2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include <linux/cgroup.h>
32 #include <linux/cred.h>
33 #include <linux/ctype.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/list.h>
38 #include <linux/magic.h>
40 #include <linux/mutex.h>
41 #include <linux/mount.h>
42 #include <linux/pagemap.h>
43 #include <linux/proc_fs.h>
44 #include <linux/rcupdate.h>
45 #include <linux/sched.h>
46 #include <linux/slab.h>
47 #include <linux/spinlock.h>
48 #include <linux/rwsem.h>
49 #include <linux/percpu-rwsem.h>
50 #include <linux/string.h>
51 #include <linux/sort.h>
52 #include <linux/kmod.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hashtable.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/idr.h>
58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
59 #include <linux/kthread.h>
60 #include <linux/delay.h>
62 #include <linux/atomic.h>
65 * pidlists linger the following amount before being destroyed. The goal
66 * is avoiding frequent destruction in the middle of consecutive read calls
67 * Expiring in the middle is a performance problem not a correctness one.
68 * 1 sec should be enough.
70 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
72 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
76 * cgroup_mutex is the master lock. Any modification to cgroup or its
77 * hierarchy must be performed while holding it.
79 * css_set_rwsem protects task->cgroups pointer, the list of css_set
80 * objects, and the chain of tasks off each css_set.
82 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
83 * cgroup.h can use them for lockdep annotations.
85 #ifdef CONFIG_PROVE_RCU
86 DEFINE_MUTEX(cgroup_mutex
);
87 DECLARE_RWSEM(css_set_rwsem
);
88 EXPORT_SYMBOL_GPL(cgroup_mutex
);
89 EXPORT_SYMBOL_GPL(css_set_rwsem
);
91 static DEFINE_MUTEX(cgroup_mutex
);
92 static DECLARE_RWSEM(css_set_rwsem
);
96 * Protects cgroup_idr and css_idr so that IDs can be released without
97 * grabbing cgroup_mutex.
99 static DEFINE_SPINLOCK(cgroup_idr_lock
);
102 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
103 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
105 static DEFINE_SPINLOCK(release_agent_path_lock
);
107 struct percpu_rw_semaphore cgroup_threadgroup_rwsem
;
109 #define cgroup_assert_mutex_or_rcu_locked() \
110 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
111 !lockdep_is_held(&cgroup_mutex), \
112 "cgroup_mutex or RCU read lock required");
115 * cgroup destruction makes heavy use of work items and there can be a lot
116 * of concurrent destructions. Use a separate workqueue so that cgroup
117 * destruction work items don't end up filling up max_active of system_wq
118 * which may lead to deadlock.
120 static struct workqueue_struct
*cgroup_destroy_wq
;
123 * pidlist destructions need to be flushed on cgroup destruction. Use a
124 * separate workqueue as flush domain.
126 static struct workqueue_struct
*cgroup_pidlist_destroy_wq
;
128 /* generate an array of cgroup subsystem pointers */
129 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
130 static struct cgroup_subsys
*cgroup_subsys
[] = {
131 #include <linux/cgroup_subsys.h>
135 /* array of cgroup subsystem names */
136 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
137 static const char *cgroup_subsys_name
[] = {
138 #include <linux/cgroup_subsys.h>
142 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
144 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
145 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
146 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
147 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
148 #include <linux/cgroup_subsys.h>
151 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
152 static struct static_key_true
*cgroup_subsys_enabled_key
[] = {
153 #include <linux/cgroup_subsys.h>
157 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
158 static struct static_key_true
*cgroup_subsys_on_dfl_key
[] = {
159 #include <linux/cgroup_subsys.h>
164 * The default hierarchy, reserved for the subsystems that are otherwise
165 * unattached - it never has more than a single cgroup, and all tasks are
166 * part of that cgroup.
168 struct cgroup_root cgrp_dfl_root
;
169 EXPORT_SYMBOL_GPL(cgrp_dfl_root
);
172 * The default hierarchy always exists but is hidden until mounted for the
173 * first time. This is for backward compatibility.
175 static bool cgrp_dfl_root_visible
;
178 * Set by the boot param of the same name and makes subsystems with NULL
179 * ->dfl_files to use ->legacy_files on the default hierarchy.
181 static bool cgroup_legacy_files_on_dfl
;
183 /* some controllers are not supported in the default hierarchy */
184 static unsigned long cgrp_dfl_root_inhibit_ss_mask
;
186 /* The list of hierarchy roots */
188 static LIST_HEAD(cgroup_roots
);
189 static int cgroup_root_count
;
191 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
192 static DEFINE_IDR(cgroup_hierarchy_idr
);
195 * Assign a monotonically increasing serial number to csses. It guarantees
196 * cgroups with bigger numbers are newer than those with smaller numbers.
197 * Also, as csses are always appended to the parent's ->children list, it
198 * guarantees that sibling csses are always sorted in the ascending serial
199 * number order on the list. Protected by cgroup_mutex.
201 static u64 css_serial_nr_next
= 1;
204 * These bitmask flags indicate whether tasks in the fork and exit paths have
205 * fork/exit handlers to call. This avoids us having to do extra work in the
206 * fork/exit path to check which subsystems have fork/exit callbacks.
208 static unsigned long have_fork_callback __read_mostly
;
209 static unsigned long have_exit_callback __read_mostly
;
211 /* Ditto for the can_fork callback. */
212 static unsigned long have_canfork_callback __read_mostly
;
214 static struct cftype cgroup_dfl_base_files
[];
215 static struct cftype cgroup_legacy_base_files
[];
217 static int rebind_subsystems(struct cgroup_root
*dst_root
,
218 unsigned long ss_mask
);
219 static int cgroup_destroy_locked(struct cgroup
*cgrp
);
220 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
222 static void css_release(struct percpu_ref
*ref
);
223 static void kill_css(struct cgroup_subsys_state
*css
);
224 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
225 struct cgroup
*cgrp
, struct cftype cfts
[],
229 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
230 * @ssid: subsys ID of interest
232 * cgroup_subsys_enabled() can only be used with literal subsys names which
233 * is fine for individual subsystems but unsuitable for cgroup core. This
234 * is slower static_key_enabled() based test indexed by @ssid.
236 static bool cgroup_ssid_enabled(int ssid
)
238 return static_key_enabled(cgroup_subsys_enabled_key
[ssid
]);
242 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
243 * @cgrp: the cgroup of interest
245 * The default hierarchy is the v2 interface of cgroup and this function
246 * can be used to test whether a cgroup is on the default hierarchy for
247 * cases where a subsystem should behave differnetly depending on the
250 * The set of behaviors which change on the default hierarchy are still
251 * being determined and the mount option is prefixed with __DEVEL__.
253 * List of changed behaviors:
255 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
256 * and "name" are disallowed.
258 * - When mounting an existing superblock, mount options should match.
260 * - Remount is disallowed.
262 * - rename(2) is disallowed.
264 * - "tasks" is removed. Everything should be at process granularity. Use
265 * "cgroup.procs" instead.
267 * - "cgroup.procs" is not sorted. pids will be unique unless they got
268 * recycled inbetween reads.
270 * - "release_agent" and "notify_on_release" are removed. Replacement
271 * notification mechanism will be implemented.
273 * - "cgroup.clone_children" is removed.
275 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
276 * and its descendants contain no task; otherwise, 1. The file also
277 * generates kernfs notification which can be monitored through poll and
278 * [di]notify when the value of the file changes.
280 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
281 * take masks of ancestors with non-empty cpus/mems, instead of being
282 * moved to an ancestor.
284 * - cpuset: a task can be moved into an empty cpuset, and again it takes
285 * masks of ancestors.
287 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
290 * - blkcg: blk-throttle becomes properly hierarchical.
292 * - debug: disallowed on the default hierarchy.
294 static bool cgroup_on_dfl(const struct cgroup
*cgrp
)
296 return cgrp
->root
== &cgrp_dfl_root
;
299 /* IDR wrappers which synchronize using cgroup_idr_lock */
300 static int cgroup_idr_alloc(struct idr
*idr
, void *ptr
, int start
, int end
,
305 idr_preload(gfp_mask
);
306 spin_lock_bh(&cgroup_idr_lock
);
307 ret
= idr_alloc(idr
, ptr
, start
, end
, gfp_mask
& ~__GFP_WAIT
);
308 spin_unlock_bh(&cgroup_idr_lock
);
313 static void *cgroup_idr_replace(struct idr
*idr
, void *ptr
, int id
)
317 spin_lock_bh(&cgroup_idr_lock
);
318 ret
= idr_replace(idr
, ptr
, id
);
319 spin_unlock_bh(&cgroup_idr_lock
);
323 static void cgroup_idr_remove(struct idr
*idr
, int id
)
325 spin_lock_bh(&cgroup_idr_lock
);
327 spin_unlock_bh(&cgroup_idr_lock
);
330 static struct cgroup
*cgroup_parent(struct cgroup
*cgrp
)
332 struct cgroup_subsys_state
*parent_css
= cgrp
->self
.parent
;
335 return container_of(parent_css
, struct cgroup
, self
);
340 * cgroup_css - obtain a cgroup's css for the specified subsystem
341 * @cgrp: the cgroup of interest
342 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
344 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
345 * function must be called either under cgroup_mutex or rcu_read_lock() and
346 * the caller is responsible for pinning the returned css if it wants to
347 * keep accessing it outside the said locks. This function may return
348 * %NULL if @cgrp doesn't have @subsys_id enabled.
350 static struct cgroup_subsys_state
*cgroup_css(struct cgroup
*cgrp
,
351 struct cgroup_subsys
*ss
)
354 return rcu_dereference_check(cgrp
->subsys
[ss
->id
],
355 lockdep_is_held(&cgroup_mutex
));
361 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
362 * @cgrp: the cgroup of interest
363 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
365 * Similar to cgroup_css() but returns the effective css, which is defined
366 * as the matching css of the nearest ancestor including self which has @ss
367 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
368 * function is guaranteed to return non-NULL css.
370 static struct cgroup_subsys_state
*cgroup_e_css(struct cgroup
*cgrp
,
371 struct cgroup_subsys
*ss
)
373 lockdep_assert_held(&cgroup_mutex
);
378 if (!(cgrp
->root
->subsys_mask
& (1 << ss
->id
)))
382 * This function is used while updating css associations and thus
383 * can't test the csses directly. Use ->child_subsys_mask.
385 while (cgroup_parent(cgrp
) &&
386 !(cgroup_parent(cgrp
)->child_subsys_mask
& (1 << ss
->id
)))
387 cgrp
= cgroup_parent(cgrp
);
389 return cgroup_css(cgrp
, ss
);
393 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
394 * @cgrp: the cgroup of interest
395 * @ss: the subsystem of interest
397 * Find and get the effective css of @cgrp for @ss. The effective css is
398 * defined as the matching css of the nearest ancestor including self which
399 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
400 * the root css is returned, so this function always returns a valid css.
401 * The returned css must be put using css_put().
403 struct cgroup_subsys_state
*cgroup_get_e_css(struct cgroup
*cgrp
,
404 struct cgroup_subsys
*ss
)
406 struct cgroup_subsys_state
*css
;
411 css
= cgroup_css(cgrp
, ss
);
413 if (css
&& css_tryget_online(css
))
415 cgrp
= cgroup_parent(cgrp
);
418 css
= init_css_set
.subsys
[ss
->id
];
425 /* convenient tests for these bits */
426 static inline bool cgroup_is_dead(const struct cgroup
*cgrp
)
428 return !(cgrp
->self
.flags
& CSS_ONLINE
);
431 static void cgroup_get(struct cgroup
*cgrp
)
433 WARN_ON_ONCE(cgroup_is_dead(cgrp
));
434 css_get(&cgrp
->self
);
437 static bool cgroup_tryget(struct cgroup
*cgrp
)
439 return css_tryget(&cgrp
->self
);
442 static void cgroup_put(struct cgroup
*cgrp
)
444 css_put(&cgrp
->self
);
447 struct cgroup_subsys_state
*of_css(struct kernfs_open_file
*of
)
449 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
450 struct cftype
*cft
= of_cft(of
);
453 * This is open and unprotected implementation of cgroup_css().
454 * seq_css() is only called from a kernfs file operation which has
455 * an active reference on the file. Because all the subsystem
456 * files are drained before a css is disassociated with a cgroup,
457 * the matching css from the cgroup's subsys table is guaranteed to
458 * be and stay valid until the enclosing operation is complete.
461 return rcu_dereference_raw(cgrp
->subsys
[cft
->ss
->id
]);
465 EXPORT_SYMBOL_GPL(of_css
);
468 * cgroup_is_descendant - test ancestry
469 * @cgrp: the cgroup to be tested
470 * @ancestor: possible ancestor of @cgrp
472 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
473 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
474 * and @ancestor are accessible.
476 bool cgroup_is_descendant(struct cgroup
*cgrp
, struct cgroup
*ancestor
)
479 if (cgrp
== ancestor
)
481 cgrp
= cgroup_parent(cgrp
);
486 static int notify_on_release(const struct cgroup
*cgrp
)
488 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
492 * for_each_css - iterate all css's of a cgroup
493 * @css: the iteration cursor
494 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
495 * @cgrp: the target cgroup to iterate css's of
497 * Should be called under cgroup_[tree_]mutex.
499 #define for_each_css(css, ssid, cgrp) \
500 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
501 if (!((css) = rcu_dereference_check( \
502 (cgrp)->subsys[(ssid)], \
503 lockdep_is_held(&cgroup_mutex)))) { } \
507 * for_each_e_css - iterate all effective css's of a cgroup
508 * @css: the iteration cursor
509 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
510 * @cgrp: the target cgroup to iterate css's of
512 * Should be called under cgroup_[tree_]mutex.
514 #define for_each_e_css(css, ssid, cgrp) \
515 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
516 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
521 * for_each_subsys - iterate all enabled cgroup subsystems
522 * @ss: the iteration cursor
523 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
525 #define for_each_subsys(ss, ssid) \
526 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
527 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
530 * for_each_subsys_which - filter for_each_subsys with a bitmask
531 * @ss: the iteration cursor
532 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
533 * @ss_maskp: a pointer to the bitmask
535 * The block will only run for cases where the ssid-th bit (1 << ssid) of
538 #define for_each_subsys_which(ss, ssid, ss_maskp) \
539 if (!CGROUP_SUBSYS_COUNT) /* to avoid spurious gcc warning */ \
542 for_each_set_bit(ssid, ss_maskp, CGROUP_SUBSYS_COUNT) \
543 if (((ss) = cgroup_subsys[ssid]) && false) \
547 /* iterate across the hierarchies */
548 #define for_each_root(root) \
549 list_for_each_entry((root), &cgroup_roots, root_list)
551 /* iterate over child cgrps, lock should be held throughout iteration */
552 #define cgroup_for_each_live_child(child, cgrp) \
553 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
554 if (({ lockdep_assert_held(&cgroup_mutex); \
555 cgroup_is_dead(child); })) \
559 static void cgroup_release_agent(struct work_struct
*work
);
560 static void check_for_release(struct cgroup
*cgrp
);
563 * A cgroup can be associated with multiple css_sets as different tasks may
564 * belong to different cgroups on different hierarchies. In the other
565 * direction, a css_set is naturally associated with multiple cgroups.
566 * This M:N relationship is represented by the following link structure
567 * which exists for each association and allows traversing the associations
570 struct cgrp_cset_link
{
571 /* the cgroup and css_set this link associates */
573 struct css_set
*cset
;
575 /* list of cgrp_cset_links anchored at cgrp->cset_links */
576 struct list_head cset_link
;
578 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
579 struct list_head cgrp_link
;
583 * The default css_set - used by init and its children prior to any
584 * hierarchies being mounted. It contains a pointer to the root state
585 * for each subsystem. Also used to anchor the list of css_sets. Not
586 * reference-counted, to improve performance when child cgroups
587 * haven't been created.
589 struct css_set init_css_set
= {
590 .refcount
= ATOMIC_INIT(1),
591 .cgrp_links
= LIST_HEAD_INIT(init_css_set
.cgrp_links
),
592 .tasks
= LIST_HEAD_INIT(init_css_set
.tasks
),
593 .mg_tasks
= LIST_HEAD_INIT(init_css_set
.mg_tasks
),
594 .mg_preload_node
= LIST_HEAD_INIT(init_css_set
.mg_preload_node
),
595 .mg_node
= LIST_HEAD_INIT(init_css_set
.mg_node
),
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_rwsem
);
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_rwsem
);
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_rwsem
);
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 but the caller
679 * is responsible for managing @from_cset and @to_cset's reference counts.
681 static void css_set_move_task(struct task_struct
*task
,
682 struct css_set
*from_cset
, struct css_set
*to_cset
,
685 lockdep_assert_held(&css_set_rwsem
);
688 WARN_ON_ONCE(list_empty(&task
->cg_list
));
689 list_del_init(&task
->cg_list
);
690 if (!css_set_populated(from_cset
))
691 css_set_update_populated(from_cset
, false);
693 WARN_ON_ONCE(!list_empty(&task
->cg_list
));
698 * We are synchronized through cgroup_threadgroup_rwsem
699 * against PF_EXITING setting such that we can't race
700 * against cgroup_exit() changing the css_set to
701 * init_css_set and dropping the old one.
703 WARN_ON_ONCE(task
->flags
& PF_EXITING
);
705 if (!css_set_populated(to_cset
))
706 css_set_update_populated(to_cset
, true);
707 rcu_assign_pointer(task
->cgroups
, to_cset
);
708 list_add_tail(&task
->cg_list
, use_mg_tasks
? &to_cset
->mg_tasks
:
714 * hash table for cgroup groups. This improves the performance to find
715 * an existing css_set. This hash doesn't (currently) take into
716 * account cgroups in empty hierarchies.
718 #define CSS_SET_HASH_BITS 7
719 static DEFINE_HASHTABLE(css_set_table
, CSS_SET_HASH_BITS
);
721 static unsigned long css_set_hash(struct cgroup_subsys_state
*css
[])
723 unsigned long key
= 0UL;
724 struct cgroup_subsys
*ss
;
727 for_each_subsys(ss
, i
)
728 key
+= (unsigned long)css
[i
];
729 key
= (key
>> 16) ^ key
;
734 static void put_css_set_locked(struct css_set
*cset
)
736 struct cgrp_cset_link
*link
, *tmp_link
;
737 struct cgroup_subsys
*ss
;
740 lockdep_assert_held(&css_set_rwsem
);
742 if (!atomic_dec_and_test(&cset
->refcount
))
745 /* This css_set is dead. unlink it and release cgroup refcounts */
746 for_each_subsys(ss
, ssid
)
747 list_del(&cset
->e_cset_node
[ssid
]);
748 hash_del(&cset
->hlist
);
751 list_for_each_entry_safe(link
, tmp_link
, &cset
->cgrp_links
, cgrp_link
) {
752 list_del(&link
->cset_link
);
753 list_del(&link
->cgrp_link
);
754 if (cgroup_parent(link
->cgrp
))
755 cgroup_put(link
->cgrp
);
759 kfree_rcu(cset
, rcu_head
);
762 static void put_css_set(struct css_set
*cset
)
765 * Ensure that the refcount doesn't hit zero while any readers
766 * can see it. Similar to atomic_dec_and_lock(), but for an
769 if (atomic_add_unless(&cset
->refcount
, -1, 1))
772 down_write(&css_set_rwsem
);
773 put_css_set_locked(cset
);
774 up_write(&css_set_rwsem
);
778 * refcounted get/put for css_set objects
780 static inline void get_css_set(struct css_set
*cset
)
782 atomic_inc(&cset
->refcount
);
786 * compare_css_sets - helper function for find_existing_css_set().
787 * @cset: candidate css_set being tested
788 * @old_cset: existing css_set for a task
789 * @new_cgrp: cgroup that's being entered by the task
790 * @template: desired set of css pointers in css_set (pre-calculated)
792 * Returns true if "cset" matches "old_cset" except for the hierarchy
793 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
795 static bool compare_css_sets(struct css_set
*cset
,
796 struct css_set
*old_cset
,
797 struct cgroup
*new_cgrp
,
798 struct cgroup_subsys_state
*template[])
800 struct list_head
*l1
, *l2
;
803 * On the default hierarchy, there can be csets which are
804 * associated with the same set of cgroups but different csses.
805 * Let's first ensure that csses match.
807 if (memcmp(template, cset
->subsys
, sizeof(cset
->subsys
)))
811 * Compare cgroup pointers in order to distinguish between
812 * different cgroups in hierarchies. As different cgroups may
813 * share the same effective css, this comparison is always
816 l1
= &cset
->cgrp_links
;
817 l2
= &old_cset
->cgrp_links
;
819 struct cgrp_cset_link
*link1
, *link2
;
820 struct cgroup
*cgrp1
, *cgrp2
;
824 /* See if we reached the end - both lists are equal length. */
825 if (l1
== &cset
->cgrp_links
) {
826 BUG_ON(l2
!= &old_cset
->cgrp_links
);
829 BUG_ON(l2
== &old_cset
->cgrp_links
);
831 /* Locate the cgroups associated with these links. */
832 link1
= list_entry(l1
, struct cgrp_cset_link
, cgrp_link
);
833 link2
= list_entry(l2
, struct cgrp_cset_link
, cgrp_link
);
836 /* Hierarchies should be linked in the same order. */
837 BUG_ON(cgrp1
->root
!= cgrp2
->root
);
840 * If this hierarchy is the hierarchy of the cgroup
841 * that's changing, then we need to check that this
842 * css_set points to the new cgroup; if it's any other
843 * hierarchy, then this css_set should point to the
844 * same cgroup as the old css_set.
