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 "cgroup-internal.h"
33 #include <linux/cred.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/magic.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/sched/task.h>
45 #include <linux/slab.h>
46 #include <linux/spinlock.h>
47 #include <linux/percpu-rwsem.h>
48 #include <linux/string.h>
49 #include <linux/hashtable.h>
50 #include <linux/idr.h>
51 #include <linux/kthread.h>
52 #include <linux/atomic.h>
53 #include <linux/cpuset.h>
54 #include <linux/proc_ns.h>
55 #include <linux/nsproxy.h>
56 #include <linux/file.h>
57 #include <linux/fs_parser.h>
58 #include <linux/sched/cputime.h>
59 #include <linux/psi.h>
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/cgroup.h>
65 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
67 /* let's not notify more than 100 times per second */
68 #define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100)
71 * cgroup_mutex is the master lock. Any modification to cgroup or its
72 * hierarchy must be performed while holding it.
74 * css_set_lock protects task->cgroups pointer, the list of css_set
75 * objects, and the chain of tasks off each css_set.
77 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
78 * cgroup.h can use them for lockdep annotations.
80 DEFINE_MUTEX(cgroup_mutex);
81 DEFINE_SPINLOCK(css_set_lock);
83 #ifdef CONFIG_PROVE_RCU
84 EXPORT_SYMBOL_GPL(cgroup_mutex);
85 EXPORT_SYMBOL_GPL(css_set_lock);
88 DEFINE_SPINLOCK(trace_cgroup_path_lock);
89 char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
90 bool cgroup_debug __read_mostly;
93 * Protects cgroup_idr and css_idr so that IDs can be released without
94 * grabbing cgroup_mutex.
96 static DEFINE_SPINLOCK(cgroup_idr_lock);
99 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
100 * against file removal/re-creation across css hiding.
102 static DEFINE_SPINLOCK(cgroup_file_kn_lock);
104 DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem);
106 #define cgroup_assert_mutex_or_rcu_locked() \
107 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
108 !lockdep_is_held(&cgroup_mutex), \
109 "cgroup_mutex or RCU read lock required");
112 * cgroup destruction makes heavy use of work items and there can be a lot
113 * of concurrent destructions. Use a separate workqueue so that cgroup
114 * destruction work items don't end up filling up max_active of system_wq
115 * which may lead to deadlock.
117 static struct workqueue_struct *cgroup_destroy_wq;
119 /* generate an array of cgroup subsystem pointers */
120 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
121 struct cgroup_subsys *cgroup_subsys[] = {
122 #include <linux/cgroup_subsys.h>
126 /* array of cgroup subsystem names */
127 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
128 static const char *cgroup_subsys_name[] = {
129 #include <linux/cgroup_subsys.h>
133 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
135 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
136 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
137 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
138 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
139 #include <linux/cgroup_subsys.h>
142 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
143 static struct static_key_true *cgroup_subsys_enabled_key[] = {
144 #include <linux/cgroup_subsys.h>
148 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
149 static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
150 #include <linux/cgroup_subsys.h>
154 static DEFINE_PER_CPU(struct cgroup_rstat_cpu, cgrp_dfl_root_rstat_cpu);
156 /* the default hierarchy */
157 struct cgroup_root cgrp_dfl_root = { .cgrp.rstat_cpu = &cgrp_dfl_root_rstat_cpu };
158 EXPORT_SYMBOL_GPL(cgrp_dfl_root);
161 * The default hierarchy always exists but is hidden until mounted for the
162 * first time. This is for backward compatibility.
164 static bool cgrp_dfl_visible;
166 /* some controllers are not supported in the default hierarchy */
167 static u16 cgrp_dfl_inhibit_ss_mask;
169 /* some controllers are implicitly enabled on the default hierarchy */
170 static u16 cgrp_dfl_implicit_ss_mask;
172 /* some controllers can be threaded on the default hierarchy */
173 static u16 cgrp_dfl_threaded_ss_mask;
175 /* The list of hierarchy roots */
176 LIST_HEAD(cgroup_roots);
177 static int cgroup_root_count;
179 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
180 static DEFINE_IDR(cgroup_hierarchy_idr);
183 * Assign a monotonically increasing serial number to csses. It guarantees
184 * cgroups with bigger numbers are newer than those with smaller numbers.
185 * Also, as csses are always appended to the parent's ->children list, it
186 * guarantees that sibling csses are always sorted in the ascending serial
187 * number order on the list. Protected by cgroup_mutex.
189 static u64 css_serial_nr_next = 1;
192 * These bitmasks identify subsystems with specific features to avoid
193 * having to do iterative checks repeatedly.
195 static u16 have_fork_callback __read_mostly;
196 static u16 have_exit_callback __read_mostly;
197 static u16 have_release_callback __read_mostly;
198 static u16 have_canfork_callback __read_mostly;
200 /* cgroup namespace for init task */
201 struct cgroup_namespace init_cgroup_ns = {
202 .ns.count = REFCOUNT_INIT(2),
203 .user_ns = &init_user_ns,
204 .ns.ops = &cgroupns_operations,
205 .ns.inum = PROC_CGROUP_INIT_INO,
206 .root_cset = &init_css_set,
209 static struct file_system_type cgroup2_fs_type;
210 static struct cftype cgroup_base_files[];
212 static int cgroup_apply_control(struct cgroup *cgrp);
213 static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
214 static void css_task_iter_skip(struct css_task_iter *it,
215 struct task_struct *task);
216 static int cgroup_destroy_locked(struct cgroup *cgrp);
217 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
218 struct cgroup_subsys *ss);
219 static void css_release(struct percpu_ref *ref);
220 static void kill_css(struct cgroup_subsys_state *css);
221 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
222 struct cgroup *cgrp, struct cftype cfts[],
226 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
227 * @ssid: subsys ID of interest
229 * cgroup_subsys_enabled() can only be used with literal subsys names which
230 * is fine for individual subsystems but unsuitable for cgroup core. This
231 * is slower static_key_enabled() based test indexed by @ssid.
233 bool cgroup_ssid_enabled(int ssid)
235 if (CGROUP_SUBSYS_COUNT == 0)
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 differently depending on the
250 * List of changed behaviors:
252 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
253 * and "name" are disallowed.
255 * - When mounting an existing superblock, mount options should match.
257 * - Remount is disallowed.
259 * - rename(2) is disallowed.
261 * - "tasks" is removed. Everything should be at process granularity. Use
262 * "cgroup.procs" instead.
264 * - "cgroup.procs" is not sorted. pids will be unique unless they got
265 * recycled in-between reads.
267 * - "release_agent" and "notify_on_release" are removed. Replacement
268 * notification mechanism will be implemented.
270 * - "cgroup.clone_children" is removed.
272 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
273 * and its descendants contain no task; otherwise, 1. The file also
274 * generates kernfs notification which can be monitored through poll and
275 * [di]notify when the value of the file changes.
277 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
278 * take masks of ancestors with non-empty cpus/mems, instead of being
279 * moved to an ancestor.
281 * - cpuset: a task can be moved into an empty cpuset, and again it takes
282 * masks of ancestors.
284 * - blkcg: blk-throttle becomes properly hierarchical.
286 * - debug: disallowed on the default hierarchy.
288 bool cgroup_on_dfl(const struct cgroup *cgrp)
290 return cgrp->root == &cgrp_dfl_root;
293 /* IDR wrappers which synchronize using cgroup_idr_lock */
294 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
299 idr_preload(gfp_mask);
300 spin_lock_bh(&cgroup_idr_lock);
301 ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
302 spin_unlock_bh(&cgroup_idr_lock);
307 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
311 spin_lock_bh(&cgroup_idr_lock);
312 ret = idr_replace(idr, ptr, id);
313 spin_unlock_bh(&cgroup_idr_lock);
317 static void cgroup_idr_remove(struct idr *idr, int id)
319 spin_lock_bh(&cgroup_idr_lock);
321 spin_unlock_bh(&cgroup_idr_lock);
324 static bool cgroup_has_tasks(struct cgroup *cgrp)
326 return cgrp->nr_populated_csets;
329 bool cgroup_is_threaded(struct cgroup *cgrp)
331 return cgrp->dom_cgrp != cgrp;
334 /* can @cgrp host both domain and threaded children? */
335 static bool cgroup_is_mixable(struct cgroup *cgrp)
338 * Root isn't under domain level resource control exempting it from
339 * the no-internal-process constraint, so it can serve as a thread
340 * root and a parent of resource domains at the same time.
342 return !cgroup_parent(cgrp);
345 /* can @cgrp become a thread root? Should always be true for a thread root */
346 static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
348 /* mixables don't care */
349 if (cgroup_is_mixable(cgrp))
352 /* domain roots can't be nested under threaded */
353 if (cgroup_is_threaded(cgrp))
356 /* can only have either domain or threaded children */
357 if (cgrp->nr_populated_domain_children)
360 /* and no domain controllers can be enabled */
361 if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
367 /* is @cgrp root of a threaded subtree? */
368 bool cgroup_is_thread_root(struct cgroup *cgrp)
370 /* thread root should be a domain */
371 if (cgroup_is_threaded(cgrp))
374 /* a domain w/ threaded children is a thread root */
375 if (cgrp->nr_threaded_children)
379 * A domain which has tasks and explicit threaded controllers
380 * enabled is a thread root.
382 if (cgroup_has_tasks(cgrp) &&
383 (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
389 /* a domain which isn't connected to the root w/o brekage can't be used */
390 static bool cgroup_is_valid_domain(struct cgroup *cgrp)
392 /* the cgroup itself can be a thread root */
393 if (cgroup_is_threaded(cgrp))
396 /* but the ancestors can't be unless mixable */
397 while ((cgrp = cgroup_parent(cgrp))) {
398 if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
400 if (cgroup_is_threaded(cgrp))
407 /* subsystems visibly enabled on a cgroup */
408 static u16 cgroup_control(struct cgroup *cgrp)
410 struct cgroup *parent = cgroup_parent(cgrp);
411 u16 root_ss_mask = cgrp->root->subsys_mask;
414 u16 ss_mask = parent->subtree_control;
416 /* threaded cgroups can only have threaded controllers */
417 if (cgroup_is_threaded(cgrp))
418 ss_mask &= cgrp_dfl_threaded_ss_mask;
422 if (cgroup_on_dfl(cgrp))
423 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
424 cgrp_dfl_implicit_ss_mask);
428 /* subsystems enabled on a cgroup */
429 static u16 cgroup_ss_mask(struct cgroup *cgrp)
431 struct cgroup *parent = cgroup_parent(cgrp);
434 u16 ss_mask = parent->subtree_ss_mask;
436 /* threaded cgroups can only have threaded controllers */
437 if (cgroup_is_threaded(cgrp))
438 ss_mask &= cgrp_dfl_threaded_ss_mask;
442 return cgrp->root->subsys_mask;
446 * cgroup_css - obtain a cgroup's css for the specified subsystem
447 * @cgrp: the cgroup of interest
448 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
450 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
451 * function must be called either under cgroup_mutex or rcu_read_lock() and
452 * the caller is responsible for pinning the returned css if it wants to
453 * keep accessing it outside the said locks. This function may return
454 * %NULL if @cgrp doesn't have @subsys_id enabled.
456 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
457 struct cgroup_subsys *ss)
460 return rcu_dereference_check(cgrp->subsys[ss->id],
461 lockdep_is_held(&cgroup_mutex));
467 * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
468 * @cgrp: the cgroup of interest
469 * @ss: the subsystem of interest
471 * Find and get @cgrp's css associated with @ss. If the css doesn't exist
472 * or is offline, %NULL is returned.
474 static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp,
475 struct cgroup_subsys *ss)
477 struct cgroup_subsys_state *css;
480 css = cgroup_css(cgrp, ss);
481 if (css && !css_tryget_online(css))
489 * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
490 * @cgrp: the cgroup of interest
491 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
493 * Similar to cgroup_css() but returns the effective css, which is defined
494 * as the matching css of the nearest ancestor including self which has @ss
495 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
496 * function is guaranteed to return non-NULL css.
498 static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp,
499 struct cgroup_subsys *ss)
501 lockdep_assert_held(&cgroup_mutex);
507 * This function is used while updating css associations and thus
508 * can't test the csses directly. Test ss_mask.
510 while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
511 cgrp = cgroup_parent(cgrp);
516 return cgroup_css(cgrp, ss);
520 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
521 * @cgrp: the cgroup of interest
522 * @ss: the subsystem of interest
524 * Find and get the effective css of @cgrp for @ss. The effective css is
525 * defined as the matching css of the nearest ancestor including self which
526 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
527 * the root css is returned, so this function always returns a valid css.
529 * The returned css is not guaranteed to be online, and therefore it is the
530 * callers responsibility to try get a reference for it.
532 struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
533 struct cgroup_subsys *ss)
535 struct cgroup_subsys_state *css;
538 css = cgroup_css(cgrp, ss);
542 cgrp = cgroup_parent(cgrp);
545 return init_css_set.subsys[ss->id];
549 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
550 * @cgrp: the cgroup of interest
551 * @ss: the subsystem of interest
553 * Find and get the effective css of @cgrp for @ss. The effective css is
554 * defined as the matching css of the nearest ancestor including self which
555 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
556 * the root css is returned, so this function always returns a valid css.
557 * The returned css must be put using css_put().
559 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
560 struct cgroup_subsys *ss)
562 struct cgroup_subsys_state *css;
567 css = cgroup_css(cgrp, ss);
569 if (css && css_tryget_online(css))
571 cgrp = cgroup_parent(cgrp);
574 css = init_css_set.subsys[ss->id];
580 EXPORT_SYMBOL_GPL(cgroup_get_e_css);
582 static void cgroup_get_live(struct cgroup *cgrp)
584 WARN_ON_ONCE(cgroup_is_dead(cgrp));
585 css_get(&cgrp->self);
589 * __cgroup_task_count - count the number of tasks in a cgroup. The caller
590 * is responsible for taking the css_set_lock.
591 * @cgrp: the cgroup in question
593 int __cgroup_task_count(const struct cgroup *cgrp)
596 struct cgrp_cset_link *link;
598 lockdep_assert_held(&css_set_lock);
600 list_for_each_entry(link, &cgrp->cset_links, cset_link)
601 count += link->cset->nr_tasks;
607 * cgroup_task_count - count the number of tasks in a cgroup.
608 * @cgrp: the cgroup in question
610 int cgroup_task_count(const struct cgroup *cgrp)
614 spin_lock_irq(&css_set_lock);
615 count = __cgroup_task_count(cgrp);
616 spin_unlock_irq(&css_set_lock);
621 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
623 struct cgroup *cgrp = of->kn->parent->priv;
624 struct cftype *cft = of_cft(of);
627 * This is open and unprotected implementation of cgroup_css().
628 * seq_css() is only called from a kernfs file operation which has
629 * an active reference on the file. Because all the subsystem
630 * files are drained before a css is disassociated with a cgroup,
631 * the matching css from the cgroup's subsys table is guaranteed to
632 * be and stay valid until the enclosing operation is complete.
635 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
639 EXPORT_SYMBOL_GPL(of_css);
642 * for_each_css - iterate all css's of a cgroup
643 * @css: the iteration cursor
644 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
645 * @cgrp: the target cgroup to iterate css's of
647 * Should be called under cgroup_[tree_]mutex.
649 #define for_each_css(css, ssid, cgrp) \
650 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
651 if (!((css) = rcu_dereference_check( \
652 (cgrp)->subsys[(ssid)], \
653 lockdep_is_held(&cgroup_mutex)))) { } \
657 * for_each_e_css - iterate all effective css's of a cgroup
658 * @css: the iteration cursor
659 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
660 * @cgrp: the target cgroup to iterate css's of
662 * Should be called under cgroup_[tree_]mutex.
664 #define for_each_e_css(css, ssid, cgrp) \
665 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
666 if (!((css) = cgroup_e_css_by_mask(cgrp, \
667 cgroup_subsys[(ssid)]))) \
672 * do_each_subsys_mask - filter for_each_subsys with a bitmask
673 * @ss: the iteration cursor
674 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
675 * @ss_mask: the bitmask
677 * The block will only run for cases where the ssid-th bit (1 << ssid) of
680 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
681 unsigned long __ss_mask = (ss_mask); \
682 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
686 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
687 (ss) = cgroup_subsys[ssid]; \
690 #define while_each_subsys_mask() \
695 /* iterate over child cgrps, lock should be held throughout iteration */
696 #define cgroup_for_each_live_child(child, cgrp) \
697 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
698 if (({ lockdep_assert_held(&cgroup_mutex); \
699 cgroup_is_dead(child); })) \
703 /* walk live descendants in pre order */
704 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
705 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
706 if (({ lockdep_assert_held(&cgroup_mutex); \
707 (dsct) = (d_css)->cgroup; \
708 cgroup_is_dead(dsct); })) \
712 /* walk live descendants in postorder */
713 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
714 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
715 if (({ lockdep_assert_held(&cgroup_mutex); \
716 (dsct) = (d_css)->cgroup; \
717 cgroup_is_dead(dsct); })) \
722 * The default css_set - used by init and its children prior to any
723 * hierarchies being mounted. It contains a pointer to the root state
724 * for each subsystem. Also used to anchor the list of css_sets. Not
725 * reference-counted, to improve performance when child cgroups
726 * haven't been created.
728 struct css_set init_css_set = {
729 .refcount = REFCOUNT_INIT(1),
730 .dom_cset = &init_css_set,
731 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
732 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
733 .dying_tasks = LIST_HEAD_INIT(init_css_set.dying_tasks),
734 .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
735 .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets),
736 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
737 .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
738 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
741 * The following field is re-initialized when this cset gets linked
742 * in cgroup_init(). However, let's initialize the field
743 * statically too so that the default cgroup can be accessed safely
746 .dfl_cgrp = &cgrp_dfl_root.cgrp,
749 static int css_set_count = 1; /* 1 for init_css_set */
751 static bool css_set_threaded(struct css_set *cset)
753 return cset->dom_cset != cset;
757 * css_set_populated - does a css_set contain any tasks?
758 * @cset: target css_set
760 * css_set_populated() should be the same as !!cset->nr_tasks at steady
761 * state. However, css_set_populated() can be called while a task is being
762 * added to or removed from the linked list before the nr_tasks is
763 * properly updated. Hence, we can't just look at ->nr_tasks here.
765 static bool css_set_populated(struct css_set *cset)
767 lockdep_assert_held(&css_set_lock);
769 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
773 * cgroup_update_populated - update the populated count of a cgroup
774 * @cgrp: the target cgroup
775 * @populated: inc or dec populated count
777 * One of the css_sets associated with @cgrp is either getting its first
778 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
779 * count is propagated towards root so that a given cgroup's
780 * nr_populated_children is zero iff none of its descendants contain any
783 * @cgrp's interface file "cgroup.populated" is zero if both
784 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
785 * 1 otherwise. When the sum changes from or to zero, userland is notified
786 * that the content of the interface file has changed. This can be used to
787 * detect when @cgrp and its descendants become populated or empty.
789 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
791 struct cgroup *child = NULL;
792 int adj = populated ? 1 : -1;
794 lockdep_assert_held(&css_set_lock);
797 bool was_populated = cgroup_is_populated(cgrp);
800 cgrp->nr_populated_csets += adj;
802 if (cgroup_is_threaded(child))
803 cgrp->nr_populated_threaded_children += adj;
805 cgrp->nr_populated_domain_children += adj;
808 if (was_populated == cgroup_is_populated(cgrp))
811 cgroup1_check_for_release(cgrp);
812 TRACE_CGROUP_PATH(notify_populated, cgrp,
813 cgroup_is_populated(cgrp));
814 cgroup_file_notify(&cgrp->events_file);
817 cgrp = cgroup_parent(cgrp);
822 * css_set_update_populated - update populated state of a css_set
823 * @cset: target css_set
824 * @populated: whether @cset is populated or depopulated
826 * @cset is either getting the first task or losing the last. Update the
827 * populated counters of all associated cgroups accordingly.
