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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/cgroup.h>
32 #include <linux/cred.h>
33 #include <linux/ctype.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/list.h>
38 #include <linux/magic.h>
40 #include <linux/mutex.h>
41 #include <linux/mount.h>
42 #include <linux/pagemap.h>
43 #include <linux/proc_fs.h>
44 #include <linux/rcupdate.h>
45 #include <linux/sched.h>
46 #include <linux/slab.h>
47 #include <linux/spinlock.h>
48 #include <linux/rwsem.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/pid_namespace.h>
56 #include <linux/idr.h>
57 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
58 #include <linux/kthread.h>
59 #include <linux/delay.h>
61 #include <linux/atomic.h>
64 * pidlists linger the following amount before being destroyed. The goal
65 * is avoiding frequent destruction in the middle of consecutive read calls
66 * Expiring in the middle is a performance problem not a correctness one.
67 * 1 sec should be enough.
69 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
71 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
75 * cgroup_mutex is the master lock. Any modification to cgroup or its
76 * hierarchy must be performed while holding it.
78 * css_set_rwsem protects task->cgroups pointer, the list of css_set
79 * objects, and the chain of tasks off each css_set.
81 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
82 * cgroup.h can use them for lockdep annotations.
84 #ifdef CONFIG_PROVE_RCU
85 DEFINE_MUTEX(cgroup_mutex
);
86 DECLARE_RWSEM(css_set_rwsem
);
87 EXPORT_SYMBOL_GPL(cgroup_mutex
);
88 EXPORT_SYMBOL_GPL(css_set_rwsem
);
90 static DEFINE_MUTEX(cgroup_mutex
);
91 static DECLARE_RWSEM(css_set_rwsem
);
95 * Protects cgroup_idr and css_idr so that IDs can be released without
96 * grabbing cgroup_mutex.
98 static DEFINE_SPINLOCK(cgroup_idr_lock
);
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_mutex_or_rcu_locked() \
107 rcu_lockdep_assert(rcu_read_lock_held() || \
108 lockdep_is_held(&cgroup_mutex), \
109 "cgroup_mutex or RCU read lock required");
112 * cgroup destruction makes heavy use of work items and there can be a lot
113 * of concurrent destructions. Use a separate workqueue so that cgroup
114 * destruction work items don't end up filling up max_active of system_wq
115 * which may lead to deadlock.
117 static struct workqueue_struct
*cgroup_destroy_wq
;
120 * pidlist destructions need to be flushed on cgroup destruction. Use a
121 * separate workqueue as flush domain.
123 static struct workqueue_struct
*cgroup_pidlist_destroy_wq
;
125 /* generate an array of cgroup subsystem pointers */
126 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
127 static struct cgroup_subsys
*cgroup_subsys
[] = {
128 #include <linux/cgroup_subsys.h>
132 /* array of cgroup subsystem names */
133 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
134 static const char *cgroup_subsys_name
[] = {
135 #include <linux/cgroup_subsys.h>
140 * The default hierarchy, reserved for the subsystems that are otherwise
141 * unattached - it never has more than a single cgroup, and all tasks are
142 * part of that cgroup.
144 struct cgroup_root cgrp_dfl_root
;
147 * The default hierarchy always exists but is hidden until mounted for the
148 * first time. This is for backward compatibility.
150 static bool cgrp_dfl_root_visible
;
153 * Set by the boot param of the same name and makes subsystems with NULL
154 * ->dfl_files to use ->legacy_files on the default hierarchy.
156 static bool cgroup_legacy_files_on_dfl
;
158 /* some controllers are not supported in the default hierarchy */
159 static const unsigned int cgrp_dfl_root_inhibit_ss_mask
= 0
160 #ifdef CONFIG_CGROUP_DEBUG
161 | (1 << debug_cgrp_id
)
165 /* The list of hierarchy roots */
167 static LIST_HEAD(cgroup_roots
);
168 static int cgroup_root_count
;
170 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
171 static DEFINE_IDR(cgroup_hierarchy_idr
);
174 * Assign a monotonically increasing serial number to csses. It guarantees
175 * cgroups with bigger numbers are newer than those with smaller numbers.
176 * Also, as csses are always appended to the parent's ->children list, it
177 * guarantees that sibling csses are always sorted in the ascending serial
178 * number order on the list. Protected by cgroup_mutex.
180 static u64 css_serial_nr_next
= 1;
182 /* This flag indicates whether tasks in the fork and exit paths should
183 * check for fork/exit handlers to call. This avoids us having to do
184 * extra work in the fork/exit path if none of the subsystems need to
187 static int need_forkexit_callback __read_mostly
;
189 static struct cftype cgroup_dfl_base_files
[];
190 static struct cftype cgroup_legacy_base_files
[];
192 static void cgroup_put(struct cgroup
*cgrp
);
193 static int rebind_subsystems(struct cgroup_root
*dst_root
,
194 unsigned int ss_mask
);
195 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
196 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
198 static void css_release(struct percpu_ref
*ref
);
199 static void kill_css(struct cgroup_subsys_state
*css
);
200 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
202 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
);
204 /* IDR wrappers which synchronize using cgroup_idr_lock */
205 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
210 idr_preload(gfp_mask
);
211 spin_lock_bh(&cgroup_idr_lock
);
212 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
);
213 spin_unlock_bh(&cgroup_idr_lock
);
218 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
222 spin_lock_bh(&cgroup_idr_lock
);
223 ret
= idr_replace(idr
, ptr
, id
);
224 spin_unlock_bh(&cgroup_idr_lock
);
228 static void cgroup_idr_remove(struct idr
*idr
, int id
)
230 spin_lock_bh(&cgroup_idr_lock
);
232 spin_unlock_bh(&cgroup_idr_lock
);
235 static struct cgroup
*cgroup_parent(struct cgroup
*cgrp
)
237 struct cgroup_subsys_state
*parent_css
= cgrp
->self
.parent
;
240 return container_of(parent_css
, struct cgroup
, self
);
245 * cgroup_css - obtain a cgroup's css for the specified subsystem
246 * @cgrp: the cgroup of interest
247 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
249 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
250 * function must be called either under cgroup_mutex or rcu_read_lock() and
251 * the caller is responsible for pinning the returned css if it wants to
252 * keep accessing it outside the said locks. This function may return
253 * %NULL if @cgrp doesn't have @subsys_id enabled.
255 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
256 struct cgroup_subsys
*ss
)
259 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
260 lockdep_is_held(&cgroup_mutex
));
266 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
267 * @cgrp: the cgroup of interest
268 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
270 * Similar to cgroup_css() but returns the effctive css, which is defined
271 * as the matching css of the nearest ancestor including self which has @ss
272 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
273 * function is guaranteed to return non-NULL css.
275 static struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
276 struct cgroup_subsys
*ss
)
278 lockdep_assert_held(&cgroup_mutex
);
283 if (!(cgrp
->root
->subsys_mask
& (1 << ss
->id
)))
286 while (cgroup_parent(cgrp
) &&
287 !(cgroup_parent(cgrp
)->child_subsys_mask
& (1 << ss
->id
)))
288 cgrp
= cgroup_parent(cgrp
);
290 return cgroup_css(cgrp
, ss
);
293 /* convenient tests for these bits */
294 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
296 return !(cgrp
->self
.flags
& CSS_ONLINE
);
299 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
301 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
302 struct cftype
*cft
= of_cft(of
);
305 * This is open and unprotected implementation of cgroup_css().
306 * seq_css() is only called from a kernfs file operation which has
307 * an active reference on the file. Because all the subsystem
308 * files are drained before a css is disassociated with a cgroup,
309 * the matching css from the cgroup's subsys table is guaranteed to
310 * be and stay valid until the enclosing operation is complete.
313 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
317 EXPORT_SYMBOL_GPL(of_css
);
320 * cgroup_is_descendant - test ancestry
321 * @cgrp: the cgroup to be tested
322 * @ancestor: possible ancestor of @cgrp
324 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
325 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
326 * and @ancestor are accessible.
328 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
331 if (cgrp
== ancestor
)
333 cgrp
= cgroup_parent(cgrp
);
338 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
341 (1 << CGRP_RELEASABLE
) |
342 (1 << CGRP_NOTIFY_ON_RELEASE
);
343 return (cgrp
->flags
& bits
) == bits
;
346 static int notify_on_release(const struct cgroup
*cgrp
)
348 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
352 * for_each_css - iterate all css's of a cgroup
353 * @css: the iteration cursor
354 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
355 * @cgrp: the target cgroup to iterate css's of
357 * Should be called under cgroup_[tree_]mutex.
359 #define for_each_css(css, ssid, cgrp) \
360 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
361 if (!((css) = rcu_dereference_check( \
362 (cgrp)->subsys[(ssid)], \
363 lockdep_is_held(&cgroup_mutex)))) { } \
367 * for_each_e_css - iterate all effective css's of a cgroup
368 * @css: the iteration cursor
369 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
370 * @cgrp: the target cgroup to iterate css's of
372 * Should be called under cgroup_[tree_]mutex.
374 #define for_each_e_css(css, ssid, cgrp) \
375 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
376 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
381 * for_each_subsys - iterate all enabled cgroup subsystems
382 * @ss: the iteration cursor
383 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
385 #define for_each_subsys(ss, ssid) \
386 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
387 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
389 /* iterate across the hierarchies */
390 #define for_each_root(root) \
391 list_for_each_entry((root), &cgroup_roots, root_list)
393 /* iterate over child cgrps, lock should be held throughout iteration */
394 #define cgroup_for_each_live_child(child, cgrp) \
395 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
396 if (({ lockdep_assert_held(&cgroup_mutex); \
397 cgroup_is_dead(child); })) \
401 /* the list of cgroups eligible for automatic release. Protected by
402 * release_list_lock */
403 static LIST_HEAD(release_list
);
404 static DEFINE_RAW_SPINLOCK(release_list_lock
);
405 static void cgroup_release_agent(struct work_struct
*work
);
406 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
407 static void check_for_release(struct cgroup
*cgrp
);
410 * A cgroup can be associated with multiple css_sets as different tasks may
411 * belong to different cgroups on different hierarchies. In the other
412 * direction, a css_set is naturally associated with multiple cgroups.
413 * This M:N relationship is represented by the following link structure
414 * which exists for each association and allows traversing the associations
417 struct cgrp_cset_link
{
418 /* the cgroup and css_set this link associates */
420 struct css_set
*cset
;
422 /* list of cgrp_cset_links anchored at cgrp->cset_links */
423 struct list_head cset_link
;
425 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
426 struct list_head cgrp_link
;
430 * The default css_set - used by init and its children prior to any
431 * hierarchies being mounted. It contains a pointer to the root state
432 * for each subsystem. Also used to anchor the list of css_sets. Not
433 * reference-counted, to improve performance when child cgroups
434 * haven't been created.
436 struct css_set init_css_set
= {
437 .refcount
= ATOMIC_INIT(1),
438 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
439 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
440 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
441 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
442 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
445 static int css_set_count
= 1; /* 1 for init_css_set */
448 * cgroup_update_populated - updated populated count of a cgroup
449 * @cgrp: the target cgroup
450 * @populated: inc or dec populated count
452 * @cgrp is either getting the first task (css_set) or losing the last.
453 * Update @cgrp->populated_cnt accordingly. The count is propagated
454 * towards root so that a given cgroup's populated_cnt is zero iff the
455 * cgroup and all its descendants are empty.
457 * @cgrp's interface file "cgroup.populated" is zero if
458 * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
459 * changes from or to zero, userland is notified that the content of the
460 * interface file has changed. This can be used to detect when @cgrp and
461 * its descendants become populated or empty.
463 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
465 lockdep_assert_held(&css_set_rwsem
);
471 trigger
= !cgrp
->populated_cnt
++;
473 trigger
= !--cgrp
->populated_cnt
;
478 if (cgrp
->populated_kn
)
479 kernfs_notify(cgrp
->populated_kn
);
480 cgrp
= cgroup_parent(cgrp
);
485 * hash table for cgroup groups. This improves the performance to find
486 * an existing css_set. This hash doesn't (currently) take into
487 * account cgroups in empty hierarchies.
489 #define CSS_SET_HASH_BITS 7
490 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
492 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
494 unsigned long key
= 0UL;
495 struct cgroup_subsys
*ss
;
498 for_each_subsys(ss
, i
)
499 key
+= (unsigned long)css
[i
];
500 key
= (key
>> 16) ^ key
;
505 static void put_css_set_locked(struct css_set
*cset
, bool taskexit
)
507 struct cgrp_cset_link
*link
, *tmp_link
;
508 struct cgroup_subsys
*ss
;
511 lockdep_assert_held(&css_set_rwsem
);
513 if (!atomic_dec_and_test(&cset
->refcount
))
516 /* This css_set is dead. unlink it and release cgroup refcounts */
517 for_each_subsys(ss
, ssid
)
518 list_del(&cset
->e_cset_node
[ssid
]);
519 hash_del(&cset
->hlist
);
522 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
523 struct cgroup
*cgrp
= link
->cgrp
;
525 list_del(&link
->cset_link
);
526 list_del(&link
->cgrp_link
);
528 /* @cgrp can't go away while we're holding css_set_rwsem */
529 if (list_empty(&cgrp
->cset_links
)) {
530 cgroup_update_populated(cgrp
, false);
531 if (notify_on_release(cgrp
)) {
533 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
534 check_for_release(cgrp
);
541 kfree_rcu(cset
, rcu_head
);
544 static void put_css_set(struct css_set
*cset
, bool taskexit
)
547 * Ensure that the refcount doesn't hit zero while any readers
548 * can see it. Similar to atomic_dec_and_lock(), but for an
551 if (atomic_add_unless(&cset
->refcount
, -1, 1))
554 down_write(&css_set_rwsem
);
555 put_css_set_locked(cset
, taskexit
);
556 up_write(&css_set_rwsem
);
560 * refcounted get/put for css_set objects
562 static inline void get_css_set(struct css_set
*cset
)
564 atomic_inc(&cset
->refcount
);
568 * compare_css_sets - helper function for find_existing_css_set().
569 * @cset: candidate css_set being tested
570 * @old_cset: existing css_set for a task
571 * @new_cgrp: cgroup that's being entered by the task
572 * @template: desired set of css pointers in css_set (pre-calculated)
574 * Returns true if "cset" matches "old_cset" except for the hierarchy
575 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
577 static bool compare_css_sets(struct css_set
*cset
,
578 struct css_set
*old_cset
,
579 struct cgroup
*new_cgrp
,
580 struct cgroup_subsys_state
*template[])
582 struct list_head
*l1
, *l2
;
585 * On the default hierarchy, there can be csets which are
586 * associated with the same set of cgroups but different csses.
587 * Let's first ensure that csses match.
589 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
593 * Compare cgroup pointers in order to distinguish between
594 * different cgroups in hierarchies. As different cgroups may
595 * share the same effective css, this comparison is always
598 l1
= &cset
->cgrp_links
;
599 l2
= &old_cset
->cgrp_links
;
601 struct cgrp_cset_link
*link1
, *link2
;
602 struct cgroup
*cgrp1
, *cgrp2
;
606 /* See if we reached the end - both lists are equal length. */
607 if (l1
== &cset
->cgrp_links
) {
608 BUG_ON(l2
!= &old_cset
->cgrp_links
);
611 BUG_ON(l2
== &old_cset
->cgrp_links
);
613 /* Locate the cgroups associated with these links. */
614 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
615 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
618 /* Hierarchies should be linked in the same order. */
619 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
622 * If this hierarchy is the hierarchy of the cgroup
623 * that's changing, then we need to check that this
624 * css_set points to the new cgroup; if it's any other
625 * hierarchy, then this css_set should point to the
626 * same cgroup as the old css_set.
628 if (cgrp1
->root
== new_cgrp
->root
) {
629 if (cgrp1
!= new_cgrp
)
640 * find_existing_css_set - init css array and find the matching css_set
641 * @old_cset: the css_set that we're using before the cgroup transition
642 * @cgrp: the cgroup that we're moving into
643 * @template: out param for the new set of csses, should be clear on entry
645 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
647 struct cgroup_subsys_state
*template[])
649 struct cgroup_root
*root
= cgrp
->root
;
650 struct cgroup_subsys
*ss
;
651 struct css_set
*cset
;
656 * Build the set of subsystem state objects that we want to see in the
657 * new css_set. while subsystems can change globally, the entries here
658 * won't change, so no need for locking.
660 for_each_subsys(ss
, i
) {
661 if (root
->subsys_mask
& (1UL << i
)) {
663 * @ss is in this hierarchy, so we want the
664 * effective css from @cgrp.
666 template[i
] = cgroup_e_css(cgrp
, ss
);
669 * @ss is not in this hierarchy, so we don't want
672 template[i
] = old_cset
->subsys
[i
];
676 key
= css_set_hash(template);
677 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
678 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
681 /* This css_set matches what we need */
685 /* No existing cgroup group matched */
689 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
691 struct cgrp_cset_link
*link
, *tmp_link
;
693 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
694 list_del(&link
->cset_link
);
700 * allocate_cgrp_cset_links - allocate cgrp_cset_links
701 * @count: the number of links to allocate
702 * @tmp_links: list_head the allocated links are put on
704 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
705 * through ->cset_link. Returns 0 on success or -errno.
