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 <linux/cgroup.h>
32 #include <linux/cred.h>
33 #include <linux/ctype.h>
34 #include <linux/errno.h>
35 #include <linux/init_task.h>
36 #include <linux/kernel.h>
37 #include <linux/list.h>
38 #include <linux/magic.h>
40 #include <linux/mutex.h>
41 #include <linux/mount.h>
42 #include <linux/pagemap.h>
43 #include <linux/proc_fs.h>
44 #include <linux/rcupdate.h>
45 #include <linux/sched.h>
46 #include <linux/slab.h>
47 #include <linux/spinlock.h>
48 #include <linux/percpu-rwsem.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/pid_namespace.h>
56 #include <linux/idr.h>
57 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
58 #include <linux/kthread.h>
59 #include <linux/delay.h>
60 #include <linux/atomic.h>
64 * pidlists linger the following amount before being destroyed. The goal
65 * is avoiding frequent destruction in the middle of consecutive read calls
66 * Expiring in the middle is a performance problem not a correctness one.
67 * 1 sec should be enough.
69 #define CGROUP_PIDLIST_DESTROY_DELAY HZ
71 #define CGROUP_FILE_NAME_MAX (MAX_CGROUP_TYPE_NAMELEN + \
75 * cgroup_mutex is the master lock. Any modification to cgroup or its
76 * hierarchy must be performed while holding it.
78 * css_set_lock protects task->cgroups pointer, the list of css_set
79 * objects, and the chain of tasks off each css_set.
81 * These locks are exported if CONFIG_PROVE_RCU so that accessors in
82 * cgroup.h can use them for lockdep annotations.
84 #ifdef CONFIG_PROVE_RCU
85 DEFINE_MUTEX(cgroup_mutex);
86 DEFINE_SPINLOCK(css_set_lock);
87 EXPORT_SYMBOL_GPL(cgroup_mutex);
88 EXPORT_SYMBOL_GPL(css_set_lock);
90 static DEFINE_MUTEX(cgroup_mutex);
91 static DEFINE_SPINLOCK(css_set_lock);
95 * Protects cgroup_idr and css_idr so that IDs can be released without
96 * grabbing cgroup_mutex.
98 static DEFINE_SPINLOCK(cgroup_idr_lock);
101 * Protects cgroup_file->kn for !self csses. It synchronizes notifications
102 * against file removal/re-creation across css hiding.
104 static DEFINE_SPINLOCK(cgroup_file_kn_lock);
107 * Protects cgroup_subsys->release_agent_path. Modifying it also requires
108 * cgroup_mutex. Reading requires either cgroup_mutex or this spinlock.
110 static DEFINE_SPINLOCK(release_agent_path_lock);
112 struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
114 #define cgroup_assert_mutex_or_rcu_locked() \
115 RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
116 !lockdep_is_held(&cgroup_mutex), \
117 "cgroup_mutex or RCU read lock required");
120 * cgroup destruction makes heavy use of work items and there can be a lot
121 * of concurrent destructions. Use a separate workqueue so that cgroup
122 * destruction work items don't end up filling up max_active of system_wq
123 * which may lead to deadlock.
125 static struct workqueue_struct *cgroup_destroy_wq;
128 * pidlist destructions need to be flushed on cgroup destruction. Use a
129 * separate workqueue as flush domain.
131 static struct workqueue_struct *cgroup_pidlist_destroy_wq;
133 /* generate an array of cgroup subsystem pointers */
134 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys,
135 static struct cgroup_subsys *cgroup_subsys[] = {
136 #include <linux/cgroup_subsys.h>
140 /* array of cgroup subsystem names */
141 #define SUBSYS(_x) [_x ## _cgrp_id] = #_x,
142 static const char *cgroup_subsys_name[] = {
143 #include <linux/cgroup_subsys.h>
147 /* array of static_keys for cgroup_subsys_enabled() and cgroup_subsys_on_dfl() */
149 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_enabled_key); \
150 DEFINE_STATIC_KEY_TRUE(_x ## _cgrp_subsys_on_dfl_key); \
151 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_enabled_key); \
152 EXPORT_SYMBOL_GPL(_x ## _cgrp_subsys_on_dfl_key);
153 #include <linux/cgroup_subsys.h>
156 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_enabled_key,
157 static struct static_key_true *cgroup_subsys_enabled_key[] = {
158 #include <linux/cgroup_subsys.h>
162 #define SUBSYS(_x) [_x ## _cgrp_id] = &_x ## _cgrp_subsys_on_dfl_key,
163 static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
164 #include <linux/cgroup_subsys.h>
169 * The default hierarchy, reserved for the subsystems that are otherwise
170 * unattached - it never has more than a single cgroup, and all tasks are
171 * part of that cgroup.
173 struct cgroup_root cgrp_dfl_root;
174 EXPORT_SYMBOL_GPL(cgrp_dfl_root);
177 * The default hierarchy always exists but is hidden until mounted for the
178 * first time. This is for backward compatibility.
180 static bool cgrp_dfl_root_visible;
182 /* some controllers are not supported in the default hierarchy */
183 static unsigned long cgrp_dfl_root_inhibit_ss_mask;
185 /* The list of hierarchy roots */
187 static LIST_HEAD(cgroup_roots);
188 static int cgroup_root_count;
190 /* hierarchy ID allocation and mapping, protected by cgroup_mutex */
191 static DEFINE_IDR(cgroup_hierarchy_idr);
194 * Assign a monotonically increasing serial number to csses. It guarantees
195 * cgroups with bigger numbers are newer than those with smaller numbers.
196 * Also, as csses are always appended to the parent's ->children list, it
197 * guarantees that sibling csses are always sorted in the ascending serial
198 * number order on the list. Protected by cgroup_mutex.
200 static u64 css_serial_nr_next = 1;
203 * These bitmask flags indicate whether tasks in the fork and exit paths have
204 * fork/exit handlers to call. This avoids us having to do extra work in the
205 * fork/exit path to check which subsystems have fork/exit callbacks.
207 static unsigned long have_fork_callback __read_mostly;
208 static unsigned long have_exit_callback __read_mostly;
209 static unsigned long have_free_callback __read_mostly;
211 /* Ditto for the can_fork callback. */
212 static unsigned long have_canfork_callback __read_mostly;
214 static struct file_system_type cgroup2_fs_type;
215 static struct cftype cgroup_dfl_base_files[];
216 static struct cftype cgroup_legacy_base_files[];
218 static int rebind_subsystems(struct cgroup_root *dst_root,
219 unsigned long ss_mask);
220 static void css_task_iter_advance(struct css_task_iter *it);
221 static int cgroup_destroy_locked(struct cgroup *cgrp);
222 static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss,
224 static void css_release(struct percpu_ref *ref);
225 static void kill_css(struct cgroup_subsys_state *css);
226 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
227 struct cgroup *cgrp, struct cftype cfts[],
231 * cgroup_ssid_enabled - cgroup subsys enabled test by subsys ID
232 * @ssid: subsys ID of interest
234 * cgroup_subsys_enabled() can only be used with literal subsys names which
235 * is fine for individual subsystems but unsuitable for cgroup core. This
236 * is slower static_key_enabled() based test indexed by @ssid.
238 static bool cgroup_ssid_enabled(int ssid)
240 return static_key_enabled(cgroup_subsys_enabled_key[ssid]);
244 * cgroup_on_dfl - test whether a cgroup is on the default hierarchy
245 * @cgrp: the cgroup of interest
247 * The default hierarchy is the v2 interface of cgroup and this function
248 * can be used to test whether a cgroup is on the default hierarchy for
249 * cases where a subsystem should behave differnetly depending on the
252 * The set of behaviors which change on the default hierarchy are still
253 * being determined and the mount option is prefixed with __DEVEL__.
255 * List of changed behaviors:
257 * - Mount options "noprefix", "xattr", "clone_children", "release_agent"
258 * and "name" are disallowed.
260 * - When mounting an existing superblock, mount options should match.
262 * - Remount is disallowed.
264 * - rename(2) is disallowed.
266 * - "tasks" is removed. Everything should be at process granularity. Use
267 * "cgroup.procs" instead.
269 * - "cgroup.procs" is not sorted. pids will be unique unless they got
270 * recycled inbetween reads.
272 * - "release_agent" and "notify_on_release" are removed. Replacement
273 * notification mechanism will be implemented.
275 * - "cgroup.clone_children" is removed.
277 * - "cgroup.subtree_populated" is available. Its value is 0 if the cgroup
278 * and its descendants contain no task; otherwise, 1. The file also
279 * generates kernfs notification which can be monitored through poll and
280 * [di]notify when the value of the file changes.
282 * - cpuset: tasks will be kept in empty cpusets when hotplug happens and
283 * take masks of ancestors with non-empty cpus/mems, instead of being
284 * moved to an ancestor.
286 * - cpuset: a task can be moved into an empty cpuset, and again it takes
287 * masks of ancestors.
289 * - memcg: use_hierarchy is on by default and the cgroup file for the flag
292 * - blkcg: blk-throttle becomes properly hierarchical.
294 * - debug: disallowed on the default hierarchy.
296 static bool cgroup_on_dfl(const struct cgroup *cgrp)
298 return cgrp->root == &cgrp_dfl_root;
301 /* IDR wrappers which synchronize using cgroup_idr_lock */
302 static int cgroup_idr_alloc(struct idr *idr, void *ptr, int start, int end,
307 idr_preload(gfp_mask);
308 spin_lock_bh(&cgroup_idr_lock);
309 ret = idr_alloc(idr, ptr, start, end, gfp_mask & ~__GFP_DIRECT_RECLAIM);
310 spin_unlock_bh(&cgroup_idr_lock);
315 static void *cgroup_idr_replace(struct idr *idr, void *ptr, int id)
319 spin_lock_bh(&cgroup_idr_lock);
320 ret = idr_replace(idr, ptr, id);
321 spin_unlock_bh(&cgroup_idr_lock);
325 static void cgroup_idr_remove(struct idr *idr, int id)
327 spin_lock_bh(&cgroup_idr_lock);
329 spin_unlock_bh(&cgroup_idr_lock);
332 static struct cgroup *cgroup_parent(struct cgroup *cgrp)
334 struct cgroup_subsys_state *parent_css = cgrp->self.parent;
337 return container_of(parent_css, struct cgroup, self);
342 * cgroup_css - obtain a cgroup's css for the specified subsystem
343 * @cgrp: the cgroup of interest
344 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
346 * Return @cgrp's css (cgroup_subsys_state) associated with @ss. This
347 * function must be called either under cgroup_mutex or rcu_read_lock() and
348 * the caller is responsible for pinning the returned css if it wants to
349 * keep accessing it outside the said locks. This function may return
350 * %NULL if @cgrp doesn't have @subsys_id enabled.
352 static struct cgroup_subsys_state *cgroup_css(struct cgroup *cgrp,
353 struct cgroup_subsys *ss)
356 return rcu_dereference_check(cgrp->subsys[ss->id],
357 lockdep_is_held(&cgroup_mutex));
363 * cgroup_e_css - obtain a cgroup's effective css for the specified subsystem
364 * @cgrp: the cgroup of interest
365 * @ss: the subsystem of interest (%NULL returns @cgrp->self)
367 * Similar to cgroup_css() but returns the effective css, which is defined
368 * as the matching css of the nearest ancestor including self which has @ss
369 * enabled. If @ss is associated with the hierarchy @cgrp is on, this
370 * function is guaranteed to return non-NULL css.
372 static struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgrp,
373 struct cgroup_subsys *ss)
375 lockdep_assert_held(&cgroup_mutex);
380 if (!(cgrp->root->subsys_mask & (1 << ss->id)))
384 * This function is used while updating css associations and thus
385 * can't test the csses directly. Use ->child_subsys_mask.
387 while (cgroup_parent(cgrp) &&
388 !(cgroup_parent(cgrp)->child_subsys_mask & (1 << ss->id)))
389 cgrp = cgroup_parent(cgrp);
391 return cgroup_css(cgrp, ss);
395 * cgroup_get_e_css - get a cgroup's effective css for the specified subsystem
396 * @cgrp: the cgroup of interest
397 * @ss: the subsystem of interest
399 * Find and get the effective css of @cgrp for @ss. The effective css is
400 * defined as the matching css of the nearest ancestor including self which
401 * has @ss enabled. If @ss is not mounted on the hierarchy @cgrp is on,
402 * the root css is returned, so this function always returns a valid css.
403 * The returned css must be put using css_put().
405 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgrp,
406 struct cgroup_subsys *ss)
408 struct cgroup_subsys_state *css;
413 css = cgroup_css(cgrp, ss);
415 if (css && css_tryget_online(css))
417 cgrp = cgroup_parent(cgrp);
420 css = init_css_set.subsys[ss->id];
427 /* convenient tests for these bits */
428 static inline bool cgroup_is_dead(const struct cgroup *cgrp)
430 return !(cgrp->self.flags & CSS_ONLINE);
433 static void cgroup_get(struct cgroup *cgrp)
435 WARN_ON_ONCE(cgroup_is_dead(cgrp));
436 css_get(&cgrp->self);
439 static bool cgroup_tryget(struct cgroup *cgrp)
441 return css_tryget(&cgrp->self);
444 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of)
446 struct cgroup *cgrp = of->kn->parent->priv;
447 struct cftype *cft = of_cft(of);
450 * This is open and unprotected implementation of cgroup_css().
451 * seq_css() is only called from a kernfs file operation which has
452 * an active reference on the file. Because all the subsystem
453 * files are drained before a css is disassociated with a cgroup,
454 * the matching css from the cgroup's subsys table is guaranteed to
455 * be and stay valid until the enclosing operation is complete.
458 return rcu_dereference_raw(cgrp->subsys[cft->ss->id]);
462 EXPORT_SYMBOL_GPL(of_css);
464 static int notify_on_release(const struct cgroup *cgrp)
466 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
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 * for_each_subsys - iterate all enabled cgroup subsystems
500 * @ss: the iteration cursor
501 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
503 #define for_each_subsys(ss, ssid) \
504 for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
505 (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
508 * for_each_subsys_which - filter for_each_subsys with a bitmask
509 * @ss: the iteration cursor
510 * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
511 * @ss_maskp: a pointer to the bitmask
513 * The block will only run for cases where the ssid-th bit (1 << ssid) of
516 #define for_each_subsys_which(ss, ssid, ss_maskp) \
517 if (!CGROUP_SUBSYS_COUNT) /* to avoid spurious gcc warning */ \
520 for_each_set_bit(ssid, ss_maskp, CGROUP_SUBSYS_COUNT) \
521 if (((ss) = cgroup_subsys[ssid]) && false) \
525 /* iterate across the hierarchies */
526 #define for_each_root(root) \
527 list_for_each_entry((root), &cgroup_roots, root_list)
529 /* iterate over child cgrps, lock should be held throughout iteration */
530 #define cgroup_for_each_live_child(child, cgrp) \
531 list_for_each_entry((child), &(cgrp)->self.children, self.sibling) \
532 if (({ lockdep_assert_held(&cgroup_mutex); \
533 cgroup_is_dead(child); })) \
537 static void cgroup_release_agent(struct work_struct *work);
538 static void check_for_release(struct cgroup *cgrp);
541 * A cgroup can be associated with multiple css_sets as different tasks may
542 * belong to different cgroups on different hierarchies. In the other
543 * direction, a css_set is naturally associated with multiple cgroups.
544 * This M:N relationship is represented by the following link structure
545 * which exists for each association and allows traversing the associations
548 struct cgrp_cset_link {
549 /* the cgroup and css_set this link associates */
551 struct css_set *cset;
553 /* list of cgrp_cset_links anchored at cgrp->cset_links */
554 struct list_head cset_link;
556 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
557 struct list_head cgrp_link;
561 * The default css_set - used by init and its children prior to any
562 * hierarchies being mounted. It contains a pointer to the root state
563 * for each subsystem. Also used to anchor the list of css_sets. Not
564 * reference-counted, to improve performance when child cgroups
565 * haven't been created.
567 struct css_set init_css_set = {
568 .refcount = ATOMIC_INIT(1),
569 .cgrp_links = LIST_HEAD_INIT(init_css_set.cgrp_links),
570 .tasks = LIST_HEAD_INIT(init_css_set.tasks),
571 .mg_tasks = LIST_HEAD_INIT(init_css_set.mg_tasks),
572 .mg_preload_node = LIST_HEAD_INIT(init_css_set.mg_preload_node),
573 .mg_node = LIST_HEAD_INIT(init_css_set.mg_node),
574 .task_iters = LIST_HEAD_INIT(init_css_set.task_iters),
577 static int css_set_count = 1; /* 1 for init_css_set */
580 * css_set_populated - does a css_set contain any tasks?
581 * @cset: target css_set
583 static bool css_set_populated(struct css_set *cset)
585 lockdep_assert_held(&css_set_lock);
587 return !list_empty(&cset->tasks) || !list_empty(&cset->mg_tasks);
591 * cgroup_update_populated - updated populated count of a cgroup
592 * @cgrp: the target cgroup
593 * @populated: inc or dec populated count
595 * One of the css_sets associated with @cgrp is either getting its first
596 * task or losing the last. Update @cgrp->populated_cnt accordingly. The
597 * count is propagated towards root so that a given cgroup's populated_cnt
598 * is zero iff the cgroup and all its descendants don't contain any tasks.
600 * @cgrp's interface file "cgroup.populated" is zero if
601 * @cgrp->populated_cnt is zero and 1 otherwise. When @cgrp->populated_cnt
602 * changes from or to zero, userland is notified that the content of the
603 * interface file has changed. This can be used to detect when @cgrp and
604 * its descendants become populated or empty.
606 static void cgroup_update_populated(struct cgroup *cgrp, bool populated)
608 lockdep_assert_held(&css_set_lock);
614 trigger = !cgrp->populated_cnt++;
616 trigger = !--cgrp->populated_cnt;
621 check_for_release(cgrp);
622 cgroup_file_notify(&cgrp->events_file);
624 cgrp = cgroup_parent(cgrp);
629 * css_set_update_populated - update populated state of a css_set
630 * @cset: target css_set
631 * @populated: whether @cset is populated or depopulated
633 * @cset is either getting the first task or losing the last. Update the
634 * ->populated_cnt of all associated cgroups accordingly.
636 static void css_set_update_populated(struct css_set *cset, bool populated)
638 struct cgrp_cset_link *link;
640 lockdep_assert_held(&css_set_lock);
642 list_for_each_entry(link, &cset->cgrp_links, cgrp_link)
643 cgroup_update_populated(link->cgrp, populated);
647 * css_set_move_task - move a task from one css_set to another
648 * @task: task being moved
649 * @from_cset: css_set @task currently belongs to (may be NULL)
650 * @to_cset: new css_set @task is being moved to (may be NULL)
651 * @use_mg_tasks: move to @to_cset->mg_tasks instead of ->tasks
653 * Move @task from @from_cset to @to_cset. If @task didn't belong to any
654 * css_set, @from_cset can be NULL. If @task is being disassociated
655 * instead of moved, @to_cset can be NULL.
657 * This function automatically handles populated_cnt updates and
658 * css_task_iter adjustments but the caller is responsible for managing
659 * @from_cset and @to_cset's reference counts.
661 static void css_set_move_task(struct task_struct *task,
662 struct css_set *from_cset, struct css_set *to_cset,
665 lockdep_assert_held(&css_set_lock);
668 struct css_task_iter *it, *pos;
670 WARN_ON_ONCE(list_empty(&task->cg_list));
673 * @task is leaving, advance task iterators which are
674 * pointing to it so that they can resume at the next
675 * position. Advancing an iterator might remove it from
676 * the list, use safe walk. See css_task_iter_advance*()
679 list_for_each_entry_safe(it, pos, &from_cset->task_iters,
681 if (it->task_pos == &task->cg_list)
682 css_task_iter_advance(it);
684 list_del_init(&task->cg_list);
685 if (!css_set_populated(from_cset))
686 css_set_update_populated(from_cset, false);
688 WARN_ON_ONCE(!list_empty(&task->cg_list));
693 * We are synchronized through cgroup_threadgroup_rwsem
694 * against PF_EXITING setting such that we can't race
695 * against cgroup_exit() changing the css_set to
696 * init_css_set and dropping the old one.
698 WARN_ON_ONCE(task->flags & PF_EXITING);
700 if (!css_set_populated(to_cset))
701 css_set_update_populated(to_cset, true);
702 rcu_assign_pointer(task->cgroups, to_cset);
703 list_add_tail(&task->cg_list, use_mg_tasks ? &to_cset->mg_tasks :
709 * hash table for cgroup groups. This improves the performance to find
710 * an existing css_set. This hash doesn't (currently) take into
711 * account cgroups in empty hierarchies.
