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>
145 static DEFINE_PER_CPU(struct cgroup_cpu_stat, cgrp_dfl_root_cpu_stat);
148 * The default hierarchy, reserved for the subsystems that are otherwise
149 * unattached - it never has more than a single cgroup, and all tasks are
150 * part of that cgroup.
152 struct cgroup_root cgrp_dfl_root = { .cgrp.cpu_stat = &cgrp_dfl_root_cpu_stat };
153 EXPORT_SYMBOL_GPL(cgrp_dfl_root);
156 * The default hierarchy always exists but is hidden until mounted for the
157 * first time. This is for backward compatibility.
159 static bool cgrp_dfl_visible;
161 /* some controllers are not supported in the default hierarchy */
162 static u16 cgrp_dfl_inhibit_ss_mask;
164 /* some controllers are implicitly enabled on the default hierarchy */
165 static u16 cgrp_dfl_implicit_ss_mask;
167 /* some controllers can be threaded on the default hierarchy */
168 static u16 cgrp_dfl_threaded_ss_mask;
170 /* The list of hierarchy roots */
171 LIST_HEAD(cgroup_roots);
172 static int cgroup_root_count;
174 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
175 static DEFINE_IDR(cgroup_hierarchy_idr);
178 * Assign a monotonically increasing serial number to csses. It guarantees
179 * cgroups with bigger numbers are newer than those with smaller numbers.
180 * Also, as csses are always appended to the parent's ->children list, it
181 * guarantees that sibling csses are always sorted in the ascending serial
182 * number order on the list. Protected by cgroup_mutex.
184 static u64 css_serial_nr_next = 1;
187 * These bitmasks identify subsystems with specific features to avoid
188 * having to do iterative checks repeatedly.
190 static u16 have_fork_callback __read_mostly;
191 static u16 have_exit_callback __read_mostly;
192 static u16 have_free_callback __read_mostly;
193 static u16 have_canfork_callback __read_mostly;
195 /* cgroup namespace for init task */
196 struct cgroup_namespace init_cgroup_ns = {
197 .count = REFCOUNT_INIT(2),
198 .user_ns = &init_user_ns,
199 .ns.ops = &cgroupns_operations,
200 .ns.inum = PROC_CGROUP_INIT_INO,
201 .root_cset = &init_css_set,
204 static struct file_system_type cgroup2_fs_type;
205 static struct cftype cgroup_base_files[];
207 static int cgroup_apply_control(struct cgroup *cgrp);
208 static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
209 static void css_task_iter_advance(struct css_task_iter *it);
210 static int cgroup_destroy_locked(struct cgroup *cgrp);
211 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
212 struct cgroup_subsys *ss);
213 static void css_release(struct percpu_ref *ref);
214 static void kill_css(struct cgroup_subsys_state *css);
215 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
216 struct cgroup *cgrp, struct cftype cfts[],
220 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
221 * @ssid: subsys ID of interest
223 * cgroup_subsys_enabled() can only be used with literal subsys names which
224 * is fine for individual subsystems but unsuitable for cgroup core. This
225 * is slower static_key_enabled() based test indexed by @ssid.
227 bool cgroup_ssid_enabled(int ssid)
229 if (CGROUP_SUBSYS_COUNT == 0)
232 return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
236 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
237 * @cgrp: the cgroup of interest
239 * The default hierarchy is the v2 interface of cgroup and this function
240 * can be used to test whether a cgroup is on the default hierarchy for
241 * cases where a subsystem should behave differnetly depending on the
244 * The set of behaviors which change on the default hierarchy are still
245 * being determined and the mount option is prefixed with __DEVEL__.
247 * List of changed behaviors:
249 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
250 * and "name" are disallowed.
252 * - When mounting an existing superblock, mount options should match.
254 * - Remount is disallowed.
256 * - rename(2) is disallowed.
258 * - "tasks" is removed. Everything should be at process granularity. Use
259 * "cgroup.procs" instead.
261 * - "cgroup.procs" is not sorted. pids will be unique unless they got
262 * recycled inbetween reads.
264 * - "release_agent" and "notify_on_release" are removed. Replacement
265 * notification mechanism will be implemented.
267 * - "cgroup.clone_children" is removed.
269 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
270 * and its descendants contain no task; otherwise, 1. The file also
271 * generates kernfs notification which can be monitored through poll and
272 * [di]notify when the value of the file changes.
274 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
275 * take masks of ancestors with non-empty cpus/mems, instead of being
276 * moved to an ancestor.
278 * - cpuset: a task can be moved into an empty cpuset, and again it takes
279 * masks of ancestors.
281 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
284 * - blkcg: blk-throttle becomes properly hierarchical.
286 * - debug: disallowed on the default hierarchy.
288 bool cgroup_on_dfl(const struct cgroup *cgrp)
290 return cgrp->root == &cgrp_dfl_root;
293 /* IDR wrappers which synchronize using cgroup_idr_lock */
294 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
299 idr_preload(gfp_mask);
300 spin_lock_bh(&cgroup_idr_lock);
301 ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
302 spin_unlock_bh(&cgroup_idr_lock);
307 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
311 spin_lock_bh(&cgroup_idr_lock);
312 ret = idr_replace(idr, ptr, id);
313 spin_unlock_bh(&cgroup_idr_lock);
317 static void cgroup_idr_remove(struct idr *idr, int id)
319 spin_lock_bh(&cgroup_idr_lock);
321 spin_unlock_bh(&cgroup_idr_lock);
324 static bool cgroup_has_tasks(struct cgroup *cgrp)
326 return cgrp->nr_populated_csets;
329 bool cgroup_is_threaded(struct cgroup *cgrp)
331 return cgrp->dom_cgrp != cgrp;
334 /* can @cgrp host both domain and threaded children? */
335 static bool cgroup_is_mixable(struct cgroup *cgrp)
338 * Root isn't under domain level resource control exempting it from
339 * the no-internal-process constraint, so it can serve as a thread
340 * root and a parent of resource domains at the same time.
342 return !cgroup_parent(cgrp);
345 /* can @cgrp become a thread root? should always be true for a thread root */
346 static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
348 /* mixables don't care */
349 if (cgroup_is_mixable(cgrp))
352 /* domain roots can't be nested under threaded */
353 if (cgroup_is_threaded(cgrp))
356 /* can only have either domain or threaded children */
357 if (cgrp->nr_populated_domain_children)
360 /* and no domain controllers can be enabled */
361 if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
367 /* is @cgrp root of a threaded subtree? */
368 bool cgroup_is_thread_root(struct cgroup *cgrp)
370 /* thread root should be a domain */
371 if (cgroup_is_threaded(cgrp))
374 /* a domain w/ threaded children is a thread root */
375 if (cgrp->nr_threaded_children)
379 * A domain which has tasks and explicit threaded controllers
380 * enabled is a thread root.
382 if (cgroup_has_tasks(cgrp) &&
383 (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
389 /* a domain which isn't connected to the root w/o brekage can't be used */
390 static bool cgroup_is_valid_domain(struct cgroup *cgrp)
392 /* the cgroup itself can be a thread root */
393 if (cgroup_is_threaded(cgrp))
396 /* but the ancestors can't be unless mixable */
397 while ((cgrp = cgroup_parent(cgrp))) {
398 if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
400 if (cgroup_is_threaded(cgrp))
407 /* subsystems visibly enabled on a cgroup */
408 static u16 cgroup_control(struct cgroup *cgrp)
410 struct cgroup *parent = cgroup_parent(cgrp);
411 u16 root_ss_mask = cgrp->root->subsys_mask;
414 u16 ss_mask = parent->subtree_control;
416 /* threaded cgroups can only have threaded controllers */
417 if (cgroup_is_threaded(cgrp))
418 ss_mask &= cgrp_dfl_threaded_ss_mask;
422 if (cgroup_on_dfl(cgrp))
423 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
424 cgrp_dfl_implicit_ss_mask);
428 /* subsystems enabled on a cgroup */
429 static u16 cgroup_ss_mask(struct cgroup *cgrp)
431 struct cgroup *parent = cgroup_parent(cgrp);
434 u16 ss_mask = parent->subtree_ss_mask;
436 /* threaded cgroups can only have threaded controllers */
437 if (cgroup_is_threaded(cgrp))
438 ss_mask &= cgrp_dfl_threaded_ss_mask;
442 return cgrp->root->subsys_mask;
446 * cgroup_css - obtain a cgroup's css for the specified subsystem
447 * @cgrp: the cgroup of interest
448 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
450 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
451 * function must be called either under cgroup_mutex or rcu_read_lock() and
452 * the caller is responsible for pinning the returned css if it wants to
453 * keep accessing it outside the said locks. This function may return
454 * %NULL if @cgrp doesn't have @subsys_id enabled.
456 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
457 struct cgroup_subsys *ss)
460 return rcu_dereference_check(cgrp->subsys[ss->id],
461 lockdep_is_held(&cgroup_mutex));
467 * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
468 * @cgrp: the cgroup of interest
469 * @ss: the subsystem of interest
471 * Find and get @cgrp's css assocaited with @ss. If the css doesn't exist
472 * or is offline, %NULL is returned.
474 static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp,
475 struct cgroup_subsys *ss)
477 struct cgroup_subsys_state *css;
480 css = cgroup_css(cgrp, ss);
481 if (!css || !css_tryget_online(css))
489 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
490 * @cgrp: the cgroup of interest
491 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
493 * Similar to cgroup_css() but returns the effective css, which is defined
494 * as the matching css of the nearest ancestor including self which has @ss
495 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
496 * function is guaranteed to return non-NULL css.
498 static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
499 struct cgroup_subsys *ss)
501 lockdep_assert_held(&cgroup_mutex);
507 * This function is used while updating css associations and thus
508 * can't test the csses directly. Test ss_mask.
510 while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
511 cgrp = cgroup_parent(cgrp);
516 return cgroup_css(cgrp, ss);
520 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
521 * @cgrp: the cgroup of interest
522 * @ss: the subsystem of interest
524 * Find and get the effective css of @cgrp for @ss. The effective css is
525 * defined as the matching css of the nearest ancestor including self which
526 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
527 * the root css is returned, so this function always returns a valid css.
528 * The returned css must be put using css_put().
530 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
531 struct cgroup_subsys *ss)
533 struct cgroup_subsys_state *css;
538 css = cgroup_css(cgrp, ss);
540 if (css && css_tryget_online(css))
542 cgrp = cgroup_parent(cgrp);
545 css = init_css_set.subsys[ss->id];
552 static void cgroup_get_live(struct cgroup *cgrp)
554 WARN_ON_ONCE(cgroup_is_dead(cgrp));
555 css_get(&cgrp->self);
558 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
560 struct cgroup *cgrp = of->kn->parent->priv;
561 struct cftype *cft = of_cft(of);
564 * This is open and unprotected implementation of cgroup_css().
565 * seq_css() is only called from a kernfs file operation which has
566 * an active reference on the file. Because all the subsystem
567 * files are drained before a css is disassociated with a cgroup,
568 * the matching css from the cgroup's subsys table is guaranteed to
569 * be and stay valid until the enclosing operation is complete.
572 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
576 EXPORT_SYMBOL_GPL(of_css);
579 * for_each_css - iterate all css's of a cgroup
580 * @css: the iteration cursor
581 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
582 * @cgrp: the target cgroup to iterate css's of
584 * Should be called under cgroup_[tree_]mutex.
586 #define for_each_css(css, ssid, cgrp) \
587 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
588 if (!((css) = rcu_dereference_check( \
589 (cgrp)->subsys[(ssid)], \
590 lockdep_is_held(&cgroup_mutex)))) { } \
594 * for_each_e_css - iterate all effective css's of a cgroup
595 * @css: the iteration cursor
596 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
597 * @cgrp: the target cgroup to iterate css's of
599 * Should be called under cgroup_[tree_]mutex.
601 #define for_each_e_css(css, ssid, cgrp) \
602 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
603 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
608 * do_each_subsys_mask - filter for_each_subsys with a bitmask
609 * @ss: the iteration cursor
610 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
611 * @ss_mask: the bitmask
613 * The block will only run for cases where the ssid-th bit (1 << ssid) of
616 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
617 unsigned long __ss_mask = (ss_mask); \
618 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
622 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
623 (ss) = cgroup_subsys[ssid]; \
626 #define while_each_subsys_mask() \
631 /* iterate over child cgrps, lock should be held throughout iteration */
632 #define cgroup_for_each_live_child(child, cgrp) \
633 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
634 if (({ lockdep_assert_held(&cgroup_mutex); \
635 cgroup_is_dead(child); })) \
639 /* walk live descendants in preorder */
640 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
641 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
642 if (({ lockdep_assert_held(&cgroup_mutex); \
643 (dsct) = (d_css)->cgroup; \
644 cgroup_is_dead(dsct); })) \
648 /* walk live descendants in postorder */
649 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
650 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
651 if (({ lockdep_assert_held(&cgroup_mutex); \
652 (dsct) = (d_css)->cgroup; \
653 cgroup_is_dead(dsct); })) \
658 * The default css_set - used by init and its children prior to any
659 * hierarchies being mounted. It contains a pointer to the root state
660 * for each subsystem. Also used to anchor the list of css_sets. Not
661 * reference-counted, to improve performance when child cgroups
662 * haven't been created.
664 struct css_set init_css_set = {
665 .refcount = REFCOUNT_INIT(1),
666 .dom_cset = &init_css_set,
667 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
668 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
669 .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
670 .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets),
671 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
672 .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
673 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
676 * The following field is re-initialized when this cset gets linked
677 * in cgroup_init(). However, let's initialize the field
678 * statically too so that the default cgroup can be accessed safely
681 .dfl_cgrp = &cgrp_dfl_root.cgrp,
684 static int css_set_count = 1; /* 1 for init_css_set */
686 static bool css_set_threaded(struct css_set *cset)
688 return cset->dom_cset != cset;
692 * css_set_populated - does a css_set contain any tasks?
693 * @cset: target css_set
695 * css_set_populated() should be the same as !!cset->nr_tasks at steady
696 * state. However, css_set_populated() can be called while a task is being
697 * added to or removed from the linked list before the nr_tasks is
698 * properly updated. Hence, we can't just look at ->nr_tasks here.
700 static bool css_set_populated(struct css_set *cset)
702 lockdep_assert_held(&css_set_lock);
704 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
708 * cgroup_update_populated - update the populated count of a cgroup
709 * @cgrp: the target cgroup
710 * @populated: inc or dec populated count
712 * One of the css_sets associated with @cgrp is either getting its first
713 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
714 * count is propagated towards root so that a given cgroup's
715 * nr_populated_children is zero iff none of its descendants contain any
718 * @cgrp's interface file "cgroup.populated" is zero if both
719 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
720 * 1 otherwise. When the sum changes from or to zero, userland is notified
721 * that the content of the interface file has changed. This can be used to
722 * detect when @cgrp and its descendants become populated or empty.
724 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
726 struct cgroup *child = NULL;
727 int adj = populated ? 1 : -1;
729 lockdep_assert_held(&css_set_lock);
732 bool was_populated = cgroup_is_populated(cgrp);
735 cgrp->nr_populated_csets += adj;
737 if (cgroup_is_threaded(child))
738 cgrp->nr_populated_threaded_children += adj;
740 cgrp->nr_populated_domain_children += adj;
743 if (was_populated == cgroup_is_populated(cgrp))
746 cgroup1_check_for_release(cgrp);
747 cgroup_file_notify(&cgrp->events_file);
750 cgrp = cgroup_parent(cgrp);
755 * css_set_update_populated - update populated state of a css_set
756 * @cset: target css_set
757 * @populated: whether @cset is populated or depopulated
759 * @cset is either getting the first task or losing the last. Update the
760 * populated counters of all associated cgroups accordingly.
762 static void css_set_update_populated(struct css_set *cset, bool populated)
764 struct cgrp_cset_link *link;
766 lockdep_assert_held(&css_set_lock);
768 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
769 cgroup_update_populated(link->cgrp, populated);
773 * css_set_move_task - move a task from one css_set to another
774 * @task: task being moved
775 * @from_cset: css_set @task currently belongs to (may be NULL)
776 * @to_cset: new css_set @task is being moved to (may be NULL)
777 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
779 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
780 * css_set, @from_cset can be NULL. If @task is being disassociated
781 * instead of moved, @to_cset can be NULL.
783 * This function automatically handles populated counter updates and
784 * css_task_iter adjustments but the caller is responsible for managing
785 * @from_cset and @to_cset's reference counts.
787 static void css_set_move_task(struct task_struct *task,
788 struct css_set *from_cset, struct css_set *to_cset,
791 lockdep_assert_held(&css_set_lock);
793 if (to_cset && !css_set_populated(to_cset))
794 css_set_update_populated(to_cset, true);
797 struct css_task_iter *it, *pos;
799 WARN_ON_ONCE(list_empty(&task->cg_list));
802 * @task is leaving, advance task iterators which are
803 * pointing to it so that they can resume at the next
804 * position. Advancing an iterator might remove it from
805 * the list, use safe walk. See css_task_iter_advance*()
808 list_for_each_entry_safe(it, pos, &from_cset->task_iters,
810 if (it->task_pos == &task->cg_list)
811 css_task_iter_advance(it);
813 list_del_init(&task->cg_list);
814 if (!css_set_populated(from_cset))
815 css_set_update_populated(from_cset, false);
817 WARN_ON_ONCE(!list_empty(&task->cg_list));
822 * We are synchronized through cgroup_threadgroup_rwsem
823 * against PF_EXITING setting such that we can't race
824 * against cgroup_exit() changing the css_set to
825 * init_css_set and dropping the old one.
