2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
33 #include <linux/init_task.h>
34 #include <linux/kernel.h>
35 #include <linux/list.h>
37 #include <linux/mutex.h>
38 #include <linux/mount.h>
39 #include <linux/pagemap.h>
40 #include <linux/proc_fs.h>
41 #include <linux/rcupdate.h>
42 #include <linux/sched.h>
43 #include <linux/backing-dev.h>
44 #include <linux/seq_file.h>
45 #include <linux/slab.h>
46 #include <linux/magic.h>
47 #include <linux/spinlock.h>
48 #include <linux/string.h>
49 #include <linux/sort.h>
50 #include <linux/kmod.h>
51 #include <linux/module.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/namei.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/idr.h>
58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
59 #include <linux/eventfd.h>
60 #include <linux/poll.h>
61 #include <linux/flex_array.h> /* used in cgroup_attach_task */
62 #include <linux/kthread.h>
64 #include <linux/atomic.h>
67 * cgroup_mutex is the master lock. Any modification to cgroup or its
68 * hierarchy must be performed while holding it.
70 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
71 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
72 * release_agent_path and so on. Modifying requires both cgroup_mutex and
73 * cgroup_root_mutex. Readers can acquire either of the two. This is to
74 * break the following locking order cycle.
76 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
77 * B. namespace_sem -> cgroup_mutex
79 * B happens only through cgroup_show_options() and using cgroup_root_mutex
82 #ifdef CONFIG_PROVE_RCU
83 DEFINE_MUTEX(cgroup_mutex
);
84 EXPORT_SYMBOL_GPL(cgroup_mutex
); /* only for task_subsys_state_check() */
86 static DEFINE_MUTEX(cgroup_mutex
);
89 static DEFINE_MUTEX(cgroup_root_mutex
);
92 * Generate an array of cgroup subsystem pointers. At boot time, this is
93 * populated with the built in subsystems, and modular subsystems are
94 * registered after that. The mutable section of this array is protected by
97 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
98 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
99 static struct cgroup_subsys
*cgroup_subsys
[CGROUP_SUBSYS_COUNT
] = {
100 #include <linux/cgroup_subsys.h>
104 * The dummy hierarchy, reserved for the subsystems that are otherwise
105 * unattached - it never has more than a single cgroup, and all tasks are
106 * part of that cgroup.
108 static struct cgroupfs_root cgroup_dummy_root
;
110 /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */
111 static struct cgroup
* const cgroup_dummy_top
= &cgroup_dummy_root
.top_cgroup
;
114 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
117 struct list_head node
;
118 struct dentry
*dentry
;
122 struct simple_xattrs xattrs
;
126 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
127 * cgroup_subsys->use_id != 0.
129 #define CSS_ID_MAX (65535)
132 * The css to which this ID points. This pointer is set to valid value
133 * after cgroup is populated. If cgroup is removed, this will be NULL.
134 * This pointer is expected to be RCU-safe because destroy()
135 * is called after synchronize_rcu(). But for safe use, css_tryget()
136 * should be used for avoiding race.
138 struct cgroup_subsys_state __rcu
*css
;
144 * Depth in hierarchy which this ID belongs to.
146 unsigned short depth
;
148 * ID is freed by RCU. (and lookup routine is RCU safe.)
150 struct rcu_head rcu_head
;
152 * Hierarchy of CSS ID belongs to.
154 unsigned short stack
[0]; /* Array of Length (depth+1) */
158 * cgroup_event represents events which userspace want to receive.
160 struct cgroup_event
{
162 * Cgroup which the event belongs to.
166 * Control file which the event associated.
170 * eventfd to signal userspace about the event.
172 struct eventfd_ctx
*eventfd
;
174 * Each of these stored in a list by the cgroup.
176 struct list_head list
;
178 * All fields below needed to unregister event when
179 * userspace closes eventfd.
182 wait_queue_head_t
*wqh
;
184 struct work_struct remove
;
187 /* The list of hierarchy roots */
189 static LIST_HEAD(cgroup_roots
);
190 static int cgroup_root_count
;
193 * Hierarchy ID allocation and mapping. It follows the same exclusion
194 * rules as other root ops - both cgroup_mutex and cgroup_root_mutex for
195 * writes, either for reads.
197 static DEFINE_IDR(cgroup_hierarchy_idr
);
199 static struct cgroup_name root_cgroup_name
= { .name
= "/" };
202 * Assign a monotonically increasing serial number to cgroups. It
203 * guarantees cgroups with bigger numbers are newer than those with smaller
204 * numbers. Also, as cgroups are always appended to the parent's
205 * ->children list, it guarantees that sibling cgroups are always sorted in
206 * the ascending serial number order on the list. Protected by
209 static u64 cgroup_serial_nr_next
= 1;
211 /* This flag indicates whether tasks in the fork and exit paths should
212 * check for fork/exit handlers to call. This avoids us having to do
213 * extra work in the fork/exit path if none of the subsystems need to
216 static int need_forkexit_callback __read_mostly
;
218 static struct cftype cgroup_base_files
[];
220 static void cgroup_offline_fn(struct work_struct
*work
);
221 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
222 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
223 struct cftype cfts
[], bool is_add
);
225 /* convenient tests for these bits */
226 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
228 return test_bit(CGRP_DEAD
, &cgrp
->flags
);
232 * cgroup_is_descendant - test ancestry
233 * @cgrp: the cgroup to be tested
234 * @ancestor: possible ancestor of @cgrp
236 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
237 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
238 * and @ancestor are accessible.
240 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
243 if (cgrp
== ancestor
)
249 EXPORT_SYMBOL_GPL(cgroup_is_descendant
);
251 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
254 (1 << CGRP_RELEASABLE
) |
255 (1 << CGRP_NOTIFY_ON_RELEASE
);
256 return (cgrp
->flags
& bits
) == bits
;
259 static int notify_on_release(const struct cgroup
*cgrp
)
261 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
265 * for_each_subsys - iterate all loaded cgroup subsystems
266 * @ss: the iteration cursor
267 * @i: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
269 * Should be called under cgroup_mutex.
271 #define for_each_subsys(ss, i) \
272 for ((i) = 0; (i) < CGROUP_SUBSYS_COUNT; (i)++) \
273 if (({ lockdep_assert_held(&cgroup_mutex); \
274 !((ss) = cgroup_subsys[i]); })) { } \
278 * for_each_builtin_subsys - iterate all built-in cgroup subsystems
279 * @ss: the iteration cursor
280 * @i: the index of @ss, CGROUP_BUILTIN_SUBSYS_COUNT after reaching the end
282 * Bulit-in subsystems are always present and iteration itself doesn't
283 * require any synchronization.
285 #define for_each_builtin_subsys(ss, i) \
286 for ((i) = 0; (i) < CGROUP_BUILTIN_SUBSYS_COUNT && \
287 (((ss) = cgroup_subsys[i]) || true); (i)++)
289 /* iterate each subsystem attached to a hierarchy */
290 #define for_each_root_subsys(root, ss) \
291 list_for_each_entry((ss), &(root)->subsys_list, sibling)
293 /* iterate across the active hierarchies */
294 #define for_each_active_root(root) \
295 list_for_each_entry((root), &cgroup_roots, root_list)
297 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
299 return dentry
->d_fsdata
;
302 static inline struct cfent
*__d_cfe(struct dentry
*dentry
)
304 return dentry
->d_fsdata
;
307 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
309 return __d_cfe(dentry
)->type
;
313 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
314 * @cgrp: the cgroup to be checked for liveness
316 * On success, returns true; the mutex should be later unlocked. On
317 * failure returns false with no lock held.
319 static bool cgroup_lock_live_group(struct cgroup
*cgrp
)
321 mutex_lock(&cgroup_mutex
);
322 if (cgroup_is_dead(cgrp
)) {
323 mutex_unlock(&cgroup_mutex
);
329 /* the list of cgroups eligible for automatic release. Protected by
330 * release_list_lock */
331 static LIST_HEAD(release_list
);
332 static DEFINE_RAW_SPINLOCK(release_list_lock
);
333 static void cgroup_release_agent(struct work_struct
*work
);
334 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
335 static void check_for_release(struct cgroup
*cgrp
);
338 * A cgroup can be associated with multiple css_sets as different tasks may
339 * belong to different cgroups on different hierarchies. In the other
340 * direction, a css_set is naturally associated with multiple cgroups.
341 * This M:N relationship is represented by the following link structure
342 * which exists for each association and allows traversing the associations
345 struct cgrp_cset_link
{
346 /* the cgroup and css_set this link associates */
348 struct css_set
*cset
;
350 /* list of cgrp_cset_links anchored at cgrp->cset_links */
351 struct list_head cset_link
;
353 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
354 struct list_head cgrp_link
;
357 /* The default css_set - used by init and its children prior to any
358 * hierarchies being mounted. It contains a pointer to the root state
359 * for each subsystem. Also used to anchor the list of css_sets. Not
360 * reference-counted, to improve performance when child cgroups
361 * haven't been created.
364 static struct css_set init_css_set
;
365 static struct cgrp_cset_link init_cgrp_cset_link
;
367 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
368 struct cgroup_subsys_state
*css
);
370 /* css_set_lock protects the list of css_set objects, and the
371 * chain of tasks off each css_set. Nests outside task->alloc_lock
372 * due to cgroup_iter_start() */
373 static DEFINE_RWLOCK(css_set_lock
);
374 static int css_set_count
;
377 * hash table for cgroup groups. This improves the performance to find
378 * an existing css_set. This hash doesn't (currently) take into
379 * account cgroups in empty hierarchies.
381 #define CSS_SET_HASH_BITS 7
382 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
384 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
386 unsigned long key
= 0UL;
387 struct cgroup_subsys
*ss
;
390 for_each_subsys(ss
, i
)
391 key
+= (unsigned long)css
[i
];
392 key
= (key
>> 16) ^ key
;
397 /* We don't maintain the lists running through each css_set to its
398 * task until after the first call to cgroup_iter_start(). This
399 * reduces the fork()/exit() overhead for people who have cgroups
400 * compiled into their kernel but not actually in use */
401 static int use_task_css_set_links __read_mostly
;
403 static void __put_css_set(struct css_set
*cset
, int taskexit
)
405 struct cgrp_cset_link
*link
, *tmp_link
;
408 * Ensure that the refcount doesn't hit zero while any readers
409 * can see it. Similar to atomic_dec_and_lock(), but for an
412 if (atomic_add_unless(&cset
->refcount
, -1, 1))
414 write_lock(&css_set_lock
);
415 if (!atomic_dec_and_test(&cset
->refcount
)) {
416 write_unlock(&css_set_lock
);
420 /* This css_set is dead. unlink it and release cgroup refcounts */
421 hash_del(&cset
->hlist
);
424 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
425 struct cgroup
*cgrp
= link
->cgrp
;
427 list_del(&link
->cset_link
);
428 list_del(&link
->cgrp_link
);
430 /* @cgrp can't go away while we're holding css_set_lock */
431 if (list_empty(&cgrp
->cset_links
) && notify_on_release(cgrp
)) {
433 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
434 check_for_release(cgrp
);
440 write_unlock(&css_set_lock
);
441 kfree_rcu(cset
, rcu_head
);
445 * refcounted get/put for css_set objects
447 static inline void get_css_set(struct css_set
*cset
)
449 atomic_inc(&cset
->refcount
);
452 static inline void put_css_set(struct css_set
*cset
)
454 __put_css_set(cset
, 0);
457 static inline void put_css_set_taskexit(struct css_set
*cset
)
459 __put_css_set(cset
, 1);
463 * compare_css_sets - helper function for find_existing_css_set().
464 * @cset: candidate css_set being tested
465 * @old_cset: existing css_set for a task
466 * @new_cgrp: cgroup that's being entered by the task
467 * @template: desired set of css pointers in css_set (pre-calculated)
469 * Returns true if "cg" matches "old_cg" except for the hierarchy
470 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
472 static bool compare_css_sets(struct css_set
*cset
,
473 struct css_set
*old_cset
,
474 struct cgroup
*new_cgrp
,
475 struct cgroup_subsys_state
*template[])
477 struct list_head
*l1
, *l2
;
479 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
))) {
480 /* Not all subsystems matched */
485 * Compare cgroup pointers in order to distinguish between
486 * different cgroups in heirarchies with no subsystems. We
487 * could get by with just this check alone (and skip the
488 * memcmp above) but on most setups the memcmp check will
489 * avoid the need for this more expensive check on almost all
493 l1
= &cset
->cgrp_links
;
494 l2
= &old_cset
->cgrp_links
;
496 struct cgrp_cset_link
*link1
, *link2
;
497 struct cgroup
*cgrp1
, *cgrp2
;
501 /* See if we reached the end - both lists are equal length. */
502 if (l1
== &cset
->cgrp_links
) {
503 BUG_ON(l2
!= &old_cset
->cgrp_links
);
506 BUG_ON(l2
== &old_cset
->cgrp_links
);
508 /* Locate the cgroups associated with these links. */
509 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
510 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
513 /* Hierarchies should be linked in the same order. */
514 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
517 * If this hierarchy is the hierarchy of the cgroup
518 * that's changing, then we need to check that this
519 * css_set points to the new cgroup; if it's any other
520 * hierarchy, then this css_set should point to the
521 * same cgroup as the old css_set.
523 if (cgrp1
->root
== new_cgrp
->root
) {
524 if (cgrp1
!= new_cgrp
)
535 * find_existing_css_set - init css array and find the matching css_set
536 * @old_cset: the css_set that we're using before the cgroup transition
537 * @cgrp: the cgroup that we're moving into
538 * @template: out param for the new set of csses, should be clear on entry
540 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
542 struct cgroup_subsys_state
*template[])
544 struct cgroupfs_root
*root
= cgrp
->root
;
545 struct cgroup_subsys
*ss
;
546 struct css_set
*cset
;
551 * Build the set of subsystem state objects that we want to see in the
552 * new css_set. while subsystems can change globally, the entries here
553 * won't change, so no need for locking.
555 for_each_subsys(ss
, i
) {
556 if (root
->subsys_mask
& (1UL << i
)) {
557 /* Subsystem is in this hierarchy. So we want
558 * the subsystem state from the new
560 template[i
] = cgrp
->subsys
[i
];
562 /* Subsystem is not in this hierarchy, so we
563 * don't want to change the subsystem state */
564 template[i
] = old_cset
->subsys
[i
];
568 key
= css_set_hash(template);
569 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
570 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
573 /* This css_set matches what we need */
577 /* No existing cgroup group matched */
581 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
583 struct cgrp_cset_link
*link
, *tmp_link
;
585 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
586 list_del(&link
->cset_link
);
592 * allocate_cgrp_cset_links - allocate cgrp_cset_links
593 * @count: the number of links to allocate
594 * @tmp_links: list_head the allocated links are put on
596 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
597 * through ->cset_link. Returns 0 on success or -errno.
599 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
601 struct cgrp_cset_link
*link
;
604 INIT_LIST_HEAD(tmp_links
);
606 for (i
= 0; i
< count
; i
++) {
607 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
609 free_cgrp_cset_links(tmp_links
);
612 list_add(&link
->cset_link
, tmp_links
);
618 * link_css_set - a helper function to link a css_set to a cgroup
619 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
620 * @cset: the css_set to be linked
621 * @cgrp: the destination cgroup
623 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
626 struct cgrp_cset_link
*link
;
628 BUG_ON(list_empty(tmp_links
));
629 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
632 list_move(&link
->cset_link
, &cgrp
->cset_links
);
634 * Always add links to the tail of the list so that the list
635 * is sorted by order of hierarchy creation
637 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
641 * find_css_set - return a new css_set with one cgroup updated
642 * @old_cset: the baseline css_set
643 * @cgrp: the cgroup to be updated
645 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
646 * substituted into the appropriate hierarchy.
648 static struct css_set
*find_css_set(struct css_set
*old_cset
,
651 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
652 struct css_set
*cset
;
653 struct list_head tmp_links
;
654 struct cgrp_cset_link
*link
;
657 lockdep_assert_held(&cgroup_mutex
);
659 /* First see if we already have a cgroup group that matches
661 read_lock(&css_set_lock
);
662 cset
= find_existing_css_set(old_cset
, cgrp
, template);
665 read_unlock(&css_set_lock
);
670 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
674 /* Allocate all the cgrp_cset_link objects that we'll need */
675 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
680 atomic_set(&cset
->refcount
, 1);
681 INIT_LIST_HEAD(&cset
->cgrp_links
);
682 INIT_LIST_HEAD(&cset
->tasks
);
683 INIT_HLIST_NODE(&cset
->hlist
);
685 /* Copy the set of subsystem state objects generated in
686 * find_existing_css_set() */
687 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
689 write_lock(&css_set_lock
);
690 /* Add reference counts and links from the new css_set. */
691 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
692 struct cgroup
*c
= link
->cgrp
;
694 if (c
->root
== cgrp
->root
)
696 link_css_set(&tmp_links
, cset
, c
);
699 BUG_ON(!list_empty(&tmp_links
));
703 /* Add this cgroup group to the hash table */
704 key
= css_set_hash(cset
->subsys
);
705 hash_add(css_set_table
, &cset
->hlist
, key
);
707 write_unlock(&css_set_lock
);
713 * Return the cgroup for "task" from the given hierarchy. Must be
714 * called with cgroup_mutex held.
716 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
717 struct cgroupfs_root
*root
)
719 struct css_set
*cset
;
720 struct cgroup
*res
= NULL
;
722 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
723 read_lock(&css_set_lock
);
725 * No need to lock the task - since we hold cgroup_mutex the
726 * task can't change groups, so the only thing that can happen
727 * is that it exits and its css is set back to init_css_set.
729 cset
= task_css_set(task
);
730 if (cset
== &init_css_set
) {
731 res
= &root
->top_cgroup
;
733 struct cgrp_cset_link
*link
;
735 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
736 struct cgroup
*c
= link
->cgrp
;
738 if (c
->root
== root
) {
744 read_unlock(&css_set_lock
);
750 * There is one global cgroup mutex. We also require taking
751 * task_lock() when dereferencing a task's cgroup subsys pointers.
752 * See "The task_lock() exception", at the end of this comment.
