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 .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 differnetly 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 inbetween 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 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
287 * - blkcg: blk-throttle becomes properly hierarchical.
289 * - debug: disallowed on the default hierarchy.
291 bool cgroup_on_dfl(const struct cgroup *cgrp)
293 return cgrp->root == &cgrp_dfl_root;
296 /* IDR wrappers which synchronize using cgroup_idr_lock */
297 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
302 idr_preload(gfp_mask);
303 spin_lock_bh(&cgroup_idr_lock);
304 ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
305 spin_unlock_bh(&cgroup_idr_lock);
310 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
314 spin_lock_bh(&cgroup_idr_lock);
315 ret = idr_replace(idr, ptr, id);
316 spin_unlock_bh(&cgroup_idr_lock);
320 static void cgroup_idr_remove(struct idr *idr, int id)
322 spin_lock_bh(&cgroup_idr_lock);
324 spin_unlock_bh(&cgroup_idr_lock);
327 static bool cgroup_has_tasks(struct cgroup *cgrp)
329 return cgrp->nr_populated_csets;
332 bool cgroup_is_threaded(struct cgroup *cgrp)
334 return cgrp->dom_cgrp != cgrp;
337 /* can @cgrp host both domain and threaded children? */
338 static bool cgroup_is_mixable(struct cgroup *cgrp)
341 * Root isn't under domain level resource control exempting it from
342 * the no-internal-process constraint, so it can serve as a thread
343 * root and a parent of resource domains at the same time.
345 return !cgroup_parent(cgrp);
348 /* can @cgrp become a thread root? should always be true for a thread root */
349 static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
351 /* mixables don't care */
352 if (cgroup_is_mixable(cgrp))
355 /* domain roots can't be nested under threaded */
356 if (cgroup_is_threaded(cgrp))
359 /* can only have either domain or threaded children */
360 if (cgrp->nr_populated_domain_children)
363 /* and no domain controllers can be enabled */
364 if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
370 /* is @cgrp root of a threaded subtree? */
371 bool cgroup_is_thread_root(struct cgroup *cgrp)
373 /* thread root should be a domain */
374 if (cgroup_is_threaded(cgrp))
377 /* a domain w/ threaded children is a thread root */
378 if (cgrp->nr_threaded_children)
382 * A domain which has tasks and explicit threaded controllers
383 * enabled is a thread root.
385 if (cgroup_has_tasks(cgrp) &&
386 (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
392 /* a domain which isn't connected to the root w/o brekage can't be used */
393 static bool cgroup_is_valid_domain(struct cgroup *cgrp)
395 /* the cgroup itself can be a thread root */
396 if (cgroup_is_threaded(cgrp))
399 /* but the ancestors can't be unless mixable */
400 while ((cgrp = cgroup_parent(cgrp))) {
401 if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
403 if (cgroup_is_threaded(cgrp))
410 /* subsystems visibly enabled on a cgroup */
411 static u16 cgroup_control(struct cgroup *cgrp)
413 struct cgroup *parent = cgroup_parent(cgrp);
414 u16 root_ss_mask = cgrp->root->subsys_mask;
417 u16 ss_mask = parent->subtree_control;
419 /* threaded cgroups can only have threaded controllers */
420 if (cgroup_is_threaded(cgrp))
421 ss_mask &= cgrp_dfl_threaded_ss_mask;
425 if (cgroup_on_dfl(cgrp))
426 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
427 cgrp_dfl_implicit_ss_mask);
431 /* subsystems enabled on a cgroup */
432 static u16 cgroup_ss_mask(struct cgroup *cgrp)
434 struct cgroup *parent = cgroup_parent(cgrp);
437 u16 ss_mask = parent->subtree_ss_mask;
439 /* threaded cgroups can only have threaded controllers */
440 if (cgroup_is_threaded(cgrp))
441 ss_mask &= cgrp_dfl_threaded_ss_mask;
445 return cgrp->root->subsys_mask;
449 * cgroup_css - obtain a cgroup's css for the specified subsystem
450 * @cgrp: the cgroup of interest
451 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
453 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
454 * function must be called either under cgroup_mutex or rcu_read_lock() and
455 * the caller is responsible for pinning the returned css if it wants to
456 * keep accessing it outside the said locks. This function may return
457 * %NULL if @cgrp doesn't have @subsys_id enabled.
459 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
460 struct cgroup_subsys *ss)
463 return rcu_dereference_check(cgrp->subsys[ss->id],
464 lockdep_is_held(&cgroup_mutex));
470 * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
471 * @cgrp: the cgroup of interest
472 * @ss: the subsystem of interest
474 * Find and get @cgrp's css assocaited with @ss. If the css doesn't exist
475 * or is offline, %NULL is returned.
477 static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp,
478 struct cgroup_subsys *ss)
480 struct cgroup_subsys_state *css;
483 css = cgroup_css(cgrp, ss);
484 if (css && !css_tryget_online(css))
492 * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
493 * @cgrp: the cgroup of interest
494 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
496 * Similar to cgroup_css() but returns the effective css, which is defined
497 * as the matching css of the nearest ancestor including self which has @ss
498 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
499 * function is guaranteed to return non-NULL css.
501 static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp,
502 struct cgroup_subsys *ss)
504 lockdep_assert_held(&cgroup_mutex);
510 * This function is used while updating css associations and thus
511 * can't test the csses directly. Test ss_mask.
513 while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
514 cgrp = cgroup_parent(cgrp);
519 return cgroup_css(cgrp, ss);
523 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
524 * @cgrp: the cgroup of interest
525 * @ss: the subsystem of interest
527 * Find and get the effective css of @cgrp for @ss. The effective css is
528 * defined as the matching css of the nearest ancestor including self which
529 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
530 * the root css is returned, so this function always returns a valid css.
532 * The returned css is not guaranteed to be online, and therefore it is the
533 * callers responsiblity to tryget a reference for it.
535 struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
536 struct cgroup_subsys *ss)
538 struct cgroup_subsys_state *css;
541 css = cgroup_css(cgrp, ss);
545 cgrp = cgroup_parent(cgrp);
548 return init_css_set.subsys[ss->id];
552 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
553 * @cgrp: the cgroup of interest
554 * @ss: the subsystem of interest
556 * Find and get the effective css of @cgrp for @ss. The effective css is
557 * defined as the matching css of the nearest ancestor including self which
558 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
559 * the root css is returned, so this function always returns a valid css.
560 * The returned css must be put using css_put().
562 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
563 struct cgroup_subsys *ss)
565 struct cgroup_subsys_state *css;
570 css = cgroup_css(cgrp, ss);
572 if (css && css_tryget_online(css))
574 cgrp = cgroup_parent(cgrp);
577 css = init_css_set.subsys[ss->id];
584 static void cgroup_get_live(struct cgroup *cgrp)
586 WARN_ON_ONCE(cgroup_is_dead(cgrp));
587 css_get(&cgrp->self);
591 * __cgroup_task_count - count the number of tasks in a cgroup. The caller
592 * is responsible for taking the css_set_lock.
593 * @cgrp: the cgroup in question
595 int __cgroup_task_count(const struct cgroup *cgrp)
598 struct cgrp_cset_link *link;
600 lockdep_assert_held(&css_set_lock);
602 list_for_each_entry(link, &cgrp->cset_links, cset_link)
603 count += link->cset->nr_tasks;
609 * cgroup_task_count - count the number of tasks in a cgroup.
610 * @cgrp: the cgroup in question
612 int cgroup_task_count(const struct cgroup *cgrp)
616 spin_lock_irq(&css_set_lock);
617 count = __cgroup_task_count(cgrp);
618 spin_unlock_irq(&css_set_lock);
623 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
625 struct cgroup *cgrp = of->kn->parent->priv;
626 struct cftype *cft = of_cft(of);
629 * This is open and unprotected implementation of cgroup_css().
630 * seq_css() is only called from a kernfs file operation which has
631 * an active reference on the file. Because all the subsystem
632 * files are drained before a css is disassociated with a cgroup,
633 * the matching css from the cgroup's subsys table is guaranteed to
634 * be and stay valid until the enclosing operation is complete.
637 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
641 EXPORT_SYMBOL_GPL(of_css);
644 * for_each_css - iterate all css's of a cgroup
645 * @css: the iteration cursor
646 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
647 * @cgrp: the target cgroup to iterate css's of
649 * Should be called under cgroup_[tree_]mutex.
651 #define for_each_css(css, ssid, cgrp) \
652 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
653 if (!((css) = rcu_dereference_check( \
654 (cgrp)->subsys[(ssid)], \
655 lockdep_is_held(&cgroup_mutex)))) { } \
659 * for_each_e_css - iterate all effective css's of a cgroup
660 * @css: the iteration cursor
661 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
662 * @cgrp: the target cgroup to iterate css's of
664 * Should be called under cgroup_[tree_]mutex.
666 #define for_each_e_css(css, ssid, cgrp) \
667 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
668 if (!((css) = cgroup_e_css_by_mask(cgrp, \
669 cgroup_subsys[(ssid)]))) \
674 * do_each_subsys_mask - filter for_each_subsys with a bitmask
675 * @ss: the iteration cursor
676 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
677 * @ss_mask: the bitmask
679 * The block will only run for cases where the ssid-th bit (1 << ssid) of
682 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
683 unsigned long __ss_mask = (ss_mask); \
684 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
688 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
689 (ss) = cgroup_subsys[ssid]; \
692 #define while_each_subsys_mask() \
697 /* iterate over child cgrps, lock should be held throughout iteration */
698 #define cgroup_for_each_live_child(child, cgrp) \
699 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
700 if (({ lockdep_assert_held(&cgroup_mutex); \
701 cgroup_is_dead(child); })) \
705 /* walk live descendants in preorder */
706 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
707 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
708 if (({ lockdep_assert_held(&cgroup_mutex); \
709 (dsct) = (d_css)->cgroup; \
710 cgroup_is_dead(dsct); })) \
714 /* walk live descendants in postorder */
715 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
716 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
717 if (({ lockdep_assert_held(&cgroup_mutex); \
718 (dsct) = (d_css)->cgroup; \
719 cgroup_is_dead(dsct); })) \
724 * The default css_set - used by init and its children prior to any
725 * hierarchies being mounted. It contains a pointer to the root state
726 * for each subsystem. Also used to anchor the list of css_sets. Not
727 * reference-counted, to improve performance when child cgroups
728 * haven't been created.
730 struct css_set init_css_set = {
731 .refcount = REFCOUNT_INIT(1),
732 .dom_cset = &init_css_set,
733 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
734 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
735 .dying_tasks = LIST_HEAD_INIT(init_css_set.dying_tasks),
736 .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
737 .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets),
738 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
739 .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
740 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
743 * The following field is re-initialized when this cset gets linked
744 * in cgroup_init(). However, let's initialize the field
745 * statically too so that the default cgroup can be accessed safely
748 .dfl_cgrp = &cgrp_dfl_root.cgrp,
751 static int css_set_count = 1; /* 1 for init_css_set */
753 static bool css_set_threaded(struct css_set *cset)
755 return cset->dom_cset != cset;
759 * css_set_populated - does a css_set contain any tasks?
760 * @cset: target css_set
762 * css_set_populated() should be the same as !!cset->nr_tasks at steady
763 * state. However, css_set_populated() can be called while a task is being
764 * added to or removed from the linked list before the nr_tasks is
765 * properly updated. Hence, we can't just look at ->nr_tasks here.
767 static bool css_set_populated(struct css_set *cset)
769 lockdep_assert_held(&css_set_lock);
771 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
775 * cgroup_update_populated - update the populated count of a cgroup
776 * @cgrp: the target cgroup
777 * @populated: inc or dec populated count
779 * One of the css_sets associated with @cgrp is either getting its first
780 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
781 * count is propagated towards root so that a given cgroup's
782 * nr_populated_children is zero iff none of its descendants contain any
785 * @cgrp's interface file "cgroup.populated" is zero if both
786 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
787 * 1 otherwise. When the sum changes from or to zero, userland is notified
788 * that the content of the interface file has changed. This can be used to
789 * detect when @cgrp and its descendants become populated or empty.
791 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
793 struct cgroup *child = NULL;
794 int adj = populated ? 1 : -1;
796 lockdep_assert_held(&css_set_lock);
799 bool was_populated = cgroup_is_populated(cgrp);
802 cgrp->nr_populated_csets += adj;
804 if (cgroup_is_threaded(child))
805 cgrp->nr_populated_threaded_children += adj;
807 cgrp->nr_populated_domain_children += adj;
810 if (was_populated == cgroup_is_populated(cgrp))
813 cgroup1_check_for_release(cgrp);
814 TRACE_CGROUP_PATH(notify_populated, cgrp,
815 cgroup_is_populated(cgrp));
816 cgroup_file_notify(&cgrp->events_file);
819 cgrp = cgroup_parent(cgrp);
824 * css_set_update_populated - update populated state of a css_set
825 * @cset: target css_set
826 * @populated: whether @cset is populated or depopulated
828 * @cset is either getting the first task or losing the last. Update the
829 * populated counters of all associated cgroups accordingly.
831 static void css_set_update_populated(struct css_set *cset, bool populated)
833 struct cgrp_cset_link *link;
835 lockdep_assert_held(&css_set_lock);
837 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
838 cgroup_update_populated(link->cgrp, populated);
842 * @task is leaving, advance task iterators which are pointing to it so
843 * that they can resume at the next position. Advancing an iterator might
844 * remove it from the list, use safe walk. See css_task_iter_skip() for
847 static void css_set_skip_task_iters(struct css_set *cset,
848 struct task_struct *task)
850 struct css_task_iter *it, *pos;
852 list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node)
853 css_task_iter_skip(it, task);
857 * css_set_move_task - move a task from one css_set to another
858 * @task: task being moved
859 * @from_cset: css_set @task currently belongs to (may be NULL)
860 * @to_cset: new css_set @task is being moved to (may be NULL)
861 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
863 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
864 * css_set, @from_cset can be NULL. If @task is being disassociated
865 * instead of moved, @to_cset can be NULL.
867 * This function automatically handles populated counter updates and
868 * css_task_iter adjustments but the caller is responsible for managing
869 * @from_cset and @to_cset's reference counts.
871 static void css_set_move_task(struct task_struct *task,
872 struct css_set *from_cset, struct css_set *to_cset,
875 lockdep_assert_held(&css_set_lock);
877 if (to_cset && !css_set_populated(to_cset))
878 css_set_update_populated(to_cset, true);
881 WARN_ON_ONCE(list_empty(&task->cg_list));
883 css_set_skip_task_iters(from_cset, task);
884 list_del_init(&task->cg_list);
885 if (!css_set_populated(from_cset))
886 css_set_update_populated(from_cset, false);
888 WARN_ON_ONCE(!list_empty(&task->cg_list));
893 * We are synchronized through cgroup_threadgroup_rwsem
894 * against PF_EXITING setting such that we can't race
895 * against cgroup_exit()/cgroup_free() dropping the css_set.
897 WARN_ON_ONCE(task->flags & PF_EXITING);
899 cgroup_move_task(task, to_cset);
900 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
906 * hash table for cgroup groups. This improves the performance to find
907 * an existing css_set. This hash doesn't (currently) take into
908 * account cgroups in empty hierarchies.
910 #define CSS_SET_HASH_BITS 7
911 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
913 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
915 unsigned long key = 0UL;
916 struct cgroup_subsys *ss;
919 for_each_subsys(ss, i)
920 key += (unsigned long)css[i];
921 key = (key >> 16) ^ key;
926 void put_css_set_locked(struct css_set *cset)
928 struct cgrp_cset_link *link, *tmp_link;
929 struct cgroup_subsys *ss;
932 lockdep_assert_held(&css_set_lock);
934 if (!refcount_dec_and_test(&cset->refcount))
937 WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
939 /* This css_set is dead. unlink it and release cgroup and css refs */
940 for_each_subsys(ss, ssid) {
941 list_del(&cset->e_cset_node[ssid]);
942 css_put(cset->subsys[ssid]);
944 hash_del(&cset->hlist);
947 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
948 list_del(&link->cset_link);
949 list_del(&link->cgrp_link);
950 if (cgroup_parent(link->cgrp))
951 cgroup_put(link->cgrp);
955 if (css_set_threaded(cset)) {
956 list_del(&cset->threaded_csets_node);
957 put_css_set_locked(cset->dom_cset);
960 kfree_rcu(cset, rcu_head);
964 * compare_css_sets - helper function for find_existing_css_set().
965 * @cset: candidate css_set being tested
966 * @old_cset: existing css_set for a task
967 * @new_cgrp: cgroup that's being entered by the task
968 * @template: desired set of css pointers in css_set (pre-calculated)
970 * Returns true if "cset" matches "old_cset" except for the hierarchy
971 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
973 static bool compare_css_sets(struct css_set *cset,
974 struct css_set *old_cset,
975 struct cgroup *new_cgrp,
976 struct cgroup_subsys_state *template[])
978 struct cgroup *new_dfl_cgrp;
979 struct list_head *l1, *l2;
982 * On the default hierarchy, there can be csets which are
983 * associated with the same set of cgroups but different csses.
984 * Let's first ensure that csses match.
986 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
990 /* @cset's domain should match the default cgroup's */
991 if (cgroup_on_dfl(new_cgrp))
992 new_dfl_cgrp = new_cgrp;
994 new_dfl_cgrp = old_cset->dfl_cgrp;
996 if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
1000 * Compare cgroup pointers in order to distinguish between
1001 * different cgroups in hierarchies. As different cgroups may
1002 * share the same effective css, this comparison is always
1005 l1 = &cset->cgrp_links;
1006 l2 = &old_cset->cgrp_links;
1008 struct cgrp_cset_link *link1, *link2;
1009 struct cgroup *cgrp1, *cgrp2;
1013 /* See if we reached the end - both lists are equal length. */
1014 if (l1 == &cset->cgrp_links) {
1015 BUG_ON(l2 != &old_cset->cgrp_links);
1018 BUG_ON(l2 == &old_cset->cgrp_links);
1020 /* Locate the cgroups associated with these links. */
1021 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
1022 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
1023 cgrp1 = link1->cgrp;
1024 cgrp2 = link2->cgrp;
1025 /* Hierarchies should be linked in the same order. */
1026 BUG_ON(cgrp1->root != cgrp2->root);
1029 * If this hierarchy is the hierarchy of the cgroup
1030 * that's changing, then we need to check that this
1031 * css_set points to the new cgroup; if it's any other
1032 * hierarchy, then this css_set should point to the
1033 * same cgroup as the old css_set.
1035 if (cgrp1->root == new_cgrp->root) {
1036 if (cgrp1 != new_cgrp)
1047 * find_existing_css_set - init css array and find the matching css_set
1048 * @old_cset: the css_set that we're using before the cgroup transition
1049 * @cgrp: the cgroup that we're moving into
1050 * @template: out param for the new set of csses, should be clear on entry
1052 static struct css_set *find_existing_css_set(struct css_set *old_cset,
1053 struct cgroup *cgrp,
1054 struct cgroup_subsys_state *template[])
1056 struct cgroup_root *root = cgrp->root;
1057 struct cgroup_subsys *ss;
1058 struct css_set *cset;
1063 * Build the set of subsystem state objects that we want to see in the
1064 * new css_set. while subsystems can change globally, the entries here
1065 * won't change, so no need for locking.
