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 "cset" matches "old_cset" except for the hierarchy
470 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
472 static bool compare_css_sets(struct css_set
*cset
,
473 struct css_set
*old_cset
,
474 struct cgroup
*new_cgrp
,
475 struct cgroup_subsys_state
*template[])
477 struct list_head
*l1
, *l2
;
479 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
))) {
480 /* Not all subsystems matched */
485 * Compare cgroup pointers in order to distinguish between
486 * different cgroups in heirarchies with no subsystems. We
487 * could get by with just this check alone (and skip the
488 * memcmp above) but on most setups the memcmp check will
489 * avoid the need for this more expensive check on almost all
493 l1
= &cset
->cgrp_links
;
494 l2
= &old_cset
->cgrp_links
;
496 struct cgrp_cset_link
*link1
, *link2
;
497 struct cgroup
*cgrp1
, *cgrp2
;
501 /* See if we reached the end - both lists are equal length. */
502 if (l1
== &cset
->cgrp_links
) {
503 BUG_ON(l2
!= &old_cset
->cgrp_links
);
506 BUG_ON(l2
== &old_cset
->cgrp_links
);
508 /* Locate the cgroups associated with these links. */
509 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
510 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
513 /* Hierarchies should be linked in the same order. */
514 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
517 * If this hierarchy is the hierarchy of the cgroup
518 * that's changing, then we need to check that this
519 * css_set points to the new cgroup; if it's any other
520 * hierarchy, then this css_set should point to the
521 * same cgroup as the old css_set.
523 if (cgrp1
->root
== new_cgrp
->root
) {
524 if (cgrp1
!= new_cgrp
)
535 * find_existing_css_set - init css array and find the matching css_set
536 * @old_cset: the css_set that we're using before the cgroup transition
537 * @cgrp: the cgroup that we're moving into
538 * @template: out param for the new set of csses, should be clear on entry
540 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
542 struct cgroup_subsys_state
*template[])
544 struct cgroupfs_root
*root
= cgrp
->root
;
545 struct cgroup_subsys
*ss
;
546 struct css_set
*cset
;
551 * Build the set of subsystem state objects that we want to see in the
552 * new css_set. while subsystems can change globally, the entries here
553 * won't change, so no need for locking.
555 for_each_subsys(ss
, i
) {
556 if (root
->subsys_mask
& (1UL << i
)) {
557 /* Subsystem is in this hierarchy. So we want
558 * the subsystem state from the new
560 template[i
] = cgrp
->subsys
[i
];
562 /* Subsystem is not in this hierarchy, so we
563 * don't want to change the subsystem state */
564 template[i
] = old_cset
->subsys
[i
];
568 key
= css_set_hash(template);
569 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
570 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
573 /* This css_set matches what we need */
577 /* No existing cgroup group matched */
581 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
583 struct cgrp_cset_link
*link
, *tmp_link
;
585 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
586 list_del(&link
->cset_link
);
592 * allocate_cgrp_cset_links - allocate cgrp_cset_links
593 * @count: the number of links to allocate
594 * @tmp_links: list_head the allocated links are put on
596 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
597 * through ->cset_link. Returns 0 on success or -errno.
599 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
601 struct cgrp_cset_link
*link
;
604 INIT_LIST_HEAD(tmp_links
);
606 for (i
= 0; i
< count
; i
++) {
607 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
609 free_cgrp_cset_links(tmp_links
);
612 list_add(&link
->cset_link
, tmp_links
);
618 * link_css_set - a helper function to link a css_set to a cgroup
619 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
620 * @cset: the css_set to be linked
621 * @cgrp: the destination cgroup
623 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
626 struct cgrp_cset_link
*link
;
628 BUG_ON(list_empty(tmp_links
));
629 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
632 list_move(&link
->cset_link
, &cgrp
->cset_links
);
634 * Always add links to the tail of the list so that the list
635 * is sorted by order of hierarchy creation
637 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
641 * find_css_set - return a new css_set with one cgroup updated
642 * @old_cset: the baseline css_set
643 * @cgrp: the cgroup to be updated
645 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
646 * substituted into the appropriate hierarchy.
648 static struct css_set
*find_css_set(struct css_set
*old_cset
,
651 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
652 struct css_set
*cset
;
653 struct list_head tmp_links
;
654 struct cgrp_cset_link
*link
;
657 lockdep_assert_held(&cgroup_mutex
);
659 /* First see if we already have a cgroup group that matches
661 read_lock(&css_set_lock
);
662 cset
= find_existing_css_set(old_cset
, cgrp
, template);
665 read_unlock(&css_set_lock
);
670 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
674 /* Allocate all the cgrp_cset_link objects that we'll need */
675 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
680 atomic_set(&cset
->refcount
, 1);
681 INIT_LIST_HEAD(&cset
->cgrp_links
);
682 INIT_LIST_HEAD(&cset
->tasks
);
683 INIT_HLIST_NODE(&cset
->hlist
);
685 /* Copy the set of subsystem state objects generated in
686 * find_existing_css_set() */
687 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
689 write_lock(&css_set_lock
);
690 /* Add reference counts and links from the new css_set. */
691 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
692 struct cgroup
*c
= link
->cgrp
;
694 if (c
->root
== cgrp
->root
)
696 link_css_set(&tmp_links
, cset
, c
);
699 BUG_ON(!list_empty(&tmp_links
));
703 /* Add this cgroup group to the hash table */
704 key
= css_set_hash(cset
->subsys
);
705 hash_add(css_set_table
, &cset
->hlist
, key
);
707 write_unlock(&css_set_lock
);
713 * Return the cgroup for "task" from the given hierarchy. Must be
714 * called with cgroup_mutex held.
716 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
717 struct cgroupfs_root
*root
)
719 struct css_set
*cset
;
720 struct cgroup
*res
= NULL
;
722 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
723 read_lock(&css_set_lock
);
725 * No need to lock the task - since we hold cgroup_mutex the
726 * task can't change groups, so the only thing that can happen
727 * is that it exits and its css is set back to init_css_set.
729 cset
= task_css_set(task
);
730 if (cset
== &init_css_set
) {
731 res
= &root
->top_cgroup
;
733 struct cgrp_cset_link
*link
;
735 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
736 struct cgroup
*c
= link
->cgrp
;
738 if (c
->root
== root
) {
744 read_unlock(&css_set_lock
);
750 * There is one global cgroup mutex. We also require taking
751 * task_lock() when dereferencing a task's cgroup subsys pointers.
752 * See "The task_lock() exception", at the end of this comment.
754 * A task must hold cgroup_mutex to modify cgroups.
756 * Any task can increment and decrement the count field without lock.
757 * So in general, code holding cgroup_mutex can't rely on the count
758 * field not changing. However, if the count goes to zero, then only
759 * cgroup_attach_task() can increment it again. Because a count of zero
760 * means that no tasks are currently attached, therefore there is no
761 * way a task attached to that cgroup can fork (the other way to
762 * increment the count). So code holding cgroup_mutex can safely
763 * assume that if the count is zero, it will stay zero. Similarly, if
764 * a task holds cgroup_mutex on a cgroup with zero count, it
765 * knows that the cgroup won't be removed, as cgroup_rmdir()
768 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
769 * (usually) take cgroup_mutex. These are the two most performance
770 * critical pieces of code here. The exception occurs on cgroup_exit(),
771 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
772 * is taken, and if the cgroup count is zero, a usermode call made
773 * to the release agent with the name of the cgroup (path relative to
774 * the root of cgroup file system) as the argument.
776 * A cgroup can only be deleted if both its 'count' of using tasks
777 * is zero, and its list of 'children' cgroups is empty. Since all
778 * tasks in the system use _some_ cgroup, and since there is always at
779 * least one task in the system (init, pid == 1), therefore, top_cgroup
780 * always has either children cgroups and/or using tasks. So we don't
781 * need a special hack to ensure that top_cgroup cannot be deleted.
783 * The task_lock() exception
785 * The need for this exception arises from the action of
786 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
787 * another. It does so using cgroup_mutex, however there are
788 * several performance critical places that need to reference
789 * task->cgroup without the expense of grabbing a system global
790 * mutex. Therefore except as noted below, when dereferencing or, as
791 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
792 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
793 * the task_struct routinely used for such matters.
795 * P.S. One more locking exception. RCU is used to guard the
796 * update of a tasks cgroup pointer by cgroup_attach_task()
800 * A couple of forward declarations required, due to cyclic reference loop:
801 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
802 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
806 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
807 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, unsigned int);
808 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
809 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
);
810 static const struct inode_operations cgroup_dir_inode_operations
;
811 static const struct file_operations proc_cgroupstats_operations
;
813 static struct backing_dev_info cgroup_backing_dev_info
= {
815 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
818 static int alloc_css_id(struct cgroup_subsys
*ss
,
819 struct cgroup
*parent
, struct cgroup
*child
);
821 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
823 struct inode
*inode
= new_inode(sb
);
826 inode
->i_ino
= get_next_ino();
827 inode
->i_mode
= mode
;
828 inode
->i_uid
= current_fsuid();
829 inode
->i_gid
= current_fsgid();
830 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
831 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
836 static struct cgroup_name
*cgroup_alloc_name(struct dentry
*dentry
)
838 struct cgroup_name
*name
;
840 name
= kmalloc(sizeof(*name
) + dentry
->d_name
.len
+ 1, GFP_KERNEL
);
843 strcpy(name
->name
, dentry
->d_name
.name
);
847 static void cgroup_free_fn(struct work_struct
*work
)
849 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, destroy_work
);
850 struct cgroup_subsys
*ss
;
852 mutex_lock(&cgroup_mutex
);
854 * Release the subsystem state objects.
856 for_each_root_subsys(cgrp
->root
, ss
)
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
);
870 * Drop the active superblock reference that we took when we
871 * created the cgroup. This will free cgrp->root, if we are
872 * holding the last reference to @sb.
874 deactivate_super(cgrp
->root
->sb
);
877 * if we're getting rid of the cgroup, refcount should ensure
878 * that there are no pidlists left.
880 BUG_ON(!list_empty(&cgrp
->pidlists
));
882 simple_xattrs_free(&cgrp
->xattrs
);
884 kfree(rcu_dereference_raw(cgrp
->name
));
888 static void cgroup_free_rcu(struct rcu_head
*head
)
890 struct cgroup
*cgrp
= container_of(head
, struct cgroup
, rcu_head
);
892 INIT_WORK(&cgrp
->destroy_work
, cgroup_free_fn
);
893 schedule_work(&cgrp
->destroy_work
);
896 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
898 /* is dentry a directory ? if so, kfree() associated cgroup */
899 if (S_ISDIR(inode
->i_mode
)) {
900 struct cgroup
*cgrp
= dentry
->d_fsdata
;
902 BUG_ON(!(cgroup_is_dead(cgrp
)));
903 call_rcu(&cgrp
->rcu_head
, cgroup_free_rcu
);
905 struct cfent
*cfe
= __d_cfe(dentry
);
906 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
908 WARN_ONCE(!list_empty(&cfe
->node
) &&
909 cgrp
!= &cgrp
->root
->top_cgroup
,
910 "cfe still linked for %s\n", cfe
->type
->name
);
911 simple_xattrs_free(&cfe
->xattrs
);
917 static int cgroup_delete(const struct dentry
*d
)
922 static void remove_dir(struct dentry
*d
)
924 struct dentry
*parent
= dget(d
->d_parent
);
927 simple_rmdir(parent
->d_inode
, d
);
931 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
935 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
936 lockdep_assert_held(&cgroup_mutex
);
939 * If we're doing cleanup due to failure of cgroup_create(),
940 * the corresponding @cfe may not exist.
942 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
943 struct dentry
*d
= cfe
->dentry
;
945 if (cft
&& cfe
->type
!= cft
)
950 simple_unlink(cgrp
->dentry
->d_inode
, d
);
951 list_del_init(&cfe
->node
);
959 * cgroup_clear_dir - remove subsys files in a cgroup directory
960 * @cgrp: target cgroup
961 * @subsys_mask: mask of the subsystem ids whose files should be removed
963 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
965 struct cgroup_subsys
*ss
;
968 for_each_subsys(ss
, i
) {
969 struct cftype_set
*set
;
971 if (!test_bit(i
, &subsys_mask
))
973 list_for_each_entry(set
, &ss
->cftsets
, node
)
974 cgroup_addrm_files(cgrp
, NULL
, set
->cfts
, false);
979 * NOTE : the dentry must have been dget()'ed
981 static void cgroup_d_remove_dir(struct dentry
*dentry
)
983 struct dentry
*parent
;
985 parent
= dentry
->d_parent
;
986 spin_lock(&parent
->d_lock
);
987 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
988 list_del_init(&dentry
->d_u
.d_child
);
989 spin_unlock(&dentry
->d_lock
);
990 spin_unlock(&parent
->d_lock
);
995 * Call with cgroup_mutex held. Drops reference counts on modules, including
996 * any duplicate ones that parse_cgroupfs_options took. If this function
997 * returns an error, no reference counts are touched.
999 static int rebind_subsystems(struct cgroupfs_root
*root
,
1000 unsigned long added_mask
, unsigned removed_mask
)
1002 struct cgroup
*cgrp
= &root
->top_cgroup
;
1003 struct cgroup_subsys
*ss
;
1004 unsigned long pinned
= 0;
1007 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1008 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1010 /* Check that any added subsystems are currently free */
1011 for_each_subsys(ss
, i
) {
1012 if (!(added_mask
& (1 << i
)))
1015 /* is the subsystem mounted elsewhere? */
1016 if (ss
->root
!= &cgroup_dummy_root
) {
1021 /* pin the module */
1022 if (!try_module_get(ss
->module
)) {
1029 /* subsys could be missing if unloaded between parsing and here */
1030 if (added_mask
!= pinned
) {
1035 ret
= cgroup_populate_dir(cgrp
, added_mask
);
1040 * Nothing can fail from this point on. Remove files for the
1041 * removed subsystems and rebind each subsystem.
1043 cgroup_clear_dir(cgrp
, removed_mask
);
1045 for_each_subsys(ss
, i
) {
1046 unsigned long bit
= 1UL << i
;
1048 if (bit
& added_mask
) {
1049 /* We're binding this subsystem to this hierarchy */
1050 BUG_ON(cgrp
->subsys
[i
]);
1051 BUG_ON(!cgroup_dummy_top
->subsys
[i
]);
1052 BUG_ON(cgroup_dummy_top
->subsys
[i
]->cgroup
!= cgroup_dummy_top
);
1054 cgrp
->subsys
[i
] = cgroup_dummy_top
->subsys
[i
];
1055 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1056 list_move(&ss
->sibling
, &root
->subsys_list
);
1061 /* refcount was already taken, and we're keeping it */
1062 root
->subsys_mask
|= bit
;
1063 } else if (bit
& removed_mask
) {
1064 /* We're removing this subsystem */
1065 BUG_ON(cgrp
->subsys
[i
] != cgroup_dummy_top
->subsys
[i
]);
1066 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1069 ss
->bind(cgroup_dummy_top
);
1070 cgroup_dummy_top
->subsys
[i
]->cgroup
= cgroup_dummy_top
;
1071 cgrp
->subsys
[i
] = NULL
;
1072 cgroup_subsys
[i
]->root
= &cgroup_dummy_root
;
1073 list_move(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
1075 /* subsystem is now free - drop reference on module */
1076 module_put(ss
->module
);
1077 root
->subsys_mask
&= ~bit
;
1082 * Mark @root has finished binding subsystems. @root->subsys_mask
1083 * now matches the bound subsystems.
1085 root
->flags
|= CGRP_ROOT_SUBSYS_BOUND
;
1090 for_each_subsys(ss
, i
)
1091 if (pinned
& (1 << i
))
1092 module_put(ss
->module
);
1096 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1098 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1099 struct cgroup_subsys
*ss
;
1101 mutex_lock(&cgroup_root_mutex
);
1102 for_each_root_subsys(root
, ss
)
1103 seq_printf(seq
, ",%s", ss
->name
);
1104 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
)
1105 seq_puts(seq
, ",sane_behavior");
1106 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1107 seq_puts(seq
, ",noprefix");
1108 if (root
->flags
& CGRP_ROOT_XATTR
)
1109 seq_puts(seq
, ",xattr");
1110 if (strlen(root
->release_agent_path
))
1111 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1112 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
))
1113 seq_puts(seq
, ",clone_children");
1114 if (strlen(root
->name
))
1115 seq_printf(seq
, ",name=%s", root
->name
);
1116 mutex_unlock(&cgroup_root_mutex
);
1120 struct cgroup_sb_opts
{
1121 unsigned long subsys_mask
;
1122 unsigned long flags
;
1123 char *release_agent
;
1124 bool cpuset_clone_children
;
1126 /* User explicitly requested empty subsystem */
1129 struct cgroupfs_root
*new_root
;
1134 * Convert a hierarchy specifier into a bitmask of subsystems and
1135 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1136 * array. This function takes refcounts on subsystems to be used, unless it
1137 * returns error, in which case no refcounts are taken.
