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
;
152 /* some controllers are not supported in the default hierarchy */
153 static const unsigned int cgrp_dfl_root_inhibit_ss_mask
= 0
154 #ifdef CONFIG_CGROUP_DEBUG
155 | (1 << debug_cgrp_id
)
159 /* The list of hierarchy roots */
161 static LIST_HEAD(cgroup_roots
);
162 static int cgroup_root_count
;
164 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
165 static DEFINE_IDR(cgroup_hierarchy_idr
);
168 * Assign a monotonically increasing serial number to csses. It guarantees
169 * cgroups with bigger numbers are newer than those with smaller numbers.
170 * Also, as csses are always appended to the parent's ->children list, it
171 * guarantees that sibling csses are always sorted in the ascending serial
172 * number order on the list. Protected by cgroup_mutex.
174 static u64 css_serial_nr_next
= 1;
176 /* This flag indicates whether tasks in the fork and exit paths should
177 * check for fork/exit handlers to call. This avoids us having to do
178 * extra work in the fork/exit path if none of the subsystems need to
181 static int need_forkexit_callback __read_mostly
;
183 static struct cftype cgroup_base_files
[];
185 static void cgroup_put(struct cgroup
*cgrp
);
186 static int rebind_subsystems(struct cgroup_root
*dst_root
,
187 unsigned int ss_mask
);
188 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
189 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
191 static void css_release(struct percpu_ref
*ref
);
192 static void kill_css(struct cgroup_subsys_state
*css
);
193 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
195 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
);
197 /* IDR wrappers which synchronize using cgroup_idr_lock */
198 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
203 idr_preload(gfp_mask
);
204 spin_lock_bh(&cgroup_idr_lock
);
205 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
);
206 spin_unlock_bh(&cgroup_idr_lock
);
211 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
215 spin_lock_bh(&cgroup_idr_lock
);
216 ret
= idr_replace(idr
, ptr
, id
);
217 spin_unlock_bh(&cgroup_idr_lock
);
221 static void cgroup_idr_remove(struct idr
*idr
, int id
)
223 spin_lock_bh(&cgroup_idr_lock
);
225 spin_unlock_bh(&cgroup_idr_lock
);
228 static struct cgroup
*cgroup_parent(struct cgroup
*cgrp
)
230 struct cgroup_subsys_state
*parent_css
= cgrp
->self
.parent
;
233 return container_of(parent_css
, struct cgroup
, self
);
238 * cgroup_css - obtain a cgroup's css for the specified subsystem
239 * @cgrp: the cgroup of interest
240 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
242 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
243 * function must be called either under cgroup_mutex or rcu_read_lock() and
244 * the caller is responsible for pinning the returned css if it wants to
245 * keep accessing it outside the said locks. This function may return
246 * %NULL if @cgrp doesn't have @subsys_id enabled.
248 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
249 struct cgroup_subsys
*ss
)
252 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
253 lockdep_is_held(&cgroup_mutex
));
259 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
260 * @cgrp: the cgroup of interest
261 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
263 * Similar to cgroup_css() but returns the effctive css, which is defined
264 * as the matching css of the nearest ancestor including self which has @ss
265 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
266 * function is guaranteed to return non-NULL css.
268 static struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
269 struct cgroup_subsys
*ss
)
271 lockdep_assert_held(&cgroup_mutex
);
276 if (!(cgrp
->root
->subsys_mask
& (1 << ss
->id
)))
279 while (cgroup_parent(cgrp
) &&
280 !(cgroup_parent(cgrp
)->child_subsys_mask
& (1 << ss
->id
)))
281 cgrp
= cgroup_parent(cgrp
);
283 return cgroup_css(cgrp
, ss
);
286 /* convenient tests for these bits */
287 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
289 return !(cgrp
->self
.flags
& CSS_ONLINE
);
292 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
294 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
295 struct cftype
*cft
= of_cft(of
);
298 * This is open and unprotected implementation of cgroup_css().
299 * seq_css() is only called from a kernfs file operation which has
300 * an active reference on the file. Because all the subsystem
301 * files are drained before a css is disassociated with a cgroup,
302 * the matching css from the cgroup's subsys table is guaranteed to
303 * be and stay valid until the enclosing operation is complete.
306 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
310 EXPORT_SYMBOL_GPL(of_css
);
313 * cgroup_is_descendant - test ancestry
314 * @cgrp: the cgroup to be tested
315 * @ancestor: possible ancestor of @cgrp
317 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
318 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
319 * and @ancestor are accessible.
321 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
324 if (cgrp
== ancestor
)
326 cgrp
= cgroup_parent(cgrp
);
331 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
334 (1 << CGRP_RELEASABLE
) |
335 (1 << CGRP_NOTIFY_ON_RELEASE
);
336 return (cgrp
->flags
& bits
) == bits
;
339 static int notify_on_release(const struct cgroup
*cgrp
)
341 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
345 * for_each_css - iterate all css's of a cgroup
346 * @css: the iteration cursor
347 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
348 * @cgrp: the target cgroup to iterate css's of
350 * Should be called under cgroup_[tree_]mutex.
352 #define for_each_css(css, ssid, cgrp) \
353 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
354 if (!((css) = rcu_dereference_check( \
355 (cgrp)->subsys[(ssid)], \
356 lockdep_is_held(&cgroup_mutex)))) { } \
360 * for_each_e_css - iterate all effective css's of a cgroup
361 * @css: the iteration cursor
362 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
363 * @cgrp: the target cgroup to iterate css's of
365 * Should be called under cgroup_[tree_]mutex.
367 #define for_each_e_css(css, ssid, cgrp) \
368 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
369 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
374 * for_each_subsys - iterate all enabled cgroup subsystems
375 * @ss: the iteration cursor
376 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
378 #define for_each_subsys(ss, ssid) \
379 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
380 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
382 /* iterate across the hierarchies */
383 #define for_each_root(root) \
384 list_for_each_entry((root), &cgroup_roots, root_list)
386 /* iterate over child cgrps, lock should be held throughout iteration */
387 #define cgroup_for_each_live_child(child, cgrp) \
388 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
389 if (({ lockdep_assert_held(&cgroup_mutex); \
390 cgroup_is_dead(child); })) \
394 /* the list of cgroups eligible for automatic release. Protected by
395 * release_list_lock */
396 static LIST_HEAD(release_list
);
397 static DEFINE_RAW_SPINLOCK(release_list_lock
);
398 static void cgroup_release_agent(struct work_struct
*work
);
399 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
400 static void check_for_release(struct cgroup
*cgrp
);
403 * A cgroup can be associated with multiple css_sets as different tasks may
404 * belong to different cgroups on different hierarchies. In the other
405 * direction, a css_set is naturally associated with multiple cgroups.
406 * This M:N relationship is represented by the following link structure
407 * which exists for each association and allows traversing the associations
410 struct cgrp_cset_link
{
411 /* the cgroup and css_set this link associates */
413 struct css_set
*cset
;
415 /* list of cgrp_cset_links anchored at cgrp->cset_links */
416 struct list_head cset_link
;
418 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
419 struct list_head cgrp_link
;
423 * The default css_set - used by init and its children prior to any
424 * hierarchies being mounted. It contains a pointer to the root state
425 * for each subsystem. Also used to anchor the list of css_sets. Not
426 * reference-counted, to improve performance when child cgroups
427 * haven't been created.
429 struct css_set init_css_set
= {
430 .refcount
= ATOMIC_INIT(1),
431 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
432 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
433 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
434 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
435 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
438 static int css_set_count
= 1; /* 1 for init_css_set */
441 * cgroup_update_populated - updated populated count of a cgroup
442 * @cgrp: the target cgroup
443 * @populated: inc or dec populated count
445 * @cgrp is either getting the first task (css_set) or losing the last.
446 * Update @cgrp->populated_cnt accordingly. The count is propagated
447 * towards root so that a given cgroup's populated_cnt is zero iff the
448 * cgroup and all its descendants are empty.
450 * @cgrp's interface file "cgroup.populated" is zero if
451 * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
452 * changes from or to zero, userland is notified that the content of the
453 * interface file has changed. This can be used to detect when @cgrp and
454 * its descendants become populated or empty.
456 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
458 lockdep_assert_held(&css_set_rwsem
);
464 trigger
= !cgrp
->populated_cnt
++;
466 trigger
= !--cgrp
->populated_cnt
;
471 if (cgrp
->populated_kn
)
472 kernfs_notify(cgrp
->populated_kn
);
473 cgrp
= cgroup_parent(cgrp
);
478 * hash table for cgroup groups. This improves the performance to find
479 * an existing css_set. This hash doesn't (currently) take into
480 * account cgroups in empty hierarchies.
482 #define CSS_SET_HASH_BITS 7
483 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
485 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
487 unsigned long key
= 0UL;
488 struct cgroup_subsys
*ss
;
491 for_each_subsys(ss
, i
)
492 key
+= (unsigned long)css
[i
];
493 key
= (key
>> 16) ^ key
;
498 static void put_css_set_locked(struct css_set
*cset
, bool taskexit
)
500 struct cgrp_cset_link
*link
, *tmp_link
;
501 struct cgroup_subsys
*ss
;
504 lockdep_assert_held(&css_set_rwsem
);
506 if (!atomic_dec_and_test(&cset
->refcount
))
509 /* This css_set is dead. unlink it and release cgroup refcounts */
510 for_each_subsys(ss
, ssid
)
511 list_del(&cset
->e_cset_node
[ssid
]);
512 hash_del(&cset
->hlist
);
515 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
516 struct cgroup
*cgrp
= link
->cgrp
;
518 list_del(&link
->cset_link
);
519 list_del(&link
->cgrp_link
);
521 /* @cgrp can't go away while we're holding css_set_rwsem */
522 if (list_empty(&cgrp
->cset_links
)) {
523 cgroup_update_populated(cgrp
, false);
524 if (notify_on_release(cgrp
)) {
526 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
527 check_for_release(cgrp
);
534 kfree_rcu(cset
, rcu_head
);
537 static void put_css_set(struct css_set
*cset
, bool taskexit
)
540 * Ensure that the refcount doesn't hit zero while any readers
541 * can see it. Similar to atomic_dec_and_lock(), but for an
544 if (atomic_add_unless(&cset
->refcount
, -1, 1))
547 down_write(&css_set_rwsem
);
548 put_css_set_locked(cset
, taskexit
);
549 up_write(&css_set_rwsem
);
553 * refcounted get/put for css_set objects
555 static inline void get_css_set(struct css_set
*cset
)
557 atomic_inc(&cset
->refcount
);
561 * compare_css_sets - helper function for find_existing_css_set().
562 * @cset: candidate css_set being tested
563 * @old_cset: existing css_set for a task
564 * @new_cgrp: cgroup that's being entered by the task
565 * @template: desired set of css pointers in css_set (pre-calculated)
567 * Returns true if "cset" matches "old_cset" except for the hierarchy
568 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
570 static bool compare_css_sets(struct css_set
*cset
,
571 struct css_set
*old_cset
,
572 struct cgroup
*new_cgrp
,
573 struct cgroup_subsys_state
*template[])
575 struct list_head
*l1
, *l2
;
578 * On the default hierarchy, there can be csets which are
579 * associated with the same set of cgroups but different csses.
580 * Let's first ensure that csses match.
582 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
586 * Compare cgroup pointers in order to distinguish between
587 * different cgroups in hierarchies. As different cgroups may
588 * share the same effective css, this comparison is always
591 l1
= &cset
->cgrp_links
;
592 l2
= &old_cset
->cgrp_links
;
594 struct cgrp_cset_link
*link1
, *link2
;
595 struct cgroup
*cgrp1
, *cgrp2
;
599 /* See if we reached the end - both lists are equal length. */
600 if (l1
== &cset
->cgrp_links
) {
601 BUG_ON(l2
!= &old_cset
->cgrp_links
);
604 BUG_ON(l2
== &old_cset
->cgrp_links
);
606 /* Locate the cgroups associated with these links. */
607 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
608 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
611 /* Hierarchies should be linked in the same order. */
612 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
615 * If this hierarchy is the hierarchy of the cgroup
616 * that's changing, then we need to check that this
617 * css_set points to the new cgroup; if it's any other
618 * hierarchy, then this css_set should point to the
619 * same cgroup as the old css_set.
621 if (cgrp1
->root
== new_cgrp
->root
) {
622 if (cgrp1
!= new_cgrp
)
633 * find_existing_css_set - init css array and find the matching css_set
634 * @old_cset: the css_set that we're using before the cgroup transition
635 * @cgrp: the cgroup that we're moving into
636 * @template: out param for the new set of csses, should be clear on entry
638 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
640 struct cgroup_subsys_state
*template[])
642 struct cgroup_root
*root
= cgrp
->root
;
643 struct cgroup_subsys
*ss
;
644 struct css_set
*cset
;
649 * Build the set of subsystem state objects that we want to see in the
650 * new css_set. while subsystems can change globally, the entries here
651 * won't change, so no need for locking.
653 for_each_subsys(ss
, i
) {
654 if (root
->subsys_mask
& (1UL << i
)) {
656 * @ss is in this hierarchy, so we want the
657 * effective css from @cgrp.
659 template[i
] = cgroup_e_css(cgrp
, ss
);
662 * @ss is not in this hierarchy, so we don't want
665 template[i
] = old_cset
->subsys
[i
];
669 key
= css_set_hash(template);
670 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
671 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
674 /* This css_set matches what we need */
678 /* No existing cgroup group matched */
682 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
684 struct cgrp_cset_link
*link
, *tmp_link
;
686 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
687 list_del(&link
->cset_link
);
693 * allocate_cgrp_cset_links - allocate cgrp_cset_links
694 * @count: the number of links to allocate
695 * @tmp_links: list_head the allocated links are put on
697 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
698 * through ->cset_link. Returns 0 on success or -errno.
700 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
702 struct cgrp_cset_link
*link
;
705 INIT_LIST_HEAD(tmp_links
);
707 for (i
= 0; i
< count
; i
++) {
708 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
710 free_cgrp_cset_links(tmp_links
);
713 list_add(&link
->cset_link
, tmp_links
);
719 * link_css_set - a helper function to link a css_set to a cgroup
720 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
721 * @cset: the css_set to be linked
722 * @cgrp: the destination cgroup
724 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
727 struct cgrp_cset_link
*link
;
729 BUG_ON(list_empty(tmp_links
));
731 if (cgroup_on_dfl(cgrp
))
732 cset
->dfl_cgrp
= cgrp
;
734 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
738 if (list_empty(&cgrp
->cset_links
))
739 cgroup_update_populated(cgrp
, true);
740 list_move(&link
->cset_link
, &cgrp
->cset_links
);
743 * Always add links to the tail of the list so that the list
744 * is sorted by order of hierarchy creation
746 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
750 * find_css_set - return a new css_set with one cgroup updated
751 * @old_cset: the baseline css_set
752 * @cgrp: the cgroup to be updated
754 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
755 * substituted into the appropriate hierarchy.
757 static struct css_set
*find_css_set(struct css_set
*old_cset
,
760 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
761 struct css_set
*cset
;
762 struct list_head tmp_links
;
763 struct cgrp_cset_link
*link
;
764 struct cgroup_subsys
*ss
;
768 lockdep_assert_held(&cgroup_mutex
);
770 /* First see if we already have a cgroup group that matches
772 down_read(&css_set_rwsem
);
773 cset
= find_existing_css_set(old_cset
, cgrp
, template);
776 up_read(&css_set_rwsem
);
781 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
785 /* Allocate all the cgrp_cset_link objects that we'll need */
786 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
791 atomic_set(&cset
->refcount
, 1);
792 INIT_LIST_HEAD(&cset
->cgrp_links
);
793 INIT_LIST_HEAD(&cset
->tasks
);
794 INIT_LIST_HEAD(&cset
->mg_tasks
);
795 INIT_LIST_HEAD(&cset
->mg_preload_node
);
796 INIT_LIST_HEAD(&cset
->mg_node
);
797 INIT_HLIST_NODE(&cset
->hlist
);
799 /* Copy the set of subsystem state objects generated in
800 * find_existing_css_set() */
801 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
803 down_write(&css_set_rwsem
);
804 /* Add reference counts and links from the new css_set. */
805 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
806 struct cgroup
*c
= link
->cgrp
;
808 if (c
->root
== cgrp
->root
)
810 link_css_set(&tmp_links
, cset
, c
);
813 BUG_ON(!list_empty(&tmp_links
));
817 /* Add @cset to the hash table */
818 key
= css_set_hash(cset
->subsys
);
819 hash_add(css_set_table
, &cset
->hlist
, key
);
821 for_each_subsys(ss
, ssid
)
822 list_add_tail(&cset
->e_cset_node
[ssid
],
823 &cset
->subsys
[ssid
]->cgroup
->e_csets
[ssid
]);
825 up_write(&css_set_rwsem
);
830 static struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
832 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
834 return root_cgrp
->root
;
837 static int cgroup_init_root_id(struct cgroup_root
*root
)
841 lockdep_assert_held(&cgroup_mutex
);
843 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
847 root
->hierarchy_id
= id
;
851 static void cgroup_exit_root_id(struct cgroup_root
*root
)
853 lockdep_assert_held(&cgroup_mutex
);
855 if (root
->hierarchy_id
) {
856 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
857 root
->hierarchy_id
= 0;
861 static void cgroup_free_root(struct cgroup_root
*root
)
864 /* hierarhcy ID shoulid already have been released */
865 WARN_ON_ONCE(root
->hierarchy_id
);
867 idr_destroy(&root
->cgroup_idr
);
872 static void cgroup_destroy_root(struct cgroup_root
*root
)
874 struct cgroup
*cgrp
= &root
->cgrp
;
875 struct cgrp_cset_link
*link
, *tmp_link
;
877 mutex_lock(&cgroup_mutex
);
879 BUG_ON(atomic_read(&root
->nr_cgrps
));
880 BUG_ON(!list_empty(&cgrp
->self
.children
));
882 /* Rebind all subsystems back to the default hierarchy */
883 rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
);
886 * Release all the links from cset_links to this hierarchy's
889 down_write(&css_set_rwsem
);
891 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
892 list_del(&link
->cset_link
);
893 list_del(&link
->cgrp_link
);
896 up_write(&css_set_rwsem
);
898 if (!list_empty(&root
->root_list
)) {
899 list_del(&root
->root_list
);
903 cgroup_exit_root_id(root
);
905 mutex_unlock(&cgroup_mutex
);
907 kernfs_destroy_root(root
->kf_root
);
908 cgroup_free_root(root
);
911 /* look up cgroup associated with given css_set on the specified hierarchy */
912 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
913 struct cgroup_root
*root
)
915 struct cgroup
*res
= NULL
;
917 lockdep_assert_held(&cgroup_mutex
);
918 lockdep_assert_held(&css_set_rwsem
);
920 if (cset
== &init_css_set
) {
923 struct cgrp_cset_link
*link
;
925 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
926 struct cgroup
*c
= link
->cgrp
;
928 if (c
->root
== root
) {
940 * Return the cgroup for "task" from the given hierarchy. Must be
941 * called with cgroup_mutex and css_set_rwsem held.