1067 for_each_subsys(ss, i) {
1068 if (root->subsys_mask & (1UL << i)) {
1070 * @ss is in this hierarchy, so we want the
1071 * effective css from @cgrp.
1073 template[i] = cgroup_e_css_by_mask(cgrp, ss);
1076 * @ss is not in this hierarchy, so we don't want
1077 * to change the css.
1079 template[i] = old_cset->subsys[i];
1083 key = css_set_hash(template);
1084 hash_for_each_possible(css_set_table, cset, hlist, key) {
1085 if (!compare_css_sets(cset, old_cset, cgrp, template))
1088 /* This css_set matches what we need */
1092 /* No existing cgroup group matched */
1096 static void free_cgrp_cset_links(struct list_head *links_to_free)
1098 struct cgrp_cset_link *link, *tmp_link;
1100 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
1101 list_del(&link->cset_link);
1107 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1108 * @count: the number of links to allocate
1109 * @tmp_links: list_head the allocated links are put on
1111 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1112 * through ->cset_link. Returns 0 on success or -errno.
1114 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1116 struct cgrp_cset_link *link;
1119 INIT_LIST_HEAD(tmp_links);
1121 for (i = 0; i < count; i++) {
1122 link = kzalloc(sizeof(*link), GFP_KERNEL);
1124 free_cgrp_cset_links(tmp_links);
1127 list_add(&link->cset_link, tmp_links);
1133 * link_css_set - a helper function to link a css_set to a cgroup
1134 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1135 * @cset: the css_set to be linked
1136 * @cgrp: the destination cgroup
1138 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1139 struct cgroup *cgrp)
1141 struct cgrp_cset_link *link;
1143 BUG_ON(list_empty(tmp_links));
1145 if (cgroup_on_dfl(cgrp))
1146 cset->dfl_cgrp = cgrp;
1148 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1153 * Always add links to the tail of the lists so that the lists are
1154 * in choronological order.
1156 list_move_tail(&link->cset_link, &cgrp->cset_links);
1157 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1159 if (cgroup_parent(cgrp))
1160 cgroup_get_live(cgrp);
1164 * find_css_set - return a new css_set with one cgroup updated
1165 * @old_cset: the baseline css_set
1166 * @cgrp: the cgroup to be updated
1168 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1169 * substituted into the appropriate hierarchy.
1171 static struct css_set *find_css_set(struct css_set *old_cset,
1172 struct cgroup *cgrp)
1174 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1175 struct css_set *cset;
1176 struct list_head tmp_links;
1177 struct cgrp_cset_link *link;
1178 struct cgroup_subsys *ss;
1182 lockdep_assert_held(&cgroup_mutex);
1184 /* First see if we already have a cgroup group that matches
1185 * the desired set */
1186 spin_lock_irq(&css_set_lock);
1187 cset = find_existing_css_set(old_cset, cgrp, template);
1190 spin_unlock_irq(&css_set_lock);
1195 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1199 /* Allocate all the cgrp_cset_link objects that we'll need */
1200 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1205 refcount_set(&cset->refcount, 1);
1206 cset->dom_cset = cset;
1207 INIT_LIST_HEAD(&cset->tasks);
1208 INIT_LIST_HEAD(&cset->mg_tasks);
1209 INIT_LIST_HEAD(&cset->dying_tasks);
1210 INIT_LIST_HEAD(&cset->task_iters);
1211 INIT_LIST_HEAD(&cset->threaded_csets);
1212 INIT_HLIST_NODE(&cset->hlist);
1213 INIT_LIST_HEAD(&cset->cgrp_links);
1214 INIT_LIST_HEAD(&cset->mg_preload_node);
1215 INIT_LIST_HEAD(&cset->mg_node);
1217 /* Copy the set of subsystem state objects generated in
1218 * find_existing_css_set() */
1219 memcpy(cset->subsys, template, sizeof(cset->subsys));
1221 spin_lock_irq(&css_set_lock);
1222 /* Add reference counts and links from the new css_set. */
1223 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1224 struct cgroup *c = link->cgrp;
1226 if (c->root == cgrp->root)
1228 link_css_set(&tmp_links, cset, c);
1231 BUG_ON(!list_empty(&tmp_links));
1235 /* Add @cset to the hash table */
1236 key = css_set_hash(cset->subsys);
1237 hash_add(css_set_table, &cset->hlist, key);
1239 for_each_subsys(ss, ssid) {
1240 struct cgroup_subsys_state *css = cset->subsys[ssid];
1242 list_add_tail(&cset->e_cset_node[ssid],
1243 &css->cgroup->e_csets[ssid]);
1247 spin_unlock_irq(&css_set_lock);
1250 * If @cset should be threaded, look up the matching dom_cset and
1251 * link them up. We first fully initialize @cset then look for the
1252 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1253 * to stay empty until we return.
1255 if (cgroup_is_threaded(cset->dfl_cgrp)) {
1256 struct css_set *dcset;
1258 dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
1264 spin_lock_irq(&css_set_lock);
1265 cset->dom_cset = dcset;
1266 list_add_tail(&cset->threaded_csets_node,
1267 &dcset->threaded_csets);
1268 spin_unlock_irq(&css_set_lock);
1274 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1276 struct cgroup *root_cgrp = kf_root->kn->priv;
1278 return root_cgrp->root;
1281 static int cgroup_init_root_id(struct cgroup_root *root)
1285 lockdep_assert_held(&cgroup_mutex);
1287 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1291 root->hierarchy_id = id;
1295 static void cgroup_exit_root_id(struct cgroup_root *root)
1297 lockdep_assert_held(&cgroup_mutex);
1299 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1302 void cgroup_free_root(struct cgroup_root *root)
1307 static void cgroup_destroy_root(struct cgroup_root *root)
1309 struct cgroup *cgrp = &root->cgrp;
1310 struct cgrp_cset_link *link, *tmp_link;
1312 trace_cgroup_destroy_root(root);
1314 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1316 BUG_ON(atomic_read(&root->nr_cgrps));
1317 BUG_ON(!list_empty(&cgrp->self.children));
1319 /* Rebind all subsystems back to the default hierarchy */
1320 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1323 * Release all the links from cset_links to this hierarchy's
1326 spin_lock_irq(&css_set_lock);
1328 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1329 list_del(&link->cset_link);
1330 list_del(&link->cgrp_link);
1334 spin_unlock_irq(&css_set_lock);
1336 if (!list_empty(&root->root_list)) {
1337 list_del(&root->root_list);
1338 cgroup_root_count--;
1341 cgroup_exit_root_id(root);
1343 mutex_unlock(&cgroup_mutex);
1345 kernfs_destroy_root(root->kf_root);
1346 cgroup_free_root(root);
1350 * look up cgroup associated with current task's cgroup namespace on the
1351 * specified hierarchy
1353 static struct cgroup *
1354 current_cgns_cgroup_from_root(struct cgroup_root *root)
1356 struct cgroup *res = NULL;
1357 struct css_set *cset;
1359 lockdep_assert_held(&css_set_lock);
1363 cset = current->nsproxy->cgroup_ns->root_cset;
1364 if (cset == &init_css_set) {
1366 } else if (root == &cgrp_dfl_root) {
1367 res = cset->dfl_cgrp;
1369 struct cgrp_cset_link *link;
1371 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1372 struct cgroup *c = link->cgrp;
1374 if (c->root == root) {
1386 /* look up cgroup associated with given css_set on the specified hierarchy */
1387 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1388 struct cgroup_root *root)
1390 struct cgroup *res = NULL;
1392 lockdep_assert_held(&cgroup_mutex);
1393 lockdep_assert_held(&css_set_lock);
1395 if (cset == &init_css_set) {
1397 } else if (root == &cgrp_dfl_root) {
1398 res = cset->dfl_cgrp;
1400 struct cgrp_cset_link *link;
1402 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1403 struct cgroup *c = link->cgrp;
1405 if (c->root == root) {
1417 * Return the cgroup for "task" from the given hierarchy. Must be
1418 * called with cgroup_mutex and css_set_lock held.
1420 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1421 struct cgroup_root *root)
1424 * No need to lock the task - since we hold css_set_lock the
1425 * task can't change groups.
1427 return cset_cgroup_from_root(task_css_set(task), root);
1431 * A task must hold cgroup_mutex to modify cgroups.
1433 * Any task can increment and decrement the count field without lock.
1434 * So in general, code holding cgroup_mutex can't rely on the count
1435 * field not changing. However, if the count goes to zero, then only
1436 * cgroup_attach_task() can increment it again. Because a count of zero
1437 * means that no tasks are currently attached, therefore there is no
1438 * way a task attached to that cgroup can fork (the other way to
1439 * increment the count). So code holding cgroup_mutex can safely
1440 * assume that if the count is zero, it will stay zero. Similarly, if
1441 * a task holds cgroup_mutex on a cgroup with zero count, it
1442 * knows that the cgroup won't be removed, as cgroup_rmdir()
1445 * A cgroup can only be deleted if both its 'count' of using tasks
1446 * is zero, and its list of 'children' cgroups is empty. Since all
1447 * tasks in the system use _some_ cgroup, and since there is always at
1448 * least one task in the system (init, pid == 1), therefore, root cgroup
1449 * always has either children cgroups and/or using tasks. So we don't
1450 * need a special hack to ensure that root cgroup cannot be deleted.
1452 * P.S. One more locking exception. RCU is used to guard the
1453 * update of a tasks cgroup pointer by cgroup_attach_task()
1456 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1458 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1461 struct cgroup_subsys *ss = cft->ss;
1463 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1464 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
1465 const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : "";
1467 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s",
1468 dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1471 strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1477 * cgroup_file_mode - deduce file mode of a control file
1478 * @cft: the control file in question
1480 * S_IRUGO for read, S_IWUSR for write.
1482 static umode_t cgroup_file_mode(const struct cftype *cft)
1486 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1489 if (cft->write_u64 || cft->write_s64 || cft->write) {
1490 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1500 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1501 * @subtree_control: the new subtree_control mask to consider
1502 * @this_ss_mask: available subsystems
1504 * On the default hierarchy, a subsystem may request other subsystems to be
1505 * enabled together through its ->depends_on mask. In such cases, more
1506 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1508 * This function calculates which subsystems need to be enabled if
1509 * @subtree_control is to be applied while restricted to @this_ss_mask.
1511 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1513 u16 cur_ss_mask = subtree_control;
1514 struct cgroup_subsys *ss;
1517 lockdep_assert_held(&cgroup_mutex);
1519 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1522 u16 new_ss_mask = cur_ss_mask;
1524 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1525 new_ss_mask |= ss->depends_on;
1526 } while_each_subsys_mask();
1529 * Mask out subsystems which aren't available. This can
1530 * happen only if some depended-upon subsystems were bound
1531 * to non-default hierarchies.
1533 new_ss_mask &= this_ss_mask;
1535 if (new_ss_mask == cur_ss_mask)
1537 cur_ss_mask = new_ss_mask;
1544 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1545 * @kn: the kernfs_node being serviced
1547 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1548 * the method finishes if locking succeeded. Note that once this function
1549 * returns the cgroup returned by cgroup_kn_lock_live() may become
1550 * inaccessible any time. If the caller intends to continue to access the
1551 * cgroup, it should pin it before invoking this function.
1553 void cgroup_kn_unlock(struct kernfs_node *kn)
1555 struct cgroup *cgrp;
1557 if (kernfs_type(kn) == KERNFS_DIR)
1560 cgrp = kn->parent->priv;
1562 mutex_unlock(&cgroup_mutex);
1564 kernfs_unbreak_active_protection(kn);
1569 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1570 * @kn: the kernfs_node being serviced
1571 * @drain_offline: perform offline draining on the cgroup
1573 * This helper is to be used by a cgroup kernfs method currently servicing
1574 * @kn. It breaks the active protection, performs cgroup locking and
1575 * verifies that the associated cgroup is alive. Returns the cgroup if
1576 * alive; otherwise, %NULL. A successful return should be undone by a
1577 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1578 * cgroup is drained of offlining csses before return.
1580 * Any cgroup kernfs method implementation which requires locking the
1581 * associated cgroup should use this helper. It avoids nesting cgroup
1582 * locking under kernfs active protection and allows all kernfs operations
1583 * including self-removal.
1585 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1587 struct cgroup *cgrp;
1589 if (kernfs_type(kn) == KERNFS_DIR)
1592 cgrp = kn->parent->priv;
1595 * We're gonna grab cgroup_mutex which nests outside kernfs
1596 * active_ref. cgroup liveliness check alone provides enough
1597 * protection against removal. Ensure @cgrp stays accessible and
1598 * break the active_ref protection.
1600 if (!cgroup_tryget(cgrp))
1602 kernfs_break_active_protection(kn);
1605 cgroup_lock_and_drain_offline(cgrp);
1607 mutex_lock(&cgroup_mutex);
1609 if (!cgroup_is_dead(cgrp))
1612 cgroup_kn_unlock(kn);
1616 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1618 char name[CGROUP_FILE_NAME_MAX];
1620 lockdep_assert_held(&cgroup_mutex);
1622 if (cft->file_offset) {
1623 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1624 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1626 spin_lock_irq(&cgroup_file_kn_lock);
1628 spin_unlock_irq(&cgroup_file_kn_lock);
1630 del_timer_sync(&cfile->notify_timer);
1633 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1637 * css_clear_dir - remove subsys files in a cgroup directory
1640 static void css_clear_dir(struct cgroup_subsys_state *css)
1642 struct cgroup *cgrp = css->cgroup;
1643 struct cftype *cfts;
1645 if (!(css->flags & CSS_VISIBLE))
1648 css->flags &= ~CSS_VISIBLE;
1651 if (cgroup_on_dfl(cgrp))
1652 cfts = cgroup_base_files;
1654 cfts = cgroup1_base_files;
1656 cgroup_addrm_files(css, cgrp, cfts, false);
1658 list_for_each_entry(cfts, &css->ss->cfts, node)
1659 cgroup_addrm_files(css, cgrp, cfts, false);
1664 * css_populate_dir - create subsys files in a cgroup directory
1667 * On failure, no file is added.
1669 static int css_populate_dir(struct cgroup_subsys_state *css)
1671 struct cgroup *cgrp = css->cgroup;
1672 struct cftype *cfts, *failed_cfts;
1675 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1679 if (cgroup_on_dfl(cgrp))
1680 cfts = cgroup_base_files;
1682 cfts = cgroup1_base_files;
1684 ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1688 list_for_each_entry(cfts, &css->ss->cfts, node) {
1689 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1697 css->flags |= CSS_VISIBLE;
1701 list_for_each_entry(cfts, &css->ss->cfts, node) {
1702 if (cfts == failed_cfts)
1704 cgroup_addrm_files(css, cgrp, cfts, false);
1709 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1711 struct cgroup *dcgrp = &dst_root->cgrp;
1712 struct cgroup_subsys *ss;
1714 u16 dfl_disable_ss_mask = 0;
1716 lockdep_assert_held(&cgroup_mutex);
1718 do_each_subsys_mask(ss, ssid, ss_mask) {
1720 * If @ss has non-root csses attached to it, can't move.
1721 * If @ss is an implicit controller, it is exempt from this
1722 * rule and can be stolen.
1724 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1725 !ss->implicit_on_dfl)
1728 /* can't move between two non-dummy roots either */
1729 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1733 * Collect ssid's that need to be disabled from default
1736 if (ss->root == &cgrp_dfl_root)
1737 dfl_disable_ss_mask |= 1 << ssid;
1739 } while_each_subsys_mask();
1741 if (dfl_disable_ss_mask) {
1742 struct cgroup *scgrp = &cgrp_dfl_root.cgrp;
1745 * Controllers from default hierarchy that need to be rebound
1746 * are all disabled together in one go.
1748 cgrp_dfl_root.subsys_mask &= ~dfl_disable_ss_mask;
1749 WARN_ON(cgroup_apply_control(scgrp));
1750 cgroup_finalize_control(scgrp, 0);
1753 do_each_subsys_mask(ss, ssid, ss_mask) {
1754 struct cgroup_root *src_root = ss->root;
1755 struct cgroup *scgrp = &src_root->cgrp;
1756 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1757 struct css_set *cset;
1759 WARN_ON(!css || cgroup_css(dcgrp, ss));
1761 if (src_root != &cgrp_dfl_root) {
1762 /* disable from the source */
1763 src_root->subsys_mask &= ~(1 << ssid);
1764 WARN_ON(cgroup_apply_control(scgrp));
1765 cgroup_finalize_control(scgrp, 0);
1769 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1770 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1771 ss->root = dst_root;
1772 css->cgroup = dcgrp;
1774 spin_lock_irq(&css_set_lock);
1775 hash_for_each(css_set_table, i, cset, hlist)
1776 list_move_tail(&cset->e_cset_node[ss->id],
1777 &dcgrp->e_csets[ss->id]);
1778 spin_unlock_irq(&css_set_lock);
1780 /* default hierarchy doesn't enable controllers by default */
1781 dst_root->subsys_mask |= 1 << ssid;
1782 if (dst_root == &cgrp_dfl_root) {
1783 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1785 dcgrp->subtree_control |= 1 << ssid;
1786 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1789 ret = cgroup_apply_control(dcgrp);
1791 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1796 } while_each_subsys_mask();
1798 kernfs_activate(dcgrp->kn);
1802 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1803 struct kernfs_root *kf_root)
1807 struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1808 struct cgroup *ns_cgroup;
1810 buf = kmalloc(PATH_MAX, GFP_KERNEL);
1814 spin_lock_irq(&css_set_lock);
1815 ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1816 len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1817 spin_unlock_irq(&css_set_lock);
1819 if (len >= PATH_MAX)
1822 seq_escape(sf, buf, " \t\n\\");
1829 enum cgroup2_param {
1831 Opt_memory_localevents,
1832 Opt_memory_recursiveprot,
1836 static const struct fs_parameter_spec cgroup2_fs_parameters[] = {
1837 fsparam_flag("nsdelegate", Opt_nsdelegate),
1838 fsparam_flag("memory_localevents", Opt_memory_localevents),
1839 fsparam_flag("memory_recursiveprot", Opt_memory_recursiveprot),
1843 static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1845 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1846 struct fs_parse_result result;
1849 opt = fs_parse(fc, cgroup2_fs_parameters, param, &result);
1854 case Opt_nsdelegate:
1855 ctx->flags |= CGRP_ROOT_NS_DELEGATE;
1857 case Opt_memory_localevents:
1858 ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1860 case Opt_memory_recursiveprot:
1861 ctx->flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1867 static void apply_cgroup_root_flags(unsigned int root_flags)
1869 if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
1870 if (root_flags & CGRP_ROOT_NS_DELEGATE)
1871 cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
1873 cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
1875 if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1876 cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1878 cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1880 if (root_flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1881 cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1883 cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1887 static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
1889 if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
1890 seq_puts(seq, ",nsdelegate");
1891 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1892 seq_puts(seq, ",memory_localevents");
1893 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1894 seq_puts(seq, ",memory_recursiveprot");
1898 static int cgroup_reconfigure(struct fs_context *fc)
1900 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1902 apply_cgroup_root_flags(ctx->flags);
1906 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1908 struct cgroup_subsys *ss;
1911 INIT_LIST_HEAD(&cgrp->self.sibling);
1912 INIT_LIST_HEAD(&cgrp->self.children);
1913 INIT_LIST_HEAD(&cgrp->cset_links);
1914 INIT_LIST_HEAD(&cgrp->pidlists);
1915 mutex_init(&cgrp->pidlist_mutex);
1916 cgrp->self.cgroup = cgrp;
1917 cgrp->self.flags |= CSS_ONLINE;
1918 cgrp->dom_cgrp = cgrp;
1919 cgrp->max_descendants = INT_MAX;
1920 cgrp->max_depth = INT_MAX;
1921 INIT_LIST_HEAD(&cgrp->rstat_css_list);
1922 prev_cputime_init(&cgrp->prev_cputime);
1924 for_each_subsys(ss, ssid)
1925 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1927 init_waitqueue_head(&cgrp->offline_waitq);
1928 INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1931 void init_cgroup_root(struct cgroup_fs_context *ctx)
1933 struct cgroup_root *root = ctx->root;
1934 struct cgroup *cgrp = &root->cgrp;
1936 INIT_LIST_HEAD(&root->root_list);
1937 atomic_set(&root->nr_cgrps, 1);
1939 init_cgroup_housekeeping(cgrp);
1941 root->flags = ctx->flags;
1942 if (ctx->release_agent)
1943 strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX);
1945 strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN);
1946 if (ctx->cpuset_clone_children)
1947 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1950 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1952 LIST_HEAD(tmp_links);
1953 struct cgroup *root_cgrp = &root->cgrp;
1954 struct kernfs_syscall_ops *kf_sops;
1955 struct css_set *cset;
1958 lockdep_assert_held(&cgroup_mutex);
1960 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
1966 * We're accessing css_set_count without locking css_set_lock here,
1967 * but that's OK - it can only be increased by someone holding
1968 * cgroup_lock, and that's us. Later rebinding may disable
1969 * controllers on the default hierarchy and thus create new csets,
1970 * which can't be more than the existing ones. Allocate 2x.
