2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
33 #include <linux/init_task.h>
34 #include <linux/kernel.h>
35 #include <linux/list.h>
37 #include <linux/mutex.h>
38 #include <linux/mount.h>
39 #include <linux/pagemap.h>
40 #include <linux/proc_fs.h>
41 #include <linux/rcupdate.h>
42 #include <linux/sched.h>
43 #include <linux/backing-dev.h>
44 #include <linux/seq_file.h>
45 #include <linux/slab.h>
46 #include <linux/magic.h>
47 #include <linux/spinlock.h>
48 #include <linux/string.h>
49 #include <linux/sort.h>
50 #include <linux/kmod.h>
51 #include <linux/module.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/namei.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/idr.h>
58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
59 #include <linux/eventfd.h>
60 #include <linux/poll.h>
61 #include <linux/flex_array.h> /* used in cgroup_attach_task */
62 #include <linux/kthread.h>
63 #include <linux/file.h>
65 #include <linux/atomic.h>
68 * cgroup_mutex is the master lock. Any modification to cgroup or its
69 * hierarchy must be performed while holding it.
71 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
72 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
73 * release_agent_path and so on. Modifying requires both cgroup_mutex and
74 * cgroup_root_mutex. Readers can acquire either of the two. This is to
75 * break the following locking order cycle.
77 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
78 * B. namespace_sem -> cgroup_mutex
80 * B happens only through cgroup_show_options() and using cgroup_root_mutex
83 #ifdef CONFIG_PROVE_RCU
84 DEFINE_MUTEX(cgroup_mutex
);
85 EXPORT_SYMBOL_GPL(cgroup_mutex
); /* only for lockdep */
87 static DEFINE_MUTEX(cgroup_mutex
);
90 static DEFINE_MUTEX(cgroup_root_mutex
);
93 * cgroup destruction makes heavy use of work items and there can be a lot
94 * of concurrent destructions. Use a separate workqueue so that cgroup
95 * destruction work items don't end up filling up max_active of system_wq
96 * which may lead to deadlock.
98 static struct workqueue_struct
*cgroup_destroy_wq
;
101 * Generate an array of cgroup subsystem pointers. At boot time, this is
102 * populated with the built in subsystems, and modular subsystems are
103 * registered after that. The mutable section of this array is protected by
106 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
107 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
108 static struct cgroup_subsys
*cgroup_subsys
[CGROUP_SUBSYS_COUNT
] = {
109 #include <linux/cgroup_subsys.h>
113 * The dummy hierarchy, reserved for the subsystems that are otherwise
114 * unattached - it never has more than a single cgroup, and all tasks are
115 * part of that cgroup.
117 static struct cgroupfs_root cgroup_dummy_root
;
119 /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */
120 static struct cgroup
* const cgroup_dummy_top
= &cgroup_dummy_root
.top_cgroup
;
123 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
126 struct list_head node
;
127 struct dentry
*dentry
;
129 struct cgroup_subsys_state
*css
;
132 struct simple_xattrs xattrs
;
136 * cgroup_event represents events which userspace want to receive.
138 struct cgroup_event
{
140 * css which the event belongs to.
142 struct cgroup_subsys_state
*css
;
144 * Control file which the event associated.
148 * eventfd to signal userspace about the event.
150 struct eventfd_ctx
*eventfd
;
152 * Each of these stored in a list by the cgroup.
154 struct list_head list
;
156 * All fields below needed to unregister event when
157 * userspace closes eventfd.
160 wait_queue_head_t
*wqh
;
162 struct work_struct remove
;
165 /* The list of hierarchy roots */
167 static LIST_HEAD(cgroup_roots
);
168 static int cgroup_root_count
;
171 * Hierarchy ID allocation and mapping. It follows the same exclusion
172 * rules as other root ops - both cgroup_mutex and cgroup_root_mutex for
173 * writes, either for reads.
175 static DEFINE_IDR(cgroup_hierarchy_idr
);
177 static struct cgroup_name root_cgroup_name
= { .name
= "/" };
180 * Assign a monotonically increasing serial number to cgroups. It
181 * guarantees cgroups with bigger numbers are newer than those with smaller
182 * numbers. Also, as cgroups are always appended to the parent's
183 * ->children list, it guarantees that sibling cgroups are always sorted in
184 * the ascending serial number order on the list. Protected by
187 static u64 cgroup_serial_nr_next
= 1;
189 /* This flag indicates whether tasks in the fork and exit paths should
190 * check for fork/exit handlers to call. This avoids us having to do
191 * extra work in the fork/exit path if none of the subsystems need to
194 static int need_forkexit_callback __read_mostly
;
196 static struct cftype cgroup_base_files
[];
198 static void cgroup_destroy_css_killed(struct cgroup
*cgrp
);
199 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
200 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
202 static int cgroup_file_release(struct inode
*inode
, struct file
*file
);
205 * cgroup_css - obtain a cgroup's css for the specified subsystem
206 * @cgrp: the cgroup of interest
207 * @ss: the subsystem of interest (%NULL returns the dummy_css)
209 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
210 * function must be called either under cgroup_mutex or rcu_read_lock() and
211 * the caller is responsible for pinning the returned css if it wants to
212 * keep accessing it outside the said locks. This function may return
213 * %NULL if @cgrp doesn't have @subsys_id enabled.
215 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
216 struct cgroup_subsys
*ss
)
219 return rcu_dereference_check(cgrp
->subsys
[ss
->subsys_id
],
220 lockdep_is_held(&cgroup_mutex
));
222 return &cgrp
->dummy_css
;
225 /* convenient tests for these bits */
226 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
228 return test_bit(CGRP_DEAD
, &cgrp
->flags
);
232 * cgroup_is_descendant - test ancestry
233 * @cgrp: the cgroup to be tested
234 * @ancestor: possible ancestor of @cgrp
236 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
237 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
238 * and @ancestor are accessible.
240 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
243 if (cgrp
== ancestor
)
249 EXPORT_SYMBOL_GPL(cgroup_is_descendant
);
251 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
254 (1 << CGRP_RELEASABLE
) |
255 (1 << CGRP_NOTIFY_ON_RELEASE
);
256 return (cgrp
->flags
& bits
) == bits
;
259 static int notify_on_release(const struct cgroup
*cgrp
)
261 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
265 * for_each_subsys - iterate all loaded cgroup subsystems
266 * @ss: the iteration cursor
267 * @i: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
269 * Should be called under cgroup_mutex.
271 #define for_each_subsys(ss, i) \
272 for ((i) = 0; (i) < CGROUP_SUBSYS_COUNT; (i)++) \
273 if (({ lockdep_assert_held(&cgroup_mutex); \
274 !((ss) = cgroup_subsys[i]); })) { } \
278 * for_each_builtin_subsys - iterate all built-in cgroup subsystems
279 * @ss: the iteration cursor
280 * @i: the index of @ss, CGROUP_BUILTIN_SUBSYS_COUNT after reaching the end
282 * Bulit-in subsystems are always present and iteration itself doesn't
283 * require any synchronization.
285 #define for_each_builtin_subsys(ss, i) \
286 for ((i) = 0; (i) < CGROUP_BUILTIN_SUBSYS_COUNT && \
287 (((ss) = cgroup_subsys[i]) || true); (i)++)
289 /* iterate each subsystem attached to a hierarchy */
290 #define for_each_root_subsys(root, ss) \
291 list_for_each_entry((ss), &(root)->subsys_list, sibling)
293 /* iterate across the active hierarchies */
294 #define for_each_active_root(root) \
295 list_for_each_entry((root), &cgroup_roots, root_list)
297 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
299 return dentry
->d_fsdata
;
302 static inline struct cfent
*__d_cfe(struct dentry
*dentry
)
304 return dentry
->d_fsdata
;
307 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
309 return __d_cfe(dentry
)->type
;
313 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
314 * @cgrp: the cgroup to be checked for liveness
316 * On success, returns true; the mutex should be later unlocked. On
317 * failure returns false with no lock held.
319 static bool cgroup_lock_live_group(struct cgroup
*cgrp
)
321 mutex_lock(&cgroup_mutex
);
322 if (cgroup_is_dead(cgrp
)) {
323 mutex_unlock(&cgroup_mutex
);
329 /* the list of cgroups eligible for automatic release. Protected by
330 * release_list_lock */
331 static LIST_HEAD(release_list
);
332 static DEFINE_RAW_SPINLOCK(release_list_lock
);
333 static void cgroup_release_agent(struct work_struct
*work
);
334 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
335 static void check_for_release(struct cgroup
*cgrp
);
338 * A cgroup can be associated with multiple css_sets as different tasks may
339 * belong to different cgroups on different hierarchies. In the other
340 * direction, a css_set is naturally associated with multiple cgroups.
341 * This M:N relationship is represented by the following link structure
342 * which exists for each association and allows traversing the associations
345 struct cgrp_cset_link
{
346 /* the cgroup and css_set this link associates */
348 struct css_set
*cset
;
350 /* list of cgrp_cset_links anchored at cgrp->cset_links */
351 struct list_head cset_link
;
353 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
354 struct list_head cgrp_link
;
357 /* The default css_set - used by init and its children prior to any
358 * hierarchies being mounted. It contains a pointer to the root state
359 * for each subsystem. Also used to anchor the list of css_sets. Not
360 * reference-counted, to improve performance when child cgroups
361 * haven't been created.
364 static struct css_set init_css_set
;
365 static struct cgrp_cset_link init_cgrp_cset_link
;
368 * css_set_lock protects the list of css_set objects, and the chain of
369 * tasks off each css_set. Nests outside task->alloc_lock due to
370 * css_task_iter_start().
372 static DEFINE_RWLOCK(css_set_lock
);
373 static int css_set_count
;
376 * hash table for cgroup groups. This improves the performance to find
377 * an existing css_set. This hash doesn't (currently) take into
378 * account cgroups in empty hierarchies.
380 #define CSS_SET_HASH_BITS 7
381 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
383 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
385 unsigned long key
= 0UL;
386 struct cgroup_subsys
*ss
;
389 for_each_subsys(ss
, i
)
390 key
+= (unsigned long)css
[i
];
391 key
= (key
>> 16) ^ key
;
397 * We don't maintain the lists running through each css_set to its task
398 * until after the first call to css_task_iter_start(). This reduces the
399 * fork()/exit() overhead for people who have cgroups compiled into their
400 * kernel but not actually in use.
402 static int use_task_css_set_links __read_mostly
;
404 static void __put_css_set(struct css_set
*cset
, int taskexit
)
406 struct cgrp_cset_link
*link
, *tmp_link
;
409 * Ensure that the refcount doesn't hit zero while any readers
410 * can see it. Similar to atomic_dec_and_lock(), but for an
413 if (atomic_add_unless(&cset
->refcount
, -1, 1))
415 write_lock(&css_set_lock
);
416 if (!atomic_dec_and_test(&cset
->refcount
)) {
417 write_unlock(&css_set_lock
);
421 /* This css_set is dead. unlink it and release cgroup refcounts */
422 hash_del(&cset
->hlist
);
425 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
426 struct cgroup
*cgrp
= link
->cgrp
;
428 list_del(&link
->cset_link
);
429 list_del(&link
->cgrp_link
);
431 /* @cgrp can't go away while we're holding css_set_lock */
432 if (list_empty(&cgrp
->cset_links
) && notify_on_release(cgrp
)) {
434 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
435 check_for_release(cgrp
);
441 write_unlock(&css_set_lock
);
442 kfree_rcu(cset
, rcu_head
);
446 * refcounted get/put for css_set objects
448 static inline void get_css_set(struct css_set
*cset
)
450 atomic_inc(&cset
->refcount
);
453 static inline void put_css_set(struct css_set
*cset
)
455 __put_css_set(cset
, 0);
458 static inline void put_css_set_taskexit(struct css_set
*cset
)
460 __put_css_set(cset
, 1);
464 * compare_css_sets - helper function for find_existing_css_set().
465 * @cset: candidate css_set being tested
466 * @old_cset: existing css_set for a task
467 * @new_cgrp: cgroup that's being entered by the task
468 * @template: desired set of css pointers in css_set (pre-calculated)
470 * Returns true if "cset" matches "old_cset" except for the hierarchy
471 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
473 static bool compare_css_sets(struct css_set
*cset
,
474 struct css_set
*old_cset
,
475 struct cgroup
*new_cgrp
,
476 struct cgroup_subsys_state
*template[])
478 struct list_head
*l1
, *l2
;
480 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
))) {
481 /* Not all subsystems matched */
486 * Compare cgroup pointers in order to distinguish between
487 * different cgroups in heirarchies with no subsystems. We
488 * could get by with just this check alone (and skip the
489 * memcmp above) but on most setups the memcmp check will
490 * avoid the need for this more expensive check on almost all
494 l1
= &cset
->cgrp_links
;
495 l2
= &old_cset
->cgrp_links
;
497 struct cgrp_cset_link
*link1
, *link2
;
498 struct cgroup
*cgrp1
, *cgrp2
;
502 /* See if we reached the end - both lists are equal length. */
503 if (l1
== &cset
->cgrp_links
) {
504 BUG_ON(l2
!= &old_cset
->cgrp_links
);
507 BUG_ON(l2
== &old_cset
->cgrp_links
);
509 /* Locate the cgroups associated with these links. */
510 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
511 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
514 /* Hierarchies should be linked in the same order. */
515 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
518 * If this hierarchy is the hierarchy of the cgroup
519 * that's changing, then we need to check that this
520 * css_set points to the new cgroup; if it's any other
521 * hierarchy, then this css_set should point to the
522 * same cgroup as the old css_set.
524 if (cgrp1
->root
== new_cgrp
->root
) {
525 if (cgrp1
!= new_cgrp
)
536 * find_existing_css_set - init css array and find the matching css_set
537 * @old_cset: the css_set that we're using before the cgroup transition
538 * @cgrp: the cgroup that we're moving into
539 * @template: out param for the new set of csses, should be clear on entry
541 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
543 struct cgroup_subsys_state
*template[])
545 struct cgroupfs_root
*root
= cgrp
->root
;
546 struct cgroup_subsys
*ss
;
547 struct css_set
*cset
;
552 * Build the set of subsystem state objects that we want to see in the
553 * new css_set. while subsystems can change globally, the entries here
554 * won't change, so no need for locking.
556 for_each_subsys(ss
, i
) {
557 if (root
->subsys_mask
& (1UL << i
)) {
558 /* Subsystem is in this hierarchy. So we want
559 * the subsystem state from the new
561 template[i
] = cgroup_css(cgrp
, ss
);
563 /* Subsystem is not in this hierarchy, so we
564 * don't want to change the subsystem state */
565 template[i
] = old_cset
->subsys
[i
];
569 key
= css_set_hash(template);
570 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
571 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
574 /* This css_set matches what we need */
578 /* No existing cgroup group matched */
582 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
584 struct cgrp_cset_link
*link
, *tmp_link
;
586 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
587 list_del(&link
->cset_link
);
593 * allocate_cgrp_cset_links - allocate cgrp_cset_links
594 * @count: the number of links to allocate
595 * @tmp_links: list_head the allocated links are put on
597 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
598 * through ->cset_link. Returns 0 on success or -errno.
600 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
602 struct cgrp_cset_link
*link
;
605 INIT_LIST_HEAD(tmp_links
);
607 for (i
= 0; i
< count
; i
++) {
608 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
610 free_cgrp_cset_links(tmp_links
);
613 list_add(&link
->cset_link
, tmp_links
);
619 * link_css_set - a helper function to link a css_set to a cgroup
620 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
621 * @cset: the css_set to be linked
622 * @cgrp: the destination cgroup
624 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
627 struct cgrp_cset_link
*link
;
629 BUG_ON(list_empty(tmp_links
));
630 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
633 list_move(&link
->cset_link
, &cgrp
->cset_links
);
635 * Always add links to the tail of the list so that the list
636 * is sorted by order of hierarchy creation
638 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
642 * find_css_set - return a new css_set with one cgroup updated
643 * @old_cset: the baseline css_set
644 * @cgrp: the cgroup to be updated
646 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
647 * substituted into the appropriate hierarchy.
649 static struct css_set
*find_css_set(struct css_set
*old_cset
,
652 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
653 struct css_set
*cset
;
654 struct list_head tmp_links
;
655 struct cgrp_cset_link
*link
;
658 lockdep_assert_held(&cgroup_mutex
);
660 /* First see if we already have a cgroup group that matches
662 read_lock(&css_set_lock
);
663 cset
= find_existing_css_set(old_cset
, cgrp
, template);
666 read_unlock(&css_set_lock
);
671 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
675 /* Allocate all the cgrp_cset_link objects that we'll need */
676 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
681 atomic_set(&cset
->refcount
, 1);
682 INIT_LIST_HEAD(&cset
->cgrp_links
);
683 INIT_LIST_HEAD(&cset
->tasks
);
684 INIT_HLIST_NODE(&cset
->hlist
);
686 /* Copy the set of subsystem state objects generated in
687 * find_existing_css_set() */
688 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
690 write_lock(&css_set_lock
);
691 /* Add reference counts and links from the new css_set. */
692 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
693 struct cgroup
*c
= link
->cgrp
;
695 if (c
->root
== cgrp
->root
)
697 link_css_set(&tmp_links
, cset
, c
);
700 BUG_ON(!list_empty(&tmp_links
));
704 /* Add this cgroup group to the hash table */
705 key
= css_set_hash(cset
->subsys
);
706 hash_add(css_set_table
, &cset
->hlist
, key
);
708 write_unlock(&css_set_lock
);
714 * Return the cgroup for "task" from the given hierarchy. Must be
715 * called with cgroup_mutex held.
717 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
718 struct cgroupfs_root
*root
)
720 struct css_set
*cset
;
721 struct cgroup
*res
= NULL
;
723 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
724 read_lock(&css_set_lock
);
726 * No need to lock the task - since we hold cgroup_mutex the
727 * task can't change groups, so the only thing that can happen
728 * is that it exits and its css is set back to init_css_set.
730 cset
= task_css_set(task
);
731 if (cset
== &init_css_set
) {
732 res
= &root
->top_cgroup
;
734 struct cgrp_cset_link
*link
;
736 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
737 struct cgroup
*c
= link
->cgrp
;
739 if (c
->root
== root
) {
745 read_unlock(&css_set_lock
);
751 * There is one global cgroup mutex. We also require taking
752 * task_lock() when dereferencing a task's cgroup subsys pointers.
