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 task_subsys_state_check() */
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 cgroup_subsys
*subsys
,
223 struct cftype cfts
[], bool is_add
);
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 "cg" matches "old_cg" 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
)
859 cgrp
->root
->number_of_cgroups
--;
860 mutex_unlock(&cgroup_mutex
);
863 * We get a ref to the parent's dentry, and put the ref when
864 * this cgroup is being freed, so it's guaranteed that the
865 * parent won't be destroyed before its children.
867 dput(cgrp
->parent
->dentry
);
869 ida_simple_remove(&cgrp
->root
->cgroup_ida
, cgrp
->id
);
872 * Drop the active superblock reference that we took when we
873 * created the cgroup. This will free cgrp->root, if we are
874 * holding the last reference to @sb.
876 deactivate_super(cgrp
->root
->sb
);
879 * if we're getting rid of the cgroup, refcount should ensure
880 * that there are no pidlists left.
882 BUG_ON(!list_empty(&cgrp
->pidlists
));
884 simple_xattrs_free(&cgrp
->xattrs
);
886 kfree(rcu_dereference_raw(cgrp
->name
));
890 static void cgroup_free_rcu(struct rcu_head
*head
)
892 struct cgroup
*cgrp
= container_of(head
, struct cgroup
, rcu_head
);
894 INIT_WORK(&cgrp
->destroy_work
, cgroup_free_fn
);
895 schedule_work(&cgrp
->destroy_work
);
898 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
900 /* is dentry a directory ? if so, kfree() associated cgroup */
901 if (S_ISDIR(inode
->i_mode
)) {
902 struct cgroup
*cgrp
= dentry
->d_fsdata
;
904 BUG_ON(!(cgroup_is_dead(cgrp
)));
905 call_rcu(&cgrp
->rcu_head
, cgroup_free_rcu
);
907 struct cfent
*cfe
= __d_cfe(dentry
);
908 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
910 WARN_ONCE(!list_empty(&cfe
->node
) &&
911 cgrp
!= &cgrp
->root
->top_cgroup
,
912 "cfe still linked for %s\n", cfe
->type
->name
);
913 simple_xattrs_free(&cfe
->xattrs
);
919 static int cgroup_delete(const struct dentry
*d
)
924 static void remove_dir(struct dentry
*d
)
926 struct dentry
*parent
= dget(d
->d_parent
);
929 simple_rmdir(parent
->d_inode
, d
);
933 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
937 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
938 lockdep_assert_held(&cgroup_mutex
);
941 * If we're doing cleanup due to failure of cgroup_create(),
942 * the corresponding @cfe may not exist.
944 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
945 struct dentry
*d
= cfe
->dentry
;
947 if (cft
&& cfe
->type
!= cft
)
952 simple_unlink(cgrp
->dentry
->d_inode
, d
);
953 list_del_init(&cfe
->node
);
961 * cgroup_clear_dir - remove subsys files in a cgroup directory
962 * @cgrp: target cgroup
963 * @subsys_mask: mask of the subsystem ids whose files should be removed
965 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
967 struct cgroup_subsys
*ss
;
970 for_each_subsys(ss
, i
) {
971 struct cftype_set
*set
;
973 if (!test_bit(i
, &subsys_mask
))
975 list_for_each_entry(set
, &ss
->cftsets
, node
)
976 cgroup_addrm_files(cgrp
, NULL
, set
->cfts
, false);
981 * NOTE : the dentry must have been dget()'ed
983 static void cgroup_d_remove_dir(struct dentry
*dentry
)
985 struct dentry
*parent
;
987 parent
= dentry
->d_parent
;
988 spin_lock(&parent
->d_lock
);
989 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
990 list_del_init(&dentry
->d_u
.d_child
);
991 spin_unlock(&dentry
->d_lock
);
992 spin_unlock(&parent
->d_lock
);
997 * Call with cgroup_mutex held. Drops reference counts on modules, including
998 * any duplicate ones that parse_cgroupfs_options took. If this function
999 * returns an error, no reference counts are touched.
1001 static int rebind_subsystems(struct cgroupfs_root
*root
,
1002 unsigned long added_mask
, unsigned removed_mask
)
1004 struct cgroup
*cgrp
= &root
->top_cgroup
;
1005 struct cgroup_subsys
*ss
;
1008 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1009 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1011 /* Check that any added subsystems are currently free */
1012 for_each_subsys(ss
, i
) {
1013 unsigned long bit
= 1UL << i
;
1015 if (!(bit
& added_mask
))
1018 if (ss
->root
!= &cgroup_dummy_root
) {
1019 /* Subsystem isn't free */
1024 /* Currently we don't handle adding/removing subsystems when
1025 * any child cgroups exist. This is theoretically supportable
1026 * but involves complex error handling, so it's being left until
1028 if (root
->number_of_cgroups
> 1)
1031 ret
= cgroup_populate_dir(cgrp
, added_mask
);
1036 * Nothing can fail from this point on. Remove files for the
1037 * removed subsystems and rebind each subsystem.
1039 cgroup_clear_dir(cgrp
, removed_mask
);
1041 for_each_subsys(ss
, i
) {
1042 unsigned long bit
= 1UL << i
;
1044 if (bit
& added_mask
) {
1045 /* We're binding this subsystem to this hierarchy */
1046 BUG_ON(cgrp
->subsys
[i
]);
1047 BUG_ON(!cgroup_dummy_top
->subsys
[i
]);
1048 BUG_ON(cgroup_dummy_top
->subsys
[i
]->cgroup
!= cgroup_dummy_top
);
1050 cgrp
->subsys
[i
] = cgroup_dummy_top
->subsys
[i
];
1051 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1052 list_move(&ss
->sibling
, &root
->subsys_list
);
1057 /* refcount was already taken, and we're keeping it */
1058 root
->subsys_mask
|= bit
;
1059 } else if (bit
& removed_mask
) {
1060 /* We're removing this subsystem */
1061 BUG_ON(cgrp
->subsys
[i
] != cgroup_dummy_top
->subsys
[i
]);
1062 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1065 ss
->bind(cgroup_dummy_top
);
1066 cgroup_dummy_top
->subsys
[i
]->cgroup
= cgroup_dummy_top
;
1067 cgrp
->subsys
[i
] = NULL
;
1068 cgroup_subsys
[i
]->root
= &cgroup_dummy_root
;
1069 list_move(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
1071 /* subsystem is now free - drop reference on module */
1072 module_put(ss
->module
);
1073 root
->subsys_mask
&= ~bit
;
1074 } else if (bit
& root
->subsys_mask
) {
1075 /* Subsystem state should already exist */
1076 BUG_ON(!cgrp
->subsys
[i
]);
1078 * a refcount was taken, but we already had one, so
1079 * drop the extra reference.
1081 module_put(ss
->module
);
1082 #ifdef CONFIG_MODULE_UNLOAD
1083 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1086 /* Subsystem state shouldn't exist */
1087 BUG_ON(cgrp
->subsys
[i
]);
1092 * Mark @root has finished binding subsystems. @root->subsys_mask
1093 * now matches the bound subsystems.
1095 root
->flags
|= CGRP_ROOT_SUBSYS_BOUND
;
1100 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1102 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1103 struct cgroup_subsys
*ss
;
1105 mutex_lock(&cgroup_root_mutex
);
1106 for_each_root_subsys(root
, ss
)
1107 seq_printf(seq
, ",%s", ss
->name
);
1108 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
)
1109 seq_puts(seq
, ",sane_behavior");
1110 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1111 seq_puts(seq
, ",noprefix");
1112 if (root
->flags
& CGRP_ROOT_XATTR
)
1113 seq_puts(seq
, ",xattr");
1114 if (strlen(root
->release_agent_path
))
1115 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1116 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
))
1117 seq_puts(seq
, ",clone_children");
1118 if (strlen(root
->name
))
1119 seq_printf(seq
, ",name=%s", root
->name
);
1120 mutex_unlock(&cgroup_root_mutex
);
1124 struct cgroup_sb_opts
{
1125 unsigned long subsys_mask
;
1126 unsigned long flags
;
1127 char *release_agent
;
1128 bool cpuset_clone_children
;
1130 /* User explicitly requested empty subsystem */
1133 struct cgroupfs_root
*new_root
;
1138 * Convert a hierarchy specifier into a bitmask of subsystems and
1139 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1140 * array. This function takes refcounts on subsystems to be used, unless it
1141 * returns error, in which case no refcounts are taken.
1143 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1145 char *token
, *o
= data
;
1146 bool all_ss
= false, one_ss
= false;
1147 unsigned long mask
= (unsigned long)-1;
1148 bool module_pin_failed
= false;
1149 struct cgroup_subsys
*ss
;
1152 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1154 #ifdef CONFIG_CPUSETS
1155 mask
= ~(1UL << cpuset_subsys_id
);
1158 memset(opts
, 0, sizeof(*opts
));
1160 while ((token
= strsep(&o
, ",")) != NULL
) {
1163 if (!strcmp(token
, "none")) {
1164 /* Explicitly have no subsystems */
1168 if (!strcmp(token
, "all")) {
1169 /* Mutually exclusive option 'all' + subsystem name */
1175 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1176 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1179 if (!strcmp(token
, "noprefix")) {
1180 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1183 if (!strcmp(token
, "clone_children")) {
1184 opts
->cpuset_clone_children
= true;
1187 if (!strcmp(token
, "xattr")) {
1188 opts
->flags
|= CGRP_ROOT_XATTR
;
1191 if (!strncmp(token
, "release_agent=", 14)) {
1192 /* Specifying two release agents is forbidden */
1193 if (opts
->release_agent
)
1195 opts
->release_agent
=
1196 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1197 if (!opts
->release_agent
)
1201 if (!strncmp(token
, "name=", 5)) {
1202 const char *name
= token
+ 5;
1203 /* Can't specify an empty name */
1206 /* Must match [\w.-]+ */
1207 for (i
= 0; i
< strlen(name
); i
++) {
1211 if ((c
== '.') || (c
== '-') || (c
== '_'))
1215 /* Specifying two names is forbidden */
1218 opts
->name
= kstrndup(name
,
1219 MAX_CGROUP_ROOT_NAMELEN
- 1,
1227 for_each_subsys(ss
, i
) {
1228 if (strcmp(token
, ss
->name
))
1233 /* Mutually exclusive option 'all' + subsystem name */
1236 set_bit(i
, &opts
->subsys_mask
);
1241 if (i
== CGROUP_SUBSYS_COUNT
)
1246 * If the 'all' option was specified select all the subsystems,
1247 * otherwise if 'none', 'name=' and a subsystem name options
1248 * were not specified, let's default to 'all'
1250 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1251 for_each_subsys(ss
, i
)
1253 set_bit(i
, &opts
->subsys_mask
);
1255 /* Consistency checks */
1257 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1258 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1260 if (opts
->flags
& CGRP_ROOT_NOPREFIX
) {
1261 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1265 if (opts
->cpuset_clone_children
) {
1266 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1272 * Option noprefix was introduced just for backward compatibility
1273 * with the old cpuset, so we allow noprefix only if mounting just
1274 * the cpuset subsystem.
1276 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1280 /* Can't specify "none" and some subsystems */
1281 if (opts
->subsys_mask
&& opts
->none
)
1285 * We either have to specify by name or by subsystems. (So all
1286 * empty hierarchies must have a name).
1288 if (!opts
->subsys_mask
&& !opts
->name
)
1292 * Grab references on all the modules we'll need, so the subsystems
1293 * don't dance around before rebind_subsystems attaches them. This may
1294 * take duplicate reference counts on a subsystem that's already used,
1295 * but rebind_subsystems handles this case.
1297 for_each_subsys(ss
, i
) {
1298 if (!(opts
->subsys_mask
& (1UL << i
)))
1300 if (!try_module_get(cgroup_subsys
[i
]->module
)) {
1301 module_pin_failed
= true;
1305 if (module_pin_failed
) {
1307 * oops, one of the modules was going away. this means that we
1308 * raced with a module_delete call, and to the user this is
1309 * essentially a "subsystem doesn't exist" case.
1311 for (i
--; i
>= 0; i
--) {
1312 /* drop refcounts only on the ones we took */
1313 unsigned long bit
= 1UL << i
;
1315 if (!(bit
& opts
->subsys_mask
))
1317 module_put(cgroup_subsys
[i
]->module
);
1325 static void drop_parsed_module_refcounts(unsigned long subsys_mask
)
1327 struct cgroup_subsys
*ss
;
1330 mutex_lock(&cgroup_mutex
);
1331 for_each_subsys(ss
, i
)
1332 if (subsys_mask
& (1UL << i
))
1333 module_put(cgroup_subsys
[i
]->module
);
1334 mutex_unlock(&cgroup_mutex
);
1337 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1340 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1341 struct cgroup
*cgrp
= &root
->top_cgroup
;
1342 struct cgroup_sb_opts opts
;
1343 unsigned long added_mask
, removed_mask
;
1345 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1346 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1350 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1351 mutex_lock(&cgroup_mutex
);
1352 mutex_lock(&cgroup_root_mutex
);
1354 /* See what subsystems are wanted */
1355 ret
= parse_cgroupfs_options(data
, &opts
);
1359 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1360 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1361 task_tgid_nr(current
), current
->comm
);
1363 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1364 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1366 /* Don't allow flags or name to change at remount */
1367 if (((opts
.flags
^ root
->flags
) & CGRP_ROOT_OPTION_MASK
) ||
1368 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1369 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1370 opts
.flags
& CGRP_ROOT_OPTION_MASK
, opts
.name
?: "",
1371 root
->flags
& CGRP_ROOT_OPTION_MASK
, root
->name
);
1376 ret
= rebind_subsystems(root
, added_mask
, removed_mask
);
1380 if (opts
.release_agent
)
1381 strcpy(root
->release_agent_path
, opts
.release_agent
);
1383 kfree(opts
.release_agent
);
1385 mutex_unlock(&cgroup_root_mutex
);
1386 mutex_unlock(&cgroup_mutex
);
1387 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1389 drop_parsed_module_refcounts(opts
.subsys_mask
);
1393 static const struct super_operations cgroup_ops
= {
1394 .statfs
= simple_statfs
,
1395 .drop_inode
= generic_delete_inode
,
1396 .show_options
= cgroup_show_options
,
1397 .remount_fs
= cgroup_remount
,
1400 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1402 INIT_LIST_HEAD(&cgrp
->sibling
);
1403 INIT_LIST_HEAD(&cgrp
->children
);
1404 INIT_LIST_HEAD(&cgrp
->files
);
1405 INIT_LIST_HEAD(&cgrp
->cset_links
);
1406 INIT_LIST_HEAD(&cgrp
->release_list
);
1407 INIT_LIST_HEAD(&cgrp
->pidlists
);
1408 mutex_init(&cgrp
->pidlist_mutex
);
1409 INIT_LIST_HEAD(&cgrp
->event_list
);
1410 spin_lock_init(&cgrp
->event_list_lock
);
1411 simple_xattrs_init(&cgrp
->xattrs
);
1414 static void init_cgroup_root(struct cgroupfs_root
*root
)
1416 struct cgroup
*cgrp
= &root
->top_cgroup
;
1418 INIT_LIST_HEAD(&root
->subsys_list
);
1419 INIT_LIST_HEAD(&root
->root_list
);
1420 root
->number_of_cgroups
= 1;
1422 RCU_INIT_POINTER(cgrp
->name
, &root_cgroup_name
);
1423 init_cgroup_housekeeping(cgrp
);
1426 static int cgroup_init_root_id(struct cgroupfs_root
*root
, int start
, int end
)
1430 lockdep_assert_held(&cgroup_mutex
);
1431 lockdep_assert_held(&cgroup_root_mutex
);
1433 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, start
, end
,
1438 root
->hierarchy_id
= id
;
1442 static void cgroup_exit_root_id(struct cgroupfs_root
*root
)
1444 lockdep_assert_held(&cgroup_mutex
);
1445 lockdep_assert_held(&cgroup_root_mutex
);
1447 if (root
->hierarchy_id
) {
1448 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1449 root
->hierarchy_id
= 0;
1453 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1455 struct cgroup_sb_opts
*opts
= data
;
1456 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1458 /* If we asked for a name then it must match */
1459 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1463 * If we asked for subsystems (or explicitly for no
1464 * subsystems) then they must match
1466 if ((opts
->subsys_mask
|| opts
->none
)
1467 && (opts
->subsys_mask
!= root
->subsys_mask
))
1473 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1475 struct cgroupfs_root
*root
;
1477 if (!opts
->subsys_mask
&& !opts
->none
)
1480 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1482 return ERR_PTR(-ENOMEM
);
1484 init_cgroup_root(root
);
1487 * We need to set @root->subsys_mask now so that @root can be
1488 * matched by cgroup_test_super() before it finishes
1489 * initialization; otherwise, competing mounts with the same
1490 * options may try to bind the same subsystems instead of waiting
1491 * for the first one leading to unexpected mount errors.
