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/slab.h>
45 #include <linux/spinlock.h>
46 #include <linux/percpu-rwsem.h>
47 #include <linux/string.h>
48 #include <linux/hashtable.h>
49 #include <linux/idr.h>
50 #include <linux/kthread.h>
51 #include <linux/atomic.h>
52 #include <linux/cpuset.h>
53 #include <linux/proc_ns.h>
54 #include <linux/nsproxy.h>
55 #include <linux/file.h>
58 #define CREATE_TRACE_POINTS
59 #include <trace/events/cgroup.h>
61 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
65 * cgroup_mutex is the master lock. Any modification to cgroup or its
66 * hierarchy must be performed while holding it.
68 * css_set_lock protects task->cgroups pointer, the list of css_set
69 * objects, and the chain of tasks off each css_set.
71 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
72 * cgroup.h can use them for lockdep annotations.
74 DEFINE_MUTEX(cgroup_mutex);
75 DEFINE_SPINLOCK(css_set_lock);
77 #ifdef CONFIG_PROVE_RCU
78 EXPORT_SYMBOL_GPL(cgroup_mutex);
79 EXPORT_SYMBOL_GPL(css_set_lock);
83 * Protects cgroup_idr and css_idr so that IDs can be released without
84 * grabbing cgroup_mutex.
86 static DEFINE_SPINLOCK(cgroup_idr_lock);
89 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
90 * against file removal/re-creation across css hiding.
92 static DEFINE_SPINLOCK(cgroup_file_kn_lock);
94 struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
96 #define cgroup_assert_mutex_or_rcu_locked() \
97 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
98 !lockdep_is_held(&cgroup_mutex), \
99 "cgroup_mutex or RCU read lock required");
102 * cgroup destruction makes heavy use of work items and there can be a lot
103 * of concurrent destructions. Use a separate workqueue so that cgroup
104 * destruction work items don't end up filling up max_active of system_wq
105 * which may lead to deadlock.
107 static struct workqueue_struct *cgroup_destroy_wq;
109 /* generate an array of cgroup subsystem pointers */
110 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
111 struct cgroup_subsys *cgroup_subsys[] = {
112 #include <linux/cgroup_subsys.h>
116 /* array of cgroup subsystem names */
117 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
118 static const char *cgroup_subsys_name[] = {
119 #include <linux/cgroup_subsys.h>
123 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
125 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
126 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
127 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
128 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
129 #include <linux/cgroup_subsys.h>
132 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
133 static struct static_key_true *cgroup_subsys_enabled_key[] = {
134 #include <linux/cgroup_subsys.h>
138 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
139 static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
140 #include <linux/cgroup_subsys.h>
145 * The default hierarchy, reserved for the subsystems that are otherwise
146 * unattached - it never has more than a single cgroup, and all tasks are
147 * part of that cgroup.
149 struct cgroup_root cgrp_dfl_root;
150 EXPORT_SYMBOL_GPL(cgrp_dfl_root);
153 * The default hierarchy always exists but is hidden until mounted for the
154 * first time. This is for backward compatibility.
156 static bool cgrp_dfl_visible;
158 /* some controllers are not supported in the default hierarchy */
159 static u16 cgrp_dfl_inhibit_ss_mask;
161 /* some controllers are implicitly enabled on the default hierarchy */
162 static unsigned long cgrp_dfl_implicit_ss_mask;
164 /* The list of hierarchy roots */
165 LIST_HEAD(cgroup_roots);
166 static int cgroup_root_count;
168 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
169 static DEFINE_IDR(cgroup_hierarchy_idr);
172 * Assign a monotonically increasing serial number to csses. It guarantees
173 * cgroups with bigger numbers are newer than those with smaller numbers.
174 * Also, as csses are always appended to the parent's ->children list, it
175 * guarantees that sibling csses are always sorted in the ascending serial
176 * number order on the list. Protected by cgroup_mutex.
178 static u64 css_serial_nr_next = 1;
181 * These bitmask flags indicate whether tasks in the fork and exit paths have
182 * fork/exit handlers to call. This avoids us having to do extra work in the
183 * fork/exit path to check which subsystems have fork/exit callbacks.
185 static u16 have_fork_callback __read_mostly;
186 static u16 have_exit_callback __read_mostly;
187 static u16 have_free_callback __read_mostly;
189 /* cgroup namespace for init task */
190 struct cgroup_namespace init_cgroup_ns = {
191 .count = { .counter = 2, },
192 .user_ns = &init_user_ns,
193 .ns.ops = &cgroupns_operations,
194 .ns.inum = PROC_CGROUP_INIT_INO,
195 .root_cset = &init_css_set,
198 /* Ditto for the can_fork callback. */
199 static u16 have_canfork_callback __read_mostly;
201 static struct file_system_type cgroup2_fs_type;
202 static struct cftype cgroup_base_files[];
204 static int cgroup_apply_control(struct cgroup *cgrp);
205 static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
206 static void css_task_iter_advance(struct css_task_iter *it);
207 static int cgroup_destroy_locked(struct cgroup *cgrp);
208 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
209 struct cgroup_subsys *ss);
210 static void css_release(struct percpu_ref *ref);
211 static void kill_css(struct cgroup_subsys_state *css);
212 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
213 struct cgroup *cgrp, struct cftype cfts[],
217 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
218 * @ssid: subsys ID of interest
220 * cgroup_subsys_enabled() can only be used with literal subsys names which
221 * is fine for individual subsystems but unsuitable for cgroup core. This
222 * is slower static_key_enabled() based test indexed by @ssid.
224 bool cgroup_ssid_enabled(int ssid)
226 if (CGROUP_SUBSYS_COUNT == 0)
229 return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
233 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
234 * @cgrp: the cgroup of interest
236 * The default hierarchy is the v2 interface of cgroup and this function
237 * can be used to test whether a cgroup is on the default hierarchy for
238 * cases where a subsystem should behave differnetly depending on the
241 * The set of behaviors which change on the default hierarchy are still
242 * being determined and the mount option is prefixed with __DEVEL__.
244 * List of changed behaviors:
246 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
247 * and "name" are disallowed.
249 * - When mounting an existing superblock, mount options should match.
251 * - Remount is disallowed.
253 * - rename(2) is disallowed.
255 * - "tasks" is removed. Everything should be at process granularity. Use
256 * "cgroup.procs" instead.
258 * - "cgroup.procs" is not sorted. pids will be unique unless they got
259 * recycled inbetween reads.
261 * - "release_agent" and "notify_on_release" are removed. Replacement
262 * notification mechanism will be implemented.
264 * - "cgroup.clone_children" is removed.
266 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
267 * and its descendants contain no task; otherwise, 1. The file also
268 * generates kernfs notification which can be monitored through poll and
269 * [di]notify when the value of the file changes.
271 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
272 * take masks of ancestors with non-empty cpus/mems, instead of being
273 * moved to an ancestor.
275 * - cpuset: a task can be moved into an empty cpuset, and again it takes
276 * masks of ancestors.
278 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
281 * - blkcg: blk-throttle becomes properly hierarchical.
283 * - debug: disallowed on the default hierarchy.
285 bool cgroup_on_dfl(const struct cgroup *cgrp)
287 return cgrp->root == &cgrp_dfl_root;
290 /* IDR wrappers which synchronize using cgroup_idr_lock */
291 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
296 idr_preload(gfp_mask);
297 spin_lock_bh(&cgroup_idr_lock);
298 ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
299 spin_unlock_bh(&cgroup_idr_lock);
304 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
308 spin_lock_bh(&cgroup_idr_lock);
309 ret = idr_replace(idr, ptr, id);
310 spin_unlock_bh(&cgroup_idr_lock);
314 static void cgroup_idr_remove(struct idr *idr, int id)
316 spin_lock_bh(&cgroup_idr_lock);
318 spin_unlock_bh(&cgroup_idr_lock);
321 static struct cgroup *cgroup_parent(struct cgroup *cgrp)
323 struct cgroup_subsys_state *parent_css = cgrp->self.parent;
326 return container_of(parent_css, struct cgroup, self);
330 /* subsystems visibly enabled on a cgroup */
331 static u16 cgroup_control(struct cgroup *cgrp)
333 struct cgroup *parent = cgroup_parent(cgrp);
334 u16 root_ss_mask = cgrp->root->subsys_mask;
337 return parent->subtree_control;
339 if (cgroup_on_dfl(cgrp))
340 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
341 cgrp_dfl_implicit_ss_mask);
345 /* subsystems enabled on a cgroup */
346 static u16 cgroup_ss_mask(struct cgroup *cgrp)
348 struct cgroup *parent = cgroup_parent(cgrp);
351 return parent->subtree_ss_mask;
353 return cgrp->root->subsys_mask;
357 * cgroup_css - obtain a cgroup's css for the specified subsystem
358 * @cgrp: the cgroup of interest
359 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
361 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
362 * function must be called either under cgroup_mutex or rcu_read_lock() and
363 * the caller is responsible for pinning the returned css if it wants to
364 * keep accessing it outside the said locks. This function may return
365 * %NULL if @cgrp doesn't have @subsys_id enabled.
367 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
368 struct cgroup_subsys *ss)
371 return rcu_dereference_check(cgrp->subsys[ss->id],
372 lockdep_is_held(&cgroup_mutex));
378 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
379 * @cgrp: the cgroup of interest
380 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
382 * Similar to cgroup_css() but returns the effective css, which is defined
383 * as the matching css of the nearest ancestor including self which has @ss
384 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
385 * function is guaranteed to return non-NULL css.
387 static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
388 struct cgroup_subsys *ss)
390 lockdep_assert_held(&cgroup_mutex);
396 * This function is used while updating css associations and thus
397 * can't test the csses directly. Test ss_mask.
399 while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
400 cgrp = cgroup_parent(cgrp);
405 return cgroup_css(cgrp, ss);
409 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
410 * @cgrp: the cgroup of interest
411 * @ss: the subsystem of interest
413 * Find and get the effective css of @cgrp for @ss. The effective css is
414 * defined as the matching css of the nearest ancestor including self which
415 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
416 * the root css is returned, so this function always returns a valid css.
417 * The returned css must be put using css_put().
419 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
420 struct cgroup_subsys *ss)
422 struct cgroup_subsys_state *css;
427 css = cgroup_css(cgrp, ss);
429 if (css && css_tryget_online(css))
431 cgrp = cgroup_parent(cgrp);
434 css = init_css_set.subsys[ss->id];
441 static void cgroup_get(struct cgroup *cgrp)
443 WARN_ON_ONCE(cgroup_is_dead(cgrp));
444 css_get(&cgrp->self);
447 static bool cgroup_tryget(struct cgroup *cgrp)
449 return css_tryget(&cgrp->self);
452 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
454 struct cgroup *cgrp = of->kn->parent->priv;
455 struct cftype *cft = of_cft(of);
458 * This is open and unprotected implementation of cgroup_css().
459 * seq_css() is only called from a kernfs file operation which has
460 * an active reference on the file. Because all the subsystem
461 * files are drained before a css is disassociated with a cgroup,
462 * the matching css from the cgroup's subsys table is guaranteed to
463 * be and stay valid until the enclosing operation is complete.
466 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
470 EXPORT_SYMBOL_GPL(of_css);
473 * for_each_css - iterate all css's of a cgroup
474 * @css: the iteration cursor
475 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
476 * @cgrp: the target cgroup to iterate css's of
478 * Should be called under cgroup_[tree_]mutex.
480 #define for_each_css(css, ssid, cgrp) \
481 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
482 if (!((css) = rcu_dereference_check( \
483 (cgrp)->subsys[(ssid)], \
484 lockdep_is_held(&cgroup_mutex)))) { } \
488 * for_each_e_css - iterate all effective css's of a cgroup
489 * @css: the iteration cursor
490 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
491 * @cgrp: the target cgroup to iterate css's of
493 * Should be called under cgroup_[tree_]mutex.
495 #define for_each_e_css(css, ssid, cgrp) \
496 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
497 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
502 * do_each_subsys_mask - filter for_each_subsys with a bitmask
503 * @ss: the iteration cursor
504 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
505 * @ss_mask: the bitmask
507 * The block will only run for cases where the ssid-th bit (1 << ssid) of
510 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
511 unsigned long __ss_mask = (ss_mask); \
512 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
516 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
517 (ss) = cgroup_subsys[ssid]; \
520 #define while_each_subsys_mask() \
525 /* iterate over child cgrps, lock should be held throughout iteration */
526 #define cgroup_for_each_live_child(child, cgrp) \
527 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
528 if (({ lockdep_assert_held(&cgroup_mutex); \
529 cgroup_is_dead(child); })) \
533 /* walk live descendants in preorder */
534 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
535 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
536 if (({ lockdep_assert_held(&cgroup_mutex); \
537 (dsct) = (d_css)->cgroup; \
538 cgroup_is_dead(dsct); })) \
542 /* walk live descendants in postorder */
543 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
544 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
545 if (({ lockdep_assert_held(&cgroup_mutex); \
546 (dsct) = (d_css)->cgroup; \
547 cgroup_is_dead(dsct); })) \
552 * The default css_set - used by init and its children prior to any
553 * hierarchies being mounted. It contains a pointer to the root state
554 * for each subsystem. Also used to anchor the list of css_sets. Not
555 * reference-counted, to improve performance when child cgroups
556 * haven't been created.
558 struct css_set init_css_set = {
559 .refcount = ATOMIC_INIT(1),
560 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
561 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
562 .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
563 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
564 .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
565 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
568 static int css_set_count = 1; /* 1 for init_css_set */
571 * css_set_populated - does a css_set contain any tasks?
572 * @cset: target css_set
574 static bool css_set_populated(struct css_set *cset)
576 lockdep_assert_held(&css_set_lock);
578 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
582 * cgroup_update_populated - updated populated count of a cgroup
583 * @cgrp: the target cgroup
584 * @populated: inc or dec populated count
586 * One of the css_sets associated with @cgrp is either getting its first
587 * task or losing the last. Update @cgrp->populated_cnt accordingly. The
588 * count is propagated towards root so that a given cgroup's populated_cnt
589 * is zero iff the cgroup and all its descendants don't contain any tasks.
591 * @cgrp's interface file "cgroup.populated" is zero if
592 * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
593 * changes from or to zero, userland is notified that the content of the
594 * interface file has changed. This can be used to detect when @cgrp and
595 * its descendants become populated or empty.
597 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
599 lockdep_assert_held(&css_set_lock);
605 trigger = !cgrp->populated_cnt++;
607 trigger = !--cgrp->populated_cnt;
612 cgroup1_check_for_release(cgrp);
613 cgroup_file_notify(&cgrp->events_file);
615 cgrp = cgroup_parent(cgrp);
620 * css_set_update_populated - update populated state of a css_set
621 * @cset: target css_set
622 * @populated: whether @cset is populated or depopulated
624 * @cset is either getting the first task or losing the last. Update the
625 * ->populated_cnt of all associated cgroups accordingly.
627 static void css_set_update_populated(struct css_set *cset, bool populated)
629 struct cgrp_cset_link *link;
631 lockdep_assert_held(&css_set_lock);
633 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
634 cgroup_update_populated(link->cgrp, populated);
638 * css_set_move_task - move a task from one css_set to another
639 * @task: task being moved
640 * @from_cset: css_set @task currently belongs to (may be NULL)
641 * @to_cset: new css_set @task is being moved to (may be NULL)
642 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
644 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
645 * css_set, @from_cset can be NULL. If @task is being disassociated
646 * instead of moved, @to_cset can be NULL.