846 if (cgrp1
->root
== new_cgrp
->root
) {
847 if (cgrp1
!= new_cgrp
)
858 * find_existing_css_set - init css array and find the matching css_set
859 * @old_cset: the css_set that we're using before the cgroup transition
860 * @cgrp: the cgroup that we're moving into
861 * @template: out param for the new set of csses, should be clear on entry
863 static struct css_set
*find_existing_css_set(struct css_set
*old_cset
,
865 struct cgroup_subsys_state
*template[])
867 struct cgroup_root
*root
= cgrp
->root
;
868 struct cgroup_subsys
*ss
;
869 struct css_set
*cset
;
874 * Build the set of subsystem state objects that we want to see in the
875 * new css_set. while subsystems can change globally, the entries here
876 * won't change, so no need for locking.
878 for_each_subsys(ss
, i
) {
879 if (root
->subsys_mask
& (1UL << i
)) {
881 * @ss is in this hierarchy, so we want the
882 * effective css from @cgrp.
884 template[i
] = cgroup_e_css(cgrp
, ss
);
887 * @ss is not in this hierarchy, so we don't want
890 template[i
] = old_cset
->subsys
[i
];
894 key
= css_set_hash(template);
895 hash_for_each_possible(css_set_table
, cset
, hlist
, key
) {
896 if (!compare_css_sets(cset
, old_cset
, cgrp
, template))
899 /* This css_set matches what we need */
903 /* No existing cgroup group matched */
907 static void free_cgrp_cset_links(struct list_head
*links_to_free
)
909 struct cgrp_cset_link
*link
, *tmp_link
;
911 list_for_each_entry_safe(link
, tmp_link
, links_to_free
, cset_link
) {
912 list_del(&link
->cset_link
);
918 * allocate_cgrp_cset_links - allocate cgrp_cset_links
919 * @count: the number of links to allocate
920 * @tmp_links: list_head the allocated links are put on
922 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
923 * through ->cset_link. Returns 0 on success or -errno.
925 static int allocate_cgrp_cset_links(int count
, struct list_head
*tmp_links
)
927 struct cgrp_cset_link
*link
;
930 INIT_LIST_HEAD(tmp_links
);
932 for (i
= 0; i
< count
; i
++) {
933 link
= kzalloc(sizeof(*link
), GFP_KERNEL
);
935 free_cgrp_cset_links(tmp_links
);
938 list_add(&link
->cset_link
, tmp_links
);
944 * link_css_set - a helper function to link a css_set to a cgroup
945 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
946 * @cset: the css_set to be linked
947 * @cgrp: the destination cgroup
949 static void link_css_set(struct list_head
*tmp_links
, struct css_set
*cset
,
952 struct cgrp_cset_link
*link
;
954 BUG_ON(list_empty(tmp_links
));
956 if (cgroup_on_dfl(cgrp
))
957 cset
->dfl_cgrp
= cgrp
;
959 link
= list_first_entry(tmp_links
, struct cgrp_cset_link
, cset_link
);
964 * Always add links to the tail of the lists so that the lists are
965 * in choronological order.
967 list_move_tail(&link
->cset_link
, &cgrp
->cset_links
);
968 list_add_tail(&link
->cgrp_link
, &cset
->cgrp_links
);
970 if (cgroup_parent(cgrp
))
975 * find_css_set - return a new css_set with one cgroup updated
976 * @old_cset: the baseline css_set
977 * @cgrp: the cgroup to be updated
979 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
980 * substituted into the appropriate hierarchy.
982 static struct css_set
*find_css_set(struct css_set
*old_cset
,
985 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
] = { };
986 struct css_set
*cset
;
987 struct list_head tmp_links
;
988 struct cgrp_cset_link
*link
;
989 struct cgroup_subsys
*ss
;
993 lockdep_assert_held(&cgroup_mutex
);
995 /* First see if we already have a cgroup group that matches
997 down_read(&css_set_rwsem
);
998 cset
= find_existing_css_set(old_cset
, cgrp
, template);
1001 up_read(&css_set_rwsem
);
1006 cset
= kzalloc(sizeof(*cset
), GFP_KERNEL
);
1010 /* Allocate all the cgrp_cset_link objects that we'll need */
1011 if (allocate_cgrp_cset_links(cgroup_root_count
, &tmp_links
) < 0) {
1016 atomic_set(&cset
->refcount
, 1);
1017 INIT_LIST_HEAD(&cset
->cgrp_links
);
1018 INIT_LIST_HEAD(&cset
->tasks
);
1019 INIT_LIST_HEAD(&cset
->mg_tasks
);
1020 INIT_LIST_HEAD(&cset
->mg_preload_node
);
1021 INIT_LIST_HEAD(&cset
->mg_node
);
1022 INIT_HLIST_NODE(&cset
->hlist
);
1024 /* Copy the set of subsystem state objects generated in
1025 * find_existing_css_set() */
1026 memcpy(cset
->subsys
, template, sizeof(cset
->subsys
));
1028 down_write(&css_set_rwsem
);
1029 /* Add reference counts and links from the new css_set. */
1030 list_for_each_entry(link
, &old_cset
->cgrp_links
, cgrp_link
) {
1031 struct cgroup
*c
= link
->cgrp
;
1033 if (c
->root
== cgrp
->root
)
1035 link_css_set(&tmp_links
, cset
, c
);
1038 BUG_ON(!list_empty(&tmp_links
));
1042 /* Add @cset to the hash table */
1043 key
= css_set_hash(cset
->subsys
);
1044 hash_add(css_set_table
, &cset
->hlist
, key
);
1046 for_each_subsys(ss
, ssid
)
1047 list_add_tail(&cset
->e_cset_node
[ssid
],
1048 &cset
->subsys
[ssid
]->cgroup
->e_csets
[ssid
]);
1050 up_write(&css_set_rwsem
);
1055 static struct cgroup_root
*cgroup_root_from_kf(struct kernfs_root
*kf_root
)
1057 struct cgroup
*root_cgrp
= kf_root
->kn
->priv
;
1059 return root_cgrp
->root
;
1062 static int cgroup_init_root_id(struct cgroup_root
*root
)
1066 lockdep_assert_held(&cgroup_mutex
);
1068 id
= idr_alloc_cyclic(&cgroup_hierarchy_idr
, root
, 0, 0, GFP_KERNEL
);
1072 root
->hierarchy_id
= id
;
1076 static void cgroup_exit_root_id(struct cgroup_root
*root
)
1078 lockdep_assert_held(&cgroup_mutex
);
1080 if (root
->hierarchy_id
) {
1081 idr_remove(&cgroup_hierarchy_idr
, root
->hierarchy_id
);
1082 root
->hierarchy_id
= 0;
1086 static void cgroup_free_root(struct cgroup_root
*root
)
1089 /* hierarchy ID should already have been released */
1090 WARN_ON_ONCE(root
->hierarchy_id
);
1092 idr_destroy(&root
->cgroup_idr
);
1097 static void cgroup_destroy_root(struct cgroup_root
*root
)
1099 struct cgroup
*cgrp
= &root
->cgrp
;
1100 struct cgrp_cset_link
*link
, *tmp_link
;
1102 mutex_lock(&cgroup_mutex
);
1104 BUG_ON(atomic_read(&root
->nr_cgrps
));
1105 BUG_ON(!list_empty(&cgrp
->self
.children
));
1107 /* Rebind all subsystems back to the default hierarchy */
1108 rebind_subsystems(&cgrp_dfl_root
, root
->subsys_mask
);
1111 * Release all the links from cset_links to this hierarchy's
1114 down_write(&css_set_rwsem
);
1116 list_for_each_entry_safe(link
, tmp_link
, &cgrp
->cset_links
, cset_link
) {
1117 list_del(&link
->cset_link
);
1118 list_del(&link
->cgrp_link
);
1121 up_write(&css_set_rwsem
);
1123 if (!list_empty(&root
->root_list
)) {
1124 list_del(&root
->root_list
);
1125 cgroup_root_count
--;
1128 cgroup_exit_root_id(root
);
1130 mutex_unlock(&cgroup_mutex
);
1132 kernfs_destroy_root(root
->kf_root
);
1133 cgroup_free_root(root
);
1136 /* look up cgroup associated with given css_set on the specified hierarchy */
1137 static struct cgroup
*cset_cgroup_from_root(struct css_set
*cset
,
1138 struct cgroup_root
*root
)
1140 struct cgroup
*res
= NULL
;
1142 lockdep_assert_held(&cgroup_mutex
);
1143 lockdep_assert_held(&css_set_rwsem
);
1145 if (cset
== &init_css_set
) {
1148 struct cgrp_cset_link
*link
;
1150 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
1151 struct cgroup
*c
= link
->cgrp
;
1153 if (c
->root
== root
) {
1165 * Return the cgroup for "task" from the given hierarchy. Must be
1166 * called with cgroup_mutex and css_set_rwsem held.
1168 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
1169 struct cgroup_root
*root
)
1172 * No need to lock the task - since we hold cgroup_mutex the
1173 * task can't change groups, so the only thing that can happen
1174 * is that it exits and its css is set back to init_css_set.
1176 return cset_cgroup_from_root(task_css_set(task
), root
);
1180 * A task must hold cgroup_mutex to modify cgroups.
1182 * Any task can increment and decrement the count field without lock.
1183 * So in general, code holding cgroup_mutex can't rely on the count
1184 * field not changing. However, if the count goes to zero, then only
1185 * cgroup_attach_task() can increment it again. Because a count of zero
1186 * means that no tasks are currently attached, therefore there is no
1187 * way a task attached to that cgroup can fork (the other way to
1188 * increment the count). So code holding cgroup_mutex can safely
1189 * assume that if the count is zero, it will stay zero. Similarly, if
1190 * a task holds cgroup_mutex on a cgroup with zero count, it
1191 * knows that the cgroup won't be removed, as cgroup_rmdir()
1194 * A cgroup can only be deleted if both its 'count' of using tasks
1195 * is zero, and its list of 'children' cgroups is empty. Since all
1196 * tasks in the system use _some_ cgroup, and since there is always at
1197 * least one task in the system (init, pid == 1), therefore, root cgroup
1198 * always has either children cgroups and/or using tasks. So we don't
1199 * need a special hack to ensure that root cgroup cannot be deleted.
1201 * P.S. One more locking exception. RCU is used to guard the
1202 * update of a tasks cgroup pointer by cgroup_attach_task()
1205 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
;
1206 static const struct file_operations proc_cgroupstats_operations
;
1208 static char *cgroup_file_name(struct cgroup
*cgrp
, const struct cftype
*cft
,
1211 struct cgroup_subsys
*ss
= cft
->ss
;
1213 if (cft
->ss
&& !(cft
->flags
& CFTYPE_NO_PREFIX
) &&
1214 !(cgrp
->root
->flags
& CGRP_ROOT_NOPREFIX
))
1215 snprintf(buf
, CGROUP_FILE_NAME_MAX
, "%s.%s",
1216 cgroup_on_dfl(cgrp
) ? ss
->name
: ss
->legacy_name
,
1219 strncpy(buf
, cft
->name
, CGROUP_FILE_NAME_MAX
);
1224 * cgroup_file_mode - deduce file mode of a control file
1225 * @cft: the control file in question
1227 * S_IRUGO for read, S_IWUSR for write.
1229 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
1233 if (cft
->read_u64
|| cft
->read_s64
|| cft
->seq_show
)
1236 if (cft
->write_u64
|| cft
->write_s64
|| cft
->write
) {
1237 if (cft
->flags
& CFTYPE_WORLD_WRITABLE
)
1247 * cgroup_calc_child_subsys_mask - calculate child_subsys_mask
1248 * @cgrp: the target cgroup
1249 * @subtree_control: the new subtree_control mask to consider
1251 * On the default hierarchy, a subsystem may request other subsystems to be
1252 * enabled together through its ->depends_on mask. In such cases, more
1253 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1255 * This function calculates which subsystems need to be enabled if
1256 * @subtree_control is to be applied to @cgrp. The returned mask is always
1257 * a superset of @subtree_control and follows the usual hierarchy rules.
1259 static unsigned long cgroup_calc_child_subsys_mask(struct cgroup
*cgrp
,
1260 unsigned long subtree_control
)
1262 struct cgroup
*parent
= cgroup_parent(cgrp
);
1263 unsigned long cur_ss_mask
= subtree_control
;
1264 struct cgroup_subsys
*ss
;
1267 lockdep_assert_held(&cgroup_mutex
);
1269 if (!cgroup_on_dfl(cgrp
))
1273 unsigned long new_ss_mask
= cur_ss_mask
;
1275 for_each_subsys_which(ss
, ssid
, &cur_ss_mask
)
1276 new_ss_mask
|= ss
->depends_on
;
1279 * Mask out subsystems which aren't available. This can
1280 * happen only if some depended-upon subsystems were bound
1281 * to non-default hierarchies.
1284 new_ss_mask
&= parent
->child_subsys_mask
;
1286 new_ss_mask
&= cgrp
->root
->subsys_mask
;
1288 if (new_ss_mask
== cur_ss_mask
)
1290 cur_ss_mask
= new_ss_mask
;
1297 * cgroup_refresh_child_subsys_mask - update child_subsys_mask
1298 * @cgrp: the target cgroup
1300 * Update @cgrp->child_subsys_mask according to the current
1301 * @cgrp->subtree_control using cgroup_calc_child_subsys_mask().
1303 static void cgroup_refresh_child_subsys_mask(struct cgroup
*cgrp
)
1305 cgrp
->child_subsys_mask
=
1306 cgroup_calc_child_subsys_mask(cgrp
, cgrp
->subtree_control
);
1310 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1311 * @kn: the kernfs_node being serviced
1313 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1314 * the method finishes if locking succeeded. Note that once this function
1315 * returns the cgroup returned by cgroup_kn_lock_live() may become
1316 * inaccessible any time. If the caller intends to continue to access the
1317 * cgroup, it should pin it before invoking this function.
1319 static void cgroup_kn_unlock(struct kernfs_node
*kn
)
1321 struct cgroup
*cgrp
;
1323 if (kernfs_type(kn
) == KERNFS_DIR
)
1326 cgrp
= kn
->parent
->priv
;
1328 mutex_unlock(&cgroup_mutex
);
1330 kernfs_unbreak_active_protection(kn
);
1335 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1336 * @kn: the kernfs_node being serviced
1338 * This helper is to be used by a cgroup kernfs method currently servicing
1339 * @kn. It breaks the active protection, performs cgroup locking and
1340 * verifies that the associated cgroup is alive. Returns the cgroup if
1341 * alive; otherwise, %NULL. A successful return should be undone by a
1342 * matching cgroup_kn_unlock() invocation.
1344 * Any cgroup kernfs method implementation which requires locking the
1345 * associated cgroup should use this helper. It avoids nesting cgroup
1346 * locking under kernfs active protection and allows all kernfs operations
1347 * including self-removal.
1349 static struct cgroup
*cgroup_kn_lock_live(struct kernfs_node
*kn
)
1351 struct cgroup
*cgrp
;
1353 if (kernfs_type(kn
) == KERNFS_DIR
)
1356 cgrp
= kn
->parent
->priv
;
1359 * We're gonna grab cgroup_mutex which nests outside kernfs
1360 * active_ref. cgroup liveliness check alone provides enough
1361 * protection against removal. Ensure @cgrp stays accessible and
1362 * break the active_ref protection.
1364 if (!cgroup_tryget(cgrp
))
1366 kernfs_break_active_protection(kn
);
1368 mutex_lock(&cgroup_mutex
);
1370 if (!cgroup_is_dead(cgrp
))
1373 cgroup_kn_unlock(kn
);
1377 static void cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
1379 char name
[CGROUP_FILE_NAME_MAX
];
1381 lockdep_assert_held(&cgroup_mutex
);
1382 kernfs_remove_by_name(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
));
1386 * css_clear_dir - remove subsys files in a cgroup directory
1388 * @cgrp_override: specify if target cgroup is different from css->cgroup
1390 static void css_clear_dir(struct cgroup_subsys_state
*css
,
1391 struct cgroup
*cgrp_override
)
1393 struct cgroup
*cgrp
= cgrp_override
?: css
->cgroup
;
1394 struct cftype
*cfts
;
1396 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
)
1397 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1401 * css_populate_dir - create subsys files in a cgroup directory
1403 * @cgrp_overried: specify if target cgroup is different from css->cgroup
1405 * On failure, no file is added.
1407 static int css_populate_dir(struct cgroup_subsys_state
*css
,
1408 struct cgroup
*cgrp_override
)
1410 struct cgroup
*cgrp
= cgrp_override
?: css
->cgroup
;
1411 struct cftype
*cfts
, *failed_cfts
;
1415 if (cgroup_on_dfl(cgrp
))
1416 cfts
= cgroup_dfl_base_files
;
1418 cfts
= cgroup_legacy_base_files
;
1420 return cgroup_addrm_files(&cgrp
->self
, cgrp
, cfts
, true);
1423 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1424 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, true);
1432 list_for_each_entry(cfts
, &css
->ss
->cfts
, node
) {
1433 if (cfts
== failed_cfts
)
1435 cgroup_addrm_files(css
, cgrp
, cfts
, false);
1440 static int rebind_subsystems(struct cgroup_root
*dst_root
,
1441 unsigned long ss_mask
)
1443 struct cgroup
*dcgrp
= &dst_root
->cgrp
;
1444 struct cgroup_subsys
*ss
;
1445 unsigned long tmp_ss_mask
;
1448 lockdep_assert_held(&cgroup_mutex
);
1450 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
1451 /* if @ss has non-root csses attached to it, can't move */
1452 if (css_next_child(NULL
, cgroup_css(&ss
->root
->cgrp
, ss
)))
1455 /* can't move between two non-dummy roots either */
1456 if (ss
->root
!= &cgrp_dfl_root
&& dst_root
!= &cgrp_dfl_root
)
1460 /* skip creating root files on dfl_root for inhibited subsystems */
1461 tmp_ss_mask
= ss_mask
;
1462 if (dst_root
== &cgrp_dfl_root
)
1463 tmp_ss_mask
&= ~cgrp_dfl_root_inhibit_ss_mask
;
1465 for_each_subsys_which(ss
, ssid
, &tmp_ss_mask
) {
1466 struct cgroup
*scgrp
= &ss
->root
->cgrp
;
1469 ret
= css_populate_dir(cgroup_css(scgrp
, ss
), dcgrp
);
1474 * Rebinding back to the default root is not allowed to
1475 * fail. Using both default and non-default roots should
1476 * be rare. Moving subsystems back and forth even more so.
1477 * Just warn about it and continue.
1479 if (dst_root
== &cgrp_dfl_root
) {
1480 if (cgrp_dfl_root_visible
) {
1481 pr_warn("failed to create files (%d) while rebinding 0x%lx to default root\n",
1483 pr_warn("you may retry by moving them to a different hierarchy and unbinding\n");
1488 for_each_subsys_which(ss
, tssid
, &tmp_ss_mask
) {
1491 css_clear_dir(cgroup_css(scgrp
, ss
), dcgrp
);
1497 * Nothing can fail from this point on. Remove files for the
1498 * removed subsystems and rebind each subsystem.