943 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
944 struct cgroup_root
*root
)
947 * No need to lock the task - since we hold cgroup_mutex the
948 * task can't change groups, so the only thing that can happen
949 * is that it exits and its css is set back to init_css_set.
951 return cset_cgroup_from_root(task_css_set(task
), root
);
955 * A task must hold cgroup_mutex to modify cgroups.
957 * Any task can increment and decrement the count field without lock.
958 * So in general, code holding cgroup_mutex can't rely on the count
959 * field not changing. However, if the count goes to zero, then only
960 * cgroup_attach_task() can increment it again. Because a count of zero
961 * means that no tasks are currently attached, therefore there is no
962 * way a task attached to that cgroup can fork (the other way to
963 * increment the count). So code holding cgroup_mutex can safely
964 * assume that if the count is zero, it will stay zero. Similarly, if
965 * a task holds cgroup_mutex on a cgroup with zero count, it
966 * knows that the cgroup won't be removed, as cgroup_rmdir()
969 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
970 * (usually) take cgroup_mutex. These are the two most performance
971 * critical pieces of code here. The exception occurs on cgroup_exit(),
972 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
973 * is taken, and if the cgroup count is zero, a usermode call made
974 * to the release agent with the name of the cgroup (path relative to
975 * the root of cgroup file system) as the argument.
977 * A cgroup can only be deleted if both its 'count' of using tasks
978 * is zero, and its list of 'children' cgroups is empty. Since all
979 * tasks in the system use _some_ cgroup, and since there is always at
980 * least one task in the system (init, pid == 1), therefore, root cgroup
981 * always has either children cgroups and/or using tasks. So we don't
982 * need a special hack to ensure that root cgroup cannot be deleted.
984 * P.S. One more locking exception. RCU is used to guard the
985 * update of a tasks cgroup pointer by cgroup_attach_task()
988 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned int subsys_mask
);
989 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
990 static const struct file_operations proc_cgroupstats_operations
;
992 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
995 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
996 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
))
997 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s.%s",
998 cft
->ss
->name
, cft
->name
);
1000 strncpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1005 * cgroup_file_mode - deduce file mode of a control file
1006 * @cft: the control file in question
1008 * returns cft->mode if ->mode is not 0
1009 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
1010 * returns S_IRUGO if it has only a read handler
1011 * returns S_IWUSR if it has only a write hander
1013 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1020 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1023 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
)
1029 static void cgroup_get(struct cgroup
*cgrp
)
1031 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
1032 css_get(&cgrp
->self
);
1035 static void cgroup_put(struct cgroup
*cgrp
)
1037 css_put(&cgrp
->self
);
1041 * cgroup_refresh_child_subsys_mask - update child_subsys_mask
1042 * @cgrp: the target cgroup
1044 * On the default hierarchy, a subsystem may request other subsystems to be
1045 * enabled together through its ->depends_on mask. In such cases, more
1046 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1048 * This function determines which subsystems need to be enabled given the
1049 * current @cgrp->subtree_control and records it in
1050 * @cgrp->child_subsys_mask. The resulting mask is always a superset of
1051 * @cgrp->subtree_control and follows the usual hierarchy rules.
1053 static void cgroup_refresh_child_subsys_mask(struct cgroup
*cgrp
)
1055 struct cgroup
*parent
= cgroup_parent(cgrp
);
1056 unsigned int cur_ss_mask
= cgrp
->subtree_control
;
1057 struct cgroup_subsys
*ss
;
1060 lockdep_assert_held(&cgroup_mutex
);
1062 if (!cgroup_on_dfl(cgrp
)) {
1063 cgrp
->child_subsys_mask
= cur_ss_mask
;
1068 unsigned int new_ss_mask
= cur_ss_mask
;
1070 for_each_subsys(ss
, ssid
)
1071 if (cur_ss_mask
& (1 << ssid
))
1072 new_ss_mask
|= ss
->depends_on
;
1075 * Mask out subsystems which aren't available. This can
1076 * happen only if some depended-upon subsystems were bound
1077 * to non-default hierarchies.
1080 new_ss_mask
&= parent
->child_subsys_mask
;
1082 new_ss_mask
&= cgrp
->root
->subsys_mask
;
1084 if (new_ss_mask
== cur_ss_mask
)
1086 cur_ss_mask
= new_ss_mask
;
1089 cgrp
->child_subsys_mask
= cur_ss_mask
;
1093 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1094 * @kn: the kernfs_node being serviced
1096 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1097 * the method finishes if locking succeeded. Note that once this function
1098 * returns the cgroup returned by cgroup_kn_lock_live() may become
1099 * inaccessible any time. If the caller intends to continue to access the
1100 * cgroup, it should pin it before invoking this function.
1102 static void cgroup_kn_unlock(struct kernfs_node
*kn
)
1104 struct cgroup
*cgrp
;
1106 if (kernfs_type(kn
) == KERNFS_DIR
)
1109 cgrp
= kn
->parent
->priv
;
1111 mutex_unlock(&cgroup_mutex
);
1113 kernfs_unbreak_active_protection(kn
);
1118 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1119 * @kn: the kernfs_node being serviced
1121 * This helper is to be used by a cgroup kernfs method currently servicing
1122 * @kn. It breaks the active protection, performs cgroup locking and
1123 * verifies that the associated cgroup is alive. Returns the cgroup if
1124 * alive; otherwise, %NULL. A successful return should be undone by a
1125 * matching cgroup_kn_unlock() invocation.
1127 * Any cgroup kernfs method implementation which requires locking the
1128 * associated cgroup should use this helper. It avoids nesting cgroup
1129 * locking under kernfs active protection and allows all kernfs operations
1130 * including self-removal.
1132 static struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
)
1134 struct cgroup
*cgrp
;
1136 if (kernfs_type(kn
) == KERNFS_DIR
)
1139 cgrp
= kn
->parent
->priv
;
1142 * We're gonna grab cgroup_mutex which nests outside kernfs
1143 * active_ref. cgroup liveliness check alone provides enough
1144 * protection against removal. Ensure @cgrp stays accessible and
1145 * break the active_ref protection.
1148 kernfs_break_active_protection(kn
);
1150 mutex_lock(&cgroup_mutex
);
1152 if (!cgroup_is_dead(cgrp
))
1155 cgroup_kn_unlock(kn
);
1159 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1161 char name
[CGROUP_FILE_NAME_MAX
];
1163 lockdep_assert_held(&cgroup_mutex
);
1164 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1168 * cgroup_clear_dir - remove subsys files in a cgroup directory
1169 * @cgrp: target cgroup
1170 * @subsys_mask: mask of the subsystem ids whose files should be removed
1172 static void cgroup_clear_dir(struct cgroup
*cgrp
, unsigned int subsys_mask
)
1174 struct cgroup_subsys
*ss
;
1177 for_each_subsys(ss
, i
) {
1178 struct cftype
*cfts
;
1180 if (!(subsys_mask
& (1 << i
)))
1182 list_for_each_entry(cfts
, &ss
->cfts
, node
)
1183 cgroup_addrm_files(cgrp
, cfts
, false);
1187 static int rebind_subsystems(struct cgroup_root
*dst_root
, unsigned int ss_mask
)
1189 struct cgroup_subsys
*ss
;
1190 unsigned int tmp_ss_mask
;
1193 lockdep_assert_held(&cgroup_mutex
);
1195 for_each_subsys(ss
, ssid
) {
1196 if (!(ss_mask
& (1 << ssid
)))
1199 /* if @ss has non-root csses attached to it, can't move */
1200 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)))
1203 /* can't move between two non-dummy roots either */
1204 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1208 /* skip creating root files on dfl_root for inhibited subsystems */
1209 tmp_ss_mask
= ss_mask
;
1210 if (dst_root
== &cgrp_dfl_root
)
1211 tmp_ss_mask
&= ~cgrp_dfl_root_inhibit_ss_mask
;
1213 ret
= cgroup_populate_dir(&dst_root
->cgrp
, tmp_ss_mask
);
1215 if (dst_root
!= &cgrp_dfl_root
)
1219 * Rebinding back to the default root is not allowed to
1220 * fail. Using both default and non-default roots should
1221 * be rare. Moving subsystems back and forth even more so.
1222 * Just warn about it and continue.
1224 if (cgrp_dfl_root_visible
) {
1225 pr_warn("failed to create files (%d) while rebinding 0x%x to default root\n",
1227 pr_warn("you may retry by moving them to a different hierarchy and unbinding\n");
1232 * Nothing can fail from this point on. Remove files for the
1233 * removed subsystems and rebind each subsystem.
1235 for_each_subsys(ss
, ssid
)
1236 if (ss_mask
& (1 << ssid
))
1237 cgroup_clear_dir(&ss
->root
->cgrp
, 1 << ssid
);
1239 for_each_subsys(ss
, ssid
) {
1240 struct cgroup_root
*src_root
;
1241 struct cgroup_subsys_state
*css
;
1242 struct css_set
*cset
;
1244 if (!(ss_mask
& (1 << ssid
)))
1247 src_root
= ss
->root
;
1248 css
= cgroup_css(&src_root
->cgrp
, ss
);
1250 WARN_ON(!css
|| cgroup_css(&dst_root
->cgrp
, ss
));
1252 RCU_INIT_POINTER(src_root
->cgrp
.subsys
[ssid
], NULL
);
1253 rcu_assign_pointer(dst_root
->cgrp
.subsys
[ssid
], css
);
1254 ss
->root
= dst_root
;
1255 css
->cgroup
= &dst_root
->cgrp
;
1257 down_write(&css_set_rwsem
);
1258 hash_for_each(css_set_table
, i
, cset
, hlist
)
1259 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1260 &dst_root
->cgrp
.e_csets
[ss
->id
]);
1261 up_write(&css_set_rwsem
);
1263 src_root
->subsys_mask
&= ~(1 << ssid
);
1264 src_root
->cgrp
.subtree_control
&= ~(1 << ssid
);
1265 cgroup_refresh_child_subsys_mask(&src_root
->cgrp
);
1267 /* default hierarchy doesn't enable controllers by default */
1268 dst_root
->subsys_mask
|= 1 << ssid
;
1269 if (dst_root
!= &cgrp_dfl_root
) {
1270 dst_root
->cgrp
.subtree_control
|= 1 << ssid
;
1271 cgroup_refresh_child_subsys_mask(&dst_root
->cgrp
);
1278 kernfs_activate(dst_root
->cgrp
.kn
);
1282 static int cgroup_show_options(struct seq_file
*seq
,
1283 struct kernfs_root
*kf_root
)
1285 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1286 struct cgroup_subsys
*ss
;
1289 for_each_subsys(ss
, ssid
)
1290 if (root
->subsys_mask
& (1 << ssid
))
1291 seq_printf(seq
, ",%s", ss
->name
);
1292 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
)
1293 seq_puts(seq
, ",sane_behavior");
1294 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1295 seq_puts(seq
, ",noprefix");
1296 if (root
->flags
& CGRP_ROOT_XATTR
)
1297 seq_puts(seq
, ",xattr");
1299 spin_lock(&release_agent_path_lock
);
1300 if (strlen(root
->release_agent_path
))
1301 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1302 spin_unlock(&release_agent_path_lock
);
1304 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
1305 seq_puts(seq
, ",clone_children");
1306 if (strlen(root
->name
))
1307 seq_printf(seq
, ",name=%s", root
->name
);
1311 struct cgroup_sb_opts
{
1312 unsigned int subsys_mask
;
1314 char *release_agent
;
1315 bool cpuset_clone_children
;
1317 /* User explicitly requested empty subsystem */
1321 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1323 char *token
, *o
= data
;
1324 bool all_ss
= false, one_ss
= false;
1325 unsigned int mask
= -1U;
1326 struct cgroup_subsys
*ss
;
1329 #ifdef CONFIG_CPUSETS
1330 mask
= ~(1U << cpuset_cgrp_id
);
1333 memset(opts
, 0, sizeof(*opts
));
1335 while ((token
= strsep(&o
, ",")) != NULL
) {
1338 if (!strcmp(token
, "none")) {
1339 /* Explicitly have no subsystems */
1343 if (!strcmp(token
, "all")) {
1344 /* Mutually exclusive option 'all' + subsystem name */
1350 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1351 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1354 if (!strcmp(token
, "noprefix")) {
1355 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1358 if (!strcmp(token
, "clone_children")) {
1359 opts
->cpuset_clone_children
= true;
1362 if (!strcmp(token
, "xattr")) {
1363 opts
->flags
|= CGRP_ROOT_XATTR
;
1366 if (!strncmp(token
, "release_agent=", 14)) {
1367 /* Specifying two release agents is forbidden */
1368 if (opts
->release_agent
)
1370 opts
->release_agent
=
1371 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1372 if (!opts
->release_agent
)
1376 if (!strncmp(token
, "name=", 5)) {
1377 const char *name
= token
+ 5;
1378 /* Can't specify an empty name */
1381 /* Must match [\w.-]+ */
1382 for (i
= 0; i
< strlen(name
); i
++) {
1386 if ((c
== '.') || (c
== '-') || (c
== '_'))
1390 /* Specifying two names is forbidden */
1393 opts
->name
= kstrndup(name
,
1394 MAX_CGROUP_ROOT_NAMELEN
- 1,
1402 for_each_subsys(ss
, i
) {
1403 if (strcmp(token
, ss
->name
))
1408 /* Mutually exclusive option 'all' + subsystem name */
1411 opts
->subsys_mask
|= (1 << i
);
1416 if (i
== CGROUP_SUBSYS_COUNT
)
1420 /* Consistency checks */
1422 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1423 pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1425 if ((opts
->flags
& (CGRP_ROOT_NOPREFIX
| CGRP_ROOT_XATTR
)) ||
1426 opts
->cpuset_clone_children
|| opts
->release_agent
||
1428 pr_err("sane_behavior: noprefix, xattr, clone_children, release_agent and name are not allowed\n");
1433 * If the 'all' option was specified select all the
1434 * subsystems, otherwise if 'none', 'name=' and a subsystem
1435 * name options were not specified, let's default to 'all'
1437 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1438 for_each_subsys(ss
, i
)
1440 opts
->subsys_mask
|= (1 << i
);
1443 * We either have to specify by name or by subsystems. (So
1444 * all empty hierarchies must have a name).
1446 if (!opts
->subsys_mask
&& !opts
->name
)
1451 * Option noprefix was introduced just for backward compatibility
1452 * with the old cpuset, so we allow noprefix only if mounting just
1453 * the cpuset subsystem.
