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>
59 #define CREATE_TRACE_POINTS
60 #include <trace/events/cgroup.h>
62 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
66 * cgroup_mutex is the master lock. Any modification to cgroup or its
67 * hierarchy must be performed while holding it.
69 * css_set_lock protects task->cgroups pointer, the list of css_set
70 * objects, and the chain of tasks off each css_set.
72 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
73 * cgroup.h can use them for lockdep annotations.
75 DEFINE_MUTEX(cgroup_mutex);
76 DEFINE_SPINLOCK(css_set_lock);
78 #ifdef CONFIG_PROVE_RCU
79 EXPORT_SYMBOL_GPL(cgroup_mutex);
80 EXPORT_SYMBOL_GPL(css_set_lock);
84 * Protects cgroup_idr and css_idr so that IDs can be released without
85 * grabbing cgroup_mutex.
87 static DEFINE_SPINLOCK(cgroup_idr_lock);
90 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
91 * against file removal/re-creation across css hiding.
93 static DEFINE_SPINLOCK(cgroup_file_kn_lock);
95 struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
97 #define cgroup_assert_mutex_or_rcu_locked() \
98 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
99 !lockdep_is_held(&cgroup_mutex), \
100 "cgroup_mutex or RCU read lock required");
103 * cgroup destruction makes heavy use of work items and there can be a lot
104 * of concurrent destructions. Use a separate workqueue so that cgroup
105 * destruction work items don't end up filling up max_active of system_wq
106 * which may lead to deadlock.
108 static struct workqueue_struct *cgroup_destroy_wq;
110 /* generate an array of cgroup subsystem pointers */
111 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
112 struct cgroup_subsys *cgroup_subsys[] = {
113 #include <linux/cgroup_subsys.h>
117 /* array of cgroup subsystem names */
118 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
119 static const char *cgroup_subsys_name[] = {
120 #include <linux/cgroup_subsys.h>
124 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
126 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
127 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
128 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
129 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
130 #include <linux/cgroup_subsys.h>
133 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
134 static struct static_key_true *cgroup_subsys_enabled_key[] = {
135 #include <linux/cgroup_subsys.h>
139 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
140 static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
141 #include <linux/cgroup_subsys.h>
146 * The default hierarchy, reserved for the subsystems that are otherwise
147 * unattached - it never has more than a single cgroup, and all tasks are
148 * part of that cgroup.
150 struct cgroup_root cgrp_dfl_root;
151 EXPORT_SYMBOL_GPL(cgrp_dfl_root);
154 * The default hierarchy always exists but is hidden until mounted for the
155 * first time. This is for backward compatibility.
157 static bool cgrp_dfl_visible;
159 /* some controllers are not supported in the default hierarchy */
160 static u16 cgrp_dfl_inhibit_ss_mask;
162 /* some controllers are implicitly enabled on the default hierarchy */
163 static u16 cgrp_dfl_implicit_ss_mask;
165 /* some controllers can be threaded on the default hierarchy */
166 static u16 cgrp_dfl_threaded_ss_mask;
168 /* The list of hierarchy roots */
169 LIST_HEAD(cgroup_roots);
170 static int cgroup_root_count;
172 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
173 static DEFINE_IDR(cgroup_hierarchy_idr);
176 * Assign a monotonically increasing serial number to csses. It guarantees
177 * cgroups with bigger numbers are newer than those with smaller numbers.
178 * Also, as csses are always appended to the parent's ->children list, it
179 * guarantees that sibling csses are always sorted in the ascending serial
180 * number order on the list. Protected by cgroup_mutex.
182 static u64 css_serial_nr_next = 1;
185 * These bitmasks identify subsystems with specific features to avoid
186 * having to do iterative checks repeatedly.
188 static u16 have_fork_callback __read_mostly;
189 static u16 have_exit_callback __read_mostly;
190 static u16 have_free_callback __read_mostly;
191 static u16 have_canfork_callback __read_mostly;
193 /* cgroup namespace for init task */
194 struct cgroup_namespace init_cgroup_ns = {
195 .count = REFCOUNT_INIT(2),
196 .user_ns = &init_user_ns,
197 .ns.ops = &cgroupns_operations,
198 .ns.inum = PROC_CGROUP_INIT_INO,
199 .root_cset = &init_css_set,
202 static struct file_system_type cgroup2_fs_type;
203 static struct cftype cgroup_base_files[];
205 static int cgroup_apply_control(struct cgroup *cgrp);
206 static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
207 static void css_task_iter_advance(struct css_task_iter *it);
208 static int cgroup_destroy_locked(struct cgroup *cgrp);
209 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
210 struct cgroup_subsys *ss);
211 static void css_release(struct percpu_ref *ref);
212 static void kill_css(struct cgroup_subsys_state *css);
213 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
214 struct cgroup *cgrp, struct cftype cfts[],
218 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
219 * @ssid: subsys ID of interest
221 * cgroup_subsys_enabled() can only be used with literal subsys names which
222 * is fine for individual subsystems but unsuitable for cgroup core. This
223 * is slower static_key_enabled() based test indexed by @ssid.
225 bool cgroup_ssid_enabled(int ssid)
227 if (CGROUP_SUBSYS_COUNT == 0)
230 return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
234 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
235 * @cgrp: the cgroup of interest
237 * The default hierarchy is the v2 interface of cgroup and this function
238 * can be used to test whether a cgroup is on the default hierarchy for
239 * cases where a subsystem should behave differnetly depending on the
242 * The set of behaviors which change on the default hierarchy are still
243 * being determined and the mount option is prefixed with __DEVEL__.
245 * List of changed behaviors:
247 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
248 * and "name" are disallowed.
250 * - When mounting an existing superblock, mount options should match.
252 * - Remount is disallowed.
254 * - rename(2) is disallowed.
256 * - "tasks" is removed. Everything should be at process granularity. Use
257 * "cgroup.procs" instead.
259 * - "cgroup.procs" is not sorted. pids will be unique unless they got
260 * recycled inbetween reads.
262 * - "release_agent" and "notify_on_release" are removed. Replacement
263 * notification mechanism will be implemented.
265 * - "cgroup.clone_children" is removed.
267 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
268 * and its descendants contain no task; otherwise, 1. The file also
269 * generates kernfs notification which can be monitored through poll and
270 * [di]notify when the value of the file changes.
272 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
273 * take masks of ancestors with non-empty cpus/mems, instead of being
274 * moved to an ancestor.
276 * - cpuset: a task can be moved into an empty cpuset, and again it takes
277 * masks of ancestors.
279 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
282 * - blkcg: blk-throttle becomes properly hierarchical.
284 * - debug: disallowed on the default hierarchy.
286 bool cgroup_on_dfl(const struct cgroup *cgrp)
288 return cgrp->root == &cgrp_dfl_root;
291 /* IDR wrappers which synchronize using cgroup_idr_lock */
292 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
297 idr_preload(gfp_mask);
298 spin_lock_bh(&cgroup_idr_lock);
299 ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
300 spin_unlock_bh(&cgroup_idr_lock);
305 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
309 spin_lock_bh(&cgroup_idr_lock);
310 ret = idr_replace(idr, ptr, id);
311 spin_unlock_bh(&cgroup_idr_lock);
315 static void cgroup_idr_remove(struct idr *idr, int id)
317 spin_lock_bh(&cgroup_idr_lock);
319 spin_unlock_bh(&cgroup_idr_lock);
322 static bool cgroup_has_tasks(struct cgroup *cgrp)
324 return cgrp->nr_populated_csets;
327 bool cgroup_is_threaded(struct cgroup *cgrp)
329 return cgrp->dom_cgrp != cgrp;
332 /* can @cgrp host both domain and threaded children? */
333 static bool cgroup_is_mixable(struct cgroup *cgrp)
336 * Root isn't under domain level resource control exempting it from
337 * the no-internal-process constraint, so it can serve as a thread
338 * root and a parent of resource domains at the same time.
340 return !cgroup_parent(cgrp);
343 /* can @cgrp become a thread root? should always be true for a thread root */
344 static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
346 /* mixables don't care */
347 if (cgroup_is_mixable(cgrp))
350 /* domain roots can't be nested under threaded */
351 if (cgroup_is_threaded(cgrp))
354 /* can only have either domain or threaded children */
355 if (cgrp->nr_populated_domain_children)
358 /* and no domain controllers can be enabled */
359 if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
365 /* is @cgrp root of a threaded subtree? */
366 bool cgroup_is_thread_root(struct cgroup *cgrp)
368 /* thread root should be a domain */
369 if (cgroup_is_threaded(cgrp))
372 /* a domain w/ threaded children is a thread root */
373 if (cgrp->nr_threaded_children)
377 * A domain which has tasks and explicit threaded controllers
378 * enabled is a thread root.
380 if (cgroup_has_tasks(cgrp) &&
381 (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
387 /* a domain which isn't connected to the root w/o brekage can't be used */
388 static bool cgroup_is_valid_domain(struct cgroup *cgrp)
390 /* the cgroup itself can be a thread root */
391 if (cgroup_is_threaded(cgrp))
394 /* but the ancestors can't be unless mixable */
395 while ((cgrp = cgroup_parent(cgrp))) {
396 if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
398 if (cgroup_is_threaded(cgrp))
405 /* subsystems visibly enabled on a cgroup */
406 static u16 cgroup_control(struct cgroup *cgrp)
408 struct cgroup *parent = cgroup_parent(cgrp);
409 u16 root_ss_mask = cgrp->root->subsys_mask;
412 u16 ss_mask = parent->subtree_control;
414 /* threaded cgroups can only have threaded controllers */
415 if (cgroup_is_threaded(cgrp))
416 ss_mask &= cgrp_dfl_threaded_ss_mask;
420 if (cgroup_on_dfl(cgrp))
421 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
422 cgrp_dfl_implicit_ss_mask);
426 /* subsystems enabled on a cgroup */
427 static u16 cgroup_ss_mask(struct cgroup *cgrp)
429 struct cgroup *parent = cgroup_parent(cgrp);
432 u16 ss_mask = parent->subtree_ss_mask;
434 /* threaded cgroups can only have threaded controllers */
435 if (cgroup_is_threaded(cgrp))
436 ss_mask &= cgrp_dfl_threaded_ss_mask;
440 return cgrp->root->subsys_mask;
444 * cgroup_css - obtain a cgroup's css for the specified subsystem
445 * @cgrp: the cgroup of interest
446 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
448 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
449 * function must be called either under cgroup_mutex or rcu_read_lock() and
450 * the caller is responsible for pinning the returned css if it wants to
451 * keep accessing it outside the said locks. This function may return
452 * %NULL if @cgrp doesn't have @subsys_id enabled.
454 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
455 struct cgroup_subsys *ss)
458 return rcu_dereference_check(cgrp->subsys[ss->id],
459 lockdep_is_held(&cgroup_mutex));
465 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
466 * @cgrp: the cgroup of interest
467 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
469 * Similar to cgroup_css() but returns the effective css, which is defined
470 * as the matching css of the nearest ancestor including self which has @ss
471 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
472 * function is guaranteed to return non-NULL css.
474 static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
475 struct cgroup_subsys *ss)
477 lockdep_assert_held(&cgroup_mutex);
483 * This function is used while updating css associations and thus
484 * can't test the csses directly. Test ss_mask.
486 while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
487 cgrp = cgroup_parent(cgrp);
492 return cgroup_css(cgrp, ss);
496 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
497 * @cgrp: the cgroup of interest
498 * @ss: the subsystem of interest
500 * Find and get the effective css of @cgrp for @ss. The effective css is
501 * defined as the matching css of the nearest ancestor including self which
502 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
503 * the root css is returned, so this function always returns a valid css.
504 * The returned css must be put using css_put().
506 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
507 struct cgroup_subsys *ss)
509 struct cgroup_subsys_state *css;
514 css = cgroup_css(cgrp, ss);
516 if (css && css_tryget_online(css))
518 cgrp = cgroup_parent(cgrp);
521 css = init_css_set.subsys[ss->id];
528 static void cgroup_get_live(struct cgroup *cgrp)
530 WARN_ON_ONCE(cgroup_is_dead(cgrp));
531 css_get(&cgrp->self);
534 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
536 struct cgroup *cgrp = of->kn->parent->priv;
537 struct cftype *cft = of_cft(of);
540 * This is open and unprotected implementation of cgroup_css().
541 * seq_css() is only called from a kernfs file operation which has
542 * an active reference on the file. Because all the subsystem
543 * files are drained before a css is disassociated with a cgroup,
544 * the matching css from the cgroup's subsys table is guaranteed to
545 * be and stay valid until the enclosing operation is complete.
548 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
552 EXPORT_SYMBOL_GPL(of_css);
555 * for_each_css - iterate all css's of a cgroup
556 * @css: the iteration cursor
557 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
558 * @cgrp: the target cgroup to iterate css's of
560 * Should be called under cgroup_[tree_]mutex.
562 #define for_each_css(css, ssid, cgrp) \
563 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
564 if (!((css) = rcu_dereference_check( \
565 (cgrp)->subsys[(ssid)], \
566 lockdep_is_held(&cgroup_mutex)))) { } \
570 * for_each_e_css - iterate all effective css's of a cgroup
571 * @css: the iteration cursor
572 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
573 * @cgrp: the target cgroup to iterate css's of
575 * Should be called under cgroup_[tree_]mutex.
577 #define for_each_e_css(css, ssid, cgrp) \
578 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
579 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
584 * do_each_subsys_mask - filter for_each_subsys with a bitmask
585 * @ss: the iteration cursor
586 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
587 * @ss_mask: the bitmask
589 * The block will only run for cases where the ssid-th bit (1 << ssid) of
592 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
593 unsigned long __ss_mask = (ss_mask); \
594 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
598 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
599 (ss) = cgroup_subsys[ssid]; \
602 #define while_each_subsys_mask() \
607 /* iterate over child cgrps, lock should be held throughout iteration */
608 #define cgroup_for_each_live_child(child, cgrp) \
609 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
610 if (({ lockdep_assert_held(&cgroup_mutex); \
611 cgroup_is_dead(child); })) \
615 /* walk live descendants in preorder */
616 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
617 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
618 if (({ lockdep_assert_held(&cgroup_mutex); \
619 (dsct) = (d_css)->cgroup; \
620 cgroup_is_dead(dsct); })) \
624 /* walk live descendants in postorder */
625 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
626 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
627 if (({ lockdep_assert_held(&cgroup_mutex); \
628 (dsct) = (d_css)->cgroup; \
629 cgroup_is_dead(dsct); })) \
634 * The default css_set - used by init and its children prior to any
635 * hierarchies being mounted. It contains a pointer to the root state
636 * for each subsystem. Also used to anchor the list of css_sets. Not
637 * reference-counted, to improve performance when child cgroups
638 * haven't been created.
640 struct css_set init_css_set = {
641 .refcount = REFCOUNT_INIT(1),
642 .dom_cset = &init_css_set,
643 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
644 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
645 .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
646 .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets),
647 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
648 .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
649 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
652 static int css_set_count = 1; /* 1 for init_css_set */
654 static bool css_set_threaded(struct css_set *cset)
656 return cset->dom_cset != cset;
660 * css_set_populated - does a css_set contain any tasks?
661 * @cset: target css_set
663 * css_set_populated() should be the same as !!cset->nr_tasks at steady
664 * state. However, css_set_populated() can be called while a task is being
665 * added to or removed from the linked list before the nr_tasks is
666 * properly updated. Hence, we can't just look at ->nr_tasks here.
668 static bool css_set_populated(struct css_set *cset)
670 lockdep_assert_held(&css_set_lock);
672 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
676 * cgroup_update_populated - update the populated count of a cgroup
677 * @cgrp: the target cgroup
678 * @populated: inc or dec populated count
680 * One of the css_sets associated with @cgrp is either getting its first
681 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
682 * count is propagated towards root so that a given cgroup's
683 * nr_populated_children is zero iff none of its descendants contain any
686 * @cgrp's interface file "cgroup.populated" is zero if both
687 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
688 * 1 otherwise. When the sum changes from or to zero, userland is notified
689 * that the content of the interface file has changed. This can be used to
690 * detect when @cgrp and its descendants become populated or empty.
692 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
694 struct cgroup *child = NULL;
695 int adj = populated ? 1 : -1;
697 lockdep_assert_held(&css_set_lock);
700 bool was_populated = cgroup_is_populated(cgrp);
703 cgrp->nr_populated_csets += adj;
705 if (cgroup_is_threaded(child))
706 cgrp->nr_populated_threaded_children += adj;
708 cgrp->nr_populated_domain_children += adj;
711 if (was_populated == cgroup_is_populated(cgrp))
714 cgroup1_check_for_release(cgrp);
715 cgroup_file_notify(&cgrp->events_file);
718 cgrp = cgroup_parent(cgrp);
723 * css_set_update_populated - update populated state of a css_set
724 * @cset: target css_set
725 * @populated: whether @cset is populated or depopulated
727 * @cset is either getting the first task or losing the last. Update the
728 * populated counters of all associated cgroups accordingly.