829 static void css_set_update_populated(struct css_set *cset, bool populated)
831 struct cgrp_cset_link *link;
833 lockdep_assert_held(&css_set_lock);
835 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
836 cgroup_update_populated(link->cgrp, populated);
840 * @task is leaving, advance task iterators which are pointing to it so
841 * that they can resume at the next position. Advancing an iterator might
842 * remove it from the list, use safe walk. See css_task_iter_skip() for
845 static void css_set_skip_task_iters(struct css_set *cset,
846 struct task_struct *task)
848 struct css_task_iter *it, *pos;
850 list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node)
851 css_task_iter_skip(it, task);
855 * css_set_move_task - move a task from one css_set to another
856 * @task: task being moved
857 * @from_cset: css_set @task currently belongs to (may be NULL)
858 * @to_cset: new css_set @task is being moved to (may be NULL)
859 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
861 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
862 * css_set, @from_cset can be NULL. If @task is being disassociated
863 * instead of moved, @to_cset can be NULL.
865 * This function automatically handles populated counter updates and
866 * css_task_iter adjustments but the caller is responsible for managing
867 * @from_cset and @to_cset's reference counts.
869 static void css_set_move_task(struct task_struct *task,
870 struct css_set *from_cset, struct css_set *to_cset,
873 lockdep_assert_held(&css_set_lock);
875 if (to_cset && !css_set_populated(to_cset))
876 css_set_update_populated(to_cset, true);
879 WARN_ON_ONCE(list_empty(&task->cg_list));
881 css_set_skip_task_iters(from_cset, task);
882 list_del_init(&task->cg_list);
883 if (!css_set_populated(from_cset))
884 css_set_update_populated(from_cset, false);
886 WARN_ON_ONCE(!list_empty(&task->cg_list));
891 * We are synchronized through cgroup_threadgroup_rwsem
892 * against PF_EXITING setting such that we can't race
893 * against cgroup_exit()/cgroup_free() dropping the css_set.
895 WARN_ON_ONCE(task->flags & PF_EXITING);
897 cgroup_move_task(task, to_cset);
898 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
904 * hash table for cgroup groups. This improves the performance to find
905 * an existing css_set. This hash doesn't (currently) take into
906 * account cgroups in empty hierarchies.
908 #define CSS_SET_HASH_BITS 7
909 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
911 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
913 unsigned long key = 0UL;
914 struct cgroup_subsys *ss;
917 for_each_subsys(ss, i)
918 key += (unsigned long)css[i];
919 key = (key >> 16) ^ key;
924 void put_css_set_locked(struct css_set *cset)
926 struct cgrp_cset_link *link, *tmp_link;
927 struct cgroup_subsys *ss;
930 lockdep_assert_held(&css_set_lock);
932 if (!refcount_dec_and_test(&cset->refcount))
935 WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
937 /* This css_set is dead. Unlink it and release cgroup and css refs */
938 for_each_subsys(ss, ssid) {
939 list_del(&cset->e_cset_node[ssid]);
940 css_put(cset->subsys[ssid]);
942 hash_del(&cset->hlist);
945 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
946 list_del(&link->cset_link);
947 list_del(&link->cgrp_link);
948 if (cgroup_parent(link->cgrp))
949 cgroup_put(link->cgrp);
953 if (css_set_threaded(cset)) {
954 list_del(&cset->threaded_csets_node);
955 put_css_set_locked(cset->dom_cset);
958 kfree_rcu(cset, rcu_head);
962 * compare_css_sets - helper function for find_existing_css_set().
963 * @cset: candidate css_set being tested
964 * @old_cset: existing css_set for a task
965 * @new_cgrp: cgroup that's being entered by the task
966 * @template: desired set of css pointers in css_set (pre-calculated)
968 * Returns true if "cset" matches "old_cset" except for the hierarchy
969 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
971 static bool compare_css_sets(struct css_set *cset,
972 struct css_set *old_cset,
973 struct cgroup *new_cgrp,
974 struct cgroup_subsys_state *template[])
976 struct cgroup *new_dfl_cgrp;
977 struct list_head *l1, *l2;
980 * On the default hierarchy, there can be csets which are
981 * associated with the same set of cgroups but different csses.
982 * Let's first ensure that csses match.
984 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
988 /* @cset's domain should match the default cgroup's */
989 if (cgroup_on_dfl(new_cgrp))
990 new_dfl_cgrp = new_cgrp;
992 new_dfl_cgrp = old_cset->dfl_cgrp;
994 if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
998 * Compare cgroup pointers in order to distinguish between
999 * different cgroups in hierarchies. As different cgroups may
1000 * share the same effective css, this comparison is always
1003 l1 = &cset->cgrp_links;
1004 l2 = &old_cset->cgrp_links;
1006 struct cgrp_cset_link *link1, *link2;
1007 struct cgroup *cgrp1, *cgrp2;
1011 /* See if we reached the end - both lists are equal length. */
1012 if (l1 == &cset->cgrp_links) {
1013 BUG_ON(l2 != &old_cset->cgrp_links);
1016 BUG_ON(l2 == &old_cset->cgrp_links);
1018 /* Locate the cgroups associated with these links. */
1019 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
1020 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
1021 cgrp1 = link1->cgrp;
1022 cgrp2 = link2->cgrp;
1023 /* Hierarchies should be linked in the same order. */
1024 BUG_ON(cgrp1->root != cgrp2->root);
1027 * If this hierarchy is the hierarchy of the cgroup
1028 * that's changing, then we need to check that this
1029 * css_set points to the new cgroup; if it's any other
1030 * hierarchy, then this css_set should point to the
1031 * same cgroup as the old css_set.
1033 if (cgrp1->root == new_cgrp->root) {
1034 if (cgrp1 != new_cgrp)
1045 * find_existing_css_set - init css array and find the matching css_set
1046 * @old_cset: the css_set that we're using before the cgroup transition
1047 * @cgrp: the cgroup that we're moving into
1048 * @template: out param for the new set of csses, should be clear on entry
1050 static struct css_set *find_existing_css_set(struct css_set *old_cset,
1051 struct cgroup *cgrp,
1052 struct cgroup_subsys_state *template[])
1054 struct cgroup_root *root = cgrp->root;
1055 struct cgroup_subsys *ss;
1056 struct css_set *cset;
1061 * Build the set of subsystem state objects that we want to see in the
1062 * new css_set. While subsystems can change globally, the entries here
1063 * won't change, so no need for locking.
1065 for_each_subsys(ss, i) {
1066 if (root->subsys_mask & (1UL << i)) {
1068 * @ss is in this hierarchy, so we want the
1069 * effective css from @cgrp.
1071 template[i] = cgroup_e_css_by_mask(cgrp, ss);
1074 * @ss is not in this hierarchy, so we don't want
1075 * to change the css.
1077 template[i] = old_cset->subsys[i];
1081 key = css_set_hash(template);
1082 hash_for_each_possible(css_set_table, cset, hlist, key) {
1083 if (!compare_css_sets(cset, old_cset, cgrp, template))
1086 /* This css_set matches what we need */
1090 /* No existing cgroup group matched */
1094 static void free_cgrp_cset_links(struct list_head *links_to_free)
1096 struct cgrp_cset_link *link, *tmp_link;
1098 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
1099 list_del(&link->cset_link);
1105 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1106 * @count: the number of links to allocate
1107 * @tmp_links: list_head the allocated links are put on
1109 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1110 * through ->cset_link. Returns 0 on success or -errno.
1112 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1114 struct cgrp_cset_link *link;
1117 INIT_LIST_HEAD(tmp_links);
1119 for (i = 0; i < count; i++) {
1120 link = kzalloc(sizeof(*link), GFP_KERNEL);
1122 free_cgrp_cset_links(tmp_links);
1125 list_add(&link->cset_link, tmp_links);
1131 * link_css_set - a helper function to link a css_set to a cgroup
1132 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1133 * @cset: the css_set to be linked
1134 * @cgrp: the destination cgroup
1136 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1137 struct cgroup *cgrp)
1139 struct cgrp_cset_link *link;
1141 BUG_ON(list_empty(tmp_links));
1143 if (cgroup_on_dfl(cgrp))
1144 cset->dfl_cgrp = cgrp;
1146 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1151 * Always add links to the tail of the lists so that the lists are
1152 * in chronological order.
1154 list_move_tail(&link->cset_link, &cgrp->cset_links);
1155 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1157 if (cgroup_parent(cgrp))
1158 cgroup_get_live(cgrp);
1162 * find_css_set - return a new css_set with one cgroup updated
1163 * @old_cset: the baseline css_set
1164 * @cgrp: the cgroup to be updated
1166 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1167 * substituted into the appropriate hierarchy.
1169 static struct css_set *find_css_set(struct css_set *old_cset,
1170 struct cgroup *cgrp)
1172 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1173 struct css_set *cset;
1174 struct list_head tmp_links;
1175 struct cgrp_cset_link *link;
1176 struct cgroup_subsys *ss;
1180 lockdep_assert_held(&cgroup_mutex);
1182 /* First see if we already have a cgroup group that matches
1183 * the desired set */
1184 spin_lock_irq(&css_set_lock);
1185 cset = find_existing_css_set(old_cset, cgrp, template);
1188 spin_unlock_irq(&css_set_lock);
1193 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1197 /* Allocate all the cgrp_cset_link objects that we'll need */
1198 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1203 refcount_set(&cset->refcount, 1);
1204 cset->dom_cset = cset;
1205 INIT_LIST_HEAD(&cset->tasks);
1206 INIT_LIST_HEAD(&cset->mg_tasks);
1207 INIT_LIST_HEAD(&cset->dying_tasks);
1208 INIT_LIST_HEAD(&cset->task_iters);
1209 INIT_LIST_HEAD(&cset->threaded_csets);
1210 INIT_HLIST_NODE(&cset->hlist);
1211 INIT_LIST_HEAD(&cset->cgrp_links);
1212 INIT_LIST_HEAD(&cset->mg_preload_node);
1213 INIT_LIST_HEAD(&cset->mg_node);
1215 /* Copy the set of subsystem state objects generated in
1216 * find_existing_css_set() */
1217 memcpy(cset->subsys, template, sizeof(cset->subsys));
1219 spin_lock_irq(&css_set_lock);
1220 /* Add reference counts and links from the new css_set. */
1221 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1222 struct cgroup *c = link->cgrp;
1224 if (c->root == cgrp->root)
1226 link_css_set(&tmp_links, cset, c);
1229 BUG_ON(!list_empty(&tmp_links));
1233 /* Add @cset to the hash table */
1234 key = css_set_hash(cset->subsys);
1235 hash_add(css_set_table, &cset->hlist, key);
1237 for_each_subsys(ss, ssid) {
1238 struct cgroup_subsys_state *css = cset->subsys[ssid];
1240 list_add_tail(&cset->e_cset_node[ssid],
1241 &css->cgroup->e_csets[ssid]);
1245 spin_unlock_irq(&css_set_lock);
1248 * If @cset should be threaded, look up the matching dom_cset and
1249 * link them up. We first fully initialize @cset then look for the
1250 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1251 * to stay empty until we return.
1253 if (cgroup_is_threaded(cset->dfl_cgrp)) {
1254 struct css_set *dcset;
1256 dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
1262 spin_lock_irq(&css_set_lock);
1263 cset->dom_cset = dcset;
1264 list_add_tail(&cset->threaded_csets_node,
1265 &dcset->threaded_csets);
1266 spin_unlock_irq(&css_set_lock);
1272 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1274 struct cgroup *root_cgrp = kf_root->kn->priv;
1276 return root_cgrp->root;
1279 static int cgroup_init_root_id(struct cgroup_root *root)
1283 lockdep_assert_held(&cgroup_mutex);
1285 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1289 root->hierarchy_id = id;
1293 static void cgroup_exit_root_id(struct cgroup_root *root)
1295 lockdep_assert_held(&cgroup_mutex);
1297 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1300 void cgroup_free_root(struct cgroup_root *root)
1305 static void cgroup_destroy_root(struct cgroup_root *root)
1307 struct cgroup *cgrp = &root->cgrp;
1308 struct cgrp_cset_link *link, *tmp_link;
1310 trace_cgroup_destroy_root(root);
1312 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1314 BUG_ON(atomic_read(&root->nr_cgrps));
1315 BUG_ON(!list_empty(&cgrp->self.children));
1317 /* Rebind all subsystems back to the default hierarchy */
1318 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1321 * Release all the links from cset_links to this hierarchy's
1324 spin_lock_irq(&css_set_lock);
1326 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1327 list_del(&link->cset_link);
1328 list_del(&link->cgrp_link);
1332 spin_unlock_irq(&css_set_lock);
1334 if (!list_empty(&root->root_list)) {
1335 list_del(&root->root_list);
1336 cgroup_root_count--;
1339 cgroup_exit_root_id(root);
1341 mutex_unlock(&cgroup_mutex);
1343 cgroup_rstat_exit(cgrp);
1344 kernfs_destroy_root(root->kf_root);
1345 cgroup_free_root(root);
1349 * look up cgroup associated with current task's cgroup namespace on the
1350 * specified hierarchy
1352 static struct cgroup *
1353 current_cgns_cgroup_from_root(struct cgroup_root *root)
1355 struct cgroup *res = NULL;
1356 struct css_set *cset;
1358 lockdep_assert_held(&css_set_lock);
1362 cset = current->nsproxy->cgroup_ns->root_cset;
1363 if (cset == &init_css_set) {
1365 } else if (root == &cgrp_dfl_root) {
1366 res = cset->dfl_cgrp;
1368 struct cgrp_cset_link *link;
1370 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1371 struct cgroup *c = link->cgrp;
1373 if (c->root == root) {
1385 /* look up cgroup associated with given css_set on the specified hierarchy */
1386 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1387 struct cgroup_root *root)
1389 struct cgroup *res = NULL;
1391 lockdep_assert_held(&cgroup_mutex);
1392 lockdep_assert_held(&css_set_lock);
1394 if (cset == &init_css_set) {
1396 } else if (root == &cgrp_dfl_root) {
1397 res = cset->dfl_cgrp;
1399 struct cgrp_cset_link *link;
1401 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1402 struct cgroup *c = link->cgrp;
1404 if (c->root == root) {
1416 * Return the cgroup for "task" from the given hierarchy. Must be
1417 * called with cgroup_mutex and css_set_lock held.
1419 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1420 struct cgroup_root *root)
1423 * No need to lock the task - since we hold css_set_lock the
1424 * task can't change groups.
1426 return cset_cgroup_from_root(task_css_set(task), root);
1430 * A task must hold cgroup_mutex to modify cgroups.
1432 * Any task can increment and decrement the count field without lock.
1433 * So in general, code holding cgroup_mutex can't rely on the count
1434 * field not changing. However, if the count goes to zero, then only
1435 * cgroup_attach_task() can increment it again. Because a count of zero
1436 * means that no tasks are currently attached, therefore there is no
1437 * way a task attached to that cgroup can fork (the other way to
1438 * increment the count). So code holding cgroup_mutex can safely
1439 * assume that if the count is zero, it will stay zero. Similarly, if
1440 * a task holds cgroup_mutex on a cgroup with zero count, it
1441 * knows that the cgroup won't be removed, as cgroup_rmdir()
1444 * A cgroup can only be deleted if both its 'count' of using tasks
1445 * is zero, and its list of 'children' cgroups is empty. Since all
1446 * tasks in the system use _some_ cgroup, and since there is always at
1447 * least one task in the system (init, pid == 1), therefore, root cgroup
1448 * always has either children cgroups and/or using tasks. So we don't
1449 * need a special hack to ensure that root cgroup cannot be deleted.
1451 * P.S. One more locking exception. RCU is used to guard the
1452 * update of a tasks cgroup pointer by cgroup_attach_task()
1455 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1457 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1460 struct cgroup_subsys *ss = cft->ss;
1462 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1463 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
1464 const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : "";
1466 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s",
1467 dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1470 strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1476 * cgroup_file_mode - deduce file mode of a control file
1477 * @cft: the control file in question
1479 * S_IRUGO for read, S_IWUSR for write.
1481 static umode_t cgroup_file_mode(const struct cftype *cft)
1485 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1488 if (cft->write_u64 || cft->write_s64 || cft->write) {
1489 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1499 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1500 * @subtree_control: the new subtree_control mask to consider
1501 * @this_ss_mask: available subsystems
1503 * On the default hierarchy, a subsystem may request other subsystems to be
1504 * enabled together through its ->depends_on mask. In such cases, more
1505 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1507 * This function calculates which subsystems need to be enabled if
1508 * @subtree_control is to be applied while restricted to @this_ss_mask.
1510 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1512 u16 cur_ss_mask = subtree_control;
1513 struct cgroup_subsys *ss;
1516 lockdep_assert_held(&cgroup_mutex);
1518 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1521 u16 new_ss_mask = cur_ss_mask;
1523 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1524 new_ss_mask |= ss->depends_on;
1525 } while_each_subsys_mask();
1528 * Mask out subsystems which aren't available. This can
1529 * happen only if some depended-upon subsystems were bound
1530 * to non-default hierarchies.
1532 new_ss_mask &= this_ss_mask;
1534 if (new_ss_mask == cur_ss_mask)
1536 cur_ss_mask = new_ss_mask;
1543 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1544 * @kn: the kernfs_node being serviced
1546 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1547 * the method finishes if locking succeeded. Note that once this function
1548 * returns the cgroup returned by cgroup_kn_lock_live() may become
1549 * inaccessible any time. If the caller intends to continue to access the
1550 * cgroup, it should pin it before invoking this function.
1552 void cgroup_kn_unlock(struct kernfs_node *kn)
1554 struct cgroup *cgrp;
1556 if (kernfs_type(kn) == KERNFS_DIR)
1559 cgrp = kn->parent->priv;
1561 mutex_unlock(&cgroup_mutex);
1563 kernfs_unbreak_active_protection(kn);
1568 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1569 * @kn: the kernfs_node being serviced
1570 * @drain_offline: perform offline draining on the cgroup
1572 * This helper is to be used by a cgroup kernfs method currently servicing
1573 * @kn. It breaks the active protection, performs cgroup locking and
1574 * verifies that the associated cgroup is alive. Returns the cgroup if
1575 * alive; otherwise, %NULL. A successful return should be undone by a
1576 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1577 * cgroup is drained of offlining csses before return.
1579 * Any cgroup kernfs method implementation which requires locking the
1580 * associated cgroup should use this helper. It avoids nesting cgroup
1581 * locking under kernfs active protection and allows all kernfs operations
1582 * including self-removal.
1584 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1586 struct cgroup *cgrp;
1588 if (kernfs_type(kn) == KERNFS_DIR)
1591 cgrp = kn->parent->priv;
1594 * We're gonna grab cgroup_mutex which nests outside kernfs
1595 * active_ref. cgroup liveliness check alone provides enough
1596 * protection against removal. Ensure @cgrp stays accessible and
1597 * break the active_ref protection.
