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_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
768 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
771 * The following field is re-initialized when this cset gets linked
772 * in cgroup_init(). However, let's initialize the field
773 * statically too so that the default cgroup can be accessed safely
776 .dfl_cgrp = &cgrp_dfl_root.cgrp,
779 static int css_set_count = 1; /* 1 for init_css_set */
781 static bool css_set_threaded(struct css_set *cset)
783 return cset->dom_cset != cset;
787 * css_set_populated - does a css_set contain any tasks?
788 * @cset: target css_set
790 * css_set_populated() should be the same as !!cset->nr_tasks at steady
791 * state. However, css_set_populated() can be called while a task is being
792 * added to or removed from the linked list before the nr_tasks is
793 * properly updated. Hence, we can't just look at ->nr_tasks here.
795 static bool css_set_populated(struct css_set *cset)
797 lockdep_assert_held(&css_set_lock);
799 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
803 * cgroup_update_populated - update the populated count of a cgroup
804 * @cgrp: the target cgroup
805 * @populated: inc or dec populated count
807 * One of the css_sets associated with @cgrp is either getting its first
808 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
809 * count is propagated towards root so that a given cgroup's
810 * nr_populated_children is zero iff none of its descendants contain any
813 * @cgrp's interface file "cgroup.populated" is zero if both
814 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
815 * 1 otherwise. When the sum changes from or to zero, userland is notified
816 * that the content of the interface file has changed. This can be used to
817 * detect when @cgrp and its descendants become populated or empty.
819 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
821 struct cgroup *child = NULL;
822 int adj = populated ? 1 : -1;
824 lockdep_assert_held(&css_set_lock);
827 bool was_populated = cgroup_is_populated(cgrp);
830 cgrp->nr_populated_csets += adj;
832 if (cgroup_is_threaded(child))
833 cgrp->nr_populated_threaded_children += adj;
835 cgrp->nr_populated_domain_children += adj;
838 if (was_populated == cgroup_is_populated(cgrp))
841 cgroup1_check_for_release(cgrp);
842 TRACE_CGROUP_PATH(notify_populated, cgrp,
843 cgroup_is_populated(cgrp));
844 cgroup_file_notify(&cgrp->events_file);
847 cgrp = cgroup_parent(cgrp);
852 * css_set_update_populated - update populated state of a css_set
853 * @cset: target css_set
854 * @populated: whether @cset is populated or depopulated
856 * @cset is either getting the first task or losing the last. Update the
857 * populated counters of all associated cgroups accordingly.
859 static void css_set_update_populated(struct css_set *cset, bool populated)
861 struct cgrp_cset_link *link;
863 lockdep_assert_held(&css_set_lock);
865 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
866 cgroup_update_populated(link->cgrp, populated);
870 * @task is leaving, advance task iterators which are pointing to it so
871 * that they can resume at the next position. Advancing an iterator might
872 * remove it from the list, use safe walk. See css_task_iter_skip() for
875 static void css_set_skip_task_iters(struct css_set *cset,
876 struct task_struct *task)
878 struct css_task_iter *it, *pos;
880 list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node)
881 css_task_iter_skip(it, task);
885 * css_set_move_task - move a task from one css_set to another
886 * @task: task being moved
887 * @from_cset: css_set @task currently belongs to (may be NULL)
888 * @to_cset: new css_set @task is being moved to (may be NULL)
889 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
891 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
892 * css_set, @from_cset can be NULL. If @task is being disassociated
893 * instead of moved, @to_cset can be NULL.
895 * This function automatically handles populated counter updates and
896 * css_task_iter adjustments but the caller is responsible for managing
897 * @from_cset and @to_cset's reference counts.
899 static void css_set_move_task(struct task_struct *task,
900 struct css_set *from_cset, struct css_set *to_cset,
903 lockdep_assert_held(&css_set_lock);
905 if (to_cset && !css_set_populated(to_cset))
906 css_set_update_populated(to_cset, true);
909 WARN_ON_ONCE(list_empty(&task->cg_list));
911 css_set_skip_task_iters(from_cset, task);
912 list_del_init(&task->cg_list);
913 if (!css_set_populated(from_cset))
914 css_set_update_populated(from_cset, false);
916 WARN_ON_ONCE(!list_empty(&task->cg_list));
921 * We are synchronized through cgroup_threadgroup_rwsem
922 * against PF_EXITING setting such that we can't race
923 * against cgroup_exit()/cgroup_free() dropping the css_set.
925 WARN_ON_ONCE(task->flags & PF_EXITING);
927 cgroup_move_task(task, to_cset);
928 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
934 * hash table for cgroup groups. This improves the performance to find
935 * an existing css_set. This hash doesn't (currently) take into
936 * account cgroups in empty hierarchies.
938 #define CSS_SET_HASH_BITS 7
939 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
941 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
943 unsigned long key = 0UL;
944 struct cgroup_subsys *ss;
947 for_each_subsys(ss, i)
948 key += (unsigned long)css[i];
949 key = (key >> 16) ^ key;
954 void put_css_set_locked(struct css_set *cset)
956 struct cgrp_cset_link *link, *tmp_link;
957 struct cgroup_subsys *ss;
960 lockdep_assert_held(&css_set_lock);
962 if (!refcount_dec_and_test(&cset->refcount))
965 WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
967 /* This css_set is dead. Unlink it and release cgroup and css refs */
968 for_each_subsys(ss, ssid) {
969 list_del(&cset->e_cset_node[ssid]);
970 css_put(cset->subsys[ssid]);
972 hash_del(&cset->hlist);
975 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
976 list_del(&link->cset_link);
977 list_del(&link->cgrp_link);
978 if (cgroup_parent(link->cgrp))
979 cgroup_put(link->cgrp);
983 if (css_set_threaded(cset)) {
984 list_del(&cset->threaded_csets_node);
985 put_css_set_locked(cset->dom_cset);
988 kfree_rcu(cset, rcu_head);
992 * compare_css_sets - helper function for find_existing_css_set().
993 * @cset: candidate css_set being tested
994 * @old_cset: existing css_set for a task
995 * @new_cgrp: cgroup that's being entered by the task
996 * @template: desired set of css pointers in css_set (pre-calculated)
998 * Returns true if "cset" matches "old_cset" except for the hierarchy
999 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
1001 static bool compare_css_sets(struct css_set *cset,
1002 struct css_set *old_cset,
1003 struct cgroup *new_cgrp,
1004 struct cgroup_subsys_state *template[])
1006 struct cgroup *new_dfl_cgrp;
1007 struct list_head *l1, *l2;
1010 * On the default hierarchy, there can be csets which are
1011 * associated with the same set of cgroups but different csses.
1012 * Let's first ensure that csses match.
1014 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
1018 /* @cset's domain should match the default cgroup's */
1019 if (cgroup_on_dfl(new_cgrp))
1020 new_dfl_cgrp = new_cgrp;
1022 new_dfl_cgrp = old_cset->dfl_cgrp;
1024 if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
1028 * Compare cgroup pointers in order to distinguish between
1029 * different cgroups in hierarchies. As different cgroups may
1030 * share the same effective css, this comparison is always
1033 l1 = &cset->cgrp_links;
1034 l2 = &old_cset->cgrp_links;
1036 struct cgrp_cset_link *link1, *link2;
1037 struct cgroup *cgrp1, *cgrp2;
1041 /* See if we reached the end - both lists are equal length. */
1042 if (l1 == &cset->cgrp_links) {
1043 BUG_ON(l2 != &old_cset->cgrp_links);
1046 BUG_ON(l2 == &old_cset->cgrp_links);
1048 /* Locate the cgroups associated with these links. */
1049 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
1050 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
1051 cgrp1 = link1->cgrp;
1052 cgrp2 = link2->cgrp;
1053 /* Hierarchies should be linked in the same order. */
1054 BUG_ON(cgrp1->root != cgrp2->root);
1057 * If this hierarchy is the hierarchy of the cgroup
1058 * that's changing, then we need to check that this
1059 * css_set points to the new cgroup; if it's any other
1060 * hierarchy, then this css_set should point to the
1061 * same cgroup as the old css_set.
1063 if (cgrp1->root == new_cgrp->root) {
1064 if (cgrp1 != new_cgrp)
1075 * find_existing_css_set - init css array and find the matching css_set
1076 * @old_cset: the css_set that we're using before the cgroup transition
1077 * @cgrp: the cgroup that we're moving into
1078 * @template: out param for the new set of csses, should be clear on entry
1080 static struct css_set *find_existing_css_set(struct css_set *old_cset,
1081 struct cgroup *cgrp,
1082 struct cgroup_subsys_state *template[])
1084 struct cgroup_root *root = cgrp->root;
1085 struct cgroup_subsys *ss;
1086 struct css_set *cset;
1091 * Build the set of subsystem state objects that we want to see in the
1092 * new css_set. While subsystems can change globally, the entries here
1093 * won't change, so no need for locking.
1095 for_each_subsys(ss, i) {
1096 if (root->subsys_mask & (1UL << i)) {
1098 * @ss is in this hierarchy, so we want the
1099 * effective css from @cgrp.
1101 template[i] = cgroup_e_css_by_mask(cgrp, ss);
1104 * @ss is not in this hierarchy, so we don't want
1105 * to change the css.
1107 template[i] = old_cset->subsys[i];
1111 key = css_set_hash(template);
1112 hash_for_each_possible(css_set_table, cset, hlist, key) {
1113 if (!compare_css_sets(cset, old_cset, cgrp, template))
1116 /* This css_set matches what we need */
1120 /* No existing cgroup group matched */
1124 static void free_cgrp_cset_links(struct list_head *links_to_free)
1126 struct cgrp_cset_link *link, *tmp_link;
1128 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
1129 list_del(&link->cset_link);
1135 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1136 * @count: the number of links to allocate
1137 * @tmp_links: list_head the allocated links are put on
1139 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1140 * through ->cset_link. Returns 0 on success or -errno.
1142 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1144 struct cgrp_cset_link *link;
1147 INIT_LIST_HEAD(tmp_links);
1149 for (i = 0; i < count; i++) {
1150 link = kzalloc(sizeof(*link), GFP_KERNEL);
1152 free_cgrp_cset_links(tmp_links);
1155 list_add(&link->cset_link, tmp_links);
1161 * link_css_set - a helper function to link a css_set to a cgroup
1162 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1163 * @cset: the css_set to be linked
1164 * @cgrp: the destination cgroup
1166 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1167 struct cgroup *cgrp)
1169 struct cgrp_cset_link *link;
1171 BUG_ON(list_empty(tmp_links));
1173 if (cgroup_on_dfl(cgrp))
1174 cset->dfl_cgrp = cgrp;
1176 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1181 * Always add links to the tail of the lists so that the lists are
1182 * in chronological order.
1184 list_move_tail(&link->cset_link, &cgrp->cset_links);
1185 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1187 if (cgroup_parent(cgrp))
1188 cgroup_get_live(cgrp);
1192 * find_css_set - return a new css_set with one cgroup updated
1193 * @old_cset: the baseline css_set
1194 * @cgrp: the cgroup to be updated
1196 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1197 * substituted into the appropriate hierarchy.
1199 static struct css_set *find_css_set(struct css_set *old_cset,
1200 struct cgroup *cgrp)
1202 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1203 struct css_set *cset;
1204 struct list_head tmp_links;
1205 struct cgrp_cset_link *link;
1206 struct cgroup_subsys *ss;
1210 lockdep_assert_held(&cgroup_mutex);
1212 /* First see if we already have a cgroup group that matches
1213 * the desired set */
1214 spin_lock_irq(&css_set_lock);
1215 cset = find_existing_css_set(old_cset, cgrp, template);
1218 spin_unlock_irq(&css_set_lock);
1223 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1227 /* Allocate all the cgrp_cset_link objects that we'll need */
1228 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1233 refcount_set(&cset->refcount, 1);
1234 cset->dom_cset = cset;
1235 INIT_LIST_HEAD(&cset->tasks);
1236 INIT_LIST_HEAD(&cset->mg_tasks);
1237 INIT_LIST_HEAD(&cset->dying_tasks);
1238 INIT_LIST_HEAD(&cset->task_iters);
1239 INIT_LIST_HEAD(&cset->threaded_csets);
1240 INIT_HLIST_NODE(&cset->hlist);
1241 INIT_LIST_HEAD(&cset->cgrp_links);
1242 INIT_LIST_HEAD(&cset->mg_preload_node);
1243 INIT_LIST_HEAD(&cset->mg_node);
1245 /* Copy the set of subsystem state objects generated in
1246 * find_existing_css_set() */
1247 memcpy(cset->subsys, template, sizeof(cset->subsys));
1249 spin_lock_irq(&css_set_lock);
1250 /* Add reference counts and links from the new css_set. */
1251 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1252 struct cgroup *c = link->cgrp;
1254 if (c->root == cgrp->root)
1256 link_css_set(&tmp_links, cset, c);
1259 BUG_ON(!list_empty(&tmp_links));
1263 /* Add @cset to the hash table */
1264 key = css_set_hash(cset->subsys);
1265 hash_add(css_set_table, &cset->hlist, key);
1267 for_each_subsys(ss, ssid) {
1268 struct cgroup_subsys_state *css = cset->subsys[ssid];
1270 list_add_tail(&cset->e_cset_node[ssid],
1271 &css->cgroup->e_csets[ssid]);
1275 spin_unlock_irq(&css_set_lock);
1278 * If @cset should be threaded, look up the matching dom_cset and
1279 * link them up. We first fully initialize @cset then look for the
1280 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1281 * to stay empty until we return.
1283 if (cgroup_is_threaded(cset->dfl_cgrp)) {
1284 struct css_set *dcset;
1286 dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
1292 spin_lock_irq(&css_set_lock);
1293 cset->dom_cset = dcset;
1294 list_add_tail(&cset->threaded_csets_node,
1295 &dcset->threaded_csets);
1296 spin_unlock_irq(&css_set_lock);
1302 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1304 struct cgroup *root_cgrp = kf_root->kn->priv;
1306 return root_cgrp->root;
1309 static int cgroup_init_root_id(struct cgroup_root *root)
1313 lockdep_assert_held(&cgroup_mutex);
1315 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1319 root->hierarchy_id = id;
1323 static void cgroup_exit_root_id(struct cgroup_root *root)
1325 lockdep_assert_held(&cgroup_mutex);
1327 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1330 void cgroup_free_root(struct cgroup_root *root)
1335 static void cgroup_destroy_root(struct cgroup_root *root)
1337 struct cgroup *cgrp = &root->cgrp;
1338 struct cgrp_cset_link *link, *tmp_link;
1340 trace_cgroup_destroy_root(root);
1342 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1344 BUG_ON(atomic_read(&root->nr_cgrps));
1345 BUG_ON(!list_empty(&cgrp->self.children));
1347 /* Rebind all subsystems back to the default hierarchy */
1348 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1351 * Release all the links from cset_links to this hierarchy's
1354 spin_lock_irq(&css_set_lock);
1356 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1357 list_del(&link->cset_link);
1358 list_del(&link->cgrp_link);
1362 spin_unlock_irq(&css_set_lock);
1364 if (!list_empty(&root->root_list)) {
1365 list_del(&root->root_list);
1366 cgroup_root_count--;
1369 cgroup_exit_root_id(root);
1371 mutex_unlock(&cgroup_mutex);
1373 cgroup_rstat_exit(cgrp);
1374 kernfs_destroy_root(root->kf_root);
1375 cgroup_free_root(root);
1379 * look up cgroup associated with current task's cgroup namespace on the
1380 * specified hierarchy
1382 static struct cgroup *
1383 current_cgns_cgroup_from_root(struct cgroup_root *root)
1385 struct cgroup *res = NULL;
1386 struct css_set *cset;
1388 lockdep_assert_held(&css_set_lock);
1392 cset = current->nsproxy->cgroup_ns->root_cset;
1393 if (cset == &init_css_set) {
1395 } else if (root == &cgrp_dfl_root) {
1396 res = cset->dfl_cgrp;
1398 struct cgrp_cset_link *link;
1400 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1401 struct cgroup *c = link->cgrp;
1403 if (c->root == root) {
1415 /* look up cgroup associated with given css_set on the specified hierarchy */
1416 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1417 struct cgroup_root *root)
1419 struct cgroup *res = NULL;
1421 lockdep_assert_held(&cgroup_mutex);
1422 lockdep_assert_held(&css_set_lock);
1424 if (cset == &init_css_set) {
1426 } else if (root == &cgrp_dfl_root) {
1427 res = cset->dfl_cgrp;
1429 struct cgrp_cset_link *link;
1431 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1432 struct cgroup *c = link->cgrp;
1434 if (c->root == root) {
1446 * Return the cgroup for "task" from the given hierarchy. Must be
1447 * called with cgroup_mutex and css_set_lock held.
1449 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1450 struct cgroup_root *root)
1453 * No need to lock the task - since we hold css_set_lock the
1454 * task can't change groups.
1456 return cset_cgroup_from_root(task_css_set(task), root);
1460 * A task must hold cgroup_mutex to modify cgroups.
1462 * Any task can increment and decrement the count field without lock.
1463 * So in general, code holding cgroup_mutex can't rely on the count
1464 * field not changing. However, if the count goes to zero, then only
1465 * cgroup_attach_task() can increment it again. Because a count of zero
1466 * means that no tasks are currently attached, therefore there is no
1467 * way a task attached to that cgroup can fork (the other way to
1468 * increment the count). So code holding cgroup_mutex can safely
1469 * assume that if the count is zero, it will stay zero. Similarly, if
1470 * a task holds cgroup_mutex on a cgroup with zero count, it
1471 * knows that the cgroup won't be removed, as cgroup_rmdir()
1474 * A cgroup can only be deleted if both its 'count' of using tasks
1475 * is zero, and its list of 'children' cgroups is empty. Since all
1476 * tasks in the system use _some_ cgroup, and since there is always at
1477 * least one task in the system (init, pid == 1), therefore, root cgroup
1478 * always has either children cgroups and/or using tasks. So we don't
1479 * need a special hack to ensure that root cgroup cannot be deleted.
