2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
33 #include <linux/init_task.h>
34 #include <linux/kernel.h>
35 #include <linux/list.h>
37 #include <linux/mutex.h>
38 #include <linux/mount.h>
39 #include <linux/pagemap.h>
40 #include <linux/proc_fs.h>
41 #include <linux/rcupdate.h>
42 #include <linux/sched.h>
43 #include <linux/backing-dev.h>
44 #include <linux/seq_file.h>
45 #include <linux/slab.h>
46 #include <linux/magic.h>
47 #include <linux/spinlock.h>
48 #include <linux/string.h>
49 #include <linux/sort.h>
50 #include <linux/kmod.h>
51 #include <linux/module.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/namei.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/idr.h>
58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
59 #include <linux/eventfd.h>
60 #include <linux/poll.h>
61 #include <linux/flex_array.h> /* used in cgroup_attach_task */
62 #include <linux/kthread.h>
64 #include <linux/atomic.h>
67 * cgroup_mutex is the master lock. Any modification to cgroup or its
68 * hierarchy must be performed while holding it.
70 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
71 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
72 * release_agent_path and so on. Modifying requires both cgroup_mutex and
73 * cgroup_root_mutex. Readers can acquire either of the two. This is to
74 * break the following locking order cycle.
76 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
77 * B. namespace_sem -> cgroup_mutex
79 * B happens only through cgroup_show_options() and using cgroup_root_mutex
82 #ifdef CONFIG_PROVE_RCU
83 DEFINE_MUTEX(cgroup_mutex
);
84 EXPORT_SYMBOL_GPL(cgroup_mutex
); /* only for lockdep */
86 static DEFINE_MUTEX(cgroup_mutex
);
89 static DEFINE_MUTEX(cgroup_root_mutex
);
92 * Generate an array of cgroup subsystem pointers. At boot time, this is
93 * populated with the built in subsystems, and modular subsystems are
94 * registered after that. The mutable section of this array is protected by
97 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
98 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
99 static struct cgroup_subsys
*cgroup_subsys
[CGROUP_SUBSYS_COUNT
] = {
100 #include <linux/cgroup_subsys.h>
104 * The dummy hierarchy, reserved for the subsystems that are otherwise
105 * unattached - it never has more than a single cgroup, and all tasks are
106 * part of that cgroup.
108 static struct cgroupfs_root cgroup_dummy_root
;
110 /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */
111 static struct cgroup
* const cgroup_dummy_top
= &cgroup_dummy_root
.top_cgroup
;
114 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
117 struct list_head node
;
118 struct dentry
*dentry
;
122 struct simple_xattrs xattrs
;
126 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
127 * cgroup_subsys->use_id != 0.
129 #define CSS_ID_MAX (65535)
132 * The css to which this ID points. This pointer is set to valid value
133 * after cgroup is populated. If cgroup is removed, this will be NULL.
134 * This pointer is expected to be RCU-safe because destroy()
135 * is called after synchronize_rcu(). But for safe use, css_tryget()
136 * should be used for avoiding race.
138 struct cgroup_subsys_state __rcu
*css
;
144 * Depth in hierarchy which this ID belongs to.
146 unsigned short depth
;
148 * ID is freed by RCU. (and lookup routine is RCU safe.)
150 struct rcu_head rcu_head
;
152 * Hierarchy of CSS ID belongs to.
154 unsigned short stack
[0]; /* Array of Length (depth+1) */
158 * cgroup_event represents events which userspace want to receive.
160 struct cgroup_event
{
162 * Cgroup which the event belongs to.
166 * Control file which the event associated.
170 * eventfd to signal userspace about the event.
172 struct eventfd_ctx
*eventfd
;
174 * Each of these stored in a list by the cgroup.
176 struct list_head list
;
178 * All fields below needed to unregister event when
179 * userspace closes eventfd.
182 wait_queue_head_t
*wqh
;
184 struct work_struct remove
;
187 /* The list of hierarchy roots */
189 static LIST_HEAD(cgroup_roots
);
190 static int cgroup_root_count
;
193 * Hierarchy ID allocation and mapping. It follows the same exclusion
194 * rules as other root ops - both cgroup_mutex and cgroup_root_mutex for
195 * writes, either for reads.
197 static DEFINE_IDR(cgroup_hierarchy_idr
);
199 static struct cgroup_name root_cgroup_name
= { .name
= "/" };
202 * Assign a monotonically increasing serial number to cgroups. It
203 * guarantees cgroups with bigger numbers are newer than those with smaller
204 * numbers. Also, as cgroups are always appended to the parent's
205 * ->children list, it guarantees that sibling cgroups are always sorted in
206 * the ascending serial number order on the list. Protected by
209 static u64 cgroup_serial_nr_next
= 1;
211 /* This flag indicates whether tasks in the fork and exit paths should
212 * check for fork/exit handlers to call. This avoids us having to do
213 * extra work in the fork/exit path if none of the subsystems need to
216 static int need_forkexit_callback __read_mostly
;
218 static struct cftype cgroup_base_files
[];
220 static void cgroup_offline_fn(struct work_struct
*work
);
221 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
222 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
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
;
367 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
368 struct cgroup_subsys_state
*css
);
370 /* css_set_lock protects the list of css_set objects, and the
371 * chain of tasks off each css_set. Nests outside task->alloc_lock
372 * due to cgroup_iter_start() */
373 static DEFINE_RWLOCK(css_set_lock
);
374 static int css_set_count
;
377 * hash table for cgroup groups. This improves the performance to find
378 * an existing css_set. This hash doesn't (currently) take into
379 * account cgroups in empty hierarchies.
381 #define CSS_SET_HASH_BITS 7
382 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
384 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
386 unsigned long key
= 0UL;
387 struct cgroup_subsys
*ss
;
390 for_each_subsys(ss
, i
)
391 key
+= (unsigned long)css
[i
];
392 key
= (key
>> 16) ^ key
;
397 /* We don't maintain the lists running through each css_set to its
398 * task until after the first call to cgroup_iter_start(). This
399 * reduces the fork()/exit() overhead for people who have cgroups
400 * compiled into their kernel but not actually in use */
401 static int use_task_css_set_links __read_mostly
;
403 static void __put_css_set(struct css_set
*cset
, int taskexit
)
405 struct cgrp_cset_link
*link
, *tmp_link
;
408 * Ensure that the refcount doesn't hit zero while any readers
409 * can see it. Similar to atomic_dec_and_lock(), but for an
412 if (atomic_add_unless(&cset
->refcount
, -1, 1))
414 write_lock(&css_set_lock
);
415 if (!atomic_dec_and_test(&cset
->refcount
)) {
416 write_unlock(&css_set_lock
);
420 /* This css_set is dead. unlink it and release cgroup refcounts */
421 hash_del(&cset
->hlist
);
424 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
425 struct cgroup
*cgrp
= link
->cgrp
;
427 list_del(&link
->cset_link
);
428 list_del(&link
->cgrp_link
);
430 /* @cgrp can't go away while we're holding css_set_lock */
431 if (list_empty(&cgrp
->cset_links
) && notify_on_release(cgrp
)) {
433 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
434 check_for_release(cgrp
);
440 write_unlock(&css_set_lock
);
441 kfree_rcu(cset
, rcu_head
);
445 * refcounted get/put for css_set objects
447 static inline void get_css_set(struct css_set
*cset
)
449 atomic_inc(&cset
->refcount
);
452 static inline void put_css_set(struct css_set
*cset
)
454 __put_css_set(cset
, 0);
457 static inline void put_css_set_taskexit(struct css_set
*cset
)
459 __put_css_set(cset
, 1);
463 * compare_css_sets - helper function for find_existing_css_set().
464 * @cset: candidate css_set being tested
465 * @old_cset: existing css_set for a task
466 * @new_cgrp: cgroup that's being entered by the task
467 * @template: desired set of css pointers in css_set (pre-calculated)
469 * Returns true if "cset" matches "old_cset" except for the hierarchy
470 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
472 static bool compare_css_sets(struct css_set
*cset
,
473 struct css_set
*old_cset
,
474 struct cgroup
*new_cgrp
,
475 struct cgroup_subsys_state
*template[])
477 struct list_head
*l1
, *l2
;
479 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
))) {
480 /* Not all subsystems matched */
485 * Compare cgroup pointers in order to distinguish between
486 * different cgroups in heirarchies with no subsystems. We
487 * could get by with just this check alone (and skip the
488 * memcmp above) but on most setups the memcmp check will
489 * avoid the need for this more expensive check on almost all
493 l1
= &cset
->cgrp_links
;
494 l2
= &old_cset
->cgrp_links
;
496 struct cgrp_cset_link
*link1
, *link2
;
497 struct cgroup
*cgrp1
, *cgrp2
;
501 /* See if we reached the end - both lists are equal length. */
502 if (l1
== &cset
->cgrp_links
) {
503 BUG_ON(l2
!= &old_cset
->cgrp_links
);
506 BUG_ON(l2
== &old_cset
->cgrp_links
);
508 /* Locate the cgroups associated with these links. */
509 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
510 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
513 /* Hierarchies should be linked in the same order. */
514 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
517 * If this hierarchy is the hierarchy of the cgroup
518 * that's changing, then we need to check that this
519 * css_set points to the new cgroup; if it's any other
520 * hierarchy, then this css_set should point to the
521 * same cgroup as the old css_set.
523 if (cgrp1
->root
== new_cgrp
->root
) {
524 if (cgrp1
!= new_cgrp
)
535 * find_existing_css_set - init css array and find the matching css_set
536 * @old_cset: the css_set that we're using before the cgroup transition
537 * @cgrp: the cgroup that we're moving into
538 * @template: out param for the new set of csses, should be clear on entry
540 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
542 struct cgroup_subsys_state
*template[])
544 struct cgroupfs_root
*root
= cgrp
->root
;
545 struct cgroup_subsys
*ss
;
546 struct css_set
*cset
;
551 * Build the set of subsystem state objects that we want to see in the
552 * new css_set. while subsystems can change globally, the entries here
553 * won't change, so no need for locking.
555 for_each_subsys(ss
, i
) {
556 if (root
->subsys_mask
& (1UL << i
)) {
557 /* Subsystem is in this hierarchy. So we want
558 * the subsystem state from the new
560 template[i
] = cgrp
->subsys
[i
];
562 /* Subsystem is not in this hierarchy, so we
563 * don't want to change the subsystem state */
564 template[i
] = old_cset
->subsys
[i
];
568 key
= css_set_hash(template);
569 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
570 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
573 /* This css_set matches what we need */
577 /* No existing cgroup group matched */
581 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
583 struct cgrp_cset_link
*link
, *tmp_link
;
585 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
586 list_del(&link
->cset_link
);
592 * allocate_cgrp_cset_links - allocate cgrp_cset_links
593 * @count: the number of links to allocate
594 * @tmp_links: list_head the allocated links are put on
596 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
597 * through ->cset_link. Returns 0 on success or -errno.
599 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
601 struct cgrp_cset_link
*link
;
604 INIT_LIST_HEAD(tmp_links
);
606 for (i
= 0; i
< count
; i
++) {
607 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
609 free_cgrp_cset_links(tmp_links
);
612 list_add(&link
->cset_link
, tmp_links
);
618 * link_css_set - a helper function to link a css_set to a cgroup
619 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
620 * @cset: the css_set to be linked
621 * @cgrp: the destination cgroup
623 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
626 struct cgrp_cset_link
*link
;
628 BUG_ON(list_empty(tmp_links
));
629 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
632 list_move(&link
->cset_link
, &cgrp
->cset_links
);
634 * Always add links to the tail of the list so that the list
635 * is sorted by order of hierarchy creation
637 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
641 * find_css_set - return a new css_set with one cgroup updated
642 * @old_cset: the baseline css_set
643 * @cgrp: the cgroup to be updated
645 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
646 * substituted into the appropriate hierarchy.
648 static struct css_set
*find_css_set(struct css_set
*old_cset
,
651 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
652 struct css_set
*cset
;
653 struct list_head tmp_links
;
654 struct cgrp_cset_link
*link
;
657 lockdep_assert_held(&cgroup_mutex
);
659 /* First see if we already have a cgroup group that matches
661 read_lock(&css_set_lock
);
662 cset
= find_existing_css_set(old_cset
, cgrp
, template);
665 read_unlock(&css_set_lock
);
670 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
674 /* Allocate all the cgrp_cset_link objects that we'll need */
675 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
680 atomic_set(&cset
->refcount
, 1);
681 INIT_LIST_HEAD(&cset
->cgrp_links
);
682 INIT_LIST_HEAD(&cset
->tasks
);
683 INIT_HLIST_NODE(&cset
->hlist
);
685 /* Copy the set of subsystem state objects generated in
686 * find_existing_css_set() */
687 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
689 write_lock(&css_set_lock
);
690 /* Add reference counts and links from the new css_set. */
691 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
692 struct cgroup
*c
= link
->cgrp
;
694 if (c
->root
== cgrp
->root
)
696 link_css_set(&tmp_links
, cset
, c
);
699 BUG_ON(!list_empty(&tmp_links
));
703 /* Add this cgroup group to the hash table */
704 key
= css_set_hash(cset
->subsys
);
705 hash_add(css_set_table
, &cset
->hlist
, key
);
707 write_unlock(&css_set_lock
);
713 * Return the cgroup for "task" from the given hierarchy. Must be
714 * called with cgroup_mutex held.
716 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
717 struct cgroupfs_root
*root
)
719 struct css_set
*cset
;
720 struct cgroup
*res
= NULL
;
722 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
723 read_lock(&css_set_lock
);
725 * No need to lock the task - since we hold cgroup_mutex the
726 * task can't change groups, so the only thing that can happen
727 * is that it exits and its css is set back to init_css_set.
729 cset
= task_css_set(task
);
730 if (cset
== &init_css_set
) {
731 res
= &root
->top_cgroup
;
733 struct cgrp_cset_link
*link
;
735 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
736 struct cgroup
*c
= link
->cgrp
;
738 if (c
->root
== root
) {
744 read_unlock(&css_set_lock
);
750 * There is one global cgroup mutex. We also require taking
751 * task_lock() when dereferencing a task's cgroup subsys pointers.
752 * See "The task_lock() exception", at the end of this comment.
754 * A task must hold cgroup_mutex to modify cgroups.
756 * Any task can increment and decrement the count field without lock.
757 * So in general, code holding cgroup_mutex can't rely on the count
758 * field not changing. However, if the count goes to zero, then only
759 * cgroup_attach_task() can increment it again. Because a count of zero
760 * means that no tasks are currently attached, therefore there is no
761 * way a task attached to that cgroup can fork (the other way to
762 * increment the count). So code holding cgroup_mutex can safely
763 * assume that if the count is zero, it will stay zero. Similarly, if
764 * a task holds cgroup_mutex on a cgroup with zero count, it
765 * knows that the cgroup won't be removed, as cgroup_rmdir()
768 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
769 * (usually) take cgroup_mutex. These are the two most performance
770 * critical pieces of code here. The exception occurs on cgroup_exit(),
771 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
772 * is taken, and if the cgroup count is zero, a usermode call made
773 * to the release agent with the name of the cgroup (path relative to
774 * the root of cgroup file system) as the argument.
776 * A cgroup can only be deleted if both its 'count' of using tasks
777 * is zero, and its list of 'children' cgroups is empty. Since all
778 * tasks in the system use _some_ cgroup, and since there is always at
779 * least one task in the system (init, pid == 1), therefore, top_cgroup
780 * always has either children cgroups and/or using tasks. So we don't
781 * need a special hack to ensure that top_cgroup cannot be deleted.
783 * The task_lock() exception
785 * The need for this exception arises from the action of
786 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
787 * another. It does so using cgroup_mutex, however there are
788 * several performance critical places that need to reference
789 * task->cgroup without the expense of grabbing a system global
790 * mutex. Therefore except as noted below, when dereferencing or, as
791 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
792 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
793 * the task_struct routinely used for such matters.
795 * P.S. One more locking exception. RCU is used to guard the
796 * update of a tasks cgroup pointer by cgroup_attach_task()
800 * A couple of forward declarations required, due to cyclic reference loop:
801 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
802 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
806 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
807 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, unsigned int);
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 int alloc_css_id(struct cgroup_subsys
*ss
,
819 struct cgroup
*parent
, struct cgroup
*child
);
821 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
823 struct inode
*inode
= new_inode(sb
);
826 inode
->i_ino
= get_next_ino();
827 inode
->i_mode
= mode
;
828 inode
->i_uid
= current_fsuid();
829 inode
->i_gid
= current_fsgid();
830 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
831 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
836 static struct cgroup_name
*cgroup_alloc_name(struct dentry
*dentry
)
838 struct cgroup_name
*name
;
840 name
= kmalloc(sizeof(*name
) + dentry
->d_name
.len
+ 1, GFP_KERNEL
);
843 strcpy(name
->name
, dentry
->d_name
.name
);
847 static void cgroup_free_fn(struct work_struct
*work
)
849 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, destroy_work
);
850 struct cgroup_subsys
*ss
;
852 mutex_lock(&cgroup_mutex
);
854 * Release the subsystem state objects.
856 for_each_root_subsys(cgrp
->root
, ss
) {
857 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
862 cgrp
->root
->number_of_cgroups
--;
863 mutex_unlock(&cgroup_mutex
);
866 * We get a ref to the parent's dentry, and put the ref when
867 * this cgroup is being freed, so it's guaranteed that the
868 * parent won't be destroyed before its children.
870 dput(cgrp
->parent
->dentry
);
873 * Drop the active superblock reference that we took when we
874 * created the cgroup. This will free cgrp->root, if we are
875 * holding the last reference to @sb.
877 deactivate_super(cgrp
->root
->sb
);
880 * if we're getting rid of the cgroup, refcount should ensure
881 * that there are no pidlists left.
