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>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hash.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_proc */
64 #include <linux/atomic.h>
67 * cgroup_mutex is the master lock. Any modification to cgroup or its
68 * hierarchy must be performed while holding it.
70 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
71 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
72 * release_agent_path and so on. Modifying requires both cgroup_mutex and
73 * cgroup_root_mutex. Readers can acquire either of the two. This is to
74 * break the following locking order cycle.
76 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
77 * B. namespace_sem -> cgroup_mutex
79 * B happens only through cgroup_show_options() and using cgroup_root_mutex
82 static DEFINE_MUTEX(cgroup_mutex);
83 static DEFINE_MUTEX(cgroup_root_mutex);
86 * Generate an array of cgroup subsystem pointers. At boot time, this is
87 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
88 * registered after that. The mutable section of this array is protected by
91 #define SUBSYS(_x) &_x ## _subsys,
92 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
93 #include <linux/cgroup_subsys.h>
96 #define MAX_CGROUP_ROOT_NAMELEN 64
99 * A cgroupfs_root represents the root of a cgroup hierarchy,
100 * and may be associated with a superblock to form an active
103 struct cgroupfs_root {
104 struct super_block *sb;
107 * The bitmask of subsystems intended to be attached to this
110 unsigned long subsys_bits;
112 /* Unique id for this hierarchy. */
115 /* The bitmask of subsystems currently attached to this hierarchy */
116 unsigned long actual_subsys_bits;
118 /* A list running through the attached subsystems */
119 struct list_head subsys_list;
121 /* The root cgroup for this hierarchy */
122 struct cgroup top_cgroup;
124 /* Tracks how many cgroups are currently defined in hierarchy.*/
125 int number_of_cgroups;
127 /* A list running through the active hierarchies */
128 struct list_head root_list;
130 /* Hierarchy-specific flags */
133 /* The path to use for release notifications. */
134 char release_agent_path[PATH_MAX];
136 /* The name for this hierarchy - may be empty */
137 char name[MAX_CGROUP_ROOT_NAMELEN];
141 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
142 * subsystems that are otherwise unattached - it never has more than a
143 * single cgroup, and all tasks are part of that cgroup.
145 static struct cgroupfs_root rootnode;
148 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
149 * cgroup_subsys->use_id != 0.
151 #define CSS_ID_MAX (65535)
154 * The css to which this ID points. This pointer is set to valid value
155 * after cgroup is populated. If cgroup is removed, this will be NULL.
156 * This pointer is expected to be RCU-safe because destroy()
157 * is called after synchronize_rcu(). But for safe use, css_is_removed()
158 * css_tryget() should be used for avoiding race.
160 struct cgroup_subsys_state __rcu *css;
166 * Depth in hierarchy which this ID belongs to.
168 unsigned short depth;
170 * ID is freed by RCU. (and lookup routine is RCU safe.)
172 struct rcu_head rcu_head;
174 * Hierarchy of CSS ID belongs to.
176 unsigned short stack[0]; /* Array of Length (depth+1) */
180 * cgroup_event represents events which userspace want to receive.
182 struct cgroup_event {
184 * Cgroup which the event belongs to.
188 * Control file which the event associated.
192 * eventfd to signal userspace about the event.
194 struct eventfd_ctx *eventfd;
196 * Each of these stored in a list by the cgroup.
198 struct list_head list;
200 * All fields below needed to unregister event when
201 * userspace closes eventfd.