713 #define CSS_SET_HASH_BITS 7
714 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
716 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
718 unsigned long key = 0UL;
719 struct cgroup_subsys *ss;
722 for_each_subsys(ss, i)
723 key += (unsigned long)css[i];
724 key = (key >> 16) ^ key;
729 static void put_css_set_locked(struct css_set *cset)
731 struct cgrp_cset_link *link, *tmp_link;
732 struct cgroup_subsys *ss;
735 lockdep_assert_held(&css_set_lock);
737 if (!atomic_dec_and_test(&cset->refcount))
740 /* This css_set is dead. unlink it and release cgroup and css refs */
741 for_each_subsys(ss, ssid) {
742 list_del(&cset->e_cset_node[ssid]);
743 css_put(cset->subsys[ssid]);
745 hash_del(&cset->hlist);
748 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
749 list_del(&link->cset_link);
750 list_del(&link->cgrp_link);
751 if (cgroup_parent(link->cgrp))
752 cgroup_put(link->cgrp);
756 kfree_rcu(cset, rcu_head);
759 static void put_css_set(struct css_set *cset)
762 * Ensure that the refcount doesn't hit zero while any readers
763 * can see it. Similar to atomic_dec_and_lock(), but for an
766 if (atomic_add_unless(&cset->refcount, -1, 1))
769 spin_lock_bh(&css_set_lock);
770 put_css_set_locked(cset);
771 spin_unlock_bh(&css_set_lock);
775 * refcounted get/put for css_set objects
777 static inline void get_css_set(struct css_set *cset)
779 atomic_inc(&cset->refcount);
783 * compare_css_sets - helper function for find_existing_css_set().
784 * @cset: candidate css_set being tested
785 * @old_cset: existing css_set for a task
786 * @new_cgrp: cgroup that's being entered by the task
787 * @template: desired set of css pointers in css_set (pre-calculated)
789 * Returns true if "cset" matches "old_cset" except for the hierarchy
790 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
792 static bool compare_css_sets(struct css_set *cset,
793 struct css_set *old_cset,
794 struct cgroup *new_cgrp,
795 struct cgroup_subsys_state *template[])
797 struct list_head *l1, *l2;
800 * On the default hierarchy, there can be csets which are
801 * associated with the same set of cgroups but different csses.
802 * Let's first ensure that csses match.
804 if (memcmp(template, cset->subsys, sizeof(cset->subsys)))
808 * Compare cgroup pointers in order to distinguish between
809 * different cgroups in hierarchies. As different cgroups may
810 * share the same effective css, this comparison is always
813 l1 = &cset->cgrp_links;
814 l2 = &old_cset->cgrp_links;
816 struct cgrp_cset_link *link1, *link2;
817 struct cgroup *cgrp1, *cgrp2;
821 /* See if we reached the end - both lists are equal length. */
822 if (l1 == &cset->cgrp_links) {
823 BUG_ON(l2 != &old_cset->cgrp_links);
826 BUG_ON(l2 == &old_cset->cgrp_links);
828 /* Locate the cgroups associated with these links. */
829 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
830 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
833 /* Hierarchies should be linked in the same order. */
834 BUG_ON(cgrp1->root != cgrp2->root);
837 * If this hierarchy is the hierarchy of the cgroup
838 * that's changing, then we need to check that this
839 * css_set points to the new cgroup; if it's any other
840 * hierarchy, then this css_set should point to the
841 * same cgroup as the old css_set.
843 if (cgrp1->root == new_cgrp->root) {
844 if (cgrp1 != new_cgrp)
855 * find_existing_css_set - init css array and find the matching css_set
856 * @old_cset: the css_set that we're using before the cgroup transition
857 * @cgrp: the cgroup that we're moving into
858 * @template: out param for the new set of csses, should be clear on entry
860 static struct css_set *find_existing_css_set(struct css_set *old_cset,
862 struct cgroup_subsys_state *template[])
864 struct cgroup_root *root = cgrp->root;
865 struct cgroup_subsys *ss;
866 struct css_set *cset;
871 * Build the set of subsystem state objects that we want to see in the
872 * new css_set. while subsystems can change globally, the entries here
873 * won't change, so no need for locking.
875 for_each_subsys(ss, i) {
876 if (root->subsys_mask & (1UL << i)) {
878 * @ss is in this hierarchy, so we want the
879 * effective css from @cgrp.
881 template[i] = cgroup_e_css(cgrp, ss);
884 * @ss is not in this hierarchy, so we don't want
887 template[i] = old_cset->subsys[i];
891 key = css_set_hash(template);
892 hash_for_each_possible(css_set_table, cset, hlist, key) {
893 if (!compare_css_sets(cset, old_cset, cgrp, template))
896 /* This css_set matches what we need */
900 /* No existing cgroup group matched */
904 static void free_cgrp_cset_links(struct list_head *links_to_free)
906 struct cgrp_cset_link *link, *tmp_link;
908 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
909 list_del(&link->cset_link);
915 * allocate_cgrp_cset_links - allocate cgrp_cset_links
916 * @count: the number of links to allocate
917 * @tmp_links: list_head the allocated links are put on
919 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
920 * through ->cset_link. Returns 0 on success or -errno.
922 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
924 struct cgrp_cset_link *link;
927 INIT_LIST_HEAD(tmp_links);
929 for (i = 0; i < count; i++) {
930 link = kzalloc(sizeof(*link), GFP_KERNEL);
932 free_cgrp_cset_links(tmp_links);
935 list_add(&link->cset_link, tmp_links);
941 * link_css_set - a helper function to link a css_set to a cgroup
942 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
943 * @cset: the css_set to be linked
944 * @cgrp: the destination cgroup
946 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
949 struct cgrp_cset_link *link;
951 BUG_ON(list_empty(tmp_links));
953 if (cgroup_on_dfl(cgrp))
954 cset->dfl_cgrp = cgrp;
956 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
961 * Always add links to the tail of the lists so that the lists are
962 * in choronological order.
964 list_move_tail(&link->cset_link, &cgrp->cset_links);
965 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
967 if (cgroup_parent(cgrp))
972 * find_css_set - return a new css_set with one cgroup updated
973 * @old_cset: the baseline css_set
974 * @cgrp: the cgroup to be updated
976 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
977 * substituted into the appropriate hierarchy.
979 static struct css_set *find_css_set(struct css_set *old_cset,
982 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
983 struct css_set *cset;
984 struct list_head tmp_links;
985 struct cgrp_cset_link *link;
986 struct cgroup_subsys *ss;
990 lockdep_assert_held(&cgroup_mutex);
992 /* First see if we already have a cgroup group that matches
994 spin_lock_bh(&css_set_lock);
995 cset = find_existing_css_set(old_cset, cgrp, template);
998 spin_unlock_bh(&css_set_lock);
1003 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
1007 /* Allocate all the cgrp_cset_link objects that we'll need */
1008 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
1013 atomic_set(&cset->refcount, 1);
1014 INIT_LIST_HEAD(&cset->cgrp_links);
1015 INIT_LIST_HEAD(&cset->tasks);
1016 INIT_LIST_HEAD(&cset->mg_tasks);
1017 INIT_LIST_HEAD(&cset->mg_preload_node);
1018 INIT_LIST_HEAD(&cset->mg_node);
1019 INIT_LIST_HEAD(&cset->task_iters);
1020 INIT_HLIST_NODE(&cset->hlist);
1022 /* Copy the set of subsystem state objects generated in
1023 * find_existing_css_set() */
1024 memcpy(cset->subsys, template, sizeof(cset->subsys));
1026 spin_lock_bh(&css_set_lock);
1027 /* Add reference counts and links from the new css_set. */
1028 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
1029 struct cgroup *c = link->cgrp;
1031 if (c->root == cgrp->root)
1033 link_css_set(&tmp_links, cset, c);
1036 BUG_ON(!list_empty(&tmp_links));
1040 /* Add @cset to the hash table */
1041 key = css_set_hash(cset->subsys);
1042 hash_add(css_set_table, &cset->hlist, key);
1044 for_each_subsys(ss, ssid) {
1045 struct cgroup_subsys_state *css = cset->subsys[ssid];
1047 list_add_tail(&cset->e_cset_node[ssid],
1048 &css->cgroup->e_csets[ssid]);
1052 spin_unlock_bh(&css_set_lock);
1057 static struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root)
1059 struct cgroup *root_cgrp = kf_root->kn->priv;
1061 return root_cgrp->root;
1064 static int cgroup_init_root_id(struct cgroup_root *root)
1068 lockdep_assert_held(&cgroup_mutex);
1070 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 0, 0, GFP_KERNEL);
1074 root->hierarchy_id = id;
1078 static void cgroup_exit_root_id(struct cgroup_root *root)
1080 lockdep_assert_held(&cgroup_mutex);
1082 if (root->hierarchy_id) {
1083 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1084 root->hierarchy_id = 0;
1088 static void cgroup_free_root(struct cgroup_root *root)
1091 /* hierarchy ID should already have been released */
1092 WARN_ON_ONCE(root->hierarchy_id);
1094 idr_destroy(&root->cgroup_idr);
1099 static void cgroup_destroy_root(struct cgroup_root *root)
1101 struct cgroup *cgrp = &root->cgrp;
1102 struct cgrp_cset_link *link, *tmp_link;
1104 mutex_lock(&cgroup_mutex);
1106 BUG_ON(atomic_read(&root->nr_cgrps));
1107 BUG_ON(!list_empty(&cgrp->self.children));
1109 /* Rebind all subsystems back to the default hierarchy */
1110 rebind_subsystems(&cgrp_dfl_root, root->subsys_mask);
1113 * Release all the links from cset_links to this hierarchy's
1116 spin_lock_bh(&css_set_lock);
1118 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1119 list_del(&link->cset_link);
1120 list_del(&link->cgrp_link);
1124 spin_unlock_bh(&css_set_lock);
1126 if (!list_empty(&root->root_list)) {
1127 list_del(&root->root_list);
1128 cgroup_root_count--;
1131 cgroup_exit_root_id(root);
1133 mutex_unlock(&cgroup_mutex);
1135 kernfs_destroy_root(root->kf_root);
1136 cgroup_free_root(root);
1139 /* look up cgroup associated with given css_set on the specified hierarchy */
1140 static struct cgroup *cset_cgroup_from_root(struct css_set *cset,
1141 struct cgroup_root *root)
1143 struct cgroup *res = NULL;
1145 lockdep_assert_held(&cgroup_mutex);
1146 lockdep_assert_held(&css_set_lock);
1148 if (cset == &init_css_set) {
1151 struct cgrp_cset_link *link;
1153 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
1154 struct cgroup *c = link->cgrp;
1156 if (c->root == root) {
1168 * Return the cgroup for "task" from the given hierarchy. Must be
1169 * called with cgroup_mutex and css_set_lock held.
1171 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
1172 struct cgroup_root *root)
1175 * No need to lock the task - since we hold cgroup_mutex the
1176 * task can't change groups, so the only thing that can happen
1177 * is that it exits and its css is set back to init_css_set.
1179 return cset_cgroup_from_root(task_css_set(task), root);
1183 * A task must hold cgroup_mutex to modify cgroups.
1185 * Any task can increment and decrement the count field without lock.
1186 * So in general, code holding cgroup_mutex can't rely on the count
1187 * field not changing. However, if the count goes to zero, then only
1188 * cgroup_attach_task() can increment it again. Because a count of zero
1189 * means that no tasks are currently attached, therefore there is no
1190 * way a task attached to that cgroup can fork (the other way to
1191 * increment the count). So code holding cgroup_mutex can safely
1192 * assume that if the count is zero, it will stay zero. Similarly, if
1193 * a task holds cgroup_mutex on a cgroup with zero count, it
1194 * knows that the cgroup won't be removed, as cgroup_rmdir()
1197 * A cgroup can only be deleted if both its 'count' of using tasks
1198 * is zero, and its list of 'children' cgroups is empty. Since all
1199 * tasks in the system use _some_ cgroup, and since there is always at
1200 * least one task in the system (init, pid == 1), therefore, root cgroup
1201 * always has either children cgroups and/or using tasks. So we don't
1202 * need a special hack to ensure that root cgroup cannot be deleted.
1204 * P.S. One more locking exception. RCU is used to guard the
1205 * update of a tasks cgroup pointer by cgroup_attach_task()
1208 static struct kernfs_syscall_ops cgroup_kf_syscall_ops;
1209 static const struct file_operations proc_cgroupstats_operations;
1211 static char *cgroup_file_name(struct cgroup *cgrp, const struct cftype *cft,
1214 struct cgroup_subsys *ss = cft->ss;
1216 if (cft->ss && !(cft->flags & CFTYPE_NO_PREFIX) &&
1217 !(cgrp->root->flags & CGRP_ROOT_NOPREFIX))
1218 snprintf(buf, CGROUP_FILE_NAME_MAX, "%s.%s",
1219 cgroup_on_dfl(cgrp) ? ss->name : ss->legacy_name,
1222 strncpy(buf, cft->name, CGROUP_FILE_NAME_MAX);
1227 * cgroup_file_mode - deduce file mode of a control file
1228 * @cft: the control file in question
1230 * S_IRUGO for read, S_IWUSR for write.
1232 static umode_t cgroup_file_mode(const struct cftype *cft)
1236 if (cft->read_u64 || cft->read_s64 || cft->seq_show)
1239 if (cft->write_u64 || cft->write_s64 || cft->write) {
1240 if (cft->flags & CFTYPE_WORLD_WRITABLE)
1250 * cgroup_calc_child_subsys_mask - calculate child_subsys_mask
1251 * @cgrp: the target cgroup
1252 * @subtree_control: the new subtree_control mask to consider
1254 * On the default hierarchy, a subsystem may request other subsystems to be
1255 * enabled together through its ->depends_on mask. In such cases, more
1256 * subsystems than specified in "cgroup.subtree_control" may be enabled.
1258 * This function calculates which subsystems need to be enabled if
1259 * @subtree_control is to be applied to @cgrp. The returned mask is always
1260 * a superset of @subtree_control and follows the usual hierarchy rules.
1262 static unsigned long cgroup_calc_child_subsys_mask(struct cgroup *cgrp,
1263 unsigned long subtree_control)
1265 struct cgroup *parent = cgroup_parent(cgrp);
1266 unsigned long cur_ss_mask = subtree_control;
1267 struct cgroup_subsys *ss;
1270 lockdep_assert_held(&cgroup_mutex);
1272 if (!cgroup_on_dfl(cgrp))
1276 unsigned long new_ss_mask = cur_ss_mask;
1278 for_each_subsys_which(ss, ssid, &cur_ss_mask)
1279 new_ss_mask |= ss->depends_on;
1282 * Mask out subsystems which aren't available. This can
1283 * happen only if some depended-upon subsystems were bound
1284 * to non-default hierarchies.
1287 new_ss_mask &= parent->child_subsys_mask;
1289 new_ss_mask &= cgrp->root->subsys_mask;
1291 if (new_ss_mask == cur_ss_mask)
1293 cur_ss_mask = new_ss_mask;
1300 * cgroup_refresh_child_subsys_mask - update child_subsys_mask
1301 * @cgrp: the target cgroup
1303 * Update @cgrp->child_subsys_mask according to the current
1304 * @cgrp->subtree_control using cgroup_calc_child_subsys_mask().
1306 static void cgroup_refresh_child_subsys_mask(struct cgroup *cgrp)
1308 cgrp->child_subsys_mask =
1309 cgroup_calc_child_subsys_mask(cgrp, cgrp->subtree_control);
1313 * cgroup_kn_unlock - unlocking helper for cgroup kernfs methods
1314 * @kn: the kernfs_node being serviced
1316 * This helper undoes cgroup_kn_lock_live() and should be invoked before
1317 * the method finishes if locking succeeded. Note that once this function
1318 * returns the cgroup returned by cgroup_kn_lock_live() may become
1319 * inaccessible any time. If the caller intends to continue to access the
1320 * cgroup, it should pin it before invoking this function.
1322 static void cgroup_kn_unlock(struct kernfs_node *kn)
1324 struct cgroup *cgrp;
1326 if (kernfs_type(kn) == KERNFS_DIR)
1329 cgrp = kn->parent->priv;
1331 mutex_unlock(&cgroup_mutex);
1333 kernfs_unbreak_active_protection(kn);
1338 * cgroup_kn_lock_live - locking helper for cgroup kernfs methods
1339 * @kn: the kernfs_node being serviced
1341 * This helper is to be used by a cgroup kernfs method currently servicing
1342 * @kn. It breaks the active protection, performs cgroup locking and
1343 * verifies that the associated cgroup is alive. Returns the cgroup if
1344 * alive; otherwise, %NULL. A successful return should be undone by a
1345 * matching cgroup_kn_unlock() invocation.
1347 * Any cgroup kernfs method implementation which requires locking the
1348 * associated cgroup should use this helper. It avoids nesting cgroup
1349 * locking under kernfs active protection and allows all kernfs operations
1350 * including self-removal.
1352 static struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn)
1354 struct cgroup *cgrp;
1356 if (kernfs_type(kn) == KERNFS_DIR)
1359 cgrp = kn->parent->priv;
1362 * We're gonna grab cgroup_mutex which nests outside kernfs
1363 * active_ref. cgroup liveliness check alone provides enough
1364 * protection against removal. Ensure @cgrp stays accessible and
1365 * break the active_ref protection.
1367 if (!cgroup_tryget(cgrp))
1369 kernfs_break_active_protection(kn);
1371 mutex_lock(&cgroup_mutex);
1373 if (!cgroup_is_dead(cgrp))
1376 cgroup_kn_unlock(kn);
1380 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
1382 char name[CGROUP_FILE_NAME_MAX];
1384 lockdep_assert_held(&cgroup_mutex);
1386 if (cft->file_offset) {
1387 struct cgroup_subsys_state *css = cgroup_css(cgrp, cft->ss);
1388 struct cgroup_file *cfile = (void *)css + cft->file_offset;
1390 spin_lock_irq(&cgroup_file_kn_lock);
1392 spin_unlock_irq(&cgroup_file_kn_lock);
1395 kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
1399 * css_clear_dir - remove subsys files in a cgroup directory
1401 * @cgrp_override: specify if target cgroup is different from css->cgroup
1403 static void css_clear_dir(struct cgroup_subsys_state *css,
1404 struct cgroup *cgrp_override)
1406 struct cgroup *cgrp = cgrp_override ?: css->cgroup;
1407 struct cftype *cfts;
1409 list_for_each_entry(cfts, &css->ss->cfts, node)
1410 cgroup_addrm_files(css, cgrp, cfts, false);
1414 * css_populate_dir - create subsys files in a cgroup directory
1416 * @cgrp_overried: specify if target cgroup is different from css->cgroup
1418 * On failure, no file is added.
1420 static int css_populate_dir(struct cgroup_subsys_state *css,
1421 struct cgroup *cgrp_override)
1423 struct cgroup *cgrp = cgrp_override ?: css->cgroup;
1424 struct cftype *cfts, *failed_cfts;
1428 if (cgroup_on_dfl(cgrp))
1429 cfts = cgroup_dfl_base_files;
1431 cfts = cgroup_legacy_base_files;
1433 return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
1436 list_for_each_entry(cfts, &css->ss->cfts, node) {
1437 ret = cgroup_addrm_files(css, cgrp, cfts, true);
1445 list_for_each_entry(cfts, &css->ss->cfts, node) {
1446 if (cfts == failed_cfts)
1448 cgroup_addrm_files(css, cgrp, cfts, false);
1453 static int rebind_subsystems(struct cgroup_root *dst_root,
1454 unsigned long ss_mask)
1456 struct cgroup *dcgrp = &dst_root->cgrp;
1457 struct cgroup_subsys *ss;
1458 unsigned long tmp_ss_mask;
1461 lockdep_assert_held(&cgroup_mutex);
1463 for_each_subsys_which(ss, ssid, &ss_mask) {
1464 /* if @ss has non-root csses attached to it, can't move */
1465 if (css_next_child(NULL, cgroup_css(&ss->root->cgrp, ss)))
1468 /* can't move between two non-dummy roots either */
1469 if (ss->root != &cgrp_dfl_root && dst_root != &cgrp_dfl_root)
1473 /* skip creating root files on dfl_root for inhibited subsystems */
1474 tmp_ss_mask = ss_mask;
1475 if (dst_root == &cgrp_dfl_root)
1476 tmp_ss_mask &= ~cgrp_dfl_root_inhibit_ss_mask;
1478 for_each_subsys_which(ss, ssid, &tmp_ss_mask) {
1479 struct cgroup *scgrp = &ss->root->cgrp;
1482 ret = css_populate_dir(cgroup_css(scgrp, ss), dcgrp);
1487 * Rebinding back to the default root is not allowed to
1488 * fail. Using both default and non-default roots should
1489 * be rare. Moving subsystems back and forth even more so.
1490 * Just warn about it and continue.
1492 if (dst_root == &cgrp_dfl_root) {
1493 if (cgrp_dfl_root_visible) {
1494 pr_warn("failed to create files (%d) while rebinding 0x%lx to default root\n",
1496 pr_warn("you may retry by moving them to a different hierarchy and unbinding\n");
1501 for_each_subsys_which(ss, tssid, &tmp_ss_mask) {
1504 css_clear_dir(cgroup_css(scgrp, ss), dcgrp);
1510 * Nothing can fail from this point on. Remove files for the
1511 * removed subsystems and rebind each subsystem.