707 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
709 struct cgrp_cset_link
*link
;
712 INIT_LIST_HEAD(tmp_links
);
714 for (i
= 0; i
< count
; i
++) {
715 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
717 free_cgrp_cset_links(tmp_links
);
720 list_add(&link
->cset_link
, tmp_links
);
726 * link_css_set - a helper function to link a css_set to a cgroup
727 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
728 * @cset: the css_set to be linked
729 * @cgrp: the destination cgroup
731 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
734 struct cgrp_cset_link
*link
;
736 BUG_ON(list_empty(tmp_links
));
738 if (cgroup_on_dfl(cgrp
))
739 cset
->dfl_cgrp
= cgrp
;
741 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
745 if (list_empty(&cgrp
->cset_links
))
746 cgroup_update_populated(cgrp
, true);
747 list_move(&link
->cset_link
, &cgrp
->cset_links
);
750 * Always add links to the tail of the list so that the list
751 * is sorted by order of hierarchy creation
753 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
757 * find_css_set - return a new css_set with one cgroup updated
758 * @old_cset: the baseline css_set
759 * @cgrp: the cgroup to be updated
761 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
762 * substituted into the appropriate hierarchy.
764 static struct css_set
*find_css_set(struct css_set
*old_cset
,
767 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
768 struct css_set
*cset
;
769 struct list_head tmp_links
;
770 struct cgrp_cset_link
*link
;
771 struct cgroup_subsys
*ss
;
775 lockdep_assert_held(&cgroup_mutex
);
777 /* First see if we already have a cgroup group that matches
779 down_read(&css_set_rwsem
);
780 cset
= find_existing_css_set(old_cset
, cgrp
, template);
783 up_read(&css_set_rwsem
);
788 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
792 /* Allocate all the cgrp_cset_link objects that we'll need */
793 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
798 atomic_set(&cset
->refcount
, 1);
799 INIT_LIST_HEAD(&cset
->cgrp_links
);
800 INIT_LIST_HEAD(&cset
->tasks
);
801 INIT_LIST_HEAD(&cset
->mg_tasks
);
802 INIT_LIST_HEAD(&cset
->mg_preload_node
);
803 INIT_LIST_HEAD(&cset
->mg_node
);
804 INIT_HLIST_NODE(&cset
->hlist
);
806 /* Copy the set of subsystem state objects generated in
807 * find_existing_css_set() */
808 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
810 down_write(&css_set_rwsem
);
811 /* Add reference counts and links from the new css_set. */
812 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
813 struct cgroup
*c
= link
->cgrp
;
815 if (c
->root
== cgrp
->root
)
817 link_css_set(&tmp_links
, cset
, c
);
820 BUG_ON(!list_empty(&tmp_links
));
824 /* Add @cset to the hash table */
825 key
= css_set_hash(cset
->subsys
);
826 hash_add(css_set_table
, &cset
->hlist
, key
);
828 for_each_subsys(ss
, ssid
)
829 list_add_tail(&cset
->e_cset_node
[ssid
],
830 &cset
->subsys
[ssid
]->cgroup
->e_csets
[ssid
]);
832 up_write(&css_set_rwsem
);
837 static struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
839 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
841 return root_cgrp
->root
;
844 static int cgroup_init_root_id(struct cgroup_root
*root
)
848 lockdep_assert_held(&cgroup_mutex
);
850 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
854 root
->hierarchy_id
= id
;
858 static void cgroup_exit_root_id(struct cgroup_root
*root
)
860 lockdep_assert_held(&cgroup_mutex
);
862 if (root
->hierarchy_id
) {
863 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
864 root
->hierarchy_id
= 0;
868 static void cgroup_free_root(struct cgroup_root
*root
)
871 /* hierarhcy ID shoulid already have been released */
872 WARN_ON_ONCE(root
->hierarchy_id
);
874 idr_destroy(&root
->cgroup_idr
);
879 static void cgroup_destroy_root(struct cgroup_root
*root
)
881 struct cgroup
*cgrp
= &root
->cgrp
;
882 struct cgrp_cset_link
*link
, *tmp_link
;
884 mutex_lock(&cgroup_mutex
);
886 BUG_ON(atomic_read(&root
->nr_cgrps
));
887 BUG_ON(!list_empty(&cgrp
->self
.children
));
889 /* Rebind all subsystems back to the default hierarchy */
890 rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
);
893 * Release all the links from cset_links to this hierarchy's
896 down_write(&css_set_rwsem
);
898 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
899 list_del(&link
->cset_link
);
900 list_del(&link
->cgrp_link
);
903 up_write(&css_set_rwsem
);
905 if (!list_empty(&root
->root_list
)) {
906 list_del(&root
->root_list
);
910 cgroup_exit_root_id(root
);
912 mutex_unlock(&cgroup_mutex
);
914 kernfs_destroy_root(root
->kf_root
);
915 cgroup_free_root(root
);
918 /* look up cgroup associated with given css_set on the specified hierarchy */
919 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
920 struct cgroup_root
*root
)
922 struct cgroup
*res
= NULL
;
924 lockdep_assert_held(&cgroup_mutex
);
925 lockdep_assert_held(&css_set_rwsem
);
927 if (cset
== &init_css_set
) {
930 struct cgrp_cset_link
*link
;
932 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
933 struct cgroup
*c
= link
->cgrp
;
935 if (c
->root
== root
) {
947 * Return the cgroup for "task" from the given hierarchy. Must be
948 * called with cgroup_mutex and css_set_rwsem held.
950 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
951 struct cgroup_root
*root
)
954 * No need to lock the task - since we hold cgroup_mutex the
955 * task can't change groups, so the only thing that can happen
956 * is that it exits and its css is set back to init_css_set.
958 return cset_cgroup_from_root(task_css_set(task
), root
);
962 * A task must hold cgroup_mutex to modify cgroups.
964 * Any task can increment and decrement the count field without lock.
965 * So in general, code holding cgroup_mutex can't rely on the count
966 * field not changing. However, if the count goes to zero, then only
967 * cgroup_attach_task() can increment it again. Because a count of zero
968 * means that no tasks are currently attached, therefore there is no
969 * way a task attached to that cgroup can fork (the other way to
970 * increment the count). So code holding cgroup_mutex can safely
971 * assume that if the count is zero, it will stay zero. Similarly, if
972 * a task holds cgroup_mutex on a cgroup with zero count, it
973 * knows that the cgroup won't be removed, as cgroup_rmdir()
976 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
977 * (usually) take cgroup_mutex. These are the two most performance
978 * critical pieces of code here. The exception occurs on cgroup_exit(),
979 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
980 * is taken, and if the cgroup count is zero, a usermode call made
981 * to the release agent with the name of the cgroup (path relative to
982 * the root of cgroup file system) as the argument.
984 * A cgroup can only be deleted if both its 'count' of using tasks
985 * is zero, and its list of 'children' cgroups is empty. Since all
986 * tasks in the system use _some_ cgroup, and since there is always at
987 * least one task in the system (init, pid == 1), therefore, root cgroup
988 * always has either children cgroups and/or using tasks. So we don't
989 * need a special hack to ensure that root cgroup cannot be deleted.
991 * P.S. One more locking exception. RCU is used to guard the
992 * update of a tasks cgroup pointer by cgroup_attach_task()
995 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned int subsys_mask
);
996 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
997 static const struct file_operations proc_cgroupstats_operations
;
999 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1002 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1003 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
))
1004 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s.%s",
1005 cft
->ss
->name
, cft
->name
);
1007 strncpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1012 * cgroup_file_mode - deduce file mode of a control file
1013 * @cft: the control file in question
1015 * returns cft->mode if ->mode is not 0
1016 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
1017 * returns S_IRUGO if it has only a read handler
1018 * returns S_IWUSR if it has only a write hander
1020 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1027 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1030 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
)
1036 static void cgroup_get(struct cgroup
*cgrp
)
1038 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
1039 css_get(&cgrp
->self
);
1042 static void cgroup_put(struct cgroup
*cgrp
)
1044 css_put(&cgrp
->self
);
1048 * cgroup_refresh_child_subsys_mask - update child_subsys_mask
1049 * @cgrp: the target cgroup
1051 * On the default hierarchy, a subsystem may request other subsystems to be
1052 * enabled together through its ->depends_on mask. In such cases, more
1053 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1055 * This function determines which subsystems need to be enabled given the
1056 * current @cgrp->subtree_control and records it in
1057 * @cgrp->child_subsys_mask. The resulting mask is always a superset of
1058 * @cgrp->subtree_control and follows the usual hierarchy rules.
1060 static void cgroup_refresh_child_subsys_mask(struct cgroup
*cgrp
)
1062 struct cgroup
*parent
= cgroup_parent(cgrp
);
1063 unsigned int cur_ss_mask
= cgrp
->subtree_control
;
1064 struct cgroup_subsys
*ss
;
1067 lockdep_assert_held(&cgroup_mutex
);
1069 if (!cgroup_on_dfl(cgrp
)) {
1070 cgrp
->child_subsys_mask
= cur_ss_mask
;
1075 unsigned int new_ss_mask
= cur_ss_mask
;
1077 for_each_subsys(ss
, ssid
)
1078 if (cur_ss_mask
& (1 << ssid
))
1079 new_ss_mask
|= ss
->depends_on
;
1082 * Mask out subsystems which aren't available. This can
1083 * happen only if some depended-upon subsystems were bound
1084 * to non-default hierarchies.
1087 new_ss_mask
&= parent
->child_subsys_mask
;
1089 new_ss_mask
&= cgrp
->root
->subsys_mask
;
1091 if (new_ss_mask
== cur_ss_mask
)
1093 cur_ss_mask
= new_ss_mask
;
1096 cgrp
->child_subsys_mask
= cur_ss_mask
;
1100 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1101 * @kn: the kernfs_node being serviced
1103 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1104 * the method finishes if locking succeeded. Note that once this function
1105 * returns the cgroup returned by cgroup_kn_lock_live() may become
1106 * inaccessible any time. If the caller intends to continue to access the
1107 * cgroup, it should pin it before invoking this function.
1109 static void cgroup_kn_unlock(struct kernfs_node
*kn
)
1111 struct cgroup
*cgrp
;
1113 if (kernfs_type(kn
) == KERNFS_DIR
)
1116 cgrp
= kn
->parent
->priv
;
1118 mutex_unlock(&cgroup_mutex
);
1120 kernfs_unbreak_active_protection(kn
);
1125 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1126 * @kn: the kernfs_node being serviced
1128 * This helper is to be used by a cgroup kernfs method currently servicing
1129 * @kn. It breaks the active protection, performs cgroup locking and
1130 * verifies that the associated cgroup is alive. Returns the cgroup if
1131 * alive; otherwise, %NULL. A successful return should be undone by a
1132 * matching cgroup_kn_unlock() invocation.
1134 * Any cgroup kernfs method implementation which requires locking the
1135 * associated cgroup should use this helper. It avoids nesting cgroup
1136 * locking under kernfs active protection and allows all kernfs operations
1137 * including self-removal.
1139 static struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
)
1141 struct cgroup
*cgrp
;
1143 if (kernfs_type(kn
) == KERNFS_DIR
)
1146 cgrp
= kn
->parent
->priv
;
1149 * We're gonna grab cgroup_mutex which nests outside kernfs
1150 * active_ref. cgroup liveliness check alone provides enough
1151 * protection against removal. Ensure @cgrp stays accessible and
1152 * break the active_ref protection.
1155 kernfs_break_active_protection(kn
);
1157 mutex_lock(&cgroup_mutex
);
1159 if (!cgroup_is_dead(cgrp
))
1162 cgroup_kn_unlock(kn
);
1166 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1168 char name
[CGROUP_FILE_NAME_MAX
];
1170 lockdep_assert_held(&cgroup_mutex
);
1171 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1175 * cgroup_clear_dir - remove subsys files in a cgroup directory
1176 * @cgrp: target cgroup
1177 * @subsys_mask: mask of the subsystem ids whose files should be removed
1179 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned int subsys_mask
)
1181 struct cgroup_subsys
*ss
;
1184 for_each_subsys(ss
, i
) {
1185 struct cftype
*cfts
;
1187 if (!(subsys_mask
& (1 << i
)))
1189 list_for_each_entry(cfts
, &ss
->cfts
, node
)
1190 cgroup_addrm_files(cgrp
, cfts
, false);
1194 static int rebind_subsystems(struct cgroup_root
*dst_root
, unsigned int ss_mask
)
1196 struct cgroup_subsys
*ss
;
1197 unsigned int tmp_ss_mask
;
1200 lockdep_assert_held(&cgroup_mutex
);
1202 for_each_subsys(ss
, ssid
) {
1203 if (!(ss_mask
& (1 << ssid
)))
1206 /* if @ss has non-root csses attached to it, can't move */
1207 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)))
1210 /* can't move between two non-dummy roots either */
1211 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1215 /* skip creating root files on dfl_root for inhibited subsystems */
1216 tmp_ss_mask
= ss_mask
;
1217 if (dst_root
== &cgrp_dfl_root
)
1218 tmp_ss_mask
&= ~cgrp_dfl_root_inhibit_ss_mask
;
1220 ret
= cgroup_populate_dir(&dst_root
->cgrp
, tmp_ss_mask
);
1222 if (dst_root
!= &cgrp_dfl_root
)
1226 * Rebinding back to the default root is not allowed to
1227 * fail. Using both default and non-default roots should
1228 * be rare. Moving subsystems back and forth even more so.
1229 * Just warn about it and continue.
1231 if (cgrp_dfl_root_visible
) {
1232 pr_warn("failed to create files (%d) while rebinding 0x%x to default root\n",
1234 pr_warn("you may retry by moving them to a different hierarchy and unbinding\n");
1239 * Nothing can fail from this point on. Remove files for the
1240 * removed subsystems and rebind each subsystem.
1242 for_each_subsys(ss
, ssid
)
1243 if (ss_mask
& (1 << ssid
))
1244 cgroup_clear_dir(&ss
->root
->cgrp
, 1 << ssid
);
1246 for_each_subsys(ss
, ssid
) {
1247 struct cgroup_root
*src_root
;
1248 struct cgroup_subsys_state
*css
;
1249 struct css_set
*cset
;
1251 if (!(ss_mask
& (1 << ssid
)))
1254 src_root
= ss
->root
;
1255 css
= cgroup_css(&src_root
->cgrp
, ss
);
1257 WARN_ON(!css
|| cgroup_css(&dst_root
->cgrp
, ss
));
1259 RCU_INIT_POINTER(src_root
->cgrp
.subsys
[ssid
], NULL
);
1260 rcu_assign_pointer(dst_root
->cgrp
.subsys
[ssid
], css
);
1261 ss
->root
= dst_root
;
1262 css
->cgroup
= &dst_root
->cgrp
;
1264 down_write(&css_set_rwsem
);
1265 hash_for_each(css_set_table
, i
, cset
, hlist
)
1266 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1267 &dst_root
->cgrp
.e_csets
[ss
->id
]);
1268 up_write(&css_set_rwsem
);
1270 src_root
->subsys_mask
&= ~(1 << ssid
);
1271 src_root
->cgrp
.subtree_control
&= ~(1 << ssid
);
1272 cgroup_refresh_child_subsys_mask(&src_root
->cgrp
);
1274 /* default hierarchy doesn't enable controllers by default */
1275 dst_root
->subsys_mask
|= 1 << ssid
;
1276 if (dst_root
!= &cgrp_dfl_root
) {
1277 dst_root
->cgrp
.subtree_control
|= 1 << ssid
;
1278 cgroup_refresh_child_subsys_mask(&dst_root
->cgrp
);
1285 kernfs_activate(dst_root
->cgrp
.kn
);
1289 static int cgroup_show_options(struct seq_file
*seq
,
1290 struct kernfs_root
*kf_root
)
1292 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1293 struct cgroup_subsys
*ss
;
1296 for_each_subsys(ss
, ssid
)
1297 if (root
->subsys_mask
& (1 << ssid
))
1298 seq_printf(seq
, ",%s", ss
->name
);
1299 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1300 seq_puts(seq
, ",noprefix");
1301 if (root
->flags
& CGRP_ROOT_XATTR
)
1302 seq_puts(seq
, ",xattr");
1304 spin_lock(&release_agent_path_lock
);
1305 if (strlen(root
->release_agent_path
))
1306 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1307 spin_unlock(&release_agent_path_lock
);
1309 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
1310 seq_puts(seq
, ",clone_children");
1311 if (strlen(root
->name
))
1312 seq_printf(seq
, ",name=%s", root
->name
);
1316 struct cgroup_sb_opts
{
1317 unsigned int subsys_mask
;
1319 char *release_agent
;
1320 bool cpuset_clone_children
;
1322 /* User explicitly requested empty subsystem */
1326 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1328 char *token
, *o
= data
;
1329 bool all_ss
= false, one_ss
= false;
1330 unsigned int mask
= -1U;
1331 struct cgroup_subsys
*ss
;
1335 #ifdef CONFIG_CPUSETS
1336 mask
= ~(1U << cpuset_cgrp_id
);
1339 memset(opts
, 0, sizeof(*opts
));
1341 while ((token
= strsep(&o
, ",")) != NULL
) {
1346 if (!strcmp(token
, "none")) {
1347 /* Explicitly have no subsystems */
1351 if (!strcmp(token
, "all")) {
1352 /* Mutually exclusive option 'all' + subsystem name */
1358 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1359 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1362 if (!strcmp(token
, "noprefix")) {
1363 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1366 if (!strcmp(token
, "clone_children")) {
1367 opts
->cpuset_clone_children
= true;
1370 if (!strcmp(token
, "xattr")) {
1371 opts
->flags
|= CGRP_ROOT_XATTR
;
1374 if (!strncmp(token
, "release_agent=", 14)) {
1375 /* Specifying two release agents is forbidden */
1376 if (opts
->release_agent
)
1378 opts
->release_agent
=
1379 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1380 if (!opts
->release_agent
)
1384 if (!strncmp(token
, "name=", 5)) {
1385 const char *name
= token
+ 5;
1386 /* Can't specify an empty name */
1389 /* Must match [\w.-]+ */
1390 for (i
= 0; i
< strlen(name
); i
++) {
1394 if ((c
== '.') || (c
== '-') || (c
== '_'))
1398 /* Specifying two names is forbidden */
1401 opts
->name
= kstrndup(name
,
1402 MAX_CGROUP_ROOT_NAMELEN
- 1,
1410 for_each_subsys(ss
, i
) {
1411 if (strcmp(token
, ss
->name
))
1416 /* Mutually exclusive option 'all' + subsystem name */
1419 opts
->subsys_mask
|= (1 << i
);
1424 if (i
== CGROUP_SUBSYS_COUNT
)
1428 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1429 pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1431 pr_err("sane_behavior: no other mount options allowed\n");
1438 * If the 'all' option was specified select all the subsystems,
1439 * otherwise if 'none', 'name=' and a subsystem name options were
1440 * not specified, let's default to 'all'
1442 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1443 for_each_subsys(ss
, i
)
1445 opts
->subsys_mask
|= (1 << i
);
1448 * We either have to specify by name or by subsystems. (So all
1449 * empty hierarchies must have a name).
