2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
31 #include "cgroup-internal.h"
33 #include <linux/cred.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/magic.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/sched/task.h>
45 #include <linux/slab.h>
46 #include <linux/spinlock.h>
47 #include <linux/percpu-rwsem.h>
48 #include <linux/string.h>
49 #include <linux/hashtable.h>
50 #include <linux/idr.h>
51 #include <linux/kthread.h>
52 #include <linux/atomic.h>
53 #include <linux/cpuset.h>
54 #include <linux/proc_ns.h>
55 #include <linux/nsproxy.h>
56 #include <linux/file.h>
57 #include <linux/fs_parser.h>
58 #include <linux/sched/cputime.h>
59 #include <linux/psi.h>
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/cgroup.h>
65 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
67 /* let's not notify more than 100 times per second */
68 #define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100)
71 * To avoid confusing the compiler (and generating warnings) with code
72 * that attempts to access what would be a 0-element array (i.e. sized
73 * to a potentially empty array when CGROUP_SUBSYS_COUNT == 0), this
74 * constant expression can be added.
76 #define CGROUP_HAS_SUBSYS_CONFIG (CGROUP_SUBSYS_COUNT > 0)
79 * cgroup_mutex is the master lock. Any modification to cgroup or its
80 * hierarchy must be performed while holding it.
82 * css_set_lock protects task->cgroups pointer, the list of css_set
83 * objects, and the chain of tasks off each css_set.
85 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
86 * cgroup.h can use them for lockdep annotations.
88 DEFINE_MUTEX(cgroup_mutex);
89 DEFINE_SPINLOCK(css_set_lock);
91 #ifdef CONFIG_PROVE_RCU
92 EXPORT_SYMBOL_GPL(cgroup_mutex);
93 EXPORT_SYMBOL_GPL(css_set_lock);
96 DEFINE_SPINLOCK(trace_cgroup_path_lock);
97 char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
98 bool cgroup_debug __read_mostly;
101 * Protects cgroup_idr and css_idr so that IDs can be released without
102 * grabbing cgroup_mutex.
104 static DEFINE_SPINLOCK(cgroup_idr_lock);
107 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
108 * against file removal/re-creation across css hiding.
110 static DEFINE_SPINLOCK(cgroup_file_kn_lock);
112 DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem);
114 #define cgroup_assert_mutex_or_rcu_locked() \
115 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
116 !lockdep_is_held(&cgroup_mutex), \
117 "cgroup_mutex or RCU read lock required");
120 * cgroup destruction makes heavy use of work items and there can be a lot
121 * of concurrent destructions. Use a separate workqueue so that cgroup
122 * destruction work items don't end up filling up max_active of system_wq
123 * which may lead to deadlock.
125 static struct workqueue_struct *cgroup_destroy_wq;
127 /* generate an array of cgroup subsystem pointers */
128 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
129 struct cgroup_subsys *cgroup_subsys[] = {
130 #include <linux/cgroup_subsys.h>
134 /* array of cgroup subsystem names */
135 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
136 static const char *cgroup_subsys_name[] = {
137 #include <linux/cgroup_subsys.h>
141 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
143 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
144 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
145 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
146 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
147 #include <linux/cgroup_subsys.h>
150 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
151 static struct static_key_true *cgroup_subsys_enabled_key[] = {
152 #include <linux/cgroup_subsys.h>
156 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
157 static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
158 #include <linux/cgroup_subsys.h>
162 static DEFINE_PER_CPU(struct cgroup_rstat_cpu, cgrp_dfl_root_rstat_cpu);
164 /* the default hierarchy */
165 struct cgroup_root cgrp_dfl_root = { .cgrp.rstat_cpu = &cgrp_dfl_root_rstat_cpu };
166 EXPORT_SYMBOL_GPL(cgrp_dfl_root);
169 * The default hierarchy always exists but is hidden until mounted for the
170 * first time. This is for backward compatibility.
172 static bool cgrp_dfl_visible;
174 /* some controllers are not supported in the default hierarchy */
175 static u16 cgrp_dfl_inhibit_ss_mask;
177 /* some controllers are implicitly enabled on the default hierarchy */
178 static u16 cgrp_dfl_implicit_ss_mask;
180 /* some controllers can be threaded on the default hierarchy */
181 static u16 cgrp_dfl_threaded_ss_mask;
183 /* The list of hierarchy roots */
184 LIST_HEAD(cgroup_roots);
185 static int cgroup_root_count;
187 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
188 static DEFINE_IDR(cgroup_hierarchy_idr);
191 * Assign a monotonically increasing serial number to csses. It guarantees
192 * cgroups with bigger numbers are newer than those with smaller numbers.
193 * Also, as csses are always appended to the parent's ->children list, it
194 * guarantees that sibling csses are always sorted in the ascending serial
195 * number order on the list. Protected by cgroup_mutex.
197 static u64 css_serial_nr_next = 1;
200 * These bitmasks identify subsystems with specific features to avoid
201 * having to do iterative checks repeatedly.
203 static u16 have_fork_callback __read_mostly;
204 static u16 have_exit_callback __read_mostly;
205 static u16 have_release_callback __read_mostly;
206 static u16 have_canfork_callback __read_mostly;
208 /* cgroup namespace for init task */
209 struct cgroup_namespace init_cgroup_ns = {
210 .ns.count = REFCOUNT_INIT(2),
211 .user_ns = &init_user_ns,
212 .ns.ops = &cgroupns_operations,
213 .ns.inum = PROC_CGROUP_INIT_INO,
214 .root_cset = &init_css_set,
217 static struct file_system_type cgroup2_fs_type;
218 static struct cftype cgroup_base_files[];
220 /* cgroup optional features */
221 enum cgroup_opt_features {
223 OPT_FEATURE_PRESSURE,
228 static const char *cgroup_opt_feature_names[OPT_FEATURE_COUNT] = {
234 static u16 cgroup_feature_disable_mask __read_mostly;
236 static int cgroup_apply_control(struct cgroup *cgrp);
237 static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
238 static void css_task_iter_skip(struct css_task_iter *it,
239 struct task_struct *task);
240 static int cgroup_destroy_locked(struct cgroup *cgrp);
241 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
242 struct cgroup_subsys *ss);
243 static void css_release(struct percpu_ref *ref);
244 static void kill_css(struct cgroup_subsys_state *css);
245 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
246 struct cgroup *cgrp, struct cftype cfts[],
250 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
251 * @ssid: subsys ID of interest
253 * cgroup_subsys_enabled() can only be used with literal subsys names which
254 * is fine for individual subsystems but unsuitable for cgroup core. This
255 * is slower static_key_enabled() based test indexed by @ssid.
257 bool cgroup_ssid_enabled(int ssid)
259 if (!CGROUP_HAS_SUBSYS_CONFIG)
262 return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
266 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
267 * @cgrp: the cgroup of interest
269 * The default hierarchy is the v2 interface of cgroup and this function
270 * can be used to test whether a cgroup is on the default hierarchy for
271 * cases where a subsystem should behave differently depending on the
274 * List of changed behaviors:
276 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
277 * and "name" are disallowed.
279 * - When mounting an existing superblock, mount options should match.
281 * - Remount is disallowed.
283 * - rename(2) is disallowed.
285 * - "tasks" is removed. Everything should be at process granularity. Use
286 * "cgroup.procs" instead.
288 * - "cgroup.procs" is not sorted. pids will be unique unless they got
289 * recycled in-between reads.
291 * - "release_agent" and "notify_on_release" are removed. Replacement
292 * notification mechanism will be implemented.
294 * - "cgroup.clone_children" is removed.
296 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
297 * and its descendants contain no task; otherwise, 1. The file also
298 * generates kernfs notification which can be monitored through poll and
299 * [di]notify when the value of the file changes.
301 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
302 * take masks of ancestors with non-empty cpus/mems, instead of being
303 * moved to an ancestor.
305 * - cpuset: a task can be moved into an empty cpuset, and again it takes
306 * masks of ancestors.
308 * - blkcg: blk-throttle becomes properly hierarchical.
310 * - debug: disallowed on the default hierarchy.
312 bool cgroup_on_dfl(const struct cgroup *cgrp)
314 return cgrp->root == &cgrp_dfl_root;
317 /* IDR wrappers which synchronize using cgroup_idr_lock */
318 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
323 idr_preload(gfp_mask);
324 spin_lock_bh(&cgroup_idr_lock);
325 ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
326 spin_unlock_bh(&cgroup_idr_lock);
331 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
335 spin_lock_bh(&cgroup_idr_lock);
336 ret = idr_replace(idr, ptr, id);
337 spin_unlock_bh(&cgroup_idr_lock);
341 static void cgroup_idr_remove(struct idr *idr, int id)
343 spin_lock_bh(&cgroup_idr_lock);
345 spin_unlock_bh(&cgroup_idr_lock);
348 static bool cgroup_has_tasks(struct cgroup *cgrp)
350 return cgrp->nr_populated_csets;
353 bool cgroup_is_threaded(struct cgroup *cgrp)
355 return cgrp->dom_cgrp != cgrp;
358 /* can @cgrp host both domain and threaded children? */
359 static bool cgroup_is_mixable(struct cgroup *cgrp)
362 * Root isn't under domain level resource control exempting it from
363 * the no-internal-process constraint, so it can serve as a thread
364 * root and a parent of resource domains at the same time.
366 return !cgroup_parent(cgrp);
369 /* can @cgrp become a thread root? Should always be true for a thread root */
370 static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
372 /* mixables don't care */
373 if (cgroup_is_mixable(cgrp))
376 /* domain roots can't be nested under threaded */
377 if (cgroup_is_threaded(cgrp))
380 /* can only have either domain or threaded children */
381 if (cgrp->nr_populated_domain_children)
384 /* and no domain controllers can be enabled */
385 if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
391 /* is @cgrp root of a threaded subtree? */
392 bool cgroup_is_thread_root(struct cgroup *cgrp)
394 /* thread root should be a domain */
395 if (cgroup_is_threaded(cgrp))
398 /* a domain w/ threaded children is a thread root */
399 if (cgrp->nr_threaded_children)
403 * A domain which has tasks and explicit threaded controllers
404 * enabled is a thread root.
406 if (cgroup_has_tasks(cgrp) &&
407 (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
413 /* a domain which isn't connected to the root w/o brekage can't be used */
414 static bool cgroup_is_valid_domain(struct cgroup *cgrp)
416 /* the cgroup itself can be a thread root */
417 if (cgroup_is_threaded(cgrp))
420 /* but the ancestors can't be unless mixable */
421 while ((cgrp = cgroup_parent(cgrp))) {
422 if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
424 if (cgroup_is_threaded(cgrp))
431 /* subsystems visibly enabled on a cgroup */
432 static u16 cgroup_control(struct cgroup *cgrp)
434 struct cgroup *parent = cgroup_parent(cgrp);
435 u16 root_ss_mask = cgrp->root->subsys_mask;
438 u16 ss_mask = parent->subtree_control;
440 /* threaded cgroups can only have threaded controllers */
441 if (cgroup_is_threaded(cgrp))
442 ss_mask &= cgrp_dfl_threaded_ss_mask;
446 if (cgroup_on_dfl(cgrp))
447 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
448 cgrp_dfl_implicit_ss_mask);
452 /* subsystems enabled on a cgroup */
453 static u16 cgroup_ss_mask(struct cgroup *cgrp)
455 struct cgroup *parent = cgroup_parent(cgrp);
458 u16 ss_mask = parent->subtree_ss_mask;
460 /* threaded cgroups can only have threaded controllers */
461 if (cgroup_is_threaded(cgrp))
462 ss_mask &= cgrp_dfl_threaded_ss_mask;
466 return cgrp->root->subsys_mask;
470 * cgroup_css - obtain a cgroup's css for the specified subsystem
471 * @cgrp: the cgroup of interest
472 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
474 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
475 * function must be called either under cgroup_mutex or rcu_read_lock() and
476 * the caller is responsible for pinning the returned css if it wants to
477 * keep accessing it outside the said locks. This function may return
478 * %NULL if @cgrp doesn't have @subsys_id enabled.
480 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
481 struct cgroup_subsys *ss)
483 if (CGROUP_HAS_SUBSYS_CONFIG && ss)
484 return rcu_dereference_check(cgrp->subsys[ss->id],
485 lockdep_is_held(&cgroup_mutex));
491 * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
492 * @cgrp: the cgroup of interest
493 * @ss: the subsystem of interest
495 * Find and get @cgrp's css associated with @ss. If the css doesn't exist
496 * or is offline, %NULL is returned.
498 static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp,
499 struct cgroup_subsys *ss)
501 struct cgroup_subsys_state *css;
504 css = cgroup_css(cgrp, ss);
505 if (css && !css_tryget_online(css))
513 * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
514 * @cgrp: the cgroup of interest
515 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
517 * Similar to cgroup_css() but returns the effective css, which is defined
518 * as the matching css of the nearest ancestor including self which has @ss
519 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
520 * function is guaranteed to return non-NULL css.
522 static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp,
523 struct cgroup_subsys *ss)
525 lockdep_assert_held(&cgroup_mutex);
531 * This function is used while updating css associations and thus
532 * can't test the csses directly. Test ss_mask.
534 while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
535 cgrp = cgroup_parent(cgrp);
540 return cgroup_css(cgrp, ss);
544 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
545 * @cgrp: the cgroup of interest
546 * @ss: the subsystem of interest
548 * Find and get the effective css of @cgrp for @ss. The effective css is
549 * defined as the matching css of the nearest ancestor including self which
550 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
551 * the root css is returned, so this function always returns a valid css.
553 * The returned css is not guaranteed to be online, and therefore it is the
554 * callers responsibility to try get a reference for it.
556 struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
557 struct cgroup_subsys *ss)
559 struct cgroup_subsys_state *css;
561 if (!CGROUP_HAS_SUBSYS_CONFIG)
565 css = cgroup_css(cgrp, ss);
569 cgrp = cgroup_parent(cgrp);
572 return init_css_set.subsys[ss->id];
576 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
577 * @cgrp: the cgroup of interest
578 * @ss: the subsystem of interest
580 * Find and get the effective css of @cgrp for @ss. The effective css is
581 * defined as the matching css of the nearest ancestor including self which
582 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
583 * the root css is returned, so this function always returns a valid css.
584 * The returned css must be put using css_put().
586 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
587 struct cgroup_subsys *ss)
589 struct cgroup_subsys_state *css;
591 if (!CGROUP_HAS_SUBSYS_CONFIG)
597 css = cgroup_css(cgrp, ss);
599 if (css && css_tryget_online(css))
601 cgrp = cgroup_parent(cgrp);
604 css = init_css_set.subsys[ss->id];
610 EXPORT_SYMBOL_GPL(cgroup_get_e_css);
612 static void cgroup_get_live(struct cgroup *cgrp)
614 WARN_ON_ONCE(cgroup_is_dead(cgrp));
615 css_get(&cgrp->self);
619 * __cgroup_task_count - count the number of tasks in a cgroup. The caller
620 * is responsible for taking the css_set_lock.
621 * @cgrp: the cgroup in question
623 int __cgroup_task_count(const struct cgroup *cgrp)
626 struct cgrp_cset_link *link;
628 lockdep_assert_held(&css_set_lock);
630 list_for_each_entry(link, &cgrp->cset_links, cset_link)
631 count += link->cset->nr_tasks;
637 * cgroup_task_count - count the number of tasks in a cgroup.
638 * @cgrp: the cgroup in question
640 int cgroup_task_count(const struct cgroup *cgrp)
644 spin_lock_irq(&css_set_lock);
645 count = __cgroup_task_count(cgrp);
646 spin_unlock_irq(&css_set_lock);
651 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
653 struct cgroup *cgrp = of->kn->parent->priv;
654 struct cftype *cft = of_cft(of);
657 * This is open and unprotected implementation of cgroup_css().
658 * seq_css() is only called from a kernfs file operation which has
659 * an active reference on the file. Because all the subsystem
660 * files are drained before a css is disassociated with a cgroup,
661 * the matching css from the cgroup's subsys table is guaranteed to
662 * be and stay valid until the enclosing operation is complete.
664 if (CGROUP_HAS_SUBSYS_CONFIG && cft->ss)
665 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
669 EXPORT_SYMBOL_GPL(of_css);
672 * for_each_css - iterate all css's of a cgroup
673 * @css: the iteration cursor
674 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
675 * @cgrp: the target cgroup to iterate css's of
677 * Should be called under cgroup_[tree_]mutex.
679 #define for_each_css(css, ssid, cgrp) \
680 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
681 if (!((css) = rcu_dereference_check( \
682 (cgrp)->subsys[(ssid)], \
683 lockdep_is_held(&cgroup_mutex)))) { } \
687 * for_each_e_css - iterate all effective css's of a cgroup
688 * @css: the iteration cursor
689 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
690 * @cgrp: the target cgroup to iterate css's of
692 * Should be called under cgroup_[tree_]mutex.
694 #define for_each_e_css(css, ssid, cgrp) \
695 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
696 if (!((css) = cgroup_e_css_by_mask(cgrp, \
697 cgroup_subsys[(ssid)]))) \
702 * do_each_subsys_mask - filter for_each_subsys with a bitmask
703 * @ss: the iteration cursor
704 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
705 * @ss_mask: the bitmask
707 * The block will only run for cases where the ssid-th bit (1 << ssid) of
710 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
711 unsigned long __ss_mask = (ss_mask); \
712 if (!CGROUP_HAS_SUBSYS_CONFIG) { \
716 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
717 (ss) = cgroup_subsys[ssid]; \
720 #define while_each_subsys_mask() \
725 /* iterate over child cgrps, lock should be held throughout iteration */
726 #define cgroup_for_each_live_child(child, cgrp) \
727 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
728 if (({ lockdep_assert_held(&cgroup_mutex); \
729 cgroup_is_dead(child); })) \
733 /* walk live descendants in pre order */
734 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
735 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
736 if (({ lockdep_assert_held(&cgroup_mutex); \
737 (dsct) = (d_css)->cgroup; \
738 cgroup_is_dead(dsct); })) \
742 /* walk live descendants in postorder */
743 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
744 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
745 if (({ lockdep_assert_held(&cgroup_mutex); \
746 (dsct) = (d_css)->cgroup; \
747 cgroup_is_dead(dsct); })) \
752 * The default css_set - used by init and its children prior to any
753 * hierarchies being mounted. It contains a pointer to the root state
754 * for each subsystem. Also used to anchor the list of css_sets. Not
755 * reference-counted, to improve performance when child cgroups
756 * haven't been created.
758 struct css_set init_css_set = {
759 .refcount = REFCOUNT_INIT(1),
760 .dom_cset = &init_css_set,
761 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
762 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
763 .dying_tasks = LIST_HEAD_INIT(init_css_set.dying_tasks),
764 .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
765 .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets),
766 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
767 .mg_src_preload_node = LIST_HEAD_INIT(init_css_set.mg_src_preload_node),
768 .mg_dst_preload_node = LIST_HEAD_INIT(init_css_set.mg_dst_preload_node),
769 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
772 * The following field is re-initialized when this cset gets linked
773 * in cgroup_init(). However, let's initialize the field
774 * statically too so that the default cgroup can be accessed safely
777 .dfl_cgrp = &cgrp_dfl_root.cgrp,
780 static int css_set_count = 1; /* 1 for init_css_set */
782 static bool css_set_threaded(struct css_set *cset)
784 return cset->dom_cset != cset;
788 * css_set_populated - does a css_set contain any tasks?
789 * @cset: target css_set
791 * css_set_populated() should be the same as !!cset->nr_tasks at steady
792 * state. However, css_set_populated() can be called while a task is being
793 * added to or removed from the linked list before the nr_tasks is
794 * properly updated. Hence, we can't just look at ->nr_tasks here.
796 static bool css_set_populated(struct css_set *cset)
798 lockdep_assert_held(&css_set_lock);
800 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
804 * cgroup_update_populated - update the populated count of a cgroup
805 * @cgrp: the target cgroup
806 * @populated: inc or dec populated count
808 * One of the css_sets associated with @cgrp is either getting its first
809 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
810 * count is propagated towards root so that a given cgroup's
811 * nr_populated_children is zero iff none of its descendants contain any
814 * @cgrp's interface file "cgroup.populated" is zero if both
815 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
816 * 1 otherwise. When the sum changes from or to zero, userland is notified
817 * that the content of the interface file has changed. This can be used to
818 * detect when @cgrp and its descendants become populated or empty.
820 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
822 struct cgroup *child = NULL;
823 int adj = populated ? 1 : -1;
825 lockdep_assert_held(&css_set_lock);
828 bool was_populated = cgroup_is_populated(cgrp);
831 cgrp->nr_populated_csets += adj;
833 if (cgroup_is_threaded(child))
834 cgrp->nr_populated_threaded_children += adj;
836 cgrp->nr_populated_domain_children += adj;
839 if (was_populated == cgroup_is_populated(cgrp))
842 cgroup1_check_for_release(cgrp);
843 TRACE_CGROUP_PATH(notify_populated, cgrp,
844 cgroup_is_populated(cgrp));
845 cgroup_file_notify(&cgrp->events_file);
848 cgrp = cgroup_parent(cgrp);
853 * css_set_update_populated - update populated state of a css_set
854 * @cset: target css_set
855 * @populated: whether @cset is populated or depopulated
857 * @cset is either getting the first task or losing the last. Update the
858 * populated counters of all associated cgroups accordingly.