1139 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1141 char *token
, *o
= data
;
1142 bool all_ss
= false, one_ss
= false;
1143 unsigned long mask
= (unsigned long)-1;
1144 struct cgroup_subsys
*ss
;
1147 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1149 #ifdef CONFIG_CPUSETS
1150 mask
= ~(1UL << cpuset_subsys_id
);
1153 memset(opts
, 0, sizeof(*opts
));
1155 while ((token
= strsep(&o
, ",")) != NULL
) {
1158 if (!strcmp(token
, "none")) {
1159 /* Explicitly have no subsystems */
1163 if (!strcmp(token
, "all")) {
1164 /* Mutually exclusive option 'all' + subsystem name */
1170 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1171 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1174 if (!strcmp(token
, "noprefix")) {
1175 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1178 if (!strcmp(token
, "clone_children")) {
1179 opts
->cpuset_clone_children
= true;
1182 if (!strcmp(token
, "xattr")) {
1183 opts
->flags
|= CGRP_ROOT_XATTR
;
1186 if (!strncmp(token
, "release_agent=", 14)) {
1187 /* Specifying two release agents is forbidden */
1188 if (opts
->release_agent
)
1190 opts
->release_agent
=
1191 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1192 if (!opts
->release_agent
)
1196 if (!strncmp(token
, "name=", 5)) {
1197 const char *name
= token
+ 5;
1198 /* Can't specify an empty name */
1201 /* Must match [\w.-]+ */
1202 for (i
= 0; i
< strlen(name
); i
++) {
1206 if ((c
== '.') || (c
== '-') || (c
== '_'))
1210 /* Specifying two names is forbidden */
1213 opts
->name
= kstrndup(name
,
1214 MAX_CGROUP_ROOT_NAMELEN
- 1,
1222 for_each_subsys(ss
, i
) {
1223 if (strcmp(token
, ss
->name
))
1228 /* Mutually exclusive option 'all' + subsystem name */
1231 set_bit(i
, &opts
->subsys_mask
);
1236 if (i
== CGROUP_SUBSYS_COUNT
)
1241 * If the 'all' option was specified select all the subsystems,
1242 * otherwise if 'none', 'name=' and a subsystem name options
1243 * were not specified, let's default to 'all'
1245 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1246 for_each_subsys(ss
, i
)
1248 set_bit(i
, &opts
->subsys_mask
);
1250 /* Consistency checks */
1252 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1253 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1255 if (opts
->flags
& CGRP_ROOT_NOPREFIX
) {
1256 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1260 if (opts
->cpuset_clone_children
) {
1261 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1267 * Option noprefix was introduced just for backward compatibility
1268 * with the old cpuset, so we allow noprefix only if mounting just
1269 * the cpuset subsystem.
1271 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1275 /* Can't specify "none" and some subsystems */
1276 if (opts
->subsys_mask
&& opts
->none
)
1280 * We either have to specify by name or by subsystems. (So all
1281 * empty hierarchies must have a name).
1283 if (!opts
->subsys_mask
&& !opts
->name
)
1289 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1292 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1293 struct cgroup
*cgrp
= &root
->top_cgroup
;
1294 struct cgroup_sb_opts opts
;
1295 unsigned long added_mask
, removed_mask
;
1297 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1298 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1302 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1303 mutex_lock(&cgroup_mutex
);
1304 mutex_lock(&cgroup_root_mutex
);
1306 /* See what subsystems are wanted */
1307 ret
= parse_cgroupfs_options(data
, &opts
);
1311 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1312 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1313 task_tgid_nr(current
), current
->comm
);
1315 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1316 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1318 /* Don't allow flags or name to change at remount */
1319 if (((opts
.flags
^ root
->flags
) & CGRP_ROOT_OPTION_MASK
) ||
1320 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1321 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1322 opts
.flags
& CGRP_ROOT_OPTION_MASK
, opts
.name
?: "",
1323 root
->flags
& CGRP_ROOT_OPTION_MASK
, root
->name
);
1328 /* remounting is not allowed for populated hierarchies */
1329 if (root
->number_of_cgroups
> 1) {
1334 ret
= rebind_subsystems(root
, added_mask
, removed_mask
);
1338 if (opts
.release_agent
)
1339 strcpy(root
->release_agent_path
, opts
.release_agent
);
1341 kfree(opts
.release_agent
);
1343 mutex_unlock(&cgroup_root_mutex
);
1344 mutex_unlock(&cgroup_mutex
);
1345 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1349 static const struct super_operations cgroup_ops
= {
1350 .statfs
= simple_statfs
,
1351 .drop_inode
= generic_delete_inode
,
1352 .show_options
= cgroup_show_options
,
1353 .remount_fs
= cgroup_remount
,
1356 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1358 INIT_LIST_HEAD(&cgrp
->sibling
);
1359 INIT_LIST_HEAD(&cgrp
->children
);
1360 INIT_LIST_HEAD(&cgrp
->files
);
1361 INIT_LIST_HEAD(&cgrp
->cset_links
);
1362 INIT_LIST_HEAD(&cgrp
->release_list
);
1363 INIT_LIST_HEAD(&cgrp
->pidlists
);
1364 mutex_init(&cgrp
->pidlist_mutex
);
1365 INIT_LIST_HEAD(&cgrp
->event_list
);
1366 spin_lock_init(&cgrp
->event_list_lock
);
1367 simple_xattrs_init(&cgrp
->xattrs
);
1370 static void init_cgroup_root(struct cgroupfs_root
*root
)
1372 struct cgroup
*cgrp
= &root
->top_cgroup
;
1374 INIT_LIST_HEAD(&root
->subsys_list
);
1375 INIT_LIST_HEAD(&root
->root_list
);
1376 root
->number_of_cgroups
= 1;
1378 RCU_INIT_POINTER(cgrp
->name
, &root_cgroup_name
);
1379 init_cgroup_housekeeping(cgrp
);
1380 idr_init(&root
->cgroup_idr
);
1383 static int cgroup_init_root_id(struct cgroupfs_root
*root
, int start
, int end
)
1387 lockdep_assert_held(&cgroup_mutex
);
1388 lockdep_assert_held(&cgroup_root_mutex
);
1390 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, start
, end
,
1395 root
->hierarchy_id
= id
;
1399 static void cgroup_exit_root_id(struct cgroupfs_root
*root
)
1401 lockdep_assert_held(&cgroup_mutex
);
1402 lockdep_assert_held(&cgroup_root_mutex
);
1404 if (root
->hierarchy_id
) {
1405 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1406 root
->hierarchy_id
= 0;
1410 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1412 struct cgroup_sb_opts
*opts
= data
;
1413 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1415 /* If we asked for a name then it must match */
1416 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1420 * If we asked for subsystems (or explicitly for no
1421 * subsystems) then they must match
1423 if ((opts
->subsys_mask
|| opts
->none
)
1424 && (opts
->subsys_mask
!= root
->subsys_mask
))
1430 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1432 struct cgroupfs_root
*root
;
1434 if (!opts
->subsys_mask
&& !opts
->none
)
1437 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1439 return ERR_PTR(-ENOMEM
);
1441 init_cgroup_root(root
);
1444 * We need to set @root->subsys_mask now so that @root can be
1445 * matched by cgroup_test_super() before it finishes
1446 * initialization; otherwise, competing mounts with the same
1447 * options may try to bind the same subsystems instead of waiting
1448 * for the first one leading to unexpected mount errors.
1449 * SUBSYS_BOUND will be set once actual binding is complete.
1451 root
->subsys_mask
= opts
->subsys_mask
;
1452 root
->flags
= opts
->flags
;
1453 if (opts
->release_agent
)
1454 strcpy(root
->release_agent_path
, opts
->release_agent
);
1456 strcpy(root
->name
, opts
->name
);
1457 if (opts
->cpuset_clone_children
)
1458 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1462 static void cgroup_free_root(struct cgroupfs_root
*root
)
1465 /* hierarhcy ID shoulid already have been released */
1466 WARN_ON_ONCE(root
->hierarchy_id
);
1468 idr_destroy(&root
->cgroup_idr
);
1473 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1476 struct cgroup_sb_opts
*opts
= data
;
1478 /* If we don't have a new root, we can't set up a new sb */
1479 if (!opts
->new_root
)
1482 BUG_ON(!opts
->subsys_mask
&& !opts
->none
);
1484 ret
= set_anon_super(sb
, NULL
);
1488 sb
->s_fs_info
= opts
->new_root
;
1489 opts
->new_root
->sb
= sb
;
1491 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1492 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1493 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1494 sb
->s_op
= &cgroup_ops
;
1499 static int cgroup_get_rootdir(struct super_block
*sb
)
1501 static const struct dentry_operations cgroup_dops
= {
1502 .d_iput
= cgroup_diput
,
1503 .d_delete
= cgroup_delete
,
1506 struct inode
*inode
=
1507 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1512 inode
->i_fop
= &simple_dir_operations
;
1513 inode
->i_op
= &cgroup_dir_inode_operations
;
1514 /* directories start off with i_nlink == 2 (for "." entry) */
1516 sb
->s_root
= d_make_root(inode
);
1519 /* for everything else we want ->d_op set */
1520 sb
->s_d_op
= &cgroup_dops
;
1524 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1525 int flags
, const char *unused_dev_name
,
1528 struct cgroup_sb_opts opts
;
1529 struct cgroupfs_root
*root
;
1531 struct super_block
*sb
;
1532 struct cgroupfs_root
*new_root
;
1533 struct list_head tmp_links
;
1534 struct inode
*inode
;
1535 const struct cred
*cred
;
1537 /* First find the desired set of subsystems */
1538 mutex_lock(&cgroup_mutex
);
1539 ret
= parse_cgroupfs_options(data
, &opts
);
1540 mutex_unlock(&cgroup_mutex
);
1545 * Allocate a new cgroup root. We may not need it if we're
1546 * reusing an existing hierarchy.
1548 new_root
= cgroup_root_from_opts(&opts
);
1549 if (IS_ERR(new_root
)) {
1550 ret
= PTR_ERR(new_root
);
1553 opts
.new_root
= new_root
;
1555 /* Locate an existing or new sb for this hierarchy */
1556 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, 0, &opts
);
1559 cgroup_free_root(opts
.new_root
);
1563 root
= sb
->s_fs_info
;
1565 if (root
== opts
.new_root
) {
1566 /* We used the new root structure, so this is a new hierarchy */
1567 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1568 struct cgroupfs_root
*existing_root
;
1570 struct css_set
*cset
;
1572 BUG_ON(sb
->s_root
!= NULL
);
1574 ret
= cgroup_get_rootdir(sb
);
1576 goto drop_new_super
;
1577 inode
= sb
->s_root
->d_inode
;
1579 mutex_lock(&inode
->i_mutex
);
1580 mutex_lock(&cgroup_mutex
);
1581 mutex_lock(&cgroup_root_mutex
);
1583 root_cgrp
->id
= idr_alloc(&root
->cgroup_idr
, root_cgrp
,
1585 if (root_cgrp
->id
< 0)
1588 /* Check for name clashes with existing mounts */
1590 if (strlen(root
->name
))
1591 for_each_active_root(existing_root
)
1592 if (!strcmp(existing_root
->name
, root
->name
))
1596 * We're accessing css_set_count without locking
1597 * css_set_lock here, but that's OK - it can only be
1598 * increased by someone holding cgroup_lock, and
1599 * that's us. The worst that can happen is that we
1600 * have some link structures left over
1602 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1606 /* ID 0 is reserved for dummy root, 1 for unified hierarchy */
1607 ret
= cgroup_init_root_id(root
, 2, 0);
1611 sb
->s_root
->d_fsdata
= root_cgrp
;
1612 root_cgrp
->dentry
= sb
->s_root
;
1615 * We're inside get_sb() and will call lookup_one_len() to
1616 * create the root files, which doesn't work if SELinux is
1617 * in use. The following cred dancing somehow works around
1618 * it. See 2ce9738ba ("cgroupfs: use init_cred when
1619 * populating new cgroupfs mount") for more details.
1621 cred
= override_creds(&init_cred
);
1623 ret
= cgroup_addrm_files(root_cgrp
, NULL
, cgroup_base_files
, true);
1627 ret
= rebind_subsystems(root
, root
->subsys_mask
, 0);
1634 * There must be no failure case after here, since rebinding
1635 * takes care of subsystems' refcounts, which are explicitly
1636 * dropped in the failure exit path.
1639 list_add(&root
->root_list
, &cgroup_roots
);
1640 cgroup_root_count
++;
1642 /* Link the top cgroup in this hierarchy into all
1643 * the css_set objects */
1644 write_lock(&css_set_lock
);
1645 hash_for_each(css_set_table
, i
, cset
, hlist
)
1646 link_css_set(&tmp_links
, cset
, root_cgrp
);
1647 write_unlock(&css_set_lock
);
1649 free_cgrp_cset_links(&tmp_links
);
1651 BUG_ON(!list_empty(&root_cgrp
->children
));
1652 BUG_ON(root
->number_of_cgroups
!= 1);
1654 mutex_unlock(&cgroup_root_mutex
);
1655 mutex_unlock(&cgroup_mutex
);
1656 mutex_unlock(&inode
->i_mutex
);
1659 * We re-used an existing hierarchy - the new root (if
1660 * any) is not needed
1662 cgroup_free_root(opts
.new_root
);
1664 if ((root
->flags
^ opts
.flags
) & CGRP_ROOT_OPTION_MASK
) {
1665 if ((root
->flags
| opts
.flags
) & CGRP_ROOT_SANE_BEHAVIOR
) {
1666 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1668 goto drop_new_super
;
1670 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1675 kfree(opts
.release_agent
);
1677 return dget(sb
->s_root
);
1680 free_cgrp_cset_links(&tmp_links
);
1681 cgroup_addrm_files(&root
->top_cgroup
, NULL
, cgroup_base_files
, false);
1684 cgroup_exit_root_id(root
);
1685 mutex_unlock(&cgroup_root_mutex
);
1686 mutex_unlock(&cgroup_mutex
);
1687 mutex_unlock(&inode
->i_mutex
);
1689 deactivate_locked_super(sb
);
1691 kfree(opts
.release_agent
);
1693 return ERR_PTR(ret
);
1696 static void cgroup_kill_sb(struct super_block
*sb
) {
1697 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1698 struct cgroup
*cgrp
= &root
->top_cgroup
;
1699 struct cgrp_cset_link
*link
, *tmp_link
;
1704 BUG_ON(root
->number_of_cgroups
!= 1);
1705 BUG_ON(!list_empty(&cgrp
->children
));
1707 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1708 mutex_lock(&cgroup_mutex
);
1709 mutex_lock(&cgroup_root_mutex
);
1711 /* Rebind all subsystems back to the default hierarchy */
1712 if (root
->flags
& CGRP_ROOT_SUBSYS_BOUND
) {
1713 ret
= rebind_subsystems(root
, 0, root
->subsys_mask
);
1714 /* Shouldn't be able to fail ... */
1719 * Release all the links from cset_links to this hierarchy's
1722 write_lock(&css_set_lock
);
1724 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1725 list_del(&link
->cset_link
);
1726 list_del(&link
->cgrp_link
);
1729 write_unlock(&css_set_lock
);
1731 if (!list_empty(&root
->root_list
)) {
1732 list_del(&root
->root_list
);
1733 cgroup_root_count
--;
1736 cgroup_exit_root_id(root
);
1738 mutex_unlock(&cgroup_root_mutex
);
1739 mutex_unlock(&cgroup_mutex
);
1740 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1742 simple_xattrs_free(&cgrp
->xattrs
);
1744 kill_litter_super(sb
);
1745 cgroup_free_root(root
);
1748 static struct file_system_type cgroup_fs_type
= {
1750 .mount
= cgroup_mount
,
1751 .kill_sb
= cgroup_kill_sb
,
1754 static struct kobject
*cgroup_kobj
;
1757 * cgroup_path - generate the path of a cgroup
1758 * @cgrp: the cgroup in question
1759 * @buf: the buffer to write the path into
1760 * @buflen: the length of the buffer
1762 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1764 * We can't generate cgroup path using dentry->d_name, as accessing
1765 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1766 * inode's i_mutex, while on the other hand cgroup_path() can be called
1767 * with some irq-safe spinlocks held.