1492 * SUBSYS_BOUND will be set once actual binding is complete.
1494 root
->subsys_mask
= opts
->subsys_mask
;
1495 root
->flags
= opts
->flags
;
1496 ida_init(&root
->cgroup_ida
);
1497 if (opts
->release_agent
)
1498 strcpy(root
->release_agent_path
, opts
->release_agent
);
1500 strcpy(root
->name
, opts
->name
);
1501 if (opts
->cpuset_clone_children
)
1502 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1506 static void cgroup_free_root(struct cgroupfs_root
*root
)
1509 /* hierarhcy ID shoulid already have been released */
1510 WARN_ON_ONCE(root
->hierarchy_id
);
1512 ida_destroy(&root
->cgroup_ida
);
1517 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1520 struct cgroup_sb_opts
*opts
= data
;
1522 /* If we don't have a new root, we can't set up a new sb */
1523 if (!opts
->new_root
)
1526 BUG_ON(!opts
->subsys_mask
&& !opts
->none
);
1528 ret
= set_anon_super(sb
, NULL
);
1532 sb
->s_fs_info
= opts
->new_root
;
1533 opts
->new_root
->sb
= sb
;
1535 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1536 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1537 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1538 sb
->s_op
= &cgroup_ops
;
1543 static int cgroup_get_rootdir(struct super_block
*sb
)
1545 static const struct dentry_operations cgroup_dops
= {
1546 .d_iput
= cgroup_diput
,
1547 .d_delete
= cgroup_delete
,
1550 struct inode
*inode
=
1551 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1556 inode
->i_fop
= &simple_dir_operations
;
1557 inode
->i_op
= &cgroup_dir_inode_operations
;
1558 /* directories start off with i_nlink == 2 (for "." entry) */
1560 sb
->s_root
= d_make_root(inode
);
1563 /* for everything else we want ->d_op set */
1564 sb
->s_d_op
= &cgroup_dops
;
1568 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1569 int flags
, const char *unused_dev_name
,
1572 struct cgroup_sb_opts opts
;
1573 struct cgroupfs_root
*root
;
1575 struct super_block
*sb
;
1576 struct cgroupfs_root
*new_root
;
1577 struct list_head tmp_links
;
1578 struct inode
*inode
;
1579 const struct cred
*cred
;
1581 /* First find the desired set of subsystems */
1582 mutex_lock(&cgroup_mutex
);
1583 ret
= parse_cgroupfs_options(data
, &opts
);
1584 mutex_unlock(&cgroup_mutex
);
1589 * Allocate a new cgroup root. We may not need it if we're
1590 * reusing an existing hierarchy.
1592 new_root
= cgroup_root_from_opts(&opts
);
1593 if (IS_ERR(new_root
)) {
1594 ret
= PTR_ERR(new_root
);
1597 opts
.new_root
= new_root
;
1599 /* Locate an existing or new sb for this hierarchy */
1600 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, 0, &opts
);
1603 cgroup_free_root(opts
.new_root
);
1607 root
= sb
->s_fs_info
;
1609 if (root
== opts
.new_root
) {
1610 /* We used the new root structure, so this is a new hierarchy */
1611 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1612 struct cgroupfs_root
*existing_root
;
1614 struct css_set
*cset
;
1616 BUG_ON(sb
->s_root
!= NULL
);
1618 ret
= cgroup_get_rootdir(sb
);
1620 goto drop_new_super
;
1621 inode
= sb
->s_root
->d_inode
;
1623 mutex_lock(&inode
->i_mutex
);
1624 mutex_lock(&cgroup_mutex
);
1625 mutex_lock(&cgroup_root_mutex
);
1627 /* Check for name clashes with existing mounts */
1629 if (strlen(root
->name
))
1630 for_each_active_root(existing_root
)
1631 if (!strcmp(existing_root
->name
, root
->name
))
1635 * We're accessing css_set_count without locking
1636 * css_set_lock here, but that's OK - it can only be
1637 * increased by someone holding cgroup_lock, and
1638 * that's us. The worst that can happen is that we
1639 * have some link structures left over
1641 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1645 /* ID 0 is reserved for dummy root, 1 for unified hierarchy */
1646 ret
= cgroup_init_root_id(root
, 2, 0);
1650 sb
->s_root
->d_fsdata
= root_cgrp
;
1651 root_cgrp
->dentry
= sb
->s_root
;
1654 * We're inside get_sb() and will call lookup_one_len() to
1655 * create the root files, which doesn't work if SELinux is
1656 * in use. The following cred dancing somehow works around
1657 * it. See 2ce9738ba ("cgroupfs: use init_cred when
1658 * populating new cgroupfs mount") for more details.
1660 cred
= override_creds(&init_cred
);
1662 ret
= cgroup_addrm_files(root_cgrp
, NULL
, cgroup_base_files
, true);
1666 ret
= rebind_subsystems(root
, root
->subsys_mask
, 0);
1673 * There must be no failure case after here, since rebinding
1674 * takes care of subsystems' refcounts, which are explicitly
1675 * dropped in the failure exit path.
1678 list_add(&root
->root_list
, &cgroup_roots
);
1679 cgroup_root_count
++;
1681 /* Link the top cgroup in this hierarchy into all
1682 * the css_set objects */
1683 write_lock(&css_set_lock
);
1684 hash_for_each(css_set_table
, i
, cset
, hlist
)
1685 link_css_set(&tmp_links
, cset
, root_cgrp
);
1686 write_unlock(&css_set_lock
);
1688 free_cgrp_cset_links(&tmp_links
);
1690 BUG_ON(!list_empty(&root_cgrp
->children
));
1691 BUG_ON(root
->number_of_cgroups
!= 1);
1693 mutex_unlock(&cgroup_root_mutex
);
1694 mutex_unlock(&cgroup_mutex
);
1695 mutex_unlock(&inode
->i_mutex
);
1698 * We re-used an existing hierarchy - the new root (if
1699 * any) is not needed
1701 cgroup_free_root(opts
.new_root
);
1703 if ((root
->flags
^ opts
.flags
) & CGRP_ROOT_OPTION_MASK
) {
1704 if ((root
->flags
| opts
.flags
) & CGRP_ROOT_SANE_BEHAVIOR
) {
1705 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1707 goto drop_new_super
;
1709 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1713 /* no subsys rebinding, so refcounts don't change */
1714 drop_parsed_module_refcounts(opts
.subsys_mask
);
1717 kfree(opts
.release_agent
);
1719 return dget(sb
->s_root
);
1722 free_cgrp_cset_links(&tmp_links
);
1723 cgroup_addrm_files(&root
->top_cgroup
, NULL
, cgroup_base_files
, false);
1726 cgroup_exit_root_id(root
);
1727 mutex_unlock(&cgroup_root_mutex
);
1728 mutex_unlock(&cgroup_mutex
);
1729 mutex_unlock(&inode
->i_mutex
);
1731 deactivate_locked_super(sb
);
1733 drop_parsed_module_refcounts(opts
.subsys_mask
);
1735 kfree(opts
.release_agent
);
1737 return ERR_PTR(ret
);
1740 static void cgroup_kill_sb(struct super_block
*sb
) {
1741 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1742 struct cgroup
*cgrp
= &root
->top_cgroup
;
1743 struct cgrp_cset_link
*link
, *tmp_link
;
1748 BUG_ON(root
->number_of_cgroups
!= 1);
1749 BUG_ON(!list_empty(&cgrp
->children
));
1751 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1752 mutex_lock(&cgroup_mutex
);
1753 mutex_lock(&cgroup_root_mutex
);
1755 /* Rebind all subsystems back to the default hierarchy */
1756 if (root
->flags
& CGRP_ROOT_SUBSYS_BOUND
) {
1757 ret
= rebind_subsystems(root
, 0, root
->subsys_mask
);
1758 /* Shouldn't be able to fail ... */
1763 * Release all the links from cset_links to this hierarchy's
1766 write_lock(&css_set_lock
);
1768 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1769 list_del(&link
->cset_link
);
1770 list_del(&link
->cgrp_link
);
1773 write_unlock(&css_set_lock
);
1775 if (!list_empty(&root
->root_list
)) {
1776 list_del(&root
->root_list
);
1777 cgroup_root_count
--;
1780 cgroup_exit_root_id(root
);
1782 mutex_unlock(&cgroup_root_mutex
);
1783 mutex_unlock(&cgroup_mutex
);
1784 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1786 simple_xattrs_free(&cgrp
->xattrs
);
1788 kill_litter_super(sb
);
1789 cgroup_free_root(root
);
1792 static struct file_system_type cgroup_fs_type
= {
1794 .mount
= cgroup_mount
,
1795 .kill_sb
= cgroup_kill_sb
,
1798 static struct kobject
*cgroup_kobj
;
1801 * cgroup_path - generate the path of a cgroup
1802 * @cgrp: the cgroup in question
1803 * @buf: the buffer to write the path into
1804 * @buflen: the length of the buffer
1806 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1808 * We can't generate cgroup path using dentry->d_name, as accessing
1809 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1810 * inode's i_mutex, while on the other hand cgroup_path() can be called
1811 * with some irq-safe spinlocks held.
1813 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1815 int ret
= -ENAMETOOLONG
;
1818 if (!cgrp
->parent
) {
1819 if (strlcpy(buf
, "/", buflen
) >= buflen
)
1820 return -ENAMETOOLONG
;
1824 start
= buf
+ buflen
- 1;
1829 const char *name
= cgroup_name(cgrp
);
1833 if ((start
-= len
) < buf
)
1835 memcpy(start
, name
, len
);
1841 cgrp
= cgrp
->parent
;
1842 } while (cgrp
->parent
);
1844 memmove(buf
, start
, buf
+ buflen
- start
);
1849 EXPORT_SYMBOL_GPL(cgroup_path
);
1852 * task_cgroup_path_from_hierarchy - cgroup path of a task on a hierarchy
1853 * @task: target task
1854 * @hierarchy_id: the hierarchy to look up @task's cgroup from
1855 * @buf: the buffer to write the path into
1856 * @buflen: the length of the buffer
1858 * Determine @task's cgroup on the hierarchy specified by @hierarchy_id and
1859 * copy its path into @buf. This function grabs cgroup_mutex and shouldn't
1860 * be used inside locks used by cgroup controller callbacks.
1862 int task_cgroup_path_from_hierarchy(struct task_struct
*task
, int hierarchy_id
,
1863 char *buf
, size_t buflen
)
1865 struct cgroupfs_root
*root
;
1866 struct cgroup
*cgrp
= NULL
;
1869 mutex_lock(&cgroup_mutex
);
1871 root
= idr_find(&cgroup_hierarchy_idr
, hierarchy_id
);
1873 cgrp
= task_cgroup_from_root(task
, root
);
1874 ret
= cgroup_path(cgrp
, buf
, buflen
);
1877 mutex_unlock(&cgroup_mutex
);
1881 EXPORT_SYMBOL_GPL(task_cgroup_path_from_hierarchy
);
1884 * Control Group taskset
1886 struct task_and_cgroup
{
1887 struct task_struct
*task
;
1888 struct cgroup
*cgrp
;
1892 struct cgroup_taskset
{
1893 struct task_and_cgroup single
;
1894 struct flex_array
*tc_array
;
1897 struct cgroup
*cur_cgrp
;
1901 * cgroup_taskset_first - reset taskset and return the first task
1902 * @tset: taskset of interest
1904 * @tset iteration is initialized and the first task is returned.
1906 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1908 if (tset
->tc_array
) {
1910 return cgroup_taskset_next(tset
);
1912 tset
->cur_cgrp
= tset
->single
.cgrp
;
1913 return tset
->single
.task
;
1916 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1919 * cgroup_taskset_next - iterate to the next task in taskset
1920 * @tset: taskset of interest
1922 * Return the next task in @tset. Iteration must have been initialized
1923 * with cgroup_taskset_first().
1925 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1927 struct task_and_cgroup
*tc
;
1929 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1932 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1933 tset
->cur_cgrp
= tc
->cgrp
;
1936 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1939 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1940 * @tset: taskset of interest
1942 * Return the cgroup for the current (last returned) task of @tset. This
1943 * function must be preceded by either cgroup_taskset_first() or
1944 * cgroup_taskset_next().
1946 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1948 return tset
->cur_cgrp
;
1950 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1953 * cgroup_taskset_size - return the number of tasks in taskset
1954 * @tset: taskset of interest
1956 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1958 return tset
->tc_array
? tset
->tc_array_len
: 1;
1960 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1964 * cgroup_task_migrate - move a task from one cgroup to another.
1966 * Must be called with cgroup_mutex and threadgroup locked.
1968 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
1969 struct task_struct
*tsk
,
1970 struct css_set
*new_cset
)
1972 struct css_set
*old_cset
;
1975 * We are synchronized through threadgroup_lock() against PF_EXITING
1976 * setting such that we can't race against cgroup_exit() changing the
1977 * css_set to init_css_set and dropping the old one.
1979 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1980 old_cset
= task_css_set(tsk
);
1983 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
1986 /* Update the css_set linked lists if we're using them */
1987 write_lock(&css_set_lock
);
1988 if (!list_empty(&tsk
->cg_list
))
1989 list_move(&tsk
->cg_list
, &new_cset
->tasks
);
1990 write_unlock(&css_set_lock
);
1993 * We just gained a reference on old_cset by taking it from the
1994 * task. As trading it for new_cset is protected by cgroup_mutex,
1995 * we're safe to drop it here; it will be freed under RCU.
1997 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
1998 put_css_set(old_cset
);
2002 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2003 * @cgrp: the cgroup to attach to
2004 * @tsk: the task or the leader of the threadgroup to be attached
2005 * @threadgroup: attach the whole threadgroup?
2007 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2008 * task_lock of @tsk or each thread in the threadgroup individually in turn.
