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/hashtable.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_task */
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 static struct cgroup_name root_cgroup_name
= { .name
= "/" };
243 /* This flag indicates whether tasks in the fork and exit paths should
244 * check for fork/exit handlers to call. This avoids us having to do
245 * extra work in the fork/exit path if none of the subsystems need to
248 static int need_forkexit_callback __read_mostly
;
250 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
251 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
252 struct cftype cfts
[], bool is_add
);
254 #ifdef CONFIG_PROVE_LOCKING
255 int cgroup_lock_is_held(void)
257 return lockdep_is_held(&cgroup_mutex
);
259 #else /* #ifdef CONFIG_PROVE_LOCKING */
260 int cgroup_lock_is_held(void)
262 return mutex_is_locked(&cgroup_mutex
);
264 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
266 EXPORT_SYMBOL_GPL(cgroup_lock_is_held
);
268 static int css_unbias_refcnt(int refcnt
)
270 return refcnt
>= 0 ? refcnt
: refcnt
- CSS_DEACT_BIAS
;
273 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
274 static int css_refcnt(struct cgroup_subsys_state
*css
)
276 int v
= atomic_read(&css
->refcnt
);
278 return css_unbias_refcnt(v
);
281 /* convenient tests for these bits */
282 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
284 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
287 /* bits in struct cgroupfs_root flags field */
289 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
290 ROOT_XATTR
, /* supports extended attributes */
293 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
296 (1 << CGRP_RELEASABLE
) |
297 (1 << CGRP_NOTIFY_ON_RELEASE
);
298 return (cgrp
->flags
& bits
) == bits
;
301 static int notify_on_release(const struct cgroup
*cgrp
)
303 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
307 * for_each_subsys() allows you to iterate on each subsystem attached to
308 * an active hierarchy
310 #define for_each_subsys(_root, _ss) \
311 list_for_each_entry(_ss, &_root->subsys_list, sibling)
313 /* for_each_active_root() allows you to iterate across the active hierarchies */
314 #define for_each_active_root(_root) \
315 list_for_each_entry(_root, &roots, root_list)
317 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
319 return dentry
->d_fsdata
;
322 static inline struct cfent
*__d_cfe(struct dentry
*dentry
)
324 return dentry
->d_fsdata
;
327 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
329 return __d_cfe(dentry
)->type
;
333 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
334 * @cgrp: the cgroup to be checked for liveness
336 * On success, returns true; the lock should be later released with
337 * cgroup_unlock(). On failure returns false with no lock held.
339 static bool cgroup_lock_live_group(struct cgroup
*cgrp
)
341 mutex_lock(&cgroup_mutex
);
342 if (cgroup_is_removed(cgrp
)) {
343 mutex_unlock(&cgroup_mutex
);
349 /* the list of cgroups eligible for automatic release. Protected by
350 * release_list_lock */
351 static LIST_HEAD(release_list
);
352 static DEFINE_RAW_SPINLOCK(release_list_lock
);
353 static void cgroup_release_agent(struct work_struct
*work
);
354 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
355 static void check_for_release(struct cgroup
*cgrp
);
357 /* Link structure for associating css_set objects with cgroups */
358 struct cg_cgroup_link
{
360 * List running through cg_cgroup_links associated with a
361 * cgroup, anchored on cgroup->css_sets
363 struct list_head cgrp_link_list
;
366 * List running through cg_cgroup_links pointing at a
367 * single css_set object, anchored on css_set->cg_links
369 struct list_head cg_link_list
;
373 /* The default css_set - used by init and its children prior to any
374 * hierarchies being mounted. It contains a pointer to the root state
375 * for each subsystem. Also used to anchor the list of css_sets. Not
376 * reference-counted, to improve performance when child cgroups
377 * haven't been created.
380 static struct css_set init_css_set
;
381 static struct cg_cgroup_link init_css_set_link
;
383 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
384 struct cgroup_subsys_state
*css
);
386 /* css_set_lock protects the list of css_set objects, and the
387 * chain of tasks off each css_set. Nests outside task->alloc_lock
388 * due to cgroup_iter_start() */
389 static DEFINE_RWLOCK(css_set_lock
);
390 static int css_set_count
;
393 * hash table for cgroup groups. This improves the performance to find
394 * an existing css_set. This hash doesn't (currently) take into
395 * account cgroups in empty hierarchies.
397 #define CSS_SET_HASH_BITS 7
398 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
400 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
403 unsigned long key
= 0UL;
405 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
406 key
+= (unsigned long)css
[i
];
407 key
= (key
>> 16) ^ key
;
412 /* We don't maintain the lists running through each css_set to its
413 * task until after the first call to cgroup_iter_start(). This
414 * reduces the fork()/exit() overhead for people who have cgroups
415 * compiled into their kernel but not actually in use */
416 static int use_task_css_set_links __read_mostly
;
418 static void __put_css_set(struct css_set
*cg
, int taskexit
)
420 struct cg_cgroup_link
*link
;
421 struct cg_cgroup_link
*saved_link
;
423 * Ensure that the refcount doesn't hit zero while any readers
424 * can see it. Similar to atomic_dec_and_lock(), but for an
427 if (atomic_add_unless(&cg
->refcount
, -1, 1))
429 write_lock(&css_set_lock
);
430 if (!atomic_dec_and_test(&cg
->refcount
)) {
431 write_unlock(&css_set_lock
);
435 /* This css_set is dead. unlink it and release cgroup refcounts */
436 hash_del(&cg
->hlist
);
439 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
441 struct cgroup
*cgrp
= link
->cgrp
;
442 list_del(&link
->cg_link_list
);
443 list_del(&link
->cgrp_link_list
);
446 * We may not be holding cgroup_mutex, and if cgrp->count is
447 * dropped to 0 the cgroup can be destroyed at any time, hence
448 * rcu_read_lock is used to keep it alive.
451 if (atomic_dec_and_test(&cgrp
->count
) &&
452 notify_on_release(cgrp
)) {
454 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
455 check_for_release(cgrp
);
462 write_unlock(&css_set_lock
);
463 kfree_rcu(cg
, rcu_head
);
467 * refcounted get/put for css_set objects
469 static inline void get_css_set(struct css_set
*cg
)
471 atomic_inc(&cg
->refcount
);
474 static inline void put_css_set(struct css_set
*cg
)
476 __put_css_set(cg
, 0);
479 static inline void put_css_set_taskexit(struct css_set
*cg
)
481 __put_css_set(cg
, 1);
485 * compare_css_sets - helper function for find_existing_css_set().
486 * @cg: candidate css_set being tested
487 * @old_cg: existing css_set for a task
488 * @new_cgrp: cgroup that's being entered by the task
489 * @template: desired set of css pointers in css_set (pre-calculated)
491 * Returns true if "cg" matches "old_cg" except for the hierarchy
492 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
494 static bool compare_css_sets(struct css_set
*cg
,
495 struct css_set
*old_cg
,
496 struct cgroup
*new_cgrp
,
497 struct cgroup_subsys_state
*template[])
499 struct list_head
*l1
, *l2
;
501 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
502 /* Not all subsystems matched */
507 * Compare cgroup pointers in order to distinguish between
508 * different cgroups in heirarchies with no subsystems. We
509 * could get by with just this check alone (and skip the
510 * memcmp above) but on most setups the memcmp check will
511 * avoid the need for this more expensive check on almost all
516 l2
= &old_cg
->cg_links
;
518 struct cg_cgroup_link
*cgl1
, *cgl2
;
519 struct cgroup
*cg1
, *cg2
;
523 /* See if we reached the end - both lists are equal length. */
524 if (l1
== &cg
->cg_links
) {
525 BUG_ON(l2
!= &old_cg
->cg_links
);
528 BUG_ON(l2
== &old_cg
->cg_links
);
530 /* Locate the cgroups associated with these links. */
531 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
532 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
535 /* Hierarchies should be linked in the same order. */
536 BUG_ON(cg1
->root
!= cg2
->root
);
539 * If this hierarchy is the hierarchy of the cgroup
540 * that's changing, then we need to check that this
541 * css_set points to the new cgroup; if it's any other
542 * hierarchy, then this css_set should point to the
543 * same cgroup as the old css_set.
545 if (cg1
->root
== new_cgrp
->root
) {
557 * find_existing_css_set() is a helper for
558 * find_css_set(), and checks to see whether an existing
559 * css_set is suitable.
561 * oldcg: the cgroup group that we're using before the cgroup
564 * cgrp: the cgroup that we're moving into
566 * template: location in which to build the desired set of subsystem
567 * state objects for the new cgroup group
569 static struct css_set
*find_existing_css_set(
570 struct css_set
*oldcg
,
572 struct cgroup_subsys_state
*template[])
575 struct cgroupfs_root
*root
= cgrp
->root
;
580 * Build the set of subsystem state objects that we want to see in the
581 * new css_set. while subsystems can change globally, the entries here
582 * won't change, so no need for locking.
584 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
585 if (root
->subsys_mask
& (1UL << i
)) {
586 /* Subsystem is in this hierarchy. So we want
587 * the subsystem state from the new
589 template[i
] = cgrp
->subsys
[i
];
591 /* Subsystem is not in this hierarchy, so we
592 * don't want to change the subsystem state */
593 template[i
] = oldcg
->subsys
[i
];
597 key
= css_set_hash(template);
598 hash_for_each_possible(css_set_table
, cg
, hlist
, key
) {
599 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
602 /* This css_set matches what we need */
606 /* No existing cgroup group matched */
610 static void free_cg_links(struct list_head
*tmp
)
612 struct cg_cgroup_link
*link
;
613 struct cg_cgroup_link
*saved_link
;
615 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
616 list_del(&link
->cgrp_link_list
);
622 * allocate_cg_links() allocates "count" cg_cgroup_link structures
623 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
624 * success or a negative error
626 static int allocate_cg_links(int count
, struct list_head
*tmp
)
628 struct cg_cgroup_link
*link
;
631 for (i
= 0; i
< count
; i
++) {
632 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
637 list_add(&link
->cgrp_link_list
, tmp
);
643 * link_css_set - a helper function to link a css_set to a cgroup
644 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
645 * @cg: the css_set to be linked
646 * @cgrp: the destination cgroup
648 static void link_css_set(struct list_head
*tmp_cg_links
,
649 struct css_set
*cg
, struct cgroup
*cgrp
)
651 struct cg_cgroup_link
*link
;
653 BUG_ON(list_empty(tmp_cg_links
));
654 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
658 atomic_inc(&cgrp
->count
);
659 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
661 * Always add links to the tail of the list so that the list
662 * is sorted by order of hierarchy creation
664 list_add_tail(&link
->cg_link_list
, &cg
->cg_links
);
668 * find_css_set() takes an existing cgroup group and a
669 * cgroup object, and returns a css_set object that's
670 * equivalent to the old group, but with the given cgroup
671 * substituted into the appropriate hierarchy. Must be called with
674 static struct css_set
*find_css_set(
675 struct css_set
*oldcg
, struct cgroup
*cgrp
)
678 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
680 struct list_head tmp_cg_links
;
682 struct cg_cgroup_link
*link
;
685 /* First see if we already have a cgroup group that matches
687 read_lock(&css_set_lock
);
688 res
= find_existing_css_set(oldcg
, cgrp
, template);
691 read_unlock(&css_set_lock
);
696 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
700 /* Allocate all the cg_cgroup_link objects that we'll need */
701 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
706 atomic_set(&res
->refcount
, 1);
707 INIT_LIST_HEAD(&res
->cg_links
);
708 INIT_LIST_HEAD(&res
->tasks
);
709 INIT_HLIST_NODE(&res
->hlist
);
711 /* Copy the set of subsystem state objects generated in
712 * find_existing_css_set() */
713 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
715 write_lock(&css_set_lock
);
716 /* Add reference counts and links from the new css_set. */
717 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
718 struct cgroup
*c
= link
->cgrp
;
719 if (c
->root
== cgrp
->root
)
721 link_css_set(&tmp_cg_links
, res
, c
);
724 BUG_ON(!list_empty(&tmp_cg_links
));
728 /* Add this cgroup group to the hash table */
729 key
= css_set_hash(res
->subsys
);
730 hash_add(css_set_table
, &res
->hlist
, key
);
732 write_unlock(&css_set_lock
);
738 * Return the cgroup for "task" from the given hierarchy. Must be
739 * called with cgroup_mutex held.
741 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
742 struct cgroupfs_root
*root
)
745 struct cgroup
*res
= NULL
;
747 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
748 read_lock(&css_set_lock
);
750 * No need to lock the task - since we hold cgroup_mutex the
751 * task can't change groups, so the only thing that can happen
752 * is that it exits and its css is set back to init_css_set.
755 if (css
== &init_css_set
) {
756 res
= &root
->top_cgroup
;
758 struct cg_cgroup_link
*link
;
759 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
760 struct cgroup
*c
= link
->cgrp
;
761 if (c
->root
== root
) {
767 read_unlock(&css_set_lock
);
773 * There is one global cgroup mutex. We also require taking
774 * task_lock() when dereferencing a task's cgroup subsys pointers.
775 * See "The task_lock() exception", at the end of this comment.
777 * A task must hold cgroup_mutex to modify cgroups.
779 * Any task can increment and decrement the count field without lock.
780 * So in general, code holding cgroup_mutex can't rely on the count
781 * field not changing. However, if the count goes to zero, then only
782 * cgroup_attach_task() can increment it again. Because a count of zero
783 * means that no tasks are currently attached, therefore there is no
784 * way a task attached to that cgroup can fork (the other way to
785 * increment the count). So code holding cgroup_mutex can safely
786 * assume that if the count is zero, it will stay zero. Similarly, if
787 * a task holds cgroup_mutex on a cgroup with zero count, it
788 * knows that the cgroup won't be removed, as cgroup_rmdir()
791 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
792 * (usually) take cgroup_mutex. These are the two most performance
793 * critical pieces of code here. The exception occurs on cgroup_exit(),
794 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
795 * is taken, and if the cgroup count is zero, a usermode call made
796 * to the release agent with the name of the cgroup (path relative to
797 * the root of cgroup file system) as the argument.
799 * A cgroup can only be deleted if both its 'count' of using tasks
800 * is zero, and its list of 'children' cgroups is empty. Since all
801 * tasks in the system use _some_ cgroup, and since there is always at
802 * least one task in the system (init, pid == 1), therefore, top_cgroup
803 * always has either children cgroups and/or using tasks. So we don't
804 * need a special hack to ensure that top_cgroup cannot be deleted.
806 * The task_lock() exception
808 * The need for this exception arises from the action of
809 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
810 * another. It does so using cgroup_mutex, however there are
811 * several performance critical places that need to reference
812 * task->cgroup without the expense of grabbing a system global
813 * mutex. Therefore except as noted below, when dereferencing or, as
814 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
815 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
816 * the task_struct routinely used for such matters.
818 * P.S. One more locking exception. RCU is used to guard the
819 * update of a tasks cgroup pointer by cgroup_attach_task()
823 * cgroup_lock - lock out any changes to cgroup structures
826 static void cgroup_lock(void)
828 mutex_lock(&cgroup_mutex
);
832 * cgroup_unlock - release lock on cgroup changes
834 * Undo the lock taken in a previous cgroup_lock() call.
836 static void cgroup_unlock(void)
838 mutex_unlock(&cgroup_mutex
);
842 * A couple of forward declarations required, due to cyclic reference loop:
843 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
844 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
848 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
849 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, unsigned int);
850 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
851 static int cgroup_populate_dir(struct cgroup
*cgrp
, bool base_files
,
852 unsigned long subsys_mask
);
853 static const struct inode_operations cgroup_dir_inode_operations
;
854 static const struct file_operations proc_cgroupstats_operations
;
856 static struct backing_dev_info cgroup_backing_dev_info
= {
858 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
861 static int alloc_css_id(struct cgroup_subsys
*ss
,
862 struct cgroup
*parent
, struct cgroup
*child
);
864 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
866 struct inode
*inode
= new_inode(sb
);
869 inode
->i_ino
= get_next_ino();
870 inode
->i_mode
= mode
;
871 inode
->i_uid
= current_fsuid();
872 inode
->i_gid
= current_fsgid();
873 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
874 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
879 static struct cgroup_name
*cgroup_alloc_name(struct dentry
*dentry
)
881 struct cgroup_name
*name
;
883 name
= kmalloc(sizeof(*name
) + dentry
->d_name
.len
+ 1, GFP_KERNEL
);
886 strcpy(name
->name
, dentry
->d_name
.name
);
890 static void cgroup_free_fn(struct work_struct
*work
)
892 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, free_work
);
893 struct cgroup_subsys
*ss
;
895 mutex_lock(&cgroup_mutex
);
897 * Release the subsystem state objects.
