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/sched/deadline.h>
60 #include <linux/psi.h>
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/cgroup.h>
66 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
68 /* let's not notify more than 100 times per second */
69 #define CGROUP_FILE_NOTIFY_MIN_INTV DIV_ROUND_UP(HZ, 100)
72 * cgroup_mutex is the master lock. Any modification to cgroup or its
73 * hierarchy must be performed while holding it.
75 * css_set_lock protects task->cgroups pointer, the list of css_set
76 * objects, and the chain of tasks off each css_set.
78 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
79 * cgroup.h can use them for lockdep annotations.
81 DEFINE_MUTEX(cgroup_mutex);
82 DEFINE_SPINLOCK(css_set_lock);
84 #ifdef CONFIG_PROVE_RCU
85 EXPORT_SYMBOL_GPL(cgroup_mutex);
86 EXPORT_SYMBOL_GPL(css_set_lock);
89 DEFINE_SPINLOCK(trace_cgroup_path_lock);
90 char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
91 bool cgroup_debug __read_mostly;
94 * Protects cgroup_idr and css_idr so that IDs can be released without
95 * grabbing cgroup_mutex.
97 static DEFINE_SPINLOCK(cgroup_idr_lock);
100 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
101 * against file removal/re-creation across css hiding.
103 static DEFINE_SPINLOCK(cgroup_file_kn_lock);
105 DEFINE_PERCPU_RWSEM(cgroup_threadgroup_rwsem);
107 #define cgroup_assert_mutex_or_rcu_locked() \
108 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
109 !lockdep_is_held(&cgroup_mutex), \
110 "cgroup_mutex or RCU read lock required");
113 * cgroup destruction makes heavy use of work items and there can be a lot
114 * of concurrent destructions. Use a separate workqueue so that cgroup
115 * destruction work items don't end up filling up max_active of system_wq
116 * which may lead to deadlock.
118 static struct workqueue_struct *cgroup_destroy_wq;
120 /* generate an array of cgroup subsystem pointers */
121 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
122 struct cgroup_subsys *cgroup_subsys[] = {
123 #include <linux/cgroup_subsys.h>
127 /* array of cgroup subsystem names */
128 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
129 static const char *cgroup_subsys_name[] = {
130 #include <linux/cgroup_subsys.h>
134 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
136 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
137 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
138 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
139 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
140 #include <linux/cgroup_subsys.h>
143 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
144 static struct static_key_true *cgroup_subsys_enabled_key[] = {
145 #include <linux/cgroup_subsys.h>
149 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
150 static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
151 #include <linux/cgroup_subsys.h>
155 static DEFINE_PER_CPU(struct cgroup_rstat_cpu, cgrp_dfl_root_rstat_cpu);
157 /* the default hierarchy */
158 struct cgroup_root cgrp_dfl_root = { .cgrp.rstat_cpu = &cgrp_dfl_root_rstat_cpu };
159 EXPORT_SYMBOL_GPL(cgrp_dfl_root);
162 * The default hierarchy always exists but is hidden until mounted for the
163 * first time. This is for backward compatibility.
165 static bool cgrp_dfl_visible;
167 /* some controllers are not supported in the default hierarchy */
168 static u16 cgrp_dfl_inhibit_ss_mask;
170 /* some controllers are implicitly enabled on the default hierarchy */
171 static u16 cgrp_dfl_implicit_ss_mask;
173 /* some controllers can be threaded on the default hierarchy */
174 static u16 cgrp_dfl_threaded_ss_mask;
176 /* The list of hierarchy roots */
177 LIST_HEAD(cgroup_roots);
178 static int cgroup_root_count;
180 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
181 static DEFINE_IDR(cgroup_hierarchy_idr);
184 * Assign a monotonically increasing serial number to csses. It guarantees
185 * cgroups with bigger numbers are newer than those with smaller numbers.
186 * Also, as csses are always appended to the parent's ->children list, it
187 * guarantees that sibling csses are always sorted in the ascending serial
188 * number order on the list. Protected by cgroup_mutex.
190 static u64 css_serial_nr_next = 1;
193 * These bitmasks identify subsystems with specific features to avoid
194 * having to do iterative checks repeatedly.
196 static u16 have_fork_callback __read_mostly;
197 static u16 have_exit_callback __read_mostly;
198 static u16 have_release_callback __read_mostly;
199 static u16 have_canfork_callback __read_mostly;
201 /* cgroup namespace for init task */
202 struct cgroup_namespace init_cgroup_ns = {
203 .count = REFCOUNT_INIT(2),
204 .user_ns = &init_user_ns,
205 .ns.ops = &cgroupns_operations,
206 .ns.inum = PROC_CGROUP_INIT_INO,
207 .root_cset = &init_css_set,
210 static struct file_system_type cgroup2_fs_type;
211 static struct cftype cgroup_base_files[];
213 static int cgroup_apply_control(struct cgroup *cgrp);
214 static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
215 static void css_task_iter_skip(struct css_task_iter *it,
216 struct task_struct *task);
217 static int cgroup_destroy_locked(struct cgroup *cgrp);
218 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
219 struct cgroup_subsys *ss);
220 static void css_release(struct percpu_ref *ref);
221 static void kill_css(struct cgroup_subsys_state *css);
222 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
223 struct cgroup *cgrp, struct cftype cfts[],
227 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
228 * @ssid: subsys ID of interest
230 * cgroup_subsys_enabled() can only be used with literal subsys names which
231 * is fine for individual subsystems but unsuitable for cgroup core. This
232 * is slower static_key_enabled() based test indexed by @ssid.
234 bool cgroup_ssid_enabled(int ssid)
236 if (CGROUP_SUBSYS_COUNT == 0)
239 return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
243 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
244 * @cgrp: the cgroup of interest
246 * The default hierarchy is the v2 interface of cgroup and this function
247 * can be used to test whether a cgroup is on the default hierarchy for
248 * cases where a subsystem should behave differnetly depending on the
251 * List of changed behaviors:
253 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
254 * and "name" are disallowed.
256 * - When mounting an existing superblock, mount options should match.
258 * - Remount is disallowed.
260 * - rename(2) is disallowed.
262 * - "tasks" is removed. Everything should be at process granularity. Use
263 * "cgroup.procs" instead.
265 * - "cgroup.procs" is not sorted. pids will be unique unless they got
266 * recycled inbetween reads.
268 * - "release_agent" and "notify_on_release" are removed. Replacement
269 * notification mechanism will be implemented.
271 * - "cgroup.clone_children" is removed.
273 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
274 * and its descendants contain no task; otherwise, 1. The file also
275 * generates kernfs notification which can be monitored through poll and
276 * [di]notify when the value of the file changes.
278 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
279 * take masks of ancestors with non-empty cpus/mems, instead of being
280 * moved to an ancestor.
282 * - cpuset: a task can be moved into an empty cpuset, and again it takes
283 * masks of ancestors.
285 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
288 * - blkcg: blk-throttle becomes properly hierarchical.
290 * - debug: disallowed on the default hierarchy.
292 bool cgroup_on_dfl(const struct cgroup *cgrp)
294 return cgrp->root == &cgrp_dfl_root;
297 /* IDR wrappers which synchronize using cgroup_idr_lock */
298 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
303 idr_preload(gfp_mask);
304 spin_lock_bh(&cgroup_idr_lock);
305 ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
306 spin_unlock_bh(&cgroup_idr_lock);
311 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
315 spin_lock_bh(&cgroup_idr_lock);
316 ret = idr_replace(idr, ptr, id);
317 spin_unlock_bh(&cgroup_idr_lock);
321 static void cgroup_idr_remove(struct idr *idr, int id)
323 spin_lock_bh(&cgroup_idr_lock);
325 spin_unlock_bh(&cgroup_idr_lock);
328 static bool cgroup_has_tasks(struct cgroup *cgrp)
330 return cgrp->nr_populated_csets;
333 bool cgroup_is_threaded(struct cgroup *cgrp)
335 return cgrp->dom_cgrp != cgrp;
338 /* can @cgrp host both domain and threaded children? */
339 static bool cgroup_is_mixable(struct cgroup *cgrp)
342 * Root isn't under domain level resource control exempting it from
343 * the no-internal-process constraint, so it can serve as a thread
344 * root and a parent of resource domains at the same time.
346 return !cgroup_parent(cgrp);
349 /* can @cgrp become a thread root? should always be true for a thread root */
350 static bool cgroup_can_be_thread_root(struct cgroup *cgrp)
352 /* mixables don't care */
353 if (cgroup_is_mixable(cgrp))
356 /* domain roots can't be nested under threaded */
357 if (cgroup_is_threaded(cgrp))
360 /* can only have either domain or threaded children */
361 if (cgrp->nr_populated_domain_children)
364 /* and no domain controllers can be enabled */
365 if (cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
371 /* is @cgrp root of a threaded subtree? */
372 bool cgroup_is_thread_root(struct cgroup *cgrp)
374 /* thread root should be a domain */
375 if (cgroup_is_threaded(cgrp))
378 /* a domain w/ threaded children is a thread root */
379 if (cgrp->nr_threaded_children)
383 * A domain which has tasks and explicit threaded controllers
384 * enabled is a thread root.
386 if (cgroup_has_tasks(cgrp) &&
387 (cgrp->subtree_control & cgrp_dfl_threaded_ss_mask))
393 /* a domain which isn't connected to the root w/o brekage can't be used */
394 static bool cgroup_is_valid_domain(struct cgroup *cgrp)
396 /* the cgroup itself can be a thread root */
397 if (cgroup_is_threaded(cgrp))
400 /* but the ancestors can't be unless mixable */
401 while ((cgrp = cgroup_parent(cgrp))) {
402 if (!cgroup_is_mixable(cgrp) && cgroup_is_thread_root(cgrp))
404 if (cgroup_is_threaded(cgrp))
411 /* subsystems visibly enabled on a cgroup */
412 static u16 cgroup_control(struct cgroup *cgrp)
414 struct cgroup *parent = cgroup_parent(cgrp);
415 u16 root_ss_mask = cgrp->root->subsys_mask;
418 u16 ss_mask = parent->subtree_control;
420 /* threaded cgroups can only have threaded controllers */
421 if (cgroup_is_threaded(cgrp))
422 ss_mask &= cgrp_dfl_threaded_ss_mask;
426 if (cgroup_on_dfl(cgrp))
427 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
428 cgrp_dfl_implicit_ss_mask);
432 /* subsystems enabled on a cgroup */
433 static u16 cgroup_ss_mask(struct cgroup *cgrp)
435 struct cgroup *parent = cgroup_parent(cgrp);
438 u16 ss_mask = parent->subtree_ss_mask;
440 /* threaded cgroups can only have threaded controllers */
441 if (cgroup_is_threaded(cgrp))
442 ss_mask &= cgrp_dfl_threaded_ss_mask;
446 return cgrp->root->subsys_mask;
450 * cgroup_css - obtain a cgroup's css for the specified subsystem
451 * @cgrp: the cgroup of interest
452 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
454 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
455 * function must be called either under cgroup_mutex or rcu_read_lock() and
456 * the caller is responsible for pinning the returned css if it wants to
457 * keep accessing it outside the said locks. This function may return
458 * %NULL if @cgrp doesn't have @subsys_id enabled.
460 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
461 struct cgroup_subsys *ss)
464 return rcu_dereference_check(cgrp->subsys[ss->id],
465 lockdep_is_held(&cgroup_mutex));
471 * cgroup_tryget_css - try to get a cgroup's css for the specified subsystem
472 * @cgrp: the cgroup of interest
473 * @ss: the subsystem of interest
475 * Find and get @cgrp's css assocaited with @ss. If the css doesn't exist
476 * or is offline, %NULL is returned.
478 static struct cgroup_subsys_state *cgroup_tryget_css(struct cgroup *cgrp,
479 struct cgroup_subsys *ss)
481 struct cgroup_subsys_state *css;
484 css = cgroup_css(cgrp, ss);
485 if (css && !css_tryget_online(css))
493 * cgroup_e_css_by_mask - obtain a cgroup's effective css for the specified ss
494 * @cgrp: the cgroup of interest
495 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
497 * Similar to cgroup_css() but returns the effective css, which is defined
498 * as the matching css of the nearest ancestor including self which has @ss
499 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
500 * function is guaranteed to return non-NULL css.
502 static struct cgroup_subsys_state *cgroup_e_css_by_mask(struct cgroup *cgrp,
503 struct cgroup_subsys *ss)
505 lockdep_assert_held(&cgroup_mutex);
511 * This function is used while updating css associations and thus
512 * can't test the csses directly. Test ss_mask.
514 while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
515 cgrp = cgroup_parent(cgrp);
520 return cgroup_css(cgrp, ss);
524 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
525 * @cgrp: the cgroup of interest
526 * @ss: the subsystem of interest
528 * Find and get the effective css of @cgrp for @ss. The effective css is
529 * defined as the matching css of the nearest ancestor including self which
530 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
531 * the root css is returned, so this function always returns a valid css.
533 * The returned css is not guaranteed to be online, and therefore it is the
534 * callers responsiblity to tryget a reference for it.
536 struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
537 struct cgroup_subsys *ss)
539 struct cgroup_subsys_state *css;
542 css = cgroup_css(cgrp, ss);
546 cgrp = cgroup_parent(cgrp);
549 return init_css_set.subsys[ss->id];
553 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
554 * @cgrp: the cgroup of interest
555 * @ss: the subsystem of interest
557 * Find and get the effective css of @cgrp for @ss. The effective css is
558 * defined as the matching css of the nearest ancestor including self which
559 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
560 * the root css is returned, so this function always returns a valid css.
561 * The returned css must be put using css_put().
563 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
564 struct cgroup_subsys *ss)
566 struct cgroup_subsys_state *css;
571 css = cgroup_css(cgrp, ss);
573 if (css && css_tryget_online(css))
575 cgrp = cgroup_parent(cgrp);
578 css = init_css_set.subsys[ss->id];
585 static void cgroup_get_live(struct cgroup *cgrp)
587 WARN_ON_ONCE(cgroup_is_dead(cgrp));
588 css_get(&cgrp->self);
592 * __cgroup_task_count - count the number of tasks in a cgroup. The caller
593 * is responsible for taking the css_set_lock.
594 * @cgrp: the cgroup in question
596 int __cgroup_task_count(const struct cgroup *cgrp)
599 struct cgrp_cset_link *link;
601 lockdep_assert_held(&css_set_lock);
603 list_for_each_entry(link, &cgrp->cset_links, cset_link)
604 count += link->cset->nr_tasks;
610 * cgroup_task_count - count the number of tasks in a cgroup.
611 * @cgrp: the cgroup in question
613 int cgroup_task_count(const struct cgroup *cgrp)
617 spin_lock_irq(&css_set_lock);
618 count = __cgroup_task_count(cgrp);
619 spin_unlock_irq(&css_set_lock);
624 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
626 struct cgroup *cgrp = of->kn->parent->priv;
627 struct cftype *cft = of_cft(of);
630 * This is open and unprotected implementation of cgroup_css().
631 * seq_css() is only called from a kernfs file operation which has
632 * an active reference on the file. Because all the subsystem
633 * files are drained before a css is disassociated with a cgroup,
634 * the matching css from the cgroup's subsys table is guaranteed to
635 * be and stay valid until the enclosing operation is complete.
638 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
642 EXPORT_SYMBOL_GPL(of_css);
645 * for_each_css - iterate all css's of a cgroup
646 * @css: the iteration cursor
647 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
648 * @cgrp: the target cgroup to iterate css's of
650 * Should be called under cgroup_[tree_]mutex.
652 #define for_each_css(css, ssid, cgrp) \
653 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
654 if (!((css) = rcu_dereference_check( \
655 (cgrp)->subsys[(ssid)], \
656 lockdep_is_held(&cgroup_mutex)))) { } \
660 * for_each_e_css - iterate all effective css's of a cgroup
661 * @css: the iteration cursor
662 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
663 * @cgrp: the target cgroup to iterate css's of
665 * Should be called under cgroup_[tree_]mutex.
667 #define for_each_e_css(css, ssid, cgrp) \
668 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
669 if (!((css) = cgroup_e_css_by_mask(cgrp, \
670 cgroup_subsys[(ssid)]))) \
675 * do_each_subsys_mask - filter for_each_subsys with a bitmask
676 * @ss: the iteration cursor
677 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
678 * @ss_mask: the bitmask
680 * The block will only run for cases where the ssid-th bit (1 << ssid) of
683 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
684 unsigned long __ss_mask = (ss_mask); \
685 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
689 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
690 (ss) = cgroup_subsys[ssid]; \
693 #define while_each_subsys_mask() \
698 /* iterate over child cgrps, lock should be held throughout iteration */
699 #define cgroup_for_each_live_child(child, cgrp) \
700 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
701 if (({ lockdep_assert_held(&cgroup_mutex); \
702 cgroup_is_dead(child); })) \
706 /* walk live descendants in preorder */
707 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
708 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
709 if (({ lockdep_assert_held(&cgroup_mutex); \
710 (dsct) = (d_css)->cgroup; \
711 cgroup_is_dead(dsct); })) \
715 /* walk live descendants in postorder */
716 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
717 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
718 if (({ lockdep_assert_held(&cgroup_mutex); \
719 (dsct) = (d_css)->cgroup; \
720 cgroup_is_dead(dsct); })) \
725 * The default css_set - used by init and its children prior to any
726 * hierarchies being mounted. It contains a pointer to the root state
727 * for each subsystem. Also used to anchor the list of css_sets. Not
728 * reference-counted, to improve performance when child cgroups
729 * haven't been created.
731 struct css_set init_css_set = {
732 .refcount = REFCOUNT_INIT(1),
733 .dom_cset = &init_css_set,
734 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
735 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
736 .dying_tasks = LIST_HEAD_INIT(init_css_set.dying_tasks),
737 .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
738 .threaded_csets = LIST_HEAD_INIT(init_css_set.threaded_csets),
739 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
740 .mg_src_preload_node = LIST_HEAD_INIT(init_css_set.mg_src_preload_node),
741 .mg_dst_preload_node = LIST_HEAD_INIT(init_css_set.mg_dst_preload_node),
742 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
745 * The following field is re-initialized when this cset gets linked
746 * in cgroup_init(). However, let's initialize the field
747 * statically too so that the default cgroup can be accessed safely
750 .dfl_cgrp = &cgrp_dfl_root.cgrp,
753 static int css_set_count = 1; /* 1 for init_css_set */
755 static bool css_set_threaded(struct css_set *cset)
757 return cset->dom_cset != cset;
761 * css_set_populated - does a css_set contain any tasks?
762 * @cset: target css_set
764 * css_set_populated() should be the same as !!cset->nr_tasks at steady
765 * state. However, css_set_populated() can be called while a task is being
766 * added to or removed from the linked list before the nr_tasks is
767 * properly updated. Hence, we can't just look at ->nr_tasks here.
769 static bool css_set_populated(struct css_set *cset)
771 lockdep_assert_held(&css_set_lock);
773 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
777 * cgroup_update_populated - update the populated count of a cgroup
778 * @cgrp: the target cgroup
779 * @populated: inc or dec populated count
781 * One of the css_sets associated with @cgrp is either getting its first
782 * task or losing the last. Update @cgrp->nr_populated_* accordingly. The
783 * count is propagated towards root so that a given cgroup's
784 * nr_populated_children is zero iff none of its descendants contain any
787 * @cgrp's interface file "cgroup.populated" is zero if both
788 * @cgrp->nr_populated_csets and @cgrp->nr_populated_children are zero and
789 * 1 otherwise. When the sum changes from or to zero, userland is notified
790 * that the content of the interface file has changed. This can be used to
791 * detect when @cgrp and its descendants become populated or empty.
793 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
795 struct cgroup *child = NULL;
796 int adj = populated ? 1 : -1;
798 lockdep_assert_held(&css_set_lock);
801 bool was_populated = cgroup_is_populated(cgrp);
804 cgrp->nr_populated_csets += adj;
806 if (cgroup_is_threaded(child))
807 cgrp->nr_populated_threaded_children += adj;
809 cgrp->nr_populated_domain_children += adj;
812 if (was_populated == cgroup_is_populated(cgrp))
815 cgroup1_check_for_release(cgrp);
816 TRACE_CGROUP_PATH(notify_populated, cgrp,
817 cgroup_is_populated(cgrp));
818 cgroup_file_notify(&cgrp->events_file);
821 cgrp = cgroup_parent(cgrp);
826 * css_set_update_populated - update populated state of a css_set
827 * @cset: target css_set
828 * @populated: whether @cset is populated or depopulated
830 * @cset is either getting the first task or losing the last. Update the
831 * populated counters of all associated cgroups accordingly.
833 static void css_set_update_populated(struct css_set *cset, bool populated)
835 struct cgrp_cset_link *link;
837 lockdep_assert_held(&css_set_lock);
839 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
840 cgroup_update_populated(link->cgrp, populated);
844 * @task is leaving, advance task iterators which are pointing to it so
845 * that they can resume at the next position. Advancing an iterator might
846 * remove it from the list, use safe walk. See css_task_iter_skip() for
849 static void css_set_skip_task_iters(struct css_set *cset,
850 struct task_struct *task)
852 struct css_task_iter *it, *pos;
854 list_for_each_entry_safe(it, pos, &cset->task_iters, iters_node)
855 css_task_iter_skip(it, task);
859 * css_set_move_task - move a task from one css_set to another
860 * @task: task being moved
861 * @from_cset: css_set @task currently belongs to (may be NULL)
862 * @to_cset: new css_set @task is being moved to (may be NULL)
863 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
865 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
866 * css_set, @from_cset can be NULL. If @task is being disassociated
867 * instead of moved, @to_cset can be NULL.