860 static void css_set_update_populated(struct css_set *cset, bool populated)
862 struct cgrp_cset_link *link;
864 lockdep_assert_held(&css_set_lock);
866 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
867 cgroup_update_populated(link->cgrp, populated);
871 * @task is leaving, advance task iterators which are pointing to it so
872 * that they can resume at the next position. Advancing an iterator might
873 * remove it from the list, use safe walk. See css_task_iter_skip() for
876 static void css_set_skip_task_iters(struct css_set *cset,
877 struct task_struct *task)
879 struct css_task_iter *it, *pos;
881 list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node)
882 css_task_iter_skip(it, task);
886 * css_set_move_task - move a task from one css_set to another
887 * @task: task being moved
888 * @from_cset: css_set @task currently belongs to (may be NULL)
889 * @to_cset: new css_set @task is being moved to (may be NULL)
890 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
892 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
893 * css_set, @from_cset can be NULL. If @task is being disassociated
894 * instead of moved, @to_cset can be NULL.
896 * This function automatically handles populated counter updates and
897 * css_task_iter adjustments but the caller is responsible for managing
898 * @from_cset and @to_cset's reference counts.
900 static void css_set_move_task(struct task_struct *task,
901 struct css_set *from_cset, struct css_set *to_cset,
904 lockdep_assert_held(&css_set_lock);
906 if (to_cset && !css_set_populated(to_cset))
907 css_set_update_populated(to_cset, true);
910 WARN_ON_ONCE(list_empty(&task->cg_list));
912 css_set_skip_task_iters(from_cset, task);
913 list_del_init(&task->cg_list);
914 if (!css_set_populated(from_cset))
915 css_set_update_populated(from_cset, false);
917 WARN_ON_ONCE(!list_empty(&task->cg_list));
922 * We are synchronized through cgroup_threadgroup_rwsem
923 * against PF_EXITING setting such that we can't race
924 * against cgroup_exit()/cgroup_free() dropping the css_set.
926 WARN_ON_ONCE(task->flags & PF_EXITING);
928 cgroup_move_task(task, to_cset);
929 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
935 * hash table for cgroup groups. This improves the performance to find
936 * an existing css_set. This hash doesn't (currently) take into
937 * account cgroups in empty hierarchies.
939 #define CSS_SET_HASH_BITS 7
940 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
942 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
944 unsigned long key = 0UL;
945 struct cgroup_subsys *ss;
948 for_each_subsys(ss, i)
949 key += (unsigned long)css[i];
950 key = (key >> 16) ^ key;
955 void put_css_set_locked(struct css_set *cset)
957 struct cgrp_cset_link *link, *tmp_link;
958 struct cgroup_subsys *ss;
961 lockdep_assert_held(&css_set_lock);
963 if (!refcount_dec_and_test(&cset->refcount))
966 WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
968 /* This css_set is dead. Unlink it and release cgroup and css refs */
969 for_each_subsys(ss, ssid) {
970 list_del(&cset->e_cset_node[ssid]);
971 css_put(cset->subsys[ssid]);
973 hash_del(&cset->hlist);
976 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
977 list_del(&link->cset_link);
978 list_del(&link->cgrp_link);
979 if (cgroup_parent(link->cgrp))
980 cgroup_put(link->cgrp);
984 if (css_set_threaded(cset)) {
985 list_del(&cset->threaded_csets_node);
986 put_css_set_locked(cset->dom_cset);
989 kfree_rcu(cset, rcu_head);
993 * compare_css_sets - helper function for find_existing_css_set().
994 * @cset: candidate css_set being tested
995 * @old_cset: existing css_set for a task
996 * @new_cgrp: cgroup that's being entered by the task
997 * @template: desired set of css pointers in css_set (pre-calculated)
999 * Returns true if "cset" matches "old_cset" except for the hierarchy
1000 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
1002 static bool compare_css_sets(struct css_set *cset,
1003 struct css_set *old_cset,
1004 struct cgroup *new_cgrp,
1005 struct cgroup_subsys_state *template[])
1007 struct cgroup *new_dfl_cgrp;
1008 struct list_head *l1, *l2;
1011 * On the default hierarchy, there can be csets which are
1012 * associated with the same set of cgroups but different csses.
1013 * Let's first ensure that csses match.
1015 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
1019 /* @cset's domain should match the default cgroup's */
1020 if (cgroup_on_dfl(new_cgrp))
1021 new_dfl_cgrp = new_cgrp;
1023 new_dfl_cgrp = old_cset->dfl_cgrp;
1025 if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
1029 * Compare cgroup pointers in order to distinguish between
1030 * different cgroups in hierarchies. As different cgroups may
1031 * share the same effective css, this comparison is always
1034 l1 = &cset->cgrp_links;
1035 l2 = &old_cset->cgrp_links;
1037 struct cgrp_cset_link *link1, *link2;
1038 struct cgroup *cgrp1, *cgrp2;
1042 /* See if we reached the end - both lists are equal length. */
1043 if (l1 == &cset->cgrp_links) {
1044 BUG_ON(l2 != &old_cset->cgrp_links);
1047 BUG_ON(l2 == &old_cset->cgrp_links);
1049 /* Locate the cgroups associated with these links. */
1050 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
1051 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
1052 cgrp1 = link1->cgrp;
1053 cgrp2 = link2->cgrp;
1054 /* Hierarchies should be linked in the same order. */
1055 BUG_ON(cgrp1->root != cgrp2->root);
1058 * If this hierarchy is the hierarchy of the cgroup
1059 * that's changing, then we need to check that this
1060 * css_set points to the new cgroup; if it's any other
1061 * hierarchy, then this css_set should point to the
1062 * same cgroup as the old css_set.
1064 if (cgrp1->root == new_cgrp->root) {
1065 if (cgrp1 != new_cgrp)
1076 * find_existing_css_set - init css array and find the matching css_set
1077 * @old_cset: the css_set that we're using before the cgroup transition
1078 * @cgrp: the cgroup that we're moving into
1079 * @template: out param for the new set of csses, should be clear on entry
1081 static struct css_set *find_existing_css_set(struct css_set *old_cset,
1082 struct cgroup *cgrp,
1083 struct cgroup_subsys_state *template[])
1085 struct cgroup_root *root = cgrp->root;
1086 struct cgroup_subsys *ss;
1087 struct css_set *cset;
1092 * Build the set of subsystem state objects that we want to see in the
1093 * new css_set. While subsystems can change globally, the entries here
1094 * won't change, so no need for locking.
1096 for_each_subsys(ss, i) {
1097 if (root->subsys_mask & (1UL << i)) {
1099 * @ss is in this hierarchy, so we want the
1100 * effective css from @cgrp.
1102 template[i] = cgroup_e_css_by_mask(cgrp, ss);
1105 * @ss is not in this hierarchy, so we don't want
1106 * to change the css.
1108 template[i] = old_cset->subsys[i];
1112 key = css_set_hash(template);
1113 hash_for_each_possible(css_set_table, cset, hlist, key) {
1114 if (!compare_css_sets(cset, old_cset, cgrp, template))
1117 /* This css_set matches what we need */
1121 /* No existing cgroup group matched */
1125 static void free_cgrp_cset_links(struct list_head *links_to_free)
1127 struct cgrp_cset_link *link, *tmp_link;
1129 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
1130 list_del(&link->cset_link);
1136 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1137 * @count: the number of links to allocate
1138 * @tmp_links: list_head the allocated links are put on
1140 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1141 * through ->cset_link. Returns 0 on success or -errno.
1143 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1145 struct cgrp_cset_link *link;
1148 INIT_LIST_HEAD(tmp_links);
1150 for (i = 0; i < count; i++) {
1151 link = kzalloc(sizeof(*link), GFP_KERNEL);
1153 free_cgrp_cset_links(tmp_links);
1156 list_add(&link->cset_link, tmp_links);
1162 * link_css_set - a helper function to link a css_set to a cgroup
1163 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1164 * @cset: the css_set to be linked
1165 * @cgrp: the destination cgroup
1167 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1168 struct cgroup *cgrp)
1170 struct cgrp_cset_link *link;
1172 BUG_ON(list_empty(tmp_links));
1174 if (cgroup_on_dfl(cgrp))
1175 cset->dfl_cgrp = cgrp;
1177 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1182 * Always add links to the tail of the lists so that the lists are
1183 * in chronological order.
1185 list_move_tail(&link->cset_link, &cgrp->cset_links);
1186 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1188 if (cgroup_parent(cgrp))
1189 cgroup_get_live(cgrp);
1193 * find_css_set - return a new css_set with one cgroup updated
1194 * @old_cset: the baseline css_set
1195 * @cgrp: the cgroup to be updated
1197 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1198 * substituted into the appropriate hierarchy.
1200 static struct css_set *find_css_set(struct css_set *old_cset,
1201 struct cgroup *cgrp)
1203 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1204 struct css_set *cset;
1205 struct list_head tmp_links;
1206 struct cgrp_cset_link *link;
1207 struct cgroup_subsys *ss;
1211 lockdep_assert_held(&cgroup_mutex);
1213 /* First see if we already have a cgroup group that matches
1214 * the desired set */
1215 spin_lock_irq(&css_set_lock);
1216 cset = find_existing_css_set(old_cset, cgrp, template);
1219 spin_unlock_irq(&css_set_lock);
1224 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1228 /* Allocate all the cgrp_cset_link objects that we'll need */
1229 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1234 refcount_set(&cset->refcount, 1);
1235 cset->dom_cset = cset;
1236 INIT_LIST_HEAD(&cset->tasks);
1237 INIT_LIST_HEAD(&cset->mg_tasks);
1238 INIT_LIST_HEAD(&cset->dying_tasks);
1239 INIT_LIST_HEAD(&cset->task_iters);
1240 INIT_LIST_HEAD(&cset->threaded_csets);
1241 INIT_HLIST_NODE(&cset->hlist);
1242 INIT_LIST_HEAD(&cset->cgrp_links);
1243 INIT_LIST_HEAD(&cset->mg_src_preload_node);
1244 INIT_LIST_HEAD(&cset->mg_dst_preload_node);
1245 INIT_LIST_HEAD(&cset->mg_node);
1247 /* Copy the set of subsystem state objects generated in
1248 * find_existing_css_set() */
1249 memcpy(cset->subsys, template, sizeof(cset->subsys));
1251 spin_lock_irq(&css_set_lock);
1252 /* Add reference counts and links from the new css_set. */
1253 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1254 struct cgroup *c = link->cgrp;
1256 if (c->root == cgrp->root)
1258 link_css_set(&tmp_links, cset, c);
1261 BUG_ON(!list_empty(&tmp_links));
1265 /* Add @cset to the hash table */
1266 key = css_set_hash(cset->subsys);
1267 hash_add(css_set_table, &cset->hlist, key);
1269 for_each_subsys(ss, ssid) {
1270 struct cgroup_subsys_state *css = cset->subsys[ssid];
1272 list_add_tail(&cset->e_cset_node[ssid],
1273 &css->cgroup->e_csets[ssid]);
1277 spin_unlock_irq(&css_set_lock);
1280 * If @cset should be threaded, look up the matching dom_cset and
1281 * link them up. We first fully initialize @cset then look for the
1282 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1283 * to stay empty until we return.
1285 if (cgroup_is_threaded(cset->dfl_cgrp)) {
1286 struct css_set *dcset;
1288 dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
1294 spin_lock_irq(&css_set_lock);
1295 cset->dom_cset = dcset;
1296 list_add_tail(&cset->threaded_csets_node,
1297 &dcset->threaded_csets);
1298 spin_unlock_irq(&css_set_lock);
1304 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1306 struct cgroup *root_cgrp = kf_root->kn->priv;
1308 return root_cgrp->root;
1311 static int cgroup_init_root_id(struct cgroup_root *root)
1315 lockdep_assert_held(&cgroup_mutex);
1317 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1321 root->hierarchy_id = id;
1325 static void cgroup_exit_root_id(struct cgroup_root *root)
1327 lockdep_assert_held(&cgroup_mutex);
1329 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1332 void cgroup_free_root(struct cgroup_root *root)
1337 static void cgroup_destroy_root(struct cgroup_root *root)
1339 struct cgroup *cgrp = &root->cgrp;
1340 struct cgrp_cset_link *link, *tmp_link;
1342 trace_cgroup_destroy_root(root);
1344 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1346 BUG_ON(atomic_read(&root->nr_cgrps));
1347 BUG_ON(!list_empty(&cgrp->self.children));
1349 /* Rebind all subsystems back to the default hierarchy */
1350 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1353 * Release all the links from cset_links to this hierarchy's
1356 spin_lock_irq(&css_set_lock);
1358 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1359 list_del(&link->cset_link);
1360 list_del(&link->cgrp_link);
1364 spin_unlock_irq(&css_set_lock);
1366 if (!list_empty(&root->root_list)) {
1367 list_del(&root->root_list);
1368 cgroup_root_count--;
1371 cgroup_exit_root_id(root);
1373 mutex_unlock(&cgroup_mutex);
1375 cgroup_rstat_exit(cgrp);
1376 kernfs_destroy_root(root->kf_root);
1377 cgroup_free_root(root);
1381 * look up cgroup associated with current task's cgroup namespace on the
1382 * specified hierarchy
1384 static struct cgroup *
1385 current_cgns_cgroup_from_root(struct cgroup_root *root)
1387 struct cgroup *res = NULL;
1388 struct css_set *cset;
1390 lockdep_assert_held(&css_set_lock);
1394 cset = current->nsproxy->cgroup_ns->root_cset;
1395 if (cset == &init_css_set) {
1397 } else if (root == &cgrp_dfl_root) {
1398 res = cset->dfl_cgrp;
1400 struct cgrp_cset_link *link;
1402 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1403 struct cgroup *c = link->cgrp;
1405 if (c->root == root) {
1417 /* look up cgroup associated with given css_set on the specified hierarchy */
1418 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1419 struct cgroup_root *root)
1421 struct cgroup *res = NULL;
1423 lockdep_assert_held(&cgroup_mutex);
1424 lockdep_assert_held(&css_set_lock);
1426 if (cset == &init_css_set) {
1428 } else if (root == &cgrp_dfl_root) {
1429 res = cset->dfl_cgrp;
1431 struct cgrp_cset_link *link;
1433 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1434 struct cgroup *c = link->cgrp;
1436 if (c->root == root) {
1448 * Return the cgroup for "task" from the given hierarchy. Must be
1449 * called with cgroup_mutex and css_set_lock held.
1451 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1452 struct cgroup_root *root)
1455 * No need to lock the task - since we hold css_set_lock the
1456 * task can't change groups.
1458 return cset_cgroup_from_root(task_css_set(task), root);
1462 * A task must hold cgroup_mutex to modify cgroups.
1464 * Any task can increment and decrement the count field without lock.
1465 * So in general, code holding cgroup_mutex can't rely on the count
1466 * field not changing. However, if the count goes to zero, then only
1467 * cgroup_attach_task() can increment it again. Because a count of zero
1468 * means that no tasks are currently attached, therefore there is no
1469 * way a task attached to that cgroup can fork (the other way to
1470 * increment the count). So code holding cgroup_mutex can safely
1471 * assume that if the count is zero, it will stay zero. Similarly, if
1472 * a task holds cgroup_mutex on a cgroup with zero count, it
1473 * knows that the cgroup won't be removed, as cgroup_rmdir()
1476 * A cgroup can only be deleted if both its 'count' of using tasks
1477 * is zero, and its list of 'children' cgroups is empty. Since all
1478 * tasks in the system use _some_ cgroup, and since there is always at
1479 * least one task in the system (init, pid == 1), therefore, root cgroup
1480 * always has either children cgroups and/or using tasks. So we don't
1481 * need a special hack to ensure that root cgroup cannot be deleted.
1483 * P.S. One more locking exception. RCU is used to guard the
1484 * update of a tasks cgroup pointer by cgroup_attach_task()
1487 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1489 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1492 struct cgroup_subsys *ss = cft->ss;
1494 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1495 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
1496 const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : "";
1498 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s",
1499 dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1502 strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1508 * cgroup_file_mode - deduce file mode of a control file
1509 * @cft: the control file in question
1511 * S_IRUGO for read, S_IWUSR for write.
1513 static umode_t cgroup_file_mode(const struct cftype *cft)
1517 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1520 if (cft->write_u64 || cft->write_s64 || cft->write) {
1521 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1531 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1532 * @subtree_control: the new subtree_control mask to consider
1533 * @this_ss_mask: available subsystems
1535 * On the default hierarchy, a subsystem may request other subsystems to be
1536 * enabled together through its ->depends_on mask. In such cases, more
1537 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1539 * This function calculates which subsystems need to be enabled if
1540 * @subtree_control is to be applied while restricted to @this_ss_mask.
1542 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1544 u16 cur_ss_mask = subtree_control;
1545 struct cgroup_subsys *ss;
1548 lockdep_assert_held(&cgroup_mutex);
1550 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1553 u16 new_ss_mask = cur_ss_mask;
1555 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1556 new_ss_mask |= ss->depends_on;
1557 } while_each_subsys_mask();
1560 * Mask out subsystems which aren't available. This can
1561 * happen only if some depended-upon subsystems were bound
1562 * to non-default hierarchies.
1564 new_ss_mask &= this_ss_mask;
1566 if (new_ss_mask == cur_ss_mask)
1568 cur_ss_mask = new_ss_mask;
1575 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1576 * @kn: the kernfs_node being serviced
1578 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1579 * the method finishes if locking succeeded. Note that once this function
1580 * returns the cgroup returned by cgroup_kn_lock_live() may become
1581 * inaccessible any time. If the caller intends to continue to access the
1582 * cgroup, it should pin it before invoking this function.
1584 void cgroup_kn_unlock(struct kernfs_node *kn)
1586 struct cgroup *cgrp;
1588 if (kernfs_type(kn) == KERNFS_DIR)
1591 cgrp = kn->parent->priv;
1593 mutex_unlock(&cgroup_mutex);
1595 kernfs_unbreak_active_protection(kn);
1600 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1601 * @kn: the kernfs_node being serviced
1602 * @drain_offline: perform offline draining on the cgroup
1604 * This helper is to be used by a cgroup kernfs method currently servicing
1605 * @kn. It breaks the active protection, performs cgroup locking and
1606 * verifies that the associated cgroup is alive. Returns the cgroup if
1607 * alive; otherwise, %NULL. A successful return should be undone by a
1608 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1609 * cgroup is drained of offlining csses before return.
1611 * Any cgroup kernfs method implementation which requires locking the
1612 * associated cgroup should use this helper. It avoids nesting cgroup
1613 * locking under kernfs active protection and allows all kernfs operations
1614 * including self-removal.
1616 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1618 struct cgroup *cgrp;
1620 if (kernfs_type(kn) == KERNFS_DIR)
1623 cgrp = kn->parent->priv;
1626 * We're gonna grab cgroup_mutex which nests outside kernfs
1627 * active_ref. cgroup liveliness check alone provides enough
1628 * protection against removal. Ensure @cgrp stays accessible and
1629 * break the active_ref protection.
1631 if (!cgroup_tryget(cgrp))
1633 kernfs_break_active_protection(kn);
1636 cgroup_lock_and_drain_offline(cgrp);
1638 mutex_lock(&cgroup_mutex);
1640 if (!cgroup_is_dead(cgrp))
1643 cgroup_kn_unlock(kn);
1647 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1649 char name[CGROUP_FILE_NAME_MAX];
1651 lockdep_assert_held(&cgroup_mutex);
1653 if (cft->file_offset) {
1654 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1655 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1657 spin_lock_irq(&cgroup_file_kn_lock);
1659 spin_unlock_irq(&cgroup_file_kn_lock);
1661 del_timer_sync(&cfile->notify_timer);
1664 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1668 * css_clear_dir - remove subsys files in a cgroup directory
1671 static void css_clear_dir(struct cgroup_subsys_state *css)
1673 struct cgroup *cgrp = css->cgroup;
1674 struct cftype *cfts;
1676 if (!(css->flags & CSS_VISIBLE))
1679 css->flags &= ~CSS_VISIBLE;
1682 if (cgroup_on_dfl(cgrp))
1683 cfts = cgroup_base_files;
1685 cfts = cgroup1_base_files;
1687 cgroup_addrm_files(css, cgrp, cfts, false);
1689 list_for_each_entry(cfts, &css->ss->cfts, node)
1690 cgroup_addrm_files(css, cgrp, cfts, false);
1695 * css_populate_dir - create subsys files in a cgroup directory
1698 * On failure, no file is added.