753 * See "The task_lock() exception", at the end of this comment.
755 * A task must hold cgroup_mutex to modify cgroups.
757 * Any task can increment and decrement the count field without lock.
758 * So in general, code holding cgroup_mutex can't rely on the count
759 * field not changing. However, if the count goes to zero, then only
760 * cgroup_attach_task() can increment it again. Because a count of zero
761 * means that no tasks are currently attached, therefore there is no
762 * way a task attached to that cgroup can fork (the other way to
763 * increment the count). So code holding cgroup_mutex can safely
764 * assume that if the count is zero, it will stay zero. Similarly, if
765 * a task holds cgroup_mutex on a cgroup with zero count, it
766 * knows that the cgroup won't be removed, as cgroup_rmdir()
769 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
770 * (usually) take cgroup_mutex. These are the two most performance
771 * critical pieces of code here. The exception occurs on cgroup_exit(),
772 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
773 * is taken, and if the cgroup count is zero, a usermode call made
774 * to the release agent with the name of the cgroup (path relative to
775 * the root of cgroup file system) as the argument.
777 * A cgroup can only be deleted if both its 'count' of using tasks
778 * is zero, and its list of 'children' cgroups is empty. Since all
779 * tasks in the system use _some_ cgroup, and since there is always at
780 * least one task in the system (init, pid == 1), therefore, top_cgroup
781 * always has either children cgroups and/or using tasks. So we don't
782 * need a special hack to ensure that top_cgroup cannot be deleted.
784 * The task_lock() exception
786 * The need for this exception arises from the action of
787 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
788 * another. It does so using cgroup_mutex, however there are
789 * several performance critical places that need to reference
790 * task->cgroup without the expense of grabbing a system global
791 * mutex. Therefore except as noted below, when dereferencing or, as
792 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
793 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
794 * the task_struct routinely used for such matters.
796 * P.S. One more locking exception. RCU is used to guard the
797 * update of a tasks cgroup pointer by cgroup_attach_task()
801 * A couple of forward declarations required, due to cyclic reference loop:
802 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
803 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
807 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
808 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
809 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
);
810 static const struct inode_operations cgroup_dir_inode_operations
;
811 static const struct file_operations proc_cgroupstats_operations
;
813 static struct backing_dev_info cgroup_backing_dev_info
= {
815 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
818 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
820 struct inode
*inode
= new_inode(sb
);
823 inode
->i_ino
= get_next_ino();
824 inode
->i_mode
= mode
;
825 inode
->i_uid
= current_fsuid();
826 inode
->i_gid
= current_fsgid();
827 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
828 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
833 static struct cgroup_name
*cgroup_alloc_name(struct dentry
*dentry
)
835 struct cgroup_name
*name
;
837 name
= kmalloc(sizeof(*name
) + dentry
->d_name
.len
+ 1, GFP_KERNEL
);
840 strcpy(name
->name
, dentry
->d_name
.name
);
844 static void cgroup_free_fn(struct work_struct
*work
)
846 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, destroy_work
);
848 mutex_lock(&cgroup_mutex
);
849 cgrp
->root
->number_of_cgroups
--;
850 mutex_unlock(&cgroup_mutex
);
853 * We get a ref to the parent's dentry, and put the ref when
854 * this cgroup is being freed, so it's guaranteed that the
855 * parent won't be destroyed before its children.
857 dput(cgrp
->parent
->dentry
);
860 * Drop the active superblock reference that we took when we
861 * created the cgroup. This will free cgrp->root, if we are
862 * holding the last reference to @sb.
864 deactivate_super(cgrp
->root
->sb
);
867 * if we're getting rid of the cgroup, refcount should ensure
868 * that there are no pidlists left.
870 BUG_ON(!list_empty(&cgrp
->pidlists
));
872 simple_xattrs_free(&cgrp
->xattrs
);
874 kfree(rcu_dereference_raw(cgrp
->name
));
878 static void cgroup_free_rcu(struct rcu_head
*head
)
880 struct cgroup
*cgrp
= container_of(head
, struct cgroup
, rcu_head
);
882 INIT_WORK(&cgrp
->destroy_work
, cgroup_free_fn
);
883 queue_work(cgroup_destroy_wq
, &cgrp
->destroy_work
);
886 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
888 /* is dentry a directory ? if so, kfree() associated cgroup */
889 if (S_ISDIR(inode
->i_mode
)) {
890 struct cgroup
*cgrp
= dentry
->d_fsdata
;
892 BUG_ON(!(cgroup_is_dead(cgrp
)));
895 * XXX: cgrp->id is only used to look up css's. As cgroup
896 * and css's lifetimes will be decoupled, it should be made
897 * per-subsystem and moved to css->id so that lookups are
898 * successful until the target css is released.
900 idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
903 call_rcu(&cgrp
->rcu_head
, cgroup_free_rcu
);
905 struct cfent
*cfe
= __d_cfe(dentry
);
906 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
908 WARN_ONCE(!list_empty(&cfe
->node
) &&
909 cgrp
!= &cgrp
->root
->top_cgroup
,
910 "cfe still linked for %s\n", cfe
->type
->name
);
911 simple_xattrs_free(&cfe
->xattrs
);
917 static void remove_dir(struct dentry
*d
)
919 struct dentry
*parent
= dget(d
->d_parent
);
922 simple_rmdir(parent
->d_inode
, d
);
926 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
930 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
931 lockdep_assert_held(&cgroup_mutex
);
934 * If we're doing cleanup due to failure of cgroup_create(),
935 * the corresponding @cfe may not exist.
937 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
938 struct dentry
*d
= cfe
->dentry
;
940 if (cft
&& cfe
->type
!= cft
)
945 simple_unlink(cgrp
->dentry
->d_inode
, d
);
946 list_del_init(&cfe
->node
);
954 * cgroup_clear_dir - remove subsys files in a cgroup directory
955 * @cgrp: target cgroup
956 * @subsys_mask: mask of the subsystem ids whose files should be removed
958 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
960 struct cgroup_subsys
*ss
;
963 for_each_subsys(ss
, i
) {
964 struct cftype_set
*set
;
966 if (!test_bit(i
, &subsys_mask
))
968 list_for_each_entry(set
, &ss
->cftsets
, node
)
969 cgroup_addrm_files(cgrp
, set
->cfts
, false);
974 * NOTE : the dentry must have been dget()'ed
976 static void cgroup_d_remove_dir(struct dentry
*dentry
)
978 struct dentry
*parent
;
980 parent
= dentry
->d_parent
;
981 spin_lock(&parent
->d_lock
);
982 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
983 list_del_init(&dentry
->d_u
.d_child
);
984 spin_unlock(&dentry
->d_lock
);
985 spin_unlock(&parent
->d_lock
);
990 * Call with cgroup_mutex held. Drops reference counts on modules, including
991 * any duplicate ones that parse_cgroupfs_options took. If this function
992 * returns an error, no reference counts are touched.
994 static int rebind_subsystems(struct cgroupfs_root
*root
,
995 unsigned long added_mask
, unsigned removed_mask
)
997 struct cgroup
*cgrp
= &root
->top_cgroup
;
998 struct cgroup_subsys
*ss
;
999 unsigned long pinned
= 0;
1002 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1003 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1005 /* Check that any added subsystems are currently free */
1006 for_each_subsys(ss
, i
) {
1007 if (!(added_mask
& (1 << i
)))
1010 /* is the subsystem mounted elsewhere? */
1011 if (ss
->root
!= &cgroup_dummy_root
) {
1016 /* pin the module */
1017 if (!try_module_get(ss
->module
)) {
1024 /* subsys could be missing if unloaded between parsing and here */
1025 if (added_mask
!= pinned
) {
1030 ret
= cgroup_populate_dir(cgrp
, added_mask
);
1035 * Nothing can fail from this point on. Remove files for the
1036 * removed subsystems and rebind each subsystem.
1038 cgroup_clear_dir(cgrp
, removed_mask
);
1040 for_each_subsys(ss
, i
) {
1041 unsigned long bit
= 1UL << i
;
1043 if (bit
& added_mask
) {
1044 /* We're binding this subsystem to this hierarchy */
1045 BUG_ON(cgroup_css(cgrp
, ss
));
1046 BUG_ON(!cgroup_css(cgroup_dummy_top
, ss
));
1047 BUG_ON(cgroup_css(cgroup_dummy_top
, ss
)->cgroup
!= cgroup_dummy_top
);
1049 rcu_assign_pointer(cgrp
->subsys
[i
],
1050 cgroup_css(cgroup_dummy_top
, ss
));
1051 cgroup_css(cgrp
, ss
)->cgroup
= cgrp
;
1053 list_move(&ss
->sibling
, &root
->subsys_list
);
1056 ss
->bind(cgroup_css(cgrp
, ss
));
1058 /* refcount was already taken, and we're keeping it */
1059 root
->subsys_mask
|= bit
;
1060 } else if (bit
& removed_mask
) {
1061 /* We're removing this subsystem */
1062 BUG_ON(cgroup_css(cgrp
, ss
) != cgroup_css(cgroup_dummy_top
, ss
));
1063 BUG_ON(cgroup_css(cgrp
, ss
)->cgroup
!= cgrp
);
1066 ss
->bind(cgroup_css(cgroup_dummy_top
, ss
));
1068 cgroup_css(cgroup_dummy_top
, ss
)->cgroup
= cgroup_dummy_top
;
1069 RCU_INIT_POINTER(cgrp
->subsys
[i
], NULL
);
1071 cgroup_subsys
[i
]->root
= &cgroup_dummy_root
;
1072 list_move(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
1074 /* subsystem is now free - drop reference on module */
1075 module_put(ss
->module
);
1076 root
->subsys_mask
&= ~bit
;
1081 * Mark @root has finished binding subsystems. @root->subsys_mask
1082 * now matches the bound subsystems.
1084 root
->flags
|= CGRP_ROOT_SUBSYS_BOUND
;
1089 for_each_subsys(ss
, i
)
1090 if (pinned
& (1 << i
))
1091 module_put(ss
->module
);
1095 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1097 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1098 struct cgroup_subsys
*ss
;
1100 mutex_lock(&cgroup_root_mutex
);
1101 for_each_root_subsys(root
, ss
)
1102 seq_printf(seq
, ",%s", ss
->name
);
1103 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
)
1104 seq_puts(seq
, ",sane_behavior");
1105 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1106 seq_puts(seq
, ",noprefix");
1107 if (root
->flags
& CGRP_ROOT_XATTR
)
1108 seq_puts(seq
, ",xattr");
1109 if (strlen(root
->release_agent_path
))
1110 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1111 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
))
1112 seq_puts(seq
, ",clone_children");
1113 if (strlen(root
->name
))
1114 seq_printf(seq
, ",name=%s", root
->name
);
1115 mutex_unlock(&cgroup_root_mutex
);
1119 struct cgroup_sb_opts
{
1120 unsigned long subsys_mask
;
1121 unsigned long flags
;
1122 char *release_agent
;
1123 bool cpuset_clone_children
;
1125 /* User explicitly requested empty subsystem */
1128 struct cgroupfs_root
*new_root
;
1133 * Convert a hierarchy specifier into a bitmask of subsystems and
1134 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1135 * array. This function takes refcounts on subsystems to be used, unless it
1136 * returns error, in which case no refcounts are taken.
1138 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1140 char *token
, *o
= data
;
1141 bool all_ss
= false, one_ss
= false;
1142 unsigned long mask
= (unsigned long)-1;
1143 struct cgroup_subsys
*ss
;
1146 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1148 #ifdef CONFIG_CPUSETS
1149 mask
= ~(1UL << cpuset_subsys_id
);
1152 memset(opts
, 0, sizeof(*opts
));
1154 while ((token
= strsep(&o
, ",")) != NULL
) {
1157 if (!strcmp(token
, "none")) {
1158 /* Explicitly have no subsystems */
1162 if (!strcmp(token
, "all")) {
1163 /* Mutually exclusive option 'all' + subsystem name */
1169 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1170 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1173 if (!strcmp(token
, "noprefix")) {
1174 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1177 if (!strcmp(token
, "clone_children")) {
1178 opts
->cpuset_clone_children
= true;
1181 if (!strcmp(token
, "xattr")) {
1182 opts
->flags
|= CGRP_ROOT_XATTR
;
1185 if (!strncmp(token
, "release_agent=", 14)) {
1186 /* Specifying two release agents is forbidden */
1187 if (opts
->release_agent
)
1189 opts
->release_agent
=
1190 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1191 if (!opts
->release_agent
)
1195 if (!strncmp(token
, "name=", 5)) {
1196 const char *name
= token
+ 5;
1197 /* Can't specify an empty name */
1200 /* Must match [\w.-]+ */
1201 for (i
= 0; i
< strlen(name
); i
++) {
1205 if ((c
== '.') || (c
== '-') || (c
== '_'))
1209 /* Specifying two names is forbidden */
1212 opts
->name
= kstrndup(name
,
1213 MAX_CGROUP_ROOT_NAMELEN
- 1,
1221 for_each_subsys(ss
, i
) {
1222 if (strcmp(token
, ss
->name
))
1227 /* Mutually exclusive option 'all' + subsystem name */
1230 set_bit(i
, &opts
->subsys_mask
);
1235 if (i
== CGROUP_SUBSYS_COUNT
)
1240 * If the 'all' option was specified select all the subsystems,
1241 * otherwise if 'none', 'name=' and a subsystem name options
1242 * were not specified, let's default to 'all'
1244 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1245 for_each_subsys(ss
, i
)
1247 set_bit(i
, &opts
->subsys_mask
);
1249 /* Consistency checks */
1251 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1252 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1254 if (opts
->flags
& CGRP_ROOT_NOPREFIX
) {
1255 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1259 if (opts
->cpuset_clone_children
) {
1260 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1266 * Option noprefix was introduced just for backward compatibility
1267 * with the old cpuset, so we allow noprefix only if mounting just
1268 * the cpuset subsystem.
1270 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1274 /* Can't specify "none" and some subsystems */
1275 if (opts
->subsys_mask
&& opts
->none
)
1279 * We either have to specify by name or by subsystems. (So all
1280 * empty hierarchies must have a name).
1282 if (!opts
->subsys_mask
&& !opts
->name
)
1288 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1291 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1292 struct cgroup
*cgrp
= &root
->top_cgroup
;
1293 struct cgroup_sb_opts opts
;
1294 unsigned long added_mask
, removed_mask
;
1296 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1297 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1301 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1302 mutex_lock(&cgroup_mutex
);
1303 mutex_lock(&cgroup_root_mutex
);
1305 /* See what subsystems are wanted */
1306 ret
= parse_cgroupfs_options(data
, &opts
);
1310 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1311 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1312 task_tgid_nr(current
), current
->comm
);
1314 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1315 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1317 /* Don't allow flags or name to change at remount */
1318 if (((opts
.flags
^ root
->flags
) & CGRP_ROOT_OPTION_MASK
) ||
1319 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1320 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1321 opts
.flags
& CGRP_ROOT_OPTION_MASK
, opts
.name
?: "",
1322 root
->flags
& CGRP_ROOT_OPTION_MASK
, root
->name
);
1327 /* remounting is not allowed for populated hierarchies */
1328 if (root
->number_of_cgroups
> 1) {
1333 ret
= rebind_subsystems(root
, added_mask
, removed_mask
);
1337 if (opts
.release_agent
)
1338 strcpy(root
->release_agent_path
, opts
.release_agent
);
1340 kfree(opts
.release_agent
);
1342 mutex_unlock(&cgroup_root_mutex
);
1343 mutex_unlock(&cgroup_mutex
);
1344 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1348 static const struct super_operations cgroup_ops
= {
1349 .statfs
= simple_statfs
,
1350 .drop_inode
= generic_delete_inode
,
1351 .show_options
= cgroup_show_options
,
1352 .remount_fs
= cgroup_remount
,
1355 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1357 INIT_LIST_HEAD(&cgrp
->sibling
);
1358 INIT_LIST_HEAD(&cgrp
->children
);
1359 INIT_LIST_HEAD(&cgrp
->files
);
1360 INIT_LIST_HEAD(&cgrp
->cset_links
);
1361 INIT_LIST_HEAD(&cgrp
->release_list
);
1362 INIT_LIST_HEAD(&cgrp
->pidlists
);
1363 mutex_init(&cgrp
->pidlist_mutex
);
1364 cgrp
->dummy_css
.cgroup
= cgrp
;
1365 INIT_LIST_HEAD(&cgrp
->event_list
);
1366 spin_lock_init(&cgrp
->event_list_lock
);
1367 simple_xattrs_init(&cgrp
->xattrs
);
1370 static void init_cgroup_root(struct cgroupfs_root
*root
)
1372 struct cgroup
*cgrp
= &root
->top_cgroup
;
1374 INIT_LIST_HEAD(&root
->subsys_list
);
1375 INIT_LIST_HEAD(&root
->root_list
);
1376 root
->number_of_cgroups
= 1;
1378 RCU_INIT_POINTER(cgrp
->name
, &root_cgroup_name
);
1379 init_cgroup_housekeeping(cgrp
);
1380 idr_init(&root
->cgroup_idr
);
1383 static int cgroup_init_root_id(struct cgroupfs_root
*root
, int start
, int end
)
1387 lockdep_assert_held(&cgroup_mutex
);
1388 lockdep_assert_held(&cgroup_root_mutex
);
1390 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, start
, end
,
1395 root
->hierarchy_id
= id
;
1399 static void cgroup_exit_root_id(struct cgroupfs_root
*root
)
1401 lockdep_assert_held(&cgroup_mutex
);
1402 lockdep_assert_held(&cgroup_root_mutex
);
1404 if (root
->hierarchy_id
) {
1405 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1406 root
->hierarchy_id
= 0;
1410 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1412 struct cgroup_sb_opts
*opts
= data
;
1413 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1415 /* If we asked for a name then it must match */
1416 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1420 * If we asked for subsystems (or explicitly for no
1421 * subsystems) then they must match
1423 if ((opts
->subsys_mask
|| opts
->none
)
1424 && (opts
->subsys_mask
!= root
->subsys_mask
))
1430 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1432 struct cgroupfs_root
*root
;
1434 if (!opts
->subsys_mask
&& !opts
->none
)
1437 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1439 return ERR_PTR(-ENOMEM
);
1441 init_cgroup_root(root
);
1444 * We need to set @root->subsys_mask now so that @root can be
1445 * matched by cgroup_test_super() before it finishes
1446 * initialization; otherwise, competing mounts with the same
1447 * options may try to bind the same subsystems instead of waiting
1448 * for the first one leading to unexpected mount errors.