869 * This function automatically handles populated counter updates and
870 * css_task_iter adjustments but the caller is responsible for managing
871 * @from_cset and @to_cset's reference counts.
873 static void css_set_move_task(struct task_struct *task,
874 struct css_set *from_cset, struct css_set *to_cset,
877 lockdep_assert_held(&css_set_lock);
879 if (to_cset && !css_set_populated(to_cset))
880 css_set_update_populated(to_cset, true);
883 WARN_ON_ONCE(list_empty(&task->cg_list));
885 css_set_skip_task_iters(from_cset, task);
886 list_del_init(&task->cg_list);
887 if (!css_set_populated(from_cset))
888 css_set_update_populated(from_cset, false);
890 WARN_ON_ONCE(!list_empty(&task->cg_list));
895 * We are synchronized through cgroup_threadgroup_rwsem
896 * against PF_EXITING setting such that we can't race
897 * against cgroup_exit()/cgroup_free() dropping the css_set.
899 WARN_ON_ONCE(task->flags & PF_EXITING);
901 cgroup_move_task(task, to_cset);
902 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
908 * hash table for cgroup groups. This improves the performance to find
909 * an existing css_set. This hash doesn't (currently) take into
910 * account cgroups in empty hierarchies.
912 #define CSS_SET_HASH_BITS 7
913 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
915 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
917 unsigned long key = 0UL;
918 struct cgroup_subsys *ss;
921 for_each_subsys(ss, i)
922 key += (unsigned long)css[i];
923 key = (key >> 16) ^ key;
928 void put_css_set_locked(struct css_set *cset)
930 struct cgrp_cset_link *link, *tmp_link;
931 struct cgroup_subsys *ss;
934 lockdep_assert_held(&css_set_lock);
936 if (!refcount_dec_and_test(&cset->refcount))
939 WARN_ON_ONCE(!list_empty(&cset->threaded_csets));
941 /* This css_set is dead. unlink it and release cgroup and css refs */
942 for_each_subsys(ss, ssid) {
943 list_del(&cset->e_cset_node[ssid]);
944 css_put(cset->subsys[ssid]);
946 hash_del(&cset->hlist);
949 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
950 list_del(&link->cset_link);
951 list_del(&link->cgrp_link);
952 if (cgroup_parent(link->cgrp))
953 cgroup_put(link->cgrp);
957 if (css_set_threaded(cset)) {
958 list_del(&cset->threaded_csets_node);
959 put_css_set_locked(cset->dom_cset);
962 kfree_rcu(cset, rcu_head);
966 * compare_css_sets - helper function for find_existing_css_set().
967 * @cset: candidate css_set being tested
968 * @old_cset: existing css_set for a task
969 * @new_cgrp: cgroup that's being entered by the task
970 * @template: desired set of css pointers in css_set (pre-calculated)
972 * Returns true if "cset" matches "old_cset" except for the hierarchy
973 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
975 static bool compare_css_sets(struct css_set *cset,
976 struct css_set *old_cset,
977 struct cgroup *new_cgrp,
978 struct cgroup_subsys_state *template[])
980 struct cgroup *new_dfl_cgrp;
981 struct list_head *l1, *l2;
984 * On the default hierarchy, there can be csets which are
985 * associated with the same set of cgroups but different csses.
986 * Let's first ensure that csses match.
988 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
992 /* @cset's domain should match the default cgroup's */
993 if (cgroup_on_dfl(new_cgrp))
994 new_dfl_cgrp = new_cgrp;
996 new_dfl_cgrp = old_cset->dfl_cgrp;
998 if (new_dfl_cgrp->dom_cgrp != cset->dom_cset->dfl_cgrp)
1002 * Compare cgroup pointers in order to distinguish between
1003 * different cgroups in hierarchies. As different cgroups may
1004 * share the same effective css, this comparison is always
1007 l1 = &cset->cgrp_links;
1008 l2 = &old_cset->cgrp_links;
1010 struct cgrp_cset_link *link1, *link2;
1011 struct cgroup *cgrp1, *cgrp2;
1015 /* See if we reached the end - both lists are equal length. */
1016 if (l1 == &cset->cgrp_links) {
1017 BUG_ON(l2 != &old_cset->cgrp_links);
1020 BUG_ON(l2 == &old_cset->cgrp_links);
1022 /* Locate the cgroups associated with these links. */
1023 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
1024 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
1025 cgrp1 = link1->cgrp;
1026 cgrp2 = link2->cgrp;
1027 /* Hierarchies should be linked in the same order. */
1028 BUG_ON(cgrp1->root != cgrp2->root);
1031 * If this hierarchy is the hierarchy of the cgroup
1032 * that's changing, then we need to check that this
1033 * css_set points to the new cgroup; if it's any other
1034 * hierarchy, then this css_set should point to the
1035 * same cgroup as the old css_set.
1037 if (cgrp1->root == new_cgrp->root) {
1038 if (cgrp1 != new_cgrp)
1049 * find_existing_css_set - init css array and find the matching css_set
1050 * @old_cset: the css_set that we're using before the cgroup transition
1051 * @cgrp: the cgroup that we're moving into
1052 * @template: out param for the new set of csses, should be clear on entry
1054 static struct css_set *find_existing_css_set(struct css_set *old_cset,
1055 struct cgroup *cgrp,
1056 struct cgroup_subsys_state *template[])
1058 struct cgroup_root *root = cgrp->root;
1059 struct cgroup_subsys *ss;
1060 struct css_set *cset;
1065 * Build the set of subsystem state objects that we want to see in the
1066 * new css_set. while subsystems can change globally, the entries here
1067 * won't change, so no need for locking.
1069 for_each_subsys(ss, i) {
1070 if (root->subsys_mask & (1UL << i)) {
1072 * @ss is in this hierarchy, so we want the
1073 * effective css from @cgrp.
1075 template[i] = cgroup_e_css_by_mask(cgrp, ss);
1078 * @ss is not in this hierarchy, so we don't want
1079 * to change the css.
1081 template[i] = old_cset->subsys[i];
1085 key = css_set_hash(template);
1086 hash_for_each_possible(css_set_table, cset, hlist, key) {
1087 if (!compare_css_sets(cset, old_cset, cgrp, template))
1090 /* This css_set matches what we need */
1094 /* No existing cgroup group matched */
1098 static void free_cgrp_cset_links(struct list_head *links_to_free)
1100 struct cgrp_cset_link *link, *tmp_link;
1102 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
1103 list_del(&link->cset_link);
1109 * allocate_cgrp_cset_links - allocate cgrp_cset_links
1110 * @count: the number of links to allocate
1111 * @tmp_links: list_head the allocated links are put on
1113 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
1114 * through ->cset_link. Returns 0 on success or -errno.
1116 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
1118 struct cgrp_cset_link *link;
1121 INIT_LIST_HEAD(tmp_links);
1123 for (i = 0; i < count; i++) {
1124 link = kzalloc(sizeof(*link), GFP_KERNEL);
1126 free_cgrp_cset_links(tmp_links);
1129 list_add(&link->cset_link, tmp_links);
1135 * link_css_set - a helper function to link a css_set to a cgroup
1136 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
1137 * @cset: the css_set to be linked
1138 * @cgrp: the destination cgroup
1140 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
1141 struct cgroup *cgrp)
1143 struct cgrp_cset_link *link;
1145 BUG_ON(list_empty(tmp_links));
1147 if (cgroup_on_dfl(cgrp))
1148 cset->dfl_cgrp = cgrp;
1150 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
1155 * Always add links to the tail of the lists so that the lists are
1156 * in choronological order.
1158 list_move_tail(&link->cset_link, &cgrp->cset_links);
1159 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
1161 if (cgroup_parent(cgrp))
1162 cgroup_get_live(cgrp);
1166 * find_css_set - return a new css_set with one cgroup updated
1167 * @old_cset: the baseline css_set
1168 * @cgrp: the cgroup to be updated
1170 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
1171 * substituted into the appropriate hierarchy.
1173 static struct css_set *find_css_set(struct css_set *old_cset,
1174 struct cgroup *cgrp)
1176 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
1177 struct css_set *cset;
1178 struct list_head tmp_links;
1179 struct cgrp_cset_link *link;
1180 struct cgroup_subsys *ss;
1184 lockdep_assert_held(&cgroup_mutex);
1186 /* First see if we already have a cgroup group that matches
1187 * the desired set */
1188 spin_lock_irq(&css_set_lock);
1189 cset = find_existing_css_set(old_cset, cgrp, template);
1192 spin_unlock_irq(&css_set_lock);
1197 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1201 /* Allocate all the cgrp_cset_link objects that we'll need */
1202 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1207 refcount_set(&cset->refcount, 1);
1208 cset->dom_cset = cset;
1209 INIT_LIST_HEAD(&cset->tasks);
1210 INIT_LIST_HEAD(&cset->mg_tasks);
1211 INIT_LIST_HEAD(&cset->dying_tasks);
1212 INIT_LIST_HEAD(&cset->task_iters);
1213 INIT_LIST_HEAD(&cset->threaded_csets);
1214 INIT_HLIST_NODE(&cset->hlist);
1215 INIT_LIST_HEAD(&cset->cgrp_links);
1216 INIT_LIST_HEAD(&cset->mg_src_preload_node);
1217 INIT_LIST_HEAD(&cset->mg_dst_preload_node);
1218 INIT_LIST_HEAD(&cset->mg_node);
1220 /* Copy the set of subsystem state objects generated in
1221 * find_existing_css_set() */
1222 memcpy(cset->subsys, template, sizeof(cset->subsys));
1224 spin_lock_irq(&css_set_lock);
1225 /* Add reference counts and links from the new css_set. */
1226 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1227 struct cgroup *c = link->cgrp;
1229 if (c->root == cgrp->root)
1231 link_css_set(&tmp_links, cset, c);
1234 BUG_ON(!list_empty(&tmp_links));
1238 /* Add @cset to the hash table */
1239 key = css_set_hash(cset->subsys);
1240 hash_add(css_set_table, &cset->hlist, key);
1242 for_each_subsys(ss, ssid) {
1243 struct cgroup_subsys_state *css = cset->subsys[ssid];
1245 list_add_tail(&cset->e_cset_node[ssid],
1246 &css->cgroup->e_csets[ssid]);
1250 spin_unlock_irq(&css_set_lock);
1253 * If @cset should be threaded, look up the matching dom_cset and
1254 * link them up. We first fully initialize @cset then look for the
1255 * dom_cset. It's simpler this way and safe as @cset is guaranteed
1256 * to stay empty until we return.
1258 if (cgroup_is_threaded(cset->dfl_cgrp)) {
1259 struct css_set *dcset;
1261 dcset = find_css_set(cset, cset->dfl_cgrp->dom_cgrp);
1267 spin_lock_irq(&css_set_lock);
1268 cset->dom_cset = dcset;
1269 list_add_tail(&cset->threaded_csets_node,
1270 &dcset->threaded_csets);
1271 spin_unlock_irq(&css_set_lock);
1277 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1279 struct cgroup *root_cgrp = kf_root->kn->priv;
1281 return root_cgrp->root;
1284 static int cgroup_init_root_id(struct cgroup_root *root)
1288 lockdep_assert_held(&cgroup_mutex);
1290 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1294 root->hierarchy_id = id;
1298 static void cgroup_exit_root_id(struct cgroup_root *root)
1300 lockdep_assert_held(&cgroup_mutex);
1302 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1305 void cgroup_free_root(struct cgroup_root *root)
1310 static void cgroup_destroy_root(struct cgroup_root *root)
1312 struct cgroup *cgrp = &root->cgrp;
1313 struct cgrp_cset_link *link, *tmp_link;
1315 trace_cgroup_destroy_root(root);
1317 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1319 BUG_ON(atomic_read(&root->nr_cgrps));
1320 BUG_ON(!list_empty(&cgrp->self.children));
1322 /* Rebind all subsystems back to the default hierarchy */
1323 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1326 * Release all the links from cset_links to this hierarchy's
1329 spin_lock_irq(&css_set_lock);
1331 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1332 list_del(&link->cset_link);
1333 list_del(&link->cgrp_link);
1337 spin_unlock_irq(&css_set_lock);
1339 if (!list_empty(&root->root_list)) {
1340 list_del(&root->root_list);
1341 cgroup_root_count--;
1344 cgroup_exit_root_id(root);
1346 mutex_unlock(&cgroup_mutex);
1348 kernfs_destroy_root(root->kf_root);
1349 cgroup_free_root(root);
1353 * look up cgroup associated with current task's cgroup namespace on the
1354 * specified hierarchy
1356 static struct cgroup *
1357 current_cgns_cgroup_from_root(struct cgroup_root *root)
1359 struct cgroup *res = NULL;
1360 struct css_set *cset;
1362 lockdep_assert_held(&css_set_lock);
1366 cset = current->nsproxy->cgroup_ns->root_cset;
1367 if (cset == &init_css_set) {
1369 } else if (root == &cgrp_dfl_root) {
1370 res = cset->dfl_cgrp;
1372 struct cgrp_cset_link *link;
1374 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1375 struct cgroup *c = link->cgrp;
1377 if (c->root == root) {
1389 /* look up cgroup associated with given css_set on the specified hierarchy */
1390 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1391 struct cgroup_root *root)
1393 struct cgroup *res = NULL;
1395 lockdep_assert_held(&cgroup_mutex);
1396 lockdep_assert_held(&css_set_lock);
1398 if (cset == &init_css_set) {
1400 } else if (root == &cgrp_dfl_root) {
1401 res = cset->dfl_cgrp;
1403 struct cgrp_cset_link *link;
1405 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1406 struct cgroup *c = link->cgrp;
1408 if (c->root == root) {
1420 * Return the cgroup for "task" from the given hierarchy. Must be
1421 * called with cgroup_mutex and css_set_lock held.
1423 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1424 struct cgroup_root *root)
1427 * No need to lock the task - since we hold css_set_lock the
1428 * task can't change groups.
1430 return cset_cgroup_from_root(task_css_set(task), root);
1434 * A task must hold cgroup_mutex to modify cgroups.
1436 * Any task can increment and decrement the count field without lock.
1437 * So in general, code holding cgroup_mutex can't rely on the count
1438 * field not changing. However, if the count goes to zero, then only
1439 * cgroup_attach_task() can increment it again. Because a count of zero
1440 * means that no tasks are currently attached, therefore there is no
1441 * way a task attached to that cgroup can fork (the other way to
1442 * increment the count). So code holding cgroup_mutex can safely
1443 * assume that if the count is zero, it will stay zero. Similarly, if
1444 * a task holds cgroup_mutex on a cgroup with zero count, it
1445 * knows that the cgroup won't be removed, as cgroup_rmdir()
1448 * A cgroup can only be deleted if both its 'count' of using tasks
1449 * is zero, and its list of 'children' cgroups is empty. Since all
1450 * tasks in the system use _some_ cgroup, and since there is always at
1451 * least one task in the system (init, pid == 1), therefore, root cgroup
1452 * always has either children cgroups and/or using tasks. So we don't
1453 * need a special hack to ensure that root cgroup cannot be deleted.
1455 * P.S. One more locking exception. RCU is used to guard the
1456 * update of a tasks cgroup pointer by cgroup_attach_task()
1459 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1461 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1464 struct cgroup_subsys *ss = cft->ss;
1466 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1467 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
1468 const char *dbg = (cft->flags & CFTYPE_DEBUG) ? ".__DEBUG__." : "";
1470 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s%s.%s",
1471 dbg, cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1474 strscpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1480 * cgroup_file_mode - deduce file mode of a control file
1481 * @cft: the control file in question
1483 * S_IRUGO for read, S_IWUSR for write.
1485 static umode_t cgroup_file_mode(const struct cftype *cft)
1489 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1492 if (cft->write_u64 || cft->write_s64 || cft->write) {
1493 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1503 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1504 * @subtree_control: the new subtree_control mask to consider
1505 * @this_ss_mask: available subsystems
1507 * On the default hierarchy, a subsystem may request other subsystems to be
1508 * enabled together through its ->depends_on mask. In such cases, more
1509 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1511 * This function calculates which subsystems need to be enabled if
1512 * @subtree_control is to be applied while restricted to @this_ss_mask.
1514 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1516 u16 cur_ss_mask = subtree_control;
1517 struct cgroup_subsys *ss;
1520 lockdep_assert_held(&cgroup_mutex);
1522 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1525 u16 new_ss_mask = cur_ss_mask;
1527 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1528 new_ss_mask |= ss->depends_on;
1529 } while_each_subsys_mask();
1532 * Mask out subsystems which aren't available. This can
1533 * happen only if some depended-upon subsystems were bound
1534 * to non-default hierarchies.
1536 new_ss_mask &= this_ss_mask;
1538 if (new_ss_mask == cur_ss_mask)
1540 cur_ss_mask = new_ss_mask;
1547 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1548 * @kn: the kernfs_node being serviced
1550 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1551 * the method finishes if locking succeeded. Note that once this function
1552 * returns the cgroup returned by cgroup_kn_lock_live() may become
1553 * inaccessible any time. If the caller intends to continue to access the
1554 * cgroup, it should pin it before invoking this function.
1556 void cgroup_kn_unlock(struct kernfs_node *kn)
1558 struct cgroup *cgrp;
1560 if (kernfs_type(kn) == KERNFS_DIR)
1563 cgrp = kn->parent->priv;
1565 mutex_unlock(&cgroup_mutex);
1567 kernfs_unbreak_active_protection(kn);
1572 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1573 * @kn: the kernfs_node being serviced
1574 * @drain_offline: perform offline draining on the cgroup
1576 * This helper is to be used by a cgroup kernfs method currently servicing
1577 * @kn. It breaks the active protection, performs cgroup locking and
1578 * verifies that the associated cgroup is alive. Returns the cgroup if
1579 * alive; otherwise, %NULL. A successful return should be undone by a
1580 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1581 * cgroup is drained of offlining csses before return.
1583 * Any cgroup kernfs method implementation which requires locking the
1584 * associated cgroup should use this helper. It avoids nesting cgroup
1585 * locking under kernfs active protection and allows all kernfs operations
1586 * including self-removal.
1588 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1590 struct cgroup *cgrp;
1592 if (kernfs_type(kn) == KERNFS_DIR)
1595 cgrp = kn->parent->priv;
1598 * We're gonna grab cgroup_mutex which nests outside kernfs
1599 * active_ref. cgroup liveliness check alone provides enough
1600 * protection against removal. Ensure @cgrp stays accessible and
1601 * break the active_ref protection.
1603 if (!cgroup_tryget(cgrp))
1605 kernfs_break_active_protection(kn);
1608 cgroup_lock_and_drain_offline(cgrp);
1610 mutex_lock(&cgroup_mutex);
1612 if (!cgroup_is_dead(cgrp))
1615 cgroup_kn_unlock(kn);
1619 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1621 char name[CGROUP_FILE_NAME_MAX];
1623 lockdep_assert_held(&cgroup_mutex);
1625 if (cft->file_offset) {
1626 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1627 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1629 spin_lock_irq(&cgroup_file_kn_lock);
1631 spin_unlock_irq(&cgroup_file_kn_lock);
1633 del_timer_sync(&cfile->notify_timer);
1636 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1640 * css_clear_dir - remove subsys files in a cgroup directory
1643 static void css_clear_dir(struct cgroup_subsys_state *css)
1645 struct cgroup *cgrp = css->cgroup;
1646 struct cftype *cfts;
1648 if (!(css->flags & CSS_VISIBLE))
1651 css->flags &= ~CSS_VISIBLE;
1654 if (cgroup_on_dfl(cgrp))
1655 cfts = cgroup_base_files;
1657 cfts = cgroup1_base_files;
1659 cgroup_addrm_files(css, cgrp, cfts, false);
1661 list_for_each_entry(cfts, &css->ss->cfts, node)
1662 cgroup_addrm_files(css, cgrp, cfts, false);
1667 * css_populate_dir - create subsys files in a cgroup directory
1670 * On failure, no file is added.
1672 static int css_populate_dir(struct cgroup_subsys_state *css)
1674 struct cgroup *cgrp = css->cgroup;
1675 struct cftype *cfts, *failed_cfts;
1678 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1682 if (cgroup_on_dfl(cgrp))
1683 cfts = cgroup_base_files;
1685 cfts = cgroup1_base_files;
1687 ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1691 list_for_each_entry(cfts, &css->ss->cfts, node) {
1692 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1700 css->flags |= CSS_VISIBLE;
1704 list_for_each_entry(cfts, &css->ss->cfts, node) {
1705 if (cfts == failed_cfts)
1707 cgroup_addrm_files(css, cgrp, cfts, false);
1712 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1714 struct cgroup *dcgrp = &dst_root->cgrp;
1715 struct cgroup_subsys *ss;
1717 u16 dfl_disable_ss_mask = 0;
1719 lockdep_assert_held(&cgroup_mutex);
1721 do_each_subsys_mask(ss, ssid, ss_mask) {
1723 * If @ss has non-root csses attached to it, can't move.
1724 * If @ss is an implicit controller, it is exempt from this
1725 * rule and can be stolen.
1727 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1728 !ss->implicit_on_dfl)
1731 /* can't move between two non-dummy roots either */
1732 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1736 * Collect ssid's that need to be disabled from default
1739 if (ss->root == &cgrp_dfl_root)
1740 dfl_disable_ss_mask |= 1 << ssid;
1742 } while_each_subsys_mask();
1744 if (dfl_disable_ss_mask) {
1745 struct cgroup *scgrp = &cgrp_dfl_root.cgrp;
1748 * Controllers from default hierarchy that need to be rebound
1749 * are all disabled together in one go.
