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 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
33 #include <linux/init_task.h>
34 #include <linux/kernel.h>
35 #include <linux/list.h>
37 #include <linux/mutex.h>
38 #include <linux/mount.h>
39 #include <linux/pagemap.h>
40 #include <linux/proc_fs.h>
41 #include <linux/rcupdate.h>
42 #include <linux/sched.h>
43 #include <linux/backing-dev.h>
44 #include <linux/seq_file.h>
45 #include <linux/slab.h>
46 #include <linux/magic.h>
47 #include <linux/spinlock.h>
48 #include <linux/string.h>
49 #include <linux/sort.h>
50 #include <linux/kmod.h>
51 #include <linux/module.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/namei.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/idr.h>
58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
59 #include <linux/eventfd.h>
60 #include <linux/poll.h>
61 #include <linux/flex_array.h> /* used in cgroup_attach_task */
62 #include <linux/kthread.h>
64 #include <linux/atomic.h>
67 * cgroup_mutex is the master lock. Any modification to cgroup or its
68 * hierarchy must be performed while holding it.
70 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
71 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
72 * release_agent_path and so on. Modifying requires both cgroup_mutex and
73 * cgroup_root_mutex. Readers can acquire either of the two. This is to
74 * break the following locking order cycle.
76 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
77 * B. namespace_sem -> cgroup_mutex
79 * B happens only through cgroup_show_options() and using cgroup_root_mutex
82 #ifdef CONFIG_PROVE_RCU
83 DEFINE_MUTEX(cgroup_mutex);
84 EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for task_subsys_state_check() */
86 static DEFINE_MUTEX(cgroup_mutex);
89 static DEFINE_MUTEX(cgroup_root_mutex);
92 * Generate an array of cgroup subsystem pointers. At boot time, this is
93 * populated with the built in subsystems, and modular subsystems are
94 * registered after that. The mutable section of this array is protected by
97 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
98 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
99 static struct cgroup_subsys *cgroup_subsys[CGROUP_SUBSYS_COUNT] = {
100 #include <linux/cgroup_subsys.h>
104 * The dummy hierarchy, reserved for the subsystems that are otherwise
105 * unattached - it never has more than a single cgroup, and all tasks are
106 * part of that cgroup.
108 static struct cgroupfs_root cgroup_dummy_root;
110 /* dummy_top is a shorthand for the dummy hierarchy's top cgroup */
111 static struct cgroup * const cgroup_dummy_top = &cgroup_dummy_root.top_cgroup;
114 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
117 struct list_head node;
118 struct dentry *dentry;
122 struct simple_xattrs xattrs;
126 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
127 * cgroup_subsys->use_id != 0.
129 #define CSS_ID_MAX (65535)
132 * The css to which this ID points. This pointer is set to valid value
133 * after cgroup is populated. If cgroup is removed, this will be NULL.
134 * This pointer is expected to be RCU-safe because destroy()
135 * is called after synchronize_rcu(). But for safe use, css_tryget()
136 * should be used for avoiding race.
138 struct cgroup_subsys_state __rcu *css;
144 * Depth in hierarchy which this ID belongs to.
146 unsigned short depth;
148 * ID is freed by RCU. (and lookup routine is RCU safe.)
150 struct rcu_head rcu_head;
152 * Hierarchy of CSS ID belongs to.
154 unsigned short stack[0]; /* Array of Length (depth+1) */
158 * cgroup_event represents events which userspace want to receive.
160 struct cgroup_event {
162 * Cgroup which the event belongs to.
166 * Control file which the event associated.
170 * eventfd to signal userspace about the event.
172 struct eventfd_ctx *eventfd;
174 * Each of these stored in a list by the cgroup.
176 struct list_head list;
178 * All fields below needed to unregister event when
179 * userspace closes eventfd.
182 wait_queue_head_t *wqh;
184 struct work_struct remove;
187 /* The list of hierarchy roots */
189 static LIST_HEAD(cgroup_roots);
190 static int cgroup_root_count;
193 * Hierarchy ID allocation and mapping. It follows the same exclusion
194 * rules as other root ops - both cgroup_mutex and cgroup_root_mutex for
195 * writes, either for reads.
197 static DEFINE_IDR(cgroup_hierarchy_idr);
199 static struct cgroup_name root_cgroup_name = { .name = "/" };
202 * Assign a monotonically increasing serial number to cgroups. It
203 * guarantees cgroups with bigger numbers are newer than those with smaller
204 * numbers. Also, as cgroups are always appended to the parent's
205 * ->children list, it guarantees that sibling cgroups are always sorted in
206 * the ascending serial number order on the list. Protected by
209 static u64 cgroup_serial_nr_next = 1;
211 /* This flag indicates whether tasks in the fork and exit paths should
212 * check for fork/exit handlers to call. This avoids us having to do
213 * extra work in the fork/exit path if none of the subsystems need to
216 static int need_forkexit_callback __read_mostly;
218 static void cgroup_offline_fn(struct work_struct *work);
219 static int cgroup_destroy_locked(struct cgroup *cgrp);
220 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
221 struct cftype cfts[], bool is_add);
223 /* convenient tests for these bits */
224 static inline bool cgroup_is_dead(const struct cgroup *cgrp)
226 return test_bit(CGRP_DEAD, &cgrp->flags);
230 * cgroup_is_descendant - test ancestry
231 * @cgrp: the cgroup to be tested
232 * @ancestor: possible ancestor of @cgrp
234 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
235 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
236 * and @ancestor are accessible.
238 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
241 if (cgrp == ancestor)
247 EXPORT_SYMBOL_GPL(cgroup_is_descendant);
249 static int cgroup_is_releasable(const struct cgroup *cgrp)
252 (1 << CGRP_RELEASABLE) |
253 (1 << CGRP_NOTIFY_ON_RELEASE);
254 return (cgrp->flags & bits) == bits;
257 static int notify_on_release(const struct cgroup *cgrp)
259 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
263 * for_each_subsys - iterate all loaded cgroup subsystems
264 * @ss: the iteration cursor
265 * @i: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
267 * Should be called under cgroup_mutex.
269 #define for_each_subsys(ss, i) \
270 for ((i) = 0; (i) < CGROUP_SUBSYS_COUNT; (i)++) \
271 if (({ lockdep_assert_held(&cgroup_mutex); \
272 !((ss) = cgroup_subsys[i]); })) { } \
276 * for_each_builtin_subsys - iterate all built-in cgroup subsystems
277 * @ss: the iteration cursor
278 * @i: the index of @ss, CGROUP_BUILTIN_SUBSYS_COUNT after reaching the end
280 * Bulit-in subsystems are always present and iteration itself doesn't
281 * require any synchronization.
283 #define for_each_builtin_subsys(ss, i) \
284 for ((i) = 0; (i) < CGROUP_BUILTIN_SUBSYS_COUNT && \
285 (((ss) = cgroup_subsys[i]) || true); (i)++)
287 /* iterate each subsystem attached to a hierarchy */
288 #define for_each_root_subsys(root, ss) \
289 list_for_each_entry((ss), &(root)->subsys_list, sibling)
291 /* iterate across the active hierarchies */
292 #define for_each_active_root(root) \
293 list_for_each_entry((root), &cgroup_roots, root_list)
295 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
297 return dentry->d_fsdata;
300 static inline struct cfent *__d_cfe(struct dentry *dentry)
302 return dentry->d_fsdata;
305 static inline struct cftype *__d_cft(struct dentry *dentry)
307 return __d_cfe(dentry)->type;
311 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
312 * @cgrp: the cgroup to be checked for liveness
314 * On success, returns true; the mutex should be later unlocked. On
315 * failure returns false with no lock held.
317 static bool cgroup_lock_live_group(struct cgroup *cgrp)
319 mutex_lock(&cgroup_mutex);
320 if (cgroup_is_dead(cgrp)) {
321 mutex_unlock(&cgroup_mutex);
327 /* the list of cgroups eligible for automatic release. Protected by
328 * release_list_lock */
329 static LIST_HEAD(release_list);
330 static DEFINE_RAW_SPINLOCK(release_list_lock);
331 static void cgroup_release_agent(struct work_struct *work);
332 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
333 static void check_for_release(struct cgroup *cgrp);
336 * A cgroup can be associated with multiple css_sets as different tasks may
337 * belong to different cgroups on different hierarchies. In the other
338 * direction, a css_set is naturally associated with multiple cgroups.
339 * This M:N relationship is represented by the following link structure
340 * which exists for each association and allows traversing the associations
343 struct cgrp_cset_link {
344 /* the cgroup and css_set this link associates */
346 struct css_set *cset;
348 /* list of cgrp_cset_links anchored at cgrp->cset_links */
349 struct list_head cset_link;
351 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
352 struct list_head cgrp_link;
355 /* The default css_set - used by init and its children prior to any
356 * hierarchies being mounted. It contains a pointer to the root state
357 * for each subsystem. Also used to anchor the list of css_sets. Not
358 * reference-counted, to improve performance when child cgroups
359 * haven't been created.
362 static struct css_set init_css_set;
363 static struct cgrp_cset_link init_cgrp_cset_link;
365 static int cgroup_init_idr(struct cgroup_subsys *ss,
366 struct cgroup_subsys_state *css);
368 /* css_set_lock protects the list of css_set objects, and the
369 * chain of tasks off each css_set. Nests outside task->alloc_lock
370 * due to cgroup_iter_start() */
371 static DEFINE_RWLOCK(css_set_lock);
372 static int css_set_count;
375 * hash table for cgroup groups. This improves the performance to find
376 * an existing css_set. This hash doesn't (currently) take into
377 * account cgroups in empty hierarchies.
379 #define CSS_SET_HASH_BITS 7
380 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
382 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
384 unsigned long key = 0UL;
385 struct cgroup_subsys *ss;
388 for_each_subsys(ss, i)
389 key += (unsigned long)css[i];
390 key = (key >> 16) ^ key;
395 /* We don't maintain the lists running through each css_set to its
396 * task until after the first call to cgroup_iter_start(). This
397 * reduces the fork()/exit() overhead for people who have cgroups
398 * compiled into their kernel but not actually in use */
399 static int use_task_css_set_links __read_mostly;
401 static void __put_css_set(struct css_set *cset, int taskexit)
403 struct cgrp_cset_link *link, *tmp_link;
406 * Ensure that the refcount doesn't hit zero while any readers
407 * can see it. Similar to atomic_dec_and_lock(), but for an
410 if (atomic_add_unless(&cset->refcount, -1, 1))
412 write_lock(&css_set_lock);
413 if (!atomic_dec_and_test(&cset->refcount)) {
414 write_unlock(&css_set_lock);
418 /* This css_set is dead. unlink it and release cgroup refcounts */
419 hash_del(&cset->hlist);
422 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
423 struct cgroup *cgrp = link->cgrp;
425 list_del(&link->cset_link);
426 list_del(&link->cgrp_link);
428 /* @cgrp can't go away while we're holding css_set_lock */
429 if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) {
431 set_bit(CGRP_RELEASABLE, &cgrp->flags);
432 check_for_release(cgrp);
438 write_unlock(&css_set_lock);
439 kfree_rcu(cset, rcu_head);
443 * refcounted get/put for css_set objects
445 static inline void get_css_set(struct css_set *cset)
447 atomic_inc(&cset->refcount);
450 static inline void put_css_set(struct css_set *cset)
452 __put_css_set(cset, 0);
455 static inline void put_css_set_taskexit(struct css_set *cset)
457 __put_css_set(cset, 1);
461 * compare_css_sets - helper function for find_existing_css_set().
462 * @cset: candidate css_set being tested
463 * @old_cset: existing css_set for a task
464 * @new_cgrp: cgroup that's being entered by the task
465 * @template: desired set of css pointers in css_set (pre-calculated)
467 * Returns true if "cg" matches "old_cg" except for the hierarchy
468 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
470 static bool compare_css_sets(struct css_set *cset,
471 struct css_set *old_cset,
472 struct cgroup *new_cgrp,
473 struct cgroup_subsys_state *template[])
475 struct list_head *l1, *l2;
477 if (memcmp(template, cset->subsys, sizeof(cset->subsys))) {
478 /* Not all subsystems matched */
483 * Compare cgroup pointers in order to distinguish between
484 * different cgroups in heirarchies with no subsystems. We
485 * could get by with just this check alone (and skip the
486 * memcmp above) but on most setups the memcmp check will
487 * avoid the need for this more expensive check on almost all
491 l1 = &cset->cgrp_links;
492 l2 = &old_cset->cgrp_links;
494 struct cgrp_cset_link *link1, *link2;
495 struct cgroup *cgrp1, *cgrp2;
499 /* See if we reached the end - both lists are equal length. */
500 if (l1 == &cset->cgrp_links) {
501 BUG_ON(l2 != &old_cset->cgrp_links);
504 BUG_ON(l2 == &old_cset->cgrp_links);
506 /* Locate the cgroups associated with these links. */
507 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
508 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
511 /* Hierarchies should be linked in the same order. */
512 BUG_ON(cgrp1->root != cgrp2->root);
515 * If this hierarchy is the hierarchy of the cgroup
516 * that's changing, then we need to check that this
517 * css_set points to the new cgroup; if it's any other
518 * hierarchy, then this css_set should point to the
519 * same cgroup as the old css_set.
521 if (cgrp1->root == new_cgrp->root) {
522 if (cgrp1 != new_cgrp)
533 * find_existing_css_set - init css array and find the matching css_set
534 * @old_cset: the css_set that we're using before the cgroup transition
535 * @cgrp: the cgroup that we're moving into
536 * @template: out param for the new set of csses, should be clear on entry
538 static struct css_set *find_existing_css_set(struct css_set *old_cset,
540 struct cgroup_subsys_state *template[])
542 struct cgroupfs_root *root = cgrp->root;
543 struct cgroup_subsys *ss;
544 struct css_set *cset;
549 * Build the set of subsystem state objects that we want to see in the
550 * new css_set. while subsystems can change globally, the entries here
551 * won't change, so no need for locking.
553 for_each_subsys(ss, i) {
554 if (root->subsys_mask & (1UL << i)) {
555 /* Subsystem is in this hierarchy. So we want
556 * the subsystem state from the new
558 template[i] = cgrp->subsys[i];
560 /* Subsystem is not in this hierarchy, so we
561 * don't want to change the subsystem state */
562 template[i] = old_cset->subsys[i];
566 key = css_set_hash(template);
567 hash_for_each_possible(css_set_table, cset, hlist, key) {
568 if (!compare_css_sets(cset, old_cset, cgrp, template))
571 /* This css_set matches what we need */
575 /* No existing cgroup group matched */
579 static void free_cgrp_cset_links(struct list_head *links_to_free)
581 struct cgrp_cset_link *link, *tmp_link;
583 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
584 list_del(&link->cset_link);
590 * allocate_cgrp_cset_links - allocate cgrp_cset_links
591 * @count: the number of links to allocate
592 * @tmp_links: list_head the allocated links are put on
594 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
595 * through ->cset_link. Returns 0 on success or -errno.
597 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
599 struct cgrp_cset_link *link;
602 INIT_LIST_HEAD(tmp_links);
604 for (i = 0; i < count; i++) {
605 link = kzalloc(sizeof(*link), GFP_KERNEL);
607 free_cgrp_cset_links(tmp_links);
610 list_add(&link->cset_link, tmp_links);
616 * link_css_set - a helper function to link a css_set to a cgroup
617 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
618 * @cset: the css_set to be linked
619 * @cgrp: the destination cgroup
621 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
624 struct cgrp_cset_link *link;
626 BUG_ON(list_empty(tmp_links));
627 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
630 list_move(&link->cset_link, &cgrp->cset_links);
632 * Always add links to the tail of the list so that the list
633 * is sorted by order of hierarchy creation
635 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
639 * find_css_set - return a new css_set with one cgroup updated
640 * @old_cset: the baseline css_set
641 * @cgrp: the cgroup to be updated
643 * Return a new css_set that's equivalent to @old_cset, but with @cgrp
644 * substituted into the appropriate hierarchy.
646 static struct css_set *find_css_set(struct css_set *old_cset,
649 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT] = { };
650 struct css_set *cset;
651 struct list_head tmp_links;
652 struct cgrp_cset_link *link;
655 lockdep_assert_held(&cgroup_mutex);
657 /* First see if we already have a cgroup group that matches
659 read_lock(&css_set_lock);
660 cset = find_existing_css_set(old_cset, cgrp, template);
663 read_unlock(&css_set_lock);
668 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
672 /* Allocate all the cgrp_cset_link objects that we'll need */
673 if (allocate_cgrp_cset_links(cgroup_root_count, &tmp_links) < 0) {
678 atomic_set(&cset->refcount, 1);
679 INIT_LIST_HEAD(&cset->cgrp_links);
680 INIT_LIST_HEAD(&cset->tasks);
681 INIT_HLIST_NODE(&cset->hlist);
683 /* Copy the set of subsystem state objects generated in
684 * find_existing_css_set() */
685 memcpy(cset->subsys, template, sizeof(cset->subsys));
687 write_lock(&css_set_lock);
688 /* Add reference counts and links from the new css_set. */
689 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
690 struct cgroup *c = link->cgrp;
692 if (c->root == cgrp->root)
694 link_css_set(&tmp_links, cset, c);
697 BUG_ON(!list_empty(&tmp_links));
701 /* Add this cgroup group to the hash table */
702 key = css_set_hash(cset->subsys);
703 hash_add(css_set_table, &cset->hlist, key);
705 write_unlock(&css_set_lock);
711 * Return the cgroup for "task" from the given hierarchy. Must be
712 * called with cgroup_mutex held.