1449 * SUBSYS_BOUND will be set once actual binding is complete.
1451 root
->subsys_mask
= opts
->subsys_mask
;
1452 root
->flags
= opts
->flags
;
1453 if (opts
->release_agent
)
1454 strcpy(root
->release_agent_path
, opts
->release_agent
);
1456 strcpy(root
->name
, opts
->name
);
1457 if (opts
->cpuset_clone_children
)
1458 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1462 static void cgroup_free_root(struct cgroupfs_root
*root
)
1465 /* hierarhcy ID shoulid already have been released */
1466 WARN_ON_ONCE(root
->hierarchy_id
);
1468 idr_destroy(&root
->cgroup_idr
);
1473 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1476 struct cgroup_sb_opts
*opts
= data
;
1478 /* If we don't have a new root, we can't set up a new sb */
1479 if (!opts
->new_root
)
1482 BUG_ON(!opts
->subsys_mask
&& !opts
->none
);
1484 ret
= set_anon_super(sb
, NULL
);
1488 sb
->s_fs_info
= opts
->new_root
;
1489 opts
->new_root
->sb
= sb
;
1491 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1492 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1493 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1494 sb
->s_op
= &cgroup_ops
;
1499 static int cgroup_get_rootdir(struct super_block
*sb
)
1501 static const struct dentry_operations cgroup_dops
= {
1502 .d_iput
= cgroup_diput
,
1503 .d_delete
= always_delete_dentry
,
1506 struct inode
*inode
=
1507 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1512 inode
->i_fop
= &simple_dir_operations
;
1513 inode
->i_op
= &cgroup_dir_inode_operations
;
1514 /* directories start off with i_nlink == 2 (for "." entry) */
1516 sb
->s_root
= d_make_root(inode
);
1519 /* for everything else we want ->d_op set */
1520 sb
->s_d_op
= &cgroup_dops
;
1524 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1525 int flags
, const char *unused_dev_name
,
1528 struct cgroup_sb_opts opts
;
1529 struct cgroupfs_root
*root
;
1531 struct super_block
*sb
;
1532 struct cgroupfs_root
*new_root
;
1533 struct list_head tmp_links
;
1534 struct inode
*inode
;
1535 const struct cred
*cred
;
1537 /* First find the desired set of subsystems */
1538 mutex_lock(&cgroup_mutex
);
1539 ret
= parse_cgroupfs_options(data
, &opts
);
1540 mutex_unlock(&cgroup_mutex
);
1545 * Allocate a new cgroup root. We may not need it if we're
1546 * reusing an existing hierarchy.
1548 new_root
= cgroup_root_from_opts(&opts
);
1549 if (IS_ERR(new_root
)) {
1550 ret
= PTR_ERR(new_root
);
1553 opts
.new_root
= new_root
;
1555 /* Locate an existing or new sb for this hierarchy */
1556 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, 0, &opts
);
1559 cgroup_free_root(opts
.new_root
);
1563 root
= sb
->s_fs_info
;
1565 if (root
== opts
.new_root
) {
1566 /* We used the new root structure, so this is a new hierarchy */
1567 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1568 struct cgroupfs_root
*existing_root
;
1570 struct css_set
*cset
;
1572 BUG_ON(sb
->s_root
!= NULL
);
1574 ret
= cgroup_get_rootdir(sb
);
1576 goto drop_new_super
;
1577 inode
= sb
->s_root
->d_inode
;
1579 mutex_lock(&inode
->i_mutex
);
1580 mutex_lock(&cgroup_mutex
);
1581 mutex_lock(&cgroup_root_mutex
);
1583 root_cgrp
->id
= idr_alloc(&root
->cgroup_idr
, root_cgrp
,
1585 if (root_cgrp
->id
< 0)
1588 /* Check for name clashes with existing mounts */
1590 if (strlen(root
->name
))
1591 for_each_active_root(existing_root
)
1592 if (!strcmp(existing_root
->name
, root
->name
))
1596 * We're accessing css_set_count without locking
1597 * css_set_lock here, but that's OK - it can only be
1598 * increased by someone holding cgroup_lock, and
1599 * that's us. The worst that can happen is that we
1600 * have some link structures left over
1602 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1606 /* ID 0 is reserved for dummy root, 1 for unified hierarchy */
1607 ret
= cgroup_init_root_id(root
, 2, 0);
1611 sb
->s_root
->d_fsdata
= root_cgrp
;
1612 root_cgrp
->dentry
= sb
->s_root
;
1615 * We're inside get_sb() and will call lookup_one_len() to
1616 * create the root files, which doesn't work if SELinux is
1617 * in use. The following cred dancing somehow works around
1618 * it. See 2ce9738ba ("cgroupfs: use init_cred when
1619 * populating new cgroupfs mount") for more details.
1621 cred
= override_creds(&init_cred
);
1623 ret
= cgroup_addrm_files(root_cgrp
, cgroup_base_files
, true);
1627 ret
= rebind_subsystems(root
, root
->subsys_mask
, 0);
1634 * There must be no failure case after here, since rebinding
1635 * takes care of subsystems' refcounts, which are explicitly
1636 * dropped in the failure exit path.
1639 list_add(&root
->root_list
, &cgroup_roots
);
1640 cgroup_root_count
++;
1642 /* Link the top cgroup in this hierarchy into all
1643 * the css_set objects */
1644 write_lock(&css_set_lock
);
1645 hash_for_each(css_set_table
, i
, cset
, hlist
)
1646 link_css_set(&tmp_links
, cset
, root_cgrp
);
1647 write_unlock(&css_set_lock
);
1649 free_cgrp_cset_links(&tmp_links
);
1651 BUG_ON(!list_empty(&root_cgrp
->children
));
1652 BUG_ON(root
->number_of_cgroups
!= 1);
1654 mutex_unlock(&cgroup_root_mutex
);
1655 mutex_unlock(&cgroup_mutex
);
1656 mutex_unlock(&inode
->i_mutex
);
1659 * We re-used an existing hierarchy - the new root (if
1660 * any) is not needed
1662 cgroup_free_root(opts
.new_root
);
1664 if ((root
->flags
^ opts
.flags
) & CGRP_ROOT_OPTION_MASK
) {
1665 if ((root
->flags
| opts
.flags
) & CGRP_ROOT_SANE_BEHAVIOR
) {
1666 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1668 goto drop_new_super
;
1670 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1675 kfree(opts
.release_agent
);
1677 return dget(sb
->s_root
);
1680 free_cgrp_cset_links(&tmp_links
);
1681 cgroup_addrm_files(&root
->top_cgroup
, cgroup_base_files
, false);
1684 cgroup_exit_root_id(root
);
1685 mutex_unlock(&cgroup_root_mutex
);
1686 mutex_unlock(&cgroup_mutex
);
1687 mutex_unlock(&inode
->i_mutex
);
1689 deactivate_locked_super(sb
);
1691 kfree(opts
.release_agent
);
1693 return ERR_PTR(ret
);
1696 static void cgroup_kill_sb(struct super_block
*sb
) {
1697 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1698 struct cgroup
*cgrp
= &root
->top_cgroup
;
1699 struct cgrp_cset_link
*link
, *tmp_link
;
1704 BUG_ON(root
->number_of_cgroups
!= 1);
1705 BUG_ON(!list_empty(&cgrp
->children
));
1707 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1708 mutex_lock(&cgroup_mutex
);
1709 mutex_lock(&cgroup_root_mutex
);
1711 /* Rebind all subsystems back to the default hierarchy */
1712 if (root
->flags
& CGRP_ROOT_SUBSYS_BOUND
) {
1713 ret
= rebind_subsystems(root
, 0, root
->subsys_mask
);
1714 /* Shouldn't be able to fail ... */
1719 * Release all the links from cset_links to this hierarchy's
1722 write_lock(&css_set_lock
);
1724 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1725 list_del(&link
->cset_link
);
1726 list_del(&link
->cgrp_link
);
1729 write_unlock(&css_set_lock
);
1731 if (!list_empty(&root
->root_list
)) {
1732 list_del(&root
->root_list
);
1733 cgroup_root_count
--;
1736 cgroup_exit_root_id(root
);
1738 mutex_unlock(&cgroup_root_mutex
);
1739 mutex_unlock(&cgroup_mutex
);
1740 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1742 simple_xattrs_free(&cgrp
->xattrs
);
1744 kill_litter_super(sb
);
1745 cgroup_free_root(root
);
1748 static struct file_system_type cgroup_fs_type
= {
1750 .mount
= cgroup_mount
,
1751 .kill_sb
= cgroup_kill_sb
,
1754 static struct kobject
*cgroup_kobj
;
1757 * cgroup_path - generate the path of a cgroup
1758 * @cgrp: the cgroup in question
1759 * @buf: the buffer to write the path into
1760 * @buflen: the length of the buffer
1762 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1764 * We can't generate cgroup path using dentry->d_name, as accessing
1765 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1766 * inode's i_mutex, while on the other hand cgroup_path() can be called
1767 * with some irq-safe spinlocks held.
1769 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1771 int ret
= -ENAMETOOLONG
;
1774 if (!cgrp
->parent
) {
1775 if (strlcpy(buf
, "/", buflen
) >= buflen
)
1776 return -ENAMETOOLONG
;
1780 start
= buf
+ buflen
- 1;
1785 const char *name
= cgroup_name(cgrp
);
1789 if ((start
-= len
) < buf
)
1791 memcpy(start
, name
, len
);
1797 cgrp
= cgrp
->parent
;
1798 } while (cgrp
->parent
);
1800 memmove(buf
, start
, buf
+ buflen
- start
);
1805 EXPORT_SYMBOL_GPL(cgroup_path
);
1808 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1809 * @task: target task
1810 * @buf: the buffer to write the path into
1811 * @buflen: the length of the buffer
1813 * Determine @task's cgroup on the first (the one with the lowest non-zero
1814 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1815 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1816 * cgroup controller callbacks.
1818 * Returns 0 on success, fails with -%ENAMETOOLONG if @buflen is too short.
1820 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
1822 struct cgroupfs_root
*root
;
1823 struct cgroup
*cgrp
;
1824 int hierarchy_id
= 1, ret
= 0;
1827 return -ENAMETOOLONG
;
1829 mutex_lock(&cgroup_mutex
);
1831 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
1834 cgrp
= task_cgroup_from_root(task
, root
);
1835 ret
= cgroup_path(cgrp
, buf
, buflen
);
1837 /* if no hierarchy exists, everyone is in "/" */
1838 memcpy(buf
, "/", 2);
1841 mutex_unlock(&cgroup_mutex
);
1844 EXPORT_SYMBOL_GPL(task_cgroup_path
);
1847 * Control Group taskset
1849 struct task_and_cgroup
{
1850 struct task_struct
*task
;
1851 struct cgroup
*cgrp
;
1852 struct css_set
*cset
;
1855 struct cgroup_taskset
{
1856 struct task_and_cgroup single
;
1857 struct flex_array
*tc_array
;
1860 struct cgroup
*cur_cgrp
;
1864 * cgroup_taskset_first - reset taskset and return the first task
1865 * @tset: taskset of interest
1867 * @tset iteration is initialized and the first task is returned.
1869 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1871 if (tset
->tc_array
) {
1873 return cgroup_taskset_next(tset
);
1875 tset
->cur_cgrp
= tset
->single
.cgrp
;
1876 return tset
->single
.task
;
1879 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1882 * cgroup_taskset_next - iterate to the next task in taskset
1883 * @tset: taskset of interest
1885 * Return the next task in @tset. Iteration must have been initialized
1886 * with cgroup_taskset_first().
1888 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1890 struct task_and_cgroup
*tc
;
1892 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1895 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1896 tset
->cur_cgrp
= tc
->cgrp
;
1899 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1902 * cgroup_taskset_cur_css - return the matching css for the current task
1903 * @tset: taskset of interest
1904 * @subsys_id: the ID of the target subsystem
1906 * Return the css for the current (last returned) task of @tset for
1907 * subsystem specified by @subsys_id. This function must be preceded by
1908 * either cgroup_taskset_first() or cgroup_taskset_next().
1910 struct cgroup_subsys_state
*cgroup_taskset_cur_css(struct cgroup_taskset
*tset
,
1913 return cgroup_css(tset
->cur_cgrp
, cgroup_subsys
[subsys_id
]);
1915 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_css
);
1918 * cgroup_taskset_size - return the number of tasks in taskset
1919 * @tset: taskset of interest
1921 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1923 return tset
->tc_array
? tset
->tc_array_len
: 1;
1925 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1929 * cgroup_task_migrate - move a task from one cgroup to another.
1931 * Must be called with cgroup_mutex and threadgroup locked.
1933 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
1934 struct task_struct
*tsk
,
1935 struct css_set
*new_cset
)
1937 struct css_set
*old_cset
;
1940 * We are synchronized through threadgroup_lock() against PF_EXITING
1941 * setting such that we can't race against cgroup_exit() changing the
1942 * css_set to init_css_set and dropping the old one.
1944 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1945 old_cset
= task_css_set(tsk
);
1948 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
1951 /* Update the css_set linked lists if we're using them */
1952 write_lock(&css_set_lock
);
1953 if (!list_empty(&tsk
->cg_list
))
1954 list_move(&tsk
->cg_list
, &new_cset
->tasks
);
1955 write_unlock(&css_set_lock
);
1958 * We just gained a reference on old_cset by taking it from the
1959 * task. As trading it for new_cset is protected by cgroup_mutex,
1960 * we're safe to drop it here; it will be freed under RCU.
1962 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
1963 put_css_set(old_cset
);
1967 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1968 * @cgrp: the cgroup to attach to
1969 * @tsk: the task or the leader of the threadgroup to be attached
1970 * @threadgroup: attach the whole threadgroup?
1972 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1973 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1975 static int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
,
1978 int retval
, i
, group_size
;
1979 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1980 struct cgroupfs_root
*root
= cgrp
->root
;
1981 /* threadgroup list cursor and array */
1982 struct task_struct
*leader
= tsk
;
1983 struct task_and_cgroup
*tc
;
1984 struct flex_array
*group
;
1985 struct cgroup_taskset tset
= { };
1988 * step 0: in order to do expensive, possibly blocking operations for
1989 * every thread, we cannot iterate the thread group list, since it needs
1990 * rcu or tasklist locked. instead, build an array of all threads in the
1991 * group - group_rwsem prevents new threads from appearing, and if
1992 * threads exit, this will just be an over-estimate.
1995 group_size
= get_nr_threads(tsk
);
1998 /* flex_array supports very large thread-groups better than kmalloc. */
1999 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2002 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2003 retval
= flex_array_prealloc(group
, 0, group_size
, GFP_KERNEL
);
2005 goto out_free_group_list
;
2009 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2010 * already PF_EXITING could be freed from underneath us unless we
2011 * take an rcu_read_lock.
2015 struct task_and_cgroup ent
;
2017 /* @tsk either already exited or can't exit until the end */
2018 if (tsk
->flags
& PF_EXITING
)
2021 /* as per above, nr_threads may decrease, but not increase. */
2022 BUG_ON(i
>= group_size
);
2024 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2025 /* nothing to do if this task is already in the cgroup */
2026 if (ent
.cgrp
== cgrp
)
2029 * saying GFP_ATOMIC has no effect here because we did prealloc
2030 * earlier, but it's good form to communicate our expectations.
2032 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2033 BUG_ON(retval
!= 0);
2038 } while_each_thread(leader
, tsk
);
2040 /* remember the number of threads in the array for later. */
2042 tset
.tc_array
= group
;
2043 tset
.tc_array_len
= group_size
;
2045 /* methods shouldn't be called if no task is actually migrating */
2048 goto out_free_group_list
;
2051 * step 1: check that we can legitimately attach to the cgroup.
2053 for_each_root_subsys(root
, ss
) {
2054 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, ss
);
2056 if (ss
->can_attach
) {
2057 retval
= ss
->can_attach(css
, &tset
);
2060 goto out_cancel_attach
;
2066 * step 2: make sure css_sets exist for all threads to be migrated.
2067 * we use find_css_set, which allocates a new one if necessary.
2069 for (i
= 0; i
< group_size
; i
++) {
2070 struct css_set
*old_cset
;
2072 tc
= flex_array_get(group
, i
);
2073 old_cset
= task_css_set(tc
->task
);
2074 tc
->cset
= find_css_set(old_cset
, cgrp
);
2077 goto out_put_css_set_refs
;
2082 * step 3: now that we're guaranteed success wrt the css_sets,
2083 * proceed to move all tasks to the new cgroup. There are no
2084 * failure cases after here, so this is the commit point.
2086 for (i
= 0; i
< group_size
; i
++) {
2087 tc
= flex_array_get(group
, i
);
2088 cgroup_task_migrate(tc
->cgrp
, tc
->task
, tc
->cset
);
2090 /* nothing is sensitive to fork() after this point. */
2093 * step 4: do subsystem attach callbacks.
2095 for_each_root_subsys(root
, ss
) {
2096 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, ss
);
2099 ss
->attach(css
, &tset
);
2103 * step 5: success! and cleanup
2106 out_put_css_set_refs
:
2108 for (i
= 0; i
< group_size
; i
++) {
2109 tc
= flex_array_get(group
, i
);
2112 put_css_set(tc
->cset
);
2117 for_each_root_subsys(root
, ss
) {
2118 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, ss
);
2120 if (ss
== failed_ss
)
2122 if (ss
->cancel_attach
)
2123 ss
->cancel_attach(css
, &tset
);
2126 out_free_group_list
:
2127 flex_array_free(group
);
2132 * Find the task_struct of the task to attach by vpid and pass it along to the
2133 * function to attach either it or all tasks in its threadgroup. Will lock
2134 * cgroup_mutex and threadgroup; may take task_lock of task.
2136 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2138 struct task_struct
*tsk
;
2139 const struct cred
*cred
= current_cred(), *tcred
;
2142 if (!cgroup_lock_live_group(cgrp
))
2148 tsk
= find_task_by_vpid(pid
);
2152 goto out_unlock_cgroup
;
2155 * even if we're attaching all tasks in the thread group, we
2156 * only need to check permissions on one of them.