754 * A task must hold cgroup_mutex to modify cgroups.
756 * Any task can increment and decrement the count field without lock.
757 * So in general, code holding cgroup_mutex can't rely on the count
758 * field not changing. However, if the count goes to zero, then only
759 * cgroup_attach_task() can increment it again. Because a count of zero
760 * means that no tasks are currently attached, therefore there is no
761 * way a task attached to that cgroup can fork (the other way to
762 * increment the count). So code holding cgroup_mutex can safely
763 * assume that if the count is zero, it will stay zero. Similarly, if
764 * a task holds cgroup_mutex on a cgroup with zero count, it
765 * knows that the cgroup won't be removed, as cgroup_rmdir()
768 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
769 * (usually) take cgroup_mutex. These are the two most performance
770 * critical pieces of code here. The exception occurs on cgroup_exit(),
771 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
772 * is taken, and if the cgroup count is zero, a usermode call made
773 * to the release agent with the name of the cgroup (path relative to
774 * the root of cgroup file system) as the argument.
776 * A cgroup can only be deleted if both its 'count' of using tasks
777 * is zero, and its list of 'children' cgroups is empty. Since all
778 * tasks in the system use _some_ cgroup, and since there is always at
779 * least one task in the system (init, pid == 1), therefore, top_cgroup
780 * always has either children cgroups and/or using tasks. So we don't
781 * need a special hack to ensure that top_cgroup cannot be deleted.
783 * The task_lock() exception
785 * The need for this exception arises from the action of
786 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
787 * another. It does so using cgroup_mutex, however there are
788 * several performance critical places that need to reference
789 * task->cgroup without the expense of grabbing a system global
790 * mutex. Therefore except as noted below, when dereferencing or, as
791 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
792 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
793 * the task_struct routinely used for such matters.
795 * P.S. One more locking exception. RCU is used to guard the
796 * update of a tasks cgroup pointer by cgroup_attach_task()
800 * A couple of forward declarations required, due to cyclic reference loop:
801 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
802 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
806 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
807 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, unsigned int);
808 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
809 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
);
810 static const struct inode_operations cgroup_dir_inode_operations
;
811 static const struct file_operations proc_cgroupstats_operations
;
813 static struct backing_dev_info cgroup_backing_dev_info
= {
815 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
818 static int alloc_css_id(struct cgroup_subsys
*ss
,
819 struct cgroup
*parent
, struct cgroup
*child
);
821 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
823 struct inode
*inode
= new_inode(sb
);
826 inode
->i_ino
= get_next_ino();
827 inode
->i_mode
= mode
;
828 inode
->i_uid
= current_fsuid();
829 inode
->i_gid
= current_fsgid();
830 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
831 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
836 static struct cgroup_name
*cgroup_alloc_name(struct dentry
*dentry
)
838 struct cgroup_name
*name
;
840 name
= kmalloc(sizeof(*name
) + dentry
->d_name
.len
+ 1, GFP_KERNEL
);
843 strcpy(name
->name
, dentry
->d_name
.name
);
847 static void cgroup_free_fn(struct work_struct
*work
)
849 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, destroy_work
);
850 struct cgroup_subsys
*ss
;
852 mutex_lock(&cgroup_mutex
);
854 * Release the subsystem state objects.
856 for_each_root_subsys(cgrp
->root
, ss
)
859 cgrp
->root
->number_of_cgroups
--;
860 mutex_unlock(&cgroup_mutex
);
863 * We get a ref to the parent's dentry, and put the ref when
864 * this cgroup is being freed, so it's guaranteed that the
865 * parent won't be destroyed before its children.
867 dput(cgrp
->parent
->dentry
);
869 ida_simple_remove(&cgrp
->root
->cgroup_ida
, cgrp
->id
);
872 * Drop the active superblock reference that we took when we
873 * created the cgroup. This will free cgrp->root, if we are
874 * holding the last reference to @sb.
876 deactivate_super(cgrp
->root
->sb
);
879 * if we're getting rid of the cgroup, refcount should ensure
880 * that there are no pidlists left.
882 BUG_ON(!list_empty(&cgrp
->pidlists
));
884 simple_xattrs_free(&cgrp
->xattrs
);
886 kfree(rcu_dereference_raw(cgrp
->name
));
890 static void cgroup_free_rcu(struct rcu_head
*head
)
892 struct cgroup
*cgrp
= container_of(head
, struct cgroup
, rcu_head
);
894 INIT_WORK(&cgrp
->destroy_work
, cgroup_free_fn
);
895 schedule_work(&cgrp
->destroy_work
);
898 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
900 /* is dentry a directory ? if so, kfree() associated cgroup */
901 if (S_ISDIR(inode
->i_mode
)) {
902 struct cgroup
*cgrp
= dentry
->d_fsdata
;
904 BUG_ON(!(cgroup_is_dead(cgrp
)));
905 call_rcu(&cgrp
->rcu_head
, cgroup_free_rcu
);
907 struct cfent
*cfe
= __d_cfe(dentry
);
908 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
910 WARN_ONCE(!list_empty(&cfe
->node
) &&
911 cgrp
!= &cgrp
->root
->top_cgroup
,
912 "cfe still linked for %s\n", cfe
->type
->name
);
913 simple_xattrs_free(&cfe
->xattrs
);
919 static int cgroup_delete(const struct dentry
*d
)
924 static void remove_dir(struct dentry
*d
)
926 struct dentry
*parent
= dget(d
->d_parent
);
929 simple_rmdir(parent
->d_inode
, d
);
933 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
937 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
938 lockdep_assert_held(&cgroup_mutex
);
941 * If we're doing cleanup due to failure of cgroup_create(),
942 * the corresponding @cfe may not exist.
944 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
945 struct dentry
*d
= cfe
->dentry
;
947 if (cft
&& cfe
->type
!= cft
)
952 simple_unlink(cgrp
->dentry
->d_inode
, d
);
953 list_del_init(&cfe
->node
);
961 * cgroup_clear_dir - remove subsys files in a cgroup directory
962 * @cgrp: target cgroup
963 * @subsys_mask: mask of the subsystem ids whose files should be removed
965 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
967 struct cgroup_subsys
*ss
;
970 for_each_subsys(ss
, i
) {
971 struct cftype_set
*set
;
973 if (!test_bit(i
, &subsys_mask
))
975 list_for_each_entry(set
, &ss
->cftsets
, node
)
976 cgroup_addrm_files(cgrp
, NULL
, set
->cfts
, false);
981 * NOTE : the dentry must have been dget()'ed
983 static void cgroup_d_remove_dir(struct dentry
*dentry
)
985 struct dentry
*parent
;
987 parent
= dentry
->d_parent
;
988 spin_lock(&parent
->d_lock
);
989 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
990 list_del_init(&dentry
->d_u
.d_child
);
991 spin_unlock(&dentry
->d_lock
);
992 spin_unlock(&parent
->d_lock
);
997 * Call with cgroup_mutex held. Drops reference counts on modules, including
998 * any duplicate ones that parse_cgroupfs_options took. If this function
999 * returns an error, no reference counts are touched.
1001 static int rebind_subsystems(struct cgroupfs_root
*root
,
1002 unsigned long added_mask
, unsigned removed_mask
)
1004 struct cgroup
*cgrp
= &root
->top_cgroup
;
1005 struct cgroup_subsys
*ss
;
1008 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1009 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1011 /* Check that any added subsystems are currently free */
1012 for_each_subsys(ss
, i
) {
1013 unsigned long bit
= 1UL << i
;
1015 if (!(bit
& added_mask
))
1018 if (ss
->root
!= &cgroup_dummy_root
) {
1019 /* Subsystem isn't free */
1024 /* Currently we don't handle adding/removing subsystems when
1025 * any child cgroups exist. This is theoretically supportable
1026 * but involves complex error handling, so it's being left until
1028 if (root
->number_of_cgroups
> 1)
1031 /* Process each subsystem */
1032 for_each_subsys(ss
, i
) {
1033 unsigned long bit
= 1UL << i
;
1035 if (bit
& added_mask
) {
1036 /* We're binding this subsystem to this hierarchy */
1037 BUG_ON(cgrp
->subsys
[i
]);
1038 BUG_ON(!cgroup_dummy_top
->subsys
[i
]);
1039 BUG_ON(cgroup_dummy_top
->subsys
[i
]->cgroup
!= cgroup_dummy_top
);
1041 cgrp
->subsys
[i
] = cgroup_dummy_top
->subsys
[i
];
1042 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1043 list_move(&ss
->sibling
, &root
->subsys_list
);
1048 /* refcount was already taken, and we're keeping it */
1049 root
->subsys_mask
|= bit
;
1050 } else if (bit
& removed_mask
) {
1051 /* We're removing this subsystem */
1052 BUG_ON(cgrp
->subsys
[i
] != cgroup_dummy_top
->subsys
[i
]);
1053 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1056 ss
->bind(cgroup_dummy_top
);
1057 cgroup_dummy_top
->subsys
[i
]->cgroup
= cgroup_dummy_top
;
1058 cgrp
->subsys
[i
] = NULL
;
1059 cgroup_subsys
[i
]->root
= &cgroup_dummy_root
;
1060 list_move(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
1062 /* subsystem is now free - drop reference on module */
1063 module_put(ss
->module
);
1064 root
->subsys_mask
&= ~bit
;
1065 } else if (bit
& root
->subsys_mask
) {
1066 /* Subsystem state should already exist */
1067 BUG_ON(!cgrp
->subsys
[i
]);
1069 * a refcount was taken, but we already had one, so
1070 * drop the extra reference.
1072 module_put(ss
->module
);
1073 #ifdef CONFIG_MODULE_UNLOAD
1074 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1077 /* Subsystem state shouldn't exist */
1078 BUG_ON(cgrp
->subsys
[i
]);
1083 * Mark @root has finished binding subsystems. @root->subsys_mask
1084 * now matches the bound subsystems.
1086 root
->flags
|= CGRP_ROOT_SUBSYS_BOUND
;
1091 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1093 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1094 struct cgroup_subsys
*ss
;
1096 mutex_lock(&cgroup_root_mutex
);
1097 for_each_root_subsys(root
, ss
)
1098 seq_printf(seq
, ",%s", ss
->name
);
1099 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
)
1100 seq_puts(seq
, ",sane_behavior");
1101 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1102 seq_puts(seq
, ",noprefix");
1103 if (root
->flags
& CGRP_ROOT_XATTR
)
1104 seq_puts(seq
, ",xattr");
1105 if (strlen(root
->release_agent_path
))
1106 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1107 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
))
1108 seq_puts(seq
, ",clone_children");
1109 if (strlen(root
->name
))
1110 seq_printf(seq
, ",name=%s", root
->name
);
1111 mutex_unlock(&cgroup_root_mutex
);
1115 struct cgroup_sb_opts
{
1116 unsigned long subsys_mask
;
1117 unsigned long flags
;
1118 char *release_agent
;
1119 bool cpuset_clone_children
;
1121 /* User explicitly requested empty subsystem */
1124 struct cgroupfs_root
*new_root
;
1129 * Convert a hierarchy specifier into a bitmask of subsystems and
1130 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1131 * array. This function takes refcounts on subsystems to be used, unless it
1132 * returns error, in which case no refcounts are taken.
1134 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1136 char *token
, *o
= data
;
1137 bool all_ss
= false, one_ss
= false;
1138 unsigned long mask
= (unsigned long)-1;
1139 bool module_pin_failed
= false;
1140 struct cgroup_subsys
*ss
;
1143 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1145 #ifdef CONFIG_CPUSETS
1146 mask
= ~(1UL << cpuset_subsys_id
);
1149 memset(opts
, 0, sizeof(*opts
));
1151 while ((token
= strsep(&o
, ",")) != NULL
) {
1154 if (!strcmp(token
, "none")) {
1155 /* Explicitly have no subsystems */
1159 if (!strcmp(token
, "all")) {
1160 /* Mutually exclusive option 'all' + subsystem name */
1166 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1167 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1170 if (!strcmp(token
, "noprefix")) {
1171 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1174 if (!strcmp(token
, "clone_children")) {
1175 opts
->cpuset_clone_children
= true;
1178 if (!strcmp(token
, "xattr")) {
1179 opts
->flags
|= CGRP_ROOT_XATTR
;
1182 if (!strncmp(token
, "release_agent=", 14)) {
1183 /* Specifying two release agents is forbidden */
1184 if (opts
->release_agent
)
1186 opts
->release_agent
=
1187 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1188 if (!opts
->release_agent
)
1192 if (!strncmp(token
, "name=", 5)) {
1193 const char *name
= token
+ 5;
1194 /* Can't specify an empty name */
1197 /* Must match [\w.-]+ */
1198 for (i
= 0; i
< strlen(name
); i
++) {
1202 if ((c
== '.') || (c
== '-') || (c
== '_'))
1206 /* Specifying two names is forbidden */
1209 opts
->name
= kstrndup(name
,
1210 MAX_CGROUP_ROOT_NAMELEN
- 1,
1218 for_each_subsys(ss
, i
) {
1219 if (strcmp(token
, ss
->name
))
1224 /* Mutually exclusive option 'all' + subsystem name */
1227 set_bit(i
, &opts
->subsys_mask
);
1232 if (i
== CGROUP_SUBSYS_COUNT
)
1237 * If the 'all' option was specified select all the subsystems,
1238 * otherwise if 'none', 'name=' and a subsystem name options
1239 * were not specified, let's default to 'all'
1241 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1242 for_each_subsys(ss
, i
)
1244 set_bit(i
, &opts
->subsys_mask
);
1246 /* Consistency checks */
1248 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1249 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1251 if (opts
->flags
& CGRP_ROOT_NOPREFIX
) {
1252 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1256 if (opts
->cpuset_clone_children
) {
1257 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1263 * Option noprefix was introduced just for backward compatibility
1264 * with the old cpuset, so we allow noprefix only if mounting just
1265 * the cpuset subsystem.
1267 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1271 /* Can't specify "none" and some subsystems */
1272 if (opts
->subsys_mask
&& opts
->none
)
1276 * We either have to specify by name or by subsystems. (So all
1277 * empty hierarchies must have a name).
1279 if (!opts
->subsys_mask
&& !opts
->name
)
1283 * Grab references on all the modules we'll need, so the subsystems
1284 * don't dance around before rebind_subsystems attaches them. This may
1285 * take duplicate reference counts on a subsystem that's already used,
1286 * but rebind_subsystems handles this case.
1288 for_each_subsys(ss
, i
) {
1289 if (!(opts
->subsys_mask
& (1UL << i
)))
1291 if (!try_module_get(cgroup_subsys
[i
]->module
)) {
1292 module_pin_failed
= true;
1296 if (module_pin_failed
) {
1298 * oops, one of the modules was going away. this means that we
1299 * raced with a module_delete call, and to the user this is
1300 * essentially a "subsystem doesn't exist" case.
1302 for (i
--; i
>= 0; i
--) {
1303 /* drop refcounts only on the ones we took */
1304 unsigned long bit
= 1UL << i
;
1306 if (!(bit
& opts
->subsys_mask
))
1308 module_put(cgroup_subsys
[i
]->module
);
1316 static void drop_parsed_module_refcounts(unsigned long subsys_mask
)
1318 struct cgroup_subsys
*ss
;
1321 mutex_lock(&cgroup_mutex
);
1322 for_each_subsys(ss
, i
)
1323 if (subsys_mask
& (1UL << i
))
1324 module_put(cgroup_subsys
[i
]->module
);
1325 mutex_unlock(&cgroup_mutex
);
1328 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1331 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1332 struct cgroup
*cgrp
= &root
->top_cgroup
;
1333 struct cgroup_sb_opts opts
;
1334 unsigned long added_mask
, removed_mask
;
1336 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1337 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1341 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1342 mutex_lock(&cgroup_mutex
);
1343 mutex_lock(&cgroup_root_mutex
);
1345 /* See what subsystems are wanted */
1346 ret
= parse_cgroupfs_options(data
, &opts
);
1350 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1351 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1352 task_tgid_nr(current
), current
->comm
);
1354 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1355 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1357 /* Don't allow flags or name to change at remount */
1358 if (((opts
.flags
^ root
->flags
) & CGRP_ROOT_OPTION_MASK
) ||
1359 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1360 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1361 opts
.flags
& CGRP_ROOT_OPTION_MASK
, opts
.name
?: "",
1362 root
->flags
& CGRP_ROOT_OPTION_MASK
, root
->name
);
1368 * Clear out the files of subsystems that should be removed, do
1369 * this before rebind_subsystems, since rebind_subsystems may
1370 * change this hierarchy's subsys_list.