2010 static int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
,
2013 int retval
, i
, group_size
;
2014 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
2015 struct cgroupfs_root
*root
= cgrp
->root
;
2016 /* threadgroup list cursor and array */
2017 struct task_struct
*leader
= tsk
;
2018 struct task_and_cgroup
*tc
;
2019 struct flex_array
*group
;
2020 struct cgroup_taskset tset
= { };
2023 * step 0: in order to do expensive, possibly blocking operations for
2024 * every thread, we cannot iterate the thread group list, since it needs
2025 * rcu or tasklist locked. instead, build an array of all threads in the
2026 * group - group_rwsem prevents new threads from appearing, and if
2027 * threads exit, this will just be an over-estimate.
2030 group_size
= get_nr_threads(tsk
);
2033 /* flex_array supports very large thread-groups better than kmalloc. */
2034 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2037 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2038 retval
= flex_array_prealloc(group
, 0, group_size
, GFP_KERNEL
);
2040 goto out_free_group_list
;
2044 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2045 * already PF_EXITING could be freed from underneath us unless we
2046 * take an rcu_read_lock.
2050 struct task_and_cgroup ent
;
2052 /* @tsk either already exited or can't exit until the end */
2053 if (tsk
->flags
& PF_EXITING
)
2056 /* as per above, nr_threads may decrease, but not increase. */
2057 BUG_ON(i
>= group_size
);
2059 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2060 /* nothing to do if this task is already in the cgroup */
2061 if (ent
.cgrp
== cgrp
)
2064 * saying GFP_ATOMIC has no effect here because we did prealloc
2065 * earlier, but it's good form to communicate our expectations.
2067 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2068 BUG_ON(retval
!= 0);
2073 } while_each_thread(leader
, tsk
);
2075 /* remember the number of threads in the array for later. */
2077 tset
.tc_array
= group
;
2078 tset
.tc_array_len
= group_size
;
2080 /* methods shouldn't be called if no task is actually migrating */
2083 goto out_free_group_list
;
2086 * step 1: check that we can legitimately attach to the cgroup.
2088 for_each_root_subsys(root
, ss
) {
2089 if (ss
->can_attach
) {
2090 retval
= ss
->can_attach(cgrp
, &tset
);
2093 goto out_cancel_attach
;
2099 * step 2: make sure css_sets exist for all threads to be migrated.
2100 * we use find_css_set, which allocates a new one if necessary.
2102 for (i
= 0; i
< group_size
; i
++) {
2103 struct css_set
*old_cset
;
2105 tc
= flex_array_get(group
, i
);
2106 old_cset
= task_css_set(tc
->task
);
2107 tc
->cg
= find_css_set(old_cset
, cgrp
);
2110 goto out_put_css_set_refs
;
2115 * step 3: now that we're guaranteed success wrt the css_sets,
2116 * proceed to move all tasks to the new cgroup. There are no
2117 * failure cases after here, so this is the commit point.
2119 for (i
= 0; i
< group_size
; i
++) {
2120 tc
= flex_array_get(group
, i
);
2121 cgroup_task_migrate(tc
->cgrp
, tc
->task
, tc
->cg
);
2123 /* nothing is sensitive to fork() after this point. */
2126 * step 4: do subsystem attach callbacks.
2128 for_each_root_subsys(root
, ss
) {
2130 ss
->attach(cgrp
, &tset
);
2134 * step 5: success! and cleanup
2137 out_put_css_set_refs
:
2139 for (i
= 0; i
< group_size
; i
++) {
2140 tc
= flex_array_get(group
, i
);
2143 put_css_set(tc
->cg
);
2148 for_each_root_subsys(root
, ss
) {
2149 if (ss
== failed_ss
)
2151 if (ss
->cancel_attach
)
2152 ss
->cancel_attach(cgrp
, &tset
);
2155 out_free_group_list
:
2156 flex_array_free(group
);
2161 * Find the task_struct of the task to attach by vpid and pass it along to the
2162 * function to attach either it or all tasks in its threadgroup. Will lock
2163 * cgroup_mutex and threadgroup; may take task_lock of task.
2165 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2167 struct task_struct
*tsk
;
2168 const struct cred
*cred
= current_cred(), *tcred
;
2171 if (!cgroup_lock_live_group(cgrp
))
2177 tsk
= find_task_by_vpid(pid
);
2181 goto out_unlock_cgroup
;
2184 * even if we're attaching all tasks in the thread group, we
2185 * only need to check permissions on one of them.
2187 tcred
= __task_cred(tsk
);
2188 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2189 !uid_eq(cred
->euid
, tcred
->uid
) &&
2190 !uid_eq(cred
->euid
, tcred
->suid
)) {
2193 goto out_unlock_cgroup
;
2199 tsk
= tsk
->group_leader
;
2202 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2203 * trapped in a cpuset, or RT worker may be born in a cgroup
2204 * with no rt_runtime allocated. Just say no.
2206 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2209 goto out_unlock_cgroup
;
2212 get_task_struct(tsk
);
2215 threadgroup_lock(tsk
);
2217 if (!thread_group_leader(tsk
)) {
2219 * a race with de_thread from another thread's exec()
2220 * may strip us of our leadership, if this happens,
2221 * there is no choice but to throw this task away and
2222 * try again; this is
2223 * "double-double-toil-and-trouble-check locking".
2225 threadgroup_unlock(tsk
);
2226 put_task_struct(tsk
);
2227 goto retry_find_task
;
2231 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2233 threadgroup_unlock(tsk
);
2235 put_task_struct(tsk
);
2237 mutex_unlock(&cgroup_mutex
);
2242 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2243 * @from: attach to all cgroups of a given task
2244 * @tsk: the task to be attached
2246 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2248 struct cgroupfs_root
*root
;
2251 mutex_lock(&cgroup_mutex
);
2252 for_each_active_root(root
) {
2253 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
2255 retval
= cgroup_attach_task(from_cg
, tsk
, false);
2259 mutex_unlock(&cgroup_mutex
);
2263 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2265 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2267 return attach_task_by_pid(cgrp
, pid
, false);
2270 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2272 return attach_task_by_pid(cgrp
, tgid
, true);
2275 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2278 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2279 if (strlen(buffer
) >= PATH_MAX
)
2281 if (!cgroup_lock_live_group(cgrp
))
2283 mutex_lock(&cgroup_root_mutex
);
2284 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2285 mutex_unlock(&cgroup_root_mutex
);
2286 mutex_unlock(&cgroup_mutex
);
2290 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2291 struct seq_file
*seq
)
2293 if (!cgroup_lock_live_group(cgrp
))
2295 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2296 seq_putc(seq
, '\n');
2297 mutex_unlock(&cgroup_mutex
);
2301 static int cgroup_sane_behavior_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2302 struct seq_file
*seq
)
2304 seq_printf(seq
, "%d\n", cgroup_sane_behavior(cgrp
));
2308 /* A buffer size big enough for numbers or short strings */
2309 #define CGROUP_LOCAL_BUFFER_SIZE 64
2311 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2313 const char __user
*userbuf
,
2314 size_t nbytes
, loff_t
*unused_ppos
)
2316 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2322 if (nbytes
>= sizeof(buffer
))
2324 if (copy_from_user(buffer
, userbuf
, nbytes
))
2327 buffer
[nbytes
] = 0; /* nul-terminate */
2328 if (cft
->write_u64
) {
2329 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2332 retval
= cft
->write_u64(cgrp
, cft
, val
);
2334 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2337 retval
= cft
->write_s64(cgrp
, cft
, val
);
2344 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2346 const char __user
*userbuf
,
2347 size_t nbytes
, loff_t
*unused_ppos
)
2349 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2351 size_t max_bytes
= cft
->max_write_len
;
2352 char *buffer
= local_buffer
;
2355 max_bytes
= sizeof(local_buffer
) - 1;
2356 if (nbytes
>= max_bytes
)
2358 /* Allocate a dynamic buffer if we need one */
2359 if (nbytes
>= sizeof(local_buffer
)) {
2360 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2364 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2369 buffer
[nbytes
] = 0; /* nul-terminate */
2370 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2374 if (buffer
!= local_buffer
)
2379 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2380 size_t nbytes
, loff_t
*ppos
)
2382 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2383 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2385 if (cgroup_is_dead(cgrp
))
2388 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2389 if (cft
->write_u64
|| cft
->write_s64
)
2390 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2391 if (cft
->write_string
)
2392 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2394 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2395 return ret
? ret
: nbytes
;
2400 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2402 char __user
*buf
, size_t nbytes
,
2405 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2406 u64 val
= cft
->read_u64(cgrp
, cft
);
2407 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2409 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2412 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2414 char __user
*buf
, size_t nbytes
,
2417 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2418 s64 val
= cft
->read_s64(cgrp
, cft
);
2419 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2421 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2424 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2425 size_t nbytes
, loff_t
*ppos
)
2427 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2428 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2430 if (cgroup_is_dead(cgrp
))
2434 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2436 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2438 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2443 * seqfile ops/methods for returning structured data. Currently just
2444 * supports string->u64 maps, but can be extended in future.
2447 struct cgroup_seqfile_state
{
2449 struct cgroup
*cgroup
;
2452 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2454 struct seq_file
*sf
= cb
->state
;
2455 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2458 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2460 struct cgroup_seqfile_state
*state
= m
->private;
2461 struct cftype
*cft
= state
->cft
;
2462 if (cft
->read_map
) {
2463 struct cgroup_map_cb cb
= {
2464 .fill
= cgroup_map_add
,
2467 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2469 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2472 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2474 struct seq_file
*seq
= file
->private_data
;
2475 kfree(seq
->private);
2476 return single_release(inode
, file
);
2479 static const struct file_operations cgroup_seqfile_operations
= {
2481 .write
= cgroup_file_write
,
2482 .llseek
= seq_lseek
,
2483 .release
= cgroup_seqfile_release
,
2486 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2491 err
= generic_file_open(inode
, file
);
2494 cft
= __d_cft(file
->f_dentry
);
2496 if (cft
->read_map
|| cft
->read_seq_string
) {
2497 struct cgroup_seqfile_state
*state
;
2499 state
= kzalloc(sizeof(*state
), GFP_USER
);
2504 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2505 file
->f_op
= &cgroup_seqfile_operations
;
2506 err
= single_open(file
, cgroup_seqfile_show
, state
);
2509 } else if (cft
->open
)
2510 err
= cft
->open(inode
, file
);
2517 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2519 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2521 return cft
->release(inode
, file
);
2526 * cgroup_rename - Only allow simple rename of directories in place.
2528 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2529 struct inode
*new_dir
, struct dentry
*new_dentry
)
2532 struct cgroup_name
*name
, *old_name
;
2533 struct cgroup
*cgrp
;
2536 * It's convinient to use parent dir's i_mutex to protected
2539 lockdep_assert_held(&old_dir
->i_mutex
);
2541 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2543 if (new_dentry
->d_inode
)
2545 if (old_dir
!= new_dir
)
2548 cgrp
= __d_cgrp(old_dentry
);
2551 * This isn't a proper migration and its usefulness is very
2552 * limited. Disallow if sane_behavior.
2554 if (cgroup_sane_behavior(cgrp
))
2557 name
= cgroup_alloc_name(new_dentry
);
2561 ret
= simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2567 old_name
= rcu_dereference_protected(cgrp
->name
, true);
2568 rcu_assign_pointer(cgrp
->name
, name
);
2570 kfree_rcu(old_name
, rcu_head
);
2574 static struct simple_xattrs
*__d_xattrs(struct dentry
*dentry
)
2576 if (S_ISDIR(dentry
->d_inode
->i_mode
))
2577 return &__d_cgrp(dentry
)->xattrs
;
2579 return &__d_cfe(dentry
)->xattrs
;
2582 static inline int xattr_enabled(struct dentry
*dentry
)
2584 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
2585 return root
->flags
& CGRP_ROOT_XATTR
;
2588 static bool is_valid_xattr(const char *name
)
2590 if (!strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) ||
2591 !strncmp(name
, XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
))
2596 static int cgroup_setxattr(struct dentry
*dentry
, const char *name
,
2597 const void *val
, size_t size
, int flags
)
2599 if (!xattr_enabled(dentry
))
2601 if (!is_valid_xattr(name
))
2603 return simple_xattr_set(__d_xattrs(dentry
), name
, val
, size
, flags
);
2606 static int cgroup_removexattr(struct dentry
*dentry
, const char *name
)
2608 if (!xattr_enabled(dentry
))
2610 if (!is_valid_xattr(name
))
2612 return simple_xattr_remove(__d_xattrs(dentry
), name
);
2615 static ssize_t
cgroup_getxattr(struct dentry
*dentry
, const char *name
,
2616 void *buf
, size_t size
)
2618 if (!xattr_enabled(dentry
))
2620 if (!is_valid_xattr(name
))
2622 return simple_xattr_get(__d_xattrs(dentry
), name
, buf
, size
);
2625 static ssize_t
cgroup_listxattr(struct dentry
*dentry
, char *buf
, size_t size
)
2627 if (!xattr_enabled(dentry
))
2629 return simple_xattr_list(__d_xattrs(dentry
), buf
, size
);
2632 static const struct file_operations cgroup_file_operations
= {
2633 .read
= cgroup_file_read
,
2634 .write
= cgroup_file_write
,
2635 .llseek
= generic_file_llseek
,
2636 .open
= cgroup_file_open
,
2637 .release
= cgroup_file_release
,
2640 static const struct inode_operations cgroup_file_inode_operations
= {
2641 .setxattr
= cgroup_setxattr
,
2642 .getxattr
= cgroup_getxattr
,
2643 .listxattr
= cgroup_listxattr
,
2644 .removexattr
= cgroup_removexattr
,
2647 static const struct inode_operations cgroup_dir_inode_operations
= {
2648 .lookup
= cgroup_lookup
,
2649 .mkdir
= cgroup_mkdir
,
2650 .rmdir
= cgroup_rmdir
,
2651 .rename
= cgroup_rename
,
2652 .setxattr
= cgroup_setxattr
,
2653 .getxattr
= cgroup_getxattr
,
2654 .listxattr
= cgroup_listxattr
,
2655 .removexattr
= cgroup_removexattr
,
2658 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
2660 if (dentry
->d_name
.len
> NAME_MAX
)
2661 return ERR_PTR(-ENAMETOOLONG
);
2662 d_add(dentry
, NULL
);
2667 * Check if a file is a control file
2669 static inline struct cftype
*__file_cft(struct file
*file
)
2671 if (file_inode(file
)->i_fop
!= &cgroup_file_operations
)
2672 return ERR_PTR(-EINVAL
);
2673 return __d_cft(file
->f_dentry
);
2676 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2677 struct super_block
*sb
)
2679 struct inode
*inode
;
2683 if (dentry
->d_inode
)
2686 inode
= cgroup_new_inode(mode
, sb
);
2690 if (S_ISDIR(mode
)) {
2691 inode
->i_op
= &cgroup_dir_inode_operations
;
2692 inode
->i_fop
= &simple_dir_operations
;
2694 /* start off with i_nlink == 2 (for "." entry) */
2696 inc_nlink(dentry
->d_parent
->d_inode
);
2699 * Control reaches here with cgroup_mutex held.