899 for_each_subsys(cgrp
->root
, ss
)
902 cgrp
->root
->number_of_cgroups
--;
903 mutex_unlock(&cgroup_mutex
);
906 * Drop the active superblock reference that we took when we
909 deactivate_super(cgrp
->root
->sb
);
912 * if we're getting rid of the cgroup, refcount should ensure
913 * that there are no pidlists left.
915 BUG_ON(!list_empty(&cgrp
->pidlists
));
917 simple_xattrs_free(&cgrp
->xattrs
);
919 ida_simple_remove(&cgrp
->root
->cgroup_ida
, cgrp
->id
);
920 kfree(rcu_dereference_raw(cgrp
->name
));
924 static void cgroup_free_rcu(struct rcu_head
*head
)
926 struct cgroup
*cgrp
= container_of(head
, struct cgroup
, rcu_head
);
928 schedule_work(&cgrp
->free_work
);
931 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
933 /* is dentry a directory ? if so, kfree() associated cgroup */
934 if (S_ISDIR(inode
->i_mode
)) {
935 struct cgroup
*cgrp
= dentry
->d_fsdata
;
937 BUG_ON(!(cgroup_is_removed(cgrp
)));
938 call_rcu(&cgrp
->rcu_head
, cgroup_free_rcu
);
940 struct cfent
*cfe
= __d_cfe(dentry
);
941 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
942 struct cftype
*cft
= cfe
->type
;
944 WARN_ONCE(!list_empty(&cfe
->node
) &&
945 cgrp
!= &cgrp
->root
->top_cgroup
,
946 "cfe still linked for %s\n", cfe
->type
->name
);
948 simple_xattrs_free(&cft
->xattrs
);
953 static int cgroup_delete(const struct dentry
*d
)
958 static void remove_dir(struct dentry
*d
)
960 struct dentry
*parent
= dget(d
->d_parent
);
963 simple_rmdir(parent
->d_inode
, d
);
967 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
971 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
972 lockdep_assert_held(&cgroup_mutex
);
975 * If we're doing cleanup due to failure of cgroup_create(),
976 * the corresponding @cfe may not exist.
978 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
979 struct dentry
*d
= cfe
->dentry
;
981 if (cft
&& cfe
->type
!= cft
)
986 simple_unlink(cgrp
->dentry
->d_inode
, d
);
987 list_del_init(&cfe
->node
);
995 * cgroup_clear_directory - selective removal of base and subsystem files
996 * @dir: directory containing the files
997 * @base_files: true if the base files should be removed
998 * @subsys_mask: mask of the subsystem ids whose files should be removed
1000 static void cgroup_clear_directory(struct dentry
*dir
, bool base_files
,
1001 unsigned long subsys_mask
)
1003 struct cgroup
*cgrp
= __d_cgrp(dir
);
1004 struct cgroup_subsys
*ss
;
1006 for_each_subsys(cgrp
->root
, ss
) {
1007 struct cftype_set
*set
;
1008 if (!test_bit(ss
->subsys_id
, &subsys_mask
))
1010 list_for_each_entry(set
, &ss
->cftsets
, node
)
1011 cgroup_addrm_files(cgrp
, NULL
, set
->cfts
, false);
1014 while (!list_empty(&cgrp
->files
))
1015 cgroup_rm_file(cgrp
, NULL
);
1020 * NOTE : the dentry must have been dget()'ed
1022 static void cgroup_d_remove_dir(struct dentry
*dentry
)
1024 struct dentry
*parent
;
1025 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1027 cgroup_clear_directory(dentry
, true, root
->subsys_mask
);
1029 parent
= dentry
->d_parent
;
1030 spin_lock(&parent
->d_lock
);
1031 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1032 list_del_init(&dentry
->d_u
.d_child
);
1033 spin_unlock(&dentry
->d_lock
);
1034 spin_unlock(&parent
->d_lock
);
1039 * Call with cgroup_mutex held. Drops reference counts on modules, including
1040 * any duplicate ones that parse_cgroupfs_options took. If this function
1041 * returns an error, no reference counts are touched.
1043 static int rebind_subsystems(struct cgroupfs_root
*root
,
1044 unsigned long final_subsys_mask
)
1046 unsigned long added_mask
, removed_mask
;
1047 struct cgroup
*cgrp
= &root
->top_cgroup
;
1050 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1051 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1053 removed_mask
= root
->actual_subsys_mask
& ~final_subsys_mask
;
1054 added_mask
= final_subsys_mask
& ~root
->actual_subsys_mask
;
1055 /* Check that any added subsystems are currently free */
1056 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1057 unsigned long bit
= 1UL << i
;
1058 struct cgroup_subsys
*ss
= subsys
[i
];
1059 if (!(bit
& added_mask
))
1062 * Nobody should tell us to do a subsys that doesn't exist:
1063 * parse_cgroupfs_options should catch that case and refcounts
1064 * ensure that subsystems won't disappear once selected.
1067 if (ss
->root
!= &rootnode
) {
1068 /* Subsystem isn't free */
1073 /* Currently we don't handle adding/removing subsystems when
1074 * any child cgroups exist. This is theoretically supportable
1075 * but involves complex error handling, so it's being left until
1077 if (root
->number_of_cgroups
> 1)
1080 /* Process each subsystem */
1081 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1082 struct cgroup_subsys
*ss
= subsys
[i
];
1083 unsigned long bit
= 1UL << i
;
1084 if (bit
& added_mask
) {
1085 /* We're binding this subsystem to this hierarchy */
1087 BUG_ON(cgrp
->subsys
[i
]);
1088 BUG_ON(!dummytop
->subsys
[i
]);
1089 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
1090 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
1091 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1092 list_move(&ss
->sibling
, &root
->subsys_list
);
1096 /* refcount was already taken, and we're keeping it */
1097 } else if (bit
& removed_mask
) {
1098 /* We're removing this subsystem */
1100 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1101 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1104 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1105 cgrp
->subsys
[i
] = NULL
;
1106 subsys
[i
]->root
= &rootnode
;
1107 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1108 /* subsystem is now free - drop reference on module */
1109 module_put(ss
->module
);
1110 } else if (bit
& final_subsys_mask
) {
1111 /* Subsystem state should already exist */
1113 BUG_ON(!cgrp
->subsys
[i
]);
1115 * a refcount was taken, but we already had one, so
1116 * drop the extra reference.
1118 module_put(ss
->module
);
1119 #ifdef CONFIG_MODULE_UNLOAD
1120 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1123 /* Subsystem state shouldn't exist */
1124 BUG_ON(cgrp
->subsys
[i
]);
1127 root
->subsys_mask
= root
->actual_subsys_mask
= final_subsys_mask
;
1132 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1134 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1135 struct cgroup_subsys
*ss
;
1137 mutex_lock(&cgroup_root_mutex
);
1138 for_each_subsys(root
, ss
)
1139 seq_printf(seq
, ",%s", ss
->name
);
1140 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1141 seq_puts(seq
, ",noprefix");
1142 if (test_bit(ROOT_XATTR
, &root
->flags
))
1143 seq_puts(seq
, ",xattr");
1144 if (strlen(root
->release_agent_path
))
1145 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1146 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
))
1147 seq_puts(seq
, ",clone_children");
1148 if (strlen(root
->name
))
1149 seq_printf(seq
, ",name=%s", root
->name
);
1150 mutex_unlock(&cgroup_root_mutex
);
1154 struct cgroup_sb_opts
{
1155 unsigned long subsys_mask
;
1156 unsigned long flags
;
1157 char *release_agent
;
1158 bool cpuset_clone_children
;
1160 /* User explicitly requested empty subsystem */
1163 struct cgroupfs_root
*new_root
;
1168 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1169 * with cgroup_mutex held to protect the subsys[] array. This function takes
1170 * refcounts on subsystems to be used, unless it returns error, in which case
1171 * no refcounts are taken.
1173 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1175 char *token
, *o
= data
;
1176 bool all_ss
= false, one_ss
= false;
1177 unsigned long mask
= (unsigned long)-1;
1179 bool module_pin_failed
= false;
1181 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1183 #ifdef CONFIG_CPUSETS
1184 mask
= ~(1UL << cpuset_subsys_id
);
1187 memset(opts
, 0, sizeof(*opts
));
1189 while ((token
= strsep(&o
, ",")) != NULL
) {
1192 if (!strcmp(token
, "none")) {
1193 /* Explicitly have no subsystems */
1197 if (!strcmp(token
, "all")) {
1198 /* Mutually exclusive option 'all' + subsystem name */
1204 if (!strcmp(token
, "noprefix")) {
1205 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1208 if (!strcmp(token
, "clone_children")) {
1209 opts
->cpuset_clone_children
= true;
1212 if (!strcmp(token
, "xattr")) {
1213 set_bit(ROOT_XATTR
, &opts
->flags
);
1216 if (!strncmp(token
, "release_agent=", 14)) {
1217 /* Specifying two release agents is forbidden */
1218 if (opts
->release_agent
)
1220 opts
->release_agent
=
1221 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1222 if (!opts
->release_agent
)
1226 if (!strncmp(token
, "name=", 5)) {
1227 const char *name
= token
+ 5;
1228 /* Can't specify an empty name */
1231 /* Must match [\w.-]+ */
1232 for (i
= 0; i
< strlen(name
); i
++) {
1236 if ((c
== '.') || (c
== '-') || (c
== '_'))
1240 /* Specifying two names is forbidden */
1243 opts
->name
= kstrndup(name
,
1244 MAX_CGROUP_ROOT_NAMELEN
- 1,
1252 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1253 struct cgroup_subsys
*ss
= subsys
[i
];
1256 if (strcmp(token
, ss
->name
))
1261 /* Mutually exclusive option 'all' + subsystem name */
1264 set_bit(i
, &opts
->subsys_mask
);
1269 if (i
== CGROUP_SUBSYS_COUNT
)
1274 * If the 'all' option was specified select all the subsystems,
1275 * otherwise if 'none', 'name=' and a subsystem name options
1276 * were not specified, let's default to 'all'
1278 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
)) {
1279 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1280 struct cgroup_subsys
*ss
= subsys
[i
];
1285 set_bit(i
, &opts
->subsys_mask
);
1289 /* Consistency checks */
1292 * Option noprefix was introduced just for backward compatibility
1293 * with the old cpuset, so we allow noprefix only if mounting just
1294 * the cpuset subsystem.
1296 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1297 (opts
->subsys_mask
& mask
))
1301 /* Can't specify "none" and some subsystems */
1302 if (opts
->subsys_mask
&& opts
->none
)
1306 * We either have to specify by name or by subsystems. (So all
1307 * empty hierarchies must have a name).
1309 if (!opts
->subsys_mask
&& !opts
->name
)
1313 * Grab references on all the modules we'll need, so the subsystems
1314 * don't dance around before rebind_subsystems attaches them. This may
1315 * take duplicate reference counts on a subsystem that's already used,
1316 * but rebind_subsystems handles this case.
1318 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1319 unsigned long bit
= 1UL << i
;
1321 if (!(bit
& opts
->subsys_mask
))
1323 if (!try_module_get(subsys
[i
]->module
)) {
1324 module_pin_failed
= true;
1328 if (module_pin_failed
) {
1330 * oops, one of the modules was going away. this means that we
1331 * raced with a module_delete call, and to the user this is
1332 * essentially a "subsystem doesn't exist" case.
1334 for (i
--; i
>= 0; i
--) {
1335 /* drop refcounts only on the ones we took */
1336 unsigned long bit
= 1UL << i
;
1338 if (!(bit
& opts
->subsys_mask
))
1340 module_put(subsys
[i
]->module
);
1348 static void drop_parsed_module_refcounts(unsigned long subsys_mask
)
1351 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1352 unsigned long bit
= 1UL << i
;
1354 if (!(bit
& subsys_mask
))
1356 module_put(subsys
[i
]->module
);
1360 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1363 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1364 struct cgroup
*cgrp
= &root
->top_cgroup
;
1365 struct cgroup_sb_opts opts
;
1366 unsigned long added_mask
, removed_mask
;
1368 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1369 mutex_lock(&cgroup_mutex
);
1370 mutex_lock(&cgroup_root_mutex
);
1372 /* See what subsystems are wanted */
1373 ret
= parse_cgroupfs_options(data
, &opts
);
1377 if (opts
.subsys_mask
!= root
->actual_subsys_mask
|| opts
.release_agent
)
1378 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1379 task_tgid_nr(current
), current
->comm
);
1381 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1382 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1384 /* Don't allow flags or name to change at remount */
1385 if (opts
.flags
!= root
->flags
||
1386 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1388 drop_parsed_module_refcounts(opts
.subsys_mask
);
1393 * Clear out the files of subsystems that should be removed, do
1394 * this before rebind_subsystems, since rebind_subsystems may
1395 * change this hierarchy's subsys_list.