883 BUG_ON(!list_empty(&cgrp
->pidlists
));
885 simple_xattrs_free(&cgrp
->xattrs
);
887 kfree(rcu_dereference_raw(cgrp
->name
));
891 static void cgroup_free_rcu(struct rcu_head
*head
)
893 struct cgroup
*cgrp
= container_of(head
, struct cgroup
, rcu_head
);
895 INIT_WORK(&cgrp
->destroy_work
, cgroup_free_fn
);
896 schedule_work(&cgrp
->destroy_work
);
899 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
901 /* is dentry a directory ? if so, kfree() associated cgroup */
902 if (S_ISDIR(inode
->i_mode
)) {
903 struct cgroup
*cgrp
= dentry
->d_fsdata
;
905 BUG_ON(!(cgroup_is_dead(cgrp
)));
906 call_rcu(&cgrp
->rcu_head
, cgroup_free_rcu
);
908 struct cfent
*cfe
= __d_cfe(dentry
);
909 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
911 WARN_ONCE(!list_empty(&cfe
->node
) &&
912 cgrp
!= &cgrp
->root
->top_cgroup
,
913 "cfe still linked for %s\n", cfe
->type
->name
);
914 simple_xattrs_free(&cfe
->xattrs
);
920 static int cgroup_delete(const struct dentry
*d
)
925 static void remove_dir(struct dentry
*d
)
927 struct dentry
*parent
= dget(d
->d_parent
);
930 simple_rmdir(parent
->d_inode
, d
);
934 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
938 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
939 lockdep_assert_held(&cgroup_mutex
);
942 * If we're doing cleanup due to failure of cgroup_create(),
943 * the corresponding @cfe may not exist.
945 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
946 struct dentry
*d
= cfe
->dentry
;
948 if (cft
&& cfe
->type
!= cft
)
953 simple_unlink(cgrp
->dentry
->d_inode
, d
);
954 list_del_init(&cfe
->node
);
962 * cgroup_clear_dir - remove subsys files in a cgroup directory
963 * @cgrp: target cgroup
964 * @subsys_mask: mask of the subsystem ids whose files should be removed
966 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
968 struct cgroup_subsys
*ss
;
971 for_each_subsys(ss
, i
) {
972 struct cftype_set
*set
;
974 if (!test_bit(i
, &subsys_mask
))
976 list_for_each_entry(set
, &ss
->cftsets
, node
)
977 cgroup_addrm_files(cgrp
, set
->cfts
, false);
982 * NOTE : the dentry must have been dget()'ed
984 static void cgroup_d_remove_dir(struct dentry
*dentry
)
986 struct dentry
*parent
;
988 parent
= dentry
->d_parent
;
989 spin_lock(&parent
->d_lock
);
990 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
991 list_del_init(&dentry
->d_u
.d_child
);
992 spin_unlock(&dentry
->d_lock
);
993 spin_unlock(&parent
->d_lock
);
998 * Call with cgroup_mutex held. Drops reference counts on modules, including
999 * any duplicate ones that parse_cgroupfs_options took. If this function
1000 * returns an error, no reference counts are touched.
1002 static int rebind_subsystems(struct cgroupfs_root
*root
,
1003 unsigned long added_mask
, unsigned removed_mask
)
1005 struct cgroup
*cgrp
= &root
->top_cgroup
;
1006 struct cgroup_subsys
*ss
;
1007 unsigned long pinned
= 0;
1010 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1011 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1013 /* Check that any added subsystems are currently free */
1014 for_each_subsys(ss
, i
) {
1015 if (!(added_mask
& (1 << i
)))
1018 /* is the subsystem mounted elsewhere? */
1019 if (ss
->root
!= &cgroup_dummy_root
) {
1024 /* pin the module */
1025 if (!try_module_get(ss
->module
)) {
1032 /* subsys could be missing if unloaded between parsing and here */
1033 if (added_mask
!= pinned
) {
1038 ret
= cgroup_populate_dir(cgrp
, added_mask
);
1043 * Nothing can fail from this point on. Remove files for the
1044 * removed subsystems and rebind each subsystem.
1046 cgroup_clear_dir(cgrp
, removed_mask
);
1048 for_each_subsys(ss
, i
) {
1049 unsigned long bit
= 1UL << i
;
1051 if (bit
& added_mask
) {
1052 /* We're binding this subsystem to this hierarchy */
1053 BUG_ON(cgrp
->subsys
[i
]);
1054 BUG_ON(!cgroup_dummy_top
->subsys
[i
]);
1055 BUG_ON(cgroup_dummy_top
->subsys
[i
]->cgroup
!= cgroup_dummy_top
);
1057 cgrp
->subsys
[i
] = cgroup_dummy_top
->subsys
[i
];
1058 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1059 list_move(&ss
->sibling
, &root
->subsys_list
);
1062 ss
->bind(cgrp
->subsys
[i
]);
1064 /* refcount was already taken, and we're keeping it */
1065 root
->subsys_mask
|= bit
;
1066 } else if (bit
& removed_mask
) {
1067 /* We're removing this subsystem */
1068 BUG_ON(cgrp
->subsys
[i
] != cgroup_dummy_top
->subsys
[i
]);
1069 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1072 ss
->bind(cgroup_dummy_top
->subsys
[i
]);
1073 cgroup_dummy_top
->subsys
[i
]->cgroup
= cgroup_dummy_top
;
1074 cgrp
->subsys
[i
] = NULL
;
1075 cgroup_subsys
[i
]->root
= &cgroup_dummy_root
;
1076 list_move(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
1078 /* subsystem is now free - drop reference on module */
1079 module_put(ss
->module
);
1080 root
->subsys_mask
&= ~bit
;
1085 * Mark @root has finished binding subsystems. @root->subsys_mask
1086 * now matches the bound subsystems.
1088 root
->flags
|= CGRP_ROOT_SUBSYS_BOUND
;
1093 for_each_subsys(ss
, i
)
1094 if (pinned
& (1 << i
))
1095 module_put(ss
->module
);
1099 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1101 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1102 struct cgroup_subsys
*ss
;
1104 mutex_lock(&cgroup_root_mutex
);
1105 for_each_root_subsys(root
, ss
)
1106 seq_printf(seq
, ",%s", ss
->name
);
1107 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
)
1108 seq_puts(seq
, ",sane_behavior");
1109 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1110 seq_puts(seq
, ",noprefix");
1111 if (root
->flags
& CGRP_ROOT_XATTR
)
1112 seq_puts(seq
, ",xattr");
1113 if (strlen(root
->release_agent_path
))
1114 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1115 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
))
1116 seq_puts(seq
, ",clone_children");
1117 if (strlen(root
->name
))
1118 seq_printf(seq
, ",name=%s", root
->name
);
1119 mutex_unlock(&cgroup_root_mutex
);
1123 struct cgroup_sb_opts
{
1124 unsigned long subsys_mask
;
1125 unsigned long flags
;
1126 char *release_agent
;
1127 bool cpuset_clone_children
;
1129 /* User explicitly requested empty subsystem */
1132 struct cgroupfs_root
*new_root
;
1137 * Convert a hierarchy specifier into a bitmask of subsystems and
1138 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1139 * array. This function takes refcounts on subsystems to be used, unless it
1140 * returns error, in which case no refcounts are taken.
1142 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1144 char *token
, *o
= data
;
1145 bool all_ss
= false, one_ss
= false;
1146 unsigned long mask
= (unsigned long)-1;
1147 struct cgroup_subsys
*ss
;
1150 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1152 #ifdef CONFIG_CPUSETS
1153 mask
= ~(1UL << cpuset_subsys_id
);
1156 memset(opts
, 0, sizeof(*opts
));
1158 while ((token
= strsep(&o
, ",")) != NULL
) {
1161 if (!strcmp(token
, "none")) {
1162 /* Explicitly have no subsystems */
1166 if (!strcmp(token
, "all")) {
1167 /* Mutually exclusive option 'all' + subsystem name */
1173 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1174 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1177 if (!strcmp(token
, "noprefix")) {
1178 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1181 if (!strcmp(token
, "clone_children")) {
1182 opts
->cpuset_clone_children
= true;
1185 if (!strcmp(token
, "xattr")) {
1186 opts
->flags
|= CGRP_ROOT_XATTR
;
1189 if (!strncmp(token
, "release_agent=", 14)) {
1190 /* Specifying two release agents is forbidden */
1191 if (opts
->release_agent
)
1193 opts
->release_agent
=
1194 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1195 if (!opts
->release_agent
)
1199 if (!strncmp(token
, "name=", 5)) {
1200 const char *name
= token
+ 5;
1201 /* Can't specify an empty name */
1204 /* Must match [\w.-]+ */
1205 for (i
= 0; i
< strlen(name
); i
++) {
1209 if ((c
== '.') || (c
== '-') || (c
== '_'))
1213 /* Specifying two names is forbidden */
1216 opts
->name
= kstrndup(name
,
1217 MAX_CGROUP_ROOT_NAMELEN
- 1,
1225 for_each_subsys(ss
, i
) {
1226 if (strcmp(token
, ss
->name
))
1231 /* Mutually exclusive option 'all' + subsystem name */
1234 set_bit(i
, &opts
->subsys_mask
);
1239 if (i
== CGROUP_SUBSYS_COUNT
)
1244 * If the 'all' option was specified select all the subsystems,
1245 * otherwise if 'none', 'name=' and a subsystem name options
1246 * were not specified, let's default to 'all'
1248 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1249 for_each_subsys(ss
, i
)
1251 set_bit(i
, &opts
->subsys_mask
);
1253 /* Consistency checks */
1255 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1256 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1258 if (opts
->flags
& CGRP_ROOT_NOPREFIX
) {
1259 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1263 if (opts
->cpuset_clone_children
) {
1264 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1270 * Option noprefix was introduced just for backward compatibility
1271 * with the old cpuset, so we allow noprefix only if mounting just
1272 * the cpuset subsystem.
1274 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1278 /* Can't specify "none" and some subsystems */
1279 if (opts
->subsys_mask
&& opts
->none
)
1283 * We either have to specify by name or by subsystems. (So all
1284 * empty hierarchies must have a name).
1286 if (!opts
->subsys_mask
&& !opts
->name
)
1292 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1295 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1296 struct cgroup
*cgrp
= &root
->top_cgroup
;
1297 struct cgroup_sb_opts opts
;
1298 unsigned long added_mask
, removed_mask
;
1300 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1301 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1305 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1306 mutex_lock(&cgroup_mutex
);
1307 mutex_lock(&cgroup_root_mutex
);
1309 /* See what subsystems are wanted */
1310 ret
= parse_cgroupfs_options(data
, &opts
);
1314 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1315 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1316 task_tgid_nr(current
), current
->comm
);
1318 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1319 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1321 /* Don't allow flags or name to change at remount */
1322 if (((opts
.flags
^ root
->flags
) & CGRP_ROOT_OPTION_MASK
) ||
1323 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1324 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1325 opts
.flags
& CGRP_ROOT_OPTION_MASK
, opts
.name
?: "",
1326 root
->flags
& CGRP_ROOT_OPTION_MASK
, root
->name
);
1331 /* remounting is not allowed for populated hierarchies */
1332 if (root
->number_of_cgroups
> 1) {
1337 ret
= rebind_subsystems(root
, added_mask
, removed_mask
);
1341 if (opts
.release_agent
)
1342 strcpy(root
->release_agent_path
, opts
.release_agent
);
1344 kfree(opts
.release_agent
);
1346 mutex_unlock(&cgroup_root_mutex
);
1347 mutex_unlock(&cgroup_mutex
);
1348 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1352 static const struct super_operations cgroup_ops
= {
1353 .statfs
= simple_statfs
,
1354 .drop_inode
= generic_delete_inode
,
1355 .show_options
= cgroup_show_options
,
1356 .remount_fs
= cgroup_remount
,
1359 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1361 INIT_LIST_HEAD(&cgrp
->sibling
);
1362 INIT_LIST_HEAD(&cgrp
->children
);
1363 INIT_LIST_HEAD(&cgrp
->files
);
1364 INIT_LIST_HEAD(&cgrp
->cset_links
);
1365 INIT_LIST_HEAD(&cgrp
->release_list
);
1366 INIT_LIST_HEAD(&cgrp
->pidlists
);
1367 mutex_init(&cgrp
->pidlist_mutex
);
1368 cgrp
->dummy_css
.cgroup
= cgrp
;
1369 INIT_LIST_HEAD(&cgrp
->event_list
);
1370 spin_lock_init(&cgrp
->event_list_lock
);
1371 simple_xattrs_init(&cgrp
->xattrs
);
1374 static void init_cgroup_root(struct cgroupfs_root
*root
)
1376 struct cgroup
*cgrp
= &root
->top_cgroup
;
1378 INIT_LIST_HEAD(&root
->subsys_list
);
1379 INIT_LIST_HEAD(&root
->root_list
);
1380 root
->number_of_cgroups
= 1;
1382 RCU_INIT_POINTER(cgrp
->name
, &root_cgroup_name
);
1383 init_cgroup_housekeeping(cgrp
);
1384 idr_init(&root
->cgroup_idr
);
1387 static int cgroup_init_root_id(struct cgroupfs_root
*root
, int start
, int end
)
1391 lockdep_assert_held(&cgroup_mutex
);
1392 lockdep_assert_held(&cgroup_root_mutex
);
1394 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, start
, end
,
1399 root
->hierarchy_id
= id
;
1403 static void cgroup_exit_root_id(struct cgroupfs_root
*root
)
1405 lockdep_assert_held(&cgroup_mutex
);
1406 lockdep_assert_held(&cgroup_root_mutex
);
1408 if (root
->hierarchy_id
) {
1409 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1410 root
->hierarchy_id
= 0;
1414 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1416 struct cgroup_sb_opts
*opts
= data
;
1417 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1419 /* If we asked for a name then it must match */
1420 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1424 * If we asked for subsystems (or explicitly for no
1425 * subsystems) then they must match
1427 if ((opts
->subsys_mask
|| opts
->none
)
1428 && (opts
->subsys_mask
!= root
->subsys_mask
))
1434 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1436 struct cgroupfs_root
*root
;
1438 if (!opts
->subsys_mask
&& !opts
->none
)
1441 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1443 return ERR_PTR(-ENOMEM
);
1445 init_cgroup_root(root
);
1448 * We need to set @root->subsys_mask now so that @root can be
1449 * matched by cgroup_test_super() before it finishes
1450 * initialization; otherwise, competing mounts with the same
1451 * options may try to bind the same subsystems instead of waiting
1452 * for the first one leading to unexpected mount errors.
1453 * SUBSYS_BOUND will be set once actual binding is complete.
1455 root
->subsys_mask
= opts
->subsys_mask
;
1456 root
->flags
= opts
->flags
;
1457 if (opts
->release_agent
)
1458 strcpy(root
->release_agent_path
, opts
->release_agent
);
1460 strcpy(root
->name
, opts
->name
);
1461 if (opts
->cpuset_clone_children
)
1462 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1466 static void cgroup_free_root(struct cgroupfs_root
*root
)
1469 /* hierarhcy ID shoulid already have been released */
1470 WARN_ON_ONCE(root
->hierarchy_id
);
1472 idr_destroy(&root
->cgroup_idr
);
1477 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1480 struct cgroup_sb_opts
*opts
= data
;
1482 /* If we don't have a new root, we can't set up a new sb */
1483 if (!opts
->new_root
)
1486 BUG_ON(!opts
->subsys_mask
&& !opts
->none
);
1488 ret
= set_anon_super(sb
, NULL
);
1492 sb
->s_fs_info
= opts
->new_root
;
1493 opts
->new_root
->sb
= sb
;
1495 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1496 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1497 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1498 sb
->s_op
= &cgroup_ops
;
1503 static int cgroup_get_rootdir(struct super_block
*sb
)
1505 static const struct dentry_operations cgroup_dops
= {
1506 .d_iput
= cgroup_diput
,
1507 .d_delete
= cgroup_delete
,
1510 struct inode
*inode
=
1511 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1516 inode
->i_fop
= &simple_dir_operations
;
1517 inode
->i_op
= &cgroup_dir_inode_operations
;
1518 /* directories start off with i_nlink == 2 (for "." entry) */
1520 sb
->s_root
= d_make_root(inode
);
1523 /* for everything else we want ->d_op set */
1524 sb
->s_d_op
= &cgroup_dops
;
1528 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1529 int flags
, const char *unused_dev_name
,
1532 struct cgroup_sb_opts opts
;
1533 struct cgroupfs_root
*root
;
1535 struct super_block
*sb
;
1536 struct cgroupfs_root
*new_root
;
1537 struct list_head tmp_links
;
1538 struct inode
*inode
;
1539 const struct cred
*cred
;
1541 /* First find the desired set of subsystems */
1542 mutex_lock(&cgroup_mutex
);
1543 ret
= parse_cgroupfs_options(data
, &opts
);
1544 mutex_unlock(&cgroup_mutex
);
1549 * Allocate a new cgroup root. We may not need it if we're
1550 * reusing an existing hierarchy.