1513 for_each_subsys_which(ss, ssid, &ss_mask) {
1514 struct cgroup_root *src_root = ss->root;
1515 struct cgroup *scgrp = &src_root->cgrp;
1516 struct cgroup_subsys_state *css = cgroup_css(scgrp, ss);
1517 struct css_set *cset;
1519 WARN_ON(!css || cgroup_css(dcgrp, ss));
1521 css_clear_dir(css, NULL);
1523 RCU_INIT_POINTER(scgrp->subsys[ssid], NULL);
1524 rcu_assign_pointer(dcgrp->subsys[ssid], css);
1525 ss->root = dst_root;
1526 css->cgroup = dcgrp;
1528 spin_lock_bh(&css_set_lock);
1529 hash_for_each(css_set_table, i, cset, hlist)
1530 list_move_tail(&cset->e_cset_node[ss->id],
1531 &dcgrp->e_csets[ss->id]);
1532 spin_unlock_bh(&css_set_lock);
1534 src_root->subsys_mask &= ~(1 << ssid);
1535 scgrp->subtree_control &= ~(1 << ssid);
1536 cgroup_refresh_child_subsys_mask(scgrp);
1538 /* default hierarchy doesn't enable controllers by default */
1539 dst_root->subsys_mask |= 1 << ssid;
1540 if (dst_root == &cgrp_dfl_root) {
1541 static_branch_enable(cgroup_subsys_on_dfl_key[ssid]);
1543 dcgrp->subtree_control |= 1 << ssid;
1544 cgroup_refresh_child_subsys_mask(dcgrp);
1545 static_branch_disable(cgroup_subsys_on_dfl_key[ssid]);
1552 kernfs_activate(dcgrp->kn);
1556 static int cgroup_show_options(struct seq_file *seq,
1557 struct kernfs_root *kf_root)
1559 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1560 struct cgroup_subsys *ss;
1563 if (root != &cgrp_dfl_root)
1564 for_each_subsys(ss, ssid)
1565 if (root->subsys_mask & (1 << ssid))
1566 seq_show_option(seq, ss->legacy_name, NULL);
1567 if (root->flags & CGRP_ROOT_NOPREFIX)
1568 seq_puts(seq, ",noprefix");
1569 if (root->flags & CGRP_ROOT_XATTR)
1570 seq_puts(seq, ",xattr");
1572 spin_lock(&release_agent_path_lock);
1573 if (strlen(root->release_agent_path))
1574 seq_show_option(seq, "release_agent",
1575 root->release_agent_path);
1576 spin_unlock(&release_agent_path_lock);
1578 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags))
1579 seq_puts(seq, ",clone_children");
1580 if (strlen(root->name))
1581 seq_show_option(seq, "name", root->name);
1585 struct cgroup_sb_opts {
1586 unsigned long subsys_mask;
1588 char *release_agent;
1589 bool cpuset_clone_children;
1591 /* User explicitly requested empty subsystem */
1595 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1597 char *token, *o = data;
1598 bool all_ss = false, one_ss = false;
1599 unsigned long mask = -1UL;
1600 struct cgroup_subsys *ss;
1604 #ifdef CONFIG_CPUSETS
1605 mask = ~(1U << cpuset_cgrp_id);
1608 memset(opts, 0, sizeof(*opts));
1610 while ((token = strsep(&o, ",")) != NULL) {
1615 if (!strcmp(token, "none")) {
1616 /* Explicitly have no subsystems */
1620 if (!strcmp(token, "all")) {
1621 /* Mutually exclusive option 'all' + subsystem name */
1627 if (!strcmp(token, "noprefix")) {
1628 opts->flags |= CGRP_ROOT_NOPREFIX;
1631 if (!strcmp(token, "clone_children")) {
1632 opts->cpuset_clone_children = true;
1635 if (!strcmp(token, "xattr")) {
1636 opts->flags |= CGRP_ROOT_XATTR;
1639 if (!strncmp(token, "release_agent=", 14)) {
1640 /* Specifying two release agents is forbidden */
1641 if (opts->release_agent)
1643 opts->release_agent =
1644 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1645 if (!opts->release_agent)
1649 if (!strncmp(token, "name=", 5)) {
1650 const char *name = token + 5;
1651 /* Can't specify an empty name */
1654 /* Must match [\w.-]+ */
1655 for (i = 0; i < strlen(name); i++) {
1659 if ((c == '.') || (c == '-') || (c == '_'))
1663 /* Specifying two names is forbidden */
1666 opts->name = kstrndup(name,
1667 MAX_CGROUP_ROOT_NAMELEN - 1,
1675 for_each_subsys(ss, i) {
1676 if (strcmp(token, ss->legacy_name))
1678 if (!cgroup_ssid_enabled(i))
1681 /* Mutually exclusive option 'all' + subsystem name */
1684 opts->subsys_mask |= (1 << i);
1689 if (i == CGROUP_SUBSYS_COUNT)
1694 * If the 'all' option was specified select all the subsystems,
1695 * otherwise if 'none', 'name=' and a subsystem name options were
1696 * not specified, let's default to 'all'
1698 if (all_ss || (!one_ss && !opts->none && !opts->name))
1699 for_each_subsys(ss, i)
1700 if (cgroup_ssid_enabled(i))
1701 opts->subsys_mask |= (1 << i);
1704 * We either have to specify by name or by subsystems. (So all
1705 * empty hierarchies must have a name).
1707 if (!opts->subsys_mask && !opts->name)
1711 * Option noprefix was introduced just for backward compatibility
1712 * with the old cpuset, so we allow noprefix only if mounting just
1713 * the cpuset subsystem.
1715 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1718 /* Can't specify "none" and some subsystems */
1719 if (opts->subsys_mask && opts->none)
1725 static int cgroup_remount(struct kernfs_root *kf_root, int *flags, char *data)
1728 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
1729 struct cgroup_sb_opts opts;
1730 unsigned long added_mask, removed_mask;
1732 if (root == &cgrp_dfl_root) {
1733 pr_err("remount is not allowed\n");
1737 mutex_lock(&cgroup_mutex);
1739 /* See what subsystems are wanted */
1740 ret = parse_cgroupfs_options(data, &opts);
1744 if (opts.subsys_mask != root->subsys_mask || opts.release_agent)
1745 pr_warn("option changes via remount are deprecated (pid=%d comm=%s)\n",
1746 task_tgid_nr(current), current->comm);
1748 added_mask = opts.subsys_mask & ~root->subsys_mask;
1749 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1751 /* Don't allow flags or name to change at remount */
1752 if ((opts.flags ^ root->flags) ||
1753 (opts.name && strcmp(opts.name, root->name))) {
1754 pr_err("option or name mismatch, new: 0x%x \"%s\", old: 0x%x \"%s\"\n",
1755 opts.flags, opts.name ?: "", root->flags, root->name);
1760 /* remounting is not allowed for populated hierarchies */
1761 if (!list_empty(&root->cgrp.self.children)) {
1766 ret = rebind_subsystems(root, added_mask);
1770 rebind_subsystems(&cgrp_dfl_root, removed_mask);
1772 if (opts.release_agent) {
1773 spin_lock(&release_agent_path_lock);
1774 strcpy(root->release_agent_path, opts.release_agent);
1775 spin_unlock(&release_agent_path_lock);
1778 kfree(opts.release_agent);
1780 mutex_unlock(&cgroup_mutex);
1785 * To reduce the fork() overhead for systems that are not actually using
1786 * their cgroups capability, we don't maintain the lists running through
1787 * each css_set to its tasks until we see the list actually used - in other
1788 * words after the first mount.
1790 static bool use_task_css_set_links __read_mostly;
1792 static void cgroup_enable_task_cg_lists(void)
1794 struct task_struct *p, *g;
1796 spin_lock_bh(&css_set_lock);
1798 if (use_task_css_set_links)
1801 use_task_css_set_links = true;
1804 * We need tasklist_lock because RCU is not safe against
1805 * while_each_thread(). Besides, a forking task that has passed
1806 * cgroup_post_fork() without seeing use_task_css_set_links = 1
1807 * is not guaranteed to have its child immediately visible in the
1808 * tasklist if we walk through it with RCU.
1810 read_lock(&tasklist_lock);
1811 do_each_thread(g, p) {
1812 WARN_ON_ONCE(!list_empty(&p->cg_list) ||
1813 task_css_set(p) != &init_css_set);
1816 * We should check if the process is exiting, otherwise
1817 * it will race with cgroup_exit() in that the list
1818 * entry won't be deleted though the process has exited.
1819 * Do it while holding siglock so that we don't end up
1820 * racing against cgroup_exit().
1822 spin_lock_irq(&p->sighand->siglock);
1823 if (!(p->flags & PF_EXITING)) {
1824 struct css_set *cset = task_css_set(p);
1826 if (!css_set_populated(cset))
1827 css_set_update_populated(cset, true);
1828 list_add_tail(&p->cg_list, &cset->tasks);
1831 spin_unlock_irq(&p->sighand->siglock);
1832 } while_each_thread(g, p);
1833 read_unlock(&tasklist_lock);
1835 spin_unlock_bh(&css_set_lock);
1838 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1840 struct cgroup_subsys *ss;
1843 INIT_LIST_HEAD(&cgrp->self.sibling);
1844 INIT_LIST_HEAD(&cgrp->self.children);
1845 INIT_LIST_HEAD(&cgrp->cset_links);
1846 INIT_LIST_HEAD(&cgrp->pidlists);
1847 mutex_init(&cgrp->pidlist_mutex);
1848 cgrp->self.cgroup = cgrp;
1849 cgrp->self.flags |= CSS_ONLINE;
1851 for_each_subsys(ss, ssid)
1852 INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
1854 init_waitqueue_head(&cgrp->offline_waitq);
1855 INIT_WORK(&cgrp->release_agent_work, cgroup_release_agent);
1858 static void init_cgroup_root(struct cgroup_root *root,
1859 struct cgroup_sb_opts *opts)
1861 struct cgroup *cgrp = &root->cgrp;
1863 INIT_LIST_HEAD(&root->root_list);
1864 atomic_set(&root->nr_cgrps, 1);
1866 init_cgroup_housekeeping(cgrp);
1867 idr_init(&root->cgroup_idr);
1869 root->flags = opts->flags;
1870 if (opts->release_agent)
1871 strcpy(root->release_agent_path, opts->release_agent);
1873 strcpy(root->name, opts->name);
1874 if (opts->cpuset_clone_children)
1875 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->cgrp.flags);
1878 static int cgroup_setup_root(struct cgroup_root *root, unsigned long ss_mask)
1880 LIST_HEAD(tmp_links);
1881 struct cgroup *root_cgrp = &root->cgrp;
1882 struct css_set *cset;
1885 lockdep_assert_held(&cgroup_mutex);
1887 ret = cgroup_idr_alloc(&root->cgroup_idr, root_cgrp, 1, 2, GFP_KERNEL);
1890 root_cgrp->id = ret;
1891 root_cgrp->ancestor_ids[0] = ret;
1893 ret = percpu_ref_init(&root_cgrp->self.refcnt, css_release, 0,
1899 * We're accessing css_set_count without locking css_set_lock here,
1900 * but that's OK - it can only be increased by someone holding
1901 * cgroup_lock, and that's us. The worst that can happen is that we
1902 * have some link structures left over
1904 ret = allocate_cgrp_cset_links(css_set_count, &tmp_links);
1908 ret = cgroup_init_root_id(root);
1912 root->kf_root = kernfs_create_root(&cgroup_kf_syscall_ops,
1913 KERNFS_ROOT_CREATE_DEACTIVATED,
1915 if (IS_ERR(root->kf_root)) {
1916 ret = PTR_ERR(root->kf_root);
1919 root_cgrp->kn = root->kf_root->kn;
1921 ret = css_populate_dir(&root_cgrp->self, NULL);
1925 ret = rebind_subsystems(root, ss_mask);
1930 * There must be no failure case after here, since rebinding takes
1931 * care of subsystems' refcounts, which are explicitly dropped in
1932 * the failure exit path.
1934 list_add(&root->root_list, &cgroup_roots);
1935 cgroup_root_count++;
1938 * Link the root cgroup in this hierarchy into all the css_set
1941 spin_lock_bh(&css_set_lock);
1942 hash_for_each(css_set_table, i, cset, hlist) {
1943 link_css_set(&tmp_links, cset, root_cgrp);
1944 if (css_set_populated(cset))
1945 cgroup_update_populated(root_cgrp, true);
1947 spin_unlock_bh(&css_set_lock);
1949 BUG_ON(!list_empty(&root_cgrp->self.children));
1950 BUG_ON(atomic_read(&root->nr_cgrps) != 1);
1952 kernfs_activate(root_cgrp->kn);
1957 kernfs_destroy_root(root->kf_root);
1958 root->kf_root = NULL;
1960 cgroup_exit_root_id(root);
1962 percpu_ref_exit(&root_cgrp->self.refcnt);
1964 free_cgrp_cset_links(&tmp_links);
1968 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1969 int flags, const char *unused_dev_name,
1972 bool is_v2 = fs_type == &cgroup2_fs_type;
1973 struct super_block *pinned_sb = NULL;
1974 struct cgroup_subsys *ss;
1975 struct cgroup_root *root;
1976 struct cgroup_sb_opts opts;
1977 struct dentry *dentry;
1983 * The first time anyone tries to mount a cgroup, enable the list
1984 * linking each css_set to its tasks and fix up all existing tasks.
1986 if (!use_task_css_set_links)
1987 cgroup_enable_task_cg_lists();
1991 pr_err("cgroup2: unknown option \"%s\"\n", (char *)data);
1992 return ERR_PTR(-EINVAL);
1994 cgrp_dfl_root_visible = true;
1995 root = &cgrp_dfl_root;
1996 cgroup_get(&root->cgrp);
2000 mutex_lock(&cgroup_mutex);
2002 /* First find the desired set of subsystems */
2003 ret = parse_cgroupfs_options(data, &opts);
2008 * Destruction of cgroup root is asynchronous, so subsystems may
2009 * still be dying after the previous unmount. Let's drain the
2010 * dying subsystems. We just need to ensure that the ones
2011 * unmounted previously finish dying and don't care about new ones
2012 * starting. Testing ref liveliness is good enough.
2014 for_each_subsys(ss, i) {
2015 if (!(opts.subsys_mask & (1 << i)) ||
2016 ss->root == &cgrp_dfl_root)
2019 if (!percpu_ref_tryget_live(&ss->root->cgrp.self.refcnt)) {
2020 mutex_unlock(&cgroup_mutex);
2022 ret = restart_syscall();
2025 cgroup_put(&ss->root->cgrp);
2028 for_each_root(root) {
2029 bool name_match = false;
2031 if (root == &cgrp_dfl_root)
2035 * If we asked for a name then it must match. Also, if
2036 * name matches but sybsys_mask doesn't, we should fail.
2037 * Remember whether name matched.
2040 if (strcmp(opts.name, root->name))
2046 * If we asked for subsystems (or explicitly for no
2047 * subsystems) then they must match.
2049 if ((opts.subsys_mask || opts.none) &&
2050 (opts.subsys_mask != root->subsys_mask)) {
2057 if (root->flags ^ opts.flags)
2058 pr_warn("new mount options do not match the existing superblock, will be ignored\n");
2061 * We want to reuse @root whose lifetime is governed by its
2062 * ->cgrp. Let's check whether @root is alive and keep it
2063 * that way. As cgroup_kill_sb() can happen anytime, we
2064 * want to block it by pinning the sb so that @root doesn't
2065 * get killed before mount is complete.
2067 * With the sb pinned, tryget_live can reliably indicate
2068 * whether @root can be reused. If it's being killed,
2069 * drain it. We can use wait_queue for the wait but this
2070 * path is super cold. Let's just sleep a bit and retry.
2072 pinned_sb = kernfs_pin_sb(root->kf_root, NULL);
2073 if (IS_ERR(pinned_sb) ||
2074 !percpu_ref_tryget_live(&root->cgrp.self.refcnt)) {
2075 mutex_unlock(&cgroup_mutex);
2076 if (!IS_ERR_OR_NULL(pinned_sb))
2077 deactivate_super(pinned_sb);
2079 ret = restart_syscall();
2088 * No such thing, create a new one. name= matching without subsys
2089 * specification is allowed for already existing hierarchies but we
2090 * can't create new one without subsys specification.
2092 if (!opts.subsys_mask && !opts.none) {
2097 root = kzalloc(sizeof(*root), GFP_KERNEL);
2103 init_cgroup_root(root, &opts);
2105 ret = cgroup_setup_root(root, opts.subsys_mask);
2107 cgroup_free_root(root);
2110 mutex_unlock(&cgroup_mutex);
2112 kfree(opts.release_agent);
2116 return ERR_PTR(ret);
2118 dentry = kernfs_mount(fs_type, flags, root->kf_root,
2119 is_v2 ? CGROUP2_SUPER_MAGIC : CGROUP_SUPER_MAGIC,
2121 if (IS_ERR(dentry) || !new_sb)
2122 cgroup_put(&root->cgrp);
2125 * If @pinned_sb, we're reusing an existing root and holding an
2126 * extra ref on its sb. Mount is complete. Put the extra ref.
2130 deactivate_super(pinned_sb);
2136 static void cgroup_kill_sb(struct super_block *sb)
2138 struct kernfs_root *kf_root = kernfs_root_from_sb(sb);
2139 struct cgroup_root *root = cgroup_root_from_kf(kf_root);
2142 * If @root doesn't have any mounts or children, start killing it.
2143 * This prevents new mounts by disabling percpu_ref_tryget_live().
2144 * cgroup_mount() may wait for @root's release.
2146 * And don't kill the default root.
2148 if (!list_empty(&root->cgrp.self.children) ||
2149 root == &cgrp_dfl_root)
2150 cgroup_put(&root->cgrp);
2152 percpu_ref_kill(&root->cgrp.self.refcnt);
2157 static struct file_system_type cgroup_fs_type = {
2159 .mount = cgroup_mount,
2160 .kill_sb = cgroup_kill_sb,
2163 static struct file_system_type cgroup2_fs_type = {
2165 .mount = cgroup_mount,
2166 .kill_sb = cgroup_kill_sb,
2170 * task_cgroup_path - cgroup path of a task in the first cgroup hierarchy
2171 * @task: target task
2172 * @buf: the buffer to write the path into
2173 * @buflen: the length of the buffer
2175 * Determine @task's cgroup on the first (the one with the lowest non-zero
2176 * hierarchy_id) cgroup hierarchy and copy its path into @buf. This
2177 * function grabs cgroup_mutex and shouldn't be used inside locks used by
2178 * cgroup controller callbacks.
2180 * Return value is the same as kernfs_path().
2182 char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen)
2184 struct cgroup_root *root;
2185 struct cgroup *cgrp;
2186 int hierarchy_id = 1;
2189 mutex_lock(&cgroup_mutex);
2190 spin_lock_bh(&css_set_lock);
2192 root = idr_get_next(&cgroup_hierarchy_idr, &hierarchy_id);
2195 cgrp = task_cgroup_from_root(task, root);
2196 path = cgroup_path(cgrp, buf, buflen);
2198 /* if no hierarchy exists, everyone is in "/" */
2199 if (strlcpy(buf, "/", buflen) < buflen)
2203 spin_unlock_bh(&css_set_lock);
2204 mutex_unlock(&cgroup_mutex);
2207 EXPORT_SYMBOL_GPL(task_cgroup_path);
2209 /* used to track tasks and other necessary states during migration */
2210 struct cgroup_taskset {
2211 /* the src and dst cset list running through cset->mg_node */
2212 struct list_head src_csets;
2213 struct list_head dst_csets;
2215 /* the subsys currently being processed */
2219 * Fields for cgroup_taskset_*() iteration.
2221 * Before migration is committed, the target migration tasks are on
2222 * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
2223 * the csets on ->dst_csets. ->csets point to either ->src_csets
2224 * or ->dst_csets depending on whether migration is committed.
2226 * ->cur_csets and ->cur_task point to the current task position
2229 struct list_head *csets;
2230 struct css_set *cur_cset;
2231 struct task_struct *cur_task;
2234 #define CGROUP_TASKSET_INIT(tset) (struct cgroup_taskset){ \
2235 .src_csets = LIST_HEAD_INIT(tset.src_csets), \
2236 .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \
2237 .csets = &tset.src_csets, \
2241 * cgroup_taskset_add - try to add a migration target task to a taskset
2242 * @task: target task
2243 * @tset: target taskset
2245 * Add @task, which is a migration target, to @tset. This function becomes
2246 * noop if @task doesn't need to be migrated. @task's css_set should have
2247 * been added as a migration source and @task->cg_list will be moved from
2248 * the css_set's tasks list to mg_tasks one.