1751 cgrp_dfl_root.subsys_mask &= ~dfl_disable_ss_mask;
1752 WARN_ON(cgroup_apply_control(scgrp));
1753 cgroup_finalize_control(scgrp, 0);
1756 do_each_subsys_mask(ss, ssid, ss_mask) {
1757 struct cgroup_root *src_root = ss->root;
1758 struct cgroup *scgrp = &src_root->cgrp;
1759 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1760 struct css_set *cset, *cset_pos;
1761 struct css_task_iter *it;
1763 WARN_ON(!css || cgroup_css(dcgrp, ss));
1765 if (src_root != &cgrp_dfl_root) {
1766 /* disable from the source */
1767 src_root->subsys_mask &= ~(1 << ssid);
1768 WARN_ON(cgroup_apply_control(scgrp));
1769 cgroup_finalize_control(scgrp, 0);
1773 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1774 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1775 ss->root = dst_root;
1776 css->cgroup = dcgrp;
1778 spin_lock_irq(&css_set_lock);
1779 WARN_ON(!list_empty(&dcgrp->e_csets[ss->id]));
1780 list_for_each_entry_safe(cset, cset_pos, &scgrp->e_csets[ss->id],
1781 e_cset_node[ss->id]) {
1782 list_move_tail(&cset->e_cset_node[ss->id],
1783 &dcgrp->e_csets[ss->id]);
1785 * all css_sets of scgrp together in same order to dcgrp,
1786 * patch in-flight iterators to preserve correct iteration.
1787 * since the iterator is always advanced right away and
1788 * finished when it->cset_pos meets it->cset_head, so only
1789 * update it->cset_head is enough here.
1791 list_for_each_entry(it, &cset->task_iters, iters_node)
1792 if (it->cset_head == &scgrp->e_csets[ss->id])
1793 it->cset_head = &dcgrp->e_csets[ss->id];
1795 spin_unlock_irq(&css_set_lock);
1797 /* default hierarchy doesn't enable controllers by default */
1798 dst_root->subsys_mask |= 1 << ssid;
1799 if (dst_root == &cgrp_dfl_root) {
1800 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1802 dcgrp->subtree_control |= 1 << ssid;
1803 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1806 ret = cgroup_apply_control(dcgrp);
1808 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1813 } while_each_subsys_mask();
1815 kernfs_activate(dcgrp->kn);
1819 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1820 struct kernfs_root *kf_root)
1824 struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1825 struct cgroup *ns_cgroup;
1827 buf = kmalloc(PATH_MAX, GFP_KERNEL);
1831 spin_lock_irq(&css_set_lock);
1832 ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1833 len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1834 spin_unlock_irq(&css_set_lock);
1836 if (len >= PATH_MAX)
1839 seq_escape(sf, buf, " \t\n\\");
1846 enum cgroup2_param {
1848 Opt_memory_localevents,
1849 Opt_memory_recursiveprot,
1853 static const struct fs_parameter_spec cgroup2_fs_parameters[] = {
1854 fsparam_flag("nsdelegate", Opt_nsdelegate),
1855 fsparam_flag("memory_localevents", Opt_memory_localevents),
1856 fsparam_flag("memory_recursiveprot", Opt_memory_recursiveprot),
1860 static int cgroup2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1862 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1863 struct fs_parse_result result;
1866 opt = fs_parse(fc, cgroup2_fs_parameters, param, &result);
1871 case Opt_nsdelegate:
1872 ctx->flags |= CGRP_ROOT_NS_DELEGATE;
1874 case Opt_memory_localevents:
1875 ctx->flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1877 case Opt_memory_recursiveprot:
1878 ctx->flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1884 static void apply_cgroup_root_flags(unsigned int root_flags)
1886 if (current->nsproxy->cgroup_ns == &init_cgroup_ns) {
1887 if (root_flags & CGRP_ROOT_NS_DELEGATE)
1888 cgrp_dfl_root.flags |= CGRP_ROOT_NS_DELEGATE;
1890 cgrp_dfl_root.flags &= ~CGRP_ROOT_NS_DELEGATE;
1892 if (root_flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1893 cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1895 cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_LOCAL_EVENTS;
1897 if (root_flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1898 cgrp_dfl_root.flags |= CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1900 cgrp_dfl_root.flags &= ~CGRP_ROOT_MEMORY_RECURSIVE_PROT;
1904 static int cgroup_show_options(struct seq_file *seq, struct kernfs_root *kf_root)
1906 if (cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE)
1907 seq_puts(seq, ",nsdelegate");
1908 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_LOCAL_EVENTS)
1909 seq_puts(seq, ",memory_localevents");
1910 if (cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)
1911 seq_puts(seq, ",memory_recursiveprot");
1915 static int cgroup_reconfigure(struct fs_context *fc)
1917 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
1919 apply_cgroup_root_flags(ctx->flags);
1923 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1925 struct cgroup_subsys *ss;
1928 INIT_LIST_HEAD(&cgrp->self.sibling);
1929 INIT_LIST_HEAD(&cgrp->self.children);
1930 INIT_LIST_HEAD(&cgrp->cset_links);
1931 INIT_LIST_HEAD(&cgrp->pidlists);
1932 mutex_init(&cgrp->pidlist_mutex);
1933 cgrp->self.cgroup = cgrp;
1934 cgrp->self.flags |= CSS_ONLINE;
1935 cgrp->dom_cgrp = cgrp;
1936 cgrp->max_descendants = INT_MAX;
1937 cgrp->max_depth = INT_MAX;
1938 INIT_LIST_HEAD(&cgrp->rstat_css_list);
1939 prev_cputime_init(&cgrp->prev_cputime);
1941 for_each_subsys(ss, ssid)
1942 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1944 init_waitqueue_head(&cgrp->offline_waitq);
1945 INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1948 void init_cgroup_root(struct cgroup_fs_context *ctx)
1950 struct cgroup_root *root = ctx->root;
1951 struct cgroup *cgrp = &root->cgrp;
1953 INIT_LIST_HEAD(&root->root_list);
1954 atomic_set(&root->nr_cgrps, 1);
1956 init_cgroup_housekeeping(cgrp);
1958 root->flags = ctx->flags;
1959 if (ctx->release_agent)
1960 strscpy(root->release_agent_path, ctx->release_agent, PATH_MAX);
1962 strscpy(root->name, ctx->name, MAX_CGROUP_ROOT_NAMELEN);
1963 if (ctx->cpuset_clone_children)
1964 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1967 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1969 LIST_HEAD(tmp_links);
1970 struct cgroup *root_cgrp = &root->cgrp;
1971 struct kernfs_syscall_ops *kf_sops;
1972 struct css_set *cset;
1975 lockdep_assert_held(&cgroup_mutex);
1977 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release,
1983 * We're accessing css_set_count without locking css_set_lock here,
1984 * but that's OK - it can only be increased by someone holding
1985 * cgroup_lock, and that's us. Later rebinding may disable
1986 * controllers on the default hierarchy and thus create new csets,
1987 * which can't be more than the existing ones. Allocate 2x.
1989 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
1993 ret = cgroup_init_root_id(root);
1997 kf_sops = root == &cgrp_dfl_root ?
1998 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
2000 root->kf_root = kernfs_create_root(kf_sops,
2001 KERNFS_ROOT_CREATE_DEACTIVATED |
2002 KERNFS_ROOT_SUPPORT_EXPORTOP |
2003 KERNFS_ROOT_SUPPORT_USER_XATTR,
2005 if (IS_ERR(root->kf_root)) {
2006 ret = PTR_ERR(root->kf_root);
2009 root_cgrp->kn = root->kf_root->kn;
2010 WARN_ON_ONCE(cgroup_ino(root_cgrp) != 1);
2011 root_cgrp->ancestor_ids[0] = cgroup_id(root_cgrp);
2013 ret = css_populate_dir(&root_cgrp->self);
2017 ret = rebind_subsystems(root, ss_mask);
2021 ret = cgroup_bpf_inherit(root_cgrp);
2024 trace_cgroup_setup_root(root);
2027 * There must be no failure case after here, since rebinding takes
2028 * care of subsystems' refcounts, which are explicitly dropped in
2029 * the failure exit path.
2031 list_add(&root->root_list, &cgroup_roots);
2032 cgroup_root_count++;
2035 * Link the root cgroup in this hierarchy into all the css_set
2038 spin_lock_irq(&css_set_lock);
2039 hash_for_each(css_set_table, i, cset, hlist) {
2040 link_css_set(&tmp_links, cset, root_cgrp);
2041 if (css_set_populated(cset))
2042 cgroup_update_populated(root_cgrp, true);
2044 spin_unlock_irq(&css_set_lock);
2046 BUG_ON(!list_empty(&root_cgrp->self.children));
2047 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
2053 kernfs_destroy_root(root->kf_root);
2054 root->kf_root = NULL;
2056 cgroup_exit_root_id(root);
2058 percpu_ref_exit(&root_cgrp->self.refcnt);
2060 free_cgrp_cset_links(&tmp_links);
2064 int cgroup_do_get_tree(struct fs_context *fc)
2066 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2069 ctx->kfc.root = ctx->root->kf_root;
2070 if (fc->fs_type == &cgroup2_fs_type)
2071 ctx->kfc.magic = CGROUP2_SUPER_MAGIC;
2073 ctx->kfc.magic = CGROUP_SUPER_MAGIC;
2074 ret = kernfs_get_tree(fc);
2077 * In non-init cgroup namespace, instead of root cgroup's dentry,
2078 * we return the dentry corresponding to the cgroupns->root_cgrp.
2080 if (!ret && ctx->ns != &init_cgroup_ns) {
2081 struct dentry *nsdentry;
2082 struct super_block *sb = fc->root->d_sb;
2083 struct cgroup *cgrp;
2085 mutex_lock(&cgroup_mutex);
2086 spin_lock_irq(&css_set_lock);
2088 cgrp = cset_cgroup_from_root(ctx->ns->root_cset, ctx->root);
2090 spin_unlock_irq(&css_set_lock);
2091 mutex_unlock(&cgroup_mutex);
2093 nsdentry = kernfs_node_dentry(cgrp->kn, sb);
2095 if (IS_ERR(nsdentry)) {
2096 deactivate_locked_super(sb);
2097 ret = PTR_ERR(nsdentry);
2100 fc->root = nsdentry;
2103 if (!ctx->kfc.new_sb_created)
2104 cgroup_put(&ctx->root->cgrp);
2110 * Destroy a cgroup filesystem context.
2112 static void cgroup_fs_context_free(struct fs_context *fc)
2114 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2117 kfree(ctx->release_agent);
2118 put_cgroup_ns(ctx->ns);
2119 kernfs_free_fs_context(fc);
2123 static int cgroup_get_tree(struct fs_context *fc)
2125 struct cgroup_fs_context *ctx = cgroup_fc2context(fc);
2128 cgrp_dfl_visible = true;
2129 cgroup_get_live(&cgrp_dfl_root.cgrp);
2130 ctx->root = &cgrp_dfl_root;
2132 ret = cgroup_do_get_tree(fc);
2134 apply_cgroup_root_flags(ctx->flags);
2138 static const struct fs_context_operations cgroup_fs_context_ops = {
2139 .free = cgroup_fs_context_free,
2140 .parse_param = cgroup2_parse_param,
2141 .get_tree = cgroup_get_tree,
2142 .reconfigure = cgroup_reconfigure,
2145 static const struct fs_context_operations cgroup1_fs_context_ops = {
2146 .free = cgroup_fs_context_free,
2147 .parse_param = cgroup1_parse_param,
2148 .get_tree = cgroup1_get_tree,
2149 .reconfigure = cgroup1_reconfigure,
2153 * Initialise the cgroup filesystem creation/reconfiguration context. Notably,
2154 * we select the namespace we're going to use.
2156 static int cgroup_init_fs_context(struct fs_context *fc)
2158 struct cgroup_fs_context *ctx;
2160 ctx = kzalloc(sizeof(struct cgroup_fs_context), GFP_KERNEL);
2164 ctx->ns = current->nsproxy->cgroup_ns;
2165 get_cgroup_ns(ctx->ns);
2166 fc->fs_private = &ctx->kfc;
2167 if (fc->fs_type == &cgroup2_fs_type)
2168 fc->ops = &cgroup_fs_context_ops;
2170 fc->ops = &cgroup1_fs_context_ops;
2171 put_user_ns(fc->user_ns);
2172 fc->user_ns = get_user_ns(ctx->ns->user_ns);
2177 static void cgroup_kill_sb(struct super_block *sb)
2179 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2180 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2183 * If @root doesn't have any children, start killing it.
2184 * This prevents new mounts by disabling percpu_ref_tryget_live().
2185 * cgroup_mount() may wait for @root's release.
2187 * And don't kill the default root.
2189 if (list_empty(&root->cgrp.self.children) && root != &cgrp_dfl_root &&
2190 !percpu_ref_is_dying(&root->cgrp.self.refcnt)) {
2191 cgroup_bpf_offline(&root->cgrp);
2192 percpu_ref_kill(&root->cgrp.self.refcnt);
2194 cgroup_put(&root->cgrp);
2198 struct file_system_type cgroup_fs_type = {
2200 .init_fs_context = cgroup_init_fs_context,
2201 .parameters = cgroup1_fs_parameters,
2202 .kill_sb = cgroup_kill_sb,
2203 .fs_flags = FS_USERNS_MOUNT,
2206 static struct file_system_type cgroup2_fs_type = {
2208 .init_fs_context = cgroup_init_fs_context,
2209 .parameters = cgroup2_fs_parameters,
2210 .kill_sb = cgroup_kill_sb,
2211 .fs_flags = FS_USERNS_MOUNT,
2214 #ifdef CONFIG_CPUSETS
2215 static const struct fs_context_operations cpuset_fs_context_ops = {
2216 .get_tree = cgroup1_get_tree,
2217 .free = cgroup_fs_context_free,
2221 * This is ugly, but preserves the userspace API for existing cpuset
2222 * users. If someone tries to mount the "cpuset" filesystem, we
2223 * silently switch it to mount "cgroup" instead
2225 static int cpuset_init_fs_context(struct fs_context *fc)
2227 char *agent = kstrdup("/sbin/cpuset_release_agent", GFP_USER);
2228 struct cgroup_fs_context *ctx;
2231 err = cgroup_init_fs_context(fc);
2237 fc->ops = &cpuset_fs_context_ops;
2239 ctx = cgroup_fc2context(fc);
2240 ctx->subsys_mask = 1 << cpuset_cgrp_id;
2241 ctx->flags |= CGRP_ROOT_NOPREFIX;
2242 ctx->release_agent = agent;
2244 get_filesystem(&cgroup_fs_type);
2245 put_filesystem(fc->fs_type);
2246 fc->fs_type = &cgroup_fs_type;
2251 static struct file_system_type cpuset_fs_type = {
2253 .init_fs_context = cpuset_init_fs_context,
2254 .fs_flags = FS_USERNS_MOUNT,
2258 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
2259 struct cgroup_namespace *ns)
2261 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
2263 return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
2266 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
2267 struct cgroup_namespace *ns)
2271 mutex_lock(&cgroup_mutex);
2272 spin_lock_irq(&css_set_lock);
2274 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
2276 spin_unlock_irq(&css_set_lock);
2277 mutex_unlock(&cgroup_mutex);
2281 EXPORT_SYMBOL_GPL(cgroup_path_ns);
2284 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2285 * @task: target task
2286 * @buf: the buffer to write the path into
2287 * @buflen: the length of the buffer
2289 * Determine @task's cgroup on the first (the one with the lowest non-zero
2290 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2291 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2292 * cgroup controller callbacks.
2294 * Return value is the same as kernfs_path().
2296 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2298 struct cgroup_root *root;
2299 struct cgroup *cgrp;
2300 int hierarchy_id = 1;
2303 mutex_lock(&cgroup_mutex);
2304 spin_lock_irq(&css_set_lock);
2306 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2309 cgrp = task_cgroup_from_root(task, root);
2310 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
2312 /* if no hierarchy exists, everyone is in "/" */
2313 ret = strlcpy(buf, "/", buflen);
2316 spin_unlock_irq(&css_set_lock);
2317 mutex_unlock(&cgroup_mutex);
2320 EXPORT_SYMBOL_GPL(task_cgroup_path);
2323 * cgroup_attach_lock - Lock for ->attach()
2324 * @lock_threadgroup: whether to down_write cgroup_threadgroup_rwsem
2326 * cgroup migration sometimes needs to stabilize threadgroups against forks and
2327 * exits by write-locking cgroup_threadgroup_rwsem. However, some ->attach()
2328 * implementations (e.g. cpuset), also need to disable CPU hotplug.
2329 * Unfortunately, letting ->attach() operations acquire cpus_read_lock() can
2330 * lead to deadlocks.
2332 * Bringing up a CPU may involve creating and destroying tasks which requires
2333 * read-locking threadgroup_rwsem, so threadgroup_rwsem nests inside
2334 * cpus_read_lock(). If we call an ->attach() which acquires the cpus lock while
2335 * write-locking threadgroup_rwsem, the locking order is reversed and we end up
2336 * waiting for an on-going CPU hotplug operation which in turn is waiting for
2337 * the threadgroup_rwsem to be released to create new tasks. For more details:
2339 * http://lkml.kernel.org/r/20220711174629.uehfmqegcwn2lqzu@wubuntu
2341 * Resolve the situation by always acquiring cpus_read_lock() before optionally
2342 * write-locking cgroup_threadgroup_rwsem. This allows ->attach() to assume that
2343 * CPU hotplug is disabled on entry.
2345 static void cgroup_attach_lock(bool lock_threadgroup)
2348 if (lock_threadgroup)
2349 percpu_down_write(&cgroup_threadgroup_rwsem);
2353 * cgroup_attach_unlock - Undo cgroup_attach_lock()
2354 * @lock_threadgroup: whether to up_write cgroup_threadgroup_rwsem
2356 static void cgroup_attach_unlock(bool lock_threadgroup)
2358 if (lock_threadgroup)
2359 percpu_up_write(&cgroup_threadgroup_rwsem);
2364 * cgroup_migrate_add_task - add a migration target task to a migration context
2365 * @task: target task
2366 * @mgctx: target migration context
2368 * Add @task, which is a migration target, to @mgctx->tset. This function
2369 * becomes noop if @task doesn't need to be migrated. @task's css_set
2370 * should have been added as a migration source and @task->cg_list will be
2371 * moved from the css_set's tasks list to mg_tasks one.
2373 static void cgroup_migrate_add_task(struct task_struct *task,
2374 struct cgroup_mgctx *mgctx)
2376 struct css_set *cset;
2378 lockdep_assert_held(&css_set_lock);
2380 /* @task either already exited or can't exit until the end */
2381 if (task->flags & PF_EXITING)
2384 /* cgroup_threadgroup_rwsem protects racing against forks */
2385 WARN_ON_ONCE(list_empty(&task->cg_list));
2387 cset = task_css_set(task);
2388 if (!cset->mg_src_cgrp)
2391 mgctx->tset.nr_tasks++;
2393 list_move_tail(&task->cg_list, &cset->mg_tasks);
2394 if (list_empty(&cset->mg_node))
2395 list_add_tail(&cset->mg_node,
2396 &mgctx->tset.src_csets);
2397 if (list_empty(&cset->mg_dst_cset->mg_node))
2398 list_add_tail(&cset->mg_dst_cset->mg_node,
2399 &mgctx->tset.dst_csets);
2403 * cgroup_taskset_first - reset taskset and return the first task
2404 * @tset: taskset of interest
2405 * @dst_cssp: output variable for the destination css
2407 * @tset iteration is initialized and the first task is returned.
2409 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2410 struct cgroup_subsys_state **dst_cssp)
2412 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2413 tset->cur_task = NULL;
2415 return cgroup_taskset_next(tset, dst_cssp);
2419 * cgroup_taskset_next - iterate to the next task in taskset
2420 * @tset: taskset of interest
2421 * @dst_cssp: output variable for the destination css
2423 * Return the next task in @tset. Iteration must have been initialized
2424 * with cgroup_taskset_first().
2426 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2427 struct cgroup_subsys_state **dst_cssp)
2429 struct css_set *cset = tset->cur_cset;
2430 struct task_struct *task = tset->cur_task;
2432 while (&cset->mg_node != tset->csets) {
2434 task = list_first_entry(&cset->mg_tasks,
2435 struct task_struct, cg_list);
2437 task = list_next_entry(task, cg_list);
2439 if (&task->cg_list != &cset->mg_tasks) {
2440 tset->cur_cset = cset;
2441 tset->cur_task = task;
2444 * This function may be called both before and
2445 * after cgroup_taskset_migrate(). The two cases
2446 * can be distinguished by looking at whether @cset
2447 * has its ->mg_dst_cset set.
2449 if (cset->mg_dst_cset)
2450 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2452 *dst_cssp = cset->subsys[tset->ssid];
2457 cset = list_next_entry(cset, mg_node);
2465 * cgroup_taskset_migrate - migrate a taskset
2466 * @mgctx: migration context
2468 * Migrate tasks in @mgctx as setup by migration preparation functions.
2469 * This function fails iff one of the ->can_attach callbacks fails and
2470 * guarantees that either all or none of the tasks in @mgctx are migrated.
2471 * @mgctx is consumed regardless of success.
2473 static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2475 struct cgroup_taskset *tset = &mgctx->tset;
2476 struct cgroup_subsys *ss;
2477 struct task_struct *task, *tmp_task;
2478 struct css_set *cset, *tmp_cset;
2479 int ssid, failed_ssid, ret;
2481 /* check that we can legitimately attach to the cgroup */
2482 if (tset->nr_tasks) {
2483 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2484 if (ss->can_attach) {
2486 ret = ss->can_attach(tset);
2489 goto out_cancel_attach;
2492 } while_each_subsys_mask();
2496 * Now that we're guaranteed success, proceed to move all tasks to
2497 * the new cgroup. There are no failure cases after here, so this
2498 * is the commit point.