1451 if (!opts
->subsys_mask
&& !opts
->name
)
1455 * Option noprefix was introduced just for backward compatibility
1456 * with the old cpuset, so we allow noprefix only if mounting just
1457 * the cpuset subsystem.
1459 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1462 /* Can't specify "none" and some subsystems */
1463 if (opts
->subsys_mask
&& opts
->none
)
1469 static int cgroup_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1472 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1473 struct cgroup_sb_opts opts
;
1474 unsigned int added_mask
, removed_mask
;
1476 if (root
== &cgrp_dfl_root
) {
1477 pr_err("remount is not allowed\n");
1481 mutex_lock(&cgroup_mutex
);
1483 /* See what subsystems are wanted */
1484 ret
= parse_cgroupfs_options(data
, &opts
);
1488 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1489 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1490 task_tgid_nr(current
), current
->comm
);
1492 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1493 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1495 /* Don't allow flags or name to change at remount */
1496 if ((opts
.flags
^ root
->flags
) ||
1497 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1498 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1499 opts
.flags
, opts
.name
?: "", root
->flags
, root
->name
);
1504 /* remounting is not allowed for populated hierarchies */
1505 if (!list_empty(&root
->cgrp
.self
.children
)) {
1510 ret
= rebind_subsystems(root
, added_mask
);
1514 rebind_subsystems(&cgrp_dfl_root
, removed_mask
);
1516 if (opts
.release_agent
) {
1517 spin_lock(&release_agent_path_lock
);
1518 strcpy(root
->release_agent_path
, opts
.release_agent
);
1519 spin_unlock(&release_agent_path_lock
);
1522 kfree(opts
.release_agent
);
1524 mutex_unlock(&cgroup_mutex
);
1529 * To reduce the fork() overhead for systems that are not actually using
1530 * their cgroups capability, we don't maintain the lists running through
1531 * each css_set to its tasks until we see the list actually used - in other
1532 * words after the first mount.
1534 static bool use_task_css_set_links __read_mostly
;
1536 static void cgroup_enable_task_cg_lists(void)
1538 struct task_struct
*p
, *g
;
1540 down_write(&css_set_rwsem
);
1542 if (use_task_css_set_links
)
1545 use_task_css_set_links
= true;
1548 * We need tasklist_lock because RCU is not safe against
1549 * while_each_thread(). Besides, a forking task that has passed
1550 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1551 * is not guaranteed to have its child immediately visible in the
1552 * tasklist if we walk through it with RCU.
1554 read_lock(&tasklist_lock
);
1555 do_each_thread(g
, p
) {
1556 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1557 task_css_set(p
) != &init_css_set
);
1560 * We should check if the process is exiting, otherwise
1561 * it will race with cgroup_exit() in that the list
1562 * entry won't be deleted though the process has exited.
1563 * Do it while holding siglock so that we don't end up
1564 * racing against cgroup_exit().
1566 spin_lock_irq(&p
->sighand
->siglock
);
1567 if (!(p
->flags
& PF_EXITING
)) {
1568 struct css_set
*cset
= task_css_set(p
);
1570 list_add(&p
->cg_list
, &cset
->tasks
);
1573 spin_unlock_irq(&p
->sighand
->siglock
);
1574 } while_each_thread(g
, p
);
1575 read_unlock(&tasklist_lock
);
1577 up_write(&css_set_rwsem
);
1580 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1582 struct cgroup_subsys
*ss
;
1585 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1586 INIT_LIST_HEAD(&cgrp
->self
.children
);
1587 INIT_LIST_HEAD(&cgrp
->cset_links
);
1588 INIT_LIST_HEAD(&cgrp
->release_list
);
1589 INIT_LIST_HEAD(&cgrp
->pidlists
);
1590 mutex_init(&cgrp
->pidlist_mutex
);
1591 cgrp
->self
.cgroup
= cgrp
;
1592 cgrp
->self
.flags
|= CSS_ONLINE
;
1594 for_each_subsys(ss
, ssid
)
1595 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1597 init_waitqueue_head(&cgrp
->offline_waitq
);
1600 static void init_cgroup_root(struct cgroup_root
*root
,
1601 struct cgroup_sb_opts
*opts
)
1603 struct cgroup
*cgrp
= &root
->cgrp
;
1605 INIT_LIST_HEAD(&root
->root_list
);
1606 atomic_set(&root
->nr_cgrps
, 1);
1608 init_cgroup_housekeeping(cgrp
);
1609 idr_init(&root
->cgroup_idr
);
1611 root
->flags
= opts
->flags
;
1612 if (opts
->release_agent
)
1613 strcpy(root
->release_agent_path
, opts
->release_agent
);
1615 strcpy(root
->name
, opts
->name
);
1616 if (opts
->cpuset_clone_children
)
1617 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1620 static int cgroup_setup_root(struct cgroup_root
*root
, unsigned int ss_mask
)
1622 LIST_HEAD(tmp_links
);
1623 struct cgroup
*root_cgrp
= &root
->cgrp
;
1624 struct cftype
*base_files
;
1625 struct css_set
*cset
;
1628 lockdep_assert_held(&cgroup_mutex
);
1630 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_NOWAIT
);
1633 root_cgrp
->id
= ret
;
1635 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
);
1640 * We're accessing css_set_count without locking css_set_rwsem here,
1641 * but that's OK - it can only be increased by someone holding
1642 * cgroup_lock, and that's us. The worst that can happen is that we
1643 * have some link structures left over
1645 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1649 ret
= cgroup_init_root_id(root
);
1653 root
->kf_root
= kernfs_create_root(&cgroup_kf_syscall_ops
,
1654 KERNFS_ROOT_CREATE_DEACTIVATED
,
1656 if (IS_ERR(root
->kf_root
)) {
1657 ret
= PTR_ERR(root
->kf_root
);
1660 root_cgrp
->kn
= root
->kf_root
->kn
;
1662 if (root
== &cgrp_dfl_root
)
1663 base_files
= cgroup_dfl_base_files
;
1665 base_files
= cgroup_legacy_base_files
;
1667 ret
= cgroup_addrm_files(root_cgrp
, base_files
, true);
1671 ret
= rebind_subsystems(root
, ss_mask
);
1676 * There must be no failure case after here, since rebinding takes
1677 * care of subsystems' refcounts, which are explicitly dropped in
1678 * the failure exit path.
1680 list_add(&root
->root_list
, &cgroup_roots
);
1681 cgroup_root_count
++;
1684 * Link the root cgroup in this hierarchy into all the css_set
1687 down_write(&css_set_rwsem
);
1688 hash_for_each(css_set_table
, i
, cset
, hlist
)
1689 link_css_set(&tmp_links
, cset
, root_cgrp
);
1690 up_write(&css_set_rwsem
);
1692 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
1693 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
1695 kernfs_activate(root_cgrp
->kn
);
1700 kernfs_destroy_root(root
->kf_root
);
1701 root
->kf_root
= NULL
;
1703 cgroup_exit_root_id(root
);
1705 percpu_ref_cancel_init(&root_cgrp
->self
.refcnt
);
1707 free_cgrp_cset_links(&tmp_links
);
1711 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1712 int flags
, const char *unused_dev_name
,
1715 struct cgroup_root
*root
;
1716 struct cgroup_sb_opts opts
;
1717 struct dentry
*dentry
;
1722 * The first time anyone tries to mount a cgroup, enable the list
1723 * linking each css_set to its tasks and fix up all existing tasks.
1725 if (!use_task_css_set_links
)
1726 cgroup_enable_task_cg_lists();
1728 mutex_lock(&cgroup_mutex
);
1730 /* First find the desired set of subsystems */
1731 ret
= parse_cgroupfs_options(data
, &opts
);
1735 /* look for a matching existing root */
1736 if (opts
.flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1737 cgrp_dfl_root_visible
= true;
1738 root
= &cgrp_dfl_root
;
1739 cgroup_get(&root
->cgrp
);
1744 for_each_root(root
) {
1745 bool name_match
= false;
1747 if (root
== &cgrp_dfl_root
)
1751 * If we asked for a name then it must match. Also, if
1752 * name matches but sybsys_mask doesn't, we should fail.
1753 * Remember whether name matched.
1756 if (strcmp(opts
.name
, root
->name
))
1762 * If we asked for subsystems (or explicitly for no
1763 * subsystems) then they must match.
1765 if ((opts
.subsys_mask
|| opts
.none
) &&
1766 (opts
.subsys_mask
!= root
->subsys_mask
)) {
1773 if (root
->flags
^ opts
.flags
)
1774 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1777 * A root's lifetime is governed by its root cgroup.
1778 * tryget_live failure indicate that the root is being
1779 * destroyed. Wait for destruction to complete so that the
1780 * subsystems are free. We can use wait_queue for the wait
1781 * but this path is super cold. Let's just sleep for a bit
1784 if (!percpu_ref_tryget_live(&root
->cgrp
.self
.refcnt
)) {
1785 mutex_unlock(&cgroup_mutex
);
1787 ret
= restart_syscall();
1796 * No such thing, create a new one. name= matching without subsys
1797 * specification is allowed for already existing hierarchies but we
1798 * can't create new one without subsys specification.
1800 if (!opts
.subsys_mask
&& !opts
.none
) {
1805 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1811 init_cgroup_root(root
, &opts
);
1813 ret
= cgroup_setup_root(root
, opts
.subsys_mask
);
1815 cgroup_free_root(root
);
1818 mutex_unlock(&cgroup_mutex
);
1820 kfree(opts
.release_agent
);
1824 return ERR_PTR(ret
);
1826 dentry
= kernfs_mount(fs_type
, flags
, root
->kf_root
,
1827 CGROUP_SUPER_MAGIC
, &new_sb
);
1828 if (IS_ERR(dentry
) || !new_sb
)
1829 cgroup_put(&root
->cgrp
);
1833 static void cgroup_kill_sb(struct super_block
*sb
)
1835 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
1836 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1839 * If @root doesn't have any mounts or children, start killing it.
1840 * This prevents new mounts by disabling percpu_ref_tryget_live().
1841 * cgroup_mount() may wait for @root's release.
1843 * And don't kill the default root.
1845 if (css_has_online_children(&root
->cgrp
.self
) ||
1846 root
== &cgrp_dfl_root
)
1847 cgroup_put(&root
->cgrp
);
1849 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
1854 static struct file_system_type cgroup_fs_type
= {
1856 .mount
= cgroup_mount
,
1857 .kill_sb
= cgroup_kill_sb
,
1860 static struct kobject
*cgroup_kobj
;
1863 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1864 * @task: target task
1865 * @buf: the buffer to write the path into
1866 * @buflen: the length of the buffer
1868 * Determine @task's cgroup on the first (the one with the lowest non-zero
1869 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1870 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1871 * cgroup controller callbacks.
1873 * Return value is the same as kernfs_path().
1875 char *task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
1877 struct cgroup_root
*root
;
1878 struct cgroup
*cgrp
;
1879 int hierarchy_id
= 1;
1882 mutex_lock(&cgroup_mutex
);
1883 down_read(&css_set_rwsem
);
1885 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
1888 cgrp
= task_cgroup_from_root(task
, root
);
1889 path
= cgroup_path(cgrp
, buf
, buflen
);
1891 /* if no hierarchy exists, everyone is in "/" */
1892 if (strlcpy(buf
, "/", buflen
) < buflen
)
1896 up_read(&css_set_rwsem
);
1897 mutex_unlock(&cgroup_mutex
);
1900 EXPORT_SYMBOL_GPL(task_cgroup_path
);
1902 /* used to track tasks and other necessary states during migration */
1903 struct cgroup_taskset
{
1904 /* the src and dst cset list running through cset->mg_node */
1905 struct list_head src_csets
;
1906 struct list_head dst_csets
;
1909 * Fields for cgroup_taskset_*() iteration.
1911 * Before migration is committed, the target migration tasks are on
1912 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
1913 * the csets on ->dst_csets. ->csets point to either ->src_csets
1914 * or ->dst_csets depending on whether migration is committed.
1916 * ->cur_csets and ->cur_task point to the current task position
1919 struct list_head
*csets
;
1920 struct css_set
*cur_cset
;
1921 struct task_struct
*cur_task
;
1925 * cgroup_taskset_first - reset taskset and return the first task
1926 * @tset: taskset of interest
1928 * @tset iteration is initialized and the first task is returned.
1930 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1932 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
1933 tset
->cur_task
= NULL
;
1935 return cgroup_taskset_next(tset
);
1939 * cgroup_taskset_next - iterate to the next task in taskset
1940 * @tset: taskset of interest
1942 * Return the next task in @tset. Iteration must have been initialized
1943 * with cgroup_taskset_first().
1945 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1947 struct css_set
*cset
= tset
->cur_cset
;
1948 struct task_struct
*task
= tset
->cur_task
;
1950 while (&cset
->mg_node
!= tset
->csets
) {
1952 task
= list_first_entry(&cset
->mg_tasks
,
1953 struct task_struct
, cg_list
);
1955 task
= list_next_entry(task
, cg_list
);
1957 if (&task
->cg_list
!= &cset
->mg_tasks
) {
1958 tset
->cur_cset
= cset
;
1959 tset
->cur_task
= task
;
1963 cset
= list_next_entry(cset
, mg_node
);
1971 * cgroup_task_migrate - move a task from one cgroup to another.
1972 * @old_cgrp: the cgroup @tsk is being migrated from
1973 * @tsk: the task being migrated
1974 * @new_cset: the new css_set @tsk is being attached to
1976 * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
1978 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
1979 struct task_struct
*tsk
,
1980 struct css_set
*new_cset
)
1982 struct css_set
*old_cset
;
1984 lockdep_assert_held(&cgroup_mutex
);
1985 lockdep_assert_held(&css_set_rwsem
);
1988 * We are synchronized through threadgroup_lock() against PF_EXITING
1989 * setting such that we can't race against cgroup_exit() changing the
1990 * css_set to init_css_set and dropping the old one.
1992 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1993 old_cset
= task_css_set(tsk
);
1995 get_css_set(new_cset
);
1996 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
1999 * Use move_tail so that cgroup_taskset_first() still returns the
2000 * leader after migration. This works because cgroup_migrate()
2001 * ensures that the dst_cset of the leader is the first on the
2002 * tset's dst_csets list.
2004 list_move_tail(&tsk
->cg_list
, &new_cset
->mg_tasks
);
2007 * We just gained a reference on old_cset by taking it from the
2008 * task. As trading it for new_cset is protected by cgroup_mutex,
2009 * we're safe to drop it here; it will be freed under RCU.
2011 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
2012 put_css_set_locked(old_cset
, false);
2016 * cgroup_migrate_finish - cleanup after attach
2017 * @preloaded_csets: list of preloaded css_sets
2019 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2020 * those functions for details.
2022 static void cgroup_migrate_finish(struct list_head
*preloaded_csets
)
2024 struct css_set
*cset
, *tmp_cset
;
2026 lockdep_assert_held(&cgroup_mutex
);
2028 down_write(&css_set_rwsem
);
2029 list_for_each_entry_safe(cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2030 cset
->mg_src_cgrp
= NULL
;
2031 cset
->mg_dst_cset
= NULL
;
2032 list_del_init(&cset
->mg_preload_node
);
2033 put_css_set_locked(cset
, false);
2035 up_write(&css_set_rwsem
);
2039 * cgroup_migrate_add_src - add a migration source css_set
2040 * @src_cset: the source css_set to add
2041 * @dst_cgrp: the destination cgroup
2042 * @preloaded_csets: list of preloaded css_sets
2044 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2045 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2046 * up by cgroup_migrate_finish().
2048 * This function may be called without holding threadgroup_lock even if the
2049 * target is a process. Threads may be created and destroyed but as long
2050 * as cgroup_mutex is not dropped, no new css_set can be put into play and
2051 * the preloaded css_sets are guaranteed to cover all migrations.
