgrp
= task_cgroup_from_root(tsk
, root
);
4838 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4846 mutex_unlock(&cgroup_mutex
);
4847 put_task_struct(tsk
);
4854 /* Display information about each subsystem and each hierarchy */
4855 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4859 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4861 * ideally we don't want subsystems moving around while we do this.
4862 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4863 * subsys/hierarchy state.
4865 mutex_lock(&cgroup_mutex
);
4866 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4867 struct cgroup_subsys
*ss
= subsys
[i
];
4870 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4871 ss
->name
, ss
->root
->hierarchy_id
,
4872 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4874 mutex_unlock(&cgroup_mutex
);
4878 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4880 return single_open(file
, proc_cgroupstats_show
, NULL
);
4883 static const struct file_operations proc_cgroupstats_operations
= {
4884 .open
= cgroupstats_open
,
4886 .llseek
= seq_lseek
,
4887 .release
= single_release
,
4891 * cgroup_fork - attach newly forked task to its parents cgroup.
4892 * @child: pointer to task_struct of forking parent process.
4894 * Description: A task inherits its parent's cgroup at fork().
4896 * A pointer to the shared css_set was automatically copied in
4897 * fork.c by dup_task_struct(). However, we ignore that copy, since
4898 * it was not made under the protection of RCU or cgroup_mutex, so
4899 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4900 * have already changed current->cgroups, allowing the previously
4901 * referenced cgroup group to be removed and freed.
4903 * At the point that cgroup_fork() is called, 'current' is the parent
4904 * task, and the passed argument 'child' points to the child task.
4906 void cgroup_fork(struct task_struct
*child
)
4909 child
->cgroups
= current
->cgroups
;
4910 get_css_set(child
->cgroups
);
4911 task_unlock(current
);
4912 INIT_LIST_HEAD(&child
->cg_list
);
4916 * cgroup_post_fork - called on a new task after adding it to the task list
4917 * @child: the task in question
4919 * Adds the task to the list running through its css_set if necessary and
4920 * call the subsystem fork() callbacks. Has to be after the task is
4921 * visible on the task list in case we race with the first call to
4922 * cgroup_iter_start() - to guarantee that the new task ends up on its
4925 void cgroup_post_fork(struct task_struct
*child
)
4930 * use_task_css_set_links is set to 1 before we walk the tasklist
4931 * under the tasklist_lock and we read it here after we added the child
4932 * to the tasklist under the tasklist_lock as well. If the child wasn't
4933 * yet in the tasklist when we walked through it from
4934 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4935 * should be visible now due to the paired locking and barriers implied
4936 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4937 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4940 if (use_task_css_set_links
) {
4941 write_lock(&css_set_lock
);
4943 if (list_empty(&child
->cg_list
))
4944 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4946 write_unlock(&css_set_lock
);
4950 * Call ss->fork(). This must happen after @child is linked on
4951 * css_set; otherwise, @child might change state between ->fork()
4952 * and addition to css_set.
4954 if (need_forkexit_callback
) {
4956 * fork/exit callbacks are supported only for builtin
4957 * subsystems, and the builtin section of the subsys
4958 * array is immutable, so we don't need to lock the
4959 * subsys array here. On the other hand, modular section
4960 * of the array can be freed at module unload, so we
4963 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4964 struct cgroup_subsys
*ss
= subsys
[i
];
4973 * cgroup_exit - detach cgroup from exiting task
4974 * @tsk: pointer to task_struct of exiting process
4975 * @run_callback: run exit callbacks?
4977 * Description: Detach cgroup from @tsk and release it.
4979 * Note that cgroups marked notify_on_release force every task in
4980 * them to take the global cgroup_mutex mutex when exiting.
4981 * This could impact scaling on very large systems. Be reluctant to
4982 * use notify_on_release cgroups where very high task exit scaling
4983 * is required on large systems.
4985 * the_top_cgroup_hack:
4987 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4989 * We call cgroup_exit() while the task is still competent to
4990 * handle notify_on_release(), then leave the task attached to the
4991 * root cgroup in each hierarchy for the remainder of its exit.
4993 * To do this properly, we would increment the reference count on
4994 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4995 * code we would add a second cgroup function call, to drop that
4996 * reference. This would just create an unnecessary hot spot on
4997 * the top_cgroup reference count, to no avail.