204 wait_queue_head_t *wqh;
206 struct work_struct remove;
209 /* The list of hierarchy roots */
211 static LIST_HEAD(roots);
212 static int root_count;
214 static DEFINE_IDA(hierarchy_ida);
215 static int next_hierarchy_id;
216 static DEFINE_SPINLOCK(hierarchy_id_lock);
218 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
219 #define dummytop (&rootnode.top_cgroup)
221 /* This flag indicates whether tasks in the fork and exit paths should
222 * check for fork/exit handlers to call. This avoids us having to do
223 * extra work in the fork/exit path if none of the subsystems need to
226 static int need_forkexit_callback __read_mostly;
228 #ifdef CONFIG_PROVE_LOCKING
229 int cgroup_lock_is_held(void)
231 return lockdep_is_held(&cgroup_mutex);
233 #else /* #ifdef CONFIG_PROVE_LOCKING */
234 int cgroup_lock_is_held(void)
236 return mutex_is_locked(&cgroup_mutex);
238 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
240 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
242 /* convenient tests for these bits */
243 inline int cgroup_is_removed(const struct cgroup *cgrp)
245 return test_bit(CGRP_REMOVED, &cgrp->flags);
248 /* bits in struct cgroupfs_root flags field */
250 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
253 static int cgroup_is_releasable(const struct cgroup *cgrp)
256 (1 << CGRP_RELEASABLE) |
257 (1 << CGRP_NOTIFY_ON_RELEASE);
258 return (cgrp->flags & bits) == bits;
261 static int notify_on_release(const struct cgroup *cgrp)
263 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
266 static int clone_children(const struct cgroup *cgrp)
268 return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
272 * for_each_subsys() allows you to iterate on each subsystem attached to
273 * an active hierarchy
275 #define for_each_subsys(_root, _ss) \
276 list_for_each_entry(_ss, &_root->subsys_list, sibling)
278 /* for_each_active_root() allows you to iterate across the active hierarchies */
279 #define for_each_active_root(_root) \
280 list_for_each_entry(_root, &roots, root_list)
282 /* the list of cgroups eligible for automatic release. Protected by
283 * release_list_lock */
284 static LIST_HEAD(release_list);
285 static DEFINE_RAW_SPINLOCK(release_list_lock);
286 static void cgroup_release_agent(struct work_struct *work);
287 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
288 static void check_for_release(struct cgroup *cgrp);
291 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
292 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
293 * reference to css->refcnt. In general, this refcnt is expected to goes down
296 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
298 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
300 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
302 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
303 wake_up_all(&cgroup_rmdir_waitq);
306 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
311 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
313 cgroup_wakeup_rmdir_waiter(css->cgroup);
317 /* Link structure for associating css_set objects with cgroups */
318 struct cg_cgroup_link {
320 * List running through cg_cgroup_links associated with a
321 * cgroup, anchored on cgroup->css_sets
323 struct list_head cgrp_link_list;
326 * List running through cg_cgroup_links pointing at a
327 * single css_set object, anchored on css_set->cg_links
329 struct list_head cg_link_list;
333 /* The default css_set - used by init and its children prior to any
334 * hierarchies being mounted. It contains a pointer to the root state
335 * for each subsystem. Also used to anchor the list of css_sets. Not
336 * reference-counted, to improve performance when child cgroups
337 * haven't been created.
340 static struct css_set init_css_set;
341 static struct cg_cgroup_link init_css_set_link;
343 static int cgroup_init_idr(struct cgroup_subsys *ss,
344 struct cgroup_subsys_state *css);
346 /* css_set_lock protects the list of css_set objects, and the
347 * chain of tasks off each css_set. Nests outside task->alloc_lock
348 * due to cgroup_iter_start() */
349 static DEFINE_RWLOCK(css_set_lock);
350 static int css_set_count;
353 * hash table for cgroup groups. This improves the performance to find
354 * an existing css_set. This hash doesn't (currently) take into
355 * account cgroups in empty hierarchies.