714 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
715 struct cgroupfs_root *root)
717 struct css_set *cset;
718 struct cgroup *res = NULL;
720 BUG_ON(!mutex_is_locked(&cgroup_mutex));
721 read_lock(&css_set_lock);
723 * No need to lock the task - since we hold cgroup_mutex the
724 * task can't change groups, so the only thing that can happen
725 * is that it exits and its css is set back to init_css_set.
727 cset = task_css_set(task);
728 if (cset == &init_css_set) {
729 res = &root->top_cgroup;
731 struct cgrp_cset_link *link;
733 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
734 struct cgroup *c = link->cgrp;
736 if (c->root == root) {
742 read_unlock(&css_set_lock);
748 * There is one global cgroup mutex. We also require taking
749 * task_lock() when dereferencing a task's cgroup subsys pointers.
750 * See "The task_lock() exception", at the end of this comment.
752 * A task must hold cgroup_mutex to modify cgroups.
754 * Any task can increment and decrement the count field without lock.
755 * So in general, code holding cgroup_mutex can't rely on the count
756 * field not changing. However, if the count goes to zero, then only
757 * cgroup_attach_task() can increment it again. Because a count of zero
758 * means that no tasks are currently attached, therefore there is no
759 * way a task attached to that cgroup can fork (the other way to
760 * increment the count). So code holding cgroup_mutex can safely
761 * assume that if the count is zero, it will stay zero. Similarly, if
762 * a task holds cgroup_mutex on a cgroup with zero count, it
763 * knows that the cgroup won't be removed, as cgroup_rmdir()
766 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
767 * (usually) take cgroup_mutex. These are the two most performance
768 * critical pieces of code here. The exception occurs on cgroup_exit(),
769 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
770 * is taken, and if the cgroup count is zero, a usermode call made
771 * to the release agent with the name of the cgroup (path relative to
772 * the root of cgroup file system) as the argument.
774 * A cgroup can only be deleted if both its 'count' of using tasks
775 * is zero, and its list of 'children' cgroups is empty. Since all
776 * tasks in the system use _some_ cgroup, and since there is always at
777 * least one task in the system (init, pid == 1), therefore, top_cgroup
778 * always has either children cgroups and/or using tasks. So we don't
779 * need a special hack to ensure that top_cgroup cannot be deleted.
781 * The task_lock() exception
783 * The need for this exception arises from the action of
784 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
785 * another. It does so using cgroup_mutex, however there are
786 * several performance critical places that need to reference
787 * task->cgroup without the expense of grabbing a system global
788 * mutex. Therefore except as noted below, when dereferencing or, as
789 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
790 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
791 * the task_struct routinely used for such matters.
793 * P.S. One more locking exception. RCU is used to guard the
794 * update of a tasks cgroup pointer by cgroup_attach_task()
798 * A couple of forward declarations required, due to cyclic reference loop:
799 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
800 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
804 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
805 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
806 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
807 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
808 unsigned long subsys_mask);
809 static const struct inode_operations cgroup_dir_inode_operations;
810 static const struct file_operations proc_cgroupstats_operations;
812 static struct backing_dev_info cgroup_backing_dev_info = {
814 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
817 static int alloc_css_id(struct cgroup_subsys *ss,
818 struct cgroup *parent, struct cgroup *child);
820 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
822 struct inode *inode = new_inode(sb);
825 inode->i_ino = get_next_ino();
826 inode->i_mode = mode;
827 inode->i_uid = current_fsuid();
828 inode->i_gid = current_fsgid();
829 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
830 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
835 static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
837 struct cgroup_name *name;
839 name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL);
842 strcpy(name->name, dentry->d_name.name);
846 static void cgroup_free_fn(struct work_struct *work)
848 struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work);
849 struct cgroup_subsys *ss;
851 mutex_lock(&cgroup_mutex);
853 * Release the subsystem state objects.
855 for_each_root_subsys(cgrp->root, ss)
858 cgrp->root->number_of_cgroups--;
859 mutex_unlock(&cgroup_mutex);
862 * We get a ref to the parent's dentry, and put the ref when
863 * this cgroup is being freed, so it's guaranteed that the
864 * parent won't be destroyed before its children.
866 dput(cgrp->parent->dentry);
868 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
871 * Drop the active superblock reference that we took when we
872 * created the cgroup. This will free cgrp->root, if we are
873 * holding the last reference to @sb.
875 deactivate_super(cgrp->root->sb);
878 * if we're getting rid of the cgroup, refcount should ensure
879 * that there are no pidlists left.
881 BUG_ON(!list_empty(&cgrp->pidlists));
883 simple_xattrs_free(&cgrp->xattrs);
885 kfree(rcu_dereference_raw(cgrp->name));
889 static void cgroup_free_rcu(struct rcu_head *head)
891 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
893 INIT_WORK(&cgrp->destroy_work, cgroup_free_fn);
894 schedule_work(&cgrp->destroy_work);
897 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
899 /* is dentry a directory ? if so, kfree() associated cgroup */
900 if (S_ISDIR(inode->i_mode)) {
901 struct cgroup *cgrp = dentry->d_fsdata;
903 BUG_ON(!(cgroup_is_dead(cgrp)));
904 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
906 struct cfent *cfe = __d_cfe(dentry);
907 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
909 WARN_ONCE(!list_empty(&cfe->node) &&
910 cgrp != &cgrp->root->top_cgroup,
911 "cfe still linked for %s\n", cfe->type->name);
912 simple_xattrs_free(&cfe->xattrs);
918 static int cgroup_delete(const struct dentry *d)
923 static void remove_dir(struct dentry *d)
925 struct dentry *parent = dget(d->d_parent);
928 simple_rmdir(parent->d_inode, d);
932 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
936 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
937 lockdep_assert_held(&cgroup_mutex);
940 * If we're doing cleanup due to failure of cgroup_create(),
941 * the corresponding @cfe may not exist.
943 list_for_each_entry(cfe, &cgrp->files, node) {
944 struct dentry *d = cfe->dentry;
946 if (cft && cfe->type != cft)
951 simple_unlink(cgrp->dentry->d_inode, d);
952 list_del_init(&cfe->node);
960 * cgroup_clear_dir - selective removal of base and subsystem files
961 * @cgrp: target cgroup
962 * @base_files: true if the base files should be removed
963 * @subsys_mask: mask of the subsystem ids whose files should be removed
965 static void cgroup_clear_dir(struct cgroup *cgrp, bool base_files,
966 unsigned long subsys_mask)
968 struct cgroup_subsys *ss;
970 for_each_root_subsys(cgrp->root, ss) {
971 struct cftype_set *set;
972 if (!test_bit(ss->subsys_id, &subsys_mask))
974 list_for_each_entry(set, &ss->cftsets, node)
975 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
978 while (!list_empty(&cgrp->files))
979 cgroup_rm_file(cgrp, NULL);
984 * NOTE : the dentry must have been dget()'ed
986 static void cgroup_d_remove_dir(struct dentry *dentry)
988 struct dentry *parent;
990 parent = dentry->d_parent;
991 spin_lock(&parent->d_lock);
992 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
993 list_del_init(&dentry->d_u.d_child);
994 spin_unlock(&dentry->d_lock);
995 spin_unlock(&parent->d_lock);
1000 * Call with cgroup_mutex held. Drops reference counts on modules, including
1001 * any duplicate ones that parse_cgroupfs_options took. If this function
1002 * returns an error, no reference counts are touched.
1004 static int rebind_subsystems(struct cgroupfs_root *root,
1005 unsigned long added_mask, unsigned removed_mask)
1007 struct cgroup *cgrp = &root->top_cgroup;
1008 struct cgroup_subsys *ss;
1011 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1012 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1014 /* Check that any added subsystems are currently free */
1015 for_each_subsys(ss, i) {
1016 unsigned long bit = 1UL << i;
1018 if (!(bit & added_mask))
1021 if (ss->root != &cgroup_dummy_root) {
1022 /* Subsystem isn't free */
1027 /* Currently we don't handle adding/removing subsystems when
1028 * any child cgroups exist. This is theoretically supportable
1029 * but involves complex error handling, so it's being left until
1031 if (root->number_of_cgroups > 1)
1034 /* Process each subsystem */
1035 for_each_subsys(ss, i) {
1036 unsigned long bit = 1UL << i;
1038 if (bit & added_mask) {
1039 /* We're binding this subsystem to this hierarchy */
1040 BUG_ON(cgrp->subsys[i]);
1041 BUG_ON(!cgroup_dummy_top->subsys[i]);
1042 BUG_ON(cgroup_dummy_top->subsys[i]->cgroup != cgroup_dummy_top);
1044 cgrp->subsys[i] = cgroup_dummy_top->subsys[i];
1045 cgrp->subsys[i]->cgroup = cgrp;
1046 list_move(&ss->sibling, &root->subsys_list);
1051 /* refcount was already taken, and we're keeping it */
1052 root->subsys_mask |= bit;
1053 } else if (bit & removed_mask) {
1054 /* We're removing this subsystem */
1055 BUG_ON(cgrp->subsys[i] != cgroup_dummy_top->subsys[i]);
1056 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1059 ss->bind(cgroup_dummy_top);
1060 cgroup_dummy_top->subsys[i]->cgroup = cgroup_dummy_top;
1061 cgrp->subsys[i] = NULL;
1062 cgroup_subsys[i]->root = &cgroup_dummy_root;
1063 list_move(&ss->sibling, &cgroup_dummy_root.subsys_list);
1065 /* subsystem is now free - drop reference on module */
1066 module_put(ss->module);
1067 root->subsys_mask &= ~bit;
1068 } else if (bit & root->subsys_mask) {
1069 /* Subsystem state should already exist */
1070 BUG_ON(!cgrp->subsys[i]);
1072 * a refcount was taken, but we already had one, so
1073 * drop the extra reference.
1075 module_put(ss->module);
1076 #ifdef CONFIG_MODULE_UNLOAD
1077 BUG_ON(ss->module && !module_refcount(ss->module));
1080 /* Subsystem state shouldn't exist */
1081 BUG_ON(cgrp->subsys[i]);
1086 * Mark @root has finished binding subsystems. @root->subsys_mask
1087 * now matches the bound subsystems.
1089 root->flags |= CGRP_ROOT_SUBSYS_BOUND;
1094 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1096 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1097 struct cgroup_subsys *ss;
1099 mutex_lock(&cgroup_root_mutex);
1100 for_each_root_subsys(root, ss)
1101 seq_printf(seq, ",%s", ss->name);
1102 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR)
1103 seq_puts(seq, ",sane_behavior");
1104 if (root->flags & CGRP_ROOT_NOPREFIX)
1105 seq_puts(seq, ",noprefix");
1106 if (root->flags & CGRP_ROOT_XATTR)
1107 seq_puts(seq, ",xattr");
1108 if (strlen(root->release_agent_path))
1109 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1110 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1111 seq_puts(seq, ",clone_children");
1112 if (strlen(root->name))
1113 seq_printf(seq, ",name=%s", root->name);
1114 mutex_unlock(&cgroup_root_mutex);
1118 struct cgroup_sb_opts {
1119 unsigned long subsys_mask;
1120 unsigned long flags;
1121 char *release_agent;
1122 bool cpuset_clone_children;
1124 /* User explicitly requested empty subsystem */
1127 struct cgroupfs_root *new_root;
1132 * Convert a hierarchy specifier into a bitmask of subsystems and
1133 * flags. Call with cgroup_mutex held to protect the cgroup_subsys[]
1134 * array. This function takes refcounts on subsystems to be used, unless it
1135 * returns error, in which case no refcounts are taken.
1137 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1139 char *token, *o = data;
1140 bool all_ss = false, one_ss = false;
1141 unsigned long mask = (unsigned long)-1;
1142 bool module_pin_failed = false;
1143 struct cgroup_subsys *ss;
1146 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1148 #ifdef CONFIG_CPUSETS
1149 mask = ~(1UL << cpuset_subsys_id);
1152 memset(opts, 0, sizeof(*opts));
1154 while ((token = strsep(&o, ",")) != NULL) {
1157 if (!strcmp(token, "none")) {
1158 /* Explicitly have no subsystems */
1162 if (!strcmp(token, "all")) {
1163 /* Mutually exclusive option 'all' + subsystem name */
1169 if (!strcmp(token, "__DEVEL__sane_behavior")) {
1170 opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
1173 if (!strcmp(token, "noprefix")) {
1174 opts->flags |= CGRP_ROOT_NOPREFIX;
1177 if (!strcmp(token, "clone_children")) {
1178 opts->cpuset_clone_children = true;
1181 if (!strcmp(token, "xattr")) {
1182 opts->flags |= CGRP_ROOT_XATTR;
1185 if (!strncmp(token, "release_agent=", 14)) {
1186 /* Specifying two release agents is forbidden */
1187 if (opts->release_agent)
1189 opts->release_agent =
1190 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1191 if (!opts->release_agent)
1195 if (!strncmp(token, "name=", 5)) {
1196 const char *name = token + 5;
1197 /* Can't specify an empty name */
1200 /* Must match [\w.-]+ */
1201 for (i = 0; i < strlen(name); i++) {
1205 if ((c == '.') || (c == '-') || (c == '_'))
1209 /* Specifying two names is forbidden */
1212 opts->name = kstrndup(name,
1213 MAX_CGROUP_ROOT_NAMELEN - 1,
1221 for_each_subsys(ss, i) {
1222 if (strcmp(token, ss->name))
1227 /* Mutually exclusive option 'all' + subsystem name */
1230 set_bit(i, &opts->subsys_mask);
1235 if (i == CGROUP_SUBSYS_COUNT)
1240 * If the 'all' option was specified select all the subsystems,
1241 * otherwise if 'none', 'name=' and a subsystem name options
1242 * were not specified, let's default to 'all'
1244 if (all_ss || (!one_ss && !opts->none && !opts->name))
1245 for_each_subsys(ss, i)
1247 set_bit(i, &opts->subsys_mask);
1249 /* Consistency checks */
1251 if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1252 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1254 if (opts->flags & CGRP_ROOT_NOPREFIX) {
1255 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1259 if (opts->cpuset_clone_children) {
1260 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1266 * Option noprefix was introduced just for backward compatibility
1267 * with the old cpuset, so we allow noprefix only if mounting just
1268 * the cpuset subsystem.
1270 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1274 /* Can't specify "none" and some subsystems */
1275 if (opts->subsys_mask && opts->none)
1279 * We either have to specify by name or by subsystems. (So all
1280 * empty hierarchies must have a name).
1282 if (!opts->subsys_mask && !opts->name)
1286 * Grab references on all the modules we'll need, so the subsystems
1287 * don't dance around before rebind_subsystems attaches them. This may
1288 * take duplicate reference counts on a subsystem that's already used,
1289 * but rebind_subsystems handles this case.
1291 for_each_subsys(ss, i) {
1292 if (!(opts->subsys_mask & (1UL << i)))
1294 if (!try_module_get(cgroup_subsys[i]->module)) {
1295 module_pin_failed = true;
1299 if (module_pin_failed) {
1301 * oops, one of the modules was going away. this means that we
1302 * raced with a module_delete call, and to the user this is
1303 * essentially a "subsystem doesn't exist" case.
1305 for (i--; i >= 0; i--) {
1306 /* drop refcounts only on the ones we took */
1307 unsigned long bit = 1UL << i;
1309 if (!(bit & opts->subsys_mask))
1311 module_put(cgroup_subsys[i]->module);
1319 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1321 struct cgroup_subsys *ss;
1324 mutex_lock(&cgroup_mutex);
1325 for_each_subsys(ss, i)
1326 if (subsys_mask & (1UL << i))
1327 module_put(cgroup_subsys[i]->module);
1328 mutex_unlock(&cgroup_mutex);
1331 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1334 struct cgroupfs_root *root = sb->s_fs_info;
1335 struct cgroup *cgrp = &root->top_cgroup;
1336 struct cgroup_sb_opts opts;
1337 unsigned long added_mask, removed_mask;
1339 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1340 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1344 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1345 mutex_lock(&cgroup_mutex);
1346 mutex_lock(&cgroup_root_mutex);
1348 /* See what subsystems are wanted */
1349 ret = parse_cgroupfs_options(data, &opts);
1353 if (opts.subsys_mask != root->subsys_mask || opts.release_agent)
1354 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1355 task_tgid_nr(current), current->comm);
1357 added_mask = opts.subsys_mask & ~root->subsys_mask;
1358 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1360 /* Don't allow flags or name to change at remount */
1361 if (((opts.flags ^ root->flags) & CGRP_ROOT_OPTION_MASK) ||
1362 (opts.name && strcmp(opts.name, root->name))) {
1363 pr_err("cgroup: option or name mismatch, new: 0x%lx \"%s\", old: 0x%lx \"%s\"\n",
1364 opts.flags & CGRP_ROOT_OPTION_MASK, opts.name ?: "",
1365 root->flags & CGRP_ROOT_OPTION_MASK, root->name);
1371 * Clear out the files of subsystems that should be removed, do
1372 * this before rebind_subsystems, since rebind_subsystems may
1373 * change this hierarchy's subsys_list.
