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/slab.h>
44 #include <linux/spinlock.h>
45 #include <linux/rwsem.h>
46 #include <linux/string.h>
47 #include <linux/sort.h>
48 #include <linux/kmod.h>
49 #include <linux/delayacct.h>
50 #include <linux/cgroupstats.h>
51 #include <linux/hashtable.h>
52 #include <linux/pid_namespace.h>
53 #include <linux/idr.h>
54 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
55 #include <linux/kthread.h>
56 #include <linux/delay.h>
58 #include <linux/atomic.h>
61 * pidlists linger the following amount before being destroyed. The goal
62 * is avoiding frequent destruction in the middle of consecutive read calls
63 * Expiring in the middle is a performance problem not a correctness one.
64 * 1 sec should be enough.
66 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
68 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
72 * cgroup_tree_mutex nests above cgroup_mutex and protects cftypes, file
73 * creation/removal and hierarchy changing operations including cgroup
74 * creation, removal, css association and controller rebinding. This outer
75 * lock is needed mainly to resolve the circular dependency between kernfs
76 * active ref and cgroup_mutex. cgroup_tree_mutex nests above both.
78 static DEFINE_MUTEX(cgroup_tree_mutex
);
81 * cgroup_mutex is the master lock. Any modification to cgroup or its
82 * hierarchy must be performed while holding it.
84 * css_set_rwsem protects task->cgroups pointer, the list of css_set
85 * objects, and the chain of tasks off each css_set.
87 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
88 * cgroup.h can use them for lockdep annotations.
90 #ifdef CONFIG_PROVE_RCU
91 DEFINE_MUTEX(cgroup_mutex
);
92 DECLARE_RWSEM(css_set_rwsem
);
93 EXPORT_SYMBOL_GPL(cgroup_mutex
);
94 EXPORT_SYMBOL_GPL(css_set_rwsem
);
96 static DEFINE_MUTEX(cgroup_mutex
);
97 static DECLARE_RWSEM(css_set_rwsem
);
101 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
102 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
104 static DEFINE_SPINLOCK(release_agent_path_lock
);
106 #define cgroup_assert_mutexes_or_rcu_locked() \
107 rcu_lockdep_assert(rcu_read_lock_held() || \
108 lockdep_is_held(&cgroup_tree_mutex) || \
109 lockdep_is_held(&cgroup_mutex), \
110 "cgroup_[tree_]mutex or RCU read lock required");
113 * cgroup destruction makes heavy use of work items and there can be a lot
114 * of concurrent destructions. Use a separate workqueue so that cgroup
115 * destruction work items don't end up filling up max_active of system_wq
116 * which may lead to deadlock.
118 static struct workqueue_struct
*cgroup_destroy_wq
;
121 * pidlist destructions need to be flushed on cgroup destruction. Use a
122 * separate workqueue as flush domain.
124 static struct workqueue_struct
*cgroup_pidlist_destroy_wq
;
126 /* generate an array of cgroup subsystem pointers */
127 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
128 static struct cgroup_subsys
*cgroup_subsys
[] = {
129 #include <linux/cgroup_subsys.h>
133 /* array of cgroup subsystem names */
134 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
135 static const char *cgroup_subsys_name
[] = {
136 #include <linux/cgroup_subsys.h>
141 * The default hierarchy, reserved for the subsystems that are otherwise
142 * unattached - it never has more than a single cgroup, and all tasks are
143 * part of that cgroup.
145 struct cgroup_root cgrp_dfl_root
;
148 * The default hierarchy always exists but is hidden until mounted for the
149 * first time. This is for backward compatibility.
151 static bool cgrp_dfl_root_visible
;
153 /* The list of hierarchy roots */
155 static LIST_HEAD(cgroup_roots
);
156 static int cgroup_root_count
;
158 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
159 static DEFINE_IDR(cgroup_hierarchy_idr
);
162 * Assign a monotonically increasing serial number to cgroups. It
163 * guarantees cgroups with bigger numbers are newer than those with smaller
164 * numbers. Also, as cgroups are always appended to the parent's
165 * ->children list, it guarantees that sibling cgroups are always sorted in
166 * the ascending serial number order on the list. Protected by
169 static u64 cgroup_serial_nr_next
= 1;
171 /* This flag indicates whether tasks in the fork and exit paths should
172 * check for fork/exit handlers to call. This avoids us having to do
173 * extra work in the fork/exit path if none of the subsystems need to
176 static int need_forkexit_callback __read_mostly
;
178 static struct cftype cgroup_base_files
[];
180 static void cgroup_put(struct cgroup
*cgrp
);
181 static int rebind_subsystems(struct cgroup_root
*dst_root
,
182 unsigned long ss_mask
);
183 static void cgroup_destroy_css_killed(struct cgroup
*cgrp
);
184 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
185 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
187 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
);
190 * cgroup_css - obtain a cgroup's css for the specified subsystem
191 * @cgrp: the cgroup of interest
192 * @ss: the subsystem of interest (%NULL returns the dummy_css)
194 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
195 * function must be called either under cgroup_mutex or rcu_read_lock() and
196 * the caller is responsible for pinning the returned css if it wants to
197 * keep accessing it outside the said locks. This function may return
198 * %NULL if @cgrp doesn't have @subsys_id enabled.
200 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
201 struct cgroup_subsys
*ss
)
204 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
205 lockdep_is_held(&cgroup_tree_mutex
) ||
206 lockdep_is_held(&cgroup_mutex
));
208 return &cgrp
->dummy_css
;
211 /* convenient tests for these bits */
212 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
214 return test_bit(CGRP_DEAD
, &cgrp
->flags
);
217 struct cgroup_subsys_state
*seq_css(struct seq_file
*seq
)
219 struct kernfs_open_file
*of
= seq
->private;
220 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
221 struct cftype
*cft
= seq_cft(seq
);
224 * This is open and unprotected implementation of cgroup_css().
225 * seq_css() is only called from a kernfs file operation which has
226 * an active reference on the file. Because all the subsystem
227 * files are drained before a css is disassociated with a cgroup,
228 * the matching css from the cgroup's subsys table is guaranteed to
229 * be and stay valid until the enclosing operation is complete.
232 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
234 return &cgrp
->dummy_css
;
236 EXPORT_SYMBOL_GPL(seq_css
);
239 * cgroup_is_descendant - test ancestry
240 * @cgrp: the cgroup to be tested
241 * @ancestor: possible ancestor of @cgrp
243 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
244 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
245 * and @ancestor are accessible.
247 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
250 if (cgrp
== ancestor
)
257 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
260 (1 << CGRP_RELEASABLE
) |
261 (1 << CGRP_NOTIFY_ON_RELEASE
);
262 return (cgrp
->flags
& bits
) == bits
;
265 static int notify_on_release(const struct cgroup
*cgrp
)
267 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
271 * for_each_css - iterate all css's of a cgroup
272 * @css: the iteration cursor
273 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
274 * @cgrp: the target cgroup to iterate css's of
276 * Should be called under cgroup_mutex.
278 #define for_each_css(css, ssid, cgrp) \
279 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
280 if (!((css) = rcu_dereference_check( \
281 (cgrp)->subsys[(ssid)], \
282 lockdep_is_held(&cgroup_tree_mutex) || \
283 lockdep_is_held(&cgroup_mutex)))) { } \
287 * for_each_subsys - iterate all enabled cgroup subsystems
288 * @ss: the iteration cursor
289 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
291 #define for_each_subsys(ss, ssid) \
292 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
293 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
295 /* iterate across the hierarchies */
296 #define for_each_root(root) \
297 list_for_each_entry((root), &cgroup_roots, root_list)
300 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
301 * @cgrp: the cgroup to be checked for liveness
303 * On success, returns true; the mutex should be later unlocked. On
304 * failure returns false with no lock held.
306 static bool cgroup_lock_live_group(struct cgroup
*cgrp
)
308 mutex_lock(&cgroup_mutex
);
309 if (cgroup_is_dead(cgrp
)) {
310 mutex_unlock(&cgroup_mutex
);
316 /* the list of cgroups eligible for automatic release. Protected by
317 * release_list_lock */
318 static LIST_HEAD(release_list
);
319 static DEFINE_RAW_SPINLOCK(release_list_lock
);
320 static void cgroup_release_agent(struct work_struct
*work
);
321 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
322 static void check_for_release(struct cgroup
*cgrp
);
325 * A cgroup can be associated with multiple css_sets as different tasks may
326 * belong to different cgroups on different hierarchies. In the other
327 * direction, a css_set is naturally associated with multiple cgroups.
328 * This M:N relationship is represented by the following link structure
329 * which exists for each association and allows traversing the associations
332 struct cgrp_cset_link
{
333 /* the cgroup and css_set this link associates */
335 struct css_set
*cset
;
337 /* list of cgrp_cset_links anchored at cgrp->cset_links */
338 struct list_head cset_link
;
340 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
341 struct list_head cgrp_link
;
345 * The default css_set - used by init and its children prior to any
346 * hierarchies being mounted. It contains a pointer to the root state
347 * for each subsystem. Also used to anchor the list of css_sets. Not
348 * reference-counted, to improve performance when child cgroups
349 * haven't been created.
351 static struct css_set init_css_set
= {
352 .refcount
= ATOMIC_INIT(1),
353 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
354 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
355 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
356 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
357 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
360 static int css_set_count
= 1; /* 1 for init_css_set */
363 * hash table for cgroup groups. This improves the performance to find
364 * an existing css_set. This hash doesn't (currently) take into
365 * account cgroups in empty hierarchies.
367 #define CSS_SET_HASH_BITS 7
368 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
370 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
372 unsigned long key
= 0UL;
373 struct cgroup_subsys
*ss
;
376 for_each_subsys(ss
, i
)
377 key
+= (unsigned long)css
[i
];
378 key
= (key
>> 16) ^ key
;
383 static void put_css_set_locked(struct css_set
*cset
, bool taskexit
)
385 struct cgrp_cset_link
*link
, *tmp_link
;
387 lockdep_assert_held(&css_set_rwsem
);
389 if (!atomic_dec_and_test(&cset
->refcount
))
392 /* This css_set is dead. unlink it and release cgroup refcounts */
393 hash_del(&cset
->hlist
);
396 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
397 struct cgroup
*cgrp
= link
->cgrp
;
399 list_del(&link
->cset_link
);
400 list_del(&link
->cgrp_link
);
402 /* @cgrp can't go away while we're holding css_set_rwsem */
403 if (list_empty(&cgrp
->cset_links
) && notify_on_release(cgrp
)) {
405 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
406 check_for_release(cgrp
);
412 kfree_rcu(cset
, rcu_head
);
415 static void put_css_set(struct css_set
*cset
, bool taskexit
)
418 * Ensure that the refcount doesn't hit zero while any readers
419 * can see it. Similar to atomic_dec_and_lock(), but for an
422 if (atomic_add_unless(&cset
->refcount
, -1, 1))
425 down_write(&css_set_rwsem
);
426 put_css_set_locked(cset
, taskexit
);
427 up_write(&css_set_rwsem
);
431 * refcounted get/put for css_set objects
433 static inline void get_css_set(struct css_set
*cset
)
435 atomic_inc(&cset
->refcount
);
439 * compare_css_sets - helper function for find_existing_css_set().
440 * @cset: candidate css_set being tested
441 * @old_cset: existing css_set for a task
442 * @new_cgrp: cgroup that's being entered by the task
443 * @template: desired set of css pointers in css_set (pre-calculated)
445 * Returns true if "cset" matches "old_cset" except for the hierarchy
446 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
448 static bool compare_css_sets(struct css_set
*cset
,
449 struct css_set
*old_cset
,
450 struct cgroup
*new_cgrp
,
451 struct cgroup_subsys_state
*template[])
453 struct list_head
*l1
, *l2
;
455 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
))) {
456 /* Not all subsystems matched */
461 * Compare cgroup pointers in order to distinguish between
462 * different cgroups in heirarchies with no subsystems. We
463 * could get by with just this check alone (and skip the
464 * memcmp above) but on most setups the memcmp check will
465 * avoid the need for this more expensive check on almost all
469 l1
= &cset
->cgrp_links
;
470 l2
= &old_cset
->cgrp_links
;
472 struct cgrp_cset_link
*link1
, *link2
;
473 struct cgroup
*cgrp1
, *cgrp2
;
477 /* See if we reached the end - both lists are equal length. */
478 if (l1
== &cset
->cgrp_links
) {
479 BUG_ON(l2
!= &old_cset
->cgrp_links
);
482 BUG_ON(l2
== &old_cset
->cgrp_links
);
484 /* Locate the cgroups associated with these links. */
485 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
486 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
489 /* Hierarchies should be linked in the same order. */
490 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
493 * If this hierarchy is the hierarchy of the cgroup
494 * that's changing, then we need to check that this
495 * css_set points to the new cgroup; if it's any other
496 * hierarchy, then this css_set should point to the
497 * same cgroup as the old css_set.
499 if (cgrp1
->root
== new_cgrp
->root
) {
500 if (cgrp1
!= new_cgrp
)
511 * find_existing_css_set - init css array and find the matching css_set
512 * @old_cset: the css_set that we're using before the cgroup transition
513 * @cgrp: the cgroup that we're moving into
514 * @template: out param for the new set of csses, should be clear on entry
516 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
518 struct cgroup_subsys_state
*template[])
520 struct cgroup_root
*root
= cgrp
->root
;
521 struct cgroup_subsys
*ss
;
522 struct css_set
*cset
;
527 * Build the set of subsystem state objects that we want to see in the
528 * new css_set. while subsystems can change globally, the entries here
529 * won't change, so no need for locking.
531 for_each_subsys(ss
, i
) {
532 if (root
->cgrp
.subsys_mask
& (1UL << i
)) {
533 /* Subsystem is in this hierarchy. So we want
534 * the subsystem state from the new
536 template[i
] = cgroup_css(cgrp
, ss
);
538 /* Subsystem is not in this hierarchy, so we
539 * don't want to change the subsystem state */
540 template[i
] = old_cset
->subsys
[i
];
544 key
= css_set_hash(template);
545 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
546 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
549 /* This css_set matches what we need */
553 /* No existing cgroup group matched */
557 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
559 struct cgrp_cset_link
*link
, *tmp_link
;
561 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
562 list_del(&link
->cset_link
);
568 * allocate_cgrp_cset_links - allocate cgrp_cset_links
569 * @count: the number of links to allocate
570 * @tmp_links: list_head the allocated links are put on
572 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
573 * through ->cset_link. Returns 0 on success or -errno.
575 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
577 struct cgrp_cset_link
*link
;
580 INIT_LIST_HEAD(tmp_links
);
582 for (i
= 0; i
< count
; i
++) {
583 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
585 free_cgrp_cset_links(tmp_links
);
588 list_add(&link
->cset_link
, tmp_links
);
594 * link_css_set - a helper function to link a css_set to a cgroup
595 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
596 * @cset: the css_set to be linked
597 * @cgrp: the destination cgroup
599 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
602 struct cgrp_cset_link
*link
;
604 BUG_ON(list_empty(tmp_links
));
605 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
608 list_move(&link
->cset_link
, &cgrp
->cset_links
);
610 * Always add links to the tail of the list so that the list
611 * is sorted by order of hierarchy creation
613 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
617 * find_css_set - return a new css_set with one cgroup updated
618 * @old_cset: the baseline css_set
619 * @cgrp: the cgroup to be updated
621 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
622 * substituted into the appropriate hierarchy.