357 #define CSS_SET_HASH_BITS 7
358 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
359 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
361 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
365 unsigned long tmp = 0UL;
367 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
368 tmp += (unsigned long)css[i];
369 tmp = (tmp >> 16) ^ tmp;
371 index = hash_long(tmp, CSS_SET_HASH_BITS);
373 return &css_set_table[index];
376 static void free_css_set_work(struct work_struct *work)
378 struct css_set *cg = container_of(work, struct css_set, work);
379 struct cg_cgroup_link *link;
380 struct cg_cgroup_link *saved_link;
382 write_lock(&css_set_lock);
383 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
385 struct cgroup *cgrp = link->cgrp;
386 list_del(&link->cg_link_list);
387 list_del(&link->cgrp_link_list);
388 if (atomic_dec_and_test(&cgrp->count)) {
389 check_for_release(cgrp);
390 cgroup_wakeup_rmdir_waiter(cgrp);
394 write_unlock(&css_set_lock);
399 static void free_css_set_rcu(struct rcu_head *obj)
401 struct css_set *cg = container_of(obj, struct css_set, rcu_head);
403 INIT_WORK(&cg->work, free_css_set_work);
404 schedule_work(&cg->work);
407 /* We don't maintain the lists running through each css_set to its
408 * task until after the first call to cgroup_iter_start(). This
409 * reduces the fork()/exit() overhead for people who have cgroups
410 * compiled into their kernel but not actually in use */
411 static int use_task_css_set_links __read_mostly;
414 * refcounted get/put for css_set objects
416 static inline void get_css_set(struct css_set *cg)
418 atomic_inc(&cg->refcount);
421 static void put_css_set(struct css_set *cg)
424 * Ensure that the refcount doesn't hit zero while any readers
425 * can see it. Similar to atomic_dec_and_lock(), but for an
428 if (atomic_add_unless(&cg->refcount, -1, 1))
430 write_lock(&css_set_lock);
431 if (!atomic_dec_and_test(&cg->refcount)) {
432 write_unlock(&css_set_lock);
436 hlist_del(&cg->hlist);
439 write_unlock(&css_set_lock);
440 call_rcu(&cg->rcu_head, free_css_set_rcu);
443 /* We don't maintain the lists running through each css_set to its
444 * task until after the first call to cgroup_iter_start(). This
445 * reduces the fork()/exit() overhead for people who have cgroups
446 * compiled into their kernel but not actually in use */
447 static int use_task_css_set_links __read_mostly;
450 * compare_css_sets - helper function for find_existing_css_set().
451 * @cg: candidate css_set being tested
452 * @old_cg: 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 *cg,
460 struct css_set *old_cg,
461 struct cgroup *new_cgrp,
462 struct cgroup_subsys_state *template[])
464 struct list_head *l1, *l2;
466 if (memcmp(template, cg->subsys, sizeof(cg->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
481 l2 = &old_cg->cg_links;
483 struct cg_cgroup_link *cgl1, *cgl2;
484 struct cgroup *cg1, *cg2;
488 /* See if we reached the end - both lists are equal length. */
489 if (l1 == &cg->cg_links) {
490 BUG_ON(l2 != &old_cg->cg_links);
493 BUG_ON(l2 == &old_cg->cg_links);
495 /* Locate the cgroups associated with these links. */
496 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
497 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
500 /* Hierarchies should be linked in the same order. */
501 BUG_ON(cg1->root != cg2->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 (cg1->root == new_cgrp->root) {
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(
535 struct css_set *oldcg,
537 struct cgroup_subsys_state *template[])
540 struct cgroupfs_root *root = cgrp->root;
541 struct hlist_head *hhead;
542 struct hlist_node *node;
546 * Build the set of subsystem state objects that we want to see in the
547 * new css_set. while subsystems can change globally, the entries here
548 * won't change, so no need for locking.
