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/percpu-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_lock 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 DEFINE_SPINLOCK(css_set_lock
);
87 EXPORT_SYMBOL_GPL(cgroup_mutex
);
88 EXPORT_SYMBOL_GPL(css_set_lock
);
90 static DEFINE_MUTEX(cgroup_mutex
);
91 static DEFINE_SPINLOCK(css_set_lock
);
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 struct percpu_rw_semaphore cgroup_threadgroup_rwsem
;
108 #define cgroup_assert_mutex_or_rcu_locked() \
109 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
110 !lockdep_is_held(&cgroup_mutex), \
111 "cgroup_mutex or RCU read lock required");
114 * cgroup destruction makes heavy use of work items and there can be a lot
115 * of concurrent destructions. Use a separate workqueue so that cgroup
116 * destruction work items don't end up filling up max_active of system_wq
117 * which may lead to deadlock.
119 static struct workqueue_struct
*cgroup_destroy_wq
;
122 * pidlist destructions need to be flushed on cgroup destruction. Use a
123 * separate workqueue as flush domain.
125 static struct workqueue_struct
*cgroup_pidlist_destroy_wq
;
127 /* generate an array of cgroup subsystem pointers */
128 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
129 static struct cgroup_subsys
*cgroup_subsys
[] = {
130 #include <linux/cgroup_subsys.h>
134 /* array of cgroup subsystem names */
135 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
136 static const char *cgroup_subsys_name
[] = {
137 #include <linux/cgroup_subsys.h>
141 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
143 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
144 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
145 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
146 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
147 #include <linux/cgroup_subsys.h>
150 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
151 static struct static_key_true
*cgroup_subsys_enabled_key
[] = {
152 #include <linux/cgroup_subsys.h>
156 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
157 static struct static_key_true
*cgroup_subsys_on_dfl_key
[] = {
158 #include <linux/cgroup_subsys.h>
163 * The default hierarchy, reserved for the subsystems that are otherwise
164 * unattached - it never has more than a single cgroup, and all tasks are
165 * part of that cgroup.
167 struct cgroup_root cgrp_dfl_root
;
168 EXPORT_SYMBOL_GPL(cgrp_dfl_root
);
171 * The default hierarchy always exists but is hidden until mounted for the
172 * first time. This is for backward compatibility.
174 static bool cgrp_dfl_root_visible
;
176 /* some controllers are not supported in the default hierarchy */
177 static unsigned long cgrp_dfl_root_inhibit_ss_mask
;
179 /* The list of hierarchy roots */
181 static LIST_HEAD(cgroup_roots
);
182 static int cgroup_root_count
;
184 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
185 static DEFINE_IDR(cgroup_hierarchy_idr
);
188 * Assign a monotonically increasing serial number to csses. It guarantees
189 * cgroups with bigger numbers are newer than those with smaller numbers.
190 * Also, as csses are always appended to the parent's ->children list, it
191 * guarantees that sibling csses are always sorted in the ascending serial
192 * number order on the list. Protected by cgroup_mutex.
194 static u64 css_serial_nr_next
= 1;
197 * These bitmask flags indicate whether tasks in the fork and exit paths have
198 * fork/exit handlers to call. This avoids us having to do extra work in the
199 * fork/exit path to check which subsystems have fork/exit callbacks.
201 static unsigned long have_fork_callback __read_mostly
;
202 static unsigned long have_exit_callback __read_mostly
;
203 static unsigned long have_free_callback __read_mostly
;
205 /* Ditto for the can_fork callback. */
206 static unsigned long have_canfork_callback __read_mostly
;
208 static struct cftype cgroup_dfl_base_files
[];
209 static struct cftype cgroup_legacy_base_files
[];
211 static int rebind_subsystems(struct cgroup_root
*dst_root
,
212 unsigned long ss_mask
);
213 static void css_task_iter_advance(struct css_task_iter
*it
);
214 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
215 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
217 static void css_release(struct percpu_ref
*ref
);
218 static void kill_css(struct cgroup_subsys_state
*css
);
219 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
220 struct cgroup
*cgrp
, struct cftype cfts
[],
224 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
225 * @ssid: subsys ID of interest
227 * cgroup_subsys_enabled() can only be used with literal subsys names which
228 * is fine for individual subsystems but unsuitable for cgroup core. This
229 * is slower static_key_enabled() based test indexed by @ssid.
231 static bool cgroup_ssid_enabled(int ssid
)
233 return static_key_enabled(cgroup_subsys_enabled_key
[ssid
]);
237 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
238 * @cgrp: the cgroup of interest
240 * The default hierarchy is the v2 interface of cgroup and this function
241 * can be used to test whether a cgroup is on the default hierarchy for
242 * cases where a subsystem should behave differnetly depending on the
245 * The set of behaviors which change on the default hierarchy are still
246 * being determined and the mount option is prefixed with __DEVEL__.
248 * List of changed behaviors:
250 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
251 * and "name" are disallowed.
253 * - When mounting an existing superblock, mount options should match.
255 * - Remount is disallowed.
257 * - rename(2) is disallowed.
259 * - "tasks" is removed. Everything should be at process granularity. Use
260 * "cgroup.procs" instead.
262 * - "cgroup.procs" is not sorted. pids will be unique unless they got
263 * recycled inbetween reads.
265 * - "release_agent" and "notify_on_release" are removed. Replacement
266 * notification mechanism will be implemented.
268 * - "cgroup.clone_children" is removed.
270 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
271 * and its descendants contain no task; otherwise, 1. The file also
272 * generates kernfs notification which can be monitored through poll and
273 * [di]notify when the value of the file changes.
275 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
276 * take masks of ancestors with non-empty cpus/mems, instead of being
277 * moved to an ancestor.
279 * - cpuset: a task can be moved into an empty cpuset, and again it takes
280 * masks of ancestors.
282 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
285 * - blkcg: blk-throttle becomes properly hierarchical.
287 * - debug: disallowed on the default hierarchy.
289 static bool cgroup_on_dfl(const struct cgroup
*cgrp
)
291 return cgrp
->root
== &cgrp_dfl_root
;
294 /* IDR wrappers which synchronize using cgroup_idr_lock */
295 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
300 idr_preload(gfp_mask
);
301 spin_lock_bh(&cgroup_idr_lock
);
302 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
& ~__GFP_DIRECT_RECLAIM
);
303 spin_unlock_bh(&cgroup_idr_lock
);
308 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
312 spin_lock_bh(&cgroup_idr_lock
);
313 ret
= idr_replace(idr
, ptr
, id
);
314 spin_unlock_bh(&cgroup_idr_lock
);
318 static void cgroup_idr_remove(struct idr
*idr
, int id
)
320 spin_lock_bh(&cgroup_idr_lock
);
322 spin_unlock_bh(&cgroup_idr_lock
);
325 static struct cgroup
*cgroup_parent(struct cgroup
*cgrp
)
327 struct cgroup_subsys_state
*parent_css
= cgrp
->self
.parent
;
330 return container_of(parent_css
, struct cgroup
, self
);
335 * cgroup_css - obtain a cgroup's css for the specified subsystem
336 * @cgrp: the cgroup of interest
337 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
339 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
340 * function must be called either under cgroup_mutex or rcu_read_lock() and
341 * the caller is responsible for pinning the returned css if it wants to
342 * keep accessing it outside the said locks. This function may return
343 * %NULL if @cgrp doesn't have @subsys_id enabled.
345 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
346 struct cgroup_subsys
*ss
)
349 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
350 lockdep_is_held(&cgroup_mutex
));
356 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
357 * @cgrp: the cgroup of interest
358 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
360 * Similar to cgroup_css() but returns the effective css, which is defined
361 * as the matching css of the nearest ancestor including self which has @ss
362 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
363 * function is guaranteed to return non-NULL css.
365 static struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
366 struct cgroup_subsys
*ss
)
368 lockdep_assert_held(&cgroup_mutex
);
373 if (!(cgrp
->root
->subsys_mask
& (1 << ss
->id
)))
377 * This function is used while updating css associations and thus
378 * can't test the csses directly. Use ->child_subsys_mask.
380 while (cgroup_parent(cgrp
) &&
381 !(cgroup_parent(cgrp
)->child_subsys_mask
& (1 << ss
->id
)))
382 cgrp
= cgroup_parent(cgrp
);
384 return cgroup_css(cgrp
, ss
);
388 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
389 * @cgrp: the cgroup of interest
390 * @ss: the subsystem of interest
392 * Find and get the effective css of @cgrp for @ss. The effective css is
393 * defined as the matching css of the nearest ancestor including self which
394 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
395 * the root css is returned, so this function always returns a valid css.
396 * The returned css must be put using css_put().
398 struct cgroup_subsys_state
*cgroup_get_e_css(struct cgroup
*cgrp
,
399 struct cgroup_subsys
*ss
)
401 struct cgroup_subsys_state
*css
;
406 css
= cgroup_css(cgrp
, ss
);
408 if (css
&& css_tryget_online(css
))
410 cgrp
= cgroup_parent(cgrp
);
413 css
= init_css_set
.subsys
[ss
->id
];
420 /* convenient tests for these bits */
421 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
423 return !(cgrp
->self
.flags
& CSS_ONLINE
);
426 static void cgroup_get(struct cgroup
*cgrp
)
428 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
429 css_get(&cgrp
->self
);
432 static bool cgroup_tryget(struct cgroup
*cgrp
)
434 return css_tryget(&cgrp
->self
);
437 static void cgroup_put(struct cgroup
*cgrp
)
439 css_put(&cgrp
->self
);
442 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
444 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
445 struct cftype
*cft
= of_cft(of
);
448 * This is open and unprotected implementation of cgroup_css().
449 * seq_css() is only called from a kernfs file operation which has
450 * an active reference on the file. Because all the subsystem
451 * files are drained before a css is disassociated with a cgroup,
452 * the matching css from the cgroup's subsys table is guaranteed to
453 * be and stay valid until the enclosing operation is complete.
456 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
460 EXPORT_SYMBOL_GPL(of_css
);
463 * cgroup_is_descendant - test ancestry
464 * @cgrp: the cgroup to be tested
465 * @ancestor: possible ancestor of @cgrp
467 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
468 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
469 * and @ancestor are accessible.
471 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
474 if (cgrp
== ancestor
)
476 cgrp
= cgroup_parent(cgrp
);
481 static int notify_on_release(const struct cgroup
*cgrp
)
483 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
487 * for_each_css - iterate all css's of a cgroup
488 * @css: the iteration cursor
489 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
490 * @cgrp: the target cgroup to iterate css's of
492 * Should be called under cgroup_[tree_]mutex.
494 #define for_each_css(css, ssid, cgrp) \
495 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
496 if (!((css) = rcu_dereference_check( \
497 (cgrp)->subsys[(ssid)], \
498 lockdep_is_held(&cgroup_mutex)))) { } \
502 * for_each_e_css - iterate all effective css's of a cgroup
503 * @css: the iteration cursor
504 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
505 * @cgrp: the target cgroup to iterate css's of
507 * Should be called under cgroup_[tree_]mutex.
509 #define for_each_e_css(css, ssid, cgrp) \
510 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
511 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
516 * for_each_subsys - iterate all enabled cgroup subsystems
517 * @ss: the iteration cursor
518 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
520 #define for_each_subsys(ss, ssid) \
521 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
522 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
525 * for_each_subsys_which - filter for_each_subsys with a bitmask
526 * @ss: the iteration cursor
527 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
528 * @ss_maskp: a pointer to the bitmask
530 * The block will only run for cases where the ssid-th bit (1 << ssid) of
533 #define for_each_subsys_which(ss, ssid, ss_maskp) \
534 if (!CGROUP_SUBSYS_COUNT) /* to avoid spurious gcc warning */ \
537 for_each_set_bit(ssid, ss_maskp, CGROUP_SUBSYS_COUNT) \
538 if (((ss) = cgroup_subsys[ssid]) && false) \
542 /* iterate across the hierarchies */
543 #define for_each_root(root) \
544 list_for_each_entry((root), &cgroup_roots, root_list)
546 /* iterate over child cgrps, lock should be held throughout iteration */
547 #define cgroup_for_each_live_child(child, cgrp) \
548 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
549 if (({ lockdep_assert_held(&cgroup_mutex); \
550 cgroup_is_dead(child); })) \
554 static void cgroup_release_agent(struct work_struct
*work
);
555 static void check_for_release(struct cgroup
*cgrp
);
558 * A cgroup can be associated with multiple css_sets as different tasks may
559 * belong to different cgroups on different hierarchies. In the other
560 * direction, a css_set is naturally associated with multiple cgroups.
561 * This M:N relationship is represented by the following link structure
562 * which exists for each association and allows traversing the associations
565 struct cgrp_cset_link
{
566 /* the cgroup and css_set this link associates */
568 struct css_set
*cset
;
570 /* list of cgrp_cset_links anchored at cgrp->cset_links */
571 struct list_head cset_link
;
573 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
574 struct list_head cgrp_link
;
578 * The default css_set - used by init and its children prior to any
579 * hierarchies being mounted. It contains a pointer to the root state
580 * for each subsystem. Also used to anchor the list of css_sets. Not
581 * reference-counted, to improve performance when child cgroups
582 * haven't been created.
584 struct css_set init_css_set
= {
585 .refcount
= ATOMIC_INIT(1),
586 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
587 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
588 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
589 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
590 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
591 .task_iters
= LIST_HEAD_INIT(init_css_set
.task_iters
),
594 static int css_set_count
= 1; /* 1 for init_css_set */
597 * css_set_populated - does a css_set contain any tasks?
598 * @cset: target css_set
600 static bool css_set_populated(struct css_set
*cset
)
602 lockdep_assert_held(&css_set_lock
);
604 return !list_empty(&cset
->tasks
) || !list_empty(&cset
->mg_tasks
);
608 * cgroup_update_populated - updated populated count of a cgroup
609 * @cgrp: the target cgroup
610 * @populated: inc or dec populated count
612 * One of the css_sets associated with @cgrp is either getting its first
613 * task or losing the last. Update @cgrp->populated_cnt accordingly. The
614 * count is propagated towards root so that a given cgroup's populated_cnt
615 * is zero iff the cgroup and all its descendants don't contain any tasks.
617 * @cgrp's interface file "cgroup.populated" is zero if
618 * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
619 * changes from or to zero, userland is notified that the content of the
620 * interface file has changed. This can be used to detect when @cgrp and
621 * its descendants become populated or empty.
623 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
625 lockdep_assert_held(&css_set_lock
);
631 trigger
= !cgrp
->populated_cnt
++;
633 trigger
= !--cgrp
->populated_cnt
;
638 check_for_release(cgrp
);
639 cgroup_file_notify(&cgrp
->events_file
);
641 cgrp
= cgroup_parent(cgrp
);
646 * css_set_update_populated - update populated state of a css_set
647 * @cset: target css_set
648 * @populated: whether @cset is populated or depopulated
650 * @cset is either getting the first task or losing the last. Update the
651 * ->populated_cnt of all associated cgroups accordingly.
653 static void css_set_update_populated(struct css_set
*cset
, bool populated
)
655 struct cgrp_cset_link
*link
;
657 lockdep_assert_held(&css_set_lock
);
659 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
)
660 cgroup_update_populated(link
->cgrp
, populated
);
664 * css_set_move_task - move a task from one css_set to another
665 * @task: task being moved
666 * @from_cset: css_set @task currently belongs to (may be NULL)
667 * @to_cset: new css_set @task is being moved to (may be NULL)
668 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
670 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
671 * css_set, @from_cset can be NULL. If @task is being disassociated
672 * instead of moved, @to_cset can be NULL.
674 * This function automatically handles populated_cnt updates and
675 * css_task_iter adjustments but the caller is responsible for managing
676 * @from_cset and @to_cset's reference counts.
678 static void css_set_move_task(struct task_struct
*task
,
679 struct css_set
*from_cset
, struct css_set
*to_cset
,
682 lockdep_assert_held(&css_set_lock
);
685 struct css_task_iter
*it
, *pos
;
687 WARN_ON_ONCE(list_empty(&task
->cg_list
));
690 * @task is leaving, advance task iterators which are
691 * pointing to it so that they can resume at the next
692 * position. Advancing an iterator might remove it from
693 * the list, use safe walk. See css_task_iter_advance*()
696 list_for_each_entry_safe(it
, pos
, &from_cset
->task_iters
,
698 if (it
->task_pos
== &task
->cg_list
)
699 css_task_iter_advance(it
);
701 list_del_init(&task
->cg_list
);
702 if (!css_set_populated(from_cset
))
703 css_set_update_populated(from_cset
, false);
705 WARN_ON_ONCE(!list_empty(&task
->cg_list
));
710 * We are synchronized through cgroup_threadgroup_rwsem
711 * against PF_EXITING setting such that we can't race
712 * against cgroup_exit() changing the css_set to
713 * init_css_set and dropping the old one.
715 WARN_ON_ONCE(task
->flags
& PF_EXITING
);
717 if (!css_set_populated(to_cset
))
718 css_set_update_populated(to_cset
, true);
719 rcu_assign_pointer(task
->cgroups
, to_cset
);
720 list_add_tail(&task
->cg_list
, use_mg_tasks
? &to_cset
->mg_tasks
:
726 * hash table for cgroup groups. This improves the performance to find
727 * an existing css_set. This hash doesn't (currently) take into
728 * account cgroups in empty hierarchies.
730 #define CSS_SET_HASH_BITS 7
731 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
733 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
735 unsigned long key
= 0UL;
736 struct cgroup_subsys
*ss
;
739 for_each_subsys(ss
, i
)
740 key
+= (unsigned long)css
[i
];
741 key
= (key
>> 16) ^ key
;
746 static void put_css_set_locked(struct css_set
*cset
)
748 struct cgrp_cset_link
*link
, *tmp_link
;
749 struct cgroup_subsys
*ss
;
752 lockdep_assert_held(&css_set_lock
);
754 if (!atomic_dec_and_test(&cset
->refcount
))
757 /* This css_set is dead. unlink it and release cgroup refcounts */
758 for_each_subsys(ss
, ssid
)
759 list_del(&cset
->e_cset_node
[ssid
]);
760 hash_del(&cset
->hlist
);
763 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
764 list_del(&link
->cset_link
);
765 list_del(&link
->cgrp_link
);
766 if (cgroup_parent(link
->cgrp
))
767 cgroup_put(link
->cgrp
);
771 kfree_rcu(cset
, rcu_head
);
774 static void put_css_set(struct css_set
*cset
)
777 * Ensure that the refcount doesn't hit zero while any readers
778 * can see it. Similar to atomic_dec_and_lock(), but for an
781 if (atomic_add_unless(&cset
->refcount
, -1, 1))
784 spin_lock_bh(&css_set_lock
);
785 put_css_set_locked(cset
);
786 spin_unlock_bh(&css_set_lock
);
790 * refcounted get/put for css_set objects
792 static inline void get_css_set(struct css_set
*cset
)
794 atomic_inc(&cset
->refcount
);
798 * compare_css_sets - helper function for find_existing_css_set().