1972 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
1976 ret = cgroup_init_root_id(root);
1980 kf_sops = root == &cgrp_dfl_root ?
1981 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
1983 root->kf_root = kernfs_create_root(kf_sops,
1984 KERNFS_ROOT_CREATE_DEACTIVATED |
1985 KERNFS_ROOT_SUPPORT_EXPORTOP |
1986 KERNFS_ROOT_SUPPORT_USER_XATTR,
1988 if (IS_ERR(root->kf_root)) {
1989 ret = PTR_ERR(root->kf_root);
1992 root_cgrp->kn = root->kf_root->kn;
1993 WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
1994 root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
1996 ret = css_populate_dir(&root_cgrp->self);
2000 ret = rebind_subsystems(root, ss_mask);
2004 ret = cgroup_bpf_inherit(root_cgrp);
2007 trace_cgroup_setup_root(root);
2010 * There must be no failure case after here, since rebinding takes
2011 * care of subsystems' refcounts, which are explicitly dropped in
2012 * the failure exit path.
2014 list_add(&root->root_list, &cgroup_roots);
2015 cgroup_root_count++;
2018 * Link the root cgroup in this hierarchy into all the css_set
2021 spin_lock_irq(&css_set_lock);
2022 hash_for_each(css_set_table, i, cset, hlist) {
2023 link_css_set(&tmp_links, cset, root_cgrp);
2024 if (css_set_populated(cset))
2025 cgroup_update_populated(root_cgrp, true);
2027 spin_unlock_irq(&css_set_lock);
2029 BUG_ON(!list_empty(&root_cgrp->self.children));
2030 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
2036 kernfs_destroy_root(root->kf_root);
2037 root->kf_root = NULL;
2039 cgroup_exit_root_id(root);
2041 percpu_ref_exit(&root_cgrp->self.refcnt);
2043 free_cgrp_cset_links(&tmp_links);
2047 int cgroup_do_get_tree(struct fs_context *fc)
2049 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2052 ctx->kfc.root = ctx->root->kf_root;
2053 if (fc->fs_type == &cgroup2_fs_type)
2054 ctx->kfc.magic = CGROUP2_SUPER_MAGIC;
2056 ctx->kfc.magic = CGROUP_SUPER_MAGIC;
2057 ret = kernfs_get_tree(fc);
2060 * In non-init cgroup namespace, instead of root cgroup's dentry,
2061 * we return the dentry corresponding to the cgroupns->root_cgrp.
2063 if (!ret && ctx->ns != &init_cgroup_ns) {
2064 struct dentry *nsdentry;
2065 struct super_block *sb = fc->root->d_sb;
2066 struct cgroup *cgrp;
2068 mutex_lock(&cgroup_mutex);
2069 spin_lock_irq(&css_set_lock);
2071 cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root);
2073 spin_unlock_irq(&css_set_lock);
2074 mutex_unlock(&cgroup_mutex);
2076 nsdentry = kernfs_node_dentry(cgrp->kn, sb);
2078 if (IS_ERR(nsdentry)) {
2079 deactivate_locked_super(sb);
2080 ret = PTR_ERR(nsdentry);
2083 fc->root = nsdentry;
2086 if (!ctx->kfc.new_sb_created)
2087 cgroup_put(&ctx->root->cgrp);
2093 * Destroy a cgroup filesystem context.
2095 static void cgroup_fs_context_free(struct fs_context *fc)
2097 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2100 kfree(ctx->release_agent);
2101 put_cgroup_ns(ctx->ns);
2102 kernfs_free_fs_context(fc);
2106 static int cgroup_get_tree(struct fs_context *fc)
2108 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2111 cgrp_dfl_visible = true;
2112 cgroup_get_live(&cgrp_dfl_root.cgrp);
2113 ctx->root = &cgrp_dfl_root;
2115 ret = cgroup_do_get_tree(fc);
2117 apply_cgroup_root_flags(ctx->flags);
2121 static const struct fs_context_operations cgroup_fs_context_ops = {
2122 .free = cgroup_fs_context_free,
2123 .parse_param = cgroup2_parse_param,
2124 .get_tree = cgroup_get_tree,
2125 .reconfigure = cgroup_reconfigure,
2128 static const struct fs_context_operations cgroup1_fs_context_ops = {
2129 .free = cgroup_fs_context_free,
2130 .parse_param = cgroup1_parse_param,
2131 .get_tree = cgroup1_get_tree,
2132 .reconfigure = cgroup1_reconfigure,
2136 * Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2137 * we select the namespace we're going to use.
2139 static int cgroup_init_fs_context(struct fs_context *fc)
2141 struct cgroup_fs_context *ctx;
2143 ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL);
2147 ctx->ns = current->nsproxy->cgroup_ns;
2148 get_cgroup_ns(ctx->ns);
2149 fc->fs_private = &ctx->kfc;
2150 if (fc->fs_type == &cgroup2_fs_type)
2151 fc->ops = &cgroup_fs_context_ops;
2153 fc->ops = &cgroup1_fs_context_ops;
2154 put_user_ns(fc->user_ns);
2155 fc->user_ns = get_user_ns(ctx->ns->user_ns);
2160 static void cgroup_kill_sb(struct super_block *sb)
2162 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2163 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2166 * If @root doesn't have any children, start killing it.
2167 * This prevents new mounts by disabling percpu_ref_tryget_live().
2168 * cgroup_mount() may wait for @root's release.
2170 * And don't kill the default root.
2172 if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root &&
2173 !percpu_ref_is_dying(&root->cgrp.self.refcnt)) {
2174 cgroup_bpf_offline(&root->cgrp);
2175 percpu_ref_kill(&root->cgrp.self.refcnt);
2177 cgroup_put(&root->cgrp);
2181 struct file_system_type cgroup_fs_type = {
2183 .init_fs_context = cgroup_init_fs_context,
2184 .parameters = cgroup1_fs_parameters,
2185 .kill_sb = cgroup_kill_sb,
2186 .fs_flags = FS_USERNS_MOUNT,
2189 static struct file_system_type cgroup2_fs_type = {
2191 .init_fs_context = cgroup_init_fs_context,
2192 .parameters = cgroup2_fs_parameters,
2193 .kill_sb = cgroup_kill_sb,
2194 .fs_flags = FS_USERNS_MOUNT,
2197 #ifdef CONFIG_CPUSETS
2198 static const struct fs_context_operations cpuset_fs_context_ops = {
2199 .get_tree = cgroup1_get_tree,
2200 .free = cgroup_fs_context_free,
2204 * This is ugly, but preserves the userspace API for existing cpuset
2205 * users. If someone tries to mount the "cpuset" filesystem, we
2206 * silently switch it to mount "cgroup" instead
2208 static int cpuset_init_fs_context(struct fs_context *fc)
2210 char *agent = kstrdup("/sbin/cpuset_release_agent", GFP_USER);
2211 struct cgroup_fs_context *ctx;
2214 err = cgroup_init_fs_context(fc);
2220 fc->ops = &cpuset_fs_context_ops;
2222 ctx = cgroup_fc2context(fc);
2223 ctx->subsys_mask = 1 << cpuset_cgrp_id;
2224 ctx->flags |= CGRP_ROOT_NOPREFIX;
2225 ctx->release_agent = agent;
2227 get_filesystem(&cgroup_fs_type);
2228 put_filesystem(fc->fs_type);
2229 fc->fs_type = &cgroup_fs_type;
2234 static struct file_system_type cpuset_fs_type = {
2236 .init_fs_context = cpuset_init_fs_context,
2237 .fs_flags = FS_USERNS_MOUNT,
2241 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2242 struct cgroup_namespace *ns)
2244 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2246 return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2249 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2250 struct cgroup_namespace *ns)
2254 mutex_lock(&cgroup_mutex);
2255 spin_lock_irq(&css_set_lock);
2257 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2259 spin_unlock_irq(&css_set_lock);
2260 mutex_unlock(&cgroup_mutex);
2264 EXPORT_SYMBOL_GPL(cgroup_path_ns);
2267 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2268 * @task: target task
2269 * @buf: the buffer to write the path into
2270 * @buflen: the length of the buffer
2272 * Determine @task's cgroup on the first (the one with the lowest non-zero
2273 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2274 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2275 * cgroup controller callbacks.
2277 * Return value is the same as kernfs_path().
2279 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2281 struct cgroup_root *root;
2282 struct cgroup *cgrp;
2283 int hierarchy_id = 1;
2286 mutex_lock(&cgroup_mutex);
2287 spin_lock_irq(&css_set_lock);
2289 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2292 cgrp = task_cgroup_from_root(task, root);
2293 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2295 /* if no hierarchy exists, everyone is in "/" */
2296 ret = strlcpy(buf, "/", buflen);
2299 spin_unlock_irq(&css_set_lock);
2300 mutex_unlock(&cgroup_mutex);
2303 EXPORT_SYMBOL_GPL(task_cgroup_path);
2306 * cgroup_migrate_add_task - add a migration target task to a migration context
2307 * @task: target task
2308 * @mgctx: target migration context
2310 * Add @task, which is a migration target, to @mgctx->tset. This function
2311 * becomes noop if @task doesn't need to be migrated. @task's css_set
2312 * should have been added as a migration source and @task->cg_list will be
2313 * moved from the css_set's tasks list to mg_tasks one.
2315 static void cgroup_migrate_add_task(struct task_struct *task,
2316 struct cgroup_mgctx *mgctx)
2318 struct css_set *cset;
2320 lockdep_assert_held(&css_set_lock);
2322 /* @task either already exited or can't exit until the end */
2323 if (task->flags & PF_EXITING)
2326 /* cgroup_threadgroup_rwsem protects racing against forks */
2327 WARN_ON_ONCE(list_empty(&task->cg_list));
2329 cset = task_css_set(task);
2330 if (!cset->mg_src_cgrp)
2333 mgctx->tset.nr_tasks++;
2335 list_move_tail(&task->cg_list, &cset->mg_tasks);
2336 if (list_empty(&cset->mg_node))
2337 list_add_tail(&cset->mg_node,
2338 &mgctx->tset.src_csets);
2339 if (list_empty(&cset->mg_dst_cset->mg_node))
2340 list_add_tail(&cset->mg_dst_cset->mg_node,
2341 &mgctx->tset.dst_csets);
2345 * cgroup_taskset_first - reset taskset and return the first task
2346 * @tset: taskset of interest
2347 * @dst_cssp: output variable for the destination css
2349 * @tset iteration is initialized and the first task is returned.
2351 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2352 struct cgroup_subsys_state **dst_cssp)
2354 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2355 tset->cur_task = NULL;
2357 return cgroup_taskset_next(tset, dst_cssp);
2361 * cgroup_taskset_next - iterate to the next task in taskset
2362 * @tset: taskset of interest
2363 * @dst_cssp: output variable for the destination css
2365 * Return the next task in @tset. Iteration must have been initialized
2366 * with cgroup_taskset_first().
2368 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2369 struct cgroup_subsys_state **dst_cssp)
2371 struct css_set *cset = tset->cur_cset;
2372 struct task_struct *task = tset->cur_task;
2374 while (&cset->mg_node != tset->csets) {
2376 task = list_first_entry(&cset->mg_tasks,
2377 struct task_struct, cg_list);
2379 task = list_next_entry(task, cg_list);
2381 if (&task->cg_list != &cset->mg_tasks) {
2382 tset->cur_cset = cset;
2383 tset->cur_task = task;
2386 * This function may be called both before and
2387 * after cgroup_taskset_migrate(). The two cases
2388 * can be distinguished by looking at whether @cset
2389 * has its ->mg_dst_cset set.
2391 if (cset->mg_dst_cset)
2392 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2394 *dst_cssp = cset->subsys[tset->ssid];
2399 cset = list_next_entry(cset, mg_node);
2407 * cgroup_taskset_migrate - migrate a taskset
2408 * @mgctx: migration context
2410 * Migrate tasks in @mgctx as setup by migration preparation functions.
2411 * This function fails iff one of the ->can_attach callbacks fails and
2412 * guarantees that either all or none of the tasks in @mgctx are migrated.
2413 * @mgctx is consumed regardless of success.
2415 static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2417 struct cgroup_taskset *tset = &mgctx->tset;
2418 struct cgroup_subsys *ss;
2419 struct task_struct *task, *tmp_task;
2420 struct css_set *cset, *tmp_cset;
2421 int ssid, failed_ssid, ret;
2423 /* check that we can legitimately attach to the cgroup */
2424 if (tset->nr_tasks) {
2425 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2426 if (ss->can_attach) {
2428 ret = ss->can_attach(tset);
2431 goto out_cancel_attach;
2434 } while_each_subsys_mask();
2438 * Now that we're guaranteed success, proceed to move all tasks to
2439 * the new cgroup. There are no failure cases after here, so this
2440 * is the commit point.
2442 spin_lock_irq(&css_set_lock);
2443 list_for_each_entry(cset, &tset->src_csets, mg_node) {
2444 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2445 struct css_set *from_cset = task_css_set(task);
2446 struct css_set *to_cset = cset->mg_dst_cset;
2448 get_css_set(to_cset);
2449 to_cset->nr_tasks++;
2450 css_set_move_task(task, from_cset, to_cset, true);
2451 from_cset->nr_tasks--;
2453 * If the source or destination cgroup is frozen,
2454 * the task might require to change its state.
2456 cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp,
2458 put_css_set_locked(from_cset);
2462 spin_unlock_irq(&css_set_lock);
2465 * Migration is committed, all target tasks are now on dst_csets.
2466 * Nothing is sensitive to fork() after this point. Notify
2467 * controllers that migration is complete.
2469 tset->csets = &tset->dst_csets;
2471 if (tset->nr_tasks) {
2472 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2477 } while_each_subsys_mask();
2481 goto out_release_tset;
2484 if (tset->nr_tasks) {
2485 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2486 if (ssid == failed_ssid)
2488 if (ss->cancel_attach) {
2490 ss->cancel_attach(tset);
2492 } while_each_subsys_mask();
2495 spin_lock_irq(&css_set_lock);
2496 list_splice_init(&tset->dst_csets, &tset->src_csets);
2497 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2498 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2499 list_del_init(&cset->mg_node);
2501 spin_unlock_irq(&css_set_lock);
2504 * Re-initialize the cgroup_taskset structure in case it is reused
2505 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2509 tset->csets = &tset->src_csets;
2514 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2515 * @dst_cgrp: destination cgroup to test
2517 * On the default hierarchy, except for the mixable, (possible) thread root
2518 * and threaded cgroups, subtree_control must be zero for migration
2519 * destination cgroups with tasks so that child cgroups don't compete
2522 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2524 /* v1 doesn't have any restriction */
2525 if (!cgroup_on_dfl(dst_cgrp))
2528 /* verify @dst_cgrp can host resources */
2529 if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
2532 /* mixables don't care */
2533 if (cgroup_is_mixable(dst_cgrp))
2537 * If @dst_cgrp is already or can become a thread root or is
2538 * threaded, it doesn't matter.
2540 if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
2543 /* apply no-internal-process constraint */
2544 if (dst_cgrp->subtree_control)
2551 * cgroup_migrate_finish - cleanup after attach
2552 * @mgctx: migration context
2554 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2555 * those functions for details.
2557 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2559 LIST_HEAD(preloaded);
2560 struct css_set *cset, *tmp_cset;
2562 lockdep_assert_held(&cgroup_mutex);
2564 spin_lock_irq(&css_set_lock);
2566 list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded);
2567 list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded);
2569 list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) {
2570 cset->mg_src_cgrp = NULL;
2571 cset->mg_dst_cgrp = NULL;
2572 cset->mg_dst_cset = NULL;
2573 list_del_init(&cset->mg_preload_node);
2574 put_css_set_locked(cset);
2577 spin_unlock_irq(&css_set_lock);
2581 * cgroup_migrate_add_src - add a migration source css_set
2582 * @src_cset: the source css_set to add
2583 * @dst_cgrp: the destination cgroup
2584 * @mgctx: migration context
2586 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2587 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2588 * up by cgroup_migrate_finish().
2590 * This function may be called without holding cgroup_threadgroup_rwsem
2591 * even if the target is a process. Threads may be created and destroyed
2592 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2593 * into play and the preloaded css_sets are guaranteed to cover all
2596 void cgroup_migrate_add_src(struct css_set *src_cset,
2597 struct cgroup *dst_cgrp,
2598 struct cgroup_mgctx *mgctx)
2600 struct cgroup *src_cgrp;
2602 lockdep_assert_held(&cgroup_mutex);
2603 lockdep_assert_held(&css_set_lock);
2606 * If ->dead, @src_set is associated with one or more dead cgroups
2607 * and doesn't contain any migratable tasks. Ignore it early so
2608 * that the rest of migration path doesn't get confused by it.