827 WARN_ON_ONCE(task->flags & PF_EXITING);
829 rcu_assign_pointer(task->cgroups, to_cset);
830 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
836 * hash table for cgroup groups. This improves the performance to find
837 * an existing css_set. This hash doesn't (currently) take into
838 * account cgroups in empty hierarchies.
840 #define CSS_SET_HASH_BITS 7
841 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
843 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
845 unsigned long key = 0UL;
846 struct cgroup_subsys *ss;
849 for_each_subsys(ss, i)
850 key += (unsigned long)css[i];
851 key = (key >> 16) ^ key;
856 void put_css_set_locked(struct css_set *cset)
858 struct cgrp_cset_link *link, *tmp_link;
859 struct cgroup_subsys *ss;
862 lockdep_assert_held(&css_set_lock);
864 if (!refcount_dec_and_test(&cset->refcount))
867 WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
869 /* This css_set is dead. unlink it and release cgroup and css refs */
870 for_each_subsys(ss, ssid) {
871 list_del(&cset->e_cset_node[ssid]);
872 css_put(cset->subsys[ssid]);
874 hash_del(&cset->hlist);
877 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
878 list_del(&link->cset_link);
879 list_del(&link->cgrp_link);
880 if (cgroup_parent(link->cgrp))
881 cgroup_put(link->cgrp);
885 if (css_set_threaded(cset)) {
886 list_del(&cset->threaded_csets_node);
887 put_css_set_locked(cset->dom_cset);
890 kfree_rcu(cset, rcu_head);
894 * compare_css_sets - helper function for find_existing_css_set().
895 * @cset: candidate css_set being tested
896 * @old_cset: existing css_set for a task
897 * @new_cgrp: cgroup that's being entered by the task
898 * @template: desired set of css pointers in css_set (pre-calculated)
900 * Returns true if "cset" matches "old_cset" except for the hierarchy
901 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
903 static bool compare_css_sets(struct css_set *cset,
904 struct css_set *old_cset,
905 struct cgroup *new_cgrp,
906 struct cgroup_subsys_state *template[])
908 struct cgroup *new_dfl_cgrp;
909 struct list_head *l1, *l2;
912 * On the default hierarchy, there can be csets which are
913 * associated with the same set of cgroups but different csses.
914 * Let's first ensure that csses match.
916 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
920 /* @cset's domain should match the default cgroup's */
921 if (cgroup_on_dfl(new_cgrp))
922 new_dfl_cgrp = new_cgrp;
924 new_dfl_cgrp = old_cset->dfl_cgrp;
926 if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
930 * Compare cgroup pointers in order to distinguish between
931 * different cgroups in hierarchies. As different cgroups may
932 * share the same effective css, this comparison is always
935 l1 = &cset->cgrp_links;
936 l2 = &old_cset->cgrp_links;
938 struct cgrp_cset_link *link1, *link2;
939 struct cgroup *cgrp1, *cgrp2;
943 /* See if we reached the end - both lists are equal length. */
944 if (l1 == &cset->cgrp_links) {
945 BUG_ON(l2 != &old_cset->cgrp_links);
948 BUG_ON(l2 == &old_cset->cgrp_links);
950 /* Locate the cgroups associated with these links. */
951 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
952 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
955 /* Hierarchies should be linked in the same order. */
956 BUG_ON(cgrp1->root != cgrp2->root);
959 * If this hierarchy is the hierarchy of the cgroup
960 * that's changing, then we need to check that this
961 * css_set points to the new cgroup; if it's any other
962 * hierarchy, then this css_set should point to the
963 * same cgroup as the old css_set.
965 if (cgrp1->root == new_cgrp->root) {
966 if (cgrp1 != new_cgrp)
977 * find_existing_css_set - init css array and find the matching css_set
978 * @old_cset: the css_set that we're using before the cgroup transition
979 * @cgrp: the cgroup that we're moving into
980 * @template: out param for the new set of csses, should be clear on entry
982 static struct css_set *find_existing_css_set(struct css_set *old_cset,
984 struct cgroup_subsys_state *template[])
986 struct cgroup_root *root = cgrp->root;
987 struct cgroup_subsys *ss;
988 struct css_set *cset;
993 * Build the set of subsystem state objects that we want to see in the
994 * new css_set. while subsystems can change globally, the entries here
995 * won't change, so no need for locking.
997 for_each_subsys(ss, i) {
998 if (root->subsys_mask & (1UL << i)) {
1000 * @ss is in this hierarchy, so we want the
1001 * effective css from @cgrp.
1003 template[i] = cgroup_e_css(cgrp, ss);
1006 * @ss is not in this hierarchy, so we don't want
1007 * to change the css.
1009 template[i] = old_cset->subsys[i];
1013 key = css_set_hash(template);
1014 hash_for_each_possible(css_set_table, cset, hlist, key) {
1015 if (!compare_css_sets(cset, old_cset, cgrp, template))
1018 /* This css_set matches what we need */
1022 /* No existing cgroup group matched */
1026 static void free_cgrp_cset_links(struct list_head *links_to_free)
1028 struct cgrp_cset_link *link, *tmp_link;
1030 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
1031 list_del(&link->cset_link);
1037 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1038 * @count: the number of links to allocate
1039 * @tmp_links: list_head the allocated links are put on
1041 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1042 * through ->cset_link. Returns 0 on success or -errno.
1044 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1046 struct cgrp_cset_link *link;
1049 INIT_LIST_HEAD(tmp_links);
1051 for (i = 0; i < count; i++) {
1052 link = kzalloc(sizeof(*link), GFP_KERNEL);
1054 free_cgrp_cset_links(tmp_links);
1057 list_add(&link->cset_link, tmp_links);
1063 * link_css_set - a helper function to link a css_set to a cgroup
1064 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1065 * @cset: the css_set to be linked
1066 * @cgrp: the destination cgroup
1068 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1069 struct cgroup *cgrp)
1071 struct cgrp_cset_link *link;
1073 BUG_ON(list_empty(tmp_links));
1075 if (cgroup_on_dfl(cgrp))
1076 cset->dfl_cgrp = cgrp;
1078 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1083 * Always add links to the tail of the lists so that the lists are
1084 * in choronological order.
1086 list_move_tail(&link->cset_link, &cgrp->cset_links);
1087 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1089 if (cgroup_parent(cgrp))
1090 cgroup_get_live(cgrp);
1094 * find_css_set - return a new css_set with one cgroup updated
1095 * @old_cset: the baseline css_set
1096 * @cgrp: the cgroup to be updated
1098 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1099 * substituted into the appropriate hierarchy.
1101 static struct css_set *find_css_set(struct css_set *old_cset,
1102 struct cgroup *cgrp)
1104 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1105 struct css_set *cset;
1106 struct list_head tmp_links;
1107 struct cgrp_cset_link *link;
1108 struct cgroup_subsys *ss;
1112 lockdep_assert_held(&cgroup_mutex);
1114 /* First see if we already have a cgroup group that matches
1115 * the desired set */
1116 spin_lock_irq(&css_set_lock);
1117 cset = find_existing_css_set(old_cset, cgrp, template);
1120 spin_unlock_irq(&css_set_lock);
1125 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1129 /* Allocate all the cgrp_cset_link objects that we'll need */
1130 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1135 refcount_set(&cset->refcount, 1);
1136 cset->dom_cset = cset;
1137 INIT_LIST_HEAD(&cset->tasks);
1138 INIT_LIST_HEAD(&cset->mg_tasks);
1139 INIT_LIST_HEAD(&cset->task_iters);
1140 INIT_LIST_HEAD(&cset->threaded_csets);
1141 INIT_HLIST_NODE(&cset->hlist);
1142 INIT_LIST_HEAD(&cset->cgrp_links);
1143 INIT_LIST_HEAD(&cset->mg_preload_node);
1144 INIT_LIST_HEAD(&cset->mg_node);
1146 /* Copy the set of subsystem state objects generated in
1147 * find_existing_css_set() */
1148 memcpy(cset->subsys, template, sizeof(cset->subsys));
1150 spin_lock_irq(&css_set_lock);
1151 /* Add reference counts and links from the new css_set. */
1152 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1153 struct cgroup *c = link->cgrp;
1155 if (c->root == cgrp->root)
1157 link_css_set(&tmp_links, cset, c);
1160 BUG_ON(!list_empty(&tmp_links));
1164 /* Add @cset to the hash table */
1165 key = css_set_hash(cset->subsys);
1166 hash_add(css_set_table, &cset->hlist, key);
1168 for_each_subsys(ss, ssid) {
1169 struct cgroup_subsys_state *css = cset->subsys[ssid];
1171 list_add_tail(&cset->e_cset_node[ssid],
1172 &css->cgroup->e_csets[ssid]);
1176 spin_unlock_irq(&css_set_lock);
1179 * If @cset should be threaded, look up the matching dom_cset and
1180 * link them up. We first fully initialize @cset then look for the
1181 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1182 * to stay empty until we return.
1184 if (cgroup_is_threaded(cset->dfl_cgrp)) {
1185 struct css_set *dcset;
1187 dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
1193 spin_lock_irq(&css_set_lock);
1194 cset->dom_cset = dcset;
1195 list_add_tail(&cset->threaded_csets_node,
1196 &dcset->threaded_csets);
1197 spin_unlock_irq(&css_set_lock);
1203 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1205 struct cgroup *root_cgrp = kf_root->kn->priv;
1207 return root_cgrp->root;
1210 static int cgroup_init_root_id(struct cgroup_root *root)
1214 lockdep_assert_held(&cgroup_mutex);
1216 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1220 root->hierarchy_id = id;
1224 static void cgroup_exit_root_id(struct cgroup_root *root)
1226 lockdep_assert_held(&cgroup_mutex);
1228 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1231 void cgroup_free_root(struct cgroup_root *root)
1234 idr_destroy(&root->cgroup_idr);
1239 static void cgroup_destroy_root(struct cgroup_root *root)
1241 struct cgroup *cgrp = &root->cgrp;
1242 struct cgrp_cset_link *link, *tmp_link;
1244 trace_cgroup_destroy_root(root);
1246 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1248 BUG_ON(atomic_read(&root->nr_cgrps));
1249 BUG_ON(!list_empty(&cgrp->self.children));
1251 /* Rebind all subsystems back to the default hierarchy */
1252 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1255 * Release all the links from cset_links to this hierarchy's
1258 spin_lock_irq(&css_set_lock);
1260 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1261 list_del(&link->cset_link);
1262 list_del(&link->cgrp_link);
1266 spin_unlock_irq(&css_set_lock);
1268 if (!list_empty(&root->root_list)) {
1269 list_del(&root->root_list);
1270 cgroup_root_count--;
1273 cgroup_exit_root_id(root);
1275 mutex_unlock(&cgroup_mutex);
1277 kernfs_destroy_root(root->kf_root);
1278 cgroup_free_root(root);
1282 * look up cgroup associated with current task's cgroup namespace on the
1283 * specified hierarchy
1285 static struct cgroup *
1286 current_cgns_cgroup_from_root(struct cgroup_root *root)
1288 struct cgroup *res = NULL;
1289 struct css_set *cset;
1291 lockdep_assert_held(&css_set_lock);
1295 cset = current->nsproxy->cgroup_ns->root_cset;
1296 if (cset == &init_css_set) {
1299 struct cgrp_cset_link *link;
1301 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1302 struct cgroup *c = link->cgrp;
1304 if (c->root == root) {
1316 /* look up cgroup associated with given css_set on the specified hierarchy */
1317 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1318 struct cgroup_root *root)
1320 struct cgroup *res = NULL;
1322 lockdep_assert_held(&cgroup_mutex);
1323 lockdep_assert_held(&css_set_lock);
1325 if (cset == &init_css_set) {
1327 } else if (root == &cgrp_dfl_root) {
1328 res = cset->dfl_cgrp;
1330 struct cgrp_cset_link *link;
1332 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1333 struct cgroup *c = link->cgrp;
1335 if (c->root == root) {
1347 * Return the cgroup for "task" from the given hierarchy. Must be
1348 * called with cgroup_mutex and css_set_lock held.
1350 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1351 struct cgroup_root *root)
1354 * No need to lock the task - since we hold cgroup_mutex the
1355 * task can't change groups, so the only thing that can happen
1356 * is that it exits and its css is set back to init_css_set.
1358 return cset_cgroup_from_root(task_css_set(task), root);
1362 * A task must hold cgroup_mutex to modify cgroups.
1364 * Any task can increment and decrement the count field without lock.
1365 * So in general, code holding cgroup_mutex can't rely on the count
1366 * field not changing. However, if the count goes to zero, then only
1367 * cgroup_attach_task() can increment it again. Because a count of zero
1368 * means that no tasks are currently attached, therefore there is no
1369 * way a task attached to that cgroup can fork (the other way to
1370 * increment the count). So code holding cgroup_mutex can safely
1371 * assume that if the count is zero, it will stay zero. Similarly, if
1372 * a task holds cgroup_mutex on a cgroup with zero count, it
1373 * knows that the cgroup won't be removed, as cgroup_rmdir()
1376 * A cgroup can only be deleted if both its 'count' of using tasks
1377 * is zero, and its list of 'children' cgroups is empty. Since all
1378 * tasks in the system use _some_ cgroup, and since there is always at
1379 * least one task in the system (init, pid == 1), therefore, root cgroup
1380 * always has either children cgroups and/or using tasks. So we don't
1381 * need a special hack to ensure that root cgroup cannot be deleted.
1383 * P.S. One more locking exception. RCU is used to guard the
1384 * update of a tasks cgroup pointer by cgroup_attach_task()
1387 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1389 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1392 struct cgroup_subsys *ss = cft->ss;
1394 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1395 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
1396 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
1397 cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1400 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1405 * cgroup_file_mode - deduce file mode of a control file
1406 * @cft: the control file in question
1408 * S_IRUGO for read, S_IWUSR for write.
1410 static umode_t cgroup_file_mode(const struct cftype *cft)
1414 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1417 if (cft->write_u64 || cft->write_s64 || cft->write) {
1418 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1428 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1429 * @subtree_control: the new subtree_control mask to consider
1430 * @this_ss_mask: available subsystems
1432 * On the default hierarchy, a subsystem may request other subsystems to be
1433 * enabled together through its ->depends_on mask. In such cases, more
1434 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1436 * This function calculates which subsystems need to be enabled if
1437 * @subtree_control is to be applied while restricted to @this_ss_mask.
1439 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1441 u16 cur_ss_mask = subtree_control;
1442 struct cgroup_subsys *ss;
1445 lockdep_assert_held(&cgroup_mutex);
1447 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1450 u16 new_ss_mask = cur_ss_mask;
1452 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1453 new_ss_mask |= ss->depends_on;
1454 } while_each_subsys_mask();
1457 * Mask out subsystems which aren't available. This can
1458 * happen only if some depended-upon subsystems were bound
1459 * to non-default hierarchies.
1461 new_ss_mask &= this_ss_mask;
1463 if (new_ss_mask == cur_ss_mask)
1465 cur_ss_mask = new_ss_mask;
1472 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1473 * @kn: the kernfs_node being serviced
1475 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1476 * the method finishes if locking succeeded. Note that once this function
1477 * returns the cgroup returned by cgroup_kn_lock_live() may become
1478 * inaccessible any time. If the caller intends to continue to access the
1479 * cgroup, it should pin it before invoking this function.
1481 void cgroup_kn_unlock(struct kernfs_node *kn)
1483 struct cgroup *cgrp;
1485 if (kernfs_type(kn) == KERNFS_DIR)
1488 cgrp = kn->parent->priv;
1490 mutex_unlock(&cgroup_mutex);
1492 kernfs_unbreak_active_protection(kn);
1497 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1498 * @kn: the kernfs_node being serviced
1499 * @drain_offline: perform offline draining on the cgroup
1501 * This helper is to be used by a cgroup kernfs method currently servicing
1502 * @kn. It breaks the active protection, performs cgroup locking and
1503 * verifies that the associated cgroup is alive. Returns the cgroup if
1504 * alive; otherwise, %NULL. A successful return should be undone by a
1505 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1506 * cgroup is drained of offlining csses before return.
1508 * Any cgroup kernfs method implementation which requires locking the
1509 * associated cgroup should use this helper. It avoids nesting cgroup
1510 * locking under kernfs active protection and allows all kernfs operations
1511 * including self-removal.
1513 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1515 struct cgroup *cgrp;
1517 if (kernfs_type(kn) == KERNFS_DIR)
1520 cgrp = kn->parent->priv;
1523 * We're gonna grab cgroup_mutex which nests outside kernfs
1524 * active_ref. cgroup liveliness check alone provides enough
1525 * protection against removal. Ensure @cgrp stays accessible and
1526 * break the active_ref protection.