624 static struct css_set
*find_css_set(struct css_set
*old_cset
,
627 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
628 struct css_set
*cset
;
629 struct list_head tmp_links
;
630 struct cgrp_cset_link
*link
;
633 lockdep_assert_held(&cgroup_mutex
);
635 /* First see if we already have a cgroup group that matches
637 down_read(&css_set_rwsem
);
638 cset
= find_existing_css_set(old_cset
, cgrp
, template);
641 up_read(&css_set_rwsem
);
646 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
650 /* Allocate all the cgrp_cset_link objects that we'll need */
651 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
656 atomic_set(&cset
->refcount
, 1);
657 INIT_LIST_HEAD(&cset
->cgrp_links
);
658 INIT_LIST_HEAD(&cset
->tasks
);
659 INIT_LIST_HEAD(&cset
->mg_tasks
);
660 INIT_LIST_HEAD(&cset
->mg_preload_node
);
661 INIT_LIST_HEAD(&cset
->mg_node
);
662 INIT_HLIST_NODE(&cset
->hlist
);
664 /* Copy the set of subsystem state objects generated in
665 * find_existing_css_set() */
666 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
668 down_write(&css_set_rwsem
);
669 /* Add reference counts and links from the new css_set. */
670 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
671 struct cgroup
*c
= link
->cgrp
;
673 if (c
->root
== cgrp
->root
)
675 link_css_set(&tmp_links
, cset
, c
);
678 BUG_ON(!list_empty(&tmp_links
));
682 /* Add this cgroup group to the hash table */
683 key
= css_set_hash(cset
->subsys
);
684 hash_add(css_set_table
, &cset
->hlist
, key
);
686 up_write(&css_set_rwsem
);
691 static struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
693 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
695 return root_cgrp
->root
;
698 static int cgroup_init_root_id(struct cgroup_root
*root
)
702 lockdep_assert_held(&cgroup_mutex
);
704 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
708 root
->hierarchy_id
= id
;
712 static void cgroup_exit_root_id(struct cgroup_root
*root
)
714 lockdep_assert_held(&cgroup_mutex
);
716 if (root
->hierarchy_id
) {
717 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
718 root
->hierarchy_id
= 0;
722 static void cgroup_free_root(struct cgroup_root
*root
)
725 /* hierarhcy ID shoulid already have been released */
726 WARN_ON_ONCE(root
->hierarchy_id
);
728 idr_destroy(&root
->cgroup_idr
);
733 static void cgroup_destroy_root(struct cgroup_root
*root
)
735 struct cgroup
*cgrp
= &root
->cgrp
;
736 struct cgrp_cset_link
*link
, *tmp_link
;
738 mutex_lock(&cgroup_tree_mutex
);
739 mutex_lock(&cgroup_mutex
);
741 BUG_ON(atomic_read(&root
->nr_cgrps
));
742 BUG_ON(!list_empty(&cgrp
->children
));
744 /* Rebind all subsystems back to the default hierarchy */
745 rebind_subsystems(&cgrp_dfl_root
, cgrp
->subsys_mask
);
748 * Release all the links from cset_links to this hierarchy's
751 down_write(&css_set_rwsem
);
753 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
754 list_del(&link
->cset_link
);
755 list_del(&link
->cgrp_link
);
758 up_write(&css_set_rwsem
);
760 if (!list_empty(&root
->root_list
)) {
761 list_del(&root
->root_list
);
765 cgroup_exit_root_id(root
);
767 mutex_unlock(&cgroup_mutex
);
768 mutex_unlock(&cgroup_tree_mutex
);
770 kernfs_destroy_root(root
->kf_root
);
771 cgroup_free_root(root
);
774 /* look up cgroup associated with given css_set on the specified hierarchy */
775 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
776 struct cgroup_root
*root
)
778 struct cgroup
*res
= NULL
;
780 lockdep_assert_held(&cgroup_mutex
);
781 lockdep_assert_held(&css_set_rwsem
);
783 if (cset
== &init_css_set
) {
786 struct cgrp_cset_link
*link
;
788 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
789 struct cgroup
*c
= link
->cgrp
;
791 if (c
->root
== root
) {
803 * Return the cgroup for "task" from the given hierarchy. Must be
804 * called with cgroup_mutex and css_set_rwsem held.
806 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
807 struct cgroup_root
*root
)
810 * No need to lock the task - since we hold cgroup_mutex the
811 * task can't change groups, so the only thing that can happen
812 * is that it exits and its css is set back to init_css_set.
814 return cset_cgroup_from_root(task_css_set(task
), root
);
818 * A task must hold cgroup_mutex to modify cgroups.
820 * Any task can increment and decrement the count field without lock.
821 * So in general, code holding cgroup_mutex can't rely on the count
822 * field not changing. However, if the count goes to zero, then only
823 * cgroup_attach_task() can increment it again. Because a count of zero
824 * means that no tasks are currently attached, therefore there is no
825 * way a task attached to that cgroup can fork (the other way to
826 * increment the count). So code holding cgroup_mutex can safely
827 * assume that if the count is zero, it will stay zero. Similarly, if
828 * a task holds cgroup_mutex on a cgroup with zero count, it
829 * knows that the cgroup won't be removed, as cgroup_rmdir()
832 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
833 * (usually) take cgroup_mutex. These are the two most performance
834 * critical pieces of code here. The exception occurs on cgroup_exit(),
835 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
836 * is taken, and if the cgroup count is zero, a usermode call made
837 * to the release agent with the name of the cgroup (path relative to
838 * the root of cgroup file system) as the argument.
840 * A cgroup can only be deleted if both its 'count' of using tasks
841 * is zero, and its list of 'children' cgroups is empty. Since all
842 * tasks in the system use _some_ cgroup, and since there is always at
843 * least one task in the system (init, pid == 1), therefore, root cgroup
844 * always has either children cgroups and/or using tasks. So we don't
845 * need a special hack to ensure that root cgroup cannot be deleted.
847 * P.S. One more locking exception. RCU is used to guard the
848 * update of a tasks cgroup pointer by cgroup_attach_task()
851 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
);
852 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
853 static const struct file_operations proc_cgroupstats_operations
;
855 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
858 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
859 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
))
860 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s.%s",
861 cft
->ss
->name
, cft
->name
);
863 strncpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
868 * cgroup_file_mode - deduce file mode of a control file
869 * @cft: the control file in question
871 * returns cft->mode if ->mode is not 0
872 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
873 * returns S_IRUGO if it has only a read handler
874 * returns S_IWUSR if it has only a write hander
876 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
883 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
886 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write_string
||
893 static void cgroup_free_fn(struct work_struct
*work
)
895 struct cgroup
*cgrp
= container_of(work
, struct cgroup
, destroy_work
);
897 atomic_dec(&cgrp
->root
->nr_cgrps
);
898 cgroup_pidlist_destroy_all(cgrp
);
902 * We get a ref to the parent, and put the ref when this
903 * cgroup is being freed, so it's guaranteed that the
904 * parent won't be destroyed before its children.
906 cgroup_put(cgrp
->parent
);
907 kernfs_put(cgrp
->kn
);
911 * This is root cgroup's refcnt reaching zero, which
912 * indicates that the root should be released.
914 cgroup_destroy_root(cgrp
->root
);
918 static void cgroup_free_rcu(struct rcu_head
*head
)
920 struct cgroup
*cgrp
= container_of(head
, struct cgroup
, rcu_head
);
922 INIT_WORK(&cgrp
->destroy_work
, cgroup_free_fn
);
923 queue_work(cgroup_destroy_wq
, &cgrp
->destroy_work
);
926 static void cgroup_get(struct cgroup
*cgrp
)
928 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
929 WARN_ON_ONCE(atomic_read(&cgrp
->refcnt
) <= 0);
930 atomic_inc(&cgrp
->refcnt
);
933 static void cgroup_put(struct cgroup
*cgrp
)
935 if (!atomic_dec_and_test(&cgrp
->refcnt
))
937 if (WARN_ON_ONCE(cgrp
->parent
&& !cgroup_is_dead(cgrp
)))
941 * XXX: cgrp->id is only used to look up css's. As cgroup and
942 * css's lifetimes will be decoupled, it should be made
943 * per-subsystem and moved to css->id so that lookups are
944 * successful until the target css is released.
946 mutex_lock(&cgroup_mutex
);
947 idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
948 mutex_unlock(&cgroup_mutex
);
951 call_rcu(&cgrp
->rcu_head
, cgroup_free_rcu
);
954 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
956 char name
[CGROUP_FILE_NAME_MAX
];
958 lockdep_assert_held(&cgroup_tree_mutex
);
959 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
963 * cgroup_clear_dir - remove subsys files in a cgroup directory
964 * @cgrp: target cgroup
965 * @subsys_mask: mask of the subsystem ids whose files should be removed
967 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
969 struct cgroup_subsys
*ss
;
972 for_each_subsys(ss
, i
) {
975 if (!test_bit(i
, &subsys_mask
))
977 list_for_each_entry(cfts
, &ss
->cfts
, node
)
978 cgroup_addrm_files(cgrp
, cfts
, false);
982 static int rebind_subsystems(struct cgroup_root
*dst_root
,
983 unsigned long ss_mask
)
985 struct cgroup_subsys
*ss
;
988 lockdep_assert_held(&cgroup_tree_mutex
);
989 lockdep_assert_held(&cgroup_mutex
);
991 for_each_subsys(ss
, ssid
) {
992 if (!(ss_mask
& (1 << ssid
)))
995 /* if @ss is on the dummy_root, we can always move it */
996 if (ss
->root
== &cgrp_dfl_root
)
999 /* if @ss has non-root cgroups attached to it, can't move */
1000 if (!list_empty(&ss
->root
->cgrp
.children
))
1003 /* can't move between two non-dummy roots either */
1004 if (dst_root
!= &cgrp_dfl_root
)
1008 ret
= cgroup_populate_dir(&dst_root
->cgrp
, ss_mask
);
1010 if (dst_root
!= &cgrp_dfl_root
)
1014 * Rebinding back to the default root is not allowed to
1015 * fail. Using both default and non-default roots should
1016 * be rare. Moving subsystems back and forth even more so.
1017 * Just warn about it and continue.
1019 if (cgrp_dfl_root_visible
) {
1020 pr_warning("cgroup: failed to create files (%d) while rebinding 0x%lx to default root\n",
1022 pr_warning("cgroup: you may retry by moving them to a different hierarchy and unbinding\n");
1027 * Nothing can fail from this point on. Remove files for the
1028 * removed subsystems and rebind each subsystem.
1030 mutex_unlock(&cgroup_mutex
);
1031 for_each_subsys(ss
, ssid
)
1032 if (ss_mask
& (1 << ssid
))
1033 cgroup_clear_dir(&ss
->root
->cgrp
, 1 << ssid
);
1034 mutex_lock(&cgroup_mutex
);
1036 for_each_subsys(ss
, ssid
) {
1037 struct cgroup_root
*src_root
;
1038 struct cgroup_subsys_state
*css
;
1040 if (!(ss_mask
& (1 << ssid
)))
1043 src_root
= ss
->root
;
1044 css
= cgroup_css(&src_root
->cgrp
, ss
);
1046 WARN_ON(!css
|| cgroup_css(&dst_root
->cgrp
, ss
));
1048 RCU_INIT_POINTER(src_root
->cgrp
.subsys
[ssid
], NULL
);
1049 rcu_assign_pointer(dst_root
->cgrp
.subsys
[ssid
], css
);
1050 ss
->root
= dst_root
;
1051 css
->cgroup
= &dst_root
->cgrp
;
1053 src_root
->cgrp
.subsys_mask
&= ~(1 << ssid
);
1054 dst_root
->cgrp
.subsys_mask
|= 1 << ssid
;
1060 kernfs_activate(dst_root
->cgrp
.kn
);
1064 static int cgroup_show_options(struct seq_file
*seq
,
1065 struct kernfs_root
*kf_root
)
1067 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1068 struct cgroup_subsys
*ss
;
1071 for_each_subsys(ss
, ssid
)
1072 if (root
->cgrp
.subsys_mask
& (1 << ssid
))
1073 seq_printf(seq
, ",%s", ss
->name
);
1074 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
)
1075 seq_puts(seq
, ",sane_behavior");
1076 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1077 seq_puts(seq
, ",noprefix");
1078 if (root
->flags
& CGRP_ROOT_XATTR
)
1079 seq_puts(seq
, ",xattr");
1081 spin_lock(&release_agent_path_lock
);
1082 if (strlen(root
->release_agent_path
))
1083 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1084 spin_unlock(&release_agent_path_lock
);
1086 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
1087 seq_puts(seq
, ",clone_children");
1088 if (strlen(root
->name
))
1089 seq_printf(seq
, ",name=%s", root
->name
);
1093 struct cgroup_sb_opts
{
1094 unsigned long subsys_mask
;
1095 unsigned long flags
;
1096 char *release_agent
;
1097 bool cpuset_clone_children
;
1099 /* User explicitly requested empty subsystem */
1104 * Convert a hierarchy specifier into a bitmask of subsystems and
1105 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1106 * array. This function takes refcounts on subsystems to be used, unless it
1107 * returns error, in which case no refcounts are taken.
1109 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1111 char *token
, *o
= data
;
1112 bool all_ss
= false, one_ss
= false;
1113 unsigned long mask
= (unsigned long)-1;
1114 struct cgroup_subsys
*ss
;
1117 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1119 #ifdef CONFIG_CPUSETS
1120 mask
= ~(1UL << cpuset_cgrp_id
);
1123 memset(opts
, 0, sizeof(*opts
));
1125 while ((token
= strsep(&o
, ",")) != NULL
) {
1128 if (!strcmp(token
, "none")) {
1129 /* Explicitly have no subsystems */
1133 if (!strcmp(token
, "all")) {
1134 /* Mutually exclusive option 'all' + subsystem name */
1140 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1141 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1144 if (!strcmp(token
, "noprefix")) {
1145 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1148 if (!strcmp(token
, "clone_children")) {
1149 opts
->cpuset_clone_children
= true;
1152 if (!strcmp(token
, "xattr")) {
1153 opts
->flags
|= CGRP_ROOT_XATTR
;
1156 if (!strncmp(token
, "release_agent=", 14)) {
1157 /* Specifying two release agents is forbidden */
1158 if (opts
->release_agent
)
1160 opts
->release_agent
=
1161 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1162 if (!opts
->release_agent
)
1166 if (!strncmp(token
, "name=", 5)) {
1167 const char *name
= token
+ 5;
1168 /* Can't specify an empty name */
1171 /* Must match [\w.-]+ */
1172 for (i
= 0; i
< strlen(name
); i
++) {
1176 if ((c
== '.') || (c
== '-') || (c
== '_'))
1180 /* Specifying two names is forbidden */
1183 opts
->name
= kstrndup(name
,
1184 MAX_CGROUP_ROOT_NAMELEN
- 1,
1192 for_each_subsys(ss
, i
) {
1193 if (strcmp(token
, ss
->name
))
1198 /* Mutually exclusive option 'all' + subsystem name */
1201 set_bit(i
, &opts
->subsys_mask
);
1206 if (i
== CGROUP_SUBSYS_COUNT
)
1210 /* Consistency checks */
1212 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1213 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1215 if ((opts
->flags
& (CGRP_ROOT_NOPREFIX
| CGRP_ROOT_XATTR
)) ||
1216 opts
->cpuset_clone_children
|| opts
->release_agent
||
1218 pr_err("cgroup: sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n");
1223 * If the 'all' option was specified select all the
1224 * subsystems, otherwise if 'none', 'name=' and a subsystem
1225 * name options were not specified, let's default to 'all'
1227 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1228 for_each_subsys(ss
, i
)
1230 set_bit(i
, &opts
->subsys_mask
);
1233 * We either have to specify by name or by subsystems. (So
1234 * all empty hierarchies must have a name).
1236 if (!opts
->subsys_mask
&& !opts
->name
)
1241 * Option noprefix was introduced just for backward compatibility
1242 * with the old cpuset, so we allow noprefix only if mounting just
1243 * the cpuset subsystem.