2613 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2615 if (!list_empty(&src_cset->mg_preload_node))
2618 WARN_ON(src_cset->mg_src_cgrp);
2619 WARN_ON(src_cset->mg_dst_cgrp);
2620 WARN_ON(!list_empty(&src_cset->mg_tasks));
2621 WARN_ON(!list_empty(&src_cset->mg_node));
2623 src_cset->mg_src_cgrp = src_cgrp;
2624 src_cset->mg_dst_cgrp = dst_cgrp;
2625 get_css_set(src_cset);
2626 list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets);
2630 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2631 * @mgctx: migration context
2633 * Tasks are about to be moved and all the source css_sets have been
2634 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2635 * pins all destination css_sets, links each to its source, and append them
2636 * to @mgctx->preloaded_dst_csets.
2638 * This function must be called after cgroup_migrate_add_src() has been
2639 * called on each migration source css_set. After migration is performed
2640 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2643 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2645 struct css_set *src_cset, *tmp_cset;
2647 lockdep_assert_held(&cgroup_mutex);
2649 /* look up the dst cset for each src cset and link it to src */
2650 list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2652 struct css_set *dst_cset;
2653 struct cgroup_subsys *ss;
2656 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2660 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2663 * If src cset equals dst, it's noop. Drop the src.
2664 * cgroup_migrate() will skip the cset too. Note that we
2665 * can't handle src == dst as some nodes are used by both.
2667 if (src_cset == dst_cset) {
2668 src_cset->mg_src_cgrp = NULL;
2669 src_cset->mg_dst_cgrp = NULL;
2670 list_del_init(&src_cset->mg_preload_node);
2671 put_css_set(src_cset);
2672 put_css_set(dst_cset);
2676 src_cset->mg_dst_cset = dst_cset;
2678 if (list_empty(&dst_cset->mg_preload_node))
2679 list_add_tail(&dst_cset->mg_preload_node,
2680 &mgctx->preloaded_dst_csets);
2682 put_css_set(dst_cset);
2684 for_each_subsys(ss, ssid)
2685 if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2686 mgctx->ss_mask |= 1 << ssid;
2693 * cgroup_migrate - migrate a process or task to a cgroup
2694 * @leader: the leader of the process or the task to migrate
2695 * @threadgroup: whether @leader points to the whole process or a single task
2696 * @mgctx: migration context
2698 * Migrate a process or task denoted by @leader. If migrating a process,
2699 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2700 * responsible for invoking cgroup_migrate_add_src() and
2701 * cgroup_migrate_prepare_dst() on the targets before invoking this
2702 * function and following up with cgroup_migrate_finish().
2704 * As long as a controller's ->can_attach() doesn't fail, this function is
2705 * guaranteed to succeed. This means that, excluding ->can_attach()
2706 * failure, when migrating multiple targets, the success or failure can be
2707 * decided for all targets by invoking group_migrate_prepare_dst() before
2708 * actually starting migrating.
2710 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2711 struct cgroup_mgctx *mgctx)
2713 struct task_struct *task;
2716 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2717 * already PF_EXITING could be freed from underneath us unless we
2718 * take an rcu_read_lock.
2720 spin_lock_irq(&css_set_lock);
2724 cgroup_migrate_add_task(task, mgctx);
2727 } while_each_thread(leader, task);
2729 spin_unlock_irq(&css_set_lock);
2731 return cgroup_migrate_execute(mgctx);
2735 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2736 * @dst_cgrp: the cgroup to attach to
2737 * @leader: the task or the leader of the threadgroup to be attached
2738 * @threadgroup: attach the whole threadgroup?
2740 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2742 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2745 DEFINE_CGROUP_MGCTX(mgctx);
2746 struct task_struct *task;
2749 /* look up all src csets */
2750 spin_lock_irq(&css_set_lock);
2754 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2757 } while_each_thread(leader, task);
2759 spin_unlock_irq(&css_set_lock);
2761 /* prepare dst csets and commit */
2762 ret = cgroup_migrate_prepare_dst(&mgctx);
2764 ret = cgroup_migrate(leader, threadgroup, &mgctx);
2766 cgroup_migrate_finish(&mgctx);
2769 TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup);
2774 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
2776 __acquires(&cgroup_threadgroup_rwsem)
2778 struct task_struct *tsk;
2781 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2782 return ERR_PTR(-EINVAL);
2785 * If we migrate a single thread, we don't care about threadgroup
2786 * stability. If the thread is `current`, it won't exit(2) under our
2787 * hands or change PID through exec(2). We exclude
2788 * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write
2789 * callers by cgroup_mutex.
2790 * Therefore, we can skip the global lock.
2792 lockdep_assert_held(&cgroup_mutex);
2793 if (pid || threadgroup) {
2794 percpu_down_write(&cgroup_threadgroup_rwsem);
2802 tsk = find_task_by_vpid(pid);
2804 tsk = ERR_PTR(-ESRCH);
2805 goto out_unlock_threadgroup;
2812 tsk = tsk->group_leader;
2815 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2816 * If userland migrates such a kthread to a non-root cgroup, it can
2817 * become trapped in a cpuset, or RT kthread may be born in a
2818 * cgroup with no rt_runtime allocated. Just say no.
2820 if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2821 tsk = ERR_PTR(-EINVAL);
2822 goto out_unlock_threadgroup;
2825 get_task_struct(tsk);
2826 goto out_unlock_rcu;
2828 out_unlock_threadgroup:
2830 percpu_up_write(&cgroup_threadgroup_rwsem);
2838 void cgroup_procs_write_finish(struct task_struct *task, bool locked)
2839 __releases(&cgroup_threadgroup_rwsem)
2841 struct cgroup_subsys *ss;
2844 /* release reference from cgroup_procs_write_start() */
2845 put_task_struct(task);
2848 percpu_up_write(&cgroup_threadgroup_rwsem);
2849 for_each_subsys(ss, ssid)
2850 if (ss->post_attach)
2854 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2856 struct cgroup_subsys *ss;
2857 bool printed = false;
2860 do_each_subsys_mask(ss, ssid, ss_mask) {
2863 seq_puts(seq, ss->name);
2865 } while_each_subsys_mask();
2867 seq_putc(seq, '\n');
2870 /* show controllers which are enabled from the parent */
2871 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2873 struct cgroup *cgrp = seq_css(seq)->cgroup;
2875 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2879 /* show controllers which are enabled for a given cgroup's children */
2880 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2882 struct cgroup *cgrp = seq_css(seq)->cgroup;
2884 cgroup_print_ss_mask(seq, cgrp->subtree_control);
2889 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2890 * @cgrp: root of the subtree to update csses for
2892 * @cgrp's control masks have changed and its subtree's css associations
2893 * need to be updated accordingly. This function looks up all css_sets
2894 * which are attached to the subtree, creates the matching updated css_sets
2895 * and migrates the tasks to the new ones.
2897 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2899 DEFINE_CGROUP_MGCTX(mgctx);
2900 struct cgroup_subsys_state *d_css;
2901 struct cgroup *dsct;
2902 struct css_set *src_cset;
2905 lockdep_assert_held(&cgroup_mutex);
2907 percpu_down_write(&cgroup_threadgroup_rwsem);
2909 /* look up all csses currently attached to @cgrp's subtree */
2910 spin_lock_irq(&css_set_lock);
2911 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2912 struct cgrp_cset_link *link;
2914 list_for_each_entry(link, &dsct->cset_links, cset_link)
2915 cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2917 spin_unlock_irq(&css_set_lock);
2919 /* NULL dst indicates self on default hierarchy */
2920 ret = cgroup_migrate_prepare_dst(&mgctx);
2924 spin_lock_irq(&css_set_lock);
2925 list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) {
2926 struct task_struct *task, *ntask;
2928 /* all tasks in src_csets need to be migrated */
2929 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2930 cgroup_migrate_add_task(task, &mgctx);
2932 spin_unlock_irq(&css_set_lock);
2934 ret = cgroup_migrate_execute(&mgctx);
2936 cgroup_migrate_finish(&mgctx);
2937 percpu_up_write(&cgroup_threadgroup_rwsem);
2942 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2943 * @cgrp: root of the target subtree
2945 * Because css offlining is asynchronous, userland may try to re-enable a
2946 * controller while the previous css is still around. This function grabs
2947 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2949 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
2950 __acquires(&cgroup_mutex)
2952 struct cgroup *dsct;
2953 struct cgroup_subsys_state *d_css;
2954 struct cgroup_subsys *ss;
2958 mutex_lock(&cgroup_mutex);
2960 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2961 for_each_subsys(ss, ssid) {
2962 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2965 if (!css || !percpu_ref_is_dying(&css->refcnt))
2968 cgroup_get_live(dsct);
2969 prepare_to_wait(&dsct->offline_waitq, &wait,
2970 TASK_UNINTERRUPTIBLE);
2972 mutex_unlock(&cgroup_mutex);
2974 finish_wait(&dsct->offline_waitq, &wait);
2983 * cgroup_save_control - save control masks and dom_cgrp of a subtree
2984 * @cgrp: root of the target subtree
2986 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
2987 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
2990 static void cgroup_save_control(struct cgroup *cgrp)
2992 struct cgroup *dsct;
2993 struct cgroup_subsys_state *d_css;
2995 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2996 dsct->old_subtree_control = dsct->subtree_control;
2997 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
2998 dsct->old_dom_cgrp = dsct->dom_cgrp;
3003 * cgroup_propagate_control - refresh control masks of a subtree
3004 * @cgrp: root of the target subtree
3006 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3007 * ->subtree_control and propagate controller availability through the
3008 * subtree so that descendants don't have unavailable controllers enabled.
3010 static void cgroup_propagate_control(struct cgroup *cgrp)
3012 struct cgroup *dsct;
3013 struct cgroup_subsys_state *d_css;
3015 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3016 dsct->subtree_control &= cgroup_control(dsct);
3017 dsct->subtree_ss_mask =
3018 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3019 cgroup_ss_mask(dsct));
3024 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3025 * @cgrp: root of the target subtree
3027 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3028 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3031 static void cgroup_restore_control(struct cgroup *cgrp)
3033 struct cgroup *dsct;
3034 struct cgroup_subsys_state *d_css;
3036 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3037 dsct->subtree_control = dsct->old_subtree_control;
3038 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3039 dsct->dom_cgrp = dsct->old_dom_cgrp;
3043 static bool css_visible(struct cgroup_subsys_state *css)
3045 struct cgroup_subsys *ss = css->ss;
3046 struct cgroup *cgrp = css->cgroup;
3048 if (cgroup_control(cgrp) & (1 << ss->id))
3050 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3052 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3056 * cgroup_apply_control_enable - enable or show csses according to control
3057 * @cgrp: root of the target subtree
3059 * Walk @cgrp's subtree and create new csses or make the existing ones
3060 * visible. A css is created invisible if it's being implicitly enabled
3061 * through dependency. An invisible css is made visible when the userland
3062 * explicitly enables it.
3064 * Returns 0 on success, -errno on failure. On failure, csses which have
3065 * been processed already aren't cleaned up. The caller is responsible for
3066 * cleaning up with cgroup_apply_control_disable().
3068 static int cgroup_apply_control_enable(struct cgroup *cgrp)
3070 struct cgroup *dsct;
3071 struct cgroup_subsys_state *d_css;
3072 struct cgroup_subsys *ss;
3075 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3076 for_each_subsys(ss, ssid) {
3077 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3079 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3083 css = css_create(dsct, ss);
3085 return PTR_ERR(css);
3088 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3090 if (css_visible(css)) {
3091 ret = css_populate_dir(css);
3102 * cgroup_apply_control_disable - kill or hide csses according to control
3103 * @cgrp: root of the target subtree
3105 * Walk @cgrp's subtree and kill and hide csses so that they match
3106 * cgroup_ss_mask() and cgroup_visible_mask().
3108 * A css is hidden when the userland requests it to be disabled while other
3109 * subsystems are still depending on it. The css must not actively control
3110 * resources and be in the vanilla state if it's made visible again later.
3111 * Controllers which may be depended upon should provide ->css_reset() for
3114 static void cgroup_apply_control_disable(struct cgroup *cgrp)
3116 struct cgroup *dsct;
3117 struct cgroup_subsys_state *d_css;
3118 struct cgroup_subsys *ss;
3121 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3122 for_each_subsys(ss, ssid) {
3123 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3128 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3131 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
3133 } else if (!css_visible(css)) {
3143 * cgroup_apply_control - apply control mask updates to the subtree
3144 * @cgrp: root of the target subtree
3146 * subsystems can be enabled and disabled in a subtree using the following
3149 * 1. Call cgroup_save_control() to stash the current state.
3150 * 2. Update ->subtree_control masks in the subtree as desired.
3151 * 3. Call cgroup_apply_control() to apply the changes.
3152 * 4. Optionally perform other related operations.
3153 * 5. Call cgroup_finalize_control() to finish up.
3155 * This function implements step 3 and propagates the mask changes
3156 * throughout @cgrp's subtree, updates csses accordingly and perform
3157 * process migrations.
3159 static int cgroup_apply_control(struct cgroup *cgrp)
3163 cgroup_propagate_control(cgrp);
3165 ret = cgroup_apply_control_enable(cgrp);
3170 * At this point, cgroup_e_css_by_mask() results reflect the new csses
3171 * making the following cgroup_update_dfl_csses() properly update
3172 * css associations of all tasks in the subtree.
3174 ret = cgroup_update_dfl_csses(cgrp);
3182 * cgroup_finalize_control - finalize control mask update
3183 * @cgrp: root of the target subtree
3184 * @ret: the result of the update
3186 * Finalize control mask update. See cgroup_apply_control() for more info.
3188 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3191 cgroup_restore_control(cgrp);
3192 cgroup_propagate_control(cgrp);
3195 cgroup_apply_control_disable(cgrp);
3198 static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
3200 u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
3202 /* if nothing is getting enabled, nothing to worry about */
3206 /* can @cgrp host any resources? */
3207 if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
3210 /* mixables don't care */
3211 if (cgroup_is_mixable(cgrp))
3214 if (domain_enable) {
3215 /* can't enable domain controllers inside a thread subtree */
3216 if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3220 * Threaded controllers can handle internal competitions
3221 * and are always allowed inside a (prospective) thread
3224 if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3229 * Controllers can't be enabled for a cgroup with tasks to avoid
3230 * child cgroups competing against tasks.
3232 if (cgroup_has_tasks(cgrp))
3238 /* change the enabled child controllers for a cgroup in the default hierarchy */
3239 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3240 char *buf, size_t nbytes,
3243 u16 enable = 0, disable = 0;
3244 struct cgroup *cgrp, *child;
3245 struct cgroup_subsys *ss;
3250 * Parse input - space separated list of subsystem names prefixed
3251 * with either + or -.
3253 buf = strstrip(buf);
3254 while ((tok = strsep(&buf, " "))) {
3257 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3258 if (!cgroup_ssid_enabled(ssid) ||
3259 strcmp(tok + 1, ss->name))
3263 enable |= 1 << ssid;
3264 disable &= ~(1 << ssid);
3265 } else if (*tok == '-') {
3266 disable |= 1 << ssid;
3267 enable &= ~(1 << ssid);
3272 } while_each_subsys_mask();
3273 if (ssid == CGROUP_SUBSYS_COUNT)
3277 cgrp = cgroup_kn_lock_live(of->kn, true);
3281 for_each_subsys(ss, ssid) {
3282 if (enable & (1 << ssid)) {
3283 if (cgrp->subtree_control & (1 << ssid)) {
3284 enable &= ~(1 << ssid);
3288 if (!(cgroup_control(cgrp) & (1 << ssid))) {
3292 } else if (disable & (1 << ssid)) {
3293 if (!(cgrp->subtree_control & (1 << ssid))) {
3294 disable &= ~(1 << ssid);
3298 /* a child has it enabled? */
3299 cgroup_for_each_live_child(child, cgrp) {
3300 if (child->subtree_control & (1 << ssid)) {
3308 if (!enable && !disable) {
3313 ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3317 /* save and update control masks and prepare csses */
3318 cgroup_save_control(cgrp);
3320 cgrp->subtree_control |= enable;
3321 cgrp->subtree_control &= ~disable;
3323 ret = cgroup_apply_control(cgrp);
3324 cgroup_finalize_control(cgrp, ret);
3328 kernfs_activate(cgrp->kn);
3330 cgroup_kn_unlock(of->kn);
3331 return ret ?: nbytes;
3335 * cgroup_enable_threaded - make @cgrp threaded
3336 * @cgrp: the target cgroup
3338 * Called when "threaded" is written to the cgroup.type interface file and
3339 * tries to make @cgrp threaded and join the parent's resource domain.
3340 * This function is never called on the root cgroup as cgroup.type doesn't
3343 static int cgroup_enable_threaded(struct cgroup *cgrp)
3345 struct cgroup *parent = cgroup_parent(cgrp);
3346 struct cgroup *dom_cgrp = parent->dom_cgrp;
3347 struct cgroup *dsct;
3348 struct cgroup_subsys_state *d_css;
3351 lockdep_assert_held(&cgroup_mutex);
3353 /* noop if already threaded */
3354 if (cgroup_is_threaded(cgrp))
3358 * If @cgroup is populated or has domain controllers enabled, it
3359 * can't be switched. While the below cgroup_can_be_thread_root()
3360 * test can catch the same conditions, that's only when @parent is
3361 * not mixable, so let's check it explicitly.
