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>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hash.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_proc */
63 #include <linux/kthread.h>
65 #include <linux/atomic.h>
67 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
68 #define CSS_DEACT_BIAS INT_MIN
71 * cgroup_mutex is the master lock. Any modification to cgroup or its
72 * hierarchy must be performed while holding it.
74 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
75 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
76 * release_agent_path and so on. Modifying requires both cgroup_mutex and
77 * cgroup_root_mutex. Readers can acquire either of the two. This is to
78 * break the following locking order cycle.
80 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
81 * B. namespace_sem -> cgroup_mutex
83 * B happens only through cgroup_show_options() and using cgroup_root_mutex
86 static DEFINE_MUTEX(cgroup_mutex
);
87 static DEFINE_MUTEX(cgroup_root_mutex
);
90 * Generate an array of cgroup subsystem pointers. At boot time, this is
91 * populated with the built in subsystems, and modular subsystems are
92 * registered after that. The mutable section of this array is protected by
95 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
96 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
97 static struct cgroup_subsys
*subsys
[CGROUP_SUBSYS_COUNT
] = {
98 #include <linux/cgroup_subsys.h>
101 #define MAX_CGROUP_ROOT_NAMELEN 64
104 * A cgroupfs_root represents the root of a cgroup hierarchy,
105 * and may be associated with a superblock to form an active
108 struct cgroupfs_root
{
109 struct super_block
*sb
;
112 * The bitmask of subsystems intended to be attached to this
115 unsigned long subsys_mask
;
117 /* Unique id for this hierarchy. */
120 /* The bitmask of subsystems currently attached to this hierarchy */
121 unsigned long actual_subsys_mask
;
123 /* A list running through the attached subsystems */
124 struct list_head subsys_list
;
126 /* The root cgroup for this hierarchy */
127 struct cgroup top_cgroup
;
129 /* Tracks how many cgroups are currently defined in hierarchy.*/
130 int number_of_cgroups
;
132 /* A list running through the active hierarchies */
133 struct list_head root_list
;
135 /* All cgroups on this root, cgroup_mutex protected */
136 struct list_head allcg_list
;
138 /* Hierarchy-specific flags */
141 /* IDs for cgroups in this hierarchy */
142 struct ida cgroup_ida
;
144 /* The path to use for release notifications. */
145 char release_agent_path
[PATH_MAX
];
147 /* The name for this hierarchy - may be empty */
148 char name
[MAX_CGROUP_ROOT_NAMELEN
];
152 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
153 * subsystems that are otherwise unattached - it never has more than a
154 * single cgroup, and all tasks are part of that cgroup.
156 static struct cgroupfs_root rootnode
;
159 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
162 struct list_head node
;
163 struct dentry
*dentry
;
168 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
169 * cgroup_subsys->use_id != 0.
171 #define CSS_ID_MAX (65535)
174 * The css to which this ID points. This pointer is set to valid value
175 * after cgroup is populated. If cgroup is removed, this will be NULL.
176 * This pointer is expected to be RCU-safe because destroy()
177 * is called after synchronize_rcu(). But for safe use, css_tryget()
178 * should be used for avoiding race.
180 struct cgroup_subsys_state __rcu
*css
;
186 * Depth in hierarchy which this ID belongs to.
188 unsigned short depth
;
190 * ID is freed by RCU. (and lookup routine is RCU safe.)
192 struct rcu_head rcu_head
;
194 * Hierarchy of CSS ID belongs to.
196 unsigned short stack
[0]; /* Array of Length (depth+1) */
200 * cgroup_event represents events which userspace want to receive.
202 struct cgroup_event
{
204 * Cgroup which the event belongs to.
208 * Control file which the event associated.
212 * eventfd to signal userspace about the event.
214 struct eventfd_ctx
*eventfd
;
216 * Each of these stored in a list by the cgroup.
218 struct list_head list
;
220 * All fields below needed to unregister event when
221 * userspace closes eventfd.
224 wait_queue_head_t
*wqh
;
226 struct work_struct remove
;
229 /* The list of hierarchy roots */
231 static LIST_HEAD(roots
);
232 static int root_count
;
234 static DEFINE_IDA(hierarchy_ida
);
235 static int next_hierarchy_id
;
236 static DEFINE_SPINLOCK(hierarchy_id_lock
);
238 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
239 #define dummytop (&rootnode.top_cgroup)
241 /* This flag indicates whether tasks in the fork and exit paths should
242 * check for fork/exit handlers to call. This avoids us having to do
243 * extra work in the fork/exit path if none of the subsystems need to
246 static int need_forkexit_callback __read_mostly
;
248 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
249 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
250 struct cftype cfts
[], bool is_add
);
252 #ifdef CONFIG_PROVE_LOCKING
253 int cgroup_lock_is_held(void)
255 return lockdep_is_held(&cgroup_mutex
);
257 #else /* #ifdef CONFIG_PROVE_LOCKING */
258 int cgroup_lock_is_held(void)
260 return mutex_is_locked(&cgroup_mutex
);
262 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
264 EXPORT_SYMBOL_GPL(cgroup_lock_is_held
);
266 static int css_unbias_refcnt(int refcnt
)
268 return refcnt
>= 0 ? refcnt
: refcnt
- CSS_DEACT_BIAS
;
271 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
272 static int css_refcnt(struct cgroup_subsys_state
*css
)
274 int v
= atomic_read(&css
->refcnt
);
276 return css_unbias_refcnt(v
);
279 /* convenient tests for these bits */
280 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
282 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
285 /* bits in struct cgroupfs_root flags field */
287 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
288 ROOT_XATTR
, /* supports extended attributes */
291 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
294 (1 << CGRP_RELEASABLE
) |
295 (1 << CGRP_NOTIFY_ON_RELEASE
);
296 return (cgrp
->flags
& bits
) == bits
;
299 static int notify_on_release(const struct cgroup
*cgrp
)
301 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
305 * for_each_subsys() allows you to iterate on each subsystem attached to
306 * an active hierarchy
308 #define for_each_subsys(_root, _ss) \
309 list_for_each_entry(_ss, &_root->subsys_list, sibling)
311 /* for_each_active_root() allows you to iterate across the active hierarchies */
312 #define for_each_active_root(_root) \
313 list_for_each_entry(_root, &roots, root_list)
315 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
317 return dentry
->d_fsdata
;
320 static inline struct cfent
*__d_cfe(struct dentry
*dentry
)
322 return dentry
->d_fsdata
;
325 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
327 return __d_cfe(dentry
)->type
;
330 /* the list of cgroups eligible for automatic release. Protected by
331 * release_list_lock */
332 static LIST_HEAD(release_list
);
333 static DEFINE_RAW_SPINLOCK(release_list_lock
);
334 static void cgroup_release_agent(struct work_struct
*work
);
335 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
336 static void check_for_release(struct cgroup
*cgrp
);
338 /* Link structure for associating css_set objects with cgroups */
339 struct cg_cgroup_link
{
341 * List running through cg_cgroup_links associated with a
342 * cgroup, anchored on cgroup->css_sets
344 struct list_head cgrp_link_list
;
347 * List running through cg_cgroup_links pointing at a
348 * single css_set object, anchored on css_set->cg_links
350 struct list_head cg_link_list
;
354 /* The default css_set - used by init and its children prior to any
355 * hierarchies being mounted. It contains a pointer to the root state
356 * for each subsystem. Also used to anchor the list of css_sets. Not
357 * reference-counted, to improve performance when child cgroups
358 * haven't been created.
361 static struct css_set init_css_set
;
362 static struct cg_cgroup_link init_css_set_link
;
364 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
365 struct cgroup_subsys_state
*css
);
367 /* css_set_lock protects the list of css_set objects, and the
368 * chain of tasks off each css_set. Nests outside task->alloc_lock
369 * due to cgroup_iter_start() */
370 static DEFINE_RWLOCK(css_set_lock
);
371 static int css_set_count
;
374 * hash table for cgroup groups. This improves the performance to find
375 * an existing css_set. This hash doesn't (currently) take into
376 * account cgroups in empty hierarchies.
378 #define CSS_SET_HASH_BITS 7
379 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
380 static struct hlist_head css_set_table
[CSS_SET_TABLE_SIZE
];
382 static struct hlist_head
*css_set_hash(struct cgroup_subsys_state
*css
[])
386 unsigned long tmp
= 0UL;
388 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
389 tmp
+= (unsigned long)css
[i
];
390 tmp
= (tmp
>> 16) ^ tmp
;
392 index
= hash_long(tmp
, CSS_SET_HASH_BITS
);
394 return &css_set_table
[index
];
397 /* We don't maintain the lists running through each css_set to its
398 * task until after the first call to cgroup_iter_start(). This
399 * reduces the fork()/exit() overhead for people who have cgroups
400 * compiled into their kernel but not actually in use */
401 static int use_task_css_set_links __read_mostly
;
403 static void __put_css_set(struct css_set
*cg
, int taskexit
)
405 struct cg_cgroup_link
*link
;
406 struct cg_cgroup_link
*saved_link
;
408 * Ensure that the refcount doesn't hit zero while any readers
409 * can see it. Similar to atomic_dec_and_lock(), but for an
412 if (atomic_add_unless(&cg
->refcount
, -1, 1))
414 write_lock(&css_set_lock
);
415 if (!atomic_dec_and_test(&cg
->refcount
)) {
416 write_unlock(&css_set_lock
);
420 /* This css_set is dead. unlink it and release cgroup refcounts */
421 hlist_del(&cg
->hlist
);
424 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
426 struct cgroup
*cgrp
= link
->cgrp
;
427 list_del(&link
->cg_link_list
);
428 list_del(&link
->cgrp_link_list
);
429 if (atomic_dec_and_test(&cgrp
->count
) &&
430 notify_on_release(cgrp
)) {
432 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
433 check_for_release(cgrp
);
439 write_unlock(&css_set_lock
);
440 kfree_rcu(cg
, rcu_head
);
444 * refcounted get/put for css_set objects
446 static inline void get_css_set(struct css_set
*cg
)
448 atomic_inc(&cg
->refcount
);
451 static inline void put_css_set(struct css_set
*cg
)
453 __put_css_set(cg
, 0);
456 static inline void put_css_set_taskexit(struct css_set
*cg
)
458 __put_css_set(cg
, 1);
462 * compare_css_sets - helper function for find_existing_css_set().
463 * @cg: candidate css_set being tested
464 * @old_cg: existing css_set for a task
465 * @new_cgrp: cgroup that's being entered by the task
466 * @template: desired set of css pointers in css_set (pre-calculated)
468 * Returns true if "cg" matches "old_cg" except for the hierarchy
469 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
471 static bool compare_css_sets(struct css_set
*cg
,
472 struct css_set
*old_cg
,
473 struct cgroup
*new_cgrp
,
474 struct cgroup_subsys_state
*template[])
476 struct list_head
*l1
, *l2
;
478 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
479 /* Not all subsystems matched */
484 * Compare cgroup pointers in order to distinguish between
485 * different cgroups in heirarchies with no subsystems. We
486 * could get by with just this check alone (and skip the
487 * memcmp above) but on most setups the memcmp check will
488 * avoid the need for this more expensive check on almost all
493 l2
= &old_cg
->cg_links
;
495 struct cg_cgroup_link
*cgl1
, *cgl2
;
496 struct cgroup
*cg1
, *cg2
;
500 /* See if we reached the end - both lists are equal length. */
501 if (l1
== &cg
->cg_links
) {
502 BUG_ON(l2
!= &old_cg
->cg_links
);
505 BUG_ON(l2
== &old_cg
->cg_links
);
507 /* Locate the cgroups associated with these links. */
508 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
509 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
512 /* Hierarchies should be linked in the same order. */
513 BUG_ON(cg1
->root
!= cg2
->root
);
516 * If this hierarchy is the hierarchy of the cgroup
517 * that's changing, then we need to check that this
518 * css_set points to the new cgroup; if it's any other
519 * hierarchy, then this css_set should point to the
520 * same cgroup as the old css_set.
522 if (cg1
->root
== new_cgrp
->root
) {
534 * find_existing_css_set() is a helper for
535 * find_css_set(), and checks to see whether an existing
536 * css_set is suitable.
538 * oldcg: the cgroup group that we're using before the cgroup
541 * cgrp: the cgroup that we're moving into
543 * template: location in which to build the desired set of subsystem
544 * state objects for the new cgroup group
546 static struct css_set
*find_existing_css_set(
547 struct css_set
*oldcg
,
549 struct cgroup_subsys_state
*template[])
552 struct cgroupfs_root
*root
= cgrp
->root
;
553 struct hlist_head
*hhead
;
554 struct hlist_node
*node
;
558 * Build the set of subsystem state objects that we want to see in the
559 * new css_set. while subsystems can change globally, the entries here
560 * won't change, so no need for locking.
562 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
563 if (root
->subsys_mask
& (1UL << i
)) {
564 /* Subsystem is in this hierarchy. So we want
565 * the subsystem state from the new
567 template[i
] = cgrp
->subsys
[i
];
569 /* Subsystem is not in this hierarchy, so we
570 * don't want to change the subsystem state */
571 template[i
] = oldcg
->subsys
[i
];
575 hhead
= css_set_hash(template);
576 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
577 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
580 /* This css_set matches what we need */
584 /* No existing cgroup group matched */
588 static void free_cg_links(struct list_head
*tmp
)
590 struct cg_cgroup_link
*link
;
591 struct cg_cgroup_link
*saved_link
;
593 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
594 list_del(&link
->cgrp_link_list
);
600 * allocate_cg_links() allocates "count" cg_cgroup_link structures
601 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
602 * success or a negative error
604 static int allocate_cg_links(int count
, struct list_head
*tmp
)
606 struct cg_cgroup_link
*link
;
609 for (i
= 0; i
< count
; i
++) {
610 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
615 list_add(&link
->cgrp_link_list
, tmp
);
621 * link_css_set - a helper function to link a css_set to a cgroup
622 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
623 * @cg: the css_set to be linked
624 * @cgrp: the destination cgroup
626 static void link_css_set(struct list_head
*tmp_cg_links
,
627 struct css_set
*cg
, struct cgroup
*cgrp
)
629 struct cg_cgroup_link
*link
;
631 BUG_ON(list_empty(tmp_cg_links
));
632 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
636 atomic_inc(&cgrp
->count
);
637 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
639 * Always add links to the tail of the list so that the list
640 * is sorted by order of hierarchy creation
642 list_add_tail(&link
->cg_link_list
, &cg
->cg_links
);
646 * find_css_set() takes an existing cgroup group and a
647 * cgroup object, and returns a css_set object that's
648 * equivalent to the old group, but with the given cgroup
649 * substituted into the appropriate hierarchy. Must be called with
652 static struct css_set
*find_css_set(
653 struct css_set
*oldcg
, struct cgroup
*cgrp
)
656 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
658 struct list_head tmp_cg_links
;
660 struct hlist_head
*hhead
;
661 struct cg_cgroup_link
*link
;
663 /* First see if we already have a cgroup group that matches
665 read_lock(&css_set_lock
);
666 res
= find_existing_css_set(oldcg
, cgrp
, template);
669 read_unlock(&css_set_lock
);
674 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
678 /* Allocate all the cg_cgroup_link objects that we'll need */
679 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
684 atomic_set(&res
->refcount
, 1);
685 INIT_LIST_HEAD(&res
->cg_links
);
686 INIT_LIST_HEAD(&res
->tasks
);
687 INIT_HLIST_NODE(&res
->hlist
);
689 /* Copy the set of subsystem state objects generated in
690 * find_existing_css_set() */
691 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
693 write_lock(&css_set_lock
);
694 /* Add reference counts and links from the new css_set. */
695 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
696 struct cgroup
*c
= link
->cgrp
;
697 if (c
->root
== cgrp
->root
)
699 link_css_set(&tmp_cg_links
, res
, c
);
702 BUG_ON(!list_empty(&tmp_cg_links
));
706 /* Add this cgroup group to the hash table */
707 hhead
= css_set_hash(res
->subsys
);
708 hlist_add_head(&res
->hlist
, hhead
);
710 write_unlock(&css_set_lock
);
716 * Return the cgroup for "task" from the given hierarchy. Must be
717 * called with cgroup_mutex held.
719 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
720 struct cgroupfs_root
*root
)
723 struct cgroup
*res
= NULL
;
725 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
726 read_lock(&css_set_lock
);
728 * No need to lock the task - since we hold cgroup_mutex the
729 * task can't change groups, so the only thing that can happen
730 * is that it exits and its css is set back to init_css_set.
733 if (css
== &init_css_set
) {
734 res
= &root
->top_cgroup
;
736 struct cg_cgroup_link
*link
;
737 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
738 struct cgroup
*c
= link
->cgrp
;
739 if (c
->root
== root
) {
745 read_unlock(&css_set_lock
);
751 * There is one global cgroup mutex. We also require taking
752 * task_lock() when dereferencing a task's cgroup subsys pointers.
