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>
60 #include <linux/atomic.h>
61 #include <linux/cpuset.h>
62 #include <linux/proc_ns.h>
63 #include <linux/nsproxy.h>
64 #include <linux/proc_ns.h>
68 * pidlists linger the following amount before being destroyed. The goal
69 * is avoiding frequent destruction in the middle of consecutive read calls
70 * Expiring in the middle is a performance problem not a correctness one.
71 * 1 sec should be enough.
73 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
75 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
79 * cgroup_mutex is the master lock. Any modification to cgroup or its
80 * hierarchy must be performed while holding it.
82 * css_set_lock protects task->cgroups pointer, the list of css_set
83 * objects, and the chain of tasks off each css_set.
85 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
86 * cgroup.h can use them for lockdep annotations.
88 #ifdef CONFIG_PROVE_RCU
89 DEFINE_MUTEX(cgroup_mutex
);
90 DEFINE_SPINLOCK(css_set_lock
);
91 EXPORT_SYMBOL_GPL(cgroup_mutex
);
92 EXPORT_SYMBOL_GPL(css_set_lock
);
94 static DEFINE_MUTEX(cgroup_mutex
);
95 static DEFINE_SPINLOCK(css_set_lock
);
99 * Protects cgroup_idr and css_idr so that IDs can be released without
100 * grabbing cgroup_mutex.
102 static DEFINE_SPINLOCK(cgroup_idr_lock
);
105 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
106 * against file removal/re-creation across css hiding.
108 static DEFINE_SPINLOCK(cgroup_file_kn_lock
);
111 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
112 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
114 static DEFINE_SPINLOCK(release_agent_path_lock
);
116 struct percpu_rw_semaphore cgroup_threadgroup_rwsem
;
118 #define cgroup_assert_mutex_or_rcu_locked() \
119 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
120 !lockdep_is_held(&cgroup_mutex), \
121 "cgroup_mutex or RCU read lock required");
124 * cgroup destruction makes heavy use of work items and there can be a lot
125 * of concurrent destructions. Use a separate workqueue so that cgroup
126 * destruction work items don't end up filling up max_active of system_wq
127 * which may lead to deadlock.
129 static struct workqueue_struct
*cgroup_destroy_wq
;
132 * pidlist destructions need to be flushed on cgroup destruction. Use a
133 * separate workqueue as flush domain.
135 static struct workqueue_struct
*cgroup_pidlist_destroy_wq
;
137 /* generate an array of cgroup subsystem pointers */
138 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
139 static struct cgroup_subsys
*cgroup_subsys
[] = {
140 #include <linux/cgroup_subsys.h>
144 /* array of cgroup subsystem names */
145 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
146 static const char *cgroup_subsys_name
[] = {
147 #include <linux/cgroup_subsys.h>
151 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
153 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
154 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
155 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
156 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
157 #include <linux/cgroup_subsys.h>
160 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
161 static struct static_key_true
*cgroup_subsys_enabled_key
[] = {
162 #include <linux/cgroup_subsys.h>
166 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
167 static struct static_key_true
*cgroup_subsys_on_dfl_key
[] = {
168 #include <linux/cgroup_subsys.h>
173 * The default hierarchy, reserved for the subsystems that are otherwise
174 * unattached - it never has more than a single cgroup, and all tasks are
175 * part of that cgroup.
177 struct cgroup_root cgrp_dfl_root
;
178 EXPORT_SYMBOL_GPL(cgrp_dfl_root
);
181 * The default hierarchy always exists but is hidden until mounted for the
182 * first time. This is for backward compatibility.
184 static bool cgrp_dfl_root_visible
;
186 /* Controllers blocked by the commandline in v1 */
187 static unsigned long cgroup_no_v1_mask
;
189 /* some controllers are not supported in the default hierarchy */
190 static unsigned long cgrp_dfl_root_inhibit_ss_mask
;
192 /* The list of hierarchy roots */
194 static LIST_HEAD(cgroup_roots
);
195 static int cgroup_root_count
;
197 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
198 static DEFINE_IDR(cgroup_hierarchy_idr
);
201 * Assign a monotonically increasing serial number to csses. It guarantees
202 * cgroups with bigger numbers are newer than those with smaller numbers.
203 * Also, as csses are always appended to the parent's ->children list, it
204 * guarantees that sibling csses are always sorted in the ascending serial
205 * number order on the list. Protected by cgroup_mutex.
207 static u64 css_serial_nr_next
= 1;
210 * These bitmask flags indicate whether tasks in the fork and exit paths have
211 * fork/exit handlers to call. This avoids us having to do extra work in the
212 * fork/exit path to check which subsystems have fork/exit callbacks.
214 static unsigned long have_fork_callback __read_mostly
;
215 static unsigned long have_exit_callback __read_mostly
;
216 static unsigned long have_free_callback __read_mostly
;
218 /* cgroup namespace for init task */
219 struct cgroup_namespace init_cgroup_ns
= {
220 .count
= { .counter
= 2, },
221 .user_ns
= &init_user_ns
,
222 .ns
.ops
= &cgroupns_operations
,
223 .ns
.inum
= PROC_CGROUP_INIT_INO
,
224 .root_cset
= &init_css_set
,
227 /* Ditto for the can_fork callback. */
228 static unsigned long have_canfork_callback __read_mostly
;
230 static struct file_system_type cgroup2_fs_type
;
231 static struct cftype cgroup_dfl_base_files
[];
232 static struct cftype cgroup_legacy_base_files
[];
234 static int rebind_subsystems(struct cgroup_root
*dst_root
,
235 unsigned long ss_mask
);
236 static void css_task_iter_advance(struct css_task_iter
*it
);
237 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
238 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
240 static void css_release(struct percpu_ref
*ref
);
241 static void kill_css(struct cgroup_subsys_state
*css
);
242 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
243 struct cgroup
*cgrp
, struct cftype cfts
[],
247 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
248 * @ssid: subsys ID of interest
250 * cgroup_subsys_enabled() can only be used with literal subsys names which
251 * is fine for individual subsystems but unsuitable for cgroup core. This
252 * is slower static_key_enabled() based test indexed by @ssid.
254 static bool cgroup_ssid_enabled(int ssid
)
256 return static_key_enabled(cgroup_subsys_enabled_key
[ssid
]);
259 static bool cgroup_ssid_no_v1(int ssid
)
261 return cgroup_no_v1_mask
& (1 << ssid
);
265 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
266 * @cgrp: the cgroup of interest
268 * The default hierarchy is the v2 interface of cgroup and this function
269 * can be used to test whether a cgroup is on the default hierarchy for
270 * cases where a subsystem should behave differnetly depending on the
273 * The set of behaviors which change on the default hierarchy are still
274 * being determined and the mount option is prefixed with __DEVEL__.
276 * List of changed behaviors:
278 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
279 * and "name" are disallowed.
281 * - When mounting an existing superblock, mount options should match.
283 * - Remount is disallowed.
285 * - rename(2) is disallowed.
287 * - "tasks" is removed. Everything should be at process granularity. Use
288 * "cgroup.procs" instead.
290 * - "cgroup.procs" is not sorted. pids will be unique unless they got
291 * recycled inbetween reads.
293 * - "release_agent" and "notify_on_release" are removed. Replacement
294 * notification mechanism will be implemented.
296 * - "cgroup.clone_children" is removed.
298 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
299 * and its descendants contain no task; otherwise, 1. The file also
300 * generates kernfs notification which can be monitored through poll and
301 * [di]notify when the value of the file changes.
303 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
304 * take masks of ancestors with non-empty cpus/mems, instead of being
305 * moved to an ancestor.
307 * - cpuset: a task can be moved into an empty cpuset, and again it takes
308 * masks of ancestors.
310 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
313 * - blkcg: blk-throttle becomes properly hierarchical.
315 * - debug: disallowed on the default hierarchy.
317 static bool cgroup_on_dfl(const struct cgroup
*cgrp
)
319 return cgrp
->root
== &cgrp_dfl_root
;
322 /* IDR wrappers which synchronize using cgroup_idr_lock */
323 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
328 idr_preload(gfp_mask
);
329 spin_lock_bh(&cgroup_idr_lock
);
330 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
& ~__GFP_DIRECT_RECLAIM
);
331 spin_unlock_bh(&cgroup_idr_lock
);
336 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
340 spin_lock_bh(&cgroup_idr_lock
);
341 ret
= idr_replace(idr
, ptr
, id
);
342 spin_unlock_bh(&cgroup_idr_lock
);
346 static void cgroup_idr_remove(struct idr
*idr
, int id
)
348 spin_lock_bh(&cgroup_idr_lock
);
350 spin_unlock_bh(&cgroup_idr_lock
);
353 static struct cgroup
*cgroup_parent(struct cgroup
*cgrp
)
355 struct cgroup_subsys_state
*parent_css
= cgrp
->self
.parent
;
358 return container_of(parent_css
, struct cgroup
, self
);
363 * cgroup_css - obtain a cgroup's css for the specified subsystem
364 * @cgrp: the cgroup of interest
365 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
367 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
368 * function must be called either under cgroup_mutex or rcu_read_lock() and
369 * the caller is responsible for pinning the returned css if it wants to
370 * keep accessing it outside the said locks. This function may return
371 * %NULL if @cgrp doesn't have @subsys_id enabled.
373 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
374 struct cgroup_subsys
*ss
)
377 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
378 lockdep_is_held(&cgroup_mutex
));
384 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
385 * @cgrp: the cgroup of interest
386 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
388 * Similar to cgroup_css() but returns the effective css, which is defined
389 * as the matching css of the nearest ancestor including self which has @ss
390 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
391 * function is guaranteed to return non-NULL css.
393 static struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
394 struct cgroup_subsys
*ss
)
396 lockdep_assert_held(&cgroup_mutex
);
401 if (!(cgrp
->root
->subsys_mask
& (1 << ss
->id
)))
405 * This function is used while updating css associations and thus
406 * can't test the csses directly. Use ->child_subsys_mask.
408 while (cgroup_parent(cgrp
) &&
409 !(cgroup_parent(cgrp
)->child_subsys_mask
& (1 << ss
->id
)))
410 cgrp
= cgroup_parent(cgrp
);
412 return cgroup_css(cgrp
, ss
);
416 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
417 * @cgrp: the cgroup of interest
418 * @ss: the subsystem of interest
420 * Find and get the effective css of @cgrp for @ss. The effective css is
421 * defined as the matching css of the nearest ancestor including self which
422 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
423 * the root css is returned, so this function always returns a valid css.
424 * The returned css must be put using css_put().
426 struct cgroup_subsys_state
*cgroup_get_e_css(struct cgroup
*cgrp
,
427 struct cgroup_subsys
*ss
)
429 struct cgroup_subsys_state
*css
;
434 css
= cgroup_css(cgrp
, ss
);
436 if (css
&& css_tryget_online(css
))
438 cgrp
= cgroup_parent(cgrp
);
441 css
= init_css_set
.subsys
[ss
->id
];
448 /* convenient tests for these bits */
449 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
451 return !(cgrp
->self
.flags
& CSS_ONLINE
);
454 static void cgroup_get(struct cgroup
*cgrp
)
456 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
457 css_get(&cgrp
->self
);
460 static bool cgroup_tryget(struct cgroup
*cgrp
)
462 return css_tryget(&cgrp
->self
);
465 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
467 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
468 struct cftype
*cft
= of_cft(of
);
471 * This is open and unprotected implementation of cgroup_css().
472 * seq_css() is only called from a kernfs file operation which has
473 * an active reference on the file. Because all the subsystem
474 * files are drained before a css is disassociated with a cgroup,
475 * the matching css from the cgroup's subsys table is guaranteed to
476 * be and stay valid until the enclosing operation is complete.
479 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
483 EXPORT_SYMBOL_GPL(of_css
);
485 static int notify_on_release(const struct cgroup
*cgrp
)
487 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
491 * for_each_css - iterate all css's of a cgroup
492 * @css: the iteration cursor
493 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
494 * @cgrp: the target cgroup to iterate css's of
496 * Should be called under cgroup_[tree_]mutex.
498 #define for_each_css(css, ssid, cgrp) \
499 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
500 if (!((css) = rcu_dereference_check( \
501 (cgrp)->subsys[(ssid)], \
502 lockdep_is_held(&cgroup_mutex)))) { } \
506 * for_each_e_css - iterate all effective css's of a cgroup
507 * @css: the iteration cursor
508 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
509 * @cgrp: the target cgroup to iterate css's of
511 * Should be called under cgroup_[tree_]mutex.
513 #define for_each_e_css(css, ssid, cgrp) \
514 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
515 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
520 * for_each_subsys - iterate all enabled cgroup subsystems
521 * @ss: the iteration cursor
522 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
524 #define for_each_subsys(ss, ssid) \
525 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
526 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
529 * for_each_subsys_which - filter for_each_subsys with a bitmask
530 * @ss: the iteration cursor
531 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
532 * @ss_maskp: a pointer to the bitmask
534 * The block will only run for cases where the ssid-th bit (1 << ssid) of
537 #define for_each_subsys_which(ss, ssid, ss_maskp) \
538 if (!CGROUP_SUBSYS_COUNT) /* to avoid spurious gcc warning */ \
541 for_each_set_bit(ssid, ss_maskp, CGROUP_SUBSYS_COUNT) \
542 if (((ss) = cgroup_subsys[ssid]) && false) \
546 /* iterate across the hierarchies */
547 #define for_each_root(root) \
548 list_for_each_entry((root), &cgroup_roots, root_list)
550 /* iterate over child cgrps, lock should be held throughout iteration */
551 #define cgroup_for_each_live_child(child, cgrp) \
552 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
553 if (({ lockdep_assert_held(&cgroup_mutex); \
554 cgroup_is_dead(child); })) \
558 static void cgroup_release_agent(struct work_struct
*work
);
559 static void check_for_release(struct cgroup
*cgrp
);
562 * A cgroup can be associated with multiple css_sets as different tasks may
563 * belong to different cgroups on different hierarchies. In the other
564 * direction, a css_set is naturally associated with multiple cgroups.
565 * This M:N relationship is represented by the following link structure
566 * which exists for each association and allows traversing the associations
569 struct cgrp_cset_link
{
570 /* the cgroup and css_set this link associates */
572 struct css_set
*cset
;
574 /* list of cgrp_cset_links anchored at cgrp->cset_links */
575 struct list_head cset_link
;
577 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
578 struct list_head cgrp_link
;
582 * The default css_set - used by init and its children prior to any
583 * hierarchies being mounted. It contains a pointer to the root state
584 * for each subsystem. Also used to anchor the list of css_sets. Not
585 * reference-counted, to improve performance when child cgroups
586 * haven't been created.
588 struct css_set init_css_set
= {
589 .refcount
= ATOMIC_INIT(1),
590 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
591 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
592 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
593 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
594 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
595 .task_iters
= LIST_HEAD_INIT(init_css_set
.task_iters
),
598 static int css_set_count
= 1; /* 1 for init_css_set */
601 * css_set_populated - does a css_set contain any tasks?
602 * @cset: target css_set
604 static bool css_set_populated(struct css_set
*cset
)
606 lockdep_assert_held(&css_set_lock
);
608 return !list_empty(&cset
->tasks
) || !list_empty(&cset
->mg_tasks
);
612 * cgroup_update_populated - updated populated count of a cgroup
613 * @cgrp: the target cgroup
614 * @populated: inc or dec populated count
616 * One of the css_sets associated with @cgrp is either getting its first
617 * task or losing the last. Update @cgrp->populated_cnt accordingly. The
618 * count is propagated towards root so that a given cgroup's populated_cnt
619 * is zero iff the cgroup and all its descendants don't contain any tasks.
621 * @cgrp's interface file "cgroup.populated" is zero if
622 * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
623 * changes from or to zero, userland is notified that the content of the
624 * interface file has changed. This can be used to detect when @cgrp and
625 * its descendants become populated or empty.
627 static void cgroup_update_populated(struct cgroup
*cgrp
, bool populated
)
629 lockdep_assert_held(&css_set_lock
);
635 trigger
= !cgrp
->populated_cnt
++;
637 trigger
= !--cgrp
->populated_cnt
;
642 check_for_release(cgrp
);
643 cgroup_file_notify(&cgrp
->events_file
);
645 cgrp
= cgroup_parent(cgrp
);
650 * css_set_update_populated - update populated state of a css_set
651 * @cset: target css_set
652 * @populated: whether @cset is populated or depopulated
654 * @cset is either getting the first task or losing the last. Update the
655 * ->populated_cnt of all associated cgroups accordingly.
657 static void css_set_update_populated(struct css_set
*cset
, bool populated
)
659 struct cgrp_cset_link
*link
;
661 lockdep_assert_held(&css_set_lock
);
663 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
)
664 cgroup_update_populated(link
->cgrp
, populated
);
668 * css_set_move_task - move a task from one css_set to another
669 * @task: task being moved
670 * @from_cset: css_set @task currently belongs to (may be NULL)
671 * @to_cset: new css_set @task is being moved to (may be NULL)
672 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
674 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
675 * css_set, @from_cset can be NULL. If @task is being disassociated
676 * instead of moved, @to_cset can be NULL.
678 * This function automatically handles populated_cnt updates and
679 * css_task_iter adjustments but the caller is responsible for managing
680 * @from_cset and @to_cset's reference counts.
682 static void css_set_move_task(struct task_struct
*task
,
683 struct css_set
*from_cset
, struct css_set
*to_cset
,
686 lockdep_assert_held(&css_set_lock
);
689 struct css_task_iter
*it
, *pos
;
691 WARN_ON_ONCE(list_empty(&task
->cg_list
));
694 * @task is leaving, advance task iterators which are
695 * pointing to it so that they can resume at the next
696 * position. Advancing an iterator might remove it from
697 * the list, use safe walk. See css_task_iter_advance*()
700 list_for_each_entry_safe(it
, pos
, &from_cset
->task_iters
,
702 if (it
->task_pos
== &task
->cg_list
)
703 css_task_iter_advance(it
);
705 list_del_init(&task
->cg_list
);
706 if (!css_set_populated(from_cset
))
707 css_set_update_populated(from_cset
, false);
709 WARN_ON_ONCE(!list_empty(&task
->cg_list
));
714 * We are synchronized through cgroup_threadgroup_rwsem
715 * against PF_EXITING setting such that we can't race
716 * against cgroup_exit() changing the css_set to
717 * init_css_set and dropping the old one.
719 WARN_ON_ONCE(task
->flags
& PF_EXITING
);
721 if (!css_set_populated(to_cset
))
722 css_set_update_populated(to_cset
, true);
723 rcu_assign_pointer(task
->cgroups
, to_cset
);
724 list_add_tail(&task
->cg_list
, use_mg_tasks
? &to_cset
->mg_tasks
:
730 * hash table for cgroup groups. This improves the performance to find
731 * an existing css_set. This hash doesn't (currently) take into
732 * account cgroups in empty hierarchies.
734 #define CSS_SET_HASH_BITS 7
735 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
737 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
739 unsigned long key
= 0UL;
740 struct cgroup_subsys
*ss
;
743 for_each_subsys(ss
, i
)
744 key
+= (unsigned long)css
[i
];
745 key
= (key
>> 16) ^ key
;
750 static void put_css_set_locked(struct css_set
*cset
)
752 struct cgrp_cset_link
*link
, *tmp_link
;
753 struct cgroup_subsys
*ss
;
756 lockdep_assert_held(&css_set_lock
);
758 if (!atomic_dec_and_test(&cset
->refcount
))
761 /* This css_set is dead. unlink it and release cgroup and css refs */
762 for_each_subsys(ss
, ssid
) {
763 list_del(&cset
->e_cset_node
[ssid
]);
764 css_put(cset
->subsys
[ssid
]);
766 hash_del(&cset
->hlist
);
769 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
770 list_del(&link
->cset_link
);
771 list_del(&link
->cgrp_link
);
772 if (cgroup_parent(link
->cgrp
))
773 cgroup_put(link
->cgrp
);
777 kfree_rcu(cset
, rcu_head
);
780 static void put_css_set(struct css_set
*cset
)
783 * Ensure that the refcount doesn't hit zero while any readers
784 * can see it. Similar to atomic_dec_and_lock(), but for an
787 if (atomic_add_unless(&cset
->refcount
, -1, 1))
790 spin_lock_bh(&css_set_lock
);
791 put_css_set_locked(cset
);
792 spin_unlock_bh(&css_set_lock
);
796 * refcounted get/put for css_set objects
798 static inline void get_css_set(struct css_set
*cset
)
800 atomic_inc(&cset
->refcount
);
804 * compare_css_sets - helper function for find_existing_css_set().
