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>
66 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
67 #define CSS_DEACT_BIAS INT_MIN
70 * cgroup_mutex is the master lock. Any modification to cgroup or its
71 * hierarchy must be performed while holding it.
73 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
74 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
75 * release_agent_path and so on. Modifying requires both cgroup_mutex and
76 * cgroup_root_mutex. Readers can acquire either of the two. This is to
77 * break the following locking order cycle.
79 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
80 * B. namespace_sem -> cgroup_mutex
82 * B happens only through cgroup_show_options() and using cgroup_root_mutex
85 #ifdef CONFIG_PROVE_RCU
86 DEFINE_MUTEX(cgroup_mutex);
87 EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for task_subsys_state_check() */
89 static DEFINE_MUTEX(cgroup_mutex);
92 static DEFINE_MUTEX(cgroup_root_mutex);
95 * Generate an array of cgroup subsystem pointers. At boot time, this is
96 * populated with the built in subsystems, and modular subsystems are
97 * registered after that. The mutable section of this array is protected by
100 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
101 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
102 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
103 #include <linux/cgroup_subsys.h>
107 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
108 * subsystems that are otherwise unattached - it never has more than a
109 * single cgroup, and all tasks are part of that cgroup.
111 static struct cgroupfs_root rootnode;
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(roots);
190 static int 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 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
200 #define dummytop (&rootnode.top_cgroup)
202 static struct cgroup_name root_cgroup_name = { .name = "/" };
204 /* This flag indicates whether tasks in the fork and exit paths should
205 * check for fork/exit handlers to call. This avoids us having to do
206 * extra work in the fork/exit path if none of the subsystems need to
209 static int need_forkexit_callback __read_mostly;
211 static int cgroup_destroy_locked(struct cgroup *cgrp);
212 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
213 struct cftype cfts[], bool is_add);
215 static int css_unbias_refcnt(int refcnt)
217 return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
220 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
221 static int css_refcnt(struct cgroup_subsys_state *css)
223 int v = atomic_read(&css->refcnt);
225 return css_unbias_refcnt(v);
228 /* convenient tests for these bits */
229 static inline bool cgroup_is_removed(const struct cgroup *cgrp)
231 return test_bit(CGRP_REMOVED, &cgrp->flags);
235 * cgroup_is_descendant - test ancestry
236 * @cgrp: the cgroup to be tested
237 * @ancestor: possible ancestor of @cgrp
239 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
240 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
241 * and @ancestor are accessible.
243 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
246 if (cgrp == ancestor)
252 EXPORT_SYMBOL_GPL(cgroup_is_descendant);
254 static int cgroup_is_releasable(const struct cgroup *cgrp)
257 (1 << CGRP_RELEASABLE) |
258 (1 << CGRP_NOTIFY_ON_RELEASE);
259 return (cgrp->flags & bits) == bits;
262 static int notify_on_release(const struct cgroup *cgrp)
264 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
268 * for_each_subsys() allows you to iterate on each subsystem attached to
269 * an active hierarchy
271 #define for_each_subsys(_root, _ss) \
272 list_for_each_entry(_ss, &_root->subsys_list, sibling)
274 /* for_each_active_root() allows you to iterate across the active hierarchies */
275 #define for_each_active_root(_root) \
276 list_for_each_entry(_root, &roots, root_list)
278 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
280 return dentry->d_fsdata;
283 static inline struct cfent *__d_cfe(struct dentry *dentry)
285 return dentry->d_fsdata;
288 static inline struct cftype *__d_cft(struct dentry *dentry)
290 return __d_cfe(dentry)->type;
294 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
295 * @cgrp: the cgroup to be checked for liveness
297 * On success, returns true; the mutex should be later unlocked. On
298 * failure returns false with no lock held.
300 static bool cgroup_lock_live_group(struct cgroup *cgrp)
302 mutex_lock(&cgroup_mutex);
303 if (cgroup_is_removed(cgrp)) {
304 mutex_unlock(&cgroup_mutex);
310 /* the list of cgroups eligible for automatic release. Protected by
311 * release_list_lock */
312 static LIST_HEAD(release_list);
313 static DEFINE_RAW_SPINLOCK(release_list_lock);
314 static void cgroup_release_agent(struct work_struct *work);
315 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
316 static void check_for_release(struct cgroup *cgrp);
319 * A cgroup can be associated with multiple css_sets as different tasks may
320 * belong to different cgroups on different hierarchies. In the other
321 * direction, a css_set is naturally associated with multiple cgroups.
322 * This M:N relationship is represented by the following link structure
323 * which exists for each association and allows traversing the associations
326 struct cgrp_cset_link {
327 /* the cgroup and css_set this link associates */
329 struct css_set *cset;
331 /* list of cgrp_cset_links anchored at cgrp->cset_links */
332 struct list_head cset_link;
334 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
335 struct list_head cgrp_link;
338 /* The default css_set - used by init and its children prior to any
339 * hierarchies being mounted. It contains a pointer to the root state
340 * for each subsystem. Also used to anchor the list of css_sets. Not
341 * reference-counted, to improve performance when child cgroups
342 * haven't been created.
345 static struct css_set init_css_set;
346 static struct cgrp_cset_link init_cgrp_cset_link;
348 static int cgroup_init_idr(struct cgroup_subsys *ss,
349 struct cgroup_subsys_state *css);
351 /* css_set_lock protects the list of css_set objects, and the
352 * chain of tasks off each css_set. Nests outside task->alloc_lock
353 * due to cgroup_iter_start() */
354 static DEFINE_RWLOCK(css_set_lock);
355 static int css_set_count;
358 * hash table for cgroup groups. This improves the performance to find
359 * an existing css_set. This hash doesn't (currently) take into
360 * account cgroups in empty hierarchies.
362 #define CSS_SET_HASH_BITS 7
363 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
365 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
368 unsigned long key = 0UL;
370 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
371 key += (unsigned long)css[i];
372 key = (key >> 16) ^ key;
377 /* We don't maintain the lists running through each css_set to its
378 * task until after the first call to cgroup_iter_start(). This
379 * reduces the fork()/exit() overhead for people who have cgroups
380 * compiled into their kernel but not actually in use */
381 static int use_task_css_set_links __read_mostly;
383 static void __put_css_set(struct css_set *cset, int taskexit)
385 struct cgrp_cset_link *link, *tmp_link;
388 * Ensure that the refcount doesn't hit zero while any readers
389 * can see it. Similar to atomic_dec_and_lock(), but for an
392 if (atomic_add_unless(&cset->refcount, -1, 1))
394 write_lock(&css_set_lock);
395 if (!atomic_dec_and_test(&cset->refcount)) {
396 write_unlock(&css_set_lock);
400 /* This css_set is dead. unlink it and release cgroup refcounts */
401 hash_del(&cset->hlist);
404 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
405 struct cgroup *cgrp = link->cgrp;
407 list_del(&link->cset_link);
408 list_del(&link->cgrp_link);
411 * We may not be holding cgroup_mutex, and if cgrp->count is
412 * dropped to 0 the cgroup can be destroyed at any time, hence
413 * rcu_read_lock is used to keep it alive.
416 if (atomic_dec_and_test(&cgrp->count) &&
417 notify_on_release(cgrp)) {
419 set_bit(CGRP_RELEASABLE, &cgrp->flags);
420 check_for_release(cgrp);
427 write_unlock(&css_set_lock);
428 kfree_rcu(cset, rcu_head);
432 * refcounted get/put for css_set objects
434 static inline void get_css_set(struct css_set *cset)
436 atomic_inc(&cset->refcount);
439 static inline void put_css_set(struct css_set *cset)
441 __put_css_set(cset, 0);
444 static inline void put_css_set_taskexit(struct css_set *cset)
446 __put_css_set(cset, 1);
450 * compare_css_sets - helper function for find_existing_css_set().
451 * @cset: candidate css_set being tested
452 * @old_cset: existing css_set for a task
453 * @new_cgrp: cgroup that's being entered by the task
454 * @template: desired set of css pointers in css_set (pre-calculated)
456 * Returns true if "cg" matches "old_cg" except for the hierarchy
457 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
459 static bool compare_css_sets(struct css_set *cset,
460 struct css_set *old_cset,
461 struct cgroup *new_cgrp,
462 struct cgroup_subsys_state *template[])
464 struct list_head *l1, *l2;
466 if (memcmp(template, cset->subsys, sizeof(cset->subsys))) {
467 /* Not all subsystems matched */
472 * Compare cgroup pointers in order to distinguish between
473 * different cgroups in heirarchies with no subsystems. We
474 * could get by with just this check alone (and skip the
475 * memcmp above) but on most setups the memcmp check will
476 * avoid the need for this more expensive check on almost all
480 l1 = &cset->cgrp_links;
481 l2 = &old_cset->cgrp_links;
483 struct cgrp_cset_link *link1, *link2;
484 struct cgroup *cgrp1, *cgrp2;
488 /* See if we reached the end - both lists are equal length. */
489 if (l1 == &cset->cgrp_links) {
490 BUG_ON(l2 != &old_cset->cgrp_links);
493 BUG_ON(l2 == &old_cset->cgrp_links);
495 /* Locate the cgroups associated with these links. */
496 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
497 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
500 /* Hierarchies should be linked in the same order. */
501 BUG_ON(cgrp1->root != cgrp2->root);
504 * If this hierarchy is the hierarchy of the cgroup
505 * that's changing, then we need to check that this
506 * css_set points to the new cgroup; if it's any other
507 * hierarchy, then this css_set should point to the
508 * same cgroup as the old css_set.
510 if (cgrp1->root == new_cgrp->root) {
511 if (cgrp1 != new_cgrp)
522 * find_existing_css_set() is a helper for
523 * find_css_set(), and checks to see whether an existing
524 * css_set is suitable.
526 * oldcg: the cgroup group that we're using before the cgroup
529 * cgrp: the cgroup that we're moving into
531 * template: location in which to build the desired set of subsystem
532 * state objects for the new cgroup group
534 static struct css_set *find_existing_css_set(struct css_set *old_cset,
536 struct cgroup_subsys_state *template[])
539 struct cgroupfs_root *root = cgrp->root;
540 struct css_set *cset;
544 * Build the set of subsystem state objects that we want to see in the
545 * new css_set. while subsystems can change globally, the entries here
546 * won't change, so no need for locking.
548 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
549 if (root->subsys_mask & (1UL << i)) {
550 /* Subsystem is in this hierarchy. So we want
551 * the subsystem state from the new
553 template[i] = cgrp->subsys[i];
555 /* Subsystem is not in this hierarchy, so we
556 * don't want to change the subsystem state */
557 template[i] = old_cset->subsys[i];
561 key = css_set_hash(template);
562 hash_for_each_possible(css_set_table, cset, hlist, key) {
563 if (!compare_css_sets(cset, old_cset, cgrp, template))
566 /* This css_set matches what we need */
570 /* No existing cgroup group matched */
574 static void free_cgrp_cset_links(struct list_head *links_to_free)
576 struct cgrp_cset_link *link, *tmp_link;
578 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
579 list_del(&link->cset_link);
585 * allocate_cgrp_cset_links - allocate cgrp_cset_links
586 * @count: the number of links to allocate
587 * @tmp_links: list_head the allocated links are put on
589 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
590 * through ->cset_link. Returns 0 on success or -errno.
592 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
594 struct cgrp_cset_link *link;
597 INIT_LIST_HEAD(tmp_links);
599 for (i = 0; i < count; i++) {
600 link = kzalloc(sizeof(*link), GFP_KERNEL);
602 free_cgrp_cset_links(tmp_links);
605 list_add(&link->cset_link, tmp_links);
611 * link_css_set - a helper function to link a css_set to a cgroup
612 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
613 * @cset: the css_set to be linked
614 * @cgrp: the destination cgroup
616 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
619 struct cgrp_cset_link *link;
621 BUG_ON(list_empty(tmp_links));
622 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
625 atomic_inc(&cgrp->count);
626 list_move(&link->cset_link, &cgrp->cset_links);
628 * Always add links to the tail of the list so that the list
629 * is sorted by order of hierarchy creation
631 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
635 * find_css_set() takes an existing cgroup group and a
636 * cgroup object, and returns a css_set object that's
637 * equivalent to the old group, but with the given cgroup
638 * substituted into the appropriate hierarchy. Must be called with
641 static struct css_set *find_css_set(struct css_set *old_cset,
644 struct css_set *cset;
645 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
646 struct list_head tmp_links;
647 struct cgrp_cset_link *link;
650 /* First see if we already have a cgroup group that matches
652 read_lock(&css_set_lock);
653 cset = find_existing_css_set(old_cset, cgrp, template);
656 read_unlock(&css_set_lock);
661 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
665 /* Allocate all the cgrp_cset_link objects that we'll need */
666 if (allocate_cgrp_cset_links(root_count, &tmp_links) < 0) {
671 atomic_set(&cset->refcount, 1);
672 INIT_LIST_HEAD(&cset->cgrp_links);
673 INIT_LIST_HEAD(&cset->tasks);
674 INIT_HLIST_NODE(&cset->hlist);
676 /* Copy the set of subsystem state objects generated in
677 * find_existing_css_set() */
678 memcpy(cset->subsys, template, sizeof(cset->subsys));
680 write_lock(&css_set_lock);
681 /* Add reference counts and links from the new css_set. */
682 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
683 struct cgroup *c = link->cgrp;
685 if (c->root == cgrp->root)
687 link_css_set(&tmp_links, cset, c);
690 BUG_ON(!list_empty(&tmp_links));
694 /* Add this cgroup group to the hash table */
695 key = css_set_hash(cset->subsys);
696 hash_add(css_set_table, &cset->hlist, key);
698 write_unlock(&css_set_lock);
704 * Return the cgroup for "task" from the given hierarchy. Must be
705 * called with cgroup_mutex held.
707 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
708 struct cgroupfs_root *root)
710 struct css_set *cset;
711 struct cgroup *res = NULL;
713 BUG_ON(!mutex_is_locked(&cgroup_mutex));
714 read_lock(&css_set_lock);
716 * No need to lock the task - since we hold cgroup_mutex the
717 * task can't change groups, so the only thing that can happen
718 * is that it exits and its css is set back to init_css_set.
720 cset = task->cgroups;
721 if (cset == &init_css_set) {
722 res = &root->top_cgroup;
724 struct cgrp_cset_link *link;
726 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
727 struct cgroup *c = link->cgrp;
729 if (c->root == root) {
735 read_unlock(&css_set_lock);
741 * There is one global cgroup mutex. We also require taking
742 * task_lock() when dereferencing a task's cgroup subsys pointers.
743 * See "The task_lock() exception", at the end of this comment.
745 * A task must hold cgroup_mutex to modify cgroups.
747 * Any task can increment and decrement the count field without lock.
748 * So in general, code holding cgroup_mutex can't rely on the count
749 * field not changing. However, if the count goes to zero, then only
750 * cgroup_attach_task() can increment it again. Because a count of zero
751 * means that no tasks are currently attached, therefore there is no
752 * way a task attached to that cgroup can fork (the other way to
753 * increment the count). So code holding cgroup_mutex can safely
754 * assume that if the count is zero, it will stay zero. Similarly, if
755 * a task holds cgroup_mutex on a cgroup with zero count, it
756 * knows that the cgroup won't be removed, as cgroup_rmdir()
759 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
760 * (usually) take cgroup_mutex. These are the two most performance
761 * critical pieces of code here. The exception occurs on cgroup_exit(),
762 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
763 * is taken, and if the cgroup count is zero, a usermode call made
764 * to the release agent with the name of the cgroup (path relative to
765 * the root of cgroup file system) as the argument.
767 * A cgroup can only be deleted if both its 'count' of using tasks
768 * is zero, and its list of 'children' cgroups is empty. Since all
769 * tasks in the system use _some_ cgroup, and since there is always at
770 * least one task in the system (init, pid == 1), therefore, top_cgroup
771 * always has either children cgroups and/or using tasks. So we don't
772 * need a special hack to ensure that top_cgroup cannot be deleted.
