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 u16 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 bitmasks identify subsystems with specific features to avoid
182 * having to do iterative checks repeatedly.
184 static u16 have_fork_callback __read_mostly;
185 static u16 have_exit_callback __read_mostly;
186 static u16 have_free_callback __read_mostly;
187 static u16 have_canfork_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 static struct file_system_type cgroup2_fs_type;
199 static struct cftype cgroup_base_files[];
201 static int cgroup_apply_control(struct cgroup *cgrp);
202 static void cgroup_finalize_control(struct cgroup *cgrp, int ret);
203 static void css_task_iter_advance(struct css_task_iter *it);
204 static int cgroup_destroy_locked(struct cgroup *cgrp);
205 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
206 struct cgroup_subsys *ss);
207 static void css_release(struct percpu_ref *ref);
208 static void kill_css(struct cgroup_subsys_state *css);
209 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
210 struct cgroup *cgrp, struct cftype cfts[],
214 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
215 * @ssid: subsys ID of interest
217 * cgroup_subsys_enabled() can only be used with literal subsys names which
218 * is fine for individual subsystems but unsuitable for cgroup core. This
219 * is slower static_key_enabled() based test indexed by @ssid.
221 bool cgroup_ssid_enabled(int ssid)
223 if (CGROUP_SUBSYS_COUNT == 0)
226 return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
230 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
231 * @cgrp: the cgroup of interest
233 * The default hierarchy is the v2 interface of cgroup and this function
234 * can be used to test whether a cgroup is on the default hierarchy for
235 * cases where a subsystem should behave differnetly depending on the
238 * The set of behaviors which change on the default hierarchy are still
239 * being determined and the mount option is prefixed with __DEVEL__.
241 * List of changed behaviors:
243 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
244 * and "name" are disallowed.
246 * - When mounting an existing superblock, mount options should match.
248 * - Remount is disallowed.
250 * - rename(2) is disallowed.
252 * - "tasks" is removed. Everything should be at process granularity. Use
253 * "cgroup.procs" instead.
255 * - "cgroup.procs" is not sorted. pids will be unique unless they got
256 * recycled inbetween reads.
258 * - "release_agent" and "notify_on_release" are removed. Replacement
259 * notification mechanism will be implemented.
261 * - "cgroup.clone_children" is removed.
263 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
264 * and its descendants contain no task; otherwise, 1. The file also
265 * generates kernfs notification which can be monitored through poll and
266 * [di]notify when the value of the file changes.
268 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
269 * take masks of ancestors with non-empty cpus/mems, instead of being
270 * moved to an ancestor.
272 * - cpuset: a task can be moved into an empty cpuset, and again it takes
273 * masks of ancestors.
275 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
278 * - blkcg: blk-throttle becomes properly hierarchical.
280 * - debug: disallowed on the default hierarchy.
282 bool cgroup_on_dfl(const struct cgroup *cgrp)
284 return cgrp->root == &cgrp_dfl_root;
287 /* IDR wrappers which synchronize using cgroup_idr_lock */
288 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
293 idr_preload(gfp_mask);
294 spin_lock_bh(&cgroup_idr_lock);
295 ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
296 spin_unlock_bh(&cgroup_idr_lock);
301 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
305 spin_lock_bh(&cgroup_idr_lock);
306 ret = idr_replace(idr, ptr, id);
307 spin_unlock_bh(&cgroup_idr_lock);
311 static void cgroup_idr_remove(struct idr *idr, int id)
313 spin_lock_bh(&cgroup_idr_lock);
315 spin_unlock_bh(&cgroup_idr_lock);
318 static struct cgroup *cgroup_parent(struct cgroup *cgrp)
320 struct cgroup_subsys_state *parent_css = cgrp->self.parent;
323 return container_of(parent_css, struct cgroup, self);
327 /* subsystems visibly enabled on a cgroup */
328 static u16 cgroup_control(struct cgroup *cgrp)
330 struct cgroup *parent = cgroup_parent(cgrp);
331 u16 root_ss_mask = cgrp->root->subsys_mask;
334 return parent->subtree_control;
336 if (cgroup_on_dfl(cgrp))
337 root_ss_mask &= ~(cgrp_dfl_inhibit_ss_mask |
338 cgrp_dfl_implicit_ss_mask);
342 /* subsystems enabled on a cgroup */
343 static u16 cgroup_ss_mask(struct cgroup *cgrp)
345 struct cgroup *parent = cgroup_parent(cgrp);
348 return parent->subtree_ss_mask;
350 return cgrp->root->subsys_mask;
354 * cgroup_css - obtain a cgroup's css for the specified subsystem
355 * @cgrp: the cgroup of interest
356 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
358 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
359 * function must be called either under cgroup_mutex or rcu_read_lock() and
360 * the caller is responsible for pinning the returned css if it wants to
361 * keep accessing it outside the said locks. This function may return
362 * %NULL if @cgrp doesn't have @subsys_id enabled.
364 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
365 struct cgroup_subsys *ss)
368 return rcu_dereference_check(cgrp->subsys[ss->id],
369 lockdep_is_held(&cgroup_mutex));
375 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
376 * @cgrp: the cgroup of interest
377 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
379 * Similar to cgroup_css() but returns the effective css, which is defined
380 * as the matching css of the nearest ancestor including self which has @ss
381 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
382 * function is guaranteed to return non-NULL css.
384 static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
385 struct cgroup_subsys *ss)
387 lockdep_assert_held(&cgroup_mutex);
393 * This function is used while updating css associations and thus
394 * can't test the csses directly. Test ss_mask.
396 while (!(cgroup_ss_mask(cgrp) & (1 << ss->id))) {
397 cgrp = cgroup_parent(cgrp);
402 return cgroup_css(cgrp, ss);
406 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
407 * @cgrp: the cgroup of interest
408 * @ss: the subsystem of interest
410 * Find and get the effective css of @cgrp for @ss. The effective css is
411 * defined as the matching css of the nearest ancestor including self which
412 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
413 * the root css is returned, so this function always returns a valid css.
414 * The returned css must be put using css_put().
416 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
417 struct cgroup_subsys *ss)
419 struct cgroup_subsys_state *css;
424 css = cgroup_css(cgrp, ss);
426 if (css && css_tryget_online(css))
428 cgrp = cgroup_parent(cgrp);
431 css = init_css_set.subsys[ss->id];
438 static void cgroup_get(struct cgroup *cgrp)
440 WARN_ON_ONCE(cgroup_is_dead(cgrp));
441 css_get(&cgrp->self);
444 static bool cgroup_tryget(struct cgroup *cgrp)
446 return css_tryget(&cgrp->self);
449 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
451 struct cgroup *cgrp = of->kn->parent->priv;
452 struct cftype *cft = of_cft(of);
455 * This is open and unprotected implementation of cgroup_css().
456 * seq_css() is only called from a kernfs file operation which has
457 * an active reference on the file. Because all the subsystem
458 * files are drained before a css is disassociated with a cgroup,
459 * the matching css from the cgroup's subsys table is guaranteed to
460 * be and stay valid until the enclosing operation is complete.
463 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
467 EXPORT_SYMBOL_GPL(of_css);
470 * for_each_css - iterate all css's of a cgroup
471 * @css: the iteration cursor
472 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
473 * @cgrp: the target cgroup to iterate css's of
475 * Should be called under cgroup_[tree_]mutex.
477 #define for_each_css(css, ssid, cgrp) \
478 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
479 if (!((css) = rcu_dereference_check( \
480 (cgrp)->subsys[(ssid)], \
481 lockdep_is_held(&cgroup_mutex)))) { } \
485 * for_each_e_css - iterate all effective css's of a cgroup
486 * @css: the iteration cursor
487 * @ssid: the index of the subsystem, CGROUP_SUBSYS_COUNT after reaching the end
488 * @cgrp: the target cgroup to iterate css's of
490 * Should be called under cgroup_[tree_]mutex.
492 #define for_each_e_css(css, ssid, cgrp) \
493 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT; (ssid)++) \
494 if (!((css) = cgroup_e_css(cgrp, cgroup_subsys[(ssid)]))) \
499 * do_each_subsys_mask - filter for_each_subsys with a bitmask
500 * @ss: the iteration cursor
501 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
502 * @ss_mask: the bitmask
504 * The block will only run for cases where the ssid-th bit (1 << ssid) of
507 #define do_each_subsys_mask(ss, ssid, ss_mask) do { \
508 unsigned long __ss_mask = (ss_mask); \
509 if (!CGROUP_SUBSYS_COUNT) { /* to avoid spurious gcc warning */ \
513 for_each_set_bit(ssid, &__ss_mask, CGROUP_SUBSYS_COUNT) { \
514 (ss) = cgroup_subsys[ssid]; \
517 #define while_each_subsys_mask() \
522 /* iterate over child cgrps, lock should be held throughout iteration */
523 #define cgroup_for_each_live_child(child, cgrp) \
524 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
525 if (({ lockdep_assert_held(&cgroup_mutex); \
526 cgroup_is_dead(child); })) \
530 /* walk live descendants in preorder */
531 #define cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) \
532 css_for_each_descendant_pre((d_css), cgroup_css((cgrp), NULL)) \
533 if (({ lockdep_assert_held(&cgroup_mutex); \
534 (dsct) = (d_css)->cgroup; \
535 cgroup_is_dead(dsct); })) \
539 /* walk live descendants in postorder */
540 #define cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) \
541 css_for_each_descendant_post((d_css), cgroup_css((cgrp), NULL)) \
542 if (({ lockdep_assert_held(&cgroup_mutex); \
543 (dsct) = (d_css)->cgroup; \
544 cgroup_is_dead(dsct); })) \
549 * The default css_set - used by init and its children prior to any
550 * hierarchies being mounted. It contains a pointer to the root state
551 * for each subsystem. Also used to anchor the list of css_sets. Not
552 * reference-counted, to improve performance when child cgroups
553 * haven't been created.
555 struct css_set init_css_set = {
556 .refcount = ATOMIC_INIT(1),
557 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
558 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
559 .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
560 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
561 .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
562 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
565 static int css_set_count = 1; /* 1 for init_css_set */
568 * css_set_populated - does a css_set contain any tasks?
569 * @cset: target css_set
571 static bool css_set_populated(struct css_set *cset)
573 lockdep_assert_held(&css_set_lock);
575 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
579 * cgroup_update_populated - updated populated count of a cgroup
580 * @cgrp: the target cgroup
581 * @populated: inc or dec populated count
583 * One of the css_sets associated with @cgrp is either getting its first
584 * task or losing the last. Update @cgrp->populated_cnt accordingly. The
585 * count is propagated towards root so that a given cgroup's populated_cnt
586 * is zero iff the cgroup and all its descendants don't contain any tasks.
588 * @cgrp's interface file "cgroup.populated" is zero if
589 * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
590 * changes from or to zero, userland is notified that the content of the
591 * interface file has changed. This can be used to detect when @cgrp and
592 * its descendants become populated or empty.
594 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
596 lockdep_assert_held(&css_set_lock);
602 trigger = !cgrp->populated_cnt++;
604 trigger = !--cgrp->populated_cnt;
609 cgroup1_check_for_release(cgrp);
610 cgroup_file_notify(&cgrp->events_file);
612 cgrp = cgroup_parent(cgrp);
617 * css_set_update_populated - update populated state of a css_set
618 * @cset: target css_set
619 * @populated: whether @cset is populated or depopulated
621 * @cset is either getting the first task or losing the last. Update the
622 * ->populated_cnt of all associated cgroups accordingly.
624 static void css_set_update_populated(struct css_set *cset, bool populated)
626 struct cgrp_cset_link *link;
628 lockdep_assert_held(&css_set_lock);
630 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
631 cgroup_update_populated(link->cgrp, populated);
635 * css_set_move_task - move a task from one css_set to another
636 * @task: task being moved
637 * @from_cset: css_set @task currently belongs to (may be NULL)
638 * @to_cset: new css_set @task is being moved to (may be NULL)
639 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
641 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
642 * css_set, @from_cset can be NULL. If @task is being disassociated
643 * instead of moved, @to_cset can be NULL.
645 * This function automatically handles populated_cnt updates and
646 * css_task_iter adjustments but the caller is responsible for managing
647 * @from_cset and @to_cset's reference counts.
649 static void css_set_move_task(struct task_struct *task,
650 struct css_set *from_cset, struct css_set *to_cset,
653 lockdep_assert_held(&css_set_lock);
655 if (to_cset && !css_set_populated(to_cset))
656 css_set_update_populated(to_cset, true);
659 struct css_task_iter *it, *pos;
661 WARN_ON_ONCE(list_empty(&task->cg_list));
664 * @task is leaving, advance task iterators which are
665 * pointing to it so that they can resume at the next
666 * position. Advancing an iterator might remove it from
667 * the list, use safe walk. See css_task_iter_advance*()
670 list_for_each_entry_safe(it, pos, &from_cset->task_iters,
672 if (it->task_pos == &task->cg_list)
673 css_task_iter_advance(it);
675 list_del_init(&task->cg_list);
676 if (!css_set_populated(from_cset))
677 css_set_update_populated(from_cset, false);
679 WARN_ON_ONCE(!list_empty(&task->cg_list));
684 * We are synchronized through cgroup_threadgroup_rwsem
685 * against PF_EXITING setting such that we can't race
686 * against cgroup_exit() changing the css_set to
687 * init_css_set and dropping the old one.
689 WARN_ON_ONCE(task->flags & PF_EXITING);
691 rcu_assign_pointer(task->cgroups, to_cset);
692 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
698 * hash table for cgroup groups. This improves the performance to find
699 * an existing css_set. This hash doesn't (currently) take into
700 * account cgroups in empty hierarchies.
702 #define CSS_SET_HASH_BITS 7
703 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
705 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
707 unsigned long key = 0UL;
708 struct cgroup_subsys *ss;
711 for_each_subsys(ss, i)
712 key += (unsigned long)css[i];
713 key = (key >> 16) ^ key;
718 void put_css_set_locked(struct css_set *cset)
720 struct cgrp_cset_link *link, *tmp_link;
721 struct cgroup_subsys *ss;
724 lockdep_assert_held(&css_set_lock);
726 if (!atomic_dec_and_test(&cset->refcount))
729 /* This css_set is dead. unlink it and release cgroup and css refs */
730 for_each_subsys(ss, ssid) {
731 list_del(&cset->e_cset_node[ssid]);
732 css_put(cset->subsys[ssid]);
734 hash_del(&cset->hlist);
737 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
738 list_del(&link->cset_link);
739 list_del(&link->cgrp_link);
740 if (cgroup_parent(link->cgrp))
741 cgroup_put(link->cgrp);
745 kfree_rcu(cset, rcu_head);
749 * compare_css_sets - helper function for find_existing_css_set().
750 * @cset: candidate css_set being tested
751 * @old_cset: existing css_set for a task
752 * @new_cgrp: cgroup that's being entered by the task
753 * @template: desired set of css pointers in css_set (pre-calculated)
755 * Returns true if "cset" matches "old_cset" except for the hierarchy
756 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
758 static bool compare_css_sets(struct css_set *cset,
759 struct css_set *old_cset,
760 struct cgroup *new_cgrp,
761 struct cgroup_subsys_state *template[])
763 struct list_head *l1, *l2;
766 * On the default hierarchy, there can be csets which are
767 * associated with the same set of cgroups but different csses.
768 * Let's first ensure that csses match.