805 * @cset: candidate css_set being tested
806 * @old_cset: existing css_set for a task
807 * @new_cgrp: cgroup that's being entered by the task
808 * @template: desired set of css pointers in css_set (pre-calculated)
810 * Returns true if "cset" matches "old_cset" except for the hierarchy
811 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
813 static bool compare_css_sets(struct css_set
*cset
,
814 struct css_set
*old_cset
,
815 struct cgroup
*new_cgrp
,
816 struct cgroup_subsys_state
*template[])
818 struct list_head
*l1
, *l2
;
821 * On the default hierarchy, there can be csets which are
822 * associated with the same set of cgroups but different csses.
823 * Let's first ensure that csses match.
825 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
829 * Compare cgroup pointers in order to distinguish between
830 * different cgroups in hierarchies. As different cgroups may
831 * share the same effective css, this comparison is always
834 l1
= &cset
->cgrp_links
;
835 l2
= &old_cset
->cgrp_links
;
837 struct cgrp_cset_link
*link1
, *link2
;
838 struct cgroup
*cgrp1
, *cgrp2
;
842 /* See if we reached the end - both lists are equal length. */
843 if (l1
== &cset
->cgrp_links
) {
844 BUG_ON(l2
!= &old_cset
->cgrp_links
);
847 BUG_ON(l2
== &old_cset
->cgrp_links
);
849 /* Locate the cgroups associated with these links. */
850 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
851 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
854 /* Hierarchies should be linked in the same order. */
855 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
858 * If this hierarchy is the hierarchy of the cgroup
859 * that's changing, then we need to check that this
860 * css_set points to the new cgroup; if it's any other
861 * hierarchy, then this css_set should point to the
862 * same cgroup as the old css_set.
864 if (cgrp1
->root
== new_cgrp
->root
) {
865 if (cgrp1
!= new_cgrp
)
876 * find_existing_css_set - init css array and find the matching css_set
877 * @old_cset: the css_set that we're using before the cgroup transition
878 * @cgrp: the cgroup that we're moving into
879 * @template: out param for the new set of csses, should be clear on entry
881 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
883 struct cgroup_subsys_state
*template[])
885 struct cgroup_root
*root
= cgrp
->root
;
886 struct cgroup_subsys
*ss
;
887 struct css_set
*cset
;
892 * Build the set of subsystem state objects that we want to see in the
893 * new css_set. while subsystems can change globally, the entries here
894 * won't change, so no need for locking.
896 for_each_subsys(ss
, i
) {
897 if (root
->subsys_mask
& (1UL << i
)) {
899 * @ss is in this hierarchy, so we want the
900 * effective css from @cgrp.
902 template[i
] = cgroup_e_css(cgrp
, ss
);
905 * @ss is not in this hierarchy, so we don't want
908 template[i
] = old_cset
->subsys
[i
];
912 key
= css_set_hash(template);
913 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
914 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
917 /* This css_set matches what we need */
921 /* No existing cgroup group matched */
925 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
927 struct cgrp_cset_link
*link
, *tmp_link
;
929 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
930 list_del(&link
->cset_link
);
936 * allocate_cgrp_cset_links - allocate cgrp_cset_links
937 * @count: the number of links to allocate
938 * @tmp_links: list_head the allocated links are put on
940 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
941 * through ->cset_link. Returns 0 on success or -errno.
943 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
945 struct cgrp_cset_link
*link
;
948 INIT_LIST_HEAD(tmp_links
);
950 for (i
= 0; i
< count
; i
++) {
951 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
953 free_cgrp_cset_links(tmp_links
);
956 list_add(&link
->cset_link
, tmp_links
);
962 * link_css_set - a helper function to link a css_set to a cgroup
963 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
964 * @cset: the css_set to be linked
965 * @cgrp: the destination cgroup
967 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
970 struct cgrp_cset_link
*link
;
972 BUG_ON(list_empty(tmp_links
));
974 if (cgroup_on_dfl(cgrp
))
975 cset
->dfl_cgrp
= cgrp
;
977 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
982 * Always add links to the tail of the lists so that the lists are
983 * in choronological order.
985 list_move_tail(&link
->cset_link
, &cgrp
->cset_links
);
986 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
988 if (cgroup_parent(cgrp
))
993 * find_css_set - return a new css_set with one cgroup updated
994 * @old_cset: the baseline css_set
995 * @cgrp: the cgroup to be updated
997 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
998 * substituted into the appropriate hierarchy.
1000 static struct css_set
*find_css_set(struct css_set
*old_cset
,
1001 struct cgroup
*cgrp
)
1003 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
1004 struct css_set
*cset
;
1005 struct list_head tmp_links
;
1006 struct cgrp_cset_link
*link
;
1007 struct cgroup_subsys
*ss
;
1011 lockdep_assert_held(&cgroup_mutex
);
1013 /* First see if we already have a cgroup group that matches
1014 * the desired set */
1015 spin_lock_bh(&css_set_lock
);
1016 cset
= find_existing_css_set(old_cset
, cgrp
, template);
1019 spin_unlock_bh(&css_set_lock
);
1024 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
1028 /* Allocate all the cgrp_cset_link objects that we'll need */
1029 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
1034 atomic_set(&cset
->refcount
, 1);
1035 INIT_LIST_HEAD(&cset
->cgrp_links
);
1036 INIT_LIST_HEAD(&cset
->tasks
);
1037 INIT_LIST_HEAD(&cset
->mg_tasks
);
1038 INIT_LIST_HEAD(&cset
->mg_preload_node
);
1039 INIT_LIST_HEAD(&cset
->mg_node
);
1040 INIT_LIST_HEAD(&cset
->task_iters
);
1041 INIT_HLIST_NODE(&cset
->hlist
);
1043 /* Copy the set of subsystem state objects generated in
1044 * find_existing_css_set() */
1045 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
1047 spin_lock_bh(&css_set_lock
);
1048 /* Add reference counts and links from the new css_set. */
1049 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
1050 struct cgroup
*c
= link
->cgrp
;
1052 if (c
->root
== cgrp
->root
)
1054 link_css_set(&tmp_links
, cset
, c
);
1057 BUG_ON(!list_empty(&tmp_links
));
1061 /* Add @cset to the hash table */
1062 key
= css_set_hash(cset
->subsys
);
1063 hash_add(css_set_table
, &cset
->hlist
, key
);
1065 for_each_subsys(ss
, ssid
) {
1066 struct cgroup_subsys_state
*css
= cset
->subsys
[ssid
];
1068 list_add_tail(&cset
->e_cset_node
[ssid
],
1069 &css
->cgroup
->e_csets
[ssid
]);
1073 spin_unlock_bh(&css_set_lock
);
1078 static struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
1080 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
1082 return root_cgrp
->root
;
1085 static int cgroup_init_root_id(struct cgroup_root
*root
)
1089 lockdep_assert_held(&cgroup_mutex
);
1091 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
1095 root
->hierarchy_id
= id
;
1099 static void cgroup_exit_root_id(struct cgroup_root
*root
)
1101 lockdep_assert_held(&cgroup_mutex
);
1103 if (root
->hierarchy_id
) {
1104 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1105 root
->hierarchy_id
= 0;
1109 static void cgroup_free_root(struct cgroup_root
*root
)
1112 /* hierarchy ID should already have been released */
1113 WARN_ON_ONCE(root
->hierarchy_id
);
1115 idr_destroy(&root
->cgroup_idr
);
1120 static void cgroup_destroy_root(struct cgroup_root
*root
)
1122 struct cgroup
*cgrp
= &root
->cgrp
;
1123 struct cgrp_cset_link
*link
, *tmp_link
;
1125 mutex_lock(&cgroup_mutex
);
1127 BUG_ON(atomic_read(&root
->nr_cgrps
));
1128 BUG_ON(!list_empty(&cgrp
->self
.children
));
1130 /* Rebind all subsystems back to the default hierarchy */
1131 rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
);
1134 * Release all the links from cset_links to this hierarchy's
1137 spin_lock_bh(&css_set_lock
);
1139 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1140 list_del(&link
->cset_link
);
1141 list_del(&link
->cgrp_link
);
1145 spin_unlock_bh(&css_set_lock
);
1147 if (!list_empty(&root
->root_list
)) {
1148 list_del(&root
->root_list
);
1149 cgroup_root_count
--;
1152 cgroup_exit_root_id(root
);
1154 mutex_unlock(&cgroup_mutex
);
1156 kernfs_destroy_root(root
->kf_root
);
1157 cgroup_free_root(root
);
1160 /* look up cgroup associated with given css_set on the specified hierarchy */
1161 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
1162 struct cgroup_root
*root
)
1164 struct cgroup
*res
= NULL
;
1166 lockdep_assert_held(&cgroup_mutex
);
1167 lockdep_assert_held(&css_set_lock
);
1169 if (cset
== &init_css_set
) {
1172 struct cgrp_cset_link
*link
;
1174 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1175 struct cgroup
*c
= link
->cgrp
;
1177 if (c
->root
== root
) {
1189 * Return the cgroup for "task" from the given hierarchy. Must be
1190 * called with cgroup_mutex and css_set_lock held.
1192 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
1193 struct cgroup_root
*root
)
1196 * No need to lock the task - since we hold cgroup_mutex the
1197 * task can't change groups, so the only thing that can happen
1198 * is that it exits and its css is set back to init_css_set.
1200 return cset_cgroup_from_root(task_css_set(task
), root
);
1204 * A task must hold cgroup_mutex to modify cgroups.
1206 * Any task can increment and decrement the count field without lock.
1207 * So in general, code holding cgroup_mutex can't rely on the count
1208 * field not changing. However, if the count goes to zero, then only
1209 * cgroup_attach_task() can increment it again. Because a count of zero
1210 * means that no tasks are currently attached, therefore there is no
1211 * way a task attached to that cgroup can fork (the other way to
1212 * increment the count). So code holding cgroup_mutex can safely
1213 * assume that if the count is zero, it will stay zero. Similarly, if
1214 * a task holds cgroup_mutex on a cgroup with zero count, it
1215 * knows that the cgroup won't be removed, as cgroup_rmdir()
1218 * A cgroup can only be deleted if both its 'count' of using tasks
1219 * is zero, and its list of 'children' cgroups is empty. Since all
1220 * tasks in the system use _some_ cgroup, and since there is always at
1221 * least one task in the system (init, pid == 1), therefore, root cgroup
1222 * always has either children cgroups and/or using tasks. So we don't
1223 * need a special hack to ensure that root cgroup cannot be deleted.
1225 * P.S. One more locking exception. RCU is used to guard the
1226 * update of a tasks cgroup pointer by cgroup_attach_task()
1229 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
1230 static const struct file_operations proc_cgroupstats_operations
;
1232 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1235 struct cgroup_subsys
*ss
= cft
->ss
;
1237 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1238 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
))
1239 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s.%s",
1240 cgroup_on_dfl(cgrp
) ? ss
->name
: ss
->legacy_name
,
1243 strncpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1248 * cgroup_file_mode - deduce file mode of a control file
1249 * @cft: the control file in question
1251 * S_IRUGO for read, S_IWUSR for write.
1253 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1257 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1260 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
) {
1261 if (cft
->flags
& CFTYPE_WORLD_WRITABLE
)
1271 * cgroup_calc_child_subsys_mask - calculate child_subsys_mask
1272 * @cgrp: the target cgroup
1273 * @subtree_control: the new subtree_control mask to consider
1275 * On the default hierarchy, a subsystem may request other subsystems to be
1276 * enabled together through its ->depends_on mask. In such cases, more
1277 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1279 * This function calculates which subsystems need to be enabled if
1280 * @subtree_control is to be applied to @cgrp. The returned mask is always
1281 * a superset of @subtree_control and follows the usual hierarchy rules.
1283 static unsigned long cgroup_calc_child_subsys_mask(struct cgroup
*cgrp
,
1284 unsigned long subtree_control
)
1286 struct cgroup
*parent
= cgroup_parent(cgrp
);
1287 unsigned long cur_ss_mask
= subtree_control
;
1288 struct cgroup_subsys
*ss
;
1291 lockdep_assert_held(&cgroup_mutex
);
1293 if (!cgroup_on_dfl(cgrp
))
1297 unsigned long new_ss_mask
= cur_ss_mask
;
1299 for_each_subsys_which(ss
, ssid
, &cur_ss_mask
)
1300 new_ss_mask
|= ss
->depends_on
;
1303 * Mask out subsystems which aren't available. This can
1304 * happen only if some depended-upon subsystems were bound
1305 * to non-default hierarchies.
1308 new_ss_mask
&= parent
->child_subsys_mask
;
1310 new_ss_mask
&= cgrp
->root
->subsys_mask
;
1312 if (new_ss_mask
== cur_ss_mask
)
1314 cur_ss_mask
= new_ss_mask
;
1321 * cgroup_refresh_child_subsys_mask - update child_subsys_mask
1322 * @cgrp: the target cgroup
1324 * Update @cgrp->child_subsys_mask according to the current
1325 * @cgrp->subtree_control using cgroup_calc_child_subsys_mask().
1327 static void cgroup_refresh_child_subsys_mask(struct cgroup
*cgrp
)
1329 cgrp
->child_subsys_mask
=
1330 cgroup_calc_child_subsys_mask(cgrp
, cgrp
->subtree_control
);
1334 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1335 * @kn: the kernfs_node being serviced
1337 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1338 * the method finishes if locking succeeded. Note that once this function
1339 * returns the cgroup returned by cgroup_kn_lock_live() may become
1340 * inaccessible any time. If the caller intends to continue to access the
1341 * cgroup, it should pin it before invoking this function.
1343 static void cgroup_kn_unlock(struct kernfs_node
*kn
)
1345 struct cgroup
*cgrp
;
1347 if (kernfs_type(kn
) == KERNFS_DIR
)
1350 cgrp
= kn
->parent
->priv
;
1352 mutex_unlock(&cgroup_mutex
);
1354 kernfs_unbreak_active_protection(kn
);
1359 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1360 * @kn: the kernfs_node being serviced
1362 * This helper is to be used by a cgroup kernfs method currently servicing
1363 * @kn. It breaks the active protection, performs cgroup locking and
1364 * verifies that the associated cgroup is alive. Returns the cgroup if
1365 * alive; otherwise, %NULL. A successful return should be undone by a
1366 * matching cgroup_kn_unlock() invocation.
1368 * Any cgroup kernfs method implementation which requires locking the
1369 * associated cgroup should use this helper. It avoids nesting cgroup
1370 * locking under kernfs active protection and allows all kernfs operations
1371 * including self-removal.
1373 static struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
)
1375 struct cgroup
*cgrp
;
1377 if (kernfs_type(kn
) == KERNFS_DIR
)
1380 cgrp
= kn
->parent
->priv
;
1383 * We're gonna grab cgroup_mutex which nests outside kernfs
1384 * active_ref. cgroup liveliness check alone provides enough
1385 * protection against removal. Ensure @cgrp stays accessible and
1386 * break the active_ref protection.
1388 if (!cgroup_tryget(cgrp
))
1390 kernfs_break_active_protection(kn
);
1392 mutex_lock(&cgroup_mutex
);
1394 if (!cgroup_is_dead(cgrp
))
1397 cgroup_kn_unlock(kn
);
1401 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1403 char name
[CGROUP_FILE_NAME_MAX
];
1405 lockdep_assert_held(&cgroup_mutex
);
1407 if (cft
->file_offset
) {
1408 struct cgroup_subsys_state
*css
= cgroup_css(cgrp
, cft
->ss
);
1409 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
1411 spin_lock_irq(&cgroup_file_kn_lock
);
1413 spin_unlock_irq(&cgroup_file_kn_lock
);
1416 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1420 * css_clear_dir - remove subsys files in a cgroup directory
1422 * @cgrp_override: specify if target cgroup is different from css->cgroup
1424 static void css_clear_dir(struct cgroup_subsys_state
*css
,
1425 struct cgroup
*cgrp_override
)
1427 struct cgroup
*cgrp
= cgrp_override
?: css
->cgroup
;
1428 struct cftype
*cfts
;
1430 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
)
1431 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1435 * css_populate_dir - create subsys files in a cgroup directory
1437 * @cgrp_overried: specify if target cgroup is different from css->cgroup
1439 * On failure, no file is added.
1441 static int css_populate_dir(struct cgroup_subsys_state
*css
,
1442 struct cgroup
*cgrp_override
)
1444 struct cgroup
*cgrp
= cgrp_override
?: css
->cgroup
;
1445 struct cftype
*cfts
, *failed_cfts
;
1449 if (cgroup_on_dfl(cgrp
))
1450 cfts
= cgroup_dfl_base_files
;
1452 cfts
= cgroup_legacy_base_files
;
1454 return cgroup_addrm_files(&cgrp
->self
, cgrp
, cfts
, true);
1457 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1458 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, true);
1466 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1467 if (cfts
== failed_cfts
)
1469 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1474 static int rebind_subsystems(struct cgroup_root
*dst_root
,
1475 unsigned long ss_mask
)
1477 struct cgroup
*dcgrp
= &dst_root
->cgrp
;
1478 struct cgroup_subsys
*ss
;
1479 unsigned long tmp_ss_mask
;
1482 lockdep_assert_held(&cgroup_mutex
);
1484 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
1485 /* if @ss has non-root csses attached to it, can't move */
1486 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)))
1489 /* can't move between two non-dummy roots either */
1490 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1494 /* skip creating root files on dfl_root for inhibited subsystems */
1495 tmp_ss_mask
= ss_mask
;
1496 if (dst_root
== &cgrp_dfl_root
)
1497 tmp_ss_mask
&= ~cgrp_dfl_root_inhibit_ss_mask
;
1499 for_each_subsys_which(ss
, ssid
, &tmp_ss_mask
) {
1500 struct cgroup
*scgrp
= &ss
->root
->cgrp
;
1503 ret
= css_populate_dir(cgroup_css(scgrp
, ss
), dcgrp
);
1508 * Rebinding back to the default root is not allowed to
1509 * fail. Using both default and non-default roots should
1510 * be rare. Moving subsystems back and forth even more so.
1511 * Just warn about it and continue.
1513 if (dst_root
== &cgrp_dfl_root
) {
1514 if (cgrp_dfl_root_visible
) {
1515 pr_warn("failed to create files (%d) while rebinding 0x%lx to default root\n",
1517 pr_warn("you may retry by moving them to a different hierarchy and unbinding\n");
1522 for_each_subsys_which(ss
, tssid
, &tmp_ss_mask
) {
1525 css_clear_dir(cgroup_css(scgrp
, ss
), dcgrp
);
1531 * Nothing can fail from this point on. Remove files for the
1532 * removed subsystems and rebind each subsystem.