1500 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
1501 struct cgroup_root
*src_root
= ss
->root
;
1502 struct cgroup
*scgrp
= &src_root
->cgrp
;
1503 struct cgroup_subsys_state
*css
= cgroup_css(scgrp
, ss
);
1504 struct css_set
*cset
;
1506 WARN_ON(!css
|| cgroup_css(dcgrp
, ss
));
1508 css_clear_dir(css
, NULL
);
1510 RCU_INIT_POINTER(scgrp
->subsys
[ssid
], NULL
);
1511 rcu_assign_pointer(dcgrp
->subsys
[ssid
], css
);
1512 ss
->root
= dst_root
;
1513 css
->cgroup
= dcgrp
;
1515 down_write(&css_set_rwsem
);
1516 hash_for_each(css_set_table
, i
, cset
, hlist
)
1517 list_move_tail(&cset
->e_cset_node
[ss
->id
],
1518 &dcgrp
->e_csets
[ss
->id
]);
1519 up_write(&css_set_rwsem
);
1521 src_root
->subsys_mask
&= ~(1 << ssid
);
1522 scgrp
->subtree_control
&= ~(1 << ssid
);
1523 cgroup_refresh_child_subsys_mask(scgrp
);
1525 /* default hierarchy doesn't enable controllers by default */
1526 dst_root
->subsys_mask
|= 1 << ssid
;
1527 if (dst_root
== &cgrp_dfl_root
) {
1528 static_branch_enable(cgroup_subsys_on_dfl_key
[ssid
]);
1530 dcgrp
->subtree_control
|= 1 << ssid
;
1531 cgroup_refresh_child_subsys_mask(dcgrp
);
1532 static_branch_disable(cgroup_subsys_on_dfl_key
[ssid
]);
1539 kernfs_activate(dcgrp
->kn
);
1543 static int cgroup_show_options(struct seq_file
*seq
,
1544 struct kernfs_root
*kf_root
)
1546 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1547 struct cgroup_subsys
*ss
;
1550 if (root
!= &cgrp_dfl_root
)
1551 for_each_subsys(ss
, ssid
)
1552 if (root
->subsys_mask
& (1 << ssid
))
1553 seq_show_option(seq
, ss
->legacy_name
, NULL
);
1554 if (root
->flags
& CGRP_ROOT_NOPREFIX
)
1555 seq_puts(seq
, ",noprefix");
1556 if (root
->flags
& CGRP_ROOT_XATTR
)
1557 seq_puts(seq
, ",xattr");
1559 spin_lock(&release_agent_path_lock
);
1560 if (strlen(root
->release_agent_path
))
1561 seq_show_option(seq
, "release_agent",
1562 root
->release_agent_path
);
1563 spin_unlock(&release_agent_path_lock
);
1565 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
))
1566 seq_puts(seq
, ",clone_children");
1567 if (strlen(root
->name
))
1568 seq_show_option(seq
, "name", root
->name
);
1572 struct cgroup_sb_opts
{
1573 unsigned long subsys_mask
;
1575 char *release_agent
;
1576 bool cpuset_clone_children
;
1578 /* User explicitly requested empty subsystem */
1582 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1584 char *token
, *o
= data
;
1585 bool all_ss
= false, one_ss
= false;
1586 unsigned long mask
= -1UL;
1587 struct cgroup_subsys
*ss
;
1591 #ifdef CONFIG_CPUSETS
1592 mask
= ~(1U << cpuset_cgrp_id
);
1595 memset(opts
, 0, sizeof(*opts
));
1597 while ((token
= strsep(&o
, ",")) != NULL
) {
1602 if (!strcmp(token
, "none")) {
1603 /* Explicitly have no subsystems */
1607 if (!strcmp(token
, "all")) {
1608 /* Mutually exclusive option 'all' + subsystem name */
1614 if (!strcmp(token
, "__DEVEL__sane_behavior")) {
1615 opts
->flags
|= CGRP_ROOT_SANE_BEHAVIOR
;
1618 if (!strcmp(token
, "noprefix")) {
1619 opts
->flags
|= CGRP_ROOT_NOPREFIX
;
1622 if (!strcmp(token
, "clone_children")) {
1623 opts
->cpuset_clone_children
= true;
1626 if (!strcmp(token
, "xattr")) {
1627 opts
->flags
|= CGRP_ROOT_XATTR
;
1630 if (!strncmp(token
, "release_agent=", 14)) {
1631 /* Specifying two release agents is forbidden */
1632 if (opts
->release_agent
)
1634 opts
->release_agent
=
1635 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1636 if (!opts
->release_agent
)
1640 if (!strncmp(token
, "name=", 5)) {
1641 const char *name
= token
+ 5;
1642 /* Can't specify an empty name */
1645 /* Must match [\w.-]+ */
1646 for (i
= 0; i
< strlen(name
); i
++) {
1650 if ((c
== '.') || (c
== '-') || (c
== '_'))
1654 /* Specifying two names is forbidden */
1657 opts
->name
= kstrndup(name
,
1658 MAX_CGROUP_ROOT_NAMELEN
- 1,
1666 for_each_subsys(ss
, i
) {
1667 if (strcmp(token
, ss
->legacy_name
))
1669 if (!cgroup_ssid_enabled(i
))
1672 /* Mutually exclusive option 'all' + subsystem name */
1675 opts
->subsys_mask
|= (1 << i
);
1680 if (i
== CGROUP_SUBSYS_COUNT
)
1684 if (opts
->flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1685 pr_warn("sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1687 pr_err("sane_behavior: no other mount options allowed\n");
1694 * If the 'all' option was specified select all the subsystems,
1695 * otherwise if 'none', 'name=' and a subsystem name options were
1696 * not specified, let's default to 'all'
1698 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
))
1699 for_each_subsys(ss
, i
)
1700 if (cgroup_ssid_enabled(i
))
1701 opts
->subsys_mask
|= (1 << i
);
1704 * We either have to specify by name or by subsystems. (So all
1705 * empty hierarchies must have a name).
1707 if (!opts
->subsys_mask
&& !opts
->name
)
1711 * Option noprefix was introduced just for backward compatibility
1712 * with the old cpuset, so we allow noprefix only if mounting just
1713 * the cpuset subsystem.
1715 if ((opts
->flags
& CGRP_ROOT_NOPREFIX
) && (opts
->subsys_mask
& mask
))
1718 /* Can't specify "none" and some subsystems */
1719 if (opts
->subsys_mask
&& opts
->none
)
1725 static int cgroup_remount(struct kernfs_root
*kf_root
, int *flags
, char *data
)
1728 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
1729 struct cgroup_sb_opts opts
;
1730 unsigned long added_mask
, removed_mask
;
1732 if (root
== &cgrp_dfl_root
) {
1733 pr_err("remount is not allowed\n");
1737 mutex_lock(&cgroup_mutex
);
1739 /* See what subsystems are wanted */
1740 ret
= parse_cgroupfs_options(data
, &opts
);
1744 if (opts
.subsys_mask
!= root
->subsys_mask
|| opts
.release_agent
)
1745 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1746 task_tgid_nr(current
), current
->comm
);
1748 added_mask
= opts
.subsys_mask
& ~root
->subsys_mask
;
1749 removed_mask
= root
->subsys_mask
& ~opts
.subsys_mask
;
1751 /* Don't allow flags or name to change at remount */
1752 if ((opts
.flags
^ root
->flags
) ||
1753 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1754 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1755 opts
.flags
, opts
.name
?: "", root
->flags
, root
->name
);
1760 /* remounting is not allowed for populated hierarchies */
1761 if (!list_empty(&root
->cgrp
.self
.children
)) {
1766 ret
= rebind_subsystems(root
, added_mask
);
1770 rebind_subsystems(&cgrp_dfl_root
, removed_mask
);
1772 if (opts
.release_agent
) {
1773 spin_lock(&release_agent_path_lock
);
1774 strcpy(root
->release_agent_path
, opts
.release_agent
);
1775 spin_unlock(&release_agent_path_lock
);
1778 kfree(opts
.release_agent
);
1780 mutex_unlock(&cgroup_mutex
);
1785 * To reduce the fork() overhead for systems that are not actually using
1786 * their cgroups capability, we don't maintain the lists running through
1787 * each css_set to its tasks until we see the list actually used - in other
1788 * words after the first mount.
1790 static bool use_task_css_set_links __read_mostly
;
1792 static void cgroup_enable_task_cg_lists(void)
1794 struct task_struct
*p
, *g
;
1796 down_write(&css_set_rwsem
);
1798 if (use_task_css_set_links
)
1801 use_task_css_set_links
= true;
1804 * We need tasklist_lock because RCU is not safe against
1805 * while_each_thread(). Besides, a forking task that has passed
1806 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1807 * is not guaranteed to have its child immediately visible in the
1808 * tasklist if we walk through it with RCU.
1810 read_lock(&tasklist_lock
);
1811 do_each_thread(g
, p
) {
1812 WARN_ON_ONCE(!list_empty(&p
->cg_list
) ||
1813 task_css_set(p
) != &init_css_set
);
1816 * We should check if the process is exiting, otherwise
1817 * it will race with cgroup_exit() in that the list
1818 * entry won't be deleted though the process has exited.
1819 * Do it while holding siglock so that we don't end up
1820 * racing against cgroup_exit().
1822 spin_lock_irq(&p
->sighand
->siglock
);
1823 if (!(p
->flags
& PF_EXITING
)) {
1824 struct css_set
*cset
= task_css_set(p
);
1826 if (!css_set_populated(cset
))
1827 css_set_update_populated(cset
, true);
1828 list_add_tail(&p
->cg_list
, &cset
->tasks
);
1831 spin_unlock_irq(&p
->sighand
->siglock
);
1832 } while_each_thread(g
, p
);
1833 read_unlock(&tasklist_lock
);
1835 up_write(&css_set_rwsem
);
1838 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1840 struct cgroup_subsys
*ss
;
1843 INIT_LIST_HEAD(&cgrp
->self
.sibling
);
1844 INIT_LIST_HEAD(&cgrp
->self
.children
);
1845 INIT_LIST_HEAD(&cgrp
->self
.files
);
1846 INIT_LIST_HEAD(&cgrp
->cset_links
);
1847 INIT_LIST_HEAD(&cgrp
->pidlists
);
1848 mutex_init(&cgrp
->pidlist_mutex
);
1849 cgrp
->self
.cgroup
= cgrp
;
1850 cgrp
->self
.flags
|= CSS_ONLINE
;
1852 for_each_subsys(ss
, ssid
)
1853 INIT_LIST_HEAD(&cgrp
->e_csets
[ssid
]);
1855 init_waitqueue_head(&cgrp
->offline_waitq
);
1856 INIT_WORK(&cgrp
->release_agent_work
, cgroup_release_agent
);
1859 static void init_cgroup_root(struct cgroup_root
*root
,
1860 struct cgroup_sb_opts
*opts
)
1862 struct cgroup
*cgrp
= &root
->cgrp
;
1864 INIT_LIST_HEAD(&root
->root_list
);
1865 atomic_set(&root
->nr_cgrps
, 1);
1867 init_cgroup_housekeeping(cgrp
);
1868 idr_init(&root
->cgroup_idr
);
1870 root
->flags
= opts
->flags
;
1871 if (opts
->release_agent
)
1872 strcpy(root
->release_agent_path
, opts
->release_agent
);
1874 strcpy(root
->name
, opts
->name
);
1875 if (opts
->cpuset_clone_children
)
1876 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &root
->cgrp
.flags
);
1879 static int cgroup_setup_root(struct cgroup_root
*root
, unsigned long ss_mask
)
1881 LIST_HEAD(tmp_links
);
1882 struct cgroup
*root_cgrp
= &root
->cgrp
;
1883 struct css_set
*cset
;
1886 lockdep_assert_held(&cgroup_mutex
);
1888 ret
= cgroup_idr_alloc(&root
->cgroup_idr
, root_cgrp
, 1, 2, GFP_KERNEL
);
1891 root_cgrp
->id
= ret
;
1893 ret
= percpu_ref_init(&root_cgrp
->self
.refcnt
, css_release
, 0,
1899 * We're accessing css_set_count without locking css_set_rwsem here,
1900 * but that's OK - it can only be increased by someone holding
1901 * cgroup_lock, and that's us. The worst that can happen is that we
1902 * have some link structures left over
1904 ret
= allocate_cgrp_cset_links(css_set_count
, &tmp_links
);
1908 ret
= cgroup_init_root_id(root
);
1912 root
->kf_root
= kernfs_create_root(&cgroup_kf_syscall_ops
,
1913 KERNFS_ROOT_CREATE_DEACTIVATED
,
1915 if (IS_ERR(root
->kf_root
)) {
1916 ret
= PTR_ERR(root
->kf_root
);
1919 root_cgrp
->kn
= root
->kf_root
->kn
;
1921 ret
= css_populate_dir(&root_cgrp
->self
, NULL
);
1925 ret
= rebind_subsystems(root
, ss_mask
);
1930 * There must be no failure case after here, since rebinding takes
1931 * care of subsystems' refcounts, which are explicitly dropped in
1932 * the failure exit path.
1934 list_add(&root
->root_list
, &cgroup_roots
);
1935 cgroup_root_count
++;
1938 * Link the root cgroup in this hierarchy into all the css_set
1941 down_write(&css_set_rwsem
);
1942 hash_for_each(css_set_table
, i
, cset
, hlist
) {
1943 link_css_set(&tmp_links
, cset
, root_cgrp
);
1944 if (css_set_populated(cset
))
1945 cgroup_update_populated(root_cgrp
, true);
1947 up_write(&css_set_rwsem
);
1949 BUG_ON(!list_empty(&root_cgrp
->self
.children
));
1950 BUG_ON(atomic_read(&root
->nr_cgrps
) != 1);
1952 kernfs_activate(root_cgrp
->kn
);
1957 kernfs_destroy_root(root
->kf_root
);
1958 root
->kf_root
= NULL
;
1960 cgroup_exit_root_id(root
);
1962 percpu_ref_exit(&root_cgrp
->self
.refcnt
);
1964 free_cgrp_cset_links(&tmp_links
);
1968 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1969 int flags
, const char *unused_dev_name
,
1972 struct super_block
*pinned_sb
= NULL
;
1973 struct cgroup_subsys
*ss
;
1974 struct cgroup_root
*root
;
1975 struct cgroup_sb_opts opts
;
1976 struct dentry
*dentry
;
1982 * The first time anyone tries to mount a cgroup, enable the list
1983 * linking each css_set to its tasks and fix up all existing tasks.
1985 if (!use_task_css_set_links
)
1986 cgroup_enable_task_cg_lists();
1988 mutex_lock(&cgroup_mutex
);
1990 /* First find the desired set of subsystems */
1991 ret
= parse_cgroupfs_options(data
, &opts
);
1995 /* look for a matching existing root */
1996 if (opts
.flags
& CGRP_ROOT_SANE_BEHAVIOR
) {
1997 cgrp_dfl_root_visible
= true;
1998 root
= &cgrp_dfl_root
;
1999 cgroup_get(&root
->cgrp
);
2005 * Destruction of cgroup root is asynchronous, so subsystems may
2006 * still be dying after the previous unmount. Let's drain the
2007 * dying subsystems. We just need to ensure that the ones
2008 * unmounted previously finish dying and don't care about new ones
2009 * starting. Testing ref liveliness is good enough.
2011 for_each_subsys(ss
, i
) {
2012 if (!(opts
.subsys_mask
& (1 << i
)) ||
2013 ss
->root
== &cgrp_dfl_root
)
2016 if (!percpu_ref_tryget_live(&ss
->root
->cgrp
.self
.refcnt
)) {
2017 mutex_unlock(&cgroup_mutex
);
2019 ret
= restart_syscall();
2022 cgroup_put(&ss
->root
->cgrp
);
2025 for_each_root(root
) {
2026 bool name_match
= false;
2028 if (root
== &cgrp_dfl_root
)
2032 * If we asked for a name then it must match. Also, if
2033 * name matches but sybsys_mask doesn't, we should fail.
2034 * Remember whether name matched.
2037 if (strcmp(opts
.name
, root
->name
))
2043 * If we asked for subsystems (or explicitly for no
2044 * subsystems) then they must match.
2046 if ((opts
.subsys_mask
|| opts
.none
) &&
2047 (opts
.subsys_mask
!= root
->subsys_mask
)) {
2054 if (root
->flags
^ opts
.flags
)
2055 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
2058 * We want to reuse @root whose lifetime is governed by its
2059 * ->cgrp. Let's check whether @root is alive and keep it
2060 * that way. As cgroup_kill_sb() can happen anytime, we
2061 * want to block it by pinning the sb so that @root doesn't
2062 * get killed before mount is complete.
2064 * With the sb pinned, tryget_live can reliably indicate
2065 * whether @root can be reused. If it's being killed,
2066 * drain it. We can use wait_queue for the wait but this
2067 * path is super cold. Let's just sleep a bit and retry.
2069 pinned_sb
= kernfs_pin_sb(root
->kf_root
, NULL
);
2070 if (IS_ERR(pinned_sb
) ||
2071 !percpu_ref_tryget_live(&root
->cgrp
.self
.refcnt
)) {
2072 mutex_unlock(&cgroup_mutex
);
2073 if (!IS_ERR_OR_NULL(pinned_sb
))
2074 deactivate_super(pinned_sb
);
2076 ret
= restart_syscall();
2085 * No such thing, create a new one. name= matching without subsys
2086 * specification is allowed for already existing hierarchies but we
2087 * can't create new one without subsys specification.
2089 if (!opts
.subsys_mask
&& !opts
.none
) {
2094 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
2100 init_cgroup_root(root
, &opts
);
2102 ret
= cgroup_setup_root(root
, opts
.subsys_mask
);
2104 cgroup_free_root(root
);
2107 mutex_unlock(&cgroup_mutex
);
2109 kfree(opts
.release_agent
);
2113 return ERR_PTR(ret
);
2115 dentry
= kernfs_mount(fs_type
, flags
, root
->kf_root
,
2116 CGROUP_SUPER_MAGIC
, &new_sb
);
2117 if (IS_ERR(dentry
) || !new_sb
)
2118 cgroup_put(&root
->cgrp
);
2121 * If @pinned_sb, we're reusing an existing root and holding an
2122 * extra ref on its sb. Mount is complete. Put the extra ref.
2126 deactivate_super(pinned_sb
);
2132 static void cgroup_kill_sb(struct super_block
*sb
)
2134 struct kernfs_root
*kf_root
= kernfs_root_from_sb(sb
);
2135 struct cgroup_root
*root
= cgroup_root_from_kf(kf_root
);
2138 * If @root doesn't have any mounts or children, start killing it.
2139 * This prevents new mounts by disabling percpu_ref_tryget_live().
2140 * cgroup_mount() may wait for @root's release.
2142 * And don't kill the default root.
2144 if (!list_empty(&root
->cgrp
.self
.children
) ||
2145 root
== &cgrp_dfl_root
)
2146 cgroup_put(&root
->cgrp
);
2148 percpu_ref_kill(&root
->cgrp
.self
.refcnt
);
2153 static struct file_system_type cgroup_fs_type
= {
2155 .mount
= cgroup_mount
,
2156 .kill_sb
= cgroup_kill_sb
,
2160 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2161 * @task: target task
2162 * @buf: the buffer to write the path into
2163 * @buflen: the length of the buffer
2165 * Determine @task's cgroup on the first (the one with the lowest non-zero
2166 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2167 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2168 * cgroup controller callbacks.
2170 * Return value is the same as kernfs_path().
2172 char *task_cgroup_path(struct task_struct
*task
, char *buf
, size_t buflen
)
2174 struct cgroup_root
*root
;
2175 struct cgroup
*cgrp
;
2176 int hierarchy_id
= 1;
2179 mutex_lock(&cgroup_mutex
);
2180 down_read(&css_set_rwsem
);
2182 root
= idr_get_next(&cgroup_hierarchy_idr
, &hierarchy_id
);
2185 cgrp
= task_cgroup_from_root(task
, root
);
2186 path
= cgroup_path(cgrp
, buf
, buflen
);
2188 /* if no hierarchy exists, everyone is in "/" */
2189 if (strlcpy(buf
, "/", buflen
) < buflen
)
2193 up_read(&css_set_rwsem
);
2194 mutex_unlock(&cgroup_mutex
);
2197 EXPORT_SYMBOL_GPL(task_cgroup_path
);
2199 /* used to track tasks and other necessary states during migration */
2200 struct cgroup_taskset
{
2201 /* the src and dst cset list running through cset->mg_node */
2202 struct list_head src_csets
;
2203 struct list_head dst_csets
;
2206 * Fields for cgroup_taskset_*() iteration.
2208 * Before migration is committed, the target migration tasks are on
2209 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
2210 * the csets on ->dst_csets. ->csets point to either ->src_csets
2211 * or ->dst_csets depending on whether migration is committed.
2213 * ->cur_csets and ->cur_task point to the current task position
2216 struct list_head
*csets
;
2217 struct css_set
*cur_cset
;
2218 struct task_struct
*cur_task
;
2221 #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \
2222 .src_csets = LIST_HEAD_INIT(tset.src_csets), \
2223 .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \
2224 .csets = &tset.src_csets, \
2228 * cgroup_taskset_add - try to add a migration target task to a taskset
2229 * @task: target task
2230 * @tset: target taskset
2232 * Add @task, which is a migration target, to @tset. This function becomes
2233 * noop if @task doesn't need to be migrated. @task's css_set should have
2234 * been added as a migration source and @task->cg_list will be moved from
2235 * the css_set's tasks list to mg_tasks one.
2237 static void cgroup_taskset_add(struct task_struct
*task
,
2238 struct cgroup_taskset
*tset
)
2240 struct css_set
*cset
;
2242 lockdep_assert_held(&css_set_rwsem
);
2244 /* @task either already exited or can't exit until the end */
2245 if (task
->flags
& PF_EXITING
)
2248 /* leave @task alone if post_fork() hasn't linked it yet */
2249 if (list_empty(&task
->cg_list
))
2252 cset
= task_css_set(task
);
2253 if (!cset
->mg_src_cgrp
)
2256 list_move_tail(&task
->cg_list
, &cset
->mg_tasks
);
2257 if (list_empty(&cset
->mg_node
))
2258 list_add_tail(&cset
->mg_node
, &tset
->src_csets
);
2259 if (list_empty(&cset
->mg_dst_cset
->mg_node
))
2260 list_move_tail(&cset
->mg_dst_cset
->mg_node
,
2265 * cgroup_taskset_first - reset taskset and return the first task
2266 * @tset: taskset of interest
2268 * @tset iteration is initialized and the first task is returned.