770 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
774 * Compare cgroup pointers in order to distinguish between
775 * different cgroups in hierarchies. As different cgroups may
776 * share the same effective css, this comparison is always
779 l1 = &cset->cgrp_links;
780 l2 = &old_cset->cgrp_links;
782 struct cgrp_cset_link *link1, *link2;
783 struct cgroup *cgrp1, *cgrp2;
787 /* See if we reached the end - both lists are equal length. */
788 if (l1 == &cset->cgrp_links) {
789 BUG_ON(l2 != &old_cset->cgrp_links);
792 BUG_ON(l2 == &old_cset->cgrp_links);
794 /* Locate the cgroups associated with these links. */
795 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
796 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
799 /* Hierarchies should be linked in the same order. */
800 BUG_ON(cgrp1->root != cgrp2->root);
803 * If this hierarchy is the hierarchy of the cgroup
804 * that's changing, then we need to check that this
805 * css_set points to the new cgroup; if it's any other
806 * hierarchy, then this css_set should point to the
807 * same cgroup as the old css_set.
809 if (cgrp1->root == new_cgrp->root) {
810 if (cgrp1 != new_cgrp)
821 * find_existing_css_set - init css array and find the matching css_set
822 * @old_cset: the css_set that we're using before the cgroup transition
823 * @cgrp: the cgroup that we're moving into
824 * @template: out param for the new set of csses, should be clear on entry
826 static struct css_set *find_existing_css_set(struct css_set *old_cset,
828 struct cgroup_subsys_state *template[])
830 struct cgroup_root *root = cgrp->root;
831 struct cgroup_subsys *ss;
832 struct css_set *cset;
837 * Build the set of subsystem state objects that we want to see in the
838 * new css_set. while subsystems can change globally, the entries here
839 * won't change, so no need for locking.
841 for_each_subsys(ss, i) {
842 if (root->subsys_mask & (1UL << i)) {
844 * @ss is in this hierarchy, so we want the
845 * effective css from @cgrp.
847 template[i] = cgroup_e_css(cgrp, ss);
850 * @ss is not in this hierarchy, so we don't want
853 template[i] = old_cset->subsys[i];
857 key = css_set_hash(template);
858 hash_for_each_possible(css_set_table, cset, hlist, key) {
859 if (!compare_css_sets(cset, old_cset, cgrp, template))
862 /* This css_set matches what we need */
866 /* No existing cgroup group matched */
870 static void free_cgrp_cset_links(struct list_head *links_to_free)
872 struct cgrp_cset_link *link, *tmp_link;
874 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
875 list_del(&link->cset_link);
881 * allocate_cgrp_cset_links - allocate cgrp_cset_links
882 * @count: the number of links to allocate
883 * @tmp_links: list_head the allocated links are put on
885 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
886 * through ->cset_link. Returns 0 on success or -errno.
888 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
890 struct cgrp_cset_link *link;
893 INIT_LIST_HEAD(tmp_links);
895 for (i = 0; i < count; i++) {
896 link = kzalloc(sizeof(*link), GFP_KERNEL);
898 free_cgrp_cset_links(tmp_links);
901 list_add(&link->cset_link, tmp_links);
907 * link_css_set - a helper function to link a css_set to a cgroup
908 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
909 * @cset: the css_set to be linked
910 * @cgrp: the destination cgroup
912 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
915 struct cgrp_cset_link *link;
917 BUG_ON(list_empty(tmp_links));
919 if (cgroup_on_dfl(cgrp))
920 cset->dfl_cgrp = cgrp;
922 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
927 * Always add links to the tail of the lists so that the lists are
928 * in choronological order.
930 list_move_tail(&link->cset_link, &cgrp->cset_links);
931 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
933 if (cgroup_parent(cgrp))
938 * find_css_set - return a new css_set with one cgroup updated
939 * @old_cset: the baseline css_set
940 * @cgrp: the cgroup to be updated
942 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
943 * substituted into the appropriate hierarchy.
945 static struct css_set *find_css_set(struct css_set *old_cset,
948 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
949 struct css_set *cset;
950 struct list_head tmp_links;
951 struct cgrp_cset_link *link;
952 struct cgroup_subsys *ss;
956 lockdep_assert_held(&cgroup_mutex);
958 /* First see if we already have a cgroup group that matches
960 spin_lock_irq(&css_set_lock);
961 cset = find_existing_css_set(old_cset, cgrp, template);
964 spin_unlock_irq(&css_set_lock);
969 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
973 /* Allocate all the cgrp_cset_link objects that we'll need */
974 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
979 atomic_set(&cset->refcount, 1);
980 INIT_LIST_HEAD(&cset->tasks);
981 INIT_LIST_HEAD(&cset->mg_tasks);
982 INIT_LIST_HEAD(&cset->task_iters);
983 INIT_HLIST_NODE(&cset->hlist);
984 INIT_LIST_HEAD(&cset->cgrp_links);
985 INIT_LIST_HEAD(&cset->mg_preload_node);
986 INIT_LIST_HEAD(&cset->mg_node);
988 /* Copy the set of subsystem state objects generated in
989 * find_existing_css_set() */
990 memcpy(cset->subsys, template, sizeof(cset->subsys));
992 spin_lock_irq(&css_set_lock);
993 /* Add reference counts and links from the new css_set. */
994 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
995 struct cgroup *c = link->cgrp;
997 if (c->root == cgrp->root)
999 link_css_set(&tmp_links, cset, c);
1002 BUG_ON(!list_empty(&tmp_links));
1006 /* Add @cset to the hash table */
1007 key = css_set_hash(cset->subsys);
1008 hash_add(css_set_table, &cset->hlist, key);
1010 for_each_subsys(ss, ssid) {
1011 struct cgroup_subsys_state *css = cset->subsys[ssid];
1013 list_add_tail(&cset->e_cset_node[ssid],
1014 &css->cgroup->e_csets[ssid]);
1018 spin_unlock_irq(&css_set_lock);
1023 struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1025 struct cgroup *root_cgrp = kf_root->kn->priv;
1027 return root_cgrp->root;
1030 static int cgroup_init_root_id(struct cgroup_root *root)
1034 lockdep_assert_held(&cgroup_mutex);
1036 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1040 root->hierarchy_id = id;
1044 static void cgroup_exit_root_id(struct cgroup_root *root)
1046 lockdep_assert_held(&cgroup_mutex);
1048 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1051 void cgroup_free_root(struct cgroup_root *root)
1054 idr_destroy(&root->cgroup_idr);
1059 static void cgroup_destroy_root(struct cgroup_root *root)
1061 struct cgroup *cgrp = &root->cgrp;
1062 struct cgrp_cset_link *link, *tmp_link;
1064 trace_cgroup_destroy_root(root);
1066 cgroup_lock_and_drain_offline(&cgrp_dfl_root.cgrp);
1068 BUG_ON(atomic_read(&root->nr_cgrps));
1069 BUG_ON(!list_empty(&cgrp->self.children));
1071 /* Rebind all subsystems back to the default hierarchy */
1072 WARN_ON(rebind_subsystems(&cgrp_dfl_root, root->subsys_mask));
1075 * Release all the links from cset_links to this hierarchy's
1078 spin_lock_irq(&css_set_lock);
1080 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1081 list_del(&link->cset_link);
1082 list_del(&link->cgrp_link);
1086 spin_unlock_irq(&css_set_lock);
1088 if (!list_empty(&root->root_list)) {
1089 list_del(&root->root_list);
1090 cgroup_root_count--;
1093 cgroup_exit_root_id(root);
1095 mutex_unlock(&cgroup_mutex);
1097 kernfs_destroy_root(root->kf_root);
1098 cgroup_free_root(root);
1102 * look up cgroup associated with current task's cgroup namespace on the
1103 * specified hierarchy
1105 static struct cgroup *
1106 current_cgns_cgroup_from_root(struct cgroup_root *root)
1108 struct cgroup *res = NULL;
1109 struct css_set *cset;
1111 lockdep_assert_held(&css_set_lock);
1115 cset = current->nsproxy->cgroup_ns->root_cset;
1116 if (cset == &init_css_set) {
1119 struct cgrp_cset_link *link;
1121 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1122 struct cgroup *c = link->cgrp;
1124 if (c->root == root) {
1136 /* look up cgroup associated with given css_set on the specified hierarchy */
1137 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1138 struct cgroup_root *root)
1140 struct cgroup *res = NULL;
1142 lockdep_assert_held(&cgroup_mutex);
1143 lockdep_assert_held(&css_set_lock);
1145 if (cset == &init_css_set) {
1148 struct cgrp_cset_link *link;
1150 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1151 struct cgroup *c = link->cgrp;
1153 if (c->root == root) {
1165 * Return the cgroup for "task" from the given hierarchy. Must be
1166 * called with cgroup_mutex and css_set_lock held.
1168 struct cgroup *task_cgroup_from_root(struct task_struct *task,
1169 struct cgroup_root *root)
1172 * No need to lock the task - since we hold cgroup_mutex the
1173 * task can't change groups, so the only thing that can happen
1174 * is that it exits and its css is set back to init_css_set.
1176 return cset_cgroup_from_root(task_css_set(task), root);
1180 * A task must hold cgroup_mutex to modify cgroups.
1182 * Any task can increment and decrement the count field without lock.
1183 * So in general, code holding cgroup_mutex can't rely on the count
1184 * field not changing. However, if the count goes to zero, then only
1185 * cgroup_attach_task() can increment it again. Because a count of zero
1186 * means that no tasks are currently attached, therefore there is no
1187 * way a task attached to that cgroup can fork (the other way to
1188 * increment the count). So code holding cgroup_mutex can safely
1189 * assume that if the count is zero, it will stay zero. Similarly, if
1190 * a task holds cgroup_mutex on a cgroup with zero count, it
1191 * knows that the cgroup won't be removed, as cgroup_rmdir()
1194 * A cgroup can only be deleted if both its 'count' of using tasks
1195 * is zero, and its list of 'children' cgroups is empty. Since all
1196 * tasks in the system use _some_ cgroup, and since there is always at
1197 * least one task in the system (init, pid == 1), therefore, root cgroup
1198 * always has either children cgroups and/or using tasks. So we don't
1199 * need a special hack to ensure that root cgroup cannot be deleted.
1201 * P.S. One more locking exception. RCU is used to guard the
1202 * update of a tasks cgroup pointer by cgroup_attach_task()
1205 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1207 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1210 struct cgroup_subsys *ss = cft->ss;
1212 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1213 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
1214 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
1215 cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1218 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1223 * cgroup_file_mode - deduce file mode of a control file
1224 * @cft: the control file in question
1226 * S_IRUGO for read, S_IWUSR for write.
1228 static umode_t cgroup_file_mode(const struct cftype *cft)
1232 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1235 if (cft->write_u64 || cft->write_s64 || cft->write) {
1236 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1246 * cgroup_calc_subtree_ss_mask - calculate subtree_ss_mask
1247 * @subtree_control: the new subtree_control mask to consider
1248 * @this_ss_mask: available subsystems
1250 * On the default hierarchy, a subsystem may request other subsystems to be
1251 * enabled together through its ->depends_on mask. In such cases, more
1252 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1254 * This function calculates which subsystems need to be enabled if
1255 * @subtree_control is to be applied while restricted to @this_ss_mask.
1257 static u16 cgroup_calc_subtree_ss_mask(u16 subtree_control, u16 this_ss_mask)
1259 u16 cur_ss_mask = subtree_control;
1260 struct cgroup_subsys *ss;
1263 lockdep_assert_held(&cgroup_mutex);
1265 cur_ss_mask |= cgrp_dfl_implicit_ss_mask;
1268 u16 new_ss_mask = cur_ss_mask;
1270 do_each_subsys_mask(ss, ssid, cur_ss_mask) {
1271 new_ss_mask |= ss->depends_on;
1272 } while_each_subsys_mask();
1275 * Mask out subsystems which aren't available. This can
1276 * happen only if some depended-upon subsystems were bound
1277 * to non-default hierarchies.
1279 new_ss_mask &= this_ss_mask;
1281 if (new_ss_mask == cur_ss_mask)
1283 cur_ss_mask = new_ss_mask;
1290 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1291 * @kn: the kernfs_node being serviced
1293 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1294 * the method finishes if locking succeeded. Note that once this function
1295 * returns the cgroup returned by cgroup_kn_lock_live() may become
1296 * inaccessible any time. If the caller intends to continue to access the
1297 * cgroup, it should pin it before invoking this function.
1299 void cgroup_kn_unlock(struct kernfs_node *kn)
1301 struct cgroup *cgrp;
1303 if (kernfs_type(kn) == KERNFS_DIR)
1306 cgrp = kn->parent->priv;
1308 mutex_unlock(&cgroup_mutex);
1310 kernfs_unbreak_active_protection(kn);
1315 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1316 * @kn: the kernfs_node being serviced
1317 * @drain_offline: perform offline draining on the cgroup
1319 * This helper is to be used by a cgroup kernfs method currently servicing
1320 * @kn. It breaks the active protection, performs cgroup locking and
1321 * verifies that the associated cgroup is alive. Returns the cgroup if
1322 * alive; otherwise, %NULL. A successful return should be undone by a
1323 * matching cgroup_kn_unlock() invocation. If @drain_offline is %true, the
1324 * cgroup is drained of offlining csses before return.
1326 * Any cgroup kernfs method implementation which requires locking the
1327 * associated cgroup should use this helper. It avoids nesting cgroup
1328 * locking under kernfs active protection and allows all kernfs operations
1329 * including self-removal.
1331 struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline)
1333 struct cgroup *cgrp;
1335 if (kernfs_type(kn) == KERNFS_DIR)
1338 cgrp = kn->parent->priv;
1341 * We're gonna grab cgroup_mutex which nests outside kernfs
1342 * active_ref. cgroup liveliness check alone provides enough
1343 * protection against removal. Ensure @cgrp stays accessible and
1344 * break the active_ref protection.
1346 if (!cgroup_tryget(cgrp))
1348 kernfs_break_active_protection(kn);
1351 cgroup_lock_and_drain_offline(cgrp);
1353 mutex_lock(&cgroup_mutex);
1355 if (!cgroup_is_dead(cgrp))
1358 cgroup_kn_unlock(kn);
1362 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1364 char name[CGROUP_FILE_NAME_MAX];
1366 lockdep_assert_held(&cgroup_mutex);
1368 if (cft->file_offset) {
1369 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1370 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1372 spin_lock_irq(&cgroup_file_kn_lock);
1374 spin_unlock_irq(&cgroup_file_kn_lock);
1377 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1381 * css_clear_dir - remove subsys files in a cgroup directory
1384 static void css_clear_dir(struct cgroup_subsys_state *css)
1386 struct cgroup *cgrp = css->cgroup;
1387 struct cftype *cfts;
1389 if (!(css->flags & CSS_VISIBLE))
1392 css->flags &= ~CSS_VISIBLE;
1394 list_for_each_entry(cfts, &css->ss->cfts, node)
1395 cgroup_addrm_files(css, cgrp, cfts, false);
1399 * css_populate_dir - create subsys files in a cgroup directory
1402 * On failure, no file is added.
1404 static int css_populate_dir(struct cgroup_subsys_state *css)
1406 struct cgroup *cgrp = css->cgroup;
1407 struct cftype *cfts, *failed_cfts;
1410 if ((css->flags & CSS_VISIBLE) || !cgrp->kn)
1414 if (cgroup_on_dfl(cgrp))
1415 cfts = cgroup_base_files;
1417 cfts = cgroup1_base_files;
1419 return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1422 list_for_each_entry(cfts, &css->ss->cfts, node) {
1423 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1430 css->flags |= CSS_VISIBLE;
1434 list_for_each_entry(cfts, &css->ss->cfts, node) {
1435 if (cfts == failed_cfts)
1437 cgroup_addrm_files(css, cgrp, cfts, false);
1442 int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask)
1444 struct cgroup *dcgrp = &dst_root->cgrp;
1445 struct cgroup_subsys *ss;
1448 lockdep_assert_held(&cgroup_mutex);
1450 do_each_subsys_mask(ss, ssid, ss_mask) {
1452 * If @ss has non-root csses attached to it, can't move.