1481 * P.S. One more locking exception. RCU is used to guard the
1482 * update of a tasks cgroup pointer by cgroup_attach_task()
1485 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1487 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1490 struct cgroup_subsys *ss = cft->ss;
1492 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1493 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
1494 const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : "";
1496 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s",
1497 dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1500 strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1506 * cgroup_file_mode - deduce file mode of a control file
1507 * @cft: the control file in question
1509 * S_IRUGO for read, S_IWUSR for write.
1511 static umode_t cgroup_file_mode(const struct cftype *cft)
1515 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1518 if (cft->write_u64 || cft->write_s64 || cft->write) {
1519 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1529 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1530 * @subtree_control: the new subtree_control mask to consider
1531 * @this_ss_mask: available subsystems
1533 * On the default hierarchy, a subsystem may request other subsystems to be
1534 * enabled together through its ->depends_on mask. In such cases, more
1535 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1537 * This function calculates which subsystems need to be enabled if
1538 * @subtree_control is to be applied while restricted to @this_ss_mask.
1540 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1542 u16 cur_ss_mask = subtree_control;
1543 struct cgroup_subsys *ss;
1546 lockdep_assert_held(&cgroup_mutex);
1548 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1551 u16 new_ss_mask = cur_ss_mask;
1553 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1554 new_ss_mask |= ss->depends_on;
1555 } while_each_subsys_mask();
1558 * Mask out subsystems which aren't available. This can
1559 * happen only if some depended-upon subsystems were bound
1560 * to non-default hierarchies.
1562 new_ss_mask &= this_ss_mask;
1564 if (new_ss_mask == cur_ss_mask)
1566 cur_ss_mask = new_ss_mask;
1573 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1574 * @kn: the kernfs_node being serviced
1576 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1577 * the method finishes if locking succeeded. Note that once this function
1578 * returns the cgroup returned by cgroup_kn_lock_live() may become
1579 * inaccessible any time. If the caller intends to continue to access the
1580 * cgroup, it should pin it before invoking this function.
1582 void cgroup_kn_unlock(struct kernfs_node *kn)
1584 struct cgroup *cgrp;
1586 if (kernfs_type(kn) == KERNFS_DIR)
1589 cgrp = kn->parent->priv;
1591 mutex_unlock(&cgroup_mutex);
1593 kernfs_unbreak_active_protection(kn);
1598 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1599 * @kn: the kernfs_node being serviced
1600 * @drain_offline: perform offline draining on the cgroup
1602 * This helper is to be used by a cgroup kernfs method currently servicing
1603 * @kn. It breaks the active protection, performs cgroup locking and
1604 * verifies that the associated cgroup is alive. Returns the cgroup if
1605 * alive; otherwise, %NULL. A successful return should be undone by a
1606 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1607 * cgroup is drained of offlining csses before return.
1609 * Any cgroup kernfs method implementation which requires locking the
1610 * associated cgroup should use this helper. It avoids nesting cgroup
1611 * locking under kernfs active protection and allows all kernfs operations
1612 * including self-removal.
1614 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1616 struct cgroup *cgrp;
1618 if (kernfs_type(kn) == KERNFS_DIR)
1621 cgrp = kn->parent->priv;
1624 * We're gonna grab cgroup_mutex which nests outside kernfs
1625 * active_ref. cgroup liveliness check alone provides enough
1626 * protection against removal. Ensure @cgrp stays accessible and
1627 * break the active_ref protection.
1629 if (!cgroup_tryget(cgrp))
1631 kernfs_break_active_protection(kn);
1634 cgroup_lock_and_drain_offline(cgrp);
1636 mutex_lock(&cgroup_mutex);
1638 if (!cgroup_is_dead(cgrp))
1641 cgroup_kn_unlock(kn);
1645 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1647 char name[CGROUP_FILE_NAME_MAX];
1649 lockdep_assert_held(&cgroup_mutex);
1651 if (cft->file_offset) {
1652 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1653 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1655 spin_lock_irq(&cgroup_file_kn_lock);
1657 spin_unlock_irq(&cgroup_file_kn_lock);
1659 del_timer_sync(&cfile->notify_timer);
1662 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1666 * css_clear_dir - remove subsys files in a cgroup directory
1669 static void css_clear_dir(struct cgroup_subsys_state *css)
1671 struct cgroup *cgrp = css->cgroup;
1672 struct cftype *cfts;
1674 if (!(css->flags & CSS_VISIBLE))
1677 css->flags &= ~CSS_VISIBLE;
1680 if (cgroup_on_dfl(cgrp))
1681 cfts = cgroup_base_files;
1683 cfts = cgroup1_base_files;
1685 cgroup_addrm_files(css, cgrp, cfts, false);
1687 list_for_each_entry(cfts, &css->ss->cfts, node)
1688 cgroup_addrm_files(css, cgrp, cfts, false);
1693 * css_populate_dir - create subsys files in a cgroup directory
1696 * On failure, no file is added.
1698 static int css_populate_dir(struct cgroup_subsys_state *css)
1700 struct cgroup *cgrp = css->cgroup;
1701 struct cftype *cfts, *failed_cfts;
1704 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1708 if (cgroup_on_dfl(cgrp))
1709 cfts = cgroup_base_files;
1711 cfts = cgroup1_base_files;
1713 ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1717 list_for_each_entry(cfts, &css->ss->cfts, node) {
1718 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1726 css->flags |= CSS_VISIBLE;
1730 list_for_each_entry(cfts, &css->ss->cfts, node) {
1731 if (cfts == failed_cfts)
1733 cgroup_addrm_files(css, cgrp, cfts, false);
1738 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1740 struct cgroup *dcgrp = &dst_root->cgrp;
1741 struct cgroup_subsys *ss;
1743 u16 dfl_disable_ss_mask = 0;
1745 lockdep_assert_held(&cgroup_mutex);
1747 do_each_subsys_mask(ss, ssid, ss_mask) {
1749 * If @ss has non-root csses attached to it, can't move.
1750 * If @ss is an implicit controller, it is exempt from this
1751 * rule and can be stolen.
1753 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1754 !ss->implicit_on_dfl)
1757 /* can't move between two non-dummy roots either */
1758 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1762 * Collect ssid's that need to be disabled from default
1765 if (ss->root == &cgrp_dfl_root)
1766 dfl_disable_ss_mask |= 1 << ssid;
1768 } while_each_subsys_mask();
1770 if (dfl_disable_ss_mask) {
1771 struct cgroup *scgrp = &cgrp_dfl_root.cgrp;
1774 * Controllers from default hierarchy that need to be rebound
1775 * are all disabled together in one go.
1777 cgrp_dfl_root.subsys_mask &= ~dfl_disable_ss_mask;
1778 WARN_ON(cgroup_apply_control(scgrp));
1779 cgroup_finalize_control(scgrp, 0);
1782 do_each_subsys_mask(ss, ssid, ss_mask) {
1783 struct cgroup_root *src_root = ss->root;
1784 struct cgroup *scgrp = &src_root->cgrp;
1785 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1786 struct css_set *cset;
1788 WARN_ON(!css || cgroup_css(dcgrp, ss));
1790 if (src_root != &cgrp_dfl_root) {
1791 /* disable from the source */
1792 src_root->subsys_mask &= ~(1 << ssid);
1793 WARN_ON(cgroup_apply_control(scgrp));
1794 cgroup_finalize_control(scgrp, 0);
1798 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1799 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1800 ss->root = dst_root;
1801 css->cgroup = dcgrp;
1803 spin_lock_irq(&css_set_lock);
1804 hash_for_each(css_set_table, i, cset, hlist)
1805 list_move_tail(&cset->e_cset_node[ss->id],
1806 &dcgrp->e_csets[ss->id]);
1807 spin_unlock_irq(&css_set_lock);
1809 if (ss->css_rstat_flush) {
1810 list_del_rcu(&css->rstat_css_node);
1811 list_add_rcu(&css->rstat_css_node,
1812 &dcgrp->rstat_css_list);
1815 /* default hierarchy doesn't enable controllers by default */
1816 dst_root->subsys_mask |= 1 << ssid;
1817 if (dst_root == &cgrp_dfl_root) {
1818 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1820 dcgrp->subtree_control |= 1 << ssid;
1821 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1824 ret = cgroup_apply_control(dcgrp);
1826 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1831 } while_each_subsys_mask();
1833 kernfs_activate(dcgrp->kn);
1837 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1838 struct kernfs_root *kf_root)
1842 struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1843 struct cgroup *ns_cgroup;
1845 buf = kmalloc(PATH_MAX, GFP_KERNEL);
1849 spin_lock_irq(&css_set_lock);
1850 ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1851 len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1852 spin_unlock_irq(&css_set_lock);
1854 if (len >= PATH_MAX)
1857 seq_escape(sf, buf, " \t\n\\");
1864 enum cgroup2_param {
1866 Opt_memory_localevents,
1867 Opt_memory_recursiveprot,
1871 static const struct fs_parameter_spec cgroup2_fs_parameters[] = {
1872 fsparam_flag("nsdelegate", Opt_nsdelegate),
1873 fsparam_flag("memory_localevents", Opt_memory_localevents),
1874 fsparam_flag("memory_recursiveprot", Opt_memory_recursiveprot),
1878 static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1880 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1881 struct fs_parse_result result;
1884 opt = fs_parse(fc, cgroup2_fs_parameters, param, &result);
1889 case Opt_nsdelegate:
1890 ctx->flags |= CGRP_ROOT_NS_DELEGATE;
1892 case Opt_memory_localevents:
1893 ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1895 case Opt_memory_recursiveprot:
1896 ctx->flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1902 static void apply_cgroup_root_flags(unsigned int root_flags)
1904 if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
1905 if (root_flags & CGRP_ROOT_NS_DELEGATE)
1906 cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
1908 cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
1910 if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1911 cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1913 cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1915 if (root_flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1916 cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1918 cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1922 static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
1924 if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
1925 seq_puts(seq, ",nsdelegate");
1926 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1927 seq_puts(seq, ",memory_localevents");
1928 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1929 seq_puts(seq, ",memory_recursiveprot");
1933 static int cgroup_reconfigure(struct fs_context *fc)
1935 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1937 apply_cgroup_root_flags(ctx->flags);
1941 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1943 struct cgroup_subsys *ss;
1946 INIT_LIST_HEAD(&cgrp->self.sibling);
1947 INIT_LIST_HEAD(&cgrp->self.children);
1948 INIT_LIST_HEAD(&cgrp->cset_links);
1949 INIT_LIST_HEAD(&cgrp->pidlists);
1950 mutex_init(&cgrp->pidlist_mutex);
1951 cgrp->self.cgroup = cgrp;
1952 cgrp->self.flags |= CSS_ONLINE;
1953 cgrp->dom_cgrp = cgrp;
1954 cgrp->max_descendants = INT_MAX;
1955 cgrp->max_depth = INT_MAX;
1956 INIT_LIST_HEAD(&cgrp->rstat_css_list);
1957 prev_cputime_init(&cgrp->prev_cputime);
1959 for_each_subsys(ss, ssid)
1960 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1962 init_waitqueue_head(&cgrp->offline_waitq);
1963 INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1966 void init_cgroup_root(struct cgroup_fs_context *ctx)
1968 struct cgroup_root *root = ctx->root;
1969 struct cgroup *cgrp = &root->cgrp;
1971 INIT_LIST_HEAD(&root->root_list);
1972 atomic_set(&root->nr_cgrps, 1);
1974 init_cgroup_housekeeping(cgrp);
1976 root->flags = ctx->flags;
1977 if (ctx->release_agent)
1978 strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX);
1980 strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN);
1981 if (ctx->cpuset_clone_children)
1982 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1985 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1987 LIST_HEAD(tmp_links);
1988 struct cgroup *root_cgrp = &root->cgrp;
1989 struct kernfs_syscall_ops *kf_sops;
1990 struct css_set *cset;
1993 lockdep_assert_held(&cgroup_mutex);
1995 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
2001 * We're accessing css_set_count without locking css_set_lock here,
2002 * but that's OK - it can only be increased by someone holding
2003 * cgroup_lock, and that's us. Later rebinding may disable
2004 * controllers on the default hierarchy and thus create new csets,
2005 * which can't be more than the existing ones. Allocate 2x.
2007 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
2011 ret = cgroup_init_root_id(root);
2015 kf_sops = root == &cgrp_dfl_root ?
2016 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
2018 root->kf_root = kernfs_create_root(kf_sops,
2019 KERNFS_ROOT_CREATE_DEACTIVATED |
2020 KERNFS_ROOT_SUPPORT_EXPORTOP |
2021 KERNFS_ROOT_SUPPORT_USER_XATTR,
2023 if (IS_ERR(root->kf_root)) {
2024 ret = PTR_ERR(root->kf_root);
2027 root_cgrp->kn = root->kf_root->kn;
2028 WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
2029 root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
2031 ret = css_populate_dir(&root_cgrp->self);
2035 ret = cgroup_rstat_init(root_cgrp);
2039 ret = rebind_subsystems(root, ss_mask);
2043 ret = cgroup_bpf_inherit(root_cgrp);
2046 trace_cgroup_setup_root(root);
2049 * There must be no failure case after here, since rebinding takes
2050 * care of subsystems' refcounts, which are explicitly dropped in
2051 * the failure exit path.
2053 list_add(&root->root_list, &cgroup_roots);
2054 cgroup_root_count++;
2057 * Link the root cgroup in this hierarchy into all the css_set
2060 spin_lock_irq(&css_set_lock);
2061 hash_for_each(css_set_table, i, cset, hlist) {
2062 link_css_set(&tmp_links, cset, root_cgrp);
2063 if (css_set_populated(cset))
2064 cgroup_update_populated(root_cgrp, true);
2066 spin_unlock_irq(&css_set_lock);
2068 BUG_ON(!list_empty(&root_cgrp->self.children));
2069 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
2075 cgroup_rstat_exit(root_cgrp);
2077 kernfs_destroy_root(root->kf_root);
2078 root->kf_root = NULL;
2080 cgroup_exit_root_id(root);
2082 percpu_ref_exit(&root_cgrp->self.refcnt);
2084 free_cgrp_cset_links(&tmp_links);
2088 int cgroup_do_get_tree(struct fs_context *fc)
2090 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2093 ctx->kfc.root = ctx->root->kf_root;
2094 if (fc->fs_type == &cgroup2_fs_type)
2095 ctx->kfc.magic = CGROUP2_SUPER_MAGIC;
2097 ctx->kfc.magic = CGROUP_SUPER_MAGIC;
2098 ret = kernfs_get_tree(fc);
2101 * In non-init cgroup namespace, instead of root cgroup's dentry,
2102 * we return the dentry corresponding to the cgroupns->root_cgrp.
2104 if (!ret && ctx->ns != &init_cgroup_ns) {
2105 struct dentry *nsdentry;
2106 struct super_block *sb = fc->root->d_sb;
2107 struct cgroup *cgrp;
2109 mutex_lock(&cgroup_mutex);
2110 spin_lock_irq(&css_set_lock);
2112 cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root);
2114 spin_unlock_irq(&css_set_lock);
2115 mutex_unlock(&cgroup_mutex);
2117 nsdentry = kernfs_node_dentry(cgrp->kn, sb);
2119 if (IS_ERR(nsdentry)) {
2120 deactivate_locked_super(sb);
2121 ret = PTR_ERR(nsdentry);
2124 fc->root = nsdentry;
2127 if (!ctx->kfc.new_sb_created)
2128 cgroup_put(&ctx->root->cgrp);
2134 * Destroy a cgroup filesystem context.
2136 static void cgroup_fs_context_free(struct fs_context *fc)
2138 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2141 kfree(ctx->release_agent);
2142 put_cgroup_ns(ctx->ns);
2143 kernfs_free_fs_context(fc);
2147 static int cgroup_get_tree(struct fs_context *fc)
2149 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2152 cgrp_dfl_visible = true;
2153 cgroup_get_live(&cgrp_dfl_root.cgrp);
2154 ctx->root = &cgrp_dfl_root;
2156 ret = cgroup_do_get_tree(fc);
2158 apply_cgroup_root_flags(ctx->flags);
2162 static const struct fs_context_operations cgroup_fs_context_ops = {
2163 .free = cgroup_fs_context_free,
2164 .parse_param = cgroup2_parse_param,
2165 .get_tree = cgroup_get_tree,
2166 .reconfigure = cgroup_reconfigure,
2169 static const struct fs_context_operations cgroup1_fs_context_ops = {
2170 .free = cgroup_fs_context_free,
2171 .parse_param = cgroup1_parse_param,
2172 .get_tree = cgroup1_get_tree,
2173 .reconfigure = cgroup1_reconfigure,
2177 * Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2178 * we select the namespace we're going to use.
2180 static int cgroup_init_fs_context(struct fs_context *fc)
2182 struct cgroup_fs_context *ctx;
2184 ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL);
2188 ctx->ns = current->nsproxy->cgroup_ns;
2189 get_cgroup_ns(ctx->ns);
2190 fc->fs_private = &ctx->kfc;
2191 if (fc->fs_type == &cgroup2_fs_type)
2192 fc->ops = &cgroup_fs_context_ops;
2194 fc->ops = &cgroup1_fs_context_ops;
2195 put_user_ns(fc->user_ns);
2196 fc->user_ns = get_user_ns(ctx->ns->user_ns);
2201 static void cgroup_kill_sb(struct super_block *sb)
2203 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2204 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2207 * If @root doesn't have any children, start killing it.
2208 * This prevents new mounts by disabling percpu_ref_tryget_live().
2210 * And don't kill the default root.