1375 cgroup_clear_dir(cgrp, false, removed_mask);
1377 ret = rebind_subsystems(root, added_mask, removed_mask);
1379 /* rebind_subsystems failed, re-populate the removed files */
1380 cgroup_populate_dir(cgrp, false, removed_mask);
1384 /* re-populate subsystem files */
1385 cgroup_populate_dir(cgrp, false, added_mask);
1387 if (opts.release_agent)
1388 strcpy(root->release_agent_path, opts.release_agent);
1390 kfree(opts.release_agent);
1392 mutex_unlock(&cgroup_root_mutex);
1393 mutex_unlock(&cgroup_mutex);
1394 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1396 drop_parsed_module_refcounts(opts.subsys_mask);
1400 static const struct super_operations cgroup_ops = {
1401 .statfs = simple_statfs,
1402 .drop_inode = generic_delete_inode,
1403 .show_options = cgroup_show_options,
1404 .remount_fs = cgroup_remount,
1407 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1409 INIT_LIST_HEAD(&cgrp->sibling);
1410 INIT_LIST_HEAD(&cgrp->children);
1411 INIT_LIST_HEAD(&cgrp->files);
1412 INIT_LIST_HEAD(&cgrp->cset_links);
1413 INIT_LIST_HEAD(&cgrp->release_list);
1414 INIT_LIST_HEAD(&cgrp->pidlists);
1415 mutex_init(&cgrp->pidlist_mutex);
1416 INIT_LIST_HEAD(&cgrp->event_list);
1417 spin_lock_init(&cgrp->event_list_lock);
1418 simple_xattrs_init(&cgrp->xattrs);
1421 static void init_cgroup_root(struct cgroupfs_root *root)
1423 struct cgroup *cgrp = &root->top_cgroup;
1425 INIT_LIST_HEAD(&root->subsys_list);
1426 INIT_LIST_HEAD(&root->root_list);
1427 root->number_of_cgroups = 1;
1429 RCU_INIT_POINTER(cgrp->name, &root_cgroup_name);
1430 init_cgroup_housekeeping(cgrp);
1433 static int cgroup_init_root_id(struct cgroupfs_root *root, int start, int end)
1437 lockdep_assert_held(&cgroup_mutex);
1438 lockdep_assert_held(&cgroup_root_mutex);
1440 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, start, end,
1445 root->hierarchy_id = id;
1449 static void cgroup_exit_root_id(struct cgroupfs_root *root)
1451 lockdep_assert_held(&cgroup_mutex);
1452 lockdep_assert_held(&cgroup_root_mutex);
1454 if (root->hierarchy_id) {
1455 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1456 root->hierarchy_id = 0;
1460 static int cgroup_test_super(struct super_block *sb, void *data)
1462 struct cgroup_sb_opts *opts = data;
1463 struct cgroupfs_root *root = sb->s_fs_info;
1465 /* If we asked for a name then it must match */
1466 if (opts->name && strcmp(opts->name, root->name))
1470 * If we asked for subsystems (or explicitly for no
1471 * subsystems) then they must match
1473 if ((opts->subsys_mask || opts->none)
1474 && (opts->subsys_mask != root->subsys_mask))
1480 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1482 struct cgroupfs_root *root;
1484 if (!opts->subsys_mask && !opts->none)
1487 root = kzalloc(sizeof(*root), GFP_KERNEL);
1489 return ERR_PTR(-ENOMEM);
1491 init_cgroup_root(root);
1494 * We need to set @root->subsys_mask now so that @root can be
1495 * matched by cgroup_test_super() before it finishes
1496 * initialization; otherwise, competing mounts with the same
1497 * options may try to bind the same subsystems instead of waiting
1498 * for the first one leading to unexpected mount errors.
1499 * SUBSYS_BOUND will be set once actual binding is complete.
1501 root->subsys_mask = opts->subsys_mask;
1502 root->flags = opts->flags;
1503 ida_init(&root->cgroup_ida);
1504 if (opts->release_agent)
1505 strcpy(root->release_agent_path, opts->release_agent);
1507 strcpy(root->name, opts->name);
1508 if (opts->cpuset_clone_children)
1509 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1513 static void cgroup_free_root(struct cgroupfs_root *root)
1516 /* hierarhcy ID shoulid already have been released */
1517 WARN_ON_ONCE(root->hierarchy_id);
1519 ida_destroy(&root->cgroup_ida);
1524 static int cgroup_set_super(struct super_block *sb, void *data)
1527 struct cgroup_sb_opts *opts = data;
1529 /* If we don't have a new root, we can't set up a new sb */
1530 if (!opts->new_root)
1533 BUG_ON(!opts->subsys_mask && !opts->none);
1535 ret = set_anon_super(sb, NULL);
1539 sb->s_fs_info = opts->new_root;
1540 opts->new_root->sb = sb;
1542 sb->s_blocksize = PAGE_CACHE_SIZE;
1543 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1544 sb->s_magic = CGROUP_SUPER_MAGIC;
1545 sb->s_op = &cgroup_ops;
1550 static int cgroup_get_rootdir(struct super_block *sb)
1552 static const struct dentry_operations cgroup_dops = {
1553 .d_iput = cgroup_diput,
1554 .d_delete = cgroup_delete,
1557 struct inode *inode =
1558 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1563 inode->i_fop = &simple_dir_operations;
1564 inode->i_op = &cgroup_dir_inode_operations;
1565 /* directories start off with i_nlink == 2 (for "." entry) */
1567 sb->s_root = d_make_root(inode);
1570 /* for everything else we want ->d_op set */
1571 sb->s_d_op = &cgroup_dops;
1575 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1576 int flags, const char *unused_dev_name,
1579 struct cgroup_sb_opts opts;
1580 struct cgroupfs_root *root;
1582 struct super_block *sb;
1583 struct cgroupfs_root *new_root;
1584 struct inode *inode;
1586 /* First find the desired set of subsystems */
1587 mutex_lock(&cgroup_mutex);
1588 ret = parse_cgroupfs_options(data, &opts);
1589 mutex_unlock(&cgroup_mutex);
1594 * Allocate a new cgroup root. We may not need it if we're
1595 * reusing an existing hierarchy.
1597 new_root = cgroup_root_from_opts(&opts);
1598 if (IS_ERR(new_root)) {
1599 ret = PTR_ERR(new_root);
1602 opts.new_root = new_root;
1604 /* Locate an existing or new sb for this hierarchy */
1605 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1608 cgroup_free_root(opts.new_root);
1612 root = sb->s_fs_info;
1614 if (root == opts.new_root) {
1615 /* We used the new root structure, so this is a new hierarchy */
1616 struct list_head tmp_links;
1617 struct cgroup *root_cgrp = &root->top_cgroup;
1618 struct cgroupfs_root *existing_root;
1619 const struct cred *cred;
1621 struct css_set *cset;
1623 BUG_ON(sb->s_root != NULL);
1625 ret = cgroup_get_rootdir(sb);
1627 goto drop_new_super;
1628 inode = sb->s_root->d_inode;
1630 mutex_lock(&inode->i_mutex);
1631 mutex_lock(&cgroup_mutex);
1632 mutex_lock(&cgroup_root_mutex);
1634 /* Check for name clashes with existing mounts */
1636 if (strlen(root->name))
1637 for_each_active_root(existing_root)
1638 if (!strcmp(existing_root->name, root->name))
1642 * We're accessing css_set_count without locking
1643 * css_set_lock here, but that's OK - it can only be
1644 * increased by someone holding cgroup_lock, and
1645 * that's us. The worst that can happen is that we
1646 * have some link structures left over
1648 ret = allocate_cgrp_cset_links(css_set_count, &tmp_links);
1652 /* ID 0 is reserved for dummy root, 1 for unified hierarchy */
1653 ret = cgroup_init_root_id(root, 2, 0);
1657 ret = rebind_subsystems(root, root->subsys_mask, 0);
1658 if (ret == -EBUSY) {
1659 free_cgrp_cset_links(&tmp_links);
1663 * There must be no failure case after here, since rebinding
1664 * takes care of subsystems' refcounts, which are explicitly
1665 * dropped in the failure exit path.
1668 /* EBUSY should be the only error here */
1671 list_add(&root->root_list, &cgroup_roots);
1672 cgroup_root_count++;
1674 sb->s_root->d_fsdata = root_cgrp;
1675 root->top_cgroup.dentry = sb->s_root;
1677 /* Link the top cgroup in this hierarchy into all
1678 * the css_set objects */
1679 write_lock(&css_set_lock);
1680 hash_for_each(css_set_table, i, cset, hlist)
1681 link_css_set(&tmp_links, cset, root_cgrp);
1682 write_unlock(&css_set_lock);
1684 free_cgrp_cset_links(&tmp_links);
1686 BUG_ON(!list_empty(&root_cgrp->children));
1687 BUG_ON(root->number_of_cgroups != 1);
1689 cred = override_creds(&init_cred);
1690 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1692 mutex_unlock(&cgroup_root_mutex);
1693 mutex_unlock(&cgroup_mutex);
1694 mutex_unlock(&inode->i_mutex);
1697 * We re-used an existing hierarchy - the new root (if
1698 * any) is not needed
1700 cgroup_free_root(opts.new_root);
1702 if ((root->flags ^ opts.flags) & CGRP_ROOT_OPTION_MASK) {
1703 if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) {
1704 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1706 goto drop_new_super;
1708 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1712 /* no subsys rebinding, so refcounts don't change */
1713 drop_parsed_module_refcounts(opts.subsys_mask);
1716 kfree(opts.release_agent);
1718 return dget(sb->s_root);
1721 cgroup_exit_root_id(root);
1722 mutex_unlock(&cgroup_root_mutex);
1723 mutex_unlock(&cgroup_mutex);
1724 mutex_unlock(&inode->i_mutex);
1726 deactivate_locked_super(sb);
1728 drop_parsed_module_refcounts(opts.subsys_mask);
1730 kfree(opts.release_agent);
1732 return ERR_PTR(ret);
1735 static void cgroup_kill_sb(struct super_block *sb) {
1736 struct cgroupfs_root *root = sb->s_fs_info;
1737 struct cgroup *cgrp = &root->top_cgroup;
1738 struct cgrp_cset_link *link, *tmp_link;
1743 BUG_ON(root->number_of_cgroups != 1);
1744 BUG_ON(!list_empty(&cgrp->children));
1746 mutex_lock(&cgroup_mutex);
1747 mutex_lock(&cgroup_root_mutex);
1749 /* Rebind all subsystems back to the default hierarchy */
1750 if (root->flags & CGRP_ROOT_SUBSYS_BOUND) {
1751 ret = rebind_subsystems(root, 0, root->subsys_mask);
1752 /* Shouldn't be able to fail ... */
1757 * Release all the links from cset_links to this hierarchy's
1760 write_lock(&css_set_lock);
1762 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1763 list_del(&link->cset_link);
1764 list_del(&link->cgrp_link);
1767 write_unlock(&css_set_lock);
1769 if (!list_empty(&root->root_list)) {
1770 list_del(&root->root_list);
1771 cgroup_root_count--;
1774 cgroup_exit_root_id(root);
1776 mutex_unlock(&cgroup_root_mutex);
1777 mutex_unlock(&cgroup_mutex);
1779 simple_xattrs_free(&cgrp->xattrs);
1781 kill_litter_super(sb);
1782 cgroup_free_root(root);
1785 static struct file_system_type cgroup_fs_type = {
1787 .mount = cgroup_mount,
1788 .kill_sb = cgroup_kill_sb,
1791 static struct kobject *cgroup_kobj;
1794 * cgroup_path - generate the path of a cgroup
1795 * @cgrp: the cgroup in question
1796 * @buf: the buffer to write the path into
1797 * @buflen: the length of the buffer
1799 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1801 * We can't generate cgroup path using dentry->d_name, as accessing
1802 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1803 * inode's i_mutex, while on the other hand cgroup_path() can be called
1804 * with some irq-safe spinlocks held.
1806 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1808 int ret = -ENAMETOOLONG;
1811 if (!cgrp->parent) {
1812 if (strlcpy(buf, "/", buflen) >= buflen)
1813 return -ENAMETOOLONG;
1817 start = buf + buflen - 1;
1822 const char *name = cgroup_name(cgrp);
1826 if ((start -= len) < buf)
1828 memcpy(start, name, len);
1834 cgrp = cgrp->parent;
1835 } while (cgrp->parent);
1837 memmove(buf, start, buf + buflen - start);
1842 EXPORT_SYMBOL_GPL(cgroup_path);
1845 * task_cgroup_path_from_hierarchy - cgroup path of a task on a hierarchy
1846 * @task: target task
1847 * @hierarchy_id: the hierarchy to look up @task's cgroup from
1848 * @buf: the buffer to write the path into
1849 * @buflen: the length of the buffer
1851 * Determine @task's cgroup on the hierarchy specified by @hierarchy_id and
1852 * copy its path into @buf. This function grabs cgroup_mutex and shouldn't
1853 * be used inside locks used by cgroup controller callbacks.
1855 int task_cgroup_path_from_hierarchy(struct task_struct *task, int hierarchy_id,
1856 char *buf, size_t buflen)
1858 struct cgroupfs_root *root;
1859 struct cgroup *cgrp = NULL;
1862 mutex_lock(&cgroup_mutex);
1864 root = idr_find(&cgroup_hierarchy_idr, hierarchy_id);
1866 cgrp = task_cgroup_from_root(task, root);
1867 ret = cgroup_path(cgrp, buf, buflen);
1870 mutex_unlock(&cgroup_mutex);
1874 EXPORT_SYMBOL_GPL(task_cgroup_path_from_hierarchy);
1877 * Control Group taskset
1879 struct task_and_cgroup {
1880 struct task_struct *task;
1881 struct cgroup *cgrp;
1885 struct cgroup_taskset {
1886 struct task_and_cgroup single;
1887 struct flex_array *tc_array;
1890 struct cgroup *cur_cgrp;
1894 * cgroup_taskset_first - reset taskset and return the first task
1895 * @tset: taskset of interest
1897 * @tset iteration is initialized and the first task is returned.
1899 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1901 if (tset->tc_array) {
1903 return cgroup_taskset_next(tset);
1905 tset->cur_cgrp = tset->single.cgrp;
1906 return tset->single.task;
1909 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1912 * cgroup_taskset_next - iterate to the next task in taskset
1913 * @tset: taskset of interest
1915 * Return the next task in @tset. Iteration must have been initialized
1916 * with cgroup_taskset_first().
1918 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1920 struct task_and_cgroup *tc;
1922 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1925 tc = flex_array_get(tset->tc_array, tset->idx++);
1926 tset->cur_cgrp = tc->cgrp;
1929 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1932 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1933 * @tset: taskset of interest
1935 * Return the cgroup for the current (last returned) task of @tset. This
1936 * function must be preceded by either cgroup_taskset_first() or
1937 * cgroup_taskset_next().
1939 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1941 return tset->cur_cgrp;
1943 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1946 * cgroup_taskset_size - return the number of tasks in taskset
1947 * @tset: taskset of interest
1949 int cgroup_taskset_size(struct cgroup_taskset *tset)
1951 return tset->tc_array ? tset->tc_array_len : 1;
1953 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1957 * cgroup_task_migrate - move a task from one cgroup to another.
1959 * Must be called with cgroup_mutex and threadgroup locked.
1961 static void cgroup_task_migrate(struct cgroup *old_cgrp,
1962 struct task_struct *tsk,
1963 struct css_set *new_cset)
1965 struct css_set *old_cset;
1968 * We are synchronized through threadgroup_lock() against PF_EXITING
1969 * setting such that we can't race against cgroup_exit() changing the
1970 * css_set to init_css_set and dropping the old one.
1972 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1973 old_cset = task_css_set(tsk);
1976 rcu_assign_pointer(tsk->cgroups, new_cset);
1979 /* Update the css_set linked lists if we're using them */
1980 write_lock(&css_set_lock);
1981 if (!list_empty(&tsk->cg_list))
1982 list_move(&tsk->cg_list, &new_cset->tasks);
1983 write_unlock(&css_set_lock);
1986 * We just gained a reference on old_cset by taking it from the
1987 * task. As trading it for new_cset is protected by cgroup_mutex,
1988 * we're safe to drop it here; it will be freed under RCU.
1990 set_bit(CGRP_RELEASABLE, &old_cgrp->flags);
1991 put_css_set(old_cset);
1995 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1996 * @cgrp: the cgroup to attach to
1997 * @tsk: the task or the leader of the threadgroup to be attached
1998 * @threadgroup: attach the whole threadgroup?
2000 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2001 * task_lock of @tsk or each thread in the threadgroup individually in turn.
2003 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
2006 int retval, i, group_size;
2007 struct cgroup_subsys *ss, *failed_ss = NULL;
2008 struct cgroupfs_root *root = cgrp->root;
2009 /* threadgroup list cursor and array */
2010 struct task_struct *leader = tsk;
2011 struct task_and_cgroup *tc;
2012 struct flex_array *group;
2013 struct cgroup_taskset tset = { };
2016 * step 0: in order to do expensive, possibly blocking operations for
2017 * every thread, we cannot iterate the thread group list, since it needs
2018 * rcu or tasklist locked. instead, build an array of all threads in the
2019 * group - group_rwsem prevents new threads from appearing, and if
2020 * threads exit, this will just be an over-estimate.
2023 group_size = get_nr_threads(tsk);
2026 /* flex_array supports very large thread-groups better than kmalloc. */
2027 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2030 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2031 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
2033 goto out_free_group_list;
2037 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2038 * already PF_EXITING could be freed from underneath us unless we
2039 * take an rcu_read_lock.