2270 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
2272 tset
->cur_cset
= list_first_entry(tset
->csets
, struct css_set
, mg_node
);
2273 tset
->cur_task
= NULL
;
2275 return cgroup_taskset_next(tset
);
2279 * cgroup_taskset_next - iterate to the next task in taskset
2280 * @tset: taskset of interest
2282 * Return the next task in @tset. Iteration must have been initialized
2283 * with cgroup_taskset_first().
2285 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
2287 struct css_set
*cset
= tset
->cur_cset
;
2288 struct task_struct
*task
= tset
->cur_task
;
2290 while (&cset
->mg_node
!= tset
->csets
) {
2292 task
= list_first_entry(&cset
->mg_tasks
,
2293 struct task_struct
, cg_list
);
2295 task
= list_next_entry(task
, cg_list
);
2297 if (&task
->cg_list
!= &cset
->mg_tasks
) {
2298 tset
->cur_cset
= cset
;
2299 tset
->cur_task
= task
;
2303 cset
= list_next_entry(cset
, mg_node
);
2311 * cgroup_taskset_migrate - migrate a taskset to a cgroup
2312 * @tset: taget taskset
2313 * @dst_cgrp: destination cgroup
2315 * Migrate tasks in @tset to @dst_cgrp. This function fails iff one of the
2316 * ->can_attach callbacks fails and guarantees that either all or none of
2317 * the tasks in @tset are migrated. @tset is consumed regardless of
2320 static int cgroup_taskset_migrate(struct cgroup_taskset
*tset
,
2321 struct cgroup
*dst_cgrp
)
2323 struct cgroup_subsys_state
*css
, *failed_css
= NULL
;
2324 struct task_struct
*task
, *tmp_task
;
2325 struct css_set
*cset
, *tmp_cset
;
2328 /* methods shouldn't be called if no task is actually migrating */
2329 if (list_empty(&tset
->src_csets
))
2332 /* check that we can legitimately attach to the cgroup */
2333 for_each_e_css(css
, i
, dst_cgrp
) {
2334 if (css
->ss
->can_attach
) {
2335 ret
= css
->ss
->can_attach(css
, tset
);
2338 goto out_cancel_attach
;
2344 * Now that we're guaranteed success, proceed to move all tasks to
2345 * the new cgroup. There are no failure cases after here, so this
2346 * is the commit point.
2348 down_write(&css_set_rwsem
);
2349 list_for_each_entry(cset
, &tset
->src_csets
, mg_node
) {
2350 list_for_each_entry_safe(task
, tmp_task
, &cset
->mg_tasks
, cg_list
) {
2351 struct css_set
*from_cset
= task_css_set(task
);
2352 struct css_set
*to_cset
= cset
->mg_dst_cset
;
2354 get_css_set(to_cset
);
2355 css_set_move_task(task
, from_cset
, to_cset
, true);
2356 put_css_set_locked(from_cset
);
2359 up_write(&css_set_rwsem
);
2362 * Migration is committed, all target tasks are now on dst_csets.
2363 * Nothing is sensitive to fork() after this point. Notify
2364 * controllers that migration is complete.
2366 tset
->csets
= &tset
->dst_csets
;
2368 for_each_e_css(css
, i
, dst_cgrp
)
2369 if (css
->ss
->attach
)
2370 css
->ss
->attach(css
, tset
);
2373 goto out_release_tset
;
2376 for_each_e_css(css
, i
, dst_cgrp
) {
2377 if (css
== failed_css
)
2379 if (css
->ss
->cancel_attach
)
2380 css
->ss
->cancel_attach(css
, tset
);
2383 down_write(&css_set_rwsem
);
2384 list_splice_init(&tset
->dst_csets
, &tset
->src_csets
);
2385 list_for_each_entry_safe(cset
, tmp_cset
, &tset
->src_csets
, mg_node
) {
2386 list_splice_tail_init(&cset
->mg_tasks
, &cset
->tasks
);
2387 list_del_init(&cset
->mg_node
);
2389 up_write(&css_set_rwsem
);
2394 * cgroup_migrate_finish - cleanup after attach
2395 * @preloaded_csets: list of preloaded css_sets
2397 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2398 * those functions for details.
2400 static void cgroup_migrate_finish(struct list_head
*preloaded_csets
)
2402 struct css_set
*cset
, *tmp_cset
;
2404 lockdep_assert_held(&cgroup_mutex
);
2406 down_write(&css_set_rwsem
);
2407 list_for_each_entry_safe(cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2408 cset
->mg_src_cgrp
= NULL
;
2409 cset
->mg_dst_cset
= NULL
;
2410 list_del_init(&cset
->mg_preload_node
);
2411 put_css_set_locked(cset
);
2413 up_write(&css_set_rwsem
);
2417 * cgroup_migrate_add_src - add a migration source css_set
2418 * @src_cset: the source css_set to add
2419 * @dst_cgrp: the destination cgroup
2420 * @preloaded_csets: list of preloaded css_sets
2422 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2423 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2424 * up by cgroup_migrate_finish().
2426 * This function may be called without holding cgroup_threadgroup_rwsem
2427 * even if the target is a process. Threads may be created and destroyed
2428 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2429 * into play and the preloaded css_sets are guaranteed to cover all
2432 static void cgroup_migrate_add_src(struct css_set
*src_cset
,
2433 struct cgroup
*dst_cgrp
,
2434 struct list_head
*preloaded_csets
)
2436 struct cgroup
*src_cgrp
;
2438 lockdep_assert_held(&cgroup_mutex
);
2439 lockdep_assert_held(&css_set_rwsem
);
2441 src_cgrp
= cset_cgroup_from_root(src_cset
, dst_cgrp
->root
);
2443 if (!list_empty(&src_cset
->mg_preload_node
))
2446 WARN_ON(src_cset
->mg_src_cgrp
);
2447 WARN_ON(!list_empty(&src_cset
->mg_tasks
));
2448 WARN_ON(!list_empty(&src_cset
->mg_node
));
2450 src_cset
->mg_src_cgrp
= src_cgrp
;
2451 get_css_set(src_cset
);
2452 list_add(&src_cset
->mg_preload_node
, preloaded_csets
);
2456 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2457 * @dst_cgrp: the destination cgroup (may be %NULL)
2458 * @preloaded_csets: list of preloaded source css_sets
2460 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
2461 * have been preloaded to @preloaded_csets. This function looks up and
2462 * pins all destination css_sets, links each to its source, and append them
2463 * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each
2464 * source css_set is assumed to be its cgroup on the default hierarchy.
2466 * This function must be called after cgroup_migrate_add_src() has been
2467 * called on each migration source css_set. After migration is performed
2468 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2471 static int cgroup_migrate_prepare_dst(struct cgroup
*dst_cgrp
,
2472 struct list_head
*preloaded_csets
)
2475 struct css_set
*src_cset
, *tmp_cset
;
2477 lockdep_assert_held(&cgroup_mutex
);
2480 * Except for the root, child_subsys_mask must be zero for a cgroup
2481 * with tasks so that child cgroups don't compete against tasks.
2483 if (dst_cgrp
&& cgroup_on_dfl(dst_cgrp
) && cgroup_parent(dst_cgrp
) &&
2484 dst_cgrp
->child_subsys_mask
)
2487 /* look up the dst cset for each src cset and link it to src */
2488 list_for_each_entry_safe(src_cset
, tmp_cset
, preloaded_csets
, mg_preload_node
) {
2489 struct css_set
*dst_cset
;
2491 dst_cset
= find_css_set(src_cset
,
2492 dst_cgrp
?: src_cset
->dfl_cgrp
);
2496 WARN_ON_ONCE(src_cset
->mg_dst_cset
|| dst_cset
->mg_dst_cset
);
2499 * If src cset equals dst, it's noop. Drop the src.
2500 * cgroup_migrate() will skip the cset too. Note that we
2501 * can't handle src == dst as some nodes are used by both.
2503 if (src_cset
== dst_cset
) {
2504 src_cset
->mg_src_cgrp
= NULL
;
2505 list_del_init(&src_cset
->mg_preload_node
);
2506 put_css_set(src_cset
);
2507 put_css_set(dst_cset
);
2511 src_cset
->mg_dst_cset
= dst_cset
;
2513 if (list_empty(&dst_cset
->mg_preload_node
))
2514 list_add(&dst_cset
->mg_preload_node
, &csets
);
2516 put_css_set(dst_cset
);
2519 list_splice_tail(&csets
, preloaded_csets
);
2522 cgroup_migrate_finish(&csets
);
2527 * cgroup_migrate - migrate a process or task to a cgroup
2528 * @leader: the leader of the process or the task to migrate
2529 * @threadgroup: whether @leader points to the whole process or a single task
2530 * @cgrp: the destination cgroup
2532 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
2533 * process, the caller must be holding cgroup_threadgroup_rwsem. The
2534 * caller is also responsible for invoking cgroup_migrate_add_src() and
2535 * cgroup_migrate_prepare_dst() on the targets before invoking this
2536 * function and following up with cgroup_migrate_finish().
2538 * As long as a controller's ->can_attach() doesn't fail, this function is
2539 * guaranteed to succeed. This means that, excluding ->can_attach()
2540 * failure, when migrating multiple targets, the success or failure can be
2541 * decided for all targets by invoking group_migrate_prepare_dst() before
2542 * actually starting migrating.
2544 static int cgroup_migrate(struct task_struct
*leader
, bool threadgroup
,
2545 struct cgroup
*cgrp
)
2547 struct cgroup_taskset tset
= CGROUP_TASKSET_INIT(tset
);
2548 struct task_struct
*task
;
2551 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2552 * already PF_EXITING could be freed from underneath us unless we
2553 * take an rcu_read_lock.
2555 down_write(&css_set_rwsem
);
2559 cgroup_taskset_add(task
, &tset
);
2562 } while_each_thread(leader
, task
);
2564 up_write(&css_set_rwsem
);
2566 return cgroup_taskset_migrate(&tset
, cgrp
);
2570 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2571 * @dst_cgrp: the cgroup to attach to
2572 * @leader: the task or the leader of the threadgroup to be attached
2573 * @threadgroup: attach the whole threadgroup?
2575 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2577 static int cgroup_attach_task(struct cgroup
*dst_cgrp
,
2578 struct task_struct
*leader
, bool threadgroup
)
2580 LIST_HEAD(preloaded_csets
);
2581 struct task_struct
*task
;
2584 /* look up all src csets */
2585 down_read(&css_set_rwsem
);
2589 cgroup_migrate_add_src(task_css_set(task
), dst_cgrp
,
2593 } while_each_thread(leader
, task
);
2595 up_read(&css_set_rwsem
);
2597 /* prepare dst csets and commit */
2598 ret
= cgroup_migrate_prepare_dst(dst_cgrp
, &preloaded_csets
);
2600 ret
= cgroup_migrate(leader
, threadgroup
, dst_cgrp
);
2602 cgroup_migrate_finish(&preloaded_csets
);
2606 static int cgroup_procs_write_permission(struct task_struct
*task
,
2607 struct cgroup
*dst_cgrp
,
2608 struct kernfs_open_file
*of
)
2610 const struct cred
*cred
= current_cred();
2611 const struct cred
*tcred
= get_task_cred(task
);
2615 * even if we're attaching all tasks in the thread group, we only
2616 * need to check permissions on one of them.
2618 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2619 !uid_eq(cred
->euid
, tcred
->uid
) &&
2620 !uid_eq(cred
->euid
, tcred
->suid
))
2623 if (!ret
&& cgroup_on_dfl(dst_cgrp
)) {
2624 struct super_block
*sb
= of
->file
->f_path
.dentry
->d_sb
;
2625 struct cgroup
*cgrp
;
2626 struct inode
*inode
;
2628 down_read(&css_set_rwsem
);
2629 cgrp
= task_cgroup_from_root(task
, &cgrp_dfl_root
);
2630 up_read(&css_set_rwsem
);
2632 while (!cgroup_is_descendant(dst_cgrp
, cgrp
))
2633 cgrp
= cgroup_parent(cgrp
);
2636 inode
= kernfs_get_inode(sb
, cgrp
->procs_file
.kn
);
2638 ret
= inode_permission(inode
, MAY_WRITE
);
2648 * Find the task_struct of the task to attach by vpid and pass it along to the
2649 * function to attach either it or all tasks in its threadgroup. Will lock
2650 * cgroup_mutex and threadgroup.
2652 static ssize_t
__cgroup_procs_write(struct kernfs_open_file
*of
, char *buf
,
2653 size_t nbytes
, loff_t off
, bool threadgroup
)
2655 struct task_struct
*tsk
;
2656 struct cgroup
*cgrp
;
2660 if (kstrtoint(strstrip(buf
), 0, &pid
) || pid
< 0)
2663 cgrp
= cgroup_kn_lock_live(of
->kn
);
2667 percpu_down_write(&cgroup_threadgroup_rwsem
);
2670 tsk
= find_task_by_vpid(pid
);
2673 goto out_unlock_rcu
;
2680 tsk
= tsk
->group_leader
;
2683 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2684 * trapped in a cpuset, or RT worker may be born in a cgroup
2685 * with no rt_runtime allocated. Just say no.
2687 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_NO_SETAFFINITY
)) {
2689 goto out_unlock_rcu
;
2692 get_task_struct(tsk
);
2695 ret
= cgroup_procs_write_permission(tsk
, cgrp
, of
);
2697 ret
= cgroup_attach_task(cgrp
, tsk
, threadgroup
);
2699 put_task_struct(tsk
);
2700 goto out_unlock_threadgroup
;
2704 out_unlock_threadgroup
:
2705 percpu_up_write(&cgroup_threadgroup_rwsem
);
2706 cgroup_kn_unlock(of
->kn
);
2707 return ret
?: nbytes
;
2711 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2712 * @from: attach to all cgroups of a given task
2713 * @tsk: the task to be attached
2715 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2717 struct cgroup_root
*root
;
2720 mutex_lock(&cgroup_mutex
);
2721 for_each_root(root
) {
2722 struct cgroup
*from_cgrp
;
2724 if (root
== &cgrp_dfl_root
)
2727 down_read(&css_set_rwsem
);
2728 from_cgrp
= task_cgroup_from_root(from
, root
);
2729 up_read(&css_set_rwsem
);
2731 retval
= cgroup_attach_task(from_cgrp
, tsk
, false);
2735 mutex_unlock(&cgroup_mutex
);
2739 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2741 static ssize_t
cgroup_tasks_write(struct kernfs_open_file
*of
,
2742 char *buf
, size_t nbytes
, loff_t off
)
2744 return __cgroup_procs_write(of
, buf
, nbytes
, off
, false);
2747 static ssize_t
cgroup_procs_write(struct kernfs_open_file
*of
,
2748 char *buf
, size_t nbytes
, loff_t off
)
2750 return __cgroup_procs_write(of
, buf
, nbytes
, off
, true);
2753 static ssize_t
cgroup_release_agent_write(struct kernfs_open_file
*of
,
2754 char *buf
, size_t nbytes
, loff_t off
)
2756 struct cgroup
*cgrp
;
2758 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2760 cgrp
= cgroup_kn_lock_live(of
->kn
);
2763 spin_lock(&release_agent_path_lock
);
2764 strlcpy(cgrp
->root
->release_agent_path
, strstrip(buf
),
2765 sizeof(cgrp
->root
->release_agent_path
));
2766 spin_unlock(&release_agent_path_lock
);
2767 cgroup_kn_unlock(of
->kn
);
2771 static int cgroup_release_agent_show(struct seq_file
*seq
, void *v
)
2773 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2775 spin_lock(&release_agent_path_lock
);
2776 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2777 spin_unlock(&release_agent_path_lock
);
2778 seq_putc(seq
, '\n');
2782 static int cgroup_sane_behavior_show(struct seq_file
*seq
, void *v
)
2784 seq_puts(seq
, "0\n");
2788 static void cgroup_print_ss_mask(struct seq_file
*seq
, unsigned long ss_mask
)
2790 struct cgroup_subsys
*ss
;
2791 bool printed
= false;
2794 for_each_subsys_which(ss
, ssid
, &ss_mask
) {
2797 seq_printf(seq
, "%s", ss
->name
);
2801 seq_putc(seq
, '\n');
2804 /* show controllers which are currently attached to the default hierarchy */
2805 static int cgroup_root_controllers_show(struct seq_file
*seq
, void *v
)
2807 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2809 cgroup_print_ss_mask(seq
, cgrp
->root
->subsys_mask
&
2810 ~cgrp_dfl_root_inhibit_ss_mask
);
2814 /* show controllers which are enabled from the parent */
2815 static int cgroup_controllers_show(struct seq_file
*seq
, void *v
)
2817 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2819 cgroup_print_ss_mask(seq
, cgroup_parent(cgrp
)->subtree_control
);
2823 /* show controllers which are enabled for a given cgroup's children */
2824 static int cgroup_subtree_control_show(struct seq_file
*seq
, void *v
)
2826 struct cgroup
*cgrp
= seq_css(seq
)->cgroup
;
2828 cgroup_print_ss_mask(seq
, cgrp
->subtree_control
);
2833 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2834 * @cgrp: root of the subtree to update csses for
2836 * @cgrp's child_subsys_mask has changed and its subtree's (self excluded)
2837 * css associations need to be updated accordingly. This function looks up
2838 * all css_sets which are attached to the subtree, creates the matching
2839 * updated css_sets and migrates the tasks to the new ones.
2841 static int cgroup_update_dfl_csses(struct cgroup
*cgrp
)
2843 LIST_HEAD(preloaded_csets
);
2844 struct cgroup_taskset tset
= CGROUP_TASKSET_INIT(tset
);
2845 struct cgroup_subsys_state
*css
;
2846 struct css_set
*src_cset
;
2849 lockdep_assert_held(&cgroup_mutex
);
2851 percpu_down_write(&cgroup_threadgroup_rwsem
);
2853 /* look up all csses currently attached to @cgrp's subtree */
2854 down_read(&css_set_rwsem
);
2855 css_for_each_descendant_pre(css
, cgroup_css(cgrp
, NULL
)) {
2856 struct cgrp_cset_link
*link
;
2858 /* self is not affected by child_subsys_mask change */
2859 if (css
->cgroup
== cgrp
)
2862 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
)
2863 cgroup_migrate_add_src(link
->cset
, cgrp
,
2866 up_read(&css_set_rwsem
);
2868 /* NULL dst indicates self on default hierarchy */
2869 ret
= cgroup_migrate_prepare_dst(NULL
, &preloaded_csets
);
2873 down_write(&css_set_rwsem
);
2874 list_for_each_entry(src_cset
, &preloaded_csets
, mg_preload_node
) {
2875 struct task_struct
*task
, *ntask
;
2877 /* src_csets precede dst_csets, break on the first dst_cset */
2878 if (!src_cset
->mg_src_cgrp
)
2881 /* all tasks in src_csets need to be migrated */
2882 list_for_each_entry_safe(task
, ntask
, &src_cset
->tasks
, cg_list
)
2883 cgroup_taskset_add(task
, &tset
);
2885 up_write(&css_set_rwsem
);
2887 ret
= cgroup_taskset_migrate(&tset
, cgrp
);
2889 cgroup_migrate_finish(&preloaded_csets
);
2890 percpu_up_write(&cgroup_threadgroup_rwsem
);
2894 /* change the enabled child controllers for a cgroup in the default hierarchy */
2895 static ssize_t
cgroup_subtree_control_write(struct kernfs_open_file
*of
,
2896 char *buf
, size_t nbytes
,
2899 unsigned long enable
= 0, disable
= 0;
2900 unsigned long css_enable
, css_disable
, old_sc
, new_sc
, old_ss
, new_ss
;
2901 struct cgroup
*cgrp
, *child
;
2902 struct cgroup_subsys
*ss
;
2907 * Parse input - space separated list of subsystem names prefixed
2908 * with either + or -.
2910 buf
= strstrip(buf
);
2911 while ((tok
= strsep(&buf
, " "))) {
2912 unsigned long tmp_ss_mask
= ~cgrp_dfl_root_inhibit_ss_mask
;
2916 for_each_subsys_which(ss
, ssid
, &tmp_ss_mask
) {
2917 if (!cgroup_ssid_enabled(ssid
) ||
2918 strcmp(tok
+ 1, ss
->name
))
2922 enable
|= 1 << ssid
;
2923 disable
&= ~(1 << ssid
);
2924 } else if (*tok
== '-') {
2925 disable
|= 1 << ssid
;
2926 enable
&= ~(1 << ssid
);
2932 if (ssid
== CGROUP_SUBSYS_COUNT
)
2936 cgrp
= cgroup_kn_lock_live(of
->kn
);
2940 for_each_subsys(ss
, ssid
) {
2941 if (enable
& (1 << ssid
)) {
2942 if (cgrp
->subtree_control
& (1 << ssid
)) {
2943 enable
&= ~(1 << ssid
);
2947 /* unavailable or not enabled on the parent? */
2948 if (!(cgrp_dfl_root
.subsys_mask
& (1 << ssid
)) ||
2949 (cgroup_parent(cgrp
) &&
2950 !(cgroup_parent(cgrp
)->subtree_control
& (1 << ssid
)))) {
2954 } else if (disable
& (1 << ssid
)) {
2955 if (!(cgrp
->subtree_control
& (1 << ssid
))) {
2956 disable
&= ~(1 << ssid
);
2960 /* a child has it enabled? */
2961 cgroup_for_each_live_child(child
, cgrp
) {
2962 if (child
->subtree_control
& (1 << ssid
)) {
2970 if (!enable
&& !disable
) {
2976 * Except for the root, subtree_control must be zero for a cgroup
2977 * with tasks so that child cgroups don't compete against tasks.