1769 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1771 int ret
= -ENAMETOOLONG
;
1774 if (!cgrp
->parent
) {
1775 if (strlcpy(buf
, "/", buflen
) >= buflen
)
1776 return -ENAMETOOLONG
;
1780 start
= buf
+ buflen
- 1;
1785 const char *name
= cgroup_name(cgrp
);
1789 if ((start
-= len
) < buf
)
1791 memcpy(start
, name
, len
);
1797 cgrp
= cgrp
->parent
;
1798 } while (cgrp
->parent
);
1800 memmove(buf
, start
, buf
+ buflen
- start
);
1805 EXPORT_SYMBOL_GPL(cgroup_path
);
1808 * task_cgroup_path_from_hierarchy - cgroup path of a task on a hierarchy
1809 * @task: target task
1810 * @hierarchy_id: the hierarchy to look up @task's cgroup from
1811 * @buf: the buffer to write the path into
1812 * @buflen: the length of the buffer
1814 * Determine @task's cgroup on the hierarchy specified by @hierarchy_id and
1815 * copy its path into @buf. This function grabs cgroup_mutex and shouldn't
1816 * be used inside locks used by cgroup controller callbacks.
1818 int task_cgroup_path_from_hierarchy(struct task_struct
*task
, int hierarchy_id
,
1819 char *buf
, size_t buflen
)
1821 struct cgroupfs_root
*root
;
1822 struct cgroup
*cgrp
= NULL
;
1825 mutex_lock(&cgroup_mutex
);
1827 root
= idr_find(&cgroup_hierarchy_idr
, hierarchy_id
);
1829 cgrp
= task_cgroup_from_root(task
, root
);
1830 ret
= cgroup_path(cgrp
, buf
, buflen
);
1833 mutex_unlock(&cgroup_mutex
);
1837 EXPORT_SYMBOL_GPL(task_cgroup_path_from_hierarchy
);
1840 * Control Group taskset
1842 struct task_and_cgroup
{
1843 struct task_struct
*task
;
1844 struct cgroup
*cgrp
;
1845 struct css_set
*cset
;
1848 struct cgroup_taskset
{
1849 struct task_and_cgroup single
;
1850 struct flex_array
*tc_array
;
1853 struct cgroup
*cur_cgrp
;
1857 * cgroup_taskset_first - reset taskset and return the first task
1858 * @tset: taskset of interest
1860 * @tset iteration is initialized and the first task is returned.
1862 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1864 if (tset
->tc_array
) {
1866 return cgroup_taskset_next(tset
);
1868 tset
->cur_cgrp
= tset
->single
.cgrp
;
1869 return tset
->single
.task
;
1872 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1875 * cgroup_taskset_next - iterate to the next task in taskset
1876 * @tset: taskset of interest
1878 * Return the next task in @tset. Iteration must have been initialized
1879 * with cgroup_taskset_first().
1881 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1883 struct task_and_cgroup
*tc
;
1885 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1888 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1889 tset
->cur_cgrp
= tc
->cgrp
;
1892 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1895 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1896 * @tset: taskset of interest
1898 * Return the cgroup for the current (last returned) task of @tset. This
1899 * function must be preceded by either cgroup_taskset_first() or
1900 * cgroup_taskset_next().
1902 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1904 return tset
->cur_cgrp
;
1906 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1909 * cgroup_taskset_size - return the number of tasks in taskset
1910 * @tset: taskset of interest
1912 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1914 return tset
->tc_array
? tset
->tc_array_len
: 1;
1916 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1920 * cgroup_task_migrate - move a task from one cgroup to another.
1922 * Must be called with cgroup_mutex and threadgroup locked.
1924 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
1925 struct task_struct
*tsk
,
1926 struct css_set
*new_cset
)
1928 struct css_set
*old_cset
;
1931 * We are synchronized through threadgroup_lock() against PF_EXITING
1932 * setting such that we can't race against cgroup_exit() changing the
1933 * css_set to init_css_set and dropping the old one.
1935 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1936 old_cset
= task_css_set(tsk
);
1939 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
1942 /* Update the css_set linked lists if we're using them */
1943 write_lock(&css_set_lock
);
1944 if (!list_empty(&tsk
->cg_list
))
1945 list_move(&tsk
->cg_list
, &new_cset
->tasks
);
1946 write_unlock(&css_set_lock
);
1949 * We just gained a reference on old_cset by taking it from the
1950 * task. As trading it for new_cset is protected by cgroup_mutex,
1951 * we're safe to drop it here; it will be freed under RCU.
1953 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
1954 put_css_set(old_cset
);
1958 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1959 * @cgrp: the cgroup to attach to
1960 * @tsk: the task or the leader of the threadgroup to be attached
1961 * @threadgroup: attach the whole threadgroup?
1963 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1964 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1966 static int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
,
1969 int retval
, i
, group_size
;
1970 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1971 struct cgroupfs_root
*root
= cgrp
->root
;
1972 /* threadgroup list cursor and array */
1973 struct task_struct
*leader
= tsk
;
1974 struct task_and_cgroup
*tc
;
1975 struct flex_array
*group
;
1976 struct cgroup_taskset tset
= { };
1979 * step 0: in order to do expensive, possibly blocking operations for
1980 * every thread, we cannot iterate the thread group list, since it needs
1981 * rcu or tasklist locked. instead, build an array of all threads in the
1982 * group - group_rwsem prevents new threads from appearing, and if
1983 * threads exit, this will just be an over-estimate.
1986 group_size
= get_nr_threads(tsk
);
1989 /* flex_array supports very large thread-groups better than kmalloc. */
1990 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
1993 /* pre-allocate to guarantee space while iterating in rcu read-side. */
1994 retval
= flex_array_prealloc(group
, 0, group_size
, GFP_KERNEL
);
1996 goto out_free_group_list
;
2000 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2001 * already PF_EXITING could be freed from underneath us unless we
2002 * take an rcu_read_lock.
2006 struct task_and_cgroup ent
;
2008 /* @tsk either already exited or can't exit until the end */
2009 if (tsk
->flags
& PF_EXITING
)
2012 /* as per above, nr_threads may decrease, but not increase. */
2013 BUG_ON(i
>= group_size
);
2015 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2016 /* nothing to do if this task is already in the cgroup */
2017 if (ent
.cgrp
== cgrp
)
2020 * saying GFP_ATOMIC has no effect here because we did prealloc
2021 * earlier, but it's good form to communicate our expectations.
2023 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2024 BUG_ON(retval
!= 0);
2029 } while_each_thread(leader
, tsk
);
2031 /* remember the number of threads in the array for later. */
2033 tset
.tc_array
= group
;
2034 tset
.tc_array_len
= group_size
;
2036 /* methods shouldn't be called if no task is actually migrating */
2039 goto out_free_group_list
;
2042 * step 1: check that we can legitimately attach to the cgroup.
2044 for_each_root_subsys(root
, ss
) {
2045 if (ss
->can_attach
) {
2046 retval
= ss
->can_attach(cgrp
, &tset
);
2049 goto out_cancel_attach
;
2055 * step 2: make sure css_sets exist for all threads to be migrated.
2056 * we use find_css_set, which allocates a new one if necessary.
2058 for (i
= 0; i
< group_size
; i
++) {
2059 struct css_set
*old_cset
;
2061 tc
= flex_array_get(group
, i
);
2062 old_cset
= task_css_set(tc
->task
);
2063 tc
->cset
= find_css_set(old_cset
, cgrp
);
2066 goto out_put_css_set_refs
;
2071 * step 3: now that we're guaranteed success wrt the css_sets,
2072 * proceed to move all tasks to the new cgroup. There are no
2073 * failure cases after here, so this is the commit point.
2075 for (i
= 0; i
< group_size
; i
++) {
2076 tc
= flex_array_get(group
, i
);
2077 cgroup_task_migrate(tc
->cgrp
, tc
->task
, tc
->cset
);
2079 /* nothing is sensitive to fork() after this point. */
2082 * step 4: do subsystem attach callbacks.
2084 for_each_root_subsys(root
, ss
) {
2086 ss
->attach(cgrp
, &tset
);
2090 * step 5: success! and cleanup
2093 out_put_css_set_refs
:
2095 for (i
= 0; i
< group_size
; i
++) {
2096 tc
= flex_array_get(group
, i
);
2099 put_css_set(tc
->cset
);
2104 for_each_root_subsys(root
, ss
) {
2105 if (ss
== failed_ss
)
2107 if (ss
->cancel_attach
)
2108 ss
->cancel_attach(cgrp
, &tset
);
2111 out_free_group_list
:
2112 flex_array_free(group
);
2117 * Find the task_struct of the task to attach by vpid and pass it along to the
2118 * function to attach either it or all tasks in its threadgroup. Will lock
2119 * cgroup_mutex and threadgroup; may take task_lock of task.
2121 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2123 struct task_struct
*tsk
;
2124 const struct cred
*cred
= current_cred(), *tcred
;
2127 if (!cgroup_lock_live_group(cgrp
))
2133 tsk
= find_task_by_vpid(pid
);
2137 goto out_unlock_cgroup
;
2140 * even if we're attaching all tasks in the thread group, we
2141 * only need to check permissions on one of them.
2143 tcred
= __task_cred(tsk
);
2144 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2145 !uid_eq(cred
->euid
, tcred
->uid
) &&
2146 !uid_eq(cred
->euid
, tcred
->suid
)) {
2149 goto out_unlock_cgroup
;
2155 tsk
= tsk
->group_leader
;
2158 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2159 * trapped in a cpuset, or RT worker may be born in a cgroup
2160 * with no rt_runtime allocated. Just say no.
2162 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2165 goto out_unlock_cgroup
;
2168 get_task_struct(tsk
);
2171 threadgroup_lock(tsk
);
2173 if (!thread_group_leader(tsk
)) {
2175 * a race with de_thread from another thread's exec()
2176 * may strip us of our leadership, if this happens,
2177 * there is no choice but to throw this task away and
2178 * try again; this is
2179 * "double-double-toil-and-trouble-check locking".
2181 threadgroup_unlock(tsk
);
2182 put_task_struct(tsk
);
2183 goto retry_find_task
;
2187 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2189 threadgroup_unlock(tsk
);
2191 put_task_struct(tsk
);
2193 mutex_unlock(&cgroup_mutex
);
2198 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2199 * @from: attach to all cgroups of a given task
2200 * @tsk: the task to be attached
2202 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2204 struct cgroupfs_root
*root
;
2207 mutex_lock(&cgroup_mutex
);
2208 for_each_active_root(root
) {
2209 struct cgroup
*from_cgrp
= task_cgroup_from_root(from
, root
);
2211 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2215 mutex_unlock(&cgroup_mutex
);
2219 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2221 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2223 return attach_task_by_pid(cgrp
, pid
, false);
2226 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2228 return attach_task_by_pid(cgrp
, tgid
, true);
2231 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2234 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2235 if (strlen(buffer
) >= PATH_MAX
)
2237 if (!cgroup_lock_live_group(cgrp
))
2239 mutex_lock(&cgroup_root_mutex
);
2240 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2241 mutex_unlock(&cgroup_root_mutex
);
2242 mutex_unlock(&cgroup_mutex
);
2246 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2247 struct seq_file
*seq
)
2249 if (!cgroup_lock_live_group(cgrp
))
2251 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2252 seq_putc(seq
, '\n');
2253 mutex_unlock(&cgroup_mutex
);
2257 static int cgroup_sane_behavior_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2258 struct seq_file
*seq
)
2260 seq_printf(seq
, "%d\n", cgroup_sane_behavior(cgrp
));
2264 /* A buffer size big enough for numbers or short strings */
2265 #define CGROUP_LOCAL_BUFFER_SIZE 64
2267 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2269 const char __user
*userbuf
,
2270 size_t nbytes
, loff_t
*unused_ppos
)
2272 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2278 if (nbytes
>= sizeof(buffer
))
2280 if (copy_from_user(buffer
, userbuf
, nbytes
))
2283 buffer
[nbytes
] = 0; /* nul-terminate */
2284 if (cft
->write_u64
) {
2285 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2288 retval
= cft
->write_u64(cgrp
, cft
, val
);
2290 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2293 retval
= cft
->write_s64(cgrp
, cft
, val
);
2300 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2302 const char __user
*userbuf
,
2303 size_t nbytes
, loff_t
*unused_ppos
)
2305 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2307 size_t max_bytes
= cft
->max_write_len
;
2308 char *buffer
= local_buffer
;
2311 max_bytes
= sizeof(local_buffer
) - 1;
2312 if (nbytes
>= max_bytes
)
2314 /* Allocate a dynamic buffer if we need one */
2315 if (nbytes
>= sizeof(local_buffer
)) {
2316 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2320 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2325 buffer
[nbytes
] = 0; /* nul-terminate */
2326 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2330 if (buffer
!= local_buffer
)
2335 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2336 size_t nbytes
, loff_t
*ppos
)
2338 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2339 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2341 if (cgroup_is_dead(cgrp
))
2344 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2345 if (cft
->write_u64
|| cft
->write_s64
)
2346 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2347 if (cft
->write_string
)
2348 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2350 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2351 return ret
? ret
: nbytes
;
2356 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2358 char __user
*buf
, size_t nbytes
,
2361 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2362 u64 val
= cft
->read_u64(cgrp
, cft
);
2363 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2365 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2368 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2370 char __user
*buf
, size_t nbytes
,
2373 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2374 s64 val
= cft
->read_s64(cgrp
, cft
);
2375 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2377 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2380 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2381 size_t nbytes
, loff_t
*ppos
)
2383 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2384 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2386 if (cgroup_is_dead(cgrp
))
2390 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2392 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2394 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2399 * seqfile ops/methods for returning structured data. Currently just
2400 * supports string->u64 maps, but can be extended in future.
2403 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2405 struct seq_file
*sf
= cb
->state
;
2406 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2409 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2411 struct cfent
*cfe
= m
->private;
2412 struct cftype
*cft
= cfe
->type
;
2413 struct cgroup
*cgrp
= __d_cgrp(cfe
->dentry
->d_parent
);
2415 if (cft
->read_map
) {
2416 struct cgroup_map_cb cb
= {
2417 .fill
= cgroup_map_add
,
2420 return cft
->read_map(cgrp
, cft
, &cb
);
2422 return cft
->read_seq_string(cgrp
, cft
, m
);
2425 static const struct file_operations cgroup_seqfile_operations
= {
2427 .write
= cgroup_file_write
,
2428 .llseek
= seq_lseek
,
2429 .release
= single_release
,
2432 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2438 err
= generic_file_open(inode
, file
);
2441 cfe
= __d_cfe(file
->f_dentry
);
2444 if (cft
->read_map
|| cft
->read_seq_string
) {
2445 file
->f_op
= &cgroup_seqfile_operations
;
2446 err
= single_open(file
, cgroup_seqfile_show
, cfe
);
2447 } else if (cft
->open
) {
2448 err
= cft
->open(inode
, file
);
2454 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2456 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2458 return cft
->release(inode
, file
);
2463 * cgroup_rename - Only allow simple rename of directories in place.
2465 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2466 struct inode
*new_dir
, struct dentry
*new_dentry
)
2469 struct cgroup_name
*name
, *old_name
;
2470 struct cgroup
*cgrp
;
2473 * It's convinient to use parent dir's i_mutex to protected
2476 lockdep_assert_held(&old_dir
->i_mutex
);
2478 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2480 if (new_dentry
->d_inode
)
2482 if (old_dir
!= new_dir
)
2485 cgrp
= __d_cgrp(old_dentry
);
2488 * This isn't a proper migration and its usefulness is very
2489 * limited. Disallow if sane_behavior.