1528 if (!cgroup_tryget(cgrp))
1530 kernfs_break_active_protection(kn);
1533 cgroup_lock_and_drain_offline(cgrp);
1535 mutex_lock(&cgroup_mutex);
1537 if (!cgroup_is_dead(cgrp))
1540 cgroup_kn_unlock(kn);
1544 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1546 char name[CGROUP_FILE_NAME_MAX];
1548 lockdep_assert_held(&cgroup_mutex);
1550 if (cft->file_offset) {
1551 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1552 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1554 spin_lock_irq(&cgroup_file_kn_lock);
1556 spin_unlock_irq(&cgroup_file_kn_lock);
1559 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1563 * css_clear_dir - remove subsys files in a cgroup directory
1566 static void css_clear_dir(struct cgroup_subsys_state *css)
1568 struct cgroup *cgrp = css->cgroup;
1569 struct cftype *cfts;
1571 if (!(css->flags & CSS_VISIBLE))
1574 css->flags &= ~CSS_VISIBLE;
1576 list_for_each_entry(cfts, &css->ss->cfts, node)
1577 cgroup_addrm_files(css, cgrp, cfts, false);
1581 * css_populate_dir - create subsys files in a cgroup directory
1584 * On failure, no file is added.
1586 static int css_populate_dir(struct cgroup_subsys_state *css)
1588 struct cgroup *cgrp = css->cgroup;
1589 struct cftype *cfts, *failed_cfts;
1592 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1596 if (cgroup_on_dfl(cgrp))
1597 cfts = cgroup_base_files;
1599 cfts = cgroup1_base_files;
1601 return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1604 list_for_each_entry(cfts, &css->ss->cfts, node) {
1605 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1612 css->flags |= CSS_VISIBLE;
1616 list_for_each_entry(cfts, &css->ss->cfts, node) {
1617 if (cfts == failed_cfts)
1619 cgroup_addrm_files(css, cgrp, cfts, false);
1624 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1626 struct cgroup *dcgrp = &dst_root->cgrp;
1627 struct cgroup_subsys *ss;
1630 lockdep_assert_held(&cgroup_mutex);
1632 do_each_subsys_mask(ss, ssid, ss_mask) {
1634 * If @ss has non-root csses attached to it, can't move.
1635 * If @ss is an implicit controller, it is exempt from this
1636 * rule and can be stolen.
1638 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1639 !ss->implicit_on_dfl)
1642 /* can't move between two non-dummy roots either */
1643 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1645 } while_each_subsys_mask();
1647 do_each_subsys_mask(ss, ssid, ss_mask) {
1648 struct cgroup_root *src_root = ss->root;
1649 struct cgroup *scgrp = &src_root->cgrp;
1650 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1651 struct css_set *cset;
1653 WARN_ON(!css || cgroup_css(dcgrp, ss));
1655 /* disable from the source */
1656 src_root->subsys_mask &= ~(1 << ssid);
1657 WARN_ON(cgroup_apply_control(scgrp));
1658 cgroup_finalize_control(scgrp, 0);
1661 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1662 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1663 ss->root = dst_root;
1664 css->cgroup = dcgrp;
1666 spin_lock_irq(&css_set_lock);
1667 hash_for_each(css_set_table, i, cset, hlist)
1668 list_move_tail(&cset->e_cset_node[ss->id],
1669 &dcgrp->e_csets[ss->id]);
1670 spin_unlock_irq(&css_set_lock);
1672 /* default hierarchy doesn't enable controllers by default */
1673 dst_root->subsys_mask |= 1 << ssid;
1674 if (dst_root == &cgrp_dfl_root) {
1675 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1677 dcgrp->subtree_control |= 1 << ssid;
1678 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1681 ret = cgroup_apply_control(dcgrp);
1683 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1688 } while_each_subsys_mask();
1690 kernfs_activate(dcgrp->kn);
1694 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1695 struct kernfs_root *kf_root)
1699 struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1700 struct cgroup *ns_cgroup;
1702 buf = kmalloc(PATH_MAX, GFP_KERNEL);
1706 spin_lock_irq(&css_set_lock);
1707 ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1708 len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1709 spin_unlock_irq(&css_set_lock);
1711 if (len >= PATH_MAX)
1714 seq_escape(sf, buf, " \t\n\\");
1721 static int parse_cgroup_root_flags(char *data, unsigned int *root_flags)
1730 while ((token = strsep(&data, ",")) != NULL) {
1731 if (!strcmp(token, "nsdelegate")) {
1732 *root_flags |= CGRP_ROOT_NS_DELEGATE;
1736 pr_err("cgroup2: unknown option \"%s\"\n", token);
1743 static void apply_cgroup_root_flags(unsigned int root_flags)
1745 if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
1746 if (root_flags & CGRP_ROOT_NS_DELEGATE)
1747 cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
1749 cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
1753 static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
1755 if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
1756 seq_puts(seq, ",nsdelegate");
1760 static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
1762 unsigned int root_flags;
1765 ret = parse_cgroup_root_flags(data, &root_flags);
1769 apply_cgroup_root_flags(root_flags);
1774 * To reduce the fork() overhead for systems that are not actually using
1775 * their cgroups capability, we don't maintain the lists running through
1776 * each css_set to its tasks until we see the list actually used - in other
1777 * words after the first mount.
1779 static bool use_task_css_set_links __read_mostly;
1781 static void cgroup_enable_task_cg_lists(void)
1783 struct task_struct *p, *g;
1785 spin_lock_irq(&css_set_lock);
1787 if (use_task_css_set_links)
1790 use_task_css_set_links = true;
1793 * We need tasklist_lock because RCU is not safe against
1794 * while_each_thread(). Besides, a forking task that has passed
1795 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1796 * is not guaranteed to have its child immediately visible in the
1797 * tasklist if we walk through it with RCU.
1799 read_lock(&tasklist_lock);
1800 do_each_thread(g, p) {
1801 WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1802 task_css_set(p) != &init_css_set);
1805 * We should check if the process is exiting, otherwise
1806 * it will race with cgroup_exit() in that the list
1807 * entry won't be deleted though the process has exited.
1808 * Do it while holding siglock so that we don't end up
1809 * racing against cgroup_exit().
1811 * Interrupts were already disabled while acquiring
1812 * the css_set_lock, so we do not need to disable it
1813 * again when acquiring the sighand->siglock here.
1815 spin_lock(&p->sighand->siglock);
1816 if (!(p->flags & PF_EXITING)) {
1817 struct css_set *cset = task_css_set(p);
1819 if (!css_set_populated(cset))
1820 css_set_update_populated(cset, true);
1821 list_add_tail(&p->cg_list, &cset->tasks);
1825 spin_unlock(&p->sighand->siglock);
1826 } while_each_thread(g, p);
1827 read_unlock(&tasklist_lock);
1829 spin_unlock_irq(&css_set_lock);
1832 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1834 struct cgroup_subsys *ss;
1837 INIT_LIST_HEAD(&cgrp->self.sibling);
1838 INIT_LIST_HEAD(&cgrp->self.children);
1839 INIT_LIST_HEAD(&cgrp->cset_links);
1840 INIT_LIST_HEAD(&cgrp->pidlists);
1841 mutex_init(&cgrp->pidlist_mutex);
1842 cgrp->self.cgroup = cgrp;
1843 cgrp->self.flags |= CSS_ONLINE;
1844 cgrp->dom_cgrp = cgrp;
1845 cgrp->max_descendants = INT_MAX;
1846 cgrp->max_depth = INT_MAX;
1848 for_each_subsys(ss, ssid)
1849 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1851 init_waitqueue_head(&cgrp->offline_waitq);
1852 INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1855 void init_cgroup_root(struct cgroup_root *root, struct cgroup_sb_opts *opts)
1857 struct cgroup *cgrp = &root->cgrp;
1859 INIT_LIST_HEAD(&root->root_list);
1860 atomic_set(&root->nr_cgrps, 1);
1862 init_cgroup_housekeeping(cgrp);
1863 idr_init(&root->cgroup_idr);
1865 root->flags = opts->flags;
1866 if (opts->release_agent)
1867 strcpy(root->release_agent_path, opts->release_agent);
1869 strcpy(root->name, opts->name);
1870 if (opts->cpuset_clone_children)
1871 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1874 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask, int ref_flags)
1876 LIST_HEAD(tmp_links);
1877 struct cgroup *root_cgrp = &root->cgrp;
1878 struct kernfs_syscall_ops *kf_sops;
1879 struct css_set *cset;
1882 lockdep_assert_held(&cgroup_mutex);
1884 ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL);
1887 root_cgrp->id = ret;
1888 root_cgrp->ancestor_ids[0] = ret;
1890 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
1891 ref_flags, GFP_KERNEL);
1896 * We're accessing css_set_count without locking css_set_lock here,
1897 * but that's OK - it can only be increased by someone holding
1898 * cgroup_lock, and that's us. Later rebinding may disable
1899 * controllers on the default hierarchy and thus create new csets,
1900 * which can't be more than the existing ones. Allocate 2x.
1902 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
1906 ret = cgroup_init_root_id(root);
1910 kf_sops = root == &cgrp_dfl_root ?
1911 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
1913 root->kf_root = kernfs_create_root(kf_sops,
1914 KERNFS_ROOT_CREATE_DEACTIVATED |
1915 KERNFS_ROOT_SUPPORT_EXPORTOP,
1917 if (IS_ERR(root->kf_root)) {
1918 ret = PTR_ERR(root->kf_root);
1921 root_cgrp->kn = root->kf_root->kn;
1923 ret = css_populate_dir(&root_cgrp->self);
1927 ret = rebind_subsystems(root, ss_mask);
1931 trace_cgroup_setup_root(root);
1934 * There must be no failure case after here, since rebinding takes
1935 * care of subsystems' refcounts, which are explicitly dropped in
1936 * the failure exit path.
1938 list_add(&root->root_list, &cgroup_roots);
1939 cgroup_root_count++;
1942 * Link the root cgroup in this hierarchy into all the css_set
1945 spin_lock_irq(&css_set_lock);
1946 hash_for_each(css_set_table, i, cset, hlist) {
1947 link_css_set(&tmp_links, cset, root_cgrp);
1948 if (css_set_populated(cset))
1949 cgroup_update_populated(root_cgrp, true);
1951 spin_unlock_irq(&css_set_lock);
1953 BUG_ON(!list_empty(&root_cgrp->self.children));
1954 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
1956 kernfs_activate(root_cgrp->kn);
1961 kernfs_destroy_root(root->kf_root);
1962 root->kf_root = NULL;
1964 cgroup_exit_root_id(root);
1966 percpu_ref_exit(&root_cgrp->self.refcnt);
1968 free_cgrp_cset_links(&tmp_links);
1972 struct dentry *cgroup_do_mount(struct file_system_type *fs_type, int flags,
1973 struct cgroup_root *root, unsigned long magic,
1974 struct cgroup_namespace *ns)
1976 struct dentry *dentry;
1979 dentry = kernfs_mount(fs_type, flags, root->kf_root, magic, &new_sb);
1982 * In non-init cgroup namespace, instead of root cgroup's dentry,
1983 * we return the dentry corresponding to the cgroupns->root_cgrp.
1985 if (!IS_ERR(dentry) && ns != &init_cgroup_ns) {
1986 struct dentry *nsdentry;
1987 struct cgroup *cgrp;
1989 mutex_lock(&cgroup_mutex);
1990 spin_lock_irq(&css_set_lock);
1992 cgrp = cset_cgroup_from_root(ns->root_cset, root);
1994 spin_unlock_irq(&css_set_lock);
1995 mutex_unlock(&cgroup_mutex);
1997 nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb);
2002 if (IS_ERR(dentry) || !new_sb)
2003 cgroup_put(&root->cgrp);
2008 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
2009 int flags, const char *unused_dev_name,
2012 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
2013 struct dentry *dentry;
2018 /* Check if the caller has permission to mount. */
2019 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) {
2021 return ERR_PTR(-EPERM);
2025 * The first time anyone tries to mount a cgroup, enable the list
2026 * linking each css_set to its tasks and fix up all existing tasks.
2028 if (!use_task_css_set_links)
2029 cgroup_enable_task_cg_lists();
2031 if (fs_type == &cgroup2_fs_type) {
2032 unsigned int root_flags;
2034 ret = parse_cgroup_root_flags(data, &root_flags);
2037 return ERR_PTR(ret);
2040 cgrp_dfl_visible = true;
2041 cgroup_get_live(&cgrp_dfl_root.cgrp);
2043 dentry = cgroup_do_mount(&cgroup2_fs_type, flags, &cgrp_dfl_root,
2044 CGROUP2_SUPER_MAGIC, ns);
2045 if (!IS_ERR(dentry))
2046 apply_cgroup_root_flags(root_flags);
2048 dentry = cgroup1_mount(&cgroup_fs_type, flags, data,
2049 CGROUP_SUPER_MAGIC, ns);
2056 static void cgroup_kill_sb(struct super_block *sb)
2058 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2059 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2062 * If @root doesn't have any mounts or children, start killing it.
2063 * This prevents new mounts by disabling percpu_ref_tryget_live().
2064 * cgroup_mount() may wait for @root's release.
2066 * And don't kill the default root.
2068 if (!list_empty(&root->cgrp.self.children) ||
2069 root == &cgrp_dfl_root)
2070 cgroup_put(&root->cgrp);
2072 percpu_ref_kill(&root->cgrp.self.refcnt);
2077 struct file_system_type cgroup_fs_type = {
2079 .mount = cgroup_mount,
2080 .kill_sb = cgroup_kill_sb,
2081 .fs_flags = FS_USERNS_MOUNT,
2084 static struct file_system_type cgroup2_fs_type = {
2086 .mount = cgroup_mount,
2087 .kill_sb = cgroup_kill_sb,
2088 .fs_flags = FS_USERNS_MOUNT,
2091 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2092 struct cgroup_namespace *ns)
2094 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2096 return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2099 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2100 struct cgroup_namespace *ns)
2104 mutex_lock(&cgroup_mutex);
2105 spin_lock_irq(&css_set_lock);
2107 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2109 spin_unlock_irq(&css_set_lock);
2110 mutex_unlock(&cgroup_mutex);
2114 EXPORT_SYMBOL_GPL(cgroup_path_ns);
2117 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2118 * @task: target task
2119 * @buf: the buffer to write the path into
2120 * @buflen: the length of the buffer
2122 * Determine @task's cgroup on the first (the one with the lowest non-zero
2123 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2124 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2125 * cgroup controller callbacks.
2127 * Return value is the same as kernfs_path().
2129 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2131 struct cgroup_root *root;
2132 struct cgroup *cgrp;
2133 int hierarchy_id = 1;
2136 mutex_lock(&cgroup_mutex);
2137 spin_lock_irq(&css_set_lock);
2139 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2142 cgrp = task_cgroup_from_root(task, root);
2143 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2145 /* if no hierarchy exists, everyone is in "/" */
2146 ret = strlcpy(buf, "/", buflen);
2149 spin_unlock_irq(&css_set_lock);
2150 mutex_unlock(&cgroup_mutex);
2153 EXPORT_SYMBOL_GPL(task_cgroup_path);
2156 * cgroup_migrate_add_task - add a migration target task to a migration context
2157 * @task: target task
2158 * @mgctx: target migration context
2160 * Add @task, which is a migration target, to @mgctx->tset. This function
2161 * becomes noop if @task doesn't need to be migrated. @task's css_set
2162 * should have been added as a migration source and @task->cg_list will be
2163 * moved from the css_set's tasks list to mg_tasks one.
2165 static void cgroup_migrate_add_task(struct task_struct *task,
2166 struct cgroup_mgctx *mgctx)
2168 struct css_set *cset;
2170 lockdep_assert_held(&css_set_lock);
2172 /* @task either already exited or can't exit until the end */
2173 if (task->flags & PF_EXITING)
2176 /* leave @task alone if post_fork() hasn't linked it yet */
2177 if (list_empty(&task->cg_list))
2180 cset = task_css_set(task);
2181 if (!cset->mg_src_cgrp)
2184 mgctx->tset.nr_tasks++;
2186 list_move_tail(&task->cg_list, &cset->mg_tasks);
2187 if (list_empty(&cset->mg_node))
2188 list_add_tail(&cset->mg_node,
2189 &mgctx->tset.src_csets);
2190 if (list_empty(&cset->mg_dst_cset->mg_node))
2191 list_add_tail(&cset->mg_dst_cset->mg_node,
2192 &mgctx->tset.dst_csets);
2196 * cgroup_taskset_first - reset taskset and return the first task
2197 * @tset: taskset of interest
2198 * @dst_cssp: output variable for the destination css
2200 * @tset iteration is initialized and the first task is returned.
2202 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2203 struct cgroup_subsys_state **dst_cssp)
2205 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2206 tset->cur_task = NULL;
2208 return cgroup_taskset_next(tset, dst_cssp);
2212 * cgroup_taskset_next - iterate to the next task in taskset
2213 * @tset: taskset of interest
2214 * @dst_cssp: output variable for the destination css
2216 * Return the next task in @tset. Iteration must have been initialized
2217 * with cgroup_taskset_first().
2219 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2220 struct cgroup_subsys_state **dst_cssp)
2222 struct css_set *cset = tset->cur_cset;
2223 struct task_struct *task = tset->cur_task;
2225 while (&cset->mg_node != tset->csets) {
2227 task = list_first_entry(&cset->mg_tasks,
2228 struct task_struct, cg_list);
2230 task = list_next_entry(task, cg_list);
2232 if (&task->cg_list != &cset->mg_tasks) {
2233 tset->cur_cset = cset;
2234 tset->cur_task = task;
2237 * This function may be called both before and
2238 * after cgroup_taskset_migrate(). The two cases
2239 * can be distinguished by looking at whether @cset
2240 * has its ->mg_dst_cset set.