799 * @cset: candidate css_set being tested
800 * @old_cset: existing css_set for a task
801 * @new_cgrp: cgroup that's being entered by the task
802 * @template: desired set of css pointers in css_set (pre-calculated)
804 * Returns true if "cset" matches "old_cset" except for the hierarchy
805 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
807 static bool compare_css_sets(struct css_set
*cset
,
808 struct css_set
*old_cset
,
809 struct cgroup
*new_cgrp
,
810 struct cgroup_subsys_state
*template[])
812 struct list_head
*l1
, *l2
;
815 * On the default hierarchy, there can be csets which are
816 * associated with the same set of cgroups but different csses.
817 * Let's first ensure that csses match.
819 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
823 * Compare cgroup pointers in order to distinguish between
824 * different cgroups in hierarchies. As different cgroups may
825 * share the same effective css, this comparison is always
828 l1
= &cset
->cgrp_links
;
829 l2
= &old_cset
->cgrp_links
;
831 struct cgrp_cset_link
*link1
, *link2
;
832 struct cgroup
*cgrp1
, *cgrp2
;
836 /* See if we reached the end - both lists are equal length. */
837 if (l1
== &cset
->cgrp_links
) {
838 BUG_ON(l2
!= &old_cset
->cgrp_links
);
841 BUG_ON(l2
== &old_cset
->cgrp_links
);
843 /* Locate the cgroups associated with these links. */
844 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
845 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
848 /* Hierarchies should be linked in the same order. */
849 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
852 * If this hierarchy is the hierarchy of the cgroup
853 * that's changing, then we need to check that this
854 * css_set points to the new cgroup; if it's any other
855 * hierarchy, then this css_set should point to the
856 * same cgroup as the old css_set.
858 if (cgrp1
->root
== new_cgrp
->root
) {
859 if (cgrp1
!= new_cgrp
)
870 * find_existing_css_set - init css array and find the matching css_set
871 * @old_cset: the css_set that we're using before the cgroup transition
872 * @cgrp: the cgroup that we're moving into
873 * @template: out param for the new set of csses, should be clear on entry
875 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
877 struct cgroup_subsys_state
*template[])
879 struct cgroup_root
*root
= cgrp
->root
;
880 struct cgroup_subsys
*ss
;
881 struct css_set
*cset
;
886 * Build the set of subsystem state objects that we want to see in the
887 * new css_set. while subsystems can change globally, the entries here
888 * won't change, so no need for locking.
890 for_each_subsys(ss
, i
) {
891 if (root
->subsys_mask
& (1UL << i
)) {
893 * @ss is in this hierarchy, so we want the
894 * effective css from @cgrp.
896 template[i
] = cgroup_e_css(cgrp
, ss
);
899 * @ss is not in this hierarchy, so we don't want
902 template[i
] = old_cset
->subsys
[i
];
906 key
= css_set_hash(template);
907 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
908 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
911 /* This css_set matches what we need */
915 /* No existing cgroup group matched */
919 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
921 struct cgrp_cset_link
*link
, *tmp_link
;
923 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
924 list_del(&link
->cset_link
);
930 * allocate_cgrp_cset_links - allocate cgrp_cset_links
931 * @count: the number of links to allocate
932 * @tmp_links: list_head the allocated links are put on
934 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
935 * through ->cset_link. Returns 0 on success or -errno.
937 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
939 struct cgrp_cset_link
*link
;
942 INIT_LIST_HEAD(tmp_links
);
944 for (i
= 0; i
< count
; i
++) {
945 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
947 free_cgrp_cset_links(tmp_links
);
950 list_add(&link
->cset_link
, tmp_links
);
956 * link_css_set - a helper function to link a css_set to a cgroup
957 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
958 * @cset: the css_set to be linked
959 * @cgrp: the destination cgroup
961 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
964 struct cgrp_cset_link
*link
;
966 BUG_ON(list_empty(tmp_links
));
968 if (cgroup_on_dfl(cgrp
))
969 cset
->dfl_cgrp
= cgrp
;
971 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
976 * Always add links to the tail of the lists so that the lists are
977 * in choronological order.
979 list_move_tail(&link
->cset_link
, &cgrp
->cset_links
);
980 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
982 if (cgroup_parent(cgrp
))
987 * find_css_set - return a new css_set with one cgroup updated
988 * @old_cset: the baseline css_set
989 * @cgrp: the cgroup to be updated
991 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
992 * substituted into the appropriate hierarchy.
994 static struct css_set
*find_css_set(struct css_set
*old_cset
,
997 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
998 struct css_set
*cset
;
999 struct list_head tmp_links
;
1000 struct cgrp_cset_link
*link
;
1001 struct cgroup_subsys
*ss
;
1005 lockdep_assert_held(&cgroup_mutex
);
1007 /* First see if we already have a cgroup group that matches
1008 * the desired set */
1009 spin_lock_bh(&css_set_lock
);
1010 cset
= find_existing_css_set(old_cset
, cgrp
, template);
1013 spin_unlock_bh(&css_set_lock
);
1018 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
1022 /* Allocate all the cgrp_cset_link objects that we'll need */
1023 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
1028 atomic_set(&cset
->refcount
, 1);
1029 INIT_LIST_HEAD(&cset
->cgrp_links
);
1030 INIT_LIST_HEAD(&cset
->tasks
);
1031 INIT_LIST_HEAD(&cset
->mg_tasks
);
1032 INIT_LIST_HEAD(&cset
->mg_preload_node
);
1033 INIT_LIST_HEAD(&cset
->mg_node
);
1034 INIT_LIST_HEAD(&cset
->task_iters
);
1035 INIT_HLIST_NODE(&cset
->hlist
);
1037 /* Copy the set of subsystem state objects generated in
1038 * find_existing_css_set() */
1039 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
1041 spin_lock_bh(&css_set_lock
);
1042 /* Add reference counts and links from the new css_set. */
1043 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
1044 struct cgroup
*c
= link
->cgrp
;
1046 if (c
->root
== cgrp
->root
)
1048 link_css_set(&tmp_links
, cset
, c
);
1051 BUG_ON(!list_empty(&tmp_links
));
1055 /* Add @cset to the hash table */
1056 key
= css_set_hash(cset
->subsys
);
1057 hash_add(css_set_table
, &cset
->hlist
, key
);
1059 for_each_subsys(ss
, ssid
)
1060 list_add_tail(&cset
->e_cset_node
[ssid
],
1061 &cset
->subsys
[ssid
]->cgroup
->e_csets
[ssid
]);
1063 spin_unlock_bh(&css_set_lock
);
1068 static struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
1070 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
1072 return root_cgrp
->root
;
1075 static int cgroup_init_root_id(struct cgroup_root
*root
)
1079 lockdep_assert_held(&cgroup_mutex
);
1081 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
1085 root
->hierarchy_id
= id
;
1089 static void cgroup_exit_root_id(struct cgroup_root
*root
)
1091 lockdep_assert_held(&cgroup_mutex
);
1093 if (root
->hierarchy_id
) {
1094 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1095 root
->hierarchy_id
= 0;
1099 static void cgroup_free_root(struct cgroup_root
*root
)
1102 /* hierarchy ID should already have been released */
1103 WARN_ON_ONCE(root
->hierarchy_id
);
1105 idr_destroy(&root
->cgroup_idr
);
1110 static void cgroup_destroy_root(struct cgroup_root
*root
)
1112 struct cgroup
*cgrp
= &root
->cgrp
;
1113 struct cgrp_cset_link
*link
, *tmp_link
;
1115 mutex_lock(&cgroup_mutex
);
1117 BUG_ON(atomic_read(&root
->nr_cgrps
));
1118 BUG_ON(!list_empty(&cgrp
->self
.children
));
1120 /* Rebind all subsystems back to the default hierarchy */
1121 rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
);
1124 * Release all the links from cset_links to this hierarchy's
1127 spin_lock_bh(&css_set_lock
);
1129 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1130 list_del(&link
->cset_link
);
1131 list_del(&link
->cgrp_link
);
1135 spin_unlock_bh(&css_set_lock
);
1137 if (!list_empty(&root
->root_list
)) {
1138 list_del(&root
->root_list
);
1139 cgroup_root_count
--;
1142 cgroup_exit_root_id(root
);
1144 mutex_unlock(&cgroup_mutex
);
1146 kernfs_destroy_root(root
->kf_root
);
1147 cgroup_free_root(root
);
1150 /* look up cgroup associated with given css_set on the specified hierarchy */
1151 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
1152 struct cgroup_root
*root
)
1154 struct cgroup
*res
= NULL
;
1156 lockdep_assert_held(&cgroup_mutex
);
1157 lockdep_assert_held(&css_set_lock
);
1159 if (cset
== &init_css_set
) {
1162 struct cgrp_cset_link
*link
;
1164 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1165 struct cgroup
*c
= link
->cgrp
;
1167 if (c
->root
== root
) {
1179 * Return the cgroup for "task" from the given hierarchy. Must be
1180 * called with cgroup_mutex and css_set_lock held.
1182 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
1183 struct cgroup_root
*root
)
1186 * No need to lock the task - since we hold cgroup_mutex the
1187 * task can't change groups, so the only thing that can happen
1188 * is that it exits and its css is set back to init_css_set.
1190 return cset_cgroup_from_root(task_css_set(task
), root
);
1194 * A task must hold cgroup_mutex to modify cgroups.
1196 * Any task can increment and decrement the count field without lock.
1197 * So in general, code holding cgroup_mutex can't rely on the count
1198 * field not changing. However, if the count goes to zero, then only
1199 * cgroup_attach_task() can increment it again. Because a count of zero
1200 * means that no tasks are currently attached, therefore there is no
1201 * way a task attached to that cgroup can fork (the other way to
1202 * increment the count). So code holding cgroup_mutex can safely
1203 * assume that if the count is zero, it will stay zero. Similarly, if
1204 * a task holds cgroup_mutex on a cgroup with zero count, it
1205 * knows that the cgroup won't be removed, as cgroup_rmdir()
1208 * A cgroup can only be deleted if both its 'count' of using tasks
1209 * is zero, and its list of 'children' cgroups is empty. Since all
1210 * tasks in the system use _some_ cgroup, and since there is always at
1211 * least one task in the system (init, pid == 1), therefore, root cgroup
1212 * always has either children cgroups and/or using tasks. So we don't
1213 * need a special hack to ensure that root cgroup cannot be deleted.
1215 * P.S. One more locking exception. RCU is used to guard the
1216 * update of a tasks cgroup pointer by cgroup_attach_task()
1219 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
1220 static const struct file_operations proc_cgroupstats_operations
;
1222 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1225 struct cgroup_subsys
*ss
= cft
->ss
;
1227 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1228 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
))
1229 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s.%s",
1230 cgroup_on_dfl(cgrp
) ? ss
->name
: ss
->legacy_name
,
1233 strncpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1238 * cgroup_file_mode - deduce file mode of a control file
1239 * @cft: the control file in question
1241 * S_IRUGO for read, S_IWUSR for write.
1243 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1247 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1250 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
) {
1251 if (cft
->flags
& CFTYPE_WORLD_WRITABLE
)
1261 * cgroup_calc_child_subsys_mask - calculate child_subsys_mask
1262 * @cgrp: the target cgroup
1263 * @subtree_control: the new subtree_control mask to consider
1265 * On the default hierarchy, a subsystem may request other subsystems to be
1266 * enabled together through its ->depends_on mask. In such cases, more
1267 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1269 * This function calculates which subsystems need to be enabled if
1270 * @subtree_control is to be applied to @cgrp. The returned mask is always
1271 * a superset of @subtree_control and follows the usual hierarchy rules.
1273 static unsigned long cgroup_calc_child_subsys_mask(struct cgroup
*cgrp
,
1274 unsigned long subtree_control
)
1276 struct cgroup
*parent
= cgroup_parent(cgrp
);
1277 unsigned long cur_ss_mask
= subtree_control
;
1278 struct cgroup_subsys
*ss
;
1281 lockdep_assert_held(&cgroup_mutex
);
1283 if (!cgroup_on_dfl(cgrp
))
1287 unsigned long new_ss_mask
= cur_ss_mask
;
1289 for_each_subsys_which(ss
, ssid
, &cur_ss_mask
)
1290 new_ss_mask
|= ss
->depends_on
;
1293 * Mask out subsystems which aren't available. This can
1294 * happen only if some depended-upon subsystems were bound
1295 * to non-default hierarchies.
1298 new_ss_mask
&= parent
->child_subsys_mask
;
1300 new_ss_mask
&= cgrp
->root
->subsys_mask
;
1302 if (new_ss_mask
== cur_ss_mask
)
1304 cur_ss_mask
= new_ss_mask
;
1311 * cgroup_refresh_child_subsys_mask - update child_subsys_mask
1312 * @cgrp: the target cgroup
1314 * Update @cgrp->child_subsys_mask according to the current
1315 * @cgrp->subtree_control using cgroup_calc_child_subsys_mask().
1317 static void cgroup_refresh_child_subsys_mask(struct cgroup
*cgrp
)
1319 cgrp
->child_subsys_mask
=
1320 cgroup_calc_child_subsys_mask(cgrp
, cgrp
->subtree_control
);
1324 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1325 * @kn: the kernfs_node being serviced
1327 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1328 * the method finishes if locking succeeded. Note that once this function
1329 * returns the cgroup returned by cgroup_kn_lock_live() may become
1330 * inaccessible any time. If the caller intends to continue to access the
1331 * cgroup, it should pin it before invoking this function.
1333 static void cgroup_kn_unlock(struct kernfs_node
*kn
)
1335 struct cgroup
*cgrp
;
1337 if (kernfs_type(kn
) == KERNFS_DIR
)
1340 cgrp
= kn
->parent
->priv
;
1342 mutex_unlock(&cgroup_mutex
);
1344 kernfs_unbreak_active_protection(kn
);
1349 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1350 * @kn: the kernfs_node being serviced
1352 * This helper is to be used by a cgroup kernfs method currently servicing
1353 * @kn. It breaks the active protection, performs cgroup locking and
1354 * verifies that the associated cgroup is alive. Returns the cgroup if
1355 * alive; otherwise, %NULL. A successful return should be undone by a
1356 * matching cgroup_kn_unlock() invocation.
1358 * Any cgroup kernfs method implementation which requires locking the
1359 * associated cgroup should use this helper. It avoids nesting cgroup
1360 * locking under kernfs active protection and allows all kernfs operations
1361 * including self-removal.
1363 static struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
)
1365 struct cgroup
*cgrp
;
1367 if (kernfs_type(kn
) == KERNFS_DIR
)
1370 cgrp
= kn
->parent
->priv
;
1373 * We're gonna grab cgroup_mutex which nests outside kernfs
1374 * active_ref. cgroup liveliness check alone provides enough
1375 * protection against removal. Ensure @cgrp stays accessible and
1376 * break the active_ref protection.
1378 if (!cgroup_tryget(cgrp
))
1380 kernfs_break_active_protection(kn
);
1382 mutex_lock(&cgroup_mutex
);
1384 if (!cgroup_is_dead(cgrp
))
1387 cgroup_kn_unlock(kn
);
1391 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1393 char name
[CGROUP_FILE_NAME_MAX
];
1395 lockdep_assert_held(&cgroup_mutex
);
1396 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1400 * css_clear_dir - remove subsys files in a cgroup directory
1402 * @cgrp_override: specify if target cgroup is different from css->cgroup
1404 static void css_clear_dir(struct cgroup_subsys_state
*css
,
1405 struct cgroup
*cgrp_override
)
1407 struct cgroup
*cgrp
= cgrp_override
?: css
->cgroup
;
1408 struct cftype
*cfts
;
1410 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
)
1411 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1415 * css_populate_dir - create subsys files in a cgroup directory
1417 * @cgrp_overried: specify if target cgroup is different from css->cgroup
1419 * On failure, no file is added.
1421 static int css_populate_dir(struct cgroup_subsys_state
*css
,
1422 struct cgroup
*cgrp_override
)
1424 struct cgroup
*cgrp
= cgrp_override
?: css
->cgroup
;
1425 struct cftype
*cfts
, *failed_cfts
;
1429 if (cgroup_on_dfl(cgrp
))
1430 cfts
= cgroup_dfl_base_files
;
1432 cfts
= cgroup_legacy_base_files
;
1434 return cgroup_addrm_files(&cgrp
->self
, cgrp
, cfts
, true);
1437 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1438 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, true);
1446 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1447 if (cfts
== failed_cfts
)
1449 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1454 static int rebind_subsystems(struct cgroup_root
*dst_root
,
1455 unsigned long ss_mask
)
1457 struct cgroup
*dcgrp
= &dst_root
->cgrp
;
1458 struct cgroup_subsys
*ss
;
1459 unsigned long tmp_ss_mask
;
1462 lockdep_assert_held(&cgroup_mutex
);
1464 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
1465 /* if @ss has non-root csses attached to it, can't move */
1466 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)))
1469 /* can't move between two non-dummy roots either */
1470 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1474 /* skip creating root files on dfl_root for inhibited subsystems */
1475 tmp_ss_mask
= ss_mask
;
1476 if (dst_root
== &cgrp_dfl_root
)
1477 tmp_ss_mask
&= ~cgrp_dfl_root_inhibit_ss_mask
;
1479 for_each_subsys_which(ss
, ssid
, &tmp_ss_mask
) {
1480 struct cgroup
*scgrp
= &ss
->root
->cgrp
;
1483 ret
= css_populate_dir(cgroup_css(scgrp
, ss
), dcgrp
);
1488 * Rebinding back to the default root is not allowed to
1489 * fail. Using both default and non-default roots should
1490 * be rare. Moving subsystems back and forth even more so.
1491 * Just warn about it and continue.
1493 if (dst_root
== &cgrp_dfl_root
) {
1494 if (cgrp_dfl_root_visible
) {
1495 pr_warn("failed to create files (%d) while rebinding 0x%lx to default root\n",
1497 pr_warn("you may retry by moving them to a different hierarchy and unbinding\n");
1502 for_each_subsys_which(ss
, tssid
, &tmp_ss_mask
) {
1505 css_clear_dir(cgroup_css(scgrp
, ss
), dcgrp
);
1511 * Nothing can fail from this point on. Remove files for the
1512 * removed subsystems and rebind each subsystem.