774 * The task_lock() exception
776 * The need for this exception arises from the action of
777 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
778 * another. It does so using cgroup_mutex, however there are
779 * several performance critical places that need to reference
780 * task->cgroup without the expense of grabbing a system global
781 * mutex. Therefore except as noted below, when dereferencing or, as
782 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
783 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
784 * the task_struct routinely used for such matters.
786 * P.S. One more locking exception. RCU is used to guard the
787 * update of a tasks cgroup pointer by cgroup_attach_task()
791 * A couple of forward declarations required, due to cyclic reference loop:
792 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
793 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
797 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
798 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
799 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
800 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
801 unsigned long subsys_mask);
802 static const struct inode_operations cgroup_dir_inode_operations;
803 static const struct file_operations proc_cgroupstats_operations;
805 static struct backing_dev_info cgroup_backing_dev_info = {
807 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
810 static int alloc_css_id(struct cgroup_subsys *ss,
811 struct cgroup *parent, struct cgroup *child);
813 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
815 struct inode *inode = new_inode(sb);
818 inode->i_ino = get_next_ino();
819 inode->i_mode = mode;
820 inode->i_uid = current_fsuid();
821 inode->i_gid = current_fsgid();
822 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
823 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
828 static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
830 struct cgroup_name *name;
832 name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL);
835 strcpy(name->name, dentry->d_name.name);
839 static void cgroup_free_fn(struct work_struct *work)
841 struct cgroup *cgrp = container_of(work, struct cgroup, free_work);
842 struct cgroup_subsys *ss;
844 mutex_lock(&cgroup_mutex);
846 * Release the subsystem state objects.
848 for_each_subsys(cgrp->root, ss)
851 cgrp->root->number_of_cgroups--;
852 mutex_unlock(&cgroup_mutex);
855 * We get a ref to the parent's dentry, and put the ref when
856 * this cgroup is being freed, so it's guaranteed that the
857 * parent won't be destroyed before its children.
859 dput(cgrp->parent->dentry);
861 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
864 * Drop the active superblock reference that we took when we
865 * created the cgroup. This will free cgrp->root, if we are
866 * holding the last reference to @sb.
868 deactivate_super(cgrp->root->sb);
871 * if we're getting rid of the cgroup, refcount should ensure
872 * that there are no pidlists left.
874 BUG_ON(!list_empty(&cgrp->pidlists));
876 simple_xattrs_free(&cgrp->xattrs);
878 kfree(rcu_dereference_raw(cgrp->name));
882 static void cgroup_free_rcu(struct rcu_head *head)
884 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
886 schedule_work(&cgrp->free_work);
889 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
891 /* is dentry a directory ? if so, kfree() associated cgroup */
892 if (S_ISDIR(inode->i_mode)) {
893 struct cgroup *cgrp = dentry->d_fsdata;
895 BUG_ON(!(cgroup_is_removed(cgrp)));
896 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
898 struct cfent *cfe = __d_cfe(dentry);
899 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
901 WARN_ONCE(!list_empty(&cfe->node) &&
902 cgrp != &cgrp->root->top_cgroup,
903 "cfe still linked for %s\n", cfe->type->name);
904 simple_xattrs_free(&cfe->xattrs);
910 static int cgroup_delete(const struct dentry *d)
915 static void remove_dir(struct dentry *d)
917 struct dentry *parent = dget(d->d_parent);
920 simple_rmdir(parent->d_inode, d);
924 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
928 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
929 lockdep_assert_held(&cgroup_mutex);
932 * If we're doing cleanup due to failure of cgroup_create(),
933 * the corresponding @cfe may not exist.
935 list_for_each_entry(cfe, &cgrp->files, node) {
936 struct dentry *d = cfe->dentry;
938 if (cft && cfe->type != cft)
943 simple_unlink(cgrp->dentry->d_inode, d);
944 list_del_init(&cfe->node);
952 * cgroup_clear_directory - selective removal of base and subsystem files
953 * @dir: directory containing the files
954 * @base_files: true if the base files should be removed
955 * @subsys_mask: mask of the subsystem ids whose files should be removed
957 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
958 unsigned long subsys_mask)
960 struct cgroup *cgrp = __d_cgrp(dir);
961 struct cgroup_subsys *ss;
963 for_each_subsys(cgrp->root, ss) {
964 struct cftype_set *set;
965 if (!test_bit(ss->subsys_id, &subsys_mask))
967 list_for_each_entry(set, &ss->cftsets, node)
968 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
971 while (!list_empty(&cgrp->files))
972 cgroup_rm_file(cgrp, NULL);
977 * NOTE : the dentry must have been dget()'ed
979 static void cgroup_d_remove_dir(struct dentry *dentry)
981 struct dentry *parent;
982 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
984 cgroup_clear_directory(dentry, true, root->subsys_mask);
986 parent = dentry->d_parent;
987 spin_lock(&parent->d_lock);
988 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
989 list_del_init(&dentry->d_u.d_child);
990 spin_unlock(&dentry->d_lock);
991 spin_unlock(&parent->d_lock);
996 * Call with cgroup_mutex held. Drops reference counts on modules, including
997 * any duplicate ones that parse_cgroupfs_options took. If this function
998 * returns an error, no reference counts are touched.
1000 static int rebind_subsystems(struct cgroupfs_root *root,
1001 unsigned long final_subsys_mask)
1003 unsigned long added_mask, removed_mask;
1004 struct cgroup *cgrp = &root->top_cgroup;
1007 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1008 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1010 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
1011 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1012 /* Check that any added subsystems are currently free */
1013 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1014 unsigned long bit = 1UL << i;
1015 struct cgroup_subsys *ss = subsys[i];
1016 if (!(bit & added_mask))
1019 * Nobody should tell us to do a subsys that doesn't exist:
1020 * parse_cgroupfs_options should catch that case and refcounts
1021 * ensure that subsystems won't disappear once selected.
1024 if (ss->root != &rootnode) {
1025 /* Subsystem isn't free */
1030 /* Currently we don't handle adding/removing subsystems when
1031 * any child cgroups exist. This is theoretically supportable
1032 * but involves complex error handling, so it's being left until
1034 if (root->number_of_cgroups > 1)
1037 /* Process each subsystem */
1038 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1039 struct cgroup_subsys *ss = subsys[i];
1040 unsigned long bit = 1UL << i;
1041 if (bit & added_mask) {
1042 /* We're binding this subsystem to this hierarchy */
1044 BUG_ON(cgrp->subsys[i]);
1045 BUG_ON(!dummytop->subsys[i]);
1046 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1047 cgrp->subsys[i] = dummytop->subsys[i];
1048 cgrp->subsys[i]->cgroup = cgrp;
1049 list_move(&ss->sibling, &root->subsys_list);
1053 /* refcount was already taken, and we're keeping it */
1054 } else if (bit & removed_mask) {
1055 /* We're removing this subsystem */
1057 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1058 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1061 dummytop->subsys[i]->cgroup = dummytop;
1062 cgrp->subsys[i] = NULL;
1063 subsys[i]->root = &rootnode;
1064 list_move(&ss->sibling, &rootnode.subsys_list);
1065 /* subsystem is now free - drop reference on module */
1066 module_put(ss->module);
1067 } else if (bit & final_subsys_mask) {
1068 /* 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]);
1084 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1089 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1091 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1092 struct cgroup_subsys *ss;
1094 mutex_lock(&cgroup_root_mutex);
1095 for_each_subsys(root, ss)
1096 seq_printf(seq, ",%s", ss->name);
1097 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR)
1098 seq_puts(seq, ",sane_behavior");
1099 if (root->flags & CGRP_ROOT_NOPREFIX)
1100 seq_puts(seq, ",noprefix");
1101 if (root->flags & CGRP_ROOT_XATTR)
1102 seq_puts(seq, ",xattr");
1103 if (strlen(root->release_agent_path))
1104 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1105 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1106 seq_puts(seq, ",clone_children");
1107 if (strlen(root->name))
1108 seq_printf(seq, ",name=%s", root->name);
1109 mutex_unlock(&cgroup_root_mutex);
1113 struct cgroup_sb_opts {
1114 unsigned long subsys_mask;
1115 unsigned long flags;
1116 char *release_agent;
1117 bool cpuset_clone_children;
1119 /* User explicitly requested empty subsystem */
1122 struct cgroupfs_root *new_root;
1127 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1128 * with cgroup_mutex held to protect the subsys[] array. This function takes
1129 * refcounts on subsystems to be used, unless it returns error, in which case
1130 * no refcounts are taken.
1132 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1134 char *token, *o = data;
1135 bool all_ss = false, one_ss = false;
1136 unsigned long mask = (unsigned long)-1;
1138 bool module_pin_failed = false;
1140 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1142 #ifdef CONFIG_CPUSETS
1143 mask = ~(1UL << cpuset_subsys_id);
1146 memset(opts, 0, sizeof(*opts));
1148 while ((token = strsep(&o, ",")) != NULL) {
1151 if (!strcmp(token, "none")) {
1152 /* Explicitly have no subsystems */
1156 if (!strcmp(token, "all")) {
1157 /* Mutually exclusive option 'all' + subsystem name */
1163 if (!strcmp(token, "__DEVEL__sane_behavior")) {
1164 opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
1167 if (!strcmp(token, "noprefix")) {
1168 opts->flags |= CGRP_ROOT_NOPREFIX;
1171 if (!strcmp(token, "clone_children")) {
1172 opts->cpuset_clone_children = true;
1175 if (!strcmp(token, "xattr")) {
1176 opts->flags |= CGRP_ROOT_XATTR;
1179 if (!strncmp(token, "release_agent=", 14)) {
1180 /* Specifying two release agents is forbidden */
1181 if (opts->release_agent)
1183 opts->release_agent =
1184 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1185 if (!opts->release_agent)
1189 if (!strncmp(token, "name=", 5)) {
1190 const char *name = token + 5;
1191 /* Can't specify an empty name */
1194 /* Must match [\w.-]+ */
1195 for (i = 0; i < strlen(name); i++) {
1199 if ((c == '.') || (c == '-') || (c == '_'))
1203 /* Specifying two names is forbidden */
1206 opts->name = kstrndup(name,
1207 MAX_CGROUP_ROOT_NAMELEN - 1,
1215 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1216 struct cgroup_subsys *ss = subsys[i];
1219 if (strcmp(token, ss->name))
1224 /* Mutually exclusive option 'all' + subsystem name */
1227 set_bit(i, &opts->subsys_mask);
1232 if (i == CGROUP_SUBSYS_COUNT)
1237 * If the 'all' option was specified select all the subsystems,
1238 * otherwise if 'none', 'name=' and a subsystem name options
1239 * were not specified, let's default to 'all'
1241 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1242 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1243 struct cgroup_subsys *ss = subsys[i];
1248 set_bit(i, &opts->subsys_mask);
1252 /* Consistency checks */
1254 if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1255 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1257 if (opts->flags & CGRP_ROOT_NOPREFIX) {
1258 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1262 if (opts->cpuset_clone_children) {
1263 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1269 * Option noprefix was introduced just for backward compatibility
1270 * with the old cpuset, so we allow noprefix only if mounting just
1271 * the cpuset subsystem.
1273 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1277 /* Can't specify "none" and some subsystems */
1278 if (opts->subsys_mask && opts->none)
1282 * We either have to specify by name or by subsystems. (So all
1283 * empty hierarchies must have a name).
1285 if (!opts->subsys_mask && !opts->name)
1289 * Grab references on all the modules we'll need, so the subsystems
1290 * don't dance around before rebind_subsystems attaches them. This may
1291 * take duplicate reference counts on a subsystem that's already used,
1292 * but rebind_subsystems handles this case.
1294 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1295 unsigned long bit = 1UL << i;
1297 if (!(bit & opts->subsys_mask))
1299 if (!try_module_get(subsys[i]->module)) {
1300 module_pin_failed = true;
1304 if (module_pin_failed) {
1306 * oops, one of the modules was going away. this means that we
1307 * raced with a module_delete call, and to the user this is
1308 * essentially a "subsystem doesn't exist" case.
1310 for (i--; i >= 0; i--) {
1311 /* drop refcounts only on the ones we took */
1312 unsigned long bit = 1UL << i;
1314 if (!(bit & opts->subsys_mask))
1316 module_put(subsys[i]->module);
1324 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1327 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1328 unsigned long bit = 1UL << i;
1330 if (!(bit & subsys_mask))
1332 module_put(subsys[i]->module);
1336 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1339 struct cgroupfs_root *root = sb->s_fs_info;
1340 struct cgroup *cgrp = &root->top_cgroup;
1341 struct cgroup_sb_opts opts;
1342 unsigned long added_mask, removed_mask;
1344 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1345 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1349 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1350 mutex_lock(&cgroup_mutex);
1351 mutex_lock(&cgroup_root_mutex);
1353 /* See what subsystems are wanted */
1354 ret = parse_cgroupfs_options(data, &opts);
1358 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1359 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1360 task_tgid_nr(current), current->comm);
1362 added_mask = opts.subsys_mask & ~root->subsys_mask;
1363 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1365 /* Don't allow flags or name to change at remount */
1366 if (opts.flags != root->flags ||
1367 (opts.name && strcmp(opts.name, root->name))) {
1369 drop_parsed_module_refcounts(opts.subsys_mask);
1374 * Clear out the files of subsystems that should be removed, do
1375 * this before rebind_subsystems, since rebind_subsystems may
1376 * change this hierarchy's subsys_list.