648 * This function automatically handles populated_cnt updates and
649 * css_task_iter adjustments but the caller is responsible for managing
650 * @from_cset and @to_cset's reference counts.
652 static void css_set_move_task(struct task_struct *task,
653 struct css_set *from_cset, struct css_set *to_cset,
656 lockdep_assert_held(&css_set_lock);
658 if (to_cset && !css_set_populated(to_cset))
659 css_set_update_populated(to_cset, true);
662 struct css_task_iter *it, *pos;
664 WARN_ON_ONCE(list_empty(&task->cg_list));
667 * @task is leaving, advance task iterators which are
668 * pointing to it so that they can resume at the next
669 * position. Advancing an iterator might remove it from
670 * the list, use safe walk. See css_task_iter_advance*()
673 list_for_each_entry_safe(it, pos, &from_cset->task_iters,
675 if (it->task_pos == &task->cg_list)
676 css_task_iter_advance(it);
678 list_del_init(&task->cg_list);
679 if (!css_set_populated(from_cset))
680 css_set_update_populated(from_cset, false);
682 WARN_ON_ONCE(!list_empty(&task->cg_list));
687 * We are synchronized through cgroup_threadgroup_rwsem
688 * against PF_EXITING setting such that we can't race
689 * against cgroup_exit() changing the css_set to
690 * init_css_set and dropping the old one.
692 WARN_ON_ONCE(task->flags & PF_EXITING);
694 rcu_assign_pointer(task->cgroups, to_cset);
695 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
701 * hash table for cgroup groups. This improves the performance to find
702 * an existing css_set. This hash doesn't (currently) take into
703 * account cgroups in empty hierarchies.
705 #define CSS_SET_HASH_BITS 7
706 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
708 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
710 unsigned long key = 0UL;
711 struct cgroup_subsys *ss;
714 for_each_subsys(ss, i)
715 key += (unsigned long)css[i];
716 key = (key >> 16) ^ key;
721 void put_css_set_locked(struct css_set *cset)
723 struct cgrp_cset_link *link, *tmp_link;
724 struct cgroup_subsys *ss;
727 lockdep_assert_held(&css_set_lock);
729 if (!atomic_dec_and_test(&cset->refcount))
732 /* This css_set is dead. unlink it and release cgroup and css refs */
733 for_each_subsys(ss, ssid) {
734 list_del(&cset->e_cset_node[ssid]);
735 css_put(cset->subsys[ssid]);
737 hash_del(&cset->hlist);
740 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
741 list_del(&link->cset_link);
742 list_del(&link->cgrp_link);
743 if (cgroup_parent(link->cgrp))
744 cgroup_put(link->cgrp);
748 kfree_rcu(cset, rcu_head);
752 * compare_css_sets - helper function for find_existing_css_set().
753 * @cset: candidate css_set being tested
754 * @old_cset: existing css_set for a task
755 * @new_cgrp: cgroup that's being entered by the task
756 * @template: desired set of css pointers in css_set (pre-calculated)
758 * Returns true if "cset" matches "old_cset" except for the hierarchy
759 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
761 static bool compare_css_sets(struct css_set *cset,
762 struct css_set *old_cset,
763 struct cgroup *new_cgrp,
764 struct cgroup_subsys_state *template[])
766 struct list_head *l1, *l2;
769 * On the default hierarchy, there can be csets which are
770 * associated with the same set of cgroups but different csses.
771 * Let's first ensure that csses match.
773 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
777 * Compare cgroup pointers in order to distinguish between
778 * different cgroups in hierarchies. As different cgroups may
779 * share the same effective css, this comparison is always
782 l1 = &cset->cgrp_links;
783 l2 = &old_cset->cgrp_links;
785 struct cgrp_cset_link *link1, *link2;
786 struct cgroup *cgrp1, *cgrp2;
790 /* See if we reached the end - both lists are equal length. */
791 if (l1 == &cset->cgrp_links) {
792 BUG_ON(l2 != &old_cset->cgrp_links);
795 BUG_ON(l2 == &old_cset->cgrp_links);
797 /* Locate the cgroups associated with these links. */
798 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
799 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
802 /* Hierarchies should be linked in the same order. */
803 BUG_ON(cgrp1->root != cgrp2->root);
806 * If this hierarchy is the hierarchy of the cgroup
807 * that's changing, then we need to check that this
808 * css_set points to the new cgroup; if it's any other
809 * hierarchy, then this css_set should point to the
810 * same cgroup as the old css_set.
812 if (cgrp1->root == new_cgrp->root) {
813 if (cgrp1 != new_cgrp)
824 * find_existing_css_set - init css array and find the matching css_set
825 * @old_cset: the css_set that we're using before the cgroup transition
826 * @cgrp: the cgroup that we're moving into
827 * @template: out param for the new set of csses, should be clear on entry
829 static struct css_set *find_existing_css_set(struct css_set *old_cset,
831 struct cgroup_subsys_state *template[])
833 struct cgroup_root *root = cgrp->root;
834 struct cgroup_subsys *ss;
835 struct css_set *cset;
840 * Build the set of subsystem state objects that we want to see in the
841 * new css_set. while subsystems can change globally, the entries here
842 * won't change, so no need for locking.
844 for_each_subsys(ss, i) {
845 if (root->subsys_mask & (1UL << i)) {
847 * @ss is in this hierarchy, so we want the
848 * effective css from @cgrp.
850 template[i] = cgroup_e_css(cgrp, ss);
853 * @ss is not in this hierarchy, so we don't want
856 template[i] = old_cset->subsys[i];
860 key = css_set_hash(template);
861 hash_for_each_possible(css_set_table, cset, hlist, key) {
862 if (!compare_css_sets(cset, old_cset, cgrp, template))
865 /* This css_set matches what we need */
869 /* No existing cgroup group matched */
873 static void free_cgrp_cset_links(struct list_head *links_to_free)
875 struct cgrp_cset_link *link, *tmp_link;
877 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
878 list_del(&link->cset_link);
884 * allocate_cgrp_cset_links - allocate cgrp_cset_links
885 * @count: the number of links to allocate
886 * @tmp_links: list_head the allocated links are put on
888 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
889 * through ->cset_link. Returns 0 on success or -errno.
891 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
893 struct cgrp_cset_link *link;
896 INIT_LIST_HEAD(tmp_links);
898 for (i = 0; i < count; i++) {
899 link = kzalloc(sizeof(*link), GFP_KERNEL);
901 free_cgrp_cset_links(tmp_links);
904 list_add(&link->cset_link, tmp_links);
910 * link_css_set - a helper function to link a css_set to a cgroup
911 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
912 * @cset: the css_set to be linked
913 * @cgrp: the destination cgroup
915 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
918 struct cgrp_cset_link *link;
920 BUG_ON(list_empty(tmp_links));
922 if (cgroup_on_dfl(cgrp))
923 cset->dfl_cgrp = cgrp;
925 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
930 * Always add links to the tail of the lists so that the lists are
931 * in choronological order.
933 list_move_tail(&link->cset_link, &cgrp->cset_links);
934 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
936 if (cgroup_parent(cgrp))
941 * find_css_set - return a new css_set with one cgroup updated
942 * @old_cset: the baseline css_set
943 * @cgrp: the cgroup to be updated
945 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
946 * substituted into the appropriate hierarchy.
948 static struct css_set *find_css_set(struct css_set *old_cset,
951 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
952 struct css_set *cset;
953 struct list_head tmp_links;
954 struct cgrp_cset_link *link;
955 struct cgroup_subsys *ss;
959 lockdep_assert_held(&cgroup_mutex);
961 /* First see if we already have a cgroup group that matches
963 spin_lock_irq(&css_set_lock);
964 cset = find_existing_css_set(old_cset, cgrp, template);
967 spin_unlock_irq(&css_set_lock);
972 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
976 /* Allocate all the cgrp_cset_link objects that we'll need */
977 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
982 atomic_set(&cset->refcount, 1);
983 INIT_LIST_HEAD(&cset->tasks);
984 INIT_LIST_HEAD(&cset->mg_tasks);
985 INIT_LIST_HEAD(&cset->task_iters);
986 INIT_HLIST_NODE(&cset->hlist);
987 INIT_LIST_HEAD(&cset->cgrp_links);
988 INIT_LIST_HEAD(&cset->mg_preload_node);
989 INIT_LIST_HEAD(&cset->mg_node);
991 /* Copy the set of subsystem state objects generated in
992 * find_existing_css_set() */
993 memcpy(cset->subsys, template, sizeof(cset->subsys));
995 spin_lock_irq(&css_set_lock);
996 /* Add reference counts and links from the new css_set. */
997 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
998 struct cgroup *c = link->cgrp;
1000 if (c->root == cgrp->root)
1002 link_css_set(&tmp_links, cset, c);
1005 BUG_ON(!list_empty(&tmp_links));
1009 /* Add @cset to the hash table */
1010 key = css_set_hash(cset->subsys);
1011 hash_add(css_set_table, &cset->hlist, key);
1013 for_each_subsys(ss, ssid) {
1014 struct cgroup_subsys_state *css = cset->subsys[ssid];
1016 list_add_tail(&cset->e_cset_node[ssid],
1017 &css->cgroup->e_csets[ssid]);
1021 spin_unlock_irq(&css_set_lock);
1026 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1028 struct cgroup *root_cgrp = kf_root->kn->priv;
1030 return root_cgrp->root;
1033 static int cgroup_init_root_id(struct cgroup_root *root)
1037 lockdep_assert_held(&cgroup_mutex);
1039 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1043 root->hierarchy_id = id;
1047 static void cgroup_exit_root_id(struct cgroup_root *root)
1049 lockdep_assert_held(&cgroup_mutex);
1051 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1054 void cgroup_free_root(struct cgroup_root *root)
1057 idr_destroy(&root->cgroup_idr);
1062 static void cgroup_destroy_root(struct cgroup_root *root)
1064 struct cgroup *cgrp = &root->cgrp;
1065 struct cgrp_cset_link *link, *tmp_link;
1067 trace_cgroup_destroy_root(root);
1069 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1071 BUG_ON(atomic_read(&root->nr_cgrps));
1072 BUG_ON(!list_empty(&cgrp->self.children));
1074 /* Rebind all subsystems back to the default hierarchy */
1075 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1078 * Release all the links from cset_links to this hierarchy's
1081 spin_lock_irq(&css_set_lock);
1083 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1084 list_del(&link->cset_link);
1085 list_del(&link->cgrp_link);
1089 spin_unlock_irq(&css_set_lock);
1091 if (!list_empty(&root->root_list)) {
1092 list_del(&root->root_list);
1093 cgroup_root_count--;
1096 cgroup_exit_root_id(root);
1098 mutex_unlock(&cgroup_mutex);
1100 kernfs_destroy_root(root->kf_root);
1101 cgroup_free_root(root);
1105 * look up cgroup associated with current task's cgroup namespace on the
1106 * specified hierarchy
1108 static struct cgroup *
1109 current_cgns_cgroup_from_root(struct cgroup_root *root)
1111 struct cgroup *res = NULL;
1112 struct css_set *cset;
1114 lockdep_assert_held(&css_set_lock);
1118 cset = current->nsproxy->cgroup_ns->root_cset;
1119 if (cset == &init_css_set) {
1122 struct cgrp_cset_link *link;
1124 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1125 struct cgroup *c = link->cgrp;
1127 if (c->root == root) {
1139 /* look up cgroup associated with given css_set on the specified hierarchy */
1140 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1141 struct cgroup_root *root)
1143 struct cgroup *res = NULL;
1145 lockdep_assert_held(&cgroup_mutex);
1146 lockdep_assert_held(&css_set_lock);
1148 if (cset == &init_css_set) {
1151 struct cgrp_cset_link *link;
1153 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1154 struct cgroup *c = link->cgrp;
1156 if (c->root == root) {
1168 * Return the cgroup for "task" from the given hierarchy. Must be
1169 * called with cgroup_mutex and css_set_lock held.
1171 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1172 struct cgroup_root *root)
1175 * No need to lock the task - since we hold cgroup_mutex the
1176 * task can't change groups, so the only thing that can happen
1177 * is that it exits and its css is set back to init_css_set.
1179 return cset_cgroup_from_root(task_css_set(task), root);
1183 * A task must hold cgroup_mutex to modify cgroups.
1185 * Any task can increment and decrement the count field without lock.
1186 * So in general, code holding cgroup_mutex can't rely on the count
1187 * field not changing. However, if the count goes to zero, then only
1188 * cgroup_attach_task() can increment it again. Because a count of zero
1189 * means that no tasks are currently attached, therefore there is no
1190 * way a task attached to that cgroup can fork (the other way to
1191 * increment the count). So code holding cgroup_mutex can safely
1192 * assume that if the count is zero, it will stay zero. Similarly, if
1193 * a task holds cgroup_mutex on a cgroup with zero count, it
1194 * knows that the cgroup won't be removed, as cgroup_rmdir()
1197 * A cgroup can only be deleted if both its 'count' of using tasks
1198 * is zero, and its list of 'children' cgroups is empty. Since all
1199 * tasks in the system use _some_ cgroup, and since there is always at
1200 * least one task in the system (init, pid == 1), therefore, root cgroup
1201 * always has either children cgroups and/or using tasks. So we don't
1202 * need a special hack to ensure that root cgroup cannot be deleted.
1204 * P.S. One more locking exception. RCU is used to guard the
1205 * update of a tasks cgroup pointer by cgroup_attach_task()
1208 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1210 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1213 struct cgroup_subsys *ss = cft->ss;
1215 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1216 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
1217 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
1218 cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1221 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1226 * cgroup_file_mode - deduce file mode of a control file
1227 * @cft: the control file in question
1229 * S_IRUGO for read, S_IWUSR for write.
1231 static umode_t cgroup_file_mode(const struct cftype *cft)
1235 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1238 if (cft->write_u64 || cft->write_s64 || cft->write) {
1239 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1249 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1250 * @subtree_control: the new subtree_control mask to consider
1251 * @this_ss_mask: available subsystems
1253 * On the default hierarchy, a subsystem may request other subsystems to be
1254 * enabled together through its ->depends_on mask. In such cases, more
1255 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1257 * This function calculates which subsystems need to be enabled if
1258 * @subtree_control is to be applied while restricted to @this_ss_mask.
1260 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1262 u16 cur_ss_mask = subtree_control;
1263 struct cgroup_subsys *ss;
1266 lockdep_assert_held(&cgroup_mutex);
1268 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1271 u16 new_ss_mask = cur_ss_mask;
1273 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1274 new_ss_mask |= ss->depends_on;
1275 } while_each_subsys_mask();
1278 * Mask out subsystems which aren't available. This can
1279 * happen only if some depended-upon subsystems were bound
1280 * to non-default hierarchies.
1282 new_ss_mask &= this_ss_mask;
1284 if (new_ss_mask == cur_ss_mask)
1286 cur_ss_mask = new_ss_mask;
1293 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1294 * @kn: the kernfs_node being serviced
1296 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1297 * the method finishes if locking succeeded. Note that once this function
1298 * returns the cgroup returned by cgroup_kn_lock_live() may become
1299 * inaccessible any time. If the caller intends to continue to access the
1300 * cgroup, it should pin it before invoking this function.
1302 void cgroup_kn_unlock(struct kernfs_node *kn)
1304 struct cgroup *cgrp;
1306 if (kernfs_type(kn) == KERNFS_DIR)
1309 cgrp = kn->parent->priv;
1311 mutex_unlock(&cgroup_mutex);
1313 kernfs_unbreak_active_protection(kn);
1318 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1319 * @kn: the kernfs_node being serviced
1320 * @drain_offline: perform offline draining on the cgroup
1322 * This helper is to be used by a cgroup kernfs method currently servicing
1323 * @kn. It breaks the active protection, performs cgroup locking and
1324 * verifies that the associated cgroup is alive. Returns the cgroup if
1325 * alive; otherwise, %NULL. A successful return should be undone by a
1326 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1327 * cgroup is drained of offlining csses before return.