1453 * If @ss is an implicit controller, it is exempt from this
1454 * rule and can be stolen.
1456 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)) &&
1457 !ss->implicit_on_dfl)
1460 /* can't move between two non-dummy roots either */
1461 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1463 } while_each_subsys_mask();
1465 do_each_subsys_mask(ss, ssid, ss_mask) {
1466 struct cgroup_root *src_root = ss->root;
1467 struct cgroup *scgrp = &src_root->cgrp;
1468 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1469 struct css_set *cset;
1471 WARN_ON(!css || cgroup_css(dcgrp, ss));
1473 /* disable from the source */
1474 src_root->subsys_mask &= ~(1 << ssid);
1475 WARN_ON(cgroup_apply_control(scgrp));
1476 cgroup_finalize_control(scgrp, 0);
1479 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1480 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1481 ss->root = dst_root;
1482 css->cgroup = dcgrp;
1484 spin_lock_irq(&css_set_lock);
1485 hash_for_each(css_set_table, i, cset, hlist)
1486 list_move_tail(&cset->e_cset_node[ss->id],
1487 &dcgrp->e_csets[ss->id]);
1488 spin_unlock_irq(&css_set_lock);
1490 /* default hierarchy doesn't enable controllers by default */
1491 dst_root->subsys_mask |= 1 << ssid;
1492 if (dst_root == &cgrp_dfl_root) {
1493 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1495 dcgrp->subtree_control |= 1 << ssid;
1496 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1499 ret = cgroup_apply_control(dcgrp);
1501 pr_warn("partial failure to rebind %s controller (err=%d)\n",
1506 } while_each_subsys_mask();
1508 kernfs_activate(dcgrp->kn);
1512 int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
1513 struct kernfs_root *kf_root)
1517 struct cgroup_root *kf_cgroot = cgroup_root_from_kf(kf_root);
1518 struct cgroup *ns_cgroup;
1520 buf = kmalloc(PATH_MAX, GFP_KERNEL);
1524 spin_lock_irq(&css_set_lock);
1525 ns_cgroup = current_cgns_cgroup_from_root(kf_cgroot);
1526 len = kernfs_path_from_node(kf_node, ns_cgroup->kn, buf, PATH_MAX);
1527 spin_unlock_irq(&css_set_lock);
1529 if (len >= PATH_MAX)
1532 seq_escape(sf, buf, " \t\n\\");
1539 static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
1541 pr_err("remount is not allowed\n");
1546 * To reduce the fork() overhead for systems that are not actually using
1547 * their cgroups capability, we don't maintain the lists running through
1548 * each css_set to its tasks until we see the list actually used - in other
1549 * words after the first mount.
1551 static bool use_task_css_set_links __read_mostly;
1553 static void cgroup_enable_task_cg_lists(void)
1555 struct task_struct *p, *g;
1557 spin_lock_irq(&css_set_lock);
1559 if (use_task_css_set_links)
1562 use_task_css_set_links = true;
1565 * We need tasklist_lock because RCU is not safe against
1566 * while_each_thread(). Besides, a forking task that has passed
1567 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1568 * is not guaranteed to have its child immediately visible in the
1569 * tasklist if we walk through it with RCU.
1571 read_lock(&tasklist_lock);
1572 do_each_thread(g, p) {
1573 WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1574 task_css_set(p) != &init_css_set);
1577 * We should check if the process is exiting, otherwise
1578 * it will race with cgroup_exit() in that the list
1579 * entry won't be deleted though the process has exited.
1580 * Do it while holding siglock so that we don't end up
1581 * racing against cgroup_exit().
1583 * Interrupts were already disabled while acquiring
1584 * the css_set_lock, so we do not need to disable it
1585 * again when acquiring the sighand->siglock here.
1587 spin_lock(&p->sighand->siglock);
1588 if (!(p->flags & PF_EXITING)) {
1589 struct css_set *cset = task_css_set(p);
1591 if (!css_set_populated(cset))
1592 css_set_update_populated(cset, true);
1593 list_add_tail(&p->cg_list, &cset->tasks);
1596 spin_unlock(&p->sighand->siglock);
1597 } while_each_thread(g, p);
1598 read_unlock(&tasklist_lock);
1600 spin_unlock_irq(&css_set_lock);
1603 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1605 struct cgroup_subsys *ss;
1608 INIT_LIST_HEAD(&cgrp->self.sibling);
1609 INIT_LIST_HEAD(&cgrp->self.children);
1610 INIT_LIST_HEAD(&cgrp->cset_links);
1611 INIT_LIST_HEAD(&cgrp->pidlists);
1612 mutex_init(&cgrp->pidlist_mutex);
1613 cgrp->self.cgroup = cgrp;
1614 cgrp->self.flags |= CSS_ONLINE;
1616 for_each_subsys(ss, ssid)
1617 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1619 init_waitqueue_head(&cgrp->offline_waitq);
1620 INIT_WORK(&cgrp->release_agent_work, cgroup1_release_agent);
1623 void init_cgroup_root(struct cgroup_root *root, struct cgroup_sb_opts *opts)
1625 struct cgroup *cgrp = &root->cgrp;
1627 INIT_LIST_HEAD(&root->root_list);
1628 atomic_set(&root->nr_cgrps, 1);
1630 init_cgroup_housekeeping(cgrp);
1631 idr_init(&root->cgroup_idr);
1633 root->flags = opts->flags;
1634 if (opts->release_agent)
1635 strcpy(root->release_agent_path, opts->release_agent);
1637 strcpy(root->name, opts->name);
1638 if (opts->cpuset_clone_children)
1639 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1642 int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask)
1644 LIST_HEAD(tmp_links);
1645 struct cgroup *root_cgrp = &root->cgrp;
1646 struct kernfs_syscall_ops *kf_sops;
1647 struct css_set *cset;
1650 lockdep_assert_held(&cgroup_mutex);
1652 ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL);
1655 root_cgrp->id = ret;
1656 root_cgrp->ancestor_ids[0] = ret;
1658 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0,
1664 * We're accessing css_set_count without locking css_set_lock here,
1665 * but that's OK - it can only be increased by someone holding
1666 * cgroup_lock, and that's us. Later rebinding may disable
1667 * controllers on the default hierarchy and thus create new csets,
1668 * which can't be more than the existing ones. Allocate 2x.
1670 ret = allocate_cgrp_cset_links(2 * css_set_count, &tmp_links);
1674 ret = cgroup_init_root_id(root);
1678 kf_sops = root == &cgrp_dfl_root ?
1679 &cgroup_kf_syscall_ops : &cgroup1_kf_syscall_ops;
1681 root->kf_root = kernfs_create_root(kf_sops,
1682 KERNFS_ROOT_CREATE_DEACTIVATED,
1684 if (IS_ERR(root->kf_root)) {
1685 ret = PTR_ERR(root->kf_root);
1688 root_cgrp->kn = root->kf_root->kn;
1690 ret = css_populate_dir(&root_cgrp->self);
1694 ret = rebind_subsystems(root, ss_mask);
1698 trace_cgroup_setup_root(root);
1701 * There must be no failure case after here, since rebinding takes
1702 * care of subsystems' refcounts, which are explicitly dropped in
1703 * the failure exit path.
1705 list_add(&root->root_list, &cgroup_roots);
1706 cgroup_root_count++;
1709 * Link the root cgroup in this hierarchy into all the css_set
1712 spin_lock_irq(&css_set_lock);
1713 hash_for_each(css_set_table, i, cset, hlist) {
1714 link_css_set(&tmp_links, cset, root_cgrp);
1715 if (css_set_populated(cset))
1716 cgroup_update_populated(root_cgrp, true);
1718 spin_unlock_irq(&css_set_lock);
1720 BUG_ON(!list_empty(&root_cgrp->self.children));
1721 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
1723 kernfs_activate(root_cgrp->kn);
1728 kernfs_destroy_root(root->kf_root);
1729 root->kf_root = NULL;
1731 cgroup_exit_root_id(root);
1733 percpu_ref_exit(&root_cgrp->self.refcnt);
1735 free_cgrp_cset_links(&tmp_links);
1739 struct dentry *cgroup_do_mount(struct file_system_type *fs_type, int flags,
1740 struct cgroup_root *root, unsigned long magic,
1741 struct cgroup_namespace *ns)
1743 struct dentry *dentry;
1746 dentry = kernfs_mount(fs_type, flags, root->kf_root, magic, &new_sb);
1749 * In non-init cgroup namespace, instead of root cgroup's dentry,
1750 * we return the dentry corresponding to the cgroupns->root_cgrp.
1752 if (!IS_ERR(dentry) && ns != &init_cgroup_ns) {
1753 struct dentry *nsdentry;
1754 struct cgroup *cgrp;
1756 mutex_lock(&cgroup_mutex);
1757 spin_lock_irq(&css_set_lock);
1759 cgrp = cset_cgroup_from_root(ns->root_cset, root);
1761 spin_unlock_irq(&css_set_lock);
1762 mutex_unlock(&cgroup_mutex);
1764 nsdentry = kernfs_node_dentry(cgrp->kn, dentry->d_sb);
1769 if (IS_ERR(dentry) || !new_sb)
1770 cgroup_put(&root->cgrp);
1775 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1776 int flags, const char *unused_dev_name,
1779 struct cgroup_namespace *ns = current->nsproxy->cgroup_ns;
1780 struct dentry *dentry;
1784 /* Check if the caller has permission to mount. */
1785 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN)) {
1787 return ERR_PTR(-EPERM);
1791 * The first time anyone tries to mount a cgroup, enable the list
1792 * linking each css_set to its tasks and fix up all existing tasks.
1794 if (!use_task_css_set_links)
1795 cgroup_enable_task_cg_lists();
1797 if (fs_type == &cgroup2_fs_type) {
1799 pr_err("cgroup2: unknown option \"%s\"\n", (char *)data);
1801 return ERR_PTR(-EINVAL);
1803 cgrp_dfl_visible = true;
1804 cgroup_get(&cgrp_dfl_root.cgrp);
1806 dentry = cgroup_do_mount(&cgroup2_fs_type, flags, &cgrp_dfl_root,
1807 CGROUP2_SUPER_MAGIC, ns);
1809 dentry = cgroup1_mount(&cgroup_fs_type, flags, data,
1810 CGROUP_SUPER_MAGIC, ns);
1817 static void cgroup_kill_sb(struct super_block *sb)
1819 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
1820 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1823 * If @root doesn't have any mounts or children, start killing it.
1824 * This prevents new mounts by disabling percpu_ref_tryget_live().
1825 * cgroup_mount() may wait for @root's release.
1827 * And don't kill the default root.
1829 if (!list_empty(&root->cgrp.self.children) ||
1830 root == &cgrp_dfl_root)
1831 cgroup_put(&root->cgrp);
1833 percpu_ref_kill(&root->cgrp.self.refcnt);
1838 struct file_system_type cgroup_fs_type = {
1840 .mount = cgroup_mount,
1841 .kill_sb = cgroup_kill_sb,
1842 .fs_flags = FS_USERNS_MOUNT,
1845 static struct file_system_type cgroup2_fs_type = {
1847 .mount = cgroup_mount,
1848 .kill_sb = cgroup_kill_sb,
1849 .fs_flags = FS_USERNS_MOUNT,
1852 int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
1853 struct cgroup_namespace *ns)
1855 struct cgroup *root = cset_cgroup_from_root(ns->root_cset, cgrp->root);
1857 return kernfs_path_from_node(cgrp->kn, root->kn, buf, buflen);
1860 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen,
1861 struct cgroup_namespace *ns)
1865 mutex_lock(&cgroup_mutex);
1866 spin_lock_irq(&css_set_lock);
1868 ret = cgroup_path_ns_locked(cgrp, buf, buflen, ns);
1870 spin_unlock_irq(&css_set_lock);
1871 mutex_unlock(&cgroup_mutex);
1875 EXPORT_SYMBOL_GPL(cgroup_path_ns);
1878 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
1879 * @task: target task
1880 * @buf: the buffer to write the path into
1881 * @buflen: the length of the buffer
1883 * Determine @task's cgroup on the first (the one with the lowest non-zero
1884 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
1885 * function grabs cgroup_mutex and shouldn't be used inside locks used by
1886 * cgroup controller callbacks.
1888 * Return value is the same as kernfs_path().
1890 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
1892 struct cgroup_root *root;
1893 struct cgroup *cgrp;
1894 int hierarchy_id = 1;
1897 mutex_lock(&cgroup_mutex);
1898 spin_lock_irq(&css_set_lock);
1900 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
1903 cgrp = task_cgroup_from_root(task, root);
1904 ret = cgroup_path_ns_locked(cgrp, buf, buflen, &init_cgroup_ns);
1906 /* if no hierarchy exists, everyone is in "/" */
1907 ret = strlcpy(buf, "/", buflen);
1910 spin_unlock_irq(&css_set_lock);
1911 mutex_unlock(&cgroup_mutex);
1914 EXPORT_SYMBOL_GPL(task_cgroup_path);
1917 * cgroup_migrate_add_task - add a migration target task to a migration context
1918 * @task: target task
1919 * @mgctx: target migration context
1921 * Add @task, which is a migration target, to @mgctx->tset. This function
1922 * becomes noop if @task doesn't need to be migrated. @task's css_set
1923 * should have been added as a migration source and @task->cg_list will be
1924 * moved from the css_set's tasks list to mg_tasks one.
1926 static void cgroup_migrate_add_task(struct task_struct *task,
1927 struct cgroup_mgctx *mgctx)
1929 struct css_set *cset;
1931 lockdep_assert_held(&css_set_lock);
1933 /* @task either already exited or can't exit until the end */
1934 if (task->flags & PF_EXITING)
1937 /* leave @task alone if post_fork() hasn't linked it yet */
1938 if (list_empty(&task->cg_list))
1941 cset = task_css_set(task);
1942 if (!cset->mg_src_cgrp)
1945 list_move_tail(&task->cg_list, &cset->mg_tasks);
1946 if (list_empty(&cset->mg_node))
1947 list_add_tail(&cset->mg_node,
1948 &mgctx->tset.src_csets);
1949 if (list_empty(&cset->mg_dst_cset->mg_node))
1950 list_add_tail(&cset->mg_dst_cset->mg_node,
1951 &mgctx->tset.dst_csets);
1955 * cgroup_taskset_first - reset taskset and return the first task
1956 * @tset: taskset of interest
1957 * @dst_cssp: output variable for the destination css
1959 * @tset iteration is initialized and the first task is returned.
1961 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
1962 struct cgroup_subsys_state **dst_cssp)
1964 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
1965 tset->cur_task = NULL;
1967 return cgroup_taskset_next(tset, dst_cssp);
1971 * cgroup_taskset_next - iterate to the next task in taskset
1972 * @tset: taskset of interest
1973 * @dst_cssp: output variable for the destination css
1975 * Return the next task in @tset. Iteration must have been initialized
1976 * with cgroup_taskset_first().
1978 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
1979 struct cgroup_subsys_state **dst_cssp)
1981 struct css_set *cset = tset->cur_cset;
1982 struct task_struct *task = tset->cur_task;
1984 while (&cset->mg_node != tset->csets) {
1986 task = list_first_entry(&cset->mg_tasks,
1987 struct task_struct, cg_list);
1989 task = list_next_entry(task, cg_list);
1991 if (&task->cg_list != &cset->mg_tasks) {
1992 tset->cur_cset = cset;
1993 tset->cur_task = task;
1996 * This function may be called both before and
1997 * after cgroup_taskset_migrate(). The two cases
1998 * can be distinguished by looking at whether @cset
1999 * has its ->mg_dst_cset set.