2043 struct task_and_cgroup ent;
2045 /* @tsk either already exited or can't exit until the end */
2046 if (tsk->flags & PF_EXITING)
2049 /* as per above, nr_threads may decrease, but not increase. */
2050 BUG_ON(i >= group_size);
2052 ent.cgrp = task_cgroup_from_root(tsk, root);
2053 /* nothing to do if this task is already in the cgroup */
2054 if (ent.cgrp == cgrp)
2057 * saying GFP_ATOMIC has no effect here because we did prealloc
2058 * earlier, but it's good form to communicate our expectations.
2060 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2061 BUG_ON(retval != 0);
2066 } while_each_thread(leader, tsk);
2068 /* remember the number of threads in the array for later. */
2070 tset.tc_array = group;
2071 tset.tc_array_len = group_size;
2073 /* methods shouldn't be called if no task is actually migrating */
2076 goto out_free_group_list;
2079 * step 1: check that we can legitimately attach to the cgroup.
2081 for_each_root_subsys(root, ss) {
2082 if (ss->can_attach) {
2083 retval = ss->can_attach(cgrp, &tset);
2086 goto out_cancel_attach;
2092 * step 2: make sure css_sets exist for all threads to be migrated.
2093 * we use find_css_set, which allocates a new one if necessary.
2095 for (i = 0; i < group_size; i++) {
2096 struct css_set *old_cset;
2098 tc = flex_array_get(group, i);
2099 old_cset = task_css_set(tc->task);
2100 tc->cg = find_css_set(old_cset, cgrp);
2103 goto out_put_css_set_refs;
2108 * step 3: now that we're guaranteed success wrt the css_sets,
2109 * proceed to move all tasks to the new cgroup. There are no
2110 * failure cases after here, so this is the commit point.
2112 for (i = 0; i < group_size; i++) {
2113 tc = flex_array_get(group, i);
2114 cgroup_task_migrate(tc->cgrp, tc->task, tc->cg);
2116 /* nothing is sensitive to fork() after this point. */
2119 * step 4: do subsystem attach callbacks.
2121 for_each_root_subsys(root, ss) {
2123 ss->attach(cgrp, &tset);
2127 * step 5: success! and cleanup
2130 out_put_css_set_refs:
2132 for (i = 0; i < group_size; i++) {
2133 tc = flex_array_get(group, i);
2136 put_css_set(tc->cg);
2141 for_each_root_subsys(root, ss) {
2142 if (ss == failed_ss)
2144 if (ss->cancel_attach)
2145 ss->cancel_attach(cgrp, &tset);
2148 out_free_group_list:
2149 flex_array_free(group);
2154 * Find the task_struct of the task to attach by vpid and pass it along to the
2155 * function to attach either it or all tasks in its threadgroup. Will lock
2156 * cgroup_mutex and threadgroup; may take task_lock of task.
2158 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2160 struct task_struct *tsk;
2161 const struct cred *cred = current_cred(), *tcred;
2164 if (!cgroup_lock_live_group(cgrp))
2170 tsk = find_task_by_vpid(pid);
2174 goto out_unlock_cgroup;
2177 * even if we're attaching all tasks in the thread group, we
2178 * only need to check permissions on one of them.
2180 tcred = __task_cred(tsk);
2181 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2182 !uid_eq(cred->euid, tcred->uid) &&
2183 !uid_eq(cred->euid, tcred->suid)) {
2186 goto out_unlock_cgroup;
2192 tsk = tsk->group_leader;
2195 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2196 * trapped in a cpuset, or RT worker may be born in a cgroup
2197 * with no rt_runtime allocated. Just say no.
2199 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
2202 goto out_unlock_cgroup;
2205 get_task_struct(tsk);
2208 threadgroup_lock(tsk);
2210 if (!thread_group_leader(tsk)) {
2212 * a race with de_thread from another thread's exec()
2213 * may strip us of our leadership, if this happens,
2214 * there is no choice but to throw this task away and
2215 * try again; this is
2216 * "double-double-toil-and-trouble-check locking".
2218 threadgroup_unlock(tsk);
2219 put_task_struct(tsk);
2220 goto retry_find_task;
2224 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2226 threadgroup_unlock(tsk);
2228 put_task_struct(tsk);
2230 mutex_unlock(&cgroup_mutex);
2235 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2236 * @from: attach to all cgroups of a given task
2237 * @tsk: the task to be attached
2239 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2241 struct cgroupfs_root *root;
2244 mutex_lock(&cgroup_mutex);
2245 for_each_active_root(root) {
2246 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2248 retval = cgroup_attach_task(from_cg, tsk, false);
2252 mutex_unlock(&cgroup_mutex);
2256 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2258 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2260 return attach_task_by_pid(cgrp, pid, false);
2263 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2265 return attach_task_by_pid(cgrp, tgid, true);
2268 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2271 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2272 if (strlen(buffer) >= PATH_MAX)
2274 if (!cgroup_lock_live_group(cgrp))
2276 mutex_lock(&cgroup_root_mutex);
2277 strcpy(cgrp->root->release_agent_path, buffer);
2278 mutex_unlock(&cgroup_root_mutex);
2279 mutex_unlock(&cgroup_mutex);
2283 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2284 struct seq_file *seq)
2286 if (!cgroup_lock_live_group(cgrp))
2288 seq_puts(seq, cgrp->root->release_agent_path);
2289 seq_putc(seq, '\n');
2290 mutex_unlock(&cgroup_mutex);
2294 static int cgroup_sane_behavior_show(struct cgroup *cgrp, struct cftype *cft,
2295 struct seq_file *seq)
2297 seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp));
2301 /* A buffer size big enough for numbers or short strings */
2302 #define CGROUP_LOCAL_BUFFER_SIZE 64
2304 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2306 const char __user *userbuf,
2307 size_t nbytes, loff_t *unused_ppos)
2309 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2315 if (nbytes >= sizeof(buffer))
2317 if (copy_from_user(buffer, userbuf, nbytes))
2320 buffer[nbytes] = 0; /* nul-terminate */
2321 if (cft->write_u64) {
2322 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2325 retval = cft->write_u64(cgrp, cft, val);
2327 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2330 retval = cft->write_s64(cgrp, cft, val);
2337 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2339 const char __user *userbuf,
2340 size_t nbytes, loff_t *unused_ppos)
2342 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2344 size_t max_bytes = cft->max_write_len;
2345 char *buffer = local_buffer;
2348 max_bytes = sizeof(local_buffer) - 1;
2349 if (nbytes >= max_bytes)
2351 /* Allocate a dynamic buffer if we need one */
2352 if (nbytes >= sizeof(local_buffer)) {
2353 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2357 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2362 buffer[nbytes] = 0; /* nul-terminate */
2363 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2367 if (buffer != local_buffer)
2372 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2373 size_t nbytes, loff_t *ppos)
2375 struct cftype *cft = __d_cft(file->f_dentry);
2376 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2378 if (cgroup_is_dead(cgrp))
2381 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2382 if (cft->write_u64 || cft->write_s64)
2383 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2384 if (cft->write_string)
2385 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2387 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2388 return ret ? ret : nbytes;
2393 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2395 char __user *buf, size_t nbytes,
2398 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2399 u64 val = cft->read_u64(cgrp, cft);
2400 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2402 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2405 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2407 char __user *buf, size_t nbytes,
2410 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2411 s64 val = cft->read_s64(cgrp, cft);
2412 int len = sprintf(tmp, "%lld\n", (long long) val);
2414 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2417 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2418 size_t nbytes, loff_t *ppos)
2420 struct cftype *cft = __d_cft(file->f_dentry);
2421 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2423 if (cgroup_is_dead(cgrp))
2427 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2429 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2431 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2436 * seqfile ops/methods for returning structured data. Currently just
2437 * supports string->u64 maps, but can be extended in future.
2440 struct cgroup_seqfile_state {
2442 struct cgroup *cgroup;
2445 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2447 struct seq_file *sf = cb->state;
2448 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2451 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2453 struct cgroup_seqfile_state *state = m->private;
2454 struct cftype *cft = state->cft;
2455 if (cft->read_map) {
2456 struct cgroup_map_cb cb = {
2457 .fill = cgroup_map_add,
2460 return cft->read_map(state->cgroup, cft, &cb);
2462 return cft->read_seq_string(state->cgroup, cft, m);
2465 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2467 struct seq_file *seq = file->private_data;
2468 kfree(seq->private);
2469 return single_release(inode, file);
2472 static const struct file_operations cgroup_seqfile_operations = {
2474 .write = cgroup_file_write,
2475 .llseek = seq_lseek,
2476 .release = cgroup_seqfile_release,
2479 static int cgroup_file_open(struct inode *inode, struct file *file)
2484 err = generic_file_open(inode, file);
2487 cft = __d_cft(file->f_dentry);
2489 if (cft->read_map || cft->read_seq_string) {
2490 struct cgroup_seqfile_state *state;
2492 state = kzalloc(sizeof(*state), GFP_USER);
2497 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2498 file->f_op = &cgroup_seqfile_operations;
2499 err = single_open(file, cgroup_seqfile_show, state);
2502 } else if (cft->open)
2503 err = cft->open(inode, file);
2510 static int cgroup_file_release(struct inode *inode, struct file *file)
2512 struct cftype *cft = __d_cft(file->f_dentry);
2514 return cft->release(inode, file);
2519 * cgroup_rename - Only allow simple rename of directories in place.
2521 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2522 struct inode *new_dir, struct dentry *new_dentry)
2525 struct cgroup_name *name, *old_name;
2526 struct cgroup *cgrp;
2529 * It's convinient to use parent dir's i_mutex to protected
2532 lockdep_assert_held(&old_dir->i_mutex);
2534 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2536 if (new_dentry->d_inode)
2538 if (old_dir != new_dir)
2541 cgrp = __d_cgrp(old_dentry);
2544 * This isn't a proper migration and its usefulness is very
2545 * limited. Disallow if sane_behavior.
2547 if (cgroup_sane_behavior(cgrp))
2550 name = cgroup_alloc_name(new_dentry);
2554 ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2560 old_name = rcu_dereference_protected(cgrp->name, true);
2561 rcu_assign_pointer(cgrp->name, name);
2563 kfree_rcu(old_name, rcu_head);
2567 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2569 if (S_ISDIR(dentry->d_inode->i_mode))
2570 return &__d_cgrp(dentry)->xattrs;
2572 return &__d_cfe(dentry)->xattrs;
2575 static inline int xattr_enabled(struct dentry *dentry)
2577 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2578 return root->flags & CGRP_ROOT_XATTR;
2581 static bool is_valid_xattr(const char *name)
2583 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2584 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2589 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2590 const void *val, size_t size, int flags)
2592 if (!xattr_enabled(dentry))
2594 if (!is_valid_xattr(name))
2596 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2599 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2601 if (!xattr_enabled(dentry))
2603 if (!is_valid_xattr(name))
2605 return simple_xattr_remove(__d_xattrs(dentry), name);
2608 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2609 void *buf, size_t size)
2611 if (!xattr_enabled(dentry))
2613 if (!is_valid_xattr(name))
2615 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2618 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2620 if (!xattr_enabled(dentry))
2622 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2625 static const struct file_operations cgroup_file_operations = {
2626 .read = cgroup_file_read,
2627 .write = cgroup_file_write,
2628 .llseek = generic_file_llseek,
2629 .open = cgroup_file_open,
2630 .release = cgroup_file_release,
2633 static const struct inode_operations cgroup_file_inode_operations = {
2634 .setxattr = cgroup_setxattr,
2635 .getxattr = cgroup_getxattr,
2636 .listxattr = cgroup_listxattr,
2637 .removexattr = cgroup_removexattr,
2640 static const struct inode_operations cgroup_dir_inode_operations = {
2641 .lookup = cgroup_lookup,
2642 .mkdir = cgroup_mkdir,
2643 .rmdir = cgroup_rmdir,
2644 .rename = cgroup_rename,
2645 .setxattr = cgroup_setxattr,
2646 .getxattr = cgroup_getxattr,
2647 .listxattr = cgroup_listxattr,
2648 .removexattr = cgroup_removexattr,
2651 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2653 if (dentry->d_name.len > NAME_MAX)
2654 return ERR_PTR(-ENAMETOOLONG);
2655 d_add(dentry, NULL);
2660 * Check if a file is a control file
2662 static inline struct cftype *__file_cft(struct file *file)
2664 if (file_inode(file)->i_fop != &cgroup_file_operations)
2665 return ERR_PTR(-EINVAL);
2666 return __d_cft(file->f_dentry);
2669 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2670 struct super_block *sb)
2672 struct inode *inode;
2676 if (dentry->d_inode)
2679 inode = cgroup_new_inode(mode, sb);
2683 if (S_ISDIR(mode)) {
2684 inode->i_op = &cgroup_dir_inode_operations;
2685 inode->i_fop = &simple_dir_operations;
2687 /* start off with i_nlink == 2 (for "." entry) */
2689 inc_nlink(dentry->d_parent->d_inode);
2692 * Control reaches here with cgroup_mutex held.
2693 * @inode->i_mutex should nest outside cgroup_mutex but we
2694 * want to populate it immediately without releasing
2695 * cgroup_mutex. As @inode isn't visible to anyone else
2696 * yet, trylock will always succeed without affecting
2699 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2700 } else if (S_ISREG(mode)) {
2702 inode->i_fop = &cgroup_file_operations;
2703 inode->i_op = &cgroup_file_inode_operations;
2705 d_instantiate(dentry, inode);
2706 dget(dentry); /* Extra count - pin the dentry in core */
2711 * cgroup_file_mode - deduce file mode of a control file
2712 * @cft: the control file in question
2714 * returns cft->mode if ->mode is not 0
2715 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2716 * returns S_IRUGO if it has only a read handler
2717 * returns S_IWUSR if it has only a write hander
2719 static umode_t cgroup_file_mode(const struct cftype *cft)
2726 if (cft->read || cft->read_u64 || cft->read_s64 ||
2727 cft->read_map || cft->read_seq_string)
2730 if (cft->write || cft->write_u64 || cft->write_s64 ||
2731 cft->write_string || cft->trigger)
2737 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2740 struct dentry *dir = cgrp->dentry;
2741 struct cgroup *parent = __d_cgrp(dir);
2742 struct dentry *dentry;
2746 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2748 if (subsys && !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
2749 strcpy(name, subsys->name);
2752 strcat(name, cft->name);
2754 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2756 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2760 dentry = lookup_one_len(name, dir, strlen(name));
2761 if (IS_ERR(dentry)) {
2762 error = PTR_ERR(dentry);
2766 cfe->type = (void *)cft;
2767 cfe->dentry = dentry;
2768 dentry->d_fsdata = cfe;
2769 simple_xattrs_init(&cfe->xattrs);
2771 mode = cgroup_file_mode(cft);
2772 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2774 list_add_tail(&cfe->node, &parent->files);
2784 * cgroup_addrm_files - add or remove files to a cgroup directory
2785 * @cgrp: the target cgroup
2786 * @subsys: the subsystem of files to be added
2787 * @cfts: array of cftypes to be added
2788 * @is_add: whether to add or remove
2790 * Depending on @is_add, add or remove files defined by @cfts on @cgrp.
2791 * All @cfts should belong to @subsys. For removals, this function never
2792 * fails. If addition fails, this function doesn't remove files already
2793 * added. The caller is responsible for cleaning up.
2795 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2796 struct cftype cfts[], bool is_add)
2801 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
2802 lockdep_assert_held(&cgroup_mutex);
2804 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2805 /* does cft->flags tell us to skip this file on @cgrp? */
2806 if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp))
2808 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2810 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2814 ret = cgroup_add_file(cgrp, subsys, cft);
2816 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2821 cgroup_rm_file(cgrp, cft);
2827 static void cgroup_cfts_prepare(void)
2828 __acquires(&cgroup_mutex)
2831 * Thanks to the entanglement with vfs inode locking, we can't walk
2832 * the existing cgroups under cgroup_mutex and create files.
2833 * Instead, we use cgroup_for_each_descendant_pre() and drop RCU
2834 * read lock before calling cgroup_addrm_files().
2836 mutex_lock(&cgroup_mutex);
2839 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2840 struct cftype *cfts, bool is_add)
2841 __releases(&cgroup_mutex)
2844 struct cgroup *cgrp, *root = &ss->root->top_cgroup;
2845 struct super_block *sb = ss->root->sb;
2846 struct dentry *prev = NULL;
2847 struct inode *inode;
2850 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2851 if (!cfts || ss->root == &cgroup_dummy_root ||
2852 !atomic_inc_not_zero(&sb->s_active)) {
2853 mutex_unlock(&cgroup_mutex);
2858 * All cgroups which are created after we drop cgroup_mutex will
2859 * have the updated set of files, so we only need to update the
2860 * cgroups created before the current @cgroup_serial_nr_next.
2862 update_before = cgroup_serial_nr_next;
2864 mutex_unlock(&cgroup_mutex);
2866 /* @root always needs to be updated */
2867 inode = root->dentry->d_inode;
2868 mutex_lock(&inode->i_mutex);
2869 mutex_lock(&cgroup_mutex);
2870 cgroup_addrm_files(root, ss, cfts, is_add);
2871 mutex_unlock(&cgroup_mutex);
2872 mutex_unlock(&inode->i_mutex);
2874 /* add/rm files for all cgroups created before */
2876 cgroup_for_each_descendant_pre(cgrp, root) {
2877 if (cgroup_is_dead(cgrp))
2880 inode = cgrp->dentry->d_inode;
2885 prev = cgrp->dentry;
2887 mutex_lock(&inode->i_mutex);
2888 mutex_lock(&cgroup_mutex);
2889 if (cgrp->serial_nr < update_before && !cgroup_is_dead(cgrp))
2890 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2891 mutex_unlock(&cgroup_mutex);
2892 mutex_unlock(&inode->i_mutex);
2898 deactivate_super(sb);
2902 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2903 * @ss: target cgroup subsystem
2904 * @cfts: zero-length name terminated array of cftypes
2906 * Register @cfts to @ss. Files described by @cfts are created for all
2907 * existing cgroups to which @ss is attached and all future cgroups will
2908 * have them too. This function can be called anytime whether @ss is
2911 * Returns 0 on successful registration, -errno on failure. Note that this
2912 * function currently returns 0 as long as @cfts registration is successful
2913 * even if some file creation attempts on existing cgroups fail.