2212 if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root &&
2213 !percpu_ref_is_dying(&root->cgrp.self.refcnt)) {
2214 cgroup_bpf_offline(&root->cgrp);
2215 percpu_ref_kill(&root->cgrp.self.refcnt);
2217 cgroup_put(&root->cgrp);
2221 struct file_system_type cgroup_fs_type = {
2223 .init_fs_context = cgroup_init_fs_context,
2224 .parameters = cgroup1_fs_parameters,
2225 .kill_sb = cgroup_kill_sb,
2226 .fs_flags = FS_USERNS_MOUNT,
2229 static struct file_system_type cgroup2_fs_type = {
2231 .init_fs_context = cgroup_init_fs_context,
2232 .parameters = cgroup2_fs_parameters,
2233 .kill_sb = cgroup_kill_sb,
2234 .fs_flags = FS_USERNS_MOUNT,
2237 #ifdef CONFIG_CPUSETS
2238 static const struct fs_context_operations cpuset_fs_context_ops = {
2239 .get_tree = cgroup1_get_tree,
2240 .free = cgroup_fs_context_free,
2244 * This is ugly, but preserves the userspace API for existing cpuset
2245 * users. If someone tries to mount the "cpuset" filesystem, we
2246 * silently switch it to mount "cgroup" instead
2248 static int cpuset_init_fs_context(struct fs_context *fc)
2250 char *agent = kstrdup("/sbin/cpuset_release_agent", GFP_USER);
2251 struct cgroup_fs_context *ctx;
2254 err = cgroup_init_fs_context(fc);
2260 fc->ops = &cpuset_fs_context_ops;
2262 ctx = cgroup_fc2context(fc);
2263 ctx->subsys_mask = 1 << cpuset_cgrp_id;
2264 ctx->flags |= CGRP_ROOT_NOPREFIX;
2265 ctx->release_agent = agent;
2267 get_filesystem(&cgroup_fs_type);
2268 put_filesystem(fc->fs_type);
2269 fc->fs_type = &cgroup_fs_type;
2274 static struct file_system_type cpuset_fs_type = {
2276 .init_fs_context = cpuset_init_fs_context,
2277 .fs_flags = FS_USERNS_MOUNT,
2281 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2282 struct cgroup_namespace *ns)
2284 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2286 return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2289 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2290 struct cgroup_namespace *ns)
2294 mutex_lock(&cgroup_mutex);
2295 spin_lock_irq(&css_set_lock);
2297 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2299 spin_unlock_irq(&css_set_lock);
2300 mutex_unlock(&cgroup_mutex);
2304 EXPORT_SYMBOL_GPL(cgroup_path_ns);
2307 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2308 * @task: target task
2309 * @buf: the buffer to write the path into
2310 * @buflen: the length of the buffer
2312 * Determine @task's cgroup on the first (the one with the lowest non-zero
2313 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2314 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2315 * cgroup controller callbacks.
2317 * Return value is the same as kernfs_path().
2319 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2321 struct cgroup_root *root;
2322 struct cgroup *cgrp;
2323 int hierarchy_id = 1;
2326 mutex_lock(&cgroup_mutex);
2327 spin_lock_irq(&css_set_lock);
2329 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2332 cgrp = task_cgroup_from_root(task, root);
2333 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2335 /* if no hierarchy exists, everyone is in "/" */
2336 ret = strlcpy(buf, "/", buflen);
2339 spin_unlock_irq(&css_set_lock);
2340 mutex_unlock(&cgroup_mutex);
2343 EXPORT_SYMBOL_GPL(task_cgroup_path);
2346 * cgroup_migrate_add_task - add a migration target task to a migration context
2347 * @task: target task
2348 * @mgctx: target migration context
2350 * Add @task, which is a migration target, to @mgctx->tset. This function
2351 * becomes noop if @task doesn't need to be migrated. @task's css_set
2352 * should have been added as a migration source and @task->cg_list will be
2353 * moved from the css_set's tasks list to mg_tasks one.
2355 static void cgroup_migrate_add_task(struct task_struct *task,
2356 struct cgroup_mgctx *mgctx)
2358 struct css_set *cset;
2360 lockdep_assert_held(&css_set_lock);
2362 /* @task either already exited or can't exit until the end */
2363 if (task->flags & PF_EXITING)
2366 /* cgroup_threadgroup_rwsem protects racing against forks */
2367 WARN_ON_ONCE(list_empty(&task->cg_list));
2369 cset = task_css_set(task);
2370 if (!cset->mg_src_cgrp)
2373 mgctx->tset.nr_tasks++;
2375 list_move_tail(&task->cg_list, &cset->mg_tasks);
2376 if (list_empty(&cset->mg_node))
2377 list_add_tail(&cset->mg_node,
2378 &mgctx->tset.src_csets);
2379 if (list_empty(&cset->mg_dst_cset->mg_node))
2380 list_add_tail(&cset->mg_dst_cset->mg_node,
2381 &mgctx->tset.dst_csets);
2385 * cgroup_taskset_first - reset taskset and return the first task
2386 * @tset: taskset of interest
2387 * @dst_cssp: output variable for the destination css
2389 * @tset iteration is initialized and the first task is returned.
2391 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2392 struct cgroup_subsys_state **dst_cssp)
2394 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2395 tset->cur_task = NULL;
2397 return cgroup_taskset_next(tset, dst_cssp);
2401 * cgroup_taskset_next - iterate to the next task in taskset
2402 * @tset: taskset of interest
2403 * @dst_cssp: output variable for the destination css
2405 * Return the next task in @tset. Iteration must have been initialized
2406 * with cgroup_taskset_first().
2408 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2409 struct cgroup_subsys_state **dst_cssp)
2411 struct css_set *cset = tset->cur_cset;
2412 struct task_struct *task = tset->cur_task;
2414 while (CGROUP_HAS_SUBSYS_CONFIG && &cset->mg_node != tset->csets) {
2416 task = list_first_entry(&cset->mg_tasks,
2417 struct task_struct, cg_list);
2419 task = list_next_entry(task, cg_list);
2421 if (&task->cg_list != &cset->mg_tasks) {
2422 tset->cur_cset = cset;
2423 tset->cur_task = task;
2426 * This function may be called both before and
2427 * after cgroup_taskset_migrate(). The two cases
2428 * can be distinguished by looking at whether @cset
2429 * has its ->mg_dst_cset set.
2431 if (cset->mg_dst_cset)
2432 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2434 *dst_cssp = cset->subsys[tset->ssid];
2439 cset = list_next_entry(cset, mg_node);
2447 * cgroup_migrate_execute - migrate a taskset
2448 * @mgctx: migration context
2450 * Migrate tasks in @mgctx as setup by migration preparation functions.
2451 * This function fails iff one of the ->can_attach callbacks fails and
2452 * guarantees that either all or none of the tasks in @mgctx are migrated.
2453 * @mgctx is consumed regardless of success.
2455 static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2457 struct cgroup_taskset *tset = &mgctx->tset;
2458 struct cgroup_subsys *ss;
2459 struct task_struct *task, *tmp_task;
2460 struct css_set *cset, *tmp_cset;
2461 int ssid, failed_ssid, ret;
2463 /* check that we can legitimately attach to the cgroup */
2464 if (tset->nr_tasks) {
2465 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2466 if (ss->can_attach) {
2468 ret = ss->can_attach(tset);
2471 goto out_cancel_attach;
2474 } while_each_subsys_mask();
2478 * Now that we're guaranteed success, proceed to move all tasks to
2479 * the new cgroup. There are no failure cases after here, so this
2480 * is the commit point.
2482 spin_lock_irq(&css_set_lock);
2483 list_for_each_entry(cset, &tset->src_csets, mg_node) {
2484 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2485 struct css_set *from_cset = task_css_set(task);
2486 struct css_set *to_cset = cset->mg_dst_cset;
2488 get_css_set(to_cset);
2489 to_cset->nr_tasks++;
2490 css_set_move_task(task, from_cset, to_cset, true);
2491 from_cset->nr_tasks--;
2493 * If the source or destination cgroup is frozen,
2494 * the task might require to change its state.
2496 cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp,
2498 put_css_set_locked(from_cset);
2502 spin_unlock_irq(&css_set_lock);
2505 * Migration is committed, all target tasks are now on dst_csets.
2506 * Nothing is sensitive to fork() after this point. Notify
2507 * controllers that migration is complete.
2509 tset->csets = &tset->dst_csets;
2511 if (tset->nr_tasks) {
2512 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2517 } while_each_subsys_mask();
2521 goto out_release_tset;
2524 if (tset->nr_tasks) {
2525 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2526 if (ssid == failed_ssid)
2528 if (ss->cancel_attach) {
2530 ss->cancel_attach(tset);
2532 } while_each_subsys_mask();
2535 spin_lock_irq(&css_set_lock);
2536 list_splice_init(&tset->dst_csets, &tset->src_csets);
2537 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2538 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2539 list_del_init(&cset->mg_node);
2541 spin_unlock_irq(&css_set_lock);
2544 * Re-initialize the cgroup_taskset structure in case it is reused
2545 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2549 tset->csets = &tset->src_csets;
2554 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2555 * @dst_cgrp: destination cgroup to test
2557 * On the default hierarchy, except for the mixable, (possible) thread root
2558 * and threaded cgroups, subtree_control must be zero for migration
2559 * destination cgroups with tasks so that child cgroups don't compete
2562 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2564 /* v1 doesn't have any restriction */
2565 if (!cgroup_on_dfl(dst_cgrp))
2568 /* verify @dst_cgrp can host resources */
2569 if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
2572 /* mixables don't care */
2573 if (cgroup_is_mixable(dst_cgrp))
2577 * If @dst_cgrp is already or can become a thread root or is
2578 * threaded, it doesn't matter.
2580 if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
2583 /* apply no-internal-process constraint */
2584 if (dst_cgrp->subtree_control)
2591 * cgroup_migrate_finish - cleanup after attach
2592 * @mgctx: migration context
2594 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2595 * those functions for details.
2597 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2599 LIST_HEAD(preloaded);
2600 struct css_set *cset, *tmp_cset;
2602 lockdep_assert_held(&cgroup_mutex);
2604 spin_lock_irq(&css_set_lock);
2606 list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded);
2607 list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded);
2609 list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) {
2610 cset->mg_src_cgrp = NULL;
2611 cset->mg_dst_cgrp = NULL;
2612 cset->mg_dst_cset = NULL;
2613 list_del_init(&cset->mg_preload_node);
2614 put_css_set_locked(cset);
2617 spin_unlock_irq(&css_set_lock);
2621 * cgroup_migrate_add_src - add a migration source css_set
2622 * @src_cset: the source css_set to add
2623 * @dst_cgrp: the destination cgroup
2624 * @mgctx: migration context
2626 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2627 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2628 * up by cgroup_migrate_finish().
2630 * This function may be called without holding cgroup_threadgroup_rwsem
2631 * even if the target is a process. Threads may be created and destroyed
2632 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2633 * into play and the preloaded css_sets are guaranteed to cover all
2636 void cgroup_migrate_add_src(struct css_set *src_cset,
2637 struct cgroup *dst_cgrp,
2638 struct cgroup_mgctx *mgctx)
2640 struct cgroup *src_cgrp;
2642 lockdep_assert_held(&cgroup_mutex);
2643 lockdep_assert_held(&css_set_lock);
2646 * If ->dead, @src_set is associated with one or more dead cgroups
2647 * and doesn't contain any migratable tasks. Ignore it early so
2648 * that the rest of migration path doesn't get confused by it.
2653 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2655 if (!list_empty(&src_cset->mg_preload_node))
2658 WARN_ON(src_cset->mg_src_cgrp);
2659 WARN_ON(src_cset->mg_dst_cgrp);
2660 WARN_ON(!list_empty(&src_cset->mg_tasks));
2661 WARN_ON(!list_empty(&src_cset->mg_node));
2663 src_cset->mg_src_cgrp = src_cgrp;
2664 src_cset->mg_dst_cgrp = dst_cgrp;
2665 get_css_set(src_cset);
2666 list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets);
2670 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2671 * @mgctx: migration context
2673 * Tasks are about to be moved and all the source css_sets have been
2674 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2675 * pins all destination css_sets, links each to its source, and append them
2676 * to @mgctx->preloaded_dst_csets.
2678 * This function must be called after cgroup_migrate_add_src() has been
2679 * called on each migration source css_set. After migration is performed
2680 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2683 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2685 struct css_set *src_cset, *tmp_cset;
2687 lockdep_assert_held(&cgroup_mutex);
2689 /* look up the dst cset for each src cset and link it to src */
2690 list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2692 struct css_set *dst_cset;
2693 struct cgroup_subsys *ss;
2696 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2700 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2703 * If src cset equals dst, it's noop. Drop the src.
2704 * cgroup_migrate() will skip the cset too. Note that we
2705 * can't handle src == dst as some nodes are used by both.
2707 if (src_cset == dst_cset) {
2708 src_cset->mg_src_cgrp = NULL;
2709 src_cset->mg_dst_cgrp = NULL;
2710 list_del_init(&src_cset->mg_preload_node);
2711 put_css_set(src_cset);
2712 put_css_set(dst_cset);
2716 src_cset->mg_dst_cset = dst_cset;
2718 if (list_empty(&dst_cset->mg_preload_node))
2719 list_add_tail(&dst_cset->mg_preload_node,
2720 &mgctx->preloaded_dst_csets);
2722 put_css_set(dst_cset);
2724 for_each_subsys(ss, ssid)
2725 if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2726 mgctx->ss_mask |= 1 << ssid;
2733 * cgroup_migrate - migrate a process or task to a cgroup
2734 * @leader: the leader of the process or the task to migrate
2735 * @threadgroup: whether @leader points to the whole process or a single task
2736 * @mgctx: migration context
2738 * Migrate a process or task denoted by @leader. If migrating a process,
2739 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2740 * responsible for invoking cgroup_migrate_add_src() and
2741 * cgroup_migrate_prepare_dst() on the targets before invoking this
2742 * function and following up with cgroup_migrate_finish().
2744 * As long as a controller's ->can_attach() doesn't fail, this function is
2745 * guaranteed to succeed. This means that, excluding ->can_attach()
2746 * failure, when migrating multiple targets, the success or failure can be
2747 * decided for all targets by invoking group_migrate_prepare_dst() before
2748 * actually starting migrating.
2750 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2751 struct cgroup_mgctx *mgctx)
2753 struct task_struct *task;
2756 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2757 * already PF_EXITING could be freed from underneath us unless we
2758 * take an rcu_read_lock.
2760 spin_lock_irq(&css_set_lock);
2764 cgroup_migrate_add_task(task, mgctx);
2767 } while_each_thread(leader, task);
2769 spin_unlock_irq(&css_set_lock);
2771 return cgroup_migrate_execute(mgctx);
2775 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2776 * @dst_cgrp: the cgroup to attach to
2777 * @leader: the task or the leader of the threadgroup to be attached
2778 * @threadgroup: attach the whole threadgroup?
2780 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2782 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2785 DEFINE_CGROUP_MGCTX(mgctx);
2786 struct task_struct *task;
2789 /* look up all src csets */
2790 spin_lock_irq(&css_set_lock);
2794 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2797 } while_each_thread(leader, task);
2799 spin_unlock_irq(&css_set_lock);
2801 /* prepare dst csets and commit */
2802 ret = cgroup_migrate_prepare_dst(&mgctx);
2804 ret = cgroup_migrate(leader, threadgroup, &mgctx);
2806 cgroup_migrate_finish(&mgctx);
2809 TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup);
2814 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
2816 __acquires(&cgroup_threadgroup_rwsem)
2818 struct task_struct *tsk;
2821 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2822 return ERR_PTR(-EINVAL);
2825 * If we migrate a single thread, we don't care about threadgroup
2826 * stability. If the thread is `current`, it won't exit(2) under our
2827 * hands or change PID through exec(2). We exclude
2828 * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write
2829 * callers by cgroup_mutex.
2830 * Therefore, we can skip the global lock.
2832 lockdep_assert_held(&cgroup_mutex);
2833 if (pid || threadgroup) {
2834 percpu_down_write(&cgroup_threadgroup_rwsem);
2842 tsk = find_task_by_vpid(pid);
2844 tsk = ERR_PTR(-ESRCH);
2845 goto out_unlock_threadgroup;
2852 tsk = tsk->group_leader;
2855 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2856 * If userland migrates such a kthread to a non-root cgroup, it can
2857 * become trapped in a cpuset, or RT kthread may be born in a
2858 * cgroup with no rt_runtime allocated. Just say no.
2860 if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2861 tsk = ERR_PTR(-EINVAL);
2862 goto out_unlock_threadgroup;
2865 get_task_struct(tsk);
2866 goto out_unlock_rcu;
2868 out_unlock_threadgroup:
2870 percpu_up_write(&cgroup_threadgroup_rwsem);
2878 void cgroup_procs_write_finish(struct task_struct *task, bool locked)
2879 __releases(&cgroup_threadgroup_rwsem)
2881 struct cgroup_subsys *ss;
2884 /* release reference from cgroup_procs_write_start() */
2885 put_task_struct(task);
2888 percpu_up_write(&cgroup_threadgroup_rwsem);
2889 for_each_subsys(ss, ssid)
2890 if (ss->post_attach)
2894 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2896 struct cgroup_subsys *ss;
2897 bool printed = false;
2900 do_each_subsys_mask(ss, ssid, ss_mask) {
2903 seq_puts(seq, ss->name);
2905 } while_each_subsys_mask();
2907 seq_putc(seq, '\n');
2910 /* show controllers which are enabled from the parent */
2911 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2913 struct cgroup *cgrp = seq_css(seq)->cgroup;
2915 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2919 /* show controllers which are enabled for a given cgroup's children */
2920 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2922 struct cgroup *cgrp = seq_css(seq)->cgroup;
2924 cgroup_print_ss_mask(seq, cgrp->subtree_control);
2929 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2930 * @cgrp: root of the subtree to update csses for
2932 * @cgrp's control masks have changed and its subtree's css associations
2933 * need to be updated accordingly. This function looks up all css_sets
2934 * which are attached to the subtree, creates the matching updated css_sets
2935 * and migrates the tasks to the new ones.