2250 static void cgroup_taskset_add(struct task_struct *task,
2251 struct cgroup_taskset *tset)
2253 struct css_set *cset;
2255 lockdep_assert_held(&css_set_lock);
2257 /* @task either already exited or can't exit until the end */
2258 if (task->flags & PF_EXITING)
2261 /* leave @task alone if post_fork() hasn't linked it yet */
2262 if (list_empty(&task->cg_list))
2265 cset = task_css_set(task);
2266 if (!cset->mg_src_cgrp)
2269 list_move_tail(&task->cg_list, &cset->mg_tasks);
2270 if (list_empty(&cset->mg_node))
2271 list_add_tail(&cset->mg_node, &tset->src_csets);
2272 if (list_empty(&cset->mg_dst_cset->mg_node))
2273 list_move_tail(&cset->mg_dst_cset->mg_node,
2278 * cgroup_taskset_first - reset taskset and return the first task
2279 * @tset: taskset of interest
2280 * @dst_cssp: output variable for the destination css
2282 * @tset iteration is initialized and the first task is returned.
2284 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset,
2285 struct cgroup_subsys_state **dst_cssp)
2287 tset->cur_cset = list_first_entry(tset->csets, struct css_set, mg_node);
2288 tset->cur_task = NULL;
2290 return cgroup_taskset_next(tset, dst_cssp);
2294 * cgroup_taskset_next - iterate to the next task in taskset
2295 * @tset: taskset of interest
2296 * @dst_cssp: output variable for the destination css
2298 * Return the next task in @tset. Iteration must have been initialized
2299 * with cgroup_taskset_first().
2301 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset,
2302 struct cgroup_subsys_state **dst_cssp)
2304 struct css_set *cset = tset->cur_cset;
2305 struct task_struct *task = tset->cur_task;
2307 while (&cset->mg_node != tset->csets) {
2309 task = list_first_entry(&cset->mg_tasks,
2310 struct task_struct, cg_list);
2312 task = list_next_entry(task, cg_list);
2314 if (&task->cg_list != &cset->mg_tasks) {
2315 tset->cur_cset = cset;
2316 tset->cur_task = task;
2319 * This function may be called both before and
2320 * after cgroup_taskset_migrate(). The two cases
2321 * can be distinguished by looking at whether @cset
2322 * has its ->mg_dst_cset set.
2324 if (cset->mg_dst_cset)
2325 *dst_cssp = cset->mg_dst_cset->subsys[tset->ssid];
2327 *dst_cssp = cset->subsys[tset->ssid];
2332 cset = list_next_entry(cset, mg_node);
2340 * cgroup_taskset_migrate - migrate a taskset to a cgroup
2341 * @tset: taget taskset
2342 * @dst_cgrp: destination cgroup
2344 * Migrate tasks in @tset to @dst_cgrp. This function fails iff one of the
2345 * ->can_attach callbacks fails and guarantees that either all or none of
2346 * the tasks in @tset are migrated. @tset is consumed regardless of
2349 static int cgroup_taskset_migrate(struct cgroup_taskset *tset,
2350 struct cgroup *dst_cgrp)
2352 struct cgroup_subsys_state *css, *failed_css = NULL;
2353 struct task_struct *task, *tmp_task;
2354 struct css_set *cset, *tmp_cset;
2357 /* methods shouldn't be called if no task is actually migrating */
2358 if (list_empty(&tset->src_csets))
2361 /* check that we can legitimately attach to the cgroup */
2362 for_each_e_css(css, i, dst_cgrp) {
2363 if (css->ss->can_attach) {
2365 ret = css->ss->can_attach(tset);
2368 goto out_cancel_attach;
2374 * Now that we're guaranteed success, proceed to move all tasks to
2375 * the new cgroup. There are no failure cases after here, so this
2376 * is the commit point.
2378 spin_lock_bh(&css_set_lock);
2379 list_for_each_entry(cset, &tset->src_csets, mg_node) {
2380 list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list) {
2381 struct css_set *from_cset = task_css_set(task);
2382 struct css_set *to_cset = cset->mg_dst_cset;
2384 get_css_set(to_cset);
2385 css_set_move_task(task, from_cset, to_cset, true);
2386 put_css_set_locked(from_cset);
2389 spin_unlock_bh(&css_set_lock);
2392 * Migration is committed, all target tasks are now on dst_csets.
2393 * Nothing is sensitive to fork() after this point. Notify
2394 * controllers that migration is complete.
2396 tset->csets = &tset->dst_csets;
2398 for_each_e_css(css, i, dst_cgrp) {
2399 if (css->ss->attach) {
2401 css->ss->attach(tset);
2406 goto out_release_tset;
2409 for_each_e_css(css, i, dst_cgrp) {
2410 if (css == failed_css)
2412 if (css->ss->cancel_attach) {
2414 css->ss->cancel_attach(tset);
2418 spin_lock_bh(&css_set_lock);
2419 list_splice_init(&tset->dst_csets, &tset->src_csets);
2420 list_for_each_entry_safe(cset, tmp_cset, &tset->src_csets, mg_node) {
2421 list_splice_tail_init(&cset->mg_tasks, &cset->tasks);
2422 list_del_init(&cset->mg_node);
2424 spin_unlock_bh(&css_set_lock);
2429 * cgroup_migrate_finish - cleanup after attach
2430 * @preloaded_csets: list of preloaded css_sets
2432 * Undo cgroup_migrate_add_src() and cgroup_migrate_prepare_dst(). See
2433 * those functions for details.
2435 static void cgroup_migrate_finish(struct list_head *preloaded_csets)
2437 struct css_set *cset, *tmp_cset;
2439 lockdep_assert_held(&cgroup_mutex);
2441 spin_lock_bh(&css_set_lock);
2442 list_for_each_entry_safe(cset, tmp_cset, preloaded_csets, mg_preload_node) {
2443 cset->mg_src_cgrp = NULL;
2444 cset->mg_dst_cset = NULL;
2445 list_del_init(&cset->mg_preload_node);
2446 put_css_set_locked(cset);
2448 spin_unlock_bh(&css_set_lock);
2452 * cgroup_migrate_add_src - add a migration source css_set
2453 * @src_cset: the source css_set to add
2454 * @dst_cgrp: the destination cgroup
2455 * @preloaded_csets: list of preloaded css_sets
2457 * Tasks belonging to @src_cset are about to be migrated to @dst_cgrp. Pin
2458 * @src_cset and add it to @preloaded_csets, which should later be cleaned
2459 * up by cgroup_migrate_finish().
2461 * This function may be called without holding cgroup_threadgroup_rwsem
2462 * even if the target is a process. Threads may be created and destroyed
2463 * but as long as cgroup_mutex is not dropped, no new css_set can be put
2464 * into play and the preloaded css_sets are guaranteed to cover all
2467 static void cgroup_migrate_add_src(struct css_set *src_cset,
2468 struct cgroup *dst_cgrp,
2469 struct list_head *preloaded_csets)
2471 struct cgroup *src_cgrp;
2473 lockdep_assert_held(&cgroup_mutex);
2474 lockdep_assert_held(&css_set_lock);
2476 src_cgrp = cset_cgroup_from_root(src_cset, dst_cgrp->root);
2478 if (!list_empty(&src_cset->mg_preload_node))
2481 WARN_ON(src_cset->mg_src_cgrp);
2482 WARN_ON(!list_empty(&src_cset->mg_tasks));
2483 WARN_ON(!list_empty(&src_cset->mg_node));
2485 src_cset->mg_src_cgrp = src_cgrp;
2486 get_css_set(src_cset);
2487 list_add(&src_cset->mg_preload_node, preloaded_csets);
2491 * cgroup_migrate_prepare_dst - prepare destination css_sets for migration
2492 * @dst_cgrp: the destination cgroup (may be %NULL)
2493 * @preloaded_csets: list of preloaded source css_sets
2495 * Tasks are about to be moved to @dst_cgrp and all the source css_sets
2496 * have been preloaded to @preloaded_csets. This function looks up and
2497 * pins all destination css_sets, links each to its source, and append them
2498 * to @preloaded_csets. If @dst_cgrp is %NULL, the destination of each
2499 * source css_set is assumed to be its cgroup on the default hierarchy.
2501 * This function must be called after cgroup_migrate_add_src() has been
2502 * called on each migration source css_set. After migration is performed
2503 * using cgroup_migrate(), cgroup_migrate_finish() must be called on
2506 static int cgroup_migrate_prepare_dst(struct cgroup *dst_cgrp,
2507 struct list_head *preloaded_csets)
2510 struct css_set *src_cset, *tmp_cset;
2512 lockdep_assert_held(&cgroup_mutex);
2515 * Except for the root, child_subsys_mask must be zero for a cgroup
2516 * with tasks so that child cgroups don't compete against tasks.
2518 if (dst_cgrp && cgroup_on_dfl(dst_cgrp) && cgroup_parent(dst_cgrp) &&
2519 dst_cgrp->child_subsys_mask)
2522 /* look up the dst cset for each src cset and link it to src */
2523 list_for_each_entry_safe(src_cset, tmp_cset, preloaded_csets, mg_preload_node) {
2524 struct css_set *dst_cset;
2526 dst_cset = find_css_set(src_cset,
2527 dst_cgrp ?: src_cset->dfl_cgrp);
2531 WARN_ON_ONCE(src_cset->mg_dst_cset || dst_cset->mg_dst_cset);
2534 * If src cset equals dst, it's noop. Drop the src.
2535 * cgroup_migrate() will skip the cset too. Note that we
2536 * can't handle src == dst as some nodes are used by both.
2538 if (src_cset == dst_cset) {
2539 src_cset->mg_src_cgrp = NULL;
2540 list_del_init(&src_cset->mg_preload_node);
2541 put_css_set(src_cset);
2542 put_css_set(dst_cset);
2546 src_cset->mg_dst_cset = dst_cset;
2548 if (list_empty(&dst_cset->mg_preload_node))
2549 list_add(&dst_cset->mg_preload_node, &csets);
2551 put_css_set(dst_cset);
2554 list_splice_tail(&csets, preloaded_csets);
2557 cgroup_migrate_finish(&csets);
2562 * cgroup_migrate - migrate a process or task to a cgroup
2563 * @leader: the leader of the process or the task to migrate
2564 * @threadgroup: whether @leader points to the whole process or a single task
2565 * @cgrp: the destination cgroup
2567 * Migrate a process or task denoted by @leader to @cgrp. If migrating a
2568 * process, the caller must be holding cgroup_threadgroup_rwsem. The
2569 * caller is also responsible for invoking cgroup_migrate_add_src() and
2570 * cgroup_migrate_prepare_dst() on the targets before invoking this
2571 * function and following up with cgroup_migrate_finish().
2573 * As long as a controller's ->can_attach() doesn't fail, this function is
2574 * guaranteed to succeed. This means that, excluding ->can_attach()
2575 * failure, when migrating multiple targets, the success or failure can be
2576 * decided for all targets by invoking group_migrate_prepare_dst() before
2577 * actually starting migrating.
2579 static int cgroup_migrate(struct task_struct *leader, bool threadgroup,
2580 struct cgroup *cgrp)
2582 struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
2583 struct task_struct *task;
2586 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2587 * already PF_EXITING could be freed from underneath us unless we
2588 * take an rcu_read_lock.
2590 spin_lock_bh(&css_set_lock);
2594 cgroup_taskset_add(task, &tset);
2597 } while_each_thread(leader, task);
2599 spin_unlock_bh(&css_set_lock);
2601 return cgroup_taskset_migrate(&tset, cgrp);
2605 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
2606 * @dst_cgrp: the cgroup to attach to
2607 * @leader: the task or the leader of the threadgroup to be attached
2608 * @threadgroup: attach the whole threadgroup?
2610 * Call holding cgroup_mutex and cgroup_threadgroup_rwsem.
2612 static int cgroup_attach_task(struct cgroup *dst_cgrp,
2613 struct task_struct *leader, bool threadgroup)
2615 LIST_HEAD(preloaded_csets);
2616 struct task_struct *task;
2619 /* look up all src csets */
2620 spin_lock_bh(&css_set_lock);
2624 cgroup_migrate_add_src(task_css_set(task), dst_cgrp,
2628 } while_each_thread(leader, task);
2630 spin_unlock_bh(&css_set_lock);
2632 /* prepare dst csets and commit */
2633 ret = cgroup_migrate_prepare_dst(dst_cgrp, &preloaded_csets);
2635 ret = cgroup_migrate(leader, threadgroup, dst_cgrp);
2637 cgroup_migrate_finish(&preloaded_csets);
2641 static int cgroup_procs_write_permission(struct task_struct *task,
2642 struct cgroup *dst_cgrp,
2643 struct kernfs_open_file *of)
2645 const struct cred *cred = current_cred();
2646 const struct cred *tcred = get_task_cred(task);
2650 * even if we're attaching all tasks in the thread group, we only
2651 * need to check permissions on one of them.
2653 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2654 !uid_eq(cred->euid, tcred->uid) &&
2655 !uid_eq(cred->euid, tcred->suid))
2658 if (!ret && cgroup_on_dfl(dst_cgrp)) {
2659 struct super_block *sb = of->file->f_path.dentry->d_sb;
2660 struct cgroup *cgrp;
2661 struct inode *inode;
2663 spin_lock_bh(&css_set_lock);
2664 cgrp = task_cgroup_from_root(task, &cgrp_dfl_root);
2665 spin_unlock_bh(&css_set_lock);
2667 while (!cgroup_is_descendant(dst_cgrp, cgrp))
2668 cgrp = cgroup_parent(cgrp);
2671 inode = kernfs_get_inode(sb, cgrp->procs_file.kn);
2673 ret = inode_permission(inode, MAY_WRITE);
2683 * Find the task_struct of the task to attach by vpid and pass it along to the
2684 * function to attach either it or all tasks in its threadgroup. Will lock
2685 * cgroup_mutex and threadgroup.
2687 static ssize_t __cgroup_procs_write(struct kernfs_open_file *of, char *buf,
2688 size_t nbytes, loff_t off, bool threadgroup)
2690 struct task_struct *tsk;
2691 struct cgroup *cgrp;
2695 if (kstrtoint(strstrip(buf), 0, &pid) || pid < 0)
2698 cgrp = cgroup_kn_lock_live(of->kn);
2702 percpu_down_write(&cgroup_threadgroup_rwsem);
2705 tsk = find_task_by_vpid(pid);
2708 goto out_unlock_rcu;
2715 tsk = tsk->group_leader;
2718 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2719 * trapped in a cpuset, or RT worker may be born in a cgroup
2720 * with no rt_runtime allocated. Just say no.
2722 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
2724 goto out_unlock_rcu;
2727 get_task_struct(tsk);
2730 ret = cgroup_procs_write_permission(tsk, cgrp, of);
2732 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2734 put_task_struct(tsk);
2735 goto out_unlock_threadgroup;
2739 out_unlock_threadgroup:
2740 percpu_up_write(&cgroup_threadgroup_rwsem);
2741 cgroup_kn_unlock(of->kn);
2742 return ret ?: nbytes;
2746 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2747 * @from: attach to all cgroups of a given task
2748 * @tsk: the task to be attached
2750 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2752 struct cgroup_root *root;
2755 mutex_lock(&cgroup_mutex);
2756 for_each_root(root) {
2757 struct cgroup *from_cgrp;
2759 if (root == &cgrp_dfl_root)
2762 spin_lock_bh(&css_set_lock);
2763 from_cgrp = task_cgroup_from_root(from, root);
2764 spin_unlock_bh(&css_set_lock);
2766 retval = cgroup_attach_task(from_cgrp, tsk, false);
2770 mutex_unlock(&cgroup_mutex);
2774 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2776 static ssize_t cgroup_tasks_write(struct kernfs_open_file *of,
2777 char *buf, size_t nbytes, loff_t off)
2779 return __cgroup_procs_write(of, buf, nbytes, off, false);
2782 static ssize_t cgroup_procs_write(struct kernfs_open_file *of,
2783 char *buf, size_t nbytes, loff_t off)
2785 return __cgroup_procs_write(of, buf, nbytes, off, true);
2788 static ssize_t cgroup_release_agent_write(struct kernfs_open_file *of,
2789 char *buf, size_t nbytes, loff_t off)
2791 struct cgroup *cgrp;
2793 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2795 cgrp = cgroup_kn_lock_live(of->kn);
2798 spin_lock(&release_agent_path_lock);
2799 strlcpy(cgrp->root->release_agent_path, strstrip(buf),
2800 sizeof(cgrp->root->release_agent_path));
2801 spin_unlock(&release_agent_path_lock);
2802 cgroup_kn_unlock(of->kn);
2806 static int cgroup_release_agent_show(struct seq_file *seq, void *v)
2808 struct cgroup *cgrp = seq_css(seq)->cgroup;
2810 spin_lock(&release_agent_path_lock);
2811 seq_puts(seq, cgrp->root->release_agent_path);
2812 spin_unlock(&release_agent_path_lock);
2813 seq_putc(seq, '\n');
2817 static int cgroup_sane_behavior_show(struct seq_file *seq, void *v)
2819 seq_puts(seq, "0\n");
2823 static void cgroup_print_ss_mask(struct seq_file *seq, unsigned long ss_mask)
2825 struct cgroup_subsys *ss;
2826 bool printed = false;
2829 for_each_subsys_which(ss, ssid, &ss_mask) {
2832 seq_printf(seq, "%s", ss->name);
2836 seq_putc(seq, '\n');
2839 /* show controllers which are currently attached to the default hierarchy */
2840 static int cgroup_root_controllers_show(struct seq_file *seq, void *v)
2842 struct cgroup *cgrp = seq_css(seq)->cgroup;
2844 cgroup_print_ss_mask(seq, cgrp->root->subsys_mask &
2845 ~cgrp_dfl_root_inhibit_ss_mask);
2849 /* show controllers which are enabled from the parent */
2850 static int cgroup_controllers_show(struct seq_file *seq, void *v)
2852 struct cgroup *cgrp = seq_css(seq)->cgroup;
2854 cgroup_print_ss_mask(seq, cgroup_parent(cgrp)->subtree_control);
2858 /* show controllers which are enabled for a given cgroup's children */
2859 static int cgroup_subtree_control_show(struct seq_file *seq, void *v)
2861 struct cgroup *cgrp = seq_css(seq)->cgroup;
2863 cgroup_print_ss_mask(seq, cgrp->subtree_control);
2868 * cgroup_update_dfl_csses - update css assoc of a subtree in default hierarchy
2869 * @cgrp: root of the subtree to update csses for
2871 * @cgrp's child_subsys_mask has changed and its subtree's (self excluded)
2872 * css associations need to be updated accordingly. This function looks up
2873 * all css_sets which are attached to the subtree, creates the matching
2874 * updated css_sets and migrates the tasks to the new ones.
2876 static int cgroup_update_dfl_csses(struct cgroup *cgrp)
2878 LIST_HEAD(preloaded_csets);
2879 struct cgroup_taskset tset = CGROUP_TASKSET_INIT(tset);
2880 struct cgroup_subsys_state *css;
2881 struct css_set *src_cset;
2884 lockdep_assert_held(&cgroup_mutex);
2886 percpu_down_write(&cgroup_threadgroup_rwsem);
2888 /* look up all csses currently attached to @cgrp's subtree */
2889 spin_lock_bh(&css_set_lock);
2890 css_for_each_descendant_pre(css, cgroup_css(cgrp, NULL)) {
2891 struct cgrp_cset_link *link;
2893 /* self is not affected by child_subsys_mask change */
2894 if (css->cgroup == cgrp)
2897 list_for_each_entry(link, &css->cgroup->cset_links, cset_link)
2898 cgroup_migrate_add_src(link->cset, cgrp,
2901 spin_unlock_bh(&css_set_lock);
2903 /* NULL dst indicates self on default hierarchy */
2904 ret = cgroup_migrate_prepare_dst(NULL, &preloaded_csets);
2908 spin_lock_bh(&css_set_lock);
2909 list_for_each_entry(src_cset, &preloaded_csets, mg_preload_node) {
2910 struct task_struct *task, *ntask;
2912 /* src_csets precede dst_csets, break on the first dst_cset */
2913 if (!src_cset->mg_src_cgrp)
2916 /* all tasks in src_csets need to be migrated */
2917 list_for_each_entry_safe(task, ntask, &src_cset->tasks, cg_list)
2918 cgroup_taskset_add(task, &tset);
2920 spin_unlock_bh(&css_set_lock);
2922 ret = cgroup_taskset_migrate(&tset, cgrp);
2924 cgroup_migrate_finish(&preloaded_csets);
2925 percpu_up_write(&cgroup_threadgroup_rwsem);
2929 /* change the enabled child controllers for a cgroup in the default hierarchy */
2930 static ssize_t cgroup_subtree_control_write(struct kernfs_open_file *of,
2931 char *buf, size_t nbytes,
2934 unsigned long enable = 0, disable = 0;
2935 unsigned long css_enable, css_disable, old_sc, new_sc, old_ss, new_ss;
2936 struct cgroup *cgrp, *child;
2937 struct cgroup_subsys *ss;
2942 * Parse input - space separated list of subsystem names prefixed
2943 * with either + or -.