730 static void css_set_update_populated(struct css_set *cset, bool populated)
732 struct cgrp_cset_link *link;
734 lockdep_assert_held(&css_set_lock);
736 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
737 cgroup_update_populated(link->cgrp, populated);
741 * css_set_move_task - move a task from one css_set to another
742 * @task: task being moved
743 * @from_cset: css_set @task currently belongs to (may be NULL)
744 * @to_cset: new css_set @task is being moved to (may be NULL)
745 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
747 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
748 * css_set, @from_cset can be NULL. If @task is being disassociated
749 * instead of moved, @to_cset can be NULL.
751 * This function automatically handles populated counter updates and
752 * css_task_iter adjustments but the caller is responsible for managing
753 * @from_cset and @to_cset's reference counts.
755 static void css_set_move_task(struct task_struct *task,
756 struct css_set *from_cset, struct css_set *to_cset,
759 lockdep_assert_held(&css_set_lock);
761 if (to_cset && !css_set_populated(to_cset))
762 css_set_update_populated(to_cset, true);
765 struct css_task_iter *it, *pos;
767 WARN_ON_ONCE(list_empty(&task->cg_list));
770 * @task is leaving, advance task iterators which are
771 * pointing to it so that they can resume at the next
772 * position. Advancing an iterator might remove it from
773 * the list, use safe walk. See css_task_iter_advance*()
776 list_for_each_entry_safe(it, pos, &from_cset->task_iters,
778 if (it->task_pos == &task->cg_list)
779 css_task_iter_advance(it);
781 list_del_init(&task->cg_list);
782 if (!css_set_populated(from_cset))
783 css_set_update_populated(from_cset, false);
785 WARN_ON_ONCE(!list_empty(&task->cg_list));
790 * We are synchronized through cgroup_threadgroup_rwsem
791 * against PF_EXITING setting such that we can't race
792 * against cgroup_exit() changing the css_set to
793 * init_css_set and dropping the old one.
795 WARN_ON_ONCE(task->flags & PF_EXITING);
797 rcu_assign_pointer(task->cgroups, to_cset);
798 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
804 * hash table for cgroup groups. This improves the performance to find
805 * an existing css_set. This hash doesn't (currently) take into
806 * account cgroups in empty hierarchies.
808 #define CSS_SET_HASH_BITS 7
809 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
811 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
813 unsigned long key = 0UL;
814 struct cgroup_subsys *ss;
817 for_each_subsys(ss, i)
818 key += (unsigned long)css[i];
819 key = (key >> 16) ^ key;
824 void put_css_set_locked(struct css_set *cset)
826 struct cgrp_cset_link *link, *tmp_link;
827 struct cgroup_subsys *ss;
830 lockdep_assert_held(&css_set_lock);
832 if (!refcount_dec_and_test(&cset->refcount))
835 WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
837 /* This css_set is dead. unlink it and release cgroup and css refs */
838 for_each_subsys(ss, ssid) {
839 list_del(&cset->e_cset_node[ssid]);
840 css_put(cset->subsys[ssid]);
842 hash_del(&cset->hlist);
845 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
846 list_del(&link->cset_link);
847 list_del(&link->cgrp_link);
848 if (cgroup_parent(link->cgrp))
849 cgroup_put(link->cgrp);
853 if (css_set_threaded(cset)) {
854 list_del(&cset->threaded_csets_node);
855 put_css_set_locked(cset->dom_cset);
858 kfree_rcu(cset, rcu_head);
862 * compare_css_sets - helper function for find_existing_css_set().
863 * @cset: candidate css_set being tested
864 * @old_cset: existing css_set for a task
865 * @new_cgrp: cgroup that's being entered by the task
866 * @template: desired set of css pointers in css_set (pre-calculated)
868 * Returns true if "cset" matches "old_cset" except for the hierarchy
869 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
871 static bool compare_css_sets(struct css_set *cset,
872 struct css_set *old_cset,
873 struct cgroup *new_cgrp,
874 struct cgroup_subsys_state *template[])
876 struct cgroup *new_dfl_cgrp;
877 struct list_head *l1, *l2;
880 * On the default hierarchy, there can be csets which are
881 * associated with the same set of cgroups but different csses.
882 * Let's first ensure that csses match.
884 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
888 /* @cset's domain should match the default cgroup's */
889 if (cgroup_on_dfl(new_cgrp))
890 new_dfl_cgrp = new_cgrp;
892 new_dfl_cgrp = old_cset->dfl_cgrp;
894 if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
898 * Compare cgroup pointers in order to distinguish between
899 * different cgroups in hierarchies. As different cgroups may
900 * share the same effective css, this comparison is always
903 l1 = &cset->cgrp_links;
904 l2 = &old_cset->cgrp_links;
906 struct cgrp_cset_link *link1, *link2;
907 struct cgroup *cgrp1, *cgrp2;
911 /* See if we reached the end - both lists are equal length. */
912 if (l1 == &cset->cgrp_links) {
913 BUG_ON(l2 != &old_cset->cgrp_links);
916 BUG_ON(l2 == &old_cset->cgrp_links);
918 /* Locate the cgroups associated with these links. */
919 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
920 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
923 /* Hierarchies should be linked in the same order. */
924 BUG_ON(cgrp1->root != cgrp2->root);
927 * If this hierarchy is the hierarchy of the cgroup
928 * that's changing, then we need to check that this
929 * css_set points to the new cgroup; if it's any other
930 * hierarchy, then this css_set should point to the
931 * same cgroup as the old css_set.
933 if (cgrp1->root == new_cgrp->root) {
934 if (cgrp1 != new_cgrp)
945 * find_existing_css_set - init css array and find the matching css_set
946 * @old_cset: the css_set that we're using before the cgroup transition
947 * @cgrp: the cgroup that we're moving into
948 * @template: out param for the new set of csses, should be clear on entry
950 static struct css_set *find_existing_css_set(struct css_set *old_cset,
952 struct cgroup_subsys_state *template[])
954 struct cgroup_root *root = cgrp->root;
955 struct cgroup_subsys *ss;
956 struct css_set *cset;
961 * Build the set of subsystem state objects that we want to see in the
962 * new css_set. while subsystems can change globally, the entries here
963 * won't change, so no need for locking.
965 for_each_subsys(ss, i) {
966 if (root->subsys_mask & (1UL << i)) {
968 * @ss is in this hierarchy, so we want the
969 * effective css from @cgrp.
971 template[i] = cgroup_e_css(cgrp, ss);
974 * @ss is not in this hierarchy, so we don't want
977 template[i] = old_cset->subsys[i];
981 key = css_set_hash(template);
982 hash_for_each_possible(css_set_table, cset, hlist, key) {
983 if (!compare_css_sets(cset, old_cset, cgrp, template))
986 /* This css_set matches what we need */
990 /* No existing cgroup group matched */
994 static void free_cgrp_cset_links(struct list_head *links_to_free)
996 struct cgrp_cset_link *link, *tmp_link;
998 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
999 list_del(&link->cset_link);
1005 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1006 * @count: the number of links to allocate
1007 * @tmp_links: list_head the allocated links are put on
1009 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1010 * through ->cset_link. Returns 0 on success or -errno.
1012 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1014 struct cgrp_cset_link *link;
1017 INIT_LIST_HEAD(tmp_links);
1019 for (i = 0; i < count; i++) {
1020 link = kzalloc(sizeof(*link), GFP_KERNEL);
1022 free_cgrp_cset_links(tmp_links);
1025 list_add(&link->cset_link, tmp_links);
1031 * link_css_set - a helper function to link a css_set to a cgroup
1032 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1033 * @cset: the css_set to be linked
1034 * @cgrp: the destination cgroup
1036 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1037 struct cgroup *cgrp)
1039 struct cgrp_cset_link *link;
1041 BUG_ON(list_empty(tmp_links));
1043 if (cgroup_on_dfl(cgrp))
1044 cset->dfl_cgrp = cgrp;
1046 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1051 * Always add links to the tail of the lists so that the lists are
1052 * in choronological order.
1054 list_move_tail(&link->cset_link, &cgrp->cset_links);
1055 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1057 if (cgroup_parent(cgrp))
1058 cgroup_get_live(cgrp);
1062 * find_css_set - return a new css_set with one cgroup updated
1063 * @old_cset: the baseline css_set
1064 * @cgrp: the cgroup to be updated
1066 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1067 * substituted into the appropriate hierarchy.
1069 static struct css_set *find_css_set(struct css_set *old_cset,
1070 struct cgroup *cgrp)
1072 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1073 struct css_set *cset;
1074 struct list_head tmp_links;
1075 struct cgrp_cset_link *link;
1076 struct cgroup_subsys *ss;
1080 lockdep_assert_held(&cgroup_mutex);
1082 /* First see if we already have a cgroup group that matches
1083 * the desired set */
1084 spin_lock_irq(&css_set_lock);
1085 cset = find_existing_css_set(old_cset, cgrp, template);
1088 spin_unlock_irq(&css_set_lock);
1093 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1097 /* Allocate all the cgrp_cset_link objects that we'll need */
1098 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1103 refcount_set(&cset->refcount, 1);
1104 cset->dom_cset = cset;
1105 INIT_LIST_HEAD(&cset->tasks);
1106 INIT_LIST_HEAD(&cset->mg_tasks);
1107 INIT_LIST_HEAD(&cset->task_iters);
1108 INIT_LIST_HEAD(&cset->threaded_csets);
1109 INIT_HLIST_NODE(&cset->hlist);
1110 INIT_LIST_HEAD(&cset->cgrp_links);
1111 INIT_LIST_HEAD(&cset->mg_preload_node);
1112 INIT_LIST_HEAD(&cset->mg_node);
1114 /* Copy the set of subsystem state objects generated in
1115 * find_existing_css_set() */
1116 memcpy(cset->subsys, template, sizeof(cset->subsys));
1118 spin_lock_irq(&css_set_lock);
1119 /* Add reference counts and links from the new css_set. */
1120 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1121 struct cgroup *c = link->cgrp;
1123 if (c->root == cgrp->root)
1125 link_css_set(&tmp_links, cset, c);
1128 BUG_ON(!list_empty(&tmp_links));
1132 /* Add @cset to the hash table */
1133 key = css_set_hash(cset->subsys);
1134 hash_add(css_set_table, &cset->hlist, key);
1136 for_each_subsys(ss, ssid) {
1137 struct cgroup_subsys_state *css = cset->subsys[ssid];
1139 list_add_tail(&cset->e_cset_node[ssid],
1140 &css->cgroup->e_csets[ssid]);
1144 spin_unlock_irq(&css_set_lock);
1147 * If @cset should be threaded, look up the matching dom_cset and
1148 * link them up. We first fully initialize @cset then look for the
1149 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1150 * to stay empty until we return.
1152 if (cgroup_is_threaded(cset->dfl_cgrp)) {
1153 struct css_set *dcset;
1155 dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
1161 spin_lock_irq(&css_set_lock);
1162 cset->dom_cset = dcset;
1163 list_add_tail(&cset->threaded_csets_node,
1164 &dcset->threaded_csets);
1165 spin_unlock_irq(&css_set_lock);
1171 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1173 struct cgroup *root_cgrp = kf_root->kn->priv;
1175 return root_cgrp->root;
1178 static int cgroup_init_root_id(struct cgroup_root *root)
1182 lockdep_assert_held(&cgroup_mutex);
1184 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1188 root->hierarchy_id = id;
1192 static void cgroup_exit_root_id(struct cgroup_root *root)
1194 lockdep_assert_held(&cgroup_mutex);
1196 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1199 void cgroup_free_root(struct cgroup_root *root)
1202 idr_destroy(&root->cgroup_idr);
1207 static void cgroup_destroy_root(struct cgroup_root *root)
1209 struct cgroup *cgrp = &root->cgrp;
1210 struct cgrp_cset_link *link, *tmp_link;
1212 trace_cgroup_destroy_root(root);
1214 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1216 BUG_ON(atomic_read(&root->nr_cgrps));
1217 BUG_ON(!list_empty(&cgrp->self.children));
1219 /* Rebind all subsystems back to the default hierarchy */
1220 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1223 * Release all the links from cset_links to this hierarchy's
1226 spin_lock_irq(&css_set_lock);
1228 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1229 list_del(&link->cset_link);
1230 list_del(&link->cgrp_link);
1234 spin_unlock_irq(&css_set_lock);
1236 if (!list_empty(&root->root_list)) {
1237 list_del(&root->root_list);
1238 cgroup_root_count--;
1241 cgroup_exit_root_id(root);
1243 mutex_unlock(&cgroup_mutex);
1245 kernfs_destroy_root(root->kf_root);
1246 cgroup_free_root(root);
1250 * look up cgroup associated with current task's cgroup namespace on the
1251 * specified hierarchy
1253 static struct cgroup *
1254 current_cgns_cgroup_from_root(struct cgroup_root *root)
1256 struct cgroup *res = NULL;
1257 struct css_set *cset;
1259 lockdep_assert_held(&css_set_lock);
1263 cset = current->nsproxy->cgroup_ns->root_cset;
1264 if (cset == &init_css_set) {
1267 struct cgrp_cset_link *link;
1269 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1270 struct cgroup *c = link->cgrp;
1272 if (c->root == root) {
1284 /* look up cgroup associated with given css_set on the specified hierarchy */
1285 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1286 struct cgroup_root *root)
1288 struct cgroup *res = NULL;
1290 lockdep_assert_held(&cgroup_mutex);
1291 lockdep_assert_held(&css_set_lock);
1293 if (cset == &init_css_set) {
1295 } else if (root == &cgrp_dfl_root) {
1296 res = cset->dfl_cgrp;
1298 struct cgrp_cset_link *link;
1300 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1301 struct cgroup *c = link->cgrp;
1303 if (c->root == root) {
1315 * Return the cgroup for "task" from the given hierarchy. Must be
1316 * called with cgroup_mutex and css_set_lock held.
1318 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1319 struct cgroup_root *root)
1322 * No need to lock the task - since we hold cgroup_mutex the
1323 * task can't change groups, so the only thing that can happen
1324 * is that it exits and its css is set back to init_css_set.
1326 return cset_cgroup_from_root(task_css_set(task), root);
1330 * A task must hold cgroup_mutex to modify cgroups.
1332 * Any task can increment and decrement the count field without lock.
1333 * So in general, code holding cgroup_mutex can't rely on the count
1334 * field not changing. However, if the count goes to zero, then only
1335 * cgroup_attach_task() can increment it again. Because a count of zero
1336 * means that no tasks are currently attached, therefore there is no
1337 * way a task attached to that cgroup can fork (the other way to
1338 * increment the count). So code holding cgroup_mutex can safely
1339 * assume that if the count is zero, it will stay zero. Similarly, if
1340 * a task holds cgroup_mutex on a cgroup with zero count, it
1341 * knows that the cgroup won't be removed, as cgroup_rmdir()
1344 * A cgroup can only be deleted if both its 'count' of using tasks
1345 * is zero, and its list of 'children' cgroups is empty. Since all
1346 * tasks in the system use _some_ cgroup, and since there is always at
1347 * least one task in the system (init, pid == 1), therefore, root cgroup
1348 * always has either children cgroups and/or using tasks. So we don't
1349 * need a special hack to ensure that root cgroup cannot be deleted.
1351 * P.S. One more locking exception. RCU is used to guard the
1352 * update of a tasks cgroup pointer by cgroup_attach_task()
1355 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1357 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1360 struct cgroup_subsys *ss = cft->ss;
1362 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1363 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
1364 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
1365 cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1368 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1373 * cgroup_file_mode - deduce file mode of a control file
1374 * @cft: the control file in question
1376 * S_IRUGO for read, S_IWUSR for write.
1378 static umode_t cgroup_file_mode(const struct cftype *cft)
1382 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1385 if (cft->write_u64 || cft->write_s64 || cft->write) {
1386 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1396 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1397 * @subtree_control: the new subtree_control mask to consider
1398 * @this_ss_mask: available subsystems
1400 * On the default hierarchy, a subsystem may request other subsystems to be
1401 * enabled together through its ->depends_on mask. In such cases, more
1402 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1404 * This function calculates which subsystems need to be enabled if
1405 * @subtree_control is to be applied while restricted to @this_ss_mask.
1407 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1409 u16 cur_ss_mask = subtree_control;
1410 struct cgroup_subsys *ss;
1413 lockdep_assert_held(&cgroup_mutex);
1415 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1418 u16 new_ss_mask = cur_ss_mask;
1420 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1421 new_ss_mask |= ss->depends_on;
1422 } while_each_subsys_mask();
1425 * Mask out subsystems which aren't available. This can
1426 * happen only if some depended-upon subsystems were bound
1427 * to non-default hierarchies.
1429 new_ss_mask &= this_ss_mask;
1431 if (new_ss_mask == cur_ss_mask)
1433 cur_ss_mask = new_ss_mask;
1440 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1441 * @kn: the kernfs_node being serviced
1443 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1444 * the method finishes if locking succeeded. Note that once this function
1445 * returns the cgroup returned by cgroup_kn_lock_live() may become
1446 * inaccessible any time. If the caller intends to continue to access the
1447 * cgroup, it should pin it before invoking this function.