2242 if (cset->mg_dst_cset)
2243 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2245 *dst_cssp = cset->subsys[tset->ssid];
2250 cset = list_next_entry(cset, mg_node);
2258 * cgroup_taskset_migrate - migrate a taskset
2259 * @mgctx: migration context
2261 * Migrate tasks in @mgctx as setup by migration preparation functions.
2262 * This function fails iff one of the ->can_attach callbacks fails and
2263 * guarantees that either all or none of the tasks in @mgctx are migrated.
2264 * @mgctx is consumed regardless of success.
2266 static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2268 struct cgroup_taskset *tset = &mgctx->tset;
2269 struct cgroup_subsys *ss;
2270 struct task_struct *task, *tmp_task;
2271 struct css_set *cset, *tmp_cset;
2272 int ssid, failed_ssid, ret;
2274 /* check that we can legitimately attach to the cgroup */
2275 if (tset->nr_tasks) {
2276 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2277 if (ss->can_attach) {
2279 ret = ss->can_attach(tset);
2282 goto out_cancel_attach;
2285 } while_each_subsys_mask();
2289 * Now that we're guaranteed success, proceed to move all tasks to
2290 * the new cgroup. There are no failure cases after here, so this
2291 * is the commit point.
2293 spin_lock_irq(&css_set_lock);
2294 list_for_each_entry(cset, &tset->src_csets, mg_node) {
2295 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2296 struct css_set *from_cset = task_css_set(task);
2297 struct css_set *to_cset = cset->mg_dst_cset;
2299 get_css_set(to_cset);
2300 to_cset->nr_tasks++;
2301 css_set_move_task(task, from_cset, to_cset, true);
2302 put_css_set_locked(from_cset);
2303 from_cset->nr_tasks--;
2306 spin_unlock_irq(&css_set_lock);
2309 * Migration is committed, all target tasks are now on dst_csets.
2310 * Nothing is sensitive to fork() after this point. Notify
2311 * controllers that migration is complete.
2313 tset->csets = &tset->dst_csets;
2315 if (tset->nr_tasks) {
2316 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2321 } while_each_subsys_mask();
2325 goto out_release_tset;
2328 if (tset->nr_tasks) {
2329 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2330 if (ssid == failed_ssid)
2332 if (ss->cancel_attach) {
2334 ss->cancel_attach(tset);
2336 } while_each_subsys_mask();
2339 spin_lock_irq(&css_set_lock);
2340 list_splice_init(&tset->dst_csets, &tset->src_csets);
2341 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2342 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2343 list_del_init(&cset->mg_node);
2345 spin_unlock_irq(&css_set_lock);
2350 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2351 * @dst_cgrp: destination cgroup to test
2353 * On the default hierarchy, except for the mixable, (possible) thread root
2354 * and threaded cgroups, subtree_control must be zero for migration
2355 * destination cgroups with tasks so that child cgroups don't compete
2358 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2360 /* v1 doesn't have any restriction */
2361 if (!cgroup_on_dfl(dst_cgrp))
2364 /* verify @dst_cgrp can host resources */
2365 if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
2368 /* mixables don't care */
2369 if (cgroup_is_mixable(dst_cgrp))
2373 * If @dst_cgrp is already or can become a thread root or is
2374 * threaded, it doesn't matter.
2376 if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
2379 /* apply no-internal-process constraint */
2380 if (dst_cgrp->subtree_control)
2387 * cgroup_migrate_finish - cleanup after attach
2388 * @mgctx: migration context
2390 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2391 * those functions for details.
2393 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2395 LIST_HEAD(preloaded);
2396 struct css_set *cset, *tmp_cset;
2398 lockdep_assert_held(&cgroup_mutex);
2400 spin_lock_irq(&css_set_lock);
2402 list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded);
2403 list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded);
2405 list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) {
2406 cset->mg_src_cgrp = NULL;
2407 cset->mg_dst_cgrp = NULL;
2408 cset->mg_dst_cset = NULL;
2409 list_del_init(&cset->mg_preload_node);
2410 put_css_set_locked(cset);
2413 spin_unlock_irq(&css_set_lock);
2417 * cgroup_migrate_add_src - add a migration source css_set
2418 * @src_cset: the source css_set to add
2419 * @dst_cgrp: the destination cgroup
2420 * @mgctx: migration context
2422 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2423 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2424 * up by cgroup_migrate_finish().
2426 * This function may be called without holding cgroup_threadgroup_rwsem
2427 * even if the target is a process. Threads may be created and destroyed
2428 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2429 * into play and the preloaded css_sets are guaranteed to cover all
2432 void cgroup_migrate_add_src(struct css_set *src_cset,
2433 struct cgroup *dst_cgrp,
2434 struct cgroup_mgctx *mgctx)
2436 struct cgroup *src_cgrp;
2438 lockdep_assert_held(&cgroup_mutex);
2439 lockdep_assert_held(&css_set_lock);
2442 * If ->dead, @src_set is associated with one or more dead cgroups
2443 * and doesn't contain any migratable tasks. Ignore it early so
2444 * that the rest of migration path doesn't get confused by it.
2449 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2451 if (!list_empty(&src_cset->mg_preload_node))
2454 WARN_ON(src_cset->mg_src_cgrp);
2455 WARN_ON(src_cset->mg_dst_cgrp);
2456 WARN_ON(!list_empty(&src_cset->mg_tasks));
2457 WARN_ON(!list_empty(&src_cset->mg_node));
2459 src_cset->mg_src_cgrp = src_cgrp;
2460 src_cset->mg_dst_cgrp = dst_cgrp;
2461 get_css_set(src_cset);
2462 list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets);
2466 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2467 * @mgctx: migration context
2469 * Tasks are about to be moved and all the source css_sets have been
2470 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2471 * pins all destination css_sets, links each to its source, and append them
2472 * to @mgctx->preloaded_dst_csets.
2474 * This function must be called after cgroup_migrate_add_src() has been
2475 * called on each migration source css_set. After migration is performed
2476 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2479 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2481 struct css_set *src_cset, *tmp_cset;
2483 lockdep_assert_held(&cgroup_mutex);
2485 /* look up the dst cset for each src cset and link it to src */
2486 list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2488 struct css_set *dst_cset;
2489 struct cgroup_subsys *ss;
2492 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2496 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2499 * If src cset equals dst, it's noop. Drop the src.
2500 * cgroup_migrate() will skip the cset too. Note that we
2501 * can't handle src == dst as some nodes are used by both.
2503 if (src_cset == dst_cset) {
2504 src_cset->mg_src_cgrp = NULL;
2505 src_cset->mg_dst_cgrp = NULL;
2506 list_del_init(&src_cset->mg_preload_node);
2507 put_css_set(src_cset);
2508 put_css_set(dst_cset);
2512 src_cset->mg_dst_cset = dst_cset;
2514 if (list_empty(&dst_cset->mg_preload_node))
2515 list_add_tail(&dst_cset->mg_preload_node,
2516 &mgctx->preloaded_dst_csets);
2518 put_css_set(dst_cset);
2520 for_each_subsys(ss, ssid)
2521 if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2522 mgctx->ss_mask |= 1 << ssid;
2527 cgroup_migrate_finish(mgctx);
2532 * cgroup_migrate - migrate a process or task to a cgroup
2533 * @leader: the leader of the process or the task to migrate
2534 * @threadgroup: whether @leader points to the whole process or a single task
2535 * @mgctx: migration context
2537 * Migrate a process or task denoted by @leader. If migrating a process,
2538 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2539 * responsible for invoking cgroup_migrate_add_src() and
2540 * cgroup_migrate_prepare_dst() on the targets before invoking this
2541 * function and following up with cgroup_migrate_finish().
2543 * As long as a controller's ->can_attach() doesn't fail, this function is
2544 * guaranteed to succeed. This means that, excluding ->can_attach()
2545 * failure, when migrating multiple targets, the success or failure can be
2546 * decided for all targets by invoking group_migrate_prepare_dst() before
2547 * actually starting migrating.
2549 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2550 struct cgroup_mgctx *mgctx)
2552 struct task_struct *task;
2555 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2556 * already PF_EXITING could be freed from underneath us unless we
2557 * take an rcu_read_lock.
2559 spin_lock_irq(&css_set_lock);
2563 cgroup_migrate_add_task(task, mgctx);
2566 } while_each_thread(leader, task);
2568 spin_unlock_irq(&css_set_lock);
2570 return cgroup_migrate_execute(mgctx);
2574 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2575 * @dst_cgrp: the cgroup to attach to
2576 * @leader: the task or the leader of the threadgroup to be attached
2577 * @threadgroup: attach the whole threadgroup?
2579 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2581 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2584 DEFINE_CGROUP_MGCTX(mgctx);
2585 struct task_struct *task;
2588 ret = cgroup_migrate_vet_dst(dst_cgrp);
2592 /* look up all src csets */
2593 spin_lock_irq(&css_set_lock);
2597 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2600 } while_each_thread(leader, task);
2602 spin_unlock_irq(&css_set_lock);
2604 /* prepare dst csets and commit */
2605 ret = cgroup_migrate_prepare_dst(&mgctx);
2607 ret = cgroup_migrate(leader, threadgroup, &mgctx);
2609 cgroup_migrate_finish(&mgctx);
2612 trace_cgroup_attach_task(dst_cgrp, leader, threadgroup);
2617 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup)
2618 __acquires(&cgroup_threadgroup_rwsem)
2620 struct task_struct *tsk;
2623 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2624 return ERR_PTR(-EINVAL);
2626 percpu_down_write(&cgroup_threadgroup_rwsem);
2630 tsk = find_task_by_vpid(pid);
2632 tsk = ERR_PTR(-ESRCH);
2633 goto out_unlock_threadgroup;
2640 tsk = tsk->group_leader;
2643 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2644 * If userland migrates such a kthread to a non-root cgroup, it can
2645 * become trapped in a cpuset, or RT kthread may be born in a
2646 * cgroup with no rt_runtime allocated. Just say no.
2648 if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2649 tsk = ERR_PTR(-EINVAL);
2650 goto out_unlock_threadgroup;
2653 get_task_struct(tsk);
2654 goto out_unlock_rcu;
2656 out_unlock_threadgroup:
2657 percpu_up_write(&cgroup_threadgroup_rwsem);
2663 void cgroup_procs_write_finish(struct task_struct *task)
2664 __releases(&cgroup_threadgroup_rwsem)
2666 struct cgroup_subsys *ss;
2669 /* release reference from cgroup_procs_write_start() */
2670 put_task_struct(task);
2672 percpu_up_write(&cgroup_threadgroup_rwsem);
2673 for_each_subsys(ss, ssid)
2674 if (ss->post_attach)
2678 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2680 struct cgroup_subsys *ss;
2681 bool printed = false;
2684 do_each_subsys_mask(ss, ssid, ss_mask) {
2687 seq_printf(seq, "%s", ss->name);
2689 } while_each_subsys_mask();
2691 seq_putc(seq, '\n');
2694 /* show controllers which are enabled from the parent */
2695 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2697 struct cgroup *cgrp = seq_css(seq)->cgroup;
2699 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2703 /* show controllers which are enabled for a given cgroup's children */
2704 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2706 struct cgroup *cgrp = seq_css(seq)->cgroup;
2708 cgroup_print_ss_mask(seq, cgrp->subtree_control);
2713 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2714 * @cgrp: root of the subtree to update csses for
2716 * @cgrp's control masks have changed and its subtree's css associations
2717 * need to be updated accordingly. This function looks up all css_sets
2718 * which are attached to the subtree, creates the matching updated css_sets
2719 * and migrates the tasks to the new ones.
2721 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2723 DEFINE_CGROUP_MGCTX(mgctx);
2724 struct cgroup_subsys_state *d_css;
2725 struct cgroup *dsct;
2726 struct css_set *src_cset;
2729 lockdep_assert_held(&cgroup_mutex);
2731 percpu_down_write(&cgroup_threadgroup_rwsem);
2733 /* look up all csses currently attached to @cgrp's subtree */
2734 spin_lock_irq(&css_set_lock);
2735 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2736 struct cgrp_cset_link *link;
2738 list_for_each_entry(link, &dsct->cset_links, cset_link)
2739 cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2741 spin_unlock_irq(&css_set_lock);
2743 /* NULL dst indicates self on default hierarchy */
2744 ret = cgroup_migrate_prepare_dst(&mgctx);
2748 spin_lock_irq(&css_set_lock);
2749 list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) {
2750 struct task_struct *task, *ntask;
2752 /* all tasks in src_csets need to be migrated */
2753 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2754 cgroup_migrate_add_task(task, &mgctx);
2756 spin_unlock_irq(&css_set_lock);
2758 ret = cgroup_migrate_execute(&mgctx);
2760 cgroup_migrate_finish(&mgctx);
2761 percpu_up_write(&cgroup_threadgroup_rwsem);
2766 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2767 * @cgrp: root of the target subtree
2769 * Because css offlining is asynchronous, userland may try to re-enable a
2770 * controller while the previous css is still around. This function grabs
2771 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2773 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
2774 __acquires(&cgroup_mutex)
2776 struct cgroup *dsct;
2777 struct cgroup_subsys_state *d_css;
2778 struct cgroup_subsys *ss;
2782 mutex_lock(&cgroup_mutex);
2784 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2785 for_each_subsys(ss, ssid) {
2786 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2789 if (!css || !percpu_ref_is_dying(&css->refcnt))
2792 cgroup_get_live(dsct);
2793 prepare_to_wait(&dsct->offline_waitq, &wait,
2794 TASK_UNINTERRUPTIBLE);
2796 mutex_unlock(&cgroup_mutex);
2798 finish_wait(&dsct->offline_waitq, &wait);
2807 * cgroup_save_control - save control masks of a subtree
2808 * @cgrp: root of the target subtree
2810 * Save ->subtree_control and ->subtree_ss_mask to the respective old_
2811 * prefixed fields for @cgrp's subtree including @cgrp itself.
2813 static void cgroup_save_control(struct cgroup *cgrp)
2815 struct cgroup *dsct;
2816 struct cgroup_subsys_state *d_css;
2818 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2819 dsct->old_subtree_control = dsct->subtree_control;
2820 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
2825 * cgroup_propagate_control - refresh control masks of a subtree
2826 * @cgrp: root of the target subtree
2828 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
2829 * ->subtree_control and propagate controller availability through the
2830 * subtree so that descendants don't have unavailable controllers enabled.
2832 static void cgroup_propagate_control(struct cgroup *cgrp)
2834 struct cgroup *dsct;
2835 struct cgroup_subsys_state *d_css;
2837 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2838 dsct->subtree_control &= cgroup_control(dsct);
2839 dsct->subtree_ss_mask =
2840 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
2841 cgroup_ss_mask(dsct));
2846 * cgroup_restore_control - restore control masks of a subtree
2847 * @cgrp: root of the target subtree
2849 * Restore ->subtree_control and ->subtree_ss_mask from the respective old_
2850 * prefixed fields for @cgrp's subtree including @cgrp itself.
2852 static void cgroup_restore_control(struct cgroup *cgrp)
2854 struct cgroup *dsct;
2855 struct cgroup_subsys_state *d_css;
2857 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2858 dsct->subtree_control = dsct->old_subtree_control;
2859 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
2863 static bool css_visible(struct cgroup_subsys_state *css)
2865 struct cgroup_subsys *ss = css->ss;
2866 struct cgroup *cgrp = css->cgroup;
2868 if (cgroup_control(cgrp) & (1 << ss->id))
2870 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
2872 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
2876 * cgroup_apply_control_enable - enable or show csses according to control
2877 * @cgrp: root of the target subtree
2879 * Walk @cgrp's subtree and create new csses or make the existing ones
2880 * visible. A css is created invisible if it's being implicitly enabled
2881 * through dependency. An invisible css is made visible when the userland
2882 * explicitly enables it.
2884 * Returns 0 on success, -errno on failure. On failure, csses which have
2885 * been processed already aren't cleaned up. The caller is responsible for
2886 * cleaning up with cgroup_apply_control_disable().
2888 static int cgroup_apply_control_enable(struct cgroup *cgrp)
2890 struct cgroup *dsct;
2891 struct cgroup_subsys_state *d_css;
2892 struct cgroup_subsys *ss;
2895 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2896 for_each_subsys(ss, ssid) {
2897 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2899 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
2901 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
2905 css = css_create(dsct, ss);
2907 return PTR_ERR(css);
2910 if (css_visible(css)) {
2911 ret = css_populate_dir(css);
2922 * cgroup_apply_control_disable - kill or hide csses according to control
2923 * @cgrp: root of the target subtree
2925 * Walk @cgrp's subtree and kill and hide csses so that they match
2926 * cgroup_ss_mask() and cgroup_visible_mask().
2928 * A css is hidden when the userland requests it to be disabled while other
2929 * subsystems are still depending on it. The css must not actively control
2930 * resources and be in the vanilla state if it's made visible again later.