1372 cgroup_clear_dir(cgrp
, removed_mask
);
1374 ret
= rebind_subsystems(root
, added_mask
, removed_mask
);
1376 /* rebind_subsystems failed, re-populate the removed files */
1377 cgroup_populate_dir(cgrp
, removed_mask
);
1381 /* re-populate subsystem files */
1382 cgroup_populate_dir(cgrp
, added_mask
);
1384 if (opts
.release_agent
)
1385 strcpy(root
->release_agent_path
, opts
.release_agent
);
1387 kfree(opts
.release_agent
);
1389 mutex_unlock(&cgroup_root_mutex
);
1390 mutex_unlock(&cgroup_mutex
);
1391 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1393 drop_parsed_module_refcounts(opts
.subsys_mask
);
1397 static const struct super_operations cgroup_ops
= {
1398 .statfs
= simple_statfs
,
1399 .drop_inode
= generic_delete_inode
,
1400 .show_options
= cgroup_show_options
,
1401 .remount_fs
= cgroup_remount
,
1404 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1406 INIT_LIST_HEAD(&cgrp
->sibling
);
1407 INIT_LIST_HEAD(&cgrp
->children
);
1408 INIT_LIST_HEAD(&cgrp
->files
);
1409 INIT_LIST_HEAD(&cgrp
->cset_links
);
1410 INIT_LIST_HEAD(&cgrp
->release_list
);
1411 INIT_LIST_HEAD(&cgrp
->pidlists
);
1412 mutex_init(&cgrp
->pidlist_mutex
);
1413 INIT_LIST_HEAD(&cgrp
->event_list
);
1414 spin_lock_init(&cgrp
->event_list_lock
);
1415 simple_xattrs_init(&cgrp
->xattrs
);
1418 static void init_cgroup_root(struct cgroupfs_root
*root
)
1420 struct cgroup
*cgrp
= &root
->top_cgroup
;
1422 INIT_LIST_HEAD(&root
->subsys_list
);
1423 INIT_LIST_HEAD(&root
->root_list
);
1424 root
->number_of_cgroups
= 1;
1426 RCU_INIT_POINTER(cgrp
->name
, &root_cgroup_name
);
1427 init_cgroup_housekeeping(cgrp
);
1430 static int cgroup_init_root_id(struct cgroupfs_root
*root
, int start
, int end
)
1434 lockdep_assert_held(&cgroup_mutex
);
1435 lockdep_assert_held(&cgroup_root_mutex
);
1437 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, start
, end
,
1442 root
->hierarchy_id
= id
;
1446 static void cgroup_exit_root_id(struct cgroupfs_root
*root
)
1448 lockdep_assert_held(&cgroup_mutex
);
1449 lockdep_assert_held(&cgroup_root_mutex
);
1451 if (root
->hierarchy_id
) {
1452 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1453 root
->hierarchy_id
= 0;
1457 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1459 struct cgroup_sb_opts
*opts
= data
;
1460 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1462 /* If we asked for a name then it must match */
1463 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1467 * If we asked for subsystems (or explicitly for no
1468 * subsystems) then they must match
1470 if ((opts
->subsys_mask
|| opts
->none
)
1471 && (opts
->subsys_mask
!= root
->subsys_mask
))
1477 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1479 struct cgroupfs_root
*root
;
1481 if (!opts
->subsys_mask
&& !opts
->none
)
1484 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1486 return ERR_PTR(-ENOMEM
);
1488 init_cgroup_root(root
);
1491 * We need to set @root->subsys_mask now so that @root can be
1492 * matched by cgroup_test_super() before it finishes
1493 * initialization; otherwise, competing mounts with the same
1494 * options may try to bind the same subsystems instead of waiting
1495 * for the first one leading to unexpected mount errors.
1496 * SUBSYS_BOUND will be set once actual binding is complete.
1498 root
->subsys_mask
= opts
->subsys_mask
;
1499 root
->flags
= opts
->flags
;
1500 ida_init(&root
->cgroup_ida
);
1501 if (opts
->release_agent
)
1502 strcpy(root
->release_agent_path
, opts
->release_agent
);
1504 strcpy(root
->name
, opts
->name
);
1505 if (opts
->cpuset_clone_children
)
1506 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1510 static void cgroup_free_root(struct cgroupfs_root
*root
)
1513 /* hierarhcy ID shoulid already have been released */
1514 WARN_ON_ONCE(root
->hierarchy_id
);
1516 ida_destroy(&root
->cgroup_ida
);
1521 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1524 struct cgroup_sb_opts
*opts
= data
;
1526 /* If we don't have a new root, we can't set up a new sb */
1527 if (!opts
->new_root
)
1530 BUG_ON(!opts
->subsys_mask
&& !opts
->none
);
1532 ret
= set_anon_super(sb
, NULL
);
1536 sb
->s_fs_info
= opts
->new_root
;
1537 opts
->new_root
->sb
= sb
;
1539 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1540 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1541 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1542 sb
->s_op
= &cgroup_ops
;
1547 static int cgroup_get_rootdir(struct super_block
*sb
)
1549 static const struct dentry_operations cgroup_dops
= {
1550 .d_iput
= cgroup_diput
,
1551 .d_delete
= cgroup_delete
,
1554 struct inode
*inode
=
1555 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1560 inode
->i_fop
= &simple_dir_operations
;
1561 inode
->i_op
= &cgroup_dir_inode_operations
;
1562 /* directories start off with i_nlink == 2 (for "." entry) */
1564 sb
->s_root
= d_make_root(inode
);
1567 /* for everything else we want ->d_op set */
1568 sb
->s_d_op
= &cgroup_dops
;
1572 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1573 int flags
, const char *unused_dev_name
,
1576 struct cgroup_sb_opts opts
;
1577 struct cgroupfs_root
*root
;
1579 struct super_block
*sb
;
1580 struct cgroupfs_root
*new_root
;
1581 struct inode
*inode
;
1583 /* First find the desired set of subsystems */
1584 mutex_lock(&cgroup_mutex
);
1585 ret
= parse_cgroupfs_options(data
, &opts
);
1586 mutex_unlock(&cgroup_mutex
);
1591 * Allocate a new cgroup root. We may not need it if we're
1592 * reusing an existing hierarchy.
1594 new_root
= cgroup_root_from_opts(&opts
);
1595 if (IS_ERR(new_root
)) {
1596 ret
= PTR_ERR(new_root
);
1599 opts
.new_root
= new_root
;
1601 /* Locate an existing or new sb for this hierarchy */
1602 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, 0, &opts
);
1605 cgroup_free_root(opts
.new_root
);
1609 root
= sb
->s_fs_info
;
1611 if (root
== opts
.new_root
) {
1612 /* We used the new root structure, so this is a new hierarchy */
1613 struct list_head tmp_links
;
1614 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1615 struct cgroupfs_root
*existing_root
;
1616 const struct cred
*cred
;
1618 struct css_set
*cset
;
1620 BUG_ON(sb
->s_root
!= NULL
);
1622 ret
= cgroup_get_rootdir(sb
);
1624 goto drop_new_super
;
1625 inode
= sb
->s_root
->d_inode
;
1627 mutex_lock(&inode
->i_mutex
);
1628 mutex_lock(&cgroup_mutex
);
1629 mutex_lock(&cgroup_root_mutex
);
1631 /* Check for name clashes with existing mounts */
1633 if (strlen(root
->name
))
1634 for_each_active_root(existing_root
)
1635 if (!strcmp(existing_root
->name
, root
->name
))
1639 * We're accessing css_set_count without locking
1640 * css_set_lock here, but that's OK - it can only be
1641 * increased by someone holding cgroup_lock, and
1642 * that's us. The worst that can happen is that we
1643 * have some link structures left over
1645 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1649 /* ID 0 is reserved for dummy root, 1 for unified hierarchy */
1650 ret
= cgroup_init_root_id(root
, 2, 0);
1654 ret
= rebind_subsystems(root
, root
->subsys_mask
, 0);
1655 if (ret
== -EBUSY
) {
1656 free_cgrp_cset_links(&tmp_links
);
1660 * There must be no failure case after here, since rebinding
1661 * takes care of subsystems' refcounts, which are explicitly
1662 * dropped in the failure exit path.
1665 /* EBUSY should be the only error here */
1668 list_add(&root
->root_list
, &cgroup_roots
);
1669 cgroup_root_count
++;
1671 sb
->s_root
->d_fsdata
= root_cgrp
;
1672 root
->top_cgroup
.dentry
= sb
->s_root
;
1674 /* Link the top cgroup in this hierarchy into all
1675 * the css_set objects */
1676 write_lock(&css_set_lock
);
1677 hash_for_each(css_set_table
, i
, cset
, hlist
)
1678 link_css_set(&tmp_links
, cset
, root_cgrp
);
1679 write_unlock(&css_set_lock
);
1681 free_cgrp_cset_links(&tmp_links
);
1683 BUG_ON(!list_empty(&root_cgrp
->children
));
1684 BUG_ON(root
->number_of_cgroups
!= 1);
1686 cred
= override_creds(&init_cred
);
1687 cgroup_addrm_files(root_cgrp
, NULL
, cgroup_base_files
, true);
1688 cgroup_populate_dir(root_cgrp
, root
->subsys_mask
);
1690 mutex_unlock(&cgroup_root_mutex
);
1691 mutex_unlock(&cgroup_mutex
);
1692 mutex_unlock(&inode
->i_mutex
);
1695 * We re-used an existing hierarchy - the new root (if
1696 * any) is not needed
1698 cgroup_free_root(opts
.new_root
);
1700 if ((root
->flags
^ opts
.flags
) & CGRP_ROOT_OPTION_MASK
) {
1701 if ((root
->flags
| opts
.flags
) & CGRP_ROOT_SANE_BEHAVIOR
) {
1702 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1704 goto drop_new_super
;
1706 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1710 /* no subsys rebinding, so refcounts don't change */
1711 drop_parsed_module_refcounts(opts
.subsys_mask
);
1714 kfree(opts
.release_agent
);
1716 return dget(sb
->s_root
);
1719 cgroup_exit_root_id(root
);
1720 mutex_unlock(&cgroup_root_mutex
);
1721 mutex_unlock(&cgroup_mutex
);
1722 mutex_unlock(&inode
->i_mutex
);
1724 deactivate_locked_super(sb
);
1726 drop_parsed_module_refcounts(opts
.subsys_mask
);
1728 kfree(opts
.release_agent
);
1730 return ERR_PTR(ret
);
1733 static void cgroup_kill_sb(struct super_block
*sb
) {
1734 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1735 struct cgroup
*cgrp
= &root
->top_cgroup
;
1736 struct cgrp_cset_link
*link
, *tmp_link
;
1741 BUG_ON(root
->number_of_cgroups
!= 1);
1742 BUG_ON(!list_empty(&cgrp
->children
));
1744 mutex_lock(&cgroup_mutex
);
1745 mutex_lock(&cgroup_root_mutex
);
1747 /* Rebind all subsystems back to the default hierarchy */
1748 if (root
->flags
& CGRP_ROOT_SUBSYS_BOUND
) {
1749 ret
= rebind_subsystems(root
, 0, root
->subsys_mask
);
1750 /* Shouldn't be able to fail ... */
1755 * Release all the links from cset_links to this hierarchy's
1758 write_lock(&css_set_lock
);
1760 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1761 list_del(&link
->cset_link
);
1762 list_del(&link
->cgrp_link
);
1765 write_unlock(&css_set_lock
);
1767 if (!list_empty(&root
->root_list
)) {
1768 list_del(&root
->root_list
);
1769 cgroup_root_count
--;
1772 cgroup_exit_root_id(root
);
1774 mutex_unlock(&cgroup_root_mutex
);
1775 mutex_unlock(&cgroup_mutex
);
1777 simple_xattrs_free(&cgrp
->xattrs
);
1779 kill_litter_super(sb
);
1780 cgroup_free_root(root
);
1783 static struct file_system_type cgroup_fs_type
= {
1785 .mount
= cgroup_mount
,
1786 .kill_sb
= cgroup_kill_sb
,
1789 static struct kobject
*cgroup_kobj
;
1792 * cgroup_path - generate the path of a cgroup
1793 * @cgrp: the cgroup in question
1794 * @buf: the buffer to write the path into
1795 * @buflen: the length of the buffer
1797 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1799 * We can't generate cgroup path using dentry->d_name, as accessing
1800 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1801 * inode's i_mutex, while on the other hand cgroup_path() can be called
1802 * with some irq-safe spinlocks held.
1804 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1806 int ret
= -ENAMETOOLONG
;
1809 if (!cgrp
->parent
) {
1810 if (strlcpy(buf
, "/", buflen
) >= buflen
)
1811 return -ENAMETOOLONG
;
1815 start
= buf
+ buflen
- 1;
1820 const char *name
= cgroup_name(cgrp
);
1824 if ((start
-= len
) < buf
)
1826 memcpy(start
, name
, len
);
1832 cgrp
= cgrp
->parent
;
1833 } while (cgrp
->parent
);
1835 memmove(buf
, start
, buf
+ buflen
- start
);
1840 EXPORT_SYMBOL_GPL(cgroup_path
);
1843 * task_cgroup_path_from_hierarchy - cgroup path of a task on a hierarchy
1844 * @task: target task
1845 * @hierarchy_id: the hierarchy to look up @task's cgroup from
1846 * @buf: the buffer to write the path into
1847 * @buflen: the length of the buffer
1849 * Determine @task's cgroup on the hierarchy specified by @hierarchy_id and
1850 * copy its path into @buf. This function grabs cgroup_mutex and shouldn't
1851 * be used inside locks used by cgroup controller callbacks.
1853 int task_cgroup_path_from_hierarchy(struct task_struct
*task
, int hierarchy_id
,
1854 char *buf
, size_t buflen
)
1856 struct cgroupfs_root
*root
;
1857 struct cgroup
*cgrp
= NULL
;
1860 mutex_lock(&cgroup_mutex
);
1862 root
= idr_find(&cgroup_hierarchy_idr
, hierarchy_id
);
1864 cgrp
= task_cgroup_from_root(task
, root
);
1865 ret
= cgroup_path(cgrp
, buf
, buflen
);
1868 mutex_unlock(&cgroup_mutex
);
1872 EXPORT_SYMBOL_GPL(task_cgroup_path_from_hierarchy
);
1875 * Control Group taskset
1877 struct task_and_cgroup
{
1878 struct task_struct
*task
;
1879 struct cgroup
*cgrp
;
1883 struct cgroup_taskset
{
1884 struct task_and_cgroup single
;
1885 struct flex_array
*tc_array
;
1888 struct cgroup
*cur_cgrp
;
1892 * cgroup_taskset_first - reset taskset and return the first task
1893 * @tset: taskset of interest
1895 * @tset iteration is initialized and the first task is returned.
1897 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1899 if (tset
->tc_array
) {
1901 return cgroup_taskset_next(tset
);
1903 tset
->cur_cgrp
= tset
->single
.cgrp
;
1904 return tset
->single
.task
;
1907 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1910 * cgroup_taskset_next - iterate to the next task in taskset
1911 * @tset: taskset of interest
1913 * Return the next task in @tset. Iteration must have been initialized
1914 * with cgroup_taskset_first().
1916 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1918 struct task_and_cgroup
*tc
;
1920 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1923 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1924 tset
->cur_cgrp
= tc
->cgrp
;
1927 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1930 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1931 * @tset: taskset of interest
1933 * Return the cgroup for the current (last returned) task of @tset. This
1934 * function must be preceded by either cgroup_taskset_first() or
1935 * cgroup_taskset_next().
1937 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1939 return tset
->cur_cgrp
;
1941 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1944 * cgroup_taskset_size - return the number of tasks in taskset
1945 * @tset: taskset of interest
1947 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1949 return tset
->tc_array
? tset
->tc_array_len
: 1;
1951 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1955 * cgroup_task_migrate - move a task from one cgroup to another.
1957 * Must be called with cgroup_mutex and threadgroup locked.
1959 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
1960 struct task_struct
*tsk
,
1961 struct css_set
*new_cset
)
1963 struct css_set
*old_cset
;
1966 * We are synchronized through threadgroup_lock() against PF_EXITING
1967 * setting such that we can't race against cgroup_exit() changing the
1968 * css_set to init_css_set and dropping the old one.
1970 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1971 old_cset
= task_css_set(tsk
);
1974 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
1977 /* Update the css_set linked lists if we're using them */
1978 write_lock(&css_set_lock
);
1979 if (!list_empty(&tsk
->cg_list
))
1980 list_move(&tsk
->cg_list
, &new_cset
->tasks
);
1981 write_unlock(&css_set_lock
);
1984 * We just gained a reference on old_cset by taking it from the
1985 * task. As trading it for new_cset is protected by cgroup_mutex,
1986 * we're safe to drop it here; it will be freed under RCU.
1988 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
1989 put_css_set(old_cset
);
1993 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1994 * @cgrp: the cgroup to attach to
1995 * @tsk: the task or the leader of the threadgroup to be attached
1996 * @threadgroup: attach the whole threadgroup?
1998 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1999 * task_lock of @tsk or each thread in the threadgroup individually in turn.
2001 static int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
,
2004 int retval
, i
, group_size
;
2005 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
2006 struct cgroupfs_root
*root
= cgrp
->root
;
2007 /* threadgroup list cursor and array */
2008 struct task_struct
*leader
= tsk
;
2009 struct task_and_cgroup
*tc
;
2010 struct flex_array
*group
;
2011 struct cgroup_taskset tset
= { };
2014 * step 0: in order to do expensive, possibly blocking operations for
2015 * every thread, we cannot iterate the thread group list, since it needs
2016 * rcu or tasklist locked. instead, build an array of all threads in the
2017 * group - group_rwsem prevents new threads from appearing, and if
2018 * threads exit, this will just be an over-estimate.
2021 group_size
= get_nr_threads(tsk
);
2024 /* flex_array supports very large thread-groups better than kmalloc. */
2025 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2028 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2029 retval
= flex_array_prealloc(group
, 0, group_size
, GFP_KERNEL
);
2031 goto out_free_group_list
;
2035 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2036 * already PF_EXITING could be freed from underneath us unless we
2037 * take an rcu_read_lock.
2041 struct task_and_cgroup ent
;
2043 /* @tsk either already exited or can't exit until the end */
2044 if (tsk
->flags
& PF_EXITING
)
2047 /* as per above, nr_threads may decrease, but not increase. */
2048 BUG_ON(i
>= group_size
);
2050 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2051 /* nothing to do if this task is already in the cgroup */
2052 if (ent
.cgrp
== cgrp
)
2055 * saying GFP_ATOMIC has no effect here because we did prealloc
2056 * earlier, but it's good form to communicate our expectations.
2058 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2059 BUG_ON(retval
!= 0);
2064 } while_each_thread(leader
, tsk
);
2066 /* remember the number of threads in the array for later. */
2068 tset
.tc_array
= group
;
2069 tset
.tc_array_len
= group_size
;
2071 /* methods shouldn't be called if no task is actually migrating */
2074 goto out_free_group_list
;
2077 * step 1: check that we can legitimately attach to the cgroup.
2079 for_each_root_subsys(root
, ss
) {
2080 if (ss
->can_attach
) {
2081 retval
= ss
->can_attach(cgrp
, &tset
);
2084 goto out_cancel_attach
;
2090 * step 2: make sure css_sets exist for all threads to be migrated.
2091 * we use find_css_set, which allocates a new one if necessary.
2093 for (i
= 0; i
< group_size
; i
++) {
2094 struct css_set
*old_cset
;
2096 tc
= flex_array_get(group
, i
);
2097 old_cset
= task_css_set(tc
->task
);
2098 tc
->cg
= find_css_set(old_cset
, cgrp
);
2101 goto out_put_css_set_refs
;
2106 * step 3: now that we're guaranteed success wrt the css_sets,
2107 * proceed to move all tasks to the new cgroup. There are no
2108 * failure cases after here, so this is the commit point.