1455 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1459 /* Can't specify "none" and some subsystems */
1460 if (opts
->subsys_mask
&& opts
->none
)
1466 static int cgroup_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1469 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1470 struct cgroup_sb_opts opts
;
1471 unsigned int added_mask
, removed_mask
;
1473 if (root
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1474 pr_err("sane_behavior: remount is not allowed\n");
1478 mutex_lock(&cgroup_mutex
);
1480 /* See what subsystems are wanted */
1481 ret
= parse_cgroupfs_options(data
, &opts
);
1485 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1486 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1487 task_tgid_nr(current
), current
->comm
);
1489 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1490 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1492 /* Don't allow flags or name to change at remount */
1493 if (((opts
.flags
^ root
->flags
) & CGRP_ROOT_OPTION_MASK
) ||
1494 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1495 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1496 opts
.flags
& CGRP_ROOT_OPTION_MASK
, opts
.name
?: "",
1497 root
->flags
& CGRP_ROOT_OPTION_MASK
, root
->name
);
1502 /* remounting is not allowed for populated hierarchies */
1503 if (!list_empty(&root
->cgrp
.self
.children
)) {
1508 ret
= rebind_subsystems(root
, added_mask
);
1512 rebind_subsystems(&cgrp_dfl_root
, removed_mask
);
1514 if (opts
.release_agent
) {
1515 spin_lock(&release_agent_path_lock
);
1516 strcpy(root
->release_agent_path
, opts
.release_agent
);
1517 spin_unlock(&release_agent_path_lock
);
1520 kfree(opts
.release_agent
);
1522 mutex_unlock(&cgroup_mutex
);
1527 * To reduce the fork() overhead for systems that are not actually using
1528 * their cgroups capability, we don't maintain the lists running through
1529 * each css_set to its tasks until we see the list actually used - in other
1530 * words after the first mount.
1532 static bool use_task_css_set_links __read_mostly
;
1534 static void cgroup_enable_task_cg_lists(void)
1536 struct task_struct
*p
, *g
;
1538 down_write(&css_set_rwsem
);
1540 if (use_task_css_set_links
)
1543 use_task_css_set_links
= true;
1546 * We need tasklist_lock because RCU is not safe against
1547 * while_each_thread(). Besides, a forking task that has passed
1548 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1549 * is not guaranteed to have its child immediately visible in the
1550 * tasklist if we walk through it with RCU.
1552 read_lock(&tasklist_lock
);
1553 do_each_thread(g
, p
) {
1554 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1555 task_css_set(p
) != &init_css_set
);
1558 * We should check if the process is exiting, otherwise
1559 * it will race with cgroup_exit() in that the list
1560 * entry won't be deleted though the process has exited.
1561 * Do it while holding siglock so that we don't end up
1562 * racing against cgroup_exit().
1564 spin_lock_irq(&p
->sighand
->siglock
);
1565 if (!(p
->flags
& PF_EXITING
)) {
1566 struct css_set
*cset
= task_css_set(p
);
1568 list_add(&p
->cg_list
, &cset
->tasks
);
1571 spin_unlock_irq(&p
->sighand
->siglock
);
1572 } while_each_thread(g
, p
);
1573 read_unlock(&tasklist_lock
);
1575 up_write(&css_set_rwsem
);
1578 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1580 struct cgroup_subsys
*ss
;
1583 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1584 INIT_LIST_HEAD(&cgrp
->self
.children
);
1585 INIT_LIST_HEAD(&cgrp
->cset_links
);
1586 INIT_LIST_HEAD(&cgrp
->release_list
);
1587 INIT_LIST_HEAD(&cgrp
->pidlists
);
1588 mutex_init(&cgrp
->pidlist_mutex
);
1589 cgrp
->self
.cgroup
= cgrp
;
1590 cgrp
->self
.flags
|= CSS_ONLINE
;
1592 for_each_subsys(ss
, ssid
)
1593 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1595 init_waitqueue_head(&cgrp
->offline_waitq
);
1598 static void init_cgroup_root(struct cgroup_root
*root
,
1599 struct cgroup_sb_opts
*opts
)
1601 struct cgroup
*cgrp
= &root
->cgrp
;
1603 INIT_LIST_HEAD(&root
->root_list
);
1604 atomic_set(&root
->nr_cgrps
, 1);
1606 init_cgroup_housekeeping(cgrp
);
1607 idr_init(&root
->cgroup_idr
);
1609 root
->flags
= opts
->flags
;
1610 if (opts
->release_agent
)
1611 strcpy(root
->release_agent_path
, opts
->release_agent
);
1613 strcpy(root
->name
, opts
->name
);
1614 if (opts
->cpuset_clone_children
)
1615 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1618 static int cgroup_setup_root(struct cgroup_root
*root
, unsigned int ss_mask
)
1620 LIST_HEAD(tmp_links
);
1621 struct cgroup
*root_cgrp
= &root
->cgrp
;
1622 struct css_set
*cset
;
1625 lockdep_assert_held(&cgroup_mutex
);
1627 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_NOWAIT
);
1630 root_cgrp
->id
= ret
;
1632 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
);
1637 * We're accessing css_set_count without locking css_set_rwsem here,
1638 * but that's OK - it can only be increased by someone holding
1639 * cgroup_lock, and that's us. The worst that can happen is that we
1640 * have some link structures left over
1642 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1646 ret
= cgroup_init_root_id(root
);
1650 root
->kf_root
= kernfs_create_root(&cgroup_kf_syscall_ops
,
1651 KERNFS_ROOT_CREATE_DEACTIVATED
,
1653 if (IS_ERR(root
->kf_root
)) {
1654 ret
= PTR_ERR(root
->kf_root
);
1657 root_cgrp
->kn
= root
->kf_root
->kn
;
1659 ret
= cgroup_addrm_files(root_cgrp
, cgroup_base_files
, true);
1663 ret
= rebind_subsystems(root
, ss_mask
);
1668 * There must be no failure case after here, since rebinding takes
1669 * care of subsystems' refcounts, which are explicitly dropped in
1670 * the failure exit path.
1672 list_add(&root
->root_list
, &cgroup_roots
);
1673 cgroup_root_count
++;
1676 * Link the root cgroup in this hierarchy into all the css_set
1679 down_write(&css_set_rwsem
);
1680 hash_for_each(css_set_table
, i
, cset
, hlist
)
1681 link_css_set(&tmp_links
, cset
, root_cgrp
);
1682 up_write(&css_set_rwsem
);
1684 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
1685 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
1687 kernfs_activate(root_cgrp
->kn
);
1692 kernfs_destroy_root(root
->kf_root
);
1693 root
->kf_root
= NULL
;
1695 cgroup_exit_root_id(root
);
1697 percpu_ref_cancel_init(&root_cgrp
->self
.refcnt
);
1699 free_cgrp_cset_links(&tmp_links
);
1703 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1704 int flags
, const char *unused_dev_name
,
1707 struct cgroup_root
*root
;
1708 struct cgroup_sb_opts opts
;
1709 struct dentry
*dentry
;
1714 * The first time anyone tries to mount a cgroup, enable the list
1715 * linking each css_set to its tasks and fix up all existing tasks.
1717 if (!use_task_css_set_links
)
1718 cgroup_enable_task_cg_lists();
1720 mutex_lock(&cgroup_mutex
);
1722 /* First find the desired set of subsystems */
1723 ret
= parse_cgroupfs_options(data
, &opts
);
1727 /* look for a matching existing root */
1728 if (!opts
.subsys_mask
&& !opts
.none
&& !opts
.name
) {
1729 cgrp_dfl_root_visible
= true;
1730 root
= &cgrp_dfl_root
;
1731 cgroup_get(&root
->cgrp
);
1736 for_each_root(root
) {
1737 bool name_match
= false;
1739 if (root
== &cgrp_dfl_root
)
1743 * If we asked for a name then it must match. Also, if
1744 * name matches but sybsys_mask doesn't, we should fail.
1745 * Remember whether name matched.
1748 if (strcmp(opts
.name
, root
->name
))
1754 * If we asked for subsystems (or explicitly for no
1755 * subsystems) then they must match.
1757 if ((opts
.subsys_mask
|| opts
.none
) &&
1758 (opts
.subsys_mask
!= root
->subsys_mask
)) {
1765 if ((root
->flags
^ opts
.flags
) & CGRP_ROOT_OPTION_MASK
) {
1766 if ((root
->flags
| opts
.flags
) & CGRP_ROOT_SANE_BEHAVIOR
) {
1767 pr_err("sane_behavior: new mount options should match the existing superblock\n");
1771 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
1776 * A root's lifetime is governed by its root cgroup.
1777 * tryget_live failure indicate that the root is being
1778 * destroyed. Wait for destruction to complete so that the
1779 * subsystems are free. We can use wait_queue for the wait
1780 * but this path is super cold. Let's just sleep for a bit
1783 if (!percpu_ref_tryget_live(&root
->cgrp
.self
.refcnt
)) {
1784 mutex_unlock(&cgroup_mutex
);
1786 ret
= restart_syscall();
1795 * No such thing, create a new one. name= matching without subsys
1796 * specification is allowed for already existing hierarchies but we
1797 * can't create new one without subsys specification.
1799 if (!opts
.subsys_mask
&& !opts
.none
) {
1804 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1810 init_cgroup_root(root
, &opts
);
1812 ret
= cgroup_setup_root(root
, opts
.subsys_mask
);
1814 cgroup_free_root(root
);
1817 mutex_unlock(&cgroup_mutex
);
1819 kfree(opts
.release_agent
);
1823 return ERR_PTR(ret
);
1825 dentry
= kernfs_mount(fs_type
, flags
, root
->kf_root
,
1826 CGROUP_SUPER_MAGIC
, &new_sb
);
1827 if (IS_ERR(dentry
) || !new_sb
)
1828 cgroup_put(&root
->cgrp
);
1832 static void cgroup_kill_sb(struct super_block
*sb
)
1834 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
1835 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1838 * If @root doesn't have any mounts or children, start killing it.
1839 * This prevents new mounts by disabling percpu_ref_tryget_live().
1840 * cgroup_mount() may wait for @root's release.
1842 * And don't kill the default root.
1844 if (css_has_online_children(&root
->cgrp
.self
) ||
1845 root
== &cgrp_dfl_root
)
1846 cgroup_put(&root
->cgrp
);
1848 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
1853 static struct file_system_type cgroup_fs_type
= {
1855 .mount
= cgroup_mount
,
1856 .kill_sb
= cgroup_kill_sb
,
1859 static struct kobject
*cgroup_kobj
;
1862 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1863 * @task: target task
1864 * @buf: the buffer to write the path into
1865 * @buflen: the length of the buffer
1867 * Determine @task's cgroup on the first (the one with the lowest non-zero
1868 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1869 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1870 * cgroup controller callbacks.
1872 * Return value is the same as kernfs_path().
1874 char *task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
1876 struct cgroup_root
*root
;
1877 struct cgroup
*cgrp
;
1878 int hierarchy_id
= 1;
1881 mutex_lock(&cgroup_mutex
);
1882 down_read(&css_set_rwsem
);
1884 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
1887 cgrp
= task_cgroup_from_root(task
, root
);
1888 path
= cgroup_path(cgrp
, buf
, buflen
);
1890 /* if no hierarchy exists, everyone is in "/" */
1891 if (strlcpy(buf
, "/", buflen
) < buflen
)
1895 up_read(&css_set_rwsem
);
1896 mutex_unlock(&cgroup_mutex
);
1899 EXPORT_SYMBOL_GPL(task_cgroup_path
);
1901 /* used to track tasks and other necessary states during migration */
1902 struct cgroup_taskset
{
1903 /* the src and dst cset list running through cset->mg_node */
1904 struct list_head src_csets
;
1905 struct list_head dst_csets
;
1908 * Fields for cgroup_taskset_*() iteration.
1910 * Before migration is committed, the target migration tasks are on
1911 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
1912 * the csets on ->dst_csets. ->csets point to either ->src_csets
1913 * or ->dst_csets depending on whether migration is committed.
1915 * ->cur_csets and ->cur_task point to the current task position
1918 struct list_head
*csets
;
1919 struct css_set
*cur_cset
;
1920 struct task_struct
*cur_task
;
1924 * cgroup_taskset_first - reset taskset and return the first task
1925 * @tset: taskset of interest
1927 * @tset iteration is initialized and the first task is returned.
1929 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1931 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
1932 tset
->cur_task
= NULL
;
1934 return cgroup_taskset_next(tset
);
1938 * cgroup_taskset_next - iterate to the next task in taskset
1939 * @tset: taskset of interest
1941 * Return the next task in @tset. Iteration must have been initialized
1942 * with cgroup_taskset_first().
1944 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1946 struct css_set
*cset
= tset
->cur_cset
;
1947 struct task_struct
*task
= tset
->cur_task
;
1949 while (&cset
->mg_node
!= tset
->csets
) {
1951 task
= list_first_entry(&cset
->mg_tasks
,
1952 struct task_struct
, cg_list
);
1954 task
= list_next_entry(task
, cg_list
);
1956 if (&task
->cg_list
!= &cset
->mg_tasks
) {
1957 tset
->cur_cset
= cset
;
1958 tset
->cur_task
= task
;
1962 cset
= list_next_entry(cset
, mg_node
);
1970 * cgroup_task_migrate - move a task from one cgroup to another.
1971 * @old_cgrp: the cgroup @tsk is being migrated from
1972 * @tsk: the task being migrated
1973 * @new_cset: the new css_set @tsk is being attached to
1975 * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
1977 static void cgroup_task_migrate(struct cgroup
*old_cgrp
,
1978 struct task_struct
*tsk
,
1979 struct css_set
*new_cset
)
1981 struct css_set
*old_cset
;
1983 lockdep_assert_held(&cgroup_mutex
);
1984 lockdep_assert_held(&css_set_rwsem
);
1987 * We are synchronized through threadgroup_lock() against PF_EXITING
1988 * setting such that we can't race against cgroup_exit() changing the
1989 * css_set to init_css_set and dropping the old one.
1991 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1992 old_cset
= task_css_set(tsk
);
1994 get_css_set(new_cset
);
1995 rcu_assign_pointer(tsk
->cgroups
, new_cset
);
1998 * Use move_tail so that cgroup_taskset_first() still returns the
1999 * leader after migration. This works because cgroup_migrate()
2000 * ensures that the dst_cset of the leader is the first on the
2001 * tset's dst_csets list.
2003 list_move_tail(&tsk
->cg_list
, &new_cset
->mg_tasks
);
2006 * We just gained a reference on old_cset by taking it from the
2007 * task. As trading it for new_cset is protected by cgroup_mutex,
2008 * we're safe to drop it here; it will be freed under RCU.
2010 set_bit(CGRP_RELEASABLE
, &old_cgrp
->flags
);
2011 put_css_set_locked(old_cset
, false);
2015 * cgroup_migrate_finish - cleanup after attach
2016 * @preloaded_csets: list of preloaded css_sets
2018 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2019 * those functions for details.
2021 static void cgroup_migrate_finish(struct list_head
*preloaded_csets
)
2023 struct css_set
*cset
, *tmp_cset
;
2025 lockdep_assert_held(&cgroup_mutex
);
2027 down_write(&css_set_rwsem
);
2028 list_for_each_entry_safe(cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2029 cset
->mg_src_cgrp
= NULL
;
2030 cset
->mg_dst_cset
= NULL
;
2031 list_del_init(&cset
->mg_preload_node
);
2032 put_css_set_locked(cset
, false);
2034 up_write(&css_set_rwsem
);
2038 * cgroup_migrate_add_src - add a migration source css_set
2039 * @src_cset: the source css_set to add
2040 * @dst_cgrp: the destination cgroup
2041 * @preloaded_csets: list of preloaded css_sets
2043 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2044 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2045 * up by cgroup_migrate_finish().
2047 * This function may be called without holding threadgroup_lock even if the
2048 * target is a process. Threads may be created and destroyed but as long
2049 * as cgroup_mutex is not dropped, no new css_set can be put into play and
2050 * the preloaded css_sets are guaranteed to cover all migrations.
2052 static void cgroup_migrate_add_src(struct css_set
*src_cset
,
2053 struct cgroup
*dst_cgrp
,
2054 struct list_head
*preloaded_csets
)
2056 struct cgroup
*src_cgrp
;
2058 lockdep_assert_held(&cgroup_mutex
);
2059 lockdep_assert_held(&css_set_rwsem
);
2061 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2063 if (!list_empty(&src_cset
->mg_preload_node
))
2066 WARN_ON(src_cset
->mg_src_cgrp
);
2067 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2068 WARN_ON(!list_empty(&src_cset
->mg_node
));
2070 src_cset
->mg_src_cgrp
= src_cgrp
;
2071 get_css_set(src_cset
);
2072 list_add(&src_cset
->mg_preload_node
, preloaded_csets
);
2076 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2077 * @dst_cgrp: the destination cgroup (may be %NULL)
2078 * @preloaded_csets: list of preloaded source css_sets
2080 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
2081 * have been preloaded to @preloaded_csets. This function looks up and
2082 * pins all destination css_sets, links each to its source, and append them
2083 * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each
2084 * source css_set is assumed to be its cgroup on the default hierarchy.
2086 * This function must be called after cgroup_migrate_add_src() has been
2087 * called on each migration source css_set. After migration is performed
2088 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2091 static int cgroup_migrate_prepare_dst(struct cgroup
*dst_cgrp
,
2092 struct list_head
*preloaded_csets
)
2095 struct css_set
*src_cset
, *tmp_cset
;
2097 lockdep_assert_held(&cgroup_mutex
);
2100 * Except for the root, child_subsys_mask must be zero for a cgroup
2101 * with tasks so that child cgroups don't compete against tasks.