1449 void cgroup_kn_unlock(struct kernfs_node *kn)
1451 struct cgroup *cgrp;
1453 if (kernfs_type(kn) == KERNFS_DIR)
1456 cgrp = kn->parent->priv;
1458 mutex_unlock(&cgroup_mutex);
1460 kernfs_unbreak_active_protection(kn);
1465 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1466 * @kn: the kernfs_node being serviced
1467 * @drain_offline: perform offline draining on the cgroup
1469 * This helper is to be used by a cgroup kernfs method currently servicing
1470 * @kn. It breaks the active protection, performs cgroup locking and
1471 * verifies that the associated cgroup is alive. Returns the cgroup if
1472 * alive; otherwise, %NULL. A successful return should be undone by a
1473 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1474 * cgroup is drained of offlining csses before return.
1476 * Any cgroup kernfs method implementation which requires locking the
1477 * associated cgroup should use this helper. It avoids nesting cgroup
1478 * locking under kernfs active protection and allows all kernfs operations
1479 * including self-removal.
1481 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1483 struct cgroup *cgrp;
1485 if (kernfs_type(kn) == KERNFS_DIR)
1488 cgrp = kn->parent->priv;
1491 * We're gonna grab cgroup_mutex which nests outside kernfs
1492 * active_ref. cgroup liveliness check alone provides enough
1493 * protection against removal. Ensure @cgrp stays accessible and
1494 * break the active_ref protection.
1496 if (!cgroup_tryget(cgrp))
1498 kernfs_break_active_protection(kn);
1501 cgroup_lock_and_drain_offline(cgrp);
1503 mutex_lock(&cgroup_mutex);
1505 if (!cgroup_is_dead(cgrp))
1508 cgroup_kn_unlock(kn);
1512 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1514 char name[CGROUP_FILE_NAME_MAX];
1516 lockdep_assert_held(&cgroup_mutex);
1518 if (cft->file_offset) {
1519 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1520 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1522 spin_lock_irq(&cgroup_file_kn_lock);
1524 spin_unlock_irq(&cgroup_file_kn_lock);
1527 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1531 * css_clear_dir - remove subsys files in a cgroup directory
1534 static void css_clear_dir(struct cgroup_subsys_state *css)
1536 struct cgroup *cgrp = css->cgroup;
1537 struct cftype *cfts;
1539 if (!(css->flags & CSS_VISIBLE))
1542 css->flags &= ~CSS_VISIBLE;
1544 list_for_each_entry(cfts, &css->ss->cfts, node)
1545 cgroup_addrm_files(css, cgrp, cfts, false);
1549 * css_populate_dir - create subsys files in a cgroup directory
1552 * On failure, no file is added.
1554 static int css_populate_dir(struct cgroup_subsys_state *css)
1556 struct cgroup *cgrp = css->cgroup;
1557 struct cftype *cfts, *failed_cfts;
1560 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1564 if (cgroup_on_dfl(cgrp))
1565 cfts = cgroup_base_files;
1567 cfts = cgroup1_base_files;
1569 return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1572 list_for_each_entry(cfts, &css->ss->cfts, node) {
1573 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1580 css->flags |= CSS_VISIBLE;
1584 list_for_each_entry(cfts, &css->ss->cfts, node) {
1585 if (cfts == failed_cfts)
1587 cgroup_addrm_files(css, cgrp, cfts, false);
1592 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1594 struct cgroup *dcgrp = &dst_root->cgrp;
1595 struct cgroup_subsys *ss;
1598 lockdep_assert_held(&cgroup_mutex);
1600 do_each_subsys_mask(ss, ssid, ss_mask) {
1602 * If @ss has non-root csses attached to it, can't move.
1603 * If @ss is an implicit controller, it is exempt from this
1604 * rule and can be stolen.
1606 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1607 !ss->implicit_on_dfl)
1610 /* can't move between two non-dummy roots either */
1611 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1613 } while_each_subsys_mask();
1615 do_each_subsys_mask(ss, ssid, ss_mask) {
1616 struct cgroup_root *src_root = ss->root;
1617 struct cgroup *scgrp = &src_root->cgrp;
1618 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1619 struct css_set *cset;
1621 WARN_ON(!css || cgroup_css(dcgrp, ss));
1623 /* disable from the source */
1624 src_root->subsys_mask &= ~(1 << ssid);
1625 WARN_ON(cgroup_apply_control(scgrp));
1626 cgroup_finalize_control(scgrp, 0);
1629 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1630 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1631 ss->root = dst_root;
1632 css->cgroup = dcgrp;
1634 spin_lock_irq(&css_set_lock);
1635 hash_for_each(css_set_table, i, cset, hlist)
1636 list_move_tail(&cset->e_cset_node[ss->id],
1637 &dcgrp->e_csets[ss->id]);
1638 spin_unlock_irq(&css_set_lock);
1640 /* default hierarchy doesn't enable controllers by default */
1641 dst_root->subsys_mask |= 1 << ssid;
1642 if (dst_root == &cgrp_dfl_root) {
1643 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1645 dcgrp->subtree_control |= 1 << ssid;
1646 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1649 ret = cgroup_apply_control(dcgrp);
1651 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1656 } while_each_subsys_mask();
1658 kernfs_activate(dcgrp->kn);
1662 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1663 struct kernfs_root *kf_root)
1667 struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1668 struct cgroup *ns_cgroup;
1670 buf = kmalloc(PATH_MAX, GFP_KERNEL);
1674 spin_lock_irq(&css_set_lock);
1675 ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1676 len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1677 spin_unlock_irq(&css_set_lock);
1679 if (len >= PATH_MAX)
1682 seq_escape(sf, buf, " \t\n\\");
1689 static int parse_cgroup_root_flags(char *data, unsigned int *root_flags)
1698 while ((token = strsep(&data, ",")) != NULL) {
1699 if (!strcmp(token, "nsdelegate")) {
1700 *root_flags |= CGRP_ROOT_NS_DELEGATE;
1704 pr_err("cgroup2: unknown option \"%s\"\n", token);
1711 static void apply_cgroup_root_flags(unsigned int root_flags)
1713 if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
1714 if (root_flags & CGRP_ROOT_NS_DELEGATE)
1715 cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
1717 cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
1721 static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
1723 if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
1724 seq_puts(seq, ",nsdelegate");
1728 static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
1730 unsigned int root_flags;
1733 ret = parse_cgroup_root_flags(data, &root_flags);
1737 apply_cgroup_root_flags(root_flags);
1742 * To reduce the fork() overhead for systems that are not actually using
1743 * their cgroups capability, we don't maintain the lists running through
1744 * each css_set to its tasks until we see the list actually used - in other
1745 * words after the first mount.
1747 static bool use_task_css_set_links __read_mostly;
1749 static void cgroup_enable_task_cg_lists(void)
1751 struct task_struct *p, *g;
1753 spin_lock_irq(&css_set_lock);
1755 if (use_task_css_set_links)
1758 use_task_css_set_links = true;
1761 * We need tasklist_lock because RCU is not safe against
1762 * while_each_thread(). Besides, a forking task that has passed
1763 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1764 * is not guaranteed to have its child immediately visible in the
1765 * tasklist if we walk through it with RCU.
1767 read_lock(&tasklist_lock);
1768 do_each_thread(g, p) {
1769 WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1770 task_css_set(p) != &init_css_set);
1773 * We should check if the process is exiting, otherwise
1774 * it will race with cgroup_exit() in that the list
1775 * entry won't be deleted though the process has exited.
1776 * Do it while holding siglock so that we don't end up
1777 * racing against cgroup_exit().
1779 * Interrupts were already disabled while acquiring
1780 * the css_set_lock, so we do not need to disable it
1781 * again when acquiring the sighand->siglock here.
1783 spin_lock(&p->sighand->siglock);
1784 if (!(p->flags & PF_EXITING)) {
1785 struct css_set *cset = task_css_set(p);
1787 if (!css_set_populated(cset))
1788 css_set_update_populated(cset, true);
1789 list_add_tail(&p->cg_list, &cset->tasks);
1793 spin_unlock(&p->sighand->siglock);
1794 } while_each_thread(g, p);
1795 read_unlock(&tasklist_lock);
1797 spin_unlock_irq(&css_set_lock);
1800 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1802 struct cgroup_subsys *ss;
1805 INIT_LIST_HEAD(&cgrp->self.sibling);
1806 INIT_LIST_HEAD(&cgrp->self.children);
1807 INIT_LIST_HEAD(&cgrp->cset_links);
1808 INIT_LIST_HEAD(&cgrp->pidlists);
1809 mutex_init(&cgrp->pidlist_mutex);
1810 cgrp->self.cgroup = cgrp;
1811 cgrp->self.flags |= CSS_ONLINE;
1812 cgrp->dom_cgrp = cgrp;
1813 cgrp->max_descendants = INT_MAX;
1814 cgrp->max_depth = INT_MAX;
1816 for_each_subsys(ss, ssid)
1817 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1819 init_waitqueue_head(&cgrp->offline_waitq);
1820 INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1823 void init_cgroup_root(struct cgroup_root *root, struct cgroup_sb_opts *opts)
1825 struct cgroup *cgrp = &root->cgrp;
1827 INIT_LIST_HEAD(&root->root_list);
1828 atomic_set(&root->nr_cgrps, 1);
1830 init_cgroup_housekeeping(cgrp);
1831 idr_init(&root->cgroup_idr);
1833 root->flags = opts->flags;
1834 if (opts->release_agent)
1835 strcpy(root->release_agent_path, opts->release_agent);
1837 strcpy(root->name, opts->name);
1838 if (opts->cpuset_clone_children)
1839 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1842 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask, int ref_flags)
1844 LIST_HEAD(tmp_links);
1845 struct cgroup *root_cgrp = &root->cgrp;
1846 struct kernfs_syscall_ops *kf_sops;
1847 struct css_set *cset;
1850 lockdep_assert_held(&cgroup_mutex);
1852 ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL);
1855 root_cgrp->id = ret;
1856 root_cgrp->ancestor_ids[0] = ret;
1858 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
1859 ref_flags, GFP_KERNEL);
1864 * We're accessing css_set_count without locking css_set_lock here,
1865 * but that's OK - it can only be increased by someone holding
1866 * cgroup_lock, and that's us. Later rebinding may disable
1867 * controllers on the default hierarchy and thus create new csets,
1868 * which can't be more than the existing ones. Allocate 2x.
1870 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
1874 ret = cgroup_init_root_id(root);
1878 kf_sops = root == &cgrp_dfl_root ?
1879 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
1881 root->kf_root = kernfs_create_root(kf_sops,
1882 KERNFS_ROOT_CREATE_DEACTIVATED |
1883 KERNFS_ROOT_SUPPORT_EXPORTOP,
1885 if (IS_ERR(root->kf_root)) {
1886 ret = PTR_ERR(root->kf_root);
1889 root_cgrp->kn = root->kf_root->kn;
1891 ret = css_populate_dir(&root_cgrp->self);
1895 ret = rebind_subsystems(root, ss_mask);
1899 trace_cgroup_setup_root(root);
1902 * There must be no failure case after here, since rebinding takes
1903 * care of subsystems' refcounts, which are explicitly dropped in
1904 * the failure exit path.
1906 list_add(&root->root_list, &cgroup_roots);
1907 cgroup_root_count++;
1910 * Link the root cgroup in this hierarchy into all the css_set
1913 spin_lock_irq(&css_set_lock);
1914 hash_for_each(css_set_table, i, cset, hlist) {
1915 link_css_set(&tmp_links, cset, root_cgrp);
1916 if (css_set_populated(cset))
1917 cgroup_update_populated(root_cgrp, true);
1919 spin_unlock_irq(&css_set_lock);
1921 BUG_ON(!list_empty(&root_cgrp->self.children));
1922 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
1924 kernfs_activate(root_cgrp->kn);
1929 kernfs_destroy_root(root->kf_root);
1930 root->kf_root = NULL;
1932 cgroup_exit_root_id(root);
1934 percpu_ref_exit(&root_cgrp->self.refcnt);
1936 free_cgrp_cset_links(&tmp_links);
1940 struct dentry *cgroup_do_mount(struct file_system_type *fs_type, int flags,
1941 struct cgroup_root *root, unsigned long magic,
1942 struct cgroup_namespace *ns)
1944 struct dentry *dentry;
1947 dentry = kernfs_mount(fs_type, flags, root->kf_root, magic, &new_sb);
1950 * In non-init cgroup namespace, instead of root cgroup's dentry,
1951 * we return the dentry corresponding to the cgroupns->root_cgrp.
1953 if (!IS_ERR(dentry) && ns != &init_cgroup_ns) {
1954 struct dentry *nsdentry;
1955 struct cgroup *cgrp;
1957 mutex_lock(&cgroup_mutex);
1958 spin_lock_irq(&css_set_lock);
1960 cgrp = cset_cgroup_from_root(ns->root_cset, root);
1962 spin_unlock_irq(&css_set_lock);
1963 mutex_unlock(&cgroup_mutex);
1965 nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb);
1970 if (IS_ERR(dentry) || !new_sb)
1971 cgroup_put(&root->cgrp);
1976 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1977 int flags, const char *unused_dev_name,
1980 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
1981 struct dentry *dentry;
1986 /* Check if the caller has permission to mount. */
1987 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) {
1989 return ERR_PTR(-EPERM);
1993 * The first time anyone tries to mount a cgroup, enable the list
1994 * linking each css_set to its tasks and fix up all existing tasks.
1996 if (!use_task_css_set_links)
1997 cgroup_enable_task_cg_lists();
1999 if (fs_type == &cgroup2_fs_type) {
2000 unsigned int root_flags;
2002 ret = parse_cgroup_root_flags(data, &root_flags);
2005 return ERR_PTR(ret);
2008 cgrp_dfl_visible = true;
2009 cgroup_get_live(&cgrp_dfl_root.cgrp);
2011 dentry = cgroup_do_mount(&cgroup2_fs_type, flags, &cgrp_dfl_root,
2012 CGROUP2_SUPER_MAGIC, ns);
2013 if (!IS_ERR(dentry))
2014 apply_cgroup_root_flags(root_flags);
2016 dentry = cgroup1_mount(&cgroup_fs_type, flags, data,
2017 CGROUP_SUPER_MAGIC, ns);
2024 static void cgroup_kill_sb(struct super_block *sb)
2026 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2027 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2030 * If @root doesn't have any mounts or children, start killing it.
2031 * This prevents new mounts by disabling percpu_ref_tryget_live().
2032 * cgroup_mount() may wait for @root's release.
2034 * And don't kill the default root.
2036 if (!list_empty(&root->cgrp.self.children) ||
2037 root == &cgrp_dfl_root)
2038 cgroup_put(&root->cgrp);
2040 percpu_ref_kill(&root->cgrp.self.refcnt);
2045 struct file_system_type cgroup_fs_type = {
2047 .mount = cgroup_mount,
2048 .kill_sb = cgroup_kill_sb,
2049 .fs_flags = FS_USERNS_MOUNT,
2052 static struct file_system_type cgroup2_fs_type = {
2054 .mount = cgroup_mount,
2055 .kill_sb = cgroup_kill_sb,
2056 .fs_flags = FS_USERNS_MOUNT,
2059 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2060 struct cgroup_namespace *ns)
2062 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2064 return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2067 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2068 struct cgroup_namespace *ns)
2072 mutex_lock(&cgroup_mutex);
2073 spin_lock_irq(&css_set_lock);
2075 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2077 spin_unlock_irq(&css_set_lock);
2078 mutex_unlock(&cgroup_mutex);
2082 EXPORT_SYMBOL_GPL(cgroup_path_ns);
2085 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2086 * @task: target task
2087 * @buf: the buffer to write the path into
2088 * @buflen: the length of the buffer
2090 * Determine @task's cgroup on the first (the one with the lowest non-zero
2091 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2092 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2093 * cgroup controller callbacks.
2095 * Return value is the same as kernfs_path().
2097 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2099 struct cgroup_root *root;
2100 struct cgroup *cgrp;
2101 int hierarchy_id = 1;
2104 mutex_lock(&cgroup_mutex);
2105 spin_lock_irq(&css_set_lock);
2107 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2110 cgrp = task_cgroup_from_root(task, root);
2111 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2113 /* if no hierarchy exists, everyone is in "/" */
2114 ret = strlcpy(buf, "/", buflen);
2117 spin_unlock_irq(&css_set_lock);
2118 mutex_unlock(&cgroup_mutex);
2121 EXPORT_SYMBOL_GPL(task_cgroup_path);
2124 * cgroup_migrate_add_task - add a migration target task to a migration context
2125 * @task: target task
2126 * @mgctx: target migration context
2128 * Add @task, which is a migration target, to @mgctx->tset. This function
2129 * becomes noop if @task doesn't need to be migrated. @task's css_set
2130 * should have been added as a migration source and @task->cg_list will be
2131 * moved from the css_set's tasks list to mg_tasks one.