2110 for (i
= 0; i
< group_size
; i
++) {
2111 tc
= flex_array_get(group
, i
);
2112 cgroup_task_migrate(tc
->cgrp
, tc
->task
, tc
->cg
);
2114 /* nothing is sensitive to fork() after this point. */
2117 * step 4: do subsystem attach callbacks.
2119 for_each_root_subsys(root
, ss
) {
2121 ss
->attach(cgrp
, &tset
);
2125 * step 5: success! and cleanup
2128 out_put_css_set_refs
:
2130 for (i
= 0; i
< group_size
; i
++) {
2131 tc
= flex_array_get(group
, i
);
2134 put_css_set(tc
->cg
);
2139 for_each_root_subsys(root
, ss
) {
2140 if (ss
== failed_ss
)
2142 if (ss
->cancel_attach
)
2143 ss
->cancel_attach(cgrp
, &tset
);
2146 out_free_group_list
:
2147 flex_array_free(group
);
2152 * Find the task_struct of the task to attach by vpid and pass it along to the
2153 * function to attach either it or all tasks in its threadgroup. Will lock
2154 * cgroup_mutex and threadgroup; may take task_lock of task.
2156 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2158 struct task_struct
*tsk
;
2159 const struct cred
*cred
= current_cred(), *tcred
;
2162 if (!cgroup_lock_live_group(cgrp
))
2168 tsk
= find_task_by_vpid(pid
);
2172 goto out_unlock_cgroup
;
2175 * even if we're attaching all tasks in the thread group, we
2176 * only need to check permissions on one of them.
2178 tcred
= __task_cred(tsk
);
2179 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2180 !uid_eq(cred
->euid
, tcred
->uid
) &&
2181 !uid_eq(cred
->euid
, tcred
->suid
)) {
2184 goto out_unlock_cgroup
;
2190 tsk
= tsk
->group_leader
;
2193 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2194 * trapped in a cpuset, or RT worker may be born in a cgroup
2195 * with no rt_runtime allocated. Just say no.
2197 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2200 goto out_unlock_cgroup
;
2203 get_task_struct(tsk
);
2206 threadgroup_lock(tsk
);
2208 if (!thread_group_leader(tsk
)) {
2210 * a race with de_thread from another thread's exec()
2211 * may strip us of our leadership, if this happens,
2212 * there is no choice but to throw this task away and
2213 * try again; this is
2214 * "double-double-toil-and-trouble-check locking".
2216 threadgroup_unlock(tsk
);
2217 put_task_struct(tsk
);
2218 goto retry_find_task
;
2222 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2224 threadgroup_unlock(tsk
);
2226 put_task_struct(tsk
);
2228 mutex_unlock(&cgroup_mutex
);
2233 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2234 * @from: attach to all cgroups of a given task
2235 * @tsk: the task to be attached
2237 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2239 struct cgroupfs_root
*root
;
2242 mutex_lock(&cgroup_mutex
);
2243 for_each_active_root(root
) {
2244 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
2246 retval
= cgroup_attach_task(from_cg
, tsk
, false);
2250 mutex_unlock(&cgroup_mutex
);
2254 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2256 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2258 return attach_task_by_pid(cgrp
, pid
, false);
2261 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2263 return attach_task_by_pid(cgrp
, tgid
, true);
2266 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2269 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2270 if (strlen(buffer
) >= PATH_MAX
)
2272 if (!cgroup_lock_live_group(cgrp
))
2274 mutex_lock(&cgroup_root_mutex
);
2275 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2276 mutex_unlock(&cgroup_root_mutex
);
2277 mutex_unlock(&cgroup_mutex
);
2281 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2282 struct seq_file
*seq
)
2284 if (!cgroup_lock_live_group(cgrp
))
2286 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2287 seq_putc(seq
, '\n');
2288 mutex_unlock(&cgroup_mutex
);
2292 static int cgroup_sane_behavior_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2293 struct seq_file
*seq
)
2295 seq_printf(seq
, "%d\n", cgroup_sane_behavior(cgrp
));
2299 /* A buffer size big enough for numbers or short strings */
2300 #define CGROUP_LOCAL_BUFFER_SIZE 64
2302 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2304 const char __user
*userbuf
,
2305 size_t nbytes
, loff_t
*unused_ppos
)
2307 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2313 if (nbytes
>= sizeof(buffer
))
2315 if (copy_from_user(buffer
, userbuf
, nbytes
))
2318 buffer
[nbytes
] = 0; /* nul-terminate */
2319 if (cft
->write_u64
) {
2320 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2323 retval
= cft
->write_u64(cgrp
, cft
, val
);
2325 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2328 retval
= cft
->write_s64(cgrp
, cft
, val
);
2335 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2337 const char __user
*userbuf
,
2338 size_t nbytes
, loff_t
*unused_ppos
)
2340 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2342 size_t max_bytes
= cft
->max_write_len
;
2343 char *buffer
= local_buffer
;
2346 max_bytes
= sizeof(local_buffer
) - 1;
2347 if (nbytes
>= max_bytes
)
2349 /* Allocate a dynamic buffer if we need one */
2350 if (nbytes
>= sizeof(local_buffer
)) {
2351 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2355 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2360 buffer
[nbytes
] = 0; /* nul-terminate */
2361 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2365 if (buffer
!= local_buffer
)
2370 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2371 size_t nbytes
, loff_t
*ppos
)
2373 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2374 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2376 if (cgroup_is_dead(cgrp
))
2379 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2380 if (cft
->write_u64
|| cft
->write_s64
)
2381 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2382 if (cft
->write_string
)
2383 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2385 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2386 return ret
? ret
: nbytes
;
2391 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2393 char __user
*buf
, size_t nbytes
,
2396 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2397 u64 val
= cft
->read_u64(cgrp
, cft
);
2398 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2400 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2403 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2405 char __user
*buf
, size_t nbytes
,
2408 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2409 s64 val
= cft
->read_s64(cgrp
, cft
);
2410 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2412 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2415 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2416 size_t nbytes
, loff_t
*ppos
)
2418 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2419 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2421 if (cgroup_is_dead(cgrp
))
2425 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2427 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2429 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2434 * seqfile ops/methods for returning structured data. Currently just
2435 * supports string->u64 maps, but can be extended in future.
2438 struct cgroup_seqfile_state
{
2440 struct cgroup
*cgroup
;
2443 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2445 struct seq_file
*sf
= cb
->state
;
2446 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2449 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2451 struct cgroup_seqfile_state
*state
= m
->private;
2452 struct cftype
*cft
= state
->cft
;
2453 if (cft
->read_map
) {
2454 struct cgroup_map_cb cb
= {
2455 .fill
= cgroup_map_add
,
2458 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2460 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2463 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2465 struct seq_file
*seq
= file
->private_data
;
2466 kfree(seq
->private);
2467 return single_release(inode
, file
);
2470 static const struct file_operations cgroup_seqfile_operations
= {
2472 .write
= cgroup_file_write
,
2473 .llseek
= seq_lseek
,
2474 .release
= cgroup_seqfile_release
,
2477 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2482 err
= generic_file_open(inode
, file
);
2485 cft
= __d_cft(file
->f_dentry
);
2487 if (cft
->read_map
|| cft
->read_seq_string
) {
2488 struct cgroup_seqfile_state
*state
;
2490 state
= kzalloc(sizeof(*state
), GFP_USER
);
2495 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2496 file
->f_op
= &cgroup_seqfile_operations
;
2497 err
= single_open(file
, cgroup_seqfile_show
, state
);
2500 } else if (cft
->open
)
2501 err
= cft
->open(inode
, file
);
2508 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2510 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2512 return cft
->release(inode
, file
);
2517 * cgroup_rename - Only allow simple rename of directories in place.
2519 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2520 struct inode
*new_dir
, struct dentry
*new_dentry
)
2523 struct cgroup_name
*name
, *old_name
;
2524 struct cgroup
*cgrp
;
2527 * It's convinient to use parent dir's i_mutex to protected
2530 lockdep_assert_held(&old_dir
->i_mutex
);
2532 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2534 if (new_dentry
->d_inode
)
2536 if (old_dir
!= new_dir
)
2539 cgrp
= __d_cgrp(old_dentry
);
2542 * This isn't a proper migration and its usefulness is very
2543 * limited. Disallow if sane_behavior.
2545 if (cgroup_sane_behavior(cgrp
))
2548 name
= cgroup_alloc_name(new_dentry
);
2552 ret
= simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2558 old_name
= rcu_dereference_protected(cgrp
->name
, true);
2559 rcu_assign_pointer(cgrp
->name
, name
);
2561 kfree_rcu(old_name
, rcu_head
);
2565 static struct simple_xattrs
*__d_xattrs(struct dentry
*dentry
)
2567 if (S_ISDIR(dentry
->d_inode
->i_mode
))
2568 return &__d_cgrp(dentry
)->xattrs
;
2570 return &__d_cfe(dentry
)->xattrs
;
2573 static inline int xattr_enabled(struct dentry
*dentry
)
2575 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
2576 return root
->flags
& CGRP_ROOT_XATTR
;
2579 static bool is_valid_xattr(const char *name
)
2581 if (!strncmp(name
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) ||
2582 !strncmp(name
, XATTR_SECURITY_PREFIX
, XATTR_SECURITY_PREFIX_LEN
))
2587 static int cgroup_setxattr(struct dentry
*dentry
, const char *name
,
2588 const void *val
, size_t size
, int flags
)
2590 if (!xattr_enabled(dentry
))
2592 if (!is_valid_xattr(name
))
2594 return simple_xattr_set(__d_xattrs(dentry
), name
, val
, size
, flags
);
2597 static int cgroup_removexattr(struct dentry
*dentry
, const char *name
)
2599 if (!xattr_enabled(dentry
))
2601 if (!is_valid_xattr(name
))
2603 return simple_xattr_remove(__d_xattrs(dentry
), name
);
2606 static ssize_t
cgroup_getxattr(struct dentry
*dentry
, const char *name
,
2607 void *buf
, size_t size
)
2609 if (!xattr_enabled(dentry
))
2611 if (!is_valid_xattr(name
))
2613 return simple_xattr_get(__d_xattrs(dentry
), name
, buf
, size
);
2616 static ssize_t
cgroup_listxattr(struct dentry
*dentry
, char *buf
, size_t size
)
2618 if (!xattr_enabled(dentry
))
2620 return simple_xattr_list(__d_xattrs(dentry
), buf
, size
);
2623 static const struct file_operations cgroup_file_operations
= {
2624 .read
= cgroup_file_read
,
2625 .write
= cgroup_file_write
,
2626 .llseek
= generic_file_llseek
,
2627 .open
= cgroup_file_open
,
2628 .release
= cgroup_file_release
,
2631 static const struct inode_operations cgroup_file_inode_operations
= {
2632 .setxattr
= cgroup_setxattr
,
2633 .getxattr
= cgroup_getxattr
,
2634 .listxattr
= cgroup_listxattr
,
2635 .removexattr
= cgroup_removexattr
,
2638 static const struct inode_operations cgroup_dir_inode_operations
= {
2639 .lookup
= cgroup_lookup
,
2640 .mkdir
= cgroup_mkdir
,
2641 .rmdir
= cgroup_rmdir
,
2642 .rename
= cgroup_rename
,
2643 .setxattr
= cgroup_setxattr
,
2644 .getxattr
= cgroup_getxattr
,
2645 .listxattr
= cgroup_listxattr
,
2646 .removexattr
= cgroup_removexattr
,
2649 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
)
2651 if (dentry
->d_name
.len
> NAME_MAX
)
2652 return ERR_PTR(-ENAMETOOLONG
);
2653 d_add(dentry
, NULL
);
2658 * Check if a file is a control file
2660 static inline struct cftype
*__file_cft(struct file
*file
)
2662 if (file_inode(file
)->i_fop
!= &cgroup_file_operations
)
2663 return ERR_PTR(-EINVAL
);
2664 return __d_cft(file
->f_dentry
);
2667 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2668 struct super_block
*sb
)
2670 struct inode
*inode
;
2674 if (dentry
->d_inode
)
2677 inode
= cgroup_new_inode(mode
, sb
);
2681 if (S_ISDIR(mode
)) {
2682 inode
->i_op
= &cgroup_dir_inode_operations
;
2683 inode
->i_fop
= &simple_dir_operations
;
2685 /* start off with i_nlink == 2 (for "." entry) */
2687 inc_nlink(dentry
->d_parent
->d_inode
);
2690 * Control reaches here with cgroup_mutex held.
2691 * @inode->i_mutex should nest outside cgroup_mutex but we
2692 * want to populate it immediately without releasing
2693 * cgroup_mutex. As @inode isn't visible to anyone else
2694 * yet, trylock will always succeed without affecting
2697 WARN_ON_ONCE(!mutex_trylock(&inode
->i_mutex
));
2698 } else if (S_ISREG(mode
)) {
2700 inode
->i_fop
= &cgroup_file_operations
;
2701 inode
->i_op
= &cgroup_file_inode_operations
;
2703 d_instantiate(dentry
, inode
);
2704 dget(dentry
); /* Extra count - pin the dentry in core */
2709 * cgroup_file_mode - deduce file mode of a control file
2710 * @cft: the control file in question
2712 * returns cft->mode if ->mode is not 0
2713 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2714 * returns S_IRUGO if it has only a read handler
2715 * returns S_IWUSR if it has only a write hander
2717 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2724 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2725 cft
->read_map
|| cft
->read_seq_string
)
2728 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2729 cft
->write_string
|| cft
->trigger
)
2735 static int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2738 struct dentry
*dir
= cgrp
->dentry
;
2739 struct cgroup
*parent
= __d_cgrp(dir
);
2740 struct dentry
*dentry
;
2744 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2746 if (subsys
&& !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
)) {
2747 strcpy(name
, subsys
->name
);
2750 strcat(name
, cft
->name
);
2752 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2754 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2758 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2759 if (IS_ERR(dentry
)) {
2760 error
= PTR_ERR(dentry
);
2764 cfe
->type
= (void *)cft
;
2765 cfe
->dentry
= dentry
;
2766 dentry
->d_fsdata
= cfe
;
2767 simple_xattrs_init(&cfe
->xattrs
);
2769 mode
= cgroup_file_mode(cft
);
2770 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2772 list_add_tail(&cfe
->node
, &parent
->files
);
2782 * cgroup_addrm_files - add or remove files to a cgroup directory
2783 * @cgrp: the target cgroup
2784 * @subsys: the subsystem of files to be added
2785 * @cfts: array of cftypes to be added
2786 * @is_add: whether to add or remove
2788 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2789 * All @cfts should belong to @subsys. For removals, this function never
2790 * fails. If addition fails, this function doesn't remove files already
2791 * added. The caller is responsible for cleaning up.
2793 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2794 struct cftype cfts
[], bool is_add
)
2799 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
2800 lockdep_assert_held(&cgroup_mutex
);
2802 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2803 /* does cft->flags tell us to skip this file on @cgrp? */
2804 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_sane_behavior(cgrp
))
2806 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2808 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2812 ret
= cgroup_add_file(cgrp
, subsys
, cft
);
2814 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2819 cgroup_rm_file(cgrp
, cft
);
2825 static void cgroup_cfts_prepare(void)
2826 __acquires(&cgroup_mutex
)
2829 * Thanks to the entanglement with vfs inode locking, we can't walk
2830 * the existing cgroups under cgroup_mutex and create files.
2831 * Instead, we use cgroup_for_each_descendant_pre() and drop RCU
2832 * read lock before calling cgroup_addrm_files().
2834 mutex_lock(&cgroup_mutex
);
2837 static int cgroup_cfts_commit(struct cgroup_subsys
*ss
,
2838 struct cftype
*cfts
, bool is_add
)
2839 __releases(&cgroup_mutex
)
2842 struct cgroup
*cgrp
, *root
= &ss
->root
->top_cgroup
;
2843 struct super_block
*sb
= ss
->root
->sb
;
2844 struct dentry
*prev
= NULL
;
2845 struct inode
*inode
;
2849 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2850 if (!cfts
|| ss
->root
== &cgroup_dummy_root
||
2851 !atomic_inc_not_zero(&sb
->s_active
)) {
2852 mutex_unlock(&cgroup_mutex
);
2857 * All cgroups which are created after we drop cgroup_mutex will
2858 * have the updated set of files, so we only need to update the
2859 * cgroups created before the current @cgroup_serial_nr_next.
2861 update_before
= cgroup_serial_nr_next
;
2863 mutex_unlock(&cgroup_mutex
);
2865 /* @root always needs to be updated */
2866 inode
= root
->dentry
->d_inode
;
2867 mutex_lock(&inode
->i_mutex
);
2868 mutex_lock(&cgroup_mutex
);
2869 ret
= cgroup_addrm_files(root
, ss
, cfts
, is_add
);
2870 mutex_unlock(&cgroup_mutex
);
2871 mutex_unlock(&inode
->i_mutex
);
2876 /* add/rm files for all cgroups created before */
2878 cgroup_for_each_descendant_pre(cgrp
, root
) {
2879 if (cgroup_is_dead(cgrp
))
2882 inode
= cgrp
->dentry
->d_inode
;
2887 prev
= cgrp
->dentry
;
2889 mutex_lock(&inode
->i_mutex
);
2890 mutex_lock(&cgroup_mutex
);
2891 if (cgrp
->serial_nr
< update_before
&& !cgroup_is_dead(cgrp
))
2892 ret
= cgroup_addrm_files(cgrp
, ss
, cfts
, is_add
);
2893 mutex_unlock(&cgroup_mutex
);
2894 mutex_unlock(&inode
->i_mutex
);
2903 deactivate_super(sb
);
2908 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2909 * @ss: target cgroup subsystem
2910 * @cfts: zero-length name terminated array of cftypes
2912 * Register @cfts to @ss. Files described by @cfts are created for all
2913 * existing cgroups to which @ss is attached and all future cgroups will
2914 * have them too. This function can be called anytime whether @ss is
2917 * Returns 0 on successful registration, -errno on failure. Note that this
2918 * function currently returns 0 as long as @cfts registration is successful
2919 * even if some file creation attempts on existing cgroups fail.