753 * See "The task_lock() exception", at the end of this comment.
755 * A task must hold cgroup_mutex to modify cgroups.
757 * Any task can increment and decrement the count field without lock.
758 * So in general, code holding cgroup_mutex can't rely on the count
759 * field not changing. However, if the count goes to zero, then only
760 * cgroup_attach_task() can increment it again. Because a count of zero
761 * means that no tasks are currently attached, therefore there is no
762 * way a task attached to that cgroup can fork (the other way to
763 * increment the count). So code holding cgroup_mutex can safely
764 * assume that if the count is zero, it will stay zero. Similarly, if
765 * a task holds cgroup_mutex on a cgroup with zero count, it
766 * knows that the cgroup won't be removed, as cgroup_rmdir()
769 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
770 * (usually) take cgroup_mutex. These are the two most performance
771 * critical pieces of code here. The exception occurs on cgroup_exit(),
772 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
773 * is taken, and if the cgroup count is zero, a usermode call made
774 * to the release agent with the name of the cgroup (path relative to
775 * the root of cgroup file system) as the argument.
777 * A cgroup can only be deleted if both its 'count' of using tasks
778 * is zero, and its list of 'children' cgroups is empty. Since all
779 * tasks in the system use _some_ cgroup, and since there is always at
780 * least one task in the system (init, pid == 1), therefore, top_cgroup
781 * always has either children cgroups and/or using tasks. So we don't
782 * need a special hack to ensure that top_cgroup cannot be deleted.
784 * The task_lock() exception
786 * The need for this exception arises from the action of
787 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
788 * another. It does so using cgroup_mutex, however there are
789 * several performance critical places that need to reference
790 * task->cgroup without the expense of grabbing a system global
791 * mutex. Therefore except as noted below, when dereferencing or, as
792 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
793 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
794 * the task_struct routinely used for such matters.
796 * P.S. One more locking exception. RCU is used to guard the
797 * update of a tasks cgroup pointer by cgroup_attach_task()
801 * cgroup_lock - lock out any changes to cgroup structures
804 void cgroup_lock(void)
806 mutex_lock(&cgroup_mutex
);
808 EXPORT_SYMBOL_GPL(cgroup_lock
);
811 * cgroup_unlock - release lock on cgroup changes
813 * Undo the lock taken in a previous cgroup_lock() call.
815 void cgroup_unlock(void)
817 mutex_unlock(&cgroup_mutex
);
819 EXPORT_SYMBOL_GPL(cgroup_unlock
);
822 * A couple of forward declarations required, due to cyclic reference loop:
823 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
824 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
828 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
829 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, unsigned int);
830 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
831 static int cgroup_populate_dir(struct cgroup
*cgrp
, bool base_files
,
832 unsigned long subsys_mask
);
833 static const struct inode_operations cgroup_dir_inode_operations
;
834 static const struct file_operations proc_cgroupstats_operations
;
836 static struct backing_dev_info cgroup_backing_dev_info
= {
838 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
841 static int alloc_css_id(struct cgroup_subsys
*ss
,
842 struct cgroup
*parent
, struct cgroup
*child
);
844 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
846 struct inode
*inode
= new_inode(sb
);
849 inode
->i_ino
= get_next_ino();
850 inode
->i_mode
= mode
;
851 inode
->i_uid
= current_fsuid();
852 inode
->i_gid
= current_fsgid();
853 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
854 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
859 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
861 /* is dentry a directory ? if so, kfree() associated cgroup */
862 if (S_ISDIR(inode
->i_mode
)) {
863 struct cgroup
*cgrp
= dentry
->d_fsdata
;
864 struct cgroup_subsys
*ss
;
865 BUG_ON(!(cgroup_is_removed(cgrp
)));
866 /* It's possible for external users to be holding css
867 * reference counts on a cgroup; css_put() needs to
868 * be able to access the cgroup after decrementing
869 * the reference count in order to know if it needs to
870 * queue the cgroup to be handled by the release
874 mutex_lock(&cgroup_mutex
);
876 * Release the subsystem state objects.
878 for_each_subsys(cgrp
->root
, ss
)
881 cgrp
->root
->number_of_cgroups
--;
882 mutex_unlock(&cgroup_mutex
);
885 * Drop the active superblock reference that we took when we
888 deactivate_super(cgrp
->root
->sb
);
891 * if we're getting rid of the cgroup, refcount should ensure
892 * that there are no pidlists left.
894 BUG_ON(!list_empty(&cgrp
->pidlists
));
896 simple_xattrs_free(&cgrp
->xattrs
);
898 ida_simple_remove(&cgrp
->root
->cgroup_ida
, cgrp
->id
);
899 kfree_rcu(cgrp
, rcu_head
);
901 struct cfent
*cfe
= __d_cfe(dentry
);
902 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
903 struct cftype
*cft
= cfe
->type
;
905 WARN_ONCE(!list_empty(&cfe
->node
) &&
906 cgrp
!= &cgrp
->root
->top_cgroup
,
907 "cfe still linked for %s\n", cfe
->type
->name
);
909 simple_xattrs_free(&cft
->xattrs
);
914 static int cgroup_delete(const struct dentry
*d
)
919 static void remove_dir(struct dentry
*d
)
921 struct dentry
*parent
= dget(d
->d_parent
);
924 simple_rmdir(parent
->d_inode
, d
);
928 static int cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
932 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
933 lockdep_assert_held(&cgroup_mutex
);
935 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
936 struct dentry
*d
= cfe
->dentry
;
938 if (cft
&& cfe
->type
!= cft
)
943 simple_unlink(cgrp
->dentry
->d_inode
, d
);
944 list_del_init(&cfe
->node
);
953 * cgroup_clear_directory - selective removal of base and subsystem files
954 * @dir: directory containing the files
955 * @base_files: true if the base files should be removed
956 * @subsys_mask: mask of the subsystem ids whose files should be removed
958 static void cgroup_clear_directory(struct dentry
*dir
, bool base_files
,
959 unsigned long subsys_mask
)
961 struct cgroup
*cgrp
= __d_cgrp(dir
);
962 struct cgroup_subsys
*ss
;
964 for_each_subsys(cgrp
->root
, ss
) {
965 struct cftype_set
*set
;
966 if (!test_bit(ss
->subsys_id
, &subsys_mask
))
968 list_for_each_entry(set
, &ss
->cftsets
, node
)
969 cgroup_addrm_files(cgrp
, NULL
, set
->cfts
, false);
972 while (!list_empty(&cgrp
->files
))
973 cgroup_rm_file(cgrp
, NULL
);
978 * NOTE : the dentry must have been dget()'ed
980 static void cgroup_d_remove_dir(struct dentry
*dentry
)
982 struct dentry
*parent
;
983 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
985 cgroup_clear_directory(dentry
, true, root
->subsys_mask
);
987 parent
= dentry
->d_parent
;
988 spin_lock(&parent
->d_lock
);
989 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
990 list_del_init(&dentry
->d_u
.d_child
);
991 spin_unlock(&dentry
->d_lock
);
992 spin_unlock(&parent
->d_lock
);
997 * Call with cgroup_mutex held. Drops reference counts on modules, including
998 * any duplicate ones that parse_cgroupfs_options took. If this function
999 * returns an error, no reference counts are touched.
1001 static int rebind_subsystems(struct cgroupfs_root
*root
,
1002 unsigned long final_subsys_mask
)
1004 unsigned long added_mask
, removed_mask
;
1005 struct cgroup
*cgrp
= &root
->top_cgroup
;
1008 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1009 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1011 removed_mask
= root
->actual_subsys_mask
& ~final_subsys_mask
;
1012 added_mask
= final_subsys_mask
& ~root
->actual_subsys_mask
;
1013 /* Check that any added subsystems are currently free */
1014 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1015 unsigned long bit
= 1UL << i
;
1016 struct cgroup_subsys
*ss
= subsys
[i
];
1017 if (!(bit
& added_mask
))
1020 * Nobody should tell us to do a subsys that doesn't exist:
1021 * parse_cgroupfs_options should catch that case and refcounts
1022 * ensure that subsystems won't disappear once selected.
1025 if (ss
->root
!= &rootnode
) {
1026 /* Subsystem isn't free */
1031 /* Currently we don't handle adding/removing subsystems when
1032 * any child cgroups exist. This is theoretically supportable
1033 * but involves complex error handling, so it's being left until
1035 if (root
->number_of_cgroups
> 1)
1038 /* Process each subsystem */
1039 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1040 struct cgroup_subsys
*ss
= subsys
[i
];
1041 unsigned long bit
= 1UL << i
;
1042 if (bit
& added_mask
) {
1043 /* We're binding this subsystem to this hierarchy */
1045 BUG_ON(cgrp
->subsys
[i
]);
1046 BUG_ON(!dummytop
->subsys
[i
]);
1047 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
1048 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
1049 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1050 list_move(&ss
->sibling
, &root
->subsys_list
);
1054 /* refcount was already taken, and we're keeping it */
1055 } else if (bit
& removed_mask
) {
1056 /* We're removing this subsystem */
1058 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1059 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1062 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1063 cgrp
->subsys
[i
] = NULL
;
1064 subsys
[i
]->root
= &rootnode
;
1065 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1066 /* subsystem is now free - drop reference on module */
1067 module_put(ss
->module
);
1068 } else if (bit
& final_subsys_mask
) {
1069 /* Subsystem state should already exist */
1071 BUG_ON(!cgrp
->subsys
[i
]);
1073 * a refcount was taken, but we already had one, so
1074 * drop the extra reference.
1076 module_put(ss
->module
);
1077 #ifdef CONFIG_MODULE_UNLOAD
1078 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1081 /* Subsystem state shouldn't exist */
1082 BUG_ON(cgrp
->subsys
[i
]);
1085 root
->subsys_mask
= root
->actual_subsys_mask
= final_subsys_mask
;
1091 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1093 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1094 struct cgroup_subsys
*ss
;
1096 mutex_lock(&cgroup_root_mutex
);
1097 for_each_subsys(root
, ss
)
1098 seq_printf(seq
, ",%s", ss
->name
);
1099 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1100 seq_puts(seq
, ",noprefix");
1101 if (test_bit(ROOT_XATTR
, &root
->flags
))
1102 seq_puts(seq
, ",xattr");
1103 if (strlen(root
->release_agent_path
))
1104 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1105 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
))
1106 seq_puts(seq
, ",clone_children");
1107 if (strlen(root
->name
))
1108 seq_printf(seq
, ",name=%s", root
->name
);
1109 mutex_unlock(&cgroup_root_mutex
);
1113 struct cgroup_sb_opts
{
1114 unsigned long subsys_mask
;
1115 unsigned long flags
;
1116 char *release_agent
;
1117 bool cpuset_clone_children
;
1119 /* User explicitly requested empty subsystem */
1122 struct cgroupfs_root
*new_root
;
1127 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1128 * with cgroup_mutex held to protect the subsys[] array. This function takes
1129 * refcounts on subsystems to be used, unless it returns error, in which case
1130 * no refcounts are taken.
1132 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1134 char *token
, *o
= data
;
1135 bool all_ss
= false, one_ss
= false;
1136 unsigned long mask
= (unsigned long)-1;
1138 bool module_pin_failed
= false;
1140 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1142 #ifdef CONFIG_CPUSETS
1143 mask
= ~(1UL << cpuset_subsys_id
);
1146 memset(opts
, 0, sizeof(*opts
));
1148 while ((token
= strsep(&o
, ",")) != NULL
) {
1151 if (!strcmp(token
, "none")) {
1152 /* Explicitly have no subsystems */
1156 if (!strcmp(token
, "all")) {
1157 /* Mutually exclusive option 'all' + subsystem name */
1163 if (!strcmp(token
, "noprefix")) {
1164 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1167 if (!strcmp(token
, "clone_children")) {
1168 opts
->cpuset_clone_children
= true;
1171 if (!strcmp(token
, "xattr")) {
1172 set_bit(ROOT_XATTR
, &opts
->flags
);
1175 if (!strncmp(token
, "release_agent=", 14)) {
1176 /* Specifying two release agents is forbidden */
1177 if (opts
->release_agent
)
1179 opts
->release_agent
=
1180 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1181 if (!opts
->release_agent
)
1185 if (!strncmp(token
, "name=", 5)) {
1186 const char *name
= token
+ 5;
1187 /* Can't specify an empty name */
1190 /* Must match [\w.-]+ */
1191 for (i
= 0; i
< strlen(name
); i
++) {
1195 if ((c
== '.') || (c
== '-') || (c
== '_'))
1199 /* Specifying two names is forbidden */
1202 opts
->name
= kstrndup(name
,
1203 MAX_CGROUP_ROOT_NAMELEN
- 1,
1211 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1212 struct cgroup_subsys
*ss
= subsys
[i
];
1215 if (strcmp(token
, ss
->name
))
1220 /* Mutually exclusive option 'all' + subsystem name */
1223 set_bit(i
, &opts
->subsys_mask
);
1228 if (i
== CGROUP_SUBSYS_COUNT
)
1233 * If the 'all' option was specified select all the subsystems,
1234 * otherwise if 'none', 'name=' and a subsystem name options
1235 * were not specified, let's default to 'all'
1237 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
)) {
1238 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1239 struct cgroup_subsys
*ss
= subsys
[i
];
1244 set_bit(i
, &opts
->subsys_mask
);
1248 /* Consistency checks */
1251 * Option noprefix was introduced just for backward compatibility
1252 * with the old cpuset, so we allow noprefix only if mounting just
1253 * the cpuset subsystem.
1255 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1256 (opts
->subsys_mask
& mask
))
1260 /* Can't specify "none" and some subsystems */
1261 if (opts
->subsys_mask
&& opts
->none
)
1265 * We either have to specify by name or by subsystems. (So all
1266 * empty hierarchies must have a name).
1268 if (!opts
->subsys_mask
&& !opts
->name
)
1272 * Grab references on all the modules we'll need, so the subsystems
1273 * don't dance around before rebind_subsystems attaches them. This may
1274 * take duplicate reference counts on a subsystem that's already used,
1275 * but rebind_subsystems handles this case.
1277 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1278 unsigned long bit
= 1UL << i
;
1280 if (!(bit
& opts
->subsys_mask
))
1282 if (!try_module_get(subsys
[i
]->module
)) {
1283 module_pin_failed
= true;
1287 if (module_pin_failed
) {
1289 * oops, one of the modules was going away. this means that we
1290 * raced with a module_delete call, and to the user this is
1291 * essentially a "subsystem doesn't exist" case.
1293 for (i
--; i
>= 0; i
--) {
1294 /* drop refcounts only on the ones we took */
1295 unsigned long bit
= 1UL << i
;
1297 if (!(bit
& opts
->subsys_mask
))
1299 module_put(subsys
[i
]->module
);
1307 static void drop_parsed_module_refcounts(unsigned long subsys_mask
)
1310 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1311 unsigned long bit
= 1UL << i
;
1313 if (!(bit
& subsys_mask
))
1315 module_put(subsys
[i
]->module
);
1319 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1322 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1323 struct cgroup
*cgrp
= &root
->top_cgroup
;
1324 struct cgroup_sb_opts opts
;
1325 unsigned long added_mask
, removed_mask
;
1327 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1328 mutex_lock(&cgroup_mutex
);
1329 mutex_lock(&cgroup_root_mutex
);
1331 /* See what subsystems are wanted */
1332 ret
= parse_cgroupfs_options(data
, &opts
);
1336 /* See feature-removal-schedule.txt */
1337 if (opts
.subsys_mask
!= root
->actual_subsys_mask
|| opts
.release_agent
)
1338 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1339 task_tgid_nr(current
), current
->comm
);
1341 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1342 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1344 /* Don't allow flags or name to change at remount */
1345 if (opts
.flags
!= root
->flags
||
1346 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1348 drop_parsed_module_refcounts(opts
.subsys_mask
);
1353 * Clear out the files of subsystems that should be removed, do
1354 * this before rebind_subsystems, since rebind_subsystems may
1355 * change this hierarchy's subsys_list.