1329 * Any cgroup kernfs method implementation which requires locking the
1330 * associated cgroup should use this helper. It avoids nesting cgroup
1331 * locking under kernfs active protection and allows all kernfs operations
1332 * including self-removal.
1334 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1336 struct cgroup *cgrp;
1338 if (kernfs_type(kn) == KERNFS_DIR)
1341 cgrp = kn->parent->priv;
1344 * We're gonna grab cgroup_mutex which nests outside kernfs
1345 * active_ref. cgroup liveliness check alone provides enough
1346 * protection against removal. Ensure @cgrp stays accessible and
1347 * break the active_ref protection.
1349 if (!cgroup_tryget(cgrp))
1351 kernfs_break_active_protection(kn);
1354 cgroup_lock_and_drain_offline(cgrp);
1356 mutex_lock(&cgroup_mutex);
1358 if (!cgroup_is_dead(cgrp))
1361 cgroup_kn_unlock(kn);
1365 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1367 char name[CGROUP_FILE_NAME_MAX];
1369 lockdep_assert_held(&cgroup_mutex);
1371 if (cft->file_offset) {
1372 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1373 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1375 spin_lock_irq(&cgroup_file_kn_lock);
1377 spin_unlock_irq(&cgroup_file_kn_lock);
1380 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1384 * css_clear_dir - remove subsys files in a cgroup directory
1387 static void css_clear_dir(struct cgroup_subsys_state *css)
1389 struct cgroup *cgrp = css->cgroup;
1390 struct cftype *cfts;
1392 if (!(css->flags & CSS_VISIBLE))
1395 css->flags &= ~CSS_VISIBLE;
1397 list_for_each_entry(cfts, &css->ss->cfts, node)
1398 cgroup_addrm_files(css, cgrp, cfts, false);
1402 * css_populate_dir - create subsys files in a cgroup directory
1405 * On failure, no file is added.
1407 static int css_populate_dir(struct cgroup_subsys_state *css)
1409 struct cgroup *cgrp = css->cgroup;
1410 struct cftype *cfts, *failed_cfts;
1413 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1417 if (cgroup_on_dfl(cgrp))
1418 cfts = cgroup_base_files;
1420 cfts = cgroup1_base_files;
1422 return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1425 list_for_each_entry(cfts, &css->ss->cfts, node) {
1426 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1433 css->flags |= CSS_VISIBLE;
1437 list_for_each_entry(cfts, &css->ss->cfts, node) {
1438 if (cfts == failed_cfts)
1440 cgroup_addrm_files(css, cgrp, cfts, false);
1445 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1447 struct cgroup *dcgrp = &dst_root->cgrp;
1448 struct cgroup_subsys *ss;
1451 lockdep_assert_held(&cgroup_mutex);
1453 do_each_subsys_mask(ss, ssid, ss_mask) {
1455 * If @ss has non-root csses attached to it, can't move.
1456 * If @ss is an implicit controller, it is exempt from this
1457 * rule and can be stolen.
1459 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1460 !ss->implicit_on_dfl)
1463 /* can't move between two non-dummy roots either */
1464 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1466 } while_each_subsys_mask();
1468 do_each_subsys_mask(ss, ssid, ss_mask) {
1469 struct cgroup_root *src_root = ss->root;
1470 struct cgroup *scgrp = &src_root->cgrp;
1471 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1472 struct css_set *cset;
1474 WARN_ON(!css || cgroup_css(dcgrp, ss));
1476 /* disable from the source */
1477 src_root->subsys_mask &= ~(1 << ssid);
1478 WARN_ON(cgroup_apply_control(scgrp));
1479 cgroup_finalize_control(scgrp, 0);
1482 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1483 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1484 ss->root = dst_root;
1485 css->cgroup = dcgrp;
1487 spin_lock_irq(&css_set_lock);
1488 hash_for_each(css_set_table, i, cset, hlist)
1489 list_move_tail(&cset->e_cset_node[ss->id],
1490 &dcgrp->e_csets[ss->id]);
1491 spin_unlock_irq(&css_set_lock);
1493 /* default hierarchy doesn't enable controllers by default */
1494 dst_root->subsys_mask |= 1 << ssid;
1495 if (dst_root == &cgrp_dfl_root) {
1496 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1498 dcgrp->subtree_control |= 1 << ssid;
1499 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1502 ret = cgroup_apply_control(dcgrp);
1504 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1509 } while_each_subsys_mask();
1511 kernfs_activate(dcgrp->kn);
1515 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1516 struct kernfs_root *kf_root)
1520 struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1521 struct cgroup *ns_cgroup;
1523 buf = kmalloc(PATH_MAX, GFP_KERNEL);
1527 spin_lock_irq(&css_set_lock);
1528 ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1529 len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1530 spin_unlock_irq(&css_set_lock);
1532 if (len >= PATH_MAX)
1535 seq_escape(sf, buf, " \t\n\\");
1542 static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
1544 pr_err("remount is not allowed\n");
1549 * To reduce the fork() overhead for systems that are not actually using
1550 * their cgroups capability, we don't maintain the lists running through
1551 * each css_set to its tasks until we see the list actually used - in other
1552 * words after the first mount.
1554 static bool use_task_css_set_links __read_mostly;
1556 static void cgroup_enable_task_cg_lists(void)
1558 struct task_struct *p, *g;
1560 spin_lock_irq(&css_set_lock);
1562 if (use_task_css_set_links)
1565 use_task_css_set_links = true;
1568 * We need tasklist_lock because RCU is not safe against
1569 * while_each_thread(). Besides, a forking task that has passed
1570 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1571 * is not guaranteed to have its child immediately visible in the
1572 * tasklist if we walk through it with RCU.
1574 read_lock(&tasklist_lock);
1575 do_each_thread(g, p) {
1576 WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1577 task_css_set(p) != &init_css_set);
1580 * We should check if the process is exiting, otherwise
1581 * it will race with cgroup_exit() in that the list
1582 * entry won't be deleted though the process has exited.
1583 * Do it while holding siglock so that we don't end up
1584 * racing against cgroup_exit().
1586 * Interrupts were already disabled while acquiring
1587 * the css_set_lock, so we do not need to disable it
1588 * again when acquiring the sighand->siglock here.
1590 spin_lock(&p->sighand->siglock);
1591 if (!(p->flags & PF_EXITING)) {
1592 struct css_set *cset = task_css_set(p);
1594 if (!css_set_populated(cset))
1595 css_set_update_populated(cset, true);
1596 list_add_tail(&p->cg_list, &cset->tasks);
1599 spin_unlock(&p->sighand->siglock);
1600 } while_each_thread(g, p);
1601 read_unlock(&tasklist_lock);
1603 spin_unlock_irq(&css_set_lock);
1606 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1608 struct cgroup_subsys *ss;
1611 INIT_LIST_HEAD(&cgrp->self.sibling);
1612 INIT_LIST_HEAD(&cgrp->self.children);
1613 INIT_LIST_HEAD(&cgrp->cset_links);
1614 INIT_LIST_HEAD(&cgrp->pidlists);
1615 mutex_init(&cgrp->pidlist_mutex);
1616 cgrp->self.cgroup = cgrp;
1617 cgrp->self.flags |= CSS_ONLINE;
1619 for_each_subsys(ss, ssid)
1620 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1622 init_waitqueue_head(&cgrp->offline_waitq);
1623 INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1626 void init_cgroup_root(struct cgroup_root *root, struct cgroup_sb_opts *opts)
1628 struct cgroup *cgrp = &root->cgrp;
1630 INIT_LIST_HEAD(&root->root_list);
1631 atomic_set(&root->nr_cgrps, 1);
1633 init_cgroup_housekeeping(cgrp);
1634 idr_init(&root->cgroup_idr);
1636 root->flags = opts->flags;
1637 if (opts->release_agent)
1638 strcpy(root->release_agent_path, opts->release_agent);
1640 strcpy(root->name, opts->name);
1641 if (opts->cpuset_clone_children)
1642 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1645 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1647 LIST_HEAD(tmp_links);
1648 struct cgroup *root_cgrp = &root->cgrp;
1649 struct kernfs_syscall_ops *kf_sops;
1650 struct css_set *cset;
1653 lockdep_assert_held(&cgroup_mutex);
1655 ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL);
1658 root_cgrp->id = ret;
1659 root_cgrp->ancestor_ids[0] = ret;
1661 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0,
1667 * We're accessing css_set_count without locking css_set_lock here,
1668 * but that's OK - it can only be increased by someone holding
1669 * cgroup_lock, and that's us. Later rebinding may disable
1670 * controllers on the default hierarchy and thus create new csets,
1671 * which can't be more than the existing ones. Allocate 2x.
1673 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
1677 ret = cgroup_init_root_id(root);
1681 kf_sops = root == &cgrp_dfl_root ?
1682 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
1684 root->kf_root = kernfs_create_root(kf_sops,
1685 KERNFS_ROOT_CREATE_DEACTIVATED,
1687 if (IS_ERR(root->kf_root)) {
1688 ret = PTR_ERR(root->kf_root);
1691 root_cgrp->kn = root->kf_root->kn;
1693 ret = css_populate_dir(&root_cgrp->self);
1697 ret = rebind_subsystems(root, ss_mask);
1701 trace_cgroup_setup_root(root);
1704 * There must be no failure case after here, since rebinding takes
1705 * care of subsystems' refcounts, which are explicitly dropped in
1706 * the failure exit path.
1708 list_add(&root->root_list, &cgroup_roots);
1709 cgroup_root_count++;
1712 * Link the root cgroup in this hierarchy into all the css_set
1715 spin_lock_irq(&css_set_lock);
1716 hash_for_each(css_set_table, i, cset, hlist) {
1717 link_css_set(&tmp_links, cset, root_cgrp);
1718 if (css_set_populated(cset))
1719 cgroup_update_populated(root_cgrp, true);
1721 spin_unlock_irq(&css_set_lock);
1723 BUG_ON(!list_empty(&root_cgrp->self.children));
1724 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
1726 kernfs_activate(root_cgrp->kn);
1731 kernfs_destroy_root(root->kf_root);
1732 root->kf_root = NULL;
1734 cgroup_exit_root_id(root);
1736 percpu_ref_exit(&root_cgrp->self.refcnt);
1738 free_cgrp_cset_links(&tmp_links);
1742 struct dentry *cgroup_do_mount(struct file_system_type *fs_type, int flags,
1743 struct cgroup_root *root, unsigned long magic,
1744 struct cgroup_namespace *ns)
1746 struct dentry *dentry;
1749 dentry = kernfs_mount(fs_type, flags, root->kf_root, magic, &new_sb);
1752 * In non-init cgroup namespace, instead of root cgroup's dentry,
1753 * we return the dentry corresponding to the cgroupns->root_cgrp.
1755 if (!IS_ERR(dentry) && ns != &init_cgroup_ns) {
1756 struct dentry *nsdentry;
1757 struct cgroup *cgrp;
1759 mutex_lock(&cgroup_mutex);
1760 spin_lock_irq(&css_set_lock);
1762 cgrp = cset_cgroup_from_root(ns->root_cset, root);
1764 spin_unlock_irq(&css_set_lock);
1765 mutex_unlock(&cgroup_mutex);
1767 nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb);
1772 if (IS_ERR(dentry) || !new_sb)
1773 cgroup_put(&root->cgrp);
1778 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1779 int flags, const char *unused_dev_name,
1782 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
1783 struct dentry *dentry;
1787 /* Check if the caller has permission to mount. */
1788 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) {
1790 return ERR_PTR(-EPERM);
1794 * The first time anyone tries to mount a cgroup, enable the list
1795 * linking each css_set to its tasks and fix up all existing tasks.
1797 if (!use_task_css_set_links)
1798 cgroup_enable_task_cg_lists();
1800 if (fs_type == &cgroup2_fs_type) {
1802 pr_err("cgroup2: unknown option \"%s\"\n", (char *)data);
1804 return ERR_PTR(-EINVAL);
1806 cgrp_dfl_visible = true;
1807 cgroup_get(&cgrp_dfl_root.cgrp);
1809 dentry = cgroup_do_mount(&cgroup2_fs_type, flags, &cgrp_dfl_root,
1810 CGROUP2_SUPER_MAGIC, ns);
1812 dentry = cgroup1_mount(&cgroup_fs_type, flags, data,
1813 CGROUP_SUPER_MAGIC, ns);
1820 static void cgroup_kill_sb(struct super_block *sb)
1822 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
1823 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1826 * If @root doesn't have any mounts or children, start killing it.
1827 * This prevents new mounts by disabling percpu_ref_tryget_live().
1828 * cgroup_mount() may wait for @root's release.
1830 * And don't kill the default root.
1832 if (!list_empty(&root->cgrp.self.children) ||
1833 root == &cgrp_dfl_root)
1834 cgroup_put(&root->cgrp);
1836 percpu_ref_kill(&root->cgrp.self.refcnt);
1841 struct file_system_type cgroup_fs_type = {
1843 .mount = cgroup_mount,
1844 .kill_sb = cgroup_kill_sb,
1845 .fs_flags = FS_USERNS_MOUNT,
1848 static struct file_system_type cgroup2_fs_type = {
1850 .mount = cgroup_mount,
1851 .kill_sb = cgroup_kill_sb,
1852 .fs_flags = FS_USERNS_MOUNT,
1855 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
1856 struct cgroup_namespace *ns)
1858 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
1860 return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
1863 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
1864 struct cgroup_namespace *ns)
1868 mutex_lock(&cgroup_mutex);
1869 spin_lock_irq(&css_set_lock);
1871 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
1873 spin_unlock_irq(&css_set_lock);
1874 mutex_unlock(&cgroup_mutex);
1878 EXPORT_SYMBOL_GPL(cgroup_path_ns);
1881 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1882 * @task: target task
1883 * @buf: the buffer to write the path into
1884 * @buflen: the length of the buffer
1886 * Determine @task's cgroup on the first (the one with the lowest non-zero
1887 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1888 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1889 * cgroup controller callbacks.
1891 * Return value is the same as kernfs_path().
1893 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
1895 struct cgroup_root *root;
1896 struct cgroup *cgrp;
1897 int hierarchy_id = 1;
1900 mutex_lock(&cgroup_mutex);
1901 spin_lock_irq(&css_set_lock);
1903 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
1906 cgrp = task_cgroup_from_root(task, root);
1907 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
1909 /* if no hierarchy exists, everyone is in "/" */
1910 ret = strlcpy(buf, "/", buflen);
1913 spin_unlock_irq(&css_set_lock);
1914 mutex_unlock(&cgroup_mutex);
1917 EXPORT_SYMBOL_GPL(task_cgroup_path);
1919 /* used to track tasks and other necessary states during migration */
1920 struct cgroup_taskset {
1921 /* the src and dst cset list running through cset->mg_node */
1922 struct list_head src_csets;
1923 struct list_head dst_csets;
1925 /* the subsys currently being processed */
1929 * Fields for cgroup_taskset_*() iteration.
1931 * Before migration is committed, the target migration tasks are on
1932 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
1933 * the csets on ->dst_csets. ->csets point to either ->src_csets
1934 * or ->dst_csets depending on whether migration is committed.