2500 spin_lock_irq(&css_set_lock);
2501 list_for_each_entry(cset, &tset->src_csets, mg_node) {
2502 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2503 struct css_set *from_cset = task_css_set(task);
2504 struct css_set *to_cset = cset->mg_dst_cset;
2506 get_css_set(to_cset);
2507 to_cset->nr_tasks++;
2508 css_set_move_task(task, from_cset, to_cset, true);
2509 from_cset->nr_tasks--;
2511 * If the source or destination cgroup is frozen,
2512 * the task might require to change its state.
2514 cgroup_freezer_migrate_task(task, from_cset->dfl_cgrp,
2516 put_css_set_locked(from_cset);
2520 spin_unlock_irq(&css_set_lock);
2523 * Migration is committed, all target tasks are now on dst_csets.
2524 * Nothing is sensitive to fork() after this point. Notify
2525 * controllers that migration is complete.
2527 tset->csets = &tset->dst_csets;
2529 if (tset->nr_tasks) {
2530 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2535 } while_each_subsys_mask();
2539 goto out_release_tset;
2542 if (tset->nr_tasks) {
2543 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2544 if (ssid == failed_ssid)
2546 if (ss->cancel_attach) {
2548 ss->cancel_attach(tset);
2550 } while_each_subsys_mask();
2553 spin_lock_irq(&css_set_lock);
2554 list_splice_init(&tset->dst_csets, &tset->src_csets);
2555 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2556 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2557 list_del_init(&cset->mg_node);
2559 spin_unlock_irq(&css_set_lock);
2562 * Re-initialize the cgroup_taskset structure in case it is reused
2563 * again in another cgroup_migrate_add_task()/cgroup_migrate_execute()
2567 tset->csets = &tset->src_csets;
2572 * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination
2573 * @dst_cgrp: destination cgroup to test
2575 * On the default hierarchy, except for the mixable, (possible) thread root
2576 * and threaded cgroups, subtree_control must be zero for migration
2577 * destination cgroups with tasks so that child cgroups don't compete
2580 int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)
2582 /* v1 doesn't have any restriction */
2583 if (!cgroup_on_dfl(dst_cgrp))
2586 /* verify @dst_cgrp can host resources */
2587 if (!cgroup_is_valid_domain(dst_cgrp->dom_cgrp))
2590 /* mixables don't care */
2591 if (cgroup_is_mixable(dst_cgrp))
2595 * If @dst_cgrp is already or can become a thread root or is
2596 * threaded, it doesn't matter.
2598 if (cgroup_can_be_thread_root(dst_cgrp) || cgroup_is_threaded(dst_cgrp))
2601 /* apply no-internal-process constraint */
2602 if (dst_cgrp->subtree_control)
2609 * cgroup_migrate_finish - cleanup after attach
2610 * @mgctx: migration context
2612 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2613 * those functions for details.
2615 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2617 struct css_set *cset, *tmp_cset;
2619 lockdep_assert_held(&cgroup_mutex);
2621 spin_lock_irq(&css_set_lock);
2623 list_for_each_entry_safe(cset, tmp_cset, &mgctx->preloaded_src_csets,
2624 mg_src_preload_node) {
2625 cset->mg_src_cgrp = NULL;
2626 cset->mg_dst_cgrp = NULL;
2627 cset->mg_dst_cset = NULL;
2628 list_del_init(&cset->mg_src_preload_node);
2629 put_css_set_locked(cset);
2632 list_for_each_entry_safe(cset, tmp_cset, &mgctx->preloaded_dst_csets,
2633 mg_dst_preload_node) {
2634 cset->mg_src_cgrp = NULL;
2635 cset->mg_dst_cgrp = NULL;
2636 cset->mg_dst_cset = NULL;
2637 list_del_init(&cset->mg_dst_preload_node);
2638 put_css_set_locked(cset);
2641 spin_unlock_irq(&css_set_lock);
2645 * cgroup_migrate_add_src - add a migration source css_set
2646 * @src_cset: the source css_set to add
2647 * @dst_cgrp: the destination cgroup
2648 * @mgctx: migration context
2650 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2651 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2652 * up by cgroup_migrate_finish().
2654 * This function may be called without holding cgroup_threadgroup_rwsem
2655 * even if the target is a process. Threads may be created and destroyed
2656 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2657 * into play and the preloaded css_sets are guaranteed to cover all
2660 void cgroup_migrate_add_src(struct css_set *src_cset,
2661 struct cgroup *dst_cgrp,
2662 struct cgroup_mgctx *mgctx)
2664 struct cgroup *src_cgrp;
2666 lockdep_assert_held(&cgroup_mutex);
2667 lockdep_assert_held(&css_set_lock);
2670 * If ->dead, @src_set is associated with one or more dead cgroups
2671 * and doesn't contain any migratable tasks. Ignore it early so
2672 * that the rest of migration path doesn't get confused by it.
2677 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2679 if (!list_empty(&src_cset->mg_src_preload_node))
2682 WARN_ON(src_cset->mg_src_cgrp);
2683 WARN_ON(src_cset->mg_dst_cgrp);
2684 WARN_ON(!list_empty(&src_cset->mg_tasks));
2685 WARN_ON(!list_empty(&src_cset->mg_node));
2687 src_cset->mg_src_cgrp = src_cgrp;
2688 src_cset->mg_dst_cgrp = dst_cgrp;
2689 get_css_set(src_cset);
2690 list_add_tail(&src_cset->mg_src_preload_node, &mgctx->preloaded_src_csets);
2694 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2695 * @mgctx: migration context
2697 * Tasks are about to be moved and all the source css_sets have been
2698 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2699 * pins all destination css_sets, links each to its source, and append them
2700 * to @mgctx->preloaded_dst_csets.
2702 * This function must be called after cgroup_migrate_add_src() has been
2703 * called on each migration source css_set. After migration is performed
2704 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2707 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2709 struct css_set *src_cset, *tmp_cset;
2711 lockdep_assert_held(&cgroup_mutex);
2713 /* look up the dst cset for each src cset and link it to src */
2714 list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2715 mg_src_preload_node) {
2716 struct css_set *dst_cset;
2717 struct cgroup_subsys *ss;
2720 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2724 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2727 * If src cset equals dst, it's noop. Drop the src.
2728 * cgroup_migrate() will skip the cset too. Note that we
2729 * can't handle src == dst as some nodes are used by both.
2731 if (src_cset == dst_cset) {
2732 src_cset->mg_src_cgrp = NULL;
2733 src_cset->mg_dst_cgrp = NULL;
2734 list_del_init(&src_cset->mg_src_preload_node);
2735 put_css_set(src_cset);
2736 put_css_set(dst_cset);
2740 src_cset->mg_dst_cset = dst_cset;
2742 if (list_empty(&dst_cset->mg_dst_preload_node))
2743 list_add_tail(&dst_cset->mg_dst_preload_node,
2744 &mgctx->preloaded_dst_csets);
2746 put_css_set(dst_cset);
2748 for_each_subsys(ss, ssid)
2749 if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2750 mgctx->ss_mask |= 1 << ssid;
2757 * cgroup_migrate - migrate a process or task to a cgroup
2758 * @leader: the leader of the process or the task to migrate
2759 * @threadgroup: whether @leader points to the whole process or a single task
2760 * @mgctx: migration context
2762 * Migrate a process or task denoted by @leader. If migrating a process,
2763 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2764 * responsible for invoking cgroup_migrate_add_src() and
2765 * cgroup_migrate_prepare_dst() on the targets before invoking this
2766 * function and following up with cgroup_migrate_finish().
2768 * As long as a controller's ->can_attach() doesn't fail, this function is
2769 * guaranteed to succeed. This means that, excluding ->can_attach()
2770 * failure, when migrating multiple targets, the success or failure can be
2771 * decided for all targets by invoking group_migrate_prepare_dst() before
2772 * actually starting migrating.
2774 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2775 struct cgroup_mgctx *mgctx)
2777 struct task_struct *task;
2780 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2781 * already PF_EXITING could be freed from underneath us unless we
2782 * take an rcu_read_lock.
2784 spin_lock_irq(&css_set_lock);
2788 cgroup_migrate_add_task(task, mgctx);
2791 } while_each_thread(leader, task);
2793 spin_unlock_irq(&css_set_lock);
2795 return cgroup_migrate_execute(mgctx);
2799 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2800 * @dst_cgrp: the cgroup to attach to
2801 * @leader: the task or the leader of the threadgroup to be attached
2802 * @threadgroup: attach the whole threadgroup?
2804 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2806 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2809 DEFINE_CGROUP_MGCTX(mgctx);
2810 struct task_struct *task;
2813 /* look up all src csets */
2814 spin_lock_irq(&css_set_lock);
2818 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2821 } while_each_thread(leader, task);
2823 spin_unlock_irq(&css_set_lock);
2825 /* prepare dst csets and commit */
2826 ret = cgroup_migrate_prepare_dst(&mgctx);
2828 ret = cgroup_migrate(leader, threadgroup, &mgctx);
2830 cgroup_migrate_finish(&mgctx);
2833 TRACE_CGROUP_PATH(attach_task, dst_cgrp, leader, threadgroup);
2838 struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
2839 bool *threadgroup_locked)
2841 struct task_struct *tsk;
2844 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2845 return ERR_PTR(-EINVAL);
2848 * If we migrate a single thread, we don't care about threadgroup
2849 * stability. If the thread is `current`, it won't exit(2) under our
2850 * hands or change PID through exec(2). We exclude
2851 * cgroup_update_dfl_csses and other cgroup_{proc,thread}s_write
2852 * callers by cgroup_mutex.
2853 * Therefore, we can skip the global lock.
2855 lockdep_assert_held(&cgroup_mutex);
2856 *threadgroup_locked = pid || threadgroup;
2857 cgroup_attach_lock(*threadgroup_locked);
2861 tsk = find_task_by_vpid(pid);
2863 tsk = ERR_PTR(-ESRCH);
2864 goto out_unlock_threadgroup;
2871 tsk = tsk->group_leader;
2874 * kthreads may acquire PF_NO_SETAFFINITY during initialization.
2875 * If userland migrates such a kthread to a non-root cgroup, it can
2876 * become trapped in a cpuset, or RT kthread may be born in a
2877 * cgroup with no rt_runtime allocated. Just say no.
2879 if (tsk->no_cgroup_migration || (tsk->flags & PF_NO_SETAFFINITY)) {
2880 tsk = ERR_PTR(-EINVAL);
2881 goto out_unlock_threadgroup;
2884 get_task_struct(tsk);
2885 goto out_unlock_rcu;
2887 out_unlock_threadgroup:
2888 cgroup_attach_unlock(*threadgroup_locked);
2889 *threadgroup_locked = false;
2895 void cgroup_procs_write_finish(struct task_struct *task, bool threadgroup_locked)
2897 struct cgroup_subsys *ss;
2900 /* release reference from cgroup_procs_write_start() */
2901 put_task_struct(task);
2903 cgroup_attach_unlock(threadgroup_locked);
2905 for_each_subsys(ss, ssid)
2906 if (ss->post_attach)
2910 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2912 struct cgroup_subsys *ss;
2913 bool printed = false;
2916 do_each_subsys_mask(ss, ssid, ss_mask) {
2919 seq_puts(seq, ss->name);
2921 } while_each_subsys_mask();
2923 seq_putc(seq, '\n');
2926 /* show controllers which are enabled from the parent */
2927 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2929 struct cgroup *cgrp = seq_css(seq)->cgroup;
2931 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2935 /* show controllers which are enabled for a given cgroup's children */
2936 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2938 struct cgroup *cgrp = seq_css(seq)->cgroup;
2940 cgroup_print_ss_mask(seq, cgrp->subtree_control);
2945 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2946 * @cgrp: root of the subtree to update csses for
2948 * @cgrp's control masks have changed and its subtree's css associations
2949 * need to be updated accordingly. This function looks up all css_sets
2950 * which are attached to the subtree, creates the matching updated css_sets
2951 * and migrates the tasks to the new ones.
2953 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2955 DEFINE_CGROUP_MGCTX(mgctx);
2956 struct cgroup_subsys_state *d_css;
2957 struct cgroup *dsct;
2958 struct css_set *src_cset;
2962 lockdep_assert_held(&cgroup_mutex);
2964 /* look up all csses currently attached to @cgrp's subtree */
2965 spin_lock_irq(&css_set_lock);
2966 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2967 struct cgrp_cset_link *link;
2969 list_for_each_entry(link, &dsct->cset_links, cset_link)
2970 cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2972 spin_unlock_irq(&css_set_lock);
2975 * We need to write-lock threadgroup_rwsem while migrating tasks.
2976 * However, if there are no source csets for @cgrp, changing its
2977 * controllers isn't gonna produce any task migrations and the
2978 * write-locking can be skipped safely.
2980 has_tasks = !list_empty(&mgctx.preloaded_src_csets);
2981 cgroup_attach_lock(has_tasks);
2983 /* NULL dst indicates self on default hierarchy */
2984 ret = cgroup_migrate_prepare_dst(&mgctx);
2988 spin_lock_irq(&css_set_lock);
2989 list_for_each_entry(src_cset, &mgctx.preloaded_src_csets,
2990 mg_src_preload_node) {
2991 struct task_struct *task, *ntask;
2993 /* all tasks in src_csets need to be migrated */
2994 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2995 cgroup_migrate_add_task(task, &mgctx);
2997 spin_unlock_irq(&css_set_lock);
2999 ret = cgroup_migrate_execute(&mgctx);
3001 cgroup_migrate_finish(&mgctx);
3002 cgroup_attach_unlock(has_tasks);
3007 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
3008 * @cgrp: root of the target subtree
3010 * Because css offlining is asynchronous, userland may try to re-enable a
3011 * controller while the previous css is still around. This function grabs
3012 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
3014 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
3015 __acquires(&cgroup_mutex)
3017 struct cgroup *dsct;
3018 struct cgroup_subsys_state *d_css;
3019 struct cgroup_subsys *ss;
3023 mutex_lock(&cgroup_mutex);
3025 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3026 for_each_subsys(ss, ssid) {
3027 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3030 if (!css || !percpu_ref_is_dying(&css->refcnt))
3033 cgroup_get_live(dsct);
3034 prepare_to_wait(&dsct->offline_waitq, &wait,
3035 TASK_UNINTERRUPTIBLE);
3037 mutex_unlock(&cgroup_mutex);
3039 finish_wait(&dsct->offline_waitq, &wait);
3048 * cgroup_save_control - save control masks and dom_cgrp of a subtree
3049 * @cgrp: root of the target subtree
3051 * Save ->subtree_control, ->subtree_ss_mask and ->dom_cgrp to the
3052 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3055 static void cgroup_save_control(struct cgroup *cgrp)
3057 struct cgroup *dsct;
3058 struct cgroup_subsys_state *d_css;
3060 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3061 dsct->old_subtree_control = dsct->subtree_control;
3062 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
3063 dsct->old_dom_cgrp = dsct->dom_cgrp;
3068 * cgroup_propagate_control - refresh control masks of a subtree
3069 * @cgrp: root of the target subtree
3071 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
3072 * ->subtree_control and propagate controller availability through the
3073 * subtree so that descendants don't have unavailable controllers enabled.
3075 static void cgroup_propagate_control(struct cgroup *cgrp)
3077 struct cgroup *dsct;
3078 struct cgroup_subsys_state *d_css;
3080 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3081 dsct->subtree_control &= cgroup_control(dsct);
3082 dsct->subtree_ss_mask =
3083 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
3084 cgroup_ss_mask(dsct));
3089 * cgroup_restore_control - restore control masks and dom_cgrp of a subtree
3090 * @cgrp: root of the target subtree
3092 * Restore ->subtree_control, ->subtree_ss_mask and ->dom_cgrp from the
3093 * respective old_ prefixed fields for @cgrp's subtree including @cgrp
3096 static void cgroup_restore_control(struct cgroup *cgrp)
3098 struct cgroup *dsct;
3099 struct cgroup_subsys_state *d_css;
3101 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3102 dsct->subtree_control = dsct->old_subtree_control;
3103 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
3104 dsct->dom_cgrp = dsct->old_dom_cgrp;
3108 static bool css_visible(struct cgroup_subsys_state *css)
3110 struct cgroup_subsys *ss = css->ss;
3111 struct cgroup *cgrp = css->cgroup;
3113 if (cgroup_control(cgrp) & (1 << ss->id))
3115 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
3117 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
3121 * cgroup_apply_control_enable - enable or show csses according to control
3122 * @cgrp: root of the target subtree
3124 * Walk @cgrp's subtree and create new csses or make the existing ones
3125 * visible. A css is created invisible if it's being implicitly enabled
3126 * through dependency. An invisible css is made visible when the userland
3127 * explicitly enables it.
3129 * Returns 0 on success, -errno on failure. On failure, csses which have
3130 * been processed already aren't cleaned up. The caller is responsible for
3131 * cleaning up with cgroup_apply_control_disable().
3133 static int cgroup_apply_control_enable(struct cgroup *cgrp)
3135 struct cgroup *dsct;
3136 struct cgroup_subsys_state *d_css;
3137 struct cgroup_subsys *ss;
3140 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
3141 for_each_subsys(ss, ssid) {
3142 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3144 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
3148 css = css_create(dsct, ss);
3150 return PTR_ERR(css);
3153 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3155 if (css_visible(css)) {
3156 ret = css_populate_dir(css);
3167 * cgroup_apply_control_disable - kill or hide csses according to control
3168 * @cgrp: root of the target subtree
3170 * Walk @cgrp's subtree and kill and hide csses so that they match
3171 * cgroup_ss_mask() and cgroup_visible_mask().
3173 * A css is hidden when the userland requests it to be disabled while other
3174 * subsystems are still depending on it. The css must not actively control
3175 * resources and be in the vanilla state if it's made visible again later.
3176 * Controllers which may be depended upon should provide ->css_reset() for
3179 static void cgroup_apply_control_disable(struct cgroup *cgrp)
3181 struct cgroup *dsct;
3182 struct cgroup_subsys_state *d_css;
3183 struct cgroup_subsys *ss;
3186 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
3187 for_each_subsys(ss, ssid) {
3188 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
3193 WARN_ON_ONCE(percpu_ref_is_dying(&css->refcnt));
3196 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
3198 } else if (!css_visible(css)) {
3208 * cgroup_apply_control - apply control mask updates to the subtree
3209 * @cgrp: root of the target subtree
3211 * subsystems can be enabled and disabled in a subtree using the following
3214 * 1. Call cgroup_save_control() to stash the current state.
3215 * 2. Update ->subtree_control masks in the subtree as desired.
3216 * 3. Call cgroup_apply_control() to apply the changes.
3217 * 4. Optionally perform other related operations.
3218 * 5. Call cgroup_finalize_control() to finish up.
3220 * This function implements step 3 and propagates the mask changes
3221 * throughout @cgrp's subtree, updates csses accordingly and perform
3222 * process migrations.
3224 static int cgroup_apply_control(struct cgroup *cgrp)
3228 cgroup_propagate_control(cgrp);
3230 ret = cgroup_apply_control_enable(cgrp);
3235 * At this point, cgroup_e_css_by_mask() results reflect the new csses
3236 * making the following cgroup_update_dfl_csses() properly update
3237 * css associations of all tasks in the subtree.
3239 ret = cgroup_update_dfl_csses(cgrp);
3247 * cgroup_finalize_control - finalize control mask update
3248 * @cgrp: root of the target subtree
3249 * @ret: the result of the update
3251 * Finalize control mask update. See cgroup_apply_control() for more info.
3253 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
3256 cgroup_restore_control(cgrp);
3257 cgroup_propagate_control(cgrp);
3260 cgroup_apply_control_disable(cgrp);
3263 static int cgroup_vet_subtree_control_enable(struct cgroup *cgrp, u16 enable)
3265 u16 domain_enable = enable & ~cgrp_dfl_threaded_ss_mask;
3267 /* if nothing is getting enabled, nothing to worry about */
3271 /* can @cgrp host any resources? */
3272 if (!cgroup_is_valid_domain(cgrp->dom_cgrp))
3275 /* mixables don't care */
3276 if (cgroup_is_mixable(cgrp))
3279 if (domain_enable) {
3280 /* can't enable domain controllers inside a thread subtree */
3281 if (cgroup_is_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3285 * Threaded controllers can handle internal competitions
3286 * and are always allowed inside a (prospective) thread
3289 if (cgroup_can_be_thread_root(cgrp) || cgroup_is_threaded(cgrp))
3294 * Controllers can't be enabled for a cgroup with tasks to avoid
3295 * child cgroups competing against tasks.
3297 if (cgroup_has_tasks(cgrp))
3303 /* change the enabled child controllers for a cgroup in the default hierarchy */
3304 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
3305 char *buf, size_t nbytes,
3308 u16 enable = 0, disable = 0;
3309 struct cgroup *cgrp, *child;
3310 struct cgroup_subsys *ss;
3315 * Parse input - space separated list of subsystem names prefixed
3316 * with either + or -.