1378 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1380 ret = rebind_subsystems(root, opts.subsys_mask);
1382 /* rebind_subsystems failed, re-populate the removed files */
1383 cgroup_populate_dir(cgrp, false, removed_mask);
1384 drop_parsed_module_refcounts(opts.subsys_mask);
1388 /* re-populate subsystem files */
1389 cgroup_populate_dir(cgrp, false, added_mask);
1391 if (opts.release_agent)
1392 strcpy(root->release_agent_path, opts.release_agent);
1394 kfree(opts.release_agent);
1396 mutex_unlock(&cgroup_root_mutex);
1397 mutex_unlock(&cgroup_mutex);
1398 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1402 static const struct super_operations cgroup_ops = {
1403 .statfs = simple_statfs,
1404 .drop_inode = generic_delete_inode,
1405 .show_options = cgroup_show_options,
1406 .remount_fs = cgroup_remount,
1409 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1411 INIT_LIST_HEAD(&cgrp->sibling);
1412 INIT_LIST_HEAD(&cgrp->children);
1413 INIT_LIST_HEAD(&cgrp->files);
1414 INIT_LIST_HEAD(&cgrp->cset_links);
1415 INIT_LIST_HEAD(&cgrp->allcg_node);
1416 INIT_LIST_HEAD(&cgrp->release_list);
1417 INIT_LIST_HEAD(&cgrp->pidlists);
1418 INIT_WORK(&cgrp->free_work, cgroup_free_fn);
1419 mutex_init(&cgrp->pidlist_mutex);
1420 INIT_LIST_HEAD(&cgrp->event_list);
1421 spin_lock_init(&cgrp->event_list_lock);
1422 simple_xattrs_init(&cgrp->xattrs);
1425 static void init_cgroup_root(struct cgroupfs_root *root)
1427 struct cgroup *cgrp = &root->top_cgroup;
1429 INIT_LIST_HEAD(&root->subsys_list);
1430 INIT_LIST_HEAD(&root->root_list);
1431 INIT_LIST_HEAD(&root->allcg_list);
1432 root->number_of_cgroups = 1;
1434 cgrp->name = &root_cgroup_name;
1435 init_cgroup_housekeeping(cgrp);
1436 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1439 static int cgroup_init_root_id(struct cgroupfs_root *root)
1443 lockdep_assert_held(&cgroup_mutex);
1444 lockdep_assert_held(&cgroup_root_mutex);
1446 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 2, 0, GFP_KERNEL);
1450 root->hierarchy_id = id;
1454 static void cgroup_exit_root_id(struct cgroupfs_root *root)
1456 lockdep_assert_held(&cgroup_mutex);
1457 lockdep_assert_held(&cgroup_root_mutex);
1459 if (root->hierarchy_id) {
1460 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1461 root->hierarchy_id = 0;
1465 static int cgroup_test_super(struct super_block *sb, void *data)
1467 struct cgroup_sb_opts *opts = data;
1468 struct cgroupfs_root *root = sb->s_fs_info;
1470 /* If we asked for a name then it must match */
1471 if (opts->name && strcmp(opts->name, root->name))
1475 * If we asked for subsystems (or explicitly for no
1476 * subsystems) then they must match
1478 if ((opts->subsys_mask || opts->none)
1479 && (opts->subsys_mask != root->subsys_mask))
1485 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1487 struct cgroupfs_root *root;
1489 if (!opts->subsys_mask && !opts->none)
1492 root = kzalloc(sizeof(*root), GFP_KERNEL);
1494 return ERR_PTR(-ENOMEM);
1496 init_cgroup_root(root);
1498 root->subsys_mask = opts->subsys_mask;
1499 root->flags = opts->flags;
1500 ida_init(&root->cgroup_ida);
1501 if (opts->release_agent)
1502 strcpy(root->release_agent_path, opts->release_agent);
1504 strcpy(root->name, opts->name);
1505 if (opts->cpuset_clone_children)
1506 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1510 static void cgroup_free_root(struct cgroupfs_root *root)
1513 /* hierarhcy ID shoulid already have been released */
1514 WARN_ON_ONCE(root->hierarchy_id);
1516 ida_destroy(&root->cgroup_ida);
1521 static int cgroup_set_super(struct super_block *sb, void *data)
1524 struct cgroup_sb_opts *opts = data;
1526 /* If we don't have a new root, we can't set up a new sb */
1527 if (!opts->new_root)
1530 BUG_ON(!opts->subsys_mask && !opts->none);
1532 ret = set_anon_super(sb, NULL);
1536 sb->s_fs_info = opts->new_root;
1537 opts->new_root->sb = sb;
1539 sb->s_blocksize = PAGE_CACHE_SIZE;
1540 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1541 sb->s_magic = CGROUP_SUPER_MAGIC;
1542 sb->s_op = &cgroup_ops;
1547 static int cgroup_get_rootdir(struct super_block *sb)
1549 static const struct dentry_operations cgroup_dops = {
1550 .d_iput = cgroup_diput,
1551 .d_delete = cgroup_delete,
1554 struct inode *inode =
1555 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1560 inode->i_fop = &simple_dir_operations;
1561 inode->i_op = &cgroup_dir_inode_operations;
1562 /* directories start off with i_nlink == 2 (for "." entry) */
1564 sb->s_root = d_make_root(inode);
1567 /* for everything else we want ->d_op set */
1568 sb->s_d_op = &cgroup_dops;
1572 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1573 int flags, const char *unused_dev_name,
1576 struct cgroup_sb_opts opts;
1577 struct cgroupfs_root *root;
1579 struct super_block *sb;
1580 struct cgroupfs_root *new_root;
1581 struct inode *inode;
1583 /* First find the desired set of subsystems */
1584 mutex_lock(&cgroup_mutex);
1585 ret = parse_cgroupfs_options(data, &opts);
1586 mutex_unlock(&cgroup_mutex);
1591 * Allocate a new cgroup root. We may not need it if we're
1592 * reusing an existing hierarchy.
1594 new_root = cgroup_root_from_opts(&opts);
1595 if (IS_ERR(new_root)) {
1596 ret = PTR_ERR(new_root);
1599 opts.new_root = new_root;
1601 /* Locate an existing or new sb for this hierarchy */
1602 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1605 cgroup_free_root(opts.new_root);
1609 root = sb->s_fs_info;
1611 if (root == opts.new_root) {
1612 /* We used the new root structure, so this is a new hierarchy */
1613 struct list_head tmp_links;
1614 struct cgroup *root_cgrp = &root->top_cgroup;
1615 struct cgroupfs_root *existing_root;
1616 const struct cred *cred;
1618 struct css_set *cset;
1620 BUG_ON(sb->s_root != NULL);
1622 ret = cgroup_get_rootdir(sb);
1624 goto drop_new_super;
1625 inode = sb->s_root->d_inode;
1627 mutex_lock(&inode->i_mutex);
1628 mutex_lock(&cgroup_mutex);
1629 mutex_lock(&cgroup_root_mutex);
1631 /* Check for name clashes with existing mounts */
1633 if (strlen(root->name))
1634 for_each_active_root(existing_root)
1635 if (!strcmp(existing_root->name, root->name))
1639 * We're accessing css_set_count without locking
1640 * css_set_lock here, but that's OK - it can only be
1641 * increased by someone holding cgroup_lock, and
1642 * that's us. The worst that can happen is that we
1643 * have some link structures left over
1645 ret = allocate_cgrp_cset_links(css_set_count, &tmp_links);
1649 ret = cgroup_init_root_id(root);
1653 ret = rebind_subsystems(root, root->subsys_mask);
1654 if (ret == -EBUSY) {
1655 free_cgrp_cset_links(&tmp_links);
1659 * There must be no failure case after here, since rebinding
1660 * takes care of subsystems' refcounts, which are explicitly
1661 * dropped in the failure exit path.
1664 /* EBUSY should be the only error here */
1667 list_add(&root->root_list, &roots);
1670 sb->s_root->d_fsdata = root_cgrp;
1671 root->top_cgroup.dentry = sb->s_root;
1673 /* Link the top cgroup in this hierarchy into all
1674 * the css_set objects */
1675 write_lock(&css_set_lock);
1676 hash_for_each(css_set_table, i, cset, hlist)
1677 link_css_set(&tmp_links, cset, root_cgrp);
1678 write_unlock(&css_set_lock);
1680 free_cgrp_cset_links(&tmp_links);
1682 BUG_ON(!list_empty(&root_cgrp->children));
1683 BUG_ON(root->number_of_cgroups != 1);
1685 cred = override_creds(&init_cred);
1686 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1688 mutex_unlock(&cgroup_root_mutex);
1689 mutex_unlock(&cgroup_mutex);
1690 mutex_unlock(&inode->i_mutex);
1693 * We re-used an existing hierarchy - the new root (if
1694 * any) is not needed
1696 cgroup_free_root(opts.new_root);
1698 if (((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) &&
1699 root->flags != opts.flags) {
1700 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1702 goto drop_new_super;
1705 /* no subsys rebinding, so refcounts don't change */
1706 drop_parsed_module_refcounts(opts.subsys_mask);
1709 kfree(opts.release_agent);
1711 return dget(sb->s_root);
1714 cgroup_exit_root_id(root);
1715 mutex_unlock(&cgroup_root_mutex);
1716 mutex_unlock(&cgroup_mutex);
1717 mutex_unlock(&inode->i_mutex);
1719 deactivate_locked_super(sb);
1721 drop_parsed_module_refcounts(opts.subsys_mask);
1723 kfree(opts.release_agent);
1725 return ERR_PTR(ret);
1728 static void cgroup_kill_sb(struct super_block *sb) {
1729 struct cgroupfs_root *root = sb->s_fs_info;
1730 struct cgroup *cgrp = &root->top_cgroup;
1731 struct cgrp_cset_link *link, *tmp_link;
1736 BUG_ON(root->number_of_cgroups != 1);
1737 BUG_ON(!list_empty(&cgrp->children));
1739 mutex_lock(&cgroup_mutex);
1740 mutex_lock(&cgroup_root_mutex);
1742 /* Rebind all subsystems back to the default hierarchy */
1743 ret = rebind_subsystems(root, 0);
1744 /* Shouldn't be able to fail ... */
1748 * Release all the links from cset_links to this hierarchy's
1751 write_lock(&css_set_lock);
1753 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1754 list_del(&link->cset_link);
1755 list_del(&link->cgrp_link);
1758 write_unlock(&css_set_lock);
1760 if (!list_empty(&root->root_list)) {
1761 list_del(&root->root_list);
1765 cgroup_exit_root_id(root);
1767 mutex_unlock(&cgroup_root_mutex);
1768 mutex_unlock(&cgroup_mutex);
1770 simple_xattrs_free(&cgrp->xattrs);
1772 kill_litter_super(sb);
1773 cgroup_free_root(root);
1776 static struct file_system_type cgroup_fs_type = {
1778 .mount = cgroup_mount,
1779 .kill_sb = cgroup_kill_sb,
1782 static struct kobject *cgroup_kobj;
1785 * cgroup_path - generate the path of a cgroup
1786 * @cgrp: the cgroup in question
1787 * @buf: the buffer to write the path into
1788 * @buflen: the length of the buffer
1790 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1792 * We can't generate cgroup path using dentry->d_name, as accessing
1793 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1794 * inode's i_mutex, while on the other hand cgroup_path() can be called
1795 * with some irq-safe spinlocks held.
1797 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1799 int ret = -ENAMETOOLONG;
1802 if (!cgrp->parent) {
1803 if (strlcpy(buf, "/", buflen) >= buflen)
1804 return -ENAMETOOLONG;
1808 start = buf + buflen - 1;
1813 const char *name = cgroup_name(cgrp);
1817 if ((start -= len) < buf)
1819 memcpy(start, name, len);
1825 cgrp = cgrp->parent;
1826 } while (cgrp->parent);
1828 memmove(buf, start, buf + buflen - start);
1833 EXPORT_SYMBOL_GPL(cgroup_path);
1836 * task_cgroup_path_from_hierarchy - cgroup path of a task on a hierarchy
1837 * @task: target task
1838 * @hierarchy_id: the hierarchy to look up @task's cgroup from
1839 * @buf: the buffer to write the path into
1840 * @buflen: the length of the buffer
1842 * Determine @task's cgroup on the hierarchy specified by @hierarchy_id and
1843 * copy its path into @buf. This function grabs cgroup_mutex and shouldn't
1844 * be used inside locks used by cgroup controller callbacks.
1846 int task_cgroup_path_from_hierarchy(struct task_struct *task, int hierarchy_id,
1847 char *buf, size_t buflen)
1849 struct cgroupfs_root *root;
1850 struct cgroup *cgrp = NULL;
1853 mutex_lock(&cgroup_mutex);
1855 root = idr_find(&cgroup_hierarchy_idr, hierarchy_id);
1857 cgrp = task_cgroup_from_root(task, root);
1858 ret = cgroup_path(cgrp, buf, buflen);
1861 mutex_unlock(&cgroup_mutex);
1865 EXPORT_SYMBOL_GPL(task_cgroup_path_from_hierarchy);
1868 * Control Group taskset
1870 struct task_and_cgroup {
1871 struct task_struct *task;
1872 struct cgroup *cgrp;
1876 struct cgroup_taskset {
1877 struct task_and_cgroup single;
1878 struct flex_array *tc_array;
1881 struct cgroup *cur_cgrp;
1885 * cgroup_taskset_first - reset taskset and return the first task
1886 * @tset: taskset of interest
1888 * @tset iteration is initialized and the first task is returned.
1890 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1892 if (tset->tc_array) {
1894 return cgroup_taskset_next(tset);
1896 tset->cur_cgrp = tset->single.cgrp;
1897 return tset->single.task;
1900 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1903 * cgroup_taskset_next - iterate to the next task in taskset
1904 * @tset: taskset of interest
1906 * Return the next task in @tset. Iteration must have been initialized
1907 * with cgroup_taskset_first().
1909 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1911 struct task_and_cgroup *tc;
1913 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1916 tc = flex_array_get(tset->tc_array, tset->idx++);
1917 tset->cur_cgrp = tc->cgrp;
1920 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1923 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1924 * @tset: taskset of interest
1926 * Return the cgroup for the current (last returned) task of @tset. This
1927 * function must be preceded by either cgroup_taskset_first() or
1928 * cgroup_taskset_next().
1930 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1932 return tset->cur_cgrp;
1934 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1937 * cgroup_taskset_size - return the number of tasks in taskset
1938 * @tset: taskset of interest
1940 int cgroup_taskset_size(struct cgroup_taskset *tset)
1942 return tset->tc_array ? tset->tc_array_len : 1;
1944 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1948 * cgroup_task_migrate - move a task from one cgroup to another.
1950 * Must be called with cgroup_mutex and threadgroup locked.
1952 static void cgroup_task_migrate(struct cgroup *old_cgrp,
1953 struct task_struct *tsk,
1954 struct css_set *new_cset)
1956 struct css_set *old_cset;
1959 * We are synchronized through threadgroup_lock() against PF_EXITING
1960 * setting such that we can't race against cgroup_exit() changing the
1961 * css_set to init_css_set and dropping the old one.
1963 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1964 old_cset = tsk->cgroups;
1967 rcu_assign_pointer(tsk->cgroups, new_cset);
1970 /* Update the css_set linked lists if we're using them */
1971 write_lock(&css_set_lock);
1972 if (!list_empty(&tsk->cg_list))
1973 list_move(&tsk->cg_list, &new_cset->tasks);
1974 write_unlock(&css_set_lock);
1977 * We just gained a reference on old_cset by taking it from the
1978 * task. As trading it for new_cset is protected by cgroup_mutex,
1979 * we're safe to drop it here; it will be freed under RCU.
1981 set_bit(CGRP_RELEASABLE, &old_cgrp->flags);
1982 put_css_set(old_cset);
1986 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1987 * @cgrp: the cgroup to attach to
1988 * @tsk: the task or the leader of the threadgroup to be attached
1989 * @threadgroup: attach the whole threadgroup?
1991 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1992 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1994 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
1997 int retval, i, group_size;
1998 struct cgroup_subsys *ss, *failed_ss = NULL;
1999 struct cgroupfs_root *root = cgrp->root;
2000 /* threadgroup list cursor and array */
2001 struct task_struct *leader = tsk;
2002 struct task_and_cgroup *tc;
2003 struct flex_array *group;
2004 struct cgroup_taskset tset = { };
2007 * step 0: in order to do expensive, possibly blocking operations for
2008 * every thread, we cannot iterate the thread group list, since it needs
2009 * rcu or tasklist locked. instead, build an array of all threads in the
2010 * group - group_rwsem prevents new threads from appearing, and if
2011 * threads exit, this will just be an over-estimate.
2014 group_size = get_nr_threads(tsk);
2017 /* flex_array supports very large thread-groups better than kmalloc. */
2018 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2021 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2022 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
2024 goto out_free_group_list;
2028 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2029 * already PF_EXITING could be freed from underneath us unless we
2030 * take an rcu_read_lock.
2034 struct task_and_cgroup ent;
2036 /* @tsk either already exited or can't exit until the end */
2037 if (tsk->flags & PF_EXITING)
2040 /* as per above, nr_threads may decrease, but not increase. */
2041 BUG_ON(i >= group_size);
2043 ent.cgrp = task_cgroup_from_root(tsk, root);
2044 /* nothing to do if this task is already in the cgroup */
2045 if (ent.cgrp == cgrp)
2048 * saying GFP_ATOMIC has no effect here because we did prealloc
2049 * earlier, but it's good form to communicate our expectations.
2051 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2052 BUG_ON(retval != 0);
2057 } while_each_thread(leader, tsk);
2059 /* remember the number of threads in the array for later. */
2061 tset.tc_array = group;
2062 tset.tc_array_len = group_size;
2064 /* methods shouldn't be called if no task is actually migrating */
2067 goto out_free_group_list;
2070 * step 1: check that we can legitimately attach to the cgroup.
2072 for_each_subsys(root, ss) {
2073 if (ss->can_attach) {
2074 retval = ss->can_attach(cgrp, &tset);
2077 goto out_cancel_attach;
2083 * step 2: make sure css_sets exist for all threads to be migrated.
2084 * we use find_css_set, which allocates a new one if necessary.
2086 for (i = 0; i < group_size; i++) {
2087 tc = flex_array_get(group, i);
2088 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2091 goto out_put_css_set_refs;
2096 * step 3: now that we're guaranteed success wrt the css_sets,
2097 * proceed to move all tasks to the new cgroup. There are no
2098 * failure cases after here, so this is the commit point.