1599 if (!cgroup_tryget(cgrp))
1601 kernfs_break_active_protection(kn);
1604 cgroup_lock_and_drain_offline(cgrp);
1606 mutex_lock(&cgroup_mutex);
1608 if (!cgroup_is_dead(cgrp))
1611 cgroup_kn_unlock(kn);
1615 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1617 char name[CGROUP_FILE_NAME_MAX];
1619 lockdep_assert_held(&cgroup_mutex);
1621 if (cft->file_offset) {
1622 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1623 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1625 spin_lock_irq(&cgroup_file_kn_lock);
1627 spin_unlock_irq(&cgroup_file_kn_lock);
1629 del_timer_sync(&cfile->notify_timer);
1632 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1636 * css_clear_dir - remove subsys files in a cgroup directory
1639 static void css_clear_dir(struct cgroup_subsys_state *css)
1641 struct cgroup *cgrp = css->cgroup;
1642 struct cftype *cfts;
1644 if (!(css->flags & CSS_VISIBLE))
1647 css->flags &= ~CSS_VISIBLE;
1650 if (cgroup_on_dfl(cgrp))
1651 cfts = cgroup_base_files;
1653 cfts = cgroup1_base_files;
1655 cgroup_addrm_files(css, cgrp, cfts, false);
1657 list_for_each_entry(cfts, &css->ss->cfts, node)
1658 cgroup_addrm_files(css, cgrp, cfts, false);
1663 * css_populate_dir - create subsys files in a cgroup directory
1666 * On failure, no file is added.
1668 static int css_populate_dir(struct cgroup_subsys_state *css)
1670 struct cgroup *cgrp = css->cgroup;
1671 struct cftype *cfts, *failed_cfts;
1674 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1678 if (cgroup_on_dfl(cgrp))
1679 cfts = cgroup_base_files;
1681 cfts = cgroup1_base_files;
1683 ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1687 list_for_each_entry(cfts, &css->ss->cfts, node) {
1688 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1696 css->flags |= CSS_VISIBLE;
1700 list_for_each_entry(cfts, &css->ss->cfts, node) {
1701 if (cfts == failed_cfts)
1703 cgroup_addrm_files(css, cgrp, cfts, false);
1708 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1710 struct cgroup *dcgrp = &dst_root->cgrp;
1711 struct cgroup_subsys *ss;
1714 lockdep_assert_held(&cgroup_mutex);
1716 do_each_subsys_mask(ss, ssid, ss_mask) {
1718 * If @ss has non-root csses attached to it, can't move.
1719 * If @ss is an implicit controller, it is exempt from this
1720 * rule and can be stolen.
1722 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1723 !ss->implicit_on_dfl)
1726 /* can't move between two non-dummy roots either */
1727 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1729 } while_each_subsys_mask();
1731 do_each_subsys_mask(ss, ssid, ss_mask) {
1732 struct cgroup_root *src_root = ss->root;
1733 struct cgroup *scgrp = &src_root->cgrp;
1734 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1735 struct css_set *cset;
1737 WARN_ON(!css || cgroup_css(dcgrp, ss));
1739 /* disable from the source */
1740 src_root->subsys_mask &= ~(1 << ssid);
1741 WARN_ON(cgroup_apply_control(scgrp));
1742 cgroup_finalize_control(scgrp, 0);
1745 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1746 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1747 ss->root = dst_root;
1748 css->cgroup = dcgrp;
1750 spin_lock_irq(&css_set_lock);
1751 hash_for_each(css_set_table, i, cset, hlist)
1752 list_move_tail(&cset->e_cset_node[ss->id],
1753 &dcgrp->e_csets[ss->id]);
1754 spin_unlock_irq(&css_set_lock);
1756 if (ss->css_rstat_flush) {
1757 list_del_rcu(&css->rstat_css_node);
1758 list_add_rcu(&css->rstat_css_node,
1759 &dcgrp->rstat_css_list);
1762 /* default hierarchy doesn't enable controllers by default */
1763 dst_root->subsys_mask |= 1 << ssid;
1764 if (dst_root == &cgrp_dfl_root) {
1765 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1767 dcgrp->subtree_control |= 1 << ssid;
1768 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1771 ret = cgroup_apply_control(dcgrp);
1773 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1778 } while_each_subsys_mask();
1780 kernfs_activate(dcgrp->kn);
1784 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1785 struct kernfs_root *kf_root)
1789 struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1790 struct cgroup *ns_cgroup;
1792 buf = kmalloc(PATH_MAX, GFP_KERNEL);
1796 spin_lock_irq(&css_set_lock);
1797 ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1798 len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1799 spin_unlock_irq(&css_set_lock);
1801 if (len >= PATH_MAX)
1804 seq_escape(sf, buf, " \t\n\\");
1811 enum cgroup2_param {
1813 Opt_memory_localevents,
1814 Opt_memory_recursiveprot,
1818 static const struct fs_parameter_spec cgroup2_fs_parameters[] = {
1819 fsparam_flag("nsdelegate", Opt_nsdelegate),
1820 fsparam_flag("memory_localevents", Opt_memory_localevents),
1821 fsparam_flag("memory_recursiveprot", Opt_memory_recursiveprot),
1825 static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1827 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1828 struct fs_parse_result result;
1831 opt = fs_parse(fc, cgroup2_fs_parameters, param, &result);
1836 case Opt_nsdelegate:
1837 ctx->flags |= CGRP_ROOT_NS_DELEGATE;
1839 case Opt_memory_localevents:
1840 ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1842 case Opt_memory_recursiveprot:
1843 ctx->flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1849 static void apply_cgroup_root_flags(unsigned int root_flags)
1851 if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
1852 if (root_flags & CGRP_ROOT_NS_DELEGATE)
1853 cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
1855 cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
1857 if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1858 cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1860 cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1862 if (root_flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1863 cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1865 cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1869 static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
1871 if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
1872 seq_puts(seq, ",nsdelegate");
1873 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1874 seq_puts(seq, ",memory_localevents");
1875 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1876 seq_puts(seq, ",memory_recursiveprot");
1880 static int cgroup_reconfigure(struct fs_context *fc)
1882 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1884 apply_cgroup_root_flags(ctx->flags);
1888 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1890 struct cgroup_subsys *ss;
1893 INIT_LIST_HEAD(&cgrp->self.sibling);
1894 INIT_LIST_HEAD(&cgrp->self.children);
1895 INIT_LIST_HEAD(&cgrp->cset_links);
1896 INIT_LIST_HEAD(&cgrp->pidlists);
1897 mutex_init(&cgrp->pidlist_mutex);
1898 cgrp->self.cgroup = cgrp;
1899 cgrp->self.flags |= CSS_ONLINE;
1900 cgrp->dom_cgrp = cgrp;
1901 cgrp->max_descendants = INT_MAX;
1902 cgrp->max_depth = INT_MAX;
1903 INIT_LIST_HEAD(&cgrp->rstat_css_list);
1904 prev_cputime_init(&cgrp->prev_cputime);
1906 for_each_subsys(ss, ssid)
1907 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1909 init_waitqueue_head(&cgrp->offline_waitq);
1910 INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1913 void init_cgroup_root(struct cgroup_fs_context *ctx)
1915 struct cgroup_root *root = ctx->root;
1916 struct cgroup *cgrp = &root->cgrp;
1918 INIT_LIST_HEAD(&root->root_list);
1919 atomic_set(&root->nr_cgrps, 1);
1921 init_cgroup_housekeeping(cgrp);
1923 root->flags = ctx->flags;
1924 if (ctx->release_agent)
1925 strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX);
1927 strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN);
1928 if (ctx->cpuset_clone_children)
1929 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1932 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1934 LIST_HEAD(tmp_links);
1935 struct cgroup *root_cgrp = &root->cgrp;
1936 struct kernfs_syscall_ops *kf_sops;
1937 struct css_set *cset;
1940 lockdep_assert_held(&cgroup_mutex);
1942 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
1948 * We're accessing css_set_count without locking css_set_lock here,
1949 * but that's OK - it can only be increased by someone holding
1950 * cgroup_lock, and that's us. Later rebinding may disable
1951 * controllers on the default hierarchy and thus create new csets,
1952 * which can't be more than the existing ones. Allocate 2x.
1954 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
1958 ret = cgroup_init_root_id(root);
1962 kf_sops = root == &cgrp_dfl_root ?
1963 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
1965 root->kf_root = kernfs_create_root(kf_sops,
1966 KERNFS_ROOT_CREATE_DEACTIVATED |
1967 KERNFS_ROOT_SUPPORT_EXPORTOP |
1968 KERNFS_ROOT_SUPPORT_USER_XATTR,
1970 if (IS_ERR(root->kf_root)) {
1971 ret = PTR_ERR(root->kf_root);
1974 root_cgrp->kn = root->kf_root->kn;
1975 WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
1976 root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
1978 ret = css_populate_dir(&root_cgrp->self);
1982 ret = cgroup_rstat_init(root_cgrp);
1986 ret = rebind_subsystems(root, ss_mask);
1990 ret = cgroup_bpf_inherit(root_cgrp);
1993 trace_cgroup_setup_root(root);
1996 * There must be no failure case after here, since rebinding takes
1997 * care of subsystems' refcounts, which are explicitly dropped in
1998 * the failure exit path.
2000 list_add(&root->root_list, &cgroup_roots);
2001 cgroup_root_count++;
2004 * Link the root cgroup in this hierarchy into all the css_set
2007 spin_lock_irq(&css_set_lock);
2008 hash_for_each(css_set_table, i, cset, hlist) {
2009 link_css_set(&tmp_links, cset, root_cgrp);
2010 if (css_set_populated(cset))
2011 cgroup_update_populated(root_cgrp, true);
2013 spin_unlock_irq(&css_set_lock);
2015 BUG_ON(!list_empty(&root_cgrp->self.children));
2016 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
2022 cgroup_rstat_exit(root_cgrp);
2024 kernfs_destroy_root(root->kf_root);
2025 root->kf_root = NULL;
2027 cgroup_exit_root_id(root);
2029 percpu_ref_exit(&root_cgrp->self.refcnt);
2031 free_cgrp_cset_links(&tmp_links);
2035 int cgroup_do_get_tree(struct fs_context *fc)
2037 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2040 ctx->kfc.root = ctx->root->kf_root;
2041 if (fc->fs_type == &cgroup2_fs_type)
2042 ctx->kfc.magic = CGROUP2_SUPER_MAGIC;
2044 ctx->kfc.magic = CGROUP_SUPER_MAGIC;
2045 ret = kernfs_get_tree(fc);
2048 * In non-init cgroup namespace, instead of root cgroup's dentry,
2049 * we return the dentry corresponding to the cgroupns->root_cgrp.
2051 if (!ret && ctx->ns != &init_cgroup_ns) {
2052 struct dentry *nsdentry;
2053 struct super_block *sb = fc->root->d_sb;
2054 struct cgroup *cgrp;
2056 mutex_lock(&cgroup_mutex);
2057 spin_lock_irq(&css_set_lock);
2059 cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root);
2061 spin_unlock_irq(&css_set_lock);
2062 mutex_unlock(&cgroup_mutex);
2064 nsdentry = kernfs_node_dentry(cgrp->kn, sb);
2066 if (IS_ERR(nsdentry)) {
2067 deactivate_locked_super(sb);
2068 ret = PTR_ERR(nsdentry);
2071 fc->root = nsdentry;
2074 if (!ctx->kfc.new_sb_created)
2075 cgroup_put(&ctx->root->cgrp);
2081 * Destroy a cgroup filesystem context.
2083 static void cgroup_fs_context_free(struct fs_context *fc)
2085 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2088 kfree(ctx->release_agent);
2089 put_cgroup_ns(ctx->ns);
2090 kernfs_free_fs_context(fc);
2094 static int cgroup_get_tree(struct fs_context *fc)
2096 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2099 cgrp_dfl_visible = true;
2100 cgroup_get_live(&cgrp_dfl_root.cgrp);
2101 ctx->root = &cgrp_dfl_root;
2103 ret = cgroup_do_get_tree(fc);
2105 apply_cgroup_root_flags(ctx->flags);
2109 static const struct fs_context_operations cgroup_fs_context_ops = {
2110 .free = cgroup_fs_context_free,
2111 .parse_param = cgroup2_parse_param,
2112 .get_tree = cgroup_get_tree,
2113 .reconfigure = cgroup_reconfigure,
2116 static const struct fs_context_operations cgroup1_fs_context_ops = {
2117 .free = cgroup_fs_context_free,
2118 .parse_param = cgroup1_parse_param,
2119 .get_tree = cgroup1_get_tree,
2120 .reconfigure = cgroup1_reconfigure,
2124 * Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2125 * we select the namespace we're going to use.
2127 static int cgroup_init_fs_context(struct fs_context *fc)
2129 struct cgroup_fs_context *ctx;
2131 ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL);
2135 ctx->ns = current->nsproxy->cgroup_ns;
2136 get_cgroup_ns(ctx->ns);
2137 fc->fs_private = &ctx->kfc;
2138 if (fc->fs_type == &cgroup2_fs_type)
2139 fc->ops = &cgroup_fs_context_ops;
2141 fc->ops = &cgroup1_fs_context_ops;
2142 put_user_ns(fc->user_ns);
2143 fc->user_ns = get_user_ns(ctx->ns->user_ns);
2148 static void cgroup_kill_sb(struct super_block *sb)
2150 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2151 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2154 * If @root doesn't have any children, start killing it.
2155 * This prevents new mounts by disabling percpu_ref_tryget_live().
2156 * cgroup_mount() may wait for @root's release.
2158 * And don't kill the default root.
2160 if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root &&
2161 !percpu_ref_is_dying(&root->cgrp.self.refcnt))
2162 percpu_ref_kill(&root->cgrp.self.refcnt);
2163 cgroup_put(&root->cgrp);
2167 struct file_system_type cgroup_fs_type = {
2169 .init_fs_context = cgroup_init_fs_context,
2170 .parameters = cgroup1_fs_parameters,
2171 .kill_sb = cgroup_kill_sb,
2172 .fs_flags = FS_USERNS_MOUNT,
2175 static struct file_system_type cgroup2_fs_type = {
2177 .init_fs_context = cgroup_init_fs_context,
2178 .parameters = cgroup2_fs_parameters,
2179 .kill_sb = cgroup_kill_sb,
2180 .fs_flags = FS_USERNS_MOUNT,
2183 #ifdef CONFIG_CPUSETS
2184 static const struct fs_context_operations cpuset_fs_context_ops = {
2185 .get_tree = cgroup1_get_tree,
2186 .free = cgroup_fs_context_free,
2190 * This is ugly, but preserves the userspace API for existing cpuset
2191 * users. If someone tries to mount the "cpuset" filesystem, we
2192 * silently switch it to mount "cgroup" instead
2194 static int cpuset_init_fs_context(struct fs_context *fc)
2196 char *agent = kstrdup("/sbin/cpuset_release_agent", GFP_USER);
2197 struct cgroup_fs_context *ctx;
2200 err = cgroup_init_fs_context(fc);
2206 fc->ops = &cpuset_fs_context_ops;
2208 ctx = cgroup_fc2context(fc);
2209 ctx->subsys_mask = 1 << cpuset_cgrp_id;
2210 ctx->flags |= CGRP_ROOT_NOPREFIX;
2211 ctx->release_agent = agent;
2213 get_filesystem(&cgroup_fs_type);
2214 put_filesystem(fc->fs_type);
2215 fc->fs_type = &cgroup_fs_type;
2220 static struct file_system_type cpuset_fs_type = {
2222 .init_fs_context = cpuset_init_fs_context,
2223 .fs_flags = FS_USERNS_MOUNT,
2227 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2228 struct cgroup_namespace *ns)
2230 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2232 return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2235 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2236 struct cgroup_namespace *ns)
2240 mutex_lock(&cgroup_mutex);
2241 spin_lock_irq(&css_set_lock);
2243 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2245 spin_unlock_irq(&css_set_lock);
2246 mutex_unlock(&cgroup_mutex);
2250 EXPORT_SYMBOL_GPL(cgroup_path_ns);
2253 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2254 * @task: target task
2255 * @buf: the buffer to write the path into
2256 * @buflen: the length of the buffer
2258 * Determine @task's cgroup on the first (the one with the lowest non-zero
2259 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2260 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2261 * cgroup controller callbacks.
2263 * Return value is the same as kernfs_path().
2265 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2267 struct cgroup_root *root;
2268 struct cgroup *cgrp;
2269 int hierarchy_id = 1;
2272 mutex_lock(&cgroup_mutex);
2273 spin_lock_irq(&css_set_lock);
2275 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2278 cgrp = task_cgroup_from_root(task, root);
2279 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2281 /* if no hierarchy exists, everyone is in "/" */
2282 ret = strlcpy(buf, "/", buflen);
2285 spin_unlock_irq(&css_set_lock);
2286 mutex_unlock(&cgroup_mutex);
2289 EXPORT_SYMBOL_GPL(task_cgroup_path);
2292 * cgroup_migrate_add_task - add a migration target task to a migration context
2293 * @task: target task
2294 * @mgctx: target migration context
2296 * Add @task, which is a migration target, to @mgctx->tset. This function
2297 * becomes noop if @task doesn't need to be migrated. @task's css_set
2298 * should have been added as a migration source and @task->cg_list will be
2299 * moved from the css_set's tasks list to mg_tasks one.
2301 static void cgroup_migrate_add_task(struct task_struct *task,
2302 struct cgroup_mgctx *mgctx)
2304 struct css_set *cset;
2306 lockdep_assert_held(&css_set_lock);
2308 /* @task either already exited or can't exit until the end */
2309 if (task->flags & PF_EXITING)
2312 /* cgroup_threadgroup_rwsem protects racing against forks */
2313 WARN_ON_ONCE(list_empty(&task->cg_list));
2315 cset = task_css_set(task);
2316 if (!cset->mg_src_cgrp)
2319 mgctx->tset.nr_tasks++;
2321 list_move_tail(&task->cg_list, &cset->mg_tasks);
2322 if (list_empty(&cset->mg_node))
2323 list_add_tail(&cset->mg_node,
2324 &mgctx->tset.src_csets);
2325 if (list_empty(&cset->mg_dst_cset->mg_node))
2326 list_add_tail(&cset->mg_dst_cset->mg_node,
2327 &mgctx->tset.dst_csets);
2331 * cgroup_taskset_first - reset taskset and return the first task
2332 * @tset: taskset of interest
2333 * @dst_cssp: output variable for the destination css
2335 * @tset iteration is initialized and the first task is returned.
2337 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2338 struct cgroup_subsys_state **dst_cssp)
2340 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2341 tset->cur_task = NULL;
2343 return cgroup_taskset_next(tset, dst_cssp);
2347 * cgroup_taskset_next - iterate to the next task in taskset
2348 * @tset: taskset of interest
2349 * @dst_cssp: output variable for the destination css
2351 * Return the next task in @tset. Iteration must have been initialized
2352 * with cgroup_taskset_first().
2354 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2355 struct cgroup_subsys_state **dst_cssp)
2357 struct css_set *cset = tset->cur_cset;
2358 struct task_struct *task = tset->cur_task;
2360 while (&cset->mg_node != tset->csets) {
2362 task = list_first_entry(&cset->mg_tasks,
2363 struct task_struct, cg_list);
2365 task = list_next_entry(task, cg_list);
2367 if (&task->cg_list != &cset->mg_tasks) {
2368 tset->cur_cset = cset;
2369 tset->cur_task = task;
2372 * This function may be called both before and
2373 * after cgroup_taskset_migrate(). The two cases
2374 * can be distinguished by looking at whether @cset
2375 * has its ->mg_dst_cset set.
2377 if (cset->mg_dst_cset)
2378 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2380 *dst_cssp = cset->subsys[tset->ssid];
2385 cset = list_next_entry(cset, mg_node);
2393 * cgroup_taskset_migrate - migrate a taskset
2394 * @mgctx: migration context
2396 * Migrate tasks in @mgctx as setup by migration preparation functions.
2397 * This function fails iff one of the ->can_attach callbacks fails and
2398 * guarantees that either all or none of the tasks in @mgctx are migrated.
2399 * @mgctx is consumed regardless of success.