3318 buf = strstrip(buf);
3319 while ((tok = strsep(&buf, " "))) {
3322 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
3323 if (!cgroup_ssid_enabled(ssid) ||
3324 strcmp(tok + 1, ss->name))
3328 enable |= 1 << ssid;
3329 disable &= ~(1 << ssid);
3330 } else if (*tok == '-') {
3331 disable |= 1 << ssid;
3332 enable &= ~(1 << ssid);
3337 } while_each_subsys_mask();
3338 if (ssid == CGROUP_SUBSYS_COUNT)
3342 cgrp = cgroup_kn_lock_live(of->kn, true);
3346 for_each_subsys(ss, ssid) {
3347 if (enable & (1 << ssid)) {
3348 if (cgrp->subtree_control & (1 << ssid)) {
3349 enable &= ~(1 << ssid);
3353 if (!(cgroup_control(cgrp) & (1 << ssid))) {
3357 } else if (disable & (1 << ssid)) {
3358 if (!(cgrp->subtree_control & (1 << ssid))) {
3359 disable &= ~(1 << ssid);
3363 /* a child has it enabled? */
3364 cgroup_for_each_live_child(child, cgrp) {
3365 if (child->subtree_control & (1 << ssid)) {
3373 if (!enable && !disable) {
3378 ret = cgroup_vet_subtree_control_enable(cgrp, enable);
3382 /* save and update control masks and prepare csses */
3383 cgroup_save_control(cgrp);
3385 cgrp->subtree_control |= enable;
3386 cgrp->subtree_control &= ~disable;
3388 ret = cgroup_apply_control(cgrp);
3389 cgroup_finalize_control(cgrp, ret);
3393 kernfs_activate(cgrp->kn);
3395 cgroup_kn_unlock(of->kn);
3396 return ret ?: nbytes;
3400 * cgroup_enable_threaded - make @cgrp threaded
3401 * @cgrp: the target cgroup
3403 * Called when "threaded" is written to the cgroup.type interface file and
3404 * tries to make @cgrp threaded and join the parent's resource domain.
3405 * This function is never called on the root cgroup as cgroup.type doesn't
3408 static int cgroup_enable_threaded(struct cgroup *cgrp)
3410 struct cgroup *parent = cgroup_parent(cgrp);
3411 struct cgroup *dom_cgrp = parent->dom_cgrp;
3412 struct cgroup *dsct;
3413 struct cgroup_subsys_state *d_css;
3416 lockdep_assert_held(&cgroup_mutex);
3418 /* noop if already threaded */
3419 if (cgroup_is_threaded(cgrp))
3423 * If @cgroup is populated or has domain controllers enabled, it
3424 * can't be switched. While the below cgroup_can_be_thread_root()
3425 * test can catch the same conditions, that's only when @parent is
3426 * not mixable, so let's check it explicitly.
3428 if (cgroup_is_populated(cgrp) ||
3429 cgrp->subtree_control & ~cgrp_dfl_threaded_ss_mask)
3432 /* we're joining the parent's domain, ensure its validity */
3433 if (!cgroup_is_valid_domain(dom_cgrp) ||
3434 !cgroup_can_be_thread_root(dom_cgrp))
3438 * The following shouldn't cause actual migrations and should
3441 cgroup_save_control(cgrp);
3443 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp)
3444 if (dsct == cgrp || cgroup_is_threaded(dsct))
3445 dsct->dom_cgrp = dom_cgrp;
3447 ret = cgroup_apply_control(cgrp);
3449 parent->nr_threaded_children++;
3451 cgroup_finalize_control(cgrp, ret);
3455 static int cgroup_type_show(struct seq_file *seq, void *v)
3457 struct cgroup *cgrp = seq_css(seq)->cgroup;
3459 if (cgroup_is_threaded(cgrp))
3460 seq_puts(seq, "threaded\n");
3461 else if (!cgroup_is_valid_domain(cgrp))
3462 seq_puts(seq, "domain invalid\n");
3463 else if (cgroup_is_thread_root(cgrp))
3464 seq_puts(seq, "domain threaded\n");
3466 seq_puts(seq, "domain\n");
3471 static ssize_t cgroup_type_write(struct kernfs_open_file *of, char *buf,
3472 size_t nbytes, loff_t off)
3474 struct cgroup *cgrp;
3477 /* only switching to threaded mode is supported */
3478 if (strcmp(strstrip(buf), "threaded"))
3481 /* drain dying csses before we re-apply (threaded) subtree control */
3482 cgrp = cgroup_kn_lock_live(of->kn, true);
3486 /* threaded can only be enabled */
3487 ret = cgroup_enable_threaded(cgrp);
3489 cgroup_kn_unlock(of->kn);
3490 return ret ?: nbytes;
3493 static int cgroup_max_descendants_show(struct seq_file *seq, void *v)
3495 struct cgroup *cgrp = seq_css(seq)->cgroup;
3496 int descendants = READ_ONCE(cgrp->max_descendants);
3498 if (descendants == INT_MAX)
3499 seq_puts(seq, "max\n");
3501 seq_printf(seq, "%d\n", descendants);
3506 static ssize_t cgroup_max_descendants_write(struct kernfs_open_file *of,
3507 char *buf, size_t nbytes, loff_t off)
3509 struct cgroup *cgrp;
3513 buf = strstrip(buf);
3514 if (!strcmp(buf, "max")) {
3515 descendants = INT_MAX;
3517 ret = kstrtoint(buf, 0, &descendants);
3522 if (descendants < 0)
3525 cgrp = cgroup_kn_lock_live(of->kn, false);
3529 cgrp->max_descendants = descendants;
3531 cgroup_kn_unlock(of->kn);
3536 static int cgroup_max_depth_show(struct seq_file *seq, void *v)
3538 struct cgroup *cgrp = seq_css(seq)->cgroup;
3539 int depth = READ_ONCE(cgrp->max_depth);
3541 if (depth == INT_MAX)
3542 seq_puts(seq, "max\n");
3544 seq_printf(seq, "%d\n", depth);
3549 static ssize_t cgroup_max_depth_write(struct kernfs_open_file *of,
3550 char *buf, size_t nbytes, loff_t off)
3552 struct cgroup *cgrp;
3556 buf = strstrip(buf);
3557 if (!strcmp(buf, "max")) {
3560 ret = kstrtoint(buf, 0, &depth);
3568 cgrp = cgroup_kn_lock_live(of->kn, false);
3572 cgrp->max_depth = depth;
3574 cgroup_kn_unlock(of->kn);
3579 static int cgroup_events_show(struct seq_file *seq, void *v)
3581 struct cgroup *cgrp = seq_css(seq)->cgroup;
3583 seq_printf(seq, "populated %d\n", cgroup_is_populated(cgrp));
3584 seq_printf(seq, "frozen %d\n", test_bit(CGRP_FROZEN, &cgrp->flags));
3589 static int cgroup_stat_show(struct seq_file *seq, void *v)
3591 struct cgroup *cgroup = seq_css(seq)->cgroup;
3593 seq_printf(seq, "nr_descendants %d\n",
3594 cgroup->nr_descendants);
3595 seq_printf(seq, "nr_dying_descendants %d\n",
3596 cgroup->nr_dying_descendants);
3601 static int __maybe_unused cgroup_extra_stat_show(struct seq_file *seq,
3602 struct cgroup *cgrp, int ssid)
3604 struct cgroup_subsys *ss = cgroup_subsys[ssid];
3605 struct cgroup_subsys_state *css;
3608 if (!ss->css_extra_stat_show)
3611 css = cgroup_tryget_css(cgrp, ss);
3615 ret = ss->css_extra_stat_show(seq, css);
3620 static int cpu_stat_show(struct seq_file *seq, void *v)
3622 struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
3625 cgroup_base_stat_cputime_show(seq);
3626 #ifdef CONFIG_CGROUP_SCHED
3627 ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id);
3633 static int cgroup_io_pressure_show(struct seq_file *seq, void *v)
3635 struct cgroup *cgrp = seq_css(seq)->cgroup;
3636 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3638 return psi_show(seq, psi, PSI_IO);
3640 static int cgroup_memory_pressure_show(struct seq_file *seq, void *v)
3642 struct cgroup *cgrp = seq_css(seq)->cgroup;
3643 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3645 return psi_show(seq, psi, PSI_MEM);
3647 static int cgroup_cpu_pressure_show(struct seq_file *seq, void *v)
3649 struct cgroup *cgrp = seq_css(seq)->cgroup;
3650 struct psi_group *psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3652 return psi_show(seq, psi, PSI_CPU);
3655 static ssize_t cgroup_pressure_write(struct kernfs_open_file *of, char *buf,
3656 size_t nbytes, enum psi_res res)
3658 struct cgroup_file_ctx *ctx = of->priv;
3659 struct psi_trigger *new;
3660 struct cgroup *cgrp;
3661 struct psi_group *psi;
3663 cgrp = cgroup_kn_lock_live(of->kn, false);
3668 cgroup_kn_unlock(of->kn);
3670 /* Allow only one trigger per file descriptor */
3671 if (ctx->psi.trigger) {
3676 psi = cgroup_ino(cgrp) == 1 ? &psi_system : &cgrp->psi;
3677 new = psi_trigger_create(psi, buf, nbytes, res);
3680 return PTR_ERR(new);
3683 smp_store_release(&ctx->psi.trigger, new);
3689 static ssize_t cgroup_io_pressure_write(struct kernfs_open_file *of,
3690 char *buf, size_t nbytes,
3693 return cgroup_pressure_write(of, buf, nbytes, PSI_IO);
3696 static ssize_t cgroup_memory_pressure_write(struct kernfs_open_file *of,
3697 char *buf, size_t nbytes,
3700 return cgroup_pressure_write(of, buf, nbytes, PSI_MEM);
3703 static ssize_t cgroup_cpu_pressure_write(struct kernfs_open_file *of,
3704 char *buf, size_t nbytes,
3707 return cgroup_pressure_write(of, buf, nbytes, PSI_CPU);
3710 static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of,
3713 struct cgroup_file_ctx *ctx = of->priv;
3714 return psi_trigger_poll(&ctx->psi.trigger, of->file, pt);
3717 static void cgroup_pressure_release(struct kernfs_open_file *of)
3719 struct cgroup_file_ctx *ctx = of->priv;
3721 psi_trigger_destroy(ctx->psi.trigger);
3723 #endif /* CONFIG_PSI */
3725 static int cgroup_freeze_show(struct seq_file *seq, void *v)
3727 struct cgroup *cgrp = seq_css(seq)->cgroup;
3729 seq_printf(seq, "%d\n", cgrp->freezer.freeze);
3734 static ssize_t cgroup_freeze_write(struct kernfs_open_file *of,
3735 char *buf, size_t nbytes, loff_t off)
3737 struct cgroup *cgrp;
3741 ret = kstrtoint(strstrip(buf), 0, &freeze);
3745 if (freeze < 0 || freeze > 1)
3748 cgrp = cgroup_kn_lock_live(of->kn, false);
3752 cgroup_freeze(cgrp, freeze);
3754 cgroup_kn_unlock(of->kn);
3759 static int cgroup_file_open(struct kernfs_open_file *of)
3761 struct cftype *cft = of->kn->priv;
3762 struct cgroup_file_ctx *ctx;
3765 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
3769 ctx->ns = current->nsproxy->cgroup_ns;
3770 get_cgroup_ns(ctx->ns);
3776 ret = cft->open(of);
3778 put_cgroup_ns(ctx->ns);
3784 static void cgroup_file_release(struct kernfs_open_file *of)
3786 struct cftype *cft = of->kn->priv;
3787 struct cgroup_file_ctx *ctx = of->priv;
3791 put_cgroup_ns(ctx->ns);
3795 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3796 size_t nbytes, loff_t off)
3798 struct cgroup_file_ctx *ctx = of->priv;
3799 struct cgroup *cgrp = of->kn->parent->priv;
3800 struct cftype *cft = of->kn->priv;
3801 struct cgroup_subsys_state *css;
3808 * If namespaces are delegation boundaries, disallow writes to
3809 * files in an non-init namespace root from inside the namespace
3810 * except for the files explicitly marked delegatable -
3811 * cgroup.procs and cgroup.subtree_control.
3813 if ((cgrp->root->flags & CGRP_ROOT_NS_DELEGATE) &&
3814 !(cft->flags & CFTYPE_NS_DELEGATABLE) &&
3815 ctx->ns != &init_cgroup_ns && ctx->ns->root_cset->dfl_cgrp == cgrp)
3819 return cft->write(of, buf, nbytes, off);
3822 * kernfs guarantees that a file isn't deleted with operations in
3823 * flight, which means that the matching css is and stays alive and
3824 * doesn't need to be pinned. The RCU locking is not necessary
3825 * either. It's just for the convenience of using cgroup_css().
3828 css = cgroup_css(cgrp, cft->ss);
3831 if (cft->write_u64) {
3832 unsigned long long v;
3833 ret = kstrtoull(buf, 0, &v);
3835 ret = cft->write_u64(css, cft, v);
3836 } else if (cft->write_s64) {
3838 ret = kstrtoll(buf, 0, &v);
3840 ret = cft->write_s64(css, cft, v);
3845 return ret ?: nbytes;
3848 static __poll_t cgroup_file_poll(struct kernfs_open_file *of, poll_table *pt)
3850 struct cftype *cft = of->kn->priv;
3853 return cft->poll(of, pt);
3855 return kernfs_generic_poll(of, pt);
3858 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3860 return seq_cft(seq)->seq_start(seq, ppos);
3863 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3865 return seq_cft(seq)->seq_next(seq, v, ppos);
3868 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3870 if (seq_cft(seq)->seq_stop)
3871 seq_cft(seq)->seq_stop(seq, v);
3874 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3876 struct cftype *cft = seq_cft(m);
3877 struct cgroup_subsys_state *css = seq_css(m);
3880 return cft->seq_show(m, arg);
3883 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3884 else if (cft->read_s64)
3885 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3891 static struct kernfs_ops cgroup_kf_single_ops = {
3892 .atomic_write_len = PAGE_SIZE,
3893 .open = cgroup_file_open,
3894 .release = cgroup_file_release,
3895 .write = cgroup_file_write,
3896 .poll = cgroup_file_poll,
3897 .seq_show = cgroup_seqfile_show,
3900 static struct kernfs_ops cgroup_kf_ops = {
3901 .atomic_write_len = PAGE_SIZE,
3902 .open = cgroup_file_open,
3903 .release = cgroup_file_release,
3904 .write = cgroup_file_write,
3905 .poll = cgroup_file_poll,
3906 .seq_start = cgroup_seqfile_start,
3907 .seq_next = cgroup_seqfile_next,
3908 .seq_stop = cgroup_seqfile_stop,
3909 .seq_show = cgroup_seqfile_show,
3912 /* set uid and gid of cgroup dirs and files to that of the creator */
3913 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3915 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3916 .ia_uid = current_fsuid(),
3917 .ia_gid = current_fsgid(), };
3919 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3920 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3923 return kernfs_setattr(kn, &iattr);
3926 static void cgroup_file_notify_timer(struct timer_list *timer)
3928 cgroup_file_notify(container_of(timer, struct cgroup_file,
3932 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3935 char name[CGROUP_FILE_NAME_MAX];
3936 struct kernfs_node *kn;
3937 struct lock_class_key *key = NULL;
3940 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3941 key = &cft->lockdep_key;
3943 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3944 cgroup_file_mode(cft),
3945 GLOBAL_ROOT_UID, GLOBAL_ROOT_GID,
3946 0, cft->kf_ops, cft,
3951 ret = cgroup_kn_set_ugid(kn);
3957 if (cft->file_offset) {
3958 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3960 timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0);
3962 spin_lock_irq(&cgroup_file_kn_lock);
3964 spin_unlock_irq(&cgroup_file_kn_lock);
3971 * cgroup_addrm_files - add or remove files to a cgroup directory
3972 * @css: the target css
3973 * @cgrp: the target cgroup (usually css->cgroup)
3974 * @cfts: array of cftypes to be added
3975 * @is_add: whether to add or remove
3977 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3978 * For removals, this function never fails.
3980 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3981 struct cgroup *cgrp, struct cftype cfts[],
3984 struct cftype *cft, *cft_end = NULL;
3987 lockdep_assert_held(&cgroup_mutex);
3990 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3991 /* does cft->flags tell us to skip this file on @cgrp? */
3992 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3994 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3996 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3998 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
4000 if ((cft->flags & CFTYPE_DEBUG) && !cgroup_debug)
4003 ret = cgroup_add_file(css, cgrp, cft);
4005 pr_warn("%s: failed to add %s, err=%d\n",
4006 __func__, cft->name, ret);
4012 cgroup_rm_file(cgrp, cft);
4018 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
4020 struct cgroup_subsys *ss = cfts[0].ss;
4021 struct cgroup *root = &ss->root->cgrp;
4022 struct cgroup_subsys_state *css;
4025 lockdep_assert_held(&cgroup_mutex);
4027 /* add/rm files for all cgroups created before */
4028 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
4029 struct cgroup *cgrp = css->cgroup;
4031 if (!(css->flags & CSS_VISIBLE))
4034 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
4040 kernfs_activate(root->kn);
4044 static void cgroup_exit_cftypes(struct cftype *cfts)
4048 for (cft = cfts; cft->name[0] != '\0'; cft++) {
4049 /* free copy for custom atomic_write_len, see init_cftypes() */
4050 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
4055 /* revert flags set by cgroup core while adding @cfts */
4056 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
4060 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4064 for (cft = cfts; cft->name[0] != '\0'; cft++) {
4065 struct kernfs_ops *kf_ops;
4067 WARN_ON(cft->ss || cft->kf_ops);
4070 kf_ops = &cgroup_kf_ops;
4072 kf_ops = &cgroup_kf_single_ops;
4075 * Ugh... if @cft wants a custom max_write_len, we need to
4076 * make a copy of kf_ops to set its atomic_write_len.
4078 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
4079 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
4081 cgroup_exit_cftypes(cfts);
4084 kf_ops->atomic_write_len = cft->max_write_len;
4087 cft->kf_ops = kf_ops;
4094 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
4096 lockdep_assert_held(&cgroup_mutex);
4098 if (!cfts || !cfts[0].ss)
4101 list_del(&cfts->node);
4102 cgroup_apply_cftypes(cfts, false);
4103 cgroup_exit_cftypes(cfts);
4108 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
4109 * @cfts: zero-length name terminated array of cftypes
4111 * Unregister @cfts. Files described by @cfts are removed from all
4112 * existing cgroups and all future cgroups won't have them either. This
4113 * function can be called anytime whether @cfts' subsys is attached or not.
4115 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
4118 int cgroup_rm_cftypes(struct cftype *cfts)
4122 mutex_lock(&cgroup_mutex);
4123 ret = cgroup_rm_cftypes_locked(cfts);
4124 mutex_unlock(&cgroup_mutex);
4129 * cgroup_add_cftypes - add an array of cftypes to a subsystem
4130 * @ss: target cgroup subsystem
4131 * @cfts: zero-length name terminated array of cftypes
4133 * Register @cfts to @ss. Files described by @cfts are created for all
4134 * existing cgroups to which @ss is attached and all future cgroups will
4135 * have them too. This function can be called anytime whether @ss is
4138 * Returns 0 on successful registration, -errno on failure. Note that this
4139 * function currently returns 0 as long as @cfts registration is successful
4140 * even if some file creation attempts on existing cgroups fail.
4142 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4146 if (!cgroup_ssid_enabled(ss->id))
4149 if (!cfts || cfts[0].name[0] == '\0')
4152 ret = cgroup_init_cftypes(ss, cfts);
4156 mutex_lock(&cgroup_mutex);
4158 list_add_tail(&cfts->node, &ss->cfts);
4159 ret = cgroup_apply_cftypes(cfts, true);
4161 cgroup_rm_cftypes_locked(cfts);
4163 mutex_unlock(&cgroup_mutex);
4168 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
4169 * @ss: target cgroup subsystem
4170 * @cfts: zero-length name terminated array of cftypes
4172 * Similar to cgroup_add_cftypes() but the added files are only used for
4173 * the default hierarchy.
4175 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4179 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4180 cft->flags |= __CFTYPE_ONLY_ON_DFL;
4181 return cgroup_add_cftypes(ss, cfts);
4185 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
4186 * @ss: target cgroup subsystem
4187 * @cfts: zero-length name terminated array of cftypes
4189 * Similar to cgroup_add_cftypes() but the added files are only used for
4190 * the legacy hierarchies.
4192 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
4196 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
4197 cft->flags |= __CFTYPE_NOT_ON_DFL;
4198 return cgroup_add_cftypes(ss, cfts);
4202 * cgroup_file_notify - generate a file modified event for a cgroup_file
4203 * @cfile: target cgroup_file
4205 * @cfile must have been obtained by setting cftype->file_offset.
4207 void cgroup_file_notify(struct cgroup_file *cfile)
4209 unsigned long flags;
4211 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
4213 unsigned long last = cfile->notified_at;
4214 unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV;
4216 if (time_in_range(jiffies, last, next)) {
4217 timer_reduce(&cfile->notify_timer, next);
4219 kernfs_notify(cfile->kn);
4220 cfile->notified_at = jiffies;
4223 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
4227 * css_next_child - find the next child of a given css
4228 * @pos: the current position (%NULL to initiate traversal)
4229 * @parent: css whose children to walk
4231 * This function returns the next child of @parent and should be called
4232 * under either cgroup_mutex or RCU read lock. The only requirement is
4233 * that @parent and @pos are accessible. The next sibling is guaranteed to
4234 * be returned regardless of their states.
4236 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4237 * css which finished ->css_online() is guaranteed to be visible in the
4238 * future iterations and will stay visible until the last reference is put.
4239 * A css which hasn't finished ->css_online() or already finished
4240 * ->css_offline() may show up during traversal. It's each subsystem's
4241 * responsibility to synchronize against on/offlining.
4243 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
4244 struct cgroup_subsys_state *parent)
4246 struct cgroup_subsys_state *next;
4248 cgroup_assert_mutex_or_rcu_locked();
4251 * @pos could already have been unlinked from the sibling list.