550 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
551 if (root->subsys_bits & (1UL << i)) {
552 /* Subsystem is in this hierarchy. So we want
553 * the subsystem state from the new
555 template[i] = cgrp->subsys[i];
557 /* Subsystem is not in this hierarchy, so we
558 * don't want to change the subsystem state */
559 template[i] = oldcg->subsys[i];
563 hhead = css_set_hash(template);
564 hlist_for_each_entry(cg, node, hhead, hlist) {
565 if (!compare_css_sets(cg, oldcg, cgrp, template))
568 /* This css_set matches what we need */
572 /* No existing cgroup group matched */
576 static void free_cg_links(struct list_head *tmp)
578 struct cg_cgroup_link *link;
579 struct cg_cgroup_link *saved_link;
581 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
582 list_del(&link->cgrp_link_list);
588 * allocate_cg_links() allocates "count" cg_cgroup_link structures
589 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
590 * success or a negative error
592 static int allocate_cg_links(int count, struct list_head *tmp)
594 struct cg_cgroup_link *link;
597 for (i = 0; i < count; i++) {
598 link = kmalloc(sizeof(*link), GFP_KERNEL);
603 list_add(&link->cgrp_link_list, tmp);
609 * link_css_set - a helper function to link a css_set to a cgroup
610 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
611 * @cg: the css_set to be linked
612 * @cgrp: the destination cgroup
614 static void link_css_set(struct list_head *tmp_cg_links,
615 struct css_set *cg, struct cgroup *cgrp)
617 struct cg_cgroup_link *link;
619 BUG_ON(list_empty(tmp_cg_links));
620 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
624 atomic_inc(&cgrp->count);
625 list_move(&link->cgrp_link_list, &cgrp->css_sets);
627 * Always add links to the tail of the list so that the list
628 * is sorted by order of hierarchy creation
630 list_add_tail(&link->cg_link_list, &cg->cg_links);
634 * find_css_set() takes an existing cgroup group and a
635 * cgroup object, and returns a css_set object that's
636 * equivalent to the old group, but with the given cgroup
637 * substituted into the appropriate hierarchy. Must be called with
640 static struct css_set *find_css_set(
641 struct css_set *oldcg, struct cgroup *cgrp)
644 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
646 struct list_head tmp_cg_links;
648 struct hlist_head *hhead;
649 struct cg_cgroup_link *link;
651 /* First see if we already have a cgroup group that matches
653 read_lock(&css_set_lock);
654 res = find_existing_css_set(oldcg, cgrp, template);
657 read_unlock(&css_set_lock);
662 res = kmalloc(sizeof(*res), GFP_KERNEL);
666 /* Allocate all the cg_cgroup_link objects that we'll need */
667 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
672 atomic_set(&res->refcount, 1);
673 INIT_LIST_HEAD(&res->cg_links);
674 INIT_LIST_HEAD(&res->tasks);
675 INIT_HLIST_NODE(&res->hlist);
677 /* Copy the set of subsystem state objects generated in
678 * find_existing_css_set() */
679 memcpy(res->subsys, template, sizeof(res->subsys));
681 write_lock(&css_set_lock);
682 /* Add reference counts and links from the new css_set. */
683 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
684 struct cgroup *c = link->cgrp;
685 if (c->root == cgrp->root)
687 link_css_set(&tmp_cg_links, res, c);
690 BUG_ON(!list_empty(&tmp_cg_links));
694 /* Add this cgroup group to the hash table */
695 hhead = css_set_hash(res->subsys);
696 hlist_add_head(&res->hlist, hhead);
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)
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.
721 if (css == &init_css_set) {
722 res = &root->top_cgroup;
724 struct cg_cgroup_link *link;
725 list_for_each_entry(link, &css->cg_links, cg_link_list) {
726 struct cgroup *c = link->cgrp;
727 if (c->root == root) {
733 read_unlock(&css_set_lock);
739 * There is one global cgroup mutex. We also require taking
740 * task_lock() when dereferencing a task's cgroup subsys pointers.
741 * See "The task_lock() exception", at the end of this comment.
743 * A task must hold cgroup_mutex to modify cgroups.
745 * Any task can increment and decrement the count field without lock.