2700 * @inode->i_mutex should nest outside cgroup_mutex but we
2701 * want to populate it immediately without releasing
2702 * cgroup_mutex. As @inode isn't visible to anyone else
2703 * yet, trylock will always succeed without affecting
2706 WARN_ON_ONCE(!mutex_trylock(&inode
->i_mutex
));
2707 } else if (S_ISREG(mode
)) {
2709 inode
->i_fop
= &cgroup_file_operations
;
2710 inode
->i_op
= &cgroup_file_inode_operations
;
2712 d_instantiate(dentry
, inode
);
2713 dget(dentry
); /* Extra count - pin the dentry in core */
2718 * cgroup_file_mode - deduce file mode of a control file
2719 * @cft: the control file in question
2721 * returns cft->mode if ->mode is not 0
2722 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2723 * returns S_IRUGO if it has only a read handler
2724 * returns S_IWUSR if it has only a write hander
2726 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2733 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2734 cft
->read_map
|| cft
->read_seq_string
)
2737 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2738 cft
->write_string
|| cft
->trigger
)
2744 static int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2747 struct dentry
*dir
= cgrp
->dentry
;
2748 struct cgroup
*parent
= __d_cgrp(dir
);
2749 struct dentry
*dentry
;
2753 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2755 if (subsys
&& !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
2756 strcpy(name
, subsys
->name
);
2759 strcat(name
, cft
->name
);
2761 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2763 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2767 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2768 if (IS_ERR(dentry
)) {
2769 error
= PTR_ERR(dentry
);
2773 cfe
->type
= (void *)cft
;
2774 cfe
->dentry
= dentry
;
2775 dentry
->d_fsdata
= cfe
;
2776 simple_xattrs_init(&cfe
->xattrs
);
2778 mode
= cgroup_file_mode(cft
);
2779 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2781 list_add_tail(&cfe
->node
, &parent
->files
);
2791 * cgroup_addrm_files - add or remove files to a cgroup directory
2792 * @cgrp: the target cgroup
2793 * @subsys: the subsystem of files to be added
2794 * @cfts: array of cftypes to be added
2795 * @is_add: whether to add or remove
2797 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2798 * All @cfts should belong to @subsys. For removals, this function never
2799 * fails. If addition fails, this function doesn't remove files already
2800 * added. The caller is responsible for cleaning up.
2802 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2803 struct cftype cfts
[], bool is_add
)
2808 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
2809 lockdep_assert_held(&cgroup_mutex
);
2811 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2812 /* does cft->flags tell us to skip this file on @cgrp? */
2813 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_sane_behavior(cgrp
))
2815 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2817 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2821 ret
= cgroup_add_file(cgrp
, subsys
, cft
);
2823 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2828 cgroup_rm_file(cgrp
, cft
);
2834 static void cgroup_cfts_prepare(void)
2835 __acquires(&cgroup_mutex
)
2838 * Thanks to the entanglement with vfs inode locking, we can't walk
2839 * the existing cgroups under cgroup_mutex and create files.
2840 * Instead, we use cgroup_for_each_descendant_pre() and drop RCU
2841 * read lock before calling cgroup_addrm_files().
2843 mutex_lock(&cgroup_mutex
);
2846 static int cgroup_cfts_commit(struct cgroup_subsys
*ss
,
2847 struct cftype
*cfts
, bool is_add
)
2848 __releases(&cgroup_mutex
)
2851 struct cgroup
*cgrp
, *root
= &ss
->root
->top_cgroup
;
2852 struct super_block
*sb
= ss
->root
->sb
;
2853 struct dentry
*prev
= NULL
;
2854 struct inode
*inode
;
2858 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2859 if (!cfts
|| ss
->root
== &cgroup_dummy_root
||
2860 !atomic_inc_not_zero(&sb
->s_active
)) {
2861 mutex_unlock(&cgroup_mutex
);
2866 * All cgroups which are created after we drop cgroup_mutex will
2867 * have the updated set of files, so we only need to update the
2868 * cgroups created before the current @cgroup_serial_nr_next.
2870 update_before
= cgroup_serial_nr_next
;
2872 mutex_unlock(&cgroup_mutex
);
2874 /* @root always needs to be updated */
2875 inode
= root
->dentry
->d_inode
;
2876 mutex_lock(&inode
->i_mutex
);
2877 mutex_lock(&cgroup_mutex
);
2878 ret
= cgroup_addrm_files(root
, ss
, cfts
, is_add
);
2879 mutex_unlock(&cgroup_mutex
);
2880 mutex_unlock(&inode
->i_mutex
);
2885 /* add/rm files for all cgroups created before */
2887 cgroup_for_each_descendant_pre(cgrp
, root
) {
2888 if (cgroup_is_dead(cgrp
))
2891 inode
= cgrp
->dentry
->d_inode
;
2896 prev
= cgrp
->dentry
;
2898 mutex_lock(&inode
->i_mutex
);
2899 mutex_lock(&cgroup_mutex
);
2900 if (cgrp
->serial_nr
< update_before
&& !cgroup_is_dead(cgrp
))
2901 ret
= cgroup_addrm_files(cgrp
, ss
, cfts
, is_add
);
2902 mutex_unlock(&cgroup_mutex
);
2903 mutex_unlock(&inode
->i_mutex
);
2912 deactivate_super(sb
);
2917 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2918 * @ss: target cgroup subsystem
2919 * @cfts: zero-length name terminated array of cftypes
2921 * Register @cfts to @ss. Files described by @cfts are created for all
2922 * existing cgroups to which @ss is attached and all future cgroups will
2923 * have them too. This function can be called anytime whether @ss is
2926 * Returns 0 on successful registration, -errno on failure. Note that this
2927 * function currently returns 0 as long as @cfts registration is successful
2928 * even if some file creation attempts on existing cgroups fail.
2930 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2932 struct cftype_set
*set
;
2935 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2939 cgroup_cfts_prepare();
2941 list_add_tail(&set
->node
, &ss
->cftsets
);
2942 ret
= cgroup_cfts_commit(ss
, cfts
, true);
2944 cgroup_rm_cftypes(ss
, cfts
);
2947 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2950 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2951 * @ss: target cgroup subsystem
2952 * @cfts: zero-length name terminated array of cftypes
2954 * Unregister @cfts from @ss. Files described by @cfts are removed from
2955 * all existing cgroups to which @ss is attached and all future cgroups
2956 * won't have them either. This function can be called anytime whether @ss
2957 * is attached or not.
2959 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2960 * registered with @ss.
2962 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2964 struct cftype_set
*set
;
2966 cgroup_cfts_prepare();
2968 list_for_each_entry(set
, &ss
->cftsets
, node
) {
2969 if (set
->cfts
== cfts
) {
2970 list_del(&set
->node
);
2972 cgroup_cfts_commit(ss
, cfts
, false);
2977 cgroup_cfts_commit(ss
, NULL
, false);
2982 * cgroup_task_count - count the number of tasks in a cgroup.
2983 * @cgrp: the cgroup in question
2985 * Return the number of tasks in the cgroup.
2987 int cgroup_task_count(const struct cgroup
*cgrp
)
2990 struct cgrp_cset_link
*link
;
2992 read_lock(&css_set_lock
);
2993 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
2994 count
+= atomic_read(&link
->cset
->refcount
);
2995 read_unlock(&css_set_lock
);
3000 * Advance a list_head iterator. The iterator should be positioned at
3001 * the start of a css_set
3003 static void cgroup_advance_iter(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3005 struct list_head
*l
= it
->cset_link
;
3006 struct cgrp_cset_link
*link
;
3007 struct css_set
*cset
;
3009 /* Advance to the next non-empty css_set */
3012 if (l
== &cgrp
->cset_links
) {
3013 it
->cset_link
= NULL
;
3016 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3018 } while (list_empty(&cset
->tasks
));
3020 it
->task
= cset
->tasks
.next
;
3024 * To reduce the fork() overhead for systems that are not actually
3025 * using their cgroups capability, we don't maintain the lists running
3026 * through each css_set to its tasks until we see the list actually
3027 * used - in other words after the first call to cgroup_iter_start().
3029 static void cgroup_enable_task_cg_lists(void)
3031 struct task_struct
*p
, *g
;
3032 write_lock(&css_set_lock
);
3033 use_task_css_set_links
= 1;
3035 * We need tasklist_lock because RCU is not safe against
3036 * while_each_thread(). Besides, a forking task that has passed
3037 * cgroup_post_fork() without seeing use_task_css_set_links = 1
3038 * is not guaranteed to have its child immediately visible in the
3039 * tasklist if we walk through it with RCU.
3041 read_lock(&tasklist_lock
);
3042 do_each_thread(g
, p
) {
3045 * We should check if the process is exiting, otherwise
3046 * it will race with cgroup_exit() in that the list
3047 * entry won't be deleted though the process has exited.
3049 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
3050 list_add(&p
->cg_list
, &task_css_set(p
)->tasks
);
3052 } while_each_thread(g
, p
);
3053 read_unlock(&tasklist_lock
);
3054 write_unlock(&css_set_lock
);
3058 * cgroup_next_sibling - find the next sibling of a given cgroup
3059 * @pos: the current cgroup
3061 * This function returns the next sibling of @pos and should be called
3062 * under RCU read lock. The only requirement is that @pos is accessible.
3063 * The next sibling is guaranteed to be returned regardless of @pos's
3066 struct cgroup
*cgroup_next_sibling(struct cgroup
*pos
)
3068 struct cgroup
*next
;
3070 WARN_ON_ONCE(!rcu_read_lock_held());
3073 * @pos could already have been removed. Once a cgroup is removed,
3074 * its ->sibling.next is no longer updated when its next sibling
3075 * changes. As CGRP_DEAD assertion is serialized and happens
3076 * before the cgroup is taken off the ->sibling list, if we see it
3077 * unasserted, it's guaranteed that the next sibling hasn't
3078 * finished its grace period even if it's already removed, and thus
3079 * safe to dereference from this RCU critical section. If
3080 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
3081 * to be visible as %true here.
3083 if (likely(!cgroup_is_dead(pos
))) {
3084 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
, sibling
);
3085 if (&next
->sibling
!= &pos
->parent
->children
)
3091 * Can't dereference the next pointer. Each cgroup is given a
3092 * monotonically increasing unique serial number and always
3093 * appended to the sibling list, so the next one can be found by
3094 * walking the parent's children until we see a cgroup with higher
3095 * serial number than @pos's.
3097 * While this path can be slow, it's taken only when either the
3098 * current cgroup is removed or iteration and removal race.
3100 list_for_each_entry_rcu(next
, &pos
->parent
->children
, sibling
)
3101 if (next
->serial_nr
> pos
->serial_nr
)
3105 EXPORT_SYMBOL_GPL(cgroup_next_sibling
);
3108 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
3109 * @pos: the current position (%NULL to initiate traversal)
3110 * @cgroup: cgroup whose descendants to walk
3112 * To be used by cgroup_for_each_descendant_pre(). Find the next
3113 * descendant to visit for pre-order traversal of @cgroup's descendants.
3115 * While this function requires RCU read locking, it doesn't require the
3116 * whole traversal to be contained in a single RCU critical section. This
3117 * function will return the correct next descendant as long as both @pos
3118 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3120 struct cgroup
*cgroup_next_descendant_pre(struct cgroup
*pos
,
3121 struct cgroup
*cgroup
)
3123 struct cgroup
*next
;
3125 WARN_ON_ONCE(!rcu_read_lock_held());
3127 /* if first iteration, pretend we just visited @cgroup */
3131 /* visit the first child if exists */
3132 next
= list_first_or_null_rcu(&pos
->children
, struct cgroup
, sibling
);
3136 /* no child, visit my or the closest ancestor's next sibling */
3137 while (pos
!= cgroup
) {
3138 next
= cgroup_next_sibling(pos
);
3146 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre
);
3149 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
3150 * @pos: cgroup of interest
3152 * Return the rightmost descendant of @pos. If there's no descendant,
3153 * @pos is returned. This can be used during pre-order traversal to skip
3156 * While this function requires RCU read locking, it doesn't require the
3157 * whole traversal to be contained in a single RCU critical section. This
3158 * function will return the correct rightmost descendant as long as @pos is
3161 struct cgroup
*cgroup_rightmost_descendant(struct cgroup
*pos
)
3163 struct cgroup
*last
, *tmp
;
3165 WARN_ON_ONCE(!rcu_read_lock_held());
3169 /* ->prev isn't RCU safe, walk ->next till the end */
3171 list_for_each_entry_rcu(tmp
, &last
->children
, sibling
)
3177 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant
);
3179 static struct cgroup
*cgroup_leftmost_descendant(struct cgroup
*pos
)
3181 struct cgroup
*last
;
3185 pos
= list_first_or_null_rcu(&pos
->children
, struct cgroup
,
3193 * cgroup_next_descendant_post - find the next descendant for post-order walk
3194 * @pos: the current position (%NULL to initiate traversal)
3195 * @cgroup: cgroup whose descendants to walk
3197 * To be used by cgroup_for_each_descendant_post(). Find the next
3198 * descendant to visit for post-order traversal of @cgroup's descendants.
3200 * While this function requires RCU read locking, it doesn't require the
3201 * whole traversal to be contained in a single RCU critical section. This
3202 * function will return the correct next descendant as long as both @pos
3203 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3205 struct cgroup
*cgroup_next_descendant_post(struct cgroup
*pos
,
3206 struct cgroup
*cgroup
)
3208 struct cgroup
*next
;
3210 WARN_ON_ONCE(!rcu_read_lock_held());
3212 /* if first iteration, visit the leftmost descendant */
3214 next
= cgroup_leftmost_descendant(cgroup
);
3215 return next
!= cgroup
? next
: NULL
;
3218 /* if there's an unvisited sibling, visit its leftmost descendant */
3219 next
= cgroup_next_sibling(pos
);
3221 return cgroup_leftmost_descendant(next
);
3223 /* no sibling left, visit parent */
3225 return next
!= cgroup
? next
: NULL
;
3227 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post
);
3229 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3230 __acquires(css_set_lock
)
3233 * The first time anyone tries to iterate across a cgroup,
3234 * we need to enable the list linking each css_set to its
3235 * tasks, and fix up all existing tasks.
3237 if (!use_task_css_set_links
)
3238 cgroup_enable_task_cg_lists();
3240 read_lock(&css_set_lock
);
3241 it
->cset_link
= &cgrp
->cset_links
;
3242 cgroup_advance_iter(cgrp
, it
);
3245 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
3246 struct cgroup_iter
*it
)
3248 struct task_struct
*res
;
3249 struct list_head
*l
= it
->task
;
3250 struct cgrp_cset_link
*link
;
3252 /* If the iterator cg is NULL, we have no tasks */
3255 res
= list_entry(l
, struct task_struct
, cg_list
);
3256 /* Advance iterator to find next entry */
3258 link
= list_entry(it
->cset_link
, struct cgrp_cset_link
, cset_link
);
3259 if (l
== &link
->cset
->tasks
) {
3260 /* We reached the end of this task list - move on to
3261 * the next cg_cgroup_link */
3262 cgroup_advance_iter(cgrp
, it
);
3269 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3270 __releases(css_set_lock
)
3272 read_unlock(&css_set_lock
);
3275 static inline int started_after_time(struct task_struct
*t1
,
3276 struct timespec
*time
,
3277 struct task_struct
*t2
)
3279 int start_diff
= timespec_compare(&t1
->start_time
, time
);
3280 if (start_diff
> 0) {
3282 } else if (start_diff
< 0) {
3286 * Arbitrarily, if two processes started at the same
3287 * time, we'll say that the lower pointer value
3288 * started first. Note that t2 may have exited by now
3289 * so this may not be a valid pointer any longer, but
3290 * that's fine - it still serves to distinguish
3291 * between two tasks started (effectively) simultaneously.