2491 if (cgroup_sane_behavior(cgrp
))
2494 name
= cgroup_alloc_name(new_dentry
);
2498 ret
= simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2504 old_name
= rcu_dereference_protected(cgrp
->name
, true);
2505 rcu_assign_pointer(cgrp
->name
, name
);
2507 kfree_rcu(old_name
, rcu_head
);
2511 static struct simple_xattrs
*__d_xattrs(struct dentry
*dentry
)
2513 if (S_ISDIR(dentry
->d_inode
->i_mode
))
2514 return &__d_cgrp(dentry
)->xattrs
;
2516 return &__d_cfe(dentry
)->xattrs
;
2519 static inline int xattr_enabled(struct dentry
*dentry
)
2521 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
2522 return root
->flags
& CGRP_ROOT_XATTR
;
2525 static bool is_valid_xattr(const char *name
)
2527 if (!strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) ||
2528 !strncmp(name
, XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
))
2533 static int cgroup_setxattr(struct dentry
*dentry
, const char *name
,
2534 const void *val
, size_t size
, int flags
)
2536 if (!xattr_enabled(dentry
))
2538 if (!is_valid_xattr(name
))
2540 return simple_xattr_set(__d_xattrs(dentry
), name
, val
, size
, flags
);
2543 static int cgroup_removexattr(struct dentry
*dentry
, const char *name
)
2545 if (!xattr_enabled(dentry
))
2547 if (!is_valid_xattr(name
))
2549 return simple_xattr_remove(__d_xattrs(dentry
), name
);
2552 static ssize_t
cgroup_getxattr(struct dentry
*dentry
, const char *name
,
2553 void *buf
, size_t size
)
2555 if (!xattr_enabled(dentry
))
2557 if (!is_valid_xattr(name
))
2559 return simple_xattr_get(__d_xattrs(dentry
), name
, buf
, size
);
2562 static ssize_t
cgroup_listxattr(struct dentry
*dentry
, char *buf
, size_t size
)
2564 if (!xattr_enabled(dentry
))
2566 return simple_xattr_list(__d_xattrs(dentry
), buf
, size
);
2569 static const struct file_operations cgroup_file_operations
= {
2570 .read
= cgroup_file_read
,
2571 .write
= cgroup_file_write
,
2572 .llseek
= generic_file_llseek
,
2573 .open
= cgroup_file_open
,
2574 .release
= cgroup_file_release
,
2577 static const struct inode_operations cgroup_file_inode_operations
= {
2578 .setxattr
= cgroup_setxattr
,
2579 .getxattr
= cgroup_getxattr
,
2580 .listxattr
= cgroup_listxattr
,
2581 .removexattr
= cgroup_removexattr
,
2584 static const struct inode_operations cgroup_dir_inode_operations
= {
2585 .lookup
= cgroup_lookup
,
2586 .mkdir
= cgroup_mkdir
,
2587 .rmdir
= cgroup_rmdir
,
2588 .rename
= cgroup_rename
,
2589 .setxattr
= cgroup_setxattr
,
2590 .getxattr
= cgroup_getxattr
,
2591 .listxattr
= cgroup_listxattr
,
2592 .removexattr
= cgroup_removexattr
,
2595 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
2597 if (dentry
->d_name
.len
> NAME_MAX
)
2598 return ERR_PTR(-ENAMETOOLONG
);
2599 d_add(dentry
, NULL
);
2604 * Check if a file is a control file
2606 static inline struct cftype
*__file_cft(struct file
*file
)
2608 if (file_inode(file
)->i_fop
!= &cgroup_file_operations
)
2609 return ERR_PTR(-EINVAL
);
2610 return __d_cft(file
->f_dentry
);
2613 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2614 struct super_block
*sb
)
2616 struct inode
*inode
;
2620 if (dentry
->d_inode
)
2623 inode
= cgroup_new_inode(mode
, sb
);
2627 if (S_ISDIR(mode
)) {
2628 inode
->i_op
= &cgroup_dir_inode_operations
;
2629 inode
->i_fop
= &simple_dir_operations
;
2631 /* start off with i_nlink == 2 (for "." entry) */
2633 inc_nlink(dentry
->d_parent
->d_inode
);
2636 * Control reaches here with cgroup_mutex held.
2637 * @inode->i_mutex should nest outside cgroup_mutex but we
2638 * want to populate it immediately without releasing
2639 * cgroup_mutex. As @inode isn't visible to anyone else
2640 * yet, trylock will always succeed without affecting
2643 WARN_ON_ONCE(!mutex_trylock(&inode
->i_mutex
));
2644 } else if (S_ISREG(mode
)) {
2646 inode
->i_fop
= &cgroup_file_operations
;
2647 inode
->i_op
= &cgroup_file_inode_operations
;
2649 d_instantiate(dentry
, inode
);
2650 dget(dentry
); /* Extra count - pin the dentry in core */
2655 * cgroup_file_mode - deduce file mode of a control file
2656 * @cft: the control file in question
2658 * returns cft->mode if ->mode is not 0
2659 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2660 * returns S_IRUGO if it has only a read handler
2661 * returns S_IWUSR if it has only a write hander
2663 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2670 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2671 cft
->read_map
|| cft
->read_seq_string
)
2674 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2675 cft
->write_string
|| cft
->trigger
)
2681 static int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2684 struct dentry
*dir
= cgrp
->dentry
;
2685 struct cgroup
*parent
= __d_cgrp(dir
);
2686 struct dentry
*dentry
;
2690 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2692 if (subsys
&& !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
2693 strcpy(name
, subsys
->name
);
2696 strcat(name
, cft
->name
);
2698 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2700 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2704 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2705 if (IS_ERR(dentry
)) {
2706 error
= PTR_ERR(dentry
);
2710 cfe
->type
= (void *)cft
;
2711 cfe
->dentry
= dentry
;
2712 dentry
->d_fsdata
= cfe
;
2713 simple_xattrs_init(&cfe
->xattrs
);
2715 mode
= cgroup_file_mode(cft
);
2716 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2718 list_add_tail(&cfe
->node
, &parent
->files
);
2728 * cgroup_addrm_files - add or remove files to a cgroup directory
2729 * @cgrp: the target cgroup
2730 * @subsys: the subsystem of files to be added
2731 * @cfts: array of cftypes to be added
2732 * @is_add: whether to add or remove
2734 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2735 * All @cfts should belong to @subsys. For removals, this function never
2736 * fails. If addition fails, this function doesn't remove files already
2737 * added. The caller is responsible for cleaning up.
2739 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2740 struct cftype cfts
[], bool is_add
)
2745 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
2746 lockdep_assert_held(&cgroup_mutex
);
2748 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2749 /* does cft->flags tell us to skip this file on @cgrp? */
2750 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_sane_behavior(cgrp
))
2752 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2754 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2758 ret
= cgroup_add_file(cgrp
, subsys
, cft
);
2760 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2765 cgroup_rm_file(cgrp
, cft
);
2771 static void cgroup_cfts_prepare(void)
2772 __acquires(&cgroup_mutex
)
2775 * Thanks to the entanglement with vfs inode locking, we can't walk
2776 * the existing cgroups under cgroup_mutex and create files.
2777 * Instead, we use cgroup_for_each_descendant_pre() and drop RCU
2778 * read lock before calling cgroup_addrm_files().
2780 mutex_lock(&cgroup_mutex
);
2783 static int cgroup_cfts_commit(struct cgroup_subsys
*ss
,
2784 struct cftype
*cfts
, bool is_add
)
2785 __releases(&cgroup_mutex
)
2788 struct cgroup
*cgrp
, *root
= &ss
->root
->top_cgroup
;
2789 struct super_block
*sb
= ss
->root
->sb
;
2790 struct dentry
*prev
= NULL
;
2791 struct inode
*inode
;
2795 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2796 if (!cfts
|| ss
->root
== &cgroup_dummy_root
||
2797 !atomic_inc_not_zero(&sb
->s_active
)) {
2798 mutex_unlock(&cgroup_mutex
);
2803 * All cgroups which are created after we drop cgroup_mutex will
2804 * have the updated set of files, so we only need to update the
2805 * cgroups created before the current @cgroup_serial_nr_next.
2807 update_before
= cgroup_serial_nr_next
;
2809 mutex_unlock(&cgroup_mutex
);
2811 /* @root always needs to be updated */
2812 inode
= root
->dentry
->d_inode
;
2813 mutex_lock(&inode
->i_mutex
);
2814 mutex_lock(&cgroup_mutex
);
2815 ret
= cgroup_addrm_files(root
, ss
, cfts
, is_add
);
2816 mutex_unlock(&cgroup_mutex
);
2817 mutex_unlock(&inode
->i_mutex
);
2822 /* add/rm files for all cgroups created before */
2824 cgroup_for_each_descendant_pre(cgrp
, root
) {
2825 if (cgroup_is_dead(cgrp
))
2828 inode
= cgrp
->dentry
->d_inode
;
2833 prev
= cgrp
->dentry
;
2835 mutex_lock(&inode
->i_mutex
);
2836 mutex_lock(&cgroup_mutex
);
2837 if (cgrp
->serial_nr
< update_before
&& !cgroup_is_dead(cgrp
))
2838 ret
= cgroup_addrm_files(cgrp
, ss
, cfts
, is_add
);
2839 mutex_unlock(&cgroup_mutex
);
2840 mutex_unlock(&inode
->i_mutex
);
2849 deactivate_super(sb
);
2854 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2855 * @ss: target cgroup subsystem
2856 * @cfts: zero-length name terminated array of cftypes
2858 * Register @cfts to @ss. Files described by @cfts are created for all
2859 * existing cgroups to which @ss is attached and all future cgroups will
2860 * have them too. This function can be called anytime whether @ss is
2863 * Returns 0 on successful registration, -errno on failure. Note that this
2864 * function currently returns 0 as long as @cfts registration is successful
2865 * even if some file creation attempts on existing cgroups fail.
2867 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2869 struct cftype_set
*set
;
2872 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2876 cgroup_cfts_prepare();
2878 list_add_tail(&set
->node
, &ss
->cftsets
);
2879 ret
= cgroup_cfts_commit(ss
, cfts
, true);
2881 cgroup_rm_cftypes(ss
, cfts
);
2884 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2887 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2888 * @ss: target cgroup subsystem
2889 * @cfts: zero-length name terminated array of cftypes
2891 * Unregister @cfts from @ss. Files described by @cfts are removed from
2892 * all existing cgroups to which @ss is attached and all future cgroups
2893 * won't have them either. This function can be called anytime whether @ss
2894 * is attached or not.
2896 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2897 * registered with @ss.
2899 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2901 struct cftype_set
*set
;
2903 cgroup_cfts_prepare();
2905 list_for_each_entry(set
, &ss
->cftsets
, node
) {
2906 if (set
->cfts
== cfts
) {
2907 list_del(&set
->node
);
2909 cgroup_cfts_commit(ss
, cfts
, false);
2914 cgroup_cfts_commit(ss
, NULL
, false);
2919 * cgroup_task_count - count the number of tasks in a cgroup.
2920 * @cgrp: the cgroup in question
2922 * Return the number of tasks in the cgroup.
2924 int cgroup_task_count(const struct cgroup
*cgrp
)
2927 struct cgrp_cset_link
*link
;
2929 read_lock(&css_set_lock
);
2930 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
2931 count
+= atomic_read(&link
->cset
->refcount
);
2932 read_unlock(&css_set_lock
);
2937 * Advance a list_head iterator. The iterator should be positioned at
2938 * the start of a css_set
2940 static void cgroup_advance_iter(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2942 struct list_head
*l
= it
->cset_link
;
2943 struct cgrp_cset_link
*link
;
2944 struct css_set
*cset
;
2946 /* Advance to the next non-empty css_set */
2949 if (l
== &cgrp
->cset_links
) {
2950 it
->cset_link
= NULL
;
2953 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
2955 } while (list_empty(&cset
->tasks
));
2957 it
->task
= cset
->tasks
.next
;
2961 * To reduce the fork() overhead for systems that are not actually
2962 * using their cgroups capability, we don't maintain the lists running
2963 * through each css_set to its tasks until we see the list actually
2964 * used - in other words after the first call to cgroup_iter_start().
2966 static void cgroup_enable_task_cg_lists(void)
2968 struct task_struct
*p
, *g
;
2969 write_lock(&css_set_lock
);
2970 use_task_css_set_links
= 1;
2972 * We need tasklist_lock because RCU is not safe against
2973 * while_each_thread(). Besides, a forking task that has passed
2974 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2975 * is not guaranteed to have its child immediately visible in the
2976 * tasklist if we walk through it with RCU.
2978 read_lock(&tasklist_lock
);
2979 do_each_thread(g
, p
) {
2982 * We should check if the process is exiting, otherwise
2983 * it will race with cgroup_exit() in that the list
2984 * entry won't be deleted though the process has exited.
2986 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2987 list_add(&p
->cg_list
, &task_css_set(p
)->tasks
);
2989 } while_each_thread(g
, p
);
2990 read_unlock(&tasklist_lock
);
2991 write_unlock(&css_set_lock
);
2995 * cgroup_next_sibling - find the next sibling of a given cgroup
2996 * @pos: the current cgroup
2998 * This function returns the next sibling of @pos and should be called
2999 * under RCU read lock. The only requirement is that @pos is accessible.
3000 * The next sibling is guaranteed to be returned regardless of @pos's
3003 struct cgroup
*cgroup_next_sibling(struct cgroup
*pos
)
3005 struct cgroup
*next
;
3007 WARN_ON_ONCE(!rcu_read_lock_held());
3010 * @pos could already have been removed. Once a cgroup is removed,
3011 * its ->sibling.next is no longer updated when its next sibling
3012 * changes. As CGRP_DEAD assertion is serialized and happens
3013 * before the cgroup is taken off the ->sibling list, if we see it
3014 * unasserted, it's guaranteed that the next sibling hasn't
3015 * finished its grace period even if it's already removed, and thus
3016 * safe to dereference from this RCU critical section. If
3017 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
3018 * to be visible as %true here.
3020 if (likely(!cgroup_is_dead(pos
))) {
3021 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
, sibling
);
3022 if (&next
->sibling
!= &pos
->parent
->children
)
3028 * Can't dereference the next pointer. Each cgroup is given a
3029 * monotonically increasing unique serial number and always
3030 * appended to the sibling list, so the next one can be found by
3031 * walking the parent's children until we see a cgroup with higher
3032 * serial number than @pos's.
3034 * While this path can be slow, it's taken only when either the
3035 * current cgroup is removed or iteration and removal race.
3037 list_for_each_entry_rcu(next
, &pos
->parent
->children
, sibling
)
3038 if (next
->serial_nr
> pos
->serial_nr
)
3042 EXPORT_SYMBOL_GPL(cgroup_next_sibling
);
3045 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
3046 * @pos: the current position (%NULL to initiate traversal)
3047 * @cgroup: cgroup whose descendants to walk
3049 * To be used by cgroup_for_each_descendant_pre(). Find the next
3050 * descendant to visit for pre-order traversal of @cgroup's descendants.
3052 * While this function requires RCU read locking, it doesn't require the
3053 * whole traversal to be contained in a single RCU critical section. This
3054 * function will return the correct next descendant as long as both @pos
3055 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3057 struct cgroup
*cgroup_next_descendant_pre(struct cgroup
*pos
,
3058 struct cgroup
*cgroup
)
3060 struct cgroup
*next
;
3062 WARN_ON_ONCE(!rcu_read_lock_held());
3064 /* if first iteration, pretend we just visited @cgroup */
3068 /* visit the first child if exists */
3069 next
= list_first_or_null_rcu(&pos
->children
, struct cgroup
, sibling
);
3073 /* no child, visit my or the closest ancestor's next sibling */
3074 while (pos
!= cgroup
) {
3075 next
= cgroup_next_sibling(pos
);
3083 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre
);
3086 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
3087 * @pos: cgroup of interest
3089 * Return the rightmost descendant of @pos. If there's no descendant,
3090 * @pos is returned. This can be used during pre-order traversal to skip
3093 * While this function requires RCU read locking, it doesn't require the
3094 * whole traversal to be contained in a single RCU critical section. This
3095 * function will return the correct rightmost descendant as long as @pos is
3098 struct cgroup
*cgroup_rightmost_descendant(struct cgroup
*pos
)
3100 struct cgroup
*last
, *tmp
;
3102 WARN_ON_ONCE(!rcu_read_lock_held());
3106 /* ->prev isn't RCU safe, walk ->next till the end */
3108 list_for_each_entry_rcu(tmp
, &last
->children
, sibling
)
3114 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant
);
3116 static struct cgroup
*cgroup_leftmost_descendant(struct cgroup
*pos
)
3118 struct cgroup
*last
;
3122 pos
= list_first_or_null_rcu(&pos
->children
, struct cgroup
,
3130 * cgroup_next_descendant_post - find the next descendant for post-order walk
3131 * @pos: the current position (%NULL to initiate traversal)
3132 * @cgroup: cgroup whose descendants to walk
3134 * To be used by cgroup_for_each_descendant_post(). Find the next
3135 * descendant to visit for post-order traversal of @cgroup's descendants.