2921 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2923 struct cftype_set
*set
;
2926 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2930 cgroup_cfts_prepare();
2932 list_add_tail(&set
->node
, &ss
->cftsets
);
2933 ret
= cgroup_cfts_commit(ss
, cfts
, true);
2935 cgroup_rm_cftypes(ss
, cfts
);
2938 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2941 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2942 * @ss: target cgroup subsystem
2943 * @cfts: zero-length name terminated array of cftypes
2945 * Unregister @cfts from @ss. Files described by @cfts are removed from
2946 * all existing cgroups to which @ss is attached and all future cgroups
2947 * won't have them either. This function can be called anytime whether @ss
2948 * is attached or not.
2950 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2951 * registered with @ss.
2953 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2955 struct cftype_set
*set
;
2957 cgroup_cfts_prepare();
2959 list_for_each_entry(set
, &ss
->cftsets
, node
) {
2960 if (set
->cfts
== cfts
) {
2961 list_del(&set
->node
);
2963 cgroup_cfts_commit(ss
, cfts
, false);
2968 cgroup_cfts_commit(ss
, NULL
, false);
2973 * cgroup_task_count - count the number of tasks in a cgroup.
2974 * @cgrp: the cgroup in question
2976 * Return the number of tasks in the cgroup.
2978 int cgroup_task_count(const struct cgroup
*cgrp
)
2981 struct cgrp_cset_link
*link
;
2983 read_lock(&css_set_lock
);
2984 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
2985 count
+= atomic_read(&link
->cset
->refcount
);
2986 read_unlock(&css_set_lock
);
2991 * Advance a list_head iterator. The iterator should be positioned at
2992 * the start of a css_set
2994 static void cgroup_advance_iter(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2996 struct list_head
*l
= it
->cset_link
;
2997 struct cgrp_cset_link
*link
;
2998 struct css_set
*cset
;
3000 /* Advance to the next non-empty css_set */
3003 if (l
== &cgrp
->cset_links
) {
3004 it
->cset_link
= NULL
;
3007 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3009 } while (list_empty(&cset
->tasks
));
3011 it
->task
= cset
->tasks
.next
;
3015 * To reduce the fork() overhead for systems that are not actually
3016 * using their cgroups capability, we don't maintain the lists running
3017 * through each css_set to its tasks until we see the list actually
3018 * used - in other words after the first call to cgroup_iter_start().
3020 static void cgroup_enable_task_cg_lists(void)
3022 struct task_struct
*p
, *g
;
3023 write_lock(&css_set_lock
);
3024 use_task_css_set_links
= 1;
3026 * We need tasklist_lock because RCU is not safe against
3027 * while_each_thread(). Besides, a forking task that has passed
3028 * cgroup_post_fork() without seeing use_task_css_set_links = 1
3029 * is not guaranteed to have its child immediately visible in the
3030 * tasklist if we walk through it with RCU.
3032 read_lock(&tasklist_lock
);
3033 do_each_thread(g
, p
) {
3036 * We should check if the process is exiting, otherwise
3037 * it will race with cgroup_exit() in that the list
3038 * entry won't be deleted though the process has exited.
3040 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
3041 list_add(&p
->cg_list
, &task_css_set(p
)->tasks
);
3043 } while_each_thread(g
, p
);
3044 read_unlock(&tasklist_lock
);
3045 write_unlock(&css_set_lock
);
3049 * cgroup_next_sibling - find the next sibling of a given cgroup
3050 * @pos: the current cgroup
3052 * This function returns the next sibling of @pos and should be called
3053 * under RCU read lock. The only requirement is that @pos is accessible.
3054 * The next sibling is guaranteed to be returned regardless of @pos's
3057 struct cgroup
*cgroup_next_sibling(struct cgroup
*pos
)
3059 struct cgroup
*next
;
3061 WARN_ON_ONCE(!rcu_read_lock_held());
3064 * @pos could already have been removed. Once a cgroup is removed,
3065 * its ->sibling.next is no longer updated when its next sibling
3066 * changes. As CGRP_DEAD assertion is serialized and happens
3067 * before the cgroup is taken off the ->sibling list, if we see it
3068 * unasserted, it's guaranteed that the next sibling hasn't
3069 * finished its grace period even if it's already removed, and thus
3070 * safe to dereference from this RCU critical section. If
3071 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
3072 * to be visible as %true here.
3074 if (likely(!cgroup_is_dead(pos
))) {
3075 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
, sibling
);
3076 if (&next
->sibling
!= &pos
->parent
->children
)
3082 * Can't dereference the next pointer. Each cgroup is given a
3083 * monotonically increasing unique serial number and always
3084 * appended to the sibling list, so the next one can be found by
3085 * walking the parent's children until we see a cgroup with higher
3086 * serial number than @pos's.
3088 * While this path can be slow, it's taken only when either the
3089 * current cgroup is removed or iteration and removal race.
3091 list_for_each_entry_rcu(next
, &pos
->parent
->children
, sibling
)
3092 if (next
->serial_nr
> pos
->serial_nr
)
3096 EXPORT_SYMBOL_GPL(cgroup_next_sibling
);
3099 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
3100 * @pos: the current position (%NULL to initiate traversal)
3101 * @cgroup: cgroup whose descendants to walk
3103 * To be used by cgroup_for_each_descendant_pre(). Find the next
3104 * descendant to visit for pre-order traversal of @cgroup's descendants.
3106 * While this function requires RCU read locking, it doesn't require the
3107 * whole traversal to be contained in a single RCU critical section. This
3108 * function will return the correct next descendant as long as both @pos
3109 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3111 struct cgroup
*cgroup_next_descendant_pre(struct cgroup
*pos
,
3112 struct cgroup
*cgroup
)
3114 struct cgroup
*next
;
3116 WARN_ON_ONCE(!rcu_read_lock_held());
3118 /* if first iteration, pretend we just visited @cgroup */
3122 /* visit the first child if exists */
3123 next
= list_first_or_null_rcu(&pos
->children
, struct cgroup
, sibling
);
3127 /* no child, visit my or the closest ancestor's next sibling */
3128 while (pos
!= cgroup
) {
3129 next
= cgroup_next_sibling(pos
);
3137 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre
);
3140 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
3141 * @pos: cgroup of interest
3143 * Return the rightmost descendant of @pos. If there's no descendant,
3144 * @pos is returned. This can be used during pre-order traversal to skip
3147 * While this function requires RCU read locking, it doesn't require the
3148 * whole traversal to be contained in a single RCU critical section. This
3149 * function will return the correct rightmost descendant as long as @pos is
3152 struct cgroup
*cgroup_rightmost_descendant(struct cgroup
*pos
)
3154 struct cgroup
*last
, *tmp
;
3156 WARN_ON_ONCE(!rcu_read_lock_held());
3160 /* ->prev isn't RCU safe, walk ->next till the end */
3162 list_for_each_entry_rcu(tmp
, &last
->children
, sibling
)
3168 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant
);
3170 static struct cgroup
*cgroup_leftmost_descendant(struct cgroup
*pos
)
3172 struct cgroup
*last
;
3176 pos
= list_first_or_null_rcu(&pos
->children
, struct cgroup
,
3184 * cgroup_next_descendant_post - find the next descendant for post-order walk
3185 * @pos: the current position (%NULL to initiate traversal)
3186 * @cgroup: cgroup whose descendants to walk
3188 * To be used by cgroup_for_each_descendant_post(). Find the next
3189 * descendant to visit for post-order traversal of @cgroup's descendants.
3191 * While this function requires RCU read locking, it doesn't require the
3192 * whole traversal to be contained in a single RCU critical section. This
3193 * function will return the correct next descendant as long as both @pos
3194 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3196 struct cgroup
*cgroup_next_descendant_post(struct cgroup
*pos
,
3197 struct cgroup
*cgroup
)
3199 struct cgroup
*next
;
3201 WARN_ON_ONCE(!rcu_read_lock_held());
3203 /* if first iteration, visit the leftmost descendant */
3205 next
= cgroup_leftmost_descendant(cgroup
);
3206 return next
!= cgroup
? next
: NULL
;
3209 /* if there's an unvisited sibling, visit its leftmost descendant */
3210 next
= cgroup_next_sibling(pos
);
3212 return cgroup_leftmost_descendant(next
);
3214 /* no sibling left, visit parent */
3216 return next
!= cgroup
? next
: NULL
;
3218 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post
);
3220 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3221 __acquires(css_set_lock
)
3224 * The first time anyone tries to iterate across a cgroup,
3225 * we need to enable the list linking each css_set to its
3226 * tasks, and fix up all existing tasks.
3228 if (!use_task_css_set_links
)
3229 cgroup_enable_task_cg_lists();
3231 read_lock(&css_set_lock
);
3232 it
->cset_link
= &cgrp
->cset_links
;
3233 cgroup_advance_iter(cgrp
, it
);
3236 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
3237 struct cgroup_iter
*it
)
3239 struct task_struct
*res
;
3240 struct list_head
*l
= it
->task
;
3241 struct cgrp_cset_link
*link
;
3243 /* If the iterator cg is NULL, we have no tasks */
3246 res
= list_entry(l
, struct task_struct
, cg_list
);
3247 /* Advance iterator to find next entry */
3249 link
= list_entry(it
->cset_link
, struct cgrp_cset_link
, cset_link
);
3250 if (l
== &link
->cset
->tasks
) {
3251 /* We reached the end of this task list - move on to
3252 * the next cg_cgroup_link */
3253 cgroup_advance_iter(cgrp
, it
);
3260 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
3261 __releases(css_set_lock
)
3263 read_unlock(&css_set_lock
);
3266 static inline int started_after_time(struct task_struct
*t1
,
3267 struct timespec
*time
,
3268 struct task_struct
*t2
)
3270 int start_diff
= timespec_compare(&t1
->start_time
, time
);
3271 if (start_diff
> 0) {
3273 } else if (start_diff
< 0) {
3277 * Arbitrarily, if two processes started at the same
3278 * time, we'll say that the lower pointer value
3279 * started first. Note that t2 may have exited by now
3280 * so this may not be a valid pointer any longer, but
3281 * that's fine - it still serves to distinguish
3282 * between two tasks started (effectively) simultaneously.
3289 * This function is a callback from heap_insert() and is used to order
3291 * In this case we order the heap in descending task start time.
3293 static inline int started_after(void *p1
, void *p2
)
3295 struct task_struct
*t1
= p1
;
3296 struct task_struct
*t2
= p2
;
3297 return started_after_time(t1
, &t2
->start_time
, t2
);
3301 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3302 * @scan: struct cgroup_scanner containing arguments for the scan
3304 * Arguments include pointers to callback functions test_task() and
3306 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3307 * and if it returns true, call process_task() for it also.
3308 * The test_task pointer may be NULL, meaning always true (select all tasks).
3309 * Effectively duplicates cgroup_iter_{start,next,end}()
3310 * but does not lock css_set_lock for the call to process_task().
3311 * The struct cgroup_scanner may be embedded in any structure of the caller's
3313 * It is guaranteed that process_task() will act on every task that
3314 * is a member of the cgroup for the duration of this call. This
3315 * function may or may not call process_task() for tasks that exit
3316 * or move to a different cgroup during the call, or are forked or
3317 * move into the cgroup during the call.
3319 * Note that test_task() may be called with locks held, and may in some
3320 * situations be called multiple times for the same task, so it should
3322 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3323 * pre-allocated and will be used for heap operations (and its "gt" member will
3324 * be overwritten), else a temporary heap will be used (allocation of which
3325 * may cause this function to fail).
3327 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
3330 struct cgroup_iter it
;
3331 struct task_struct
*p
, *dropped
;
3332 /* Never dereference latest_task, since it's not refcounted */
3333 struct task_struct
*latest_task
= NULL
;
3334 struct ptr_heap tmp_heap
;
3335 struct ptr_heap
*heap
;
3336 struct timespec latest_time
= { 0, 0 };
3339 /* The caller supplied our heap and pre-allocated its memory */
3341 heap
->gt
= &started_after
;
3343 /* We need to allocate our own heap memory */
3345 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3347 /* cannot allocate the heap */
3353 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3354 * to determine which are of interest, and using the scanner's
3355 * "process_task" callback to process any of them that need an update.
3356 * Since we don't want to hold any locks during the task updates,
3357 * gather tasks to be processed in a heap structure.
3358 * The heap is sorted by descending task start time.
3359 * If the statically-sized heap fills up, we overflow tasks that
3360 * started later, and in future iterations only consider tasks that
3361 * started after the latest task in the previous pass. This
3362 * guarantees forward progress and that we don't miss any tasks.
3365 cgroup_iter_start(scan
->cg
, &it
);
3366 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
3368 * Only affect tasks that qualify per the caller's callback,
3369 * if he provided one
3371 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
3374 * Only process tasks that started after the last task
3377 if (!started_after_time(p
, &latest_time
, latest_task
))
3379 dropped
= heap_insert(heap
, p
);
3380 if (dropped
== NULL
) {
3382 * The new task was inserted; the heap wasn't
3386 } else if (dropped
!= p
) {
3388 * The new task was inserted, and pushed out a
3392 put_task_struct(dropped
);
3395 * Else the new task was newer than anything already in
3396 * the heap and wasn't inserted
3399 cgroup_iter_end(scan
->cg
, &it
);
3402 for (i
= 0; i
< heap
->size
; i
++) {
3403 struct task_struct
*q
= heap
->ptrs
[i
];
3405 latest_time
= q
->start_time
;
3408 /* Process the task per the caller's callback */
3409 scan
->process_task(q
, scan
);
3413 * If we had to process any tasks at all, scan again
3414 * in case some of them were in the middle of forking
3415 * children that didn't get processed.
3416 * Not the most efficient way to do it, but it avoids
3417 * having to take callback_mutex in the fork path
3421 if (heap
== &tmp_heap
)
3422 heap_free(&tmp_heap
);
3426 static void cgroup_transfer_one_task(struct task_struct
*task
,
3427 struct cgroup_scanner
*scan
)
3429 struct cgroup
*new_cgroup
= scan
->data
;
3431 mutex_lock(&cgroup_mutex
);
3432 cgroup_attach_task(new_cgroup
, task
, false);
3433 mutex_unlock(&cgroup_mutex
);
3437 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3438 * @to: cgroup to which the tasks will be moved
3439 * @from: cgroup in which the tasks currently reside
3441 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3443 struct cgroup_scanner scan
;
3446 scan
.test_task
= NULL
; /* select all tasks in cgroup */
3447 scan
.process_task
= cgroup_transfer_one_task
;
3451 return cgroup_scan_tasks(&scan
);
3455 * Stuff for reading the 'tasks'/'procs' files.
3457 * Reading this file can return large amounts of data if a cgroup has
3458 * *lots* of attached tasks. So it may need several calls to read(),
3459 * but we cannot guarantee that the information we produce is correct
3460 * unless we produce it entirely atomically.
3464 /* which pidlist file are we talking about? */
3465 enum cgroup_filetype
{
3471 * A pidlist is a list of pids that virtually represents the contents of one
3472 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3473 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3476 struct cgroup_pidlist
{
3478 * used to find which pidlist is wanted. doesn't change as long as
3479 * this particular list stays in the list.
3481 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3484 /* how many elements the above list has */
3486 /* how many files are using the current array */
3488 /* each of these stored in a list by its cgroup */
3489 struct list_head links
;
3490 /* pointer to the cgroup we belong to, for list removal purposes */
3491 struct cgroup
*owner
;
3492 /* protects the other fields */
3493 struct rw_semaphore mutex
;
3497 * The following two functions "fix" the issue where there are more pids
3498 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3499 * TODO: replace with a kernel-wide solution to this problem
3501 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3502 static void *pidlist_allocate(int count
)
3504 if (PIDLIST_TOO_LARGE(count
))
3505 return vmalloc(count
* sizeof(pid_t
));
3507 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3509 static void pidlist_free(void *p
)
3511 if (is_vmalloc_addr(p
))
3518 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3519 * Returns the number of unique elements.
3521 static int pidlist_uniq(pid_t
*list
, int length
)
3526 * we presume the 0th element is unique, so i starts at 1. trivial
3527 * edge cases first; no work needs to be done for either
3529 if (length
== 0 || length
== 1)
3531 /* src and dest walk down the list; dest counts unique elements */
3532 for (src
= 1; src
< length
; src
++) {
3533 /* find next unique element */
3534 while (list
[src
] == list
[src
-1]) {
3539 /* dest always points to where the next unique element goes */
3540 list
[dest
] = list
[src
];
3547 static int cmppid(const void *a
, const void *b
)
3549 return *(pid_t
*)a
- *(pid_t
*)b
;
3553 * find the appropriate pidlist for our purpose (given procs vs tasks)
3554 * returns with the lock on that pidlist already held, and takes care
3555 * of the use count, or returns NULL with no locks held if we're out of
3558 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3559 enum cgroup_filetype type
)
3561 struct cgroup_pidlist
*l
;
3562 /* don't need task_nsproxy() if we're looking at ourself */
3563 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3566 * We can't drop the pidlist_mutex before taking the l->mutex in case
3567 * the last ref-holder is trying to remove l from the list at the same
3568 * time. Holding the pidlist_mutex precludes somebody taking whichever
3569 * list we find out from under us - compare release_pid_array().
3571 mutex_lock(&cgrp
->pidlist_mutex
);
3572 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3573 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3574 /* make sure l doesn't vanish out from under us */
3575 down_write(&l
->mutex
);
3576 mutex_unlock(&cgrp
->pidlist_mutex
);
3580 /* entry not found; create a new one */
3581 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3583 mutex_unlock(&cgrp
->pidlist_mutex
);
3586 init_rwsem(&l
->mutex
);
3587 down_write(&l
->mutex
);
3589 l
->key
.ns
= get_pid_ns(ns
);
3591 list_add(&l
->links
, &cgrp
->pidlists
);
3592 mutex_unlock(&cgrp
->pidlist_mutex
);
3597 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3599 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3600 struct cgroup_pidlist
**lp
)
3604 int pid
, n
= 0; /* used for populating the array */
3605 struct cgroup_iter it
;
3606 struct task_struct
*tsk
;
3607 struct cgroup_pidlist
*l
;
3610 * If cgroup gets more users after we read count, we won't have
3611 * enough space - tough. This race is indistinguishable to the
3612 * caller from the case that the additional cgroup users didn't
3613 * show up until sometime later on.