1700 static int css_populate_dir(struct cgroup_subsys_state *css)
1702 struct cgroup *cgrp = css->cgroup;
1703 struct cftype *cfts, *failed_cfts;
1706 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1710 if (cgroup_on_dfl(cgrp))
1711 cfts = cgroup_base_files;
1713 cfts = cgroup1_base_files;
1715 ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1719 list_for_each_entry(cfts, &css->ss->cfts, node) {
1720 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1728 css->flags |= CSS_VISIBLE;
1732 list_for_each_entry(cfts, &css->ss->cfts, node) {
1733 if (cfts == failed_cfts)
1735 cgroup_addrm_files(css, cgrp, cfts, false);
1740 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1742 struct cgroup *dcgrp = &dst_root->cgrp;
1743 struct cgroup_subsys *ss;
1745 u16 dfl_disable_ss_mask = 0;
1747 lockdep_assert_held(&cgroup_mutex);
1749 do_each_subsys_mask(ss, ssid, ss_mask) {
1751 * If @ss has non-root csses attached to it, can't move.
1752 * If @ss is an implicit controller, it is exempt from this
1753 * rule and can be stolen.
1755 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1756 !ss->implicit_on_dfl)
1759 /* can't move between two non-dummy roots either */
1760 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1764 * Collect ssid's that need to be disabled from default
1767 if (ss->root == &cgrp_dfl_root)
1768 dfl_disable_ss_mask |= 1 << ssid;
1770 } while_each_subsys_mask();
1772 if (dfl_disable_ss_mask) {
1773 struct cgroup *scgrp = &cgrp_dfl_root.cgrp;
1776 * Controllers from default hierarchy that need to be rebound
1777 * are all disabled together in one go.
1779 cgrp_dfl_root.subsys_mask &= ~dfl_disable_ss_mask;
1780 WARN_ON(cgroup_apply_control(scgrp));
1781 cgroup_finalize_control(scgrp, 0);
1784 do_each_subsys_mask(ss, ssid, ss_mask) {
1785 struct cgroup_root *src_root = ss->root;
1786 struct cgroup *scgrp = &src_root->cgrp;
1787 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1788 struct css_set *cset;
1790 WARN_ON(!css || cgroup_css(dcgrp, ss));
1792 if (src_root != &cgrp_dfl_root) {
1793 /* disable from the source */
1794 src_root->subsys_mask &= ~(1 << ssid);
1795 WARN_ON(cgroup_apply_control(scgrp));
1796 cgroup_finalize_control(scgrp, 0);
1800 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1801 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1802 ss->root = dst_root;
1803 css->cgroup = dcgrp;
1805 spin_lock_irq(&css_set_lock);
1806 hash_for_each(css_set_table, i, cset, hlist)
1807 list_move_tail(&cset->e_cset_node[ss->id],
1808 &dcgrp->e_csets[ss->id]);
1809 spin_unlock_irq(&css_set_lock);
1811 if (ss->css_rstat_flush) {
1812 list_del_rcu(&css->rstat_css_node);
1814 list_add_rcu(&css->rstat_css_node,
1815 &dcgrp->rstat_css_list);
1818 /* default hierarchy doesn't enable controllers by default */
1819 dst_root->subsys_mask |= 1 << ssid;
1820 if (dst_root == &cgrp_dfl_root) {
1821 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1823 dcgrp->subtree_control |= 1 << ssid;
1824 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1827 ret = cgroup_apply_control(dcgrp);
1829 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1834 } while_each_subsys_mask();
1836 kernfs_activate(dcgrp->kn);
1840 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1841 struct kernfs_root *kf_root)
1845 struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1846 struct cgroup *ns_cgroup;
1848 buf = kmalloc(PATH_MAX, GFP_KERNEL);
1852 spin_lock_irq(&css_set_lock);
1853 ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1854 len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1855 spin_unlock_irq(&css_set_lock);
1857 if (len >= PATH_MAX)
1860 seq_escape(sf, buf, " \t\n\\");
1867 enum cgroup2_param {
1869 Opt_memory_localevents,
1870 Opt_memory_recursiveprot,
1874 static const struct fs_parameter_spec cgroup2_fs_parameters[] = {
1875 fsparam_flag("nsdelegate", Opt_nsdelegate),
1876 fsparam_flag("memory_localevents", Opt_memory_localevents),
1877 fsparam_flag("memory_recursiveprot", Opt_memory_recursiveprot),
1881 static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1883 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1884 struct fs_parse_result result;
1887 opt = fs_parse(fc, cgroup2_fs_parameters, param, &result);
1892 case Opt_nsdelegate:
1893 ctx->flags |= CGRP_ROOT_NS_DELEGATE;
1895 case Opt_memory_localevents:
1896 ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1898 case Opt_memory_recursiveprot:
1899 ctx->flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1905 static void apply_cgroup_root_flags(unsigned int root_flags)
1907 if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
1908 if (root_flags & CGRP_ROOT_NS_DELEGATE)
1909 cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
1911 cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
1913 if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1914 cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1916 cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1918 if (root_flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1919 cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1921 cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1925 static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
1927 if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
1928 seq_puts(seq, ",nsdelegate");
1929 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1930 seq_puts(seq, ",memory_localevents");
1931 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1932 seq_puts(seq, ",memory_recursiveprot");
1936 static int cgroup_reconfigure(struct fs_context *fc)
1938 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1940 apply_cgroup_root_flags(ctx->flags);
1944 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1946 struct cgroup_subsys *ss;
1949 INIT_LIST_HEAD(&cgrp->self.sibling);
1950 INIT_LIST_HEAD(&cgrp->self.children);
1951 INIT_LIST_HEAD(&cgrp->cset_links);
1952 INIT_LIST_HEAD(&cgrp->pidlists);
1953 mutex_init(&cgrp->pidlist_mutex);
1954 cgrp->self.cgroup = cgrp;
1955 cgrp->self.flags |= CSS_ONLINE;
1956 cgrp->dom_cgrp = cgrp;
1957 cgrp->max_descendants = INT_MAX;
1958 cgrp->max_depth = INT_MAX;
1959 INIT_LIST_HEAD(&cgrp->rstat_css_list);
1960 prev_cputime_init(&cgrp->prev_cputime);
1962 for_each_subsys(ss, ssid)
1963 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1965 init_waitqueue_head(&cgrp->offline_waitq);
1966 INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1969 void init_cgroup_root(struct cgroup_fs_context *ctx)
1971 struct cgroup_root *root = ctx->root;
1972 struct cgroup *cgrp = &root->cgrp;
1974 INIT_LIST_HEAD(&root->root_list);
1975 atomic_set(&root->nr_cgrps, 1);
1977 init_cgroup_housekeeping(cgrp);
1979 root->flags = ctx->flags;
1980 if (ctx->release_agent)
1981 strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX);
1983 strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN);
1984 if (ctx->cpuset_clone_children)
1985 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1988 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1990 LIST_HEAD(tmp_links);
1991 struct cgroup *root_cgrp = &root->cgrp;
1992 struct kernfs_syscall_ops *kf_sops;
1993 struct css_set *cset;
1996 lockdep_assert_held(&cgroup_mutex);
1998 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
2004 * We're accessing css_set_count without locking css_set_lock here,
2005 * but that's OK - it can only be increased by someone holding
2006 * cgroup_lock, and that's us. Later rebinding may disable
2007 * controllers on the default hierarchy and thus create new csets,
2008 * which can't be more than the existing ones. Allocate 2x.
2010 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
2014 ret = cgroup_init_root_id(root);
2018 kf_sops = root == &cgrp_dfl_root ?
2019 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
2021 root->kf_root = kernfs_create_root(kf_sops,
2022 KERNFS_ROOT_CREATE_DEACTIVATED |
2023 KERNFS_ROOT_SUPPORT_EXPORTOP |
2024 KERNFS_ROOT_SUPPORT_USER_XATTR,
2026 if (IS_ERR(root->kf_root)) {
2027 ret = PTR_ERR(root->kf_root);
2030 root_cgrp->kn = root->kf_root->kn;
2031 WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
2032 root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
2034 ret = css_populate_dir(&root_cgrp->self);
2038 ret = cgroup_rstat_init(root_cgrp);
2042 ret = rebind_subsystems(root, ss_mask);
2046 ret = cgroup_bpf_inherit(root_cgrp);
2049 trace_cgroup_setup_root(root);
2052 * There must be no failure case after here, since rebinding takes
2053 * care of subsystems' refcounts, which are explicitly dropped in
2054 * the failure exit path.
2056 list_add(&root->root_list, &cgroup_roots);
2057 cgroup_root_count++;
2060 * Link the root cgroup in this hierarchy into all the css_set
2063 spin_lock_irq(&css_set_lock);
2064 hash_for_each(css_set_table, i, cset, hlist) {
2065 link_css_set(&tmp_links, cset, root_cgrp);
2066 if (css_set_populated(cset))
2067 cgroup_update_populated(root_cgrp, true);
2069 spin_unlock_irq(&css_set_lock);
2071 BUG_ON(!list_empty(&root_cgrp->self.children));
2072 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
2078 cgroup_rstat_exit(root_cgrp);
2080 kernfs_destroy_root(root->kf_root);
2081 root->kf_root = NULL;
2083 cgroup_exit_root_id(root);
2085 percpu_ref_exit(&root_cgrp->self.refcnt);
2087 free_cgrp_cset_links(&tmp_links);
2091 int cgroup_do_get_tree(struct fs_context *fc)
2093 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2096 ctx->kfc.root = ctx->root->kf_root;
2097 if (fc->fs_type == &cgroup2_fs_type)
2098 ctx->kfc.magic = CGROUP2_SUPER_MAGIC;
2100 ctx->kfc.magic = CGROUP_SUPER_MAGIC;
2101 ret = kernfs_get_tree(fc);
2104 * In non-init cgroup namespace, instead of root cgroup's dentry,
2105 * we return the dentry corresponding to the cgroupns->root_cgrp.
2107 if (!ret && ctx->ns != &init_cgroup_ns) {
2108 struct dentry *nsdentry;
2109 struct super_block *sb = fc->root->d_sb;
2110 struct cgroup *cgrp;
2112 mutex_lock(&cgroup_mutex);
2113 spin_lock_irq(&css_set_lock);
2115 cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root);
2117 spin_unlock_irq(&css_set_lock);
2118 mutex_unlock(&cgroup_mutex);
2120 nsdentry = kernfs_node_dentry(cgrp->kn, sb);
2122 if (IS_ERR(nsdentry)) {
2123 deactivate_locked_super(sb);
2124 ret = PTR_ERR(nsdentry);
2127 fc->root = nsdentry;
2130 if (!ctx->kfc.new_sb_created)
2131 cgroup_put(&ctx->root->cgrp);
2137 * Destroy a cgroup filesystem context.
2139 static void cgroup_fs_context_free(struct fs_context *fc)
2141 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2144 kfree(ctx->release_agent);
2145 put_cgroup_ns(ctx->ns);
2146 kernfs_free_fs_context(fc);
2150 static int cgroup_get_tree(struct fs_context *fc)
2152 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2155 cgrp_dfl_visible = true;
2156 cgroup_get_live(&cgrp_dfl_root.cgrp);
2157 ctx->root = &cgrp_dfl_root;
2159 ret = cgroup_do_get_tree(fc);
2161 apply_cgroup_root_flags(ctx->flags);
2165 static const struct fs_context_operations cgroup_fs_context_ops = {
2166 .free = cgroup_fs_context_free,
2167 .parse_param = cgroup2_parse_param,
2168 .get_tree = cgroup_get_tree,
2169 .reconfigure = cgroup_reconfigure,
2172 static const struct fs_context_operations cgroup1_fs_context_ops = {
2173 .free = cgroup_fs_context_free,
2174 .parse_param = cgroup1_parse_param,
2175 .get_tree = cgroup1_get_tree,
2176 .reconfigure = cgroup1_reconfigure,
2180 * Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2181 * we select the namespace we're going to use.
2183 static int cgroup_init_fs_context(struct fs_context *fc)
2185 struct cgroup_fs_context *ctx;
2187 ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL);
2191 ctx->ns = current->nsproxy->cgroup_ns;
2192 get_cgroup_ns(ctx->ns);
2193 fc->fs_private = &ctx->kfc;
2194 if (fc->fs_type == &cgroup2_fs_type)
2195 fc->ops = &cgroup_fs_context_ops;
2197 fc->ops = &cgroup1_fs_context_ops;
2198 put_user_ns(fc->user_ns);
2199 fc->user_ns = get_user_ns(ctx->ns->user_ns);
2204 static void cgroup_kill_sb(struct super_block *sb)
2206 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2207 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2210 * If @root doesn't have any children, start killing it.
2211 * This prevents new mounts by disabling percpu_ref_tryget_live().
2213 * And don't kill the default root.
2215 if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root &&
2216 !percpu_ref_is_dying(&root->cgrp.self.refcnt)) {
2217 cgroup_bpf_offline(&root->cgrp);
2218 percpu_ref_kill(&root->cgrp.self.refcnt);
2220 cgroup_put(&root->cgrp);
2224 struct file_system_type cgroup_fs_type = {
2226 .init_fs_context = cgroup_init_fs_context,
2227 .parameters = cgroup1_fs_parameters,
2228 .kill_sb = cgroup_kill_sb,
2229 .fs_flags = FS_USERNS_MOUNT,
2232 static struct file_system_type cgroup2_fs_type = {
2234 .init_fs_context = cgroup_init_fs_context,
2235 .parameters = cgroup2_fs_parameters,
2236 .kill_sb = cgroup_kill_sb,
2237 .fs_flags = FS_USERNS_MOUNT,
2240 #ifdef CONFIG_CPUSETS
2241 static const struct fs_context_operations cpuset_fs_context_ops = {
2242 .get_tree = cgroup1_get_tree,
2243 .free = cgroup_fs_context_free,
2247 * This is ugly, but preserves the userspace API for existing cpuset
2248 * users. If someone tries to mount the "cpuset" filesystem, we
2249 * silently switch it to mount "cgroup" instead
2251 static int cpuset_init_fs_context(struct fs_context *fc)
2253 char *agent = kstrdup("/sbin/cpuset_release_agent", GFP_USER);
2254 struct cgroup_fs_context *ctx;
2257 err = cgroup_init_fs_context(fc);
2263 fc->ops = &cpuset_fs_context_ops;
2265 ctx = cgroup_fc2context(fc);
2266 ctx->subsys_mask = 1 << cpuset_cgrp_id;
2267 ctx->flags |= CGRP_ROOT_NOPREFIX;
2268 ctx->release_agent = agent;
2270 get_filesystem(&cgroup_fs_type);
2271 put_filesystem(fc->fs_type);
2272 fc->fs_type = &cgroup_fs_type;
2277 static struct file_system_type cpuset_fs_type = {
2279 .init_fs_context = cpuset_init_fs_context,
2280 .fs_flags = FS_USERNS_MOUNT,
2284 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2285 struct cgroup_namespace *ns)
2287 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2289 return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2292 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2293 struct cgroup_namespace *ns)
2297 mutex_lock(&cgroup_mutex);
2298 spin_lock_irq(&css_set_lock);
2300 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2302 spin_unlock_irq(&css_set_lock);
2303 mutex_unlock(&cgroup_mutex);
2307 EXPORT_SYMBOL_GPL(cgroup_path_ns);
2310 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2311 * @task: target task
2312 * @buf: the buffer to write the path into
2313 * @buflen: the length of the buffer
2315 * Determine @task's cgroup on the first (the one with the lowest non-zero
2316 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2317 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2318 * cgroup controller callbacks.
2320 * Return value is the same as kernfs_path().
2322 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2324 struct cgroup_root *root;
2325 struct cgroup *cgrp;
2326 int hierarchy_id = 1;
2329 mutex_lock(&cgroup_mutex);
2330 spin_lock_irq(&css_set_lock);
2332 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2335 cgrp = task_cgroup_from_root(task, root);
2336 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2338 /* if no hierarchy exists, everyone is in "/" */
2339 ret = strlcpy(buf, "/", buflen);
2342 spin_unlock_irq(&css_set_lock);
2343 mutex_unlock(&cgroup_mutex);
2346 EXPORT_SYMBOL_GPL(task_cgroup_path);
2349 * cgroup_attach_lock - Lock for ->attach()
2350 * @lock_threadgroup: whether to down_write cgroup_threadgroup_rwsem
2352 * cgroup migration sometimes needs to stabilize threadgroups against forks and
2353 * exits by write-locking cgroup_threadgroup_rwsem. However, some ->attach()
2354 * implementations (e.g. cpuset), also need to disable CPU hotplug.
2355 * Unfortunately, letting ->attach() operations acquire cpus_read_lock() can
2356 * lead to deadlocks.
2358 * Bringing up a CPU may involve creating and destroying tasks which requires
2359 * read-locking threadgroup_rwsem, so threadgroup_rwsem nests inside
2360 * cpus_read_lock(). If we call an ->attach() which acquires the cpus lock while
2361 * write-locking threadgroup_rwsem, the locking order is reversed and we end up
2362 * waiting for an on-going CPU hotplug operation which in turn is waiting for
2363 * the threadgroup_rwsem to be released to create new tasks. For more details:
2365 * http://lkml.kernel.org/r/20220711174629.uehfmqegcwn2lqzu@wubuntu
2367 * Resolve the situation by always acquiring cpus_read_lock() before optionally
2368 * write-locking cgroup_threadgroup_rwsem. This allows ->attach() to assume that
2369 * CPU hotplug is disabled on entry.
2371 static void cgroup_attach_lock(bool lock_threadgroup)
2374 if (lock_threadgroup)
2375 percpu_down_write(&cgroup_threadgroup_rwsem);
2379 * cgroup_attach_unlock - Undo cgroup_attach_lock()
2380 * @lock_threadgroup: whether to up_write cgroup_threadgroup_rwsem
2382 static void cgroup_attach_unlock(bool lock_threadgroup)
2384 if (lock_threadgroup)
2385 percpu_up_write(&cgroup_threadgroup_rwsem);
2390 * cgroup_migrate_add_task - add a migration target task to a migration context
2391 * @task: target task
2392 * @mgctx: target migration context
2394 * Add @task, which is a migration target, to @mgctx->tset. This function
2395 * becomes noop if @task doesn't need to be migrated. @task's css_set
2396 * should have been added as a migration source and @task->cg_list will be
2397 * moved from the css_set's tasks list to mg_tasks one.
2399 static void cgroup_migrate_add_task(struct task_struct *task,
2400 struct cgroup_mgctx *mgctx)
2402 struct css_set *cset;
2404 lockdep_assert_held(&css_set_lock);
2406 /* @task either already exited or can't exit until the end */
2407 if (task->flags & PF_EXITING)
2410 /* cgroup_threadgroup_rwsem protects racing against forks */
2411 WARN_ON_ONCE(list_empty(&task->cg_list));
2413 cset = task_css_set(task);
2414 if (!cset->mg_src_cgrp)
2417 mgctx->tset.nr_tasks++;
2419 list_move_tail(&task->cg_list, &cset->mg_tasks);
2420 if (list_empty(&cset->mg_node))
2421 list_add_tail(&cset->mg_node,
2422 &mgctx->tset.src_csets);
2423 if (list_empty(&cset->mg_dst_cset->mg_node))
2424 list_add_tail(&cset->mg_dst_cset->mg_node,
2425 &mgctx->tset.dst_csets);
2429 * cgroup_taskset_first - reset taskset and return the first task
2430 * @tset: taskset of interest
2431 * @dst_cssp: output variable for the destination css
2433 * @tset iteration is initialized and the first task is returned.
2435 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2436 struct cgroup_subsys_state **dst_cssp)
2438 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2439 tset->cur_task = NULL;
2441 return cgroup_taskset_next(tset, dst_cssp);
2445 * cgroup_taskset_next - iterate to the next task in taskset
2446 * @tset: taskset of interest
2447 * @dst_cssp: output variable for the destination css
2449 * Return the next task in @tset. Iteration must have been initialized
2450 * with cgroup_taskset_first().
2452 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2453 struct cgroup_subsys_state **dst_cssp)
2455 struct css_set *cset = tset->cur_cset;
2456 struct task_struct *task = tset->cur_task;
2458 while (CGROUP_HAS_SUBSYS_CONFIG && &cset->mg_node != tset->csets) {
2460 task = list_first_entry(&cset->mg_tasks,
2461 struct task_struct, cg_list);
2463 task = list_next_entry(task, cg_list);
2465 if (&task->cg_list != &cset->mg_tasks) {
2466 tset->cur_cset = cset;
2467 tset->cur_task = task;
2470 * This function may be called both before and
2471 * after cgroup_taskset_migrate(). The two cases
2472 * can be distinguished by looking at whether @cset
2473 * has its ->mg_dst_cset set.
2475 if (cset->mg_dst_cset)
2476 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2478 *dst_cssp = cset->subsys[tset->ssid];
2483 cset = list_next_entry(cset, mg_node);
2491 * cgroup_migrate_execute - migrate a taskset
2492 * @mgctx: migration context
2494 * Migrate tasks in @mgctx as setup by migration preparation functions.
2495 * This function fails iff one of the ->can_attach callbacks fails and
2496 * guarantees that either all or none of the tasks in @mgctx are migrated.
2497 * @mgctx is consumed regardless of success.