2001 if (cset->mg_dst_cset)
2002 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2004 *dst_cssp = cset->subsys[tset->ssid];
2009 cset = list_next_entry(cset, mg_node);
2017 * cgroup_taskset_migrate - migrate a taskset
2018 * @mgctx: migration context
2020 * Migrate tasks in @mgctx as setup by migration preparation functions.
2021 * This function fails iff one of the ->can_attach callbacks fails and
2022 * guarantees that either all or none of the tasks in @mgctx are migrated.
2023 * @mgctx is consumed regardless of success.
2025 static int cgroup_migrate_execute(struct cgroup_mgctx *mgctx)
2027 struct cgroup_taskset *tset = &mgctx->tset;
2028 struct cgroup_subsys *ss;
2029 struct task_struct *task, *tmp_task;
2030 struct css_set *cset, *tmp_cset;
2031 int ssid, failed_ssid, ret;
2033 /* methods shouldn't be called if no task is actually migrating */
2034 if (list_empty(&tset->src_csets))
2037 /* check that we can legitimately attach to the cgroup */
2038 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2039 if (ss->can_attach) {
2041 ret = ss->can_attach(tset);
2044 goto out_cancel_attach;
2047 } while_each_subsys_mask();
2050 * Now that we're guaranteed success, proceed to move all tasks to
2051 * the new cgroup. There are no failure cases after here, so this
2052 * is the commit point.
2054 spin_lock_irq(&css_set_lock);
2055 list_for_each_entry(cset, &tset->src_csets, mg_node) {
2056 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2057 struct css_set *from_cset = task_css_set(task);
2058 struct css_set *to_cset = cset->mg_dst_cset;
2060 get_css_set(to_cset);
2061 css_set_move_task(task, from_cset, to_cset, true);
2062 put_css_set_locked(from_cset);
2065 spin_unlock_irq(&css_set_lock);
2068 * Migration is committed, all target tasks are now on dst_csets.
2069 * Nothing is sensitive to fork() after this point. Notify
2070 * controllers that migration is complete.
2072 tset->csets = &tset->dst_csets;
2074 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2079 } while_each_subsys_mask();
2082 goto out_release_tset;
2085 do_each_subsys_mask(ss, ssid, mgctx->ss_mask) {
2086 if (ssid == failed_ssid)
2088 if (ss->cancel_attach) {
2090 ss->cancel_attach(tset);
2092 } while_each_subsys_mask();
2094 spin_lock_irq(&css_set_lock);
2095 list_splice_init(&tset->dst_csets, &tset->src_csets);
2096 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2097 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2098 list_del_init(&cset->mg_node);
2100 spin_unlock_irq(&css_set_lock);
2105 * cgroup_may_migrate_to - verify whether a cgroup can be migration destination
2106 * @dst_cgrp: destination cgroup to test
2108 * On the default hierarchy, except for the root, subtree_control must be
2109 * zero for migration destination cgroups with tasks so that child cgroups
2110 * don't compete against tasks.
2112 bool cgroup_may_migrate_to(struct cgroup *dst_cgrp)
2114 return !cgroup_on_dfl(dst_cgrp) || !cgroup_parent(dst_cgrp) ||
2115 !dst_cgrp->subtree_control;
2119 * cgroup_migrate_finish - cleanup after attach
2120 * @mgctx: migration context
2122 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2123 * those functions for details.
2125 void cgroup_migrate_finish(struct cgroup_mgctx *mgctx)
2127 LIST_HEAD(preloaded);
2128 struct css_set *cset, *tmp_cset;
2130 lockdep_assert_held(&cgroup_mutex);
2132 spin_lock_irq(&css_set_lock);
2134 list_splice_tail_init(&mgctx->preloaded_src_csets, &preloaded);
2135 list_splice_tail_init(&mgctx->preloaded_dst_csets, &preloaded);
2137 list_for_each_entry_safe(cset, tmp_cset, &preloaded, mg_preload_node) {
2138 cset->mg_src_cgrp = NULL;
2139 cset->mg_dst_cgrp = NULL;
2140 cset->mg_dst_cset = NULL;
2141 list_del_init(&cset->mg_preload_node);
2142 put_css_set_locked(cset);
2145 spin_unlock_irq(&css_set_lock);
2149 * cgroup_migrate_add_src - add a migration source css_set
2150 * @src_cset: the source css_set to add
2151 * @dst_cgrp: the destination cgroup
2152 * @mgctx: migration context
2154 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2155 * @src_cset and add it to @mgctx->src_csets, which should later be cleaned
2156 * up by cgroup_migrate_finish().
2158 * This function may be called without holding cgroup_threadgroup_rwsem
2159 * even if the target is a process. Threads may be created and destroyed
2160 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2161 * into play and the preloaded css_sets are guaranteed to cover all
2164 void cgroup_migrate_add_src(struct css_set *src_cset,
2165 struct cgroup *dst_cgrp,
2166 struct cgroup_mgctx *mgctx)
2168 struct cgroup *src_cgrp;
2170 lockdep_assert_held(&cgroup_mutex);
2171 lockdep_assert_held(&css_set_lock);
2174 * If ->dead, @src_set is associated with one or more dead cgroups
2175 * and doesn't contain any migratable tasks. Ignore it early so
2176 * that the rest of migration path doesn't get confused by it.
2181 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2183 if (!list_empty(&src_cset->mg_preload_node))
2186 WARN_ON(src_cset->mg_src_cgrp);
2187 WARN_ON(src_cset->mg_dst_cgrp);
2188 WARN_ON(!list_empty(&src_cset->mg_tasks));
2189 WARN_ON(!list_empty(&src_cset->mg_node));
2191 src_cset->mg_src_cgrp = src_cgrp;
2192 src_cset->mg_dst_cgrp = dst_cgrp;
2193 get_css_set(src_cset);
2194 list_add_tail(&src_cset->mg_preload_node, &mgctx->preloaded_src_csets);
2198 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2199 * @mgctx: migration context
2201 * Tasks are about to be moved and all the source css_sets have been
2202 * preloaded to @mgctx->preloaded_src_csets. This function looks up and
2203 * pins all destination css_sets, links each to its source, and append them
2204 * to @mgctx->preloaded_dst_csets.
2206 * This function must be called after cgroup_migrate_add_src() has been
2207 * called on each migration source css_set. After migration is performed
2208 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2211 int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx)
2213 struct css_set *src_cset, *tmp_cset;
2215 lockdep_assert_held(&cgroup_mutex);
2217 /* look up the dst cset for each src cset and link it to src */
2218 list_for_each_entry_safe(src_cset, tmp_cset, &mgctx->preloaded_src_csets,
2220 struct css_set *dst_cset;
2221 struct cgroup_subsys *ss;
2224 dst_cset = find_css_set(src_cset, src_cset->mg_dst_cgrp);
2228 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2231 * If src cset equals dst, it's noop. Drop the src.
2232 * cgroup_migrate() will skip the cset too. Note that we
2233 * can't handle src == dst as some nodes are used by both.
2235 if (src_cset == dst_cset) {
2236 src_cset->mg_src_cgrp = NULL;
2237 src_cset->mg_dst_cgrp = NULL;
2238 list_del_init(&src_cset->mg_preload_node);
2239 put_css_set(src_cset);
2240 put_css_set(dst_cset);
2244 src_cset->mg_dst_cset = dst_cset;
2246 if (list_empty(&dst_cset->mg_preload_node))
2247 list_add_tail(&dst_cset->mg_preload_node,
2248 &mgctx->preloaded_dst_csets);
2250 put_css_set(dst_cset);
2252 for_each_subsys(ss, ssid)
2253 if (src_cset->subsys[ssid] != dst_cset->subsys[ssid])
2254 mgctx->ss_mask |= 1 << ssid;
2259 cgroup_migrate_finish(mgctx);
2264 * cgroup_migrate - migrate a process or task to a cgroup
2265 * @leader: the leader of the process or the task to migrate
2266 * @threadgroup: whether @leader points to the whole process or a single task
2267 * @mgctx: migration context
2269 * Migrate a process or task denoted by @leader. If migrating a process,
2270 * the caller must be holding cgroup_threadgroup_rwsem. The caller is also
2271 * responsible for invoking cgroup_migrate_add_src() and
2272 * cgroup_migrate_prepare_dst() on the targets before invoking this
2273 * function and following up with cgroup_migrate_finish().
2275 * As long as a controller's ->can_attach() doesn't fail, this function is
2276 * guaranteed to succeed. This means that, excluding ->can_attach()
2277 * failure, when migrating multiple targets, the success or failure can be
2278 * decided for all targets by invoking group_migrate_prepare_dst() before
2279 * actually starting migrating.
2281 int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2282 struct cgroup_mgctx *mgctx)
2284 struct task_struct *task;
2287 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2288 * already PF_EXITING could be freed from underneath us unless we
2289 * take an rcu_read_lock.
2291 spin_lock_irq(&css_set_lock);
2295 cgroup_migrate_add_task(task, mgctx);
2298 } while_each_thread(leader, task);
2300 spin_unlock_irq(&css_set_lock);
2302 return cgroup_migrate_execute(mgctx);
2306 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2307 * @dst_cgrp: the cgroup to attach to
2308 * @leader: the task or the leader of the threadgroup to be attached
2309 * @threadgroup: attach the whole threadgroup?
2311 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2313 int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
2316 DEFINE_CGROUP_MGCTX(mgctx);
2317 struct task_struct *task;
2320 if (!cgroup_may_migrate_to(dst_cgrp))
2323 /* look up all src csets */
2324 spin_lock_irq(&css_set_lock);
2328 cgroup_migrate_add_src(task_css_set(task), dst_cgrp, &mgctx);
2331 } while_each_thread(leader, task);
2333 spin_unlock_irq(&css_set_lock);
2335 /* prepare dst csets and commit */
2336 ret = cgroup_migrate_prepare_dst(&mgctx);
2338 ret = cgroup_migrate(leader, threadgroup, &mgctx);
2340 cgroup_migrate_finish(&mgctx);
2343 trace_cgroup_attach_task(dst_cgrp, leader, threadgroup);
2348 static int cgroup_procs_write_permission(struct task_struct *task,
2349 struct cgroup *dst_cgrp,
2350 struct kernfs_open_file *of)
2354 if (cgroup_on_dfl(dst_cgrp)) {
2355 struct super_block *sb = of->file->f_path.dentry->d_sb;
2356 struct cgroup *cgrp;
2357 struct inode *inode;
2359 spin_lock_irq(&css_set_lock);
2360 cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
2361 spin_unlock_irq(&css_set_lock);
2363 while (!cgroup_is_descendant(dst_cgrp, cgrp))
2364 cgrp = cgroup_parent(cgrp);
2367 inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
2369 ret = inode_permission(inode, MAY_WRITE);
2373 const struct cred *cred = current_cred();
2374 const struct cred *tcred = get_task_cred(task);
2377 * even if we're attaching all tasks in the thread group,
2378 * we only need to check permissions on one of them.
2380 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2381 !uid_eq(cred->euid, tcred->uid) &&
2382 !uid_eq(cred->euid, tcred->suid))
2391 * Find the task_struct of the task to attach by vpid and pass it along to the
2392 * function to attach either it or all tasks in its threadgroup. Will lock
2393 * cgroup_mutex and threadgroup.
2395 ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
2396 size_t nbytes, loff_t off, bool threadgroup)
2398 struct task_struct *tsk;
2399 struct cgroup_subsys *ss;
2400 struct cgroup *cgrp;
2404 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2407 cgrp = cgroup_kn_lock_live(of->kn, false);
2411 percpu_down_write(&cgroup_threadgroup_rwsem);
2414 tsk = find_task_by_vpid(pid);
2417 goto out_unlock_rcu;
2424 tsk = tsk->group_leader;
2427 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2428 * trapped in a cpuset, or RT worker may be born in a cgroup
2429 * with no rt_runtime allocated. Just say no.
2431 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
2433 goto out_unlock_rcu;
2436 get_task_struct(tsk);
2439 ret = cgroup_procs_write_permission(tsk, cgrp, of);
2441 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2443 put_task_struct(tsk);
2444 goto out_unlock_threadgroup;
2448 out_unlock_threadgroup:
2449 percpu_up_write(&cgroup_threadgroup_rwsem);
2450 for_each_subsys(ss, ssid)
2451 if (ss->post_attach)
2453 cgroup_kn_unlock(of->kn);
2454 return ret ?: nbytes;
2457 ssize_t cgroup_procs_write(struct kernfs_open_file *of, char *buf, size_t nbytes,
2460 return __cgroup_procs_write(of, buf, nbytes, off, true);
2463 static void cgroup_print_ss_mask(struct seq_file *seq, u16 ss_mask)
2465 struct cgroup_subsys *ss;
2466 bool printed = false;
2469 do_each_subsys_mask(ss, ssid, ss_mask) {
2472 seq_printf(seq, "%s", ss->name);
2474 } while_each_subsys_mask();
2476 seq_putc(seq, '\n');
2479 /* show controllers which are enabled from the parent */
2480 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2482 struct cgroup *cgrp = seq_css(seq)->cgroup;
2484 cgroup_print_ss_mask(seq, cgroup_control(cgrp));
2488 /* show controllers which are enabled for a given cgroup's children */
2489 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2491 struct cgroup *cgrp = seq_css(seq)->cgroup;
2493 cgroup_print_ss_mask(seq, cgrp->subtree_control);
2498 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2499 * @cgrp: root of the subtree to update csses for
2501 * @cgrp's control masks have changed and its subtree's css associations
2502 * need to be updated accordingly. This function looks up all css_sets
2503 * which are attached to the subtree, creates the matching updated css_sets
2504 * and migrates the tasks to the new ones.
2506 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2508 DEFINE_CGROUP_MGCTX(mgctx);
2509 struct cgroup_subsys_state *d_css;
2510 struct cgroup *dsct;
2511 struct css_set *src_cset;
2514 lockdep_assert_held(&cgroup_mutex);
2516 percpu_down_write(&cgroup_threadgroup_rwsem);
2518 /* look up all csses currently attached to @cgrp's subtree */
2519 spin_lock_irq(&css_set_lock);
2520 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2521 struct cgrp_cset_link *link;
2523 list_for_each_entry(link, &dsct->cset_links, cset_link)
2524 cgroup_migrate_add_src(link->cset, dsct, &mgctx);
2526 spin_unlock_irq(&css_set_lock);
2528 /* NULL dst indicates self on default hierarchy */
2529 ret = cgroup_migrate_prepare_dst(&mgctx);
2533 spin_lock_irq(&css_set_lock);
2534 list_for_each_entry(src_cset, &mgctx.preloaded_src_csets, mg_preload_node) {
2535 struct task_struct *task, *ntask;
2537 /* all tasks in src_csets need to be migrated */
2538 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2539 cgroup_migrate_add_task(task, &mgctx);
2541 spin_unlock_irq(&css_set_lock);
2543 ret = cgroup_migrate_execute(&mgctx);
2545 cgroup_migrate_finish(&mgctx);
2546 percpu_up_write(&cgroup_threadgroup_rwsem);
2551 * cgroup_lock_and_drain_offline - lock cgroup_mutex and drain offlined csses
2552 * @cgrp: root of the target subtree
2554 * Because css offlining is asynchronous, userland may try to re-enable a
2555 * controller while the previous css is still around. This function grabs
2556 * cgroup_mutex and drains the previous css instances of @cgrp's subtree.