3615 length
= cgroup_task_count(cgrp
);
3616 array
= pidlist_allocate(length
);
3619 /* now, populate the array */
3620 cgroup_iter_start(cgrp
, &it
);
3621 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3622 if (unlikely(n
== length
))
3624 /* get tgid or pid for procs or tasks file respectively */
3625 if (type
== CGROUP_FILE_PROCS
)
3626 pid
= task_tgid_vnr(tsk
);
3628 pid
= task_pid_vnr(tsk
);
3629 if (pid
> 0) /* make sure to only use valid results */
3632 cgroup_iter_end(cgrp
, &it
);
3634 /* now sort & (if procs) strip out duplicates */
3635 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3636 if (type
== CGROUP_FILE_PROCS
)
3637 length
= pidlist_uniq(array
, length
);
3638 l
= cgroup_pidlist_find(cgrp
, type
);
3640 pidlist_free(array
);
3643 /* store array, freeing old if necessary - lock already held */
3644 pidlist_free(l
->list
);
3648 up_write(&l
->mutex
);
3654 * cgroupstats_build - build and fill cgroupstats
3655 * @stats: cgroupstats to fill information into
3656 * @dentry: A dentry entry belonging to the cgroup for which stats have
3659 * Build and fill cgroupstats so that taskstats can export it to user
3662 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3665 struct cgroup
*cgrp
;
3666 struct cgroup_iter it
;
3667 struct task_struct
*tsk
;
3670 * Validate dentry by checking the superblock operations,
3671 * and make sure it's a directory.
3673 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3674 !S_ISDIR(dentry
->d_inode
->i_mode
))
3678 cgrp
= dentry
->d_fsdata
;
3680 cgroup_iter_start(cgrp
, &it
);
3681 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3682 switch (tsk
->state
) {
3684 stats
->nr_running
++;
3686 case TASK_INTERRUPTIBLE
:
3687 stats
->nr_sleeping
++;
3689 case TASK_UNINTERRUPTIBLE
:
3690 stats
->nr_uninterruptible
++;
3693 stats
->nr_stopped
++;
3696 if (delayacct_is_task_waiting_on_io(tsk
))
3697 stats
->nr_io_wait
++;
3701 cgroup_iter_end(cgrp
, &it
);
3709 * seq_file methods for the tasks/procs files. The seq_file position is the
3710 * next pid to display; the seq_file iterator is a pointer to the pid
3711 * in the cgroup->l->list array.
3714 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3717 * Initially we receive a position value that corresponds to
3718 * one more than the last pid shown (or 0 on the first call or
3719 * after a seek to the start). Use a binary-search to find the
3720 * next pid to display, if any
3722 struct cgroup_pidlist
*l
= s
->private;
3723 int index
= 0, pid
= *pos
;
3726 down_read(&l
->mutex
);
3728 int end
= l
->length
;
3730 while (index
< end
) {
3731 int mid
= (index
+ end
) / 2;
3732 if (l
->list
[mid
] == pid
) {
3735 } else if (l
->list
[mid
] <= pid
)
3741 /* If we're off the end of the array, we're done */
3742 if (index
>= l
->length
)
3744 /* Update the abstract position to be the actual pid that we found */
3745 iter
= l
->list
+ index
;
3750 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3752 struct cgroup_pidlist
*l
= s
->private;
3756 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3758 struct cgroup_pidlist
*l
= s
->private;
3760 pid_t
*end
= l
->list
+ l
->length
;
3762 * Advance to the next pid in the array. If this goes off the
3774 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3776 return seq_printf(s
, "%d\n", *(int *)v
);
3780 * seq_operations functions for iterating on pidlists through seq_file -
3781 * independent of whether it's tasks or procs
3783 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3784 .start
= cgroup_pidlist_start
,
3785 .stop
= cgroup_pidlist_stop
,
3786 .next
= cgroup_pidlist_next
,
3787 .show
= cgroup_pidlist_show
,
3790 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3793 * the case where we're the last user of this particular pidlist will
3794 * have us remove it from the cgroup's list, which entails taking the
3795 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3796 * pidlist_mutex, we have to take pidlist_mutex first.
3798 mutex_lock(&l
->owner
->pidlist_mutex
);
3799 down_write(&l
->mutex
);
3800 BUG_ON(!l
->use_count
);
3801 if (!--l
->use_count
) {
3802 /* we're the last user if refcount is 0; remove and free */
3803 list_del(&l
->links
);
3804 mutex_unlock(&l
->owner
->pidlist_mutex
);
3805 pidlist_free(l
->list
);
3806 put_pid_ns(l
->key
.ns
);
3807 up_write(&l
->mutex
);
3811 mutex_unlock(&l
->owner
->pidlist_mutex
);
3812 up_write(&l
->mutex
);
3815 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3817 struct cgroup_pidlist
*l
;
3818 if (!(file
->f_mode
& FMODE_READ
))
3821 * the seq_file will only be initialized if the file was opened for
3822 * reading; hence we check if it's not null only in that case.
3824 l
= ((struct seq_file
*)file
->private_data
)->private;
3825 cgroup_release_pid_array(l
);
3826 return seq_release(inode
, file
);
3829 static const struct file_operations cgroup_pidlist_operations
= {
3831 .llseek
= seq_lseek
,
3832 .write
= cgroup_file_write
,
3833 .release
= cgroup_pidlist_release
,
3837 * The following functions handle opens on a file that displays a pidlist
3838 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3841 /* helper function for the two below it */
3842 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3844 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3845 struct cgroup_pidlist
*l
;
3848 /* Nothing to do for write-only files */
3849 if (!(file
->f_mode
& FMODE_READ
))
3852 /* have the array populated */
3853 retval
= pidlist_array_load(cgrp
, type
, &l
);
3856 /* configure file information */
3857 file
->f_op
= &cgroup_pidlist_operations
;
3859 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3861 cgroup_release_pid_array(l
);
3864 ((struct seq_file
*)file
->private_data
)->private = l
;
3867 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3869 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3871 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3873 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3876 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3879 return notify_on_release(cgrp
);
3882 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3886 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3888 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3890 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3895 * When dput() is called asynchronously, if umount has been done and
3896 * then deactivate_super() in cgroup_free_fn() kills the superblock,
3897 * there's a small window that vfs will see the root dentry with non-zero
3898 * refcnt and trigger BUG().
3900 * That's why we hold a reference before dput() and drop it right after.
3902 static void cgroup_dput(struct cgroup
*cgrp
)
3904 struct super_block
*sb
= cgrp
->root
->sb
;
3906 atomic_inc(&sb
->s_active
);
3908 deactivate_super(sb
);
3912 * Unregister event and free resources.
3914 * Gets called from workqueue.
3916 static void cgroup_event_remove(struct work_struct
*work
)
3918 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3920 struct cgroup
*cgrp
= event
->cgrp
;
3922 remove_wait_queue(event
->wqh
, &event
->wait
);
3924 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3926 /* Notify userspace the event is going away. */
3927 eventfd_signal(event
->eventfd
, 1);
3929 eventfd_ctx_put(event
->eventfd
);
3935 * Gets called on POLLHUP on eventfd when user closes it.
3937 * Called with wqh->lock held and interrupts disabled.
3939 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3940 int sync
, void *key
)
3942 struct cgroup_event
*event
= container_of(wait
,
3943 struct cgroup_event
, wait
);
3944 struct cgroup
*cgrp
= event
->cgrp
;
3945 unsigned long flags
= (unsigned long)key
;
3947 if (flags
& POLLHUP
) {
3949 * If the event has been detached at cgroup removal, we
3950 * can simply return knowing the other side will cleanup
3953 * We can't race against event freeing since the other
3954 * side will require wqh->lock via remove_wait_queue(),
3957 spin_lock(&cgrp
->event_list_lock
);
3958 if (!list_empty(&event
->list
)) {
3959 list_del_init(&event
->list
);
3961 * We are in atomic context, but cgroup_event_remove()
3962 * may sleep, so we have to call it in workqueue.
3964 schedule_work(&event
->remove
);
3966 spin_unlock(&cgrp
->event_list_lock
);
3972 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3973 wait_queue_head_t
*wqh
, poll_table
*pt
)
3975 struct cgroup_event
*event
= container_of(pt
,
3976 struct cgroup_event
, pt
);
3979 add_wait_queue(wqh
, &event
->wait
);
3983 * Parse input and register new cgroup event handler.
3985 * Input must be in format '<event_fd> <control_fd> <args>'.
3986 * Interpretation of args is defined by control file implementation.
3988 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3991 struct cgroup_event
*event
= NULL
;
3992 struct cgroup
*cgrp_cfile
;
3993 unsigned int efd
, cfd
;
3994 struct file
*efile
= NULL
;
3995 struct file
*cfile
= NULL
;
3999 efd
= simple_strtoul(buffer
, &endp
, 10);
4004 cfd
= simple_strtoul(buffer
, &endp
, 10);
4005 if ((*endp
!= ' ') && (*endp
!= '\0'))
4009 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
4013 INIT_LIST_HEAD(&event
->list
);
4014 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
4015 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
4016 INIT_WORK(&event
->remove
, cgroup_event_remove
);
4018 efile
= eventfd_fget(efd
);
4019 if (IS_ERR(efile
)) {
4020 ret
= PTR_ERR(efile
);
4024 event
->eventfd
= eventfd_ctx_fileget(efile
);
4025 if (IS_ERR(event
->eventfd
)) {
4026 ret
= PTR_ERR(event
->eventfd
);
4036 /* the process need read permission on control file */
4037 /* AV: shouldn't we check that it's been opened for read instead? */
4038 ret
= inode_permission(file_inode(cfile
), MAY_READ
);
4042 event
->cft
= __file_cft(cfile
);
4043 if (IS_ERR(event
->cft
)) {
4044 ret
= PTR_ERR(event
->cft
);
4049 * The file to be monitored must be in the same cgroup as
4050 * cgroup.event_control is.
4052 cgrp_cfile
= __d_cgrp(cfile
->f_dentry
->d_parent
);
4053 if (cgrp_cfile
!= cgrp
) {
4058 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
4063 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
4064 event
->eventfd
, buffer
);
4068 efile
->f_op
->poll(efile
, &event
->pt
);
4071 * Events should be removed after rmdir of cgroup directory, but before
4072 * destroying subsystem state objects. Let's take reference to cgroup
4073 * directory dentry to do that.
4077 spin_lock(&cgrp
->event_list_lock
);
4078 list_add(&event
->list
, &cgrp
->event_list
);
4079 spin_unlock(&cgrp
->event_list_lock
);
4090 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
4091 eventfd_ctx_put(event
->eventfd
);
4093 if (!IS_ERR_OR_NULL(efile
))
4101 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
4104 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4107 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
4112 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4114 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4118 static struct cftype cgroup_base_files
[] = {
4120 .name
= "cgroup.procs",
4121 .open
= cgroup_procs_open
,
4122 .write_u64
= cgroup_procs_write
,
4123 .release
= cgroup_pidlist_release
,
4124 .mode
= S_IRUGO
| S_IWUSR
,
4127 .name
= "cgroup.event_control",
4128 .write_string
= cgroup_write_event_control
,
4132 .name
= "cgroup.clone_children",
4133 .flags
= CFTYPE_INSANE
,
4134 .read_u64
= cgroup_clone_children_read
,
4135 .write_u64
= cgroup_clone_children_write
,
4138 .name
= "cgroup.sane_behavior",
4139 .flags
= CFTYPE_ONLY_ON_ROOT
,
4140 .read_seq_string
= cgroup_sane_behavior_show
,
4144 * Historical crazy stuff. These don't have "cgroup." prefix and
4145 * don't exist if sane_behavior. If you're depending on these, be
4146 * prepared to be burned.
4150 .flags
= CFTYPE_INSANE
, /* use "procs" instead */
4151 .open
= cgroup_tasks_open
,
4152 .write_u64
= cgroup_tasks_write
,
4153 .release
= cgroup_pidlist_release
,
4154 .mode
= S_IRUGO
| S_IWUSR
,
4157 .name
= "notify_on_release",
4158 .flags
= CFTYPE_INSANE
,
4159 .read_u64
= cgroup_read_notify_on_release
,
4160 .write_u64
= cgroup_write_notify_on_release
,
4163 .name
= "release_agent",
4164 .flags
= CFTYPE_INSANE
| CFTYPE_ONLY_ON_ROOT
,
4165 .read_seq_string
= cgroup_release_agent_show
,
4166 .write_string
= cgroup_release_agent_write
,
4167 .max_write_len
= PATH_MAX
,
4173 * cgroup_populate_dir - create subsys files in a cgroup directory
4174 * @cgrp: target cgroup
4175 * @subsys_mask: mask of the subsystem ids whose files should be added
4177 * On failure, no file is added.
4179 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
4181 struct cgroup_subsys
*ss
;
4184 /* process cftsets of each subsystem */
4185 for_each_subsys(ss
, i
) {
4186 struct cftype_set
*set
;
4188 if (!test_bit(i
, &subsys_mask
))
4191 list_for_each_entry(set
, &ss
->cftsets
, node
) {
4192 ret
= cgroup_addrm_files(cgrp
, ss
, set
->cfts
, true);
4198 /* This cgroup is ready now */
4199 for_each_root_subsys(cgrp
->root
, ss
) {
4200 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4201 struct css_id
*id
= rcu_dereference_protected(css
->id
, true);
4204 * Update id->css pointer and make this css visible from
4205 * CSS ID functions. This pointer will be dereferened
4206 * from RCU-read-side without locks.
4209 rcu_assign_pointer(id
->css
, css
);
4214 cgroup_clear_dir(cgrp
, subsys_mask
);
4218 static void css_dput_fn(struct work_struct
*work
)
4220 struct cgroup_subsys_state
*css
=
4221 container_of(work
, struct cgroup_subsys_state
, dput_work
);
4223 cgroup_dput(css
->cgroup
);
4226 static void css_release(struct percpu_ref
*ref
)
4228 struct cgroup_subsys_state
*css
=
4229 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4231 schedule_work(&css
->dput_work
);
4234 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
4235 struct cgroup_subsys
*ss
,
4236 struct cgroup
*cgrp
)
4241 if (cgrp
== cgroup_dummy_top
)
4242 css
->flags
|= CSS_ROOT
;
4243 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
4244 cgrp
->subsys
[ss
->subsys_id
] = css
;
4247 * css holds an extra ref to @cgrp->dentry which is put on the last
4248 * css_put(). dput() requires process context, which css_put() may
4249 * be called without. @css->dput_work will be used to invoke
4250 * dput() asynchronously from css_put().
4252 INIT_WORK(&css
->dput_work
, css_dput_fn
);
4255 /* invoke ->post_create() on a new CSS and mark it online if successful */
4256 static int online_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4260 lockdep_assert_held(&cgroup_mutex
);
4263 ret
= ss
->css_online(cgrp
);
4265 cgrp
->subsys
[ss
->subsys_id
]->flags
|= CSS_ONLINE
;
4269 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4270 static void offline_css(struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4271 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4273 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4275 lockdep_assert_held(&cgroup_mutex
);
4277 if (!(css
->flags
& CSS_ONLINE
))
4280 if (ss
->css_offline
)
4281 ss
->css_offline(cgrp
);
4283 cgrp
->subsys
[ss
->subsys_id
]->flags
&= ~CSS_ONLINE
;
4287 * cgroup_create - create a cgroup
4288 * @parent: cgroup that will be parent of the new cgroup
4289 * @dentry: dentry of the new cgroup
4290 * @mode: mode to set on new inode
4292 * Must be called with the mutex on the parent inode held
4294 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
4297 struct cgroup
*cgrp
;
4298 struct cgroup_name
*name
;
4299 struct cgroupfs_root
*root
= parent
->root
;
4301 struct cgroup_subsys
*ss
;
4302 struct super_block
*sb
= root
->sb
;
4304 /* allocate the cgroup and its ID, 0 is reserved for the root */
4305 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4309 name
= cgroup_alloc_name(dentry
);
4312 rcu_assign_pointer(cgrp
->name
, name
);
4314 cgrp
->id
= ida_simple_get(&root
->cgroup_ida
, 1, 0, GFP_KERNEL
);
4319 * Only live parents can have children. Note that the liveliness
4320 * check isn't strictly necessary because cgroup_mkdir() and
4321 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4322 * anyway so that locking is contained inside cgroup proper and we
4323 * don't get nasty surprises if we ever grow another caller.
4325 if (!cgroup_lock_live_group(parent
)) {
4330 /* Grab a reference on the superblock so the hierarchy doesn't
4331 * get deleted on unmount if there are child cgroups. This
4332 * can be done outside cgroup_mutex, since the sb can't
4333 * disappear while someone has an open control file on the
4335 atomic_inc(&sb
->s_active
);
4337 init_cgroup_housekeeping(cgrp
);
4339 dentry
->d_fsdata
= cgrp
;
4340 cgrp
->dentry
= dentry
;
4342 cgrp
->parent
= parent
;
4343 cgrp
->root
= parent
->root
;
4345 if (notify_on_release(parent
))
4346 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4348 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4349 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4351 for_each_root_subsys(root
, ss
) {
4352 struct cgroup_subsys_state
*css
;
4354 css
= ss
->css_alloc(cgrp
);
4360 err
= percpu_ref_init(&css
->refcnt
, css_release
);
4364 init_cgroup_css(css
, ss
, cgrp
);
4367 err
= alloc_css_id(ss
, parent
, cgrp
);
4374 * Create directory. cgroup_create_file() returns with the new
4375 * directory locked on success so that it can be populated without
4376 * dropping cgroup_mutex.