2053 static void cgroup_migrate_add_src(struct css_set
*src_cset
,
2054 struct cgroup
*dst_cgrp
,
2055 struct list_head
*preloaded_csets
)
2057 struct cgroup
*src_cgrp
;
2059 lockdep_assert_held(&cgroup_mutex
);
2060 lockdep_assert_held(&css_set_rwsem
);
2062 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2064 if (!list_empty(&src_cset
->mg_preload_node
))
2067 WARN_ON(src_cset
->mg_src_cgrp
);
2068 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2069 WARN_ON(!list_empty(&src_cset
->mg_node
));
2071 src_cset
->mg_src_cgrp
= src_cgrp
;
2072 get_css_set(src_cset
);
2073 list_add(&src_cset
->mg_preload_node
, preloaded_csets
);
2077 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2078 * @dst_cgrp: the destination cgroup (may be %NULL)
2079 * @preloaded_csets: list of preloaded source css_sets
2081 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
2082 * have been preloaded to @preloaded_csets. This function looks up and
2083 * pins all destination css_sets, links each to its source, and append them
2084 * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each
2085 * source css_set is assumed to be its cgroup on the default hierarchy.
2087 * This function must be called after cgroup_migrate_add_src() has been
2088 * called on each migration source css_set. After migration is performed
2089 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2092 static int cgroup_migrate_prepare_dst(struct cgroup
*dst_cgrp
,
2093 struct list_head
*preloaded_csets
)
2096 struct css_set
*src_cset
, *tmp_cset
;
2098 lockdep_assert_held(&cgroup_mutex
);
2101 * Except for the root, child_subsys_mask must be zero for a cgroup
2102 * with tasks so that child cgroups don't compete against tasks.
2104 if (dst_cgrp
&& cgroup_on_dfl(dst_cgrp
) && cgroup_parent(dst_cgrp
) &&
2105 dst_cgrp
->child_subsys_mask
)
2108 /* look up the dst cset for each src cset and link it to src */
2109 list_for_each_entry_safe(src_cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2110 struct css_set
*dst_cset
;
2112 dst_cset
= find_css_set(src_cset
,
2113 dst_cgrp
?: src_cset
->dfl_cgrp
);
2117 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2120 * If src cset equals dst, it's noop. Drop the src.
2121 * cgroup_migrate() will skip the cset too. Note that we
2122 * can't handle src == dst as some nodes are used by both.
2124 if (src_cset
== dst_cset
) {
2125 src_cset
->mg_src_cgrp
= NULL
;
2126 list_del_init(&src_cset
->mg_preload_node
);
2127 put_css_set(src_cset
, false);
2128 put_css_set(dst_cset
, false);
2132 src_cset
->mg_dst_cset
= dst_cset
;
2134 if (list_empty(&dst_cset
->mg_preload_node
))
2135 list_add(&dst_cset
->mg_preload_node
, &csets
);
2137 put_css_set(dst_cset
, false);
2140 list_splice_tail(&csets
, preloaded_csets
);
2143 cgroup_migrate_finish(&csets
);
2148 * cgroup_migrate - migrate a process or task to a cgroup
2149 * @cgrp: the destination cgroup
2150 * @leader: the leader of the process or the task to migrate
2151 * @threadgroup: whether @leader points to the whole process or a single task
2153 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
2154 * process, the caller must be holding threadgroup_lock of @leader. The
2155 * caller is also responsible for invoking cgroup_migrate_add_src() and
2156 * cgroup_migrate_prepare_dst() on the targets before invoking this
2157 * function and following up with cgroup_migrate_finish().
2159 * As long as a controller's ->can_attach() doesn't fail, this function is
2160 * guaranteed to succeed. This means that, excluding ->can_attach()
2161 * failure, when migrating multiple targets, the success or failure can be
2162 * decided for all targets by invoking group_migrate_prepare_dst() before
2163 * actually starting migrating.
2165 static int cgroup_migrate(struct cgroup
*cgrp
, struct task_struct
*leader
,
2168 struct cgroup_taskset tset
= {
2169 .src_csets
= LIST_HEAD_INIT(tset
.src_csets
),
2170 .dst_csets
= LIST_HEAD_INIT(tset
.dst_csets
),
2171 .csets
= &tset
.src_csets
,
2173 struct cgroup_subsys_state
*css
, *failed_css
= NULL
;
2174 struct css_set
*cset
, *tmp_cset
;
2175 struct task_struct
*task
, *tmp_task
;
2179 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2180 * already PF_EXITING could be freed from underneath us unless we
2181 * take an rcu_read_lock.
2183 down_write(&css_set_rwsem
);
2187 /* @task either already exited or can't exit until the end */
2188 if (task
->flags
& PF_EXITING
)
2191 /* leave @task alone if post_fork() hasn't linked it yet */
2192 if (list_empty(&task
->cg_list
))
2195 cset
= task_css_set(task
);
2196 if (!cset
->mg_src_cgrp
)
2200 * cgroup_taskset_first() must always return the leader.
2201 * Take care to avoid disturbing the ordering.
2203 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2204 if (list_empty(&cset
->mg_node
))
2205 list_add_tail(&cset
->mg_node
, &tset
.src_csets
);
2206 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2207 list_move_tail(&cset
->mg_dst_cset
->mg_node
,
2212 } while_each_thread(leader
, task
);
2214 up_write(&css_set_rwsem
);
2216 /* methods shouldn't be called if no task is actually migrating */
2217 if (list_empty(&tset
.src_csets
))
2220 /* check that we can legitimately attach to the cgroup */
2221 for_each_e_css(css
, i
, cgrp
) {
2222 if (css
->ss
->can_attach
) {
2223 ret
= css
->ss
->can_attach(css
, &tset
);
2226 goto out_cancel_attach
;
2232 * Now that we're guaranteed success, proceed to move all tasks to
2233 * the new cgroup. There are no failure cases after here, so this
2234 * is the commit point.
2236 down_write(&css_set_rwsem
);
2237 list_for_each_entry(cset
, &tset
.src_csets
, mg_node
) {
2238 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
)
2239 cgroup_task_migrate(cset
->mg_src_cgrp
, task
,
2242 up_write(&css_set_rwsem
);
2245 * Migration is committed, all target tasks are now on dst_csets.
2246 * Nothing is sensitive to fork() after this point. Notify
2247 * controllers that migration is complete.
2249 tset
.csets
= &tset
.dst_csets
;
2251 for_each_e_css(css
, i
, cgrp
)
2252 if (css
->ss
->attach
)
2253 css
->ss
->attach(css
, &tset
);
2256 goto out_release_tset
;
2259 for_each_e_css(css
, i
, cgrp
) {
2260 if (css
== failed_css
)
2262 if (css
->ss
->cancel_attach
)
2263 css
->ss
->cancel_attach(css
, &tset
);
2266 down_write(&css_set_rwsem
);
2267 list_splice_init(&tset
.dst_csets
, &tset
.src_csets
);
2268 list_for_each_entry_safe(cset
, tmp_cset
, &tset
.src_csets
, mg_node
) {
2269 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2270 list_del_init(&cset
->mg_node
);
2272 up_write(&css_set_rwsem
);
2277 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2278 * @dst_cgrp: the cgroup to attach to
2279 * @leader: the task or the leader of the threadgroup to be attached
2280 * @threadgroup: attach the whole threadgroup?
2282 * Call holding cgroup_mutex and threadgroup_lock of @leader.
2284 static int cgroup_attach_task(struct cgroup
*dst_cgrp
,
2285 struct task_struct
*leader
, bool threadgroup
)
2287 LIST_HEAD(preloaded_csets
);
2288 struct task_struct
*task
;
2291 /* look up all src csets */
2292 down_read(&css_set_rwsem
);
2296 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
,
2300 } while_each_thread(leader
, task
);
2302 up_read(&css_set_rwsem
);
2304 /* prepare dst csets and commit */
2305 ret
= cgroup_migrate_prepare_dst(dst_cgrp
, &preloaded_csets
);
2307 ret
= cgroup_migrate(dst_cgrp
, leader
, threadgroup
);
2309 cgroup_migrate_finish(&preloaded_csets
);
2314 * Find the task_struct of the task to attach by vpid and pass it along to the
2315 * function to attach either it or all tasks in its threadgroup. Will lock
2316 * cgroup_mutex and threadgroup.
2318 static ssize_t
__cgroup_procs_write(struct kernfs_open_file
*of
, char *buf
,
2319 size_t nbytes
, loff_t off
, bool threadgroup
)
2321 struct task_struct
*tsk
;
2322 const struct cred
*cred
= current_cred(), *tcred
;
2323 struct cgroup
*cgrp
;
2327 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2330 cgrp
= cgroup_kn_lock_live(of
->kn
);
2337 tsk
= find_task_by_vpid(pid
);
2341 goto out_unlock_cgroup
;
2344 * even if we're attaching all tasks in the thread group, we
2345 * only need to check permissions on one of them.
2347 tcred
= __task_cred(tsk
);
2348 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2349 !uid_eq(cred
->euid
, tcred
->uid
) &&
2350 !uid_eq(cred
->euid
, tcred
->suid
)) {
2353 goto out_unlock_cgroup
;
2359 tsk
= tsk
->group_leader
;
2362 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2363 * trapped in a cpuset, or RT worker may be born in a cgroup
2364 * with no rt_runtime allocated. Just say no.
2366 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2369 goto out_unlock_cgroup
;
2372 get_task_struct(tsk
);
2375 threadgroup_lock(tsk
);
2377 if (!thread_group_leader(tsk
)) {
2379 * a race with de_thread from another thread's exec()
2380 * may strip us of our leadership, if this happens,
2381 * there is no choice but to throw this task away and
2382 * try again; this is
2383 * "double-double-toil-and-trouble-check locking".
2385 threadgroup_unlock(tsk
);
2386 put_task_struct(tsk
);
2387 goto retry_find_task
;
2391 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2393 threadgroup_unlock(tsk
);
2395 put_task_struct(tsk
);
2397 cgroup_kn_unlock(of
->kn
);
2398 return ret
?: nbytes
;
2402 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2403 * @from: attach to all cgroups of a given task
2404 * @tsk: the task to be attached
2406 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2408 struct cgroup_root
*root
;
2411 mutex_lock(&cgroup_mutex
);
2412 for_each_root(root
) {
2413 struct cgroup
*from_cgrp
;
2415 if (root
== &cgrp_dfl_root
)
2418 down_read(&css_set_rwsem
);
2419 from_cgrp
= task_cgroup_from_root(from
, root
);
2420 up_read(&css_set_rwsem
);
2422 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2426 mutex_unlock(&cgroup_mutex
);
2430 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2432 static ssize_t
cgroup_tasks_write(struct kernfs_open_file
*of
,
2433 char *buf
, size_t nbytes
, loff_t off
)
2435 return __cgroup_procs_write(of
, buf
, nbytes
, off
, false);
2438 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
2439 char *buf
, size_t nbytes
, loff_t off
)
2441 return __cgroup_procs_write(of
, buf
, nbytes
, off
, true);
2444 static ssize_t
cgroup_release_agent_write(struct kernfs_open_file
*of
,
2445 char *buf
, size_t nbytes
, loff_t off
)
2447 struct cgroup
*cgrp
;
2449 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2451 cgrp
= cgroup_kn_lock_live(of
->kn
);
2454 spin_lock(&release_agent_path_lock
);
2455 strlcpy(cgrp
->root
->release_agent_path
, strstrip(buf
),
2456 sizeof(cgrp
->root
->release_agent_path
));
2457 spin_unlock(&release_agent_path_lock
);
2458 cgroup_kn_unlock(of
->kn
);
2462 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
2464 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2466 spin_lock(&release_agent_path_lock
);
2467 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2468 spin_unlock(&release_agent_path_lock
);
2469 seq_putc(seq
, '\n');
2473 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
2475 seq_puts(seq
, "0\n");
2479 static void cgroup_print_ss_mask(struct seq_file
*seq
, unsigned int ss_mask
)
2481 struct cgroup_subsys
*ss
;
2482 bool printed
= false;
2485 for_each_subsys(ss
, ssid
) {
2486 if (ss_mask
& (1 << ssid
)) {
2489 seq_printf(seq
, "%s", ss
->name
);
2494 seq_putc(seq
, '\n');
2497 /* show controllers which are currently attached to the default hierarchy */
2498 static int cgroup_root_controllers_show(struct seq_file
*seq
, void *v
)
2500 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2502 cgroup_print_ss_mask(seq
, cgrp
->root
->subsys_mask
&
2503 ~cgrp_dfl_root_inhibit_ss_mask
);
2507 /* show controllers which are enabled from the parent */
2508 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2510 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2512 cgroup_print_ss_mask(seq
, cgroup_parent(cgrp
)->subtree_control
);
2516 /* show controllers which are enabled for a given cgroup's children */
2517 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2519 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2521 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2526 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2527 * @cgrp: root of the subtree to update csses for
2529 * @cgrp's child_subsys_mask has changed and its subtree's (self excluded)
2530 * css associations need to be updated accordingly. This function looks up
2531 * all css_sets which are attached to the subtree, creates the matching
2532 * updated css_sets and migrates the tasks to the new ones.
2534 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2536 LIST_HEAD(preloaded_csets
);
2537 struct cgroup_subsys_state
*css
;
2538 struct css_set
*src_cset
;
2541 lockdep_assert_held(&cgroup_mutex
);
2543 /* look up all csses currently attached to @cgrp's subtree */
2544 down_read(&css_set_rwsem
);
2545 css_for_each_descendant_pre(css
, cgroup_css(cgrp
, NULL
)) {
2546 struct cgrp_cset_link
*link
;
2548 /* self is not affected by child_subsys_mask change */
2549 if (css
->cgroup
== cgrp
)
2552 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
)
2553 cgroup_migrate_add_src(link
->cset
, cgrp
,
2556 up_read(&css_set_rwsem
);
2558 /* NULL dst indicates self on default hierarchy */
2559 ret
= cgroup_migrate_prepare_dst(NULL
, &preloaded_csets
);
2563 list_for_each_entry(src_cset
, &preloaded_csets
, mg_preload_node
) {
2564 struct task_struct
*last_task
= NULL
, *task
;
2566 /* src_csets precede dst_csets, break on the first dst_cset */
2567 if (!src_cset
->mg_src_cgrp
)
2571 * All tasks in src_cset need to be migrated to the
2572 * matching dst_cset. Empty it process by process. We
2573 * walk tasks but migrate processes. The leader might even
2574 * belong to a different cset but such src_cset would also
2575 * be among the target src_csets because the default
2576 * hierarchy enforces per-process membership.
2579 down_read(&css_set_rwsem
);
2580 task
= list_first_entry_or_null(&src_cset
->tasks
,
2581 struct task_struct
, cg_list
);
2583 task
= task
->group_leader
;
2584 WARN_ON_ONCE(!task_css_set(task
)->mg_src_cgrp
);
2585 get_task_struct(task
);
2587 up_read(&css_set_rwsem
);
2592 /* guard against possible infinite loop */
2593 if (WARN(last_task
== task
,
2594 "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n"))
2598 threadgroup_lock(task
);
2599 /* raced against de_thread() from another thread? */
2600 if (!thread_group_leader(task
)) {
2601 threadgroup_unlock(task
);
2602 put_task_struct(task
);
2606 ret
= cgroup_migrate(src_cset
->dfl_cgrp
, task
, true);
2608 threadgroup_unlock(task
);
2609 put_task_struct(task
);
2611 if (WARN(ret
, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret
))
2617 cgroup_migrate_finish(&preloaded_csets
);
2621 /* change the enabled child controllers for a cgroup in the default hierarchy */
2622 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
2623 char *buf
, size_t nbytes
,
2626 unsigned int enable
= 0, disable
= 0;
2627 unsigned int css_enable
, css_disable
, old_ctrl
, new_ctrl
;
2628 struct cgroup
*cgrp
, *child
;
2629 struct cgroup_subsys
*ss
;
2634 * Parse input - space separated list of subsystem names prefixed
2635 * with either + or -.
2637 buf
= strstrip(buf
);
2638 while ((tok
= strsep(&buf
, " "))) {
2641 for_each_subsys(ss
, ssid
) {
2642 if (ss
->disabled
|| strcmp(tok
+ 1, ss
->name
) ||
2643 ((1 << ss
->id
) & cgrp_dfl_root_inhibit_ss_mask
))
2647 enable
|= 1 << ssid
;
2648 disable
&= ~(1 << ssid
);
2649 } else if (*tok
== '-') {
2650 disable
|= 1 << ssid
;
2651 enable
&= ~(1 << ssid
);
2657 if (ssid
== CGROUP_SUBSYS_COUNT
)
2661 cgrp
= cgroup_kn_lock_live(of
->kn
);
2665 for_each_subsys(ss
, ssid
) {
2666 if (enable
& (1 << ssid
)) {
2667 if (cgrp
->subtree_control
& (1 << ssid
)) {
2668 enable
&= ~(1 << ssid
);
2672 /* unavailable or not enabled on the parent? */
2673 if (!(cgrp_dfl_root
.subsys_mask
& (1 << ssid
)) ||
2674 (cgroup_parent(cgrp
) &&
2675 !(cgroup_parent(cgrp
)->subtree_control
& (1 << ssid
)))) {
2681 * @ss is already enabled through dependency and
2682 * we'll just make it visible. Skip draining.
2684 if (cgrp
->child_subsys_mask
& (1 << ssid
))
2688 * Because css offlining is asynchronous, userland
2689 * might try to re-enable the same controller while
2690 * the previous instance is still around. In such
2691 * cases, wait till it's gone using offline_waitq.