2931 * Controllers which may be depended upon should provide ->css_reset() for
2934 static void cgroup_apply_control_disable(struct cgroup *cgrp)
2936 struct cgroup *dsct;
2937 struct cgroup_subsys_state *d_css;
2938 struct cgroup_subsys *ss;
2941 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2942 for_each_subsys(ss, ssid) {
2943 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2945 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
2951 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
2953 } else if (!css_visible(css)) {
2963 * cgroup_apply_control - apply control mask updates to the subtree
2964 * @cgrp: root of the target subtree
2966 * subsystems can be enabled and disabled in a subtree using the following
2969 * 1. Call cgroup_save_control() to stash the current state.
2970 * 2. Update ->subtree_control masks in the subtree as desired.
2971 * 3. Call cgroup_apply_control() to apply the changes.
2972 * 4. Optionally perform other related operations.
2973 * 5. Call cgroup_finalize_control() to finish up.
2975 * This function implements step 3 and propagates the mask changes
2976 * throughout @cgrp's subtree, updates csses accordingly and perform
2977 * process migrations.
2979 static int cgroup_apply_control(struct cgroup *cgrp)
2983 cgroup_propagate_control(cgrp);
2985 ret = cgroup_apply_control_enable(cgrp);
2990 * At this point, cgroup_e_css() results reflect the new csses
2991 * making the following cgroup_update_dfl_csses() properly update
2992 * css associations of all tasks in the subtree.
2994 ret = cgroup_update_dfl_csses(cgrp);
3002 * cgroup_finalize_control - finalize control mask update
3003 * @cgrp: root of the target subtree
3004 * @ret: the result of the update
3006 * Finalize control mask update. See cgroup_apply_control() for more info.
3008 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3011 cgroup_restore_control(cgrp);
3012 cgroup_propagate_control(cgrp);
3015 cgroup_apply_control_disable(cgrp);
3018 static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
3020 u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
3022 /* if nothing is getting enabled, nothing to worry about */
3026 /* can @cgrp host any resources? */
3027 if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
3030 /* mixables don't care */
3031 if (cgroup_is_mixable(cgrp))
3034 if (domain_enable) {
3035 /* can't enable domain controllers inside a thread subtree */
3036 if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3040 * Threaded controllers can handle internal competitions
3041 * and are always allowed inside a (prospective) thread
3044 if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3049 * Controllers can't be enabled for a cgroup with tasks to avoid
3050 * child cgroups competing against tasks.
3052 if (cgroup_has_tasks(cgrp))
3058 /* change the enabled child controllers for a cgroup in the default hierarchy */
3059 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3060 char *buf, size_t nbytes,
3063 u16 enable = 0, disable = 0;
3064 struct cgroup *cgrp, *child;
3065 struct cgroup_subsys *ss;
3070 * Parse input - space separated list of subsystem names prefixed
3071 * with either + or -.
3073 buf = strstrip(buf);
3074 while ((tok = strsep(&buf, " "))) {
3077 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3078 if (!cgroup_ssid_enabled(ssid) ||
3079 strcmp(tok + 1, ss->name))
3083 enable |= 1 << ssid;
3084 disable &= ~(1 << ssid);
3085 } else if (*tok == '-') {
3086 disable |= 1 << ssid;
3087 enable &= ~(1 << ssid);
3092 } while_each_subsys_mask();
3093 if (ssid == CGROUP_SUBSYS_COUNT)
3097 cgrp = cgroup_kn_lock_live(of->kn, true);
3101 for_each_subsys(ss, ssid) {
3102 if (enable & (1 << ssid)) {
3103 if (cgrp->subtree_control & (1 << ssid)) {
3104 enable &= ~(1 << ssid);
3108 if (!(cgroup_control(cgrp) & (1 << ssid))) {
3112 } else if (disable & (1 << ssid)) {
3113 if (!(cgrp->subtree_control & (1 << ssid))) {
3114 disable &= ~(1 << ssid);
3118 /* a child has it enabled? */
3119 cgroup_for_each_live_child(child, cgrp) {
3120 if (child->subtree_control & (1 << ssid)) {
3128 if (!enable && !disable) {
3133 ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3137 /* save and update control masks and prepare csses */
3138 cgroup_save_control(cgrp);
3140 cgrp->subtree_control |= enable;
3141 cgrp->subtree_control &= ~disable;
3143 ret = cgroup_apply_control(cgrp);
3144 cgroup_finalize_control(cgrp, ret);
3148 kernfs_activate(cgrp->kn);
3150 cgroup_kn_unlock(of->kn);
3151 return ret ?: nbytes;
3155 * cgroup_enable_threaded - make @cgrp threaded
3156 * @cgrp: the target cgroup
3158 * Called when "threaded" is written to the cgroup.type interface file and
3159 * tries to make @cgrp threaded and join the parent's resource domain.
3160 * This function is never called on the root cgroup as cgroup.type doesn't
3163 static int cgroup_enable_threaded(struct cgroup *cgrp)
3165 struct cgroup *parent = cgroup_parent(cgrp);
3166 struct cgroup *dom_cgrp = parent->dom_cgrp;
3169 lockdep_assert_held(&cgroup_mutex);
3171 /* noop if already threaded */
3172 if (cgroup_is_threaded(cgrp))
3175 /* we're joining the parent's domain, ensure its validity */
3176 if (!cgroup_is_valid_domain(dom_cgrp) ||
3177 !cgroup_can_be_thread_root(dom_cgrp))
3181 * The following shouldn't cause actual migrations and should
3184 cgroup_save_control(cgrp);
3186 cgrp->dom_cgrp = dom_cgrp;
3187 ret = cgroup_apply_control(cgrp);
3189 parent->nr_threaded_children++;
3191 cgrp->dom_cgrp = cgrp;
3193 cgroup_finalize_control(cgrp, ret);
3197 static int cgroup_type_show(struct seq_file *seq, void *v)
3199 struct cgroup *cgrp = seq_css(seq)->cgroup;
3201 if (cgroup_is_threaded(cgrp))
3202 seq_puts(seq, "threaded\n");
3203 else if (!cgroup_is_valid_domain(cgrp))
3204 seq_puts(seq, "domain invalid\n");
3205 else if (cgroup_is_thread_root(cgrp))
3206 seq_puts(seq, "domain threaded\n");
3208 seq_puts(seq, "domain\n");
3213 static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3214 size_t nbytes, loff_t off)
3216 struct cgroup *cgrp;
3219 /* only switching to threaded mode is supported */
3220 if (strcmp(strstrip(buf), "threaded"))
3223 cgrp = cgroup_kn_lock_live(of->kn, false);
3227 /* threaded can only be enabled */
3228 ret = cgroup_enable_threaded(cgrp);
3230 cgroup_kn_unlock(of->kn);
3231 return ret ?: nbytes;
3234 static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3236 struct cgroup *cgrp = seq_css(seq)->cgroup;
3237 int descendants = READ_ONCE(cgrp->max_descendants);
3239 if (descendants == INT_MAX)
3240 seq_puts(seq, "max\n");
3242 seq_printf(seq, "%d\n", descendants);
3247 static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3248 char *buf, size_t nbytes, loff_t off)
3250 struct cgroup *cgrp;
3254 buf = strstrip(buf);
3255 if (!strcmp(buf, "max")) {
3256 descendants = INT_MAX;
3258 ret = kstrtoint(buf, 0, &descendants);
3263 if (descendants < 0)
3266 cgrp = cgroup_kn_lock_live(of->kn, false);
3270 cgrp->max_descendants = descendants;
3272 cgroup_kn_unlock(of->kn);
3277 static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3279 struct cgroup *cgrp = seq_css(seq)->cgroup;
3280 int depth = READ_ONCE(cgrp->max_depth);
3282 if (depth == INT_MAX)
3283 seq_puts(seq, "max\n");
3285 seq_printf(seq, "%d\n", depth);
3290 static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3291 char *buf, size_t nbytes, loff_t off)
3293 struct cgroup *cgrp;
3297 buf = strstrip(buf);
3298 if (!strcmp(buf, "max")) {
3301 ret = kstrtoint(buf, 0, &depth);
3309 cgrp = cgroup_kn_lock_live(of->kn, false);
3313 cgrp->max_depth = depth;
3315 cgroup_kn_unlock(of->kn);
3320 static int cgroup_events_show(struct seq_file *seq, void *v)
3322 seq_printf(seq, "populated %d\n",
3323 cgroup_is_populated(seq_css(seq)->cgroup));
3327 static int cgroup_stat_show(struct seq_file *seq, void *v)
3329 struct cgroup *cgroup = seq_css(seq)->cgroup;
3331 seq_printf(seq, "nr_descendants %d\n",
3332 cgroup->nr_descendants);
3333 seq_printf(seq, "nr_dying_descendants %d\n",
3334 cgroup->nr_dying_descendants);
3339 static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq,
3340 struct cgroup *cgrp, int ssid)
3342 struct cgroup_subsys *ss = cgroup_subsys[ssid];
3343 struct cgroup_subsys_state *css;
3346 if (!ss->css_extra_stat_show)
3349 css = cgroup_tryget_css(cgrp, ss);
3353 ret = ss->css_extra_stat_show(seq, css);
3358 static int cpu_stat_show(struct seq_file *seq, void *v)
3360 struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
3363 cgroup_stat_show_cputime(seq);
3364 #ifdef CONFIG_CGROUP_SCHED
3365 ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id);
3370 static int cgroup_file_open(struct kernfs_open_file *of)
3372 struct cftype *cft = of->kn->priv;
3375 return cft->open(of);
3379 static void cgroup_file_release(struct kernfs_open_file *of)
3381 struct cftype *cft = of->kn->priv;
3387 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3388 size_t nbytes, loff_t off)
3390 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
3391 struct cgroup *cgrp = of->kn->parent->priv;
3392 struct cftype *cft = of->kn->priv;
3393 struct cgroup_subsys_state *css;
3397 * If namespaces are delegation boundaries, disallow writes to
3398 * files in an non-init namespace root from inside the namespace
3399 * except for the files explicitly marked delegatable -
3400 * cgroup.procs and cgroup.subtree_control.
3402 if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
3403 !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
3404 ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp)
3408 return cft->write(of, buf, nbytes, off);
3411 * kernfs guarantees that a file isn't deleted with operations in
3412 * flight, which means that the matching css is and stays alive and
3413 * doesn't need to be pinned. The RCU locking is not necessary
3414 * either. It's just for the convenience of using cgroup_css().
3417 css = cgroup_css(cgrp, cft->ss);
3420 if (cft->write_u64) {
3421 unsigned long long v;
3422 ret = kstrtoull(buf, 0, &v);
3424 ret = cft->write_u64(css, cft, v);
3425 } else if (cft->write_s64) {
3427 ret = kstrtoll(buf, 0, &v);
3429 ret = cft->write_s64(css, cft, v);
3434 return ret ?: nbytes;
3437 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3439 return seq_cft(seq)->seq_start(seq, ppos);
3442 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3444 return seq_cft(seq)->seq_next(seq, v, ppos);
3447 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3449 if (seq_cft(seq)->seq_stop)
3450 seq_cft(seq)->seq_stop(seq, v);
3453 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3455 struct cftype *cft = seq_cft(m);
3456 struct cgroup_subsys_state *css = seq_css(m);
3459 return cft->seq_show(m, arg);
3462 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3463 else if (cft->read_s64)
3464 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3470 static struct kernfs_ops cgroup_kf_single_ops = {
3471 .atomic_write_len = PAGE_SIZE,
3472 .open = cgroup_file_open,
3473 .release = cgroup_file_release,
3474 .write = cgroup_file_write,
3475 .seq_show = cgroup_seqfile_show,
3478 static struct kernfs_ops cgroup_kf_ops = {
3479 .atomic_write_len = PAGE_SIZE,
3480 .open = cgroup_file_open,
3481 .release = cgroup_file_release,
3482 .write = cgroup_file_write,
3483 .seq_start = cgroup_seqfile_start,
3484 .seq_next = cgroup_seqfile_next,
3485 .seq_stop = cgroup_seqfile_stop,
3486 .seq_show = cgroup_seqfile_show,
3489 /* set uid and gid of cgroup dirs and files to that of the creator */
3490 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3492 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3493 .ia_uid = current_fsuid(),
3494 .ia_gid = current_fsgid(), };
3496 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3497 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3500 return kernfs_setattr(kn, &iattr);
3503 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3506 char name[CGROUP_FILE_NAME_MAX];
3507 struct kernfs_node *kn;
3508 struct lock_class_key *key = NULL;
3511 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3512 key = &cft->lockdep_key;
3514 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3515 cgroup_file_mode(cft), 0, cft->kf_ops, cft,
3520 ret = cgroup_kn_set_ugid(kn);
3526 if (cft->file_offset) {
3527 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3529 spin_lock_irq(&cgroup_file_kn_lock);
3531 spin_unlock_irq(&cgroup_file_kn_lock);
3538 * cgroup_addrm_files - add or remove files to a cgroup directory
3539 * @css: the target css
3540 * @cgrp: the target cgroup (usually css->cgroup)
3541 * @cfts: array of cftypes to be added
3542 * @is_add: whether to add or remove
3544 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3545 * For removals, this function never fails.
3547 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3548 struct cgroup *cgrp, struct cftype cfts[],
3551 struct cftype *cft, *cft_end = NULL;
3554 lockdep_assert_held(&cgroup_mutex);
3557 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3558 /* does cft->flags tell us to skip this file on @cgrp? */
3559 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3561 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3563 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3565 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3569 ret = cgroup_add_file(css, cgrp, cft);
3571 pr_warn("%s: failed to add %s, err=%d\n",
3572 __func__, cft->name, ret);
3578 cgroup_rm_file(cgrp, cft);
3584 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3586 struct cgroup_subsys *ss = cfts[0].ss;
3587 struct cgroup *root = &ss->root->cgrp;
3588 struct cgroup_subsys_state *css;
3591 lockdep_assert_held(&cgroup_mutex);
3593 /* add/rm files for all cgroups created before */
3594 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3595 struct cgroup *cgrp = css->cgroup;
3597 if (!(css->flags & CSS_VISIBLE))
3600 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3606 kernfs_activate(root->kn);
3610 static void cgroup_exit_cftypes(struct cftype *cfts)
3614 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3615 /* free copy for custom atomic_write_len, see init_cftypes() */
3616 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3621 /* revert flags set by cgroup core while adding @cfts */
3622 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
3626 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3630 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3631 struct kernfs_ops *kf_ops;
3633 WARN_ON(cft->ss || cft->kf_ops);
3636 kf_ops = &cgroup_kf_ops;
3638 kf_ops = &cgroup_kf_single_ops;
3641 * Ugh... if @cft wants a custom max_write_len, we need to
3642 * make a copy of kf_ops to set its atomic_write_len.
3644 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3645 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3647 cgroup_exit_cftypes(cfts);
3650 kf_ops->atomic_write_len = cft->max_write_len;
3653 cft->kf_ops = kf_ops;
3660 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3662 lockdep_assert_held(&cgroup_mutex);
3664 if (!cfts || !cfts[0].ss)
3667 list_del(&cfts->node);
3668 cgroup_apply_cftypes(cfts, false);
3669 cgroup_exit_cftypes(cfts);
3674 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3675 * @cfts: zero-length name terminated array of cftypes
3677 * Unregister @cfts. Files described by @cfts are removed from all
3678 * existing cgroups and all future cgroups won't have them either. This
3679 * function can be called anytime whether @cfts' subsys is attached or not.
3681 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3684 int cgroup_rm_cftypes(struct cftype *cfts)
3688 mutex_lock(&cgroup_mutex);
3689 ret = cgroup_rm_cftypes_locked(cfts);
3690 mutex_unlock(&cgroup_mutex);
3695 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3696 * @ss: target cgroup subsystem
3697 * @cfts: zero-length name terminated array of cftypes
3699 * Register @cfts to @ss. Files described by @cfts are created for all
3700 * existing cgroups to which @ss is attached and all future cgroups will
3701 * have them too. This function can be called anytime whether @ss is
3704 * Returns 0 on successful registration, -errno on failure. Note that this
3705 * function currently returns 0 as long as @cfts registration is successful
3706 * even if some file creation attempts on existing cgroups fail.
3708 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3712 if (!cgroup_ssid_enabled(ss->id))
3715 if (!cfts || cfts[0].name[0] == '\0')
3718 ret = cgroup_init_cftypes(ss, cfts);
3722 mutex_lock(&cgroup_mutex);
3724 list_add_tail(&cfts->node, &ss->cfts);
3725 ret = cgroup_apply_cftypes(cfts, true);
3727 cgroup_rm_cftypes_locked(cfts);
3729 mutex_unlock(&cgroup_mutex);
3734 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3735 * @ss: target cgroup subsystem
3736 * @cfts: zero-length name terminated array of cftypes
3738 * Similar to cgroup_add_cftypes() but the added files are only used for
3739 * the default hierarchy.
3741 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3745 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3746 cft->flags |= __CFTYPE_ONLY_ON_DFL;
3747 return cgroup_add_cftypes(ss, cfts);
3751 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3752 * @ss: target cgroup subsystem
3753 * @cfts: zero-length name terminated array of cftypes
3755 * Similar to cgroup_add_cftypes() but the added files are only used for
3756 * the legacy hierarchies.
3758 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3762 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3763 cft->flags |= __CFTYPE_NOT_ON_DFL;
3764 return cgroup_add_cftypes(ss, cfts);
3768 * cgroup_file_notify - generate a file modified event for a cgroup_file
3769 * @cfile: target cgroup_file
3771 * @cfile must have been obtained by setting cftype->file_offset.