2499 static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2501 struct cgroup_taskset *tset = &mgctx->tset;
2502 struct cgroup_subsys *ss;
2503 struct task_struct *task, *tmp_task;
2504 struct css_set *cset, *tmp_cset;
2505 int ssid, failed_ssid, ret;
2507 /* check that we can legitimately attach to the cgroup */
2508 if (tset->nr_tasks) {
2509 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2510 if (ss->can_attach) {
2512 ret = ss->can_attach(tset);
2515 goto out_cancel_attach;
2518 } while_each_subsys_mask();
2522 * Now that we're guaranteed success, proceed to move all tasks to
2523 * the new cgroup. There are no failure cases after here, so this
2524 * is the commit point.
2526 spin_lock_irq(&css_set_lock);
2527 list_for_each_entry(cset, &tset->src_csets, mg_node) {
2528 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2529 struct css_set *from_cset = task_css_set(task);
2530 struct css_set *to_cset = cset->mg_dst_cset;
2532 get_css_set(to_cset);
2533 to_cset->nr_tasks++;
2534 css_set_move_task(task, from_cset, to_cset, true);
2535 from_cset->nr_tasks--;
2537 * If the source or destination cgroup is frozen,
2538 * the task might require to change its state.
2540 cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp,
2542 put_css_set_locked(from_cset);
2546 spin_unlock_irq(&css_set_lock);
2549 * Migration is committed, all target tasks are now on dst_csets.
2550 * Nothing is sensitive to fork() after this point. Notify
2551 * controllers that migration is complete.
2553 tset->csets = &tset->dst_csets;
2555 if (tset->nr_tasks) {
2556 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2561 } while_each_subsys_mask();
2565 goto out_release_tset;
2568 if (tset->nr_tasks) {
2569 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2570 if (ssid == failed_ssid)
2572 if (ss->cancel_attach) {
2574 ss->cancel_attach(tset);
2576 } while_each_subsys_mask();
2579 spin_lock_irq(&css_set_lock);
2580 list_splice_init(&tset->dst_csets, &tset->src_csets);
2581 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2582 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2583 list_del_init(&cset->mg_node);
2585 spin_unlock_irq(&css_set_lock);
2588 * Re-initialize the cgroup_taskset structure in case it is reused
2589 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2593 tset->csets = &tset->src_csets;
2598 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2599 * @dst_cgrp: destination cgroup to test
2601 * On the default hierarchy, except for the mixable, (possible) thread root
2602 * and threaded cgroups, subtree_control must be zero for migration
2603 * destination cgroups with tasks so that child cgroups don't compete
2606 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2608 /* v1 doesn't have any restriction */
2609 if (!cgroup_on_dfl(dst_cgrp))
2612 /* verify @dst_cgrp can host resources */
2613 if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
2616 /* mixables don't care */
2617 if (cgroup_is_mixable(dst_cgrp))
2621 * If @dst_cgrp is already or can become a thread root or is
2622 * threaded, it doesn't matter.
2624 if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
2627 /* apply no-internal-process constraint */
2628 if (dst_cgrp->subtree_control)
2635 * cgroup_migrate_finish - cleanup after attach
2636 * @mgctx: migration context
2638 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2639 * those functions for details.
2641 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2643 struct css_set *cset, *tmp_cset;
2645 lockdep_assert_held(&cgroup_mutex);
2647 spin_lock_irq(&css_set_lock);
2649 list_for_each_entry_safe(cset, tmp_cset, &mgctx->preloaded_src_csets,
2650 mg_src_preload_node) {
2651 cset->mg_src_cgrp = NULL;
2652 cset->mg_dst_cgrp = NULL;
2653 cset->mg_dst_cset = NULL;
2654 list_del_init(&cset->mg_src_preload_node);
2655 put_css_set_locked(cset);
2658 list_for_each_entry_safe(cset, tmp_cset, &mgctx->preloaded_dst_csets,
2659 mg_dst_preload_node) {
2660 cset->mg_src_cgrp = NULL;
2661 cset->mg_dst_cgrp = NULL;
2662 cset->mg_dst_cset = NULL;
2663 list_del_init(&cset->mg_dst_preload_node);
2664 put_css_set_locked(cset);
2667 spin_unlock_irq(&css_set_lock);
2671 * cgroup_migrate_add_src - add a migration source css_set
2672 * @src_cset: the source css_set to add
2673 * @dst_cgrp: the destination cgroup
2674 * @mgctx: migration context
2676 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2677 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2678 * up by cgroup_migrate_finish().
2680 * This function may be called without holding cgroup_threadgroup_rwsem
2681 * even if the target is a process. Threads may be created and destroyed
2682 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2683 * into play and the preloaded css_sets are guaranteed to cover all
2686 void cgroup_migrate_add_src(struct css_set *src_cset,
2687 struct cgroup *dst_cgrp,
2688 struct cgroup_mgctx *mgctx)
2690 struct cgroup *src_cgrp;
2692 lockdep_assert_held(&cgroup_mutex);
2693 lockdep_assert_held(&css_set_lock);
2696 * If ->dead, @src_set is associated with one or more dead cgroups
2697 * and doesn't contain any migratable tasks. Ignore it early so
2698 * that the rest of migration path doesn't get confused by it.
2703 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2705 if (!list_empty(&src_cset->mg_src_preload_node))
2708 WARN_ON(src_cset->mg_src_cgrp);
2709 WARN_ON(src_cset->mg_dst_cgrp);
2710 WARN_ON(!list_empty(&src_cset->mg_tasks));
2711 WARN_ON(!list_empty(&src_cset->mg_node));
2713 src_cset->mg_src_cgrp = src_cgrp;
2714 src_cset->mg_dst_cgrp = dst_cgrp;
2715 get_css_set(src_cset);
2716 list_add_tail(&src_cset->mg_src_preload_node, &mgctx->preloaded_src_csets);
2720 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2721 * @mgctx: migration context
2723 * Tasks are about to be moved and all the source css_sets have been
2724 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2725 * pins all destination css_sets, links each to its source, and append them
2726 * to @mgctx->preloaded_dst_csets.
2728 * This function must be called after cgroup_migrate_add_src() has been
2729 * called on each migration source css_set. After migration is performed
2730 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2733 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2735 struct css_set *src_cset, *tmp_cset;
2737 lockdep_assert_held(&cgroup_mutex);
2739 /* look up the dst cset for each src cset and link it to src */
2740 list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2741 mg_src_preload_node) {
2742 struct css_set *dst_cset;
2743 struct cgroup_subsys *ss;
2746 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2750 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2753 * If src cset equals dst, it's noop. Drop the src.
2754 * cgroup_migrate() will skip the cset too. Note that we
2755 * can't handle src == dst as some nodes are used by both.
2757 if (src_cset == dst_cset) {
2758 src_cset->mg_src_cgrp = NULL;
2759 src_cset->mg_dst_cgrp = NULL;
2760 list_del_init(&src_cset->mg_src_preload_node);
2761 put_css_set(src_cset);
2762 put_css_set(dst_cset);
2766 src_cset->mg_dst_cset = dst_cset;
2768 if (list_empty(&dst_cset->mg_dst_preload_node))
2769 list_add_tail(&dst_cset->mg_dst_preload_node,
2770 &mgctx->preloaded_dst_csets);
2772 put_css_set(dst_cset);
2774 for_each_subsys(ss, ssid)
2775 if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2776 mgctx->ss_mask |= 1 << ssid;
2783 * cgroup_migrate - migrate a process or task to a cgroup
2784 * @leader: the leader of the process or the task to migrate
2785 * @threadgroup: whether @leader points to the whole process or a single task
2786 * @mgctx: migration context
2788 * Migrate a process or task denoted by @leader. If migrating a process,
2789 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2790 * responsible for invoking cgroup_migrate_add_src() and
2791 * cgroup_migrate_prepare_dst() on the targets before invoking this
2792 * function and following up with cgroup_migrate_finish().
2794 * As long as a controller's ->can_attach() doesn't fail, this function is
2795 * guaranteed to succeed. This means that, excluding ->can_attach()
2796 * failure, when migrating multiple targets, the success or failure can be
2797 * decided for all targets by invoking group_migrate_prepare_dst() before
2798 * actually starting migrating.
2800 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2801 struct cgroup_mgctx *mgctx)
2803 struct task_struct *task;
2806 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2807 * already PF_EXITING could be freed from underneath us unless we
2808 * take an rcu_read_lock.
2810 spin_lock_irq(&css_set_lock);
2814 cgroup_migrate_add_task(task, mgctx);
2817 } while_each_thread(leader, task);
2819 spin_unlock_irq(&css_set_lock);
2821 return cgroup_migrate_execute(mgctx);
2825 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2826 * @dst_cgrp: the cgroup to attach to
2827 * @leader: the task or the leader of the threadgroup to be attached
2828 * @threadgroup: attach the whole threadgroup?
2830 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2832 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2835 DEFINE_CGROUP_MGCTX(mgctx);
2836 struct task_struct *task;
2839 /* look up all src csets */
2840 spin_lock_irq(&css_set_lock);
2844 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2847 } while_each_thread(leader, task);
2849 spin_unlock_irq(&css_set_lock);
2851 /* prepare dst csets and commit */
2852 ret = cgroup_migrate_prepare_dst(&mgctx);
2854 ret = cgroup_migrate(leader, threadgroup, &mgctx);
2856 cgroup_migrate_finish(&mgctx);
2859 TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup);
2864 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
2865 bool *threadgroup_locked)
2867 struct task_struct *tsk;
2870 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2871 return ERR_PTR(-EINVAL);
2874 * If we migrate a single thread, we don't care about threadgroup
2875 * stability. If the thread is `current`, it won't exit(2) under our
2876 * hands or change PID through exec(2). We exclude
2877 * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write
2878 * callers by cgroup_mutex.
2879 * Therefore, we can skip the global lock.
2881 lockdep_assert_held(&cgroup_mutex);
2882 *threadgroup_locked = pid || threadgroup;
2883 cgroup_attach_lock(*threadgroup_locked);
2887 tsk = find_task_by_vpid(pid);
2889 tsk = ERR_PTR(-ESRCH);
2890 goto out_unlock_threadgroup;
2897 tsk = tsk->group_leader;
2900 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2901 * If userland migrates such a kthread to a non-root cgroup, it can
2902 * become trapped in a cpuset, or RT kthread may be born in a
2903 * cgroup with no rt_runtime allocated. Just say no.
2905 if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2906 tsk = ERR_PTR(-EINVAL);
2907 goto out_unlock_threadgroup;
2910 get_task_struct(tsk);
2911 goto out_unlock_rcu;
2913 out_unlock_threadgroup:
2914 cgroup_attach_unlock(*threadgroup_locked);
2915 *threadgroup_locked = false;
2921 void cgroup_procs_write_finish(struct task_struct *task, bool threadgroup_locked)
2923 struct cgroup_subsys *ss;
2926 /* release reference from cgroup_procs_write_start() */
2927 put_task_struct(task);
2929 cgroup_attach_unlock(threadgroup_locked);
2931 for_each_subsys(ss, ssid)
2932 if (ss->post_attach)
2936 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2938 struct cgroup_subsys *ss;
2939 bool printed = false;
2942 do_each_subsys_mask(ss, ssid, ss_mask) {
2945 seq_puts(seq, ss->name);
2947 } while_each_subsys_mask();
2949 seq_putc(seq, '\n');
2952 /* show controllers which are enabled from the parent */
2953 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2955 struct cgroup *cgrp = seq_css(seq)->cgroup;
2957 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2961 /* show controllers which are enabled for a given cgroup's children */
2962 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2964 struct cgroup *cgrp = seq_css(seq)->cgroup;
2966 cgroup_print_ss_mask(seq, cgrp->subtree_control);
2971 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2972 * @cgrp: root of the subtree to update csses for
2974 * @cgrp's control masks have changed and its subtree's css associations
2975 * need to be updated accordingly. This function looks up all css_sets
2976 * which are attached to the subtree, creates the matching updated css_sets
2977 * and migrates the tasks to the new ones.
2979 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2981 DEFINE_CGROUP_MGCTX(mgctx);
2982 struct cgroup_subsys_state *d_css;
2983 struct cgroup *dsct;
2984 struct css_set *src_cset;
2988 lockdep_assert_held(&cgroup_mutex);
2990 /* look up all csses currently attached to @cgrp's subtree */
2991 spin_lock_irq(&css_set_lock);
2992 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2993 struct cgrp_cset_link *link;
2995 list_for_each_entry(link, &dsct->cset_links, cset_link)
2996 cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2998 spin_unlock_irq(&css_set_lock);
3001 * We need to write-lock threadgroup_rwsem while migrating tasks.
3002 * However, if there are no source csets for @cgrp, changing its
3003 * controllers isn't gonna produce any task migrations and the
3004 * write-locking can be skipped safely.
3006 has_tasks = !list_empty(&mgctx.preloaded_src_csets);
3007 cgroup_attach_lock(has_tasks);
3009 /* NULL dst indicates self on default hierarchy */
3010 ret = cgroup_migrate_prepare_dst(&mgctx);
3014 spin_lock_irq(&css_set_lock);
3015 list_for_each_entry(src_cset, &mgctx.preloaded_src_csets,
3016 mg_src_preload_node) {
3017 struct task_struct *task, *ntask;
3019 /* all tasks in src_csets need to be migrated */
3020 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
3021 cgroup_migrate_add_task(task, &mgctx);
3023 spin_unlock_irq(&css_set_lock);
3025 ret = cgroup_migrate_execute(&mgctx);
3027 cgroup_migrate_finish(&mgctx);
3028 cgroup_attach_unlock(has_tasks);
3033 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
3034 * @cgrp: root of the target subtree
3036 * Because css offlining is asynchronous, userland may try to re-enable a
3037 * controller while the previous css is still around. This function grabs
3038 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
3040 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
3041 __acquires(&cgroup_mutex)
3043 struct cgroup *dsct;
3044 struct cgroup_subsys_state *d_css;
3045 struct cgroup_subsys *ss;
3049 mutex_lock(&cgroup_mutex);
3051 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3052 for_each_subsys(ss, ssid) {
3053 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3056 if (!css || !percpu_ref_is_dying(&css->refcnt))
3059 cgroup_get_live(dsct);
3060 prepare_to_wait(&dsct->offline_waitq, &wait,
3061 TASK_UNINTERRUPTIBLE);
3063 mutex_unlock(&cgroup_mutex);
3065 finish_wait(&dsct->offline_waitq, &wait);
3074 * cgroup_save_control - save control masks and dom_cgrp of a subtree
3075 * @cgrp: root of the target subtree
3077 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
3078 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3081 static void cgroup_save_control(struct cgroup *cgrp)
3083 struct cgroup *dsct;
3084 struct cgroup_subsys_state *d_css;
3086 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3087 dsct->old_subtree_control = dsct->subtree_control;
3088 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
3089 dsct->old_dom_cgrp = dsct->dom_cgrp;
3094 * cgroup_propagate_control - refresh control masks of a subtree
3095 * @cgrp: root of the target subtree
3097 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3098 * ->subtree_control and propagate controller availability through the
3099 * subtree so that descendants don't have unavailable controllers enabled.
3101 static void cgroup_propagate_control(struct cgroup *cgrp)
3103 struct cgroup *dsct;
3104 struct cgroup_subsys_state *d_css;
3106 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3107 dsct->subtree_control &= cgroup_control(dsct);
3108 dsct->subtree_ss_mask =
3109 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3110 cgroup_ss_mask(dsct));
3115 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3116 * @cgrp: root of the target subtree
3118 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3119 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3122 static void cgroup_restore_control(struct cgroup *cgrp)
3124 struct cgroup *dsct;
3125 struct cgroup_subsys_state *d_css;
3127 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3128 dsct->subtree_control = dsct->old_subtree_control;
3129 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3130 dsct->dom_cgrp = dsct->old_dom_cgrp;
3134 static bool css_visible(struct cgroup_subsys_state *css)
3136 struct cgroup_subsys *ss = css->ss;
3137 struct cgroup *cgrp = css->cgroup;
3139 if (cgroup_control(cgrp) & (1 << ss->id))
3141 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3143 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3147 * cgroup_apply_control_enable - enable or show csses according to control
3148 * @cgrp: root of the target subtree
3150 * Walk @cgrp's subtree and create new csses or make the existing ones
3151 * visible. A css is created invisible if it's being implicitly enabled
3152 * through dependency. An invisible css is made visible when the userland
3153 * explicitly enables it.
3155 * Returns 0 on success, -errno on failure. On failure, csses which have
3156 * been processed already aren't cleaned up. The caller is responsible for
3157 * cleaning up with cgroup_apply_control_disable().
3159 static int cgroup_apply_control_enable(struct cgroup *cgrp)
3161 struct cgroup *dsct;
3162 struct cgroup_subsys_state *d_css;
3163 struct cgroup_subsys *ss;
3166 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3167 for_each_subsys(ss, ssid) {
3168 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3170 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3174 css = css_create(dsct, ss);
3176 return PTR_ERR(css);
3179 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3181 if (css_visible(css)) {
3182 ret = css_populate_dir(css);
3193 * cgroup_apply_control_disable - kill or hide csses according to control
3194 * @cgrp: root of the target subtree
3196 * Walk @cgrp's subtree and kill and hide csses so that they match
3197 * cgroup_ss_mask() and cgroup_visible_mask().
3199 * A css is hidden when the userland requests it to be disabled while other
3200 * subsystems are still depending on it. The css must not actively control
3201 * resources and be in the vanilla state if it's made visible again later.
3202 * Controllers which may be depended upon should provide ->css_reset() for
3205 static void cgroup_apply_control_disable(struct cgroup *cgrp)
3207 struct cgroup *dsct;
3208 struct cgroup_subsys_state *d_css;
3209 struct cgroup_subsys *ss;
3212 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3213 for_each_subsys(ss, ssid) {
3214 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3219 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3222 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
3224 } else if (!css_visible(css)) {
3234 * cgroup_apply_control - apply control mask updates to the subtree
3235 * @cgrp: root of the target subtree
3237 * subsystems can be enabled and disabled in a subtree using the following
3240 * 1. Call cgroup_save_control() to stash the current state.
3241 * 2. Update ->subtree_control masks in the subtree as desired.
3242 * 3. Call cgroup_apply_control() to apply the changes.
3243 * 4. Optionally perform other related operations.
3244 * 5. Call cgroup_finalize_control() to finish up.
3246 * This function implements step 3 and propagates the mask changes
3247 * throughout @cgrp's subtree, updates csses accordingly and perform
3248 * process migrations.
3250 static int cgroup_apply_control(struct cgroup *cgrp)
3254 cgroup_propagate_control(cgrp);
3256 ret = cgroup_apply_control_enable(cgrp);
3261 * At this point, cgroup_e_css_by_mask() results reflect the new csses
3262 * making the following cgroup_update_dfl_csses() properly update
3263 * css associations of all tasks in the subtree.
3265 ret = cgroup_update_dfl_csses(cgrp);
3273 * cgroup_finalize_control - finalize control mask update
3274 * @cgrp: root of the target subtree
3275 * @ret: the result of the update
3277 * Finalize control mask update. See cgroup_apply_control() for more info.
3279 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3282 cgroup_restore_control(cgrp);
3283 cgroup_propagate_control(cgrp);
3286 cgroup_apply_control_disable(cgrp);
3289 static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
3291 u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
3293 /* if nothing is getting enabled, nothing to worry about */
3297 /* can @cgrp host any resources? */
3298 if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
3301 /* mixables don't care */
3302 if (cgroup_is_mixable(cgrp))
3305 if (domain_enable) {
3306 /* can't enable domain controllers inside a thread subtree */
3307 if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3311 * Threaded controllers can handle internal competitions
3312 * and are always allowed inside a (prospective) thread
3315 if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3320 * Controllers can't be enabled for a cgroup with tasks to avoid
3321 * child cgroups competing against tasks.
3323 if (cgroup_has_tasks(cgrp))
3329 /* change the enabled child controllers for a cgroup in the default hierarchy */
3330 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3331 char *buf, size_t nbytes,
3334 u16 enable = 0, disable = 0;
3335 struct cgroup *cgrp, *child;
3336 struct cgroup_subsys *ss;
3341 * Parse input - space separated list of subsystem names prefixed
3342 * with either + or -.
3344 buf = strstrip(buf);
3345 while ((tok = strsep(&buf, " "))) {
3348 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3349 if (!cgroup_ssid_enabled(ssid) ||
3350 strcmp(tok + 1, ss->name))
3354 enable |= 1 << ssid;
3355 disable &= ~(1 << ssid);
3356 } else if (*tok == '-') {
3357 disable |= 1 << ssid;
3358 enable &= ~(1 << ssid);
3363 } while_each_subsys_mask();
3364 if (ssid == CGROUP_SUBSYS_COUNT)
3368 cgrp = cgroup_kn_lock_live(of->kn, true);
3372 for_each_subsys(ss, ssid) {
3373 if (enable & (1 << ssid)) {
3374 if (cgrp->subtree_control & (1 << ssid)) {
3375 enable &= ~(1 << ssid);
3379 if (!(cgroup_control(cgrp) & (1 << ssid))) {
3383 } else if (disable & (1 << ssid)) {
3384 if (!(cgrp->subtree_control & (1 << ssid))) {
3385 disable &= ~(1 << ssid);
3389 /* a child has it enabled? */
3390 cgroup_for_each_live_child(child, cgrp) {
3391 if (child->subtree_control & (1 << ssid)) {
3399 if (!enable && !disable) {
3404 ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3408 /* save and update control masks and prepare csses */
3409 cgroup_save_control(cgrp);
3411 cgrp->subtree_control |= enable;
3412 cgrp->subtree_control &= ~disable;
3414 ret = cgroup_apply_control(cgrp);
3415 cgroup_finalize_control(cgrp, ret);
3419 kernfs_activate(cgrp->kn);
3421 cgroup_kn_unlock(of->kn);
3422 return ret ?: nbytes;
3426 * cgroup_enable_threaded - make @cgrp threaded
3427 * @cgrp: the target cgroup
3429 * Called when "threaded" is written to the cgroup.type interface file and
3430 * tries to make @cgrp threaded and join the parent's resource domain.