4999 * Normally, holding a reference to a cgroup without bumping its
5000 * count is unsafe. The cgroup could go away, or someone could
5001 * attach us to a different cgroup, decrementing the count on
5002 * the first cgroup that we never incremented. But in this case,
5003 * top_cgroup isn't going away, and either task has PF_EXITING set,
5004 * which wards off any cgroup_attach_task() attempts, or task is a failed
5005 * fork, never visible to cgroup_attach_task.
5007 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
5013 * Unlink from the css_set task list if necessary.
5014 * Optimistically check cg_list before taking
5017 if (!list_empty(&tsk
->cg_list
)) {
5018 write_lock(&css_set_lock
);
5019 if (!list_empty(&tsk
->cg_list
))
5020 list_del_init(&tsk
->cg_list
);
5021 write_unlock(&css_set_lock
);
5024 /* Reassign the task to the init_css_set. */
5027 tsk
->cgroups
= &init_css_set
;
5029 if (run_callbacks
&& need_forkexit_callback
) {
5031 * fork/exit callbacks are supported only for builtin
5032 * subsystems, see cgroup_post_fork() for details.
5034 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
5035 struct cgroup_subsys
*ss
= subsys
[i
];
5038 struct cgroup
*old_cgrp
=
5039 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
5040 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
5041 ss
->exit(cgrp
, old_cgrp
, tsk
);
5047 put_css_set_taskexit(cg
);
5050 static void check_for_release(struct cgroup
*cgrp
)
5052 /* All of these checks rely on RCU to keep the cgroup
5053 * structure alive */
5054 if (cgroup_is_releasable(cgrp
) &&
5055 !atomic_read(&cgrp
->count
) && list_empty(&cgrp
->children
)) {
5057 * Control Group is currently removeable. If it's not
5058 * already queued for a userspace notification, queue
5061 int need_schedule_work
= 0;
5063 raw_spin_lock(&release_list_lock
);
5064 if (!cgroup_is_removed(cgrp
) &&
5065 list_empty(&cgrp
->release_list
)) {
5066 list_add(&cgrp
->release_list
, &release_list
);
5067 need_schedule_work
= 1;
5069 raw_spin_unlock(&release_list_lock
);
5070 if (need_schedule_work
)
5071 schedule_work(&release_agent_work
);
5075 /* Caller must verify that the css is not for root cgroup */
5076 bool __css_tryget(struct cgroup_subsys_state
*css
)
5081 v
= css_refcnt(css
);
5082 t
= atomic_cmpxchg(&css
->refcnt
, v
, v
+ 1);
5090 EXPORT_SYMBOL_GPL(__css_tryget
);
5092 /* Caller must verify that the css is not for root cgroup */
5093 void __css_put(struct cgroup_subsys_state
*css
)
5097 v
= css_unbias_refcnt(atomic_dec_return(&css
->refcnt
));
5099 schedule_work(&css
->dput_work
);
5101 EXPORT_SYMBOL_GPL(__css_put
);
5104 * Notify userspace when a cgroup is released, by running the
5105 * configured release agent with the name of the cgroup (path
5106 * relative to the root of cgroup file system) as the argument.
5108 * Most likely, this user command will try to rmdir this cgroup.
5110 * This races with the possibility that some other task will be
5111 * attached to this cgroup before it is removed, or that some other
5112 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5113 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5114 * unused, and this cgroup will be reprieved from its death sentence,
5115 * to continue to serve a useful existence. Next time it's released,
5116 * we will get notified again, if it still has 'notify_on_release' set.
5118 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5119 * means only wait until the task is successfully execve()'d. The
5120 * separate release agent task is forked by call_usermodehelper(),
5121 * then control in this thread returns here, without waiting for the
5122 * release agent task. We don't bother to wait because the caller of
5123 * this routine has no use for the exit status of the release agent
5124 * task, so no sense holding our caller up for that.