4252 * Once a cgroup is removed, its ->sibling.next is no longer
4253 * updated when its next sibling changes. CSS_RELEASED is set when
4254 * @pos is taken off list, at which time its next pointer is valid,
4255 * and, as releases are serialized, the one pointed to by the next
4256 * pointer is guaranteed to not have started release yet. This
4257 * implies that if we observe !CSS_RELEASED on @pos in this RCU
4258 * critical section, the one pointed to by its next pointer is
4259 * guaranteed to not have finished its RCU grace period even if we
4260 * have dropped rcu_read_lock() inbetween iterations.
4262 * If @pos has CSS_RELEASED set, its next pointer can't be
4263 * dereferenced; however, as each css is given a monotonically
4264 * increasing unique serial number and always appended to the
4265 * sibling list, the next one can be found by walking the parent's
4266 * children until the first css with higher serial number than
4267 * @pos's. While this path can be slower, it happens iff iteration
4268 * races against release and the race window is very small.
4271 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
4272 } else if (likely(!(pos->flags & CSS_RELEASED))) {
4273 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
4275 list_for_each_entry_rcu(next, &parent->children, sibling,
4276 lockdep_is_held(&cgroup_mutex))
4277 if (next->serial_nr > pos->serial_nr)
4282 * @next, if not pointing to the head, can be dereferenced and is
4285 if (&next->sibling != &parent->children)
4291 * css_next_descendant_pre - find the next descendant for pre-order walk
4292 * @pos: the current position (%NULL to initiate traversal)
4293 * @root: css whose descendants to walk
4295 * To be used by css_for_each_descendant_pre(). Find the next descendant
4296 * to visit for pre-order traversal of @root's descendants. @root is
4297 * included in the iteration and the first node to be visited.
4299 * While this function requires cgroup_mutex or RCU read locking, it
4300 * doesn't require the whole traversal to be contained in a single critical
4301 * section. This function will return the correct next descendant as long
4302 * as both @pos and @root are accessible and @pos is a descendant of @root.
4304 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4305 * css which finished ->css_online() is guaranteed to be visible in the
4306 * future iterations and will stay visible until the last reference is put.
4307 * A css which hasn't finished ->css_online() or already finished
4308 * ->css_offline() may show up during traversal. It's each subsystem's
4309 * responsibility to synchronize against on/offlining.
4311 struct cgroup_subsys_state *
4312 css_next_descendant_pre(struct cgroup_subsys_state *pos,
4313 struct cgroup_subsys_state *root)
4315 struct cgroup_subsys_state *next;
4317 cgroup_assert_mutex_or_rcu_locked();
4319 /* if first iteration, visit @root */
4323 /* visit the first child if exists */
4324 next = css_next_child(NULL, pos);
4328 /* no child, visit my or the closest ancestor's next sibling */
4329 while (pos != root) {
4330 next = css_next_child(pos, pos->parent);
4338 EXPORT_SYMBOL_GPL(css_next_descendant_pre);
4341 * css_rightmost_descendant - return the rightmost descendant of a css
4342 * @pos: css of interest
4344 * Return the rightmost descendant of @pos. If there's no descendant, @pos
4345 * is returned. This can be used during pre-order traversal to skip
4348 * While this function requires cgroup_mutex or RCU read locking, it
4349 * doesn't require the whole traversal to be contained in a single critical
4350 * section. This function will return the correct rightmost descendant as
4351 * long as @pos is accessible.
4353 struct cgroup_subsys_state *
4354 css_rightmost_descendant(struct cgroup_subsys_state *pos)
4356 struct cgroup_subsys_state *last, *tmp;
4358 cgroup_assert_mutex_or_rcu_locked();
4362 /* ->prev isn't RCU safe, walk ->next till the end */
4364 css_for_each_child(tmp, last)
4371 static struct cgroup_subsys_state *
4372 css_leftmost_descendant(struct cgroup_subsys_state *pos)
4374 struct cgroup_subsys_state *last;
4378 pos = css_next_child(NULL, pos);
4385 * css_next_descendant_post - find the next descendant for post-order walk
4386 * @pos: the current position (%NULL to initiate traversal)
4387 * @root: css whose descendants to walk
4389 * To be used by css_for_each_descendant_post(). Find the next descendant
4390 * to visit for post-order traversal of @root's descendants. @root is
4391 * included in the iteration and the last node to be visited.
4393 * While this function requires cgroup_mutex or RCU read locking, it
4394 * doesn't require the whole traversal to be contained in a single critical
4395 * section. This function will return the correct next descendant as long
4396 * as both @pos and @cgroup are accessible and @pos is a descendant of
4399 * If a subsystem synchronizes ->css_online() and the start of iteration, a
4400 * css which finished ->css_online() is guaranteed to be visible in the
4401 * future iterations and will stay visible until the last reference is put.
4402 * A css which hasn't finished ->css_online() or already finished
4403 * ->css_offline() may show up during traversal. It's each subsystem's
4404 * responsibility to synchronize against on/offlining.
4406 struct cgroup_subsys_state *
4407 css_next_descendant_post(struct cgroup_subsys_state *pos,
4408 struct cgroup_subsys_state *root)
4410 struct cgroup_subsys_state *next;
4412 cgroup_assert_mutex_or_rcu_locked();
4414 /* if first iteration, visit leftmost descendant which may be @root */
4416 return css_leftmost_descendant(root);
4418 /* if we visited @root, we're done */
4422 /* if there's an unvisited sibling, visit its leftmost descendant */
4423 next = css_next_child(pos, pos->parent);
4425 return css_leftmost_descendant(next);
4427 /* no sibling left, visit parent */
4432 * css_has_online_children - does a css have online children
4433 * @css: the target css
4435 * Returns %true if @css has any online children; otherwise, %false. This
4436 * function can be called from any context but the caller is responsible
4437 * for synchronizing against on/offlining as necessary.
4439 bool css_has_online_children(struct cgroup_subsys_state *css)
4441 struct cgroup_subsys_state *child;
4445 css_for_each_child(child, css) {
4446 if (child->flags & CSS_ONLINE) {
4455 static struct css_set *css_task_iter_next_css_set(struct css_task_iter *it)
4457 struct list_head *l;
4458 struct cgrp_cset_link *link;
4459 struct css_set *cset;
4461 lockdep_assert_held(&css_set_lock);
4463 /* find the next threaded cset */
4464 if (it->tcset_pos) {
4465 l = it->tcset_pos->next;
4467 if (l != it->tcset_head) {
4469 return container_of(l, struct css_set,
4470 threaded_csets_node);
4473 it->tcset_pos = NULL;
4476 /* find the next cset */
4479 if (l == it->cset_head) {
4480 it->cset_pos = NULL;
4485 cset = container_of(l, struct css_set, e_cset_node[it->ss->id]);
4487 link = list_entry(l, struct cgrp_cset_link, cset_link);
4493 /* initialize threaded css_set walking */
4494 if (it->flags & CSS_TASK_ITER_THREADED) {
4496 put_css_set_locked(it->cur_dcset);
4497 it->cur_dcset = cset;
4500 it->tcset_head = &cset->threaded_csets;
4501 it->tcset_pos = &cset->threaded_csets;
4508 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
4509 * @it: the iterator to advance
4511 * Advance @it to the next css_set to walk.
4513 static void css_task_iter_advance_css_set(struct css_task_iter *it)
4515 struct css_set *cset;
4517 lockdep_assert_held(&css_set_lock);
4519 /* Advance to the next non-empty css_set and find first non-empty tasks list*/
4520 while ((cset = css_task_iter_next_css_set(it))) {
4521 if (!list_empty(&cset->tasks)) {
4522 it->cur_tasks_head = &cset->tasks;
4524 } else if (!list_empty(&cset->mg_tasks)) {
4525 it->cur_tasks_head = &cset->mg_tasks;
4527 } else if (!list_empty(&cset->dying_tasks)) {
4528 it->cur_tasks_head = &cset->dying_tasks;
4533 it->task_pos = NULL;
4536 it->task_pos = it->cur_tasks_head->next;
4539 * We don't keep css_sets locked across iteration steps and thus
4540 * need to take steps to ensure that iteration can be resumed after
4541 * the lock is re-acquired. Iteration is performed at two levels -
4542 * css_sets and tasks in them.
4544 * Once created, a css_set never leaves its cgroup lists, so a
4545 * pinned css_set is guaranteed to stay put and we can resume
4546 * iteration afterwards.
4548 * Tasks may leave @cset across iteration steps. This is resolved
4549 * by registering each iterator with the css_set currently being
4550 * walked and making css_set_move_task() advance iterators whose
4551 * next task is leaving.
4554 list_del(&it->iters_node);
4555 put_css_set_locked(it->cur_cset);
4558 it->cur_cset = cset;
4559 list_add(&it->iters_node, &cset->task_iters);
4562 static void css_task_iter_skip(struct css_task_iter *it,
4563 struct task_struct *task)
4565 lockdep_assert_held(&css_set_lock);
4567 if (it->task_pos == &task->cg_list) {
4568 it->task_pos = it->task_pos->next;
4569 it->flags |= CSS_TASK_ITER_SKIPPED;
4573 static void css_task_iter_advance(struct css_task_iter *it)
4575 struct task_struct *task;
4577 lockdep_assert_held(&css_set_lock);
4581 * Advance iterator to find next entry. We go through cset
4582 * tasks, mg_tasks and dying_tasks, when consumed we move onto
4585 if (it->flags & CSS_TASK_ITER_SKIPPED)
4586 it->flags &= ~CSS_TASK_ITER_SKIPPED;
4588 it->task_pos = it->task_pos->next;
4590 if (it->task_pos == &it->cur_cset->tasks) {
4591 it->cur_tasks_head = &it->cur_cset->mg_tasks;
4592 it->task_pos = it->cur_tasks_head->next;
4594 if (it->task_pos == &it->cur_cset->mg_tasks) {
4595 it->cur_tasks_head = &it->cur_cset->dying_tasks;
4596 it->task_pos = it->cur_tasks_head->next;
4598 if (it->task_pos == &it->cur_cset->dying_tasks)
4599 css_task_iter_advance_css_set(it);
4601 /* called from start, proceed to the first cset */
4602 css_task_iter_advance_css_set(it);
4608 task = list_entry(it->task_pos, struct task_struct, cg_list);
4610 if (it->flags & CSS_TASK_ITER_PROCS) {
4611 /* if PROCS, skip over tasks which aren't group leaders */
4612 if (!thread_group_leader(task))
4615 /* and dying leaders w/o live member threads */
4616 if (it->cur_tasks_head == &it->cur_cset->dying_tasks &&
4617 !atomic_read(&task->signal->live))
4620 /* skip all dying ones */
4621 if (it->cur_tasks_head == &it->cur_cset->dying_tasks)
4627 * css_task_iter_start - initiate task iteration
4628 * @css: the css to walk tasks of
4629 * @flags: CSS_TASK_ITER_* flags
4630 * @it: the task iterator to use
4632 * Initiate iteration through the tasks of @css. The caller can call
4633 * css_task_iter_next() to walk through the tasks until the function
4634 * returns NULL. On completion of iteration, css_task_iter_end() must be
4637 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags,
4638 struct css_task_iter *it)
4640 memset(it, 0, sizeof(*it));
4642 spin_lock_irq(&css_set_lock);
4648 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
4650 it->cset_pos = &css->cgroup->cset_links;
4652 it->cset_head = it->cset_pos;
4654 css_task_iter_advance(it);
4656 spin_unlock_irq(&css_set_lock);
4660 * css_task_iter_next - return the next task for the iterator
4661 * @it: the task iterator being iterated
4663 * The "next" function for task iteration. @it should have been
4664 * initialized via css_task_iter_start(). Returns NULL when the iteration
4667 struct task_struct *css_task_iter_next(struct css_task_iter *it)
4670 put_task_struct(it->cur_task);
4671 it->cur_task = NULL;
4674 spin_lock_irq(&css_set_lock);
4676 /* @it may be half-advanced by skips, finish advancing */
4677 if (it->flags & CSS_TASK_ITER_SKIPPED)
4678 css_task_iter_advance(it);
4681 it->cur_task = list_entry(it->task_pos, struct task_struct,
4683 get_task_struct(it->cur_task);
4684 css_task_iter_advance(it);
4687 spin_unlock_irq(&css_set_lock);
4689 return it->cur_task;
4693 * css_task_iter_end - finish task iteration
4694 * @it: the task iterator to finish
4696 * Finish task iteration started by css_task_iter_start().
4698 void css_task_iter_end(struct css_task_iter *it)
4701 spin_lock_irq(&css_set_lock);
4702 list_del(&it->iters_node);
4703 put_css_set_locked(it->cur_cset);
4704 spin_unlock_irq(&css_set_lock);
4708 put_css_set(it->cur_dcset);
4711 put_task_struct(it->cur_task);
4714 static void cgroup_procs_release(struct kernfs_open_file *of)
4716 struct cgroup_file_ctx *ctx = of->priv;
4718 if (ctx->procs.started)
4719 css_task_iter_end(&ctx->procs.iter);
4722 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
4724 struct kernfs_open_file *of = s->private;
4725 struct cgroup_file_ctx *ctx = of->priv;
4730 return css_task_iter_next(&ctx->procs.iter);
4733 static void *__cgroup_procs_start(struct seq_file *s, loff_t *pos,
4734 unsigned int iter_flags)
4736 struct kernfs_open_file *of = s->private;
4737 struct cgroup *cgrp = seq_css(s)->cgroup;
4738 struct cgroup_file_ctx *ctx = of->priv;
4739 struct css_task_iter *it = &ctx->procs.iter;
4742 * When a seq_file is seeked, it's always traversed sequentially
4743 * from position 0, so we can simply keep iterating on !0 *pos.
4745 if (!ctx->procs.started) {
4746 if (WARN_ON_ONCE((*pos)))
4747 return ERR_PTR(-EINVAL);
4748 css_task_iter_start(&cgrp->self, iter_flags, it);
4749 ctx->procs.started = true;
4750 } else if (!(*pos)) {
4751 css_task_iter_end(it);
4752 css_task_iter_start(&cgrp->self, iter_flags, it);
4754 return it->cur_task;
4756 return cgroup_procs_next(s, NULL, NULL);
4759 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
4761 struct cgroup *cgrp = seq_css(s)->cgroup;
4764 * All processes of a threaded subtree belong to the domain cgroup
4765 * of the subtree. Only threads can be distributed across the
4766 * subtree. Reject reads on cgroup.procs in the subtree proper.
4767 * They're always empty anyway.
4769 if (cgroup_is_threaded(cgrp))
4770 return ERR_PTR(-EOPNOTSUPP);
4772 return __cgroup_procs_start(s, pos, CSS_TASK_ITER_PROCS |
4773 CSS_TASK_ITER_THREADED);
4776 static int cgroup_procs_show(struct seq_file *s, void *v)
4778 seq_printf(s, "%d\n", task_pid_vnr(v));
4782 static int cgroup_may_write(const struct cgroup *cgrp, struct super_block *sb)
4785 struct inode *inode;
4787 lockdep_assert_held(&cgroup_mutex);
4789 inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
4793 ret = inode_permission(inode, MAY_WRITE);
4798 static int cgroup_procs_write_permission(struct cgroup *src_cgrp,
4799 struct cgroup *dst_cgrp,
4800 struct super_block *sb,
4801 struct cgroup_namespace *ns)
4803 struct cgroup *com_cgrp = src_cgrp;
4806 lockdep_assert_held(&cgroup_mutex);
4808 /* find the common ancestor */
4809 while (!cgroup_is_descendant(dst_cgrp, com_cgrp))
4810 com_cgrp = cgroup_parent(com_cgrp);
4812 /* %current should be authorized to migrate to the common ancestor */
4813 ret = cgroup_may_write(com_cgrp, sb);
4818 * If namespaces are delegation boundaries, %current must be able
4819 * to see both source and destination cgroups from its namespace.
4821 if ((cgrp_dfl_root.flags & CGRP_ROOT_NS_DELEGATE) &&
4822 (!cgroup_is_descendant(src_cgrp, ns->root_cset->dfl_cgrp) ||
4823 !cgroup_is_descendant(dst_cgrp, ns->root_cset->dfl_cgrp)))
4829 static int cgroup_attach_permissions(struct cgroup *src_cgrp,
4830 struct cgroup *dst_cgrp,
4831 struct super_block *sb, bool threadgroup,
4832 struct cgroup_namespace *ns)
4836 ret = cgroup_procs_write_permission(src_cgrp, dst_cgrp, sb, ns);
4840 ret = cgroup_migrate_vet_dst(dst_cgrp);
4844 if (!threadgroup && (src_cgrp->dom_cgrp != dst_cgrp->dom_cgrp))
4850 static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
4851 char *buf, size_t nbytes, loff_t off)
4853 struct cgroup_file_ctx *ctx = of->priv;
4854 struct cgroup *src_cgrp, *dst_cgrp;
4855 struct task_struct *task;
4856 const struct cred *saved_cred;
4858 bool threadgroup_locked;
4860 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4864 task = cgroup_procs_write_start(buf, true, &threadgroup_locked);
4865 ret = PTR_ERR_OR_ZERO(task);
4869 /* find the source cgroup */
4870 spin_lock_irq(&css_set_lock);
4871 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4872 spin_unlock_irq(&css_set_lock);
4875 * Process and thread migrations follow same delegation rule. Check
4876 * permissions using the credentials from file open to protect against
4877 * inherited fd attacks.
4879 saved_cred = override_creds(of->file->f_cred);
4880 ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
4881 of->file->f_path.dentry->d_sb, true,
4883 revert_creds(saved_cred);
4887 ret = cgroup_attach_task(dst_cgrp, task, true);
4890 cgroup_procs_write_finish(task, threadgroup_locked);
4892 cgroup_kn_unlock(of->kn);
4894 return ret ?: nbytes;
4897 static void *cgroup_threads_start(struct seq_file *s, loff_t *pos)
4899 return __cgroup_procs_start(s, pos, 0);
4902 static ssize_t cgroup_threads_write(struct kernfs_open_file *of,
4903 char *buf, size_t nbytes, loff_t off)
4905 struct cgroup_file_ctx *ctx = of->priv;
4906 struct cgroup *src_cgrp, *dst_cgrp;
4907 struct task_struct *task;
4908 const struct cred *saved_cred;
4912 buf = strstrip(buf);
4914 dst_cgrp = cgroup_kn_lock_live(of->kn, false);
4918 task = cgroup_procs_write_start(buf, false, &locked);
4919 ret = PTR_ERR_OR_ZERO(task);
4923 /* find the source cgroup */
4924 spin_lock_irq(&css_set_lock);
4925 src_cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
4926 spin_unlock_irq(&css_set_lock);
4929 * Process and thread migrations follow same delegation rule. Check
4930 * permissions using the credentials from file open to protect against
4931 * inherited fd attacks.
4933 saved_cred = override_creds(of->file->f_cred);
4934 ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
4935 of->file->f_path.dentry->d_sb, false,
4937 revert_creds(saved_cred);
4941 ret = cgroup_attach_task(dst_cgrp, task, false);
4944 cgroup_procs_write_finish(task, locked);
4946 cgroup_kn_unlock(of->kn);
4948 return ret ?: nbytes;
4951 /* cgroup core interface files for the default hierarchy */
4952 static struct cftype cgroup_base_files[] = {
4954 .name = "cgroup.type",
4955 .flags = CFTYPE_NOT_ON_ROOT,
4956 .seq_show = cgroup_type_show,
4957 .write = cgroup_type_write,
4960 .name = "cgroup.procs",
4961 .flags = CFTYPE_NS_DELEGATABLE,
4962 .file_offset = offsetof(struct cgroup, procs_file),
4963 .release = cgroup_procs_release,
4964 .seq_start = cgroup_procs_start,
4965 .seq_next = cgroup_procs_next,
4966 .seq_show = cgroup_procs_show,
4967 .write = cgroup_procs_write,
4970 .name = "cgroup.threads",
4971 .flags = CFTYPE_NS_DELEGATABLE,
4972 .release = cgroup_procs_release,
4973 .seq_start = cgroup_threads_start,
4974 .seq_next = cgroup_procs_next,
4975 .seq_show = cgroup_procs_show,
4976 .write = cgroup_threads_write,
4979 .name = "cgroup.controllers",
4980 .seq_show = cgroup_controllers_show,
4983 .name = "cgroup.subtree_control",
4984 .flags = CFTYPE_NS_DELEGATABLE,
4985 .seq_show = cgroup_subtree_control_show,
4986 .write = cgroup_subtree_control_write,
4989 .name = "cgroup.events",
4990 .flags = CFTYPE_NOT_ON_ROOT,
4991 .file_offset = offsetof(struct cgroup, events_file),
4992 .seq_show = cgroup_events_show,
4995 .name = "cgroup.max.descendants",
4996 .seq_show = cgroup_max_descendants_show,
4997 .write = cgroup_max_descendants_write,
5000 .name = "cgroup.max.depth",
5001 .seq_show = cgroup_max_depth_show,
5002 .write = cgroup_max_depth_write,
5005 .name = "cgroup.stat",
5006 .seq_show = cgroup_stat_show,
5009 .name = "cgroup.freeze",
5010 .flags = CFTYPE_NOT_ON_ROOT,
5011 .seq_show = cgroup_freeze_show,
5012 .write = cgroup_freeze_write,
5016 .seq_show = cpu_stat_show,
5020 .name = "io.pressure",
5021 .seq_show = cgroup_io_pressure_show,
5022 .write = cgroup_io_pressure_write,
5023 .poll = cgroup_pressure_poll,
5024 .release = cgroup_pressure_release,
5027 .name = "memory.pressure",
5028 .seq_show = cgroup_memory_pressure_show,
5029 .write = cgroup_memory_pressure_write,
5030 .poll = cgroup_pressure_poll,
5031 .release = cgroup_pressure_release,
5034 .name = "cpu.pressure",
5035 .seq_show = cgroup_cpu_pressure_show,
5036 .write = cgroup_cpu_pressure_write,
5037 .poll = cgroup_pressure_poll,
5038 .release = cgroup_pressure_release,
5040 #endif /* CONFIG_PSI */
5045 * css destruction is four-stage process.