3773 void cgroup_file_notify(struct cgroup_file *cfile)
3775 unsigned long flags;
3777 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
3779 kernfs_notify(cfile->kn);
3780 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
3784 * css_next_child - find the next child of a given css
3785 * @pos: the current position (%NULL to initiate traversal)
3786 * @parent: css whose children to walk
3788 * This function returns the next child of @parent and should be called
3789 * under either cgroup_mutex or RCU read lock. The only requirement is
3790 * that @parent and @pos are accessible. The next sibling is guaranteed to
3791 * be returned regardless of their states.
3793 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3794 * css which finished ->css_online() is guaranteed to be visible in the
3795 * future iterations and will stay visible until the last reference is put.
3796 * A css which hasn't finished ->css_online() or already finished
3797 * ->css_offline() may show up during traversal. It's each subsystem's
3798 * responsibility to synchronize against on/offlining.
3800 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
3801 struct cgroup_subsys_state *parent)
3803 struct cgroup_subsys_state *next;
3805 cgroup_assert_mutex_or_rcu_locked();
3808 * @pos could already have been unlinked from the sibling list.
3809 * Once a cgroup is removed, its ->sibling.next is no longer
3810 * updated when its next sibling changes. CSS_RELEASED is set when
3811 * @pos is taken off list, at which time its next pointer is valid,
3812 * and, as releases are serialized, the one pointed to by the next
3813 * pointer is guaranteed to not have started release yet. This
3814 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3815 * critical section, the one pointed to by its next pointer is
3816 * guaranteed to not have finished its RCU grace period even if we
3817 * have dropped rcu_read_lock() inbetween iterations.
3819 * If @pos has CSS_RELEASED set, its next pointer can't be
3820 * dereferenced; however, as each css is given a monotonically
3821 * increasing unique serial number and always appended to the
3822 * sibling list, the next one can be found by walking the parent's
3823 * children until the first css with higher serial number than
3824 * @pos's. While this path can be slower, it happens iff iteration
3825 * races against release and the race window is very small.
3828 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
3829 } else if (likely(!(pos->flags & CSS_RELEASED))) {
3830 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
3832 list_for_each_entry_rcu(next, &parent->children, sibling)
3833 if (next->serial_nr > pos->serial_nr)
3838 * @next, if not pointing to the head, can be dereferenced and is
3841 if (&next->sibling != &parent->children)
3847 * css_next_descendant_pre - find the next descendant for pre-order walk
3848 * @pos: the current position (%NULL to initiate traversal)
3849 * @root: css whose descendants to walk
3851 * To be used by css_for_each_descendant_pre(). Find the next descendant
3852 * to visit for pre-order traversal of @root's descendants. @root is
3853 * included in the iteration and the first node to be visited.
3855 * While this function requires cgroup_mutex or RCU read locking, it
3856 * doesn't require the whole traversal to be contained in a single critical
3857 * section. This function will return the correct next descendant as long
3858 * as both @pos and @root are accessible and @pos is a descendant of @root.
3860 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3861 * css which finished ->css_online() is guaranteed to be visible in the
3862 * future iterations and will stay visible until the last reference is put.
3863 * A css which hasn't finished ->css_online() or already finished
3864 * ->css_offline() may show up during traversal. It's each subsystem's
3865 * responsibility to synchronize against on/offlining.
3867 struct cgroup_subsys_state *
3868 css_next_descendant_pre(struct cgroup_subsys_state *pos,
3869 struct cgroup_subsys_state *root)
3871 struct cgroup_subsys_state *next;
3873 cgroup_assert_mutex_or_rcu_locked();
3875 /* if first iteration, visit @root */
3879 /* visit the first child if exists */
3880 next = css_next_child(NULL, pos);
3884 /* no child, visit my or the closest ancestor's next sibling */
3885 while (pos != root) {
3886 next = css_next_child(pos, pos->parent);
3896 * css_rightmost_descendant - return the rightmost descendant of a css
3897 * @pos: css of interest
3899 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3900 * is returned. This can be used during pre-order traversal to skip
3903 * While this function requires cgroup_mutex or RCU read locking, it
3904 * doesn't require the whole traversal to be contained in a single critical
3905 * section. This function will return the correct rightmost descendant as
3906 * long as @pos is accessible.
3908 struct cgroup_subsys_state *
3909 css_rightmost_descendant(struct cgroup_subsys_state *pos)
3911 struct cgroup_subsys_state *last, *tmp;
3913 cgroup_assert_mutex_or_rcu_locked();
3917 /* ->prev isn't RCU safe, walk ->next till the end */
3919 css_for_each_child(tmp, last)
3926 static struct cgroup_subsys_state *
3927 css_leftmost_descendant(struct cgroup_subsys_state *pos)
3929 struct cgroup_subsys_state *last;
3933 pos = css_next_child(NULL, pos);
3940 * css_next_descendant_post - find the next descendant for post-order walk
3941 * @pos: the current position (%NULL to initiate traversal)
3942 * @root: css whose descendants to walk
3944 * To be used by css_for_each_descendant_post(). Find the next descendant
3945 * to visit for post-order traversal of @root's descendants. @root is
3946 * included in the iteration and the last node to be visited.
3948 * While this function requires cgroup_mutex or RCU read locking, it
3949 * doesn't require the whole traversal to be contained in a single critical
3950 * section. This function will return the correct next descendant as long
3951 * as both @pos and @cgroup are accessible and @pos is a descendant of
3954 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3955 * css which finished ->css_online() is guaranteed to be visible in the
3956 * future iterations and will stay visible until the last reference is put.
3957 * A css which hasn't finished ->css_online() or already finished
3958 * ->css_offline() may show up during traversal. It's each subsystem's
3959 * responsibility to synchronize against on/offlining.
3961 struct cgroup_subsys_state *
3962 css_next_descendant_post(struct cgroup_subsys_state *pos,
3963 struct cgroup_subsys_state *root)
3965 struct cgroup_subsys_state *next;
3967 cgroup_assert_mutex_or_rcu_locked();
3969 /* if first iteration, visit leftmost descendant which may be @root */
3971 return css_leftmost_descendant(root);
3973 /* if we visited @root, we're done */
3977 /* if there's an unvisited sibling, visit its leftmost descendant */
3978 next = css_next_child(pos, pos->parent);
3980 return css_leftmost_descendant(next);
3982 /* no sibling left, visit parent */
3987 * css_has_online_children - does a css have online children
3988 * @css: the target css
3990 * Returns %true if @css has any online children; otherwise, %false. This
3991 * function can be called from any context but the caller is responsible
3992 * for synchronizing against on/offlining as necessary.
3994 bool css_has_online_children(struct cgroup_subsys_state *css)
3996 struct cgroup_subsys_state *child;
4000 css_for_each_child(child, css) {
4001 if (child->flags & CSS_ONLINE) {
4010 static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
4012 struct list_head *l;
4013 struct cgrp_cset_link *link;
4014 struct css_set *cset;
4016 lockdep_assert_held(&css_set_lock);
4018 /* find the next threaded cset */
4019 if (it->tcset_pos) {
4020 l = it->tcset_pos->next;
4022 if (l != it->tcset_head) {
4024 return container_of(l, struct css_set,
4025 threaded_csets_node);
4028 it->tcset_pos = NULL;
4031 /* find the next cset */
4034 if (l == it->cset_head) {
4035 it->cset_pos = NULL;
4040 cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
4042 link = list_entry(l, struct cgrp_cset_link, cset_link);
4048 /* initialize threaded css_set walking */
4049 if (it->flags & CSS_TASK_ITER_THREADED) {
4051 put_css_set_locked(it->cur_dcset);
4052 it->cur_dcset = cset;
4055 it->tcset_head = &cset->threaded_csets;
4056 it->tcset_pos = &cset->threaded_csets;
4063 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
4064 * @it: the iterator to advance
4066 * Advance @it to the next css_set to walk.
4068 static void css_task_iter_advance_css_set(struct css_task_iter *it)
4070 struct css_set *cset;
4072 lockdep_assert_held(&css_set_lock);
4074 /* Advance to the next non-empty css_set */
4076 cset = css_task_iter_next_css_set(it);
4078 it->task_pos = NULL;
4081 } while (!css_set_populated(cset));
4083 if (!list_empty(&cset->tasks))
4084 it->task_pos = cset->tasks.next;
4086 it->task_pos = cset->mg_tasks.next;
4088 it->tasks_head = &cset->tasks;
4089 it->mg_tasks_head = &cset->mg_tasks;
4092 * We don't keep css_sets locked across iteration steps and thus
4093 * need to take steps to ensure that iteration can be resumed after
4094 * the lock is re-acquired. Iteration is performed at two levels -
4095 * css_sets and tasks in them.
4097 * Once created, a css_set never leaves its cgroup lists, so a
4098 * pinned css_set is guaranteed to stay put and we can resume
4099 * iteration afterwards.
4101 * Tasks may leave @cset across iteration steps. This is resolved
4102 * by registering each iterator with the css_set currently being
4103 * walked and making css_set_move_task() advance iterators whose
4104 * next task is leaving.
4107 list_del(&it->iters_node);
4108 put_css_set_locked(it->cur_cset);
4111 it->cur_cset = cset;
4112 list_add(&it->iters_node, &cset->task_iters);
4115 static void css_task_iter_advance(struct css_task_iter *it)
4117 struct list_head *l = it->task_pos;
4119 lockdep_assert_held(&css_set_lock);
4124 * Advance iterator to find next entry. cset->tasks is consumed
4125 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
4130 if (l == it->tasks_head)
4131 l = it->mg_tasks_head->next;
4133 if (l == it->mg_tasks_head)
4134 css_task_iter_advance_css_set(it);
4138 /* if PROCS, skip over tasks which aren't group leaders */
4139 if ((it->flags & CSS_TASK_ITER_PROCS) && it->task_pos &&
4140 !thread_group_leader(list_entry(it->task_pos, struct task_struct,
4146 * css_task_iter_start - initiate task iteration
4147 * @css: the css to walk tasks of
4148 * @flags: CSS_TASK_ITER_* flags
4149 * @it: the task iterator to use
4151 * Initiate iteration through the tasks of @css. The caller can call
4152 * css_task_iter_next() to walk through the tasks until the function
4153 * returns NULL. On completion of iteration, css_task_iter_end() must be
4156 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
4157 struct css_task_iter *it)
4159 /* no one should try to iterate before mounting cgroups */
4160 WARN_ON_ONCE(!use_task_css_set_links);
4162 memset(it, 0, sizeof(*it));
4164 spin_lock_irq(&css_set_lock);
4170 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4172 it->cset_pos = &css->cgroup->cset_links;
4174 it->cset_head = it->cset_pos;
4176 css_task_iter_advance_css_set(it);
4178 spin_unlock_irq(&css_set_lock);
4182 * css_task_iter_next - return the next task for the iterator
4183 * @it: the task iterator being iterated
4185 * The "next" function for task iteration. @it should have been
4186 * initialized via css_task_iter_start(). Returns NULL when the iteration
4189 struct task_struct *css_task_iter_next(struct css_task_iter *it)
4192 put_task_struct(it->cur_task);
4193 it->cur_task = NULL;
4196 spin_lock_irq(&css_set_lock);
4199 it->cur_task = list_entry(it->task_pos, struct task_struct,
4201 get_task_struct(it->cur_task);
4202 css_task_iter_advance(it);
4205 spin_unlock_irq(&css_set_lock);
4207 return it->cur_task;
4211 * css_task_iter_end - finish task iteration
4212 * @it: the task iterator to finish
4214 * Finish task iteration started by css_task_iter_start().
4216 void css_task_iter_end(struct css_task_iter *it)
4219 spin_lock_irq(&css_set_lock);
4220 list_del(&it->iters_node);
4221 put_css_set_locked(it->cur_cset);
4222 spin_unlock_irq(&css_set_lock);
4226 put_css_set(it->cur_dcset);
4229 put_task_struct(it->cur_task);
4232 static void cgroup_procs_release(struct kernfs_open_file *of)
4235 css_task_iter_end(of->priv);
4240 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
4242 struct kernfs_open_file *of = s->private;
4243 struct css_task_iter *it = of->priv;
4245 return css_task_iter_next(it);
4248 static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
4249 unsigned int iter_flags)
4251 struct kernfs_open_file *of = s->private;
4252 struct cgroup *cgrp = seq_css(s)->cgroup;
4253 struct css_task_iter *it = of->priv;
4256 * When a seq_file is seeked, it's always traversed sequentially
4257 * from position 0, so we can simply keep iterating on !0 *pos.
4260 if (WARN_ON_ONCE((*pos)++))
4261 return ERR_PTR(-EINVAL);
4263 it = kzalloc(sizeof(*it), GFP_KERNEL);
4265 return ERR_PTR(-ENOMEM);
4267 css_task_iter_start(&cgrp->self, iter_flags, it);
4268 } else if (!(*pos)++) {
4269 css_task_iter_end(it);
4270 css_task_iter_start(&cgrp->self, iter_flags, it);
4273 return cgroup_procs_next(s, NULL, NULL);
4276 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
4278 struct cgroup *cgrp = seq_css(s)->cgroup;
4281 * All processes of a threaded subtree belong to the domain cgroup
4282 * of the subtree. Only threads can be distributed across the
4283 * subtree. Reject reads on cgroup.procs in the subtree proper.
4284 * They're always empty anyway.
4286 if (cgroup_is_threaded(cgrp))
4287 return ERR_PTR(-EOPNOTSUPP);
4289 return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
4290 CSS_TASK_ITER_THREADED);
4293 static int cgroup_procs_show(struct seq_file *s, void *v)
4295 seq_printf(s, "%d\n", task_pid_vnr(v));
4299 static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
4300 struct cgroup *dst_cgrp,
4301 struct super_block *sb)
4303 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
4304 struct cgroup *com_cgrp = src_cgrp;
4305 struct inode *inode;
4308 lockdep_assert_held(&cgroup_mutex);
4310 /* find the common ancestor */
4311 while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
4312 com_cgrp = cgroup_parent(com_cgrp);
4314 /* %current should be authorized to migrate to the common ancestor */
4315 inode = kernfs_get_inode(sb, com_cgrp->procs_file.kn);
4319 ret = inode_permission(inode, MAY_WRITE);
4325 * If namespaces are delegation boundaries, %current must be able
4326 * to see both source and destination cgroups from its namespace.
4328 if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
4329 (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
4330 !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
4336 static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
4337 char *buf, size_t nbytes, loff_t off)
4339 struct cgroup *src_cgrp, *dst_cgrp;
4340 struct task_struct *task;
4343 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4347 task = cgroup_procs_write_start(buf, true);
4348 ret = PTR_ERR_OR_ZERO(task);
4352 /* find the source cgroup */
4353 spin_lock_irq(&css_set_lock);
4354 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4355 spin_unlock_irq(&css_set_lock);
4357 ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp,
4358 of->file->f_path.dentry->d_sb);
4362 ret = cgroup_attach_task(dst_cgrp, task, true);
4365 cgroup_procs_write_finish(task);
4367 cgroup_kn_unlock(of->kn);
4369 return ret ?: nbytes;
4372 static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
4374 return __cgroup_procs_start(s, pos, 0);
4377 static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
4378 char *buf, size_t nbytes, loff_t off)
4380 struct cgroup *src_cgrp, *dst_cgrp;
4381 struct task_struct *task;
4384 buf = strstrip(buf);
4386 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4390 task = cgroup_procs_write_start(buf, false);
4391 ret = PTR_ERR_OR_ZERO(task);
4395 /* find the source cgroup */
4396 spin_lock_irq(&css_set_lock);
4397 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4398 spin_unlock_irq(&css_set_lock);
4400 /* thread migrations follow the cgroup.procs delegation rule */
4401 ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp,
4402 of->file->f_path.dentry->d_sb);
4406 /* and must be contained in the same domain */
4408 if (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp)
4411 ret = cgroup_attach_task(dst_cgrp, task, false);
4414 cgroup_procs_write_finish(task);
4416 cgroup_kn_unlock(of->kn);
4418 return ret ?: nbytes;
4421 /* cgroup core interface files for the default hierarchy */
4422 static struct cftype cgroup_base_files[] = {
4424 .name = "cgroup.type",
4425 .flags = CFTYPE_NOT_ON_ROOT,
4426 .seq_show = cgroup_type_show,
4427 .write = cgroup_type_write,
4430 .name = "cgroup.procs",
4431 .flags = CFTYPE_NS_DELEGATABLE,
4432 .file_offset = offsetof(struct cgroup, procs_file),
4433 .release = cgroup_procs_release,
4434 .seq_start = cgroup_procs_start,
4435 .seq_next = cgroup_procs_next,
4436 .seq_show = cgroup_procs_show,
4437 .write = cgroup_procs_write,
4440 .name = "cgroup.threads",
4441 .release = cgroup_procs_release,
4442 .seq_start = cgroup_threads_start,
4443 .seq_next = cgroup_procs_next,
4444 .seq_show = cgroup_procs_show,
4445 .write = cgroup_threads_write,
4448 .name = "cgroup.controllers",
4449 .seq_show = cgroup_controllers_show,
4452 .name = "cgroup.subtree_control",
4453 .flags = CFTYPE_NS_DELEGATABLE,
4454 .seq_show = cgroup_subtree_control_show,
4455 .write = cgroup_subtree_control_write,
4458 .name = "cgroup.events",
4459 .flags = CFTYPE_NOT_ON_ROOT,
4460 .file_offset = offsetof(struct cgroup, events_file),
4461 .seq_show = cgroup_events_show,
4464 .name = "cgroup.max.descendants",
4465 .seq_show = cgroup_max_descendants_show,
4466 .write = cgroup_max_descendants_write,
4469 .name = "cgroup.max.depth",
4470 .seq_show = cgroup_max_depth_show,
4471 .write = cgroup_max_depth_write,
4474 .name = "cgroup.stat",
4475 .seq_show = cgroup_stat_show,
4479 .flags = CFTYPE_NOT_ON_ROOT,
4480 .seq_show = cpu_stat_show,
4486 * css destruction is four-stage process.