1534 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
1535 struct cgroup_root
*src_root
= ss
->root
;
1536 struct cgroup
*scgrp
= &src_root
->cgrp
;
1537 struct cgroup_subsys_state
*css
= cgroup_css(scgrp
, ss
);
1538 struct css_set
*cset
;
1540 WARN_ON(!css
|| cgroup_css(dcgrp
, ss
));
1542 css_clear_dir(css
, NULL
);
1544 RCU_INIT_POINTER(scgrp
->subsys
[ssid
], NULL
);
1545 rcu_assign_pointer(dcgrp
->subsys
[ssid
], css
);
1546 ss
->root
= dst_root
;
1547 css
->cgroup
= dcgrp
;
1549 spin_lock_bh(&css_set_lock
);
1550 hash_for_each(css_set_table
, i
, cset
, hlist
)
1551 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1552 &dcgrp
->e_csets
[ss
->id
]);
1553 spin_unlock_bh(&css_set_lock
);
1555 src_root
->subsys_mask
&= ~(1 << ssid
);
1556 scgrp
->subtree_control
&= ~(1 << ssid
);
1557 cgroup_refresh_child_subsys_mask(scgrp
);
1559 /* default hierarchy doesn't enable controllers by default */
1560 dst_root
->subsys_mask
|= 1 << ssid
;
1561 if (dst_root
== &cgrp_dfl_root
) {
1562 static_branch_enable(cgroup_subsys_on_dfl_key
[ssid
]);
1564 dcgrp
->subtree_control
|= 1 << ssid
;
1565 cgroup_refresh_child_subsys_mask(dcgrp
);
1566 static_branch_disable(cgroup_subsys_on_dfl_key
[ssid
]);
1573 kernfs_activate(dcgrp
->kn
);
1577 static int cgroup_show_options(struct seq_file
*seq
,
1578 struct kernfs_root
*kf_root
)
1580 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1581 struct cgroup_subsys
*ss
;
1584 if (root
!= &cgrp_dfl_root
)
1585 for_each_subsys(ss
, ssid
)
1586 if (root
->subsys_mask
& (1 << ssid
))
1587 seq_show_option(seq
, ss
->legacy_name
, NULL
);
1588 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1589 seq_puts(seq
, ",noprefix");
1590 if (root
->flags
& CGRP_ROOT_XATTR
)
1591 seq_puts(seq
, ",xattr");
1593 spin_lock(&release_agent_path_lock
);
1594 if (strlen(root
->release_agent_path
))
1595 seq_show_option(seq
, "release_agent",
1596 root
->release_agent_path
);
1597 spin_unlock(&release_agent_path_lock
);
1599 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
1600 seq_puts(seq
, ",clone_children");
1601 if (strlen(root
->name
))
1602 seq_show_option(seq
, "name", root
->name
);
1606 struct cgroup_sb_opts
{
1607 unsigned long subsys_mask
;
1609 char *release_agent
;
1610 bool cpuset_clone_children
;
1612 /* User explicitly requested empty subsystem */
1616 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1618 char *token
, *o
= data
;
1619 bool all_ss
= false, one_ss
= false;
1620 unsigned long mask
= -1UL;
1621 struct cgroup_subsys
*ss
;
1625 #ifdef CONFIG_CPUSETS
1626 mask
= ~(1U << cpuset_cgrp_id
);
1629 memset(opts
, 0, sizeof(*opts
));
1631 while ((token
= strsep(&o
, ",")) != NULL
) {
1636 if (!strcmp(token
, "none")) {
1637 /* Explicitly have no subsystems */
1641 if (!strcmp(token
, "all")) {
1642 /* Mutually exclusive option 'all' + subsystem name */
1648 if (!strcmp(token
, "noprefix")) {
1649 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1652 if (!strcmp(token
, "clone_children")) {
1653 opts
->cpuset_clone_children
= true;
1656 if (!strcmp(token
, "xattr")) {
1657 opts
->flags
|= CGRP_ROOT_XATTR
;
1660 if (!strncmp(token
, "release_agent=", 14)) {
1661 /* Specifying two release agents is forbidden */
1662 if (opts
->release_agent
)
1664 opts
->release_agent
=
1665 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1666 if (!opts
->release_agent
)
1670 if (!strncmp(token
, "name=", 5)) {
1671 const char *name
= token
+ 5;
1672 /* Can't specify an empty name */
1675 /* Must match [\w.-]+ */
1676 for (i
= 0; i
< strlen(name
); i
++) {
1680 if ((c
== '.') || (c
== '-') || (c
== '_'))
1684 /* Specifying two names is forbidden */
1687 opts
->name
= kstrndup(name
,
1688 MAX_CGROUP_ROOT_NAMELEN
- 1,
1696 for_each_subsys(ss
, i
) {
1697 if (strcmp(token
, ss
->legacy_name
))
1699 if (!cgroup_ssid_enabled(i
))
1701 if (cgroup_ssid_no_v1(i
))
1704 /* Mutually exclusive option 'all' + subsystem name */
1707 opts
->subsys_mask
|= (1 << i
);
1712 if (i
== CGROUP_SUBSYS_COUNT
)
1717 * If the 'all' option was specified select all the subsystems,
1718 * otherwise if 'none', 'name=' and a subsystem name options were
1719 * not specified, let's default to 'all'
1721 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1722 for_each_subsys(ss
, i
)
1723 if (cgroup_ssid_enabled(i
) && !cgroup_ssid_no_v1(i
))
1724 opts
->subsys_mask
|= (1 << i
);
1727 * We either have to specify by name or by subsystems. (So all
1728 * empty hierarchies must have a name).
1730 if (!opts
->subsys_mask
&& !opts
->name
)
1734 * Option noprefix was introduced just for backward compatibility
1735 * with the old cpuset, so we allow noprefix only if mounting just
1736 * the cpuset subsystem.
1738 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1741 /* Can't specify "none" and some subsystems */
1742 if (opts
->subsys_mask
&& opts
->none
)
1748 static int cgroup_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1751 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1752 struct cgroup_sb_opts opts
;
1753 unsigned long added_mask
, removed_mask
;
1755 if (root
== &cgrp_dfl_root
) {
1756 pr_err("remount is not allowed\n");
1760 mutex_lock(&cgroup_mutex
);
1762 /* See what subsystems are wanted */
1763 ret
= parse_cgroupfs_options(data
, &opts
);
1767 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1768 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1769 task_tgid_nr(current
), current
->comm
);
1771 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1772 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1774 /* Don't allow flags or name to change at remount */
1775 if ((opts
.flags
^ root
->flags
) ||
1776 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1777 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1778 opts
.flags
, opts
.name
?: "", root
->flags
, root
->name
);
1783 /* remounting is not allowed for populated hierarchies */
1784 if (!list_empty(&root
->cgrp
.self
.children
)) {
1789 ret
= rebind_subsystems(root
, added_mask
);
1793 rebind_subsystems(&cgrp_dfl_root
, removed_mask
);
1795 if (opts
.release_agent
) {
1796 spin_lock(&release_agent_path_lock
);
1797 strcpy(root
->release_agent_path
, opts
.release_agent
);
1798 spin_unlock(&release_agent_path_lock
);
1801 kfree(opts
.release_agent
);
1803 mutex_unlock(&cgroup_mutex
);
1808 * To reduce the fork() overhead for systems that are not actually using
1809 * their cgroups capability, we don't maintain the lists running through
1810 * each css_set to its tasks until we see the list actually used - in other
1811 * words after the first mount.
1813 static bool use_task_css_set_links __read_mostly
;
1815 static void cgroup_enable_task_cg_lists(void)
1817 struct task_struct
*p
, *g
;
1819 spin_lock_bh(&css_set_lock
);
1821 if (use_task_css_set_links
)
1824 use_task_css_set_links
= true;
1827 * We need tasklist_lock because RCU is not safe against
1828 * while_each_thread(). Besides, a forking task that has passed
1829 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1830 * is not guaranteed to have its child immediately visible in the
1831 * tasklist if we walk through it with RCU.
1833 read_lock(&tasklist_lock
);
1834 do_each_thread(g
, p
) {
1835 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1836 task_css_set(p
) != &init_css_set
);
1839 * We should check if the process is exiting, otherwise
1840 * it will race with cgroup_exit() in that the list
1841 * entry won't be deleted though the process has exited.
1842 * Do it while holding siglock so that we don't end up
1843 * racing against cgroup_exit().
1845 spin_lock_irq(&p
->sighand
->siglock
);
1846 if (!(p
->flags
& PF_EXITING
)) {
1847 struct css_set
*cset
= task_css_set(p
);
1849 if (!css_set_populated(cset
))
1850 css_set_update_populated(cset
, true);
1851 list_add_tail(&p
->cg_list
, &cset
->tasks
);
1854 spin_unlock_irq(&p
->sighand
->siglock
);
1855 } while_each_thread(g
, p
);
1856 read_unlock(&tasklist_lock
);
1858 spin_unlock_bh(&css_set_lock
);
1861 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1863 struct cgroup_subsys
*ss
;
1866 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1867 INIT_LIST_HEAD(&cgrp
->self
.children
);
1868 INIT_LIST_HEAD(&cgrp
->cset_links
);
1869 INIT_LIST_HEAD(&cgrp
->pidlists
);
1870 mutex_init(&cgrp
->pidlist_mutex
);
1871 cgrp
->self
.cgroup
= cgrp
;
1872 cgrp
->self
.flags
|= CSS_ONLINE
;
1874 for_each_subsys(ss
, ssid
)
1875 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1877 init_waitqueue_head(&cgrp
->offline_waitq
);
1878 INIT_WORK(&cgrp
->release_agent_work
, cgroup_release_agent
);
1881 static void init_cgroup_root(struct cgroup_root
*root
,
1882 struct cgroup_sb_opts
*opts
)
1884 struct cgroup
*cgrp
= &root
->cgrp
;
1886 INIT_LIST_HEAD(&root
->root_list
);
1887 atomic_set(&root
->nr_cgrps
, 1);
1889 init_cgroup_housekeeping(cgrp
);
1890 idr_init(&root
->cgroup_idr
);
1892 root
->flags
= opts
->flags
;
1893 if (opts
->release_agent
)
1894 strcpy(root
->release_agent_path
, opts
->release_agent
);
1896 strcpy(root
->name
, opts
->name
);
1897 if (opts
->cpuset_clone_children
)
1898 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1901 static int cgroup_setup_root(struct cgroup_root
*root
, unsigned long ss_mask
)
1903 LIST_HEAD(tmp_links
);
1904 struct cgroup
*root_cgrp
= &root
->cgrp
;
1905 struct css_set
*cset
;
1908 lockdep_assert_held(&cgroup_mutex
);
1910 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_KERNEL
);
1913 root_cgrp
->id
= ret
;
1914 root_cgrp
->ancestor_ids
[0] = ret
;
1916 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
, 0,
1922 * We're accessing css_set_count without locking css_set_lock here,
1923 * but that's OK - it can only be increased by someone holding
1924 * cgroup_lock, and that's us. The worst that can happen is that we
1925 * have some link structures left over
1927 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1931 ret
= cgroup_init_root_id(root
);
1935 root
->kf_root
= kernfs_create_root(&cgroup_kf_syscall_ops
,
1936 KERNFS_ROOT_CREATE_DEACTIVATED
,
1938 if (IS_ERR(root
->kf_root
)) {
1939 ret
= PTR_ERR(root
->kf_root
);
1942 root_cgrp
->kn
= root
->kf_root
->kn
;
1944 ret
= css_populate_dir(&root_cgrp
->self
, NULL
);
1948 ret
= rebind_subsystems(root
, ss_mask
);
1953 * There must be no failure case after here, since rebinding takes
1954 * care of subsystems' refcounts, which are explicitly dropped in
1955 * the failure exit path.
1957 list_add(&root
->root_list
, &cgroup_roots
);
1958 cgroup_root_count
++;
1961 * Link the root cgroup in this hierarchy into all the css_set
1964 spin_lock_bh(&css_set_lock
);
1965 hash_for_each(css_set_table
, i
, cset
, hlist
) {
1966 link_css_set(&tmp_links
, cset
, root_cgrp
);
1967 if (css_set_populated(cset
))
1968 cgroup_update_populated(root_cgrp
, true);
1970 spin_unlock_bh(&css_set_lock
);
1972 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
1973 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
1975 kernfs_activate(root_cgrp
->kn
);
1980 kernfs_destroy_root(root
->kf_root
);
1981 root
->kf_root
= NULL
;
1983 cgroup_exit_root_id(root
);
1985 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
1987 free_cgrp_cset_links(&tmp_links
);
1991 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1992 int flags
, const char *unused_dev_name
,
1995 bool is_v2
= fs_type
== &cgroup2_fs_type
;
1996 struct super_block
*pinned_sb
= NULL
;
1997 struct cgroup_namespace
*ns
= current
->nsproxy
->cgroup_ns
;
1998 struct cgroup_subsys
*ss
;
1999 struct cgroup_root
*root
;
2000 struct cgroup_sb_opts opts
;
2001 struct dentry
*dentry
;
2008 /* Check if the caller has permission to mount. */
2009 if (!ns_capable(ns
->user_ns
, CAP_SYS_ADMIN
)) {
2011 return ERR_PTR(-EPERM
);
2015 * The first time anyone tries to mount a cgroup, enable the list
2016 * linking each css_set to its tasks and fix up all existing tasks.
2018 if (!use_task_css_set_links
)
2019 cgroup_enable_task_cg_lists();
2023 pr_err("cgroup2: unknown option \"%s\"\n", (char *)data
);
2025 return ERR_PTR(-EINVAL
);
2027 cgrp_dfl_root_visible
= true;
2028 root
= &cgrp_dfl_root
;
2029 cgroup_get(&root
->cgrp
);
2033 mutex_lock(&cgroup_mutex
);
2035 /* First find the desired set of subsystems */
2036 ret
= parse_cgroupfs_options(data
, &opts
);
2041 * Destruction of cgroup root is asynchronous, so subsystems may
2042 * still be dying after the previous unmount. Let's drain the
2043 * dying subsystems. We just need to ensure that the ones
2044 * unmounted previously finish dying and don't care about new ones
2045 * starting. Testing ref liveliness is good enough.
2047 for_each_subsys(ss
, i
) {
2048 if (!(opts
.subsys_mask
& (1 << i
)) ||
2049 ss
->root
== &cgrp_dfl_root
)
2052 if (!percpu_ref_tryget_live(&ss
->root
->cgrp
.self
.refcnt
)) {
2053 mutex_unlock(&cgroup_mutex
);
2055 ret
= restart_syscall();
2058 cgroup_put(&ss
->root
->cgrp
);
2061 for_each_root(root
) {
2062 bool name_match
= false;
2064 if (root
== &cgrp_dfl_root
)
2068 * If we asked for a name then it must match. Also, if
2069 * name matches but sybsys_mask doesn't, we should fail.
2070 * Remember whether name matched.
2073 if (strcmp(opts
.name
, root
->name
))
2079 * If we asked for subsystems (or explicitly for no
2080 * subsystems) then they must match.
2082 if ((opts
.subsys_mask
|| opts
.none
) &&
2083 (opts
.subsys_mask
!= root
->subsys_mask
)) {
2090 if (root
->flags
^ opts
.flags
)
2091 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
2094 * We want to reuse @root whose lifetime is governed by its
2095 * ->cgrp. Let's check whether @root is alive and keep it
2096 * that way. As cgroup_kill_sb() can happen anytime, we
2097 * want to block it by pinning the sb so that @root doesn't
2098 * get killed before mount is complete.
2100 * With the sb pinned, tryget_live can reliably indicate
2101 * whether @root can be reused. If it's being killed,
2102 * drain it. We can use wait_queue for the wait but this
2103 * path is super cold. Let's just sleep a bit and retry.
2105 pinned_sb
= kernfs_pin_sb(root
->kf_root
, NULL
);
2106 if (IS_ERR(pinned_sb
) ||
2107 !percpu_ref_tryget_live(&root
->cgrp
.self
.refcnt
)) {
2108 mutex_unlock(&cgroup_mutex
);
2109 if (!IS_ERR_OR_NULL(pinned_sb
))
2110 deactivate_super(pinned_sb
);
2112 ret
= restart_syscall();
2121 * No such thing, create a new one. name= matching without subsys
2122 * specification is allowed for already existing hierarchies but we
2123 * can't create new one without subsys specification.
2125 if (!opts
.subsys_mask
&& !opts
.none
) {
2131 * We know this subsystem has not yet been bound. Users in a non-init
2132 * user namespace may only mount hierarchies with no bound subsystems,
2133 * i.e. 'none,name=user1'
2135 if (!opts
.none
&& !capable(CAP_SYS_ADMIN
)) {
2140 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
2146 init_cgroup_root(root
, &opts
);
2148 ret
= cgroup_setup_root(root
, opts
.subsys_mask
);
2150 cgroup_free_root(root
);
2153 mutex_unlock(&cgroup_mutex
);
2155 kfree(opts
.release_agent
);
2160 return ERR_PTR(ret
);
2163 dentry
= kernfs_mount(fs_type
, flags
, root
->kf_root
,
2164 is_v2
? CGROUP2_SUPER_MAGIC
: CGROUP_SUPER_MAGIC
,
2168 * In non-init cgroup namespace, instead of root cgroup's
2169 * dentry, we return the dentry corresponding to the
2170 * cgroupns->root_cgrp.
2172 if (!IS_ERR(dentry
) && ns
!= &init_cgroup_ns
) {
2173 struct dentry
*nsdentry
;
2174 struct cgroup
*cgrp
;
2176 mutex_lock(&cgroup_mutex
);
2177 spin_lock_bh(&css_set_lock
);
2179 cgrp
= cset_cgroup_from_root(ns
->root_cset
, root
);
2181 spin_unlock_bh(&css_set_lock
);
2182 mutex_unlock(&cgroup_mutex
);
2184 nsdentry
= kernfs_node_dentry(cgrp
->kn
, dentry
->d_sb
);
2189 if (IS_ERR(dentry
) || !new_sb
)
2190 cgroup_put(&root
->cgrp
);
2193 * If @pinned_sb, we're reusing an existing root and holding an
2194 * extra ref on its sb. Mount is complete. Put the extra ref.
2198 deactivate_super(pinned_sb
);
2205 static void cgroup_kill_sb(struct super_block
*sb
)
2207 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
2208 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
2211 * If @root doesn't have any mounts or children, start killing it.
2212 * This prevents new mounts by disabling percpu_ref_tryget_live().
2213 * cgroup_mount() may wait for @root's release.
2215 * And don't kill the default root.
2217 if (!list_empty(&root
->cgrp
.self
.children
) ||
2218 root
== &cgrp_dfl_root
)
2219 cgroup_put(&root
->cgrp
);
2221 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
2226 static struct file_system_type cgroup_fs_type
= {
2228 .mount
= cgroup_mount
,
2229 .kill_sb
= cgroup_kill_sb
,
2232 static struct file_system_type cgroup2_fs_type
= {
2234 .mount
= cgroup_mount
,
2235 .kill_sb
= cgroup_kill_sb
,
2238 static char *cgroup_path_ns_locked(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2239 struct cgroup_namespace
*ns
)
2241 struct cgroup
*root
= cset_cgroup_from_root(ns
->root_cset
, cgrp
->root
);
2244 ret
= kernfs_path_from_node(cgrp
->kn
, root
->kn
, buf
, buflen
);
2245 if (ret
< 0 || ret
>= buflen
)
2250 char *cgroup_path_ns(struct cgroup
*cgrp
, char *buf
, size_t buflen
,
2251 struct cgroup_namespace
*ns
)
2255 mutex_lock(&cgroup_mutex
);
2256 spin_lock_bh(&css_set_lock
);
2258 ret
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, ns
);
2260 spin_unlock_bh(&css_set_lock
);
2261 mutex_unlock(&cgroup_mutex
);
2265 EXPORT_SYMBOL_GPL(cgroup_path_ns
);
2268 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2269 * @task: target task
2270 * @buf: the buffer to write the path into
2271 * @buflen: the length of the buffer
2273 * Determine @task's cgroup on the first (the one with the lowest non-zero
2274 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2275 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2276 * cgroup controller callbacks.
2278 * Return value is the same as kernfs_path().
2280 char *task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
2282 struct cgroup_root
*root
;
2283 struct cgroup
*cgrp
;
2284 int hierarchy_id
= 1;
2287 mutex_lock(&cgroup_mutex
);
2288 spin_lock_bh(&css_set_lock
);
2290 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
2293 cgrp
= task_cgroup_from_root(task
, root
);
2294 path
= cgroup_path_ns_locked(cgrp
, buf
, buflen
, &init_cgroup_ns
);
2296 /* if no hierarchy exists, everyone is in "/" */
2297 if (strlcpy(buf
, "/", buflen
) < buflen
)
2301 spin_unlock_bh(&css_set_lock
);
2302 mutex_unlock(&cgroup_mutex
);
2305 EXPORT_SYMBOL_GPL(task_cgroup_path
);
2307 /* used to track tasks and other necessary states during migration */
2308 struct cgroup_taskset
{
2309 /* the src and dst cset list running through cset->mg_node */
2310 struct list_head src_csets
;
2311 struct list_head dst_csets
;
2313 /* the subsys currently being processed */
2317 * Fields for cgroup_taskset_*() iteration.
2319 * Before migration is committed, the target migration tasks are on
2320 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
2321 * the csets on ->dst_csets. ->csets point to either ->src_csets
2322 * or ->dst_csets depending on whether migration is committed.