3298 * This function is a callback from heap_insert() and is used to order
3300 * In this case we order the heap in descending task start time.
3302 static inline int started_after(void *p1
, void *p2
)
3304 struct task_struct
*t1
= p1
;
3305 struct task_struct
*t2
= p2
;
3306 return started_after_time(t1
, &t2
->start_time
, t2
);
3310 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3311 * @scan: struct cgroup_scanner containing arguments for the scan
3313 * Arguments include pointers to callback functions test_task() and
3315 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3316 * and if it returns true, call process_task() for it also.
3317 * The test_task pointer may be NULL, meaning always true (select all tasks).
3318 * Effectively duplicates cgroup_iter_{start,next,end}()
3319 * but does not lock css_set_lock for the call to process_task().
3320 * The struct cgroup_scanner may be embedded in any structure of the caller's
3322 * It is guaranteed that process_task() will act on every task that
3323 * is a member of the cgroup for the duration of this call. This
3324 * function may or may not call process_task() for tasks that exit
3325 * or move to a different cgroup during the call, or are forked or
3326 * move into the cgroup during the call.
3328 * Note that test_task() may be called with locks held, and may in some
3329 * situations be called multiple times for the same task, so it should
3331 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3332 * pre-allocated and will be used for heap operations (and its "gt" member will
3333 * be overwritten), else a temporary heap will be used (allocation of which
3334 * may cause this function to fail).
3336 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
3339 struct cgroup_iter it
;
3340 struct task_struct
*p
, *dropped
;
3341 /* Never dereference latest_task, since it's not refcounted */
3342 struct task_struct
*latest_task
= NULL
;
3343 struct ptr_heap tmp_heap
;
3344 struct ptr_heap
*heap
;
3345 struct timespec latest_time
= { 0, 0 };
3348 /* The caller supplied our heap and pre-allocated its memory */
3350 heap
->gt
= &started_after
;
3352 /* We need to allocate our own heap memory */
3354 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3356 /* cannot allocate the heap */
3362 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3363 * to determine which are of interest, and using the scanner's
3364 * "process_task" callback to process any of them that need an update.
3365 * Since we don't want to hold any locks during the task updates,
3366 * gather tasks to be processed in a heap structure.
3367 * The heap is sorted by descending task start time.
3368 * If the statically-sized heap fills up, we overflow tasks that
3369 * started later, and in future iterations only consider tasks that
3370 * started after the latest task in the previous pass. This
3371 * guarantees forward progress and that we don't miss any tasks.
3374 cgroup_iter_start(scan
->cg
, &it
);
3375 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
3377 * Only affect tasks that qualify per the caller's callback,
3378 * if he provided one
3380 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
3383 * Only process tasks that started after the last task
3386 if (!started_after_time(p
, &latest_time
, latest_task
))
3388 dropped
= heap_insert(heap
, p
);
3389 if (dropped
== NULL
) {
3391 * The new task was inserted; the heap wasn't
3395 } else if (dropped
!= p
) {
3397 * The new task was inserted, and pushed out a
3401 put_task_struct(dropped
);
3404 * Else the new task was newer than anything already in
3405 * the heap and wasn't inserted
3408 cgroup_iter_end(scan
->cg
, &it
);
3411 for (i
= 0; i
< heap
->size
; i
++) {
3412 struct task_struct
*q
= heap
->ptrs
[i
];
3414 latest_time
= q
->start_time
;
3417 /* Process the task per the caller's callback */
3418 scan
->process_task(q
, scan
);
3422 * If we had to process any tasks at all, scan again
3423 * in case some of them were in the middle of forking
3424 * children that didn't get processed.
3425 * Not the most efficient way to do it, but it avoids
3426 * having to take callback_mutex in the fork path
3430 if (heap
== &tmp_heap
)
3431 heap_free(&tmp_heap
);
3435 static void cgroup_transfer_one_task(struct task_struct
*task
,
3436 struct cgroup_scanner
*scan
)
3438 struct cgroup
*new_cgroup
= scan
->data
;
3440 mutex_lock(&cgroup_mutex
);
3441 cgroup_attach_task(new_cgroup
, task
, false);
3442 mutex_unlock(&cgroup_mutex
);
3446 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3447 * @to: cgroup to which the tasks will be moved
3448 * @from: cgroup in which the tasks currently reside
3450 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3452 struct cgroup_scanner scan
;
3455 scan
.test_task
= NULL
; /* select all tasks in cgroup */
3456 scan
.process_task
= cgroup_transfer_one_task
;
3460 return cgroup_scan_tasks(&scan
);
3464 * Stuff for reading the 'tasks'/'procs' files.
3466 * Reading this file can return large amounts of data if a cgroup has
3467 * *lots* of attached tasks. So it may need several calls to read(),
3468 * but we cannot guarantee that the information we produce is correct
3469 * unless we produce it entirely atomically.
3473 /* which pidlist file are we talking about? */
3474 enum cgroup_filetype
{
3480 * A pidlist is a list of pids that virtually represents the contents of one
3481 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3482 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3485 struct cgroup_pidlist
{
3487 * used to find which pidlist is wanted. doesn't change as long as
3488 * this particular list stays in the list.
3490 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3493 /* how many elements the above list has */
3495 /* how many files are using the current array */
3497 /* each of these stored in a list by its cgroup */
3498 struct list_head links
;
3499 /* pointer to the cgroup we belong to, for list removal purposes */
3500 struct cgroup
*owner
;
3501 /* protects the other fields */
3502 struct rw_semaphore mutex
;
3506 * The following two functions "fix" the issue where there are more pids
3507 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3508 * TODO: replace with a kernel-wide solution to this problem
3510 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3511 static void *pidlist_allocate(int count
)
3513 if (PIDLIST_TOO_LARGE(count
))
3514 return vmalloc(count
* sizeof(pid_t
));
3516 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3518 static void pidlist_free(void *p
)
3520 if (is_vmalloc_addr(p
))
3527 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3528 * Returns the number of unique elements.
3530 static int pidlist_uniq(pid_t
*list
, int length
)
3535 * we presume the 0th element is unique, so i starts at 1. trivial
3536 * edge cases first; no work needs to be done for either
3538 if (length
== 0 || length
== 1)
3540 /* src and dest walk down the list; dest counts unique elements */
3541 for (src
= 1; src
< length
; src
++) {
3542 /* find next unique element */
3543 while (list
[src
] == list
[src
-1]) {
3548 /* dest always points to where the next unique element goes */
3549 list
[dest
] = list
[src
];
3556 static int cmppid(const void *a
, const void *b
)
3558 return *(pid_t
*)a
- *(pid_t
*)b
;
3562 * find the appropriate pidlist for our purpose (given procs vs tasks)
3563 * returns with the lock on that pidlist already held, and takes care
3564 * of the use count, or returns NULL with no locks held if we're out of
3567 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3568 enum cgroup_filetype type
)
3570 struct cgroup_pidlist
*l
;
3571 /* don't need task_nsproxy() if we're looking at ourself */
3572 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3575 * We can't drop the pidlist_mutex before taking the l->mutex in case
3576 * the last ref-holder is trying to remove l from the list at the same
3577 * time. Holding the pidlist_mutex precludes somebody taking whichever
3578 * list we find out from under us - compare release_pid_array().
3580 mutex_lock(&cgrp
->pidlist_mutex
);
3581 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3582 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3583 /* make sure l doesn't vanish out from under us */
3584 down_write(&l
->mutex
);
3585 mutex_unlock(&cgrp
->pidlist_mutex
);
3589 /* entry not found; create a new one */
3590 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3592 mutex_unlock(&cgrp
->pidlist_mutex
);
3595 init_rwsem(&l
->mutex
);
3596 down_write(&l
->mutex
);
3598 l
->key
.ns
= get_pid_ns(ns
);
3600 list_add(&l
->links
, &cgrp
->pidlists
);
3601 mutex_unlock(&cgrp
->pidlist_mutex
);
3606 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3608 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3609 struct cgroup_pidlist
**lp
)
3613 int pid
, n
= 0; /* used for populating the array */
3614 struct cgroup_iter it
;
3615 struct task_struct
*tsk
;
3616 struct cgroup_pidlist
*l
;
3619 * If cgroup gets more users after we read count, we won't have
3620 * enough space - tough. This race is indistinguishable to the
3621 * caller from the case that the additional cgroup users didn't
3622 * show up until sometime later on.
3624 length
= cgroup_task_count(cgrp
);
3625 array
= pidlist_allocate(length
);
3628 /* now, populate the array */
3629 cgroup_iter_start(cgrp
, &it
);
3630 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3631 if (unlikely(n
== length
))
3633 /* get tgid or pid for procs or tasks file respectively */
3634 if (type
== CGROUP_FILE_PROCS
)
3635 pid
= task_tgid_vnr(tsk
);
3637 pid
= task_pid_vnr(tsk
);
3638 if (pid
> 0) /* make sure to only use valid results */
3641 cgroup_iter_end(cgrp
, &it
);
3643 /* now sort & (if procs) strip out duplicates */
3644 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3645 if (type
== CGROUP_FILE_PROCS
)
3646 length
= pidlist_uniq(array
, length
);
3647 l
= cgroup_pidlist_find(cgrp
, type
);
3649 pidlist_free(array
);
3652 /* store array, freeing old if necessary - lock already held */
3653 pidlist_free(l
->list
);
3657 up_write(&l
->mutex
);
3663 * cgroupstats_build - build and fill cgroupstats
3664 * @stats: cgroupstats to fill information into
3665 * @dentry: A dentry entry belonging to the cgroup for which stats have
3668 * Build and fill cgroupstats so that taskstats can export it to user
3671 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3674 struct cgroup
*cgrp
;
3675 struct cgroup_iter it
;
3676 struct task_struct
*tsk
;
3679 * Validate dentry by checking the superblock operations,
3680 * and make sure it's a directory.
3682 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3683 !S_ISDIR(dentry
->d_inode
->i_mode
))
3687 cgrp
= dentry
->d_fsdata
;
3689 cgroup_iter_start(cgrp
, &it
);
3690 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3691 switch (tsk
->state
) {
3693 stats
->nr_running
++;
3695 case TASK_INTERRUPTIBLE
:
3696 stats
->nr_sleeping
++;
3698 case TASK_UNINTERRUPTIBLE
:
3699 stats
->nr_uninterruptible
++;
3702 stats
->nr_stopped
++;
3705 if (delayacct_is_task_waiting_on_io(tsk
))
3706 stats
->nr_io_wait
++;
3710 cgroup_iter_end(cgrp
, &it
);
3718 * seq_file methods for the tasks/procs files. The seq_file position is the
3719 * next pid to display; the seq_file iterator is a pointer to the pid
3720 * in the cgroup->l->list array.
3723 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3726 * Initially we receive a position value that corresponds to
3727 * one more than the last pid shown (or 0 on the first call or
3728 * after a seek to the start). Use a binary-search to find the
3729 * next pid to display, if any
3731 struct cgroup_pidlist
*l
= s
->private;
3732 int index
= 0, pid
= *pos
;
3735 down_read(&l
->mutex
);
3737 int end
= l
->length
;
3739 while (index
< end
) {
3740 int mid
= (index
+ end
) / 2;
3741 if (l
->list
[mid
] == pid
) {
3744 } else if (l
->list
[mid
] <= pid
)
3750 /* If we're off the end of the array, we're done */
3751 if (index
>= l
->length
)
3753 /* Update the abstract position to be the actual pid that we found */
3754 iter
= l
->list
+ index
;
3759 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3761 struct cgroup_pidlist
*l
= s
->private;
3765 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3767 struct cgroup_pidlist
*l
= s
->private;
3769 pid_t
*end
= l
->list
+ l
->length
;
3771 * Advance to the next pid in the array. If this goes off the
3783 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3785 return seq_printf(s
, "%d\n", *(int *)v
);
3789 * seq_operations functions for iterating on pidlists through seq_file -
3790 * independent of whether it's tasks or procs
3792 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3793 .start
= cgroup_pidlist_start
,
3794 .stop
= cgroup_pidlist_stop
,
3795 .next
= cgroup_pidlist_next
,
3796 .show
= cgroup_pidlist_show
,
3799 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3802 * the case where we're the last user of this particular pidlist will
3803 * have us remove it from the cgroup's list, which entails taking the
3804 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3805 * pidlist_mutex, we have to take pidlist_mutex first.
3807 mutex_lock(&l
->owner
->pidlist_mutex
);
3808 down_write(&l
->mutex
);
3809 BUG_ON(!l
->use_count
);
3810 if (!--l
->use_count
) {
3811 /* we're the last user if refcount is 0; remove and free */
3812 list_del(&l
->links
);
3813 mutex_unlock(&l
->owner
->pidlist_mutex
);
3814 pidlist_free(l
->list
);
3815 put_pid_ns(l
->key
.ns
);
3816 up_write(&l
->mutex
);
3820 mutex_unlock(&l
->owner
->pidlist_mutex
);
3821 up_write(&l
->mutex
);
3824 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3826 struct cgroup_pidlist
*l
;
3827 if (!(file
->f_mode
& FMODE_READ
))
3830 * the seq_file will only be initialized if the file was opened for
3831 * reading; hence we check if it's not null only in that case.
3833 l
= ((struct seq_file
*)file
->private_data
)->private;
3834 cgroup_release_pid_array(l
);
3835 return seq_release(inode
, file
);
3838 static const struct file_operations cgroup_pidlist_operations
= {
3840 .llseek
= seq_lseek
,
3841 .write
= cgroup_file_write
,
3842 .release
= cgroup_pidlist_release
,
3846 * The following functions handle opens on a file that displays a pidlist
3847 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3850 /* helper function for the two below it */
3851 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3853 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3854 struct cgroup_pidlist
*l
;
3857 /* Nothing to do for write-only files */
3858 if (!(file
->f_mode
& FMODE_READ
))
3861 /* have the array populated */
3862 retval
= pidlist_array_load(cgrp
, type
, &l
);
3865 /* configure file information */
3866 file
->f_op
= &cgroup_pidlist_operations
;
3868 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3870 cgroup_release_pid_array(l
);
3873 ((struct seq_file
*)file
->private_data
)->private = l
;
3876 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3878 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3880 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3882 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3885 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3888 return notify_on_release(cgrp
);
3891 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3895 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3897 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3899 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3904 * When dput() is called asynchronously, if umount has been done and
3905 * then deactivate_super() in cgroup_free_fn() kills the superblock,
3906 * there's a small window that vfs will see the root dentry with non-zero
3907 * refcnt and trigger BUG().
3909 * That's why we hold a reference before dput() and drop it right after.
3911 static void cgroup_dput(struct cgroup
*cgrp
)
3913 struct super_block
*sb
= cgrp
->root
->sb
;
3915 atomic_inc(&sb
->s_active
);
3917 deactivate_super(sb
);
3921 * Unregister event and free resources.
3923 * Gets called from workqueue.