2945 buf = strstrip(buf);
2946 while ((tok = strsep(&buf, " "))) {
2947 unsigned long tmp_ss_mask = ~cgrp_dfl_root_inhibit_ss_mask;
2951 for_each_subsys_which(ss, ssid, &tmp_ss_mask) {
2952 if (!cgroup_ssid_enabled(ssid) ||
2953 strcmp(tok + 1, ss->name))
2957 enable |= 1 << ssid;
2958 disable &= ~(1 << ssid);
2959 } else if (*tok == '-') {
2960 disable |= 1 << ssid;
2961 enable &= ~(1 << ssid);
2967 if (ssid == CGROUP_SUBSYS_COUNT)
2971 cgrp = cgroup_kn_lock_live(of->kn);
2975 for_each_subsys(ss, ssid) {
2976 if (enable & (1 << ssid)) {
2977 if (cgrp->subtree_control & (1 << ssid)) {
2978 enable &= ~(1 << ssid);
2982 /* unavailable or not enabled on the parent? */
2983 if (!(cgrp_dfl_root.subsys_mask & (1 << ssid)) ||
2984 (cgroup_parent(cgrp) &&
2985 !(cgroup_parent(cgrp)->subtree_control & (1 << ssid)))) {
2989 } else if (disable & (1 << ssid)) {
2990 if (!(cgrp->subtree_control & (1 << ssid))) {
2991 disable &= ~(1 << ssid);
2995 /* a child has it enabled? */
2996 cgroup_for_each_live_child(child, cgrp) {
2997 if (child->subtree_control & (1 << ssid)) {
3005 if (!enable && !disable) {
3011 * Except for the root, subtree_control must be zero for a cgroup
3012 * with tasks so that child cgroups don't compete against tasks.
3014 if (enable && cgroup_parent(cgrp) && !list_empty(&cgrp->cset_links)) {
3020 * Update subsys masks and calculate what needs to be done. More
3021 * subsystems than specified may need to be enabled or disabled
3022 * depending on subsystem dependencies.
3024 old_sc = cgrp->subtree_control;
3025 old_ss = cgrp->child_subsys_mask;
3026 new_sc = (old_sc | enable) & ~disable;
3027 new_ss = cgroup_calc_child_subsys_mask(cgrp, new_sc);
3029 css_enable = ~old_ss & new_ss;
3030 css_disable = old_ss & ~new_ss;
3031 enable |= css_enable;
3032 disable |= css_disable;
3035 * Because css offlining is asynchronous, userland might try to
3036 * re-enable the same controller while the previous instance is
3037 * still around. In such cases, wait till it's gone using
3040 for_each_subsys_which(ss, ssid, &css_enable) {
3041 cgroup_for_each_live_child(child, cgrp) {
3044 if (!cgroup_css(child, ss))
3048 prepare_to_wait(&child->offline_waitq, &wait,
3049 TASK_UNINTERRUPTIBLE);
3050 cgroup_kn_unlock(of->kn);
3052 finish_wait(&child->offline_waitq, &wait);
3055 return restart_syscall();
3059 cgrp->subtree_control = new_sc;
3060 cgrp->child_subsys_mask = new_ss;
3063 * Create new csses or make the existing ones visible. A css is
3064 * created invisible if it's being implicitly enabled through
3065 * dependency. An invisible css is made visible when the userland
3066 * explicitly enables it.
3068 for_each_subsys(ss, ssid) {
3069 if (!(enable & (1 << ssid)))
3072 cgroup_for_each_live_child(child, cgrp) {
3073 if (css_enable & (1 << ssid))
3074 ret = create_css(child, ss,
3075 cgrp->subtree_control & (1 << ssid));
3077 ret = css_populate_dir(cgroup_css(child, ss),
3085 * At this point, cgroup_e_css() results reflect the new csses
3086 * making the following cgroup_update_dfl_csses() properly update
3087 * css associations of all tasks in the subtree.
3089 ret = cgroup_update_dfl_csses(cgrp);
3094 * All tasks are migrated out of disabled csses. Kill or hide
3095 * them. A css is hidden when the userland requests it to be
3096 * disabled while other subsystems are still depending on it. The
3097 * css must not actively control resources and be in the vanilla
3098 * state if it's made visible again later. Controllers which may
3099 * be depended upon should provide ->css_reset() for this purpose.
3101 for_each_subsys(ss, ssid) {
3102 if (!(disable & (1 << ssid)))
3105 cgroup_for_each_live_child(child, cgrp) {
3106 struct cgroup_subsys_state *css = cgroup_css(child, ss);
3108 if (css_disable & (1 << ssid)) {
3111 css_clear_dir(css, NULL);
3119 * The effective csses of all the descendants (excluding @cgrp) may
3120 * have changed. Subsystems can optionally subscribe to this event
3121 * by implementing ->css_e_css_changed() which is invoked if any of
3122 * the effective csses seen from the css's cgroup may have changed.
3124 for_each_subsys(ss, ssid) {
3125 struct cgroup_subsys_state *this_css = cgroup_css(cgrp, ss);
3126 struct cgroup_subsys_state *css;
3128 if (!ss->css_e_css_changed || !this_css)
3131 css_for_each_descendant_pre(css, this_css)
3132 if (css != this_css)
3133 ss->css_e_css_changed(css);
3136 kernfs_activate(cgrp->kn);
3139 cgroup_kn_unlock(of->kn);
3140 return ret ?: nbytes;
3143 cgrp->subtree_control = old_sc;
3144 cgrp->child_subsys_mask = old_ss;
3146 for_each_subsys(ss, ssid) {
3147 if (!(enable & (1 << ssid)))
3150 cgroup_for_each_live_child(child, cgrp) {
3151 struct cgroup_subsys_state *css = cgroup_css(child, ss);
3156 if (css_enable & (1 << ssid))
3159 css_clear_dir(css, NULL);
3165 static int cgroup_events_show(struct seq_file *seq, void *v)
3167 seq_printf(seq, "populated %d\n",
3168 cgroup_is_populated(seq_css(seq)->cgroup));
3172 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
3173 size_t nbytes, loff_t off)
3175 struct cgroup *cgrp = of->kn->parent->priv;
3176 struct cftype *cft = of->kn->priv;
3177 struct cgroup_subsys_state *css;
3181 return cft->write(of, buf, nbytes, off);
3184 * kernfs guarantees that a file isn't deleted with operations in
3185 * flight, which means that the matching css is and stays alive and
3186 * doesn't need to be pinned. The RCU locking is not necessary
3187 * either. It's just for the convenience of using cgroup_css().
3190 css = cgroup_css(cgrp, cft->ss);
3193 if (cft->write_u64) {
3194 unsigned long long v;
3195 ret = kstrtoull(buf, 0, &v);
3197 ret = cft->write_u64(css, cft, v);
3198 } else if (cft->write_s64) {
3200 ret = kstrtoll(buf, 0, &v);
3202 ret = cft->write_s64(css, cft, v);
3207 return ret ?: nbytes;
3210 static void *cgroup_seqfile_start(struct seq_file *seq, loff_t *ppos)
3212 return seq_cft(seq)->seq_start(seq, ppos);
3215 static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
3217 return seq_cft(seq)->seq_next(seq, v, ppos);
3220 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
3222 seq_cft(seq)->seq_stop(seq, v);
3225 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
3227 struct cftype *cft = seq_cft(m);
3228 struct cgroup_subsys_state *css = seq_css(m);
3231 return cft->seq_show(m, arg);
3234 seq_printf(m, "%llu\n", cft->read_u64(css, cft));
3235 else if (cft->read_s64)
3236 seq_printf(m, "%lld\n", cft->read_s64(css, cft));
3242 static struct kernfs_ops cgroup_kf_single_ops = {
3243 .atomic_write_len = PAGE_SIZE,
3244 .write = cgroup_file_write,
3245 .seq_show = cgroup_seqfile_show,
3248 static struct kernfs_ops cgroup_kf_ops = {
3249 .atomic_write_len = PAGE_SIZE,
3250 .write = cgroup_file_write,
3251 .seq_start = cgroup_seqfile_start,
3252 .seq_next = cgroup_seqfile_next,
3253 .seq_stop = cgroup_seqfile_stop,
3254 .seq_show = cgroup_seqfile_show,
3258 * cgroup_rename - Only allow simple rename of directories in place.
3260 static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
3261 const char *new_name_str)
3263 struct cgroup *cgrp = kn->priv;
3266 if (kernfs_type(kn) != KERNFS_DIR)
3268 if (kn->parent != new_parent)
3272 * This isn't a proper migration and its usefulness is very
3273 * limited. Disallow on the default hierarchy.
3275 if (cgroup_on_dfl(cgrp))
3279 * We're gonna grab cgroup_mutex which nests outside kernfs
3280 * active_ref. kernfs_rename() doesn't require active_ref
3281 * protection. Break them before grabbing cgroup_mutex.
3283 kernfs_break_active_protection(new_parent);
3284 kernfs_break_active_protection(kn);
3286 mutex_lock(&cgroup_mutex);
3288 ret = kernfs_rename(kn, new_parent, new_name_str);
3290 mutex_unlock(&cgroup_mutex);
3292 kernfs_unbreak_active_protection(kn);
3293 kernfs_unbreak_active_protection(new_parent);
3297 /* set uid and gid of cgroup dirs and files to that of the creator */
3298 static int cgroup_kn_set_ugid(struct kernfs_node *kn)
3300 struct iattr iattr = { .ia_valid = ATTR_UID | ATTR_GID,
3301 .ia_uid = current_fsuid(),
3302 .ia_gid = current_fsgid(), };
3304 if (uid_eq(iattr.ia_uid, GLOBAL_ROOT_UID) &&
3305 gid_eq(iattr.ia_gid, GLOBAL_ROOT_GID))
3308 return kernfs_setattr(kn, &iattr);
3311 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
3314 char name[CGROUP_FILE_NAME_MAX];
3315 struct kernfs_node *kn;
3316 struct lock_class_key *key = NULL;
3319 #ifdef CONFIG_DEBUG_LOCK_ALLOC
3320 key = &cft->lockdep_key;
3322 kn = __kernfs_create_file(cgrp->kn, cgroup_file_name(cgrp, cft, name),
3323 cgroup_file_mode(cft), 0, cft->kf_ops, cft,
3328 ret = cgroup_kn_set_ugid(kn);
3334 if (cft->file_offset) {
3335 struct cgroup_file *cfile = (void *)css + cft->file_offset;
3337 spin_lock_irq(&cgroup_file_kn_lock);
3339 spin_unlock_irq(&cgroup_file_kn_lock);
3346 * cgroup_addrm_files - add or remove files to a cgroup directory
3347 * @css: the target css
3348 * @cgrp: the target cgroup (usually css->cgroup)
3349 * @cfts: array of cftypes to be added
3350 * @is_add: whether to add or remove
3352 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
3353 * For removals, this function never fails.
3355 static int cgroup_addrm_files(struct cgroup_subsys_state *css,
3356 struct cgroup *cgrp, struct cftype cfts[],
3359 struct cftype *cft, *cft_end = NULL;
3362 lockdep_assert_held(&cgroup_mutex);
3365 for (cft = cfts; cft != cft_end && cft->name[0] != '\0'; cft++) {
3366 /* does cft->flags tell us to skip this file on @cgrp? */
3367 if ((cft->flags & __CFTYPE_ONLY_ON_DFL) && !cgroup_on_dfl(cgrp))
3369 if ((cft->flags & __CFTYPE_NOT_ON_DFL) && cgroup_on_dfl(cgrp))
3371 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgroup_parent(cgrp))
3373 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgroup_parent(cgrp))
3377 ret = cgroup_add_file(css, cgrp, cft);
3379 pr_warn("%s: failed to add %s, err=%d\n",
3380 __func__, cft->name, ret);
3386 cgroup_rm_file(cgrp, cft);
3392 static int cgroup_apply_cftypes(struct cftype *cfts, bool is_add)
3395 struct cgroup_subsys *ss = cfts[0].ss;
3396 struct cgroup *root = &ss->root->cgrp;
3397 struct cgroup_subsys_state *css;
3400 lockdep_assert_held(&cgroup_mutex);
3402 /* add/rm files for all cgroups created before */
3403 css_for_each_descendant_pre(css, cgroup_css(root, ss)) {
3404 struct cgroup *cgrp = css->cgroup;
3406 if (cgroup_is_dead(cgrp))
3409 ret = cgroup_addrm_files(css, cgrp, cfts, is_add);
3415 kernfs_activate(root->kn);
3419 static void cgroup_exit_cftypes(struct cftype *cfts)
3423 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3424 /* free copy for custom atomic_write_len, see init_cftypes() */
3425 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE)
3430 /* revert flags set by cgroup core while adding @cfts */
3431 cft->flags &= ~(__CFTYPE_ONLY_ON_DFL | __CFTYPE_NOT_ON_DFL);
3435 static int cgroup_init_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3439 for (cft = cfts; cft->name[0] != '\0'; cft++) {
3440 struct kernfs_ops *kf_ops;
3442 WARN_ON(cft->ss || cft->kf_ops);
3445 kf_ops = &cgroup_kf_ops;
3447 kf_ops = &cgroup_kf_single_ops;
3450 * Ugh... if @cft wants a custom max_write_len, we need to
3451 * make a copy of kf_ops to set its atomic_write_len.
3453 if (cft->max_write_len && cft->max_write_len != PAGE_SIZE) {
3454 kf_ops = kmemdup(kf_ops, sizeof(*kf_ops), GFP_KERNEL);
3456 cgroup_exit_cftypes(cfts);
3459 kf_ops->atomic_write_len = cft->max_write_len;
3462 cft->kf_ops = kf_ops;
3469 static int cgroup_rm_cftypes_locked(struct cftype *cfts)
3471 lockdep_assert_held(&cgroup_mutex);
3473 if (!cfts || !cfts[0].ss)
3476 list_del(&cfts->node);
3477 cgroup_apply_cftypes(cfts, false);
3478 cgroup_exit_cftypes(cfts);
3483 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
3484 * @cfts: zero-length name terminated array of cftypes
3486 * Unregister @cfts. Files described by @cfts are removed from all
3487 * existing cgroups and all future cgroups won't have them either. This
3488 * function can be called anytime whether @cfts' subsys is attached or not.
3490 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
3493 int cgroup_rm_cftypes(struct cftype *cfts)
3497 mutex_lock(&cgroup_mutex);
3498 ret = cgroup_rm_cftypes_locked(cfts);
3499 mutex_unlock(&cgroup_mutex);
3504 * cgroup_add_cftypes - add an array of cftypes to a subsystem
3505 * @ss: target cgroup subsystem
3506 * @cfts: zero-length name terminated array of cftypes
3508 * Register @cfts to @ss. Files described by @cfts are created for all
3509 * existing cgroups to which @ss is attached and all future cgroups will
3510 * have them too. This function can be called anytime whether @ss is
3513 * Returns 0 on successful registration, -errno on failure. Note that this
3514 * function currently returns 0 as long as @cfts registration is successful
3515 * even if some file creation attempts on existing cgroups fail.
3517 static int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3521 if (!cgroup_ssid_enabled(ss->id))
3524 if (!cfts || cfts[0].name[0] == '\0')
3527 ret = cgroup_init_cftypes(ss, cfts);
3531 mutex_lock(&cgroup_mutex);
3533 list_add_tail(&cfts->node, &ss->cfts);
3534 ret = cgroup_apply_cftypes(cfts, true);
3536 cgroup_rm_cftypes_locked(cfts);
3538 mutex_unlock(&cgroup_mutex);
3543 * cgroup_add_dfl_cftypes - add an array of cftypes for default hierarchy
3544 * @ss: target cgroup subsystem
3545 * @cfts: zero-length name terminated array of cftypes
3547 * Similar to cgroup_add_cftypes() but the added files are only used for
3548 * the default hierarchy.
3550 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3554 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3555 cft->flags |= __CFTYPE_ONLY_ON_DFL;
3556 return cgroup_add_cftypes(ss, cfts);
3560 * cgroup_add_legacy_cftypes - add an array of cftypes for legacy hierarchies
3561 * @ss: target cgroup subsystem
3562 * @cfts: zero-length name terminated array of cftypes
3564 * Similar to cgroup_add_cftypes() but the added files are only used for
3565 * the legacy hierarchies.
3567 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
3571 for (cft = cfts; cft && cft->name[0] != '\0'; cft++)
3572 cft->flags |= __CFTYPE_NOT_ON_DFL;
3573 return cgroup_add_cftypes(ss, cfts);
3577 * cgroup_file_notify - generate a file modified event for a cgroup_file
3578 * @cfile: target cgroup_file
3580 * @cfile must have been obtained by setting cftype->file_offset.
3582 void cgroup_file_notify(struct cgroup_file *cfile)
3584 unsigned long flags;
3586 spin_lock_irqsave(&cgroup_file_kn_lock, flags);
3588 kernfs_notify(cfile->kn);
3589 spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
3593 * cgroup_task_count - count the number of tasks in a cgroup.
3594 * @cgrp: the cgroup in question
3596 * Return the number of tasks in the cgroup.
3598 static int cgroup_task_count(const struct cgroup *cgrp)
3601 struct cgrp_cset_link *link;
3603 spin_lock_bh(&css_set_lock);
3604 list_for_each_entry(link, &cgrp->cset_links, cset_link)
3605 count += atomic_read(&link->cset->refcount);
3606 spin_unlock_bh(&css_set_lock);
3611 * css_next_child - find the next child of a given css
3612 * @pos: the current position (%NULL to initiate traversal)
3613 * @parent: css whose children to walk
3615 * This function returns the next child of @parent and should be called
3616 * under either cgroup_mutex or RCU read lock. The only requirement is
3617 * that @parent and @pos are accessible. The next sibling is guaranteed to
3618 * be returned regardless of their states.
3620 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3621 * css which finished ->css_online() is guaranteed to be visible in the
3622 * future iterations and will stay visible until the last reference is put.
3623 * A css which hasn't finished ->css_online() or already finished
3624 * ->css_offline() may show up during traversal. It's each subsystem's
3625 * responsibility to synchronize against on/offlining.
3627 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos,
3628 struct cgroup_subsys_state *parent)
3630 struct cgroup_subsys_state *next;
3632 cgroup_assert_mutex_or_rcu_locked();
3635 * @pos could already have been unlinked from the sibling list.
3636 * Once a cgroup is removed, its ->sibling.next is no longer
3637 * updated when its next sibling changes. CSS_RELEASED is set when
3638 * @pos is taken off list, at which time its next pointer is valid,
3639 * and, as releases are serialized, the one pointed to by the next
3640 * pointer is guaranteed to not have started release yet. This
3641 * implies that if we observe !CSS_RELEASED on @pos in this RCU
3642 * critical section, the one pointed to by its next pointer is
3643 * guaranteed to not have finished its RCU grace period even if we
3644 * have dropped rcu_read_lock() inbetween iterations.
3646 * If @pos has CSS_RELEASED set, its next pointer can't be
3647 * dereferenced; however, as each css is given a monotonically
3648 * increasing unique serial number and always appended to the
3649 * sibling list, the next one can be found by walking the parent's
3650 * children until the first css with higher serial number than
3651 * @pos's. While this path can be slower, it happens iff iteration
3652 * races against release and the race window is very small.
3655 next = list_entry_rcu(parent->children.next, struct cgroup_subsys_state, sibling);
3656 } else if (likely(!(pos->flags & CSS_RELEASED))) {
3657 next = list_entry_rcu(pos->sibling.next, struct cgroup_subsys_state, sibling);
3659 list_for_each_entry_rcu(next, &parent->children, sibling)
3660 if (next->serial_nr > pos->serial_nr)
3665 * @next, if not pointing to the head, can be dereferenced and is
3668 if (&next->sibling != &parent->children)
3674 * css_next_descendant_pre - find the next descendant for pre-order walk
3675 * @pos: the current position (%NULL to initiate traversal)
3676 * @root: css whose descendants to walk
3678 * To be used by css_for_each_descendant_pre(). Find the next descendant
3679 * to visit for pre-order traversal of @root's descendants. @root is
3680 * included in the iteration and the first node to be visited.
3682 * While this function requires cgroup_mutex or RCU read locking, it
3683 * doesn't require the whole traversal to be contained in a single critical
3684 * section. This function will return the correct next descendant as long
3685 * as both @pos and @root are accessible and @pos is a descendant of @root.
3687 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3688 * css which finished ->css_online() is guaranteed to be visible in the
3689 * future iterations and will stay visible until the last reference is put.
3690 * A css which hasn't finished ->css_online() or already finished
3691 * ->css_offline() may show up during traversal. It's each subsystem's
3692 * responsibility to synchronize against on/offlining.