2324 * ->cur_csets and ->cur_task point to the current task position
2327 struct list_head
*csets
;
2328 struct css_set
*cur_cset
;
2329 struct task_struct
*cur_task
;
2332 #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \
2333 .src_csets = LIST_HEAD_INIT(tset.src_csets), \
2334 .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \
2335 .csets = &tset.src_csets, \
2339 * cgroup_taskset_add - try to add a migration target task to a taskset
2340 * @task: target task
2341 * @tset: target taskset
2343 * Add @task, which is a migration target, to @tset. This function becomes
2344 * noop if @task doesn't need to be migrated. @task's css_set should have
2345 * been added as a migration source and @task->cg_list will be moved from
2346 * the css_set's tasks list to mg_tasks one.
2348 static void cgroup_taskset_add(struct task_struct
*task
,
2349 struct cgroup_taskset
*tset
)
2351 struct css_set
*cset
;
2353 lockdep_assert_held(&css_set_lock
);
2355 /* @task either already exited or can't exit until the end */
2356 if (task
->flags
& PF_EXITING
)
2359 /* leave @task alone if post_fork() hasn't linked it yet */
2360 if (list_empty(&task
->cg_list
))
2363 cset
= task_css_set(task
);
2364 if (!cset
->mg_src_cgrp
)
2367 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2368 if (list_empty(&cset
->mg_node
))
2369 list_add_tail(&cset
->mg_node
, &tset
->src_csets
);
2370 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2371 list_move_tail(&cset
->mg_dst_cset
->mg_node
,
2376 * cgroup_taskset_first - reset taskset and return the first task
2377 * @tset: taskset of interest
2378 * @dst_cssp: output variable for the destination css
2380 * @tset iteration is initialized and the first task is returned.
2382 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
,
2383 struct cgroup_subsys_state
**dst_cssp
)
2385 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2386 tset
->cur_task
= NULL
;
2388 return cgroup_taskset_next(tset
, dst_cssp
);
2392 * cgroup_taskset_next - iterate to the next task in taskset
2393 * @tset: taskset of interest
2394 * @dst_cssp: output variable for the destination css
2396 * Return the next task in @tset. Iteration must have been initialized
2397 * with cgroup_taskset_first().
2399 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
,
2400 struct cgroup_subsys_state
**dst_cssp
)
2402 struct css_set
*cset
= tset
->cur_cset
;
2403 struct task_struct
*task
= tset
->cur_task
;
2405 while (&cset
->mg_node
!= tset
->csets
) {
2407 task
= list_first_entry(&cset
->mg_tasks
,
2408 struct task_struct
, cg_list
);
2410 task
= list_next_entry(task
, cg_list
);
2412 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2413 tset
->cur_cset
= cset
;
2414 tset
->cur_task
= task
;
2417 * This function may be called both before and
2418 * after cgroup_taskset_migrate(). The two cases
2419 * can be distinguished by looking at whether @cset
2420 * has its ->mg_dst_cset set.
2422 if (cset
->mg_dst_cset
)
2423 *dst_cssp
= cset
->mg_dst_cset
->subsys
[tset
->ssid
];
2425 *dst_cssp
= cset
->subsys
[tset
->ssid
];
2430 cset
= list_next_entry(cset
, mg_node
);
2438 * cgroup_taskset_migrate - migrate a taskset to a cgroup
2439 * @tset: taget taskset
2440 * @dst_cgrp: destination cgroup
2442 * Migrate tasks in @tset to @dst_cgrp. This function fails iff one of the
2443 * ->can_attach callbacks fails and guarantees that either all or none of
2444 * the tasks in @tset are migrated. @tset is consumed regardless of
2447 static int cgroup_taskset_migrate(struct cgroup_taskset
*tset
,
2448 struct cgroup
*dst_cgrp
)
2450 struct cgroup_subsys_state
*css
, *failed_css
= NULL
;
2451 struct task_struct
*task
, *tmp_task
;
2452 struct css_set
*cset
, *tmp_cset
;
2455 /* methods shouldn't be called if no task is actually migrating */
2456 if (list_empty(&tset
->src_csets
))
2459 /* check that we can legitimately attach to the cgroup */
2460 for_each_e_css(css
, i
, dst_cgrp
) {
2461 if (css
->ss
->can_attach
) {
2463 ret
= css
->ss
->can_attach(tset
);
2466 goto out_cancel_attach
;
2472 * Now that we're guaranteed success, proceed to move all tasks to
2473 * the new cgroup. There are no failure cases after here, so this
2474 * is the commit point.
2476 spin_lock_bh(&css_set_lock
);
2477 list_for_each_entry(cset
, &tset
->src_csets
, mg_node
) {
2478 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
) {
2479 struct css_set
*from_cset
= task_css_set(task
);
2480 struct css_set
*to_cset
= cset
->mg_dst_cset
;
2482 get_css_set(to_cset
);
2483 css_set_move_task(task
, from_cset
, to_cset
, true);
2484 put_css_set_locked(from_cset
);
2487 spin_unlock_bh(&css_set_lock
);
2490 * Migration is committed, all target tasks are now on dst_csets.
2491 * Nothing is sensitive to fork() after this point. Notify
2492 * controllers that migration is complete.
2494 tset
->csets
= &tset
->dst_csets
;
2496 for_each_e_css(css
, i
, dst_cgrp
) {
2497 if (css
->ss
->attach
) {
2499 css
->ss
->attach(tset
);
2504 goto out_release_tset
;
2507 for_each_e_css(css
, i
, dst_cgrp
) {
2508 if (css
== failed_css
)
2510 if (css
->ss
->cancel_attach
) {
2512 css
->ss
->cancel_attach(tset
);
2516 spin_lock_bh(&css_set_lock
);
2517 list_splice_init(&tset
->dst_csets
, &tset
->src_csets
);
2518 list_for_each_entry_safe(cset
, tmp_cset
, &tset
->src_csets
, mg_node
) {
2519 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2520 list_del_init(&cset
->mg_node
);
2522 spin_unlock_bh(&css_set_lock
);
2527 * cgroup_migrate_finish - cleanup after attach
2528 * @preloaded_csets: list of preloaded css_sets
2530 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2531 * those functions for details.
2533 static void cgroup_migrate_finish(struct list_head
*preloaded_csets
)
2535 struct css_set
*cset
, *tmp_cset
;
2537 lockdep_assert_held(&cgroup_mutex
);
2539 spin_lock_bh(&css_set_lock
);
2540 list_for_each_entry_safe(cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2541 cset
->mg_src_cgrp
= NULL
;
2542 cset
->mg_dst_cset
= NULL
;
2543 list_del_init(&cset
->mg_preload_node
);
2544 put_css_set_locked(cset
);
2546 spin_unlock_bh(&css_set_lock
);
2550 * cgroup_migrate_add_src - add a migration source css_set
2551 * @src_cset: the source css_set to add
2552 * @dst_cgrp: the destination cgroup
2553 * @preloaded_csets: list of preloaded css_sets
2555 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2556 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2557 * up by cgroup_migrate_finish().
2559 * This function may be called without holding cgroup_threadgroup_rwsem
2560 * even if the target is a process. Threads may be created and destroyed
2561 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2562 * into play and the preloaded css_sets are guaranteed to cover all
2565 static void cgroup_migrate_add_src(struct css_set
*src_cset
,
2566 struct cgroup
*dst_cgrp
,
2567 struct list_head
*preloaded_csets
)
2569 struct cgroup
*src_cgrp
;
2571 lockdep_assert_held(&cgroup_mutex
);
2572 lockdep_assert_held(&css_set_lock
);
2574 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2576 if (!list_empty(&src_cset
->mg_preload_node
))
2579 WARN_ON(src_cset
->mg_src_cgrp
);
2580 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2581 WARN_ON(!list_empty(&src_cset
->mg_node
));
2583 src_cset
->mg_src_cgrp
= src_cgrp
;
2584 get_css_set(src_cset
);
2585 list_add(&src_cset
->mg_preload_node
, preloaded_csets
);
2589 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2590 * @dst_cgrp: the destination cgroup (may be %NULL)
2591 * @preloaded_csets: list of preloaded source css_sets
2593 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
2594 * have been preloaded to @preloaded_csets. This function looks up and
2595 * pins all destination css_sets, links each to its source, and append them
2596 * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each
2597 * source css_set is assumed to be its cgroup on the default hierarchy.
2599 * This function must be called after cgroup_migrate_add_src() has been
2600 * called on each migration source css_set. After migration is performed
2601 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2604 static int cgroup_migrate_prepare_dst(struct cgroup
*dst_cgrp
,
2605 struct list_head
*preloaded_csets
)
2608 struct css_set
*src_cset
, *tmp_cset
;
2610 lockdep_assert_held(&cgroup_mutex
);
2613 * Except for the root, child_subsys_mask must be zero for a cgroup
2614 * with tasks so that child cgroups don't compete against tasks.
2616 if (dst_cgrp
&& cgroup_on_dfl(dst_cgrp
) && cgroup_parent(dst_cgrp
) &&
2617 dst_cgrp
->child_subsys_mask
)
2620 /* look up the dst cset for each src cset and link it to src */
2621 list_for_each_entry_safe(src_cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2622 struct css_set
*dst_cset
;
2624 dst_cset
= find_css_set(src_cset
,
2625 dst_cgrp
?: src_cset
->dfl_cgrp
);
2629 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2632 * If src cset equals dst, it's noop. Drop the src.
2633 * cgroup_migrate() will skip the cset too. Note that we
2634 * can't handle src == dst as some nodes are used by both.
2636 if (src_cset
== dst_cset
) {
2637 src_cset
->mg_src_cgrp
= NULL
;
2638 list_del_init(&src_cset
->mg_preload_node
);
2639 put_css_set(src_cset
);
2640 put_css_set(dst_cset
);
2644 src_cset
->mg_dst_cset
= dst_cset
;
2646 if (list_empty(&dst_cset
->mg_preload_node
))
2647 list_add(&dst_cset
->mg_preload_node
, &csets
);
2649 put_css_set(dst_cset
);
2652 list_splice_tail(&csets
, preloaded_csets
);
2655 cgroup_migrate_finish(&csets
);
2660 * cgroup_migrate - migrate a process or task to a cgroup
2661 * @leader: the leader of the process or the task to migrate
2662 * @threadgroup: whether @leader points to the whole process or a single task
2663 * @cgrp: the destination cgroup
2665 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
2666 * process, the caller must be holding cgroup_threadgroup_rwsem. The
2667 * caller is also responsible for invoking cgroup_migrate_add_src() and
2668 * cgroup_migrate_prepare_dst() on the targets before invoking this
2669 * function and following up with cgroup_migrate_finish().
2671 * As long as a controller's ->can_attach() doesn't fail, this function is
2672 * guaranteed to succeed. This means that, excluding ->can_attach()
2673 * failure, when migrating multiple targets, the success or failure can be
2674 * decided for all targets by invoking group_migrate_prepare_dst() before
2675 * actually starting migrating.
2677 static int cgroup_migrate(struct task_struct
*leader
, bool threadgroup
,
2678 struct cgroup
*cgrp
)
2680 struct cgroup_taskset tset
= CGROUP_TASKSET_INIT(tset
);
2681 struct task_struct
*task
;
2684 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2685 * already PF_EXITING could be freed from underneath us unless we
2686 * take an rcu_read_lock.
2688 spin_lock_bh(&css_set_lock
);
2692 cgroup_taskset_add(task
, &tset
);
2695 } while_each_thread(leader
, task
);
2697 spin_unlock_bh(&css_set_lock
);
2699 return cgroup_taskset_migrate(&tset
, cgrp
);
2703 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2704 * @dst_cgrp: the cgroup to attach to
2705 * @leader: the task or the leader of the threadgroup to be attached
2706 * @threadgroup: attach the whole threadgroup?
2708 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2710 static int cgroup_attach_task(struct cgroup
*dst_cgrp
,
2711 struct task_struct
*leader
, bool threadgroup
)
2713 LIST_HEAD(preloaded_csets
);
2714 struct task_struct
*task
;
2717 /* look up all src csets */
2718 spin_lock_bh(&css_set_lock
);
2722 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
,
2726 } while_each_thread(leader
, task
);
2728 spin_unlock_bh(&css_set_lock
);
2730 /* prepare dst csets and commit */
2731 ret
= cgroup_migrate_prepare_dst(dst_cgrp
, &preloaded_csets
);
2733 ret
= cgroup_migrate(leader
, threadgroup
, dst_cgrp
);
2735 cgroup_migrate_finish(&preloaded_csets
);
2739 static int cgroup_procs_write_permission(struct task_struct
*task
,
2740 struct cgroup
*dst_cgrp
,
2741 struct kernfs_open_file
*of
)
2743 const struct cred
*cred
= current_cred();
2744 const struct cred
*tcred
= get_task_cred(task
);
2748 * even if we're attaching all tasks in the thread group, we only
2749 * need to check permissions on one of them.
2751 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2752 !uid_eq(cred
->euid
, tcred
->uid
) &&
2753 !uid_eq(cred
->euid
, tcred
->suid
))
2756 if (!ret
&& cgroup_on_dfl(dst_cgrp
)) {
2757 struct super_block
*sb
= of
->file
->f_path
.dentry
->d_sb
;
2758 struct cgroup
*cgrp
;
2759 struct inode
*inode
;
2761 spin_lock_bh(&css_set_lock
);
2762 cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
2763 spin_unlock_bh(&css_set_lock
);
2765 while (!cgroup_is_descendant(dst_cgrp
, cgrp
))
2766 cgrp
= cgroup_parent(cgrp
);
2769 inode
= kernfs_get_inode(sb
, cgrp
->procs_file
.kn
);
2771 ret
= inode_permission(inode
, MAY_WRITE
);
2781 * Find the task_struct of the task to attach by vpid and pass it along to the
2782 * function to attach either it or all tasks in its threadgroup. Will lock
2783 * cgroup_mutex and threadgroup.
2785 static ssize_t
__cgroup_procs_write(struct kernfs_open_file
*of
, char *buf
,
2786 size_t nbytes
, loff_t off
, bool threadgroup
)
2788 struct task_struct
*tsk
;
2789 struct cgroup
*cgrp
;
2793 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2796 cgrp
= cgroup_kn_lock_live(of
->kn
);
2800 percpu_down_write(&cgroup_threadgroup_rwsem
);
2803 tsk
= find_task_by_vpid(pid
);
2806 goto out_unlock_rcu
;
2813 tsk
= tsk
->group_leader
;
2816 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2817 * trapped in a cpuset, or RT worker may be born in a cgroup
2818 * with no rt_runtime allocated. Just say no.
2820 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2822 goto out_unlock_rcu
;
2825 get_task_struct(tsk
);
2828 ret
= cgroup_procs_write_permission(tsk
, cgrp
, of
);
2830 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2832 put_task_struct(tsk
);
2833 goto out_unlock_threadgroup
;
2837 out_unlock_threadgroup
:
2838 percpu_up_write(&cgroup_threadgroup_rwsem
);
2839 cgroup_kn_unlock(of
->kn
);
2840 cpuset_post_attach_flush();
2841 return ret
?: nbytes
;
2845 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2846 * @from: attach to all cgroups of a given task
2847 * @tsk: the task to be attached
2849 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2851 struct cgroup_root
*root
;
2854 mutex_lock(&cgroup_mutex
);
2855 for_each_root(root
) {
2856 struct cgroup
*from_cgrp
;
2858 if (root
== &cgrp_dfl_root
)
2861 spin_lock_bh(&css_set_lock
);
2862 from_cgrp
= task_cgroup_from_root(from
, root
);
2863 spin_unlock_bh(&css_set_lock
);
2865 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2869 mutex_unlock(&cgroup_mutex
);
2873 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2875 static ssize_t
cgroup_tasks_write(struct kernfs_open_file
*of
,
2876 char *buf
, size_t nbytes
, loff_t off
)
2878 return __cgroup_procs_write(of
, buf
, nbytes
, off
, false);
2881 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
2882 char *buf
, size_t nbytes
, loff_t off
)
2884 return __cgroup_procs_write(of
, buf
, nbytes
, off
, true);
2887 static ssize_t
cgroup_release_agent_write(struct kernfs_open_file
*of
,
2888 char *buf
, size_t nbytes
, loff_t off
)
2890 struct cgroup
*cgrp
;
2892 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2894 cgrp
= cgroup_kn_lock_live(of
->kn
);
2897 spin_lock(&release_agent_path_lock
);
2898 strlcpy(cgrp
->root
->release_agent_path
, strstrip(buf
),
2899 sizeof(cgrp
->root
->release_agent_path
));
2900 spin_unlock(&release_agent_path_lock
);
2901 cgroup_kn_unlock(of
->kn
);
2905 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
2907 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2909 spin_lock(&release_agent_path_lock
);
2910 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2911 spin_unlock(&release_agent_path_lock
);
2912 seq_putc(seq
, '\n');
2916 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
2918 seq_puts(seq
, "0\n");
2922 static void cgroup_print_ss_mask(struct seq_file
*seq
, unsigned long ss_mask
)
2924 struct cgroup_subsys
*ss
;
2925 bool printed
= false;
2928 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
2931 seq_printf(seq
, "%s", ss
->name
);
2935 seq_putc(seq
, '\n');
2938 /* show controllers which are currently attached to the default hierarchy */
2939 static int cgroup_root_controllers_show(struct seq_file
*seq
, void *v
)
2941 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2943 cgroup_print_ss_mask(seq
, cgrp
->root
->subsys_mask
&
2944 ~cgrp_dfl_root_inhibit_ss_mask
);
2948 /* show controllers which are enabled from the parent */
2949 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2951 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2953 cgroup_print_ss_mask(seq
, cgroup_parent(cgrp
)->subtree_control
);
2957 /* show controllers which are enabled for a given cgroup's children */
2958 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2960 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2962 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2967 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2968 * @cgrp: root of the subtree to update csses for
2970 * @cgrp's child_subsys_mask has changed and its subtree's (self excluded)
2971 * css associations need to be updated accordingly. This function looks up
2972 * all css_sets which are attached to the subtree, creates the matching
2973 * updated css_sets and migrates the tasks to the new ones.
2975 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2977 LIST_HEAD(preloaded_csets
);
2978 struct cgroup_taskset tset
= CGROUP_TASKSET_INIT(tset
);
2979 struct cgroup_subsys_state
*css
;
2980 struct css_set
*src_cset
;
2983 lockdep_assert_held(&cgroup_mutex
);
2985 percpu_down_write(&cgroup_threadgroup_rwsem
);
2987 /* look up all csses currently attached to @cgrp's subtree */
2988 spin_lock_bh(&css_set_lock
);
2989 css_for_each_descendant_pre(css
, cgroup_css(cgrp
, NULL
)) {
2990 struct cgrp_cset_link
*link
;
2992 /* self is not affected by child_subsys_mask change */
2993 if (css
->cgroup
== cgrp
)
2996 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
)
2997 cgroup_migrate_add_src(link
->cset
, cgrp
,
3000 spin_unlock_bh(&css_set_lock
);
3002 /* NULL dst indicates self on default hierarchy */
3003 ret
= cgroup_migrate_prepare_dst(NULL
, &preloaded_csets
);
3007 spin_lock_bh(&css_set_lock
);
3008 list_for_each_entry(src_cset
, &preloaded_csets
, mg_preload_node
) {
3009 struct task_struct
*task
, *ntask
;
3011 /* src_csets precede dst_csets, break on the first dst_cset */
3012 if (!src_cset
->mg_src_cgrp
)
3015 /* all tasks in src_csets need to be migrated */
3016 list_for_each_entry_safe(task
, ntask
, &src_cset
->tasks
, cg_list
)
3017 cgroup_taskset_add(task
, &tset
);
3019 spin_unlock_bh(&css_set_lock
);
3021 ret
= cgroup_taskset_migrate(&tset
, cgrp
);
3023 cgroup_migrate_finish(&preloaded_csets
);
3024 percpu_up_write(&cgroup_threadgroup_rwsem
);
3028 /* change the enabled child controllers for a cgroup in the default hierarchy */
3029 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
3030 char *buf
, size_t nbytes
,
3033 unsigned long enable
= 0, disable
= 0;
3034 unsigned long css_enable
, css_disable
, old_sc
, new_sc
, old_ss
, new_ss
;
3035 struct cgroup
*cgrp
, *child
;
3036 struct cgroup_subsys
*ss
;
3041 * Parse input - space separated list of subsystem names prefixed
3042 * with either + or -.