1245 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1249 /* Can't specify "none" and some subsystems */
1250 if (opts
->subsys_mask
&& opts
->none
)
1256 static int cgroup_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1259 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1260 struct cgroup_sb_opts opts
;
1261 unsigned long added_mask
, removed_mask
;
1263 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1264 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1268 mutex_lock(&cgroup_tree_mutex
);
1269 mutex_lock(&cgroup_mutex
);
1271 /* See what subsystems are wanted */
1272 ret
= parse_cgroupfs_options(data
, &opts
);
1276 if (opts
.subsys_mask
!= root
->cgrp
.subsys_mask
|| opts
.release_agent
)
1277 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1278 task_tgid_nr(current
), current
->comm
);
1280 added_mask
= opts
.subsys_mask
& ~root
->cgrp
.subsys_mask
;
1281 removed_mask
= root
->cgrp
.subsys_mask
& ~opts
.subsys_mask
;
1283 /* Don't allow flags or name to change at remount */
1284 if (((opts
.flags
^ root
->flags
) & CGRP_ROOT_OPTION_MASK
) ||
1285 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1286 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1287 opts
.flags
& CGRP_ROOT_OPTION_MASK
, opts
.name
?: "",
1288 root
->flags
& CGRP_ROOT_OPTION_MASK
, root
->name
);
1293 /* remounting is not allowed for populated hierarchies */
1294 if (!list_empty(&root
->cgrp
.children
)) {
1299 ret
= rebind_subsystems(root
, added_mask
);
1303 rebind_subsystems(&cgrp_dfl_root
, removed_mask
);
1305 if (opts
.release_agent
) {
1306 spin_lock(&release_agent_path_lock
);
1307 strcpy(root
->release_agent_path
, opts
.release_agent
);
1308 spin_unlock(&release_agent_path_lock
);
1311 kfree(opts
.release_agent
);
1313 mutex_unlock(&cgroup_mutex
);
1314 mutex_unlock(&cgroup_tree_mutex
);
1319 * To reduce the fork() overhead for systems that are not actually using
1320 * their cgroups capability, we don't maintain the lists running through
1321 * each css_set to its tasks until we see the list actually used - in other
1322 * words after the first mount.
1324 static bool use_task_css_set_links __read_mostly
;
1326 static void cgroup_enable_task_cg_lists(void)
1328 struct task_struct
*p
, *g
;
1330 down_write(&css_set_rwsem
);
1332 if (use_task_css_set_links
)
1335 use_task_css_set_links
= true;
1338 * We need tasklist_lock because RCU is not safe against
1339 * while_each_thread(). Besides, a forking task that has passed
1340 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1341 * is not guaranteed to have its child immediately visible in the
1342 * tasklist if we walk through it with RCU.
1344 read_lock(&tasklist_lock
);
1345 do_each_thread(g
, p
) {
1346 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1347 task_css_set(p
) != &init_css_set
);
1350 * We should check if the process is exiting, otherwise
1351 * it will race with cgroup_exit() in that the list
1352 * entry won't be deleted though the process has exited.
1353 * Do it while holding siglock so that we don't end up
1354 * racing against cgroup_exit().
1356 spin_lock_irq(&p
->sighand
->siglock
);
1357 if (!(p
->flags
& PF_EXITING
)) {
1358 struct css_set
*cset
= task_css_set(p
);
1360 list_add(&p
->cg_list
, &cset
->tasks
);
1363 spin_unlock_irq(&p
->sighand
->siglock
);
1364 } while_each_thread(g
, p
);
1365 read_unlock(&tasklist_lock
);
1367 up_write(&css_set_rwsem
);
1370 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1372 atomic_set(&cgrp
->refcnt
, 1);
1373 INIT_LIST_HEAD(&cgrp
->sibling
);
1374 INIT_LIST_HEAD(&cgrp
->children
);
1375 INIT_LIST_HEAD(&cgrp
->cset_links
);
1376 INIT_LIST_HEAD(&cgrp
->release_list
);
1377 INIT_LIST_HEAD(&cgrp
->pidlists
);
1378 mutex_init(&cgrp
->pidlist_mutex
);
1379 cgrp
->dummy_css
.cgroup
= cgrp
;
1382 static void init_cgroup_root(struct cgroup_root
*root
,
1383 struct cgroup_sb_opts
*opts
)
1385 struct cgroup
*cgrp
= &root
->cgrp
;
1387 INIT_LIST_HEAD(&root
->root_list
);
1388 atomic_set(&root
->nr_cgrps
, 1);
1390 init_cgroup_housekeeping(cgrp
);
1391 idr_init(&root
->cgroup_idr
);
1393 root
->flags
= opts
->flags
;
1394 if (opts
->release_agent
)
1395 strcpy(root
->release_agent_path
, opts
->release_agent
);
1397 strcpy(root
->name
, opts
->name
);
1398 if (opts
->cpuset_clone_children
)
1399 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1402 static int cgroup_setup_root(struct cgroup_root
*root
, unsigned long ss_mask
)
1404 LIST_HEAD(tmp_links
);
1405 struct cgroup
*root_cgrp
= &root
->cgrp
;
1406 struct css_set
*cset
;
1409 lockdep_assert_held(&cgroup_tree_mutex
);
1410 lockdep_assert_held(&cgroup_mutex
);
1412 ret
= idr_alloc(&root
->cgroup_idr
, root_cgrp
, 0, 1, GFP_KERNEL
);
1415 root_cgrp
->id
= ret
;
1418 * We're accessing css_set_count without locking css_set_rwsem here,
1419 * but that's OK - it can only be increased by someone holding
1420 * cgroup_lock, and that's us. The worst that can happen is that we
1421 * have some link structures left over
1423 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1427 ret
= cgroup_init_root_id(root
);
1431 root
->kf_root
= kernfs_create_root(&cgroup_kf_syscall_ops
,
1432 KERNFS_ROOT_CREATE_DEACTIVATED
,
1434 if (IS_ERR(root
->kf_root
)) {
1435 ret
= PTR_ERR(root
->kf_root
);
1438 root_cgrp
->kn
= root
->kf_root
->kn
;
1440 ret
= cgroup_addrm_files(root_cgrp
, cgroup_base_files
, true);
1444 ret
= rebind_subsystems(root
, ss_mask
);
1449 * There must be no failure case after here, since rebinding takes
1450 * care of subsystems' refcounts, which are explicitly dropped in
1451 * the failure exit path.
1453 list_add(&root
->root_list
, &cgroup_roots
);
1454 cgroup_root_count
++;
1457 * Link the root cgroup in this hierarchy into all the css_set
1460 down_write(&css_set_rwsem
);
1461 hash_for_each(css_set_table
, i
, cset
, hlist
)
1462 link_css_set(&tmp_links
, cset
, root_cgrp
);
1463 up_write(&css_set_rwsem
);
1465 BUG_ON(!list_empty(&root_cgrp
->children
));
1466 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
1468 kernfs_activate(root_cgrp
->kn
);
1473 kernfs_destroy_root(root
->kf_root
);
1474 root
->kf_root
= NULL
;
1476 cgroup_exit_root_id(root
);
1478 free_cgrp_cset_links(&tmp_links
);
1482 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1483 int flags
, const char *unused_dev_name
,
1486 struct cgroup_root
*root
;
1487 struct cgroup_sb_opts opts
;
1488 struct dentry
*dentry
;
1492 * The first time anyone tries to mount a cgroup, enable the list
1493 * linking each css_set to its tasks and fix up all existing tasks.
1495 if (!use_task_css_set_links
)
1496 cgroup_enable_task_cg_lists();
1498 mutex_lock(&cgroup_tree_mutex
);
1499 mutex_lock(&cgroup_mutex
);
1501 /* First find the desired set of subsystems */
1502 ret
= parse_cgroupfs_options(data
, &opts
);
1506 /* look for a matching existing root */
1507 if (!opts
.subsys_mask
&& !opts
.none
&& !opts
.name
) {
1508 cgrp_dfl_root_visible
= true;
1509 root
= &cgrp_dfl_root
;
1510 cgroup_get(&root
->cgrp
);
1515 for_each_root(root
) {
1516 bool name_match
= false;
1518 if (root
== &cgrp_dfl_root
)
1522 * If we asked for a name then it must match. Also, if
1523 * name matches but sybsys_mask doesn't, we should fail.
1524 * Remember whether name matched.
1527 if (strcmp(opts
.name
, root
->name
))
1533 * If we asked for subsystems (or explicitly for no
1534 * subsystems) then they must match.
1536 if ((opts
.subsys_mask
|| opts
.none
) &&
1537 (opts
.subsys_mask
!= root
->cgrp
.subsys_mask
)) {
1544 if ((root
->flags
^ opts
.flags
) & CGRP_ROOT_OPTION_MASK
) {
1545 if ((root
->flags
| opts
.flags
) & CGRP_ROOT_SANE_BEHAVIOR
) {
1546 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1550 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1555 * A root's lifetime is governed by its root cgroup. Zero
1556 * ref indicate that the root is being destroyed. Wait for
1557 * destruction to complete so that the subsystems are free.
1558 * We can use wait_queue for the wait but this path is
1559 * super cold. Let's just sleep for a bit and retry.
1561 if (!atomic_inc_not_zero(&root
->cgrp
.refcnt
)) {
1562 mutex_unlock(&cgroup_mutex
);
1563 mutex_unlock(&cgroup_tree_mutex
);
1564 kfree(opts
.release_agent
);
1575 * No such thing, create a new one. name= matching without subsys
1576 * specification is allowed for already existing hierarchies but we
1577 * can't create new one without subsys specification.
1579 if (!opts
.subsys_mask
&& !opts
.none
) {
1584 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1590 init_cgroup_root(root
, &opts
);
1592 ret
= cgroup_setup_root(root
, opts
.subsys_mask
);
1594 cgroup_free_root(root
);
1597 mutex_unlock(&cgroup_mutex
);
1598 mutex_unlock(&cgroup_tree_mutex
);
1600 kfree(opts
.release_agent
);
1604 return ERR_PTR(ret
);
1606 dentry
= kernfs_mount(fs_type
, flags
, root
->kf_root
, NULL
);
1608 cgroup_put(&root
->cgrp
);
1612 static void cgroup_kill_sb(struct super_block
*sb
)
1614 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
1615 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1617 cgroup_put(&root
->cgrp
);
1621 static struct file_system_type cgroup_fs_type
= {
1623 .mount
= cgroup_mount
,
1624 .kill_sb
= cgroup_kill_sb
,
1627 static struct kobject
*cgroup_kobj
;
1630 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1631 * @task: target task
1632 * @buf: the buffer to write the path into
1633 * @buflen: the length of the buffer
1635 * Determine @task's cgroup on the first (the one with the lowest non-zero
1636 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1637 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1638 * cgroup controller callbacks.
1640 * Return value is the same as kernfs_path().
1642 char *task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
1644 struct cgroup_root
*root
;
1645 struct cgroup
*cgrp
;
1646 int hierarchy_id
= 1;
1649 mutex_lock(&cgroup_mutex
);
1650 down_read(&css_set_rwsem
);
1652 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
1655 cgrp
= task_cgroup_from_root(task
, root
);
1656 path
= cgroup_path(cgrp
, buf
, buflen
);
1658 /* if no hierarchy exists, everyone is in "/" */
1659 if (strlcpy(buf
, "/", buflen
) < buflen
)
1663 up_read(&css_set_rwsem
);
1664 mutex_unlock(&cgroup_mutex
);
1667 EXPORT_SYMBOL_GPL(task_cgroup_path
);
1669 /* used to track tasks and other necessary states during migration */
1670 struct cgroup_taskset
{
1671 /* the src and dst cset list running through cset->mg_node */
1672 struct list_head src_csets
;
1673 struct list_head dst_csets
;
1676 * Fields for cgroup_taskset_*() iteration.
1678 * Before migration is committed, the target migration tasks are on
1679 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
1680 * the csets on ->dst_csets. ->csets point to either ->src_csets
1681 * or ->dst_csets depending on whether migration is committed.
1683 * ->cur_csets and ->cur_task point to the current task position
1686 struct list_head
*csets
;
1687 struct css_set
*cur_cset
;
1688 struct task_struct
*cur_task
;
1692 * cgroup_taskset_first - reset taskset and return the first task
1693 * @tset: taskset of interest
1695 * @tset iteration is initialized and the first task is returned.
1697 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1699 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
1700 tset
->cur_task
= NULL
;
1702 return cgroup_taskset_next(tset
);
1706 * cgroup_taskset_next - iterate to the next task in taskset
1707 * @tset: taskset of interest
1709 * Return the next task in @tset. Iteration must have been initialized
1710 * with cgroup_taskset_first().
1712 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1714 struct css_set
*cset
= tset
->cur_cset
;
1715 struct task_struct
*task
= tset
->cur_task
;
1717 while (&cset
->mg_node
!= tset
->csets
) {
1719 task
= list_first_entry(&cset
->mg_tasks
,
1720 struct task_struct
, cg_list
);
1722 task
= list_next_entry(task
, cg_list
);
1724 if (&task
->cg_list
!= &cset
->mg_tasks
) {
1725 tset
->cur_cset
= cset
;
1726 tset
->cur_task
= task
;
1730 cset
= list_next_entry(cset
, mg_node
);
1738 * cgroup_task_migrate - move a task from one cgroup to another.
1739 * @old_cgrp; the cgroup @tsk is being migrated from
1740 * @tsk: the task being migrated
1741 * @new_cset: the new css_set @tsk is being attached to
1743 * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
1745 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
1746 struct task_struct
*tsk
,
1747 struct css_set
*new_cset
)
1749 struct css_set
*old_cset
;
1751 lockdep_assert_held(&cgroup_mutex
);
1752 lockdep_assert_held(&css_set_rwsem
);
1755 * We are synchronized through threadgroup_lock() against PF_EXITING
1756 * setting such that we can't race against cgroup_exit() changing the
1757 * css_set to init_css_set and dropping the old one.
1759 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1760 old_cset
= task_css_set(tsk
);
1762 get_css_set(new_cset
);
1763 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
1766 * Use move_tail so that cgroup_taskset_first() still returns the
1767 * leader after migration. This works because cgroup_migrate()
1768 * ensures that the dst_cset of the leader is the first on the
1769 * tset's dst_csets list.
1771 list_move_tail(&tsk
->cg_list
, &new_cset
->mg_tasks
);
1774 * We just gained a reference on old_cset by taking it from the
1775 * task. As trading it for new_cset is protected by cgroup_mutex,
1776 * we're safe to drop it here; it will be freed under RCU.
1778 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
1779 put_css_set_locked(old_cset
, false);
1783 * cgroup_migrate_finish - cleanup after attach
1784 * @preloaded_csets: list of preloaded css_sets
1786 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
1787 * those functions for details.
1789 static void cgroup_migrate_finish(struct list_head
*preloaded_csets
)
1791 struct css_set
*cset
, *tmp_cset
;
1793 lockdep_assert_held(&cgroup_mutex
);
1795 down_write(&css_set_rwsem
);
1796 list_for_each_entry_safe(cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
1797 cset
->mg_src_cgrp
= NULL
;
1798 cset
->mg_dst_cset
= NULL
;
1799 list_del_init(&cset
->mg_preload_node
);
1800 put_css_set_locked(cset
, false);
1802 up_write(&css_set_rwsem
);
1806 * cgroup_migrate_add_src - add a migration source css_set
1807 * @src_cset: the source css_set to add
1808 * @dst_cgrp: the destination cgroup
1809 * @preloaded_csets: list of preloaded css_sets
1811 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
1812 * @src_cset and add it to @preloaded_csets, which should later be cleaned
1813 * up by cgroup_migrate_finish().