1552 new_root
= cgroup_root_from_opts(&opts
);
1553 if (IS_ERR(new_root
)) {
1554 ret
= PTR_ERR(new_root
);
1557 opts
.new_root
= new_root
;
1559 /* Locate an existing or new sb for this hierarchy */
1560 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, 0, &opts
);
1563 cgroup_free_root(opts
.new_root
);
1567 root
= sb
->s_fs_info
;
1569 if (root
== opts
.new_root
) {
1570 /* We used the new root structure, so this is a new hierarchy */
1571 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1572 struct cgroupfs_root
*existing_root
;
1574 struct css_set
*cset
;
1576 BUG_ON(sb
->s_root
!= NULL
);
1578 ret
= cgroup_get_rootdir(sb
);
1580 goto drop_new_super
;
1581 inode
= sb
->s_root
->d_inode
;
1583 mutex_lock(&inode
->i_mutex
);
1584 mutex_lock(&cgroup_mutex
);
1585 mutex_lock(&cgroup_root_mutex
);
1587 root_cgrp
->id
= idr_alloc(&root
->cgroup_idr
, root_cgrp
,
1589 if (root_cgrp
->id
< 0)
1592 /* Check for name clashes with existing mounts */
1594 if (strlen(root
->name
))
1595 for_each_active_root(existing_root
)
1596 if (!strcmp(existing_root
->name
, root
->name
))
1600 * We're accessing css_set_count without locking
1601 * css_set_lock here, but that's OK - it can only be
1602 * increased by someone holding cgroup_lock, and
1603 * that's us. The worst that can happen is that we
1604 * have some link structures left over
1606 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1610 /* ID 0 is reserved for dummy root, 1 for unified hierarchy */
1611 ret
= cgroup_init_root_id(root
, 2, 0);
1615 sb
->s_root
->d_fsdata
= root_cgrp
;
1616 root_cgrp
->dentry
= sb
->s_root
;
1619 * We're inside get_sb() and will call lookup_one_len() to
1620 * create the root files, which doesn't work if SELinux is
1621 * in use. The following cred dancing somehow works around
1622 * it. See 2ce9738ba ("cgroupfs: use init_cred when
1623 * populating new cgroupfs mount") for more details.
1625 cred
= override_creds(&init_cred
);
1627 ret
= cgroup_addrm_files(root_cgrp
, cgroup_base_files
, true);
1631 ret
= rebind_subsystems(root
, root
->subsys_mask
, 0);
1638 * There must be no failure case after here, since rebinding
1639 * takes care of subsystems' refcounts, which are explicitly
1640 * dropped in the failure exit path.
1643 list_add(&root
->root_list
, &cgroup_roots
);
1644 cgroup_root_count
++;
1646 /* Link the top cgroup in this hierarchy into all
1647 * the css_set objects */
1648 write_lock(&css_set_lock
);
1649 hash_for_each(css_set_table
, i
, cset
, hlist
)
1650 link_css_set(&tmp_links
, cset
, root_cgrp
);
1651 write_unlock(&css_set_lock
);
1653 free_cgrp_cset_links(&tmp_links
);
1655 BUG_ON(!list_empty(&root_cgrp
->children
));
1656 BUG_ON(root
->number_of_cgroups
!= 1);
1658 mutex_unlock(&cgroup_root_mutex
);
1659 mutex_unlock(&cgroup_mutex
);
1660 mutex_unlock(&inode
->i_mutex
);
1663 * We re-used an existing hierarchy - the new root (if
1664 * any) is not needed
1666 cgroup_free_root(opts
.new_root
);
1668 if ((root
->flags
^ opts
.flags
) & CGRP_ROOT_OPTION_MASK
) {
1669 if ((root
->flags
| opts
.flags
) & CGRP_ROOT_SANE_BEHAVIOR
) {
1670 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1672 goto drop_new_super
;
1674 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1679 kfree(opts
.release_agent
);
1681 return dget(sb
->s_root
);
1684 free_cgrp_cset_links(&tmp_links
);
1685 cgroup_addrm_files(&root
->top_cgroup
, cgroup_base_files
, false);
1688 cgroup_exit_root_id(root
);
1689 mutex_unlock(&cgroup_root_mutex
);
1690 mutex_unlock(&cgroup_mutex
);
1691 mutex_unlock(&inode
->i_mutex
);
1693 deactivate_locked_super(sb
);
1695 kfree(opts
.release_agent
);
1697 return ERR_PTR(ret
);
1700 static void cgroup_kill_sb(struct super_block
*sb
) {
1701 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1702 struct cgroup
*cgrp
= &root
->top_cgroup
;
1703 struct cgrp_cset_link
*link
, *tmp_link
;
1708 BUG_ON(root
->number_of_cgroups
!= 1);
1709 BUG_ON(!list_empty(&cgrp
->children
));
1711 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1712 mutex_lock(&cgroup_mutex
);
1713 mutex_lock(&cgroup_root_mutex
);
1715 /* Rebind all subsystems back to the default hierarchy */
1716 if (root
->flags
& CGRP_ROOT_SUBSYS_BOUND
) {
1717 ret
= rebind_subsystems(root
, 0, root
->subsys_mask
);
1718 /* Shouldn't be able to fail ... */
1723 * Release all the links from cset_links to this hierarchy's
1726 write_lock(&css_set_lock
);
1728 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1729 list_del(&link
->cset_link
);
1730 list_del(&link
->cgrp_link
);
1733 write_unlock(&css_set_lock
);
1735 if (!list_empty(&root
->root_list
)) {
1736 list_del(&root
->root_list
);
1737 cgroup_root_count
--;
1740 cgroup_exit_root_id(root
);
1742 mutex_unlock(&cgroup_root_mutex
);
1743 mutex_unlock(&cgroup_mutex
);
1744 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1746 simple_xattrs_free(&cgrp
->xattrs
);
1748 kill_litter_super(sb
);
1749 cgroup_free_root(root
);
1752 static struct file_system_type cgroup_fs_type
= {
1754 .mount
= cgroup_mount
,
1755 .kill_sb
= cgroup_kill_sb
,
1758 static struct kobject
*cgroup_kobj
;
1761 * cgroup_path - generate the path of a cgroup
1762 * @cgrp: the cgroup in question
1763 * @buf: the buffer to write the path into
1764 * @buflen: the length of the buffer
1766 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1768 * We can't generate cgroup path using dentry->d_name, as accessing
1769 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1770 * inode's i_mutex, while on the other hand cgroup_path() can be called
1771 * with some irq-safe spinlocks held.
1773 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1775 int ret
= -ENAMETOOLONG
;
1778 if (!cgrp
->parent
) {
1779 if (strlcpy(buf
, "/", buflen
) >= buflen
)
1780 return -ENAMETOOLONG
;
1784 start
= buf
+ buflen
- 1;
1789 const char *name
= cgroup_name(cgrp
);
1793 if ((start
-= len
) < buf
)
1795 memcpy(start
, name
, len
);
1801 cgrp
= cgrp
->parent
;
1802 } while (cgrp
->parent
);
1804 memmove(buf
, start
, buf
+ buflen
- start
);
1809 EXPORT_SYMBOL_GPL(cgroup_path
);
1812 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1813 * @task: target task
1814 * @buf: the buffer to write the path into
1815 * @buflen: the length of the buffer
1817 * Determine @task's cgroup on the first (the one with the lowest non-zero
1818 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1819 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1820 * cgroup controller callbacks.
1822 * Returns 0 on success, fails with -%ENAMETOOLONG if @buflen is too short.
1824 int task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
1826 struct cgroupfs_root
*root
;
1827 struct cgroup
*cgrp
;
1828 int hierarchy_id
= 1, ret
= 0;
1831 return -ENAMETOOLONG
;
1833 mutex_lock(&cgroup_mutex
);
1835 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
1838 cgrp
= task_cgroup_from_root(task
, root
);
1839 ret
= cgroup_path(cgrp
, buf
, buflen
);
1841 /* if no hierarchy exists, everyone is in "/" */
1842 memcpy(buf
, "/", 2);
1845 mutex_unlock(&cgroup_mutex
);
1848 EXPORT_SYMBOL_GPL(task_cgroup_path
);
1851 * Control Group taskset
1853 struct task_and_cgroup
{
1854 struct task_struct
*task
;
1855 struct cgroup
*cgrp
;
1856 struct css_set
*cset
;
1859 struct cgroup_taskset
{
1860 struct task_and_cgroup single
;
1861 struct flex_array
*tc_array
;
1864 struct cgroup
*cur_cgrp
;
1868 * cgroup_taskset_first - reset taskset and return the first task
1869 * @tset: taskset of interest
1871 * @tset iteration is initialized and the first task is returned.
1873 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1875 if (tset
->tc_array
) {
1877 return cgroup_taskset_next(tset
);
1879 tset
->cur_cgrp
= tset
->single
.cgrp
;
1880 return tset
->single
.task
;
1883 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1886 * cgroup_taskset_next - iterate to the next task in taskset
1887 * @tset: taskset of interest
1889 * Return the next task in @tset. Iteration must have been initialized
1890 * with cgroup_taskset_first().
1892 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1894 struct task_and_cgroup
*tc
;
1896 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1899 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1900 tset
->cur_cgrp
= tc
->cgrp
;
1903 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1906 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1907 * @tset: taskset of interest
1909 * Return the cgroup for the current (last returned) task of @tset. This
1910 * function must be preceded by either cgroup_taskset_first() or
1911 * cgroup_taskset_next().
1913 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1915 return tset
->cur_cgrp
;
1917 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1920 * cgroup_taskset_size - return the number of tasks in taskset
1921 * @tset: taskset of interest
1923 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1925 return tset
->tc_array
? tset
->tc_array_len
: 1;
1927 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1931 * cgroup_task_migrate - move a task from one cgroup to another.
1933 * Must be called with cgroup_mutex and threadgroup locked.
1935 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
1936 struct task_struct
*tsk
,
1937 struct css_set
*new_cset
)
1939 struct css_set
*old_cset
;
1942 * We are synchronized through threadgroup_lock() against PF_EXITING
1943 * setting such that we can't race against cgroup_exit() changing the
1944 * css_set to init_css_set and dropping the old one.
1946 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1947 old_cset
= task_css_set(tsk
);
1950 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
1953 /* Update the css_set linked lists if we're using them */
1954 write_lock(&css_set_lock
);
1955 if (!list_empty(&tsk
->cg_list
))
1956 list_move(&tsk
->cg_list
, &new_cset
->tasks
);
1957 write_unlock(&css_set_lock
);
1960 * We just gained a reference on old_cset by taking it from the
1961 * task. As trading it for new_cset is protected by cgroup_mutex,
1962 * we're safe to drop it here; it will be freed under RCU.
1964 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
1965 put_css_set(old_cset
);
1969 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1970 * @cgrp: the cgroup to attach to
1971 * @tsk: the task or the leader of the threadgroup to be attached
1972 * @threadgroup: attach the whole threadgroup?
1974 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1975 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1977 static int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
,
1980 int retval
, i
, group_size
;
1981 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1982 struct cgroupfs_root
*root
= cgrp
->root
;
1983 /* threadgroup list cursor and array */
1984 struct task_struct
*leader
= tsk
;
1985 struct task_and_cgroup
*tc
;
1986 struct flex_array
*group
;
1987 struct cgroup_taskset tset
= { };
1990 * step 0: in order to do expensive, possibly blocking operations for
1991 * every thread, we cannot iterate the thread group list, since it needs
1992 * rcu or tasklist locked. instead, build an array of all threads in the
1993 * group - group_rwsem prevents new threads from appearing, and if
1994 * threads exit, this will just be an over-estimate.
1997 group_size
= get_nr_threads(tsk
);
2000 /* flex_array supports very large thread-groups better than kmalloc. */
2001 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2004 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2005 retval
= flex_array_prealloc(group
, 0, group_size
, GFP_KERNEL
);
2007 goto out_free_group_list
;
2011 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2012 * already PF_EXITING could be freed from underneath us unless we
2013 * take an rcu_read_lock.
2017 struct task_and_cgroup ent
;
2019 /* @tsk either already exited or can't exit until the end */
2020 if (tsk
->flags
& PF_EXITING
)
2023 /* as per above, nr_threads may decrease, but not increase. */
2024 BUG_ON(i
>= group_size
);
2026 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2027 /* nothing to do if this task is already in the cgroup */
2028 if (ent
.cgrp
== cgrp
)
2031 * saying GFP_ATOMIC has no effect here because we did prealloc
2032 * earlier, but it's good form to communicate our expectations.
2034 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2035 BUG_ON(retval
!= 0);
2040 } while_each_thread(leader
, tsk
);
2042 /* remember the number of threads in the array for later. */
2044 tset
.tc_array
= group
;
2045 tset
.tc_array_len
= group_size
;
2047 /* methods shouldn't be called if no task is actually migrating */
2050 goto out_free_group_list
;
2053 * step 1: check that we can legitimately attach to the cgroup.
2055 for_each_root_subsys(root
, ss
) {
2056 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
2058 if (ss
->can_attach
) {
2059 retval
= ss
->can_attach(css
, &tset
);
2062 goto out_cancel_attach
;
2068 * step 2: make sure css_sets exist for all threads to be migrated.
2069 * we use find_css_set, which allocates a new one if necessary.
2071 for (i
= 0; i
< group_size
; i
++) {
2072 struct css_set
*old_cset
;
2074 tc
= flex_array_get(group
, i
);
2075 old_cset
= task_css_set(tc
->task
);
2076 tc
->cset
= find_css_set(old_cset
, cgrp
);
2079 goto out_put_css_set_refs
;
2084 * step 3: now that we're guaranteed success wrt the css_sets,
2085 * proceed to move all tasks to the new cgroup. There are no
2086 * failure cases after here, so this is the commit point.
2088 for (i
= 0; i
< group_size
; i
++) {
2089 tc
= flex_array_get(group
, i
);
2090 cgroup_task_migrate(tc
->cgrp
, tc
->task
, tc
->cset
);
2092 /* nothing is sensitive to fork() after this point. */
2095 * step 4: do subsystem attach callbacks.
2097 for_each_root_subsys(root
, ss
) {
2098 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
2101 ss
->attach(css
, &tset
);
2105 * step 5: success! and cleanup
2108 out_put_css_set_refs
:
2110 for (i
= 0; i
< group_size
; i
++) {
2111 tc
= flex_array_get(group
, i
);
2114 put_css_set(tc
->cset
);
2119 for_each_root_subsys(root
, ss
) {
2120 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
2122 if (ss
== failed_ss
)
2124 if (ss
->cancel_attach
)
2125 ss
->cancel_attach(css
, &tset
);
2128 out_free_group_list
:
2129 flex_array_free(group
);
2134 * Find the task_struct of the task to attach by vpid and pass it along to the
2135 * function to attach either it or all tasks in its threadgroup. Will lock
2136 * cgroup_mutex and threadgroup; may take task_lock of task.
2138 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2140 struct task_struct
*tsk
;
2141 const struct cred
*cred
= current_cred(), *tcred
;
2144 if (!cgroup_lock_live_group(cgrp
))
2150 tsk
= find_task_by_vpid(pid
);
2154 goto out_unlock_cgroup
;
2157 * even if we're attaching all tasks in the thread group, we
2158 * only need to check permissions on one of them.
2160 tcred
= __task_cred(tsk
);
2161 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2162 !uid_eq(cred
->euid
, tcred
->uid
) &&
2163 !uid_eq(cred
->euid
, tcred
->suid
)) {
2166 goto out_unlock_cgroup
;
2172 tsk
= tsk
->group_leader
;
2175 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2176 * trapped in a cpuset, or RT worker may be born in a cgroup
2177 * with no rt_runtime allocated. Just say no.
2179 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2182 goto out_unlock_cgroup
;
2185 get_task_struct(tsk
);
2188 threadgroup_lock(tsk
);
2190 if (!thread_group_leader(tsk
)) {
2192 * a race with de_thread from another thread's exec()
2193 * may strip us of our leadership, if this happens,
2194 * there is no choice but to throw this task away and
2195 * try again; this is
2196 * "double-double-toil-and-trouble-check locking".