2937 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2939 DEFINE_CGROUP_MGCTX(mgctx);
2940 struct cgroup_subsys_state *d_css;
2941 struct cgroup *dsct;
2942 struct css_set *src_cset;
2945 lockdep_assert_held(&cgroup_mutex);
2947 percpu_down_write(&cgroup_threadgroup_rwsem);
2949 /* look up all csses currently attached to @cgrp's subtree */
2950 spin_lock_irq(&css_set_lock);
2951 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2952 struct cgrp_cset_link *link;
2954 list_for_each_entry(link, &dsct->cset_links, cset_link)
2955 cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2957 spin_unlock_irq(&css_set_lock);
2959 /* NULL dst indicates self on default hierarchy */
2960 ret = cgroup_migrate_prepare_dst(&mgctx);
2964 spin_lock_irq(&css_set_lock);
2965 list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) {
2966 struct task_struct *task, *ntask;
2968 /* all tasks in src_csets need to be migrated */
2969 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2970 cgroup_migrate_add_task(task, &mgctx);
2972 spin_unlock_irq(&css_set_lock);
2974 ret = cgroup_migrate_execute(&mgctx);
2976 cgroup_migrate_finish(&mgctx);
2977 percpu_up_write(&cgroup_threadgroup_rwsem);
2982 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2983 * @cgrp: root of the target subtree
2985 * Because css offlining is asynchronous, userland may try to re-enable a
2986 * controller while the previous css is still around. This function grabs
2987 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2989 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
2990 __acquires(&cgroup_mutex)
2992 struct cgroup *dsct;
2993 struct cgroup_subsys_state *d_css;
2994 struct cgroup_subsys *ss;
2998 mutex_lock(&cgroup_mutex);
3000 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3001 for_each_subsys(ss, ssid) {
3002 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3005 if (!css || !percpu_ref_is_dying(&css->refcnt))
3008 cgroup_get_live(dsct);
3009 prepare_to_wait(&dsct->offline_waitq, &wait,
3010 TASK_UNINTERRUPTIBLE);
3012 mutex_unlock(&cgroup_mutex);
3014 finish_wait(&dsct->offline_waitq, &wait);
3023 * cgroup_save_control - save control masks and dom_cgrp of a subtree
3024 * @cgrp: root of the target subtree
3026 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
3027 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3030 static void cgroup_save_control(struct cgroup *cgrp)
3032 struct cgroup *dsct;
3033 struct cgroup_subsys_state *d_css;
3035 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3036 dsct->old_subtree_control = dsct->subtree_control;
3037 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
3038 dsct->old_dom_cgrp = dsct->dom_cgrp;
3043 * cgroup_propagate_control - refresh control masks of a subtree
3044 * @cgrp: root of the target subtree
3046 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3047 * ->subtree_control and propagate controller availability through the
3048 * subtree so that descendants don't have unavailable controllers enabled.
3050 static void cgroup_propagate_control(struct cgroup *cgrp)
3052 struct cgroup *dsct;
3053 struct cgroup_subsys_state *d_css;
3055 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3056 dsct->subtree_control &= cgroup_control(dsct);
3057 dsct->subtree_ss_mask =
3058 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3059 cgroup_ss_mask(dsct));
3064 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3065 * @cgrp: root of the target subtree
3067 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3068 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3071 static void cgroup_restore_control(struct cgroup *cgrp)
3073 struct cgroup *dsct;
3074 struct cgroup_subsys_state *d_css;
3076 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3077 dsct->subtree_control = dsct->old_subtree_control;
3078 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3079 dsct->dom_cgrp = dsct->old_dom_cgrp;
3083 static bool css_visible(struct cgroup_subsys_state *css)
3085 struct cgroup_subsys *ss = css->ss;
3086 struct cgroup *cgrp = css->cgroup;
3088 if (cgroup_control(cgrp) & (1 << ss->id))
3090 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3092 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3096 * cgroup_apply_control_enable - enable or show csses according to control
3097 * @cgrp: root of the target subtree
3099 * Walk @cgrp's subtree and create new csses or make the existing ones
3100 * visible. A css is created invisible if it's being implicitly enabled
3101 * through dependency. An invisible css is made visible when the userland
3102 * explicitly enables it.
3104 * Returns 0 on success, -errno on failure. On failure, csses which have
3105 * been processed already aren't cleaned up. The caller is responsible for
3106 * cleaning up with cgroup_apply_control_disable().
3108 static int cgroup_apply_control_enable(struct cgroup *cgrp)
3110 struct cgroup *dsct;
3111 struct cgroup_subsys_state *d_css;
3112 struct cgroup_subsys *ss;
3115 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3116 for_each_subsys(ss, ssid) {
3117 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3119 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3123 css = css_create(dsct, ss);
3125 return PTR_ERR(css);
3128 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3130 if (css_visible(css)) {
3131 ret = css_populate_dir(css);
3142 * cgroup_apply_control_disable - kill or hide csses according to control
3143 * @cgrp: root of the target subtree
3145 * Walk @cgrp's subtree and kill and hide csses so that they match
3146 * cgroup_ss_mask() and cgroup_visible_mask().
3148 * A css is hidden when the userland requests it to be disabled while other
3149 * subsystems are still depending on it. The css must not actively control
3150 * resources and be in the vanilla state if it's made visible again later.
3151 * Controllers which may be depended upon should provide ->css_reset() for
3154 static void cgroup_apply_control_disable(struct cgroup *cgrp)
3156 struct cgroup *dsct;
3157 struct cgroup_subsys_state *d_css;
3158 struct cgroup_subsys *ss;
3161 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3162 for_each_subsys(ss, ssid) {
3163 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3168 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3171 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
3173 } else if (!css_visible(css)) {
3183 * cgroup_apply_control - apply control mask updates to the subtree
3184 * @cgrp: root of the target subtree
3186 * subsystems can be enabled and disabled in a subtree using the following
3189 * 1. Call cgroup_save_control() to stash the current state.
3190 * 2. Update ->subtree_control masks in the subtree as desired.
3191 * 3. Call cgroup_apply_control() to apply the changes.
3192 * 4. Optionally perform other related operations.
3193 * 5. Call cgroup_finalize_control() to finish up.
3195 * This function implements step 3 and propagates the mask changes
3196 * throughout @cgrp's subtree, updates csses accordingly and perform
3197 * process migrations.
3199 static int cgroup_apply_control(struct cgroup *cgrp)
3203 cgroup_propagate_control(cgrp);
3205 ret = cgroup_apply_control_enable(cgrp);
3210 * At this point, cgroup_e_css_by_mask() results reflect the new csses
3211 * making the following cgroup_update_dfl_csses() properly update
3212 * css associations of all tasks in the subtree.
3214 ret = cgroup_update_dfl_csses(cgrp);
3222 * cgroup_finalize_control - finalize control mask update
3223 * @cgrp: root of the target subtree
3224 * @ret: the result of the update
3226 * Finalize control mask update. See cgroup_apply_control() for more info.
3228 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3231 cgroup_restore_control(cgrp);
3232 cgroup_propagate_control(cgrp);
3235 cgroup_apply_control_disable(cgrp);
3238 static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
3240 u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
3242 /* if nothing is getting enabled, nothing to worry about */
3246 /* can @cgrp host any resources? */
3247 if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
3250 /* mixables don't care */
3251 if (cgroup_is_mixable(cgrp))
3254 if (domain_enable) {
3255 /* can't enable domain controllers inside a thread subtree */
3256 if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3260 * Threaded controllers can handle internal competitions
3261 * and are always allowed inside a (prospective) thread
3264 if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3269 * Controllers can't be enabled for a cgroup with tasks to avoid
3270 * child cgroups competing against tasks.
3272 if (cgroup_has_tasks(cgrp))
3278 /* change the enabled child controllers for a cgroup in the default hierarchy */
3279 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3280 char *buf, size_t nbytes,
3283 u16 enable = 0, disable = 0;
3284 struct cgroup *cgrp, *child;
3285 struct cgroup_subsys *ss;
3290 * Parse input - space separated list of subsystem names prefixed
3291 * with either + or -.
3293 buf = strstrip(buf);
3294 while ((tok = strsep(&buf, " "))) {
3297 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3298 if (!cgroup_ssid_enabled(ssid) ||
3299 strcmp(tok + 1, ss->name))
3303 enable |= 1 << ssid;
3304 disable &= ~(1 << ssid);
3305 } else if (*tok == '-') {
3306 disable |= 1 << ssid;
3307 enable &= ~(1 << ssid);
3312 } while_each_subsys_mask();
3313 if (ssid == CGROUP_SUBSYS_COUNT)
3317 cgrp = cgroup_kn_lock_live(of->kn, true);
3321 for_each_subsys(ss, ssid) {
3322 if (enable & (1 << ssid)) {
3323 if (cgrp->subtree_control & (1 << ssid)) {
3324 enable &= ~(1 << ssid);
3328 if (!(cgroup_control(cgrp) & (1 << ssid))) {
3332 } else if (disable & (1 << ssid)) {
3333 if (!(cgrp->subtree_control & (1 << ssid))) {
3334 disable &= ~(1 << ssid);
3338 /* a child has it enabled? */
3339 cgroup_for_each_live_child(child, cgrp) {
3340 if (child->subtree_control & (1 << ssid)) {
3348 if (!enable && !disable) {
3353 ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3357 /* save and update control masks and prepare csses */
3358 cgroup_save_control(cgrp);
3360 cgrp->subtree_control |= enable;
3361 cgrp->subtree_control &= ~disable;
3363 ret = cgroup_apply_control(cgrp);
3364 cgroup_finalize_control(cgrp, ret);
3368 kernfs_activate(cgrp->kn);
3370 cgroup_kn_unlock(of->kn);
3371 return ret ?: nbytes;
3375 * cgroup_enable_threaded - make @cgrp threaded
3376 * @cgrp: the target cgroup
3378 * Called when "threaded" is written to the cgroup.type interface file and
3379 * tries to make @cgrp threaded and join the parent's resource domain.
3380 * This function is never called on the root cgroup as cgroup.type doesn't
3383 static int cgroup_enable_threaded(struct cgroup *cgrp)
3385 struct cgroup *parent = cgroup_parent(cgrp);
3386 struct cgroup *dom_cgrp = parent->dom_cgrp;
3387 struct cgroup *dsct;
3388 struct cgroup_subsys_state *d_css;
3391 lockdep_assert_held(&cgroup_mutex);
3393 /* noop if already threaded */
3394 if (cgroup_is_threaded(cgrp))
3398 * If @cgroup is populated or has domain controllers enabled, it
3399 * can't be switched. While the below cgroup_can_be_thread_root()
3400 * test can catch the same conditions, that's only when @parent is
3401 * not mixable, so let's check it explicitly.
3403 if (cgroup_is_populated(cgrp) ||
3404 cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
3407 /* we're joining the parent's domain, ensure its validity */
3408 if (!cgroup_is_valid_domain(dom_cgrp) ||
3409 !cgroup_can_be_thread_root(dom_cgrp))
3413 * The following shouldn't cause actual migrations and should
3416 cgroup_save_control(cgrp);
3418 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)
3419 if (dsct == cgrp || cgroup_is_threaded(dsct))
3420 dsct->dom_cgrp = dom_cgrp;
3422 ret = cgroup_apply_control(cgrp);
3424 parent->nr_threaded_children++;
3426 cgroup_finalize_control(cgrp, ret);
3430 static int cgroup_type_show(struct seq_file *seq, void *v)
3432 struct cgroup *cgrp = seq_css(seq)->cgroup;
3434 if (cgroup_is_threaded(cgrp))
3435 seq_puts(seq, "threaded\n");
3436 else if (!cgroup_is_valid_domain(cgrp))
3437 seq_puts(seq, "domain invalid\n");
3438 else if (cgroup_is_thread_root(cgrp))
3439 seq_puts(seq, "domain threaded\n");
3441 seq_puts(seq, "domain\n");
3446 static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3447 size_t nbytes, loff_t off)
3449 struct cgroup *cgrp;
3452 /* only switching to threaded mode is supported */
3453 if (strcmp(strstrip(buf), "threaded"))
3456 /* drain dying csses before we re-apply (threaded) subtree control */
3457 cgrp = cgroup_kn_lock_live(of->kn, true);
3461 /* threaded can only be enabled */
3462 ret = cgroup_enable_threaded(cgrp);
3464 cgroup_kn_unlock(of->kn);
3465 return ret ?: nbytes;
3468 static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3470 struct cgroup *cgrp = seq_css(seq)->cgroup;
3471 int descendants = READ_ONCE(cgrp->max_descendants);
3473 if (descendants == INT_MAX)
3474 seq_puts(seq, "max\n");
3476 seq_printf(seq, "%d\n", descendants);
3481 static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3482 char *buf, size_t nbytes, loff_t off)
3484 struct cgroup *cgrp;
3488 buf = strstrip(buf);
3489 if (!strcmp(buf, "max")) {
3490 descendants = INT_MAX;
3492 ret = kstrtoint(buf, 0, &descendants);
3497 if (descendants < 0)
3500 cgrp = cgroup_kn_lock_live(of->kn, false);
3504 cgrp->max_descendants = descendants;
3506 cgroup_kn_unlock(of->kn);
3511 static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3513 struct cgroup *cgrp = seq_css(seq)->cgroup;
3514 int depth = READ_ONCE(cgrp->max_depth);
3516 if (depth == INT_MAX)
3517 seq_puts(seq, "max\n");
3519 seq_printf(seq, "%d\n", depth);
3524 static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3525 char *buf, size_t nbytes, loff_t off)
3527 struct cgroup *cgrp;
3531 buf = strstrip(buf);
3532 if (!strcmp(buf, "max")) {
3535 ret = kstrtoint(buf, 0, &depth);
3543 cgrp = cgroup_kn_lock_live(of->kn, false);
3547 cgrp->max_depth = depth;
3549 cgroup_kn_unlock(of->kn);
3554 static int cgroup_events_show(struct seq_file *seq, void *v)
3556 struct cgroup *cgrp = seq_css(seq)->cgroup;
3558 seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp));
3559 seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags));
3564 static int cgroup_stat_show(struct seq_file *seq, void *v)
3566 struct cgroup *cgroup = seq_css(seq)->cgroup;
3568 seq_printf(seq, "nr_descendants %d\n",
3569 cgroup->nr_descendants);
3570 seq_printf(seq, "nr_dying_descendants %d\n",
3571 cgroup->nr_dying_descendants);
3576 static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq,
3577 struct cgroup *cgrp, int ssid)
3579 struct cgroup_subsys *ss = cgroup_subsys[ssid];
3580 struct cgroup_subsys_state *css;
3583 if (!ss->css_extra_stat_show)
3586 css = cgroup_tryget_css(cgrp, ss);
3590 ret = ss->css_extra_stat_show(seq, css);
3595 static int cpu_stat_show(struct seq_file *seq, void *v)
3597 struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
3600 cgroup_base_stat_cputime_show(seq);
3601 #ifdef CONFIG_CGROUP_SCHED
3602 ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id);
3608 static int cgroup_io_pressure_show(struct seq_file *seq, void *v)
3610 struct cgroup *cgrp = seq_css(seq)->cgroup;
3611 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3613 return psi_show(seq, psi, PSI_IO);
3615 static int cgroup_memory_pressure_show(struct seq_file *seq, void *v)
3617 struct cgroup *cgrp = seq_css(seq)->cgroup;
3618 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3620 return psi_show(seq, psi, PSI_MEM);
3622 static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
3624 struct cgroup *cgrp = seq_css(seq)->cgroup;
3625 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3627 return psi_show(seq, psi, PSI_CPU);
3630 static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, char *buf,
3631 size_t nbytes, enum psi_res res)
3633 struct psi_trigger *new;
3634 struct cgroup *cgrp;
3635 struct psi_group *psi;
3637 cgrp = cgroup_kn_lock_live(of->kn, false);
3642 cgroup_kn_unlock(of->kn);
3644 psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3645 new = psi_trigger_create(psi, buf, nbytes, res);
3648 return PTR_ERR(new);
3651 psi_trigger_replace(&of->priv, new);
3658 static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of,
3659 char *buf, size_t nbytes,
3662 return cgroup_pressure_write(of, buf, nbytes, PSI_IO);
3665 static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of,
3666 char *buf, size_t nbytes,
3669 return cgroup_pressure_write(of, buf, nbytes, PSI_MEM);
3672 static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of,
3673 char *buf, size_t nbytes,
3676 return cgroup_pressure_write(of, buf, nbytes, PSI_CPU);
3679 static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of,
3682 return psi_trigger_poll(&of->priv, of->file, pt);
3685 static void cgroup_pressure_release(struct kernfs_open_file *of)
3687 psi_trigger_replace(&of->priv, NULL);
3690 bool cgroup_psi_enabled(void)
3692 return (cgroup_feature_disable_mask & (1 << OPT_FEATURE_PRESSURE)) == 0;
3695 #else /* CONFIG_PSI */
3696 bool cgroup_psi_enabled(void)
3701 #endif /* CONFIG_PSI */
3703 static int cgroup_freeze_show(struct seq_file *seq, void *v)
3705 struct cgroup *cgrp = seq_css(seq)->cgroup;
3707 seq_printf(seq, "%d\n", cgrp->freezer.freeze);
3712 static ssize_t cgroup_freeze_write(struct kernfs_open_file *of,
3713 char *buf, size_t nbytes, loff_t off)
3715 struct cgroup *cgrp;
3719 ret = kstrtoint(strstrip(buf), 0, &freeze);
3723 if (freeze < 0 || freeze > 1)
3726 cgrp = cgroup_kn_lock_live(of->kn, false);
3730 cgroup_freeze(cgrp, freeze);
3732 cgroup_kn_unlock(of->kn);
3737 static void __cgroup_kill(struct cgroup *cgrp)
3739 struct css_task_iter it;
3740 struct task_struct *task;
3742 lockdep_assert_held(&cgroup_mutex);
3744 spin_lock_irq(&css_set_lock);
3745 set_bit(CGRP_KILL, &cgrp->flags);
3746 spin_unlock_irq(&css_set_lock);
3748 css_task_iter_start(&cgrp->self, CSS_TASK_ITER_PROCS | CSS_TASK_ITER_THREADED, &it);
3749 while ((task = css_task_iter_next(&it))) {
3750 /* Ignore kernel threads here. */
3751 if (task->flags & PF_KTHREAD)
3754 /* Skip tasks that are already dying. */
3755 if (__fatal_signal_pending(task))
3758 send_sig(SIGKILL, task, 0);
3760 css_task_iter_end(&it);
3762 spin_lock_irq(&css_set_lock);
3763 clear_bit(CGRP_KILL, &cgrp->flags);
3764 spin_unlock_irq(&css_set_lock);
3767 static void cgroup_kill(struct cgroup *cgrp)
3769 struct cgroup_subsys_state *css;
3770 struct cgroup *dsct;
3772 lockdep_assert_held(&cgroup_mutex);
3774 cgroup_for_each_live_descendant_pre(dsct, css, cgrp)
3775 __cgroup_kill(dsct);
3778 static ssize_t cgroup_kill_write(struct kernfs_open_file *of, char *buf,
3779 size_t nbytes, loff_t off)
3783 struct cgroup *cgrp;
3785 ret = kstrtoint(strstrip(buf), 0, &kill);
3792 cgrp = cgroup_kn_lock_live(of->kn, false);
3797 * Killing is a process directed operation, i.e. the whole thread-group
3798 * is taken down so act like we do for cgroup.procs and only make this
3799 * writable in non-threaded cgroups.