3363 if (cgroup_is_populated(cgrp) ||
3364 cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
3367 /* we're joining the parent's domain, ensure its validity */
3368 if (!cgroup_is_valid_domain(dom_cgrp) ||
3369 !cgroup_can_be_thread_root(dom_cgrp))
3373 * The following shouldn't cause actual migrations and should
3376 cgroup_save_control(cgrp);
3378 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)
3379 if (dsct == cgrp || cgroup_is_threaded(dsct))
3380 dsct->dom_cgrp = dom_cgrp;
3382 ret = cgroup_apply_control(cgrp);
3384 parent->nr_threaded_children++;
3386 cgroup_finalize_control(cgrp, ret);
3390 static int cgroup_type_show(struct seq_file *seq, void *v)
3392 struct cgroup *cgrp = seq_css(seq)->cgroup;
3394 if (cgroup_is_threaded(cgrp))
3395 seq_puts(seq, "threaded\n");
3396 else if (!cgroup_is_valid_domain(cgrp))
3397 seq_puts(seq, "domain invalid\n");
3398 else if (cgroup_is_thread_root(cgrp))
3399 seq_puts(seq, "domain threaded\n");
3401 seq_puts(seq, "domain\n");
3406 static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3407 size_t nbytes, loff_t off)
3409 struct cgroup *cgrp;
3412 /* only switching to threaded mode is supported */
3413 if (strcmp(strstrip(buf), "threaded"))
3416 /* drain dying csses before we re-apply (threaded) subtree control */
3417 cgrp = cgroup_kn_lock_live(of->kn, true);
3421 /* threaded can only be enabled */
3422 ret = cgroup_enable_threaded(cgrp);
3424 cgroup_kn_unlock(of->kn);
3425 return ret ?: nbytes;
3428 static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3430 struct cgroup *cgrp = seq_css(seq)->cgroup;
3431 int descendants = READ_ONCE(cgrp->max_descendants);
3433 if (descendants == INT_MAX)
3434 seq_puts(seq, "max\n");
3436 seq_printf(seq, "%d\n", descendants);
3441 static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3442 char *buf, size_t nbytes, loff_t off)
3444 struct cgroup *cgrp;
3448 buf = strstrip(buf);
3449 if (!strcmp(buf, "max")) {
3450 descendants = INT_MAX;
3452 ret = kstrtoint(buf, 0, &descendants);
3457 if (descendants < 0)
3460 cgrp = cgroup_kn_lock_live(of->kn, false);
3464 cgrp->max_descendants = descendants;
3466 cgroup_kn_unlock(of->kn);
3471 static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3473 struct cgroup *cgrp = seq_css(seq)->cgroup;
3474 int depth = READ_ONCE(cgrp->max_depth);
3476 if (depth == INT_MAX)
3477 seq_puts(seq, "max\n");
3479 seq_printf(seq, "%d\n", depth);
3484 static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3485 char *buf, size_t nbytes, loff_t off)
3487 struct cgroup *cgrp;
3491 buf = strstrip(buf);
3492 if (!strcmp(buf, "max")) {
3495 ret = kstrtoint(buf, 0, &depth);
3503 cgrp = cgroup_kn_lock_live(of->kn, false);
3507 cgrp->max_depth = depth;
3509 cgroup_kn_unlock(of->kn);
3514 static int cgroup_events_show(struct seq_file *seq, void *v)
3516 struct cgroup *cgrp = seq_css(seq)->cgroup;
3518 seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp));
3519 seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags));
3524 static int cgroup_stat_show(struct seq_file *seq, void *v)
3526 struct cgroup *cgroup = seq_css(seq)->cgroup;
3528 seq_printf(seq, "nr_descendants %d\n",
3529 cgroup->nr_descendants);
3530 seq_printf(seq, "nr_dying_descendants %d\n",
3531 cgroup->nr_dying_descendants);
3536 static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq,
3537 struct cgroup *cgrp, int ssid)
3539 struct cgroup_subsys *ss = cgroup_subsys[ssid];
3540 struct cgroup_subsys_state *css;
3543 if (!ss->css_extra_stat_show)
3546 css = cgroup_tryget_css(cgrp, ss);
3550 ret = ss->css_extra_stat_show(seq, css);
3555 static int cpu_stat_show(struct seq_file *seq, void *v)
3557 struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
3560 cgroup_base_stat_cputime_show(seq);
3561 #ifdef CONFIG_CGROUP_SCHED
3562 ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id);
3568 static int cgroup_io_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_IO);
3575 static int cgroup_memory_pressure_show(struct seq_file *seq, void *v)
3577 struct cgroup *cgrp = seq_css(seq)->cgroup;
3578 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3580 return psi_show(seq, psi, PSI_MEM);
3582 static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
3584 struct cgroup *cgrp = seq_css(seq)->cgroup;
3585 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3587 return psi_show(seq, psi, PSI_CPU);
3590 static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, char *buf,
3591 size_t nbytes, enum psi_res res)
3593 struct psi_trigger *new;
3594 struct cgroup *cgrp;
3595 struct psi_group *psi;
3597 cgrp = cgroup_kn_lock_live(of->kn, false);
3602 cgroup_kn_unlock(of->kn);
3604 psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3605 new = psi_trigger_create(psi, buf, nbytes, res);
3608 return PTR_ERR(new);
3611 psi_trigger_replace(&of->priv, new);
3618 static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of,
3619 char *buf, size_t nbytes,
3622 return cgroup_pressure_write(of, buf, nbytes, PSI_IO);
3625 static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of,
3626 char *buf, size_t nbytes,
3629 return cgroup_pressure_write(of, buf, nbytes, PSI_MEM);
3632 static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of,
3633 char *buf, size_t nbytes,
3636 return cgroup_pressure_write(of, buf, nbytes, PSI_CPU);
3639 static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of,
3642 return psi_trigger_poll(&of->priv, of->file, pt);
3645 static void cgroup_pressure_release(struct kernfs_open_file *of)
3647 psi_trigger_replace(&of->priv, NULL);
3649 #endif /* CONFIG_PSI */
3651 static int cgroup_freeze_show(struct seq_file *seq, void *v)
3653 struct cgroup *cgrp = seq_css(seq)->cgroup;
3655 seq_printf(seq, "%d\n", cgrp->freezer.freeze);
3660 static ssize_t cgroup_freeze_write(struct kernfs_open_file *of,
3661 char *buf, size_t nbytes, loff_t off)
3663 struct cgroup *cgrp;
3667 ret = kstrtoint(strstrip(buf), 0, &freeze);
3671 if (freeze < 0 || freeze > 1)
3674 cgrp = cgroup_kn_lock_live(of->kn, false);
3678 cgroup_freeze(cgrp, freeze);
3680 cgroup_kn_unlock(of->kn);
3685 static int cgroup_file_open(struct kernfs_open_file *of)
3687 struct cftype *cft = of->kn->priv;
3690 return cft->open(of);
3694 static void cgroup_file_release(struct kernfs_open_file *of)
3696 struct cftype *cft = of->kn->priv;
3702 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3703 size_t nbytes, loff_t off)
3705 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
3706 struct cgroup *cgrp = of->kn->parent->priv;
3707 struct cftype *cft = of->kn->priv;
3708 struct cgroup_subsys_state *css;
3715 * If namespaces are delegation boundaries, disallow writes to
3716 * files in an non-init namespace root from inside the namespace
3717 * except for the files explicitly marked delegatable -
3718 * cgroup.procs and cgroup.subtree_control.
3720 if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
3721 !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
3722 ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp)
3726 return cft->write(of, buf, nbytes, off);
3729 * kernfs guarantees that a file isn't deleted with operations in
3730 * flight, which means that the matching css is and stays alive and
3731 * doesn't need to be pinned. The RCU locking is not necessary
3732 * either. It's just for the convenience of using cgroup_css().
3735 css = cgroup_css(cgrp, cft->ss);
3738 if (cft->write_u64) {
3739 unsigned long long v;
3740 ret = kstrtoull(buf, 0, &v);
3742 ret = cft->write_u64(css, cft, v);
3743 } else if (cft->write_s64) {
3745 ret = kstrtoll(buf, 0, &v);
3747 ret = cft->write_s64(css, cft, v);
3752 return ret ?: nbytes;
3755 static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt)
3757 struct cftype *cft = of->kn->priv;
3760 return cft->poll(of, pt);
3762 return kernfs_generic_poll(of, pt);
3765 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3767 return seq_cft(seq)->seq_start(seq, ppos);
3770 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3772 return seq_cft(seq)->seq_next(seq, v, ppos);
3775 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3777 if (seq_cft(seq)->seq_stop)
3778 seq_cft(seq)->seq_stop(seq, v);
3781 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3783 struct cftype *cft = seq_cft(m);
3784 struct cgroup_subsys_state *css = seq_css(m);
3787 return cft->seq_show(m, arg);
3790 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3791 else if (cft->read_s64)
3792 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3798 static struct kernfs_ops cgroup_kf_single_ops = {
3799 .atomic_write_len = PAGE_SIZE,
3800 .open = cgroup_file_open,
3801 .release = cgroup_file_release,
3802 .write = cgroup_file_write,
3803 .poll = cgroup_file_poll,
3804 .seq_show = cgroup_seqfile_show,
3807 static struct kernfs_ops cgroup_kf_ops = {
3808 .atomic_write_len = PAGE_SIZE,
3809 .open = cgroup_file_open,
3810 .release = cgroup_file_release,
3811 .write = cgroup_file_write,
3812 .poll = cgroup_file_poll,
3813 .seq_start = cgroup_seqfile_start,
3814 .seq_next = cgroup_seqfile_next,
3815 .seq_stop = cgroup_seqfile_stop,
3816 .seq_show = cgroup_seqfile_show,
3819 /* set uid and gid of cgroup dirs and files to that of the creator */
3820 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3822 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3823 .ia_uid = current_fsuid(),
3824 .ia_gid = current_fsgid(), };
3826 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3827 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3830 return kernfs_setattr(kn, &iattr);
3833 static void cgroup_file_notify_timer(struct timer_list *timer)
3835 cgroup_file_notify(container_of(timer, struct cgroup_file,
3839 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3842 char name[CGROUP_FILE_NAME_MAX];
3843 struct kernfs_node *kn;
3844 struct lock_class_key *key = NULL;
3847 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3848 key = &cft->lockdep_key;
3850 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3851 cgroup_file_mode(cft),
3852 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
3853 0, cft->kf_ops, cft,
3858 ret = cgroup_kn_set_ugid(kn);
3864 if (cft->file_offset) {
3865 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3867 timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0);
3869 spin_lock_irq(&cgroup_file_kn_lock);
3871 spin_unlock_irq(&cgroup_file_kn_lock);
3878 * cgroup_addrm_files - add or remove files to a cgroup directory
3879 * @css: the target css
3880 * @cgrp: the target cgroup (usually css->cgroup)
3881 * @cfts: array of cftypes to be added
3882 * @is_add: whether to add or remove
3884 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3885 * For removals, this function never fails.
3887 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3888 struct cgroup *cgrp, struct cftype cfts[],
3891 struct cftype *cft, *cft_end = NULL;
3894 lockdep_assert_held(&cgroup_mutex);
3897 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3898 /* does cft->flags tell us to skip this file on @cgrp? */
3899 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3901 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3903 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3905 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3907 if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug)
3910 ret = cgroup_add_file(css, cgrp, cft);
3912 pr_warn("%s: failed to add %s, err=%d\n",
3913 __func__, cft->name, ret);
3919 cgroup_rm_file(cgrp, cft);
3925 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3927 struct cgroup_subsys *ss = cfts[0].ss;
3928 struct cgroup *root = &ss->root->cgrp;
3929 struct cgroup_subsys_state *css;
3932 lockdep_assert_held(&cgroup_mutex);
3934 /* add/rm files for all cgroups created before */
3935 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3936 struct cgroup *cgrp = css->cgroup;
3938 if (!(css->flags & CSS_VISIBLE))
3941 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3947 kernfs_activate(root->kn);
3951 static void cgroup_exit_cftypes(struct cftype *cfts)
3955 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3956 /* free copy for custom atomic_write_len, see init_cftypes() */
3957 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3962 /* revert flags set by cgroup core while adding @cfts */
3963 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
3967 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3971 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3972 struct kernfs_ops *kf_ops;
3974 WARN_ON(cft->ss || cft->kf_ops);
3977 kf_ops = &cgroup_kf_ops;
3979 kf_ops = &cgroup_kf_single_ops;
3982 * Ugh... if @cft wants a custom max_write_len, we need to
3983 * make a copy of kf_ops to set its atomic_write_len.
3985 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3986 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3988 cgroup_exit_cftypes(cfts);
3991 kf_ops->atomic_write_len = cft->max_write_len;
3994 cft->kf_ops = kf_ops;
4001 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
4003 lockdep_assert_held(&cgroup_mutex);
4005 if (!cfts || !cfts[0].ss)
4008 list_del(&cfts->node);
4009 cgroup_apply_cftypes(cfts, false);
4010 cgroup_exit_cftypes(cfts);
4015 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4016 * @cfts: zero-length name terminated array of cftypes
4018 * Unregister @cfts. Files described by @cfts are removed from all
4019 * existing cgroups and all future cgroups won't have them either. This
4020 * function can be called anytime whether @cfts' subsys is attached or not.
4022 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4025 int cgroup_rm_cftypes(struct cftype *cfts)
4029 mutex_lock(&cgroup_mutex);
4030 ret = cgroup_rm_cftypes_locked(cfts);
4031 mutex_unlock(&cgroup_mutex);
4036 * cgroup_add_cftypes - add an array of cftypes to a subsystem
4037 * @ss: target cgroup subsystem
4038 * @cfts: zero-length name terminated array of cftypes
4040 * Register @cfts to @ss. Files described by @cfts are created for all
4041 * existing cgroups to which @ss is attached and all future cgroups will
4042 * have them too. This function can be called anytime whether @ss is
4045 * Returns 0 on successful registration, -errno on failure. Note that this
4046 * function currently returns 0 as long as @cfts registration is successful
4047 * even if some file creation attempts on existing cgroups fail.
4049 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4053 if (!cgroup_ssid_enabled(ss->id))
4056 if (!cfts || cfts[0].name[0] == '\0')
4059 ret = cgroup_init_cftypes(ss, cfts);
4063 mutex_lock(&cgroup_mutex);
4065 list_add_tail(&cfts->node, &ss->cfts);
4066 ret = cgroup_apply_cftypes(cfts, true);
4068 cgroup_rm_cftypes_locked(cfts);
4070 mutex_unlock(&cgroup_mutex);
4075 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4076 * @ss: target cgroup subsystem
4077 * @cfts: zero-length name terminated array of cftypes
4079 * Similar to cgroup_add_cftypes() but the added files are only used for
4080 * the default hierarchy.
4082 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4086 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4087 cft->flags |= __CFTYPE_ONLY_ON_DFL;
4088 return cgroup_add_cftypes(ss, cfts);
4092 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4093 * @ss: target cgroup subsystem
4094 * @cfts: zero-length name terminated array of cftypes
4096 * Similar to cgroup_add_cftypes() but the added files are only used for
4097 * the legacy hierarchies.
4099 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4103 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4104 cft->flags |= __CFTYPE_NOT_ON_DFL;
4105 return cgroup_add_cftypes(ss, cfts);
4109 * cgroup_file_notify - generate a file modified event for a cgroup_file
4110 * @cfile: target cgroup_file
4112 * @cfile must have been obtained by setting cftype->file_offset.
4114 void cgroup_file_notify(struct cgroup_file *cfile)
4116 unsigned long flags;
4118 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
4120 unsigned long last = cfile->notified_at;
4121 unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV;
4123 if (time_in_range(jiffies, last, next)) {
4124 timer_reduce(&cfile->notify_timer, next);
4126 kernfs_notify(cfile->kn);
4127 cfile->notified_at = jiffies;
4130 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
4134 * css_next_child - find the next child of a given css
4135 * @pos: the current position (%NULL to initiate traversal)
4136 * @parent: css whose children to walk
4138 * This function returns the next child of @parent and should be called
4139 * under either cgroup_mutex or RCU read lock. The only requirement is
4140 * that @parent and @pos are accessible. The next sibling is guaranteed to
4141 * be returned regardless of their states.
4143 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4144 * css which finished ->css_online() is guaranteed to be visible in the
4145 * future iterations and will stay visible until the last reference is put.
4146 * A css which hasn't finished ->css_online() or already finished
4147 * ->css_offline() may show up during traversal. It's each subsystem's
4148 * responsibility to synchronize against on/offlining.
4150 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
4151 struct cgroup_subsys_state *parent)
4153 struct cgroup_subsys_state *next;
4155 cgroup_assert_mutex_or_rcu_locked();
4158 * @pos could already have been unlinked from the sibling list.
4159 * Once a cgroup is removed, its ->sibling.next is no longer
4160 * updated when its next sibling changes. CSS_RELEASED is set when
4161 * @pos is taken off list, at which time its next pointer is valid,
4162 * and, as releases are serialized, the one pointed to by the next
4163 * pointer is guaranteed to not have started release yet. This
4164 * implies that if we observe !CSS_RELEASED on @pos in this RCU
4165 * critical section, the one pointed to by its next pointer is
4166 * guaranteed to not have finished its RCU grace period even if we
4167 * have dropped rcu_read_lock() inbetween iterations.
4169 * If @pos has CSS_RELEASED set, its next pointer can't be
4170 * dereferenced; however, as each css is given a monotonically
4171 * increasing unique serial number and always appended to the
4172 * sibling list, the next one can be found by walking the parent's
4173 * children until the first css with higher serial number than
4174 * @pos's. While this path can be slower, it happens iff iteration
4175 * races against release and the race window is very small.
4178 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
4179 } else if (likely(!(pos->flags & CSS_RELEASED))) {
4180 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
4182 list_for_each_entry_rcu(next, &parent->children, sibling,
4183 lockdep_is_held(&cgroup_mutex))
4184 if (next->serial_nr > pos->serial_nr)
4189 * @next, if not pointing to the head, can be dereferenced and is
4192 if (&next->sibling != &parent->children)
4198 * css_next_descendant_pre - find the next descendant for pre-order walk
4199 * @pos: the current position (%NULL to initiate traversal)
4200 * @root: css whose descendants to walk
4202 * To be used by css_for_each_descendant_pre(). Find the next descendant
4203 * to visit for pre-order traversal of @root's descendants. @root is
4204 * included in the iteration and the first node to be visited.
4206 * While this function requires cgroup_mutex or RCU read locking, it
4207 * doesn't require the whole traversal to be contained in a single critical
4208 * section. This function will return the correct next descendant as long
4209 * as both @pos and @root are accessible and @pos is a descendant of @root.
4211 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4212 * css which finished ->css_online() is guaranteed to be visible in the
4213 * future iterations and will stay visible until the last reference is put.
4214 * A css which hasn't finished ->css_online() or already finished
4215 * ->css_offline() may show up during traversal. It's each subsystem's
4216 * responsibility to synchronize against on/offlining.
4218 struct cgroup_subsys_state *
4219 css_next_descendant_pre(struct cgroup_subsys_state *pos,
4220 struct cgroup_subsys_state *root)
4222 struct cgroup_subsys_state *next;
4224 cgroup_assert_mutex_or_rcu_locked();
4226 /* if first iteration, visit @root */
4230 /* visit the first child if exists */
4231 next = css_next_child(NULL, pos);
4235 /* no child, visit my or the closest ancestor's next sibling */
4236 while (pos != root) {
4237 next = css_next_child(pos, pos->parent);
4245 EXPORT_SYMBOL_GPL(css_next_descendant_pre);
4248 * css_rightmost_descendant - return the rightmost descendant of a css
4249 * @pos: css of interest
4251 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4252 * is returned. This can be used during pre-order traversal to skip
4255 * While this function requires cgroup_mutex or RCU read locking, it
4256 * doesn't require the whole traversal to be contained in a single critical
4257 * section. This function will return the correct rightmost descendant as
4258 * long as @pos is accessible.
4260 struct cgroup_subsys_state *
4261 css_rightmost_descendant(struct cgroup_subsys_state *pos)
4263 struct cgroup_subsys_state *last, *tmp;
4265 cgroup_assert_mutex_or_rcu_locked();
4269 /* ->prev isn't RCU safe, walk ->next till the end */
4271 css_for_each_child(tmp, last)
4278 static struct cgroup_subsys_state *
4279 css_leftmost_descendant(struct cgroup_subsys_state *pos)
4281 struct cgroup_subsys_state *last;
4285 pos = css_next_child(NULL, pos);
4292 * css_next_descendant_post - find the next descendant for post-order walk
4293 * @pos: the current position (%NULL to initiate traversal)
4294 * @root: css whose descendants to walk
4296 * To be used by css_for_each_descendant_post(). Find the next descendant
4297 * to visit for post-order traversal of @root's descendants. @root is
4298 * included in the iteration and the last node to be visited.