1815 * This function may be called without holding threadgroup_lock even if the
1816 * target is a process. Threads may be created and destroyed but as long
1817 * as cgroup_mutex is not dropped, no new css_set can be put into play and
1818 * the preloaded css_sets are guaranteed to cover all migrations.
1820 static void cgroup_migrate_add_src(struct css_set
*src_cset
,
1821 struct cgroup
*dst_cgrp
,
1822 struct list_head
*preloaded_csets
)
1824 struct cgroup
*src_cgrp
;
1826 lockdep_assert_held(&cgroup_mutex
);
1827 lockdep_assert_held(&css_set_rwsem
);
1829 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
1831 /* nothing to do if this cset already belongs to the cgroup */
1832 if (src_cgrp
== dst_cgrp
)
1835 if (!list_empty(&src_cset
->mg_preload_node
))
1838 WARN_ON(src_cset
->mg_src_cgrp
);
1839 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
1840 WARN_ON(!list_empty(&src_cset
->mg_node
));
1842 src_cset
->mg_src_cgrp
= src_cgrp
;
1843 get_css_set(src_cset
);
1844 list_add(&src_cset
->mg_preload_node
, preloaded_csets
);
1848 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
1849 * @dst_cgrp: the destination cgroup
1850 * @preloaded_csets: list of preloaded source css_sets
1852 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
1853 * have been preloaded to @preloaded_csets. This function looks up and
1854 * pins all destination css_sets, links each to its source, and put them on
1857 * This function must be called after cgroup_migrate_add_src() has been
1858 * called on each migration source css_set. After migration is performed
1859 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
1862 static int cgroup_migrate_prepare_dst(struct cgroup
*dst_cgrp
,
1863 struct list_head
*preloaded_csets
)
1866 struct css_set
*src_cset
;
1868 lockdep_assert_held(&cgroup_mutex
);
1870 /* look up the dst cset for each src cset and link it to src */
1871 list_for_each_entry(src_cset
, preloaded_csets
, mg_preload_node
) {
1872 struct css_set
*dst_cset
;
1874 dst_cset
= find_css_set(src_cset
, dst_cgrp
);
1878 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
1879 src_cset
->mg_dst_cset
= dst_cset
;
1881 if (list_empty(&dst_cset
->mg_preload_node
))
1882 list_add(&dst_cset
->mg_preload_node
, &csets
);
1884 put_css_set(dst_cset
, false);
1887 list_splice(&csets
, preloaded_csets
);
1890 cgroup_migrate_finish(&csets
);
1895 * cgroup_migrate - migrate a process or task to a cgroup
1896 * @cgrp: the destination cgroup
1897 * @leader: the leader of the process or the task to migrate
1898 * @threadgroup: whether @leader points to the whole process or a single task
1900 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
1901 * process, the caller must be holding threadgroup_lock of @leader. The
1902 * caller is also responsible for invoking cgroup_migrate_add_src() and
1903 * cgroup_migrate_prepare_dst() on the targets before invoking this
1904 * function and following up with cgroup_migrate_finish().
1906 * As long as a controller's ->can_attach() doesn't fail, this function is
1907 * guaranteed to succeed. This means that, excluding ->can_attach()
1908 * failure, when migrating multiple targets, the success or failure can be
1909 * decided for all targets by invoking group_migrate_prepare_dst() before
1910 * actually starting migrating.
1912 static int cgroup_migrate(struct cgroup
*cgrp
, struct task_struct
*leader
,
1915 struct cgroup_taskset tset
= {
1916 .src_csets
= LIST_HEAD_INIT(tset
.src_csets
),
1917 .dst_csets
= LIST_HEAD_INIT(tset
.dst_csets
),
1918 .csets
= &tset
.src_csets
,
1920 struct cgroup_subsys_state
*css
, *failed_css
= NULL
;
1921 struct css_set
*cset
, *tmp_cset
;
1922 struct task_struct
*task
, *tmp_task
;
1926 * Prevent freeing of tasks while we take a snapshot. Tasks that are
1927 * already PF_EXITING could be freed from underneath us unless we
1928 * take an rcu_read_lock.
1930 down_write(&css_set_rwsem
);
1934 /* @task either already exited or can't exit until the end */
1935 if (task
->flags
& PF_EXITING
)
1938 /* leave @task alone if post_fork() hasn't linked it yet */
1939 if (list_empty(&task
->cg_list
))
1942 cset
= task_css_set(task
);
1943 if (!cset
->mg_src_cgrp
)
1947 * cgroup_taskset_first() must always return the leader.
1948 * Take care to avoid disturbing the ordering.
1950 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
1951 if (list_empty(&cset
->mg_node
))
1952 list_add_tail(&cset
->mg_node
, &tset
.src_csets
);
1953 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
1954 list_move_tail(&cset
->mg_dst_cset
->mg_node
,
1959 } while_each_thread(leader
, task
);
1961 up_write(&css_set_rwsem
);
1963 /* methods shouldn't be called if no task is actually migrating */
1964 if (list_empty(&tset
.src_csets
))
1967 /* check that we can legitimately attach to the cgroup */
1968 for_each_css(css
, i
, cgrp
) {
1969 if (css
->ss
->can_attach
) {
1970 ret
= css
->ss
->can_attach(css
, &tset
);
1973 goto out_cancel_attach
;
1979 * Now that we're guaranteed success, proceed to move all tasks to
1980 * the new cgroup. There are no failure cases after here, so this
1981 * is the commit point.
1983 down_write(&css_set_rwsem
);
1984 list_for_each_entry(cset
, &tset
.src_csets
, mg_node
) {
1985 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
)
1986 cgroup_task_migrate(cset
->mg_src_cgrp
, task
,
1989 up_write(&css_set_rwsem
);
1992 * Migration is committed, all target tasks are now on dst_csets.
1993 * Nothing is sensitive to fork() after this point. Notify
1994 * controllers that migration is complete.
1996 tset
.csets
= &tset
.dst_csets
;
1998 for_each_css(css
, i
, cgrp
)
1999 if (css
->ss
->attach
)
2000 css
->ss
->attach(css
, &tset
);
2003 goto out_release_tset
;
2006 for_each_css(css
, i
, cgrp
) {
2007 if (css
== failed_css
)
2009 if (css
->ss
->cancel_attach
)
2010 css
->ss
->cancel_attach(css
, &tset
);
2013 down_write(&css_set_rwsem
);
2014 list_splice_init(&tset
.dst_csets
, &tset
.src_csets
);
2015 list_for_each_entry_safe(cset
, tmp_cset
, &tset
.src_csets
, mg_node
) {
2016 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2017 list_del_init(&cset
->mg_node
);
2019 up_write(&css_set_rwsem
);
2024 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2025 * @dst_cgrp: the cgroup to attach to
2026 * @leader: the task or the leader of the threadgroup to be attached
2027 * @threadgroup: attach the whole threadgroup?
2029 * Call holding cgroup_mutex and threadgroup_lock of @leader.
2031 static int cgroup_attach_task(struct cgroup
*dst_cgrp
,
2032 struct task_struct
*leader
, bool threadgroup
)
2034 LIST_HEAD(preloaded_csets
);
2035 struct task_struct
*task
;
2038 /* look up all src csets */
2039 down_read(&css_set_rwsem
);
2043 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
,
2047 } while_each_thread(leader
, task
);
2049 up_read(&css_set_rwsem
);
2051 /* prepare dst csets and commit */
2052 ret
= cgroup_migrate_prepare_dst(dst_cgrp
, &preloaded_csets
);
2054 ret
= cgroup_migrate(dst_cgrp
, leader
, threadgroup
);
2056 cgroup_migrate_finish(&preloaded_csets
);
2061 * Find the task_struct of the task to attach by vpid and pass it along to the
2062 * function to attach either it or all tasks in its threadgroup. Will lock
2063 * cgroup_mutex and threadgroup.
2065 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2067 struct task_struct
*tsk
;
2068 const struct cred
*cred
= current_cred(), *tcred
;
2071 if (!cgroup_lock_live_group(cgrp
))
2077 tsk
= find_task_by_vpid(pid
);
2081 goto out_unlock_cgroup
;
2084 * even if we're attaching all tasks in the thread group, we
2085 * only need to check permissions on one of them.
2087 tcred
= __task_cred(tsk
);
2088 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2089 !uid_eq(cred
->euid
, tcred
->uid
) &&
2090 !uid_eq(cred
->euid
, tcred
->suid
)) {
2093 goto out_unlock_cgroup
;
2099 tsk
= tsk
->group_leader
;
2102 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2103 * trapped in a cpuset, or RT worker may be born in a cgroup
2104 * with no rt_runtime allocated. Just say no.
2106 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2109 goto out_unlock_cgroup
;
2112 get_task_struct(tsk
);
2115 threadgroup_lock(tsk
);
2117 if (!thread_group_leader(tsk
)) {
2119 * a race with de_thread from another thread's exec()
2120 * may strip us of our leadership, if this happens,
2121 * there is no choice but to throw this task away and
2122 * try again; this is
2123 * "double-double-toil-and-trouble-check locking".
2125 threadgroup_unlock(tsk
);
2126 put_task_struct(tsk
);
2127 goto retry_find_task
;
2131 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2133 threadgroup_unlock(tsk
);
2135 put_task_struct(tsk
);
2137 mutex_unlock(&cgroup_mutex
);
2142 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2143 * @from: attach to all cgroups of a given task
2144 * @tsk: the task to be attached
2146 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2148 struct cgroup_root
*root
;
2151 mutex_lock(&cgroup_mutex
);
2152 for_each_root(root
) {
2153 struct cgroup
*from_cgrp
;
2155 if (root
== &cgrp_dfl_root
)
2158 down_read(&css_set_rwsem
);
2159 from_cgrp
= task_cgroup_from_root(from
, root
);
2160 up_read(&css_set_rwsem
);
2162 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2166 mutex_unlock(&cgroup_mutex
);
2170 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2172 static int cgroup_tasks_write(struct cgroup_subsys_state
*css
,
2173 struct cftype
*cft
, u64 pid
)
2175 return attach_task_by_pid(css
->cgroup
, pid
, false);
2178 static int cgroup_procs_write(struct cgroup_subsys_state
*css
,
2179 struct cftype
*cft
, u64 tgid
)
2181 return attach_task_by_pid(css
->cgroup
, tgid
, true);
2184 static int cgroup_release_agent_write(struct cgroup_subsys_state
*css
,
2185 struct cftype
*cft
, char *buffer
)
2187 struct cgroup_root
*root
= css
->cgroup
->root
;
2189 BUILD_BUG_ON(sizeof(root
->release_agent_path
) < PATH_MAX
);
2190 if (!cgroup_lock_live_group(css
->cgroup
))
2192 spin_lock(&release_agent_path_lock
);
2193 strlcpy(root
->release_agent_path
, buffer
,
2194 sizeof(root
->release_agent_path
));
2195 spin_unlock(&release_agent_path_lock
);
2196 mutex_unlock(&cgroup_mutex
);
2200 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
2202 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2204 if (!cgroup_lock_live_group(cgrp
))
2206 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2207 seq_putc(seq
, '\n');
2208 mutex_unlock(&cgroup_mutex
);
2212 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
2214 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2216 seq_printf(seq
, "%d\n", cgroup_sane_behavior(cgrp
));
2220 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
2221 size_t nbytes
, loff_t off
)
2223 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
2224 struct cftype
*cft
= of
->kn
->priv
;
2225 struct cgroup_subsys_state
*css
;
2229 * kernfs guarantees that a file isn't deleted with operations in
2230 * flight, which means that the matching css is and stays alive and
2231 * doesn't need to be pinned. The RCU locking is not necessary
2232 * either. It's just for the convenience of using cgroup_css().
2235 css
= cgroup_css(cgrp
, cft
->ss
);
2238 if (cft
->write_string
) {
2239 ret
= cft
->write_string(css
, cft
, strstrip(buf
));
2240 } else if (cft
->write_u64
) {
2241 unsigned long long v
;
2242 ret
= kstrtoull(buf
, 0, &v
);
2244 ret
= cft
->write_u64(css
, cft
, v
);
2245 } else if (cft
->write_s64
) {
2247 ret
= kstrtoll(buf
, 0, &v
);
2249 ret
= cft
->write_s64(css
, cft
, v
);
2250 } else if (cft
->trigger
) {
2251 ret
= cft
->trigger(css
, (unsigned int)cft
->private);
2256 return ret
?: nbytes
;
2259 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
2261 return seq_cft(seq
)->seq_start(seq
, ppos
);
2264 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
2266 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
2269 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
2271 seq_cft(seq
)->seq_stop(seq
, v
);
2274 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2276 struct cftype
*cft
= seq_cft(m
);
2277 struct cgroup_subsys_state
*css
= seq_css(m
);
2280 return cft
->seq_show(m
, arg
);
2283 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
2284 else if (cft
->read_s64
)
2285 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
2291 static struct kernfs_ops cgroup_kf_single_ops
= {
2292 .atomic_write_len
= PAGE_SIZE
,
2293 .write
= cgroup_file_write
,
2294 .seq_show
= cgroup_seqfile_show
,
2297 static struct kernfs_ops cgroup_kf_ops
= {
2298 .atomic_write_len
= PAGE_SIZE
,
2299 .write
= cgroup_file_write
,
2300 .seq_start
= cgroup_seqfile_start
,
2301 .seq_next
= cgroup_seqfile_next
,
2302 .seq_stop
= cgroup_seqfile_stop
,
2303 .seq_show
= cgroup_seqfile_show
,
2307 * cgroup_rename - Only allow simple rename of directories in place.
2309 static int cgroup_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
2310 const char *new_name_str
)
2312 struct cgroup
*cgrp
= kn
->priv
;
2315 if (kernfs_type(kn
) != KERNFS_DIR
)
2317 if (kn
->parent
!= new_parent
)
2321 * This isn't a proper migration and its usefulness is very
2322 * limited. Disallow if sane_behavior.
2324 if (cgroup_sane_behavior(cgrp
))
2328 * We're gonna grab cgroup_tree_mutex which nests outside kernfs
2329 * active_ref. kernfs_rename() doesn't require active_ref
2330 * protection. Break them before grabbing cgroup_tree_mutex.
2332 kernfs_break_active_protection(new_parent
);
2333 kernfs_break_active_protection(kn
);
2335 mutex_lock(&cgroup_tree_mutex
);
2336 mutex_lock(&cgroup_mutex
);
2338 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
2340 mutex_unlock(&cgroup_mutex
);
2341 mutex_unlock(&cgroup_tree_mutex
);
2343 kernfs_unbreak_active_protection(kn
);
2344 kernfs_unbreak_active_protection(new_parent
);
2348 static int cgroup_add_file(struct cgroup
*cgrp
, struct cftype
*cft
)
2350 char name
[CGROUP_FILE_NAME_MAX
];
2351 struct kernfs_node
*kn
;
2352 struct lock_class_key
*key
= NULL
;
2354 #ifdef CONFIG_DEBUG_LOCK_ALLOC
2355 key
= &cft
->lockdep_key
;
2357 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
2358 cgroup_file_mode(cft
), 0, cft
->kf_ops
, cft
,
2360 return PTR_ERR_OR_ZERO(kn
);
2364 * cgroup_addrm_files - add or remove files to a cgroup directory
2365 * @cgrp: the target cgroup
2366 * @cfts: array of cftypes to be added
2367 * @is_add: whether to add or remove
2369 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2370 * For removals, this function never fails. If addition fails, this
2371 * function doesn't remove files already added. The caller is responsible
2374 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
2380 lockdep_assert_held(&cgroup_tree_mutex
);
2382 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2383 /* does cft->flags tell us to skip this file on @cgrp? */
2384 if ((cft
->flags
& CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
2386 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_sane_behavior(cgrp
))
2388 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2390 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2394 ret
= cgroup_add_file(cgrp
, cft
);
2396 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2401 cgroup_rm_file(cgrp
, cft
);
2407 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
2410 struct cgroup_subsys
*ss
= cfts
[0].ss
;
2411 struct cgroup
*root
= &ss
->root
->cgrp
;
2412 struct cgroup_subsys_state
*css
;
2415 lockdep_assert_held(&cgroup_tree_mutex
);
2417 /* don't bother if @ss isn't attached */
2418 if (ss
->root
== &cgrp_dfl_root
)
2421 /* add/rm files for all cgroups created before */
2422 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
2423 struct cgroup
*cgrp
= css
->cgroup
;
2425 if (cgroup_is_dead(cgrp
))
2428 ret
= cgroup_addrm_files(cgrp
, cfts
, is_add
);
2434 kernfs_activate(root
->kn
);
2438 static void cgroup_exit_cftypes(struct cftype
*cfts
)
2442 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2443 /* free copy for custom atomic_write_len, see init_cftypes() */
2444 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
2451 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2455 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2456 struct kernfs_ops
*kf_ops
;
2458 WARN_ON(cft
->ss
|| cft
->kf_ops
);
2461 kf_ops
= &cgroup_kf_ops
;
2463 kf_ops
= &cgroup_kf_single_ops
;
2466 * Ugh... if @cft wants a custom max_write_len, we need to
2467 * make a copy of kf_ops to set its atomic_write_len.