1936 * ->cur_csets and ->cur_task point to the current task position
1939 struct list_head *csets;
1940 struct css_set *cur_cset;
1941 struct task_struct *cur_task;
1944 #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \
1945 .src_csets = LIST_HEAD_INIT(tset.src_csets), \
1946 .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \
1947 .csets = &tset.src_csets, \
1951 * cgroup_taskset_add - try to add a migration target task to a taskset
1952 * @task: target task
1953 * @tset: target taskset
1955 * Add @task, which is a migration target, to @tset. This function becomes
1956 * noop if @task doesn't need to be migrated. @task's css_set should have
1957 * been added as a migration source and @task->cg_list will be moved from
1958 * the css_set's tasks list to mg_tasks one.
1960 static void cgroup_taskset_add(struct task_struct *task,
1961 struct cgroup_taskset *tset)
1963 struct css_set *cset;
1965 lockdep_assert_held(&css_set_lock);
1967 /* @task either already exited or can't exit until the end */
1968 if (task->flags & PF_EXITING)
1971 /* leave @task alone if post_fork() hasn't linked it yet */
1972 if (list_empty(&task->cg_list))
1975 cset = task_css_set(task);
1976 if (!cset->mg_src_cgrp)
1979 list_move_tail(&task->cg_list, &cset->mg_tasks);
1980 if (list_empty(&cset->mg_node))
1981 list_add_tail(&cset->mg_node, &tset->src_csets);
1982 if (list_empty(&cset->mg_dst_cset->mg_node))
1983 list_move_tail(&cset->mg_dst_cset->mg_node,
1988 * cgroup_taskset_first - reset taskset and return the first task
1989 * @tset: taskset of interest
1990 * @dst_cssp: output variable for the destination css
1992 * @tset iteration is initialized and the first task is returned.
1994 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
1995 struct cgroup_subsys_state **dst_cssp)
1997 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
1998 tset->cur_task = NULL;
2000 return cgroup_taskset_next(tset, dst_cssp);
2004 * cgroup_taskset_next - iterate to the next task in taskset
2005 * @tset: taskset of interest
2006 * @dst_cssp: output variable for the destination css
2008 * Return the next task in @tset. Iteration must have been initialized
2009 * with cgroup_taskset_first().
2011 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2012 struct cgroup_subsys_state **dst_cssp)
2014 struct css_set *cset = tset->cur_cset;
2015 struct task_struct *task = tset->cur_task;
2017 while (&cset->mg_node != tset->csets) {
2019 task = list_first_entry(&cset->mg_tasks,
2020 struct task_struct, cg_list);
2022 task = list_next_entry(task, cg_list);
2024 if (&task->cg_list != &cset->mg_tasks) {
2025 tset->cur_cset = cset;
2026 tset->cur_task = task;
2029 * This function may be called both before and
2030 * after cgroup_taskset_migrate(). The two cases
2031 * can be distinguished by looking at whether @cset
2032 * has its ->mg_dst_cset set.
2034 if (cset->mg_dst_cset)
2035 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2037 *dst_cssp = cset->subsys[tset->ssid];
2042 cset = list_next_entry(cset, mg_node);
2050 * cgroup_taskset_migrate - migrate a taskset
2051 * @tset: taget taskset
2052 * @root: cgroup root the migration is taking place on
2054 * Migrate tasks in @tset as setup by migration preparation functions.
2055 * This function fails iff one of the ->can_attach callbacks fails and
2056 * guarantees that either all or none of the tasks in @tset are migrated.
2057 * @tset is consumed regardless of success.
2059 static int cgroup_taskset_migrate(struct cgroup_taskset *tset,
2060 struct cgroup_root *root)
2062 struct cgroup_subsys *ss;
2063 struct task_struct *task, *tmp_task;
2064 struct css_set *cset, *tmp_cset;
2065 int ssid, failed_ssid, ret;
2067 /* methods shouldn't be called if no task is actually migrating */
2068 if (list_empty(&tset->src_csets))
2071 /* check that we can legitimately attach to the cgroup */
2072 do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2073 if (ss->can_attach) {
2075 ret = ss->can_attach(tset);
2078 goto out_cancel_attach;
2081 } while_each_subsys_mask();
2084 * Now that we're guaranteed success, proceed to move all tasks to
2085 * the new cgroup. There are no failure cases after here, so this
2086 * is the commit point.
2088 spin_lock_irq(&css_set_lock);
2089 list_for_each_entry(cset, &tset->src_csets, mg_node) {
2090 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2091 struct css_set *from_cset = task_css_set(task);
2092 struct css_set *to_cset = cset->mg_dst_cset;
2094 get_css_set(to_cset);
2095 css_set_move_task(task, from_cset, to_cset, true);
2096 put_css_set_locked(from_cset);
2099 spin_unlock_irq(&css_set_lock);
2102 * Migration is committed, all target tasks are now on dst_csets.
2103 * Nothing is sensitive to fork() after this point. Notify
2104 * controllers that migration is complete.
2106 tset->csets = &tset->dst_csets;
2108 do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2113 } while_each_subsys_mask();
2116 goto out_release_tset;
2119 do_each_subsys_mask(ss, ssid, root->subsys_mask) {
2120 if (ssid == failed_ssid)
2122 if (ss->cancel_attach) {
2124 ss->cancel_attach(tset);
2126 } while_each_subsys_mask();
2128 spin_lock_irq(&css_set_lock);
2129 list_splice_init(&tset->dst_csets, &tset->src_csets);
2130 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2131 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2132 list_del_init(&cset->mg_node);
2134 spin_unlock_irq(&css_set_lock);
2139 * cgroup_may_migrate_to - verify whether a cgroup can be migration destination
2140 * @dst_cgrp: destination cgroup to test
2142 * On the default hierarchy, except for the root, subtree_control must be
2143 * zero for migration destination cgroups with tasks so that child cgroups
2144 * don't compete against tasks.
2146 bool cgroup_may_migrate_to(struct cgroup *dst_cgrp)
2148 return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) ||
2149 !dst_cgrp->subtree_control;
2153 * cgroup_migrate_finish - cleanup after attach
2154 * @preloaded_csets: list of preloaded css_sets
2156 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2157 * those functions for details.
2159 void cgroup_migrate_finish(struct list_head *preloaded_csets)
2161 struct css_set *cset, *tmp_cset;
2163 lockdep_assert_held(&cgroup_mutex);
2165 spin_lock_irq(&css_set_lock);
2166 list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) {
2167 cset->mg_src_cgrp = NULL;
2168 cset->mg_dst_cgrp = NULL;
2169 cset->mg_dst_cset = NULL;
2170 list_del_init(&cset->mg_preload_node);
2171 put_css_set_locked(cset);
2173 spin_unlock_irq(&css_set_lock);
2177 * cgroup_migrate_add_src - add a migration source css_set
2178 * @src_cset: the source css_set to add
2179 * @dst_cgrp: the destination cgroup
2180 * @preloaded_csets: list of preloaded css_sets
2182 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2183 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2184 * up by cgroup_migrate_finish().
2186 * This function may be called without holding cgroup_threadgroup_rwsem
2187 * even if the target is a process. Threads may be created and destroyed
2188 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2189 * into play and the preloaded css_sets are guaranteed to cover all
2192 void cgroup_migrate_add_src(struct css_set *src_cset,
2193 struct cgroup *dst_cgrp,
2194 struct list_head *preloaded_csets)
2196 struct cgroup *src_cgrp;
2198 lockdep_assert_held(&cgroup_mutex);
2199 lockdep_assert_held(&css_set_lock);
2202 * If ->dead, @src_set is associated with one or more dead cgroups
2203 * and doesn't contain any migratable tasks. Ignore it early so
2204 * that the rest of migration path doesn't get confused by it.
2209 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2211 if (!list_empty(&src_cset->mg_preload_node))
2214 WARN_ON(src_cset->mg_src_cgrp);
2215 WARN_ON(src_cset->mg_dst_cgrp);
2216 WARN_ON(!list_empty(&src_cset->mg_tasks));
2217 WARN_ON(!list_empty(&src_cset->mg_node));
2219 src_cset->mg_src_cgrp = src_cgrp;
2220 src_cset->mg_dst_cgrp = dst_cgrp;
2221 get_css_set(src_cset);
2222 list_add(&src_cset->mg_preload_node, preloaded_csets);
2226 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2227 * @preloaded_csets: list of preloaded source css_sets
2229 * Tasks are about to be moved and all the source css_sets have been
2230 * preloaded to @preloaded_csets. This function looks up and pins all
2231 * destination css_sets, links each to its source, and append them to
2234 * This function must be called after cgroup_migrate_add_src() has been
2235 * called on each migration source css_set. After migration is performed
2236 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2239 int cgroup_migrate_prepare_dst(struct list_head *preloaded_csets)
2242 struct css_set *src_cset, *tmp_cset;
2244 lockdep_assert_held(&cgroup_mutex);
2246 /* look up the dst cset for each src cset and link it to src */
2247 list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) {
2248 struct css_set *dst_cset;
2250 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2254 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2257 * If src cset equals dst, it's noop. Drop the src.
2258 * cgroup_migrate() will skip the cset too. Note that we
2259 * can't handle src == dst as some nodes are used by both.
2261 if (src_cset == dst_cset) {
2262 src_cset->mg_src_cgrp = NULL;
2263 src_cset->mg_dst_cgrp = NULL;
2264 list_del_init(&src_cset->mg_preload_node);
2265 put_css_set(src_cset);
2266 put_css_set(dst_cset);
2270 src_cset->mg_dst_cset = dst_cset;
2272 if (list_empty(&dst_cset->mg_preload_node))
2273 list_add(&dst_cset->mg_preload_node, &csets);
2275 put_css_set(dst_cset);
2278 list_splice_tail(&csets, preloaded_csets);
2281 cgroup_migrate_finish(&csets);
2286 * cgroup_migrate - migrate a process or task to a cgroup
2287 * @leader: the leader of the process or the task to migrate
2288 * @threadgroup: whether @leader points to the whole process or a single task
2289 * @root: cgroup root migration is taking place on
2291 * Migrate a process or task denoted by @leader. If migrating a process,
2292 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2293 * responsible for invoking cgroup_migrate_add_src() and
2294 * cgroup_migrate_prepare_dst() on the targets before invoking this
2295 * function and following up with cgroup_migrate_finish().
2297 * As long as a controller's ->can_attach() doesn't fail, this function is
2298 * guaranteed to succeed. This means that, excluding ->can_attach()
2299 * failure, when migrating multiple targets, the success or failure can be
2300 * decided for all targets by invoking group_migrate_prepare_dst() before
2301 * actually starting migrating.
2303 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2304 struct cgroup_root *root)
2306 struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
2307 struct task_struct *task;
2310 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2311 * already PF_EXITING could be freed from underneath us unless we
2312 * take an rcu_read_lock.
2314 spin_lock_irq(&css_set_lock);
2318 cgroup_taskset_add(task, &tset);
2321 } while_each_thread(leader, task);
2323 spin_unlock_irq(&css_set_lock);
2325 return cgroup_taskset_migrate(&tset, root);
2329 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2330 * @dst_cgrp: the cgroup to attach to
2331 * @leader: the task or the leader of the threadgroup to be attached
2332 * @threadgroup: attach the whole threadgroup?
2334 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2336 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2339 LIST_HEAD(preloaded_csets);
2340 struct task_struct *task;
2343 if (!cgroup_may_migrate_to(dst_cgrp))
2346 /* look up all src csets */
2347 spin_lock_irq(&css_set_lock);
2351 cgroup_migrate_add_src(task_css_set(task), dst_cgrp,
2355 } while_each_thread(leader, task);
2357 spin_unlock_irq(&css_set_lock);
2359 /* prepare dst csets and commit */
2360 ret = cgroup_migrate_prepare_dst(&preloaded_csets);
2362 ret = cgroup_migrate(leader, threadgroup, dst_cgrp->root);
2364 cgroup_migrate_finish(&preloaded_csets);
2367 trace_cgroup_attach_task(dst_cgrp, leader, threadgroup);
2372 static int cgroup_procs_write_permission(struct task_struct *task,
2373 struct cgroup *dst_cgrp,
2374 struct kernfs_open_file *of)
2376 const struct cred *cred = current_cred();
2377 const struct cred *tcred = get_task_cred(task);
2381 * even if we're attaching all tasks in the thread group, we only
2382 * need to check permissions on one of them.
2384 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2385 !uid_eq(cred->euid, tcred->uid) &&
2386 !uid_eq(cred->euid, tcred->suid))
2389 if (!ret && cgroup_on_dfl(dst_cgrp)) {
2390 struct super_block *sb = of->file->f_path.dentry->d_sb;
2391 struct cgroup *cgrp;
2392 struct inode *inode;
2394 spin_lock_irq(&css_set_lock);
2395 cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
2396 spin_unlock_irq(&css_set_lock);
2398 while (!cgroup_is_descendant(dst_cgrp, cgrp))
2399 cgrp = cgroup_parent(cgrp);
2402 inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
2404 ret = inode_permission(inode, MAY_WRITE);
2414 * Find the task_struct of the task to attach by vpid and pass it along to the
2415 * function to attach either it or all tasks in its threadgroup. Will lock
2416 * cgroup_mutex and threadgroup.
2418 ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
2419 size_t nbytes, loff_t off, bool threadgroup)
2421 struct task_struct *tsk;
2422 struct cgroup_subsys *ss;
2423 struct cgroup *cgrp;
2427 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2430 cgrp = cgroup_kn_lock_live(of->kn, false);
2434 percpu_down_write(&cgroup_threadgroup_rwsem);
2437 tsk = find_task_by_vpid(pid);
2440 goto out_unlock_rcu;
2447 tsk = tsk->group_leader;
2450 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2451 * trapped in a cpuset, or RT worker may be born in a cgroup
2452 * with no rt_runtime allocated. Just say no.
2454 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
2456 goto out_unlock_rcu;
2459 get_task_struct(tsk);
2462 ret = cgroup_procs_write_permission(tsk, cgrp, of);
2464 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2466 put_task_struct(tsk);
2467 goto out_unlock_threadgroup;
2471 out_unlock_threadgroup:
2472 percpu_up_write(&cgroup_threadgroup_rwsem);
2473 for_each_subsys(ss, ssid)
2474 if (ss->post_attach)
2476 cgroup_kn_unlock(of->kn);
2477 return ret ?: nbytes;
2480 ssize_t cgroup_procs_write(struct kernfs_open_file *of, char *buf, size_t nbytes,
2483 return __cgroup_procs_write(of, buf, nbytes, off, true);
2486 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2488 struct cgroup_subsys *ss;
2489 bool printed = false;
2492 do_each_subsys_mask(ss, ssid, ss_mask) {
2495 seq_printf(seq, "%s", ss->name);
2497 } while_each_subsys_mask();
2499 seq_putc(seq, '\n');
2502 /* show controllers which are enabled from the parent */
2503 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2505 struct cgroup *cgrp = seq_css(seq)->cgroup;
2507 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2511 /* show controllers which are enabled for a given cgroup's children */
2512 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2514 struct cgroup *cgrp = seq_css(seq)->cgroup;
2516 cgroup_print_ss_mask(seq, cgrp->subtree_control);
2521 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2522 * @cgrp: root of the subtree to update csses for
2524 * @cgrp's control masks have changed and its subtree's css associations
2525 * need to be updated accordingly. This function looks up all css_sets
2526 * which are attached to the subtree, creates the matching updated css_sets
2527 * and migrates the tasks to the new ones.