3694 struct cgroup_subsys_state *
3695 css_next_descendant_pre(struct cgroup_subsys_state *pos,
3696 struct cgroup_subsys_state *root)
3698 struct cgroup_subsys_state *next;
3700 cgroup_assert_mutex_or_rcu_locked();
3702 /* if first iteration, visit @root */
3706 /* visit the first child if exists */
3707 next = css_next_child(NULL, pos);
3711 /* no child, visit my or the closest ancestor's next sibling */
3712 while (pos != root) {
3713 next = css_next_child(pos, pos->parent);
3723 * css_rightmost_descendant - return the rightmost descendant of a css
3724 * @pos: css of interest
3726 * Return the rightmost descendant of @pos. If there's no descendant, @pos
3727 * is returned. This can be used during pre-order traversal to skip
3730 * While this function requires cgroup_mutex or RCU read locking, it
3731 * doesn't require the whole traversal to be contained in a single critical
3732 * section. This function will return the correct rightmost descendant as
3733 * long as @pos is accessible.
3735 struct cgroup_subsys_state *
3736 css_rightmost_descendant(struct cgroup_subsys_state *pos)
3738 struct cgroup_subsys_state *last, *tmp;
3740 cgroup_assert_mutex_or_rcu_locked();
3744 /* ->prev isn't RCU safe, walk ->next till the end */
3746 css_for_each_child(tmp, last)
3753 static struct cgroup_subsys_state *
3754 css_leftmost_descendant(struct cgroup_subsys_state *pos)
3756 struct cgroup_subsys_state *last;
3760 pos = css_next_child(NULL, pos);
3767 * css_next_descendant_post - find the next descendant for post-order walk
3768 * @pos: the current position (%NULL to initiate traversal)
3769 * @root: css whose descendants to walk
3771 * To be used by css_for_each_descendant_post(). Find the next descendant
3772 * to visit for post-order traversal of @root's descendants. @root is
3773 * included in the iteration and the last node to be visited.
3775 * While this function requires cgroup_mutex or RCU read locking, it
3776 * doesn't require the whole traversal to be contained in a single critical
3777 * section. This function will return the correct next descendant as long
3778 * as both @pos and @cgroup are accessible and @pos is a descendant of
3781 * If a subsystem synchronizes ->css_online() and the start of iteration, a
3782 * css which finished ->css_online() is guaranteed to be visible in the
3783 * future iterations and will stay visible until the last reference is put.
3784 * A css which hasn't finished ->css_online() or already finished
3785 * ->css_offline() may show up during traversal. It's each subsystem's
3786 * responsibility to synchronize against on/offlining.
3788 struct cgroup_subsys_state *
3789 css_next_descendant_post(struct cgroup_subsys_state *pos,
3790 struct cgroup_subsys_state *root)
3792 struct cgroup_subsys_state *next;
3794 cgroup_assert_mutex_or_rcu_locked();
3796 /* if first iteration, visit leftmost descendant which may be @root */
3798 return css_leftmost_descendant(root);
3800 /* if we visited @root, we're done */
3804 /* if there's an unvisited sibling, visit its leftmost descendant */
3805 next = css_next_child(pos, pos->parent);
3807 return css_leftmost_descendant(next);
3809 /* no sibling left, visit parent */
3814 * css_has_online_children - does a css have online children
3815 * @css: the target css
3817 * Returns %true if @css has any online children; otherwise, %false. This
3818 * function can be called from any context but the caller is responsible
3819 * for synchronizing against on/offlining as necessary.
3821 bool css_has_online_children(struct cgroup_subsys_state *css)
3823 struct cgroup_subsys_state *child;
3827 css_for_each_child(child, css) {
3828 if (child->flags & CSS_ONLINE) {
3838 * css_task_iter_advance_css_set - advance a task itererator to the next css_set
3839 * @it: the iterator to advance
3841 * Advance @it to the next css_set to walk.
3843 static void css_task_iter_advance_css_set(struct css_task_iter *it)
3845 struct list_head *l = it->cset_pos;
3846 struct cgrp_cset_link *link;
3847 struct css_set *cset;
3849 lockdep_assert_held(&css_set_lock);
3851 /* Advance to the next non-empty css_set */
3854 if (l == it->cset_head) {
3855 it->cset_pos = NULL;
3856 it->task_pos = NULL;
3861 cset = container_of(l, struct css_set,
3862 e_cset_node[it->ss->id]);
3864 link = list_entry(l, struct cgrp_cset_link, cset_link);
3867 } while (!css_set_populated(cset));
3871 if (!list_empty(&cset->tasks))
3872 it->task_pos = cset->tasks.next;
3874 it->task_pos = cset->mg_tasks.next;
3876 it->tasks_head = &cset->tasks;
3877 it->mg_tasks_head = &cset->mg_tasks;
3880 * We don't keep css_sets locked across iteration steps and thus
3881 * need to take steps to ensure that iteration can be resumed after
3882 * the lock is re-acquired. Iteration is performed at two levels -
3883 * css_sets and tasks in them.
3885 * Once created, a css_set never leaves its cgroup lists, so a
3886 * pinned css_set is guaranteed to stay put and we can resume
3887 * iteration afterwards.
3889 * Tasks may leave @cset across iteration steps. This is resolved
3890 * by registering each iterator with the css_set currently being
3891 * walked and making css_set_move_task() advance iterators whose
3892 * next task is leaving.
3895 list_del(&it->iters_node);
3896 put_css_set_locked(it->cur_cset);
3899 it->cur_cset = cset;
3900 list_add(&it->iters_node, &cset->task_iters);
3903 static void css_task_iter_advance(struct css_task_iter *it)
3905 struct list_head *l = it->task_pos;
3907 lockdep_assert_held(&css_set_lock);
3911 * Advance iterator to find next entry. cset->tasks is consumed
3912 * first and then ->mg_tasks. After ->mg_tasks, we move onto the
3917 if (l == it->tasks_head)
3918 l = it->mg_tasks_head->next;
3920 if (l == it->mg_tasks_head)
3921 css_task_iter_advance_css_set(it);
3927 * css_task_iter_start - initiate task iteration
3928 * @css: the css to walk tasks of
3929 * @it: the task iterator to use
3931 * Initiate iteration through the tasks of @css. The caller can call
3932 * css_task_iter_next() to walk through the tasks until the function
3933 * returns NULL. On completion of iteration, css_task_iter_end() must be
3936 void css_task_iter_start(struct cgroup_subsys_state *css,
3937 struct css_task_iter *it)
3939 /* no one should try to iterate before mounting cgroups */
3940 WARN_ON_ONCE(!use_task_css_set_links);
3942 memset(it, 0, sizeof(*it));
3944 spin_lock_bh(&css_set_lock);
3949 it->cset_pos = &css->cgroup->e_csets[css->ss->id];
3951 it->cset_pos = &css->cgroup->cset_links;
3953 it->cset_head = it->cset_pos;
3955 css_task_iter_advance_css_set(it);
3957 spin_unlock_bh(&css_set_lock);
3961 * css_task_iter_next - return the next task for the iterator
3962 * @it: the task iterator being iterated
3964 * The "next" function for task iteration. @it should have been
3965 * initialized via css_task_iter_start(). Returns NULL when the iteration
3968 struct task_struct *css_task_iter_next(struct css_task_iter *it)
3971 put_task_struct(it->cur_task);
3972 it->cur_task = NULL;
3975 spin_lock_bh(&css_set_lock);
3978 it->cur_task = list_entry(it->task_pos, struct task_struct,
3980 get_task_struct(it->cur_task);
3981 css_task_iter_advance(it);
3984 spin_unlock_bh(&css_set_lock);
3986 return it->cur_task;
3990 * css_task_iter_end - finish task iteration
3991 * @it: the task iterator to finish
3993 * Finish task iteration started by css_task_iter_start().
3995 void css_task_iter_end(struct css_task_iter *it)
3998 spin_lock_bh(&css_set_lock);
3999 list_del(&it->iters_node);
4000 put_css_set_locked(it->cur_cset);
4001 spin_unlock_bh(&css_set_lock);
4005 put_task_struct(it->cur_task);
4009 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
4010 * @to: cgroup to which the tasks will be moved
4011 * @from: cgroup in which the tasks currently reside
4013 * Locking rules between cgroup_post_fork() and the migration path
4014 * guarantee that, if a task is forking while being migrated, the new child
4015 * is guaranteed to be either visible in the source cgroup after the
4016 * parent's migration is complete or put into the target cgroup. No task
4017 * can slip out of migration through forking.
4019 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
4021 LIST_HEAD(preloaded_csets);
4022 struct cgrp_cset_link *link;
4023 struct css_task_iter it;
4024 struct task_struct *task;
4027 mutex_lock(&cgroup_mutex);
4029 /* all tasks in @from are being moved, all csets are source */
4030 spin_lock_bh(&css_set_lock);
4031 list_for_each_entry(link, &from->cset_links, cset_link)
4032 cgroup_migrate_add_src(link->cset, to, &preloaded_csets);
4033 spin_unlock_bh(&css_set_lock);
4035 ret = cgroup_migrate_prepare_dst(to, &preloaded_csets);
4040 * Migrate tasks one-by-one until @from is empty. This fails iff
4041 * ->can_attach() fails.
4044 css_task_iter_start(&from->self, &it);
4045 task = css_task_iter_next(&it);
4047 get_task_struct(task);
4048 css_task_iter_end(&it);
4051 ret = cgroup_migrate(task, false, to);
4052 put_task_struct(task);
4054 } while (task && !ret);
4056 cgroup_migrate_finish(&preloaded_csets);
4057 mutex_unlock(&cgroup_mutex);
4062 * Stuff for reading the 'tasks'/'procs' files.
4064 * Reading this file can return large amounts of data if a cgroup has
4065 * *lots* of attached tasks. So it may need several calls to read(),
4066 * but we cannot guarantee that the information we produce is correct
4067 * unless we produce it entirely atomically.
4071 /* which pidlist file are we talking about? */
4072 enum cgroup_filetype {
4078 * A pidlist is a list of pids that virtually represents the contents of one
4079 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
4080 * a pair (one each for procs, tasks) for each pid namespace that's relevant
4083 struct cgroup_pidlist {
4085 * used to find which pidlist is wanted. doesn't change as long as
4086 * this particular list stays in the list.
4088 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
4091 /* how many elements the above list has */
4093 /* each of these stored in a list by its cgroup */
4094 struct list_head links;
4095 /* pointer to the cgroup we belong to, for list removal purposes */
4096 struct cgroup *owner;
4097 /* for delayed destruction */
4098 struct delayed_work destroy_dwork;
4102 * The following two functions "fix" the issue where there are more pids
4103 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
4104 * TODO: replace with a kernel-wide solution to this problem
4106 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
4107 static void *pidlist_allocate(int count)
4109 if (PIDLIST_TOO_LARGE(count))
4110 return vmalloc(count * sizeof(pid_t));
4112 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
4115 static void pidlist_free(void *p)
4121 * Used to destroy all pidlists lingering waiting for destroy timer. None
4122 * should be left afterwards.
4124 static void cgroup_pidlist_destroy_all(struct cgroup *cgrp)
4126 struct cgroup_pidlist *l, *tmp_l;
4128 mutex_lock(&cgrp->pidlist_mutex);
4129 list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
4130 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
4131 mutex_unlock(&cgrp->pidlist_mutex);
4133 flush_workqueue(cgroup_pidlist_destroy_wq);
4134 BUG_ON(!list_empty(&cgrp->pidlists));
4137 static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
4139 struct delayed_work *dwork = to_delayed_work(work);
4140 struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
4142 struct cgroup_pidlist *tofree = NULL;
4144 mutex_lock(&l->owner->pidlist_mutex);
4147 * Destroy iff we didn't get queued again. The state won't change
4148 * as destroy_dwork can only be queued while locked.
4150 if (!delayed_work_pending(dwork)) {
4151 list_del(&l->links);
4152 pidlist_free(l->list);
4153 put_pid_ns(l->key.ns);
4157 mutex_unlock(&l->owner->pidlist_mutex);
4162 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
4163 * Returns the number of unique elements.
4165 static int pidlist_uniq(pid_t *list, int length)
4170 * we presume the 0th element is unique, so i starts at 1. trivial
4171 * edge cases first; no work needs to be done for either
4173 if (length == 0 || length == 1)
4175 /* src and dest walk down the list; dest counts unique elements */
4176 for (src = 1; src < length; src++) {
4177 /* find next unique element */
4178 while (list[src] == list[src-1]) {
4183 /* dest always points to where the next unique element goes */
4184 list[dest] = list[src];
4192 * The two pid files - task and cgroup.procs - guaranteed that the result
4193 * is sorted, which forced this whole pidlist fiasco. As pid order is
4194 * different per namespace, each namespace needs differently sorted list,
4195 * making it impossible to use, for example, single rbtree of member tasks
4196 * sorted by task pointer. As pidlists can be fairly large, allocating one
4197 * per open file is dangerous, so cgroup had to implement shared pool of
4198 * pidlists keyed by cgroup and namespace.
4200 * All this extra complexity was caused by the original implementation
4201 * committing to an entirely unnecessary property. In the long term, we
4202 * want to do away with it. Explicitly scramble sort order if on the
4203 * default hierarchy so that no such expectation exists in the new
4206 * Scrambling is done by swapping every two consecutive bits, which is
4207 * non-identity one-to-one mapping which disturbs sort order sufficiently.
4209 static pid_t pid_fry(pid_t pid)
4211 unsigned a = pid & 0x55555555;
4212 unsigned b = pid & 0xAAAAAAAA;
4214 return (a << 1) | (b >> 1);
4217 static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid)
4219 if (cgroup_on_dfl(cgrp))
4220 return pid_fry(pid);
4225 static int cmppid(const void *a, const void *b)
4227 return *(pid_t *)a - *(pid_t *)b;
4230 static int fried_cmppid(const void *a, const void *b)
4232 return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b);
4235 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
4236 enum cgroup_filetype type)
4238 struct cgroup_pidlist *l;
4239 /* don't need task_nsproxy() if we're looking at ourself */
4240 struct pid_namespace *ns = task_active_pid_ns(current);
4242 lockdep_assert_held(&cgrp->pidlist_mutex);
4244 list_for_each_entry(l, &cgrp->pidlists, links)
4245 if (l->key.type == type && l->key.ns == ns)
4251 * find the appropriate pidlist for our purpose (given procs vs tasks)
4252 * returns with the lock on that pidlist already held, and takes care
4253 * of the use count, or returns NULL with no locks held if we're out of
4256 static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
4257 enum cgroup_filetype type)
4259 struct cgroup_pidlist *l;
4261 lockdep_assert_held(&cgrp->pidlist_mutex);
4263 l = cgroup_pidlist_find(cgrp, type);
4267 /* entry not found; create a new one */
4268 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
4272 INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
4274 /* don't need task_nsproxy() if we're looking at ourself */
4275 l->key.ns = get_pid_ns(task_active_pid_ns(current));
4277 list_add(&l->links, &cgrp->pidlists);
4282 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
4284 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
4285 struct cgroup_pidlist **lp)
4289 int pid, n = 0; /* used for populating the array */
4290 struct css_task_iter it;
4291 struct task_struct *tsk;
4292 struct cgroup_pidlist *l;
4294 lockdep_assert_held(&cgrp->pidlist_mutex);
4297 * If cgroup gets more users after we read count, we won't have
4298 * enough space - tough. This race is indistinguishable to the
4299 * caller from the case that the additional cgroup users didn't
4300 * show up until sometime later on.
4302 length = cgroup_task_count(cgrp);
4303 array = pidlist_allocate(length);
4306 /* now, populate the array */
4307 css_task_iter_start(&cgrp->self, &it);
4308 while ((tsk = css_task_iter_next(&it))) {
4309 if (unlikely(n == length))
4311 /* get tgid or pid for procs or tasks file respectively */
4312 if (type == CGROUP_FILE_PROCS)
4313 pid = task_tgid_vnr(tsk);
4315 pid = task_pid_vnr(tsk);
4316 if (pid > 0) /* make sure to only use valid results */
4319 css_task_iter_end(&it);
4321 /* now sort & (if procs) strip out duplicates */
4322 if (cgroup_on_dfl(cgrp))
4323 sort(array, length, sizeof(pid_t), fried_cmppid, NULL);
4325 sort(array, length, sizeof(pid_t), cmppid, NULL);
4326 if (type == CGROUP_FILE_PROCS)
4327 length = pidlist_uniq(array, length);
4329 l = cgroup_pidlist_find_create(cgrp, type);
4331 pidlist_free(array);
4335 /* store array, freeing old if necessary */
4336 pidlist_free(l->list);
4344 * cgroupstats_build - build and fill cgroupstats
4345 * @stats: cgroupstats to fill information into
4346 * @dentry: A dentry entry belonging to the cgroup for which stats have
4349 * Build and fill cgroupstats so that taskstats can export it to user
4352 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
4354 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
4355 struct cgroup *cgrp;
4356 struct css_task_iter it;
4357 struct task_struct *tsk;
4359 /* it should be kernfs_node belonging to cgroupfs and is a directory */
4360 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
4361 kernfs_type(kn) != KERNFS_DIR)
4364 mutex_lock(&cgroup_mutex);
4367 * We aren't being called from kernfs and there's no guarantee on
4368 * @kn->priv's validity. For this and css_tryget_online_from_dir(),
4369 * @kn->priv is RCU safe. Let's do the RCU dancing.
4372 cgrp = rcu_dereference(kn->priv);
4373 if (!cgrp || cgroup_is_dead(cgrp)) {
4375 mutex_unlock(&cgroup_mutex);
4380 css_task_iter_start(&cgrp->self, &it);
4381 while ((tsk = css_task_iter_next(&it))) {
4382 switch (tsk->state) {
4384 stats->nr_running++;
4386 case TASK_INTERRUPTIBLE:
4387 stats->nr_sleeping++;
4389 case TASK_UNINTERRUPTIBLE:
4390 stats->nr_uninterruptible++;
4393 stats->nr_stopped++;
4396 if (delayacct_is_task_waiting_on_io(tsk))
4397 stats->nr_io_wait++;
4401 css_task_iter_end(&it);
4403 mutex_unlock(&cgroup_mutex);
4409 * seq_file methods for the tasks/procs files. The seq_file position is the
4410 * next pid to display; the seq_file iterator is a pointer to the pid
4411 * in the cgroup->l->list array.
4414 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
4417 * Initially we receive a position value that corresponds to
4418 * one more than the last pid shown (or 0 on the first call or
4419 * after a seek to the start). Use a binary-search to find the
4420 * next pid to display, if any
4422 struct kernfs_open_file *of = s->private;
4423 struct cgroup *cgrp = seq_css(s)->cgroup;
4424 struct cgroup_pidlist *l;
4425 enum cgroup_filetype type = seq_cft(s)->private;
4426 int index = 0, pid = *pos;
4429 mutex_lock(&cgrp->pidlist_mutex);
4432 * !NULL @of->priv indicates that this isn't the first start()
4433 * after open. If the matching pidlist is around, we can use that.
4434 * Look for it. Note that @of->priv can't be used directly. It
4435 * could already have been destroyed.
4438 of->priv = cgroup_pidlist_find(cgrp, type);
4441 * Either this is the first start() after open or the matching
4442 * pidlist has been destroyed inbetween. Create a new one.