2979 if (enable
&& cgroup_parent(cgrp
) && !list_empty(&cgrp
->cset_links
)) {
2985 * Update subsys masks and calculate what needs to be done. More
2986 * subsystems than specified may need to be enabled or disabled
2987 * depending on subsystem dependencies.
2989 old_sc
= cgrp
->subtree_control
;
2990 old_ss
= cgrp
->child_subsys_mask
;
2991 new_sc
= (old_sc
| enable
) & ~disable
;
2992 new_ss
= cgroup_calc_child_subsys_mask(cgrp
, new_sc
);
2994 css_enable
= ~old_ss
& new_ss
;
2995 css_disable
= old_ss
& ~new_ss
;
2996 enable
|= css_enable
;
2997 disable
|= css_disable
;
3000 * Because css offlining is asynchronous, userland might try to
3001 * re-enable the same controller while the previous instance is
3002 * still around. In such cases, wait till it's gone using
3005 for_each_subsys_which(ss
, ssid
, &css_enable
) {
3006 cgroup_for_each_live_child(child
, cgrp
) {
3009 if (!cgroup_css(child
, ss
))
3013 prepare_to_wait(&child
->offline_waitq
, &wait
,
3014 TASK_UNINTERRUPTIBLE
);
3015 cgroup_kn_unlock(of
->kn
);
3017 finish_wait(&child
->offline_waitq
, &wait
);
3020 return restart_syscall();
3024 cgrp
->subtree_control
= new_sc
;
3025 cgrp
->child_subsys_mask
= new_ss
;
3028 * Create new csses or make the existing ones visible. A css is
3029 * created invisible if it's being implicitly enabled through
3030 * dependency. An invisible css is made visible when the userland
3031 * explicitly enables it.
3033 for_each_subsys(ss
, ssid
) {
3034 if (!(enable
& (1 << ssid
)))
3037 cgroup_for_each_live_child(child
, cgrp
) {
3038 if (css_enable
& (1 << ssid
))
3039 ret
= create_css(child
, ss
,
3040 cgrp
->subtree_control
& (1 << ssid
));
3042 ret
= css_populate_dir(cgroup_css(child
, ss
),
3050 * At this point, cgroup_e_css() results reflect the new csses
3051 * making the following cgroup_update_dfl_csses() properly update
3052 * css associations of all tasks in the subtree.
3054 ret
= cgroup_update_dfl_csses(cgrp
);
3059 * All tasks are migrated out of disabled csses. Kill or hide
3060 * them. A css is hidden when the userland requests it to be
3061 * disabled while other subsystems are still depending on it. The
3062 * css must not actively control resources and be in the vanilla
3063 * state if it's made visible again later. Controllers which may
3064 * be depended upon should provide ->css_reset() for this purpose.
3066 for_each_subsys(ss
, ssid
) {
3067 if (!(disable
& (1 << ssid
)))
3070 cgroup_for_each_live_child(child
, cgrp
) {
3071 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
3073 if (css_disable
& (1 << ssid
)) {
3076 css_clear_dir(css
, NULL
);
3084 * The effective csses of all the descendants (excluding @cgrp) may
3085 * have changed. Subsystems can optionally subscribe to this event
3086 * by implementing ->css_e_css_changed() which is invoked if any of
3087 * the effective csses seen from the css's cgroup may have changed.
3089 for_each_subsys(ss
, ssid
) {
3090 struct cgroup_subsys_state
*this_css
= cgroup_css(cgrp
, ss
);
3091 struct cgroup_subsys_state
*css
;
3093 if (!ss
->css_e_css_changed
|| !this_css
)
3096 css_for_each_descendant_pre(css
, this_css
)
3097 if (css
!= this_css
)
3098 ss
->css_e_css_changed(css
);
3101 kernfs_activate(cgrp
->kn
);
3104 cgroup_kn_unlock(of
->kn
);
3105 return ret
?: nbytes
;
3108 cgrp
->subtree_control
= old_sc
;
3109 cgrp
->child_subsys_mask
= old_ss
;
3111 for_each_subsys(ss
, ssid
) {
3112 if (!(enable
& (1 << ssid
)))
3115 cgroup_for_each_live_child(child
, cgrp
) {
3116 struct cgroup_subsys_state
*css
= cgroup_css(child
, ss
);
3121 if (css_enable
& (1 << ssid
))
3124 css_clear_dir(css
, NULL
);
3130 static int cgroup_events_show(struct seq_file
*seq
, void *v
)
3132 seq_printf(seq
, "populated %d\n",
3133 cgroup_is_populated(seq_css(seq
)->cgroup
));
3137 static ssize_t
cgroup_file_write(struct kernfs_open_file
*of
, char *buf
,
3138 size_t nbytes
, loff_t off
)
3140 struct cgroup
*cgrp
= of
->kn
->parent
->priv
;
3141 struct cftype
*cft
= of
->kn
->priv
;
3142 struct cgroup_subsys_state
*css
;
3146 return cft
->write(of
, buf
, nbytes
, off
);
3149 * kernfs guarantees that a file isn't deleted with operations in
3150 * flight, which means that the matching css is and stays alive and
3151 * doesn't need to be pinned. The RCU locking is not necessary
3152 * either. It's just for the convenience of using cgroup_css().
3155 css
= cgroup_css(cgrp
, cft
->ss
);
3158 if (cft
->write_u64
) {
3159 unsigned long long v
;
3160 ret
= kstrtoull(buf
, 0, &v
);
3162 ret
= cft
->write_u64(css
, cft
, v
);
3163 } else if (cft
->write_s64
) {
3165 ret
= kstrtoll(buf
, 0, &v
);
3167 ret
= cft
->write_s64(css
, cft
, v
);
3172 return ret
?: nbytes
;
3175 static void *cgroup_seqfile_start(struct seq_file
*seq
, loff_t
*ppos
)
3177 return seq_cft(seq
)->seq_start(seq
, ppos
);
3180 static void *cgroup_seqfile_next(struct seq_file
*seq
, void *v
, loff_t
*ppos
)
3182 return seq_cft(seq
)->seq_next(seq
, v
, ppos
);
3185 static void cgroup_seqfile_stop(struct seq_file
*seq
, void *v
)
3187 seq_cft(seq
)->seq_stop(seq
, v
);
3190 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
3192 struct cftype
*cft
= seq_cft(m
);
3193 struct cgroup_subsys_state
*css
= seq_css(m
);
3196 return cft
->seq_show(m
, arg
);
3199 seq_printf(m
, "%llu\n", cft
->read_u64(css
, cft
));
3200 else if (cft
->read_s64
)
3201 seq_printf(m
, "%lld\n", cft
->read_s64(css
, cft
));
3207 static struct kernfs_ops cgroup_kf_single_ops
= {
3208 .atomic_write_len
= PAGE_SIZE
,
3209 .write
= cgroup_file_write
,
3210 .seq_show
= cgroup_seqfile_show
,
3213 static struct kernfs_ops cgroup_kf_ops
= {
3214 .atomic_write_len
= PAGE_SIZE
,
3215 .write
= cgroup_file_write
,
3216 .seq_start
= cgroup_seqfile_start
,
3217 .seq_next
= cgroup_seqfile_next
,
3218 .seq_stop
= cgroup_seqfile_stop
,
3219 .seq_show
= cgroup_seqfile_show
,
3223 * cgroup_rename - Only allow simple rename of directories in place.
3225 static int cgroup_rename(struct kernfs_node
*kn
, struct kernfs_node
*new_parent
,
3226 const char *new_name_str
)
3228 struct cgroup
*cgrp
= kn
->priv
;
3231 if (kernfs_type(kn
) != KERNFS_DIR
)
3233 if (kn
->parent
!= new_parent
)
3237 * This isn't a proper migration and its usefulness is very
3238 * limited. Disallow on the default hierarchy.
3240 if (cgroup_on_dfl(cgrp
))
3244 * We're gonna grab cgroup_mutex which nests outside kernfs
3245 * active_ref. kernfs_rename() doesn't require active_ref
3246 * protection. Break them before grabbing cgroup_mutex.
3248 kernfs_break_active_protection(new_parent
);
3249 kernfs_break_active_protection(kn
);
3251 mutex_lock(&cgroup_mutex
);
3253 ret
= kernfs_rename(kn
, new_parent
, new_name_str
);
3255 mutex_unlock(&cgroup_mutex
);
3257 kernfs_unbreak_active_protection(kn
);
3258 kernfs_unbreak_active_protection(new_parent
);
3262 /* set uid and gid of cgroup dirs and files to that of the creator */
3263 static int cgroup_kn_set_ugid(struct kernfs_node
*kn
)
3265 struct iattr iattr
= { .ia_valid
= ATTR_UID
| ATTR_GID
,
3266 .ia_uid
= current_fsuid(),
3267 .ia_gid
= current_fsgid(), };
3269 if (uid_eq(iattr
.ia_uid
, GLOBAL_ROOT_UID
) &&
3270 gid_eq(iattr
.ia_gid
, GLOBAL_ROOT_GID
))
3273 return kernfs_setattr(kn
, &iattr
);
3276 static int cgroup_add_file(struct cgroup_subsys_state
*css
, struct cgroup
*cgrp
,
3279 char name
[CGROUP_FILE_NAME_MAX
];
3280 struct kernfs_node
*kn
;
3281 struct lock_class_key
*key
= NULL
;
3284 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3285 key
= &cft
->lockdep_key
;
3287 kn
= __kernfs_create_file(cgrp
->kn
, cgroup_file_name(cgrp
, cft
, name
),
3288 cgroup_file_mode(cft
), 0, cft
->kf_ops
, cft
,
3293 ret
= cgroup_kn_set_ugid(kn
);
3299 if (cft
->file_offset
) {
3300 struct cgroup_file
*cfile
= (void *)css
+ cft
->file_offset
;
3304 list_add(&cfile
->node
, &css
->files
);
3311 * cgroup_addrm_files - add or remove files to a cgroup directory
3312 * @css: the target css
3313 * @cgrp: the target cgroup (usually css->cgroup)
3314 * @cfts: array of cftypes to be added
3315 * @is_add: whether to add or remove
3317 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3318 * For removals, this function never fails.
3320 static int cgroup_addrm_files(struct cgroup_subsys_state
*css
,
3321 struct cgroup
*cgrp
, struct cftype cfts
[],
3324 struct cftype
*cft
, *cft_end
= NULL
;
3327 lockdep_assert_held(&cgroup_mutex
);
3330 for (cft
= cfts
; cft
!= cft_end
&& cft
->name
[0] != '\0'; cft
++) {
3331 /* does cft->flags tell us to skip this file on @cgrp? */
3332 if ((cft
->flags
& __CFTYPE_ONLY_ON_DFL
) && !cgroup_on_dfl(cgrp
))
3334 if ((cft
->flags
& __CFTYPE_NOT_ON_DFL
) && cgroup_on_dfl(cgrp
))
3336 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgroup_parent(cgrp
))
3338 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgroup_parent(cgrp
))
3342 ret
= cgroup_add_file(css
, cgrp
, cft
);
3344 pr_warn("%s: failed to add %s, err=%d\n",
3345 __func__
, cft
->name
, ret
);
3351 cgroup_rm_file(cgrp
, cft
);
3357 static int cgroup_apply_cftypes(struct cftype
*cfts
, bool is_add
)
3360 struct cgroup_subsys
*ss
= cfts
[0].ss
;
3361 struct cgroup
*root
= &ss
->root
->cgrp
;
3362 struct cgroup_subsys_state
*css
;
3365 lockdep_assert_held(&cgroup_mutex
);
3367 /* add/rm files for all cgroups created before */
3368 css_for_each_descendant_pre(css
, cgroup_css(root
, ss
)) {
3369 struct cgroup
*cgrp
= css
->cgroup
;
3371 if (cgroup_is_dead(cgrp
))
3374 ret
= cgroup_addrm_files(css
, cgrp
, cfts
, is_add
);
3380 kernfs_activate(root
->kn
);
3384 static void cgroup_exit_cftypes(struct cftype
*cfts
)
3388 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3389 /* free copy for custom atomic_write_len, see init_cftypes() */
3390 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
)
3395 /* revert flags set by cgroup core while adding @cfts */
3396 cft
->flags
&= ~(__CFTYPE_ONLY_ON_DFL
| __CFTYPE_NOT_ON_DFL
);
3400 static int cgroup_init_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3404 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
3405 struct kernfs_ops
*kf_ops
;
3407 WARN_ON(cft
->ss
|| cft
->kf_ops
);
3410 kf_ops
= &cgroup_kf_ops
;
3412 kf_ops
= &cgroup_kf_single_ops
;
3415 * Ugh... if @cft wants a custom max_write_len, we need to
3416 * make a copy of kf_ops to set its atomic_write_len.
3418 if (cft
->max_write_len
&& cft
->max_write_len
!= PAGE_SIZE
) {
3419 kf_ops
= kmemdup(kf_ops
, sizeof(*kf_ops
), GFP_KERNEL
);
3421 cgroup_exit_cftypes(cfts
);
3424 kf_ops
->atomic_write_len
= cft
->max_write_len
;
3427 cft
->kf_ops
= kf_ops
;
3434 static int cgroup_rm_cftypes_locked(struct cftype
*cfts
)
3436 lockdep_assert_held(&cgroup_mutex
);
3438 if (!cfts
|| !cfts
[0].ss
)
3441 list_del(&cfts
->node
);
3442 cgroup_apply_cftypes(cfts
, false);
3443 cgroup_exit_cftypes(cfts
);
3448 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3449 * @cfts: zero-length name terminated array of cftypes
3451 * Unregister @cfts. Files described by @cfts are removed from all
3452 * existing cgroups and all future cgroups won't have them either. This
3453 * function can be called anytime whether @cfts' subsys is attached or not.
3455 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3458 int cgroup_rm_cftypes(struct cftype
*cfts
)
3462 mutex_lock(&cgroup_mutex
);
3463 ret
= cgroup_rm_cftypes_locked(cfts
);
3464 mutex_unlock(&cgroup_mutex
);
3469 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3470 * @ss: target cgroup subsystem
3471 * @cfts: zero-length name terminated array of cftypes
3473 * Register @cfts to @ss. Files described by @cfts are created for all
3474 * existing cgroups to which @ss is attached and all future cgroups will
3475 * have them too. This function can be called anytime whether @ss is
3478 * Returns 0 on successful registration, -errno on failure. Note that this
3479 * function currently returns 0 as long as @cfts registration is successful
3480 * even if some file creation attempts on existing cgroups fail.
3482 static int cgroup_add_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3486 if (!cgroup_ssid_enabled(ss
->id
))
3489 if (!cfts
|| cfts
[0].name
[0] == '\0')
3492 ret
= cgroup_init_cftypes(ss
, cfts
);
3496 mutex_lock(&cgroup_mutex
);
3498 list_add_tail(&cfts
->node
, &ss
->cfts
);
3499 ret
= cgroup_apply_cftypes(cfts
, true);
3501 cgroup_rm_cftypes_locked(cfts
);
3503 mutex_unlock(&cgroup_mutex
);
3508 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3509 * @ss: target cgroup subsystem
3510 * @cfts: zero-length name terminated array of cftypes
3512 * Similar to cgroup_add_cftypes() but the added files are only used for
3513 * the default hierarchy.
3515 int cgroup_add_dfl_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3519 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3520 cft
->flags
|= __CFTYPE_ONLY_ON_DFL
;
3521 return cgroup_add_cftypes(ss
, cfts
);
3525 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3526 * @ss: target cgroup subsystem
3527 * @cfts: zero-length name terminated array of cftypes
3529 * Similar to cgroup_add_cftypes() but the added files are only used for
3530 * the legacy hierarchies.
3532 int cgroup_add_legacy_cftypes(struct cgroup_subsys
*ss
, struct cftype
*cfts
)
3537 * If legacy_flies_on_dfl, we want to show the legacy files on the
3538 * dfl hierarchy but iff the target subsystem hasn't been updated
3539 * for the dfl hierarchy yet.
3541 if (!cgroup_legacy_files_on_dfl
||
3542 ss
->dfl_cftypes
!= ss
->legacy_cftypes
) {
3543 for (cft
= cfts
; cft
&& cft
->name
[0] != '\0'; cft
++)
3544 cft
->flags
|= __CFTYPE_NOT_ON_DFL
;
3547 return cgroup_add_cftypes(ss
, cfts
);
3551 * cgroup_task_count - count the number of tasks in a cgroup.
3552 * @cgrp: the cgroup in question
3554 * Return the number of tasks in the cgroup.
3556 static int cgroup_task_count(const struct cgroup
*cgrp
)
3559 struct cgrp_cset_link
*link
;
3561 down_read(&css_set_rwsem
);
3562 list_for_each_entry(link
, &cgrp
->cset_links
, cset_link
)
3563 count
+= atomic_read(&link
->cset
->refcount
);
3564 up_read(&css_set_rwsem
);
3569 * css_next_child - find the next child of a given css
3570 * @pos: the current position (%NULL to initiate traversal)
3571 * @parent: css whose children to walk
3573 * This function returns the next child of @parent and should be called
3574 * under either cgroup_mutex or RCU read lock. The only requirement is
3575 * that @parent and @pos are accessible. The next sibling is guaranteed to
3576 * be returned regardless of their states.
3578 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3579 * css which finished ->css_online() is guaranteed to be visible in the
3580 * future iterations and will stay visible until the last reference is put.
3581 * A css which hasn't finished ->css_online() or already finished
3582 * ->css_offline() may show up during traversal. It's each subsystem's
3583 * responsibility to synchronize against on/offlining.
3585 struct cgroup_subsys_state
*css_next_child(struct cgroup_subsys_state
*pos
,
3586 struct cgroup_subsys_state
*parent
)
3588 struct cgroup_subsys_state
*next
;
3590 cgroup_assert_mutex_or_rcu_locked();
3593 * @pos could already have been unlinked from the sibling list.
3594 * Once a cgroup is removed, its ->sibling.next is no longer
3595 * updated when its next sibling changes. CSS_RELEASED is set when
3596 * @pos is taken off list, at which time its next pointer is valid,
3597 * and, as releases are serialized, the one pointed to by the next
3598 * pointer is guaranteed to not have started release yet. This
3599 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3600 * critical section, the one pointed to by its next pointer is
3601 * guaranteed to not have finished its RCU grace period even if we
3602 * have dropped rcu_read_lock() inbetween iterations.
3604 * If @pos has CSS_RELEASED set, its next pointer can't be
3605 * dereferenced; however, as each css is given a monotonically
3606 * increasing unique serial number and always appended to the
3607 * sibling list, the next one can be found by walking the parent's
3608 * children until the first css with higher serial number than
3609 * @pos's. While this path can be slower, it happens iff iteration
3610 * races against release and the race window is very small.
3613 next
= list_entry_rcu(parent
->children
.next
, struct cgroup_subsys_state
, sibling
);
3614 } else if (likely(!(pos
->flags
& CSS_RELEASED
))) {
3615 next
= list_entry_rcu(pos
->sibling
.next
, struct cgroup_subsys_state
, sibling
);
3617 list_for_each_entry_rcu(next
, &parent
->children
, sibling
)
3618 if (next
->serial_nr
> pos
->serial_nr
)
3623 * @next, if not pointing to the head, can be dereferenced and is
3626 if (&next
->sibling
!= &parent
->children
)
3632 * css_next_descendant_pre - find the next descendant for pre-order walk
3633 * @pos: the current position (%NULL to initiate traversal)
3634 * @root: css whose descendants to walk
3636 * To be used by css_for_each_descendant_pre(). Find the next descendant
3637 * to visit for pre-order traversal of @root's descendants. @root is
3638 * included in the iteration and the first node to be visited.
3640 * While this function requires cgroup_mutex or RCU read locking, it
3641 * doesn't require the whole traversal to be contained in a single critical
3642 * section. This function will return the correct next descendant as long
3643 * as both @pos and @root are accessible and @pos is a descendant of @root.
3645 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3646 * css which finished ->css_online() is guaranteed to be visible in the
3647 * future iterations and will stay visible until the last reference is put.
3648 * A css which hasn't finished ->css_online() or already finished
3649 * ->css_offline() may show up during traversal. It's each subsystem's
3650 * responsibility to synchronize against on/offlining.