3925 static void cgroup_event_remove(struct work_struct
*work
)
3927 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3929 struct cgroup
*cgrp
= event
->cgrp
;
3931 remove_wait_queue(event
->wqh
, &event
->wait
);
3933 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3935 /* Notify userspace the event is going away. */
3936 eventfd_signal(event
->eventfd
, 1);
3938 eventfd_ctx_put(event
->eventfd
);
3944 * Gets called on POLLHUP on eventfd when user closes it.
3946 * Called with wqh->lock held and interrupts disabled.
3948 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3949 int sync
, void *key
)
3951 struct cgroup_event
*event
= container_of(wait
,
3952 struct cgroup_event
, wait
);
3953 struct cgroup
*cgrp
= event
->cgrp
;
3954 unsigned long flags
= (unsigned long)key
;
3956 if (flags
& POLLHUP
) {
3958 * If the event has been detached at cgroup removal, we
3959 * can simply return knowing the other side will cleanup
3962 * We can't race against event freeing since the other
3963 * side will require wqh->lock via remove_wait_queue(),
3966 spin_lock(&cgrp
->event_list_lock
);
3967 if (!list_empty(&event
->list
)) {
3968 list_del_init(&event
->list
);
3970 * We are in atomic context, but cgroup_event_remove()
3971 * may sleep, so we have to call it in workqueue.
3973 schedule_work(&event
->remove
);
3975 spin_unlock(&cgrp
->event_list_lock
);
3981 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3982 wait_queue_head_t
*wqh
, poll_table
*pt
)
3984 struct cgroup_event
*event
= container_of(pt
,
3985 struct cgroup_event
, pt
);
3988 add_wait_queue(wqh
, &event
->wait
);
3992 * Parse input and register new cgroup event handler.
3994 * Input must be in format '<event_fd> <control_fd> <args>'.
3995 * Interpretation of args is defined by control file implementation.
3997 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
4000 struct cgroup_event
*event
= NULL
;
4001 struct cgroup
*cgrp_cfile
;
4002 unsigned int efd
, cfd
;
4003 struct file
*efile
= NULL
;
4004 struct file
*cfile
= NULL
;
4008 efd
= simple_strtoul(buffer
, &endp
, 10);
4013 cfd
= simple_strtoul(buffer
, &endp
, 10);
4014 if ((*endp
!= ' ') && (*endp
!= '\0'))
4018 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
4022 INIT_LIST_HEAD(&event
->list
);
4023 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
4024 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
4025 INIT_WORK(&event
->remove
, cgroup_event_remove
);
4027 efile
= eventfd_fget(efd
);
4028 if (IS_ERR(efile
)) {
4029 ret
= PTR_ERR(efile
);
4033 event
->eventfd
= eventfd_ctx_fileget(efile
);
4034 if (IS_ERR(event
->eventfd
)) {
4035 ret
= PTR_ERR(event
->eventfd
);
4045 /* the process need read permission on control file */
4046 /* AV: shouldn't we check that it's been opened for read instead? */
4047 ret
= inode_permission(file_inode(cfile
), MAY_READ
);
4051 event
->cft
= __file_cft(cfile
);
4052 if (IS_ERR(event
->cft
)) {
4053 ret
= PTR_ERR(event
->cft
);
4058 * The file to be monitored must be in the same cgroup as
4059 * cgroup.event_control is.
4061 cgrp_cfile
= __d_cgrp(cfile
->f_dentry
->d_parent
);
4062 if (cgrp_cfile
!= cgrp
) {
4067 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
4072 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
4073 event
->eventfd
, buffer
);
4077 efile
->f_op
->poll(efile
, &event
->pt
);
4080 * Events should be removed after rmdir of cgroup directory, but before
4081 * destroying subsystem state objects. Let's take reference to cgroup
4082 * directory dentry to do that.
4086 spin_lock(&cgrp
->event_list_lock
);
4087 list_add(&event
->list
, &cgrp
->event_list
);
4088 spin_unlock(&cgrp
->event_list_lock
);
4099 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
4100 eventfd_ctx_put(event
->eventfd
);
4102 if (!IS_ERR_OR_NULL(efile
))
4110 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
4113 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4116 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
4121 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4123 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4127 static struct cftype cgroup_base_files
[] = {
4129 .name
= "cgroup.procs",
4130 .open
= cgroup_procs_open
,
4131 .write_u64
= cgroup_procs_write
,
4132 .release
= cgroup_pidlist_release
,
4133 .mode
= S_IRUGO
| S_IWUSR
,
4136 .name
= "cgroup.event_control",
4137 .write_string
= cgroup_write_event_control
,
4141 .name
= "cgroup.clone_children",
4142 .flags
= CFTYPE_INSANE
,
4143 .read_u64
= cgroup_clone_children_read
,
4144 .write_u64
= cgroup_clone_children_write
,
4147 .name
= "cgroup.sane_behavior",
4148 .flags
= CFTYPE_ONLY_ON_ROOT
,
4149 .read_seq_string
= cgroup_sane_behavior_show
,
4153 * Historical crazy stuff. These don't have "cgroup." prefix and
4154 * don't exist if sane_behavior. If you're depending on these, be
4155 * prepared to be burned.
4159 .flags
= CFTYPE_INSANE
, /* use "procs" instead */
4160 .open
= cgroup_tasks_open
,
4161 .write_u64
= cgroup_tasks_write
,
4162 .release
= cgroup_pidlist_release
,
4163 .mode
= S_IRUGO
| S_IWUSR
,
4166 .name
= "notify_on_release",
4167 .flags
= CFTYPE_INSANE
,
4168 .read_u64
= cgroup_read_notify_on_release
,
4169 .write_u64
= cgroup_write_notify_on_release
,
4172 .name
= "release_agent",
4173 .flags
= CFTYPE_INSANE
| CFTYPE_ONLY_ON_ROOT
,
4174 .read_seq_string
= cgroup_release_agent_show
,
4175 .write_string
= cgroup_release_agent_write
,
4176 .max_write_len
= PATH_MAX
,
4182 * cgroup_populate_dir - create subsys files in a cgroup directory
4183 * @cgrp: target cgroup
4184 * @subsys_mask: mask of the subsystem ids whose files should be added
4186 * On failure, no file is added.
4188 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
4190 struct cgroup_subsys
*ss
;
4193 /* process cftsets of each subsystem */
4194 for_each_subsys(ss
, i
) {
4195 struct cftype_set
*set
;
4197 if (!test_bit(i
, &subsys_mask
))
4200 list_for_each_entry(set
, &ss
->cftsets
, node
) {
4201 ret
= cgroup_addrm_files(cgrp
, ss
, set
->cfts
, true);
4207 /* This cgroup is ready now */
4208 for_each_root_subsys(cgrp
->root
, ss
) {
4209 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4210 struct css_id
*id
= rcu_dereference_protected(css
->id
, true);
4213 * Update id->css pointer and make this css visible from
4214 * CSS ID functions. This pointer will be dereferened
4215 * from RCU-read-side without locks.
4218 rcu_assign_pointer(id
->css
, css
);
4223 cgroup_clear_dir(cgrp
, subsys_mask
);
4227 static void css_dput_fn(struct work_struct
*work
)
4229 struct cgroup_subsys_state
*css
=
4230 container_of(work
, struct cgroup_subsys_state
, dput_work
);
4232 cgroup_dput(css
->cgroup
);
4235 static void css_release(struct percpu_ref
*ref
)
4237 struct cgroup_subsys_state
*css
=
4238 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4240 schedule_work(&css
->dput_work
);
4243 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
4244 struct cgroup_subsys
*ss
,
4245 struct cgroup
*cgrp
)
4250 if (cgrp
== cgroup_dummy_top
)
4251 css
->flags
|= CSS_ROOT
;
4252 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
4253 cgrp
->subsys
[ss
->subsys_id
] = css
;
4256 * css holds an extra ref to @cgrp->dentry which is put on the last
4257 * css_put(). dput() requires process context, which css_put() may
4258 * be called without. @css->dput_work will be used to invoke
4259 * dput() asynchronously from css_put().
4261 INIT_WORK(&css
->dput_work
, css_dput_fn
);
4264 /* invoke ->post_create() on a new CSS and mark it online if successful */
4265 static int online_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4269 lockdep_assert_held(&cgroup_mutex
);
4272 ret
= ss
->css_online(cgrp
);
4274 cgrp
->subsys
[ss
->subsys_id
]->flags
|= CSS_ONLINE
;
4278 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4279 static void offline_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4280 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4282 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4284 lockdep_assert_held(&cgroup_mutex
);
4286 if (!(css
->flags
& CSS_ONLINE
))
4289 if (ss
->css_offline
)
4290 ss
->css_offline(cgrp
);
4292 cgrp
->subsys
[ss
->subsys_id
]->flags
&= ~CSS_ONLINE
;
4296 * cgroup_create - create a cgroup
4297 * @parent: cgroup that will be parent of the new cgroup
4298 * @dentry: dentry of the new cgroup
4299 * @mode: mode to set on new inode
4301 * Must be called with the mutex on the parent inode held
4303 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
4306 struct cgroup
*cgrp
;
4307 struct cgroup_name
*name
;
4308 struct cgroupfs_root
*root
= parent
->root
;
4310 struct cgroup_subsys
*ss
;
4311 struct super_block
*sb
= root
->sb
;
4313 /* allocate the cgroup and its ID, 0 is reserved for the root */
4314 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4318 name
= cgroup_alloc_name(dentry
);
4321 rcu_assign_pointer(cgrp
->name
, name
);
4323 cgrp
->id
= ida_simple_get(&root
->cgroup_ida
, 1, 0, GFP_KERNEL
);
4328 * Only live parents can have children. Note that the liveliness
4329 * check isn't strictly necessary because cgroup_mkdir() and
4330 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4331 * anyway so that locking is contained inside cgroup proper and we
4332 * don't get nasty surprises if we ever grow another caller.
4334 if (!cgroup_lock_live_group(parent
)) {
4339 /* Grab a reference on the superblock so the hierarchy doesn't
4340 * get deleted on unmount if there are child cgroups. This
4341 * can be done outside cgroup_mutex, since the sb can't
4342 * disappear while someone has an open control file on the
4344 atomic_inc(&sb
->s_active
);
4346 init_cgroup_housekeeping(cgrp
);
4348 dentry
->d_fsdata
= cgrp
;
4349 cgrp
->dentry
= dentry
;
4351 cgrp
->parent
= parent
;
4352 cgrp
->root
= parent
->root
;
4354 if (notify_on_release(parent
))
4355 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4357 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4358 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4360 for_each_root_subsys(root
, ss
) {
4361 struct cgroup_subsys_state
*css
;
4363 css
= ss
->css_alloc(cgrp
);
4369 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 err
= cgroup_addrm_files(cgrp
, NULL
, cgroup_base_files
, true);
4425 err
= cgroup_populate_dir(cgrp
, root
->subsys_mask
);
4429 mutex_unlock(&cgroup_mutex
);
4430 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4435 for_each_root_subsys(root
, ss
) {
4436 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4439 percpu_ref_cancel_init(&css
->refcnt
);
4443 mutex_unlock(&cgroup_mutex
);
4444 /* Release the reference count that we took on the superblock */
4445 deactivate_super(sb
);
4447 ida_simple_remove(&root
->cgroup_ida
, cgrp
->id
);
4449 kfree(rcu_dereference_raw(cgrp
->name
));
4455 cgroup_destroy_locked(cgrp
);
4456 mutex_unlock(&cgroup_mutex
);
4457 mutex_unlock(&dentry
->d_inode
->i_mutex
);
4461 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4463 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4465 /* the vfs holds inode->i_mutex already */
4466 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4469 static void cgroup_css_killed(struct cgroup
*cgrp
)
4471 if (!atomic_dec_and_test(&cgrp
->css_kill_cnt
))
4474 /* percpu ref's of all css's are killed, kick off the next step */
4475 INIT_WORK(&cgrp
->destroy_work
, cgroup_offline_fn
);
4476 schedule_work(&cgrp
->destroy_work
);
4479 static void css_ref_killed_fn(struct percpu_ref
*ref
)
4481 struct cgroup_subsys_state
*css
=
4482 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4484 cgroup_css_killed(css
->cgroup
);
4488 * cgroup_destroy_locked - the first stage of cgroup destruction
4489 * @cgrp: cgroup to be destroyed
4491 * css's make use of percpu refcnts whose killing latency shouldn't be
4492 * exposed to userland and are RCU protected. Also, cgroup core needs to
4493 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
4494 * invoked. To satisfy all the requirements, destruction is implemented in
4495 * the following two steps.
4497 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4498 * userland visible parts and start killing the percpu refcnts of
4499 * css's. Set up so that the next stage will be kicked off once all
4500 * the percpu refcnts are confirmed to be killed.
4502 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4503 * rest of destruction. Once all cgroup references are gone, the
4504 * cgroup is RCU-freed.
4506 * This function implements s1. After this step, @cgrp is gone as far as
4507 * the userland is concerned and a new cgroup with the same name may be
4508 * created. As cgroup doesn't care about the names internally, this
4509 * doesn't cause any problem.
4511 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4512 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4514 struct dentry
*d
= cgrp
->dentry
;
4515 struct cgroup_event
*event
, *tmp
;
4516 struct cgroup_subsys
*ss
;
4519 lockdep_assert_held(&d
->d_inode
->i_mutex
);
4520 lockdep_assert_held(&cgroup_mutex
);
4523 * css_set_lock synchronizes access to ->cset_links and prevents
4524 * @cgrp from being removed while __put_css_set() is in progress.
4526 read_lock(&css_set_lock
);
4527 empty
= list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
);
4528 read_unlock(&css_set_lock
);
4533 * Block new css_tryget() by killing css refcnts. cgroup core
4534 * guarantees that, by the time ->css_offline() is invoked, no new
4535 * css reference will be given out via css_tryget(). We can't
4536 * simply call percpu_ref_kill() and proceed to offlining css's
4537 * because percpu_ref_kill() doesn't guarantee that the ref is seen
4538 * as killed on all CPUs on return.
4540 * Use percpu_ref_kill_and_confirm() to get notifications as each
4541 * css is confirmed to be seen as killed on all CPUs. The
4542 * notification callback keeps track of the number of css's to be
4543 * killed and schedules cgroup_offline_fn() to perform the rest of
4544 * destruction once the percpu refs of all css's are confirmed to
4547 atomic_set(&cgrp
->css_kill_cnt
, 1);
4548 for_each_root_subsys(cgrp
->root
, ss
) {
4549 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4552 * Killing would put the base ref, but we need to keep it
4553 * alive until after ->css_offline.
4555 percpu_ref_get(&css
->refcnt
);
4557 atomic_inc(&cgrp
->css_kill_cnt
);
4558 percpu_ref_kill_and_confirm(&css
->refcnt
, css_ref_killed_fn
);
4560 cgroup_css_killed(cgrp
);
4563 * Mark @cgrp dead. This prevents further task migration and child
4564 * creation by disabling cgroup_lock_live_group(). Note that
4565 * CGRP_DEAD assertion is depended upon by cgroup_next_sibling() to
4566 * resume iteration after dropping RCU read lock. See
4567 * cgroup_next_sibling() for details.