2103 if (dst_cgrp
&& cgroup_on_dfl(dst_cgrp
) && cgroup_parent(dst_cgrp
) &&
2104 dst_cgrp
->child_subsys_mask
)
2107 /* look up the dst cset for each src cset and link it to src */
2108 list_for_each_entry_safe(src_cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2109 struct css_set
*dst_cset
;
2111 dst_cset
= find_css_set(src_cset
,
2112 dst_cgrp
?: src_cset
->dfl_cgrp
);
2116 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2119 * If src cset equals dst, it's noop. Drop the src.
2120 * cgroup_migrate() will skip the cset too. Note that we
2121 * can't handle src == dst as some nodes are used by both.
2123 if (src_cset
== dst_cset
) {
2124 src_cset
->mg_src_cgrp
= NULL
;
2125 list_del_init(&src_cset
->mg_preload_node
);
2126 put_css_set(src_cset
, false);
2127 put_css_set(dst_cset
, false);
2131 src_cset
->mg_dst_cset
= dst_cset
;
2133 if (list_empty(&dst_cset
->mg_preload_node
))
2134 list_add(&dst_cset
->mg_preload_node
, &csets
);
2136 put_css_set(dst_cset
, false);
2139 list_splice_tail(&csets
, preloaded_csets
);
2142 cgroup_migrate_finish(&csets
);
2147 * cgroup_migrate - migrate a process or task to a cgroup
2148 * @cgrp: the destination cgroup
2149 * @leader: the leader of the process or the task to migrate
2150 * @threadgroup: whether @leader points to the whole process or a single task
2152 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
2153 * process, the caller must be holding threadgroup_lock of @leader. The
2154 * caller is also responsible for invoking cgroup_migrate_add_src() and
2155 * cgroup_migrate_prepare_dst() on the targets before invoking this
2156 * function and following up with cgroup_migrate_finish().
2158 * As long as a controller's ->can_attach() doesn't fail, this function is
2159 * guaranteed to succeed. This means that, excluding ->can_attach()
2160 * failure, when migrating multiple targets, the success or failure can be
2161 * decided for all targets by invoking group_migrate_prepare_dst() before
2162 * actually starting migrating.
2164 static int cgroup_migrate(struct cgroup
*cgrp
, struct task_struct
*leader
,
2167 struct cgroup_taskset tset
= {
2168 .src_csets
= LIST_HEAD_INIT(tset
.src_csets
),
2169 .dst_csets
= LIST_HEAD_INIT(tset
.dst_csets
),
2170 .csets
= &tset
.src_csets
,
2172 struct cgroup_subsys_state
*css
, *failed_css
= NULL
;
2173 struct css_set
*cset
, *tmp_cset
;
2174 struct task_struct
*task
, *tmp_task
;
2178 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2179 * already PF_EXITING could be freed from underneath us unless we
2180 * take an rcu_read_lock.
2182 down_write(&css_set_rwsem
);
2186 /* @task either already exited or can't exit until the end */
2187 if (task
->flags
& PF_EXITING
)
2190 /* leave @task alone if post_fork() hasn't linked it yet */
2191 if (list_empty(&task
->cg_list
))
2194 cset
= task_css_set(task
);
2195 if (!cset
->mg_src_cgrp
)
2199 * cgroup_taskset_first() must always return the leader.
2200 * Take care to avoid disturbing the ordering.
2202 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2203 if (list_empty(&cset
->mg_node
))
2204 list_add_tail(&cset
->mg_node
, &tset
.src_csets
);
2205 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2206 list_move_tail(&cset
->mg_dst_cset
->mg_node
,
2211 } while_each_thread(leader
, task
);
2213 up_write(&css_set_rwsem
);
2215 /* methods shouldn't be called if no task is actually migrating */
2216 if (list_empty(&tset
.src_csets
))
2219 /* check that we can legitimately attach to the cgroup */
2220 for_each_e_css(css
, i
, cgrp
) {
2221 if (css
->ss
->can_attach
) {
2222 ret
= css
->ss
->can_attach(css
, &tset
);
2225 goto out_cancel_attach
;
2231 * Now that we're guaranteed success, proceed to move all tasks to
2232 * the new cgroup. There are no failure cases after here, so this
2233 * is the commit point.
2235 down_write(&css_set_rwsem
);
2236 list_for_each_entry(cset
, &tset
.src_csets
, mg_node
) {
2237 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
)
2238 cgroup_task_migrate(cset
->mg_src_cgrp
, task
,
2241 up_write(&css_set_rwsem
);
2244 * Migration is committed, all target tasks are now on dst_csets.
2245 * Nothing is sensitive to fork() after this point. Notify
2246 * controllers that migration is complete.
2248 tset
.csets
= &tset
.dst_csets
;
2250 for_each_e_css(css
, i
, cgrp
)
2251 if (css
->ss
->attach
)
2252 css
->ss
->attach(css
, &tset
);
2255 goto out_release_tset
;
2258 for_each_e_css(css
, i
, cgrp
) {
2259 if (css
== failed_css
)
2261 if (css
->ss
->cancel_attach
)
2262 css
->ss
->cancel_attach(css
, &tset
);
2265 down_write(&css_set_rwsem
);
2266 list_splice_init(&tset
.dst_csets
, &tset
.src_csets
);
2267 list_for_each_entry_safe(cset
, tmp_cset
, &tset
.src_csets
, mg_node
) {
2268 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2269 list_del_init(&cset
->mg_node
);
2271 up_write(&css_set_rwsem
);
2276 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2277 * @dst_cgrp: the cgroup to attach to
2278 * @leader: the task or the leader of the threadgroup to be attached
2279 * @threadgroup: attach the whole threadgroup?
2281 * Call holding cgroup_mutex and threadgroup_lock of @leader.
2283 static int cgroup_attach_task(struct cgroup
*dst_cgrp
,
2284 struct task_struct
*leader
, bool threadgroup
)
2286 LIST_HEAD(preloaded_csets
);
2287 struct task_struct
*task
;
2290 /* look up all src csets */
2291 down_read(&css_set_rwsem
);
2295 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
,
2299 } while_each_thread(leader
, task
);
2301 up_read(&css_set_rwsem
);
2303 /* prepare dst csets and commit */
2304 ret
= cgroup_migrate_prepare_dst(dst_cgrp
, &preloaded_csets
);
2306 ret
= cgroup_migrate(dst_cgrp
, leader
, threadgroup
);
2308 cgroup_migrate_finish(&preloaded_csets
);
2313 * Find the task_struct of the task to attach by vpid and pass it along to the
2314 * function to attach either it or all tasks in its threadgroup. Will lock
2315 * cgroup_mutex and threadgroup.
2317 static ssize_t
__cgroup_procs_write(struct kernfs_open_file
*of
, char *buf
,
2318 size_t nbytes
, loff_t off
, bool threadgroup
)
2320 struct task_struct
*tsk
;
2321 const struct cred
*cred
= current_cred(), *tcred
;
2322 struct cgroup
*cgrp
;
2326 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2329 cgrp
= cgroup_kn_lock_live(of
->kn
);
2336 tsk
= find_task_by_vpid(pid
);
2340 goto out_unlock_cgroup
;
2343 * even if we're attaching all tasks in the thread group, we
2344 * only need to check permissions on one of them.
2346 tcred
= __task_cred(tsk
);
2347 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2348 !uid_eq(cred
->euid
, tcred
->uid
) &&
2349 !uid_eq(cred
->euid
, tcred
->suid
)) {
2352 goto out_unlock_cgroup
;
2358 tsk
= tsk
->group_leader
;
2361 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2362 * trapped in a cpuset, or RT worker may be born in a cgroup
2363 * with no rt_runtime allocated. Just say no.
2365 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2368 goto out_unlock_cgroup
;
2371 get_task_struct(tsk
);
2374 threadgroup_lock(tsk
);
2376 if (!thread_group_leader(tsk
)) {
2378 * a race with de_thread from another thread's exec()
2379 * may strip us of our leadership, if this happens,
2380 * there is no choice but to throw this task away and
2381 * try again; this is
2382 * "double-double-toil-and-trouble-check locking".
2384 threadgroup_unlock(tsk
);
2385 put_task_struct(tsk
);
2386 goto retry_find_task
;
2390 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2392 threadgroup_unlock(tsk
);
2394 put_task_struct(tsk
);
2396 cgroup_kn_unlock(of
->kn
);
2397 return ret
?: nbytes
;
2401 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2402 * @from: attach to all cgroups of a given task
2403 * @tsk: the task to be attached
2405 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2407 struct cgroup_root
*root
;
2410 mutex_lock(&cgroup_mutex
);
2411 for_each_root(root
) {
2412 struct cgroup
*from_cgrp
;
2414 if (root
== &cgrp_dfl_root
)
2417 down_read(&css_set_rwsem
);
2418 from_cgrp
= task_cgroup_from_root(from
, root
);
2419 up_read(&css_set_rwsem
);
2421 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2425 mutex_unlock(&cgroup_mutex
);
2429 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2431 static ssize_t
cgroup_tasks_write(struct kernfs_open_file
*of
,
2432 char *buf
, size_t nbytes
, loff_t off
)
2434 return __cgroup_procs_write(of
, buf
, nbytes
, off
, false);
2437 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
2438 char *buf
, size_t nbytes
, loff_t off
)
2440 return __cgroup_procs_write(of
, buf
, nbytes
, off
, true);
2443 static ssize_t
cgroup_release_agent_write(struct kernfs_open_file
*of
,
2444 char *buf
, size_t nbytes
, loff_t off
)
2446 struct cgroup
*cgrp
;
2448 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2450 cgrp
= cgroup_kn_lock_live(of
->kn
);
2453 spin_lock(&release_agent_path_lock
);
2454 strlcpy(cgrp
->root
->release_agent_path
, strstrip(buf
),
2455 sizeof(cgrp
->root
->release_agent_path
));
2456 spin_unlock(&release_agent_path_lock
);
2457 cgroup_kn_unlock(of
->kn
);
2461 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
2463 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2465 spin_lock(&release_agent_path_lock
);
2466 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2467 spin_unlock(&release_agent_path_lock
);
2468 seq_putc(seq
, '\n');
2472 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
2474 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2476 seq_printf(seq
, "%d\n", cgroup_sane_behavior(cgrp
));
2480 static void cgroup_print_ss_mask(struct seq_file
*seq
, unsigned int ss_mask
)
2482 struct cgroup_subsys
*ss
;
2483 bool printed
= false;
2486 for_each_subsys(ss
, ssid
) {
2487 if (ss_mask
& (1 << ssid
)) {
2490 seq_printf(seq
, "%s", ss
->name
);
2495 seq_putc(seq
, '\n');
2498 /* show controllers which are currently attached to the default hierarchy */
2499 static int cgroup_root_controllers_show(struct seq_file
*seq
, void *v
)
2501 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2503 cgroup_print_ss_mask(seq
, cgrp
->root
->subsys_mask
&
2504 ~cgrp_dfl_root_inhibit_ss_mask
);
2508 /* show controllers which are enabled from the parent */
2509 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2511 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2513 cgroup_print_ss_mask(seq
, cgroup_parent(cgrp
)->subtree_control
);
2517 /* show controllers which are enabled for a given cgroup's children */
2518 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2520 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2522 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2527 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2528 * @cgrp: root of the subtree to update csses for
2530 * @cgrp's child_subsys_mask has changed and its subtree's (self excluded)
2531 * css associations need to be updated accordingly. This function looks up
2532 * all css_sets which are attached to the subtree, creates the matching
2533 * updated css_sets and migrates the tasks to the new ones.
2535 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2537 LIST_HEAD(preloaded_csets
);
2538 struct cgroup_subsys_state
*css
;
2539 struct css_set
*src_cset
;
2542 lockdep_assert_held(&cgroup_mutex
);
2544 /* look up all csses currently attached to @cgrp's subtree */
2545 down_read(&css_set_rwsem
);
2546 css_for_each_descendant_pre(css
, cgroup_css(cgrp
, NULL
)) {
2547 struct cgrp_cset_link
*link
;
2549 /* self is not affected by child_subsys_mask change */
2550 if (css
->cgroup
== cgrp
)
2553 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
)
2554 cgroup_migrate_add_src(link
->cset
, cgrp
,
2557 up_read(&css_set_rwsem
);
2559 /* NULL dst indicates self on default hierarchy */
2560 ret
= cgroup_migrate_prepare_dst(NULL
, &preloaded_csets
);
2564 list_for_each_entry(src_cset
, &preloaded_csets
, mg_preload_node
) {
2565 struct task_struct
*last_task
= NULL
, *task
;
2567 /* src_csets precede dst_csets, break on the first dst_cset */
2568 if (!src_cset
->mg_src_cgrp
)
2572 * All tasks in src_cset need to be migrated to the
2573 * matching dst_cset. Empty it process by process. We
2574 * walk tasks but migrate processes. The leader might even
2575 * belong to a different cset but such src_cset would also
2576 * be among the target src_csets because the default
2577 * hierarchy enforces per-process membership.
2580 down_read(&css_set_rwsem
);
2581 task
= list_first_entry_or_null(&src_cset
->tasks
,
2582 struct task_struct
, cg_list
);
2584 task
= task
->group_leader
;
2585 WARN_ON_ONCE(!task_css_set(task
)->mg_src_cgrp
);
2586 get_task_struct(task
);
2588 up_read(&css_set_rwsem
);
2593 /* guard against possible infinite loop */
2594 if (WARN(last_task
== task
,
2595 "cgroup: update_dfl_csses failed to make progress, aborting in inconsistent state\n"))
2599 threadgroup_lock(task
);
2600 /* raced against de_thread() from another thread? */
2601 if (!thread_group_leader(task
)) {
2602 threadgroup_unlock(task
);
2603 put_task_struct(task
);
2607 ret
= cgroup_migrate(src_cset
->dfl_cgrp
, task
, true);
2609 threadgroup_unlock(task
);
2610 put_task_struct(task
);
2612 if (WARN(ret
, "cgroup: failed to update controllers for the default hierarchy (%d), further operations may crash or hang\n", ret
))
2618 cgroup_migrate_finish(&preloaded_csets
);
2622 /* change the enabled child controllers for a cgroup in the default hierarchy */
2623 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
2624 char *buf
, size_t nbytes
,
2627 unsigned int enable
= 0, disable
= 0;
2628 unsigned int css_enable
, css_disable
, old_ctrl
, new_ctrl
;
2629 struct cgroup
*cgrp
, *child
;
2630 struct cgroup_subsys
*ss
;
2635 * Parse input - space separated list of subsystem names prefixed
2636 * with either + or -.
2638 buf
= strstrip(buf
);
2639 while ((tok
= strsep(&buf
, " "))) {
2642 for_each_subsys(ss
, ssid
) {
2643 if (ss
->disabled
|| strcmp(tok
+ 1, ss
->name
) ||
2644 ((1 << ss
->id
) & cgrp_dfl_root_inhibit_ss_mask
))
2648 enable
|= 1 << ssid
;
2649 disable
&= ~(1 << ssid
);
2650 } else if (*tok
== '-') {
2651 disable
|= 1 << ssid
;
2652 enable
&= ~(1 << ssid
);
2658 if (ssid
== CGROUP_SUBSYS_COUNT
)
2662 cgrp
= cgroup_kn_lock_live(of
->kn
);
2666 for_each_subsys(ss
, ssid
) {
2667 if (enable
& (1 << ssid
)) {
2668 if (cgrp
->subtree_control
& (1 << ssid
)) {
2669 enable
&= ~(1 << ssid
);
2673 /* unavailable or not enabled on the parent? */
2674 if (!(cgrp_dfl_root
.subsys_mask
& (1 << ssid
)) ||
2675 (cgroup_parent(cgrp
) &&
2676 !(cgroup_parent(cgrp
)->subtree_control
& (1 << ssid
)))) {
2682 * @ss is already enabled through dependency and
2683 * we'll just make it visible. Skip draining.
2685 if (cgrp
->child_subsys_mask
& (1 << ssid
))
2689 * Because css offlining is asynchronous, userland
2690 * might try to re-enable the same controller while
2691 * the previous instance is still around. In such
2692 * cases, wait till it's gone using offline_waitq.
2694 cgroup_for_each_live_child(child
, cgrp
) {
2697 if (!cgroup_css(child
, ss
))
2701 prepare_to_wait(&child
->offline_waitq
, &wait
,
2702 TASK_UNINTERRUPTIBLE
);
2703 cgroup_kn_unlock(of
->kn
);
2705 finish_wait(&child
->offline_waitq
, &wait
);
2708 return restart_syscall();
2710 } else if (disable
& (1 << ssid
)) {
2711 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
2712 disable
&= ~(1 << ssid
);
2716 /* a child has it enabled? */
2717 cgroup_for_each_live_child(child
, cgrp
) {
2718 if (child
->subtree_control
& (1 << ssid
)) {
2726 if (!enable
&& !disable
) {
2732 * Except for the root, subtree_control must be zero for a cgroup
2733 * with tasks so that child cgroups don't compete against tasks.
2735 if (enable
&& cgroup_parent(cgrp
) && !list_empty(&cgrp
->cset_links
)) {
2741 * Update subsys masks and calculate what needs to be done. More
2742 * subsystems than specified may need to be enabled or disabled
2743 * depending on subsystem dependencies.