2158 tcred
= __task_cred(tsk
);
2159 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2160 !uid_eq(cred
->euid
, tcred
->uid
) &&
2161 !uid_eq(cred
->euid
, tcred
->suid
)) {
2164 goto out_unlock_cgroup
;
2170 tsk
= tsk
->group_leader
;
2173 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2174 * trapped in a cpuset, or RT worker may be born in a cgroup
2175 * with no rt_runtime allocated. Just say no.
2177 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2180 goto out_unlock_cgroup
;
2183 get_task_struct(tsk
);
2186 threadgroup_lock(tsk
);
2188 if (!thread_group_leader(tsk
)) {
2190 * a race with de_thread from another thread's exec()
2191 * may strip us of our leadership, if this happens,
2192 * there is no choice but to throw this task away and
2193 * try again; this is
2194 * "double-double-toil-and-trouble-check locking".
2196 threadgroup_unlock(tsk
);
2197 put_task_struct(tsk
);
2198 goto retry_find_task
;
2202 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2204 threadgroup_unlock(tsk
);
2206 put_task_struct(tsk
);
2208 mutex_unlock(&cgroup_mutex
);
2213 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2214 * @from: attach to all cgroups of a given task
2215 * @tsk: the task to be attached
2217 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2219 struct cgroupfs_root
*root
;
2222 mutex_lock(&cgroup_mutex
);
2223 for_each_active_root(root
) {
2224 struct cgroup
*from_cgrp
= task_cgroup_from_root(from
, root
);
2226 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2230 mutex_unlock(&cgroup_mutex
);
2234 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2236 static int cgroup_tasks_write(struct cgroup_subsys_state
*css
,
2237 struct cftype
*cft
, u64 pid
)
2239 return attach_task_by_pid(css
->cgroup
, pid
, false);
2242 static int cgroup_procs_write(struct cgroup_subsys_state
*css
,
2243 struct cftype
*cft
, u64 tgid
)
2245 return attach_task_by_pid(css
->cgroup
, tgid
, true);
2248 static int cgroup_release_agent_write(struct cgroup_subsys_state
*css
,
2249 struct cftype
*cft
, const char *buffer
)
2251 BUILD_BUG_ON(sizeof(css
->cgroup
->root
->release_agent_path
) < PATH_MAX
);
2252 if (strlen(buffer
) >= PATH_MAX
)
2254 if (!cgroup_lock_live_group(css
->cgroup
))
2256 mutex_lock(&cgroup_root_mutex
);
2257 strcpy(css
->cgroup
->root
->release_agent_path
, buffer
);
2258 mutex_unlock(&cgroup_root_mutex
);
2259 mutex_unlock(&cgroup_mutex
);
2263 static int cgroup_release_agent_show(struct cgroup_subsys_state
*css
,
2264 struct cftype
*cft
, struct seq_file
*seq
)
2266 struct cgroup
*cgrp
= css
->cgroup
;
2268 if (!cgroup_lock_live_group(cgrp
))
2270 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2271 seq_putc(seq
, '\n');
2272 mutex_unlock(&cgroup_mutex
);
2276 static int cgroup_sane_behavior_show(struct cgroup_subsys_state
*css
,
2277 struct cftype
*cft
, struct seq_file
*seq
)
2279 seq_printf(seq
, "%d\n", cgroup_sane_behavior(css
->cgroup
));
2283 /* A buffer size big enough for numbers or short strings */
2284 #define CGROUP_LOCAL_BUFFER_SIZE 64
2286 static ssize_t
cgroup_write_X64(struct cgroup_subsys_state
*css
,
2287 struct cftype
*cft
, struct file
*file
,
2288 const char __user
*userbuf
, size_t nbytes
,
2289 loff_t
*unused_ppos
)
2291 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2297 if (nbytes
>= sizeof(buffer
))
2299 if (copy_from_user(buffer
, userbuf
, nbytes
))
2302 buffer
[nbytes
] = 0; /* nul-terminate */
2303 if (cft
->write_u64
) {
2304 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2307 retval
= cft
->write_u64(css
, cft
, val
);
2309 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2312 retval
= cft
->write_s64(css
, cft
, val
);
2319 static ssize_t
cgroup_write_string(struct cgroup_subsys_state
*css
,
2320 struct cftype
*cft
, struct file
*file
,
2321 const char __user
*userbuf
, size_t nbytes
,
2322 loff_t
*unused_ppos
)
2324 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2326 size_t max_bytes
= cft
->max_write_len
;
2327 char *buffer
= local_buffer
;
2330 max_bytes
= sizeof(local_buffer
) - 1;
2331 if (nbytes
>= max_bytes
)
2333 /* Allocate a dynamic buffer if we need one */
2334 if (nbytes
>= sizeof(local_buffer
)) {
2335 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2339 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2344 buffer
[nbytes
] = 0; /* nul-terminate */
2345 retval
= cft
->write_string(css
, cft
, strstrip(buffer
));
2349 if (buffer
!= local_buffer
)
2354 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2355 size_t nbytes
, loff_t
*ppos
)
2357 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2358 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2359 struct cgroup_subsys_state
*css
= cfe
->css
;
2362 return cft
->write(css
, cft
, file
, buf
, nbytes
, ppos
);
2363 if (cft
->write_u64
|| cft
->write_s64
)
2364 return cgroup_write_X64(css
, cft
, file
, buf
, nbytes
, ppos
);
2365 if (cft
->write_string
)
2366 return cgroup_write_string(css
, cft
, file
, buf
, nbytes
, ppos
);
2368 int ret
= cft
->trigger(css
, (unsigned int)cft
->private);
2369 return ret
? ret
: nbytes
;
2374 static ssize_t
cgroup_read_u64(struct cgroup_subsys_state
*css
,
2375 struct cftype
*cft
, struct file
*file
,
2376 char __user
*buf
, size_t nbytes
, loff_t
*ppos
)
2378 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2379 u64 val
= cft
->read_u64(css
, cft
);
2380 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2382 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2385 static ssize_t
cgroup_read_s64(struct cgroup_subsys_state
*css
,
2386 struct cftype
*cft
, struct file
*file
,
2387 char __user
*buf
, size_t nbytes
, loff_t
*ppos
)
2389 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2390 s64 val
= cft
->read_s64(css
, cft
);
2391 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2393 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2396 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2397 size_t nbytes
, loff_t
*ppos
)
2399 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2400 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2401 struct cgroup_subsys_state
*css
= cfe
->css
;
2404 return cft
->read(css
, cft
, file
, buf
, nbytes
, ppos
);
2406 return cgroup_read_u64(css
, cft
, file
, buf
, nbytes
, ppos
);
2408 return cgroup_read_s64(css
, cft
, file
, buf
, nbytes
, ppos
);
2413 * seqfile ops/methods for returning structured data. Currently just
2414 * supports string->u64 maps, but can be extended in future.
2417 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2419 struct seq_file
*sf
= cb
->state
;
2420 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2423 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2425 struct cfent
*cfe
= m
->private;
2426 struct cftype
*cft
= cfe
->type
;
2427 struct cgroup_subsys_state
*css
= cfe
->css
;
2429 if (cft
->read_map
) {
2430 struct cgroup_map_cb cb
= {
2431 .fill
= cgroup_map_add
,
2434 return cft
->read_map(css
, cft
, &cb
);
2436 return cft
->read_seq_string(css
, cft
, m
);
2439 static const struct file_operations cgroup_seqfile_operations
= {
2441 .write
= cgroup_file_write
,
2442 .llseek
= seq_lseek
,
2443 .release
= cgroup_file_release
,
2446 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2448 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2449 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2450 struct cgroup
*cgrp
= __d_cgrp(cfe
->dentry
->d_parent
);
2451 struct cgroup_subsys_state
*css
;
2454 err
= generic_file_open(inode
, file
);
2459 * If the file belongs to a subsystem, pin the css. Will be
2460 * unpinned either on open failure or release. This ensures that
2461 * @css stays alive for all file operations.
2464 css
= cgroup_css(cgrp
, cft
->ss
);
2465 if (cft
->ss
&& !css_tryget(css
))
2473 * @cfe->css is used by read/write/close to determine the
2474 * associated css. @file->private_data would be a better place but
2475 * that's already used by seqfile. Multiple accessors may use it
2476 * simultaneously which is okay as the association never changes.
2478 WARN_ON_ONCE(cfe
->css
&& cfe
->css
!= css
);
2481 if (cft
->read_map
|| cft
->read_seq_string
) {
2482 file
->f_op
= &cgroup_seqfile_operations
;
2483 err
= single_open(file
, cgroup_seqfile_show
, cfe
);
2484 } else if (cft
->open
) {
2485 err
= cft
->open(inode
, file
);
2493 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2495 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2496 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2497 struct cgroup_subsys_state
*css
= cfe
->css
;
2501 ret
= cft
->release(inode
, file
);
2504 if (file
->f_op
== &cgroup_seqfile_operations
)
2505 single_release(inode
, file
);
2510 * cgroup_rename - Only allow simple rename of directories in place.
2512 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2513 struct inode
*new_dir
, struct dentry
*new_dentry
)
2516 struct cgroup_name
*name
, *old_name
;
2517 struct cgroup
*cgrp
;
2520 * It's convinient to use parent dir's i_mutex to protected
2523 lockdep_assert_held(&old_dir
->i_mutex
);
2525 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2527 if (new_dentry
->d_inode
)
2529 if (old_dir
!= new_dir
)
2532 cgrp
= __d_cgrp(old_dentry
);
2535 * This isn't a proper migration and its usefulness is very
2536 * limited. Disallow if sane_behavior.
2538 if (cgroup_sane_behavior(cgrp
))
2541 name
= cgroup_alloc_name(new_dentry
);
2545 ret
= simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2551 old_name
= rcu_dereference_protected(cgrp
->name
, true);
2552 rcu_assign_pointer(cgrp
->name
, name
);
2554 kfree_rcu(old_name
, rcu_head
);
2558 static struct simple_xattrs
*__d_xattrs(struct dentry
*dentry
)
2560 if (S_ISDIR(dentry
->d_inode
->i_mode
))
2561 return &__d_cgrp(dentry
)->xattrs
;
2563 return &__d_cfe(dentry
)->xattrs
;
2566 static inline int xattr_enabled(struct dentry
*dentry
)
2568 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
2569 return root
->flags
& CGRP_ROOT_XATTR
;
2572 static bool is_valid_xattr(const char *name
)
2574 if (!strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) ||
2575 !strncmp(name
, XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
))
2580 static int cgroup_setxattr(struct dentry
*dentry
, const char *name
,
2581 const void *val
, size_t size
, int flags
)
2583 if (!xattr_enabled(dentry
))
2585 if (!is_valid_xattr(name
))
2587 return simple_xattr_set(__d_xattrs(dentry
), name
, val
, size
, flags
);
2590 static int cgroup_removexattr(struct dentry
*dentry
, const char *name
)
2592 if (!xattr_enabled(dentry
))
2594 if (!is_valid_xattr(name
))
2596 return simple_xattr_remove(__d_xattrs(dentry
), name
);
2599 static ssize_t
cgroup_getxattr(struct dentry
*dentry
, const char *name
,
2600 void *buf
, size_t size
)
2602 if (!xattr_enabled(dentry
))
2604 if (!is_valid_xattr(name
))
2606 return simple_xattr_get(__d_xattrs(dentry
), name
, buf
, size
);
2609 static ssize_t
cgroup_listxattr(struct dentry
*dentry
, char *buf
, size_t size
)
2611 if (!xattr_enabled(dentry
))
2613 return simple_xattr_list(__d_xattrs(dentry
), buf
, size
);
2616 static const struct file_operations cgroup_file_operations
= {
2617 .read
= cgroup_file_read
,
2618 .write
= cgroup_file_write
,
2619 .llseek
= generic_file_llseek
,
2620 .open
= cgroup_file_open
,
2621 .release
= cgroup_file_release
,
2624 static const struct inode_operations cgroup_file_inode_operations
= {
2625 .setxattr
= cgroup_setxattr
,
2626 .getxattr
= cgroup_getxattr
,
2627 .listxattr
= cgroup_listxattr
,
2628 .removexattr
= cgroup_removexattr
,
2631 static const struct inode_operations cgroup_dir_inode_operations
= {
2632 .lookup
= simple_lookup
,
2633 .mkdir
= cgroup_mkdir
,
2634 .rmdir
= cgroup_rmdir
,
2635 .rename
= cgroup_rename
,
2636 .setxattr
= cgroup_setxattr
,
2637 .getxattr
= cgroup_getxattr
,
2638 .listxattr
= cgroup_listxattr
,
2639 .removexattr
= cgroup_removexattr
,
2643 * Check if a file is a control file
2645 static inline struct cftype
*__file_cft(struct file
*file
)
2647 if (file_inode(file
)->i_fop
!= &cgroup_file_operations
)
2648 return ERR_PTR(-EINVAL
);
2649 return __d_cft(file
->f_dentry
);
2652 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2653 struct super_block
*sb
)
2655 struct inode
*inode
;
2659 if (dentry
->d_inode
)
2662 inode
= cgroup_new_inode(mode
, sb
);
2666 if (S_ISDIR(mode
)) {
2667 inode
->i_op
= &cgroup_dir_inode_operations
;
2668 inode
->i_fop
= &simple_dir_operations
;
2670 /* start off with i_nlink == 2 (for "." entry) */
2672 inc_nlink(dentry
->d_parent
->d_inode
);
2675 * Control reaches here with cgroup_mutex held.
2676 * @inode->i_mutex should nest outside cgroup_mutex but we
2677 * want to populate it immediately without releasing
2678 * cgroup_mutex. As @inode isn't visible to anyone else
2679 * yet, trylock will always succeed without affecting
2682 WARN_ON_ONCE(!mutex_trylock(&inode
->i_mutex
));
2683 } else if (S_ISREG(mode
)) {
2685 inode
->i_fop
= &cgroup_file_operations
;
2686 inode
->i_op
= &cgroup_file_inode_operations
;
2688 d_instantiate(dentry
, inode
);
2689 dget(dentry
); /* Extra count - pin the dentry in core */
2694 * cgroup_file_mode - deduce file mode of a control file
2695 * @cft: the control file in question
2697 * returns cft->mode if ->mode is not 0
2698 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2699 * returns S_IRUGO if it has only a read handler
2700 * returns S_IWUSR if it has only a write hander
2702 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2709 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2710 cft
->read_map
|| cft
->read_seq_string
)
2713 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2714 cft
->write_string
|| cft
->trigger
)
2720 static int cgroup_add_file(struct cgroup
*cgrp
, struct cftype
*cft
)
2722 struct dentry
*dir
= cgrp
->dentry
;
2723 struct cgroup
*parent
= __d_cgrp(dir
);
2724 struct dentry
*dentry
;
2728 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2730 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
2731 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
2732 strcpy(name
, cft
->ss
->name
);
2735 strcat(name
, cft
->name
);
2737 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2739 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2743 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2744 if (IS_ERR(dentry
)) {
2745 error
= PTR_ERR(dentry
);
2749 cfe
->type
= (void *)cft
;
2750 cfe
->dentry
= dentry
;
2751 dentry
->d_fsdata
= cfe
;
2752 simple_xattrs_init(&cfe
->xattrs
);
2754 mode
= cgroup_file_mode(cft
);
2755 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2757 list_add_tail(&cfe
->node
, &parent
->files
);
2767 * cgroup_addrm_files - add or remove files to a cgroup directory
2768 * @cgrp: the target cgroup
2769 * @cfts: array of cftypes to be added
2770 * @is_add: whether to add or remove
2772 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2773 * For removals, this function never fails. If addition fails, this
2774 * function doesn't remove files already added. The caller is responsible
2777 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
2783 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
2784 lockdep_assert_held(&cgroup_mutex
);
2786 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2787 /* does cft->flags tell us to skip this file on @cgrp? */
2788 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_sane_behavior(cgrp
))
2790 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2792 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2796 ret
= cgroup_add_file(cgrp
, cft
);
2798 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2803 cgroup_rm_file(cgrp
, cft
);
2809 static void cgroup_cfts_prepare(void)
2810 __acquires(&cgroup_mutex
)
2813 * Thanks to the entanglement with vfs inode locking, we can't walk
2814 * the existing cgroups under cgroup_mutex and create files.
2815 * Instead, we use css_for_each_descendant_pre() and drop RCU read
2816 * lock before calling cgroup_addrm_files().
2818 mutex_lock(&cgroup_mutex
);
2821 static int cgroup_cfts_commit(struct cftype
*cfts
, bool is_add
)
2822 __releases(&cgroup_mutex
)
2825 struct cgroup_subsys
*ss
= cfts
[0].ss
;
2826 struct cgroup
*root
= &ss
->root
->top_cgroup
;
2827 struct super_block
*sb
= ss
->root
->sb
;
2828 struct dentry
*prev
= NULL
;
2829 struct inode
*inode
;
2830 struct cgroup_subsys_state
*css
;
2834 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2835 if (!cfts
|| ss
->root
== &cgroup_dummy_root
||
2836 !atomic_inc_not_zero(&sb
->s_active
)) {
2837 mutex_unlock(&cgroup_mutex
);
2842 * All cgroups which are created after we drop cgroup_mutex will
2843 * have the updated set of files, so we only need to update the
2844 * cgroups created before the current @cgroup_serial_nr_next.
2846 update_before
= cgroup_serial_nr_next
;
2848 mutex_unlock(&cgroup_mutex
);
2850 /* add/rm files for all cgroups created before */
2852 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
2853 struct cgroup
*cgrp
= css
->cgroup
;
2855 if (cgroup_is_dead(cgrp
))
2858 inode
= cgrp
->dentry
->d_inode
;
2863 prev
= cgrp
->dentry
;
2865 mutex_lock(&inode
->i_mutex
);
2866 mutex_lock(&cgroup_mutex
);
2867 if (cgrp
->serial_nr
< update_before
&& !cgroup_is_dead(cgrp
))
2868 ret
= cgroup_addrm_files(cgrp
, cfts
, is_add
);
2869 mutex_unlock(&cgroup_mutex
);
2870 mutex_unlock(&inode
->i_mutex
);
2878 deactivate_super(sb
);
2883 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2884 * @ss: target cgroup subsystem
2885 * @cfts: zero-length name terminated array of cftypes
2887 * Register @cfts to @ss. Files described by @cfts are created for all
2888 * existing cgroups to which @ss is attached and all future cgroups will
2889 * have them too. This function can be called anytime whether @ss is
2892 * Returns 0 on successful registration, -errno on failure. Note that this
2893 * function currently returns 0 as long as @cfts registration is successful
2894 * even if some file creation attempts on existing cgroups fail.