3431 * This function is never called on the root cgroup as cgroup.type doesn't
3434 static int cgroup_enable_threaded(struct cgroup *cgrp)
3436 struct cgroup *parent = cgroup_parent(cgrp);
3437 struct cgroup *dom_cgrp = parent->dom_cgrp;
3438 struct cgroup *dsct;
3439 struct cgroup_subsys_state *d_css;
3442 lockdep_assert_held(&cgroup_mutex);
3444 /* noop if already threaded */
3445 if (cgroup_is_threaded(cgrp))
3449 * If @cgroup is populated or has domain controllers enabled, it
3450 * can't be switched. While the below cgroup_can_be_thread_root()
3451 * test can catch the same conditions, that's only when @parent is
3452 * not mixable, so let's check it explicitly.
3454 if (cgroup_is_populated(cgrp) ||
3455 cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
3458 /* we're joining the parent's domain, ensure its validity */
3459 if (!cgroup_is_valid_domain(dom_cgrp) ||
3460 !cgroup_can_be_thread_root(dom_cgrp))
3464 * The following shouldn't cause actual migrations and should
3467 cgroup_save_control(cgrp);
3469 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)
3470 if (dsct == cgrp || cgroup_is_threaded(dsct))
3471 dsct->dom_cgrp = dom_cgrp;
3473 ret = cgroup_apply_control(cgrp);
3475 parent->nr_threaded_children++;
3477 cgroup_finalize_control(cgrp, ret);
3481 static int cgroup_type_show(struct seq_file *seq, void *v)
3483 struct cgroup *cgrp = seq_css(seq)->cgroup;
3485 if (cgroup_is_threaded(cgrp))
3486 seq_puts(seq, "threaded\n");
3487 else if (!cgroup_is_valid_domain(cgrp))
3488 seq_puts(seq, "domain invalid\n");
3489 else if (cgroup_is_thread_root(cgrp))
3490 seq_puts(seq, "domain threaded\n");
3492 seq_puts(seq, "domain\n");
3497 static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3498 size_t nbytes, loff_t off)
3500 struct cgroup *cgrp;
3503 /* only switching to threaded mode is supported */
3504 if (strcmp(strstrip(buf), "threaded"))
3507 /* drain dying csses before we re-apply (threaded) subtree control */
3508 cgrp = cgroup_kn_lock_live(of->kn, true);
3512 /* threaded can only be enabled */
3513 ret = cgroup_enable_threaded(cgrp);
3515 cgroup_kn_unlock(of->kn);
3516 return ret ?: nbytes;
3519 static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3521 struct cgroup *cgrp = seq_css(seq)->cgroup;
3522 int descendants = READ_ONCE(cgrp->max_descendants);
3524 if (descendants == INT_MAX)
3525 seq_puts(seq, "max\n");
3527 seq_printf(seq, "%d\n", descendants);
3532 static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3533 char *buf, size_t nbytes, loff_t off)
3535 struct cgroup *cgrp;
3539 buf = strstrip(buf);
3540 if (!strcmp(buf, "max")) {
3541 descendants = INT_MAX;
3543 ret = kstrtoint(buf, 0, &descendants);
3548 if (descendants < 0)
3551 cgrp = cgroup_kn_lock_live(of->kn, false);
3555 cgrp->max_descendants = descendants;
3557 cgroup_kn_unlock(of->kn);
3562 static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3564 struct cgroup *cgrp = seq_css(seq)->cgroup;
3565 int depth = READ_ONCE(cgrp->max_depth);
3567 if (depth == INT_MAX)
3568 seq_puts(seq, "max\n");
3570 seq_printf(seq, "%d\n", depth);
3575 static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3576 char *buf, size_t nbytes, loff_t off)
3578 struct cgroup *cgrp;
3582 buf = strstrip(buf);
3583 if (!strcmp(buf, "max")) {
3586 ret = kstrtoint(buf, 0, &depth);
3594 cgrp = cgroup_kn_lock_live(of->kn, false);
3598 cgrp->max_depth = depth;
3600 cgroup_kn_unlock(of->kn);
3605 static int cgroup_events_show(struct seq_file *seq, void *v)
3607 struct cgroup *cgrp = seq_css(seq)->cgroup;
3609 seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp));
3610 seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags));
3615 static int cgroup_stat_show(struct seq_file *seq, void *v)
3617 struct cgroup *cgroup = seq_css(seq)->cgroup;
3619 seq_printf(seq, "nr_descendants %d\n",
3620 cgroup->nr_descendants);
3621 seq_printf(seq, "nr_dying_descendants %d\n",
3622 cgroup->nr_dying_descendants);
3627 static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq,
3628 struct cgroup *cgrp, int ssid)
3630 struct cgroup_subsys *ss = cgroup_subsys[ssid];
3631 struct cgroup_subsys_state *css;
3634 if (!ss->css_extra_stat_show)
3637 css = cgroup_tryget_css(cgrp, ss);
3641 ret = ss->css_extra_stat_show(seq, css);
3646 static int cpu_stat_show(struct seq_file *seq, void *v)
3648 struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
3651 cgroup_base_stat_cputime_show(seq);
3652 #ifdef CONFIG_CGROUP_SCHED
3653 ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id);
3659 static int cgroup_io_pressure_show(struct seq_file *seq, void *v)
3661 struct cgroup *cgrp = seq_css(seq)->cgroup;
3662 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3664 return psi_show(seq, psi, PSI_IO);
3666 static int cgroup_memory_pressure_show(struct seq_file *seq, void *v)
3668 struct cgroup *cgrp = seq_css(seq)->cgroup;
3669 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3671 return psi_show(seq, psi, PSI_MEM);
3673 static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
3675 struct cgroup *cgrp = seq_css(seq)->cgroup;
3676 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3678 return psi_show(seq, psi, PSI_CPU);
3681 static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, char *buf,
3682 size_t nbytes, enum psi_res res)
3684 struct cgroup_file_ctx *ctx = of->priv;
3685 struct psi_trigger *new;
3686 struct cgroup *cgrp;
3687 struct psi_group *psi;
3689 cgrp = cgroup_kn_lock_live(of->kn, false);
3694 cgroup_kn_unlock(of->kn);
3696 /* Allow only one trigger per file descriptor */
3697 if (ctx->psi.trigger) {
3702 psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3703 new = psi_trigger_create(psi, buf, nbytes, res);
3706 return PTR_ERR(new);
3709 smp_store_release(&ctx->psi.trigger, new);
3715 static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of,
3716 char *buf, size_t nbytes,
3719 return cgroup_pressure_write(of, buf, nbytes, PSI_IO);
3722 static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of,
3723 char *buf, size_t nbytes,
3726 return cgroup_pressure_write(of, buf, nbytes, PSI_MEM);
3729 static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of,
3730 char *buf, size_t nbytes,
3733 return cgroup_pressure_write(of, buf, nbytes, PSI_CPU);
3736 static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of,
3739 struct cgroup_file_ctx *ctx = of->priv;
3741 return psi_trigger_poll(&ctx->psi.trigger, of->file, pt);
3744 static void cgroup_pressure_release(struct kernfs_open_file *of)
3746 struct cgroup_file_ctx *ctx = of->priv;
3748 psi_trigger_destroy(ctx->psi.trigger);
3751 bool cgroup_psi_enabled(void)
3753 return (cgroup_feature_disable_mask & (1 << OPT_FEATURE_PRESSURE)) == 0;
3756 #else /* CONFIG_PSI */
3757 bool cgroup_psi_enabled(void)
3762 #endif /* CONFIG_PSI */
3764 static int cgroup_freeze_show(struct seq_file *seq, void *v)
3766 struct cgroup *cgrp = seq_css(seq)->cgroup;
3768 seq_printf(seq, "%d\n", cgrp->freezer.freeze);
3773 static ssize_t cgroup_freeze_write(struct kernfs_open_file *of,
3774 char *buf, size_t nbytes, loff_t off)
3776 struct cgroup *cgrp;
3780 ret = kstrtoint(strstrip(buf), 0, &freeze);
3784 if (freeze < 0 || freeze > 1)
3787 cgrp = cgroup_kn_lock_live(of->kn, false);
3791 cgroup_freeze(cgrp, freeze);
3793 cgroup_kn_unlock(of->kn);
3798 static void __cgroup_kill(struct cgroup *cgrp)
3800 struct css_task_iter it;
3801 struct task_struct *task;
3803 lockdep_assert_held(&cgroup_mutex);
3805 spin_lock_irq(&css_set_lock);
3806 set_bit(CGRP_KILL, &cgrp->flags);
3807 spin_unlock_irq(&css_set_lock);
3809 css_task_iter_start(&cgrp->self, CSS_TASK_ITER_PROCS | CSS_TASK_ITER_THREADED, &it);
3810 while ((task = css_task_iter_next(&it))) {
3811 /* Ignore kernel threads here. */
3812 if (task->flags & PF_KTHREAD)
3815 /* Skip tasks that are already dying. */
3816 if (__fatal_signal_pending(task))
3819 send_sig(SIGKILL, task, 0);
3821 css_task_iter_end(&it);
3823 spin_lock_irq(&css_set_lock);
3824 clear_bit(CGRP_KILL, &cgrp->flags);
3825 spin_unlock_irq(&css_set_lock);
3828 static void cgroup_kill(struct cgroup *cgrp)
3830 struct cgroup_subsys_state *css;
3831 struct cgroup *dsct;
3833 lockdep_assert_held(&cgroup_mutex);
3835 cgroup_for_each_live_descendant_pre(dsct, css, cgrp)
3836 __cgroup_kill(dsct);
3839 static ssize_t cgroup_kill_write(struct kernfs_open_file *of, char *buf,
3840 size_t nbytes, loff_t off)
3844 struct cgroup *cgrp;
3846 ret = kstrtoint(strstrip(buf), 0, &kill);
3853 cgrp = cgroup_kn_lock_live(of->kn, false);
3858 * Killing is a process directed operation, i.e. the whole thread-group
3859 * is taken down so act like we do for cgroup.procs and only make this
3860 * writable in non-threaded cgroups.
3862 if (cgroup_is_threaded(cgrp))
3867 cgroup_kn_unlock(of->kn);
3869 return ret ?: nbytes;
3872 static int cgroup_file_open(struct kernfs_open_file *of)
3874 struct cftype *cft = of_cft(of);
3875 struct cgroup_file_ctx *ctx;
3878 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
3882 ctx->ns = current->nsproxy->cgroup_ns;
3883 get_cgroup_ns(ctx->ns);
3889 ret = cft->open(of);
3891 put_cgroup_ns(ctx->ns);
3897 static void cgroup_file_release(struct kernfs_open_file *of)
3899 struct cftype *cft = of_cft(of);
3900 struct cgroup_file_ctx *ctx = of->priv;
3904 put_cgroup_ns(ctx->ns);
3908 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3909 size_t nbytes, loff_t off)
3911 struct cgroup_file_ctx *ctx = of->priv;
3912 struct cgroup *cgrp = of->kn->parent->priv;
3913 struct cftype *cft = of_cft(of);
3914 struct cgroup_subsys_state *css;
3921 * If namespaces are delegation boundaries, disallow writes to
3922 * files in an non-init namespace root from inside the namespace
3923 * except for the files explicitly marked delegatable -
3924 * cgroup.procs and cgroup.subtree_control.
3926 if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
3927 !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
3928 ctx->ns != &init_cgroup_ns && ctx->ns->root_cset->dfl_cgrp == cgrp)
3932 return cft->write(of, buf, nbytes, off);
3935 * kernfs guarantees that a file isn't deleted with operations in
3936 * flight, which means that the matching css is and stays alive and
3937 * doesn't need to be pinned. The RCU locking is not necessary
3938 * either. It's just for the convenience of using cgroup_css().
3941 css = cgroup_css(cgrp, cft->ss);
3944 if (cft->write_u64) {
3945 unsigned long long v;
3946 ret = kstrtoull(buf, 0, &v);
3948 ret = cft->write_u64(css, cft, v);
3949 } else if (cft->write_s64) {
3951 ret = kstrtoll(buf, 0, &v);
3953 ret = cft->write_s64(css, cft, v);
3958 return ret ?: nbytes;
3961 static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt)
3963 struct cftype *cft = of_cft(of);
3966 return cft->poll(of, pt);
3968 return kernfs_generic_poll(of, pt);
3971 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3973 return seq_cft(seq)->seq_start(seq, ppos);
3976 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3978 return seq_cft(seq)->seq_next(seq, v, ppos);
3981 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3983 if (seq_cft(seq)->seq_stop)
3984 seq_cft(seq)->seq_stop(seq, v);
3987 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3989 struct cftype *cft = seq_cft(m);
3990 struct cgroup_subsys_state *css = seq_css(m);
3993 return cft->seq_show(m, arg);
3996 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3997 else if (cft->read_s64)
3998 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
4004 static struct kernfs_ops cgroup_kf_single_ops = {
4005 .atomic_write_len = PAGE_SIZE,
4006 .open = cgroup_file_open,
4007 .release = cgroup_file_release,
4008 .write = cgroup_file_write,
4009 .poll = cgroup_file_poll,
4010 .seq_show = cgroup_seqfile_show,
4013 static struct kernfs_ops cgroup_kf_ops = {
4014 .atomic_write_len = PAGE_SIZE,
4015 .open = cgroup_file_open,
4016 .release = cgroup_file_release,
4017 .write = cgroup_file_write,
4018 .poll = cgroup_file_poll,
4019 .seq_start = cgroup_seqfile_start,
4020 .seq_next = cgroup_seqfile_next,
4021 .seq_stop = cgroup_seqfile_stop,
4022 .seq_show = cgroup_seqfile_show,
4025 /* set uid and gid of cgroup dirs and files to that of the creator */
4026 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
4028 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
4029 .ia_uid = current_fsuid(),
4030 .ia_gid = current_fsgid(), };
4032 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
4033 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
4036 return kernfs_setattr(kn, &iattr);
4039 static void cgroup_file_notify_timer(struct timer_list *timer)
4041 cgroup_file_notify(container_of(timer, struct cgroup_file,
4045 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
4048 char name[CGROUP_FILE_NAME_MAX];
4049 struct kernfs_node *kn;
4050 struct lock_class_key *key = NULL;
4053 #ifdef CONFIG_DEBUG_LOCK_ALLOC
4054 key = &cft->lockdep_key;
4056 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
4057 cgroup_file_mode(cft),
4058 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
4059 0, cft->kf_ops, cft,
4064 ret = cgroup_kn_set_ugid(kn);
4070 if (cft->file_offset) {
4071 struct cgroup_file *cfile = (void *)css + cft->file_offset;
4073 timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0);
4075 spin_lock_irq(&cgroup_file_kn_lock);
4077 spin_unlock_irq(&cgroup_file_kn_lock);
4084 * cgroup_addrm_files - add or remove files to a cgroup directory
4085 * @css: the target css
4086 * @cgrp: the target cgroup (usually css->cgroup)
4087 * @cfts: array of cftypes to be added
4088 * @is_add: whether to add or remove
4090 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
4091 * For removals, this function never fails.
4093 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
4094 struct cgroup *cgrp, struct cftype cfts[],
4097 struct cftype *cft, *cft_end = NULL;
4100 lockdep_assert_held(&cgroup_mutex);
4103 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
4104 /* does cft->flags tell us to skip this file on @cgrp? */
4105 if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
4107 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
4109 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
4111 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
4113 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
4115 if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug)
4118 ret = cgroup_add_file(css, cgrp, cft);
4120 pr_warn("%s: failed to add %s, err=%d\n",
4121 __func__, cft->name, ret);
4127 cgroup_rm_file(cgrp, cft);
4133 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
4135 struct cgroup_subsys *ss = cfts[0].ss;
4136 struct cgroup *root = &ss->root->cgrp;
4137 struct cgroup_subsys_state *css;
4140 lockdep_assert_held(&cgroup_mutex);
4142 /* add/rm files for all cgroups created before */
4143 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
4144 struct cgroup *cgrp = css->cgroup;
4146 if (!(css->flags & CSS_VISIBLE))
4149 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
4155 kernfs_activate(root->kn);
4159 static void cgroup_exit_cftypes(struct cftype *cfts)
4163 for (cft = cfts; cft->name[0] != '\0'; cft++) {
4164 /* free copy for custom atomic_write_len, see init_cftypes() */
4165 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
4170 /* revert flags set by cgroup core while adding @cfts */
4171 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
4175 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4179 for (cft = cfts; cft->name[0] != '\0'; cft++) {
4180 struct kernfs_ops *kf_ops;
4182 WARN_ON(cft->ss || cft->kf_ops);
4184 if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
4188 kf_ops = &cgroup_kf_ops;
4190 kf_ops = &cgroup_kf_single_ops;
4193 * Ugh... if @cft wants a custom max_write_len, we need to
4194 * make a copy of kf_ops to set its atomic_write_len.
4196 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
4197 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
4199 cgroup_exit_cftypes(cfts);
4202 kf_ops->atomic_write_len = cft->max_write_len;
4205 cft->kf_ops = kf_ops;
4212 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
4214 lockdep_assert_held(&cgroup_mutex);
4216 if (!cfts || !cfts[0].ss)
4219 list_del(&cfts->node);
4220 cgroup_apply_cftypes(cfts, false);
4221 cgroup_exit_cftypes(cfts);
4226 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4227 * @cfts: zero-length name terminated array of cftypes
4229 * Unregister @cfts. Files described by @cfts are removed from all
4230 * existing cgroups and all future cgroups won't have them either. This
4231 * function can be called anytime whether @cfts' subsys is attached or not.
4233 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4236 int cgroup_rm_cftypes(struct cftype *cfts)
4240 mutex_lock(&cgroup_mutex);
4241 ret = cgroup_rm_cftypes_locked(cfts);
4242 mutex_unlock(&cgroup_mutex);
4247 * cgroup_add_cftypes - add an array of cftypes to a subsystem
4248 * @ss: target cgroup subsystem
4249 * @cfts: zero-length name terminated array of cftypes
4251 * Register @cfts to @ss. Files described by @cfts are created for all
4252 * existing cgroups to which @ss is attached and all future cgroups will
4253 * have them too. This function can be called anytime whether @ss is
4256 * Returns 0 on successful registration, -errno on failure. Note that this
4257 * function currently returns 0 as long as @cfts registration is successful
4258 * even if some file creation attempts on existing cgroups fail.
4260 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4264 if (!cgroup_ssid_enabled(ss->id))
4267 if (!cfts || cfts[0].name[0] == '\0')
4270 ret = cgroup_init_cftypes(ss, cfts);
4274 mutex_lock(&cgroup_mutex);
4276 list_add_tail(&cfts->node, &ss->cfts);
4277 ret = cgroup_apply_cftypes(cfts, true);
4279 cgroup_rm_cftypes_locked(cfts);
4281 mutex_unlock(&cgroup_mutex);
4286 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4287 * @ss: target cgroup subsystem
4288 * @cfts: zero-length name terminated array of cftypes
4290 * Similar to cgroup_add_cftypes() but the added files are only used for
4291 * the default hierarchy.
4293 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4297 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4298 cft->flags |= __CFTYPE_ONLY_ON_DFL;
4299 return cgroup_add_cftypes(ss, cfts);
4303 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4304 * @ss: target cgroup subsystem
4305 * @cfts: zero-length name terminated array of cftypes
4307 * Similar to cgroup_add_cftypes() but the added files are only used for
4308 * the legacy hierarchies.
4310 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4314 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4315 cft->flags |= __CFTYPE_NOT_ON_DFL;
4316 return cgroup_add_cftypes(ss, cfts);
4320 * cgroup_file_notify - generate a file modified event for a cgroup_file
4321 * @cfile: target cgroup_file
4323 * @cfile must have been obtained by setting cftype->file_offset.
4325 void cgroup_file_notify(struct cgroup_file *cfile)
4327 unsigned long flags;
4329 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
4331 unsigned long last = cfile->notified_at;
4332 unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV;
4334 if (time_in_range(jiffies, last, next)) {
4335 timer_reduce(&cfile->notify_timer, next);
4337 kernfs_notify(cfile->kn);
4338 cfile->notified_at = jiffies;
4341 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
4345 * css_next_child - find the next child of a given css
4346 * @pos: the current position (%NULL to initiate traversal)
4347 * @parent: css whose children to walk
4349 * This function returns the next child of @parent and should be called
4350 * under either cgroup_mutex or RCU read lock. The only requirement is
4351 * that @parent and @pos are accessible. The next sibling is guaranteed to
4352 * be returned regardless of their states.