2745 cgrp
->subtree_control
|= enable
;
2746 cgrp
->subtree_control
&= ~disable
;
2748 old_ctrl
= cgrp
->child_subsys_mask
;
2749 cgroup_refresh_child_subsys_mask(cgrp
);
2750 new_ctrl
= cgrp
->child_subsys_mask
;
2752 css_enable
= ~old_ctrl
& new_ctrl
;
2753 css_disable
= old_ctrl
& ~new_ctrl
;
2754 enable
|= css_enable
;
2755 disable
|= css_disable
;
2758 * Create new csses or make the existing ones visible. A css is
2759 * created invisible if it's being implicitly enabled through
2760 * dependency. An invisible css is made visible when the userland
2761 * explicitly enables it.
2763 for_each_subsys(ss
, ssid
) {
2764 if (!(enable
& (1 << ssid
)))
2767 cgroup_for_each_live_child(child
, cgrp
) {
2768 if (css_enable
& (1 << ssid
))
2769 ret
= create_css(child
, ss
,
2770 cgrp
->subtree_control
& (1 << ssid
));
2772 ret
= cgroup_populate_dir(child
, 1 << ssid
);
2779 * At this point, cgroup_e_css() results reflect the new csses
2780 * making the following cgroup_update_dfl_csses() properly update
2781 * css associations of all tasks in the subtree.
2783 ret
= cgroup_update_dfl_csses(cgrp
);
2788 * All tasks are migrated out of disabled csses. Kill or hide
2789 * them. A css is hidden when the userland requests it to be
2790 * disabled while other subsystems are still depending on it. The
2791 * css must not actively control resources and be in the vanilla
2792 * state if it's made visible again later. Controllers which may
2793 * be depended upon should provide ->css_reset() for this purpose.
2795 for_each_subsys(ss
, ssid
) {
2796 if (!(disable
& (1 << ssid
)))
2799 cgroup_for_each_live_child(child
, cgrp
) {
2800 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
2802 if (css_disable
& (1 << ssid
)) {
2805 cgroup_clear_dir(child
, 1 << ssid
);
2812 kernfs_activate(cgrp
->kn
);
2815 cgroup_kn_unlock(of
->kn
);
2816 return ret
?: nbytes
;
2819 cgrp
->subtree_control
&= ~enable
;
2820 cgrp
->subtree_control
|= disable
;
2821 cgroup_refresh_child_subsys_mask(cgrp
);
2823 for_each_subsys(ss
, ssid
) {
2824 if (!(enable
& (1 << ssid
)))
2827 cgroup_for_each_live_child(child
, cgrp
) {
2828 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
2833 if (css_enable
& (1 << ssid
))
2836 cgroup_clear_dir(child
, 1 << ssid
);
2842 static int cgroup_populated_show(struct seq_file
*seq
, void *v
)
2844 seq_printf(seq
, "%d\n", (bool)seq_css(seq
)->cgroup
->populated_cnt
);
2848 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
2849 size_t nbytes
, loff_t off
)
2851 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
2852 struct cftype
*cft
= of
->kn
->priv
;
2853 struct cgroup_subsys_state
*css
;
2857 return cft
->write(of
, buf
, nbytes
, off
);
2860 * kernfs guarantees that a file isn't deleted with operations in
2861 * flight, which means that the matching css is and stays alive and
2862 * doesn't need to be pinned. The RCU locking is not necessary
2863 * either. It's just for the convenience of using cgroup_css().
2866 css
= cgroup_css(cgrp
, cft
->ss
);
2869 if (cft
->write_u64
) {
2870 unsigned long long v
;
2871 ret
= kstrtoull(buf
, 0, &v
);
2873 ret
= cft
->write_u64(css
, cft
, v
);
2874 } else if (cft
->write_s64
) {
2876 ret
= kstrtoll(buf
, 0, &v
);
2878 ret
= cft
->write_s64(css
, cft
, v
);
2883 return ret
?: nbytes
;
2886 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
2888 return seq_cft(seq
)->seq_start(seq
, ppos
);
2891 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
2893 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
2896 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
2898 seq_cft(seq
)->seq_stop(seq
, v
);
2901 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2903 struct cftype
*cft
= seq_cft(m
);
2904 struct cgroup_subsys_state
*css
= seq_css(m
);
2907 return cft
->seq_show(m
, arg
);
2910 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
2911 else if (cft
->read_s64
)
2912 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
2918 static struct kernfs_ops cgroup_kf_single_ops
= {
2919 .atomic_write_len
= PAGE_SIZE
,
2920 .write
= cgroup_file_write
,
2921 .seq_show
= cgroup_seqfile_show
,
2924 static struct kernfs_ops cgroup_kf_ops
= {
2925 .atomic_write_len
= PAGE_SIZE
,
2926 .write
= cgroup_file_write
,
2927 .seq_start
= cgroup_seqfile_start
,
2928 .seq_next
= cgroup_seqfile_next
,
2929 .seq_stop
= cgroup_seqfile_stop
,
2930 .seq_show
= cgroup_seqfile_show
,
2934 * cgroup_rename - Only allow simple rename of directories in place.
2936 static int cgroup_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
2937 const char *new_name_str
)
2939 struct cgroup
*cgrp
= kn
->priv
;
2942 if (kernfs_type(kn
) != KERNFS_DIR
)
2944 if (kn
->parent
!= new_parent
)
2948 * This isn't a proper migration and its usefulness is very
2949 * limited. Disallow if sane_behavior.
2951 if (cgroup_sane_behavior(cgrp
))
2955 * We're gonna grab cgroup_mutex which nests outside kernfs
2956 * active_ref. kernfs_rename() doesn't require active_ref
2957 * protection. Break them before grabbing cgroup_mutex.
2959 kernfs_break_active_protection(new_parent
);
2960 kernfs_break_active_protection(kn
);
2962 mutex_lock(&cgroup_mutex
);
2964 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
2966 mutex_unlock(&cgroup_mutex
);
2968 kernfs_unbreak_active_protection(kn
);
2969 kernfs_unbreak_active_protection(new_parent
);
2973 /* set uid and gid of cgroup dirs and files to that of the creator */
2974 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
2976 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
2977 .ia_uid
= current_fsuid(),
2978 .ia_gid
= current_fsgid(), };
2980 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
2981 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
2984 return kernfs_setattr(kn
, &iattr
);
2987 static int cgroup_add_file(struct cgroup
*cgrp
, struct cftype
*cft
)
2989 char name
[CGROUP_FILE_NAME_MAX
];
2990 struct kernfs_node
*kn
;
2991 struct lock_class_key
*key
= NULL
;
2994 #ifdef CONFIG_DEBUG_LOCK_ALLOC
2995 key
= &cft
->lockdep_key
;
2997 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
2998 cgroup_file_mode(cft
), 0, cft
->kf_ops
, cft
,
3003 ret
= cgroup_kn_set_ugid(kn
);
3009 if (cft
->seq_show
== cgroup_populated_show
)
3010 cgrp
->populated_kn
= kn
;
3015 * cgroup_addrm_files - add or remove files to a cgroup directory
3016 * @cgrp: the target cgroup
3017 * @cfts: array of cftypes to be added
3018 * @is_add: whether to add or remove
3020 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3021 * For removals, this function never fails. If addition fails, this
3022 * function doesn't remove files already added. The caller is responsible
3025 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cftype cfts
[],
3031 lockdep_assert_held(&cgroup_mutex
);
3033 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3034 /* does cft->flags tell us to skip this file on @cgrp? */
3035 if ((cft
->flags
& CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3037 if ((cft
->flags
& CFTYPE_INSANE
) && cgroup_sane_behavior(cgrp
))
3039 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3041 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3045 ret
= cgroup_add_file(cgrp
, cft
);
3047 pr_warn("%s: failed to add %s, err=%d\n",
3048 __func__
, cft
->name
, ret
);
3052 cgroup_rm_file(cgrp
, cft
);
3058 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3061 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3062 struct cgroup
*root
= &ss
->root
->cgrp
;
3063 struct cgroup_subsys_state
*css
;
3066 lockdep_assert_held(&cgroup_mutex
);
3068 /* add/rm files for all cgroups created before */
3069 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3070 struct cgroup
*cgrp
= css
->cgroup
;
3072 if (cgroup_is_dead(cgrp
))
3075 ret
= cgroup_addrm_files(cgrp
, cfts
, is_add
);
3081 kernfs_activate(root
->kn
);
3085 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3089 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3090 /* free copy for custom atomic_write_len, see init_cftypes() */
3091 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3098 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3102 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3103 struct kernfs_ops
*kf_ops
;
3105 WARN_ON(cft
->ss
|| cft
->kf_ops
);
3108 kf_ops
= &cgroup_kf_ops
;
3110 kf_ops
= &cgroup_kf_single_ops
;
3113 * Ugh... if @cft wants a custom max_write_len, we need to
3114 * make a copy of kf_ops to set its atomic_write_len.
3116 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
3117 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
3119 cgroup_exit_cftypes(cfts
);
3122 kf_ops
->atomic_write_len
= cft
->max_write_len
;
3125 cft
->kf_ops
= kf_ops
;
3132 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
3134 lockdep_assert_held(&cgroup_mutex
);
3136 if (!cfts
|| !cfts
[0].ss
)
3139 list_del(&cfts
->node
);
3140 cgroup_apply_cftypes(cfts
, false);
3141 cgroup_exit_cftypes(cfts
);
3146 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3147 * @cfts: zero-length name terminated array of cftypes
3149 * Unregister @cfts. Files described by @cfts are removed from all
3150 * existing cgroups and all future cgroups won't have them either. This
3151 * function can be called anytime whether @cfts' subsys is attached or not.
3153 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3156 int cgroup_rm_cftypes(struct cftype
*cfts
)
3160 mutex_lock(&cgroup_mutex
);
3161 ret
= cgroup_rm_cftypes_locked(cfts
);
3162 mutex_unlock(&cgroup_mutex
);
3167 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3168 * @ss: target cgroup subsystem
3169 * @cfts: zero-length name terminated array of cftypes
3171 * Register @cfts to @ss. Files described by @cfts are created for all
3172 * existing cgroups to which @ss is attached and all future cgroups will
3173 * have them too. This function can be called anytime whether @ss is
3176 * Returns 0 on successful registration, -errno on failure. Note that this
3177 * function currently returns 0 as long as @cfts registration is successful
3178 * even if some file creation attempts on existing cgroups fail.
3180 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3187 if (!cfts
|| cfts
[0].name
[0] == '\0')
3190 ret
= cgroup_init_cftypes(ss
, cfts
);
3194 mutex_lock(&cgroup_mutex
);
3196 list_add_tail(&cfts
->node
, &ss
->cfts
);
3197 ret
= cgroup_apply_cftypes(cfts
, true);
3199 cgroup_rm_cftypes_locked(cfts
);
3201 mutex_unlock(&cgroup_mutex
);
3206 * cgroup_task_count - count the number of tasks in a cgroup.
3207 * @cgrp: the cgroup in question
3209 * Return the number of tasks in the cgroup.
3211 static int cgroup_task_count(const struct cgroup
*cgrp
)
3214 struct cgrp_cset_link
*link
;
3216 down_read(&css_set_rwsem
);
3217 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
3218 count
+= atomic_read(&link
->cset
->refcount
);
3219 up_read(&css_set_rwsem
);
3224 * css_next_child - find the next child of a given css
3225 * @pos: the current position (%NULL to initiate traversal)
3226 * @parent: css whose children to walk
3228 * This function returns the next child of @parent and should be called
3229 * under either cgroup_mutex or RCU read lock. The only requirement is
3230 * that @parent and @pos are accessible. The next sibling is guaranteed to
3231 * be returned regardless of their states.
3233 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3234 * css which finished ->css_online() is guaranteed to be visible in the
3235 * future iterations and will stay visible until the last reference is put.
3236 * A css which hasn't finished ->css_online() or already finished
3237 * ->css_offline() may show up during traversal. It's each subsystem's
3238 * responsibility to synchronize against on/offlining.
3240 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
3241 struct cgroup_subsys_state
*parent
)
3243 struct cgroup_subsys_state
*next
;
3245 cgroup_assert_mutex_or_rcu_locked();
3248 * @pos could already have been unlinked from the sibling list.
3249 * Once a cgroup is removed, its ->sibling.next is no longer
3250 * updated when its next sibling changes. CSS_RELEASED is set when
3251 * @pos is taken off list, at which time its next pointer is valid,
3252 * and, as releases are serialized, the one pointed to by the next
3253 * pointer is guaranteed to not have started release yet. This
3254 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3255 * critical section, the one pointed to by its next pointer is
3256 * guaranteed to not have finished its RCU grace period even if we
3257 * have dropped rcu_read_lock() inbetween iterations.
3259 * If @pos has CSS_RELEASED set, its next pointer can't be
3260 * dereferenced; however, as each css is given a monotonically
3261 * increasing unique serial number and always appended to the
3262 * sibling list, the next one can be found by walking the parent's
3263 * children until the first css with higher serial number than
3264 * @pos's. While this path can be slower, it happens iff iteration
3265 * races against release and the race window is very small.
3268 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
3269 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
3270 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
3272 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
3273 if (next
->serial_nr
> pos
->serial_nr
)
3278 * @next, if not pointing to the head, can be dereferenced and is
3281 if (&next
->sibling
!= &parent
->children
)
3287 * css_next_descendant_pre - find the next descendant for pre-order walk
3288 * @pos: the current position (%NULL to initiate traversal)
3289 * @root: css whose descendants to walk
3291 * To be used by css_for_each_descendant_pre(). Find the next descendant
3292 * to visit for pre-order traversal of @root's descendants. @root is
3293 * included in the iteration and the first node to be visited.
3295 * While this function requires cgroup_mutex or RCU read locking, it
3296 * doesn't require the whole traversal to be contained in a single critical
3297 * section. This function will return the correct next descendant as long
3298 * as both @pos and @root are accessible and @pos is a descendant of @root.
3300 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3301 * css which finished ->css_online() is guaranteed to be visible in the
3302 * future iterations and will stay visible until the last reference is put.
3303 * A css which hasn't finished ->css_online() or already finished
3304 * ->css_offline() may show up during traversal. It's each subsystem's
3305 * responsibility to synchronize against on/offlining.
3307 struct cgroup_subsys_state
*
3308 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3309 struct cgroup_subsys_state
*root
)
3311 struct cgroup_subsys_state
*next
;
3313 cgroup_assert_mutex_or_rcu_locked();
3315 /* if first iteration, visit @root */
3319 /* visit the first child if exists */
3320 next
= css_next_child(NULL
, pos
);
3324 /* no child, visit my or the closest ancestor's next sibling */
3325 while (pos
!= root
) {
3326 next
= css_next_child(pos
, pos
->parent
);
3336 * css_rightmost_descendant - return the rightmost descendant of a css
3337 * @pos: css of interest
3339 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3340 * is returned. This can be used during pre-order traversal to skip
3343 * While this function requires cgroup_mutex or RCU read locking, it
3344 * doesn't require the whole traversal to be contained in a single critical
3345 * section. This function will return the correct rightmost descendant as
3346 * long as @pos is accessible.
3348 struct cgroup_subsys_state
*
3349 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3351 struct cgroup_subsys_state
*last
, *tmp
;
3353 cgroup_assert_mutex_or_rcu_locked();
3357 /* ->prev isn't RCU safe, walk ->next till the end */
3359 css_for_each_child(tmp
, last
)
3366 static struct cgroup_subsys_state
*
3367 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3369 struct cgroup_subsys_state
*last
;
3373 pos
= css_next_child(NULL
, pos
);
3380 * css_next_descendant_post - find the next descendant for post-order walk
3381 * @pos: the current position (%NULL to initiate traversal)
3382 * @root: css whose descendants to walk
3384 * To be used by css_for_each_descendant_post(). Find the next descendant
3385 * to visit for post-order traversal of @root's descendants. @root is
3386 * included in the iteration and the last node to be visited.
3388 * While this function requires cgroup_mutex or RCU read locking, it
3389 * doesn't require the whole traversal to be contained in a single critical
3390 * section. This function will return the correct next descendant as long
3391 * as both @pos and @cgroup are accessible and @pos is a descendant of
3394 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3395 * css which finished ->css_online() is guaranteed to be visible in the
3396 * future iterations and will stay visible until the last reference is put.
3397 * A css which hasn't finished ->css_online() or already finished
3398 * ->css_offline() may show up during traversal. It's each subsystem's
3399 * responsibility to synchronize against on/offlining.
3401 struct cgroup_subsys_state
*
3402 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3403 struct cgroup_subsys_state
*root
)
3405 struct cgroup_subsys_state
*next
;
3407 cgroup_assert_mutex_or_rcu_locked();
3409 /* if first iteration, visit leftmost descendant which may be @root */
3411 return css_leftmost_descendant(root
);
3413 /* if we visited @root, we're done */
3417 /* if there's an unvisited sibling, visit its leftmost descendant */
3418 next
= css_next_child(pos
, pos
->parent
);
3420 return css_leftmost_descendant(next
);
3422 /* no sibling left, visit parent */
3427 * css_has_online_children - does a css have online children
3428 * @css: the target css
3430 * Returns %true if @css has any online children; otherwise, %false. This
3431 * function can be called from any context but the caller is responsible
3432 * for synchronizing against on/offlining as necessary.