3044 buf
= strstrip(buf
);
3045 while ((tok
= strsep(&buf
, " "))) {
3046 unsigned long tmp_ss_mask
= ~cgrp_dfl_root_inhibit_ss_mask
;
3050 for_each_subsys_which(ss
, ssid
, &tmp_ss_mask
) {
3051 if (!cgroup_ssid_enabled(ssid
) ||
3052 strcmp(tok
+ 1, ss
->name
))
3056 enable
|= 1 << ssid
;
3057 disable
&= ~(1 << ssid
);
3058 } else if (*tok
== '-') {
3059 disable
|= 1 << ssid
;
3060 enable
&= ~(1 << ssid
);
3066 if (ssid
== CGROUP_SUBSYS_COUNT
)
3070 cgrp
= cgroup_kn_lock_live(of
->kn
);
3074 for_each_subsys(ss
, ssid
) {
3075 if (enable
& (1 << ssid
)) {
3076 if (cgrp
->subtree_control
& (1 << ssid
)) {
3077 enable
&= ~(1 << ssid
);
3081 /* unavailable or not enabled on the parent? */
3082 if (!(cgrp_dfl_root
.subsys_mask
& (1 << ssid
)) ||
3083 (cgroup_parent(cgrp
) &&
3084 !(cgroup_parent(cgrp
)->subtree_control
& (1 << ssid
)))) {
3088 } else if (disable
& (1 << ssid
)) {
3089 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
3090 disable
&= ~(1 << ssid
);
3094 /* a child has it enabled? */
3095 cgroup_for_each_live_child(child
, cgrp
) {
3096 if (child
->subtree_control
& (1 << ssid
)) {
3104 if (!enable
&& !disable
) {
3110 * Except for the root, subtree_control must be zero for a cgroup
3111 * with tasks so that child cgroups don't compete against tasks.
3113 if (enable
&& cgroup_parent(cgrp
) && !list_empty(&cgrp
->cset_links
)) {
3119 * Update subsys masks and calculate what needs to be done. More
3120 * subsystems than specified may need to be enabled or disabled
3121 * depending on subsystem dependencies.
3123 old_sc
= cgrp
->subtree_control
;
3124 old_ss
= cgrp
->child_subsys_mask
;
3125 new_sc
= (old_sc
| enable
) & ~disable
;
3126 new_ss
= cgroup_calc_child_subsys_mask(cgrp
, new_sc
);
3128 css_enable
= ~old_ss
& new_ss
;
3129 css_disable
= old_ss
& ~new_ss
;
3130 enable
|= css_enable
;
3131 disable
|= css_disable
;
3134 * Because css offlining is asynchronous, userland might try to
3135 * re-enable the same controller while the previous instance is
3136 * still around. In such cases, wait till it's gone using
3139 for_each_subsys_which(ss
, ssid
, &css_enable
) {
3140 cgroup_for_each_live_child(child
, cgrp
) {
3143 if (!cgroup_css(child
, ss
))
3147 prepare_to_wait(&child
->offline_waitq
, &wait
,
3148 TASK_UNINTERRUPTIBLE
);
3149 cgroup_kn_unlock(of
->kn
);
3151 finish_wait(&child
->offline_waitq
, &wait
);
3154 return restart_syscall();
3158 cgrp
->subtree_control
= new_sc
;
3159 cgrp
->child_subsys_mask
= new_ss
;
3162 * Create new csses or make the existing ones visible. A css is
3163 * created invisible if it's being implicitly enabled through
3164 * dependency. An invisible css is made visible when the userland
3165 * explicitly enables it.
3167 for_each_subsys(ss
, ssid
) {
3168 if (!(enable
& (1 << ssid
)))
3171 cgroup_for_each_live_child(child
, cgrp
) {
3172 if (css_enable
& (1 << ssid
))
3173 ret
= create_css(child
, ss
,
3174 cgrp
->subtree_control
& (1 << ssid
));
3176 ret
= css_populate_dir(cgroup_css(child
, ss
),
3184 * At this point, cgroup_e_css() results reflect the new csses
3185 * making the following cgroup_update_dfl_csses() properly update
3186 * css associations of all tasks in the subtree.
3188 ret
= cgroup_update_dfl_csses(cgrp
);
3193 * All tasks are migrated out of disabled csses. Kill or hide
3194 * them. A css is hidden when the userland requests it to be
3195 * disabled while other subsystems are still depending on it. The
3196 * css must not actively control resources and be in the vanilla
3197 * state if it's made visible again later. Controllers which may
3198 * be depended upon should provide ->css_reset() for this purpose.
3200 for_each_subsys(ss
, ssid
) {
3201 if (!(disable
& (1 << ssid
)))
3204 cgroup_for_each_live_child(child
, cgrp
) {
3205 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
3207 if (css_disable
& (1 << ssid
)) {
3210 css_clear_dir(css
, NULL
);
3218 * The effective csses of all the descendants (excluding @cgrp) may
3219 * have changed. Subsystems can optionally subscribe to this event
3220 * by implementing ->css_e_css_changed() which is invoked if any of
3221 * the effective csses seen from the css's cgroup may have changed.
3223 for_each_subsys(ss
, ssid
) {
3224 struct cgroup_subsys_state
*this_css
= cgroup_css(cgrp
, ss
);
3225 struct cgroup_subsys_state
*css
;
3227 if (!ss
->css_e_css_changed
|| !this_css
)
3230 css_for_each_descendant_pre(css
, this_css
)
3231 if (css
!= this_css
)
3232 ss
->css_e_css_changed(css
);
3235 kernfs_activate(cgrp
->kn
);
3238 cgroup_kn_unlock(of
->kn
);
3239 return ret
?: nbytes
;
3242 cgrp
->subtree_control
= old_sc
;
3243 cgrp
->child_subsys_mask
= old_ss
;
3245 for_each_subsys(ss
, ssid
) {
3246 if (!(enable
& (1 << ssid
)))
3249 cgroup_for_each_live_child(child
, cgrp
) {
3250 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
3255 if (css_enable
& (1 << ssid
))
3258 css_clear_dir(css
, NULL
);
3264 static int cgroup_events_show(struct seq_file
*seq
, void *v
)
3266 seq_printf(seq
, "populated %d\n",
3267 cgroup_is_populated(seq_css(seq
)->cgroup
));
3271 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
3272 size_t nbytes
, loff_t off
)
3274 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
3275 struct cftype
*cft
= of
->kn
->priv
;
3276 struct cgroup_subsys_state
*css
;
3280 return cft
->write(of
, buf
, nbytes
, off
);
3283 * kernfs guarantees that a file isn't deleted with operations in
3284 * flight, which means that the matching css is and stays alive and
3285 * doesn't need to be pinned. The RCU locking is not necessary
3286 * either. It's just for the convenience of using cgroup_css().
3289 css
= cgroup_css(cgrp
, cft
->ss
);
3292 if (cft
->write_u64
) {
3293 unsigned long long v
;
3294 ret
= kstrtoull(buf
, 0, &v
);
3296 ret
= cft
->write_u64(css
, cft
, v
);
3297 } else if (cft
->write_s64
) {
3299 ret
= kstrtoll(buf
, 0, &v
);
3301 ret
= cft
->write_s64(css
, cft
, v
);
3306 return ret
?: nbytes
;
3309 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
3311 return seq_cft(seq
)->seq_start(seq
, ppos
);
3314 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
3316 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
3319 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
3321 seq_cft(seq
)->seq_stop(seq
, v
);
3324 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
3326 struct cftype
*cft
= seq_cft(m
);
3327 struct cgroup_subsys_state
*css
= seq_css(m
);
3330 return cft
->seq_show(m
, arg
);
3333 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
3334 else if (cft
->read_s64
)
3335 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
3341 static struct kernfs_ops cgroup_kf_single_ops
= {
3342 .atomic_write_len
= PAGE_SIZE
,
3343 .write
= cgroup_file_write
,
3344 .seq_show
= cgroup_seqfile_show
,
3347 static struct kernfs_ops cgroup_kf_ops
= {
3348 .atomic_write_len
= PAGE_SIZE
,
3349 .write
= cgroup_file_write
,
3350 .seq_start
= cgroup_seqfile_start
,
3351 .seq_next
= cgroup_seqfile_next
,
3352 .seq_stop
= cgroup_seqfile_stop
,
3353 .seq_show
= cgroup_seqfile_show
,
3357 * cgroup_rename - Only allow simple rename of directories in place.
3359 static int cgroup_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
3360 const char *new_name_str
)
3362 struct cgroup
*cgrp
= kn
->priv
;
3365 if (kernfs_type(kn
) != KERNFS_DIR
)
3367 if (kn
->parent
!= new_parent
)
3371 * This isn't a proper migration and its usefulness is very
3372 * limited. Disallow on the default hierarchy.
3374 if (cgroup_on_dfl(cgrp
))
3378 * We're gonna grab cgroup_mutex which nests outside kernfs
3379 * active_ref. kernfs_rename() doesn't require active_ref
3380 * protection. Break them before grabbing cgroup_mutex.
3382 kernfs_break_active_protection(new_parent
);
3383 kernfs_break_active_protection(kn
);
3385 mutex_lock(&cgroup_mutex
);
3387 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
3389 mutex_unlock(&cgroup_mutex
);
3391 kernfs_unbreak_active_protection(kn
);
3392 kernfs_unbreak_active_protection(new_parent
);
3396 /* set uid and gid of cgroup dirs and files to that of the creator */
3397 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3399 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3400 .ia_uid
= current_fsuid(),
3401 .ia_gid
= current_fsgid(), };
3403 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3404 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3407 return kernfs_setattr(kn
, &iattr
);
3410 static int cgroup_add_file(struct cgroup_subsys_state
*css
, struct cgroup
*cgrp
,
3413 char name
[CGROUP_FILE_NAME_MAX
];
3414 struct kernfs_node
*kn
;
3415 struct lock_class_key
*key
= NULL
;
3418 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3419 key
= &cft
->lockdep_key
;
3421 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3422 cgroup_file_mode(cft
), 0, cft
->kf_ops
, cft
,
3427 ret
= cgroup_kn_set_ugid(kn
);
3433 if (cft
->file_offset
) {
3434 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
3436 spin_lock_irq(&cgroup_file_kn_lock
);
3438 spin_unlock_irq(&cgroup_file_kn_lock
);
3445 * cgroup_addrm_files - add or remove files to a cgroup directory
3446 * @css: the target css
3447 * @cgrp: the target cgroup (usually css->cgroup)
3448 * @cfts: array of cftypes to be added
3449 * @is_add: whether to add or remove
3451 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3452 * For removals, this function never fails.
3454 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
3455 struct cgroup
*cgrp
, struct cftype cfts
[],
3458 struct cftype
*cft
, *cft_end
= NULL
;
3461 lockdep_assert_held(&cgroup_mutex
);
3464 for (cft
= cfts
; cft
!= cft_end
&& cft
->name
[0] != '\0'; cft
++) {
3465 /* does cft->flags tell us to skip this file on @cgrp? */
3466 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3468 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
3470 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3472 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3476 ret
= cgroup_add_file(css
, cgrp
, cft
);
3478 pr_warn("%s: failed to add %s, err=%d\n",
3479 __func__
, cft
->name
, ret
);
3485 cgroup_rm_file(cgrp
, cft
);
3491 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3494 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3495 struct cgroup
*root
= &ss
->root
->cgrp
;
3496 struct cgroup_subsys_state
*css
;
3499 lockdep_assert_held(&cgroup_mutex
);
3501 /* add/rm files for all cgroups created before */
3502 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3503 struct cgroup
*cgrp
= css
->cgroup
;
3505 if (cgroup_is_dead(cgrp
))
3508 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, is_add
);
3514 kernfs_activate(root
->kn
);
3518 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3522 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3523 /* free copy for custom atomic_write_len, see init_cftypes() */
3524 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3529 /* revert flags set by cgroup core while adding @cfts */
3530 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
3534 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3538 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3539 struct kernfs_ops
*kf_ops
;
3541 WARN_ON(cft
->ss
|| cft
->kf_ops
);
3544 kf_ops
= &cgroup_kf_ops
;
3546 kf_ops
= &cgroup_kf_single_ops
;
3549 * Ugh... if @cft wants a custom max_write_len, we need to
3550 * make a copy of kf_ops to set its atomic_write_len.
3552 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
3553 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
3555 cgroup_exit_cftypes(cfts
);
3558 kf_ops
->atomic_write_len
= cft
->max_write_len
;
3561 cft
->kf_ops
= kf_ops
;
3568 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
3570 lockdep_assert_held(&cgroup_mutex
);
3572 if (!cfts
|| !cfts
[0].ss
)
3575 list_del(&cfts
->node
);
3576 cgroup_apply_cftypes(cfts
, false);
3577 cgroup_exit_cftypes(cfts
);
3582 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3583 * @cfts: zero-length name terminated array of cftypes
3585 * Unregister @cfts. Files described by @cfts are removed from all
3586 * existing cgroups and all future cgroups won't have them either. This
3587 * function can be called anytime whether @cfts' subsys is attached or not.
3589 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3592 int cgroup_rm_cftypes(struct cftype
*cfts
)
3596 mutex_lock(&cgroup_mutex
);
3597 ret
= cgroup_rm_cftypes_locked(cfts
);
3598 mutex_unlock(&cgroup_mutex
);
3603 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3604 * @ss: target cgroup subsystem
3605 * @cfts: zero-length name terminated array of cftypes
3607 * Register @cfts to @ss. Files described by @cfts are created for all
3608 * existing cgroups to which @ss is attached and all future cgroups will
3609 * have them too. This function can be called anytime whether @ss is
3612 * Returns 0 on successful registration, -errno on failure. Note that this
3613 * function currently returns 0 as long as @cfts registration is successful
3614 * even if some file creation attempts on existing cgroups fail.
3616 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3620 if (!cgroup_ssid_enabled(ss
->id
))
3623 if (!cfts
|| cfts
[0].name
[0] == '\0')
3626 ret
= cgroup_init_cftypes(ss
, cfts
);
3630 mutex_lock(&cgroup_mutex
);
3632 list_add_tail(&cfts
->node
, &ss
->cfts
);
3633 ret
= cgroup_apply_cftypes(cfts
, true);
3635 cgroup_rm_cftypes_locked(cfts
);
3637 mutex_unlock(&cgroup_mutex
);
3642 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3643 * @ss: target cgroup subsystem
3644 * @cfts: zero-length name terminated array of cftypes
3646 * Similar to cgroup_add_cftypes() but the added files are only used for
3647 * the default hierarchy.
3649 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3653 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3654 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
3655 return cgroup_add_cftypes(ss
, cfts
);
3659 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3660 * @ss: target cgroup subsystem
3661 * @cfts: zero-length name terminated array of cftypes
3663 * Similar to cgroup_add_cftypes() but the added files are only used for
3664 * the legacy hierarchies.
3666 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3670 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3671 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
3672 return cgroup_add_cftypes(ss
, cfts
);
3676 * cgroup_file_notify - generate a file modified event for a cgroup_file
3677 * @cfile: target cgroup_file
3679 * @cfile must have been obtained by setting cftype->file_offset.
3681 void cgroup_file_notify(struct cgroup_file
*cfile
)
3683 unsigned long flags
;
3685 spin_lock_irqsave(&cgroup_file_kn_lock
, flags
);
3687 kernfs_notify(cfile
->kn
);
3688 spin_unlock_irqrestore(&cgroup_file_kn_lock
, flags
);
3692 * cgroup_task_count - count the number of tasks in a cgroup.
3693 * @cgrp: the cgroup in question
3695 * Return the number of tasks in the cgroup.
3697 static int cgroup_task_count(const struct cgroup
*cgrp
)
3700 struct cgrp_cset_link
*link
;
3702 spin_lock_bh(&css_set_lock
);
3703 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
3704 count
+= atomic_read(&link
->cset
->refcount
);
3705 spin_unlock_bh(&css_set_lock
);
3710 * css_next_child - find the next child of a given css
3711 * @pos: the current position (%NULL to initiate traversal)
3712 * @parent: css whose children to walk
3714 * This function returns the next child of @parent and should be called
3715 * under either cgroup_mutex or RCU read lock. The only requirement is
3716 * that @parent and @pos are accessible. The next sibling is guaranteed to
3717 * be returned regardless of their states.
3719 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3720 * css which finished ->css_online() is guaranteed to be visible in the
3721 * future iterations and will stay visible until the last reference is put.
3722 * A css which hasn't finished ->css_online() or already finished
3723 * ->css_offline() may show up during traversal. It's each subsystem's
3724 * responsibility to synchronize against on/offlining.
3726 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
3727 struct cgroup_subsys_state
*parent
)
3729 struct cgroup_subsys_state
*next
;
3731 cgroup_assert_mutex_or_rcu_locked();
3734 * @pos could already have been unlinked from the sibling list.
3735 * Once a cgroup is removed, its ->sibling.next is no longer
3736 * updated when its next sibling changes. CSS_RELEASED is set when
3737 * @pos is taken off list, at which time its next pointer is valid,
3738 * and, as releases are serialized, the one pointed to by the next
3739 * pointer is guaranteed to not have started release yet. This
3740 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3741 * critical section, the one pointed to by its next pointer is
3742 * guaranteed to not have finished its RCU grace period even if we
3743 * have dropped rcu_read_lock() inbetween iterations.
3745 * If @pos has CSS_RELEASED set, its next pointer can't be
3746 * dereferenced; however, as each css is given a monotonically
3747 * increasing unique serial number and always appended to the
3748 * sibling list, the next one can be found by walking the parent's
3749 * children until the first css with higher serial number than
3750 * @pos's. While this path can be slower, it happens iff iteration
3751 * races against release and the race window is very small.
3754 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
3755 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
3756 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
3758 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
3759 if (next
->serial_nr
> pos
->serial_nr
)
3764 * @next, if not pointing to the head, can be dereferenced and is
3767 if (&next
->sibling
!= &parent
->children
)
3773 * css_next_descendant_pre - find the next descendant for pre-order walk
3774 * @pos: the current position (%NULL to initiate traversal)
3775 * @root: css whose descendants to walk
3777 * To be used by css_for_each_descendant_pre(). Find the next descendant
3778 * to visit for pre-order traversal of @root's descendants. @root is
3779 * included in the iteration and the first node to be visited.
3781 * While this function requires cgroup_mutex or RCU read locking, it
3782 * doesn't require the whole traversal to be contained in a single critical
3783 * section. This function will return the correct next descendant as long
3784 * as both @pos and @root are accessible and @pos is a descendant of @root.
3786 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3787 * css which finished ->css_online() is guaranteed to be visible in the
3788 * future iterations and will stay visible until the last reference is put.
3789 * A css which hasn't finished ->css_online() or already finished
3790 * ->css_offline() may show up during traversal. It's each subsystem's
3791 * responsibility to synchronize against on/offlining.
3793 struct cgroup_subsys_state
*
3794 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3795 struct cgroup_subsys_state
*root
)
3797 struct cgroup_subsys_state
*next
;
3799 cgroup_assert_mutex_or_rcu_locked();
3801 /* if first iteration, visit @root */
3805 /* visit the first child if exists */
3806 next
= css_next_child(NULL
, pos
);
3810 /* no child, visit my or the closest ancestor's next sibling */
3811 while (pos
!= root
) {
3812 next
= css_next_child(pos
, pos
->parent
);
3822 * css_rightmost_descendant - return the rightmost descendant of a css
3823 * @pos: css of interest
3825 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3826 * is returned. This can be used during pre-order traversal to skip
3829 * While this function requires cgroup_mutex or RCU read locking, it
3830 * doesn't require the whole traversal to be contained in a single critical
3831 * section. This function will return the correct rightmost descendant as
3832 * long as @pos is accessible.
3834 struct cgroup_subsys_state
*
3835 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3837 struct cgroup_subsys_state
*last
, *tmp
;
3839 cgroup_assert_mutex_or_rcu_locked();
3843 /* ->prev isn't RCU safe, walk ->next till the end */
3845 css_for_each_child(tmp
, last
)
3852 static struct cgroup_subsys_state
*
3853 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3855 struct cgroup_subsys_state
*last
;
3859 pos
= css_next_child(NULL
, pos
);
3866 * css_next_descendant_post - find the next descendant for post-order walk
3867 * @pos: the current position (%NULL to initiate traversal)
3868 * @root: css whose descendants to walk
3870 * To be used by css_for_each_descendant_post(). Find the next descendant
3871 * to visit for post-order traversal of @root's descendants. @root is
3872 * included in the iteration and the last node to be visited.