1514 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
1515 struct cgroup_root
*src_root
= ss
->root
;
1516 struct cgroup
*scgrp
= &src_root
->cgrp
;
1517 struct cgroup_subsys_state
*css
= cgroup_css(scgrp
, ss
);
1518 struct css_set
*cset
;
1520 WARN_ON(!css
|| cgroup_css(dcgrp
, ss
));
1522 css_clear_dir(css
, NULL
);
1524 RCU_INIT_POINTER(scgrp
->subsys
[ssid
], NULL
);
1525 rcu_assign_pointer(dcgrp
->subsys
[ssid
], css
);
1526 ss
->root
= dst_root
;
1527 css
->cgroup
= dcgrp
;
1529 spin_lock_bh(&css_set_lock
);
1530 hash_for_each(css_set_table
, i
, cset
, hlist
)
1531 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1532 &dcgrp
->e_csets
[ss
->id
]);
1533 spin_unlock_bh(&css_set_lock
);
1535 src_root
->subsys_mask
&= ~(1 << ssid
);
1536 scgrp
->subtree_control
&= ~(1 << ssid
);
1537 cgroup_refresh_child_subsys_mask(scgrp
);
1539 /* default hierarchy doesn't enable controllers by default */
1540 dst_root
->subsys_mask
|= 1 << ssid
;
1541 if (dst_root
== &cgrp_dfl_root
) {
1542 static_branch_enable(cgroup_subsys_on_dfl_key
[ssid
]);
1544 dcgrp
->subtree_control
|= 1 << ssid
;
1545 cgroup_refresh_child_subsys_mask(dcgrp
);
1546 static_branch_disable(cgroup_subsys_on_dfl_key
[ssid
]);
1553 kernfs_activate(dcgrp
->kn
);
1557 static int cgroup_show_options(struct seq_file
*seq
,
1558 struct kernfs_root
*kf_root
)
1560 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1561 struct cgroup_subsys
*ss
;
1564 if (root
!= &cgrp_dfl_root
)
1565 for_each_subsys(ss
, ssid
)
1566 if (root
->subsys_mask
& (1 << ssid
))
1567 seq_show_option(seq
, ss
->legacy_name
, NULL
);
1568 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1569 seq_puts(seq
, ",noprefix");
1570 if (root
->flags
& CGRP_ROOT_XATTR
)
1571 seq_puts(seq
, ",xattr");
1573 spin_lock(&release_agent_path_lock
);
1574 if (strlen(root
->release_agent_path
))
1575 seq_show_option(seq
, "release_agent",
1576 root
->release_agent_path
);
1577 spin_unlock(&release_agent_path_lock
);
1579 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
1580 seq_puts(seq
, ",clone_children");
1581 if (strlen(root
->name
))
1582 seq_show_option(seq
, "name", root
->name
);
1586 struct cgroup_sb_opts
{
1587 unsigned long subsys_mask
;
1589 char *release_agent
;
1590 bool cpuset_clone_children
;
1592 /* User explicitly requested empty subsystem */
1596 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1598 char *token
, *o
= data
;
1599 bool all_ss
= false, one_ss
= false;
1600 unsigned long mask
= -1UL;
1601 struct cgroup_subsys
*ss
;
1605 #ifdef CONFIG_CPUSETS
1606 mask
= ~(1U << cpuset_cgrp_id
);
1609 memset(opts
, 0, sizeof(*opts
));
1611 while ((token
= strsep(&o
, ",")) != NULL
) {
1616 if (!strcmp(token
, "none")) {
1617 /* Explicitly have no subsystems */
1621 if (!strcmp(token
, "all")) {
1622 /* Mutually exclusive option 'all' + subsystem name */
1628 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1629 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1632 if (!strcmp(token
, "noprefix")) {
1633 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1636 if (!strcmp(token
, "clone_children")) {
1637 opts
->cpuset_clone_children
= true;
1640 if (!strcmp(token
, "xattr")) {
1641 opts
->flags
|= CGRP_ROOT_XATTR
;
1644 if (!strncmp(token
, "release_agent=", 14)) {
1645 /* Specifying two release agents is forbidden */
1646 if (opts
->release_agent
)
1648 opts
->release_agent
=
1649 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1650 if (!opts
->release_agent
)
1654 if (!strncmp(token
, "name=", 5)) {
1655 const char *name
= token
+ 5;
1656 /* Can't specify an empty name */
1659 /* Must match [\w.-]+ */
1660 for (i
= 0; i
< strlen(name
); i
++) {
1664 if ((c
== '.') || (c
== '-') || (c
== '_'))
1668 /* Specifying two names is forbidden */
1671 opts
->name
= kstrndup(name
,
1672 MAX_CGROUP_ROOT_NAMELEN
- 1,
1680 for_each_subsys(ss
, i
) {
1681 if (strcmp(token
, ss
->legacy_name
))
1683 if (!cgroup_ssid_enabled(i
))
1686 /* Mutually exclusive option 'all' + subsystem name */
1689 opts
->subsys_mask
|= (1 << i
);
1694 if (i
== CGROUP_SUBSYS_COUNT
)
1698 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1699 pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1701 pr_err("sane_behavior: no other mount options allowed\n");
1708 * If the 'all' option was specified select all the subsystems,
1709 * otherwise if 'none', 'name=' and a subsystem name options were
1710 * not specified, let's default to 'all'
1712 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1713 for_each_subsys(ss
, i
)
1714 if (cgroup_ssid_enabled(i
))
1715 opts
->subsys_mask
|= (1 << i
);
1718 * We either have to specify by name or by subsystems. (So all
1719 * empty hierarchies must have a name).
1721 if (!opts
->subsys_mask
&& !opts
->name
)
1725 * Option noprefix was introduced just for backward compatibility
1726 * with the old cpuset, so we allow noprefix only if mounting just
1727 * the cpuset subsystem.
1729 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1732 /* Can't specify "none" and some subsystems */
1733 if (opts
->subsys_mask
&& opts
->none
)
1739 static int cgroup_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1742 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1743 struct cgroup_sb_opts opts
;
1744 unsigned long added_mask
, removed_mask
;
1746 if (root
== &cgrp_dfl_root
) {
1747 pr_err("remount is not allowed\n");
1751 mutex_lock(&cgroup_mutex
);
1753 /* See what subsystems are wanted */
1754 ret
= parse_cgroupfs_options(data
, &opts
);
1758 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1759 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1760 task_tgid_nr(current
), current
->comm
);
1762 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1763 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1765 /* Don't allow flags or name to change at remount */
1766 if ((opts
.flags
^ root
->flags
) ||
1767 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1768 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1769 opts
.flags
, opts
.name
?: "", root
->flags
, root
->name
);
1774 /* remounting is not allowed for populated hierarchies */
1775 if (!list_empty(&root
->cgrp
.self
.children
)) {
1780 ret
= rebind_subsystems(root
, added_mask
);
1784 rebind_subsystems(&cgrp_dfl_root
, removed_mask
);
1786 if (opts
.release_agent
) {
1787 spin_lock(&release_agent_path_lock
);
1788 strcpy(root
->release_agent_path
, opts
.release_agent
);
1789 spin_unlock(&release_agent_path_lock
);
1792 kfree(opts
.release_agent
);
1794 mutex_unlock(&cgroup_mutex
);
1799 * To reduce the fork() overhead for systems that are not actually using
1800 * their cgroups capability, we don't maintain the lists running through
1801 * each css_set to its tasks until we see the list actually used - in other
1802 * words after the first mount.
1804 static bool use_task_css_set_links __read_mostly
;
1806 static void cgroup_enable_task_cg_lists(void)
1808 struct task_struct
*p
, *g
;
1810 spin_lock_bh(&css_set_lock
);
1812 if (use_task_css_set_links
)
1815 use_task_css_set_links
= true;
1818 * We need tasklist_lock because RCU is not safe against
1819 * while_each_thread(). Besides, a forking task that has passed
1820 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1821 * is not guaranteed to have its child immediately visible in the
1822 * tasklist if we walk through it with RCU.
1824 read_lock(&tasklist_lock
);
1825 do_each_thread(g
, p
) {
1826 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1827 task_css_set(p
) != &init_css_set
);
1830 * We should check if the process is exiting, otherwise
1831 * it will race with cgroup_exit() in that the list
1832 * entry won't be deleted though the process has exited.
1833 * Do it while holding siglock so that we don't end up
1834 * racing against cgroup_exit().
1836 spin_lock_irq(&p
->sighand
->siglock
);
1837 if (!(p
->flags
& PF_EXITING
)) {
1838 struct css_set
*cset
= task_css_set(p
);
1840 if (!css_set_populated(cset
))
1841 css_set_update_populated(cset
, true);
1842 list_add_tail(&p
->cg_list
, &cset
->tasks
);
1845 spin_unlock_irq(&p
->sighand
->siglock
);
1846 } while_each_thread(g
, p
);
1847 read_unlock(&tasklist_lock
);
1849 spin_unlock_bh(&css_set_lock
);
1852 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1854 struct cgroup_subsys
*ss
;
1857 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1858 INIT_LIST_HEAD(&cgrp
->self
.children
);
1859 INIT_LIST_HEAD(&cgrp
->self
.files
);
1860 INIT_LIST_HEAD(&cgrp
->cset_links
);
1861 INIT_LIST_HEAD(&cgrp
->pidlists
);
1862 mutex_init(&cgrp
->pidlist_mutex
);
1863 cgrp
->self
.cgroup
= cgrp
;
1864 cgrp
->self
.flags
|= CSS_ONLINE
;
1866 for_each_subsys(ss
, ssid
)
1867 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1869 init_waitqueue_head(&cgrp
->offline_waitq
);
1870 INIT_WORK(&cgrp
->release_agent_work
, cgroup_release_agent
);
1873 static void init_cgroup_root(struct cgroup_root
*root
,
1874 struct cgroup_sb_opts
*opts
)
1876 struct cgroup
*cgrp
= &root
->cgrp
;
1878 INIT_LIST_HEAD(&root
->root_list
);
1879 atomic_set(&root
->nr_cgrps
, 1);
1881 init_cgroup_housekeeping(cgrp
);
1882 idr_init(&root
->cgroup_idr
);
1884 root
->flags
= opts
->flags
;
1885 if (opts
->release_agent
)
1886 strcpy(root
->release_agent_path
, opts
->release_agent
);
1888 strcpy(root
->name
, opts
->name
);
1889 if (opts
->cpuset_clone_children
)
1890 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1893 static int cgroup_setup_root(struct cgroup_root
*root
, unsigned long ss_mask
)
1895 LIST_HEAD(tmp_links
);
1896 struct cgroup
*root_cgrp
= &root
->cgrp
;
1897 struct css_set
*cset
;
1900 lockdep_assert_held(&cgroup_mutex
);
1902 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_KERNEL
);
1905 root_cgrp
->id
= ret
;
1907 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
, 0,
1913 * We're accessing css_set_count without locking css_set_lock here,
1914 * but that's OK - it can only be increased by someone holding
1915 * cgroup_lock, and that's us. The worst that can happen is that we
1916 * have some link structures left over
1918 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1922 ret
= cgroup_init_root_id(root
);
1926 root
->kf_root
= kernfs_create_root(&cgroup_kf_syscall_ops
,
1927 KERNFS_ROOT_CREATE_DEACTIVATED
,
1929 if (IS_ERR(root
->kf_root
)) {
1930 ret
= PTR_ERR(root
->kf_root
);
1933 root_cgrp
->kn
= root
->kf_root
->kn
;
1935 ret
= css_populate_dir(&root_cgrp
->self
, NULL
);
1939 ret
= rebind_subsystems(root
, ss_mask
);
1944 * There must be no failure case after here, since rebinding takes
1945 * care of subsystems' refcounts, which are explicitly dropped in
1946 * the failure exit path.
1948 list_add(&root
->root_list
, &cgroup_roots
);
1949 cgroup_root_count
++;
1952 * Link the root cgroup in this hierarchy into all the css_set
1955 spin_lock_bh(&css_set_lock
);
1956 hash_for_each(css_set_table
, i
, cset
, hlist
) {
1957 link_css_set(&tmp_links
, cset
, root_cgrp
);
1958 if (css_set_populated(cset
))
1959 cgroup_update_populated(root_cgrp
, true);
1961 spin_unlock_bh(&css_set_lock
);
1963 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
1964 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
1966 kernfs_activate(root_cgrp
->kn
);
1971 kernfs_destroy_root(root
->kf_root
);
1972 root
->kf_root
= NULL
;
1974 cgroup_exit_root_id(root
);
1976 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
1978 free_cgrp_cset_links(&tmp_links
);
1982 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1983 int flags
, const char *unused_dev_name
,
1986 struct super_block
*pinned_sb
= NULL
;
1987 struct cgroup_subsys
*ss
;
1988 struct cgroup_root
*root
;
1989 struct cgroup_sb_opts opts
;
1990 struct dentry
*dentry
;
1996 * The first time anyone tries to mount a cgroup, enable the list
1997 * linking each css_set to its tasks and fix up all existing tasks.
1999 if (!use_task_css_set_links
)
2000 cgroup_enable_task_cg_lists();
2002 mutex_lock(&cgroup_mutex
);
2004 /* First find the desired set of subsystems */
2005 ret
= parse_cgroupfs_options(data
, &opts
);
2009 /* look for a matching existing root */
2010 if (opts
.flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
2011 cgrp_dfl_root_visible
= true;
2012 root
= &cgrp_dfl_root
;
2013 cgroup_get(&root
->cgrp
);
2019 * Destruction of cgroup root is asynchronous, so subsystems may
2020 * still be dying after the previous unmount. Let's drain the
2021 * dying subsystems. We just need to ensure that the ones
2022 * unmounted previously finish dying and don't care about new ones
2023 * starting. Testing ref liveliness is good enough.
2025 for_each_subsys(ss
, i
) {
2026 if (!(opts
.subsys_mask
& (1 << i
)) ||
2027 ss
->root
== &cgrp_dfl_root
)
2030 if (!percpu_ref_tryget_live(&ss
->root
->cgrp
.self
.refcnt
)) {
2031 mutex_unlock(&cgroup_mutex
);
2033 ret
= restart_syscall();
2036 cgroup_put(&ss
->root
->cgrp
);
2039 for_each_root(root
) {
2040 bool name_match
= false;
2042 if (root
== &cgrp_dfl_root
)
2046 * If we asked for a name then it must match. Also, if
2047 * name matches but sybsys_mask doesn't, we should fail.
2048 * Remember whether name matched.
2051 if (strcmp(opts
.name
, root
->name
))
2057 * If we asked for subsystems (or explicitly for no
2058 * subsystems) then they must match.
2060 if ((opts
.subsys_mask
|| opts
.none
) &&
2061 (opts
.subsys_mask
!= root
->subsys_mask
)) {
2068 if (root
->flags
^ opts
.flags
)
2069 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
2072 * We want to reuse @root whose lifetime is governed by its
2073 * ->cgrp. Let's check whether @root is alive and keep it
2074 * that way. As cgroup_kill_sb() can happen anytime, we
2075 * want to block it by pinning the sb so that @root doesn't
2076 * get killed before mount is complete.
2078 * With the sb pinned, tryget_live can reliably indicate
2079 * whether @root can be reused. If it's being killed,
2080 * drain it. We can use wait_queue for the wait but this
2081 * path is super cold. Let's just sleep a bit and retry.
2083 pinned_sb
= kernfs_pin_sb(root
->kf_root
, NULL
);
2084 if (IS_ERR(pinned_sb
) ||
2085 !percpu_ref_tryget_live(&root
->cgrp
.self
.refcnt
)) {
2086 mutex_unlock(&cgroup_mutex
);
2087 if (!IS_ERR_OR_NULL(pinned_sb
))
2088 deactivate_super(pinned_sb
);
2090 ret
= restart_syscall();
2099 * No such thing, create a new one. name= matching without subsys
2100 * specification is allowed for already existing hierarchies but we
2101 * can't create new one without subsys specification.
2103 if (!opts
.subsys_mask
&& !opts
.none
) {
2108 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
2114 init_cgroup_root(root
, &opts
);
2116 ret
= cgroup_setup_root(root
, opts
.subsys_mask
);
2118 cgroup_free_root(root
);
2121 mutex_unlock(&cgroup_mutex
);
2123 kfree(opts
.release_agent
);
2127 return ERR_PTR(ret
);
2129 dentry
= kernfs_mount(fs_type
, flags
, root
->kf_root
,
2130 CGROUP_SUPER_MAGIC
, &new_sb
);
2131 if (IS_ERR(dentry
) || !new_sb
)
2132 cgroup_put(&root
->cgrp
);
2135 * If @pinned_sb, we're reusing an existing root and holding an
2136 * extra ref on its sb. Mount is complete. Put the extra ref.
2140 deactivate_super(pinned_sb
);
2146 static void cgroup_kill_sb(struct super_block
*sb
)
2148 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
2149 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
2152 * If @root doesn't have any mounts or children, start killing it.
2153 * This prevents new mounts by disabling percpu_ref_tryget_live().
2154 * cgroup_mount() may wait for @root's release.
2156 * And don't kill the default root.
2158 if (!list_empty(&root
->cgrp
.self
.children
) ||
2159 root
== &cgrp_dfl_root
)
2160 cgroup_put(&root
->cgrp
);
2162 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
2167 static struct file_system_type cgroup_fs_type
= {
2169 .mount
= cgroup_mount
,
2170 .kill_sb
= cgroup_kill_sb
,
2174 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2175 * @task: target task
2176 * @buf: the buffer to write the path into
2177 * @buflen: the length of the buffer
2179 * Determine @task's cgroup on the first (the one with the lowest non-zero
2180 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2181 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2182 * cgroup controller callbacks.
2184 * Return value is the same as kernfs_path().
2186 char *task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
2188 struct cgroup_root
*root
;
2189 struct cgroup
*cgrp
;
2190 int hierarchy_id
= 1;
2193 mutex_lock(&cgroup_mutex
);
2194 spin_lock_bh(&css_set_lock
);
2196 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
2199 cgrp
= task_cgroup_from_root(task
, root
);
2200 path
= cgroup_path(cgrp
, buf
, buflen
);
2202 /* if no hierarchy exists, everyone is in "/" */
2203 if (strlcpy(buf
, "/", buflen
) < buflen
)
2207 spin_unlock_bh(&css_set_lock
);
2208 mutex_unlock(&cgroup_mutex
);
2211 EXPORT_SYMBOL_GPL(task_cgroup_path
);
2213 /* used to track tasks and other necessary states during migration */
2214 struct cgroup_taskset
{
2215 /* the src and dst cset list running through cset->mg_node */
2216 struct list_head src_csets
;
2217 struct list_head dst_csets
;
2220 * Fields for cgroup_taskset_*() iteration.
2222 * Before migration is committed, the target migration tasks are on
2223 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
2224 * the csets on ->dst_csets. ->csets point to either ->src_csets
2225 * or ->dst_csets depending on whether migration is committed.