3434 bool css_has_online_children(struct cgroup_subsys_state
*css
)
3436 struct cgroup_subsys_state
*child
;
3440 css_for_each_child(child
, css
) {
3441 if (css
->flags
& CSS_ONLINE
) {
3451 * css_advance_task_iter - advance a task itererator to the next css_set
3452 * @it: the iterator to advance
3454 * Advance @it to the next css_set to walk.
3456 static void css_advance_task_iter(struct css_task_iter
*it
)
3458 struct list_head
*l
= it
->cset_pos
;
3459 struct cgrp_cset_link
*link
;
3460 struct css_set
*cset
;
3462 /* Advance to the next non-empty css_set */
3465 if (l
== it
->cset_head
) {
3466 it
->cset_pos
= NULL
;
3471 cset
= container_of(l
, struct css_set
,
3472 e_cset_node
[it
->ss
->id
]);
3474 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3477 } while (list_empty(&cset
->tasks
) && list_empty(&cset
->mg_tasks
));
3481 if (!list_empty(&cset
->tasks
))
3482 it
->task_pos
= cset
->tasks
.next
;
3484 it
->task_pos
= cset
->mg_tasks
.next
;
3486 it
->tasks_head
= &cset
->tasks
;
3487 it
->mg_tasks_head
= &cset
->mg_tasks
;
3491 * css_task_iter_start - initiate task iteration
3492 * @css: the css to walk tasks of
3493 * @it: the task iterator to use
3495 * Initiate iteration through the tasks of @css. The caller can call
3496 * css_task_iter_next() to walk through the tasks until the function
3497 * returns NULL. On completion of iteration, css_task_iter_end() must be
3500 * Note that this function acquires a lock which is released when the
3501 * iteration finishes. The caller can't sleep while iteration is in
3504 void css_task_iter_start(struct cgroup_subsys_state
*css
,
3505 struct css_task_iter
*it
)
3506 __acquires(css_set_rwsem
)
3508 /* no one should try to iterate before mounting cgroups */
3509 WARN_ON_ONCE(!use_task_css_set_links
);
3511 down_read(&css_set_rwsem
);
3516 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
3518 it
->cset_pos
= &css
->cgroup
->cset_links
;
3520 it
->cset_head
= it
->cset_pos
;
3522 css_advance_task_iter(it
);
3526 * css_task_iter_next - return the next task for the iterator
3527 * @it: the task iterator being iterated
3529 * The "next" function for task iteration. @it should have been
3530 * initialized via css_task_iter_start(). Returns NULL when the iteration
3533 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
3535 struct task_struct
*res
;
3536 struct list_head
*l
= it
->task_pos
;
3538 /* If the iterator cg is NULL, we have no tasks */
3541 res
= list_entry(l
, struct task_struct
, cg_list
);
3544 * Advance iterator to find next entry. cset->tasks is consumed
3545 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
3550 if (l
== it
->tasks_head
)
3551 l
= it
->mg_tasks_head
->next
;
3553 if (l
== it
->mg_tasks_head
)
3554 css_advance_task_iter(it
);
3562 * css_task_iter_end - finish task iteration
3563 * @it: the task iterator to finish
3565 * Finish task iteration started by css_task_iter_start().
3567 void css_task_iter_end(struct css_task_iter
*it
)
3568 __releases(css_set_rwsem
)
3570 up_read(&css_set_rwsem
);
3574 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3575 * @to: cgroup to which the tasks will be moved
3576 * @from: cgroup in which the tasks currently reside
3578 * Locking rules between cgroup_post_fork() and the migration path
3579 * guarantee that, if a task is forking while being migrated, the new child
3580 * is guaranteed to be either visible in the source cgroup after the
3581 * parent's migration is complete or put into the target cgroup. No task
3582 * can slip out of migration through forking.
3584 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3586 LIST_HEAD(preloaded_csets
);
3587 struct cgrp_cset_link
*link
;
3588 struct css_task_iter it
;
3589 struct task_struct
*task
;
3592 mutex_lock(&cgroup_mutex
);
3594 /* all tasks in @from are being moved, all csets are source */
3595 down_read(&css_set_rwsem
);
3596 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
3597 cgroup_migrate_add_src(link
->cset
, to
, &preloaded_csets
);
3598 up_read(&css_set_rwsem
);
3600 ret
= cgroup_migrate_prepare_dst(to
, &preloaded_csets
);
3605 * Migrate tasks one-by-one until @form is empty. This fails iff
3606 * ->can_attach() fails.
3609 css_task_iter_start(&from
->self
, &it
);
3610 task
= css_task_iter_next(&it
);
3612 get_task_struct(task
);
3613 css_task_iter_end(&it
);
3616 ret
= cgroup_migrate(to
, task
, false);
3617 put_task_struct(task
);
3619 } while (task
&& !ret
);
3621 cgroup_migrate_finish(&preloaded_csets
);
3622 mutex_unlock(&cgroup_mutex
);
3627 * Stuff for reading the 'tasks'/'procs' files.
3629 * Reading this file can return large amounts of data if a cgroup has
3630 * *lots* of attached tasks. So it may need several calls to read(),
3631 * but we cannot guarantee that the information we produce is correct
3632 * unless we produce it entirely atomically.
3636 /* which pidlist file are we talking about? */
3637 enum cgroup_filetype
{
3643 * A pidlist is a list of pids that virtually represents the contents of one
3644 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3645 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3648 struct cgroup_pidlist
{
3650 * used to find which pidlist is wanted. doesn't change as long as
3651 * this particular list stays in the list.
3653 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3656 /* how many elements the above list has */
3658 /* each of these stored in a list by its cgroup */
3659 struct list_head links
;
3660 /* pointer to the cgroup we belong to, for list removal purposes */
3661 struct cgroup
*owner
;
3662 /* for delayed destruction */
3663 struct delayed_work destroy_dwork
;
3667 * The following two functions "fix" the issue where there are more pids
3668 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3669 * TODO: replace with a kernel-wide solution to this problem
3671 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3672 static void *pidlist_allocate(int count
)
3674 if (PIDLIST_TOO_LARGE(count
))
3675 return vmalloc(count
* sizeof(pid_t
));
3677 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3680 static void pidlist_free(void *p
)
3682 if (is_vmalloc_addr(p
))
3689 * Used to destroy all pidlists lingering waiting for destroy timer. None
3690 * should be left afterwards.
3692 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
)
3694 struct cgroup_pidlist
*l
, *tmp_l
;
3696 mutex_lock(&cgrp
->pidlist_mutex
);
3697 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
3698 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
3699 mutex_unlock(&cgrp
->pidlist_mutex
);
3701 flush_workqueue(cgroup_pidlist_destroy_wq
);
3702 BUG_ON(!list_empty(&cgrp
->pidlists
));
3705 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
3707 struct delayed_work
*dwork
= to_delayed_work(work
);
3708 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
3710 struct cgroup_pidlist
*tofree
= NULL
;
3712 mutex_lock(&l
->owner
->pidlist_mutex
);
3715 * Destroy iff we didn't get queued again. The state won't change
3716 * as destroy_dwork can only be queued while locked.
3718 if (!delayed_work_pending(dwork
)) {
3719 list_del(&l
->links
);
3720 pidlist_free(l
->list
);
3721 put_pid_ns(l
->key
.ns
);
3725 mutex_unlock(&l
->owner
->pidlist_mutex
);
3730 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3731 * Returns the number of unique elements.
3733 static int pidlist_uniq(pid_t
*list
, int length
)
3738 * we presume the 0th element is unique, so i starts at 1. trivial
3739 * edge cases first; no work needs to be done for either
3741 if (length
== 0 || length
== 1)
3743 /* src and dest walk down the list; dest counts unique elements */
3744 for (src
= 1; src
< length
; src
++) {
3745 /* find next unique element */
3746 while (list
[src
] == list
[src
-1]) {
3751 /* dest always points to where the next unique element goes */
3752 list
[dest
] = list
[src
];
3760 * The two pid files - task and cgroup.procs - guaranteed that the result
3761 * is sorted, which forced this whole pidlist fiasco. As pid order is
3762 * different per namespace, each namespace needs differently sorted list,
3763 * making it impossible to use, for example, single rbtree of member tasks
3764 * sorted by task pointer. As pidlists can be fairly large, allocating one
3765 * per open file is dangerous, so cgroup had to implement shared pool of
3766 * pidlists keyed by cgroup and namespace.
3768 * All this extra complexity was caused by the original implementation
3769 * committing to an entirely unnecessary property. In the long term, we
3770 * want to do away with it. Explicitly scramble sort order if
3771 * sane_behavior so that no such expectation exists in the new interface.
3773 * Scrambling is done by swapping every two consecutive bits, which is
3774 * non-identity one-to-one mapping which disturbs sort order sufficiently.
3776 static pid_t
pid_fry(pid_t pid
)
3778 unsigned a
= pid
& 0x55555555;
3779 unsigned b
= pid
& 0xAAAAAAAA;
3781 return (a
<< 1) | (b
>> 1);
3784 static pid_t
cgroup_pid_fry(struct cgroup
*cgrp
, pid_t pid
)
3786 if (cgroup_sane_behavior(cgrp
))
3787 return pid_fry(pid
);
3792 static int cmppid(const void *a
, const void *b
)
3794 return *(pid_t
*)a
- *(pid_t
*)b
;
3797 static int fried_cmppid(const void *a
, const void *b
)
3799 return pid_fry(*(pid_t
*)a
) - pid_fry(*(pid_t
*)b
);
3802 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3803 enum cgroup_filetype type
)
3805 struct cgroup_pidlist
*l
;
3806 /* don't need task_nsproxy() if we're looking at ourself */
3807 struct pid_namespace
*ns
= task_active_pid_ns(current
);
3809 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3811 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
3812 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
3818 * find the appropriate pidlist for our purpose (given procs vs tasks)
3819 * returns with the lock on that pidlist already held, and takes care
3820 * of the use count, or returns NULL with no locks held if we're out of
3823 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
3824 enum cgroup_filetype type
)
3826 struct cgroup_pidlist
*l
;
3828 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3830 l
= cgroup_pidlist_find(cgrp
, type
);
3834 /* entry not found; create a new one */
3835 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3839 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
3841 /* don't need task_nsproxy() if we're looking at ourself */
3842 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
3844 list_add(&l
->links
, &cgrp
->pidlists
);
3849 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3851 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3852 struct cgroup_pidlist
**lp
)
3856 int pid
, n
= 0; /* used for populating the array */
3857 struct css_task_iter it
;
3858 struct task_struct
*tsk
;
3859 struct cgroup_pidlist
*l
;
3861 lockdep_assert_held(&cgrp
->pidlist_mutex
);
3864 * If cgroup gets more users after we read count, we won't have
3865 * enough space - tough. This race is indistinguishable to the
3866 * caller from the case that the additional cgroup users didn't
3867 * show up until sometime later on.
3869 length
= cgroup_task_count(cgrp
);
3870 array
= pidlist_allocate(length
);
3873 /* now, populate the array */
3874 css_task_iter_start(&cgrp
->self
, &it
);
3875 while ((tsk
= css_task_iter_next(&it
))) {
3876 if (unlikely(n
== length
))
3878 /* get tgid or pid for procs or tasks file respectively */
3879 if (type
== CGROUP_FILE_PROCS
)
3880 pid
= task_tgid_vnr(tsk
);
3882 pid
= task_pid_vnr(tsk
);
3883 if (pid
> 0) /* make sure to only use valid results */
3886 css_task_iter_end(&it
);
3888 /* now sort & (if procs) strip out duplicates */
3889 if (cgroup_sane_behavior(cgrp
))
3890 sort(array
, length
, sizeof(pid_t
), fried_cmppid
, NULL
);
3892 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3893 if (type
== CGROUP_FILE_PROCS
)
3894 length
= pidlist_uniq(array
, length
);
3896 l
= cgroup_pidlist_find_create(cgrp
, type
);
3898 mutex_unlock(&cgrp
->pidlist_mutex
);
3899 pidlist_free(array
);
3903 /* store array, freeing old if necessary */
3904 pidlist_free(l
->list
);
3912 * cgroupstats_build - build and fill cgroupstats
3913 * @stats: cgroupstats to fill information into
3914 * @dentry: A dentry entry belonging to the cgroup for which stats have
3917 * Build and fill cgroupstats so that taskstats can export it to user
3920 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3922 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
3923 struct cgroup
*cgrp
;
3924 struct css_task_iter it
;
3925 struct task_struct
*tsk
;
3927 /* it should be kernfs_node belonging to cgroupfs and is a directory */
3928 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
3929 kernfs_type(kn
) != KERNFS_DIR
)
3932 mutex_lock(&cgroup_mutex
);
3935 * We aren't being called from kernfs and there's no guarantee on
3936 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
3937 * @kn->priv is RCU safe. Let's do the RCU dancing.
3940 cgrp
= rcu_dereference(kn
->priv
);
3941 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
3943 mutex_unlock(&cgroup_mutex
);
3948 css_task_iter_start(&cgrp
->self
, &it
);
3949 while ((tsk
= css_task_iter_next(&it
))) {
3950 switch (tsk
->state
) {
3952 stats
->nr_running
++;
3954 case TASK_INTERRUPTIBLE
:
3955 stats
->nr_sleeping
++;
3957 case TASK_UNINTERRUPTIBLE
:
3958 stats
->nr_uninterruptible
++;
3961 stats
->nr_stopped
++;
3964 if (delayacct_is_task_waiting_on_io(tsk
))
3965 stats
->nr_io_wait
++;
3969 css_task_iter_end(&it
);
3971 mutex_unlock(&cgroup_mutex
);
3977 * seq_file methods for the tasks/procs files. The seq_file position is the
3978 * next pid to display; the seq_file iterator is a pointer to the pid
3979 * in the cgroup->l->list array.
3982 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3985 * Initially we receive a position value that corresponds to
3986 * one more than the last pid shown (or 0 on the first call or
3987 * after a seek to the start). Use a binary-search to find the
3988 * next pid to display, if any
3990 struct kernfs_open_file
*of
= s
->private;
3991 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
3992 struct cgroup_pidlist
*l
;
3993 enum cgroup_filetype type
= seq_cft(s
)->private;
3994 int index
= 0, pid
= *pos
;
3997 mutex_lock(&cgrp
->pidlist_mutex
);
4000 * !NULL @of->priv indicates that this isn't the first start()
4001 * after open. If the matching pidlist is around, we can use that.
4002 * Look for it. Note that @of->priv can't be used directly. It
4003 * could already have been destroyed.
4006 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
4009 * Either this is the first start() after open or the matching
4010 * pidlist has been destroyed inbetween. Create a new one.
4013 ret
= pidlist_array_load(cgrp
, type
,
4014 (struct cgroup_pidlist
**)&of
->priv
);
4016 return ERR_PTR(ret
);
4021 int end
= l
->length
;
4023 while (index
< end
) {
4024 int mid
= (index
+ end
) / 2;
4025 if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) == pid
) {
4028 } else if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) <= pid
)
4034 /* If we're off the end of the array, we're done */
4035 if (index
>= l
->length
)
4037 /* Update the abstract position to be the actual pid that we found */
4038 iter
= l
->list
+ index
;
4039 *pos
= cgroup_pid_fry(cgrp
, *iter
);
4043 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
4045 struct kernfs_open_file
*of
= s
->private;
4046 struct cgroup_pidlist
*l
= of
->priv
;
4049 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
4050 CGROUP_PIDLIST_DESTROY_DELAY
);
4051 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
4054 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4056 struct kernfs_open_file
*of
= s
->private;
4057 struct cgroup_pidlist
*l
= of
->priv
;
4059 pid_t
*end
= l
->list
+ l
->length
;
4061 * Advance to the next pid in the array. If this goes off the
4068 *pos
= cgroup_pid_fry(seq_css(s
)->cgroup
, *p
);
4073 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
4075 return seq_printf(s
, "%d\n", *(int *)v
);
4078 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
4081 return notify_on_release(css
->cgroup
);
4084 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
4085 struct cftype
*cft
, u64 val
)
4087 clear_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
4089 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4091 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4095 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
4098 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4101 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
4102 struct cftype
*cft
, u64 val
)
4105 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4107 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4111 static struct cftype cgroup_base_files
[] = {
4113 .name
= "cgroup.procs",
4114 .seq_start
= cgroup_pidlist_start
,
4115 .seq_next
= cgroup_pidlist_next
,
4116 .seq_stop
= cgroup_pidlist_stop
,
4117 .seq_show
= cgroup_pidlist_show
,
4118 .private = CGROUP_FILE_PROCS
,
4119 .write
= cgroup_procs_write
,
4120 .mode
= S_IRUGO
| S_IWUSR
,
4123 .name
= "cgroup.clone_children",
4124 .flags
= CFTYPE_INSANE
,
4125 .read_u64
= cgroup_clone_children_read
,
4126 .write_u64
= cgroup_clone_children_write
,
4129 .name
= "cgroup.sane_behavior",
4130 .flags
= CFTYPE_ONLY_ON_ROOT
,
4131 .seq_show
= cgroup_sane_behavior_show
,
4134 .name
= "cgroup.controllers",
4135 .flags
= CFTYPE_ONLY_ON_DFL
| CFTYPE_ONLY_ON_ROOT
,
4136 .seq_show
= cgroup_root_controllers_show
,
4139 .name
= "cgroup.controllers",
4140 .flags
= CFTYPE_ONLY_ON_DFL
| CFTYPE_NOT_ON_ROOT
,
4141 .seq_show
= cgroup_controllers_show
,
4144 .name
= "cgroup.subtree_control",
4145 .flags
= CFTYPE_ONLY_ON_DFL
,
4146 .seq_show
= cgroup_subtree_control_show
,
4147 .write
= cgroup_subtree_control_write
,
4150 .name
= "cgroup.populated",
4151 .flags
= CFTYPE_ONLY_ON_DFL
| CFTYPE_NOT_ON_ROOT
,
4152 .seq_show
= cgroup_populated_show
,
4156 * Historical crazy stuff. These don't have "cgroup." prefix and
4157 * don't exist if sane_behavior. If you're depending on these, be
4158 * prepared to be burned.