746 * So in general, code holding cgroup_mutex can't rely on the count
747 * field not changing. However, if the count goes to zero, then only
748 * cgroup_attach_task() can increment it again. Because a count of zero
749 * means that no tasks are currently attached, therefore there is no
750 * way a task attached to that cgroup can fork (the other way to
751 * increment the count). So code holding cgroup_mutex can safely
752 * assume that if the count is zero, it will stay zero. Similarly, if
753 * a task holds cgroup_mutex on a cgroup with zero count, it
754 * knows that the cgroup won't be removed, as cgroup_rmdir()
757 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
758 * (usually) take cgroup_mutex. These are the two most performance
759 * critical pieces of code here. The exception occurs on cgroup_exit(),
760 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
761 * is taken, and if the cgroup count is zero, a usermode call made
762 * to the release agent with the name of the cgroup (path relative to
763 * the root of cgroup file system) as the argument.
765 * A cgroup can only be deleted if both its 'count' of using tasks
766 * is zero, and its list of 'children' cgroups is empty. Since all
767 * tasks in the system use _some_ cgroup, and since there is always at
768 * least one task in the system (init, pid == 1), therefore, top_cgroup
769 * always has either children cgroups and/or using tasks. So we don't
770 * need a special hack to ensure that top_cgroup cannot be deleted.
772 * The task_lock() exception
774 * The need for this exception arises from the action of
775 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
776 * another. It does so using cgroup_mutex, however there are
777 * several performance critical places that need to reference
778 * task->cgroups without the expense of grabbing a system global
779 * mutex. Therefore except as noted below, when dereferencing or, as
780 * in cgroup_attach_task(), modifying a task's cgroups pointer we use
781 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
782 * the task_struct routinely used for such matters.
784 * P.S. One more locking exception. RCU is used to guard the
785 * update of a tasks cgroup pointer by cgroup_attach_task()
789 * cgroup_lock - lock out any changes to cgroup structures
792 void cgroup_lock(void)
794 mutex_lock(&cgroup_mutex);
796 EXPORT_SYMBOL_GPL(cgroup_lock);
799 * cgroup_unlock - release lock on cgroup changes
801 * Undo the lock taken in a previous cgroup_lock() call.
803 void cgroup_unlock(void)
805 mutex_unlock(&cgroup_mutex);
807 EXPORT_SYMBOL_GPL(cgroup_unlock);
810 * A couple of forward declarations required, due to cyclic reference loop:
811 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
812 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
816 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
817 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, struct nameidata *);
818 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
819 static int cgroup_populate_dir(struct cgroup *cgrp);
820 static const struct inode_operations cgroup_dir_inode_operations;
821 static const struct file_operations proc_cgroupstats_operations;
823 static struct backing_dev_info cgroup_backing_dev_info = {
825 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
828 static int alloc_css_id(struct cgroup_subsys *ss,
829 struct cgroup *parent, struct cgroup *child);
831 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
833 struct inode *inode = new_inode(sb);
836 inode->i_ino = get_next_ino();
837 inode->i_mode = mode;
838 inode->i_uid = current_fsuid();
839 inode->i_gid = current_fsgid();
840 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
841 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
847 * Call subsys's pre_destroy handler.
848 * This is called before css refcnt check.
850 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
852 struct cgroup_subsys *ss;
855 for_each_subsys(cgrp->root, ss)
856 if (ss->pre_destroy) {
857 ret = ss->pre_destroy(cgrp);
865 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
867 /* is dentry a directory ? if so, kfree() associated cgroup */
868 if (S_ISDIR(inode->i_mode)) {
869 struct cgroup *cgrp = dentry->d_fsdata;
870 struct cgroup_subsys *ss;
871 BUG_ON(!(cgroup_is_removed(cgrp)));
872 /* It's possible for external users to be holding css
873 * reference counts on a cgroup; css_put() needs to
874 * be able to access the cgroup after decrementing
875 * the reference count in order to know if it needs to
876 * queue the cgroup to be handled by the release
880 mutex_lock(&cgroup_mutex);
882 * Release the subsystem state objects.