3137 * While this function requires RCU read locking, it doesn't require the
3138 * whole traversal to be contained in a single RCU critical section. This
3139 * function will return the correct next descendant as long as both @pos
3140 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3142 struct cgroup
*cgroup_next_descendant_post(struct cgroup
*pos
,
3143 struct cgroup
*cgroup
)
3145 struct cgroup
*next
;
3147 WARN_ON_ONCE(!rcu_read_lock_held());
3149 /* if first iteration, visit the leftmost descendant */
3151 next
= cgroup_leftmost_descendant(cgroup
);
3152 return next
!= cgroup
? next
: NULL
;
3155 /* if there's an unvisited sibling, visit its leftmost descendant */
3156 next
= cgroup_next_sibling(pos
);
3158 return cgroup_leftmost_descendant(next
);
3160 /* no sibling left, visit parent */
3162 return next
!= cgroup
? next
: NULL
;
3164 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post
);
3166 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3167 __acquires(css_set_lock
)
3170 * The first time anyone tries to iterate across a cgroup,
3171 * we need to enable the list linking each css_set to its
3172 * tasks, and fix up all existing tasks.
3174 if (!use_task_css_set_links
)
3175 cgroup_enable_task_cg_lists();
3177 read_lock(&css_set_lock
);
3178 it
->cset_link
= &cgrp
->cset_links
;
3179 cgroup_advance_iter(cgrp
, it
);
3182 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
3183 struct cgroup_iter
*it
)
3185 struct task_struct
*res
;
3186 struct list_head
*l
= it
->task
;
3187 struct cgrp_cset_link
*link
;
3189 /* If the iterator cg is NULL, we have no tasks */
3192 res
= list_entry(l
, struct task_struct
, cg_list
);
3193 /* Advance iterator to find next entry */
3195 link
= list_entry(it
->cset_link
, struct cgrp_cset_link
, cset_link
);
3196 if (l
== &link
->cset
->tasks
) {
3197 /* We reached the end of this task list - move on to
3198 * the next cg_cgroup_link */
3199 cgroup_advance_iter(cgrp
, it
);
3206 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3207 __releases(css_set_lock
)
3209 read_unlock(&css_set_lock
);
3212 static inline int started_after_time(struct task_struct
*t1
,
3213 struct timespec
*time
,
3214 struct task_struct
*t2
)
3216 int start_diff
= timespec_compare(&t1
->start_time
, time
);
3217 if (start_diff
> 0) {
3219 } else if (start_diff
< 0) {
3223 * Arbitrarily, if two processes started at the same
3224 * time, we'll say that the lower pointer value
3225 * started first. Note that t2 may have exited by now
3226 * so this may not be a valid pointer any longer, but
3227 * that's fine - it still serves to distinguish
3228 * between two tasks started (effectively) simultaneously.
3235 * This function is a callback from heap_insert() and is used to order
3237 * In this case we order the heap in descending task start time.
3239 static inline int started_after(void *p1
, void *p2
)
3241 struct task_struct
*t1
= p1
;
3242 struct task_struct
*t2
= p2
;
3243 return started_after_time(t1
, &t2
->start_time
, t2
);
3247 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3248 * @scan: struct cgroup_scanner containing arguments for the scan
3250 * Arguments include pointers to callback functions test_task() and
3252 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3253 * and if it returns true, call process_task() for it also.
3254 * The test_task pointer may be NULL, meaning always true (select all tasks).
3255 * Effectively duplicates cgroup_iter_{start,next,end}()
3256 * but does not lock css_set_lock for the call to process_task().
3257 * The struct cgroup_scanner may be embedded in any structure of the caller's
3259 * It is guaranteed that process_task() will act on every task that
3260 * is a member of the cgroup for the duration of this call. This
3261 * function may or may not call process_task() for tasks that exit
3262 * or move to a different cgroup during the call, or are forked or
3263 * move into the cgroup during the call.
3265 * Note that test_task() may be called with locks held, and may in some
3266 * situations be called multiple times for the same task, so it should
3268 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3269 * pre-allocated and will be used for heap operations (and its "gt" member will
3270 * be overwritten), else a temporary heap will be used (allocation of which
3271 * may cause this function to fail).
3273 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
3276 struct cgroup_iter it
;
3277 struct task_struct
*p
, *dropped
;
3278 /* Never dereference latest_task, since it's not refcounted */
3279 struct task_struct
*latest_task
= NULL
;
3280 struct ptr_heap tmp_heap
;
3281 struct ptr_heap
*heap
;
3282 struct timespec latest_time
= { 0, 0 };
3285 /* The caller supplied our heap and pre-allocated its memory */
3287 heap
->gt
= &started_after
;
3289 /* We need to allocate our own heap memory */
3291 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3293 /* cannot allocate the heap */
3299 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3300 * to determine which are of interest, and using the scanner's
3301 * "process_task" callback to process any of them that need an update.
3302 * Since we don't want to hold any locks during the task updates,
3303 * gather tasks to be processed in a heap structure.
3304 * The heap is sorted by descending task start time.
3305 * If the statically-sized heap fills up, we overflow tasks that
3306 * started later, and in future iterations only consider tasks that
3307 * started after the latest task in the previous pass. This
3308 * guarantees forward progress and that we don't miss any tasks.
3311 cgroup_iter_start(scan
->cgrp
, &it
);
3312 while ((p
= cgroup_iter_next(scan
->cgrp
, &it
))) {
3314 * Only affect tasks that qualify per the caller's callback,
3315 * if he provided one
3317 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
3320 * Only process tasks that started after the last task
3323 if (!started_after_time(p
, &latest_time
, latest_task
))
3325 dropped
= heap_insert(heap
, p
);
3326 if (dropped
== NULL
) {
3328 * The new task was inserted; the heap wasn't
3332 } else if (dropped
!= p
) {
3334 * The new task was inserted, and pushed out a
3338 put_task_struct(dropped
);
3341 * Else the new task was newer than anything already in
3342 * the heap and wasn't inserted
3345 cgroup_iter_end(scan
->cgrp
, &it
);
3348 for (i
= 0; i
< heap
->size
; i
++) {
3349 struct task_struct
*q
= heap
->ptrs
[i
];
3351 latest_time
= q
->start_time
;
3354 /* Process the task per the caller's callback */
3355 scan
->process_task(q
, scan
);
3359 * If we had to process any tasks at all, scan again
3360 * in case some of them were in the middle of forking
3361 * children that didn't get processed.
3362 * Not the most efficient way to do it, but it avoids
3363 * having to take callback_mutex in the fork path
3367 if (heap
== &tmp_heap
)
3368 heap_free(&tmp_heap
);
3372 static void cgroup_transfer_one_task(struct task_struct
*task
,
3373 struct cgroup_scanner
*scan
)
3375 struct cgroup
*new_cgroup
= scan
->data
;
3377 mutex_lock(&cgroup_mutex
);
3378 cgroup_attach_task(new_cgroup
, task
, false);
3379 mutex_unlock(&cgroup_mutex
);
3383 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3384 * @to: cgroup to which the tasks will be moved
3385 * @from: cgroup in which the tasks currently reside
3387 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3389 struct cgroup_scanner scan
;
3392 scan
.test_task
= NULL
; /* select all tasks in cgroup */
3393 scan
.process_task
= cgroup_transfer_one_task
;
3397 return cgroup_scan_tasks(&scan
);
3401 * Stuff for reading the 'tasks'/'procs' files.
3403 * Reading this file can return large amounts of data if a cgroup has
3404 * *lots* of attached tasks. So it may need several calls to read(),
3405 * but we cannot guarantee that the information we produce is correct
3406 * unless we produce it entirely atomically.
3410 /* which pidlist file are we talking about? */
3411 enum cgroup_filetype
{
3417 * A pidlist is a list of pids that virtually represents the contents of one
3418 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3419 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3422 struct cgroup_pidlist
{
3424 * used to find which pidlist is wanted. doesn't change as long as
3425 * this particular list stays in the list.
3427 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3430 /* how many elements the above list has */
3432 /* how many files are using the current array */
3434 /* each of these stored in a list by its cgroup */
3435 struct list_head links
;
3436 /* pointer to the cgroup we belong to, for list removal purposes */
3437 struct cgroup
*owner
;
3438 /* protects the other fields */
3439 struct rw_semaphore mutex
;
3443 * The following two functions "fix" the issue where there are more pids
3444 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3445 * TODO: replace with a kernel-wide solution to this problem
3447 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3448 static void *pidlist_allocate(int count
)
3450 if (PIDLIST_TOO_LARGE(count
))
3451 return vmalloc(count
* sizeof(pid_t
));
3453 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3455 static void pidlist_free(void *p
)
3457 if (is_vmalloc_addr(p
))
3464 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3465 * Returns the number of unique elements.
3467 static int pidlist_uniq(pid_t
*list
, int length
)
3472 * we presume the 0th element is unique, so i starts at 1. trivial
3473 * edge cases first; no work needs to be done for either
3475 if (length
== 0 || length
== 1)
3477 /* src and dest walk down the list; dest counts unique elements */
3478 for (src
= 1; src
< length
; src
++) {
3479 /* find next unique element */
3480 while (list
[src
] == list
[src
-1]) {
3485 /* dest always points to where the next unique element goes */
3486 list
[dest
] = list
[src
];
3493 static int cmppid(const void *a
, const void *b
)
3495 return *(pid_t
*)a
- *(pid_t
*)b
;
3499 * find the appropriate pidlist for our purpose (given procs vs tasks)
3500 * returns with the lock on that pidlist already held, and takes care
3501 * of the use count, or returns NULL with no locks held if we're out of
3504 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3505 enum cgroup_filetype type
)
3507 struct cgroup_pidlist
*l
;
3508 /* don't need task_nsproxy() if we're looking at ourself */
3509 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3512 * We can't drop the pidlist_mutex before taking the l->mutex in case
3513 * the last ref-holder is trying to remove l from the list at the same
3514 * time. Holding the pidlist_mutex precludes somebody taking whichever
3515 * list we find out from under us - compare release_pid_array().
3517 mutex_lock(&cgrp
->pidlist_mutex
);
3518 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3519 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3520 /* make sure l doesn't vanish out from under us */
3521 down_write(&l
->mutex
);
3522 mutex_unlock(&cgrp
->pidlist_mutex
);
3526 /* entry not found; create a new one */
3527 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3529 mutex_unlock(&cgrp
->pidlist_mutex
);
3532 init_rwsem(&l
->mutex
);
3533 down_write(&l
->mutex
);
3535 l
->key
.ns
= get_pid_ns(ns
);
3537 list_add(&l
->links
, &cgrp
->pidlists
);
3538 mutex_unlock(&cgrp
->pidlist_mutex
);
3543 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3545 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3546 struct cgroup_pidlist
**lp
)
3550 int pid
, n
= 0; /* used for populating the array */
3551 struct cgroup_iter it
;
3552 struct task_struct
*tsk
;
3553 struct cgroup_pidlist
*l
;
3556 * If cgroup gets more users after we read count, we won't have
3557 * enough space - tough. This race is indistinguishable to the
3558 * caller from the case that the additional cgroup users didn't
3559 * show up until sometime later on.
3561 length
= cgroup_task_count(cgrp
);
3562 array
= pidlist_allocate(length
);
3565 /* now, populate the array */
3566 cgroup_iter_start(cgrp
, &it
);
3567 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3568 if (unlikely(n
== length
))
3570 /* get tgid or pid for procs or tasks file respectively */
3571 if (type
== CGROUP_FILE_PROCS
)
3572 pid
= task_tgid_vnr(tsk
);
3574 pid
= task_pid_vnr(tsk
);
3575 if (pid
> 0) /* make sure to only use valid results */
3578 cgroup_iter_end(cgrp
, &it
);
3580 /* now sort & (if procs) strip out duplicates */
3581 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3582 if (type
== CGROUP_FILE_PROCS
)
3583 length
= pidlist_uniq(array
, length
);
3584 l
= cgroup_pidlist_find(cgrp
, type
);
3586 pidlist_free(array
);
3589 /* store array, freeing old if necessary - lock already held */
3590 pidlist_free(l
->list
);
3594 up_write(&l
->mutex
);
3600 * cgroupstats_build - build and fill cgroupstats
3601 * @stats: cgroupstats to fill information into
3602 * @dentry: A dentry entry belonging to the cgroup for which stats have
3605 * Build and fill cgroupstats so that taskstats can export it to user
3608 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3611 struct cgroup
*cgrp
;
3612 struct cgroup_iter it
;
3613 struct task_struct
*tsk
;
3616 * Validate dentry by checking the superblock operations,
3617 * and make sure it's a directory.
3619 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3620 !S_ISDIR(dentry
->d_inode
->i_mode
))
3624 cgrp
= dentry
->d_fsdata
;
3626 cgroup_iter_start(cgrp
, &it
);
3627 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3628 switch (tsk
->state
) {
3630 stats
->nr_running
++;
3632 case TASK_INTERRUPTIBLE
:
3633 stats
->nr_sleeping
++;
3635 case TASK_UNINTERRUPTIBLE
:
3636 stats
->nr_uninterruptible
++;
3639 stats
->nr_stopped
++;
3642 if (delayacct_is_task_waiting_on_io(tsk
))
3643 stats
->nr_io_wait
++;
3647 cgroup_iter_end(cgrp
, &it
);
3655 * seq_file methods for the tasks/procs files. The seq_file position is the
3656 * next pid to display; the seq_file iterator is a pointer to the pid
3657 * in the cgroup->l->list array.
3660 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3663 * Initially we receive a position value that corresponds to
3664 * one more than the last pid shown (or 0 on the first call or
3665 * after a seek to the start). Use a binary-search to find the
3666 * next pid to display, if any
3668 struct cgroup_pidlist
*l
= s
->private;
3669 int index
= 0, pid
= *pos
;
3672 down_read(&l
->mutex
);
3674 int end
= l
->length
;
3676 while (index
< end
) {
3677 int mid
= (index
+ end
) / 2;
3678 if (l
->list
[mid
] == pid
) {
3681 } else if (l
->list
[mid
] <= pid
)
3687 /* If we're off the end of the array, we're done */
3688 if (index
>= l
->length
)
3690 /* Update the abstract position to be the actual pid that we found */
3691 iter
= l
->list
+ index
;
3696 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3698 struct cgroup_pidlist
*l
= s
->private;
3702 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3704 struct cgroup_pidlist
*l
= s
->private;
3706 pid_t
*end
= l
->list
+ l
->length
;
3708 * Advance to the next pid in the array. If this goes off the
3720 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3722 return seq_printf(s
, "%d\n", *(int *)v
);
3726 * seq_operations functions for iterating on pidlists through seq_file -
3727 * independent of whether it's tasks or procs
3729 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3730 .start
= cgroup_pidlist_start
,
3731 .stop
= cgroup_pidlist_stop
,
3732 .next
= cgroup_pidlist_next
,
3733 .show
= cgroup_pidlist_show
,
3736 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3739 * the case where we're the last user of this particular pidlist will
3740 * have us remove it from the cgroup's list, which entails taking the
3741 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3742 * pidlist_mutex, we have to take pidlist_mutex first.
3744 mutex_lock(&l
->owner
->pidlist_mutex
);
3745 down_write(&l
->mutex
);
3746 BUG_ON(!l
->use_count
);
3747 if (!--l
->use_count
) {
3748 /* we're the last user if refcount is 0; remove and free */
3749 list_del(&l
->links
);
3750 mutex_unlock(&l
->owner
->pidlist_mutex
);
3751 pidlist_free(l
->list
);
3752 put_pid_ns(l
->key
.ns
);
3753 up_write(&l
->mutex
);
3757 mutex_unlock(&l
->owner
->pidlist_mutex
);
3758 up_write(&l
->mutex
);
3761 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3763 struct cgroup_pidlist
*l
;
3764 if (!(file
->f_mode
& FMODE_READ
))
3767 * the seq_file will only be initialized if the file was opened for
3768 * reading; hence we check if it's not null only in that case.