2133 static void cgroup_migrate_add_task(struct task_struct *task,
2134 struct cgroup_mgctx *mgctx)
2136 struct css_set *cset;
2138 lockdep_assert_held(&css_set_lock);
2140 /* @task either already exited or can't exit until the end */
2141 if (task->flags & PF_EXITING)
2144 /* leave @task alone if post_fork() hasn't linked it yet */
2145 if (list_empty(&task->cg_list))
2148 cset = task_css_set(task);
2149 if (!cset->mg_src_cgrp)
2152 mgctx->tset.nr_tasks++;
2154 list_move_tail(&task->cg_list, &cset->mg_tasks);
2155 if (list_empty(&cset->mg_node))
2156 list_add_tail(&cset->mg_node,
2157 &mgctx->tset.src_csets);
2158 if (list_empty(&cset->mg_dst_cset->mg_node))
2159 list_add_tail(&cset->mg_dst_cset->mg_node,
2160 &mgctx->tset.dst_csets);
2164 * cgroup_taskset_first - reset taskset and return the first task
2165 * @tset: taskset of interest
2166 * @dst_cssp: output variable for the destination css
2168 * @tset iteration is initialized and the first task is returned.
2170 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2171 struct cgroup_subsys_state **dst_cssp)
2173 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2174 tset->cur_task = NULL;
2176 return cgroup_taskset_next(tset, dst_cssp);
2180 * cgroup_taskset_next - iterate to the next task in taskset
2181 * @tset: taskset of interest
2182 * @dst_cssp: output variable for the destination css
2184 * Return the next task in @tset. Iteration must have been initialized
2185 * with cgroup_taskset_first().
2187 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2188 struct cgroup_subsys_state **dst_cssp)
2190 struct css_set *cset = tset->cur_cset;
2191 struct task_struct *task = tset->cur_task;
2193 while (&cset->mg_node != tset->csets) {
2195 task = list_first_entry(&cset->mg_tasks,
2196 struct task_struct, cg_list);
2198 task = list_next_entry(task, cg_list);
2200 if (&task->cg_list != &cset->mg_tasks) {
2201 tset->cur_cset = cset;
2202 tset->cur_task = task;
2205 * This function may be called both before and
2206 * after cgroup_taskset_migrate(). The two cases
2207 * can be distinguished by looking at whether @cset
2208 * has its ->mg_dst_cset set.
2210 if (cset->mg_dst_cset)
2211 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2213 *dst_cssp = cset->subsys[tset->ssid];
2218 cset = list_next_entry(cset, mg_node);
2226 * cgroup_taskset_migrate - migrate a taskset
2227 * @mgctx: migration context
2229 * Migrate tasks in @mgctx as setup by migration preparation functions.
2230 * This function fails iff one of the ->can_attach callbacks fails and
2231 * guarantees that either all or none of the tasks in @mgctx are migrated.
2232 * @mgctx is consumed regardless of success.
2234 static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2236 struct cgroup_taskset *tset = &mgctx->tset;
2237 struct cgroup_subsys *ss;
2238 struct task_struct *task, *tmp_task;
2239 struct css_set *cset, *tmp_cset;
2240 int ssid, failed_ssid, ret;
2242 /* check that we can legitimately attach to the cgroup */
2243 if (tset->nr_tasks) {
2244 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2245 if (ss->can_attach) {
2247 ret = ss->can_attach(tset);
2250 goto out_cancel_attach;
2253 } while_each_subsys_mask();
2257 * Now that we're guaranteed success, proceed to move all tasks to
2258 * the new cgroup. There are no failure cases after here, so this
2259 * is the commit point.
2261 spin_lock_irq(&css_set_lock);
2262 list_for_each_entry(cset, &tset->src_csets, mg_node) {
2263 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2264 struct css_set *from_cset = task_css_set(task);
2265 struct css_set *to_cset = cset->mg_dst_cset;
2267 get_css_set(to_cset);
2268 to_cset->nr_tasks++;
2269 css_set_move_task(task, from_cset, to_cset, true);
2270 put_css_set_locked(from_cset);
2271 from_cset->nr_tasks--;
2274 spin_unlock_irq(&css_set_lock);
2277 * Migration is committed, all target tasks are now on dst_csets.
2278 * Nothing is sensitive to fork() after this point. Notify
2279 * controllers that migration is complete.
2281 tset->csets = &tset->dst_csets;
2283 if (tset->nr_tasks) {
2284 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2289 } while_each_subsys_mask();
2293 goto out_release_tset;
2296 if (tset->nr_tasks) {
2297 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2298 if (ssid == failed_ssid)
2300 if (ss->cancel_attach) {
2302 ss->cancel_attach(tset);
2304 } while_each_subsys_mask();
2307 spin_lock_irq(&css_set_lock);
2308 list_splice_init(&tset->dst_csets, &tset->src_csets);
2309 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2310 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2311 list_del_init(&cset->mg_node);
2313 spin_unlock_irq(&css_set_lock);
2318 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2319 * @dst_cgrp: destination cgroup to test
2321 * On the default hierarchy, except for the mixable, (possible) thread root
2322 * and threaded cgroups, subtree_control must be zero for migration
2323 * destination cgroups with tasks so that child cgroups don't compete
2326 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2328 /* v1 doesn't have any restriction */
2329 if (!cgroup_on_dfl(dst_cgrp))
2332 /* verify @dst_cgrp can host resources */
2333 if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
2336 /* mixables don't care */
2337 if (cgroup_is_mixable(dst_cgrp))
2341 * If @dst_cgrp is already or can become a thread root or is
2342 * threaded, it doesn't matter.
2344 if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
2347 /* apply no-internal-process constraint */
2348 if (dst_cgrp->subtree_control)
2355 * cgroup_migrate_finish - cleanup after attach
2356 * @mgctx: migration context
2358 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2359 * those functions for details.
2361 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2363 LIST_HEAD(preloaded);
2364 struct css_set *cset, *tmp_cset;
2366 lockdep_assert_held(&cgroup_mutex);
2368 spin_lock_irq(&css_set_lock);
2370 list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded);
2371 list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded);
2373 list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) {
2374 cset->mg_src_cgrp = NULL;
2375 cset->mg_dst_cgrp = NULL;
2376 cset->mg_dst_cset = NULL;
2377 list_del_init(&cset->mg_preload_node);
2378 put_css_set_locked(cset);
2381 spin_unlock_irq(&css_set_lock);
2385 * cgroup_migrate_add_src - add a migration source css_set
2386 * @src_cset: the source css_set to add
2387 * @dst_cgrp: the destination cgroup
2388 * @mgctx: migration context
2390 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2391 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2392 * up by cgroup_migrate_finish().
2394 * This function may be called without holding cgroup_threadgroup_rwsem
2395 * even if the target is a process. Threads may be created and destroyed
2396 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2397 * into play and the preloaded css_sets are guaranteed to cover all
2400 void cgroup_migrate_add_src(struct css_set *src_cset,
2401 struct cgroup *dst_cgrp,
2402 struct cgroup_mgctx *mgctx)
2404 struct cgroup *src_cgrp;
2406 lockdep_assert_held(&cgroup_mutex);
2407 lockdep_assert_held(&css_set_lock);
2410 * If ->dead, @src_set is associated with one or more dead cgroups
2411 * and doesn't contain any migratable tasks. Ignore it early so
2412 * that the rest of migration path doesn't get confused by it.
2417 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2419 if (!list_empty(&src_cset->mg_preload_node))
2422 WARN_ON(src_cset->mg_src_cgrp);
2423 WARN_ON(src_cset->mg_dst_cgrp);
2424 WARN_ON(!list_empty(&src_cset->mg_tasks));
2425 WARN_ON(!list_empty(&src_cset->mg_node));
2427 src_cset->mg_src_cgrp = src_cgrp;
2428 src_cset->mg_dst_cgrp = dst_cgrp;
2429 get_css_set(src_cset);
2430 list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets);
2434 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2435 * @mgctx: migration context
2437 * Tasks are about to be moved and all the source css_sets have been
2438 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2439 * pins all destination css_sets, links each to its source, and append them
2440 * to @mgctx->preloaded_dst_csets.
2442 * This function must be called after cgroup_migrate_add_src() has been
2443 * called on each migration source css_set. After migration is performed
2444 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2447 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2449 struct css_set *src_cset, *tmp_cset;
2451 lockdep_assert_held(&cgroup_mutex);
2453 /* look up the dst cset for each src cset and link it to src */
2454 list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2456 struct css_set *dst_cset;
2457 struct cgroup_subsys *ss;
2460 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2464 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2467 * If src cset equals dst, it's noop. Drop the src.
2468 * cgroup_migrate() will skip the cset too. Note that we
2469 * can't handle src == dst as some nodes are used by both.
2471 if (src_cset == dst_cset) {
2472 src_cset->mg_src_cgrp = NULL;
2473 src_cset->mg_dst_cgrp = NULL;
2474 list_del_init(&src_cset->mg_preload_node);
2475 put_css_set(src_cset);
2476 put_css_set(dst_cset);
2480 src_cset->mg_dst_cset = dst_cset;
2482 if (list_empty(&dst_cset->mg_preload_node))
2483 list_add_tail(&dst_cset->mg_preload_node,
2484 &mgctx->preloaded_dst_csets);
2486 put_css_set(dst_cset);
2488 for_each_subsys(ss, ssid)
2489 if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2490 mgctx->ss_mask |= 1 << ssid;
2495 cgroup_migrate_finish(mgctx);
2500 * cgroup_migrate - migrate a process or task to a cgroup
2501 * @leader: the leader of the process or the task to migrate
2502 * @threadgroup: whether @leader points to the whole process or a single task
2503 * @mgctx: migration context
2505 * Migrate a process or task denoted by @leader. If migrating a process,
2506 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2507 * responsible for invoking cgroup_migrate_add_src() and
2508 * cgroup_migrate_prepare_dst() on the targets before invoking this
2509 * function and following up with cgroup_migrate_finish().
2511 * As long as a controller's ->can_attach() doesn't fail, this function is
2512 * guaranteed to succeed. This means that, excluding ->can_attach()
2513 * failure, when migrating multiple targets, the success or failure can be
2514 * decided for all targets by invoking group_migrate_prepare_dst() before
2515 * actually starting migrating.
2517 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2518 struct cgroup_mgctx *mgctx)
2520 struct task_struct *task;
2523 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2524 * already PF_EXITING could be freed from underneath us unless we
2525 * take an rcu_read_lock.
2527 spin_lock_irq(&css_set_lock);
2531 cgroup_migrate_add_task(task, mgctx);
2534 } while_each_thread(leader, task);
2536 spin_unlock_irq(&css_set_lock);
2538 return cgroup_migrate_execute(mgctx);
2542 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2543 * @dst_cgrp: the cgroup to attach to
2544 * @leader: the task or the leader of the threadgroup to be attached
2545 * @threadgroup: attach the whole threadgroup?
2547 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2549 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2552 DEFINE_CGROUP_MGCTX(mgctx);
2553 struct task_struct *task;
2556 ret = cgroup_migrate_vet_dst(dst_cgrp);
2560 /* look up all src csets */
2561 spin_lock_irq(&css_set_lock);
2565 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2568 } while_each_thread(leader, task);
2570 spin_unlock_irq(&css_set_lock);
2572 /* prepare dst csets and commit */
2573 ret = cgroup_migrate_prepare_dst(&mgctx);
2575 ret = cgroup_migrate(leader, threadgroup, &mgctx);
2577 cgroup_migrate_finish(&mgctx);
2580 trace_cgroup_attach_task(dst_cgrp, leader, threadgroup);
2585 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup)
2586 __acquires(&cgroup_threadgroup_rwsem)
2588 struct task_struct *tsk;
2591 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2592 return ERR_PTR(-EINVAL);
2594 percpu_down_write(&cgroup_threadgroup_rwsem);
2598 tsk = find_task_by_vpid(pid);
2600 tsk = ERR_PTR(-ESRCH);
2601 goto out_unlock_threadgroup;
2608 tsk = tsk->group_leader;
2611 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2612 * If userland migrates such a kthread to a non-root cgroup, it can
2613 * become trapped in a cpuset, or RT kthread may be born in a
2614 * cgroup with no rt_runtime allocated. Just say no.
2616 if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2617 tsk = ERR_PTR(-EINVAL);
2618 goto out_unlock_threadgroup;
2621 get_task_struct(tsk);
2622 goto out_unlock_rcu;
2624 out_unlock_threadgroup:
2625 percpu_up_write(&cgroup_threadgroup_rwsem);
2631 void cgroup_procs_write_finish(struct task_struct *task)
2632 __releases(&cgroup_threadgroup_rwsem)
2634 struct cgroup_subsys *ss;
2637 /* release reference from cgroup_procs_write_start() */
2638 put_task_struct(task);
2640 percpu_up_write(&cgroup_threadgroup_rwsem);
2641 for_each_subsys(ss, ssid)
2642 if (ss->post_attach)
2646 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2648 struct cgroup_subsys *ss;
2649 bool printed = false;
2652 do_each_subsys_mask(ss, ssid, ss_mask) {
2655 seq_printf(seq, "%s", ss->name);
2657 } while_each_subsys_mask();
2659 seq_putc(seq, '\n');
2662 /* show controllers which are enabled from the parent */
2663 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2665 struct cgroup *cgrp = seq_css(seq)->cgroup;
2667 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2671 /* show controllers which are enabled for a given cgroup's children */
2672 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2674 struct cgroup *cgrp = seq_css(seq)->cgroup;
2676 cgroup_print_ss_mask(seq, cgrp->subtree_control);
2681 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2682 * @cgrp: root of the subtree to update csses for
2684 * @cgrp's control masks have changed and its subtree's css associations
2685 * need to be updated accordingly. This function looks up all css_sets
2686 * which are attached to the subtree, creates the matching updated css_sets
2687 * and migrates the tasks to the new ones.
2689 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2691 DEFINE_CGROUP_MGCTX(mgctx);
2692 struct cgroup_subsys_state *d_css;
2693 struct cgroup *dsct;
2694 struct css_set *src_cset;
2697 lockdep_assert_held(&cgroup_mutex);
2699 percpu_down_write(&cgroup_threadgroup_rwsem);
2701 /* look up all csses currently attached to @cgrp's subtree */
2702 spin_lock_irq(&css_set_lock);
2703 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2704 struct cgrp_cset_link *link;
2706 list_for_each_entry(link, &dsct->cset_links, cset_link)
2707 cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2709 spin_unlock_irq(&css_set_lock);
2711 /* NULL dst indicates self on default hierarchy */
2712 ret = cgroup_migrate_prepare_dst(&mgctx);
2716 spin_lock_irq(&css_set_lock);
2717 list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) {
2718 struct task_struct *task, *ntask;
2720 /* all tasks in src_csets need to be migrated */
2721 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2722 cgroup_migrate_add_task(task, &mgctx);
2724 spin_unlock_irq(&css_set_lock);
2726 ret = cgroup_migrate_execute(&mgctx);
2728 cgroup_migrate_finish(&mgctx);
2729 percpu_up_write(&cgroup_threadgroup_rwsem);
2734 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2735 * @cgrp: root of the target subtree
2737 * Because css offlining is asynchronous, userland may try to re-enable a
2738 * controller while the previous css is still around. This function grabs
2739 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2741 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
2742 __acquires(&cgroup_mutex)
2744 struct cgroup *dsct;
2745 struct cgroup_subsys_state *d_css;
2746 struct cgroup_subsys *ss;
2750 mutex_lock(&cgroup_mutex);
2752 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2753 for_each_subsys(ss, ssid) {
2754 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2757 if (!css || !percpu_ref_is_dying(&css->refcnt))
2760 cgroup_get_live(dsct);
2761 prepare_to_wait(&dsct->offline_waitq, &wait,
2762 TASK_UNINTERRUPTIBLE);
2764 mutex_unlock(&cgroup_mutex);
2766 finish_wait(&dsct->offline_waitq, &wait);
2775 * cgroup_save_control - save control masks of a subtree
2776 * @cgrp: root of the target subtree
2778 * Save ->subtree_control and ->subtree_ss_mask to the respective old_
2779 * prefixed fields for @cgrp's subtree including @cgrp itself.
2781 static void cgroup_save_control(struct cgroup *cgrp)
2783 struct cgroup *dsct;
2784 struct cgroup_subsys_state *d_css;
2786 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2787 dsct->old_subtree_control = dsct->subtree_control;
2788 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
2793 * cgroup_propagate_control - refresh control masks of a subtree
2794 * @cgrp: root of the target subtree
2796 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
2797 * ->subtree_control and propagate controller availability through the
2798 * subtree so that descendants don't have unavailable controllers enabled.
2800 static void cgroup_propagate_control(struct cgroup *cgrp)
2802 struct cgroup *dsct;
2803 struct cgroup_subsys_state *d_css;
2805 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2806 dsct->subtree_control &= cgroup_control(dsct);
2807 dsct->subtree_ss_mask =
2808 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
2809 cgroup_ss_mask(dsct));
2814 * cgroup_restore_control - restore control masks of a subtree
2815 * @cgrp: root of the target subtree
2817 * Restore ->subtree_control and ->subtree_ss_mask from the respective old_
2818 * prefixed fields for @cgrp's subtree including @cgrp itself.
2820 static void cgroup_restore_control(struct cgroup *cgrp)
2822 struct cgroup *dsct;
2823 struct cgroup_subsys_state *d_css;
2825 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2826 dsct->subtree_control = dsct->old_subtree_control;
2827 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
2831 static bool css_visible(struct cgroup_subsys_state *css)
2833 struct cgroup_subsys *ss = css->ss;
2834 struct cgroup *cgrp = css->cgroup;
2836 if (cgroup_control(cgrp) & (1 << ss->id))
2838 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
2840 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
2844 * cgroup_apply_control_enable - enable or show csses according to control
2845 * @cgrp: root of the target subtree
2847 * Walk @cgrp's subtree and create new csses or make the existing ones
2848 * visible. A css is created invisible if it's being implicitly enabled
2849 * through dependency. An invisible css is made visible when the userland
2850 * explicitly enables it.