2693 cgroup_for_each_live_child(child
, cgrp
) {
2696 if (!cgroup_css(child
, ss
))
2700 prepare_to_wait(&child
->offline_waitq
, &wait
,
2701 TASK_UNINTERRUPTIBLE
);
2702 cgroup_kn_unlock(of
->kn
);
2704 finish_wait(&child
->offline_waitq
, &wait
);
2707 return restart_syscall();
2709 } else if (disable
& (1 << ssid
)) {
2710 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
2711 disable
&= ~(1 << ssid
);
2715 /* a child has it enabled? */
2716 cgroup_for_each_live_child(child
, cgrp
) {
2717 if (child
->subtree_control
& (1 << ssid
)) {
2725 if (!enable
&& !disable
) {
2731 * Except for the root, subtree_control must be zero for a cgroup
2732 * with tasks so that child cgroups don't compete against tasks.
2734 if (enable
&& cgroup_parent(cgrp
) && !list_empty(&cgrp
->cset_links
)) {
2740 * Update subsys masks and calculate what needs to be done. More
2741 * subsystems than specified may need to be enabled or disabled
2742 * depending on subsystem dependencies.
2744 cgrp
->subtree_control
|= enable
;
2745 cgrp
->subtree_control
&= ~disable
;
2747 old_ctrl
= cgrp
->child_subsys_mask
;
2748 cgroup_refresh_child_subsys_mask(cgrp
);
2749 new_ctrl
= cgrp
->child_subsys_mask
;
2751 css_enable
= ~old_ctrl
& new_ctrl
;
2752 css_disable
= old_ctrl
& ~new_ctrl
;
2753 enable
|= css_enable
;
2754 disable
|= css_disable
;
2757 * Create new csses or make the existing ones visible. A css is
2758 * created invisible if it's being implicitly enabled through
2759 * dependency. An invisible css is made visible when the userland
2760 * explicitly enables it.
2762 for_each_subsys(ss
, ssid
) {
2763 if (!(enable
& (1 << ssid
)))
2766 cgroup_for_each_live_child(child
, cgrp
) {
2767 if (css_enable
& (1 << ssid
))
2768 ret
= create_css(child
, ss
,
2769 cgrp
->subtree_control
& (1 << ssid
));
2771 ret
= cgroup_populate_dir(child
, 1 << ssid
);
2778 * At this point, cgroup_e_css() results reflect the new csses
2779 * making the following cgroup_update_dfl_csses() properly update
2780 * css associations of all tasks in the subtree.
2782 ret
= cgroup_update_dfl_csses(cgrp
);
2787 * All tasks are migrated out of disabled csses. Kill or hide
2788 * them. A css is hidden when the userland requests it to be
2789 * disabled while other subsystems are still depending on it. The
2790 * css must not actively control resources and be in the vanilla
2791 * state if it's made visible again later. Controllers which may
2792 * be depended upon should provide ->css_reset() for this purpose.
2794 for_each_subsys(ss
, ssid
) {
2795 if (!(disable
& (1 << ssid
)))
2798 cgroup_for_each_live_child(child
, cgrp
) {
2799 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
2801 if (css_disable
& (1 << ssid
)) {
2804 cgroup_clear_dir(child
, 1 << ssid
);
2811 kernfs_activate(cgrp
->kn
);
2814 cgroup_kn_unlock(of
->kn
);
2815 return ret
?: nbytes
;
2818 cgrp
->subtree_control
&= ~enable
;
2819 cgrp
->subtree_control
|= disable
;
2820 cgroup_refresh_child_subsys_mask(cgrp
);
2822 for_each_subsys(ss
, ssid
) {
2823 if (!(enable
& (1 << ssid
)))
2826 cgroup_for_each_live_child(child
, cgrp
) {
2827 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
2832 if (css_enable
& (1 << ssid
))
2835 cgroup_clear_dir(child
, 1 << ssid
);
2841 static int cgroup_populated_show(struct seq_file
*seq
, void *v
)
2843 seq_printf(seq
, "%d\n", (bool)seq_css(seq
)->cgroup
->populated_cnt
);
2847 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
2848 size_t nbytes
, loff_t off
)
2850 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
2851 struct cftype
*cft
= of
->kn
->priv
;
2852 struct cgroup_subsys_state
*css
;
2856 return cft
->write(of
, buf
, nbytes
, off
);
2859 * kernfs guarantees that a file isn't deleted with operations in
2860 * flight, which means that the matching css is and stays alive and
2861 * doesn't need to be pinned. The RCU locking is not necessary
2862 * either. It's just for the convenience of using cgroup_css().
2865 css
= cgroup_css(cgrp
, cft
->ss
);
2868 if (cft
->write_u64
) {
2869 unsigned long long v
;
2870 ret
= kstrtoull(buf
, 0, &v
);
2872 ret
= cft
->write_u64(css
, cft
, v
);
2873 } else if (cft
->write_s64
) {
2875 ret
= kstrtoll(buf
, 0, &v
);
2877 ret
= cft
->write_s64(css
, cft
, v
);
2882 return ret
?: nbytes
;
2885 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
2887 return seq_cft(seq
)->seq_start(seq
, ppos
);
2890 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
2892 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
2895 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
2897 seq_cft(seq
)->seq_stop(seq
, v
);
2900 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2902 struct cftype
*cft
= seq_cft(m
);
2903 struct cgroup_subsys_state
*css
= seq_css(m
);
2906 return cft
->seq_show(m
, arg
);
2909 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
2910 else if (cft
->read_s64
)
2911 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
2917 static struct kernfs_ops cgroup_kf_single_ops
= {
2918 .atomic_write_len
= PAGE_SIZE
,
2919 .write
= cgroup_file_write
,
2920 .seq_show
= cgroup_seqfile_show
,
2923 static struct kernfs_ops cgroup_kf_ops
= {
2924 .atomic_write_len
= PAGE_SIZE
,
2925 .write
= cgroup_file_write
,
2926 .seq_start
= cgroup_seqfile_start
,
2927 .seq_next
= cgroup_seqfile_next
,
2928 .seq_stop
= cgroup_seqfile_stop
,
2929 .seq_show
= cgroup_seqfile_show
,
2933 * cgroup_rename - Only allow simple rename of directories in place.
2935 static int cgroup_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
2936 const char *new_name_str
)
2938 struct cgroup
*cgrp
= kn
->priv
;
2941 if (kernfs_type(kn
) != KERNFS_DIR
)
2943 if (kn
->parent
!= new_parent
)
2947 * This isn't a proper migration and its usefulness is very
2948 * limited. Disallow on the default hierarchy.
2950 if (cgroup_on_dfl(cgrp
))
2954 * We're gonna grab cgroup_mutex which nests outside kernfs
2955 * active_ref. kernfs_rename() doesn't require active_ref
2956 * protection. Break them before grabbing cgroup_mutex.
2958 kernfs_break_active_protection(new_parent
);
2959 kernfs_break_active_protection(kn
);
2961 mutex_lock(&cgroup_mutex
);
2963 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
2965 mutex_unlock(&cgroup_mutex
);
2967 kernfs_unbreak_active_protection(kn
);
2968 kernfs_unbreak_active_protection(new_parent
);
2972 /* set uid and gid of cgroup dirs and files to that of the creator */
2973 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
2975 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
2976 .ia_uid
= current_fsuid(),
2977 .ia_gid
= current_fsgid(), };
2979 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
2980 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
2983 return kernfs_setattr(kn
, &iattr
);
2986 static int cgroup_add_file(struct cgroup
*cgrp
, struct cftype
*cft
)
2988 char name
[CGROUP_FILE_NAME_MAX
];
2989 struct kernfs_node
*kn
;
2990 struct lock_class_key
*key
= NULL
;
2993 #ifdef CONFIG_DEBUG_LOCK_ALLOC
2994 key
= &cft
->lockdep_key
;
2996 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
2997 cgroup_file_mode(cft
), 0, cft
->kf_ops
, cft
,
3002 ret
= cgroup_kn_set_ugid(kn
);
3008 if (cft
->seq_show
== cgroup_populated_show
)
3009 cgrp
->populated_kn
= kn
;
3014 * cgroup_addrm_files - add or remove files to a cgroup directory
3015 * @cgrp: the target cgroup
3016 * @cfts: array of cftypes to be added
3017 * @is_add: whether to add or remove
3019 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3020 * For removals, this function never fails. If addition fails, this
3021 * function doesn't remove files already added. The caller is responsible
3024 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
3030 lockdep_assert_held(&cgroup_mutex
);
3032 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3033 /* does cft->flags tell us to skip this file on @cgrp? */
3034 if ((cft
->flags
& CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3036 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_on_dfl(cgrp
))
3038 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3040 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3044 ret
= cgroup_add_file(cgrp
, cft
);
3046 pr_warn("%s: failed to add %s, err=%d\n",
3047 __func__
, cft
->name
, ret
);
3051 cgroup_rm_file(cgrp
, cft
);
3057 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3060 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3061 struct cgroup
*root
= &ss
->root
->cgrp
;
3062 struct cgroup_subsys_state
*css
;
3065 lockdep_assert_held(&cgroup_mutex
);
3067 /* add/rm files for all cgroups created before */
3068 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3069 struct cgroup
*cgrp
= css
->cgroup
;
3071 if (cgroup_is_dead(cgrp
))
3074 ret
= cgroup_addrm_files(cgrp
, cfts
, is_add
);
3080 kernfs_activate(root
->kn
);
3084 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3088 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3089 /* free copy for custom atomic_write_len, see init_cftypes() */
3090 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3095 /* revert flags set by cgroup core while adding @cfts */
3096 cft
->flags
&= ~(CFTYPE_ONLY_ON_DFL
| CFTYPE_INSANE
);
3100 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3104 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3105 struct kernfs_ops
*kf_ops
;
3107 WARN_ON(cft
->ss
|| cft
->kf_ops
);
3110 kf_ops
= &cgroup_kf_ops
;
3112 kf_ops
= &cgroup_kf_single_ops
;
3115 * Ugh... if @cft wants a custom max_write_len, we need to
3116 * make a copy of kf_ops to set its atomic_write_len.
3118 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
3119 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
3121 cgroup_exit_cftypes(cfts
);
3124 kf_ops
->atomic_write_len
= cft
->max_write_len
;
3127 cft
->kf_ops
= kf_ops
;
3134 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
3136 lockdep_assert_held(&cgroup_mutex
);
3138 if (!cfts
|| !cfts
[0].ss
)
3141 list_del(&cfts
->node
);
3142 cgroup_apply_cftypes(cfts
, false);
3143 cgroup_exit_cftypes(cfts
);
3148 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3149 * @cfts: zero-length name terminated array of cftypes
3151 * Unregister @cfts. Files described by @cfts are removed from all
3152 * existing cgroups and all future cgroups won't have them either. This
3153 * function can be called anytime whether @cfts' subsys is attached or not.
3155 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3158 int cgroup_rm_cftypes(struct cftype
*cfts
)
3162 mutex_lock(&cgroup_mutex
);
3163 ret
= cgroup_rm_cftypes_locked(cfts
);
3164 mutex_unlock(&cgroup_mutex
);
3169 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3170 * @ss: target cgroup subsystem
3171 * @cfts: zero-length name terminated array of cftypes
3173 * Register @cfts to @ss. Files described by @cfts are created for all
3174 * existing cgroups to which @ss is attached and all future cgroups will
3175 * have them too. This function can be called anytime whether @ss is
3178 * Returns 0 on successful registration, -errno on failure. Note that this
3179 * function currently returns 0 as long as @cfts registration is successful
3180 * even if some file creation attempts on existing cgroups fail.
3182 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3189 if (!cfts
|| cfts
[0].name
[0] == '\0')
3192 ret
= cgroup_init_cftypes(ss
, cfts
);
3196 mutex_lock(&cgroup_mutex
);
3198 list_add_tail(&cfts
->node
, &ss
->cfts
);
3199 ret
= cgroup_apply_cftypes(cfts
, true);
3201 cgroup_rm_cftypes_locked(cfts
);
3203 mutex_unlock(&cgroup_mutex
);
3208 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3209 * @ss: target cgroup subsystem
3210 * @cfts: zero-length name terminated array of cftypes
3212 * Similar to cgroup_add_cftypes() but the added files are only used for
3213 * the default hierarchy.
3215 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3219 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3220 cft
->flags
|= CFTYPE_ONLY_ON_DFL
;
3221 return cgroup_add_cftypes(ss
, cfts
);
3225 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3226 * @ss: target cgroup subsystem
3227 * @cfts: zero-length name terminated array of cftypes
3229 * Similar to cgroup_add_cftypes() but the added files are only used for
3230 * the legacy hierarchies.
3232 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3236 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3237 cft
->flags
|= CFTYPE_INSANE
;
3238 return cgroup_add_cftypes(ss
, cfts
);
3242 * cgroup_task_count - count the number of tasks in a cgroup.
3243 * @cgrp: the cgroup in question
3245 * Return the number of tasks in the cgroup.
3247 static int cgroup_task_count(const struct cgroup
*cgrp
)
3250 struct cgrp_cset_link
*link
;
3252 down_read(&css_set_rwsem
);
3253 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
3254 count
+= atomic_read(&link
->cset
->refcount
);
3255 up_read(&css_set_rwsem
);
3260 * css_next_child - find the next child of a given css
3261 * @pos: the current position (%NULL to initiate traversal)
3262 * @parent: css whose children to walk
3264 * This function returns the next child of @parent and should be called
3265 * under either cgroup_mutex or RCU read lock. The only requirement is
3266 * that @parent and @pos are accessible. The next sibling is guaranteed to
3267 * be returned regardless of their states.
3269 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3270 * css which finished ->css_online() is guaranteed to be visible in the
3271 * future iterations and will stay visible until the last reference is put.
3272 * A css which hasn't finished ->css_online() or already finished
3273 * ->css_offline() may show up during traversal. It's each subsystem's
3274 * responsibility to synchronize against on/offlining.
3276 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
3277 struct cgroup_subsys_state
*parent
)
3279 struct cgroup_subsys_state
*next
;
3281 cgroup_assert_mutex_or_rcu_locked();
3284 * @pos could already have been unlinked from the sibling list.
3285 * Once a cgroup is removed, its ->sibling.next is no longer
3286 * updated when its next sibling changes. CSS_RELEASED is set when
3287 * @pos is taken off list, at which time its next pointer is valid,
3288 * and, as releases are serialized, the one pointed to by the next
3289 * pointer is guaranteed to not have started release yet. This
3290 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3291 * critical section, the one pointed to by its next pointer is
3292 * guaranteed to not have finished its RCU grace period even if we
3293 * have dropped rcu_read_lock() inbetween iterations.
3295 * If @pos has CSS_RELEASED set, its next pointer can't be
3296 * dereferenced; however, as each css is given a monotonically
3297 * increasing unique serial number and always appended to the
3298 * sibling list, the next one can be found by walking the parent's
3299 * children until the first css with higher serial number than
3300 * @pos's. While this path can be slower, it happens iff iteration
3301 * races against release and the race window is very small.
3304 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
3305 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
3306 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
3308 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
3309 if (next
->serial_nr
> pos
->serial_nr
)
3314 * @next, if not pointing to the head, can be dereferenced and is
3317 if (&next
->sibling
!= &parent
->children
)
3323 * css_next_descendant_pre - find the next descendant for pre-order walk
3324 * @pos: the current position (%NULL to initiate traversal)
3325 * @root: css whose descendants to walk
3327 * To be used by css_for_each_descendant_pre(). Find the next descendant
3328 * to visit for pre-order traversal of @root's descendants. @root is
3329 * included in the iteration and the first node to be visited.
3331 * While this function requires cgroup_mutex or RCU read locking, it
3332 * doesn't require the whole traversal to be contained in a single critical
3333 * section. This function will return the correct next descendant as long
3334 * as both @pos and @root are accessible and @pos is a descendant of @root.
3336 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3337 * css which finished ->css_online() is guaranteed to be visible in the
3338 * future iterations and will stay visible until the last reference is put.
3339 * A css which hasn't finished ->css_online() or already finished
3340 * ->css_offline() may show up during traversal. It's each subsystem's
3341 * responsibility to synchronize against on/offlining.
3343 struct cgroup_subsys_state
*
3344 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3345 struct cgroup_subsys_state
*root
)
3347 struct cgroup_subsys_state
*next
;
3349 cgroup_assert_mutex_or_rcu_locked();
3351 /* if first iteration, visit @root */
3355 /* visit the first child if exists */
3356 next
= css_next_child(NULL
, pos
);
3360 /* no child, visit my or the closest ancestor's next sibling */
3361 while (pos
!= root
) {
3362 next
= css_next_child(pos
, pos
->parent
);
3372 * css_rightmost_descendant - return the rightmost descendant of a css
3373 * @pos: css of interest
3375 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3376 * is returned. This can be used during pre-order traversal to skip
3379 * While this function requires cgroup_mutex or RCU read locking, it
3380 * doesn't require the whole traversal to be contained in a single critical
3381 * section. This function will return the correct rightmost descendant as
3382 * long as @pos is accessible.
3384 struct cgroup_subsys_state
*
3385 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3387 struct cgroup_subsys_state
*last
, *tmp
;
3389 cgroup_assert_mutex_or_rcu_locked();
3393 /* ->prev isn't RCU safe, walk ->next till the end */
3395 css_for_each_child(tmp
, last
)
3402 static struct cgroup_subsys_state
*
3403 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3405 struct cgroup_subsys_state
*last
;
3409 pos
= css_next_child(NULL
, pos
);
3416 * css_next_descendant_post - find the next descendant for post-order walk
3417 * @pos: the current position (%NULL to initiate traversal)
3418 * @root: css whose descendants to walk
3420 * To be used by css_for_each_descendant_post(). Find the next descendant
3421 * to visit for post-order traversal of @root's descendants. @root is
3422 * included in the iteration and the last node to be visited.