3770 l
= ((struct seq_file
*)file
->private_data
)->private;
3771 cgroup_release_pid_array(l
);
3772 return seq_release(inode
, file
);
3775 static const struct file_operations cgroup_pidlist_operations
= {
3777 .llseek
= seq_lseek
,
3778 .write
= cgroup_file_write
,
3779 .release
= cgroup_pidlist_release
,
3783 * The following functions handle opens on a file that displays a pidlist
3784 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3787 /* helper function for the two below it */
3788 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3790 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3791 struct cgroup_pidlist
*l
;
3794 /* Nothing to do for write-only files */
3795 if (!(file
->f_mode
& FMODE_READ
))
3798 /* have the array populated */
3799 retval
= pidlist_array_load(cgrp
, type
, &l
);
3802 /* configure file information */
3803 file
->f_op
= &cgroup_pidlist_operations
;
3805 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3807 cgroup_release_pid_array(l
);
3810 ((struct seq_file
*)file
->private_data
)->private = l
;
3813 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3815 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3817 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3819 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3822 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3825 return notify_on_release(cgrp
);
3828 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3832 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3834 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3836 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3841 * When dput() is called asynchronously, if umount has been done and
3842 * then deactivate_super() in cgroup_free_fn() kills the superblock,
3843 * there's a small window that vfs will see the root dentry with non-zero
3844 * refcnt and trigger BUG().
3846 * That's why we hold a reference before dput() and drop it right after.
3848 static void cgroup_dput(struct cgroup
*cgrp
)
3850 struct super_block
*sb
= cgrp
->root
->sb
;
3852 atomic_inc(&sb
->s_active
);
3854 deactivate_super(sb
);
3858 * Unregister event and free resources.
3860 * Gets called from workqueue.
3862 static void cgroup_event_remove(struct work_struct
*work
)
3864 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3866 struct cgroup
*cgrp
= event
->cgrp
;
3868 remove_wait_queue(event
->wqh
, &event
->wait
);
3870 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3872 /* Notify userspace the event is going away. */
3873 eventfd_signal(event
->eventfd
, 1);
3875 eventfd_ctx_put(event
->eventfd
);
3881 * Gets called on POLLHUP on eventfd when user closes it.
3883 * Called with wqh->lock held and interrupts disabled.
3885 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3886 int sync
, void *key
)
3888 struct cgroup_event
*event
= container_of(wait
,
3889 struct cgroup_event
, wait
);
3890 struct cgroup
*cgrp
= event
->cgrp
;
3891 unsigned long flags
= (unsigned long)key
;
3893 if (flags
& POLLHUP
) {
3895 * If the event has been detached at cgroup removal, we
3896 * can simply return knowing the other side will cleanup
3899 * We can't race against event freeing since the other
3900 * side will require wqh->lock via remove_wait_queue(),
3903 spin_lock(&cgrp
->event_list_lock
);
3904 if (!list_empty(&event
->list
)) {
3905 list_del_init(&event
->list
);
3907 * We are in atomic context, but cgroup_event_remove()
3908 * may sleep, so we have to call it in workqueue.
3910 schedule_work(&event
->remove
);
3912 spin_unlock(&cgrp
->event_list_lock
);
3918 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3919 wait_queue_head_t
*wqh
, poll_table
*pt
)
3921 struct cgroup_event
*event
= container_of(pt
,
3922 struct cgroup_event
, pt
);
3925 add_wait_queue(wqh
, &event
->wait
);
3929 * Parse input and register new cgroup event handler.
3931 * Input must be in format '<event_fd> <control_fd> <args>'.
3932 * Interpretation of args is defined by control file implementation.
3934 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3937 struct cgroup_event
*event
;
3938 struct cgroup
*cgrp_cfile
;
3939 unsigned int efd
, cfd
;
3945 efd
= simple_strtoul(buffer
, &endp
, 10);
3950 cfd
= simple_strtoul(buffer
, &endp
, 10);
3951 if ((*endp
!= ' ') && (*endp
!= '\0'))
3955 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3959 INIT_LIST_HEAD(&event
->list
);
3960 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3961 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3962 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3964 efile
= eventfd_fget(efd
);
3965 if (IS_ERR(efile
)) {
3966 ret
= PTR_ERR(efile
);
3970 event
->eventfd
= eventfd_ctx_fileget(efile
);
3971 if (IS_ERR(event
->eventfd
)) {
3972 ret
= PTR_ERR(event
->eventfd
);
3979 goto out_put_eventfd
;
3982 /* the process need read permission on control file */
3983 /* AV: shouldn't we check that it's been opened for read instead? */
3984 ret
= inode_permission(file_inode(cfile
), MAY_READ
);
3988 event
->cft
= __file_cft(cfile
);
3989 if (IS_ERR(event
->cft
)) {
3990 ret
= PTR_ERR(event
->cft
);
3995 * The file to be monitored must be in the same cgroup as
3996 * cgroup.event_control is.
3998 cgrp_cfile
= __d_cgrp(cfile
->f_dentry
->d_parent
);
3999 if (cgrp_cfile
!= cgrp
) {
4004 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
4009 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
4010 event
->eventfd
, buffer
);
4014 efile
->f_op
->poll(efile
, &event
->pt
);
4017 * Events should be removed after rmdir of cgroup directory, but before
4018 * destroying subsystem state objects. Let's take reference to cgroup
4019 * directory dentry to do that.
4023 spin_lock(&cgrp
->event_list_lock
);
4024 list_add(&event
->list
, &cgrp
->event_list
);
4025 spin_unlock(&cgrp
->event_list_lock
);
4035 eventfd_ctx_put(event
->eventfd
);
4044 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
4047 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4050 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
4055 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4057 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4061 static struct cftype cgroup_base_files
[] = {
4063 .name
= "cgroup.procs",
4064 .open
= cgroup_procs_open
,
4065 .write_u64
= cgroup_procs_write
,
4066 .release
= cgroup_pidlist_release
,
4067 .mode
= S_IRUGO
| S_IWUSR
,
4070 .name
= "cgroup.event_control",
4071 .write_string
= cgroup_write_event_control
,
4075 .name
= "cgroup.clone_children",
4076 .flags
= CFTYPE_INSANE
,
4077 .read_u64
= cgroup_clone_children_read
,
4078 .write_u64
= cgroup_clone_children_write
,
4081 .name
= "cgroup.sane_behavior",
4082 .flags
= CFTYPE_ONLY_ON_ROOT
,
4083 .read_seq_string
= cgroup_sane_behavior_show
,
4087 * Historical crazy stuff. These don't have "cgroup." prefix and
4088 * don't exist if sane_behavior. If you're depending on these, be
4089 * prepared to be burned.
4093 .flags
= CFTYPE_INSANE
, /* use "procs" instead */
4094 .open
= cgroup_tasks_open
,
4095 .write_u64
= cgroup_tasks_write
,
4096 .release
= cgroup_pidlist_release
,
4097 .mode
= S_IRUGO
| S_IWUSR
,
4100 .name
= "notify_on_release",
4101 .flags
= CFTYPE_INSANE
,
4102 .read_u64
= cgroup_read_notify_on_release
,
4103 .write_u64
= cgroup_write_notify_on_release
,
4106 .name
= "release_agent",
4107 .flags
= CFTYPE_INSANE
| CFTYPE_ONLY_ON_ROOT
,
4108 .read_seq_string
= cgroup_release_agent_show
,
4109 .write_string
= cgroup_release_agent_write
,
4110 .max_write_len
= PATH_MAX
,
4116 * cgroup_populate_dir - create subsys files in a cgroup directory
4117 * @cgrp: target cgroup
4118 * @subsys_mask: mask of the subsystem ids whose files should be added
4120 * On failure, no file is added.
4122 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
4124 struct cgroup_subsys
*ss
;
4127 /* process cftsets of each subsystem */
4128 for_each_subsys(ss
, i
) {
4129 struct cftype_set
*set
;
4131 if (!test_bit(i
, &subsys_mask
))
4134 list_for_each_entry(set
, &ss
->cftsets
, node
) {
4135 ret
= cgroup_addrm_files(cgrp
, ss
, set
->cfts
, true);
4141 /* This cgroup is ready now */
4142 for_each_root_subsys(cgrp
->root
, ss
) {
4143 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4144 struct css_id
*id
= rcu_dereference_protected(css
->id
, true);
4147 * Update id->css pointer and make this css visible from
4148 * CSS ID functions. This pointer will be dereferened
4149 * from RCU-read-side without locks.
4152 rcu_assign_pointer(id
->css
, css
);
4157 cgroup_clear_dir(cgrp
, subsys_mask
);
4161 static void css_dput_fn(struct work_struct
*work
)
4163 struct cgroup_subsys_state
*css
=
4164 container_of(work
, struct cgroup_subsys_state
, dput_work
);
4166 cgroup_dput(css
->cgroup
);
4169 static void css_release(struct percpu_ref
*ref
)
4171 struct cgroup_subsys_state
*css
=
4172 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4174 schedule_work(&css
->dput_work
);
4177 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
4178 struct cgroup_subsys
*ss
,
4179 struct cgroup
*cgrp
)
4184 if (cgrp
== cgroup_dummy_top
)
4185 css
->flags
|= CSS_ROOT
;
4186 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
4187 cgrp
->subsys
[ss
->subsys_id
] = css
;
4190 * css holds an extra ref to @cgrp->dentry which is put on the last
4191 * css_put(). dput() requires process context, which css_put() may
4192 * be called without. @css->dput_work will be used to invoke
4193 * dput() asynchronously from css_put().
4195 INIT_WORK(&css
->dput_work
, css_dput_fn
);
4198 /* invoke ->css_online() on a new CSS and mark it online if successful */
4199 static int online_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4203 lockdep_assert_held(&cgroup_mutex
);
4206 ret
= ss
->css_online(cgrp
);
4208 cgrp
->subsys
[ss
->subsys_id
]->flags
|= CSS_ONLINE
;
4212 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4213 static void offline_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4215 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4217 lockdep_assert_held(&cgroup_mutex
);
4219 if (!(css
->flags
& CSS_ONLINE
))
4222 if (ss
->css_offline
)
4223 ss
->css_offline(cgrp
);
4225 cgrp
->subsys
[ss
->subsys_id
]->flags
&= ~CSS_ONLINE
;
4229 * cgroup_create - create a cgroup
4230 * @parent: cgroup that will be parent of the new cgroup
4231 * @dentry: dentry of the new cgroup
4232 * @mode: mode to set on new inode
4234 * Must be called with the mutex on the parent inode held
4236 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
4239 struct cgroup
*cgrp
;
4240 struct cgroup_name
*name
;
4241 struct cgroupfs_root
*root
= parent
->root
;
4243 struct cgroup_subsys
*ss
;
4244 struct super_block
*sb
= root
->sb
;
4246 /* allocate the cgroup and its ID, 0 is reserved for the root */
4247 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4251 name
= cgroup_alloc_name(dentry
);
4254 rcu_assign_pointer(cgrp
->name
, name
);
4257 * Temporarily set the pointer to NULL, so idr_find() won't return
4258 * a half-baked cgroup.
4260 cgrp
->id
= idr_alloc(&root
->cgroup_idr
, NULL
, 1, 0, GFP_KERNEL
);
4265 * Only live parents can have children. Note that the liveliness
4266 * check isn't strictly necessary because cgroup_mkdir() and
4267 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4268 * anyway so that locking is contained inside cgroup proper and we
4269 * don't get nasty surprises if we ever grow another caller.
4271 if (!cgroup_lock_live_group(parent
)) {
4276 /* Grab a reference on the superblock so the hierarchy doesn't
4277 * get deleted on unmount if there are child cgroups. This
4278 * can be done outside cgroup_mutex, since the sb can't
4279 * disappear while someone has an open control file on the
4281 atomic_inc(&sb
->s_active
);
4283 init_cgroup_housekeeping(cgrp
);
4285 dentry
->d_fsdata
= cgrp
;
4286 cgrp
->dentry
= dentry
;
4288 cgrp
->parent
= parent
;
4289 cgrp
->root
= parent
->root
;
4291 if (notify_on_release(parent
))
4292 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4294 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4295 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4297 for_each_root_subsys(root
, ss
) {
4298 struct cgroup_subsys_state
*css
;
4300 css
= ss
->css_alloc(cgrp
);
4306 err
= percpu_ref_init(&css
->refcnt
, css_release
);
4310 init_cgroup_css(css
, ss
, cgrp
);
4313 err
= alloc_css_id(ss
, parent
, cgrp
);
4320 * Create directory. cgroup_create_file() returns with the new
4321 * directory locked on success so that it can be populated without
4322 * dropping cgroup_mutex.
4324 err
= cgroup_create_file(dentry
, S_IFDIR
| mode
, sb
);
4327 lockdep_assert_held(&dentry
->d_inode
->i_mutex
);
4329 cgrp
->serial_nr
= cgroup_serial_nr_next
++;
4331 /* allocation complete, commit to creation */
4332 list_add_tail_rcu(&cgrp
->sibling
, &cgrp
->parent
->children
);
4333 root
->number_of_cgroups
++;
4335 /* each css holds a ref to the cgroup's dentry */
4336 for_each_root_subsys(root
, ss
)
4339 /* hold a ref to the parent's dentry */
4340 dget(parent
->dentry
);
4342 /* creation succeeded, notify subsystems */
4343 for_each_root_subsys(root
, ss
) {
4344 err
= online_css(ss
, cgrp
);
4348 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4350 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",
4351 current
->comm
, current
->pid
, ss
->name
);
4352 if (!strcmp(ss
->name
, "memory"))
4353 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4354 ss
->warned_broken_hierarchy
= true;
4358 idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4360 err
= cgroup_addrm_files(cgrp
, NULL
, cgroup_base_files
, true);
4364 err
= cgroup_populate_dir(cgrp
, root
->subsys_mask
);
4368 mutex_unlock(&cgroup_mutex
);
4369 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4374 for_each_root_subsys(root
, ss
) {
4375 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4378 percpu_ref_cancel_init(&css
->refcnt
);
4382 mutex_unlock(&cgroup_mutex
);
4383 /* Release the reference count that we took on the superblock */
4384 deactivate_super(sb
);
4386 idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4388 kfree(rcu_dereference_raw(cgrp
->name
));
4394 cgroup_destroy_locked(cgrp
);
4395 mutex_unlock(&cgroup_mutex
);
4396 mutex_unlock(&dentry
->d_inode
->i_mutex
);
4400 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4402 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4404 /* the vfs holds inode->i_mutex already */
4405 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4408 static void cgroup_css_killed(struct cgroup
*cgrp
)
4410 if (!atomic_dec_and_test(&cgrp
->css_kill_cnt
))
4413 /* percpu ref's of all css's are killed, kick off the next step */
4414 INIT_WORK(&cgrp
->destroy_work
, cgroup_offline_fn
);
4415 schedule_work(&cgrp
->destroy_work
);
4418 static void css_ref_killed_fn(struct percpu_ref
*ref
)
4420 struct cgroup_subsys_state
*css
=
4421 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4423 cgroup_css_killed(css
->cgroup
);
4427 * cgroup_destroy_locked - the first stage of cgroup destruction
4428 * @cgrp: cgroup to be destroyed
4430 * css's make use of percpu refcnts whose killing latency shouldn't be
4431 * exposed to userland and are RCU protected. Also, cgroup core needs to
4432 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
4433 * invoked. To satisfy all the requirements, destruction is implemented in
4434 * the following two steps.
4436 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4437 * userland visible parts and start killing the percpu refcnts of
4438 * css's. Set up so that the next stage will be kicked off once all
4439 * the percpu refcnts are confirmed to be killed.
4441 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4442 * rest of destruction. Once all cgroup references are gone, the
4443 * cgroup is RCU-freed.
4445 * This function implements s1. After this step, @cgrp is gone as far as
4446 * the userland is concerned and a new cgroup with the same name may be
4447 * created. As cgroup doesn't care about the names internally, this
4448 * doesn't cause any problem.
4450 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4451 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4453 struct dentry
*d
= cgrp
->dentry
;
4454 struct cgroup_event
*event
, *tmp
;
4455 struct cgroup_subsys
*ss
;
4458 lockdep_assert_held(&d
->d_inode
->i_mutex
);
4459 lockdep_assert_held(&cgroup_mutex
);
4462 * css_set_lock synchronizes access to ->cset_links and prevents
4463 * @cgrp from being removed while __put_css_set() is in progress.