1357 cgroup_clear_directory(cgrp
->dentry
, false, removed_mask
);
1359 ret
= rebind_subsystems(root
, opts
.subsys_mask
);
1361 /* rebind_subsystems failed, re-populate the removed files */
1362 cgroup_populate_dir(cgrp
, false, removed_mask
);
1363 drop_parsed_module_refcounts(opts
.subsys_mask
);
1367 /* re-populate subsystem files */
1368 cgroup_populate_dir(cgrp
, false, added_mask
);
1370 if (opts
.release_agent
)
1371 strcpy(root
->release_agent_path
, opts
.release_agent
);
1373 kfree(opts
.release_agent
);
1375 mutex_unlock(&cgroup_root_mutex
);
1376 mutex_unlock(&cgroup_mutex
);
1377 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1381 static const struct super_operations cgroup_ops
= {
1382 .statfs
= simple_statfs
,
1383 .drop_inode
= generic_delete_inode
,
1384 .show_options
= cgroup_show_options
,
1385 .remount_fs
= cgroup_remount
,
1388 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1390 INIT_LIST_HEAD(&cgrp
->sibling
);
1391 INIT_LIST_HEAD(&cgrp
->children
);
1392 INIT_LIST_HEAD(&cgrp
->files
);
1393 INIT_LIST_HEAD(&cgrp
->css_sets
);
1394 INIT_LIST_HEAD(&cgrp
->allcg_node
);
1395 INIT_LIST_HEAD(&cgrp
->release_list
);
1396 INIT_LIST_HEAD(&cgrp
->pidlists
);
1397 mutex_init(&cgrp
->pidlist_mutex
);
1398 INIT_LIST_HEAD(&cgrp
->event_list
);
1399 spin_lock_init(&cgrp
->event_list_lock
);
1400 simple_xattrs_init(&cgrp
->xattrs
);
1403 static void init_cgroup_root(struct cgroupfs_root
*root
)
1405 struct cgroup
*cgrp
= &root
->top_cgroup
;
1407 INIT_LIST_HEAD(&root
->subsys_list
);
1408 INIT_LIST_HEAD(&root
->root_list
);
1409 INIT_LIST_HEAD(&root
->allcg_list
);
1410 root
->number_of_cgroups
= 1;
1412 cgrp
->top_cgroup
= cgrp
;
1413 init_cgroup_housekeeping(cgrp
);
1414 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
1417 static bool init_root_id(struct cgroupfs_root
*root
)
1422 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1424 spin_lock(&hierarchy_id_lock
);
1425 /* Try to allocate the next unused ID */
1426 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1427 &root
->hierarchy_id
);
1429 /* Try again starting from 0 */
1430 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1432 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1433 } else if (ret
!= -EAGAIN
) {
1434 /* Can only get here if the 31-bit IDR is full ... */
1437 spin_unlock(&hierarchy_id_lock
);
1442 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1444 struct cgroup_sb_opts
*opts
= data
;
1445 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1447 /* If we asked for a name then it must match */
1448 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1452 * If we asked for subsystems (or explicitly for no
1453 * subsystems) then they must match
1455 if ((opts
->subsys_mask
|| opts
->none
)
1456 && (opts
->subsys_mask
!= root
->subsys_mask
))
1462 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1464 struct cgroupfs_root
*root
;
1466 if (!opts
->subsys_mask
&& !opts
->none
)
1469 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1471 return ERR_PTR(-ENOMEM
);
1473 if (!init_root_id(root
)) {
1475 return ERR_PTR(-ENOMEM
);
1477 init_cgroup_root(root
);
1479 root
->subsys_mask
= opts
->subsys_mask
;
1480 root
->flags
= opts
->flags
;
1481 ida_init(&root
->cgroup_ida
);
1482 if (opts
->release_agent
)
1483 strcpy(root
->release_agent_path
, opts
->release_agent
);
1485 strcpy(root
->name
, opts
->name
);
1486 if (opts
->cpuset_clone_children
)
1487 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1491 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1496 BUG_ON(!root
->hierarchy_id
);
1497 spin_lock(&hierarchy_id_lock
);
1498 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1499 spin_unlock(&hierarchy_id_lock
);
1500 ida_destroy(&root
->cgroup_ida
);
1504 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1507 struct cgroup_sb_opts
*opts
= data
;
1509 /* If we don't have a new root, we can't set up a new sb */
1510 if (!opts
->new_root
)
1513 BUG_ON(!opts
->subsys_mask
&& !opts
->none
);
1515 ret
= set_anon_super(sb
, NULL
);
1519 sb
->s_fs_info
= opts
->new_root
;
1520 opts
->new_root
->sb
= sb
;
1522 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1523 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1524 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1525 sb
->s_op
= &cgroup_ops
;
1530 static int cgroup_get_rootdir(struct super_block
*sb
)
1532 static const struct dentry_operations cgroup_dops
= {
1533 .d_iput
= cgroup_diput
,
1534 .d_delete
= cgroup_delete
,
1537 struct inode
*inode
=
1538 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1543 inode
->i_fop
= &simple_dir_operations
;
1544 inode
->i_op
= &cgroup_dir_inode_operations
;
1545 /* directories start off with i_nlink == 2 (for "." entry) */
1547 sb
->s_root
= d_make_root(inode
);
1550 /* for everything else we want ->d_op set */
1551 sb
->s_d_op
= &cgroup_dops
;
1555 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1556 int flags
, const char *unused_dev_name
,
1559 struct cgroup_sb_opts opts
;
1560 struct cgroupfs_root
*root
;
1562 struct super_block
*sb
;
1563 struct cgroupfs_root
*new_root
;
1564 struct inode
*inode
;
1566 /* First find the desired set of subsystems */
1567 mutex_lock(&cgroup_mutex
);
1568 ret
= parse_cgroupfs_options(data
, &opts
);
1569 mutex_unlock(&cgroup_mutex
);
1574 * Allocate a new cgroup root. We may not need it if we're
1575 * reusing an existing hierarchy.
1577 new_root
= cgroup_root_from_opts(&opts
);
1578 if (IS_ERR(new_root
)) {
1579 ret
= PTR_ERR(new_root
);
1582 opts
.new_root
= new_root
;
1584 /* Locate an existing or new sb for this hierarchy */
1585 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, 0, &opts
);
1588 cgroup_drop_root(opts
.new_root
);
1592 root
= sb
->s_fs_info
;
1594 if (root
== opts
.new_root
) {
1595 /* We used the new root structure, so this is a new hierarchy */
1596 struct list_head tmp_cg_links
;
1597 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1598 struct cgroupfs_root
*existing_root
;
1599 const struct cred
*cred
;
1602 BUG_ON(sb
->s_root
!= NULL
);
1604 ret
= cgroup_get_rootdir(sb
);
1606 goto drop_new_super
;
1607 inode
= sb
->s_root
->d_inode
;
1609 mutex_lock(&inode
->i_mutex
);
1610 mutex_lock(&cgroup_mutex
);
1611 mutex_lock(&cgroup_root_mutex
);
1613 /* Check for name clashes with existing mounts */
1615 if (strlen(root
->name
))
1616 for_each_active_root(existing_root
)
1617 if (!strcmp(existing_root
->name
, root
->name
))
1621 * We're accessing css_set_count without locking
1622 * css_set_lock here, but that's OK - it can only be
1623 * increased by someone holding cgroup_lock, and
1624 * that's us. The worst that can happen is that we
1625 * have some link structures left over
1627 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1631 ret
= rebind_subsystems(root
, root
->subsys_mask
);
1632 if (ret
== -EBUSY
) {
1633 free_cg_links(&tmp_cg_links
);
1637 * There must be no failure case after here, since rebinding
1638 * takes care of subsystems' refcounts, which are explicitly
1639 * dropped in the failure exit path.
1642 /* EBUSY should be the only error here */
1645 list_add(&root
->root_list
, &roots
);
1648 sb
->s_root
->d_fsdata
= root_cgrp
;
1649 root
->top_cgroup
.dentry
= sb
->s_root
;
1651 /* Link the top cgroup in this hierarchy into all
1652 * the css_set objects */
1653 write_lock(&css_set_lock
);
1654 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1655 struct hlist_head
*hhead
= &css_set_table
[i
];
1656 struct hlist_node
*node
;
1659 hlist_for_each_entry(cg
, node
, hhead
, hlist
)
1660 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1662 write_unlock(&css_set_lock
);
1664 free_cg_links(&tmp_cg_links
);
1666 BUG_ON(!list_empty(&root_cgrp
->children
));
1667 BUG_ON(root
->number_of_cgroups
!= 1);
1669 cred
= override_creds(&init_cred
);
1670 cgroup_populate_dir(root_cgrp
, true, root
->subsys_mask
);
1672 mutex_unlock(&cgroup_root_mutex
);
1673 mutex_unlock(&cgroup_mutex
);
1674 mutex_unlock(&inode
->i_mutex
);
1677 * We re-used an existing hierarchy - the new root (if
1678 * any) is not needed
1680 cgroup_drop_root(opts
.new_root
);
1681 /* no subsys rebinding, so refcounts don't change */
1682 drop_parsed_module_refcounts(opts
.subsys_mask
);
1685 kfree(opts
.release_agent
);
1687 return dget(sb
->s_root
);
1690 mutex_unlock(&cgroup_root_mutex
);
1691 mutex_unlock(&cgroup_mutex
);
1692 mutex_unlock(&inode
->i_mutex
);
1694 deactivate_locked_super(sb
);
1696 drop_parsed_module_refcounts(opts
.subsys_mask
);
1698 kfree(opts
.release_agent
);
1700 return ERR_PTR(ret
);
1703 static void cgroup_kill_sb(struct super_block
*sb
) {
1704 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1705 struct cgroup
*cgrp
= &root
->top_cgroup
;
1707 struct cg_cgroup_link
*link
;
1708 struct cg_cgroup_link
*saved_link
;
1712 BUG_ON(root
->number_of_cgroups
!= 1);
1713 BUG_ON(!list_empty(&cgrp
->children
));
1715 mutex_lock(&cgroup_mutex
);
1716 mutex_lock(&cgroup_root_mutex
);
1718 /* Rebind all subsystems back to the default hierarchy */
1719 ret
= rebind_subsystems(root
, 0);
1720 /* Shouldn't be able to fail ... */
1724 * Release all the links from css_sets to this hierarchy's
1727 write_lock(&css_set_lock
);
1729 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1731 list_del(&link
->cg_link_list
);
1732 list_del(&link
->cgrp_link_list
);
1735 write_unlock(&css_set_lock
);
1737 if (!list_empty(&root
->root_list
)) {
1738 list_del(&root
->root_list
);
1742 mutex_unlock(&cgroup_root_mutex
);
1743 mutex_unlock(&cgroup_mutex
);
1745 simple_xattrs_free(&cgrp
->xattrs
);
1747 kill_litter_super(sb
);
1748 cgroup_drop_root(root
);
1751 static struct file_system_type cgroup_fs_type
= {
1753 .mount
= cgroup_mount
,
1754 .kill_sb
= cgroup_kill_sb
,
1757 static struct kobject
*cgroup_kobj
;
1760 * cgroup_path - generate the path of a cgroup
1761 * @cgrp: the cgroup in question
1762 * @buf: the buffer to write the path into
1763 * @buflen: the length of the buffer
1765 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1766 * reference. Writes path of cgroup into buf. Returns 0 on success,
1769 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1771 struct dentry
*dentry
= cgrp
->dentry
;
1774 rcu_lockdep_assert(rcu_read_lock_held() || cgroup_lock_is_held(),
1775 "cgroup_path() called without proper locking");
1777 if (!dentry
|| cgrp
== dummytop
) {
1779 * Inactive subsystems have no dentry for their root
1786 start
= buf
+ buflen
- 1;
1790 int len
= dentry
->d_name
.len
;
1792 if ((start
-= len
) < buf
)
1793 return -ENAMETOOLONG
;
1794 memcpy(start
, dentry
->d_name
.name
, len
);
1795 cgrp
= cgrp
->parent
;
1799 dentry
= cgrp
->dentry
;
1803 return -ENAMETOOLONG
;
1806 memmove(buf
, start
, buf
+ buflen
- start
);
1809 EXPORT_SYMBOL_GPL(cgroup_path
);
1812 * Control Group taskset
1814 struct task_and_cgroup
{
1815 struct task_struct
*task
;
1816 struct cgroup
*cgrp
;
1820 struct cgroup_taskset
{
1821 struct task_and_cgroup single
;
1822 struct flex_array
*tc_array
;
1825 struct cgroup
*cur_cgrp
;
1829 * cgroup_taskset_first - reset taskset and return the first task
1830 * @tset: taskset of interest
1832 * @tset iteration is initialized and the first task is returned.
1834 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1836 if (tset
->tc_array
) {
1838 return cgroup_taskset_next(tset
);
1840 tset
->cur_cgrp
= tset
->single
.cgrp
;
1841 return tset
->single
.task
;
1844 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1847 * cgroup_taskset_next - iterate to the next task in taskset
1848 * @tset: taskset of interest
1850 * Return the next task in @tset. Iteration must have been initialized
1851 * with cgroup_taskset_first().
1853 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1855 struct task_and_cgroup
*tc
;
1857 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1860 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1861 tset
->cur_cgrp
= tc
->cgrp
;
1864 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1867 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1868 * @tset: taskset of interest
1870 * Return the cgroup for the current (last returned) task of @tset. This
1871 * function must be preceded by either cgroup_taskset_first() or
1872 * cgroup_taskset_next().
1874 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1876 return tset
->cur_cgrp
;
1878 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1881 * cgroup_taskset_size - return the number of tasks in taskset
1882 * @tset: taskset of interest
1884 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1886 return tset
->tc_array
? tset
->tc_array_len
: 1;
1888 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1892 * cgroup_task_migrate - move a task from one cgroup to another.
1894 * Must be called with cgroup_mutex and threadgroup locked.
1896 static void cgroup_task_migrate(struct cgroup
*cgrp
, struct cgroup
*oldcgrp
,
1897 struct task_struct
*tsk
, struct css_set
*newcg
)
1899 struct css_set
*oldcg
;
1902 * We are synchronized through threadgroup_lock() against PF_EXITING
1903 * setting such that we can't race against cgroup_exit() changing the
1904 * css_set to init_css_set and dropping the old one.
1906 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1907 oldcg
= tsk
->cgroups
;
1910 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1913 /* Update the css_set linked lists if we're using them */
1914 write_lock(&css_set_lock
);
1915 if (!list_empty(&tsk
->cg_list
))
1916 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1917 write_unlock(&css_set_lock
);
1920 * We just gained a reference on oldcg by taking it from the task. As
1921 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1922 * it here; it will be freed under RCU.
1924 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1929 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1930 * @cgrp: the cgroup the task is attaching to
1931 * @tsk: the task to be attached
1933 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1936 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1939 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1940 struct cgroup
*oldcgrp
;
1941 struct cgroupfs_root
*root
= cgrp
->root
;
1942 struct cgroup_taskset tset
= { };
1943 struct css_set
*newcg
;
1945 /* @tsk either already exited or can't exit until the end */
1946 if (tsk
->flags
& PF_EXITING
)
1949 /* Nothing to do if the task is already in that cgroup */
1950 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1951 if (cgrp
== oldcgrp
)
1954 tset
.single
.task
= tsk
;
1955 tset
.single
.cgrp
= oldcgrp
;
1957 for_each_subsys(root
, ss
) {
1958 if (ss
->can_attach
) {
1959 retval
= ss
->can_attach(cgrp
, &tset
);
1962 * Remember on which subsystem the can_attach()
1963 * failed, so that we only call cancel_attach()
1964 * against the subsystems whose can_attach()
1965 * succeeded. (See below)
1973 newcg
= find_css_set(tsk
->cgroups
, cgrp
);
1979 cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, newcg
);
1981 for_each_subsys(root
, ss
) {
1983 ss
->attach(cgrp
, &tset
);
1989 for_each_subsys(root
, ss
) {
1990 if (ss
== failed_ss
)
1992 * This subsystem was the one that failed the
1993 * can_attach() check earlier, so we don't need
1994 * to call cancel_attach() against it or any
1995 * remaining subsystems.
1998 if (ss
->cancel_attach
)
1999 ss
->cancel_attach(cgrp
, &tset
);
2006 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2007 * @from: attach to all cgroups of a given task
2008 * @tsk: the task to be attached
2010 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2012 struct cgroupfs_root
*root
;
2016 for_each_active_root(root
) {
2017 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
2019 retval
= cgroup_attach_task(from_cg
, tsk
);
2027 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2030 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2031 * @cgrp: the cgroup to attach to
2032 * @leader: the threadgroup leader task_struct of the group to be attached
2034 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2035 * task_lock of each thread in leader's threadgroup individually in turn.
2037 static int cgroup_attach_proc(struct cgroup
*cgrp
, struct task_struct
*leader
)
2039 int retval
, i
, group_size
;
2040 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
2041 /* guaranteed to be initialized later, but the compiler needs this */
2042 struct cgroupfs_root
*root
= cgrp
->root
;
2043 /* threadgroup list cursor and array */
2044 struct task_struct
*tsk
;
2045 struct task_and_cgroup
*tc
;
2046 struct flex_array
*group
;
2047 struct cgroup_taskset tset
= { };
2050 * step 0: in order to do expensive, possibly blocking operations for
2051 * every thread, we cannot iterate the thread group list, since it needs
2052 * rcu or tasklist locked. instead, build an array of all threads in the
2053 * group - group_rwsem prevents new threads from appearing, and if
2054 * threads exit, this will just be an over-estimate.
2056 group_size
= get_nr_threads(leader
);
2057 /* flex_array supports very large thread-groups better than kmalloc. */
2058 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2061 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2062 retval
= flex_array_prealloc(group
, 0, group_size
- 1, GFP_KERNEL
);
2064 goto out_free_group_list
;
2069 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2070 * already PF_EXITING could be freed from underneath us unless we
2071 * take an rcu_read_lock.