2469 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
2470 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
2472 cgroup_exit_cftypes(cfts
);
2475 kf_ops
->atomic_write_len
= cft
->max_write_len
;
2478 cft
->kf_ops
= kf_ops
;
2485 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
2487 lockdep_assert_held(&cgroup_tree_mutex
);
2489 if (!cfts
|| !cfts
[0].ss
)
2492 list_del(&cfts
->node
);
2493 cgroup_apply_cftypes(cfts
, false);
2494 cgroup_exit_cftypes(cfts
);
2499 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2500 * @cfts: zero-length name terminated array of cftypes
2502 * Unregister @cfts. Files described by @cfts are removed from all
2503 * existing cgroups and all future cgroups won't have them either. This
2504 * function can be called anytime whether @cfts' subsys is attached or not.
2506 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2509 int cgroup_rm_cftypes(struct cftype
*cfts
)
2513 mutex_lock(&cgroup_tree_mutex
);
2514 ret
= cgroup_rm_cftypes_locked(cfts
);
2515 mutex_unlock(&cgroup_tree_mutex
);
2520 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2521 * @ss: target cgroup subsystem
2522 * @cfts: zero-length name terminated array of cftypes
2524 * Register @cfts to @ss. Files described by @cfts are created for all
2525 * existing cgroups to which @ss is attached and all future cgroups will
2526 * have them too. This function can be called anytime whether @ss is
2529 * Returns 0 on successful registration, -errno on failure. Note that this
2530 * function currently returns 0 as long as @cfts registration is successful
2531 * even if some file creation attempts on existing cgroups fail.
2533 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
2537 if (!cfts
|| cfts
[0].name
[0] == '\0')
2540 ret
= cgroup_init_cftypes(ss
, cfts
);
2544 mutex_lock(&cgroup_tree_mutex
);
2546 list_add_tail(&cfts
->node
, &ss
->cfts
);
2547 ret
= cgroup_apply_cftypes(cfts
, true);
2549 cgroup_rm_cftypes_locked(cfts
);
2551 mutex_unlock(&cgroup_tree_mutex
);
2556 * cgroup_task_count - count the number of tasks in a cgroup.
2557 * @cgrp: the cgroup in question
2559 * Return the number of tasks in the cgroup.
2561 static int cgroup_task_count(const struct cgroup
*cgrp
)
2564 struct cgrp_cset_link
*link
;
2566 down_read(&css_set_rwsem
);
2567 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
2568 count
+= atomic_read(&link
->cset
->refcount
);
2569 up_read(&css_set_rwsem
);
2574 * css_next_child - find the next child of a given css
2575 * @pos_css: the current position (%NULL to initiate traversal)
2576 * @parent_css: css whose children to walk
2578 * This function returns the next child of @parent_css and should be called
2579 * under either cgroup_mutex or RCU read lock. The only requirement is
2580 * that @parent_css and @pos_css are accessible. The next sibling is
2581 * guaranteed to be returned regardless of their states.
2583 struct cgroup_subsys_state
*
2584 css_next_child(struct cgroup_subsys_state
*pos_css
,
2585 struct cgroup_subsys_state
*parent_css
)
2587 struct cgroup
*pos
= pos_css
? pos_css
->cgroup
: NULL
;
2588 struct cgroup
*cgrp
= parent_css
->cgroup
;
2589 struct cgroup
*next
;
2591 cgroup_assert_mutexes_or_rcu_locked();
2594 * @pos could already have been removed. Once a cgroup is removed,
2595 * its ->sibling.next is no longer updated when its next sibling
2596 * changes. As CGRP_DEAD assertion is serialized and happens
2597 * before the cgroup is taken off the ->sibling list, if we see it
2598 * unasserted, it's guaranteed that the next sibling hasn't
2599 * finished its grace period even if it's already removed, and thus
2600 * safe to dereference from this RCU critical section. If
2601 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
2602 * to be visible as %true here.
2604 * If @pos is dead, its next pointer can't be dereferenced;
2605 * however, as each cgroup is given a monotonically increasing
2606 * unique serial number and always appended to the sibling list,
2607 * the next one can be found by walking the parent's children until
2608 * we see a cgroup with higher serial number than @pos's. While
2609 * this path can be slower, it's taken only when either the current
2610 * cgroup is removed or iteration and removal race.
2613 next
= list_entry_rcu(cgrp
->children
.next
, struct cgroup
, sibling
);
2614 } else if (likely(!cgroup_is_dead(pos
))) {
2615 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup
, sibling
);
2617 list_for_each_entry_rcu(next
, &cgrp
->children
, sibling
)
2618 if (next
->serial_nr
> pos
->serial_nr
)
2622 if (&next
->sibling
== &cgrp
->children
)
2625 return cgroup_css(next
, parent_css
->ss
);
2629 * css_next_descendant_pre - find the next descendant for pre-order walk
2630 * @pos: the current position (%NULL to initiate traversal)
2631 * @root: css whose descendants to walk
2633 * To be used by css_for_each_descendant_pre(). Find the next descendant
2634 * to visit for pre-order traversal of @root's descendants. @root is
2635 * included in the iteration and the first node to be visited.
2637 * While this function requires cgroup_mutex or RCU read locking, it
2638 * doesn't require the whole traversal to be contained in a single critical
2639 * section. This function will return the correct next descendant as long
2640 * as both @pos and @root are accessible and @pos is a descendant of @root.
2642 struct cgroup_subsys_state
*
2643 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
2644 struct cgroup_subsys_state
*root
)
2646 struct cgroup_subsys_state
*next
;
2648 cgroup_assert_mutexes_or_rcu_locked();
2650 /* if first iteration, visit @root */
2654 /* visit the first child if exists */
2655 next
= css_next_child(NULL
, pos
);
2659 /* no child, visit my or the closest ancestor's next sibling */
2660 while (pos
!= root
) {
2661 next
= css_next_child(pos
, css_parent(pos
));
2664 pos
= css_parent(pos
);
2671 * css_rightmost_descendant - return the rightmost descendant of a css
2672 * @pos: css of interest
2674 * Return the rightmost descendant of @pos. If there's no descendant, @pos
2675 * is returned. This can be used during pre-order traversal to skip
2678 * While this function requires cgroup_mutex or RCU read locking, it
2679 * doesn't require the whole traversal to be contained in a single critical
2680 * section. This function will return the correct rightmost descendant as
2681 * long as @pos is accessible.
2683 struct cgroup_subsys_state
*
2684 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
2686 struct cgroup_subsys_state
*last
, *tmp
;
2688 cgroup_assert_mutexes_or_rcu_locked();
2692 /* ->prev isn't RCU safe, walk ->next till the end */
2694 css_for_each_child(tmp
, last
)
2701 static struct cgroup_subsys_state
*
2702 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
2704 struct cgroup_subsys_state
*last
;
2708 pos
= css_next_child(NULL
, pos
);
2715 * css_next_descendant_post - find the next descendant for post-order walk
2716 * @pos: the current position (%NULL to initiate traversal)
2717 * @root: css whose descendants to walk
2719 * To be used by css_for_each_descendant_post(). Find the next descendant
2720 * to visit for post-order traversal of @root's descendants. @root is
2721 * included in the iteration and the last node to be visited.
2723 * While this function requires cgroup_mutex or RCU read locking, it
2724 * doesn't require the whole traversal to be contained in a single critical
2725 * section. This function will return the correct next descendant as long
2726 * as both @pos and @cgroup are accessible and @pos is a descendant of
2729 struct cgroup_subsys_state
*
2730 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
2731 struct cgroup_subsys_state
*root
)
2733 struct cgroup_subsys_state
*next
;
2735 cgroup_assert_mutexes_or_rcu_locked();
2737 /* if first iteration, visit leftmost descendant which may be @root */
2739 return css_leftmost_descendant(root
);
2741 /* if we visited @root, we're done */
2745 /* if there's an unvisited sibling, visit its leftmost descendant */
2746 next
= css_next_child(pos
, css_parent(pos
));
2748 return css_leftmost_descendant(next
);
2750 /* no sibling left, visit parent */
2751 return css_parent(pos
);
2755 * css_advance_task_iter - advance a task itererator to the next css_set
2756 * @it: the iterator to advance
2758 * Advance @it to the next css_set to walk.
2760 static void css_advance_task_iter(struct css_task_iter
*it
)
2762 struct list_head
*l
= it
->cset_link
;
2763 struct cgrp_cset_link
*link
;
2764 struct css_set
*cset
;
2766 /* Advance to the next non-empty css_set */
2769 if (l
== &it
->origin_css
->cgroup
->cset_links
) {
2770 it
->cset_link
= NULL
;
2773 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
2775 } while (list_empty(&cset
->tasks
) && list_empty(&cset
->mg_tasks
));
2779 if (!list_empty(&cset
->tasks
))
2780 it
->task
= cset
->tasks
.next
;
2782 it
->task
= cset
->mg_tasks
.next
;
2786 * css_task_iter_start - initiate task iteration
2787 * @css: the css to walk tasks of
2788 * @it: the task iterator to use
2790 * Initiate iteration through the tasks of @css. The caller can call
2791 * css_task_iter_next() to walk through the tasks until the function
2792 * returns NULL. On completion of iteration, css_task_iter_end() must be
2795 * Note that this function acquires a lock which is released when the
2796 * iteration finishes. The caller can't sleep while iteration is in
2799 void css_task_iter_start(struct cgroup_subsys_state
*css
,
2800 struct css_task_iter
*it
)
2801 __acquires(css_set_rwsem
)
2803 /* no one should try to iterate before mounting cgroups */
2804 WARN_ON_ONCE(!use_task_css_set_links
);
2806 down_read(&css_set_rwsem
);
2808 it
->origin_css
= css
;
2809 it
->cset_link
= &css
->cgroup
->cset_links
;
2811 css_advance_task_iter(it
);
2815 * css_task_iter_next - return the next task for the iterator
2816 * @it: the task iterator being iterated
2818 * The "next" function for task iteration. @it should have been
2819 * initialized via css_task_iter_start(). Returns NULL when the iteration
2822 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
2824 struct task_struct
*res
;
2825 struct list_head
*l
= it
->task
;
2826 struct cgrp_cset_link
*link
= list_entry(it
->cset_link
,
2827 struct cgrp_cset_link
, cset_link
);
2829 /* If the iterator cg is NULL, we have no tasks */
2832 res
= list_entry(l
, struct task_struct
, cg_list
);
2835 * Advance iterator to find next entry. cset->tasks is consumed
2836 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
2841 if (l
== &link
->cset
->tasks
)
2842 l
= link
->cset
->mg_tasks
.next
;
2844 if (l
== &link
->cset
->mg_tasks
)
2845 css_advance_task_iter(it
);
2853 * css_task_iter_end - finish task iteration
2854 * @it: the task iterator to finish
2856 * Finish task iteration started by css_task_iter_start().
2858 void css_task_iter_end(struct css_task_iter
*it
)
2859 __releases(css_set_rwsem
)
2861 up_read(&css_set_rwsem
);
2865 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
2866 * @to: cgroup to which the tasks will be moved
2867 * @from: cgroup in which the tasks currently reside
2869 * Locking rules between cgroup_post_fork() and the migration path
2870 * guarantee that, if a task is forking while being migrated, the new child
2871 * is guaranteed to be either visible in the source cgroup after the
2872 * parent's migration is complete or put into the target cgroup. No task
2873 * can slip out of migration through forking.
2875 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
2877 LIST_HEAD(preloaded_csets
);
2878 struct cgrp_cset_link
*link
;
2879 struct css_task_iter it
;
2880 struct task_struct
*task
;
2883 mutex_lock(&cgroup_mutex
);
2885 /* all tasks in @from are being moved, all csets are source */
2886 down_read(&css_set_rwsem
);
2887 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
2888 cgroup_migrate_add_src(link
->cset
, to
, &preloaded_csets
);
2889 up_read(&css_set_rwsem
);
2891 ret
= cgroup_migrate_prepare_dst(to
, &preloaded_csets
);
2896 * Migrate tasks one-by-one until @form is empty. This fails iff
2897 * ->can_attach() fails.
2900 css_task_iter_start(&from
->dummy_css
, &it
);
2901 task
= css_task_iter_next(&it
);
2903 get_task_struct(task
);
2904 css_task_iter_end(&it
);
2907 ret
= cgroup_migrate(to
, task
, false);
2908 put_task_struct(task
);
2910 } while (task
&& !ret
);
2912 cgroup_migrate_finish(&preloaded_csets
);
2913 mutex_unlock(&cgroup_mutex
);
2918 * Stuff for reading the 'tasks'/'procs' files.
2920 * Reading this file can return large amounts of data if a cgroup has
2921 * *lots* of attached tasks. So it may need several calls to read(),
2922 * but we cannot guarantee that the information we produce is correct
2923 * unless we produce it entirely atomically.
2927 /* which pidlist file are we talking about? */
2928 enum cgroup_filetype
{
2934 * A pidlist is a list of pids that virtually represents the contents of one
2935 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
2936 * a pair (one each for procs, tasks) for each pid namespace that's relevant
2939 struct cgroup_pidlist
{
2941 * used to find which pidlist is wanted. doesn't change as long as
2942 * this particular list stays in the list.
2944 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
2947 /* how many elements the above list has */
2949 /* each of these stored in a list by its cgroup */
2950 struct list_head links
;
2951 /* pointer to the cgroup we belong to, for list removal purposes */
2952 struct cgroup
*owner
;
2953 /* for delayed destruction */
2954 struct delayed_work destroy_dwork
;
2958 * The following two functions "fix" the issue where there are more pids
2959 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2960 * TODO: replace with a kernel-wide solution to this problem
2962 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2963 static void *pidlist_allocate(int count
)
2965 if (PIDLIST_TOO_LARGE(count
))
2966 return vmalloc(count
* sizeof(pid_t
));
2968 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
2971 static void pidlist_free(void *p
)
2973 if (is_vmalloc_addr(p
))
2980 * Used to destroy all pidlists lingering waiting for destroy timer. None
2981 * should be left afterwards.