3801 if (cgroup_is_threaded(cgrp))
3806 cgroup_kn_unlock(of->kn);
3808 return ret ?: nbytes;
3811 static int cgroup_file_open(struct kernfs_open_file *of)
3813 struct cftype *cft = of_cft(of);
3816 return cft->open(of);
3820 static void cgroup_file_release(struct kernfs_open_file *of)
3822 struct cftype *cft = of_cft(of);
3828 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3829 size_t nbytes, loff_t off)
3831 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
3832 struct cgroup *cgrp = of->kn->parent->priv;
3833 struct cftype *cft = of_cft(of);
3834 struct cgroup_subsys_state *css;
3841 * If namespaces are delegation boundaries, disallow writes to
3842 * files in an non-init namespace root from inside the namespace
3843 * except for the files explicitly marked delegatable -
3844 * cgroup.procs and cgroup.subtree_control.
3846 if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
3847 !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
3848 ns != &init_cgroup_ns && ns->root_cset->dfl_cgrp == cgrp)
3852 return cft->write(of, buf, nbytes, off);
3855 * kernfs guarantees that a file isn't deleted with operations in
3856 * flight, which means that the matching css is and stays alive and
3857 * doesn't need to be pinned. The RCU locking is not necessary
3858 * either. It's just for the convenience of using cgroup_css().
3861 css = cgroup_css(cgrp, cft->ss);
3864 if (cft->write_u64) {
3865 unsigned long long v;
3866 ret = kstrtoull(buf, 0, &v);
3868 ret = cft->write_u64(css, cft, v);
3869 } else if (cft->write_s64) {
3871 ret = kstrtoll(buf, 0, &v);
3873 ret = cft->write_s64(css, cft, v);
3878 return ret ?: nbytes;
3881 static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt)
3883 struct cftype *cft = of_cft(of);
3886 return cft->poll(of, pt);
3888 return kernfs_generic_poll(of, pt);
3891 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3893 return seq_cft(seq)->seq_start(seq, ppos);
3896 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3898 return seq_cft(seq)->seq_next(seq, v, ppos);
3901 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3903 if (seq_cft(seq)->seq_stop)
3904 seq_cft(seq)->seq_stop(seq, v);
3907 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3909 struct cftype *cft = seq_cft(m);
3910 struct cgroup_subsys_state *css = seq_css(m);
3913 return cft->seq_show(m, arg);
3916 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3917 else if (cft->read_s64)
3918 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3924 static struct kernfs_ops cgroup_kf_single_ops = {
3925 .atomic_write_len = PAGE_SIZE,
3926 .open = cgroup_file_open,
3927 .release = cgroup_file_release,
3928 .write = cgroup_file_write,
3929 .poll = cgroup_file_poll,
3930 .seq_show = cgroup_seqfile_show,
3933 static struct kernfs_ops cgroup_kf_ops = {
3934 .atomic_write_len = PAGE_SIZE,
3935 .open = cgroup_file_open,
3936 .release = cgroup_file_release,
3937 .write = cgroup_file_write,
3938 .poll = cgroup_file_poll,
3939 .seq_start = cgroup_seqfile_start,
3940 .seq_next = cgroup_seqfile_next,
3941 .seq_stop = cgroup_seqfile_stop,
3942 .seq_show = cgroup_seqfile_show,
3945 /* set uid and gid of cgroup dirs and files to that of the creator */
3946 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3948 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3949 .ia_uid = current_fsuid(),
3950 .ia_gid = current_fsgid(), };
3952 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3953 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3956 return kernfs_setattr(kn, &iattr);
3959 static void cgroup_file_notify_timer(struct timer_list *timer)
3961 cgroup_file_notify(container_of(timer, struct cgroup_file,
3965 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3968 char name[CGROUP_FILE_NAME_MAX];
3969 struct kernfs_node *kn;
3970 struct lock_class_key *key = NULL;
3973 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3974 key = &cft->lockdep_key;
3976 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3977 cgroup_file_mode(cft),
3978 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
3979 0, cft->kf_ops, cft,
3984 ret = cgroup_kn_set_ugid(kn);
3990 if (cft->file_offset) {
3991 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3993 timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0);
3995 spin_lock_irq(&cgroup_file_kn_lock);
3997 spin_unlock_irq(&cgroup_file_kn_lock);
4004 * cgroup_addrm_files - add or remove files to a cgroup directory
4005 * @css: the target css
4006 * @cgrp: the target cgroup (usually css->cgroup)
4007 * @cfts: array of cftypes to be added
4008 * @is_add: whether to add or remove
4010 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
4011 * For removals, this function never fails.
4013 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
4014 struct cgroup *cgrp, struct cftype cfts[],
4017 struct cftype *cft, *cft_end = NULL;
4020 lockdep_assert_held(&cgroup_mutex);
4023 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
4024 /* does cft->flags tell us to skip this file on @cgrp? */
4025 if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
4027 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
4029 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
4031 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
4033 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
4035 if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug)
4038 ret = cgroup_add_file(css, cgrp, cft);
4040 pr_warn("%s: failed to add %s, err=%d\n",
4041 __func__, cft->name, ret);
4047 cgroup_rm_file(cgrp, cft);
4053 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
4055 struct cgroup_subsys *ss = cfts[0].ss;
4056 struct cgroup *root = &ss->root->cgrp;
4057 struct cgroup_subsys_state *css;
4060 lockdep_assert_held(&cgroup_mutex);
4062 /* add/rm files for all cgroups created before */
4063 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
4064 struct cgroup *cgrp = css->cgroup;
4066 if (!(css->flags & CSS_VISIBLE))
4069 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
4075 kernfs_activate(root->kn);
4079 static void cgroup_exit_cftypes(struct cftype *cfts)
4083 for (cft = cfts; cft->name[0] != '\0'; cft++) {
4084 /* free copy for custom atomic_write_len, see init_cftypes() */
4085 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
4090 /* revert flags set by cgroup core while adding @cfts */
4091 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
4095 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4099 for (cft = cfts; cft->name[0] != '\0'; cft++) {
4100 struct kernfs_ops *kf_ops;
4102 WARN_ON(cft->ss || cft->kf_ops);
4104 if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
4108 kf_ops = &cgroup_kf_ops;
4110 kf_ops = &cgroup_kf_single_ops;
4113 * Ugh... if @cft wants a custom max_write_len, we need to
4114 * make a copy of kf_ops to set its atomic_write_len.
4116 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
4117 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
4119 cgroup_exit_cftypes(cfts);
4122 kf_ops->atomic_write_len = cft->max_write_len;
4125 cft->kf_ops = kf_ops;
4132 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
4134 lockdep_assert_held(&cgroup_mutex);
4136 if (!cfts || !cfts[0].ss)
4139 list_del(&cfts->node);
4140 cgroup_apply_cftypes(cfts, false);
4141 cgroup_exit_cftypes(cfts);
4146 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4147 * @cfts: zero-length name terminated array of cftypes
4149 * Unregister @cfts. Files described by @cfts are removed from all
4150 * existing cgroups and all future cgroups won't have them either. This
4151 * function can be called anytime whether @cfts' subsys is attached or not.
4153 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4156 int cgroup_rm_cftypes(struct cftype *cfts)
4160 mutex_lock(&cgroup_mutex);
4161 ret = cgroup_rm_cftypes_locked(cfts);
4162 mutex_unlock(&cgroup_mutex);
4167 * cgroup_add_cftypes - add an array of cftypes to a subsystem
4168 * @ss: target cgroup subsystem
4169 * @cfts: zero-length name terminated array of cftypes
4171 * Register @cfts to @ss. Files described by @cfts are created for all
4172 * existing cgroups to which @ss is attached and all future cgroups will
4173 * have them too. This function can be called anytime whether @ss is
4176 * Returns 0 on successful registration, -errno on failure. Note that this
4177 * function currently returns 0 as long as @cfts registration is successful
4178 * even if some file creation attempts on existing cgroups fail.
4180 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4184 if (!cgroup_ssid_enabled(ss->id))
4187 if (!cfts || cfts[0].name[0] == '\0')
4190 ret = cgroup_init_cftypes(ss, cfts);
4194 mutex_lock(&cgroup_mutex);
4196 list_add_tail(&cfts->node, &ss->cfts);
4197 ret = cgroup_apply_cftypes(cfts, true);
4199 cgroup_rm_cftypes_locked(cfts);
4201 mutex_unlock(&cgroup_mutex);
4206 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4207 * @ss: target cgroup subsystem
4208 * @cfts: zero-length name terminated array of cftypes
4210 * Similar to cgroup_add_cftypes() but the added files are only used for
4211 * the default hierarchy.
4213 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4217 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4218 cft->flags |= __CFTYPE_ONLY_ON_DFL;
4219 return cgroup_add_cftypes(ss, cfts);
4223 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4224 * @ss: target cgroup subsystem
4225 * @cfts: zero-length name terminated array of cftypes
4227 * Similar to cgroup_add_cftypes() but the added files are only used for
4228 * the legacy hierarchies.
4230 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4234 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4235 cft->flags |= __CFTYPE_NOT_ON_DFL;
4236 return cgroup_add_cftypes(ss, cfts);
4240 * cgroup_file_notify - generate a file modified event for a cgroup_file
4241 * @cfile: target cgroup_file
4243 * @cfile must have been obtained by setting cftype->file_offset.
4245 void cgroup_file_notify(struct cgroup_file *cfile)
4247 unsigned long flags;
4249 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
4251 unsigned long last = cfile->notified_at;
4252 unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV;
4254 if (time_in_range(jiffies, last, next)) {
4255 timer_reduce(&cfile->notify_timer, next);
4257 kernfs_notify(cfile->kn);
4258 cfile->notified_at = jiffies;
4261 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
4265 * css_next_child - find the next child of a given css
4266 * @pos: the current position (%NULL to initiate traversal)
4267 * @parent: css whose children to walk
4269 * This function returns the next child of @parent and should be called
4270 * under either cgroup_mutex or RCU read lock. The only requirement is
4271 * that @parent and @pos are accessible. The next sibling is guaranteed to
4272 * be returned regardless of their states.
4274 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4275 * css which finished ->css_online() is guaranteed to be visible in the
4276 * future iterations and will stay visible until the last reference is put.
4277 * A css which hasn't finished ->css_online() or already finished
4278 * ->css_offline() may show up during traversal. It's each subsystem's
4279 * responsibility to synchronize against on/offlining.
4281 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
4282 struct cgroup_subsys_state *parent)
4284 struct cgroup_subsys_state *next;
4286 cgroup_assert_mutex_or_rcu_locked();
4289 * @pos could already have been unlinked from the sibling list.
4290 * Once a cgroup is removed, its ->sibling.next is no longer
4291 * updated when its next sibling changes. CSS_RELEASED is set when
4292 * @pos is taken off list, at which time its next pointer is valid,
4293 * and, as releases are serialized, the one pointed to by the next
4294 * pointer is guaranteed to not have started release yet. This
4295 * implies that if we observe !CSS_RELEASED on @pos in this RCU
4296 * critical section, the one pointed to by its next pointer is
4297 * guaranteed to not have finished its RCU grace period even if we
4298 * have dropped rcu_read_lock() in-between iterations.
4300 * If @pos has CSS_RELEASED set, its next pointer can't be
4301 * dereferenced; however, as each css is given a monotonically
4302 * increasing unique serial number and always appended to the
4303 * sibling list, the next one can be found by walking the parent's
4304 * children until the first css with higher serial number than
4305 * @pos's. While this path can be slower, it happens iff iteration
4306 * races against release and the race window is very small.
4309 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
4310 } else if (likely(!(pos->flags & CSS_RELEASED))) {
4311 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
4313 list_for_each_entry_rcu(next, &parent->children, sibling,
4314 lockdep_is_held(&cgroup_mutex))
4315 if (next->serial_nr > pos->serial_nr)
4320 * @next, if not pointing to the head, can be dereferenced and is
4323 if (&next->sibling != &parent->children)
4329 * css_next_descendant_pre - find the next descendant for pre-order walk
4330 * @pos: the current position (%NULL to initiate traversal)
4331 * @root: css whose descendants to walk
4333 * To be used by css_for_each_descendant_pre(). Find the next descendant
4334 * to visit for pre-order traversal of @root's descendants. @root is
4335 * included in the iteration and the first node to be visited.
4337 * While this function requires cgroup_mutex or RCU read locking, it
4338 * doesn't require the whole traversal to be contained in a single critical
4339 * section. This function will return the correct next descendant as long
4340 * as both @pos and @root are accessible and @pos is a descendant of @root.
4342 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4343 * css which finished ->css_online() is guaranteed to be visible in the
4344 * future iterations and will stay visible until the last reference is put.
4345 * A css which hasn't finished ->css_online() or already finished
4346 * ->css_offline() may show up during traversal. It's each subsystem's
4347 * responsibility to synchronize against on/offlining.
4349 struct cgroup_subsys_state *
4350 css_next_descendant_pre(struct cgroup_subsys_state *pos,
4351 struct cgroup_subsys_state *root)
4353 struct cgroup_subsys_state *next;
4355 cgroup_assert_mutex_or_rcu_locked();
4357 /* if first iteration, visit @root */
4361 /* visit the first child if exists */
4362 next = css_next_child(NULL, pos);
4366 /* no child, visit my or the closest ancestor's next sibling */
4367 while (pos != root) {
4368 next = css_next_child(pos, pos->parent);
4376 EXPORT_SYMBOL_GPL(css_next_descendant_pre);
4379 * css_rightmost_descendant - return the rightmost descendant of a css
4380 * @pos: css of interest
4382 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4383 * is returned. This can be used during pre-order traversal to skip
4386 * While this function requires cgroup_mutex or RCU read locking, it
4387 * doesn't require the whole traversal to be contained in a single critical
4388 * section. This function will return the correct rightmost descendant as
4389 * long as @pos is accessible.
4391 struct cgroup_subsys_state *
4392 css_rightmost_descendant(struct cgroup_subsys_state *pos)
4394 struct cgroup_subsys_state *last, *tmp;
4396 cgroup_assert_mutex_or_rcu_locked();
4400 /* ->prev isn't RCU safe, walk ->next till the end */
4402 css_for_each_child(tmp, last)
4409 static struct cgroup_subsys_state *
4410 css_leftmost_descendant(struct cgroup_subsys_state *pos)
4412 struct cgroup_subsys_state *last;
4416 pos = css_next_child(NULL, pos);
4423 * css_next_descendant_post - find the next descendant for post-order walk
4424 * @pos: the current position (%NULL to initiate traversal)
4425 * @root: css whose descendants to walk
4427 * To be used by css_for_each_descendant_post(). Find the next descendant
4428 * to visit for post-order traversal of @root's descendants. @root is
4429 * included in the iteration and the last node to be visited.
4431 * While this function requires cgroup_mutex or RCU read locking, it
4432 * doesn't require the whole traversal to be contained in a single critical
4433 * section. This function will return the correct next descendant as long
4434 * as both @pos and @cgroup are accessible and @pos is a descendant of
4437 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4438 * css which finished ->css_online() is guaranteed to be visible in the
4439 * future iterations and will stay visible until the last reference is put.
4440 * A css which hasn't finished ->css_online() or already finished
4441 * ->css_offline() may show up during traversal. It's each subsystem's
4442 * responsibility to synchronize against on/offlining.
4444 struct cgroup_subsys_state *
4445 css_next_descendant_post(struct cgroup_subsys_state *pos,
4446 struct cgroup_subsys_state *root)
4448 struct cgroup_subsys_state *next;
4450 cgroup_assert_mutex_or_rcu_locked();
4452 /* if first iteration, visit leftmost descendant which may be @root */
4454 return css_leftmost_descendant(root);
4456 /* if we visited @root, we're done */
4460 /* if there's an unvisited sibling, visit its leftmost descendant */
4461 next = css_next_child(pos, pos->parent);
4463 return css_leftmost_descendant(next);
4465 /* no sibling left, visit parent */
4470 * css_has_online_children - does a css have online children
4471 * @css: the target css
4473 * Returns %true if @css has any online children; otherwise, %false. This
4474 * function can be called from any context but the caller is responsible
4475 * for synchronizing against on/offlining as necessary.
4477 bool css_has_online_children(struct cgroup_subsys_state *css)
4479 struct cgroup_subsys_state *child;
4483 css_for_each_child(child, css) {
4484 if (child->flags & CSS_ONLINE) {
4493 static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
4495 struct list_head *l;
4496 struct cgrp_cset_link *link;
4497 struct css_set *cset;
4499 lockdep_assert_held(&css_set_lock);
4501 /* find the next threaded cset */
4502 if (it->tcset_pos) {
4503 l = it->tcset_pos->next;
4505 if (l != it->tcset_head) {
4507 return container_of(l, struct css_set,
4508 threaded_csets_node);
4511 it->tcset_pos = NULL;
4514 /* find the next cset */
4517 if (l == it->cset_head) {
4518 it->cset_pos = NULL;
4523 cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
4525 link = list_entry(l, struct cgrp_cset_link, cset_link);
4531 /* initialize threaded css_set walking */
4532 if (it->flags & CSS_TASK_ITER_THREADED) {
4534 put_css_set_locked(it->cur_dcset);
4535 it->cur_dcset = cset;
4538 it->tcset_head = &cset->threaded_csets;
4539 it->tcset_pos = &cset->threaded_csets;
4546 * css_task_iter_advance_css_set - advance a task iterator to the next css_set
4547 * @it: the iterator to advance
4549 * Advance @it to the next css_set to walk.