4488 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4489 * Implemented in kill_css().
4491 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4492 * and thus css_tryget_online() is guaranteed to fail, the css can be
4493 * offlined by invoking offline_css(). After offlining, the base ref is
4494 * put. Implemented in css_killed_work_fn().
4496 * 3. When the percpu_ref reaches zero, the only possible remaining
4497 * accessors are inside RCU read sections. css_release() schedules the
4500 * 4. After the grace period, the css can be freed. Implemented in
4501 * css_free_work_fn().
4503 * It is actually hairier because both step 2 and 4 require process context
4504 * and thus involve punting to css->destroy_work adding two additional
4505 * steps to the already complex sequence.
4507 static void css_free_work_fn(struct work_struct *work)
4509 struct cgroup_subsys_state *css =
4510 container_of(work, struct cgroup_subsys_state, destroy_work);
4511 struct cgroup_subsys *ss = css->ss;
4512 struct cgroup *cgrp = css->cgroup;
4514 percpu_ref_exit(&css->refcnt);
4518 struct cgroup_subsys_state *parent = css->parent;
4522 cgroup_idr_remove(&ss->css_idr, id);
4528 /* cgroup free path */
4529 atomic_dec(&cgrp->root->nr_cgrps);
4530 cgroup1_pidlist_destroy_all(cgrp);
4531 cancel_work_sync(&cgrp->release_agent_work);
4533 if (cgroup_parent(cgrp)) {
4535 * We get a ref to the parent, and put the ref when
4536 * this cgroup is being freed, so it's guaranteed
4537 * that the parent won't be destroyed before its
4540 cgroup_put(cgroup_parent(cgrp));
4541 kernfs_put(cgrp->kn);
4542 if (cgroup_on_dfl(cgrp))
4543 cgroup_stat_exit(cgrp);
4547 * This is root cgroup's refcnt reaching zero,
4548 * which indicates that the root should be
4551 cgroup_destroy_root(cgrp->root);
4556 static void css_free_rcu_fn(struct rcu_head *rcu_head)
4558 struct cgroup_subsys_state *css =
4559 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
4561 INIT_WORK(&css->destroy_work, css_free_work_fn);
4562 queue_work(cgroup_destroy_wq, &css->destroy_work);
4565 static void css_release_work_fn(struct work_struct *work)
4567 struct cgroup_subsys_state *css =
4568 container_of(work, struct cgroup_subsys_state, destroy_work);
4569 struct cgroup_subsys *ss = css->ss;
4570 struct cgroup *cgrp = css->cgroup;
4572 mutex_lock(&cgroup_mutex);
4574 css->flags |= CSS_RELEASED;
4575 list_del_rcu(&css->sibling);
4578 /* css release path */
4579 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
4580 if (ss->css_released)
4581 ss->css_released(css);
4583 struct cgroup *tcgrp;
4585 /* cgroup release path */
4586 trace_cgroup_release(cgrp);
4588 if (cgroup_on_dfl(cgrp))
4589 cgroup_stat_flush(cgrp);
4591 for (tcgrp = cgroup_parent(cgrp); tcgrp;
4592 tcgrp = cgroup_parent(tcgrp))
4593 tcgrp->nr_dying_descendants--;
4595 cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
4599 * There are two control paths which try to determine
4600 * cgroup from dentry without going through kernfs -
4601 * cgroupstats_build() and css_tryget_online_from_dir().
4602 * Those are supported by RCU protecting clearing of
4603 * cgrp->kn->priv backpointer.
4606 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
4609 cgroup_bpf_put(cgrp);
4612 mutex_unlock(&cgroup_mutex);
4614 call_rcu(&css->rcu_head, css_free_rcu_fn);
4617 static void css_release(struct percpu_ref *ref)
4619 struct cgroup_subsys_state *css =
4620 container_of(ref, struct cgroup_subsys_state, refcnt);
4622 INIT_WORK(&css->destroy_work, css_release_work_fn);
4623 queue_work(cgroup_destroy_wq, &css->destroy_work);
4626 static void init_and_link_css(struct cgroup_subsys_state *css,
4627 struct cgroup_subsys *ss, struct cgroup *cgrp)
4629 lockdep_assert_held(&cgroup_mutex);
4631 cgroup_get_live(cgrp);
4633 memset(css, 0, sizeof(*css));
4637 INIT_LIST_HEAD(&css->sibling);
4638 INIT_LIST_HEAD(&css->children);
4639 css->serial_nr = css_serial_nr_next++;
4640 atomic_set(&css->online_cnt, 0);
4642 if (cgroup_parent(cgrp)) {
4643 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
4644 css_get(css->parent);
4647 BUG_ON(cgroup_css(cgrp, ss));
4650 /* invoke ->css_online() on a new CSS and mark it online if successful */
4651 static int online_css(struct cgroup_subsys_state *css)
4653 struct cgroup_subsys *ss = css->ss;
4656 lockdep_assert_held(&cgroup_mutex);
4659 ret = ss->css_online(css);
4661 css->flags |= CSS_ONLINE;
4662 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
4664 atomic_inc(&css->online_cnt);
4666 atomic_inc(&css->parent->online_cnt);
4671 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4672 static void offline_css(struct cgroup_subsys_state *css)
4674 struct cgroup_subsys *ss = css->ss;
4676 lockdep_assert_held(&cgroup_mutex);
4678 if (!(css->flags & CSS_ONLINE))
4681 if (ss->css_offline)
4682 ss->css_offline(css);
4684 css->flags &= ~CSS_ONLINE;
4685 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
4687 wake_up_all(&css->cgroup->offline_waitq);
4691 * css_create - create a cgroup_subsys_state
4692 * @cgrp: the cgroup new css will be associated with
4693 * @ss: the subsys of new css
4695 * Create a new css associated with @cgrp - @ss pair. On success, the new
4696 * css is online and installed in @cgrp. This function doesn't create the
4697 * interface files. Returns 0 on success, -errno on failure.
4699 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
4700 struct cgroup_subsys *ss)
4702 struct cgroup *parent = cgroup_parent(cgrp);
4703 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
4704 struct cgroup_subsys_state *css;
4707 lockdep_assert_held(&cgroup_mutex);
4709 css = ss->css_alloc(parent_css);
4711 css = ERR_PTR(-ENOMEM);
4715 init_and_link_css(css, ss, cgrp);
4717 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
4721 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
4726 /* @css is ready to be brought online now, make it visible */
4727 list_add_tail_rcu(&css->sibling, &parent_css->children);
4728 cgroup_idr_replace(&ss->css_idr, css, css->id);
4730 err = online_css(css);
4734 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4735 cgroup_parent(parent)) {
4736 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4737 current->comm, current->pid, ss->name);
4738 if (!strcmp(ss->name, "memory"))
4739 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4740 ss->warned_broken_hierarchy = true;
4746 list_del_rcu(&css->sibling);
4748 call_rcu(&css->rcu_head, css_free_rcu_fn);
4749 return ERR_PTR(err);
4753 * The returned cgroup is fully initialized including its control mask, but
4754 * it isn't associated with its kernfs_node and doesn't have the control
4757 static struct cgroup *cgroup_create(struct cgroup *parent)
4759 struct cgroup_root *root = parent->root;
4760 struct cgroup *cgrp, *tcgrp;
4761 int level = parent->level + 1;
4764 /* allocate the cgroup and its ID, 0 is reserved for the root */
4765 cgrp = kzalloc(sizeof(*cgrp) +
4766 sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL);
4768 return ERR_PTR(-ENOMEM);
4770 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
4774 if (cgroup_on_dfl(parent)) {
4775 ret = cgroup_stat_init(cgrp);
4777 goto out_cancel_ref;
4781 * Temporarily set the pointer to NULL, so idr_find() won't return
4782 * a half-baked cgroup.
4784 cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL);
4790 init_cgroup_housekeeping(cgrp);
4792 cgrp->self.parent = &parent->self;
4794 cgrp->level = level;
4796 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
4797 cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
4800 tcgrp->nr_descendants++;
4803 if (notify_on_release(parent))
4804 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4806 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4807 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4809 cgrp->self.serial_nr = css_serial_nr_next++;
4811 /* allocation complete, commit to creation */
4812 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
4813 atomic_inc(&root->nr_cgrps);
4814 cgroup_get_live(parent);
4817 * @cgrp is now fully operational. If something fails after this
4818 * point, it'll be released via the normal destruction path.
4820 cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
4823 * On the default hierarchy, a child doesn't automatically inherit
4824 * subtree_control from the parent. Each is configured manually.
4826 if (!cgroup_on_dfl(cgrp))
4827 cgrp->subtree_control = cgroup_control(cgrp);
4830 cgroup_bpf_inherit(cgrp, parent);
4832 cgroup_propagate_control(cgrp);
4837 if (cgroup_on_dfl(parent))
4838 cgroup_stat_exit(cgrp);
4840 percpu_ref_exit(&cgrp->self.refcnt);
4843 return ERR_PTR(ret);
4846 static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
4848 struct cgroup *cgroup;
4852 lockdep_assert_held(&cgroup_mutex);
4854 for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
4855 if (cgroup->nr_descendants >= cgroup->max_descendants)
4858 if (level > cgroup->max_depth)
4869 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
4871 struct cgroup *parent, *cgrp;
4872 struct kernfs_node *kn;
4875 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
4876 if (strchr(name, '\n'))
4879 parent = cgroup_kn_lock_live(parent_kn, false);
4883 if (!cgroup_check_hierarchy_limits(parent)) {
4888 cgrp = cgroup_create(parent);
4890 ret = PTR_ERR(cgrp);
4894 /* create the directory */
4895 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
4903 * This extra ref will be put in cgroup_free_fn() and guarantees
4904 * that @cgrp->kn is always accessible.
4908 ret = cgroup_kn_set_ugid(kn);
4912 ret = css_populate_dir(&cgrp->self);
4916 ret = cgroup_apply_control_enable(cgrp);
4920 trace_cgroup_mkdir(cgrp);
4922 /* let's create and online css's */
4923 kernfs_activate(kn);
4929 cgroup_destroy_locked(cgrp);
4931 cgroup_kn_unlock(parent_kn);
4936 * This is called when the refcnt of a css is confirmed to be killed.
4937 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4938 * initate destruction and put the css ref from kill_css().
4940 static void css_killed_work_fn(struct work_struct *work)
4942 struct cgroup_subsys_state *css =
4943 container_of(work, struct cgroup_subsys_state, destroy_work);
4945 mutex_lock(&cgroup_mutex);
4950 /* @css can't go away while we're holding cgroup_mutex */
4952 } while (css && atomic_dec_and_test(&css->online_cnt));
4954 mutex_unlock(&cgroup_mutex);
4957 /* css kill confirmation processing requires process context, bounce */
4958 static void css_killed_ref_fn(struct percpu_ref *ref)
4960 struct cgroup_subsys_state *css =
4961 container_of(ref, struct cgroup_subsys_state, refcnt);
4963 if (atomic_dec_and_test(&css->online_cnt)) {
4964 INIT_WORK(&css->destroy_work, css_killed_work_fn);
4965 queue_work(cgroup_destroy_wq, &css->destroy_work);
4970 * kill_css - destroy a css
4971 * @css: css to destroy
4973 * This function initiates destruction of @css by removing cgroup interface
4974 * files and putting its base reference. ->css_offline() will be invoked
4975 * asynchronously once css_tryget_online() is guaranteed to fail and when
4976 * the reference count reaches zero, @css will be released.
4978 static void kill_css(struct cgroup_subsys_state *css)
4980 lockdep_assert_held(&cgroup_mutex);
4982 if (css->flags & CSS_DYING)
4985 css->flags |= CSS_DYING;
4988 * This must happen before css is disassociated with its cgroup.
4989 * See seq_css() for details.
4994 * Killing would put the base ref, but we need to keep it alive
4995 * until after ->css_offline().
5000 * cgroup core guarantees that, by the time ->css_offline() is
5001 * invoked, no new css reference will be given out via
5002 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5003 * proceed to offlining css's because percpu_ref_kill() doesn't
5004 * guarantee that the ref is seen as killed on all CPUs on return.
5006 * Use percpu_ref_kill_and_confirm() to get notifications as each
5007 * css is confirmed to be seen as killed on all CPUs.
5009 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5013 * cgroup_destroy_locked - the first stage of cgroup destruction
5014 * @cgrp: cgroup to be destroyed
5016 * css's make use of percpu refcnts whose killing latency shouldn't be
5017 * exposed to userland and are RCU protected. Also, cgroup core needs to
5018 * guarantee that css_tryget_online() won't succeed by the time
5019 * ->css_offline() is invoked. To satisfy all the requirements,
5020 * destruction is implemented in the following two steps.
5022 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5023 * userland visible parts and start killing the percpu refcnts of
5024 * css's. Set up so that the next stage will be kicked off once all
5025 * the percpu refcnts are confirmed to be killed.
5027 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5028 * rest of destruction. Once all cgroup references are gone, the
5029 * cgroup is RCU-freed.
5031 * This function implements s1. After this step, @cgrp is gone as far as
5032 * the userland is concerned and a new cgroup with the same name may be
5033 * created. As cgroup doesn't care about the names internally, this
5034 * doesn't cause any problem.
5036 static int cgroup_destroy_locked(struct cgroup *cgrp)
5037 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5039 struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
5040 struct cgroup_subsys_state *css;
5041 struct cgrp_cset_link *link;
5044 lockdep_assert_held(&cgroup_mutex);
5047 * Only migration can raise populated from zero and we're already
5048 * holding cgroup_mutex.
5050 if (cgroup_is_populated(cgrp))
5054 * Make sure there's no live children. We can't test emptiness of
5055 * ->self.children as dead children linger on it while being
5056 * drained; otherwise, "rmdir parent/child parent" may fail.
5058 if (css_has_online_children(&cgrp->self))
5062 * Mark @cgrp and the associated csets dead. The former prevents
5063 * further task migration and child creation by disabling
5064 * cgroup_lock_live_group(). The latter makes the csets ignored by
5065 * the migration path.
5067 cgrp->self.flags &= ~CSS_ONLINE;
5069 spin_lock_irq(&css_set_lock);
5070 list_for_each_entry(link, &cgrp->cset_links, cset_link)
5071 link->cset->dead = true;
5072 spin_unlock_irq(&css_set_lock);
5074 /* initiate massacre of all css's */
5075 for_each_css(css, ssid, cgrp)
5079 * Remove @cgrp directory along with the base files. @cgrp has an
5080 * extra ref on its kn.
5082 kernfs_remove(cgrp->kn);
5084 if (parent && cgroup_is_threaded(cgrp))
5085 parent->nr_threaded_children--;
5087 for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5088 tcgrp->nr_descendants--;
5089 tcgrp->nr_dying_descendants++;
5092 cgroup1_check_for_release(parent);
5094 /* put the base reference */
5095 percpu_ref_kill(&cgrp->self.refcnt);
5100 int cgroup_rmdir(struct kernfs_node *kn)
5102 struct cgroup *cgrp;
5105 cgrp = cgroup_kn_lock_live(kn, false);
5109 ret = cgroup_destroy_locked(cgrp);
5112 trace_cgroup_rmdir(cgrp);
5114 cgroup_kn_unlock(kn);
5118 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5119 .show_options = cgroup_show_options,
5120 .remount_fs = cgroup_remount,
5121 .mkdir = cgroup_mkdir,
5122 .rmdir = cgroup_rmdir,
5123 .show_path = cgroup_show_path,
5126 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5128 struct cgroup_subsys_state *css;
5130 pr_debug("Initializing cgroup subsys %s\n", ss->name);
5132 mutex_lock(&cgroup_mutex);
5134 idr_init(&ss->css_idr);
5135 INIT_LIST_HEAD(&ss->cfts);
5137 /* Create the root cgroup state for this subsystem */
5138 ss->root = &cgrp_dfl_root;
5139 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5140 /* We don't handle early failures gracefully */
5141 BUG_ON(IS_ERR(css));
5142 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5145 * Root csses are never destroyed and we can't initialize
5146 * percpu_ref during early init. Disable refcnting.