2896 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2898 struct cftype_set
*set
;
2902 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2906 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++)
2909 cgroup_cfts_prepare();
2911 list_add_tail(&set
->node
, &ss
->cftsets
);
2912 ret
= cgroup_cfts_commit(cfts
, true);
2914 cgroup_rm_cftypes(cfts
);
2917 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2920 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2921 * @cfts: zero-length name terminated array of cftypes
2923 * Unregister @cfts. Files described by @cfts are removed from all
2924 * existing cgroups and all future cgroups won't have them either. This
2925 * function can be called anytime whether @cfts' subsys is attached or not.
2927 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2930 int cgroup_rm_cftypes(struct cftype
*cfts
)
2932 struct cftype_set
*set
;
2934 if (!cfts
|| !cfts
[0].ss
)
2937 cgroup_cfts_prepare();
2939 list_for_each_entry(set
, &cfts
[0].ss
->cftsets
, node
) {
2940 if (set
->cfts
== cfts
) {
2941 list_del(&set
->node
);
2943 cgroup_cfts_commit(cfts
, false);
2948 cgroup_cfts_commit(NULL
, false);
2953 * cgroup_task_count - count the number of tasks in a cgroup.
2954 * @cgrp: the cgroup in question
2956 * Return the number of tasks in the cgroup.
2958 int cgroup_task_count(const struct cgroup
*cgrp
)
2961 struct cgrp_cset_link
*link
;
2963 read_lock(&css_set_lock
);
2964 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
2965 count
+= atomic_read(&link
->cset
->refcount
);
2966 read_unlock(&css_set_lock
);
2971 * To reduce the fork() overhead for systems that are not actually using
2972 * their cgroups capability, we don't maintain the lists running through
2973 * each css_set to its tasks until we see the list actually used - in other
2974 * words after the first call to css_task_iter_start().
2976 static void cgroup_enable_task_cg_lists(void)
2978 struct task_struct
*p
, *g
;
2979 write_lock(&css_set_lock
);
2980 use_task_css_set_links
= 1;
2982 * We need tasklist_lock because RCU is not safe against
2983 * while_each_thread(). Besides, a forking task that has passed
2984 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2985 * is not guaranteed to have its child immediately visible in the
2986 * tasklist if we walk through it with RCU.
2988 read_lock(&tasklist_lock
);
2989 do_each_thread(g
, p
) {
2992 * We should check if the process is exiting, otherwise
2993 * it will race with cgroup_exit() in that the list
2994 * entry won't be deleted though the process has exited.
2996 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2997 list_add(&p
->cg_list
, &task_css_set(p
)->tasks
);
2999 } while_each_thread(g
, p
);
3000 read_unlock(&tasklist_lock
);
3001 write_unlock(&css_set_lock
);
3005 * css_next_child - find the next child of a given css
3006 * @pos_css: the current position (%NULL to initiate traversal)
3007 * @parent_css: css whose children to walk
3009 * This function returns the next child of @parent_css and should be called
3010 * under RCU read lock. The only requirement is that @parent_css and
3011 * @pos_css are accessible. The next sibling is guaranteed to be returned
3012 * regardless of their states.
3014 struct cgroup_subsys_state
*
3015 css_next_child(struct cgroup_subsys_state
*pos_css
,
3016 struct cgroup_subsys_state
*parent_css
)
3018 struct cgroup
*pos
= pos_css
? pos_css
->cgroup
: NULL
;
3019 struct cgroup
*cgrp
= parent_css
->cgroup
;
3020 struct cgroup
*next
;
3022 WARN_ON_ONCE(!rcu_read_lock_held());
3025 * @pos could already have been removed. Once a cgroup is removed,
3026 * its ->sibling.next is no longer updated when its next sibling
3027 * changes. As CGRP_DEAD assertion is serialized and happens
3028 * before the cgroup is taken off the ->sibling list, if we see it
3029 * unasserted, it's guaranteed that the next sibling hasn't
3030 * finished its grace period even if it's already removed, and thus
3031 * safe to dereference from this RCU critical section. If
3032 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
3033 * to be visible as %true here.
3035 * If @pos is dead, its next pointer can't be dereferenced;
3036 * however, as each cgroup is given a monotonically increasing
3037 * unique serial number and always appended to the sibling list,
3038 * the next one can be found by walking the parent's children until
3039 * we see a cgroup with higher serial number than @pos's. While
3040 * this path can be slower, it's taken only when either the current
3041 * cgroup is removed or iteration and removal race.
3044 next
= list_entry_rcu(cgrp
->children
.next
, struct cgroup
, sibling
);
3045 } else if (likely(!cgroup_is_dead(pos
))) {
3046 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
, sibling
);
3048 list_for_each_entry_rcu(next
, &cgrp
->children
, sibling
)
3049 if (next
->serial_nr
> pos
->serial_nr
)
3053 if (&next
->sibling
== &cgrp
->children
)
3056 return cgroup_css(next
, parent_css
->ss
);
3058 EXPORT_SYMBOL_GPL(css_next_child
);
3061 * css_next_descendant_pre - find the next descendant for pre-order walk
3062 * @pos: the current position (%NULL to initiate traversal)
3063 * @root: css whose descendants to walk
3065 * To be used by css_for_each_descendant_pre(). Find the next descendant
3066 * to visit for pre-order traversal of @root's descendants. @root is
3067 * included in the iteration and the first node to be visited.
3069 * While this function requires RCU read locking, it doesn't require the
3070 * whole traversal to be contained in a single RCU critical section. This
3071 * function will return the correct next descendant as long as both @pos
3072 * and @root are accessible and @pos is a descendant of @root.
3074 struct cgroup_subsys_state
*
3075 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3076 struct cgroup_subsys_state
*root
)
3078 struct cgroup_subsys_state
*next
;
3080 WARN_ON_ONCE(!rcu_read_lock_held());
3082 /* if first iteration, visit @root */
3086 /* visit the first child if exists */
3087 next
= css_next_child(NULL
, pos
);
3091 /* no child, visit my or the closest ancestor's next sibling */
3092 while (pos
!= root
) {
3093 next
= css_next_child(pos
, css_parent(pos
));
3096 pos
= css_parent(pos
);
3101 EXPORT_SYMBOL_GPL(css_next_descendant_pre
);
3104 * css_rightmost_descendant - return the rightmost descendant of a css
3105 * @pos: css of interest
3107 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3108 * is returned. This can be used during pre-order traversal to skip
3111 * While this function requires RCU read locking, it doesn't require the
3112 * whole traversal to be contained in a single RCU critical section. This
3113 * function will return the correct rightmost descendant as long as @pos is
3116 struct cgroup_subsys_state
*
3117 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3119 struct cgroup_subsys_state
*last
, *tmp
;
3121 WARN_ON_ONCE(!rcu_read_lock_held());
3125 /* ->prev isn't RCU safe, walk ->next till the end */
3127 css_for_each_child(tmp
, last
)
3133 EXPORT_SYMBOL_GPL(css_rightmost_descendant
);
3135 static struct cgroup_subsys_state
*
3136 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3138 struct cgroup_subsys_state
*last
;
3142 pos
= css_next_child(NULL
, pos
);
3149 * css_next_descendant_post - find the next descendant for post-order walk
3150 * @pos: the current position (%NULL to initiate traversal)
3151 * @root: css whose descendants to walk
3153 * To be used by css_for_each_descendant_post(). Find the next descendant
3154 * to visit for post-order traversal of @root's descendants. @root is
3155 * included in the iteration and the last node to be visited.
3157 * While this function requires RCU read locking, it doesn't require the
3158 * whole traversal to be contained in a single RCU critical section. This
3159 * function will return the correct next descendant as long as both @pos
3160 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3162 struct cgroup_subsys_state
*
3163 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3164 struct cgroup_subsys_state
*root
)
3166 struct cgroup_subsys_state
*next
;
3168 WARN_ON_ONCE(!rcu_read_lock_held());
3170 /* if first iteration, visit leftmost descendant which may be @root */
3172 return css_leftmost_descendant(root
);
3174 /* if we visited @root, we're done */
3178 /* if there's an unvisited sibling, visit its leftmost descendant */
3179 next
= css_next_child(pos
, css_parent(pos
));
3181 return css_leftmost_descendant(next
);
3183 /* no sibling left, visit parent */
3184 return css_parent(pos
);
3186 EXPORT_SYMBOL_GPL(css_next_descendant_post
);
3189 * css_advance_task_iter - advance a task itererator to the next css_set
3190 * @it: the iterator to advance
3192 * Advance @it to the next css_set to walk.
3194 static void css_advance_task_iter(struct css_task_iter
*it
)
3196 struct list_head
*l
= it
->cset_link
;
3197 struct cgrp_cset_link
*link
;
3198 struct css_set
*cset
;
3200 /* Advance to the next non-empty css_set */
3203 if (l
== &it
->origin_css
->cgroup
->cset_links
) {
3204 it
->cset_link
= NULL
;
3207 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3209 } while (list_empty(&cset
->tasks
));
3211 it
->task
= cset
->tasks
.next
;
3215 * css_task_iter_start - initiate task iteration
3216 * @css: the css to walk tasks of
3217 * @it: the task iterator to use
3219 * Initiate iteration through the tasks of @css. The caller can call
3220 * css_task_iter_next() to walk through the tasks until the function
3221 * returns NULL. On completion of iteration, css_task_iter_end() must be
3224 * Note that this function acquires a lock which is released when the
3225 * iteration finishes. The caller can't sleep while iteration is in
3228 void css_task_iter_start(struct cgroup_subsys_state
*css
,
3229 struct css_task_iter
*it
)
3230 __acquires(css_set_lock
)
3233 * The first time anyone tries to iterate across a css, we need to
3234 * enable the list linking each css_set to its tasks, and fix up
3235 * all existing tasks.
3237 if (!use_task_css_set_links
)
3238 cgroup_enable_task_cg_lists();
3240 read_lock(&css_set_lock
);
3242 it
->origin_css
= css
;
3243 it
->cset_link
= &css
->cgroup
->cset_links
;
3245 css_advance_task_iter(it
);
3249 * css_task_iter_next - return the next task for the iterator
3250 * @it: the task iterator being iterated
3252 * The "next" function for task iteration. @it should have been
3253 * initialized via css_task_iter_start(). Returns NULL when the iteration
3256 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
3258 struct task_struct
*res
;
3259 struct list_head
*l
= it
->task
;
3260 struct cgrp_cset_link
*link
;
3262 /* If the iterator cg is NULL, we have no tasks */
3265 res
= list_entry(l
, struct task_struct
, cg_list
);
3266 /* Advance iterator to find next entry */
3268 link
= list_entry(it
->cset_link
, struct cgrp_cset_link
, cset_link
);
3269 if (l
== &link
->cset
->tasks
) {
3271 * We reached the end of this task list - move on to the
3272 * next cgrp_cset_link.
3274 css_advance_task_iter(it
);
3282 * css_task_iter_end - finish task iteration
3283 * @it: the task iterator to finish
3285 * Finish task iteration started by css_task_iter_start().
3287 void css_task_iter_end(struct css_task_iter
*it
)
3288 __releases(css_set_lock
)
3290 read_unlock(&css_set_lock
);
3293 static inline int started_after_time(struct task_struct
*t1
,
3294 struct timespec
*time
,
3295 struct task_struct
*t2
)
3297 int start_diff
= timespec_compare(&t1
->start_time
, time
);
3298 if (start_diff
> 0) {
3300 } else if (start_diff
< 0) {
3304 * Arbitrarily, if two processes started at the same
3305 * time, we'll say that the lower pointer value
3306 * started first. Note that t2 may have exited by now
3307 * so this may not be a valid pointer any longer, but
3308 * that's fine - it still serves to distinguish
3309 * between two tasks started (effectively) simultaneously.
3316 * This function is a callback from heap_insert() and is used to order
3318 * In this case we order the heap in descending task start time.
3320 static inline int started_after(void *p1
, void *p2
)
3322 struct task_struct
*t1
= p1
;
3323 struct task_struct
*t2
= p2
;
3324 return started_after_time(t1
, &t2
->start_time
, t2
);
3328 * css_scan_tasks - iterate though all the tasks in a css
3329 * @css: the css to iterate tasks of
3330 * @test: optional test callback
3331 * @process: process callback
3332 * @data: data passed to @test and @process
3333 * @heap: optional pre-allocated heap used for task iteration
3335 * Iterate through all the tasks in @css, calling @test for each, and if it
3336 * returns %true, call @process for it also.
3338 * @test may be NULL, meaning always true (select all tasks), which
3339 * effectively duplicates css_task_iter_{start,next,end}() but does not
3340 * lock css_set_lock for the call to @process.
3342 * It is guaranteed that @process will act on every task that is a member
3343 * of @css for the duration of this call. This function may or may not
3344 * call @process for tasks that exit or move to a different css during the
3345 * call, or are forked or move into the css during the call.
3347 * Note that @test may be called with locks held, and may in some
3348 * situations be called multiple times for the same task, so it should be
3351 * If @heap is non-NULL, a heap has been pre-allocated and will be used for
3352 * heap operations (and its "gt" member will be overwritten), else a
3353 * temporary heap will be used (allocation of which may cause this function
3356 int css_scan_tasks(struct cgroup_subsys_state
*css
,
3357 bool (*test
)(struct task_struct
*, void *),
3358 void (*process
)(struct task_struct
*, void *),
3359 void *data
, struct ptr_heap
*heap
)
3362 struct css_task_iter it
;
3363 struct task_struct
*p
, *dropped
;
3364 /* Never dereference latest_task, since it's not refcounted */
3365 struct task_struct
*latest_task
= NULL
;
3366 struct ptr_heap tmp_heap
;
3367 struct timespec latest_time
= { 0, 0 };
3370 /* The caller supplied our heap and pre-allocated its memory */
3371 heap
->gt
= &started_after
;
3373 /* We need to allocate our own heap memory */
3375 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3377 /* cannot allocate the heap */
3383 * Scan tasks in the css, using the @test callback to determine
3384 * which are of interest, and invoking @process callback on the
3385 * ones which need an update. Since we don't want to hold any
3386 * locks during the task updates, gather tasks to be processed in a
3387 * heap structure. The heap is sorted by descending task start
3388 * time. If the statically-sized heap fills up, we overflow tasks
3389 * that started later, and in future iterations only consider tasks
3390 * that started after the latest task in the previous pass. This
3391 * guarantees forward progress and that we don't miss any tasks.
3394 css_task_iter_start(css
, &it
);
3395 while ((p
= css_task_iter_next(&it
))) {
3397 * Only affect tasks that qualify per the caller's callback,
3398 * if he provided one
3400 if (test
&& !test(p
, data
))
3403 * Only process tasks that started after the last task
3406 if (!started_after_time(p
, &latest_time
, latest_task
))
3408 dropped
= heap_insert(heap
, p
);
3409 if (dropped
== NULL
) {
3411 * The new task was inserted; the heap wasn't
3415 } else if (dropped
!= p
) {
3417 * The new task was inserted, and pushed out a
3421 put_task_struct(dropped
);
3424 * Else the new task was newer than anything already in
3425 * the heap and wasn't inserted
3428 css_task_iter_end(&it
);
3431 for (i
= 0; i
< heap
->size
; i
++) {
3432 struct task_struct
*q
= heap
->ptrs
[i
];
3434 latest_time
= q
->start_time
;
3437 /* Process the task per the caller's callback */
3442 * If we had to process any tasks at all, scan again
3443 * in case some of them were in the middle of forking
3444 * children that didn't get processed.
3445 * Not the most efficient way to do it, but it avoids
3446 * having to take callback_mutex in the fork path
3450 if (heap
== &tmp_heap
)
3451 heap_free(&tmp_heap
);
3455 static void cgroup_transfer_one_task(struct task_struct
*task
, void *data
)
3457 struct cgroup
*new_cgroup
= data
;
3459 mutex_lock(&cgroup_mutex
);
3460 cgroup_attach_task(new_cgroup
, task
, false);
3461 mutex_unlock(&cgroup_mutex
);
3465 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3466 * @to: cgroup to which the tasks will be moved
3467 * @from: cgroup in which the tasks currently reside
3469 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3471 return css_scan_tasks(&from
->dummy_css
, NULL
, cgroup_transfer_one_task
,
3476 * Stuff for reading the 'tasks'/'procs' files.
3478 * Reading this file can return large amounts of data if a cgroup has
3479 * *lots* of attached tasks. So it may need several calls to read(),
3480 * but we cannot guarantee that the information we produce is correct
3481 * unless we produce it entirely atomically.
3485 /* which pidlist file are we talking about? */
3486 enum cgroup_filetype
{
3492 * A pidlist is a list of pids that virtually represents the contents of one
3493 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3494 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3497 struct cgroup_pidlist
{
3499 * used to find which pidlist is wanted. doesn't change as long as
3500 * this particular list stays in the list.
3502 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3505 /* how many elements the above list has */
3507 /* how many files are using the current array */
3509 /* each of these stored in a list by its cgroup */
3510 struct list_head links
;
3511 /* pointer to the cgroup we belong to, for list removal purposes */
3512 struct cgroup
*owner
;
3513 /* protects the other fields */
3514 struct rw_semaphore rwsem
;
3518 * The following two functions "fix" the issue where there are more pids
3519 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3520 * TODO: replace with a kernel-wide solution to this problem
3522 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3523 static void *pidlist_allocate(int count
)
3525 if (PIDLIST_TOO_LARGE(count
))
3526 return vmalloc(count
* sizeof(pid_t
));
3528 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3530 static void pidlist_free(void *p
)
3532 if (is_vmalloc_addr(p
))
3539 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3540 * Returns the number of unique elements.