1397 cgroup_clear_directory(cgrp
->dentry
, false, removed_mask
);
1399 ret
= rebind_subsystems(root
, opts
.subsys_mask
);
1401 /* rebind_subsystems failed, re-populate the removed files */
1402 cgroup_populate_dir(cgrp
, false, removed_mask
);
1403 drop_parsed_module_refcounts(opts
.subsys_mask
);
1407 /* re-populate subsystem files */
1408 cgroup_populate_dir(cgrp
, false, added_mask
);
1410 if (opts
.release_agent
)
1411 strcpy(root
->release_agent_path
, opts
.release_agent
);
1413 kfree(opts
.release_agent
);
1415 mutex_unlock(&cgroup_root_mutex
);
1416 mutex_unlock(&cgroup_mutex
);
1417 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1421 static const struct super_operations cgroup_ops
= {
1422 .statfs
= simple_statfs
,
1423 .drop_inode
= generic_delete_inode
,
1424 .show_options
= cgroup_show_options
,
1425 .remount_fs
= cgroup_remount
,
1428 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1430 INIT_LIST_HEAD(&cgrp
->sibling
);
1431 INIT_LIST_HEAD(&cgrp
->children
);
1432 INIT_LIST_HEAD(&cgrp
->files
);
1433 INIT_LIST_HEAD(&cgrp
->css_sets
);
1434 INIT_LIST_HEAD(&cgrp
->allcg_node
);
1435 INIT_LIST_HEAD(&cgrp
->release_list
);
1436 INIT_LIST_HEAD(&cgrp
->pidlists
);
1437 INIT_WORK(&cgrp
->free_work
, cgroup_free_fn
);
1438 mutex_init(&cgrp
->pidlist_mutex
);
1439 INIT_LIST_HEAD(&cgrp
->event_list
);
1440 spin_lock_init(&cgrp
->event_list_lock
);
1441 simple_xattrs_init(&cgrp
->xattrs
);
1444 static void init_cgroup_root(struct cgroupfs_root
*root
)
1446 struct cgroup
*cgrp
= &root
->top_cgroup
;
1448 INIT_LIST_HEAD(&root
->subsys_list
);
1449 INIT_LIST_HEAD(&root
->root_list
);
1450 INIT_LIST_HEAD(&root
->allcg_list
);
1451 root
->number_of_cgroups
= 1;
1453 cgrp
->name
= &root_cgroup_name
;
1454 cgrp
->top_cgroup
= cgrp
;
1455 init_cgroup_housekeeping(cgrp
);
1456 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
1459 static bool init_root_id(struct cgroupfs_root
*root
)
1464 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1466 spin_lock(&hierarchy_id_lock
);
1467 /* Try to allocate the next unused ID */
1468 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1469 &root
->hierarchy_id
);
1471 /* Try again starting from 0 */
1472 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1474 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1475 } else if (ret
!= -EAGAIN
) {
1476 /* Can only get here if the 31-bit IDR is full ... */
1479 spin_unlock(&hierarchy_id_lock
);
1484 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1486 struct cgroup_sb_opts
*opts
= data
;
1487 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1489 /* If we asked for a name then it must match */
1490 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1494 * If we asked for subsystems (or explicitly for no
1495 * subsystems) then they must match
1497 if ((opts
->subsys_mask
|| opts
->none
)
1498 && (opts
->subsys_mask
!= root
->subsys_mask
))
1504 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1506 struct cgroupfs_root
*root
;
1508 if (!opts
->subsys_mask
&& !opts
->none
)
1511 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1513 return ERR_PTR(-ENOMEM
);
1515 if (!init_root_id(root
)) {
1517 return ERR_PTR(-ENOMEM
);
1519 init_cgroup_root(root
);
1521 root
->subsys_mask
= opts
->subsys_mask
;
1522 root
->flags
= opts
->flags
;
1523 ida_init(&root
->cgroup_ida
);
1524 if (opts
->release_agent
)
1525 strcpy(root
->release_agent_path
, opts
->release_agent
);
1527 strcpy(root
->name
, opts
->name
);
1528 if (opts
->cpuset_clone_children
)
1529 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1533 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1538 BUG_ON(!root
->hierarchy_id
);
1539 spin_lock(&hierarchy_id_lock
);
1540 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1541 spin_unlock(&hierarchy_id_lock
);
1542 ida_destroy(&root
->cgroup_ida
);
1546 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1549 struct cgroup_sb_opts
*opts
= data
;
1551 /* If we don't have a new root, we can't set up a new sb */
1552 if (!opts
->new_root
)
1555 BUG_ON(!opts
->subsys_mask
&& !opts
->none
);
1557 ret
= set_anon_super(sb
, NULL
);
1561 sb
->s_fs_info
= opts
->new_root
;
1562 opts
->new_root
->sb
= sb
;
1564 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1565 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1566 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1567 sb
->s_op
= &cgroup_ops
;
1572 static int cgroup_get_rootdir(struct super_block
*sb
)
1574 static const struct dentry_operations cgroup_dops
= {
1575 .d_iput
= cgroup_diput
,
1576 .d_delete
= cgroup_delete
,
1579 struct inode
*inode
=
1580 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1585 inode
->i_fop
= &simple_dir_operations
;
1586 inode
->i_op
= &cgroup_dir_inode_operations
;
1587 /* directories start off with i_nlink == 2 (for "." entry) */
1589 sb
->s_root
= d_make_root(inode
);
1592 /* for everything else we want ->d_op set */
1593 sb
->s_d_op
= &cgroup_dops
;
1597 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1598 int flags
, const char *unused_dev_name
,
1601 struct cgroup_sb_opts opts
;
1602 struct cgroupfs_root
*root
;
1604 struct super_block
*sb
;
1605 struct cgroupfs_root
*new_root
;
1606 struct inode
*inode
;
1608 /* First find the desired set of subsystems */
1609 mutex_lock(&cgroup_mutex
);
1610 ret
= parse_cgroupfs_options(data
, &opts
);
1611 mutex_unlock(&cgroup_mutex
);
1616 * Allocate a new cgroup root. We may not need it if we're
1617 * reusing an existing hierarchy.
1619 new_root
= cgroup_root_from_opts(&opts
);
1620 if (IS_ERR(new_root
)) {
1621 ret
= PTR_ERR(new_root
);
1624 opts
.new_root
= new_root
;
1626 /* Locate an existing or new sb for this hierarchy */
1627 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, 0, &opts
);
1630 cgroup_drop_root(opts
.new_root
);
1634 root
= sb
->s_fs_info
;
1636 if (root
== opts
.new_root
) {
1637 /* We used the new root structure, so this is a new hierarchy */
1638 struct list_head tmp_cg_links
;
1639 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1640 struct cgroupfs_root
*existing_root
;
1641 const struct cred
*cred
;
1645 BUG_ON(sb
->s_root
!= NULL
);
1647 ret
= cgroup_get_rootdir(sb
);
1649 goto drop_new_super
;
1650 inode
= sb
->s_root
->d_inode
;
1652 mutex_lock(&inode
->i_mutex
);
1653 mutex_lock(&cgroup_mutex
);
1654 mutex_lock(&cgroup_root_mutex
);
1656 /* Check for name clashes with existing mounts */
1658 if (strlen(root
->name
))
1659 for_each_active_root(existing_root
)
1660 if (!strcmp(existing_root
->name
, root
->name
))
1664 * We're accessing css_set_count without locking
1665 * css_set_lock here, but that's OK - it can only be
1666 * increased by someone holding cgroup_lock, and
1667 * that's us. The worst that can happen is that we
1668 * have some link structures left over
1670 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1674 ret
= rebind_subsystems(root
, root
->subsys_mask
);
1675 if (ret
== -EBUSY
) {
1676 free_cg_links(&tmp_cg_links
);
1680 * There must be no failure case after here, since rebinding
1681 * takes care of subsystems' refcounts, which are explicitly
1682 * dropped in the failure exit path.
1685 /* EBUSY should be the only error here */
1688 list_add(&root
->root_list
, &roots
);
1691 sb
->s_root
->d_fsdata
= root_cgrp
;
1692 root
->top_cgroup
.dentry
= sb
->s_root
;
1694 /* Link the top cgroup in this hierarchy into all
1695 * the css_set objects */
1696 write_lock(&css_set_lock
);
1697 hash_for_each(css_set_table
, i
, cg
, hlist
)
1698 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1699 write_unlock(&css_set_lock
);
1701 free_cg_links(&tmp_cg_links
);
1703 BUG_ON(!list_empty(&root_cgrp
->children
));
1704 BUG_ON(root
->number_of_cgroups
!= 1);
1706 cred
= override_creds(&init_cred
);
1707 cgroup_populate_dir(root_cgrp
, true, root
->subsys_mask
);
1709 mutex_unlock(&cgroup_root_mutex
);
1710 mutex_unlock(&cgroup_mutex
);
1711 mutex_unlock(&inode
->i_mutex
);
1714 * We re-used an existing hierarchy - the new root (if
1715 * any) is not needed
1717 cgroup_drop_root(opts
.new_root
);
1718 /* no subsys rebinding, so refcounts don't change */
1719 drop_parsed_module_refcounts(opts
.subsys_mask
);
1722 kfree(opts
.release_agent
);
1724 return dget(sb
->s_root
);
1727 mutex_unlock(&cgroup_root_mutex
);
1728 mutex_unlock(&cgroup_mutex
);
1729 mutex_unlock(&inode
->i_mutex
);
1731 deactivate_locked_super(sb
);
1733 drop_parsed_module_refcounts(opts
.subsys_mask
);
1735 kfree(opts
.release_agent
);
1737 return ERR_PTR(ret
);
1740 static void cgroup_kill_sb(struct super_block
*sb
) {
1741 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1742 struct cgroup
*cgrp
= &root
->top_cgroup
;
1744 struct cg_cgroup_link
*link
;
1745 struct cg_cgroup_link
*saved_link
;
1749 BUG_ON(root
->number_of_cgroups
!= 1);
1750 BUG_ON(!list_empty(&cgrp
->children
));
1752 mutex_lock(&cgroup_mutex
);
1753 mutex_lock(&cgroup_root_mutex
);
1755 /* Rebind all subsystems back to the default hierarchy */
1756 ret
= rebind_subsystems(root
, 0);
1757 /* Shouldn't be able to fail ... */
1761 * Release all the links from css_sets to this hierarchy's
1764 write_lock(&css_set_lock
);
1766 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1768 list_del(&link
->cg_link_list
);
1769 list_del(&link
->cgrp_link_list
);
1772 write_unlock(&css_set_lock
);
1774 if (!list_empty(&root
->root_list
)) {
1775 list_del(&root
->root_list
);
1779 mutex_unlock(&cgroup_root_mutex
);
1780 mutex_unlock(&cgroup_mutex
);
1782 simple_xattrs_free(&cgrp
->xattrs
);
1784 kill_litter_super(sb
);
1785 cgroup_drop_root(root
);
1788 static struct file_system_type cgroup_fs_type
= {
1790 .mount
= cgroup_mount
,
1791 .kill_sb
= cgroup_kill_sb
,
1794 static struct kobject
*cgroup_kobj
;
1797 * cgroup_path - generate the path of a cgroup
1798 * @cgrp: the cgroup in question
1799 * @buf: the buffer to write the path into
1800 * @buflen: the length of the buffer
1802 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1804 * We can't generate cgroup path using dentry->d_name, as accessing
1805 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1806 * inode's i_mutex, while on the other hand cgroup_path() can be called
1807 * with some irq-safe spinlocks held.
1809 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1811 int ret
= -ENAMETOOLONG
;
1814 start
= buf
+ buflen
- 1;
1819 const char *name
= cgroup_name(cgrp
);
1823 if ((start
-= len
) < buf
)
1825 memcpy(start
, name
, len
);
1834 cgrp
= cgrp
->parent
;
1837 memmove(buf
, start
, buf
+ buflen
- start
);
1842 EXPORT_SYMBOL_GPL(cgroup_path
);
1845 * Control Group taskset
1847 struct task_and_cgroup
{
1848 struct task_struct
*task
;
1849 struct cgroup
*cgrp
;
1853 struct cgroup_taskset
{
1854 struct task_and_cgroup single
;
1855 struct flex_array
*tc_array
;
1858 struct cgroup
*cur_cgrp
;
1862 * cgroup_taskset_first - reset taskset and return the first task
1863 * @tset: taskset of interest
1865 * @tset iteration is initialized and the first task is returned.
1867 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1869 if (tset
->tc_array
) {
1871 return cgroup_taskset_next(tset
);
1873 tset
->cur_cgrp
= tset
->single
.cgrp
;
1874 return tset
->single
.task
;
1877 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1880 * cgroup_taskset_next - iterate to the next task in taskset
1881 * @tset: taskset of interest
1883 * Return the next task in @tset. Iteration must have been initialized
1884 * with cgroup_taskset_first().
1886 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1888 struct task_and_cgroup
*tc
;
1890 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1893 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1894 tset
->cur_cgrp
= tc
->cgrp
;
1897 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1900 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1901 * @tset: taskset of interest
1903 * Return the cgroup for the current (last returned) task of @tset. This
1904 * function must be preceded by either cgroup_taskset_first() or
1905 * cgroup_taskset_next().
1907 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1909 return tset
->cur_cgrp
;
1911 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1914 * cgroup_taskset_size - return the number of tasks in taskset
1915 * @tset: taskset of interest
1917 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1919 return tset
->tc_array
? tset
->tc_array_len
: 1;
1921 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1925 * cgroup_task_migrate - move a task from one cgroup to another.
1927 * Must be called with cgroup_mutex and threadgroup locked.
1929 static void cgroup_task_migrate(struct cgroup
*oldcgrp
,
1930 struct task_struct
*tsk
, struct css_set
*newcg
)
1932 struct css_set
*oldcg
;
1935 * We are synchronized through threadgroup_lock() against PF_EXITING
1936 * setting such that we can't race against cgroup_exit() changing the
1937 * css_set to init_css_set and dropping the old one.
1939 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1940 oldcg
= tsk
->cgroups
;
1943 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1946 /* Update the css_set linked lists if we're using them */
1947 write_lock(&css_set_lock
);
1948 if (!list_empty(&tsk
->cg_list
))
1949 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1950 write_unlock(&css_set_lock
);
1953 * We just gained a reference on oldcg by taking it from the task. As
1954 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1955 * it here; it will be freed under RCU.
1957 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1962 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1963 * @cgrp: the cgroup to attach to
1964 * @tsk: the task or the leader of the threadgroup to be attached
1965 * @threadgroup: attach the whole threadgroup?
1967 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1968 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1970 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
,
1973 int retval
, i
, group_size
;
1974 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1975 struct cgroupfs_root
*root
= cgrp
->root
;
1976 /* threadgroup list cursor and array */
1977 struct task_struct
*leader
= tsk
;
1978 struct task_and_cgroup
*tc
;
1979 struct flex_array
*group
;
1980 struct cgroup_taskset tset
= { };
1983 * step 0: in order to do expensive, possibly blocking operations for
1984 * every thread, we cannot iterate the thread group list, since it needs
1985 * rcu or tasklist locked. instead, build an array of all threads in the
1986 * group - group_rwsem prevents new threads from appearing, and if
1987 * threads exit, this will just be an over-estimate.
1990 group_size
= get_nr_threads(tsk
);
1993 /* flex_array supports very large thread-groups better than kmalloc. */
1994 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
1997 /* pre-allocate to guarantee space while iterating in rcu read-side. */
1998 retval
= flex_array_prealloc(group
, 0, group_size
, GFP_KERNEL
);
2000 goto out_free_group_list
;
2004 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2005 * already PF_EXITING could be freed from underneath us unless we
2006 * take an rcu_read_lock.
2010 struct task_and_cgroup ent
;
2012 /* @tsk either already exited or can't exit until the end */
2013 if (tsk
->flags
& PF_EXITING
)
2016 /* as per above, nr_threads may decrease, but not increase. */
2017 BUG_ON(i
>= group_size
);
2019 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2020 /* nothing to do if this task is already in the cgroup */
2021 if (ent
.cgrp
== cgrp
)
2024 * saying GFP_ATOMIC has no effect here because we did prealloc
2025 * earlier, but it's good form to communicate our expectations.
2027 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2028 BUG_ON(retval
!= 0);
2033 } while_each_thread(leader
, tsk
);
2035 /* remember the number of threads in the array for later. */
2037 tset
.tc_array
= group
;
2038 tset
.tc_array_len
= group_size
;
2040 /* methods shouldn't be called if no task is actually migrating */
2043 goto out_free_group_list
;
2046 * step 1: check that we can legitimately attach to the cgroup.
2048 for_each_subsys(root
, ss
) {
2049 if (ss
->can_attach
) {
2050 retval
= ss
->can_attach(cgrp
, &tset
);
2053 goto out_cancel_attach
;
2059 * step 2: make sure css_sets exist for all threads to be migrated.
2060 * we use find_css_set, which allocates a new one if necessary.
2062 for (i
= 0; i
< group_size
; i
++) {
2063 tc
= flex_array_get(group
, i
);
2064 tc
->cg
= find_css_set(tc
->task
->cgroups
, cgrp
);
2067 goto out_put_css_set_refs
;
2072 * step 3: now that we're guaranteed success wrt the css_sets,
2073 * proceed to move all tasks to the new cgroup. There are no
2074 * failure cases after here, so this is the commit point.
2076 for (i
= 0; i
< group_size
; i
++) {
2077 tc
= flex_array_get(group
, i
);
2078 cgroup_task_migrate(tc
->cgrp
, tc
->task
, tc
->cg
);
2080 /* nothing is sensitive to fork() after this point. */
2083 * step 4: do subsystem attach callbacks.
2085 for_each_subsys(root
, ss
) {
2087 ss
->attach(cgrp
, &tset
);
2091 * step 5: success! and cleanup
2094 out_put_css_set_refs
:
2096 for (i
= 0; i
< group_size
; i
++) {
2097 tc
= flex_array_get(group
, i
);
2100 put_css_set(tc
->cg
);
2105 for_each_subsys(root
, ss
) {
2106 if (ss
== failed_ss
)
2108 if (ss
->cancel_attach
)
2109 ss
->cancel_attach(cgrp
, &tset
);
2112 out_free_group_list
:
2113 flex_array_free(group
);
2118 * Find the task_struct of the task to attach by vpid and pass it along to the
2119 * function to attach either it or all tasks in its threadgroup. Will lock
2120 * cgroup_mutex and threadgroup; may take task_lock of task.