4465 read_lock(&css_set_lock
);
4466 empty
= list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
);
4467 read_unlock(&css_set_lock
);
4472 * Block new css_tryget() by killing css refcnts. cgroup core
4473 * guarantees that, by the time ->css_offline() is invoked, no new
4474 * css reference will be given out via css_tryget(). We can't
4475 * simply call percpu_ref_kill() and proceed to offlining css's
4476 * because percpu_ref_kill() doesn't guarantee that the ref is seen
4477 * as killed on all CPUs on return.
4479 * Use percpu_ref_kill_and_confirm() to get notifications as each
4480 * css is confirmed to be seen as killed on all CPUs. The
4481 * notification callback keeps track of the number of css's to be
4482 * killed and schedules cgroup_offline_fn() to perform the rest of
4483 * destruction once the percpu refs of all css's are confirmed to
4486 atomic_set(&cgrp
->css_kill_cnt
, 1);
4487 for_each_root_subsys(cgrp
->root
, ss
) {
4488 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4491 * Killing would put the base ref, but we need to keep it
4492 * alive until after ->css_offline.
4494 percpu_ref_get(&css
->refcnt
);
4496 atomic_inc(&cgrp
->css_kill_cnt
);
4497 percpu_ref_kill_and_confirm(&css
->refcnt
, css_ref_killed_fn
);
4499 cgroup_css_killed(cgrp
);
4502 * Mark @cgrp dead. This prevents further task migration and child
4503 * creation by disabling cgroup_lock_live_group(). Note that
4504 * CGRP_DEAD assertion is depended upon by cgroup_next_sibling() to
4505 * resume iteration after dropping RCU read lock. See
4506 * cgroup_next_sibling() for details.
4508 set_bit(CGRP_DEAD
, &cgrp
->flags
);
4510 /* CGRP_DEAD is set, remove from ->release_list for the last time */
4511 raw_spin_lock(&release_list_lock
);
4512 if (!list_empty(&cgrp
->release_list
))
4513 list_del_init(&cgrp
->release_list
);
4514 raw_spin_unlock(&release_list_lock
);
4517 * Clear and remove @cgrp directory. The removal puts the base ref
4518 * but we aren't quite done with @cgrp yet, so hold onto it.
4520 cgroup_clear_dir(cgrp
, cgrp
->root
->subsys_mask
);
4521 cgroup_addrm_files(cgrp
, NULL
, cgroup_base_files
, false);
4523 cgroup_d_remove_dir(d
);
4526 * Unregister events and notify userspace.
4527 * Notify userspace about cgroup removing only after rmdir of cgroup
4528 * directory to avoid race between userspace and kernelspace.
4530 spin_lock(&cgrp
->event_list_lock
);
4531 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4532 list_del_init(&event
->list
);
4533 schedule_work(&event
->remove
);
4535 spin_unlock(&cgrp
->event_list_lock
);
4541 * cgroup_offline_fn - the second step of cgroup destruction
4542 * @work: cgroup->destroy_free_work
4544 * This function is invoked from a work item for a cgroup which is being
4545 * destroyed after the percpu refcnts of all css's are guaranteed to be
4546 * seen as killed on all CPUs, and performs the rest of destruction. This
4547 * is the second step of destruction described in the comment above
4548 * cgroup_destroy_locked().
4550 static void cgroup_offline_fn(struct work_struct
*work
)
4552 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, destroy_work
);
4553 struct cgroup
*parent
= cgrp
->parent
;
4554 struct dentry
*d
= cgrp
->dentry
;
4555 struct cgroup_subsys
*ss
;
4557 mutex_lock(&cgroup_mutex
);
4560 * css_tryget() is guaranteed to fail now. Tell subsystems to
4561 * initate destruction.
4563 for_each_root_subsys(cgrp
->root
, ss
)
4564 offline_css(ss
, cgrp
);
4567 * Put the css refs from cgroup_destroy_locked(). Each css holds
4568 * an extra reference to the cgroup's dentry and cgroup removal
4569 * proceeds regardless of css refs. On the last put of each css,
4570 * whenever that may be, the extra dentry ref is put so that dentry
4571 * destruction happens only after all css's are released.
4573 for_each_root_subsys(cgrp
->root
, ss
)
4574 css_put(cgrp
->subsys
[ss
->subsys_id
]);
4576 /* delete this cgroup from parent->children */
4577 list_del_rcu(&cgrp
->sibling
);
4580 * We should remove the cgroup object from idr before its grace
4581 * period starts, so we won't be looking up a cgroup while the
4582 * cgroup is being freed.
4584 idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4589 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4590 check_for_release(parent
);
4592 mutex_unlock(&cgroup_mutex
);
4595 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4599 mutex_lock(&cgroup_mutex
);
4600 ret
= cgroup_destroy_locked(dentry
->d_fsdata
);
4601 mutex_unlock(&cgroup_mutex
);
4606 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4608 INIT_LIST_HEAD(&ss
->cftsets
);
4611 * base_cftset is embedded in subsys itself, no need to worry about
4614 if (ss
->base_cftypes
) {
4615 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4616 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4620 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4622 struct cgroup_subsys_state
*css
;
4624 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4626 mutex_lock(&cgroup_mutex
);
4628 /* init base cftset */
4629 cgroup_init_cftsets(ss
);
4631 /* Create the top cgroup state for this subsystem */
4632 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4633 ss
->root
= &cgroup_dummy_root
;
4634 css
= ss
->css_alloc(cgroup_dummy_top
);
4635 /* We don't handle early failures gracefully */
4636 BUG_ON(IS_ERR(css
));
4637 init_cgroup_css(css
, ss
, cgroup_dummy_top
);
4639 /* Update the init_css_set to contain a subsys
4640 * pointer to this state - since the subsystem is
4641 * newly registered, all tasks and hence the
4642 * init_css_set is in the subsystem's top cgroup. */
4643 init_css_set
.subsys
[ss
->subsys_id
] = css
;
4645 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4647 /* At system boot, before all subsystems have been
4648 * registered, no tasks have been forked, so we don't
4649 * need to invoke fork callbacks here. */
4650 BUG_ON(!list_empty(&init_task
.tasks
));
4652 BUG_ON(online_css(ss
, cgroup_dummy_top
));
4654 mutex_unlock(&cgroup_mutex
);
4656 /* this function shouldn't be used with modular subsystems, since they
4657 * need to register a subsys_id, among other things */
4662 * cgroup_load_subsys: load and register a modular subsystem at runtime
4663 * @ss: the subsystem to load
4665 * This function should be called in a modular subsystem's initcall. If the
4666 * subsystem is built as a module, it will be assigned a new subsys_id and set
4667 * up for use. If the subsystem is built-in anyway, work is delegated to the
4668 * simpler cgroup_init_subsys.
4670 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4672 struct cgroup_subsys_state
*css
;
4674 struct hlist_node
*tmp
;
4675 struct css_set
*cset
;
4678 /* check name and function validity */
4679 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4680 ss
->css_alloc
== NULL
|| ss
->css_free
== NULL
)
4684 * we don't support callbacks in modular subsystems. this check is
4685 * before the ss->module check for consistency; a subsystem that could
4686 * be a module should still have no callbacks even if the user isn't
4687 * compiling it as one.
4689 if (ss
->fork
|| ss
->exit
)
4693 * an optionally modular subsystem is built-in: we want to do nothing,
4694 * since cgroup_init_subsys will have already taken care of it.
4696 if (ss
->module
== NULL
) {
4697 /* a sanity check */
4698 BUG_ON(cgroup_subsys
[ss
->subsys_id
] != ss
);
4702 /* init base cftset */
4703 cgroup_init_cftsets(ss
);
4705 mutex_lock(&cgroup_mutex
);
4706 cgroup_subsys
[ss
->subsys_id
] = ss
;
4709 * no ss->css_alloc seems to need anything important in the ss
4710 * struct, so this can happen first (i.e. before the dummy root
4713 css
= ss
->css_alloc(cgroup_dummy_top
);
4715 /* failure case - need to deassign the cgroup_subsys[] slot. */
4716 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4717 mutex_unlock(&cgroup_mutex
);
4718 return PTR_ERR(css
);
4721 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4722 ss
->root
= &cgroup_dummy_root
;
4724 /* our new subsystem will be attached to the dummy hierarchy. */
4725 init_cgroup_css(css
, ss
, cgroup_dummy_top
);
4726 /* init_idr must be after init_cgroup_css because it sets css->id. */
4728 ret
= cgroup_init_idr(ss
, css
);
4734 * Now we need to entangle the css into the existing css_sets. unlike
4735 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4736 * will need a new pointer to it; done by iterating the css_set_table.
4737 * furthermore, modifying the existing css_sets will corrupt the hash
4738 * table state, so each changed css_set will need its hash recomputed.
4739 * this is all done under the css_set_lock.
4741 write_lock(&css_set_lock
);
4742 hash_for_each_safe(css_set_table
, i
, tmp
, cset
, hlist
) {
4743 /* skip entries that we already rehashed */
4744 if (cset
->subsys
[ss
->subsys_id
])
4746 /* remove existing entry */
4747 hash_del(&cset
->hlist
);
4749 cset
->subsys
[ss
->subsys_id
] = css
;
4750 /* recompute hash and restore entry */
4751 key
= css_set_hash(cset
->subsys
);
4752 hash_add(css_set_table
, &cset
->hlist
, key
);
4754 write_unlock(&css_set_lock
);
4756 ret
= online_css(ss
, cgroup_dummy_top
);
4761 mutex_unlock(&cgroup_mutex
);
4765 mutex_unlock(&cgroup_mutex
);
4766 /* @ss can't be mounted here as try_module_get() would fail */
4767 cgroup_unload_subsys(ss
);
4770 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4773 * cgroup_unload_subsys: unload a modular subsystem
4774 * @ss: the subsystem to unload
4776 * This function should be called in a modular subsystem's exitcall. When this
4777 * function is invoked, the refcount on the subsystem's module will be 0, so
4778 * the subsystem will not be attached to any hierarchy.
4780 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4782 struct cgrp_cset_link
*link
;
4784 BUG_ON(ss
->module
== NULL
);
4787 * we shouldn't be called if the subsystem is in use, and the use of
4788 * try_module_get() in rebind_subsystems() should ensure that it
4789 * doesn't start being used while we're killing it off.
4791 BUG_ON(ss
->root
!= &cgroup_dummy_root
);
4793 mutex_lock(&cgroup_mutex
);
4795 offline_css(ss
, cgroup_dummy_top
);
4798 idr_destroy(&ss
->idr
);
4800 /* deassign the subsys_id */
4801 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4803 /* remove subsystem from the dummy root's list of subsystems */
4804 list_del_init(&ss
->sibling
);
4807 * disentangle the css from all css_sets attached to the dummy
4808 * top. as in loading, we need to pay our respects to the hashtable
4811 write_lock(&css_set_lock
);
4812 list_for_each_entry(link
, &cgroup_dummy_top
->cset_links
, cset_link
) {
4813 struct css_set
*cset
= link
->cset
;
4816 hash_del(&cset
->hlist
);
4817 cset
->subsys
[ss
->subsys_id
] = NULL
;
4818 key
= css_set_hash(cset
->subsys
);
4819 hash_add(css_set_table
, &cset
->hlist
, key
);
4821 write_unlock(&css_set_lock
);
4824 * remove subsystem's css from the cgroup_dummy_top and free it -
4825 * need to free before marking as null because ss->css_free needs
4826 * the cgrp->subsys pointer to find their state. note that this
4827 * also takes care of freeing the css_id.
4829 ss
->css_free(cgroup_dummy_top
);
4830 cgroup_dummy_top
->subsys
[ss
->subsys_id
] = NULL
;
4832 mutex_unlock(&cgroup_mutex
);
4834 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4837 * cgroup_init_early - cgroup initialization at system boot
4839 * Initialize cgroups at system boot, and initialize any
4840 * subsystems that request early init.
4842 int __init
cgroup_init_early(void)
4844 struct cgroup_subsys
*ss
;
4847 atomic_set(&init_css_set
.refcount
, 1);
4848 INIT_LIST_HEAD(&init_css_set
.cgrp_links
);
4849 INIT_LIST_HEAD(&init_css_set
.tasks
);
4850 INIT_HLIST_NODE(&init_css_set
.hlist
);
4852 init_cgroup_root(&cgroup_dummy_root
);
4853 cgroup_root_count
= 1;
4854 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
4856 init_cgrp_cset_link
.cset
= &init_css_set
;
4857 init_cgrp_cset_link
.cgrp
= cgroup_dummy_top
;
4858 list_add(&init_cgrp_cset_link
.cset_link
, &cgroup_dummy_top
->cset_links
);
4859 list_add(&init_cgrp_cset_link
.cgrp_link
, &init_css_set
.cgrp_links
);
4861 /* at bootup time, we don't worry about modular subsystems */
4862 for_each_builtin_subsys(ss
, i
) {
4864 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4865 BUG_ON(!ss
->css_alloc
);
4866 BUG_ON(!ss
->css_free
);
4867 if (ss
->subsys_id
!= i
) {
4868 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4869 ss
->name
, ss
->subsys_id
);
4874 cgroup_init_subsys(ss
);
4880 * cgroup_init - cgroup initialization
4882 * Register cgroup filesystem and /proc file, and initialize
4883 * any subsystems that didn't request early init.
4885 int __init
cgroup_init(void)
4887 struct cgroup_subsys
*ss
;
4891 err
= bdi_init(&cgroup_backing_dev_info
);
4895 for_each_builtin_subsys(ss
, i
) {
4896 if (!ss
->early_init
)
4897 cgroup_init_subsys(ss
);
4899 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4902 /* allocate id for the dummy hierarchy */
4903 mutex_lock(&cgroup_mutex
);
4904 mutex_lock(&cgroup_root_mutex
);
4906 /* Add init_css_set to the hash table */
4907 key
= css_set_hash(init_css_set
.subsys
);
4908 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4910 BUG_ON(cgroup_init_root_id(&cgroup_dummy_root
, 0, 1));
4912 err
= idr_alloc(&cgroup_dummy_root
.cgroup_idr
, cgroup_dummy_top
,
4916 mutex_unlock(&cgroup_root_mutex
);
4917 mutex_unlock(&cgroup_mutex
);
4919 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4925 err
= register_filesystem(&cgroup_fs_type
);
4927 kobject_put(cgroup_kobj
);
4931 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4935 bdi_destroy(&cgroup_backing_dev_info
);
4941 * proc_cgroup_show()
4942 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4943 * - Used for /proc/<pid>/cgroup.
4944 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4945 * doesn't really matter if tsk->cgroup changes after we read it,
4946 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4947 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4948 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4949 * cgroup to top_cgroup.
4952 /* TODO: Use a proper seq_file iterator */
4953 int proc_cgroup_show(struct seq_file
*m
, void *v
)
4956 struct task_struct
*tsk
;
4959 struct cgroupfs_root
*root
;
4962 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4968 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4974 mutex_lock(&cgroup_mutex
);
4976 for_each_active_root(root
) {
4977 struct cgroup_subsys
*ss
;
4978 struct cgroup
*cgrp
;
4981 seq_printf(m
, "%d:", root
->hierarchy_id
);
4982 for_each_root_subsys(root
, ss
)
4983 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4984 if (strlen(root
->name
))
4985 seq_printf(m
, "%sname=%s", count
? "," : "",
4988 cgrp
= task_cgroup_from_root(tsk
, root
);
4989 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4997 mutex_unlock(&cgroup_mutex
);
4998 put_task_struct(tsk
);
5005 /* Display information about each subsystem and each hierarchy */
5006 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5008 struct cgroup_subsys
*ss
;
5011 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5013 * ideally we don't want subsystems moving around while we do this.
5014 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5015 * subsys/hierarchy state.
5017 mutex_lock(&cgroup_mutex
);
5019 for_each_subsys(ss
, i
)
5020 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5021 ss
->name
, ss
->root
->hierarchy_id
,
5022 ss
->root
->number_of_cgroups
, !ss
->disabled
);
5024 mutex_unlock(&cgroup_mutex
);
5028 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5030 return single_open(file
, proc_cgroupstats_show
, NULL
);
5033 static const struct file_operations proc_cgroupstats_operations
= {
5034 .open
= cgroupstats_open
,
5036 .llseek
= seq_lseek
,
5037 .release
= single_release
,
5041 * cgroup_fork - attach newly forked task to its parents cgroup.