2227 * ->cur_csets and ->cur_task point to the current task position
2230 struct list_head
*csets
;
2231 struct css_set
*cur_cset
;
2232 struct task_struct
*cur_task
;
2235 #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \
2236 .src_csets = LIST_HEAD_INIT(tset.src_csets), \
2237 .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \
2238 .csets = &tset.src_csets, \
2242 * cgroup_taskset_add - try to add a migration target task to a taskset
2243 * @task: target task
2244 * @tset: target taskset
2246 * Add @task, which is a migration target, to @tset. This function becomes
2247 * noop if @task doesn't need to be migrated. @task's css_set should have
2248 * been added as a migration source and @task->cg_list will be moved from
2249 * the css_set's tasks list to mg_tasks one.
2251 static void cgroup_taskset_add(struct task_struct
*task
,
2252 struct cgroup_taskset
*tset
)
2254 struct css_set
*cset
;
2256 lockdep_assert_held(&css_set_lock
);
2258 /* @task either already exited or can't exit until the end */
2259 if (task
->flags
& PF_EXITING
)
2262 /* leave @task alone if post_fork() hasn't linked it yet */
2263 if (list_empty(&task
->cg_list
))
2266 cset
= task_css_set(task
);
2267 if (!cset
->mg_src_cgrp
)
2270 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2271 if (list_empty(&cset
->mg_node
))
2272 list_add_tail(&cset
->mg_node
, &tset
->src_csets
);
2273 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2274 list_move_tail(&cset
->mg_dst_cset
->mg_node
,
2279 * cgroup_taskset_first - reset taskset and return the first task
2280 * @tset: taskset of interest
2282 * @tset iteration is initialized and the first task is returned.
2284 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
2286 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2287 tset
->cur_task
= NULL
;
2289 return cgroup_taskset_next(tset
);
2293 * cgroup_taskset_next - iterate to the next task in taskset
2294 * @tset: taskset of interest
2296 * Return the next task in @tset. Iteration must have been initialized
2297 * with cgroup_taskset_first().
2299 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
2301 struct css_set
*cset
= tset
->cur_cset
;
2302 struct task_struct
*task
= tset
->cur_task
;
2304 while (&cset
->mg_node
!= tset
->csets
) {
2306 task
= list_first_entry(&cset
->mg_tasks
,
2307 struct task_struct
, cg_list
);
2309 task
= list_next_entry(task
, cg_list
);
2311 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2312 tset
->cur_cset
= cset
;
2313 tset
->cur_task
= task
;
2317 cset
= list_next_entry(cset
, mg_node
);
2325 * cgroup_taskset_migrate - migrate a taskset to a cgroup
2326 * @tset: taget taskset
2327 * @dst_cgrp: destination cgroup
2329 * Migrate tasks in @tset to @dst_cgrp. This function fails iff one of the
2330 * ->can_attach callbacks fails and guarantees that either all or none of
2331 * the tasks in @tset are migrated. @tset is consumed regardless of
2334 static int cgroup_taskset_migrate(struct cgroup_taskset
*tset
,
2335 struct cgroup
*dst_cgrp
)
2337 struct cgroup_subsys_state
*css
, *failed_css
= NULL
;
2338 struct task_struct
*task
, *tmp_task
;
2339 struct css_set
*cset
, *tmp_cset
;
2342 /* methods shouldn't be called if no task is actually migrating */
2343 if (list_empty(&tset
->src_csets
))
2346 /* check that we can legitimately attach to the cgroup */
2347 for_each_e_css(css
, i
, dst_cgrp
) {
2348 if (css
->ss
->can_attach
) {
2349 ret
= css
->ss
->can_attach(css
, tset
);
2352 goto out_cancel_attach
;
2358 * Now that we're guaranteed success, proceed to move all tasks to
2359 * the new cgroup. There are no failure cases after here, so this
2360 * is the commit point.
2362 spin_lock_bh(&css_set_lock
);
2363 list_for_each_entry(cset
, &tset
->src_csets
, mg_node
) {
2364 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
) {
2365 struct css_set
*from_cset
= task_css_set(task
);
2366 struct css_set
*to_cset
= cset
->mg_dst_cset
;
2368 get_css_set(to_cset
);
2369 css_set_move_task(task
, from_cset
, to_cset
, true);
2370 put_css_set_locked(from_cset
);
2373 spin_unlock_bh(&css_set_lock
);
2376 * Migration is committed, all target tasks are now on dst_csets.
2377 * Nothing is sensitive to fork() after this point. Notify
2378 * controllers that migration is complete.
2380 tset
->csets
= &tset
->dst_csets
;
2382 for_each_e_css(css
, i
, dst_cgrp
)
2383 if (css
->ss
->attach
)
2384 css
->ss
->attach(css
, tset
);
2387 goto out_release_tset
;
2390 for_each_e_css(css
, i
, dst_cgrp
) {
2391 if (css
== failed_css
)
2393 if (css
->ss
->cancel_attach
)
2394 css
->ss
->cancel_attach(css
, tset
);
2397 spin_lock_bh(&css_set_lock
);
2398 list_splice_init(&tset
->dst_csets
, &tset
->src_csets
);
2399 list_for_each_entry_safe(cset
, tmp_cset
, &tset
->src_csets
, mg_node
) {
2400 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2401 list_del_init(&cset
->mg_node
);
2403 spin_unlock_bh(&css_set_lock
);
2408 * cgroup_migrate_finish - cleanup after attach
2409 * @preloaded_csets: list of preloaded css_sets
2411 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2412 * those functions for details.
2414 static void cgroup_migrate_finish(struct list_head
*preloaded_csets
)
2416 struct css_set
*cset
, *tmp_cset
;
2418 lockdep_assert_held(&cgroup_mutex
);
2420 spin_lock_bh(&css_set_lock
);
2421 list_for_each_entry_safe(cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2422 cset
->mg_src_cgrp
= NULL
;
2423 cset
->mg_dst_cset
= NULL
;
2424 list_del_init(&cset
->mg_preload_node
);
2425 put_css_set_locked(cset
);
2427 spin_unlock_bh(&css_set_lock
);
2431 * cgroup_migrate_add_src - add a migration source css_set
2432 * @src_cset: the source css_set to add
2433 * @dst_cgrp: the destination cgroup
2434 * @preloaded_csets: list of preloaded css_sets
2436 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2437 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2438 * up by cgroup_migrate_finish().
2440 * This function may be called without holding cgroup_threadgroup_rwsem
2441 * even if the target is a process. Threads may be created and destroyed
2442 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2443 * into play and the preloaded css_sets are guaranteed to cover all
2446 static void cgroup_migrate_add_src(struct css_set
*src_cset
,
2447 struct cgroup
*dst_cgrp
,
2448 struct list_head
*preloaded_csets
)
2450 struct cgroup
*src_cgrp
;
2452 lockdep_assert_held(&cgroup_mutex
);
2453 lockdep_assert_held(&css_set_lock
);
2455 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2457 if (!list_empty(&src_cset
->mg_preload_node
))
2460 WARN_ON(src_cset
->mg_src_cgrp
);
2461 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2462 WARN_ON(!list_empty(&src_cset
->mg_node
));
2464 src_cset
->mg_src_cgrp
= src_cgrp
;
2465 get_css_set(src_cset
);
2466 list_add(&src_cset
->mg_preload_node
, preloaded_csets
);
2470 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2471 * @dst_cgrp: the destination cgroup (may be %NULL)
2472 * @preloaded_csets: list of preloaded source css_sets
2474 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
2475 * have been preloaded to @preloaded_csets. This function looks up and
2476 * pins all destination css_sets, links each to its source, and append them
2477 * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each
2478 * source css_set is assumed to be its cgroup on the default hierarchy.
2480 * This function must be called after cgroup_migrate_add_src() has been
2481 * called on each migration source css_set. After migration is performed
2482 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2485 static int cgroup_migrate_prepare_dst(struct cgroup
*dst_cgrp
,
2486 struct list_head
*preloaded_csets
)
2489 struct css_set
*src_cset
, *tmp_cset
;
2491 lockdep_assert_held(&cgroup_mutex
);
2494 * Except for the root, child_subsys_mask must be zero for a cgroup
2495 * with tasks so that child cgroups don't compete against tasks.
2497 if (dst_cgrp
&& cgroup_on_dfl(dst_cgrp
) && cgroup_parent(dst_cgrp
) &&
2498 dst_cgrp
->child_subsys_mask
)
2501 /* look up the dst cset for each src cset and link it to src */
2502 list_for_each_entry_safe(src_cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2503 struct css_set
*dst_cset
;
2505 dst_cset
= find_css_set(src_cset
,
2506 dst_cgrp
?: src_cset
->dfl_cgrp
);
2510 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2513 * If src cset equals dst, it's noop. Drop the src.
2514 * cgroup_migrate() will skip the cset too. Note that we
2515 * can't handle src == dst as some nodes are used by both.
2517 if (src_cset
== dst_cset
) {
2518 src_cset
->mg_src_cgrp
= NULL
;
2519 list_del_init(&src_cset
->mg_preload_node
);
2520 put_css_set(src_cset
);
2521 put_css_set(dst_cset
);
2525 src_cset
->mg_dst_cset
= dst_cset
;
2527 if (list_empty(&dst_cset
->mg_preload_node
))
2528 list_add(&dst_cset
->mg_preload_node
, &csets
);
2530 put_css_set(dst_cset
);
2533 list_splice_tail(&csets
, preloaded_csets
);
2536 cgroup_migrate_finish(&csets
);
2541 * cgroup_migrate - migrate a process or task to a cgroup
2542 * @leader: the leader of the process or the task to migrate
2543 * @threadgroup: whether @leader points to the whole process or a single task
2544 * @cgrp: the destination cgroup
2546 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
2547 * process, the caller must be holding cgroup_threadgroup_rwsem. The
2548 * caller is also responsible for invoking cgroup_migrate_add_src() and
2549 * cgroup_migrate_prepare_dst() on the targets before invoking this
2550 * function and following up with cgroup_migrate_finish().
2552 * As long as a controller's ->can_attach() doesn't fail, this function is
2553 * guaranteed to succeed. This means that, excluding ->can_attach()
2554 * failure, when migrating multiple targets, the success or failure can be
2555 * decided for all targets by invoking group_migrate_prepare_dst() before
2556 * actually starting migrating.
2558 static int cgroup_migrate(struct task_struct
*leader
, bool threadgroup
,
2559 struct cgroup
*cgrp
)
2561 struct cgroup_taskset tset
= CGROUP_TASKSET_INIT(tset
);
2562 struct task_struct
*task
;
2565 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2566 * already PF_EXITING could be freed from underneath us unless we
2567 * take an rcu_read_lock.
2569 spin_lock_bh(&css_set_lock
);
2573 cgroup_taskset_add(task
, &tset
);
2576 } while_each_thread(leader
, task
);
2578 spin_unlock_bh(&css_set_lock
);
2580 return cgroup_taskset_migrate(&tset
, cgrp
);
2584 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2585 * @dst_cgrp: the cgroup to attach to
2586 * @leader: the task or the leader of the threadgroup to be attached
2587 * @threadgroup: attach the whole threadgroup?
2589 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2591 static int cgroup_attach_task(struct cgroup
*dst_cgrp
,
2592 struct task_struct
*leader
, bool threadgroup
)
2594 LIST_HEAD(preloaded_csets
);
2595 struct task_struct
*task
;
2598 /* look up all src csets */
2599 spin_lock_bh(&css_set_lock
);
2603 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
,
2607 } while_each_thread(leader
, task
);
2609 spin_unlock_bh(&css_set_lock
);
2611 /* prepare dst csets and commit */
2612 ret
= cgroup_migrate_prepare_dst(dst_cgrp
, &preloaded_csets
);
2614 ret
= cgroup_migrate(leader
, threadgroup
, dst_cgrp
);
2616 cgroup_migrate_finish(&preloaded_csets
);
2620 static int cgroup_procs_write_permission(struct task_struct
*task
,
2621 struct cgroup
*dst_cgrp
,
2622 struct kernfs_open_file
*of
)
2624 const struct cred
*cred
= current_cred();
2625 const struct cred
*tcred
= get_task_cred(task
);
2629 * even if we're attaching all tasks in the thread group, we only
2630 * need to check permissions on one of them.
2632 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2633 !uid_eq(cred
->euid
, tcred
->uid
) &&
2634 !uid_eq(cred
->euid
, tcred
->suid
))
2637 if (!ret
&& cgroup_on_dfl(dst_cgrp
)) {
2638 struct super_block
*sb
= of
->file
->f_path
.dentry
->d_sb
;
2639 struct cgroup
*cgrp
;
2640 struct inode
*inode
;
2642 spin_lock_bh(&css_set_lock
);
2643 cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
2644 spin_unlock_bh(&css_set_lock
);
2646 while (!cgroup_is_descendant(dst_cgrp
, cgrp
))
2647 cgrp
= cgroup_parent(cgrp
);
2650 inode
= kernfs_get_inode(sb
, cgrp
->procs_file
.kn
);
2652 ret
= inode_permission(inode
, MAY_WRITE
);
2662 * Find the task_struct of the task to attach by vpid and pass it along to the
2663 * function to attach either it or all tasks in its threadgroup. Will lock
2664 * cgroup_mutex and threadgroup.
2666 static ssize_t
__cgroup_procs_write(struct kernfs_open_file
*of
, char *buf
,
2667 size_t nbytes
, loff_t off
, bool threadgroup
)
2669 struct task_struct
*tsk
;
2670 struct cgroup
*cgrp
;
2674 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2677 cgrp
= cgroup_kn_lock_live(of
->kn
);
2681 percpu_down_write(&cgroup_threadgroup_rwsem
);
2684 tsk
= find_task_by_vpid(pid
);
2687 goto out_unlock_rcu
;
2694 tsk
= tsk
->group_leader
;
2697 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2698 * trapped in a cpuset, or RT worker may be born in a cgroup
2699 * with no rt_runtime allocated. Just say no.
2701 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2703 goto out_unlock_rcu
;
2706 get_task_struct(tsk
);
2709 ret
= cgroup_procs_write_permission(tsk
, cgrp
, of
);
2711 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2713 put_task_struct(tsk
);
2714 goto out_unlock_threadgroup
;
2718 out_unlock_threadgroup
:
2719 percpu_up_write(&cgroup_threadgroup_rwsem
);
2720 cgroup_kn_unlock(of
->kn
);
2721 return ret
?: nbytes
;
2725 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2726 * @from: attach to all cgroups of a given task
2727 * @tsk: the task to be attached
2729 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2731 struct cgroup_root
*root
;
2734 mutex_lock(&cgroup_mutex
);
2735 for_each_root(root
) {
2736 struct cgroup
*from_cgrp
;
2738 if (root
== &cgrp_dfl_root
)
2741 spin_lock_bh(&css_set_lock
);
2742 from_cgrp
= task_cgroup_from_root(from
, root
);
2743 spin_unlock_bh(&css_set_lock
);
2745 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2749 mutex_unlock(&cgroup_mutex
);
2753 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2755 static ssize_t
cgroup_tasks_write(struct kernfs_open_file
*of
,
2756 char *buf
, size_t nbytes
, loff_t off
)
2758 return __cgroup_procs_write(of
, buf
, nbytes
, off
, false);
2761 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
2762 char *buf
, size_t nbytes
, loff_t off
)
2764 return __cgroup_procs_write(of
, buf
, nbytes
, off
, true);
2767 static ssize_t
cgroup_release_agent_write(struct kernfs_open_file
*of
,
2768 char *buf
, size_t nbytes
, loff_t off
)
2770 struct cgroup
*cgrp
;
2772 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2774 cgrp
= cgroup_kn_lock_live(of
->kn
);
2777 spin_lock(&release_agent_path_lock
);
2778 strlcpy(cgrp
->root
->release_agent_path
, strstrip(buf
),
2779 sizeof(cgrp
->root
->release_agent_path
));
2780 spin_unlock(&release_agent_path_lock
);
2781 cgroup_kn_unlock(of
->kn
);
2785 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
2787 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2789 spin_lock(&release_agent_path_lock
);
2790 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2791 spin_unlock(&release_agent_path_lock
);
2792 seq_putc(seq
, '\n');
2796 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
2798 seq_puts(seq
, "0\n");
2802 static void cgroup_print_ss_mask(struct seq_file
*seq
, unsigned long ss_mask
)
2804 struct cgroup_subsys
*ss
;
2805 bool printed
= false;
2808 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
2811 seq_printf(seq
, "%s", ss
->name
);
2815 seq_putc(seq
, '\n');
2818 /* show controllers which are currently attached to the default hierarchy */
2819 static int cgroup_root_controllers_show(struct seq_file
*seq
, void *v
)
2821 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2823 cgroup_print_ss_mask(seq
, cgrp
->root
->subsys_mask
&
2824 ~cgrp_dfl_root_inhibit_ss_mask
);
2828 /* show controllers which are enabled from the parent */
2829 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2831 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2833 cgroup_print_ss_mask(seq
, cgroup_parent(cgrp
)->subtree_control
);
2837 /* show controllers which are enabled for a given cgroup's children */
2838 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2840 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2842 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2847 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2848 * @cgrp: root of the subtree to update csses for
2850 * @cgrp's child_subsys_mask has changed and its subtree's (self excluded)
2851 * css associations need to be updated accordingly. This function looks up
2852 * all css_sets which are attached to the subtree, creates the matching
2853 * updated css_sets and migrates the tasks to the new ones.
2855 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2857 LIST_HEAD(preloaded_csets
);
2858 struct cgroup_taskset tset
= CGROUP_TASKSET_INIT(tset
);
2859 struct cgroup_subsys_state
*css
;
2860 struct css_set
*src_cset
;
2863 lockdep_assert_held(&cgroup_mutex
);
2865 percpu_down_write(&cgroup_threadgroup_rwsem
);
2867 /* look up all csses currently attached to @cgrp's subtree */
2868 spin_lock_bh(&css_set_lock
);
2869 css_for_each_descendant_pre(css
, cgroup_css(cgrp
, NULL
)) {
2870 struct cgrp_cset_link
*link
;
2872 /* self is not affected by child_subsys_mask change */
2873 if (css
->cgroup
== cgrp
)
2876 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
)
2877 cgroup_migrate_add_src(link
->cset
, cgrp
,
2880 spin_unlock_bh(&css_set_lock
);
2882 /* NULL dst indicates self on default hierarchy */
2883 ret
= cgroup_migrate_prepare_dst(NULL
, &preloaded_csets
);
2887 spin_lock_bh(&css_set_lock
);
2888 list_for_each_entry(src_cset
, &preloaded_csets
, mg_preload_node
) {
2889 struct task_struct
*task
, *ntask
;
2891 /* src_csets precede dst_csets, break on the first dst_cset */
2892 if (!src_cset
->mg_src_cgrp
)
2895 /* all tasks in src_csets need to be migrated */
2896 list_for_each_entry_safe(task
, ntask
, &src_cset
->tasks
, cg_list
)
2897 cgroup_taskset_add(task
, &tset
);
2899 spin_unlock_bh(&css_set_lock
);
2901 ret
= cgroup_taskset_migrate(&tset
, cgrp
);
2903 cgroup_migrate_finish(&preloaded_csets
);
2904 percpu_up_write(&cgroup_threadgroup_rwsem
);
2908 /* change the enabled child controllers for a cgroup in the default hierarchy */
2909 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
2910 char *buf
, size_t nbytes
,
2913 unsigned long enable
= 0, disable
= 0;
2914 unsigned long css_enable
, css_disable
, old_sc
, new_sc
, old_ss
, new_ss
;
2915 struct cgroup
*cgrp
, *child
;
2916 struct cgroup_subsys
*ss
;
2921 * Parse input - space separated list of subsystem names prefixed
2922 * with either + or -.