2852 * Returns 0 on success, -errno on failure. On failure, csses which have
2853 * been processed already aren't cleaned up. The caller is responsible for
2854 * cleaning up with cgroup_apply_control_disable().
2856 static int cgroup_apply_control_enable(struct cgroup *cgrp)
2858 struct cgroup *dsct;
2859 struct cgroup_subsys_state *d_css;
2860 struct cgroup_subsys *ss;
2863 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2864 for_each_subsys(ss, ssid) {
2865 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2867 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
2869 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
2873 css = css_create(dsct, ss);
2875 return PTR_ERR(css);
2878 if (css_visible(css)) {
2879 ret = css_populate_dir(css);
2890 * cgroup_apply_control_disable - kill or hide csses according to control
2891 * @cgrp: root of the target subtree
2893 * Walk @cgrp's subtree and kill and hide csses so that they match
2894 * cgroup_ss_mask() and cgroup_visible_mask().
2896 * A css is hidden when the userland requests it to be disabled while other
2897 * subsystems are still depending on it. The css must not actively control
2898 * resources and be in the vanilla state if it's made visible again later.
2899 * Controllers which may be depended upon should provide ->css_reset() for
2902 static void cgroup_apply_control_disable(struct cgroup *cgrp)
2904 struct cgroup *dsct;
2905 struct cgroup_subsys_state *d_css;
2906 struct cgroup_subsys *ss;
2909 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2910 for_each_subsys(ss, ssid) {
2911 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2913 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
2919 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
2921 } else if (!css_visible(css)) {
2931 * cgroup_apply_control - apply control mask updates to the subtree
2932 * @cgrp: root of the target subtree
2934 * subsystems can be enabled and disabled in a subtree using the following
2937 * 1. Call cgroup_save_control() to stash the current state.
2938 * 2. Update ->subtree_control masks in the subtree as desired.
2939 * 3. Call cgroup_apply_control() to apply the changes.
2940 * 4. Optionally perform other related operations.
2941 * 5. Call cgroup_finalize_control() to finish up.
2943 * This function implements step 3 and propagates the mask changes
2944 * throughout @cgrp's subtree, updates csses accordingly and perform
2945 * process migrations.
2947 static int cgroup_apply_control(struct cgroup *cgrp)
2951 cgroup_propagate_control(cgrp);
2953 ret = cgroup_apply_control_enable(cgrp);
2958 * At this point, cgroup_e_css() results reflect the new csses
2959 * making the following cgroup_update_dfl_csses() properly update
2960 * css associations of all tasks in the subtree.
2962 ret = cgroup_update_dfl_csses(cgrp);
2970 * cgroup_finalize_control - finalize control mask update
2971 * @cgrp: root of the target subtree
2972 * @ret: the result of the update
2974 * Finalize control mask update. See cgroup_apply_control() for more info.
2976 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
2979 cgroup_restore_control(cgrp);
2980 cgroup_propagate_control(cgrp);
2983 cgroup_apply_control_disable(cgrp);
2986 static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
2988 u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
2990 /* if nothing is getting enabled, nothing to worry about */
2994 /* can @cgrp host any resources? */
2995 if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
2998 /* mixables don't care */
2999 if (cgroup_is_mixable(cgrp))
3002 if (domain_enable) {
3003 /* can't enable domain controllers inside a thread subtree */
3004 if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3008 * Threaded controllers can handle internal competitions
3009 * and are always allowed inside a (prospective) thread
3012 if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3017 * Controllers can't be enabled for a cgroup with tasks to avoid
3018 * child cgroups competing against tasks.
3020 if (cgroup_has_tasks(cgrp))
3026 /* change the enabled child controllers for a cgroup in the default hierarchy */
3027 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3028 char *buf, size_t nbytes,
3031 u16 enable = 0, disable = 0;
3032 struct cgroup *cgrp, *child;
3033 struct cgroup_subsys *ss;
3038 * Parse input - space separated list of subsystem names prefixed
3039 * with either + or -.
3041 buf = strstrip(buf);
3042 while ((tok = strsep(&buf, " "))) {
3045 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3046 if (!cgroup_ssid_enabled(ssid) ||
3047 strcmp(tok + 1, ss->name))
3051 enable |= 1 << ssid;
3052 disable &= ~(1 << ssid);
3053 } else if (*tok == '-') {
3054 disable |= 1 << ssid;
3055 enable &= ~(1 << ssid);
3060 } while_each_subsys_mask();
3061 if (ssid == CGROUP_SUBSYS_COUNT)
3065 cgrp = cgroup_kn_lock_live(of->kn, true);
3069 for_each_subsys(ss, ssid) {
3070 if (enable & (1 << ssid)) {
3071 if (cgrp->subtree_control & (1 << ssid)) {
3072 enable &= ~(1 << ssid);
3076 if (!(cgroup_control(cgrp) & (1 << ssid))) {
3080 } else if (disable & (1 << ssid)) {
3081 if (!(cgrp->subtree_control & (1 << ssid))) {
3082 disable &= ~(1 << ssid);
3086 /* a child has it enabled? */
3087 cgroup_for_each_live_child(child, cgrp) {
3088 if (child->subtree_control & (1 << ssid)) {
3096 if (!enable && !disable) {
3101 ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3105 /* save and update control masks and prepare csses */
3106 cgroup_save_control(cgrp);
3108 cgrp->subtree_control |= enable;
3109 cgrp->subtree_control &= ~disable;
3111 ret = cgroup_apply_control(cgrp);
3112 cgroup_finalize_control(cgrp, ret);
3116 kernfs_activate(cgrp->kn);
3118 cgroup_kn_unlock(of->kn);
3119 return ret ?: nbytes;
3123 * cgroup_enable_threaded - make @cgrp threaded
3124 * @cgrp: the target cgroup
3126 * Called when "threaded" is written to the cgroup.type interface file and
3127 * tries to make @cgrp threaded and join the parent's resource domain.
3128 * This function is never called on the root cgroup as cgroup.type doesn't
3131 static int cgroup_enable_threaded(struct cgroup *cgrp)
3133 struct cgroup *parent = cgroup_parent(cgrp);
3134 struct cgroup *dom_cgrp = parent->dom_cgrp;
3137 lockdep_assert_held(&cgroup_mutex);
3139 /* noop if already threaded */
3140 if (cgroup_is_threaded(cgrp))
3143 /* we're joining the parent's domain, ensure its validity */
3144 if (!cgroup_is_valid_domain(dom_cgrp) ||
3145 !cgroup_can_be_thread_root(dom_cgrp))
3149 * The following shouldn't cause actual migrations and should
3152 cgroup_save_control(cgrp);
3154 cgrp->dom_cgrp = dom_cgrp;
3155 ret = cgroup_apply_control(cgrp);
3157 parent->nr_threaded_children++;
3159 cgrp->dom_cgrp = cgrp;
3161 cgroup_finalize_control(cgrp, ret);
3165 static int cgroup_type_show(struct seq_file *seq, void *v)
3167 struct cgroup *cgrp = seq_css(seq)->cgroup;
3169 if (cgroup_is_threaded(cgrp))
3170 seq_puts(seq, "threaded\n");
3171 else if (!cgroup_is_valid_domain(cgrp))
3172 seq_puts(seq, "domain invalid\n");
3173 else if (cgroup_is_thread_root(cgrp))
3174 seq_puts(seq, "domain threaded\n");
3176 seq_puts(seq, "domain\n");
3181 static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3182 size_t nbytes, loff_t off)
3184 struct cgroup *cgrp;
3187 /* only switching to threaded mode is supported */
3188 if (strcmp(strstrip(buf), "threaded"))
3191 cgrp = cgroup_kn_lock_live(of->kn, false);
3195 /* threaded can only be enabled */
3196 ret = cgroup_enable_threaded(cgrp);
3198 cgroup_kn_unlock(of->kn);
3199 return ret ?: nbytes;
3202 static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3204 struct cgroup *cgrp = seq_css(seq)->cgroup;
3205 int descendants = READ_ONCE(cgrp->max_descendants);
3207 if (descendants == INT_MAX)
3208 seq_puts(seq, "max\n");
3210 seq_printf(seq, "%d\n", descendants);
3215 static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3216 char *buf, size_t nbytes, loff_t off)
3218 struct cgroup *cgrp;
3222 buf = strstrip(buf);
3223 if (!strcmp(buf, "max")) {
3224 descendants = INT_MAX;
3226 ret = kstrtoint(buf, 0, &descendants);
3231 if (descendants < 0)
3234 cgrp = cgroup_kn_lock_live(of->kn, false);
3238 cgrp->max_descendants = descendants;
3240 cgroup_kn_unlock(of->kn);
3245 static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3247 struct cgroup *cgrp = seq_css(seq)->cgroup;
3248 int depth = READ_ONCE(cgrp->max_depth);
3250 if (depth == INT_MAX)
3251 seq_puts(seq, "max\n");
3253 seq_printf(seq, "%d\n", depth);
3258 static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3259 char *buf, size_t nbytes, loff_t off)
3261 struct cgroup *cgrp;
3265 buf = strstrip(buf);
3266 if (!strcmp(buf, "max")) {
3269 ret = kstrtoint(buf, 0, &depth);
3277 cgrp = cgroup_kn_lock_live(of->kn, false);
3281 cgrp->max_depth = depth;
3283 cgroup_kn_unlock(of->kn);
3288 static int cgroup_events_show(struct seq_file *seq, void *v)
3290 seq_printf(seq, "populated %d\n",
3291 cgroup_is_populated(seq_css(seq)->cgroup));
3295 static int cgroup_stat_show(struct seq_file *seq, void *v)
3297 struct cgroup *cgroup = seq_css(seq)->cgroup;
3299 seq_printf(seq, "nr_descendants %d\n",
3300 cgroup->nr_descendants);
3301 seq_printf(seq, "nr_dying_descendants %d\n",
3302 cgroup->nr_dying_descendants);
3307 static int cgroup_file_open(struct kernfs_open_file *of)
3309 struct cftype *cft = of->kn->priv;
3312 return cft->open(of);
3316 static void cgroup_file_release(struct kernfs_open_file *of)
3318 struct cftype *cft = of->kn->priv;
3324 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3325 size_t nbytes, loff_t off)
3327 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
3328 struct cgroup *cgrp = of->kn->parent->priv;
3329 struct cftype *cft = of->kn->priv;
3330 struct cgroup_subsys_state *css;
3334 * If namespaces are delegation boundaries, disallow writes to
3335 * files in an non-init namespace root from inside the namespace
3336 * except for the files explicitly marked delegatable -
3337 * cgroup.procs and cgroup.subtree_control.
3339 if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
3340 !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
3341 ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp)
3345 return cft->write(of, buf, nbytes, off);
3348 * kernfs guarantees that a file isn't deleted with operations in
3349 * flight, which means that the matching css is and stays alive and
3350 * doesn't need to be pinned. The RCU locking is not necessary
3351 * either. It's just for the convenience of using cgroup_css().
3354 css = cgroup_css(cgrp, cft->ss);
3357 if (cft->write_u64) {
3358 unsigned long long v;
3359 ret = kstrtoull(buf, 0, &v);
3361 ret = cft->write_u64(css, cft, v);
3362 } else if (cft->write_s64) {
3364 ret = kstrtoll(buf, 0, &v);
3366 ret = cft->write_s64(css, cft, v);
3371 return ret ?: nbytes;
3374 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3376 return seq_cft(seq)->seq_start(seq, ppos);
3379 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3381 return seq_cft(seq)->seq_next(seq, v, ppos);
3384 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3386 if (seq_cft(seq)->seq_stop)
3387 seq_cft(seq)->seq_stop(seq, v);
3390 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3392 struct cftype *cft = seq_cft(m);
3393 struct cgroup_subsys_state *css = seq_css(m);
3396 return cft->seq_show(m, arg);
3399 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3400 else if (cft->read_s64)
3401 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3407 static struct kernfs_ops cgroup_kf_single_ops = {
3408 .atomic_write_len = PAGE_SIZE,
3409 .open = cgroup_file_open,
3410 .release = cgroup_file_release,
3411 .write = cgroup_file_write,
3412 .seq_show = cgroup_seqfile_show,
3415 static struct kernfs_ops cgroup_kf_ops = {
3416 .atomic_write_len = PAGE_SIZE,
3417 .open = cgroup_file_open,
3418 .release = cgroup_file_release,
3419 .write = cgroup_file_write,
3420 .seq_start = cgroup_seqfile_start,
3421 .seq_next = cgroup_seqfile_next,
3422 .seq_stop = cgroup_seqfile_stop,
3423 .seq_show = cgroup_seqfile_show,
3426 /* set uid and gid of cgroup dirs and files to that of the creator */
3427 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3429 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3430 .ia_uid = current_fsuid(),
3431 .ia_gid = current_fsgid(), };
3433 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3434 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3437 return kernfs_setattr(kn, &iattr);
3440 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3443 char name[CGROUP_FILE_NAME_MAX];
3444 struct kernfs_node *kn;
3445 struct lock_class_key *key = NULL;
3448 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3449 key = &cft->lockdep_key;
3451 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3452 cgroup_file_mode(cft), 0, cft->kf_ops, cft,
3457 ret = cgroup_kn_set_ugid(kn);
3463 if (cft->file_offset) {
3464 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3466 spin_lock_irq(&cgroup_file_kn_lock);
3468 spin_unlock_irq(&cgroup_file_kn_lock);
3475 * cgroup_addrm_files - add or remove files to a cgroup directory
3476 * @css: the target css
3477 * @cgrp: the target cgroup (usually css->cgroup)
3478 * @cfts: array of cftypes to be added
3479 * @is_add: whether to add or remove
3481 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3482 * For removals, this function never fails.
3484 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3485 struct cgroup *cgrp, struct cftype cfts[],
3488 struct cftype *cft, *cft_end = NULL;
3491 lockdep_assert_held(&cgroup_mutex);
3494 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3495 /* does cft->flags tell us to skip this file on @cgrp? */
3496 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3498 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3500 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3502 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3506 ret = cgroup_add_file(css, cgrp, cft);
3508 pr_warn("%s: failed to add %s, err=%d\n",
3509 __func__, cft->name, ret);
3515 cgroup_rm_file(cgrp, cft);
3521 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3523 struct cgroup_subsys *ss = cfts[0].ss;
3524 struct cgroup *root = &ss->root->cgrp;
3525 struct cgroup_subsys_state *css;
3528 lockdep_assert_held(&cgroup_mutex);
3530 /* add/rm files for all cgroups created before */
3531 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3532 struct cgroup *cgrp = css->cgroup;
3534 if (!(css->flags & CSS_VISIBLE))
3537 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3543 kernfs_activate(root->kn);
3547 static void cgroup_exit_cftypes(struct cftype *cfts)
3551 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3552 /* free copy for custom atomic_write_len, see init_cftypes() */
3553 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3558 /* revert flags set by cgroup core while adding @cfts */
3559 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
3563 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3567 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3568 struct kernfs_ops *kf_ops;
3570 WARN_ON(cft->ss || cft->kf_ops);
3573 kf_ops = &cgroup_kf_ops;
3575 kf_ops = &cgroup_kf_single_ops;
3578 * Ugh... if @cft wants a custom max_write_len, we need to
3579 * make a copy of kf_ops to set its atomic_write_len.
3581 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3582 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3584 cgroup_exit_cftypes(cfts);
3587 kf_ops->atomic_write_len = cft->max_write_len;
3590 cft->kf_ops = kf_ops;
3597 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3599 lockdep_assert_held(&cgroup_mutex);
3601 if (!cfts || !cfts[0].ss)
3604 list_del(&cfts->node);
3605 cgroup_apply_cftypes(cfts, false);
3606 cgroup_exit_cftypes(cfts);
3611 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3612 * @cfts: zero-length name terminated array of cftypes
3614 * Unregister @cfts. Files described by @cfts are removed from all
3615 * existing cgroups and all future cgroups won't have them either. This
3616 * function can be called anytime whether @cfts' subsys is attached or not.
3618 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3621 int cgroup_rm_cftypes(struct cftype *cfts)
3625 mutex_lock(&cgroup_mutex);
3626 ret = cgroup_rm_cftypes_locked(cfts);
3627 mutex_unlock(&cgroup_mutex);
3632 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3633 * @ss: target cgroup subsystem
3634 * @cfts: zero-length name terminated array of cftypes
3636 * Register @cfts to @ss. Files described by @cfts are created for all
3637 * existing cgroups to which @ss is attached and all future cgroups will
3638 * have them too. This function can be called anytime whether @ss is
3641 * Returns 0 on successful registration, -errno on failure. Note that this
3642 * function currently returns 0 as long as @cfts registration is successful
3643 * even if some file creation attempts on existing cgroups fail.