2401 static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2403 struct cgroup_taskset *tset = &mgctx->tset;
2404 struct cgroup_subsys *ss;
2405 struct task_struct *task, *tmp_task;
2406 struct css_set *cset, *tmp_cset;
2407 int ssid, failed_ssid, ret;
2409 /* check that we can legitimately attach to the cgroup */
2410 if (tset->nr_tasks) {
2411 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2412 if (ss->can_attach) {
2414 ret = ss->can_attach(tset);
2417 goto out_cancel_attach;
2420 } while_each_subsys_mask();
2424 * Now that we're guaranteed success, proceed to move all tasks to
2425 * the new cgroup. There are no failure cases after here, so this
2426 * is the commit point.
2428 spin_lock_irq(&css_set_lock);
2429 list_for_each_entry(cset, &tset->src_csets, mg_node) {
2430 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2431 struct css_set *from_cset = task_css_set(task);
2432 struct css_set *to_cset = cset->mg_dst_cset;
2434 get_css_set(to_cset);
2435 to_cset->nr_tasks++;
2436 css_set_move_task(task, from_cset, to_cset, true);
2437 from_cset->nr_tasks--;
2439 * If the source or destination cgroup is frozen,
2440 * the task might require to change its state.
2442 cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp,
2444 put_css_set_locked(from_cset);
2448 spin_unlock_irq(&css_set_lock);
2451 * Migration is committed, all target tasks are now on dst_csets.
2452 * Nothing is sensitive to fork() after this point. Notify
2453 * controllers that migration is complete.
2455 tset->csets = &tset->dst_csets;
2457 if (tset->nr_tasks) {
2458 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2463 } while_each_subsys_mask();
2467 goto out_release_tset;
2470 if (tset->nr_tasks) {
2471 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2472 if (ssid == failed_ssid)
2474 if (ss->cancel_attach) {
2476 ss->cancel_attach(tset);
2478 } while_each_subsys_mask();
2481 spin_lock_irq(&css_set_lock);
2482 list_splice_init(&tset->dst_csets, &tset->src_csets);
2483 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2484 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2485 list_del_init(&cset->mg_node);
2487 spin_unlock_irq(&css_set_lock);
2490 * Re-initialize the cgroup_taskset structure in case it is reused
2491 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2495 tset->csets = &tset->src_csets;
2500 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2501 * @dst_cgrp: destination cgroup to test
2503 * On the default hierarchy, except for the mixable, (possible) thread root
2504 * and threaded cgroups, subtree_control must be zero for migration
2505 * destination cgroups with tasks so that child cgroups don't compete
2508 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2510 /* v1 doesn't have any restriction */
2511 if (!cgroup_on_dfl(dst_cgrp))
2514 /* verify @dst_cgrp can host resources */
2515 if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
2518 /* mixables don't care */
2519 if (cgroup_is_mixable(dst_cgrp))
2523 * If @dst_cgrp is already or can become a thread root or is
2524 * threaded, it doesn't matter.
2526 if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
2529 /* apply no-internal-process constraint */
2530 if (dst_cgrp->subtree_control)
2537 * cgroup_migrate_finish - cleanup after attach
2538 * @mgctx: migration context
2540 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2541 * those functions for details.
2543 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2545 LIST_HEAD(preloaded);
2546 struct css_set *cset, *tmp_cset;
2548 lockdep_assert_held(&cgroup_mutex);
2550 spin_lock_irq(&css_set_lock);
2552 list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded);
2553 list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded);
2555 list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) {
2556 cset->mg_src_cgrp = NULL;
2557 cset->mg_dst_cgrp = NULL;
2558 cset->mg_dst_cset = NULL;
2559 list_del_init(&cset->mg_preload_node);
2560 put_css_set_locked(cset);
2563 spin_unlock_irq(&css_set_lock);
2567 * cgroup_migrate_add_src - add a migration source css_set
2568 * @src_cset: the source css_set to add
2569 * @dst_cgrp: the destination cgroup
2570 * @mgctx: migration context
2572 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2573 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2574 * up by cgroup_migrate_finish().
2576 * This function may be called without holding cgroup_threadgroup_rwsem
2577 * even if the target is a process. Threads may be created and destroyed
2578 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2579 * into play and the preloaded css_sets are guaranteed to cover all
2582 void cgroup_migrate_add_src(struct css_set *src_cset,
2583 struct cgroup *dst_cgrp,
2584 struct cgroup_mgctx *mgctx)
2586 struct cgroup *src_cgrp;
2588 lockdep_assert_held(&cgroup_mutex);
2589 lockdep_assert_held(&css_set_lock);
2592 * If ->dead, @src_set is associated with one or more dead cgroups
2593 * and doesn't contain any migratable tasks. Ignore it early so
2594 * that the rest of migration path doesn't get confused by it.
2599 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2601 if (!list_empty(&src_cset->mg_preload_node))
2604 WARN_ON(src_cset->mg_src_cgrp);
2605 WARN_ON(src_cset->mg_dst_cgrp);
2606 WARN_ON(!list_empty(&src_cset->mg_tasks));
2607 WARN_ON(!list_empty(&src_cset->mg_node));
2609 src_cset->mg_src_cgrp = src_cgrp;
2610 src_cset->mg_dst_cgrp = dst_cgrp;
2611 get_css_set(src_cset);
2612 list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets);
2616 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2617 * @mgctx: migration context
2619 * Tasks are about to be moved and all the source css_sets have been
2620 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2621 * pins all destination css_sets, links each to its source, and append them
2622 * to @mgctx->preloaded_dst_csets.
2624 * This function must be called after cgroup_migrate_add_src() has been
2625 * called on each migration source css_set. After migration is performed
2626 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2629 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2631 struct css_set *src_cset, *tmp_cset;
2633 lockdep_assert_held(&cgroup_mutex);
2635 /* look up the dst cset for each src cset and link it to src */
2636 list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2638 struct css_set *dst_cset;
2639 struct cgroup_subsys *ss;
2642 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2646 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2649 * If src cset equals dst, it's noop. Drop the src.
2650 * cgroup_migrate() will skip the cset too. Note that we
2651 * can't handle src == dst as some nodes are used by both.
2653 if (src_cset == dst_cset) {
2654 src_cset->mg_src_cgrp = NULL;
2655 src_cset->mg_dst_cgrp = NULL;
2656 list_del_init(&src_cset->mg_preload_node);
2657 put_css_set(src_cset);
2658 put_css_set(dst_cset);
2662 src_cset->mg_dst_cset = dst_cset;
2664 if (list_empty(&dst_cset->mg_preload_node))
2665 list_add_tail(&dst_cset->mg_preload_node,
2666 &mgctx->preloaded_dst_csets);
2668 put_css_set(dst_cset);
2670 for_each_subsys(ss, ssid)
2671 if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2672 mgctx->ss_mask |= 1 << ssid;
2679 * cgroup_migrate - migrate a process or task to a cgroup
2680 * @leader: the leader of the process or the task to migrate
2681 * @threadgroup: whether @leader points to the whole process or a single task
2682 * @mgctx: migration context
2684 * Migrate a process or task denoted by @leader. If migrating a process,
2685 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2686 * responsible for invoking cgroup_migrate_add_src() and
2687 * cgroup_migrate_prepare_dst() on the targets before invoking this
2688 * function and following up with cgroup_migrate_finish().
2690 * As long as a controller's ->can_attach() doesn't fail, this function is
2691 * guaranteed to succeed. This means that, excluding ->can_attach()
2692 * failure, when migrating multiple targets, the success or failure can be
2693 * decided for all targets by invoking group_migrate_prepare_dst() before
2694 * actually starting migrating.
2696 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2697 struct cgroup_mgctx *mgctx)
2699 struct task_struct *task;
2702 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2703 * already PF_EXITING could be freed from underneath us unless we
2704 * take an rcu_read_lock.
2706 spin_lock_irq(&css_set_lock);
2710 cgroup_migrate_add_task(task, mgctx);
2713 } while_each_thread(leader, task);
2715 spin_unlock_irq(&css_set_lock);
2717 return cgroup_migrate_execute(mgctx);
2721 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2722 * @dst_cgrp: the cgroup to attach to
2723 * @leader: the task or the leader of the threadgroup to be attached
2724 * @threadgroup: attach the whole threadgroup?
2726 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2728 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2731 DEFINE_CGROUP_MGCTX(mgctx);
2732 struct task_struct *task;
2735 /* look up all src csets */
2736 spin_lock_irq(&css_set_lock);
2740 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2743 } while_each_thread(leader, task);
2745 spin_unlock_irq(&css_set_lock);
2747 /* prepare dst csets and commit */
2748 ret = cgroup_migrate_prepare_dst(&mgctx);
2750 ret = cgroup_migrate(leader, threadgroup, &mgctx);
2752 cgroup_migrate_finish(&mgctx);
2755 TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup);
2760 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
2762 __acquires(&cgroup_threadgroup_rwsem)
2764 struct task_struct *tsk;
2767 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2768 return ERR_PTR(-EINVAL);
2771 * If we migrate a single thread, we don't care about threadgroup
2772 * stability. If the thread is `current`, it won't exit(2) under our
2773 * hands or change PID through exec(2). We exclude
2774 * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write
2775 * callers by cgroup_mutex.
2776 * Therefore, we can skip the global lock.
2778 lockdep_assert_held(&cgroup_mutex);
2779 if (pid || threadgroup) {
2780 percpu_down_write(&cgroup_threadgroup_rwsem);
2788 tsk = find_task_by_vpid(pid);
2790 tsk = ERR_PTR(-ESRCH);
2791 goto out_unlock_threadgroup;
2798 tsk = tsk->group_leader;
2801 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2802 * If userland migrates such a kthread to a non-root cgroup, it can
2803 * become trapped in a cpuset, or RT kthread may be born in a
2804 * cgroup with no rt_runtime allocated. Just say no.
2806 if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2807 tsk = ERR_PTR(-EINVAL);
2808 goto out_unlock_threadgroup;
2811 get_task_struct(tsk);
2812 goto out_unlock_rcu;
2814 out_unlock_threadgroup:
2816 percpu_up_write(&cgroup_threadgroup_rwsem);
2824 void cgroup_procs_write_finish(struct task_struct *task, bool locked)
2825 __releases(&cgroup_threadgroup_rwsem)
2827 struct cgroup_subsys *ss;
2830 /* release reference from cgroup_procs_write_start() */
2831 put_task_struct(task);
2834 percpu_up_write(&cgroup_threadgroup_rwsem);
2835 for_each_subsys(ss, ssid)
2836 if (ss->post_attach)
2840 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2842 struct cgroup_subsys *ss;
2843 bool printed = false;
2846 do_each_subsys_mask(ss, ssid, ss_mask) {
2849 seq_puts(seq, ss->name);
2851 } while_each_subsys_mask();
2853 seq_putc(seq, '\n');
2856 /* show controllers which are enabled from the parent */
2857 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2859 struct cgroup *cgrp = seq_css(seq)->cgroup;
2861 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2865 /* show controllers which are enabled for a given cgroup's children */
2866 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2868 struct cgroup *cgrp = seq_css(seq)->cgroup;
2870 cgroup_print_ss_mask(seq, cgrp->subtree_control);
2875 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2876 * @cgrp: root of the subtree to update csses for
2878 * @cgrp's control masks have changed and its subtree's css associations
2879 * need to be updated accordingly. This function looks up all css_sets
2880 * which are attached to the subtree, creates the matching updated css_sets
2881 * and migrates the tasks to the new ones.
2883 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2885 DEFINE_CGROUP_MGCTX(mgctx);
2886 struct cgroup_subsys_state *d_css;
2887 struct cgroup *dsct;
2888 struct css_set *src_cset;
2891 lockdep_assert_held(&cgroup_mutex);
2893 percpu_down_write(&cgroup_threadgroup_rwsem);
2895 /* look up all csses currently attached to @cgrp's subtree */
2896 spin_lock_irq(&css_set_lock);
2897 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2898 struct cgrp_cset_link *link;
2900 list_for_each_entry(link, &dsct->cset_links, cset_link)
2901 cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2903 spin_unlock_irq(&css_set_lock);
2905 /* NULL dst indicates self on default hierarchy */
2906 ret = cgroup_migrate_prepare_dst(&mgctx);
2910 spin_lock_irq(&css_set_lock);
2911 list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) {
2912 struct task_struct *task, *ntask;
2914 /* all tasks in src_csets need to be migrated */
2915 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2916 cgroup_migrate_add_task(task, &mgctx);
2918 spin_unlock_irq(&css_set_lock);
2920 ret = cgroup_migrate_execute(&mgctx);
2922 cgroup_migrate_finish(&mgctx);
2923 percpu_up_write(&cgroup_threadgroup_rwsem);
2928 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2929 * @cgrp: root of the target subtree
2931 * Because css offlining is asynchronous, userland may try to re-enable a
2932 * controller while the previous css is still around. This function grabs
2933 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2935 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
2936 __acquires(&cgroup_mutex)
2938 struct cgroup *dsct;
2939 struct cgroup_subsys_state *d_css;
2940 struct cgroup_subsys *ss;
2944 mutex_lock(&cgroup_mutex);
2946 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2947 for_each_subsys(ss, ssid) {
2948 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2951 if (!css || !percpu_ref_is_dying(&css->refcnt))
2954 cgroup_get_live(dsct);
2955 prepare_to_wait(&dsct->offline_waitq, &wait,
2956 TASK_UNINTERRUPTIBLE);
2958 mutex_unlock(&cgroup_mutex);
2960 finish_wait(&dsct->offline_waitq, &wait);
2969 * cgroup_save_control - save control masks and dom_cgrp of a subtree
2970 * @cgrp: root of the target subtree
2972 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
2973 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
2976 static void cgroup_save_control(struct cgroup *cgrp)
2978 struct cgroup *dsct;
2979 struct cgroup_subsys_state *d_css;
2981 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2982 dsct->old_subtree_control = dsct->subtree_control;
2983 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
2984 dsct->old_dom_cgrp = dsct->dom_cgrp;
2989 * cgroup_propagate_control - refresh control masks of a subtree
2990 * @cgrp: root of the target subtree
2992 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
2993 * ->subtree_control and propagate controller availability through the
2994 * subtree so that descendants don't have unavailable controllers enabled.
2996 static void cgroup_propagate_control(struct cgroup *cgrp)
2998 struct cgroup *dsct;
2999 struct cgroup_subsys_state *d_css;
3001 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3002 dsct->subtree_control &= cgroup_control(dsct);
3003 dsct->subtree_ss_mask =
3004 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3005 cgroup_ss_mask(dsct));
3010 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3011 * @cgrp: root of the target subtree
3013 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3014 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3017 static void cgroup_restore_control(struct cgroup *cgrp)
3019 struct cgroup *dsct;
3020 struct cgroup_subsys_state *d_css;
3022 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3023 dsct->subtree_control = dsct->old_subtree_control;
3024 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3025 dsct->dom_cgrp = dsct->old_dom_cgrp;
3029 static bool css_visible(struct cgroup_subsys_state *css)
3031 struct cgroup_subsys *ss = css->ss;
3032 struct cgroup *cgrp = css->cgroup;
3034 if (cgroup_control(cgrp) & (1 << ss->id))
3036 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3038 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3042 * cgroup_apply_control_enable - enable or show csses according to control
3043 * @cgrp: root of the target subtree
3045 * Walk @cgrp's subtree and create new csses or make the existing ones
3046 * visible. A css is created invisible if it's being implicitly enabled
3047 * through dependency. An invisible css is made visible when the userland
3048 * explicitly enables it.
3050 * Returns 0 on success, -errno on failure. On failure, csses which have
3051 * been processed already aren't cleaned up. The caller is responsible for
3052 * cleaning up with cgroup_apply_control_disable().
3054 static int cgroup_apply_control_enable(struct cgroup *cgrp)
3056 struct cgroup *dsct;
3057 struct cgroup_subsys_state *d_css;
3058 struct cgroup_subsys *ss;
3061 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3062 for_each_subsys(ss, ssid) {
3063 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3065 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3069 css = css_create(dsct, ss);
3071 return PTR_ERR(css);
3074 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3076 if (css_visible(css)) {
3077 ret = css_populate_dir(css);
3088 * cgroup_apply_control_disable - kill or hide csses according to control
3089 * @cgrp: root of the target subtree
3091 * Walk @cgrp's subtree and kill and hide csses so that they match
3092 * cgroup_ss_mask() and cgroup_visible_mask().
3094 * A css is hidden when the userland requests it to be disabled while other
3095 * subsystems are still depending on it. The css must not actively control
3096 * resources and be in the vanilla state if it's made visible again later.
3097 * Controllers which may be depended upon should provide ->css_reset() for
3100 static void cgroup_apply_control_disable(struct cgroup *cgrp)
3102 struct cgroup *dsct;
3103 struct cgroup_subsys_state *d_css;
3104 struct cgroup_subsys *ss;
3107 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3108 for_each_subsys(ss, ssid) {
3109 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3114 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3117 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
3119 } else if (!css_visible(css)) {
3129 * cgroup_apply_control - apply control mask updates to the subtree
3130 * @cgrp: root of the target subtree
3132 * subsystems can be enabled and disabled in a subtree using the following
3135 * 1. Call cgroup_save_control() to stash the current state.
3136 * 2. Update ->subtree_control masks in the subtree as desired.
3137 * 3. Call cgroup_apply_control() to apply the changes.
3138 * 4. Optionally perform other related operations.
3139 * 5. Call cgroup_finalize_control() to finish up.
3141 * This function implements step 3 and propagates the mask changes
3142 * throughout @cgrp's subtree, updates csses accordingly and perform
3143 * process migrations.
3145 static int cgroup_apply_control(struct cgroup *cgrp)
3149 cgroup_propagate_control(cgrp);
3151 ret = cgroup_apply_control_enable(cgrp);
3156 * At this point, cgroup_e_css_by_mask() results reflect the new csses
3157 * making the following cgroup_update_dfl_csses() properly update
3158 * css associations of all tasks in the subtree.
3160 ret = cgroup_update_dfl_csses(cgrp);
3168 * cgroup_finalize_control - finalize control mask update
3169 * @cgrp: root of the target subtree
3170 * @ret: the result of the update
3172 * Finalize control mask update. See cgroup_apply_control() for more info.
3174 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3177 cgroup_restore_control(cgrp);
3178 cgroup_propagate_control(cgrp);
3181 cgroup_apply_control_disable(cgrp);
3184 static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
3186 u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
3188 /* if nothing is getting enabled, nothing to worry about */
3192 /* can @cgrp host any resources? */
3193 if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
3196 /* mixables don't care */
3197 if (cgroup_is_mixable(cgrp))
3200 if (domain_enable) {
3201 /* can't enable domain controllers inside a thread subtree */
3202 if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3206 * Threaded controllers can handle internal competitions
3207 * and are always allowed inside a (prospective) thread
3210 if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3215 * Controllers can't be enabled for a cgroup with tasks to avoid
3216 * child cgroups competing against tasks.
3218 if (cgroup_has_tasks(cgrp))
3224 /* change the enabled child controllers for a cgroup in the default hierarchy */
3225 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3226 char *buf, size_t nbytes,
3229 u16 enable = 0, disable = 0;
3230 struct cgroup *cgrp, *child;
3231 struct cgroup_subsys *ss;
3236 * Parse input - space separated list of subsystem names prefixed
3237 * with either + or -.