4445 ret = pidlist_array_load(cgrp, type,
4446 (struct cgroup_pidlist **)&of->priv);
4448 return ERR_PTR(ret);
4453 int end = l->length;
4455 while (index < end) {
4456 int mid = (index + end) / 2;
4457 if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) {
4460 } else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid)
4466 /* If we're off the end of the array, we're done */
4467 if (index >= l->length)
4469 /* Update the abstract position to be the actual pid that we found */
4470 iter = l->list + index;
4471 *pos = cgroup_pid_fry(cgrp, *iter);
4475 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
4477 struct kernfs_open_file *of = s->private;
4478 struct cgroup_pidlist *l = of->priv;
4481 mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
4482 CGROUP_PIDLIST_DESTROY_DELAY);
4483 mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
4486 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
4488 struct kernfs_open_file *of = s->private;
4489 struct cgroup_pidlist *l = of->priv;
4491 pid_t *end = l->list + l->length;
4493 * Advance to the next pid in the array. If this goes off the
4500 *pos = cgroup_pid_fry(seq_css(s)->cgroup, *p);
4505 static int cgroup_pidlist_show(struct seq_file *s, void *v)
4507 seq_printf(s, "%d\n", *(int *)v);
4512 static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
4515 return notify_on_release(css->cgroup);
4518 static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
4519 struct cftype *cft, u64 val)
4522 set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
4524 clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
4528 static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
4531 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4534 static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
4535 struct cftype *cft, u64 val)
4538 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4540 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
4544 /* cgroup core interface files for the default hierarchy */
4545 static struct cftype cgroup_dfl_base_files[] = {
4547 .name = "cgroup.procs",
4548 .file_offset = offsetof(struct cgroup, procs_file),
4549 .seq_start = cgroup_pidlist_start,
4550 .seq_next = cgroup_pidlist_next,
4551 .seq_stop = cgroup_pidlist_stop,
4552 .seq_show = cgroup_pidlist_show,
4553 .private = CGROUP_FILE_PROCS,
4554 .write = cgroup_procs_write,
4557 .name = "cgroup.controllers",
4558 .flags = CFTYPE_ONLY_ON_ROOT,
4559 .seq_show = cgroup_root_controllers_show,
4562 .name = "cgroup.controllers",
4563 .flags = CFTYPE_NOT_ON_ROOT,
4564 .seq_show = cgroup_controllers_show,
4567 .name = "cgroup.subtree_control",
4568 .seq_show = cgroup_subtree_control_show,
4569 .write = cgroup_subtree_control_write,
4572 .name = "cgroup.events",
4573 .flags = CFTYPE_NOT_ON_ROOT,
4574 .file_offset = offsetof(struct cgroup, events_file),
4575 .seq_show = cgroup_events_show,
4580 /* cgroup core interface files for the legacy hierarchies */
4581 static struct cftype cgroup_legacy_base_files[] = {
4583 .name = "cgroup.procs",
4584 .seq_start = cgroup_pidlist_start,
4585 .seq_next = cgroup_pidlist_next,
4586 .seq_stop = cgroup_pidlist_stop,
4587 .seq_show = cgroup_pidlist_show,
4588 .private = CGROUP_FILE_PROCS,
4589 .write = cgroup_procs_write,
4592 .name = "cgroup.clone_children",
4593 .read_u64 = cgroup_clone_children_read,
4594 .write_u64 = cgroup_clone_children_write,
4597 .name = "cgroup.sane_behavior",
4598 .flags = CFTYPE_ONLY_ON_ROOT,
4599 .seq_show = cgroup_sane_behavior_show,
4603 .seq_start = cgroup_pidlist_start,
4604 .seq_next = cgroup_pidlist_next,
4605 .seq_stop = cgroup_pidlist_stop,
4606 .seq_show = cgroup_pidlist_show,
4607 .private = CGROUP_FILE_TASKS,
4608 .write = cgroup_tasks_write,
4611 .name = "notify_on_release",
4612 .read_u64 = cgroup_read_notify_on_release,
4613 .write_u64 = cgroup_write_notify_on_release,
4616 .name = "release_agent",
4617 .flags = CFTYPE_ONLY_ON_ROOT,
4618 .seq_show = cgroup_release_agent_show,
4619 .write = cgroup_release_agent_write,
4620 .max_write_len = PATH_MAX - 1,
4626 * css destruction is four-stage process.
4628 * 1. Destruction starts. Killing of the percpu_ref is initiated.
4629 * Implemented in kill_css().
4631 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
4632 * and thus css_tryget_online() is guaranteed to fail, the css can be
4633 * offlined by invoking offline_css(). After offlining, the base ref is
4634 * put. Implemented in css_killed_work_fn().
4636 * 3. When the percpu_ref reaches zero, the only possible remaining
4637 * accessors are inside RCU read sections. css_release() schedules the
4640 * 4. After the grace period, the css can be freed. Implemented in
4641 * css_free_work_fn().
4643 * It is actually hairier because both step 2 and 4 require process context
4644 * and thus involve punting to css->destroy_work adding two additional
4645 * steps to the already complex sequence.
4647 static void css_free_work_fn(struct work_struct *work)
4649 struct cgroup_subsys_state *css =
4650 container_of(work, struct cgroup_subsys_state, destroy_work);
4651 struct cgroup_subsys *ss = css->ss;
4652 struct cgroup *cgrp = css->cgroup;
4654 percpu_ref_exit(&css->refcnt);
4661 css_put(css->parent);
4664 cgroup_idr_remove(&ss->css_idr, id);
4667 /* cgroup free path */
4668 atomic_dec(&cgrp->root->nr_cgrps);
4669 cgroup_pidlist_destroy_all(cgrp);
4670 cancel_work_sync(&cgrp->release_agent_work);
4672 if (cgroup_parent(cgrp)) {
4674 * We get a ref to the parent, and put the ref when
4675 * this cgroup is being freed, so it's guaranteed
4676 * that the parent won't be destroyed before its
4679 cgroup_put(cgroup_parent(cgrp));
4680 kernfs_put(cgrp->kn);
4684 * This is root cgroup's refcnt reaching zero,
4685 * which indicates that the root should be
4688 cgroup_destroy_root(cgrp->root);
4693 static void css_free_rcu_fn(struct rcu_head *rcu_head)
4695 struct cgroup_subsys_state *css =
4696 container_of(rcu_head, struct cgroup_subsys_state, rcu_head);
4698 INIT_WORK(&css->destroy_work, css_free_work_fn);
4699 queue_work(cgroup_destroy_wq, &css->destroy_work);
4702 static void css_release_work_fn(struct work_struct *work)
4704 struct cgroup_subsys_state *css =
4705 container_of(work, struct cgroup_subsys_state, destroy_work);
4706 struct cgroup_subsys *ss = css->ss;
4707 struct cgroup *cgrp = css->cgroup;
4709 mutex_lock(&cgroup_mutex);
4711 css->flags |= CSS_RELEASED;
4712 list_del_rcu(&css->sibling);
4715 /* css release path */
4716 cgroup_idr_replace(&ss->css_idr, NULL, css->id);
4717 if (ss->css_released)
4718 ss->css_released(css);
4720 /* cgroup release path */
4721 cgroup_idr_remove(&cgrp->root->cgroup_idr, cgrp->id);
4725 * There are two control paths which try to determine
4726 * cgroup from dentry without going through kernfs -
4727 * cgroupstats_build() and css_tryget_online_from_dir().
4728 * Those are supported by RCU protecting clearing of
4729 * cgrp->kn->priv backpointer.
4731 RCU_INIT_POINTER(*(void __rcu __force **)&cgrp->kn->priv, NULL);
4734 mutex_unlock(&cgroup_mutex);
4736 call_rcu(&css->rcu_head, css_free_rcu_fn);
4739 static void css_release(struct percpu_ref *ref)
4741 struct cgroup_subsys_state *css =
4742 container_of(ref, struct cgroup_subsys_state, refcnt);
4744 INIT_WORK(&css->destroy_work, css_release_work_fn);
4745 queue_work(cgroup_destroy_wq, &css->destroy_work);
4748 static void init_and_link_css(struct cgroup_subsys_state *css,
4749 struct cgroup_subsys *ss, struct cgroup *cgrp)
4751 lockdep_assert_held(&cgroup_mutex);
4755 memset(css, 0, sizeof(*css));
4758 INIT_LIST_HEAD(&css->sibling);
4759 INIT_LIST_HEAD(&css->children);
4760 css->serial_nr = css_serial_nr_next++;
4762 if (cgroup_parent(cgrp)) {
4763 css->parent = cgroup_css(cgroup_parent(cgrp), ss);
4764 css_get(css->parent);
4767 BUG_ON(cgroup_css(cgrp, ss));
4770 /* invoke ->css_online() on a new CSS and mark it online if successful */
4771 static int online_css(struct cgroup_subsys_state *css)
4773 struct cgroup_subsys *ss = css->ss;
4776 lockdep_assert_held(&cgroup_mutex);
4779 ret = ss->css_online(css);
4781 css->flags |= CSS_ONLINE;
4782 rcu_assign_pointer(css->cgroup->subsys[ss->id], css);
4787 /* if the CSS is online, invoke ->css_offline() on it and mark it offline */
4788 static void offline_css(struct cgroup_subsys_state *css)
4790 struct cgroup_subsys *ss = css->ss;
4792 lockdep_assert_held(&cgroup_mutex);
4794 if (!(css->flags & CSS_ONLINE))
4797 if (ss->css_offline)
4798 ss->css_offline(css);
4800 css->flags &= ~CSS_ONLINE;
4801 RCU_INIT_POINTER(css->cgroup->subsys[ss->id], NULL);
4803 wake_up_all(&css->cgroup->offline_waitq);
4807 * create_css - create a cgroup_subsys_state
4808 * @cgrp: the cgroup new css will be associated with
4809 * @ss: the subsys of new css
4810 * @visible: whether to create control knobs for the new css or not
4812 * Create a new css associated with @cgrp - @ss pair. On success, the new
4813 * css is online and installed in @cgrp with all interface files created if
4814 * @visible. Returns 0 on success, -errno on failure.
4816 static int create_css(struct cgroup *cgrp, struct cgroup_subsys *ss,
4819 struct cgroup *parent = cgroup_parent(cgrp);
4820 struct cgroup_subsys_state *parent_css = cgroup_css(parent, ss);
4821 struct cgroup_subsys_state *css;
4824 lockdep_assert_held(&cgroup_mutex);
4826 css = ss->css_alloc(parent_css);
4828 return PTR_ERR(css);
4830 init_and_link_css(css, ss, cgrp);
4832 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
4836 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
4838 goto err_free_percpu_ref;
4842 err = css_populate_dir(css, NULL);
4847 /* @css is ready to be brought online now, make it visible */
4848 list_add_tail_rcu(&css->sibling, &parent_css->children);
4849 cgroup_idr_replace(&ss->css_idr, css, css->id);
4851 err = online_css(css);
4855 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4856 cgroup_parent(parent)) {
4857 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4858 current->comm, current->pid, ss->name);
4859 if (!strcmp(ss->name, "memory"))
4860 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
4861 ss->warned_broken_hierarchy = true;
4867 list_del_rcu(&css->sibling);
4868 css_clear_dir(css, NULL);
4870 cgroup_idr_remove(&ss->css_idr, css->id);
4871 err_free_percpu_ref:
4872 percpu_ref_exit(&css->refcnt);
4874 call_rcu(&css->rcu_head, css_free_rcu_fn);
4878 static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
4881 struct cgroup *parent, *cgrp, *tcgrp;
4882 struct cgroup_root *root;
4883 struct cgroup_subsys *ss;
4884 struct kernfs_node *kn;
4885 int level, ssid, ret;
4887 /* Do not accept '\n' to prevent making /proc/<pid>/cgroup unparsable.
4889 if (strchr(name, '\n'))
4892 parent = cgroup_kn_lock_live(parent_kn);
4895 root = parent->root;
4896 level = parent->level + 1;
4898 /* allocate the cgroup and its ID, 0 is reserved for the root */
4899 cgrp = kzalloc(sizeof(*cgrp) +
4900 sizeof(cgrp->ancestor_ids[0]) * (level + 1), GFP_KERNEL);
4906 ret = percpu_ref_init(&cgrp->self.refcnt, css_release, 0, GFP_KERNEL);
4911 * Temporarily set the pointer to NULL, so idr_find() won't return
4912 * a half-baked cgroup.
4914 cgrp->id = cgroup_idr_alloc(&root->cgroup_idr, NULL, 2, 0, GFP_KERNEL);
4917 goto out_cancel_ref;
4920 init_cgroup_housekeeping(cgrp);
4922 cgrp->self.parent = &parent->self;
4924 cgrp->level = level;
4926 for (tcgrp = cgrp; tcgrp; tcgrp = cgroup_parent(tcgrp))
4927 cgrp->ancestor_ids[tcgrp->level] = tcgrp->id;
4929 if (notify_on_release(parent))
4930 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4932 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4933 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4935 /* create the directory */
4936 kn = kernfs_create_dir(parent->kn, name, mode, cgrp);
4944 * This extra ref will be put in cgroup_free_fn() and guarantees
4945 * that @cgrp->kn is always accessible.
4949 cgrp->self.serial_nr = css_serial_nr_next++;
4951 /* allocation complete, commit to creation */
4952 list_add_tail_rcu(&cgrp->self.sibling, &cgroup_parent(cgrp)->self.children);
4953 atomic_inc(&root->nr_cgrps);
4957 * @cgrp is now fully operational. If something fails after this
4958 * point, it'll be released via the normal destruction path.
4960 cgroup_idr_replace(&root->cgroup_idr, cgrp, cgrp->id);
4962 ret = cgroup_kn_set_ugid(kn);
4966 ret = css_populate_dir(&cgrp->self, NULL);
4970 /* let's create and online css's */
4971 for_each_subsys(ss, ssid) {
4972 if (parent->child_subsys_mask & (1 << ssid)) {
4973 ret = create_css(cgrp, ss,
4974 parent->subtree_control & (1 << ssid));
4981 * On the default hierarchy, a child doesn't automatically inherit
4982 * subtree_control from the parent. Each is configured manually.
4984 if (!cgroup_on_dfl(cgrp)) {
4985 cgrp->subtree_control = parent->subtree_control;
4986 cgroup_refresh_child_subsys_mask(cgrp);
4989 kernfs_activate(kn);
4995 cgroup_idr_remove(&root->cgroup_idr, cgrp->id);
4997 percpu_ref_exit(&cgrp->self.refcnt);
5001 cgroup_kn_unlock(parent_kn);
5005 cgroup_destroy_locked(cgrp);
5010 * This is called when the refcnt of a css is confirmed to be killed.
5011 * css_tryget_online() is now guaranteed to fail. Tell the subsystem to
5012 * initate destruction and put the css ref from kill_css().
5014 static void css_killed_work_fn(struct work_struct *work)
5016 struct cgroup_subsys_state *css =
5017 container_of(work, struct cgroup_subsys_state, destroy_work);
5019 mutex_lock(&cgroup_mutex);
5021 mutex_unlock(&cgroup_mutex);
5026 /* css kill confirmation processing requires process context, bounce */
5027 static void css_killed_ref_fn(struct percpu_ref *ref)
5029 struct cgroup_subsys_state *css =
5030 container_of(ref, struct cgroup_subsys_state, refcnt);
5032 INIT_WORK(&css->destroy_work, css_killed_work_fn);
5033 queue_work(cgroup_destroy_wq, &css->destroy_work);
5037 * kill_css - destroy a css
5038 * @css: css to destroy
5040 * This function initiates destruction of @css by removing cgroup interface
5041 * files and putting its base reference. ->css_offline() will be invoked
5042 * asynchronously once css_tryget_online() is guaranteed to fail and when
5043 * the reference count reaches zero, @css will be released.
5045 static void kill_css(struct cgroup_subsys_state *css)
5047 lockdep_assert_held(&cgroup_mutex);
5050 * This must happen before css is disassociated with its cgroup.
5051 * See seq_css() for details.
5053 css_clear_dir(css, NULL);
5056 * Killing would put the base ref, but we need to keep it alive
5057 * until after ->css_offline().
5062 * cgroup core guarantees that, by the time ->css_offline() is
5063 * invoked, no new css reference will be given out via
5064 * css_tryget_online(). We can't simply call percpu_ref_kill() and
5065 * proceed to offlining css's because percpu_ref_kill() doesn't
5066 * guarantee that the ref is seen as killed on all CPUs on return.
5068 * Use percpu_ref_kill_and_confirm() to get notifications as each
5069 * css is confirmed to be seen as killed on all CPUs.
5071 percpu_ref_kill_and_confirm(&css->refcnt, css_killed_ref_fn);
5075 * cgroup_destroy_locked - the first stage of cgroup destruction
5076 * @cgrp: cgroup to be destroyed
5078 * css's make use of percpu refcnts whose killing latency shouldn't be
5079 * exposed to userland and are RCU protected. Also, cgroup core needs to
5080 * guarantee that css_tryget_online() won't succeed by the time
5081 * ->css_offline() is invoked. To satisfy all the requirements,
5082 * destruction is implemented in the following two steps.
5084 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
5085 * userland visible parts and start killing the percpu refcnts of
5086 * css's. Set up so that the next stage will be kicked off once all
5087 * the percpu refcnts are confirmed to be killed.
5089 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
5090 * rest of destruction. Once all cgroup references are gone, the
5091 * cgroup is RCU-freed.
5093 * This function implements s1. After this step, @cgrp is gone as far as
5094 * the userland is concerned and a new cgroup with the same name may be
5095 * created. As cgroup doesn't care about the names internally, this
5096 * doesn't cause any problem.
5098 static int cgroup_destroy_locked(struct cgroup *cgrp)
5099 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
5101 struct cgroup_subsys_state *css;
5104 lockdep_assert_held(&cgroup_mutex);
5107 * Only migration can raise populated from zero and we're already
5108 * holding cgroup_mutex.
5110 if (cgroup_is_populated(cgrp))
5114 * Make sure there's no live children. We can't test emptiness of
5115 * ->self.children as dead children linger on it while being
5116 * drained; otherwise, "rmdir parent/child parent" may fail.
5118 if (css_has_online_children(&cgrp->self))
5122 * Mark @cgrp dead. This prevents further task migration and child
5123 * creation by disabling cgroup_lock_live_group().
5125 cgrp->self.flags &= ~CSS_ONLINE;
5127 /* initiate massacre of all css's */
5128 for_each_css(css, ssid, cgrp)
5132 * Remove @cgrp directory along with the base files. @cgrp has an
5133 * extra ref on its kn.
5135 kernfs_remove(cgrp->kn);
5137 check_for_release(cgroup_parent(cgrp));
5139 /* put the base reference */
5140 percpu_ref_kill(&cgrp->self.refcnt);
5145 static int cgroup_rmdir(struct kernfs_node *kn)
5147 struct cgroup *cgrp;
5150 cgrp = cgroup_kn_lock_live(kn);
5154 ret = cgroup_destroy_locked(cgrp);
5156 cgroup_kn_unlock(kn);
5160 static struct kernfs_syscall_ops cgroup_kf_syscall_ops = {
5161 .remount_fs = cgroup_remount,
5162 .show_options = cgroup_show_options,
5163 .mkdir = cgroup_mkdir,
5164 .rmdir = cgroup_rmdir,
5165 .rename = cgroup_rename,
5168 static void __init cgroup_init_subsys(struct cgroup_subsys *ss, bool early)
5170 struct cgroup_subsys_state *css;
5172 pr_debug("Initializing cgroup subsys %s\n", ss->name);
5174 mutex_lock(&cgroup_mutex);
5176 idr_init(&ss->css_idr);
5177 INIT_LIST_HEAD(&ss->cfts);
5179 /* Create the root cgroup state for this subsystem */
5180 ss->root = &cgrp_dfl_root;
5181 css = ss->css_alloc(cgroup_css(&cgrp_dfl_root.cgrp, ss));
5182 /* We don't handle early failures gracefully */
5183 BUG_ON(IS_ERR(css));
5184 init_and_link_css(css, ss, &cgrp_dfl_root.cgrp);
5187 * Root csses are never destroyed and we can't initialize
5188 * percpu_ref during early init. Disable refcnting.
5190 css->flags |= CSS_NO_REF;
5193 /* allocation can't be done safely during early init */
5196 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2, GFP_KERNEL);
5197 BUG_ON(css->id < 0);
5200 /* Update the init_css_set to contain a subsys
5201 * pointer to this state - since the subsystem is
5202 * newly registered, all tasks and hence the
5203 * init_css_set is in the subsystem's root cgroup. */
5204 init_css_set.subsys[ss->id] = css;
5206 have_fork_callback |= (bool)ss->fork << ss->id;
5207 have_exit_callback |= (bool)ss->exit << ss->id;
5208 have_free_callback |= (bool)ss->free << ss->id;
5209 have_canfork_callback |= (bool)ss->can_fork << ss->id;
5211 /* At system boot, before all subsystems have been
5212 * registered, no tasks have been forked, so we don't
5213 * need to invoke fork callbacks here. */
5214 BUG_ON(!list_empty(&init_task.tasks));
5216 BUG_ON(online_css(css));
5218 mutex_unlock(&cgroup_mutex);
5222 * cgroup_init_early - cgroup initialization at system boot
5224 * Initialize cgroups at system boot, and initialize any
5225 * subsystems that request early init.
5227 int __init cgroup_init_early(void)
5229 static struct cgroup_sb_opts __initdata opts;
5230 struct cgroup_subsys *ss;
5233 init_cgroup_root(&cgrp_dfl_root, &opts);
5234 cgrp_dfl_root.cgrp.self.flags |= CSS_NO_REF;
5236 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
5238 for_each_subsys(ss, i) {
5239 WARN(!ss->css_alloc || !ss->css_free || ss->name || ss->id,
5240 "invalid cgroup_subsys %d:%s css_alloc=%p css_free=%p name:id=%d:%s\n",
5241 i, cgroup_subsys_name[i], ss->css_alloc, ss->css_free,
5243 WARN(strlen(cgroup_subsys_name[i]) > MAX_CGROUP_TYPE_NAMELEN,
5244 "cgroup_subsys_name %s too long\n", cgroup_subsys_name[i]);
5247 ss->name = cgroup_subsys_name[i];
5248 if (!ss->legacy_name)
5249 ss->legacy_name = cgroup_subsys_name[i];
5252 cgroup_init_subsys(ss, true);
5257 static unsigned long cgroup_disable_mask __initdata;
5260 * cgroup_init - cgroup initialization
5262 * Register cgroup filesystem and /proc file, and initialize
5263 * any subsystems that didn't request early init.