2924 buf
= strstrip(buf
);
2925 while ((tok
= strsep(&buf
, " "))) {
2926 unsigned long tmp_ss_mask
= ~cgrp_dfl_root_inhibit_ss_mask
;
2930 for_each_subsys_which(ss
, ssid
, &tmp_ss_mask
) {
2931 if (!cgroup_ssid_enabled(ssid
) ||
2932 strcmp(tok
+ 1, ss
->name
))
2936 enable
|= 1 << ssid
;
2937 disable
&= ~(1 << ssid
);
2938 } else if (*tok
== '-') {
2939 disable
|= 1 << ssid
;
2940 enable
&= ~(1 << ssid
);
2946 if (ssid
== CGROUP_SUBSYS_COUNT
)
2950 cgrp
= cgroup_kn_lock_live(of
->kn
);
2954 for_each_subsys(ss
, ssid
) {
2955 if (enable
& (1 << ssid
)) {
2956 if (cgrp
->subtree_control
& (1 << ssid
)) {
2957 enable
&= ~(1 << ssid
);
2961 /* unavailable or not enabled on the parent? */
2962 if (!(cgrp_dfl_root
.subsys_mask
& (1 << ssid
)) ||
2963 (cgroup_parent(cgrp
) &&
2964 !(cgroup_parent(cgrp
)->subtree_control
& (1 << ssid
)))) {
2968 } else if (disable
& (1 << ssid
)) {
2969 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
2970 disable
&= ~(1 << ssid
);
2974 /* a child has it enabled? */
2975 cgroup_for_each_live_child(child
, cgrp
) {
2976 if (child
->subtree_control
& (1 << ssid
)) {
2984 if (!enable
&& !disable
) {
2990 * Except for the root, subtree_control must be zero for a cgroup
2991 * with tasks so that child cgroups don't compete against tasks.
2993 if (enable
&& cgroup_parent(cgrp
) && !list_empty(&cgrp
->cset_links
)) {
2999 * Update subsys masks and calculate what needs to be done. More
3000 * subsystems than specified may need to be enabled or disabled
3001 * depending on subsystem dependencies.
3003 old_sc
= cgrp
->subtree_control
;
3004 old_ss
= cgrp
->child_subsys_mask
;
3005 new_sc
= (old_sc
| enable
) & ~disable
;
3006 new_ss
= cgroup_calc_child_subsys_mask(cgrp
, new_sc
);
3008 css_enable
= ~old_ss
& new_ss
;
3009 css_disable
= old_ss
& ~new_ss
;
3010 enable
|= css_enable
;
3011 disable
|= css_disable
;
3014 * Because css offlining is asynchronous, userland might try to
3015 * re-enable the same controller while the previous instance is
3016 * still around. In such cases, wait till it's gone using
3019 for_each_subsys_which(ss
, ssid
, &css_enable
) {
3020 cgroup_for_each_live_child(child
, cgrp
) {
3023 if (!cgroup_css(child
, ss
))
3027 prepare_to_wait(&child
->offline_waitq
, &wait
,
3028 TASK_UNINTERRUPTIBLE
);
3029 cgroup_kn_unlock(of
->kn
);
3031 finish_wait(&child
->offline_waitq
, &wait
);
3034 return restart_syscall();
3038 cgrp
->subtree_control
= new_sc
;
3039 cgrp
->child_subsys_mask
= new_ss
;
3042 * Create new csses or make the existing ones visible. A css is
3043 * created invisible if it's being implicitly enabled through
3044 * dependency. An invisible css is made visible when the userland
3045 * explicitly enables it.
3047 for_each_subsys(ss
, ssid
) {
3048 if (!(enable
& (1 << ssid
)))
3051 cgroup_for_each_live_child(child
, cgrp
) {
3052 if (css_enable
& (1 << ssid
))
3053 ret
= create_css(child
, ss
,
3054 cgrp
->subtree_control
& (1 << ssid
));
3056 ret
= css_populate_dir(cgroup_css(child
, ss
),
3064 * At this point, cgroup_e_css() results reflect the new csses
3065 * making the following cgroup_update_dfl_csses() properly update
3066 * css associations of all tasks in the subtree.
3068 ret
= cgroup_update_dfl_csses(cgrp
);
3073 * All tasks are migrated out of disabled csses. Kill or hide
3074 * them. A css is hidden when the userland requests it to be
3075 * disabled while other subsystems are still depending on it. The
3076 * css must not actively control resources and be in the vanilla
3077 * state if it's made visible again later. Controllers which may
3078 * be depended upon should provide ->css_reset() for this purpose.
3080 for_each_subsys(ss
, ssid
) {
3081 if (!(disable
& (1 << ssid
)))
3084 cgroup_for_each_live_child(child
, cgrp
) {
3085 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
3087 if (css_disable
& (1 << ssid
)) {
3090 css_clear_dir(css
, NULL
);
3098 * The effective csses of all the descendants (excluding @cgrp) may
3099 * have changed. Subsystems can optionally subscribe to this event
3100 * by implementing ->css_e_css_changed() which is invoked if any of
3101 * the effective csses seen from the css's cgroup may have changed.
3103 for_each_subsys(ss
, ssid
) {
3104 struct cgroup_subsys_state
*this_css
= cgroup_css(cgrp
, ss
);
3105 struct cgroup_subsys_state
*css
;
3107 if (!ss
->css_e_css_changed
|| !this_css
)
3110 css_for_each_descendant_pre(css
, this_css
)
3111 if (css
!= this_css
)
3112 ss
->css_e_css_changed(css
);
3115 kernfs_activate(cgrp
->kn
);
3118 cgroup_kn_unlock(of
->kn
);
3119 return ret
?: nbytes
;
3122 cgrp
->subtree_control
= old_sc
;
3123 cgrp
->child_subsys_mask
= old_ss
;
3125 for_each_subsys(ss
, ssid
) {
3126 if (!(enable
& (1 << ssid
)))
3129 cgroup_for_each_live_child(child
, cgrp
) {
3130 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
3135 if (css_enable
& (1 << ssid
))
3138 css_clear_dir(css
, NULL
);
3144 static int cgroup_events_show(struct seq_file
*seq
, void *v
)
3146 seq_printf(seq
, "populated %d\n",
3147 cgroup_is_populated(seq_css(seq
)->cgroup
));
3151 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
3152 size_t nbytes
, loff_t off
)
3154 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
3155 struct cftype
*cft
= of
->kn
->priv
;
3156 struct cgroup_subsys_state
*css
;
3160 return cft
->write(of
, buf
, nbytes
, off
);
3163 * kernfs guarantees that a file isn't deleted with operations in
3164 * flight, which means that the matching css is and stays alive and
3165 * doesn't need to be pinned. The RCU locking is not necessary
3166 * either. It's just for the convenience of using cgroup_css().
3169 css
= cgroup_css(cgrp
, cft
->ss
);
3172 if (cft
->write_u64
) {
3173 unsigned long long v
;
3174 ret
= kstrtoull(buf
, 0, &v
);
3176 ret
= cft
->write_u64(css
, cft
, v
);
3177 } else if (cft
->write_s64
) {
3179 ret
= kstrtoll(buf
, 0, &v
);
3181 ret
= cft
->write_s64(css
, cft
, v
);
3186 return ret
?: nbytes
;
3189 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
3191 return seq_cft(seq
)->seq_start(seq
, ppos
);
3194 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
3196 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
3199 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
3201 seq_cft(seq
)->seq_stop(seq
, v
);
3204 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
3206 struct cftype
*cft
= seq_cft(m
);
3207 struct cgroup_subsys_state
*css
= seq_css(m
);
3210 return cft
->seq_show(m
, arg
);
3213 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
3214 else if (cft
->read_s64
)
3215 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
3221 static struct kernfs_ops cgroup_kf_single_ops
= {
3222 .atomic_write_len
= PAGE_SIZE
,
3223 .write
= cgroup_file_write
,
3224 .seq_show
= cgroup_seqfile_show
,
3227 static struct kernfs_ops cgroup_kf_ops
= {
3228 .atomic_write_len
= PAGE_SIZE
,
3229 .write
= cgroup_file_write
,
3230 .seq_start
= cgroup_seqfile_start
,
3231 .seq_next
= cgroup_seqfile_next
,
3232 .seq_stop
= cgroup_seqfile_stop
,
3233 .seq_show
= cgroup_seqfile_show
,
3237 * cgroup_rename - Only allow simple rename of directories in place.
3239 static int cgroup_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
3240 const char *new_name_str
)
3242 struct cgroup
*cgrp
= kn
->priv
;
3245 if (kernfs_type(kn
) != KERNFS_DIR
)
3247 if (kn
->parent
!= new_parent
)
3251 * This isn't a proper migration and its usefulness is very
3252 * limited. Disallow on the default hierarchy.
3254 if (cgroup_on_dfl(cgrp
))
3258 * We're gonna grab cgroup_mutex which nests outside kernfs
3259 * active_ref. kernfs_rename() doesn't require active_ref
3260 * protection. Break them before grabbing cgroup_mutex.
3262 kernfs_break_active_protection(new_parent
);
3263 kernfs_break_active_protection(kn
);
3265 mutex_lock(&cgroup_mutex
);
3267 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
3269 mutex_unlock(&cgroup_mutex
);
3271 kernfs_unbreak_active_protection(kn
);
3272 kernfs_unbreak_active_protection(new_parent
);
3276 /* set uid and gid of cgroup dirs and files to that of the creator */
3277 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3279 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3280 .ia_uid
= current_fsuid(),
3281 .ia_gid
= current_fsgid(), };
3283 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3284 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3287 return kernfs_setattr(kn
, &iattr
);
3290 static int cgroup_add_file(struct cgroup_subsys_state
*css
, struct cgroup
*cgrp
,
3293 char name
[CGROUP_FILE_NAME_MAX
];
3294 struct kernfs_node
*kn
;
3295 struct lock_class_key
*key
= NULL
;
3298 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3299 key
= &cft
->lockdep_key
;
3301 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3302 cgroup_file_mode(cft
), 0, cft
->kf_ops
, cft
,
3307 ret
= cgroup_kn_set_ugid(kn
);
3313 if (cft
->file_offset
) {
3314 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
3318 list_add(&cfile
->node
, &css
->files
);
3325 * cgroup_addrm_files - add or remove files to a cgroup directory
3326 * @css: the target css
3327 * @cgrp: the target cgroup (usually css->cgroup)
3328 * @cfts: array of cftypes to be added
3329 * @is_add: whether to add or remove
3331 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3332 * For removals, this function never fails.
3334 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
3335 struct cgroup
*cgrp
, struct cftype cfts
[],
3338 struct cftype
*cft
, *cft_end
= NULL
;
3341 lockdep_assert_held(&cgroup_mutex
);
3344 for (cft
= cfts
; cft
!= cft_end
&& cft
->name
[0] != '\0'; cft
++) {
3345 /* does cft->flags tell us to skip this file on @cgrp? */
3346 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3348 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
3350 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3352 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3356 ret
= cgroup_add_file(css
, cgrp
, cft
);
3358 pr_warn("%s: failed to add %s, err=%d\n",
3359 __func__
, cft
->name
, ret
);
3365 cgroup_rm_file(cgrp
, cft
);
3371 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3374 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3375 struct cgroup
*root
= &ss
->root
->cgrp
;
3376 struct cgroup_subsys_state
*css
;
3379 lockdep_assert_held(&cgroup_mutex
);
3381 /* add/rm files for all cgroups created before */
3382 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3383 struct cgroup
*cgrp
= css
->cgroup
;
3385 if (cgroup_is_dead(cgrp
))
3388 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, is_add
);
3394 kernfs_activate(root
->kn
);
3398 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3402 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3403 /* free copy for custom atomic_write_len, see init_cftypes() */
3404 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3409 /* revert flags set by cgroup core while adding @cfts */
3410 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
3414 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3418 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3419 struct kernfs_ops
*kf_ops
;
3421 WARN_ON(cft
->ss
|| cft
->kf_ops
);
3424 kf_ops
= &cgroup_kf_ops
;
3426 kf_ops
= &cgroup_kf_single_ops
;
3429 * Ugh... if @cft wants a custom max_write_len, we need to
3430 * make a copy of kf_ops to set its atomic_write_len.
3432 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
3433 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
3435 cgroup_exit_cftypes(cfts
);
3438 kf_ops
->atomic_write_len
= cft
->max_write_len
;
3441 cft
->kf_ops
= kf_ops
;
3448 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
3450 lockdep_assert_held(&cgroup_mutex
);
3452 if (!cfts
|| !cfts
[0].ss
)
3455 list_del(&cfts
->node
);
3456 cgroup_apply_cftypes(cfts
, false);
3457 cgroup_exit_cftypes(cfts
);
3462 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3463 * @cfts: zero-length name terminated array of cftypes
3465 * Unregister @cfts. Files described by @cfts are removed from all
3466 * existing cgroups and all future cgroups won't have them either. This
3467 * function can be called anytime whether @cfts' subsys is attached or not.
3469 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3472 int cgroup_rm_cftypes(struct cftype
*cfts
)
3476 mutex_lock(&cgroup_mutex
);
3477 ret
= cgroup_rm_cftypes_locked(cfts
);
3478 mutex_unlock(&cgroup_mutex
);
3483 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3484 * @ss: target cgroup subsystem
3485 * @cfts: zero-length name terminated array of cftypes
3487 * Register @cfts to @ss. Files described by @cfts are created for all
3488 * existing cgroups to which @ss is attached and all future cgroups will
3489 * have them too. This function can be called anytime whether @ss is
3492 * Returns 0 on successful registration, -errno on failure. Note that this
3493 * function currently returns 0 as long as @cfts registration is successful
3494 * even if some file creation attempts on existing cgroups fail.
3496 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3500 if (!cgroup_ssid_enabled(ss
->id
))
3503 if (!cfts
|| cfts
[0].name
[0] == '\0')
3506 ret
= cgroup_init_cftypes(ss
, cfts
);
3510 mutex_lock(&cgroup_mutex
);
3512 list_add_tail(&cfts
->node
, &ss
->cfts
);
3513 ret
= cgroup_apply_cftypes(cfts
, true);
3515 cgroup_rm_cftypes_locked(cfts
);
3517 mutex_unlock(&cgroup_mutex
);
3522 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3523 * @ss: target cgroup subsystem
3524 * @cfts: zero-length name terminated array of cftypes
3526 * Similar to cgroup_add_cftypes() but the added files are only used for
3527 * the default hierarchy.
3529 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3533 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3534 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
3535 return cgroup_add_cftypes(ss
, cfts
);
3539 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3540 * @ss: target cgroup subsystem
3541 * @cfts: zero-length name terminated array of cftypes
3543 * Similar to cgroup_add_cftypes() but the added files are only used for
3544 * the legacy hierarchies.
3546 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3550 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3551 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
3552 return cgroup_add_cftypes(ss
, cfts
);
3556 * cgroup_task_count - count the number of tasks in a cgroup.
3557 * @cgrp: the cgroup in question
3559 * Return the number of tasks in the cgroup.
3561 static int cgroup_task_count(const struct cgroup
*cgrp
)
3564 struct cgrp_cset_link
*link
;
3566 spin_lock_bh(&css_set_lock
);
3567 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
3568 count
+= atomic_read(&link
->cset
->refcount
);
3569 spin_unlock_bh(&css_set_lock
);
3574 * css_next_child - find the next child of a given css
3575 * @pos: the current position (%NULL to initiate traversal)
3576 * @parent: css whose children to walk
3578 * This function returns the next child of @parent and should be called
3579 * under either cgroup_mutex or RCU read lock. The only requirement is
3580 * that @parent and @pos are accessible. The next sibling is guaranteed to
3581 * be returned regardless of their states.
3583 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3584 * css which finished ->css_online() is guaranteed to be visible in the
3585 * future iterations and will stay visible until the last reference is put.
3586 * A css which hasn't finished ->css_online() or already finished
3587 * ->css_offline() may show up during traversal. It's each subsystem's
3588 * responsibility to synchronize against on/offlining.
3590 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
3591 struct cgroup_subsys_state
*parent
)
3593 struct cgroup_subsys_state
*next
;
3595 cgroup_assert_mutex_or_rcu_locked();
3598 * @pos could already have been unlinked from the sibling list.
3599 * Once a cgroup is removed, its ->sibling.next is no longer
3600 * updated when its next sibling changes. CSS_RELEASED is set when
3601 * @pos is taken off list, at which time its next pointer is valid,
3602 * and, as releases are serialized, the one pointed to by the next
3603 * pointer is guaranteed to not have started release yet. This
3604 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3605 * critical section, the one pointed to by its next pointer is
3606 * guaranteed to not have finished its RCU grace period even if we
3607 * have dropped rcu_read_lock() inbetween iterations.
3609 * If @pos has CSS_RELEASED set, its next pointer can't be
3610 * dereferenced; however, as each css is given a monotonically
3611 * increasing unique serial number and always appended to the
3612 * sibling list, the next one can be found by walking the parent's
3613 * children until the first css with higher serial number than
3614 * @pos's. While this path can be slower, it happens iff iteration
3615 * races against release and the race window is very small.
3618 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
3619 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
3620 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
3622 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
3623 if (next
->serial_nr
> pos
->serial_nr
)
3628 * @next, if not pointing to the head, can be dereferenced and is
3631 if (&next
->sibling
!= &parent
->children
)
3637 * css_next_descendant_pre - find the next descendant for pre-order walk
3638 * @pos: the current position (%NULL to initiate traversal)
3639 * @root: css whose descendants to walk
3641 * To be used by css_for_each_descendant_pre(). Find the next descendant
3642 * to visit for pre-order traversal of @root's descendants. @root is
3643 * included in the iteration and the first node to be visited.
3645 * While this function requires cgroup_mutex or RCU read locking, it
3646 * doesn't require the whole traversal to be contained in a single critical
3647 * section. This function will return the correct next descendant as long
3648 * as both @pos and @root are accessible and @pos is a descendant of @root.
3650 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3651 * css which finished ->css_online() is guaranteed to be visible in the
3652 * future iterations and will stay visible until the last reference is put.