5126 static void cgroup_release_agent(struct work_struct
*work
)
5128 BUG_ON(work
!= &release_agent_work
);
5129 mutex_lock(&cgroup_mutex
);
5130 raw_spin_lock(&release_list_lock
);
5131 while (!list_empty(&release_list
)) {
5132 char *argv
[3], *envp
[3];
5134 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5135 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5138 list_del_init(&cgrp
->release_list
);
5139 raw_spin_unlock(&release_list_lock
);
5140 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5143 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5145 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5150 argv
[i
++] = agentbuf
;
5151 argv
[i
++] = pathbuf
;
5155 /* minimal command environment */
5156 envp
[i
++] = "HOME=/";
5157 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5160 /* Drop the lock while we invoke the usermode helper,
5161 * since the exec could involve hitting disk and hence
5162 * be a slow process */
5163 mutex_unlock(&cgroup_mutex
);
5164 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5165 mutex_lock(&cgroup_mutex
);
5169 raw_spin_lock(&release_list_lock
);
5171 raw_spin_unlock(&release_list_lock
);
5172 mutex_unlock(&cgroup_mutex
);
5175 static int __init
cgroup_disable(char *str
)
5180 while ((token
= strsep(&str
, ",")) != NULL
) {
5183 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
5184 struct cgroup_subsys
*ss
= subsys
[i
];
5187 * cgroup_disable, being at boot time, can't
5188 * know about module subsystems, so we don't
5191 if (!ss
|| ss
->module
)
5194 if (!strcmp(token
, ss
->name
)) {
5196 printk(KERN_INFO
"Disabling %s control group"
5197 " subsystem\n", ss
->name
);
5204 __setup("cgroup_disable=", cgroup_disable
);
5207 * Functons for CSS ID.
5211 *To get ID other than 0, this should be called when !cgroup_is_removed().
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
);
5230 unsigned short css_depth(struct cgroup_subsys_state
*css
)
5232 struct css_id
*cssid
;
5234 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5237 return cssid
->depth
;
5240 EXPORT_SYMBOL_GPL(css_depth
);
5243 * css_is_ancestor - test "root" css is an ancestor of "child"
5244 * @child: the css to be tested.
5245 * @root: the css supporsed to be an ancestor of the child.
5247 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5248 * this function reads css->id, the caller must hold rcu_read_lock().
5249 * But, considering usual usage, the csses should be valid objects after test.
5250 * Assuming that the caller will do some action to the child if this returns
5251 * returns true, the caller must take "child";s reference count.
5252 * If "child" is valid object and this returns true, "root" is valid, too.
5255 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5256 const struct cgroup_subsys_state
*root
)
5258 struct css_id
*child_id
;
5259 struct css_id
*root_id
;
5261 child_id
= rcu_dereference(child
->id
);
5264 root_id
= rcu_dereference(root
->id
);
5267 if (child_id
->depth
< root_id
->depth
)
5269 if (child_id
->stack
[root_id
->depth
] != root_id
->id
)
5274 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5276 struct css_id
*id
= css
->id
;
5277 /* When this is called before css_id initialization, id can be NULL */
5281 BUG_ON(!ss
->use_id
);
5283 rcu_assign_pointer(id
->css
, NULL
);
5284 rcu_assign_pointer(css
->id
, NULL
);
5285 spin_lock(&ss
->id_lock
);
5286 idr_remove(&ss
->idr
, id
->id
);
5287 spin_unlock(&ss
->id_lock
);
5288 kfree_rcu(id
, rcu_head
);
5290 EXPORT_SYMBOL_GPL(free_css_id
);
5293 * This is called by init or create(). Then, calls to this function are
5294 * always serialized (By cgroup_mutex() at create()).
5297 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5299 struct css_id
*newid
;
5302 BUG_ON(!ss
->use_id
);
5304 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5305 newid
= kzalloc(size
, GFP_KERNEL
);
5307 return ERR_PTR(-ENOMEM
);
5309 idr_preload(GFP_KERNEL
);
5310 spin_lock(&ss
->id_lock
);
5311 /* Don't use 0. allocates an ID of 1-65535 */
5312 ret
= idr_alloc(&ss
->idr
, newid
, 1, CSS_ID_MAX
+ 1, GFP_NOWAIT
);
5313 spin_unlock(&ss
->id_lock
);
5316 /* Returns error when there are no free spaces for new ID.*/
5321 newid
->depth
= depth
;
5325 return ERR_PTR(ret
);
5329 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5330 struct cgroup_subsys_state
*rootcss
)
5332 struct css_id
*newid
;
5334 spin_lock_init(&ss
->id_lock
);
5337 newid
= get_new_cssid(ss
, 0);
5339 return PTR_ERR(newid
);
5341 newid
->stack
[0] = newid
->id
;
5342 newid
->css
= rootcss
;
5343 rootcss
->id
= newid
;
5347 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5348 struct cgroup
*child
)
5350 int subsys_id
, i
, depth
= 0;
5351 struct cgroup_subsys_state
*parent_css
, *child_css
;
5352 struct css_id
*child_id
, *parent_id
;
5354 subsys_id
= ss
->subsys_id
;
5355 parent_css
= parent
->subsys
[subsys_id
];
5356 child_css
= child
->subsys
[subsys_id
];
5357 parent_id
= parent_css
->id
;
5358 depth
= parent_id
->depth
+ 1;
5360 child_id
= get_new_cssid(ss
, depth
);
5361 if (IS_ERR(child_id
))
5362 return PTR_ERR(child_id
);
5364 for (i
= 0; i
< depth
; i
++)
5365 child_id
->stack
[i
] = parent_id
->stack
[i
];
5366 child_id
->stack
[depth
] = child_id
->id
;
5368 * child_id->css pointer will be set after this cgroup is available
5369 * see cgroup_populate_dir()
5371 rcu_assign_pointer(child_css
->id
, child_id
);
5377 * css_lookup - lookup css by id
5378 * @ss: cgroup subsys to be looked into.