4300 * While this function requires cgroup_mutex or RCU read locking, it
4301 * doesn't require the whole traversal to be contained in a single critical
4302 * section. This function will return the correct next descendant as long
4303 * as both @pos and @cgroup are accessible and @pos is a descendant of
4306 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4307 * css which finished ->css_online() is guaranteed to be visible in the
4308 * future iterations and will stay visible until the last reference is put.
4309 * A css which hasn't finished ->css_online() or already finished
4310 * ->css_offline() may show up during traversal. It's each subsystem's
4311 * responsibility to synchronize against on/offlining.
4313 struct cgroup_subsys_state *
4314 css_next_descendant_post(struct cgroup_subsys_state *pos,
4315 struct cgroup_subsys_state *root)
4317 struct cgroup_subsys_state *next;
4319 cgroup_assert_mutex_or_rcu_locked();
4321 /* if first iteration, visit leftmost descendant which may be @root */
4323 return css_leftmost_descendant(root);
4325 /* if we visited @root, we're done */
4329 /* if there's an unvisited sibling, visit its leftmost descendant */
4330 next = css_next_child(pos, pos->parent);
4332 return css_leftmost_descendant(next);
4334 /* no sibling left, visit parent */
4339 * css_has_online_children - does a css have online children
4340 * @css: the target css
4342 * Returns %true if @css has any online children; otherwise, %false. This
4343 * function can be called from any context but the caller is responsible
4344 * for synchronizing against on/offlining as necessary.
4346 bool css_has_online_children(struct cgroup_subsys_state *css)
4348 struct cgroup_subsys_state *child;
4352 css_for_each_child(child, css) {
4353 if (child->flags & CSS_ONLINE) {
4362 static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
4364 struct list_head *l;
4365 struct cgrp_cset_link *link;
4366 struct css_set *cset;
4368 lockdep_assert_held(&css_set_lock);
4370 /* find the next threaded cset */
4371 if (it->tcset_pos) {
4372 l = it->tcset_pos->next;
4374 if (l != it->tcset_head) {
4376 return container_of(l, struct css_set,
4377 threaded_csets_node);
4380 it->tcset_pos = NULL;
4383 /* find the next cset */
4386 if (l == it->cset_head) {
4387 it->cset_pos = NULL;
4392 cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
4394 link = list_entry(l, struct cgrp_cset_link, cset_link);
4400 /* initialize threaded css_set walking */
4401 if (it->flags & CSS_TASK_ITER_THREADED) {
4403 put_css_set_locked(it->cur_dcset);
4404 it->cur_dcset = cset;
4407 it->tcset_head = &cset->threaded_csets;
4408 it->tcset_pos = &cset->threaded_csets;
4415 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
4416 * @it: the iterator to advance
4418 * Advance @it to the next css_set to walk.
4420 static void css_task_iter_advance_css_set(struct css_task_iter *it)
4422 struct css_set *cset;
4424 lockdep_assert_held(&css_set_lock);
4426 /* Advance to the next non-empty css_set and find first non-empty tasks list*/
4427 while ((cset = css_task_iter_next_css_set(it))) {
4428 if (!list_empty(&cset->tasks)) {
4429 it->cur_tasks_head = &cset->tasks;
4431 } else if (!list_empty(&cset->mg_tasks)) {
4432 it->cur_tasks_head = &cset->mg_tasks;
4434 } else if (!list_empty(&cset->dying_tasks)) {
4435 it->cur_tasks_head = &cset->dying_tasks;
4440 it->task_pos = NULL;
4443 it->task_pos = it->cur_tasks_head->next;
4446 * We don't keep css_sets locked across iteration steps and thus
4447 * need to take steps to ensure that iteration can be resumed after
4448 * the lock is re-acquired. Iteration is performed at two levels -
4449 * css_sets and tasks in them.
4451 * Once created, a css_set never leaves its cgroup lists, so a
4452 * pinned css_set is guaranteed to stay put and we can resume
4453 * iteration afterwards.
4455 * Tasks may leave @cset across iteration steps. This is resolved
4456 * by registering each iterator with the css_set currently being
4457 * walked and making css_set_move_task() advance iterators whose
4458 * next task is leaving.
4461 list_del(&it->iters_node);
4462 put_css_set_locked(it->cur_cset);
4465 it->cur_cset = cset;
4466 list_add(&it->iters_node, &cset->task_iters);
4469 static void css_task_iter_skip(struct css_task_iter *it,
4470 struct task_struct *task)
4472 lockdep_assert_held(&css_set_lock);
4474 if (it->task_pos == &task->cg_list) {
4475 it->task_pos = it->task_pos->next;
4476 it->flags |= CSS_TASK_ITER_SKIPPED;
4480 static void css_task_iter_advance(struct css_task_iter *it)
4482 struct task_struct *task;
4484 lockdep_assert_held(&css_set_lock);
4488 * Advance iterator to find next entry. We go through cset
4489 * tasks, mg_tasks and dying_tasks, when consumed we move onto
4492 if (it->flags & CSS_TASK_ITER_SKIPPED)
4493 it->flags &= ~CSS_TASK_ITER_SKIPPED;
4495 it->task_pos = it->task_pos->next;
4497 if (it->task_pos == &it->cur_cset->tasks) {
4498 it->cur_tasks_head = &it->cur_cset->mg_tasks;
4499 it->task_pos = it->cur_tasks_head->next;
4501 if (it->task_pos == &it->cur_cset->mg_tasks) {
4502 it->cur_tasks_head = &it->cur_cset->dying_tasks;
4503 it->task_pos = it->cur_tasks_head->next;
4505 if (it->task_pos == &it->cur_cset->dying_tasks)
4506 css_task_iter_advance_css_set(it);
4508 /* called from start, proceed to the first cset */
4509 css_task_iter_advance_css_set(it);
4515 task = list_entry(it->task_pos, struct task_struct, cg_list);
4517 if (it->flags & CSS_TASK_ITER_PROCS) {
4518 /* if PROCS, skip over tasks which aren't group leaders */
4519 if (!thread_group_leader(task))
4522 /* and dying leaders w/o live member threads */
4523 if (it->cur_tasks_head == &it->cur_cset->dying_tasks &&
4524 !atomic_read(&task->signal->live))
4527 /* skip all dying ones */
4528 if (it->cur_tasks_head == &it->cur_cset->dying_tasks)
4534 * css_task_iter_start - initiate task iteration
4535 * @css: the css to walk tasks of
4536 * @flags: CSS_TASK_ITER_* flags
4537 * @it: the task iterator to use
4539 * Initiate iteration through the tasks of @css. The caller can call
4540 * css_task_iter_next() to walk through the tasks until the function
4541 * returns NULL. On completion of iteration, css_task_iter_end() must be
4544 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
4545 struct css_task_iter *it)
4547 memset(it, 0, sizeof(*it));
4549 spin_lock_irq(&css_set_lock);
4555 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4557 it->cset_pos = &css->cgroup->cset_links;
4559 it->cset_head = it->cset_pos;
4561 css_task_iter_advance(it);
4563 spin_unlock_irq(&css_set_lock);
4567 * css_task_iter_next - return the next task for the iterator
4568 * @it: the task iterator being iterated
4570 * The "next" function for task iteration. @it should have been
4571 * initialized via css_task_iter_start(). Returns NULL when the iteration
4574 struct task_struct *css_task_iter_next(struct css_task_iter *it)
4577 put_task_struct(it->cur_task);
4578 it->cur_task = NULL;
4581 spin_lock_irq(&css_set_lock);
4583 /* @it may be half-advanced by skips, finish advancing */
4584 if (it->flags & CSS_TASK_ITER_SKIPPED)
4585 css_task_iter_advance(it);
4588 it->cur_task = list_entry(it->task_pos, struct task_struct,
4590 get_task_struct(it->cur_task);
4591 css_task_iter_advance(it);
4594 spin_unlock_irq(&css_set_lock);
4596 return it->cur_task;
4600 * css_task_iter_end - finish task iteration
4601 * @it: the task iterator to finish
4603 * Finish task iteration started by css_task_iter_start().
4605 void css_task_iter_end(struct css_task_iter *it)
4608 spin_lock_irq(&css_set_lock);
4609 list_del(&it->iters_node);
4610 put_css_set_locked(it->cur_cset);
4611 spin_unlock_irq(&css_set_lock);
4615 put_css_set(it->cur_dcset);
4618 put_task_struct(it->cur_task);
4621 static void cgroup_procs_release(struct kernfs_open_file *of)
4624 css_task_iter_end(of->priv);
4629 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
4631 struct kernfs_open_file *of = s->private;
4632 struct css_task_iter *it = of->priv;
4637 return css_task_iter_next(it);
4640 static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
4641 unsigned int iter_flags)
4643 struct kernfs_open_file *of = s->private;
4644 struct cgroup *cgrp = seq_css(s)->cgroup;
4645 struct css_task_iter *it = of->priv;
4648 * When a seq_file is seeked, it's always traversed sequentially
4649 * from position 0, so we can simply keep iterating on !0 *pos.
4652 if (WARN_ON_ONCE((*pos)))
4653 return ERR_PTR(-EINVAL);
4655 it = kzalloc(sizeof(*it), GFP_KERNEL);
4657 return ERR_PTR(-ENOMEM);
4659 css_task_iter_start(&cgrp->self, iter_flags, it);
4660 } else if (!(*pos)) {
4661 css_task_iter_end(it);
4662 css_task_iter_start(&cgrp->self, iter_flags, it);
4664 return it->cur_task;
4666 return cgroup_procs_next(s, NULL, NULL);
4669 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
4671 struct cgroup *cgrp = seq_css(s)->cgroup;
4674 * All processes of a threaded subtree belong to the domain cgroup
4675 * of the subtree. Only threads can be distributed across the
4676 * subtree. Reject reads on cgroup.procs in the subtree proper.
4677 * They're always empty anyway.
4679 if (cgroup_is_threaded(cgrp))
4680 return ERR_PTR(-EOPNOTSUPP);
4682 return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
4683 CSS_TASK_ITER_THREADED);
4686 static int cgroup_procs_show(struct seq_file *s, void *v)
4688 seq_printf(s, "%d\n", task_pid_vnr(v));
4692 static int cgroup_may_write(const struct cgroup *cgrp, struct super_block *sb)
4695 struct inode *inode;
4697 lockdep_assert_held(&cgroup_mutex);
4699 inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
4703 ret = inode_permission(inode, MAY_WRITE);
4708 static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
4709 struct cgroup *dst_cgrp,
4710 struct super_block *sb)
4712 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
4713 struct cgroup *com_cgrp = src_cgrp;
4716 lockdep_assert_held(&cgroup_mutex);
4718 /* find the common ancestor */
4719 while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
4720 com_cgrp = cgroup_parent(com_cgrp);
4722 /* %current should be authorized to migrate to the common ancestor */
4723 ret = cgroup_may_write(com_cgrp, sb);
4728 * If namespaces are delegation boundaries, %current must be able
4729 * to see both source and destination cgroups from its namespace.
4731 if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
4732 (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
4733 !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
4739 static int cgroup_attach_permissions(struct cgroup *src_cgrp,
4740 struct cgroup *dst_cgrp,
4741 struct super_block *sb, bool threadgroup)
4745 ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, sb);
4749 ret = cgroup_migrate_vet_dst(dst_cgrp);
4753 if (!threadgroup && (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp))
4759 static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
4760 char *buf, size_t nbytes, loff_t off)
4762 struct cgroup *src_cgrp, *dst_cgrp;
4763 struct task_struct *task;
4767 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4771 task = cgroup_procs_write_start(buf, true, &locked);
4772 ret = PTR_ERR_OR_ZERO(task);
4776 /* find the source cgroup */
4777 spin_lock_irq(&css_set_lock);
4778 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4779 spin_unlock_irq(&css_set_lock);
4781 ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
4782 of->file->f_path.dentry->d_sb, true);
4786 ret = cgroup_attach_task(dst_cgrp, task, true);
4789 cgroup_procs_write_finish(task, locked);
4791 cgroup_kn_unlock(of->kn);
4793 return ret ?: nbytes;
4796 static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
4798 return __cgroup_procs_start(s, pos, 0);
4801 static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
4802 char *buf, size_t nbytes, loff_t off)
4804 struct cgroup *src_cgrp, *dst_cgrp;
4805 struct task_struct *task;
4809 buf = strstrip(buf);
4811 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4815 task = cgroup_procs_write_start(buf, false, &locked);
4816 ret = PTR_ERR_OR_ZERO(task);
4820 /* find the source cgroup */
4821 spin_lock_irq(&css_set_lock);
4822 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4823 spin_unlock_irq(&css_set_lock);
4825 /* thread migrations follow the cgroup.procs delegation rule */
4826 ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
4827 of->file->f_path.dentry->d_sb, false);
4831 ret = cgroup_attach_task(dst_cgrp, task, false);
4834 cgroup_procs_write_finish(task, locked);
4836 cgroup_kn_unlock(of->kn);
4838 return ret ?: nbytes;
4841 /* cgroup core interface files for the default hierarchy */
4842 static struct cftype cgroup_base_files[] = {
4844 .name = "cgroup.type",
4845 .flags = CFTYPE_NOT_ON_ROOT,
4846 .seq_show = cgroup_type_show,
4847 .write = cgroup_type_write,
4850 .name = "cgroup.procs",
4851 .flags = CFTYPE_NS_DELEGATABLE,
4852 .file_offset = offsetof(struct cgroup, procs_file),
4853 .release = cgroup_procs_release,
4854 .seq_start = cgroup_procs_start,
4855 .seq_next = cgroup_procs_next,
4856 .seq_show = cgroup_procs_show,
4857 .write = cgroup_procs_write,
4860 .name = "cgroup.threads",
4861 .flags = CFTYPE_NS_DELEGATABLE,
4862 .release = cgroup_procs_release,
4863 .seq_start = cgroup_threads_start,
4864 .seq_next = cgroup_procs_next,
4865 .seq_show = cgroup_procs_show,
4866 .write = cgroup_threads_write,
4869 .name = "cgroup.controllers",
4870 .seq_show = cgroup_controllers_show,
4873 .name = "cgroup.subtree_control",
4874 .flags = CFTYPE_NS_DELEGATABLE,
4875 .seq_show = cgroup_subtree_control_show,
4876 .write = cgroup_subtree_control_write,
4879 .name = "cgroup.events",
4880 .flags = CFTYPE_NOT_ON_ROOT,
4881 .file_offset = offsetof(struct cgroup, events_file),
4882 .seq_show = cgroup_events_show,
4885 .name = "cgroup.max.descendants",
4886 .seq_show = cgroup_max_descendants_show,
4887 .write = cgroup_max_descendants_write,
4890 .name = "cgroup.max.depth",
4891 .seq_show = cgroup_max_depth_show,
4892 .write = cgroup_max_depth_write,
4895 .name = "cgroup.stat",
4896 .seq_show = cgroup_stat_show,
4899 .name = "cgroup.freeze",
4900 .flags = CFTYPE_NOT_ON_ROOT,
4901 .seq_show = cgroup_freeze_show,
4902 .write = cgroup_freeze_write,
4906 .seq_show = cpu_stat_show,
4910 .name = "io.pressure",
4911 .seq_show = cgroup_io_pressure_show,
4912 .write = cgroup_io_pressure_write,
4913 .poll = cgroup_pressure_poll,
4914 .release = cgroup_pressure_release,
4917 .name = "memory.pressure",
4918 .seq_show = cgroup_memory_pressure_show,
4919 .write = cgroup_memory_pressure_write,
4920 .poll = cgroup_pressure_poll,
4921 .release = cgroup_pressure_release,
4924 .name = "cpu.pressure",
4925 .seq_show = cgroup_cpu_pressure_show,
4926 .write = cgroup_cpu_pressure_write,
4927 .poll = cgroup_pressure_poll,
4928 .release = cgroup_pressure_release,
4930 #endif /* CONFIG_PSI */
4935 * css destruction is four-stage process.
4937 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4938 * Implemented in kill_css().
4940 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4941 * and thus css_tryget_online() is guaranteed to fail, the css can be
4942 * offlined by invoking offline_css(). After offlining, the base ref is
4943 * put. Implemented in css_killed_work_fn().
4945 * 3. When the percpu_ref reaches zero, the only possible remaining
4946 * accessors are inside RCU read sections. css_release() schedules the
4949 * 4. After the grace period, the css can be freed. Implemented in
4950 * css_free_work_fn().
4952 * It is actually hairier because both step 2 and 4 require process context
4953 * and thus involve punting to css->destroy_work adding two additional
4954 * steps to the already complex sequence.
4956 static void css_free_rwork_fn(struct work_struct *work)
4958 struct cgroup_subsys_state *css = container_of(to_rcu_work(work),
4959 struct cgroup_subsys_state, destroy_rwork);
4960 struct cgroup_subsys *ss = css->ss;
4961 struct cgroup *cgrp = css->cgroup;
4963 percpu_ref_exit(&css->refcnt);
4967 struct cgroup_subsys_state *parent = css->parent;
4971 cgroup_idr_remove(&ss->css_idr, id);
4977 /* cgroup free path */
4978 atomic_dec(&cgrp->root->nr_cgrps);
4979 cgroup1_pidlist_destroy_all(cgrp);
4980 cancel_work_sync(&cgrp->release_agent_work);
4982 if (cgroup_parent(cgrp)) {
4984 * We get a ref to the parent, and put the ref when
4985 * this cgroup is being freed, so it's guaranteed
4986 * that the parent won't be destroyed before its
4989 cgroup_put(cgroup_parent(cgrp));
4990 kernfs_put(cgrp->kn);
4991 psi_cgroup_free(cgrp);
4992 if (cgroup_on_dfl(cgrp))
4993 cgroup_rstat_exit(cgrp);
4997 * This is root cgroup's refcnt reaching zero,
4998 * which indicates that the root should be
5001 cgroup_destroy_root(cgrp->root);
5006 static void css_release_work_fn(struct work_struct *work)
5008 struct cgroup_subsys_state *css =
5009 container_of(work, struct cgroup_subsys_state, destroy_work);
5010 struct cgroup_subsys *ss = css->ss;
5011 struct cgroup *cgrp = css->cgroup;
5013 mutex_lock(&cgroup_mutex);
5015 css->flags |= CSS_RELEASED;
5016 list_del_rcu(&css->sibling);
5019 /* css release path */
5020 if (!list_empty(&css->rstat_css_node)) {
5021 cgroup_rstat_flush(cgrp);
5022 list_del_rcu(&css->rstat_css_node);
5025 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
5026 if (ss->css_released)
5027 ss->css_released(css);
5029 struct cgroup *tcgrp;
5031 /* cgroup release path */
5032 TRACE_CGROUP_PATH(release, cgrp);
5034 if (cgroup_on_dfl(cgrp))
5035 cgroup_rstat_flush(cgrp);
5037 spin_lock_irq(&css_set_lock);
5038 for (tcgrp = cgroup_parent(cgrp); tcgrp;
5039 tcgrp = cgroup_parent(tcgrp))
5040 tcgrp->nr_dying_descendants--;
5041 spin_unlock_irq(&css_set_lock);
5044 * There are two control paths which try to determine
5045 * cgroup from dentry without going through kernfs -
5046 * cgroupstats_build() and css_tryget_online_from_dir().