3653 * A css which hasn't finished ->css_online() or already finished
3654 * ->css_offline() may show up during traversal. It's each subsystem's
3655 * responsibility to synchronize against on/offlining.
3657 struct cgroup_subsys_state
*
3658 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3659 struct cgroup_subsys_state
*root
)
3661 struct cgroup_subsys_state
*next
;
3663 cgroup_assert_mutex_or_rcu_locked();
3665 /* if first iteration, visit @root */
3669 /* visit the first child if exists */
3670 next
= css_next_child(NULL
, pos
);
3674 /* no child, visit my or the closest ancestor's next sibling */
3675 while (pos
!= root
) {
3676 next
= css_next_child(pos
, pos
->parent
);
3686 * css_rightmost_descendant - return the rightmost descendant of a css
3687 * @pos: css of interest
3689 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3690 * is returned. This can be used during pre-order traversal to skip
3693 * While this function requires cgroup_mutex or RCU read locking, it
3694 * doesn't require the whole traversal to be contained in a single critical
3695 * section. This function will return the correct rightmost descendant as
3696 * long as @pos is accessible.
3698 struct cgroup_subsys_state
*
3699 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3701 struct cgroup_subsys_state
*last
, *tmp
;
3703 cgroup_assert_mutex_or_rcu_locked();
3707 /* ->prev isn't RCU safe, walk ->next till the end */
3709 css_for_each_child(tmp
, last
)
3716 static struct cgroup_subsys_state
*
3717 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3719 struct cgroup_subsys_state
*last
;
3723 pos
= css_next_child(NULL
, pos
);
3730 * css_next_descendant_post - find the next descendant for post-order walk
3731 * @pos: the current position (%NULL to initiate traversal)
3732 * @root: css whose descendants to walk
3734 * To be used by css_for_each_descendant_post(). Find the next descendant
3735 * to visit for post-order traversal of @root's descendants. @root is
3736 * included in the iteration and the last node to be visited.
3738 * While this function requires cgroup_mutex or RCU read locking, it
3739 * doesn't require the whole traversal to be contained in a single critical
3740 * section. This function will return the correct next descendant as long
3741 * as both @pos and @cgroup are accessible and @pos is a descendant of
3744 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3745 * css which finished ->css_online() is guaranteed to be visible in the
3746 * future iterations and will stay visible until the last reference is put.
3747 * A css which hasn't finished ->css_online() or already finished
3748 * ->css_offline() may show up during traversal. It's each subsystem's
3749 * responsibility to synchronize against on/offlining.
3751 struct cgroup_subsys_state
*
3752 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3753 struct cgroup_subsys_state
*root
)
3755 struct cgroup_subsys_state
*next
;
3757 cgroup_assert_mutex_or_rcu_locked();
3759 /* if first iteration, visit leftmost descendant which may be @root */
3761 return css_leftmost_descendant(root
);
3763 /* if we visited @root, we're done */
3767 /* if there's an unvisited sibling, visit its leftmost descendant */
3768 next
= css_next_child(pos
, pos
->parent
);
3770 return css_leftmost_descendant(next
);
3772 /* no sibling left, visit parent */
3777 * css_has_online_children - does a css have online children
3778 * @css: the target css
3780 * Returns %true if @css has any online children; otherwise, %false. This
3781 * function can be called from any context but the caller is responsible
3782 * for synchronizing against on/offlining as necessary.
3784 bool css_has_online_children(struct cgroup_subsys_state
*css
)
3786 struct cgroup_subsys_state
*child
;
3790 css_for_each_child(child
, css
) {
3791 if (child
->flags
& CSS_ONLINE
) {
3801 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
3802 * @it: the iterator to advance
3804 * Advance @it to the next css_set to walk.
3806 static void css_task_iter_advance_css_set(struct css_task_iter
*it
)
3808 struct list_head
*l
= it
->cset_pos
;
3809 struct cgrp_cset_link
*link
;
3810 struct css_set
*cset
;
3812 lockdep_assert_held(&css_set_lock
);
3814 /* Advance to the next non-empty css_set */
3817 if (l
== it
->cset_head
) {
3818 it
->cset_pos
= NULL
;
3819 it
->task_pos
= NULL
;
3824 cset
= container_of(l
, struct css_set
,
3825 e_cset_node
[it
->ss
->id
]);
3827 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3830 } while (!css_set_populated(cset
));
3834 if (!list_empty(&cset
->tasks
))
3835 it
->task_pos
= cset
->tasks
.next
;
3837 it
->task_pos
= cset
->mg_tasks
.next
;
3839 it
->tasks_head
= &cset
->tasks
;
3840 it
->mg_tasks_head
= &cset
->mg_tasks
;
3843 * We don't keep css_sets locked across iteration steps and thus
3844 * need to take steps to ensure that iteration can be resumed after
3845 * the lock is re-acquired. Iteration is performed at two levels -
3846 * css_sets and tasks in them.
3848 * Once created, a css_set never leaves its cgroup lists, so a
3849 * pinned css_set is guaranteed to stay put and we can resume
3850 * iteration afterwards.
3852 * Tasks may leave @cset across iteration steps. This is resolved
3853 * by registering each iterator with the css_set currently being
3854 * walked and making css_set_move_task() advance iterators whose
3855 * next task is leaving.
3858 list_del(&it
->iters_node
);
3859 put_css_set_locked(it
->cur_cset
);
3862 it
->cur_cset
= cset
;
3863 list_add(&it
->iters_node
, &cset
->task_iters
);
3866 static void css_task_iter_advance(struct css_task_iter
*it
)
3868 struct list_head
*l
= it
->task_pos
;
3870 lockdep_assert_held(&css_set_lock
);
3874 * Advance iterator to find next entry. cset->tasks is consumed
3875 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
3880 if (l
== it
->tasks_head
)
3881 l
= it
->mg_tasks_head
->next
;
3883 if (l
== it
->mg_tasks_head
)
3884 css_task_iter_advance_css_set(it
);
3890 * css_task_iter_start - initiate task iteration
3891 * @css: the css to walk tasks of
3892 * @it: the task iterator to use
3894 * Initiate iteration through the tasks of @css. The caller can call
3895 * css_task_iter_next() to walk through the tasks until the function
3896 * returns NULL. On completion of iteration, css_task_iter_end() must be
3899 void css_task_iter_start(struct cgroup_subsys_state
*css
,
3900 struct css_task_iter
*it
)
3902 /* no one should try to iterate before mounting cgroups */
3903 WARN_ON_ONCE(!use_task_css_set_links
);
3905 memset(it
, 0, sizeof(*it
));
3907 spin_lock_bh(&css_set_lock
);
3912 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
3914 it
->cset_pos
= &css
->cgroup
->cset_links
;
3916 it
->cset_head
= it
->cset_pos
;
3918 css_task_iter_advance_css_set(it
);
3920 spin_unlock_bh(&css_set_lock
);
3924 * css_task_iter_next - return the next task for the iterator
3925 * @it: the task iterator being iterated
3927 * The "next" function for task iteration. @it should have been
3928 * initialized via css_task_iter_start(). Returns NULL when the iteration
3931 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
3934 put_task_struct(it
->cur_task
);
3935 it
->cur_task
= NULL
;
3938 spin_lock_bh(&css_set_lock
);
3941 it
->cur_task
= list_entry(it
->task_pos
, struct task_struct
,
3943 get_task_struct(it
->cur_task
);
3944 css_task_iter_advance(it
);
3947 spin_unlock_bh(&css_set_lock
);
3949 return it
->cur_task
;
3953 * css_task_iter_end - finish task iteration
3954 * @it: the task iterator to finish
3956 * Finish task iteration started by css_task_iter_start().
3958 void css_task_iter_end(struct css_task_iter
*it
)
3961 spin_lock_bh(&css_set_lock
);
3962 list_del(&it
->iters_node
);
3963 put_css_set_locked(it
->cur_cset
);
3964 spin_unlock_bh(&css_set_lock
);
3968 put_task_struct(it
->cur_task
);
3972 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3973 * @to: cgroup to which the tasks will be moved
3974 * @from: cgroup in which the tasks currently reside
3976 * Locking rules between cgroup_post_fork() and the migration path
3977 * guarantee that, if a task is forking while being migrated, the new child
3978 * is guaranteed to be either visible in the source cgroup after the
3979 * parent's migration is complete or put into the target cgroup. No task
3980 * can slip out of migration through forking.
3982 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3984 LIST_HEAD(preloaded_csets
);
3985 struct cgrp_cset_link
*link
;
3986 struct css_task_iter it
;
3987 struct task_struct
*task
;
3990 mutex_lock(&cgroup_mutex
);
3992 /* all tasks in @from are being moved, all csets are source */
3993 spin_lock_bh(&css_set_lock
);
3994 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
3995 cgroup_migrate_add_src(link
->cset
, to
, &preloaded_csets
);
3996 spin_unlock_bh(&css_set_lock
);
3998 ret
= cgroup_migrate_prepare_dst(to
, &preloaded_csets
);
4003 * Migrate tasks one-by-one until @form is empty. This fails iff
4004 * ->can_attach() fails.
4007 css_task_iter_start(&from
->self
, &it
);
4008 task
= css_task_iter_next(&it
);
4010 get_task_struct(task
);
4011 css_task_iter_end(&it
);
4014 ret
= cgroup_migrate(task
, false, to
);
4015 put_task_struct(task
);
4017 } while (task
&& !ret
);
4019 cgroup_migrate_finish(&preloaded_csets
);
4020 mutex_unlock(&cgroup_mutex
);
4025 * Stuff for reading the 'tasks'/'procs' files.
4027 * Reading this file can return large amounts of data if a cgroup has
4028 * *lots* of attached tasks. So it may need several calls to read(),
4029 * but we cannot guarantee that the information we produce is correct
4030 * unless we produce it entirely atomically.
4034 /* which pidlist file are we talking about? */
4035 enum cgroup_filetype
{
4041 * A pidlist is a list of pids that virtually represents the contents of one
4042 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
4043 * a pair (one each for procs, tasks) for each pid namespace that's relevant
4046 struct cgroup_pidlist
{
4048 * used to find which pidlist is wanted. doesn't change as long as
4049 * this particular list stays in the list.
4051 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
4054 /* how many elements the above list has */
4056 /* each of these stored in a list by its cgroup */
4057 struct list_head links
;
4058 /* pointer to the cgroup we belong to, for list removal purposes */
4059 struct cgroup
*owner
;
4060 /* for delayed destruction */
4061 struct delayed_work destroy_dwork
;
4065 * The following two functions "fix" the issue where there are more pids
4066 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
4067 * TODO: replace with a kernel-wide solution to this problem
4069 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
4070 static void *pidlist_allocate(int count
)
4072 if (PIDLIST_TOO_LARGE(count
))
4073 return vmalloc(count
* sizeof(pid_t
));
4075 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
4078 static void pidlist_free(void *p
)
4084 * Used to destroy all pidlists lingering waiting for destroy timer. None
4085 * should be left afterwards.
4087 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
)
4089 struct cgroup_pidlist
*l
, *tmp_l
;
4091 mutex_lock(&cgrp
->pidlist_mutex
);
4092 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
4093 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
4094 mutex_unlock(&cgrp
->pidlist_mutex
);
4096 flush_workqueue(cgroup_pidlist_destroy_wq
);
4097 BUG_ON(!list_empty(&cgrp
->pidlists
));
4100 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
4102 struct delayed_work
*dwork
= to_delayed_work(work
);
4103 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
4105 struct cgroup_pidlist
*tofree
= NULL
;
4107 mutex_lock(&l
->owner
->pidlist_mutex
);
4110 * Destroy iff we didn't get queued again. The state won't change
4111 * as destroy_dwork can only be queued while locked.
4113 if (!delayed_work_pending(dwork
)) {
4114 list_del(&l
->links
);
4115 pidlist_free(l
->list
);
4116 put_pid_ns(l
->key
.ns
);
4120 mutex_unlock(&l
->owner
->pidlist_mutex
);
4125 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
4126 * Returns the number of unique elements.
4128 static int pidlist_uniq(pid_t
*list
, int length
)
4133 * we presume the 0th element is unique, so i starts at 1. trivial
4134 * edge cases first; no work needs to be done for either
4136 if (length
== 0 || length
== 1)
4138 /* src and dest walk down the list; dest counts unique elements */
4139 for (src
= 1; src
< length
; src
++) {
4140 /* find next unique element */
4141 while (list
[src
] == list
[src
-1]) {
4146 /* dest always points to where the next unique element goes */
4147 list
[dest
] = list
[src
];
4155 * The two pid files - task and cgroup.procs - guaranteed that the result
4156 * is sorted, which forced this whole pidlist fiasco. As pid order is
4157 * different per namespace, each namespace needs differently sorted list,
4158 * making it impossible to use, for example, single rbtree of member tasks
4159 * sorted by task pointer. As pidlists can be fairly large, allocating one
4160 * per open file is dangerous, so cgroup had to implement shared pool of
4161 * pidlists keyed by cgroup and namespace.
4163 * All this extra complexity was caused by the original implementation
4164 * committing to an entirely unnecessary property. In the long term, we
4165 * want to do away with it. Explicitly scramble sort order if on the
4166 * default hierarchy so that no such expectation exists in the new
4169 * Scrambling is done by swapping every two consecutive bits, which is
4170 * non-identity one-to-one mapping which disturbs sort order sufficiently.
4172 static pid_t
pid_fry(pid_t pid
)
4174 unsigned a
= pid
& 0x55555555;
4175 unsigned b
= pid
& 0xAAAAAAAA;
4177 return (a
<< 1) | (b
>> 1);
4180 static pid_t
cgroup_pid_fry(struct cgroup
*cgrp
, pid_t pid
)
4182 if (cgroup_on_dfl(cgrp
))
4183 return pid_fry(pid
);
4188 static int cmppid(const void *a
, const void *b
)
4190 return *(pid_t
*)a
- *(pid_t
*)b
;
4193 static int fried_cmppid(const void *a
, const void *b
)
4195 return pid_fry(*(pid_t
*)a
) - pid_fry(*(pid_t
*)b
);
4198 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
4199 enum cgroup_filetype type
)
4201 struct cgroup_pidlist
*l
;
4202 /* don't need task_nsproxy() if we're looking at ourself */
4203 struct pid_namespace
*ns
= task_active_pid_ns(current
);
4205 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4207 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
4208 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
4214 * find the appropriate pidlist for our purpose (given procs vs tasks)
4215 * returns with the lock on that pidlist already held, and takes care
4216 * of the use count, or returns NULL with no locks held if we're out of
4219 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
4220 enum cgroup_filetype type
)
4222 struct cgroup_pidlist
*l
;
4224 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4226 l
= cgroup_pidlist_find(cgrp
, type
);
4230 /* entry not found; create a new one */
4231 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
4235 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
4237 /* don't need task_nsproxy() if we're looking at ourself */
4238 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
4240 list_add(&l
->links
, &cgrp
->pidlists
);
4245 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
4247 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
4248 struct cgroup_pidlist
**lp
)
4252 int pid
, n
= 0; /* used for populating the array */
4253 struct css_task_iter it
;
4254 struct task_struct
*tsk
;
4255 struct cgroup_pidlist
*l
;
4257 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4260 * If cgroup gets more users after we read count, we won't have
4261 * enough space - tough. This race is indistinguishable to the
4262 * caller from the case that the additional cgroup users didn't
4263 * show up until sometime later on.
4265 length
= cgroup_task_count(cgrp
);
4266 array
= pidlist_allocate(length
);
4269 /* now, populate the array */
4270 css_task_iter_start(&cgrp
->self
, &it
);
4271 while ((tsk
= css_task_iter_next(&it
))) {
4272 if (unlikely(n
== length
))
4274 /* get tgid or pid for procs or tasks file respectively */
4275 if (type
== CGROUP_FILE_PROCS
)
4276 pid
= task_tgid_vnr(tsk
);
4278 pid
= task_pid_vnr(tsk
);
4279 if (pid
> 0) /* make sure to only use valid results */
4282 css_task_iter_end(&it
);
4284 /* now sort & (if procs) strip out duplicates */
4285 if (cgroup_on_dfl(cgrp
))
4286 sort(array
, length
, sizeof(pid_t
), fried_cmppid
, NULL
);
4288 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
4289 if (type
== CGROUP_FILE_PROCS
)
4290 length
= pidlist_uniq(array
, length
);
4292 l
= cgroup_pidlist_find_create(cgrp
, type
);
4294 pidlist_free(array
);
4298 /* store array, freeing old if necessary */
4299 pidlist_free(l
->list
);
4307 * cgroupstats_build - build and fill cgroupstats
4308 * @stats: cgroupstats to fill information into
4309 * @dentry: A dentry entry belonging to the cgroup for which stats have
4312 * Build and fill cgroupstats so that taskstats can export it to user
4315 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
4317 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
4318 struct cgroup
*cgrp
;
4319 struct css_task_iter it
;
4320 struct task_struct
*tsk
;
4322 /* it should be kernfs_node belonging to cgroupfs and is a directory */
4323 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
4324 kernfs_type(kn
) != KERNFS_DIR
)
4327 mutex_lock(&cgroup_mutex
);
4330 * We aren't being called from kernfs and there's no guarantee on
4331 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
4332 * @kn->priv is RCU safe. Let's do the RCU dancing.
4335 cgrp
= rcu_dereference(kn
->priv
);
4336 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
4338 mutex_unlock(&cgroup_mutex
);
4343 css_task_iter_start(&cgrp
->self
, &it
);
4344 while ((tsk
= css_task_iter_next(&it
))) {
4345 switch (tsk
->state
) {
4347 stats
->nr_running
++;
4349 case TASK_INTERRUPTIBLE
:
4350 stats
->nr_sleeping
++;
4352 case TASK_UNINTERRUPTIBLE
:
4353 stats
->nr_uninterruptible
++;
4356 stats
->nr_stopped
++;
4359 if (delayacct_is_task_waiting_on_io(tsk
))
4360 stats
->nr_io_wait
++;
4364 css_task_iter_end(&it
);
4366 mutex_unlock(&cgroup_mutex
);
4372 * seq_file methods for the tasks/procs files. The seq_file position is the
4373 * next pid to display; the seq_file iterator is a pointer to the pid
4374 * in the cgroup->l->list array.
4377 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
4380 * Initially we receive a position value that corresponds to
4381 * one more than the last pid shown (or 0 on the first call or
4382 * after a seek to the start). Use a binary-search to find the
4383 * next pid to display, if any
4385 struct kernfs_open_file
*of
= s
->private;
4386 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4387 struct cgroup_pidlist
*l
;
4388 enum cgroup_filetype type
= seq_cft(s
)->private;
4389 int index
= 0, pid
= *pos
;
4392 mutex_lock(&cgrp
->pidlist_mutex
);
4395 * !NULL @of->priv indicates that this isn't the first start()
4396 * after open. If the matching pidlist is around, we can use that.