3874 * While this function requires cgroup_mutex or RCU read locking, it
3875 * doesn't require the whole traversal to be contained in a single critical
3876 * section. This function will return the correct next descendant as long
3877 * as both @pos and @cgroup are accessible and @pos is a descendant of
3880 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3881 * css which finished ->css_online() is guaranteed to be visible in the
3882 * future iterations and will stay visible until the last reference is put.
3883 * A css which hasn't finished ->css_online() or already finished
3884 * ->css_offline() may show up during traversal. It's each subsystem's
3885 * responsibility to synchronize against on/offlining.
3887 struct cgroup_subsys_state
*
3888 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3889 struct cgroup_subsys_state
*root
)
3891 struct cgroup_subsys_state
*next
;
3893 cgroup_assert_mutex_or_rcu_locked();
3895 /* if first iteration, visit leftmost descendant which may be @root */
3897 return css_leftmost_descendant(root
);
3899 /* if we visited @root, we're done */
3903 /* if there's an unvisited sibling, visit its leftmost descendant */
3904 next
= css_next_child(pos
, pos
->parent
);
3906 return css_leftmost_descendant(next
);
3908 /* no sibling left, visit parent */
3913 * css_has_online_children - does a css have online children
3914 * @css: the target css
3916 * Returns %true if @css has any online children; otherwise, %false. This
3917 * function can be called from any context but the caller is responsible
3918 * for synchronizing against on/offlining as necessary.
3920 bool css_has_online_children(struct cgroup_subsys_state
*css
)
3922 struct cgroup_subsys_state
*child
;
3926 css_for_each_child(child
, css
) {
3927 if (child
->flags
& CSS_ONLINE
) {
3937 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
3938 * @it: the iterator to advance
3940 * Advance @it to the next css_set to walk.
3942 static void css_task_iter_advance_css_set(struct css_task_iter
*it
)
3944 struct list_head
*l
= it
->cset_pos
;
3945 struct cgrp_cset_link
*link
;
3946 struct css_set
*cset
;
3948 lockdep_assert_held(&css_set_lock
);
3950 /* Advance to the next non-empty css_set */
3953 if (l
== it
->cset_head
) {
3954 it
->cset_pos
= NULL
;
3955 it
->task_pos
= NULL
;
3960 cset
= container_of(l
, struct css_set
,
3961 e_cset_node
[it
->ss
->id
]);
3963 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3966 } while (!css_set_populated(cset
));
3970 if (!list_empty(&cset
->tasks
))
3971 it
->task_pos
= cset
->tasks
.next
;
3973 it
->task_pos
= cset
->mg_tasks
.next
;
3975 it
->tasks_head
= &cset
->tasks
;
3976 it
->mg_tasks_head
= &cset
->mg_tasks
;
3979 * We don't keep css_sets locked across iteration steps and thus
3980 * need to take steps to ensure that iteration can be resumed after
3981 * the lock is re-acquired. Iteration is performed at two levels -
3982 * css_sets and tasks in them.
3984 * Once created, a css_set never leaves its cgroup lists, so a
3985 * pinned css_set is guaranteed to stay put and we can resume
3986 * iteration afterwards.
3988 * Tasks may leave @cset across iteration steps. This is resolved
3989 * by registering each iterator with the css_set currently being
3990 * walked and making css_set_move_task() advance iterators whose
3991 * next task is leaving.
3994 list_del(&it
->iters_node
);
3995 put_css_set_locked(it
->cur_cset
);
3998 it
->cur_cset
= cset
;
3999 list_add(&it
->iters_node
, &cset
->task_iters
);
4002 static void css_task_iter_advance(struct css_task_iter
*it
)
4004 struct list_head
*l
= it
->task_pos
;
4006 lockdep_assert_held(&css_set_lock
);
4010 * Advance iterator to find next entry. cset->tasks is consumed
4011 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
4016 if (l
== it
->tasks_head
)
4017 l
= it
->mg_tasks_head
->next
;
4019 if (l
== it
->mg_tasks_head
)
4020 css_task_iter_advance_css_set(it
);
4026 * css_task_iter_start - initiate task iteration
4027 * @css: the css to walk tasks of
4028 * @it: the task iterator to use
4030 * Initiate iteration through the tasks of @css. The caller can call
4031 * css_task_iter_next() to walk through the tasks until the function
4032 * returns NULL. On completion of iteration, css_task_iter_end() must be
4035 void css_task_iter_start(struct cgroup_subsys_state
*css
,
4036 struct css_task_iter
*it
)
4038 /* no one should try to iterate before mounting cgroups */
4039 WARN_ON_ONCE(!use_task_css_set_links
);
4041 memset(it
, 0, sizeof(*it
));
4043 spin_lock_bh(&css_set_lock
);
4048 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
4050 it
->cset_pos
= &css
->cgroup
->cset_links
;
4052 it
->cset_head
= it
->cset_pos
;
4054 css_task_iter_advance_css_set(it
);
4056 spin_unlock_bh(&css_set_lock
);
4060 * css_task_iter_next - return the next task for the iterator
4061 * @it: the task iterator being iterated
4063 * The "next" function for task iteration. @it should have been
4064 * initialized via css_task_iter_start(). Returns NULL when the iteration
4067 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
4070 put_task_struct(it
->cur_task
);
4071 it
->cur_task
= NULL
;
4074 spin_lock_bh(&css_set_lock
);
4077 it
->cur_task
= list_entry(it
->task_pos
, struct task_struct
,
4079 get_task_struct(it
->cur_task
);
4080 css_task_iter_advance(it
);
4083 spin_unlock_bh(&css_set_lock
);
4085 return it
->cur_task
;
4089 * css_task_iter_end - finish task iteration
4090 * @it: the task iterator to finish
4092 * Finish task iteration started by css_task_iter_start().
4094 void css_task_iter_end(struct css_task_iter
*it
)
4097 spin_lock_bh(&css_set_lock
);
4098 list_del(&it
->iters_node
);
4099 put_css_set_locked(it
->cur_cset
);
4100 spin_unlock_bh(&css_set_lock
);
4104 put_task_struct(it
->cur_task
);
4108 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
4109 * @to: cgroup to which the tasks will be moved
4110 * @from: cgroup in which the tasks currently reside
4112 * Locking rules between cgroup_post_fork() and the migration path
4113 * guarantee that, if a task is forking while being migrated, the new child
4114 * is guaranteed to be either visible in the source cgroup after the
4115 * parent's migration is complete or put into the target cgroup. No task
4116 * can slip out of migration through forking.
4118 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
4120 LIST_HEAD(preloaded_csets
);
4121 struct cgrp_cset_link
*link
;
4122 struct css_task_iter it
;
4123 struct task_struct
*task
;
4126 mutex_lock(&cgroup_mutex
);
4128 /* all tasks in @from are being moved, all csets are source */
4129 spin_lock_bh(&css_set_lock
);
4130 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
4131 cgroup_migrate_add_src(link
->cset
, to
, &preloaded_csets
);
4132 spin_unlock_bh(&css_set_lock
);
4134 ret
= cgroup_migrate_prepare_dst(to
, &preloaded_csets
);
4139 * Migrate tasks one-by-one until @from is empty. This fails iff
4140 * ->can_attach() fails.
4143 css_task_iter_start(&from
->self
, &it
);
4144 task
= css_task_iter_next(&it
);
4146 get_task_struct(task
);
4147 css_task_iter_end(&it
);
4150 ret
= cgroup_migrate(task
, false, to
);
4151 put_task_struct(task
);
4153 } while (task
&& !ret
);
4155 cgroup_migrate_finish(&preloaded_csets
);
4156 mutex_unlock(&cgroup_mutex
);
4161 * Stuff for reading the 'tasks'/'procs' files.
4163 * Reading this file can return large amounts of data if a cgroup has
4164 * *lots* of attached tasks. So it may need several calls to read(),
4165 * but we cannot guarantee that the information we produce is correct
4166 * unless we produce it entirely atomically.
4170 /* which pidlist file are we talking about? */
4171 enum cgroup_filetype
{
4177 * A pidlist is a list of pids that virtually represents the contents of one
4178 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
4179 * a pair (one each for procs, tasks) for each pid namespace that's relevant
4182 struct cgroup_pidlist
{
4184 * used to find which pidlist is wanted. doesn't change as long as
4185 * this particular list stays in the list.
4187 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
4190 /* how many elements the above list has */
4192 /* each of these stored in a list by its cgroup */
4193 struct list_head links
;
4194 /* pointer to the cgroup we belong to, for list removal purposes */
4195 struct cgroup
*owner
;
4196 /* for delayed destruction */
4197 struct delayed_work destroy_dwork
;
4201 * The following two functions "fix" the issue where there are more pids
4202 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
4203 * TODO: replace with a kernel-wide solution to this problem
4205 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
4206 static void *pidlist_allocate(int count
)
4208 if (PIDLIST_TOO_LARGE(count
))
4209 return vmalloc(count
* sizeof(pid_t
));
4211 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
4214 static void pidlist_free(void *p
)
4220 * Used to destroy all pidlists lingering waiting for destroy timer. None
4221 * should be left afterwards.
4223 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
)
4225 struct cgroup_pidlist
*l
, *tmp_l
;
4227 mutex_lock(&cgrp
->pidlist_mutex
);
4228 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
4229 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
4230 mutex_unlock(&cgrp
->pidlist_mutex
);
4232 flush_workqueue(cgroup_pidlist_destroy_wq
);
4233 BUG_ON(!list_empty(&cgrp
->pidlists
));
4236 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
4238 struct delayed_work
*dwork
= to_delayed_work(work
);
4239 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
4241 struct cgroup_pidlist
*tofree
= NULL
;
4243 mutex_lock(&l
->owner
->pidlist_mutex
);
4246 * Destroy iff we didn't get queued again. The state won't change
4247 * as destroy_dwork can only be queued while locked.
4249 if (!delayed_work_pending(dwork
)) {
4250 list_del(&l
->links
);
4251 pidlist_free(l
->list
);
4252 put_pid_ns(l
->key
.ns
);
4256 mutex_unlock(&l
->owner
->pidlist_mutex
);
4261 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
4262 * Returns the number of unique elements.
4264 static int pidlist_uniq(pid_t
*list
, int length
)
4269 * we presume the 0th element is unique, so i starts at 1. trivial
4270 * edge cases first; no work needs to be done for either
4272 if (length
== 0 || length
== 1)
4274 /* src and dest walk down the list; dest counts unique elements */
4275 for (src
= 1; src
< length
; src
++) {
4276 /* find next unique element */
4277 while (list
[src
] == list
[src
-1]) {
4282 /* dest always points to where the next unique element goes */
4283 list
[dest
] = list
[src
];
4291 * The two pid files - task and cgroup.procs - guaranteed that the result
4292 * is sorted, which forced this whole pidlist fiasco. As pid order is
4293 * different per namespace, each namespace needs differently sorted list,
4294 * making it impossible to use, for example, single rbtree of member tasks
4295 * sorted by task pointer. As pidlists can be fairly large, allocating one
4296 * per open file is dangerous, so cgroup had to implement shared pool of
4297 * pidlists keyed by cgroup and namespace.
4299 * All this extra complexity was caused by the original implementation
4300 * committing to an entirely unnecessary property. In the long term, we
4301 * want to do away with it. Explicitly scramble sort order if on the
4302 * default hierarchy so that no such expectation exists in the new
4305 * Scrambling is done by swapping every two consecutive bits, which is
4306 * non-identity one-to-one mapping which disturbs sort order sufficiently.
4308 static pid_t
pid_fry(pid_t pid
)
4310 unsigned a
= pid
& 0x55555555;
4311 unsigned b
= pid
& 0xAAAAAAAA;
4313 return (a
<< 1) | (b
>> 1);
4316 static pid_t
cgroup_pid_fry(struct cgroup
*cgrp
, pid_t pid
)
4318 if (cgroup_on_dfl(cgrp
))
4319 return pid_fry(pid
);
4324 static int cmppid(const void *a
, const void *b
)
4326 return *(pid_t
*)a
- *(pid_t
*)b
;
4329 static int fried_cmppid(const void *a
, const void *b
)
4331 return pid_fry(*(pid_t
*)a
) - pid_fry(*(pid_t
*)b
);
4334 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
4335 enum cgroup_filetype type
)
4337 struct cgroup_pidlist
*l
;
4338 /* don't need task_nsproxy() if we're looking at ourself */
4339 struct pid_namespace
*ns
= task_active_pid_ns(current
);
4341 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4343 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
4344 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
4350 * find the appropriate pidlist for our purpose (given procs vs tasks)
4351 * returns with the lock on that pidlist already held, and takes care
4352 * of the use count, or returns NULL with no locks held if we're out of
4355 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
4356 enum cgroup_filetype type
)
4358 struct cgroup_pidlist
*l
;
4360 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4362 l
= cgroup_pidlist_find(cgrp
, type
);
4366 /* entry not found; create a new one */
4367 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
4371 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
4373 /* don't need task_nsproxy() if we're looking at ourself */
4374 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
4376 list_add(&l
->links
, &cgrp
->pidlists
);
4381 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
4383 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
4384 struct cgroup_pidlist
**lp
)
4388 int pid
, n
= 0; /* used for populating the array */
4389 struct css_task_iter it
;
4390 struct task_struct
*tsk
;
4391 struct cgroup_pidlist
*l
;
4393 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4396 * If cgroup gets more users after we read count, we won't have
4397 * enough space - tough. This race is indistinguishable to the
4398 * caller from the case that the additional cgroup users didn't
4399 * show up until sometime later on.
4401 length
= cgroup_task_count(cgrp
);
4402 array
= pidlist_allocate(length
);
4405 /* now, populate the array */
4406 css_task_iter_start(&cgrp
->self
, &it
);
4407 while ((tsk
= css_task_iter_next(&it
))) {
4408 if (unlikely(n
== length
))
4410 /* get tgid or pid for procs or tasks file respectively */
4411 if (type
== CGROUP_FILE_PROCS
)
4412 pid
= task_tgid_vnr(tsk
);
4414 pid
= task_pid_vnr(tsk
);
4415 if (pid
> 0) /* make sure to only use valid results */
4418 css_task_iter_end(&it
);
4420 /* now sort & (if procs) strip out duplicates */
4421 if (cgroup_on_dfl(cgrp
))
4422 sort(array
, length
, sizeof(pid_t
), fried_cmppid
, NULL
);
4424 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
4425 if (type
== CGROUP_FILE_PROCS
)
4426 length
= pidlist_uniq(array
, length
);
4428 l
= cgroup_pidlist_find_create(cgrp
, type
);
4430 pidlist_free(array
);
4434 /* store array, freeing old if necessary */
4435 pidlist_free(l
->list
);
4443 * cgroupstats_build - build and fill cgroupstats
4444 * @stats: cgroupstats to fill information into
4445 * @dentry: A dentry entry belonging to the cgroup for which stats have
4448 * Build and fill cgroupstats so that taskstats can export it to user
4451 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
4453 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
4454 struct cgroup
*cgrp
;
4455 struct css_task_iter it
;
4456 struct task_struct
*tsk
;
4458 /* it should be kernfs_node belonging to cgroupfs and is a directory */
4459 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
4460 kernfs_type(kn
) != KERNFS_DIR
)
4463 mutex_lock(&cgroup_mutex
);
4466 * We aren't being called from kernfs and there's no guarantee on
4467 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
4468 * @kn->priv is RCU safe. Let's do the RCU dancing.
4471 cgrp
= rcu_dereference(kn
->priv
);
4472 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
4474 mutex_unlock(&cgroup_mutex
);
4479 css_task_iter_start(&cgrp
->self
, &it
);
4480 while ((tsk
= css_task_iter_next(&it
))) {
4481 switch (tsk
->state
) {
4483 stats
->nr_running
++;
4485 case TASK_INTERRUPTIBLE
:
4486 stats
->nr_sleeping
++;
4488 case TASK_UNINTERRUPTIBLE
:
4489 stats
->nr_uninterruptible
++;
4492 stats
->nr_stopped
++;
4495 if (delayacct_is_task_waiting_on_io(tsk
))
4496 stats
->nr_io_wait
++;
4500 css_task_iter_end(&it
);
4502 mutex_unlock(&cgroup_mutex
);
4508 * seq_file methods for the tasks/procs files. The seq_file position is the
4509 * next pid to display; the seq_file iterator is a pointer to the pid
4510 * in the cgroup->l->list array.
4513 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
4516 * Initially we receive a position value that corresponds to
4517 * one more than the last pid shown (or 0 on the first call or
4518 * after a seek to the start). Use a binary-search to find the
4519 * next pid to display, if any
4521 struct kernfs_open_file
*of
= s
->private;
4522 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4523 struct cgroup_pidlist
*l
;
4524 enum cgroup_filetype type
= seq_cft(s
)->private;
4525 int index
= 0, pid
= *pos
;
4528 mutex_lock(&cgrp
->pidlist_mutex
);
4531 * !NULL @of->priv indicates that this isn't the first start()
4532 * after open. If the matching pidlist is around, we can use that.
4533 * Look for it. Note that @of->priv can't be used directly. It
4534 * could already have been destroyed.
4537 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
4540 * Either this is the first start() after open or the matching
4541 * pidlist has been destroyed inbetween. Create a new one.
4544 ret
= pidlist_array_load(cgrp
, type
,
4545 (struct cgroup_pidlist
**)&of
->priv
);
4547 return ERR_PTR(ret
);
4552 int end
= l
->length
;
4554 while (index
< end
) {
4555 int mid
= (index
+ end
) / 2;
4556 if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) == pid
) {
4559 } else if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) <= pid
)
4565 /* If we're off the end of the array, we're done */
4566 if (index
>= l
->length
)
4568 /* Update the abstract position to be the actual pid that we found */
4569 iter
= l
->list
+ index
;
4570 *pos
= cgroup_pid_fry(cgrp
, *iter
);
4574 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
4576 struct kernfs_open_file
*of
= s
->private;
4577 struct cgroup_pidlist
*l
= of
->priv
;
4580 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
4581 CGROUP_PIDLIST_DESTROY_DELAY
);
4582 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
4585 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4587 struct kernfs_open_file
*of
= s
->private;
4588 struct cgroup_pidlist
*l
= of
->priv
;
4590 pid_t
*end
= l
->list
+ l
->length
;
4592 * Advance to the next pid in the array. If this goes off the
4599 *pos
= cgroup_pid_fry(seq_css(s
)->cgroup
, *p
);
4604 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
4606 seq_printf(s
, "%d\n", *(int *)v
);
4611 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
4614 return notify_on_release(css
->cgroup
);
4617 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
4618 struct cftype
*cft
, u64 val
)
4621 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4623 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4627 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
4630 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4633 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
4634 struct cftype
*cft
, u64 val
)
4637 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4639 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4643 /* cgroup core interface files for the default hierarchy */
4644 static struct cftype cgroup_dfl_base_files
[] = {
4646 .name
= "cgroup.procs",
4647 .file_offset
= offsetof(struct cgroup
, procs_file
),
4648 .seq_start
= cgroup_pidlist_start
,
4649 .seq_next
= cgroup_pidlist_next
,
4650 .seq_stop
= cgroup_pidlist_stop
,
4651 .seq_show
= cgroup_pidlist_show
,
4652 .private = CGROUP_FILE_PROCS
,
4653 .write
= cgroup_procs_write
,
4656 .name
= "cgroup.controllers",
4657 .flags
= CFTYPE_ONLY_ON_ROOT
,
4658 .seq_show
= cgroup_root_controllers_show
,
4661 .name
= "cgroup.controllers",
4662 .flags
= CFTYPE_NOT_ON_ROOT
,
4663 .seq_show
= cgroup_controllers_show
,
4666 .name
= "cgroup.subtree_control",
4667 .seq_show
= cgroup_subtree_control_show
,
4668 .write
= cgroup_subtree_control_write
,
4671 .name
= "cgroup.events",
4672 .flags
= CFTYPE_NOT_ON_ROOT
,
4673 .file_offset
= offsetof(struct cgroup
, events_file
),
4674 .seq_show
= cgroup_events_show
,
4679 /* cgroup core interface files for the legacy hierarchies */
4680 static struct cftype cgroup_legacy_base_files
[] = {
4682 .name
= "cgroup.procs",
4683 .seq_start
= cgroup_pidlist_start
,
4684 .seq_next
= cgroup_pidlist_next
,
4685 .seq_stop
= cgroup_pidlist_stop
,
4686 .seq_show
= cgroup_pidlist_show
,
4687 .private = CGROUP_FILE_PROCS
,
4688 .write
= cgroup_procs_write
,
4691 .name
= "cgroup.clone_children",
4692 .read_u64
= cgroup_clone_children_read
,
4693 .write_u64
= cgroup_clone_children_write
,
4696 .name
= "cgroup.sane_behavior",
4697 .flags
= CFTYPE_ONLY_ON_ROOT
,
4698 .seq_show
= cgroup_sane_behavior_show
,
4702 .seq_start
= cgroup_pidlist_start
,
4703 .seq_next
= cgroup_pidlist_next
,
4704 .seq_stop
= cgroup_pidlist_stop
,
4705 .seq_show
= cgroup_pidlist_show
,
4706 .private = CGROUP_FILE_TASKS
,
4707 .write
= cgroup_tasks_write
,
4710 .name
= "notify_on_release",
4711 .read_u64
= cgroup_read_notify_on_release
,
4712 .write_u64
= cgroup_write_notify_on_release
,
4715 .name
= "release_agent",
4716 .flags
= CFTYPE_ONLY_ON_ROOT
,
4717 .seq_show
= cgroup_release_agent_show
,
4718 .write
= cgroup_release_agent_write
,
4719 .max_write_len
= PATH_MAX
- 1,
4725 * css destruction is four-stage process.