3424 * While this function requires cgroup_mutex or RCU read locking, it
3425 * doesn't require the whole traversal to be contained in a single critical
3426 * section. This function will return the correct next descendant as long
3427 * as both @pos and @cgroup are accessible and @pos is a descendant of
3430 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3431 * css which finished ->css_online() is guaranteed to be visible in the
3432 * future iterations and will stay visible until the last reference is put.
3433 * A css which hasn't finished ->css_online() or already finished
3434 * ->css_offline() may show up during traversal. It's each subsystem's
3435 * responsibility to synchronize against on/offlining.
3437 struct cgroup_subsys_state
*
3438 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3439 struct cgroup_subsys_state
*root
)
3441 struct cgroup_subsys_state
*next
;
3443 cgroup_assert_mutex_or_rcu_locked();
3445 /* if first iteration, visit leftmost descendant which may be @root */
3447 return css_leftmost_descendant(root
);
3449 /* if we visited @root, we're done */
3453 /* if there's an unvisited sibling, visit its leftmost descendant */
3454 next
= css_next_child(pos
, pos
->parent
);
3456 return css_leftmost_descendant(next
);
3458 /* no sibling left, visit parent */
3463 * css_has_online_children - does a css have online children
3464 * @css: the target css
3466 * Returns %true if @css has any online children; otherwise, %false. This
3467 * function can be called from any context but the caller is responsible
3468 * for synchronizing against on/offlining as necessary.
3470 bool css_has_online_children(struct cgroup_subsys_state
*css
)
3472 struct cgroup_subsys_state
*child
;
3476 css_for_each_child(child
, css
) {
3477 if (css
->flags
& CSS_ONLINE
) {
3487 * css_advance_task_iter - advance a task itererator to the next css_set
3488 * @it: the iterator to advance
3490 * Advance @it to the next css_set to walk.
3492 static void css_advance_task_iter(struct css_task_iter
*it
)
3494 struct list_head
*l
= it
->cset_pos
;
3495 struct cgrp_cset_link
*link
;
3496 struct css_set
*cset
;
3498 /* Advance to the next non-empty css_set */
3501 if (l
== it
->cset_head
) {
3502 it
->cset_pos
= NULL
;
3507 cset
= container_of(l
, struct css_set
,
3508 e_cset_node
[it
->ss
->id
]);
3510 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3513 } while (list_empty(&cset
->tasks
) && list_empty(&cset
->mg_tasks
));
3517 if (!list_empty(&cset
->tasks
))
3518 it
->task_pos
= cset
->tasks
.next
;
3520 it
->task_pos
= cset
->mg_tasks
.next
;
3522 it
->tasks_head
= &cset
->tasks
;
3523 it
->mg_tasks_head
= &cset
->mg_tasks
;
3527 * css_task_iter_start - initiate task iteration
3528 * @css: the css to walk tasks of
3529 * @it: the task iterator to use
3531 * Initiate iteration through the tasks of @css. The caller can call
3532 * css_task_iter_next() to walk through the tasks until the function
3533 * returns NULL. On completion of iteration, css_task_iter_end() must be
3536 * Note that this function acquires a lock which is released when the
3537 * iteration finishes. The caller can't sleep while iteration is in
3540 void css_task_iter_start(struct cgroup_subsys_state
*css
,
3541 struct css_task_iter
*it
)
3542 __acquires(css_set_rwsem
)
3544 /* no one should try to iterate before mounting cgroups */
3545 WARN_ON_ONCE(!use_task_css_set_links
);
3547 down_read(&css_set_rwsem
);
3552 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
3554 it
->cset_pos
= &css
->cgroup
->cset_links
;
3556 it
->cset_head
= it
->cset_pos
;
3558 css_advance_task_iter(it
);
3562 * css_task_iter_next - return the next task for the iterator
3563 * @it: the task iterator being iterated
3565 * The "next" function for task iteration. @it should have been
3566 * initialized via css_task_iter_start(). Returns NULL when the iteration
3569 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
3571 struct task_struct
*res
;
3572 struct list_head
*l
= it
->task_pos
;
3574 /* If the iterator cg is NULL, we have no tasks */
3577 res
= list_entry(l
, struct task_struct
, cg_list
);
3580 * Advance iterator to find next entry. cset->tasks is consumed
3581 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
3586 if (l
== it
->tasks_head
)
3587 l
= it
->mg_tasks_head
->next
;
3589 if (l
== it
->mg_tasks_head
)
3590 css_advance_task_iter(it
);
3598 * css_task_iter_end - finish task iteration
3599 * @it: the task iterator to finish
3601 * Finish task iteration started by css_task_iter_start().
3603 void css_task_iter_end(struct css_task_iter
*it
)
3604 __releases(css_set_rwsem
)
3606 up_read(&css_set_rwsem
);
3610 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3611 * @to: cgroup to which the tasks will be moved
3612 * @from: cgroup in which the tasks currently reside
3614 * Locking rules between cgroup_post_fork() and the migration path
3615 * guarantee that, if a task is forking while being migrated, the new child
3616 * is guaranteed to be either visible in the source cgroup after the
3617 * parent's migration is complete or put into the target cgroup. No task
3618 * can slip out of migration through forking.
3620 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3622 LIST_HEAD(preloaded_csets
);
3623 struct cgrp_cset_link
*link
;
3624 struct css_task_iter it
;
3625 struct task_struct
*task
;
3628 mutex_lock(&cgroup_mutex
);
3630 /* all tasks in @from are being moved, all csets are source */
3631 down_read(&css_set_rwsem
);
3632 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
3633 cgroup_migrate_add_src(link
->cset
, to
, &preloaded_csets
);
3634 up_read(&css_set_rwsem
);
3636 ret
= cgroup_migrate_prepare_dst(to
, &preloaded_csets
);
3641 * Migrate tasks one-by-one until @form is empty. This fails iff
3642 * ->can_attach() fails.
3645 css_task_iter_start(&from
->self
, &it
);
3646 task
= css_task_iter_next(&it
);
3648 get_task_struct(task
);
3649 css_task_iter_end(&it
);
3652 ret
= cgroup_migrate(to
, task
, false);
3653 put_task_struct(task
);
3655 } while (task
&& !ret
);
3657 cgroup_migrate_finish(&preloaded_csets
);
3658 mutex_unlock(&cgroup_mutex
);
3663 * Stuff for reading the 'tasks'/'procs' files.
3665 * Reading this file can return large amounts of data if a cgroup has
3666 * *lots* of attached tasks. So it may need several calls to read(),
3667 * but we cannot guarantee that the information we produce is correct
3668 * unless we produce it entirely atomically.
3672 /* which pidlist file are we talking about? */
3673 enum cgroup_filetype
{
3679 * A pidlist is a list of pids that virtually represents the contents of one
3680 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3681 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3684 struct cgroup_pidlist
{
3686 * used to find which pidlist is wanted. doesn't change as long as
3687 * this particular list stays in the list.
3689 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3692 /* how many elements the above list has */
3694 /* each of these stored in a list by its cgroup */
3695 struct list_head links
;
3696 /* pointer to the cgroup we belong to, for list removal purposes */
3697 struct cgroup
*owner
;
3698 /* for delayed destruction */
3699 struct delayed_work destroy_dwork
;
3703 * The following two functions "fix" the issue where there are more pids
3704 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3705 * TODO: replace with a kernel-wide solution to this problem
3707 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3708 static void *pidlist_allocate(int count
)
3710 if (PIDLIST_TOO_LARGE(count
))
3711 return vmalloc(count
* sizeof(pid_t
));
3713 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3716 static void pidlist_free(void *p
)
3718 if (is_vmalloc_addr(p
))
3725 * Used to destroy all pidlists lingering waiting for destroy timer. None
3726 * should be left afterwards.
3728 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
)
3730 struct cgroup_pidlist
*l
, *tmp_l
;
3732 mutex_lock(&cgrp
->pidlist_mutex
);
3733 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
3734 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
3735 mutex_unlock(&cgrp
->pidlist_mutex
);
3737 flush_workqueue(cgroup_pidlist_destroy_wq
);
3738 BUG_ON(!list_empty(&cgrp
->pidlists
));
3741 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
3743 struct delayed_work
*dwork
= to_delayed_work(work
);
3744 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
3746 struct cgroup_pidlist
*tofree
= NULL
;
3748 mutex_lock(&l
->owner
->pidlist_mutex
);
3751 * Destroy iff we didn't get queued again. The state won't change
3752 * as destroy_dwork can only be queued while locked.
3754 if (!delayed_work_pending(dwork
)) {
3755 list_del(&l
->links
);
3756 pidlist_free(l
->list
);
3757 put_pid_ns(l
->key
.ns
);
3761 mutex_unlock(&l
->owner
->pidlist_mutex
);
3766 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3767 * Returns the number of unique elements.
3769 static int pidlist_uniq(pid_t
*list
, int length
)
3774 * we presume the 0th element is unique, so i starts at 1. trivial
3775 * edge cases first; no work needs to be done for either
3777 if (length
== 0 || length
== 1)
3779 /* src and dest walk down the list; dest counts unique elements */
3780 for (src
= 1; src
< length
; src
++) {
3781 /* find next unique element */
3782 while (list
[src
] == list
[src
-1]) {
3787 /* dest always points to where the next unique element goes */
3788 list
[dest
] = list
[src
];
3796 * The two pid files - task and cgroup.procs - guaranteed that the result
3797 * is sorted, which forced this whole pidlist fiasco. As pid order is
3798 * different per namespace, each namespace needs differently sorted list,
3799 * making it impossible to use, for example, single rbtree of member tasks
3800 * sorted by task pointer. As pidlists can be fairly large, allocating one
3801 * per open file is dangerous, so cgroup had to implement shared pool of
3802 * pidlists keyed by cgroup and namespace.
3804 * All this extra complexity was caused by the original implementation
3805 * committing to an entirely unnecessary property. In the long term, we
3806 * want to do away with it. Explicitly scramble sort order if on the
3807 * default hierarchy so that no such expectation exists in the new
3810 * Scrambling is done by swapping every two consecutive bits, which is
3811 * non-identity one-to-one mapping which disturbs sort order sufficiently.
3813 static pid_t
pid_fry(pid_t pid
)
3815 unsigned a
= pid
& 0x55555555;
3816 unsigned b
= pid
& 0xAAAAAAAA;
3818 return (a
<< 1) | (b
>> 1);
3821 static pid_t
cgroup_pid_fry(struct cgroup
*cgrp
, pid_t pid
)
3823 if (cgroup_on_dfl(cgrp
))
3824 return pid_fry(pid
);
3829 static int cmppid(const void *a
, const void *b
)
3831 return *(pid_t
*)a
- *(pid_t
*)b
;
3834 static int fried_cmppid(const void *a
, const void *b
)
3836 return pid_fry(*(pid_t
*)a
) - pid_fry(*(pid_t
*)b
);
3839 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3840 enum cgroup_filetype type
)
3842 struct cgroup_pidlist
*l
;
3843 /* don't need task_nsproxy() if we're looking at ourself */
3844 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3846 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3848 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
3849 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
3855 * find the appropriate pidlist for our purpose (given procs vs tasks)
3856 * returns with the lock on that pidlist already held, and takes care
3857 * of the use count, or returns NULL with no locks held if we're out of
3860 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
3861 enum cgroup_filetype type
)
3863 struct cgroup_pidlist
*l
;
3865 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3867 l
= cgroup_pidlist_find(cgrp
, type
);
3871 /* entry not found; create a new one */
3872 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3876 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
3878 /* don't need task_nsproxy() if we're looking at ourself */
3879 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
3881 list_add(&l
->links
, &cgrp
->pidlists
);
3886 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3888 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3889 struct cgroup_pidlist
**lp
)
3893 int pid
, n
= 0; /* used for populating the array */
3894 struct css_task_iter it
;
3895 struct task_struct
*tsk
;
3896 struct cgroup_pidlist
*l
;
3898 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3901 * If cgroup gets more users after we read count, we won't have
3902 * enough space - tough. This race is indistinguishable to the
3903 * caller from the case that the additional cgroup users didn't
3904 * show up until sometime later on.
3906 length
= cgroup_task_count(cgrp
);
3907 array
= pidlist_allocate(length
);
3910 /* now, populate the array */
3911 css_task_iter_start(&cgrp
->self
, &it
);
3912 while ((tsk
= css_task_iter_next(&it
))) {
3913 if (unlikely(n
== length
))
3915 /* get tgid or pid for procs or tasks file respectively */
3916 if (type
== CGROUP_FILE_PROCS
)
3917 pid
= task_tgid_vnr(tsk
);
3919 pid
= task_pid_vnr(tsk
);
3920 if (pid
> 0) /* make sure to only use valid results */
3923 css_task_iter_end(&it
);
3925 /* now sort & (if procs) strip out duplicates */
3926 if (cgroup_on_dfl(cgrp
))
3927 sort(array
, length
, sizeof(pid_t
), fried_cmppid
, NULL
);
3929 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3930 if (type
== CGROUP_FILE_PROCS
)
3931 length
= pidlist_uniq(array
, length
);
3933 l
= cgroup_pidlist_find_create(cgrp
, type
);
3935 mutex_unlock(&cgrp
->pidlist_mutex
);
3936 pidlist_free(array
);
3940 /* store array, freeing old if necessary */
3941 pidlist_free(l
->list
);
3949 * cgroupstats_build - build and fill cgroupstats
3950 * @stats: cgroupstats to fill information into
3951 * @dentry: A dentry entry belonging to the cgroup for which stats have
3954 * Build and fill cgroupstats so that taskstats can export it to user
3957 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3959 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
3960 struct cgroup
*cgrp
;
3961 struct css_task_iter it
;
3962 struct task_struct
*tsk
;
3964 /* it should be kernfs_node belonging to cgroupfs and is a directory */
3965 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
3966 kernfs_type(kn
) != KERNFS_DIR
)
3969 mutex_lock(&cgroup_mutex
);
3972 * We aren't being called from kernfs and there's no guarantee on
3973 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
3974 * @kn->priv is RCU safe. Let's do the RCU dancing.
3977 cgrp
= rcu_dereference(kn
->priv
);
3978 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
3980 mutex_unlock(&cgroup_mutex
);
3985 css_task_iter_start(&cgrp
->self
, &it
);
3986 while ((tsk
= css_task_iter_next(&it
))) {
3987 switch (tsk
->state
) {
3989 stats
->nr_running
++;
3991 case TASK_INTERRUPTIBLE
:
3992 stats
->nr_sleeping
++;
3994 case TASK_UNINTERRUPTIBLE
:
3995 stats
->nr_uninterruptible
++;
3998 stats
->nr_stopped
++;
4001 if (delayacct_is_task_waiting_on_io(tsk
))
4002 stats
->nr_io_wait
++;
4006 css_task_iter_end(&it
);
4008 mutex_unlock(&cgroup_mutex
);
4014 * seq_file methods for the tasks/procs files. The seq_file position is the
4015 * next pid to display; the seq_file iterator is a pointer to the pid
4016 * in the cgroup->l->list array.
4019 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
4022 * Initially we receive a position value that corresponds to
4023 * one more than the last pid shown (or 0 on the first call or
4024 * after a seek to the start). Use a binary-search to find the
4025 * next pid to display, if any
4027 struct kernfs_open_file
*of
= s
->private;
4028 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4029 struct cgroup_pidlist
*l
;
4030 enum cgroup_filetype type
= seq_cft(s
)->private;
4031 int index
= 0, pid
= *pos
;
4034 mutex_lock(&cgrp
->pidlist_mutex
);
4037 * !NULL @of->priv indicates that this isn't the first start()
4038 * after open. If the matching pidlist is around, we can use that.
4039 * Look for it. Note that @of->priv can't be used directly. It
4040 * could already have been destroyed.
4043 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
4046 * Either this is the first start() after open or the matching
4047 * pidlist has been destroyed inbetween. Create a new one.