2100 for (i = 0; i < group_size; i++) {
2101 tc = flex_array_get(group, i);
2102 cgroup_task_migrate(tc->cgrp, tc->task, tc->cg);
2104 /* nothing is sensitive to fork() after this point. */
2107 * step 4: do subsystem attach callbacks.
2109 for_each_subsys(root, ss) {
2111 ss->attach(cgrp, &tset);
2115 * step 5: success! and cleanup
2118 out_put_css_set_refs:
2120 for (i = 0; i < group_size; i++) {
2121 tc = flex_array_get(group, i);
2124 put_css_set(tc->cg);
2129 for_each_subsys(root, ss) {
2130 if (ss == failed_ss)
2132 if (ss->cancel_attach)
2133 ss->cancel_attach(cgrp, &tset);
2136 out_free_group_list:
2137 flex_array_free(group);
2142 * Find the task_struct of the task to attach by vpid and pass it along to the
2143 * function to attach either it or all tasks in its threadgroup. Will lock
2144 * cgroup_mutex and threadgroup; may take task_lock of task.
2146 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2148 struct task_struct *tsk;
2149 const struct cred *cred = current_cred(), *tcred;
2152 if (!cgroup_lock_live_group(cgrp))
2158 tsk = find_task_by_vpid(pid);
2162 goto out_unlock_cgroup;
2165 * even if we're attaching all tasks in the thread group, we
2166 * only need to check permissions on one of them.
2168 tcred = __task_cred(tsk);
2169 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2170 !uid_eq(cred->euid, tcred->uid) &&
2171 !uid_eq(cred->euid, tcred->suid)) {
2174 goto out_unlock_cgroup;
2180 tsk = tsk->group_leader;
2183 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2184 * trapped in a cpuset, or RT worker may be born in a cgroup
2185 * with no rt_runtime allocated. Just say no.
2187 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
2190 goto out_unlock_cgroup;
2193 get_task_struct(tsk);
2196 threadgroup_lock(tsk);
2198 if (!thread_group_leader(tsk)) {
2200 * a race with de_thread from another thread's exec()
2201 * may strip us of our leadership, if this happens,
2202 * there is no choice but to throw this task away and
2203 * try again; this is
2204 * "double-double-toil-and-trouble-check locking".
2206 threadgroup_unlock(tsk);
2207 put_task_struct(tsk);
2208 goto retry_find_task;
2212 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2214 threadgroup_unlock(tsk);
2216 put_task_struct(tsk);
2218 mutex_unlock(&cgroup_mutex);
2223 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2224 * @from: attach to all cgroups of a given task
2225 * @tsk: the task to be attached
2227 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2229 struct cgroupfs_root *root;
2232 mutex_lock(&cgroup_mutex);
2233 for_each_active_root(root) {
2234 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2236 retval = cgroup_attach_task(from_cg, tsk, false);
2240 mutex_unlock(&cgroup_mutex);
2244 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2246 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2248 return attach_task_by_pid(cgrp, pid, false);
2251 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2253 return attach_task_by_pid(cgrp, tgid, true);
2256 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2259 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2260 if (strlen(buffer) >= PATH_MAX)
2262 if (!cgroup_lock_live_group(cgrp))
2264 mutex_lock(&cgroup_root_mutex);
2265 strcpy(cgrp->root->release_agent_path, buffer);
2266 mutex_unlock(&cgroup_root_mutex);
2267 mutex_unlock(&cgroup_mutex);
2271 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2272 struct seq_file *seq)
2274 if (!cgroup_lock_live_group(cgrp))
2276 seq_puts(seq, cgrp->root->release_agent_path);
2277 seq_putc(seq, '\n');
2278 mutex_unlock(&cgroup_mutex);
2282 static int cgroup_sane_behavior_show(struct cgroup *cgrp, struct cftype *cft,
2283 struct seq_file *seq)
2285 seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp));
2289 /* A buffer size big enough for numbers or short strings */
2290 #define CGROUP_LOCAL_BUFFER_SIZE 64
2292 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2294 const char __user *userbuf,
2295 size_t nbytes, loff_t *unused_ppos)
2297 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2303 if (nbytes >= sizeof(buffer))
2305 if (copy_from_user(buffer, userbuf, nbytes))
2308 buffer[nbytes] = 0; /* nul-terminate */
2309 if (cft->write_u64) {
2310 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2313 retval = cft->write_u64(cgrp, cft, val);
2315 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2318 retval = cft->write_s64(cgrp, cft, val);
2325 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2327 const char __user *userbuf,
2328 size_t nbytes, loff_t *unused_ppos)
2330 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2332 size_t max_bytes = cft->max_write_len;
2333 char *buffer = local_buffer;
2336 max_bytes = sizeof(local_buffer) - 1;
2337 if (nbytes >= max_bytes)
2339 /* Allocate a dynamic buffer if we need one */
2340 if (nbytes >= sizeof(local_buffer)) {
2341 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2345 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2350 buffer[nbytes] = 0; /* nul-terminate */
2351 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2355 if (buffer != local_buffer)
2360 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2361 size_t nbytes, loff_t *ppos)
2363 struct cftype *cft = __d_cft(file->f_dentry);
2364 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2366 if (cgroup_is_removed(cgrp))
2369 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2370 if (cft->write_u64 || cft->write_s64)
2371 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2372 if (cft->write_string)
2373 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2375 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2376 return ret ? ret : nbytes;
2381 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2383 char __user *buf, size_t nbytes,
2386 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2387 u64 val = cft->read_u64(cgrp, cft);
2388 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2390 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2393 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2395 char __user *buf, size_t nbytes,
2398 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2399 s64 val = cft->read_s64(cgrp, cft);
2400 int len = sprintf(tmp, "%lld\n", (long long) val);
2402 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2405 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2406 size_t nbytes, loff_t *ppos)
2408 struct cftype *cft = __d_cft(file->f_dentry);
2409 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2411 if (cgroup_is_removed(cgrp))
2415 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2417 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2419 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2424 * seqfile ops/methods for returning structured data. Currently just
2425 * supports string->u64 maps, but can be extended in future.
2428 struct cgroup_seqfile_state {
2430 struct cgroup *cgroup;
2433 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2435 struct seq_file *sf = cb->state;
2436 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2439 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2441 struct cgroup_seqfile_state *state = m->private;
2442 struct cftype *cft = state->cft;
2443 if (cft->read_map) {
2444 struct cgroup_map_cb cb = {
2445 .fill = cgroup_map_add,
2448 return cft->read_map(state->cgroup, cft, &cb);
2450 return cft->read_seq_string(state->cgroup, cft, m);
2453 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2455 struct seq_file *seq = file->private_data;
2456 kfree(seq->private);
2457 return single_release(inode, file);
2460 static const struct file_operations cgroup_seqfile_operations = {
2462 .write = cgroup_file_write,
2463 .llseek = seq_lseek,
2464 .release = cgroup_seqfile_release,
2467 static int cgroup_file_open(struct inode *inode, struct file *file)
2472 err = generic_file_open(inode, file);
2475 cft = __d_cft(file->f_dentry);
2477 if (cft->read_map || cft->read_seq_string) {
2478 struct cgroup_seqfile_state *state;
2480 state = kzalloc(sizeof(*state), GFP_USER);
2485 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2486 file->f_op = &cgroup_seqfile_operations;
2487 err = single_open(file, cgroup_seqfile_show, state);
2490 } else if (cft->open)
2491 err = cft->open(inode, file);
2498 static int cgroup_file_release(struct inode *inode, struct file *file)
2500 struct cftype *cft = __d_cft(file->f_dentry);
2502 return cft->release(inode, file);
2507 * cgroup_rename - Only allow simple rename of directories in place.
2509 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2510 struct inode *new_dir, struct dentry *new_dentry)
2513 struct cgroup_name *name, *old_name;
2514 struct cgroup *cgrp;
2517 * It's convinient to use parent dir's i_mutex to protected
2520 lockdep_assert_held(&old_dir->i_mutex);
2522 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2524 if (new_dentry->d_inode)
2526 if (old_dir != new_dir)
2529 cgrp = __d_cgrp(old_dentry);
2531 name = cgroup_alloc_name(new_dentry);
2535 ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2541 old_name = cgrp->name;
2542 rcu_assign_pointer(cgrp->name, name);
2544 kfree_rcu(old_name, rcu_head);
2548 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2550 if (S_ISDIR(dentry->d_inode->i_mode))
2551 return &__d_cgrp(dentry)->xattrs;
2553 return &__d_cfe(dentry)->xattrs;
2556 static inline int xattr_enabled(struct dentry *dentry)
2558 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2559 return root->flags & CGRP_ROOT_XATTR;
2562 static bool is_valid_xattr(const char *name)
2564 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2565 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2570 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2571 const void *val, size_t size, int flags)
2573 if (!xattr_enabled(dentry))
2575 if (!is_valid_xattr(name))
2577 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2580 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2582 if (!xattr_enabled(dentry))
2584 if (!is_valid_xattr(name))
2586 return simple_xattr_remove(__d_xattrs(dentry), name);
2589 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2590 void *buf, size_t size)
2592 if (!xattr_enabled(dentry))
2594 if (!is_valid_xattr(name))
2596 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2599 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2601 if (!xattr_enabled(dentry))
2603 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2606 static const struct file_operations cgroup_file_operations = {
2607 .read = cgroup_file_read,
2608 .write = cgroup_file_write,
2609 .llseek = generic_file_llseek,
2610 .open = cgroup_file_open,
2611 .release = cgroup_file_release,
2614 static const struct inode_operations cgroup_file_inode_operations = {
2615 .setxattr = cgroup_setxattr,
2616 .getxattr = cgroup_getxattr,
2617 .listxattr = cgroup_listxattr,
2618 .removexattr = cgroup_removexattr,
2621 static const struct inode_operations cgroup_dir_inode_operations = {
2622 .lookup = cgroup_lookup,
2623 .mkdir = cgroup_mkdir,
2624 .rmdir = cgroup_rmdir,
2625 .rename = cgroup_rename,
2626 .setxattr = cgroup_setxattr,
2627 .getxattr = cgroup_getxattr,
2628 .listxattr = cgroup_listxattr,
2629 .removexattr = cgroup_removexattr,
2632 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2634 if (dentry->d_name.len > NAME_MAX)
2635 return ERR_PTR(-ENAMETOOLONG);
2636 d_add(dentry, NULL);
2641 * Check if a file is a control file
2643 static inline struct cftype *__file_cft(struct file *file)
2645 if (file_inode(file)->i_fop != &cgroup_file_operations)
2646 return ERR_PTR(-EINVAL);
2647 return __d_cft(file->f_dentry);
2650 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2651 struct super_block *sb)
2653 struct inode *inode;
2657 if (dentry->d_inode)
2660 inode = cgroup_new_inode(mode, sb);
2664 if (S_ISDIR(mode)) {
2665 inode->i_op = &cgroup_dir_inode_operations;
2666 inode->i_fop = &simple_dir_operations;
2668 /* start off with i_nlink == 2 (for "." entry) */
2670 inc_nlink(dentry->d_parent->d_inode);
2673 * Control reaches here with cgroup_mutex held.
2674 * @inode->i_mutex should nest outside cgroup_mutex but we
2675 * want to populate it immediately without releasing
2676 * cgroup_mutex. As @inode isn't visible to anyone else
2677 * yet, trylock will always succeed without affecting
2680 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2681 } else if (S_ISREG(mode)) {
2683 inode->i_fop = &cgroup_file_operations;
2684 inode->i_op = &cgroup_file_inode_operations;
2686 d_instantiate(dentry, inode);
2687 dget(dentry); /* Extra count - pin the dentry in core */
2692 * cgroup_file_mode - deduce file mode of a control file
2693 * @cft: the control file in question
2695 * returns cft->mode if ->mode is not 0
2696 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2697 * returns S_IRUGO if it has only a read handler
2698 * returns S_IWUSR if it has only a write hander
2700 static umode_t cgroup_file_mode(const struct cftype *cft)
2707 if (cft->read || cft->read_u64 || cft->read_s64 ||
2708 cft->read_map || cft->read_seq_string)
2711 if (cft->write || cft->write_u64 || cft->write_s64 ||
2712 cft->write_string || cft->trigger)
2718 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2721 struct dentry *dir = cgrp->dentry;
2722 struct cgroup *parent = __d_cgrp(dir);
2723 struct dentry *dentry;
2727 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2729 if (subsys && !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
2730 strcpy(name, subsys->name);
2733 strcat(name, cft->name);
2735 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2737 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2741 dentry = lookup_one_len(name, dir, strlen(name));
2742 if (IS_ERR(dentry)) {
2743 error = PTR_ERR(dentry);
2747 cfe->type = (void *)cft;
2748 cfe->dentry = dentry;
2749 dentry->d_fsdata = cfe;
2750 simple_xattrs_init(&cfe->xattrs);
2752 mode = cgroup_file_mode(cft);
2753 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2755 list_add_tail(&cfe->node, &parent->files);
2764 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2765 struct cftype cfts[], bool is_add)
2770 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2771 /* does cft->flags tell us to skip this file on @cgrp? */
2772 if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp))
2774 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2776 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2780 err = cgroup_add_file(cgrp, subsys, cft);
2782 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2786 cgroup_rm_file(cgrp, cft);
2792 static DEFINE_MUTEX(cgroup_cft_mutex);
2794 static void cgroup_cfts_prepare(void)
2795 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2798 * Thanks to the entanglement with vfs inode locking, we can't walk
2799 * the existing cgroups under cgroup_mutex and create files.
2800 * Instead, we increment reference on all cgroups and build list of
2801 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2802 * exclusive access to the field.
2804 mutex_lock(&cgroup_cft_mutex);
2805 mutex_lock(&cgroup_mutex);
2808 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2809 struct cftype *cfts, bool is_add)
2810 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2813 struct cgroup *cgrp, *n;
2815 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2816 if (cfts && ss->root != &rootnode) {
2817 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2819 list_add_tail(&cgrp->cft_q_node, &pending);
2823 mutex_unlock(&cgroup_mutex);
2826 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2827 * files for all cgroups which were created before.
2829 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2830 struct inode *inode = cgrp->dentry->d_inode;
2832 mutex_lock(&inode->i_mutex);
2833 mutex_lock(&cgroup_mutex);
2834 if (!cgroup_is_removed(cgrp))
2835 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2836 mutex_unlock(&cgroup_mutex);
2837 mutex_unlock(&inode->i_mutex);
2839 list_del_init(&cgrp->cft_q_node);
2843 mutex_unlock(&cgroup_cft_mutex);
2847 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2848 * @ss: target cgroup subsystem
2849 * @cfts: zero-length name terminated array of cftypes
2851 * Register @cfts to @ss. Files described by @cfts are created for all
2852 * existing cgroups to which @ss is attached and all future cgroups will
2853 * have them too. This function can be called anytime whether @ss is
2856 * Returns 0 on successful registration, -errno on failure. Note that this
2857 * function currently returns 0 as long as @cfts registration is successful
2858 * even if some file creation attempts on existing cgroups fail.
2860 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2862 struct cftype_set *set;
2864 set = kzalloc(sizeof(*set), GFP_KERNEL);
2868 cgroup_cfts_prepare();
2870 list_add_tail(&set->node, &ss->cftsets);
2871 cgroup_cfts_commit(ss, cfts, true);
2875 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2878 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2879 * @ss: target cgroup subsystem
2880 * @cfts: zero-length name terminated array of cftypes
2882 * Unregister @cfts from @ss. Files described by @cfts are removed from
2883 * all existing cgroups to which @ss is attached and all future cgroups
2884 * won't have them either. This function can be called anytime whether @ss
2885 * is attached or not.
2887 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2888 * registered with @ss.
2890 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2892 struct cftype_set *set;
2894 cgroup_cfts_prepare();
2896 list_for_each_entry(set, &ss->cftsets, node) {
2897 if (set->cfts == cfts) {
2898 list_del_init(&set->node);
2899 cgroup_cfts_commit(ss, cfts, false);
2904 cgroup_cfts_commit(ss, NULL, false);
2909 * cgroup_task_count - count the number of tasks in a cgroup.