4354 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4355 * css which finished ->css_online() is guaranteed to be visible in the
4356 * future iterations and will stay visible until the last reference is put.
4357 * A css which hasn't finished ->css_online() or already finished
4358 * ->css_offline() may show up during traversal. It's each subsystem's
4359 * responsibility to synchronize against on/offlining.
4361 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
4362 struct cgroup_subsys_state *parent)
4364 struct cgroup_subsys_state *next;
4366 cgroup_assert_mutex_or_rcu_locked();
4369 * @pos could already have been unlinked from the sibling list.
4370 * Once a cgroup is removed, its ->sibling.next is no longer
4371 * updated when its next sibling changes. CSS_RELEASED is set when
4372 * @pos is taken off list, at which time its next pointer is valid,
4373 * and, as releases are serialized, the one pointed to by the next
4374 * pointer is guaranteed to not have started release yet. This
4375 * implies that if we observe !CSS_RELEASED on @pos in this RCU
4376 * critical section, the one pointed to by its next pointer is
4377 * guaranteed to not have finished its RCU grace period even if we
4378 * have dropped rcu_read_lock() in-between iterations.
4380 * If @pos has CSS_RELEASED set, its next pointer can't be
4381 * dereferenced; however, as each css is given a monotonically
4382 * increasing unique serial number and always appended to the
4383 * sibling list, the next one can be found by walking the parent's
4384 * children until the first css with higher serial number than
4385 * @pos's. While this path can be slower, it happens iff iteration
4386 * races against release and the race window is very small.
4389 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
4390 } else if (likely(!(pos->flags & CSS_RELEASED))) {
4391 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
4393 list_for_each_entry_rcu(next, &parent->children, sibling,
4394 lockdep_is_held(&cgroup_mutex))
4395 if (next->serial_nr > pos->serial_nr)
4400 * @next, if not pointing to the head, can be dereferenced and is
4403 if (&next->sibling != &parent->children)
4409 * css_next_descendant_pre - find the next descendant for pre-order walk
4410 * @pos: the current position (%NULL to initiate traversal)
4411 * @root: css whose descendants to walk
4413 * To be used by css_for_each_descendant_pre(). Find the next descendant
4414 * to visit for pre-order traversal of @root's descendants. @root is
4415 * included in the iteration and the first node to be visited.
4417 * While this function requires cgroup_mutex or RCU read locking, it
4418 * doesn't require the whole traversal to be contained in a single critical
4419 * section. This function will return the correct next descendant as long
4420 * as both @pos and @root are accessible and @pos is a descendant of @root.
4422 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4423 * css which finished ->css_online() is guaranteed to be visible in the
4424 * future iterations and will stay visible until the last reference is put.
4425 * A css which hasn't finished ->css_online() or already finished
4426 * ->css_offline() may show up during traversal. It's each subsystem's
4427 * responsibility to synchronize against on/offlining.
4429 struct cgroup_subsys_state *
4430 css_next_descendant_pre(struct cgroup_subsys_state *pos,
4431 struct cgroup_subsys_state *root)
4433 struct cgroup_subsys_state *next;
4435 cgroup_assert_mutex_or_rcu_locked();
4437 /* if first iteration, visit @root */
4441 /* visit the first child if exists */
4442 next = css_next_child(NULL, pos);
4446 /* no child, visit my or the closest ancestor's next sibling */
4447 while (pos != root) {
4448 next = css_next_child(pos, pos->parent);
4456 EXPORT_SYMBOL_GPL(css_next_descendant_pre);
4459 * css_rightmost_descendant - return the rightmost descendant of a css
4460 * @pos: css of interest
4462 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4463 * is returned. This can be used during pre-order traversal to skip
4466 * While this function requires cgroup_mutex or RCU read locking, it
4467 * doesn't require the whole traversal to be contained in a single critical
4468 * section. This function will return the correct rightmost descendant as
4469 * long as @pos is accessible.
4471 struct cgroup_subsys_state *
4472 css_rightmost_descendant(struct cgroup_subsys_state *pos)
4474 struct cgroup_subsys_state *last, *tmp;
4476 cgroup_assert_mutex_or_rcu_locked();
4480 /* ->prev isn't RCU safe, walk ->next till the end */
4482 css_for_each_child(tmp, last)
4489 static struct cgroup_subsys_state *
4490 css_leftmost_descendant(struct cgroup_subsys_state *pos)
4492 struct cgroup_subsys_state *last;
4496 pos = css_next_child(NULL, pos);
4503 * css_next_descendant_post - find the next descendant for post-order walk
4504 * @pos: the current position (%NULL to initiate traversal)
4505 * @root: css whose descendants to walk
4507 * To be used by css_for_each_descendant_post(). Find the next descendant
4508 * to visit for post-order traversal of @root's descendants. @root is
4509 * included in the iteration and the last node to be visited.
4511 * While this function requires cgroup_mutex or RCU read locking, it
4512 * doesn't require the whole traversal to be contained in a single critical
4513 * section. This function will return the correct next descendant as long
4514 * as both @pos and @cgroup are accessible and @pos is a descendant of
4517 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4518 * css which finished ->css_online() is guaranteed to be visible in the
4519 * future iterations and will stay visible until the last reference is put.
4520 * A css which hasn't finished ->css_online() or already finished
4521 * ->css_offline() may show up during traversal. It's each subsystem's
4522 * responsibility to synchronize against on/offlining.
4524 struct cgroup_subsys_state *
4525 css_next_descendant_post(struct cgroup_subsys_state *pos,
4526 struct cgroup_subsys_state *root)
4528 struct cgroup_subsys_state *next;
4530 cgroup_assert_mutex_or_rcu_locked();
4532 /* if first iteration, visit leftmost descendant which may be @root */
4534 return css_leftmost_descendant(root);
4536 /* if we visited @root, we're done */
4540 /* if there's an unvisited sibling, visit its leftmost descendant */
4541 next = css_next_child(pos, pos->parent);
4543 return css_leftmost_descendant(next);
4545 /* no sibling left, visit parent */
4550 * css_has_online_children - does a css have online children
4551 * @css: the target css
4553 * Returns %true if @css has any online children; otherwise, %false. This
4554 * function can be called from any context but the caller is responsible
4555 * for synchronizing against on/offlining as necessary.
4557 bool css_has_online_children(struct cgroup_subsys_state *css)
4559 struct cgroup_subsys_state *child;
4563 css_for_each_child(child, css) {
4564 if (child->flags & CSS_ONLINE) {
4573 static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
4575 struct list_head *l;
4576 struct cgrp_cset_link *link;
4577 struct css_set *cset;
4579 lockdep_assert_held(&css_set_lock);
4581 /* find the next threaded cset */
4582 if (it->tcset_pos) {
4583 l = it->tcset_pos->next;
4585 if (l != it->tcset_head) {
4587 return container_of(l, struct css_set,
4588 threaded_csets_node);
4591 it->tcset_pos = NULL;
4594 /* find the next cset */
4597 if (l == it->cset_head) {
4598 it->cset_pos = NULL;
4603 cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
4605 link = list_entry(l, struct cgrp_cset_link, cset_link);
4611 /* initialize threaded css_set walking */
4612 if (it->flags & CSS_TASK_ITER_THREADED) {
4614 put_css_set_locked(it->cur_dcset);
4615 it->cur_dcset = cset;
4618 it->tcset_head = &cset->threaded_csets;
4619 it->tcset_pos = &cset->threaded_csets;
4626 * css_task_iter_advance_css_set - advance a task iterator to the next css_set
4627 * @it: the iterator to advance
4629 * Advance @it to the next css_set to walk.
4631 static void css_task_iter_advance_css_set(struct css_task_iter *it)
4633 struct css_set *cset;
4635 lockdep_assert_held(&css_set_lock);
4637 /* Advance to the next non-empty css_set and find first non-empty tasks list*/
4638 while ((cset = css_task_iter_next_css_set(it))) {
4639 if (!list_empty(&cset->tasks)) {
4640 it->cur_tasks_head = &cset->tasks;
4642 } else if (!list_empty(&cset->mg_tasks)) {
4643 it->cur_tasks_head = &cset->mg_tasks;
4645 } else if (!list_empty(&cset->dying_tasks)) {
4646 it->cur_tasks_head = &cset->dying_tasks;
4651 it->task_pos = NULL;
4654 it->task_pos = it->cur_tasks_head->next;
4657 * We don't keep css_sets locked across iteration steps and thus
4658 * need to take steps to ensure that iteration can be resumed after
4659 * the lock is re-acquired. Iteration is performed at two levels -
4660 * css_sets and tasks in them.
4662 * Once created, a css_set never leaves its cgroup lists, so a
4663 * pinned css_set is guaranteed to stay put and we can resume
4664 * iteration afterwards.
4666 * Tasks may leave @cset across iteration steps. This is resolved
4667 * by registering each iterator with the css_set currently being
4668 * walked and making css_set_move_task() advance iterators whose
4669 * next task is leaving.
4672 list_del(&it->iters_node);
4673 put_css_set_locked(it->cur_cset);
4676 it->cur_cset = cset;
4677 list_add(&it->iters_node, &cset->task_iters);
4680 static void css_task_iter_skip(struct css_task_iter *it,
4681 struct task_struct *task)
4683 lockdep_assert_held(&css_set_lock);
4685 if (it->task_pos == &task->cg_list) {
4686 it->task_pos = it->task_pos->next;
4687 it->flags |= CSS_TASK_ITER_SKIPPED;
4691 static void css_task_iter_advance(struct css_task_iter *it)
4693 struct task_struct *task;
4695 lockdep_assert_held(&css_set_lock);
4699 * Advance iterator to find next entry. We go through cset
4700 * tasks, mg_tasks and dying_tasks, when consumed we move onto
4703 if (it->flags & CSS_TASK_ITER_SKIPPED)
4704 it->flags &= ~CSS_TASK_ITER_SKIPPED;
4706 it->task_pos = it->task_pos->next;
4708 if (it->task_pos == &it->cur_cset->tasks) {
4709 it->cur_tasks_head = &it->cur_cset->mg_tasks;
4710 it->task_pos = it->cur_tasks_head->next;
4712 if (it->task_pos == &it->cur_cset->mg_tasks) {
4713 it->cur_tasks_head = &it->cur_cset->dying_tasks;
4714 it->task_pos = it->cur_tasks_head->next;
4716 if (it->task_pos == &it->cur_cset->dying_tasks)
4717 css_task_iter_advance_css_set(it);
4719 /* called from start, proceed to the first cset */
4720 css_task_iter_advance_css_set(it);
4726 task = list_entry(it->task_pos, struct task_struct, cg_list);
4728 if (it->flags & CSS_TASK_ITER_PROCS) {
4729 /* if PROCS, skip over tasks which aren't group leaders */
4730 if (!thread_group_leader(task))
4733 /* and dying leaders w/o live member threads */
4734 if (it->cur_tasks_head == &it->cur_cset->dying_tasks &&
4735 !atomic_read(&task->signal->live))
4738 /* skip all dying ones */
4739 if (it->cur_tasks_head == &it->cur_cset->dying_tasks)
4745 * css_task_iter_start - initiate task iteration
4746 * @css: the css to walk tasks of
4747 * @flags: CSS_TASK_ITER_* flags
4748 * @it: the task iterator to use
4750 * Initiate iteration through the tasks of @css. The caller can call
4751 * css_task_iter_next() to walk through the tasks until the function
4752 * returns NULL. On completion of iteration, css_task_iter_end() must be
4755 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
4756 struct css_task_iter *it)
4758 memset(it, 0, sizeof(*it));
4760 spin_lock_irq(&css_set_lock);
4765 if (CGROUP_HAS_SUBSYS_CONFIG && it->ss)
4766 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4768 it->cset_pos = &css->cgroup->cset_links;
4770 it->cset_head = it->cset_pos;
4772 css_task_iter_advance(it);
4774 spin_unlock_irq(&css_set_lock);
4778 * css_task_iter_next - return the next task for the iterator
4779 * @it: the task iterator being iterated
4781 * The "next" function for task iteration. @it should have been
4782 * initialized via css_task_iter_start(). Returns NULL when the iteration
4785 struct task_struct *css_task_iter_next(struct css_task_iter *it)
4788 put_task_struct(it->cur_task);
4789 it->cur_task = NULL;
4792 spin_lock_irq(&css_set_lock);
4794 /* @it may be half-advanced by skips, finish advancing */
4795 if (it->flags & CSS_TASK_ITER_SKIPPED)
4796 css_task_iter_advance(it);
4799 it->cur_task = list_entry(it->task_pos, struct task_struct,
4801 get_task_struct(it->cur_task);
4802 css_task_iter_advance(it);
4805 spin_unlock_irq(&css_set_lock);
4807 return it->cur_task;
4811 * css_task_iter_end - finish task iteration
4812 * @it: the task iterator to finish
4814 * Finish task iteration started by css_task_iter_start().
4816 void css_task_iter_end(struct css_task_iter *it)
4819 spin_lock_irq(&css_set_lock);
4820 list_del(&it->iters_node);
4821 put_css_set_locked(it->cur_cset);
4822 spin_unlock_irq(&css_set_lock);
4826 put_css_set(it->cur_dcset);
4829 put_task_struct(it->cur_task);
4832 static void cgroup_procs_release(struct kernfs_open_file *of)
4834 struct cgroup_file_ctx *ctx = of->priv;
4836 if (ctx->procs.started)
4837 css_task_iter_end(&ctx->procs.iter);
4840 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
4842 struct kernfs_open_file *of = s->private;
4843 struct cgroup_file_ctx *ctx = of->priv;
4848 return css_task_iter_next(&ctx->procs.iter);
4851 static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
4852 unsigned int iter_flags)
4854 struct kernfs_open_file *of = s->private;
4855 struct cgroup *cgrp = seq_css(s)->cgroup;
4856 struct cgroup_file_ctx *ctx = of->priv;
4857 struct css_task_iter *it = &ctx->procs.iter;
4860 * When a seq_file is seeked, it's always traversed sequentially
4861 * from position 0, so we can simply keep iterating on !0 *pos.
4863 if (!ctx->procs.started) {
4864 if (WARN_ON_ONCE((*pos)))
4865 return ERR_PTR(-EINVAL);
4866 css_task_iter_start(&cgrp->self, iter_flags, it);
4867 ctx->procs.started = true;
4868 } else if (!(*pos)) {
4869 css_task_iter_end(it);
4870 css_task_iter_start(&cgrp->self, iter_flags, it);
4872 return it->cur_task;
4874 return cgroup_procs_next(s, NULL, NULL);
4877 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
4879 struct cgroup *cgrp = seq_css(s)->cgroup;
4882 * All processes of a threaded subtree belong to the domain cgroup
4883 * of the subtree. Only threads can be distributed across the
4884 * subtree. Reject reads on cgroup.procs in the subtree proper.
4885 * They're always empty anyway.
4887 if (cgroup_is_threaded(cgrp))
4888 return ERR_PTR(-EOPNOTSUPP);
4890 return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
4891 CSS_TASK_ITER_THREADED);
4894 static int cgroup_procs_show(struct seq_file *s, void *v)
4896 seq_printf(s, "%d\n", task_pid_vnr(v));
4900 static int cgroup_may_write(const struct cgroup *cgrp, struct super_block *sb)
4903 struct inode *inode;
4905 lockdep_assert_held(&cgroup_mutex);
4907 inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
4911 ret = inode_permission(&init_user_ns, inode, MAY_WRITE);
4916 static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
4917 struct cgroup *dst_cgrp,
4918 struct super_block *sb,
4919 struct cgroup_namespace *ns)
4921 struct cgroup *com_cgrp = src_cgrp;
4924 lockdep_assert_held(&cgroup_mutex);
4926 /* find the common ancestor */
4927 while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
4928 com_cgrp = cgroup_parent(com_cgrp);
4930 /* %current should be authorized to migrate to the common ancestor */
4931 ret = cgroup_may_write(com_cgrp, sb);
4936 * If namespaces are delegation boundaries, %current must be able
4937 * to see both source and destination cgroups from its namespace.
4939 if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
4940 (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
4941 !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
4947 static int cgroup_attach_permissions(struct cgroup *src_cgrp,
4948 struct cgroup *dst_cgrp,
4949 struct super_block *sb, bool threadgroup,
4950 struct cgroup_namespace *ns)
4954 ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, sb, ns);
4958 ret = cgroup_migrate_vet_dst(dst_cgrp);
4962 if (!threadgroup && (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp))
4968 static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
4971 struct cgroup_file_ctx *ctx = of->priv;
4972 struct cgroup *src_cgrp, *dst_cgrp;
4973 struct task_struct *task;
4974 const struct cred *saved_cred;
4976 bool threadgroup_locked;
4978 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4982 task = cgroup_procs_write_start(buf, threadgroup, &threadgroup_locked);
4983 ret = PTR_ERR_OR_ZERO(task);
4987 /* find the source cgroup */
4988 spin_lock_irq(&css_set_lock);
4989 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4990 spin_unlock_irq(&css_set_lock);
4993 * Process and thread migrations follow same delegation rule. Check
4994 * permissions using the credentials from file open to protect against
4995 * inherited fd attacks.
4997 saved_cred = override_creds(of->file->f_cred);
4998 ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
4999 of->file->f_path.dentry->d_sb,
5000 threadgroup, ctx->ns);
5001 revert_creds(saved_cred);
5005 ret = cgroup_attach_task(dst_cgrp, task, threadgroup);
5008 cgroup_procs_write_finish(task, threadgroup_locked);
5010 cgroup_kn_unlock(of->kn);
5015 static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
5016 char *buf, size_t nbytes, loff_t off)
5018 return __cgroup_procs_write(of, buf, true) ?: nbytes;
5021 static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
5023 return __cgroup_procs_start(s, pos, 0);
5026 static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
5027 char *buf, size_t nbytes, loff_t off)
5029 return __cgroup_procs_write(of, buf, false) ?: nbytes;
5032 /* cgroup core interface files for the default hierarchy */
5033 static struct cftype cgroup_base_files[] = {
5035 .name = "cgroup.type",
5036 .flags = CFTYPE_NOT_ON_ROOT,
5037 .seq_show = cgroup_type_show,
5038 .write = cgroup_type_write,
5041 .name = "cgroup.procs",
5042 .flags = CFTYPE_NS_DELEGATABLE,
5043 .file_offset = offsetof(struct cgroup, procs_file),
5044 .release = cgroup_procs_release,
5045 .seq_start = cgroup_procs_start,
5046 .seq_next = cgroup_procs_next,
5047 .seq_show = cgroup_procs_show,
5048 .write = cgroup_procs_write,
5051 .name = "cgroup.threads",
5052 .flags = CFTYPE_NS_DELEGATABLE,
5053 .release = cgroup_procs_release,
5054 .seq_start = cgroup_threads_start,
5055 .seq_next = cgroup_procs_next,
5056 .seq_show = cgroup_procs_show,
5057 .write = cgroup_threads_write,
5060 .name = "cgroup.controllers",
5061 .seq_show = cgroup_controllers_show,
5064 .name = "cgroup.subtree_control",
5065 .flags = CFTYPE_NS_DELEGATABLE,
5066 .seq_show = cgroup_subtree_control_show,
5067 .write = cgroup_subtree_control_write,
5070 .name = "cgroup.events",
5071 .flags = CFTYPE_NOT_ON_ROOT,
5072 .file_offset = offsetof(struct cgroup, events_file),
5073 .seq_show = cgroup_events_show,
5076 .name = "cgroup.max.descendants",
5077 .seq_show = cgroup_max_descendants_show,
5078 .write = cgroup_max_descendants_write,
5081 .name = "cgroup.max.depth",
5082 .seq_show = cgroup_max_depth_show,
5083 .write = cgroup_max_depth_write,
5086 .name = "cgroup.stat",
5087 .seq_show = cgroup_stat_show,
5090 .name = "cgroup.freeze",
5091 .flags = CFTYPE_NOT_ON_ROOT,
5092 .seq_show = cgroup_freeze_show,
5093 .write = cgroup_freeze_write,
5096 .name = "cgroup.kill",
5097 .flags = CFTYPE_NOT_ON_ROOT,
5098 .write = cgroup_kill_write,
5102 .seq_show = cpu_stat_show,
5106 .name = "io.pressure",
5107 .flags = CFTYPE_PRESSURE,
5108 .seq_show = cgroup_io_pressure_show,
5109 .write = cgroup_io_pressure_write,
5110 .poll = cgroup_pressure_poll,
5111 .release = cgroup_pressure_release,
5114 .name = "memory.pressure",
5115 .flags = CFTYPE_PRESSURE,
5116 .seq_show = cgroup_memory_pressure_show,
5117 .write = cgroup_memory_pressure_write,
5118 .poll = cgroup_pressure_poll,
5119 .release = cgroup_pressure_release,
5122 .name = "cpu.pressure",
5123 .flags = CFTYPE_PRESSURE,
5124 .seq_show = cgroup_cpu_pressure_show,
5125 .write = cgroup_cpu_pressure_write,
5126 .poll = cgroup_pressure_poll,
5127 .release = cgroup_pressure_release,
5129 #endif /* CONFIG_PSI */
5134 * css destruction is four-stage process.