5047 * 1. Destruction starts. Killing of the percpu_ref is initiated.
5048 * Implemented in kill_css().
5050 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
5051 * and thus css_tryget_online() is guaranteed to fail, the css can be
5052 * offlined by invoking offline_css(). After offlining, the base ref is
5053 * put. Implemented in css_killed_work_fn().
5055 * 3. When the percpu_ref reaches zero, the only possible remaining
5056 * accessors are inside RCU read sections. css_release() schedules the
5059 * 4. After the grace period, the css can be freed. Implemented in
5060 * css_free_work_fn().
5062 * It is actually hairier because both step 2 and 4 require process context
5063 * and thus involve punting to css->destroy_work adding two additional
5064 * steps to the already complex sequence.
5066 static void css_free_rwork_fn(struct work_struct *work)
5068 struct cgroup_subsys_state *css = container_of(to_rcu_work(work),
5069 struct cgroup_subsys_state, destroy_rwork);
5070 struct cgroup_subsys *ss = css->ss;
5071 struct cgroup *cgrp = css->cgroup;
5073 percpu_ref_exit(&css->refcnt);
5077 struct cgroup_subsys_state *parent = css->parent;
5081 cgroup_idr_remove(&ss->css_idr, id);
5087 /* cgroup free path */
5088 atomic_dec(&cgrp->root->nr_cgrps);
5089 cgroup1_pidlist_destroy_all(cgrp);
5090 cancel_work_sync(&cgrp->release_agent_work);
5092 if (cgroup_parent(cgrp)) {
5094 * We get a ref to the parent, and put the ref when
5095 * this cgroup is being freed, so it's guaranteed
5096 * that the parent won't be destroyed before its
5099 cgroup_put(cgroup_parent(cgrp));
5100 kernfs_put(cgrp->kn);
5101 psi_cgroup_free(cgrp);
5102 if (cgroup_on_dfl(cgrp))
5103 cgroup_rstat_exit(cgrp);
5107 * This is root cgroup's refcnt reaching zero,
5108 * which indicates that the root should be
5111 cgroup_destroy_root(cgrp->root);
5116 static void css_release_work_fn(struct work_struct *work)
5118 struct cgroup_subsys_state *css =
5119 container_of(work, struct cgroup_subsys_state, destroy_work);
5120 struct cgroup_subsys *ss = css->ss;
5121 struct cgroup *cgrp = css->cgroup;
5123 mutex_lock(&cgroup_mutex);
5125 css->flags |= CSS_RELEASED;
5126 list_del_rcu(&css->sibling);
5129 /* css release path */
5130 if (!list_empty(&css->rstat_css_node)) {
5131 cgroup_rstat_flush(cgrp);
5132 list_del_rcu(&css->rstat_css_node);
5135 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
5136 if (ss->css_released)
5137 ss->css_released(css);
5139 struct cgroup *tcgrp;
5141 /* cgroup release path */
5142 TRACE_CGROUP_PATH(release, cgrp);
5144 if (cgroup_on_dfl(cgrp))
5145 cgroup_rstat_flush(cgrp);
5147 spin_lock_irq(&css_set_lock);
5148 for (tcgrp = cgroup_parent(cgrp); tcgrp;
5149 tcgrp = cgroup_parent(tcgrp))
5150 tcgrp->nr_dying_descendants--;
5151 spin_unlock_irq(&css_set_lock);
5154 * There are two control paths which try to determine
5155 * cgroup from dentry without going through kernfs -
5156 * cgroupstats_build() and css_tryget_online_from_dir().
5157 * Those are supported by RCU protecting clearing of
5158 * cgrp->kn->priv backpointer.
5161 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
5165 mutex_unlock(&cgroup_mutex);
5167 INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5168 queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5171 static void css_release(struct percpu_ref *ref)
5173 struct cgroup_subsys_state *css =
5174 container_of(ref, struct cgroup_subsys_state, refcnt);
5176 INIT_WORK(&css->destroy_work, css_release_work_fn);
5177 queue_work(cgroup_destroy_wq, &css->destroy_work);
5180 static void init_and_link_css(struct cgroup_subsys_state *css,
5181 struct cgroup_subsys *ss, struct cgroup *cgrp)
5183 lockdep_assert_held(&cgroup_mutex);
5185 cgroup_get_live(cgrp);
5187 memset(css, 0, sizeof(*css));
5191 INIT_LIST_HEAD(&css->sibling);
5192 INIT_LIST_HEAD(&css->children);
5193 INIT_LIST_HEAD(&css->rstat_css_node);
5194 css->serial_nr = css_serial_nr_next++;
5195 atomic_set(&css->online_cnt, 0);
5197 if (cgroup_parent(cgrp)) {
5198 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
5199 css_get(css->parent);
5202 if (cgroup_on_dfl(cgrp) && ss->css_rstat_flush)
5203 list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list);
5205 BUG_ON(cgroup_css(cgrp, ss));
5208 /* invoke ->css_online() on a new CSS and mark it online if successful */
5209 static int online_css(struct cgroup_subsys_state *css)
5211 struct cgroup_subsys *ss = css->ss;
5214 lockdep_assert_held(&cgroup_mutex);
5217 ret = ss->css_online(css);
5219 css->flags |= CSS_ONLINE;
5220 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
5222 atomic_inc(&css->online_cnt);
5224 atomic_inc(&css->parent->online_cnt);
5229 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
5230 static void offline_css(struct cgroup_subsys_state *css)
5232 struct cgroup_subsys *ss = css->ss;
5234 lockdep_assert_held(&cgroup_mutex);
5236 if (!(css->flags & CSS_ONLINE))
5239 if (ss->css_offline)
5240 ss->css_offline(css);
5242 css->flags &= ~CSS_ONLINE;
5243 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
5245 wake_up_all(&css->cgroup->offline_waitq);
5249 * css_create - create a cgroup_subsys_state
5250 * @cgrp: the cgroup new css will be associated with
5251 * @ss: the subsys of new css
5253 * Create a new css associated with @cgrp - @ss pair. On success, the new
5254 * css is online and installed in @cgrp. This function doesn't create the
5255 * interface files. Returns 0 on success, -errno on failure.
5257 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
5258 struct cgroup_subsys *ss)
5260 struct cgroup *parent = cgroup_parent(cgrp);
5261 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
5262 struct cgroup_subsys_state *css;
5265 lockdep_assert_held(&cgroup_mutex);
5267 css = ss->css_alloc(parent_css);
5269 css = ERR_PTR(-ENOMEM);
5273 init_and_link_css(css, ss, cgrp);
5275 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
5279 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
5284 /* @css is ready to be brought online now, make it visible */
5285 list_add_tail_rcu(&css->sibling, &parent_css->children);
5286 cgroup_idr_replace(&ss->css_idr, css, css->id);
5288 err = online_css(css);
5292 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
5293 cgroup_parent(parent)) {
5294 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
5295 current->comm, current->pid, ss->name);
5296 if (!strcmp(ss->name, "memory"))
5297 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
5298 ss->warned_broken_hierarchy = true;
5304 list_del_rcu(&css->sibling);
5306 list_del_rcu(&css->rstat_css_node);
5307 INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
5308 queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
5309 return ERR_PTR(err);
5313 * The returned cgroup is fully initialized including its control mask, but
5314 * it isn't associated with its kernfs_node and doesn't have the control
5317 static struct cgroup *cgroup_create(struct cgroup *parent, const char *name,
5320 struct cgroup_root *root = parent->root;
5321 struct cgroup *cgrp, *tcgrp;
5322 struct kernfs_node *kn;
5323 int level = parent->level + 1;
5326 /* allocate the cgroup and its ID, 0 is reserved for the root */
5327 cgrp = kzalloc(struct_size(cgrp, ancestor_ids, (level + 1)),
5330 return ERR_PTR(-ENOMEM);
5332 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
5336 if (cgroup_on_dfl(parent)) {
5337 ret = cgroup_rstat_init(cgrp);
5339 goto out_cancel_ref;
5342 /* create the directory */
5343 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
5350 init_cgroup_housekeeping(cgrp);
5352 cgrp->self.parent = &parent->self;
5354 cgrp->level = level;
5356 ret = psi_cgroup_alloc(cgrp);
5358 goto out_kernfs_remove;
5360 ret = cgroup_bpf_inherit(cgrp);
5365 * New cgroup inherits effective freeze counter, and
5366 * if the parent has to be frozen, the child has too.
5368 cgrp->freezer.e_freeze = parent->freezer.e_freeze;
5369 if (cgrp->freezer.e_freeze) {
5371 * Set the CGRP_FREEZE flag, so when a process will be
5372 * attached to the child cgroup, it will become frozen.
5373 * At this point the new cgroup is unpopulated, so we can
5374 * consider it frozen immediately.
5376 set_bit(CGRP_FREEZE, &cgrp->flags);
5377 set_bit(CGRP_FROZEN, &cgrp->flags);
5380 spin_lock_irq(&css_set_lock);
5381 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5382 cgrp->ancestor_ids[tcgrp->level] = cgroup_id(tcgrp);
5384 if (tcgrp != cgrp) {
5385 tcgrp->nr_descendants++;
5388 * If the new cgroup is frozen, all ancestor cgroups
5389 * get a new frozen descendant, but their state can't
5390 * change because of this.
5392 if (cgrp->freezer.e_freeze)
5393 tcgrp->freezer.nr_frozen_descendants++;
5396 spin_unlock_irq(&css_set_lock);
5398 if (notify_on_release(parent))
5399 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
5401 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
5402 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
5404 cgrp->self.serial_nr = css_serial_nr_next++;
5406 /* allocation complete, commit to creation */
5407 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
5408 atomic_inc(&root->nr_cgrps);
5409 cgroup_get_live(parent);
5412 * On the default hierarchy, a child doesn't automatically inherit
5413 * subtree_control from the parent. Each is configured manually.
5415 if (!cgroup_on_dfl(cgrp))
5416 cgrp->subtree_control = cgroup_control(cgrp);
5418 cgroup_propagate_control(cgrp);
5423 psi_cgroup_free(cgrp);
5425 kernfs_remove(cgrp->kn);
5427 if (cgroup_on_dfl(parent))
5428 cgroup_rstat_exit(cgrp);
5430 percpu_ref_exit(&cgrp->self.refcnt);
5433 return ERR_PTR(ret);
5436 static bool cgroup_check_hierarchy_limits(struct cgroup *parent)
5438 struct cgroup *cgroup;
5442 lockdep_assert_held(&cgroup_mutex);
5444 for (cgroup = parent; cgroup; cgroup = cgroup_parent(cgroup)) {
5445 if (cgroup->nr_descendants >= cgroup->max_descendants)
5448 if (level > cgroup->max_depth)
5459 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
5461 struct cgroup *parent, *cgrp;
5464 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
5465 if (strchr(name, '\n'))
5468 parent = cgroup_kn_lock_live(parent_kn, false);
5472 if (!cgroup_check_hierarchy_limits(parent)) {
5477 cgrp = cgroup_create(parent, name, mode);
5479 ret = PTR_ERR(cgrp);
5484 * This extra ref will be put in cgroup_free_fn() and guarantees
5485 * that @cgrp->kn is always accessible.
5487 kernfs_get(cgrp->kn);
5489 ret = cgroup_kn_set_ugid(cgrp->kn);
5493 ret = css_populate_dir(&cgrp->self);
5497 ret = cgroup_apply_control_enable(cgrp);
5501 TRACE_CGROUP_PATH(mkdir, cgrp);
5503 /* let's create and online css's */
5504 kernfs_activate(cgrp->kn);
5510 cgroup_destroy_locked(cgrp);
5512 cgroup_kn_unlock(parent_kn);
5517 * This is called when the refcnt of a css is confirmed to be killed.
5518 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5519 * initate destruction and put the css ref from kill_css().
5521 static void css_killed_work_fn(struct work_struct *work)
5523 struct cgroup_subsys_state *css =
5524 container_of(work, struct cgroup_subsys_state, destroy_work);
5526 mutex_lock(&cgroup_mutex);
5531 /* @css can't go away while we're holding cgroup_mutex */
5533 } while (css && atomic_dec_and_test(&css->online_cnt));
5535 mutex_unlock(&cgroup_mutex);
5538 /* css kill confirmation processing requires process context, bounce */
5539 static void css_killed_ref_fn(struct percpu_ref *ref)
5541 struct cgroup_subsys_state *css =
5542 container_of(ref, struct cgroup_subsys_state, refcnt);
5544 if (atomic_dec_and_test(&css->online_cnt)) {
5545 INIT_WORK(&css->destroy_work, css_killed_work_fn);
5546 queue_work(cgroup_destroy_wq, &css->destroy_work);
5551 * kill_css - destroy a css
5552 * @css: css to destroy
5554 * This function initiates destruction of @css by removing cgroup interface
5555 * files and putting its base reference. ->css_offline() will be invoked
5556 * asynchronously once css_tryget_online() is guaranteed to fail and when
5557 * the reference count reaches zero, @css will be released.
5559 static void kill_css(struct cgroup_subsys_state *css)
5561 lockdep_assert_held(&cgroup_mutex);
5563 if (css->flags & CSS_DYING)
5566 css->flags |= CSS_DYING;
5569 * This must happen before css is disassociated with its cgroup.
5570 * See seq_css() for details.
5575 * Killing would put the base ref, but we need to keep it alive
5576 * until after ->css_offline().
5581 * cgroup core guarantees that, by the time ->css_offline() is
5582 * invoked, no new css reference will be given out via
5583 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5584 * proceed to offlining css's because percpu_ref_kill() doesn't
5585 * guarantee that the ref is seen as killed on all CPUs on return.
5587 * Use percpu_ref_kill_and_confirm() to get notifications as each
5588 * css is confirmed to be seen as killed on all CPUs.
5590 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5594 * cgroup_destroy_locked - the first stage of cgroup destruction
5595 * @cgrp: cgroup to be destroyed
5597 * css's make use of percpu refcnts whose killing latency shouldn't be
5598 * exposed to userland and are RCU protected. Also, cgroup core needs to
5599 * guarantee that css_tryget_online() won't succeed by the time
5600 * ->css_offline() is invoked. To satisfy all the requirements,
5601 * destruction is implemented in the following two steps.
5603 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5604 * userland visible parts and start killing the percpu refcnts of
5605 * css's. Set up so that the next stage will be kicked off once all
5606 * the percpu refcnts are confirmed to be killed.
5608 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5609 * rest of destruction. Once all cgroup references are gone, the
5610 * cgroup is RCU-freed.
5612 * This function implements s1. After this step, @cgrp is gone as far as
5613 * the userland is concerned and a new cgroup with the same name may be
5614 * created. As cgroup doesn't care about the names internally, this
5615 * doesn't cause any problem.
5617 static int cgroup_destroy_locked(struct cgroup *cgrp)
5618 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5620 struct cgroup *tcgrp, *parent = cgroup_parent(cgrp);
5621 struct cgroup_subsys_state *css;
5622 struct cgrp_cset_link *link;
5625 lockdep_assert_held(&cgroup_mutex);
5628 * Only migration can raise populated from zero and we're already
5629 * holding cgroup_mutex.
5631 if (cgroup_is_populated(cgrp))
5635 * Make sure there's no live children. We can't test emptiness of
5636 * ->self.children as dead children linger on it while being
5637 * drained; otherwise, "rmdir parent/child parent" may fail.
5639 if (css_has_online_children(&cgrp->self))
5643 * Mark @cgrp and the associated csets dead. The former prevents
5644 * further task migration and child creation by disabling
5645 * cgroup_lock_live_group(). The latter makes the csets ignored by
5646 * the migration path.
5648 cgrp->self.flags &= ~CSS_ONLINE;
5650 spin_lock_irq(&css_set_lock);
5651 list_for_each_entry(link, &cgrp->cset_links, cset_link)
5652 link->cset->dead = true;
5653 spin_unlock_irq(&css_set_lock);
5655 /* initiate massacre of all css's */
5656 for_each_css(css, ssid, cgrp)
5659 /* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
5660 css_clear_dir(&cgrp->self);
5661 kernfs_remove(cgrp->kn);
5663 if (parent && cgroup_is_threaded(cgrp))
5664 parent->nr_threaded_children--;
5666 spin_lock_irq(&css_set_lock);
5667 for (tcgrp = cgroup_parent(cgrp); tcgrp; tcgrp = cgroup_parent(tcgrp)) {
5668 tcgrp->nr_descendants--;
5669 tcgrp->nr_dying_descendants++;
5671 * If the dying cgroup is frozen, decrease frozen descendants
5672 * counters of ancestor cgroups.
5674 if (test_bit(CGRP_FROZEN, &cgrp->flags))
5675 tcgrp->freezer.nr_frozen_descendants--;
5677 spin_unlock_irq(&css_set_lock);
5679 cgroup1_check_for_release(parent);
5681 cgroup_bpf_offline(cgrp);
5683 /* put the base reference */
5684 percpu_ref_kill(&cgrp->self.refcnt);
5689 int cgroup_rmdir(struct kernfs_node *kn)
5691 struct cgroup *cgrp;
5694 cgrp = cgroup_kn_lock_live(kn, false);
5698 ret = cgroup_destroy_locked(cgrp);
5700 TRACE_CGROUP_PATH(rmdir, cgrp);
5702 cgroup_kn_unlock(kn);
5706 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5707 .show_options = cgroup_show_options,
5708 .mkdir = cgroup_mkdir,
5709 .rmdir = cgroup_rmdir,
5710 .show_path = cgroup_show_path,
5713 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5715 struct cgroup_subsys_state *css;
5717 pr_debug("Initializing cgroup subsys %s\n", ss->name);
5719 mutex_lock(&cgroup_mutex);
5721 idr_init(&ss->css_idr);
5722 INIT_LIST_HEAD(&ss->cfts);
5724 /* Create the root cgroup state for this subsystem */
5725 ss->root = &cgrp_dfl_root;
5726 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5727 /* We don't handle early failures gracefully */
5728 BUG_ON(IS_ERR(css));
5729 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5732 * Root csses are never destroyed and we can't initialize
5733 * percpu_ref during early init. Disable refcnting.
5735 css->flags |= CSS_NO_REF;
5738 /* allocation can't be done safely during early init */
5741 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5742 BUG_ON(css->id < 0);
5745 /* Update the init_css_set to contain a subsys
5746 * pointer to this state - since the subsystem is
5747 * newly registered, all tasks and hence the
5748 * init_css_set is in the subsystem's root cgroup. */
5749 init_css_set.subsys[ss->id] = css;
5751 have_fork_callback |= (bool)ss->fork << ss->id;
5752 have_exit_callback |= (bool)ss->exit << ss->id;
5753 have_release_callback |= (bool)ss->release << ss->id;
5754 have_canfork_callback |= (bool)ss->can_fork << ss->id;
5756 /* At system boot, before all subsystems have been
5757 * registered, no tasks have been forked, so we don't
5758 * need to invoke fork callbacks here. */
5759 BUG_ON(!list_empty(&init_task.tasks));
5761 BUG_ON(online_css(css));
5763 mutex_unlock(&cgroup_mutex);
5767 * cgroup_init_early - cgroup initialization at system boot
5769 * Initialize cgroups at system boot, and initialize any
5770 * subsystems that request early init.
5772 int __init cgroup_init_early(void)
5774 static struct cgroup_fs_context __initdata ctx;
5775 struct cgroup_subsys *ss;
5778 ctx.root = &cgrp_dfl_root;
5779 init_cgroup_root(&ctx);
5780 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5782 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5784 for_each_subsys(ss, i) {
5785 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5786 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
5787 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5789 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5790 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5793 ss->name = cgroup_subsys_name[i];
5794 if (!ss->legacy_name)
5795 ss->legacy_name = cgroup_subsys_name[i];
5798 cgroup_init_subsys(ss, true);
5804 * cgroup_init - cgroup initialization
5806 * Register cgroup filesystem and /proc file, and initialize
5807 * any subsystems that didn't request early init.
5809 int __init cgroup_init(void)
5811 struct cgroup_subsys *ss;
5814 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
5815 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
5816 BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
5818 cgroup_rstat_boot();
5821 * The latency of the synchronize_rcu() is too high for cgroups,
5822 * avoid it at the cost of forcing all readers into the slow path.
5824 rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
5826 get_user_ns(init_cgroup_ns.user_ns);
5828 mutex_lock(&cgroup_mutex);
5831 * Add init_css_set to the hash table so that dfl_root can link to
5834 hash_add(css_set_table, &init_css_set.hlist,
5835 css_set_hash(init_css_set.subsys));
5837 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
5839 mutex_unlock(&cgroup_mutex);
5841 for_each_subsys(ss, ssid) {
5842 if (ss->early_init) {
5843 struct cgroup_subsys_state *css =
5844 init_css_set.subsys[ss->id];
5846 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5848 BUG_ON(css->id < 0);
5850 cgroup_init_subsys(ss, false);
5853 list_add_tail(&init_css_set.e_cset_node[ssid],
5854 &cgrp_dfl_root.cgrp.e_csets[ssid]);
5857 * Setting dfl_root subsys_mask needs to consider the
5858 * disabled flag and cftype registration needs kmalloc,
5859 * both of which aren't available during early_init.