2198 threadgroup_unlock(tsk
);
2199 put_task_struct(tsk
);
2200 goto retry_find_task
;
2204 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2206 threadgroup_unlock(tsk
);
2208 put_task_struct(tsk
);
2210 mutex_unlock(&cgroup_mutex
);
2215 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2216 * @from: attach to all cgroups of a given task
2217 * @tsk: the task to be attached
2219 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2221 struct cgroupfs_root
*root
;
2224 mutex_lock(&cgroup_mutex
);
2225 for_each_active_root(root
) {
2226 struct cgroup
*from_cgrp
= task_cgroup_from_root(from
, root
);
2228 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2232 mutex_unlock(&cgroup_mutex
);
2236 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2238 static int cgroup_tasks_write(struct cgroup_subsys_state
*css
,
2239 struct cftype
*cft
, u64 pid
)
2241 return attach_task_by_pid(css
->cgroup
, pid
, false);
2244 static int cgroup_procs_write(struct cgroup_subsys_state
*css
,
2245 struct cftype
*cft
, u64 tgid
)
2247 return attach_task_by_pid(css
->cgroup
, tgid
, true);
2250 static int cgroup_release_agent_write(struct cgroup_subsys_state
*css
,
2251 struct cftype
*cft
, const char *buffer
)
2253 BUILD_BUG_ON(sizeof(css
->cgroup
->root
->release_agent_path
) < PATH_MAX
);
2254 if (strlen(buffer
) >= PATH_MAX
)
2256 if (!cgroup_lock_live_group(css
->cgroup
))
2258 mutex_lock(&cgroup_root_mutex
);
2259 strcpy(css
->cgroup
->root
->release_agent_path
, buffer
);
2260 mutex_unlock(&cgroup_root_mutex
);
2261 mutex_unlock(&cgroup_mutex
);
2265 static int cgroup_release_agent_show(struct cgroup_subsys_state
*css
,
2266 struct cftype
*cft
, struct seq_file
*seq
)
2268 struct cgroup
*cgrp
= css
->cgroup
;
2270 if (!cgroup_lock_live_group(cgrp
))
2272 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2273 seq_putc(seq
, '\n');
2274 mutex_unlock(&cgroup_mutex
);
2278 static int cgroup_sane_behavior_show(struct cgroup_subsys_state
*css
,
2279 struct cftype
*cft
, struct seq_file
*seq
)
2281 seq_printf(seq
, "%d\n", cgroup_sane_behavior(css
->cgroup
));
2285 /* return the css for the given cgroup file */
2286 static struct cgroup_subsys_state
*cgroup_file_css(struct cfent
*cfe
)
2288 struct cftype
*cft
= cfe
->type
;
2289 struct cgroup
*cgrp
= __d_cgrp(cfe
->dentry
->d_parent
);
2292 return cgrp
->subsys
[cft
->ss
->subsys_id
];
2293 return &cgrp
->dummy_css
;
2296 /* A buffer size big enough for numbers or short strings */
2297 #define CGROUP_LOCAL_BUFFER_SIZE 64
2299 static ssize_t
cgroup_write_X64(struct cgroup_subsys_state
*css
,
2300 struct cftype
*cft
, struct file
*file
,
2301 const char __user
*userbuf
, size_t nbytes
,
2302 loff_t
*unused_ppos
)
2304 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2310 if (nbytes
>= sizeof(buffer
))
2312 if (copy_from_user(buffer
, userbuf
, nbytes
))
2315 buffer
[nbytes
] = 0; /* nul-terminate */
2316 if (cft
->write_u64
) {
2317 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2320 retval
= cft
->write_u64(css
, cft
, val
);
2322 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2325 retval
= cft
->write_s64(css
, cft
, val
);
2332 static ssize_t
cgroup_write_string(struct cgroup_subsys_state
*css
,
2333 struct cftype
*cft
, struct file
*file
,
2334 const char __user
*userbuf
, size_t nbytes
,
2335 loff_t
*unused_ppos
)
2337 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2339 size_t max_bytes
= cft
->max_write_len
;
2340 char *buffer
= local_buffer
;
2343 max_bytes
= sizeof(local_buffer
) - 1;
2344 if (nbytes
>= max_bytes
)
2346 /* Allocate a dynamic buffer if we need one */
2347 if (nbytes
>= sizeof(local_buffer
)) {
2348 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2352 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2357 buffer
[nbytes
] = 0; /* nul-terminate */
2358 retval
= cft
->write_string(css
, cft
, strstrip(buffer
));
2362 if (buffer
!= local_buffer
)
2367 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2368 size_t nbytes
, loff_t
*ppos
)
2370 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2371 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2372 struct cgroup_subsys_state
*css
= cgroup_file_css(cfe
);
2375 return cft
->write(css
, cft
, file
, buf
, nbytes
, ppos
);
2376 if (cft
->write_u64
|| cft
->write_s64
)
2377 return cgroup_write_X64(css
, cft
, file
, buf
, nbytes
, ppos
);
2378 if (cft
->write_string
)
2379 return cgroup_write_string(css
, cft
, file
, buf
, nbytes
, ppos
);
2381 int ret
= cft
->trigger(css
, (unsigned int)cft
->private);
2382 return ret
? ret
: nbytes
;
2387 static ssize_t
cgroup_read_u64(struct cgroup_subsys_state
*css
,
2388 struct cftype
*cft
, struct file
*file
,
2389 char __user
*buf
, size_t nbytes
, loff_t
*ppos
)
2391 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2392 u64 val
= cft
->read_u64(css
, cft
);
2393 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2395 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2398 static ssize_t
cgroup_read_s64(struct cgroup_subsys_state
*css
,
2399 struct cftype
*cft
, struct file
*file
,
2400 char __user
*buf
, size_t nbytes
, loff_t
*ppos
)
2402 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2403 s64 val
= cft
->read_s64(css
, cft
);
2404 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2406 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2409 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2410 size_t nbytes
, loff_t
*ppos
)
2412 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2413 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2414 struct cgroup_subsys_state
*css
= cgroup_file_css(cfe
);
2417 return cft
->read(css
, cft
, file
, buf
, nbytes
, ppos
);
2419 return cgroup_read_u64(css
, cft
, file
, buf
, nbytes
, ppos
);
2421 return cgroup_read_s64(css
, cft
, file
, buf
, nbytes
, ppos
);
2426 * seqfile ops/methods for returning structured data. Currently just
2427 * supports string->u64 maps, but can be extended in future.
2430 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2432 struct seq_file
*sf
= cb
->state
;
2433 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2436 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2438 struct cfent
*cfe
= m
->private;
2439 struct cftype
*cft
= cfe
->type
;
2440 struct cgroup_subsys_state
*css
= cgroup_file_css(cfe
);
2442 if (cft
->read_map
) {
2443 struct cgroup_map_cb cb
= {
2444 .fill
= cgroup_map_add
,
2447 return cft
->read_map(css
, cft
, &cb
);
2449 return cft
->read_seq_string(css
, cft
, m
);
2452 static const struct file_operations cgroup_seqfile_operations
= {
2454 .write
= cgroup_file_write
,
2455 .llseek
= seq_lseek
,
2456 .release
= single_release
,
2459 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2461 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2462 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2463 struct cgroup_subsys_state
*css
= cgroup_file_css(cfe
);
2466 err
= generic_file_open(inode
, file
);
2471 * If the file belongs to a subsystem, pin the css. Will be
2472 * unpinned either on open failure or release. This ensures that
2473 * @css stays alive for all file operations.
2475 if (css
->ss
&& !css_tryget(css
))
2478 if (cft
->read_map
|| cft
->read_seq_string
) {
2479 file
->f_op
= &cgroup_seqfile_operations
;
2480 err
= single_open(file
, cgroup_seqfile_show
, cfe
);
2481 } else if (cft
->open
) {
2482 err
= cft
->open(inode
, file
);
2490 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2492 struct cfent
*cfe
= __d_cfe(file
->f_dentry
);
2493 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2494 struct cgroup_subsys_state
*css
= cgroup_file_css(cfe
);
2498 ret
= cft
->release(inode
, file
);
2505 * cgroup_rename - Only allow simple rename of directories in place.
2507 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2508 struct inode
*new_dir
, struct dentry
*new_dentry
)
2511 struct cgroup_name
*name
, *old_name
;
2512 struct cgroup
*cgrp
;
2515 * It's convinient to use parent dir's i_mutex to protected
2518 lockdep_assert_held(&old_dir
->i_mutex
);
2520 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2522 if (new_dentry
->d_inode
)
2524 if (old_dir
!= new_dir
)
2527 cgrp
= __d_cgrp(old_dentry
);
2530 * This isn't a proper migration and its usefulness is very
2531 * limited. Disallow if sane_behavior.
2533 if (cgroup_sane_behavior(cgrp
))
2536 name
= cgroup_alloc_name(new_dentry
);
2540 ret
= simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2546 old_name
= rcu_dereference_protected(cgrp
->name
, true);
2547 rcu_assign_pointer(cgrp
->name
, name
);
2549 kfree_rcu(old_name
, rcu_head
);
2553 static struct simple_xattrs
*__d_xattrs(struct dentry
*dentry
)
2555 if (S_ISDIR(dentry
->d_inode
->i_mode
))
2556 return &__d_cgrp(dentry
)->xattrs
;
2558 return &__d_cfe(dentry
)->xattrs
;
2561 static inline int xattr_enabled(struct dentry
*dentry
)
2563 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
2564 return root
->flags
& CGRP_ROOT_XATTR
;
2567 static bool is_valid_xattr(const char *name
)
2569 if (!strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) ||
2570 !strncmp(name
, XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
))
2575 static int cgroup_setxattr(struct dentry
*dentry
, const char *name
,
2576 const void *val
, size_t size
, int flags
)
2578 if (!xattr_enabled(dentry
))
2580 if (!is_valid_xattr(name
))
2582 return simple_xattr_set(__d_xattrs(dentry
), name
, val
, size
, flags
);
2585 static int cgroup_removexattr(struct dentry
*dentry
, const char *name
)
2587 if (!xattr_enabled(dentry
))
2589 if (!is_valid_xattr(name
))
2591 return simple_xattr_remove(__d_xattrs(dentry
), name
);
2594 static ssize_t
cgroup_getxattr(struct dentry
*dentry
, const char *name
,
2595 void *buf
, size_t size
)
2597 if (!xattr_enabled(dentry
))
2599 if (!is_valid_xattr(name
))
2601 return simple_xattr_get(__d_xattrs(dentry
), name
, buf
, size
);
2604 static ssize_t
cgroup_listxattr(struct dentry
*dentry
, char *buf
, size_t size
)
2606 if (!xattr_enabled(dentry
))
2608 return simple_xattr_list(__d_xattrs(dentry
), buf
, size
);
2611 static const struct file_operations cgroup_file_operations
= {
2612 .read
= cgroup_file_read
,
2613 .write
= cgroup_file_write
,
2614 .llseek
= generic_file_llseek
,
2615 .open
= cgroup_file_open
,
2616 .release
= cgroup_file_release
,
2619 static const struct inode_operations cgroup_file_inode_operations
= {
2620 .setxattr
= cgroup_setxattr
,
2621 .getxattr
= cgroup_getxattr
,
2622 .listxattr
= cgroup_listxattr
,
2623 .removexattr
= cgroup_removexattr
,
2626 static const struct inode_operations cgroup_dir_inode_operations
= {
2627 .lookup
= cgroup_lookup
,
2628 .mkdir
= cgroup_mkdir
,
2629 .rmdir
= cgroup_rmdir
,
2630 .rename
= cgroup_rename
,
2631 .setxattr
= cgroup_setxattr
,
2632 .getxattr
= cgroup_getxattr
,
2633 .listxattr
= cgroup_listxattr
,
2634 .removexattr
= cgroup_removexattr
,
2637 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
2639 if (dentry
->d_name
.len
> NAME_MAX
)
2640 return ERR_PTR(-ENAMETOOLONG
);
2641 d_add(dentry
, NULL
);
2646 * Check if a file is a control file
2648 static inline struct cftype
*__file_cft(struct file
*file
)
2650 if (file_inode(file
)->i_fop
!= &cgroup_file_operations
)
2651 return ERR_PTR(-EINVAL
);
2652 return __d_cft(file
->f_dentry
);
2655 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2656 struct super_block
*sb
)
2658 struct inode
*inode
;
2662 if (dentry
->d_inode
)
2665 inode
= cgroup_new_inode(mode
, sb
);
2669 if (S_ISDIR(mode
)) {
2670 inode
->i_op
= &cgroup_dir_inode_operations
;
2671 inode
->i_fop
= &simple_dir_operations
;
2673 /* start off with i_nlink == 2 (for "." entry) */
2675 inc_nlink(dentry
->d_parent
->d_inode
);
2678 * Control reaches here with cgroup_mutex held.
2679 * @inode->i_mutex should nest outside cgroup_mutex but we
2680 * want to populate it immediately without releasing
2681 * cgroup_mutex. As @inode isn't visible to anyone else
2682 * yet, trylock will always succeed without affecting
2685 WARN_ON_ONCE(!mutex_trylock(&inode
->i_mutex
));
2686 } else if (S_ISREG(mode
)) {
2688 inode
->i_fop
= &cgroup_file_operations
;
2689 inode
->i_op
= &cgroup_file_inode_operations
;
2691 d_instantiate(dentry
, inode
);
2692 dget(dentry
); /* Extra count - pin the dentry in core */
2697 * cgroup_file_mode - deduce file mode of a control file
2698 * @cft: the control file in question
2700 * returns cft->mode if ->mode is not 0
2701 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2702 * returns S_IRUGO if it has only a read handler
2703 * returns S_IWUSR if it has only a write hander
2705 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2712 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2713 cft
->read_map
|| cft
->read_seq_string
)
2716 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2717 cft
->write_string
|| cft
->trigger
)
2723 static int cgroup_add_file(struct cgroup
*cgrp
, struct cftype
*cft
)
2725 struct dentry
*dir
= cgrp
->dentry
;
2726 struct cgroup
*parent
= __d_cgrp(dir
);
2727 struct dentry
*dentry
;
2731 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2733 if (cft
->ss
&& !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
2734 strcpy(name
, cft
->ss
->name
);
2737 strcat(name
, cft
->name
);
2739 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2741 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2745 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2746 if (IS_ERR(dentry
)) {
2747 error
= PTR_ERR(dentry
);
2751 cfe
->type
= (void *)cft
;
2752 cfe
->dentry
= dentry
;
2753 dentry
->d_fsdata
= cfe
;
2754 simple_xattrs_init(&cfe
->xattrs
);
2756 mode
= cgroup_file_mode(cft
);
2757 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2759 list_add_tail(&cfe
->node
, &parent
->files
);
2769 * cgroup_addrm_files - add or remove files to a cgroup directory
2770 * @cgrp: the target cgroup
2771 * @cfts: array of cftypes to be added
2772 * @is_add: whether to add or remove
2774 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2775 * For removals, this function never fails. If addition fails, this
2776 * function doesn't remove files already added. The caller is responsible
2779 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
2785 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
2786 lockdep_assert_held(&cgroup_mutex
);
2788 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2789 /* does cft->flags tell us to skip this file on @cgrp? */
2790 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_sane_behavior(cgrp
))
2792 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2794 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2798 ret
= cgroup_add_file(cgrp
, cft
);
2800 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2805 cgroup_rm_file(cgrp
, cft
);
2811 static void cgroup_cfts_prepare(void)
2812 __acquires(&cgroup_mutex
)
2815 * Thanks to the entanglement with vfs inode locking, we can't walk
2816 * the existing cgroups under cgroup_mutex and create files.
2817 * Instead, we use css_for_each_descendant_pre() and drop RCU read
2818 * lock before calling cgroup_addrm_files().
2820 mutex_lock(&cgroup_mutex
);
2823 static int cgroup_cfts_commit(struct cftype
*cfts
, bool is_add
)
2824 __releases(&cgroup_mutex
)
2827 struct cgroup_subsys
*ss
= cfts
[0].ss
;
2828 struct cgroup
*root
= &ss
->root
->top_cgroup
;
2829 struct super_block
*sb
= ss
->root
->sb
;
2830 struct dentry
*prev
= NULL
;
2831 struct inode
*inode
;
2832 struct cgroup_subsys_state
*css
;
2836 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2837 if (!cfts
|| ss
->root
== &cgroup_dummy_root
||
2838 !atomic_inc_not_zero(&sb
->s_active
)) {
2839 mutex_unlock(&cgroup_mutex
);
2844 * All cgroups which are created after we drop cgroup_mutex will
2845 * have the updated set of files, so we only need to update the
2846 * cgroups created before the current @cgroup_serial_nr_next.
2848 update_before
= cgroup_serial_nr_next
;
2850 mutex_unlock(&cgroup_mutex
);
2852 /* @root always needs to be updated */
2853 inode
= root
->dentry
->d_inode
;
2854 mutex_lock(&inode
->i_mutex
);
2855 mutex_lock(&cgroup_mutex
);
2856 ret
= cgroup_addrm_files(root
, cfts
, is_add
);
2857 mutex_unlock(&cgroup_mutex
);
2858 mutex_unlock(&inode
->i_mutex
);
2863 /* add/rm files for all cgroups created before */
2865 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
->subsys_id
)) {
2866 struct cgroup
*cgrp
= css
->cgroup
;
2868 if (cgroup_is_dead(cgrp
))
2871 inode
= cgrp
->dentry
->d_inode
;
2876 prev
= cgrp
->dentry
;
2878 mutex_lock(&inode
->i_mutex
);
2879 mutex_lock(&cgroup_mutex
);
2880 if (cgrp
->serial_nr
< update_before
&& !cgroup_is_dead(cgrp
))
2881 ret
= cgroup_addrm_files(cgrp
, cfts
, is_add
);
2882 mutex_unlock(&cgroup_mutex
);
2883 mutex_unlock(&inode
->i_mutex
);
2892 deactivate_super(sb
);
2897 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2898 * @ss: target cgroup subsystem
2899 * @cfts: zero-length name terminated array of cftypes
2901 * Register @cfts to @ss. Files described by @cfts are created for all
2902 * existing cgroups to which @ss is attached and all future cgroups will
2903 * have them too. This function can be called anytime whether @ss is
2906 * Returns 0 on successful registration, -errno on failure. Note that this
2907 * function currently returns 0 as long as @cfts registration is successful
2908 * even if some file creation attempts on existing cgroups fail.
2910 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2912 struct cftype_set
*set
;
2916 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2920 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++)
2923 cgroup_cfts_prepare();
2925 list_add_tail(&set
->node
, &ss
->cftsets
);
2926 ret
= cgroup_cfts_commit(cfts
, true);
2928 cgroup_rm_cftypes(cfts
);
2931 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2934 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2935 * @cfts: zero-length name terminated array of cftypes
2937 * Unregister @cfts. Files described by @cfts are removed from all
2938 * existing cgroups and all future cgroups won't have them either. This
2939 * function can be called anytime whether @cfts' subsys is attached or not.
2941 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2944 int cgroup_rm_cftypes(struct cftype
*cfts
)
2946 struct cftype_set
*set
;
2948 if (!cfts
|| !cfts
[0].ss
)
2951 cgroup_cfts_prepare();
2953 list_for_each_entry(set
, &cfts
[0].ss
->cftsets
, node
) {
2954 if (set
->cfts
== cfts
) {
2955 list_del(&set
->node
);
2957 cgroup_cfts_commit(cfts
, false);
2962 cgroup_cfts_commit(NULL
, false);
2967 * cgroup_task_count - count the number of tasks in a cgroup.
2968 * @cgrp: the cgroup in question
2970 * Return the number of tasks in the cgroup.
2972 int cgroup_task_count(const struct cgroup
*cgrp
)
2975 struct cgrp_cset_link
*link
;
2977 read_lock(&css_set_lock
);
2978 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
2979 count
+= atomic_read(&link
->cset
->refcount
);
2980 read_unlock(&css_set_lock
);
2985 * Advance a list_head iterator. The iterator should be positioned at
2986 * the start of a css_set
2988 static void cgroup_advance_iter(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2990 struct list_head
*l
= it
->cset_link
;
2991 struct cgrp_cset_link
*link
;
2992 struct css_set
*cset
;
2994 /* Advance to the next non-empty css_set */
2997 if (l
== &cgrp
->cset_links
) {
2998 it
->cset_link
= NULL
;
3001 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3003 } while (list_empty(&cset
->tasks
));
3005 it
->task
= cset
->tasks
.next
;
3009 * To reduce the fork() overhead for systems that are not actually
3010 * using their cgroups capability, we don't maintain the lists running
3011 * through each css_set to its tasks until we see the list actually
3012 * used - in other words after the first call to cgroup_iter_start().