3542 static int pidlist_uniq(pid_t
*list
, int length
)
3547 * we presume the 0th element is unique, so i starts at 1. trivial
3548 * edge cases first; no work needs to be done for either
3550 if (length
== 0 || length
== 1)
3552 /* src and dest walk down the list; dest counts unique elements */
3553 for (src
= 1; src
< length
; src
++) {
3554 /* find next unique element */
3555 while (list
[src
] == list
[src
-1]) {
3560 /* dest always points to where the next unique element goes */
3561 list
[dest
] = list
[src
];
3568 static int cmppid(const void *a
, const void *b
)
3570 return *(pid_t
*)a
- *(pid_t
*)b
;
3574 * find the appropriate pidlist for our purpose (given procs vs tasks)
3575 * returns with the lock on that pidlist already held, and takes care
3576 * of the use count, or returns NULL with no locks held if we're out of
3579 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3580 enum cgroup_filetype type
)
3582 struct cgroup_pidlist
*l
;
3583 /* don't need task_nsproxy() if we're looking at ourself */
3584 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3587 * We can't drop the pidlist_mutex before taking the l->rwsem in case
3588 * the last ref-holder is trying to remove l from the list at the same
3589 * time. Holding the pidlist_mutex precludes somebody taking whichever
3590 * list we find out from under us - compare release_pid_array().
3592 mutex_lock(&cgrp
->pidlist_mutex
);
3593 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3594 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3595 /* make sure l doesn't vanish out from under us */
3596 down_write(&l
->rwsem
);
3597 mutex_unlock(&cgrp
->pidlist_mutex
);
3601 /* entry not found; create a new one */
3602 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3604 mutex_unlock(&cgrp
->pidlist_mutex
);
3607 init_rwsem(&l
->rwsem
);
3608 down_write(&l
->rwsem
);
3610 l
->key
.ns
= get_pid_ns(ns
);
3612 list_add(&l
->links
, &cgrp
->pidlists
);
3613 mutex_unlock(&cgrp
->pidlist_mutex
);
3618 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3620 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3621 struct cgroup_pidlist
**lp
)
3625 int pid
, n
= 0; /* used for populating the array */
3626 struct css_task_iter it
;
3627 struct task_struct
*tsk
;
3628 struct cgroup_pidlist
*l
;
3631 * If cgroup gets more users after we read count, we won't have
3632 * enough space - tough. This race is indistinguishable to the
3633 * caller from the case that the additional cgroup users didn't
3634 * show up until sometime later on.
3636 length
= cgroup_task_count(cgrp
);
3637 array
= pidlist_allocate(length
);
3640 /* now, populate the array */
3641 css_task_iter_start(&cgrp
->dummy_css
, &it
);
3642 while ((tsk
= css_task_iter_next(&it
))) {
3643 if (unlikely(n
== length
))
3645 /* get tgid or pid for procs or tasks file respectively */
3646 if (type
== CGROUP_FILE_PROCS
)
3647 pid
= task_tgid_vnr(tsk
);
3649 pid
= task_pid_vnr(tsk
);
3650 if (pid
> 0) /* make sure to only use valid results */
3653 css_task_iter_end(&it
);
3655 /* now sort & (if procs) strip out duplicates */
3656 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3657 if (type
== CGROUP_FILE_PROCS
)
3658 length
= pidlist_uniq(array
, length
);
3659 l
= cgroup_pidlist_find(cgrp
, type
);
3661 pidlist_free(array
);
3664 /* store array, freeing old if necessary - lock already held */
3665 pidlist_free(l
->list
);
3669 up_write(&l
->rwsem
);
3675 * cgroupstats_build - build and fill cgroupstats
3676 * @stats: cgroupstats to fill information into
3677 * @dentry: A dentry entry belonging to the cgroup for which stats have
3680 * Build and fill cgroupstats so that taskstats can export it to user
3683 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3686 struct cgroup
*cgrp
;
3687 struct css_task_iter it
;
3688 struct task_struct
*tsk
;
3691 * Validate dentry by checking the superblock operations,
3692 * and make sure it's a directory.
3694 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3695 !S_ISDIR(dentry
->d_inode
->i_mode
))
3699 cgrp
= dentry
->d_fsdata
;
3701 css_task_iter_start(&cgrp
->dummy_css
, &it
);
3702 while ((tsk
= css_task_iter_next(&it
))) {
3703 switch (tsk
->state
) {
3705 stats
->nr_running
++;
3707 case TASK_INTERRUPTIBLE
:
3708 stats
->nr_sleeping
++;
3710 case TASK_UNINTERRUPTIBLE
:
3711 stats
->nr_uninterruptible
++;
3714 stats
->nr_stopped
++;
3717 if (delayacct_is_task_waiting_on_io(tsk
))
3718 stats
->nr_io_wait
++;
3722 css_task_iter_end(&it
);
3730 * seq_file methods for the tasks/procs files. The seq_file position is the
3731 * next pid to display; the seq_file iterator is a pointer to the pid
3732 * in the cgroup->l->list array.
3735 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3738 * Initially we receive a position value that corresponds to
3739 * one more than the last pid shown (or 0 on the first call or
3740 * after a seek to the start). Use a binary-search to find the
3741 * next pid to display, if any
3743 struct cgroup_pidlist
*l
= s
->private;
3744 int index
= 0, pid
= *pos
;
3747 down_read(&l
->rwsem
);
3749 int end
= l
->length
;
3751 while (index
< end
) {
3752 int mid
= (index
+ end
) / 2;
3753 if (l
->list
[mid
] == pid
) {
3756 } else if (l
->list
[mid
] <= pid
)
3762 /* If we're off the end of the array, we're done */
3763 if (index
>= l
->length
)
3765 /* Update the abstract position to be the actual pid that we found */
3766 iter
= l
->list
+ index
;
3771 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3773 struct cgroup_pidlist
*l
= s
->private;
3777 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3779 struct cgroup_pidlist
*l
= s
->private;
3781 pid_t
*end
= l
->list
+ l
->length
;
3783 * Advance to the next pid in the array. If this goes off the
3795 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3797 return seq_printf(s
, "%d\n", *(int *)v
);
3801 * seq_operations functions for iterating on pidlists through seq_file -
3802 * independent of whether it's tasks or procs
3804 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3805 .start
= cgroup_pidlist_start
,
3806 .stop
= cgroup_pidlist_stop
,
3807 .next
= cgroup_pidlist_next
,
3808 .show
= cgroup_pidlist_show
,
3811 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3814 * the case where we're the last user of this particular pidlist will
3815 * have us remove it from the cgroup's list, which entails taking the
3816 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3817 * pidlist_mutex, we have to take pidlist_mutex first.
3819 mutex_lock(&l
->owner
->pidlist_mutex
);
3820 down_write(&l
->rwsem
);
3821 BUG_ON(!l
->use_count
);
3822 if (!--l
->use_count
) {
3823 /* we're the last user if refcount is 0; remove and free */
3824 list_del(&l
->links
);
3825 mutex_unlock(&l
->owner
->pidlist_mutex
);
3826 pidlist_free(l
->list
);
3827 put_pid_ns(l
->key
.ns
);
3828 up_write(&l
->rwsem
);
3832 mutex_unlock(&l
->owner
->pidlist_mutex
);
3833 up_write(&l
->rwsem
);
3836 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3838 struct cgroup_pidlist
*l
;
3839 if (!(file
->f_mode
& FMODE_READ
))
3842 * the seq_file will only be initialized if the file was opened for
3843 * reading; hence we check if it's not null only in that case.
3845 l
= ((struct seq_file
*)file
->private_data
)->private;
3846 cgroup_release_pid_array(l
);
3847 return seq_release(inode
, file
);
3850 static const struct file_operations cgroup_pidlist_operations
= {
3852 .llseek
= seq_lseek
,
3853 .write
= cgroup_file_write
,
3854 .release
= cgroup_pidlist_release
,
3858 * The following functions handle opens on a file that displays a pidlist
3859 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3862 /* helper function for the two below it */
3863 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3865 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3866 struct cgroup_pidlist
*l
;
3869 /* Nothing to do for write-only files */
3870 if (!(file
->f_mode
& FMODE_READ
))
3873 /* have the array populated */
3874 retval
= pidlist_array_load(cgrp
, type
, &l
);
3877 /* configure file information */
3878 file
->f_op
= &cgroup_pidlist_operations
;
3880 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3882 cgroup_release_pid_array(l
);
3885 ((struct seq_file
*)file
->private_data
)->private = l
;
3888 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3890 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3892 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3894 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3897 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
3900 return notify_on_release(css
->cgroup
);
3903 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
3904 struct cftype
*cft
, u64 val
)
3906 clear_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
3908 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
3910 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
3915 * When dput() is called asynchronously, if umount has been done and
3916 * then deactivate_super() in cgroup_free_fn() kills the superblock,
3917 * there's a small window that vfs will see the root dentry with non-zero
3918 * refcnt and trigger BUG().
3920 * That's why we hold a reference before dput() and drop it right after.
3922 static void cgroup_dput(struct cgroup
*cgrp
)
3924 struct super_block
*sb
= cgrp
->root
->sb
;
3926 atomic_inc(&sb
->s_active
);
3928 deactivate_super(sb
);
3932 * Unregister event and free resources.
3934 * Gets called from workqueue.
3936 static void cgroup_event_remove(struct work_struct
*work
)
3938 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3940 struct cgroup_subsys_state
*css
= event
->css
;
3942 remove_wait_queue(event
->wqh
, &event
->wait
);
3944 event
->cft
->unregister_event(css
, event
->cft
, event
->eventfd
);
3946 /* Notify userspace the event is going away. */
3947 eventfd_signal(event
->eventfd
, 1);
3949 eventfd_ctx_put(event
->eventfd
);
3955 * Gets called on POLLHUP on eventfd when user closes it.
3957 * Called with wqh->lock held and interrupts disabled.
3959 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3960 int sync
, void *key
)
3962 struct cgroup_event
*event
= container_of(wait
,
3963 struct cgroup_event
, wait
);
3964 struct cgroup
*cgrp
= event
->css
->cgroup
;
3965 unsigned long flags
= (unsigned long)key
;
3967 if (flags
& POLLHUP
) {
3969 * If the event has been detached at cgroup removal, we
3970 * can simply return knowing the other side will cleanup
3973 * We can't race against event freeing since the other
3974 * side will require wqh->lock via remove_wait_queue(),
3977 spin_lock(&cgrp
->event_list_lock
);
3978 if (!list_empty(&event
->list
)) {
3979 list_del_init(&event
->list
);
3981 * We are in atomic context, but cgroup_event_remove()
3982 * may sleep, so we have to call it in workqueue.
3984 schedule_work(&event
->remove
);
3986 spin_unlock(&cgrp
->event_list_lock
);
3992 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3993 wait_queue_head_t
*wqh
, poll_table
*pt
)
3995 struct cgroup_event
*event
= container_of(pt
,
3996 struct cgroup_event
, pt
);
3999 add_wait_queue(wqh
, &event
->wait
);
4003 * Parse input and register new cgroup event handler.
4005 * Input must be in format '<event_fd> <control_fd> <args>'.
4006 * Interpretation of args is defined by control file implementation.
4008 static int cgroup_write_event_control(struct cgroup_subsys_state
*dummy_css
,
4009 struct cftype
*cft
, const char *buffer
)
4011 struct cgroup
*cgrp
= dummy_css
->cgroup
;
4012 struct cgroup_event
*event
;
4013 struct cgroup_subsys_state
*cfile_css
;
4014 unsigned int efd
, cfd
;
4020 efd
= simple_strtoul(buffer
, &endp
, 10);
4025 cfd
= simple_strtoul(buffer
, &endp
, 10);
4026 if ((*endp
!= ' ') && (*endp
!= '\0'))
4030 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
4034 INIT_LIST_HEAD(&event
->list
);
4035 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
4036 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
4037 INIT_WORK(&event
->remove
, cgroup_event_remove
);
4045 event
->eventfd
= eventfd_ctx_fileget(efile
.file
);
4046 if (IS_ERR(event
->eventfd
)) {
4047 ret
= PTR_ERR(event
->eventfd
);
4054 goto out_put_eventfd
;
4057 /* the process need read permission on control file */
4058 /* AV: shouldn't we check that it's been opened for read instead? */
4059 ret
= inode_permission(file_inode(cfile
.file
), MAY_READ
);
4063 event
->cft
= __file_cft(cfile
.file
);
4064 if (IS_ERR(event
->cft
)) {
4065 ret
= PTR_ERR(event
->cft
);
4069 if (!event
->cft
->ss
) {
4075 * Determine the css of @cfile, verify it belongs to the same
4076 * cgroup as cgroup.event_control, and associate @event with it.
4077 * Remaining events are automatically removed on cgroup destruction
4078 * but the removal is asynchronous, so take an extra ref.
4083 event
->css
= cgroup_css(cgrp
, event
->cft
->ss
);
4084 cfile_css
= css_from_dir(cfile
.file
->f_dentry
->d_parent
, event
->cft
->ss
);
4085 if (event
->css
&& event
->css
== cfile_css
&& css_tryget(event
->css
))
4092 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
4097 ret
= event
->cft
->register_event(event
->css
, event
->cft
,
4098 event
->eventfd
, buffer
);
4102 efile
.file
->f_op
->poll(efile
.file
, &event
->pt
);
4104 spin_lock(&cgrp
->event_list_lock
);
4105 list_add(&event
->list
, &cgrp
->event_list
);
4106 spin_unlock(&cgrp
->event_list_lock
);
4114 css_put(event
->css
);
4118 eventfd_ctx_put(event
->eventfd
);
4127 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
4130 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4133 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
4134 struct cftype
*cft
, u64 val
)
4137 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4139 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4143 static struct cftype cgroup_base_files
[] = {
4145 .name
= "cgroup.procs",
4146 .open
= cgroup_procs_open
,
4147 .write_u64
= cgroup_procs_write
,
4148 .release
= cgroup_pidlist_release
,
4149 .mode
= S_IRUGO
| S_IWUSR
,
4152 .name
= "cgroup.event_control",
4153 .write_string
= cgroup_write_event_control
,
4157 .name
= "cgroup.clone_children",
4158 .flags
= CFTYPE_INSANE
,
4159 .read_u64
= cgroup_clone_children_read
,
4160 .write_u64
= cgroup_clone_children_write
,
4163 .name
= "cgroup.sane_behavior",
4164 .flags
= CFTYPE_ONLY_ON_ROOT
,
4165 .read_seq_string
= cgroup_sane_behavior_show
,
4169 * Historical crazy stuff. These don't have "cgroup." prefix and
4170 * don't exist if sane_behavior. If you're depending on these, be
4171 * prepared to be burned.
4175 .flags
= CFTYPE_INSANE
, /* use "procs" instead */
4176 .open
= cgroup_tasks_open
,
4177 .write_u64
= cgroup_tasks_write
,
4178 .release
= cgroup_pidlist_release
,
4179 .mode
= S_IRUGO
| S_IWUSR
,
4182 .name
= "notify_on_release",
4183 .flags
= CFTYPE_INSANE
,
4184 .read_u64
= cgroup_read_notify_on_release
,
4185 .write_u64
= cgroup_write_notify_on_release
,
4188 .name
= "release_agent",
4189 .flags
= CFTYPE_INSANE
| CFTYPE_ONLY_ON_ROOT
,
4190 .read_seq_string
= cgroup_release_agent_show
,
4191 .write_string
= cgroup_release_agent_write
,
4192 .max_write_len
= PATH_MAX
,
4198 * cgroup_populate_dir - create subsys files in a cgroup directory
4199 * @cgrp: target cgroup
4200 * @subsys_mask: mask of the subsystem ids whose files should be added
4202 * On failure, no file is added.
4204 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
4206 struct cgroup_subsys
*ss
;
4209 /* process cftsets of each subsystem */
4210 for_each_subsys(ss
, i
) {
4211 struct cftype_set
*set
;
4213 if (!test_bit(i
, &subsys_mask
))
4216 list_for_each_entry(set
, &ss
->cftsets
, node
) {
4217 ret
= cgroup_addrm_files(cgrp
, set
->cfts
, true);
4224 cgroup_clear_dir(cgrp
, subsys_mask
);
4229 * css destruction is four-stage process.
4231 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4232 * Implemented in kill_css().
4234 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4235 * and thus css_tryget() is guaranteed to fail, the css can be offlined
4236 * by invoking offline_css(). After offlining, the base ref is put.
4237 * Implemented in css_killed_work_fn().
4239 * 3. When the percpu_ref reaches zero, the only possible remaining
4240 * accessors are inside RCU read sections. css_release() schedules the
4243 * 4. After the grace period, the css can be freed. Implemented in
4244 * css_free_work_fn().
4246 * It is actually hairier because both step 2 and 4 require process context
4247 * and thus involve punting to css->destroy_work adding two additional
4248 * steps to the already complex sequence.
4250 static void css_free_work_fn(struct work_struct
*work
)
4252 struct cgroup_subsys_state
*css
=
4253 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4254 struct cgroup
*cgrp
= css
->cgroup
;
4257 css_put(css
->parent
);
4259 css
->ss
->css_free(css
);
4263 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
4265 struct cgroup_subsys_state
*css
=
4266 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
4269 * css holds an extra ref to @cgrp->dentry which is put on the last
4270 * css_put(). dput() requires process context which we don't have.