5861 if (!cgroup_ssid_enabled(ssid))
5864 if (cgroup1_ssid_disabled(ssid))
5865 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
5868 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5870 /* implicit controllers must be threaded too */
5871 WARN_ON(ss->implicit_on_dfl && !ss->threaded);
5873 if (ss->implicit_on_dfl)
5874 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
5875 else if (!ss->dfl_cftypes)
5876 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
5879 cgrp_dfl_threaded_ss_mask |= 1 << ss->id;
5881 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5882 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5884 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5885 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5889 ss->bind(init_css_set.subsys[ssid]);
5891 mutex_lock(&cgroup_mutex);
5892 css_populate_dir(init_css_set.subsys[ssid]);
5893 mutex_unlock(&cgroup_mutex);
5896 /* init_css_set.subsys[] has been updated, re-hash */
5897 hash_del(&init_css_set.hlist);
5898 hash_add(css_set_table, &init_css_set.hlist,
5899 css_set_hash(init_css_set.subsys));
5901 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5902 WARN_ON(register_filesystem(&cgroup_fs_type));
5903 WARN_ON(register_filesystem(&cgroup2_fs_type));
5904 WARN_ON(!proc_create_single("cgroups", 0, NULL, proc_cgroupstats_show));
5905 #ifdef CONFIG_CPUSETS
5906 WARN_ON(register_filesystem(&cpuset_fs_type));
5912 static int __init cgroup_wq_init(void)
5915 * There isn't much point in executing destruction path in
5916 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5917 * Use 1 for @max_active.
5919 * We would prefer to do this in cgroup_init() above, but that
5920 * is called before init_workqueues(): so leave this until after.
5922 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5923 BUG_ON(!cgroup_destroy_wq);
5926 core_initcall(cgroup_wq_init);
5928 void cgroup_path_from_kernfs_id(u64 id, char *buf, size_t buflen)
5930 struct kernfs_node *kn;
5932 kn = kernfs_find_and_get_node_by_id(cgrp_dfl_root.kf_root, id);
5935 kernfs_path(kn, buf, buflen);
5940 * proc_cgroup_show()
5941 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5942 * - Used for /proc/<pid>/cgroup.
5944 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5945 struct pid *pid, struct task_struct *tsk)
5949 struct cgroup_root *root;
5952 buf = kmalloc(PATH_MAX, GFP_KERNEL);
5956 mutex_lock(&cgroup_mutex);
5957 spin_lock_irq(&css_set_lock);
5959 for_each_root(root) {
5960 struct cgroup_subsys *ss;
5961 struct cgroup *cgrp;
5962 int ssid, count = 0;
5964 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
5967 seq_printf(m, "%d:", root->hierarchy_id);
5968 if (root != &cgrp_dfl_root)
5969 for_each_subsys(ss, ssid)
5970 if (root->subsys_mask & (1 << ssid))
5971 seq_printf(m, "%s%s", count++ ? "," : "",
5973 if (strlen(root->name))
5974 seq_printf(m, "%sname=%s", count ? "," : "",
5978 cgrp = task_cgroup_from_root(tsk, root);
5981 * On traditional hierarchies, all zombie tasks show up as
5982 * belonging to the root cgroup. On the default hierarchy,
5983 * while a zombie doesn't show up in "cgroup.procs" and
5984 * thus can't be migrated, its /proc/PID/cgroup keeps
5985 * reporting the cgroup it belonged to before exiting. If
5986 * the cgroup is removed before the zombie is reaped,
5987 * " (deleted)" is appended to the cgroup path.
5989 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
5990 retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
5991 current->nsproxy->cgroup_ns);
5992 if (retval >= PATH_MAX)
5993 retval = -ENAMETOOLONG;
6002 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
6003 seq_puts(m, " (deleted)\n");
6010 spin_unlock_irq(&css_set_lock);
6011 mutex_unlock(&cgroup_mutex);
6018 * cgroup_fork - initialize cgroup related fields during copy_process()
6019 * @child: pointer to task_struct of forking parent process.
6021 * A task is associated with the init_css_set until cgroup_post_fork()
6022 * attaches it to the target css_set.
6024 void cgroup_fork(struct task_struct *child)
6026 RCU_INIT_POINTER(child->cgroups, &init_css_set);
6027 INIT_LIST_HEAD(&child->cg_list);
6030 static struct cgroup *cgroup_get_from_file(struct file *f)
6032 struct cgroup_subsys_state *css;
6033 struct cgroup *cgrp;
6035 css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
6037 return ERR_CAST(css);
6040 if (!cgroup_on_dfl(cgrp)) {
6042 return ERR_PTR(-EBADF);
6049 * cgroup_css_set_fork - find or create a css_set for a child process
6050 * @kargs: the arguments passed to create the child process
6052 * This functions finds or creates a new css_set which the child
6053 * process will be attached to in cgroup_post_fork(). By default,
6054 * the child process will be given the same css_set as its parent.
6056 * If CLONE_INTO_CGROUP is specified this function will try to find an
6057 * existing css_set which includes the requested cgroup and if not create
6058 * a new css_set that the child will be attached to later. If this function
6059 * succeeds it will hold cgroup_threadgroup_rwsem on return. If
6060 * CLONE_INTO_CGROUP is requested this function will grab cgroup mutex
6061 * before grabbing cgroup_threadgroup_rwsem and will hold a reference
6062 * to the target cgroup.
6064 static int cgroup_css_set_fork(struct kernel_clone_args *kargs)
6065 __acquires(&cgroup_mutex) __acquires(&cgroup_threadgroup_rwsem)
6068 struct cgroup *dst_cgrp = NULL;
6069 struct css_set *cset;
6070 struct super_block *sb;
6073 if (kargs->flags & CLONE_INTO_CGROUP)
6074 mutex_lock(&cgroup_mutex);
6076 cgroup_threadgroup_change_begin(current);
6078 spin_lock_irq(&css_set_lock);
6079 cset = task_css_set(current);
6081 spin_unlock_irq(&css_set_lock);
6083 if (!(kargs->flags & CLONE_INTO_CGROUP)) {
6088 f = fget_raw(kargs->cgroup);
6093 sb = f->f_path.dentry->d_sb;
6095 dst_cgrp = cgroup_get_from_file(f);
6096 if (IS_ERR(dst_cgrp)) {
6097 ret = PTR_ERR(dst_cgrp);
6102 if (cgroup_is_dead(dst_cgrp)) {
6108 * Verify that we the target cgroup is writable for us. This is
6109 * usually done by the vfs layer but since we're not going through
6110 * the vfs layer here we need to do it "manually".
6112 ret = cgroup_may_write(dst_cgrp, sb);
6116 ret = cgroup_attach_permissions(cset->dfl_cgrp, dst_cgrp, sb,
6117 !(kargs->flags & CLONE_THREAD),
6118 current->nsproxy->cgroup_ns);
6122 kargs->cset = find_css_set(cset, dst_cgrp);
6130 kargs->cgrp = dst_cgrp;
6134 cgroup_threadgroup_change_end(current);
6135 mutex_unlock(&cgroup_mutex);
6139 cgroup_put(dst_cgrp);
6142 put_css_set(kargs->cset);
6147 * cgroup_css_set_put_fork - drop references we took during fork
6148 * @kargs: the arguments passed to create the child process
6150 * Drop references to the prepared css_set and target cgroup if
6151 * CLONE_INTO_CGROUP was requested.
6153 static void cgroup_css_set_put_fork(struct kernel_clone_args *kargs)
6154 __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6156 struct cgroup *cgrp = kargs->cgrp;
6157 struct css_set *cset = kargs->cset;
6159 cgroup_threadgroup_change_end(current);
6166 if (kargs->flags & CLONE_INTO_CGROUP) {
6167 mutex_unlock(&cgroup_mutex);
6176 * cgroup_can_fork - called on a new task before the process is exposed
6177 * @child: the child process
6179 * This prepares a new css_set for the child process which the child will
6180 * be attached to in cgroup_post_fork().
6181 * This calls the subsystem can_fork() callbacks. If the cgroup_can_fork()
6182 * callback returns an error, the fork aborts with that error code. This
6183 * allows for a cgroup subsystem to conditionally allow or deny new forks.
6185 int cgroup_can_fork(struct task_struct *child, struct kernel_clone_args *kargs)
6187 struct cgroup_subsys *ss;
6190 ret = cgroup_css_set_fork(kargs);
6194 do_each_subsys_mask(ss, i, have_canfork_callback) {
6195 ret = ss->can_fork(child, kargs->cset);
6198 } while_each_subsys_mask();
6203 for_each_subsys(ss, j) {
6206 if (ss->cancel_fork)
6207 ss->cancel_fork(child, kargs->cset);
6210 cgroup_css_set_put_fork(kargs);
6216 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
6217 * @child: the child process
6218 * @kargs: the arguments passed to create the child process
6220 * This calls the cancel_fork() callbacks if a fork failed *after*
6221 * cgroup_can_fork() succeded and cleans up references we took to
6222 * prepare a new css_set for the child process in cgroup_can_fork().
6224 void cgroup_cancel_fork(struct task_struct *child,
6225 struct kernel_clone_args *kargs)
6227 struct cgroup_subsys *ss;
6230 for_each_subsys(ss, i)
6231 if (ss->cancel_fork)
6232 ss->cancel_fork(child, kargs->cset);
6234 cgroup_css_set_put_fork(kargs);
6238 * cgroup_post_fork - finalize cgroup setup for the child process
6239 * @child: the child process
6241 * Attach the child process to its css_set calling the subsystem fork()
6244 void cgroup_post_fork(struct task_struct *child,
6245 struct kernel_clone_args *kargs)
6246 __releases(&cgroup_threadgroup_rwsem) __releases(&cgroup_mutex)
6248 struct cgroup_subsys *ss;
6249 struct css_set *cset;
6255 spin_lock_irq(&css_set_lock);
6257 /* init tasks are special, only link regular threads */
6258 if (likely(child->pid)) {
6259 WARN_ON_ONCE(!list_empty(&child->cg_list));
6261 css_set_move_task(child, NULL, cset, false);
6268 * If the cgroup has to be frozen, the new task has too. Let's set
6269 * the JOBCTL_TRAP_FREEZE jobctl bit to get the task into the
6272 if (unlikely(cgroup_task_freeze(child))) {
6273 spin_lock(&child->sighand->siglock);
6274 WARN_ON_ONCE(child->frozen);
6275 child->jobctl |= JOBCTL_TRAP_FREEZE;
6276 spin_unlock(&child->sighand->siglock);
6279 * Calling cgroup_update_frozen() isn't required here,
6280 * because it will be called anyway a bit later from
6281 * do_freezer_trap(). So we avoid cgroup's transient switch
6282 * from the frozen state and back.
6286 spin_unlock_irq(&css_set_lock);
6289 * Call ss->fork(). This must happen after @child is linked on
6290 * css_set; otherwise, @child might change state between ->fork()
6291 * and addition to css_set.
6293 do_each_subsys_mask(ss, i, have_fork_callback) {
6295 } while_each_subsys_mask();
6297 /* Make the new cset the root_cset of the new cgroup namespace. */
6298 if (kargs->flags & CLONE_NEWCGROUP) {
6299 struct css_set *rcset = child->nsproxy->cgroup_ns->root_cset;
6302 child->nsproxy->cgroup_ns->root_cset = cset;
6306 cgroup_css_set_put_fork(kargs);
6310 * cgroup_exit - detach cgroup from exiting task
6311 * @tsk: pointer to task_struct of exiting process
6313 * Description: Detach cgroup from @tsk.
6316 void cgroup_exit(struct task_struct *tsk)
6318 struct cgroup_subsys *ss;
6319 struct css_set *cset;
6322 spin_lock_irq(&css_set_lock);
6324 WARN_ON_ONCE(list_empty(&tsk->cg_list));
6325 cset = task_css_set(tsk);
6326 css_set_move_task(tsk, cset, NULL, false);
6327 list_add_tail(&tsk->cg_list, &cset->dying_tasks);
6331 dec_dl_tasks_cs(tsk);
6333 WARN_ON_ONCE(cgroup_task_frozen(tsk));
6334 if (unlikely(cgroup_task_freeze(tsk)))
6335 cgroup_update_frozen(task_dfl_cgroup(tsk));
6337 spin_unlock_irq(&css_set_lock);
6339 /* see cgroup_post_fork() for details */
6340 do_each_subsys_mask(ss, i, have_exit_callback) {
6342 } while_each_subsys_mask();
6345 void cgroup_release(struct task_struct *task)
6347 struct cgroup_subsys *ss;
6350 do_each_subsys_mask(ss, ssid, have_release_callback) {
6352 } while_each_subsys_mask();
6354 spin_lock_irq(&css_set_lock);
6355 css_set_skip_task_iters(task_css_set(task), task);
6356 list_del_init(&task->cg_list);
6357 spin_unlock_irq(&css_set_lock);
6360 void cgroup_free(struct task_struct *task)
6362 struct css_set *cset = task_css_set(task);
6366 static int __init cgroup_disable(char *str)
6368 struct cgroup_subsys *ss;
6372 while ((token = strsep(&str, ",")) != NULL) {
6376 for_each_subsys(ss, i) {
6377 if (strcmp(token, ss->name) &&
6378 strcmp(token, ss->legacy_name))
6381 static_branch_disable(cgroup_subsys_enabled_key[i]);
6382 pr_info("Disabling %s control group subsystem\n",
6388 __setup("cgroup_disable=", cgroup_disable);
6390 void __init __weak enable_debug_cgroup(void) { }
6392 static int __init enable_cgroup_debug(char *str)
6394 cgroup_debug = true;
6395 enable_debug_cgroup();
6398 __setup("cgroup_debug", enable_cgroup_debug);
6401 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
6402 * @dentry: directory dentry of interest
6403 * @ss: subsystem of interest
6405 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
6406 * to get the corresponding css and return it. If such css doesn't exist
6407 * or can't be pinned, an ERR_PTR value is returned.
6409 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
6410 struct cgroup_subsys *ss)
6412 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
6413 struct file_system_type *s_type = dentry->d_sb->s_type;
6414 struct cgroup_subsys_state *css = NULL;
6415 struct cgroup *cgrp;
6417 /* is @dentry a cgroup dir? */
6418 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
6419 !kn || kernfs_type(kn) != KERNFS_DIR)
6420 return ERR_PTR(-EBADF);
6425 * This path doesn't originate from kernfs and @kn could already
6426 * have been or be removed at any point. @kn->priv is RCU
6427 * protected for this access. See css_release_work_fn() for details.
6429 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
6431 css = cgroup_css(cgrp, ss);
6433 if (!css || !css_tryget_online(css))
6434 css = ERR_PTR(-ENOENT);
6441 * css_from_id - lookup css by id
6442 * @id: the cgroup id
6443 * @ss: cgroup subsys to be looked into
6445 * Returns the css if there's valid one with @id, otherwise returns NULL.
6446 * Should be called under rcu_read_lock().
6448 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
6450 WARN_ON_ONCE(!rcu_read_lock_held());
6451 return idr_find(&ss->css_idr, id);
6455 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
6456 * @path: path on the default hierarchy
6458 * Find the cgroup at @path on the default hierarchy, increment its
6459 * reference count and return it. Returns pointer to the found cgroup on
6460 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
6461 * if @path points to a non-directory.
6463 struct cgroup *cgroup_get_from_path(const char *path)
6465 struct kernfs_node *kn;
6466 struct cgroup *cgrp;
6468 mutex_lock(&cgroup_mutex);
6470 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
6472 if (kernfs_type(kn) == KERNFS_DIR) {
6474 cgroup_get_live(cgrp);
6476 cgrp = ERR_PTR(-ENOTDIR);
6480 cgrp = ERR_PTR(-ENOENT);
6483 mutex_unlock(&cgroup_mutex);
6486 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
6489 * cgroup_get_from_fd - get a cgroup pointer from a fd
6490 * @fd: fd obtained by open(cgroup2_dir)
6492 * Find the cgroup from a fd which should be obtained
6493 * by opening a cgroup directory. Returns a pointer to the
6494 * cgroup on success. ERR_PTR is returned if the cgroup
6497 struct cgroup *cgroup_get_from_fd(int fd)
6499 struct cgroup *cgrp;
6504 return ERR_PTR(-EBADF);
6506 cgrp = cgroup_get_from_file(f);
6510 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
6512 static u64 power_of_ten(int power)
6521 * cgroup_parse_float - parse a floating number
6522 * @input: input string
6523 * @dec_shift: number of decimal digits to shift
6526 * Parse a decimal floating point number in @input and store the result in
6527 * @v with decimal point right shifted @dec_shift times. For example, if
6528 * @input is "12.3456" and @dec_shift is 3, *@v will be set to 12345.
6529 * Returns 0 on success, -errno otherwise.
6531 * There's nothing cgroup specific about this function except that it's
6532 * currently the only user.
6534 int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v)
6536 s64 whole, frac = 0;
6537 int fstart = 0, fend = 0, flen;
6539 if (!sscanf(input, "%lld.%n%lld%n", &whole, &fstart, &frac, &fend))
6544 flen = fend > fstart ? fend - fstart : 0;
6545 if (flen < dec_shift)
6546 frac *= power_of_ten(dec_shift - flen);
6548 frac = DIV_ROUND_CLOSEST_ULL(frac, power_of_ten(flen - dec_shift));
6550 *v = whole * power_of_ten(dec_shift) + frac;
6555 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
6556 * definition in cgroup-defs.h.
6558 #ifdef CONFIG_SOCK_CGROUP_DATA
6560 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
6562 DEFINE_SPINLOCK(cgroup_sk_update_lock);
6563 static bool cgroup_sk_alloc_disabled __read_mostly;
6565 void cgroup_sk_alloc_disable(void)
6567 if (cgroup_sk_alloc_disabled)
6569 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
6570 cgroup_sk_alloc_disabled = true;
6575 #define cgroup_sk_alloc_disabled false
6579 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
6581 if (cgroup_sk_alloc_disabled) {
6582 skcd->no_refcnt = 1;
6586 /* Don't associate the sock with unrelated interrupted task's cgroup. */
6593 struct css_set *cset;
6595 cset = task_css_set(current);
6596 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
6597 skcd->val = (unsigned long)cset->dfl_cgrp;
6598 cgroup_bpf_get(cset->dfl_cgrp);
6607 void cgroup_sk_clone(struct sock_cgroup_data *skcd)
6610 if (skcd->no_refcnt)
6613 * We might be cloning a socket which is left in an empty
6614 * cgroup and the cgroup might have already been rmdir'd.
6615 * Don't use cgroup_get_live().
6617 cgroup_get(sock_cgroup_ptr(skcd));
6618 cgroup_bpf_get(sock_cgroup_ptr(skcd));
6622 void cgroup_sk_free(struct sock_cgroup_data *skcd)
6624 struct cgroup *cgrp = sock_cgroup_ptr(skcd);
6626 if (skcd->no_refcnt)
6628 cgroup_bpf_put(cgrp);
6632 #endif /* CONFIG_SOCK_CGROUP_DATA */
6634 #ifdef CONFIG_CGROUP_BPF
6635 int cgroup_bpf_attach(struct cgroup *cgrp,
6636 struct bpf_prog *prog, struct bpf_prog *replace_prog,
6637 struct bpf_cgroup_link *link,
6638 enum bpf_attach_type type,
6643 mutex_lock(&cgroup_mutex);
6644 ret = __cgroup_bpf_attach(cgrp, prog, replace_prog, link, type, flags);
6645 mutex_unlock(&cgroup_mutex);
6649 int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
6650 enum bpf_attach_type type)
6654 mutex_lock(&cgroup_mutex);
6655 ret = __cgroup_bpf_detach(cgrp, prog, NULL, type);
6656 mutex_unlock(&cgroup_mutex);
6660 int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
6661 union bpf_attr __user *uattr)
6665 mutex_lock(&cgroup_mutex);
6666 ret = __cgroup_bpf_query(cgrp, attr, uattr);
6667 mutex_unlock(&cgroup_mutex);
6670 #endif /* CONFIG_CGROUP_BPF */
6673 static ssize_t show_delegatable_files(struct cftype *files, char *buf,
6674 ssize_t size, const char *prefix)
6679 for (cft = files; cft && cft->name[0] != '\0'; cft++) {
6680 if (!(cft->flags & CFTYPE_NS_DELEGATABLE))
6684 ret += snprintf(buf + ret, size - ret, "%s.", prefix);
6686 ret += snprintf(buf + ret, size - ret, "%s\n", cft->name);
6688 if (WARN_ON(ret >= size))
6695 static ssize_t delegate_show(struct kobject *kobj, struct kobj_attribute *attr,
6698 struct cgroup_subsys *ss;
6702 ret = show_delegatable_files(cgroup_base_files, buf, PAGE_SIZE - ret,
6705 for_each_subsys(ss, ssid)
6706 ret += show_delegatable_files(ss->dfl_cftypes, buf + ret,
6708 cgroup_subsys_name[ssid]);
6712 static struct kobj_attribute cgroup_delegate_attr = __ATTR_RO(delegate);
6714 static ssize_t features_show(struct kobject *kobj, struct kobj_attribute *attr,
6717 return snprintf(buf, PAGE_SIZE,
6719 "memory_localevents\n"
6720 "memory_recursiveprot\n");
6722 static struct kobj_attribute cgroup_features_attr = __ATTR_RO(features);
6724 static struct attribute *cgroup_sysfs_attrs[] = {
6725 &cgroup_delegate_attr.attr,
6726 &cgroup_features_attr.attr,
6730 static const struct attribute_group cgroup_sysfs_attr_group = {
6731 .attrs = cgroup_sysfs_attrs,
6735 static int __init cgroup_sysfs_init(void)
6737 return sysfs_create_group(kernel_kobj, &cgroup_sysfs_attr_group);
6739 subsys_initcall(cgroup_sysfs_init);
6741 #endif /* CONFIG_SYSFS */