2529 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2531 LIST_HEAD(preloaded_csets);
2532 struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
2533 struct cgroup_subsys_state *d_css;
2534 struct cgroup *dsct;
2535 struct css_set *src_cset;
2538 lockdep_assert_held(&cgroup_mutex);
2540 percpu_down_write(&cgroup_threadgroup_rwsem);
2542 /* look up all csses currently attached to @cgrp's subtree */
2543 spin_lock_irq(&css_set_lock);
2544 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2545 struct cgrp_cset_link *link;
2547 list_for_each_entry(link, &dsct->cset_links, cset_link)
2548 cgroup_migrate_add_src(link->cset, dsct,
2551 spin_unlock_irq(&css_set_lock);
2553 /* NULL dst indicates self on default hierarchy */
2554 ret = cgroup_migrate_prepare_dst(&preloaded_csets);
2558 spin_lock_irq(&css_set_lock);
2559 list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) {
2560 struct task_struct *task, *ntask;
2562 /* src_csets precede dst_csets, break on the first dst_cset */
2563 if (!src_cset->mg_src_cgrp)
2566 /* all tasks in src_csets need to be migrated */
2567 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2568 cgroup_taskset_add(task, &tset);
2570 spin_unlock_irq(&css_set_lock);
2572 ret = cgroup_taskset_migrate(&tset, cgrp->root);
2574 cgroup_migrate_finish(&preloaded_csets);
2575 percpu_up_write(&cgroup_threadgroup_rwsem);
2580 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2581 * @cgrp: root of the target subtree
2583 * Because css offlining is asynchronous, userland may try to re-enable a
2584 * controller while the previous css is still around. This function grabs
2585 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2587 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
2588 __acquires(&cgroup_mutex)
2590 struct cgroup *dsct;
2591 struct cgroup_subsys_state *d_css;
2592 struct cgroup_subsys *ss;
2596 mutex_lock(&cgroup_mutex);
2598 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2599 for_each_subsys(ss, ssid) {
2600 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2603 if (!css || !percpu_ref_is_dying(&css->refcnt))
2607 prepare_to_wait(&dsct->offline_waitq, &wait,
2608 TASK_UNINTERRUPTIBLE);
2610 mutex_unlock(&cgroup_mutex);
2612 finish_wait(&dsct->offline_waitq, &wait);
2621 * cgroup_save_control - save control masks of a subtree
2622 * @cgrp: root of the target subtree
2624 * Save ->subtree_control and ->subtree_ss_mask to the respective old_
2625 * prefixed fields for @cgrp's subtree including @cgrp itself.
2627 static void cgroup_save_control(struct cgroup *cgrp)
2629 struct cgroup *dsct;
2630 struct cgroup_subsys_state *d_css;
2632 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2633 dsct->old_subtree_control = dsct->subtree_control;
2634 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
2639 * cgroup_propagate_control - refresh control masks of a subtree
2640 * @cgrp: root of the target subtree
2642 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
2643 * ->subtree_control and propagate controller availability through the
2644 * subtree so that descendants don't have unavailable controllers enabled.
2646 static void cgroup_propagate_control(struct cgroup *cgrp)
2648 struct cgroup *dsct;
2649 struct cgroup_subsys_state *d_css;
2651 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2652 dsct->subtree_control &= cgroup_control(dsct);
2653 dsct->subtree_ss_mask =
2654 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
2655 cgroup_ss_mask(dsct));
2660 * cgroup_restore_control - restore control masks of a subtree
2661 * @cgrp: root of the target subtree
2663 * Restore ->subtree_control and ->subtree_ss_mask from the respective old_
2664 * prefixed fields for @cgrp's subtree including @cgrp itself.
2666 static void cgroup_restore_control(struct cgroup *cgrp)
2668 struct cgroup *dsct;
2669 struct cgroup_subsys_state *d_css;
2671 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2672 dsct->subtree_control = dsct->old_subtree_control;
2673 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
2677 static bool css_visible(struct cgroup_subsys_state *css)
2679 struct cgroup_subsys *ss = css->ss;
2680 struct cgroup *cgrp = css->cgroup;
2682 if (cgroup_control(cgrp) & (1 << ss->id))
2684 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
2686 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
2690 * cgroup_apply_control_enable - enable or show csses according to control
2691 * @cgrp: root of the target subtree
2693 * Walk @cgrp's subtree and create new csses or make the existing ones
2694 * visible. A css is created invisible if it's being implicitly enabled
2695 * through dependency. An invisible css is made visible when the userland
2696 * explicitly enables it.
2698 * Returns 0 on success, -errno on failure. On failure, csses which have
2699 * been processed already aren't cleaned up. The caller is responsible for
2700 * cleaning up with cgroup_apply_control_disble().
2702 static int cgroup_apply_control_enable(struct cgroup *cgrp)
2704 struct cgroup *dsct;
2705 struct cgroup_subsys_state *d_css;
2706 struct cgroup_subsys *ss;
2709 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2710 for_each_subsys(ss, ssid) {
2711 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2713 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
2715 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
2719 css = css_create(dsct, ss);
2721 return PTR_ERR(css);
2724 if (css_visible(css)) {
2725 ret = css_populate_dir(css);
2736 * cgroup_apply_control_disable - kill or hide csses according to control
2737 * @cgrp: root of the target subtree
2739 * Walk @cgrp's subtree and kill and hide csses so that they match
2740 * cgroup_ss_mask() and cgroup_visible_mask().
2742 * A css is hidden when the userland requests it to be disabled while other
2743 * subsystems are still depending on it. The css must not actively control
2744 * resources and be in the vanilla state if it's made visible again later.
2745 * Controllers which may be depended upon should provide ->css_reset() for
2748 static void cgroup_apply_control_disable(struct cgroup *cgrp)
2750 struct cgroup *dsct;
2751 struct cgroup_subsys_state *d_css;
2752 struct cgroup_subsys *ss;
2755 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2756 for_each_subsys(ss, ssid) {
2757 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2759 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
2765 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
2767 } else if (!css_visible(css)) {
2777 * cgroup_apply_control - apply control mask updates to the subtree
2778 * @cgrp: root of the target subtree
2780 * subsystems can be enabled and disabled in a subtree using the following
2783 * 1. Call cgroup_save_control() to stash the current state.
2784 * 2. Update ->subtree_control masks in the subtree as desired.
2785 * 3. Call cgroup_apply_control() to apply the changes.
2786 * 4. Optionally perform other related operations.
2787 * 5. Call cgroup_finalize_control() to finish up.
2789 * This function implements step 3 and propagates the mask changes
2790 * throughout @cgrp's subtree, updates csses accordingly and perform
2791 * process migrations.
2793 static int cgroup_apply_control(struct cgroup *cgrp)
2797 cgroup_propagate_control(cgrp);
2799 ret = cgroup_apply_control_enable(cgrp);
2804 * At this point, cgroup_e_css() results reflect the new csses
2805 * making the following cgroup_update_dfl_csses() properly update
2806 * css associations of all tasks in the subtree.
2808 ret = cgroup_update_dfl_csses(cgrp);
2816 * cgroup_finalize_control - finalize control mask update
2817 * @cgrp: root of the target subtree
2818 * @ret: the result of the update
2820 * Finalize control mask update. See cgroup_apply_control() for more info.
2822 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
2825 cgroup_restore_control(cgrp);
2826 cgroup_propagate_control(cgrp);
2829 cgroup_apply_control_disable(cgrp);
2832 /* change the enabled child controllers for a cgroup in the default hierarchy */
2833 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
2834 char *buf, size_t nbytes,
2837 u16 enable = 0, disable = 0;
2838 struct cgroup *cgrp, *child;
2839 struct cgroup_subsys *ss;
2844 * Parse input - space separated list of subsystem names prefixed
2845 * with either + or -.
2847 buf = strstrip(buf);
2848 while ((tok = strsep(&buf, " "))) {
2851 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
2852 if (!cgroup_ssid_enabled(ssid) ||
2853 strcmp(tok + 1, ss->name))
2857 enable |= 1 << ssid;
2858 disable &= ~(1 << ssid);
2859 } else if (*tok == '-') {
2860 disable |= 1 << ssid;
2861 enable &= ~(1 << ssid);
2866 } while_each_subsys_mask();
2867 if (ssid == CGROUP_SUBSYS_COUNT)
2871 cgrp = cgroup_kn_lock_live(of->kn, true);
2875 for_each_subsys(ss, ssid) {
2876 if (enable & (1 << ssid)) {
2877 if (cgrp->subtree_control & (1 << ssid)) {
2878 enable &= ~(1 << ssid);
2882 if (!(cgroup_control(cgrp) & (1 << ssid))) {
2886 } else if (disable & (1 << ssid)) {
2887 if (!(cgrp->subtree_control & (1 << ssid))) {
2888 disable &= ~(1 << ssid);
2892 /* a child has it enabled? */
2893 cgroup_for_each_live_child(child, cgrp) {
2894 if (child->subtree_control & (1 << ssid)) {
2902 if (!enable && !disable) {
2908 * Except for the root, subtree_control must be zero for a cgroup
2909 * with tasks so that child cgroups don't compete against tasks.
2911 if (enable && cgroup_parent(cgrp)) {
2912 struct cgrp_cset_link *link;
2915 * Because namespaces pin csets too, @cgrp->cset_links
2916 * might not be empty even when @cgrp is empty. Walk and
2919 spin_lock_irq(&css_set_lock);
2922 list_for_each_entry(link, &cgrp->cset_links, cset_link) {
2923 if (css_set_populated(link->cset)) {
2929 spin_unlock_irq(&css_set_lock);
2935 /* save and update control masks and prepare csses */
2936 cgroup_save_control(cgrp);
2938 cgrp->subtree_control |= enable;
2939 cgrp->subtree_control &= ~disable;
2941 ret = cgroup_apply_control(cgrp);
2943 cgroup_finalize_control(cgrp, ret);
2945 kernfs_activate(cgrp->kn);
2948 cgroup_kn_unlock(of->kn);
2949 return ret ?: nbytes;
2952 static int cgroup_events_show(struct seq_file *seq, void *v)
2954 seq_printf(seq, "populated %d\n",
2955 cgroup_is_populated(seq_css(seq)->cgroup));
2959 static int cgroup_file_open(struct kernfs_open_file *of)
2961 struct cftype *cft = of->kn->priv;
2964 return cft->open(of);
2968 static void cgroup_file_release(struct kernfs_open_file *of)
2970 struct cftype *cft = of->kn->priv;
2976 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
2977 size_t nbytes, loff_t off)
2979 struct cgroup *cgrp = of->kn->parent->priv;
2980 struct cftype *cft = of->kn->priv;
2981 struct cgroup_subsys_state *css;
2985 return cft->write(of, buf, nbytes, off);
2988 * kernfs guarantees that a file isn't deleted with operations in
2989 * flight, which means that the matching css is and stays alive and
2990 * doesn't need to be pinned. The RCU locking is not necessary
2991 * either. It's just for the convenience of using cgroup_css().
2994 css = cgroup_css(cgrp, cft->ss);
2997 if (cft->write_u64) {
2998 unsigned long long v;
2999 ret = kstrtoull(buf, 0, &v);
3001 ret = cft->write_u64(css, cft, v);
3002 } else if (cft->write_s64) {
3004 ret = kstrtoll(buf, 0, &v);
3006 ret = cft->write_s64(css, cft, v);
3011 return ret ?: nbytes;
3014 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3016 return seq_cft(seq)->seq_start(seq, ppos);
3019 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3021 return seq_cft(seq)->seq_next(seq, v, ppos);
3024 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3026 if (seq_cft(seq)->seq_stop)
3027 seq_cft(seq)->seq_stop(seq, v);
3030 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3032 struct cftype *cft = seq_cft(m);
3033 struct cgroup_subsys_state *css = seq_css(m);
3036 return cft->seq_show(m, arg);
3039 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3040 else if (cft->read_s64)
3041 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3047 static struct kernfs_ops cgroup_kf_single_ops = {
3048 .atomic_write_len = PAGE_SIZE,
3049 .open = cgroup_file_open,
3050 .release = cgroup_file_release,
3051 .write = cgroup_file_write,
3052 .seq_show = cgroup_seqfile_show,
3055 static struct kernfs_ops cgroup_kf_ops = {
3056 .atomic_write_len = PAGE_SIZE,
3057 .open = cgroup_file_open,
3058 .release = cgroup_file_release,
3059 .write = cgroup_file_write,
3060 .seq_start = cgroup_seqfile_start,
3061 .seq_next = cgroup_seqfile_next,
3062 .seq_stop = cgroup_seqfile_stop,
3063 .seq_show = cgroup_seqfile_show,
3066 /* set uid and gid of cgroup dirs and files to that of the creator */
3067 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3069 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3070 .ia_uid = current_fsuid(),
3071 .ia_gid = current_fsgid(), };
3073 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3074 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3077 return kernfs_setattr(kn, &iattr);
3080 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3083 char name[CGROUP_FILE_NAME_MAX];
3084 struct kernfs_node *kn;
3085 struct lock_class_key *key = NULL;
3088 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3089 key = &cft->lockdep_key;
3091 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3092 cgroup_file_mode(cft), 0, cft->kf_ops, cft,
3097 ret = cgroup_kn_set_ugid(kn);
3103 if (cft->file_offset) {
3104 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3106 spin_lock_irq(&cgroup_file_kn_lock);
3108 spin_unlock_irq(&cgroup_file_kn_lock);
3115 * cgroup_addrm_files - add or remove files to a cgroup directory
3116 * @css: the target css
3117 * @cgrp: the target cgroup (usually css->cgroup)
3118 * @cfts: array of cftypes to be added
3119 * @is_add: whether to add or remove
3121 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3122 * For removals, this function never fails.
3124 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3125 struct cgroup *cgrp, struct cftype cfts[],
3128 struct cftype *cft, *cft_end = NULL;
3131 lockdep_assert_held(&cgroup_mutex);
3134 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3135 /* does cft->flags tell us to skip this file on @cgrp? */
3136 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3138 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3140 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3142 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3146 ret = cgroup_add_file(css, cgrp, cft);
3148 pr_warn("%s: failed to add %s, err=%d\n",
3149 __func__, cft->name, ret);
3155 cgroup_rm_file(cgrp, cft);
3161 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3164 struct cgroup_subsys *ss = cfts[0].ss;
3165 struct cgroup *root = &ss->root->cgrp;
3166 struct cgroup_subsys_state *css;
3169 lockdep_assert_held(&cgroup_mutex);
3171 /* add/rm files for all cgroups created before */
3172 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3173 struct cgroup *cgrp = css->cgroup;
3175 if (!(css->flags & CSS_VISIBLE))
3178 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3184 kernfs_activate(root->kn);
3188 static void cgroup_exit_cftypes(struct cftype *cfts)
3192 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3193 /* free copy for custom atomic_write_len, see init_cftypes() */
3194 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3199 /* revert flags set by cgroup core while adding @cfts */
3200 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
3204 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3208 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3209 struct kernfs_ops *kf_ops;
3211 WARN_ON(cft->ss || cft->kf_ops);
3214 kf_ops = &cgroup_kf_ops;
3216 kf_ops = &cgroup_kf_single_ops;
3219 * Ugh... if @cft wants a custom max_write_len, we need to
3220 * make a copy of kf_ops to set its atomic_write_len.
3222 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3223 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3225 cgroup_exit_cftypes(cfts);
3228 kf_ops->atomic_write_len = cft->max_write_len;
3231 cft->kf_ops = kf_ops;
3238 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3240 lockdep_assert_held(&cgroup_mutex);
3242 if (!cfts || !cfts[0].ss)
3245 list_del(&cfts->node);
3246 cgroup_apply_cftypes(cfts, false);
3247 cgroup_exit_cftypes(cfts);
3252 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3253 * @cfts: zero-length name terminated array of cftypes
3255 * Unregister @cfts. Files described by @cfts are removed from all
3256 * existing cgroups and all future cgroups won't have them either. This
3257 * function can be called anytime whether @cfts' subsys is attached or not.