3239 buf = strstrip(buf);
3240 while ((tok = strsep(&buf, " "))) {
3243 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3244 if (!cgroup_ssid_enabled(ssid) ||
3245 strcmp(tok + 1, ss->name))
3249 enable |= 1 << ssid;
3250 disable &= ~(1 << ssid);
3251 } else if (*tok == '-') {
3252 disable |= 1 << ssid;
3253 enable &= ~(1 << ssid);
3258 } while_each_subsys_mask();
3259 if (ssid == CGROUP_SUBSYS_COUNT)
3263 cgrp = cgroup_kn_lock_live(of->kn, true);
3267 for_each_subsys(ss, ssid) {
3268 if (enable & (1 << ssid)) {
3269 if (cgrp->subtree_control & (1 << ssid)) {
3270 enable &= ~(1 << ssid);
3274 if (!(cgroup_control(cgrp) & (1 << ssid))) {
3278 } else if (disable & (1 << ssid)) {
3279 if (!(cgrp->subtree_control & (1 << ssid))) {
3280 disable &= ~(1 << ssid);
3284 /* a child has it enabled? */
3285 cgroup_for_each_live_child(child, cgrp) {
3286 if (child->subtree_control & (1 << ssid)) {
3294 if (!enable && !disable) {
3299 ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3303 /* save and update control masks and prepare csses */
3304 cgroup_save_control(cgrp);
3306 cgrp->subtree_control |= enable;
3307 cgrp->subtree_control &= ~disable;
3309 ret = cgroup_apply_control(cgrp);
3310 cgroup_finalize_control(cgrp, ret);
3314 kernfs_activate(cgrp->kn);
3316 cgroup_kn_unlock(of->kn);
3317 return ret ?: nbytes;
3321 * cgroup_enable_threaded - make @cgrp threaded
3322 * @cgrp: the target cgroup
3324 * Called when "threaded" is written to the cgroup.type interface file and
3325 * tries to make @cgrp threaded and join the parent's resource domain.
3326 * This function is never called on the root cgroup as cgroup.type doesn't
3329 static int cgroup_enable_threaded(struct cgroup *cgrp)
3331 struct cgroup *parent = cgroup_parent(cgrp);
3332 struct cgroup *dom_cgrp = parent->dom_cgrp;
3333 struct cgroup *dsct;
3334 struct cgroup_subsys_state *d_css;
3337 lockdep_assert_held(&cgroup_mutex);
3339 /* noop if already threaded */
3340 if (cgroup_is_threaded(cgrp))
3344 * If @cgroup is populated or has domain controllers enabled, it
3345 * can't be switched. While the below cgroup_can_be_thread_root()
3346 * test can catch the same conditions, that's only when @parent is
3347 * not mixable, so let's check it explicitly.
3349 if (cgroup_is_populated(cgrp) ||
3350 cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
3353 /* we're joining the parent's domain, ensure its validity */
3354 if (!cgroup_is_valid_domain(dom_cgrp) ||
3355 !cgroup_can_be_thread_root(dom_cgrp))
3359 * The following shouldn't cause actual migrations and should
3362 cgroup_save_control(cgrp);
3364 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)
3365 if (dsct == cgrp || cgroup_is_threaded(dsct))
3366 dsct->dom_cgrp = dom_cgrp;
3368 ret = cgroup_apply_control(cgrp);
3370 parent->nr_threaded_children++;
3372 cgroup_finalize_control(cgrp, ret);
3376 static int cgroup_type_show(struct seq_file *seq, void *v)
3378 struct cgroup *cgrp = seq_css(seq)->cgroup;
3380 if (cgroup_is_threaded(cgrp))
3381 seq_puts(seq, "threaded\n");
3382 else if (!cgroup_is_valid_domain(cgrp))
3383 seq_puts(seq, "domain invalid\n");
3384 else if (cgroup_is_thread_root(cgrp))
3385 seq_puts(seq, "domain threaded\n");
3387 seq_puts(seq, "domain\n");
3392 static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3393 size_t nbytes, loff_t off)
3395 struct cgroup *cgrp;
3398 /* only switching to threaded mode is supported */
3399 if (strcmp(strstrip(buf), "threaded"))
3402 /* drain dying csses before we re-apply (threaded) subtree control */
3403 cgrp = cgroup_kn_lock_live(of->kn, true);
3407 /* threaded can only be enabled */
3408 ret = cgroup_enable_threaded(cgrp);
3410 cgroup_kn_unlock(of->kn);
3411 return ret ?: nbytes;
3414 static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3416 struct cgroup *cgrp = seq_css(seq)->cgroup;
3417 int descendants = READ_ONCE(cgrp->max_descendants);
3419 if (descendants == INT_MAX)
3420 seq_puts(seq, "max\n");
3422 seq_printf(seq, "%d\n", descendants);
3427 static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3428 char *buf, size_t nbytes, loff_t off)
3430 struct cgroup *cgrp;
3434 buf = strstrip(buf);
3435 if (!strcmp(buf, "max")) {
3436 descendants = INT_MAX;
3438 ret = kstrtoint(buf, 0, &descendants);
3443 if (descendants < 0)
3446 cgrp = cgroup_kn_lock_live(of->kn, false);
3450 cgrp->max_descendants = descendants;
3452 cgroup_kn_unlock(of->kn);
3457 static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3459 struct cgroup *cgrp = seq_css(seq)->cgroup;
3460 int depth = READ_ONCE(cgrp->max_depth);
3462 if (depth == INT_MAX)
3463 seq_puts(seq, "max\n");
3465 seq_printf(seq, "%d\n", depth);
3470 static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3471 char *buf, size_t nbytes, loff_t off)
3473 struct cgroup *cgrp;
3477 buf = strstrip(buf);
3478 if (!strcmp(buf, "max")) {
3481 ret = kstrtoint(buf, 0, &depth);
3489 cgrp = cgroup_kn_lock_live(of->kn, false);
3493 cgrp->max_depth = depth;
3495 cgroup_kn_unlock(of->kn);
3500 static int cgroup_events_show(struct seq_file *seq, void *v)
3502 struct cgroup *cgrp = seq_css(seq)->cgroup;
3504 seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp));
3505 seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags));
3510 static int cgroup_stat_show(struct seq_file *seq, void *v)
3512 struct cgroup *cgroup = seq_css(seq)->cgroup;
3514 seq_printf(seq, "nr_descendants %d\n",
3515 cgroup->nr_descendants);
3516 seq_printf(seq, "nr_dying_descendants %d\n",
3517 cgroup->nr_dying_descendants);
3522 static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq,
3523 struct cgroup *cgrp, int ssid)
3525 struct cgroup_subsys *ss = cgroup_subsys[ssid];
3526 struct cgroup_subsys_state *css;
3529 if (!ss->css_extra_stat_show)
3532 css = cgroup_tryget_css(cgrp, ss);
3536 ret = ss->css_extra_stat_show(seq, css);
3541 static int cpu_stat_show(struct seq_file *seq, void *v)
3543 struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
3546 cgroup_base_stat_cputime_show(seq);
3547 #ifdef CONFIG_CGROUP_SCHED
3548 ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id);
3554 static int cgroup_io_pressure_show(struct seq_file *seq, void *v)
3556 struct cgroup *cgrp = seq_css(seq)->cgroup;
3557 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3559 return psi_show(seq, psi, PSI_IO);
3561 static int cgroup_memory_pressure_show(struct seq_file *seq, void *v)
3563 struct cgroup *cgrp = seq_css(seq)->cgroup;
3564 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3566 return psi_show(seq, psi, PSI_MEM);
3568 static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
3570 struct cgroup *cgrp = seq_css(seq)->cgroup;
3571 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3573 return psi_show(seq, psi, PSI_CPU);
3576 static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, char *buf,
3577 size_t nbytes, enum psi_res res)
3579 struct psi_trigger *new;
3580 struct cgroup *cgrp;
3581 struct psi_group *psi;
3583 cgrp = cgroup_kn_lock_live(of->kn, false);
3588 cgroup_kn_unlock(of->kn);
3590 psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3591 new = psi_trigger_create(psi, buf, nbytes, res);
3594 return PTR_ERR(new);
3597 psi_trigger_replace(&of->priv, new);
3604 static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of,
3605 char *buf, size_t nbytes,
3608 return cgroup_pressure_write(of, buf, nbytes, PSI_IO);
3611 static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of,
3612 char *buf, size_t nbytes,
3615 return cgroup_pressure_write(of, buf, nbytes, PSI_MEM);
3618 static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of,
3619 char *buf, size_t nbytes,
3622 return cgroup_pressure_write(of, buf, nbytes, PSI_CPU);
3625 static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of,
3628 return psi_trigger_poll(&of->priv, of->file, pt);
3631 static void cgroup_pressure_release(struct kernfs_open_file *of)
3633 psi_trigger_replace(&of->priv, NULL);
3635 #endif /* CONFIG_PSI */
3637 static int cgroup_freeze_show(struct seq_file *seq, void *v)
3639 struct cgroup *cgrp = seq_css(seq)->cgroup;
3641 seq_printf(seq, "%d\n", cgrp->freezer.freeze);
3646 static ssize_t cgroup_freeze_write(struct kernfs_open_file *of,
3647 char *buf, size_t nbytes, loff_t off)
3649 struct cgroup *cgrp;
3653 ret = kstrtoint(strstrip(buf), 0, &freeze);
3657 if (freeze < 0 || freeze > 1)
3660 cgrp = cgroup_kn_lock_live(of->kn, false);
3664 cgroup_freeze(cgrp, freeze);
3666 cgroup_kn_unlock(of->kn);
3671 static int cgroup_file_open(struct kernfs_open_file *of)
3673 struct cftype *cft = of_cft(of);
3676 return cft->open(of);
3680 static void cgroup_file_release(struct kernfs_open_file *of)
3682 struct cftype *cft = of_cft(of);
3688 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3689 size_t nbytes, loff_t off)
3691 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
3692 struct cgroup *cgrp = of->kn->parent->priv;
3693 struct cftype *cft = of_cft(of);
3694 struct cgroup_subsys_state *css;
3701 * If namespaces are delegation boundaries, disallow writes to
3702 * files in an non-init namespace root from inside the namespace
3703 * except for the files explicitly marked delegatable -
3704 * cgroup.procs and cgroup.subtree_control.
3706 if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
3707 !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
3708 ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp)
3712 return cft->write(of, buf, nbytes, off);
3715 * kernfs guarantees that a file isn't deleted with operations in
3716 * flight, which means that the matching css is and stays alive and
3717 * doesn't need to be pinned. The RCU locking is not necessary
3718 * either. It's just for the convenience of using cgroup_css().
3721 css = cgroup_css(cgrp, cft->ss);
3724 if (cft->write_u64) {
3725 unsigned long long v;
3726 ret = kstrtoull(buf, 0, &v);
3728 ret = cft->write_u64(css, cft, v);
3729 } else if (cft->write_s64) {
3731 ret = kstrtoll(buf, 0, &v);
3733 ret = cft->write_s64(css, cft, v);
3738 return ret ?: nbytes;
3741 static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt)
3743 struct cftype *cft = of_cft(of);
3746 return cft->poll(of, pt);
3748 return kernfs_generic_poll(of, pt);
3751 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3753 return seq_cft(seq)->seq_start(seq, ppos);
3756 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3758 return seq_cft(seq)->seq_next(seq, v, ppos);
3761 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3763 if (seq_cft(seq)->seq_stop)
3764 seq_cft(seq)->seq_stop(seq, v);
3767 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3769 struct cftype *cft = seq_cft(m);
3770 struct cgroup_subsys_state *css = seq_css(m);
3773 return cft->seq_show(m, arg);
3776 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3777 else if (cft->read_s64)
3778 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3784 static struct kernfs_ops cgroup_kf_single_ops = {
3785 .atomic_write_len = PAGE_SIZE,
3786 .open = cgroup_file_open,
3787 .release = cgroup_file_release,
3788 .write = cgroup_file_write,
3789 .poll = cgroup_file_poll,
3790 .seq_show = cgroup_seqfile_show,
3793 static struct kernfs_ops cgroup_kf_ops = {
3794 .atomic_write_len = PAGE_SIZE,
3795 .open = cgroup_file_open,
3796 .release = cgroup_file_release,
3797 .write = cgroup_file_write,
3798 .poll = cgroup_file_poll,
3799 .seq_start = cgroup_seqfile_start,
3800 .seq_next = cgroup_seqfile_next,
3801 .seq_stop = cgroup_seqfile_stop,
3802 .seq_show = cgroup_seqfile_show,
3805 /* set uid and gid of cgroup dirs and files to that of the creator */
3806 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3808 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3809 .ia_uid = current_fsuid(),
3810 .ia_gid = current_fsgid(), };
3812 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3813 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3816 return kernfs_setattr(kn, &iattr);
3819 static void cgroup_file_notify_timer(struct timer_list *timer)
3821 cgroup_file_notify(container_of(timer, struct cgroup_file,
3825 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3828 char name[CGROUP_FILE_NAME_MAX];
3829 struct kernfs_node *kn;
3830 struct lock_class_key *key = NULL;
3833 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3834 key = &cft->lockdep_key;
3836 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3837 cgroup_file_mode(cft),
3838 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
3839 0, cft->kf_ops, cft,
3844 ret = cgroup_kn_set_ugid(kn);
3850 if (cft->file_offset) {
3851 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3853 timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0);
3855 spin_lock_irq(&cgroup_file_kn_lock);
3857 spin_unlock_irq(&cgroup_file_kn_lock);
3864 * cgroup_addrm_files - add or remove files to a cgroup directory
3865 * @css: the target css
3866 * @cgrp: the target cgroup (usually css->cgroup)
3867 * @cfts: array of cftypes to be added
3868 * @is_add: whether to add or remove
3870 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3871 * For removals, this function never fails.
3873 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3874 struct cgroup *cgrp, struct cftype cfts[],
3877 struct cftype *cft, *cft_end = NULL;
3880 lockdep_assert_held(&cgroup_mutex);
3883 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3884 /* does cft->flags tell us to skip this file on @cgrp? */
3885 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3887 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3889 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3891 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3893 if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug)
3896 ret = cgroup_add_file(css, cgrp, cft);
3898 pr_warn("%s: failed to add %s, err=%d\n",
3899 __func__, cft->name, ret);
3905 cgroup_rm_file(cgrp, cft);
3911 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3913 struct cgroup_subsys *ss = cfts[0].ss;
3914 struct cgroup *root = &ss->root->cgrp;
3915 struct cgroup_subsys_state *css;
3918 lockdep_assert_held(&cgroup_mutex);
3920 /* add/rm files for all cgroups created before */
3921 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3922 struct cgroup *cgrp = css->cgroup;
3924 if (!(css->flags & CSS_VISIBLE))
3927 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3933 kernfs_activate(root->kn);
3937 static void cgroup_exit_cftypes(struct cftype *cfts)
3941 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3942 /* free copy for custom atomic_write_len, see init_cftypes() */
3943 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3948 /* revert flags set by cgroup core while adding @cfts */
3949 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
3953 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3957 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3958 struct kernfs_ops *kf_ops;
3960 WARN_ON(cft->ss || cft->kf_ops);
3963 kf_ops = &cgroup_kf_ops;
3965 kf_ops = &cgroup_kf_single_ops;
3968 * Ugh... if @cft wants a custom max_write_len, we need to
3969 * make a copy of kf_ops to set its atomic_write_len.
3971 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3972 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3974 cgroup_exit_cftypes(cfts);
3977 kf_ops->atomic_write_len = cft->max_write_len;
3980 cft->kf_ops = kf_ops;
3987 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3989 lockdep_assert_held(&cgroup_mutex);
3991 if (!cfts || !cfts[0].ss)
3994 list_del(&cfts->node);
3995 cgroup_apply_cftypes(cfts, false);
3996 cgroup_exit_cftypes(cfts);
4001 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4002 * @cfts: zero-length name terminated array of cftypes
4004 * Unregister @cfts. Files described by @cfts are removed from all
4005 * existing cgroups and all future cgroups won't have them either. This
4006 * function can be called anytime whether @cfts' subsys is attached or not.
4008 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4011 int cgroup_rm_cftypes(struct cftype *cfts)
4015 mutex_lock(&cgroup_mutex);
4016 ret = cgroup_rm_cftypes_locked(cfts);
4017 mutex_unlock(&cgroup_mutex);
4022 * cgroup_add_cftypes - add an array of cftypes to a subsystem
4023 * @ss: target cgroup subsystem
4024 * @cfts: zero-length name terminated array of cftypes
4026 * Register @cfts to @ss. Files described by @cfts are created for all
4027 * existing cgroups to which @ss is attached and all future cgroups will
4028 * have them too. This function can be called anytime whether @ss is
4031 * Returns 0 on successful registration, -errno on failure. Note that this
4032 * function currently returns 0 as long as @cfts registration is successful
4033 * even if some file creation attempts on existing cgroups fail.
4035 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4039 if (!cgroup_ssid_enabled(ss->id))
4042 if (!cfts || cfts[0].name[0] == '\0')
4045 ret = cgroup_init_cftypes(ss, cfts);
4049 mutex_lock(&cgroup_mutex);
4051 list_add_tail(&cfts->node, &ss->cfts);
4052 ret = cgroup_apply_cftypes(cfts, true);
4054 cgroup_rm_cftypes_locked(cfts);
4056 mutex_unlock(&cgroup_mutex);
4061 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4062 * @ss: target cgroup subsystem
4063 * @cfts: zero-length name terminated array of cftypes
4065 * Similar to cgroup_add_cftypes() but the added files are only used for
4066 * the default hierarchy.
4068 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4072 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4073 cft->flags |= __CFTYPE_ONLY_ON_DFL;
4074 return cgroup_add_cftypes(ss, cfts);
4078 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4079 * @ss: target cgroup subsystem
4080 * @cfts: zero-length name terminated array of cftypes
4082 * Similar to cgroup_add_cftypes() but the added files are only used for
4083 * the legacy hierarchies.
4085 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4089 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4090 cft->flags |= __CFTYPE_NOT_ON_DFL;
4091 return cgroup_add_cftypes(ss, cfts);
4095 * cgroup_file_notify - generate a file modified event for a cgroup_file
4096 * @cfile: target cgroup_file
4098 * @cfile must have been obtained by setting cftype->file_offset.
4100 void cgroup_file_notify(struct cgroup_file *cfile)
4102 unsigned long flags;
4104 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
4106 unsigned long last = cfile->notified_at;
4107 unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV;
4109 if (time_in_range(jiffies, last, next)) {
4110 timer_reduce(&cfile->notify_timer, next);
4112 kernfs_notify(cfile->kn);
4113 cfile->notified_at = jiffies;
4116 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
4120 * css_next_child - find the next child of a given css
4121 * @pos: the current position (%NULL to initiate traversal)
4122 * @parent: css whose children to walk
4124 * This function returns the next child of @parent and should be called
4125 * under either cgroup_mutex or RCU read lock. The only requirement is
4126 * that @parent and @pos are accessible. The next sibling is guaranteed to
4127 * be returned regardless of their states.
4129 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4130 * css which finished ->css_online() is guaranteed to be visible in the
4131 * future iterations and will stay visible until the last reference is put.
4132 * A css which hasn't finished ->css_online() or already finished
4133 * ->css_offline() may show up during traversal. It's each subsystem's
4134 * responsibility to synchronize against on/offlining.
4136 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
4137 struct cgroup_subsys_state *parent)
4139 struct cgroup_subsys_state *next;
4141 cgroup_assert_mutex_or_rcu_locked();
4144 * @pos could already have been unlinked from the sibling list.
4145 * Once a cgroup is removed, its ->sibling.next is no longer
4146 * updated when its next sibling changes. CSS_RELEASED is set when
4147 * @pos is taken off list, at which time its next pointer is valid,
4148 * and, as releases are serialized, the one pointed to by the next
4149 * pointer is guaranteed to not have started release yet. This
4150 * implies that if we observe !CSS_RELEASED on @pos in this RCU
4151 * critical section, the one pointed to by its next pointer is
4152 * guaranteed to not have finished its RCU grace period even if we
4153 * have dropped rcu_read_lock() in-between iterations.
4155 * If @pos has CSS_RELEASED set, its next pointer can't be
4156 * dereferenced; however, as each css is given a monotonically
4157 * increasing unique serial number and always appended to the
4158 * sibling list, the next one can be found by walking the parent's
4159 * children until the first css with higher serial number than
4160 * @pos's. While this path can be slower, it happens iff iteration
4161 * races against release and the race window is very small.