2910 * @cgrp: the cgroup in question
2912 * Return the number of tasks in the cgroup.
2914 int cgroup_task_count(const struct cgroup *cgrp)
2917 struct cgrp_cset_link *link;
2919 read_lock(&css_set_lock);
2920 list_for_each_entry(link, &cgrp->cset_links, cset_link)
2921 count += atomic_read(&link->cset->refcount);
2922 read_unlock(&css_set_lock);
2927 * Advance a list_head iterator. The iterator should be positioned at
2928 * the start of a css_set
2930 static void cgroup_advance_iter(struct cgroup *cgrp, struct cgroup_iter *it)
2932 struct list_head *l = it->cset_link;
2933 struct cgrp_cset_link *link;
2934 struct css_set *cset;
2936 /* Advance to the next non-empty css_set */
2939 if (l == &cgrp->cset_links) {
2940 it->cset_link = NULL;
2943 link = list_entry(l, struct cgrp_cset_link, cset_link);
2945 } while (list_empty(&cset->tasks));
2947 it->task = cset->tasks.next;
2951 * To reduce the fork() overhead for systems that are not actually
2952 * using their cgroups capability, we don't maintain the lists running
2953 * through each css_set to its tasks until we see the list actually
2954 * used - in other words after the first call to cgroup_iter_start().
2956 static void cgroup_enable_task_cg_lists(void)
2958 struct task_struct *p, *g;
2959 write_lock(&css_set_lock);
2960 use_task_css_set_links = 1;
2962 * We need tasklist_lock because RCU is not safe against
2963 * while_each_thread(). Besides, a forking task that has passed
2964 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2965 * is not guaranteed to have its child immediately visible in the
2966 * tasklist if we walk through it with RCU.
2968 read_lock(&tasklist_lock);
2969 do_each_thread(g, p) {
2972 * We should check if the process is exiting, otherwise
2973 * it will race with cgroup_exit() in that the list
2974 * entry won't be deleted though the process has exited.
2976 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2977 list_add(&p->cg_list, &p->cgroups->tasks);
2979 } while_each_thread(g, p);
2980 read_unlock(&tasklist_lock);
2981 write_unlock(&css_set_lock);
2985 * cgroup_next_sibling - find the next sibling of a given cgroup
2986 * @pos: the current cgroup
2988 * This function returns the next sibling of @pos and should be called
2989 * under RCU read lock. The only requirement is that @pos is accessible.
2990 * The next sibling is guaranteed to be returned regardless of @pos's
2993 struct cgroup *cgroup_next_sibling(struct cgroup *pos)
2995 struct cgroup *next;
2997 WARN_ON_ONCE(!rcu_read_lock_held());
3000 * @pos could already have been removed. Once a cgroup is removed,
3001 * its ->sibling.next is no longer updated when its next sibling
3002 * changes. As CGRP_REMOVED is set on removal which is fully
3003 * serialized, if we see it unasserted, it's guaranteed that the
3004 * next sibling hasn't finished its grace period even if it's
3005 * already removed, and thus safe to dereference from this RCU
3006 * critical section. If ->sibling.next is inaccessible,
3007 * cgroup_is_removed() is guaranteed to be visible as %true here.
3009 if (likely(!cgroup_is_removed(pos))) {
3010 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3011 if (&next->sibling != &pos->parent->children)
3017 * Can't dereference the next pointer. Each cgroup is given a
3018 * monotonically increasing unique serial number and always
3019 * appended to the sibling list, so the next one can be found by
3020 * walking the parent's children until we see a cgroup with higher
3021 * serial number than @pos's.
3023 * While this path can be slow, it's taken only when either the
3024 * current cgroup is removed or iteration and removal race.
3026 list_for_each_entry_rcu(next, &pos->parent->children, sibling)
3027 if (next->serial_nr > pos->serial_nr)
3031 EXPORT_SYMBOL_GPL(cgroup_next_sibling);
3034 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
3035 * @pos: the current position (%NULL to initiate traversal)
3036 * @cgroup: cgroup whose descendants to walk
3038 * To be used by cgroup_for_each_descendant_pre(). Find the next
3039 * descendant to visit for pre-order traversal of @cgroup's descendants.
3041 * While this function requires RCU read locking, it doesn't require the
3042 * whole traversal to be contained in a single RCU critical section. This
3043 * function will return the correct next descendant as long as both @pos
3044 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3046 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
3047 struct cgroup *cgroup)
3049 struct cgroup *next;
3051 WARN_ON_ONCE(!rcu_read_lock_held());
3053 /* if first iteration, pretend we just visited @cgroup */
3057 /* visit the first child if exists */
3058 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
3062 /* no child, visit my or the closest ancestor's next sibling */
3063 while (pos != cgroup) {
3064 next = cgroup_next_sibling(pos);
3072 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
3075 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
3076 * @pos: cgroup of interest
3078 * Return the rightmost descendant of @pos. If there's no descendant,
3079 * @pos is returned. This can be used during pre-order traversal to skip
3082 * While this function requires RCU read locking, it doesn't require the
3083 * whole traversal to be contained in a single RCU critical section. This
3084 * function will return the correct rightmost descendant as long as @pos is
3087 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
3089 struct cgroup *last, *tmp;
3091 WARN_ON_ONCE(!rcu_read_lock_held());
3095 /* ->prev isn't RCU safe, walk ->next till the end */
3097 list_for_each_entry_rcu(tmp, &last->children, sibling)
3103 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
3105 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3107 struct cgroup *last;
3111 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3119 * cgroup_next_descendant_post - find the next descendant for post-order walk
3120 * @pos: the current position (%NULL to initiate traversal)
3121 * @cgroup: cgroup whose descendants to walk
3123 * To be used by cgroup_for_each_descendant_post(). Find the next
3124 * descendant to visit for post-order traversal of @cgroup's descendants.
3126 * While this function requires RCU read locking, it doesn't require the
3127 * whole traversal to be contained in a single RCU critical section. This
3128 * function will return the correct next descendant as long as both @pos
3129 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3131 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3132 struct cgroup *cgroup)
3134 struct cgroup *next;
3136 WARN_ON_ONCE(!rcu_read_lock_held());
3138 /* if first iteration, visit the leftmost descendant */
3140 next = cgroup_leftmost_descendant(cgroup);
3141 return next != cgroup ? next : NULL;
3144 /* if there's an unvisited sibling, visit its leftmost descendant */
3145 next = cgroup_next_sibling(pos);
3147 return cgroup_leftmost_descendant(next);
3149 /* no sibling left, visit parent */
3151 return next != cgroup ? next : NULL;
3153 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3155 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3156 __acquires(css_set_lock)
3159 * The first time anyone tries to iterate across a cgroup,
3160 * we need to enable the list linking each css_set to its
3161 * tasks, and fix up all existing tasks.
3163 if (!use_task_css_set_links)
3164 cgroup_enable_task_cg_lists();
3166 read_lock(&css_set_lock);
3167 it->cset_link = &cgrp->cset_links;
3168 cgroup_advance_iter(cgrp, it);
3171 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3172 struct cgroup_iter *it)
3174 struct task_struct *res;
3175 struct list_head *l = it->task;
3176 struct cgrp_cset_link *link;
3178 /* If the iterator cg is NULL, we have no tasks */
3181 res = list_entry(l, struct task_struct, cg_list);
3182 /* Advance iterator to find next entry */
3184 link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link);
3185 if (l == &link->cset->tasks) {
3186 /* We reached the end of this task list - move on to
3187 * the next cg_cgroup_link */
3188 cgroup_advance_iter(cgrp, it);
3195 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3196 __releases(css_set_lock)
3198 read_unlock(&css_set_lock);
3201 static inline int started_after_time(struct task_struct *t1,
3202 struct timespec *time,
3203 struct task_struct *t2)
3205 int start_diff = timespec_compare(&t1->start_time, time);
3206 if (start_diff > 0) {
3208 } else if (start_diff < 0) {
3212 * Arbitrarily, if two processes started at the same
3213 * time, we'll say that the lower pointer value
3214 * started first. Note that t2 may have exited by now
3215 * so this may not be a valid pointer any longer, but
3216 * that's fine - it still serves to distinguish
3217 * between two tasks started (effectively) simultaneously.
3224 * This function is a callback from heap_insert() and is used to order
3226 * In this case we order the heap in descending task start time.
3228 static inline int started_after(void *p1, void *p2)
3230 struct task_struct *t1 = p1;
3231 struct task_struct *t2 = p2;
3232 return started_after_time(t1, &t2->start_time, t2);
3236 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3237 * @scan: struct cgroup_scanner containing arguments for the scan
3239 * Arguments include pointers to callback functions test_task() and
3241 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3242 * and if it returns true, call process_task() for it also.
3243 * The test_task pointer may be NULL, meaning always true (select all tasks).
3244 * Effectively duplicates cgroup_iter_{start,next,end}()
3245 * but does not lock css_set_lock for the call to process_task().
3246 * The struct cgroup_scanner may be embedded in any structure of the caller's
3248 * It is guaranteed that process_task() will act on every task that
3249 * is a member of the cgroup for the duration of this call. This
3250 * function may or may not call process_task() for tasks that exit
3251 * or move to a different cgroup during the call, or are forked or
3252 * move into the cgroup during the call.
3254 * Note that test_task() may be called with locks held, and may in some
3255 * situations be called multiple times for the same task, so it should
3257 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3258 * pre-allocated and will be used for heap operations (and its "gt" member will
3259 * be overwritten), else a temporary heap will be used (allocation of which
3260 * may cause this function to fail).
3262 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3265 struct cgroup_iter it;
3266 struct task_struct *p, *dropped;
3267 /* Never dereference latest_task, since it's not refcounted */
3268 struct task_struct *latest_task = NULL;
3269 struct ptr_heap tmp_heap;
3270 struct ptr_heap *heap;
3271 struct timespec latest_time = { 0, 0 };
3274 /* The caller supplied our heap and pre-allocated its memory */
3276 heap->gt = &started_after;
3278 /* We need to allocate our own heap memory */
3280 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3282 /* cannot allocate the heap */
3288 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3289 * to determine which are of interest, and using the scanner's
3290 * "process_task" callback to process any of them that need an update.
3291 * Since we don't want to hold any locks during the task updates,
3292 * gather tasks to be processed in a heap structure.
3293 * The heap is sorted by descending task start time.
3294 * If the statically-sized heap fills up, we overflow tasks that
3295 * started later, and in future iterations only consider tasks that
3296 * started after the latest task in the previous pass. This
3297 * guarantees forward progress and that we don't miss any tasks.
3300 cgroup_iter_start(scan->cg, &it);
3301 while ((p = cgroup_iter_next(scan->cg, &it))) {
3303 * Only affect tasks that qualify per the caller's callback,
3304 * if he provided one
3306 if (scan->test_task && !scan->test_task(p, scan))
3309 * Only process tasks that started after the last task
3312 if (!started_after_time(p, &latest_time, latest_task))
3314 dropped = heap_insert(heap, p);
3315 if (dropped == NULL) {
3317 * The new task was inserted; the heap wasn't
3321 } else if (dropped != p) {
3323 * The new task was inserted, and pushed out a
3327 put_task_struct(dropped);
3330 * Else the new task was newer than anything already in
3331 * the heap and wasn't inserted
3334 cgroup_iter_end(scan->cg, &it);
3337 for (i = 0; i < heap->size; i++) {
3338 struct task_struct *q = heap->ptrs[i];
3340 latest_time = q->start_time;
3343 /* Process the task per the caller's callback */
3344 scan->process_task(q, scan);
3348 * If we had to process any tasks at all, scan again
3349 * in case some of them were in the middle of forking
3350 * children that didn't get processed.
3351 * Not the most efficient way to do it, but it avoids
3352 * having to take callback_mutex in the fork path
3356 if (heap == &tmp_heap)
3357 heap_free(&tmp_heap);
3361 static void cgroup_transfer_one_task(struct task_struct *task,
3362 struct cgroup_scanner *scan)
3364 struct cgroup *new_cgroup = scan->data;
3366 mutex_lock(&cgroup_mutex);
3367 cgroup_attach_task(new_cgroup, task, false);
3368 mutex_unlock(&cgroup_mutex);
3372 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3373 * @to: cgroup to which the tasks will be moved
3374 * @from: cgroup in which the tasks currently reside
3376 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3378 struct cgroup_scanner scan;
3381 scan.test_task = NULL; /* select all tasks in cgroup */
3382 scan.process_task = cgroup_transfer_one_task;
3386 return cgroup_scan_tasks(&scan);
3390 * Stuff for reading the 'tasks'/'procs' files.
3392 * Reading this file can return large amounts of data if a cgroup has
3393 * *lots* of attached tasks. So it may need several calls to read(),
3394 * but we cannot guarantee that the information we produce is correct
3395 * unless we produce it entirely atomically.
3399 /* which pidlist file are we talking about? */
3400 enum cgroup_filetype {
3406 * A pidlist is a list of pids that virtually represents the contents of one
3407 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3408 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3411 struct cgroup_pidlist {
3413 * used to find which pidlist is wanted. doesn't change as long as
3414 * this particular list stays in the list.
3416 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3419 /* how many elements the above list has */
3421 /* how many files are using the current array */
3423 /* each of these stored in a list by its cgroup */
3424 struct list_head links;
3425 /* pointer to the cgroup we belong to, for list removal purposes */
3426 struct cgroup *owner;
3427 /* protects the other fields */
3428 struct rw_semaphore mutex;
3432 * The following two functions "fix" the issue where there are more pids
3433 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3434 * TODO: replace with a kernel-wide solution to this problem
3436 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3437 static void *pidlist_allocate(int count)
3439 if (PIDLIST_TOO_LARGE(count))
3440 return vmalloc(count * sizeof(pid_t));
3442 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3444 static void pidlist_free(void *p)
3446 if (is_vmalloc_addr(p))
3453 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3454 * Returns the number of unique elements.
3456 static int pidlist_uniq(pid_t *list, int length)
3461 * we presume the 0th element is unique, so i starts at 1. trivial
3462 * edge cases first; no work needs to be done for either
3464 if (length == 0 || length == 1)
3466 /* src and dest walk down the list; dest counts unique elements */
3467 for (src = 1; src < length; src++) {
3468 /* find next unique element */
3469 while (list[src] == list[src-1]) {
3474 /* dest always points to where the next unique element goes */
3475 list[dest] = list[src];
3482 static int cmppid(const void *a, const void *b)
3484 return *(pid_t *)a - *(pid_t *)b;
3488 * find the appropriate pidlist for our purpose (given procs vs tasks)
3489 * returns with the lock on that pidlist already held, and takes care
3490 * of the use count, or returns NULL with no locks held if we're out of
3493 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3494 enum cgroup_filetype type)
3496 struct cgroup_pidlist *l;
3497 /* don't need task_nsproxy() if we're looking at ourself */
3498 struct pid_namespace *ns = task_active_pid_ns(current);
3501 * We can't drop the pidlist_mutex before taking the l->mutex in case
3502 * the last ref-holder is trying to remove l from the list at the same
3503 * time. Holding the pidlist_mutex precludes somebody taking whichever
3504 * list we find out from under us - compare release_pid_array().
3506 mutex_lock(&cgrp->pidlist_mutex);
3507 list_for_each_entry(l, &cgrp->pidlists, links) {
3508 if (l->key.type == type && l->key.ns == ns) {
3509 /* make sure l doesn't vanish out from under us */
3510 down_write(&l->mutex);
3511 mutex_unlock(&cgrp->pidlist_mutex);
3515 /* entry not found; create a new one */
3516 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3518 mutex_unlock(&cgrp->pidlist_mutex);
3521 init_rwsem(&l->mutex);
3522 down_write(&l->mutex);
3524 l->key.ns = get_pid_ns(ns);
3526 list_add(&l->links, &cgrp->pidlists);
3527 mutex_unlock(&cgrp->pidlist_mutex);
3532 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3534 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3535 struct cgroup_pidlist **lp)
3539 int pid, n = 0; /* used for populating the array */
3540 struct cgroup_iter it;
3541 struct task_struct *tsk;
3542 struct cgroup_pidlist *l;
3545 * If cgroup gets more users after we read count, we won't have
3546 * enough space - tough. This race is indistinguishable to the
3547 * caller from the case that the additional cgroup users didn't
3548 * show up until sometime later on.