3259 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3262 int cgroup_rm_cftypes(struct cftype *cfts)
3266 mutex_lock(&cgroup_mutex);
3267 ret = cgroup_rm_cftypes_locked(cfts);
3268 mutex_unlock(&cgroup_mutex);
3273 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3274 * @ss: target cgroup subsystem
3275 * @cfts: zero-length name terminated array of cftypes
3277 * Register @cfts to @ss. Files described by @cfts are created for all
3278 * existing cgroups to which @ss is attached and all future cgroups will
3279 * have them too. This function can be called anytime whether @ss is
3282 * Returns 0 on successful registration, -errno on failure. Note that this
3283 * function currently returns 0 as long as @cfts registration is successful
3284 * even if some file creation attempts on existing cgroups fail.
3286 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3290 if (!cgroup_ssid_enabled(ss->id))
3293 if (!cfts || cfts[0].name[0] == '\0')
3296 ret = cgroup_init_cftypes(ss, cfts);
3300 mutex_lock(&cgroup_mutex);
3302 list_add_tail(&cfts->node, &ss->cfts);
3303 ret = cgroup_apply_cftypes(cfts, true);
3305 cgroup_rm_cftypes_locked(cfts);
3307 mutex_unlock(&cgroup_mutex);
3312 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3313 * @ss: target cgroup subsystem
3314 * @cfts: zero-length name terminated array of cftypes
3316 * Similar to cgroup_add_cftypes() but the added files are only used for
3317 * the default hierarchy.
3319 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3323 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3324 cft->flags |= __CFTYPE_ONLY_ON_DFL;
3325 return cgroup_add_cftypes(ss, cfts);
3329 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3330 * @ss: target cgroup subsystem
3331 * @cfts: zero-length name terminated array of cftypes
3333 * Similar to cgroup_add_cftypes() but the added files are only used for
3334 * the legacy hierarchies.
3336 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3340 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3341 cft->flags |= __CFTYPE_NOT_ON_DFL;
3342 return cgroup_add_cftypes(ss, cfts);
3346 * cgroup_file_notify - generate a file modified event for a cgroup_file
3347 * @cfile: target cgroup_file
3349 * @cfile must have been obtained by setting cftype->file_offset.
3351 void cgroup_file_notify(struct cgroup_file *cfile)
3353 unsigned long flags;
3355 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
3357 kernfs_notify(cfile->kn);
3358 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
3362 * css_next_child - find the next child of a given css
3363 * @pos: the current position (%NULL to initiate traversal)
3364 * @parent: css whose children to walk
3366 * This function returns the next child of @parent and should be called
3367 * under either cgroup_mutex or RCU read lock. The only requirement is
3368 * that @parent and @pos are accessible. The next sibling is guaranteed to
3369 * be returned regardless of their states.
3371 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3372 * css which finished ->css_online() is guaranteed to be visible in the
3373 * future iterations and will stay visible until the last reference is put.
3374 * A css which hasn't finished ->css_online() or already finished
3375 * ->css_offline() may show up during traversal. It's each subsystem's
3376 * responsibility to synchronize against on/offlining.
3378 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
3379 struct cgroup_subsys_state *parent)
3381 struct cgroup_subsys_state *next;
3383 cgroup_assert_mutex_or_rcu_locked();
3386 * @pos could already have been unlinked from the sibling list.
3387 * Once a cgroup is removed, its ->sibling.next is no longer
3388 * updated when its next sibling changes. CSS_RELEASED is set when
3389 * @pos is taken off list, at which time its next pointer is valid,
3390 * and, as releases are serialized, the one pointed to by the next
3391 * pointer is guaranteed to not have started release yet. This
3392 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3393 * critical section, the one pointed to by its next pointer is
3394 * guaranteed to not have finished its RCU grace period even if we
3395 * have dropped rcu_read_lock() inbetween iterations.
3397 * If @pos has CSS_RELEASED set, its next pointer can't be
3398 * dereferenced; however, as each css is given a monotonically
3399 * increasing unique serial number and always appended to the
3400 * sibling list, the next one can be found by walking the parent's
3401 * children until the first css with higher serial number than
3402 * @pos's. While this path can be slower, it happens iff iteration
3403 * races against release and the race window is very small.
3406 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
3407 } else if (likely(!(pos->flags & CSS_RELEASED))) {
3408 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
3410 list_for_each_entry_rcu(next, &parent->children, sibling)
3411 if (next->serial_nr > pos->serial_nr)
3416 * @next, if not pointing to the head, can be dereferenced and is
3419 if (&next->sibling != &parent->children)
3425 * css_next_descendant_pre - find the next descendant for pre-order walk
3426 * @pos: the current position (%NULL to initiate traversal)
3427 * @root: css whose descendants to walk
3429 * To be used by css_for_each_descendant_pre(). Find the next descendant
3430 * to visit for pre-order traversal of @root's descendants. @root is
3431 * included in the iteration and the first node to be visited.
3433 * While this function requires cgroup_mutex or RCU read locking, it
3434 * doesn't require the whole traversal to be contained in a single critical
3435 * section. This function will return the correct next descendant as long
3436 * as both @pos and @root are accessible and @pos is a descendant of @root.
3438 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3439 * css which finished ->css_online() is guaranteed to be visible in the
3440 * future iterations and will stay visible until the last reference is put.
3441 * A css which hasn't finished ->css_online() or already finished
3442 * ->css_offline() may show up during traversal. It's each subsystem's
3443 * responsibility to synchronize against on/offlining.
3445 struct cgroup_subsys_state *
3446 css_next_descendant_pre(struct cgroup_subsys_state *pos,
3447 struct cgroup_subsys_state *root)
3449 struct cgroup_subsys_state *next;
3451 cgroup_assert_mutex_or_rcu_locked();
3453 /* if first iteration, visit @root */
3457 /* visit the first child if exists */
3458 next = css_next_child(NULL, pos);
3462 /* no child, visit my or the closest ancestor's next sibling */
3463 while (pos != root) {
3464 next = css_next_child(pos, pos->parent);
3474 * css_rightmost_descendant - return the rightmost descendant of a css
3475 * @pos: css of interest
3477 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3478 * is returned. This can be used during pre-order traversal to skip
3481 * While this function requires cgroup_mutex or RCU read locking, it
3482 * doesn't require the whole traversal to be contained in a single critical
3483 * section. This function will return the correct rightmost descendant as
3484 * long as @pos is accessible.
3486 struct cgroup_subsys_state *
3487 css_rightmost_descendant(struct cgroup_subsys_state *pos)
3489 struct cgroup_subsys_state *last, *tmp;
3491 cgroup_assert_mutex_or_rcu_locked();
3495 /* ->prev isn't RCU safe, walk ->next till the end */
3497 css_for_each_child(tmp, last)
3504 static struct cgroup_subsys_state *
3505 css_leftmost_descendant(struct cgroup_subsys_state *pos)
3507 struct cgroup_subsys_state *last;
3511 pos = css_next_child(NULL, pos);
3518 * css_next_descendant_post - find the next descendant for post-order walk
3519 * @pos: the current position (%NULL to initiate traversal)
3520 * @root: css whose descendants to walk
3522 * To be used by css_for_each_descendant_post(). Find the next descendant
3523 * to visit for post-order traversal of @root's descendants. @root is
3524 * included in the iteration and the last node to be visited.
3526 * While this function requires cgroup_mutex or RCU read locking, it
3527 * doesn't require the whole traversal to be contained in a single critical
3528 * section. This function will return the correct next descendant as long
3529 * as both @pos and @cgroup are accessible and @pos is a descendant of
3532 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3533 * css which finished ->css_online() is guaranteed to be visible in the
3534 * future iterations and will stay visible until the last reference is put.
3535 * A css which hasn't finished ->css_online() or already finished
3536 * ->css_offline() may show up during traversal. It's each subsystem's
3537 * responsibility to synchronize against on/offlining.
3539 struct cgroup_subsys_state *
3540 css_next_descendant_post(struct cgroup_subsys_state *pos,
3541 struct cgroup_subsys_state *root)
3543 struct cgroup_subsys_state *next;
3545 cgroup_assert_mutex_or_rcu_locked();
3547 /* if first iteration, visit leftmost descendant which may be @root */
3549 return css_leftmost_descendant(root);
3551 /* if we visited @root, we're done */
3555 /* if there's an unvisited sibling, visit its leftmost descendant */
3556 next = css_next_child(pos, pos->parent);
3558 return css_leftmost_descendant(next);
3560 /* no sibling left, visit parent */
3565 * css_has_online_children - does a css have online children
3566 * @css: the target css
3568 * Returns %true if @css has any online children; otherwise, %false. This
3569 * function can be called from any context but the caller is responsible
3570 * for synchronizing against on/offlining as necessary.
3572 bool css_has_online_children(struct cgroup_subsys_state *css)
3574 struct cgroup_subsys_state *child;
3578 css_for_each_child(child, css) {
3579 if (child->flags & CSS_ONLINE) {
3589 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
3590 * @it: the iterator to advance
3592 * Advance @it to the next css_set to walk.
3594 static void css_task_iter_advance_css_set(struct css_task_iter *it)
3596 struct list_head *l = it->cset_pos;
3597 struct cgrp_cset_link *link;
3598 struct css_set *cset;
3600 lockdep_assert_held(&css_set_lock);
3602 /* Advance to the next non-empty css_set */
3605 if (l == it->cset_head) {
3606 it->cset_pos = NULL;
3607 it->task_pos = NULL;
3612 cset = container_of(l, struct css_set,
3613 e_cset_node[it->ss->id]);
3615 link = list_entry(l, struct cgrp_cset_link, cset_link);
3618 } while (!css_set_populated(cset));
3622 if (!list_empty(&cset->tasks))
3623 it->task_pos = cset->tasks.next;
3625 it->task_pos = cset->mg_tasks.next;
3627 it->tasks_head = &cset->tasks;
3628 it->mg_tasks_head = &cset->mg_tasks;
3631 * We don't keep css_sets locked across iteration steps and thus
3632 * need to take steps to ensure that iteration can be resumed after
3633 * the lock is re-acquired. Iteration is performed at two levels -
3634 * css_sets and tasks in them.
3636 * Once created, a css_set never leaves its cgroup lists, so a
3637 * pinned css_set is guaranteed to stay put and we can resume
3638 * iteration afterwards.
3640 * Tasks may leave @cset across iteration steps. This is resolved
3641 * by registering each iterator with the css_set currently being
3642 * walked and making css_set_move_task() advance iterators whose
3643 * next task is leaving.
3646 list_del(&it->iters_node);
3647 put_css_set_locked(it->cur_cset);
3650 it->cur_cset = cset;
3651 list_add(&it->iters_node, &cset->task_iters);
3654 static void css_task_iter_advance(struct css_task_iter *it)
3656 struct list_head *l = it->task_pos;
3658 lockdep_assert_held(&css_set_lock);
3662 * Advance iterator to find next entry. cset->tasks is consumed
3663 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
3668 if (l == it->tasks_head)
3669 l = it->mg_tasks_head->next;
3671 if (l == it->mg_tasks_head)
3672 css_task_iter_advance_css_set(it);
3678 * css_task_iter_start - initiate task iteration
3679 * @css: the css to walk tasks of
3680 * @it: the task iterator to use
3682 * Initiate iteration through the tasks of @css. The caller can call
3683 * css_task_iter_next() to walk through the tasks until the function
3684 * returns NULL. On completion of iteration, css_task_iter_end() must be
3687 void css_task_iter_start(struct cgroup_subsys_state *css,
3688 struct css_task_iter *it)
3690 /* no one should try to iterate before mounting cgroups */
3691 WARN_ON_ONCE(!use_task_css_set_links);
3693 memset(it, 0, sizeof(*it));
3695 spin_lock_irq(&css_set_lock);
3700 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
3702 it->cset_pos = &css->cgroup->cset_links;
3704 it->cset_head = it->cset_pos;
3706 css_task_iter_advance_css_set(it);
3708 spin_unlock_irq(&css_set_lock);
3712 * css_task_iter_next - return the next task for the iterator
3713 * @it: the task iterator being iterated
3715 * The "next" function for task iteration. @it should have been
3716 * initialized via css_task_iter_start(). Returns NULL when the iteration
3719 struct task_struct *css_task_iter_next(struct css_task_iter *it)
3722 put_task_struct(it->cur_task);
3723 it->cur_task = NULL;
3726 spin_lock_irq(&css_set_lock);
3729 it->cur_task = list_entry(it->task_pos, struct task_struct,
3731 get_task_struct(it->cur_task);
3732 css_task_iter_advance(it);
3735 spin_unlock_irq(&css_set_lock);
3737 return it->cur_task;
3741 * css_task_iter_end - finish task iteration
3742 * @it: the task iterator to finish
3744 * Finish task iteration started by css_task_iter_start().
3746 void css_task_iter_end(struct css_task_iter *it)
3749 spin_lock_irq(&css_set_lock);
3750 list_del(&it->iters_node);
3751 put_css_set_locked(it->cur_cset);
3752 spin_unlock_irq(&css_set_lock);
3756 put_task_struct(it->cur_task);
3759 static void cgroup_procs_release(struct kernfs_open_file *of)
3762 css_task_iter_end(of->priv);
3767 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
3769 struct kernfs_open_file *of = s->private;
3770 struct css_task_iter *it = of->priv;
3771 struct task_struct *task;
3774 task = css_task_iter_next(it);
3775 } while (task && !thread_group_leader(task));
3780 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
3782 struct kernfs_open_file *of = s->private;
3783 struct cgroup *cgrp = seq_css(s)->cgroup;
3784 struct css_task_iter *it = of->priv;
3787 * When a seq_file is seeked, it's always traversed sequentially
3788 * from position 0, so we can simply keep iterating on !0 *pos.
3791 if (WARN_ON_ONCE((*pos)++))
3792 return ERR_PTR(-EINVAL);
3794 it = kzalloc(sizeof(*it), GFP_KERNEL);
3796 return ERR_PTR(-ENOMEM);
3798 css_task_iter_start(&cgrp->self, it);
3799 } else if (!(*pos)++) {
3800 css_task_iter_end(it);
3801 css_task_iter_start(&cgrp->self, it);
3804 return cgroup_procs_next(s, NULL, NULL);
3807 static int cgroup_procs_show(struct seq_file *s, void *v)
3809 seq_printf(s, "%d\n", task_tgid_vnr(v));
3813 /* cgroup core interface files for the default hierarchy */
3814 static struct cftype cgroup_base_files[] = {
3816 .name = "cgroup.procs",
3817 .file_offset = offsetof(struct cgroup, procs_file),
3818 .release = cgroup_procs_release,
3819 .seq_start = cgroup_procs_start,
3820 .seq_next = cgroup_procs_next,
3821 .seq_show = cgroup_procs_show,
3822 .write = cgroup_procs_write,
3825 .name = "cgroup.controllers",
3826 .seq_show = cgroup_controllers_show,
3829 .name = "cgroup.subtree_control",
3830 .seq_show = cgroup_subtree_control_show,
3831 .write = cgroup_subtree_control_write,
3834 .name = "cgroup.events",
3835 .flags = CFTYPE_NOT_ON_ROOT,
3836 .file_offset = offsetof(struct cgroup, events_file),
3837 .seq_show = cgroup_events_show,
3843 * css destruction is four-stage process.