2075 struct task_and_cgroup ent
;
2077 /* @tsk either already exited or can't exit until the end */
2078 if (tsk
->flags
& PF_EXITING
)
2081 /* as per above, nr_threads may decrease, but not increase. */
2082 BUG_ON(i
>= group_size
);
2084 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2085 /* nothing to do if this task is already in the cgroup */
2086 if (ent
.cgrp
== cgrp
)
2089 * saying GFP_ATOMIC has no effect here because we did prealloc
2090 * earlier, but it's good form to communicate our expectations.
2092 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2093 BUG_ON(retval
!= 0);
2095 } while_each_thread(leader
, tsk
);
2097 /* remember the number of threads in the array for later. */
2099 tset
.tc_array
= group
;
2100 tset
.tc_array_len
= group_size
;
2102 /* methods shouldn't be called if no task is actually migrating */
2105 goto out_free_group_list
;
2108 * step 1: check that we can legitimately attach to the cgroup.
2110 for_each_subsys(root
, ss
) {
2111 if (ss
->can_attach
) {
2112 retval
= ss
->can_attach(cgrp
, &tset
);
2115 goto out_cancel_attach
;
2121 * step 2: make sure css_sets exist for all threads to be migrated.
2122 * we use find_css_set, which allocates a new one if necessary.
2124 for (i
= 0; i
< group_size
; i
++) {
2125 tc
= flex_array_get(group
, i
);
2126 tc
->cg
= find_css_set(tc
->task
->cgroups
, cgrp
);
2129 goto out_put_css_set_refs
;
2134 * step 3: now that we're guaranteed success wrt the css_sets,
2135 * proceed to move all tasks to the new cgroup. There are no
2136 * failure cases after here, so this is the commit point.
2138 for (i
= 0; i
< group_size
; i
++) {
2139 tc
= flex_array_get(group
, i
);
2140 cgroup_task_migrate(cgrp
, tc
->cgrp
, tc
->task
, tc
->cg
);
2142 /* nothing is sensitive to fork() after this point. */
2145 * step 4: do subsystem attach callbacks.
2147 for_each_subsys(root
, ss
) {
2149 ss
->attach(cgrp
, &tset
);
2153 * step 5: success! and cleanup
2157 out_put_css_set_refs
:
2159 for (i
= 0; i
< group_size
; i
++) {
2160 tc
= flex_array_get(group
, i
);
2163 put_css_set(tc
->cg
);
2168 for_each_subsys(root
, ss
) {
2169 if (ss
== failed_ss
)
2171 if (ss
->cancel_attach
)
2172 ss
->cancel_attach(cgrp
, &tset
);
2175 out_free_group_list
:
2176 flex_array_free(group
);
2181 * Find the task_struct of the task to attach by vpid and pass it along to the
2182 * function to attach either it or all tasks in its threadgroup. Will lock
2183 * cgroup_mutex and threadgroup; may take task_lock of task.
2185 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2187 struct task_struct
*tsk
;
2188 const struct cred
*cred
= current_cred(), *tcred
;
2191 if (!cgroup_lock_live_group(cgrp
))
2197 tsk
= find_task_by_vpid(pid
);
2201 goto out_unlock_cgroup
;
2204 * even if we're attaching all tasks in the thread group, we
2205 * only need to check permissions on one of them.
2207 tcred
= __task_cred(tsk
);
2208 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2209 !uid_eq(cred
->euid
, tcred
->uid
) &&
2210 !uid_eq(cred
->euid
, tcred
->suid
)) {
2213 goto out_unlock_cgroup
;
2219 tsk
= tsk
->group_leader
;
2222 * Workqueue threads may acquire PF_THREAD_BOUND and become
2223 * trapped in a cpuset, or RT worker may be born in a cgroup
2224 * with no rt_runtime allocated. Just say no.
2226 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_THREAD_BOUND
)) {
2229 goto out_unlock_cgroup
;
2232 get_task_struct(tsk
);
2235 threadgroup_lock(tsk
);
2237 if (!thread_group_leader(tsk
)) {
2239 * a race with de_thread from another thread's exec()
2240 * may strip us of our leadership, if this happens,
2241 * there is no choice but to throw this task away and
2242 * try again; this is
2243 * "double-double-toil-and-trouble-check locking".
2245 threadgroup_unlock(tsk
);
2246 put_task_struct(tsk
);
2247 goto retry_find_task
;
2249 ret
= cgroup_attach_proc(cgrp
, tsk
);
2251 ret
= cgroup_attach_task(cgrp
, tsk
);
2252 threadgroup_unlock(tsk
);
2254 put_task_struct(tsk
);
2260 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2262 return attach_task_by_pid(cgrp
, pid
, false);
2265 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2267 return attach_task_by_pid(cgrp
, tgid
, true);
2271 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2272 * @cgrp: the cgroup to be checked for liveness
2274 * On success, returns true; the lock should be later released with
2275 * cgroup_unlock(). On failure returns false with no lock held.
2277 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
2279 mutex_lock(&cgroup_mutex
);
2280 if (cgroup_is_removed(cgrp
)) {
2281 mutex_unlock(&cgroup_mutex
);
2286 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
2288 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2291 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2292 if (strlen(buffer
) >= PATH_MAX
)
2294 if (!cgroup_lock_live_group(cgrp
))
2296 mutex_lock(&cgroup_root_mutex
);
2297 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2298 mutex_unlock(&cgroup_root_mutex
);
2303 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2304 struct seq_file
*seq
)
2306 if (!cgroup_lock_live_group(cgrp
))
2308 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2309 seq_putc(seq
, '\n');
2314 /* A buffer size big enough for numbers or short strings */
2315 #define CGROUP_LOCAL_BUFFER_SIZE 64
2317 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2319 const char __user
*userbuf
,
2320 size_t nbytes
, loff_t
*unused_ppos
)
2322 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2328 if (nbytes
>= sizeof(buffer
))
2330 if (copy_from_user(buffer
, userbuf
, nbytes
))
2333 buffer
[nbytes
] = 0; /* nul-terminate */
2334 if (cft
->write_u64
) {
2335 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2338 retval
= cft
->write_u64(cgrp
, cft
, val
);
2340 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2343 retval
= cft
->write_s64(cgrp
, cft
, val
);
2350 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2352 const char __user
*userbuf
,
2353 size_t nbytes
, loff_t
*unused_ppos
)
2355 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2357 size_t max_bytes
= cft
->max_write_len
;
2358 char *buffer
= local_buffer
;
2361 max_bytes
= sizeof(local_buffer
) - 1;
2362 if (nbytes
>= max_bytes
)
2364 /* Allocate a dynamic buffer if we need one */
2365 if (nbytes
>= sizeof(local_buffer
)) {
2366 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2370 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2375 buffer
[nbytes
] = 0; /* nul-terminate */
2376 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2380 if (buffer
!= local_buffer
)
2385 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2386 size_t nbytes
, loff_t
*ppos
)
2388 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2389 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2391 if (cgroup_is_removed(cgrp
))
2394 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2395 if (cft
->write_u64
|| cft
->write_s64
)
2396 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2397 if (cft
->write_string
)
2398 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2400 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2401 return ret
? ret
: nbytes
;
2406 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2408 char __user
*buf
, size_t nbytes
,
2411 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2412 u64 val
= cft
->read_u64(cgrp
, cft
);
2413 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2415 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2418 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2420 char __user
*buf
, size_t nbytes
,
2423 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2424 s64 val
= cft
->read_s64(cgrp
, cft
);
2425 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2427 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2430 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2431 size_t nbytes
, loff_t
*ppos
)
2433 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2434 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2436 if (cgroup_is_removed(cgrp
))
2440 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2442 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2444 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2449 * seqfile ops/methods for returning structured data. Currently just
2450 * supports string->u64 maps, but can be extended in future.
2453 struct cgroup_seqfile_state
{
2455 struct cgroup
*cgroup
;
2458 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2460 struct seq_file
*sf
= cb
->state
;
2461 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2464 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2466 struct cgroup_seqfile_state
*state
= m
->private;
2467 struct cftype
*cft
= state
->cft
;
2468 if (cft
->read_map
) {
2469 struct cgroup_map_cb cb
= {
2470 .fill
= cgroup_map_add
,
2473 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2475 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2478 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2480 struct seq_file
*seq
= file
->private_data
;
2481 kfree(seq
->private);
2482 return single_release(inode
, file
);
2485 static const struct file_operations cgroup_seqfile_operations
= {
2487 .write
= cgroup_file_write
,
2488 .llseek
= seq_lseek
,
2489 .release
= cgroup_seqfile_release
,
2492 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2497 err
= generic_file_open(inode
, file
);
2500 cft
= __d_cft(file
->f_dentry
);
2502 if (cft
->read_map
|| cft
->read_seq_string
) {
2503 struct cgroup_seqfile_state
*state
=
2504 kzalloc(sizeof(*state
), GFP_USER
);
2508 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2509 file
->f_op
= &cgroup_seqfile_operations
;
2510 err
= single_open(file
, cgroup_seqfile_show
, state
);
2513 } else if (cft
->open
)
2514 err
= cft
->open(inode
, file
);
2521 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2523 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2525 return cft
->release(inode
, file
);
2530 * cgroup_rename - Only allow simple rename of directories in place.
2532 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2533 struct inode
*new_dir
, struct dentry
*new_dentry
)
2535 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2537 if (new_dentry
->d_inode
)
2539 if (old_dir
!= new_dir
)
2541 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2544 static struct simple_xattrs
*__d_xattrs(struct dentry
*dentry
)
2546 if (S_ISDIR(dentry
->d_inode
->i_mode
))
2547 return &__d_cgrp(dentry
)->xattrs
;
2549 return &__d_cft(dentry
)->xattrs
;
2552 static inline int xattr_enabled(struct dentry
*dentry
)
2554 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
2555 return test_bit(ROOT_XATTR
, &root
->flags
);
2558 static bool is_valid_xattr(const char *name
)
2560 if (!strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) ||
2561 !strncmp(name
, XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
))
2566 static int cgroup_setxattr(struct dentry
*dentry
, const char *name
,
2567 const void *val
, size_t size
, int flags
)
2569 if (!xattr_enabled(dentry
))
2571 if (!is_valid_xattr(name
))
2573 return simple_xattr_set(__d_xattrs(dentry
), name
, val
, size
, flags
);
2576 static int cgroup_removexattr(struct dentry
*dentry
, const char *name
)
2578 if (!xattr_enabled(dentry
))
2580 if (!is_valid_xattr(name
))
2582 return simple_xattr_remove(__d_xattrs(dentry
), name
);
2585 static ssize_t
cgroup_getxattr(struct dentry
*dentry
, const char *name
,
2586 void *buf
, size_t size
)
2588 if (!xattr_enabled(dentry
))
2590 if (!is_valid_xattr(name
))
2592 return simple_xattr_get(__d_xattrs(dentry
), name
, buf
, size
);
2595 static ssize_t
cgroup_listxattr(struct dentry
*dentry
, char *buf
, size_t size
)
2597 if (!xattr_enabled(dentry
))
2599 return simple_xattr_list(__d_xattrs(dentry
), buf
, size
);
2602 static const struct file_operations cgroup_file_operations
= {
2603 .read
= cgroup_file_read
,
2604 .write
= cgroup_file_write
,
2605 .llseek
= generic_file_llseek
,
2606 .open
= cgroup_file_open
,
2607 .release
= cgroup_file_release
,
2610 static const struct inode_operations cgroup_file_inode_operations
= {
2611 .setxattr
= cgroup_setxattr
,
2612 .getxattr
= cgroup_getxattr
,
2613 .listxattr
= cgroup_listxattr
,
2614 .removexattr
= cgroup_removexattr
,
2617 static const struct inode_operations cgroup_dir_inode_operations
= {
2618 .lookup
= cgroup_lookup
,
2619 .mkdir
= cgroup_mkdir
,
2620 .rmdir
= cgroup_rmdir
,
2621 .rename
= cgroup_rename
,
2622 .setxattr
= cgroup_setxattr
,
2623 .getxattr
= cgroup_getxattr
,
2624 .listxattr
= cgroup_listxattr
,
2625 .removexattr
= cgroup_removexattr
,
2628 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
2630 if (dentry
->d_name
.len
> NAME_MAX
)
2631 return ERR_PTR(-ENAMETOOLONG
);
2632 d_add(dentry
, NULL
);
2637 * Check if a file is a control file
2639 static inline struct cftype
*__file_cft(struct file
*file
)
2641 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2642 return ERR_PTR(-EINVAL
);
2643 return __d_cft(file
->f_dentry
);
2646 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2647 struct super_block
*sb
)
2649 struct inode
*inode
;
2653 if (dentry
->d_inode
)
2656 inode
= cgroup_new_inode(mode
, sb
);
2660 if (S_ISDIR(mode
)) {
2661 inode
->i_op
= &cgroup_dir_inode_operations
;
2662 inode
->i_fop
= &simple_dir_operations
;
2664 /* start off with i_nlink == 2 (for "." entry) */
2666 inc_nlink(dentry
->d_parent
->d_inode
);
2669 * Control reaches here with cgroup_mutex held.
2670 * @inode->i_mutex should nest outside cgroup_mutex but we
2671 * want to populate it immediately without releasing
2672 * cgroup_mutex. As @inode isn't visible to anyone else
2673 * yet, trylock will always succeed without affecting
2676 WARN_ON_ONCE(!mutex_trylock(&inode
->i_mutex
));
2677 } else if (S_ISREG(mode
)) {
2679 inode
->i_fop
= &cgroup_file_operations
;
2680 inode
->i_op
= &cgroup_file_inode_operations
;
2682 d_instantiate(dentry
, inode
);
2683 dget(dentry
); /* Extra count - pin the dentry in core */
2688 * cgroup_file_mode - deduce file mode of a control file
2689 * @cft: the control file in question
2691 * returns cft->mode if ->mode is not 0
2692 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2693 * returns S_IRUGO if it has only a read handler
2694 * returns S_IWUSR if it has only a write hander
2696 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2703 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2704 cft
->read_map
|| cft
->read_seq_string
)
2707 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2708 cft
->write_string
|| cft
->trigger
)
2714 static int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2717 struct dentry
*dir
= cgrp
->dentry
;
2718 struct cgroup
*parent
= __d_cgrp(dir
);
2719 struct dentry
*dentry
;
2723 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2725 simple_xattrs_init(&cft
->xattrs
);
2727 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2728 strcpy(name
, subsys
->name
);
2731 strcat(name
, cft
->name
);
2733 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2735 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2739 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2740 if (IS_ERR(dentry
)) {
2741 error
= PTR_ERR(dentry
);
2745 mode
= cgroup_file_mode(cft
);
2746 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2748 cfe
->type
= (void *)cft
;
2749 cfe
->dentry
= dentry
;
2750 dentry
->d_fsdata
= cfe
;
2751 list_add_tail(&cfe
->node
, &parent
->files
);
2760 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2761 struct cftype cfts
[], bool is_add
)
2766 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2767 /* does cft->flags tell us to skip this file on @cgrp? */
2768 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2770 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2774 err
= cgroup_add_file(cgrp
, subsys
, cft
);
2776 err
= cgroup_rm_file(cgrp
, cft
);
2778 pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2779 is_add
? "add" : "remove", cft
->name
, err
);
2786 static DEFINE_MUTEX(cgroup_cft_mutex
);
2788 static void cgroup_cfts_prepare(void)
2789 __acquires(&cgroup_cft_mutex
) __acquires(&cgroup_mutex
)
2792 * Thanks to the entanglement with vfs inode locking, we can't walk
2793 * the existing cgroups under cgroup_mutex and create files.
2794 * Instead, we increment reference on all cgroups and build list of
2795 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2796 * exclusive access to the field.
2798 mutex_lock(&cgroup_cft_mutex
);
2799 mutex_lock(&cgroup_mutex
);
2802 static void cgroup_cfts_commit(struct cgroup_subsys
*ss
,
2803 struct cftype
*cfts
, bool is_add
)
2804 __releases(&cgroup_mutex
) __releases(&cgroup_cft_mutex
)
2807 struct cgroup
*cgrp
, *n
;
2809 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2810 if (cfts
&& ss
->root
!= &rootnode
) {
2811 list_for_each_entry(cgrp
, &ss
->root
->allcg_list
, allcg_node
) {
2813 list_add_tail(&cgrp
->cft_q_node
, &pending
);
2817 mutex_unlock(&cgroup_mutex
);
2820 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2821 * files for all cgroups which were created before.