4551 static void css_task_iter_advance_css_set(struct css_task_iter *it)
4553 struct css_set *cset;
4555 lockdep_assert_held(&css_set_lock);
4557 /* Advance to the next non-empty css_set and find first non-empty tasks list*/
4558 while ((cset = css_task_iter_next_css_set(it))) {
4559 if (!list_empty(&cset->tasks)) {
4560 it->cur_tasks_head = &cset->tasks;
4562 } else if (!list_empty(&cset->mg_tasks)) {
4563 it->cur_tasks_head = &cset->mg_tasks;
4565 } else if (!list_empty(&cset->dying_tasks)) {
4566 it->cur_tasks_head = &cset->dying_tasks;
4571 it->task_pos = NULL;
4574 it->task_pos = it->cur_tasks_head->next;
4577 * We don't keep css_sets locked across iteration steps and thus
4578 * need to take steps to ensure that iteration can be resumed after
4579 * the lock is re-acquired. Iteration is performed at two levels -
4580 * css_sets and tasks in them.
4582 * Once created, a css_set never leaves its cgroup lists, so a
4583 * pinned css_set is guaranteed to stay put and we can resume
4584 * iteration afterwards.
4586 * Tasks may leave @cset across iteration steps. This is resolved
4587 * by registering each iterator with the css_set currently being
4588 * walked and making css_set_move_task() advance iterators whose
4589 * next task is leaving.
4592 list_del(&it->iters_node);
4593 put_css_set_locked(it->cur_cset);
4596 it->cur_cset = cset;
4597 list_add(&it->iters_node, &cset->task_iters);
4600 static void css_task_iter_skip(struct css_task_iter *it,
4601 struct task_struct *task)
4603 lockdep_assert_held(&css_set_lock);
4605 if (it->task_pos == &task->cg_list) {
4606 it->task_pos = it->task_pos->next;
4607 it->flags |= CSS_TASK_ITER_SKIPPED;
4611 static void css_task_iter_advance(struct css_task_iter *it)
4613 struct task_struct *task;
4615 lockdep_assert_held(&css_set_lock);
4619 * Advance iterator to find next entry. We go through cset
4620 * tasks, mg_tasks and dying_tasks, when consumed we move onto
4623 if (it->flags & CSS_TASK_ITER_SKIPPED)
4624 it->flags &= ~CSS_TASK_ITER_SKIPPED;
4626 it->task_pos = it->task_pos->next;
4628 if (it->task_pos == &it->cur_cset->tasks) {
4629 it->cur_tasks_head = &it->cur_cset->mg_tasks;
4630 it->task_pos = it->cur_tasks_head->next;
4632 if (it->task_pos == &it->cur_cset->mg_tasks) {
4633 it->cur_tasks_head = &it->cur_cset->dying_tasks;
4634 it->task_pos = it->cur_tasks_head->next;
4636 if (it->task_pos == &it->cur_cset->dying_tasks)
4637 css_task_iter_advance_css_set(it);
4639 /* called from start, proceed to the first cset */
4640 css_task_iter_advance_css_set(it);
4646 task = list_entry(it->task_pos, struct task_struct, cg_list);
4648 if (it->flags & CSS_TASK_ITER_PROCS) {
4649 /* if PROCS, skip over tasks which aren't group leaders */
4650 if (!thread_group_leader(task))
4653 /* and dying leaders w/o live member threads */
4654 if (it->cur_tasks_head == &it->cur_cset->dying_tasks &&
4655 !atomic_read(&task->signal->live))
4658 /* skip all dying ones */
4659 if (it->cur_tasks_head == &it->cur_cset->dying_tasks)
4665 * css_task_iter_start - initiate task iteration
4666 * @css: the css to walk tasks of
4667 * @flags: CSS_TASK_ITER_* flags
4668 * @it: the task iterator to use
4670 * Initiate iteration through the tasks of @css. The caller can call
4671 * css_task_iter_next() to walk through the tasks until the function
4672 * returns NULL. On completion of iteration, css_task_iter_end() must be
4675 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
4676 struct css_task_iter *it)
4678 memset(it, 0, sizeof(*it));
4680 spin_lock_irq(&css_set_lock);
4685 if (CGROUP_HAS_SUBSYS_CONFIG && it->ss)
4686 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4688 it->cset_pos = &css->cgroup->cset_links;
4690 it->cset_head = it->cset_pos;
4692 css_task_iter_advance(it);
4694 spin_unlock_irq(&css_set_lock);
4698 * css_task_iter_next - return the next task for the iterator
4699 * @it: the task iterator being iterated
4701 * The "next" function for task iteration. @it should have been
4702 * initialized via css_task_iter_start(). Returns NULL when the iteration
4705 struct task_struct *css_task_iter_next(struct css_task_iter *it)
4708 put_task_struct(it->cur_task);
4709 it->cur_task = NULL;
4712 spin_lock_irq(&css_set_lock);
4714 /* @it may be half-advanced by skips, finish advancing */
4715 if (it->flags & CSS_TASK_ITER_SKIPPED)
4716 css_task_iter_advance(it);
4719 it->cur_task = list_entry(it->task_pos, struct task_struct,
4721 get_task_struct(it->cur_task);
4722 css_task_iter_advance(it);
4725 spin_unlock_irq(&css_set_lock);
4727 return it->cur_task;
4731 * css_task_iter_end - finish task iteration
4732 * @it: the task iterator to finish
4734 * Finish task iteration started by css_task_iter_start().
4736 void css_task_iter_end(struct css_task_iter *it)
4739 spin_lock_irq(&css_set_lock);
4740 list_del(&it->iters_node);
4741 put_css_set_locked(it->cur_cset);
4742 spin_unlock_irq(&css_set_lock);
4746 put_css_set(it->cur_dcset);
4749 put_task_struct(it->cur_task);
4752 static void cgroup_procs_release(struct kernfs_open_file *of)
4755 css_task_iter_end(of->priv);
4760 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
4762 struct kernfs_open_file *of = s->private;
4763 struct css_task_iter *it = of->priv;
4768 return css_task_iter_next(it);
4771 static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
4772 unsigned int iter_flags)
4774 struct kernfs_open_file *of = s->private;
4775 struct cgroup *cgrp = seq_css(s)->cgroup;
4776 struct css_task_iter *it = of->priv;
4779 * When a seq_file is seeked, it's always traversed sequentially
4780 * from position 0, so we can simply keep iterating on !0 *pos.
4783 if (WARN_ON_ONCE((*pos)))
4784 return ERR_PTR(-EINVAL);
4786 it = kzalloc(sizeof(*it), GFP_KERNEL);
4788 return ERR_PTR(-ENOMEM);
4790 css_task_iter_start(&cgrp->self, iter_flags, it);
4791 } else if (!(*pos)) {
4792 css_task_iter_end(it);
4793 css_task_iter_start(&cgrp->self, iter_flags, it);
4795 return it->cur_task;
4797 return cgroup_procs_next(s, NULL, NULL);
4800 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
4802 struct cgroup *cgrp = seq_css(s)->cgroup;
4805 * All processes of a threaded subtree belong to the domain cgroup
4806 * of the subtree. Only threads can be distributed across the
4807 * subtree. Reject reads on cgroup.procs in the subtree proper.
4808 * They're always empty anyway.
4810 if (cgroup_is_threaded(cgrp))
4811 return ERR_PTR(-EOPNOTSUPP);
4813 return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
4814 CSS_TASK_ITER_THREADED);
4817 static int cgroup_procs_show(struct seq_file *s, void *v)
4819 seq_printf(s, "%d\n", task_pid_vnr(v));
4823 static int cgroup_may_write(const struct cgroup *cgrp, struct super_block *sb)
4826 struct inode *inode;
4828 lockdep_assert_held(&cgroup_mutex);
4830 inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
4834 ret = inode_permission(&init_user_ns, inode, MAY_WRITE);
4839 static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
4840 struct cgroup *dst_cgrp,
4841 struct super_block *sb)
4843 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
4844 struct cgroup *com_cgrp = src_cgrp;
4847 lockdep_assert_held(&cgroup_mutex);
4849 /* find the common ancestor */
4850 while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
4851 com_cgrp = cgroup_parent(com_cgrp);
4853 /* %current should be authorized to migrate to the common ancestor */
4854 ret = cgroup_may_write(com_cgrp, sb);
4859 * If namespaces are delegation boundaries, %current must be able
4860 * to see both source and destination cgroups from its namespace.
4862 if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
4863 (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
4864 !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
4870 static int cgroup_attach_permissions(struct cgroup *src_cgrp,
4871 struct cgroup *dst_cgrp,
4872 struct super_block *sb, bool threadgroup)
4876 ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, sb);
4880 ret = cgroup_migrate_vet_dst(dst_cgrp);
4884 if (!threadgroup && (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp))
4890 static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
4893 struct cgroup *src_cgrp, *dst_cgrp;
4894 struct task_struct *task;
4895 const struct cred *saved_cred;
4899 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4903 task = cgroup_procs_write_start(buf, threadgroup, &locked);
4904 ret = PTR_ERR_OR_ZERO(task);
4908 /* find the source cgroup */
4909 spin_lock_irq(&css_set_lock);
4910 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4911 spin_unlock_irq(&css_set_lock);
4914 * Process and thread migrations follow same delegation rule. Check
4915 * permissions using the credentials from file open to protect against
4916 * inherited fd attacks.
4918 saved_cred = override_creds(of->file->f_cred);
4919 ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
4920 of->file->f_path.dentry->d_sb, threadgroup);
4921 revert_creds(saved_cred);
4925 ret = cgroup_attach_task(dst_cgrp, task, threadgroup);
4928 cgroup_procs_write_finish(task, locked);
4930 cgroup_kn_unlock(of->kn);
4935 static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
4936 char *buf, size_t nbytes, loff_t off)
4938 return __cgroup_procs_write(of, buf, true) ?: nbytes;
4941 static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
4943 return __cgroup_procs_start(s, pos, 0);
4946 static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
4947 char *buf, size_t nbytes, loff_t off)
4949 return __cgroup_procs_write(of, buf, false) ?: nbytes;
4952 /* cgroup core interface files for the default hierarchy */
4953 static struct cftype cgroup_base_files[] = {
4955 .name = "cgroup.type",
4956 .flags = CFTYPE_NOT_ON_ROOT,
4957 .seq_show = cgroup_type_show,
4958 .write = cgroup_type_write,
4961 .name = "cgroup.procs",
4962 .flags = CFTYPE_NS_DELEGATABLE,
4963 .file_offset = offsetof(struct cgroup, procs_file),
4964 .release = cgroup_procs_release,
4965 .seq_start = cgroup_procs_start,
4966 .seq_next = cgroup_procs_next,
4967 .seq_show = cgroup_procs_show,
4968 .write = cgroup_procs_write,
4971 .name = "cgroup.threads",
4972 .flags = CFTYPE_NS_DELEGATABLE,
4973 .release = cgroup_procs_release,
4974 .seq_start = cgroup_threads_start,
4975 .seq_next = cgroup_procs_next,
4976 .seq_show = cgroup_procs_show,
4977 .write = cgroup_threads_write,
4980 .name = "cgroup.controllers",
4981 .seq_show = cgroup_controllers_show,
4984 .name = "cgroup.subtree_control",
4985 .flags = CFTYPE_NS_DELEGATABLE,
4986 .seq_show = cgroup_subtree_control_show,
4987 .write = cgroup_subtree_control_write,
4990 .name = "cgroup.events",
4991 .flags = CFTYPE_NOT_ON_ROOT,
4992 .file_offset = offsetof(struct cgroup, events_file),
4993 .seq_show = cgroup_events_show,
4996 .name = "cgroup.max.descendants",
4997 .seq_show = cgroup_max_descendants_show,
4998 .write = cgroup_max_descendants_write,
5001 .name = "cgroup.max.depth",
5002 .seq_show = cgroup_max_depth_show,
5003 .write = cgroup_max_depth_write,
5006 .name = "cgroup.stat",
5007 .seq_show = cgroup_stat_show,
5010 .name = "cgroup.freeze",
5011 .flags = CFTYPE_NOT_ON_ROOT,
5012 .seq_show = cgroup_freeze_show,
5013 .write = cgroup_freeze_write,
5016 .name = "cgroup.kill",
5017 .flags = CFTYPE_NOT_ON_ROOT,
5018 .write = cgroup_kill_write,
5022 .seq_show = cpu_stat_show,
5026 .name = "io.pressure",
5027 .flags = CFTYPE_PRESSURE,
5028 .seq_show = cgroup_io_pressure_show,
5029 .write = cgroup_io_pressure_write,
5030 .poll = cgroup_pressure_poll,
5031 .release = cgroup_pressure_release,
5034 .name = "memory.pressure",
5035 .flags = CFTYPE_PRESSURE,
5036 .seq_show = cgroup_memory_pressure_show,
5037 .write = cgroup_memory_pressure_write,
5038 .poll = cgroup_pressure_poll,
5039 .release = cgroup_pressure_release,
5042 .name = "cpu.pressure",
5043 .flags = CFTYPE_PRESSURE,
5044 .seq_show = cgroup_cpu_pressure_show,
5045 .write = cgroup_cpu_pressure_write,
5046 .poll = cgroup_pressure_poll,
5047 .release = cgroup_pressure_release,
5049 #endif /* CONFIG_PSI */
5054 * css destruction is four-stage process.
5056 * 1. Destruction starts. Killing of the percpu_ref is initiated.
5057 * Implemented in kill_css().
5059 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
5060 * and thus css_tryget_online() is guaranteed to fail, the css can be
5061 * offlined by invoking offline_css(). After offlining, the base ref is
5062 * put. Implemented in css_killed_work_fn().
5064 * 3. When the percpu_ref reaches zero, the only possible remaining
5065 * accessors are inside RCU read sections. css_release() schedules the
5068 * 4. After the grace period, the css can be freed. Implemented in
5069 * css_free_work_fn().
5071 * It is actually hairier because both step 2 and 4 require process context
5072 * and thus involve punting to css->destroy_work adding two additional
5073 * steps to the already complex sequence.
5075 static void css_free_rwork_fn(struct work_struct *work)
5077 struct cgroup_subsys_state *css = container_of(to_rcu_work(work),
5078 struct cgroup_subsys_state, destroy_rwork);
5079 struct cgroup_subsys *ss = css->ss;
5080 struct cgroup *cgrp = css->cgroup;
5082 percpu_ref_exit(&css->refcnt);
5086 struct cgroup_subsys_state *parent = css->parent;
5090 cgroup_idr_remove(&ss->css_idr, id);
5096 /* cgroup free path */
5097 atomic_dec(&cgrp->root->nr_cgrps);
5098 cgroup1_pidlist_destroy_all(cgrp);
5099 cancel_work_sync(&cgrp->release_agent_work);
5101 if (cgroup_parent(cgrp)) {
5103 * We get a ref to the parent, and put the ref when
5104 * this cgroup is being freed, so it's guaranteed
5105 * that the parent won't be destroyed before its
5108 cgroup_put(cgroup_parent(cgrp));
5109 kernfs_put(cgrp->kn);
5110 psi_cgroup_free(cgrp);
5111 cgroup_rstat_exit(cgrp);
5115 * This is root cgroup's refcnt reaching zero,
5116 * which indicates that the root should be
5119 cgroup_destroy_root(cgrp->root);
5124 static void css_release_work_fn(struct work_struct *work)
5126 struct cgroup_subsys_state *css =
5127 container_of(work, struct cgroup_subsys_state, destroy_work);
5128 struct cgroup_subsys *ss = css->ss;
5129 struct cgroup *cgrp = css->cgroup;
5131 mutex_lock(&cgroup_mutex);
5133 css->flags |= CSS_RELEASED;
5134 list_del_rcu(&css->sibling);
5137 /* css release path */
5138 if (!list_empty(&css->rstat_css_node)) {
5139 cgroup_rstat_flush(cgrp);
5140 list_del_rcu(&css->rstat_css_node);
5143 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
5144 if (ss->css_released)
5145 ss->css_released(css);
5147 struct cgroup *tcgrp;
5149 /* cgroup release path */
5150 TRACE_CGROUP_PATH(release, cgrp);
5152 cgroup_rstat_flush(cgrp);
5154 spin_lock_irq(&css_set_lock);
5155 for (tcgrp = cgroup_parent(cgrp); tcgrp;
5156 tcgrp = cgroup_parent(tcgrp))
5157 tcgrp->nr_dying_descendants--;
5158 spin_unlock_irq(&css_set_lock);
5161 * There are two control paths which try to determine
5162 * cgroup from dentry without going through kernfs -
5163 * cgroupstats_build() and css_tryget_online_from_dir().
5164 * Those are supported by RCU protecting clearing of
5165 * cgrp->kn->priv backpointer.