3014 static void cgroup_enable_task_cg_lists(void)
3016 struct task_struct
*p
, *g
;
3017 write_lock(&css_set_lock
);
3018 use_task_css_set_links
= 1;
3020 * We need tasklist_lock because RCU is not safe against
3021 * while_each_thread(). Besides, a forking task that has passed
3022 * cgroup_post_fork() without seeing use_task_css_set_links = 1
3023 * is not guaranteed to have its child immediately visible in the
3024 * tasklist if we walk through it with RCU.
3026 read_lock(&tasklist_lock
);
3027 do_each_thread(g
, p
) {
3030 * We should check if the process is exiting, otherwise
3031 * it will race with cgroup_exit() in that the list
3032 * entry won't be deleted though the process has exited.
3034 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
3035 list_add(&p
->cg_list
, &task_css_set(p
)->tasks
);
3037 } while_each_thread(g
, p
);
3038 read_unlock(&tasklist_lock
);
3039 write_unlock(&css_set_lock
);
3043 * css_next_child - find the next child of a given css
3044 * @pos_css: the current position (%NULL to initiate traversal)
3045 * @parent_css: css whose children to walk
3047 * This function returns the next child of @parent_css and should be called
3048 * under RCU read lock. The only requirement is that @parent_css and
3049 * @pos_css are accessible. The next sibling is guaranteed to be returned
3050 * regardless of their states.
3052 struct cgroup_subsys_state
*
3053 css_next_child(struct cgroup_subsys_state
*pos_css
,
3054 struct cgroup_subsys_state
*parent_css
)
3056 struct cgroup
*pos
= pos_css
? pos_css
->cgroup
: NULL
;
3057 struct cgroup
*cgrp
= parent_css
->cgroup
;
3058 struct cgroup
*next
;
3060 WARN_ON_ONCE(!rcu_read_lock_held());
3063 * @pos could already have been removed. Once a cgroup is removed,
3064 * its ->sibling.next is no longer updated when its next sibling
3065 * changes. As CGRP_DEAD assertion is serialized and happens
3066 * before the cgroup is taken off the ->sibling list, if we see it
3067 * unasserted, it's guaranteed that the next sibling hasn't
3068 * finished its grace period even if it's already removed, and thus
3069 * safe to dereference from this RCU critical section. If
3070 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
3071 * to be visible as %true here.
3073 * If @pos is dead, its next pointer can't be dereferenced;
3074 * however, as each cgroup is given a monotonically increasing
3075 * unique serial number and always appended to the sibling list,
3076 * the next one can be found by walking the parent's children until
3077 * we see a cgroup with higher serial number than @pos's. While
3078 * this path can be slower, it's taken only when either the current
3079 * cgroup is removed or iteration and removal race.
3082 next
= list_entry_rcu(cgrp
->children
.next
, struct cgroup
, sibling
);
3083 } else if (likely(!cgroup_is_dead(pos
))) {
3084 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
, sibling
);
3086 list_for_each_entry_rcu(next
, &cgrp
->children
, sibling
)
3087 if (next
->serial_nr
> pos
->serial_nr
)
3091 if (&next
->sibling
== &cgrp
->children
)
3095 return cgroup_css(next
, parent_css
->ss
->subsys_id
);
3097 return &next
->dummy_css
;
3099 EXPORT_SYMBOL_GPL(css_next_child
);
3102 * css_next_descendant_pre - find the next descendant for pre-order walk
3103 * @pos: the current position (%NULL to initiate traversal)
3104 * @root: css whose descendants to walk
3106 * To be used by css_for_each_descendant_pre(). Find the next descendant
3107 * to visit for pre-order traversal of @root's descendants.
3109 * While this function requires RCU read locking, it doesn't require the
3110 * whole traversal to be contained in a single RCU critical section. This
3111 * function will return the correct next descendant as long as both @pos
3112 * and @root are accessible and @pos is a descendant of @root.
3114 struct cgroup_subsys_state
*
3115 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3116 struct cgroup_subsys_state
*root
)
3118 struct cgroup_subsys_state
*next
;
3120 WARN_ON_ONCE(!rcu_read_lock_held());
3122 /* if first iteration, pretend we just visited @root */
3126 /* visit the first child if exists */
3127 next
= css_next_child(NULL
, pos
);
3131 /* no child, visit my or the closest ancestor's next sibling */
3132 while (pos
!= root
) {
3133 next
= css_next_child(pos
, css_parent(pos
));
3136 pos
= css_parent(pos
);
3141 EXPORT_SYMBOL_GPL(css_next_descendant_pre
);
3144 * css_rightmost_descendant - return the rightmost descendant of a css
3145 * @pos: css of interest
3147 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3148 * is returned. This can be used during pre-order traversal to skip
3151 * While this function requires RCU read locking, it doesn't require the
3152 * whole traversal to be contained in a single RCU critical section. This
3153 * function will return the correct rightmost descendant as long as @pos is
3156 struct cgroup_subsys_state
*
3157 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3159 struct cgroup_subsys_state
*last
, *tmp
;
3161 WARN_ON_ONCE(!rcu_read_lock_held());
3165 /* ->prev isn't RCU safe, walk ->next till the end */
3167 css_for_each_child(tmp
, last
)
3173 EXPORT_SYMBOL_GPL(css_rightmost_descendant
);
3175 static struct cgroup_subsys_state
*
3176 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3178 struct cgroup_subsys_state
*last
;
3182 pos
= css_next_child(NULL
, pos
);
3189 * css_next_descendant_post - find the next descendant for post-order walk
3190 * @pos: the current position (%NULL to initiate traversal)
3191 * @root: css whose descendants to walk
3193 * To be used by css_for_each_descendant_post(). Find the next descendant
3194 * to visit for post-order traversal of @root's descendants.
3196 * While this function requires RCU read locking, it doesn't require the
3197 * whole traversal to be contained in a single RCU critical section. This
3198 * function will return the correct next descendant as long as both @pos
3199 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3201 struct cgroup_subsys_state
*
3202 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3203 struct cgroup_subsys_state
*root
)
3205 struct cgroup_subsys_state
*next
;
3207 WARN_ON_ONCE(!rcu_read_lock_held());
3209 /* if first iteration, visit the leftmost descendant */
3211 next
= css_leftmost_descendant(root
);
3212 return next
!= root
? next
: NULL
;
3215 /* if there's an unvisited sibling, visit its leftmost descendant */
3216 next
= css_next_child(pos
, css_parent(pos
));
3218 return css_leftmost_descendant(next
);
3220 /* no sibling left, visit parent */
3221 next
= css_parent(pos
);
3222 return next
!= root
? next
: NULL
;
3224 EXPORT_SYMBOL_GPL(css_next_descendant_post
);
3226 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3227 __acquires(css_set_lock
)
3230 * The first time anyone tries to iterate across a cgroup,
3231 * we need to enable the list linking each css_set to its
3232 * tasks, and fix up all existing tasks.
3234 if (!use_task_css_set_links
)
3235 cgroup_enable_task_cg_lists();
3237 read_lock(&css_set_lock
);
3238 it
->cset_link
= &cgrp
->cset_links
;
3239 cgroup_advance_iter(cgrp
, it
);
3242 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
3243 struct cgroup_iter
*it
)
3245 struct task_struct
*res
;
3246 struct list_head
*l
= it
->task
;
3247 struct cgrp_cset_link
*link
;
3249 /* If the iterator cg is NULL, we have no tasks */
3252 res
= list_entry(l
, struct task_struct
, cg_list
);
3253 /* Advance iterator to find next entry */
3255 link
= list_entry(it
->cset_link
, struct cgrp_cset_link
, cset_link
);
3256 if (l
== &link
->cset
->tasks
) {
3257 /* We reached the end of this task list - move on to
3258 * the next cg_cgroup_link */
3259 cgroup_advance_iter(cgrp
, it
);
3266 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3267 __releases(css_set_lock
)
3269 read_unlock(&css_set_lock
);
3272 static inline int started_after_time(struct task_struct
*t1
,
3273 struct timespec
*time
,
3274 struct task_struct
*t2
)
3276 int start_diff
= timespec_compare(&t1
->start_time
, time
);
3277 if (start_diff
> 0) {
3279 } else if (start_diff
< 0) {
3283 * Arbitrarily, if two processes started at the same
3284 * time, we'll say that the lower pointer value
3285 * started first. Note that t2 may have exited by now
3286 * so this may not be a valid pointer any longer, but
3287 * that's fine - it still serves to distinguish
3288 * between two tasks started (effectively) simultaneously.
3295 * This function is a callback from heap_insert() and is used to order
3297 * In this case we order the heap in descending task start time.
3299 static inline int started_after(void *p1
, void *p2
)
3301 struct task_struct
*t1
= p1
;
3302 struct task_struct
*t2
= p2
;
3303 return started_after_time(t1
, &t2
->start_time
, t2
);
3307 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3308 * @scan: struct cgroup_scanner containing arguments for the scan
3310 * Arguments include pointers to callback functions test_task() and
3312 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3313 * and if it returns true, call process_task() for it also.
3314 * The test_task pointer may be NULL, meaning always true (select all tasks).
3315 * Effectively duplicates cgroup_iter_{start,next,end}()
3316 * but does not lock css_set_lock for the call to process_task().
3317 * The struct cgroup_scanner may be embedded in any structure of the caller's
3319 * It is guaranteed that process_task() will act on every task that
3320 * is a member of the cgroup for the duration of this call. This
3321 * function may or may not call process_task() for tasks that exit
3322 * or move to a different cgroup during the call, or are forked or
3323 * move into the cgroup during the call.
3325 * Note that test_task() may be called with locks held, and may in some
3326 * situations be called multiple times for the same task, so it should
3328 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3329 * pre-allocated and will be used for heap operations (and its "gt" member will
3330 * be overwritten), else a temporary heap will be used (allocation of which
3331 * may cause this function to fail).
3333 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
3336 struct cgroup_iter it
;
3337 struct task_struct
*p
, *dropped
;
3338 /* Never dereference latest_task, since it's not refcounted */
3339 struct task_struct
*latest_task
= NULL
;
3340 struct ptr_heap tmp_heap
;
3341 struct ptr_heap
*heap
;
3342 struct timespec latest_time
= { 0, 0 };
3345 /* The caller supplied our heap and pre-allocated its memory */
3347 heap
->gt
= &started_after
;
3349 /* We need to allocate our own heap memory */
3351 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3353 /* cannot allocate the heap */
3359 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3360 * to determine which are of interest, and using the scanner's
3361 * "process_task" callback to process any of them that need an update.
3362 * Since we don't want to hold any locks during the task updates,
3363 * gather tasks to be processed in a heap structure.
3364 * The heap is sorted by descending task start time.
3365 * If the statically-sized heap fills up, we overflow tasks that
3366 * started later, and in future iterations only consider tasks that
3367 * started after the latest task in the previous pass. This
3368 * guarantees forward progress and that we don't miss any tasks.
3371 cgroup_iter_start(scan
->cgrp
, &it
);
3372 while ((p
= cgroup_iter_next(scan
->cgrp
, &it
))) {
3374 * Only affect tasks that qualify per the caller's callback,
3375 * if he provided one
3377 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
3380 * Only process tasks that started after the last task
3383 if (!started_after_time(p
, &latest_time
, latest_task
))
3385 dropped
= heap_insert(heap
, p
);
3386 if (dropped
== NULL
) {
3388 * The new task was inserted; the heap wasn't
3392 } else if (dropped
!= p
) {
3394 * The new task was inserted, and pushed out a
3398 put_task_struct(dropped
);
3401 * Else the new task was newer than anything already in
3402 * the heap and wasn't inserted
3405 cgroup_iter_end(scan
->cgrp
, &it
);
3408 for (i
= 0; i
< heap
->size
; i
++) {
3409 struct task_struct
*q
= heap
->ptrs
[i
];
3411 latest_time
= q
->start_time
;
3414 /* Process the task per the caller's callback */
3415 scan
->process_task(q
, scan
);
3419 * If we had to process any tasks at all, scan again
3420 * in case some of them were in the middle of forking
3421 * children that didn't get processed.
3422 * Not the most efficient way to do it, but it avoids
3423 * having to take callback_mutex in the fork path
3427 if (heap
== &tmp_heap
)
3428 heap_free(&tmp_heap
);
3432 static void cgroup_transfer_one_task(struct task_struct
*task
,
3433 struct cgroup_scanner
*scan
)
3435 struct cgroup
*new_cgroup
= scan
->data
;
3437 mutex_lock(&cgroup_mutex
);
3438 cgroup_attach_task(new_cgroup
, task
, false);
3439 mutex_unlock(&cgroup_mutex
);
3443 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3444 * @to: cgroup to which the tasks will be moved
3445 * @from: cgroup in which the tasks currently reside
3447 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3449 struct cgroup_scanner scan
;
3452 scan
.test_task
= NULL
; /* select all tasks in cgroup */
3453 scan
.process_task
= cgroup_transfer_one_task
;
3457 return cgroup_scan_tasks(&scan
);
3461 * Stuff for reading the 'tasks'/'procs' files.
3463 * Reading this file can return large amounts of data if a cgroup has
3464 * *lots* of attached tasks. So it may need several calls to read(),
3465 * but we cannot guarantee that the information we produce is correct
3466 * unless we produce it entirely atomically.
3470 /* which pidlist file are we talking about? */
3471 enum cgroup_filetype
{
3477 * A pidlist is a list of pids that virtually represents the contents of one
3478 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3479 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3482 struct cgroup_pidlist
{
3484 * used to find which pidlist is wanted. doesn't change as long as
3485 * this particular list stays in the list.
3487 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3490 /* how many elements the above list has */
3492 /* how many files are using the current array */
3494 /* each of these stored in a list by its cgroup */
3495 struct list_head links
;
3496 /* pointer to the cgroup we belong to, for list removal purposes */
3497 struct cgroup
*owner
;
3498 /* protects the other fields */
3499 struct rw_semaphore rwsem
;
3503 * The following two functions "fix" the issue where there are more pids
3504 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3505 * TODO: replace with a kernel-wide solution to this problem
3507 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3508 static void *pidlist_allocate(int count
)
3510 if (PIDLIST_TOO_LARGE(count
))
3511 return vmalloc(count
* sizeof(pid_t
));
3513 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3515 static void pidlist_free(void *p
)
3517 if (is_vmalloc_addr(p
))
3524 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3525 * Returns the number of unique elements.
3527 static int pidlist_uniq(pid_t
*list
, int length
)
3532 * we presume the 0th element is unique, so i starts at 1. trivial
3533 * edge cases first; no work needs to be done for either
3535 if (length
== 0 || length
== 1)
3537 /* src and dest walk down the list; dest counts unique elements */
3538 for (src
= 1; src
< length
; src
++) {
3539 /* find next unique element */
3540 while (list
[src
] == list
[src
-1]) {
3545 /* dest always points to where the next unique element goes */
3546 list
[dest
] = list
[src
];
3553 static int cmppid(const void *a
, const void *b
)
3555 return *(pid_t
*)a
- *(pid_t
*)b
;
3559 * find the appropriate pidlist for our purpose (given procs vs tasks)
3560 * returns with the lock on that pidlist already held, and takes care
3561 * of the use count, or returns NULL with no locks held if we're out of
3564 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3565 enum cgroup_filetype type
)
3567 struct cgroup_pidlist
*l
;
3568 /* don't need task_nsproxy() if we're looking at ourself */
3569 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3572 * We can't drop the pidlist_mutex before taking the l->rwsem in case
3573 * the last ref-holder is trying to remove l from the list at the same
3574 * time. Holding the pidlist_mutex precludes somebody taking whichever
3575 * list we find out from under us - compare release_pid_array().
3577 mutex_lock(&cgrp
->pidlist_mutex
);
3578 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3579 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3580 /* make sure l doesn't vanish out from under us */
3581 down_write(&l
->rwsem
);
3582 mutex_unlock(&cgrp
->pidlist_mutex
);
3586 /* entry not found; create a new one */
3587 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3589 mutex_unlock(&cgrp
->pidlist_mutex
);
3592 init_rwsem(&l
->rwsem
);
3593 down_write(&l
->rwsem
);
3595 l
->key
.ns
= get_pid_ns(ns
);
3597 list_add(&l
->links
, &cgrp
->pidlists
);
3598 mutex_unlock(&cgrp
->pidlist_mutex
);
3603 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3605 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3606 struct cgroup_pidlist
**lp
)
3610 int pid
, n
= 0; /* used for populating the array */
3611 struct cgroup_iter it
;
3612 struct task_struct
*tsk
;
3613 struct cgroup_pidlist
*l
;
3616 * If cgroup gets more users after we read count, we won't have
3617 * enough space - tough. This race is indistinguishable to the
3618 * caller from the case that the additional cgroup users didn't
3619 * show up until sometime later on.
3621 length
= cgroup_task_count(cgrp
);
3622 array
= pidlist_allocate(length
);
3625 /* now, populate the array */
3626 cgroup_iter_start(cgrp
, &it
);
3627 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3628 if (unlikely(n
== length
))
3630 /* get tgid or pid for procs or tasks file respectively */
3631 if (type
== CGROUP_FILE_PROCS
)
3632 pid
= task_tgid_vnr(tsk
);
3634 pid
= task_pid_vnr(tsk
);
3635 if (pid
> 0) /* make sure to only use valid results */
3638 cgroup_iter_end(cgrp
, &it
);
3640 /* now sort & (if procs) strip out duplicates */
3641 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3642 if (type
== CGROUP_FILE_PROCS
)
3643 length
= pidlist_uniq(array
, length
);
3644 l
= cgroup_pidlist_find(cgrp
, type
);
3646 pidlist_free(array
);
3649 /* store array, freeing old if necessary - lock already held */
3650 pidlist_free(l
->list
);
3654 up_write(&l
->rwsem
);
3660 * cgroupstats_build - build and fill cgroupstats
3661 * @stats: cgroupstats to fill information into
3662 * @dentry: A dentry entry belonging to the cgroup for which stats have
3665 * Build and fill cgroupstats so that taskstats can export it to user
3668 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3671 struct cgroup
*cgrp
;
3672 struct cgroup_iter it
;
3673 struct task_struct
*tsk
;
3676 * Validate dentry by checking the superblock operations,
3677 * and make sure it's a directory.