5047 * Those are supported by RCU protecting clearing of
5048 * cgrp->kn->priv backpointer.
5051 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
5055 mutex_unlock(&cgroup_mutex);
5057 INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5058 queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5061 static void css_release(struct percpu_ref *ref)
5063 struct cgroup_subsys_state *css =
5064 container_of(ref, struct cgroup_subsys_state, refcnt);
5066 INIT_WORK(&css->destroy_work, css_release_work_fn);
5067 queue_work(cgroup_destroy_wq, &css->destroy_work);
5070 static void init_and_link_css(struct cgroup_subsys_state *css,
5071 struct cgroup_subsys *ss, struct cgroup *cgrp)
5073 lockdep_assert_held(&cgroup_mutex);
5075 cgroup_get_live(cgrp);
5077 memset(css, 0, sizeof(*css));
5081 INIT_LIST_HEAD(&css->sibling);
5082 INIT_LIST_HEAD(&css->children);
5083 INIT_LIST_HEAD(&css->rstat_css_node);
5084 css->serial_nr = css_serial_nr_next++;
5085 atomic_set(&css->online_cnt, 0);
5087 if (cgroup_parent(cgrp)) {
5088 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
5089 css_get(css->parent);
5092 if (cgroup_on_dfl(cgrp) && ss->css_rstat_flush)
5093 list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list);
5095 BUG_ON(cgroup_css(cgrp, ss));
5098 /* invoke ->css_online() on a new CSS and mark it online if successful */
5099 static int online_css(struct cgroup_subsys_state *css)
5101 struct cgroup_subsys *ss = css->ss;
5104 lockdep_assert_held(&cgroup_mutex);
5107 ret = ss->css_online(css);
5109 css->flags |= CSS_ONLINE;
5110 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
5112 atomic_inc(&css->online_cnt);
5114 atomic_inc(&css->parent->online_cnt);
5119 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5120 static void offline_css(struct cgroup_subsys_state *css)
5122 struct cgroup_subsys *ss = css->ss;
5124 lockdep_assert_held(&cgroup_mutex);
5126 if (!(css->flags & CSS_ONLINE))
5129 if (ss->css_offline)
5130 ss->css_offline(css);
5132 css->flags &= ~CSS_ONLINE;
5133 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
5135 wake_up_all(&css->cgroup->offline_waitq);
5139 * css_create - create a cgroup_subsys_state
5140 * @cgrp: the cgroup new css will be associated with
5141 * @ss: the subsys of new css
5143 * Create a new css associated with @cgrp - @ss pair. On success, the new
5144 * css is online and installed in @cgrp. This function doesn't create the
5145 * interface files. Returns 0 on success, -errno on failure.
5147 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5148 struct cgroup_subsys *ss)
5150 struct cgroup *parent = cgroup_parent(cgrp);
5151 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
5152 struct cgroup_subsys_state *css;
5155 lockdep_assert_held(&cgroup_mutex);
5157 css = ss->css_alloc(parent_css);
5159 css = ERR_PTR(-ENOMEM);
5163 init_and_link_css(css, ss, cgrp);
5165 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
5169 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
5174 /* @css is ready to be brought online now, make it visible */
5175 list_add_tail_rcu(&css->sibling, &parent_css->children);
5176 cgroup_idr_replace(&ss->css_idr, css, css->id);
5178 err = online_css(css);
5182 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
5183 cgroup_parent(parent)) {
5184 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
5185 current->comm, current->pid, ss->name);
5186 if (!strcmp(ss->name, "memory"))
5187 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
5188 ss->warned_broken_hierarchy = true;
5194 list_del_rcu(&css->sibling);
5196 list_del_rcu(&css->rstat_css_node);
5197 INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5198 queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5199 return ERR_PTR(err);
5203 * The returned cgroup is fully initialized including its control mask, but
5204 * it isn't associated with its kernfs_node and doesn't have the control
5207 static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
5210 struct cgroup_root *root = parent->root;
5211 struct cgroup *cgrp, *tcgrp;
5212 struct kernfs_node *kn;
5213 int level = parent->level + 1;
5216 /* allocate the cgroup and its ID, 0 is reserved for the root */
5217 cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
5220 return ERR_PTR(-ENOMEM);
5222 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
5226 if (cgroup_on_dfl(parent)) {
5227 ret = cgroup_rstat_init(cgrp);
5229 goto out_cancel_ref;
5232 /* create the directory */
5233 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5240 init_cgroup_housekeeping(cgrp);
5242 cgrp->self.parent = &parent->self;
5244 cgrp->level = level;
5246 ret = psi_cgroup_alloc(cgrp);
5248 goto out_kernfs_remove;
5250 ret = cgroup_bpf_inherit(cgrp);
5255 * New cgroup inherits effective freeze counter, and
5256 * if the parent has to be frozen, the child has too.
5258 cgrp->freezer.e_freeze = parent->freezer.e_freeze;
5259 if (cgrp->freezer.e_freeze) {
5261 * Set the CGRP_FREEZE flag, so when a process will be
5262 * attached to the child cgroup, it will become frozen.
5263 * At this point the new cgroup is unpopulated, so we can
5264 * consider it frozen immediately.
5266 set_bit(CGRP_FREEZE, &cgrp->flags);
5267 set_bit(CGRP_FROZEN, &cgrp->flags);
5270 spin_lock_irq(&css_set_lock);
5271 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5272 cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
5274 if (tcgrp != cgrp) {
5275 tcgrp->nr_descendants++;
5278 * If the new cgroup is frozen, all ancestor cgroups
5279 * get a new frozen descendant, but their state can't
5280 * change because of this.
5282 if (cgrp->freezer.e_freeze)
5283 tcgrp->freezer.nr_frozen_descendants++;
5286 spin_unlock_irq(&css_set_lock);
5288 if (notify_on_release(parent))
5289 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5291 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5292 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
5294 cgrp->self.serial_nr = css_serial_nr_next++;
5296 /* allocation complete, commit to creation */
5297 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
5298 atomic_inc(&root->nr_cgrps);
5299 cgroup_get_live(parent);
5302 * On the default hierarchy, a child doesn't automatically inherit
5303 * subtree_control from the parent. Each is configured manually.
5305 if (!cgroup_on_dfl(cgrp))
5306 cgrp->subtree_control = cgroup_control(cgrp);
5308 cgroup_propagate_control(cgrp);
5313 psi_cgroup_free(cgrp);
5315 kernfs_remove(cgrp->kn);
5317 if (cgroup_on_dfl(parent))
5318 cgroup_rstat_exit(cgrp);
5320 percpu_ref_exit(&cgrp->self.refcnt);
5323 return ERR_PTR(ret);
5326 static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
5328 struct cgroup *cgroup;
5332 lockdep_assert_held(&cgroup_mutex);
5334 for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
5335 if (cgroup->nr_descendants >= cgroup->max_descendants)
5338 if (level > cgroup->max_depth)
5349 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
5351 struct cgroup *parent, *cgrp;
5354 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5355 if (strchr(name, '\n'))
5358 parent = cgroup_kn_lock_live(parent_kn, false);
5362 if (!cgroup_check_hierarchy_limits(parent)) {
5367 cgrp = cgroup_create(parent, name, mode);
5369 ret = PTR_ERR(cgrp);
5374 * This extra ref will be put in cgroup_free_fn() and guarantees
5375 * that @cgrp->kn is always accessible.
5377 kernfs_get(cgrp->kn);
5379 ret = cgroup_kn_set_ugid(cgrp->kn);
5383 ret = css_populate_dir(&cgrp->self);
5387 ret = cgroup_apply_control_enable(cgrp);
5391 TRACE_CGROUP_PATH(mkdir, cgrp);
5393 /* let's create and online css's */
5394 kernfs_activate(cgrp->kn);
5400 cgroup_destroy_locked(cgrp);
5402 cgroup_kn_unlock(parent_kn);
5407 * This is called when the refcnt of a css is confirmed to be killed.
5408 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5409 * initate destruction and put the css ref from kill_css().
5411 static void css_killed_work_fn(struct work_struct *work)
5413 struct cgroup_subsys_state *css =
5414 container_of(work, struct cgroup_subsys_state, destroy_work);
5416 mutex_lock(&cgroup_mutex);
5421 /* @css can't go away while we're holding cgroup_mutex */
5423 } while (css && atomic_dec_and_test(&css->online_cnt));
5425 mutex_unlock(&cgroup_mutex);
5428 /* css kill confirmation processing requires process context, bounce */
5429 static void css_killed_ref_fn(struct percpu_ref *ref)
5431 struct cgroup_subsys_state *css =
5432 container_of(ref, struct cgroup_subsys_state, refcnt);
5434 if (atomic_dec_and_test(&css->online_cnt)) {
5435 INIT_WORK(&css->destroy_work, css_killed_work_fn);
5436 queue_work(cgroup_destroy_wq, &css->destroy_work);
5441 * kill_css - destroy a css
5442 * @css: css to destroy
5444 * This function initiates destruction of @css by removing cgroup interface
5445 * files and putting its base reference. ->css_offline() will be invoked
5446 * asynchronously once css_tryget_online() is guaranteed to fail and when
5447 * the reference count reaches zero, @css will be released.
5449 static void kill_css(struct cgroup_subsys_state *css)
5451 lockdep_assert_held(&cgroup_mutex);
5453 if (css->flags & CSS_DYING)
5456 css->flags |= CSS_DYING;
5459 * This must happen before css is disassociated with its cgroup.
5460 * See seq_css() for details.
5465 * Killing would put the base ref, but we need to keep it alive
5466 * until after ->css_offline().
5471 * cgroup core guarantees that, by the time ->css_offline() is
5472 * invoked, no new css reference will be given out via
5473 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5474 * proceed to offlining css's because percpu_ref_kill() doesn't
5475 * guarantee that the ref is seen as killed on all CPUs on return.
5477 * Use percpu_ref_kill_and_confirm() to get notifications as each
5478 * css is confirmed to be seen as killed on all CPUs.
5480 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5484 * cgroup_destroy_locked - the first stage of cgroup destruction
5485 * @cgrp: cgroup to be destroyed
5487 * css's make use of percpu refcnts whose killing latency shouldn't be
5488 * exposed to userland and are RCU protected. Also, cgroup core needs to
5489 * guarantee that css_tryget_online() won't succeed by the time
5490 * ->css_offline() is invoked. To satisfy all the requirements,
5491 * destruction is implemented in the following two steps.
5493 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5494 * userland visible parts and start killing the percpu refcnts of
5495 * css's. Set up so that the next stage will be kicked off once all
5496 * the percpu refcnts are confirmed to be killed.
5498 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5499 * rest of destruction. Once all cgroup references are gone, the
5500 * cgroup is RCU-freed.
5502 * This function implements s1. After this step, @cgrp is gone as far as
5503 * the userland is concerned and a new cgroup with the same name may be
5504 * created. As cgroup doesn't care about the names internally, this
5505 * doesn't cause any problem.
5507 static int cgroup_destroy_locked(struct cgroup *cgrp)
5508 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5510 struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
5511 struct cgroup_subsys_state *css;
5512 struct cgrp_cset_link *link;
5515 lockdep_assert_held(&cgroup_mutex);
5518 * Only migration can raise populated from zero and we're already
5519 * holding cgroup_mutex.
5521 if (cgroup_is_populated(cgrp))
5525 * Make sure there's no live children. We can't test emptiness of
5526 * ->self.children as dead children linger on it while being
5527 * drained; otherwise, "rmdir parent/child parent" may fail.
5529 if (css_has_online_children(&cgrp->self))
5533 * Mark @cgrp and the associated csets dead. The former prevents
5534 * further task migration and child creation by disabling
5535 * cgroup_lock_live_group(). The latter makes the csets ignored by
5536 * the migration path.
5538 cgrp->self.flags &= ~CSS_ONLINE;
5540 spin_lock_irq(&css_set_lock);
5541 list_for_each_entry(link, &cgrp->cset_links, cset_link)
5542 link->cset->dead = true;
5543 spin_unlock_irq(&css_set_lock);
5545 /* initiate massacre of all css's */
5546 for_each_css(css, ssid, cgrp)
5549 /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5550 css_clear_dir(&cgrp->self);
5551 kernfs_remove(cgrp->kn);
5553 if (parent && cgroup_is_threaded(cgrp))
5554 parent->nr_threaded_children--;
5556 spin_lock_irq(&css_set_lock);
5557 for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5558 tcgrp->nr_descendants--;
5559 tcgrp->nr_dying_descendants++;
5561 * If the dying cgroup is frozen, decrease frozen descendants
5562 * counters of ancestor cgroups.
5564 if (test_bit(CGRP_FROZEN, &cgrp->flags))
5565 tcgrp->freezer.nr_frozen_descendants--;
5567 spin_unlock_irq(&css_set_lock);
5569 cgroup1_check_for_release(parent);
5571 cgroup_bpf_offline(cgrp);
5573 /* put the base reference */
5574 percpu_ref_kill(&cgrp->self.refcnt);
5579 int cgroup_rmdir(struct kernfs_node *kn)
5581 struct cgroup *cgrp;
5584 cgrp = cgroup_kn_lock_live(kn, false);
5588 ret = cgroup_destroy_locked(cgrp);
5590 TRACE_CGROUP_PATH(rmdir, cgrp);
5592 cgroup_kn_unlock(kn);
5596 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5597 .show_options = cgroup_show_options,
5598 .mkdir = cgroup_mkdir,
5599 .rmdir = cgroup_rmdir,
5600 .show_path = cgroup_show_path,
5603 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5605 struct cgroup_subsys_state *css;
5607 pr_debug("Initializing cgroup subsys %s\n", ss->name);
5609 mutex_lock(&cgroup_mutex);
5611 idr_init(&ss->css_idr);
5612 INIT_LIST_HEAD(&ss->cfts);
5614 /* Create the root cgroup state for this subsystem */
5615 ss->root = &cgrp_dfl_root;
5616 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5617 /* We don't handle early failures gracefully */
5618 BUG_ON(IS_ERR(css));
5619 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5622 * Root csses are never destroyed and we can't initialize
5623 * percpu_ref during early init. Disable refcnting.
5625 css->flags |= CSS_NO_REF;
5628 /* allocation can't be done safely during early init */
5631 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5632 BUG_ON(css->id < 0);
5635 /* Update the init_css_set to contain a subsys
5636 * pointer to this state - since the subsystem is
5637 * newly registered, all tasks and hence the
5638 * init_css_set is in the subsystem's root cgroup. */
5639 init_css_set.subsys[ss->id] = css;
5641 have_fork_callback |= (bool)ss->fork << ss->id;
5642 have_exit_callback |= (bool)ss->exit << ss->id;
5643 have_release_callback |= (bool)ss->release << ss->id;
5644 have_canfork_callback |= (bool)ss->can_fork << ss->id;
5646 /* At system boot, before all subsystems have been
5647 * registered, no tasks have been forked, so we don't
5648 * need to invoke fork callbacks here. */
5649 BUG_ON(!list_empty(&init_task.tasks));
5651 BUG_ON(online_css(css));
5653 mutex_unlock(&cgroup_mutex);
5657 * cgroup_init_early - cgroup initialization at system boot
5659 * Initialize cgroups at system boot, and initialize any
5660 * subsystems that request early init.
5662 int __init cgroup_init_early(void)
5664 static struct cgroup_fs_context __initdata ctx;
5665 struct cgroup_subsys *ss;
5668 ctx.root = &cgrp_dfl_root;
5669 init_cgroup_root(&ctx);
5670 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5672 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5674 for_each_subsys(ss, i) {
5675 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5676 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5677 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5679 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5680 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5683 ss->name = cgroup_subsys_name[i];
5684 if (!ss->legacy_name)
5685 ss->legacy_name = cgroup_subsys_name[i];
5688 cgroup_init_subsys(ss, true);
5694 * cgroup_init - cgroup initialization
5696 * Register cgroup filesystem and /proc file, and initialize
5697 * any subsystems that didn't request early init.
5699 int __init cgroup_init(void)
5701 struct cgroup_subsys *ss;
5704 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5705 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
5706 BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
5708 cgroup_rstat_boot();
5711 * The latency of the synchronize_rcu() is too high for cgroups,
5712 * avoid it at the cost of forcing all readers into the slow path.
5714 rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5716 get_user_ns(init_cgroup_ns.user_ns);
5718 mutex_lock(&cgroup_mutex);
5721 * Add init_css_set to the hash table so that dfl_root can link to
5724 hash_add(css_set_table, &init_css_set.hlist,
5725 css_set_hash(init_css_set.subsys));
5727 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
5729 mutex_unlock(&cgroup_mutex);
5731 for_each_subsys(ss, ssid) {
5732 if (ss->early_init) {
5733 struct cgroup_subsys_state *css =
5734 init_css_set.subsys[ss->id];
5736 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5738 BUG_ON(css->id < 0);
5740 cgroup_init_subsys(ss, false);
5743 list_add_tail(&init_css_set.e_cset_node[ssid],
5744 &cgrp_dfl_root.cgrp.e_csets[ssid]);
5747 * Setting dfl_root subsys_mask needs to consider the
5748 * disabled flag and cftype registration needs kmalloc,
5749 * both of which aren't available during early_init.
5751 if (!cgroup_ssid_enabled(ssid))
5754 if (cgroup1_ssid_disabled(ssid))
5755 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5758 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5760 /* implicit controllers must be threaded too */
5761 WARN_ON(ss->implicit_on_dfl && !ss->threaded);
5763 if (ss->implicit_on_dfl)
5764 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5765 else if (!ss->dfl_cftypes)
5766 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5769 cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
5771 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5772 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5774 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5775 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5779 ss->bind(init_css_set.subsys[ssid]);
5781 mutex_lock(&cgroup_mutex);
5782 css_populate_dir(init_css_set.subsys[ssid]);
5783 mutex_unlock(&cgroup_mutex);
5786 /* init_css_set.subsys[] has been updated, re-hash */
5787 hash_del(&init_css_set.hlist);
5788 hash_add(css_set_table, &init_css_set.hlist,
5789 css_set_hash(init_css_set.subsys));
5791 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5792 WARN_ON(register_filesystem(&cgroup_fs_type));
5793 WARN_ON(register_filesystem(&cgroup2_fs_type));
5794 WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show));
5795 #ifdef CONFIG_CPUSETS
5796 WARN_ON(register_filesystem(&cpuset_fs_type));
5802 static int __init cgroup_wq_init(void)
5805 * There isn't much point in executing destruction path in
5806 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5807 * Use 1 for @max_active.