2915 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2917 struct cftype_set *set;
2919 set = kzalloc(sizeof(*set), GFP_KERNEL);
2923 cgroup_cfts_prepare();
2925 list_add_tail(&set->node, &ss->cftsets);
2926 cgroup_cfts_commit(ss, cfts, true);
2930 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2933 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2934 * @ss: target cgroup subsystem
2935 * @cfts: zero-length name terminated array of cftypes
2937 * Unregister @cfts from @ss. Files described by @cfts are removed from
2938 * all existing cgroups to which @ss is attached and all future cgroups
2939 * won't have them either. This function can be called anytime whether @ss
2940 * is attached or not.
2942 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2943 * registered with @ss.
2945 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2947 struct cftype_set *set;
2949 cgroup_cfts_prepare();
2951 list_for_each_entry(set, &ss->cftsets, node) {
2952 if (set->cfts == cfts) {
2953 list_del(&set->node);
2955 cgroup_cfts_commit(ss, cfts, false);
2960 cgroup_cfts_commit(ss, NULL, false);
2965 * cgroup_task_count - count the number of tasks in a cgroup.
2966 * @cgrp: the cgroup in question
2968 * Return the number of tasks in the cgroup.
2970 int cgroup_task_count(const struct cgroup *cgrp)
2973 struct cgrp_cset_link *link;
2975 read_lock(&css_set_lock);
2976 list_for_each_entry(link, &cgrp->cset_links, cset_link)
2977 count += atomic_read(&link->cset->refcount);
2978 read_unlock(&css_set_lock);
2983 * Advance a list_head iterator. The iterator should be positioned at
2984 * the start of a css_set
2986 static void cgroup_advance_iter(struct cgroup *cgrp, struct cgroup_iter *it)
2988 struct list_head *l = it->cset_link;
2989 struct cgrp_cset_link *link;
2990 struct css_set *cset;
2992 /* Advance to the next non-empty css_set */
2995 if (l == &cgrp->cset_links) {
2996 it->cset_link = NULL;
2999 link = list_entry(l, struct cgrp_cset_link, cset_link);
3001 } while (list_empty(&cset->tasks));
3003 it->task = cset->tasks.next;
3007 * To reduce the fork() overhead for systems that are not actually
3008 * using their cgroups capability, we don't maintain the lists running
3009 * through each css_set to its tasks until we see the list actually
3010 * used - in other words after the first call to cgroup_iter_start().
3012 static void cgroup_enable_task_cg_lists(void)
3014 struct task_struct *p, *g;
3015 write_lock(&css_set_lock);
3016 use_task_css_set_links = 1;
3018 * We need tasklist_lock because RCU is not safe against
3019 * while_each_thread(). Besides, a forking task that has passed
3020 * cgroup_post_fork() without seeing use_task_css_set_links = 1
3021 * is not guaranteed to have its child immediately visible in the
3022 * tasklist if we walk through it with RCU.
3024 read_lock(&tasklist_lock);
3025 do_each_thread(g, p) {
3028 * We should check if the process is exiting, otherwise
3029 * it will race with cgroup_exit() in that the list
3030 * entry won't be deleted though the process has exited.
3032 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
3033 list_add(&p->cg_list, &task_css_set(p)->tasks);
3035 } while_each_thread(g, p);
3036 read_unlock(&tasklist_lock);
3037 write_unlock(&css_set_lock);
3041 * cgroup_next_sibling - find the next sibling of a given cgroup
3042 * @pos: the current cgroup
3044 * This function returns the next sibling of @pos and should be called
3045 * under RCU read lock. The only requirement is that @pos is accessible.
3046 * The next sibling is guaranteed to be returned regardless of @pos's
3049 struct cgroup *cgroup_next_sibling(struct cgroup *pos)
3051 struct cgroup *next;
3053 WARN_ON_ONCE(!rcu_read_lock_held());
3056 * @pos could already have been removed. Once a cgroup is removed,
3057 * its ->sibling.next is no longer updated when its next sibling
3058 * changes. As CGRP_DEAD assertion is serialized and happens
3059 * before the cgroup is taken off the ->sibling list, if we see it
3060 * unasserted, it's guaranteed that the next sibling hasn't
3061 * finished its grace period even if it's already removed, and thus
3062 * safe to dereference from this RCU critical section. If
3063 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
3064 * to be visible as %true here.
3066 if (likely(!cgroup_is_dead(pos))) {
3067 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3068 if (&next->sibling != &pos->parent->children)
3074 * Can't dereference the next pointer. Each cgroup is given a
3075 * monotonically increasing unique serial number and always
3076 * appended to the sibling list, so the next one can be found by
3077 * walking the parent's children until we see a cgroup with higher
3078 * serial number than @pos's.
3080 * While this path can be slow, it's taken only when either the
3081 * current cgroup is removed or iteration and removal race.
3083 list_for_each_entry_rcu(next, &pos->parent->children, sibling)
3084 if (next->serial_nr > pos->serial_nr)
3088 EXPORT_SYMBOL_GPL(cgroup_next_sibling);
3091 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
3092 * @pos: the current position (%NULL to initiate traversal)
3093 * @cgroup: cgroup whose descendants to walk
3095 * To be used by cgroup_for_each_descendant_pre(). Find the next
3096 * descendant to visit for pre-order traversal of @cgroup's descendants.
3098 * While this function requires RCU read locking, it doesn't require the
3099 * whole traversal to be contained in a single RCU critical section. This
3100 * function will return the correct next descendant as long as both @pos
3101 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3103 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
3104 struct cgroup *cgroup)
3106 struct cgroup *next;
3108 WARN_ON_ONCE(!rcu_read_lock_held());
3110 /* if first iteration, pretend we just visited @cgroup */
3114 /* visit the first child if exists */
3115 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
3119 /* no child, visit my or the closest ancestor's next sibling */
3120 while (pos != cgroup) {
3121 next = cgroup_next_sibling(pos);
3129 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
3132 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
3133 * @pos: cgroup of interest
3135 * Return the rightmost descendant of @pos. If there's no descendant,
3136 * @pos is returned. This can be used during pre-order traversal to skip
3139 * While this function requires RCU read locking, it doesn't require the
3140 * whole traversal to be contained in a single RCU critical section. This
3141 * function will return the correct rightmost descendant as long as @pos is
3144 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
3146 struct cgroup *last, *tmp;
3148 WARN_ON_ONCE(!rcu_read_lock_held());
3152 /* ->prev isn't RCU safe, walk ->next till the end */
3154 list_for_each_entry_rcu(tmp, &last->children, sibling)
3160 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
3162 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3164 struct cgroup *last;
3168 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3176 * cgroup_next_descendant_post - find the next descendant for post-order walk
3177 * @pos: the current position (%NULL to initiate traversal)
3178 * @cgroup: cgroup whose descendants to walk
3180 * To be used by cgroup_for_each_descendant_post(). Find the next
3181 * descendant to visit for post-order traversal of @cgroup's descendants.
3183 * While this function requires RCU read locking, it doesn't require the
3184 * whole traversal to be contained in a single RCU critical section. This
3185 * function will return the correct next descendant as long as both @pos
3186 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3188 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3189 struct cgroup *cgroup)
3191 struct cgroup *next;
3193 WARN_ON_ONCE(!rcu_read_lock_held());
3195 /* if first iteration, visit the leftmost descendant */
3197 next = cgroup_leftmost_descendant(cgroup);
3198 return next != cgroup ? next : NULL;
3201 /* if there's an unvisited sibling, visit its leftmost descendant */
3202 next = cgroup_next_sibling(pos);
3204 return cgroup_leftmost_descendant(next);
3206 /* no sibling left, visit parent */
3208 return next != cgroup ? next : NULL;
3210 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3212 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3213 __acquires(css_set_lock)
3216 * The first time anyone tries to iterate across a cgroup,
3217 * we need to enable the list linking each css_set to its
3218 * tasks, and fix up all existing tasks.
3220 if (!use_task_css_set_links)
3221 cgroup_enable_task_cg_lists();
3223 read_lock(&css_set_lock);
3224 it->cset_link = &cgrp->cset_links;
3225 cgroup_advance_iter(cgrp, it);
3228 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3229 struct cgroup_iter *it)
3231 struct task_struct *res;
3232 struct list_head *l = it->task;
3233 struct cgrp_cset_link *link;
3235 /* If the iterator cg is NULL, we have no tasks */
3238 res = list_entry(l, struct task_struct, cg_list);
3239 /* Advance iterator to find next entry */
3241 link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link);
3242 if (l == &link->cset->tasks) {
3243 /* We reached the end of this task list - move on to
3244 * the next cg_cgroup_link */
3245 cgroup_advance_iter(cgrp, it);
3252 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3253 __releases(css_set_lock)
3255 read_unlock(&css_set_lock);
3258 static inline int started_after_time(struct task_struct *t1,
3259 struct timespec *time,
3260 struct task_struct *t2)
3262 int start_diff = timespec_compare(&t1->start_time, time);
3263 if (start_diff > 0) {
3265 } else if (start_diff < 0) {
3269 * Arbitrarily, if two processes started at the same
3270 * time, we'll say that the lower pointer value
3271 * started first. Note that t2 may have exited by now
3272 * so this may not be a valid pointer any longer, but
3273 * that's fine - it still serves to distinguish
3274 * between two tasks started (effectively) simultaneously.
3281 * This function is a callback from heap_insert() and is used to order
3283 * In this case we order the heap in descending task start time.
3285 static inline int started_after(void *p1, void *p2)
3287 struct task_struct *t1 = p1;
3288 struct task_struct *t2 = p2;
3289 return started_after_time(t1, &t2->start_time, t2);
3293 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3294 * @scan: struct cgroup_scanner containing arguments for the scan
3296 * Arguments include pointers to callback functions test_task() and
3298 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3299 * and if it returns true, call process_task() for it also.
3300 * The test_task pointer may be NULL, meaning always true (select all tasks).
3301 * Effectively duplicates cgroup_iter_{start,next,end}()
3302 * but does not lock css_set_lock for the call to process_task().
3303 * The struct cgroup_scanner may be embedded in any structure of the caller's
3305 * It is guaranteed that process_task() will act on every task that
3306 * is a member of the cgroup for the duration of this call. This
3307 * function may or may not call process_task() for tasks that exit
3308 * or move to a different cgroup during the call, or are forked or
3309 * move into the cgroup during the call.
3311 * Note that test_task() may be called with locks held, and may in some
3312 * situations be called multiple times for the same task, so it should
3314 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3315 * pre-allocated and will be used for heap operations (and its "gt" member will
3316 * be overwritten), else a temporary heap will be used (allocation of which
3317 * may cause this function to fail).
3319 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3322 struct cgroup_iter it;
3323 struct task_struct *p, *dropped;
3324 /* Never dereference latest_task, since it's not refcounted */
3325 struct task_struct *latest_task = NULL;
3326 struct ptr_heap tmp_heap;
3327 struct ptr_heap *heap;
3328 struct timespec latest_time = { 0, 0 };
3331 /* The caller supplied our heap and pre-allocated its memory */
3333 heap->gt = &started_after;
3335 /* We need to allocate our own heap memory */
3337 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3339 /* cannot allocate the heap */
3345 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3346 * to determine which are of interest, and using the scanner's
3347 * "process_task" callback to process any of them that need an update.
3348 * Since we don't want to hold any locks during the task updates,
3349 * gather tasks to be processed in a heap structure.
3350 * The heap is sorted by descending task start time.
3351 * If the statically-sized heap fills up, we overflow tasks that
3352 * started later, and in future iterations only consider tasks that
3353 * started after the latest task in the previous pass. This
3354 * guarantees forward progress and that we don't miss any tasks.
3357 cgroup_iter_start(scan->cg, &it);
3358 while ((p = cgroup_iter_next(scan->cg, &it))) {
3360 * Only affect tasks that qualify per the caller's callback,
3361 * if he provided one
3363 if (scan->test_task && !scan->test_task(p, scan))
3366 * Only process tasks that started after the last task
3369 if (!started_after_time(p, &latest_time, latest_task))
3371 dropped = heap_insert(heap, p);
3372 if (dropped == NULL) {
3374 * The new task was inserted; the heap wasn't
3378 } else if (dropped != p) {
3380 * The new task was inserted, and pushed out a
3384 put_task_struct(dropped);
3387 * Else the new task was newer than anything already in
3388 * the heap and wasn't inserted
3391 cgroup_iter_end(scan->cg, &it);
3394 for (i = 0; i < heap->size; i++) {
3395 struct task_struct *q = heap->ptrs[i];
3397 latest_time = q->start_time;
3400 /* Process the task per the caller's callback */
3401 scan->process_task(q, scan);
3405 * If we had to process any tasks at all, scan again
3406 * in case some of them were in the middle of forking
3407 * children that didn't get processed.
3408 * Not the most efficient way to do it, but it avoids
3409 * having to take callback_mutex in the fork path
3413 if (heap == &tmp_heap)
3414 heap_free(&tmp_heap);
3418 static void cgroup_transfer_one_task(struct task_struct *task,
3419 struct cgroup_scanner *scan)
3421 struct cgroup *new_cgroup = scan->data;
3423 mutex_lock(&cgroup_mutex);
3424 cgroup_attach_task(new_cgroup, task, false);
3425 mutex_unlock(&cgroup_mutex);
3429 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3430 * @to: cgroup to which the tasks will be moved
3431 * @from: cgroup in which the tasks currently reside
3433 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3435 struct cgroup_scanner scan;
3438 scan.test_task = NULL; /* select all tasks in cgroup */
3439 scan.process_task = cgroup_transfer_one_task;
3443 return cgroup_scan_tasks(&scan);
3447 * Stuff for reading the 'tasks'/'procs' files.
3449 * Reading this file can return large amounts of data if a cgroup has
3450 * *lots* of attached tasks. So it may need several calls to read(),
3451 * but we cannot guarantee that the information we produce is correct
3452 * unless we produce it entirely atomically.
3456 /* which pidlist file are we talking about? */
3457 enum cgroup_filetype {
3463 * A pidlist is a list of pids that virtually represents the contents of one
3464 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3465 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3468 struct cgroup_pidlist {
3470 * used to find which pidlist is wanted. doesn't change as long as
3471 * this particular list stays in the list.
3473 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3476 /* how many elements the above list has */
3478 /* how many files are using the current array */
3480 /* each of these stored in a list by its cgroup */
3481 struct list_head links;
3482 /* pointer to the cgroup we belong to, for list removal purposes */
3483 struct cgroup *owner;
3484 /* protects the other fields */
3485 struct rw_semaphore mutex;
3489 * The following two functions "fix" the issue where there are more pids
3490 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3491 * TODO: replace with a kernel-wide solution to this problem
3493 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3494 static void *pidlist_allocate(int count)
3496 if (PIDLIST_TOO_LARGE(count))
3497 return vmalloc(count * sizeof(pid_t));
3499 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3501 static void pidlist_free(void *p)
3503 if (is_vmalloc_addr(p))
3510 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3511 * Returns the number of unique elements.
3513 static int pidlist_uniq(pid_t *list, int length)
3518 * we presume the 0th element is unique, so i starts at 1. trivial
3519 * edge cases first; no work needs to be done for either
3521 if (length == 0 || length == 1)
3523 /* src and dest walk down the list; dest counts unique elements */
3524 for (src = 1; src < length; src++) {
3525 /* find next unique element */
3526 while (list[src] == list[src-1]) {
3531 /* dest always points to where the next unique element goes */
3532 list[dest] = list[src];
3539 static int cmppid(const void *a, const void *b)
3541 return *(pid_t *)a - *(pid_t *)b;
3545 * find the appropriate pidlist for our purpose (given procs vs tasks)
3546 * returns with the lock on that pidlist already held, and takes care
3547 * of the use count, or returns NULL with no locks held if we're out of
3550 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3551 enum cgroup_filetype type)
3553 struct cgroup_pidlist *l;
3554 /* don't need task_nsproxy() if we're looking at ourself */
3555 struct pid_namespace *ns = task_active_pid_ns(current);
3558 * We can't drop the pidlist_mutex before taking the l->mutex in case
3559 * the last ref-holder is trying to remove l from the list at the same
3560 * time. Holding the pidlist_mutex precludes somebody taking whichever
3561 * list we find out from under us - compare release_pid_array().
3563 mutex_lock(&cgrp->pidlist_mutex);
3564 list_for_each_entry(l, &cgrp->pidlists, links) {
3565 if (l->key.type == type && l->key.ns == ns) {
3566 /* make sure l doesn't vanish out from under us */
3567 down_write(&l->mutex);
3568 mutex_unlock(&cgrp->pidlist_mutex);
3572 /* entry not found; create a new one */
3573 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3575 mutex_unlock(&cgrp->pidlist_mutex);
3578 init_rwsem(&l->mutex);
3579 down_write(&l->mutex);
3581 l->key.ns = get_pid_ns(ns);
3583 list_add(&l->links, &cgrp->pidlists);
3584 mutex_unlock(&cgrp->pidlist_mutex);
3589 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3591 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3592 struct cgroup_pidlist **lp)
3596 int pid, n = 0; /* used for populating the array */
3597 struct cgroup_iter it;
3598 struct task_struct *tsk;
3599 struct cgroup_pidlist *l;
3602 * If cgroup gets more users after we read count, we won't have
3603 * enough space - tough. This race is indistinguishable to the
3604 * caller from the case that the additional cgroup users didn't
3605 * show up until sometime later on.