5136 * 1. Destruction starts. Killing of the percpu_ref is initiated.
5137 * Implemented in kill_css().
5139 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
5140 * and thus css_tryget_online() is guaranteed to fail, the css can be
5141 * offlined by invoking offline_css(). After offlining, the base ref is
5142 * put. Implemented in css_killed_work_fn().
5144 * 3. When the percpu_ref reaches zero, the only possible remaining
5145 * accessors are inside RCU read sections. css_release() schedules the
5148 * 4. After the grace period, the css can be freed. Implemented in
5149 * css_free_work_fn().
5151 * It is actually hairier because both step 2 and 4 require process context
5152 * and thus involve punting to css->destroy_work adding two additional
5153 * steps to the already complex sequence.
5155 static void css_free_rwork_fn(struct work_struct *work)
5157 struct cgroup_subsys_state *css = container_of(to_rcu_work(work),
5158 struct cgroup_subsys_state, destroy_rwork);
5159 struct cgroup_subsys *ss = css->ss;
5160 struct cgroup *cgrp = css->cgroup;
5162 percpu_ref_exit(&css->refcnt);
5166 struct cgroup_subsys_state *parent = css->parent;
5170 cgroup_idr_remove(&ss->css_idr, id);
5176 /* cgroup free path */
5177 atomic_dec(&cgrp->root->nr_cgrps);
5178 cgroup1_pidlist_destroy_all(cgrp);
5179 cancel_work_sync(&cgrp->release_agent_work);
5181 if (cgroup_parent(cgrp)) {
5183 * We get a ref to the parent, and put the ref when
5184 * this cgroup is being freed, so it's guaranteed
5185 * that the parent won't be destroyed before its
5188 cgroup_put(cgroup_parent(cgrp));
5189 kernfs_put(cgrp->kn);
5190 psi_cgroup_free(cgrp);
5191 cgroup_rstat_exit(cgrp);
5195 * This is root cgroup's refcnt reaching zero,
5196 * which indicates that the root should be
5199 cgroup_destroy_root(cgrp->root);
5204 static void css_release_work_fn(struct work_struct *work)
5206 struct cgroup_subsys_state *css =
5207 container_of(work, struct cgroup_subsys_state, destroy_work);
5208 struct cgroup_subsys *ss = css->ss;
5209 struct cgroup *cgrp = css->cgroup;
5211 mutex_lock(&cgroup_mutex);
5213 css->flags |= CSS_RELEASED;
5214 list_del_rcu(&css->sibling);
5217 /* css release path */
5218 if (!list_empty(&css->rstat_css_node)) {
5219 cgroup_rstat_flush(cgrp);
5220 list_del_rcu(&css->rstat_css_node);
5223 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
5224 if (ss->css_released)
5225 ss->css_released(css);
5227 struct cgroup *tcgrp;
5229 /* cgroup release path */
5230 TRACE_CGROUP_PATH(release, cgrp);
5232 cgroup_rstat_flush(cgrp);
5234 spin_lock_irq(&css_set_lock);
5235 for (tcgrp = cgroup_parent(cgrp); tcgrp;
5236 tcgrp = cgroup_parent(tcgrp))
5237 tcgrp->nr_dying_descendants--;
5238 spin_unlock_irq(&css_set_lock);
5241 * There are two control paths which try to determine
5242 * cgroup from dentry without going through kernfs -
5243 * cgroupstats_build() and css_tryget_online_from_dir().
5244 * Those are supported by RCU protecting clearing of
5245 * cgrp->kn->priv backpointer.
5248 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
5252 mutex_unlock(&cgroup_mutex);
5254 INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5255 queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5258 static void css_release(struct percpu_ref *ref)
5260 struct cgroup_subsys_state *css =
5261 container_of(ref, struct cgroup_subsys_state, refcnt);
5263 INIT_WORK(&css->destroy_work, css_release_work_fn);
5264 queue_work(cgroup_destroy_wq, &css->destroy_work);
5267 static void init_and_link_css(struct cgroup_subsys_state *css,
5268 struct cgroup_subsys *ss, struct cgroup *cgrp)
5270 lockdep_assert_held(&cgroup_mutex);
5272 cgroup_get_live(cgrp);
5274 memset(css, 0, sizeof(*css));
5278 INIT_LIST_HEAD(&css->sibling);
5279 INIT_LIST_HEAD(&css->children);
5280 INIT_LIST_HEAD(&css->rstat_css_node);
5281 css->serial_nr = css_serial_nr_next++;
5282 atomic_set(&css->online_cnt, 0);
5284 if (cgroup_parent(cgrp)) {
5285 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
5286 css_get(css->parent);
5289 if (ss->css_rstat_flush)
5290 list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list);
5292 BUG_ON(cgroup_css(cgrp, ss));
5295 /* invoke ->css_online() on a new CSS and mark it online if successful */
5296 static int online_css(struct cgroup_subsys_state *css)
5298 struct cgroup_subsys *ss = css->ss;
5301 lockdep_assert_held(&cgroup_mutex);
5304 ret = ss->css_online(css);
5306 css->flags |= CSS_ONLINE;
5307 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
5309 atomic_inc(&css->online_cnt);
5311 atomic_inc(&css->parent->online_cnt);
5316 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5317 static void offline_css(struct cgroup_subsys_state *css)
5319 struct cgroup_subsys *ss = css->ss;
5321 lockdep_assert_held(&cgroup_mutex);
5323 if (!(css->flags & CSS_ONLINE))
5326 if (ss->css_offline)
5327 ss->css_offline(css);
5329 css->flags &= ~CSS_ONLINE;
5330 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
5332 wake_up_all(&css->cgroup->offline_waitq);
5336 * css_create - create a cgroup_subsys_state
5337 * @cgrp: the cgroup new css will be associated with
5338 * @ss: the subsys of new css
5340 * Create a new css associated with @cgrp - @ss pair. On success, the new
5341 * css is online and installed in @cgrp. This function doesn't create the
5342 * interface files. Returns 0 on success, -errno on failure.
5344 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5345 struct cgroup_subsys *ss)
5347 struct cgroup *parent = cgroup_parent(cgrp);
5348 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
5349 struct cgroup_subsys_state *css;
5352 lockdep_assert_held(&cgroup_mutex);
5354 css = ss->css_alloc(parent_css);
5356 css = ERR_PTR(-ENOMEM);
5360 init_and_link_css(css, ss, cgrp);
5362 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
5366 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
5371 /* @css is ready to be brought online now, make it visible */
5372 list_add_tail_rcu(&css->sibling, &parent_css->children);
5373 cgroup_idr_replace(&ss->css_idr, css, css->id);
5375 err = online_css(css);
5382 list_del_rcu(&css->sibling);
5384 list_del_rcu(&css->rstat_css_node);
5385 INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5386 queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5387 return ERR_PTR(err);
5391 * The returned cgroup is fully initialized including its control mask, but
5392 * it isn't associated with its kernfs_node and doesn't have the control
5395 static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
5398 struct cgroup_root *root = parent->root;
5399 struct cgroup *cgrp, *tcgrp;
5400 struct kernfs_node *kn;
5401 int level = parent->level + 1;
5404 /* allocate the cgroup and its ID, 0 is reserved for the root */
5405 cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
5408 return ERR_PTR(-ENOMEM);
5410 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
5414 ret = cgroup_rstat_init(cgrp);
5416 goto out_cancel_ref;
5418 /* create the directory */
5419 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5426 init_cgroup_housekeeping(cgrp);
5428 cgrp->self.parent = &parent->self;
5430 cgrp->level = level;
5432 ret = psi_cgroup_alloc(cgrp);
5434 goto out_kernfs_remove;
5436 ret = cgroup_bpf_inherit(cgrp);
5441 * New cgroup inherits effective freeze counter, and
5442 * if the parent has to be frozen, the child has too.
5444 cgrp->freezer.e_freeze = parent->freezer.e_freeze;
5445 if (cgrp->freezer.e_freeze) {
5447 * Set the CGRP_FREEZE flag, so when a process will be
5448 * attached to the child cgroup, it will become frozen.
5449 * At this point the new cgroup is unpopulated, so we can
5450 * consider it frozen immediately.
5452 set_bit(CGRP_FREEZE, &cgrp->flags);
5453 set_bit(CGRP_FROZEN, &cgrp->flags);
5456 spin_lock_irq(&css_set_lock);
5457 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5458 cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
5460 if (tcgrp != cgrp) {
5461 tcgrp->nr_descendants++;
5464 * If the new cgroup is frozen, all ancestor cgroups
5465 * get a new frozen descendant, but their state can't
5466 * change because of this.
5468 if (cgrp->freezer.e_freeze)
5469 tcgrp->freezer.nr_frozen_descendants++;
5472 spin_unlock_irq(&css_set_lock);
5474 if (notify_on_release(parent))
5475 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5477 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5478 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
5480 cgrp->self.serial_nr = css_serial_nr_next++;
5482 /* allocation complete, commit to creation */
5483 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
5484 atomic_inc(&root->nr_cgrps);
5485 cgroup_get_live(parent);
5488 * On the default hierarchy, a child doesn't automatically inherit
5489 * subtree_control from the parent. Each is configured manually.
5491 if (!cgroup_on_dfl(cgrp))
5492 cgrp->subtree_control = cgroup_control(cgrp);
5494 cgroup_propagate_control(cgrp);
5499 psi_cgroup_free(cgrp);
5501 kernfs_remove(cgrp->kn);
5503 cgroup_rstat_exit(cgrp);
5505 percpu_ref_exit(&cgrp->self.refcnt);
5508 return ERR_PTR(ret);
5511 static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
5513 struct cgroup *cgroup;
5517 lockdep_assert_held(&cgroup_mutex);
5519 for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
5520 if (cgroup->nr_descendants >= cgroup->max_descendants)
5523 if (level > cgroup->max_depth)
5534 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
5536 struct cgroup *parent, *cgrp;
5539 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5540 if (strchr(name, '\n'))
5543 parent = cgroup_kn_lock_live(parent_kn, false);
5547 if (!cgroup_check_hierarchy_limits(parent)) {
5552 cgrp = cgroup_create(parent, name, mode);
5554 ret = PTR_ERR(cgrp);
5559 * This extra ref will be put in cgroup_free_fn() and guarantees
5560 * that @cgrp->kn is always accessible.
5562 kernfs_get(cgrp->kn);
5564 ret = cgroup_kn_set_ugid(cgrp->kn);
5568 ret = css_populate_dir(&cgrp->self);
5572 ret = cgroup_apply_control_enable(cgrp);
5576 TRACE_CGROUP_PATH(mkdir, cgrp);
5578 /* let's create and online css's */
5579 kernfs_activate(cgrp->kn);
5585 cgroup_destroy_locked(cgrp);
5587 cgroup_kn_unlock(parent_kn);
5592 * This is called when the refcnt of a css is confirmed to be killed.
5593 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5594 * initiate destruction and put the css ref from kill_css().
5596 static void css_killed_work_fn(struct work_struct *work)
5598 struct cgroup_subsys_state *css =
5599 container_of(work, struct cgroup_subsys_state, destroy_work);
5601 mutex_lock(&cgroup_mutex);
5606 /* @css can't go away while we're holding cgroup_mutex */
5608 } while (css && atomic_dec_and_test(&css->online_cnt));
5610 mutex_unlock(&cgroup_mutex);
5613 /* css kill confirmation processing requires process context, bounce */
5614 static void css_killed_ref_fn(struct percpu_ref *ref)
5616 struct cgroup_subsys_state *css =
5617 container_of(ref, struct cgroup_subsys_state, refcnt);
5619 if (atomic_dec_and_test(&css->online_cnt)) {
5620 INIT_WORK(&css->destroy_work, css_killed_work_fn);
5621 queue_work(cgroup_destroy_wq, &css->destroy_work);
5626 * kill_css - destroy a css
5627 * @css: css to destroy
5629 * This function initiates destruction of @css by removing cgroup interface
5630 * files and putting its base reference. ->css_offline() will be invoked
5631 * asynchronously once css_tryget_online() is guaranteed to fail and when
5632 * the reference count reaches zero, @css will be released.
5634 static void kill_css(struct cgroup_subsys_state *css)
5636 lockdep_assert_held(&cgroup_mutex);
5638 if (css->flags & CSS_DYING)
5641 css->flags |= CSS_DYING;
5644 * This must happen before css is disassociated with its cgroup.
5645 * See seq_css() for details.
5650 * Killing would put the base ref, but we need to keep it alive
5651 * until after ->css_offline().
5656 * cgroup core guarantees that, by the time ->css_offline() is
5657 * invoked, no new css reference will be given out via
5658 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5659 * proceed to offlining css's because percpu_ref_kill() doesn't
5660 * guarantee that the ref is seen as killed on all CPUs on return.
5662 * Use percpu_ref_kill_and_confirm() to get notifications as each
5663 * css is confirmed to be seen as killed on all CPUs.
5665 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5669 * cgroup_destroy_locked - the first stage of cgroup destruction
5670 * @cgrp: cgroup to be destroyed
5672 * css's make use of percpu refcnts whose killing latency shouldn't be
5673 * exposed to userland and are RCU protected. Also, cgroup core needs to
5674 * guarantee that css_tryget_online() won't succeed by the time
5675 * ->css_offline() is invoked. To satisfy all the requirements,
5676 * destruction is implemented in the following two steps.
5678 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5679 * userland visible parts and start killing the percpu refcnts of
5680 * css's. Set up so that the next stage will be kicked off once all
5681 * the percpu refcnts are confirmed to be killed.
5683 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5684 * rest of destruction. Once all cgroup references are gone, the
5685 * cgroup is RCU-freed.
5687 * This function implements s1. After this step, @cgrp is gone as far as
5688 * the userland is concerned and a new cgroup with the same name may be
5689 * created. As cgroup doesn't care about the names internally, this
5690 * doesn't cause any problem.
5692 static int cgroup_destroy_locked(struct cgroup *cgrp)
5693 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5695 struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
5696 struct cgroup_subsys_state *css;
5697 struct cgrp_cset_link *link;
5700 lockdep_assert_held(&cgroup_mutex);
5703 * Only migration can raise populated from zero and we're already
5704 * holding cgroup_mutex.
5706 if (cgroup_is_populated(cgrp))
5710 * Make sure there's no live children. We can't test emptiness of
5711 * ->self.children as dead children linger on it while being
5712 * drained; otherwise, "rmdir parent/child parent" may fail.
5714 if (css_has_online_children(&cgrp->self))
5718 * Mark @cgrp and the associated csets dead. The former prevents
5719 * further task migration and child creation by disabling
5720 * cgroup_lock_live_group(). The latter makes the csets ignored by
5721 * the migration path.
5723 cgrp->self.flags &= ~CSS_ONLINE;
5725 spin_lock_irq(&css_set_lock);
5726 list_for_each_entry(link, &cgrp->cset_links, cset_link)
5727 link->cset->dead = true;
5728 spin_unlock_irq(&css_set_lock);
5730 /* initiate massacre of all css's */
5731 for_each_css(css, ssid, cgrp)
5734 /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5735 css_clear_dir(&cgrp->self);
5736 kernfs_remove(cgrp->kn);
5738 if (parent && cgroup_is_threaded(cgrp))
5739 parent->nr_threaded_children--;
5741 spin_lock_irq(&css_set_lock);
5742 for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5743 tcgrp->nr_descendants--;
5744 tcgrp->nr_dying_descendants++;
5746 * If the dying cgroup is frozen, decrease frozen descendants
5747 * counters of ancestor cgroups.
5749 if (test_bit(CGRP_FROZEN, &cgrp->flags))
5750 tcgrp->freezer.nr_frozen_descendants--;
5752 spin_unlock_irq(&css_set_lock);
5754 cgroup1_check_for_release(parent);
5756 cgroup_bpf_offline(cgrp);
5758 /* put the base reference */
5759 percpu_ref_kill(&cgrp->self.refcnt);
5764 int cgroup_rmdir(struct kernfs_node *kn)
5766 struct cgroup *cgrp;
5769 cgrp = cgroup_kn_lock_live(kn, false);
5773 ret = cgroup_destroy_locked(cgrp);
5775 TRACE_CGROUP_PATH(rmdir, cgrp);
5777 cgroup_kn_unlock(kn);
5781 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5782 .show_options = cgroup_show_options,
5783 .mkdir = cgroup_mkdir,
5784 .rmdir = cgroup_rmdir,
5785 .show_path = cgroup_show_path,
5788 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5790 struct cgroup_subsys_state *css;
5792 pr_debug("Initializing cgroup subsys %s\n", ss->name);
5794 mutex_lock(&cgroup_mutex);
5796 idr_init(&ss->css_idr);
5797 INIT_LIST_HEAD(&ss->cfts);
5799 /* Create the root cgroup state for this subsystem */
5800 ss->root = &cgrp_dfl_root;
5801 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5802 /* We don't handle early failures gracefully */
5803 BUG_ON(IS_ERR(css));
5804 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5807 * Root csses are never destroyed and we can't initialize
5808 * percpu_ref during early init. Disable refcnting.
5810 css->flags |= CSS_NO_REF;
5813 /* allocation can't be done safely during early init */
5816 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5817 BUG_ON(css->id < 0);
5820 /* Update the init_css_set to contain a subsys
5821 * pointer to this state - since the subsystem is
5822 * newly registered, all tasks and hence the
5823 * init_css_set is in the subsystem's root cgroup. */
5824 init_css_set.subsys[ss->id] = css;
5826 have_fork_callback |= (bool)ss->fork << ss->id;
5827 have_exit_callback |= (bool)ss->exit << ss->id;
5828 have_release_callback |= (bool)ss->release << ss->id;
5829 have_canfork_callback |= (bool)ss->can_fork << ss->id;
5831 /* At system boot, before all subsystems have been
5832 * registered, no tasks have been forked, so we don't
5833 * need to invoke fork callbacks here. */
5834 BUG_ON(!list_empty(&init_task.tasks));
5836 BUG_ON(online_css(css));
5838 mutex_unlock(&cgroup_mutex);
5842 * cgroup_init_early - cgroup initialization at system boot
5844 * Initialize cgroups at system boot, and initialize any
5845 * subsystems that request early init.
5847 int __init cgroup_init_early(void)
5849 static struct cgroup_fs_context __initdata ctx;
5850 struct cgroup_subsys *ss;
5853 ctx.root = &cgrp_dfl_root;
5854 init_cgroup_root(&ctx);
5855 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5857 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5859 for_each_subsys(ss, i) {
5860 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5861 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5862 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5864 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5865 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5868 ss->name = cgroup_subsys_name[i];
5869 if (!ss->legacy_name)
5870 ss->legacy_name = cgroup_subsys_name[i];
5873 cgroup_init_subsys(ss, true);
5879 * cgroup_init - cgroup initialization
5881 * Register cgroup filesystem and /proc file, and initialize
5882 * any subsystems that didn't request early init.
5884 int __init cgroup_init(void)
5886 struct cgroup_subsys *ss;
5889 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5890 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
5891 BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
5893 cgroup_rstat_boot();
5896 * The latency of the synchronize_rcu() is too high for cgroups,
5897 * avoid it at the cost of forcing all readers into the slow path.
5899 rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5901 get_user_ns(init_cgroup_ns.user_ns);
5903 mutex_lock(&cgroup_mutex);
5906 * Add init_css_set to the hash table so that dfl_root can link to
5909 hash_add(css_set_table, &init_css_set.hlist,
5910 css_set_hash(init_css_set.subsys));
5912 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
5914 mutex_unlock(&cgroup_mutex);
5916 for_each_subsys(ss, ssid) {
5917 if (ss->early_init) {
5918 struct cgroup_subsys_state *css =
5919 init_css_set.subsys[ss->id];
5921 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5923 BUG_ON(css->id < 0);
5925 cgroup_init_subsys(ss, false);
5928 list_add_tail(&init_css_set.e_cset_node[ssid],
5929 &cgrp_dfl_root.cgrp.e_csets[ssid]);
5932 * Setting dfl_root subsys_mask needs to consider the
5933 * disabled flag and cftype registration needs kmalloc,
5934 * both of which aren't available during early_init.