2122 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2124 struct task_struct
*tsk
;
2125 const struct cred
*cred
= current_cred(), *tcred
;
2128 if (!cgroup_lock_live_group(cgrp
))
2134 tsk
= find_task_by_vpid(pid
);
2138 goto out_unlock_cgroup
;
2141 * even if we're attaching all tasks in the thread group, we
2142 * only need to check permissions on one of them.
2144 tcred
= __task_cred(tsk
);
2145 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2146 !uid_eq(cred
->euid
, tcred
->uid
) &&
2147 !uid_eq(cred
->euid
, tcred
->suid
)) {
2150 goto out_unlock_cgroup
;
2156 tsk
= tsk
->group_leader
;
2159 * Workqueue threads may acquire PF_THREAD_BOUND and become
2160 * trapped in a cpuset, or RT worker may be born in a cgroup
2161 * with no rt_runtime allocated. Just say no.
2163 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_THREAD_BOUND
)) {
2166 goto out_unlock_cgroup
;
2169 get_task_struct(tsk
);
2172 threadgroup_lock(tsk
);
2174 if (!thread_group_leader(tsk
)) {
2176 * a race with de_thread from another thread's exec()
2177 * may strip us of our leadership, if this happens,
2178 * there is no choice but to throw this task away and
2179 * try again; this is
2180 * "double-double-toil-and-trouble-check locking".
2182 threadgroup_unlock(tsk
);
2183 put_task_struct(tsk
);
2184 goto retry_find_task
;
2188 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2190 threadgroup_unlock(tsk
);
2192 put_task_struct(tsk
);
2199 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2200 * @from: attach to all cgroups of a given task
2201 * @tsk: the task to be attached
2203 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2205 struct cgroupfs_root
*root
;
2209 for_each_active_root(root
) {
2210 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
2212 retval
= cgroup_attach_task(from_cg
, tsk
, false);
2220 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2222 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2224 return attach_task_by_pid(cgrp
, pid
, false);
2227 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2229 return attach_task_by_pid(cgrp
, tgid
, true);
2232 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2235 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2236 if (strlen(buffer
) >= PATH_MAX
)
2238 if (!cgroup_lock_live_group(cgrp
))
2240 mutex_lock(&cgroup_root_mutex
);
2241 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2242 mutex_unlock(&cgroup_root_mutex
);
2247 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2248 struct seq_file
*seq
)
2250 if (!cgroup_lock_live_group(cgrp
))
2252 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2253 seq_putc(seq
, '\n');
2258 /* A buffer size big enough for numbers or short strings */
2259 #define CGROUP_LOCAL_BUFFER_SIZE 64
2261 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2263 const char __user
*userbuf
,
2264 size_t nbytes
, loff_t
*unused_ppos
)
2266 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2272 if (nbytes
>= sizeof(buffer
))
2274 if (copy_from_user(buffer
, userbuf
, nbytes
))
2277 buffer
[nbytes
] = 0; /* nul-terminate */
2278 if (cft
->write_u64
) {
2279 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2282 retval
= cft
->write_u64(cgrp
, cft
, val
);
2284 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2287 retval
= cft
->write_s64(cgrp
, cft
, val
);
2294 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2296 const char __user
*userbuf
,
2297 size_t nbytes
, loff_t
*unused_ppos
)
2299 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2301 size_t max_bytes
= cft
->max_write_len
;
2302 char *buffer
= local_buffer
;
2305 max_bytes
= sizeof(local_buffer
) - 1;
2306 if (nbytes
>= max_bytes
)
2308 /* Allocate a dynamic buffer if we need one */
2309 if (nbytes
>= sizeof(local_buffer
)) {
2310 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2314 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2319 buffer
[nbytes
] = 0; /* nul-terminate */
2320 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2324 if (buffer
!= local_buffer
)
2329 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2330 size_t nbytes
, loff_t
*ppos
)
2332 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2333 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2335 if (cgroup_is_removed(cgrp
))
2338 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2339 if (cft
->write_u64
|| cft
->write_s64
)
2340 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2341 if (cft
->write_string
)
2342 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2344 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2345 return ret
? ret
: nbytes
;
2350 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2352 char __user
*buf
, size_t nbytes
,
2355 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2356 u64 val
= cft
->read_u64(cgrp
, cft
);
2357 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2359 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2362 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2364 char __user
*buf
, size_t nbytes
,
2367 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2368 s64 val
= cft
->read_s64(cgrp
, cft
);
2369 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2371 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2374 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2375 size_t nbytes
, loff_t
*ppos
)
2377 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2378 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2380 if (cgroup_is_removed(cgrp
))
2384 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2386 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2388 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2393 * seqfile ops/methods for returning structured data. Currently just
2394 * supports string->u64 maps, but can be extended in future.
2397 struct cgroup_seqfile_state
{
2399 struct cgroup
*cgroup
;
2402 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2404 struct seq_file
*sf
= cb
->state
;
2405 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2408 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2410 struct cgroup_seqfile_state
*state
= m
->private;
2411 struct cftype
*cft
= state
->cft
;
2412 if (cft
->read_map
) {
2413 struct cgroup_map_cb cb
= {
2414 .fill
= cgroup_map_add
,
2417 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2419 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2422 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2424 struct seq_file
*seq
= file
->private_data
;
2425 kfree(seq
->private);
2426 return single_release(inode
, file
);
2429 static const struct file_operations cgroup_seqfile_operations
= {
2431 .write
= cgroup_file_write
,
2432 .llseek
= seq_lseek
,
2433 .release
= cgroup_seqfile_release
,
2436 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2441 err
= generic_file_open(inode
, file
);
2444 cft
= __d_cft(file
->f_dentry
);
2446 if (cft
->read_map
|| cft
->read_seq_string
) {
2447 struct cgroup_seqfile_state
*state
=
2448 kzalloc(sizeof(*state
), GFP_USER
);
2452 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2453 file
->f_op
= &cgroup_seqfile_operations
;
2454 err
= single_open(file
, cgroup_seqfile_show
, state
);
2457 } else if (cft
->open
)
2458 err
= cft
->open(inode
, file
);
2465 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2467 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2469 return cft
->release(inode
, file
);
2474 * cgroup_rename - Only allow simple rename of directories in place.
2476 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2477 struct inode
*new_dir
, struct dentry
*new_dentry
)
2480 struct cgroup_name
*name
, *old_name
;
2481 struct cgroup
*cgrp
;
2484 * It's convinient to use parent dir's i_mutex to protected
2487 lockdep_assert_held(&old_dir
->i_mutex
);
2489 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2491 if (new_dentry
->d_inode
)
2493 if (old_dir
!= new_dir
)
2496 cgrp
= __d_cgrp(old_dentry
);
2498 name
= cgroup_alloc_name(new_dentry
);
2502 ret
= simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2508 old_name
= cgrp
->name
;
2509 rcu_assign_pointer(cgrp
->name
, name
);
2511 kfree_rcu(old_name
, rcu_head
);
2515 static struct simple_xattrs
*__d_xattrs(struct dentry
*dentry
)
2517 if (S_ISDIR(dentry
->d_inode
->i_mode
))
2518 return &__d_cgrp(dentry
)->xattrs
;
2520 return &__d_cft(dentry
)->xattrs
;
2523 static inline int xattr_enabled(struct dentry
*dentry
)
2525 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
2526 return test_bit(ROOT_XATTR
, &root
->flags
);
2529 static bool is_valid_xattr(const char *name
)
2531 if (!strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) ||
2532 !strncmp(name
, XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
))
2537 static int cgroup_setxattr(struct dentry
*dentry
, const char *name
,
2538 const void *val
, size_t size
, int flags
)
2540 if (!xattr_enabled(dentry
))
2542 if (!is_valid_xattr(name
))
2544 return simple_xattr_set(__d_xattrs(dentry
), name
, val
, size
, flags
);
2547 static int cgroup_removexattr(struct dentry
*dentry
, const char *name
)
2549 if (!xattr_enabled(dentry
))
2551 if (!is_valid_xattr(name
))
2553 return simple_xattr_remove(__d_xattrs(dentry
), name
);
2556 static ssize_t
cgroup_getxattr(struct dentry
*dentry
, const char *name
,
2557 void *buf
, size_t size
)
2559 if (!xattr_enabled(dentry
))
2561 if (!is_valid_xattr(name
))
2563 return simple_xattr_get(__d_xattrs(dentry
), name
, buf
, size
);
2566 static ssize_t
cgroup_listxattr(struct dentry
*dentry
, char *buf
, size_t size
)
2568 if (!xattr_enabled(dentry
))
2570 return simple_xattr_list(__d_xattrs(dentry
), buf
, size
);
2573 static const struct file_operations cgroup_file_operations
= {
2574 .read
= cgroup_file_read
,
2575 .write
= cgroup_file_write
,
2576 .llseek
= generic_file_llseek
,
2577 .open
= cgroup_file_open
,
2578 .release
= cgroup_file_release
,
2581 static const struct inode_operations cgroup_file_inode_operations
= {
2582 .setxattr
= cgroup_setxattr
,
2583 .getxattr
= cgroup_getxattr
,
2584 .listxattr
= cgroup_listxattr
,
2585 .removexattr
= cgroup_removexattr
,
2588 static const struct inode_operations cgroup_dir_inode_operations
= {
2589 .lookup
= cgroup_lookup
,
2590 .mkdir
= cgroup_mkdir
,
2591 .rmdir
= cgroup_rmdir
,
2592 .rename
= cgroup_rename
,
2593 .setxattr
= cgroup_setxattr
,
2594 .getxattr
= cgroup_getxattr
,
2595 .listxattr
= cgroup_listxattr
,
2596 .removexattr
= cgroup_removexattr
,
2599 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
2601 if (dentry
->d_name
.len
> NAME_MAX
)
2602 return ERR_PTR(-ENAMETOOLONG
);
2603 d_add(dentry
, NULL
);
2608 * Check if a file is a control file
2610 static inline struct cftype
*__file_cft(struct file
*file
)
2612 if (file_inode(file
)->i_fop
!= &cgroup_file_operations
)
2613 return ERR_PTR(-EINVAL
);
2614 return __d_cft(file
->f_dentry
);
2617 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2618 struct super_block
*sb
)
2620 struct inode
*inode
;
2624 if (dentry
->d_inode
)
2627 inode
= cgroup_new_inode(mode
, sb
);
2631 if (S_ISDIR(mode
)) {
2632 inode
->i_op
= &cgroup_dir_inode_operations
;
2633 inode
->i_fop
= &simple_dir_operations
;
2635 /* start off with i_nlink == 2 (for "." entry) */
2637 inc_nlink(dentry
->d_parent
->d_inode
);
2640 * Control reaches here with cgroup_mutex held.
2641 * @inode->i_mutex should nest outside cgroup_mutex but we
2642 * want to populate it immediately without releasing
2643 * cgroup_mutex. As @inode isn't visible to anyone else
2644 * yet, trylock will always succeed without affecting
2647 WARN_ON_ONCE(!mutex_trylock(&inode
->i_mutex
));
2648 } else if (S_ISREG(mode
)) {
2650 inode
->i_fop
= &cgroup_file_operations
;
2651 inode
->i_op
= &cgroup_file_inode_operations
;
2653 d_instantiate(dentry
, inode
);
2654 dget(dentry
); /* Extra count - pin the dentry in core */
2659 * cgroup_file_mode - deduce file mode of a control file
2660 * @cft: the control file in question
2662 * returns cft->mode if ->mode is not 0
2663 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2664 * returns S_IRUGO if it has only a read handler
2665 * returns S_IWUSR if it has only a write hander
2667 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2674 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2675 cft
->read_map
|| cft
->read_seq_string
)
2678 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2679 cft
->write_string
|| cft
->trigger
)
2685 static int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2688 struct dentry
*dir
= cgrp
->dentry
;
2689 struct cgroup
*parent
= __d_cgrp(dir
);
2690 struct dentry
*dentry
;
2694 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2696 simple_xattrs_init(&cft
->xattrs
);
2698 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2699 strcpy(name
, subsys
->name
);
2702 strcat(name
, cft
->name
);
2704 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2706 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2710 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2711 if (IS_ERR(dentry
)) {
2712 error
= PTR_ERR(dentry
);
2716 mode
= cgroup_file_mode(cft
);
2717 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2719 cfe
->type
= (void *)cft
;
2720 cfe
->dentry
= dentry
;
2721 dentry
->d_fsdata
= cfe
;
2722 list_add_tail(&cfe
->node
, &parent
->files
);
2731 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2732 struct cftype cfts
[], bool is_add
)
2737 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2738 /* does cft->flags tell us to skip this file on @cgrp? */
2739 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2741 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2745 err
= cgroup_add_file(cgrp
, subsys
, cft
);
2747 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2751 cgroup_rm_file(cgrp
, cft
);
2757 static DEFINE_MUTEX(cgroup_cft_mutex
);
2759 static void cgroup_cfts_prepare(void)
2760 __acquires(&cgroup_cft_mutex
) __acquires(&cgroup_mutex
)
2763 * Thanks to the entanglement with vfs inode locking, we can't walk
2764 * the existing cgroups under cgroup_mutex and create files.
2765 * Instead, we increment reference on all cgroups and build list of
2766 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2767 * exclusive access to the field.
2769 mutex_lock(&cgroup_cft_mutex
);
2770 mutex_lock(&cgroup_mutex
);
2773 static void cgroup_cfts_commit(struct cgroup_subsys
*ss
,
2774 struct cftype
*cfts
, bool is_add
)
2775 __releases(&cgroup_mutex
) __releases(&cgroup_cft_mutex
)
2778 struct cgroup
*cgrp
, *n
;
2780 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2781 if (cfts
&& ss
->root
!= &rootnode
) {
2782 list_for_each_entry(cgrp
, &ss
->root
->allcg_list
, allcg_node
) {
2784 list_add_tail(&cgrp
->cft_q_node
, &pending
);
2788 mutex_unlock(&cgroup_mutex
);
2791 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2792 * files for all cgroups which were created before.
2794 list_for_each_entry_safe(cgrp
, n
, &pending
, cft_q_node
) {
2795 struct inode
*inode
= cgrp
->dentry
->d_inode
;
2797 mutex_lock(&inode
->i_mutex
);
2798 mutex_lock(&cgroup_mutex
);
2799 if (!cgroup_is_removed(cgrp
))
2800 cgroup_addrm_files(cgrp
, ss
, cfts
, is_add
);
2801 mutex_unlock(&cgroup_mutex
);
2802 mutex_unlock(&inode
->i_mutex
);
2804 list_del_init(&cgrp
->cft_q_node
);
2808 mutex_unlock(&cgroup_cft_mutex
);
2812 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2813 * @ss: target cgroup subsystem
2814 * @cfts: zero-length name terminated array of cftypes
2816 * Register @cfts to @ss. Files described by @cfts are created for all
2817 * existing cgroups to which @ss is attached and all future cgroups will
2818 * have them too. This function can be called anytime whether @ss is
2821 * Returns 0 on successful registration, -errno on failure. Note that this
2822 * function currently returns 0 as long as @cfts registration is successful
2823 * even if some file creation attempts on existing cgroups fail.
2825 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2827 struct cftype_set
*set
;
2829 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2833 cgroup_cfts_prepare();
2835 list_add_tail(&set
->node
, &ss
->cftsets
);
2836 cgroup_cfts_commit(ss
, cfts
, true);
2840 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2843 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2844 * @ss: target cgroup subsystem
2845 * @cfts: zero-length name terminated array of cftypes
2847 * Unregister @cfts from @ss. Files described by @cfts are removed from
2848 * all existing cgroups to which @ss is attached and all future cgroups
2849 * won't have them either. This function can be called anytime whether @ss
2850 * is attached or not.
2852 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2853 * registered with @ss.
2855 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2857 struct cftype_set
*set
;
2859 cgroup_cfts_prepare();
2861 list_for_each_entry(set
, &ss
->cftsets
, node
) {
2862 if (set
->cfts
== cfts
) {
2863 list_del_init(&set
->node
);
2864 cgroup_cfts_commit(ss
, cfts
, false);
2869 cgroup_cfts_commit(ss
, NULL
, false);
2874 * cgroup_task_count - count the number of tasks in a cgroup.