3845 * 1. Destruction starts. Killing of the percpu_ref is initiated.
3846 * Implemented in kill_css().
3848 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
3849 * and thus css_tryget_online() is guaranteed to fail, the css can be
3850 * offlined by invoking offline_css(). After offlining, the base ref is
3851 * put. Implemented in css_killed_work_fn().
3853 * 3. When the percpu_ref reaches zero, the only possible remaining
3854 * accessors are inside RCU read sections. css_release() schedules the
3857 * 4. After the grace period, the css can be freed. Implemented in
3858 * css_free_work_fn().
3860 * It is actually hairier because both step 2 and 4 require process context
3861 * and thus involve punting to css->destroy_work adding two additional
3862 * steps to the already complex sequence.
3864 static void css_free_work_fn(struct work_struct *work)
3866 struct cgroup_subsys_state *css =
3867 container_of(work, struct cgroup_subsys_state, destroy_work);
3868 struct cgroup_subsys *ss = css->ss;
3869 struct cgroup *cgrp = css->cgroup;
3871 percpu_ref_exit(&css->refcnt);
3875 struct cgroup_subsys_state *parent = css->parent;
3879 cgroup_idr_remove(&ss->css_idr, id);
3885 /* cgroup free path */
3886 atomic_dec(&cgrp->root->nr_cgrps);
3887 cgroup1_pidlist_destroy_all(cgrp);
3888 cancel_work_sync(&cgrp->release_agent_work);
3890 if (cgroup_parent(cgrp)) {
3892 * We get a ref to the parent, and put the ref when
3893 * this cgroup is being freed, so it's guaranteed
3894 * that the parent won't be destroyed before its
3897 cgroup_put(cgroup_parent(cgrp));
3898 kernfs_put(cgrp->kn);
3902 * This is root cgroup's refcnt reaching zero,
3903 * which indicates that the root should be
3906 cgroup_destroy_root(cgrp->root);
3911 static void css_free_rcu_fn(struct rcu_head *rcu_head)
3913 struct cgroup_subsys_state *css =
3914 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
3916 INIT_WORK(&css->destroy_work, css_free_work_fn);
3917 queue_work(cgroup_destroy_wq, &css->destroy_work);
3920 static void css_release_work_fn(struct work_struct *work)
3922 struct cgroup_subsys_state *css =
3923 container_of(work, struct cgroup_subsys_state, destroy_work);
3924 struct cgroup_subsys *ss = css->ss;
3925 struct cgroup *cgrp = css->cgroup;
3927 mutex_lock(&cgroup_mutex);
3929 css->flags |= CSS_RELEASED;
3930 list_del_rcu(&css->sibling);
3933 /* css release path */
3934 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
3935 if (ss->css_released)
3936 ss->css_released(css);
3938 /* cgroup release path */
3939 trace_cgroup_release(cgrp);
3941 cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
3945 * There are two control paths which try to determine
3946 * cgroup from dentry without going through kernfs -
3947 * cgroupstats_build() and css_tryget_online_from_dir().
3948 * Those are supported by RCU protecting clearing of
3949 * cgrp->kn->priv backpointer.
3952 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
3955 cgroup_bpf_put(cgrp);
3958 mutex_unlock(&cgroup_mutex);
3960 call_rcu(&css->rcu_head, css_free_rcu_fn);
3963 static void css_release(struct percpu_ref *ref)
3965 struct cgroup_subsys_state *css =
3966 container_of(ref, struct cgroup_subsys_state, refcnt);
3968 INIT_WORK(&css->destroy_work, css_release_work_fn);
3969 queue_work(cgroup_destroy_wq, &css->destroy_work);
3972 static void init_and_link_css(struct cgroup_subsys_state *css,
3973 struct cgroup_subsys *ss, struct cgroup *cgrp)
3975 lockdep_assert_held(&cgroup_mutex);
3979 memset(css, 0, sizeof(*css));
3983 INIT_LIST_HEAD(&css->sibling);
3984 INIT_LIST_HEAD(&css->children);
3985 css->serial_nr = css_serial_nr_next++;
3986 atomic_set(&css->online_cnt, 0);
3988 if (cgroup_parent(cgrp)) {
3989 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
3990 css_get(css->parent);
3993 BUG_ON(cgroup_css(cgrp, ss));
3996 /* invoke ->css_online() on a new CSS and mark it online if successful */
3997 static int online_css(struct cgroup_subsys_state *css)
3999 struct cgroup_subsys *ss = css->ss;
4002 lockdep_assert_held(&cgroup_mutex);
4005 ret = ss->css_online(css);
4007 css->flags |= CSS_ONLINE;
4008 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
4010 atomic_inc(&css->online_cnt);
4012 atomic_inc(&css->parent->online_cnt);
4017 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4018 static void offline_css(struct cgroup_subsys_state *css)
4020 struct cgroup_subsys *ss = css->ss;
4022 lockdep_assert_held(&cgroup_mutex);
4024 if (!(css->flags & CSS_ONLINE))
4030 if (ss->css_offline)
4031 ss->css_offline(css);
4033 css->flags &= ~CSS_ONLINE;
4034 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
4036 wake_up_all(&css->cgroup->offline_waitq);
4040 * css_create - create a cgroup_subsys_state
4041 * @cgrp: the cgroup new css will be associated with
4042 * @ss: the subsys of new css
4044 * Create a new css associated with @cgrp - @ss pair. On success, the new
4045 * css is online and installed in @cgrp. This function doesn't create the
4046 * interface files. Returns 0 on success, -errno on failure.
4048 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
4049 struct cgroup_subsys *ss)
4051 struct cgroup *parent = cgroup_parent(cgrp);
4052 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
4053 struct cgroup_subsys_state *css;
4056 lockdep_assert_held(&cgroup_mutex);
4058 css = ss->css_alloc(parent_css);
4060 css = ERR_PTR(-ENOMEM);
4064 init_and_link_css(css, ss, cgrp);
4066 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
4070 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
4075 /* @css is ready to be brought online now, make it visible */
4076 list_add_tail_rcu(&css->sibling, &parent_css->children);
4077 cgroup_idr_replace(&ss->css_idr, css, css->id);
4079 err = online_css(css);
4083 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4084 cgroup_parent(parent)) {
4085 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4086 current->comm, current->pid, ss->name);
4087 if (!strcmp(ss->name, "memory"))
4088 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4089 ss->warned_broken_hierarchy = true;
4095 list_del_rcu(&css->sibling);
4097 call_rcu(&css->rcu_head, css_free_rcu_fn);
4098 return ERR_PTR(err);
4101 static struct cgroup *cgroup_create(struct cgroup *parent)
4103 struct cgroup_root *root = parent->root;
4104 struct cgroup *cgrp, *tcgrp;
4105 int level = parent->level + 1;
4108 /* allocate the cgroup and its ID, 0 is reserved for the root */
4109 cgrp = kzalloc(sizeof(*cgrp) +
4110 sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL);
4112 return ERR_PTR(-ENOMEM);
4114 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
4119 * Temporarily set the pointer to NULL, so idr_find() won't return
4120 * a half-baked cgroup.
4122 cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL);
4125 goto out_cancel_ref;
4128 init_cgroup_housekeeping(cgrp);
4130 cgrp->self.parent = &parent->self;
4132 cgrp->level = level;
4134 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp))
4135 cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
4137 if (notify_on_release(parent))
4138 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4140 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4141 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4143 cgrp->self.serial_nr = css_serial_nr_next++;
4145 /* allocation complete, commit to creation */
4146 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
4147 atomic_inc(&root->nr_cgrps);
4151 * @cgrp is now fully operational. If something fails after this
4152 * point, it'll be released via the normal destruction path.
4154 cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
4157 * On the default hierarchy, a child doesn't automatically inherit
4158 * subtree_control from the parent. Each is configured manually.
4160 if (!cgroup_on_dfl(cgrp))
4161 cgrp->subtree_control = cgroup_control(cgrp);
4164 cgroup_bpf_inherit(cgrp, parent);
4166 cgroup_propagate_control(cgrp);
4168 /* @cgrp doesn't have dir yet so the following will only create csses */
4169 ret = cgroup_apply_control_enable(cgrp);
4176 percpu_ref_exit(&cgrp->self.refcnt);
4179 return ERR_PTR(ret);
4181 cgroup_destroy_locked(cgrp);
4182 return ERR_PTR(ret);
4185 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
4187 struct cgroup *parent, *cgrp;
4188 struct kernfs_node *kn;
4191 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
4192 if (strchr(name, '\n'))
4195 parent = cgroup_kn_lock_live(parent_kn, false);
4199 cgrp = cgroup_create(parent);
4201 ret = PTR_ERR(cgrp);
4205 /* create the directory */
4206 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
4214 * This extra ref will be put in cgroup_free_fn() and guarantees
4215 * that @cgrp->kn is always accessible.
4219 ret = cgroup_kn_set_ugid(kn);
4223 ret = css_populate_dir(&cgrp->self);
4227 ret = cgroup_apply_control_enable(cgrp);
4231 trace_cgroup_mkdir(cgrp);
4233 /* let's create and online css's */
4234 kernfs_activate(kn);
4240 cgroup_destroy_locked(cgrp);
4242 cgroup_kn_unlock(parent_kn);
4247 * This is called when the refcnt of a css is confirmed to be killed.
4248 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4249 * initate destruction and put the css ref from kill_css().
4251 static void css_killed_work_fn(struct work_struct *work)
4253 struct cgroup_subsys_state *css =
4254 container_of(work, struct cgroup_subsys_state, destroy_work);
4256 mutex_lock(&cgroup_mutex);
4261 /* @css can't go away while we're holding cgroup_mutex */
4263 } while (css && atomic_dec_and_test(&css->online_cnt));
4265 mutex_unlock(&cgroup_mutex);
4268 /* css kill confirmation processing requires process context, bounce */
4269 static void css_killed_ref_fn(struct percpu_ref *ref)
4271 struct cgroup_subsys_state *css =
4272 container_of(ref, struct cgroup_subsys_state, refcnt);
4274 if (atomic_dec_and_test(&css->online_cnt)) {
4275 INIT_WORK(&css->destroy_work, css_killed_work_fn);
4276 queue_work(cgroup_destroy_wq, &css->destroy_work);
4281 * kill_css - destroy a css
4282 * @css: css to destroy
4284 * This function initiates destruction of @css by removing cgroup interface
4285 * files and putting its base reference. ->css_offline() will be invoked
4286 * asynchronously once css_tryget_online() is guaranteed to fail and when
4287 * the reference count reaches zero, @css will be released.
4289 static void kill_css(struct cgroup_subsys_state *css)
4291 lockdep_assert_held(&cgroup_mutex);
4294 * This must happen before css is disassociated with its cgroup.
4295 * See seq_css() for details.
4300 * Killing would put the base ref, but we need to keep it alive
4301 * until after ->css_offline().
4306 * cgroup core guarantees that, by the time ->css_offline() is
4307 * invoked, no new css reference will be given out via
4308 * css_tryget_online(). We can't simply call percpu_ref_kill() and
4309 * proceed to offlining css's because percpu_ref_kill() doesn't
4310 * guarantee that the ref is seen as killed on all CPUs on return.
4312 * Use percpu_ref_kill_and_confirm() to get notifications as each
4313 * css is confirmed to be seen as killed on all CPUs.
4315 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
4319 * cgroup_destroy_locked - the first stage of cgroup destruction
4320 * @cgrp: cgroup to be destroyed
4322 * css's make use of percpu refcnts whose killing latency shouldn't be
4323 * exposed to userland and are RCU protected. Also, cgroup core needs to
4324 * guarantee that css_tryget_online() won't succeed by the time
4325 * ->css_offline() is invoked. To satisfy all the requirements,
4326 * destruction is implemented in the following two steps.
4328 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4329 * userland visible parts and start killing the percpu refcnts of
4330 * css's. Set up so that the next stage will be kicked off once all
4331 * the percpu refcnts are confirmed to be killed.
4333 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4334 * rest of destruction. Once all cgroup references are gone, the
4335 * cgroup is RCU-freed.
4337 * This function implements s1. After this step, @cgrp is gone as far as
4338 * the userland is concerned and a new cgroup with the same name may be
4339 * created. As cgroup doesn't care about the names internally, this
4340 * doesn't cause any problem.
4342 static int cgroup_destroy_locked(struct cgroup *cgrp)
4343 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4345 struct cgroup_subsys_state *css;
4346 struct cgrp_cset_link *link;
4349 lockdep_assert_held(&cgroup_mutex);
4352 * Only migration can raise populated from zero and we're already
4353 * holding cgroup_mutex.
4355 if (cgroup_is_populated(cgrp))
4359 * Make sure there's no live children. We can't test emptiness of
4360 * ->self.children as dead children linger on it while being
4361 * drained; otherwise, "rmdir parent/child parent" may fail.
4363 if (css_has_online_children(&cgrp->self))
4367 * Mark @cgrp and the associated csets dead. The former prevents
4368 * further task migration and child creation by disabling
4369 * cgroup_lock_live_group(). The latter makes the csets ignored by
4370 * the migration path.
4372 cgrp->self.flags &= ~CSS_ONLINE;
4374 spin_lock_irq(&css_set_lock);
4375 list_for_each_entry(link, &cgrp->cset_links, cset_link)
4376 link->cset->dead = true;
4377 spin_unlock_irq(&css_set_lock);
4379 /* initiate massacre of all css's */
4380 for_each_css(css, ssid, cgrp)
4384 * Remove @cgrp directory along with the base files. @cgrp has an
4385 * extra ref on its kn.
4387 kernfs_remove(cgrp->kn);
4389 cgroup1_check_for_release(cgroup_parent(cgrp));
4391 /* put the base reference */
4392 percpu_ref_kill(&cgrp->self.refcnt);
4397 int cgroup_rmdir(struct kernfs_node *kn)
4399 struct cgroup *cgrp;
4402 cgrp = cgroup_kn_lock_live(kn, false);
4406 ret = cgroup_destroy_locked(cgrp);
4409 trace_cgroup_rmdir(cgrp);
4411 cgroup_kn_unlock(kn);
4415 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
4416 .remount_fs = cgroup_remount,
4417 .mkdir = cgroup_mkdir,
4418 .rmdir = cgroup_rmdir,
4419 .show_path = cgroup_show_path,
4422 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
4424 struct cgroup_subsys_state *css;
4426 pr_debug("Initializing cgroup subsys %s\n", ss->name);
4428 mutex_lock(&cgroup_mutex);
4430 idr_init(&ss->css_idr);
4431 INIT_LIST_HEAD(&ss->cfts);
4433 /* Create the root cgroup state for this subsystem */
4434 ss->root = &cgrp_dfl_root;
4435 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
4436 /* We don't handle early failures gracefully */
4437 BUG_ON(IS_ERR(css));
4438 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
4441 * Root csses are never destroyed and we can't initialize
4442 * percpu_ref during early init. Disable refcnting.
4444 css->flags |= CSS_NO_REF;
4447 /* allocation can't be done safely during early init */
4450 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
4451 BUG_ON(css->id < 0);
4454 /* Update the init_css_set to contain a subsys
4455 * pointer to this state - since the subsystem is
4456 * newly registered, all tasks and hence the
4457 * init_css_set is in the subsystem's root cgroup. */
4458 init_css_set.subsys[ss->id] = css;
4460 have_fork_callback |= (bool)ss->fork << ss->id;
4461 have_exit_callback |= (bool)ss->exit << ss->id;
4462 have_free_callback |= (bool)ss->free << ss->id;
4463 have_canfork_callback |= (bool)ss->can_fork << ss->id;
4465 /* At system boot, before all subsystems have been
4466 * registered, no tasks have been forked, so we don't
4467 * need to invoke fork callbacks here. */
4468 BUG_ON(!list_empty(&init_task.tasks));
4470 BUG_ON(online_css(css));
4472 mutex_unlock(&cgroup_mutex);
4476 * cgroup_init_early - cgroup initialization at system boot
4478 * Initialize cgroups at system boot, and initialize any
4479 * subsystems that request early init.