4397 * Look for it. Note that @of->priv can't be used directly. It
4398 * could already have been destroyed.
4401 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
4404 * Either this is the first start() after open or the matching
4405 * pidlist has been destroyed inbetween. Create a new one.
4408 ret
= pidlist_array_load(cgrp
, type
,
4409 (struct cgroup_pidlist
**)&of
->priv
);
4411 return ERR_PTR(ret
);
4416 int end
= l
->length
;
4418 while (index
< end
) {
4419 int mid
= (index
+ end
) / 2;
4420 if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) == pid
) {
4423 } else if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) <= pid
)
4429 /* If we're off the end of the array, we're done */
4430 if (index
>= l
->length
)
4432 /* Update the abstract position to be the actual pid that we found */
4433 iter
= l
->list
+ index
;
4434 *pos
= cgroup_pid_fry(cgrp
, *iter
);
4438 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
4440 struct kernfs_open_file
*of
= s
->private;
4441 struct cgroup_pidlist
*l
= of
->priv
;
4444 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
4445 CGROUP_PIDLIST_DESTROY_DELAY
);
4446 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
4449 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4451 struct kernfs_open_file
*of
= s
->private;
4452 struct cgroup_pidlist
*l
= of
->priv
;
4454 pid_t
*end
= l
->list
+ l
->length
;
4456 * Advance to the next pid in the array. If this goes off the
4463 *pos
= cgroup_pid_fry(seq_css(s
)->cgroup
, *p
);
4468 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
4470 seq_printf(s
, "%d\n", *(int *)v
);
4475 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
4478 return notify_on_release(css
->cgroup
);
4481 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
4482 struct cftype
*cft
, u64 val
)
4485 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4487 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4491 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
4494 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4497 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
4498 struct cftype
*cft
, u64 val
)
4501 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4503 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4507 /* cgroup core interface files for the default hierarchy */
4508 static struct cftype cgroup_dfl_base_files
[] = {
4510 .name
= "cgroup.procs",
4511 .file_offset
= offsetof(struct cgroup
, procs_file
),
4512 .seq_start
= cgroup_pidlist_start
,
4513 .seq_next
= cgroup_pidlist_next
,
4514 .seq_stop
= cgroup_pidlist_stop
,
4515 .seq_show
= cgroup_pidlist_show
,
4516 .private = CGROUP_FILE_PROCS
,
4517 .write
= cgroup_procs_write
,
4520 .name
= "cgroup.controllers",
4521 .flags
= CFTYPE_ONLY_ON_ROOT
,
4522 .seq_show
= cgroup_root_controllers_show
,
4525 .name
= "cgroup.controllers",
4526 .flags
= CFTYPE_NOT_ON_ROOT
,
4527 .seq_show
= cgroup_controllers_show
,
4530 .name
= "cgroup.subtree_control",
4531 .seq_show
= cgroup_subtree_control_show
,
4532 .write
= cgroup_subtree_control_write
,
4535 .name
= "cgroup.events",
4536 .flags
= CFTYPE_NOT_ON_ROOT
,
4537 .file_offset
= offsetof(struct cgroup
, events_file
),
4538 .seq_show
= cgroup_events_show
,
4543 /* cgroup core interface files for the legacy hierarchies */
4544 static struct cftype cgroup_legacy_base_files
[] = {
4546 .name
= "cgroup.procs",
4547 .seq_start
= cgroup_pidlist_start
,
4548 .seq_next
= cgroup_pidlist_next
,
4549 .seq_stop
= cgroup_pidlist_stop
,
4550 .seq_show
= cgroup_pidlist_show
,
4551 .private = CGROUP_FILE_PROCS
,
4552 .write
= cgroup_procs_write
,
4555 .name
= "cgroup.clone_children",
4556 .read_u64
= cgroup_clone_children_read
,
4557 .write_u64
= cgroup_clone_children_write
,
4560 .name
= "cgroup.sane_behavior",
4561 .flags
= CFTYPE_ONLY_ON_ROOT
,
4562 .seq_show
= cgroup_sane_behavior_show
,
4566 .seq_start
= cgroup_pidlist_start
,
4567 .seq_next
= cgroup_pidlist_next
,
4568 .seq_stop
= cgroup_pidlist_stop
,
4569 .seq_show
= cgroup_pidlist_show
,
4570 .private = CGROUP_FILE_TASKS
,
4571 .write
= cgroup_tasks_write
,
4574 .name
= "notify_on_release",
4575 .read_u64
= cgroup_read_notify_on_release
,
4576 .write_u64
= cgroup_write_notify_on_release
,
4579 .name
= "release_agent",
4580 .flags
= CFTYPE_ONLY_ON_ROOT
,
4581 .seq_show
= cgroup_release_agent_show
,
4582 .write
= cgroup_release_agent_write
,
4583 .max_write_len
= PATH_MAX
- 1,
4589 * css destruction is four-stage process.
4591 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4592 * Implemented in kill_css().
4594 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4595 * and thus css_tryget_online() is guaranteed to fail, the css can be
4596 * offlined by invoking offline_css(). After offlining, the base ref is
4597 * put. Implemented in css_killed_work_fn().
4599 * 3. When the percpu_ref reaches zero, the only possible remaining
4600 * accessors are inside RCU read sections. css_release() schedules the
4603 * 4. After the grace period, the css can be freed. Implemented in
4604 * css_free_work_fn().
4606 * It is actually hairier because both step 2 and 4 require process context
4607 * and thus involve punting to css->destroy_work adding two additional
4608 * steps to the already complex sequence.
4610 static void css_free_work_fn(struct work_struct
*work
)
4612 struct cgroup_subsys_state
*css
=
4613 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4614 struct cgroup_subsys
*ss
= css
->ss
;
4615 struct cgroup
*cgrp
= css
->cgroup
;
4616 struct cgroup_file
*cfile
;
4618 percpu_ref_exit(&css
->refcnt
);
4620 list_for_each_entry(cfile
, &css
->files
, node
)
4621 kernfs_put(cfile
->kn
);
4628 css_put(css
->parent
);
4631 cgroup_idr_remove(&ss
->css_idr
, id
);
4634 /* cgroup free path */
4635 atomic_dec(&cgrp
->root
->nr_cgrps
);
4636 cgroup_pidlist_destroy_all(cgrp
);
4637 cancel_work_sync(&cgrp
->release_agent_work
);
4639 if (cgroup_parent(cgrp
)) {
4641 * We get a ref to the parent, and put the ref when
4642 * this cgroup is being freed, so it's guaranteed
4643 * that the parent won't be destroyed before its
4646 cgroup_put(cgroup_parent(cgrp
));
4647 kernfs_put(cgrp
->kn
);
4651 * This is root cgroup's refcnt reaching zero,
4652 * which indicates that the root should be
4655 cgroup_destroy_root(cgrp
->root
);
4660 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
4662 struct cgroup_subsys_state
*css
=
4663 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
4665 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
4666 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4669 static void css_release_work_fn(struct work_struct
*work
)
4671 struct cgroup_subsys_state
*css
=
4672 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4673 struct cgroup_subsys
*ss
= css
->ss
;
4674 struct cgroup
*cgrp
= css
->cgroup
;
4676 mutex_lock(&cgroup_mutex
);
4678 css
->flags
|= CSS_RELEASED
;
4679 list_del_rcu(&css
->sibling
);
4682 /* css release path */
4683 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
4684 if (ss
->css_released
)
4685 ss
->css_released(css
);
4687 /* cgroup release path */
4688 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4692 * There are two control paths which try to determine
4693 * cgroup from dentry without going through kernfs -
4694 * cgroupstats_build() and css_tryget_online_from_dir().
4695 * Those are supported by RCU protecting clearing of
4696 * cgrp->kn->priv backpointer.
4698 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
, NULL
);
4701 mutex_unlock(&cgroup_mutex
);
4703 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4706 static void css_release(struct percpu_ref
*ref
)
4708 struct cgroup_subsys_state
*css
=
4709 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4711 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
4712 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4715 static void init_and_link_css(struct cgroup_subsys_state
*css
,
4716 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4718 lockdep_assert_held(&cgroup_mutex
);
4722 memset(css
, 0, sizeof(*css
));
4725 INIT_LIST_HEAD(&css
->sibling
);
4726 INIT_LIST_HEAD(&css
->children
);
4727 INIT_LIST_HEAD(&css
->files
);
4728 css
->serial_nr
= css_serial_nr_next
++;
4730 if (cgroup_parent(cgrp
)) {
4731 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
4732 css_get(css
->parent
);
4735 BUG_ON(cgroup_css(cgrp
, ss
));
4738 /* invoke ->css_online() on a new CSS and mark it online if successful */
4739 static int online_css(struct cgroup_subsys_state
*css
)
4741 struct cgroup_subsys
*ss
= css
->ss
;
4744 lockdep_assert_held(&cgroup_mutex
);
4747 ret
= ss
->css_online(css
);
4749 css
->flags
|= CSS_ONLINE
;
4750 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
4755 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4756 static void offline_css(struct cgroup_subsys_state
*css
)
4758 struct cgroup_subsys
*ss
= css
->ss
;
4760 lockdep_assert_held(&cgroup_mutex
);
4762 if (!(css
->flags
& CSS_ONLINE
))
4765 if (ss
->css_offline
)
4766 ss
->css_offline(css
);
4768 css
->flags
&= ~CSS_ONLINE
;
4769 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
4771 wake_up_all(&css
->cgroup
->offline_waitq
);
4775 * create_css - create a cgroup_subsys_state
4776 * @cgrp: the cgroup new css will be associated with
4777 * @ss: the subsys of new css
4778 * @visible: whether to create control knobs for the new css or not
4780 * Create a new css associated with @cgrp - @ss pair. On success, the new
4781 * css is online and installed in @cgrp with all interface files created if
4782 * @visible. Returns 0 on success, -errno on failure.
4784 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
4787 struct cgroup
*parent
= cgroup_parent(cgrp
);
4788 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
4789 struct cgroup_subsys_state
*css
;
4792 lockdep_assert_held(&cgroup_mutex
);
4794 css
= ss
->css_alloc(parent_css
);
4796 return PTR_ERR(css
);
4798 init_and_link_css(css
, ss
, cgrp
);
4800 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
4804 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
4806 goto err_free_percpu_ref
;
4810 err
= css_populate_dir(css
, NULL
);
4815 /* @css is ready to be brought online now, make it visible */
4816 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
4817 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
4819 err
= online_css(css
);
4823 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4824 cgroup_parent(parent
)) {
4825 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4826 current
->comm
, current
->pid
, ss
->name
);
4827 if (!strcmp(ss
->name
, "memory"))
4828 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4829 ss
->warned_broken_hierarchy
= true;
4835 list_del_rcu(&css
->sibling
);
4836 css_clear_dir(css
, NULL
);
4838 cgroup_idr_remove(&ss
->css_idr
, css
->id
);
4839 err_free_percpu_ref
:
4840 percpu_ref_exit(&css
->refcnt
);
4842 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4846 static int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
,
4849 struct cgroup
*parent
, *cgrp
;
4850 struct cgroup_root
*root
;
4851 struct cgroup_subsys
*ss
;
4852 struct kernfs_node
*kn
;
4855 /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable.
4857 if (strchr(name
, '\n'))
4860 parent
= cgroup_kn_lock_live(parent_kn
);
4863 root
= parent
->root
;
4865 /* allocate the cgroup and its ID, 0 is reserved for the root */
4866 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4872 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
4877 * Temporarily set the pointer to NULL, so idr_find() won't return
4878 * a half-baked cgroup.
4880 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_KERNEL
);
4883 goto out_cancel_ref
;
4886 init_cgroup_housekeeping(cgrp
);
4888 cgrp
->self
.parent
= &parent
->self
;
4891 if (notify_on_release(parent
))
4892 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4894 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4895 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4897 /* create the directory */
4898 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
4906 * This extra ref will be put in cgroup_free_fn() and guarantees
4907 * that @cgrp->kn is always accessible.
4911 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
4913 /* allocation complete, commit to creation */
4914 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
4915 atomic_inc(&root
->nr_cgrps
);
4919 * @cgrp is now fully operational. If something fails after this
4920 * point, it'll be released via the normal destruction path.
4922 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4924 ret
= cgroup_kn_set_ugid(kn
);
4928 ret
= css_populate_dir(&cgrp
->self
, NULL
);
4932 /* let's create and online css's */
4933 for_each_subsys(ss
, ssid
) {
4934 if (parent
->child_subsys_mask
& (1 << ssid
)) {
4935 ret
= create_css(cgrp
, ss
,
4936 parent
->subtree_control
& (1 << ssid
));
4943 * On the default hierarchy, a child doesn't automatically inherit
4944 * subtree_control from the parent. Each is configured manually.
4946 if (!cgroup_on_dfl(cgrp
)) {
4947 cgrp
->subtree_control
= parent
->subtree_control
;
4948 cgroup_refresh_child_subsys_mask(cgrp
);
4951 kernfs_activate(kn
);
4957 cgroup_idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4959 percpu_ref_exit(&cgrp
->self
.refcnt
);
4963 cgroup_kn_unlock(parent_kn
);
4967 cgroup_destroy_locked(cgrp
);
4972 * This is called when the refcnt of a css is confirmed to be killed.
4973 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4974 * initate destruction and put the css ref from kill_css().
4976 static void css_killed_work_fn(struct work_struct
*work
)
4978 struct cgroup_subsys_state
*css
=
4979 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4981 mutex_lock(&cgroup_mutex
);
4983 mutex_unlock(&cgroup_mutex
);
4988 /* css kill confirmation processing requires process context, bounce */
4989 static void css_killed_ref_fn(struct percpu_ref
*ref
)
4991 struct cgroup_subsys_state
*css
=
4992 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4994 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
4995 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4999 * kill_css - destroy a css
5000 * @css: css to destroy
5002 * This function initiates destruction of @css by removing cgroup interface
5003 * files and putting its base reference. ->css_offline() will be invoked
5004 * asynchronously once css_tryget_online() is guaranteed to fail and when
5005 * the reference count reaches zero, @css will be released.
5007 static void kill_css(struct cgroup_subsys_state
*css
)
5009 lockdep_assert_held(&cgroup_mutex
);
5012 * This must happen before css is disassociated with its cgroup.
5013 * See seq_css() for details.
5015 css_clear_dir(css
, NULL
);
5018 * Killing would put the base ref, but we need to keep it alive
5019 * until after ->css_offline().
5024 * cgroup core guarantees that, by the time ->css_offline() is
5025 * invoked, no new css reference will be given out via
5026 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5027 * proceed to offlining css's because percpu_ref_kill() doesn't
5028 * guarantee that the ref is seen as killed on all CPUs on return.
5030 * Use percpu_ref_kill_and_confirm() to get notifications as each
5031 * css is confirmed to be seen as killed on all CPUs.
5033 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
5037 * cgroup_destroy_locked - the first stage of cgroup destruction
5038 * @cgrp: cgroup to be destroyed
5040 * css's make use of percpu refcnts whose killing latency shouldn't be
5041 * exposed to userland and are RCU protected. Also, cgroup core needs to
5042 * guarantee that css_tryget_online() won't succeed by the time
5043 * ->css_offline() is invoked. To satisfy all the requirements,
5044 * destruction is implemented in the following two steps.
5046 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5047 * userland visible parts and start killing the percpu refcnts of
5048 * css's. Set up so that the next stage will be kicked off once all
5049 * the percpu refcnts are confirmed to be killed.
5051 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5052 * rest of destruction. Once all cgroup references are gone, the
5053 * cgroup is RCU-freed.
5055 * This function implements s1. After this step, @cgrp is gone as far as
5056 * the userland is concerned and a new cgroup with the same name may be
5057 * created. As cgroup doesn't care about the names internally, this
5058 * doesn't cause any problem.
5060 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
5061 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
5063 struct cgroup_subsys_state
*css
;
5066 lockdep_assert_held(&cgroup_mutex
);
5069 * Only migration can raise populated from zero and we're already
5070 * holding cgroup_mutex.
5072 if (cgroup_is_populated(cgrp
))
5076 * Make sure there's no live children. We can't test emptiness of
5077 * ->self.children as dead children linger on it while being
5078 * drained; otherwise, "rmdir parent/child parent" may fail.
5080 if (css_has_online_children(&cgrp
->self
))
5084 * Mark @cgrp dead. This prevents further task migration and child
5085 * creation by disabling cgroup_lock_live_group().
5087 cgrp
->self
.flags
&= ~CSS_ONLINE
;
5089 /* initiate massacre of all css's */
5090 for_each_css(css
, ssid
, cgrp
)
5094 * Remove @cgrp directory along with the base files. @cgrp has an
5095 * extra ref on its kn.
5097 kernfs_remove(cgrp
->kn
);
5099 check_for_release(cgroup_parent(cgrp
));
5101 /* put the base reference */
5102 percpu_ref_kill(&cgrp
->self
.refcnt
);
5107 static int cgroup_rmdir(struct kernfs_node
*kn
)
5109 struct cgroup
*cgrp
;
5112 cgrp
= cgroup_kn_lock_live(kn
);
5116 ret
= cgroup_destroy_locked(cgrp
);
5118 cgroup_kn_unlock(kn
);
5122 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
5123 .remount_fs
= cgroup_remount
,
5124 .show_options
= cgroup_show_options
,
5125 .mkdir
= cgroup_mkdir
,
5126 .rmdir
= cgroup_rmdir
,
5127 .rename
= cgroup_rename
,
5130 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
5132 struct cgroup_subsys_state
*css
;
5134 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
5136 mutex_lock(&cgroup_mutex
);
5138 idr_init(&ss
->css_idr
);
5139 INIT_LIST_HEAD(&ss
->cfts
);
5141 /* Create the root cgroup state for this subsystem */
5142 ss
->root
= &cgrp_dfl_root
;
5143 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
5144 /* We don't handle early failures gracefully */
5145 BUG_ON(IS_ERR(css
));
5146 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
5149 * Root csses are never destroyed and we can't initialize
5150 * percpu_ref during early init. Disable refcnting.
5152 css
->flags
|= CSS_NO_REF
;
5155 /* allocation can't be done safely during early init */
5158 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
5159 BUG_ON(css
->id
< 0);
5162 /* Update the init_css_set to contain a subsys
5163 * pointer to this state - since the subsystem is
5164 * newly registered, all tasks and hence the
5165 * init_css_set is in the subsystem's root cgroup. */
5166 init_css_set
.subsys
[ss
->id
] = css
;
5168 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
5169 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
5170 have_free_callback
|= (bool)ss
->free
<< ss
->id
;
5171 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
5173 /* At system boot, before all subsystems have been
5174 * registered, no tasks have been forked, so we don't
5175 * need to invoke fork callbacks here. */
5176 BUG_ON(!list_empty(&init_task
.tasks
));
5178 BUG_ON(online_css(css
));
5180 mutex_unlock(&cgroup_mutex
);
5184 * cgroup_init_early - cgroup initialization at system boot
5186 * Initialize cgroups at system boot, and initialize any
5187 * subsystems that request early init.