4050 ret
= pidlist_array_load(cgrp
, type
,
4051 (struct cgroup_pidlist
**)&of
->priv
);
4053 return ERR_PTR(ret
);
4058 int end
= l
->length
;
4060 while (index
< end
) {
4061 int mid
= (index
+ end
) / 2;
4062 if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) == pid
) {
4065 } else if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) <= pid
)
4071 /* If we're off the end of the array, we're done */
4072 if (index
>= l
->length
)
4074 /* Update the abstract position to be the actual pid that we found */
4075 iter
= l
->list
+ index
;
4076 *pos
= cgroup_pid_fry(cgrp
, *iter
);
4080 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
4082 struct kernfs_open_file
*of
= s
->private;
4083 struct cgroup_pidlist
*l
= of
->priv
;
4086 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
4087 CGROUP_PIDLIST_DESTROY_DELAY
);
4088 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
4091 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4093 struct kernfs_open_file
*of
= s
->private;
4094 struct cgroup_pidlist
*l
= of
->priv
;
4096 pid_t
*end
= l
->list
+ l
->length
;
4098 * Advance to the next pid in the array. If this goes off the
4105 *pos
= cgroup_pid_fry(seq_css(s
)->cgroup
, *p
);
4110 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
4112 return seq_printf(s
, "%d\n", *(int *)v
);
4115 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
4118 return notify_on_release(css
->cgroup
);
4121 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
4122 struct cftype
*cft
, u64 val
)
4124 clear_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
4126 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4128 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4132 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
4135 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4138 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
4139 struct cftype
*cft
, u64 val
)
4142 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4144 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4148 /* cgroup core interface files for the default hierarchy */
4149 static struct cftype cgroup_dfl_base_files
[] = {
4151 .name
= "cgroup.procs",
4152 .seq_start
= cgroup_pidlist_start
,
4153 .seq_next
= cgroup_pidlist_next
,
4154 .seq_stop
= cgroup_pidlist_stop
,
4155 .seq_show
= cgroup_pidlist_show
,
4156 .private = CGROUP_FILE_PROCS
,
4157 .write
= cgroup_procs_write
,
4158 .mode
= S_IRUGO
| S_IWUSR
,
4161 .name
= "cgroup.controllers",
4162 .flags
= CFTYPE_ONLY_ON_ROOT
,
4163 .seq_show
= cgroup_root_controllers_show
,
4166 .name
= "cgroup.controllers",
4167 .flags
= CFTYPE_NOT_ON_ROOT
,
4168 .seq_show
= cgroup_controllers_show
,
4171 .name
= "cgroup.subtree_control",
4172 .seq_show
= cgroup_subtree_control_show
,
4173 .write
= cgroup_subtree_control_write
,
4176 .name
= "cgroup.populated",
4177 .flags
= CFTYPE_NOT_ON_ROOT
,
4178 .seq_show
= cgroup_populated_show
,
4183 /* cgroup core interface files for the legacy hierarchies */
4184 static struct cftype cgroup_legacy_base_files
[] = {
4186 .name
= "cgroup.procs",
4187 .seq_start
= cgroup_pidlist_start
,
4188 .seq_next
= cgroup_pidlist_next
,
4189 .seq_stop
= cgroup_pidlist_stop
,
4190 .seq_show
= cgroup_pidlist_show
,
4191 .private = CGROUP_FILE_PROCS
,
4192 .write
= cgroup_procs_write
,
4193 .mode
= S_IRUGO
| S_IWUSR
,
4196 .name
= "cgroup.clone_children",
4197 .read_u64
= cgroup_clone_children_read
,
4198 .write_u64
= cgroup_clone_children_write
,
4201 .name
= "cgroup.sane_behavior",
4202 .flags
= CFTYPE_ONLY_ON_ROOT
,
4203 .seq_show
= cgroup_sane_behavior_show
,
4207 .seq_start
= cgroup_pidlist_start
,
4208 .seq_next
= cgroup_pidlist_next
,
4209 .seq_stop
= cgroup_pidlist_stop
,
4210 .seq_show
= cgroup_pidlist_show
,
4211 .private = CGROUP_FILE_TASKS
,
4212 .write
= cgroup_tasks_write
,
4213 .mode
= S_IRUGO
| S_IWUSR
,
4216 .name
= "notify_on_release",
4217 .read_u64
= cgroup_read_notify_on_release
,
4218 .write_u64
= cgroup_write_notify_on_release
,
4221 .name
= "release_agent",
4222 .flags
= CFTYPE_ONLY_ON_ROOT
,
4223 .seq_show
= cgroup_release_agent_show
,
4224 .write
= cgroup_release_agent_write
,
4225 .max_write_len
= PATH_MAX
- 1,
4231 * cgroup_populate_dir - create subsys files in a cgroup directory
4232 * @cgrp: target cgroup
4233 * @subsys_mask: mask of the subsystem ids whose files should be added
4235 * On failure, no file is added.
4237 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned int subsys_mask
)
4239 struct cgroup_subsys
*ss
;
4242 /* process cftsets of each subsystem */
4243 for_each_subsys(ss
, i
) {
4244 struct cftype
*cfts
;
4246 if (!(subsys_mask
& (1 << i
)))
4249 list_for_each_entry(cfts
, &ss
->cfts
, node
) {
4250 ret
= cgroup_addrm_files(cgrp
, cfts
, true);
4257 cgroup_clear_dir(cgrp
, subsys_mask
);
4262 * css destruction is four-stage process.
4264 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4265 * Implemented in kill_css().
4267 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4268 * and thus css_tryget_online() is guaranteed to fail, the css can be
4269 * offlined by invoking offline_css(). After offlining, the base ref is
4270 * put. Implemented in css_killed_work_fn().
4272 * 3. When the percpu_ref reaches zero, the only possible remaining
4273 * accessors are inside RCU read sections. css_release() schedules the
4276 * 4. After the grace period, the css can be freed. Implemented in
4277 * css_free_work_fn().
4279 * It is actually hairier because both step 2 and 4 require process context
4280 * and thus involve punting to css->destroy_work adding two additional
4281 * steps to the already complex sequence.
4283 static void css_free_work_fn(struct work_struct
*work
)
4285 struct cgroup_subsys_state
*css
=
4286 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4287 struct cgroup
*cgrp
= css
->cgroup
;
4292 css_put(css
->parent
);
4294 css
->ss
->css_free(css
);
4297 /* cgroup free path */
4298 atomic_dec(&cgrp
->root
->nr_cgrps
);
4299 cgroup_pidlist_destroy_all(cgrp
);
4301 if (cgroup_parent(cgrp
)) {
4303 * We get a ref to the parent, and put the ref when
4304 * this cgroup is being freed, so it's guaranteed
4305 * that the parent won't be destroyed before its
4308 cgroup_put(cgroup_parent(cgrp
));
4309 kernfs_put(cgrp
->kn
);
4313 * This is root cgroup's refcnt reaching zero,
4314 * which indicates that the root should be
4317 cgroup_destroy_root(cgrp
->root
);
4322 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
4324 struct cgroup_subsys_state
*css
=
4325 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
4327 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
4328 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4331 static void css_release_work_fn(struct work_struct
*work
)
4333 struct cgroup_subsys_state
*css
=
4334 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4335 struct cgroup_subsys
*ss
= css
->ss
;
4336 struct cgroup
*cgrp
= css
->cgroup
;
4338 mutex_lock(&cgroup_mutex
);
4340 css
->flags
|= CSS_RELEASED
;
4341 list_del_rcu(&css
->sibling
);
4344 /* css release path */
4345 cgroup_idr_remove(&ss
->css_idr
, css
->id
);
4347 /* cgroup release path */
4348 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4352 mutex_unlock(&cgroup_mutex
);
4354 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4357 static void css_release(struct percpu_ref
*ref
)
4359 struct cgroup_subsys_state
*css
=
4360 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4362 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
4363 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4366 static void init_and_link_css(struct cgroup_subsys_state
*css
,
4367 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4369 lockdep_assert_held(&cgroup_mutex
);
4373 memset(css
, 0, sizeof(*css
));
4376 INIT_LIST_HEAD(&css
->sibling
);
4377 INIT_LIST_HEAD(&css
->children
);
4378 css
->serial_nr
= css_serial_nr_next
++;
4380 if (cgroup_parent(cgrp
)) {
4381 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
4382 css_get(css
->parent
);
4385 BUG_ON(cgroup_css(cgrp
, ss
));
4388 /* invoke ->css_online() on a new CSS and mark it online if successful */
4389 static int online_css(struct cgroup_subsys_state
*css
)
4391 struct cgroup_subsys
*ss
= css
->ss
;
4394 lockdep_assert_held(&cgroup_mutex
);
4397 ret
= ss
->css_online(css
);
4399 css
->flags
|= CSS_ONLINE
;
4400 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
4405 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4406 static void offline_css(struct cgroup_subsys_state
*css
)
4408 struct cgroup_subsys
*ss
= css
->ss
;
4410 lockdep_assert_held(&cgroup_mutex
);
4412 if (!(css
->flags
& CSS_ONLINE
))
4415 if (ss
->css_offline
)
4416 ss
->css_offline(css
);
4418 css
->flags
&= ~CSS_ONLINE
;
4419 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
4421 wake_up_all(&css
->cgroup
->offline_waitq
);
4425 * create_css - create a cgroup_subsys_state
4426 * @cgrp: the cgroup new css will be associated with
4427 * @ss: the subsys of new css
4428 * @visible: whether to create control knobs for the new css or not
4430 * Create a new css associated with @cgrp - @ss pair. On success, the new
4431 * css is online and installed in @cgrp with all interface files created if
4432 * @visible. Returns 0 on success, -errno on failure.
4434 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
4437 struct cgroup
*parent
= cgroup_parent(cgrp
);
4438 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
4439 struct cgroup_subsys_state
*css
;
4442 lockdep_assert_held(&cgroup_mutex
);
4444 css
= ss
->css_alloc(parent_css
);
4446 return PTR_ERR(css
);
4448 init_and_link_css(css
, ss
, cgrp
);
4450 err
= percpu_ref_init(&css
->refcnt
, css_release
);
4454 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_NOWAIT
);
4456 goto err_free_percpu_ref
;
4460 err
= cgroup_populate_dir(cgrp
, 1 << ss
->id
);
4465 /* @css is ready to be brought online now, make it visible */
4466 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
4467 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
4469 err
= online_css(css
);
4473 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4474 cgroup_parent(parent
)) {
4475 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4476 current
->comm
, current
->pid
, ss
->name
);
4477 if (!strcmp(ss
->name
, "memory"))
4478 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4479 ss
->warned_broken_hierarchy
= true;
4485 list_del_rcu(&css
->sibling
);
4486 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->id
);
4488 cgroup_idr_remove(&ss
->css_idr
, css
->id
);
4489 err_free_percpu_ref
:
4490 percpu_ref_cancel_init(&css
->refcnt
);
4492 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4496 static int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
,
4499 struct cgroup
*parent
, *cgrp
;
4500 struct cgroup_root
*root
;
4501 struct cgroup_subsys
*ss
;
4502 struct kernfs_node
*kn
;
4503 struct cftype
*base_files
;
4506 parent
= cgroup_kn_lock_live(parent_kn
);
4509 root
= parent
->root
;
4511 /* allocate the cgroup and its ID, 0 is reserved for the root */
4512 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4518 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
);
4523 * Temporarily set the pointer to NULL, so idr_find() won't return
4524 * a half-baked cgroup.
4526 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_NOWAIT
);
4529 goto out_cancel_ref
;
4532 init_cgroup_housekeeping(cgrp
);
4534 cgrp
->self
.parent
= &parent
->self
;
4537 if (notify_on_release(parent
))
4538 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4540 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4541 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4543 /* create the directory */
4544 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
4552 * This extra ref will be put in cgroup_free_fn() and guarantees
4553 * that @cgrp->kn is always accessible.
4557 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
4559 /* allocation complete, commit to creation */
4560 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
4561 atomic_inc(&root
->nr_cgrps
);
4565 * @cgrp is now fully operational. If something fails after this
4566 * point, it'll be released via the normal destruction path.
4568 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4570 ret
= cgroup_kn_set_ugid(kn
);
4574 if (cgroup_on_dfl(cgrp
))
4575 base_files
= cgroup_dfl_base_files
;
4577 base_files
= cgroup_legacy_base_files
;
4579 ret
= cgroup_addrm_files(cgrp
, base_files
, true);
4583 /* let's create and online css's */
4584 for_each_subsys(ss
, ssid
) {
4585 if (parent
->child_subsys_mask
& (1 << ssid
)) {
4586 ret
= create_css(cgrp
, ss
,
4587 parent
->subtree_control
& (1 << ssid
));
4594 * On the default hierarchy, a child doesn't automatically inherit
4595 * subtree_control from the parent. Each is configured manually.
4597 if (!cgroup_on_dfl(cgrp
)) {
4598 cgrp
->subtree_control
= parent
->subtree_control
;
4599 cgroup_refresh_child_subsys_mask(cgrp
);
4602 kernfs_activate(kn
);
4608 cgroup_idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4610 percpu_ref_cancel_init(&cgrp
->self
.refcnt
);
4614 cgroup_kn_unlock(parent_kn
);
4618 cgroup_destroy_locked(cgrp
);
4623 * This is called when the refcnt of a css is confirmed to be killed.
4624 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4625 * initate destruction and put the css ref from kill_css().
4627 static void css_killed_work_fn(struct work_struct
*work
)
4629 struct cgroup_subsys_state
*css
=
4630 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4632 mutex_lock(&cgroup_mutex
);
4634 mutex_unlock(&cgroup_mutex
);
4639 /* css kill confirmation processing requires process context, bounce */
4640 static void css_killed_ref_fn(struct percpu_ref
*ref
)
4642 struct cgroup_subsys_state
*css
=
4643 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4645 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
4646 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4650 * kill_css - destroy a css
4651 * @css: css to destroy
4653 * This function initiates destruction of @css by removing cgroup interface
4654 * files and putting its base reference. ->css_offline() will be invoked
4655 * asynchronously once css_tryget_online() is guaranteed to fail and when
4656 * the reference count reaches zero, @css will be released.
4658 static void kill_css(struct cgroup_subsys_state
*css
)
4660 lockdep_assert_held(&cgroup_mutex
);
4663 * This must happen before css is disassociated with its cgroup.
4664 * See seq_css() for details.
4666 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->id
);
4669 * Killing would put the base ref, but we need to keep it alive
4670 * until after ->css_offline().
4675 * cgroup core guarantees that, by the time ->css_offline() is
4676 * invoked, no new css reference will be given out via
4677 * css_tryget_online(). We can't simply call percpu_ref_kill() and
4678 * proceed to offlining css's because percpu_ref_kill() doesn't
4679 * guarantee that the ref is seen as killed on all CPUs on return.
4681 * Use percpu_ref_kill_and_confirm() to get notifications as each
4682 * css is confirmed to be seen as killed on all CPUs.
4684 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
4688 * cgroup_destroy_locked - the first stage of cgroup destruction
4689 * @cgrp: cgroup to be destroyed
4691 * css's make use of percpu refcnts whose killing latency shouldn't be
4692 * exposed to userland and are RCU protected. Also, cgroup core needs to
4693 * guarantee that css_tryget_online() won't succeed by the time
4694 * ->css_offline() is invoked. To satisfy all the requirements,
4695 * destruction is implemented in the following two steps.
4697 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4698 * userland visible parts and start killing the percpu refcnts of
4699 * css's. Set up so that the next stage will be kicked off once all
4700 * the percpu refcnts are confirmed to be killed.
4702 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4703 * rest of destruction. Once all cgroup references are gone, the
4704 * cgroup is RCU-freed.
4706 * This function implements s1. After this step, @cgrp is gone as far as
4707 * the userland is concerned and a new cgroup with the same name may be
4708 * created. As cgroup doesn't care about the names internally, this
4709 * doesn't cause any problem.
4711 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4712 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4714 struct cgroup_subsys_state
*css
;
4718 lockdep_assert_held(&cgroup_mutex
);
4721 * css_set_rwsem synchronizes access to ->cset_links and prevents
4722 * @cgrp from being removed while put_css_set() is in progress.
4724 down_read(&css_set_rwsem
);
4725 empty
= list_empty(&cgrp
->cset_links
);
4726 up_read(&css_set_rwsem
);
4731 * Make sure there's no live children. We can't test emptiness of
4732 * ->self.children as dead children linger on it while being
4733 * drained; otherwise, "rmdir parent/child parent" may fail.
4735 if (css_has_online_children(&cgrp
->self
))
4739 * Mark @cgrp dead. This prevents further task migration and child
4740 * creation by disabling cgroup_lock_live_group().
4742 cgrp
->self
.flags
&= ~CSS_ONLINE
;
4744 /* initiate massacre of all css's */
4745 for_each_css(css
, ssid
, cgrp
)
4748 /* CSS_ONLINE is clear, remove from ->release_list for the last time */
4749 raw_spin_lock(&release_list_lock
);
4750 if (!list_empty(&cgrp
->release_list
))
4751 list_del_init(&cgrp
->release_list
);
4752 raw_spin_unlock(&release_list_lock
);
4755 * Remove @cgrp directory along with the base files. @cgrp has an
4756 * extra ref on its kn.
4758 kernfs_remove(cgrp
->kn
);
4760 set_bit(CGRP_RELEASABLE
, &cgroup_parent(cgrp
)->flags
);
4761 check_for_release(cgroup_parent(cgrp
));
4763 /* put the base reference */
4764 percpu_ref_kill(&cgrp
->self
.refcnt
);
4769 static int cgroup_rmdir(struct kernfs_node
*kn
)
4771 struct cgroup
*cgrp
;
4774 cgrp
= cgroup_kn_lock_live(kn
);
4777 cgroup_get(cgrp
); /* for @kn->priv clearing */
4779 ret
= cgroup_destroy_locked(cgrp
);
4781 cgroup_kn_unlock(kn
);
4784 * There are two control paths which try to determine cgroup from
4785 * dentry without going through kernfs - cgroupstats_build() and
4786 * css_tryget_online_from_dir(). Those are supported by RCU
4787 * protecting clearing of cgrp->kn->priv backpointer, which should
4788 * happen after all files under it have been removed.
4791 RCU_INIT_POINTER(*(void __rcu __force
**)&kn
->priv
, NULL
);
4797 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
4798 .remount_fs
= cgroup_remount
,
4799 .show_options
= cgroup_show_options
,
4800 .mkdir
= cgroup_mkdir
,
4801 .rmdir
= cgroup_rmdir
,
4802 .rename
= cgroup_rename
,
4805 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
4807 struct cgroup_subsys_state
*css
;
4809 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4811 mutex_lock(&cgroup_mutex
);
4813 idr_init(&ss
->css_idr
);
4814 INIT_LIST_HEAD(&ss
->cfts
);
4816 /* Create the root cgroup state for this subsystem */
4817 ss
->root
= &cgrp_dfl_root
;
4818 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
4819 /* We don't handle early failures gracefully */
4820 BUG_ON(IS_ERR(css
));
4821 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
4824 * Root csses are never destroyed and we can't initialize
4825 * percpu_ref during early init. Disable refcnting.