2983 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
)
2985 struct cgroup_pidlist
*l
, *tmp_l
;
2987 mutex_lock(&cgrp
->pidlist_mutex
);
2988 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
2989 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
2990 mutex_unlock(&cgrp
->pidlist_mutex
);
2992 flush_workqueue(cgroup_pidlist_destroy_wq
);
2993 BUG_ON(!list_empty(&cgrp
->pidlists
));
2996 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
2998 struct delayed_work
*dwork
= to_delayed_work(work
);
2999 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
3001 struct cgroup_pidlist
*tofree
= NULL
;
3003 mutex_lock(&l
->owner
->pidlist_mutex
);
3006 * Destroy iff we didn't get queued again. The state won't change
3007 * as destroy_dwork can only be queued while locked.
3009 if (!delayed_work_pending(dwork
)) {
3010 list_del(&l
->links
);
3011 pidlist_free(l
->list
);
3012 put_pid_ns(l
->key
.ns
);
3016 mutex_unlock(&l
->owner
->pidlist_mutex
);
3021 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3022 * Returns the number of unique elements.
3024 static int pidlist_uniq(pid_t
*list
, int length
)
3029 * we presume the 0th element is unique, so i starts at 1. trivial
3030 * edge cases first; no work needs to be done for either
3032 if (length
== 0 || length
== 1)
3034 /* src and dest walk down the list; dest counts unique elements */
3035 for (src
= 1; src
< length
; src
++) {
3036 /* find next unique element */
3037 while (list
[src
] == list
[src
-1]) {
3042 /* dest always points to where the next unique element goes */
3043 list
[dest
] = list
[src
];
3051 * The two pid files - task and cgroup.procs - guaranteed that the result
3052 * is sorted, which forced this whole pidlist fiasco. As pid order is
3053 * different per namespace, each namespace needs differently sorted list,
3054 * making it impossible to use, for example, single rbtree of member tasks
3055 * sorted by task pointer. As pidlists can be fairly large, allocating one
3056 * per open file is dangerous, so cgroup had to implement shared pool of
3057 * pidlists keyed by cgroup and namespace.
3059 * All this extra complexity was caused by the original implementation
3060 * committing to an entirely unnecessary property. In the long term, we
3061 * want to do away with it. Explicitly scramble sort order if
3062 * sane_behavior so that no such expectation exists in the new interface.
3064 * Scrambling is done by swapping every two consecutive bits, which is
3065 * non-identity one-to-one mapping which disturbs sort order sufficiently.
3067 static pid_t
pid_fry(pid_t pid
)
3069 unsigned a
= pid
& 0x55555555;
3070 unsigned b
= pid
& 0xAAAAAAAA;
3072 return (a
<< 1) | (b
>> 1);
3075 static pid_t
cgroup_pid_fry(struct cgroup
*cgrp
, pid_t pid
)
3077 if (cgroup_sane_behavior(cgrp
))
3078 return pid_fry(pid
);
3083 static int cmppid(const void *a
, const void *b
)
3085 return *(pid_t
*)a
- *(pid_t
*)b
;
3088 static int fried_cmppid(const void *a
, const void *b
)
3090 return pid_fry(*(pid_t
*)a
) - pid_fry(*(pid_t
*)b
);
3093 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3094 enum cgroup_filetype type
)
3096 struct cgroup_pidlist
*l
;
3097 /* don't need task_nsproxy() if we're looking at ourself */
3098 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3100 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3102 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
3103 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
3109 * find the appropriate pidlist for our purpose (given procs vs tasks)
3110 * returns with the lock on that pidlist already held, and takes care
3111 * of the use count, or returns NULL with no locks held if we're out of
3114 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
3115 enum cgroup_filetype type
)
3117 struct cgroup_pidlist
*l
;
3119 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3121 l
= cgroup_pidlist_find(cgrp
, type
);
3125 /* entry not found; create a new one */
3126 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3130 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
3132 /* don't need task_nsproxy() if we're looking at ourself */
3133 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
3135 list_add(&l
->links
, &cgrp
->pidlists
);
3140 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3142 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3143 struct cgroup_pidlist
**lp
)
3147 int pid
, n
= 0; /* used for populating the array */
3148 struct css_task_iter it
;
3149 struct task_struct
*tsk
;
3150 struct cgroup_pidlist
*l
;
3152 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3155 * If cgroup gets more users after we read count, we won't have
3156 * enough space - tough. This race is indistinguishable to the
3157 * caller from the case that the additional cgroup users didn't
3158 * show up until sometime later on.
3160 length
= cgroup_task_count(cgrp
);
3161 array
= pidlist_allocate(length
);
3164 /* now, populate the array */
3165 css_task_iter_start(&cgrp
->dummy_css
, &it
);
3166 while ((tsk
= css_task_iter_next(&it
))) {
3167 if (unlikely(n
== length
))
3169 /* get tgid or pid for procs or tasks file respectively */
3170 if (type
== CGROUP_FILE_PROCS
)
3171 pid
= task_tgid_vnr(tsk
);
3173 pid
= task_pid_vnr(tsk
);
3174 if (pid
> 0) /* make sure to only use valid results */
3177 css_task_iter_end(&it
);
3179 /* now sort & (if procs) strip out duplicates */
3180 if (cgroup_sane_behavior(cgrp
))
3181 sort(array
, length
, sizeof(pid_t
), fried_cmppid
, NULL
);
3183 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3184 if (type
== CGROUP_FILE_PROCS
)
3185 length
= pidlist_uniq(array
, length
);
3187 l
= cgroup_pidlist_find_create(cgrp
, type
);
3189 mutex_unlock(&cgrp
->pidlist_mutex
);
3190 pidlist_free(array
);
3194 /* store array, freeing old if necessary */
3195 pidlist_free(l
->list
);
3203 * cgroupstats_build - build and fill cgroupstats
3204 * @stats: cgroupstats to fill information into
3205 * @dentry: A dentry entry belonging to the cgroup for which stats have
3208 * Build and fill cgroupstats so that taskstats can export it to user
3211 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3213 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
3214 struct cgroup
*cgrp
;
3215 struct css_task_iter it
;
3216 struct task_struct
*tsk
;
3218 /* it should be kernfs_node belonging to cgroupfs and is a directory */
3219 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
3220 kernfs_type(kn
) != KERNFS_DIR
)
3223 mutex_lock(&cgroup_mutex
);
3226 * We aren't being called from kernfs and there's no guarantee on
3227 * @kn->priv's validity. For this and css_tryget_from_dir(),
3228 * @kn->priv is RCU safe. Let's do the RCU dancing.
3231 cgrp
= rcu_dereference(kn
->priv
);
3232 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
3234 mutex_unlock(&cgroup_mutex
);
3239 css_task_iter_start(&cgrp
->dummy_css
, &it
);
3240 while ((tsk
= css_task_iter_next(&it
))) {
3241 switch (tsk
->state
) {
3243 stats
->nr_running
++;
3245 case TASK_INTERRUPTIBLE
:
3246 stats
->nr_sleeping
++;
3248 case TASK_UNINTERRUPTIBLE
:
3249 stats
->nr_uninterruptible
++;
3252 stats
->nr_stopped
++;
3255 if (delayacct_is_task_waiting_on_io(tsk
))
3256 stats
->nr_io_wait
++;
3260 css_task_iter_end(&it
);
3262 mutex_unlock(&cgroup_mutex
);
3268 * seq_file methods for the tasks/procs files. The seq_file position is the
3269 * next pid to display; the seq_file iterator is a pointer to the pid
3270 * in the cgroup->l->list array.
3273 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3276 * Initially we receive a position value that corresponds to
3277 * one more than the last pid shown (or 0 on the first call or
3278 * after a seek to the start). Use a binary-search to find the
3279 * next pid to display, if any
3281 struct kernfs_open_file
*of
= s
->private;
3282 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
3283 struct cgroup_pidlist
*l
;
3284 enum cgroup_filetype type
= seq_cft(s
)->private;
3285 int index
= 0, pid
= *pos
;
3288 mutex_lock(&cgrp
->pidlist_mutex
);
3291 * !NULL @of->priv indicates that this isn't the first start()
3292 * after open. If the matching pidlist is around, we can use that.
3293 * Look for it. Note that @of->priv can't be used directly. It
3294 * could already have been destroyed.
3297 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
3300 * Either this is the first start() after open or the matching
3301 * pidlist has been destroyed inbetween. Create a new one.
3304 ret
= pidlist_array_load(cgrp
, type
,
3305 (struct cgroup_pidlist
**)&of
->priv
);
3307 return ERR_PTR(ret
);
3312 int end
= l
->length
;
3314 while (index
< end
) {
3315 int mid
= (index
+ end
) / 2;
3316 if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) == pid
) {
3319 } else if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) <= pid
)
3325 /* If we're off the end of the array, we're done */
3326 if (index
>= l
->length
)
3328 /* Update the abstract position to be the actual pid that we found */
3329 iter
= l
->list
+ index
;
3330 *pos
= cgroup_pid_fry(cgrp
, *iter
);
3334 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3336 struct kernfs_open_file
*of
= s
->private;
3337 struct cgroup_pidlist
*l
= of
->priv
;
3340 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
3341 CGROUP_PIDLIST_DESTROY_DELAY
);
3342 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
3345 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3347 struct kernfs_open_file
*of
= s
->private;
3348 struct cgroup_pidlist
*l
= of
->priv
;
3350 pid_t
*end
= l
->list
+ l
->length
;
3352 * Advance to the next pid in the array. If this goes off the
3359 *pos
= cgroup_pid_fry(seq_css(s
)->cgroup
, *p
);
3364 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3366 return seq_printf(s
, "%d\n", *(int *)v
);
3370 * seq_operations functions for iterating on pidlists through seq_file -
3371 * independent of whether it's tasks or procs
3373 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3374 .start
= cgroup_pidlist_start
,
3375 .stop
= cgroup_pidlist_stop
,
3376 .next
= cgroup_pidlist_next
,
3377 .show
= cgroup_pidlist_show
,
3380 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
3383 return notify_on_release(css
->cgroup
);
3386 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
3387 struct cftype
*cft
, u64 val
)
3389 clear_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
3391 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
3393 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
3397 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
3400 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
3403 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
3404 struct cftype
*cft
, u64 val
)
3407 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
3409 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
3413 static struct cftype cgroup_base_files
[] = {
3415 .name
= "cgroup.procs",
3416 .seq_start
= cgroup_pidlist_start
,
3417 .seq_next
= cgroup_pidlist_next
,
3418 .seq_stop
= cgroup_pidlist_stop
,
3419 .seq_show
= cgroup_pidlist_show
,
3420 .private = CGROUP_FILE_PROCS
,
3421 .write_u64
= cgroup_procs_write
,
3422 .mode
= S_IRUGO
| S_IWUSR
,
3425 .name
= "cgroup.clone_children",
3426 .flags
= CFTYPE_INSANE
,
3427 .read_u64
= cgroup_clone_children_read
,
3428 .write_u64
= cgroup_clone_children_write
,
3431 .name
= "cgroup.sane_behavior",
3432 .flags
= CFTYPE_ONLY_ON_ROOT
,
3433 .seq_show
= cgroup_sane_behavior_show
,
3437 * Historical crazy stuff. These don't have "cgroup." prefix and
3438 * don't exist if sane_behavior. If you're depending on these, be
3439 * prepared to be burned.
3443 .flags
= CFTYPE_INSANE
, /* use "procs" instead */
3444 .seq_start
= cgroup_pidlist_start
,
3445 .seq_next
= cgroup_pidlist_next
,
3446 .seq_stop
= cgroup_pidlist_stop
,
3447 .seq_show
= cgroup_pidlist_show
,
3448 .private = CGROUP_FILE_TASKS
,
3449 .write_u64
= cgroup_tasks_write
,
3450 .mode
= S_IRUGO
| S_IWUSR
,
3453 .name
= "notify_on_release",
3454 .flags
= CFTYPE_INSANE
,
3455 .read_u64
= cgroup_read_notify_on_release
,
3456 .write_u64
= cgroup_write_notify_on_release
,
3459 .name
= "release_agent",
3460 .flags
= CFTYPE_INSANE
| CFTYPE_ONLY_ON_ROOT
,
3461 .seq_show
= cgroup_release_agent_show
,
3462 .write_string
= cgroup_release_agent_write
,
3463 .max_write_len
= PATH_MAX
- 1,
3469 * cgroup_populate_dir - create subsys files in a cgroup directory
3470 * @cgrp: target cgroup
3471 * @subsys_mask: mask of the subsystem ids whose files should be added
3473 * On failure, no file is added.
3475 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned long subsys_mask
)
3477 struct cgroup_subsys
*ss
;
3480 /* process cftsets of each subsystem */
3481 for_each_subsys(ss
, i
) {
3482 struct cftype
*cfts
;
3484 if (!test_bit(i
, &subsys_mask
))
3487 list_for_each_entry(cfts
, &ss
->cfts
, node
) {
3488 ret
= cgroup_addrm_files(cgrp
, cfts
, true);
3495 cgroup_clear_dir(cgrp
, subsys_mask
);
3500 * css destruction is four-stage process.
3502 * 1. Destruction starts. Killing of the percpu_ref is initiated.
3503 * Implemented in kill_css().
3505 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
3506 * and thus css_tryget() is guaranteed to fail, the css can be offlined
3507 * by invoking offline_css(). After offlining, the base ref is put.
3508 * Implemented in css_killed_work_fn().
3510 * 3. When the percpu_ref reaches zero, the only possible remaining
3511 * accessors are inside RCU read sections. css_release() schedules the
3514 * 4. After the grace period, the css can be freed. Implemented in
3515 * css_free_work_fn().
3517 * It is actually hairier because both step 2 and 4 require process context
3518 * and thus involve punting to css->destroy_work adding two additional
3519 * steps to the already complex sequence.
3521 static void css_free_work_fn(struct work_struct
*work
)
3523 struct cgroup_subsys_state
*css
=
3524 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
3525 struct cgroup
*cgrp
= css
->cgroup
;
3528 css_put(css
->parent
);
3530 css
->ss
->css_free(css
);
3534 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
3536 struct cgroup_subsys_state
*css
=
3537 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
3539 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
3540 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
3543 static void css_release(struct percpu_ref
*ref
)
3545 struct cgroup_subsys_state
*css
=
3546 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
3548 RCU_INIT_POINTER(css
->cgroup
->subsys
[css
->ss
->id
], NULL
);
3549 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
3552 static void init_css(struct cgroup_subsys_state
*css
, struct cgroup_subsys
*ss
,
3553 struct cgroup
*cgrp
)
3560 css
->parent
= cgroup_css(cgrp
->parent
, ss
);
3562 css
->flags
|= CSS_ROOT
;
3564 BUG_ON(cgroup_css(cgrp
, ss
));
3567 /* invoke ->css_online() on a new CSS and mark it online if successful */
3568 static int online_css(struct cgroup_subsys_state
*css
)
3570 struct cgroup_subsys
*ss
= css
->ss
;
3573 lockdep_assert_held(&cgroup_tree_mutex
);
3574 lockdep_assert_held(&cgroup_mutex
);
3577 ret
= ss
->css_online(css
);
3579 css
->flags
|= CSS_ONLINE
;
3580 css
->cgroup
->nr_css
++;
3581 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
3586 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
3587 static void offline_css(struct cgroup_subsys_state
*css
)
3589 struct cgroup_subsys
*ss
= css
->ss
;
3591 lockdep_assert_held(&cgroup_tree_mutex
);
3592 lockdep_assert_held(&cgroup_mutex
);
3594 if (!(css
->flags
& CSS_ONLINE
))
3597 if (ss
->css_offline
)
3598 ss
->css_offline(css
);
3600 css
->flags
&= ~CSS_ONLINE
;
3601 css
->cgroup
->nr_css
--;
3602 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], css
);
3606 * create_css - create a cgroup_subsys_state
3607 * @cgrp: the cgroup new css will be associated with
3608 * @ss: the subsys of new css
3610 * Create a new css associated with @cgrp - @ss pair. On success, the new
3611 * css is online and installed in @cgrp with all interface files created.