3679 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3680 !S_ISDIR(dentry
->d_inode
->i_mode
))
3684 cgrp
= dentry
->d_fsdata
;
3686 cgroup_iter_start(cgrp
, &it
);
3687 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3688 switch (tsk
->state
) {
3690 stats
->nr_running
++;
3692 case TASK_INTERRUPTIBLE
:
3693 stats
->nr_sleeping
++;
3695 case TASK_UNINTERRUPTIBLE
:
3696 stats
->nr_uninterruptible
++;
3699 stats
->nr_stopped
++;
3702 if (delayacct_is_task_waiting_on_io(tsk
))
3703 stats
->nr_io_wait
++;
3707 cgroup_iter_end(cgrp
, &it
);
3715 * seq_file methods for the tasks/procs files. The seq_file position is the
3716 * next pid to display; the seq_file iterator is a pointer to the pid
3717 * in the cgroup->l->list array.
3720 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3723 * Initially we receive a position value that corresponds to
3724 * one more than the last pid shown (or 0 on the first call or
3725 * after a seek to the start). Use a binary-search to find the
3726 * next pid to display, if any
3728 struct cgroup_pidlist
*l
= s
->private;
3729 int index
= 0, pid
= *pos
;
3732 down_read(&l
->rwsem
);
3734 int end
= l
->length
;
3736 while (index
< end
) {
3737 int mid
= (index
+ end
) / 2;
3738 if (l
->list
[mid
] == pid
) {
3741 } else if (l
->list
[mid
] <= pid
)
3747 /* If we're off the end of the array, we're done */
3748 if (index
>= l
->length
)
3750 /* Update the abstract position to be the actual pid that we found */
3751 iter
= l
->list
+ index
;
3756 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3758 struct cgroup_pidlist
*l
= s
->private;
3762 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3764 struct cgroup_pidlist
*l
= s
->private;
3766 pid_t
*end
= l
->list
+ l
->length
;
3768 * Advance to the next pid in the array. If this goes off the
3780 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3782 return seq_printf(s
, "%d\n", *(int *)v
);
3786 * seq_operations functions for iterating on pidlists through seq_file -
3787 * independent of whether it's tasks or procs
3789 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3790 .start
= cgroup_pidlist_start
,
3791 .stop
= cgroup_pidlist_stop
,
3792 .next
= cgroup_pidlist_next
,
3793 .show
= cgroup_pidlist_show
,
3796 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3799 * the case where we're the last user of this particular pidlist will
3800 * have us remove it from the cgroup's list, which entails taking the
3801 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3802 * pidlist_mutex, we have to take pidlist_mutex first.
3804 mutex_lock(&l
->owner
->pidlist_mutex
);
3805 down_write(&l
->rwsem
);
3806 BUG_ON(!l
->use_count
);
3807 if (!--l
->use_count
) {
3808 /* we're the last user if refcount is 0; remove and free */
3809 list_del(&l
->links
);
3810 mutex_unlock(&l
->owner
->pidlist_mutex
);
3811 pidlist_free(l
->list
);
3812 put_pid_ns(l
->key
.ns
);
3813 up_write(&l
->rwsem
);
3817 mutex_unlock(&l
->owner
->pidlist_mutex
);
3818 up_write(&l
->rwsem
);
3821 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3823 struct cgroup_pidlist
*l
;
3824 if (!(file
->f_mode
& FMODE_READ
))
3827 * the seq_file will only be initialized if the file was opened for
3828 * reading; hence we check if it's not null only in that case.
3830 l
= ((struct seq_file
*)file
->private_data
)->private;
3831 cgroup_release_pid_array(l
);
3832 return seq_release(inode
, file
);
3835 static const struct file_operations cgroup_pidlist_operations
= {
3837 .llseek
= seq_lseek
,
3838 .write
= cgroup_file_write
,
3839 .release
= cgroup_pidlist_release
,
3843 * The following functions handle opens on a file that displays a pidlist
3844 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3847 /* helper function for the two below it */
3848 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3850 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3851 struct cgroup_pidlist
*l
;
3854 /* Nothing to do for write-only files */
3855 if (!(file
->f_mode
& FMODE_READ
))
3858 /* have the array populated */
3859 retval
= pidlist_array_load(cgrp
, type
, &l
);
3862 /* configure file information */
3863 file
->f_op
= &cgroup_pidlist_operations
;
3865 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3867 cgroup_release_pid_array(l
);
3870 ((struct seq_file
*)file
->private_data
)->private = l
;
3873 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3875 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3877 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3879 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3882 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
3885 return notify_on_release(css
->cgroup
);
3888 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
3889 struct cftype
*cft
, u64 val
)
3891 clear_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
3893 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
3895 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
3900 * When dput() is called asynchronously, if umount has been done and
3901 * then deactivate_super() in cgroup_free_fn() kills the superblock,
3902 * there's a small window that vfs will see the root dentry with non-zero
3903 * refcnt and trigger BUG().
3905 * That's why we hold a reference before dput() and drop it right after.
3907 static void cgroup_dput(struct cgroup
*cgrp
)
3909 struct super_block
*sb
= cgrp
->root
->sb
;
3911 atomic_inc(&sb
->s_active
);
3913 deactivate_super(sb
);
3917 * Unregister event and free resources.
3919 * Gets called from workqueue.
3921 static void cgroup_event_remove(struct work_struct
*work
)
3923 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3925 struct cgroup
*cgrp
= event
->cgrp
;
3927 remove_wait_queue(event
->wqh
, &event
->wait
);
3929 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3931 /* Notify userspace the event is going away. */
3932 eventfd_signal(event
->eventfd
, 1);
3934 eventfd_ctx_put(event
->eventfd
);
3940 * Gets called on POLLHUP on eventfd when user closes it.
3942 * Called with wqh->lock held and interrupts disabled.
3944 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3945 int sync
, void *key
)
3947 struct cgroup_event
*event
= container_of(wait
,
3948 struct cgroup_event
, wait
);
3949 struct cgroup
*cgrp
= event
->cgrp
;
3950 unsigned long flags
= (unsigned long)key
;
3952 if (flags
& POLLHUP
) {
3954 * If the event has been detached at cgroup removal, we
3955 * can simply return knowing the other side will cleanup
3958 * We can't race against event freeing since the other
3959 * side will require wqh->lock via remove_wait_queue(),
3962 spin_lock(&cgrp
->event_list_lock
);
3963 if (!list_empty(&event
->list
)) {
3964 list_del_init(&event
->list
);
3966 * We are in atomic context, but cgroup_event_remove()
3967 * may sleep, so we have to call it in workqueue.
3969 schedule_work(&event
->remove
);
3971 spin_unlock(&cgrp
->event_list_lock
);
3977 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3978 wait_queue_head_t
*wqh
, poll_table
*pt
)
3980 struct cgroup_event
*event
= container_of(pt
,
3981 struct cgroup_event
, pt
);
3984 add_wait_queue(wqh
, &event
->wait
);
3988 * Parse input and register new cgroup event handler.
3990 * Input must be in format '<event_fd> <control_fd> <args>'.
3991 * Interpretation of args is defined by control file implementation.
3993 static int cgroup_write_event_control(struct cgroup_subsys_state
*css
,
3994 struct cftype
*cft
, const char *buffer
)
3996 struct cgroup
*cgrp
= css
->cgroup
;
3997 struct cgroup_event
*event
;
3998 struct cgroup
*cgrp_cfile
;
3999 unsigned int efd
, cfd
;
4005 efd
= simple_strtoul(buffer
, &endp
, 10);
4010 cfd
= simple_strtoul(buffer
, &endp
, 10);
4011 if ((*endp
!= ' ') && (*endp
!= '\0'))
4015 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
4019 INIT_LIST_HEAD(&event
->list
);
4020 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
4021 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
4022 INIT_WORK(&event
->remove
, cgroup_event_remove
);
4024 efile
= eventfd_fget(efd
);
4025 if (IS_ERR(efile
)) {
4026 ret
= PTR_ERR(efile
);
4030 event
->eventfd
= eventfd_ctx_fileget(efile
);
4031 if (IS_ERR(event
->eventfd
)) {
4032 ret
= PTR_ERR(event
->eventfd
);
4039 goto out_put_eventfd
;
4042 /* the process need read permission on control file */
4043 /* AV: shouldn't we check that it's been opened for read instead? */
4044 ret
= inode_permission(file_inode(cfile
), MAY_READ
);
4048 event
->cft
= __file_cft(cfile
);
4049 if (IS_ERR(event
->cft
)) {
4050 ret
= PTR_ERR(event
->cft
);
4055 * The file to be monitored must be in the same cgroup as
4056 * cgroup.event_control is.
4058 cgrp_cfile
= __d_cgrp(cfile
->f_dentry
->d_parent
);
4059 if (cgrp_cfile
!= cgrp
) {
4064 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
4069 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
4070 event
->eventfd
, buffer
);
4074 efile
->f_op
->poll(efile
, &event
->pt
);
4077 * Events should be removed after rmdir of cgroup directory, but before
4078 * destroying subsystem state objects. Let's take reference to cgroup
4079 * directory dentry to do that.
4083 spin_lock(&cgrp
->event_list_lock
);
4084 list_add(&event
->list
, &cgrp
->event_list
);
4085 spin_unlock(&cgrp
->event_list_lock
);
4095 eventfd_ctx_put(event
->eventfd
);
4104 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
4107 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4110 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
4111 struct cftype
*cft
, u64 val
)
4114 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4116 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4120 static struct cftype cgroup_base_files
[] = {
4122 .name
= "cgroup.procs",
4123 .open
= cgroup_procs_open
,
4124 .write_u64
= cgroup_procs_write
,
4125 .release
= cgroup_pidlist_release
,
4126 .mode
= S_IRUGO
| S_IWUSR
,
4129 .name
= "cgroup.event_control",
4130 .write_string
= cgroup_write_event_control
,
4134 .name
= "cgroup.clone_children",
4135 .flags
= CFTYPE_INSANE
,
4136 .read_u64
= cgroup_clone_children_read
,
4137 .write_u64
= cgroup_clone_children_write
,
4140 .name
= "cgroup.sane_behavior",
4141 .flags
= CFTYPE_ONLY_ON_ROOT
,
4142 .read_seq_string
= cgroup_sane_behavior_show
,
4146 * Historical crazy stuff. These don't have "cgroup." prefix and
4147 * don't exist if sane_behavior. If you're depending on these, be
4148 * prepared to be burned.
4152 .flags
= CFTYPE_INSANE
, /* use "procs" instead */
4153 .open
= cgroup_tasks_open
,
4154 .write_u64
= cgroup_tasks_write
,
4155 .release
= cgroup_pidlist_release
,
4156 .mode
= S_IRUGO
| S_IWUSR
,
4159 .name
= "notify_on_release",
4160 .flags
= CFTYPE_INSANE
,
4161 .read_u64
= cgroup_read_notify_on_release
,
4162 .write_u64
= cgroup_write_notify_on_release
,
4165 .name
= "release_agent",
4166 .flags
= CFTYPE_INSANE
| CFTYPE_ONLY_ON_ROOT
,
4167 .read_seq_string
= cgroup_release_agent_show
,
4168 .write_string
= cgroup_release_agent_write
,
4169 .max_write_len
= PATH_MAX
,
4175 * cgroup_populate_dir - create subsys files in a cgroup directory
4176 * @cgrp: target cgroup
4177 * @subsys_mask: mask of the subsystem ids whose files should be added
4179 * On failure, no file is added.
4181 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
4183 struct cgroup_subsys
*ss
;
4186 /* process cftsets of each subsystem */
4187 for_each_subsys(ss
, i
) {
4188 struct cftype_set
*set
;
4190 if (!test_bit(i
, &subsys_mask
))
4193 list_for_each_entry(set
, &ss
->cftsets
, node
) {
4194 ret
= cgroup_addrm_files(cgrp
, set
->cfts
, true);
4200 /* This cgroup is ready now */
4201 for_each_root_subsys(cgrp
->root
, ss
) {
4202 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4203 struct css_id
*id
= rcu_dereference_protected(css
->id
, true);
4206 * Update id->css pointer and make this css visible from
4207 * CSS ID functions. This pointer will be dereferened
4208 * from RCU-read-side without locks.
4211 rcu_assign_pointer(id
->css
, css
);
4216 cgroup_clear_dir(cgrp
, subsys_mask
);
4220 static void css_dput_fn(struct work_struct
*work
)
4222 struct cgroup_subsys_state
*css
=
4223 container_of(work
, struct cgroup_subsys_state
, dput_work
);
4225 cgroup_dput(css
->cgroup
);
4228 static void css_release(struct percpu_ref
*ref
)
4230 struct cgroup_subsys_state
*css
=
4231 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4233 schedule_work(&css
->dput_work
);
4236 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
4237 struct cgroup_subsys
*ss
,
4238 struct cgroup
*cgrp
)
4244 if (cgrp
== cgroup_dummy_top
)
4245 css
->flags
|= CSS_ROOT
;
4246 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
4247 cgrp
->subsys
[ss
->subsys_id
] = css
;
4250 * css holds an extra ref to @cgrp->dentry which is put on the last
4251 * css_put(). dput() requires process context, which css_put() may
4252 * be called without. @css->dput_work will be used to invoke
4253 * dput() asynchronously from css_put().
4255 INIT_WORK(&css
->dput_work
, css_dput_fn
);
4258 /* invoke ->css_online() on a new CSS and mark it online if successful */
4259 static int online_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4261 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4264 lockdep_assert_held(&cgroup_mutex
);
4267 ret
= ss
->css_online(css
);
4269 css
->flags
|= CSS_ONLINE
;
4273 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4274 static void offline_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4276 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4278 lockdep_assert_held(&cgroup_mutex
);
4280 if (!(css
->flags
& CSS_ONLINE
))
4283 if (ss
->css_offline
)
4284 ss
->css_offline(css
);
4286 css
->flags
&= ~CSS_ONLINE
;
4290 * cgroup_create - create a cgroup
4291 * @parent: cgroup that will be parent of the new cgroup
4292 * @dentry: dentry of the new cgroup
4293 * @mode: mode to set on new inode
4295 * Must be called with the mutex on the parent inode held
4297 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
4300 struct cgroup
*cgrp
;
4301 struct cgroup_name
*name
;
4302 struct cgroupfs_root
*root
= parent
->root
;
4304 struct cgroup_subsys
*ss
;
4305 struct super_block
*sb
= root
->sb
;
4307 /* allocate the cgroup and its ID, 0 is reserved for the root */
4308 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4312 name
= cgroup_alloc_name(dentry
);
4315 rcu_assign_pointer(cgrp
->name
, name
);
4318 * Temporarily set the pointer to NULL, so idr_find() won't return
4319 * a half-baked cgroup.
4321 cgrp
->id
= idr_alloc(&root
->cgroup_idr
, NULL
, 1, 0, GFP_KERNEL
);
4326 * Only live parents can have children. Note that the liveliness
4327 * check isn't strictly necessary because cgroup_mkdir() and
4328 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4329 * anyway so that locking is contained inside cgroup proper and we
4330 * don't get nasty surprises if we ever grow another caller.
4332 if (!cgroup_lock_live_group(parent
)) {
4337 /* Grab a reference on the superblock so the hierarchy doesn't
4338 * get deleted on unmount if there are child cgroups. This
4339 * can be done outside cgroup_mutex, since the sb can't
4340 * disappear while someone has an open control file on the
4342 atomic_inc(&sb
->s_active
);
4344 init_cgroup_housekeeping(cgrp
);
4346 dentry
->d_fsdata
= cgrp
;
4347 cgrp
->dentry
= dentry
;
4349 cgrp
->parent
= parent
;
4350 cgrp
->root
= parent
->root
;
4352 if (notify_on_release(parent
))
4353 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4355 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4356 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4358 for_each_root_subsys(root
, ss
) {
4359 struct cgroup_subsys_state
*css
;
4361 css
= ss
->css_alloc(parent
->subsys
[ss
->subsys_id
]);
4367 err
= percpu_ref_init(&css
->refcnt
, css_release
);
4373 init_cgroup_css(css
, ss
, cgrp
);
4376 err
= alloc_css_id(ss
, parent
, cgrp
);
4383 * Create directory. cgroup_create_file() returns with the new
4384 * directory locked on success so that it can be populated without
4385 * dropping cgroup_mutex.