2558 void cgroup_lock_and_drain_offline(struct cgroup *cgrp)
2559 __acquires(&cgroup_mutex)
2561 struct cgroup *dsct;
2562 struct cgroup_subsys_state *d_css;
2563 struct cgroup_subsys *ss;
2567 mutex_lock(&cgroup_mutex);
2569 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2570 for_each_subsys(ss, ssid) {
2571 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2574 if (!css || !percpu_ref_is_dying(&css->refcnt))
2578 prepare_to_wait(&dsct->offline_waitq, &wait,
2579 TASK_UNINTERRUPTIBLE);
2581 mutex_unlock(&cgroup_mutex);
2583 finish_wait(&dsct->offline_waitq, &wait);
2592 * cgroup_save_control - save control masks of a subtree
2593 * @cgrp: root of the target subtree
2595 * Save ->subtree_control and ->subtree_ss_mask to the respective old_
2596 * prefixed fields for @cgrp's subtree including @cgrp itself.
2598 static void cgroup_save_control(struct cgroup *cgrp)
2600 struct cgroup *dsct;
2601 struct cgroup_subsys_state *d_css;
2603 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2604 dsct->old_subtree_control = dsct->subtree_control;
2605 dsct->old_subtree_ss_mask = dsct->subtree_ss_mask;
2610 * cgroup_propagate_control - refresh control masks of a subtree
2611 * @cgrp: root of the target subtree
2613 * For @cgrp and its subtree, ensure ->subtree_ss_mask matches
2614 * ->subtree_control and propagate controller availability through the
2615 * subtree so that descendants don't have unavailable controllers enabled.
2617 static void cgroup_propagate_control(struct cgroup *cgrp)
2619 struct cgroup *dsct;
2620 struct cgroup_subsys_state *d_css;
2622 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2623 dsct->subtree_control &= cgroup_control(dsct);
2624 dsct->subtree_ss_mask =
2625 cgroup_calc_subtree_ss_mask(dsct->subtree_control,
2626 cgroup_ss_mask(dsct));
2631 * cgroup_restore_control - restore control masks of a subtree
2632 * @cgrp: root of the target subtree
2634 * Restore ->subtree_control and ->subtree_ss_mask from the respective old_
2635 * prefixed fields for @cgrp's subtree including @cgrp itself.
2637 static void cgroup_restore_control(struct cgroup *cgrp)
2639 struct cgroup *dsct;
2640 struct cgroup_subsys_state *d_css;
2642 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2643 dsct->subtree_control = dsct->old_subtree_control;
2644 dsct->subtree_ss_mask = dsct->old_subtree_ss_mask;
2648 static bool css_visible(struct cgroup_subsys_state *css)
2650 struct cgroup_subsys *ss = css->ss;
2651 struct cgroup *cgrp = css->cgroup;
2653 if (cgroup_control(cgrp) & (1 << ss->id))
2655 if (!(cgroup_ss_mask(cgrp) & (1 << ss->id)))
2657 return cgroup_on_dfl(cgrp) && ss->implicit_on_dfl;
2661 * cgroup_apply_control_enable - enable or show csses according to control
2662 * @cgrp: root of the target subtree
2664 * Walk @cgrp's subtree and create new csses or make the existing ones
2665 * visible. A css is created invisible if it's being implicitly enabled
2666 * through dependency. An invisible css is made visible when the userland
2667 * explicitly enables it.
2669 * Returns 0 on success, -errno on failure. On failure, csses which have
2670 * been processed already aren't cleaned up. The caller is responsible for
2671 * cleaning up with cgroup_apply_control_disble().
2673 static int cgroup_apply_control_enable(struct cgroup *cgrp)
2675 struct cgroup *dsct;
2676 struct cgroup_subsys_state *d_css;
2677 struct cgroup_subsys *ss;
2680 cgroup_for_each_live_descendant_pre(dsct, d_css, cgrp) {
2681 for_each_subsys(ss, ssid) {
2682 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2684 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
2686 if (!(cgroup_ss_mask(dsct) & (1 << ss->id)))
2690 css = css_create(dsct, ss);
2692 return PTR_ERR(css);
2695 if (css_visible(css)) {
2696 ret = css_populate_dir(css);
2707 * cgroup_apply_control_disable - kill or hide csses according to control
2708 * @cgrp: root of the target subtree
2710 * Walk @cgrp's subtree and kill and hide csses so that they match
2711 * cgroup_ss_mask() and cgroup_visible_mask().
2713 * A css is hidden when the userland requests it to be disabled while other
2714 * subsystems are still depending on it. The css must not actively control
2715 * resources and be in the vanilla state if it's made visible again later.
2716 * Controllers which may be depended upon should provide ->css_reset() for
2719 static void cgroup_apply_control_disable(struct cgroup *cgrp)
2721 struct cgroup *dsct;
2722 struct cgroup_subsys_state *d_css;
2723 struct cgroup_subsys *ss;
2726 cgroup_for_each_live_descendant_post(dsct, d_css, cgrp) {
2727 for_each_subsys(ss, ssid) {
2728 struct cgroup_subsys_state *css = cgroup_css(dsct, ss);
2730 WARN_ON_ONCE(css && percpu_ref_is_dying(&css->refcnt));
2736 !(cgroup_ss_mask(dsct) & (1 << ss->id))) {
2738 } else if (!css_visible(css)) {
2748 * cgroup_apply_control - apply control mask updates to the subtree
2749 * @cgrp: root of the target subtree
2751 * subsystems can be enabled and disabled in a subtree using the following
2754 * 1. Call cgroup_save_control() to stash the current state.
2755 * 2. Update ->subtree_control masks in the subtree as desired.
2756 * 3. Call cgroup_apply_control() to apply the changes.
2757 * 4. Optionally perform other related operations.
2758 * 5. Call cgroup_finalize_control() to finish up.
2760 * This function implements step 3 and propagates the mask changes
2761 * throughout @cgrp's subtree, updates csses accordingly and perform
2762 * process migrations.
2764 static int cgroup_apply_control(struct cgroup *cgrp)
2768 cgroup_propagate_control(cgrp);
2770 ret = cgroup_apply_control_enable(cgrp);
2775 * At this point, cgroup_e_css() results reflect the new csses
2776 * making the following cgroup_update_dfl_csses() properly update
2777 * css associations of all tasks in the subtree.
2779 ret = cgroup_update_dfl_csses(cgrp);
2787 * cgroup_finalize_control - finalize control mask update
2788 * @cgrp: root of the target subtree
2789 * @ret: the result of the update
2791 * Finalize control mask update. See cgroup_apply_control() for more info.
2793 static void cgroup_finalize_control(struct cgroup *cgrp, int ret)
2796 cgroup_restore_control(cgrp);
2797 cgroup_propagate_control(cgrp);
2800 cgroup_apply_control_disable(cgrp);
2803 /* change the enabled child controllers for a cgroup in the default hierarchy */
2804 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
2805 char *buf, size_t nbytes,
2808 u16 enable = 0, disable = 0;
2809 struct cgroup *cgrp, *child;
2810 struct cgroup_subsys *ss;
2815 * Parse input - space separated list of subsystem names prefixed
2816 * with either + or -.
2818 buf = strstrip(buf);
2819 while ((tok = strsep(&buf, " "))) {
2822 do_each_subsys_mask(ss, ssid, ~cgrp_dfl_inhibit_ss_mask) {
2823 if (!cgroup_ssid_enabled(ssid) ||
2824 strcmp(tok + 1, ss->name))
2828 enable |= 1 << ssid;
2829 disable &= ~(1 << ssid);
2830 } else if (*tok == '-') {
2831 disable |= 1 << ssid;
2832 enable &= ~(1 << ssid);
2837 } while_each_subsys_mask();
2838 if (ssid == CGROUP_SUBSYS_COUNT)
2842 cgrp = cgroup_kn_lock_live(of->kn, true);
2846 for_each_subsys(ss, ssid) {
2847 if (enable & (1 << ssid)) {
2848 if (cgrp->subtree_control & (1 << ssid)) {
2849 enable &= ~(1 << ssid);
2853 if (!(cgroup_control(cgrp) & (1 << ssid))) {
2857 } else if (disable & (1 << ssid)) {
2858 if (!(cgrp->subtree_control & (1 << ssid))) {
2859 disable &= ~(1 << ssid);
2863 /* a child has it enabled? */
2864 cgroup_for_each_live_child(child, cgrp) {
2865 if (child->subtree_control & (1 << ssid)) {
2873 if (!enable && !disable) {
2879 * Except for the root, subtree_control must be zero for a cgroup
2880 * with tasks so that child cgroups don't compete against tasks.
2882 if (enable && cgroup_parent(cgrp)) {
2883 struct cgrp_cset_link *link;
2886 * Because namespaces pin csets too, @cgrp->cset_links
2887 * might not be empty even when @cgrp is empty. Walk and
2890 spin_lock_irq(&css_set_lock);
2893 list_for_each_entry(link, &cgrp->cset_links, cset_link) {
2894 if (css_set_populated(link->cset)) {
2900 spin_unlock_irq(&css_set_lock);
2906 /* save and update control masks and prepare csses */
2907 cgroup_save_control(cgrp);
2909 cgrp->subtree_control |= enable;
2910 cgrp->subtree_control &= ~disable;
2912 ret = cgroup_apply_control(cgrp);
2914 cgroup_finalize_control(cgrp, ret);
2916 kernfs_activate(cgrp->kn);
2919 cgroup_kn_unlock(of->kn);
2920 return ret ?: nbytes;
2923 static int cgroup_events_show(struct seq_file *seq, void *v)
2925 seq_printf(seq, "populated %d\n",
2926 cgroup_is_populated(seq_css(seq)->cgroup));
2930 static int cgroup_file_open(struct kernfs_open_file *of)
2932 struct cftype *cft = of->kn->priv;
2935 return cft->open(of);
2939 static void cgroup_file_release(struct kernfs_open_file *of)
2941 struct cftype *cft = of->kn->priv;
2947 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
2948 size_t nbytes, loff_t off)
2950 struct cgroup *cgrp = of->kn->parent->priv;
2951 struct cftype *cft = of->kn->priv;
2952 struct cgroup_subsys_state *css;
2956 return cft->write(of, buf, nbytes, off);
2959 * kernfs guarantees that a file isn't deleted with operations in
2960 * flight, which means that the matching css is and stays alive and
2961 * doesn't need to be pinned. The RCU locking is not necessary
2962 * either. It's just for the convenience of using cgroup_css().
2965 css = cgroup_css(cgrp, cft->ss);
2968 if (cft->write_u64) {
2969 unsigned long long v;
2970 ret = kstrtoull(buf, 0, &v);
2972 ret = cft->write_u64(css, cft, v);
2973 } else if (cft->write_s64) {
2975 ret = kstrtoll(buf, 0, &v);
2977 ret = cft->write_s64(css, cft, v);
2982 return ret ?: nbytes;
2985 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
2987 return seq_cft(seq)->seq_start(seq, ppos);
2990 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
2992 return seq_cft(seq)->seq_next(seq, v, ppos);
2995 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
2997 if (seq_cft(seq)->seq_stop)
2998 seq_cft(seq)->seq_stop(seq, v);
3001 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3003 struct cftype *cft = seq_cft(m);
3004 struct cgroup_subsys_state *css = seq_css(m);
3007 return cft->seq_show(m, arg);
3010 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3011 else if (cft->read_s64)
3012 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3018 static struct kernfs_ops cgroup_kf_single_ops = {
3019 .atomic_write_len = PAGE_SIZE,
3020 .open = cgroup_file_open,
3021 .release = cgroup_file_release,
3022 .write = cgroup_file_write,
3023 .seq_show = cgroup_seqfile_show,
3026 static struct kernfs_ops cgroup_kf_ops = {
3027 .atomic_write_len = PAGE_SIZE,
3028 .open = cgroup_file_open,
3029 .release = cgroup_file_release,
3030 .write = cgroup_file_write,
3031 .seq_start = cgroup_seqfile_start,
3032 .seq_next = cgroup_seqfile_next,
3033 .seq_stop = cgroup_seqfile_stop,
3034 .seq_show = cgroup_seqfile_show,
3037 /* set uid and gid of cgroup dirs and files to that of the creator */
3038 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3040 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3041 .ia_uid = current_fsuid(),
3042 .ia_gid = current_fsgid(), };
3044 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3045 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3048 return kernfs_setattr(kn, &iattr);
3051 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3054 char name[CGROUP_FILE_NAME_MAX];
3055 struct kernfs_node *kn;
3056 struct lock_class_key *key = NULL;
3059 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3060 key = &cft->lockdep_key;
3062 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3063 cgroup_file_mode(cft), 0, cft->kf_ops, cft,
3068 ret = cgroup_kn_set_ugid(kn);
3074 if (cft->file_offset) {
3075 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3077 spin_lock_irq(&cgroup_file_kn_lock);
3079 spin_unlock_irq(&cgroup_file_kn_lock);
3086 * cgroup_addrm_files - add or remove files to a cgroup directory
3087 * @css: the target css
3088 * @cgrp: the target cgroup (usually css->cgroup)
3089 * @cfts: array of cftypes to be added
3090 * @is_add: whether to add or remove
3092 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3093 * For removals, this function never fails.
3095 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3096 struct cgroup *cgrp, struct cftype cfts[],
3099 struct cftype *cft, *cft_end = NULL;
3102 lockdep_assert_held(&cgroup_mutex);
3105 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3106 /* does cft->flags tell us to skip this file on @cgrp? */
3107 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3109 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3111 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3113 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3117 ret = cgroup_add_file(css, cgrp, cft);
3119 pr_warn("%s: failed to add %s, err=%d\n",
3120 __func__, cft->name, ret);
3126 cgroup_rm_file(cgrp, cft);
3132 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3135 struct cgroup_subsys *ss = cfts[0].ss;
3136 struct cgroup *root = &ss->root->cgrp;
3137 struct cgroup_subsys_state *css;
3140 lockdep_assert_held(&cgroup_mutex);
3142 /* add/rm files for all cgroups created before */
3143 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3144 struct cgroup *cgrp = css->cgroup;
3146 if (!(css->flags & CSS_VISIBLE))
3149 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3155 kernfs_activate(root->kn);
3159 static void cgroup_exit_cftypes(struct cftype *cfts)
3163 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3164 /* free copy for custom atomic_write_len, see init_cftypes() */
3165 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3170 /* revert flags set by cgroup core while adding @cfts */
3171 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
3175 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3179 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3180 struct kernfs_ops *kf_ops;
3182 WARN_ON(cft->ss || cft->kf_ops);
3185 kf_ops = &cgroup_kf_ops;
3187 kf_ops = &cgroup_kf_single_ops;
3190 * Ugh... if @cft wants a custom max_write_len, we need to
3191 * make a copy of kf_ops to set its atomic_write_len.
3193 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3194 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3196 cgroup_exit_cftypes(cfts);
3199 kf_ops->atomic_write_len = cft->max_write_len;
3202 cft->kf_ops = kf_ops;
3209 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3211 lockdep_assert_held(&cgroup_mutex);
3213 if (!cfts || !cfts[0].ss)
3216 list_del(&cfts->node);
3217 cgroup_apply_cftypes(cfts, false);
3218 cgroup_exit_cftypes(cfts);
3223 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3224 * @cfts: zero-length name terminated array of cftypes
3226 * Unregister @cfts. Files described by @cfts are removed from all
3227 * existing cgroups and all future cgroups won't have them either. This
3228 * function can be called anytime whether @cfts' subsys is attached or not.