3607 length = cgroup_task_count(cgrp);
3608 array = pidlist_allocate(length);
3611 /* now, populate the array */
3612 cgroup_iter_start(cgrp, &it);
3613 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3614 if (unlikely(n == length))
3616 /* get tgid or pid for procs or tasks file respectively */
3617 if (type == CGROUP_FILE_PROCS)
3618 pid = task_tgid_vnr(tsk);
3620 pid = task_pid_vnr(tsk);
3621 if (pid > 0) /* make sure to only use valid results */
3624 cgroup_iter_end(cgrp, &it);
3626 /* now sort & (if procs) strip out duplicates */
3627 sort(array, length, sizeof(pid_t), cmppid, NULL);
3628 if (type == CGROUP_FILE_PROCS)
3629 length = pidlist_uniq(array, length);
3630 l = cgroup_pidlist_find(cgrp, type);
3632 pidlist_free(array);
3635 /* store array, freeing old if necessary - lock already held */
3636 pidlist_free(l->list);
3640 up_write(&l->mutex);
3646 * cgroupstats_build - build and fill cgroupstats
3647 * @stats: cgroupstats to fill information into
3648 * @dentry: A dentry entry belonging to the cgroup for which stats have
3651 * Build and fill cgroupstats so that taskstats can export it to user
3654 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3657 struct cgroup *cgrp;
3658 struct cgroup_iter it;
3659 struct task_struct *tsk;
3662 * Validate dentry by checking the superblock operations,
3663 * and make sure it's a directory.
3665 if (dentry->d_sb->s_op != &cgroup_ops ||
3666 !S_ISDIR(dentry->d_inode->i_mode))
3670 cgrp = dentry->d_fsdata;
3672 cgroup_iter_start(cgrp, &it);
3673 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3674 switch (tsk->state) {
3676 stats->nr_running++;
3678 case TASK_INTERRUPTIBLE:
3679 stats->nr_sleeping++;
3681 case TASK_UNINTERRUPTIBLE:
3682 stats->nr_uninterruptible++;
3685 stats->nr_stopped++;
3688 if (delayacct_is_task_waiting_on_io(tsk))
3689 stats->nr_io_wait++;
3693 cgroup_iter_end(cgrp, &it);
3701 * seq_file methods for the tasks/procs files. The seq_file position is the
3702 * next pid to display; the seq_file iterator is a pointer to the pid
3703 * in the cgroup->l->list array.
3706 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3709 * Initially we receive a position value that corresponds to
3710 * one more than the last pid shown (or 0 on the first call or
3711 * after a seek to the start). Use a binary-search to find the
3712 * next pid to display, if any
3714 struct cgroup_pidlist *l = s->private;
3715 int index = 0, pid = *pos;
3718 down_read(&l->mutex);
3720 int end = l->length;
3722 while (index < end) {
3723 int mid = (index + end) / 2;
3724 if (l->list[mid] == pid) {
3727 } else if (l->list[mid] <= pid)
3733 /* If we're off the end of the array, we're done */
3734 if (index >= l->length)
3736 /* Update the abstract position to be the actual pid that we found */
3737 iter = l->list + index;
3742 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3744 struct cgroup_pidlist *l = s->private;
3748 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3750 struct cgroup_pidlist *l = s->private;
3752 pid_t *end = l->list + l->length;
3754 * Advance to the next pid in the array. If this goes off the
3766 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3768 return seq_printf(s, "%d\n", *(int *)v);
3772 * seq_operations functions for iterating on pidlists through seq_file -
3773 * independent of whether it's tasks or procs
3775 static const struct seq_operations cgroup_pidlist_seq_operations = {
3776 .start = cgroup_pidlist_start,
3777 .stop = cgroup_pidlist_stop,
3778 .next = cgroup_pidlist_next,
3779 .show = cgroup_pidlist_show,
3782 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3785 * the case where we're the last user of this particular pidlist will
3786 * have us remove it from the cgroup's list, which entails taking the
3787 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3788 * pidlist_mutex, we have to take pidlist_mutex first.
3790 mutex_lock(&l->owner->pidlist_mutex);
3791 down_write(&l->mutex);
3792 BUG_ON(!l->use_count);
3793 if (!--l->use_count) {
3794 /* we're the last user if refcount is 0; remove and free */
3795 list_del(&l->links);
3796 mutex_unlock(&l->owner->pidlist_mutex);
3797 pidlist_free(l->list);
3798 put_pid_ns(l->key.ns);
3799 up_write(&l->mutex);
3803 mutex_unlock(&l->owner->pidlist_mutex);
3804 up_write(&l->mutex);
3807 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3809 struct cgroup_pidlist *l;
3810 if (!(file->f_mode & FMODE_READ))
3813 * the seq_file will only be initialized if the file was opened for
3814 * reading; hence we check if it's not null only in that case.
3816 l = ((struct seq_file *)file->private_data)->private;
3817 cgroup_release_pid_array(l);
3818 return seq_release(inode, file);
3821 static const struct file_operations cgroup_pidlist_operations = {
3823 .llseek = seq_lseek,
3824 .write = cgroup_file_write,
3825 .release = cgroup_pidlist_release,
3829 * The following functions handle opens on a file that displays a pidlist
3830 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3833 /* helper function for the two below it */
3834 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3836 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3837 struct cgroup_pidlist *l;
3840 /* Nothing to do for write-only files */
3841 if (!(file->f_mode & FMODE_READ))
3844 /* have the array populated */
3845 retval = pidlist_array_load(cgrp, type, &l);
3848 /* configure file information */
3849 file->f_op = &cgroup_pidlist_operations;
3851 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3853 cgroup_release_pid_array(l);
3856 ((struct seq_file *)file->private_data)->private = l;
3859 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3861 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3863 static int cgroup_procs_open(struct inode *unused, struct file *file)
3865 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3868 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3871 return notify_on_release(cgrp);
3874 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3878 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3880 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3882 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3887 * When dput() is called asynchronously, if umount has been done and
3888 * then deactivate_super() in cgroup_free_fn() kills the superblock,
3889 * there's a small window that vfs will see the root dentry with non-zero
3890 * refcnt and trigger BUG().
3892 * That's why we hold a reference before dput() and drop it right after.
3894 static void cgroup_dput(struct cgroup *cgrp)
3896 struct super_block *sb = cgrp->root->sb;
3898 atomic_inc(&sb->s_active);
3900 deactivate_super(sb);
3904 * Unregister event and free resources.
3906 * Gets called from workqueue.
3908 static void cgroup_event_remove(struct work_struct *work)
3910 struct cgroup_event *event = container_of(work, struct cgroup_event,
3912 struct cgroup *cgrp = event->cgrp;
3914 remove_wait_queue(event->wqh, &event->wait);
3916 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3918 /* Notify userspace the event is going away. */
3919 eventfd_signal(event->eventfd, 1);
3921 eventfd_ctx_put(event->eventfd);
3927 * Gets called on POLLHUP on eventfd when user closes it.
3929 * Called with wqh->lock held and interrupts disabled.
3931 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3932 int sync, void *key)
3934 struct cgroup_event *event = container_of(wait,
3935 struct cgroup_event, wait);
3936 struct cgroup *cgrp = event->cgrp;
3937 unsigned long flags = (unsigned long)key;
3939 if (flags & POLLHUP) {
3941 * If the event has been detached at cgroup removal, we
3942 * can simply return knowing the other side will cleanup
3945 * We can't race against event freeing since the other
3946 * side will require wqh->lock via remove_wait_queue(),
3949 spin_lock(&cgrp->event_list_lock);
3950 if (!list_empty(&event->list)) {
3951 list_del_init(&event->list);
3953 * We are in atomic context, but cgroup_event_remove()
3954 * may sleep, so we have to call it in workqueue.
3956 schedule_work(&event->remove);
3958 spin_unlock(&cgrp->event_list_lock);
3964 static void cgroup_event_ptable_queue_proc(struct file *file,
3965 wait_queue_head_t *wqh, poll_table *pt)
3967 struct cgroup_event *event = container_of(pt,
3968 struct cgroup_event, pt);
3971 add_wait_queue(wqh, &event->wait);
3975 * Parse input and register new cgroup event handler.
3977 * Input must be in format '<event_fd> <control_fd> <args>'.
3978 * Interpretation of args is defined by control file implementation.
3980 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3983 struct cgroup_event *event = NULL;
3984 struct cgroup *cgrp_cfile;
3985 unsigned int efd, cfd;
3986 struct file *efile = NULL;
3987 struct file *cfile = NULL;
3991 efd = simple_strtoul(buffer, &endp, 10);
3996 cfd = simple_strtoul(buffer, &endp, 10);
3997 if ((*endp != ' ') && (*endp != '\0'))
4001 event = kzalloc(sizeof(*event), GFP_KERNEL);
4005 INIT_LIST_HEAD(&event->list);
4006 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
4007 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
4008 INIT_WORK(&event->remove, cgroup_event_remove);
4010 efile = eventfd_fget(efd);
4011 if (IS_ERR(efile)) {
4012 ret = PTR_ERR(efile);
4016 event->eventfd = eventfd_ctx_fileget(efile);
4017 if (IS_ERR(event->eventfd)) {
4018 ret = PTR_ERR(event->eventfd);
4028 /* the process need read permission on control file */
4029 /* AV: shouldn't we check that it's been opened for read instead? */
4030 ret = inode_permission(file_inode(cfile), MAY_READ);
4034 event->cft = __file_cft(cfile);
4035 if (IS_ERR(event->cft)) {
4036 ret = PTR_ERR(event->cft);
4041 * The file to be monitored must be in the same cgroup as
4042 * cgroup.event_control is.
4044 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
4045 if (cgrp_cfile != cgrp) {
4050 if (!event->cft->register_event || !event->cft->unregister_event) {
4055 ret = event->cft->register_event(cgrp, event->cft,
4056 event->eventfd, buffer);
4060 efile->f_op->poll(efile, &event->pt);
4063 * Events should be removed after rmdir of cgroup directory, but before
4064 * destroying subsystem state objects. Let's take reference to cgroup
4065 * directory dentry to do that.
4069 spin_lock(&cgrp->event_list_lock);
4070 list_add(&event->list, &cgrp->event_list);
4071 spin_unlock(&cgrp->event_list_lock);
4082 if (event && event->eventfd && !IS_ERR(event->eventfd))
4083 eventfd_ctx_put(event->eventfd);
4085 if (!IS_ERR_OR_NULL(efile))
4093 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
4096 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4099 static int cgroup_clone_children_write(struct cgroup *cgrp,
4104 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4106 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4110 static struct cftype cgroup_base_files[] = {
4112 .name = "cgroup.procs",
4113 .open = cgroup_procs_open,
4114 .write_u64 = cgroup_procs_write,
4115 .release = cgroup_pidlist_release,
4116 .mode = S_IRUGO | S_IWUSR,
4119 .name = "cgroup.event_control",
4120 .write_string = cgroup_write_event_control,
4124 .name = "cgroup.clone_children",
4125 .flags = CFTYPE_INSANE,
4126 .read_u64 = cgroup_clone_children_read,
4127 .write_u64 = cgroup_clone_children_write,
4130 .name = "cgroup.sane_behavior",
4131 .flags = CFTYPE_ONLY_ON_ROOT,
4132 .read_seq_string = cgroup_sane_behavior_show,
4136 * Historical crazy stuff. These don't have "cgroup." prefix and
4137 * don't exist if sane_behavior. If you're depending on these, be
4138 * prepared to be burned.
4142 .flags = CFTYPE_INSANE, /* use "procs" instead */
4143 .open = cgroup_tasks_open,
4144 .write_u64 = cgroup_tasks_write,
4145 .release = cgroup_pidlist_release,
4146 .mode = S_IRUGO | S_IWUSR,
4149 .name = "notify_on_release",
4150 .flags = CFTYPE_INSANE,
4151 .read_u64 = cgroup_read_notify_on_release,
4152 .write_u64 = cgroup_write_notify_on_release,
4155 .name = "release_agent",
4156 .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT,
4157 .read_seq_string = cgroup_release_agent_show,
4158 .write_string = cgroup_release_agent_write,
4159 .max_write_len = PATH_MAX,
4165 * cgroup_populate_dir - selectively creation of files in a directory
4166 * @cgrp: target cgroup
4167 * @base_files: true if the base files should be added
4168 * @subsys_mask: mask of the subsystem ids whose files should be added
4170 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
4171 unsigned long subsys_mask)
4174 struct cgroup_subsys *ss;
4177 err = cgroup_addrm_files(cgrp, NULL, cgroup_base_files, true);
4182 /* process cftsets of each subsystem */
4183 for_each_root_subsys(cgrp->root, ss) {
4184 struct cftype_set *set;
4185 if (!test_bit(ss->subsys_id, &subsys_mask))
4188 list_for_each_entry(set, &ss->cftsets, node)
4189 cgroup_addrm_files(cgrp, ss, set->cfts, true);
4192 /* This cgroup is ready now */
4193 for_each_root_subsys(cgrp->root, ss) {
4194 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4195 struct css_id *id = rcu_dereference_protected(css->id, true);
4198 * Update id->css pointer and make this css visible from
4199 * CSS ID functions. This pointer will be dereferened
4200 * from RCU-read-side without locks.
4203 rcu_assign_pointer(id->css, css);
4209 static void css_dput_fn(struct work_struct *work)
4211 struct cgroup_subsys_state *css =
4212 container_of(work, struct cgroup_subsys_state, dput_work);
4214 cgroup_dput(css->cgroup);
4217 static void css_release(struct percpu_ref *ref)
4219 struct cgroup_subsys_state *css =
4220 container_of(ref, struct cgroup_subsys_state, refcnt);
4222 schedule_work(&css->dput_work);
4225 static void init_cgroup_css(struct cgroup_subsys_state *css,
4226 struct cgroup_subsys *ss,
4227 struct cgroup *cgrp)
4232 if (cgrp == cgroup_dummy_top)
4233 css->flags |= CSS_ROOT;
4234 BUG_ON(cgrp->subsys[ss->subsys_id]);
4235 cgrp->subsys[ss->subsys_id] = css;
4238 * css holds an extra ref to @cgrp->dentry which is put on the last
4239 * css_put(). dput() requires process context, which css_put() may
4240 * be called without. @css->dput_work will be used to invoke
4241 * dput() asynchronously from css_put().
4243 INIT_WORK(&css->dput_work, css_dput_fn);
4246 /* invoke ->post_create() on a new CSS and mark it online if successful */
4247 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4251 lockdep_assert_held(&cgroup_mutex);
4254 ret = ss->css_online(cgrp);
4256 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4260 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4261 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4262 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4264 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4266 lockdep_assert_held(&cgroup_mutex);
4268 if (!(css->flags & CSS_ONLINE))
4271 if (ss->css_offline)
4272 ss->css_offline(cgrp);
4274 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4278 * cgroup_create - create a cgroup
4279 * @parent: cgroup that will be parent of the new cgroup
4280 * @dentry: dentry of the new cgroup
4281 * @mode: mode to set on new inode
4283 * Must be called with the mutex on the parent inode held
4285 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4288 struct cgroup *cgrp;
4289 struct cgroup_name *name;
4290 struct cgroupfs_root *root = parent->root;
4292 struct cgroup_subsys *ss;
4293 struct super_block *sb = root->sb;
4295 /* allocate the cgroup and its ID, 0 is reserved for the root */
4296 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4300 name = cgroup_alloc_name(dentry);
4303 rcu_assign_pointer(cgrp->name, name);
4305 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4310 * Only live parents can have children. Note that the liveliness
4311 * check isn't strictly necessary because cgroup_mkdir() and
4312 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4313 * anyway so that locking is contained inside cgroup proper and we
4314 * don't get nasty surprises if we ever grow another caller.
4316 if (!cgroup_lock_live_group(parent)) {
4321 /* Grab a reference on the superblock so the hierarchy doesn't
4322 * get deleted on unmount if there are child cgroups. This
4323 * can be done outside cgroup_mutex, since the sb can't
4324 * disappear while someone has an open control file on the
4326 atomic_inc(&sb->s_active);
4328 init_cgroup_housekeeping(cgrp);
4330 dentry->d_fsdata = cgrp;
4331 cgrp->dentry = dentry;
4333 cgrp->parent = parent;
4334 cgrp->root = parent->root;
4336 if (notify_on_release(parent))
4337 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4339 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4340 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4342 for_each_root_subsys(root, ss) {
4343 struct cgroup_subsys_state *css;
4345 css = ss->css_alloc(cgrp);
4351 err = percpu_ref_init(&css->refcnt, css_release);
4355 init_cgroup_css(css, ss, cgrp);
4358 err = alloc_css_id(ss, parent, cgrp);
4365 * Create directory. cgroup_create_file() returns with the new
4366 * directory locked on success so that it can be populated without
4367 * dropping cgroup_mutex.