4164 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
4165 } else if (likely(!(pos->flags & CSS_RELEASED))) {
4166 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
4168 list_for_each_entry_rcu(next, &parent->children, sibling,
4169 lockdep_is_held(&cgroup_mutex))
4170 if (next->serial_nr > pos->serial_nr)
4175 * @next, if not pointing to the head, can be dereferenced and is
4178 if (&next->sibling != &parent->children)
4184 * css_next_descendant_pre - find the next descendant for pre-order walk
4185 * @pos: the current position (%NULL to initiate traversal)
4186 * @root: css whose descendants to walk
4188 * To be used by css_for_each_descendant_pre(). Find the next descendant
4189 * to visit for pre-order traversal of @root's descendants. @root is
4190 * included in the iteration and the first node to be visited.
4192 * While this function requires cgroup_mutex or RCU read locking, it
4193 * doesn't require the whole traversal to be contained in a single critical
4194 * section. This function will return the correct next descendant as long
4195 * as both @pos and @root are accessible and @pos is a descendant of @root.
4197 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4198 * css which finished ->css_online() is guaranteed to be visible in the
4199 * future iterations and will stay visible until the last reference is put.
4200 * A css which hasn't finished ->css_online() or already finished
4201 * ->css_offline() may show up during traversal. It's each subsystem's
4202 * responsibility to synchronize against on/offlining.
4204 struct cgroup_subsys_state *
4205 css_next_descendant_pre(struct cgroup_subsys_state *pos,
4206 struct cgroup_subsys_state *root)
4208 struct cgroup_subsys_state *next;
4210 cgroup_assert_mutex_or_rcu_locked();
4212 /* if first iteration, visit @root */
4216 /* visit the first child if exists */
4217 next = css_next_child(NULL, pos);
4221 /* no child, visit my or the closest ancestor's next sibling */
4222 while (pos != root) {
4223 next = css_next_child(pos, pos->parent);
4231 EXPORT_SYMBOL_GPL(css_next_descendant_pre);
4234 * css_rightmost_descendant - return the rightmost descendant of a css
4235 * @pos: css of interest
4237 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4238 * is returned. This can be used during pre-order traversal to skip
4241 * While this function requires cgroup_mutex or RCU read locking, it
4242 * doesn't require the whole traversal to be contained in a single critical
4243 * section. This function will return the correct rightmost descendant as
4244 * long as @pos is accessible.
4246 struct cgroup_subsys_state *
4247 css_rightmost_descendant(struct cgroup_subsys_state *pos)
4249 struct cgroup_subsys_state *last, *tmp;
4251 cgroup_assert_mutex_or_rcu_locked();
4255 /* ->prev isn't RCU safe, walk ->next till the end */
4257 css_for_each_child(tmp, last)
4264 static struct cgroup_subsys_state *
4265 css_leftmost_descendant(struct cgroup_subsys_state *pos)
4267 struct cgroup_subsys_state *last;
4271 pos = css_next_child(NULL, pos);
4278 * css_next_descendant_post - find the next descendant for post-order walk
4279 * @pos: the current position (%NULL to initiate traversal)
4280 * @root: css whose descendants to walk
4282 * To be used by css_for_each_descendant_post(). Find the next descendant
4283 * to visit for post-order traversal of @root's descendants. @root is
4284 * included in the iteration and the last node to be visited.
4286 * While this function requires cgroup_mutex or RCU read locking, it
4287 * doesn't require the whole traversal to be contained in a single critical
4288 * section. This function will return the correct next descendant as long
4289 * as both @pos and @cgroup are accessible and @pos is a descendant of
4292 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4293 * css which finished ->css_online() is guaranteed to be visible in the
4294 * future iterations and will stay visible until the last reference is put.
4295 * A css which hasn't finished ->css_online() or already finished
4296 * ->css_offline() may show up during traversal. It's each subsystem's
4297 * responsibility to synchronize against on/offlining.
4299 struct cgroup_subsys_state *
4300 css_next_descendant_post(struct cgroup_subsys_state *pos,
4301 struct cgroup_subsys_state *root)
4303 struct cgroup_subsys_state *next;
4305 cgroup_assert_mutex_or_rcu_locked();
4307 /* if first iteration, visit leftmost descendant which may be @root */
4309 return css_leftmost_descendant(root);
4311 /* if we visited @root, we're done */
4315 /* if there's an unvisited sibling, visit its leftmost descendant */
4316 next = css_next_child(pos, pos->parent);
4318 return css_leftmost_descendant(next);
4320 /* no sibling left, visit parent */
4325 * css_has_online_children - does a css have online children
4326 * @css: the target css
4328 * Returns %true if @css has any online children; otherwise, %false. This
4329 * function can be called from any context but the caller is responsible
4330 * for synchronizing against on/offlining as necessary.
4332 bool css_has_online_children(struct cgroup_subsys_state *css)
4334 struct cgroup_subsys_state *child;
4338 css_for_each_child(child, css) {
4339 if (child->flags & CSS_ONLINE) {
4348 static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
4350 struct list_head *l;
4351 struct cgrp_cset_link *link;
4352 struct css_set *cset;
4354 lockdep_assert_held(&css_set_lock);
4356 /* find the next threaded cset */
4357 if (it->tcset_pos) {
4358 l = it->tcset_pos->next;
4360 if (l != it->tcset_head) {
4362 return container_of(l, struct css_set,
4363 threaded_csets_node);
4366 it->tcset_pos = NULL;
4369 /* find the next cset */
4372 if (l == it->cset_head) {
4373 it->cset_pos = NULL;
4378 cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
4380 link = list_entry(l, struct cgrp_cset_link, cset_link);
4386 /* initialize threaded css_set walking */
4387 if (it->flags & CSS_TASK_ITER_THREADED) {
4389 put_css_set_locked(it->cur_dcset);
4390 it->cur_dcset = cset;
4393 it->tcset_head = &cset->threaded_csets;
4394 it->tcset_pos = &cset->threaded_csets;
4401 * css_task_iter_advance_css_set - advance a task iterator to the next css_set
4402 * @it: the iterator to advance
4404 * Advance @it to the next css_set to walk.
4406 static void css_task_iter_advance_css_set(struct css_task_iter *it)
4408 struct css_set *cset;
4410 lockdep_assert_held(&css_set_lock);
4412 /* Advance to the next non-empty css_set and find first non-empty tasks list*/
4413 while ((cset = css_task_iter_next_css_set(it))) {
4414 if (!list_empty(&cset->tasks)) {
4415 it->cur_tasks_head = &cset->tasks;
4417 } else if (!list_empty(&cset->mg_tasks)) {
4418 it->cur_tasks_head = &cset->mg_tasks;
4420 } else if (!list_empty(&cset->dying_tasks)) {
4421 it->cur_tasks_head = &cset->dying_tasks;
4426 it->task_pos = NULL;
4429 it->task_pos = it->cur_tasks_head->next;
4432 * We don't keep css_sets locked across iteration steps and thus
4433 * need to take steps to ensure that iteration can be resumed after
4434 * the lock is re-acquired. Iteration is performed at two levels -
4435 * css_sets and tasks in them.
4437 * Once created, a css_set never leaves its cgroup lists, so a
4438 * pinned css_set is guaranteed to stay put and we can resume
4439 * iteration afterwards.
4441 * Tasks may leave @cset across iteration steps. This is resolved
4442 * by registering each iterator with the css_set currently being
4443 * walked and making css_set_move_task() advance iterators whose
4444 * next task is leaving.
4447 list_del(&it->iters_node);
4448 put_css_set_locked(it->cur_cset);
4451 it->cur_cset = cset;
4452 list_add(&it->iters_node, &cset->task_iters);
4455 static void css_task_iter_skip(struct css_task_iter *it,
4456 struct task_struct *task)
4458 lockdep_assert_held(&css_set_lock);
4460 if (it->task_pos == &task->cg_list) {
4461 it->task_pos = it->task_pos->next;
4462 it->flags |= CSS_TASK_ITER_SKIPPED;
4466 static void css_task_iter_advance(struct css_task_iter *it)
4468 struct task_struct *task;
4470 lockdep_assert_held(&css_set_lock);
4474 * Advance iterator to find next entry. We go through cset
4475 * tasks, mg_tasks and dying_tasks, when consumed we move onto
4478 if (it->flags & CSS_TASK_ITER_SKIPPED)
4479 it->flags &= ~CSS_TASK_ITER_SKIPPED;
4481 it->task_pos = it->task_pos->next;
4483 if (it->task_pos == &it->cur_cset->tasks) {
4484 it->cur_tasks_head = &it->cur_cset->mg_tasks;
4485 it->task_pos = it->cur_tasks_head->next;
4487 if (it->task_pos == &it->cur_cset->mg_tasks) {
4488 it->cur_tasks_head = &it->cur_cset->dying_tasks;
4489 it->task_pos = it->cur_tasks_head->next;
4491 if (it->task_pos == &it->cur_cset->dying_tasks)
4492 css_task_iter_advance_css_set(it);
4494 /* called from start, proceed to the first cset */
4495 css_task_iter_advance_css_set(it);
4501 task = list_entry(it->task_pos, struct task_struct, cg_list);
4503 if (it->flags & CSS_TASK_ITER_PROCS) {
4504 /* if PROCS, skip over tasks which aren't group leaders */
4505 if (!thread_group_leader(task))
4508 /* and dying leaders w/o live member threads */
4509 if (it->cur_tasks_head == &it->cur_cset->dying_tasks &&
4510 !atomic_read(&task->signal->live))
4513 /* skip all dying ones */
4514 if (it->cur_tasks_head == &it->cur_cset->dying_tasks)
4520 * css_task_iter_start - initiate task iteration
4521 * @css: the css to walk tasks of
4522 * @flags: CSS_TASK_ITER_* flags
4523 * @it: the task iterator to use
4525 * Initiate iteration through the tasks of @css. The caller can call
4526 * css_task_iter_next() to walk through the tasks until the function
4527 * returns NULL. On completion of iteration, css_task_iter_end() must be
4530 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
4531 struct css_task_iter *it)
4533 memset(it, 0, sizeof(*it));
4535 spin_lock_irq(&css_set_lock);
4541 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4543 it->cset_pos = &css->cgroup->cset_links;
4545 it->cset_head = it->cset_pos;
4547 css_task_iter_advance(it);
4549 spin_unlock_irq(&css_set_lock);
4553 * css_task_iter_next - return the next task for the iterator
4554 * @it: the task iterator being iterated
4556 * The "next" function for task iteration. @it should have been
4557 * initialized via css_task_iter_start(). Returns NULL when the iteration
4560 struct task_struct *css_task_iter_next(struct css_task_iter *it)
4563 put_task_struct(it->cur_task);
4564 it->cur_task = NULL;
4567 spin_lock_irq(&css_set_lock);
4569 /* @it may be half-advanced by skips, finish advancing */
4570 if (it->flags & CSS_TASK_ITER_SKIPPED)
4571 css_task_iter_advance(it);
4574 it->cur_task = list_entry(it->task_pos, struct task_struct,
4576 get_task_struct(it->cur_task);
4577 css_task_iter_advance(it);
4580 spin_unlock_irq(&css_set_lock);
4582 return it->cur_task;
4586 * css_task_iter_end - finish task iteration
4587 * @it: the task iterator to finish
4589 * Finish task iteration started by css_task_iter_start().
4591 void css_task_iter_end(struct css_task_iter *it)
4594 spin_lock_irq(&css_set_lock);
4595 list_del(&it->iters_node);
4596 put_css_set_locked(it->cur_cset);
4597 spin_unlock_irq(&css_set_lock);
4601 put_css_set(it->cur_dcset);
4604 put_task_struct(it->cur_task);
4607 static void cgroup_procs_release(struct kernfs_open_file *of)
4610 css_task_iter_end(of->priv);
4615 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
4617 struct kernfs_open_file *of = s->private;
4618 struct css_task_iter *it = of->priv;
4623 return css_task_iter_next(it);
4626 static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
4627 unsigned int iter_flags)
4629 struct kernfs_open_file *of = s->private;
4630 struct cgroup *cgrp = seq_css(s)->cgroup;
4631 struct css_task_iter *it = of->priv;
4634 * When a seq_file is seeked, it's always traversed sequentially
4635 * from position 0, so we can simply keep iterating on !0 *pos.
4638 if (WARN_ON_ONCE((*pos)))
4639 return ERR_PTR(-EINVAL);
4641 it = kzalloc(sizeof(*it), GFP_KERNEL);
4643 return ERR_PTR(-ENOMEM);
4645 css_task_iter_start(&cgrp->self, iter_flags, it);
4646 } else if (!(*pos)) {
4647 css_task_iter_end(it);
4648 css_task_iter_start(&cgrp->self, iter_flags, it);
4650 return it->cur_task;
4652 return cgroup_procs_next(s, NULL, NULL);
4655 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
4657 struct cgroup *cgrp = seq_css(s)->cgroup;
4660 * All processes of a threaded subtree belong to the domain cgroup
4661 * of the subtree. Only threads can be distributed across the
4662 * subtree. Reject reads on cgroup.procs in the subtree proper.
4663 * They're always empty anyway.
4665 if (cgroup_is_threaded(cgrp))
4666 return ERR_PTR(-EOPNOTSUPP);
4668 return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
4669 CSS_TASK_ITER_THREADED);
4672 static int cgroup_procs_show(struct seq_file *s, void *v)
4674 seq_printf(s, "%d\n", task_pid_vnr(v));
4678 static int cgroup_may_write(const struct cgroup *cgrp, struct super_block *sb)
4681 struct inode *inode;
4683 lockdep_assert_held(&cgroup_mutex);
4685 inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
4689 ret = inode_permission(&init_user_ns, inode, MAY_WRITE);
4694 static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
4695 struct cgroup *dst_cgrp,
4696 struct super_block *sb)
4698 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
4699 struct cgroup *com_cgrp = src_cgrp;
4702 lockdep_assert_held(&cgroup_mutex);
4704 /* find the common ancestor */
4705 while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
4706 com_cgrp = cgroup_parent(com_cgrp);
4708 /* %current should be authorized to migrate to the common ancestor */
4709 ret = cgroup_may_write(com_cgrp, sb);
4714 * If namespaces are delegation boundaries, %current must be able
4715 * to see both source and destination cgroups from its namespace.
4717 if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
4718 (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
4719 !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
4725 static int cgroup_attach_permissions(struct cgroup *src_cgrp,
4726 struct cgroup *dst_cgrp,
4727 struct super_block *sb, bool threadgroup)
4731 ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, sb);
4735 ret = cgroup_migrate_vet_dst(dst_cgrp);
4739 if (!threadgroup && (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp))
4745 static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
4748 struct cgroup *src_cgrp, *dst_cgrp;
4749 struct task_struct *task;
4753 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4757 task = cgroup_procs_write_start(buf, threadgroup, &locked);
4758 ret = PTR_ERR_OR_ZERO(task);
4762 /* find the source cgroup */
4763 spin_lock_irq(&css_set_lock);
4764 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4765 spin_unlock_irq(&css_set_lock);
4767 /* process and thread migrations follow same delegation rule */
4768 ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
4769 of->file->f_path.dentry->d_sb, threadgroup);
4773 ret = cgroup_attach_task(dst_cgrp, task, threadgroup);
4776 cgroup_procs_write_finish(task, locked);
4778 cgroup_kn_unlock(of->kn);
4783 static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
4784 char *buf, size_t nbytes, loff_t off)
4786 return __cgroup_procs_write(of, buf, true) ?: nbytes;
4789 static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
4791 return __cgroup_procs_start(s, pos, 0);
4794 static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
4795 char *buf, size_t nbytes, loff_t off)
4797 return __cgroup_procs_write(of, buf, false) ?: nbytes;
4800 /* cgroup core interface files for the default hierarchy */
4801 static struct cftype cgroup_base_files[] = {
4803 .name = "cgroup.type",
4804 .flags = CFTYPE_NOT_ON_ROOT,
4805 .seq_show = cgroup_type_show,
4806 .write = cgroup_type_write,
4809 .name = "cgroup.procs",
4810 .flags = CFTYPE_NS_DELEGATABLE,
4811 .file_offset = offsetof(struct cgroup, procs_file),
4812 .release = cgroup_procs_release,
4813 .seq_start = cgroup_procs_start,
4814 .seq_next = cgroup_procs_next,
4815 .seq_show = cgroup_procs_show,
4816 .write = cgroup_procs_write,
4819 .name = "cgroup.threads",
4820 .flags = CFTYPE_NS_DELEGATABLE,
4821 .release = cgroup_procs_release,
4822 .seq_start = cgroup_threads_start,
4823 .seq_next = cgroup_procs_next,
4824 .seq_show = cgroup_procs_show,
4825 .write = cgroup_threads_write,
4828 .name = "cgroup.controllers",
4829 .seq_show = cgroup_controllers_show,
4832 .name = "cgroup.subtree_control",
4833 .flags = CFTYPE_NS_DELEGATABLE,
4834 .seq_show = cgroup_subtree_control_show,
4835 .write = cgroup_subtree_control_write,
4838 .name = "cgroup.events",
4839 .flags = CFTYPE_NOT_ON_ROOT,
4840 .file_offset = offsetof(struct cgroup, events_file),
4841 .seq_show = cgroup_events_show,
4844 .name = "cgroup.max.descendants",
4845 .seq_show = cgroup_max_descendants_show,
4846 .write = cgroup_max_descendants_write,
4849 .name = "cgroup.max.depth",
4850 .seq_show = cgroup_max_depth_show,
4851 .write = cgroup_max_depth_write,
4854 .name = "cgroup.stat",
4855 .seq_show = cgroup_stat_show,
4858 .name = "cgroup.freeze",
4859 .flags = CFTYPE_NOT_ON_ROOT,
4860 .seq_show = cgroup_freeze_show,
4861 .write = cgroup_freeze_write,
4865 .seq_show = cpu_stat_show,
4869 .name = "io.pressure",
4870 .seq_show = cgroup_io_pressure_show,
4871 .write = cgroup_io_pressure_write,
4872 .poll = cgroup_pressure_poll,
4873 .release = cgroup_pressure_release,
4876 .name = "memory.pressure",
4877 .seq_show = cgroup_memory_pressure_show,
4878 .write = cgroup_memory_pressure_write,
4879 .poll = cgroup_pressure_poll,
4880 .release = cgroup_pressure_release,
4883 .name = "cpu.pressure",
4884 .seq_show = cgroup_cpu_pressure_show,
4885 .write = cgroup_cpu_pressure_write,
4886 .poll = cgroup_pressure_poll,
4887 .release = cgroup_pressure_release,
4889 #endif /* CONFIG_PSI */
4894 * css destruction is four-stage process.
4896 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4897 * Implemented in kill_css().
4899 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4900 * and thus css_tryget_online() is guaranteed to fail, the css can be
4901 * offlined by invoking offline_css(). After offlining, the base ref is
4902 * put. Implemented in css_killed_work_fn().
4904 * 3. When the percpu_ref reaches zero, the only possible remaining
4905 * accessors are inside RCU read sections. css_release() schedules the
4908 * 4. After the grace period, the css can be freed. Implemented in
4909 * css_free_work_fn().
4911 * It is actually hairier because both step 2 and 4 require process context
4912 * and thus involve punting to css->destroy_work adding two additional
4913 * steps to the already complex sequence.