4387 err
= cgroup_create_file(dentry
, S_IFDIR
| mode
, sb
);
4390 lockdep_assert_held(&dentry
->d_inode
->i_mutex
);
4392 cgrp
->serial_nr
= cgroup_serial_nr_next
++;
4394 /* allocation complete, commit to creation */
4395 list_add_tail_rcu(&cgrp
->sibling
, &cgrp
->parent
->children
);
4396 root
->number_of_cgroups
++;
4398 /* each css holds a ref to the cgroup's dentry */
4399 for_each_root_subsys(root
, ss
)
4402 /* hold a ref to the parent's dentry */
4403 dget(parent
->dentry
);
4405 /* creation succeeded, notify subsystems */
4406 for_each_root_subsys(root
, ss
) {
4407 err
= online_css(ss
, cgrp
);
4411 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4413 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",
4414 current
->comm
, current
->pid
, ss
->name
);
4415 if (!strcmp(ss
->name
, "memory"))
4416 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4417 ss
->warned_broken_hierarchy
= true;
4421 idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4423 err
= cgroup_addrm_files(cgrp
, cgroup_base_files
, true);
4427 err
= cgroup_populate_dir(cgrp
, root
->subsys_mask
);
4431 mutex_unlock(&cgroup_mutex
);
4432 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4437 for_each_root_subsys(root
, ss
) {
4438 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4441 percpu_ref_cancel_init(&css
->refcnt
);
4445 mutex_unlock(&cgroup_mutex
);
4446 /* Release the reference count that we took on the superblock */
4447 deactivate_super(sb
);
4449 idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4451 kfree(rcu_dereference_raw(cgrp
->name
));
4457 cgroup_destroy_locked(cgrp
);
4458 mutex_unlock(&cgroup_mutex
);
4459 mutex_unlock(&dentry
->d_inode
->i_mutex
);
4463 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4465 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4467 /* the vfs holds inode->i_mutex already */
4468 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4471 static void cgroup_css_killed(struct cgroup
*cgrp
)
4473 if (!atomic_dec_and_test(&cgrp
->css_kill_cnt
))
4476 /* percpu ref's of all css's are killed, kick off the next step */
4477 INIT_WORK(&cgrp
->destroy_work
, cgroup_offline_fn
);
4478 schedule_work(&cgrp
->destroy_work
);
4481 static void css_ref_killed_fn(struct percpu_ref
*ref
)
4483 struct cgroup_subsys_state
*css
=
4484 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4486 cgroup_css_killed(css
->cgroup
);
4490 * cgroup_destroy_locked - the first stage of cgroup destruction
4491 * @cgrp: cgroup to be destroyed
4493 * css's make use of percpu refcnts whose killing latency shouldn't be
4494 * exposed to userland and are RCU protected. Also, cgroup core needs to
4495 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
4496 * invoked. To satisfy all the requirements, destruction is implemented in
4497 * the following two steps.
4499 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4500 * userland visible parts and start killing the percpu refcnts of
4501 * css's. Set up so that the next stage will be kicked off once all
4502 * the percpu refcnts are confirmed to be killed.
4504 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4505 * rest of destruction. Once all cgroup references are gone, the
4506 * cgroup is RCU-freed.
4508 * This function implements s1. After this step, @cgrp is gone as far as
4509 * the userland is concerned and a new cgroup with the same name may be
4510 * created. As cgroup doesn't care about the names internally, this
4511 * doesn't cause any problem.
4513 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4514 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4516 struct dentry
*d
= cgrp
->dentry
;
4517 struct cgroup_event
*event
, *tmp
;
4518 struct cgroup_subsys
*ss
;
4521 lockdep_assert_held(&d
->d_inode
->i_mutex
);
4522 lockdep_assert_held(&cgroup_mutex
);
4525 * css_set_lock synchronizes access to ->cset_links and prevents
4526 * @cgrp from being removed while __put_css_set() is in progress.
4528 read_lock(&css_set_lock
);
4529 empty
= list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
);
4530 read_unlock(&css_set_lock
);
4535 * Block new css_tryget() by killing css refcnts. cgroup core
4536 * guarantees that, by the time ->css_offline() is invoked, no new
4537 * css reference will be given out via css_tryget(). We can't
4538 * simply call percpu_ref_kill() and proceed to offlining css's
4539 * because percpu_ref_kill() doesn't guarantee that the ref is seen
4540 * as killed on all CPUs on return.
4542 * Use percpu_ref_kill_and_confirm() to get notifications as each
4543 * css is confirmed to be seen as killed on all CPUs. The
4544 * notification callback keeps track of the number of css's to be
4545 * killed and schedules cgroup_offline_fn() to perform the rest of
4546 * destruction once the percpu refs of all css's are confirmed to
4549 atomic_set(&cgrp
->css_kill_cnt
, 1);
4550 for_each_root_subsys(cgrp
->root
, ss
) {
4551 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4554 * Killing would put the base ref, but we need to keep it
4555 * alive until after ->css_offline.
4557 percpu_ref_get(&css
->refcnt
);
4559 atomic_inc(&cgrp
->css_kill_cnt
);
4560 percpu_ref_kill_and_confirm(&css
->refcnt
, css_ref_killed_fn
);
4562 cgroup_css_killed(cgrp
);
4565 * Mark @cgrp dead. This prevents further task migration and child
4566 * creation by disabling cgroup_lock_live_group(). Note that
4567 * CGRP_DEAD assertion is depended upon by css_next_child() to
4568 * resume iteration after dropping RCU read lock. See
4569 * css_next_child() for details.
4571 set_bit(CGRP_DEAD
, &cgrp
->flags
);
4573 /* CGRP_DEAD is set, remove from ->release_list for the last time */
4574 raw_spin_lock(&release_list_lock
);
4575 if (!list_empty(&cgrp
->release_list
))
4576 list_del_init(&cgrp
->release_list
);
4577 raw_spin_unlock(&release_list_lock
);
4580 * Clear and remove @cgrp directory. The removal puts the base ref
4581 * but we aren't quite done with @cgrp yet, so hold onto it.
4583 cgroup_clear_dir(cgrp
, cgrp
->root
->subsys_mask
);
4584 cgroup_addrm_files(cgrp
, cgroup_base_files
, false);
4586 cgroup_d_remove_dir(d
);
4589 * Unregister events and notify userspace.
4590 * Notify userspace about cgroup removing only after rmdir of cgroup
4591 * directory to avoid race between userspace and kernelspace.
4593 spin_lock(&cgrp
->event_list_lock
);
4594 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4595 list_del_init(&event
->list
);
4596 schedule_work(&event
->remove
);
4598 spin_unlock(&cgrp
->event_list_lock
);
4604 * cgroup_offline_fn - the second step of cgroup destruction
4605 * @work: cgroup->destroy_free_work
4607 * This function is invoked from a work item for a cgroup which is being
4608 * destroyed after the percpu refcnts of all css's are guaranteed to be
4609 * seen as killed on all CPUs, and performs the rest of destruction. This
4610 * is the second step of destruction described in the comment above
4611 * cgroup_destroy_locked().
4613 static void cgroup_offline_fn(struct work_struct
*work
)
4615 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, destroy_work
);
4616 struct cgroup
*parent
= cgrp
->parent
;
4617 struct dentry
*d
= cgrp
->dentry
;
4618 struct cgroup_subsys
*ss
;
4620 mutex_lock(&cgroup_mutex
);
4623 * css_tryget() is guaranteed to fail now. Tell subsystems to
4624 * initate destruction.
4626 for_each_root_subsys(cgrp
->root
, ss
)
4627 offline_css(ss
, cgrp
);
4630 * Put the css refs from cgroup_destroy_locked(). Each css holds
4631 * an extra reference to the cgroup's dentry and cgroup removal
4632 * proceeds regardless of css refs. On the last put of each css,
4633 * whenever that may be, the extra dentry ref is put so that dentry
4634 * destruction happens only after all css's are released.
4636 for_each_root_subsys(cgrp
->root
, ss
)
4637 css_put(cgrp
->subsys
[ss
->subsys_id
]);
4639 /* delete this cgroup from parent->children */
4640 list_del_rcu(&cgrp
->sibling
);
4643 * We should remove the cgroup object from idr before its grace
4644 * period starts, so we won't be looking up a cgroup while the
4645 * cgroup is being freed.
4647 idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4652 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4653 check_for_release(parent
);
4655 mutex_unlock(&cgroup_mutex
);
4658 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4662 mutex_lock(&cgroup_mutex
);
4663 ret
= cgroup_destroy_locked(dentry
->d_fsdata
);
4664 mutex_unlock(&cgroup_mutex
);
4669 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4671 INIT_LIST_HEAD(&ss
->cftsets
);
4674 * base_cftset is embedded in subsys itself, no need to worry about
4677 if (ss
->base_cftypes
) {
4680 for (cft
= ss
->base_cftypes
; cft
->name
[0] != '\0'; cft
++)
4683 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4684 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4688 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4690 struct cgroup_subsys_state
*css
;
4692 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4694 mutex_lock(&cgroup_mutex
);
4696 /* init base cftset */
4697 cgroup_init_cftsets(ss
);
4699 /* Create the top cgroup state for this subsystem */
4700 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4701 ss
->root
= &cgroup_dummy_root
;
4702 css
= ss
->css_alloc(cgroup_dummy_top
->subsys
[ss
->subsys_id
]);
4703 /* We don't handle early failures gracefully */
4704 BUG_ON(IS_ERR(css
));
4705 init_cgroup_css(css
, ss
, cgroup_dummy_top
);
4707 /* Update the init_css_set to contain a subsys
4708 * pointer to this state - since the subsystem is
4709 * newly registered, all tasks and hence the
4710 * init_css_set is in the subsystem's top cgroup. */
4711 init_css_set
.subsys
[ss
->subsys_id
] = css
;
4713 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4715 /* At system boot, before all subsystems have been
4716 * registered, no tasks have been forked, so we don't
4717 * need to invoke fork callbacks here. */
4718 BUG_ON(!list_empty(&init_task
.tasks
));
4720 BUG_ON(online_css(ss
, cgroup_dummy_top
));
4722 mutex_unlock(&cgroup_mutex
);
4724 /* this function shouldn't be used with modular subsystems, since they
4725 * need to register a subsys_id, among other things */
4730 * cgroup_load_subsys: load and register a modular subsystem at runtime
4731 * @ss: the subsystem to load
4733 * This function should be called in a modular subsystem's initcall. If the
4734 * subsystem is built as a module, it will be assigned a new subsys_id and set
4735 * up for use. If the subsystem is built-in anyway, work is delegated to the
4736 * simpler cgroup_init_subsys.
4738 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4740 struct cgroup_subsys_state
*css
;
4742 struct hlist_node
*tmp
;
4743 struct css_set
*cset
;
4746 /* check name and function validity */
4747 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4748 ss
->css_alloc
== NULL
|| ss
->css_free
== NULL
)
4752 * we don't support callbacks in modular subsystems. this check is
4753 * before the ss->module check for consistency; a subsystem that could
4754 * be a module should still have no callbacks even if the user isn't
4755 * compiling it as one.
4757 if (ss
->fork
|| ss
->exit
)
4761 * an optionally modular subsystem is built-in: we want to do nothing,
4762 * since cgroup_init_subsys will have already taken care of it.
4764 if (ss
->module
== NULL
) {
4765 /* a sanity check */
4766 BUG_ON(cgroup_subsys
[ss
->subsys_id
] != ss
);
4770 /* init base cftset */
4771 cgroup_init_cftsets(ss
);
4773 mutex_lock(&cgroup_mutex
);
4774 cgroup_subsys
[ss
->subsys_id
] = ss
;
4777 * no ss->css_alloc seems to need anything important in the ss
4778 * struct, so this can happen first (i.e. before the dummy root
4781 css
= ss
->css_alloc(cgroup_dummy_top
->subsys
[ss
->subsys_id
]);
4783 /* failure case - need to deassign the cgroup_subsys[] slot. */
4784 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4785 mutex_unlock(&cgroup_mutex
);
4786 return PTR_ERR(css
);
4789 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4790 ss
->root
= &cgroup_dummy_root
;
4792 /* our new subsystem will be attached to the dummy hierarchy. */
4793 init_cgroup_css(css
, ss
, cgroup_dummy_top
);
4794 /* init_idr must be after init_cgroup_css because it sets css->id. */
4796 ret
= cgroup_init_idr(ss
, css
);
4802 * Now we need to entangle the css into the existing css_sets. unlike
4803 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4804 * will need a new pointer to it; done by iterating the css_set_table.
4805 * furthermore, modifying the existing css_sets will corrupt the hash
4806 * table state, so each changed css_set will need its hash recomputed.
4807 * this is all done under the css_set_lock.
4809 write_lock(&css_set_lock
);
4810 hash_for_each_safe(css_set_table
, i
, tmp
, cset
, hlist
) {
4811 /* skip entries that we already rehashed */
4812 if (cset
->subsys
[ss
->subsys_id
])
4814 /* remove existing entry */
4815 hash_del(&cset
->hlist
);
4817 cset
->subsys
[ss
->subsys_id
] = css
;
4818 /* recompute hash and restore entry */
4819 key
= css_set_hash(cset
->subsys
);
4820 hash_add(css_set_table
, &cset
->hlist
, key
);
4822 write_unlock(&css_set_lock
);
4824 ret
= online_css(ss
, cgroup_dummy_top
);
4829 mutex_unlock(&cgroup_mutex
);
4833 mutex_unlock(&cgroup_mutex
);
4834 /* @ss can't be mounted here as try_module_get() would fail */
4835 cgroup_unload_subsys(ss
);
4838 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4841 * cgroup_unload_subsys: unload a modular subsystem
4842 * @ss: the subsystem to unload
4844 * This function should be called in a modular subsystem's exitcall. When this
4845 * function is invoked, the refcount on the subsystem's module will be 0, so
4846 * the subsystem will not be attached to any hierarchy.
4848 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4850 struct cgrp_cset_link
*link
;
4852 BUG_ON(ss
->module
== NULL
);
4855 * we shouldn't be called if the subsystem is in use, and the use of
4856 * try_module_get() in rebind_subsystems() should ensure that it
4857 * doesn't start being used while we're killing it off.
4859 BUG_ON(ss
->root
!= &cgroup_dummy_root
);
4861 mutex_lock(&cgroup_mutex
);
4863 offline_css(ss
, cgroup_dummy_top
);
4866 idr_destroy(&ss
->idr
);
4868 /* deassign the subsys_id */
4869 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4871 /* remove subsystem from the dummy root's list of subsystems */
4872 list_del_init(&ss
->sibling
);
4875 * disentangle the css from all css_sets attached to the dummy
4876 * top. as in loading, we need to pay our respects to the hashtable
4879 write_lock(&css_set_lock
);
4880 list_for_each_entry(link
, &cgroup_dummy_top
->cset_links
, cset_link
) {
4881 struct css_set
*cset
= link
->cset
;
4884 hash_del(&cset
->hlist
);
4885 cset
->subsys
[ss
->subsys_id
] = NULL
;
4886 key
= css_set_hash(cset
->subsys
);
4887 hash_add(css_set_table
, &cset
->hlist
, key
);
4889 write_unlock(&css_set_lock
);
4892 * remove subsystem's css from the cgroup_dummy_top and free it -
4893 * need to free before marking as null because ss->css_free needs
4894 * the cgrp->subsys pointer to find their state. note that this
4895 * also takes care of freeing the css_id.
4897 ss
->css_free(cgroup_dummy_top
->subsys
[ss
->subsys_id
]);
4898 cgroup_dummy_top
->subsys
[ss
->subsys_id
] = NULL
;
4900 mutex_unlock(&cgroup_mutex
);
4902 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4905 * cgroup_init_early - cgroup initialization at system boot
4907 * Initialize cgroups at system boot, and initialize any
4908 * subsystems that request early init.
4910 int __init
cgroup_init_early(void)
4912 struct cgroup_subsys
*ss
;
4915 atomic_set(&init_css_set
.refcount
, 1);
4916 INIT_LIST_HEAD(&init_css_set
.cgrp_links
);
4917 INIT_LIST_HEAD(&init_css_set
.tasks
);
4918 INIT_HLIST_NODE(&init_css_set
.hlist
);
4920 init_cgroup_root(&cgroup_dummy_root
);
4921 cgroup_root_count
= 1;
4922 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
4924 init_cgrp_cset_link
.cset
= &init_css_set
;
4925 init_cgrp_cset_link
.cgrp
= cgroup_dummy_top
;
4926 list_add(&init_cgrp_cset_link
.cset_link
, &cgroup_dummy_top
->cset_links
);
4927 list_add(&init_cgrp_cset_link
.cgrp_link
, &init_css_set
.cgrp_links
);
4929 /* at bootup time, we don't worry about modular subsystems */
4930 for_each_builtin_subsys(ss
, i
) {
4932 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4933 BUG_ON(!ss
->css_alloc
);
4934 BUG_ON(!ss
->css_free
);
4935 if (ss
->subsys_id
!= i
) {
4936 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4937 ss
->name
, ss
->subsys_id
);
4942 cgroup_init_subsys(ss
);
4948 * cgroup_init - cgroup initialization
4950 * Register cgroup filesystem and /proc file, and initialize
4951 * any subsystems that didn't request early init.
4953 int __init
cgroup_init(void)
4955 struct cgroup_subsys
*ss
;
4959 err
= bdi_init(&cgroup_backing_dev_info
);
4963 for_each_builtin_subsys(ss
, i
) {
4964 if (!ss
->early_init
)
4965 cgroup_init_subsys(ss
);
4967 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4970 /* allocate id for the dummy hierarchy */
4971 mutex_lock(&cgroup_mutex
);
4972 mutex_lock(&cgroup_root_mutex
);
4974 /* Add init_css_set to the hash table */
4975 key
= css_set_hash(init_css_set
.subsys
);
4976 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4978 BUG_ON(cgroup_init_root_id(&cgroup_dummy_root
, 0, 1));
4980 err
= idr_alloc(&cgroup_dummy_root
.cgroup_idr
, cgroup_dummy_top
,
4984 mutex_unlock(&cgroup_root_mutex
);
4985 mutex_unlock(&cgroup_mutex
);
4987 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4993 err
= register_filesystem(&cgroup_fs_type
);
4995 kobject_put(cgroup_kobj
);
4999 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
5003 bdi_destroy(&cgroup_backing_dev_info
);
5009 * proc_cgroup_show()
5010 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5011 * - Used for /proc/<pid>/cgroup.
5012 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
5013 * doesn't really matter if tsk->cgroup changes after we read it,
5014 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
5015 * anyway. No need to check that tsk->cgroup != NULL, thanks to
5016 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
5017 * cgroup to top_cgroup.