3230 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3233 int cgroup_rm_cftypes(struct cftype *cfts)
3237 mutex_lock(&cgroup_mutex);
3238 ret = cgroup_rm_cftypes_locked(cfts);
3239 mutex_unlock(&cgroup_mutex);
3244 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3245 * @ss: target cgroup subsystem
3246 * @cfts: zero-length name terminated array of cftypes
3248 * Register @cfts to @ss. Files described by @cfts are created for all
3249 * existing cgroups to which @ss is attached and all future cgroups will
3250 * have them too. This function can be called anytime whether @ss is
3253 * Returns 0 on successful registration, -errno on failure. Note that this
3254 * function currently returns 0 as long as @cfts registration is successful
3255 * even if some file creation attempts on existing cgroups fail.
3257 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3261 if (!cgroup_ssid_enabled(ss->id))
3264 if (!cfts || cfts[0].name[0] == '\0')
3267 ret = cgroup_init_cftypes(ss, cfts);
3271 mutex_lock(&cgroup_mutex);
3273 list_add_tail(&cfts->node, &ss->cfts);
3274 ret = cgroup_apply_cftypes(cfts, true);
3276 cgroup_rm_cftypes_locked(cfts);
3278 mutex_unlock(&cgroup_mutex);
3283 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3284 * @ss: target cgroup subsystem
3285 * @cfts: zero-length name terminated array of cftypes
3287 * Similar to cgroup_add_cftypes() but the added files are only used for
3288 * the default hierarchy.
3290 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3294 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3295 cft->flags |= __CFTYPE_ONLY_ON_DFL;
3296 return cgroup_add_cftypes(ss, cfts);
3300 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3301 * @ss: target cgroup subsystem
3302 * @cfts: zero-length name terminated array of cftypes
3304 * Similar to cgroup_add_cftypes() but the added files are only used for
3305 * the legacy hierarchies.
3307 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3311 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3312 cft->flags |= __CFTYPE_NOT_ON_DFL;
3313 return cgroup_add_cftypes(ss, cfts);
3317 * cgroup_file_notify - generate a file modified event for a cgroup_file
3318 * @cfile: target cgroup_file
3320 * @cfile must have been obtained by setting cftype->file_offset.
3322 void cgroup_file_notify(struct cgroup_file *cfile)
3324 unsigned long flags;
3326 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
3328 kernfs_notify(cfile->kn);
3329 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
3333 * css_next_child - find the next child of a given css
3334 * @pos: the current position (%NULL to initiate traversal)
3335 * @parent: css whose children to walk
3337 * This function returns the next child of @parent and should be called
3338 * under either cgroup_mutex or RCU read lock. The only requirement is
3339 * that @parent and @pos are accessible. The next sibling is guaranteed to
3340 * be returned regardless of their states.
3342 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3343 * css which finished ->css_online() is guaranteed to be visible in the
3344 * future iterations and will stay visible until the last reference is put.
3345 * A css which hasn't finished ->css_online() or already finished
3346 * ->css_offline() may show up during traversal. It's each subsystem's
3347 * responsibility to synchronize against on/offlining.
3349 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
3350 struct cgroup_subsys_state *parent)
3352 struct cgroup_subsys_state *next;
3354 cgroup_assert_mutex_or_rcu_locked();
3357 * @pos could already have been unlinked from the sibling list.
3358 * Once a cgroup is removed, its ->sibling.next is no longer
3359 * updated when its next sibling changes. CSS_RELEASED is set when
3360 * @pos is taken off list, at which time its next pointer is valid,
3361 * and, as releases are serialized, the one pointed to by the next
3362 * pointer is guaranteed to not have started release yet. This
3363 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3364 * critical section, the one pointed to by its next pointer is
3365 * guaranteed to not have finished its RCU grace period even if we
3366 * have dropped rcu_read_lock() inbetween iterations.
3368 * If @pos has CSS_RELEASED set, its next pointer can't be
3369 * dereferenced; however, as each css is given a monotonically
3370 * increasing unique serial number and always appended to the
3371 * sibling list, the next one can be found by walking the parent's
3372 * children until the first css with higher serial number than
3373 * @pos's. While this path can be slower, it happens iff iteration
3374 * races against release and the race window is very small.
3377 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
3378 } else if (likely(!(pos->flags & CSS_RELEASED))) {
3379 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
3381 list_for_each_entry_rcu(next, &parent->children, sibling)
3382 if (next->serial_nr > pos->serial_nr)
3387 * @next, if not pointing to the head, can be dereferenced and is
3390 if (&next->sibling != &parent->children)
3396 * css_next_descendant_pre - find the next descendant for pre-order walk
3397 * @pos: the current position (%NULL to initiate traversal)
3398 * @root: css whose descendants to walk
3400 * To be used by css_for_each_descendant_pre(). Find the next descendant
3401 * to visit for pre-order traversal of @root's descendants. @root is
3402 * included in the iteration and the first node to be visited.
3404 * While this function requires cgroup_mutex or RCU read locking, it
3405 * doesn't require the whole traversal to be contained in a single critical
3406 * section. This function will return the correct next descendant as long
3407 * as both @pos and @root are accessible and @pos is a descendant of @root.
3409 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3410 * css which finished ->css_online() is guaranteed to be visible in the
3411 * future iterations and will stay visible until the last reference is put.
3412 * A css which hasn't finished ->css_online() or already finished
3413 * ->css_offline() may show up during traversal. It's each subsystem's
3414 * responsibility to synchronize against on/offlining.
3416 struct cgroup_subsys_state *
3417 css_next_descendant_pre(struct cgroup_subsys_state *pos,
3418 struct cgroup_subsys_state *root)
3420 struct cgroup_subsys_state *next;
3422 cgroup_assert_mutex_or_rcu_locked();
3424 /* if first iteration, visit @root */
3428 /* visit the first child if exists */
3429 next = css_next_child(NULL, pos);
3433 /* no child, visit my or the closest ancestor's next sibling */
3434 while (pos != root) {
3435 next = css_next_child(pos, pos->parent);
3445 * css_rightmost_descendant - return the rightmost descendant of a css
3446 * @pos: css of interest
3448 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3449 * is returned. This can be used during pre-order traversal to skip
3452 * While this function requires cgroup_mutex or RCU read locking, it
3453 * doesn't require the whole traversal to be contained in a single critical
3454 * section. This function will return the correct rightmost descendant as
3455 * long as @pos is accessible.
3457 struct cgroup_subsys_state *
3458 css_rightmost_descendant(struct cgroup_subsys_state *pos)
3460 struct cgroup_subsys_state *last, *tmp;
3462 cgroup_assert_mutex_or_rcu_locked();
3466 /* ->prev isn't RCU safe, walk ->next till the end */
3468 css_for_each_child(tmp, last)
3475 static struct cgroup_subsys_state *
3476 css_leftmost_descendant(struct cgroup_subsys_state *pos)
3478 struct cgroup_subsys_state *last;
3482 pos = css_next_child(NULL, pos);
3489 * css_next_descendant_post - find the next descendant for post-order walk
3490 * @pos: the current position (%NULL to initiate traversal)
3491 * @root: css whose descendants to walk
3493 * To be used by css_for_each_descendant_post(). Find the next descendant
3494 * to visit for post-order traversal of @root's descendants. @root is
3495 * included in the iteration and the last node to be visited.
3497 * While this function requires cgroup_mutex or RCU read locking, it
3498 * doesn't require the whole traversal to be contained in a single critical
3499 * section. This function will return the correct next descendant as long
3500 * as both @pos and @cgroup are accessible and @pos is a descendant of
3503 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3504 * css which finished ->css_online() is guaranteed to be visible in the
3505 * future iterations and will stay visible until the last reference is put.
3506 * A css which hasn't finished ->css_online() or already finished
3507 * ->css_offline() may show up during traversal. It's each subsystem's
3508 * responsibility to synchronize against on/offlining.
3510 struct cgroup_subsys_state *
3511 css_next_descendant_post(struct cgroup_subsys_state *pos,
3512 struct cgroup_subsys_state *root)
3514 struct cgroup_subsys_state *next;
3516 cgroup_assert_mutex_or_rcu_locked();
3518 /* if first iteration, visit leftmost descendant which may be @root */
3520 return css_leftmost_descendant(root);
3522 /* if we visited @root, we're done */
3526 /* if there's an unvisited sibling, visit its leftmost descendant */
3527 next = css_next_child(pos, pos->parent);
3529 return css_leftmost_descendant(next);
3531 /* no sibling left, visit parent */
3536 * css_has_online_children - does a css have online children
3537 * @css: the target css
3539 * Returns %true if @css has any online children; otherwise, %false. This
3540 * function can be called from any context but the caller is responsible
3541 * for synchronizing against on/offlining as necessary.
3543 bool css_has_online_children(struct cgroup_subsys_state *css)
3545 struct cgroup_subsys_state *child;
3549 css_for_each_child(child, css) {
3550 if (child->flags & CSS_ONLINE) {
3560 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
3561 * @it: the iterator to advance
3563 * Advance @it to the next css_set to walk.
3565 static void css_task_iter_advance_css_set(struct css_task_iter *it)
3567 struct list_head *l = it->cset_pos;
3568 struct cgrp_cset_link *link;
3569 struct css_set *cset;
3571 lockdep_assert_held(&css_set_lock);
3573 /* Advance to the next non-empty css_set */
3576 if (l == it->cset_head) {
3577 it->cset_pos = NULL;
3578 it->task_pos = NULL;
3583 cset = container_of(l, struct css_set,
3584 e_cset_node[it->ss->id]);
3586 link = list_entry(l, struct cgrp_cset_link, cset_link);
3589 } while (!css_set_populated(cset));
3593 if (!list_empty(&cset->tasks))
3594 it->task_pos = cset->tasks.next;
3596 it->task_pos = cset->mg_tasks.next;
3598 it->tasks_head = &cset->tasks;
3599 it->mg_tasks_head = &cset->mg_tasks;
3602 * We don't keep css_sets locked across iteration steps and thus
3603 * need to take steps to ensure that iteration can be resumed after
3604 * the lock is re-acquired. Iteration is performed at two levels -
3605 * css_sets and tasks in them.
3607 * Once created, a css_set never leaves its cgroup lists, so a
3608 * pinned css_set is guaranteed to stay put and we can resume
3609 * iteration afterwards.
3611 * Tasks may leave @cset across iteration steps. This is resolved
3612 * by registering each iterator with the css_set currently being
3613 * walked and making css_set_move_task() advance iterators whose
3614 * next task is leaving.
3617 list_del(&it->iters_node);
3618 put_css_set_locked(it->cur_cset);
3621 it->cur_cset = cset;
3622 list_add(&it->iters_node, &cset->task_iters);
3625 static void css_task_iter_advance(struct css_task_iter *it)
3627 struct list_head *l = it->task_pos;
3629 lockdep_assert_held(&css_set_lock);
3633 * Advance iterator to find next entry. cset->tasks is consumed
3634 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
3639 if (l == it->tasks_head)
3640 l = it->mg_tasks_head->next;
3642 if (l == it->mg_tasks_head)
3643 css_task_iter_advance_css_set(it);
3649 * css_task_iter_start - initiate task iteration
3650 * @css: the css to walk tasks of
3651 * @it: the task iterator to use
3653 * Initiate iteration through the tasks of @css. The caller can call
3654 * css_task_iter_next() to walk through the tasks until the function
3655 * returns NULL. On completion of iteration, css_task_iter_end() must be
3658 void css_task_iter_start(struct cgroup_subsys_state *css,
3659 struct css_task_iter *it)
3661 /* no one should try to iterate before mounting cgroups */
3662 WARN_ON_ONCE(!use_task_css_set_links);
3664 memset(it, 0, sizeof(*it));
3666 spin_lock_irq(&css_set_lock);
3671 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
3673 it->cset_pos = &css->cgroup->cset_links;
3675 it->cset_head = it->cset_pos;
3677 css_task_iter_advance_css_set(it);
3679 spin_unlock_irq(&css_set_lock);
3683 * css_task_iter_next - return the next task for the iterator
3684 * @it: the task iterator being iterated
3686 * The "next" function for task iteration. @it should have been
3687 * initialized via css_task_iter_start(). Returns NULL when the iteration
3690 struct task_struct *css_task_iter_next(struct css_task_iter *it)
3693 put_task_struct(it->cur_task);
3694 it->cur_task = NULL;
3697 spin_lock_irq(&css_set_lock);
3700 it->cur_task = list_entry(it->task_pos, struct task_struct,
3702 get_task_struct(it->cur_task);
3703 css_task_iter_advance(it);
3706 spin_unlock_irq(&css_set_lock);
3708 return it->cur_task;
3712 * css_task_iter_end - finish task iteration
3713 * @it: the task iterator to finish
3715 * Finish task iteration started by css_task_iter_start().
3717 void css_task_iter_end(struct css_task_iter *it)
3720 spin_lock_irq(&css_set_lock);
3721 list_del(&it->iters_node);
3722 put_css_set_locked(it->cur_cset);
3723 spin_unlock_irq(&css_set_lock);
3727 put_task_struct(it->cur_task);
3730 static void cgroup_procs_release(struct kernfs_open_file *of)
3733 css_task_iter_end(of->priv);
3738 static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
3740 struct kernfs_open_file *of = s->private;
3741 struct css_task_iter *it = of->priv;
3742 struct task_struct *task;
3745 task = css_task_iter_next(it);
3746 } while (task && !thread_group_leader(task));
3751 static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
3753 struct kernfs_open_file *of = s->private;
3754 struct cgroup *cgrp = seq_css(s)->cgroup;
3755 struct css_task_iter *it = of->priv;
3758 * When a seq_file is seeked, it's always traversed sequentially
3759 * from position 0, so we can simply keep iterating on !0 *pos.
3762 if (WARN_ON_ONCE((*pos)++))
3763 return ERR_PTR(-EINVAL);
3765 it = kzalloc(sizeof(*it), GFP_KERNEL);
3767 return ERR_PTR(-ENOMEM);
3769 css_task_iter_start(&cgrp->self, it);
3770 } else if (!(*pos)++) {
3771 css_task_iter_end(it);
3772 css_task_iter_start(&cgrp->self, it);
3775 return cgroup_procs_next(s, NULL, NULL);
3778 static int cgroup_procs_show(struct seq_file *s, void *v)
3780 seq_printf(s, "%d\n", task_tgid_vnr(v));
3784 /* cgroup core interface files for the default hierarchy */
3785 static struct cftype cgroup_base_files[] = {
3787 .name = "cgroup.procs",
3788 .file_offset = offsetof(struct cgroup, procs_file),
3789 .release = cgroup_procs_release,
3790 .seq_start = cgroup_procs_start,
3791 .seq_next = cgroup_procs_next,
3792 .seq_show = cgroup_procs_show,
3793 .write = cgroup_procs_write,
3796 .name = "cgroup.controllers",
3797 .seq_show = cgroup_controllers_show,
3800 .name = "cgroup.subtree_control",
3801 .seq_show = cgroup_subtree_control_show,
3802 .write = cgroup_subtree_control_write,
3805 .name = "cgroup.events",
3806 .flags = CFTYPE_NOT_ON_ROOT,
3807 .file_offset = offsetof(struct cgroup, events_file),
3808 .seq_show = cgroup_events_show,
3814 * css destruction is four-stage process.
3816 * 1. Destruction starts. Killing of the percpu_ref is initiated.
3817 * Implemented in kill_css().
3819 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
3820 * and thus css_tryget_online() is guaranteed to fail, the css can be
3821 * offlined by invoking offline_css(). After offlining, the base ref is
3822 * put. Implemented in css_killed_work_fn().
3824 * 3. When the percpu_ref reaches zero, the only possible remaining
3825 * accessors are inside RCU read sections. css_release() schedules the
3828 * 4. After the grace period, the css can be freed. Implemented in
3829 * css_free_work_fn().