3550 length = cgroup_task_count(cgrp);
3551 array = pidlist_allocate(length);
3554 /* now, populate the array */
3555 cgroup_iter_start(cgrp, &it);
3556 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3557 if (unlikely(n == length))
3559 /* get tgid or pid for procs or tasks file respectively */
3560 if (type == CGROUP_FILE_PROCS)
3561 pid = task_tgid_vnr(tsk);
3563 pid = task_pid_vnr(tsk);
3564 if (pid > 0) /* make sure to only use valid results */
3567 cgroup_iter_end(cgrp, &it);
3569 /* now sort & (if procs) strip out duplicates */
3570 sort(array, length, sizeof(pid_t), cmppid, NULL);
3571 if (type == CGROUP_FILE_PROCS)
3572 length = pidlist_uniq(array, length);
3573 l = cgroup_pidlist_find(cgrp, type);
3575 pidlist_free(array);
3578 /* store array, freeing old if necessary - lock already held */
3579 pidlist_free(l->list);
3583 up_write(&l->mutex);
3589 * cgroupstats_build - build and fill cgroupstats
3590 * @stats: cgroupstats to fill information into
3591 * @dentry: A dentry entry belonging to the cgroup for which stats have
3594 * Build and fill cgroupstats so that taskstats can export it to user
3597 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3600 struct cgroup *cgrp;
3601 struct cgroup_iter it;
3602 struct task_struct *tsk;
3605 * Validate dentry by checking the superblock operations,
3606 * and make sure it's a directory.
3608 if (dentry->d_sb->s_op != &cgroup_ops ||
3609 !S_ISDIR(dentry->d_inode->i_mode))
3613 cgrp = dentry->d_fsdata;
3615 cgroup_iter_start(cgrp, &it);
3616 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3617 switch (tsk->state) {
3619 stats->nr_running++;
3621 case TASK_INTERRUPTIBLE:
3622 stats->nr_sleeping++;
3624 case TASK_UNINTERRUPTIBLE:
3625 stats->nr_uninterruptible++;
3628 stats->nr_stopped++;
3631 if (delayacct_is_task_waiting_on_io(tsk))
3632 stats->nr_io_wait++;
3636 cgroup_iter_end(cgrp, &it);
3644 * seq_file methods for the tasks/procs files. The seq_file position is the
3645 * next pid to display; the seq_file iterator is a pointer to the pid
3646 * in the cgroup->l->list array.
3649 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3652 * Initially we receive a position value that corresponds to
3653 * one more than the last pid shown (or 0 on the first call or
3654 * after a seek to the start). Use a binary-search to find the
3655 * next pid to display, if any
3657 struct cgroup_pidlist *l = s->private;
3658 int index = 0, pid = *pos;
3661 down_read(&l->mutex);
3663 int end = l->length;
3665 while (index < end) {
3666 int mid = (index + end) / 2;
3667 if (l->list[mid] == pid) {
3670 } else if (l->list[mid] <= pid)
3676 /* If we're off the end of the array, we're done */
3677 if (index >= l->length)
3679 /* Update the abstract position to be the actual pid that we found */
3680 iter = l->list + index;
3685 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3687 struct cgroup_pidlist *l = s->private;
3691 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3693 struct cgroup_pidlist *l = s->private;
3695 pid_t *end = l->list + l->length;
3697 * Advance to the next pid in the array. If this goes off the
3709 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3711 return seq_printf(s, "%d\n", *(int *)v);
3715 * seq_operations functions for iterating on pidlists through seq_file -
3716 * independent of whether it's tasks or procs
3718 static const struct seq_operations cgroup_pidlist_seq_operations = {
3719 .start = cgroup_pidlist_start,
3720 .stop = cgroup_pidlist_stop,
3721 .next = cgroup_pidlist_next,
3722 .show = cgroup_pidlist_show,
3725 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3728 * the case where we're the last user of this particular pidlist will
3729 * have us remove it from the cgroup's list, which entails taking the
3730 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3731 * pidlist_mutex, we have to take pidlist_mutex first.
3733 mutex_lock(&l->owner->pidlist_mutex);
3734 down_write(&l->mutex);
3735 BUG_ON(!l->use_count);
3736 if (!--l->use_count) {
3737 /* we're the last user if refcount is 0; remove and free */
3738 list_del(&l->links);
3739 mutex_unlock(&l->owner->pidlist_mutex);
3740 pidlist_free(l->list);
3741 put_pid_ns(l->key.ns);
3742 up_write(&l->mutex);
3746 mutex_unlock(&l->owner->pidlist_mutex);
3747 up_write(&l->mutex);
3750 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3752 struct cgroup_pidlist *l;
3753 if (!(file->f_mode & FMODE_READ))
3756 * the seq_file will only be initialized if the file was opened for
3757 * reading; hence we check if it's not null only in that case.
3759 l = ((struct seq_file *)file->private_data)->private;
3760 cgroup_release_pid_array(l);
3761 return seq_release(inode, file);
3764 static const struct file_operations cgroup_pidlist_operations = {
3766 .llseek = seq_lseek,
3767 .write = cgroup_file_write,
3768 .release = cgroup_pidlist_release,
3772 * The following functions handle opens on a file that displays a pidlist
3773 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3776 /* helper function for the two below it */
3777 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3779 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3780 struct cgroup_pidlist *l;
3783 /* Nothing to do for write-only files */
3784 if (!(file->f_mode & FMODE_READ))
3787 /* have the array populated */
3788 retval = pidlist_array_load(cgrp, type, &l);
3791 /* configure file information */
3792 file->f_op = &cgroup_pidlist_operations;
3794 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3796 cgroup_release_pid_array(l);
3799 ((struct seq_file *)file->private_data)->private = l;
3802 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3804 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3806 static int cgroup_procs_open(struct inode *unused, struct file *file)
3808 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3811 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3814 return notify_on_release(cgrp);
3817 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3821 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3823 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3825 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3830 * Unregister event and free resources.
3832 * Gets called from workqueue.
3834 static void cgroup_event_remove(struct work_struct *work)
3836 struct cgroup_event *event = container_of(work, struct cgroup_event,
3838 struct cgroup *cgrp = event->cgrp;
3840 remove_wait_queue(event->wqh, &event->wait);
3842 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3844 /* Notify userspace the event is going away. */
3845 eventfd_signal(event->eventfd, 1);
3847 eventfd_ctx_put(event->eventfd);
3853 * Gets called on POLLHUP on eventfd when user closes it.
3855 * Called with wqh->lock held and interrupts disabled.
3857 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3858 int sync, void *key)
3860 struct cgroup_event *event = container_of(wait,
3861 struct cgroup_event, wait);
3862 struct cgroup *cgrp = event->cgrp;
3863 unsigned long flags = (unsigned long)key;
3865 if (flags & POLLHUP) {
3867 * If the event has been detached at cgroup removal, we
3868 * can simply return knowing the other side will cleanup
3871 * We can't race against event freeing since the other
3872 * side will require wqh->lock via remove_wait_queue(),
3875 spin_lock(&cgrp->event_list_lock);
3876 if (!list_empty(&event->list)) {
3877 list_del_init(&event->list);
3879 * We are in atomic context, but cgroup_event_remove()
3880 * may sleep, so we have to call it in workqueue.
3882 schedule_work(&event->remove);
3884 spin_unlock(&cgrp->event_list_lock);
3890 static void cgroup_event_ptable_queue_proc(struct file *file,
3891 wait_queue_head_t *wqh, poll_table *pt)
3893 struct cgroup_event *event = container_of(pt,
3894 struct cgroup_event, pt);
3897 add_wait_queue(wqh, &event->wait);
3901 * Parse input and register new cgroup event handler.
3903 * Input must be in format '<event_fd> <control_fd> <args>'.
3904 * Interpretation of args is defined by control file implementation.
3906 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3909 struct cgroup_event *event = NULL;
3910 struct cgroup *cgrp_cfile;
3911 unsigned int efd, cfd;
3912 struct file *efile = NULL;
3913 struct file *cfile = NULL;
3917 efd = simple_strtoul(buffer, &endp, 10);
3922 cfd = simple_strtoul(buffer, &endp, 10);
3923 if ((*endp != ' ') && (*endp != '\0'))
3927 event = kzalloc(sizeof(*event), GFP_KERNEL);
3931 INIT_LIST_HEAD(&event->list);
3932 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3933 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3934 INIT_WORK(&event->remove, cgroup_event_remove);
3936 efile = eventfd_fget(efd);
3937 if (IS_ERR(efile)) {
3938 ret = PTR_ERR(efile);
3942 event->eventfd = eventfd_ctx_fileget(efile);
3943 if (IS_ERR(event->eventfd)) {
3944 ret = PTR_ERR(event->eventfd);
3954 /* the process need read permission on control file */
3955 /* AV: shouldn't we check that it's been opened for read instead? */
3956 ret = inode_permission(file_inode(cfile), MAY_READ);
3960 event->cft = __file_cft(cfile);
3961 if (IS_ERR(event->cft)) {
3962 ret = PTR_ERR(event->cft);
3967 * The file to be monitored must be in the same cgroup as
3968 * cgroup.event_control is.
3970 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
3971 if (cgrp_cfile != cgrp) {
3976 if (!event->cft->register_event || !event->cft->unregister_event) {
3981 ret = event->cft->register_event(cgrp, event->cft,
3982 event->eventfd, buffer);
3986 efile->f_op->poll(efile, &event->pt);
3989 * Events should be removed after rmdir of cgroup directory, but before
3990 * destroying subsystem state objects. Let's take reference to cgroup
3991 * directory dentry to do that.
3995 spin_lock(&cgrp->event_list_lock);
3996 list_add(&event->list, &cgrp->event_list);
3997 spin_unlock(&cgrp->event_list_lock);
4008 if (event && event->eventfd && !IS_ERR(event->eventfd))
4009 eventfd_ctx_put(event->eventfd);
4011 if (!IS_ERR_OR_NULL(efile))
4019 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
4022 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4025 static int cgroup_clone_children_write(struct cgroup *cgrp,
4030 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4032 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4036 static struct cftype cgroup_base_files[] = {
4038 .name = "cgroup.procs",
4039 .open = cgroup_procs_open,
4040 .write_u64 = cgroup_procs_write,
4041 .release = cgroup_pidlist_release,
4042 .mode = S_IRUGO | S_IWUSR,
4045 .name = "cgroup.event_control",
4046 .write_string = cgroup_write_event_control,
4050 .name = "cgroup.clone_children",
4051 .flags = CFTYPE_INSANE,
4052 .read_u64 = cgroup_clone_children_read,
4053 .write_u64 = cgroup_clone_children_write,
4056 .name = "cgroup.sane_behavior",
4057 .flags = CFTYPE_ONLY_ON_ROOT,
4058 .read_seq_string = cgroup_sane_behavior_show,
4062 * Historical crazy stuff. These don't have "cgroup." prefix and
4063 * don't exist if sane_behavior. If you're depending on these, be
4064 * prepared to be burned.
4068 .flags = CFTYPE_INSANE, /* use "procs" instead */
4069 .open = cgroup_tasks_open,
4070 .write_u64 = cgroup_tasks_write,
4071 .release = cgroup_pidlist_release,
4072 .mode = S_IRUGO | S_IWUSR,
4075 .name = "notify_on_release",
4076 .flags = CFTYPE_INSANE,
4077 .read_u64 = cgroup_read_notify_on_release,
4078 .write_u64 = cgroup_write_notify_on_release,
4081 .name = "release_agent",
4082 .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT,
4083 .read_seq_string = cgroup_release_agent_show,
4084 .write_string = cgroup_release_agent_write,
4085 .max_write_len = PATH_MAX,
4091 * cgroup_populate_dir - selectively creation of files in a directory
4092 * @cgrp: target cgroup
4093 * @base_files: true if the base files should be added
4094 * @subsys_mask: mask of the subsystem ids whose files should be added
4096 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
4097 unsigned long subsys_mask)
4100 struct cgroup_subsys *ss;
4103 err = cgroup_addrm_files(cgrp, NULL, cgroup_base_files, true);
4108 /* process cftsets of each subsystem */
4109 for_each_subsys(cgrp->root, ss) {
4110 struct cftype_set *set;
4111 if (!test_bit(ss->subsys_id, &subsys_mask))
4114 list_for_each_entry(set, &ss->cftsets, node)
4115 cgroup_addrm_files(cgrp, ss, set->cfts, true);
4118 /* This cgroup is ready now */
4119 for_each_subsys(cgrp->root, ss) {
4120 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4122 * Update id->css pointer and make this css visible from
4123 * CSS ID functions. This pointer will be dereferened
4124 * from RCU-read-side without locks.
4127 rcu_assign_pointer(css->id->css, css);
4133 static void css_dput_fn(struct work_struct *work)
4135 struct cgroup_subsys_state *css =
4136 container_of(work, struct cgroup_subsys_state, dput_work);
4137 struct dentry *dentry = css->cgroup->dentry;
4138 struct super_block *sb = dentry->d_sb;
4140 atomic_inc(&sb->s_active);
4142 deactivate_super(sb);
4145 static void init_cgroup_css(struct cgroup_subsys_state *css,
4146 struct cgroup_subsys *ss,
4147 struct cgroup *cgrp)
4150 atomic_set(&css->refcnt, 1);
4153 if (cgrp == dummytop)
4154 css->flags |= CSS_ROOT;
4155 BUG_ON(cgrp->subsys[ss->subsys_id]);
4156 cgrp->subsys[ss->subsys_id] = css;
4159 * css holds an extra ref to @cgrp->dentry which is put on the last
4160 * css_put(). dput() requires process context, which css_put() may
4161 * be called without. @css->dput_work will be used to invoke
4162 * dput() asynchronously from css_put().
4164 INIT_WORK(&css->dput_work, css_dput_fn);
4167 /* invoke ->post_create() on a new CSS and mark it online if successful */
4168 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4172 lockdep_assert_held(&cgroup_mutex);
4175 ret = ss->css_online(cgrp);
4177 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4181 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4182 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4183 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4185 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4187 lockdep_assert_held(&cgroup_mutex);
4189 if (!(css->flags & CSS_ONLINE))
4192 if (ss->css_offline)
4193 ss->css_offline(cgrp);
4195 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4199 * cgroup_create - create a cgroup
4200 * @parent: cgroup that will be parent of the new cgroup
4201 * @dentry: dentry of the new cgroup
4202 * @mode: mode to set on new inode
4204 * Must be called with the mutex on the parent inode held
4206 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4209 static atomic64_t serial_nr_cursor = ATOMIC64_INIT(0);
4210 struct cgroup *cgrp;
4211 struct cgroup_name *name;
4212 struct cgroupfs_root *root = parent->root;
4214 struct cgroup_subsys *ss;
4215 struct super_block *sb = root->sb;
4217 /* allocate the cgroup and its ID, 0 is reserved for the root */
4218 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4222 name = cgroup_alloc_name(dentry);
4225 rcu_assign_pointer(cgrp->name, name);
4227 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4232 * Only live parents can have children. Note that the liveliness
4233 * check isn't strictly necessary because cgroup_mkdir() and
4234 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4235 * anyway so that locking is contained inside cgroup proper and we
4236 * don't get nasty surprises if we ever grow another caller.