4272 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
4273 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4276 static void css_release(struct percpu_ref
*ref
)
4278 struct cgroup_subsys_state
*css
=
4279 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4281 rcu_assign_pointer(css
->cgroup
->subsys
[css
->ss
->subsys_id
], NULL
);
4282 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4285 static void init_css(struct cgroup_subsys_state
*css
, struct cgroup_subsys
*ss
,
4286 struct cgroup
*cgrp
)
4293 css
->parent
= cgroup_css(cgrp
->parent
, ss
);
4295 css
->flags
|= CSS_ROOT
;
4297 BUG_ON(cgroup_css(cgrp
, ss
));
4300 /* invoke ->css_online() on a new CSS and mark it online if successful */
4301 static int online_css(struct cgroup_subsys_state
*css
)
4303 struct cgroup_subsys
*ss
= css
->ss
;
4306 lockdep_assert_held(&cgroup_mutex
);
4309 ret
= ss
->css_online(css
);
4311 css
->flags
|= CSS_ONLINE
;
4312 css
->cgroup
->nr_css
++;
4313 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->subsys_id
], css
);
4318 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4319 static void offline_css(struct cgroup_subsys_state
*css
)
4321 struct cgroup_subsys
*ss
= css
->ss
;
4323 lockdep_assert_held(&cgroup_mutex
);
4325 if (!(css
->flags
& CSS_ONLINE
))
4328 if (ss
->css_offline
)
4329 ss
->css_offline(css
);
4331 css
->flags
&= ~CSS_ONLINE
;
4332 css
->cgroup
->nr_css
--;
4333 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->subsys_id
], css
);
4337 * cgroup_create - create a cgroup
4338 * @parent: cgroup that will be parent of the new cgroup
4339 * @dentry: dentry of the new cgroup
4340 * @mode: mode to set on new inode
4342 * Must be called with the mutex on the parent inode held
4344 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
4347 struct cgroup_subsys_state
*css_ar
[CGROUP_SUBSYS_COUNT
] = { };
4348 struct cgroup
*cgrp
;
4349 struct cgroup_name
*name
;
4350 struct cgroupfs_root
*root
= parent
->root
;
4352 struct cgroup_subsys
*ss
;
4353 struct super_block
*sb
= root
->sb
;
4355 /* allocate the cgroup and its ID, 0 is reserved for the root */
4356 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4360 name
= cgroup_alloc_name(dentry
);
4363 rcu_assign_pointer(cgrp
->name
, name
);
4366 * Temporarily set the pointer to NULL, so idr_find() won't return
4367 * a half-baked cgroup.
4369 cgrp
->id
= idr_alloc(&root
->cgroup_idr
, NULL
, 1, 0, GFP_KERNEL
);
4374 * Only live parents can have children. Note that the liveliness
4375 * check isn't strictly necessary because cgroup_mkdir() and
4376 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4377 * anyway so that locking is contained inside cgroup proper and we
4378 * don't get nasty surprises if we ever grow another caller.
4380 if (!cgroup_lock_live_group(parent
)) {
4385 /* Grab a reference on the superblock so the hierarchy doesn't
4386 * get deleted on unmount if there are child cgroups. This
4387 * can be done outside cgroup_mutex, since the sb can't
4388 * disappear while someone has an open control file on the
4390 atomic_inc(&sb
->s_active
);
4392 init_cgroup_housekeeping(cgrp
);
4394 dentry
->d_fsdata
= cgrp
;
4395 cgrp
->dentry
= dentry
;
4397 cgrp
->parent
= parent
;
4398 cgrp
->dummy_css
.parent
= &parent
->dummy_css
;
4399 cgrp
->root
= parent
->root
;
4401 if (notify_on_release(parent
))
4402 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4404 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4405 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4407 for_each_root_subsys(root
, ss
) {
4408 struct cgroup_subsys_state
*css
;
4410 css
= ss
->css_alloc(cgroup_css(parent
, ss
));
4415 css_ar
[ss
->subsys_id
] = css
;
4417 err
= percpu_ref_init(&css
->refcnt
, css_release
);
4421 init_css(css
, ss
, cgrp
);
4425 * Create directory. cgroup_create_file() returns with the new
4426 * directory locked on success so that it can be populated without
4427 * dropping cgroup_mutex.
4429 err
= cgroup_create_file(dentry
, S_IFDIR
| mode
, sb
);
4432 lockdep_assert_held(&dentry
->d_inode
->i_mutex
);
4434 cgrp
->serial_nr
= cgroup_serial_nr_next
++;
4436 /* allocation complete, commit to creation */
4437 list_add_tail_rcu(&cgrp
->sibling
, &cgrp
->parent
->children
);
4438 root
->number_of_cgroups
++;
4440 /* hold a ref to the parent's dentry */
4441 dget(parent
->dentry
);
4443 /* creation succeeded, notify subsystems */
4444 for_each_root_subsys(root
, ss
) {
4445 struct cgroup_subsys_state
*css
= css_ar
[ss
->subsys_id
];
4447 err
= online_css(css
);
4451 /* each css holds a ref to the cgroup's dentry and parent css */
4453 css_get(css
->parent
);
4455 /* mark it consumed for error path */
4456 css_ar
[ss
->subsys_id
] = NULL
;
4458 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4460 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",
4461 current
->comm
, current
->pid
, ss
->name
);
4462 if (!strcmp(ss
->name
, "memory"))
4463 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4464 ss
->warned_broken_hierarchy
= true;
4468 idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4470 err
= cgroup_addrm_files(cgrp
, cgroup_base_files
, true);
4474 err
= cgroup_populate_dir(cgrp
, root
->subsys_mask
);
4478 mutex_unlock(&cgroup_mutex
);
4479 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4484 for_each_root_subsys(root
, ss
) {
4485 struct cgroup_subsys_state
*css
= css_ar
[ss
->subsys_id
];
4488 percpu_ref_cancel_init(&css
->refcnt
);
4492 mutex_unlock(&cgroup_mutex
);
4493 /* Release the reference count that we took on the superblock */
4494 deactivate_super(sb
);
4496 idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4498 kfree(rcu_dereference_raw(cgrp
->name
));
4504 for_each_root_subsys(root
, ss
) {
4505 struct cgroup_subsys_state
*css
= css_ar
[ss
->subsys_id
];
4508 percpu_ref_cancel_init(&css
->refcnt
);
4512 cgroup_destroy_locked(cgrp
);
4513 mutex_unlock(&cgroup_mutex
);
4514 mutex_unlock(&dentry
->d_inode
->i_mutex
);
4518 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4520 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4522 /* the vfs holds inode->i_mutex already */
4523 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4527 * This is called when the refcnt of a css is confirmed to be killed.
4528 * css_tryget() is now guaranteed to fail.
4530 static void css_killed_work_fn(struct work_struct
*work
)
4532 struct cgroup_subsys_state
*css
=
4533 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4534 struct cgroup
*cgrp
= css
->cgroup
;
4536 mutex_lock(&cgroup_mutex
);
4539 * css_tryget() is guaranteed to fail now. Tell subsystems to
4540 * initate destruction.
4545 * If @cgrp is marked dead, it's waiting for refs of all css's to
4546 * be disabled before proceeding to the second phase of cgroup
4547 * destruction. If we are the last one, kick it off.
4549 if (!cgrp
->nr_css
&& cgroup_is_dead(cgrp
))
4550 cgroup_destroy_css_killed(cgrp
);
4552 mutex_unlock(&cgroup_mutex
);
4555 * Put the css refs from kill_css(). Each css holds an extra
4556 * reference to the cgroup's dentry and cgroup removal proceeds
4557 * regardless of css refs. On the last put of each css, whenever
4558 * that may be, the extra dentry ref is put so that dentry
4559 * destruction happens only after all css's are released.
4564 /* css kill confirmation processing requires process context, bounce */
4565 static void css_killed_ref_fn(struct percpu_ref
*ref
)
4567 struct cgroup_subsys_state
*css
=
4568 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4570 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
4571 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4575 * kill_css - destroy a css
4576 * @css: css to destroy
4578 * This function initiates destruction of @css by removing cgroup interface
4579 * files and putting its base reference. ->css_offline() will be invoked
4580 * asynchronously once css_tryget() is guaranteed to fail and when the
4581 * reference count reaches zero, @css will be released.
4583 static void kill_css(struct cgroup_subsys_state
*css
)
4585 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->subsys_id
);
4588 * Killing would put the base ref, but we need to keep it alive
4589 * until after ->css_offline().
4594 * cgroup core guarantees that, by the time ->css_offline() is
4595 * invoked, no new css reference will be given out via
4596 * css_tryget(). We can't simply call percpu_ref_kill() and
4597 * proceed to offlining css's because percpu_ref_kill() doesn't
4598 * guarantee that the ref is seen as killed on all CPUs on return.
4600 * Use percpu_ref_kill_and_confirm() to get notifications as each
4601 * css is confirmed to be seen as killed on all CPUs.
4603 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
4607 * cgroup_destroy_locked - the first stage of cgroup destruction
4608 * @cgrp: cgroup to be destroyed
4610 * css's make use of percpu refcnts whose killing latency shouldn't be
4611 * exposed to userland and are RCU protected. Also, cgroup core needs to
4612 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
4613 * invoked. To satisfy all the requirements, destruction is implemented in
4614 * the following two steps.
4616 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4617 * userland visible parts and start killing the percpu refcnts of
4618 * css's. Set up so that the next stage will be kicked off once all
4619 * the percpu refcnts are confirmed to be killed.
4621 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4622 * rest of destruction. Once all cgroup references are gone, the
4623 * cgroup is RCU-freed.
4625 * This function implements s1. After this step, @cgrp is gone as far as
4626 * the userland is concerned and a new cgroup with the same name may be
4627 * created. As cgroup doesn't care about the names internally, this
4628 * doesn't cause any problem.
4630 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4631 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4633 struct dentry
*d
= cgrp
->dentry
;
4634 struct cgroup_event
*event
, *tmp
;
4635 struct cgroup_subsys
*ss
;
4636 struct cgroup
*child
;
4639 lockdep_assert_held(&d
->d_inode
->i_mutex
);
4640 lockdep_assert_held(&cgroup_mutex
);
4643 * css_set_lock synchronizes access to ->cset_links and prevents
4644 * @cgrp from being removed while __put_css_set() is in progress.
4646 read_lock(&css_set_lock
);
4647 empty
= list_empty(&cgrp
->cset_links
);
4648 read_unlock(&css_set_lock
);
4653 * Make sure there's no live children. We can't test ->children
4654 * emptiness as dead children linger on it while being destroyed;
4655 * otherwise, "rmdir parent/child parent" may fail with -EBUSY.
4659 list_for_each_entry_rcu(child
, &cgrp
->children
, sibling
) {
4660 empty
= cgroup_is_dead(child
);
4669 * Initiate massacre of all css's. cgroup_destroy_css_killed()
4670 * will be invoked to perform the rest of destruction once the
4671 * percpu refs of all css's are confirmed to be killed.
4673 for_each_root_subsys(cgrp
->root
, ss
) {
4674 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, ss
);
4681 * Mark @cgrp dead. This prevents further task migration and child
4682 * creation by disabling cgroup_lock_live_group(). Note that
4683 * CGRP_DEAD assertion is depended upon by css_next_child() to
4684 * resume iteration after dropping RCU read lock. See
4685 * css_next_child() for details.
4687 set_bit(CGRP_DEAD
, &cgrp
->flags
);
4689 /* CGRP_DEAD is set, remove from ->release_list for the last time */
4690 raw_spin_lock(&release_list_lock
);
4691 if (!list_empty(&cgrp
->release_list
))
4692 list_del_init(&cgrp
->release_list
);
4693 raw_spin_unlock(&release_list_lock
);
4696 * If @cgrp has css's attached, the second stage of cgroup
4697 * destruction is kicked off from css_killed_work_fn() after the
4698 * refs of all attached css's are killed. If @cgrp doesn't have
4699 * any css, we kick it off here.
4702 cgroup_destroy_css_killed(cgrp
);
4705 * Clear the base files and remove @cgrp directory. The removal
4706 * puts the base ref but we aren't quite done with @cgrp yet, so
4709 cgroup_addrm_files(cgrp
, cgroup_base_files
, false);
4711 cgroup_d_remove_dir(d
);
4714 * Unregister events and notify userspace.
4715 * Notify userspace about cgroup removing only after rmdir of cgroup
4716 * directory to avoid race between userspace and kernelspace.
4718 spin_lock(&cgrp
->event_list_lock
);
4719 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4720 list_del_init(&event
->list
);
4721 schedule_work(&event
->remove
);
4723 spin_unlock(&cgrp
->event_list_lock
);
4729 * cgroup_destroy_css_killed - the second step of cgroup destruction
4730 * @work: cgroup->destroy_free_work
4732 * This function is invoked from a work item for a cgroup which is being
4733 * destroyed after all css's are offlined and performs the rest of
4734 * destruction. This is the second step of destruction described in the
4735 * comment above cgroup_destroy_locked().
4737 static void cgroup_destroy_css_killed(struct cgroup
*cgrp
)
4739 struct cgroup
*parent
= cgrp
->parent
;
4740 struct dentry
*d
= cgrp
->dentry
;
4742 lockdep_assert_held(&cgroup_mutex
);
4744 /* delete this cgroup from parent->children */
4745 list_del_rcu(&cgrp
->sibling
);
4749 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4750 check_for_release(parent
);
4753 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4757 mutex_lock(&cgroup_mutex
);
4758 ret
= cgroup_destroy_locked(dentry
->d_fsdata
);
4759 mutex_unlock(&cgroup_mutex
);
4764 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4766 INIT_LIST_HEAD(&ss
->cftsets
);
4769 * base_cftset is embedded in subsys itself, no need to worry about
4772 if (ss
->base_cftypes
) {
4775 for (cft
= ss
->base_cftypes
; cft
->name
[0] != '\0'; cft
++)
4778 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4779 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4783 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4785 struct cgroup_subsys_state
*css
;
4787 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4789 mutex_lock(&cgroup_mutex
);
4791 /* init base cftset */
4792 cgroup_init_cftsets(ss
);
4794 /* Create the top cgroup state for this subsystem */
4795 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4796 ss
->root
= &cgroup_dummy_root
;
4797 css
= ss
->css_alloc(cgroup_css(cgroup_dummy_top
, ss
));
4798 /* We don't handle early failures gracefully */
4799 BUG_ON(IS_ERR(css
));
4800 init_css(css
, ss
, cgroup_dummy_top
);
4802 /* Update the init_css_set to contain a subsys
4803 * pointer to this state - since the subsystem is
4804 * newly registered, all tasks and hence the
4805 * init_css_set is in the subsystem's top cgroup. */
4806 init_css_set
.subsys
[ss
->subsys_id
] = css
;
4808 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4810 /* At system boot, before all subsystems have been
4811 * registered, no tasks have been forked, so we don't
4812 * need to invoke fork callbacks here. */
4813 BUG_ON(!list_empty(&init_task
.tasks
));
4815 BUG_ON(online_css(css
));
4817 mutex_unlock(&cgroup_mutex
);
4819 /* this function shouldn't be used with modular subsystems, since they
4820 * need to register a subsys_id, among other things */
4825 * cgroup_load_subsys: load and register a modular subsystem at runtime
4826 * @ss: the subsystem to load
4828 * This function should be called in a modular subsystem's initcall. If the
4829 * subsystem is built as a module, it will be assigned a new subsys_id and set
4830 * up for use. If the subsystem is built-in anyway, work is delegated to the
4831 * simpler cgroup_init_subsys.
4833 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4835 struct cgroup_subsys_state
*css
;
4837 struct hlist_node
*tmp
;
4838 struct css_set
*cset
;
4841 /* check name and function validity */
4842 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4843 ss
->css_alloc
== NULL
|| ss
->css_free
== NULL
)
4847 * we don't support callbacks in modular subsystems. this check is
4848 * before the ss->module check for consistency; a subsystem that could
4849 * be a module should still have no callbacks even if the user isn't
4850 * compiling it as one.
4852 if (ss
->fork
|| ss
->exit
)
4856 * an optionally modular subsystem is built-in: we want to do nothing,
4857 * since cgroup_init_subsys will have already taken care of it.
4859 if (ss
->module
== NULL
) {
4860 /* a sanity check */
4861 BUG_ON(cgroup_subsys
[ss
->subsys_id
] != ss
);
4865 /* init base cftset */
4866 cgroup_init_cftsets(ss
);
4868 mutex_lock(&cgroup_mutex
);
4869 cgroup_subsys
[ss
->subsys_id
] = ss
;
4872 * no ss->css_alloc seems to need anything important in the ss
4873 * struct, so this can happen first (i.e. before the dummy root
4876 css
= ss
->css_alloc(cgroup_css(cgroup_dummy_top
, ss
));
4878 /* failure case - need to deassign the cgroup_subsys[] slot. */
4879 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4880 mutex_unlock(&cgroup_mutex
);
4881 return PTR_ERR(css
);
4884 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4885 ss
->root
= &cgroup_dummy_root
;
4887 /* our new subsystem will be attached to the dummy hierarchy. */
4888 init_css(css
, ss
, cgroup_dummy_top
);
4891 * Now we need to entangle the css into the existing css_sets. unlike
4892 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4893 * will need a new pointer to it; done by iterating the css_set_table.
4894 * furthermore, modifying the existing css_sets will corrupt the hash
4895 * table state, so each changed css_set will need its hash recomputed.
4896 * this is all done under the css_set_lock.
4898 write_lock(&css_set_lock
);
4899 hash_for_each_safe(css_set_table
, i
, tmp
, cset
, hlist
) {
4900 /* skip entries that we already rehashed */
4901 if (cset
->subsys
[ss
->subsys_id
])
4903 /* remove existing entry */
4904 hash_del(&cset
->hlist
);
4906 cset
->subsys
[ss
->subsys_id
] = css
;
4907 /* recompute hash and restore entry */
4908 key
= css_set_hash(cset
->subsys
);
4909 hash_add(css_set_table
, &cset
->hlist
, key
);
4911 write_unlock(&css_set_lock
);
4913 ret
= online_css(css
);
4918 mutex_unlock(&cgroup_mutex
);
4922 mutex_unlock(&cgroup_mutex
);
4923 /* @ss can't be mounted here as try_module_get() would fail */
4924 cgroup_unload_subsys(ss
);
4927 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4930 * cgroup_unload_subsys: unload a modular subsystem
4931 * @ss: the subsystem to unload
4933 * This function should be called in a modular subsystem's exitcall. When this
4934 * function is invoked, the refcount on the subsystem's module will be 0, so
4935 * the subsystem will not be attached to any hierarchy.
4937 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4939 struct cgrp_cset_link
*link
;
4941 BUG_ON(ss
->module
== NULL
);
4944 * we shouldn't be called if the subsystem is in use, and the use of
4945 * try_module_get() in rebind_subsystems() should ensure that it
4946 * doesn't start being used while we're killing it off.
4948 BUG_ON(ss
->root
!= &cgroup_dummy_root
);
4950 mutex_lock(&cgroup_mutex
);
4952 offline_css(cgroup_css(cgroup_dummy_top
, ss
));
4954 /* deassign the subsys_id */
4955 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4957 /* remove subsystem from the dummy root's list of subsystems */
4958 list_del_init(&ss
->sibling
);
4961 * disentangle the css from all css_sets attached to the dummy
4962 * top. as in loading, we need to pay our respects to the hashtable
4965 write_lock(&css_set_lock
);
4966 list_for_each_entry(link
, &cgroup_dummy_top
->cset_links
, cset_link
) {
4967 struct css_set
*cset
= link
->cset
;
4970 hash_del(&cset
->hlist
);
4971 cset
->subsys
[ss
->subsys_id
] = NULL
;
4972 key
= css_set_hash(cset
->subsys
);
4973 hash_add(css_set_table
, &cset
->hlist
, key
);
4975 write_unlock(&css_set_lock
);
4978 * remove subsystem's css from the cgroup_dummy_top and free it -
4979 * need to free before marking as null because ss->css_free needs
4980 * the cgrp->subsys pointer to find their state.