4162 .flags
= CFTYPE_INSANE
, /* use "procs" instead */
4163 .seq_start
= cgroup_pidlist_start
,
4164 .seq_next
= cgroup_pidlist_next
,
4165 .seq_stop
= cgroup_pidlist_stop
,
4166 .seq_show
= cgroup_pidlist_show
,
4167 .private = CGROUP_FILE_TASKS
,
4168 .write
= cgroup_tasks_write
,
4169 .mode
= S_IRUGO
| S_IWUSR
,
4172 .name
= "notify_on_release",
4173 .flags
= CFTYPE_INSANE
,
4174 .read_u64
= cgroup_read_notify_on_release
,
4175 .write_u64
= cgroup_write_notify_on_release
,
4178 .name
= "release_agent",
4179 .flags
= CFTYPE_INSANE
| CFTYPE_ONLY_ON_ROOT
,
4180 .seq_show
= cgroup_release_agent_show
,
4181 .write
= cgroup_release_agent_write
,
4182 .max_write_len
= PATH_MAX
- 1,
4188 * cgroup_populate_dir - create subsys files in a cgroup directory
4189 * @cgrp: target cgroup
4190 * @subsys_mask: mask of the subsystem ids whose files should be added
4192 * On failure, no file is added.
4194 static int cgroup_populate_dir(struct cgroup
*cgrp
, unsigned int subsys_mask
)
4196 struct cgroup_subsys
*ss
;
4199 /* process cftsets of each subsystem */
4200 for_each_subsys(ss
, i
) {
4201 struct cftype
*cfts
;
4203 if (!(subsys_mask
& (1 << i
)))
4206 list_for_each_entry(cfts
, &ss
->cfts
, node
) {
4207 ret
= cgroup_addrm_files(cgrp
, cfts
, true);
4214 cgroup_clear_dir(cgrp
, subsys_mask
);
4219 * css destruction is four-stage process.
4221 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4222 * Implemented in kill_css().
4224 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4225 * and thus css_tryget_online() is guaranteed to fail, the css can be
4226 * offlined by invoking offline_css(). After offlining, the base ref is
4227 * put. Implemented in css_killed_work_fn().
4229 * 3. When the percpu_ref reaches zero, the only possible remaining
4230 * accessors are inside RCU read sections. css_release() schedules the
4233 * 4. After the grace period, the css can be freed. Implemented in
4234 * css_free_work_fn().
4236 * It is actually hairier because both step 2 and 4 require process context
4237 * and thus involve punting to css->destroy_work adding two additional
4238 * steps to the already complex sequence.
4240 static void css_free_work_fn(struct work_struct
*work
)
4242 struct cgroup_subsys_state
*css
=
4243 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4244 struct cgroup
*cgrp
= css
->cgroup
;
4249 css_put(css
->parent
);
4251 css
->ss
->css_free(css
);
4254 /* cgroup free path */
4255 atomic_dec(&cgrp
->root
->nr_cgrps
);
4256 cgroup_pidlist_destroy_all(cgrp
);
4258 if (cgroup_parent(cgrp
)) {
4260 * We get a ref to the parent, and put the ref when
4261 * this cgroup is being freed, so it's guaranteed
4262 * that the parent won't be destroyed before its
4265 cgroup_put(cgroup_parent(cgrp
));
4266 kernfs_put(cgrp
->kn
);
4270 * This is root cgroup's refcnt reaching zero,
4271 * which indicates that the root should be
4274 cgroup_destroy_root(cgrp
->root
);
4279 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
4281 struct cgroup_subsys_state
*css
=
4282 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
4284 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
4285 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4288 static void css_release_work_fn(struct work_struct
*work
)
4290 struct cgroup_subsys_state
*css
=
4291 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4292 struct cgroup_subsys
*ss
= css
->ss
;
4293 struct cgroup
*cgrp
= css
->cgroup
;
4295 mutex_lock(&cgroup_mutex
);
4297 css
->flags
|= CSS_RELEASED
;
4298 list_del_rcu(&css
->sibling
);
4301 /* css release path */
4302 cgroup_idr_remove(&ss
->css_idr
, css
->id
);
4304 /* cgroup release path */
4305 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4309 mutex_unlock(&cgroup_mutex
);
4311 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4314 static void css_release(struct percpu_ref
*ref
)
4316 struct cgroup_subsys_state
*css
=
4317 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4319 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
4320 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4323 static void init_and_link_css(struct cgroup_subsys_state
*css
,
4324 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4326 lockdep_assert_held(&cgroup_mutex
);
4330 memset(css
, 0, sizeof(*css
));
4333 INIT_LIST_HEAD(&css
->sibling
);
4334 INIT_LIST_HEAD(&css
->children
);
4335 css
->serial_nr
= css_serial_nr_next
++;
4337 if (cgroup_parent(cgrp
)) {
4338 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
4339 css_get(css
->parent
);
4342 BUG_ON(cgroup_css(cgrp
, ss
));
4345 /* invoke ->css_online() on a new CSS and mark it online if successful */
4346 static int online_css(struct cgroup_subsys_state
*css
)
4348 struct cgroup_subsys
*ss
= css
->ss
;
4351 lockdep_assert_held(&cgroup_mutex
);
4354 ret
= ss
->css_online(css
);
4356 css
->flags
|= CSS_ONLINE
;
4357 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
4362 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4363 static void offline_css(struct cgroup_subsys_state
*css
)
4365 struct cgroup_subsys
*ss
= css
->ss
;
4367 lockdep_assert_held(&cgroup_mutex
);
4369 if (!(css
->flags
& CSS_ONLINE
))
4372 if (ss
->css_offline
)
4373 ss
->css_offline(css
);
4375 css
->flags
&= ~CSS_ONLINE
;
4376 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
4378 wake_up_all(&css
->cgroup
->offline_waitq
);
4382 * create_css - create a cgroup_subsys_state
4383 * @cgrp: the cgroup new css will be associated with
4384 * @ss: the subsys of new css
4385 * @visible: whether to create control knobs for the new css or not
4387 * Create a new css associated with @cgrp - @ss pair. On success, the new
4388 * css is online and installed in @cgrp with all interface files created if
4389 * @visible. Returns 0 on success, -errno on failure.
4391 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
4394 struct cgroup
*parent
= cgroup_parent(cgrp
);
4395 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
4396 struct cgroup_subsys_state
*css
;
4399 lockdep_assert_held(&cgroup_mutex
);
4401 css
= ss
->css_alloc(parent_css
);
4403 return PTR_ERR(css
);
4405 init_and_link_css(css
, ss
, cgrp
);
4407 err
= percpu_ref_init(&css
->refcnt
, css_release
);
4411 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_NOWAIT
);
4413 goto err_free_percpu_ref
;
4417 err
= cgroup_populate_dir(cgrp
, 1 << ss
->id
);
4422 /* @css is ready to be brought online now, make it visible */
4423 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
4424 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
4426 err
= online_css(css
);
4430 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4431 cgroup_parent(parent
)) {
4432 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4433 current
->comm
, current
->pid
, ss
->name
);
4434 if (!strcmp(ss
->name
, "memory"))
4435 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4436 ss
->warned_broken_hierarchy
= true;
4442 list_del_rcu(&css
->sibling
);
4443 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->id
);
4445 cgroup_idr_remove(&ss
->css_idr
, css
->id
);
4446 err_free_percpu_ref
:
4447 percpu_ref_cancel_init(&css
->refcnt
);
4449 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4453 static int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
,
4456 struct cgroup
*parent
, *cgrp
;
4457 struct cgroup_root
*root
;
4458 struct cgroup_subsys
*ss
;
4459 struct kernfs_node
*kn
;
4462 parent
= cgroup_kn_lock_live(parent_kn
);
4465 root
= parent
->root
;
4467 /* allocate the cgroup and its ID, 0 is reserved for the root */
4468 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4474 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
);
4479 * Temporarily set the pointer to NULL, so idr_find() won't return
4480 * a half-baked cgroup.
4482 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_NOWAIT
);
4485 goto out_cancel_ref
;
4488 init_cgroup_housekeeping(cgrp
);
4490 cgrp
->self
.parent
= &parent
->self
;
4493 if (notify_on_release(parent
))
4494 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4496 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4497 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4499 /* create the directory */
4500 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
4508 * This extra ref will be put in cgroup_free_fn() and guarantees
4509 * that @cgrp->kn is always accessible.
4513 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
4515 /* allocation complete, commit to creation */
4516 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
4517 atomic_inc(&root
->nr_cgrps
);
4521 * @cgrp is now fully operational. If something fails after this
4522 * point, it'll be released via the normal destruction path.
4524 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4526 ret
= cgroup_kn_set_ugid(kn
);
4530 ret
= cgroup_addrm_files(cgrp
, cgroup_base_files
, true);
4534 /* let's create and online css's */
4535 for_each_subsys(ss
, ssid
) {
4536 if (parent
->child_subsys_mask
& (1 << ssid
)) {
4537 ret
= create_css(cgrp
, ss
,
4538 parent
->subtree_control
& (1 << ssid
));
4545 * On the default hierarchy, a child doesn't automatically inherit
4546 * subtree_control from the parent. Each is configured manually.
4548 if (!cgroup_on_dfl(cgrp
)) {
4549 cgrp
->subtree_control
= parent
->subtree_control
;
4550 cgroup_refresh_child_subsys_mask(cgrp
);
4553 kernfs_activate(kn
);
4559 cgroup_idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4561 percpu_ref_cancel_init(&cgrp
->self
.refcnt
);
4565 cgroup_kn_unlock(parent_kn
);
4569 cgroup_destroy_locked(cgrp
);
4574 * This is called when the refcnt of a css is confirmed to be killed.
4575 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4576 * initate destruction and put the css ref from kill_css().
4578 static void css_killed_work_fn(struct work_struct
*work
)
4580 struct cgroup_subsys_state
*css
=
4581 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4583 mutex_lock(&cgroup_mutex
);
4585 mutex_unlock(&cgroup_mutex
);
4590 /* css kill confirmation processing requires process context, bounce */
4591 static void css_killed_ref_fn(struct percpu_ref
*ref
)
4593 struct cgroup_subsys_state
*css
=
4594 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4596 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
4597 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4601 * kill_css - destroy a css
4602 * @css: css to destroy
4604 * This function initiates destruction of @css by removing cgroup interface
4605 * files and putting its base reference. ->css_offline() will be invoked
4606 * asynchronously once css_tryget_online() is guaranteed to fail and when
4607 * the reference count reaches zero, @css will be released.
4609 static void kill_css(struct cgroup_subsys_state
*css
)
4611 lockdep_assert_held(&cgroup_mutex
);
4614 * This must happen before css is disassociated with its cgroup.
4615 * See seq_css() for details.
4617 cgroup_clear_dir(css
->cgroup
, 1 << css
->ss
->id
);
4620 * Killing would put the base ref, but we need to keep it alive
4621 * until after ->css_offline().
4626 * cgroup core guarantees that, by the time ->css_offline() is
4627 * invoked, no new css reference will be given out via
4628 * css_tryget_online(). We can't simply call percpu_ref_kill() and
4629 * proceed to offlining css's because percpu_ref_kill() doesn't
4630 * guarantee that the ref is seen as killed on all CPUs on return.
4632 * Use percpu_ref_kill_and_confirm() to get notifications as each
4633 * css is confirmed to be seen as killed on all CPUs.
4635 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
4639 * cgroup_destroy_locked - the first stage of cgroup destruction
4640 * @cgrp: cgroup to be destroyed
4642 * css's make use of percpu refcnts whose killing latency shouldn't be
4643 * exposed to userland and are RCU protected. Also, cgroup core needs to
4644 * guarantee that css_tryget_online() won't succeed by the time
4645 * ->css_offline() is invoked. To satisfy all the requirements,
4646 * destruction is implemented in the following two steps.
4648 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4649 * userland visible parts and start killing the percpu refcnts of
4650 * css's. Set up so that the next stage will be kicked off once all
4651 * the percpu refcnts are confirmed to be killed.
4653 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4654 * rest of destruction. Once all cgroup references are gone, the
4655 * cgroup is RCU-freed.
4657 * This function implements s1. After this step, @cgrp is gone as far as
4658 * the userland is concerned and a new cgroup with the same name may be
4659 * created. As cgroup doesn't care about the names internally, this
4660 * doesn't cause any problem.
4662 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
4663 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
4665 struct cgroup_subsys_state
*css
;
4669 lockdep_assert_held(&cgroup_mutex
);
4672 * css_set_rwsem synchronizes access to ->cset_links and prevents
4673 * @cgrp from being removed while put_css_set() is in progress.
4675 down_read(&css_set_rwsem
);
4676 empty
= list_empty(&cgrp
->cset_links
);
4677 up_read(&css_set_rwsem
);
4682 * Make sure there's no live children. We can't test emptiness of
4683 * ->self.children as dead children linger on it while being
4684 * drained; otherwise, "rmdir parent/child parent" may fail.
4686 if (css_has_online_children(&cgrp
->self
))
4690 * Mark @cgrp dead. This prevents further task migration and child
4691 * creation by disabling cgroup_lock_live_group().
4693 cgrp
->self
.flags
&= ~CSS_ONLINE
;
4695 /* initiate massacre of all css's */
4696 for_each_css(css
, ssid
, cgrp
)
4699 /* CSS_ONLINE is clear, remove from ->release_list for the last time */
4700 raw_spin_lock(&release_list_lock
);
4701 if (!list_empty(&cgrp
->release_list
))
4702 list_del_init(&cgrp
->release_list
);
4703 raw_spin_unlock(&release_list_lock
);
4706 * Remove @cgrp directory along with the base files. @cgrp has an
4707 * extra ref on its kn.
4709 kernfs_remove(cgrp
->kn
);
4711 set_bit(CGRP_RELEASABLE
, &cgroup_parent(cgrp
)->flags
);
4712 check_for_release(cgroup_parent(cgrp
));
4714 /* put the base reference */
4715 percpu_ref_kill(&cgrp
->self
.refcnt
);
4720 static int cgroup_rmdir(struct kernfs_node
*kn
)
4722 struct cgroup
*cgrp
;
4725 cgrp
= cgroup_kn_lock_live(kn
);
4728 cgroup_get(cgrp
); /* for @kn->priv clearing */
4730 ret
= cgroup_destroy_locked(cgrp
);
4732 cgroup_kn_unlock(kn
);
4735 * There are two control paths which try to determine cgroup from
4736 * dentry without going through kernfs - cgroupstats_build() and
4737 * css_tryget_online_from_dir(). Those are supported by RCU
4738 * protecting clearing of cgrp->kn->priv backpointer, which should
4739 * happen after all files under it have been removed.
4742 RCU_INIT_POINTER(*(void __rcu __force
**)&kn
->priv
, NULL
);
4748 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
4749 .remount_fs
= cgroup_remount
,
4750 .show_options
= cgroup_show_options
,
4751 .mkdir
= cgroup_mkdir
,
4752 .rmdir
= cgroup_rmdir
,
4753 .rename
= cgroup_rename
,
4756 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
4758 struct cgroup_subsys_state
*css
;
4760 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4762 mutex_lock(&cgroup_mutex
);
4764 idr_init(&ss
->css_idr
);
4765 INIT_LIST_HEAD(&ss
->cfts
);
4767 /* Create the root cgroup state for this subsystem */
4768 ss
->root
= &cgrp_dfl_root
;
4769 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
4770 /* We don't handle early failures gracefully */
4771 BUG_ON(IS_ERR(css
));
4772 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
4775 * Root csses are never destroyed and we can't initialize
4776 * percpu_ref during early init. Disable refcnting.