4827 css
->flags
|= CSS_NO_REF
;
4830 /* allocation can't be done safely during early init */
4833 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
4834 BUG_ON(css
->id
< 0);
4837 /* Update the init_css_set to contain a subsys
4838 * pointer to this state - since the subsystem is
4839 * newly registered, all tasks and hence the
4840 * init_css_set is in the subsystem's root cgroup. */
4841 init_css_set
.subsys
[ss
->id
] = css
;
4843 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4845 /* At system boot, before all subsystems have been
4846 * registered, no tasks have been forked, so we don't
4847 * need to invoke fork callbacks here. */
4848 BUG_ON(!list_empty(&init_task
.tasks
));
4850 BUG_ON(online_css(css
));
4852 mutex_unlock(&cgroup_mutex
);
4856 * cgroup_init_early - cgroup initialization at system boot
4858 * Initialize cgroups at system boot, and initialize any
4859 * subsystems that request early init.
4861 int __init
cgroup_init_early(void)
4863 static struct cgroup_sb_opts __initdata opts
;
4864 struct cgroup_subsys
*ss
;
4867 init_cgroup_root(&cgrp_dfl_root
, &opts
);
4868 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
4870 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
4872 for_each_subsys(ss
, i
) {
4873 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
4874 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
4875 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
4877 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
4878 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
4881 ss
->name
= cgroup_subsys_name
[i
];
4884 cgroup_init_subsys(ss
, true);
4890 * cgroup_init - cgroup initialization
4892 * Register cgroup filesystem and /proc file, and initialize
4893 * any subsystems that didn't request early init.
4895 int __init
cgroup_init(void)
4897 struct cgroup_subsys
*ss
;
4901 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_dfl_base_files
));
4902 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_legacy_base_files
));
4904 mutex_lock(&cgroup_mutex
);
4906 /* Add init_css_set to the hash table */
4907 key
= css_set_hash(init_css_set
.subsys
);
4908 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4910 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
4912 mutex_unlock(&cgroup_mutex
);
4914 for_each_subsys(ss
, ssid
) {
4915 if (ss
->early_init
) {
4916 struct cgroup_subsys_state
*css
=
4917 init_css_set
.subsys
[ss
->id
];
4919 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
4921 BUG_ON(css
->id
< 0);
4923 cgroup_init_subsys(ss
, false);
4926 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
4927 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
4930 * Setting dfl_root subsys_mask needs to consider the
4931 * disabled flag and cftype registration needs kmalloc,
4932 * both of which aren't available during early_init.
4937 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
4939 if (cgroup_legacy_files_on_dfl
&& !ss
->dfl_cftypes
)
4940 ss
->dfl_cftypes
= ss
->legacy_cftypes
;
4942 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
4943 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
4945 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
4946 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
4950 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4954 err
= register_filesystem(&cgroup_fs_type
);
4956 kobject_put(cgroup_kobj
);
4960 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4964 static int __init
cgroup_wq_init(void)
4967 * There isn't much point in executing destruction path in
4968 * parallel. Good chunk is serialized with cgroup_mutex anyway.
4969 * Use 1 for @max_active.
4971 * We would prefer to do this in cgroup_init() above, but that
4972 * is called before init_workqueues(): so leave this until after.
4974 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
4975 BUG_ON(!cgroup_destroy_wq
);
4978 * Used to destroy pidlists and separate to serve as flush domain.
4979 * Cap @max_active to 1 too.
4981 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
4983 BUG_ON(!cgroup_pidlist_destroy_wq
);
4987 core_initcall(cgroup_wq_init
);
4990 * proc_cgroup_show()
4991 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4992 * - Used for /proc/<pid>/cgroup.
4995 /* TODO: Use a proper seq_file iterator */
4996 int proc_cgroup_show(struct seq_file
*m
, void *v
)
4999 struct task_struct
*tsk
;
5002 struct cgroup_root
*root
;
5005 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5011 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
5017 mutex_lock(&cgroup_mutex
);
5018 down_read(&css_set_rwsem
);
5020 for_each_root(root
) {
5021 struct cgroup_subsys
*ss
;
5022 struct cgroup
*cgrp
;
5023 int ssid
, count
= 0;
5025 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_root_visible
)
5028 seq_printf(m
, "%d:", root
->hierarchy_id
);
5029 for_each_subsys(ss
, ssid
)
5030 if (root
->subsys_mask
& (1 << ssid
))
5031 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
5032 if (strlen(root
->name
))
5033 seq_printf(m
, "%sname=%s", count
? "," : "",
5036 cgrp
= task_cgroup_from_root(tsk
, root
);
5037 path
= cgroup_path(cgrp
, buf
, PATH_MAX
);
5039 retval
= -ENAMETOOLONG
;
5047 up_read(&css_set_rwsem
);
5048 mutex_unlock(&cgroup_mutex
);
5049 put_task_struct(tsk
);
5056 /* Display information about each subsystem and each hierarchy */
5057 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5059 struct cgroup_subsys
*ss
;
5062 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5064 * ideally we don't want subsystems moving around while we do this.
5065 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5066 * subsys/hierarchy state.
5068 mutex_lock(&cgroup_mutex
);
5070 for_each_subsys(ss
, i
)
5071 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5072 ss
->name
, ss
->root
->hierarchy_id
,
5073 atomic_read(&ss
->root
->nr_cgrps
), !ss
->disabled
);
5075 mutex_unlock(&cgroup_mutex
);
5079 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5081 return single_open(file
, proc_cgroupstats_show
, NULL
);
5084 static const struct file_operations proc_cgroupstats_operations
= {
5085 .open
= cgroupstats_open
,
5087 .llseek
= seq_lseek
,
5088 .release
= single_release
,
5092 * cgroup_fork - initialize cgroup related fields during copy_process()
5093 * @child: pointer to task_struct of forking parent process.
5095 * A task is associated with the init_css_set until cgroup_post_fork()
5096 * attaches it to the parent's css_set. Empty cg_list indicates that
5097 * @child isn't holding reference to its css_set.
5099 void cgroup_fork(struct task_struct
*child
)
5101 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5102 INIT_LIST_HEAD(&child
->cg_list
);
5106 * cgroup_post_fork - called on a new task after adding it to the task list
5107 * @child: the task in question
5109 * Adds the task to the list running through its css_set if necessary and
5110 * call the subsystem fork() callbacks. Has to be after the task is
5111 * visible on the task list in case we race with the first call to
5112 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5115 void cgroup_post_fork(struct task_struct
*child
)
5117 struct cgroup_subsys
*ss
;
5121 * This may race against cgroup_enable_task_cg_links(). As that
5122 * function sets use_task_css_set_links before grabbing
5123 * tasklist_lock and we just went through tasklist_lock to add
5124 * @child, it's guaranteed that either we see the set
5125 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5126 * @child during its iteration.
5128 * If we won the race, @child is associated with %current's
5129 * css_set. Grabbing css_set_rwsem guarantees both that the
5130 * association is stable, and, on completion of the parent's
5131 * migration, @child is visible in the source of migration or
5132 * already in the destination cgroup. This guarantee is necessary
5133 * when implementing operations which need to migrate all tasks of
5134 * a cgroup to another.
5136 * Note that if we lose to cgroup_enable_task_cg_links(), @child
5137 * will remain in init_css_set. This is safe because all tasks are
5138 * in the init_css_set before cg_links is enabled and there's no
5139 * operation which transfers all tasks out of init_css_set.
5141 if (use_task_css_set_links
) {
5142 struct css_set
*cset
;
5144 down_write(&css_set_rwsem
);
5145 cset
= task_css_set(current
);
5146 if (list_empty(&child
->cg_list
)) {
5147 rcu_assign_pointer(child
->cgroups
, cset
);
5148 list_add(&child
->cg_list
, &cset
->tasks
);
5151 up_write(&css_set_rwsem
);
5155 * Call ss->fork(). This must happen after @child is linked on
5156 * css_set; otherwise, @child might change state between ->fork()
5157 * and addition to css_set.
5159 if (need_forkexit_callback
) {
5160 for_each_subsys(ss
, i
)
5167 * cgroup_exit - detach cgroup from exiting task
5168 * @tsk: pointer to task_struct of exiting process
5170 * Description: Detach cgroup from @tsk and release it.
5172 * Note that cgroups marked notify_on_release force every task in
5173 * them to take the global cgroup_mutex mutex when exiting.
5174 * This could impact scaling on very large systems. Be reluctant to
5175 * use notify_on_release cgroups where very high task exit scaling
5176 * is required on large systems.
5178 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5179 * call cgroup_exit() while the task is still competent to handle
5180 * notify_on_release(), then leave the task attached to the root cgroup in
5181 * each hierarchy for the remainder of its exit. No need to bother with
5182 * init_css_set refcnting. init_css_set never goes away and we can't race
5183 * with migration path - PF_EXITING is visible to migration path.
5185 void cgroup_exit(struct task_struct
*tsk
)
5187 struct cgroup_subsys
*ss
;
5188 struct css_set
*cset
;
5189 bool put_cset
= false;
5193 * Unlink from @tsk from its css_set. As migration path can't race
5194 * with us, we can check cg_list without grabbing css_set_rwsem.
5196 if (!list_empty(&tsk
->cg_list
)) {
5197 down_write(&css_set_rwsem
);
5198 list_del_init(&tsk
->cg_list
);
5199 up_write(&css_set_rwsem
);
5203 /* Reassign the task to the init_css_set. */
5204 cset
= task_css_set(tsk
);
5205 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5207 if (need_forkexit_callback
) {
5208 /* see cgroup_post_fork() for details */
5209 for_each_subsys(ss
, i
) {
5211 struct cgroup_subsys_state
*old_css
= cset
->subsys
[i
];
5212 struct cgroup_subsys_state
*css
= task_css(tsk
, i
);
5214 ss
->exit(css
, old_css
, tsk
);
5220 put_css_set(cset
, true);
5223 static void check_for_release(struct cgroup
*cgrp
)
5225 if (cgroup_is_releasable(cgrp
) && list_empty(&cgrp
->cset_links
) &&
5226 !css_has_online_children(&cgrp
->self
)) {
5228 * Control Group is currently removeable. If it's not
5229 * already queued for a userspace notification, queue
5232 int need_schedule_work
= 0;
5234 raw_spin_lock(&release_list_lock
);
5235 if (!cgroup_is_dead(cgrp
) &&
5236 list_empty(&cgrp
->release_list
)) {
5237 list_add(&cgrp
->release_list
, &release_list
);
5238 need_schedule_work
= 1;
5240 raw_spin_unlock(&release_list_lock
);
5241 if (need_schedule_work
)
5242 schedule_work(&release_agent_work
);
5247 * Notify userspace when a cgroup is released, by running the
5248 * configured release agent with the name of the cgroup (path
5249 * relative to the root of cgroup file system) as the argument.
5251 * Most likely, this user command will try to rmdir this cgroup.
5253 * This races with the possibility that some other task will be
5254 * attached to this cgroup before it is removed, or that some other
5255 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5256 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5257 * unused, and this cgroup will be reprieved from its death sentence,
5258 * to continue to serve a useful existence. Next time it's released,
5259 * we will get notified again, if it still has 'notify_on_release' set.
5261 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5262 * means only wait until the task is successfully execve()'d. The
5263 * separate release agent task is forked by call_usermodehelper(),
5264 * then control in this thread returns here, without waiting for the
5265 * release agent task. We don't bother to wait because the caller of
5266 * this routine has no use for the exit status of the release agent
5267 * task, so no sense holding our caller up for that.
5269 static void cgroup_release_agent(struct work_struct
*work
)
5271 BUG_ON(work
!= &release_agent_work
);
5272 mutex_lock(&cgroup_mutex
);
5273 raw_spin_lock(&release_list_lock
);
5274 while (!list_empty(&release_list
)) {
5275 char *argv
[3], *envp
[3];
5277 char *pathbuf
= NULL
, *agentbuf
= NULL
, *path
;
5278 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5281 list_del_init(&cgrp
->release_list
);
5282 raw_spin_unlock(&release_list_lock
);
5283 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5286 path
= cgroup_path(cgrp
, pathbuf
, PATH_MAX
);
5289 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5294 argv
[i
++] = agentbuf
;
5299 /* minimal command environment */
5300 envp
[i
++] = "HOME=/";
5301 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5304 /* Drop the lock while we invoke the usermode helper,
5305 * since the exec could involve hitting disk and hence
5306 * be a slow process */
5307 mutex_unlock(&cgroup_mutex
);
5308 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5309 mutex_lock(&cgroup_mutex
);
5313 raw_spin_lock(&release_list_lock
);
5315 raw_spin_unlock(&release_list_lock
);
5316 mutex_unlock(&cgroup_mutex
);
5319 static int __init
cgroup_disable(char *str
)
5321 struct cgroup_subsys
*ss
;
5325 while ((token
= strsep(&str
, ",")) != NULL
) {
5329 for_each_subsys(ss
, i
) {
5330 if (!strcmp(token
, ss
->name
)) {
5332 printk(KERN_INFO
"Disabling %s control group"
5333 " subsystem\n", ss
->name
);
5340 __setup("cgroup_disable=", cgroup_disable
);
5342 static int __init
cgroup_set_legacy_files_on_dfl(char *str
)
5344 printk("cgroup: using legacy files on the default hierarchy\n");
5345 cgroup_legacy_files_on_dfl
= true;
5348 __setup("cgroup__DEVEL__legacy_files_on_dfl", cgroup_set_legacy_files_on_dfl
);
5351 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5352 * @dentry: directory dentry of interest
5353 * @ss: subsystem of interest
5355 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5356 * to get the corresponding css and return it. If such css doesn't exist
5357 * or can't be pinned, an ERR_PTR value is returned.
5359 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
5360 struct cgroup_subsys
*ss
)
5362 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
5363 struct cgroup_subsys_state
*css
= NULL
;
5364 struct cgroup
*cgrp
;
5366 /* is @dentry a cgroup dir? */
5367 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
5368 kernfs_type(kn
) != KERNFS_DIR
)
5369 return ERR_PTR(-EBADF
);
5374 * This path doesn't originate from kernfs and @kn could already
5375 * have been or be removed at any point. @kn->priv is RCU
5376 * protected for this access. See cgroup_rmdir() for details.
5378 cgrp
= rcu_dereference(kn
->priv
);
5380 css
= cgroup_css(cgrp
, ss
);
5382 if (!css
|| !css_tryget_online(css
))
5383 css
= ERR_PTR(-ENOENT
);
5390 * css_from_id - lookup css by id
5391 * @id: the cgroup id
5392 * @ss: cgroup subsys to be looked into
5394 * Returns the css if there's valid one with @id, otherwise returns NULL.
5395 * Should be called under rcu_read_lock().
5397 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5399 WARN_ON_ONCE(!rcu_read_lock_held());
5400 return idr_find(&ss
->css_idr
, id
);
5403 #ifdef CONFIG_CGROUP_DEBUG
5404 static struct cgroup_subsys_state
*
5405 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
5407 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5410 return ERR_PTR(-ENOMEM
);
5415 static void debug_css_free(struct cgroup_subsys_state
*css
)
5420 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
5423 return cgroup_task_count(css
->cgroup
);
5426 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
5429 return (u64
)(unsigned long)current
->cgroups
;
5432 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
5438 count
= atomic_read(&task_css_set(current
)->refcount
);
5443 static int current_css_set_cg_links_read(struct seq_file
*seq
, void *v
)
5445 struct cgrp_cset_link
*link
;
5446 struct css_set
*cset
;
5449 name_buf
= kmalloc(NAME_MAX
+ 1, GFP_KERNEL
);
5453 down_read(&css_set_rwsem
);
5455 cset
= rcu_dereference(current
->cgroups
);
5456 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5457 struct cgroup
*c
= link
->cgrp
;
5459 cgroup_name(c
, name_buf
, NAME_MAX
+ 1);
5460 seq_printf(seq
, "Root %d group %s\n",
5461 c
->root
->hierarchy_id
, name_buf
);
5464 up_read(&css_set_rwsem
);
5469 #define MAX_TASKS_SHOWN_PER_CSS 25
5470 static int cgroup_css_links_read(struct seq_file
*seq
, void *v
)
5472 struct cgroup_subsys_state
*css
= seq_css(seq
);
5473 struct cgrp_cset_link
*link
;
5475 down_read(&css_set_rwsem
);
5476 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
5477 struct css_set
*cset
= link
->cset
;
5478 struct task_struct
*task
;
5481 seq_printf(seq
, "css_set %p\n", cset
);
5483 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5484 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5486 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5489 list_for_each_entry(task
, &cset
->mg_tasks
, cg_list
) {
5490 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5492 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5496 seq_puts(seq
, " ...\n");
5498 up_read(&css_set_rwsem
);
5502 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
5504 return test_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
5507 static struct cftype debug_files
[] = {
5509 .name
= "taskcount",
5510 .read_u64
= debug_taskcount_read
,
5514 .name
= "current_css_set",
5515 .read_u64
= current_css_set_read
,
5519 .name
= "current_css_set_refcount",
5520 .read_u64
= current_css_set_refcount_read
,
5524 .name
= "current_css_set_cg_links",
5525 .seq_show
= current_css_set_cg_links_read
,
5529 .name
= "cgroup_css_links",
5530 .seq_show
= cgroup_css_links_read
,
5534 .name
= "releasable",
5535 .read_u64
= releasable_read
,
5541 struct cgroup_subsys debug_cgrp_subsys
= {
5542 .css_alloc
= debug_css_alloc
,
5543 .css_free
= debug_css_free
,
5544 .legacy_cftypes
= debug_files
,
5546 #endif /* CONFIG_CGROUP_DEBUG */