3612 * Returns 0 on success, -errno on failure.
3614 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
)
3616 struct cgroup
*parent
= cgrp
->parent
;
3617 struct cgroup_subsys_state
*css
;
3620 lockdep_assert_held(&cgroup_mutex
);
3622 css
= ss
->css_alloc(cgroup_css(parent
, ss
));
3624 return PTR_ERR(css
);
3626 err
= percpu_ref_init(&css
->refcnt
, css_release
);
3630 init_css(css
, ss
, cgrp
);
3632 err
= cgroup_populate_dir(cgrp
, 1 << ss
->id
);
3634 goto err_free_percpu_ref
;
3636 err
= online_css(css
);
3641 css_get(css
->parent
);
3643 cgrp
->subsys_mask
|= 1 << ss
->id
;
3645 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
3647 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",
3648 current
->comm
, current
->pid
, ss
->name
);
3649 if (!strcmp(ss
->name
, "memory"))
3650 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
3651 ss
->warned_broken_hierarchy
= true;
3657 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->id
);
3658 err_free_percpu_ref
:
3659 percpu_ref_cancel_init(&css
->refcnt
);
3666 * cgroup_create - create a cgroup
3667 * @parent: cgroup that will be parent of the new cgroup
3668 * @name: name of the new cgroup
3669 * @mode: mode to set on new cgroup
3671 static long cgroup_create(struct cgroup
*parent
, const char *name
,
3674 struct cgroup
*cgrp
;
3675 struct cgroup_root
*root
= parent
->root
;
3677 struct cgroup_subsys
*ss
;
3678 struct kernfs_node
*kn
;
3681 * XXX: The default hierarchy isn't fully implemented yet. Block
3682 * !root cgroup creation on it for now.
3684 if (root
== &cgrp_dfl_root
)
3687 /* allocate the cgroup and its ID, 0 is reserved for the root */
3688 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3692 mutex_lock(&cgroup_tree_mutex
);
3695 * Only live parents can have children. Note that the liveliness
3696 * check isn't strictly necessary because cgroup_mkdir() and
3697 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
3698 * anyway so that locking is contained inside cgroup proper and we
3699 * don't get nasty surprises if we ever grow another caller.
3701 if (!cgroup_lock_live_group(parent
)) {
3703 goto err_unlock_tree
;
3707 * Temporarily set the pointer to NULL, so idr_find() won't return
3708 * a half-baked cgroup.
3710 cgrp
->id
= idr_alloc(&root
->cgroup_idr
, NULL
, 1, 0, GFP_KERNEL
);
3716 init_cgroup_housekeeping(cgrp
);
3718 cgrp
->parent
= parent
;
3719 cgrp
->dummy_css
.parent
= &parent
->dummy_css
;
3720 cgrp
->root
= parent
->root
;
3722 if (notify_on_release(parent
))
3723 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3725 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
3726 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
3728 /* create the directory */
3729 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
3737 * This extra ref will be put in cgroup_free_fn() and guarantees
3738 * that @cgrp->kn is always accessible.
3742 cgrp
->serial_nr
= cgroup_serial_nr_next
++;
3744 /* allocation complete, commit to creation */
3745 list_add_tail_rcu(&cgrp
->sibling
, &cgrp
->parent
->children
);
3746 atomic_inc(&root
->nr_cgrps
);
3750 * @cgrp is now fully operational. If something fails after this
3751 * point, it'll be released via the normal destruction path.
3753 idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
3755 err
= cgroup_addrm_files(cgrp
, cgroup_base_files
, true);
3759 /* let's create and online css's */
3760 for_each_subsys(ss
, ssid
) {
3761 if (root
->cgrp
.subsys_mask
& (1 << ssid
)) {
3762 err
= create_css(cgrp
, ss
);
3768 kernfs_activate(kn
);
3770 mutex_unlock(&cgroup_mutex
);
3771 mutex_unlock(&cgroup_tree_mutex
);
3776 idr_remove(&root
->cgroup_idr
, cgrp
->id
);
3778 mutex_unlock(&cgroup_mutex
);
3780 mutex_unlock(&cgroup_tree_mutex
);
3785 cgroup_destroy_locked(cgrp
);
3786 mutex_unlock(&cgroup_mutex
);
3787 mutex_unlock(&cgroup_tree_mutex
);
3791 static int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
,
3794 struct cgroup
*parent
= parent_kn
->priv
;
3798 * cgroup_create() grabs cgroup_tree_mutex which nests outside
3799 * kernfs active_ref and cgroup_create() already synchronizes
3800 * properly against removal through cgroup_lock_live_group().
3801 * Break it before calling cgroup_create().
3804 kernfs_break_active_protection(parent_kn
);
3806 ret
= cgroup_create(parent
, name
, mode
);
3808 kernfs_unbreak_active_protection(parent_kn
);
3814 * This is called when the refcnt of a css is confirmed to be killed.
3815 * css_tryget() is now guaranteed to fail.
3817 static void css_killed_work_fn(struct work_struct
*work
)
3819 struct cgroup_subsys_state
*css
=
3820 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
3821 struct cgroup
*cgrp
= css
->cgroup
;
3823 mutex_lock(&cgroup_tree_mutex
);
3824 mutex_lock(&cgroup_mutex
);
3827 * css_tryget() is guaranteed to fail now. Tell subsystems to
3828 * initate destruction.
3833 * If @cgrp is marked dead, it's waiting for refs of all css's to
3834 * be disabled before proceeding to the second phase of cgroup
3835 * destruction. If we are the last one, kick it off.
3837 if (!cgrp
->nr_css
&& cgroup_is_dead(cgrp
))
3838 cgroup_destroy_css_killed(cgrp
);
3840 mutex_unlock(&cgroup_mutex
);
3841 mutex_unlock(&cgroup_tree_mutex
);
3844 * Put the css refs from kill_css(). Each css holds an extra
3845 * reference to the cgroup's dentry and cgroup removal proceeds
3846 * regardless of css refs. On the last put of each css, whenever
3847 * that may be, the extra dentry ref is put so that dentry
3848 * destruction happens only after all css's are released.
3853 /* css kill confirmation processing requires process context, bounce */
3854 static void css_killed_ref_fn(struct percpu_ref
*ref
)
3856 struct cgroup_subsys_state
*css
=
3857 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
3859 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
3860 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
3863 static void __kill_css(struct cgroup_subsys_state
*css
)
3865 lockdep_assert_held(&cgroup_tree_mutex
);
3868 * This must happen before css is disassociated with its cgroup.
3869 * See seq_css() for details.
3871 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->id
);
3874 * Killing would put the base ref, but we need to keep it alive
3875 * until after ->css_offline().
3880 * cgroup core guarantees that, by the time ->css_offline() is
3881 * invoked, no new css reference will be given out via
3882 * css_tryget(). We can't simply call percpu_ref_kill() and
3883 * proceed to offlining css's because percpu_ref_kill() doesn't
3884 * guarantee that the ref is seen as killed on all CPUs on return.
3886 * Use percpu_ref_kill_and_confirm() to get notifications as each
3887 * css is confirmed to be seen as killed on all CPUs.
3889 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
3893 * kill_css - destroy a css
3894 * @css: css to destroy
3896 * This function initiates destruction of @css by removing cgroup interface
3897 * files and putting its base reference. ->css_offline() will be invoked
3898 * asynchronously once css_tryget() is guaranteed to fail and when the
3899 * reference count reaches zero, @css will be released.
3901 static void kill_css(struct cgroup_subsys_state
*css
)
3903 struct cgroup
*cgrp
= css
->cgroup
;
3905 lockdep_assert_held(&cgroup_tree_mutex
);
3907 /* if already killed, noop */
3908 if (cgrp
->subsys_mask
& (1 << css
->ss
->id
)) {
3909 cgrp
->subsys_mask
&= ~(1 << css
->ss
->id
);
3915 * cgroup_destroy_locked - the first stage of cgroup destruction
3916 * @cgrp: cgroup to be destroyed
3918 * css's make use of percpu refcnts whose killing latency shouldn't be
3919 * exposed to userland and are RCU protected. Also, cgroup core needs to
3920 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
3921 * invoked. To satisfy all the requirements, destruction is implemented in
3922 * the following two steps.
3924 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
3925 * userland visible parts and start killing the percpu refcnts of
3926 * css's. Set up so that the next stage will be kicked off once all
3927 * the percpu refcnts are confirmed to be killed.
3929 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
3930 * rest of destruction. Once all cgroup references are gone, the
3931 * cgroup is RCU-freed.
3933 * This function implements s1. After this step, @cgrp is gone as far as
3934 * the userland is concerned and a new cgroup with the same name may be
3935 * created. As cgroup doesn't care about the names internally, this
3936 * doesn't cause any problem.
3938 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
3939 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
3941 struct cgroup
*child
;
3942 struct cgroup_subsys_state
*css
;
3946 lockdep_assert_held(&cgroup_tree_mutex
);
3947 lockdep_assert_held(&cgroup_mutex
);
3950 * css_set_rwsem synchronizes access to ->cset_links and prevents
3951 * @cgrp from being removed while put_css_set() is in progress.
3953 down_read(&css_set_rwsem
);
3954 empty
= list_empty(&cgrp
->cset_links
);
3955 up_read(&css_set_rwsem
);
3960 * Make sure there's no live children. We can't test ->children
3961 * emptiness as dead children linger on it while being destroyed;
3962 * otherwise, "rmdir parent/child parent" may fail with -EBUSY.
3966 list_for_each_entry_rcu(child
, &cgrp
->children
, sibling
) {
3967 empty
= cgroup_is_dead(child
);
3976 * Mark @cgrp dead. This prevents further task migration and child
3977 * creation by disabling cgroup_lock_live_group(). Note that
3978 * CGRP_DEAD assertion is depended upon by css_next_child() to
3979 * resume iteration after dropping RCU read lock. See
3980 * css_next_child() for details.
3982 set_bit(CGRP_DEAD
, &cgrp
->flags
);
3985 * Initiate massacre of all css's. cgroup_destroy_css_killed()
3986 * will be invoked to perform the rest of destruction once the
3987 * percpu refs of all css's are confirmed to be killed. This
3988 * involves removing the subsystem's files, drop cgroup_mutex.
3990 mutex_unlock(&cgroup_mutex
);
3991 for_each_css(css
, ssid
, cgrp
)
3993 mutex_lock(&cgroup_mutex
);
3995 /* CGRP_DEAD is set, remove from ->release_list for the last time */
3996 raw_spin_lock(&release_list_lock
);
3997 if (!list_empty(&cgrp
->release_list
))
3998 list_del_init(&cgrp
->release_list
);
3999 raw_spin_unlock(&release_list_lock
);
4002 * If @cgrp has css's attached, the second stage of cgroup
4003 * destruction is kicked off from css_killed_work_fn() after the
4004 * refs of all attached css's are killed. If @cgrp doesn't have
4005 * any css, we kick it off here.
4008 cgroup_destroy_css_killed(cgrp
);
4010 /* remove @cgrp directory along with the base files */
4011 mutex_unlock(&cgroup_mutex
);
4014 * There are two control paths which try to determine cgroup from
4015 * dentry without going through kernfs - cgroupstats_build() and
4016 * css_tryget_from_dir(). Those are supported by RCU protecting
4017 * clearing of cgrp->kn->priv backpointer, which should happen
4018 * after all files under it have been removed.
4020 kernfs_remove(cgrp
->kn
); /* @cgrp has an extra ref on its kn */
4021 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
, NULL
);
4023 mutex_lock(&cgroup_mutex
);
4029 * cgroup_destroy_css_killed - the second step of cgroup destruction
4030 * @work: cgroup->destroy_free_work
4032 * This function is invoked from a work item for a cgroup which is being
4033 * destroyed after all css's are offlined and performs the rest of
4034 * destruction. This is the second step of destruction described in the
4035 * comment above cgroup_destroy_locked().
4037 static void cgroup_destroy_css_killed(struct cgroup
*cgrp
)
4039 struct cgroup
*parent
= cgrp
->parent
;
4041 lockdep_assert_held(&cgroup_tree_mutex
);
4042 lockdep_assert_held(&cgroup_mutex
);
4044 /* delete this cgroup from parent->children */
4045 list_del_rcu(&cgrp
->sibling
);
4049 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4050 check_for_release(parent
);
4053 static int cgroup_rmdir(struct kernfs_node
*kn
)
4055 struct cgroup
*cgrp
= kn
->priv
;
4059 * This is self-destruction but @kn can't be removed while this
4060 * callback is in progress. Let's break active protection. Once
4061 * the protection is broken, @cgrp can be destroyed at any point.
4062 * Pin it so that it stays accessible.
4065 kernfs_break_active_protection(kn
);
4067 mutex_lock(&cgroup_tree_mutex
);
4068 mutex_lock(&cgroup_mutex
);
4071 * @cgrp might already have been destroyed while we're trying to
4074 if (!cgroup_is_dead(cgrp
))
4075 ret
= cgroup_destroy_locked(cgrp
);
4077 mutex_unlock(&cgroup_mutex
);
4078 mutex_unlock(&cgroup_tree_mutex
);
4080 kernfs_unbreak_active_protection(kn
);
4085 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
4086 .remount_fs
= cgroup_remount
,
4087 .show_options
= cgroup_show_options
,
4088 .mkdir
= cgroup_mkdir
,
4089 .rmdir
= cgroup_rmdir
,
4090 .rename
= cgroup_rename
,
4093 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4095 struct cgroup_subsys_state
*css
;
4097 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4099 mutex_lock(&cgroup_tree_mutex
);
4100 mutex_lock(&cgroup_mutex
);
4102 INIT_LIST_HEAD(&ss
->cfts
);
4104 /* Create the root cgroup state for this subsystem */
4105 ss
->root
= &cgrp_dfl_root
;
4106 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
4107 /* We don't handle early failures gracefully */
4108 BUG_ON(IS_ERR(css
));
4109 init_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
4111 /* Update the init_css_set to contain a subsys
4112 * pointer to this state - since the subsystem is
4113 * newly registered, all tasks and hence the
4114 * init_css_set is in the subsystem's root cgroup. */
4115 init_css_set
.subsys
[ss
->id
] = css
;
4117 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4119 /* At system boot, before all subsystems have been
4120 * registered, no tasks have been forked, so we don't
4121 * need to invoke fork callbacks here. */
4122 BUG_ON(!list_empty(&init_task
.tasks
));
4124 BUG_ON(online_css(css
));
4126 cgrp_dfl_root
.cgrp
.subsys_mask
|= 1 << ss
->id
;
4128 mutex_unlock(&cgroup_mutex
);
4129 mutex_unlock(&cgroup_tree_mutex
);
4133 * cgroup_init_early - cgroup initialization at system boot
4135 * Initialize cgroups at system boot, and initialize any
4136 * subsystems that request early init.
4138 int __init
cgroup_init_early(void)
4140 static struct cgroup_sb_opts __initdata opts
=
4141 { .flags
= CGRP_ROOT_SANE_BEHAVIOR
};
4142 struct cgroup_subsys
*ss
;
4145 init_cgroup_root(&cgrp_dfl_root
, &opts
);
4146 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
4148 for_each_subsys(ss
, i
) {
4149 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
4150 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
4151 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
4153 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
4154 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
4157 ss
->name
= cgroup_subsys_name
[i
];
4160 cgroup_init_subsys(ss
);
4166 * cgroup_init - cgroup initialization
4168 * Register cgroup filesystem and /proc file, and initialize
4169 * any subsystems that didn't request early init.