3652 struct cgroup_subsys_state
*
3653 css_next_descendant_pre(struct cgroup_subsys_state
*pos
,
3654 struct cgroup_subsys_state
*root
)
3656 struct cgroup_subsys_state
*next
;
3658 cgroup_assert_mutex_or_rcu_locked();
3660 /* if first iteration, visit @root */
3664 /* visit the first child if exists */
3665 next
= css_next_child(NULL
, pos
);
3669 /* no child, visit my or the closest ancestor's next sibling */
3670 while (pos
!= root
) {
3671 next
= css_next_child(pos
, pos
->parent
);
3681 * css_rightmost_descendant - return the rightmost descendant of a css
3682 * @pos: css of interest
3684 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3685 * is returned. This can be used during pre-order traversal to skip
3688 * While this function requires cgroup_mutex or RCU read locking, it
3689 * doesn't require the whole traversal to be contained in a single critical
3690 * section. This function will return the correct rightmost descendant as
3691 * long as @pos is accessible.
3693 struct cgroup_subsys_state
*
3694 css_rightmost_descendant(struct cgroup_subsys_state
*pos
)
3696 struct cgroup_subsys_state
*last
, *tmp
;
3698 cgroup_assert_mutex_or_rcu_locked();
3702 /* ->prev isn't RCU safe, walk ->next till the end */
3704 css_for_each_child(tmp
, last
)
3711 static struct cgroup_subsys_state
*
3712 css_leftmost_descendant(struct cgroup_subsys_state
*pos
)
3714 struct cgroup_subsys_state
*last
;
3718 pos
= css_next_child(NULL
, pos
);
3725 * css_next_descendant_post - find the next descendant for post-order walk
3726 * @pos: the current position (%NULL to initiate traversal)
3727 * @root: css whose descendants to walk
3729 * To be used by css_for_each_descendant_post(). Find the next descendant
3730 * to visit for post-order traversal of @root's descendants. @root is
3731 * included in the iteration and the last node to be visited.
3733 * While this function requires cgroup_mutex or RCU read locking, it
3734 * doesn't require the whole traversal to be contained in a single critical
3735 * section. This function will return the correct next descendant as long
3736 * as both @pos and @cgroup are accessible and @pos is a descendant of
3739 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3740 * css which finished ->css_online() is guaranteed to be visible in the
3741 * future iterations and will stay visible until the last reference is put.
3742 * A css which hasn't finished ->css_online() or already finished
3743 * ->css_offline() may show up during traversal. It's each subsystem's
3744 * responsibility to synchronize against on/offlining.
3746 struct cgroup_subsys_state
*
3747 css_next_descendant_post(struct cgroup_subsys_state
*pos
,
3748 struct cgroup_subsys_state
*root
)
3750 struct cgroup_subsys_state
*next
;
3752 cgroup_assert_mutex_or_rcu_locked();
3754 /* if first iteration, visit leftmost descendant which may be @root */
3756 return css_leftmost_descendant(root
);
3758 /* if we visited @root, we're done */
3762 /* if there's an unvisited sibling, visit its leftmost descendant */
3763 next
= css_next_child(pos
, pos
->parent
);
3765 return css_leftmost_descendant(next
);
3767 /* no sibling left, visit parent */
3772 * css_has_online_children - does a css have online children
3773 * @css: the target css
3775 * Returns %true if @css has any online children; otherwise, %false. This
3776 * function can be called from any context but the caller is responsible
3777 * for synchronizing against on/offlining as necessary.
3779 bool css_has_online_children(struct cgroup_subsys_state
*css
)
3781 struct cgroup_subsys_state
*child
;
3785 css_for_each_child(child
, css
) {
3786 if (child
->flags
& CSS_ONLINE
) {
3796 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
3797 * @it: the iterator to advance
3799 * Advance @it to the next css_set to walk.
3801 static void css_task_iter_advance_css_set(struct css_task_iter
*it
)
3803 struct list_head
*l
= it
->cset_pos
;
3804 struct cgrp_cset_link
*link
;
3805 struct css_set
*cset
;
3807 /* Advance to the next non-empty css_set */
3810 if (l
== it
->cset_head
) {
3811 it
->cset_pos
= NULL
;
3812 it
->task_pos
= NULL
;
3817 cset
= container_of(l
, struct css_set
,
3818 e_cset_node
[it
->ss
->id
]);
3820 link
= list_entry(l
, struct cgrp_cset_link
, cset_link
);
3823 } while (!css_set_populated(cset
));
3827 if (!list_empty(&cset
->tasks
))
3828 it
->task_pos
= cset
->tasks
.next
;
3830 it
->task_pos
= cset
->mg_tasks
.next
;
3832 it
->tasks_head
= &cset
->tasks
;
3833 it
->mg_tasks_head
= &cset
->mg_tasks
;
3836 static void css_task_iter_advance(struct css_task_iter
*it
)
3838 struct list_head
*l
= it
->task_pos
;
3843 * Advance iterator to find next entry. cset->tasks is consumed
3844 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
3849 if (l
== it
->tasks_head
)
3850 l
= it
->mg_tasks_head
->next
;
3852 if (l
== it
->mg_tasks_head
)
3853 css_task_iter_advance_css_set(it
);
3859 * css_task_iter_start - initiate task iteration
3860 * @css: the css to walk tasks of
3861 * @it: the task iterator to use
3863 * Initiate iteration through the tasks of @css. The caller can call
3864 * css_task_iter_next() to walk through the tasks until the function
3865 * returns NULL. On completion of iteration, css_task_iter_end() must be
3868 * Note that this function acquires a lock which is released when the
3869 * iteration finishes. The caller can't sleep while iteration is in
3872 void css_task_iter_start(struct cgroup_subsys_state
*css
,
3873 struct css_task_iter
*it
)
3874 __acquires(css_set_rwsem
)
3876 /* no one should try to iterate before mounting cgroups */
3877 WARN_ON_ONCE(!use_task_css_set_links
);
3879 down_read(&css_set_rwsem
);
3884 it
->cset_pos
= &css
->cgroup
->e_csets
[css
->ss
->id
];
3886 it
->cset_pos
= &css
->cgroup
->cset_links
;
3888 it
->cset_head
= it
->cset_pos
;
3890 css_task_iter_advance_css_set(it
);
3894 * css_task_iter_next - return the next task for the iterator
3895 * @it: the task iterator being iterated
3897 * The "next" function for task iteration. @it should have been
3898 * initialized via css_task_iter_start(). Returns NULL when the iteration
3901 struct task_struct
*css_task_iter_next(struct css_task_iter
*it
)
3903 struct task_struct
*res
;
3908 res
= list_entry(it
->task_pos
, struct task_struct
, cg_list
);
3909 css_task_iter_advance(it
);
3914 * css_task_iter_end - finish task iteration
3915 * @it: the task iterator to finish
3917 * Finish task iteration started by css_task_iter_start().
3919 void css_task_iter_end(struct css_task_iter
*it
)
3920 __releases(css_set_rwsem
)
3922 up_read(&css_set_rwsem
);
3926 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3927 * @to: cgroup to which the tasks will be moved
3928 * @from: cgroup in which the tasks currently reside
3930 * Locking rules between cgroup_post_fork() and the migration path
3931 * guarantee that, if a task is forking while being migrated, the new child
3932 * is guaranteed to be either visible in the source cgroup after the
3933 * parent's migration is complete or put into the target cgroup. No task
3934 * can slip out of migration through forking.
3936 int cgroup_transfer_tasks(struct cgroup
*to
, struct cgroup
*from
)
3938 LIST_HEAD(preloaded_csets
);
3939 struct cgrp_cset_link
*link
;
3940 struct css_task_iter it
;
3941 struct task_struct
*task
;
3944 mutex_lock(&cgroup_mutex
);
3946 /* all tasks in @from are being moved, all csets are source */
3947 down_read(&css_set_rwsem
);
3948 list_for_each_entry(link
, &from
->cset_links
, cset_link
)
3949 cgroup_migrate_add_src(link
->cset
, to
, &preloaded_csets
);
3950 up_read(&css_set_rwsem
);
3952 ret
= cgroup_migrate_prepare_dst(to
, &preloaded_csets
);
3957 * Migrate tasks one-by-one until @form is empty. This fails iff
3958 * ->can_attach() fails.
3961 css_task_iter_start(&from
->self
, &it
);
3962 task
= css_task_iter_next(&it
);
3964 get_task_struct(task
);
3965 css_task_iter_end(&it
);
3968 ret
= cgroup_migrate(task
, false, to
);
3969 put_task_struct(task
);
3971 } while (task
&& !ret
);
3973 cgroup_migrate_finish(&preloaded_csets
);
3974 mutex_unlock(&cgroup_mutex
);
3979 * Stuff for reading the 'tasks'/'procs' files.
3981 * Reading this file can return large amounts of data if a cgroup has
3982 * *lots* of attached tasks. So it may need several calls to read(),
3983 * but we cannot guarantee that the information we produce is correct
3984 * unless we produce it entirely atomically.
3988 /* which pidlist file are we talking about? */
3989 enum cgroup_filetype
{
3995 * A pidlist is a list of pids that virtually represents the contents of one
3996 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3997 * a pair (one each for procs, tasks) for each pid namespace that's relevant
4000 struct cgroup_pidlist
{
4002 * used to find which pidlist is wanted. doesn't change as long as
4003 * this particular list stays in the list.
4005 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
4008 /* how many elements the above list has */
4010 /* each of these stored in a list by its cgroup */
4011 struct list_head links
;
4012 /* pointer to the cgroup we belong to, for list removal purposes */
4013 struct cgroup
*owner
;
4014 /* for delayed destruction */
4015 struct delayed_work destroy_dwork
;
4019 * The following two functions "fix" the issue where there are more pids
4020 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
4021 * TODO: replace with a kernel-wide solution to this problem
4023 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
4024 static void *pidlist_allocate(int count
)
4026 if (PIDLIST_TOO_LARGE(count
))
4027 return vmalloc(count
* sizeof(pid_t
));
4029 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
4032 static void pidlist_free(void *p
)
4038 * Used to destroy all pidlists lingering waiting for destroy timer. None
4039 * should be left afterwards.
4041 static void cgroup_pidlist_destroy_all(struct cgroup
*cgrp
)
4043 struct cgroup_pidlist
*l
, *tmp_l
;
4045 mutex_lock(&cgrp
->pidlist_mutex
);
4046 list_for_each_entry_safe(l
, tmp_l
, &cgrp
->pidlists
, links
)
4047 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
, 0);
4048 mutex_unlock(&cgrp
->pidlist_mutex
);
4050 flush_workqueue(cgroup_pidlist_destroy_wq
);
4051 BUG_ON(!list_empty(&cgrp
->pidlists
));
4054 static void cgroup_pidlist_destroy_work_fn(struct work_struct
*work
)
4056 struct delayed_work
*dwork
= to_delayed_work(work
);
4057 struct cgroup_pidlist
*l
= container_of(dwork
, struct cgroup_pidlist
,
4059 struct cgroup_pidlist
*tofree
= NULL
;
4061 mutex_lock(&l
->owner
->pidlist_mutex
);
4064 * Destroy iff we didn't get queued again. The state won't change
4065 * as destroy_dwork can only be queued while locked.
4067 if (!delayed_work_pending(dwork
)) {
4068 list_del(&l
->links
);
4069 pidlist_free(l
->list
);
4070 put_pid_ns(l
->key
.ns
);
4074 mutex_unlock(&l
->owner
->pidlist_mutex
);
4079 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
4080 * Returns the number of unique elements.
4082 static int pidlist_uniq(pid_t
*list
, int length
)
4087 * we presume the 0th element is unique, so i starts at 1. trivial
4088 * edge cases first; no work needs to be done for either
4090 if (length
== 0 || length
== 1)
4092 /* src and dest walk down the list; dest counts unique elements */
4093 for (src
= 1; src
< length
; src
++) {
4094 /* find next unique element */
4095 while (list
[src
] == list
[src
-1]) {
4100 /* dest always points to where the next unique element goes */
4101 list
[dest
] = list
[src
];
4109 * The two pid files - task and cgroup.procs - guaranteed that the result
4110 * is sorted, which forced this whole pidlist fiasco. As pid order is
4111 * different per namespace, each namespace needs differently sorted list,
4112 * making it impossible to use, for example, single rbtree of member tasks
4113 * sorted by task pointer. As pidlists can be fairly large, allocating one
4114 * per open file is dangerous, so cgroup had to implement shared pool of
4115 * pidlists keyed by cgroup and namespace.
4117 * All this extra complexity was caused by the original implementation
4118 * committing to an entirely unnecessary property. In the long term, we
4119 * want to do away with it. Explicitly scramble sort order if on the
4120 * default hierarchy so that no such expectation exists in the new
4123 * Scrambling is done by swapping every two consecutive bits, which is
4124 * non-identity one-to-one mapping which disturbs sort order sufficiently.
4126 static pid_t
pid_fry(pid_t pid
)
4128 unsigned a
= pid
& 0x55555555;
4129 unsigned b
= pid
& 0xAAAAAAAA;
4131 return (a
<< 1) | (b
>> 1);
4134 static pid_t
cgroup_pid_fry(struct cgroup
*cgrp
, pid_t pid
)
4136 if (cgroup_on_dfl(cgrp
))
4137 return pid_fry(pid
);
4142 static int cmppid(const void *a
, const void *b
)
4144 return *(pid_t
*)a
- *(pid_t
*)b
;
4147 static int fried_cmppid(const void *a
, const void *b
)
4149 return pid_fry(*(pid_t
*)a
) - pid_fry(*(pid_t
*)b
);
4152 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
4153 enum cgroup_filetype type
)
4155 struct cgroup_pidlist
*l
;
4156 /* don't need task_nsproxy() if we're looking at ourself */
4157 struct pid_namespace
*ns
= task_active_pid_ns(current
);
4159 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4161 list_for_each_entry(l
, &cgrp
->pidlists
, links
)
4162 if (l
->key
.type
== type
&& l
->key
.ns
== ns
)
4168 * find the appropriate pidlist for our purpose (given procs vs tasks)
4169 * returns with the lock on that pidlist already held, and takes care
4170 * of the use count, or returns NULL with no locks held if we're out of
4173 static struct cgroup_pidlist
*cgroup_pidlist_find_create(struct cgroup
*cgrp
,
4174 enum cgroup_filetype type
)
4176 struct cgroup_pidlist
*l
;
4178 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4180 l
= cgroup_pidlist_find(cgrp
, type
);
4184 /* entry not found; create a new one */
4185 l
= kzalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
4189 INIT_DELAYED_WORK(&l
->destroy_dwork
, cgroup_pidlist_destroy_work_fn
);
4191 /* don't need task_nsproxy() if we're looking at ourself */
4192 l
->key
.ns
= get_pid_ns(task_active_pid_ns(current
));
4194 list_add(&l
->links
, &cgrp
->pidlists
);
4199 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
4201 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
4202 struct cgroup_pidlist
**lp
)
4206 int pid
, n
= 0; /* used for populating the array */
4207 struct css_task_iter it
;
4208 struct task_struct
*tsk
;
4209 struct cgroup_pidlist
*l
;
4211 lockdep_assert_held(&cgrp
->pidlist_mutex
);
4214 * If cgroup gets more users after we read count, we won't have
4215 * enough space - tough. This race is indistinguishable to the
4216 * caller from the case that the additional cgroup users didn't
4217 * show up until sometime later on.
4219 length
= cgroup_task_count(cgrp
);
4220 array
= pidlist_allocate(length
);
4223 /* now, populate the array */
4224 css_task_iter_start(&cgrp
->self
, &it
);
4225 while ((tsk
= css_task_iter_next(&it
))) {
4226 if (unlikely(n
== length
))
4228 /* get tgid or pid for procs or tasks file respectively */
4229 if (type
== CGROUP_FILE_PROCS
)
4230 pid
= task_tgid_vnr(tsk
);
4232 pid
= task_pid_vnr(tsk
);
4233 if (pid
> 0) /* make sure to only use valid results */
4236 css_task_iter_end(&it
);
4238 /* now sort & (if procs) strip out duplicates */
4239 if (cgroup_on_dfl(cgrp
))
4240 sort(array
, length
, sizeof(pid_t
), fried_cmppid
, NULL
);
4242 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
4243 if (type
== CGROUP_FILE_PROCS
)
4244 length
= pidlist_uniq(array
, length
);
4246 l
= cgroup_pidlist_find_create(cgrp
, type
);
4248 pidlist_free(array
);
4252 /* store array, freeing old if necessary */
4253 pidlist_free(l
->list
);
4261 * cgroupstats_build - build and fill cgroupstats
4262 * @stats: cgroupstats to fill information into
4263 * @dentry: A dentry entry belonging to the cgroup for which stats have
4266 * Build and fill cgroupstats so that taskstats can export it to user
4269 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
4271 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
4272 struct cgroup
*cgrp
;
4273 struct css_task_iter it
;
4274 struct task_struct
*tsk
;
4276 /* it should be kernfs_node belonging to cgroupfs and is a directory */
4277 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
4278 kernfs_type(kn
) != KERNFS_DIR
)
4281 mutex_lock(&cgroup_mutex
);
4284 * We aren't being called from kernfs and there's no guarantee on
4285 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
4286 * @kn->priv is RCU safe. Let's do the RCU dancing.
4289 cgrp
= rcu_dereference(kn
->priv
);
4290 if (!cgrp
|| cgroup_is_dead(cgrp
)) {
4292 mutex_unlock(&cgroup_mutex
);
4297 css_task_iter_start(&cgrp
->self
, &it
);
4298 while ((tsk
= css_task_iter_next(&it
))) {
4299 switch (tsk
->state
) {
4301 stats
->nr_running
++;
4303 case TASK_INTERRUPTIBLE
:
4304 stats
->nr_sleeping
++;
4306 case TASK_UNINTERRUPTIBLE
:
4307 stats
->nr_uninterruptible
++;
4310 stats
->nr_stopped
++;
4313 if (delayacct_is_task_waiting_on_io(tsk
))
4314 stats
->nr_io_wait
++;
4318 css_task_iter_end(&it
);
4320 mutex_unlock(&cgroup_mutex
);
4326 * seq_file methods for the tasks/procs files. The seq_file position is the
4327 * next pid to display; the seq_file iterator is a pointer to the pid
4328 * in the cgroup->l->list array.
4331 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
4334 * Initially we receive a position value that corresponds to
4335 * one more than the last pid shown (or 0 on the first call or
4336 * after a seek to the start). Use a binary-search to find the
4337 * next pid to display, if any
4339 struct kernfs_open_file
*of
= s
->private;
4340 struct cgroup
*cgrp
= seq_css(s
)->cgroup
;
4341 struct cgroup_pidlist
*l
;
4342 enum cgroup_filetype type
= seq_cft(s
)->private;
4343 int index
= 0, pid
= *pos
;
4346 mutex_lock(&cgrp
->pidlist_mutex
);
4349 * !NULL @of->priv indicates that this isn't the first start()
4350 * after open. If the matching pidlist is around, we can use that.
4351 * Look for it. Note that @of->priv can't be used directly. It
4352 * could already have been destroyed.
4355 of
->priv
= cgroup_pidlist_find(cgrp
, type
);
4358 * Either this is the first start() after open or the matching
4359 * pidlist has been destroyed inbetween. Create a new one.