5265 int __init cgroup_init(void)
5267 struct cgroup_subsys *ss;
5271 BUG_ON(percpu_init_rwsem(&cgroup_threadgroup_rwsem));
5272 BUG_ON(cgroup_init_cftypes(NULL, cgroup_dfl_base_files));
5273 BUG_ON(cgroup_init_cftypes(NULL, cgroup_legacy_base_files));
5275 mutex_lock(&cgroup_mutex);
5277 /* Add init_css_set to the hash table */
5278 key = css_set_hash(init_css_set.subsys);
5279 hash_add(css_set_table, &init_css_set.hlist, key);
5281 BUG_ON(cgroup_setup_root(&cgrp_dfl_root, 0));
5283 mutex_unlock(&cgroup_mutex);
5285 for_each_subsys(ss, ssid) {
5286 if (ss->early_init) {
5287 struct cgroup_subsys_state *css =
5288 init_css_set.subsys[ss->id];
5290 css->id = cgroup_idr_alloc(&ss->css_idr, css, 1, 2,
5292 BUG_ON(css->id < 0);
5294 cgroup_init_subsys(ss, false);
5297 list_add_tail(&init_css_set.e_cset_node[ssid],
5298 &cgrp_dfl_root.cgrp.e_csets[ssid]);
5301 * Setting dfl_root subsys_mask needs to consider the
5302 * disabled flag and cftype registration needs kmalloc,
5303 * both of which aren't available during early_init.
5305 if (cgroup_disable_mask & (1 << ssid)) {
5306 static_branch_disable(cgroup_subsys_enabled_key[ssid]);
5307 printk(KERN_INFO "Disabling %s control group subsystem\n",
5312 cgrp_dfl_root.subsys_mask |= 1 << ss->id;
5314 if (!ss->dfl_cftypes)
5315 cgrp_dfl_root_inhibit_ss_mask |= 1 << ss->id;
5317 if (ss->dfl_cftypes == ss->legacy_cftypes) {
5318 WARN_ON(cgroup_add_cftypes(ss, ss->dfl_cftypes));
5320 WARN_ON(cgroup_add_dfl_cftypes(ss, ss->dfl_cftypes));
5321 WARN_ON(cgroup_add_legacy_cftypes(ss, ss->legacy_cftypes));
5325 ss->bind(init_css_set.subsys[ssid]);
5328 WARN_ON(sysfs_create_mount_point(fs_kobj, "cgroup"));
5329 WARN_ON(register_filesystem(&cgroup_fs_type));
5330 WARN_ON(register_filesystem(&cgroup2_fs_type));
5331 WARN_ON(!proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations));
5336 static int __init cgroup_wq_init(void)
5339 * There isn't much point in executing destruction path in
5340 * parallel. Good chunk is serialized with cgroup_mutex anyway.
5341 * Use 1 for @max_active.
5343 * We would prefer to do this in cgroup_init() above, but that
5344 * is called before init_workqueues(): so leave this until after.
5346 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
5347 BUG_ON(!cgroup_destroy_wq);
5350 * Used to destroy pidlists and separate to serve as flush domain.
5351 * Cap @max_active to 1 too.
5353 cgroup_pidlist_destroy_wq = alloc_workqueue("cgroup_pidlist_destroy",
5355 BUG_ON(!cgroup_pidlist_destroy_wq);
5359 core_initcall(cgroup_wq_init);
5362 * proc_cgroup_show()
5363 * - Print task's cgroup paths into seq_file, one line for each hierarchy
5364 * - Used for /proc/<pid>/cgroup.
5366 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns,
5367 struct pid *pid, struct task_struct *tsk)
5371 struct cgroup_root *root;
5374 buf = kmalloc(PATH_MAX, GFP_KERNEL);
5378 mutex_lock(&cgroup_mutex);
5379 spin_lock_bh(&css_set_lock);
5381 for_each_root(root) {
5382 struct cgroup_subsys *ss;
5383 struct cgroup *cgrp;
5384 int ssid, count = 0;
5386 if (root == &cgrp_dfl_root && !cgrp_dfl_root_visible)
5389 seq_printf(m, "%d:", root->hierarchy_id);
5390 if (root != &cgrp_dfl_root)
5391 for_each_subsys(ss, ssid)
5392 if (root->subsys_mask & (1 << ssid))
5393 seq_printf(m, "%s%s", count++ ? "," : "",
5395 if (strlen(root->name))
5396 seq_printf(m, "%sname=%s", count ? "," : "",
5400 cgrp = task_cgroup_from_root(tsk, root);
5403 * On traditional hierarchies, all zombie tasks show up as
5404 * belonging to the root cgroup. On the default hierarchy,
5405 * while a zombie doesn't show up in "cgroup.procs" and
5406 * thus can't be migrated, its /proc/PID/cgroup keeps
5407 * reporting the cgroup it belonged to before exiting. If
5408 * the cgroup is removed before the zombie is reaped,
5409 * " (deleted)" is appended to the cgroup path.
5411 if (cgroup_on_dfl(cgrp) || !(tsk->flags & PF_EXITING)) {
5412 path = cgroup_path(cgrp, buf, PATH_MAX);
5414 retval = -ENAMETOOLONG;
5423 if (cgroup_on_dfl(cgrp) && cgroup_is_dead(cgrp))
5424 seq_puts(m, " (deleted)\n");
5431 spin_unlock_bh(&css_set_lock);
5432 mutex_unlock(&cgroup_mutex);
5438 /* Display information about each subsystem and each hierarchy */
5439 static int proc_cgroupstats_show(struct seq_file *m, void *v)
5441 struct cgroup_subsys *ss;
5444 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5446 * ideally we don't want subsystems moving around while we do this.
5447 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5448 * subsys/hierarchy state.
5450 mutex_lock(&cgroup_mutex);
5452 for_each_subsys(ss, i)
5453 seq_printf(m, "%s\t%d\t%d\t%d\n",
5454 ss->legacy_name, ss->root->hierarchy_id,
5455 atomic_read(&ss->root->nr_cgrps),
5456 cgroup_ssid_enabled(i));
5458 mutex_unlock(&cgroup_mutex);
5462 static int cgroupstats_open(struct inode *inode, struct file *file)
5464 return single_open(file, proc_cgroupstats_show, NULL);
5467 static const struct file_operations proc_cgroupstats_operations = {
5468 .open = cgroupstats_open,
5470 .llseek = seq_lseek,
5471 .release = single_release,
5475 * cgroup_fork - initialize cgroup related fields during copy_process()
5476 * @child: pointer to task_struct of forking parent process.
5478 * A task is associated with the init_css_set until cgroup_post_fork()
5479 * attaches it to the parent's css_set. Empty cg_list indicates that
5480 * @child isn't holding reference to its css_set.
5482 void cgroup_fork(struct task_struct *child)
5484 RCU_INIT_POINTER(child->cgroups, &init_css_set);
5485 INIT_LIST_HEAD(&child->cg_list);
5489 * cgroup_can_fork - called on a new task before the process is exposed
5490 * @child: the task in question.
5492 * This calls the subsystem can_fork() callbacks. If the can_fork() callback
5493 * returns an error, the fork aborts with that error code. This allows for
5494 * a cgroup subsystem to conditionally allow or deny new forks.
5496 int cgroup_can_fork(struct task_struct *child)
5498 struct cgroup_subsys *ss;
5501 for_each_subsys_which(ss, i, &have_canfork_callback) {
5502 ret = ss->can_fork(child);
5510 for_each_subsys(ss, j) {
5513 if (ss->cancel_fork)
5514 ss->cancel_fork(child);
5521 * cgroup_cancel_fork - called if a fork failed after cgroup_can_fork()
5522 * @child: the task in question
5524 * This calls the cancel_fork() callbacks if a fork failed *after*
5525 * cgroup_can_fork() succeded.
5527 void cgroup_cancel_fork(struct task_struct *child)
5529 struct cgroup_subsys *ss;
5532 for_each_subsys(ss, i)
5533 if (ss->cancel_fork)
5534 ss->cancel_fork(child);
5538 * cgroup_post_fork - called on a new task after adding it to the task list
5539 * @child: the task in question
5541 * Adds the task to the list running through its css_set if necessary and
5542 * call the subsystem fork() callbacks. Has to be after the task is
5543 * visible on the task list in case we race with the first call to
5544 * cgroup_task_iter_start() - to guarantee that the new task ends up on its
5547 void cgroup_post_fork(struct task_struct *child)
5549 struct cgroup_subsys *ss;
5553 * This may race against cgroup_enable_task_cg_lists(). As that
5554 * function sets use_task_css_set_links before grabbing
5555 * tasklist_lock and we just went through tasklist_lock to add
5556 * @child, it's guaranteed that either we see the set
5557 * use_task_css_set_links or cgroup_enable_task_cg_lists() sees
5558 * @child during its iteration.
5560 * If we won the race, @child is associated with %current's
5561 * css_set. Grabbing css_set_lock guarantees both that the
5562 * association is stable, and, on completion of the parent's
5563 * migration, @child is visible in the source of migration or
5564 * already in the destination cgroup. This guarantee is necessary
5565 * when implementing operations which need to migrate all tasks of
5566 * a cgroup to another.
5568 * Note that if we lose to cgroup_enable_task_cg_lists(), @child
5569 * will remain in init_css_set. This is safe because all tasks are
5570 * in the init_css_set before cg_links is enabled and there's no
5571 * operation which transfers all tasks out of init_css_set.
5573 if (use_task_css_set_links) {
5574 struct css_set *cset;
5576 spin_lock_bh(&css_set_lock);
5577 cset = task_css_set(current);
5578 if (list_empty(&child->cg_list)) {
5580 css_set_move_task(child, NULL, cset, false);
5582 spin_unlock_bh(&css_set_lock);
5586 * Call ss->fork(). This must happen after @child is linked on
5587 * css_set; otherwise, @child might change state between ->fork()
5588 * and addition to css_set.
5590 for_each_subsys_which(ss, i, &have_fork_callback)
5595 * cgroup_exit - detach cgroup from exiting task
5596 * @tsk: pointer to task_struct of exiting process
5598 * Description: Detach cgroup from @tsk and release it.
5600 * Note that cgroups marked notify_on_release force every task in
5601 * them to take the global cgroup_mutex mutex when exiting.
5602 * This could impact scaling on very large systems. Be reluctant to
5603 * use notify_on_release cgroups where very high task exit scaling
5604 * is required on large systems.
5606 * We set the exiting tasks cgroup to the root cgroup (top_cgroup). We
5607 * call cgroup_exit() while the task is still competent to handle
5608 * notify_on_release(), then leave the task attached to the root cgroup in
5609 * each hierarchy for the remainder of its exit. No need to bother with
5610 * init_css_set refcnting. init_css_set never goes away and we can't race
5611 * with migration path - PF_EXITING is visible to migration path.
5613 void cgroup_exit(struct task_struct *tsk)
5615 struct cgroup_subsys *ss;
5616 struct css_set *cset;
5620 * Unlink from @tsk from its css_set. As migration path can't race
5621 * with us, we can check css_set and cg_list without synchronization.
5623 cset = task_css_set(tsk);
5625 if (!list_empty(&tsk->cg_list)) {
5626 spin_lock_bh(&css_set_lock);
5627 css_set_move_task(tsk, cset, NULL, false);
5628 spin_unlock_bh(&css_set_lock);
5633 /* see cgroup_post_fork() for details */
5634 for_each_subsys_which(ss, i, &have_exit_callback)
5638 void cgroup_free(struct task_struct *task)
5640 struct css_set *cset = task_css_set(task);
5641 struct cgroup_subsys *ss;
5644 for_each_subsys_which(ss, ssid, &have_free_callback)
5650 static void check_for_release(struct cgroup *cgrp)
5652 if (notify_on_release(cgrp) && !cgroup_is_populated(cgrp) &&
5653 !css_has_online_children(&cgrp->self) && !cgroup_is_dead(cgrp))
5654 schedule_work(&cgrp->release_agent_work);
5658 * Notify userspace when a cgroup is released, by running the
5659 * configured release agent with the name of the cgroup (path
5660 * relative to the root of cgroup file system) as the argument.
5662 * Most likely, this user command will try to rmdir this cgroup.
5664 * This races with the possibility that some other task will be
5665 * attached to this cgroup before it is removed, or that some other
5666 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5667 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5668 * unused, and this cgroup will be reprieved from its death sentence,
5669 * to continue to serve a useful existence. Next time it's released,
5670 * we will get notified again, if it still has 'notify_on_release' set.
5672 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5673 * means only wait until the task is successfully execve()'d. The
5674 * separate release agent task is forked by call_usermodehelper(),
5675 * then control in this thread returns here, without waiting for the
5676 * release agent task. We don't bother to wait because the caller of
5677 * this routine has no use for the exit status of the release agent
5678 * task, so no sense holding our caller up for that.
5680 static void cgroup_release_agent(struct work_struct *work)
5682 struct cgroup *cgrp =
5683 container_of(work, struct cgroup, release_agent_work);
5684 char *pathbuf = NULL, *agentbuf = NULL, *path;
5685 char *argv[3], *envp[3];
5687 mutex_lock(&cgroup_mutex);
5689 pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
5690 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5691 if (!pathbuf || !agentbuf)
5694 path = cgroup_path(cgrp, pathbuf, PATH_MAX);
5702 /* minimal command environment */
5704 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5707 mutex_unlock(&cgroup_mutex);
5708 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5711 mutex_unlock(&cgroup_mutex);
5717 static int __init cgroup_disable(char *str)
5719 struct cgroup_subsys *ss;
5723 while ((token = strsep(&str, ",")) != NULL) {
5727 for_each_subsys(ss, i) {
5728 if (strcmp(token, ss->name) &&
5729 strcmp(token, ss->legacy_name))
5731 cgroup_disable_mask |= 1 << i;
5736 __setup("cgroup_disable=", cgroup_disable);
5739 * css_tryget_online_from_dir - get corresponding css from a cgroup dentry
5740 * @dentry: directory dentry of interest
5741 * @ss: subsystem of interest
5743 * If @dentry is a directory for a cgroup which has @ss enabled on it, try
5744 * to get the corresponding css and return it. If such css doesn't exist
5745 * or can't be pinned, an ERR_PTR value is returned.
5747 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry,
5748 struct cgroup_subsys *ss)
5750 struct kernfs_node *kn = kernfs_node_from_dentry(dentry);
5751 struct cgroup_subsys_state *css = NULL;
5752 struct cgroup *cgrp;
5754 /* is @dentry a cgroup dir? */
5755 if (dentry->d_sb->s_type != &cgroup_fs_type || !kn ||
5756 kernfs_type(kn) != KERNFS_DIR)
5757 return ERR_PTR(-EBADF);
5762 * This path doesn't originate from kernfs and @kn could already
5763 * have been or be removed at any point. @kn->priv is RCU
5764 * protected for this access. See css_release_work_fn() for details.
5766 cgrp = rcu_dereference(kn->priv);
5768 css = cgroup_css(cgrp, ss);
5770 if (!css || !css_tryget_online(css))
5771 css = ERR_PTR(-ENOENT);
5778 * css_from_id - lookup css by id
5779 * @id: the cgroup id
5780 * @ss: cgroup subsys to be looked into
5782 * Returns the css if there's valid one with @id, otherwise returns NULL.
5783 * Should be called under rcu_read_lock().
5785 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss)
5787 WARN_ON_ONCE(!rcu_read_lock_held());
5788 return id > 0 ? idr_find(&ss->css_idr, id) : NULL;
5792 * cgroup_get_from_path - lookup and get a cgroup from its default hierarchy path
5793 * @path: path on the default hierarchy
5795 * Find the cgroup at @path on the default hierarchy, increment its
5796 * reference count and return it. Returns pointer to the found cgroup on
5797 * success, ERR_PTR(-ENOENT) if @path doens't exist and ERR_PTR(-ENOTDIR)
5798 * if @path points to a non-directory.
5800 struct cgroup *cgroup_get_from_path(const char *path)
5802 struct kernfs_node *kn;
5803 struct cgroup *cgrp;
5805 mutex_lock(&cgroup_mutex);
5807 kn = kernfs_walk_and_get(cgrp_dfl_root.cgrp.kn, path);
5809 if (kernfs_type(kn) == KERNFS_DIR) {
5813 cgrp = ERR_PTR(-ENOTDIR);
5817 cgrp = ERR_PTR(-ENOENT);
5820 mutex_unlock(&cgroup_mutex);
5823 EXPORT_SYMBOL_GPL(cgroup_get_from_path);
5826 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data
5827 * definition in cgroup-defs.h.
5829 #ifdef CONFIG_SOCK_CGROUP_DATA
5831 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID)
5833 DEFINE_SPINLOCK(cgroup_sk_update_lock);
5834 static bool cgroup_sk_alloc_disabled __read_mostly;
5836 void cgroup_sk_alloc_disable(void)
5838 if (cgroup_sk_alloc_disabled)
5840 pr_info("cgroup: disabling cgroup2 socket matching due to net_prio or net_cls activation\n");
5841 cgroup_sk_alloc_disabled = true;
5846 #define cgroup_sk_alloc_disabled false
5850 void cgroup_sk_alloc(struct sock_cgroup_data *skcd)
5852 if (cgroup_sk_alloc_disabled)
5858 struct css_set *cset;
5860 cset = task_css_set(current);
5861 if (likely(cgroup_tryget(cset->dfl_cgrp))) {
5862 skcd->val = (unsigned long)cset->dfl_cgrp;
5871 void cgroup_sk_free(struct sock_cgroup_data *skcd)
5873 cgroup_put(sock_cgroup_ptr(skcd));
5876 #endif /* CONFIG_SOCK_CGROUP_DATA */
5878 #ifdef CONFIG_CGROUP_DEBUG
5879 static struct cgroup_subsys_state *
5880 debug_css_alloc(struct cgroup_subsys_state *parent_css)
5882 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5885 return ERR_PTR(-ENOMEM);
5890 static void debug_css_free(struct cgroup_subsys_state *css)
5895 static u64 debug_taskcount_read(struct cgroup_subsys_state *css,
5898 return cgroup_task_count(css->cgroup);
5901 static u64 current_css_set_read(struct cgroup_subsys_state *css,
5904 return (u64)(unsigned long)current->cgroups;
5907 static u64 current_css_set_refcount_read(struct cgroup_subsys_state *css,
5913 count = atomic_read(&task_css_set(current)->refcount);
5918 static int current_css_set_cg_links_read(struct seq_file *seq, void *v)
5920 struct cgrp_cset_link *link;
5921 struct css_set *cset;
5924 name_buf = kmalloc(NAME_MAX + 1, GFP_KERNEL);
5928 spin_lock_bh(&css_set_lock);
5930 cset = rcu_dereference(current->cgroups);
5931 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
5932 struct cgroup *c = link->cgrp;
5934 cgroup_name(c, name_buf, NAME_MAX + 1);
5935 seq_printf(seq, "Root %d group %s\n",
5936 c->root->hierarchy_id, name_buf);
5939 spin_unlock_bh(&css_set_lock);
5944 #define MAX_TASKS_SHOWN_PER_CSS 25
5945 static int cgroup_css_links_read(struct seq_file *seq, void *v)
5947 struct cgroup_subsys_state *css = seq_css(seq);
5948 struct cgrp_cset_link *link;
5950 spin_lock_bh(&css_set_lock);
5951 list_for_each_entry(link, &css->cgroup->cset_links, cset_link) {
5952 struct css_set *cset = link->cset;
5953 struct task_struct *task;
5956 seq_printf(seq, "css_set %p\n", cset);
5958 list_for_each_entry(task, &cset->tasks, cg_list) {
5959 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
5961 seq_printf(seq, " task %d\n", task_pid_vnr(task));
5964 list_for_each_entry(task, &cset->mg_tasks, cg_list) {
5965 if (count++ > MAX_TASKS_SHOWN_PER_CSS)
5967 seq_printf(seq, " task %d\n", task_pid_vnr(task));
5971 seq_puts(seq, " ...\n");
5973 spin_unlock_bh(&css_set_lock);
5977 static u64 releasable_read(struct cgroup_subsys_state *css, struct cftype *cft)
5979 return (!cgroup_is_populated(css->cgroup) &&
5980 !css_has_online_children(&css->cgroup->self));
5983 static struct cftype debug_files[] = {
5985 .name = "taskcount",
5986 .read_u64 = debug_taskcount_read,
5990 .name = "current_css_set",
5991 .read_u64 = current_css_set_read,
5995 .name = "current_css_set_refcount",
5996 .read_u64 = current_css_set_refcount_read,
6000 .name = "current_css_set_cg_links",
6001 .seq_show = current_css_set_cg_links_read,
6005 .name = "cgroup_css_links",
6006 .seq_show = cgroup_css_links_read,
6010 .name = "releasable",
6011 .read_u64 = releasable_read,
6017 struct cgroup_subsys debug_cgrp_subsys = {
6018 .css_alloc = debug_css_alloc,
6019 .css_free = debug_css_free,
6020 .legacy_cftypes = debug_files,
6022 #endif /* CONFIG_CGROUP_DEBUG */