2875 * @cgrp: the cgroup in question
2877 * Return the number of tasks in the cgroup.
2879 int cgroup_task_count(const struct cgroup
*cgrp
)
2882 struct cg_cgroup_link
*link
;
2884 read_lock(&css_set_lock
);
2885 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2886 count
+= atomic_read(&link
->cg
->refcount
);
2888 read_unlock(&css_set_lock
);
2893 * Advance a list_head iterator. The iterator should be positioned at
2894 * the start of a css_set
2896 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2897 struct cgroup_iter
*it
)
2899 struct list_head
*l
= it
->cg_link
;
2900 struct cg_cgroup_link
*link
;
2903 /* Advance to the next non-empty css_set */
2906 if (l
== &cgrp
->css_sets
) {
2910 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2912 } while (list_empty(&cg
->tasks
));
2914 it
->task
= cg
->tasks
.next
;
2918 * To reduce the fork() overhead for systems that are not actually
2919 * using their cgroups capability, we don't maintain the lists running
2920 * through each css_set to its tasks until we see the list actually
2921 * used - in other words after the first call to cgroup_iter_start().
2923 static void cgroup_enable_task_cg_lists(void)
2925 struct task_struct
*p
, *g
;
2926 write_lock(&css_set_lock
);
2927 use_task_css_set_links
= 1;
2929 * We need tasklist_lock because RCU is not safe against
2930 * while_each_thread(). Besides, a forking task that has passed
2931 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2932 * is not guaranteed to have its child immediately visible in the
2933 * tasklist if we walk through it with RCU.
2935 read_lock(&tasklist_lock
);
2936 do_each_thread(g
, p
) {
2939 * We should check if the process is exiting, otherwise
2940 * it will race with cgroup_exit() in that the list
2941 * entry won't be deleted though the process has exited.
2943 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2944 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2946 } while_each_thread(g
, p
);
2947 read_unlock(&tasklist_lock
);
2948 write_unlock(&css_set_lock
);
2952 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2953 * @pos: the current position (%NULL to initiate traversal)
2954 * @cgroup: cgroup whose descendants to walk
2956 * To be used by cgroup_for_each_descendant_pre(). Find the next
2957 * descendant to visit for pre-order traversal of @cgroup's descendants.
2959 struct cgroup
*cgroup_next_descendant_pre(struct cgroup
*pos
,
2960 struct cgroup
*cgroup
)
2962 struct cgroup
*next
;
2964 WARN_ON_ONCE(!rcu_read_lock_held());
2966 /* if first iteration, pretend we just visited @cgroup */
2968 if (list_empty(&cgroup
->children
))
2973 /* visit the first child if exists */
2974 next
= list_first_or_null_rcu(&pos
->children
, struct cgroup
, sibling
);
2978 /* no child, visit my or the closest ancestor's next sibling */
2980 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
,
2982 if (&next
->sibling
!= &pos
->parent
->children
)
2986 } while (pos
!= cgroup
);
2990 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre
);
2993 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
2994 * @pos: cgroup of interest
2996 * Return the rightmost descendant of @pos. If there's no descendant,
2997 * @pos is returned. This can be used during pre-order traversal to skip
3000 struct cgroup
*cgroup_rightmost_descendant(struct cgroup
*pos
)
3002 struct cgroup
*last
, *tmp
;
3004 WARN_ON_ONCE(!rcu_read_lock_held());
3008 /* ->prev isn't RCU safe, walk ->next till the end */
3010 list_for_each_entry_rcu(tmp
, &last
->children
, sibling
)
3016 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant
);
3018 static struct cgroup
*cgroup_leftmost_descendant(struct cgroup
*pos
)
3020 struct cgroup
*last
;
3024 pos
= list_first_or_null_rcu(&pos
->children
, struct cgroup
,
3032 * cgroup_next_descendant_post - find the next descendant for post-order walk
3033 * @pos: the current position (%NULL to initiate traversal)
3034 * @cgroup: cgroup whose descendants to walk
3036 * To be used by cgroup_for_each_descendant_post(). Find the next
3037 * descendant to visit for post-order traversal of @cgroup's descendants.
3039 struct cgroup
*cgroup_next_descendant_post(struct cgroup
*pos
,
3040 struct cgroup
*cgroup
)
3042 struct cgroup
*next
;
3044 WARN_ON_ONCE(!rcu_read_lock_held());
3046 /* if first iteration, visit the leftmost descendant */
3048 next
= cgroup_leftmost_descendant(cgroup
);
3049 return next
!= cgroup
? next
: NULL
;
3052 /* if there's an unvisited sibling, visit its leftmost descendant */
3053 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
, sibling
);
3054 if (&next
->sibling
!= &pos
->parent
->children
)
3055 return cgroup_leftmost_descendant(next
);
3057 /* no sibling left, visit parent */
3059 return next
!= cgroup
? next
: NULL
;
3061 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post
);
3063 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3064 __acquires(css_set_lock
)
3067 * The first time anyone tries to iterate across a cgroup,
3068 * we need to enable the list linking each css_set to its
3069 * tasks, and fix up all existing tasks.
3071 if (!use_task_css_set_links
)
3072 cgroup_enable_task_cg_lists();
3074 read_lock(&css_set_lock
);
3075 it
->cg_link
= &cgrp
->css_sets
;
3076 cgroup_advance_iter(cgrp
, it
);
3079 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
3080 struct cgroup_iter
*it
)
3082 struct task_struct
*res
;
3083 struct list_head
*l
= it
->task
;
3084 struct cg_cgroup_link
*link
;
3086 /* If the iterator cg is NULL, we have no tasks */
3089 res
= list_entry(l
, struct task_struct
, cg_list
);
3090 /* Advance iterator to find next entry */
3092 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
3093 if (l
== &link
->cg
->tasks
) {
3094 /* We reached the end of this task list - move on to
3095 * the next cg_cgroup_link */
3096 cgroup_advance_iter(cgrp
, it
);
3103 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3104 __releases(css_set_lock
)
3106 read_unlock(&css_set_lock
);
3109 static inline int started_after_time(struct task_struct
*t1
,
3110 struct timespec
*time
,
3111 struct task_struct
*t2
)
3113 int start_diff
= timespec_compare(&t1
->start_time
, time
);
3114 if (start_diff
> 0) {
3116 } else if (start_diff
< 0) {
3120 * Arbitrarily, if two processes started at the same
3121 * time, we'll say that the lower pointer value
3122 * started first. Note that t2 may have exited by now
3123 * so this may not be a valid pointer any longer, but
3124 * that's fine - it still serves to distinguish
3125 * between two tasks started (effectively) simultaneously.
3132 * This function is a callback from heap_insert() and is used to order
3134 * In this case we order the heap in descending task start time.
3136 static inline int started_after(void *p1
, void *p2
)
3138 struct task_struct
*t1
= p1
;
3139 struct task_struct
*t2
= p2
;
3140 return started_after_time(t1
, &t2
->start_time
, t2
);
3144 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3145 * @scan: struct cgroup_scanner containing arguments for the scan
3147 * Arguments include pointers to callback functions test_task() and
3149 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3150 * and if it returns true, call process_task() for it also.
3151 * The test_task pointer may be NULL, meaning always true (select all tasks).
3152 * Effectively duplicates cgroup_iter_{start,next,end}()
3153 * but does not lock css_set_lock for the call to process_task().
3154 * The struct cgroup_scanner may be embedded in any structure of the caller's
3156 * It is guaranteed that process_task() will act on every task that
3157 * is a member of the cgroup for the duration of this call. This
3158 * function may or may not call process_task() for tasks that exit
3159 * or move to a different cgroup during the call, or are forked or
3160 * move into the cgroup during the call.
3162 * Note that test_task() may be called with locks held, and may in some
3163 * situations be called multiple times for the same task, so it should
3165 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3166 * pre-allocated and will be used for heap operations (and its "gt" member will
3167 * be overwritten), else a temporary heap will be used (allocation of which
3168 * may cause this function to fail).
3170 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
3173 struct cgroup_iter it
;
3174 struct task_struct
*p
, *dropped
;
3175 /* Never dereference latest_task, since it's not refcounted */
3176 struct task_struct
*latest_task
= NULL
;
3177 struct ptr_heap tmp_heap
;
3178 struct ptr_heap
*heap
;
3179 struct timespec latest_time
= { 0, 0 };
3182 /* The caller supplied our heap and pre-allocated its memory */
3184 heap
->gt
= &started_after
;
3186 /* We need to allocate our own heap memory */
3188 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3190 /* cannot allocate the heap */
3196 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3197 * to determine which are of interest, and using the scanner's
3198 * "process_task" callback to process any of them that need an update.
3199 * Since we don't want to hold any locks during the task updates,
3200 * gather tasks to be processed in a heap structure.
3201 * The heap is sorted by descending task start time.
3202 * If the statically-sized heap fills up, we overflow tasks that
3203 * started later, and in future iterations only consider tasks that
3204 * started after the latest task in the previous pass. This
3205 * guarantees forward progress and that we don't miss any tasks.
3208 cgroup_iter_start(scan
->cg
, &it
);
3209 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
3211 * Only affect tasks that qualify per the caller's callback,
3212 * if he provided one
3214 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
3217 * Only process tasks that started after the last task
3220 if (!started_after_time(p
, &latest_time
, latest_task
))
3222 dropped
= heap_insert(heap
, p
);
3223 if (dropped
== NULL
) {
3225 * The new task was inserted; the heap wasn't
3229 } else if (dropped
!= p
) {
3231 * The new task was inserted, and pushed out a
3235 put_task_struct(dropped
);
3238 * Else the new task was newer than anything already in
3239 * the heap and wasn't inserted
3242 cgroup_iter_end(scan
->cg
, &it
);
3245 for (i
= 0; i
< heap
->size
; i
++) {
3246 struct task_struct
*q
= heap
->ptrs
[i
];
3248 latest_time
= q
->start_time
;
3251 /* Process the task per the caller's callback */
3252 scan
->process_task(q
, scan
);
3256 * If we had to process any tasks at all, scan again
3257 * in case some of them were in the middle of forking
3258 * children that didn't get processed.
3259 * Not the most efficient way to do it, but it avoids
3260 * having to take callback_mutex in the fork path
3264 if (heap
== &tmp_heap
)
3265 heap_free(&tmp_heap
);
3269 static void cgroup_transfer_one_task(struct task_struct
*task
,
3270 struct cgroup_scanner
*scan
)
3272 struct cgroup
*new_cgroup
= scan
->data
;
3275 cgroup_attach_task(new_cgroup
, task
, false);
3280 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3281 * @to: cgroup to which the tasks will be moved
3282 * @from: cgroup in which the tasks currently reside
3284 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3286 struct cgroup_scanner scan
;
3289 scan
.test_task
= NULL
; /* select all tasks in cgroup */
3290 scan
.process_task
= cgroup_transfer_one_task
;
3294 return cgroup_scan_tasks(&scan
);
3298 * Stuff for reading the 'tasks'/'procs' files.
3300 * Reading this file can return large amounts of data if a cgroup has
3301 * *lots* of attached tasks. So it may need several calls to read(),
3302 * but we cannot guarantee that the information we produce is correct
3303 * unless we produce it entirely atomically.
3307 /* which pidlist file are we talking about? */
3308 enum cgroup_filetype
{
3314 * A pidlist is a list of pids that virtually represents the contents of one
3315 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3316 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3319 struct cgroup_pidlist
{
3321 * used to find which pidlist is wanted. doesn't change as long as
3322 * this particular list stays in the list.
3324 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3327 /* how many elements the above list has */
3329 /* how many files are using the current array */
3331 /* each of these stored in a list by its cgroup */
3332 struct list_head links
;
3333 /* pointer to the cgroup we belong to, for list removal purposes */
3334 struct cgroup
*owner
;
3335 /* protects the other fields */
3336 struct rw_semaphore mutex
;
3340 * The following two functions "fix" the issue where there are more pids
3341 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3342 * TODO: replace with a kernel-wide solution to this problem
3344 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3345 static void *pidlist_allocate(int count
)
3347 if (PIDLIST_TOO_LARGE(count
))
3348 return vmalloc(count
* sizeof(pid_t
));
3350 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3352 static void pidlist_free(void *p
)
3354 if (is_vmalloc_addr(p
))
3361 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3362 * Returns the number of unique elements.
3364 static int pidlist_uniq(pid_t
*list
, int length
)
3369 * we presume the 0th element is unique, so i starts at 1. trivial
3370 * edge cases first; no work needs to be done for either
3372 if (length
== 0 || length
== 1)
3374 /* src and dest walk down the list; dest counts unique elements */
3375 for (src
= 1; src
< length
; src
++) {
3376 /* find next unique element */
3377 while (list
[src
] == list
[src
-1]) {
3382 /* dest always points to where the next unique element goes */
3383 list
[dest
] = list
[src
];
3390 static int cmppid(const void *a
, const void *b
)
3392 return *(pid_t
*)a
- *(pid_t
*)b
;
3396 * find the appropriate pidlist for our purpose (given procs vs tasks)
3397 * returns with the lock on that pidlist already held, and takes care
3398 * of the use count, or returns NULL with no locks held if we're out of
3401 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3402 enum cgroup_filetype type
)
3404 struct cgroup_pidlist
*l
;
3405 /* don't need task_nsproxy() if we're looking at ourself */
3406 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3409 * We can't drop the pidlist_mutex before taking the l->mutex in case
3410 * the last ref-holder is trying to remove l from the list at the same
3411 * time. Holding the pidlist_mutex precludes somebody taking whichever
3412 * list we find out from under us - compare release_pid_array().
3414 mutex_lock(&cgrp
->pidlist_mutex
);
3415 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3416 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3417 /* make sure l doesn't vanish out from under us */
3418 down_write(&l
->mutex
);
3419 mutex_unlock(&cgrp
->pidlist_mutex
);
3423 /* entry not found; create a new one */
3424 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3426 mutex_unlock(&cgrp
->pidlist_mutex
);
3429 init_rwsem(&l
->mutex
);
3430 down_write(&l
->mutex
);
3432 l
->key
.ns
= get_pid_ns(ns
);
3433 l
->use_count
= 0; /* don't increment here */
3436 list_add(&l
->links
, &cgrp
->pidlists
);
3437 mutex_unlock(&cgrp
->pidlist_mutex
);
3442 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3444 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3445 struct cgroup_pidlist
**lp
)
3449 int pid
, n
= 0; /* used for populating the array */
3450 struct cgroup_iter it
;
3451 struct task_struct
*tsk
;
3452 struct cgroup_pidlist
*l
;
3455 * If cgroup gets more users after we read count, we won't have
3456 * enough space - tough. This race is indistinguishable to the
3457 * caller from the case that the additional cgroup users didn't
3458 * show up until sometime later on.
3460 length
= cgroup_task_count(cgrp
);
3461 array
= pidlist_allocate(length
);
3464 /* now, populate the array */
3465 cgroup_iter_start(cgrp
, &it
);
3466 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3467 if (unlikely(n
== length
))
3469 /* get tgid or pid for procs or tasks file respectively */
3470 if (type
== CGROUP_FILE_PROCS
)
3471 pid
= task_tgid_vnr(tsk
);
3473 pid
= task_pid_vnr(tsk
);
3474 if (pid
> 0) /* make sure to only use valid results */
3477 cgroup_iter_end(cgrp
, &it
);
3479 /* now sort & (if procs) strip out duplicates */
3480 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3481 if (type
== CGROUP_FILE_PROCS
)
3482 length
= pidlist_uniq(array
, length
);
3483 l
= cgroup_pidlist_find(cgrp
, type
);
3485 pidlist_free(array
);
3488 /* store array, freeing old if necessary - lock already held */
3489 pidlist_free(l
->list
);
3493 up_write(&l
->mutex
);
3499 * cgroupstats_build - build and fill cgroupstats
3500 * @stats: cgroupstats to fill information into
3501 * @dentry: A dentry entry belonging to the cgroup for which stats have
3504 * Build and fill cgroupstats so that taskstats can export it to user
3507 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3510 struct cgroup
*cgrp
;
3511 struct cgroup_iter it
;
3512 struct task_struct
*tsk
;
3515 * Validate dentry by checking the superblock operations,
3516 * and make sure it's a directory.