4369 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4372 lockdep_assert_held(&dentry->d_inode->i_mutex);
4374 cgrp->serial_nr = cgroup_serial_nr_next++;
4376 /* allocation complete, commit to creation */
4377 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4378 root->number_of_cgroups++;
4380 /* each css holds a ref to the cgroup's dentry */
4381 for_each_root_subsys(root, ss)
4384 /* hold a ref to the parent's dentry */
4385 dget(parent->dentry);
4387 /* creation succeeded, notify subsystems */
4388 for_each_root_subsys(root, ss) {
4389 err = online_css(ss, cgrp);
4393 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4395 pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4396 current->comm, current->pid, ss->name);
4397 if (!strcmp(ss->name, "memory"))
4398 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4399 ss->warned_broken_hierarchy = true;
4403 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4407 mutex_unlock(&cgroup_mutex);
4408 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4413 for_each_root_subsys(root, ss) {
4414 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4417 percpu_ref_cancel_init(&css->refcnt);
4421 mutex_unlock(&cgroup_mutex);
4422 /* Release the reference count that we took on the superblock */
4423 deactivate_super(sb);
4425 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4427 kfree(rcu_dereference_raw(cgrp->name));
4433 cgroup_destroy_locked(cgrp);
4434 mutex_unlock(&cgroup_mutex);
4435 mutex_unlock(&dentry->d_inode->i_mutex);
4439 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4441 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4443 /* the vfs holds inode->i_mutex already */
4444 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4447 static void cgroup_css_killed(struct cgroup *cgrp)
4449 if (!atomic_dec_and_test(&cgrp->css_kill_cnt))
4452 /* percpu ref's of all css's are killed, kick off the next step */
4453 INIT_WORK(&cgrp->destroy_work, cgroup_offline_fn);
4454 schedule_work(&cgrp->destroy_work);
4457 static void css_ref_killed_fn(struct percpu_ref *ref)
4459 struct cgroup_subsys_state *css =
4460 container_of(ref, struct cgroup_subsys_state, refcnt);
4462 cgroup_css_killed(css->cgroup);
4466 * cgroup_destroy_locked - the first stage of cgroup destruction
4467 * @cgrp: cgroup to be destroyed
4469 * css's make use of percpu refcnts whose killing latency shouldn't be
4470 * exposed to userland and are RCU protected. Also, cgroup core needs to
4471 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
4472 * invoked. To satisfy all the requirements, destruction is implemented in
4473 * the following two steps.
4475 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4476 * userland visible parts and start killing the percpu refcnts of
4477 * css's. Set up so that the next stage will be kicked off once all
4478 * the percpu refcnts are confirmed to be killed.
4480 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4481 * rest of destruction. Once all cgroup references are gone, the
4482 * cgroup is RCU-freed.
4484 * This function implements s1. After this step, @cgrp is gone as far as
4485 * the userland is concerned and a new cgroup with the same name may be
4486 * created. As cgroup doesn't care about the names internally, this
4487 * doesn't cause any problem.
4489 static int cgroup_destroy_locked(struct cgroup *cgrp)
4490 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4492 struct dentry *d = cgrp->dentry;
4493 struct cgroup_event *event, *tmp;
4494 struct cgroup_subsys *ss;
4497 lockdep_assert_held(&d->d_inode->i_mutex);
4498 lockdep_assert_held(&cgroup_mutex);
4501 * css_set_lock synchronizes access to ->cset_links and prevents
4502 * @cgrp from being removed while __put_css_set() is in progress.
4504 read_lock(&css_set_lock);
4505 empty = list_empty(&cgrp->cset_links) && list_empty(&cgrp->children);
4506 read_unlock(&css_set_lock);
4511 * Block new css_tryget() by killing css refcnts. cgroup core
4512 * guarantees that, by the time ->css_offline() is invoked, no new
4513 * css reference will be given out via css_tryget(). We can't
4514 * simply call percpu_ref_kill() and proceed to offlining css's
4515 * because percpu_ref_kill() doesn't guarantee that the ref is seen
4516 * as killed on all CPUs on return.
4518 * Use percpu_ref_kill_and_confirm() to get notifications as each
4519 * css is confirmed to be seen as killed on all CPUs. The
4520 * notification callback keeps track of the number of css's to be
4521 * killed and schedules cgroup_offline_fn() to perform the rest of
4522 * destruction once the percpu refs of all css's are confirmed to
4525 atomic_set(&cgrp->css_kill_cnt, 1);
4526 for_each_root_subsys(cgrp->root, ss) {
4527 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4530 * Killing would put the base ref, but we need to keep it
4531 * alive until after ->css_offline.
4533 percpu_ref_get(&css->refcnt);
4535 atomic_inc(&cgrp->css_kill_cnt);
4536 percpu_ref_kill_and_confirm(&css->refcnt, css_ref_killed_fn);
4538 cgroup_css_killed(cgrp);
4541 * Mark @cgrp dead. This prevents further task migration and child
4542 * creation by disabling cgroup_lock_live_group(). Note that
4543 * CGRP_DEAD assertion is depended upon by cgroup_next_sibling() to
4544 * resume iteration after dropping RCU read lock. See
4545 * cgroup_next_sibling() for details.
4547 set_bit(CGRP_DEAD, &cgrp->flags);
4549 /* CGRP_DEAD is set, remove from ->release_list for the last time */
4550 raw_spin_lock(&release_list_lock);
4551 if (!list_empty(&cgrp->release_list))
4552 list_del_init(&cgrp->release_list);
4553 raw_spin_unlock(&release_list_lock);
4556 * Clear and remove @cgrp directory. The removal puts the base ref
4557 * but we aren't quite done with @cgrp yet, so hold onto it.
4559 cgroup_clear_dir(cgrp, true, cgrp->root->subsys_mask);
4561 cgroup_d_remove_dir(d);
4564 * Unregister events and notify userspace.
4565 * Notify userspace about cgroup removing only after rmdir of cgroup
4566 * directory to avoid race between userspace and kernelspace.
4568 spin_lock(&cgrp->event_list_lock);
4569 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4570 list_del_init(&event->list);
4571 schedule_work(&event->remove);
4573 spin_unlock(&cgrp->event_list_lock);
4579 * cgroup_offline_fn - the second step of cgroup destruction
4580 * @work: cgroup->destroy_free_work
4582 * This function is invoked from a work item for a cgroup which is being
4583 * destroyed after the percpu refcnts of all css's are guaranteed to be
4584 * seen as killed on all CPUs, and performs the rest of destruction. This
4585 * is the second step of destruction described in the comment above
4586 * cgroup_destroy_locked().
4588 static void cgroup_offline_fn(struct work_struct *work)
4590 struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work);
4591 struct cgroup *parent = cgrp->parent;
4592 struct dentry *d = cgrp->dentry;
4593 struct cgroup_subsys *ss;
4595 mutex_lock(&cgroup_mutex);
4598 * css_tryget() is guaranteed to fail now. Tell subsystems to
4599 * initate destruction.
4601 for_each_root_subsys(cgrp->root, ss)
4602 offline_css(ss, cgrp);
4605 * Put the css refs from cgroup_destroy_locked(). Each css holds
4606 * an extra reference to the cgroup's dentry and cgroup removal
4607 * proceeds regardless of css refs. On the last put of each css,
4608 * whenever that may be, the extra dentry ref is put so that dentry
4609 * destruction happens only after all css's are released.
4611 for_each_root_subsys(cgrp->root, ss)
4612 css_put(cgrp->subsys[ss->subsys_id]);
4614 /* delete this cgroup from parent->children */
4615 list_del_rcu(&cgrp->sibling);
4619 set_bit(CGRP_RELEASABLE, &parent->flags);
4620 check_for_release(parent);
4622 mutex_unlock(&cgroup_mutex);
4625 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4629 mutex_lock(&cgroup_mutex);
4630 ret = cgroup_destroy_locked(dentry->d_fsdata);
4631 mutex_unlock(&cgroup_mutex);
4636 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4638 INIT_LIST_HEAD(&ss->cftsets);
4641 * base_cftset is embedded in subsys itself, no need to worry about
4644 if (ss->base_cftypes) {
4645 ss->base_cftset.cfts = ss->base_cftypes;
4646 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4650 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4652 struct cgroup_subsys_state *css;
4654 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4656 mutex_lock(&cgroup_mutex);
4658 /* init base cftset */
4659 cgroup_init_cftsets(ss);
4661 /* Create the top cgroup state for this subsystem */
4662 list_add(&ss->sibling, &cgroup_dummy_root.subsys_list);
4663 ss->root = &cgroup_dummy_root;
4664 css = ss->css_alloc(cgroup_dummy_top);
4665 /* We don't handle early failures gracefully */
4666 BUG_ON(IS_ERR(css));
4667 init_cgroup_css(css, ss, cgroup_dummy_top);
4669 /* Update the init_css_set to contain a subsys
4670 * pointer to this state - since the subsystem is
4671 * newly registered, all tasks and hence the
4672 * init_css_set is in the subsystem's top cgroup. */
4673 init_css_set.subsys[ss->subsys_id] = css;
4675 need_forkexit_callback |= ss->fork || ss->exit;
4677 /* At system boot, before all subsystems have been
4678 * registered, no tasks have been forked, so we don't
4679 * need to invoke fork callbacks here. */
4680 BUG_ON(!list_empty(&init_task.tasks));
4682 BUG_ON(online_css(ss, cgroup_dummy_top));
4684 mutex_unlock(&cgroup_mutex);
4686 /* this function shouldn't be used with modular subsystems, since they
4687 * need to register a subsys_id, among other things */
4692 * cgroup_load_subsys: load and register a modular subsystem at runtime
4693 * @ss: the subsystem to load
4695 * This function should be called in a modular subsystem's initcall. If the
4696 * subsystem is built as a module, it will be assigned a new subsys_id and set
4697 * up for use. If the subsystem is built-in anyway, work is delegated to the
4698 * simpler cgroup_init_subsys.
4700 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4702 struct cgroup_subsys_state *css;
4704 struct hlist_node *tmp;
4705 struct css_set *cset;
4708 /* check name and function validity */
4709 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4710 ss->css_alloc == NULL || ss->css_free == NULL)
4714 * we don't support callbacks in modular subsystems. this check is
4715 * before the ss->module check for consistency; a subsystem that could
4716 * be a module should still have no callbacks even if the user isn't
4717 * compiling it as one.
4719 if (ss->fork || ss->exit)
4723 * an optionally modular subsystem is built-in: we want to do nothing,
4724 * since cgroup_init_subsys will have already taken care of it.
4726 if (ss->module == NULL) {
4727 /* a sanity check */
4728 BUG_ON(cgroup_subsys[ss->subsys_id] != ss);
4732 /* init base cftset */
4733 cgroup_init_cftsets(ss);
4735 mutex_lock(&cgroup_mutex);
4736 cgroup_subsys[ss->subsys_id] = ss;
4739 * no ss->css_alloc seems to need anything important in the ss
4740 * struct, so this can happen first (i.e. before the dummy root
4743 css = ss->css_alloc(cgroup_dummy_top);
4745 /* failure case - need to deassign the cgroup_subsys[] slot. */
4746 cgroup_subsys[ss->subsys_id] = NULL;
4747 mutex_unlock(&cgroup_mutex);
4748 return PTR_ERR(css);
4751 list_add(&ss->sibling, &cgroup_dummy_root.subsys_list);
4752 ss->root = &cgroup_dummy_root;
4754 /* our new subsystem will be attached to the dummy hierarchy. */
4755 init_cgroup_css(css, ss, cgroup_dummy_top);
4756 /* init_idr must be after init_cgroup_css because it sets css->id. */
4758 ret = cgroup_init_idr(ss, css);
4764 * Now we need to entangle the css into the existing css_sets. unlike
4765 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4766 * will need a new pointer to it; done by iterating the css_set_table.
4767 * furthermore, modifying the existing css_sets will corrupt the hash
4768 * table state, so each changed css_set will need its hash recomputed.
4769 * this is all done under the css_set_lock.
4771 write_lock(&css_set_lock);
4772 hash_for_each_safe(css_set_table, i, tmp, cset, hlist) {
4773 /* skip entries that we already rehashed */
4774 if (cset->subsys[ss->subsys_id])
4776 /* remove existing entry */
4777 hash_del(&cset->hlist);
4779 cset->subsys[ss->subsys_id] = css;
4780 /* recompute hash and restore entry */
4781 key = css_set_hash(cset->subsys);
4782 hash_add(css_set_table, &cset->hlist, key);
4784 write_unlock(&css_set_lock);
4786 ret = online_css(ss, cgroup_dummy_top);
4791 mutex_unlock(&cgroup_mutex);
4795 mutex_unlock(&cgroup_mutex);
4796 /* @ss can't be mounted here as try_module_get() would fail */
4797 cgroup_unload_subsys(ss);
4800 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4803 * cgroup_unload_subsys: unload a modular subsystem
4804 * @ss: the subsystem to unload
4806 * This function should be called in a modular subsystem's exitcall. When this
4807 * function is invoked, the refcount on the subsystem's module will be 0, so
4808 * the subsystem will not be attached to any hierarchy.
4810 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4812 struct cgrp_cset_link *link;
4814 BUG_ON(ss->module == NULL);
4817 * we shouldn't be called if the subsystem is in use, and the use of
4818 * try_module_get in parse_cgroupfs_options should ensure that it
4819 * doesn't start being used while we're killing it off.
4821 BUG_ON(ss->root != &cgroup_dummy_root);
4823 mutex_lock(&cgroup_mutex);
4825 offline_css(ss, cgroup_dummy_top);
4828 idr_destroy(&ss->idr);
4830 /* deassign the subsys_id */
4831 cgroup_subsys[ss->subsys_id] = NULL;
4833 /* remove subsystem from the dummy root's list of subsystems */
4834 list_del_init(&ss->sibling);
4837 * disentangle the css from all css_sets attached to the dummy
4838 * top. as in loading, we need to pay our respects to the hashtable
4841 write_lock(&css_set_lock);
4842 list_for_each_entry(link, &cgroup_dummy_top->cset_links, cset_link) {
4843 struct css_set *cset = link->cset;
4846 hash_del(&cset->hlist);
4847 cset->subsys[ss->subsys_id] = NULL;
4848 key = css_set_hash(cset->subsys);
4849 hash_add(css_set_table, &cset->hlist, key);
4851 write_unlock(&css_set_lock);
4854 * remove subsystem's css from the cgroup_dummy_top and free it -
4855 * need to free before marking as null because ss->css_free needs
4856 * the cgrp->subsys pointer to find their state. note that this
4857 * also takes care of freeing the css_id.
4859 ss->css_free(cgroup_dummy_top);
4860 cgroup_dummy_top->subsys[ss->subsys_id] = NULL;
4862 mutex_unlock(&cgroup_mutex);
4864 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4867 * cgroup_init_early - cgroup initialization at system boot
4869 * Initialize cgroups at system boot, and initialize any
4870 * subsystems that request early init.
4872 int __init cgroup_init_early(void)
4874 struct cgroup_subsys *ss;
4877 atomic_set(&init_css_set.refcount, 1);
4878 INIT_LIST_HEAD(&init_css_set.cgrp_links);
4879 INIT_LIST_HEAD(&init_css_set.tasks);
4880 INIT_HLIST_NODE(&init_css_set.hlist);
4882 init_cgroup_root(&cgroup_dummy_root);
4883 cgroup_root_count = 1;
4884 RCU_INIT_POINTER(init_task.cgroups, &init_css_set);
4886 init_cgrp_cset_link.cset = &init_css_set;
4887 init_cgrp_cset_link.cgrp = cgroup_dummy_top;
4888 list_add(&init_cgrp_cset_link.cset_link, &cgroup_dummy_top->cset_links);
4889 list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links);
4891 /* at bootup time, we don't worry about modular subsystems */
4892 for_each_builtin_subsys(ss, i) {
4894 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4895 BUG_ON(!ss->css_alloc);
4896 BUG_ON(!ss->css_free);
4897 if (ss->subsys_id != i) {
4898 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4899 ss->name, ss->subsys_id);
4904 cgroup_init_subsys(ss);
4910 * cgroup_init - cgroup initialization
4912 * Register cgroup filesystem and /proc file, and initialize
4913 * any subsystems that didn't request early init.
4915 int __init cgroup_init(void)
4917 struct cgroup_subsys *ss;
4921 err = bdi_init(&cgroup_backing_dev_info);
4925 for_each_builtin_subsys(ss, i) {
4926 if (!ss->early_init)
4927 cgroup_init_subsys(ss);
4929 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4932 /* allocate id for the dummy hierarchy */
4933 mutex_lock(&cgroup_mutex);
4934 mutex_lock(&cgroup_root_mutex);
4936 /* Add init_css_set to the hash table */
4937 key = css_set_hash(init_css_set.subsys);
4938 hash_add(css_set_table, &init_css_set.hlist, key);
4940 BUG_ON(cgroup_init_root_id(&cgroup_dummy_root, 0, 1));
4942 mutex_unlock(&cgroup_root_mutex);
4943 mutex_unlock(&cgroup_mutex);
4945 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4951 err = register_filesystem(&cgroup_fs_type);
4953 kobject_put(cgroup_kobj);
4957 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4961 bdi_destroy(&cgroup_backing_dev_info);
4967 * proc_cgroup_show()
4968 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4969 * - Used for /proc/<pid>/cgroup.
4970 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4971 * doesn't really matter if tsk->cgroup changes after we read it,
4972 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4973 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4974 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4975 * cgroup to top_cgroup.