4915 static void css_free_rwork_fn(struct work_struct *work)
4917 struct cgroup_subsys_state *css = container_of(to_rcu_work(work),
4918 struct cgroup_subsys_state, destroy_rwork);
4919 struct cgroup_subsys *ss = css->ss;
4920 struct cgroup *cgrp = css->cgroup;
4922 percpu_ref_exit(&css->refcnt);
4926 struct cgroup_subsys_state *parent = css->parent;
4930 cgroup_idr_remove(&ss->css_idr, id);
4936 /* cgroup free path */
4937 atomic_dec(&cgrp->root->nr_cgrps);
4938 cgroup1_pidlist_destroy_all(cgrp);
4939 cancel_work_sync(&cgrp->release_agent_work);
4941 if (cgroup_parent(cgrp)) {
4943 * We get a ref to the parent, and put the ref when
4944 * this cgroup is being freed, so it's guaranteed
4945 * that the parent won't be destroyed before its
4948 cgroup_put(cgroup_parent(cgrp));
4949 kernfs_put(cgrp->kn);
4950 psi_cgroup_free(cgrp);
4951 cgroup_rstat_exit(cgrp);
4955 * This is root cgroup's refcnt reaching zero,
4956 * which indicates that the root should be
4959 cgroup_destroy_root(cgrp->root);
4964 static void css_release_work_fn(struct work_struct *work)
4966 struct cgroup_subsys_state *css =
4967 container_of(work, struct cgroup_subsys_state, destroy_work);
4968 struct cgroup_subsys *ss = css->ss;
4969 struct cgroup *cgrp = css->cgroup;
4971 mutex_lock(&cgroup_mutex);
4973 css->flags |= CSS_RELEASED;
4974 list_del_rcu(&css->sibling);
4977 /* css release path */
4978 if (!list_empty(&css->rstat_css_node)) {
4979 cgroup_rstat_flush(cgrp);
4980 list_del_rcu(&css->rstat_css_node);
4983 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
4984 if (ss->css_released)
4985 ss->css_released(css);
4987 struct cgroup *tcgrp;
4989 /* cgroup release path */
4990 TRACE_CGROUP_PATH(release, cgrp);
4992 cgroup_rstat_flush(cgrp);
4994 spin_lock_irq(&css_set_lock);
4995 for (tcgrp = cgroup_parent(cgrp); tcgrp;
4996 tcgrp = cgroup_parent(tcgrp))
4997 tcgrp->nr_dying_descendants--;
4998 spin_unlock_irq(&css_set_lock);
5001 * There are two control paths which try to determine
5002 * cgroup from dentry without going through kernfs -
5003 * cgroupstats_build() and css_tryget_online_from_dir().
5004 * Those are supported by RCU protecting clearing of
5005 * cgrp->kn->priv backpointer.
5008 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
5012 mutex_unlock(&cgroup_mutex);
5014 INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5015 queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5018 static void css_release(struct percpu_ref *ref)
5020 struct cgroup_subsys_state *css =
5021 container_of(ref, struct cgroup_subsys_state, refcnt);
5023 INIT_WORK(&css->destroy_work, css_release_work_fn);
5024 queue_work(cgroup_destroy_wq, &css->destroy_work);
5027 static void init_and_link_css(struct cgroup_subsys_state *css,
5028 struct cgroup_subsys *ss, struct cgroup *cgrp)
5030 lockdep_assert_held(&cgroup_mutex);
5032 cgroup_get_live(cgrp);
5034 memset(css, 0, sizeof(*css));
5038 INIT_LIST_HEAD(&css->sibling);
5039 INIT_LIST_HEAD(&css->children);
5040 INIT_LIST_HEAD(&css->rstat_css_node);
5041 css->serial_nr = css_serial_nr_next++;
5042 atomic_set(&css->online_cnt, 0);
5044 if (cgroup_parent(cgrp)) {
5045 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
5046 css_get(css->parent);
5049 if (ss->css_rstat_flush)
5050 list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list);
5052 BUG_ON(cgroup_css(cgrp, ss));
5055 /* invoke ->css_online() on a new CSS and mark it online if successful */
5056 static int online_css(struct cgroup_subsys_state *css)
5058 struct cgroup_subsys *ss = css->ss;
5061 lockdep_assert_held(&cgroup_mutex);
5064 ret = ss->css_online(css);
5066 css->flags |= CSS_ONLINE;
5067 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
5069 atomic_inc(&css->online_cnt);
5071 atomic_inc(&css->parent->online_cnt);
5076 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5077 static void offline_css(struct cgroup_subsys_state *css)
5079 struct cgroup_subsys *ss = css->ss;
5081 lockdep_assert_held(&cgroup_mutex);
5083 if (!(css->flags & CSS_ONLINE))
5086 if (ss->css_offline)
5087 ss->css_offline(css);
5089 css->flags &= ~CSS_ONLINE;
5090 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
5092 wake_up_all(&css->cgroup->offline_waitq);
5096 * css_create - create a cgroup_subsys_state
5097 * @cgrp: the cgroup new css will be associated with
5098 * @ss: the subsys of new css
5100 * Create a new css associated with @cgrp - @ss pair. On success, the new
5101 * css is online and installed in @cgrp. This function doesn't create the
5102 * interface files. Returns 0 on success, -errno on failure.
5104 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5105 struct cgroup_subsys *ss)
5107 struct cgroup *parent = cgroup_parent(cgrp);
5108 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
5109 struct cgroup_subsys_state *css;
5112 lockdep_assert_held(&cgroup_mutex);
5114 css = ss->css_alloc(parent_css);
5116 css = ERR_PTR(-ENOMEM);
5120 init_and_link_css(css, ss, cgrp);
5122 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
5126 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
5131 /* @css is ready to be brought online now, make it visible */
5132 list_add_tail_rcu(&css->sibling, &parent_css->children);
5133 cgroup_idr_replace(&ss->css_idr, css, css->id);
5135 err = online_css(css);
5142 list_del_rcu(&css->sibling);
5144 list_del_rcu(&css->rstat_css_node);
5145 INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5146 queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5147 return ERR_PTR(err);
5151 * The returned cgroup is fully initialized including its control mask, but
5152 * it isn't associated with its kernfs_node and doesn't have the control
5155 static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
5158 struct cgroup_root *root = parent->root;
5159 struct cgroup *cgrp, *tcgrp;
5160 struct kernfs_node *kn;
5161 int level = parent->level + 1;
5164 /* allocate the cgroup and its ID, 0 is reserved for the root */
5165 cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
5168 return ERR_PTR(-ENOMEM);
5170 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
5174 ret = cgroup_rstat_init(cgrp);
5176 goto out_cancel_ref;
5178 /* create the directory */
5179 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5186 init_cgroup_housekeeping(cgrp);
5188 cgrp->self.parent = &parent->self;
5190 cgrp->level = level;
5192 ret = psi_cgroup_alloc(cgrp);
5194 goto out_kernfs_remove;
5196 ret = cgroup_bpf_inherit(cgrp);
5201 * New cgroup inherits effective freeze counter, and
5202 * if the parent has to be frozen, the child has too.
5204 cgrp->freezer.e_freeze = parent->freezer.e_freeze;
5205 if (cgrp->freezer.e_freeze) {
5207 * Set the CGRP_FREEZE flag, so when a process will be
5208 * attached to the child cgroup, it will become frozen.
5209 * At this point the new cgroup is unpopulated, so we can
5210 * consider it frozen immediately.
5212 set_bit(CGRP_FREEZE, &cgrp->flags);
5213 set_bit(CGRP_FROZEN, &cgrp->flags);
5216 spin_lock_irq(&css_set_lock);
5217 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5218 cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
5220 if (tcgrp != cgrp) {
5221 tcgrp->nr_descendants++;
5224 * If the new cgroup is frozen, all ancestor cgroups
5225 * get a new frozen descendant, but their state can't
5226 * change because of this.
5228 if (cgrp->freezer.e_freeze)
5229 tcgrp->freezer.nr_frozen_descendants++;
5232 spin_unlock_irq(&css_set_lock);
5234 if (notify_on_release(parent))
5235 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5237 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5238 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
5240 cgrp->self.serial_nr = css_serial_nr_next++;
5242 /* allocation complete, commit to creation */
5243 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
5244 atomic_inc(&root->nr_cgrps);
5245 cgroup_get_live(parent);
5248 * On the default hierarchy, a child doesn't automatically inherit
5249 * subtree_control from the parent. Each is configured manually.
5251 if (!cgroup_on_dfl(cgrp))
5252 cgrp->subtree_control = cgroup_control(cgrp);
5254 cgroup_propagate_control(cgrp);
5259 psi_cgroup_free(cgrp);
5261 kernfs_remove(cgrp->kn);
5263 cgroup_rstat_exit(cgrp);
5265 percpu_ref_exit(&cgrp->self.refcnt);
5268 return ERR_PTR(ret);
5271 static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
5273 struct cgroup *cgroup;
5277 lockdep_assert_held(&cgroup_mutex);
5279 for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
5280 if (cgroup->nr_descendants >= cgroup->max_descendants)
5283 if (level > cgroup->max_depth)
5294 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
5296 struct cgroup *parent, *cgrp;
5299 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5300 if (strchr(name, '\n'))
5303 parent = cgroup_kn_lock_live(parent_kn, false);
5307 if (!cgroup_check_hierarchy_limits(parent)) {
5312 cgrp = cgroup_create(parent, name, mode);
5314 ret = PTR_ERR(cgrp);
5319 * This extra ref will be put in cgroup_free_fn() and guarantees
5320 * that @cgrp->kn is always accessible.
5322 kernfs_get(cgrp->kn);
5324 ret = cgroup_kn_set_ugid(cgrp->kn);
5328 ret = css_populate_dir(&cgrp->self);
5332 ret = cgroup_apply_control_enable(cgrp);
5336 TRACE_CGROUP_PATH(mkdir, cgrp);
5338 /* let's create and online css's */
5339 kernfs_activate(cgrp->kn);
5345 cgroup_destroy_locked(cgrp);
5347 cgroup_kn_unlock(parent_kn);
5352 * This is called when the refcnt of a css is confirmed to be killed.
5353 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5354 * initiate destruction and put the css ref from kill_css().
5356 static void css_killed_work_fn(struct work_struct *work)
5358 struct cgroup_subsys_state *css =
5359 container_of(work, struct cgroup_subsys_state, destroy_work);
5361 mutex_lock(&cgroup_mutex);
5366 /* @css can't go away while we're holding cgroup_mutex */
5368 } while (css && atomic_dec_and_test(&css->online_cnt));
5370 mutex_unlock(&cgroup_mutex);
5373 /* css kill confirmation processing requires process context, bounce */
5374 static void css_killed_ref_fn(struct percpu_ref *ref)
5376 struct cgroup_subsys_state *css =
5377 container_of(ref, struct cgroup_subsys_state, refcnt);
5379 if (atomic_dec_and_test(&css->online_cnt)) {
5380 INIT_WORK(&css->destroy_work, css_killed_work_fn);
5381 queue_work(cgroup_destroy_wq, &css->destroy_work);
5386 * kill_css - destroy a css
5387 * @css: css to destroy
5389 * This function initiates destruction of @css by removing cgroup interface
5390 * files and putting its base reference. ->css_offline() will be invoked
5391 * asynchronously once css_tryget_online() is guaranteed to fail and when
5392 * the reference count reaches zero, @css will be released.
5394 static void kill_css(struct cgroup_subsys_state *css)
5396 lockdep_assert_held(&cgroup_mutex);
5398 if (css->flags & CSS_DYING)
5401 css->flags |= CSS_DYING;
5404 * This must happen before css is disassociated with its cgroup.
5405 * See seq_css() for details.
5410 * Killing would put the base ref, but we need to keep it alive
5411 * until after ->css_offline().
5416 * cgroup core guarantees that, by the time ->css_offline() is
5417 * invoked, no new css reference will be given out via
5418 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5419 * proceed to offlining css's because percpu_ref_kill() doesn't
5420 * guarantee that the ref is seen as killed on all CPUs on return.
5422 * Use percpu_ref_kill_and_confirm() to get notifications as each
5423 * css is confirmed to be seen as killed on all CPUs.
5425 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5429 * cgroup_destroy_locked - the first stage of cgroup destruction
5430 * @cgrp: cgroup to be destroyed
5432 * css's make use of percpu refcnts whose killing latency shouldn't be
5433 * exposed to userland and are RCU protected. Also, cgroup core needs to
5434 * guarantee that css_tryget_online() won't succeed by the time
5435 * ->css_offline() is invoked. To satisfy all the requirements,
5436 * destruction is implemented in the following two steps.
5438 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5439 * userland visible parts and start killing the percpu refcnts of
5440 * css's. Set up so that the next stage will be kicked off once all
5441 * the percpu refcnts are confirmed to be killed.
5443 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5444 * rest of destruction. Once all cgroup references are gone, the
5445 * cgroup is RCU-freed.
5447 * This function implements s1. After this step, @cgrp is gone as far as
5448 * the userland is concerned and a new cgroup with the same name may be
5449 * created. As cgroup doesn't care about the names internally, this
5450 * doesn't cause any problem.
5452 static int cgroup_destroy_locked(struct cgroup *cgrp)
5453 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5455 struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
5456 struct cgroup_subsys_state *css;
5457 struct cgrp_cset_link *link;
5460 lockdep_assert_held(&cgroup_mutex);
5463 * Only migration can raise populated from zero and we're already
5464 * holding cgroup_mutex.
5466 if (cgroup_is_populated(cgrp))
5470 * Make sure there's no live children. We can't test emptiness of
5471 * ->self.children as dead children linger on it while being
5472 * drained; otherwise, "rmdir parent/child parent" may fail.
5474 if (css_has_online_children(&cgrp->self))
5478 * Mark @cgrp and the associated csets dead. The former prevents
5479 * further task migration and child creation by disabling
5480 * cgroup_lock_live_group(). The latter makes the csets ignored by
5481 * the migration path.
5483 cgrp->self.flags &= ~CSS_ONLINE;
5485 spin_lock_irq(&css_set_lock);
5486 list_for_each_entry(link, &cgrp->cset_links, cset_link)
5487 link->cset->dead = true;
5488 spin_unlock_irq(&css_set_lock);
5490 /* initiate massacre of all css's */
5491 for_each_css(css, ssid, cgrp)
5494 /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5495 css_clear_dir(&cgrp->self);
5496 kernfs_remove(cgrp->kn);
5498 if (parent && cgroup_is_threaded(cgrp))
5499 parent->nr_threaded_children--;
5501 spin_lock_irq(&css_set_lock);
5502 for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5503 tcgrp->nr_descendants--;
5504 tcgrp->nr_dying_descendants++;
5506 * If the dying cgroup is frozen, decrease frozen descendants
5507 * counters of ancestor cgroups.
5509 if (test_bit(CGRP_FROZEN, &cgrp->flags))
5510 tcgrp->freezer.nr_frozen_descendants--;
5512 spin_unlock_irq(&css_set_lock);
5514 cgroup1_check_for_release(parent);
5516 cgroup_bpf_offline(cgrp);
5518 /* put the base reference */
5519 percpu_ref_kill(&cgrp->self.refcnt);
5524 int cgroup_rmdir(struct kernfs_node *kn)
5526 struct cgroup *cgrp;
5529 cgrp = cgroup_kn_lock_live(kn, false);
5533 ret = cgroup_destroy_locked(cgrp);
5535 TRACE_CGROUP_PATH(rmdir, cgrp);
5537 cgroup_kn_unlock(kn);
5541 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5542 .show_options = cgroup_show_options,
5543 .mkdir = cgroup_mkdir,
5544 .rmdir = cgroup_rmdir,
5545 .show_path = cgroup_show_path,
5548 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5550 struct cgroup_subsys_state *css;
5552 pr_debug("Initializing cgroup subsys %s\n", ss->name);
5554 mutex_lock(&cgroup_mutex);
5556 idr_init(&ss->css_idr);
5557 INIT_LIST_HEAD(&ss->cfts);
5559 /* Create the root cgroup state for this subsystem */
5560 ss->root = &cgrp_dfl_root;
5561 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5562 /* We don't handle early failures gracefully */
5563 BUG_ON(IS_ERR(css));
5564 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5567 * Root csses are never destroyed and we can't initialize
5568 * percpu_ref during early init. Disable refcnting.
5570 css->flags |= CSS_NO_REF;
5573 /* allocation can't be done safely during early init */
5576 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5577 BUG_ON(css->id < 0);
5580 /* Update the init_css_set to contain a subsys
5581 * pointer to this state - since the subsystem is
5582 * newly registered, all tasks and hence the
5583 * init_css_set is in the subsystem's root cgroup. */
5584 init_css_set.subsys[ss->id] = css;
5586 have_fork_callback |= (bool)ss->fork << ss->id;
5587 have_exit_callback |= (bool)ss->exit << ss->id;
5588 have_release_callback |= (bool)ss->release << ss->id;
5589 have_canfork_callback |= (bool)ss->can_fork << ss->id;
5591 /* At system boot, before all subsystems have been
5592 * registered, no tasks have been forked, so we don't
5593 * need to invoke fork callbacks here. */
5594 BUG_ON(!list_empty(&init_task.tasks));
5596 BUG_ON(online_css(css));
5598 mutex_unlock(&cgroup_mutex);
5602 * cgroup_init_early - cgroup initialization at system boot
5604 * Initialize cgroups at system boot, and initialize any
5605 * subsystems that request early init.
5607 int __init cgroup_init_early(void)
5609 static struct cgroup_fs_context __initdata ctx;
5610 struct cgroup_subsys *ss;
5613 ctx.root = &cgrp_dfl_root;
5614 init_cgroup_root(&ctx);
5615 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5617 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5619 for_each_subsys(ss, i) {
5620 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5621 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5622 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5624 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5625 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5628 ss->name = cgroup_subsys_name[i];
5629 if (!ss->legacy_name)
5630 ss->legacy_name = cgroup_subsys_name[i];
5633 cgroup_init_subsys(ss, true);
5639 * cgroup_init - cgroup initialization
5641 * Register cgroup filesystem and /proc file, and initialize
5642 * any subsystems that didn't request early init.
5644 int __init cgroup_init(void)
5646 struct cgroup_subsys *ss;
5649 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5650 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
5651 BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
5653 cgroup_rstat_boot();
5656 * The latency of the synchronize_rcu() is too high for cgroups,
5657 * avoid it at the cost of forcing all readers into the slow path.
5659 rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5661 get_user_ns(init_cgroup_ns.user_ns);
5663 mutex_lock(&cgroup_mutex);
5666 * Add init_css_set to the hash table so that dfl_root can link to
5669 hash_add(css_set_table, &init_css_set.hlist,
5670 css_set_hash(init_css_set.subsys));
5672 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
5674 mutex_unlock(&cgroup_mutex);
5676 for_each_subsys(ss, ssid) {
5677 if (ss->early_init) {
5678 struct cgroup_subsys_state *css =
5679 init_css_set.subsys[ss->id];
5681 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5683 BUG_ON(css->id < 0);
5685 cgroup_init_subsys(ss, false);
5688 list_add_tail(&init_css_set.e_cset_node[ssid],
5689 &cgrp_dfl_root.cgrp.e_csets[ssid]);
5692 * Setting dfl_root subsys_mask needs to consider the
5693 * disabled flag and cftype registration needs kmalloc,
5694 * both of which aren't available during early_init.
5696 if (!cgroup_ssid_enabled(ssid))
5699 if (cgroup1_ssid_disabled(ssid))
5700 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5703 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5705 /* implicit controllers must be threaded too */
5706 WARN_ON(ss->implicit_on_dfl && !ss->threaded);
5708 if (ss->implicit_on_dfl)
5709 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5710 else if (!ss->dfl_cftypes)
5711 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5714 cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
5716 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5717 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5719 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5720 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5724 ss->bind(init_css_set.subsys[ssid]);
5726 mutex_lock(&cgroup_mutex);
5727 css_populate_dir(init_css_set.subsys[ssid]);
5728 mutex_unlock(&cgroup_mutex);
5731 /* init_css_set.subsys[] has been updated, re-hash */
5732 hash_del(&init_css_set.hlist);
5733 hash_add(css_set_table, &init_css_set.hlist,
5734 css_set_hash(init_css_set.subsys));
5736 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5737 WARN_ON(register_filesystem(&cgroup_fs_type));
5738 WARN_ON(register_filesystem(&cgroup2_fs_type));
5739 WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show));
5740 #ifdef CONFIG_CPUSETS
5741 WARN_ON(register_filesystem(&cpuset_fs_type));
5747 static int __init cgroup_wq_init(void)
5750 * There isn't much point in executing destruction path in
5751 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5752 * Use 1 for @max_active.