4982 ss
->css_free(cgroup_css(cgroup_dummy_top
, ss
));
4983 RCU_INIT_POINTER(cgroup_dummy_top
->subsys
[ss
->subsys_id
], NULL
);
4985 mutex_unlock(&cgroup_mutex
);
4987 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4990 * cgroup_init_early - cgroup initialization at system boot
4992 * Initialize cgroups at system boot, and initialize any
4993 * subsystems that request early init.
4995 int __init
cgroup_init_early(void)
4997 struct cgroup_subsys
*ss
;
5000 atomic_set(&init_css_set
.refcount
, 1);
5001 INIT_LIST_HEAD(&init_css_set
.cgrp_links
);
5002 INIT_LIST_HEAD(&init_css_set
.tasks
);
5003 INIT_HLIST_NODE(&init_css_set
.hlist
);
5005 init_cgroup_root(&cgroup_dummy_root
);
5006 cgroup_root_count
= 1;
5007 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5009 init_cgrp_cset_link
.cset
= &init_css_set
;
5010 init_cgrp_cset_link
.cgrp
= cgroup_dummy_top
;
5011 list_add(&init_cgrp_cset_link
.cset_link
, &cgroup_dummy_top
->cset_links
);
5012 list_add(&init_cgrp_cset_link
.cgrp_link
, &init_css_set
.cgrp_links
);
5014 /* at bootup time, we don't worry about modular subsystems */
5015 for_each_builtin_subsys(ss
, i
) {
5017 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
5018 BUG_ON(!ss
->css_alloc
);
5019 BUG_ON(!ss
->css_free
);
5020 if (ss
->subsys_id
!= i
) {
5021 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
5022 ss
->name
, ss
->subsys_id
);
5027 cgroup_init_subsys(ss
);
5033 * cgroup_init - cgroup initialization
5035 * Register cgroup filesystem and /proc file, and initialize
5036 * any subsystems that didn't request early init.
5038 int __init
cgroup_init(void)
5040 struct cgroup_subsys
*ss
;
5044 err
= bdi_init(&cgroup_backing_dev_info
);
5048 for_each_builtin_subsys(ss
, i
) {
5049 if (!ss
->early_init
)
5050 cgroup_init_subsys(ss
);
5053 /* allocate id for the dummy hierarchy */
5054 mutex_lock(&cgroup_mutex
);
5055 mutex_lock(&cgroup_root_mutex
);
5057 /* Add init_css_set to the hash table */
5058 key
= css_set_hash(init_css_set
.subsys
);
5059 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
5061 BUG_ON(cgroup_init_root_id(&cgroup_dummy_root
, 0, 1));
5063 err
= idr_alloc(&cgroup_dummy_root
.cgroup_idr
, cgroup_dummy_top
,
5067 mutex_unlock(&cgroup_root_mutex
);
5068 mutex_unlock(&cgroup_mutex
);
5070 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
5076 err
= register_filesystem(&cgroup_fs_type
);
5078 kobject_put(cgroup_kobj
);
5082 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
5086 bdi_destroy(&cgroup_backing_dev_info
);
5091 static int __init
cgroup_wq_init(void)
5094 * There isn't much point in executing destruction path in
5095 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5096 * Use 1 for @max_active.
5098 * We would prefer to do this in cgroup_init() above, but that
5099 * is called before init_workqueues(): so leave this until after.
5101 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5102 BUG_ON(!cgroup_destroy_wq
);
5105 core_initcall(cgroup_wq_init
);
5108 * proc_cgroup_show()
5109 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5110 * - Used for /proc/<pid>/cgroup.
5111 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
5112 * doesn't really matter if tsk->cgroup changes after we read it,
5113 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
5114 * anyway. No need to check that tsk->cgroup != NULL, thanks to
5115 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
5116 * cgroup to top_cgroup.
5119 /* TODO: Use a proper seq_file iterator */
5120 int proc_cgroup_show(struct seq_file
*m
, void *v
)
5123 struct task_struct
*tsk
;
5126 struct cgroupfs_root
*root
;
5129 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5135 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
5141 mutex_lock(&cgroup_mutex
);
5143 for_each_active_root(root
) {
5144 struct cgroup_subsys
*ss
;
5145 struct cgroup
*cgrp
;
5148 seq_printf(m
, "%d:", root
->hierarchy_id
);
5149 for_each_root_subsys(root
, ss
)
5150 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
5151 if (strlen(root
->name
))
5152 seq_printf(m
, "%sname=%s", count
? "," : "",
5155 cgrp
= task_cgroup_from_root(tsk
, root
);
5156 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
5164 mutex_unlock(&cgroup_mutex
);
5165 put_task_struct(tsk
);
5172 /* Display information about each subsystem and each hierarchy */
5173 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5175 struct cgroup_subsys
*ss
;
5178 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5180 * ideally we don't want subsystems moving around while we do this.
5181 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5182 * subsys/hierarchy state.
5184 mutex_lock(&cgroup_mutex
);
5186 for_each_subsys(ss
, i
)
5187 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5188 ss
->name
, ss
->root
->hierarchy_id
,
5189 ss
->root
->number_of_cgroups
, !ss
->disabled
);
5191 mutex_unlock(&cgroup_mutex
);
5195 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5197 return single_open(file
, proc_cgroupstats_show
, NULL
);
5200 static const struct file_operations proc_cgroupstats_operations
= {
5201 .open
= cgroupstats_open
,
5203 .llseek
= seq_lseek
,
5204 .release
= single_release
,
5208 * cgroup_fork - attach newly forked task to its parents cgroup.
5209 * @child: pointer to task_struct of forking parent process.
5211 * Description: A task inherits its parent's cgroup at fork().
5213 * A pointer to the shared css_set was automatically copied in
5214 * fork.c by dup_task_struct(). However, we ignore that copy, since
5215 * it was not made under the protection of RCU or cgroup_mutex, so
5216 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
5217 * have already changed current->cgroups, allowing the previously
5218 * referenced cgroup group to be removed and freed.
5220 * At the point that cgroup_fork() is called, 'current' is the parent
5221 * task, and the passed argument 'child' points to the child task.
5223 void cgroup_fork(struct task_struct
*child
)
5226 get_css_set(task_css_set(current
));
5227 child
->cgroups
= current
->cgroups
;
5228 task_unlock(current
);
5229 INIT_LIST_HEAD(&child
->cg_list
);
5233 * cgroup_post_fork - called on a new task after adding it to the task list
5234 * @child: the task in question
5236 * Adds the task to the list running through its css_set if necessary and
5237 * call the subsystem fork() callbacks. Has to be after the task is
5238 * visible on the task list in case we race with the first call to
5239 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5242 void cgroup_post_fork(struct task_struct
*child
)
5244 struct cgroup_subsys
*ss
;
5248 * use_task_css_set_links is set to 1 before we walk the tasklist
5249 * under the tasklist_lock and we read it here after we added the child
5250 * to the tasklist under the tasklist_lock as well. If the child wasn't
5251 * yet in the tasklist when we walked through it from
5252 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
5253 * should be visible now due to the paired locking and barriers implied
5254 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
5255 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
5258 if (use_task_css_set_links
) {
5259 write_lock(&css_set_lock
);
5261 if (list_empty(&child
->cg_list
))
5262 list_add(&child
->cg_list
, &task_css_set(child
)->tasks
);
5264 write_unlock(&css_set_lock
);
5268 * Call ss->fork(). This must happen after @child is linked on
5269 * css_set; otherwise, @child might change state between ->fork()
5270 * and addition to css_set.
5272 if (need_forkexit_callback
) {
5274 * fork/exit callbacks are supported only for builtin
5275 * subsystems, and the builtin section of the subsys
5276 * array is immutable, so we don't need to lock the
5277 * subsys array here. On the other hand, modular section
5278 * of the array can be freed at module unload, so we
5281 for_each_builtin_subsys(ss
, i
)
5288 * cgroup_exit - detach cgroup from exiting task
5289 * @tsk: pointer to task_struct of exiting process
5290 * @run_callback: run exit callbacks?
5292 * Description: Detach cgroup from @tsk and release it.
5294 * Note that cgroups marked notify_on_release force every task in
5295 * them to take the global cgroup_mutex mutex when exiting.
5296 * This could impact scaling on very large systems. Be reluctant to
5297 * use notify_on_release cgroups where very high task exit scaling
5298 * is required on large systems.
5300 * the_top_cgroup_hack:
5302 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
5304 * We call cgroup_exit() while the task is still competent to
5305 * handle notify_on_release(), then leave the task attached to the
5306 * root cgroup in each hierarchy for the remainder of its exit.
5308 * To do this properly, we would increment the reference count on
5309 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
5310 * code we would add a second cgroup function call, to drop that
5311 * reference. This would just create an unnecessary hot spot on
5312 * the top_cgroup reference count, to no avail.
5314 * Normally, holding a reference to a cgroup without bumping its
5315 * count is unsafe. The cgroup could go away, or someone could
5316 * attach us to a different cgroup, decrementing the count on
5317 * the first cgroup that we never incremented. But in this case,
5318 * top_cgroup isn't going away, and either task has PF_EXITING set,
5319 * which wards off any cgroup_attach_task() attempts, or task is a failed
5320 * fork, never visible to cgroup_attach_task.
5322 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
5324 struct cgroup_subsys
*ss
;
5325 struct css_set
*cset
;
5329 * Unlink from the css_set task list if necessary.
5330 * Optimistically check cg_list before taking
5333 if (!list_empty(&tsk
->cg_list
)) {
5334 write_lock(&css_set_lock
);
5335 if (!list_empty(&tsk
->cg_list
))
5336 list_del_init(&tsk
->cg_list
);
5337 write_unlock(&css_set_lock
);
5340 /* Reassign the task to the init_css_set. */
5342 cset
= task_css_set(tsk
);
5343 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5345 if (run_callbacks
&& need_forkexit_callback
) {
5347 * fork/exit callbacks are supported only for builtin
5348 * subsystems, see cgroup_post_fork() for details.
5350 for_each_builtin_subsys(ss
, i
) {
5352 struct cgroup_subsys_state
*old_css
= cset
->subsys
[i
];
5353 struct cgroup_subsys_state
*css
= task_css(tsk
, i
);
5355 ss
->exit(css
, old_css
, tsk
);
5361 put_css_set_taskexit(cset
);
5364 static void check_for_release(struct cgroup
*cgrp
)
5366 if (cgroup_is_releasable(cgrp
) &&
5367 list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
)) {
5369 * Control Group is currently removeable. If it's not
5370 * already queued for a userspace notification, queue
5373 int need_schedule_work
= 0;
5375 raw_spin_lock(&release_list_lock
);
5376 if (!cgroup_is_dead(cgrp
) &&
5377 list_empty(&cgrp
->release_list
)) {
5378 list_add(&cgrp
->release_list
, &release_list
);
5379 need_schedule_work
= 1;
5381 raw_spin_unlock(&release_list_lock
);
5382 if (need_schedule_work
)
5383 schedule_work(&release_agent_work
);
5388 * Notify userspace when a cgroup is released, by running the
5389 * configured release agent with the name of the cgroup (path
5390 * relative to the root of cgroup file system) as the argument.
5392 * Most likely, this user command will try to rmdir this cgroup.
5394 * This races with the possibility that some other task will be
5395 * attached to this cgroup before it is removed, or that some other
5396 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5397 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5398 * unused, and this cgroup will be reprieved from its death sentence,
5399 * to continue to serve a useful existence. Next time it's released,
5400 * we will get notified again, if it still has 'notify_on_release' set.
5402 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5403 * means only wait until the task is successfully execve()'d. The
5404 * separate release agent task is forked by call_usermodehelper(),
5405 * then control in this thread returns here, without waiting for the
5406 * release agent task. We don't bother to wait because the caller of
5407 * this routine has no use for the exit status of the release agent
5408 * task, so no sense holding our caller up for that.
5410 static void cgroup_release_agent(struct work_struct
*work
)
5412 BUG_ON(work
!= &release_agent_work
);
5413 mutex_lock(&cgroup_mutex
);
5414 raw_spin_lock(&release_list_lock
);
5415 while (!list_empty(&release_list
)) {
5416 char *argv
[3], *envp
[3];
5418 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5419 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5422 list_del_init(&cgrp
->release_list
);
5423 raw_spin_unlock(&release_list_lock
);
5424 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5427 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5429 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5434 argv
[i
++] = agentbuf
;
5435 argv
[i
++] = pathbuf
;
5439 /* minimal command environment */
5440 envp
[i
++] = "HOME=/";
5441 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5444 /* Drop the lock while we invoke the usermode helper,
5445 * since the exec could involve hitting disk and hence
5446 * be a slow process */
5447 mutex_unlock(&cgroup_mutex
);
5448 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5449 mutex_lock(&cgroup_mutex
);
5453 raw_spin_lock(&release_list_lock
);
5455 raw_spin_unlock(&release_list_lock
);
5456 mutex_unlock(&cgroup_mutex
);
5459 static int __init
cgroup_disable(char *str
)
5461 struct cgroup_subsys
*ss
;
5465 while ((token
= strsep(&str
, ",")) != NULL
) {
5470 * cgroup_disable, being at boot time, can't know about
5471 * module subsystems, so we don't worry about them.
5473 for_each_builtin_subsys(ss
, i
) {
5474 if (!strcmp(token
, ss
->name
)) {
5476 printk(KERN_INFO
"Disabling %s control group"
5477 " subsystem\n", ss
->name
);
5484 __setup("cgroup_disable=", cgroup_disable
);
5487 * css_from_dir - get corresponding css from the dentry of a cgroup dir
5488 * @dentry: directory dentry of interest
5489 * @ss: subsystem of interest
5491 * Must be called under RCU read lock. The caller is responsible for
5492 * pinning the returned css if it needs to be accessed outside the RCU
5495 struct cgroup_subsys_state
*css_from_dir(struct dentry
*dentry
,
5496 struct cgroup_subsys
*ss
)
5498 struct cgroup
*cgrp
;
5500 WARN_ON_ONCE(!rcu_read_lock_held());
5502 /* is @dentry a cgroup dir? */
5503 if (!dentry
->d_inode
||
5504 dentry
->d_inode
->i_op
!= &cgroup_dir_inode_operations
)
5505 return ERR_PTR(-EBADF
);
5507 cgrp
= __d_cgrp(dentry
);
5508 return cgroup_css(cgrp
, ss
) ?: ERR_PTR(-ENOENT
);
5512 * css_from_id - lookup css by id
5513 * @id: the cgroup id
5514 * @ss: cgroup subsys to be looked into
5516 * Returns the css if there's valid one with @id, otherwise returns NULL.
5517 * Should be called under rcu_read_lock().
5519 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5521 struct cgroup
*cgrp
;
5523 rcu_lockdep_assert(rcu_read_lock_held() ||
5524 lockdep_is_held(&cgroup_mutex
),
5525 "css_from_id() needs proper protection");
5527 cgrp
= idr_find(&ss
->root
->cgroup_idr
, id
);
5529 return cgroup_css(cgrp
, ss
);
5533 #ifdef CONFIG_CGROUP_DEBUG
5534 static struct cgroup_subsys_state
*
5535 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
5537 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5540 return ERR_PTR(-ENOMEM
);
5545 static void debug_css_free(struct cgroup_subsys_state
*css
)
5550 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
5553 return cgroup_task_count(css
->cgroup
);
5556 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
5559 return (u64
)(unsigned long)current
->cgroups
;
5562 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
5568 count
= atomic_read(&task_css_set(current
)->refcount
);
5573 static int current_css_set_cg_links_read(struct cgroup_subsys_state
*css
,
5575 struct seq_file
*seq
)
5577 struct cgrp_cset_link
*link
;
5578 struct css_set
*cset
;
5580 read_lock(&css_set_lock
);
5582 cset
= rcu_dereference(current
->cgroups
);
5583 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5584 struct cgroup
*c
= link
->cgrp
;
5588 name
= c
->dentry
->d_name
.name
;
5591 seq_printf(seq
, "Root %d group %s\n",
5592 c
->root
->hierarchy_id
, name
);
5595 read_unlock(&css_set_lock
);
5599 #define MAX_TASKS_SHOWN_PER_CSS 25
5600 static int cgroup_css_links_read(struct cgroup_subsys_state
*css
,
5601 struct cftype
*cft
, struct seq_file
*seq
)
5603 struct cgrp_cset_link
*link
;
5605 read_lock(&css_set_lock
);
5606 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
5607 struct css_set
*cset
= link
->cset
;
5608 struct task_struct
*task
;
5610 seq_printf(seq
, "css_set %p\n", cset
);
5611 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5612 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5613 seq_puts(seq
, " ...\n");
5616 seq_printf(seq
, " task %d\n",
5617 task_pid_vnr(task
));
5621 read_unlock(&css_set_lock
);
5625 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
5627 return test_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
5630 static struct cftype debug_files
[] = {
5632 .name
= "taskcount",
5633 .read_u64
= debug_taskcount_read
,
5637 .name
= "current_css_set",
5638 .read_u64
= current_css_set_read
,
5642 .name
= "current_css_set_refcount",
5643 .read_u64
= current_css_set_refcount_read
,
5647 .name
= "current_css_set_cg_links",
5648 .read_seq_string
= current_css_set_cg_links_read
,
5652 .name
= "cgroup_css_links",
5653 .read_seq_string
= cgroup_css_links_read
,
5657 .name
= "releasable",
5658 .read_u64
= releasable_read
,
5664 struct cgroup_subsys debug_subsys
= {
5666 .css_alloc
= debug_css_alloc
,
5667 .css_free
= debug_css_free
,
5668 .subsys_id
= debug_subsys_id
,
5669 .base_cftypes
= debug_files
,
5671 #endif /* CONFIG_CGROUP_DEBUG */