2823 list_for_each_entry_safe(cgrp
, n
, &pending
, cft_q_node
) {
2824 struct inode
*inode
= cgrp
->dentry
->d_inode
;
2826 mutex_lock(&inode
->i_mutex
);
2827 mutex_lock(&cgroup_mutex
);
2828 if (!cgroup_is_removed(cgrp
))
2829 cgroup_addrm_files(cgrp
, ss
, cfts
, is_add
);
2830 mutex_unlock(&cgroup_mutex
);
2831 mutex_unlock(&inode
->i_mutex
);
2833 list_del_init(&cgrp
->cft_q_node
);
2837 mutex_unlock(&cgroup_cft_mutex
);
2841 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2842 * @ss: target cgroup subsystem
2843 * @cfts: zero-length name terminated array of cftypes
2845 * Register @cfts to @ss. Files described by @cfts are created for all
2846 * existing cgroups to which @ss is attached and all future cgroups will
2847 * have them too. This function can be called anytime whether @ss is
2850 * Returns 0 on successful registration, -errno on failure. Note that this
2851 * function currently returns 0 as long as @cfts registration is successful
2852 * even if some file creation attempts on existing cgroups fail.
2854 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2856 struct cftype_set
*set
;
2858 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2862 cgroup_cfts_prepare();
2864 list_add_tail(&set
->node
, &ss
->cftsets
);
2865 cgroup_cfts_commit(ss
, cfts
, true);
2869 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2872 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2873 * @ss: target cgroup subsystem
2874 * @cfts: zero-length name terminated array of cftypes
2876 * Unregister @cfts from @ss. Files described by @cfts are removed from
2877 * all existing cgroups to which @ss is attached and all future cgroups
2878 * won't have them either. This function can be called anytime whether @ss
2879 * is attached or not.
2881 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2882 * registered with @ss.
2884 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2886 struct cftype_set
*set
;
2888 cgroup_cfts_prepare();
2890 list_for_each_entry(set
, &ss
->cftsets
, node
) {
2891 if (set
->cfts
== cfts
) {
2892 list_del_init(&set
->node
);
2893 cgroup_cfts_commit(ss
, cfts
, false);
2898 cgroup_cfts_commit(ss
, NULL
, false);
2903 * cgroup_task_count - count the number of tasks in a cgroup.
2904 * @cgrp: the cgroup in question
2906 * Return the number of tasks in the cgroup.
2908 int cgroup_task_count(const struct cgroup
*cgrp
)
2911 struct cg_cgroup_link
*link
;
2913 read_lock(&css_set_lock
);
2914 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2915 count
+= atomic_read(&link
->cg
->refcount
);
2917 read_unlock(&css_set_lock
);
2922 * Advance a list_head iterator. The iterator should be positioned at
2923 * the start of a css_set
2925 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2926 struct cgroup_iter
*it
)
2928 struct list_head
*l
= it
->cg_link
;
2929 struct cg_cgroup_link
*link
;
2932 /* Advance to the next non-empty css_set */
2935 if (l
== &cgrp
->css_sets
) {
2939 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2941 } while (list_empty(&cg
->tasks
));
2943 it
->task
= cg
->tasks
.next
;
2947 * To reduce the fork() overhead for systems that are not actually
2948 * using their cgroups capability, we don't maintain the lists running
2949 * through each css_set to its tasks until we see the list actually
2950 * used - in other words after the first call to cgroup_iter_start().
2952 static void cgroup_enable_task_cg_lists(void)
2954 struct task_struct
*p
, *g
;
2955 write_lock(&css_set_lock
);
2956 use_task_css_set_links
= 1;
2958 * We need tasklist_lock because RCU is not safe against
2959 * while_each_thread(). Besides, a forking task that has passed
2960 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2961 * is not guaranteed to have its child immediately visible in the
2962 * tasklist if we walk through it with RCU.
2964 read_lock(&tasklist_lock
);
2965 do_each_thread(g
, p
) {
2968 * We should check if the process is exiting, otherwise
2969 * it will race with cgroup_exit() in that the list
2970 * entry won't be deleted though the process has exited.
2972 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2973 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2975 } while_each_thread(g
, p
);
2976 read_unlock(&tasklist_lock
);
2977 write_unlock(&css_set_lock
);
2981 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2982 * @pos: the current position (%NULL to initiate traversal)
2983 * @cgroup: cgroup whose descendants to walk
2985 * To be used by cgroup_for_each_descendant_pre(). Find the next
2986 * descendant to visit for pre-order traversal of @cgroup's descendants.
2988 struct cgroup
*cgroup_next_descendant_pre(struct cgroup
*pos
,
2989 struct cgroup
*cgroup
)
2991 struct cgroup
*next
;
2993 WARN_ON_ONCE(!rcu_read_lock_held());
2995 /* if first iteration, pretend we just visited @cgroup */
2997 if (list_empty(&cgroup
->children
))
3002 /* visit the first child if exists */
3003 next
= list_first_or_null_rcu(&pos
->children
, struct cgroup
, sibling
);
3007 /* no child, visit my or the closest ancestor's next sibling */
3009 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
,
3011 if (&next
->sibling
!= &pos
->parent
->children
)
3015 } while (pos
!= cgroup
);
3019 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre
);
3021 static struct cgroup
*cgroup_leftmost_descendant(struct cgroup
*pos
)
3023 struct cgroup
*last
;
3027 pos
= list_first_or_null_rcu(&pos
->children
, struct cgroup
,
3035 * cgroup_next_descendant_post - find the next descendant for post-order walk
3036 * @pos: the current position (%NULL to initiate traversal)
3037 * @cgroup: cgroup whose descendants to walk
3039 * To be used by cgroup_for_each_descendant_post(). Find the next
3040 * descendant to visit for post-order traversal of @cgroup's descendants.
3042 struct cgroup
*cgroup_next_descendant_post(struct cgroup
*pos
,
3043 struct cgroup
*cgroup
)
3045 struct cgroup
*next
;
3047 WARN_ON_ONCE(!rcu_read_lock_held());
3049 /* if first iteration, visit the leftmost descendant */
3051 next
= cgroup_leftmost_descendant(cgroup
);
3052 return next
!= cgroup
? next
: NULL
;
3055 /* if there's an unvisited sibling, visit its leftmost descendant */
3056 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
, sibling
);
3057 if (&next
->sibling
!= &pos
->parent
->children
)
3058 return cgroup_leftmost_descendant(next
);
3060 /* no sibling left, visit parent */
3062 return next
!= cgroup
? next
: NULL
;
3064 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post
);
3066 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3067 __acquires(css_set_lock
)
3070 * The first time anyone tries to iterate across a cgroup,
3071 * we need to enable the list linking each css_set to its
3072 * tasks, and fix up all existing tasks.
3074 if (!use_task_css_set_links
)
3075 cgroup_enable_task_cg_lists();
3077 read_lock(&css_set_lock
);
3078 it
->cg_link
= &cgrp
->css_sets
;
3079 cgroup_advance_iter(cgrp
, it
);
3082 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
3083 struct cgroup_iter
*it
)
3085 struct task_struct
*res
;
3086 struct list_head
*l
= it
->task
;
3087 struct cg_cgroup_link
*link
;
3089 /* If the iterator cg is NULL, we have no tasks */
3092 res
= list_entry(l
, struct task_struct
, cg_list
);
3093 /* Advance iterator to find next entry */
3095 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
3096 if (l
== &link
->cg
->tasks
) {
3097 /* We reached the end of this task list - move on to
3098 * the next cg_cgroup_link */
3099 cgroup_advance_iter(cgrp
, it
);
3106 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3107 __releases(css_set_lock
)
3109 read_unlock(&css_set_lock
);
3112 static inline int started_after_time(struct task_struct
*t1
,
3113 struct timespec
*time
,
3114 struct task_struct
*t2
)
3116 int start_diff
= timespec_compare(&t1
->start_time
, time
);
3117 if (start_diff
> 0) {
3119 } else if (start_diff
< 0) {
3123 * Arbitrarily, if two processes started at the same
3124 * time, we'll say that the lower pointer value
3125 * started first. Note that t2 may have exited by now
3126 * so this may not be a valid pointer any longer, but
3127 * that's fine - it still serves to distinguish
3128 * between two tasks started (effectively) simultaneously.
3135 * This function is a callback from heap_insert() and is used to order
3137 * In this case we order the heap in descending task start time.
3139 static inline int started_after(void *p1
, void *p2
)
3141 struct task_struct
*t1
= p1
;
3142 struct task_struct
*t2
= p2
;
3143 return started_after_time(t1
, &t2
->start_time
, t2
);
3147 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3148 * @scan: struct cgroup_scanner containing arguments for the scan
3150 * Arguments include pointers to callback functions test_task() and
3152 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3153 * and if it returns true, call process_task() for it also.
3154 * The test_task pointer may be NULL, meaning always true (select all tasks).
3155 * Effectively duplicates cgroup_iter_{start,next,end}()
3156 * but does not lock css_set_lock for the call to process_task().
3157 * The struct cgroup_scanner may be embedded in any structure of the caller's
3159 * It is guaranteed that process_task() will act on every task that
3160 * is a member of the cgroup for the duration of this call. This
3161 * function may or may not call process_task() for tasks that exit
3162 * or move to a different cgroup during the call, or are forked or
3163 * move into the cgroup during the call.
3165 * Note that test_task() may be called with locks held, and may in some
3166 * situations be called multiple times for the same task, so it should
3168 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3169 * pre-allocated and will be used for heap operations (and its "gt" member will
3170 * be overwritten), else a temporary heap will be used (allocation of which
3171 * may cause this function to fail).
3173 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
3176 struct cgroup_iter it
;
3177 struct task_struct
*p
, *dropped
;
3178 /* Never dereference latest_task, since it's not refcounted */
3179 struct task_struct
*latest_task
= NULL
;
3180 struct ptr_heap tmp_heap
;
3181 struct ptr_heap
*heap
;
3182 struct timespec latest_time
= { 0, 0 };
3185 /* The caller supplied our heap and pre-allocated its memory */
3187 heap
->gt
= &started_after
;
3189 /* We need to allocate our own heap memory */
3191 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3193 /* cannot allocate the heap */
3199 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3200 * to determine which are of interest, and using the scanner's
3201 * "process_task" callback to process any of them that need an update.
3202 * Since we don't want to hold any locks during the task updates,
3203 * gather tasks to be processed in a heap structure.
3204 * The heap is sorted by descending task start time.
3205 * If the statically-sized heap fills up, we overflow tasks that
3206 * started later, and in future iterations only consider tasks that
3207 * started after the latest task in the previous pass. This
3208 * guarantees forward progress and that we don't miss any tasks.
3211 cgroup_iter_start(scan
->cg
, &it
);
3212 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
3214 * Only affect tasks that qualify per the caller's callback,
3215 * if he provided one
3217 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
3220 * Only process tasks that started after the last task
3223 if (!started_after_time(p
, &latest_time
, latest_task
))
3225 dropped
= heap_insert(heap
, p
);
3226 if (dropped
== NULL
) {
3228 * The new task was inserted; the heap wasn't
3232 } else if (dropped
!= p
) {
3234 * The new task was inserted, and pushed out a
3238 put_task_struct(dropped
);
3241 * Else the new task was newer than anything already in
3242 * the heap and wasn't inserted
3245 cgroup_iter_end(scan
->cg
, &it
);
3248 for (i
= 0; i
< heap
->size
; i
++) {
3249 struct task_struct
*q
= heap
->ptrs
[i
];
3251 latest_time
= q
->start_time
;
3254 /* Process the task per the caller's callback */
3255 scan
->process_task(q
, scan
);
3259 * If we had to process any tasks at all, scan again
3260 * in case some of them were in the middle of forking
3261 * children that didn't get processed.
3262 * Not the most efficient way to do it, but it avoids
3263 * having to take callback_mutex in the fork path
3267 if (heap
== &tmp_heap
)
3268 heap_free(&tmp_heap
);
3273 * Stuff for reading the 'tasks'/'procs' files.
3275 * Reading this file can return large amounts of data if a cgroup has
3276 * *lots* of attached tasks. So it may need several calls to read(),
3277 * but we cannot guarantee that the information we produce is correct
3278 * unless we produce it entirely atomically.
3282 /* which pidlist file are we talking about? */
3283 enum cgroup_filetype
{
3289 * A pidlist is a list of pids that virtually represents the contents of one
3290 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3291 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3294 struct cgroup_pidlist
{
3296 * used to find which pidlist is wanted. doesn't change as long as
3297 * this particular list stays in the list.
3299 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3302 /* how many elements the above list has */
3304 /* how many files are using the current array */
3306 /* each of these stored in a list by its cgroup */
3307 struct list_head links
;
3308 /* pointer to the cgroup we belong to, for list removal purposes */
3309 struct cgroup
*owner
;
3310 /* protects the other fields */
3311 struct rw_semaphore mutex
;
3315 * The following two functions "fix" the issue where there are more pids
3316 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3317 * TODO: replace with a kernel-wide solution to this problem
3319 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3320 static void *pidlist_allocate(int count
)
3322 if (PIDLIST_TOO_LARGE(count
))
3323 return vmalloc(count
* sizeof(pid_t
));
3325 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3327 static void pidlist_free(void *p
)
3329 if (is_vmalloc_addr(p
))
3334 static void *pidlist_resize(void *p
, int newcount
)
3337 /* note: if new alloc fails, old p will still be valid either way */
3338 if (is_vmalloc_addr(p
)) {
3339 newlist
= vmalloc(newcount
* sizeof(pid_t
));
3342 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
3345 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
3351 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3352 * If the new stripped list is sufficiently smaller and there's enough memory
3353 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3354 * number of unique elements.
3356 /* is the size difference enough that we should re-allocate the array? */
3357 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3358 static int pidlist_uniq(pid_t
**p
, int length
)
3365 * we presume the 0th element is unique, so i starts at 1. trivial
3366 * edge cases first; no work needs to be done for either
3368 if (length
== 0 || length
== 1)
3370 /* src and dest walk down the list; dest counts unique elements */
3371 for (src
= 1; src
< length
; src
++) {
3372 /* find next unique element */
3373 while (list
[src
] == list
[src
-1]) {
3378 /* dest always points to where the next unique element goes */
3379 list
[dest
] = list
[src
];
3384 * if the length difference is large enough, we want to allocate a
3385 * smaller buffer to save memory. if this fails due to out of memory,
3386 * we'll just stay with what we've got.
3388 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
3389 newlist
= pidlist_resize(list
, dest
);
3396 static int cmppid(const void *a
, const void *b
)
3398 return *(pid_t
*)a
- *(pid_t
*)b
;
3402 * find the appropriate pidlist for our purpose (given procs vs tasks)
3403 * returns with the lock on that pidlist already held, and takes care
3404 * of the use count, or returns NULL with no locks held if we're out of
3407 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3408 enum cgroup_filetype type
)
3410 struct cgroup_pidlist
*l
;
3411 /* don't need task_nsproxy() if we're looking at ourself */
3412 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
3415 * We can't drop the pidlist_mutex before taking the l->mutex in case
3416 * the last ref-holder is trying to remove l from the list at the same
3417 * time. Holding the pidlist_mutex precludes somebody taking whichever
3418 * list we find out from under us - compare release_pid_array().
3420 mutex_lock(&cgrp
->pidlist_mutex
);
3421 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3422 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3423 /* make sure l doesn't vanish out from under us */
3424 down_write(&l
->mutex
);
3425 mutex_unlock(&cgrp
->pidlist_mutex
);
3429 /* entry not found; create a new one */
3430 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3432 mutex_unlock(&cgrp
->pidlist_mutex
);
3435 init_rwsem(&l
->mutex
);
3436 down_write(&l
->mutex
);
3438 l
->key
.ns
= get_pid_ns(ns
);
3439 l
->use_count
= 0; /* don't increment here */
3442 list_add(&l
->links
, &cgrp
->pidlists
);
3443 mutex_unlock(&cgrp
->pidlist_mutex
);
3448 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3450 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3451 struct cgroup_pidlist
**lp
)
3455 int pid
, n
= 0; /* used for populating the array */
3456 struct cgroup_iter it
;
3457 struct task_struct
*tsk
;
3458 struct cgroup_pidlist
*l
;
3461 * If cgroup gets more users after we read count, we won't have
3462 * enough space - tough. This race is indistinguishable to the
3463 * caller from the case that the additional cgroup users didn't
3464 * show up until sometime later on.
3466 length
= cgroup_task_count(cgrp
);
3467 array
= pidlist_allocate(length
);
3470 /* now, populate the array */
3471 cgroup_iter_start(cgrp
, &it
);
3472 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3473 if (unlikely(n
== length
))
3475 /* get tgid or pid for procs or tasks file respectively */
3476 if (type
== CGROUP_FILE_PROCS
)
3477 pid
= task_tgid_vnr(tsk
);
3479 pid
= task_pid_vnr(tsk
);
3480 if (pid
> 0) /* make sure to only use valid results */
3483 cgroup_iter_end(cgrp
, &it
);
3485 /* now sort & (if procs) strip out duplicates */
3486 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3487 if (type
== CGROUP_FILE_PROCS
)
3488 length
= pidlist_uniq(&array
, length
);
3489 l
= cgroup_pidlist_find(cgrp
, type
);
3491 pidlist_free(array
);
3494 /* store array, freeing old if necessary - lock already held */
3495 pidlist_free(l
->list
);
3499 up_write(&l
->mutex
);
3505 * cgroupstats_build - build and fill cgroupstats
3506 * @stats: cgroupstats to fill information into
3507 * @dentry: A dentry entry belonging to the cgroup for which stats have
3510 * Build and fill cgroupstats so that taskstats can export it to user
3513 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3516 struct cgroup
*cgrp
;
3517 struct cgroup_iter it
;
3518 struct task_struct
*tsk
;
3521 * Validate dentry by checking the superblock operations,
3522 * and make sure it's a directory.