5189 int __init
cgroup_init_early(void)
5191 static struct cgroup_sb_opts __initdata opts
;
5192 struct cgroup_subsys
*ss
;
5195 init_cgroup_root(&cgrp_dfl_root
, &opts
);
5196 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
5198 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5200 for_each_subsys(ss
, i
) {
5201 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
5202 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
5203 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5205 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5206 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5209 ss
->name
= cgroup_subsys_name
[i
];
5210 if (!ss
->legacy_name
)
5211 ss
->legacy_name
= cgroup_subsys_name
[i
];
5214 cgroup_init_subsys(ss
, true);
5219 static unsigned long cgroup_disable_mask __initdata
;
5222 * cgroup_init - cgroup initialization
5224 * Register cgroup filesystem and /proc file, and initialize
5225 * any subsystems that didn't request early init.
5227 int __init
cgroup_init(void)
5229 struct cgroup_subsys
*ss
;
5233 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem
));
5234 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_dfl_base_files
));
5235 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_legacy_base_files
));
5237 mutex_lock(&cgroup_mutex
);
5239 /* Add init_css_set to the hash table */
5240 key
= css_set_hash(init_css_set
.subsys
);
5241 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
5243 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
5245 mutex_unlock(&cgroup_mutex
);
5247 for_each_subsys(ss
, ssid
) {
5248 if (ss
->early_init
) {
5249 struct cgroup_subsys_state
*css
=
5250 init_css_set
.subsys
[ss
->id
];
5252 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5254 BUG_ON(css
->id
< 0);
5256 cgroup_init_subsys(ss
, false);
5259 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5260 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5263 * Setting dfl_root subsys_mask needs to consider the
5264 * disabled flag and cftype registration needs kmalloc,
5265 * both of which aren't available during early_init.
5267 if (cgroup_disable_mask
& (1 << ssid
)) {
5268 static_branch_disable(cgroup_subsys_enabled_key
[ssid
]);
5269 printk(KERN_INFO
"Disabling %s control group subsystem\n",
5274 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
5276 if (!ss
->dfl_cftypes
)
5277 cgrp_dfl_root_inhibit_ss_mask
|= 1 << ss
->id
;
5279 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5280 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5282 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5283 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5287 ss
->bind(init_css_set
.subsys
[ssid
]);
5290 WARN_ON(sysfs_create_mount_point(fs_kobj
, "cgroup"));
5291 WARN_ON(register_filesystem(&cgroup_fs_type
));
5292 WARN_ON(!proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
));
5297 static int __init
cgroup_wq_init(void)
5300 * There isn't much point in executing destruction path in
5301 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5302 * Use 1 for @max_active.
5304 * We would prefer to do this in cgroup_init() above, but that
5305 * is called before init_workqueues(): so leave this until after.
5307 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5308 BUG_ON(!cgroup_destroy_wq
);
5311 * Used to destroy pidlists and separate to serve as flush domain.
5312 * Cap @max_active to 1 too.
5314 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
5316 BUG_ON(!cgroup_pidlist_destroy_wq
);
5320 core_initcall(cgroup_wq_init
);
5323 * proc_cgroup_show()
5324 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5325 * - Used for /proc/<pid>/cgroup.
5327 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
5328 struct pid
*pid
, struct task_struct
*tsk
)
5332 struct cgroup_root
*root
;
5335 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5339 mutex_lock(&cgroup_mutex
);
5340 spin_lock_bh(&css_set_lock
);
5342 for_each_root(root
) {
5343 struct cgroup_subsys
*ss
;
5344 struct cgroup
*cgrp
;
5345 int ssid
, count
= 0;
5347 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_root_visible
)
5350 seq_printf(m
, "%d:", root
->hierarchy_id
);
5351 if (root
!= &cgrp_dfl_root
)
5352 for_each_subsys(ss
, ssid
)
5353 if (root
->subsys_mask
& (1 << ssid
))
5354 seq_printf(m
, "%s%s", count
++ ? "," : "",
5356 if (strlen(root
->name
))
5357 seq_printf(m
, "%sname=%s", count
? "," : "",
5361 cgrp
= task_cgroup_from_root(tsk
, root
);
5364 * On traditional hierarchies, all zombie tasks show up as
5365 * belonging to the root cgroup. On the default hierarchy,
5366 * while a zombie doesn't show up in "cgroup.procs" and
5367 * thus can't be migrated, its /proc/PID/cgroup keeps
5368 * reporting the cgroup it belonged to before exiting. If
5369 * the cgroup is removed before the zombie is reaped,
5370 * " (deleted)" is appended to the cgroup path.
5372 if (cgroup_on_dfl(cgrp
) || !(tsk
->flags
& PF_EXITING
)) {
5373 path
= cgroup_path(cgrp
, buf
, PATH_MAX
);
5375 retval
= -ENAMETOOLONG
;
5384 if (cgroup_on_dfl(cgrp
) && cgroup_is_dead(cgrp
))
5385 seq_puts(m
, " (deleted)\n");
5392 spin_unlock_bh(&css_set_lock
);
5393 mutex_unlock(&cgroup_mutex
);
5399 /* Display information about each subsystem and each hierarchy */
5400 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5402 struct cgroup_subsys
*ss
;
5405 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5407 * ideally we don't want subsystems moving around while we do this.
5408 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5409 * subsys/hierarchy state.
5411 mutex_lock(&cgroup_mutex
);
5413 for_each_subsys(ss
, i
)
5414 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5415 ss
->legacy_name
, ss
->root
->hierarchy_id
,
5416 atomic_read(&ss
->root
->nr_cgrps
),
5417 cgroup_ssid_enabled(i
));
5419 mutex_unlock(&cgroup_mutex
);
5423 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5425 return single_open(file
, proc_cgroupstats_show
, NULL
);
5428 static const struct file_operations proc_cgroupstats_operations
= {
5429 .open
= cgroupstats_open
,
5431 .llseek
= seq_lseek
,
5432 .release
= single_release
,
5435 static void **subsys_canfork_priv_p(void *ss_priv
[CGROUP_CANFORK_COUNT
], int i
)
5437 if (CGROUP_CANFORK_START
<= i
&& i
< CGROUP_CANFORK_END
)
5438 return &ss_priv
[i
- CGROUP_CANFORK_START
];
5442 static void *subsys_canfork_priv(void *ss_priv
[CGROUP_CANFORK_COUNT
], int i
)
5444 void **private = subsys_canfork_priv_p(ss_priv
, i
);
5445 return private ? *private : NULL
;
5449 * cgroup_fork - initialize cgroup related fields during copy_process()
5450 * @child: pointer to task_struct of forking parent process.
5452 * A task is associated with the init_css_set until cgroup_post_fork()
5453 * attaches it to the parent's css_set. Empty cg_list indicates that
5454 * @child isn't holding reference to its css_set.
5456 void cgroup_fork(struct task_struct
*child
)
5458 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5459 INIT_LIST_HEAD(&child
->cg_list
);
5463 * cgroup_can_fork - called on a new task before the process is exposed
5464 * @child: the task in question.
5466 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5467 * returns an error, the fork aborts with that error code. This allows for
5468 * a cgroup subsystem to conditionally allow or deny new forks.
5470 int cgroup_can_fork(struct task_struct
*child
,
5471 void *ss_priv
[CGROUP_CANFORK_COUNT
])
5473 struct cgroup_subsys
*ss
;
5476 for_each_subsys_which(ss
, i
, &have_canfork_callback
) {
5477 ret
= ss
->can_fork(child
, subsys_canfork_priv_p(ss_priv
, i
));
5485 for_each_subsys(ss
, j
) {
5488 if (ss
->cancel_fork
)
5489 ss
->cancel_fork(child
, subsys_canfork_priv(ss_priv
, j
));
5496 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5497 * @child: the task in question
5499 * This calls the cancel_fork() callbacks if a fork failed *after*
5500 * cgroup_can_fork() succeded.
5502 void cgroup_cancel_fork(struct task_struct
*child
,
5503 void *ss_priv
[CGROUP_CANFORK_COUNT
])
5505 struct cgroup_subsys
*ss
;
5508 for_each_subsys(ss
, i
)
5509 if (ss
->cancel_fork
)
5510 ss
->cancel_fork(child
, subsys_canfork_priv(ss_priv
, i
));
5514 * cgroup_post_fork - called on a new task after adding it to the task list
5515 * @child: the task in question
5517 * Adds the task to the list running through its css_set if necessary and
5518 * call the subsystem fork() callbacks. Has to be after the task is
5519 * visible on the task list in case we race with the first call to
5520 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5523 void cgroup_post_fork(struct task_struct
*child
,
5524 void *old_ss_priv
[CGROUP_CANFORK_COUNT
])
5526 struct cgroup_subsys
*ss
;
5530 * This may race against cgroup_enable_task_cg_lists(). As that
5531 * function sets use_task_css_set_links before grabbing
5532 * tasklist_lock and we just went through tasklist_lock to add
5533 * @child, it's guaranteed that either we see the set
5534 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5535 * @child during its iteration.
5537 * If we won the race, @child is associated with %current's
5538 * css_set. Grabbing css_set_lock guarantees both that the
5539 * association is stable, and, on completion of the parent's
5540 * migration, @child is visible in the source of migration or
5541 * already in the destination cgroup. This guarantee is necessary
5542 * when implementing operations which need to migrate all tasks of
5543 * a cgroup to another.
5545 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
5546 * will remain in init_css_set. This is safe because all tasks are
5547 * in the init_css_set before cg_links is enabled and there's no
5548 * operation which transfers all tasks out of init_css_set.
5550 if (use_task_css_set_links
) {
5551 struct css_set
*cset
;
5553 spin_lock_bh(&css_set_lock
);
5554 cset
= task_css_set(current
);
5555 if (list_empty(&child
->cg_list
)) {
5557 css_set_move_task(child
, NULL
, cset
, false);
5559 spin_unlock_bh(&css_set_lock
);
5563 * Call ss->fork(). This must happen after @child is linked on
5564 * css_set; otherwise, @child might change state between ->fork()
5565 * and addition to css_set.
5567 for_each_subsys_which(ss
, i
, &have_fork_callback
)
5568 ss
->fork(child
, subsys_canfork_priv(old_ss_priv
, i
));
5572 * cgroup_exit - detach cgroup from exiting task
5573 * @tsk: pointer to task_struct of exiting process
5575 * Description: Detach cgroup from @tsk and release it.
5577 * Note that cgroups marked notify_on_release force every task in
5578 * them to take the global cgroup_mutex mutex when exiting.
5579 * This could impact scaling on very large systems. Be reluctant to
5580 * use notify_on_release cgroups where very high task exit scaling
5581 * is required on large systems.
5583 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5584 * call cgroup_exit() while the task is still competent to handle
5585 * notify_on_release(), then leave the task attached to the root cgroup in
5586 * each hierarchy for the remainder of its exit. No need to bother with
5587 * init_css_set refcnting. init_css_set never goes away and we can't race
5588 * with migration path - PF_EXITING is visible to migration path.
5590 void cgroup_exit(struct task_struct
*tsk
)
5592 struct cgroup_subsys
*ss
;
5593 struct css_set
*cset
;
5597 * Unlink from @tsk from its css_set. As migration path can't race
5598 * with us, we can check css_set and cg_list without synchronization.
5600 cset
= task_css_set(tsk
);
5602 if (!list_empty(&tsk
->cg_list
)) {
5603 spin_lock_bh(&css_set_lock
);
5604 css_set_move_task(tsk
, cset
, NULL
, false);
5605 spin_unlock_bh(&css_set_lock
);
5610 /* see cgroup_post_fork() for details */
5611 for_each_subsys_which(ss
, i
, &have_exit_callback
)
5615 void cgroup_free(struct task_struct
*task
)
5617 struct css_set
*cset
= task_css_set(task
);
5618 struct cgroup_subsys
*ss
;
5621 for_each_subsys_which(ss
, ssid
, &have_free_callback
)
5627 static void check_for_release(struct cgroup
*cgrp
)
5629 if (notify_on_release(cgrp
) && !cgroup_is_populated(cgrp
) &&
5630 !css_has_online_children(&cgrp
->self
) && !cgroup_is_dead(cgrp
))
5631 schedule_work(&cgrp
->release_agent_work
);
5635 * Notify userspace when a cgroup is released, by running the
5636 * configured release agent with the name of the cgroup (path
5637 * relative to the root of cgroup file system) as the argument.
5639 * Most likely, this user command will try to rmdir this cgroup.
5641 * This races with the possibility that some other task will be
5642 * attached to this cgroup before it is removed, or that some other
5643 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5644 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5645 * unused, and this cgroup will be reprieved from its death sentence,
5646 * to continue to serve a useful existence. Next time it's released,
5647 * we will get notified again, if it still has 'notify_on_release' set.
5649 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5650 * means only wait until the task is successfully execve()'d. The
5651 * separate release agent task is forked by call_usermodehelper(),
5652 * then control in this thread returns here, without waiting for the
5653 * release agent task. We don't bother to wait because the caller of
5654 * this routine has no use for the exit status of the release agent
5655 * task, so no sense holding our caller up for that.
5657 static void cgroup_release_agent(struct work_struct
*work
)
5659 struct cgroup
*cgrp
=
5660 container_of(work
, struct cgroup
, release_agent_work
);
5661 char *pathbuf
= NULL
, *agentbuf
= NULL
, *path
;
5662 char *argv
[3], *envp
[3];
5664 mutex_lock(&cgroup_mutex
);
5666 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5667 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5668 if (!pathbuf
|| !agentbuf
)
5671 path
= cgroup_path(cgrp
, pathbuf
, PATH_MAX
);
5679 /* minimal command environment */
5681 envp
[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5684 mutex_unlock(&cgroup_mutex
);
5685 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5688 mutex_unlock(&cgroup_mutex
);
5694 static int __init
cgroup_disable(char *str
)
5696 struct cgroup_subsys
*ss
;
5700 while ((token
= strsep(&str
, ",")) != NULL
) {
5704 for_each_subsys(ss
, i
) {
5705 if (strcmp(token
, ss
->name
) &&
5706 strcmp(token
, ss
->legacy_name
))
5708 cgroup_disable_mask
|= 1 << i
;
5713 __setup("cgroup_disable=", cgroup_disable
);
5716 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5717 * @dentry: directory dentry of interest
5718 * @ss: subsystem of interest
5720 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5721 * to get the corresponding css and return it. If such css doesn't exist
5722 * or can't be pinned, an ERR_PTR value is returned.
5724 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
5725 struct cgroup_subsys
*ss
)
5727 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
5728 struct cgroup_subsys_state
*css
= NULL
;
5729 struct cgroup
*cgrp
;
5731 /* is @dentry a cgroup dir? */
5732 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
5733 kernfs_type(kn
) != KERNFS_DIR
)
5734 return ERR_PTR(-EBADF
);
5739 * This path doesn't originate from kernfs and @kn could already
5740 * have been or be removed at any point. @kn->priv is RCU
5741 * protected for this access. See css_release_work_fn() for details.
5743 cgrp
= rcu_dereference(kn
->priv
);
5745 css
= cgroup_css(cgrp
, ss
);
5747 if (!css
|| !css_tryget_online(css
))
5748 css
= ERR_PTR(-ENOENT
);
5755 * css_from_id - lookup css by id
5756 * @id: the cgroup id
5757 * @ss: cgroup subsys to be looked into
5759 * Returns the css if there's valid one with @id, otherwise returns NULL.
5760 * Should be called under rcu_read_lock().
5762 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5764 WARN_ON_ONCE(!rcu_read_lock_held());
5765 return id
> 0 ? idr_find(&ss
->css_idr
, id
) : NULL
;
5768 #ifdef CONFIG_CGROUP_DEBUG
5769 static struct cgroup_subsys_state
*
5770 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
5772 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5775 return ERR_PTR(-ENOMEM
);
5780 static void debug_css_free(struct cgroup_subsys_state
*css
)
5785 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
5788 return cgroup_task_count(css
->cgroup
);
5791 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
5794 return (u64
)(unsigned long)current
->cgroups
;
5797 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
5803 count
= atomic_read(&task_css_set(current
)->refcount
);
5808 static int current_css_set_cg_links_read(struct seq_file
*seq
, void *v
)
5810 struct cgrp_cset_link
*link
;
5811 struct css_set
*cset
;
5814 name_buf
= kmalloc(NAME_MAX
+ 1, GFP_KERNEL
);
5818 spin_lock_bh(&css_set_lock
);
5820 cset
= rcu_dereference(current
->cgroups
);
5821 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5822 struct cgroup
*c
= link
->cgrp
;
5824 cgroup_name(c
, name_buf
, NAME_MAX
+ 1);
5825 seq_printf(seq
, "Root %d group %s\n",
5826 c
->root
->hierarchy_id
, name_buf
);
5829 spin_unlock_bh(&css_set_lock
);
5834 #define MAX_TASKS_SHOWN_PER_CSS 25
5835 static int cgroup_css_links_read(struct seq_file
*seq
, void *v
)
5837 struct cgroup_subsys_state
*css
= seq_css(seq
);
5838 struct cgrp_cset_link
*link
;
5840 spin_lock_bh(&css_set_lock
);
5841 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
5842 struct css_set
*cset
= link
->cset
;
5843 struct task_struct
*task
;
5846 seq_printf(seq
, "css_set %p\n", cset
);
5848 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5849 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5851 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5854 list_for_each_entry(task
, &cset
->mg_tasks
, cg_list
) {
5855 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5857 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5861 seq_puts(seq
, " ...\n");
5863 spin_unlock_bh(&css_set_lock
);
5867 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
5869 return (!cgroup_is_populated(css
->cgroup
) &&
5870 !css_has_online_children(&css
->cgroup
->self
));
5873 static struct cftype debug_files
[] = {
5875 .name
= "taskcount",
5876 .read_u64
= debug_taskcount_read
,
5880 .name
= "current_css_set",
5881 .read_u64
= current_css_set_read
,
5885 .name
= "current_css_set_refcount",
5886 .read_u64
= current_css_set_refcount_read
,
5890 .name
= "current_css_set_cg_links",
5891 .seq_show
= current_css_set_cg_links_read
,
5895 .name
= "cgroup_css_links",
5896 .seq_show
= cgroup_css_links_read
,
5900 .name
= "releasable",
5901 .read_u64
= releasable_read
,
5907 struct cgroup_subsys debug_cgrp_subsys
= {
5908 .css_alloc
= debug_css_alloc
,
5909 .css_free
= debug_css_free
,
5910 .legacy_cftypes
= debug_files
,
5912 #endif /* CONFIG_CGROUP_DEBUG */