4778 css
->flags
|= CSS_NO_REF
;
4781 /* allocation can't be done safely during early init */
4784 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
4785 BUG_ON(css
->id
< 0);
4788 /* Update the init_css_set to contain a subsys
4789 * pointer to this state - since the subsystem is
4790 * newly registered, all tasks and hence the
4791 * init_css_set is in the subsystem's root cgroup. */
4792 init_css_set
.subsys
[ss
->id
] = css
;
4794 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4796 /* At system boot, before all subsystems have been
4797 * registered, no tasks have been forked, so we don't
4798 * need to invoke fork callbacks here. */
4799 BUG_ON(!list_empty(&init_task
.tasks
));
4801 BUG_ON(online_css(css
));
4803 mutex_unlock(&cgroup_mutex
);
4807 * cgroup_init_early - cgroup initialization at system boot
4809 * Initialize cgroups at system boot, and initialize any
4810 * subsystems that request early init.
4812 int __init
cgroup_init_early(void)
4814 static struct cgroup_sb_opts __initdata opts
=
4815 { .flags
= CGRP_ROOT_SANE_BEHAVIOR
};
4816 struct cgroup_subsys
*ss
;
4819 init_cgroup_root(&cgrp_dfl_root
, &opts
);
4820 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
4822 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
4824 for_each_subsys(ss
, i
) {
4825 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
4826 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
4827 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
4829 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
4830 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
4833 ss
->name
= cgroup_subsys_name
[i
];
4836 cgroup_init_subsys(ss
, true);
4842 * cgroup_init - cgroup initialization
4844 * Register cgroup filesystem and /proc file, and initialize
4845 * any subsystems that didn't request early init.
4847 int __init
cgroup_init(void)
4849 struct cgroup_subsys
*ss
;
4853 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_base_files
));
4855 mutex_lock(&cgroup_mutex
);
4857 /* Add init_css_set to the hash table */
4858 key
= css_set_hash(init_css_set
.subsys
);
4859 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
4861 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
4863 mutex_unlock(&cgroup_mutex
);
4865 for_each_subsys(ss
, ssid
) {
4866 if (ss
->early_init
) {
4867 struct cgroup_subsys_state
*css
=
4868 init_css_set
.subsys
[ss
->id
];
4870 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
4872 BUG_ON(css
->id
< 0);
4874 cgroup_init_subsys(ss
, false);
4877 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
4878 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
4881 * Setting dfl_root subsys_mask needs to consider the
4882 * disabled flag and cftype registration needs kmalloc,
4883 * both of which aren't available during early_init.
4885 if (!ss
->disabled
) {
4886 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
4887 WARN_ON(cgroup_add_cftypes(ss
, ss
->base_cftypes
));
4891 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4895 err
= register_filesystem(&cgroup_fs_type
);
4897 kobject_put(cgroup_kobj
);
4901 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4905 static int __init
cgroup_wq_init(void)
4908 * There isn't much point in executing destruction path in
4909 * parallel. Good chunk is serialized with cgroup_mutex anyway.
4910 * Use 1 for @max_active.
4912 * We would prefer to do this in cgroup_init() above, but that
4913 * is called before init_workqueues(): so leave this until after.
4915 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
4916 BUG_ON(!cgroup_destroy_wq
);
4919 * Used to destroy pidlists and separate to serve as flush domain.
4920 * Cap @max_active to 1 too.
4922 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
4924 BUG_ON(!cgroup_pidlist_destroy_wq
);
4928 core_initcall(cgroup_wq_init
);
4931 * proc_cgroup_show()
4932 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4933 * - Used for /proc/<pid>/cgroup.
4936 /* TODO: Use a proper seq_file iterator */
4937 int proc_cgroup_show(struct seq_file
*m
, void *v
)
4940 struct task_struct
*tsk
;
4943 struct cgroup_root
*root
;
4946 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
4952 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4958 mutex_lock(&cgroup_mutex
);
4959 down_read(&css_set_rwsem
);
4961 for_each_root(root
) {
4962 struct cgroup_subsys
*ss
;
4963 struct cgroup
*cgrp
;
4964 int ssid
, count
= 0;
4966 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_root_visible
)
4969 seq_printf(m
, "%d:", root
->hierarchy_id
);
4970 for_each_subsys(ss
, ssid
)
4971 if (root
->subsys_mask
& (1 << ssid
))
4972 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4973 if (strlen(root
->name
))
4974 seq_printf(m
, "%sname=%s", count
? "," : "",
4977 cgrp
= task_cgroup_from_root(tsk
, root
);
4978 path
= cgroup_path(cgrp
, buf
, PATH_MAX
);
4980 retval
= -ENAMETOOLONG
;
4988 up_read(&css_set_rwsem
);
4989 mutex_unlock(&cgroup_mutex
);
4990 put_task_struct(tsk
);
4997 /* Display information about each subsystem and each hierarchy */
4998 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5000 struct cgroup_subsys
*ss
;
5003 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5005 * ideally we don't want subsystems moving around while we do this.
5006 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5007 * subsys/hierarchy state.
5009 mutex_lock(&cgroup_mutex
);
5011 for_each_subsys(ss
, i
)
5012 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5013 ss
->name
, ss
->root
->hierarchy_id
,
5014 atomic_read(&ss
->root
->nr_cgrps
), !ss
->disabled
);
5016 mutex_unlock(&cgroup_mutex
);
5020 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5022 return single_open(file
, proc_cgroupstats_show
, NULL
);
5025 static const struct file_operations proc_cgroupstats_operations
= {
5026 .open
= cgroupstats_open
,
5028 .llseek
= seq_lseek
,
5029 .release
= single_release
,
5033 * cgroup_fork - initialize cgroup related fields during copy_process()
5034 * @child: pointer to task_struct of forking parent process.
5036 * A task is associated with the init_css_set until cgroup_post_fork()
5037 * attaches it to the parent's css_set. Empty cg_list indicates that
5038 * @child isn't holding reference to its css_set.
5040 void cgroup_fork(struct task_struct
*child
)
5042 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5043 INIT_LIST_HEAD(&child
->cg_list
);
5047 * cgroup_post_fork - called on a new task after adding it to the task list
5048 * @child: the task in question
5050 * Adds the task to the list running through its css_set if necessary and
5051 * call the subsystem fork() callbacks. Has to be after the task is
5052 * visible on the task list in case we race with the first call to
5053 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5056 void cgroup_post_fork(struct task_struct
*child
)
5058 struct cgroup_subsys
*ss
;
5062 * This may race against cgroup_enable_task_cg_links(). As that
5063 * function sets use_task_css_set_links before grabbing
5064 * tasklist_lock and we just went through tasklist_lock to add
5065 * @child, it's guaranteed that either we see the set
5066 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5067 * @child during its iteration.
5069 * If we won the race, @child is associated with %current's
5070 * css_set. Grabbing css_set_rwsem guarantees both that the
5071 * association is stable, and, on completion of the parent's
5072 * migration, @child is visible in the source of migration or
5073 * already in the destination cgroup. This guarantee is necessary
5074 * when implementing operations which need to migrate all tasks of
5075 * a cgroup to another.
5077 * Note that if we lose to cgroup_enable_task_cg_links(), @child
5078 * will remain in init_css_set. This is safe because all tasks are
5079 * in the init_css_set before cg_links is enabled and there's no
5080 * operation which transfers all tasks out of init_css_set.
5082 if (use_task_css_set_links
) {
5083 struct css_set
*cset
;
5085 down_write(&css_set_rwsem
);
5086 cset
= task_css_set(current
);
5087 if (list_empty(&child
->cg_list
)) {
5088 rcu_assign_pointer(child
->cgroups
, cset
);
5089 list_add(&child
->cg_list
, &cset
->tasks
);
5092 up_write(&css_set_rwsem
);
5096 * Call ss->fork(). This must happen after @child is linked on
5097 * css_set; otherwise, @child might change state between ->fork()
5098 * and addition to css_set.
5100 if (need_forkexit_callback
) {
5101 for_each_subsys(ss
, i
)
5108 * cgroup_exit - detach cgroup from exiting task
5109 * @tsk: pointer to task_struct of exiting process
5111 * Description: Detach cgroup from @tsk and release it.
5113 * Note that cgroups marked notify_on_release force every task in
5114 * them to take the global cgroup_mutex mutex when exiting.
5115 * This could impact scaling on very large systems. Be reluctant to
5116 * use notify_on_release cgroups where very high task exit scaling
5117 * is required on large systems.
5119 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5120 * call cgroup_exit() while the task is still competent to handle
5121 * notify_on_release(), then leave the task attached to the root cgroup in
5122 * each hierarchy for the remainder of its exit. No need to bother with
5123 * init_css_set refcnting. init_css_set never goes away and we can't race
5124 * with migration path - PF_EXITING is visible to migration path.
5126 void cgroup_exit(struct task_struct
*tsk
)
5128 struct cgroup_subsys
*ss
;
5129 struct css_set
*cset
;
5130 bool put_cset
= false;
5134 * Unlink from @tsk from its css_set. As migration path can't race
5135 * with us, we can check cg_list without grabbing css_set_rwsem.
5137 if (!list_empty(&tsk
->cg_list
)) {
5138 down_write(&css_set_rwsem
);
5139 list_del_init(&tsk
->cg_list
);
5140 up_write(&css_set_rwsem
);
5144 /* Reassign the task to the init_css_set. */
5145 cset
= task_css_set(tsk
);
5146 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5148 if (need_forkexit_callback
) {
5149 /* see cgroup_post_fork() for details */
5150 for_each_subsys(ss
, i
) {
5152 struct cgroup_subsys_state
*old_css
= cset
->subsys
[i
];
5153 struct cgroup_subsys_state
*css
= task_css(tsk
, i
);
5155 ss
->exit(css
, old_css
, tsk
);
5161 put_css_set(cset
, true);
5164 static void check_for_release(struct cgroup
*cgrp
)
5166 if (cgroup_is_releasable(cgrp
) && list_empty(&cgrp
->cset_links
) &&
5167 !css_has_online_children(&cgrp
->self
)) {
5169 * Control Group is currently removeable. If it's not
5170 * already queued for a userspace notification, queue
5173 int need_schedule_work
= 0;
5175 raw_spin_lock(&release_list_lock
);
5176 if (!cgroup_is_dead(cgrp
) &&
5177 list_empty(&cgrp
->release_list
)) {
5178 list_add(&cgrp
->release_list
, &release_list
);
5179 need_schedule_work
= 1;
5181 raw_spin_unlock(&release_list_lock
);
5182 if (need_schedule_work
)
5183 schedule_work(&release_agent_work
);
5188 * Notify userspace when a cgroup is released, by running the
5189 * configured release agent with the name of the cgroup (path
5190 * relative to the root of cgroup file system) as the argument.
5192 * Most likely, this user command will try to rmdir this cgroup.
5194 * This races with the possibility that some other task will be
5195 * attached to this cgroup before it is removed, or that some other
5196 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5197 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5198 * unused, and this cgroup will be reprieved from its death sentence,
5199 * to continue to serve a useful existence. Next time it's released,
5200 * we will get notified again, if it still has 'notify_on_release' set.
5202 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5203 * means only wait until the task is successfully execve()'d. The
5204 * separate release agent task is forked by call_usermodehelper(),
5205 * then control in this thread returns here, without waiting for the
5206 * release agent task. We don't bother to wait because the caller of
5207 * this routine has no use for the exit status of the release agent
5208 * task, so no sense holding our caller up for that.
5210 static void cgroup_release_agent(struct work_struct
*work
)
5212 BUG_ON(work
!= &release_agent_work
);
5213 mutex_lock(&cgroup_mutex
);
5214 raw_spin_lock(&release_list_lock
);
5215 while (!list_empty(&release_list
)) {
5216 char *argv
[3], *envp
[3];
5218 char *pathbuf
= NULL
, *agentbuf
= NULL
, *path
;
5219 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5222 list_del_init(&cgrp
->release_list
);
5223 raw_spin_unlock(&release_list_lock
);
5224 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5227 path
= cgroup_path(cgrp
, pathbuf
, PATH_MAX
);
5230 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5235 argv
[i
++] = agentbuf
;
5240 /* minimal command environment */
5241 envp
[i
++] = "HOME=/";
5242 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5245 /* Drop the lock while we invoke the usermode helper,
5246 * since the exec could involve hitting disk and hence
5247 * be a slow process */
5248 mutex_unlock(&cgroup_mutex
);
5249 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5250 mutex_lock(&cgroup_mutex
);
5254 raw_spin_lock(&release_list_lock
);
5256 raw_spin_unlock(&release_list_lock
);
5257 mutex_unlock(&cgroup_mutex
);
5260 static int __init
cgroup_disable(char *str
)
5262 struct cgroup_subsys
*ss
;
5266 while ((token
= strsep(&str
, ",")) != NULL
) {
5270 for_each_subsys(ss
, i
) {
5271 if (!strcmp(token
, ss
->name
)) {
5273 printk(KERN_INFO
"Disabling %s control group"
5274 " subsystem\n", ss
->name
);
5281 __setup("cgroup_disable=", cgroup_disable
);
5284 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5285 * @dentry: directory dentry of interest
5286 * @ss: subsystem of interest
5288 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5289 * to get the corresponding css and return it. If such css doesn't exist
5290 * or can't be pinned, an ERR_PTR value is returned.
5292 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
5293 struct cgroup_subsys
*ss
)
5295 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
5296 struct cgroup_subsys_state
*css
= NULL
;
5297 struct cgroup
*cgrp
;
5299 /* is @dentry a cgroup dir? */
5300 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
5301 kernfs_type(kn
) != KERNFS_DIR
)
5302 return ERR_PTR(-EBADF
);
5307 * This path doesn't originate from kernfs and @kn could already
5308 * have been or be removed at any point. @kn->priv is RCU
5309 * protected for this access. See cgroup_rmdir() for details.
5311 cgrp
= rcu_dereference(kn
->priv
);
5313 css
= cgroup_css(cgrp
, ss
);
5315 if (!css
|| !css_tryget_online(css
))
5316 css
= ERR_PTR(-ENOENT
);
5323 * css_from_id - lookup css by id
5324 * @id: the cgroup id
5325 * @ss: cgroup subsys to be looked into
5327 * Returns the css if there's valid one with @id, otherwise returns NULL.
5328 * Should be called under rcu_read_lock().
5330 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5332 WARN_ON_ONCE(!rcu_read_lock_held());
5333 return idr_find(&ss
->css_idr
, id
);
5336 #ifdef CONFIG_CGROUP_DEBUG
5337 static struct cgroup_subsys_state
*
5338 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
5340 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5343 return ERR_PTR(-ENOMEM
);
5348 static void debug_css_free(struct cgroup_subsys_state
*css
)
5353 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
5356 return cgroup_task_count(css
->cgroup
);
5359 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
5362 return (u64
)(unsigned long)current
->cgroups
;
5365 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
5371 count
= atomic_read(&task_css_set(current
)->refcount
);
5376 static int current_css_set_cg_links_read(struct seq_file
*seq
, void *v
)
5378 struct cgrp_cset_link
*link
;
5379 struct css_set
*cset
;
5382 name_buf
= kmalloc(NAME_MAX
+ 1, GFP_KERNEL
);
5386 down_read(&css_set_rwsem
);
5388 cset
= rcu_dereference(current
->cgroups
);
5389 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5390 struct cgroup
*c
= link
->cgrp
;
5392 cgroup_name(c
, name_buf
, NAME_MAX
+ 1);
5393 seq_printf(seq
, "Root %d group %s\n",
5394 c
->root
->hierarchy_id
, name_buf
);
5397 up_read(&css_set_rwsem
);
5402 #define MAX_TASKS_SHOWN_PER_CSS 25
5403 static int cgroup_css_links_read(struct seq_file
*seq
, void *v
)
5405 struct cgroup_subsys_state
*css
= seq_css(seq
);
5406 struct cgrp_cset_link
*link
;
5408 down_read(&css_set_rwsem
);
5409 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
5410 struct css_set
*cset
= link
->cset
;
5411 struct task_struct
*task
;
5414 seq_printf(seq
, "css_set %p\n", cset
);
5416 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5417 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5419 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5422 list_for_each_entry(task
, &cset
->mg_tasks
, cg_list
) {
5423 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5425 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5429 seq_puts(seq
, " ...\n");
5431 up_read(&css_set_rwsem
);
5435 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
5437 return test_bit(CGRP_RELEASABLE
, &css
->cgroup
->flags
);
5440 static struct cftype debug_files
[] = {
5442 .name
= "taskcount",
5443 .read_u64
= debug_taskcount_read
,
5447 .name
= "current_css_set",
5448 .read_u64
= current_css_set_read
,
5452 .name
= "current_css_set_refcount",
5453 .read_u64
= current_css_set_refcount_read
,
5457 .name
= "current_css_set_cg_links",
5458 .seq_show
= current_css_set_cg_links_read
,
5462 .name
= "cgroup_css_links",
5463 .seq_show
= cgroup_css_links_read
,
5467 .name
= "releasable",
5468 .read_u64
= releasable_read
,
5474 struct cgroup_subsys debug_cgrp_subsys
= {
5475 .css_alloc
= debug_css_alloc
,
5476 .css_free
= debug_css_free
,
5477 .base_cftypes
= debug_files
,
5479 #endif /* CONFIG_CGROUP_DEBUG */