4238 if (!cgroup_lock_live_group(parent)) {
4243 /* Grab a reference on the superblock so the hierarchy doesn't
4244 * get deleted on unmount if there are child cgroups. This
4245 * can be done outside cgroup_mutex, since the sb can't
4246 * disappear while someone has an open control file on the
4248 atomic_inc(&sb->s_active);
4250 init_cgroup_housekeeping(cgrp);
4252 dentry->d_fsdata = cgrp;
4253 cgrp->dentry = dentry;
4255 cgrp->parent = parent;
4256 cgrp->root = parent->root;
4258 if (notify_on_release(parent))
4259 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4261 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4262 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4264 for_each_subsys(root, ss) {
4265 struct cgroup_subsys_state *css;
4267 css = ss->css_alloc(cgrp);
4272 init_cgroup_css(css, ss, cgrp);
4274 err = alloc_css_id(ss, parent, cgrp);
4281 * Create directory. cgroup_create_file() returns with the new
4282 * directory locked on success so that it can be populated without
4283 * dropping cgroup_mutex.
4285 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4288 lockdep_assert_held(&dentry->d_inode->i_mutex);
4291 * Assign a monotonically increasing serial number. With the list
4292 * appending below, it guarantees that sibling cgroups are always
4293 * sorted in the ascending serial number order on the parent's
4296 cgrp->serial_nr = atomic64_inc_return(&serial_nr_cursor);
4298 /* allocation complete, commit to creation */
4299 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4300 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4301 root->number_of_cgroups++;
4303 /* each css holds a ref to the cgroup's dentry */
4304 for_each_subsys(root, ss)
4307 /* hold a ref to the parent's dentry */
4308 dget(parent->dentry);
4310 /* creation succeeded, notify subsystems */
4311 for_each_subsys(root, ss) {
4312 err = online_css(ss, cgrp);
4316 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4318 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",
4319 current->comm, current->pid, ss->name);
4320 if (!strcmp(ss->name, "memory"))
4321 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4322 ss->warned_broken_hierarchy = true;
4326 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4330 mutex_unlock(&cgroup_mutex);
4331 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4336 for_each_subsys(root, ss) {
4337 if (cgrp->subsys[ss->subsys_id])
4340 mutex_unlock(&cgroup_mutex);
4341 /* Release the reference count that we took on the superblock */
4342 deactivate_super(sb);
4344 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4346 kfree(rcu_dereference_raw(cgrp->name));
4352 cgroup_destroy_locked(cgrp);
4353 mutex_unlock(&cgroup_mutex);
4354 mutex_unlock(&dentry->d_inode->i_mutex);
4358 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4360 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4362 /* the vfs holds inode->i_mutex already */
4363 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4366 static int cgroup_destroy_locked(struct cgroup *cgrp)
4367 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4369 struct dentry *d = cgrp->dentry;
4370 struct cgroup *parent = cgrp->parent;
4371 struct cgroup_event *event, *tmp;
4372 struct cgroup_subsys *ss;
4374 lockdep_assert_held(&d->d_inode->i_mutex);
4375 lockdep_assert_held(&cgroup_mutex);
4377 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children))
4381 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4382 * removed. This makes future css_tryget() and child creation
4383 * attempts fail thus maintaining the removal conditions verified
4386 * Note that CGRP_REMVOED clearing is depended upon by
4387 * cgroup_next_sibling() to resume iteration after dropping RCU
4388 * read lock. See cgroup_next_sibling() for details.
4390 for_each_subsys(cgrp->root, ss) {
4391 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4393 WARN_ON(atomic_read(&css->refcnt) < 0);
4394 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4396 set_bit(CGRP_REMOVED, &cgrp->flags);
4398 /* tell subsystems to initate destruction */
4399 for_each_subsys(cgrp->root, ss)
4400 offline_css(ss, cgrp);
4403 * Put all the base refs. Each css holds an extra reference to the
4404 * cgroup's dentry and cgroup removal proceeds regardless of css
4405 * refs. On the last put of each css, whenever that may be, the
4406 * extra dentry ref is put so that dentry destruction happens only
4407 * after all css's are released.
4409 for_each_subsys(cgrp->root, ss)
4410 css_put(cgrp->subsys[ss->subsys_id]);
4412 raw_spin_lock(&release_list_lock);
4413 if (!list_empty(&cgrp->release_list))
4414 list_del_init(&cgrp->release_list);
4415 raw_spin_unlock(&release_list_lock);
4417 /* delete this cgroup from parent->children */
4418 list_del_rcu(&cgrp->sibling);
4419 list_del_init(&cgrp->allcg_node);
4422 cgroup_d_remove_dir(d);
4425 set_bit(CGRP_RELEASABLE, &parent->flags);
4426 check_for_release(parent);
4429 * Unregister events and notify userspace.
4430 * Notify userspace about cgroup removing only after rmdir of cgroup
4431 * directory to avoid race between userspace and kernelspace.
4433 spin_lock(&cgrp->event_list_lock);
4434 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4435 list_del_init(&event->list);
4436 schedule_work(&event->remove);
4438 spin_unlock(&cgrp->event_list_lock);
4443 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4447 mutex_lock(&cgroup_mutex);
4448 ret = cgroup_destroy_locked(dentry->d_fsdata);
4449 mutex_unlock(&cgroup_mutex);
4454 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4456 INIT_LIST_HEAD(&ss->cftsets);
4459 * base_cftset is embedded in subsys itself, no need to worry about
4462 if (ss->base_cftypes) {
4463 ss->base_cftset.cfts = ss->base_cftypes;
4464 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4468 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4470 struct cgroup_subsys_state *css;
4472 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4474 mutex_lock(&cgroup_mutex);
4476 /* init base cftset */
4477 cgroup_init_cftsets(ss);
4479 /* Create the top cgroup state for this subsystem */
4480 list_add(&ss->sibling, &rootnode.subsys_list);
4481 ss->root = &rootnode;
4482 css = ss->css_alloc(dummytop);
4483 /* We don't handle early failures gracefully */
4484 BUG_ON(IS_ERR(css));
4485 init_cgroup_css(css, ss, dummytop);
4487 /* Update the init_css_set to contain a subsys
4488 * pointer to this state - since the subsystem is
4489 * newly registered, all tasks and hence the
4490 * init_css_set is in the subsystem's top cgroup. */
4491 init_css_set.subsys[ss->subsys_id] = css;
4493 need_forkexit_callback |= ss->fork || ss->exit;
4495 /* At system boot, before all subsystems have been
4496 * registered, no tasks have been forked, so we don't
4497 * need to invoke fork callbacks here. */
4498 BUG_ON(!list_empty(&init_task.tasks));
4500 BUG_ON(online_css(ss, dummytop));
4502 mutex_unlock(&cgroup_mutex);
4504 /* this function shouldn't be used with modular subsystems, since they
4505 * need to register a subsys_id, among other things */
4510 * cgroup_load_subsys: load and register a modular subsystem at runtime
4511 * @ss: the subsystem to load
4513 * This function should be called in a modular subsystem's initcall. If the
4514 * subsystem is built as a module, it will be assigned a new subsys_id and set
4515 * up for use. If the subsystem is built-in anyway, work is delegated to the
4516 * simpler cgroup_init_subsys.
4518 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4520 struct cgroup_subsys_state *css;
4522 struct hlist_node *tmp;
4523 struct css_set *cset;
4526 /* check name and function validity */
4527 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4528 ss->css_alloc == NULL || ss->css_free == NULL)
4532 * we don't support callbacks in modular subsystems. this check is
4533 * before the ss->module check for consistency; a subsystem that could
4534 * be a module should still have no callbacks even if the user isn't
4535 * compiling it as one.
4537 if (ss->fork || ss->exit)
4541 * an optionally modular subsystem is built-in: we want to do nothing,
4542 * since cgroup_init_subsys will have already taken care of it.
4544 if (ss->module == NULL) {
4545 /* a sanity check */
4546 BUG_ON(subsys[ss->subsys_id] != ss);
4550 /* init base cftset */
4551 cgroup_init_cftsets(ss);
4553 mutex_lock(&cgroup_mutex);
4554 subsys[ss->subsys_id] = ss;
4557 * no ss->css_alloc seems to need anything important in the ss
4558 * struct, so this can happen first (i.e. before the rootnode
4561 css = ss->css_alloc(dummytop);
4563 /* failure case - need to deassign the subsys[] slot. */
4564 subsys[ss->subsys_id] = NULL;
4565 mutex_unlock(&cgroup_mutex);
4566 return PTR_ERR(css);
4569 list_add(&ss->sibling, &rootnode.subsys_list);
4570 ss->root = &rootnode;
4572 /* our new subsystem will be attached to the dummy hierarchy. */
4573 init_cgroup_css(css, ss, dummytop);
4574 /* init_idr must be after init_cgroup_css because it sets css->id. */
4576 ret = cgroup_init_idr(ss, css);
4582 * Now we need to entangle the css into the existing css_sets. unlike
4583 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4584 * will need a new pointer to it; done by iterating the css_set_table.
4585 * furthermore, modifying the existing css_sets will corrupt the hash
4586 * table state, so each changed css_set will need its hash recomputed.
4587 * this is all done under the css_set_lock.
4589 write_lock(&css_set_lock);
4590 hash_for_each_safe(css_set_table, i, tmp, cset, hlist) {
4591 /* skip entries that we already rehashed */
4592 if (cset->subsys[ss->subsys_id])
4594 /* remove existing entry */
4595 hash_del(&cset->hlist);
4597 cset->subsys[ss->subsys_id] = css;
4598 /* recompute hash and restore entry */
4599 key = css_set_hash(cset->subsys);
4600 hash_add(css_set_table, &cset->hlist, key);
4602 write_unlock(&css_set_lock);
4604 ret = online_css(ss, dummytop);
4609 mutex_unlock(&cgroup_mutex);
4613 mutex_unlock(&cgroup_mutex);
4614 /* @ss can't be mounted here as try_module_get() would fail */
4615 cgroup_unload_subsys(ss);
4618 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4621 * cgroup_unload_subsys: unload a modular subsystem
4622 * @ss: the subsystem to unload
4624 * This function should be called in a modular subsystem's exitcall. When this
4625 * function is invoked, the refcount on the subsystem's module will be 0, so
4626 * the subsystem will not be attached to any hierarchy.
4628 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4630 struct cgrp_cset_link *link;
4632 BUG_ON(ss->module == NULL);
4635 * we shouldn't be called if the subsystem is in use, and the use of
4636 * try_module_get in parse_cgroupfs_options should ensure that it
4637 * doesn't start being used while we're killing it off.
4639 BUG_ON(ss->root != &rootnode);
4641 mutex_lock(&cgroup_mutex);
4643 offline_css(ss, dummytop);
4646 idr_destroy(&ss->idr);
4648 /* deassign the subsys_id */
4649 subsys[ss->subsys_id] = NULL;
4651 /* remove subsystem from rootnode's list of subsystems */
4652 list_del_init(&ss->sibling);
4655 * disentangle the css from all css_sets attached to the dummytop. as
4656 * in loading, we need to pay our respects to the hashtable gods.
4658 write_lock(&css_set_lock);
4659 list_for_each_entry(link, &dummytop->cset_links, cset_link) {
4660 struct css_set *cset = link->cset;
4663 hash_del(&cset->hlist);
4664 cset->subsys[ss->subsys_id] = NULL;
4665 key = css_set_hash(cset->subsys);
4666 hash_add(css_set_table, &cset->hlist, key);
4668 write_unlock(&css_set_lock);
4671 * remove subsystem's css from the dummytop and free it - need to
4672 * free before marking as null because ss->css_free needs the
4673 * cgrp->subsys pointer to find their state. note that this also
4674 * takes care of freeing the css_id.
4676 ss->css_free(dummytop);
4677 dummytop->subsys[ss->subsys_id] = NULL;
4679 mutex_unlock(&cgroup_mutex);
4681 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4684 * cgroup_init_early - cgroup initialization at system boot
4686 * Initialize cgroups at system boot, and initialize any
4687 * subsystems that request early init.
4689 int __init cgroup_init_early(void)
4692 atomic_set(&init_css_set.refcount, 1);
4693 INIT_LIST_HEAD(&init_css_set.cgrp_links);
4694 INIT_LIST_HEAD(&init_css_set.tasks);
4695 INIT_HLIST_NODE(&init_css_set.hlist);
4697 init_cgroup_root(&rootnode);
4699 init_task.cgroups = &init_css_set;
4701 init_cgrp_cset_link.cset = &init_css_set;
4702 init_cgrp_cset_link.cgrp = dummytop;
4703 list_add(&init_cgrp_cset_link.cset_link, &rootnode.top_cgroup.cset_links);
4704 list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links);
4706 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4707 struct cgroup_subsys *ss = subsys[i];
4709 /* at bootup time, we don't worry about modular subsystems */
4710 if (!ss || ss->module)
4714 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4715 BUG_ON(!ss->css_alloc);
4716 BUG_ON(!ss->css_free);
4717 if (ss->subsys_id != i) {
4718 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4719 ss->name, ss->subsys_id);
4724 cgroup_init_subsys(ss);
4730 * cgroup_init - cgroup initialization
4732 * Register cgroup filesystem and /proc file, and initialize
4733 * any subsystems that didn't request early init.
4735 int __init cgroup_init(void)
4741 err = bdi_init(&cgroup_backing_dev_info);
4745 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4746 struct cgroup_subsys *ss = subsys[i];
4748 /* at bootup time, we don't worry about modular subsystems */
4749 if (!ss || ss->module)
4751 if (!ss->early_init)
4752 cgroup_init_subsys(ss);
4754 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4757 /* Add init_css_set to the hash table */
4758 key = css_set_hash(init_css_set.subsys);
4759 hash_add(css_set_table, &init_css_set.hlist, key);
4761 /* allocate id for the dummy hierarchy */
4762 mutex_lock(&cgroup_mutex);
4763 mutex_lock(&cgroup_root_mutex);
4765 BUG_ON(cgroup_init_root_id(&rootnode));
4767 mutex_unlock(&cgroup_root_mutex);
4768 mutex_unlock(&cgroup_mutex);
4770 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4776 err = register_filesystem(&cgroup_fs_type);
4778 kobject_put(cgroup_kobj);
4782 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4786 bdi_destroy(&cgroup_backing_dev_info);
4792 * proc_cgroup_show()
4793 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4794 * - Used for /proc/<pid>/cgroup.
4795 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4796 * doesn't really matter if tsk->cgroup changes after we read it,
4797 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4798 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4799 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4800 * cgroup to top_cgroup.
4803 /* TODO: Use a proper seq_file iterator */
4804 int proc_cgroup_show(struct seq_file *m, void *v)
4807 struct task_struct *tsk;
4810 struct cgroupfs_root *root;
4813 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4819 tsk = get_pid_task(pid, PIDTYPE_PID);
4825 mutex_lock(&cgroup_mutex);
4827 for_each_active_root(root) {
4828 struct cgroup_subsys *ss;
4829 struct cgroup *cgrp;
4832 seq_printf(m, "%d:", root->hierarchy_id);
4833 for_each_subsys(root, ss)
4834 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4835 if (strlen(root->name))
4836 seq_printf(m, "%sname=%s", count ? "," : "",
4839 cgrp = task_cgroup_from_root(tsk, root);
4840 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4848 mutex_unlock(&cgroup_mutex);
4849 put_task_struct(tsk);
4856 /* Display information about each subsystem and each hierarchy */
4857 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4861 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4863 * ideally we don't want subsystems moving around while we do this.
4864 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4865 * subsys/hierarchy state.
4867 mutex_lock(&cgroup_mutex);
4868 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4869 struct cgroup_subsys *ss = subsys[i];
4872 seq_printf(m, "%s\t%d\t%d\t%d\n",
4873 ss->name, ss->root->hierarchy_id,
4874 ss->root->number_of_cgroups, !ss->disabled);
4876 mutex_unlock(&cgroup_mutex);
4880 static int cgroupstats_open(struct inode *inode, struct file *file)
4882 return single_open(file, proc_cgroupstats_show, NULL);
4885 static const struct file_operations proc_cgroupstats_operations = {
4886 .open = cgroupstats_open,
4888 .llseek = seq_lseek,
4889 .release = single_release,
4893 * cgroup_fork - attach newly forked task to its parents cgroup.
4894 * @child: pointer to task_struct of forking parent process.
4896 * Description: A task inherits its parent's cgroup at fork().