4727 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4728 * Implemented in kill_css().
4730 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4731 * and thus css_tryget_online() is guaranteed to fail, the css can be
4732 * offlined by invoking offline_css(). After offlining, the base ref is
4733 * put. Implemented in css_killed_work_fn().
4735 * 3. When the percpu_ref reaches zero, the only possible remaining
4736 * accessors are inside RCU read sections. css_release() schedules the
4739 * 4. After the grace period, the css can be freed. Implemented in
4740 * css_free_work_fn().
4742 * It is actually hairier because both step 2 and 4 require process context
4743 * and thus involve punting to css->destroy_work adding two additional
4744 * steps to the already complex sequence.
4746 static void css_free_work_fn(struct work_struct
*work
)
4748 struct cgroup_subsys_state
*css
=
4749 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4750 struct cgroup_subsys
*ss
= css
->ss
;
4751 struct cgroup
*cgrp
= css
->cgroup
;
4753 percpu_ref_exit(&css
->refcnt
);
4757 struct cgroup_subsys_state
*parent
= css
->parent
;
4761 cgroup_idr_remove(&ss
->css_idr
, id
);
4767 /* cgroup free path */
4768 atomic_dec(&cgrp
->root
->nr_cgrps
);
4769 cgroup_pidlist_destroy_all(cgrp
);
4770 cancel_work_sync(&cgrp
->release_agent_work
);
4772 if (cgroup_parent(cgrp
)) {
4774 * We get a ref to the parent, and put the ref when
4775 * this cgroup is being freed, so it's guaranteed
4776 * that the parent won't be destroyed before its
4779 cgroup_put(cgroup_parent(cgrp
));
4780 kernfs_put(cgrp
->kn
);
4784 * This is root cgroup's refcnt reaching zero,
4785 * which indicates that the root should be
4788 cgroup_destroy_root(cgrp
->root
);
4793 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
4795 struct cgroup_subsys_state
*css
=
4796 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
4798 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
4799 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4802 static void css_release_work_fn(struct work_struct
*work
)
4804 struct cgroup_subsys_state
*css
=
4805 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4806 struct cgroup_subsys
*ss
= css
->ss
;
4807 struct cgroup
*cgrp
= css
->cgroup
;
4809 mutex_lock(&cgroup_mutex
);
4811 css
->flags
|= CSS_RELEASED
;
4812 list_del_rcu(&css
->sibling
);
4815 /* css release path */
4816 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
4817 if (ss
->css_released
)
4818 ss
->css_released(css
);
4820 /* cgroup release path */
4821 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4825 * There are two control paths which try to determine
4826 * cgroup from dentry without going through kernfs -
4827 * cgroupstats_build() and css_tryget_online_from_dir().
4828 * Those are supported by RCU protecting clearing of
4829 * cgrp->kn->priv backpointer.
4831 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
, NULL
);
4834 mutex_unlock(&cgroup_mutex
);
4836 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4839 static void css_release(struct percpu_ref
*ref
)
4841 struct cgroup_subsys_state
*css
=
4842 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4844 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
4845 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4848 static void init_and_link_css(struct cgroup_subsys_state
*css
,
4849 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4851 lockdep_assert_held(&cgroup_mutex
);
4855 memset(css
, 0, sizeof(*css
));
4858 INIT_LIST_HEAD(&css
->sibling
);
4859 INIT_LIST_HEAD(&css
->children
);
4860 css
->serial_nr
= css_serial_nr_next
++;
4861 atomic_set(&css
->online_cnt
, 0);
4863 if (cgroup_parent(cgrp
)) {
4864 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
4865 css_get(css
->parent
);
4868 BUG_ON(cgroup_css(cgrp
, ss
));
4871 /* invoke ->css_online() on a new CSS and mark it online if successful */
4872 static int online_css(struct cgroup_subsys_state
*css
)
4874 struct cgroup_subsys
*ss
= css
->ss
;
4877 lockdep_assert_held(&cgroup_mutex
);
4880 ret
= ss
->css_online(css
);
4882 css
->flags
|= CSS_ONLINE
;
4883 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
4885 atomic_inc(&css
->online_cnt
);
4887 atomic_inc(&css
->parent
->online_cnt
);
4892 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4893 static void offline_css(struct cgroup_subsys_state
*css
)
4895 struct cgroup_subsys
*ss
= css
->ss
;
4897 lockdep_assert_held(&cgroup_mutex
);
4899 if (!(css
->flags
& CSS_ONLINE
))
4902 if (ss
->css_offline
)
4903 ss
->css_offline(css
);
4905 css
->flags
&= ~CSS_ONLINE
;
4906 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
4908 wake_up_all(&css
->cgroup
->offline_waitq
);
4912 * create_css - create a cgroup_subsys_state
4913 * @cgrp: the cgroup new css will be associated with
4914 * @ss: the subsys of new css
4915 * @visible: whether to create control knobs for the new css or not
4917 * Create a new css associated with @cgrp - @ss pair. On success, the new
4918 * css is online and installed in @cgrp with all interface files created if
4919 * @visible. Returns 0 on success, -errno on failure.
4921 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
4924 struct cgroup
*parent
= cgroup_parent(cgrp
);
4925 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
4926 struct cgroup_subsys_state
*css
;
4929 lockdep_assert_held(&cgroup_mutex
);
4931 css
= ss
->css_alloc(parent_css
);
4933 return PTR_ERR(css
);
4935 init_and_link_css(css
, ss
, cgrp
);
4937 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
4941 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
4943 goto err_free_percpu_ref
;
4947 err
= css_populate_dir(css
, NULL
);
4952 /* @css is ready to be brought online now, make it visible */
4953 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
4954 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
4956 err
= online_css(css
);
4960 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4961 cgroup_parent(parent
)) {
4962 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4963 current
->comm
, current
->pid
, ss
->name
);
4964 if (!strcmp(ss
->name
, "memory"))
4965 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4966 ss
->warned_broken_hierarchy
= true;
4972 list_del_rcu(&css
->sibling
);
4973 css_clear_dir(css
, NULL
);
4975 cgroup_idr_remove(&ss
->css_idr
, css
->id
);
4976 err_free_percpu_ref
:
4977 percpu_ref_exit(&css
->refcnt
);
4979 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4983 static int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
,
4986 struct cgroup
*parent
, *cgrp
, *tcgrp
;
4987 struct cgroup_root
*root
;
4988 struct cgroup_subsys
*ss
;
4989 struct kernfs_node
*kn
;
4990 int level
, ssid
, ret
;
4992 /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable.
4994 if (strchr(name
, '\n'))
4997 parent
= cgroup_kn_lock_live(parent_kn
);
5000 root
= parent
->root
;
5001 level
= parent
->level
+ 1;
5003 /* allocate the cgroup and its ID, 0 is reserved for the root */
5004 cgrp
= kzalloc(sizeof(*cgrp
) +
5005 sizeof(cgrp
->ancestor_ids
[0]) * (level
+ 1), GFP_KERNEL
);
5011 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
5016 * Temporarily set the pointer to NULL, so idr_find() won't return
5017 * a half-baked cgroup.
5019 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_KERNEL
);
5022 goto out_cancel_ref
;
5025 init_cgroup_housekeeping(cgrp
);
5027 cgrp
->self
.parent
= &parent
->self
;
5029 cgrp
->level
= level
;
5031 for (tcgrp
= cgrp
; tcgrp
; tcgrp
= cgroup_parent(tcgrp
))
5032 cgrp
->ancestor_ids
[tcgrp
->level
] = tcgrp
->id
;
5034 if (notify_on_release(parent
))
5035 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
5037 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
5038 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
5040 /* create the directory */
5041 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
5049 * This extra ref will be put in cgroup_free_fn() and guarantees
5050 * that @cgrp->kn is always accessible.
5054 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
5056 /* allocation complete, commit to creation */
5057 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
5058 atomic_inc(&root
->nr_cgrps
);
5062 * @cgrp is now fully operational. If something fails after this
5063 * point, it'll be released via the normal destruction path.
5065 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
5067 ret
= cgroup_kn_set_ugid(kn
);
5071 ret
= css_populate_dir(&cgrp
->self
, NULL
);
5075 /* let's create and online css's */
5076 for_each_subsys(ss
, ssid
) {
5077 if (parent
->child_subsys_mask
& (1 << ssid
)) {
5078 ret
= create_css(cgrp
, ss
,
5079 parent
->subtree_control
& (1 << ssid
));
5086 * On the default hierarchy, a child doesn't automatically inherit
5087 * subtree_control from the parent. Each is configured manually.
5089 if (!cgroup_on_dfl(cgrp
)) {
5090 cgrp
->subtree_control
= parent
->subtree_control
;
5091 cgroup_refresh_child_subsys_mask(cgrp
);
5094 kernfs_activate(kn
);
5100 cgroup_idr_remove(&root
->cgroup_idr
, cgrp
->id
);
5102 percpu_ref_exit(&cgrp
->self
.refcnt
);
5106 cgroup_kn_unlock(parent_kn
);
5110 cgroup_destroy_locked(cgrp
);
5115 * This is called when the refcnt of a css is confirmed to be killed.
5116 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5117 * initate destruction and put the css ref from kill_css().
5119 static void css_killed_work_fn(struct work_struct
*work
)
5121 struct cgroup_subsys_state
*css
=
5122 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
5124 mutex_lock(&cgroup_mutex
);
5129 /* @css can't go away while we're holding cgroup_mutex */
5131 } while (css
&& atomic_dec_and_test(&css
->online_cnt
));
5133 mutex_unlock(&cgroup_mutex
);
5136 /* css kill confirmation processing requires process context, bounce */
5137 static void css_killed_ref_fn(struct percpu_ref
*ref
)
5139 struct cgroup_subsys_state
*css
=
5140 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
5142 if (atomic_dec_and_test(&css
->online_cnt
)) {
5143 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
5144 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
5149 * kill_css - destroy a css
5150 * @css: css to destroy
5152 * This function initiates destruction of @css by removing cgroup interface
5153 * files and putting its base reference. ->css_offline() will be invoked
5154 * asynchronously once css_tryget_online() is guaranteed to fail and when
5155 * the reference count reaches zero, @css will be released.
5157 static void kill_css(struct cgroup_subsys_state
*css
)
5159 lockdep_assert_held(&cgroup_mutex
);
5162 * This must happen before css is disassociated with its cgroup.
5163 * See seq_css() for details.
5165 css_clear_dir(css
, NULL
);
5168 * Killing would put the base ref, but we need to keep it alive
5169 * until after ->css_offline().
5174 * cgroup core guarantees that, by the time ->css_offline() is
5175 * invoked, no new css reference will be given out via
5176 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5177 * proceed to offlining css's because percpu_ref_kill() doesn't
5178 * guarantee that the ref is seen as killed on all CPUs on return.
5180 * Use percpu_ref_kill_and_confirm() to get notifications as each
5181 * css is confirmed to be seen as killed on all CPUs.
5183 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
5187 * cgroup_destroy_locked - the first stage of cgroup destruction
5188 * @cgrp: cgroup to be destroyed
5190 * css's make use of percpu refcnts whose killing latency shouldn't be
5191 * exposed to userland and are RCU protected. Also, cgroup core needs to
5192 * guarantee that css_tryget_online() won't succeed by the time
5193 * ->css_offline() is invoked. To satisfy all the requirements,
5194 * destruction is implemented in the following two steps.
5196 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5197 * userland visible parts and start killing the percpu refcnts of
5198 * css's. Set up so that the next stage will be kicked off once all
5199 * the percpu refcnts are confirmed to be killed.
5201 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5202 * rest of destruction. Once all cgroup references are gone, the
5203 * cgroup is RCU-freed.
5205 * This function implements s1. After this step, @cgrp is gone as far as
5206 * the userland is concerned and a new cgroup with the same name may be
5207 * created. As cgroup doesn't care about the names internally, this
5208 * doesn't cause any problem.
5210 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
5211 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
5213 struct cgroup_subsys_state
*css
;
5216 lockdep_assert_held(&cgroup_mutex
);
5219 * Only migration can raise populated from zero and we're already
5220 * holding cgroup_mutex.
5222 if (cgroup_is_populated(cgrp
))
5226 * Make sure there's no live children. We can't test emptiness of
5227 * ->self.children as dead children linger on it while being
5228 * drained; otherwise, "rmdir parent/child parent" may fail.
5230 if (css_has_online_children(&cgrp
->self
))
5234 * Mark @cgrp dead. This prevents further task migration and child
5235 * creation by disabling cgroup_lock_live_group().
5237 cgrp
->self
.flags
&= ~CSS_ONLINE
;
5239 /* initiate massacre of all css's */
5240 for_each_css(css
, ssid
, cgrp
)
5244 * Remove @cgrp directory along with the base files. @cgrp has an
5245 * extra ref on its kn.
5247 kernfs_remove(cgrp
->kn
);
5249 check_for_release(cgroup_parent(cgrp
));
5251 /* put the base reference */
5252 percpu_ref_kill(&cgrp
->self
.refcnt
);
5257 static int cgroup_rmdir(struct kernfs_node
*kn
)
5259 struct cgroup
*cgrp
;
5262 cgrp
= cgroup_kn_lock_live(kn
);
5266 ret
= cgroup_destroy_locked(cgrp
);
5268 cgroup_kn_unlock(kn
);
5272 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
5273 .remount_fs
= cgroup_remount
,
5274 .show_options
= cgroup_show_options
,
5275 .mkdir
= cgroup_mkdir
,
5276 .rmdir
= cgroup_rmdir
,
5277 .rename
= cgroup_rename
,
5280 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
5282 struct cgroup_subsys_state
*css
;
5284 pr_debug("Initializing cgroup subsys %s\n", ss
->name
);
5286 mutex_lock(&cgroup_mutex
);
5288 idr_init(&ss
->css_idr
);
5289 INIT_LIST_HEAD(&ss
->cfts
);
5291 /* Create the root cgroup state for this subsystem */
5292 ss
->root
= &cgrp_dfl_root
;
5293 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
5294 /* We don't handle early failures gracefully */
5295 BUG_ON(IS_ERR(css
));
5296 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
5299 * Root csses are never destroyed and we can't initialize
5300 * percpu_ref during early init. Disable refcnting.
5302 css
->flags
|= CSS_NO_REF
;
5305 /* allocation can't be done safely during early init */
5308 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
5309 BUG_ON(css
->id
< 0);
5312 /* Update the init_css_set to contain a subsys
5313 * pointer to this state - since the subsystem is
5314 * newly registered, all tasks and hence the
5315 * init_css_set is in the subsystem's root cgroup. */
5316 init_css_set
.subsys
[ss
->id
] = css
;
5318 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
5319 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
5320 have_free_callback
|= (bool)ss
->free
<< ss
->id
;
5321 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
5323 /* At system boot, before all subsystems have been
5324 * registered, no tasks have been forked, so we don't
5325 * need to invoke fork callbacks here. */
5326 BUG_ON(!list_empty(&init_task
.tasks
));
5328 BUG_ON(online_css(css
));
5330 mutex_unlock(&cgroup_mutex
);
5334 * cgroup_init_early - cgroup initialization at system boot
5336 * Initialize cgroups at system boot, and initialize any
5337 * subsystems that request early init.
5339 int __init
cgroup_init_early(void)
5341 static struct cgroup_sb_opts __initdata opts
;
5342 struct cgroup_subsys
*ss
;
5345 init_cgroup_root(&cgrp_dfl_root
, &opts
);
5346 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
5348 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5350 for_each_subsys(ss
, i
) {
5351 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
5352 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
5353 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5355 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5356 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5359 ss
->name
= cgroup_subsys_name
[i
];
5360 if (!ss
->legacy_name
)
5361 ss
->legacy_name
= cgroup_subsys_name
[i
];
5364 cgroup_init_subsys(ss
, true);
5369 static unsigned long cgroup_disable_mask __initdata
;
5372 * cgroup_init - cgroup initialization
5374 * Register cgroup filesystem and /proc file, and initialize
5375 * any subsystems that didn't request early init.
5377 int __init
cgroup_init(void)
5379 struct cgroup_subsys
*ss
;
5383 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem
));
5384 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_dfl_base_files
));
5385 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_legacy_base_files
));
5387 get_user_ns(init_cgroup_ns
.user_ns
);
5389 mutex_lock(&cgroup_mutex
);
5391 /* Add init_css_set to the hash table */
5392 key
= css_set_hash(init_css_set
.subsys
);
5393 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
5395 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
5397 mutex_unlock(&cgroup_mutex
);
5399 for_each_subsys(ss
, ssid
) {
5400 if (ss
->early_init
) {
5401 struct cgroup_subsys_state
*css
=
5402 init_css_set
.subsys
[ss
->id
];
5404 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5406 BUG_ON(css
->id
< 0);
5408 cgroup_init_subsys(ss
, false);
5411 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5412 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5415 * Setting dfl_root subsys_mask needs to consider the
5416 * disabled flag and cftype registration needs kmalloc,
5417 * both of which aren't available during early_init.
5419 if (cgroup_disable_mask
& (1 << ssid
)) {
5420 static_branch_disable(cgroup_subsys_enabled_key
[ssid
]);
5421 printk(KERN_INFO
"Disabling %s control group subsystem\n",
5426 if (cgroup_ssid_no_v1(ssid
))
5427 printk(KERN_INFO
"Disabling %s control group subsystem in v1 mounts\n",
5430 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
5432 if (!ss
->dfl_cftypes
)
5433 cgrp_dfl_root_inhibit_ss_mask
|= 1 << ss
->id
;
5435 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5436 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5438 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5439 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5443 ss
->bind(init_css_set
.subsys
[ssid
]);
5446 WARN_ON(sysfs_create_mount_point(fs_kobj
, "cgroup"));
5447 WARN_ON(register_filesystem(&cgroup_fs_type
));
5448 WARN_ON(register_filesystem(&cgroup2_fs_type
));
5449 WARN_ON(!proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
));
5454 static int __init
cgroup_wq_init(void)
5457 * There isn't much point in executing destruction path in
5458 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5459 * Use 1 for @max_active.
5461 * We would prefer to do this in cgroup_init() above, but that
5462 * is called before init_workqueues(): so leave this until after.
5464 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5465 BUG_ON(!cgroup_destroy_wq
);
5468 * Used to destroy pidlists and separate to serve as flush domain.
5469 * Cap @max_active to 1 too.
5471 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
5473 BUG_ON(!cgroup_pidlist_destroy_wq
);
5477 core_initcall(cgroup_wq_init
);
5480 * proc_cgroup_show()
5481 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5482 * - Used for /proc/<pid>/cgroup.