3524 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3525 !S_ISDIR(dentry
->d_inode
->i_mode
))
3529 cgrp
= dentry
->d_fsdata
;
3531 cgroup_iter_start(cgrp
, &it
);
3532 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3533 switch (tsk
->state
) {
3535 stats
->nr_running
++;
3537 case TASK_INTERRUPTIBLE
:
3538 stats
->nr_sleeping
++;
3540 case TASK_UNINTERRUPTIBLE
:
3541 stats
->nr_uninterruptible
++;
3544 stats
->nr_stopped
++;
3547 if (delayacct_is_task_waiting_on_io(tsk
))
3548 stats
->nr_io_wait
++;
3552 cgroup_iter_end(cgrp
, &it
);
3560 * seq_file methods for the tasks/procs files. The seq_file position is the
3561 * next pid to display; the seq_file iterator is a pointer to the pid
3562 * in the cgroup->l->list array.
3565 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3568 * Initially we receive a position value that corresponds to
3569 * one more than the last pid shown (or 0 on the first call or
3570 * after a seek to the start). Use a binary-search to find the
3571 * next pid to display, if any
3573 struct cgroup_pidlist
*l
= s
->private;
3574 int index
= 0, pid
= *pos
;
3577 down_read(&l
->mutex
);
3579 int end
= l
->length
;
3581 while (index
< end
) {
3582 int mid
= (index
+ end
) / 2;
3583 if (l
->list
[mid
] == pid
) {
3586 } else if (l
->list
[mid
] <= pid
)
3592 /* If we're off the end of the array, we're done */
3593 if (index
>= l
->length
)
3595 /* Update the abstract position to be the actual pid that we found */
3596 iter
= l
->list
+ index
;
3601 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3603 struct cgroup_pidlist
*l
= s
->private;
3607 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3609 struct cgroup_pidlist
*l
= s
->private;
3611 pid_t
*end
= l
->list
+ l
->length
;
3613 * Advance to the next pid in the array. If this goes off the
3625 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3627 return seq_printf(s
, "%d\n", *(int *)v
);
3631 * seq_operations functions for iterating on pidlists through seq_file -
3632 * independent of whether it's tasks or procs
3634 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3635 .start
= cgroup_pidlist_start
,
3636 .stop
= cgroup_pidlist_stop
,
3637 .next
= cgroup_pidlist_next
,
3638 .show
= cgroup_pidlist_show
,
3641 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3644 * the case where we're the last user of this particular pidlist will
3645 * have us remove it from the cgroup's list, which entails taking the
3646 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3647 * pidlist_mutex, we have to take pidlist_mutex first.
3649 mutex_lock(&l
->owner
->pidlist_mutex
);
3650 down_write(&l
->mutex
);
3651 BUG_ON(!l
->use_count
);
3652 if (!--l
->use_count
) {
3653 /* we're the last user if refcount is 0; remove and free */
3654 list_del(&l
->links
);
3655 mutex_unlock(&l
->owner
->pidlist_mutex
);
3656 pidlist_free(l
->list
);
3657 put_pid_ns(l
->key
.ns
);
3658 up_write(&l
->mutex
);
3662 mutex_unlock(&l
->owner
->pidlist_mutex
);
3663 up_write(&l
->mutex
);
3666 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3668 struct cgroup_pidlist
*l
;
3669 if (!(file
->f_mode
& FMODE_READ
))
3672 * the seq_file will only be initialized if the file was opened for
3673 * reading; hence we check if it's not null only in that case.
3675 l
= ((struct seq_file
*)file
->private_data
)->private;
3676 cgroup_release_pid_array(l
);
3677 return seq_release(inode
, file
);
3680 static const struct file_operations cgroup_pidlist_operations
= {
3682 .llseek
= seq_lseek
,
3683 .write
= cgroup_file_write
,
3684 .release
= cgroup_pidlist_release
,
3688 * The following functions handle opens on a file that displays a pidlist
3689 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3692 /* helper function for the two below it */
3693 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3695 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3696 struct cgroup_pidlist
*l
;
3699 /* Nothing to do for write-only files */
3700 if (!(file
->f_mode
& FMODE_READ
))
3703 /* have the array populated */
3704 retval
= pidlist_array_load(cgrp
, type
, &l
);
3707 /* configure file information */
3708 file
->f_op
= &cgroup_pidlist_operations
;
3710 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3712 cgroup_release_pid_array(l
);
3715 ((struct seq_file
*)file
->private_data
)->private = l
;
3718 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3720 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3722 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3724 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3727 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3730 return notify_on_release(cgrp
);
3733 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3737 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3739 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3741 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3746 * Unregister event and free resources.
3748 * Gets called from workqueue.
3750 static void cgroup_event_remove(struct work_struct
*work
)
3752 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3754 struct cgroup
*cgrp
= event
->cgrp
;
3756 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3758 eventfd_ctx_put(event
->eventfd
);
3764 * Gets called on POLLHUP on eventfd when user closes it.
3766 * Called with wqh->lock held and interrupts disabled.
3768 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3769 int sync
, void *key
)
3771 struct cgroup_event
*event
= container_of(wait
,
3772 struct cgroup_event
, wait
);
3773 struct cgroup
*cgrp
= event
->cgrp
;
3774 unsigned long flags
= (unsigned long)key
;
3776 if (flags
& POLLHUP
) {
3777 __remove_wait_queue(event
->wqh
, &event
->wait
);
3778 spin_lock(&cgrp
->event_list_lock
);
3779 list_del_init(&event
->list
);
3780 spin_unlock(&cgrp
->event_list_lock
);
3782 * We are in atomic context, but cgroup_event_remove() may
3783 * sleep, so we have to call it in workqueue.
3785 schedule_work(&event
->remove
);
3791 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3792 wait_queue_head_t
*wqh
, poll_table
*pt
)
3794 struct cgroup_event
*event
= container_of(pt
,
3795 struct cgroup_event
, pt
);
3798 add_wait_queue(wqh
, &event
->wait
);
3802 * Parse input and register new cgroup event handler.
3804 * Input must be in format '<event_fd> <control_fd> <args>'.
3805 * Interpretation of args is defined by control file implementation.
3807 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3810 struct cgroup_event
*event
= NULL
;
3811 unsigned int efd
, cfd
;
3812 struct file
*efile
= NULL
;
3813 struct file
*cfile
= NULL
;
3817 efd
= simple_strtoul(buffer
, &endp
, 10);
3822 cfd
= simple_strtoul(buffer
, &endp
, 10);
3823 if ((*endp
!= ' ') && (*endp
!= '\0'))
3827 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3831 INIT_LIST_HEAD(&event
->list
);
3832 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3833 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3834 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3836 efile
= eventfd_fget(efd
);
3837 if (IS_ERR(efile
)) {
3838 ret
= PTR_ERR(efile
);
3842 event
->eventfd
= eventfd_ctx_fileget(efile
);
3843 if (IS_ERR(event
->eventfd
)) {
3844 ret
= PTR_ERR(event
->eventfd
);
3854 /* the process need read permission on control file */
3855 /* AV: shouldn't we check that it's been opened for read instead? */
3856 ret
= inode_permission(cfile
->f_path
.dentry
->d_inode
, MAY_READ
);
3860 event
->cft
= __file_cft(cfile
);
3861 if (IS_ERR(event
->cft
)) {
3862 ret
= PTR_ERR(event
->cft
);
3866 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3871 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3872 event
->eventfd
, buffer
);
3876 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3877 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3883 * Events should be removed after rmdir of cgroup directory, but before
3884 * destroying subsystem state objects. Let's take reference to cgroup
3885 * directory dentry to do that.
3889 spin_lock(&cgrp
->event_list_lock
);
3890 list_add(&event
->list
, &cgrp
->event_list
);
3891 spin_unlock(&cgrp
->event_list_lock
);
3902 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3903 eventfd_ctx_put(event
->eventfd
);
3905 if (!IS_ERR_OR_NULL(efile
))
3913 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3916 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
3919 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3924 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
3926 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
3931 * for the common functions, 'private' gives the type of file
3933 /* for hysterical raisins, we can't put this on the older files */
3934 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3935 static struct cftype files
[] = {
3938 .open
= cgroup_tasks_open
,
3939 .write_u64
= cgroup_tasks_write
,
3940 .release
= cgroup_pidlist_release
,
3941 .mode
= S_IRUGO
| S_IWUSR
,
3944 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3945 .open
= cgroup_procs_open
,
3946 .write_u64
= cgroup_procs_write
,
3947 .release
= cgroup_pidlist_release
,
3948 .mode
= S_IRUGO
| S_IWUSR
,
3951 .name
= "notify_on_release",
3952 .read_u64
= cgroup_read_notify_on_release
,
3953 .write_u64
= cgroup_write_notify_on_release
,
3956 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3957 .write_string
= cgroup_write_event_control
,
3961 .name
= "cgroup.clone_children",
3962 .read_u64
= cgroup_clone_children_read
,
3963 .write_u64
= cgroup_clone_children_write
,
3966 .name
= "release_agent",
3967 .flags
= CFTYPE_ONLY_ON_ROOT
,
3968 .read_seq_string
= cgroup_release_agent_show
,
3969 .write_string
= cgroup_release_agent_write
,
3970 .max_write_len
= PATH_MAX
,
3976 * cgroup_populate_dir - selectively creation of files in a directory
3977 * @cgrp: target cgroup
3978 * @base_files: true if the base files should be added
3979 * @subsys_mask: mask of the subsystem ids whose files should be added
3981 static int cgroup_populate_dir(struct cgroup
*cgrp
, bool base_files
,
3982 unsigned long subsys_mask
)
3985 struct cgroup_subsys
*ss
;
3988 err
= cgroup_addrm_files(cgrp
, NULL
, files
, true);
3993 /* process cftsets of each subsystem */
3994 for_each_subsys(cgrp
->root
, ss
) {
3995 struct cftype_set
*set
;
3996 if (!test_bit(ss
->subsys_id
, &subsys_mask
))
3999 list_for_each_entry(set
, &ss
->cftsets
, node
)
4000 cgroup_addrm_files(cgrp
, ss
, set
->cfts
, true);
4003 /* This cgroup is ready now */
4004 for_each_subsys(cgrp
->root
, ss
) {
4005 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4007 * Update id->css pointer and make this css visible from
4008 * CSS ID functions. This pointer will be dereferened
4009 * from RCU-read-side without locks.
4012 rcu_assign_pointer(css
->id
->css
, css
);
4018 static void css_dput_fn(struct work_struct
*work
)
4020 struct cgroup_subsys_state
*css
=
4021 container_of(work
, struct cgroup_subsys_state
, dput_work
);
4022 struct dentry
*dentry
= css
->cgroup
->dentry
;
4023 struct super_block
*sb
= dentry
->d_sb
;
4025 atomic_inc(&sb
->s_active
);
4027 deactivate_super(sb
);
4030 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
4031 struct cgroup_subsys
*ss
,
4032 struct cgroup
*cgrp
)
4035 atomic_set(&css
->refcnt
, 1);
4038 if (cgrp
== dummytop
)
4039 css
->flags
|= CSS_ROOT
;
4040 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
4041 cgrp
->subsys
[ss
->subsys_id
] = css
;
4044 * css holds an extra ref to @cgrp->dentry which is put on the last
4045 * css_put(). dput() requires process context, which css_put() may
4046 * be called without. @css->dput_work will be used to invoke
4047 * dput() asynchronously from css_put().
4049 INIT_WORK(&css
->dput_work
, css_dput_fn
);
4052 /* invoke ->post_create() on a new CSS and mark it online if successful */
4053 static int online_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4057 lockdep_assert_held(&cgroup_mutex
);
4060 ret
= ss
->css_online(cgrp
);
4062 cgrp
->subsys
[ss
->subsys_id
]->flags
|= CSS_ONLINE
;
4066 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4067 static void offline_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4068 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4070 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4072 lockdep_assert_held(&cgroup_mutex
);
4074 if (!(css
->flags
& CSS_ONLINE
))
4078 * css_offline() should be called with cgroup_mutex unlocked. See
4079 * 3fa59dfbc3 ("cgroup: fix potential deadlock in pre_destroy") for
4080 * details. This temporary unlocking should go away once
4081 * cgroup_mutex is unexported from controllers.
4083 if (ss
->css_offline
) {
4084 mutex_unlock(&cgroup_mutex
);
4085 ss
->css_offline(cgrp
);
4086 mutex_lock(&cgroup_mutex
);
4089 cgrp
->subsys
[ss
->subsys_id
]->flags
&= ~CSS_ONLINE
;
4093 * cgroup_create - create a cgroup
4094 * @parent: cgroup that will be parent of the new cgroup
4095 * @dentry: dentry of the new cgroup
4096 * @mode: mode to set on new inode
4098 * Must be called with the mutex on the parent inode held
4100 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
4103 struct cgroup
*cgrp
;
4104 struct cgroupfs_root
*root
= parent
->root
;
4106 struct cgroup_subsys
*ss
;
4107 struct super_block
*sb
= root
->sb
;
4109 /* allocate the cgroup and its ID, 0 is reserved for the root */
4110 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4114 cgrp
->id
= ida_simple_get(&root
->cgroup_ida
, 1, 0, GFP_KERNEL
);
4119 * Only live parents can have children. Note that the liveliness
4120 * check isn't strictly necessary because cgroup_mkdir() and
4121 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4122 * anyway so that locking is contained inside cgroup proper and we
4123 * don't get nasty surprises if we ever grow another caller.
4125 if (!cgroup_lock_live_group(parent
)) {
4130 /* Grab a reference on the superblock so the hierarchy doesn't
4131 * get deleted on unmount if there are child cgroups. This
4132 * can be done outside cgroup_mutex, since the sb can't
4133 * disappear while someone has an open control file on the
4135 atomic_inc(&sb
->s_active
);
4137 init_cgroup_housekeeping(cgrp
);
4139 cgrp
->parent
= parent
;
4140 cgrp
->root
= parent
->root
;
4141 cgrp
->top_cgroup
= parent
->top_cgroup
;
4143 if (notify_on_release(parent
))
4144 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4146 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4147 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4149 for_each_subsys(root
, ss
) {
4150 struct cgroup_subsys_state
*css
;
4152 css
= ss
->css_alloc(cgrp
);
4157 init_cgroup_css(css
, ss
, cgrp
);
4159 err
= alloc_css_id(ss
, parent
, cgrp
);
4166 * Create directory. cgroup_create_file() returns with the new
4167 * directory locked on success so that it can be populated without
4168 * dropping cgroup_mutex.
4170 err
= cgroup_create_file(dentry
, S_IFDIR
| mode
, sb
);
4173 lockdep_assert_held(&dentry
->d_inode
->i_mutex
);
4175 /* allocation complete, commit to creation */
4176 dentry
->d_fsdata
= cgrp
;
4177 cgrp
->dentry
= dentry
;
4178 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
4179 list_add_tail_rcu(&cgrp
->sibling
, &cgrp
->parent
->children
);
4180 root
->number_of_cgroups
++;
4182 /* each css holds a ref to the cgroup's dentry */
4183 for_each_subsys(root
, ss
)
4186 /* creation succeeded, notify subsystems */
4187 for_each_subsys(root
, ss
) {
4188 err
= online_css(ss
, cgrp
);
4192 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4194 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",
4195 current
->comm
, current
->pid
, ss
->name
);
4196 if (!strcmp(ss
->name
, "memory"))
4197 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4198 ss
->warned_broken_hierarchy
= true;
4202 err
= cgroup_populate_dir(cgrp
, true, root
->subsys_mask
);
4206 mutex_unlock(&cgroup_mutex
);
4207 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4212 for_each_subsys(root
, ss
) {
4213 if (cgrp
->subsys
[ss
->subsys_id
])
4216 mutex_unlock(&cgroup_mutex
);
4217 /* Release the reference count that we took on the superblock */
4218 deactivate_super(sb
);
4220 ida_simple_remove(&root
->cgroup_ida
, cgrp
->id
);
4226 cgroup_destroy_locked(cgrp
);
4227 mutex_unlock(&cgroup_mutex
);
4228 mutex_unlock(&dentry
->d_inode
->i_mutex
);
4232 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4234 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4236 /* the vfs holds inode->i_mutex already */
4237 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4241 * Check the reference count on each subsystem. Since we already
4242 * established that there are no tasks in the cgroup, if the css refcount
4243 * is also 1, then there should be no outstanding references, so the
4244 * subsystem is safe to destroy. We scan across all subsystems rather than
4245 * using the per-hierarchy linked list of mounted subsystems since we can
4246 * be called via check_for_release() with no synchronization other than
4247 * RCU, and the subsystem linked list isn't RCU-safe.