3645 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3649 if (!cgroup_ssid_enabled(ss->id))
3652 if (!cfts || cfts[0].name[0] == '\0')
3655 ret = cgroup_init_cftypes(ss, cfts);
3659 mutex_lock(&cgroup_mutex);
3661 list_add_tail(&cfts->node, &ss->cfts);
3662 ret = cgroup_apply_cftypes(cfts, true);
3664 cgroup_rm_cftypes_locked(cfts);
3666 mutex_unlock(&cgroup_mutex);
3671 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3672 * @ss: target cgroup subsystem
3673 * @cfts: zero-length name terminated array of cftypes
3675 * Similar to cgroup_add_cftypes() but the added files are only used for
3676 * the default hierarchy.
3678 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3682 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3683 cft->flags |= __CFTYPE_ONLY_ON_DFL;
3684 return cgroup_add_cftypes(ss, cfts);
3688 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3689 * @ss: target cgroup subsystem
3690 * @cfts: zero-length name terminated array of cftypes
3692 * Similar to cgroup_add_cftypes() but the added files are only used for
3693 * the legacy hierarchies.
3695 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3699 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3700 cft->flags |= __CFTYPE_NOT_ON_DFL;
3701 return cgroup_add_cftypes(ss, cfts);
3705 * cgroup_file_notify - generate a file modified event for a cgroup_file
3706 * @cfile: target cgroup_file
3708 * @cfile must have been obtained by setting cftype->file_offset.
3710 void cgroup_file_notify(struct cgroup_file *cfile)
3712 unsigned long flags;
3714 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
3716 kernfs_notify(cfile->kn);
3717 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
3721 * css_next_child - find the next child of a given css
3722 * @pos: the current position (%NULL to initiate traversal)
3723 * @parent: css whose children to walk
3725 * This function returns the next child of @parent and should be called
3726 * under either cgroup_mutex or RCU read lock. The only requirement is
3727 * that @parent and @pos are accessible. The next sibling is guaranteed to
3728 * be returned regardless of their states.
3730 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3731 * css which finished ->css_online() is guaranteed to be visible in the
3732 * future iterations and will stay visible until the last reference is put.
3733 * A css which hasn't finished ->css_online() or already finished
3734 * ->css_offline() may show up during traversal. It's each subsystem's
3735 * responsibility to synchronize against on/offlining.
3737 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
3738 struct cgroup_subsys_state *parent)
3740 struct cgroup_subsys_state *next;
3742 cgroup_assert_mutex_or_rcu_locked();
3745 * @pos could already have been unlinked from the sibling list.
3746 * Once a cgroup is removed, its ->sibling.next is no longer
3747 * updated when its next sibling changes. CSS_RELEASED is set when
3748 * @pos is taken off list, at which time its next pointer is valid,
3749 * and, as releases are serialized, the one pointed to by the next
3750 * pointer is guaranteed to not have started release yet. This
3751 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3752 * critical section, the one pointed to by its next pointer is
3753 * guaranteed to not have finished its RCU grace period even if we
3754 * have dropped rcu_read_lock() inbetween iterations.
3756 * If @pos has CSS_RELEASED set, its next pointer can't be
3757 * dereferenced; however, as each css is given a monotonically
3758 * increasing unique serial number and always appended to the
3759 * sibling list, the next one can be found by walking the parent's
3760 * children until the first css with higher serial number than
3761 * @pos's. While this path can be slower, it happens iff iteration
3762 * races against release and the race window is very small.
3765 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
3766 } else if (likely(!(pos->flags & CSS_RELEASED))) {
3767 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
3769 list_for_each_entry_rcu(next, &parent->children, sibling)
3770 if (next->serial_nr > pos->serial_nr)
3775 * @next, if not pointing to the head, can be dereferenced and is
3778 if (&next->sibling != &parent->children)
3784 * css_next_descendant_pre - find the next descendant for pre-order walk
3785 * @pos: the current position (%NULL to initiate traversal)
3786 * @root: css whose descendants to walk
3788 * To be used by css_for_each_descendant_pre(). Find the next descendant
3789 * to visit for pre-order traversal of @root's descendants. @root is
3790 * included in the iteration and the first node to be visited.
3792 * While this function requires cgroup_mutex or RCU read locking, it
3793 * doesn't require the whole traversal to be contained in a single critical
3794 * section. This function will return the correct next descendant as long
3795 * as both @pos and @root are accessible and @pos is a descendant of @root.
3797 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3798 * css which finished ->css_online() is guaranteed to be visible in the
3799 * future iterations and will stay visible until the last reference is put.
3800 * A css which hasn't finished ->css_online() or already finished
3801 * ->css_offline() may show up during traversal. It's each subsystem's
3802 * responsibility to synchronize against on/offlining.
3804 struct cgroup_subsys_state *
3805 css_next_descendant_pre(struct cgroup_subsys_state *pos,
3806 struct cgroup_subsys_state *root)
3808 struct cgroup_subsys_state *next;
3810 cgroup_assert_mutex_or_rcu_locked();
3812 /* if first iteration, visit @root */
3816 /* visit the first child if exists */
3817 next = css_next_child(NULL, pos);
3821 /* no child, visit my or the closest ancestor's next sibling */
3822 while (pos != root) {
3823 next = css_next_child(pos, pos->parent);
3833 * css_rightmost_descendant - return the rightmost descendant of a css
3834 * @pos: css of interest
3836 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3837 * is returned. This can be used during pre-order traversal to skip
3840 * While this function requires cgroup_mutex or RCU read locking, it
3841 * doesn't require the whole traversal to be contained in a single critical
3842 * section. This function will return the correct rightmost descendant as
3843 * long as @pos is accessible.
3845 struct cgroup_subsys_state *
3846 css_rightmost_descendant(struct cgroup_subsys_state *pos)
3848 struct cgroup_subsys_state *last, *tmp;
3850 cgroup_assert_mutex_or_rcu_locked();
3854 /* ->prev isn't RCU safe, walk ->next till the end */
3856 css_for_each_child(tmp, last)
3863 static struct cgroup_subsys_state *
3864 css_leftmost_descendant(struct cgroup_subsys_state *pos)
3866 struct cgroup_subsys_state *last;
3870 pos = css_next_child(NULL, pos);
3877 * css_next_descendant_post - find the next descendant for post-order walk
3878 * @pos: the current position (%NULL to initiate traversal)
3879 * @root: css whose descendants to walk
3881 * To be used by css_for_each_descendant_post(). Find the next descendant
3882 * to visit for post-order traversal of @root's descendants. @root is
3883 * included in the iteration and the last node to be visited.
3885 * While this function requires cgroup_mutex or RCU read locking, it
3886 * doesn't require the whole traversal to be contained in a single critical
3887 * section. This function will return the correct next descendant as long
3888 * as both @pos and @cgroup are accessible and @pos is a descendant of
3891 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3892 * css which finished ->css_online() is guaranteed to be visible in the
3893 * future iterations and will stay visible until the last reference is put.
3894 * A css which hasn't finished ->css_online() or already finished
3895 * ->css_offline() may show up during traversal. It's each subsystem's
3896 * responsibility to synchronize against on/offlining.
3898 struct cgroup_subsys_state *
3899 css_next_descendant_post(struct cgroup_subsys_state *pos,
3900 struct cgroup_subsys_state *root)
3902 struct cgroup_subsys_state *next;
3904 cgroup_assert_mutex_or_rcu_locked();
3906 /* if first iteration, visit leftmost descendant which may be @root */
3908 return css_leftmost_descendant(root);
3910 /* if we visited @root, we're done */
3914 /* if there's an unvisited sibling, visit its leftmost descendant */
3915 next = css_next_child(pos, pos->parent);
3917 return css_leftmost_descendant(next);
3919 /* no sibling left, visit parent */
3924 * css_has_online_children - does a css have online children
3925 * @css: the target css
3927 * Returns %true if @css has any online children; otherwise, %false. This
3928 * function can be called from any context but the caller is responsible
3929 * for synchronizing against on/offlining as necessary.
3931 bool css_has_online_children(struct cgroup_subsys_state *css)
3933 struct cgroup_subsys_state *child;
3937 css_for_each_child(child, css) {
3938 if (child->flags & CSS_ONLINE) {
3947 static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
3949 struct list_head *l;
3950 struct cgrp_cset_link *link;
3951 struct css_set *cset;
3953 lockdep_assert_held(&css_set_lock);
3955 /* find the next threaded cset */
3956 if (it->tcset_pos) {
3957 l = it->tcset_pos->next;
3959 if (l != it->tcset_head) {
3961 return container_of(l, struct css_set,
3962 threaded_csets_node);
3965 it->tcset_pos = NULL;
3968 /* find the next cset */
3971 if (l == it->cset_head) {
3972 it->cset_pos = NULL;
3977 cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
3979 link = list_entry(l, struct cgrp_cset_link, cset_link);
3985 /* initialize threaded css_set walking */
3986 if (it->flags & CSS_TASK_ITER_THREADED) {
3988 put_css_set_locked(it->cur_dcset);
3989 it->cur_dcset = cset;
3992 it->tcset_head = &cset->threaded_csets;
3993 it->tcset_pos = &cset->threaded_csets;
4000 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
4001 * @it: the iterator to advance
4003 * Advance @it to the next css_set to walk.
4005 static void css_task_iter_advance_css_set(struct css_task_iter *it)
4007 struct css_set *cset;
4009 lockdep_assert_held(&css_set_lock);
4011 /* Advance to the next non-empty css_set */
4013 cset = css_task_iter_next_css_set(it);
4015 it->task_pos = NULL;
4018 } while (!css_set_populated(cset));
4020 if (!list_empty(&cset->tasks))
4021 it->task_pos = cset->tasks.next;
4023 it->task_pos = cset->mg_tasks.next;
4025 it->tasks_head = &cset->tasks;
4026 it->mg_tasks_head = &cset->mg_tasks;
4029 * We don't keep css_sets locked across iteration steps and thus
4030 * need to take steps to ensure that iteration can be resumed after
4031 * the lock is re-acquired. Iteration is performed at two levels -
4032 * css_sets and tasks in them.
4034 * Once created, a css_set never leaves its cgroup lists, so a
4035 * pinned css_set is guaranteed to stay put and we can resume
4036 * iteration afterwards.
4038 * Tasks may leave @cset across iteration steps. This is resolved
4039 * by registering each iterator with the css_set currently being
4040 * walked and making css_set_move_task() advance iterators whose
4041 * next task is leaving.
4044 list_del(&it->iters_node);
4045 put_css_set_locked(it->cur_cset);
4048 it->cur_cset = cset;
4049 list_add(&it->iters_node, &cset->task_iters);
4052 static void css_task_iter_advance(struct css_task_iter *it)
4054 struct list_head *l = it->task_pos;
4056 lockdep_assert_held(&css_set_lock);
4061 * Advance iterator to find next entry. cset->tasks is consumed
4062 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
4067 if (l == it->tasks_head)
4068 l = it->mg_tasks_head->next;
4070 if (l == it->mg_tasks_head)
4071 css_task_iter_advance_css_set(it);
4075 /* if PROCS, skip over tasks which aren't group leaders */
4076 if ((it->flags & CSS_TASK_ITER_PROCS) && it->task_pos &&
4077 !thread_group_leader(list_entry(it->task_pos, struct task_struct,
4083 * css_task_iter_start - initiate task iteration
4084 * @css: the css to walk tasks of
4085 * @flags: CSS_TASK_ITER_* flags
4086 * @it: the task iterator to use
4088 * Initiate iteration through the tasks of @css. The caller can call
4089 * css_task_iter_next() to walk through the tasks until the function
4090 * returns NULL. On completion of iteration, css_task_iter_end() must be
4093 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
4094 struct css_task_iter *it)
4096 /* no one should try to iterate before mounting cgroups */
4097 WARN_ON_ONCE(!use_task_css_set_links);
4099 memset(it, 0, sizeof(*it));
4101 spin_lock_irq(&css_set_lock);
4107 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4109 it->cset_pos = &css->cgroup->cset_links;
4111 it->cset_head = it->cset_pos;
4113 css_task_iter_advance_css_set(it);
4115 spin_unlock_irq(&css_set_lock);
4119 * css_task_iter_next - return the next task for the iterator
4120 * @it: the task iterator being iterated
4122 * The "next" function for task iteration. @it should have been
4123 * initialized via css_task_iter_start(). Returns NULL when the iteration
4126 struct task_struct *css_task_iter_next(struct css_task_iter *it)
4129 put_task_struct(it->cur_task);
4130 it->cur_task = NULL;
4133 spin_lock_irq(&css_set_lock);
4136 it->cur_task = list_entry(it->task_pos, struct task_struct,
4138 get_task_struct(it->cur_task);
4139 css_task_iter_advance(it);
4142 spin_unlock_irq(&css_set_lock);
4144 return it->cur_task;
4148 * css_task_iter_end - finish task iteration
4149 * @it: the task iterator to finish
4151 * Finish task iteration started by css_task_iter_start().
4153 void css_task_iter_end(struct css_task_iter *it)
4156 spin_lock_irq(&css_set_lock);
4157 list_del(&it->iters_node);
4158 put_css_set_locked(it->cur_cset);
4159 spin_unlock_irq(&css_set_lock);
4163 put_css_set(it->cur_dcset);
4166 put_task_struct(it->cur_task);
4169 static void cgroup_procs_release(struct kernfs_open_file *of)
4172 css_task_iter_end(of->priv);
4177 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
4179 struct kernfs_open_file *of = s->private;
4180 struct css_task_iter *it = of->priv;
4182 return css_task_iter_next(it);
4185 static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
4186 unsigned int iter_flags)
4188 struct kernfs_open_file *of = s->private;
4189 struct cgroup *cgrp = seq_css(s)->cgroup;
4190 struct css_task_iter *it = of->priv;
4193 * When a seq_file is seeked, it's always traversed sequentially
4194 * from position 0, so we can simply keep iterating on !0 *pos.
4197 if (WARN_ON_ONCE((*pos)++))
4198 return ERR_PTR(-EINVAL);
4200 it = kzalloc(sizeof(*it), GFP_KERNEL);
4202 return ERR_PTR(-ENOMEM);
4204 css_task_iter_start(&cgrp->self, iter_flags, it);
4205 } else if (!(*pos)++) {
4206 css_task_iter_end(it);
4207 css_task_iter_start(&cgrp->self, iter_flags, it);
4210 return cgroup_procs_next(s, NULL, NULL);
4213 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
4215 struct cgroup *cgrp = seq_css(s)->cgroup;
4218 * All processes of a threaded subtree belong to the domain cgroup
4219 * of the subtree. Only threads can be distributed across the
4220 * subtree. Reject reads on cgroup.procs in the subtree proper.
4221 * They're always empty anyway.
4223 if (cgroup_is_threaded(cgrp))
4224 return ERR_PTR(-EOPNOTSUPP);
4226 return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
4227 CSS_TASK_ITER_THREADED);
4230 static int cgroup_procs_show(struct seq_file *s, void *v)
4232 seq_printf(s, "%d\n", task_pid_vnr(v));
4236 static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
4237 struct cgroup *dst_cgrp,
4238 struct super_block *sb)
4240 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
4241 struct cgroup *com_cgrp = src_cgrp;
4242 struct inode *inode;
4245 lockdep_assert_held(&cgroup_mutex);
4247 /* find the common ancestor */
4248 while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
4249 com_cgrp = cgroup_parent(com_cgrp);
4251 /* %current should be authorized to migrate to the common ancestor */
4252 inode = kernfs_get_inode(sb, com_cgrp->procs_file.kn);
4256 ret = inode_permission(inode, MAY_WRITE);
4262 * If namespaces are delegation boundaries, %current must be able
4263 * to see both source and destination cgroups from its namespace.