5381 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5382 * NULL if not. Should be called under rcu_read_lock()
5384 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5386 struct css_id
*cssid
= NULL
;
5388 BUG_ON(!ss
->use_id
);
5389 cssid
= idr_find(&ss
->idr
, id
);
5391 if (unlikely(!cssid
))
5394 return rcu_dereference(cssid
->css
);
5396 EXPORT_SYMBOL_GPL(css_lookup
);
5399 * get corresponding css from file open on cgroupfs directory
5401 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5403 struct cgroup
*cgrp
;
5404 struct inode
*inode
;
5405 struct cgroup_subsys_state
*css
;
5407 inode
= file_inode(f
);
5408 /* check in cgroup filesystem dir */
5409 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5410 return ERR_PTR(-EBADF
);
5412 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5413 return ERR_PTR(-EINVAL
);
5416 cgrp
= __d_cgrp(f
->f_dentry
);
5417 css
= cgrp
->subsys
[id
];
5418 return css
? css
: ERR_PTR(-ENOENT
);
5421 #ifdef CONFIG_CGROUP_DEBUG
5422 static struct cgroup_subsys_state
*debug_css_alloc(struct cgroup
*cont
)
5424 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5427 return ERR_PTR(-ENOMEM
);
5432 static void debug_css_free(struct cgroup
*cont
)
5434 kfree(cont
->subsys
[debug_subsys_id
]);
5437 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5439 return atomic_read(&cont
->count
);
5442 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5444 return cgroup_task_count(cont
);
5447 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5449 return (u64
)(unsigned long)current
->cgroups
;
5452 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5458 count
= atomic_read(¤t
->cgroups
->refcount
);
5463 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5465 struct seq_file
*seq
)
5467 struct cg_cgroup_link
*link
;
5470 read_lock(&css_set_lock
);
5472 cg
= rcu_dereference(current
->cgroups
);
5473 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5474 struct cgroup
*c
= link
->cgrp
;
5478 name
= c
->dentry
->d_name
.name
;
5481 seq_printf(seq
, "Root %d group %s\n",
5482 c
->root
->hierarchy_id
, name
);
5485 read_unlock(&css_set_lock
);
5489 #define MAX_TASKS_SHOWN_PER_CSS 25
5490 static int cgroup_css_links_read(struct cgroup
*cont
,
5492 struct seq_file
*seq
)
5494 struct cg_cgroup_link
*link
;
5496 read_lock(&css_set_lock
);
5497 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5498 struct css_set
*cg
= link
->cg
;
5499 struct task_struct
*task
;
5501 seq_printf(seq
, "css_set %p\n", cg
);
5502 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5503 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5504 seq_puts(seq
, " ...\n");
5507 seq_printf(seq
, " task %d\n",
5508 task_pid_vnr(task
));
5512 read_unlock(&css_set_lock
);
5516 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5518 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5521 static struct cftype debug_files
[] = {
5523 .name
= "cgroup_refcount",
5524 .read_u64
= cgroup_refcount_read
,
5527 .name
= "taskcount",
5528 .read_u64
= debug_taskcount_read
,
5532 .name
= "current_css_set",
5533 .read_u64
= current_css_set_read
,
5537 .name
= "current_css_set_refcount",
5538 .read_u64
= current_css_set_refcount_read
,
5542 .name
= "current_css_set_cg_links",
5543 .read_seq_string
= current_css_set_cg_links_read
,
5547 .name
= "cgroup_css_links",
5548 .read_seq_string
= cgroup_css_links_read
,
5552 .name
= "releasable",
5553 .read_u64
= releasable_read
,
5559 struct cgroup_subsys debug_subsys
= {
5561 .css_alloc
= debug_css_alloc
,
5562 .css_free
= debug_css_free
,
5563 .subsys_id
= debug_subsys_id
,
5564 .base_cftypes
= debug_files
,
5566 #endif /* CONFIG_CGROUP_DEBUG */