3831 * It is actually hairier because both step 2 and 4 require process context
3832 * and thus involve punting to css->destroy_work adding two additional
3833 * steps to the already complex sequence.
3835 static void css_free_work_fn(struct work_struct *work)
3837 struct cgroup_subsys_state *css =
3838 container_of(work, struct cgroup_subsys_state, destroy_work);
3839 struct cgroup_subsys *ss = css->ss;
3840 struct cgroup *cgrp = css->cgroup;
3842 percpu_ref_exit(&css->refcnt);
3846 struct cgroup_subsys_state *parent = css->parent;
3850 cgroup_idr_remove(&ss->css_idr, id);
3856 /* cgroup free path */
3857 atomic_dec(&cgrp->root->nr_cgrps);
3858 cgroup1_pidlist_destroy_all(cgrp);
3859 cancel_work_sync(&cgrp->release_agent_work);
3861 if (cgroup_parent(cgrp)) {
3863 * We get a ref to the parent, and put the ref when
3864 * this cgroup is being freed, so it's guaranteed
3865 * that the parent won't be destroyed before its
3868 cgroup_put(cgroup_parent(cgrp));
3869 kernfs_put(cgrp->kn);
3873 * This is root cgroup's refcnt reaching zero,
3874 * which indicates that the root should be
3877 cgroup_destroy_root(cgrp->root);
3882 static void css_free_rcu_fn(struct rcu_head *rcu_head)
3884 struct cgroup_subsys_state *css =
3885 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
3887 INIT_WORK(&css->destroy_work, css_free_work_fn);
3888 queue_work(cgroup_destroy_wq, &css->destroy_work);
3891 static void css_release_work_fn(struct work_struct *work)
3893 struct cgroup_subsys_state *css =
3894 container_of(work, struct cgroup_subsys_state, destroy_work);
3895 struct cgroup_subsys *ss = css->ss;
3896 struct cgroup *cgrp = css->cgroup;
3898 mutex_lock(&cgroup_mutex);
3900 css->flags |= CSS_RELEASED;
3901 list_del_rcu(&css->sibling);
3904 /* css release path */
3905 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
3906 if (ss->css_released)
3907 ss->css_released(css);
3909 /* cgroup release path */
3910 trace_cgroup_release(cgrp);
3912 cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
3916 * There are two control paths which try to determine
3917 * cgroup from dentry without going through kernfs -
3918 * cgroupstats_build() and css_tryget_online_from_dir().
3919 * Those are supported by RCU protecting clearing of
3920 * cgrp->kn->priv backpointer.
3923 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv,
3926 cgroup_bpf_put(cgrp);
3929 mutex_unlock(&cgroup_mutex);
3931 call_rcu(&css->rcu_head, css_free_rcu_fn);
3934 static void css_release(struct percpu_ref *ref)
3936 struct cgroup_subsys_state *css =
3937 container_of(ref, struct cgroup_subsys_state, refcnt);
3939 INIT_WORK(&css->destroy_work, css_release_work_fn);
3940 queue_work(cgroup_destroy_wq, &css->destroy_work);
3943 static void init_and_link_css(struct cgroup_subsys_state *css,
3944 struct cgroup_subsys *ss, struct cgroup *cgrp)
3946 lockdep_assert_held(&cgroup_mutex);
3950 memset(css, 0, sizeof(*css));
3954 INIT_LIST_HEAD(&css->sibling);
3955 INIT_LIST_HEAD(&css->children);
3956 css->serial_nr = css_serial_nr_next++;
3957 atomic_set(&css->online_cnt, 0);
3959 if (cgroup_parent(cgrp)) {
3960 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
3961 css_get(css->parent);
3964 BUG_ON(cgroup_css(cgrp, ss));
3967 /* invoke ->css_online() on a new CSS and mark it online if successful */
3968 static int online_css(struct cgroup_subsys_state *css)
3970 struct cgroup_subsys *ss = css->ss;
3973 lockdep_assert_held(&cgroup_mutex);
3976 ret = ss->css_online(css);
3978 css->flags |= CSS_ONLINE;
3979 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
3981 atomic_inc(&css->online_cnt);
3983 atomic_inc(&css->parent->online_cnt);
3988 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
3989 static void offline_css(struct cgroup_subsys_state *css)
3991 struct cgroup_subsys *ss = css->ss;
3993 lockdep_assert_held(&cgroup_mutex);
3995 if (!(css->flags & CSS_ONLINE))
4001 if (ss->css_offline)
4002 ss->css_offline(css);
4004 css->flags &= ~CSS_ONLINE;
4005 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
4007 wake_up_all(&css->cgroup->offline_waitq);
4011 * css_create - create a cgroup_subsys_state
4012 * @cgrp: the cgroup new css will be associated with
4013 * @ss: the subsys of new css
4015 * Create a new css associated with @cgrp - @ss pair. On success, the new
4016 * css is online and installed in @cgrp. This function doesn't create the
4017 * interface files. Returns 0 on success, -errno on failure.
4019 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
4020 struct cgroup_subsys *ss)
4022 struct cgroup *parent = cgroup_parent(cgrp);
4023 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
4024 struct cgroup_subsys_state *css;
4027 lockdep_assert_held(&cgroup_mutex);
4029 css = ss->css_alloc(parent_css);
4031 css = ERR_PTR(-ENOMEM);
4035 init_and_link_css(css, ss, cgrp);
4037 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
4041 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
4046 /* @css is ready to be brought online now, make it visible */
4047 list_add_tail_rcu(&css->sibling, &parent_css->children);
4048 cgroup_idr_replace(&ss->css_idr, css, css->id);
4050 err = online_css(css);
4054 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4055 cgroup_parent(parent)) {
4056 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4057 current->comm, current->pid, ss->name);
4058 if (!strcmp(ss->name, "memory"))
4059 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4060 ss->warned_broken_hierarchy = true;
4066 list_del_rcu(&css->sibling);
4068 call_rcu(&css->rcu_head, css_free_rcu_fn);
4069 return ERR_PTR(err);
4073 * The returned cgroup is fully initialized including its control mask, but
4074 * it isn't associated with its kernfs_node and doesn't have the control
4077 static struct cgroup *cgroup_create(struct cgroup *parent)
4079 struct cgroup_root *root = parent->root;
4080 struct cgroup *cgrp, *tcgrp;
4081 int level = parent->level + 1;
4084 /* allocate the cgroup and its ID, 0 is reserved for the root */
4085 cgrp = kzalloc(sizeof(*cgrp) +
4086 sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL);
4088 return ERR_PTR(-ENOMEM);
4090 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
4095 * Temporarily set the pointer to NULL, so idr_find() won't return
4096 * a half-baked cgroup.
4098 cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL);
4101 goto out_cancel_ref;
4104 init_cgroup_housekeeping(cgrp);
4106 cgrp->self.parent = &parent->self;
4108 cgrp->level = level;
4110 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp))
4111 cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
4113 if (notify_on_release(parent))
4114 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4116 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4117 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4119 cgrp->self.serial_nr = css_serial_nr_next++;
4121 /* allocation complete, commit to creation */
4122 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
4123 atomic_inc(&root->nr_cgrps);
4127 * @cgrp is now fully operational. If something fails after this
4128 * point, it'll be released via the normal destruction path.
4130 cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
4133 * On the default hierarchy, a child doesn't automatically inherit
4134 * subtree_control from the parent. Each is configured manually.
4136 if (!cgroup_on_dfl(cgrp))
4137 cgrp->subtree_control = cgroup_control(cgrp);
4140 cgroup_bpf_inherit(cgrp, parent);
4142 cgroup_propagate_control(cgrp);
4147 percpu_ref_exit(&cgrp->self.refcnt);
4150 return ERR_PTR(ret);
4153 int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode)
4155 struct cgroup *parent, *cgrp;
4156 struct kernfs_node *kn;
4159 /* do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable */
4160 if (strchr(name, '\n'))
4163 parent = cgroup_kn_lock_live(parent_kn, false);
4167 cgrp = cgroup_create(parent);
4169 ret = PTR_ERR(cgrp);
4173 /* create the directory */
4174 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
4182 * This extra ref will be put in cgroup_free_fn() and guarantees
4183 * that @cgrp->kn is always accessible.
4187 ret = cgroup_kn_set_ugid(kn);
4191 ret = css_populate_dir(&cgrp->self);
4195 ret = cgroup_apply_control_enable(cgrp);
4199 trace_cgroup_mkdir(cgrp);
4201 /* let's create and online css's */
4202 kernfs_activate(kn);
4208 cgroup_destroy_locked(cgrp);
4210 cgroup_kn_unlock(parent_kn);
4215 * This is called when the refcnt of a css is confirmed to be killed.
4216 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
4217 * initate destruction and put the css ref from kill_css().
4219 static void css_killed_work_fn(struct work_struct *work)
4221 struct cgroup_subsys_state *css =
4222 container_of(work, struct cgroup_subsys_state, destroy_work);
4224 mutex_lock(&cgroup_mutex);
4229 /* @css can't go away while we're holding cgroup_mutex */
4231 } while (css && atomic_dec_and_test(&css->online_cnt));
4233 mutex_unlock(&cgroup_mutex);
4236 /* css kill confirmation processing requires process context, bounce */
4237 static void css_killed_ref_fn(struct percpu_ref *ref)
4239 struct cgroup_subsys_state *css =
4240 container_of(ref, struct cgroup_subsys_state, refcnt);
4242 if (atomic_dec_and_test(&css->online_cnt)) {
4243 INIT_WORK(&css->destroy_work, css_killed_work_fn);
4244 queue_work(cgroup_destroy_wq, &css->destroy_work);
4249 * kill_css - destroy a css
4250 * @css: css to destroy
4252 * This function initiates destruction of @css by removing cgroup interface
4253 * files and putting its base reference. ->css_offline() will be invoked
4254 * asynchronously once css_tryget_online() is guaranteed to fail and when
4255 * the reference count reaches zero, @css will be released.
4257 static void kill_css(struct cgroup_subsys_state *css)
4259 lockdep_assert_held(&cgroup_mutex);
4262 * This must happen before css is disassociated with its cgroup.
4263 * See seq_css() for details.
4268 * Killing would put the base ref, but we need to keep it alive
4269 * until after ->css_offline().
4274 * cgroup core guarantees that, by the time ->css_offline() is
4275 * invoked, no new css reference will be given out via
4276 * css_tryget_online(). We can't simply call percpu_ref_kill() and
4277 * proceed to offlining css's because percpu_ref_kill() doesn't
4278 * guarantee that the ref is seen as killed on all CPUs on return.
4280 * Use percpu_ref_kill_and_confirm() to get notifications as each
4281 * css is confirmed to be seen as killed on all CPUs.
4283 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
4287 * cgroup_destroy_locked - the first stage of cgroup destruction
4288 * @cgrp: cgroup to be destroyed
4290 * css's make use of percpu refcnts whose killing latency shouldn't be
4291 * exposed to userland and are RCU protected. Also, cgroup core needs to
4292 * guarantee that css_tryget_online() won't succeed by the time
4293 * ->css_offline() is invoked. To satisfy all the requirements,
4294 * destruction is implemented in the following two steps.
4296 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4297 * userland visible parts and start killing the percpu refcnts of
4298 * css's. Set up so that the next stage will be kicked off once all
4299 * the percpu refcnts are confirmed to be killed.
4301 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4302 * rest of destruction. Once all cgroup references are gone, the
4303 * cgroup is RCU-freed.
4305 * This function implements s1. After this step, @cgrp is gone as far as
4306 * the userland is concerned and a new cgroup with the same name may be
4307 * created. As cgroup doesn't care about the names internally, this
4308 * doesn't cause any problem.
4310 static int cgroup_destroy_locked(struct cgroup *cgrp)
4311 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4313 struct cgroup_subsys_state *css;
4314 struct cgrp_cset_link *link;
4317 lockdep_assert_held(&cgroup_mutex);
4320 * Only migration can raise populated from zero and we're already
4321 * holding cgroup_mutex.
4323 if (cgroup_is_populated(cgrp))
4327 * Make sure there's no live children. We can't test emptiness of
4328 * ->self.children as dead children linger on it while being
4329 * drained; otherwise, "rmdir parent/child parent" may fail.
4331 if (css_has_online_children(&cgrp->self))
4335 * Mark @cgrp and the associated csets dead. The former prevents
4336 * further task migration and child creation by disabling
4337 * cgroup_lock_live_group(). The latter makes the csets ignored by
4338 * the migration path.
4340 cgrp->self.flags &= ~CSS_ONLINE;
4342 spin_lock_irq(&css_set_lock);
4343 list_for_each_entry(link, &cgrp->cset_links, cset_link)
4344 link->cset->dead = true;
4345 spin_unlock_irq(&css_set_lock);
4347 /* initiate massacre of all css's */
4348 for_each_css(css, ssid, cgrp)
4352 * Remove @cgrp directory along with the base files. @cgrp has an
4353 * extra ref on its kn.
4355 kernfs_remove(cgrp->kn);
4357 cgroup1_check_for_release(cgroup_parent(cgrp));
4359 /* put the base reference */
4360 percpu_ref_kill(&cgrp->self.refcnt);
4365 int cgroup_rmdir(struct kernfs_node *kn)
4367 struct cgroup *cgrp;
4370 cgrp = cgroup_kn_lock_live(kn, false);
4374 ret = cgroup_destroy_locked(cgrp);
4377 trace_cgroup_rmdir(cgrp);
4379 cgroup_kn_unlock(kn);
4383 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
4384 .remount_fs = cgroup_remount,
4385 .mkdir = cgroup_mkdir,
4386 .rmdir = cgroup_rmdir,
4387 .show_path = cgroup_show_path,
4390 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
4392 struct cgroup_subsys_state *css;
4394 pr_debug("Initializing cgroup subsys %s\n", ss->name);
4396 mutex_lock(&cgroup_mutex);
4398 idr_init(&ss->css_idr);
4399 INIT_LIST_HEAD(&ss->cfts);
4401 /* Create the root cgroup state for this subsystem */
4402 ss->root = &cgrp_dfl_root;
4403 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
4404 /* We don't handle early failures gracefully */
4405 BUG_ON(IS_ERR(css));
4406 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
4409 * Root csses are never destroyed and we can't initialize
4410 * percpu_ref during early init. Disable refcnting.
4412 css->flags |= CSS_NO_REF;
4415 /* allocation can't be done safely during early init */
4418 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
4419 BUG_ON(css->id < 0);
4422 /* Update the init_css_set to contain a subsys
4423 * pointer to this state - since the subsystem is
4424 * newly registered, all tasks and hence the
4425 * init_css_set is in the subsystem's root cgroup. */
4426 init_css_set.subsys[ss->id] = css;
4428 have_fork_callback |= (bool)ss->fork << ss->id;
4429 have_exit_callback |= (bool)ss->exit << ss->id;
4430 have_free_callback |= (bool)ss->free << ss->id;
4431 have_canfork_callback |= (bool)ss->can_fork << ss->id;
4433 /* At system boot, before all subsystems have been
4434 * registered, no tasks have been forked, so we don't
4435 * need to invoke fork callbacks here. */
4436 BUG_ON(!list_empty(&init_task.tasks));
4438 BUG_ON(online_css(css));
4440 mutex_unlock(&cgroup_mutex);
4444 * cgroup_init_early - cgroup initialization at system boot
4446 * Initialize cgroups at system boot, and initialize any
4447 * subsystems that request early init.