3518 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3519 !S_ISDIR(dentry
->d_inode
->i_mode
))
3523 cgrp
= dentry
->d_fsdata
;
3525 cgroup_iter_start(cgrp
, &it
);
3526 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3527 switch (tsk
->state
) {
3529 stats
->nr_running
++;
3531 case TASK_INTERRUPTIBLE
:
3532 stats
->nr_sleeping
++;
3534 case TASK_UNINTERRUPTIBLE
:
3535 stats
->nr_uninterruptible
++;
3538 stats
->nr_stopped
++;
3541 if (delayacct_is_task_waiting_on_io(tsk
))
3542 stats
->nr_io_wait
++;
3546 cgroup_iter_end(cgrp
, &it
);
3554 * seq_file methods for the tasks/procs files. The seq_file position is the
3555 * next pid to display; the seq_file iterator is a pointer to the pid
3556 * in the cgroup->l->list array.
3559 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3562 * Initially we receive a position value that corresponds to
3563 * one more than the last pid shown (or 0 on the first call or
3564 * after a seek to the start). Use a binary-search to find the
3565 * next pid to display, if any
3567 struct cgroup_pidlist
*l
= s
->private;
3568 int index
= 0, pid
= *pos
;
3571 down_read(&l
->mutex
);
3573 int end
= l
->length
;
3575 while (index
< end
) {
3576 int mid
= (index
+ end
) / 2;
3577 if (l
->list
[mid
] == pid
) {
3580 } else if (l
->list
[mid
] <= pid
)
3586 /* If we're off the end of the array, we're done */
3587 if (index
>= l
->length
)
3589 /* Update the abstract position to be the actual pid that we found */
3590 iter
= l
->list
+ index
;
3595 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3597 struct cgroup_pidlist
*l
= s
->private;
3601 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3603 struct cgroup_pidlist
*l
= s
->private;
3605 pid_t
*end
= l
->list
+ l
->length
;
3607 * Advance to the next pid in the array. If this goes off the
3619 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3621 return seq_printf(s
, "%d\n", *(int *)v
);
3625 * seq_operations functions for iterating on pidlists through seq_file -
3626 * independent of whether it's tasks or procs
3628 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3629 .start
= cgroup_pidlist_start
,
3630 .stop
= cgroup_pidlist_stop
,
3631 .next
= cgroup_pidlist_next
,
3632 .show
= cgroup_pidlist_show
,
3635 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3638 * the case where we're the last user of this particular pidlist will
3639 * have us remove it from the cgroup's list, which entails taking the
3640 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3641 * pidlist_mutex, we have to take pidlist_mutex first.
3643 mutex_lock(&l
->owner
->pidlist_mutex
);
3644 down_write(&l
->mutex
);
3645 BUG_ON(!l
->use_count
);
3646 if (!--l
->use_count
) {
3647 /* we're the last user if refcount is 0; remove and free */
3648 list_del(&l
->links
);
3649 mutex_unlock(&l
->owner
->pidlist_mutex
);
3650 pidlist_free(l
->list
);
3651 put_pid_ns(l
->key
.ns
);
3652 up_write(&l
->mutex
);
3656 mutex_unlock(&l
->owner
->pidlist_mutex
);
3657 up_write(&l
->mutex
);
3660 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3662 struct cgroup_pidlist
*l
;
3663 if (!(file
->f_mode
& FMODE_READ
))
3666 * the seq_file will only be initialized if the file was opened for
3667 * reading; hence we check if it's not null only in that case.
3669 l
= ((struct seq_file
*)file
->private_data
)->private;
3670 cgroup_release_pid_array(l
);
3671 return seq_release(inode
, file
);
3674 static const struct file_operations cgroup_pidlist_operations
= {
3676 .llseek
= seq_lseek
,
3677 .write
= cgroup_file_write
,
3678 .release
= cgroup_pidlist_release
,
3682 * The following functions handle opens on a file that displays a pidlist
3683 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3686 /* helper function for the two below it */
3687 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3689 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3690 struct cgroup_pidlist
*l
;
3693 /* Nothing to do for write-only files */
3694 if (!(file
->f_mode
& FMODE_READ
))
3697 /* have the array populated */
3698 retval
= pidlist_array_load(cgrp
, type
, &l
);
3701 /* configure file information */
3702 file
->f_op
= &cgroup_pidlist_operations
;
3704 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3706 cgroup_release_pid_array(l
);
3709 ((struct seq_file
*)file
->private_data
)->private = l
;
3712 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3714 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3716 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3718 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3721 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3724 return notify_on_release(cgrp
);
3727 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3731 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3733 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3735 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3740 * Unregister event and free resources.
3742 * Gets called from workqueue.
3744 static void cgroup_event_remove(struct work_struct
*work
)
3746 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3748 struct cgroup
*cgrp
= event
->cgrp
;
3750 remove_wait_queue(event
->wqh
, &event
->wait
);
3752 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3754 /* Notify userspace the event is going away. */
3755 eventfd_signal(event
->eventfd
, 1);
3757 eventfd_ctx_put(event
->eventfd
);
3763 * Gets called on POLLHUP on eventfd when user closes it.
3765 * Called with wqh->lock held and interrupts disabled.
3767 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3768 int sync
, void *key
)
3770 struct cgroup_event
*event
= container_of(wait
,
3771 struct cgroup_event
, wait
);
3772 struct cgroup
*cgrp
= event
->cgrp
;
3773 unsigned long flags
= (unsigned long)key
;
3775 if (flags
& POLLHUP
) {
3777 * If the event has been detached at cgroup removal, we
3778 * can simply return knowing the other side will cleanup
3781 * We can't race against event freeing since the other
3782 * side will require wqh->lock via remove_wait_queue(),
3785 spin_lock(&cgrp
->event_list_lock
);
3786 if (!list_empty(&event
->list
)) {
3787 list_del_init(&event
->list
);
3789 * We are in atomic context, but cgroup_event_remove()
3790 * may sleep, so we have to call it in workqueue.
3792 schedule_work(&event
->remove
);
3794 spin_unlock(&cgrp
->event_list_lock
);
3800 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3801 wait_queue_head_t
*wqh
, poll_table
*pt
)
3803 struct cgroup_event
*event
= container_of(pt
,
3804 struct cgroup_event
, pt
);
3807 add_wait_queue(wqh
, &event
->wait
);
3811 * Parse input and register new cgroup event handler.
3813 * Input must be in format '<event_fd> <control_fd> <args>'.
3814 * Interpretation of args is defined by control file implementation.
3816 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3819 struct cgroup_event
*event
= NULL
;
3820 struct cgroup
*cgrp_cfile
;
3821 unsigned int efd
, cfd
;
3822 struct file
*efile
= NULL
;
3823 struct file
*cfile
= NULL
;
3827 efd
= simple_strtoul(buffer
, &endp
, 10);
3832 cfd
= simple_strtoul(buffer
, &endp
, 10);
3833 if ((*endp
!= ' ') && (*endp
!= '\0'))
3837 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3841 INIT_LIST_HEAD(&event
->list
);
3842 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3843 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3844 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3846 efile
= eventfd_fget(efd
);
3847 if (IS_ERR(efile
)) {
3848 ret
= PTR_ERR(efile
);
3852 event
->eventfd
= eventfd_ctx_fileget(efile
);
3853 if (IS_ERR(event
->eventfd
)) {
3854 ret
= PTR_ERR(event
->eventfd
);
3864 /* the process need read permission on control file */
3865 /* AV: shouldn't we check that it's been opened for read instead? */
3866 ret
= inode_permission(file_inode(cfile
), MAY_READ
);
3870 event
->cft
= __file_cft(cfile
);
3871 if (IS_ERR(event
->cft
)) {
3872 ret
= PTR_ERR(event
->cft
);
3877 * The file to be monitored must be in the same cgroup as
3878 * cgroup.event_control is.
3880 cgrp_cfile
= __d_cgrp(cfile
->f_dentry
->d_parent
);
3881 if (cgrp_cfile
!= cgrp
) {
3886 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3891 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3892 event
->eventfd
, buffer
);
3897 * Events should be removed after rmdir of cgroup directory, but before
3898 * destroying subsystem state objects. Let's take reference to cgroup
3899 * directory dentry to do that.
3903 spin_lock(&cgrp
->event_list_lock
);
3904 list_add(&event
->list
, &cgrp
->event_list
);
3905 spin_unlock(&cgrp
->event_list_lock
);
3916 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3917 eventfd_ctx_put(event
->eventfd
);
3919 if (!IS_ERR_OR_NULL(efile
))
3927 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3930 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
3933 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3938 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
3940 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
3945 * for the common functions, 'private' gives the type of file
3947 /* for hysterical raisins, we can't put this on the older files */
3948 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3949 static struct cftype files
[] = {
3952 .open
= cgroup_tasks_open
,
3953 .write_u64
= cgroup_tasks_write
,
3954 .release
= cgroup_pidlist_release
,
3955 .mode
= S_IRUGO
| S_IWUSR
,
3958 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3959 .open
= cgroup_procs_open
,
3960 .write_u64
= cgroup_procs_write
,
3961 .release
= cgroup_pidlist_release
,
3962 .mode
= S_IRUGO
| S_IWUSR
,
3965 .name
= "notify_on_release",
3966 .read_u64
= cgroup_read_notify_on_release
,
3967 .write_u64
= cgroup_write_notify_on_release
,
3970 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3971 .write_string
= cgroup_write_event_control
,
3975 .name
= "cgroup.clone_children",
3976 .read_u64
= cgroup_clone_children_read
,
3977 .write_u64
= cgroup_clone_children_write
,
3980 .name
= "release_agent",
3981 .flags
= CFTYPE_ONLY_ON_ROOT
,
3982 .read_seq_string
= cgroup_release_agent_show
,
3983 .write_string
= cgroup_release_agent_write
,
3984 .max_write_len
= PATH_MAX
,
3990 * cgroup_populate_dir - selectively creation of files in a directory
3991 * @cgrp: target cgroup
3992 * @base_files: true if the base files should be added
3993 * @subsys_mask: mask of the subsystem ids whose files should be added
3995 static int cgroup_populate_dir(struct cgroup
*cgrp
, bool base_files
,
3996 unsigned long subsys_mask
)
3999 struct cgroup_subsys
*ss
;
4002 err
= cgroup_addrm_files(cgrp
, NULL
, files
, true);
4007 /* process cftsets of each subsystem */
4008 for_each_subsys(cgrp
->root
, ss
) {
4009 struct cftype_set
*set
;
4010 if (!test_bit(ss
->subsys_id
, &subsys_mask
))
4013 list_for_each_entry(set
, &ss
->cftsets
, node
)
4014 cgroup_addrm_files(cgrp
, ss
, set
->cfts
, true);
4017 /* This cgroup is ready now */
4018 for_each_subsys(cgrp
->root
, ss
) {
4019 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4021 * Update id->css pointer and make this css visible from
4022 * CSS ID functions. This pointer will be dereferened
4023 * from RCU-read-side without locks.
4026 rcu_assign_pointer(css
->id
->css
, css
);
4032 static void css_dput_fn(struct work_struct
*work
)
4034 struct cgroup_subsys_state
*css
=
4035 container_of(work
, struct cgroup_subsys_state
, dput_work
);
4036 struct dentry
*dentry
= css
->cgroup
->dentry
;
4037 struct super_block
*sb
= dentry
->d_sb
;
4039 atomic_inc(&sb
->s_active
);
4041 deactivate_super(sb
);
4044 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
4045 struct cgroup_subsys
*ss
,
4046 struct cgroup
*cgrp
)
4049 atomic_set(&css
->refcnt
, 1);
4052 if (cgrp
== dummytop
)
4053 css
->flags
|= CSS_ROOT
;
4054 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
4055 cgrp
->subsys
[ss
->subsys_id
] = css
;
4058 * css holds an extra ref to @cgrp->dentry which is put on the last
4059 * css_put(). dput() requires process context, which css_put() may
4060 * be called without. @css->dput_work will be used to invoke
4061 * dput() asynchronously from css_put().
4063 INIT_WORK(&css
->dput_work
, css_dput_fn
);
4066 /* invoke ->post_create() on a new CSS and mark it online if successful */
4067 static int online_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4071 lockdep_assert_held(&cgroup_mutex
);
4074 ret
= ss
->css_online(cgrp
);
4076 cgrp
->subsys
[ss
->subsys_id
]->flags
|= CSS_ONLINE
;
4080 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4081 static void offline_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4082 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4084 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4086 lockdep_assert_held(&cgroup_mutex
);
4088 if (!(css
->flags
& CSS_ONLINE
))
4091 if (ss
->css_offline
)
4092 ss
->css_offline(cgrp
);
4094 cgrp
->subsys
[ss
->subsys_id
]->flags
&= ~CSS_ONLINE
;
4098 * cgroup_create - create a cgroup
4099 * @parent: cgroup that will be parent of the new cgroup
4100 * @dentry: dentry of the new cgroup
4101 * @mode: mode to set on new inode
4103 * Must be called with the mutex on the parent inode held
4105 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
4108 struct cgroup
*cgrp
;
4109 struct cgroup_name
*name
;
4110 struct cgroupfs_root
*root
= parent
->root
;
4112 struct cgroup_subsys
*ss
;
4113 struct super_block
*sb
= root
->sb
;
4115 /* allocate the cgroup and its ID, 0 is reserved for the root */
4116 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4120 name
= cgroup_alloc_name(dentry
);
4123 rcu_assign_pointer(cgrp
->name
, name
);
4125 cgrp
->id
= ida_simple_get(&root
->cgroup_ida
, 1, 0, GFP_KERNEL
);
4130 * Only live parents can have children. Note that the liveliness
4131 * check isn't strictly necessary because cgroup_mkdir() and
4132 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4133 * anyway so that locking is contained inside cgroup proper and we
4134 * don't get nasty surprises if we ever grow another caller.
4136 if (!cgroup_lock_live_group(parent
)) {
4141 /* Grab a reference on the superblock so the hierarchy doesn't
4142 * get deleted on unmount if there are child cgroups. This
4143 * can be done outside cgroup_mutex, since the sb can't
4144 * disappear while someone has an open control file on the
4146 atomic_inc(&sb
->s_active
);
4148 init_cgroup_housekeeping(cgrp
);
4150 dentry
->d_fsdata
= cgrp
;
4151 cgrp
->dentry
= dentry
;
4153 cgrp
->parent
= parent
;
4154 cgrp
->root
= parent
->root
;
4155 cgrp
->top_cgroup
= parent
->top_cgroup
;
4157 if (notify_on_release(parent
))
4158 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4160 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4161 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4163 for_each_subsys(root
, ss
) {
4164 struct cgroup_subsys_state
*css
;
4166 css
= ss
->css_alloc(cgrp
);
4171 init_cgroup_css(css
, ss
, cgrp
);
4173 err
= alloc_css_id(ss
, parent
, cgrp
);
4180 * Create directory. cgroup_create_file() returns with the new
4181 * directory locked on success so that it can be populated without
4182 * dropping cgroup_mutex.