4265 if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
4266 (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
4267 !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
4273 static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
4274 char *buf, size_t nbytes, loff_t off)
4276 struct cgroup *src_cgrp, *dst_cgrp;
4277 struct task_struct *task;
4280 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4284 task = cgroup_procs_write_start(buf, true);
4285 ret = PTR_ERR_OR_ZERO(task);
4289 /* find the source cgroup */
4290 spin_lock_irq(&css_set_lock);
4291 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4292 spin_unlock_irq(&css_set_lock);
4294 ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp,
4295 of->file->f_path.dentry->d_sb);
4299 ret = cgroup_attach_task(dst_cgrp, task, true);
4302 cgroup_procs_write_finish(task);
4304 cgroup_kn_unlock(of->kn);
4306 return ret ?: nbytes;
4309 static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
4311 return __cgroup_procs_start(s, pos, 0);
4314 static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
4315 char *buf, size_t nbytes, loff_t off)
4317 struct cgroup *src_cgrp, *dst_cgrp;
4318 struct task_struct *task;
4321 buf = strstrip(buf);
4323 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4327 task = cgroup_procs_write_start(buf, false);
4328 ret = PTR_ERR_OR_ZERO(task);
4332 /* find the source cgroup */
4333 spin_lock_irq(&css_set_lock);
4334 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4335 spin_unlock_irq(&css_set_lock);
4337 /* thread migrations follow the cgroup.procs delegation rule */
4338 ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp,
4339 of->file->f_path.dentry->d_sb);
4343 /* and must be contained in the same domain */
4345 if (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp)
4348 ret = cgroup_attach_task(dst_cgrp, task, false);
4351 cgroup_procs_write_finish(task);
4353 cgroup_kn_unlock(of->kn);
4355 return ret ?: nbytes;
4358 /* cgroup core interface files for the default hierarchy */
4359 static struct cftype cgroup_base_files[] = {
4361 .name = "cgroup.type",
4362 .flags = CFTYPE_NOT_ON_ROOT,
4363 .seq_show = cgroup_type_show,
4364 .write = cgroup_type_write,
4367 .name = "cgroup.procs",
4368 .flags = CFTYPE_NS_DELEGATABLE,
4369 .file_offset = offsetof(struct cgroup, procs_file),
4370 .release = cgroup_procs_release,
4371 .seq_start = cgroup_procs_start,
4372 .seq_next = cgroup_procs_next,
4373 .seq_show = cgroup_procs_show,
4374 .write = cgroup_procs_write,
4377 .name = "cgroup.threads",
4378 .release = cgroup_procs_release,
4379 .seq_start = cgroup_threads_start,
4380 .seq_next = cgroup_procs_next,
4381 .seq_show = cgroup_procs_show,
4382 .write = cgroup_threads_write,
4385 .name = "cgroup.controllers",
4386 .seq_show = cgroup_controllers_show,
4389 .name = "cgroup.subtree_control",
4390 .flags = CFTYPE_NS_DELEGATABLE,
4391 .seq_show = cgroup_subtree_control_show,
4392 .write = cgroup_subtree_control_write,
4395 .name = "cgroup.events",
4396 .flags = CFTYPE_NOT_ON_ROOT,
4397 .file_offset = offsetof(struct cgroup, events_file),
4398 .seq_show = cgroup_events_show,
4401 .name = "cgroup.max.descendants",
4402 .seq_show = cgroup_max_descendants_show,
4403 .write = cgroup_max_descendants_write,
4406 .name = "cgroup.max.depth",
4407 .seq_show = cgroup_max_depth_show,
4408 .write = cgroup_max_depth_write,
4411 .name = "cgroup.stat",
4412 .seq_show = cgroup_stat_show,
4418 * css destruction is four-stage process.
4420 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4421 * Implemented in kill_css().
4423 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4424 * and thus css_tryget_online() is guaranteed to fail, the css can be
4425 * offlined by invoking offline_css(). After offlining, the base ref is
4426 * put. Implemented in css_killed_work_fn().
4428 * 3. When the percpu_ref reaches zero, the only possible remaining
4429 * accessors are inside RCU read sections. css_release() schedules the
4432 * 4. After the grace period, the css can be freed. Implemented in
4433 * css_free_work_fn().
4435 * It is actually hairier because both step 2 and 4 require process context
4436 * and thus involve punting to css->destroy_work adding two additional
4437 * steps to the already complex sequence.
4439 static void css_free_work_fn(struct work_struct *work)
4441 struct cgroup_subsys_state *css =
4442 container_of(work, struct cgroup_subsys_state, destroy_work);
4443 struct cgroup_subsys *ss = css->ss;
4444 struct cgroup *cgrp = css->cgroup;
4446 percpu_ref_exit(&css->refcnt);
4450 struct cgroup_subsys_state *parent = css->parent;
4454 cgroup_idr_remove(&ss->css_idr, id);
4460 /* cgroup free path */
4461 atomic_dec(&cgrp->root->nr_cgrps);
4462 cgroup1_pidlist_destroy_all(cgrp);
4463 cancel_work_sync(&cgrp->release_agent_work);
4465 if (cgroup_parent(cgrp)) {
4467 * We get a ref to the parent, and put the ref when
4468 * this cgroup is being freed, so it's guaranteed
4469 * that the parent won't be destroyed before its
4472 cgroup_put(cgroup_parent(cgrp));
4473 kernfs_put(cgrp->kn);
4477 * This is root cgroup's refcnt reaching zero,
4478 * which indicates that the root should be
4481 cgroup_destroy_root(cgrp->root);
4486 static void css_free_rcu_fn(struct rcu_head *rcu_head)
4488 struct cgroup_subsys_state *css =
4489 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
4491 INIT_WORK(&css->destroy_work, css_free_work_fn);
4492 queue_work(cgroup_destroy_wq, &css->destroy_work);
4495 static void css_release_work_fn(struct work_struct *work)
4497 struct cgroup_subsys_state *css =
4498 container_of(work, struct cgroup_subsys_state, destroy_work);
4499 struct cgroup_subsys *ss = css->ss;
4500 struct cgroup *cgrp = css->cgroup;
4502 mutex_lock(&cgroup_mutex);
4504 css->flags |= CSS_RELEASED;
4505 list_del_rcu(&css->sibling);
4508 /* css release path */
4509 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
4510 if (ss->css_released)
4511 ss->css_released(css);
4513 struct cgroup *tcgrp;
4515 /* cgroup release path */
4516 trace_cgroup_release(cgrp);
4518 for (tcgrp = cgroup_parent(cgrp); tcgrp;
4519 tcgrp = cgroup_parent(tcgrp))
4520 tcgrp->nr_dying_descendants--;
4522 cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
4526 * There are two control paths which try to determine
4527 * cgroup from dentry without going through kernfs -
4528 * cgroupstats_build() and css_tryget_online_from_dir().
4529 * Those are supported by RCU protecting clearing of
4530 * cgrp->kn->priv backpointer.
4533 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
4536 cgroup_bpf_put(cgrp);
4539 mutex_unlock(&cgroup_mutex);
4541 call_rcu(&css->rcu_head, css_free_rcu_fn);
4544 static void css_release(struct percpu_ref *ref)
4546 struct cgroup_subsys_state *css =
4547 container_of(ref, struct cgroup_subsys_state, refcnt);
4549 INIT_WORK(&css->destroy_work, css_release_work_fn);
4550 queue_work(cgroup_destroy_wq, &css->destroy_work);
4553 static void init_and_link_css(struct cgroup_subsys_state *css,
4554 struct cgroup_subsys *ss, struct cgroup *cgrp)
4556 lockdep_assert_held(&cgroup_mutex);
4558 cgroup_get_live(cgrp);
4560 memset(css, 0, sizeof(*css));
4564 INIT_LIST_HEAD(&css->sibling);
4565 INIT_LIST_HEAD(&css->children);
4566 css->serial_nr = css_serial_nr_next++;
4567 atomic_set(&css->online_cnt, 0);
4569 if (cgroup_parent(cgrp)) {
4570 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
4571 css_get(css->parent);
4574 BUG_ON(cgroup_css(cgrp, ss));
4577 /* invoke ->css_online() on a new CSS and mark it online if successful */
4578 static int online_css(struct cgroup_subsys_state *css)
4580 struct cgroup_subsys *ss = css->ss;
4583 lockdep_assert_held(&cgroup_mutex);
4586 ret = ss->css_online(css);
4588 css->flags |= CSS_ONLINE;
4589 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
4591 atomic_inc(&css->online_cnt);
4593 atomic_inc(&css->parent->online_cnt);
4598 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4599 static void offline_css(struct cgroup_subsys_state *css)
4601 struct cgroup_subsys *ss = css->ss;
4603 lockdep_assert_held(&cgroup_mutex);
4605 if (!(css->flags & CSS_ONLINE))
4608 if (ss->css_offline)
4609 ss->css_offline(css);
4611 css->flags &= ~CSS_ONLINE;
4612 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
4614 wake_up_all(&css->cgroup->offline_waitq);
4618 * css_create - create a cgroup_subsys_state
4619 * @cgrp: the cgroup new css will be associated with
4620 * @ss: the subsys of new css
4622 * Create a new css associated with @cgrp - @ss pair. On success, the new
4623 * css is online and installed in @cgrp. This function doesn't create the
4624 * interface files. Returns 0 on success, -errno on failure.
4626 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
4627 struct cgroup_subsys *ss)
4629 struct cgroup *parent = cgroup_parent(cgrp);
4630 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
4631 struct cgroup_subsys_state *css;
4634 lockdep_assert_held(&cgroup_mutex);
4636 css = ss->css_alloc(parent_css);
4638 css = ERR_PTR(-ENOMEM);
4642 init_and_link_css(css, ss, cgrp);
4644 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
4648 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
4653 /* @css is ready to be brought online now, make it visible */
4654 list_add_tail_rcu(&css->sibling, &parent_css->children);
4655 cgroup_idr_replace(&ss->css_idr, css, css->id);
4657 err = online_css(css);
4661 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4662 cgroup_parent(parent)) {
4663 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4664 current->comm, current->pid, ss->name);
4665 if (!strcmp(ss->name, "memory"))
4666 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4667 ss->warned_broken_hierarchy = true;
4673 list_del_rcu(&css->sibling);
4675 call_rcu(&css->rcu_head, css_free_rcu_fn);
4676 return ERR_PTR(err);
4680 * The returned cgroup is fully initialized including its control mask, but
4681 * it isn't associated with its kernfs_node and doesn't have the control
4684 static struct cgroup *cgroup_create(struct cgroup *parent)
4686 struct cgroup_root *root = parent->root;
4687 struct cgroup *cgrp, *tcgrp;
4688 int level = parent->level + 1;
4691 /* allocate the cgroup and its ID, 0 is reserved for the root */
4692 cgrp = kzalloc(sizeof(*cgrp) +
4693 sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL);
4695 return ERR_PTR(-ENOMEM);
4697 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
4702 * Temporarily set the pointer to NULL, so idr_find() won't return
4703 * a half-baked cgroup.
4705 cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL);
4708 goto out_cancel_ref;
4711 init_cgroup_housekeeping(cgrp);
4713 cgrp->self.parent = &parent->self;
4715 cgrp->level = level;
4717 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
4718 cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
4721 tcgrp->nr_descendants++;
4724 if (notify_on_release(parent))
4725 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4727 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4728 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4730 cgrp->self.serial_nr = css_serial_nr_next++;
4732 /* allocation complete, commit to creation */
4733 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
4734 atomic_inc(&root->nr_cgrps);
4735 cgroup_get_live(parent);
4738 * @cgrp is now fully operational. If something fails after this
4739 * point, it'll be released via the normal destruction path.
4741 cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
4744 * On the default hierarchy, a child doesn't automatically inherit
4745 * subtree_control from the parent. Each is configured manually.
4747 if (!cgroup_on_dfl(cgrp))
4748 cgrp->subtree_control = cgroup_control(cgrp);
4751 cgroup_bpf_inherit(cgrp, parent);
4753 cgroup_propagate_control(cgrp);
4758 percpu_ref_exit(&cgrp->self.refcnt);
4761 return ERR_PTR(ret);
4764 static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
4766 struct cgroup *cgroup;
4770 lockdep_assert_held(&cgroup_mutex);
4772 for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
4773 if (cgroup->nr_descendants >= cgroup->max_descendants)
4776 if (level > cgroup->max_depth)
4787 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
4789 struct cgroup *parent, *cgrp;
4790 struct kernfs_node *kn;
4793 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
4794 if (strchr(name, '\n'))
4797 parent = cgroup_kn_lock_live(parent_kn, false);
4801 if (!cgroup_check_hierarchy_limits(parent)) {
4806 cgrp = cgroup_create(parent);
4808 ret = PTR_ERR(cgrp);
4812 /* create the directory */
4813 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
4821 * This extra ref will be put in cgroup_free_fn() and guarantees
4822 * that @cgrp->kn is always accessible.
4826 ret = cgroup_kn_set_ugid(kn);
4830 ret = css_populate_dir(&cgrp->self);
4834 ret = cgroup_apply_control_enable(cgrp);
4838 trace_cgroup_mkdir(cgrp);
4840 /* let's create and online css's */
4841 kernfs_activate(kn);
4847 cgroup_destroy_locked(cgrp);
4849 cgroup_kn_unlock(parent_kn);
4854 * This is called when the refcnt of a css is confirmed to be killed.
4855 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4856 * initate destruction and put the css ref from kill_css().
4858 static void css_killed_work_fn(struct work_struct *work)
4860 struct cgroup_subsys_state *css =
4861 container_of(work, struct cgroup_subsys_state, destroy_work);
4863 mutex_lock(&cgroup_mutex);
4868 /* @css can't go away while we're holding cgroup_mutex */
4870 } while (css && atomic_dec_and_test(&css->online_cnt));
4872 mutex_unlock(&cgroup_mutex);
4875 /* css kill confirmation processing requires process context, bounce */
4876 static void css_killed_ref_fn(struct percpu_ref *ref)
4878 struct cgroup_subsys_state *css =
4879 container_of(ref, struct cgroup_subsys_state, refcnt);
4881 if (atomic_dec_and_test(&css->online_cnt)) {
4882 INIT_WORK(&css->destroy_work, css_killed_work_fn);
4883 queue_work(cgroup_destroy_wq, &css->destroy_work);
4888 * kill_css - destroy a css
4889 * @css: css to destroy
4891 * This function initiates destruction of @css by removing cgroup interface
4892 * files and putting its base reference. ->css_offline() will be invoked
4893 * asynchronously once css_tryget_online() is guaranteed to fail and when
4894 * the reference count reaches zero, @css will be released.
4896 static void kill_css(struct cgroup_subsys_state *css)
4898 lockdep_assert_held(&cgroup_mutex);
4900 if (css->flags & CSS_DYING)
4903 css->flags |= CSS_DYING;
4906 * This must happen before css is disassociated with its cgroup.
4907 * See seq_css() for details.
4912 * Killing would put the base ref, but we need to keep it alive
4913 * until after ->css_offline().
4918 * cgroup core guarantees that, by the time ->css_offline() is
4919 * invoked, no new css reference will be given out via
4920 * css_tryget_online(). We can't simply call percpu_ref_kill() and
4921 * proceed to offlining css's because percpu_ref_kill() doesn't
4922 * guarantee that the ref is seen as killed on all CPUs on return.
4924 * Use percpu_ref_kill_and_confirm() to get notifications as each
4925 * css is confirmed to be seen as killed on all CPUs.
4927 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
4931 * cgroup_destroy_locked - the first stage of cgroup destruction
4932 * @cgrp: cgroup to be destroyed
4934 * css's make use of percpu refcnts whose killing latency shouldn't be
4935 * exposed to userland and are RCU protected. Also, cgroup core needs to
4936 * guarantee that css_tryget_online() won't succeed by the time
4937 * ->css_offline() is invoked. To satisfy all the requirements,
4938 * destruction is implemented in the following two steps.
4940 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4941 * userland visible parts and start killing the percpu refcnts of
4942 * css's. Set up so that the next stage will be kicked off once all
4943 * the percpu refcnts are confirmed to be killed.
4945 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4946 * rest of destruction. Once all cgroup references are gone, the
4947 * cgroup is RCU-freed.
4949 * This function implements s1. After this step, @cgrp is gone as far as
4950 * the userland is concerned and a new cgroup with the same name may be
4951 * created. As cgroup doesn't care about the names internally, this
4952 * doesn't cause any problem.
4954 static int cgroup_destroy_locked(struct cgroup *cgrp)
4955 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4957 struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
4958 struct cgroup_subsys_state *css;
4959 struct cgrp_cset_link *link;
4962 lockdep_assert_held(&cgroup_mutex);
4965 * Only migration can raise populated from zero and we're already
4966 * holding cgroup_mutex.
4968 if (cgroup_is_populated(cgrp))
4972 * Make sure there's no live children. We can't test emptiness of
4973 * ->self.children as dead children linger on it while being
4974 * drained; otherwise, "rmdir parent/child parent" may fail.
4976 if (css_has_online_children(&cgrp->self))
4980 * Mark @cgrp and the associated csets dead. The former prevents
4981 * further task migration and child creation by disabling
4982 * cgroup_lock_live_group(). The latter makes the csets ignored by
4983 * the migration path.
4985 cgrp->self.flags &= ~CSS_ONLINE;
4987 spin_lock_irq(&css_set_lock);
4988 list_for_each_entry(link, &cgrp->cset_links, cset_link)
4989 link->cset->dead = true;
4990 spin_unlock_irq(&css_set_lock);
4992 /* initiate massacre of all css's */
4993 for_each_css(css, ssid, cgrp)
4997 * Remove @cgrp directory along with the base files. @cgrp has an
4998 * extra ref on its kn.
5000 kernfs_remove(cgrp->kn);
5002 if (parent && cgroup_is_threaded(cgrp))
5003 parent->nr_threaded_children--;
5005 for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5006 tcgrp->nr_descendants--;
5007 tcgrp->nr_dying_descendants++;
5010 cgroup1_check_for_release(parent);
5012 /* put the base reference */
5013 percpu_ref_kill(&cgrp->self.refcnt);
5018 int cgroup_rmdir(struct kernfs_node *kn)
5020 struct cgroup *cgrp;
5023 cgrp = cgroup_kn_lock_live(kn, false);
5027 ret = cgroup_destroy_locked(cgrp);
5030 trace_cgroup_rmdir(cgrp);
5032 cgroup_kn_unlock(kn);
5036 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5037 .show_options = cgroup_show_options,
5038 .remount_fs = cgroup_remount,
5039 .mkdir = cgroup_mkdir,
5040 .rmdir = cgroup_rmdir,
5041 .show_path = cgroup_show_path,
5044 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5046 struct cgroup_subsys_state *css;
5048 pr_debug("Initializing cgroup subsys %s\n", ss->name);
5050 mutex_lock(&cgroup_mutex);
5052 idr_init(&ss->css_idr);
5053 INIT_LIST_HEAD(&ss->cfts);
5055 /* Create the root cgroup state for this subsystem */
5056 ss->root = &cgrp_dfl_root;
5057 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5058 /* We don't handle early failures gracefully */
5059 BUG_ON(IS_ERR(css));
5060 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5063 * Root csses are never destroyed and we can't initialize
5064 * percpu_ref during early init. Disable refcnting.