5936 if (!cgroup_ssid_enabled(ssid))
5939 if (cgroup1_ssid_disabled(ssid))
5940 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5943 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5945 /* implicit controllers must be threaded too */
5946 WARN_ON(ss->implicit_on_dfl && !ss->threaded);
5948 if (ss->implicit_on_dfl)
5949 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5950 else if (!ss->dfl_cftypes)
5951 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5954 cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
5956 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5957 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5959 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5960 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5964 ss->bind(init_css_set.subsys[ssid]);
5966 mutex_lock(&cgroup_mutex);
5967 css_populate_dir(init_css_set.subsys[ssid]);
5968 mutex_unlock(&cgroup_mutex);
5971 /* init_css_set.subsys[] has been updated, re-hash */
5972 hash_del(&init_css_set.hlist);
5973 hash_add(css_set_table, &init_css_set.hlist,
5974 css_set_hash(init_css_set.subsys));
5976 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5977 WARN_ON(register_filesystem(&cgroup_fs_type));
5978 WARN_ON(register_filesystem(&cgroup2_fs_type));
5979 WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show));
5980 #ifdef CONFIG_CPUSETS
5981 WARN_ON(register_filesystem(&cpuset_fs_type));
5987 static int __init cgroup_wq_init(void)
5990 * There isn't much point in executing destruction path in
5991 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5992 * Use 1 for @max_active.
5994 * We would prefer to do this in cgroup_init() above, but that
5995 * is called before init_workqueues(): so leave this until after.
5997 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5998 BUG_ON(!cgroup_destroy_wq);
6001 core_initcall(cgroup_wq_init);
6003 void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen)
6005 struct kernfs_node *kn;
6007 kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id);
6010 kernfs_path(kn, buf, buflen);
6015 * cgroup_get_from_id : get the cgroup associated with cgroup id
6017 * On success return the cgrp, on failure return NULL
6019 struct cgroup *cgroup_get_from_id(u64 id)
6021 struct kernfs_node *kn;
6022 struct cgroup *cgrp = NULL;
6024 kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id);
6028 if (kernfs_type(kn) != KERNFS_DIR)
6033 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6034 if (cgrp && !cgroup_tryget(cgrp))
6043 EXPORT_SYMBOL_GPL(cgroup_get_from_id);
6046 * proc_cgroup_show()
6047 * - Print task's cgroup paths into seq_file, one line for each hierarchy
6048 * - Used for /proc/<pid>/cgroup.
6050 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
6051 struct pid *pid, struct task_struct *tsk)
6055 struct cgroup_root *root;
6058 buf = kmalloc(PATH_MAX, GFP_KERNEL);
6062 mutex_lock(&cgroup_mutex);
6063 spin_lock_irq(&css_set_lock);
6065 for_each_root(root) {
6066 struct cgroup_subsys *ss;
6067 struct cgroup *cgrp;
6068 int ssid, count = 0;
6070 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
6073 seq_printf(m, "%d:", root->hierarchy_id);
6074 if (root != &cgrp_dfl_root)
6075 for_each_subsys(ss, ssid)
6076 if (root->subsys_mask & (1 << ssid))
6077 seq_printf(m, "%s%s", count++ ? "," : "",
6079 if (strlen(root->name))
6080 seq_printf(m, "%sname=%s", count ? "," : "",
6084 cgrp = task_cgroup_from_root(tsk, root);
6087 * On traditional hierarchies, all zombie tasks show up as
6088 * belonging to the root cgroup. On the default hierarchy,
6089 * while a zombie doesn't show up in "cgroup.procs" and
6090 * thus can't be migrated, its /proc/PID/cgroup keeps
6091 * reporting the cgroup it belonged to before exiting. If
6092 * the cgroup is removed before the zombie is reaped,
6093 * " (deleted)" is appended to the cgroup path.
6095 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
6096 retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
6097 current->nsproxy->cgroup_ns);
6098 if (retval >= PATH_MAX)
6099 retval = -ENAMETOOLONG;
6108 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
6109 seq_puts(m, " (deleted)\n");
6116 spin_unlock_irq(&css_set_lock);
6117 mutex_unlock(&cgroup_mutex);
6124 * cgroup_fork - initialize cgroup related fields during copy_process()
6125 * @child: pointer to task_struct of forking parent process.
6127 * A task is associated with the init_css_set until cgroup_post_fork()
6128 * attaches it to the target css_set.
6130 void cgroup_fork(struct task_struct *child)
6132 RCU_INIT_POINTER(child->cgroups, &init_css_set);
6133 INIT_LIST_HEAD(&child->cg_list);
6136 static struct cgroup *cgroup_get_from_file(struct file *f)
6138 struct cgroup_subsys_state *css;
6139 struct cgroup *cgrp;
6141 css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
6143 return ERR_CAST(css);
6146 if (!cgroup_on_dfl(cgrp)) {
6148 return ERR_PTR(-EBADF);
6155 * cgroup_css_set_fork - find or create a css_set for a child process
6156 * @kargs: the arguments passed to create the child process
6158 * This functions finds or creates a new css_set which the child
6159 * process will be attached to in cgroup_post_fork(). By default,
6160 * the child process will be given the same css_set as its parent.
6162 * If CLONE_INTO_CGROUP is specified this function will try to find an
6163 * existing css_set which includes the requested cgroup and if not create
6164 * a new css_set that the child will be attached to later. If this function
6165 * succeeds it will hold cgroup_threadgroup_rwsem on return. If
6166 * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
6167 * before grabbing cgroup_threadgroup_rwsem and will hold a reference
6168 * to the target cgroup.
6170 static int cgroup_css_set_fork(struct kernel_clone_args *kargs)
6171 __acquires(&cgroup_mutex) __acquires(&cgroup_threadgroup_rwsem)
6174 struct cgroup *dst_cgrp = NULL;
6175 struct css_set *cset;
6176 struct super_block *sb;
6179 if (kargs->flags & CLONE_INTO_CGROUP)
6180 mutex_lock(&cgroup_mutex);
6182 cgroup_threadgroup_change_begin(current);
6184 spin_lock_irq(&css_set_lock);
6185 cset = task_css_set(current);
6187 spin_unlock_irq(&css_set_lock);
6189 if (!(kargs->flags & CLONE_INTO_CGROUP)) {
6194 f = fget_raw(kargs->cgroup);
6199 sb = f->f_path.dentry->d_sb;
6201 dst_cgrp = cgroup_get_from_file(f);
6202 if (IS_ERR(dst_cgrp)) {
6203 ret = PTR_ERR(dst_cgrp);
6208 if (cgroup_is_dead(dst_cgrp)) {
6214 * Verify that we the target cgroup is writable for us. This is
6215 * usually done by the vfs layer but since we're not going through
6216 * the vfs layer here we need to do it "manually".
6218 ret = cgroup_may_write(dst_cgrp, sb);
6222 ret = cgroup_attach_permissions(cset->dfl_cgrp, dst_cgrp, sb,
6223 !(kargs->flags & CLONE_THREAD),
6224 current->nsproxy->cgroup_ns);
6228 kargs->cset = find_css_set(cset, dst_cgrp);
6236 kargs->cgrp = dst_cgrp;
6240 cgroup_threadgroup_change_end(current);
6241 mutex_unlock(&cgroup_mutex);
6245 cgroup_put(dst_cgrp);
6248 put_css_set(kargs->cset);
6253 * cgroup_css_set_put_fork - drop references we took during fork
6254 * @kargs: the arguments passed to create the child process
6256 * Drop references to the prepared css_set and target cgroup if
6257 * CLONE_INTO_CGROUP was requested.
6259 static void cgroup_css_set_put_fork(struct kernel_clone_args *kargs)
6260 __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6262 cgroup_threadgroup_change_end(current);
6264 if (kargs->flags & CLONE_INTO_CGROUP) {
6265 struct cgroup *cgrp = kargs->cgrp;
6266 struct css_set *cset = kargs->cset;
6268 mutex_unlock(&cgroup_mutex);
6283 * cgroup_can_fork - called on a new task before the process is exposed
6284 * @child: the child process
6286 * This prepares a new css_set for the child process which the child will
6287 * be attached to in cgroup_post_fork().
6288 * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
6289 * callback returns an error, the fork aborts with that error code. This
6290 * allows for a cgroup subsystem to conditionally allow or deny new forks.
6292 int cgroup_can_fork(struct task_struct *child, struct kernel_clone_args *kargs)
6294 struct cgroup_subsys *ss;
6297 ret = cgroup_css_set_fork(kargs);
6301 do_each_subsys_mask(ss, i, have_canfork_callback) {
6302 ret = ss->can_fork(child, kargs->cset);
6305 } while_each_subsys_mask();
6310 for_each_subsys(ss, j) {
6313 if (ss->cancel_fork)
6314 ss->cancel_fork(child, kargs->cset);
6317 cgroup_css_set_put_fork(kargs);
6323 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6324 * @child: the child process
6325 * @kargs: the arguments passed to create the child process
6327 * This calls the cancel_fork() callbacks if a fork failed *after*
6328 * cgroup_can_fork() succeeded and cleans up references we took to
6329 * prepare a new css_set for the child process in cgroup_can_fork().
6331 void cgroup_cancel_fork(struct task_struct *child,
6332 struct kernel_clone_args *kargs)
6334 struct cgroup_subsys *ss;
6337 for_each_subsys(ss, i)
6338 if (ss->cancel_fork)
6339 ss->cancel_fork(child, kargs->cset);
6341 cgroup_css_set_put_fork(kargs);
6345 * cgroup_post_fork - finalize cgroup setup for the child process
6346 * @child: the child process
6348 * Attach the child process to its css_set calling the subsystem fork()
6351 void cgroup_post_fork(struct task_struct *child,
6352 struct kernel_clone_args *kargs)
6353 __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6355 unsigned long cgrp_flags = 0;
6357 struct cgroup_subsys *ss;
6358 struct css_set *cset;
6364 spin_lock_irq(&css_set_lock);
6366 /* init tasks are special, only link regular threads */
6367 if (likely(child->pid)) {
6369 cgrp_flags = kargs->cgrp->flags;
6371 cgrp_flags = cset->dfl_cgrp->flags;
6373 WARN_ON_ONCE(!list_empty(&child->cg_list));
6375 css_set_move_task(child, NULL, cset, false);
6381 if (!(child->flags & PF_KTHREAD)) {
6382 if (unlikely(test_bit(CGRP_FREEZE, &cgrp_flags))) {
6384 * If the cgroup has to be frozen, the new task has
6385 * too. Let's set the JOBCTL_TRAP_FREEZE jobctl bit to
6386 * get the task into the frozen state.
6388 spin_lock(&child->sighand->siglock);
6389 WARN_ON_ONCE(child->frozen);
6390 child->jobctl |= JOBCTL_TRAP_FREEZE;
6391 spin_unlock(&child->sighand->siglock);
6394 * Calling cgroup_update_frozen() isn't required here,
6395 * because it will be called anyway a bit later from
6396 * do_freezer_trap(). So we avoid cgroup's transient
6397 * switch from the frozen state and back.
6402 * If the cgroup is to be killed notice it now and take the
6403 * child down right after we finished preparing it for
6406 kill = test_bit(CGRP_KILL, &cgrp_flags);
6409 spin_unlock_irq(&css_set_lock);
6412 * Call ss->fork(). This must happen after @child is linked on
6413 * css_set; otherwise, @child might change state between ->fork()
6414 * and addition to css_set.
6416 do_each_subsys_mask(ss, i, have_fork_callback) {
6418 } while_each_subsys_mask();
6420 /* Make the new cset the root_cset of the new cgroup namespace. */
6421 if (kargs->flags & CLONE_NEWCGROUP) {
6422 struct css_set *rcset = child->nsproxy->cgroup_ns->root_cset;
6425 child->nsproxy->cgroup_ns->root_cset = cset;
6429 /* Cgroup has to be killed so take down child immediately. */
6431 do_send_sig_info(SIGKILL, SEND_SIG_NOINFO, child, PIDTYPE_TGID);
6433 cgroup_css_set_put_fork(kargs);
6437 * cgroup_exit - detach cgroup from exiting task
6438 * @tsk: pointer to task_struct of exiting process
6440 * Description: Detach cgroup from @tsk.
6443 void cgroup_exit(struct task_struct *tsk)
6445 struct cgroup_subsys *ss;
6446 struct css_set *cset;
6449 spin_lock_irq(&css_set_lock);
6451 WARN_ON_ONCE(list_empty(&tsk->cg_list));
6452 cset = task_css_set(tsk);
6453 css_set_move_task(tsk, cset, NULL, false);
6454 list_add_tail(&tsk->cg_list, &cset->dying_tasks);
6457 WARN_ON_ONCE(cgroup_task_frozen(tsk));
6458 if (unlikely(!(tsk->flags & PF_KTHREAD) &&
6459 test_bit(CGRP_FREEZE, &task_dfl_cgroup(tsk)->flags)))
6460 cgroup_update_frozen(task_dfl_cgroup(tsk));
6462 spin_unlock_irq(&css_set_lock);
6464 /* see cgroup_post_fork() for details */
6465 do_each_subsys_mask(ss, i, have_exit_callback) {
6467 } while_each_subsys_mask();
6470 void cgroup_release(struct task_struct *task)
6472 struct cgroup_subsys *ss;
6475 do_each_subsys_mask(ss, ssid, have_release_callback) {
6477 } while_each_subsys_mask();
6479 spin_lock_irq(&css_set_lock);
6480 css_set_skip_task_iters(task_css_set(task), task);
6481 list_del_init(&task->cg_list);
6482 spin_unlock_irq(&css_set_lock);
6485 void cgroup_free(struct task_struct *task)
6487 struct css_set *cset = task_css_set(task);
6491 static int __init cgroup_disable(char *str)
6493 struct cgroup_subsys *ss;
6497 while ((token = strsep(&str, ",")) != NULL) {
6501 for_each_subsys(ss, i) {
6502 if (strcmp(token, ss->name) &&
6503 strcmp(token, ss->legacy_name))
6506 static_branch_disable(cgroup_subsys_enabled_key[i]);
6507 pr_info("Disabling %s control group subsystem\n",
6511 for (i = 0; i < OPT_FEATURE_COUNT; i++) {
6512 if (strcmp(token, cgroup_opt_feature_names[i]))
6514 cgroup_feature_disable_mask |= 1 << i;
6515 pr_info("Disabling %s control group feature\n",
6516 cgroup_opt_feature_names[i]);
6522 __setup("cgroup_disable=", cgroup_disable);
6524 void __init __weak enable_debug_cgroup(void) { }
6526 static int __init enable_cgroup_debug(char *str)
6528 cgroup_debug = true;
6529 enable_debug_cgroup();
6532 __setup("cgroup_debug", enable_cgroup_debug);
6535 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6536 * @dentry: directory dentry of interest
6537 * @ss: subsystem of interest
6539 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6540 * to get the corresponding css and return it. If such css doesn't exist
6541 * or can't be pinned, an ERR_PTR value is returned.
6543 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
6544 struct cgroup_subsys *ss)
6546 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
6547 struct file_system_type *s_type = dentry->d_sb->s_type;
6548 struct cgroup_subsys_state *css = NULL;
6549 struct cgroup *cgrp;
6551 /* is @dentry a cgroup dir? */
6552 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
6553 !kn || kernfs_type(kn) != KERNFS_DIR)
6554 return ERR_PTR(-EBADF);
6559 * This path doesn't originate from kernfs and @kn could already
6560 * have been or be removed at any point. @kn->priv is RCU
6561 * protected for this access. See css_release_work_fn() for details.
6563 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6565 css = cgroup_css(cgrp, ss);
6567 if (!css || !css_tryget_online(css))
6568 css = ERR_PTR(-ENOENT);
6575 * css_from_id - lookup css by id
6576 * @id: the cgroup id
6577 * @ss: cgroup subsys to be looked into
6579 * Returns the css if there's valid one with @id, otherwise returns NULL.
6580 * Should be called under rcu_read_lock().
6582 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
6584 WARN_ON_ONCE(!rcu_read_lock_held());
6585 return idr_find(&ss->css_idr, id);
6589 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6590 * @path: path on the default hierarchy
6592 * Find the cgroup at @path on the default hierarchy, increment its
6593 * reference count and return it. Returns pointer to the found cgroup on
6594 * success, ERR_PTR(-ENOENT) if @path doesn't exist or if the cgroup has already
6595 * been released and ERR_PTR(-ENOTDIR) if @path points to a non-directory.
6597 struct cgroup *cgroup_get_from_path(const char *path)
6599 struct kernfs_node *kn;
6600 struct cgroup *cgrp = ERR_PTR(-ENOENT);
6602 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6606 if (kernfs_type(kn) != KERNFS_DIR) {
6607 cgrp = ERR_PTR(-ENOTDIR);
6613 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6614 if (!cgrp || !cgroup_tryget(cgrp))
6615 cgrp = ERR_PTR(-ENOENT);
6624 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6627 * cgroup_get_from_fd - get a cgroup pointer from a fd
6628 * @fd: fd obtained by open(cgroup2_dir)
6630 * Find the cgroup from a fd which should be obtained
6631 * by opening a cgroup directory. Returns a pointer to the
6632 * cgroup on success. ERR_PTR is returned if the cgroup
6635 struct cgroup *cgroup_get_from_fd(int fd)
6637 struct cgroup *cgrp;
6642 return ERR_PTR(-EBADF);
6644 cgrp = cgroup_get_from_file(f);
6648 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
6650 static u64 power_of_ten(int power)
6659 * cgroup_parse_float - parse a floating number
6660 * @input: input string
6661 * @dec_shift: number of decimal digits to shift
6664 * Parse a decimal floating point number in @input and store the result in
6665 * @v with decimal point right shifted @dec_shift times. For example, if
6666 * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6667 * Returns 0 on success, -errno otherwise.
6669 * There's nothing cgroup specific about this function except that it's
6670 * currently the only user.
6672 int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v)
6674 s64 whole, frac = 0;
6675 int fstart = 0, fend = 0, flen;
6677 if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend))
6682 flen = fend > fstart ? fend - fstart : 0;
6683 if (flen < dec_shift)
6684 frac *= power_of_ten(dec_shift - flen);
6686 frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift));
6688 *v = whole * power_of_ten(dec_shift) + frac;
6693 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6694 * definition in cgroup-defs.h.
6696 #ifdef CONFIG_SOCK_CGROUP_DATA
6698 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6700 struct cgroup *cgroup;
6703 /* Don't associate the sock with unrelated interrupted task's cgroup. */
6704 if (in_interrupt()) {
6705 cgroup = &cgrp_dfl_root.cgrp;
6711 struct css_set *cset;
6713 cset = task_css_set(current);
6714 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6715 cgroup = cset->dfl_cgrp;
6721 skcd->cgroup = cgroup;
6722 cgroup_bpf_get(cgroup);
6726 void cgroup_sk_clone(struct sock_cgroup_data *skcd)
6728 struct cgroup *cgrp = sock_cgroup_ptr(skcd);
6731 * We might be cloning a socket which is left in an empty
6732 * cgroup and the cgroup might have already been rmdir'd.
6733 * Don't use cgroup_get_live().
6736 cgroup_bpf_get(cgrp);
6739 void cgroup_sk_free(struct sock_cgroup_data *skcd)
6741 struct cgroup *cgrp = sock_cgroup_ptr(skcd);
6743 cgroup_bpf_put(cgrp);
6747 #endif /* CONFIG_SOCK_CGROUP_DATA */
6749 #ifdef CONFIG_CGROUP_BPF
6750 int cgroup_bpf_attach(struct cgroup *cgrp,
6751 struct bpf_prog *prog, struct bpf_prog *replace_prog,
6752 struct bpf_cgroup_link *link,
6753 enum bpf_attach_type type,
6758 mutex_lock(&cgroup_mutex);
6759 ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags);
6760 mutex_unlock(&cgroup_mutex);
6764 int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
6765 enum bpf_attach_type type)
6769 mutex_lock(&cgroup_mutex);
6770 ret = __cgroup_bpf_detach(cgrp, prog, NULL, type);
6771 mutex_unlock(&cgroup_mutex);
6775 int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
6776 union bpf_attr __user *uattr)
6780 mutex_lock(&cgroup_mutex);
6781 ret = __cgroup_bpf_query(cgrp, attr, uattr);
6782 mutex_unlock(&cgroup_mutex);
6785 #endif /* CONFIG_CGROUP_BPF */
6788 static ssize_t show_delegatable_files(struct cftype *files, char *buf,
6789 ssize_t size, const char *prefix)
6794 for (cft = files; cft && cft->name[0] != '\0'; cft++) {
6795 if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
6798 if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
6802 ret += snprintf(buf + ret, size - ret, "%s.", prefix);
6804 ret += snprintf(buf + ret, size - ret, "%s\n", cft->name);
6806 if (WARN_ON(ret >= size))
6813 static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr,
6816 struct cgroup_subsys *ss;
6820 ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret,
6823 for_each_subsys(ss, ssid)
6824 ret += show_delegatable_files(ss->dfl_cftypes, buf + ret,
6826 cgroup_subsys_name[ssid]);
6830 static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate);
6832 static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr,
6835 return snprintf(buf, PAGE_SIZE,
6837 "memory_localevents\n"
6838 "memory_recursiveprot\n");
6840 static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features);
6842 static struct attribute *cgroup_sysfs_attrs[] = {
6843 &cgroup_delegate_attr.attr,
6844 &cgroup_features_attr.attr,
6848 static const struct attribute_group cgroup_sysfs_attr_group = {
6849 .attrs = cgroup_sysfs_attrs,
6853 static int __init cgroup_sysfs_init(void)
6855 return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group);
6857 subsys_initcall(cgroup_sysfs_init);
6859 #endif /* CONFIG_SYSFS */