4481 int __init cgroup_init_early(void)
4483 static struct cgroup_sb_opts __initdata opts;
4484 struct cgroup_subsys *ss;
4487 init_cgroup_root(&cgrp_dfl_root, &opts);
4488 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
4490 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
4492 for_each_subsys(ss, i) {
4493 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
4494 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
4495 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
4497 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
4498 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
4501 ss->name = cgroup_subsys_name[i];
4502 if (!ss->legacy_name)
4503 ss->legacy_name = cgroup_subsys_name[i];
4506 cgroup_init_subsys(ss, true);
4511 static u16 cgroup_disable_mask __initdata;
4514 * cgroup_init - cgroup initialization
4516 * Register cgroup filesystem and /proc file, and initialize
4517 * any subsystems that didn't request early init.
4519 int __init cgroup_init(void)
4521 struct cgroup_subsys *ss;
4524 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
4525 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
4526 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
4527 BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
4530 * The latency of the synchronize_sched() is too high for cgroups,
4531 * avoid it at the cost of forcing all readers into the slow path.
4533 rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
4535 get_user_ns(init_cgroup_ns.user_ns);
4537 mutex_lock(&cgroup_mutex);
4540 * Add init_css_set to the hash table so that dfl_root can link to
4543 hash_add(css_set_table, &init_css_set.hlist,
4544 css_set_hash(init_css_set.subsys));
4546 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
4548 mutex_unlock(&cgroup_mutex);
4550 for_each_subsys(ss, ssid) {
4551 if (ss->early_init) {
4552 struct cgroup_subsys_state *css =
4553 init_css_set.subsys[ss->id];
4555 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
4557 BUG_ON(css->id < 0);
4559 cgroup_init_subsys(ss, false);
4562 list_add_tail(&init_css_set.e_cset_node[ssid],
4563 &cgrp_dfl_root.cgrp.e_csets[ssid]);
4566 * Setting dfl_root subsys_mask needs to consider the
4567 * disabled flag and cftype registration needs kmalloc,
4568 * both of which aren't available during early_init.
4570 if (cgroup_disable_mask & (1 << ssid)) {
4571 static_branch_disable(cgroup_subsys_enabled_key[ssid]);
4572 printk(KERN_INFO "Disabling %s control group subsystem\n",
4577 if (cgroup1_ssid_disabled(ssid))
4578 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
4581 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
4583 if (ss->implicit_on_dfl)
4584 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
4585 else if (!ss->dfl_cftypes)
4586 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
4588 if (ss->dfl_cftypes == ss->legacy_cftypes) {
4589 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
4591 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
4592 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
4596 ss->bind(init_css_set.subsys[ssid]);
4599 /* init_css_set.subsys[] has been updated, re-hash */
4600 hash_del(&init_css_set.hlist);
4601 hash_add(css_set_table, &init_css_set.hlist,
4602 css_set_hash(init_css_set.subsys));
4604 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
4605 WARN_ON(register_filesystem(&cgroup_fs_type));
4606 WARN_ON(register_filesystem(&cgroup2_fs_type));
4607 WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations));
4612 static int __init cgroup_wq_init(void)
4615 * There isn't much point in executing destruction path in
4616 * parallel. Good chunk is serialized with cgroup_mutex anyway.
4617 * Use 1 for @max_active.
4619 * We would prefer to do this in cgroup_init() above, but that
4620 * is called before init_workqueues(): so leave this until after.
4622 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
4623 BUG_ON(!cgroup_destroy_wq);
4626 core_initcall(cgroup_wq_init);
4629 * proc_cgroup_show()
4630 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4631 * - Used for /proc/<pid>/cgroup.
4633 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
4634 struct pid *pid, struct task_struct *tsk)
4638 struct cgroup_root *root;
4641 buf = kmalloc(PATH_MAX, GFP_KERNEL);
4645 mutex_lock(&cgroup_mutex);
4646 spin_lock_irq(&css_set_lock);
4648 for_each_root(root) {
4649 struct cgroup_subsys *ss;
4650 struct cgroup *cgrp;
4651 int ssid, count = 0;
4653 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
4656 seq_printf(m, "%d:", root->hierarchy_id);
4657 if (root != &cgrp_dfl_root)
4658 for_each_subsys(ss, ssid)
4659 if (root->subsys_mask & (1 << ssid))
4660 seq_printf(m, "%s%s", count++ ? "," : "",
4662 if (strlen(root->name))
4663 seq_printf(m, "%sname=%s", count ? "," : "",
4667 cgrp = task_cgroup_from_root(tsk, root);
4670 * On traditional hierarchies, all zombie tasks show up as
4671 * belonging to the root cgroup. On the default hierarchy,
4672 * while a zombie doesn't show up in "cgroup.procs" and
4673 * thus can't be migrated, its /proc/PID/cgroup keeps
4674 * reporting the cgroup it belonged to before exiting. If
4675 * the cgroup is removed before the zombie is reaped,
4676 * " (deleted)" is appended to the cgroup path.
4678 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
4679 retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
4680 current->nsproxy->cgroup_ns);
4681 if (retval >= PATH_MAX)
4682 retval = -ENAMETOOLONG;
4691 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
4692 seq_puts(m, " (deleted)\n");
4699 spin_unlock_irq(&css_set_lock);
4700 mutex_unlock(&cgroup_mutex);
4707 * cgroup_fork - initialize cgroup related fields during copy_process()
4708 * @child: pointer to task_struct of forking parent process.
4710 * A task is associated with the init_css_set until cgroup_post_fork()
4711 * attaches it to the parent's css_set. Empty cg_list indicates that
4712 * @child isn't holding reference to its css_set.
4714 void cgroup_fork(struct task_struct *child)
4716 RCU_INIT_POINTER(child->cgroups, &init_css_set);
4717 INIT_LIST_HEAD(&child->cg_list);
4721 * cgroup_can_fork - called on a new task before the process is exposed
4722 * @child: the task in question.
4724 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
4725 * returns an error, the fork aborts with that error code. This allows for
4726 * a cgroup subsystem to conditionally allow or deny new forks.
4728 int cgroup_can_fork(struct task_struct *child)
4730 struct cgroup_subsys *ss;
4733 do_each_subsys_mask(ss, i, have_canfork_callback) {
4734 ret = ss->can_fork(child);
4737 } while_each_subsys_mask();
4742 for_each_subsys(ss, j) {
4745 if (ss->cancel_fork)
4746 ss->cancel_fork(child);
4753 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
4754 * @child: the task in question
4756 * This calls the cancel_fork() callbacks if a fork failed *after*
4757 * cgroup_can_fork() succeded.
4759 void cgroup_cancel_fork(struct task_struct *child)
4761 struct cgroup_subsys *ss;
4764 for_each_subsys(ss, i)
4765 if (ss->cancel_fork)
4766 ss->cancel_fork(child);
4770 * cgroup_post_fork - called on a new task after adding it to the task list
4771 * @child: the task in question
4773 * Adds the task to the list running through its css_set if necessary and
4774 * call the subsystem fork() callbacks. Has to be after the task is
4775 * visible on the task list in case we race with the first call to
4776 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
4779 void cgroup_post_fork(struct task_struct *child)
4781 struct cgroup_subsys *ss;
4785 * This may race against cgroup_enable_task_cg_lists(). As that
4786 * function sets use_task_css_set_links before grabbing
4787 * tasklist_lock and we just went through tasklist_lock to add
4788 * @child, it's guaranteed that either we see the set
4789 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
4790 * @child during its iteration.
4792 * If we won the race, @child is associated with %current's
4793 * css_set. Grabbing css_set_lock guarantees both that the
4794 * association is stable, and, on completion of the parent's
4795 * migration, @child is visible in the source of migration or
4796 * already in the destination cgroup. This guarantee is necessary
4797 * when implementing operations which need to migrate all tasks of
4798 * a cgroup to another.
4800 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
4801 * will remain in init_css_set. This is safe because all tasks are
4802 * in the init_css_set before cg_links is enabled and there's no
4803 * operation which transfers all tasks out of init_css_set.
4805 if (use_task_css_set_links) {
4806 struct css_set *cset;
4808 spin_lock_irq(&css_set_lock);
4809 cset = task_css_set(current);
4810 if (list_empty(&child->cg_list)) {
4812 css_set_move_task(child, NULL, cset, false);
4814 spin_unlock_irq(&css_set_lock);
4818 * Call ss->fork(). This must happen after @child is linked on
4819 * css_set; otherwise, @child might change state between ->fork()
4820 * and addition to css_set.
4822 do_each_subsys_mask(ss, i, have_fork_callback) {
4824 } while_each_subsys_mask();
4828 * cgroup_exit - detach cgroup from exiting task
4829 * @tsk: pointer to task_struct of exiting process
4831 * Description: Detach cgroup from @tsk and release it.
4833 * Note that cgroups marked notify_on_release force every task in
4834 * them to take the global cgroup_mutex mutex when exiting.
4835 * This could impact scaling on very large systems. Be reluctant to
4836 * use notify_on_release cgroups where very high task exit scaling
4837 * is required on large systems.
4839 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
4840 * call cgroup_exit() while the task is still competent to handle
4841 * notify_on_release(), then leave the task attached to the root cgroup in
4842 * each hierarchy for the remainder of its exit. No need to bother with
4843 * init_css_set refcnting. init_css_set never goes away and we can't race
4844 * with migration path - PF_EXITING is visible to migration path.
4846 void cgroup_exit(struct task_struct *tsk)
4848 struct cgroup_subsys *ss;
4849 struct css_set *cset;
4853 * Unlink from @tsk from its css_set. As migration path can't race
4854 * with us, we can check css_set and cg_list without synchronization.
4856 cset = task_css_set(tsk);
4858 if (!list_empty(&tsk->cg_list)) {
4859 spin_lock_irq(&css_set_lock);
4860 css_set_move_task(tsk, cset, NULL, false);
4861 spin_unlock_irq(&css_set_lock);
4866 /* see cgroup_post_fork() for details */
4867 do_each_subsys_mask(ss, i, have_exit_callback) {
4869 } while_each_subsys_mask();
4872 void cgroup_free(struct task_struct *task)
4874 struct css_set *cset = task_css_set(task);
4875 struct cgroup_subsys *ss;
4878 do_each_subsys_mask(ss, ssid, have_free_callback) {
4880 } while_each_subsys_mask();
4885 static int __init cgroup_disable(char *str)
4887 struct cgroup_subsys *ss;
4891 while ((token = strsep(&str, ",")) != NULL) {
4895 for_each_subsys(ss, i) {
4896 if (strcmp(token, ss->name) &&
4897 strcmp(token, ss->legacy_name))
4899 cgroup_disable_mask |= 1 << i;
4904 __setup("cgroup_disable=", cgroup_disable);
4907 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
4908 * @dentry: directory dentry of interest
4909 * @ss: subsystem of interest
4911 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
4912 * to get the corresponding css and return it. If such css doesn't exist
4913 * or can't be pinned, an ERR_PTR value is returned.
4915 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
4916 struct cgroup_subsys *ss)
4918 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
4919 struct file_system_type *s_type = dentry->d_sb->s_type;
4920 struct cgroup_subsys_state *css = NULL;
4921 struct cgroup *cgrp;
4923 /* is @dentry a cgroup dir? */
4924 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
4925 !kn || kernfs_type(kn) != KERNFS_DIR)
4926 return ERR_PTR(-EBADF);
4931 * This path doesn't originate from kernfs and @kn could already
4932 * have been or be removed at any point. @kn->priv is RCU
4933 * protected for this access. See css_release_work_fn() for details.
4935 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
4937 css = cgroup_css(cgrp, ss);
4939 if (!css || !css_tryget_online(css))
4940 css = ERR_PTR(-ENOENT);
4947 * css_from_id - lookup css by id
4948 * @id: the cgroup id
4949 * @ss: cgroup subsys to be looked into
4951 * Returns the css if there's valid one with @id, otherwise returns NULL.
4952 * Should be called under rcu_read_lock().
4954 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
4956 WARN_ON_ONCE(!rcu_read_lock_held());
4957 return idr_find(&ss->css_idr, id);
4961 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
4962 * @path: path on the default hierarchy
4964 * Find the cgroup at @path on the default hierarchy, increment its
4965 * reference count and return it. Returns pointer to the found cgroup on
4966 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
4967 * if @path points to a non-directory.
4969 struct cgroup *cgroup_get_from_path(const char *path)
4971 struct kernfs_node *kn;
4972 struct cgroup *cgrp;
4974 mutex_lock(&cgroup_mutex);
4976 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
4978 if (kernfs_type(kn) == KERNFS_DIR) {
4982 cgrp = ERR_PTR(-ENOTDIR);
4986 cgrp = ERR_PTR(-ENOENT);
4989 mutex_unlock(&cgroup_mutex);
4992 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
4995 * cgroup_get_from_fd - get a cgroup pointer from a fd
4996 * @fd: fd obtained by open(cgroup2_dir)
4998 * Find the cgroup from a fd which should be obtained
4999 * by opening a cgroup directory. Returns a pointer to the
5000 * cgroup on success. ERR_PTR is returned if the cgroup
5003 struct cgroup *cgroup_get_from_fd(int fd)
5005 struct cgroup_subsys_state *css;
5006 struct cgroup *cgrp;
5011 return ERR_PTR(-EBADF);
5013 css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
5016 return ERR_CAST(css);
5019 if (!cgroup_on_dfl(cgrp)) {
5021 return ERR_PTR(-EBADF);
5026 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
5029 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
5030 * definition in cgroup-defs.h.
5032 #ifdef CONFIG_SOCK_CGROUP_DATA
5034 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
5036 DEFINE_SPINLOCK(cgroup_sk_update_lock);
5037 static bool cgroup_sk_alloc_disabled __read_mostly;
5039 void cgroup_sk_alloc_disable(void)
5041 if (cgroup_sk_alloc_disabled)
5043 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
5044 cgroup_sk_alloc_disabled = true;
5049 #define cgroup_sk_alloc_disabled false
5053 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
5055 if (cgroup_sk_alloc_disabled)
5058 /* Socket clone path */
5060 cgroup_get(sock_cgroup_ptr(skcd));
5067 struct css_set *cset;
5069 cset = task_css_set(current);
5070 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
5071 skcd->val = (unsigned long)cset->dfl_cgrp;
5080 void cgroup_sk_free(struct sock_cgroup_data *skcd)
5082 cgroup_put(sock_cgroup_ptr(skcd));
5085 #endif /* CONFIG_SOCK_CGROUP_DATA */
5087 #ifdef CONFIG_CGROUP_BPF
5088 void cgroup_bpf_update(struct cgroup *cgrp,
5089 struct bpf_prog *prog,
5090 enum bpf_attach_type type)
5092 struct cgroup *parent = cgroup_parent(cgrp);
5094 mutex_lock(&cgroup_mutex);
5095 __cgroup_bpf_update(cgrp, parent, prog, type);
5096 mutex_unlock(&cgroup_mutex);
5098 #endif /* CONFIG_CGROUP_BPF */