5042 * @child: pointer to task_struct of forking parent process.
5044 * Description: A task inherits its parent's cgroup at fork().
5046 * A pointer to the shared css_set was automatically copied in
5047 * fork.c by dup_task_struct(). However, we ignore that copy, since
5048 * it was not made under the protection of RCU or cgroup_mutex, so
5049 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
5050 * have already changed current->cgroups, allowing the previously
5051 * referenced cgroup group to be removed and freed.
5053 * At the point that cgroup_fork() is called, 'current' is the parent
5054 * task, and the passed argument 'child' points to the child task.
5056 void cgroup_fork(struct task_struct
*child
)
5059 get_css_set(task_css_set(current
));
5060 child
->cgroups
= current
->cgroups
;
5061 task_unlock(current
);
5062 INIT_LIST_HEAD(&child
->cg_list
);
5066 * cgroup_post_fork - called on a new task after adding it to the task list
5067 * @child: the task in question
5069 * Adds the task to the list running through its css_set if necessary and
5070 * call the subsystem fork() callbacks. Has to be after the task is
5071 * visible on the task list in case we race with the first call to
5072 * cgroup_iter_start() - to guarantee that the new task ends up on its
5075 void cgroup_post_fork(struct task_struct
*child
)
5077 struct cgroup_subsys
*ss
;
5081 * use_task_css_set_links is set to 1 before we walk the tasklist
5082 * under the tasklist_lock and we read it here after we added the child
5083 * to the tasklist under the tasklist_lock as well. If the child wasn't
5084 * yet in the tasklist when we walked through it from
5085 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
5086 * should be visible now due to the paired locking and barriers implied
5087 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
5088 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
5091 if (use_task_css_set_links
) {
5092 write_lock(&css_set_lock
);
5094 if (list_empty(&child
->cg_list
))
5095 list_add(&child
->cg_list
, &task_css_set(child
)->tasks
);
5097 write_unlock(&css_set_lock
);
5101 * Call ss->fork(). This must happen after @child is linked on
5102 * css_set; otherwise, @child might change state between ->fork()
5103 * and addition to css_set.
5105 if (need_forkexit_callback
) {
5107 * fork/exit callbacks are supported only for builtin
5108 * subsystems, and the builtin section of the subsys
5109 * array is immutable, so we don't need to lock the
5110 * subsys array here. On the other hand, modular section
5111 * of the array can be freed at module unload, so we
5114 for_each_builtin_subsys(ss
, i
)
5121 * cgroup_exit - detach cgroup from exiting task
5122 * @tsk: pointer to task_struct of exiting process
5123 * @run_callback: run exit callbacks?
5125 * Description: Detach cgroup from @tsk and release it.
5127 * Note that cgroups marked notify_on_release force every task in
5128 * them to take the global cgroup_mutex mutex when exiting.
5129 * This could impact scaling on very large systems. Be reluctant to
5130 * use notify_on_release cgroups where very high task exit scaling
5131 * is required on large systems.
5133 * the_top_cgroup_hack:
5135 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
5137 * We call cgroup_exit() while the task is still competent to
5138 * handle notify_on_release(), then leave the task attached to the
5139 * root cgroup in each hierarchy for the remainder of its exit.
5141 * To do this properly, we would increment the reference count on
5142 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
5143 * code we would add a second cgroup function call, to drop that
5144 * reference. This would just create an unnecessary hot spot on
5145 * the top_cgroup reference count, to no avail.
5147 * Normally, holding a reference to a cgroup without bumping its
5148 * count is unsafe. The cgroup could go away, or someone could
5149 * attach us to a different cgroup, decrementing the count on
5150 * the first cgroup that we never incremented. But in this case,
5151 * top_cgroup isn't going away, and either task has PF_EXITING set,
5152 * which wards off any cgroup_attach_task() attempts, or task is a failed
5153 * fork, never visible to cgroup_attach_task.
5155 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
5157 struct cgroup_subsys
*ss
;
5158 struct css_set
*cset
;
5162 * Unlink from the css_set task list if necessary.
5163 * Optimistically check cg_list before taking
5166 if (!list_empty(&tsk
->cg_list
)) {
5167 write_lock(&css_set_lock
);
5168 if (!list_empty(&tsk
->cg_list
))
5169 list_del_init(&tsk
->cg_list
);
5170 write_unlock(&css_set_lock
);
5173 /* Reassign the task to the init_css_set. */
5175 cset
= task_css_set(tsk
);
5176 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5178 if (run_callbacks
&& need_forkexit_callback
) {
5180 * fork/exit callbacks are supported only for builtin
5181 * subsystems, see cgroup_post_fork() for details.
5183 for_each_builtin_subsys(ss
, i
) {
5185 struct cgroup
*old_cgrp
= cset
->subsys
[i
]->cgroup
;
5186 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
5188 ss
->exit(cgrp
, old_cgrp
, tsk
);
5194 put_css_set_taskexit(cset
);
5197 static void check_for_release(struct cgroup
*cgrp
)
5199 if (cgroup_is_releasable(cgrp
) &&
5200 list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
)) {
5202 * Control Group is currently removeable. If it's not
5203 * already queued for a userspace notification, queue
5206 int need_schedule_work
= 0;
5208 raw_spin_lock(&release_list_lock
);
5209 if (!cgroup_is_dead(cgrp
) &&
5210 list_empty(&cgrp
->release_list
)) {
5211 list_add(&cgrp
->release_list
, &release_list
);
5212 need_schedule_work
= 1;
5214 raw_spin_unlock(&release_list_lock
);
5215 if (need_schedule_work
)
5216 schedule_work(&release_agent_work
);
5221 * Notify userspace when a cgroup is released, by running the
5222 * configured release agent with the name of the cgroup (path
5223 * relative to the root of cgroup file system) as the argument.
5225 * Most likely, this user command will try to rmdir this cgroup.
5227 * This races with the possibility that some other task will be
5228 * attached to this cgroup before it is removed, or that some other
5229 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5230 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5231 * unused, and this cgroup will be reprieved from its death sentence,
5232 * to continue to serve a useful existence. Next time it's released,
5233 * we will get notified again, if it still has 'notify_on_release' set.
5235 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5236 * means only wait until the task is successfully execve()'d. The
5237 * separate release agent task is forked by call_usermodehelper(),
5238 * then control in this thread returns here, without waiting for the
5239 * release agent task. We don't bother to wait because the caller of
5240 * this routine has no use for the exit status of the release agent
5241 * task, so no sense holding our caller up for that.
5243 static void cgroup_release_agent(struct work_struct
*work
)
5245 BUG_ON(work
!= &release_agent_work
);
5246 mutex_lock(&cgroup_mutex
);
5247 raw_spin_lock(&release_list_lock
);
5248 while (!list_empty(&release_list
)) {
5249 char *argv
[3], *envp
[3];
5251 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5252 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5255 list_del_init(&cgrp
->release_list
);
5256 raw_spin_unlock(&release_list_lock
);
5257 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5260 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5262 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5267 argv
[i
++] = agentbuf
;
5268 argv
[i
++] = pathbuf
;
5272 /* minimal command environment */
5273 envp
[i
++] = "HOME=/";
5274 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5277 /* Drop the lock while we invoke the usermode helper,
5278 * since the exec could involve hitting disk and hence
5279 * be a slow process */
5280 mutex_unlock(&cgroup_mutex
);
5281 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5282 mutex_lock(&cgroup_mutex
);
5286 raw_spin_lock(&release_list_lock
);
5288 raw_spin_unlock(&release_list_lock
);
5289 mutex_unlock(&cgroup_mutex
);
5292 static int __init
cgroup_disable(char *str
)
5294 struct cgroup_subsys
*ss
;
5298 while ((token
= strsep(&str
, ",")) != NULL
) {
5303 * cgroup_disable, being at boot time, can't know about
5304 * module subsystems, so we don't worry about them.
5306 for_each_builtin_subsys(ss
, i
) {
5307 if (!strcmp(token
, ss
->name
)) {
5309 printk(KERN_INFO
"Disabling %s control group"
5310 " subsystem\n", ss
->name
);
5317 __setup("cgroup_disable=", cgroup_disable
);
5320 * Functons for CSS ID.
5323 /* to get ID other than 0, this should be called when !cgroup_is_dead() */
5324 unsigned short css_id(struct cgroup_subsys_state
*css
)
5326 struct css_id
*cssid
;
5329 * This css_id() can return correct value when somone has refcnt
5330 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5331 * it's unchanged until freed.
5333 cssid
= rcu_dereference_raw(css
->id
);
5339 EXPORT_SYMBOL_GPL(css_id
);
5342 * css_is_ancestor - test "root" css is an ancestor of "child"
5343 * @child: the css to be tested.
5344 * @root: the css supporsed to be an ancestor of the child.
5346 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5347 * this function reads css->id, the caller must hold rcu_read_lock().
5348 * But, considering usual usage, the csses should be valid objects after test.
5349 * Assuming that the caller will do some action to the child if this returns
5350 * returns true, the caller must take "child";s reference count.
5351 * If "child" is valid object and this returns true, "root" is valid, too.
5354 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5355 const struct cgroup_subsys_state
*root
)
5357 struct css_id
*child_id
;
5358 struct css_id
*root_id
;
5360 child_id
= rcu_dereference(child
->id
);
5363 root_id
= rcu_dereference(root
->id
);
5366 if (child_id
->depth
< root_id
->depth
)
5368 if (child_id
->stack
[root_id
->depth
] != root_id
->id
)
5373 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5375 struct css_id
*id
= rcu_dereference_protected(css
->id
, true);
5377 /* When this is called before css_id initialization, id can be NULL */
5381 BUG_ON(!ss
->use_id
);
5383 rcu_assign_pointer(id
->css
, NULL
);
5384 rcu_assign_pointer(css
->id
, NULL
);
5385 spin_lock(&ss
->id_lock
);
5386 idr_remove(&ss
->idr
, id
->id
);
5387 spin_unlock(&ss
->id_lock
);
5388 kfree_rcu(id
, rcu_head
);
5390 EXPORT_SYMBOL_GPL(free_css_id
);
5393 * This is called by init or create(). Then, calls to this function are
5394 * always serialized (By cgroup_mutex() at create()).
5397 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5399 struct css_id
*newid
;
5402 BUG_ON(!ss
->use_id
);
5404 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5405 newid
= kzalloc(size
, GFP_KERNEL
);
5407 return ERR_PTR(-ENOMEM
);
5409 idr_preload(GFP_KERNEL
);
5410 spin_lock(&ss
->id_lock
);
5411 /* Don't use 0. allocates an ID of 1-65535 */
5412 ret
= idr_alloc(&ss
->idr
, newid
, 1, CSS_ID_MAX
+ 1, GFP_NOWAIT
);
5413 spin_unlock(&ss
->id_lock
);
5416 /* Returns error when there are no free spaces for new ID.*/
5421 newid
->depth
= depth
;
5425 return ERR_PTR(ret
);
5429 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5430 struct cgroup_subsys_state
*rootcss
)
5432 struct css_id
*newid
;
5434 spin_lock_init(&ss
->id_lock
);
5437 newid
= get_new_cssid(ss
, 0);
5439 return PTR_ERR(newid
);
5441 newid
->stack
[0] = newid
->id
;
5442 RCU_INIT_POINTER(newid
->css
, rootcss
);
5443 RCU_INIT_POINTER(rootcss
->id
, newid
);
5447 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5448 struct cgroup
*child
)
5450 int subsys_id
, i
, depth
= 0;
5451 struct cgroup_subsys_state
*parent_css
, *child_css
;
5452 struct css_id
*child_id
, *parent_id
;
5454 subsys_id
= ss
->subsys_id
;
5455 parent_css
= parent
->subsys
[subsys_id
];
5456 child_css
= child
->subsys
[subsys_id
];
5457 parent_id
= rcu_dereference_protected(parent_css
->id
, true);
5458 depth
= parent_id
->depth
+ 1;
5460 child_id
= get_new_cssid(ss
, depth
);
5461 if (IS_ERR(child_id
))
5462 return PTR_ERR(child_id
);
5464 for (i
= 0; i
< depth
; i
++)
5465 child_id
->stack
[i
] = parent_id
->stack
[i
];
5466 child_id
->stack
[depth
] = child_id
->id
;
5468 * child_id->css pointer will be set after this cgroup is available
5469 * see cgroup_populate_dir()
5471 rcu_assign_pointer(child_css
->id
, child_id
);
5477 * css_lookup - lookup css by id
5478 * @ss: cgroup subsys to be looked into.
5481 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5482 * NULL if not. Should be called under rcu_read_lock()
5484 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5486 struct css_id
*cssid
= NULL
;
5488 BUG_ON(!ss
->use_id
);
5489 cssid
= idr_find(&ss
->idr
, id
);
5491 if (unlikely(!cssid
))
5494 return rcu_dereference(cssid
->css
);
5496 EXPORT_SYMBOL_GPL(css_lookup
);
5499 * get corresponding css from file open on cgroupfs directory
5501 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5503 struct cgroup
*cgrp
;
5504 struct inode
*inode
;
5505 struct cgroup_subsys_state
*css
;
5507 inode
= file_inode(f
);
5508 /* check in cgroup filesystem dir */
5509 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5510 return ERR_PTR(-EBADF
);
5512 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5513 return ERR_PTR(-EINVAL
);
5516 cgrp
= __d_cgrp(f
->f_dentry
);
5517 css
= cgrp
->subsys
[id
];
5518 return css
? css
: ERR_PTR(-ENOENT
);
5521 #ifdef CONFIG_CGROUP_DEBUG
5522 static struct cgroup_subsys_state
*debug_css_alloc(struct cgroup
*cgrp
)
5524 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5527 return ERR_PTR(-ENOMEM
);
5532 static void debug_css_free(struct cgroup
*cgrp
)
5534 kfree(cgrp
->subsys
[debug_subsys_id
]);
5537 static u64
debug_taskcount_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5539 return cgroup_task_count(cgrp
);
5542 static u64
current_css_set_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5544 return (u64
)(unsigned long)current
->cgroups
;
5547 static u64
current_css_set_refcount_read(struct cgroup
*cgrp
,
5553 count
= atomic_read(&task_css_set(current
)->refcount
);
5558 static int current_css_set_cg_links_read(struct cgroup
*cgrp
,
5560 struct seq_file
*seq
)
5562 struct cgrp_cset_link
*link
;
5563 struct css_set
*cset
;
5565 read_lock(&css_set_lock
);
5567 cset
= rcu_dereference(current
->cgroups
);
5568 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5569 struct cgroup
*c
= link
->cgrp
;
5573 name
= c
->dentry
->d_name
.name
;
5576 seq_printf(seq
, "Root %d group %s\n",
5577 c
->root
->hierarchy_id
, name
);
5580 read_unlock(&css_set_lock
);
5584 #define MAX_TASKS_SHOWN_PER_CSS 25
5585 static int cgroup_css_links_read(struct cgroup
*cgrp
,
5587 struct seq_file
*seq
)
5589 struct cgrp_cset_link
*link
;
5591 read_lock(&css_set_lock
);
5592 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
) {
5593 struct css_set
*cset
= link
->cset
;
5594 struct task_struct
*task
;
5596 seq_printf(seq
, "css_set %p\n", cset
);
5597 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5598 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5599 seq_puts(seq
, " ...\n");
5602 seq_printf(seq
, " task %d\n",
5603 task_pid_vnr(task
));
5607 read_unlock(&css_set_lock
);
5611 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5613 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5616 static struct cftype debug_files
[] = {
5618 .name
= "taskcount",
5619 .read_u64
= debug_taskcount_read
,
5623 .name
= "current_css_set",
5624 .read_u64
= current_css_set_read
,
5628 .name
= "current_css_set_refcount",
5629 .read_u64
= current_css_set_refcount_read
,
5633 .name
= "current_css_set_cg_links",
5634 .read_seq_string
= current_css_set_cg_links_read
,
5638 .name
= "cgroup_css_links",
5639 .read_seq_string
= cgroup_css_links_read
,
5643 .name
= "releasable",
5644 .read_u64
= releasable_read
,
5650 struct cgroup_subsys debug_subsys
= {
5652 .css_alloc
= debug_css_alloc
,
5653 .css_free
= debug_css_free
,
5654 .subsys_id
= debug_subsys_id
,
5655 .base_cftypes
= debug_files
,
5657 #endif /* CONFIG_CGROUP_DEBUG */