4378 err
= cgroup_create_file(dentry
, S_IFDIR
| mode
, sb
);
4381 lockdep_assert_held(&dentry
->d_inode
->i_mutex
);
4383 cgrp
->serial_nr
= cgroup_serial_nr_next
++;
4385 /* allocation complete, commit to creation */
4386 list_add_tail_rcu(&cgrp
->sibling
, &cgrp
->parent
->children
);
4387 root
->number_of_cgroups
++;
4389 /* each css holds a ref to the cgroup's dentry */
4390 for_each_root_subsys(root
, ss
)
4393 /* hold a ref to the parent's dentry */
4394 dget(parent
->dentry
);
4396 /* creation succeeded, notify subsystems */
4397 for_each_root_subsys(root
, ss
) {
4398 err
= online_css(ss
, cgrp
);
4402 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4404 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",
4405 current
->comm
, current
->pid
, ss
->name
);
4406 if (!strcmp(ss
->name
, "memory"))
4407 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4408 ss
->warned_broken_hierarchy
= true;
4412 err
= cgroup_addrm_files(cgrp
, NULL
, cgroup_base_files
, true);
4416 err
= cgroup_populate_dir(cgrp
, root
->subsys_mask
);
4420 mutex_unlock(&cgroup_mutex
);
4421 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4426 for_each_root_subsys(root
, ss
) {
4427 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4430 percpu_ref_cancel_init(&css
->refcnt
);
4434 mutex_unlock(&cgroup_mutex
);
4435 /* Release the reference count that we took on the superblock */
4436 deactivate_super(sb
);
4438 ida_simple_remove(&root
->cgroup_ida
, cgrp
->id
);
4440 kfree(rcu_dereference_raw(cgrp
->name
));
4446 cgroup_destroy_locked(cgrp
);
4447 mutex_unlock(&cgroup_mutex
);
4448 mutex_unlock(&dentry
->d_inode
->i_mutex
);
4452 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4454 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4456 /* the vfs holds inode->i_mutex already */
4457 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4460 static void cgroup_css_killed(struct cgroup
*cgrp
)
4462 if (!atomic_dec_and_test(&cgrp
->css_kill_cnt
))
4465 /* percpu ref's of all css's are killed, kick off the next step */
4466 INIT_WORK(&cgrp
->destroy_work
, cgroup_offline_fn
);
4467 schedule_work(&cgrp
->destroy_work
);
4470 static void css_ref_killed_fn(struct percpu_ref
*ref
)
4472 struct cgroup_subsys_state
*css
=
4473 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4475 cgroup_css_killed(css
->cgroup
);
4479 * cgroup_destroy_locked - the first stage of cgroup destruction
4480 * @cgrp: cgroup to be destroyed
4482 * css's make use of percpu refcnts whose killing latency shouldn't be
4483 * exposed to userland and are RCU protected. Also, cgroup core needs to
4484 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
4485 * invoked. To satisfy all the requirements, destruction is implemented in
4486 * the following two steps.
4488 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4489 * userland visible parts and start killing the percpu refcnts of
4490 * css's. Set up so that the next stage will be kicked off once all
4491 * the percpu refcnts are confirmed to be killed.
4493 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4494 * rest of destruction. Once all cgroup references are gone, the
4495 * cgroup is RCU-freed.
4497 * This function implements s1. After this step, @cgrp is gone as far as
4498 * the userland is concerned and a new cgroup with the same name may be
4499 * created. As cgroup doesn't care about the names internally, this
4500 * doesn't cause any problem.
4502 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4503 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4505 struct dentry
*d
= cgrp
->dentry
;
4506 struct cgroup_event
*event
, *tmp
;
4507 struct cgroup_subsys
*ss
;
4510 lockdep_assert_held(&d
->d_inode
->i_mutex
);
4511 lockdep_assert_held(&cgroup_mutex
);
4514 * css_set_lock synchronizes access to ->cset_links and prevents
4515 * @cgrp from being removed while __put_css_set() is in progress.
4517 read_lock(&css_set_lock
);
4518 empty
= list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
);
4519 read_unlock(&css_set_lock
);
4524 * Block new css_tryget() by killing css refcnts. cgroup core
4525 * guarantees that, by the time ->css_offline() is invoked, no new
4526 * css reference will be given out via css_tryget(). We can't
4527 * simply call percpu_ref_kill() and proceed to offlining css's
4528 * because percpu_ref_kill() doesn't guarantee that the ref is seen
4529 * as killed on all CPUs on return.
4531 * Use percpu_ref_kill_and_confirm() to get notifications as each
4532 * css is confirmed to be seen as killed on all CPUs. The
4533 * notification callback keeps track of the number of css's to be
4534 * killed and schedules cgroup_offline_fn() to perform the rest of
4535 * destruction once the percpu refs of all css's are confirmed to
4538 atomic_set(&cgrp
->css_kill_cnt
, 1);
4539 for_each_root_subsys(cgrp
->root
, ss
) {
4540 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4543 * Killing would put the base ref, but we need to keep it
4544 * alive until after ->css_offline.
4546 percpu_ref_get(&css
->refcnt
);
4548 atomic_inc(&cgrp
->css_kill_cnt
);
4549 percpu_ref_kill_and_confirm(&css
->refcnt
, css_ref_killed_fn
);
4551 cgroup_css_killed(cgrp
);
4554 * Mark @cgrp dead. This prevents further task migration and child
4555 * creation by disabling cgroup_lock_live_group(). Note that
4556 * CGRP_DEAD assertion is depended upon by cgroup_next_sibling() to
4557 * resume iteration after dropping RCU read lock. See
4558 * cgroup_next_sibling() for details.
4560 set_bit(CGRP_DEAD
, &cgrp
->flags
);
4562 /* CGRP_DEAD is set, remove from ->release_list for the last time */
4563 raw_spin_lock(&release_list_lock
);
4564 if (!list_empty(&cgrp
->release_list
))
4565 list_del_init(&cgrp
->release_list
);
4566 raw_spin_unlock(&release_list_lock
);
4569 * Clear and remove @cgrp directory. The removal puts the base ref
4570 * but we aren't quite done with @cgrp yet, so hold onto it.
4572 cgroup_clear_dir(cgrp
, cgrp
->root
->subsys_mask
);
4573 cgroup_addrm_files(cgrp
, NULL
, cgroup_base_files
, false);
4575 cgroup_d_remove_dir(d
);
4578 * Unregister events and notify userspace.
4579 * Notify userspace about cgroup removing only after rmdir of cgroup
4580 * directory to avoid race between userspace and kernelspace.
4582 spin_lock(&cgrp
->event_list_lock
);
4583 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4584 list_del_init(&event
->list
);
4585 schedule_work(&event
->remove
);
4587 spin_unlock(&cgrp
->event_list_lock
);
4593 * cgroup_offline_fn - the second step of cgroup destruction
4594 * @work: cgroup->destroy_free_work
4596 * This function is invoked from a work item for a cgroup which is being
4597 * destroyed after the percpu refcnts of all css's are guaranteed to be
4598 * seen as killed on all CPUs, and performs the rest of destruction. This
4599 * is the second step of destruction described in the comment above
4600 * cgroup_destroy_locked().
4602 static void cgroup_offline_fn(struct work_struct
*work
)
4604 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, destroy_work
);
4605 struct cgroup
*parent
= cgrp
->parent
;
4606 struct dentry
*d
= cgrp
->dentry
;
4607 struct cgroup_subsys
*ss
;
4609 mutex_lock(&cgroup_mutex
);
4612 * css_tryget() is guaranteed to fail now. Tell subsystems to
4613 * initate destruction.
4615 for_each_root_subsys(cgrp
->root
, ss
)
4616 offline_css(ss
, cgrp
);
4619 * Put the css refs from cgroup_destroy_locked(). Each css holds
4620 * an extra reference to the cgroup's dentry and cgroup removal
4621 * proceeds regardless of css refs. On the last put of each css,
4622 * whenever that may be, the extra dentry ref is put so that dentry
4623 * destruction happens only after all css's are released.
4625 for_each_root_subsys(cgrp
->root
, ss
)
4626 css_put(cgrp
->subsys
[ss
->subsys_id
]);
4628 /* delete this cgroup from parent->children */
4629 list_del_rcu(&cgrp
->sibling
);
4633 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4634 check_for_release(parent
);
4636 mutex_unlock(&cgroup_mutex
);
4639 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4643 mutex_lock(&cgroup_mutex
);
4644 ret
= cgroup_destroy_locked(dentry
->d_fsdata
);
4645 mutex_unlock(&cgroup_mutex
);
4650 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4652 INIT_LIST_HEAD(&ss
->cftsets
);
4655 * base_cftset is embedded in subsys itself, no need to worry about
4658 if (ss
->base_cftypes
) {
4659 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4660 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4664 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4666 struct cgroup_subsys_state
*css
;
4668 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4670 mutex_lock(&cgroup_mutex
);
4672 /* init base cftset */
4673 cgroup_init_cftsets(ss
);
4675 /* Create the top cgroup state for this subsystem */
4676 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4677 ss
->root
= &cgroup_dummy_root
;
4678 css
= ss
->css_alloc(cgroup_dummy_top
);
4679 /* We don't handle early failures gracefully */
4680 BUG_ON(IS_ERR(css
));
4681 init_cgroup_css(css
, ss
, cgroup_dummy_top
);
4683 /* Update the init_css_set to contain a subsys
4684 * pointer to this state - since the subsystem is
4685 * newly registered, all tasks and hence the
4686 * init_css_set is in the subsystem's top cgroup. */
4687 init_css_set
.subsys
[ss
->subsys_id
] = css
;
4689 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4691 /* At system boot, before all subsystems have been
4692 * registered, no tasks have been forked, so we don't
4693 * need to invoke fork callbacks here. */
4694 BUG_ON(!list_empty(&init_task
.tasks
));
4696 BUG_ON(online_css(ss
, cgroup_dummy_top
));
4698 mutex_unlock(&cgroup_mutex
);
4700 /* this function shouldn't be used with modular subsystems, since they
4701 * need to register a subsys_id, among other things */
4706 * cgroup_load_subsys: load and register a modular subsystem at runtime
4707 * @ss: the subsystem to load
4709 * This function should be called in a modular subsystem's initcall. If the
4710 * subsystem is built as a module, it will be assigned a new subsys_id and set
4711 * up for use. If the subsystem is built-in anyway, work is delegated to the
4712 * simpler cgroup_init_subsys.
4714 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4716 struct cgroup_subsys_state
*css
;
4718 struct hlist_node
*tmp
;
4719 struct css_set
*cset
;
4722 /* check name and function validity */
4723 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4724 ss
->css_alloc
== NULL
|| ss
->css_free
== NULL
)
4728 * we don't support callbacks in modular subsystems. this check is
4729 * before the ss->module check for consistency; a subsystem that could
4730 * be a module should still have no callbacks even if the user isn't
4731 * compiling it as one.
4733 if (ss
->fork
|| ss
->exit
)
4737 * an optionally modular subsystem is built-in: we want to do nothing,
4738 * since cgroup_init_subsys will have already taken care of it.
4740 if (ss
->module
== NULL
) {
4741 /* a sanity check */
4742 BUG_ON(cgroup_subsys
[ss
->subsys_id
] != ss
);
4746 /* init base cftset */
4747 cgroup_init_cftsets(ss
);
4749 mutex_lock(&cgroup_mutex
);
4750 cgroup_subsys
[ss
->subsys_id
] = ss
;
4753 * no ss->css_alloc seems to need anything important in the ss
4754 * struct, so this can happen first (i.e. before the dummy root
4757 css
= ss
->css_alloc(cgroup_dummy_top
);
4759 /* failure case - need to deassign the cgroup_subsys[] slot. */
4760 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4761 mutex_unlock(&cgroup_mutex
);
4762 return PTR_ERR(css
);
4765 list_add(&ss
->sibling
, &cgroup_dummy_root
.subsys_list
);
4766 ss
->root
= &cgroup_dummy_root
;
4768 /* our new subsystem will be attached to the dummy hierarchy. */
4769 init_cgroup_css(css
, ss
, cgroup_dummy_top
);
4770 /* init_idr must be after init_cgroup_css because it sets css->id. */
4772 ret
= cgroup_init_idr(ss
, css
);
4778 * Now we need to entangle the css into the existing css_sets. unlike
4779 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4780 * will need a new pointer to it; done by iterating the css_set_table.
4781 * furthermore, modifying the existing css_sets will corrupt the hash
4782 * table state, so each changed css_set will need its hash recomputed.
4783 * this is all done under the css_set_lock.
4785 write_lock(&css_set_lock
);
4786 hash_for_each_safe(css_set_table
, i
, tmp
, cset
, hlist
) {
4787 /* skip entries that we already rehashed */
4788 if (cset
->subsys
[ss
->subsys_id
])
4790 /* remove existing entry */
4791 hash_del(&cset
->hlist
);
4793 cset
->subsys
[ss
->subsys_id
] = css
;
4794 /* recompute hash and restore entry */
4795 key
= css_set_hash(cset
->subsys
);
4796 hash_add(css_set_table
, &cset
->hlist
, key
);
4798 write_unlock(&css_set_lock
);
4800 ret
= online_css(ss
, cgroup_dummy_top
);
4805 mutex_unlock(&cgroup_mutex
);
4809 mutex_unlock(&cgroup_mutex
);
4810 /* @ss can't be mounted here as try_module_get() would fail */
4811 cgroup_unload_subsys(ss
);
4814 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4817 * cgroup_unload_subsys: unload a modular subsystem
4818 * @ss: the subsystem to unload
4820 * This function should be called in a modular subsystem's exitcall. When this
4821 * function is invoked, the refcount on the subsystem's module will be 0, so
4822 * the subsystem will not be attached to any hierarchy.
4824 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4826 struct cgrp_cset_link
*link
;
4828 BUG_ON(ss
->module
== NULL
);
4831 * we shouldn't be called if the subsystem is in use, and the use of
4832 * try_module_get in parse_cgroupfs_options should ensure that it
4833 * doesn't start being used while we're killing it off.
4835 BUG_ON(ss
->root
!= &cgroup_dummy_root
);
4837 mutex_lock(&cgroup_mutex
);
4839 offline_css(ss
, cgroup_dummy_top
);
4842 idr_destroy(&ss
->idr
);
4844 /* deassign the subsys_id */
4845 cgroup_subsys
[ss
->subsys_id
] = NULL
;
4847 /* remove subsystem from the dummy root's list of subsystems */
4848 list_del_init(&ss
->sibling
);
4851 * disentangle the css from all css_sets attached to the dummy
4852 * top. as in loading, we need to pay our respects to the hashtable
4855 write_lock(&css_set_lock
);
4856 list_for_each_entry(link
, &cgroup_dummy_top
->cset_links
, cset_link
) {
4857 struct css_set
*cset
= link
->cset
;
4860 hash_del(&cset
->hlist
);
4861 cset
->subsys
[ss
->subsys_id
] = NULL
;
4862 key
= css_set_hash(cset
->subsys
);
4863 hash_add(css_set_table
, &cset
->hlist
, key
);
4865 write_unlock(&css_set_lock
);
4868 * remove subsystem's css from the cgroup_dummy_top and free it -
4869 * need to free before marking as null because ss->css_free needs
4870 * the cgrp->subsys pointer to find their state. note that this
4871 * also takes care of freeing the css_id.
4873 ss
->css_free(cgroup_dummy_top
);
4874 cgroup_dummy_top
->subsys
[ss
->subsys_id
] = NULL
;
4876 mutex_unlock(&cgroup_mutex
);
4878 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4881 * cgroup_init_early - cgroup initialization at system boot
4883 * Initialize cgroups at system boot, and initialize any
4884 * subsystems that request early init.
4886 int __init
cgroup_init_early(void)
4888 struct cgroup_subsys
*ss
;
4891 atomic_set(&init_css_set
.refcount
, 1);
4892 INIT_LIST_HEAD(&init_css_set
.cgrp_links
);
4893 INIT_LIST_HEAD(&init_css_set
.tasks
);
4894 INIT_HLIST_NODE(&init_css_set
.hlist
);
4896 init_cgroup_root(&cgroup_dummy_root
);
4897 cgroup_root_count
= 1;
4898 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
4900 init_cgrp_cset_link
.cset
= &init_css_set
;
4901 init_cgrp_cset_link
.cgrp
= cgroup_dummy_top
;
4902 list_add(&init_cgrp_cset_link
.cset_link
, &cgroup_dummy_top
->cset_links
);
4903 list_add(&init_cgrp_cset_link
.cgrp_link
, &init_css_set
.cgrp_links
);
4905 /* at bootup time, we don't worry about modular subsystems */
4906 for_each_builtin_subsys(ss
, i
) {
4908 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4909 BUG_ON(!ss
->css_alloc
);
4910 BUG_ON(!ss
->css_free
);
4911 if (ss
->subsys_id
!= i
) {
4912 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4913 ss
->name
, ss
->subsys_id
);
4918 cgroup_init_subsys(ss
);
4924 * cgroup_init - cgroup initialization
4926 * Register cgroup filesystem and /proc file, and initialize
4927 * any subsystems that didn't request early init.
4929 int __init
cgroup_init(void)
4931 struct cgroup_subsys
*ss
;
4935 err
= bdi_init(&cgroup_backing_dev_info
);
4939 for_each_builtin_subsys(ss
, i
) {
4940 if (!ss
->early_init
)
4941 cgroup_init_subsys(ss
);
4943 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4946 /* allocate id for the dummy hierarchy */
4947 mutex_lock(&cgroup_mutex
);
4948 mutex_lock(&cgroup_root_mutex
);
4950 /* Add init_css_set to the hash table */
4951 key
= css_set_hash(init_css_set
.subsys
);
4952 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4954 BUG_ON(cgroup_init_root_id(&cgroup_dummy_root
, 0, 1));
4956 mutex_unlock(&cgroup_root_mutex
);
4957 mutex_unlock(&cgroup_mutex
);
4959 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4965 err
= register_filesystem(&cgroup_fs_type
);
4967 kobject_put(cgroup_kobj
);
4971 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4975 bdi_destroy(&cgroup_backing_dev_info
);
4981 * proc_cgroup_show()
4982 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4983 * - Used for /proc/<pid>/cgroup.
4984 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4985 * doesn't really matter if tsk->cgroup changes after we read it,
4986 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4987 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4988 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4989 * cgroup to top_cgroup.