4362 ret
= pidlist_array_load(cgrp
, type
,
4363 (struct cgroup_pidlist
**)&of
->priv
);
4365 return ERR_PTR(ret
);
4370 int end
= l
->length
;
4372 while (index
< end
) {
4373 int mid
= (index
+ end
) / 2;
4374 if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) == pid
) {
4377 } else if (cgroup_pid_fry(cgrp
, l
->list
[mid
]) <= pid
)
4383 /* If we're off the end of the array, we're done */
4384 if (index
>= l
->length
)
4386 /* Update the abstract position to be the actual pid that we found */
4387 iter
= l
->list
+ index
;
4388 *pos
= cgroup_pid_fry(cgrp
, *iter
);
4392 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
4394 struct kernfs_open_file
*of
= s
->private;
4395 struct cgroup_pidlist
*l
= of
->priv
;
4398 mod_delayed_work(cgroup_pidlist_destroy_wq
, &l
->destroy_dwork
,
4399 CGROUP_PIDLIST_DESTROY_DELAY
);
4400 mutex_unlock(&seq_css(s
)->cgroup
->pidlist_mutex
);
4403 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
4405 struct kernfs_open_file
*of
= s
->private;
4406 struct cgroup_pidlist
*l
= of
->priv
;
4408 pid_t
*end
= l
->list
+ l
->length
;
4410 * Advance to the next pid in the array. If this goes off the
4417 *pos
= cgroup_pid_fry(seq_css(s
)->cgroup
, *p
);
4422 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
4424 seq_printf(s
, "%d\n", *(int *)v
);
4429 static u64
cgroup_read_notify_on_release(struct cgroup_subsys_state
*css
,
4432 return notify_on_release(css
->cgroup
);
4435 static int cgroup_write_notify_on_release(struct cgroup_subsys_state
*css
,
4436 struct cftype
*cft
, u64 val
)
4439 set_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4441 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &css
->cgroup
->flags
);
4445 static u64
cgroup_clone_children_read(struct cgroup_subsys_state
*css
,
4448 return test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4451 static int cgroup_clone_children_write(struct cgroup_subsys_state
*css
,
4452 struct cftype
*cft
, u64 val
)
4455 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4457 clear_bit(CGRP_CPUSET_CLONE_CHILDREN
, &css
->cgroup
->flags
);
4461 /* cgroup core interface files for the default hierarchy */
4462 static struct cftype cgroup_dfl_base_files
[] = {
4464 .name
= "cgroup.procs",
4465 .file_offset
= offsetof(struct cgroup
, procs_file
),
4466 .seq_start
= cgroup_pidlist_start
,
4467 .seq_next
= cgroup_pidlist_next
,
4468 .seq_stop
= cgroup_pidlist_stop
,
4469 .seq_show
= cgroup_pidlist_show
,
4470 .private = CGROUP_FILE_PROCS
,
4471 .write
= cgroup_procs_write
,
4474 .name
= "cgroup.controllers",
4475 .flags
= CFTYPE_ONLY_ON_ROOT
,
4476 .seq_show
= cgroup_root_controllers_show
,
4479 .name
= "cgroup.controllers",
4480 .flags
= CFTYPE_NOT_ON_ROOT
,
4481 .seq_show
= cgroup_controllers_show
,
4484 .name
= "cgroup.subtree_control",
4485 .seq_show
= cgroup_subtree_control_show
,
4486 .write
= cgroup_subtree_control_write
,
4489 .name
= "cgroup.events",
4490 .flags
= CFTYPE_NOT_ON_ROOT
,
4491 .file_offset
= offsetof(struct cgroup
, events_file
),
4492 .seq_show
= cgroup_events_show
,
4497 /* cgroup core interface files for the legacy hierarchies */
4498 static struct cftype cgroup_legacy_base_files
[] = {
4500 .name
= "cgroup.procs",
4501 .seq_start
= cgroup_pidlist_start
,
4502 .seq_next
= cgroup_pidlist_next
,
4503 .seq_stop
= cgroup_pidlist_stop
,
4504 .seq_show
= cgroup_pidlist_show
,
4505 .private = CGROUP_FILE_PROCS
,
4506 .write
= cgroup_procs_write
,
4509 .name
= "cgroup.clone_children",
4510 .read_u64
= cgroup_clone_children_read
,
4511 .write_u64
= cgroup_clone_children_write
,
4514 .name
= "cgroup.sane_behavior",
4515 .flags
= CFTYPE_ONLY_ON_ROOT
,
4516 .seq_show
= cgroup_sane_behavior_show
,
4520 .seq_start
= cgroup_pidlist_start
,
4521 .seq_next
= cgroup_pidlist_next
,
4522 .seq_stop
= cgroup_pidlist_stop
,
4523 .seq_show
= cgroup_pidlist_show
,
4524 .private = CGROUP_FILE_TASKS
,
4525 .write
= cgroup_tasks_write
,
4528 .name
= "notify_on_release",
4529 .read_u64
= cgroup_read_notify_on_release
,
4530 .write_u64
= cgroup_write_notify_on_release
,
4533 .name
= "release_agent",
4534 .flags
= CFTYPE_ONLY_ON_ROOT
,
4535 .seq_show
= cgroup_release_agent_show
,
4536 .write
= cgroup_release_agent_write
,
4537 .max_write_len
= PATH_MAX
- 1,
4543 * css destruction is four-stage process.
4545 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4546 * Implemented in kill_css().
4548 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4549 * and thus css_tryget_online() is guaranteed to fail, the css can be
4550 * offlined by invoking offline_css(). After offlining, the base ref is
4551 * put. Implemented in css_killed_work_fn().
4553 * 3. When the percpu_ref reaches zero, the only possible remaining
4554 * accessors are inside RCU read sections. css_release() schedules the
4557 * 4. After the grace period, the css can be freed. Implemented in
4558 * css_free_work_fn().
4560 * It is actually hairier because both step 2 and 4 require process context
4561 * and thus involve punting to css->destroy_work adding two additional
4562 * steps to the already complex sequence.
4564 static void css_free_work_fn(struct work_struct
*work
)
4566 struct cgroup_subsys_state
*css
=
4567 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4568 struct cgroup_subsys
*ss
= css
->ss
;
4569 struct cgroup
*cgrp
= css
->cgroup
;
4570 struct cgroup_file
*cfile
;
4572 percpu_ref_exit(&css
->refcnt
);
4574 list_for_each_entry(cfile
, &css
->files
, node
)
4575 kernfs_put(cfile
->kn
);
4582 css_put(css
->parent
);
4585 cgroup_idr_remove(&ss
->css_idr
, id
);
4588 /* cgroup free path */
4589 atomic_dec(&cgrp
->root
->nr_cgrps
);
4590 cgroup_pidlist_destroy_all(cgrp
);
4591 cancel_work_sync(&cgrp
->release_agent_work
);
4593 if (cgroup_parent(cgrp
)) {
4595 * We get a ref to the parent, and put the ref when
4596 * this cgroup is being freed, so it's guaranteed
4597 * that the parent won't be destroyed before its
4600 cgroup_put(cgroup_parent(cgrp
));
4601 kernfs_put(cgrp
->kn
);
4605 * This is root cgroup's refcnt reaching zero,
4606 * which indicates that the root should be
4609 cgroup_destroy_root(cgrp
->root
);
4614 static void css_free_rcu_fn(struct rcu_head
*rcu_head
)
4616 struct cgroup_subsys_state
*css
=
4617 container_of(rcu_head
, struct cgroup_subsys_state
, rcu_head
);
4619 INIT_WORK(&css
->destroy_work
, css_free_work_fn
);
4620 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4623 static void css_release_work_fn(struct work_struct
*work
)
4625 struct cgroup_subsys_state
*css
=
4626 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4627 struct cgroup_subsys
*ss
= css
->ss
;
4628 struct cgroup
*cgrp
= css
->cgroup
;
4630 mutex_lock(&cgroup_mutex
);
4632 css
->flags
|= CSS_RELEASED
;
4633 list_del_rcu(&css
->sibling
);
4636 /* css release path */
4637 cgroup_idr_replace(&ss
->css_idr
, NULL
, css
->id
);
4638 if (ss
->css_released
)
4639 ss
->css_released(css
);
4641 /* cgroup release path */
4642 cgroup_idr_remove(&cgrp
->root
->cgroup_idr
, cgrp
->id
);
4646 * There are two control paths which try to determine
4647 * cgroup from dentry without going through kernfs -
4648 * cgroupstats_build() and css_tryget_online_from_dir().
4649 * Those are supported by RCU protecting clearing of
4650 * cgrp->kn->priv backpointer.
4652 RCU_INIT_POINTER(*(void __rcu __force
**)&cgrp
->kn
->priv
, NULL
);
4655 mutex_unlock(&cgroup_mutex
);
4657 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4660 static void css_release(struct percpu_ref
*ref
)
4662 struct cgroup_subsys_state
*css
=
4663 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4665 INIT_WORK(&css
->destroy_work
, css_release_work_fn
);
4666 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4669 static void init_and_link_css(struct cgroup_subsys_state
*css
,
4670 struct cgroup_subsys
*ss
, struct cgroup
*cgrp
)
4672 lockdep_assert_held(&cgroup_mutex
);
4676 memset(css
, 0, sizeof(*css
));
4679 INIT_LIST_HEAD(&css
->sibling
);
4680 INIT_LIST_HEAD(&css
->children
);
4681 INIT_LIST_HEAD(&css
->files
);
4682 css
->serial_nr
= css_serial_nr_next
++;
4684 if (cgroup_parent(cgrp
)) {
4685 css
->parent
= cgroup_css(cgroup_parent(cgrp
), ss
);
4686 css_get(css
->parent
);
4689 BUG_ON(cgroup_css(cgrp
, ss
));
4692 /* invoke ->css_online() on a new CSS and mark it online if successful */
4693 static int online_css(struct cgroup_subsys_state
*css
)
4695 struct cgroup_subsys
*ss
= css
->ss
;
4698 lockdep_assert_held(&cgroup_mutex
);
4701 ret
= ss
->css_online(css
);
4703 css
->flags
|= CSS_ONLINE
;
4704 rcu_assign_pointer(css
->cgroup
->subsys
[ss
->id
], css
);
4709 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4710 static void offline_css(struct cgroup_subsys_state
*css
)
4712 struct cgroup_subsys
*ss
= css
->ss
;
4714 lockdep_assert_held(&cgroup_mutex
);
4716 if (!(css
->flags
& CSS_ONLINE
))
4719 if (ss
->css_offline
)
4720 ss
->css_offline(css
);
4722 css
->flags
&= ~CSS_ONLINE
;
4723 RCU_INIT_POINTER(css
->cgroup
->subsys
[ss
->id
], NULL
);
4725 wake_up_all(&css
->cgroup
->offline_waitq
);
4729 * create_css - create a cgroup_subsys_state
4730 * @cgrp: the cgroup new css will be associated with
4731 * @ss: the subsys of new css
4732 * @visible: whether to create control knobs for the new css or not
4734 * Create a new css associated with @cgrp - @ss pair. On success, the new
4735 * css is online and installed in @cgrp with all interface files created if
4736 * @visible. Returns 0 on success, -errno on failure.
4738 static int create_css(struct cgroup
*cgrp
, struct cgroup_subsys
*ss
,
4741 struct cgroup
*parent
= cgroup_parent(cgrp
);
4742 struct cgroup_subsys_state
*parent_css
= cgroup_css(parent
, ss
);
4743 struct cgroup_subsys_state
*css
;
4746 lockdep_assert_held(&cgroup_mutex
);
4748 css
= ss
->css_alloc(parent_css
);
4750 return PTR_ERR(css
);
4752 init_and_link_css(css
, ss
, cgrp
);
4754 err
= percpu_ref_init(&css
->refcnt
, css_release
, 0, GFP_KERNEL
);
4758 err
= cgroup_idr_alloc(&ss
->css_idr
, NULL
, 2, 0, GFP_KERNEL
);
4760 goto err_free_percpu_ref
;
4764 err
= css_populate_dir(css
, NULL
);
4769 /* @css is ready to be brought online now, make it visible */
4770 list_add_tail_rcu(&css
->sibling
, &parent_css
->children
);
4771 cgroup_idr_replace(&ss
->css_idr
, css
, css
->id
);
4773 err
= online_css(css
);
4777 if (ss
->broken_hierarchy
&& !ss
->warned_broken_hierarchy
&&
4778 cgroup_parent(parent
)) {
4779 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4780 current
->comm
, current
->pid
, ss
->name
);
4781 if (!strcmp(ss
->name
, "memory"))
4782 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4783 ss
->warned_broken_hierarchy
= true;
4789 list_del_rcu(&css
->sibling
);
4790 css_clear_dir(css
, NULL
);
4792 cgroup_idr_remove(&ss
->css_idr
, css
->id
);
4793 err_free_percpu_ref
:
4794 percpu_ref_exit(&css
->refcnt
);
4796 call_rcu(&css
->rcu_head
, css_free_rcu_fn
);
4800 static int cgroup_mkdir(struct kernfs_node
*parent_kn
, const char *name
,
4803 struct cgroup
*parent
, *cgrp
;
4804 struct cgroup_root
*root
;
4805 struct cgroup_subsys
*ss
;
4806 struct kernfs_node
*kn
;
4809 /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable.
4811 if (strchr(name
, '\n'))
4814 parent
= cgroup_kn_lock_live(parent_kn
);
4817 root
= parent
->root
;
4819 /* allocate the cgroup and its ID, 0 is reserved for the root */
4820 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
4826 ret
= percpu_ref_init(&cgrp
->self
.refcnt
, css_release
, 0, GFP_KERNEL
);
4831 * Temporarily set the pointer to NULL, so idr_find() won't return
4832 * a half-baked cgroup.
4834 cgrp
->id
= cgroup_idr_alloc(&root
->cgroup_idr
, NULL
, 2, 0, GFP_KERNEL
);
4837 goto out_cancel_ref
;
4840 init_cgroup_housekeeping(cgrp
);
4842 cgrp
->self
.parent
= &parent
->self
;
4845 if (notify_on_release(parent
))
4846 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
4848 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN
, &parent
->flags
))
4849 set_bit(CGRP_CPUSET_CLONE_CHILDREN
, &cgrp
->flags
);
4851 /* create the directory */
4852 kn
= kernfs_create_dir(parent
->kn
, name
, mode
, cgrp
);
4860 * This extra ref will be put in cgroup_free_fn() and guarantees
4861 * that @cgrp->kn is always accessible.
4865 cgrp
->self
.serial_nr
= css_serial_nr_next
++;
4867 /* allocation complete, commit to creation */
4868 list_add_tail_rcu(&cgrp
->self
.sibling
, &cgroup_parent(cgrp
)->self
.children
);
4869 atomic_inc(&root
->nr_cgrps
);
4873 * @cgrp is now fully operational. If something fails after this
4874 * point, it'll be released via the normal destruction path.
4876 cgroup_idr_replace(&root
->cgroup_idr
, cgrp
, cgrp
->id
);
4878 ret
= cgroup_kn_set_ugid(kn
);
4882 ret
= css_populate_dir(&cgrp
->self
, NULL
);
4886 /* let's create and online css's */
4887 for_each_subsys(ss
, ssid
) {
4888 if (parent
->child_subsys_mask
& (1 << ssid
)) {
4889 ret
= create_css(cgrp
, ss
,
4890 parent
->subtree_control
& (1 << ssid
));
4897 * On the default hierarchy, a child doesn't automatically inherit
4898 * subtree_control from the parent. Each is configured manually.
4900 if (!cgroup_on_dfl(cgrp
)) {
4901 cgrp
->subtree_control
= parent
->subtree_control
;
4902 cgroup_refresh_child_subsys_mask(cgrp
);
4905 kernfs_activate(kn
);
4911 cgroup_idr_remove(&root
->cgroup_idr
, cgrp
->id
);
4913 percpu_ref_exit(&cgrp
->self
.refcnt
);
4917 cgroup_kn_unlock(parent_kn
);
4921 cgroup_destroy_locked(cgrp
);
4926 * This is called when the refcnt of a css is confirmed to be killed.
4927 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4928 * initate destruction and put the css ref from kill_css().
4930 static void css_killed_work_fn(struct work_struct
*work
)
4932 struct cgroup_subsys_state
*css
=
4933 container_of(work
, struct cgroup_subsys_state
, destroy_work
);
4935 mutex_lock(&cgroup_mutex
);
4937 mutex_unlock(&cgroup_mutex
);
4942 /* css kill confirmation processing requires process context, bounce */
4943 static void css_killed_ref_fn(struct percpu_ref
*ref
)
4945 struct cgroup_subsys_state
*css
=
4946 container_of(ref
, struct cgroup_subsys_state
, refcnt
);
4948 INIT_WORK(&css
->destroy_work
, css_killed_work_fn
);
4949 queue_work(cgroup_destroy_wq
, &css
->destroy_work
);
4953 * kill_css - destroy a css
4954 * @css: css to destroy
4956 * This function initiates destruction of @css by removing cgroup interface
4957 * files and putting its base reference. ->css_offline() will be invoked
4958 * asynchronously once css_tryget_online() is guaranteed to fail and when
4959 * the reference count reaches zero, @css will be released.
4961 static void kill_css(struct cgroup_subsys_state
*css
)
4963 lockdep_assert_held(&cgroup_mutex
);
4966 * This must happen before css is disassociated with its cgroup.
4967 * See seq_css() for details.
4969 css_clear_dir(css
, NULL
);
4972 * Killing would put the base ref, but we need to keep it alive
4973 * until after ->css_offline().
4978 * cgroup core guarantees that, by the time ->css_offline() is
4979 * invoked, no new css reference will be given out via
4980 * css_tryget_online(). We can't simply call percpu_ref_kill() and
4981 * proceed to offlining css's because percpu_ref_kill() doesn't
4982 * guarantee that the ref is seen as killed on all CPUs on return.
4984 * Use percpu_ref_kill_and_confirm() to get notifications as each
4985 * css is confirmed to be seen as killed on all CPUs.
4987 percpu_ref_kill_and_confirm(&css
->refcnt
, css_killed_ref_fn
);
4991 * cgroup_destroy_locked - the first stage of cgroup destruction
4992 * @cgrp: cgroup to be destroyed
4994 * css's make use of percpu refcnts whose killing latency shouldn't be
4995 * exposed to userland and are RCU protected. Also, cgroup core needs to
4996 * guarantee that css_tryget_online() won't succeed by the time
4997 * ->css_offline() is invoked. To satisfy all the requirements,
4998 * destruction is implemented in the following two steps.
5000 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5001 * userland visible parts and start killing the percpu refcnts of
5002 * css's. Set up so that the next stage will be kicked off once all
5003 * the percpu refcnts are confirmed to be killed.
5005 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5006 * rest of destruction. Once all cgroup references are gone, the
5007 * cgroup is RCU-freed.
5009 * This function implements s1. After this step, @cgrp is gone as far as
5010 * the userland is concerned and a new cgroup with the same name may be
5011 * created. As cgroup doesn't care about the names internally, this
5012 * doesn't cause any problem.
5014 static int cgroup_destroy_locked(struct cgroup
*cgrp
)
5015 __releases(&cgroup_mutex
) __acquires(&cgroup_mutex
)
5017 struct cgroup_subsys_state
*css
;
5020 lockdep_assert_held(&cgroup_mutex
);
5023 * Only migration can raise populated from zero and we're already
5024 * holding cgroup_mutex.
5026 if (cgroup_is_populated(cgrp
))
5030 * Make sure there's no live children. We can't test emptiness of
5031 * ->self.children as dead children linger on it while being
5032 * drained; otherwise, "rmdir parent/child parent" may fail.
5034 if (css_has_online_children(&cgrp
->self
))
5038 * Mark @cgrp dead. This prevents further task migration and child
5039 * creation by disabling cgroup_lock_live_group().
5041 cgrp
->self
.flags
&= ~CSS_ONLINE
;
5043 /* initiate massacre of all css's */
5044 for_each_css(css
, ssid
, cgrp
)
5048 * Remove @cgrp directory along with the base files. @cgrp has an
5049 * extra ref on its kn.
5051 kernfs_remove(cgrp
->kn
);
5053 check_for_release(cgroup_parent(cgrp
));
5055 /* put the base reference */
5056 percpu_ref_kill(&cgrp
->self
.refcnt
);
5061 static int cgroup_rmdir(struct kernfs_node
*kn
)
5063 struct cgroup
*cgrp
;
5066 cgrp
= cgroup_kn_lock_live(kn
);
5070 ret
= cgroup_destroy_locked(cgrp
);
5072 cgroup_kn_unlock(kn
);
5076 static struct kernfs_syscall_ops cgroup_kf_syscall_ops
= {
5077 .remount_fs
= cgroup_remount
,
5078 .show_options
= cgroup_show_options
,
5079 .mkdir
= cgroup_mkdir
,
5080 .rmdir
= cgroup_rmdir
,
5081 .rename
= cgroup_rename
,
5084 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
, bool early
)
5086 struct cgroup_subsys_state
*css
;
5088 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
5090 mutex_lock(&cgroup_mutex
);
5092 idr_init(&ss
->css_idr
);
5093 INIT_LIST_HEAD(&ss
->cfts
);
5095 /* Create the root cgroup state for this subsystem */
5096 ss
->root
= &cgrp_dfl_root
;
5097 css
= ss
->css_alloc(cgroup_css(&cgrp_dfl_root
.cgrp
, ss
));
5098 /* We don't handle early failures gracefully */
5099 BUG_ON(IS_ERR(css
));
5100 init_and_link_css(css
, ss
, &cgrp_dfl_root
.cgrp
);
5103 * Root csses are never destroyed and we can't initialize
5104 * percpu_ref during early init. Disable refcnting.
5106 css
->flags
|= CSS_NO_REF
;
5109 /* allocation can't be done safely during early init */
5112 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2, GFP_KERNEL
);
5113 BUG_ON(css
->id
< 0);
5116 /* Update the init_css_set to contain a subsys
5117 * pointer to this state - since the subsystem is
5118 * newly registered, all tasks and hence the
5119 * init_css_set is in the subsystem's root cgroup. */
5120 init_css_set
.subsys
[ss
->id
] = css
;
5122 have_fork_callback
|= (bool)ss
->fork
<< ss
->id
;
5123 have_exit_callback
|= (bool)ss
->exit
<< ss
->id
;
5124 have_canfork_callback
|= (bool)ss
->can_fork
<< ss
->id
;
5126 /* At system boot, before all subsystems have been
5127 * registered, no tasks have been forked, so we don't
5128 * need to invoke fork callbacks here. */
5129 BUG_ON(!list_empty(&init_task
.tasks
));
5131 BUG_ON(online_css(css
));
5133 mutex_unlock(&cgroup_mutex
);
5137 * cgroup_init_early - cgroup initialization at system boot
5139 * Initialize cgroups at system boot, and initialize any
5140 * subsystems that request early init.