5168 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
5172 mutex_unlock(&cgroup_mutex);
5174 INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5175 queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5178 static void css_release(struct percpu_ref *ref)
5180 struct cgroup_subsys_state *css =
5181 container_of(ref, struct cgroup_subsys_state, refcnt);
5183 INIT_WORK(&css->destroy_work, css_release_work_fn);
5184 queue_work(cgroup_destroy_wq, &css->destroy_work);
5187 static void init_and_link_css(struct cgroup_subsys_state *css,
5188 struct cgroup_subsys *ss, struct cgroup *cgrp)
5190 lockdep_assert_held(&cgroup_mutex);
5192 cgroup_get_live(cgrp);
5194 memset(css, 0, sizeof(*css));
5198 INIT_LIST_HEAD(&css->sibling);
5199 INIT_LIST_HEAD(&css->children);
5200 INIT_LIST_HEAD(&css->rstat_css_node);
5201 css->serial_nr = css_serial_nr_next++;
5202 atomic_set(&css->online_cnt, 0);
5204 if (cgroup_parent(cgrp)) {
5205 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
5206 css_get(css->parent);
5209 if (ss->css_rstat_flush)
5210 list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list);
5212 BUG_ON(cgroup_css(cgrp, ss));
5215 /* invoke ->css_online() on a new CSS and mark it online if successful */
5216 static int online_css(struct cgroup_subsys_state *css)
5218 struct cgroup_subsys *ss = css->ss;
5221 lockdep_assert_held(&cgroup_mutex);
5224 ret = ss->css_online(css);
5226 css->flags |= CSS_ONLINE;
5227 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
5229 atomic_inc(&css->online_cnt);
5231 atomic_inc(&css->parent->online_cnt);
5236 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5237 static void offline_css(struct cgroup_subsys_state *css)
5239 struct cgroup_subsys *ss = css->ss;
5241 lockdep_assert_held(&cgroup_mutex);
5243 if (!(css->flags & CSS_ONLINE))
5246 if (ss->css_offline)
5247 ss->css_offline(css);
5249 css->flags &= ~CSS_ONLINE;
5250 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
5252 wake_up_all(&css->cgroup->offline_waitq);
5256 * css_create - create a cgroup_subsys_state
5257 * @cgrp: the cgroup new css will be associated with
5258 * @ss: the subsys of new css
5260 * Create a new css associated with @cgrp - @ss pair. On success, the new
5261 * css is online and installed in @cgrp. This function doesn't create the
5262 * interface files. Returns 0 on success, -errno on failure.
5264 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5265 struct cgroup_subsys *ss)
5267 struct cgroup *parent = cgroup_parent(cgrp);
5268 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
5269 struct cgroup_subsys_state *css;
5272 lockdep_assert_held(&cgroup_mutex);
5274 css = ss->css_alloc(parent_css);
5276 css = ERR_PTR(-ENOMEM);
5280 init_and_link_css(css, ss, cgrp);
5282 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
5286 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
5291 /* @css is ready to be brought online now, make it visible */
5292 list_add_tail_rcu(&css->sibling, &parent_css->children);
5293 cgroup_idr_replace(&ss->css_idr, css, css->id);
5295 err = online_css(css);
5302 list_del_rcu(&css->sibling);
5304 list_del_rcu(&css->rstat_css_node);
5305 INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5306 queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5307 return ERR_PTR(err);
5311 * The returned cgroup is fully initialized including its control mask, but
5312 * it isn't associated with its kernfs_node and doesn't have the control
5315 static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
5318 struct cgroup_root *root = parent->root;
5319 struct cgroup *cgrp, *tcgrp;
5320 struct kernfs_node *kn;
5321 int level = parent->level + 1;
5324 /* allocate the cgroup and its ID, 0 is reserved for the root */
5325 cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
5328 return ERR_PTR(-ENOMEM);
5330 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
5334 ret = cgroup_rstat_init(cgrp);
5336 goto out_cancel_ref;
5338 /* create the directory */
5339 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5346 init_cgroup_housekeeping(cgrp);
5348 cgrp->self.parent = &parent->self;
5350 cgrp->level = level;
5352 ret = psi_cgroup_alloc(cgrp);
5354 goto out_kernfs_remove;
5356 ret = cgroup_bpf_inherit(cgrp);
5361 * New cgroup inherits effective freeze counter, and
5362 * if the parent has to be frozen, the child has too.
5364 cgrp->freezer.e_freeze = parent->freezer.e_freeze;
5365 if (cgrp->freezer.e_freeze) {
5367 * Set the CGRP_FREEZE flag, so when a process will be
5368 * attached to the child cgroup, it will become frozen.
5369 * At this point the new cgroup is unpopulated, so we can
5370 * consider it frozen immediately.
5372 set_bit(CGRP_FREEZE, &cgrp->flags);
5373 set_bit(CGRP_FROZEN, &cgrp->flags);
5376 spin_lock_irq(&css_set_lock);
5377 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5378 cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
5380 if (tcgrp != cgrp) {
5381 tcgrp->nr_descendants++;
5384 * If the new cgroup is frozen, all ancestor cgroups
5385 * get a new frozen descendant, but their state can't
5386 * change because of this.
5388 if (cgrp->freezer.e_freeze)
5389 tcgrp->freezer.nr_frozen_descendants++;
5392 spin_unlock_irq(&css_set_lock);
5394 if (notify_on_release(parent))
5395 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5397 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5398 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
5400 cgrp->self.serial_nr = css_serial_nr_next++;
5402 /* allocation complete, commit to creation */
5403 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
5404 atomic_inc(&root->nr_cgrps);
5405 cgroup_get_live(parent);
5408 * On the default hierarchy, a child doesn't automatically inherit
5409 * subtree_control from the parent. Each is configured manually.
5411 if (!cgroup_on_dfl(cgrp))
5412 cgrp->subtree_control = cgroup_control(cgrp);
5414 cgroup_propagate_control(cgrp);
5419 psi_cgroup_free(cgrp);
5421 kernfs_remove(cgrp->kn);
5423 cgroup_rstat_exit(cgrp);
5425 percpu_ref_exit(&cgrp->self.refcnt);
5428 return ERR_PTR(ret);
5431 static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
5433 struct cgroup *cgroup;
5437 lockdep_assert_held(&cgroup_mutex);
5439 for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
5440 if (cgroup->nr_descendants >= cgroup->max_descendants)
5443 if (level > cgroup->max_depth)
5454 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
5456 struct cgroup *parent, *cgrp;
5459 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5460 if (strchr(name, '\n'))
5463 parent = cgroup_kn_lock_live(parent_kn, false);
5467 if (!cgroup_check_hierarchy_limits(parent)) {
5472 cgrp = cgroup_create(parent, name, mode);
5474 ret = PTR_ERR(cgrp);
5479 * This extra ref will be put in cgroup_free_fn() and guarantees
5480 * that @cgrp->kn is always accessible.
5482 kernfs_get(cgrp->kn);
5484 ret = cgroup_kn_set_ugid(cgrp->kn);
5488 ret = css_populate_dir(&cgrp->self);
5492 ret = cgroup_apply_control_enable(cgrp);
5496 TRACE_CGROUP_PATH(mkdir, cgrp);
5498 /* let's create and online css's */
5499 kernfs_activate(cgrp->kn);
5505 cgroup_destroy_locked(cgrp);
5507 cgroup_kn_unlock(parent_kn);
5512 * This is called when the refcnt of a css is confirmed to be killed.
5513 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5514 * initiate destruction and put the css ref from kill_css().
5516 static void css_killed_work_fn(struct work_struct *work)
5518 struct cgroup_subsys_state *css =
5519 container_of(work, struct cgroup_subsys_state, destroy_work);
5521 mutex_lock(&cgroup_mutex);
5526 /* @css can't go away while we're holding cgroup_mutex */
5528 } while (css && atomic_dec_and_test(&css->online_cnt));
5530 mutex_unlock(&cgroup_mutex);
5533 /* css kill confirmation processing requires process context, bounce */
5534 static void css_killed_ref_fn(struct percpu_ref *ref)
5536 struct cgroup_subsys_state *css =
5537 container_of(ref, struct cgroup_subsys_state, refcnt);
5539 if (atomic_dec_and_test(&css->online_cnt)) {
5540 INIT_WORK(&css->destroy_work, css_killed_work_fn);
5541 queue_work(cgroup_destroy_wq, &css->destroy_work);
5546 * kill_css - destroy a css
5547 * @css: css to destroy
5549 * This function initiates destruction of @css by removing cgroup interface
5550 * files and putting its base reference. ->css_offline() will be invoked
5551 * asynchronously once css_tryget_online() is guaranteed to fail and when
5552 * the reference count reaches zero, @css will be released.
5554 static void kill_css(struct cgroup_subsys_state *css)
5556 lockdep_assert_held(&cgroup_mutex);
5558 if (css->flags & CSS_DYING)
5561 css->flags |= CSS_DYING;
5564 * This must happen before css is disassociated with its cgroup.
5565 * See seq_css() for details.
5570 * Killing would put the base ref, but we need to keep it alive
5571 * until after ->css_offline().
5576 * cgroup core guarantees that, by the time ->css_offline() is
5577 * invoked, no new css reference will be given out via
5578 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5579 * proceed to offlining css's because percpu_ref_kill() doesn't
5580 * guarantee that the ref is seen as killed on all CPUs on return.
5582 * Use percpu_ref_kill_and_confirm() to get notifications as each
5583 * css is confirmed to be seen as killed on all CPUs.
5585 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5589 * cgroup_destroy_locked - the first stage of cgroup destruction
5590 * @cgrp: cgroup to be destroyed
5592 * css's make use of percpu refcnts whose killing latency shouldn't be
5593 * exposed to userland and are RCU protected. Also, cgroup core needs to
5594 * guarantee that css_tryget_online() won't succeed by the time
5595 * ->css_offline() is invoked. To satisfy all the requirements,
5596 * destruction is implemented in the following two steps.
5598 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5599 * userland visible parts and start killing the percpu refcnts of
5600 * css's. Set up so that the next stage will be kicked off once all
5601 * the percpu refcnts are confirmed to be killed.
5603 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5604 * rest of destruction. Once all cgroup references are gone, the
5605 * cgroup is RCU-freed.
5607 * This function implements s1. After this step, @cgrp is gone as far as
5608 * the userland is concerned and a new cgroup with the same name may be
5609 * created. As cgroup doesn't care about the names internally, this
5610 * doesn't cause any problem.
5612 static int cgroup_destroy_locked(struct cgroup *cgrp)
5613 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5615 struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
5616 struct cgroup_subsys_state *css;
5617 struct cgrp_cset_link *link;
5620 lockdep_assert_held(&cgroup_mutex);
5623 * Only migration can raise populated from zero and we're already
5624 * holding cgroup_mutex.
5626 if (cgroup_is_populated(cgrp))
5630 * Make sure there's no live children. We can't test emptiness of
5631 * ->self.children as dead children linger on it while being
5632 * drained; otherwise, "rmdir parent/child parent" may fail.
5634 if (css_has_online_children(&cgrp->self))
5638 * Mark @cgrp and the associated csets dead. The former prevents
5639 * further task migration and child creation by disabling
5640 * cgroup_lock_live_group(). The latter makes the csets ignored by
5641 * the migration path.
5643 cgrp->self.flags &= ~CSS_ONLINE;
5645 spin_lock_irq(&css_set_lock);
5646 list_for_each_entry(link, &cgrp->cset_links, cset_link)
5647 link->cset->dead = true;
5648 spin_unlock_irq(&css_set_lock);
5650 /* initiate massacre of all css's */
5651 for_each_css(css, ssid, cgrp)
5654 /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5655 css_clear_dir(&cgrp->self);
5656 kernfs_remove(cgrp->kn);
5658 if (parent && cgroup_is_threaded(cgrp))
5659 parent->nr_threaded_children--;
5661 spin_lock_irq(&css_set_lock);
5662 for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5663 tcgrp->nr_descendants--;
5664 tcgrp->nr_dying_descendants++;
5666 * If the dying cgroup is frozen, decrease frozen descendants
5667 * counters of ancestor cgroups.
5669 if (test_bit(CGRP_FROZEN, &cgrp->flags))
5670 tcgrp->freezer.nr_frozen_descendants--;
5672 spin_unlock_irq(&css_set_lock);
5674 cgroup1_check_for_release(parent);
5676 cgroup_bpf_offline(cgrp);
5678 /* put the base reference */
5679 percpu_ref_kill(&cgrp->self.refcnt);
5684 int cgroup_rmdir(struct kernfs_node *kn)
5686 struct cgroup *cgrp;
5689 cgrp = cgroup_kn_lock_live(kn, false);
5693 ret = cgroup_destroy_locked(cgrp);
5695 TRACE_CGROUP_PATH(rmdir, cgrp);
5697 cgroup_kn_unlock(kn);
5701 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5702 .show_options = cgroup_show_options,
5703 .mkdir = cgroup_mkdir,
5704 .rmdir = cgroup_rmdir,
5705 .show_path = cgroup_show_path,
5708 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5710 struct cgroup_subsys_state *css;
5712 pr_debug("Initializing cgroup subsys %s\n", ss->name);
5714 mutex_lock(&cgroup_mutex);
5716 idr_init(&ss->css_idr);
5717 INIT_LIST_HEAD(&ss->cfts);
5719 /* Create the root cgroup state for this subsystem */
5720 ss->root = &cgrp_dfl_root;
5721 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5722 /* We don't handle early failures gracefully */
5723 BUG_ON(IS_ERR(css));
5724 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5727 * Root csses are never destroyed and we can't initialize
5728 * percpu_ref during early init. Disable refcnting.
5730 css->flags |= CSS_NO_REF;
5733 /* allocation can't be done safely during early init */
5736 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5737 BUG_ON(css->id < 0);
5740 /* Update the init_css_set to contain a subsys
5741 * pointer to this state - since the subsystem is
5742 * newly registered, all tasks and hence the
5743 * init_css_set is in the subsystem's root cgroup. */
5744 init_css_set.subsys[ss->id] = css;
5746 have_fork_callback |= (bool)ss->fork << ss->id;
5747 have_exit_callback |= (bool)ss->exit << ss->id;
5748 have_release_callback |= (bool)ss->release << ss->id;
5749 have_canfork_callback |= (bool)ss->can_fork << ss->id;
5751 /* At system boot, before all subsystems have been
5752 * registered, no tasks have been forked, so we don't
5753 * need to invoke fork callbacks here. */
5754 BUG_ON(!list_empty(&init_task.tasks));
5756 BUG_ON(online_css(css));
5758 mutex_unlock(&cgroup_mutex);
5762 * cgroup_init_early - cgroup initialization at system boot
5764 * Initialize cgroups at system boot, and initialize any
5765 * subsystems that request early init.
5767 int __init cgroup_init_early(void)
5769 static struct cgroup_fs_context __initdata ctx;
5770 struct cgroup_subsys *ss;
5773 ctx.root = &cgrp_dfl_root;
5774 init_cgroup_root(&ctx);
5775 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5777 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5779 for_each_subsys(ss, i) {
5780 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5781 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5782 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5784 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5785 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5788 ss->name = cgroup_subsys_name[i];
5789 if (!ss->legacy_name)
5790 ss->legacy_name = cgroup_subsys_name[i];
5793 cgroup_init_subsys(ss, true);
5799 * cgroup_init - cgroup initialization
5801 * Register cgroup filesystem and /proc file, and initialize
5802 * any subsystems that didn't request early init.
5804 int __init cgroup_init(void)
5806 struct cgroup_subsys *ss;
5809 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5810 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
5811 BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
5813 cgroup_rstat_boot();
5816 * The latency of the synchronize_rcu() is too high for cgroups,
5817 * avoid it at the cost of forcing all readers into the slow path.
5819 rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5821 get_user_ns(init_cgroup_ns.user_ns);
5823 mutex_lock(&cgroup_mutex);
5826 * Add init_css_set to the hash table so that dfl_root can link to
5829 hash_add(css_set_table, &init_css_set.hlist,
5830 css_set_hash(init_css_set.subsys));
5832 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
5834 mutex_unlock(&cgroup_mutex);
5836 for_each_subsys(ss, ssid) {
5837 if (ss->early_init) {
5838 struct cgroup_subsys_state *css =
5839 init_css_set.subsys[ss->id];
5841 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5843 BUG_ON(css->id < 0);
5845 cgroup_init_subsys(ss, false);
5848 list_add_tail(&init_css_set.e_cset_node[ssid],
5849 &cgrp_dfl_root.cgrp.e_csets[ssid]);
5852 * Setting dfl_root subsys_mask needs to consider the
5853 * disabled flag and cftype registration needs kmalloc,
5854 * both of which aren't available during early_init.
5856 if (!cgroup_ssid_enabled(ssid))
5859 if (cgroup1_ssid_disabled(ssid))
5860 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5863 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5865 /* implicit controllers must be threaded too */
5866 WARN_ON(ss->implicit_on_dfl && !ss->threaded);
5868 if (ss->implicit_on_dfl)
5869 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5870 else if (!ss->dfl_cftypes)
5871 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5874 cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
5876 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5877 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5879 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5880 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5884 ss->bind(init_css_set.subsys[ssid]);
5886 mutex_lock(&cgroup_mutex);
5887 css_populate_dir(init_css_set.subsys[ssid]);
5888 mutex_unlock(&cgroup_mutex);
5891 /* init_css_set.subsys[] has been updated, re-hash */
5892 hash_del(&init_css_set.hlist);
5893 hash_add(css_set_table, &init_css_set.hlist,
5894 css_set_hash(init_css_set.subsys));
5896 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5897 WARN_ON(register_filesystem(&cgroup_fs_type));
5898 WARN_ON(register_filesystem(&cgroup2_fs_type));
5899 WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show));
5900 #ifdef CONFIG_CPUSETS
5901 WARN_ON(register_filesystem(&cpuset_fs_type));
5907 static int __init cgroup_wq_init(void)
5910 * There isn't much point in executing destruction path in
5911 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5912 * Use 1 for @max_active.
5914 * We would prefer to do this in cgroup_init() above, but that
5915 * is called before init_workqueues(): so leave this until after.
5917 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5918 BUG_ON(!cgroup_destroy_wq);
5921 core_initcall(cgroup_wq_init);
5923 void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen)
5925 struct kernfs_node *kn;
5927 kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id);
5930 kernfs_path(kn, buf, buflen);
5935 * cgroup_get_from_id : get the cgroup associated with cgroup id
5937 * On success return the cgrp, on failure return NULL
5939 struct cgroup *cgroup_get_from_id(u64 id)
5941 struct kernfs_node *kn;
5942 struct cgroup *cgrp = NULL;
5944 kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id);
5950 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
5951 if (cgrp && !cgroup_tryget(cgrp))
5960 EXPORT_SYMBOL_GPL(cgroup_get_from_id);
5963 * proc_cgroup_show()
5964 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5965 * - Used for /proc/<pid>/cgroup.