4449 int __init cgroup_init_early(void)
4451 static struct cgroup_sb_opts __initdata opts;
4452 struct cgroup_subsys *ss;
4455 init_cgroup_root(&cgrp_dfl_root, &opts);
4456 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
4458 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
4460 for_each_subsys(ss, i) {
4461 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
4462 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p id:name=%d:%s\n",
4463 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
4465 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
4466 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
4469 ss->name = cgroup_subsys_name[i];
4470 if (!ss->legacy_name)
4471 ss->legacy_name = cgroup_subsys_name[i];
4474 cgroup_init_subsys(ss, true);
4479 static u16 cgroup_disable_mask __initdata;
4482 * cgroup_init - cgroup initialization
4484 * Register cgroup filesystem and /proc file, and initialize
4485 * any subsystems that didn't request early init.
4487 int __init cgroup_init(void)
4489 struct cgroup_subsys *ss;
4492 BUILD_BUG_ON(CGROUP_SUBSYS_COUNT > 16);
4493 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
4494 BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
4495 BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
4498 * The latency of the synchronize_sched() is too high for cgroups,
4499 * avoid it at the cost of forcing all readers into the slow path.
4501 rcu_sync_enter_start(&cgroup_threadgroup_rwsem.rss);
4503 get_user_ns(init_cgroup_ns.user_ns);
4505 mutex_lock(&cgroup_mutex);
4508 * Add init_css_set to the hash table so that dfl_root can link to
4511 hash_add(css_set_table, &init_css_set.hlist,
4512 css_set_hash(init_css_set.subsys));
4514 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
4516 mutex_unlock(&cgroup_mutex);
4518 for_each_subsys(ss, ssid) {
4519 if (ss->early_init) {
4520 struct cgroup_subsys_state *css =
4521 init_css_set.subsys[ss->id];
4523 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
4525 BUG_ON(css->id < 0);
4527 cgroup_init_subsys(ss, false);
4530 list_add_tail(&init_css_set.e_cset_node[ssid],
4531 &cgrp_dfl_root.cgrp.e_csets[ssid]);
4534 * Setting dfl_root subsys_mask needs to consider the
4535 * disabled flag and cftype registration needs kmalloc,
4536 * both of which aren't available during early_init.
4538 if (cgroup_disable_mask & (1 << ssid)) {
4539 static_branch_disable(cgroup_subsys_enabled_key[ssid]);
4540 printk(KERN_INFO "Disabling %s control group subsystem\n",
4545 if (cgroup1_ssid_disabled(ssid))
4546 printk(KERN_INFO "Disabling %s control group subsystem in v1 mounts\n",
4549 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
4551 if (ss->implicit_on_dfl)
4552 cgrp_dfl_implicit_ss_mask |= 1 << ss->id;
4553 else if (!ss->dfl_cftypes)
4554 cgrp_dfl_inhibit_ss_mask |= 1 << ss->id;
4556 if (ss->dfl_cftypes == ss->legacy_cftypes) {
4557 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
4559 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
4560 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
4564 ss->bind(init_css_set.subsys[ssid]);
4567 /* init_css_set.subsys[] has been updated, re-hash */
4568 hash_del(&init_css_set.hlist);
4569 hash_add(css_set_table, &init_css_set.hlist,
4570 css_set_hash(init_css_set.subsys));
4572 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
4573 WARN_ON(register_filesystem(&cgroup_fs_type));
4574 WARN_ON(register_filesystem(&cgroup2_fs_type));
4575 WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations));
4580 static int __init cgroup_wq_init(void)
4583 * There isn't much point in executing destruction path in
4584 * parallel. Good chunk is serialized with cgroup_mutex anyway.
4585 * Use 1 for @max_active.
4587 * We would prefer to do this in cgroup_init() above, but that
4588 * is called before init_workqueues(): so leave this until after.
4590 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
4591 BUG_ON(!cgroup_destroy_wq);
4594 core_initcall(cgroup_wq_init);
4597 * proc_cgroup_show()
4598 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4599 * - Used for /proc/<pid>/cgroup.
4601 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
4602 struct pid *pid, struct task_struct *tsk)
4606 struct cgroup_root *root;
4609 buf = kmalloc(PATH_MAX, GFP_KERNEL);
4613 mutex_lock(&cgroup_mutex);
4614 spin_lock_irq(&css_set_lock);
4616 for_each_root(root) {
4617 struct cgroup_subsys *ss;
4618 struct cgroup *cgrp;
4619 int ssid, count = 0;
4621 if (root == &cgrp_dfl_root && !cgrp_dfl_visible)
4624 seq_printf(m, "%d:", root->hierarchy_id);
4625 if (root != &cgrp_dfl_root)
4626 for_each_subsys(ss, ssid)
4627 if (root->subsys_mask & (1 << ssid))
4628 seq_printf(m, "%s%s", count++ ? "," : "",
4630 if (strlen(root->name))
4631 seq_printf(m, "%sname=%s", count ? "," : "",
4635 cgrp = task_cgroup_from_root(tsk, root);
4638 * On traditional hierarchies, all zombie tasks show up as
4639 * belonging to the root cgroup. On the default hierarchy,
4640 * while a zombie doesn't show up in "cgroup.procs" and
4641 * thus can't be migrated, its /proc/PID/cgroup keeps
4642 * reporting the cgroup it belonged to before exiting. If
4643 * the cgroup is removed before the zombie is reaped,
4644 * " (deleted)" is appended to the cgroup path.
4646 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
4647 retval = cgroup_path_ns_locked(cgrp, buf, PATH_MAX,
4648 current->nsproxy->cgroup_ns);
4649 if (retval >= PATH_MAX)
4650 retval = -ENAMETOOLONG;
4659 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
4660 seq_puts(m, " (deleted)\n");
4667 spin_unlock_irq(&css_set_lock);
4668 mutex_unlock(&cgroup_mutex);
4675 * cgroup_fork - initialize cgroup related fields during copy_process()
4676 * @child: pointer to task_struct of forking parent process.
4678 * A task is associated with the init_css_set until cgroup_post_fork()
4679 * attaches it to the parent's css_set. Empty cg_list indicates that
4680 * @child isn't holding reference to its css_set.
4682 void cgroup_fork(struct task_struct *child)
4684 RCU_INIT_POINTER(child->cgroups, &init_css_set);
4685 INIT_LIST_HEAD(&child->cg_list);
4689 * cgroup_can_fork - called on a new task before the process is exposed
4690 * @child: the task in question.
4692 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
4693 * returns an error, the fork aborts with that error code. This allows for
4694 * a cgroup subsystem to conditionally allow or deny new forks.
4696 int cgroup_can_fork(struct task_struct *child)
4698 struct cgroup_subsys *ss;
4701 do_each_subsys_mask(ss, i, have_canfork_callback) {
4702 ret = ss->can_fork(child);
4705 } while_each_subsys_mask();
4710 for_each_subsys(ss, j) {
4713 if (ss->cancel_fork)
4714 ss->cancel_fork(child);
4721 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
4722 * @child: the task in question
4724 * This calls the cancel_fork() callbacks if a fork failed *after*
4725 * cgroup_can_fork() succeded.
4727 void cgroup_cancel_fork(struct task_struct *child)
4729 struct cgroup_subsys *ss;
4732 for_each_subsys(ss, i)
4733 if (ss->cancel_fork)
4734 ss->cancel_fork(child);
4738 * cgroup_post_fork - called on a new task after adding it to the task list
4739 * @child: the task in question
4741 * Adds the task to the list running through its css_set if necessary and
4742 * call the subsystem fork() callbacks. Has to be after the task is
4743 * visible on the task list in case we race with the first call to
4744 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
4747 void cgroup_post_fork(struct task_struct *child)
4749 struct cgroup_subsys *ss;
4753 * This may race against cgroup_enable_task_cg_lists(). As that
4754 * function sets use_task_css_set_links before grabbing
4755 * tasklist_lock and we just went through tasklist_lock to add
4756 * @child, it's guaranteed that either we see the set
4757 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
4758 * @child during its iteration.
4760 * If we won the race, @child is associated with %current's
4761 * css_set. Grabbing css_set_lock guarantees both that the
4762 * association is stable, and, on completion of the parent's
4763 * migration, @child is visible in the source of migration or
4764 * already in the destination cgroup. This guarantee is necessary
4765 * when implementing operations which need to migrate all tasks of
4766 * a cgroup to another.
4768 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
4769 * will remain in init_css_set. This is safe because all tasks are
4770 * in the init_css_set before cg_links is enabled and there's no
4771 * operation which transfers all tasks out of init_css_set.
4773 if (use_task_css_set_links) {
4774 struct css_set *cset;
4776 spin_lock_irq(&css_set_lock);
4777 cset = task_css_set(current);
4778 if (list_empty(&child->cg_list)) {
4780 css_set_move_task(child, NULL, cset, false);
4782 spin_unlock_irq(&css_set_lock);
4786 * Call ss->fork(). This must happen after @child is linked on
4787 * css_set; otherwise, @child might change state between ->fork()
4788 * and addition to css_set.
4790 do_each_subsys_mask(ss, i, have_fork_callback) {
4792 } while_each_subsys_mask();
4796 * cgroup_exit - detach cgroup from exiting task
4797 * @tsk: pointer to task_struct of exiting process
4799 * Description: Detach cgroup from @tsk and release it.
4801 * Note that cgroups marked notify_on_release force every task in
4802 * them to take the global cgroup_mutex mutex when exiting.
4803 * This could impact scaling on very large systems. Be reluctant to
4804 * use notify_on_release cgroups where very high task exit scaling
4805 * is required on large systems.
4807 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
4808 * call cgroup_exit() while the task is still competent to handle
4809 * notify_on_release(), then leave the task attached to the root cgroup in
4810 * each hierarchy for the remainder of its exit. No need to bother with
4811 * init_css_set refcnting. init_css_set never goes away and we can't race
4812 * with migration path - PF_EXITING is visible to migration path.
4814 void cgroup_exit(struct task_struct *tsk)
4816 struct cgroup_subsys *ss;
4817 struct css_set *cset;
4821 * Unlink from @tsk from its css_set. As migration path can't race
4822 * with us, we can check css_set and cg_list without synchronization.
4824 cset = task_css_set(tsk);
4826 if (!list_empty(&tsk->cg_list)) {
4827 spin_lock_irq(&css_set_lock);
4828 css_set_move_task(tsk, cset, NULL, false);
4829 spin_unlock_irq(&css_set_lock);
4834 /* see cgroup_post_fork() for details */
4835 do_each_subsys_mask(ss, i, have_exit_callback) {
4837 } while_each_subsys_mask();
4840 void cgroup_free(struct task_struct *task)
4842 struct css_set *cset = task_css_set(task);
4843 struct cgroup_subsys *ss;
4846 do_each_subsys_mask(ss, ssid, have_free_callback) {
4848 } while_each_subsys_mask();
4853 static int __init cgroup_disable(char *str)
4855 struct cgroup_subsys *ss;
4859 while ((token = strsep(&str, ",")) != NULL) {
4863 for_each_subsys(ss, i) {
4864 if (strcmp(token, ss->name) &&
4865 strcmp(token, ss->legacy_name))
4867 cgroup_disable_mask |= 1 << i;
4872 __setup("cgroup_disable=", cgroup_disable);
4875 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
4876 * @dentry: directory dentry of interest
4877 * @ss: subsystem of interest
4879 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
4880 * to get the corresponding css and return it. If such css doesn't exist
4881 * or can't be pinned, an ERR_PTR value is returned.
4883 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
4884 struct cgroup_subsys *ss)
4886 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
4887 struct file_system_type *s_type = dentry->d_sb->s_type;
4888 struct cgroup_subsys_state *css = NULL;
4889 struct cgroup *cgrp;
4891 /* is @dentry a cgroup dir? */
4892 if ((s_type != &cgroup_fs_type && s_type != &cgroup2_fs_type) ||
4893 !kn || kernfs_type(kn) != KERNFS_DIR)
4894 return ERR_PTR(-EBADF);
4899 * This path doesn't originate from kernfs and @kn could already
4900 * have been or be removed at any point. @kn->priv is RCU
4901 * protected for this access. See css_release_work_fn() for details.
4903 cgrp = rcu_dereference(*(void __rcu __force **)&kn->priv);
4905 css = cgroup_css(cgrp, ss);
4907 if (!css || !css_tryget_online(css))
4908 css = ERR_PTR(-ENOENT);
4915 * css_from_id - lookup css by id
4916 * @id: the cgroup id
4917 * @ss: cgroup subsys to be looked into
4919 * Returns the css if there's valid one with @id, otherwise returns NULL.
4920 * Should be called under rcu_read_lock().
4922 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
4924 WARN_ON_ONCE(!rcu_read_lock_held());
4925 return idr_find(&ss->css_idr, id);
4929 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
4930 * @path: path on the default hierarchy
4932 * Find the cgroup at @path on the default hierarchy, increment its
4933 * reference count and return it. Returns pointer to the found cgroup on
4934 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
4935 * if @path points to a non-directory.
4937 struct cgroup *cgroup_get_from_path(const char *path)
4939 struct kernfs_node *kn;
4940 struct cgroup *cgrp;
4942 mutex_lock(&cgroup_mutex);
4944 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
4946 if (kernfs_type(kn) == KERNFS_DIR) {
4950 cgrp = ERR_PTR(-ENOTDIR);
4954 cgrp = ERR_PTR(-ENOENT);
4957 mutex_unlock(&cgroup_mutex);
4960 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
4963 * cgroup_get_from_fd - get a cgroup pointer from a fd
4964 * @fd: fd obtained by open(cgroup2_dir)
4966 * Find the cgroup from a fd which should be obtained
4967 * by opening a cgroup directory. Returns a pointer to the
4968 * cgroup on success. ERR_PTR is returned if the cgroup
4971 struct cgroup *cgroup_get_from_fd(int fd)
4973 struct cgroup_subsys_state *css;
4974 struct cgroup *cgrp;
4979 return ERR_PTR(-EBADF);
4981 css = css_tryget_online_from_dir(f->f_path.dentry, NULL);
4984 return ERR_CAST(css);
4987 if (!cgroup_on_dfl(cgrp)) {
4989 return ERR_PTR(-EBADF);
4994 EXPORT_SYMBOL_GPL(cgroup_get_from_fd);
4997 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
4998 * definition in cgroup-defs.h.
5000 #ifdef CONFIG_SOCK_CGROUP_DATA
5002 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
5004 DEFINE_SPINLOCK(cgroup_sk_update_lock);
5005 static bool cgroup_sk_alloc_disabled __read_mostly;
5007 void cgroup_sk_alloc_disable(void)
5009 if (cgroup_sk_alloc_disabled)
5011 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
5012 cgroup_sk_alloc_disabled = true;
5017 #define cgroup_sk_alloc_disabled false
5021 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
5023 if (cgroup_sk_alloc_disabled)
5026 /* Socket clone path */
5028 cgroup_get(sock_cgroup_ptr(skcd));
5035 struct css_set *cset;
5037 cset = task_css_set(current);
5038 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
5039 skcd->val = (unsigned long)cset->dfl_cgrp;
5048 void cgroup_sk_free(struct sock_cgroup_data *skcd)
5050 cgroup_put(sock_cgroup_ptr(skcd));
5053 #endif /* CONFIG_SOCK_CGROUP_DATA */
5055 #ifdef CONFIG_CGROUP_BPF
5056 int cgroup_bpf_update(struct cgroup *cgrp, struct bpf_prog *prog,
5057 enum bpf_attach_type type, bool overridable)
5059 struct cgroup *parent = cgroup_parent(cgrp);
5062 mutex_lock(&cgroup_mutex);
5063 ret = __cgroup_bpf_update(cgrp, parent, prog, type, overridable);
5064 mutex_unlock(&cgroup_mutex);
5067 #endif /* CONFIG_CGROUP_BPF */