4978 /* TODO: Use a proper seq_file iterator */
4979 int proc_cgroup_show(struct seq_file *m, void *v)
4982 struct task_struct *tsk;
4985 struct cgroupfs_root *root;
4988 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4994 tsk = get_pid_task(pid, PIDTYPE_PID);
5000 mutex_lock(&cgroup_mutex);
5002 for_each_active_root(root) {
5003 struct cgroup_subsys *ss;
5004 struct cgroup *cgrp;
5007 seq_printf(m, "%d:", root->hierarchy_id);
5008 for_each_root_subsys(root, ss)
5009 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
5010 if (strlen(root->name))
5011 seq_printf(m, "%sname=%s", count ? "," : "",
5014 cgrp = task_cgroup_from_root(tsk, root);
5015 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
5023 mutex_unlock(&cgroup_mutex);
5024 put_task_struct(tsk);
5031 /* Display information about each subsystem and each hierarchy */
5032 static int proc_cgroupstats_show(struct seq_file *m, void *v)
5034 struct cgroup_subsys *ss;
5037 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5039 * ideally we don't want subsystems moving around while we do this.
5040 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5041 * subsys/hierarchy state.
5043 mutex_lock(&cgroup_mutex);
5045 for_each_subsys(ss, i)
5046 seq_printf(m, "%s\t%d\t%d\t%d\n",
5047 ss->name, ss->root->hierarchy_id,
5048 ss->root->number_of_cgroups, !ss->disabled);
5050 mutex_unlock(&cgroup_mutex);
5054 static int cgroupstats_open(struct inode *inode, struct file *file)
5056 return single_open(file, proc_cgroupstats_show, NULL);
5059 static const struct file_operations proc_cgroupstats_operations = {
5060 .open = cgroupstats_open,
5062 .llseek = seq_lseek,
5063 .release = single_release,
5067 * cgroup_fork - attach newly forked task to its parents cgroup.
5068 * @child: pointer to task_struct of forking parent process.
5070 * Description: A task inherits its parent's cgroup at fork().
5072 * A pointer to the shared css_set was automatically copied in
5073 * fork.c by dup_task_struct(). However, we ignore that copy, since
5074 * it was not made under the protection of RCU or cgroup_mutex, so
5075 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
5076 * have already changed current->cgroups, allowing the previously
5077 * referenced cgroup group to be removed and freed.
5079 * At the point that cgroup_fork() is called, 'current' is the parent
5080 * task, and the passed argument 'child' points to the child task.
5082 void cgroup_fork(struct task_struct *child)
5085 get_css_set(task_css_set(current));
5086 child->cgroups = current->cgroups;
5087 task_unlock(current);
5088 INIT_LIST_HEAD(&child->cg_list);
5092 * cgroup_post_fork - called on a new task after adding it to the task list
5093 * @child: the task in question
5095 * Adds the task to the list running through its css_set if necessary and
5096 * call the subsystem fork() callbacks. Has to be after the task is
5097 * visible on the task list in case we race with the first call to
5098 * cgroup_iter_start() - to guarantee that the new task ends up on its
5101 void cgroup_post_fork(struct task_struct *child)
5103 struct cgroup_subsys *ss;
5107 * use_task_css_set_links is set to 1 before we walk the tasklist
5108 * under the tasklist_lock and we read it here after we added the child
5109 * to the tasklist under the tasklist_lock as well. If the child wasn't
5110 * yet in the tasklist when we walked through it from
5111 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
5112 * should be visible now due to the paired locking and barriers implied
5113 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
5114 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
5117 if (use_task_css_set_links) {
5118 write_lock(&css_set_lock);
5120 if (list_empty(&child->cg_list))
5121 list_add(&child->cg_list, &task_css_set(child)->tasks);
5123 write_unlock(&css_set_lock);
5127 * Call ss->fork(). This must happen after @child is linked on
5128 * css_set; otherwise, @child might change state between ->fork()
5129 * and addition to css_set.
5131 if (need_forkexit_callback) {
5133 * fork/exit callbacks are supported only for builtin
5134 * subsystems, and the builtin section of the subsys
5135 * array is immutable, so we don't need to lock the
5136 * subsys array here. On the other hand, modular section
5137 * of the array can be freed at module unload, so we
5140 for_each_builtin_subsys(ss, i)
5147 * cgroup_exit - detach cgroup from exiting task
5148 * @tsk: pointer to task_struct of exiting process
5149 * @run_callback: run exit callbacks?
5151 * Description: Detach cgroup from @tsk and release it.
5153 * Note that cgroups marked notify_on_release force every task in
5154 * them to take the global cgroup_mutex mutex when exiting.
5155 * This could impact scaling on very large systems. Be reluctant to
5156 * use notify_on_release cgroups where very high task exit scaling
5157 * is required on large systems.
5159 * the_top_cgroup_hack:
5161 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
5163 * We call cgroup_exit() while the task is still competent to
5164 * handle notify_on_release(), then leave the task attached to the
5165 * root cgroup in each hierarchy for the remainder of its exit.
5167 * To do this properly, we would increment the reference count on
5168 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
5169 * code we would add a second cgroup function call, to drop that
5170 * reference. This would just create an unnecessary hot spot on
5171 * the top_cgroup reference count, to no avail.
5173 * Normally, holding a reference to a cgroup without bumping its
5174 * count is unsafe. The cgroup could go away, or someone could
5175 * attach us to a different cgroup, decrementing the count on
5176 * the first cgroup that we never incremented. But in this case,
5177 * top_cgroup isn't going away, and either task has PF_EXITING set,
5178 * which wards off any cgroup_attach_task() attempts, or task is a failed
5179 * fork, never visible to cgroup_attach_task.
5181 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
5183 struct cgroup_subsys *ss;
5184 struct css_set *cset;
5188 * Unlink from the css_set task list if necessary.
5189 * Optimistically check cg_list before taking
5192 if (!list_empty(&tsk->cg_list)) {
5193 write_lock(&css_set_lock);
5194 if (!list_empty(&tsk->cg_list))
5195 list_del_init(&tsk->cg_list);
5196 write_unlock(&css_set_lock);
5199 /* Reassign the task to the init_css_set. */
5201 cset = task_css_set(tsk);
5202 RCU_INIT_POINTER(tsk->cgroups, &init_css_set);
5204 if (run_callbacks && need_forkexit_callback) {
5206 * fork/exit callbacks are supported only for builtin
5207 * subsystems, see cgroup_post_fork() for details.
5209 for_each_builtin_subsys(ss, i) {
5211 struct cgroup *old_cgrp = cset->subsys[i]->cgroup;
5212 struct cgroup *cgrp = task_cgroup(tsk, i);
5214 ss->exit(cgrp, old_cgrp, tsk);
5220 put_css_set_taskexit(cset);
5223 static void check_for_release(struct cgroup *cgrp)
5225 if (cgroup_is_releasable(cgrp) &&
5226 list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) {
5228 * Control Group is currently removeable. If it's not
5229 * already queued for a userspace notification, queue
5232 int need_schedule_work = 0;
5234 raw_spin_lock(&release_list_lock);
5235 if (!cgroup_is_dead(cgrp) &&
5236 list_empty(&cgrp->release_list)) {
5237 list_add(&cgrp->release_list, &release_list);
5238 need_schedule_work = 1;
5240 raw_spin_unlock(&release_list_lock);
5241 if (need_schedule_work)
5242 schedule_work(&release_agent_work);
5247 * Notify userspace when a cgroup is released, by running the
5248 * configured release agent with the name of the cgroup (path
5249 * relative to the root of cgroup file system) as the argument.
5251 * Most likely, this user command will try to rmdir this cgroup.
5253 * This races with the possibility that some other task will be
5254 * attached to this cgroup before it is removed, or that some other
5255 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5256 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5257 * unused, and this cgroup will be reprieved from its death sentence,
5258 * to continue to serve a useful existence. Next time it's released,
5259 * we will get notified again, if it still has 'notify_on_release' set.
5261 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5262 * means only wait until the task is successfully execve()'d. The
5263 * separate release agent task is forked by call_usermodehelper(),
5264 * then control in this thread returns here, without waiting for the
5265 * release agent task. We don't bother to wait because the caller of
5266 * this routine has no use for the exit status of the release agent
5267 * task, so no sense holding our caller up for that.
5269 static void cgroup_release_agent(struct work_struct *work)
5271 BUG_ON(work != &release_agent_work);
5272 mutex_lock(&cgroup_mutex);
5273 raw_spin_lock(&release_list_lock);
5274 while (!list_empty(&release_list)) {
5275 char *argv[3], *envp[3];
5277 char *pathbuf = NULL, *agentbuf = NULL;
5278 struct cgroup *cgrp = list_entry(release_list.next,
5281 list_del_init(&cgrp->release_list);
5282 raw_spin_unlock(&release_list_lock);
5283 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5286 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5288 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5293 argv[i++] = agentbuf;
5294 argv[i++] = pathbuf;
5298 /* minimal command environment */
5299 envp[i++] = "HOME=/";
5300 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5303 /* Drop the lock while we invoke the usermode helper,
5304 * since the exec could involve hitting disk and hence
5305 * be a slow process */
5306 mutex_unlock(&cgroup_mutex);
5307 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5308 mutex_lock(&cgroup_mutex);
5312 raw_spin_lock(&release_list_lock);
5314 raw_spin_unlock(&release_list_lock);
5315 mutex_unlock(&cgroup_mutex);
5318 static int __init cgroup_disable(char *str)
5320 struct cgroup_subsys *ss;
5324 while ((token = strsep(&str, ",")) != NULL) {
5329 * cgroup_disable, being at boot time, can't know about
5330 * module subsystems, so we don't worry about them.
5332 for_each_builtin_subsys(ss, i) {
5333 if (!strcmp(token, ss->name)) {
5335 printk(KERN_INFO "Disabling %s control group"
5336 " subsystem\n", ss->name);
5343 __setup("cgroup_disable=", cgroup_disable);
5346 * Functons for CSS ID.
5349 /* to get ID other than 0, this should be called when !cgroup_is_dead() */
5350 unsigned short css_id(struct cgroup_subsys_state *css)
5352 struct css_id *cssid;
5355 * This css_id() can return correct value when somone has refcnt
5356 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5357 * it's unchanged until freed.
5359 cssid = rcu_dereference_raw(css->id);
5365 EXPORT_SYMBOL_GPL(css_id);
5368 * css_is_ancestor - test "root" css is an ancestor of "child"
5369 * @child: the css to be tested.
5370 * @root: the css supporsed to be an ancestor of the child.
5372 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5373 * this function reads css->id, the caller must hold rcu_read_lock().
5374 * But, considering usual usage, the csses should be valid objects after test.
5375 * Assuming that the caller will do some action to the child if this returns
5376 * returns true, the caller must take "child";s reference count.
5377 * If "child" is valid object and this returns true, "root" is valid, too.
5380 bool css_is_ancestor(struct cgroup_subsys_state *child,
5381 const struct cgroup_subsys_state *root)
5383 struct css_id *child_id;
5384 struct css_id *root_id;
5386 child_id = rcu_dereference(child->id);
5389 root_id = rcu_dereference(root->id);
5392 if (child_id->depth < root_id->depth)
5394 if (child_id->stack[root_id->depth] != root_id->id)
5399 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5401 struct css_id *id = rcu_dereference_protected(css->id, true);
5403 /* When this is called before css_id initialization, id can be NULL */
5407 BUG_ON(!ss->use_id);
5409 rcu_assign_pointer(id->css, NULL);
5410 rcu_assign_pointer(css->id, NULL);
5411 spin_lock(&ss->id_lock);
5412 idr_remove(&ss->idr, id->id);
5413 spin_unlock(&ss->id_lock);
5414 kfree_rcu(id, rcu_head);
5416 EXPORT_SYMBOL_GPL(free_css_id);
5419 * This is called by init or create(). Then, calls to this function are
5420 * always serialized (By cgroup_mutex() at create()).
5423 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5425 struct css_id *newid;
5428 BUG_ON(!ss->use_id);
5430 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5431 newid = kzalloc(size, GFP_KERNEL);
5433 return ERR_PTR(-ENOMEM);
5435 idr_preload(GFP_KERNEL);
5436 spin_lock(&ss->id_lock);
5437 /* Don't use 0. allocates an ID of 1-65535 */
5438 ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5439 spin_unlock(&ss->id_lock);
5442 /* Returns error when there are no free spaces for new ID.*/
5447 newid->depth = depth;
5451 return ERR_PTR(ret);
5455 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5456 struct cgroup_subsys_state *rootcss)
5458 struct css_id *newid;
5460 spin_lock_init(&ss->id_lock);
5463 newid = get_new_cssid(ss, 0);
5465 return PTR_ERR(newid);
5467 newid->stack[0] = newid->id;
5468 RCU_INIT_POINTER(newid->css, rootcss);
5469 RCU_INIT_POINTER(rootcss->id, newid);
5473 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5474 struct cgroup *child)
5476 int subsys_id, i, depth = 0;
5477 struct cgroup_subsys_state *parent_css, *child_css;
5478 struct css_id *child_id, *parent_id;
5480 subsys_id = ss->subsys_id;
5481 parent_css = parent->subsys[subsys_id];
5482 child_css = child->subsys[subsys_id];
5483 parent_id = rcu_dereference_protected(parent_css->id, true);
5484 depth = parent_id->depth + 1;
5486 child_id = get_new_cssid(ss, depth);
5487 if (IS_ERR(child_id))
5488 return PTR_ERR(child_id);
5490 for (i = 0; i < depth; i++)
5491 child_id->stack[i] = parent_id->stack[i];
5492 child_id->stack[depth] = child_id->id;
5494 * child_id->css pointer will be set after this cgroup is available
5495 * see cgroup_populate_dir()
5497 rcu_assign_pointer(child_css->id, child_id);
5503 * css_lookup - lookup css by id
5504 * @ss: cgroup subsys to be looked into.
5507 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5508 * NULL if not. Should be called under rcu_read_lock()
5510 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5512 struct css_id *cssid = NULL;
5514 BUG_ON(!ss->use_id);
5515 cssid = idr_find(&ss->idr, id);
5517 if (unlikely(!cssid))
5520 return rcu_dereference(cssid->css);
5522 EXPORT_SYMBOL_GPL(css_lookup);
5525 * get corresponding css from file open on cgroupfs directory
5527 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5529 struct cgroup *cgrp;
5530 struct inode *inode;
5531 struct cgroup_subsys_state *css;
5533 inode = file_inode(f);
5534 /* check in cgroup filesystem dir */
5535 if (inode->i_op != &cgroup_dir_inode_operations)
5536 return ERR_PTR(-EBADF);
5538 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5539 return ERR_PTR(-EINVAL);
5542 cgrp = __d_cgrp(f->f_dentry);
5543 css = cgrp->subsys[id];
5544 return css ? css : ERR_PTR(-ENOENT);
5547 #ifdef CONFIG_CGROUP_DEBUG
5548 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cgrp)
5550 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5553 return ERR_PTR(-ENOMEM);
5558 static void debug_css_free(struct cgroup *cgrp)
5560 kfree(cgrp->subsys[debug_subsys_id]);
5563 static u64 debug_taskcount_read(struct cgroup *cgrp, struct cftype *cft)
5565 return cgroup_task_count(cgrp);
5568 static u64 current_css_set_read(struct cgroup *cgrp, struct cftype *cft)
5570 return (u64)(unsigned long)current->cgroups;
5573 static u64 current_css_set_refcount_read(struct cgroup *cgrp,
5579 count = atomic_read(&task_css_set(current)->refcount);
5584 static int current_css_set_cg_links_read(struct cgroup *cgrp,
5586 struct seq_file *seq)
5588 struct cgrp_cset_link *link;
5589 struct css_set *cset;
5591 read_lock(&css_set_lock);
5593 cset = rcu_dereference(current->cgroups);
5594 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
5595 struct cgroup *c = link->cgrp;
5599 name = c->dentry->d_name.name;
5602 seq_printf(seq, "Root %d group %s\n",
5603 c->root->hierarchy_id, name);
5606 read_unlock(&css_set_lock);
5610 #define MAX_TASKS_SHOWN_PER_CSS 25
5611 static int cgroup_css_links_read(struct cgroup *cgrp,
5613 struct seq_file *seq)
5615 struct cgrp_cset_link *link;
5617 read_lock(&css_set_lock);
5618 list_for_each_entry(link, &cgrp->cset_links, cset_link) {
5619 struct css_set *cset = link->cset;
5620 struct task_struct *task;
5622 seq_printf(seq, "css_set %p\n", cset);
5623 list_for_each_entry(task, &cset->tasks, cg_list) {
5624 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5625 seq_puts(seq, " ...\n");
5628 seq_printf(seq, " task %d\n",
5629 task_pid_vnr(task));
5633 read_unlock(&css_set_lock);
5637 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5639 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5642 static struct cftype debug_files[] = {
5644 .name = "taskcount",
5645 .read_u64 = debug_taskcount_read,
5649 .name = "current_css_set",
5650 .read_u64 = current_css_set_read,
5654 .name = "current_css_set_refcount",
5655 .read_u64 = current_css_set_refcount_read,
5659 .name = "current_css_set_cg_links",
5660 .read_seq_string = current_css_set_cg_links_read,
5664 .name = "cgroup_css_links",
5665 .read_seq_string = cgroup_css_links_read,
5669 .name = "releasable",
5670 .read_u64 = releasable_read,
5676 struct cgroup_subsys debug_subsys = {
5678 .css_alloc = debug_css_alloc,
5679 .css_free = debug_css_free,
5680 .subsys_id = debug_subsys_id,
5681 .base_cftypes = debug_files,
5683 #endif /* CONFIG_CGROUP_DEBUG */