4249 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
4254 * We won't need to lock the subsys array, because the subsystems
4255 * we're concerned about aren't going anywhere since our cgroup root
4256 * has a reference on them.
4258 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4259 struct cgroup_subsys
*ss
= subsys
[i
];
4260 struct cgroup_subsys_state
*css
;
4262 /* Skip subsystems not present or not in this hierarchy */
4263 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
4266 css
= cgrp
->subsys
[ss
->subsys_id
];
4268 * When called from check_for_release() it's possible
4269 * that by this point the cgroup has been removed
4270 * and the css deleted. But a false-positive doesn't
4271 * matter, since it can only happen if the cgroup
4272 * has been deleted and hence no longer needs the
4273 * release agent to be called anyway.
4275 if (css
&& css_refcnt(css
) > 1)
4281 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4282 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4284 struct dentry
*d
= cgrp
->dentry
;
4285 struct cgroup
*parent
= cgrp
->parent
;
4287 struct cgroup_event
*event
, *tmp
;
4288 struct cgroup_subsys
*ss
;
4289 LIST_HEAD(tmp_list
);
4291 lockdep_assert_held(&d
->d_inode
->i_mutex
);
4292 lockdep_assert_held(&cgroup_mutex
);
4294 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
))
4298 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4299 * removed. This makes future css_tryget() and child creation
4300 * attempts fail thus maintaining the removal conditions verified
4303 for_each_subsys(cgrp
->root
, ss
) {
4304 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4306 WARN_ON(atomic_read(&css
->refcnt
) < 0);
4307 atomic_add(CSS_DEACT_BIAS
, &css
->refcnt
);
4309 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
4311 /* tell subsystems to initate destruction */
4312 for_each_subsys(cgrp
->root
, ss
)
4313 offline_css(ss
, cgrp
);
4316 * Put all the base refs. Each css holds an extra reference to the
4317 * cgroup's dentry and cgroup removal proceeds regardless of css
4318 * refs. On the last put of each css, whenever that may be, the
4319 * extra dentry ref is put so that dentry destruction happens only
4320 * after all css's are released.
4322 for_each_subsys(cgrp
->root
, ss
)
4323 css_put(cgrp
->subsys
[ss
->subsys_id
]);
4325 raw_spin_lock(&release_list_lock
);
4326 if (!list_empty(&cgrp
->release_list
))
4327 list_del_init(&cgrp
->release_list
);
4328 raw_spin_unlock(&release_list_lock
);
4330 /* delete this cgroup from parent->children */
4331 list_del_rcu(&cgrp
->sibling
);
4332 list_del_init(&cgrp
->allcg_node
);
4335 cgroup_d_remove_dir(d
);
4338 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4339 check_for_release(parent
);
4342 * Unregister events and notify userspace.
4343 * Notify userspace about cgroup removing only after rmdir of cgroup
4344 * directory to avoid race between userspace and kernelspace. Use
4345 * a temporary list to avoid a deadlock with cgroup_event_wake(). Since
4346 * cgroup_event_wake() is called with the wait queue head locked,
4347 * remove_wait_queue() cannot be called while holding event_list_lock.
4349 spin_lock(&cgrp
->event_list_lock
);
4350 list_splice_init(&cgrp
->event_list
, &tmp_list
);
4351 spin_unlock(&cgrp
->event_list_lock
);
4352 list_for_each_entry_safe(event
, tmp
, &tmp_list
, list
) {
4353 list_del_init(&event
->list
);
4354 remove_wait_queue(event
->wqh
, &event
->wait
);
4355 eventfd_signal(event
->eventfd
, 1);
4356 schedule_work(&event
->remove
);
4362 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4366 mutex_lock(&cgroup_mutex
);
4367 ret
= cgroup_destroy_locked(dentry
->d_fsdata
);
4368 mutex_unlock(&cgroup_mutex
);
4373 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4375 INIT_LIST_HEAD(&ss
->cftsets
);
4378 * base_cftset is embedded in subsys itself, no need to worry about
4381 if (ss
->base_cftypes
) {
4382 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4383 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4387 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4389 struct cgroup_subsys_state
*css
;
4391 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4393 mutex_lock(&cgroup_mutex
);
4395 /* init base cftset */
4396 cgroup_init_cftsets(ss
);
4398 /* Create the top cgroup state for this subsystem */
4399 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4400 ss
->root
= &rootnode
;
4401 css
= ss
->css_alloc(dummytop
);
4402 /* We don't handle early failures gracefully */
4403 BUG_ON(IS_ERR(css
));
4404 init_cgroup_css(css
, ss
, dummytop
);
4406 /* Update the init_css_set to contain a subsys
4407 * pointer to this state - since the subsystem is
4408 * newly registered, all tasks and hence the
4409 * init_css_set is in the subsystem's top cgroup. */
4410 init_css_set
.subsys
[ss
->subsys_id
] = css
;
4412 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4414 /* At system boot, before all subsystems have been
4415 * registered, no tasks have been forked, so we don't
4416 * need to invoke fork callbacks here. */
4417 BUG_ON(!list_empty(&init_task
.tasks
));
4420 BUG_ON(online_css(ss
, dummytop
));
4422 mutex_unlock(&cgroup_mutex
);
4424 /* this function shouldn't be used with modular subsystems, since they
4425 * need to register a subsys_id, among other things */
4430 * cgroup_load_subsys: load and register a modular subsystem at runtime
4431 * @ss: the subsystem to load
4433 * This function should be called in a modular subsystem's initcall. If the
4434 * subsystem is built as a module, it will be assigned a new subsys_id and set
4435 * up for use. If the subsystem is built-in anyway, work is delegated to the
4436 * simpler cgroup_init_subsys.
4438 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4440 struct cgroup_subsys_state
*css
;
4443 /* check name and function validity */
4444 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4445 ss
->css_alloc
== NULL
|| ss
->css_free
== NULL
)
4449 * we don't support callbacks in modular subsystems. this check is
4450 * before the ss->module check for consistency; a subsystem that could
4451 * be a module should still have no callbacks even if the user isn't
4452 * compiling it as one.
4454 if (ss
->fork
|| ss
->exit
)
4458 * an optionally modular subsystem is built-in: we want to do nothing,
4459 * since cgroup_init_subsys will have already taken care of it.
4461 if (ss
->module
== NULL
) {
4462 /* a sanity check */
4463 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4467 /* init base cftset */
4468 cgroup_init_cftsets(ss
);
4470 mutex_lock(&cgroup_mutex
);
4471 subsys
[ss
->subsys_id
] = ss
;
4474 * no ss->css_alloc seems to need anything important in the ss
4475 * struct, so this can happen first (i.e. before the rootnode
4478 css
= ss
->css_alloc(dummytop
);
4480 /* failure case - need to deassign the subsys[] slot. */
4481 subsys
[ss
->subsys_id
] = NULL
;
4482 mutex_unlock(&cgroup_mutex
);
4483 return PTR_ERR(css
);
4486 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4487 ss
->root
= &rootnode
;
4489 /* our new subsystem will be attached to the dummy hierarchy. */
4490 init_cgroup_css(css
, ss
, dummytop
);
4491 /* init_idr must be after init_cgroup_css because it sets css->id. */
4493 ret
= cgroup_init_idr(ss
, css
);
4499 * Now we need to entangle the css into the existing css_sets. unlike
4500 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4501 * will need a new pointer to it; done by iterating the css_set_table.
4502 * furthermore, modifying the existing css_sets will corrupt the hash
4503 * table state, so each changed css_set will need its hash recomputed.
4504 * this is all done under the css_set_lock.
4506 write_lock(&css_set_lock
);
4507 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
4509 struct hlist_node
*node
, *tmp
;
4510 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
4512 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
4513 /* skip entries that we already rehashed */
4514 if (cg
->subsys
[ss
->subsys_id
])
4516 /* remove existing entry */
4517 hlist_del(&cg
->hlist
);
4519 cg
->subsys
[ss
->subsys_id
] = css
;
4520 /* recompute hash and restore entry */
4521 new_bucket
= css_set_hash(cg
->subsys
);
4522 hlist_add_head(&cg
->hlist
, new_bucket
);
4525 write_unlock(&css_set_lock
);
4528 ret
= online_css(ss
, dummytop
);
4533 mutex_unlock(&cgroup_mutex
);
4537 mutex_unlock(&cgroup_mutex
);
4538 /* @ss can't be mounted here as try_module_get() would fail */
4539 cgroup_unload_subsys(ss
);
4542 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4545 * cgroup_unload_subsys: unload a modular subsystem
4546 * @ss: the subsystem to unload
4548 * This function should be called in a modular subsystem's exitcall. When this
4549 * function is invoked, the refcount on the subsystem's module will be 0, so
4550 * the subsystem will not be attached to any hierarchy.
4552 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4554 struct cg_cgroup_link
*link
;
4555 struct hlist_head
*hhead
;
4557 BUG_ON(ss
->module
== NULL
);
4560 * we shouldn't be called if the subsystem is in use, and the use of
4561 * try_module_get in parse_cgroupfs_options should ensure that it
4562 * doesn't start being used while we're killing it off.
4564 BUG_ON(ss
->root
!= &rootnode
);
4566 mutex_lock(&cgroup_mutex
);
4568 offline_css(ss
, dummytop
);
4572 idr_remove_all(&ss
->idr
);
4573 idr_destroy(&ss
->idr
);
4576 /* deassign the subsys_id */
4577 subsys
[ss
->subsys_id
] = NULL
;
4579 /* remove subsystem from rootnode's list of subsystems */
4580 list_del_init(&ss
->sibling
);
4583 * disentangle the css from all css_sets attached to the dummytop. as
4584 * in loading, we need to pay our respects to the hashtable gods.
4586 write_lock(&css_set_lock
);
4587 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4588 struct css_set
*cg
= link
->cg
;
4590 hlist_del(&cg
->hlist
);
4591 cg
->subsys
[ss
->subsys_id
] = NULL
;
4592 hhead
= css_set_hash(cg
->subsys
);
4593 hlist_add_head(&cg
->hlist
, hhead
);
4595 write_unlock(&css_set_lock
);
4598 * remove subsystem's css from the dummytop and free it - need to
4599 * free before marking as null because ss->css_free needs the
4600 * cgrp->subsys pointer to find their state. note that this also
4601 * takes care of freeing the css_id.
4603 ss
->css_free(dummytop
);
4604 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4606 mutex_unlock(&cgroup_mutex
);
4608 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4611 * cgroup_init_early - cgroup initialization at system boot
4613 * Initialize cgroups at system boot, and initialize any
4614 * subsystems that request early init.
4616 int __init
cgroup_init_early(void)
4619 atomic_set(&init_css_set
.refcount
, 1);
4620 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4621 INIT_LIST_HEAD(&init_css_set
.tasks
);
4622 INIT_HLIST_NODE(&init_css_set
.hlist
);
4624 init_cgroup_root(&rootnode
);
4626 init_task
.cgroups
= &init_css_set
;
4628 init_css_set_link
.cg
= &init_css_set
;
4629 init_css_set_link
.cgrp
= dummytop
;
4630 list_add(&init_css_set_link
.cgrp_link_list
,
4631 &rootnode
.top_cgroup
.css_sets
);
4632 list_add(&init_css_set_link
.cg_link_list
,
4633 &init_css_set
.cg_links
);
4635 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
4636 INIT_HLIST_HEAD(&css_set_table
[i
]);
4638 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4639 struct cgroup_subsys
*ss
= subsys
[i
];
4641 /* at bootup time, we don't worry about modular subsystems */
4642 if (!ss
|| ss
->module
)
4646 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4647 BUG_ON(!ss
->css_alloc
);
4648 BUG_ON(!ss
->css_free
);
4649 if (ss
->subsys_id
!= i
) {
4650 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4651 ss
->name
, ss
->subsys_id
);
4656 cgroup_init_subsys(ss
);
4662 * cgroup_init - cgroup initialization
4664 * Register cgroup filesystem and /proc file, and initialize
4665 * any subsystems that didn't request early init.
4667 int __init
cgroup_init(void)
4671 struct hlist_head
*hhead
;
4673 err
= bdi_init(&cgroup_backing_dev_info
);
4677 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4678 struct cgroup_subsys
*ss
= subsys
[i
];
4680 /* at bootup time, we don't worry about modular subsystems */
4681 if (!ss
|| ss
->module
)
4683 if (!ss
->early_init
)
4684 cgroup_init_subsys(ss
);
4686 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4689 /* Add init_css_set to the hash table */
4690 hhead
= css_set_hash(init_css_set
.subsys
);
4691 hlist_add_head(&init_css_set
.hlist
, hhead
);
4692 BUG_ON(!init_root_id(&rootnode
));
4694 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4700 err
= register_filesystem(&cgroup_fs_type
);
4702 kobject_put(cgroup_kobj
);
4706 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4710 bdi_destroy(&cgroup_backing_dev_info
);
4716 * proc_cgroup_show()
4717 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4718 * - Used for /proc/<pid>/cgroup.
4719 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4720 * doesn't really matter if tsk->cgroup changes after we read it,
4721 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4722 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4723 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4724 * cgroup to top_cgroup.
4727 /* TODO: Use a proper seq_file iterator */
4728 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4731 struct task_struct
*tsk
;
4734 struct cgroupfs_root
*root
;
4737 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4743 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4749 mutex_lock(&cgroup_mutex
);
4751 for_each_active_root(root
) {
4752 struct cgroup_subsys
*ss
;
4753 struct cgroup
*cgrp
;
4756 seq_printf(m
, "%d:", root
->hierarchy_id
);
4757 for_each_subsys(root
, ss
)
4758 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4759 if (strlen(root
->name
))
4760 seq_printf(m
, "%sname=%s", count
? "," : "",
4763 cgrp
= task_cgroup_from_root(tsk
, root
);
4764 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4772 mutex_unlock(&cgroup_mutex
);
4773 put_task_struct(tsk
);
4780 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4782 struct pid
*pid
= PROC_I(inode
)->pid
;
4783 return single_open(file
, proc_cgroup_show
, pid
);
4786 const struct file_operations proc_cgroup_operations
= {
4787 .open
= cgroup_open
,
4789 .llseek
= seq_lseek
,
4790 .release
= single_release
,
4793 /* Display information about each subsystem and each hierarchy */
4794 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4798 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4800 * ideally we don't want subsystems moving around while we do this.
4801 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4802 * subsys/hierarchy state.
4804 mutex_lock(&cgroup_mutex
);
4805 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4806 struct cgroup_subsys
*ss
= subsys
[i
];
4809 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4810 ss
->name
, ss
->root
->hierarchy_id
,
4811 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4813 mutex_unlock(&cgroup_mutex
);
4817 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4819 return single_open(file
, proc_cgroupstats_show
, NULL
);
4822 static const struct file_operations proc_cgroupstats_operations
= {
4823 .open
= cgroupstats_open
,
4825 .llseek
= seq_lseek
,
4826 .release
= single_release
,
4830 * cgroup_fork - attach newly forked task to its parents cgroup.
4831 * @child: pointer to task_struct of forking parent process.
4833 * Description: A task inherits its parent's cgroup at fork().
4835 * A pointer to the shared css_set was automatically copied in
4836 * fork.c by dup_task_struct(). However, we ignore that copy, since
4837 * it was not made under the protection of RCU or cgroup_mutex, so
4838 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4839 * have already changed current->cgroups, allowing the previously
4840 * referenced cgroup group to be removed and freed.
4842 * At the point that cgroup_fork() is called, 'current' is the parent
4843 * task, and the passed argument 'child' points to the child task.
4845 void cgroup_fork(struct task_struct
*child
)
4848 child
->cgroups
= current
->cgroups
;
4849 get_css_set(child
->cgroups
);
4850 task_unlock(current
);
4851 INIT_LIST_HEAD(&child
->cg_list
);
4855 * cgroup_post_fork - called on a new task after adding it to the task list
4856 * @child: the task in question
4858 * Adds the task to the list running through its css_set if necessary and
4859 * call the subsystem fork() callbacks. Has to be after the task is
4860 * visible on the task list in case we race with the first call to
4861 * cgroup_iter_start() - to guarantee that the new task ends up on its
4864 void cgroup_post_fork(struct task_struct
*child
)
4869 * use_task_css_set_links is set to 1 before we walk the tasklist
4870 * under the tasklist_lock and we read it here after we added the child
4871 * to the tasklist under the tasklist_lock as well. If the child wasn't
4872 * yet in the tasklist when we walked through it from
4873 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4874 * should be visible now due to the paired locking and barriers implied
4875 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4876 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4879 if (use_task_css_set_links
) {
4880 write_lock(&css_set_lock
);
4882 if (list_empty(&child
->cg_list
))
4883 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4885 write_unlock(&css_set_lock
);
4889 * Call ss->fork(). This must happen after @child is linked on
4890 * css_set; otherwise, @child might change state between ->fork()
4891 * and addition to css_set.