4171 int __init
cgroup_init(void)
4173 struct cgroup_subsys
*ss
;
4177 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_base_files
));
4179 mutex_lock(&cgroup_tree_mutex
);
4180 mutex_lock(&cgroup_mutex
);
4182 /* Add init_css_set to the hash table */
4183 key
= css_set_hash(init_css_set
.subsys
);
4184 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4186 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
4188 mutex_unlock(&cgroup_mutex
);
4189 mutex_unlock(&cgroup_tree_mutex
);
4191 for_each_subsys(ss
, ssid
) {
4192 if (!ss
->early_init
)
4193 cgroup_init_subsys(ss
);
4196 * cftype registration needs kmalloc and can't be done
4197 * during early_init. Register base cftypes separately.
4199 if (ss
->base_cftypes
)
4200 WARN_ON(cgroup_add_cftypes(ss
, ss
->base_cftypes
));
4203 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4207 err
= register_filesystem(&cgroup_fs_type
);
4209 kobject_put(cgroup_kobj
);
4213 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4217 static int __init
cgroup_wq_init(void)
4220 * There isn't much point in executing destruction path in
4221 * parallel. Good chunk is serialized with cgroup_mutex anyway.
4222 * Use 1 for @max_active.
4224 * We would prefer to do this in cgroup_init() above, but that
4225 * is called before init_workqueues(): so leave this until after.
4227 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
4228 BUG_ON(!cgroup_destroy_wq
);
4231 * Used to destroy pidlists and separate to serve as flush domain.
4232 * Cap @max_active to 1 too.
4234 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
4236 BUG_ON(!cgroup_pidlist_destroy_wq
);
4240 core_initcall(cgroup_wq_init
);
4243 * proc_cgroup_show()
4244 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4245 * - Used for /proc/<pid>/cgroup.
4248 /* TODO: Use a proper seq_file iterator */
4249 int proc_cgroup_show(struct seq_file
*m
, void *v
)
4252 struct task_struct
*tsk
;
4255 struct cgroup_root
*root
;
4258 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
4264 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4270 mutex_lock(&cgroup_mutex
);
4271 down_read(&css_set_rwsem
);
4273 for_each_root(root
) {
4274 struct cgroup_subsys
*ss
;
4275 struct cgroup
*cgrp
;
4276 int ssid
, count
= 0;
4278 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_root_visible
)
4281 seq_printf(m
, "%d:", root
->hierarchy_id
);
4282 for_each_subsys(ss
, ssid
)
4283 if (root
->cgrp
.subsys_mask
& (1 << ssid
))
4284 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4285 if (strlen(root
->name
))
4286 seq_printf(m
, "%sname=%s", count
? "," : "",
4289 cgrp
= task_cgroup_from_root(tsk
, root
);
4290 path
= cgroup_path(cgrp
, buf
, PATH_MAX
);
4292 retval
= -ENAMETOOLONG
;
4300 up_read(&css_set_rwsem
);
4301 mutex_unlock(&cgroup_mutex
);
4302 put_task_struct(tsk
);
4309 /* Display information about each subsystem and each hierarchy */
4310 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4312 struct cgroup_subsys
*ss
;
4315 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4317 * ideally we don't want subsystems moving around while we do this.
4318 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4319 * subsys/hierarchy state.
4321 mutex_lock(&cgroup_mutex
);
4323 for_each_subsys(ss
, i
)
4324 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4325 ss
->name
, ss
->root
->hierarchy_id
,
4326 atomic_read(&ss
->root
->nr_cgrps
), !ss
->disabled
);
4328 mutex_unlock(&cgroup_mutex
);
4332 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4334 return single_open(file
, proc_cgroupstats_show
, NULL
);
4337 static const struct file_operations proc_cgroupstats_operations
= {
4338 .open
= cgroupstats_open
,
4340 .llseek
= seq_lseek
,
4341 .release
= single_release
,
4345 * cgroup_fork - initialize cgroup related fields during copy_process()
4346 * @child: pointer to task_struct of forking parent process.
4348 * A task is associated with the init_css_set until cgroup_post_fork()
4349 * attaches it to the parent's css_set. Empty cg_list indicates that
4350 * @child isn't holding reference to its css_set.
4352 void cgroup_fork(struct task_struct
*child
)
4354 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
4355 INIT_LIST_HEAD(&child
->cg_list
);
4359 * cgroup_post_fork - called on a new task after adding it to the task list
4360 * @child: the task in question
4362 * Adds the task to the list running through its css_set if necessary and
4363 * call the subsystem fork() callbacks. Has to be after the task is
4364 * visible on the task list in case we race with the first call to
4365 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
4368 void cgroup_post_fork(struct task_struct
*child
)
4370 struct cgroup_subsys
*ss
;
4374 * This may race against cgroup_enable_task_cg_links(). As that
4375 * function sets use_task_css_set_links before grabbing
4376 * tasklist_lock and we just went through tasklist_lock to add
4377 * @child, it's guaranteed that either we see the set
4378 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
4379 * @child during its iteration.
4381 * If we won the race, @child is associated with %current's
4382 * css_set. Grabbing css_set_rwsem guarantees both that the
4383 * association is stable, and, on completion of the parent's
4384 * migration, @child is visible in the source of migration or
4385 * already in the destination cgroup. This guarantee is necessary
4386 * when implementing operations which need to migrate all tasks of
4387 * a cgroup to another.
4389 * Note that if we lose to cgroup_enable_task_cg_links(), @child
4390 * will remain in init_css_set. This is safe because all tasks are
4391 * in the init_css_set before cg_links is enabled and there's no
4392 * operation which transfers all tasks out of init_css_set.
4394 if (use_task_css_set_links
) {
4395 struct css_set
*cset
;
4397 down_write(&css_set_rwsem
);
4398 cset
= task_css_set(current
);
4399 if (list_empty(&child
->cg_list
)) {
4400 rcu_assign_pointer(child
->cgroups
, cset
);
4401 list_add(&child
->cg_list
, &cset
->tasks
);
4404 up_write(&css_set_rwsem
);
4408 * Call ss->fork(). This must happen after @child is linked on
4409 * css_set; otherwise, @child might change state between ->fork()
4410 * and addition to css_set.
4412 if (need_forkexit_callback
) {
4413 for_each_subsys(ss
, i
)
4420 * cgroup_exit - detach cgroup from exiting task
4421 * @tsk: pointer to task_struct of exiting process
4423 * Description: Detach cgroup from @tsk and release it.
4425 * Note that cgroups marked notify_on_release force every task in
4426 * them to take the global cgroup_mutex mutex when exiting.
4427 * This could impact scaling on very large systems. Be reluctant to
4428 * use notify_on_release cgroups where very high task exit scaling
4429 * is required on large systems.
4431 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
4432 * call cgroup_exit() while the task is still competent to handle
4433 * notify_on_release(), then leave the task attached to the root cgroup in
4434 * each hierarchy for the remainder of its exit. No need to bother with
4435 * init_css_set refcnting. init_css_set never goes away and we can't race
4436 * with migration path - PF_EXITING is visible to migration path.
4438 void cgroup_exit(struct task_struct
*tsk
)
4440 struct cgroup_subsys
*ss
;
4441 struct css_set
*cset
;
4442 bool put_cset
= false;
4446 * Unlink from @tsk from its css_set. As migration path can't race
4447 * with us, we can check cg_list without grabbing css_set_rwsem.
4449 if (!list_empty(&tsk
->cg_list
)) {
4450 down_write(&css_set_rwsem
);
4451 list_del_init(&tsk
->cg_list
);
4452 up_write(&css_set_rwsem
);
4456 /* Reassign the task to the init_css_set. */
4457 cset
= task_css_set(tsk
);
4458 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
4460 if (need_forkexit_callback
) {
4461 /* see cgroup_post_fork() for details */
4462 for_each_subsys(ss
, i
) {
4464 struct cgroup_subsys_state
*old_css
= cset
->subsys
[i
];
4465 struct cgroup_subsys_state
*css
= task_css(tsk
, i
);
4467 ss
->exit(css
, old_css
, tsk
);
4473 put_css_set(cset
, true);
4476 static void check_for_release(struct cgroup
*cgrp
)
4478 if (cgroup_is_releasable(cgrp
) &&
4479 list_empty(&cgrp
->cset_links
) && list_empty(&cgrp
->children
)) {
4481 * Control Group is currently removeable. If it's not
4482 * already queued for a userspace notification, queue
4485 int need_schedule_work
= 0;
4487 raw_spin_lock(&release_list_lock
);
4488 if (!cgroup_is_dead(cgrp
) &&
4489 list_empty(&cgrp
->release_list
)) {
4490 list_add(&cgrp
->release_list
, &release_list
);
4491 need_schedule_work
= 1;
4493 raw_spin_unlock(&release_list_lock
);
4494 if (need_schedule_work
)
4495 schedule_work(&release_agent_work
);
4500 * Notify userspace when a cgroup is released, by running the
4501 * configured release agent with the name of the cgroup (path
4502 * relative to the root of cgroup file system) as the argument.
4504 * Most likely, this user command will try to rmdir this cgroup.
4506 * This races with the possibility that some other task will be
4507 * attached to this cgroup before it is removed, or that some other
4508 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4509 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4510 * unused, and this cgroup will be reprieved from its death sentence,
4511 * to continue to serve a useful existence. Next time it's released,
4512 * we will get notified again, if it still has 'notify_on_release' set.
4514 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4515 * means only wait until the task is successfully execve()'d. The
4516 * separate release agent task is forked by call_usermodehelper(),
4517 * then control in this thread returns here, without waiting for the
4518 * release agent task. We don't bother to wait because the caller of
4519 * this routine has no use for the exit status of the release agent
4520 * task, so no sense holding our caller up for that.
4522 static void cgroup_release_agent(struct work_struct
*work
)
4524 BUG_ON(work
!= &release_agent_work
);
4525 mutex_lock(&cgroup_mutex
);
4526 raw_spin_lock(&release_list_lock
);
4527 while (!list_empty(&release_list
)) {
4528 char *argv
[3], *envp
[3];
4530 char *pathbuf
= NULL
, *agentbuf
= NULL
, *path
;
4531 struct cgroup
*cgrp
= list_entry(release_list
.next
,
4534 list_del_init(&cgrp
->release_list
);
4535 raw_spin_unlock(&release_list_lock
);
4536 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
4539 path
= cgroup_path(cgrp
, pathbuf
, PATH_MAX
);
4542 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
4547 argv
[i
++] = agentbuf
;
4552 /* minimal command environment */
4553 envp
[i
++] = "HOME=/";
4554 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4557 /* Drop the lock while we invoke the usermode helper,
4558 * since the exec could involve hitting disk and hence
4559 * be a slow process */
4560 mutex_unlock(&cgroup_mutex
);
4561 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
4562 mutex_lock(&cgroup_mutex
);
4566 raw_spin_lock(&release_list_lock
);
4568 raw_spin_unlock(&release_list_lock
);
4569 mutex_unlock(&cgroup_mutex
);
4572 static int __init
cgroup_disable(char *str
)
4574 struct cgroup_subsys
*ss
;
4578 while ((token
= strsep(&str
, ",")) != NULL
) {
4582 for_each_subsys(ss
, i
) {
4583 if (!strcmp(token
, ss
->name
)) {
4585 printk(KERN_INFO
"Disabling %s control group"
4586 " subsystem\n", ss
->name
);
4593 __setup("cgroup_disable=", cgroup_disable
);
4596 * css_tryget_from_dir - get corresponding css from the dentry of a cgroup dir
4597 * @dentry: directory dentry of interest
4598 * @ss: subsystem of interest
4600 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
4601 * to get the corresponding css and return it. If such css doesn't exist
4602 * or can't be pinned, an ERR_PTR value is returned.
4604 struct cgroup_subsys_state
*css_tryget_from_dir(struct dentry
*dentry
,
4605 struct cgroup_subsys
*ss
)
4607 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
4608 struct cgroup_subsys_state
*css
= NULL
;
4609 struct cgroup
*cgrp
;
4611 /* is @dentry a cgroup dir? */
4612 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
4613 kernfs_type(kn
) != KERNFS_DIR
)
4614 return ERR_PTR(-EBADF
);
4619 * This path doesn't originate from kernfs and @kn could already
4620 * have been or be removed at any point. @kn->priv is RCU
4621 * protected for this access. See destroy_locked() for details.
4623 cgrp
= rcu_dereference(kn
->priv
);
4625 css
= cgroup_css(cgrp
, ss
);
4627 if (!css
|| !css_tryget(css
))
4628 css
= ERR_PTR(-ENOENT
);
4635 * css_from_id - lookup css by id
4636 * @id: the cgroup id
4637 * @ss: cgroup subsys to be looked into
4639 * Returns the css if there's valid one with @id, otherwise returns NULL.
4640 * Should be called under rcu_read_lock().
4642 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
4644 struct cgroup
*cgrp
;
4646 cgroup_assert_mutexes_or_rcu_locked();
4648 cgrp
= idr_find(&ss
->root
->cgroup_idr
, id
);
4650 return cgroup_css(cgrp
, ss
);
4654 #ifdef CONFIG_CGROUP_DEBUG
4655 static struct cgroup_subsys_state
*
4656 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
4658 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
4661 return ERR_PTR(-ENOMEM
);
4666 static void debug_css_free(struct cgroup_subsys_state
*css
)
4671 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
4674 return cgroup_task_count(css
->cgroup
);
4677 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
4680 return (u64
)(unsigned long)current
->cgroups
;
4683 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
4689 count
= atomic_read(&task_css_set(current
)->refcount
);
4694 static int current_css_set_cg_links_read(struct seq_file
*seq
, void *v
)
4696 struct cgrp_cset_link
*link
;
4697 struct css_set
*cset
;
4700 name_buf
= kmalloc(NAME_MAX
+ 1, GFP_KERNEL
);
4704 down_read(&css_set_rwsem
);
4706 cset
= rcu_dereference(current
->cgroups
);
4707 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
4708 struct cgroup
*c
= link
->cgrp
;
4710 cgroup_name(c
, name_buf
, NAME_MAX
+ 1);
4711 seq_printf(seq
, "Root %d group %s\n",
4712 c
->root
->hierarchy_id
, name_buf
);
4715 up_read(&css_set_rwsem
);
4720 #define MAX_TASKS_SHOWN_PER_CSS 25
4721 static int cgroup_css_links_read(struct seq_file
*seq
, void *v
)
4723 struct cgroup_subsys_state
*css
= seq_css(seq
);
4724 struct cgrp_cset_link
*link
;
4726 down_read(&css_set_rwsem
);
4727 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
4728 struct css_set
*cset
= link
->cset
;
4729 struct task_struct
*task
;
4732 seq_printf(seq
, "css_set %p\n", cset
);
4734 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
4735 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
4737 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
4740 list_for_each_entry(task
, &cset
->mg_tasks
, cg_list
) {
4741 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
4743 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
4747 seq_puts(seq
, " ...\n");
4749 up_read(&css_set_rwsem
);
4753 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
4755 return test_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
4758 static struct cftype debug_files
[] = {
4760 .name
= "taskcount",
4761 .read_u64
= debug_taskcount_read
,
4765 .name
= "current_css_set",
4766 .read_u64
= current_css_set_read
,
4770 .name
= "current_css_set_refcount",
4771 .read_u64
= current_css_set_refcount_read
,
4775 .name
= "current_css_set_cg_links",
4776 .seq_show
= current_css_set_cg_links_read
,
4780 .name
= "cgroup_css_links",
4781 .seq_show
= cgroup_css_links_read
,
4785 .name
= "releasable",
4786 .read_u64
= releasable_read
,
4792 struct cgroup_subsys debug_cgrp_subsys
= {
4793 .css_alloc
= debug_css_alloc
,
4794 .css_free
= debug_css_free
,
4795 .base_cftypes
= debug_files
,
4797 #endif /* CONFIG_CGROUP_DEBUG */