4184 err
= cgroup_create_file(dentry
, S_IFDIR
| mode
, sb
);
4187 lockdep_assert_held(&dentry
->d_inode
->i_mutex
);
4189 /* allocation complete, commit to creation */
4190 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
4191 list_add_tail_rcu(&cgrp
->sibling
, &cgrp
->parent
->children
);
4192 root
->number_of_cgroups
++;
4194 /* each css holds a ref to the cgroup's dentry */
4195 for_each_subsys(root
, ss
)
4198 /* creation succeeded, notify subsystems */
4199 for_each_subsys(root
, ss
) {
4200 err
= online_css(ss
, cgrp
);
4204 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4206 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",
4207 current
->comm
, current
->pid
, ss
->name
);
4208 if (!strcmp(ss
->name
, "memory"))
4209 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4210 ss
->warned_broken_hierarchy
= true;
4214 err
= cgroup_populate_dir(cgrp
, true, root
->subsys_mask
);
4218 mutex_unlock(&cgroup_mutex
);
4219 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4224 for_each_subsys(root
, ss
) {
4225 if (cgrp
->subsys
[ss
->subsys_id
])
4228 mutex_unlock(&cgroup_mutex
);
4229 /* Release the reference count that we took on the superblock */
4230 deactivate_super(sb
);
4232 ida_simple_remove(&root
->cgroup_ida
, cgrp
->id
);
4234 kfree(rcu_dereference_raw(cgrp
->name
));
4240 cgroup_destroy_locked(cgrp
);
4241 mutex_unlock(&cgroup_mutex
);
4242 mutex_unlock(&dentry
->d_inode
->i_mutex
);
4246 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4248 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4250 /* the vfs holds inode->i_mutex already */
4251 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4254 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4255 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4257 struct dentry
*d
= cgrp
->dentry
;
4258 struct cgroup
*parent
= cgrp
->parent
;
4259 struct cgroup_event
*event
, *tmp
;
4260 struct cgroup_subsys
*ss
;
4262 lockdep_assert_held(&d
->d_inode
->i_mutex
);
4263 lockdep_assert_held(&cgroup_mutex
);
4265 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
))
4269 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4270 * removed. This makes future css_tryget() and child creation
4271 * attempts fail thus maintaining the removal conditions verified
4274 for_each_subsys(cgrp
->root
, ss
) {
4275 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4277 WARN_ON(atomic_read(&css
->refcnt
) < 0);
4278 atomic_add(CSS_DEACT_BIAS
, &css
->refcnt
);
4280 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
4282 /* tell subsystems to initate destruction */
4283 for_each_subsys(cgrp
->root
, ss
)
4284 offline_css(ss
, cgrp
);
4287 * Put all the base refs. Each css holds an extra reference to the
4288 * cgroup's dentry and cgroup removal proceeds regardless of css
4289 * refs. On the last put of each css, whenever that may be, the
4290 * extra dentry ref is put so that dentry destruction happens only
4291 * after all css's are released.
4293 for_each_subsys(cgrp
->root
, ss
)
4294 css_put(cgrp
->subsys
[ss
->subsys_id
]);
4296 raw_spin_lock(&release_list_lock
);
4297 if (!list_empty(&cgrp
->release_list
))
4298 list_del_init(&cgrp
->release_list
);
4299 raw_spin_unlock(&release_list_lock
);
4301 /* delete this cgroup from parent->children */
4302 list_del_rcu(&cgrp
->sibling
);
4303 list_del_init(&cgrp
->allcg_node
);
4306 cgroup_d_remove_dir(d
);
4309 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4310 check_for_release(parent
);
4313 * Unregister events and notify userspace.
4314 * Notify userspace about cgroup removing only after rmdir of cgroup
4315 * directory to avoid race between userspace and kernelspace.
4317 spin_lock(&cgrp
->event_list_lock
);
4318 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4319 list_del_init(&event
->list
);
4320 schedule_work(&event
->remove
);
4322 spin_unlock(&cgrp
->event_list_lock
);
4327 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4331 mutex_lock(&cgroup_mutex
);
4332 ret
= cgroup_destroy_locked(dentry
->d_fsdata
);
4333 mutex_unlock(&cgroup_mutex
);
4338 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4340 INIT_LIST_HEAD(&ss
->cftsets
);
4343 * base_cftset is embedded in subsys itself, no need to worry about
4346 if (ss
->base_cftypes
) {
4347 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4348 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4352 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4354 struct cgroup_subsys_state
*css
;
4356 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4358 mutex_lock(&cgroup_mutex
);
4360 /* init base cftset */
4361 cgroup_init_cftsets(ss
);
4363 /* Create the top cgroup state for this subsystem */
4364 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4365 ss
->root
= &rootnode
;
4366 css
= ss
->css_alloc(dummytop
);
4367 /* We don't handle early failures gracefully */
4368 BUG_ON(IS_ERR(css
));
4369 init_cgroup_css(css
, ss
, dummytop
);
4371 /* Update the init_css_set to contain a subsys
4372 * pointer to this state - since the subsystem is
4373 * newly registered, all tasks and hence the
4374 * init_css_set is in the subsystem's top cgroup. */
4375 init_css_set
.subsys
[ss
->subsys_id
] = css
;
4377 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4379 /* At system boot, before all subsystems have been
4380 * registered, no tasks have been forked, so we don't
4381 * need to invoke fork callbacks here. */
4382 BUG_ON(!list_empty(&init_task
.tasks
));
4385 BUG_ON(online_css(ss
, dummytop
));
4387 mutex_unlock(&cgroup_mutex
);
4389 /* this function shouldn't be used with modular subsystems, since they
4390 * need to register a subsys_id, among other things */
4395 * cgroup_load_subsys: load and register a modular subsystem at runtime
4396 * @ss: the subsystem to load
4398 * This function should be called in a modular subsystem's initcall. If the
4399 * subsystem is built as a module, it will be assigned a new subsys_id and set
4400 * up for use. If the subsystem is built-in anyway, work is delegated to the
4401 * simpler cgroup_init_subsys.
4403 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4405 struct cgroup_subsys_state
*css
;
4407 struct hlist_node
*tmp
;
4411 /* check name and function validity */
4412 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4413 ss
->css_alloc
== NULL
|| ss
->css_free
== NULL
)
4417 * we don't support callbacks in modular subsystems. this check is
4418 * before the ss->module check for consistency; a subsystem that could
4419 * be a module should still have no callbacks even if the user isn't
4420 * compiling it as one.
4422 if (ss
->fork
|| ss
->exit
)
4426 * an optionally modular subsystem is built-in: we want to do nothing,
4427 * since cgroup_init_subsys will have already taken care of it.
4429 if (ss
->module
== NULL
) {
4430 /* a sanity check */
4431 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4435 /* init base cftset */
4436 cgroup_init_cftsets(ss
);
4438 mutex_lock(&cgroup_mutex
);
4439 subsys
[ss
->subsys_id
] = ss
;
4442 * no ss->css_alloc seems to need anything important in the ss
4443 * struct, so this can happen first (i.e. before the rootnode
4446 css
= ss
->css_alloc(dummytop
);
4448 /* failure case - need to deassign the subsys[] slot. */
4449 subsys
[ss
->subsys_id
] = NULL
;
4450 mutex_unlock(&cgroup_mutex
);
4451 return PTR_ERR(css
);
4454 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4455 ss
->root
= &rootnode
;
4457 /* our new subsystem will be attached to the dummy hierarchy. */
4458 init_cgroup_css(css
, ss
, dummytop
);
4459 /* init_idr must be after init_cgroup_css because it sets css->id. */
4461 ret
= cgroup_init_idr(ss
, css
);
4467 * Now we need to entangle the css into the existing css_sets. unlike
4468 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4469 * will need a new pointer to it; done by iterating the css_set_table.
4470 * furthermore, modifying the existing css_sets will corrupt the hash
4471 * table state, so each changed css_set will need its hash recomputed.
4472 * this is all done under the css_set_lock.
4474 write_lock(&css_set_lock
);
4475 hash_for_each_safe(css_set_table
, i
, tmp
, cg
, hlist
) {
4476 /* skip entries that we already rehashed */
4477 if (cg
->subsys
[ss
->subsys_id
])
4479 /* remove existing entry */
4480 hash_del(&cg
->hlist
);
4482 cg
->subsys
[ss
->subsys_id
] = css
;
4483 /* recompute hash and restore entry */
4484 key
= css_set_hash(cg
->subsys
);
4485 hash_add(css_set_table
, &cg
->hlist
, key
);
4487 write_unlock(&css_set_lock
);
4490 ret
= online_css(ss
, dummytop
);
4495 mutex_unlock(&cgroup_mutex
);
4499 mutex_unlock(&cgroup_mutex
);
4500 /* @ss can't be mounted here as try_module_get() would fail */
4501 cgroup_unload_subsys(ss
);
4504 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4507 * cgroup_unload_subsys: unload a modular subsystem
4508 * @ss: the subsystem to unload
4510 * This function should be called in a modular subsystem's exitcall. When this
4511 * function is invoked, the refcount on the subsystem's module will be 0, so
4512 * the subsystem will not be attached to any hierarchy.
4514 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4516 struct cg_cgroup_link
*link
;
4518 BUG_ON(ss
->module
== NULL
);
4521 * we shouldn't be called if the subsystem is in use, and the use of
4522 * try_module_get in parse_cgroupfs_options should ensure that it
4523 * doesn't start being used while we're killing it off.
4525 BUG_ON(ss
->root
!= &rootnode
);
4527 mutex_lock(&cgroup_mutex
);
4529 offline_css(ss
, dummytop
);
4533 idr_destroy(&ss
->idr
);
4535 /* deassign the subsys_id */
4536 subsys
[ss
->subsys_id
] = NULL
;
4538 /* remove subsystem from rootnode's list of subsystems */
4539 list_del_init(&ss
->sibling
);
4542 * disentangle the css from all css_sets attached to the dummytop. as
4543 * in loading, we need to pay our respects to the hashtable gods.
4545 write_lock(&css_set_lock
);
4546 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4547 struct css_set
*cg
= link
->cg
;
4550 hash_del(&cg
->hlist
);
4551 cg
->subsys
[ss
->subsys_id
] = NULL
;
4552 key
= css_set_hash(cg
->subsys
);
4553 hash_add(css_set_table
, &cg
->hlist
, key
);
4555 write_unlock(&css_set_lock
);
4558 * remove subsystem's css from the dummytop and free it - need to
4559 * free before marking as null because ss->css_free needs the
4560 * cgrp->subsys pointer to find their state. note that this also
4561 * takes care of freeing the css_id.
4563 ss
->css_free(dummytop
);
4564 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4566 mutex_unlock(&cgroup_mutex
);
4568 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4571 * cgroup_init_early - cgroup initialization at system boot
4573 * Initialize cgroups at system boot, and initialize any
4574 * subsystems that request early init.
4576 int __init
cgroup_init_early(void)
4579 atomic_set(&init_css_set
.refcount
, 1);
4580 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4581 INIT_LIST_HEAD(&init_css_set
.tasks
);
4582 INIT_HLIST_NODE(&init_css_set
.hlist
);
4584 init_cgroup_root(&rootnode
);
4586 init_task
.cgroups
= &init_css_set
;
4588 init_css_set_link
.cg
= &init_css_set
;
4589 init_css_set_link
.cgrp
= dummytop
;
4590 list_add(&init_css_set_link
.cgrp_link_list
,
4591 &rootnode
.top_cgroup
.css_sets
);
4592 list_add(&init_css_set_link
.cg_link_list
,
4593 &init_css_set
.cg_links
);
4595 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4596 struct cgroup_subsys
*ss
= subsys
[i
];
4598 /* at bootup time, we don't worry about modular subsystems */
4599 if (!ss
|| ss
->module
)
4603 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4604 BUG_ON(!ss
->css_alloc
);
4605 BUG_ON(!ss
->css_free
);
4606 if (ss
->subsys_id
!= i
) {
4607 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4608 ss
->name
, ss
->subsys_id
);
4613 cgroup_init_subsys(ss
);
4619 * cgroup_init - cgroup initialization
4621 * Register cgroup filesystem and /proc file, and initialize
4622 * any subsystems that didn't request early init.
4624 int __init
cgroup_init(void)
4630 err
= bdi_init(&cgroup_backing_dev_info
);
4634 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4635 struct cgroup_subsys
*ss
= subsys
[i
];
4637 /* at bootup time, we don't worry about modular subsystems */
4638 if (!ss
|| ss
->module
)
4640 if (!ss
->early_init
)
4641 cgroup_init_subsys(ss
);
4643 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4646 /* Add init_css_set to the hash table */
4647 key
= css_set_hash(init_css_set
.subsys
);
4648 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4649 BUG_ON(!init_root_id(&rootnode
));
4651 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4657 err
= register_filesystem(&cgroup_fs_type
);
4659 kobject_put(cgroup_kobj
);
4663 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4667 bdi_destroy(&cgroup_backing_dev_info
);
4673 * proc_cgroup_show()
4674 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4675 * - Used for /proc/<pid>/cgroup.
4676 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4677 * doesn't really matter if tsk->cgroup changes after we read it,
4678 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4679 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4680 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4681 * cgroup to top_cgroup.
4684 /* TODO: Use a proper seq_file iterator */
4685 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4688 struct task_struct
*tsk
;
4691 struct cgroupfs_root
*root
;
4694 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4700 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4706 mutex_lock(&cgroup_mutex
);
4708 for_each_active_root(root
) {
4709 struct cgroup_subsys
*ss
;
4710 struct cgroup
*cgrp
;
4713 seq_printf(m
, "%d:", root
->hierarchy_id
);
4714 for_each_subsys(root
, ss
)
4715 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4716 if (strlen(root
->name
))
4717 seq_printf(m
, "%sname=%s", count
? "," : "",
4720 cgrp
= task_cgroup_from_root(tsk
, root
);
4721 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4729 mutex_unlock(&cgroup_mutex
);
4730 put_task_struct(tsk
);
4737 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4739 struct pid
*pid
= PROC_I(inode
)->pid
;
4740 return single_open(file
, proc_cgroup_show
, pid
);
4743 const struct file_operations proc_cgroup_operations
= {
4744 .open
= cgroup_open
,
4746 .llseek
= seq_lseek
,
4747 .release
= single_release
,
4750 /* Display information about each subsystem and each hierarchy */
4751 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4755 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4757 * ideally we don't want subsystems moving around while we do this.
4758 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4759 * subsys/hierarchy state.
4761 mutex_lock(&cgroup_mutex
);
4762 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4763 struct cgroup_subsys
*ss
= subsys
[i
];
4766 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4767 ss
->name
, ss
->root
->hierarchy_id
,
4768 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4770 mutex_unlock(&cgroup_mutex
);
4774 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4776 return single_open(file
, proc_cgroupstats_show
, NULL
);
4779 static const struct file_operations proc_cgroupstats_operations
= {
4780 .open
= cgroupstats_open
,
4782 .llseek
= seq_lseek
,
4783 .release
= single_release
,
4787 * cgroup_fork - attach newly forked task to its parents cgroup.
4788 * @child: pointer to task_struct of forking parent process.
4790 * Description: A task inherits its parent's cgroup at fork().
4792 * A pointer to the shared css_set was automatically copied in
4793 * fork.c by dup_task_struct(). However, we ignore that copy, since
4794 * it was not made under the protection of RCU or cgroup_mutex, so
4795 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4796 * have already changed current->cgroups, allowing the previously
4797 * referenced cgroup group to be removed and freed.
4799 * At the point that cgroup_fork() is called, 'current' is the parent
4800 * task, and the passed argument 'child' points to the child task.
4802 void cgroup_fork(struct task_struct
*child
)
4805 child
->cgroups
= current
->cgroups
;
4806 get_css_set(child
->cgroups
);
4807 task_unlock(current
);
4808 INIT_LIST_HEAD(&child
->cg_list
);
4812 * cgroup_post_fork - called on a new task after adding it to the task list
4813 * @child: the task in question
4815 * Adds the task to the list running through its css_set if necessary and
4816 * call the subsystem fork() callbacks. Has to be after the task is
4817 * visible on the task list in case we race with the first call to
4818 * cgroup_iter_start() - to guarantee that the new task ends up on its
4821 void cgroup_post_fork(struct task_struct
*child
)
4826 * use_task_css_set_links is set to 1 before we walk the tasklist
4827 * under the tasklist_lock and we read it here after we added the child
4828 * to the tasklist under the tasklist_lock as well. If the child wasn't
4829 * yet in the tasklist when we walked through it from
4830 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4831 * should be visible now due to the paired locking and barriers implied
4832 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4833 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4836 if (use_task_css_set_links
) {
4837 write_lock(&css_set_lock
);
4839 if (list_empty(&child
->cg_list
))
4840 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4842 write_unlock(&css_set_lock
);
4846 * Call ss->fork(). This must happen after @child is linked on
4847 * css_set; otherwise, @child might change state between ->fork()
4848 * and addition to css_set.