4898 * A pointer to the shared css_set was automatically copied in
4899 * fork.c by dup_task_struct(). However, we ignore that copy, since
4900 * it was not made under the protection of RCU or cgroup_mutex, so
4901 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4902 * have already changed current->cgroups, allowing the previously
4903 * referenced cgroup group to be removed and freed.
4905 * At the point that cgroup_fork() is called, 'current' is the parent
4906 * task, and the passed argument 'child' points to the child task.
4908 void cgroup_fork(struct task_struct *child)
4911 child->cgroups = current->cgroups;
4912 get_css_set(child->cgroups);
4913 task_unlock(current);
4914 INIT_LIST_HEAD(&child->cg_list);
4918 * cgroup_post_fork - called on a new task after adding it to the task list
4919 * @child: the task in question
4921 * Adds the task to the list running through its css_set if necessary and
4922 * call the subsystem fork() callbacks. Has to be after the task is
4923 * visible on the task list in case we race with the first call to
4924 * cgroup_iter_start() - to guarantee that the new task ends up on its
4927 void cgroup_post_fork(struct task_struct *child)
4932 * use_task_css_set_links is set to 1 before we walk the tasklist
4933 * under the tasklist_lock and we read it here after we added the child
4934 * to the tasklist under the tasklist_lock as well. If the child wasn't
4935 * yet in the tasklist when we walked through it from
4936 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4937 * should be visible now due to the paired locking and barriers implied
4938 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4939 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4942 if (use_task_css_set_links) {
4943 write_lock(&css_set_lock);
4945 if (list_empty(&child->cg_list))
4946 list_add(&child->cg_list, &child->cgroups->tasks);
4948 write_unlock(&css_set_lock);
4952 * Call ss->fork(). This must happen after @child is linked on
4953 * css_set; otherwise, @child might change state between ->fork()
4954 * and addition to css_set.
4956 if (need_forkexit_callback) {
4958 * fork/exit callbacks are supported only for builtin
4959 * subsystems, and the builtin section of the subsys
4960 * array is immutable, so we don't need to lock the
4961 * subsys array here. On the other hand, modular section
4962 * of the array can be freed at module unload, so we
4965 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4966 struct cgroup_subsys *ss = subsys[i];
4975 * cgroup_exit - detach cgroup from exiting task
4976 * @tsk: pointer to task_struct of exiting process
4977 * @run_callback: run exit callbacks?
4979 * Description: Detach cgroup from @tsk and release it.
4981 * Note that cgroups marked notify_on_release force every task in
4982 * them to take the global cgroup_mutex mutex when exiting.
4983 * This could impact scaling on very large systems. Be reluctant to
4984 * use notify_on_release cgroups where very high task exit scaling
4985 * is required on large systems.
4987 * the_top_cgroup_hack:
4989 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4991 * We call cgroup_exit() while the task is still competent to
4992 * handle notify_on_release(), then leave the task attached to the
4993 * root cgroup in each hierarchy for the remainder of its exit.
4995 * To do this properly, we would increment the reference count on
4996 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4997 * code we would add a second cgroup function call, to drop that
4998 * reference. This would just create an unnecessary hot spot on
4999 * the top_cgroup reference count, to no avail.
5001 * Normally, holding a reference to a cgroup without bumping its
5002 * count is unsafe. The cgroup could go away, or someone could
5003 * attach us to a different cgroup, decrementing the count on
5004 * the first cgroup that we never incremented. But in this case,
5005 * top_cgroup isn't going away, and either task has PF_EXITING set,
5006 * which wards off any cgroup_attach_task() attempts, or task is a failed
5007 * fork, never visible to cgroup_attach_task.
5009 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
5011 struct css_set *cset;
5015 * Unlink from the css_set task list if necessary.
5016 * Optimistically check cg_list before taking
5019 if (!list_empty(&tsk->cg_list)) {
5020 write_lock(&css_set_lock);
5021 if (!list_empty(&tsk->cg_list))
5022 list_del_init(&tsk->cg_list);
5023 write_unlock(&css_set_lock);
5026 /* Reassign the task to the init_css_set. */
5028 cset = tsk->cgroups;
5029 tsk->cgroups = &init_css_set;
5031 if (run_callbacks && need_forkexit_callback) {
5033 * fork/exit callbacks are supported only for builtin
5034 * subsystems, see cgroup_post_fork() for details.
5036 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
5037 struct cgroup_subsys *ss = subsys[i];
5040 struct cgroup *old_cgrp =
5041 rcu_dereference_raw(cset->subsys[i])->cgroup;
5042 struct cgroup *cgrp = task_cgroup(tsk, i);
5043 ss->exit(cgrp, old_cgrp, tsk);
5049 put_css_set_taskexit(cset);
5052 static void check_for_release(struct cgroup *cgrp)
5054 /* All of these checks rely on RCU to keep the cgroup
5055 * structure alive */
5056 if (cgroup_is_releasable(cgrp) &&
5057 !atomic_read(&cgrp->count) && list_empty(&cgrp->children)) {
5059 * Control Group is currently removeable. If it's not
5060 * already queued for a userspace notification, queue
5063 int need_schedule_work = 0;
5065 raw_spin_lock(&release_list_lock);
5066 if (!cgroup_is_removed(cgrp) &&
5067 list_empty(&cgrp->release_list)) {
5068 list_add(&cgrp->release_list, &release_list);
5069 need_schedule_work = 1;
5071 raw_spin_unlock(&release_list_lock);
5072 if (need_schedule_work)
5073 schedule_work(&release_agent_work);
5077 /* Caller must verify that the css is not for root cgroup */
5078 bool __css_tryget(struct cgroup_subsys_state *css)
5083 v = css_refcnt(css);
5084 t = atomic_cmpxchg(&css->refcnt, v, v + 1);
5092 EXPORT_SYMBOL_GPL(__css_tryget);
5094 /* Caller must verify that the css is not for root cgroup */
5095 void __css_put(struct cgroup_subsys_state *css)
5099 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
5101 schedule_work(&css->dput_work);
5103 EXPORT_SYMBOL_GPL(__css_put);
5106 * Notify userspace when a cgroup is released, by running the
5107 * configured release agent with the name of the cgroup (path
5108 * relative to the root of cgroup file system) as the argument.
5110 * Most likely, this user command will try to rmdir this cgroup.
5112 * This races with the possibility that some other task will be
5113 * attached to this cgroup before it is removed, or that some other
5114 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5115 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5116 * unused, and this cgroup will be reprieved from its death sentence,
5117 * to continue to serve a useful existence. Next time it's released,
5118 * we will get notified again, if it still has 'notify_on_release' set.
5120 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5121 * means only wait until the task is successfully execve()'d. The
5122 * separate release agent task is forked by call_usermodehelper(),
5123 * then control in this thread returns here, without waiting for the
5124 * release agent task. We don't bother to wait because the caller of
5125 * this routine has no use for the exit status of the release agent
5126 * task, so no sense holding our caller up for that.
5128 static void cgroup_release_agent(struct work_struct *work)
5130 BUG_ON(work != &release_agent_work);
5131 mutex_lock(&cgroup_mutex);
5132 raw_spin_lock(&release_list_lock);
5133 while (!list_empty(&release_list)) {
5134 char *argv[3], *envp[3];
5136 char *pathbuf = NULL, *agentbuf = NULL;
5137 struct cgroup *cgrp = list_entry(release_list.next,
5140 list_del_init(&cgrp->release_list);
5141 raw_spin_unlock(&release_list_lock);
5142 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5145 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5147 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5152 argv[i++] = agentbuf;
5153 argv[i++] = pathbuf;
5157 /* minimal command environment */
5158 envp[i++] = "HOME=/";
5159 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5162 /* Drop the lock while we invoke the usermode helper,
5163 * since the exec could involve hitting disk and hence
5164 * be a slow process */
5165 mutex_unlock(&cgroup_mutex);
5166 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5167 mutex_lock(&cgroup_mutex);
5171 raw_spin_lock(&release_list_lock);
5173 raw_spin_unlock(&release_list_lock);
5174 mutex_unlock(&cgroup_mutex);
5177 static int __init cgroup_disable(char *str)
5182 while ((token = strsep(&str, ",")) != NULL) {
5185 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5186 struct cgroup_subsys *ss = subsys[i];
5189 * cgroup_disable, being at boot time, can't
5190 * know about module subsystems, so we don't
5193 if (!ss || ss->module)
5196 if (!strcmp(token, ss->name)) {
5198 printk(KERN_INFO "Disabling %s control group"
5199 " subsystem\n", ss->name);
5206 __setup("cgroup_disable=", cgroup_disable);
5209 * Functons for CSS ID.
5213 *To get ID other than 0, this should be called when !cgroup_is_removed().
5215 unsigned short css_id(struct cgroup_subsys_state *css)
5217 struct css_id *cssid;
5220 * This css_id() can return correct value when somone has refcnt
5221 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5222 * it's unchanged until freed.
5224 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5230 EXPORT_SYMBOL_GPL(css_id);
5233 * css_is_ancestor - test "root" css is an ancestor of "child"
5234 * @child: the css to be tested.
5235 * @root: the css supporsed to be an ancestor of the child.
5237 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5238 * this function reads css->id, the caller must hold rcu_read_lock().
5239 * But, considering usual usage, the csses should be valid objects after test.
5240 * Assuming that the caller will do some action to the child if this returns
5241 * returns true, the caller must take "child";s reference count.
5242 * If "child" is valid object and this returns true, "root" is valid, too.
5245 bool css_is_ancestor(struct cgroup_subsys_state *child,
5246 const struct cgroup_subsys_state *root)
5248 struct css_id *child_id;
5249 struct css_id *root_id;
5251 child_id = rcu_dereference(child->id);
5254 root_id = rcu_dereference(root->id);
5257 if (child_id->depth < root_id->depth)
5259 if (child_id->stack[root_id->depth] != root_id->id)
5264 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5266 struct css_id *id = css->id;
5267 /* When this is called before css_id initialization, id can be NULL */
5271 BUG_ON(!ss->use_id);
5273 rcu_assign_pointer(id->css, NULL);
5274 rcu_assign_pointer(css->id, NULL);
5275 spin_lock(&ss->id_lock);
5276 idr_remove(&ss->idr, id->id);
5277 spin_unlock(&ss->id_lock);
5278 kfree_rcu(id, rcu_head);
5280 EXPORT_SYMBOL_GPL(free_css_id);
5283 * This is called by init or create(). Then, calls to this function are
5284 * always serialized (By cgroup_mutex() at create()).
5287 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5289 struct css_id *newid;
5292 BUG_ON(!ss->use_id);
5294 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5295 newid = kzalloc(size, GFP_KERNEL);
5297 return ERR_PTR(-ENOMEM);
5299 idr_preload(GFP_KERNEL);
5300 spin_lock(&ss->id_lock);
5301 /* Don't use 0. allocates an ID of 1-65535 */
5302 ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5303 spin_unlock(&ss->id_lock);
5306 /* Returns error when there are no free spaces for new ID.*/
5311 newid->depth = depth;
5315 return ERR_PTR(ret);
5319 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5320 struct cgroup_subsys_state *rootcss)
5322 struct css_id *newid;
5324 spin_lock_init(&ss->id_lock);
5327 newid = get_new_cssid(ss, 0);
5329 return PTR_ERR(newid);
5331 newid->stack[0] = newid->id;
5332 newid->css = rootcss;
5333 rootcss->id = newid;
5337 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5338 struct cgroup *child)
5340 int subsys_id, i, depth = 0;
5341 struct cgroup_subsys_state *parent_css, *child_css;
5342 struct css_id *child_id, *parent_id;
5344 subsys_id = ss->subsys_id;
5345 parent_css = parent->subsys[subsys_id];
5346 child_css = child->subsys[subsys_id];
5347 parent_id = parent_css->id;
5348 depth = parent_id->depth + 1;
5350 child_id = get_new_cssid(ss, depth);
5351 if (IS_ERR(child_id))
5352 return PTR_ERR(child_id);
5354 for (i = 0; i < depth; i++)
5355 child_id->stack[i] = parent_id->stack[i];
5356 child_id->stack[depth] = child_id->id;
5358 * child_id->css pointer will be set after this cgroup is available
5359 * see cgroup_populate_dir()
5361 rcu_assign_pointer(child_css->id, child_id);
5367 * css_lookup - lookup css by id
5368 * @ss: cgroup subsys to be looked into.
5371 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5372 * NULL if not. Should be called under rcu_read_lock()
5374 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5376 struct css_id *cssid = NULL;
5378 BUG_ON(!ss->use_id);
5379 cssid = idr_find(&ss->idr, id);
5381 if (unlikely(!cssid))
5384 return rcu_dereference(cssid->css);
5386 EXPORT_SYMBOL_GPL(css_lookup);
5389 * get corresponding css from file open on cgroupfs directory
5391 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5393 struct cgroup *cgrp;
5394 struct inode *inode;
5395 struct cgroup_subsys_state *css;
5397 inode = file_inode(f);
5398 /* check in cgroup filesystem dir */
5399 if (inode->i_op != &cgroup_dir_inode_operations)
5400 return ERR_PTR(-EBADF);
5402 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5403 return ERR_PTR(-EINVAL);
5406 cgrp = __d_cgrp(f->f_dentry);
5407 css = cgrp->subsys[id];
5408 return css ? css : ERR_PTR(-ENOENT);
5411 #ifdef CONFIG_CGROUP_DEBUG
5412 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5414 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5417 return ERR_PTR(-ENOMEM);
5422 static void debug_css_free(struct cgroup *cont)
5424 kfree(cont->subsys[debug_subsys_id]);
5427 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5429 return atomic_read(&cont->count);
5432 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5434 return cgroup_task_count(cont);
5437 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5439 return (u64)(unsigned long)current->cgroups;
5442 static u64 current_css_set_refcount_read(struct cgroup *cont,
5448 count = atomic_read(¤t->cgroups->refcount);
5453 static int current_css_set_cg_links_read(struct cgroup *cont,
5455 struct seq_file *seq)
5457 struct cgrp_cset_link *link;
5458 struct css_set *cset;
5460 read_lock(&css_set_lock);
5462 cset = rcu_dereference(current->cgroups);
5463 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
5464 struct cgroup *c = link->cgrp;
5468 name = c->dentry->d_name.name;
5471 seq_printf(seq, "Root %d group %s\n",
5472 c->root->hierarchy_id, name);
5475 read_unlock(&css_set_lock);
5479 #define MAX_TASKS_SHOWN_PER_CSS 25
5480 static int cgroup_css_links_read(struct cgroup *cont,
5482 struct seq_file *seq)
5484 struct cgrp_cset_link *link;
5486 read_lock(&css_set_lock);
5487 list_for_each_entry(link, &cont->cset_links, cset_link) {
5488 struct css_set *cset = link->cset;
5489 struct task_struct *task;
5491 seq_printf(seq, "css_set %p\n", cset);
5492 list_for_each_entry(task, &cset->tasks, cg_list) {
5493 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5494 seq_puts(seq, " ...\n");
5497 seq_printf(seq, " task %d\n",
5498 task_pid_vnr(task));
5502 read_unlock(&css_set_lock);
5506 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5508 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5511 static struct cftype debug_files[] = {
5513 .name = "cgroup_refcount",
5514 .read_u64 = cgroup_refcount_read,
5517 .name = "taskcount",
5518 .read_u64 = debug_taskcount_read,
5522 .name = "current_css_set",
5523 .read_u64 = current_css_set_read,
5527 .name = "current_css_set_refcount",
5528 .read_u64 = current_css_set_refcount_read,
5532 .name = "current_css_set_cg_links",
5533 .read_seq_string = current_css_set_cg_links_read,
5537 .name = "cgroup_css_links",
5538 .read_seq_string = cgroup_css_links_read,
5542 .name = "releasable",
5543 .read_u64 = releasable_read,
5549 struct cgroup_subsys debug_subsys = {
5551 .css_alloc = debug_css_alloc,
5552 .css_free = debug_css_free,
5553 .subsys_id = debug_subsys_id,
5554 .base_cftypes = debug_files,
5556 #endif /* CONFIG_CGROUP_DEBUG */