5809 * We would prefer to do this in cgroup_init() above, but that
5810 * is called before init_workqueues(): so leave this until after.
5812 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5813 BUG_ON(!cgroup_destroy_wq);
5816 core_initcall(cgroup_wq_init);
5818 void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen)
5820 struct kernfs_node *kn;
5822 kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id);
5825 kernfs_path(kn, buf, buflen);
5830 * proc_cgroup_show()
5831 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5832 * - Used for /proc/<pid>/cgroup.
5834 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5835 struct pid *pid, struct task_struct *tsk)
5839 struct cgroup_root *root;
5842 buf = kmalloc(PATH_MAX, GFP_KERNEL);
5846 mutex_lock(&cgroup_mutex);
5847 spin_lock_irq(&css_set_lock);
5849 for_each_root(root) {
5850 struct cgroup_subsys *ss;
5851 struct cgroup *cgrp;
5852 int ssid, count = 0;
5854 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5857 seq_printf(m, "%d:", root->hierarchy_id);
5858 if (root != &cgrp_dfl_root)
5859 for_each_subsys(ss, ssid)
5860 if (root->subsys_mask & (1 << ssid))
5861 seq_printf(m, "%s%s", count++ ? "," : "",
5863 if (strlen(root->name))
5864 seq_printf(m, "%sname=%s", count ? "," : "",
5868 cgrp = task_cgroup_from_root(tsk, root);
5871 * On traditional hierarchies, all zombie tasks show up as
5872 * belonging to the root cgroup. On the default hierarchy,
5873 * while a zombie doesn't show up in "cgroup.procs" and
5874 * thus can't be migrated, its /proc/PID/cgroup keeps
5875 * reporting the cgroup it belonged to before exiting. If
5876 * the cgroup is removed before the zombie is reaped,
5877 * " (deleted)" is appended to the cgroup path.
5879 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
5880 retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5881 current->nsproxy->cgroup_ns);
5882 if (retval >= PATH_MAX)
5883 retval = -ENAMETOOLONG;
5892 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5893 seq_puts(m, " (deleted)\n");
5900 spin_unlock_irq(&css_set_lock);
5901 mutex_unlock(&cgroup_mutex);
5908 * cgroup_fork - initialize cgroup related fields during copy_process()
5909 * @child: pointer to task_struct of forking parent process.
5911 * A task is associated with the init_css_set until cgroup_post_fork()
5912 * attaches it to the target css_set.
5914 void cgroup_fork(struct task_struct *child)
5916 RCU_INIT_POINTER(child->cgroups, &init_css_set);
5917 INIT_LIST_HEAD(&child->cg_list);
5920 static struct cgroup *cgroup_get_from_file(struct file *f)
5922 struct cgroup_subsys_state *css;
5923 struct cgroup *cgrp;
5925 css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
5927 return ERR_CAST(css);
5930 if (!cgroup_on_dfl(cgrp)) {
5932 return ERR_PTR(-EBADF);
5939 * cgroup_css_set_fork - find or create a css_set for a child process
5940 * @kargs: the arguments passed to create the child process
5942 * This functions finds or creates a new css_set which the child
5943 * process will be attached to in cgroup_post_fork(). By default,
5944 * the child process will be given the same css_set as its parent.
5946 * If CLONE_INTO_CGROUP is specified this function will try to find an
5947 * existing css_set which includes the requested cgroup and if not create
5948 * a new css_set that the child will be attached to later. If this function
5949 * succeeds it will hold cgroup_threadgroup_rwsem on return. If
5950 * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
5951 * before grabbing cgroup_threadgroup_rwsem and will hold a reference
5952 * to the target cgroup.
5954 static int cgroup_css_set_fork(struct kernel_clone_args *kargs)
5955 __acquires(&cgroup_mutex) __acquires(&cgroup_threadgroup_rwsem)
5958 struct cgroup *dst_cgrp = NULL;
5959 struct css_set *cset;
5960 struct super_block *sb;
5963 if (kargs->flags & CLONE_INTO_CGROUP)
5964 mutex_lock(&cgroup_mutex);
5966 cgroup_threadgroup_change_begin(current);
5968 spin_lock_irq(&css_set_lock);
5969 cset = task_css_set(current);
5971 spin_unlock_irq(&css_set_lock);
5973 if (!(kargs->flags & CLONE_INTO_CGROUP)) {
5978 f = fget_raw(kargs->cgroup);
5983 sb = f->f_path.dentry->d_sb;
5985 dst_cgrp = cgroup_get_from_file(f);
5986 if (IS_ERR(dst_cgrp)) {
5987 ret = PTR_ERR(dst_cgrp);
5992 if (cgroup_is_dead(dst_cgrp)) {
5998 * Verify that we the target cgroup is writable for us. This is
5999 * usually done by the vfs layer but since we're not going through
6000 * the vfs layer here we need to do it "manually".
6002 ret = cgroup_may_write(dst_cgrp, sb);
6006 ret = cgroup_attach_permissions(cset->dfl_cgrp, dst_cgrp, sb,
6007 !(kargs->flags & CLONE_THREAD));
6011 kargs->cset = find_css_set(cset, dst_cgrp);
6019 kargs->cgrp = dst_cgrp;
6023 cgroup_threadgroup_change_end(current);
6024 mutex_unlock(&cgroup_mutex);
6028 cgroup_put(dst_cgrp);
6031 put_css_set(kargs->cset);
6036 * cgroup_css_set_put_fork - drop references we took during fork
6037 * @kargs: the arguments passed to create the child process
6039 * Drop references to the prepared css_set and target cgroup if
6040 * CLONE_INTO_CGROUP was requested.
6042 static void cgroup_css_set_put_fork(struct kernel_clone_args *kargs)
6043 __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6045 cgroup_threadgroup_change_end(current);
6047 if (kargs->flags & CLONE_INTO_CGROUP) {
6048 struct cgroup *cgrp = kargs->cgrp;
6049 struct css_set *cset = kargs->cset;
6051 mutex_unlock(&cgroup_mutex);
6066 * cgroup_can_fork - called on a new task before the process is exposed
6067 * @child: the child process
6069 * This prepares a new css_set for the child process which the child will
6070 * be attached to in cgroup_post_fork().
6071 * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
6072 * callback returns an error, the fork aborts with that error code. This
6073 * allows for a cgroup subsystem to conditionally allow or deny new forks.
6075 int cgroup_can_fork(struct task_struct *child, struct kernel_clone_args *kargs)
6077 struct cgroup_subsys *ss;
6080 ret = cgroup_css_set_fork(kargs);
6084 do_each_subsys_mask(ss, i, have_canfork_callback) {
6085 ret = ss->can_fork(child, kargs->cset);
6088 } while_each_subsys_mask();
6093 for_each_subsys(ss, j) {
6096 if (ss->cancel_fork)
6097 ss->cancel_fork(child, kargs->cset);
6100 cgroup_css_set_put_fork(kargs);
6106 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6107 * @child: the child process
6108 * @kargs: the arguments passed to create the child process
6110 * This calls the cancel_fork() callbacks if a fork failed *after*
6111 * cgroup_can_fork() succeded and cleans up references we took to
6112 * prepare a new css_set for the child process in cgroup_can_fork().
6114 void cgroup_cancel_fork(struct task_struct *child,
6115 struct kernel_clone_args *kargs)
6117 struct cgroup_subsys *ss;
6120 for_each_subsys(ss, i)
6121 if (ss->cancel_fork)
6122 ss->cancel_fork(child, kargs->cset);
6124 cgroup_css_set_put_fork(kargs);
6128 * cgroup_post_fork - finalize cgroup setup for the child process
6129 * @child: the child process
6131 * Attach the child process to its css_set calling the subsystem fork()
6134 void cgroup_post_fork(struct task_struct *child,
6135 struct kernel_clone_args *kargs)
6136 __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6138 struct cgroup_subsys *ss;
6139 struct css_set *cset;
6145 spin_lock_irq(&css_set_lock);
6147 /* init tasks are special, only link regular threads */
6148 if (likely(child->pid)) {
6149 WARN_ON_ONCE(!list_empty(&child->cg_list));
6151 css_set_move_task(child, NULL, cset, false);
6158 * If the cgroup has to be frozen, the new task has too. Let's set
6159 * the JOBCTL_TRAP_FREEZE jobctl bit to get the task into the
6162 if (unlikely(cgroup_task_freeze(child))) {
6163 spin_lock(&child->sighand->siglock);
6164 WARN_ON_ONCE(child->frozen);
6165 child->jobctl |= JOBCTL_TRAP_FREEZE;
6166 spin_unlock(&child->sighand->siglock);
6169 * Calling cgroup_update_frozen() isn't required here,
6170 * because it will be called anyway a bit later from
6171 * do_freezer_trap(). So we avoid cgroup's transient switch
6172 * from the frozen state and back.
6176 spin_unlock_irq(&css_set_lock);
6179 * Call ss->fork(). This must happen after @child is linked on
6180 * css_set; otherwise, @child might change state between ->fork()
6181 * and addition to css_set.
6183 do_each_subsys_mask(ss, i, have_fork_callback) {
6185 } while_each_subsys_mask();
6187 /* Make the new cset the root_cset of the new cgroup namespace. */
6188 if (kargs->flags & CLONE_NEWCGROUP) {
6189 struct css_set *rcset = child->nsproxy->cgroup_ns->root_cset;
6192 child->nsproxy->cgroup_ns->root_cset = cset;
6196 cgroup_css_set_put_fork(kargs);
6200 * cgroup_exit - detach cgroup from exiting task
6201 * @tsk: pointer to task_struct of exiting process
6203 * Description: Detach cgroup from @tsk.
6206 void cgroup_exit(struct task_struct *tsk)
6208 struct cgroup_subsys *ss;
6209 struct css_set *cset;
6212 spin_lock_irq(&css_set_lock);
6214 WARN_ON_ONCE(list_empty(&tsk->cg_list));
6215 cset = task_css_set(tsk);
6216 css_set_move_task(tsk, cset, NULL, false);
6217 list_add_tail(&tsk->cg_list, &cset->dying_tasks);
6220 WARN_ON_ONCE(cgroup_task_frozen(tsk));
6221 if (unlikely(cgroup_task_freeze(tsk)))
6222 cgroup_update_frozen(task_dfl_cgroup(tsk));
6224 spin_unlock_irq(&css_set_lock);
6226 /* see cgroup_post_fork() for details */
6227 do_each_subsys_mask(ss, i, have_exit_callback) {
6229 } while_each_subsys_mask();
6232 void cgroup_release(struct task_struct *task)
6234 struct cgroup_subsys *ss;
6237 do_each_subsys_mask(ss, ssid, have_release_callback) {
6239 } while_each_subsys_mask();
6241 spin_lock_irq(&css_set_lock);
6242 css_set_skip_task_iters(task_css_set(task), task);
6243 list_del_init(&task->cg_list);
6244 spin_unlock_irq(&css_set_lock);
6247 void cgroup_free(struct task_struct *task)
6249 struct css_set *cset = task_css_set(task);
6253 static int __init cgroup_disable(char *str)
6255 struct cgroup_subsys *ss;
6259 while ((token = strsep(&str, ",")) != NULL) {
6263 for_each_subsys(ss, i) {
6264 if (strcmp(token, ss->name) &&
6265 strcmp(token, ss->legacy_name))
6268 static_branch_disable(cgroup_subsys_enabled_key[i]);
6269 pr_info("Disabling %s control group subsystem\n",
6275 __setup("cgroup_disable=", cgroup_disable);
6277 void __init __weak enable_debug_cgroup(void) { }
6279 static int __init enable_cgroup_debug(char *str)
6281 cgroup_debug = true;
6282 enable_debug_cgroup();
6285 __setup("cgroup_debug", enable_cgroup_debug);
6288 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6289 * @dentry: directory dentry of interest
6290 * @ss: subsystem of interest
6292 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6293 * to get the corresponding css and return it. If such css doesn't exist
6294 * or can't be pinned, an ERR_PTR value is returned.
6296 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
6297 struct cgroup_subsys *ss)
6299 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
6300 struct file_system_type *s_type = dentry->d_sb->s_type;
6301 struct cgroup_subsys_state *css = NULL;
6302 struct cgroup *cgrp;
6304 /* is @dentry a cgroup dir? */
6305 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
6306 !kn || kernfs_type(kn) != KERNFS_DIR)
6307 return ERR_PTR(-EBADF);
6312 * This path doesn't originate from kernfs and @kn could already
6313 * have been or be removed at any point. @kn->priv is RCU
6314 * protected for this access. See css_release_work_fn() for details.
6316 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6318 css = cgroup_css(cgrp, ss);
6320 if (!css || !css_tryget_online(css))
6321 css = ERR_PTR(-ENOENT);
6328 * css_from_id - lookup css by id
6329 * @id: the cgroup id
6330 * @ss: cgroup subsys to be looked into
6332 * Returns the css if there's valid one with @id, otherwise returns NULL.
6333 * Should be called under rcu_read_lock().
6335 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
6337 WARN_ON_ONCE(!rcu_read_lock_held());
6338 return idr_find(&ss->css_idr, id);
6342 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6343 * @path: path on the default hierarchy
6345 * Find the cgroup at @path on the default hierarchy, increment its
6346 * reference count and return it. Returns pointer to the found cgroup on
6347 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
6348 * if @path points to a non-directory.
6350 struct cgroup *cgroup_get_from_path(const char *path)
6352 struct kernfs_node *kn;
6353 struct cgroup *cgrp;
6355 mutex_lock(&cgroup_mutex);
6357 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6359 if (kernfs_type(kn) == KERNFS_DIR) {
6361 cgroup_get_live(cgrp);
6363 cgrp = ERR_PTR(-ENOTDIR);
6367 cgrp = ERR_PTR(-ENOENT);
6370 mutex_unlock(&cgroup_mutex);
6373 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6376 * cgroup_get_from_fd - get a cgroup pointer from a fd
6377 * @fd: fd obtained by open(cgroup2_dir)
6379 * Find the cgroup from a fd which should be obtained
6380 * by opening a cgroup directory. Returns a pointer to the
6381 * cgroup on success. ERR_PTR is returned if the cgroup
6384 struct cgroup *cgroup_get_from_fd(int fd)
6386 struct cgroup *cgrp;
6391 return ERR_PTR(-EBADF);
6393 cgrp = cgroup_get_from_file(f);
6397 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
6399 static u64 power_of_ten(int power)
6408 * cgroup_parse_float - parse a floating number
6409 * @input: input string
6410 * @dec_shift: number of decimal digits to shift
6413 * Parse a decimal floating point number in @input and store the result in
6414 * @v with decimal point right shifted @dec_shift times. For example, if
6415 * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6416 * Returns 0 on success, -errno otherwise.
6418 * There's nothing cgroup specific about this function except that it's
6419 * currently the only user.
6421 int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v)
6423 s64 whole, frac = 0;
6424 int fstart = 0, fend = 0, flen;
6426 if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend))
6431 flen = fend > fstart ? fend - fstart : 0;
6432 if (flen < dec_shift)
6433 frac *= power_of_ten(dec_shift - flen);
6435 frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift));
6437 *v = whole * power_of_ten(dec_shift) + frac;
6442 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6443 * definition in cgroup-defs.h.
6445 #ifdef CONFIG_SOCK_CGROUP_DATA
6447 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6449 DEFINE_SPINLOCK(cgroup_sk_update_lock);
6450 static bool cgroup_sk_alloc_disabled __read_mostly;
6452 void cgroup_sk_alloc_disable(void)
6454 if (cgroup_sk_alloc_disabled)
6456 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6457 cgroup_sk_alloc_disabled = true;
6462 #define cgroup_sk_alloc_disabled false
6466 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6468 if (cgroup_sk_alloc_disabled) {
6469 skcd->no_refcnt = 1;
6473 /* Don't associate the sock with unrelated interrupted task's cgroup. */
6480 struct css_set *cset;
6482 cset = task_css_set(current);
6483 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6484 skcd->val = (unsigned long)cset->dfl_cgrp;
6485 cgroup_bpf_get(cset->dfl_cgrp);
6494 void cgroup_sk_clone(struct sock_cgroup_data *skcd)
6497 if (skcd->no_refcnt)
6500 * We might be cloning a socket which is left in an empty
6501 * cgroup and the cgroup might have already been rmdir'd.
6502 * Don't use cgroup_get_live().
6504 cgroup_get(sock_cgroup_ptr(skcd));
6505 cgroup_bpf_get(sock_cgroup_ptr(skcd));
6509 void cgroup_sk_free(struct sock_cgroup_data *skcd)
6511 struct cgroup *cgrp = sock_cgroup_ptr(skcd);
6513 if (skcd->no_refcnt)
6515 cgroup_bpf_put(cgrp);
6519 #endif /* CONFIG_SOCK_CGROUP_DATA */
6521 #ifdef CONFIG_CGROUP_BPF
6522 int cgroup_bpf_attach(struct cgroup *cgrp,
6523 struct bpf_prog *prog, struct bpf_prog *replace_prog,
6524 struct bpf_cgroup_link *link,
6525 enum bpf_attach_type type,
6530 mutex_lock(&cgroup_mutex);
6531 ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags);
6532 mutex_unlock(&cgroup_mutex);
6536 int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
6537 enum bpf_attach_type type)
6541 mutex_lock(&cgroup_mutex);
6542 ret = __cgroup_bpf_detach(cgrp, prog, NULL, type);
6543 mutex_unlock(&cgroup_mutex);
6547 int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
6548 union bpf_attr __user *uattr)
6552 mutex_lock(&cgroup_mutex);
6553 ret = __cgroup_bpf_query(cgrp, attr, uattr);
6554 mutex_unlock(&cgroup_mutex);
6557 #endif /* CONFIG_CGROUP_BPF */
6560 static ssize_t show_delegatable_files(struct cftype *files, char *buf,
6561 ssize_t size, const char *prefix)
6566 for (cft = files; cft && cft->name[0] != '\0'; cft++) {
6567 if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
6571 ret += snprintf(buf + ret, size - ret, "%s.", prefix);
6573 ret += snprintf(buf + ret, size - ret, "%s\n", cft->name);
6575 if (WARN_ON(ret >= size))
6582 static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr,
6585 struct cgroup_subsys *ss;
6589 ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret,
6592 for_each_subsys(ss, ssid)
6593 ret += show_delegatable_files(ss->dfl_cftypes, buf + ret,
6595 cgroup_subsys_name[ssid]);
6599 static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate);
6601 static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr,
6604 return snprintf(buf, PAGE_SIZE,
6606 "memory_localevents\n"
6607 "memory_recursiveprot\n");
6609 static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features);
6611 static struct attribute *cgroup_sysfs_attrs[] = {
6612 &cgroup_delegate_attr.attr,
6613 &cgroup_features_attr.attr,
6617 static const struct attribute_group cgroup_sysfs_attr_group = {
6618 .attrs = cgroup_sysfs_attrs,
6622 static int __init cgroup_sysfs_init(void)
6624 return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group);
6626 subsys_initcall(cgroup_sysfs_init);
6628 #endif /* CONFIG_SYSFS */