5142 int __init
cgroup_init_early(void)
5144 static struct cgroup_sb_opts __initdata opts
;
5145 struct cgroup_subsys
*ss
;
5148 init_cgroup_root(&cgrp_dfl_root
, &opts
);
5149 cgrp_dfl_root
.cgrp
.self
.flags
|= CSS_NO_REF
;
5151 RCU_INIT_POINTER(init_task
.cgroups
, &init_css_set
);
5153 for_each_subsys(ss
, i
) {
5154 WARN(!ss
->css_alloc
|| !ss
->css_free
|| ss
->name
|| ss
->id
,
5155 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
5156 i
, cgroup_subsys_name
[i
], ss
->css_alloc
, ss
->css_free
,
5158 WARN(strlen(cgroup_subsys_name
[i
]) > MAX_CGROUP_TYPE_NAMELEN
,
5159 "cgroup_subsys_name %s too long\n", cgroup_subsys_name
[i
]);
5162 ss
->name
= cgroup_subsys_name
[i
];
5163 if (!ss
->legacy_name
)
5164 ss
->legacy_name
= cgroup_subsys_name
[i
];
5167 cgroup_init_subsys(ss
, true);
5172 static unsigned long cgroup_disable_mask __initdata
;
5175 * cgroup_init - cgroup initialization
5177 * Register cgroup filesystem and /proc file, and initialize
5178 * any subsystems that didn't request early init.
5180 int __init
cgroup_init(void)
5182 struct cgroup_subsys
*ss
;
5186 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem
));
5187 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_dfl_base_files
));
5188 BUG_ON(cgroup_init_cftypes(NULL
, cgroup_legacy_base_files
));
5190 mutex_lock(&cgroup_mutex
);
5192 /* Add init_css_set to the hash table */
5193 key
= css_set_hash(init_css_set
.subsys
);
5194 hash_add(css_set_table
, &init_css_set
.hlist
, key
);
5196 BUG_ON(cgroup_setup_root(&cgrp_dfl_root
, 0));
5198 mutex_unlock(&cgroup_mutex
);
5200 for_each_subsys(ss
, ssid
) {
5201 if (ss
->early_init
) {
5202 struct cgroup_subsys_state
*css
=
5203 init_css_set
.subsys
[ss
->id
];
5205 css
->id
= cgroup_idr_alloc(&ss
->css_idr
, css
, 1, 2,
5207 BUG_ON(css
->id
< 0);
5209 cgroup_init_subsys(ss
, false);
5212 list_add_tail(&init_css_set
.e_cset_node
[ssid
],
5213 &cgrp_dfl_root
.cgrp
.e_csets
[ssid
]);
5216 * Setting dfl_root subsys_mask needs to consider the
5217 * disabled flag and cftype registration needs kmalloc,
5218 * both of which aren't available during early_init.
5220 if (cgroup_disable_mask
& (1 << ssid
)) {
5221 static_branch_disable(cgroup_subsys_enabled_key
[ssid
]);
5222 printk(KERN_INFO
"Disabling %s control group subsystem\n",
5227 cgrp_dfl_root
.subsys_mask
|= 1 << ss
->id
;
5229 if (cgroup_legacy_files_on_dfl
&& !ss
->dfl_cftypes
)
5230 ss
->dfl_cftypes
= ss
->legacy_cftypes
;
5232 if (!ss
->dfl_cftypes
)
5233 cgrp_dfl_root_inhibit_ss_mask
|= 1 << ss
->id
;
5235 if (ss
->dfl_cftypes
== ss
->legacy_cftypes
) {
5236 WARN_ON(cgroup_add_cftypes(ss
, ss
->dfl_cftypes
));
5238 WARN_ON(cgroup_add_dfl_cftypes(ss
, ss
->dfl_cftypes
));
5239 WARN_ON(cgroup_add_legacy_cftypes(ss
, ss
->legacy_cftypes
));
5243 ss
->bind(init_css_set
.subsys
[ssid
]);
5246 err
= sysfs_create_mount_point(fs_kobj
, "cgroup");
5250 err
= register_filesystem(&cgroup_fs_type
);
5252 sysfs_remove_mount_point(fs_kobj
, "cgroup");
5256 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
5260 static int __init
cgroup_wq_init(void)
5263 * There isn't much point in executing destruction path in
5264 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5265 * Use 1 for @max_active.
5267 * We would prefer to do this in cgroup_init() above, but that
5268 * is called before init_workqueues(): so leave this until after.
5270 cgroup_destroy_wq
= alloc_workqueue("cgroup_destroy", 0, 1);
5271 BUG_ON(!cgroup_destroy_wq
);
5274 * Used to destroy pidlists and separate to serve as flush domain.
5275 * Cap @max_active to 1 too.
5277 cgroup_pidlist_destroy_wq
= alloc_workqueue("cgroup_pidlist_destroy",
5279 BUG_ON(!cgroup_pidlist_destroy_wq
);
5283 core_initcall(cgroup_wq_init
);
5286 * proc_cgroup_show()
5287 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5288 * - Used for /proc/<pid>/cgroup.
5290 int proc_cgroup_show(struct seq_file
*m
, struct pid_namespace
*ns
,
5291 struct pid
*pid
, struct task_struct
*tsk
)
5295 struct cgroup_root
*root
;
5298 buf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5302 mutex_lock(&cgroup_mutex
);
5303 down_read(&css_set_rwsem
);
5305 for_each_root(root
) {
5306 struct cgroup_subsys
*ss
;
5307 struct cgroup
*cgrp
;
5308 int ssid
, count
= 0;
5310 if (root
== &cgrp_dfl_root
&& !cgrp_dfl_root_visible
)
5313 seq_printf(m
, "%d:", root
->hierarchy_id
);
5314 if (root
!= &cgrp_dfl_root
)
5315 for_each_subsys(ss
, ssid
)
5316 if (root
->subsys_mask
& (1 << ssid
))
5317 seq_printf(m
, "%s%s", count
++ ? "," : "",
5319 if (strlen(root
->name
))
5320 seq_printf(m
, "%sname=%s", count
? "," : "",
5323 cgrp
= task_cgroup_from_root(tsk
, root
);
5324 path
= cgroup_path(cgrp
, buf
, PATH_MAX
);
5326 retval
= -ENAMETOOLONG
;
5335 up_read(&css_set_rwsem
);
5336 mutex_unlock(&cgroup_mutex
);
5342 /* Display information about each subsystem and each hierarchy */
5343 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
5345 struct cgroup_subsys
*ss
;
5348 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5350 * ideally we don't want subsystems moving around while we do this.
5351 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5352 * subsys/hierarchy state.
5354 mutex_lock(&cgroup_mutex
);
5356 for_each_subsys(ss
, i
)
5357 seq_printf(m
, "%s\t%d\t%d\t%d\n",
5358 ss
->legacy_name
, ss
->root
->hierarchy_id
,
5359 atomic_read(&ss
->root
->nr_cgrps
),
5360 cgroup_ssid_enabled(i
));
5362 mutex_unlock(&cgroup_mutex
);
5366 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
5368 return single_open(file
, proc_cgroupstats_show
, NULL
);
5371 static const struct file_operations proc_cgroupstats_operations
= {
5372 .open
= cgroupstats_open
,
5374 .llseek
= seq_lseek
,
5375 .release
= single_release
,
5378 static void **subsys_canfork_priv_p(void *ss_priv
[CGROUP_CANFORK_COUNT
], int i
)
5380 if (CGROUP_CANFORK_START
<= i
&& i
< CGROUP_CANFORK_END
)
5381 return &ss_priv
[i
- CGROUP_CANFORK_START
];
5385 static void *subsys_canfork_priv(void *ss_priv
[CGROUP_CANFORK_COUNT
], int i
)
5387 void **private = subsys_canfork_priv_p(ss_priv
, i
);
5388 return private ? *private : NULL
;
5392 * cgroup_fork - initialize cgroup related fields during copy_process()
5393 * @child: pointer to task_struct of forking parent process.
5395 * A task is associated with the init_css_set until cgroup_post_fork()
5396 * attaches it to the parent's css_set. Empty cg_list indicates that
5397 * @child isn't holding reference to its css_set.
5399 void cgroup_fork(struct task_struct
*child
)
5401 RCU_INIT_POINTER(child
->cgroups
, &init_css_set
);
5402 INIT_LIST_HEAD(&child
->cg_list
);
5406 * cgroup_can_fork - called on a new task before the process is exposed
5407 * @child: the task in question.
5409 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5410 * returns an error, the fork aborts with that error code. This allows for
5411 * a cgroup subsystem to conditionally allow or deny new forks.
5413 int cgroup_can_fork(struct task_struct
*child
,
5414 void *ss_priv
[CGROUP_CANFORK_COUNT
])
5416 struct cgroup_subsys
*ss
;
5419 for_each_subsys_which(ss
, i
, &have_canfork_callback
) {
5420 ret
= ss
->can_fork(child
, subsys_canfork_priv_p(ss_priv
, i
));
5428 for_each_subsys(ss
, j
) {
5431 if (ss
->cancel_fork
)
5432 ss
->cancel_fork(child
, subsys_canfork_priv(ss_priv
, j
));
5439 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5440 * @child: the task in question
5442 * This calls the cancel_fork() callbacks if a fork failed *after*
5443 * cgroup_can_fork() succeded.
5445 void cgroup_cancel_fork(struct task_struct
*child
,
5446 void *ss_priv
[CGROUP_CANFORK_COUNT
])
5448 struct cgroup_subsys
*ss
;
5451 for_each_subsys(ss
, i
)
5452 if (ss
->cancel_fork
)
5453 ss
->cancel_fork(child
, subsys_canfork_priv(ss_priv
, i
));
5457 * cgroup_post_fork - called on a new task after adding it to the task list
5458 * @child: the task in question
5460 * Adds the task to the list running through its css_set if necessary and
5461 * call the subsystem fork() callbacks. Has to be after the task is
5462 * visible on the task list in case we race with the first call to
5463 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5466 void cgroup_post_fork(struct task_struct
*child
,
5467 void *old_ss_priv
[CGROUP_CANFORK_COUNT
])
5469 struct cgroup_subsys
*ss
;
5473 * This may race against cgroup_enable_task_cg_lists(). As that
5474 * function sets use_task_css_set_links before grabbing
5475 * tasklist_lock and we just went through tasklist_lock to add
5476 * @child, it's guaranteed that either we see the set
5477 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5478 * @child during its iteration.
5480 * If we won the race, @child is associated with %current's
5481 * css_set. Grabbing css_set_rwsem guarantees both that the
5482 * association is stable, and, on completion of the parent's
5483 * migration, @child is visible in the source of migration or
5484 * already in the destination cgroup. This guarantee is necessary
5485 * when implementing operations which need to migrate all tasks of
5486 * a cgroup to another.
5488 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
5489 * will remain in init_css_set. This is safe because all tasks are
5490 * in the init_css_set before cg_links is enabled and there's no
5491 * operation which transfers all tasks out of init_css_set.
5493 if (use_task_css_set_links
) {
5494 struct css_set
*cset
;
5496 down_write(&css_set_rwsem
);
5497 cset
= task_css_set(current
);
5498 if (list_empty(&child
->cg_list
)) {
5500 css_set_move_task(child
, NULL
, cset
, false);
5502 up_write(&css_set_rwsem
);
5506 * Call ss->fork(). This must happen after @child is linked on
5507 * css_set; otherwise, @child might change state between ->fork()
5508 * and addition to css_set.
5510 for_each_subsys_which(ss
, i
, &have_fork_callback
)
5511 ss
->fork(child
, subsys_canfork_priv(old_ss_priv
, i
));
5515 * cgroup_exit - detach cgroup from exiting task
5516 * @tsk: pointer to task_struct of exiting process
5518 * Description: Detach cgroup from @tsk and release it.
5520 * Note that cgroups marked notify_on_release force every task in
5521 * them to take the global cgroup_mutex mutex when exiting.
5522 * This could impact scaling on very large systems. Be reluctant to
5523 * use notify_on_release cgroups where very high task exit scaling
5524 * is required on large systems.
5526 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5527 * call cgroup_exit() while the task is still competent to handle
5528 * notify_on_release(), then leave the task attached to the root cgroup in
5529 * each hierarchy for the remainder of its exit. No need to bother with
5530 * init_css_set refcnting. init_css_set never goes away and we can't race
5531 * with migration path - PF_EXITING is visible to migration path.
5533 void cgroup_exit(struct task_struct
*tsk
)
5535 struct cgroup_subsys
*ss
;
5536 struct css_set
*cset
;
5537 bool put_cset
= false;
5541 * Unlink from @tsk from its css_set. As migration path can't race
5542 * with us, we can check css_set and cg_list without synchronization.
5544 cset
= task_css_set(tsk
);
5546 if (!list_empty(&tsk
->cg_list
)) {
5547 down_write(&css_set_rwsem
);
5548 css_set_move_task(tsk
, cset
, NULL
, false);
5549 up_write(&css_set_rwsem
);
5553 /* Reassign the task to the init_css_set. */
5554 RCU_INIT_POINTER(tsk
->cgroups
, &init_css_set
);
5556 /* see cgroup_post_fork() for details */
5557 for_each_subsys_which(ss
, i
, &have_exit_callback
) {
5558 struct cgroup_subsys_state
*old_css
= cset
->subsys
[i
];
5559 struct cgroup_subsys_state
*css
= task_css(tsk
, i
);
5561 ss
->exit(css
, old_css
, tsk
);
5568 static void check_for_release(struct cgroup
*cgrp
)
5570 if (notify_on_release(cgrp
) && !cgroup_is_populated(cgrp
) &&
5571 !css_has_online_children(&cgrp
->self
) && !cgroup_is_dead(cgrp
))
5572 schedule_work(&cgrp
->release_agent_work
);
5576 * Notify userspace when a cgroup is released, by running the
5577 * configured release agent with the name of the cgroup (path
5578 * relative to the root of cgroup file system) as the argument.
5580 * Most likely, this user command will try to rmdir this cgroup.
5582 * This races with the possibility that some other task will be
5583 * attached to this cgroup before it is removed, or that some other
5584 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5585 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5586 * unused, and this cgroup will be reprieved from its death sentence,
5587 * to continue to serve a useful existence. Next time it's released,
5588 * we will get notified again, if it still has 'notify_on_release' set.
5590 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5591 * means only wait until the task is successfully execve()'d. The
5592 * separate release agent task is forked by call_usermodehelper(),
5593 * then control in this thread returns here, without waiting for the
5594 * release agent task. We don't bother to wait because the caller of
5595 * this routine has no use for the exit status of the release agent
5596 * task, so no sense holding our caller up for that.
5598 static void cgroup_release_agent(struct work_struct
*work
)
5600 struct cgroup
*cgrp
=
5601 container_of(work
, struct cgroup
, release_agent_work
);
5602 char *pathbuf
= NULL
, *agentbuf
= NULL
, *path
;
5603 char *argv
[3], *envp
[3];
5605 mutex_lock(&cgroup_mutex
);
5607 pathbuf
= kmalloc(PATH_MAX
, GFP_KERNEL
);
5608 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5609 if (!pathbuf
|| !agentbuf
)
5612 path
= cgroup_path(cgrp
, pathbuf
, PATH_MAX
);
5620 /* minimal command environment */
5622 envp
[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5625 mutex_unlock(&cgroup_mutex
);
5626 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5629 mutex_unlock(&cgroup_mutex
);
5635 static int __init
cgroup_disable(char *str
)
5637 struct cgroup_subsys
*ss
;
5641 while ((token
= strsep(&str
, ",")) != NULL
) {
5645 for_each_subsys(ss
, i
) {
5646 if (strcmp(token
, ss
->name
) &&
5647 strcmp(token
, ss
->legacy_name
))
5649 cgroup_disable_mask
|= 1 << i
;
5654 __setup("cgroup_disable=", cgroup_disable
);
5656 static int __init
cgroup_set_legacy_files_on_dfl(char *str
)
5658 printk("cgroup: using legacy files on the default hierarchy\n");
5659 cgroup_legacy_files_on_dfl
= true;
5662 __setup("cgroup__DEVEL__legacy_files_on_dfl", cgroup_set_legacy_files_on_dfl
);
5665 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5666 * @dentry: directory dentry of interest
5667 * @ss: subsystem of interest
5669 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5670 * to get the corresponding css and return it. If such css doesn't exist
5671 * or can't be pinned, an ERR_PTR value is returned.
5673 struct cgroup_subsys_state
*css_tryget_online_from_dir(struct dentry
*dentry
,
5674 struct cgroup_subsys
*ss
)
5676 struct kernfs_node
*kn
= kernfs_node_from_dentry(dentry
);
5677 struct cgroup_subsys_state
*css
= NULL
;
5678 struct cgroup
*cgrp
;
5680 /* is @dentry a cgroup dir? */
5681 if (dentry
->d_sb
->s_type
!= &cgroup_fs_type
|| !kn
||
5682 kernfs_type(kn
) != KERNFS_DIR
)
5683 return ERR_PTR(-EBADF
);
5688 * This path doesn't originate from kernfs and @kn could already
5689 * have been or be removed at any point. @kn->priv is RCU
5690 * protected for this access. See css_release_work_fn() for details.
5692 cgrp
= rcu_dereference(kn
->priv
);
5694 css
= cgroup_css(cgrp
, ss
);
5696 if (!css
|| !css_tryget_online(css
))
5697 css
= ERR_PTR(-ENOENT
);
5704 * css_from_id - lookup css by id
5705 * @id: the cgroup id
5706 * @ss: cgroup subsys to be looked into
5708 * Returns the css if there's valid one with @id, otherwise returns NULL.
5709 * Should be called under rcu_read_lock().
5711 struct cgroup_subsys_state
*css_from_id(int id
, struct cgroup_subsys
*ss
)
5713 WARN_ON_ONCE(!rcu_read_lock_held());
5714 return id
> 0 ? idr_find(&ss
->css_idr
, id
) : NULL
;
5717 #ifdef CONFIG_CGROUP_DEBUG
5718 static struct cgroup_subsys_state
*
5719 debug_css_alloc(struct cgroup_subsys_state
*parent_css
)
5721 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5724 return ERR_PTR(-ENOMEM
);
5729 static void debug_css_free(struct cgroup_subsys_state
*css
)
5734 static u64
debug_taskcount_read(struct cgroup_subsys_state
*css
,
5737 return cgroup_task_count(css
->cgroup
);
5740 static u64
current_css_set_read(struct cgroup_subsys_state
*css
,
5743 return (u64
)(unsigned long)current
->cgroups
;
5746 static u64
current_css_set_refcount_read(struct cgroup_subsys_state
*css
,
5752 count
= atomic_read(&task_css_set(current
)->refcount
);
5757 static int current_css_set_cg_links_read(struct seq_file
*seq
, void *v
)
5759 struct cgrp_cset_link
*link
;
5760 struct css_set
*cset
;
5763 name_buf
= kmalloc(NAME_MAX
+ 1, GFP_KERNEL
);
5767 down_read(&css_set_rwsem
);
5769 cset
= rcu_dereference(current
->cgroups
);
5770 list_for_each_entry(link
, &cset
->cgrp_links
, cgrp_link
) {
5771 struct cgroup
*c
= link
->cgrp
;
5773 cgroup_name(c
, name_buf
, NAME_MAX
+ 1);
5774 seq_printf(seq
, "Root %d group %s\n",
5775 c
->root
->hierarchy_id
, name_buf
);
5778 up_read(&css_set_rwsem
);
5783 #define MAX_TASKS_SHOWN_PER_CSS 25
5784 static int cgroup_css_links_read(struct seq_file
*seq
, void *v
)
5786 struct cgroup_subsys_state
*css
= seq_css(seq
);
5787 struct cgrp_cset_link
*link
;
5789 down_read(&css_set_rwsem
);
5790 list_for_each_entry(link
, &css
->cgroup
->cset_links
, cset_link
) {
5791 struct css_set
*cset
= link
->cset
;
5792 struct task_struct
*task
;
5795 seq_printf(seq
, "css_set %p\n", cset
);
5797 list_for_each_entry(task
, &cset
->tasks
, cg_list
) {
5798 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5800 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5803 list_for_each_entry(task
, &cset
->mg_tasks
, cg_list
) {
5804 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
)
5806 seq_printf(seq
, " task %d\n", task_pid_vnr(task
));
5810 seq_puts(seq
, " ...\n");
5812 up_read(&css_set_rwsem
);
5816 static u64
releasable_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
5818 return (!cgroup_is_populated(css
->cgroup
) &&
5819 !css_has_online_children(&css
->cgroup
->self
));
5822 static struct cftype debug_files
[] = {
5824 .name
= "taskcount",
5825 .read_u64
= debug_taskcount_read
,
5829 .name
= "current_css_set",
5830 .read_u64
= current_css_set_read
,
5834 .name
= "current_css_set_refcount",
5835 .read_u64
= current_css_set_refcount_read
,
5839 .name
= "current_css_set_cg_links",
5840 .seq_show
= current_css_set_cg_links_read
,
5844 .name
= "cgroup_css_links",
5845 .seq_show
= cgroup_css_links_read
,
5849 .name
= "releasable",
5850 .read_u64
= releasable_read
,
5856 struct cgroup_subsys debug_cgrp_subsys
= {
5857 .css_alloc
= debug_css_alloc
,
5858 .css_free
= debug_css_free
,
5859 .legacy_cftypes
= debug_files
,
5861 #endif /* CONFIG_CGROUP_DEBUG */