5754 * We would prefer to do this in cgroup_init() above, but that
5755 * is called before init_workqueues(): so leave this until after.
5757 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5758 BUG_ON(!cgroup_destroy_wq);
5761 core_initcall(cgroup_wq_init);
5763 void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen)
5765 struct kernfs_node *kn;
5767 kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id);
5770 kernfs_path(kn, buf, buflen);
5775 * proc_cgroup_show()
5776 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5777 * - Used for /proc/<pid>/cgroup.
5779 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5780 struct pid *pid, struct task_struct *tsk)
5784 struct cgroup_root *root;
5787 buf = kmalloc(PATH_MAX, GFP_KERNEL);
5791 mutex_lock(&cgroup_mutex);
5792 spin_lock_irq(&css_set_lock);
5794 for_each_root(root) {
5795 struct cgroup_subsys *ss;
5796 struct cgroup *cgrp;
5797 int ssid, count = 0;
5799 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5802 seq_printf(m, "%d:", root->hierarchy_id);
5803 if (root != &cgrp_dfl_root)
5804 for_each_subsys(ss, ssid)
5805 if (root->subsys_mask & (1 << ssid))
5806 seq_printf(m, "%s%s", count++ ? "," : "",
5808 if (strlen(root->name))
5809 seq_printf(m, "%sname=%s", count ? "," : "",
5813 cgrp = task_cgroup_from_root(tsk, root);
5816 * On traditional hierarchies, all zombie tasks show up as
5817 * belonging to the root cgroup. On the default hierarchy,
5818 * while a zombie doesn't show up in "cgroup.procs" and
5819 * thus can't be migrated, its /proc/PID/cgroup keeps
5820 * reporting the cgroup it belonged to before exiting. If
5821 * the cgroup is removed before the zombie is reaped,
5822 * " (deleted)" is appended to the cgroup path.
5824 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
5825 retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5826 current->nsproxy->cgroup_ns);
5827 if (retval >= PATH_MAX)
5828 retval = -ENAMETOOLONG;
5837 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5838 seq_puts(m, " (deleted)\n");
5845 spin_unlock_irq(&css_set_lock);
5846 mutex_unlock(&cgroup_mutex);
5853 * cgroup_fork - initialize cgroup related fields during copy_process()
5854 * @child: pointer to task_struct of forking parent process.
5856 * A task is associated with the init_css_set until cgroup_post_fork()
5857 * attaches it to the target css_set.
5859 void cgroup_fork(struct task_struct *child)
5861 RCU_INIT_POINTER(child->cgroups, &init_css_set);
5862 INIT_LIST_HEAD(&child->cg_list);
5865 static struct cgroup *cgroup_get_from_file(struct file *f)
5867 struct cgroup_subsys_state *css;
5868 struct cgroup *cgrp;
5870 css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
5872 return ERR_CAST(css);
5875 if (!cgroup_on_dfl(cgrp)) {
5877 return ERR_PTR(-EBADF);
5884 * cgroup_css_set_fork - find or create a css_set for a child process
5885 * @kargs: the arguments passed to create the child process
5887 * This functions finds or creates a new css_set which the child
5888 * process will be attached to in cgroup_post_fork(). By default,
5889 * the child process will be given the same css_set as its parent.
5891 * If CLONE_INTO_CGROUP is specified this function will try to find an
5892 * existing css_set which includes the requested cgroup and if not create
5893 * a new css_set that the child will be attached to later. If this function
5894 * succeeds it will hold cgroup_threadgroup_rwsem on return. If
5895 * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
5896 * before grabbing cgroup_threadgroup_rwsem and will hold a reference
5897 * to the target cgroup.
5899 static int cgroup_css_set_fork(struct kernel_clone_args *kargs)
5900 __acquires(&cgroup_mutex) __acquires(&cgroup_threadgroup_rwsem)
5903 struct cgroup *dst_cgrp = NULL;
5904 struct css_set *cset;
5905 struct super_block *sb;
5908 if (kargs->flags & CLONE_INTO_CGROUP)
5909 mutex_lock(&cgroup_mutex);
5911 cgroup_threadgroup_change_begin(current);
5913 spin_lock_irq(&css_set_lock);
5914 cset = task_css_set(current);
5916 spin_unlock_irq(&css_set_lock);
5918 if (!(kargs->flags & CLONE_INTO_CGROUP)) {
5923 f = fget_raw(kargs->cgroup);
5928 sb = f->f_path.dentry->d_sb;
5930 dst_cgrp = cgroup_get_from_file(f);
5931 if (IS_ERR(dst_cgrp)) {
5932 ret = PTR_ERR(dst_cgrp);
5937 if (cgroup_is_dead(dst_cgrp)) {
5943 * Verify that we the target cgroup is writable for us. This is
5944 * usually done by the vfs layer but since we're not going through
5945 * the vfs layer here we need to do it "manually".
5947 ret = cgroup_may_write(dst_cgrp, sb);
5951 ret = cgroup_attach_permissions(cset->dfl_cgrp, dst_cgrp, sb,
5952 !(kargs->flags & CLONE_THREAD));
5956 kargs->cset = find_css_set(cset, dst_cgrp);
5964 kargs->cgrp = dst_cgrp;
5968 cgroup_threadgroup_change_end(current);
5969 mutex_unlock(&cgroup_mutex);
5973 cgroup_put(dst_cgrp);
5976 put_css_set(kargs->cset);
5981 * cgroup_css_set_put_fork - drop references we took during fork
5982 * @kargs: the arguments passed to create the child process
5984 * Drop references to the prepared css_set and target cgroup if
5985 * CLONE_INTO_CGROUP was requested.
5987 static void cgroup_css_set_put_fork(struct kernel_clone_args *kargs)
5988 __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
5990 cgroup_threadgroup_change_end(current);
5992 if (kargs->flags & CLONE_INTO_CGROUP) {
5993 struct cgroup *cgrp = kargs->cgrp;
5994 struct css_set *cset = kargs->cset;
5996 mutex_unlock(&cgroup_mutex);
6011 * cgroup_can_fork - called on a new task before the process is exposed
6012 * @child: the child process
6014 * This prepares a new css_set for the child process which the child will
6015 * be attached to in cgroup_post_fork().
6016 * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
6017 * callback returns an error, the fork aborts with that error code. This
6018 * allows for a cgroup subsystem to conditionally allow or deny new forks.
6020 int cgroup_can_fork(struct task_struct *child, struct kernel_clone_args *kargs)
6022 struct cgroup_subsys *ss;
6025 ret = cgroup_css_set_fork(kargs);
6029 do_each_subsys_mask(ss, i, have_canfork_callback) {
6030 ret = ss->can_fork(child, kargs->cset);
6033 } while_each_subsys_mask();
6038 for_each_subsys(ss, j) {
6041 if (ss->cancel_fork)
6042 ss->cancel_fork(child, kargs->cset);
6045 cgroup_css_set_put_fork(kargs);
6051 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6052 * @child: the child process
6053 * @kargs: the arguments passed to create the child process
6055 * This calls the cancel_fork() callbacks if a fork failed *after*
6056 * cgroup_can_fork() succeeded and cleans up references we took to
6057 * prepare a new css_set for the child process in cgroup_can_fork().
6059 void cgroup_cancel_fork(struct task_struct *child,
6060 struct kernel_clone_args *kargs)
6062 struct cgroup_subsys *ss;
6065 for_each_subsys(ss, i)
6066 if (ss->cancel_fork)
6067 ss->cancel_fork(child, kargs->cset);
6069 cgroup_css_set_put_fork(kargs);
6073 * cgroup_post_fork - finalize cgroup setup for the child process
6074 * @child: the child process
6076 * Attach the child process to its css_set calling the subsystem fork()
6079 void cgroup_post_fork(struct task_struct *child,
6080 struct kernel_clone_args *kargs)
6081 __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6083 struct cgroup_subsys *ss;
6084 struct css_set *cset;
6090 spin_lock_irq(&css_set_lock);
6092 /* init tasks are special, only link regular threads */
6093 if (likely(child->pid)) {
6094 WARN_ON_ONCE(!list_empty(&child->cg_list));
6096 css_set_move_task(child, NULL, cset, false);
6103 * If the cgroup has to be frozen, the new task has too. Let's set
6104 * the JOBCTL_TRAP_FREEZE jobctl bit to get the task into the
6107 if (unlikely(cgroup_task_freeze(child))) {
6108 spin_lock(&child->sighand->siglock);
6109 WARN_ON_ONCE(child->frozen);
6110 child->jobctl |= JOBCTL_TRAP_FREEZE;
6111 spin_unlock(&child->sighand->siglock);
6114 * Calling cgroup_update_frozen() isn't required here,
6115 * because it will be called anyway a bit later from
6116 * do_freezer_trap(). So we avoid cgroup's transient switch
6117 * from the frozen state and back.
6121 spin_unlock_irq(&css_set_lock);
6124 * Call ss->fork(). This must happen after @child is linked on
6125 * css_set; otherwise, @child might change state between ->fork()
6126 * and addition to css_set.
6128 do_each_subsys_mask(ss, i, have_fork_callback) {
6130 } while_each_subsys_mask();
6132 /* Make the new cset the root_cset of the new cgroup namespace. */
6133 if (kargs->flags & CLONE_NEWCGROUP) {
6134 struct css_set *rcset = child->nsproxy->cgroup_ns->root_cset;
6137 child->nsproxy->cgroup_ns->root_cset = cset;
6141 cgroup_css_set_put_fork(kargs);
6145 * cgroup_exit - detach cgroup from exiting task
6146 * @tsk: pointer to task_struct of exiting process
6148 * Description: Detach cgroup from @tsk.
6151 void cgroup_exit(struct task_struct *tsk)
6153 struct cgroup_subsys *ss;
6154 struct css_set *cset;
6157 spin_lock_irq(&css_set_lock);
6159 WARN_ON_ONCE(list_empty(&tsk->cg_list));
6160 cset = task_css_set(tsk);
6161 css_set_move_task(tsk, cset, NULL, false);
6162 list_add_tail(&tsk->cg_list, &cset->dying_tasks);
6165 WARN_ON_ONCE(cgroup_task_frozen(tsk));
6166 if (unlikely(cgroup_task_freeze(tsk)))
6167 cgroup_update_frozen(task_dfl_cgroup(tsk));
6169 spin_unlock_irq(&css_set_lock);
6171 /* see cgroup_post_fork() for details */
6172 do_each_subsys_mask(ss, i, have_exit_callback) {
6174 } while_each_subsys_mask();
6177 void cgroup_release(struct task_struct *task)
6179 struct cgroup_subsys *ss;
6182 do_each_subsys_mask(ss, ssid, have_release_callback) {
6184 } while_each_subsys_mask();
6186 spin_lock_irq(&css_set_lock);
6187 css_set_skip_task_iters(task_css_set(task), task);
6188 list_del_init(&task->cg_list);
6189 spin_unlock_irq(&css_set_lock);
6192 void cgroup_free(struct task_struct *task)
6194 struct css_set *cset = task_css_set(task);
6198 static int __init cgroup_disable(char *str)
6200 struct cgroup_subsys *ss;
6204 while ((token = strsep(&str, ",")) != NULL) {
6208 for_each_subsys(ss, i) {
6209 if (strcmp(token, ss->name) &&
6210 strcmp(token, ss->legacy_name))
6213 static_branch_disable(cgroup_subsys_enabled_key[i]);
6214 pr_info("Disabling %s control group subsystem\n",
6220 __setup("cgroup_disable=", cgroup_disable);
6222 void __init __weak enable_debug_cgroup(void) { }
6224 static int __init enable_cgroup_debug(char *str)
6226 cgroup_debug = true;
6227 enable_debug_cgroup();
6230 __setup("cgroup_debug", enable_cgroup_debug);
6233 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6234 * @dentry: directory dentry of interest
6235 * @ss: subsystem of interest
6237 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6238 * to get the corresponding css and return it. If such css doesn't exist
6239 * or can't be pinned, an ERR_PTR value is returned.
6241 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
6242 struct cgroup_subsys *ss)
6244 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
6245 struct file_system_type *s_type = dentry->d_sb->s_type;
6246 struct cgroup_subsys_state *css = NULL;
6247 struct cgroup *cgrp;
6249 /* is @dentry a cgroup dir? */
6250 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
6251 !kn || kernfs_type(kn) != KERNFS_DIR)
6252 return ERR_PTR(-EBADF);
6257 * This path doesn't originate from kernfs and @kn could already
6258 * have been or be removed at any point. @kn->priv is RCU
6259 * protected for this access. See css_release_work_fn() for details.
6261 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6263 css = cgroup_css(cgrp, ss);
6265 if (!css || !css_tryget_online(css))
6266 css = ERR_PTR(-ENOENT);
6273 * css_from_id - lookup css by id
6274 * @id: the cgroup id
6275 * @ss: cgroup subsys to be looked into
6277 * Returns the css if there's valid one with @id, otherwise returns NULL.
6278 * Should be called under rcu_read_lock().
6280 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
6282 WARN_ON_ONCE(!rcu_read_lock_held());
6283 return idr_find(&ss->css_idr, id);
6287 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6288 * @path: path on the default hierarchy
6290 * Find the cgroup at @path on the default hierarchy, increment its
6291 * reference count and return it. Returns pointer to the found cgroup on
6292 * success, ERR_PTR(-ENOENT) if @path doesn't exist and ERR_PTR(-ENOTDIR)
6293 * if @path points to a non-directory.
6295 struct cgroup *cgroup_get_from_path(const char *path)
6297 struct kernfs_node *kn;
6298 struct cgroup *cgrp;
6300 mutex_lock(&cgroup_mutex);
6302 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6304 if (kernfs_type(kn) == KERNFS_DIR) {
6306 cgroup_get_live(cgrp);
6308 cgrp = ERR_PTR(-ENOTDIR);
6312 cgrp = ERR_PTR(-ENOENT);
6315 mutex_unlock(&cgroup_mutex);
6318 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6321 * cgroup_get_from_fd - get a cgroup pointer from a fd
6322 * @fd: fd obtained by open(cgroup2_dir)
6324 * Find the cgroup from a fd which should be obtained
6325 * by opening a cgroup directory. Returns a pointer to the
6326 * cgroup on success. ERR_PTR is returned if the cgroup
6329 struct cgroup *cgroup_get_from_fd(int fd)
6331 struct cgroup *cgrp;
6336 return ERR_PTR(-EBADF);
6338 cgrp = cgroup_get_from_file(f);
6342 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
6344 static u64 power_of_ten(int power)
6353 * cgroup_parse_float - parse a floating number
6354 * @input: input string
6355 * @dec_shift: number of decimal digits to shift
6358 * Parse a decimal floating point number in @input and store the result in
6359 * @v with decimal point right shifted @dec_shift times. For example, if
6360 * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6361 * Returns 0 on success, -errno otherwise.
6363 * There's nothing cgroup specific about this function except that it's
6364 * currently the only user.
6366 int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v)
6368 s64 whole, frac = 0;
6369 int fstart = 0, fend = 0, flen;
6371 if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend))
6376 flen = fend > fstart ? fend - fstart : 0;
6377 if (flen < dec_shift)
6378 frac *= power_of_ten(dec_shift - flen);
6380 frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift));
6382 *v = whole * power_of_ten(dec_shift) + frac;
6387 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6388 * definition in cgroup-defs.h.
6390 #ifdef CONFIG_SOCK_CGROUP_DATA
6392 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6394 DEFINE_SPINLOCK(cgroup_sk_update_lock);
6395 static bool cgroup_sk_alloc_disabled __read_mostly;
6397 void cgroup_sk_alloc_disable(void)
6399 if (cgroup_sk_alloc_disabled)
6401 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6402 cgroup_sk_alloc_disabled = true;
6407 #define cgroup_sk_alloc_disabled false
6411 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6413 if (cgroup_sk_alloc_disabled) {
6414 skcd->no_refcnt = 1;
6418 /* Don't associate the sock with unrelated interrupted task's cgroup. */
6425 struct css_set *cset;
6427 cset = task_css_set(current);
6428 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6429 skcd->val = (unsigned long)cset->dfl_cgrp;
6430 cgroup_bpf_get(cset->dfl_cgrp);
6439 void cgroup_sk_clone(struct sock_cgroup_data *skcd)
6442 if (skcd->no_refcnt)
6445 * We might be cloning a socket which is left in an empty
6446 * cgroup and the cgroup might have already been rmdir'd.
6447 * Don't use cgroup_get_live().
6449 cgroup_get(sock_cgroup_ptr(skcd));
6450 cgroup_bpf_get(sock_cgroup_ptr(skcd));
6454 void cgroup_sk_free(struct sock_cgroup_data *skcd)
6456 struct cgroup *cgrp = sock_cgroup_ptr(skcd);
6458 if (skcd->no_refcnt)
6460 cgroup_bpf_put(cgrp);
6464 #endif /* CONFIG_SOCK_CGROUP_DATA */
6466 #ifdef CONFIG_CGROUP_BPF
6467 int cgroup_bpf_attach(struct cgroup *cgrp,
6468 struct bpf_prog *prog, struct bpf_prog *replace_prog,
6469 struct bpf_cgroup_link *link,
6470 enum bpf_attach_type type,
6475 mutex_lock(&cgroup_mutex);
6476 ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags);
6477 mutex_unlock(&cgroup_mutex);
6481 int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
6482 enum bpf_attach_type type)
6486 mutex_lock(&cgroup_mutex);
6487 ret = __cgroup_bpf_detach(cgrp, prog, NULL, type);
6488 mutex_unlock(&cgroup_mutex);
6492 int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
6493 union bpf_attr __user *uattr)
6497 mutex_lock(&cgroup_mutex);
6498 ret = __cgroup_bpf_query(cgrp, attr, uattr);
6499 mutex_unlock(&cgroup_mutex);
6502 #endif /* CONFIG_CGROUP_BPF */
6505 static ssize_t show_delegatable_files(struct cftype *files, char *buf,
6506 ssize_t size, const char *prefix)
6511 for (cft = files; cft && cft->name[0] != '\0'; cft++) {
6512 if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
6516 ret += snprintf(buf + ret, size - ret, "%s.", prefix);
6518 ret += snprintf(buf + ret, size - ret, "%s\n", cft->name);
6520 if (WARN_ON(ret >= size))
6527 static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr,
6530 struct cgroup_subsys *ss;
6534 ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret,
6537 for_each_subsys(ss, ssid)
6538 ret += show_delegatable_files(ss->dfl_cftypes, buf + ret,
6540 cgroup_subsys_name[ssid]);
6544 static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate);
6546 static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr,
6549 return snprintf(buf, PAGE_SIZE,
6551 "memory_localevents\n"
6552 "memory_recursiveprot\n");
6554 static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features);
6556 static struct attribute *cgroup_sysfs_attrs[] = {
6557 &cgroup_delegate_attr.attr,
6558 &cgroup_features_attr.attr,
6562 static const struct attribute_group cgroup_sysfs_attr_group = {
6563 .attrs = cgroup_sysfs_attrs,
6567 static int __init cgroup_sysfs_init(void)
6569 return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group);
6571 subsys_initcall(cgroup_sysfs_init);
6573 #endif /* CONFIG_SYSFS */