4850 if (need_forkexit_callback
) {
4852 * fork/exit callbacks are supported only for builtin
4853 * subsystems, and the builtin section of the subsys
4854 * array is immutable, so we don't need to lock the
4855 * subsys array here. On the other hand, modular section
4856 * of the array can be freed at module unload, so we
4859 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4860 struct cgroup_subsys
*ss
= subsys
[i
];
4869 * cgroup_exit - detach cgroup from exiting task
4870 * @tsk: pointer to task_struct of exiting process
4871 * @run_callback: run exit callbacks?
4873 * Description: Detach cgroup from @tsk and release it.
4875 * Note that cgroups marked notify_on_release force every task in
4876 * them to take the global cgroup_mutex mutex when exiting.
4877 * This could impact scaling on very large systems. Be reluctant to
4878 * use notify_on_release cgroups where very high task exit scaling
4879 * is required on large systems.
4881 * the_top_cgroup_hack:
4883 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4885 * We call cgroup_exit() while the task is still competent to
4886 * handle notify_on_release(), then leave the task attached to the
4887 * root cgroup in each hierarchy for the remainder of its exit.
4889 * To do this properly, we would increment the reference count on
4890 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4891 * code we would add a second cgroup function call, to drop that
4892 * reference. This would just create an unnecessary hot spot on
4893 * the top_cgroup reference count, to no avail.
4895 * Normally, holding a reference to a cgroup without bumping its
4896 * count is unsafe. The cgroup could go away, or someone could
4897 * attach us to a different cgroup, decrementing the count on
4898 * the first cgroup that we never incremented. But in this case,
4899 * top_cgroup isn't going away, and either task has PF_EXITING set,
4900 * which wards off any cgroup_attach_task() attempts, or task is a failed
4901 * fork, never visible to cgroup_attach_task.
4903 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4909 * Unlink from the css_set task list if necessary.
4910 * Optimistically check cg_list before taking
4913 if (!list_empty(&tsk
->cg_list
)) {
4914 write_lock(&css_set_lock
);
4915 if (!list_empty(&tsk
->cg_list
))
4916 list_del_init(&tsk
->cg_list
);
4917 write_unlock(&css_set_lock
);
4920 /* Reassign the task to the init_css_set. */
4923 tsk
->cgroups
= &init_css_set
;
4925 if (run_callbacks
&& need_forkexit_callback
) {
4927 * fork/exit callbacks are supported only for builtin
4928 * subsystems, see cgroup_post_fork() for details.
4930 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4931 struct cgroup_subsys
*ss
= subsys
[i
];
4934 struct cgroup
*old_cgrp
=
4935 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4936 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4937 ss
->exit(cgrp
, old_cgrp
, tsk
);
4943 put_css_set_taskexit(cg
);
4946 static void check_for_release(struct cgroup
*cgrp
)
4948 /* All of these checks rely on RCU to keep the cgroup
4949 * structure alive */
4950 if (cgroup_is_releasable(cgrp
) &&
4951 !atomic_read(&cgrp
->count
) && list_empty(&cgrp
->children
)) {
4953 * Control Group is currently removeable. If it's not
4954 * already queued for a userspace notification, queue
4957 int need_schedule_work
= 0;
4959 raw_spin_lock(&release_list_lock
);
4960 if (!cgroup_is_removed(cgrp
) &&
4961 list_empty(&cgrp
->release_list
)) {
4962 list_add(&cgrp
->release_list
, &release_list
);
4963 need_schedule_work
= 1;
4965 raw_spin_unlock(&release_list_lock
);
4966 if (need_schedule_work
)
4967 schedule_work(&release_agent_work
);
4971 /* Caller must verify that the css is not for root cgroup */
4972 bool __css_tryget(struct cgroup_subsys_state
*css
)
4977 v
= css_refcnt(css
);
4978 t
= atomic_cmpxchg(&css
->refcnt
, v
, v
+ 1);
4986 EXPORT_SYMBOL_GPL(__css_tryget
);
4988 /* Caller must verify that the css is not for root cgroup */
4989 void __css_put(struct cgroup_subsys_state
*css
)
4993 v
= css_unbias_refcnt(atomic_dec_return(&css
->refcnt
));
4995 schedule_work(&css
->dput_work
);
4997 EXPORT_SYMBOL_GPL(__css_put
);
5000 * Notify userspace when a cgroup is released, by running the
5001 * configured release agent with the name of the cgroup (path
5002 * relative to the root of cgroup file system) as the argument.
5004 * Most likely, this user command will try to rmdir this cgroup.
5006 * This races with the possibility that some other task will be
5007 * attached to this cgroup before it is removed, or that some other
5008 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5009 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5010 * unused, and this cgroup will be reprieved from its death sentence,
5011 * to continue to serve a useful existence. Next time it's released,
5012 * we will get notified again, if it still has 'notify_on_release' set.
5014 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5015 * means only wait until the task is successfully execve()'d. The
5016 * separate release agent task is forked by call_usermodehelper(),
5017 * then control in this thread returns here, without waiting for the
5018 * release agent task. We don't bother to wait because the caller of
5019 * this routine has no use for the exit status of the release agent
5020 * task, so no sense holding our caller up for that.
5022 static void cgroup_release_agent(struct work_struct
*work
)
5024 BUG_ON(work
!= &release_agent_work
);
5025 mutex_lock(&cgroup_mutex
);
5026 raw_spin_lock(&release_list_lock
);
5027 while (!list_empty(&release_list
)) {
5028 char *argv
[3], *envp
[3];
5030 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5031 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5034 list_del_init(&cgrp
->release_list
);
5035 raw_spin_unlock(&release_list_lock
);
5036 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5039 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5041 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5046 argv
[i
++] = agentbuf
;
5047 argv
[i
++] = pathbuf
;
5051 /* minimal command environment */
5052 envp
[i
++] = "HOME=/";
5053 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5056 /* Drop the lock while we invoke the usermode helper,
5057 * since the exec could involve hitting disk and hence
5058 * be a slow process */
5059 mutex_unlock(&cgroup_mutex
);
5060 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5061 mutex_lock(&cgroup_mutex
);
5065 raw_spin_lock(&release_list_lock
);
5067 raw_spin_unlock(&release_list_lock
);
5068 mutex_unlock(&cgroup_mutex
);
5071 static int __init
cgroup_disable(char *str
)
5076 while ((token
= strsep(&str
, ",")) != NULL
) {
5079 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
5080 struct cgroup_subsys
*ss
= subsys
[i
];
5083 * cgroup_disable, being at boot time, can't
5084 * know about module subsystems, so we don't
5087 if (!ss
|| ss
->module
)
5090 if (!strcmp(token
, ss
->name
)) {
5092 printk(KERN_INFO
"Disabling %s control group"
5093 " subsystem\n", ss
->name
);
5100 __setup("cgroup_disable=", cgroup_disable
);
5103 * Functons for CSS ID.
5107 *To get ID other than 0, this should be called when !cgroup_is_removed().
5109 unsigned short css_id(struct cgroup_subsys_state
*css
)
5111 struct css_id
*cssid
;
5114 * This css_id() can return correct value when somone has refcnt
5115 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5116 * it's unchanged until freed.
5118 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5124 EXPORT_SYMBOL_GPL(css_id
);
5126 unsigned short css_depth(struct cgroup_subsys_state
*css
)
5128 struct css_id
*cssid
;
5130 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5133 return cssid
->depth
;
5136 EXPORT_SYMBOL_GPL(css_depth
);
5139 * css_is_ancestor - test "root" css is an ancestor of "child"
5140 * @child: the css to be tested.
5141 * @root: the css supporsed to be an ancestor of the child.
5143 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5144 * this function reads css->id, the caller must hold rcu_read_lock().
5145 * But, considering usual usage, the csses should be valid objects after test.
5146 * Assuming that the caller will do some action to the child if this returns
5147 * returns true, the caller must take "child";s reference count.
5148 * If "child" is valid object and this returns true, "root" is valid, too.
5151 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5152 const struct cgroup_subsys_state
*root
)
5154 struct css_id
*child_id
;
5155 struct css_id
*root_id
;
5157 child_id
= rcu_dereference(child
->id
);
5160 root_id
= rcu_dereference(root
->id
);
5163 if (child_id
->depth
< root_id
->depth
)
5165 if (child_id
->stack
[root_id
->depth
] != root_id
->id
)
5170 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5172 struct css_id
*id
= css
->id
;
5173 /* When this is called before css_id initialization, id can be NULL */
5177 BUG_ON(!ss
->use_id
);
5179 rcu_assign_pointer(id
->css
, NULL
);
5180 rcu_assign_pointer(css
->id
, NULL
);
5181 spin_lock(&ss
->id_lock
);
5182 idr_remove(&ss
->idr
, id
->id
);
5183 spin_unlock(&ss
->id_lock
);
5184 kfree_rcu(id
, rcu_head
);
5186 EXPORT_SYMBOL_GPL(free_css_id
);
5189 * This is called by init or create(). Then, calls to this function are
5190 * always serialized (By cgroup_mutex() at create()).
5193 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5195 struct css_id
*newid
;
5198 BUG_ON(!ss
->use_id
);
5200 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5201 newid
= kzalloc(size
, GFP_KERNEL
);
5203 return ERR_PTR(-ENOMEM
);
5205 idr_preload(GFP_KERNEL
);
5206 spin_lock(&ss
->id_lock
);
5207 /* Don't use 0. allocates an ID of 1-65535 */
5208 ret
= idr_alloc(&ss
->idr
, newid
, 1, CSS_ID_MAX
+ 1, GFP_NOWAIT
);
5209 spin_unlock(&ss
->id_lock
);
5212 /* Returns error when there are no free spaces for new ID.*/
5217 newid
->depth
= depth
;
5221 return ERR_PTR(ret
);
5225 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5226 struct cgroup_subsys_state
*rootcss
)
5228 struct css_id
*newid
;
5230 spin_lock_init(&ss
->id_lock
);
5233 newid
= get_new_cssid(ss
, 0);
5235 return PTR_ERR(newid
);
5237 newid
->stack
[0] = newid
->id
;
5238 newid
->css
= rootcss
;
5239 rootcss
->id
= newid
;
5243 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5244 struct cgroup
*child
)
5246 int subsys_id
, i
, depth
= 0;
5247 struct cgroup_subsys_state
*parent_css
, *child_css
;
5248 struct css_id
*child_id
, *parent_id
;
5250 subsys_id
= ss
->subsys_id
;
5251 parent_css
= parent
->subsys
[subsys_id
];
5252 child_css
= child
->subsys
[subsys_id
];
5253 parent_id
= parent_css
->id
;
5254 depth
= parent_id
->depth
+ 1;
5256 child_id
= get_new_cssid(ss
, depth
);
5257 if (IS_ERR(child_id
))
5258 return PTR_ERR(child_id
);
5260 for (i
= 0; i
< depth
; i
++)
5261 child_id
->stack
[i
] = parent_id
->stack
[i
];
5262 child_id
->stack
[depth
] = child_id
->id
;
5264 * child_id->css pointer will be set after this cgroup is available
5265 * see cgroup_populate_dir()
5267 rcu_assign_pointer(child_css
->id
, child_id
);
5273 * css_lookup - lookup css by id
5274 * @ss: cgroup subsys to be looked into.
5277 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5278 * NULL if not. Should be called under rcu_read_lock()
5280 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5282 struct css_id
*cssid
= NULL
;
5284 BUG_ON(!ss
->use_id
);
5285 cssid
= idr_find(&ss
->idr
, id
);
5287 if (unlikely(!cssid
))
5290 return rcu_dereference(cssid
->css
);
5292 EXPORT_SYMBOL_GPL(css_lookup
);
5295 * css_get_next - lookup next cgroup under specified hierarchy.
5296 * @ss: pointer to subsystem
5297 * @id: current position of iteration.
5298 * @root: pointer to css. search tree under this.
5299 * @foundid: position of found object.
5301 * Search next css under the specified hierarchy of rootid. Calling under
5302 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5304 struct cgroup_subsys_state
*
5305 css_get_next(struct cgroup_subsys
*ss
, int id
,
5306 struct cgroup_subsys_state
*root
, int *foundid
)
5308 struct cgroup_subsys_state
*ret
= NULL
;
5311 int rootid
= css_id(root
);
5312 int depth
= css_depth(root
);
5317 BUG_ON(!ss
->use_id
);
5318 WARN_ON_ONCE(!rcu_read_lock_held());
5320 /* fill start point for scan */
5324 * scan next entry from bitmap(tree), tmpid is updated after
5327 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
5330 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
5331 ret
= rcu_dereference(tmp
->css
);
5337 /* continue to scan from next id */
5344 * get corresponding css from file open on cgroupfs directory
5346 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5348 struct cgroup
*cgrp
;
5349 struct inode
*inode
;
5350 struct cgroup_subsys_state
*css
;
5352 inode
= file_inode(f
);
5353 /* check in cgroup filesystem dir */
5354 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5355 return ERR_PTR(-EBADF
);
5357 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5358 return ERR_PTR(-EINVAL
);
5361 cgrp
= __d_cgrp(f
->f_dentry
);
5362 css
= cgrp
->subsys
[id
];
5363 return css
? css
: ERR_PTR(-ENOENT
);
5366 #ifdef CONFIG_CGROUP_DEBUG
5367 static struct cgroup_subsys_state
*debug_css_alloc(struct cgroup
*cont
)
5369 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5372 return ERR_PTR(-ENOMEM
);
5377 static void debug_css_free(struct cgroup
*cont
)
5379 kfree(cont
->subsys
[debug_subsys_id
]);
5382 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5384 return atomic_read(&cont
->count
);
5387 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5389 return cgroup_task_count(cont
);
5392 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5394 return (u64
)(unsigned long)current
->cgroups
;
5397 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5403 count
= atomic_read(¤t
->cgroups
->refcount
);
5408 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5410 struct seq_file
*seq
)
5412 struct cg_cgroup_link
*link
;
5415 read_lock(&css_set_lock
);
5417 cg
= rcu_dereference(current
->cgroups
);
5418 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5419 struct cgroup
*c
= link
->cgrp
;
5423 name
= c
->dentry
->d_name
.name
;
5426 seq_printf(seq
, "Root %d group %s\n",
5427 c
->root
->hierarchy_id
, name
);
5430 read_unlock(&css_set_lock
);
5434 #define MAX_TASKS_SHOWN_PER_CSS 25
5435 static int cgroup_css_links_read(struct cgroup
*cont
,
5437 struct seq_file
*seq
)
5439 struct cg_cgroup_link
*link
;
5441 read_lock(&css_set_lock
);
5442 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5443 struct css_set
*cg
= link
->cg
;
5444 struct task_struct
*task
;
5446 seq_printf(seq
, "css_set %p\n", cg
);
5447 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5448 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5449 seq_puts(seq
, " ...\n");
5452 seq_printf(seq
, " task %d\n",
5453 task_pid_vnr(task
));
5457 read_unlock(&css_set_lock
);
5461 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5463 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5466 static struct cftype debug_files
[] = {
5468 .name
= "cgroup_refcount",
5469 .read_u64
= cgroup_refcount_read
,
5472 .name
= "taskcount",
5473 .read_u64
= debug_taskcount_read
,
5477 .name
= "current_css_set",
5478 .read_u64
= current_css_set_read
,
5482 .name
= "current_css_set_refcount",
5483 .read_u64
= current_css_set_refcount_read
,
5487 .name
= "current_css_set_cg_links",
5488 .read_seq_string
= current_css_set_cg_links_read
,
5492 .name
= "cgroup_css_links",
5493 .read_seq_string
= cgroup_css_links_read
,
5497 .name
= "releasable",
5498 .read_u64
= releasable_read
,
5504 struct cgroup_subsys debug_subsys
= {
5506 .css_alloc
= debug_css_alloc
,
5507 .css_free
= debug_css_free
,
5508 .subsys_id
= debug_subsys_id
,
5509 .base_cftypes
= debug_files
,
5511 #endif /* CONFIG_CGROUP_DEBUG */