4893 if (need_forkexit_callback
) {
4894 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4895 struct cgroup_subsys
*ss
= subsys
[i
];
4898 * fork/exit callbacks are supported only for
4899 * builtin subsystems and we don't need further
4900 * synchronization as they never go away.
4902 if (!ss
|| ss
->module
)
4912 * cgroup_exit - detach cgroup from exiting task
4913 * @tsk: pointer to task_struct of exiting process
4914 * @run_callback: run exit callbacks?
4916 * Description: Detach cgroup from @tsk and release it.
4918 * Note that cgroups marked notify_on_release force every task in
4919 * them to take the global cgroup_mutex mutex when exiting.
4920 * This could impact scaling on very large systems. Be reluctant to
4921 * use notify_on_release cgroups where very high task exit scaling
4922 * is required on large systems.
4924 * the_top_cgroup_hack:
4926 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4928 * We call cgroup_exit() while the task is still competent to
4929 * handle notify_on_release(), then leave the task attached to the
4930 * root cgroup in each hierarchy for the remainder of its exit.
4932 * To do this properly, we would increment the reference count on
4933 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4934 * code we would add a second cgroup function call, to drop that
4935 * reference. This would just create an unnecessary hot spot on
4936 * the top_cgroup reference count, to no avail.
4938 * Normally, holding a reference to a cgroup without bumping its
4939 * count is unsafe. The cgroup could go away, or someone could
4940 * attach us to a different cgroup, decrementing the count on
4941 * the first cgroup that we never incremented. But in this case,
4942 * top_cgroup isn't going away, and either task has PF_EXITING set,
4943 * which wards off any cgroup_attach_task() attempts, or task is a failed
4944 * fork, never visible to cgroup_attach_task.
4946 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4952 * Unlink from the css_set task list if necessary.
4953 * Optimistically check cg_list before taking
4956 if (!list_empty(&tsk
->cg_list
)) {
4957 write_lock(&css_set_lock
);
4958 if (!list_empty(&tsk
->cg_list
))
4959 list_del_init(&tsk
->cg_list
);
4960 write_unlock(&css_set_lock
);
4963 /* Reassign the task to the init_css_set. */
4966 tsk
->cgroups
= &init_css_set
;
4968 if (run_callbacks
&& need_forkexit_callback
) {
4969 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4970 struct cgroup_subsys
*ss
= subsys
[i
];
4972 /* modular subsystems can't use callbacks */
4973 if (!ss
|| ss
->module
)
4977 struct cgroup
*old_cgrp
=
4978 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4979 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4980 ss
->exit(cgrp
, old_cgrp
, tsk
);
4987 put_css_set_taskexit(cg
);
4991 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4992 * @cgrp: the cgroup in question
4993 * @task: the task in question
4995 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4998 * If we are sending in dummytop, then presumably we are creating
4999 * the top cgroup in the subsystem.
5001 * Called only by the ns (nsproxy) cgroup.
5003 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
5006 struct cgroup
*target
;
5008 if (cgrp
== dummytop
)
5011 target
= task_cgroup_from_root(task
, cgrp
->root
);
5012 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
5013 cgrp
= cgrp
->parent
;
5014 ret
= (cgrp
== target
);
5018 static void check_for_release(struct cgroup
*cgrp
)
5020 /* All of these checks rely on RCU to keep the cgroup
5021 * structure alive */
5022 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
5023 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
5024 /* Control Group is currently removeable. If it's not
5025 * already queued for a userspace notification, queue
5027 int need_schedule_work
= 0;
5028 raw_spin_lock(&release_list_lock
);
5029 if (!cgroup_is_removed(cgrp
) &&
5030 list_empty(&cgrp
->release_list
)) {
5031 list_add(&cgrp
->release_list
, &release_list
);
5032 need_schedule_work
= 1;
5034 raw_spin_unlock(&release_list_lock
);
5035 if (need_schedule_work
)
5036 schedule_work(&release_agent_work
);
5040 /* Caller must verify that the css is not for root cgroup */
5041 bool __css_tryget(struct cgroup_subsys_state
*css
)
5046 v
= css_refcnt(css
);
5047 t
= atomic_cmpxchg(&css
->refcnt
, v
, v
+ 1);
5055 EXPORT_SYMBOL_GPL(__css_tryget
);
5057 /* Caller must verify that the css is not for root cgroup */
5058 void __css_put(struct cgroup_subsys_state
*css
)
5060 struct cgroup
*cgrp
= css
->cgroup
;
5064 v
= css_unbias_refcnt(atomic_dec_return(&css
->refcnt
));
5068 if (notify_on_release(cgrp
)) {
5069 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5070 check_for_release(cgrp
);
5074 schedule_work(&css
->dput_work
);
5079 EXPORT_SYMBOL_GPL(__css_put
);
5082 * Notify userspace when a cgroup is released, by running the
5083 * configured release agent with the name of the cgroup (path
5084 * relative to the root of cgroup file system) as the argument.
5086 * Most likely, this user command will try to rmdir this cgroup.
5088 * This races with the possibility that some other task will be
5089 * attached to this cgroup before it is removed, or that some other
5090 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5091 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5092 * unused, and this cgroup will be reprieved from its death sentence,
5093 * to continue to serve a useful existence. Next time it's released,
5094 * we will get notified again, if it still has 'notify_on_release' set.
5096 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5097 * means only wait until the task is successfully execve()'d. The
5098 * separate release agent task is forked by call_usermodehelper(),
5099 * then control in this thread returns here, without waiting for the
5100 * release agent task. We don't bother to wait because the caller of
5101 * this routine has no use for the exit status of the release agent
5102 * task, so no sense holding our caller up for that.
5104 static void cgroup_release_agent(struct work_struct
*work
)
5106 BUG_ON(work
!= &release_agent_work
);
5107 mutex_lock(&cgroup_mutex
);
5108 raw_spin_lock(&release_list_lock
);
5109 while (!list_empty(&release_list
)) {
5110 char *argv
[3], *envp
[3];
5112 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5113 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5116 list_del_init(&cgrp
->release_list
);
5117 raw_spin_unlock(&release_list_lock
);
5118 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5121 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5123 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5128 argv
[i
++] = agentbuf
;
5129 argv
[i
++] = pathbuf
;
5133 /* minimal command environment */
5134 envp
[i
++] = "HOME=/";
5135 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5138 /* Drop the lock while we invoke the usermode helper,
5139 * since the exec could involve hitting disk and hence
5140 * be a slow process */
5141 mutex_unlock(&cgroup_mutex
);
5142 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5143 mutex_lock(&cgroup_mutex
);
5147 raw_spin_lock(&release_list_lock
);
5149 raw_spin_unlock(&release_list_lock
);
5150 mutex_unlock(&cgroup_mutex
);
5153 static int __init
cgroup_disable(char *str
)
5158 while ((token
= strsep(&str
, ",")) != NULL
) {
5161 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
5162 struct cgroup_subsys
*ss
= subsys
[i
];
5165 * cgroup_disable, being at boot time, can't
5166 * know about module subsystems, so we don't
5169 if (!ss
|| ss
->module
)
5172 if (!strcmp(token
, ss
->name
)) {
5174 printk(KERN_INFO
"Disabling %s control group"
5175 " subsystem\n", ss
->name
);
5182 __setup("cgroup_disable=", cgroup_disable
);
5185 * Functons for CSS ID.
5189 *To get ID other than 0, this should be called when !cgroup_is_removed().
5191 unsigned short css_id(struct cgroup_subsys_state
*css
)
5193 struct css_id
*cssid
;
5196 * This css_id() can return correct value when somone has refcnt
5197 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5198 * it's unchanged until freed.
5200 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5206 EXPORT_SYMBOL_GPL(css_id
);
5208 unsigned short css_depth(struct cgroup_subsys_state
*css
)
5210 struct css_id
*cssid
;
5212 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5215 return cssid
->depth
;
5218 EXPORT_SYMBOL_GPL(css_depth
);
5221 * css_is_ancestor - test "root" css is an ancestor of "child"
5222 * @child: the css to be tested.
5223 * @root: the css supporsed to be an ancestor of the child.
5225 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5226 * this function reads css->id, the caller must hold rcu_read_lock().
5227 * But, considering usual usage, the csses should be valid objects after test.
5228 * Assuming that the caller will do some action to the child if this returns
5229 * returns true, the caller must take "child";s reference count.
5230 * If "child" is valid object and this returns true, "root" is valid, too.
5233 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5234 const struct cgroup_subsys_state
*root
)
5236 struct css_id
*child_id
;
5237 struct css_id
*root_id
;
5239 child_id
= rcu_dereference(child
->id
);
5242 root_id
= rcu_dereference(root
->id
);
5245 if (child_id
->depth
< root_id
->depth
)
5247 if (child_id
->stack
[root_id
->depth
] != root_id
->id
)
5252 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5254 struct css_id
*id
= css
->id
;
5255 /* When this is called before css_id initialization, id can be NULL */
5259 BUG_ON(!ss
->use_id
);
5261 rcu_assign_pointer(id
->css
, NULL
);
5262 rcu_assign_pointer(css
->id
, NULL
);
5263 spin_lock(&ss
->id_lock
);
5264 idr_remove(&ss
->idr
, id
->id
);
5265 spin_unlock(&ss
->id_lock
);
5266 kfree_rcu(id
, rcu_head
);
5268 EXPORT_SYMBOL_GPL(free_css_id
);
5271 * This is called by init or create(). Then, calls to this function are
5272 * always serialized (By cgroup_mutex() at create()).
5275 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5277 struct css_id
*newid
;
5278 int myid
, error
, size
;
5280 BUG_ON(!ss
->use_id
);
5282 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5283 newid
= kzalloc(size
, GFP_KERNEL
);
5285 return ERR_PTR(-ENOMEM
);
5287 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
5291 spin_lock(&ss
->id_lock
);
5292 /* Don't use 0. allocates an ID of 1-65535 */
5293 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
5294 spin_unlock(&ss
->id_lock
);
5296 /* Returns error when there are no free spaces for new ID.*/
5301 if (myid
> CSS_ID_MAX
)
5305 newid
->depth
= depth
;
5309 spin_lock(&ss
->id_lock
);
5310 idr_remove(&ss
->idr
, myid
);
5311 spin_unlock(&ss
->id_lock
);
5314 return ERR_PTR(error
);
5318 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5319 struct cgroup_subsys_state
*rootcss
)
5321 struct css_id
*newid
;
5323 spin_lock_init(&ss
->id_lock
);
5326 newid
= get_new_cssid(ss
, 0);
5328 return PTR_ERR(newid
);
5330 newid
->stack
[0] = newid
->id
;
5331 newid
->css
= rootcss
;
5332 rootcss
->id
= newid
;
5336 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5337 struct cgroup
*child
)
5339 int subsys_id
, i
, depth
= 0;
5340 struct cgroup_subsys_state
*parent_css
, *child_css
;
5341 struct css_id
*child_id
, *parent_id
;
5343 subsys_id
= ss
->subsys_id
;
5344 parent_css
= parent
->subsys
[subsys_id
];
5345 child_css
= child
->subsys
[subsys_id
];
5346 parent_id
= parent_css
->id
;
5347 depth
= parent_id
->depth
+ 1;
5349 child_id
= get_new_cssid(ss
, depth
);
5350 if (IS_ERR(child_id
))
5351 return PTR_ERR(child_id
);
5353 for (i
= 0; i
< depth
; i
++)
5354 child_id
->stack
[i
] = parent_id
->stack
[i
];
5355 child_id
->stack
[depth
] = child_id
->id
;
5357 * child_id->css pointer will be set after this cgroup is available
5358 * see cgroup_populate_dir()
5360 rcu_assign_pointer(child_css
->id
, child_id
);
5366 * css_lookup - lookup css by id
5367 * @ss: cgroup subsys to be looked into.
5370 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5371 * NULL if not. Should be called under rcu_read_lock()
5373 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5375 struct css_id
*cssid
= NULL
;
5377 BUG_ON(!ss
->use_id
);
5378 cssid
= idr_find(&ss
->idr
, id
);
5380 if (unlikely(!cssid
))
5383 return rcu_dereference(cssid
->css
);
5385 EXPORT_SYMBOL_GPL(css_lookup
);
5388 * css_get_next - lookup next cgroup under specified hierarchy.
5389 * @ss: pointer to subsystem
5390 * @id: current position of iteration.
5391 * @root: pointer to css. search tree under this.
5392 * @foundid: position of found object.
5394 * Search next css under the specified hierarchy of rootid. Calling under
5395 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5397 struct cgroup_subsys_state
*
5398 css_get_next(struct cgroup_subsys
*ss
, int id
,
5399 struct cgroup_subsys_state
*root
, int *foundid
)
5401 struct cgroup_subsys_state
*ret
= NULL
;
5404 int rootid
= css_id(root
);
5405 int depth
= css_depth(root
);
5410 BUG_ON(!ss
->use_id
);
5411 WARN_ON_ONCE(!rcu_read_lock_held());
5413 /* fill start point for scan */
5417 * scan next entry from bitmap(tree), tmpid is updated after
5420 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
5423 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
5424 ret
= rcu_dereference(tmp
->css
);
5430 /* continue to scan from next id */
5437 * get corresponding css from file open on cgroupfs directory
5439 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5441 struct cgroup
*cgrp
;
5442 struct inode
*inode
;
5443 struct cgroup_subsys_state
*css
;
5445 inode
= f
->f_dentry
->d_inode
;
5446 /* check in cgroup filesystem dir */
5447 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5448 return ERR_PTR(-EBADF
);
5450 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5451 return ERR_PTR(-EINVAL
);
5454 cgrp
= __d_cgrp(f
->f_dentry
);
5455 css
= cgrp
->subsys
[id
];
5456 return css
? css
: ERR_PTR(-ENOENT
);
5459 #ifdef CONFIG_CGROUP_DEBUG
5460 static struct cgroup_subsys_state
*debug_css_alloc(struct cgroup
*cont
)
5462 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5465 return ERR_PTR(-ENOMEM
);
5470 static void debug_css_free(struct cgroup
*cont
)
5472 kfree(cont
->subsys
[debug_subsys_id
]);
5475 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5477 return atomic_read(&cont
->count
);
5480 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5482 return cgroup_task_count(cont
);
5485 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5487 return (u64
)(unsigned long)current
->cgroups
;
5490 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5496 count
= atomic_read(¤t
->cgroups
->refcount
);
5501 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5503 struct seq_file
*seq
)
5505 struct cg_cgroup_link
*link
;
5508 read_lock(&css_set_lock
);
5510 cg
= rcu_dereference(current
->cgroups
);
5511 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5512 struct cgroup
*c
= link
->cgrp
;
5516 name
= c
->dentry
->d_name
.name
;
5519 seq_printf(seq
, "Root %d group %s\n",
5520 c
->root
->hierarchy_id
, name
);
5523 read_unlock(&css_set_lock
);
5527 #define MAX_TASKS_SHOWN_PER_CSS 25
5528 static int cgroup_css_links_read(struct cgroup
*cont
,
5530 struct seq_file
*seq
)
5532 struct cg_cgroup_link
*link
;
5534 read_lock(&css_set_lock
);
5535 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5536 struct css_set
*cg
= link
->cg
;
5537 struct task_struct
*task
;
5539 seq_printf(seq
, "css_set %p\n", cg
);
5540 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5541 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5542 seq_puts(seq
, " ...\n");
5545 seq_printf(seq
, " task %d\n",
5546 task_pid_vnr(task
));
5550 read_unlock(&css_set_lock
);
5554 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5556 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5559 static struct cftype debug_files
[] = {
5561 .name
= "cgroup_refcount",
5562 .read_u64
= cgroup_refcount_read
,
5565 .name
= "taskcount",
5566 .read_u64
= debug_taskcount_read
,
5570 .name
= "current_css_set",
5571 .read_u64
= current_css_set_read
,
5575 .name
= "current_css_set_refcount",
5576 .read_u64
= current_css_set_refcount_read
,
5580 .name
= "current_css_set_cg_links",
5581 .read_seq_string
= current_css_set_cg_links_read
,
5585 .name
= "cgroup_css_links",
5586 .read_seq_string
= cgroup_css_links_read
,
5590 .name
= "releasable",
5591 .read_u64
= releasable_read
,
5597 struct cgroup_subsys debug_subsys
= {
5599 .css_alloc
= debug_css_alloc
,
5600 .css_free
= debug_css_free
,
5601 .subsys_id
= debug_subsys_id
,
5602 .base_cftypes
= debug_files
,
5604 #endif /* CONFIG_CGROUP_DEBUG */