5020 /* TODO: Use a proper seq_file iterator */
5021 int proc_cgroup_show(struct seq_file
*m
, void *v
)
5024 struct task_struct
*tsk
;
5027 struct cgroupfs_root
*root
;
5030 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5036 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
5042 mutex_lock(&cgroup_mutex
);
5044 for_each_active_root(root
) {
5045 struct cgroup_subsys
*ss
;
5046 struct cgroup
*cgrp
;
5049 seq_printf(m
, "%d:", root
->hierarchy_id
);
5050 for_each_root_subsys(root
, ss
)
5051 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
5052 if (strlen(root
->name
))
5053 seq_printf(m
, "%sname=%s", count
? "," : "",
5056 cgrp
= task_cgroup_from_root(tsk
, root
);
5057 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
5065 mutex_unlock(&cgroup_mutex
);
5066 put_task_struct(tsk
);
5073 /* Display information about each subsystem and each hierarchy */
5074 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5076 struct cgroup_subsys
*ss
;
5079 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5081 * ideally we don't want subsystems moving around while we do this.
5082 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5083 * subsys/hierarchy state.
5085 mutex_lock(&cgroup_mutex
);
5087 for_each_subsys(ss
, i
)
5088 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5089 ss
->name
, ss
->root
->hierarchy_id
,
5090 ss
->root
->number_of_cgroups
, !ss
->disabled
);
5092 mutex_unlock(&cgroup_mutex
);
5096 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5098 return single_open(file
, proc_cgroupstats_show
, NULL
);
5101 static const struct file_operations proc_cgroupstats_operations
= {
5102 .open
= cgroupstats_open
,
5104 .llseek
= seq_lseek
,
5105 .release
= single_release
,
5109 * cgroup_fork - attach newly forked task to its parents cgroup.
5110 * @child: pointer to task_struct of forking parent process.
5112 * Description: A task inherits its parent's cgroup at fork().
5114 * A pointer to the shared css_set was automatically copied in
5115 * fork.c by dup_task_struct(). However, we ignore that copy, since
5116 * it was not made under the protection of RCU or cgroup_mutex, so
5117 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
5118 * have already changed current->cgroups, allowing the previously
5119 * referenced cgroup group to be removed and freed.
5121 * At the point that cgroup_fork() is called, 'current' is the parent
5122 * task, and the passed argument 'child' points to the child task.
5124 void cgroup_fork(struct task_struct
*child
)
5127 get_css_set(task_css_set(current
));
5128 child
->cgroups
= current
->cgroups
;
5129 task_unlock(current
);
5130 INIT_LIST_HEAD(&child
->cg_list
);
5134 * cgroup_post_fork - called on a new task after adding it to the task list
5135 * @child: the task in question
5137 * Adds the task to the list running through its css_set if necessary and
5138 * call the subsystem fork() callbacks. Has to be after the task is
5139 * visible on the task list in case we race with the first call to
5140 * cgroup_iter_start() - to guarantee that the new task ends up on its
5143 void cgroup_post_fork(struct task_struct
*child
)
5145 struct cgroup_subsys
*ss
;
5149 * use_task_css_set_links is set to 1 before we walk the tasklist
5150 * under the tasklist_lock and we read it here after we added the child
5151 * to the tasklist under the tasklist_lock as well. If the child wasn't
5152 * yet in the tasklist when we walked through it from
5153 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
5154 * should be visible now due to the paired locking and barriers implied
5155 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
5156 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
5159 if (use_task_css_set_links
) {
5160 write_lock(&css_set_lock
);
5162 if (list_empty(&child
->cg_list
))
5163 list_add(&child
->cg_list
, &task_css_set(child
)->tasks
);
5165 write_unlock(&css_set_lock
);
5169 * Call ss->fork(). This must happen after @child is linked on
5170 * css_set; otherwise, @child might change state between ->fork()
5171 * and addition to css_set.
5173 if (need_forkexit_callback
) {
5175 * fork/exit callbacks are supported only for builtin
5176 * subsystems, and the builtin section of the subsys
5177 * array is immutable, so we don't need to lock the
5178 * subsys array here. On the other hand, modular section
5179 * of the array can be freed at module unload, so we
5182 for_each_builtin_subsys(ss
, i
)
5189 * cgroup_exit - detach cgroup from exiting task
5190 * @tsk: pointer to task_struct of exiting process
5191 * @run_callback: run exit callbacks?
5193 * Description: Detach cgroup from @tsk and release it.
5195 * Note that cgroups marked notify_on_release force every task in
5196 * them to take the global cgroup_mutex mutex when exiting.
5197 * This could impact scaling on very large systems. Be reluctant to
5198 * use notify_on_release cgroups where very high task exit scaling
5199 * is required on large systems.
5201 * the_top_cgroup_hack:
5203 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
5205 * We call cgroup_exit() while the task is still competent to
5206 * handle notify_on_release(), then leave the task attached to the
5207 * root cgroup in each hierarchy for the remainder of its exit.
5209 * To do this properly, we would increment the reference count on
5210 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
5211 * code we would add a second cgroup function call, to drop that
5212 * reference. This would just create an unnecessary hot spot on
5213 * the top_cgroup reference count, to no avail.
5215 * Normally, holding a reference to a cgroup without bumping its
5216 * count is unsafe. The cgroup could go away, or someone could
5217 * attach us to a different cgroup, decrementing the count on
5218 * the first cgroup that we never incremented. But in this case,
5219 * top_cgroup isn't going away, and either task has PF_EXITING set,
5220 * which wards off any cgroup_attach_task() attempts, or task is a failed
5221 * fork, never visible to cgroup_attach_task.
5223 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
5225 struct cgroup_subsys
*ss
;
5226 struct css_set
*cset
;
5230 * Unlink from the css_set task list if necessary.
5231 * Optimistically check cg_list before taking
5234 if (!list_empty(&tsk
->cg_list
)) {
5235 write_lock(&css_set_lock
);
5236 if (!list_empty(&tsk
->cg_list
))
5237 list_del_init(&tsk
->cg_list
);
5238 write_unlock(&css_set_lock
);
5241 /* Reassign the task to the init_css_set. */
5243 cset
= task_css_set(tsk
);
5244 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5246 if (run_callbacks
&& need_forkexit_callback
) {
5248 * fork/exit callbacks are supported only for builtin
5249 * subsystems, see cgroup_post_fork() for details.
5251 for_each_builtin_subsys(ss
, i
) {
5253 struct cgroup_subsys_state
*old_css
= cset
->subsys
[i
];
5254 struct cgroup_subsys_state
*css
= task_css(tsk
, i
);
5256 ss
->exit(css
, old_css
, tsk
);
5262 put_css_set_taskexit(cset
);
5265 static void check_for_release(struct cgroup
*cgrp
)
5267 if (cgroup_is_releasable(cgrp
) &&
5268 list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
)) {
5270 * Control Group is currently removeable. If it's not
5271 * already queued for a userspace notification, queue
5274 int need_schedule_work
= 0;
5276 raw_spin_lock(&release_list_lock
);
5277 if (!cgroup_is_dead(cgrp
) &&
5278 list_empty(&cgrp
->release_list
)) {
5279 list_add(&cgrp
->release_list
, &release_list
);
5280 need_schedule_work
= 1;
5282 raw_spin_unlock(&release_list_lock
);
5283 if (need_schedule_work
)
5284 schedule_work(&release_agent_work
);
5289 * Notify userspace when a cgroup is released, by running the
5290 * configured release agent with the name of the cgroup (path
5291 * relative to the root of cgroup file system) as the argument.
5293 * Most likely, this user command will try to rmdir this cgroup.
5295 * This races with the possibility that some other task will be
5296 * attached to this cgroup before it is removed, or that some other
5297 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5298 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5299 * unused, and this cgroup will be reprieved from its death sentence,
5300 * to continue to serve a useful existence. Next time it's released,
5301 * we will get notified again, if it still has 'notify_on_release' set.
5303 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5304 * means only wait until the task is successfully execve()'d. The
5305 * separate release agent task is forked by call_usermodehelper(),
5306 * then control in this thread returns here, without waiting for the
5307 * release agent task. We don't bother to wait because the caller of
5308 * this routine has no use for the exit status of the release agent
5309 * task, so no sense holding our caller up for that.
5311 static void cgroup_release_agent(struct work_struct
*work
)
5313 BUG_ON(work
!= &release_agent_work
);
5314 mutex_lock(&cgroup_mutex
);
5315 raw_spin_lock(&release_list_lock
);
5316 while (!list_empty(&release_list
)) {
5317 char *argv
[3], *envp
[3];
5319 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5320 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5323 list_del_init(&cgrp
->release_list
);
5324 raw_spin_unlock(&release_list_lock
);
5325 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5328 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5330 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5335 argv
[i
++] = agentbuf
;
5336 argv
[i
++] = pathbuf
;
5340 /* minimal command environment */
5341 envp
[i
++] = "HOME=/";
5342 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5345 /* Drop the lock while we invoke the usermode helper,
5346 * since the exec could involve hitting disk and hence
5347 * be a slow process */
5348 mutex_unlock(&cgroup_mutex
);
5349 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5350 mutex_lock(&cgroup_mutex
);
5354 raw_spin_lock(&release_list_lock
);
5356 raw_spin_unlock(&release_list_lock
);
5357 mutex_unlock(&cgroup_mutex
);
5360 static int __init
cgroup_disable(char *str
)
5362 struct cgroup_subsys
*ss
;
5366 while ((token
= strsep(&str
, ",")) != NULL
) {
5371 * cgroup_disable, being at boot time, can't know about
5372 * module subsystems, so we don't worry about them.
5374 for_each_builtin_subsys(ss
, i
) {
5375 if (!strcmp(token
, ss
->name
)) {
5377 printk(KERN_INFO
"Disabling %s control group"
5378 " subsystem\n", ss
->name
);
5385 __setup("cgroup_disable=", cgroup_disable
);
5388 * Functons for CSS ID.
5391 /* to get ID other than 0, this should be called when !cgroup_is_dead() */
5392 unsigned short css_id(struct cgroup_subsys_state
*css
)
5394 struct css_id
*cssid
;
5397 * This css_id() can return correct value when somone has refcnt
5398 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5399 * it's unchanged until freed.
5401 cssid
= rcu_dereference_raw(css
->id
);
5407 EXPORT_SYMBOL_GPL(css_id
);
5410 * css_is_ancestor - test "root" css is an ancestor of "child"
5411 * @child: the css to be tested.
5412 * @root: the css supporsed to be an ancestor of the child.
5414 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5415 * this function reads css->id, the caller must hold rcu_read_lock().
5416 * But, considering usual usage, the csses should be valid objects after test.
5417 * Assuming that the caller will do some action to the child if this returns
5418 * returns true, the caller must take "child";s reference count.
5419 * If "child" is valid object and this returns true, "root" is valid, too.
5422 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5423 const struct cgroup_subsys_state
*root
)
5425 struct css_id
*child_id
;
5426 struct css_id
*root_id
;
5428 child_id
= rcu_dereference(child
->id
);
5431 root_id
= rcu_dereference(root
->id
);
5434 if (child_id
->depth
< root_id
->depth
)
5436 if (child_id
->stack
[root_id
->depth
] != root_id
->id
)
5441 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5443 struct css_id
*id
= rcu_dereference_protected(css
->id
, true);
5445 /* When this is called before css_id initialization, id can be NULL */
5449 BUG_ON(!ss
->use_id
);
5451 rcu_assign_pointer(id
->css
, NULL
);
5452 rcu_assign_pointer(css
->id
, NULL
);
5453 spin_lock(&ss
->id_lock
);
5454 idr_remove(&ss
->idr
, id
->id
);
5455 spin_unlock(&ss
->id_lock
);
5456 kfree_rcu(id
, rcu_head
);
5458 EXPORT_SYMBOL_GPL(free_css_id
);
5461 * This is called by init or create(). Then, calls to this function are
5462 * always serialized (By cgroup_mutex() at create()).
5465 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5467 struct css_id
*newid
;
5470 BUG_ON(!ss
->use_id
);
5472 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5473 newid
= kzalloc(size
, GFP_KERNEL
);
5475 return ERR_PTR(-ENOMEM
);
5477 idr_preload(GFP_KERNEL
);
5478 spin_lock(&ss
->id_lock
);
5479 /* Don't use 0. allocates an ID of 1-65535 */
5480 ret
= idr_alloc(&ss
->idr
, newid
, 1, CSS_ID_MAX
+ 1, GFP_NOWAIT
);
5481 spin_unlock(&ss
->id_lock
);
5484 /* Returns error when there are no free spaces for new ID.*/
5489 newid
->depth
= depth
;
5493 return ERR_PTR(ret
);
5497 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5498 struct cgroup_subsys_state
*rootcss
)
5500 struct css_id
*newid
;
5502 spin_lock_init(&ss
->id_lock
);
5505 newid
= get_new_cssid(ss
, 0);
5507 return PTR_ERR(newid
);
5509 newid
->stack
[0] = newid
->id
;
5510 RCU_INIT_POINTER(newid
->css
, rootcss
);
5511 RCU_INIT_POINTER(rootcss
->id
, newid
);
5515 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5516 struct cgroup
*child
)
5518 int subsys_id
, i
, depth
= 0;
5519 struct cgroup_subsys_state
*parent_css
, *child_css
;
5520 struct css_id
*child_id
, *parent_id
;
5522 subsys_id
= ss
->subsys_id
;
5523 parent_css
= parent
->subsys
[subsys_id
];
5524 child_css
= child
->subsys
[subsys_id
];
5525 parent_id
= rcu_dereference_protected(parent_css
->id
, true);
5526 depth
= parent_id
->depth
+ 1;
5528 child_id
= get_new_cssid(ss
, depth
);
5529 if (IS_ERR(child_id
))
5530 return PTR_ERR(child_id
);
5532 for (i
= 0; i
< depth
; i
++)
5533 child_id
->stack
[i
] = parent_id
->stack
[i
];
5534 child_id
->stack
[depth
] = child_id
->id
;
5536 * child_id->css pointer will be set after this cgroup is available
5537 * see cgroup_populate_dir()
5539 rcu_assign_pointer(child_css
->id
, child_id
);
5545 * css_lookup - lookup css by id
5546 * @ss: cgroup subsys to be looked into.
5549 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5550 * NULL if not. Should be called under rcu_read_lock()
5552 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5554 struct css_id
*cssid
= NULL
;
5556 BUG_ON(!ss
->use_id
);
5557 cssid
= idr_find(&ss
->idr
, id
);
5559 if (unlikely(!cssid
))
5562 return rcu_dereference(cssid
->css
);
5564 EXPORT_SYMBOL_GPL(css_lookup
);
5567 * get corresponding css from file open on cgroupfs directory
5569 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5571 struct cgroup
*cgrp
;
5572 struct inode
*inode
;
5573 struct cgroup_subsys_state
*css
;
5575 inode
= file_inode(f
);
5576 /* check in cgroup filesystem dir */
5577 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5578 return ERR_PTR(-EBADF
);
5580 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5581 return ERR_PTR(-EINVAL
);
5584 cgrp
= __d_cgrp(f
->f_dentry
);
5585 css
= cgrp
->subsys
[id
];
5586 return css
? css
: ERR_PTR(-ENOENT
);
5589 #ifdef CONFIG_CGROUP_DEBUG
5590 static struct cgroup_subsys_state
*
5591 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
5593 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5596 return ERR_PTR(-ENOMEM
);
5601 static void debug_css_free(struct cgroup_subsys_state
*css
)
5606 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
5609 return cgroup_task_count(css
->cgroup
);
5612 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
5615 return (u64
)(unsigned long)current
->cgroups
;
5618 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
5624 count
= atomic_read(&task_css_set(current
)->refcount
);
5629 static int current_css_set_cg_links_read(struct cgroup_subsys_state
*css
,
5631 struct seq_file
*seq
)
5633 struct cgrp_cset_link
*link
;
5634 struct css_set
*cset
;
5636 read_lock(&css_set_lock
);
5638 cset
= rcu_dereference(current
->cgroups
);
5639 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5640 struct cgroup
*c
= link
->cgrp
;
5644 name
= c
->dentry
->d_name
.name
;
5647 seq_printf(seq
, "Root %d group %s\n",
5648 c
->root
->hierarchy_id
, name
);
5651 read_unlock(&css_set_lock
);
5655 #define MAX_TASKS_SHOWN_PER_CSS 25
5656 static int cgroup_css_links_read(struct cgroup_subsys_state
*css
,
5657 struct cftype
*cft
, struct seq_file
*seq
)
5659 struct cgrp_cset_link
*link
;
5661 read_lock(&css_set_lock
);
5662 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
5663 struct css_set
*cset
= link
->cset
;
5664 struct task_struct
*task
;
5666 seq_printf(seq
, "css_set %p\n", cset
);
5667 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5668 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5669 seq_puts(seq
, " ...\n");
5672 seq_printf(seq
, " task %d\n",
5673 task_pid_vnr(task
));
5677 read_unlock(&css_set_lock
);
5681 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
5683 return test_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
5686 static struct cftype debug_files
[] = {
5688 .name
= "taskcount",
5689 .read_u64
= debug_taskcount_read
,
5693 .name
= "current_css_set",
5694 .read_u64
= current_css_set_read
,
5698 .name
= "current_css_set_refcount",
5699 .read_u64
= current_css_set_refcount_read
,
5703 .name
= "current_css_set_cg_links",
5704 .read_seq_string
= current_css_set_cg_links_read
,
5708 .name
= "cgroup_css_links",
5709 .read_seq_string
= cgroup_css_links_read
,
5713 .name
= "releasable",
5714 .read_u64
= releasable_read
,
5720 struct cgroup_subsys debug_subsys
= {
5722 .css_alloc
= debug_css_alloc
,
5723 .css_free
= debug_css_free
,
5724 .subsys_id
= debug_subsys_id
,
5725 .base_cftypes
= debug_files
,
5727 #endif /* CONFIG_CGROUP_DEBUG */