4569 set_bit(CGRP_DEAD
, &cgrp
->flags
);
4571 /* CGRP_DEAD is set, remove from ->release_list for the last time */
4572 raw_spin_lock(&release_list_lock
);
4573 if (!list_empty(&cgrp
->release_list
))
4574 list_del_init(&cgrp
->release_list
);
4575 raw_spin_unlock(&release_list_lock
);
4578 * Clear and remove @cgrp directory. The removal puts the base ref
4579 * but we aren't quite done with @cgrp yet, so hold onto it.
4581 cgroup_clear_dir(cgrp
, cgrp
->root
->subsys_mask
);
4582 cgroup_addrm_files(cgrp
, NULL
, cgroup_base_files
, false);
4584 cgroup_d_remove_dir(d
);
4587 * Unregister events and notify userspace.
4588 * Notify userspace about cgroup removing only after rmdir of cgroup
4589 * directory to avoid race between userspace and kernelspace.
4591 spin_lock(&cgrp
->event_list_lock
);
4592 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4593 list_del_init(&event
->list
);
4594 schedule_work(&event
->remove
);
4596 spin_unlock(&cgrp
->event_list_lock
);
4602 * cgroup_offline_fn - the second step of cgroup destruction
4603 * @work: cgroup->destroy_free_work
4605 * This function is invoked from a work item for a cgroup which is being
4606 * destroyed after the percpu refcnts of all css's are guaranteed to be
4607 * seen as killed on all CPUs, and performs the rest of destruction. This
4608 * is the second step of destruction described in the comment above
4609 * cgroup_destroy_locked().
4611 static void cgroup_offline_fn(struct work_struct
*work
)
4613 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, destroy_work
);
4614 struct cgroup
*parent
= cgrp
->parent
;
4615 struct dentry
*d
= cgrp
->dentry
;
4616 struct cgroup_subsys
*ss
;
4618 mutex_lock(&cgroup_mutex
);
4621 * css_tryget() is guaranteed to fail now. Tell subsystems to
4622 * initate destruction.
4624 for_each_root_subsys(cgrp
->root
, ss
)
4625 offline_css(ss
, cgrp
);
4628 * Put the css refs from cgroup_destroy_locked(). Each css holds
4629 * an extra reference to the cgroup's dentry and cgroup removal
4630 * proceeds regardless of css refs. On the last put of each css,
4631 * whenever that may be, the extra dentry ref is put so that dentry
4632 * destruction happens only after all css's are released.
4634 for_each_root_subsys(cgrp
->root
, ss
)
4635 css_put(cgrp
->subsys
[ss
->subsys_id
]);
4637 /* delete this cgroup from parent->children */
4638 list_del_rcu(&cgrp
->sibling
);
4642 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4643 check_for_release(parent
);
4645 mutex_unlock(&cgroup_mutex
);
4648 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4652 mutex_lock(&cgroup_mutex
);
4653 ret
= cgroup_destroy_locked(dentry
->d_fsdata
);
4654 mutex_unlock(&cgroup_mutex
);
4659 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4661 INIT_LIST_HEAD(&ss
->cftsets
);
4664 * base_cftset is embedded in subsys itself, no need to worry about
4667 if (ss
->base_cftypes
) {
4668 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4669 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4673 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4675 struct cgroup_subsys_state
*css
;
4677 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4679 mutex_lock(&cgroup_mutex
);
4681 /* init base cftset */
4682 cgroup_init_cftsets(ss
);
4684 /* Create the top cgroup state for this subsystem */
4685 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4686 ss
->root
= &cgroup_dummy_root
;
4687 css
= ss
->css_alloc(cgroup_dummy_top
);
4688 /* We don't handle early failures gracefully */
4689 BUG_ON(IS_ERR(css
));
4690 init_cgroup_css(css
, ss
, cgroup_dummy_top
);
4692 /* Update the init_css_set to contain a subsys
4693 * pointer to this state - since the subsystem is
4694 * newly registered, all tasks and hence the
4695 * init_css_set is in the subsystem's top cgroup. */
4696 init_css_set
.subsys
[ss
->subsys_id
] = css
;
4698 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4700 /* At system boot, before all subsystems have been
4701 * registered, no tasks have been forked, so we don't
4702 * need to invoke fork callbacks here. */
4703 BUG_ON(!list_empty(&init_task
.tasks
));
4705 BUG_ON(online_css(ss
, cgroup_dummy_top
));
4707 mutex_unlock(&cgroup_mutex
);
4709 /* this function shouldn't be used with modular subsystems, since they
4710 * need to register a subsys_id, among other things */
4715 * cgroup_load_subsys: load and register a modular subsystem at runtime
4716 * @ss: the subsystem to load
4718 * This function should be called in a modular subsystem's initcall. If the
4719 * subsystem is built as a module, it will be assigned a new subsys_id and set
4720 * up for use. If the subsystem is built-in anyway, work is delegated to the
4721 * simpler cgroup_init_subsys.
4723 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4725 struct cgroup_subsys_state
*css
;
4727 struct hlist_node
*tmp
;
4728 struct css_set
*cset
;
4731 /* check name and function validity */
4732 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4733 ss
->css_alloc
== NULL
|| ss
->css_free
== NULL
)
4737 * we don't support callbacks in modular subsystems. this check is
4738 * before the ss->module check for consistency; a subsystem that could
4739 * be a module should still have no callbacks even if the user isn't
4740 * compiling it as one.
4742 if (ss
->fork
|| ss
->exit
)
4746 * an optionally modular subsystem is built-in: we want to do nothing,
4747 * since cgroup_init_subsys will have already taken care of it.
4749 if (ss
->module
== NULL
) {
4750 /* a sanity check */
4751 BUG_ON(cgroup_subsys
[ss
->subsys_id
] != ss
);
4755 /* init base cftset */
4756 cgroup_init_cftsets(ss
);
4758 mutex_lock(&cgroup_mutex
);
4759 cgroup_subsys
[ss
->subsys_id
] = ss
;
4762 * no ss->css_alloc seems to need anything important in the ss
4763 * struct, so this can happen first (i.e. before the dummy root
4766 css
= ss
->css_alloc(cgroup_dummy_top
);
4768 /* failure case - need to deassign the cgroup_subsys[] slot. */
4769 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4770 mutex_unlock(&cgroup_mutex
);
4771 return PTR_ERR(css
);
4774 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4775 ss
->root
= &cgroup_dummy_root
;
4777 /* our new subsystem will be attached to the dummy hierarchy. */
4778 init_cgroup_css(css
, ss
, cgroup_dummy_top
);
4779 /* init_idr must be after init_cgroup_css because it sets css->id. */
4781 ret
= cgroup_init_idr(ss
, css
);
4787 * Now we need to entangle the css into the existing css_sets. unlike
4788 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4789 * will need a new pointer to it; done by iterating the css_set_table.
4790 * furthermore, modifying the existing css_sets will corrupt the hash
4791 * table state, so each changed css_set will need its hash recomputed.
4792 * this is all done under the css_set_lock.
4794 write_lock(&css_set_lock
);
4795 hash_for_each_safe(css_set_table
, i
, tmp
, cset
, hlist
) {
4796 /* skip entries that we already rehashed */
4797 if (cset
->subsys
[ss
->subsys_id
])
4799 /* remove existing entry */
4800 hash_del(&cset
->hlist
);
4802 cset
->subsys
[ss
->subsys_id
] = css
;
4803 /* recompute hash and restore entry */
4804 key
= css_set_hash(cset
->subsys
);
4805 hash_add(css_set_table
, &cset
->hlist
, key
);
4807 write_unlock(&css_set_lock
);
4809 ret
= online_css(ss
, cgroup_dummy_top
);
4814 mutex_unlock(&cgroup_mutex
);
4818 mutex_unlock(&cgroup_mutex
);
4819 /* @ss can't be mounted here as try_module_get() would fail */
4820 cgroup_unload_subsys(ss
);
4823 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4826 * cgroup_unload_subsys: unload a modular subsystem
4827 * @ss: the subsystem to unload
4829 * This function should be called in a modular subsystem's exitcall. When this
4830 * function is invoked, the refcount on the subsystem's module will be 0, so
4831 * the subsystem will not be attached to any hierarchy.
4833 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4835 struct cgrp_cset_link
*link
;
4837 BUG_ON(ss
->module
== NULL
);
4840 * we shouldn't be called if the subsystem is in use, and the use of
4841 * try_module_get in parse_cgroupfs_options should ensure that it
4842 * doesn't start being used while we're killing it off.
4844 BUG_ON(ss
->root
!= &cgroup_dummy_root
);
4846 mutex_lock(&cgroup_mutex
);
4848 offline_css(ss
, cgroup_dummy_top
);
4851 idr_destroy(&ss
->idr
);
4853 /* deassign the subsys_id */
4854 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4856 /* remove subsystem from the dummy root's list of subsystems */
4857 list_del_init(&ss
->sibling
);
4860 * disentangle the css from all css_sets attached to the dummy
4861 * top. as in loading, we need to pay our respects to the hashtable
4864 write_lock(&css_set_lock
);
4865 list_for_each_entry(link
, &cgroup_dummy_top
->cset_links
, cset_link
) {
4866 struct css_set
*cset
= link
->cset
;
4869 hash_del(&cset
->hlist
);
4870 cset
->subsys
[ss
->subsys_id
] = NULL
;
4871 key
= css_set_hash(cset
->subsys
);
4872 hash_add(css_set_table
, &cset
->hlist
, key
);
4874 write_unlock(&css_set_lock
);
4877 * remove subsystem's css from the cgroup_dummy_top and free it -
4878 * need to free before marking as null because ss->css_free needs
4879 * the cgrp->subsys pointer to find their state. note that this
4880 * also takes care of freeing the css_id.
4882 ss
->css_free(cgroup_dummy_top
);
4883 cgroup_dummy_top
->subsys
[ss
->subsys_id
] = NULL
;
4885 mutex_unlock(&cgroup_mutex
);
4887 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4890 * cgroup_init_early - cgroup initialization at system boot
4892 * Initialize cgroups at system boot, and initialize any
4893 * subsystems that request early init.
4895 int __init
cgroup_init_early(void)
4897 struct cgroup_subsys
*ss
;
4900 atomic_set(&init_css_set
.refcount
, 1);
4901 INIT_LIST_HEAD(&init_css_set
.cgrp_links
);
4902 INIT_LIST_HEAD(&init_css_set
.tasks
);
4903 INIT_HLIST_NODE(&init_css_set
.hlist
);
4905 init_cgroup_root(&cgroup_dummy_root
);
4906 cgroup_root_count
= 1;
4907 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
4909 init_cgrp_cset_link
.cset
= &init_css_set
;
4910 init_cgrp_cset_link
.cgrp
= cgroup_dummy_top
;
4911 list_add(&init_cgrp_cset_link
.cset_link
, &cgroup_dummy_top
->cset_links
);
4912 list_add(&init_cgrp_cset_link
.cgrp_link
, &init_css_set
.cgrp_links
);
4914 /* at bootup time, we don't worry about modular subsystems */
4915 for_each_builtin_subsys(ss
, i
) {
4917 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4918 BUG_ON(!ss
->css_alloc
);
4919 BUG_ON(!ss
->css_free
);
4920 if (ss
->subsys_id
!= i
) {
4921 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4922 ss
->name
, ss
->subsys_id
);
4927 cgroup_init_subsys(ss
);
4933 * cgroup_init - cgroup initialization
4935 * Register cgroup filesystem and /proc file, and initialize
4936 * any subsystems that didn't request early init.
4938 int __init
cgroup_init(void)
4940 struct cgroup_subsys
*ss
;
4944 err
= bdi_init(&cgroup_backing_dev_info
);
4948 for_each_builtin_subsys(ss
, i
) {
4949 if (!ss
->early_init
)
4950 cgroup_init_subsys(ss
);
4952 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4955 /* allocate id for the dummy hierarchy */
4956 mutex_lock(&cgroup_mutex
);
4957 mutex_lock(&cgroup_root_mutex
);
4959 /* Add init_css_set to the hash table */
4960 key
= css_set_hash(init_css_set
.subsys
);
4961 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4963 BUG_ON(cgroup_init_root_id(&cgroup_dummy_root
, 0, 1));
4965 mutex_unlock(&cgroup_root_mutex
);
4966 mutex_unlock(&cgroup_mutex
);
4968 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4974 err
= register_filesystem(&cgroup_fs_type
);
4976 kobject_put(cgroup_kobj
);
4980 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4984 bdi_destroy(&cgroup_backing_dev_info
);
4990 * proc_cgroup_show()
4991 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4992 * - Used for /proc/<pid>/cgroup.
4993 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4994 * doesn't really matter if tsk->cgroup changes after we read it,
4995 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4996 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4997 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4998 * cgroup to top_cgroup.
5001 /* TODO: Use a proper seq_file iterator */
5002 int proc_cgroup_show(struct seq_file
*m
, void *v
)
5005 struct task_struct
*tsk
;
5008 struct cgroupfs_root
*root
;
5011 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5017 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
5023 mutex_lock(&cgroup_mutex
);
5025 for_each_active_root(root
) {
5026 struct cgroup_subsys
*ss
;
5027 struct cgroup
*cgrp
;
5030 seq_printf(m
, "%d:", root
->hierarchy_id
);
5031 for_each_root_subsys(root
, ss
)
5032 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
5033 if (strlen(root
->name
))
5034 seq_printf(m
, "%sname=%s", count
? "," : "",
5037 cgrp
= task_cgroup_from_root(tsk
, root
);
5038 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
5046 mutex_unlock(&cgroup_mutex
);
5047 put_task_struct(tsk
);
5054 /* Display information about each subsystem and each hierarchy */
5055 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5057 struct cgroup_subsys
*ss
;
5060 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5062 * ideally we don't want subsystems moving around while we do this.
5063 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5064 * subsys/hierarchy state.
5066 mutex_lock(&cgroup_mutex
);
5068 for_each_subsys(ss
, i
)
5069 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5070 ss
->name
, ss
->root
->hierarchy_id
,
5071 ss
->root
->number_of_cgroups
, !ss
->disabled
);
5073 mutex_unlock(&cgroup_mutex
);
5077 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5079 return single_open(file
, proc_cgroupstats_show
, NULL
);
5082 static const struct file_operations proc_cgroupstats_operations
= {
5083 .open
= cgroupstats_open
,
5085 .llseek
= seq_lseek
,
5086 .release
= single_release
,
5090 * cgroup_fork - attach newly forked task to its parents cgroup.
5091 * @child: pointer to task_struct of forking parent process.
5093 * Description: A task inherits its parent's cgroup at fork().
5095 * A pointer to the shared css_set was automatically copied in
5096 * fork.c by dup_task_struct(). However, we ignore that copy, since
5097 * it was not made under the protection of RCU or cgroup_mutex, so
5098 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
5099 * have already changed current->cgroups, allowing the previously
5100 * referenced cgroup group to be removed and freed.
5102 * At the point that cgroup_fork() is called, 'current' is the parent
5103 * task, and the passed argument 'child' points to the child task.