5148 css->flags |= CSS_NO_REF;
5151 /* allocation can't be done safely during early init */
5154 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5155 BUG_ON(css->id < 0);
5158 /* Update the init_css_set to contain a subsys
5159 * pointer to this state - since the subsystem is
5160 * newly registered, all tasks and hence the
5161 * init_css_set is in the subsystem's root cgroup. */
5162 init_css_set.subsys[ss->id] = css;
5164 have_fork_callback |= (bool)ss->fork << ss->id;
5165 have_exit_callback |= (bool)ss->exit << ss->id;
5166 have_free_callback |= (bool)ss->free << ss->id;
5167 have_canfork_callback |= (bool)ss->can_fork << ss->id;
5169 /* At system boot, before all subsystems have been
5170 * registered, no tasks have been forked, so we don't
5171 * need to invoke fork callbacks here. */
5172 BUG_ON(!list_empty(&init_task.tasks));
5174 BUG_ON(online_css(css));
5176 mutex_unlock(&cgroup_mutex);
5180 * cgroup_init_early - cgroup initialization at system boot
5182 * Initialize cgroups at system boot, and initialize any
5183 * subsystems that request early init.
5185 int __init cgroup_init_early(void)
5187 static struct cgroup_sb_opts __initdata opts;
5188 struct cgroup_subsys *ss;
5191 init_cgroup_root(&cgrp_dfl_root, &opts);
5192 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5194 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5196 for_each_subsys(ss, i) {
5197 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5198 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5199 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5201 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5202 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5205 ss->name = cgroup_subsys_name[i];
5206 if (!ss->legacy_name)
5207 ss->legacy_name = cgroup_subsys_name[i];
5210 cgroup_init_subsys(ss, true);
5215 static u16 cgroup_disable_mask __initdata;
5218 * cgroup_init - cgroup initialization
5220 * Register cgroup filesystem and /proc file, and initialize
5221 * any subsystems that didn't request early init.
5223 int __init cgroup_init(void)
5225 struct cgroup_subsys *ss;
5228 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5229 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
5230 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
5231 BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
5236 * The latency of the synchronize_sched() is too high for cgroups,
5237 * avoid it at the cost of forcing all readers into the slow path.
5239 rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5241 get_user_ns(init_cgroup_ns.user_ns);
5243 mutex_lock(&cgroup_mutex);
5246 * Add init_css_set to the hash table so that dfl_root can link to
5249 hash_add(css_set_table, &init_css_set.hlist,
5250 css_set_hash(init_css_set.subsys));
5252 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0, 0));
5254 mutex_unlock(&cgroup_mutex);
5256 for_each_subsys(ss, ssid) {
5257 if (ss->early_init) {
5258 struct cgroup_subsys_state *css =
5259 init_css_set.subsys[ss->id];
5261 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5263 BUG_ON(css->id < 0);
5265 cgroup_init_subsys(ss, false);
5268 list_add_tail(&init_css_set.e_cset_node[ssid],
5269 &cgrp_dfl_root.cgrp.e_csets[ssid]);
5272 * Setting dfl_root subsys_mask needs to consider the
5273 * disabled flag and cftype registration needs kmalloc,
5274 * both of which aren't available during early_init.
5276 if (cgroup_disable_mask & (1 << ssid)) {
5277 static_branch_disable(cgroup_subsys_enabled_key[ssid]);
5278 printk(KERN_INFO "Disabling %s control group subsystem\n",
5283 if (cgroup1_ssid_disabled(ssid))
5284 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5287 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5289 /* implicit controllers must be threaded too */
5290 WARN_ON(ss->implicit_on_dfl && !ss->threaded);
5292 if (ss->implicit_on_dfl)
5293 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5294 else if (!ss->dfl_cftypes)
5295 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5298 cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
5300 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5301 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5303 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5304 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5308 ss->bind(init_css_set.subsys[ssid]);
5310 mutex_lock(&cgroup_mutex);
5311 css_populate_dir(init_css_set.subsys[ssid]);
5312 mutex_unlock(&cgroup_mutex);
5315 /* init_css_set.subsys[] has been updated, re-hash */
5316 hash_del(&init_css_set.hlist);
5317 hash_add(css_set_table, &init_css_set.hlist,
5318 css_set_hash(init_css_set.subsys));
5320 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5321 WARN_ON(register_filesystem(&cgroup_fs_type));
5322 WARN_ON(register_filesystem(&cgroup2_fs_type));
5323 WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations));
5328 static int __init cgroup_wq_init(void)
5331 * There isn't much point in executing destruction path in
5332 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5333 * Use 1 for @max_active.
5335 * We would prefer to do this in cgroup_init() above, but that
5336 * is called before init_workqueues(): so leave this until after.
5338 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5339 BUG_ON(!cgroup_destroy_wq);
5342 core_initcall(cgroup_wq_init);
5344 void cgroup_path_from_kernfs_id(const union kernfs_node_id *id,
5345 char *buf, size_t buflen)
5347 struct kernfs_node *kn;
5349 kn = kernfs_get_node_by_id(cgrp_dfl_root.kf_root, id);
5352 kernfs_path(kn, buf, buflen);
5357 * proc_cgroup_show()
5358 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5359 * - Used for /proc/<pid>/cgroup.
5361 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5362 struct pid *pid, struct task_struct *tsk)
5366 struct cgroup_root *root;
5369 buf = kmalloc(PATH_MAX, GFP_KERNEL);
5373 mutex_lock(&cgroup_mutex);
5374 spin_lock_irq(&css_set_lock);
5376 for_each_root(root) {
5377 struct cgroup_subsys *ss;
5378 struct cgroup *cgrp;
5379 int ssid, count = 0;
5381 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5384 seq_printf(m, "%d:", root->hierarchy_id);
5385 if (root != &cgrp_dfl_root)
5386 for_each_subsys(ss, ssid)
5387 if (root->subsys_mask & (1 << ssid))
5388 seq_printf(m, "%s%s", count++ ? "," : "",
5390 if (strlen(root->name))
5391 seq_printf(m, "%sname=%s", count ? "," : "",
5395 cgrp = task_cgroup_from_root(tsk, root);
5398 * On traditional hierarchies, all zombie tasks show up as
5399 * belonging to the root cgroup. On the default hierarchy,
5400 * while a zombie doesn't show up in "cgroup.procs" and
5401 * thus can't be migrated, its /proc/PID/cgroup keeps
5402 * reporting the cgroup it belonged to before exiting. If
5403 * the cgroup is removed before the zombie is reaped,
5404 * " (deleted)" is appended to the cgroup path.
5406 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
5407 retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5408 current->nsproxy->cgroup_ns);
5409 if (retval >= PATH_MAX)
5410 retval = -ENAMETOOLONG;
5419 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5420 seq_puts(m, " (deleted)\n");
5427 spin_unlock_irq(&css_set_lock);
5428 mutex_unlock(&cgroup_mutex);
5435 * cgroup_fork - initialize cgroup related fields during copy_process()
5436 * @child: pointer to task_struct of forking parent process.
5438 * A task is associated with the init_css_set until cgroup_post_fork()
5439 * attaches it to the parent's css_set. Empty cg_list indicates that
5440 * @child isn't holding reference to its css_set.
5442 void cgroup_fork(struct task_struct *child)
5444 RCU_INIT_POINTER(child->cgroups, &init_css_set);
5445 INIT_LIST_HEAD(&child->cg_list);
5449 * cgroup_can_fork - called on a new task before the process is exposed
5450 * @child: the task in question.
5452 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5453 * returns an error, the fork aborts with that error code. This allows for
5454 * a cgroup subsystem to conditionally allow or deny new forks.
5456 int cgroup_can_fork(struct task_struct *child)
5458 struct cgroup_subsys *ss;
5461 do_each_subsys_mask(ss, i, have_canfork_callback) {
5462 ret = ss->can_fork(child);
5465 } while_each_subsys_mask();
5470 for_each_subsys(ss, j) {
5473 if (ss->cancel_fork)
5474 ss->cancel_fork(child);
5481 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5482 * @child: the task in question
5484 * This calls the cancel_fork() callbacks if a fork failed *after*
5485 * cgroup_can_fork() succeded.
5487 void cgroup_cancel_fork(struct task_struct *child)
5489 struct cgroup_subsys *ss;
5492 for_each_subsys(ss, i)
5493 if (ss->cancel_fork)
5494 ss->cancel_fork(child);
5498 * cgroup_post_fork - called on a new task after adding it to the task list
5499 * @child: the task in question
5501 * Adds the task to the list running through its css_set if necessary and
5502 * call the subsystem fork() callbacks. Has to be after the task is
5503 * visible on the task list in case we race with the first call to
5504 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5507 void cgroup_post_fork(struct task_struct *child)
5509 struct cgroup_subsys *ss;
5513 * This may race against cgroup_enable_task_cg_lists(). As that
5514 * function sets use_task_css_set_links before grabbing
5515 * tasklist_lock and we just went through tasklist_lock to add
5516 * @child, it's guaranteed that either we see the set
5517 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5518 * @child during its iteration.
5520 * If we won the race, @child is associated with %current's
5521 * css_set. Grabbing css_set_lock guarantees both that the
5522 * association is stable, and, on completion of the parent's
5523 * migration, @child is visible in the source of migration or
5524 * already in the destination cgroup. This guarantee is necessary
5525 * when implementing operations which need to migrate all tasks of
5526 * a cgroup to another.
5528 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
5529 * will remain in init_css_set. This is safe because all tasks are
5530 * in the init_css_set before cg_links is enabled and there's no
5531 * operation which transfers all tasks out of init_css_set.
5533 if (use_task_css_set_links) {
5534 struct css_set *cset;
5536 spin_lock_irq(&css_set_lock);
5537 cset = task_css_set(current);
5538 if (list_empty(&child->cg_list)) {
5541 css_set_move_task(child, NULL, cset, false);
5543 spin_unlock_irq(&css_set_lock);
5547 * Call ss->fork(). This must happen after @child is linked on
5548 * css_set; otherwise, @child might change state between ->fork()
5549 * and addition to css_set.
5551 do_each_subsys_mask(ss, i, have_fork_callback) {
5553 } while_each_subsys_mask();
5557 * cgroup_exit - detach cgroup from exiting task
5558 * @tsk: pointer to task_struct of exiting process
5560 * Description: Detach cgroup from @tsk and release it.
5562 * Note that cgroups marked notify_on_release force every task in
5563 * them to take the global cgroup_mutex mutex when exiting.
5564 * This could impact scaling on very large systems. Be reluctant to
5565 * use notify_on_release cgroups where very high task exit scaling
5566 * is required on large systems.
5568 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5569 * call cgroup_exit() while the task is still competent to handle
5570 * notify_on_release(), then leave the task attached to the root cgroup in
5571 * each hierarchy for the remainder of its exit. No need to bother with
5572 * init_css_set refcnting. init_css_set never goes away and we can't race
5573 * with migration path - PF_EXITING is visible to migration path.
5575 void cgroup_exit(struct task_struct *tsk)
5577 struct cgroup_subsys *ss;
5578 struct css_set *cset;
5582 * Unlink from @tsk from its css_set. As migration path can't race
5583 * with us, we can check css_set and cg_list without synchronization.
5585 cset = task_css_set(tsk);
5587 if (!list_empty(&tsk->cg_list)) {
5588 spin_lock_irq(&css_set_lock);
5589 css_set_move_task(tsk, cset, NULL, false);
5591 spin_unlock_irq(&css_set_lock);
5596 /* see cgroup_post_fork() for details */
5597 do_each_subsys_mask(ss, i, have_exit_callback) {
5599 } while_each_subsys_mask();
5602 void cgroup_free(struct task_struct *task)
5604 struct css_set *cset = task_css_set(task);
5605 struct cgroup_subsys *ss;
5608 do_each_subsys_mask(ss, ssid, have_free_callback) {
5610 } while_each_subsys_mask();
5615 static int __init cgroup_disable(char *str)
5617 struct cgroup_subsys *ss;
5621 while ((token = strsep(&str, ",")) != NULL) {
5625 for_each_subsys(ss, i) {
5626 if (strcmp(token, ss->name) &&
5627 strcmp(token, ss->legacy_name))
5629 cgroup_disable_mask |= 1 << i;
5634 __setup("cgroup_disable=", cgroup_disable);
5637 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5638 * @dentry: directory dentry of interest
5639 * @ss: subsystem of interest
5641 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5642 * to get the corresponding css and return it. If such css doesn't exist
5643 * or can't be pinned, an ERR_PTR value is returned.
5645 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
5646 struct cgroup_subsys *ss)
5648 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
5649 struct file_system_type *s_type = dentry->d_sb->s_type;
5650 struct cgroup_subsys_state *css = NULL;
5651 struct cgroup *cgrp;
5653 /* is @dentry a cgroup dir? */
5654 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
5655 !kn || kernfs_type(kn) != KERNFS_DIR)
5656 return ERR_PTR(-EBADF);
5661 * This path doesn't originate from kernfs and @kn could already
5662 * have been or be removed at any point. @kn->priv is RCU
5663 * protected for this access. See css_release_work_fn() for details.
5665 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
5667 css = cgroup_css(cgrp, ss);
5669 if (!css || !css_tryget_online(css))
5670 css = ERR_PTR(-ENOENT);
5677 * css_from_id - lookup css by id
5678 * @id: the cgroup id
5679 * @ss: cgroup subsys to be looked into
5681 * Returns the css if there's valid one with @id, otherwise returns NULL.
5682 * Should be called under rcu_read_lock().
5684 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
5686 WARN_ON_ONCE(!rcu_read_lock_held());
5687 return idr_find(&ss->css_idr, id);
5691 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
5692 * @path: path on the default hierarchy
5694 * Find the cgroup at @path on the default hierarchy, increment its
5695 * reference count and return it. Returns pointer to the found cgroup on
5696 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
5697 * if @path points to a non-directory.
5699 struct cgroup *cgroup_get_from_path(const char *path)
5701 struct kernfs_node *kn;
5702 struct cgroup *cgrp;
5704 mutex_lock(&cgroup_mutex);
5706 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
5708 if (kernfs_type(kn) == KERNFS_DIR) {
5710 cgroup_get_live(cgrp);
5712 cgrp = ERR_PTR(-ENOTDIR);
5716 cgrp = ERR_PTR(-ENOENT);
5719 mutex_unlock(&cgroup_mutex);
5722 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
5725 * cgroup_get_from_fd - get a cgroup pointer from a fd
5726 * @fd: fd obtained by open(cgroup2_dir)
5728 * Find the cgroup from a fd which should be obtained
5729 * by opening a cgroup directory. Returns a pointer to the
5730 * cgroup on success. ERR_PTR is returned if the cgroup
5733 struct cgroup *cgroup_get_from_fd(int fd)
5735 struct cgroup_subsys_state *css;
5736 struct cgroup *cgrp;
5741 return ERR_PTR(-EBADF);
5743 css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
5746 return ERR_CAST(css);
5749 if (!cgroup_on_dfl(cgrp)) {
5751 return ERR_PTR(-EBADF);
5756 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
5759 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
5760 * definition in cgroup-defs.h.
5762 #ifdef CONFIG_SOCK_CGROUP_DATA
5764 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
5766 DEFINE_SPINLOCK(cgroup_sk_update_lock);
5767 static bool cgroup_sk_alloc_disabled __read_mostly;
5769 void cgroup_sk_alloc_disable(void)
5771 if (cgroup_sk_alloc_disabled)
5773 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
5774 cgroup_sk_alloc_disabled = true;
5779 #define cgroup_sk_alloc_disabled false
5783 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
5785 if (cgroup_sk_alloc_disabled)
5788 /* Socket clone path */
5791 * We might be cloning a socket which is left in an empty
5792 * cgroup and the cgroup might have already been rmdir'd.
5793 * Don't use cgroup_get_live().
5795 cgroup_get(sock_cgroup_ptr(skcd));
5802 struct css_set *cset;
5804 cset = task_css_set(current);
5805 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
5806 skcd->val = (unsigned long)cset->dfl_cgrp;
5815 void cgroup_sk_free(struct sock_cgroup_data *skcd)
5817 cgroup_put(sock_cgroup_ptr(skcd));
5820 #endif /* CONFIG_SOCK_CGROUP_DATA */
5822 #ifdef CONFIG_CGROUP_BPF
5823 int cgroup_bpf_update(struct cgroup *cgrp, struct bpf_prog *prog,
5824 enum bpf_attach_type type, bool overridable)
5826 struct cgroup *parent = cgroup_parent(cgrp);
5829 mutex_lock(&cgroup_mutex);
5830 ret = __cgroup_bpf_update(cgrp, parent, prog, type, overridable);
5831 mutex_unlock(&cgroup_mutex);
5834 #endif /* CONFIG_CGROUP_BPF */
5837 static ssize_t show_delegatable_files(struct cftype *files, char *buf,
5838 ssize_t size, const char *prefix)
5843 for (cft = files; cft && cft->name[0] != '\0'; cft++) {
5844 if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
5848 ret += snprintf(buf + ret, size - ret, "%s.", prefix);
5850 ret += snprintf(buf + ret, size - ret, "%s\n", cft->name);
5852 if (unlikely(ret >= size)) {
5861 static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr,
5864 struct cgroup_subsys *ss;
5868 ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret,
5871 for_each_subsys(ss, ssid)
5872 ret += show_delegatable_files(ss->dfl_cftypes, buf + ret,
5874 cgroup_subsys_name[ssid]);
5878 static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate);
5880 static struct attribute *cgroup_sysfs_attrs[] = {
5881 &cgroup_delegate_attr.attr,
5885 static const struct attribute_group cgroup_sysfs_attr_group = {
5886 .attrs = cgroup_sysfs_attrs,
5890 static int __init cgroup_sysfs_init(void)
5892 return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group);
5894 subsys_initcall(cgroup_sysfs_init);
5895 #endif /* CONFIG_SYSFS */