5484 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
5485 struct pid
*pid
, struct task_struct
*tsk
)
5489 struct cgroup_root
*root
;
5492 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5496 mutex_lock(&cgroup_mutex
);
5497 spin_lock_bh(&css_set_lock
);
5499 for_each_root(root
) {
5500 struct cgroup_subsys
*ss
;
5501 struct cgroup
*cgrp
;
5502 int ssid
, count
= 0;
5504 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_root_visible
)
5507 seq_printf(m
, "%d:", root
->hierarchy_id
);
5508 if (root
!= &cgrp_dfl_root
)
5509 for_each_subsys(ss
, ssid
)
5510 if (root
->subsys_mask
& (1 << ssid
))
5511 seq_printf(m
, "%s%s", count
++ ? "," : "",
5513 if (strlen(root
->name
))
5514 seq_printf(m
, "%sname=%s", count
? "," : "",
5518 cgrp
= task_cgroup_from_root(tsk
, root
);
5521 * On traditional hierarchies, all zombie tasks show up as
5522 * belonging to the root cgroup. On the default hierarchy,
5523 * while a zombie doesn't show up in "cgroup.procs" and
5524 * thus can't be migrated, its /proc/PID/cgroup keeps
5525 * reporting the cgroup it belonged to before exiting. If
5526 * the cgroup is removed before the zombie is reaped,
5527 * " (deleted)" is appended to the cgroup path.
5529 if (cgroup_on_dfl(cgrp
) || !(tsk
->flags
& PF_EXITING
)) {
5530 path
= cgroup_path_ns_locked(cgrp
, buf
, PATH_MAX
,
5531 current
->nsproxy
->cgroup_ns
);
5533 retval
= -ENAMETOOLONG
;
5542 if (cgroup_on_dfl(cgrp
) && cgroup_is_dead(cgrp
))
5543 seq_puts(m
, " (deleted)\n");
5550 spin_unlock_bh(&css_set_lock
);
5551 mutex_unlock(&cgroup_mutex
);
5557 /* Display information about each subsystem and each hierarchy */
5558 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5560 struct cgroup_subsys
*ss
;
5563 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5565 * ideally we don't want subsystems moving around while we do this.
5566 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5567 * subsys/hierarchy state.
5569 mutex_lock(&cgroup_mutex
);
5571 for_each_subsys(ss
, i
)
5572 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5573 ss
->legacy_name
, ss
->root
->hierarchy_id
,
5574 atomic_read(&ss
->root
->nr_cgrps
),
5575 cgroup_ssid_enabled(i
));
5577 mutex_unlock(&cgroup_mutex
);
5581 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5583 return single_open(file
, proc_cgroupstats_show
, NULL
);
5586 static const struct file_operations proc_cgroupstats_operations
= {
5587 .open
= cgroupstats_open
,
5589 .llseek
= seq_lseek
,
5590 .release
= single_release
,
5594 * cgroup_fork - initialize cgroup related fields during copy_process()
5595 * @child: pointer to task_struct of forking parent process.
5597 * A task is associated with the init_css_set until cgroup_post_fork()
5598 * attaches it to the parent's css_set. Empty cg_list indicates that
5599 * @child isn't holding reference to its css_set.
5601 void cgroup_fork(struct task_struct
*child
)
5603 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5604 INIT_LIST_HEAD(&child
->cg_list
);
5608 * cgroup_can_fork - called on a new task before the process is exposed
5609 * @child: the task in question.
5611 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5612 * returns an error, the fork aborts with that error code. This allows for
5613 * a cgroup subsystem to conditionally allow or deny new forks.
5615 int cgroup_can_fork(struct task_struct
*child
)
5617 struct cgroup_subsys
*ss
;
5620 for_each_subsys_which(ss
, i
, &have_canfork_callback
) {
5621 ret
= ss
->can_fork(child
);
5629 for_each_subsys(ss
, j
) {
5632 if (ss
->cancel_fork
)
5633 ss
->cancel_fork(child
);
5640 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5641 * @child: the task in question
5643 * This calls the cancel_fork() callbacks if a fork failed *after*
5644 * cgroup_can_fork() succeded.
5646 void cgroup_cancel_fork(struct task_struct
*child
)
5648 struct cgroup_subsys
*ss
;
5651 for_each_subsys(ss
, i
)
5652 if (ss
->cancel_fork
)
5653 ss
->cancel_fork(child
);
5657 * cgroup_post_fork - called on a new task after adding it to the task list
5658 * @child: the task in question
5660 * Adds the task to the list running through its css_set if necessary and
5661 * call the subsystem fork() callbacks. Has to be after the task is
5662 * visible on the task list in case we race with the first call to
5663 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5666 void cgroup_post_fork(struct task_struct
*child
)
5668 struct cgroup_subsys
*ss
;
5672 * This may race against cgroup_enable_task_cg_lists(). As that
5673 * function sets use_task_css_set_links before grabbing
5674 * tasklist_lock and we just went through tasklist_lock to add
5675 * @child, it's guaranteed that either we see the set
5676 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5677 * @child during its iteration.
5679 * If we won the race, @child is associated with %current's
5680 * css_set. Grabbing css_set_lock guarantees both that the
5681 * association is stable, and, on completion of the parent's
5682 * migration, @child is visible in the source of migration or
5683 * already in the destination cgroup. This guarantee is necessary
5684 * when implementing operations which need to migrate all tasks of
5685 * a cgroup to another.
5687 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
5688 * will remain in init_css_set. This is safe because all tasks are
5689 * in the init_css_set before cg_links is enabled and there's no
5690 * operation which transfers all tasks out of init_css_set.
5692 if (use_task_css_set_links
) {
5693 struct css_set
*cset
;
5695 spin_lock_bh(&css_set_lock
);
5696 cset
= task_css_set(current
);
5697 if (list_empty(&child
->cg_list
)) {
5699 css_set_move_task(child
, NULL
, cset
, false);
5701 spin_unlock_bh(&css_set_lock
);
5705 * Call ss->fork(). This must happen after @child is linked on
5706 * css_set; otherwise, @child might change state between ->fork()
5707 * and addition to css_set.
5709 for_each_subsys_which(ss
, i
, &have_fork_callback
)
5714 * cgroup_exit - detach cgroup from exiting task
5715 * @tsk: pointer to task_struct of exiting process
5717 * Description: Detach cgroup from @tsk and release it.
5719 * Note that cgroups marked notify_on_release force every task in
5720 * them to take the global cgroup_mutex mutex when exiting.
5721 * This could impact scaling on very large systems. Be reluctant to
5722 * use notify_on_release cgroups where very high task exit scaling
5723 * is required on large systems.
5725 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5726 * call cgroup_exit() while the task is still competent to handle
5727 * notify_on_release(), then leave the task attached to the root cgroup in
5728 * each hierarchy for the remainder of its exit. No need to bother with
5729 * init_css_set refcnting. init_css_set never goes away and we can't race
5730 * with migration path - PF_EXITING is visible to migration path.
5732 void cgroup_exit(struct task_struct
*tsk
)
5734 struct cgroup_subsys
*ss
;
5735 struct css_set
*cset
;
5739 * Unlink from @tsk from its css_set. As migration path can't race
5740 * with us, we can check css_set and cg_list without synchronization.
5742 cset
= task_css_set(tsk
);
5744 if (!list_empty(&tsk
->cg_list
)) {
5745 spin_lock_bh(&css_set_lock
);
5746 css_set_move_task(tsk
, cset
, NULL
, false);
5747 spin_unlock_bh(&css_set_lock
);
5752 /* see cgroup_post_fork() for details */
5753 for_each_subsys_which(ss
, i
, &have_exit_callback
)
5757 void cgroup_free(struct task_struct
*task
)
5759 struct css_set
*cset
= task_css_set(task
);
5760 struct cgroup_subsys
*ss
;
5763 for_each_subsys_which(ss
, ssid
, &have_free_callback
)
5769 static void check_for_release(struct cgroup
*cgrp
)
5771 if (notify_on_release(cgrp
) && !cgroup_is_populated(cgrp
) &&
5772 !css_has_online_children(&cgrp
->self
) && !cgroup_is_dead(cgrp
))
5773 schedule_work(&cgrp
->release_agent_work
);
5777 * Notify userspace when a cgroup is released, by running the
5778 * configured release agent with the name of the cgroup (path
5779 * relative to the root of cgroup file system) as the argument.
5781 * Most likely, this user command will try to rmdir this cgroup.
5783 * This races with the possibility that some other task will be
5784 * attached to this cgroup before it is removed, or that some other
5785 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5786 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5787 * unused, and this cgroup will be reprieved from its death sentence,
5788 * to continue to serve a useful existence. Next time it's released,
5789 * we will get notified again, if it still has 'notify_on_release' set.
5791 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5792 * means only wait until the task is successfully execve()'d. The
5793 * separate release agent task is forked by call_usermodehelper(),
5794 * then control in this thread returns here, without waiting for the
5795 * release agent task. We don't bother to wait because the caller of
5796 * this routine has no use for the exit status of the release agent
5797 * task, so no sense holding our caller up for that.
5799 static void cgroup_release_agent(struct work_struct
*work
)
5801 struct cgroup
*cgrp
=
5802 container_of(work
, struct cgroup
, release_agent_work
);
5803 char *pathbuf
= NULL
, *agentbuf
= NULL
, *path
;
5804 char *argv
[3], *envp
[3];
5806 mutex_lock(&cgroup_mutex
);
5808 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5809 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5810 if (!pathbuf
|| !agentbuf
)
5813 spin_lock_bh(&css_set_lock
);
5814 path
= cgroup_path_ns_locked(cgrp
, pathbuf
, PATH_MAX
, &init_cgroup_ns
);
5815 spin_unlock_bh(&css_set_lock
);
5823 /* minimal command environment */
5825 envp
[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5828 mutex_unlock(&cgroup_mutex
);
5829 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5832 mutex_unlock(&cgroup_mutex
);
5838 static int __init
cgroup_disable(char *str
)
5840 struct cgroup_subsys
*ss
;
5844 while ((token
= strsep(&str
, ",")) != NULL
) {
5848 for_each_subsys(ss
, i
) {
5849 if (strcmp(token
, ss
->name
) &&
5850 strcmp(token
, ss
->legacy_name
))
5852 cgroup_disable_mask
|= 1 << i
;
5857 __setup("cgroup_disable=", cgroup_disable
);
5859 static int __init
cgroup_no_v1(char *str
)
5861 struct cgroup_subsys
*ss
;
5865 while ((token
= strsep(&str
, ",")) != NULL
) {
5869 if (!strcmp(token
, "all")) {
5870 cgroup_no_v1_mask
= ~0UL;
5874 for_each_subsys(ss
, i
) {
5875 if (strcmp(token
, ss
->name
) &&
5876 strcmp(token
, ss
->legacy_name
))
5879 cgroup_no_v1_mask
|= 1 << i
;
5884 __setup("cgroup_no_v1=", cgroup_no_v1
);
5887 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5888 * @dentry: directory dentry of interest
5889 * @ss: subsystem of interest
5891 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5892 * to get the corresponding css and return it. If such css doesn't exist
5893 * or can't be pinned, an ERR_PTR value is returned.
5895 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
5896 struct cgroup_subsys
*ss
)
5898 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
5899 struct cgroup_subsys_state
*css
= NULL
;
5900 struct cgroup
*cgrp
;
5902 /* is @dentry a cgroup dir? */
5903 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
5904 kernfs_type(kn
) != KERNFS_DIR
)
5905 return ERR_PTR(-EBADF
);
5910 * This path doesn't originate from kernfs and @kn could already
5911 * have been or be removed at any point. @kn->priv is RCU
5912 * protected for this access. See css_release_work_fn() for details.
5914 cgrp
= rcu_dereference(kn
->priv
);
5916 css
= cgroup_css(cgrp
, ss
);
5918 if (!css
|| !css_tryget_online(css
))
5919 css
= ERR_PTR(-ENOENT
);
5926 * css_from_id - lookup css by id
5927 * @id: the cgroup id
5928 * @ss: cgroup subsys to be looked into
5930 * Returns the css if there's valid one with @id, otherwise returns NULL.
5931 * Should be called under rcu_read_lock().
5933 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5935 WARN_ON_ONCE(!rcu_read_lock_held());
5936 return id
> 0 ? idr_find(&ss
->css_idr
, id
) : NULL
;
5940 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
5941 * @path: path on the default hierarchy
5943 * Find the cgroup at @path on the default hierarchy, increment its
5944 * reference count and return it. Returns pointer to the found cgroup on
5945 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
5946 * if @path points to a non-directory.
5948 struct cgroup
*cgroup_get_from_path(const char *path
)
5950 struct kernfs_node
*kn
;
5951 struct cgroup
*cgrp
;
5953 mutex_lock(&cgroup_mutex
);
5955 kn
= kernfs_walk_and_get(cgrp_dfl_root
.cgrp
.kn
, path
);
5957 if (kernfs_type(kn
) == KERNFS_DIR
) {
5961 cgrp
= ERR_PTR(-ENOTDIR
);
5965 cgrp
= ERR_PTR(-ENOENT
);
5968 mutex_unlock(&cgroup_mutex
);
5971 EXPORT_SYMBOL_GPL(cgroup_get_from_path
);
5974 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
5975 * definition in cgroup-defs.h.
5977 #ifdef CONFIG_SOCK_CGROUP_DATA
5979 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
5981 DEFINE_SPINLOCK(cgroup_sk_update_lock
);
5982 static bool cgroup_sk_alloc_disabled __read_mostly
;
5984 void cgroup_sk_alloc_disable(void)
5986 if (cgroup_sk_alloc_disabled
)
5988 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
5989 cgroup_sk_alloc_disabled
= true;
5994 #define cgroup_sk_alloc_disabled false
5998 void cgroup_sk_alloc(struct sock_cgroup_data
*skcd
)
6000 if (cgroup_sk_alloc_disabled
)
6006 struct css_set
*cset
;
6008 cset
= task_css_set(current
);
6009 if (likely(cgroup_tryget(cset
->dfl_cgrp
))) {
6010 skcd
->val
= (unsigned long)cset
->dfl_cgrp
;
6019 void cgroup_sk_free(struct sock_cgroup_data
*skcd
)
6021 cgroup_put(sock_cgroup_ptr(skcd
));
6024 #endif /* CONFIG_SOCK_CGROUP_DATA */
6026 /* cgroup namespaces */
6028 static struct cgroup_namespace
*alloc_cgroup_ns(void)
6030 struct cgroup_namespace
*new_ns
;
6033 new_ns
= kzalloc(sizeof(struct cgroup_namespace
), GFP_KERNEL
);
6035 return ERR_PTR(-ENOMEM
);
6036 ret
= ns_alloc_inum(&new_ns
->ns
);
6039 return ERR_PTR(ret
);
6041 atomic_set(&new_ns
->count
, 1);
6042 new_ns
->ns
.ops
= &cgroupns_operations
;
6046 void free_cgroup_ns(struct cgroup_namespace
*ns
)
6048 put_css_set(ns
->root_cset
);
6049 put_user_ns(ns
->user_ns
);
6050 ns_free_inum(&ns
->ns
);
6053 EXPORT_SYMBOL(free_cgroup_ns
);
6055 struct cgroup_namespace
*copy_cgroup_ns(unsigned long flags
,
6056 struct user_namespace
*user_ns
,
6057 struct cgroup_namespace
*old_ns
)
6059 struct cgroup_namespace
*new_ns
= NULL
;
6060 struct css_set
*cset
= NULL
;
6065 if (!(flags
& CLONE_NEWCGROUP
)) {
6066 get_cgroup_ns(old_ns
);
6070 /* Allow only sysadmin to create cgroup namespace. */
6072 if (!ns_capable(user_ns
, CAP_SYS_ADMIN
))
6075 mutex_lock(&cgroup_mutex
);
6076 spin_lock_bh(&css_set_lock
);
6078 cset
= task_css_set(current
);
6081 spin_unlock_bh(&css_set_lock
);
6082 mutex_unlock(&cgroup_mutex
);
6085 new_ns
= alloc_cgroup_ns();
6089 new_ns
->user_ns
= get_user_ns(user_ns
);
6090 new_ns
->root_cset
= cset
;
6098 return ERR_PTR(err
);
6101 static inline struct cgroup_namespace
*to_cg_ns(struct ns_common
*ns
)
6103 return container_of(ns
, struct cgroup_namespace
, ns
);
6106 static int cgroupns_install(struct nsproxy
*nsproxy
, struct ns_common
*ns
)
6108 struct cgroup_namespace
*cgroup_ns
= to_cg_ns(ns
);
6110 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN
) ||
6111 !ns_capable(cgroup_ns
->user_ns
, CAP_SYS_ADMIN
))
6114 /* Don't need to do anything if we are attaching to our own cgroupns. */
6115 if (cgroup_ns
== nsproxy
->cgroup_ns
)
6118 get_cgroup_ns(cgroup_ns
);
6119 put_cgroup_ns(nsproxy
->cgroup_ns
);
6120 nsproxy
->cgroup_ns
= cgroup_ns
;
6125 static struct ns_common
*cgroupns_get(struct task_struct
*task
)
6127 struct cgroup_namespace
*ns
= NULL
;
6128 struct nsproxy
*nsproxy
;
6131 nsproxy
= task
->nsproxy
;
6133 ns
= nsproxy
->cgroup_ns
;
6138 return ns
? &ns
->ns
: NULL
;
6141 static void cgroupns_put(struct ns_common
*ns
)
6143 put_cgroup_ns(to_cg_ns(ns
));
6146 const struct proc_ns_operations cgroupns_operations
= {
6148 .type
= CLONE_NEWCGROUP
,
6149 .get
= cgroupns_get
,
6150 .put
= cgroupns_put
,
6151 .install
= cgroupns_install
,
6154 static __init
int cgroup_namespaces_init(void)
6158 subsys_initcall(cgroup_namespaces_init
);
6160 #ifdef CONFIG_CGROUP_DEBUG
6161 static struct cgroup_subsys_state
*
6162 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
6164 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
6167 return ERR_PTR(-ENOMEM
);
6172 static void debug_css_free(struct cgroup_subsys_state
*css
)
6177 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
6180 return cgroup_task_count(css
->cgroup
);
6183 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
6186 return (u64
)(unsigned long)current
->cgroups
;
6189 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
6195 count
= atomic_read(&task_css_set(current
)->refcount
);
6200 static int current_css_set_cg_links_read(struct seq_file
*seq
, void *v
)
6202 struct cgrp_cset_link
*link
;
6203 struct css_set
*cset
;
6206 name_buf
= kmalloc(NAME_MAX
+ 1, GFP_KERNEL
);
6210 spin_lock_bh(&css_set_lock
);
6212 cset
= rcu_dereference(current
->cgroups
);
6213 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
6214 struct cgroup
*c
= link
->cgrp
;
6216 cgroup_name(c
, name_buf
, NAME_MAX
+ 1);
6217 seq_printf(seq
, "Root %d group %s\n",
6218 c
->root
->hierarchy_id
, name_buf
);
6221 spin_unlock_bh(&css_set_lock
);
6226 #define MAX_TASKS_SHOWN_PER_CSS 25
6227 static int cgroup_css_links_read(struct seq_file
*seq
, void *v
)
6229 struct cgroup_subsys_state
*css
= seq_css(seq
);
6230 struct cgrp_cset_link
*link
;
6232 spin_lock_bh(&css_set_lock
);
6233 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
6234 struct css_set
*cset
= link
->cset
;
6235 struct task_struct
*task
;
6238 seq_printf(seq
, "css_set %p\n", cset
);
6240 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
6241 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
6243 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
6246 list_for_each_entry(task
, &cset
->mg_tasks
, cg_list
) {
6247 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
6249 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
6253 seq_puts(seq
, " ...\n");
6255 spin_unlock_bh(&css_set_lock
);
6259 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
6261 return (!cgroup_is_populated(css
->cgroup
) &&
6262 !css_has_online_children(&css
->cgroup
->self
));
6265 static struct cftype debug_files
[] = {
6267 .name
= "taskcount",
6268 .read_u64
= debug_taskcount_read
,
6272 .name
= "current_css_set",
6273 .read_u64
= current_css_set_read
,
6277 .name
= "current_css_set_refcount",
6278 .read_u64
= current_css_set_refcount_read
,
6282 .name
= "current_css_set_cg_links",
6283 .seq_show
= current_css_set_cg_links_read
,
6287 .name
= "cgroup_css_links",
6288 .seq_show
= cgroup_css_links_read
,
6292 .name
= "releasable",
6293 .read_u64
= releasable_read
,
6299 struct cgroup_subsys debug_cgrp_subsys
= {
6300 .css_alloc
= debug_css_alloc
,
6301 .css_free
= debug_css_free
,
6302 .legacy_cftypes
= debug_files
,
6304 #endif /* CONFIG_CGROUP_DEBUG */