5066 css->flags |= CSS_NO_REF;
5069 /* allocation can't be done safely during early init */
5072 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5073 BUG_ON(css->id < 0);
5076 /* Update the init_css_set to contain a subsys
5077 * pointer to this state - since the subsystem is
5078 * newly registered, all tasks and hence the
5079 * init_css_set is in the subsystem's root cgroup. */
5080 init_css_set.subsys[ss->id] = css;
5082 have_fork_callback |= (bool)ss->fork << ss->id;
5083 have_exit_callback |= (bool)ss->exit << ss->id;
5084 have_free_callback |= (bool)ss->free << ss->id;
5085 have_canfork_callback |= (bool)ss->can_fork << ss->id;
5087 /* At system boot, before all subsystems have been
5088 * registered, no tasks have been forked, so we don't
5089 * need to invoke fork callbacks here. */
5090 BUG_ON(!list_empty(&init_task.tasks));
5092 BUG_ON(online_css(css));
5094 mutex_unlock(&cgroup_mutex);
5098 * cgroup_init_early - cgroup initialization at system boot
5100 * Initialize cgroups at system boot, and initialize any
5101 * subsystems that request early init.
5103 int __init cgroup_init_early(void)
5105 static struct cgroup_sb_opts __initdata opts;
5106 struct cgroup_subsys *ss;
5109 init_cgroup_root(&cgrp_dfl_root, &opts);
5110 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5112 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5114 for_each_subsys(ss, i) {
5115 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5116 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5117 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5119 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5120 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5123 ss->name = cgroup_subsys_name[i];
5124 if (!ss->legacy_name)
5125 ss->legacy_name = cgroup_subsys_name[i];
5128 cgroup_init_subsys(ss, true);
5133 static u16 cgroup_disable_mask __initdata;
5136 * cgroup_init - cgroup initialization
5138 * Register cgroup filesystem and /proc file, and initialize
5139 * any subsystems that didn't request early init.
5141 int __init cgroup_init(void)
5143 struct cgroup_subsys *ss;
5146 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5147 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
5148 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
5149 BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
5152 * The latency of the synchronize_sched() is too high for cgroups,
5153 * avoid it at the cost of forcing all readers into the slow path.
5155 rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5157 get_user_ns(init_cgroup_ns.user_ns);
5159 mutex_lock(&cgroup_mutex);
5162 * Add init_css_set to the hash table so that dfl_root can link to
5165 hash_add(css_set_table, &init_css_set.hlist,
5166 css_set_hash(init_css_set.subsys));
5168 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0, 0));
5170 mutex_unlock(&cgroup_mutex);
5172 for_each_subsys(ss, ssid) {
5173 if (ss->early_init) {
5174 struct cgroup_subsys_state *css =
5175 init_css_set.subsys[ss->id];
5177 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5179 BUG_ON(css->id < 0);
5181 cgroup_init_subsys(ss, false);
5184 list_add_tail(&init_css_set.e_cset_node[ssid],
5185 &cgrp_dfl_root.cgrp.e_csets[ssid]);
5188 * Setting dfl_root subsys_mask needs to consider the
5189 * disabled flag and cftype registration needs kmalloc,
5190 * both of which aren't available during early_init.
5192 if (cgroup_disable_mask & (1 << ssid)) {
5193 static_branch_disable(cgroup_subsys_enabled_key[ssid]);
5194 printk(KERN_INFO "Disabling %s control group subsystem\n",
5199 if (cgroup1_ssid_disabled(ssid))
5200 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5203 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5205 /* implicit controllers must be threaded too */
5206 WARN_ON(ss->implicit_on_dfl && !ss->threaded);
5208 if (ss->implicit_on_dfl)
5209 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5210 else if (!ss->dfl_cftypes)
5211 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5214 cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
5216 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5217 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5219 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5220 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5224 ss->bind(init_css_set.subsys[ssid]);
5226 mutex_lock(&cgroup_mutex);
5227 css_populate_dir(init_css_set.subsys[ssid]);
5228 mutex_unlock(&cgroup_mutex);
5231 /* init_css_set.subsys[] has been updated, re-hash */
5232 hash_del(&init_css_set.hlist);
5233 hash_add(css_set_table, &init_css_set.hlist,
5234 css_set_hash(init_css_set.subsys));
5236 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5237 WARN_ON(register_filesystem(&cgroup_fs_type));
5238 WARN_ON(register_filesystem(&cgroup2_fs_type));
5239 WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations));
5244 static int __init cgroup_wq_init(void)
5247 * There isn't much point in executing destruction path in
5248 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5249 * Use 1 for @max_active.
5251 * We would prefer to do this in cgroup_init() above, but that
5252 * is called before init_workqueues(): so leave this until after.
5254 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5255 BUG_ON(!cgroup_destroy_wq);
5258 core_initcall(cgroup_wq_init);
5260 void cgroup_path_from_kernfs_id(const union kernfs_node_id *id,
5261 char *buf, size_t buflen)
5263 struct kernfs_node *kn;
5265 kn = kernfs_get_node_by_id(cgrp_dfl_root.kf_root, id);
5268 kernfs_path(kn, buf, buflen);
5273 * proc_cgroup_show()
5274 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5275 * - Used for /proc/<pid>/cgroup.
5277 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5278 struct pid *pid, struct task_struct *tsk)
5282 struct cgroup_root *root;
5285 buf = kmalloc(PATH_MAX, GFP_KERNEL);
5289 mutex_lock(&cgroup_mutex);
5290 spin_lock_irq(&css_set_lock);
5292 for_each_root(root) {
5293 struct cgroup_subsys *ss;
5294 struct cgroup *cgrp;
5295 int ssid, count = 0;
5297 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5300 seq_printf(m, "%d:", root->hierarchy_id);
5301 if (root != &cgrp_dfl_root)
5302 for_each_subsys(ss, ssid)
5303 if (root->subsys_mask & (1 << ssid))
5304 seq_printf(m, "%s%s", count++ ? "," : "",
5306 if (strlen(root->name))
5307 seq_printf(m, "%sname=%s", count ? "," : "",
5311 cgrp = task_cgroup_from_root(tsk, root);
5314 * On traditional hierarchies, all zombie tasks show up as
5315 * belonging to the root cgroup. On the default hierarchy,
5316 * while a zombie doesn't show up in "cgroup.procs" and
5317 * thus can't be migrated, its /proc/PID/cgroup keeps
5318 * reporting the cgroup it belonged to before exiting. If
5319 * the cgroup is removed before the zombie is reaped,
5320 * " (deleted)" is appended to the cgroup path.
5322 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
5323 retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5324 current->nsproxy->cgroup_ns);
5325 if (retval >= PATH_MAX)
5326 retval = -ENAMETOOLONG;
5335 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5336 seq_puts(m, " (deleted)\n");
5343 spin_unlock_irq(&css_set_lock);
5344 mutex_unlock(&cgroup_mutex);
5351 * cgroup_fork - initialize cgroup related fields during copy_process()
5352 * @child: pointer to task_struct of forking parent process.
5354 * A task is associated with the init_css_set until cgroup_post_fork()
5355 * attaches it to the parent's css_set. Empty cg_list indicates that
5356 * @child isn't holding reference to its css_set.
5358 void cgroup_fork(struct task_struct *child)
5360 RCU_INIT_POINTER(child->cgroups, &init_css_set);
5361 INIT_LIST_HEAD(&child->cg_list);
5365 * cgroup_can_fork - called on a new task before the process is exposed
5366 * @child: the task in question.
5368 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5369 * returns an error, the fork aborts with that error code. This allows for
5370 * a cgroup subsystem to conditionally allow or deny new forks.
5372 int cgroup_can_fork(struct task_struct *child)
5374 struct cgroup_subsys *ss;
5377 do_each_subsys_mask(ss, i, have_canfork_callback) {
5378 ret = ss->can_fork(child);
5381 } while_each_subsys_mask();
5386 for_each_subsys(ss, j) {
5389 if (ss->cancel_fork)
5390 ss->cancel_fork(child);
5397 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5398 * @child: the task in question
5400 * This calls the cancel_fork() callbacks if a fork failed *after*
5401 * cgroup_can_fork() succeded.
5403 void cgroup_cancel_fork(struct task_struct *child)
5405 struct cgroup_subsys *ss;
5408 for_each_subsys(ss, i)
5409 if (ss->cancel_fork)
5410 ss->cancel_fork(child);
5414 * cgroup_post_fork - called on a new task after adding it to the task list
5415 * @child: the task in question
5417 * Adds the task to the list running through its css_set if necessary and
5418 * call the subsystem fork() callbacks. Has to be after the task is
5419 * visible on the task list in case we race with the first call to
5420 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5423 void cgroup_post_fork(struct task_struct *child)
5425 struct cgroup_subsys *ss;
5429 * This may race against cgroup_enable_task_cg_lists(). As that
5430 * function sets use_task_css_set_links before grabbing
5431 * tasklist_lock and we just went through tasklist_lock to add
5432 * @child, it's guaranteed that either we see the set
5433 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5434 * @child during its iteration.
5436 * If we won the race, @child is associated with %current's
5437 * css_set. Grabbing css_set_lock guarantees both that the
5438 * association is stable, and, on completion of the parent's
5439 * migration, @child is visible in the source of migration or
5440 * already in the destination cgroup. This guarantee is necessary
5441 * when implementing operations which need to migrate all tasks of
5442 * a cgroup to another.
5444 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
5445 * will remain in init_css_set. This is safe because all tasks are
5446 * in the init_css_set before cg_links is enabled and there's no
5447 * operation which transfers all tasks out of init_css_set.
5449 if (use_task_css_set_links) {
5450 struct css_set *cset;
5452 spin_lock_irq(&css_set_lock);
5453 cset = task_css_set(current);
5454 if (list_empty(&child->cg_list)) {
5457 css_set_move_task(child, NULL, cset, false);
5459 spin_unlock_irq(&css_set_lock);
5463 * Call ss->fork(). This must happen after @child is linked on
5464 * css_set; otherwise, @child might change state between ->fork()
5465 * and addition to css_set.
5467 do_each_subsys_mask(ss, i, have_fork_callback) {
5469 } while_each_subsys_mask();
5473 * cgroup_exit - detach cgroup from exiting task
5474 * @tsk: pointer to task_struct of exiting process
5476 * Description: Detach cgroup from @tsk and release it.
5478 * Note that cgroups marked notify_on_release force every task in
5479 * them to take the global cgroup_mutex mutex when exiting.
5480 * This could impact scaling on very large systems. Be reluctant to
5481 * use notify_on_release cgroups where very high task exit scaling
5482 * is required on large systems.
5484 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5485 * call cgroup_exit() while the task is still competent to handle
5486 * notify_on_release(), then leave the task attached to the root cgroup in
5487 * each hierarchy for the remainder of its exit. No need to bother with
5488 * init_css_set refcnting. init_css_set never goes away and we can't race
5489 * with migration path - PF_EXITING is visible to migration path.
5491 void cgroup_exit(struct task_struct *tsk)
5493 struct cgroup_subsys *ss;
5494 struct css_set *cset;
5498 * Unlink from @tsk from its css_set. As migration path can't race
5499 * with us, we can check css_set and cg_list without synchronization.
5501 cset = task_css_set(tsk);
5503 if (!list_empty(&tsk->cg_list)) {
5504 spin_lock_irq(&css_set_lock);
5505 css_set_move_task(tsk, cset, NULL, false);
5507 spin_unlock_irq(&css_set_lock);
5512 /* see cgroup_post_fork() for details */
5513 do_each_subsys_mask(ss, i, have_exit_callback) {
5515 } while_each_subsys_mask();
5518 void cgroup_free(struct task_struct *task)
5520 struct css_set *cset = task_css_set(task);
5521 struct cgroup_subsys *ss;
5524 do_each_subsys_mask(ss, ssid, have_free_callback) {
5526 } while_each_subsys_mask();
5531 static int __init cgroup_disable(char *str)
5533 struct cgroup_subsys *ss;
5537 while ((token = strsep(&str, ",")) != NULL) {
5541 for_each_subsys(ss, i) {
5542 if (strcmp(token, ss->name) &&
5543 strcmp(token, ss->legacy_name))
5545 cgroup_disable_mask |= 1 << i;
5550 __setup("cgroup_disable=", cgroup_disable);
5553 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5554 * @dentry: directory dentry of interest
5555 * @ss: subsystem of interest
5557 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5558 * to get the corresponding css and return it. If such css doesn't exist
5559 * or can't be pinned, an ERR_PTR value is returned.
5561 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
5562 struct cgroup_subsys *ss)
5564 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
5565 struct file_system_type *s_type = dentry->d_sb->s_type;
5566 struct cgroup_subsys_state *css = NULL;
5567 struct cgroup *cgrp;
5569 /* is @dentry a cgroup dir? */
5570 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
5571 !kn || kernfs_type(kn) != KERNFS_DIR)
5572 return ERR_PTR(-EBADF);
5577 * This path doesn't originate from kernfs and @kn could already
5578 * have been or be removed at any point. @kn->priv is RCU
5579 * protected for this access. See css_release_work_fn() for details.
5581 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
5583 css = cgroup_css(cgrp, ss);
5585 if (!css || !css_tryget_online(css))
5586 css = ERR_PTR(-ENOENT);
5593 * css_from_id - lookup css by id
5594 * @id: the cgroup id
5595 * @ss: cgroup subsys to be looked into
5597 * Returns the css if there's valid one with @id, otherwise returns NULL.
5598 * Should be called under rcu_read_lock().
5600 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
5602 WARN_ON_ONCE(!rcu_read_lock_held());
5603 return idr_find(&ss->css_idr, id);
5607 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
5608 * @path: path on the default hierarchy
5610 * Find the cgroup at @path on the default hierarchy, increment its
5611 * reference count and return it. Returns pointer to the found cgroup on
5612 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
5613 * if @path points to a non-directory.
5615 struct cgroup *cgroup_get_from_path(const char *path)
5617 struct kernfs_node *kn;
5618 struct cgroup *cgrp;
5620 mutex_lock(&cgroup_mutex);
5622 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
5624 if (kernfs_type(kn) == KERNFS_DIR) {
5626 cgroup_get_live(cgrp);
5628 cgrp = ERR_PTR(-ENOTDIR);
5632 cgrp = ERR_PTR(-ENOENT);
5635 mutex_unlock(&cgroup_mutex);
5638 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
5641 * cgroup_get_from_fd - get a cgroup pointer from a fd
5642 * @fd: fd obtained by open(cgroup2_dir)
5644 * Find the cgroup from a fd which should be obtained
5645 * by opening a cgroup directory. Returns a pointer to the
5646 * cgroup on success. ERR_PTR is returned if the cgroup
5649 struct cgroup *cgroup_get_from_fd(int fd)
5651 struct cgroup_subsys_state *css;
5652 struct cgroup *cgrp;
5657 return ERR_PTR(-EBADF);
5659 css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
5662 return ERR_CAST(css);
5665 if (!cgroup_on_dfl(cgrp)) {
5667 return ERR_PTR(-EBADF);
5672 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
5675 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
5676 * definition in cgroup-defs.h.
5678 #ifdef CONFIG_SOCK_CGROUP_DATA
5680 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
5682 DEFINE_SPINLOCK(cgroup_sk_update_lock);
5683 static bool cgroup_sk_alloc_disabled __read_mostly;
5685 void cgroup_sk_alloc_disable(void)
5687 if (cgroup_sk_alloc_disabled)
5689 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
5690 cgroup_sk_alloc_disabled = true;
5695 #define cgroup_sk_alloc_disabled false
5699 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
5701 if (cgroup_sk_alloc_disabled)
5704 /* Socket clone path */
5707 * We might be cloning a socket which is left in an empty
5708 * cgroup and the cgroup might have already been rmdir'd.
5709 * Don't use cgroup_get_live().
5711 cgroup_get(sock_cgroup_ptr(skcd));
5718 struct css_set *cset;
5720 cset = task_css_set(current);
5721 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
5722 skcd->val = (unsigned long)cset->dfl_cgrp;
5731 void cgroup_sk_free(struct sock_cgroup_data *skcd)
5733 cgroup_put(sock_cgroup_ptr(skcd));
5736 #endif /* CONFIG_SOCK_CGROUP_DATA */
5738 #ifdef CONFIG_CGROUP_BPF
5739 int cgroup_bpf_update(struct cgroup *cgrp, struct bpf_prog *prog,
5740 enum bpf_attach_type type, bool overridable)
5742 struct cgroup *parent = cgroup_parent(cgrp);
5745 mutex_lock(&cgroup_mutex);
5746 ret = __cgroup_bpf_update(cgrp, parent, prog, type, overridable);
5747 mutex_unlock(&cgroup_mutex);
5750 #endif /* CONFIG_CGROUP_BPF */