5105 void cgroup_fork(struct task_struct
*child
)
5108 get_css_set(task_css_set(current
));
5109 child
->cgroups
= current
->cgroups
;
5110 task_unlock(current
);
5111 INIT_LIST_HEAD(&child
->cg_list
);
5115 * cgroup_post_fork - called on a new task after adding it to the task list
5116 * @child: the task in question
5118 * Adds the task to the list running through its css_set if necessary and
5119 * call the subsystem fork() callbacks. Has to be after the task is
5120 * visible on the task list in case we race with the first call to
5121 * cgroup_iter_start() - to guarantee that the new task ends up on its
5124 void cgroup_post_fork(struct task_struct
*child
)
5126 struct cgroup_subsys
*ss
;
5130 * use_task_css_set_links is set to 1 before we walk the tasklist
5131 * under the tasklist_lock and we read it here after we added the child
5132 * to the tasklist under the tasklist_lock as well. If the child wasn't
5133 * yet in the tasklist when we walked through it from
5134 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
5135 * should be visible now due to the paired locking and barriers implied
5136 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
5137 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
5140 if (use_task_css_set_links
) {
5141 write_lock(&css_set_lock
);
5143 if (list_empty(&child
->cg_list
))
5144 list_add(&child
->cg_list
, &task_css_set(child
)->tasks
);
5146 write_unlock(&css_set_lock
);
5150 * Call ss->fork(). This must happen after @child is linked on
5151 * css_set; otherwise, @child might change state between ->fork()
5152 * and addition to css_set.
5154 if (need_forkexit_callback
) {
5156 * fork/exit callbacks are supported only for builtin
5157 * subsystems, and the builtin section of the subsys
5158 * array is immutable, so we don't need to lock the
5159 * subsys array here. On the other hand, modular section
5160 * of the array can be freed at module unload, so we
5163 for_each_builtin_subsys(ss
, i
)
5170 * cgroup_exit - detach cgroup from exiting task
5171 * @tsk: pointer to task_struct of exiting process
5172 * @run_callback: run exit callbacks?
5174 * Description: Detach cgroup from @tsk and release it.
5176 * Note that cgroups marked notify_on_release force every task in
5177 * them to take the global cgroup_mutex mutex when exiting.
5178 * This could impact scaling on very large systems. Be reluctant to
5179 * use notify_on_release cgroups where very high task exit scaling
5180 * is required on large systems.
5182 * the_top_cgroup_hack:
5184 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
5186 * We call cgroup_exit() while the task is still competent to
5187 * handle notify_on_release(), then leave the task attached to the
5188 * root cgroup in each hierarchy for the remainder of its exit.
5190 * To do this properly, we would increment the reference count on
5191 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
5192 * code we would add a second cgroup function call, to drop that
5193 * reference. This would just create an unnecessary hot spot on
5194 * the top_cgroup reference count, to no avail.
5196 * Normally, holding a reference to a cgroup without bumping its
5197 * count is unsafe. The cgroup could go away, or someone could
5198 * attach us to a different cgroup, decrementing the count on
5199 * the first cgroup that we never incremented. But in this case,
5200 * top_cgroup isn't going away, and either task has PF_EXITING set,
5201 * which wards off any cgroup_attach_task() attempts, or task is a failed
5202 * fork, never visible to cgroup_attach_task.
5204 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
5206 struct cgroup_subsys
*ss
;
5207 struct css_set
*cset
;
5211 * Unlink from the css_set task list if necessary.
5212 * Optimistically check cg_list before taking
5215 if (!list_empty(&tsk
->cg_list
)) {
5216 write_lock(&css_set_lock
);
5217 if (!list_empty(&tsk
->cg_list
))
5218 list_del_init(&tsk
->cg_list
);
5219 write_unlock(&css_set_lock
);
5222 /* Reassign the task to the init_css_set. */
5224 cset
= task_css_set(tsk
);
5225 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5227 if (run_callbacks
&& need_forkexit_callback
) {
5229 * fork/exit callbacks are supported only for builtin
5230 * subsystems, see cgroup_post_fork() for details.
5232 for_each_builtin_subsys(ss
, i
) {
5234 struct cgroup
*old_cgrp
= cset
->subsys
[i
]->cgroup
;
5235 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
5237 ss
->exit(cgrp
, old_cgrp
, tsk
);
5243 put_css_set_taskexit(cset
);
5246 static void check_for_release(struct cgroup
*cgrp
)
5248 if (cgroup_is_releasable(cgrp
) &&
5249 list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
)) {
5251 * Control Group is currently removeable. If it's not
5252 * already queued for a userspace notification, queue
5255 int need_schedule_work
= 0;
5257 raw_spin_lock(&release_list_lock
);
5258 if (!cgroup_is_dead(cgrp
) &&
5259 list_empty(&cgrp
->release_list
)) {
5260 list_add(&cgrp
->release_list
, &release_list
);
5261 need_schedule_work
= 1;
5263 raw_spin_unlock(&release_list_lock
);
5264 if (need_schedule_work
)
5265 schedule_work(&release_agent_work
);
5270 * Notify userspace when a cgroup is released, by running the
5271 * configured release agent with the name of the cgroup (path
5272 * relative to the root of cgroup file system) as the argument.
5274 * Most likely, this user command will try to rmdir this cgroup.
5276 * This races with the possibility that some other task will be
5277 * attached to this cgroup before it is removed, or that some other
5278 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5279 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5280 * unused, and this cgroup will be reprieved from its death sentence,
5281 * to continue to serve a useful existence. Next time it's released,
5282 * we will get notified again, if it still has 'notify_on_release' set.
5284 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5285 * means only wait until the task is successfully execve()'d. The
5286 * separate release agent task is forked by call_usermodehelper(),
5287 * then control in this thread returns here, without waiting for the
5288 * release agent task. We don't bother to wait because the caller of
5289 * this routine has no use for the exit status of the release agent
5290 * task, so no sense holding our caller up for that.
5292 static void cgroup_release_agent(struct work_struct
*work
)
5294 BUG_ON(work
!= &release_agent_work
);
5295 mutex_lock(&cgroup_mutex
);
5296 raw_spin_lock(&release_list_lock
);
5297 while (!list_empty(&release_list
)) {
5298 char *argv
[3], *envp
[3];
5300 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5301 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5304 list_del_init(&cgrp
->release_list
);
5305 raw_spin_unlock(&release_list_lock
);
5306 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5309 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5311 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5316 argv
[i
++] = agentbuf
;
5317 argv
[i
++] = pathbuf
;
5321 /* minimal command environment */
5322 envp
[i
++] = "HOME=/";
5323 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5326 /* Drop the lock while we invoke the usermode helper,
5327 * since the exec could involve hitting disk and hence
5328 * be a slow process */
5329 mutex_unlock(&cgroup_mutex
);
5330 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5331 mutex_lock(&cgroup_mutex
);
5335 raw_spin_lock(&release_list_lock
);
5337 raw_spin_unlock(&release_list_lock
);
5338 mutex_unlock(&cgroup_mutex
);
5341 static int __init
cgroup_disable(char *str
)
5343 struct cgroup_subsys
*ss
;
5347 while ((token
= strsep(&str
, ",")) != NULL
) {
5352 * cgroup_disable, being at boot time, can't know about
5353 * module subsystems, so we don't worry about them.
5355 for_each_builtin_subsys(ss
, i
) {
5356 if (!strcmp(token
, ss
->name
)) {
5358 printk(KERN_INFO
"Disabling %s control group"
5359 " subsystem\n", ss
->name
);
5366 __setup("cgroup_disable=", cgroup_disable
);
5369 * Functons for CSS ID.
5372 /* to get ID other than 0, this should be called when !cgroup_is_dead() */
5373 unsigned short css_id(struct cgroup_subsys_state
*css
)
5375 struct css_id
*cssid
;
5378 * This css_id() can return correct value when somone has refcnt
5379 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5380 * it's unchanged until freed.
5382 cssid
= rcu_dereference_raw(css
->id
);
5388 EXPORT_SYMBOL_GPL(css_id
);
5391 * css_is_ancestor - test "root" css is an ancestor of "child"
5392 * @child: the css to be tested.
5393 * @root: the css supporsed to be an ancestor of the child.
5395 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5396 * this function reads css->id, the caller must hold rcu_read_lock().
5397 * But, considering usual usage, the csses should be valid objects after test.
5398 * Assuming that the caller will do some action to the child if this returns
5399 * returns true, the caller must take "child";s reference count.
5400 * If "child" is valid object and this returns true, "root" is valid, too.
5403 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5404 const struct cgroup_subsys_state
*root
)
5406 struct css_id
*child_id
;
5407 struct css_id
*root_id
;
5409 child_id
= rcu_dereference(child
->id
);
5412 root_id
= rcu_dereference(root
->id
);
5415 if (child_id
->depth
< root_id
->depth
)
5417 if (child_id
->stack
[root_id
->depth
] != root_id
->id
)
5422 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5424 struct css_id
*id
= rcu_dereference_protected(css
->id
, true);
5426 /* When this is called before css_id initialization, id can be NULL */
5430 BUG_ON(!ss
->use_id
);
5432 rcu_assign_pointer(id
->css
, NULL
);
5433 rcu_assign_pointer(css
->id
, NULL
);
5434 spin_lock(&ss
->id_lock
);
5435 idr_remove(&ss
->idr
, id
->id
);
5436 spin_unlock(&ss
->id_lock
);
5437 kfree_rcu(id
, rcu_head
);
5439 EXPORT_SYMBOL_GPL(free_css_id
);
5442 * This is called by init or create(). Then, calls to this function are
5443 * always serialized (By cgroup_mutex() at create()).
5446 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5448 struct css_id
*newid
;
5451 BUG_ON(!ss
->use_id
);
5453 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5454 newid
= kzalloc(size
, GFP_KERNEL
);
5456 return ERR_PTR(-ENOMEM
);
5458 idr_preload(GFP_KERNEL
);
5459 spin_lock(&ss
->id_lock
);
5460 /* Don't use 0. allocates an ID of 1-65535 */
5461 ret
= idr_alloc(&ss
->idr
, newid
, 1, CSS_ID_MAX
+ 1, GFP_NOWAIT
);
5462 spin_unlock(&ss
->id_lock
);
5465 /* Returns error when there are no free spaces for new ID.*/
5470 newid
->depth
= depth
;
5474 return ERR_PTR(ret
);
5478 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5479 struct cgroup_subsys_state
*rootcss
)
5481 struct css_id
*newid
;
5483 spin_lock_init(&ss
->id_lock
);
5486 newid
= get_new_cssid(ss
, 0);
5488 return PTR_ERR(newid
);
5490 newid
->stack
[0] = newid
->id
;
5491 RCU_INIT_POINTER(newid
->css
, rootcss
);
5492 RCU_INIT_POINTER(rootcss
->id
, newid
);
5496 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5497 struct cgroup
*child
)
5499 int subsys_id
, i
, depth
= 0;
5500 struct cgroup_subsys_state
*parent_css
, *child_css
;
5501 struct css_id
*child_id
, *parent_id
;
5503 subsys_id
= ss
->subsys_id
;
5504 parent_css
= parent
->subsys
[subsys_id
];
5505 child_css
= child
->subsys
[subsys_id
];
5506 parent_id
= rcu_dereference_protected(parent_css
->id
, true);
5507 depth
= parent_id
->depth
+ 1;
5509 child_id
= get_new_cssid(ss
, depth
);
5510 if (IS_ERR(child_id
))
5511 return PTR_ERR(child_id
);
5513 for (i
= 0; i
< depth
; i
++)
5514 child_id
->stack
[i
] = parent_id
->stack
[i
];
5515 child_id
->stack
[depth
] = child_id
->id
;
5517 * child_id->css pointer will be set after this cgroup is available
5518 * see cgroup_populate_dir()
5520 rcu_assign_pointer(child_css
->id
, child_id
);
5526 * css_lookup - lookup css by id
5527 * @ss: cgroup subsys to be looked into.
5530 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5531 * NULL if not. Should be called under rcu_read_lock()
5533 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5535 struct css_id
*cssid
= NULL
;
5537 BUG_ON(!ss
->use_id
);
5538 cssid
= idr_find(&ss
->idr
, id
);
5540 if (unlikely(!cssid
))
5543 return rcu_dereference(cssid
->css
);
5545 EXPORT_SYMBOL_GPL(css_lookup
);
5548 * get corresponding css from file open on cgroupfs directory
5550 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5552 struct cgroup
*cgrp
;
5553 struct inode
*inode
;
5554 struct cgroup_subsys_state
*css
;
5556 inode
= file_inode(f
);
5557 /* check in cgroup filesystem dir */
5558 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5559 return ERR_PTR(-EBADF
);
5561 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5562 return ERR_PTR(-EINVAL
);
5565 cgrp
= __d_cgrp(f
->f_dentry
);
5566 css
= cgrp
->subsys
[id
];
5567 return css
? css
: ERR_PTR(-ENOENT
);
5570 #ifdef CONFIG_CGROUP_DEBUG
5571 static struct cgroup_subsys_state
*debug_css_alloc(struct cgroup
*cgrp
)
5573 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5576 return ERR_PTR(-ENOMEM
);
5581 static void debug_css_free(struct cgroup
*cgrp
)
5583 kfree(cgrp
->subsys
[debug_subsys_id
]);
5586 static u64
debug_taskcount_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5588 return cgroup_task_count(cgrp
);
5591 static u64
current_css_set_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5593 return (u64
)(unsigned long)current
->cgroups
;
5596 static u64
current_css_set_refcount_read(struct cgroup
*cgrp
,
5602 count
= atomic_read(&task_css_set(current
)->refcount
);
5607 static int current_css_set_cg_links_read(struct cgroup
*cgrp
,
5609 struct seq_file
*seq
)
5611 struct cgrp_cset_link
*link
;
5612 struct css_set
*cset
;
5614 read_lock(&css_set_lock
);
5616 cset
= rcu_dereference(current
->cgroups
);
5617 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5618 struct cgroup
*c
= link
->cgrp
;
5622 name
= c
->dentry
->d_name
.name
;
5625 seq_printf(seq
, "Root %d group %s\n",
5626 c
->root
->hierarchy_id
, name
);
5629 read_unlock(&css_set_lock
);
5633 #define MAX_TASKS_SHOWN_PER_CSS 25
5634 static int cgroup_css_links_read(struct cgroup
*cgrp
,
5636 struct seq_file
*seq
)
5638 struct cgrp_cset_link
*link
;
5640 read_lock(&css_set_lock
);
5641 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
) {
5642 struct css_set
*cset
= link
->cset
;
5643 struct task_struct
*task
;
5645 seq_printf(seq
, "css_set %p\n", cset
);
5646 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5647 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5648 seq_puts(seq
, " ...\n");
5651 seq_printf(seq
, " task %d\n",
5652 task_pid_vnr(task
));
5656 read_unlock(&css_set_lock
);
5660 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5662 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5665 static struct cftype debug_files
[] = {
5667 .name
= "taskcount",
5668 .read_u64
= debug_taskcount_read
,
5672 .name
= "current_css_set",
5673 .read_u64
= current_css_set_read
,
5677 .name
= "current_css_set_refcount",
5678 .read_u64
= current_css_set_refcount_read
,
5682 .name
= "current_css_set_cg_links",
5683 .read_seq_string
= current_css_set_cg_links_read
,
5687 .name
= "cgroup_css_links",
5688 .read_seq_string
= cgroup_css_links_read
,
5692 .name
= "releasable",
5693 .read_u64
= releasable_read
,
5699 struct cgroup_subsys debug_subsys
= {
5701 .css_alloc
= debug_css_alloc
,
5702 .css_free
= debug_css_free
,
5703 .subsys_id
= debug_subsys_id
,
5704 .base_cftypes
= debug_files
,
5706 #endif /* CONFIG_CGROUP_DEBUG */