5967 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5968 struct pid *pid, struct task_struct *tsk)
5972 struct cgroup_root *root;
5975 buf = kmalloc(PATH_MAX, GFP_KERNEL);
5979 mutex_lock(&cgroup_mutex);
5980 spin_lock_irq(&css_set_lock);
5982 for_each_root(root) {
5983 struct cgroup_subsys *ss;
5984 struct cgroup *cgrp;
5985 int ssid, count = 0;
5987 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5990 seq_printf(m, "%d:", root->hierarchy_id);
5991 if (root != &cgrp_dfl_root)
5992 for_each_subsys(ss, ssid)
5993 if (root->subsys_mask & (1 << ssid))
5994 seq_printf(m, "%s%s", count++ ? "," : "",
5996 if (strlen(root->name))
5997 seq_printf(m, "%sname=%s", count ? "," : "",
6001 cgrp = task_cgroup_from_root(tsk, root);
6004 * On traditional hierarchies, all zombie tasks show up as
6005 * belonging to the root cgroup. On the default hierarchy,
6006 * while a zombie doesn't show up in "cgroup.procs" and
6007 * thus can't be migrated, its /proc/PID/cgroup keeps
6008 * reporting the cgroup it belonged to before exiting. If
6009 * the cgroup is removed before the zombie is reaped,
6010 * " (deleted)" is appended to the cgroup path.
6012 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
6013 retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
6014 current->nsproxy->cgroup_ns);
6015 if (retval >= PATH_MAX)
6016 retval = -ENAMETOOLONG;
6025 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
6026 seq_puts(m, " (deleted)\n");
6033 spin_unlock_irq(&css_set_lock);
6034 mutex_unlock(&cgroup_mutex);
6041 * cgroup_fork - initialize cgroup related fields during copy_process()
6042 * @child: pointer to task_struct of forking parent process.
6044 * A task is associated with the init_css_set until cgroup_post_fork()
6045 * attaches it to the target css_set.
6047 void cgroup_fork(struct task_struct *child)
6049 RCU_INIT_POINTER(child->cgroups, &init_css_set);
6050 INIT_LIST_HEAD(&child->cg_list);
6053 static struct cgroup *cgroup_get_from_file(struct file *f)
6055 struct cgroup_subsys_state *css;
6056 struct cgroup *cgrp;
6058 css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
6060 return ERR_CAST(css);
6063 if (!cgroup_on_dfl(cgrp)) {
6065 return ERR_PTR(-EBADF);
6072 * cgroup_css_set_fork - find or create a css_set for a child process
6073 * @kargs: the arguments passed to create the child process
6075 * This functions finds or creates a new css_set which the child
6076 * process will be attached to in cgroup_post_fork(). By default,
6077 * the child process will be given the same css_set as its parent.
6079 * If CLONE_INTO_CGROUP is specified this function will try to find an
6080 * existing css_set which includes the requested cgroup and if not create
6081 * a new css_set that the child will be attached to later. If this function
6082 * succeeds it will hold cgroup_threadgroup_rwsem on return. If
6083 * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
6084 * before grabbing cgroup_threadgroup_rwsem and will hold a reference
6085 * to the target cgroup.
6087 static int cgroup_css_set_fork(struct kernel_clone_args *kargs)
6088 __acquires(&cgroup_mutex) __acquires(&cgroup_threadgroup_rwsem)
6091 struct cgroup *dst_cgrp = NULL;
6092 struct css_set *cset;
6093 struct super_block *sb;
6096 if (kargs->flags & CLONE_INTO_CGROUP)
6097 mutex_lock(&cgroup_mutex);
6099 cgroup_threadgroup_change_begin(current);
6101 spin_lock_irq(&css_set_lock);
6102 cset = task_css_set(current);
6104 spin_unlock_irq(&css_set_lock);
6106 if (!(kargs->flags & CLONE_INTO_CGROUP)) {
6111 f = fget_raw(kargs->cgroup);
6116 sb = f->f_path.dentry->d_sb;
6118 dst_cgrp = cgroup_get_from_file(f);
6119 if (IS_ERR(dst_cgrp)) {
6120 ret = PTR_ERR(dst_cgrp);
6125 if (cgroup_is_dead(dst_cgrp)) {
6131 * Verify that we the target cgroup is writable for us. This is
6132 * usually done by the vfs layer but since we're not going through
6133 * the vfs layer here we need to do it "manually".
6135 ret = cgroup_may_write(dst_cgrp, sb);
6139 ret = cgroup_attach_permissions(cset->dfl_cgrp, dst_cgrp, sb,
6140 !(kargs->flags & CLONE_THREAD));
6144 kargs->cset = find_css_set(cset, dst_cgrp);
6152 kargs->cgrp = dst_cgrp;
6156 cgroup_threadgroup_change_end(current);
6157 mutex_unlock(&cgroup_mutex);
6161 cgroup_put(dst_cgrp);
6164 put_css_set(kargs->cset);
6169 * cgroup_css_set_put_fork - drop references we took during fork
6170 * @kargs: the arguments passed to create the child process
6172 * Drop references to the prepared css_set and target cgroup if
6173 * CLONE_INTO_CGROUP was requested.
6175 static void cgroup_css_set_put_fork(struct kernel_clone_args *kargs)
6176 __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6178 cgroup_threadgroup_change_end(current);
6180 if (kargs->flags & CLONE_INTO_CGROUP) {
6181 struct cgroup *cgrp = kargs->cgrp;
6182 struct css_set *cset = kargs->cset;
6184 mutex_unlock(&cgroup_mutex);
6199 * cgroup_can_fork - called on a new task before the process is exposed
6200 * @child: the child process
6202 * This prepares a new css_set for the child process which the child will
6203 * be attached to in cgroup_post_fork().
6204 * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
6205 * callback returns an error, the fork aborts with that error code. This
6206 * allows for a cgroup subsystem to conditionally allow or deny new forks.
6208 int cgroup_can_fork(struct task_struct *child, struct kernel_clone_args *kargs)
6210 struct cgroup_subsys *ss;
6213 ret = cgroup_css_set_fork(kargs);
6217 do_each_subsys_mask(ss, i, have_canfork_callback) {
6218 ret = ss->can_fork(child, kargs->cset);
6221 } while_each_subsys_mask();
6226 for_each_subsys(ss, j) {
6229 if (ss->cancel_fork)
6230 ss->cancel_fork(child, kargs->cset);
6233 cgroup_css_set_put_fork(kargs);
6239 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6240 * @child: the child process
6241 * @kargs: the arguments passed to create the child process
6243 * This calls the cancel_fork() callbacks if a fork failed *after*
6244 * cgroup_can_fork() succeeded and cleans up references we took to
6245 * prepare a new css_set for the child process in cgroup_can_fork().
6247 void cgroup_cancel_fork(struct task_struct *child,
6248 struct kernel_clone_args *kargs)
6250 struct cgroup_subsys *ss;
6253 for_each_subsys(ss, i)
6254 if (ss->cancel_fork)
6255 ss->cancel_fork(child, kargs->cset);
6257 cgroup_css_set_put_fork(kargs);
6261 * cgroup_post_fork - finalize cgroup setup for the child process
6262 * @child: the child process
6264 * Attach the child process to its css_set calling the subsystem fork()
6267 void cgroup_post_fork(struct task_struct *child,
6268 struct kernel_clone_args *kargs)
6269 __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6271 unsigned long cgrp_flags = 0;
6273 struct cgroup_subsys *ss;
6274 struct css_set *cset;
6280 spin_lock_irq(&css_set_lock);
6282 /* init tasks are special, only link regular threads */
6283 if (likely(child->pid)) {
6285 cgrp_flags = kargs->cgrp->flags;
6287 cgrp_flags = cset->dfl_cgrp->flags;
6289 WARN_ON_ONCE(!list_empty(&child->cg_list));
6291 css_set_move_task(child, NULL, cset, false);
6297 if (!(child->flags & PF_KTHREAD)) {
6298 if (unlikely(test_bit(CGRP_FREEZE, &cgrp_flags))) {
6300 * If the cgroup has to be frozen, the new task has
6301 * too. Let's set the JOBCTL_TRAP_FREEZE jobctl bit to
6302 * get the task into the frozen state.
6304 spin_lock(&child->sighand->siglock);
6305 WARN_ON_ONCE(child->frozen);
6306 child->jobctl |= JOBCTL_TRAP_FREEZE;
6307 spin_unlock(&child->sighand->siglock);
6310 * Calling cgroup_update_frozen() isn't required here,
6311 * because it will be called anyway a bit later from
6312 * do_freezer_trap(). So we avoid cgroup's transient
6313 * switch from the frozen state and back.
6318 * If the cgroup is to be killed notice it now and take the
6319 * child down right after we finished preparing it for
6322 kill = test_bit(CGRP_KILL, &cgrp_flags);
6325 spin_unlock_irq(&css_set_lock);
6328 * Call ss->fork(). This must happen after @child is linked on
6329 * css_set; otherwise, @child might change state between ->fork()
6330 * and addition to css_set.
6332 do_each_subsys_mask(ss, i, have_fork_callback) {
6334 } while_each_subsys_mask();
6336 /* Make the new cset the root_cset of the new cgroup namespace. */
6337 if (kargs->flags & CLONE_NEWCGROUP) {
6338 struct css_set *rcset = child->nsproxy->cgroup_ns->root_cset;
6341 child->nsproxy->cgroup_ns->root_cset = cset;
6345 /* Cgroup has to be killed so take down child immediately. */
6347 do_send_sig_info(SIGKILL, SEND_SIG_NOINFO, child, PIDTYPE_TGID);
6349 cgroup_css_set_put_fork(kargs);
6353 * cgroup_exit - detach cgroup from exiting task
6354 * @tsk: pointer to task_struct of exiting process
6356 * Description: Detach cgroup from @tsk.
6359 void cgroup_exit(struct task_struct *tsk)
6361 struct cgroup_subsys *ss;
6362 struct css_set *cset;
6365 spin_lock_irq(&css_set_lock);
6367 WARN_ON_ONCE(list_empty(&tsk->cg_list));
6368 cset = task_css_set(tsk);
6369 css_set_move_task(tsk, cset, NULL, false);
6370 list_add_tail(&tsk->cg_list, &cset->dying_tasks);
6373 WARN_ON_ONCE(cgroup_task_frozen(tsk));
6374 if (unlikely(!(tsk->flags & PF_KTHREAD) &&
6375 test_bit(CGRP_FREEZE, &task_dfl_cgroup(tsk)->flags)))
6376 cgroup_update_frozen(task_dfl_cgroup(tsk));
6378 spin_unlock_irq(&css_set_lock);
6380 /* see cgroup_post_fork() for details */
6381 do_each_subsys_mask(ss, i, have_exit_callback) {
6383 } while_each_subsys_mask();
6386 void cgroup_release(struct task_struct *task)
6388 struct cgroup_subsys *ss;
6391 do_each_subsys_mask(ss, ssid, have_release_callback) {
6393 } while_each_subsys_mask();
6395 spin_lock_irq(&css_set_lock);
6396 css_set_skip_task_iters(task_css_set(task), task);
6397 list_del_init(&task->cg_list);
6398 spin_unlock_irq(&css_set_lock);
6401 void cgroup_free(struct task_struct *task)
6403 struct css_set *cset = task_css_set(task);
6407 static int __init cgroup_disable(char *str)
6409 struct cgroup_subsys *ss;
6413 while ((token = strsep(&str, ",")) != NULL) {
6417 for_each_subsys(ss, i) {
6418 if (strcmp(token, ss->name) &&
6419 strcmp(token, ss->legacy_name))
6422 static_branch_disable(cgroup_subsys_enabled_key[i]);
6423 pr_info("Disabling %s control group subsystem\n",
6427 for (i = 0; i < OPT_FEATURE_COUNT; i++) {
6428 if (strcmp(token, cgroup_opt_feature_names[i]))
6430 cgroup_feature_disable_mask |= 1 << i;
6431 pr_info("Disabling %s control group feature\n",
6432 cgroup_opt_feature_names[i]);
6438 __setup("cgroup_disable=", cgroup_disable);
6440 void __init __weak enable_debug_cgroup(void) { }
6442 static int __init enable_cgroup_debug(char *str)
6444 cgroup_debug = true;
6445 enable_debug_cgroup();
6448 __setup("cgroup_debug", enable_cgroup_debug);
6451 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6452 * @dentry: directory dentry of interest
6453 * @ss: subsystem of interest
6455 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6456 * to get the corresponding css and return it. If such css doesn't exist
6457 * or can't be pinned, an ERR_PTR value is returned.
6459 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
6460 struct cgroup_subsys *ss)
6462 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
6463 struct file_system_type *s_type = dentry->d_sb->s_type;
6464 struct cgroup_subsys_state *css = NULL;
6465 struct cgroup *cgrp;
6467 /* is @dentry a cgroup dir? */
6468 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
6469 !kn || kernfs_type(kn) != KERNFS_DIR)
6470 return ERR_PTR(-EBADF);
6475 * This path doesn't originate from kernfs and @kn could already
6476 * have been or be removed at any point. @kn->priv is RCU
6477 * protected for this access. See css_release_work_fn() for details.
6479 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6481 css = cgroup_css(cgrp, ss);
6483 if (!css || !css_tryget_online(css))
6484 css = ERR_PTR(-ENOENT);
6491 * css_from_id - lookup css by id
6492 * @id: the cgroup id
6493 * @ss: cgroup subsys to be looked into
6495 * Returns the css if there's valid one with @id, otherwise returns NULL.
6496 * Should be called under rcu_read_lock().
6498 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
6500 WARN_ON_ONCE(!rcu_read_lock_held());
6501 return idr_find(&ss->css_idr, id);
6505 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6506 * @path: path on the default hierarchy
6508 * Find the cgroup at @path on the default hierarchy, increment its
6509 * reference count and return it. Returns pointer to the found cgroup on
6510 * success, ERR_PTR(-ENOENT) if @path doesn't exist or if the cgroup has already
6511 * been released and ERR_PTR(-ENOTDIR) if @path points to a non-directory.
6513 struct cgroup *cgroup_get_from_path(const char *path)
6515 struct kernfs_node *kn;
6516 struct cgroup *cgrp = ERR_PTR(-ENOENT);
6518 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6522 if (kernfs_type(kn) != KERNFS_DIR) {
6523 cgrp = ERR_PTR(-ENOTDIR);
6529 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6530 if (!cgrp || !cgroup_tryget(cgrp))
6531 cgrp = ERR_PTR(-ENOENT);
6540 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6543 * cgroup_get_from_fd - get a cgroup pointer from a fd
6544 * @fd: fd obtained by open(cgroup2_dir)
6546 * Find the cgroup from a fd which should be obtained
6547 * by opening a cgroup directory. Returns a pointer to the
6548 * cgroup on success. ERR_PTR is returned if the cgroup
6551 struct cgroup *cgroup_get_from_fd(int fd)
6553 struct cgroup *cgrp;
6558 return ERR_PTR(-EBADF);
6560 cgrp = cgroup_get_from_file(f);
6564 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
6566 static u64 power_of_ten(int power)
6575 * cgroup_parse_float - parse a floating number
6576 * @input: input string
6577 * @dec_shift: number of decimal digits to shift
6580 * Parse a decimal floating point number in @input and store the result in
6581 * @v with decimal point right shifted @dec_shift times. For example, if
6582 * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6583 * Returns 0 on success, -errno otherwise.
6585 * There's nothing cgroup specific about this function except that it's
6586 * currently the only user.
6588 int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v)
6590 s64 whole, frac = 0;
6591 int fstart = 0, fend = 0, flen;
6593 if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend))
6598 flen = fend > fstart ? fend - fstart : 0;
6599 if (flen < dec_shift)
6600 frac *= power_of_ten(dec_shift - flen);
6602 frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift));
6604 *v = whole * power_of_ten(dec_shift) + frac;
6609 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6610 * definition in cgroup-defs.h.
6612 #ifdef CONFIG_SOCK_CGROUP_DATA
6614 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6616 struct cgroup *cgroup;
6619 /* Don't associate the sock with unrelated interrupted task's cgroup. */
6620 if (in_interrupt()) {
6621 cgroup = &cgrp_dfl_root.cgrp;
6627 struct css_set *cset;
6629 cset = task_css_set(current);
6630 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6631 cgroup = cset->dfl_cgrp;
6637 skcd->cgroup = cgroup;
6638 cgroup_bpf_get(cgroup);
6642 void cgroup_sk_clone(struct sock_cgroup_data *skcd)
6644 struct cgroup *cgrp = sock_cgroup_ptr(skcd);
6647 * We might be cloning a socket which is left in an empty
6648 * cgroup and the cgroup might have already been rmdir'd.
6649 * Don't use cgroup_get_live().
6652 cgroup_bpf_get(cgrp);
6655 void cgroup_sk_free(struct sock_cgroup_data *skcd)
6657 struct cgroup *cgrp = sock_cgroup_ptr(skcd);
6659 cgroup_bpf_put(cgrp);
6663 #endif /* CONFIG_SOCK_CGROUP_DATA */
6666 static ssize_t show_delegatable_files(struct cftype *files, char *buf,
6667 ssize_t size, const char *prefix)
6672 for (cft = files; cft && cft->name[0] != '\0'; cft++) {
6673 if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
6676 if ((cft->flags & CFTYPE_PRESSURE) && !cgroup_psi_enabled())
6680 ret += snprintf(buf + ret, size - ret, "%s.", prefix);
6682 ret += snprintf(buf + ret, size - ret, "%s\n", cft->name);
6684 if (WARN_ON(ret >= size))
6691 static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr,
6694 struct cgroup_subsys *ss;
6698 ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret,
6701 for_each_subsys(ss, ssid)
6702 ret += show_delegatable_files(ss->dfl_cftypes, buf + ret,
6704 cgroup_subsys_name[ssid]);
6708 static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate);
6710 static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr,
6713 return snprintf(buf, PAGE_SIZE,
6715 "memory_localevents\n"
6716 "memory_recursiveprot\n");
6718 static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features);
6720 static struct attribute *cgroup_sysfs_attrs[] = {
6721 &cgroup_delegate_attr.attr,
6722 &cgroup_features_attr.attr,
6726 static const struct attribute_group cgroup_sysfs_attr_group = {
6727 .attrs = cgroup_sysfs_attrs,
6731 static int __init cgroup_sysfs_init(void)
6733 return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group);
6735 subsys_initcall(cgroup_sysfs_init);
6737 #endif /* CONFIG_SYSFS */