4992 /* TODO: Use a proper seq_file iterator */
4993 int proc_cgroup_show(struct seq_file
*m
, void *v
)
4996 struct task_struct
*tsk
;
4999 struct cgroupfs_root
*root
;
5002 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5008 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
5014 mutex_lock(&cgroup_mutex
);
5016 for_each_active_root(root
) {
5017 struct cgroup_subsys
*ss
;
5018 struct cgroup
*cgrp
;
5021 seq_printf(m
, "%d:", root
->hierarchy_id
);
5022 for_each_root_subsys(root
, ss
)
5023 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
5024 if (strlen(root
->name
))
5025 seq_printf(m
, "%sname=%s", count
? "," : "",
5028 cgrp
= task_cgroup_from_root(tsk
, root
);
5029 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
5037 mutex_unlock(&cgroup_mutex
);
5038 put_task_struct(tsk
);
5045 /* Display information about each subsystem and each hierarchy */
5046 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5048 struct cgroup_subsys
*ss
;
5051 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5053 * ideally we don't want subsystems moving around while we do this.
5054 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5055 * subsys/hierarchy state.
5057 mutex_lock(&cgroup_mutex
);
5059 for_each_subsys(ss
, i
)
5060 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5061 ss
->name
, ss
->root
->hierarchy_id
,
5062 ss
->root
->number_of_cgroups
, !ss
->disabled
);
5064 mutex_unlock(&cgroup_mutex
);
5068 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5070 return single_open(file
, proc_cgroupstats_show
, NULL
);
5073 static const struct file_operations proc_cgroupstats_operations
= {
5074 .open
= cgroupstats_open
,
5076 .llseek
= seq_lseek
,
5077 .release
= single_release
,
5081 * cgroup_fork - attach newly forked task to its parents cgroup.
5082 * @child: pointer to task_struct of forking parent process.
5084 * Description: A task inherits its parent's cgroup at fork().
5086 * A pointer to the shared css_set was automatically copied in
5087 * fork.c by dup_task_struct(). However, we ignore that copy, since
5088 * it was not made under the protection of RCU or cgroup_mutex, so
5089 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
5090 * have already changed current->cgroups, allowing the previously
5091 * referenced cgroup group to be removed and freed.
5093 * At the point that cgroup_fork() is called, 'current' is the parent
5094 * task, and the passed argument 'child' points to the child task.
5096 void cgroup_fork(struct task_struct
*child
)
5099 get_css_set(task_css_set(current
));
5100 child
->cgroups
= current
->cgroups
;
5101 task_unlock(current
);
5102 INIT_LIST_HEAD(&child
->cg_list
);
5106 * cgroup_post_fork - called on a new task after adding it to the task list
5107 * @child: the task in question
5109 * Adds the task to the list running through its css_set if necessary and
5110 * call the subsystem fork() callbacks. Has to be after the task is
5111 * visible on the task list in case we race with the first call to
5112 * cgroup_iter_start() - to guarantee that the new task ends up on its
5115 void cgroup_post_fork(struct task_struct
*child
)
5117 struct cgroup_subsys
*ss
;
5121 * use_task_css_set_links is set to 1 before we walk the tasklist
5122 * under the tasklist_lock and we read it here after we added the child
5123 * to the tasklist under the tasklist_lock as well. If the child wasn't
5124 * yet in the tasklist when we walked through it from
5125 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
5126 * should be visible now due to the paired locking and barriers implied
5127 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
5128 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
5131 if (use_task_css_set_links
) {
5132 write_lock(&css_set_lock
);
5134 if (list_empty(&child
->cg_list
))
5135 list_add(&child
->cg_list
, &task_css_set(child
)->tasks
);
5137 write_unlock(&css_set_lock
);
5141 * Call ss->fork(). This must happen after @child is linked on
5142 * css_set; otherwise, @child might change state between ->fork()
5143 * and addition to css_set.
5145 if (need_forkexit_callback
) {
5147 * fork/exit callbacks are supported only for builtin
5148 * subsystems, and the builtin section of the subsys
5149 * array is immutable, so we don't need to lock the
5150 * subsys array here. On the other hand, modular section
5151 * of the array can be freed at module unload, so we
5154 for_each_builtin_subsys(ss
, i
)
5161 * cgroup_exit - detach cgroup from exiting task
5162 * @tsk: pointer to task_struct of exiting process
5163 * @run_callback: run exit callbacks?
5165 * Description: Detach cgroup from @tsk and release it.
5167 * Note that cgroups marked notify_on_release force every task in
5168 * them to take the global cgroup_mutex mutex when exiting.
5169 * This could impact scaling on very large systems. Be reluctant to
5170 * use notify_on_release cgroups where very high task exit scaling
5171 * is required on large systems.
5173 * the_top_cgroup_hack:
5175 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
5177 * We call cgroup_exit() while the task is still competent to
5178 * handle notify_on_release(), then leave the task attached to the
5179 * root cgroup in each hierarchy for the remainder of its exit.
5181 * To do this properly, we would increment the reference count on
5182 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
5183 * code we would add a second cgroup function call, to drop that
5184 * reference. This would just create an unnecessary hot spot on
5185 * the top_cgroup reference count, to no avail.
5187 * Normally, holding a reference to a cgroup without bumping its
5188 * count is unsafe. The cgroup could go away, or someone could
5189 * attach us to a different cgroup, decrementing the count on
5190 * the first cgroup that we never incremented. But in this case,
5191 * top_cgroup isn't going away, and either task has PF_EXITING set,
5192 * which wards off any cgroup_attach_task() attempts, or task is a failed
5193 * fork, never visible to cgroup_attach_task.
5195 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
5197 struct cgroup_subsys
*ss
;
5198 struct css_set
*cset
;
5202 * Unlink from the css_set task list if necessary.
5203 * Optimistically check cg_list before taking
5206 if (!list_empty(&tsk
->cg_list
)) {
5207 write_lock(&css_set_lock
);
5208 if (!list_empty(&tsk
->cg_list
))
5209 list_del_init(&tsk
->cg_list
);
5210 write_unlock(&css_set_lock
);
5213 /* Reassign the task to the init_css_set. */
5215 cset
= task_css_set(tsk
);
5216 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5218 if (run_callbacks
&& need_forkexit_callback
) {
5220 * fork/exit callbacks are supported only for builtin
5221 * subsystems, see cgroup_post_fork() for details.
5223 for_each_builtin_subsys(ss
, i
) {
5225 struct cgroup
*old_cgrp
= cset
->subsys
[i
]->cgroup
;
5226 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
5228 ss
->exit(cgrp
, old_cgrp
, tsk
);
5234 put_css_set_taskexit(cset
);
5237 static void check_for_release(struct cgroup
*cgrp
)
5239 if (cgroup_is_releasable(cgrp
) &&
5240 list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
)) {
5242 * Control Group is currently removeable. If it's not
5243 * already queued for a userspace notification, queue
5246 int need_schedule_work
= 0;
5248 raw_spin_lock(&release_list_lock
);
5249 if (!cgroup_is_dead(cgrp
) &&
5250 list_empty(&cgrp
->release_list
)) {
5251 list_add(&cgrp
->release_list
, &release_list
);
5252 need_schedule_work
= 1;
5254 raw_spin_unlock(&release_list_lock
);
5255 if (need_schedule_work
)
5256 schedule_work(&release_agent_work
);
5261 * Notify userspace when a cgroup is released, by running the
5262 * configured release agent with the name of the cgroup (path
5263 * relative to the root of cgroup file system) as the argument.
5265 * Most likely, this user command will try to rmdir this cgroup.
5267 * This races with the possibility that some other task will be
5268 * attached to this cgroup before it is removed, or that some other
5269 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5270 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5271 * unused, and this cgroup will be reprieved from its death sentence,
5272 * to continue to serve a useful existence. Next time it's released,
5273 * we will get notified again, if it still has 'notify_on_release' set.
5275 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5276 * means only wait until the task is successfully execve()'d. The
5277 * separate release agent task is forked by call_usermodehelper(),
5278 * then control in this thread returns here, without waiting for the
5279 * release agent task. We don't bother to wait because the caller of
5280 * this routine has no use for the exit status of the release agent
5281 * task, so no sense holding our caller up for that.
5283 static void cgroup_release_agent(struct work_struct
*work
)
5285 BUG_ON(work
!= &release_agent_work
);
5286 mutex_lock(&cgroup_mutex
);
5287 raw_spin_lock(&release_list_lock
);
5288 while (!list_empty(&release_list
)) {
5289 char *argv
[3], *envp
[3];
5291 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5292 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5295 list_del_init(&cgrp
->release_list
);
5296 raw_spin_unlock(&release_list_lock
);
5297 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5300 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5302 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5307 argv
[i
++] = agentbuf
;
5308 argv
[i
++] = pathbuf
;
5312 /* minimal command environment */
5313 envp
[i
++] = "HOME=/";
5314 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5317 /* Drop the lock while we invoke the usermode helper,
5318 * since the exec could involve hitting disk and hence
5319 * be a slow process */
5320 mutex_unlock(&cgroup_mutex
);
5321 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5322 mutex_lock(&cgroup_mutex
);
5326 raw_spin_lock(&release_list_lock
);
5328 raw_spin_unlock(&release_list_lock
);
5329 mutex_unlock(&cgroup_mutex
);
5332 static int __init
cgroup_disable(char *str
)
5334 struct cgroup_subsys
*ss
;
5338 while ((token
= strsep(&str
, ",")) != NULL
) {
5343 * cgroup_disable, being at boot time, can't know about
5344 * module subsystems, so we don't worry about them.
5346 for_each_builtin_subsys(ss
, i
) {
5347 if (!strcmp(token
, ss
->name
)) {
5349 printk(KERN_INFO
"Disabling %s control group"
5350 " subsystem\n", ss
->name
);
5357 __setup("cgroup_disable=", cgroup_disable
);
5360 * Functons for CSS ID.
5363 /* to get ID other than 0, this should be called when !cgroup_is_dead() */
5364 unsigned short css_id(struct cgroup_subsys_state
*css
)
5366 struct css_id
*cssid
;
5369 * This css_id() can return correct value when somone has refcnt
5370 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5371 * it's unchanged until freed.
5373 cssid
= rcu_dereference_raw(css
->id
);
5379 EXPORT_SYMBOL_GPL(css_id
);
5382 * css_is_ancestor - test "root" css is an ancestor of "child"
5383 * @child: the css to be tested.
5384 * @root: the css supporsed to be an ancestor of the child.
5386 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5387 * this function reads css->id, the caller must hold rcu_read_lock().
5388 * But, considering usual usage, the csses should be valid objects after test.
5389 * Assuming that the caller will do some action to the child if this returns
5390 * returns true, the caller must take "child";s reference count.
5391 * If "child" is valid object and this returns true, "root" is valid, too.
5394 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5395 const struct cgroup_subsys_state
*root
)
5397 struct css_id
*child_id
;
5398 struct css_id
*root_id
;
5400 child_id
= rcu_dereference(child
->id
);
5403 root_id
= rcu_dereference(root
->id
);
5406 if (child_id
->depth
< root_id
->depth
)
5408 if (child_id
->stack
[root_id
->depth
] != root_id
->id
)
5413 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5415 struct css_id
*id
= rcu_dereference_protected(css
->id
, true);
5417 /* When this is called before css_id initialization, id can be NULL */
5421 BUG_ON(!ss
->use_id
);
5423 rcu_assign_pointer(id
->css
, NULL
);
5424 rcu_assign_pointer(css
->id
, NULL
);
5425 spin_lock(&ss
->id_lock
);
5426 idr_remove(&ss
->idr
, id
->id
);
5427 spin_unlock(&ss
->id_lock
);
5428 kfree_rcu(id
, rcu_head
);
5430 EXPORT_SYMBOL_GPL(free_css_id
);
5433 * This is called by init or create(). Then, calls to this function are
5434 * always serialized (By cgroup_mutex() at create()).
5437 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5439 struct css_id
*newid
;
5442 BUG_ON(!ss
->use_id
);
5444 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5445 newid
= kzalloc(size
, GFP_KERNEL
);
5447 return ERR_PTR(-ENOMEM
);
5449 idr_preload(GFP_KERNEL
);
5450 spin_lock(&ss
->id_lock
);
5451 /* Don't use 0. allocates an ID of 1-65535 */
5452 ret
= idr_alloc(&ss
->idr
, newid
, 1, CSS_ID_MAX
+ 1, GFP_NOWAIT
);
5453 spin_unlock(&ss
->id_lock
);
5456 /* Returns error when there are no free spaces for new ID.*/
5461 newid
->depth
= depth
;
5465 return ERR_PTR(ret
);
5469 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5470 struct cgroup_subsys_state
*rootcss
)
5472 struct css_id
*newid
;
5474 spin_lock_init(&ss
->id_lock
);
5477 newid
= get_new_cssid(ss
, 0);
5479 return PTR_ERR(newid
);
5481 newid
->stack
[0] = newid
->id
;
5482 RCU_INIT_POINTER(newid
->css
, rootcss
);
5483 RCU_INIT_POINTER(rootcss
->id
, newid
);
5487 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5488 struct cgroup
*child
)
5490 int subsys_id
, i
, depth
= 0;
5491 struct cgroup_subsys_state
*parent_css
, *child_css
;
5492 struct css_id
*child_id
, *parent_id
;
5494 subsys_id
= ss
->subsys_id
;
5495 parent_css
= parent
->subsys
[subsys_id
];
5496 child_css
= child
->subsys
[subsys_id
];
5497 parent_id
= rcu_dereference_protected(parent_css
->id
, true);
5498 depth
= parent_id
->depth
+ 1;
5500 child_id
= get_new_cssid(ss
, depth
);
5501 if (IS_ERR(child_id
))
5502 return PTR_ERR(child_id
);
5504 for (i
= 0; i
< depth
; i
++)
5505 child_id
->stack
[i
] = parent_id
->stack
[i
];
5506 child_id
->stack
[depth
] = child_id
->id
;
5508 * child_id->css pointer will be set after this cgroup is available
5509 * see cgroup_populate_dir()
5511 rcu_assign_pointer(child_css
->id
, child_id
);
5517 * css_lookup - lookup css by id
5518 * @ss: cgroup subsys to be looked into.
5521 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5522 * NULL if not. Should be called under rcu_read_lock()
5524 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5526 struct css_id
*cssid
= NULL
;
5528 BUG_ON(!ss
->use_id
);
5529 cssid
= idr_find(&ss
->idr
, id
);
5531 if (unlikely(!cssid
))
5534 return rcu_dereference(cssid
->css
);
5536 EXPORT_SYMBOL_GPL(css_lookup
);
5539 * get corresponding css from file open on cgroupfs directory
5541 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5543 struct cgroup
*cgrp
;
5544 struct inode
*inode
;
5545 struct cgroup_subsys_state
*css
;
5547 inode
= file_inode(f
);
5548 /* check in cgroup filesystem dir */
5549 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5550 return ERR_PTR(-EBADF
);
5552 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5553 return ERR_PTR(-EINVAL
);
5556 cgrp
= __d_cgrp(f
->f_dentry
);
5557 css
= cgrp
->subsys
[id
];
5558 return css
? css
: ERR_PTR(-ENOENT
);
5561 #ifdef CONFIG_CGROUP_DEBUG
5562 static struct cgroup_subsys_state
*debug_css_alloc(struct cgroup
*cgrp
)
5564 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5567 return ERR_PTR(-ENOMEM
);
5572 static void debug_css_free(struct cgroup
*cgrp
)
5574 kfree(cgrp
->subsys
[debug_subsys_id
]);
5577 static u64
debug_taskcount_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5579 return cgroup_task_count(cgrp
);
5582 static u64
current_css_set_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5584 return (u64
)(unsigned long)current
->cgroups
;
5587 static u64
current_css_set_refcount_read(struct cgroup
*cgrp
,
5593 count
= atomic_read(&task_css_set(current
)->refcount
);
5598 static int current_css_set_cg_links_read(struct cgroup
*cgrp
,
5600 struct seq_file
*seq
)
5602 struct cgrp_cset_link
*link
;
5603 struct css_set
*cset
;
5605 read_lock(&css_set_lock
);
5607 cset
= rcu_dereference(current
->cgroups
);
5608 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5609 struct cgroup
*c
= link
->cgrp
;
5613 name
= c
->dentry
->d_name
.name
;
5616 seq_printf(seq
, "Root %d group %s\n",
5617 c
->root
->hierarchy_id
, name
);
5620 read_unlock(&css_set_lock
);
5624 #define MAX_TASKS_SHOWN_PER_CSS 25
5625 static int cgroup_css_links_read(struct cgroup
*cgrp
,
5627 struct seq_file
*seq
)
5629 struct cgrp_cset_link
*link
;
5631 read_lock(&css_set_lock
);
5632 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
) {
5633 struct css_set
*cset
= link
->cset
;
5634 struct task_struct
*task
;
5636 seq_printf(seq
, "css_set %p\n", cset
);
5637 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5638 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5639 seq_puts(seq
, " ...\n");
5642 seq_printf(seq
, " task %d\n",
5643 task_pid_vnr(task
));
5647 read_unlock(&css_set_lock
);
5651 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5653 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5656 static struct cftype debug_files
[] = {
5658 .name
= "taskcount",
5659 .read_u64
= debug_taskcount_read
,
5663 .name
= "current_css_set",
5664 .read_u64
= current_css_set_read
,
5668 .name
= "current_css_set_refcount",
5669 .read_u64
= current_css_set_refcount_read
,
5673 .name
= "current_css_set_cg_links",
5674 .read_seq_string
= current_css_set_cg_links_read
,
5678 .name
= "cgroup_css_links",
5679 .read_seq_string
= cgroup_css_links_read
,
5683 .name
= "releasable",
5684 .read_u64
= releasable_read
,
5690 struct cgroup_subsys debug_subsys
= {
5692 .css_alloc
= debug_css_alloc
,
5693 .css_free
= debug_css_free
,
5694 .subsys_id
= debug_subsys_id
,
5695 .base_cftypes
= debug_files
,
5697 #endif /* CONFIG_CGROUP_DEBUG */