884 for_each_subsys(cgrp->root, ss)
887 cgrp->root->number_of_cgroups--;
888 mutex_unlock(&cgroup_mutex);
891 * Drop the active superblock reference that we took when we
894 deactivate_super(cgrp->root->sb);
897 * if we're getting rid of the cgroup, refcount should ensure
898 * that there are no pidlists left.
900 BUG_ON(!list_empty(&cgrp->pidlists));
902 kfree_rcu(cgrp, rcu_head);
907 static int cgroup_delete(const struct dentry *d)
912 static void remove_dir(struct dentry *d)
914 struct dentry *parent = dget(d->d_parent);
917 simple_rmdir(parent->d_inode, d);
921 static void cgroup_clear_directory(struct dentry *dentry)
923 struct list_head *node;
925 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
926 spin_lock(&dentry->d_lock);
927 node = dentry->d_subdirs.next;
928 while (node != &dentry->d_subdirs) {
929 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
931 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
934 /* This should never be called on a cgroup
935 * directory with child cgroups */
936 BUG_ON(d->d_inode->i_mode & S_IFDIR);
938 spin_unlock(&d->d_lock);
939 spin_unlock(&dentry->d_lock);
941 simple_unlink(dentry->d_inode, d);
943 spin_lock(&dentry->d_lock);
945 spin_unlock(&d->d_lock);
946 node = dentry->d_subdirs.next;
948 spin_unlock(&dentry->d_lock);
952 * NOTE : the dentry must have been dget()'ed
954 static void cgroup_d_remove_dir(struct dentry *dentry)
956 struct dentry *parent;
958 cgroup_clear_directory(dentry);
960 parent = dentry->d_parent;
961 spin_lock(&parent->d_lock);
962 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
963 list_del_init(&dentry->d_u.d_child);
964 spin_unlock(&dentry->d_lock);
965 spin_unlock(&parent->d_lock);
970 * Call with cgroup_mutex held. Drops reference counts on modules, including
971 * any duplicate ones that parse_cgroupfs_options took. If this function
972 * returns an error, no reference counts are touched.
974 static int rebind_subsystems(struct cgroupfs_root *root,
975 unsigned long final_bits)
977 unsigned long added_bits, removed_bits;
978 struct cgroup *cgrp = &root->top_cgroup;
981 BUG_ON(!mutex_is_locked(&cgroup_mutex));
982 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
984 removed_bits = root->actual_subsys_bits & ~final_bits;
985 added_bits = final_bits & ~root->actual_subsys_bits;
986 /* Check that any added subsystems are currently free */
987 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
988 unsigned long bit = 1UL << i;
989 struct cgroup_subsys *ss = subsys[i];
990 if (!(bit & added_bits))
993 * Nobody should tell us to do a subsys that doesn't exist:
994 * parse_cgroupfs_options should catch that case and refcounts
995 * ensure that subsystems won't disappear once selected.
998 if (ss->root != &rootnode) {
999 /* Subsystem isn't free */
1004 /* Currently we don't handle adding/removing subsystems when
1005 * any child cgroups exist. This is theoretically supportable
1006 * but involves complex error handling, so it's being left until
1008 if (root->number_of_cgroups > 1)
1011 /* Process each subsystem */
1012 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1013 struct cgroup_subsys *ss = subsys[i];
1014 unsigned long bit = 1UL << i;
1015 if (bit & added_bits) {
1016 /* We're binding this subsystem to this hierarchy */
1018 BUG_ON(cgrp->subsys[i]);
1019 BUG_ON(!dummytop->subsys[i]);
1020 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1021 mutex_lock(&ss->hierarchy_mutex);
1022 cgrp->subsys[i] = dummytop->subsys[i];
1023 cgrp->subsys[i]->cgroup = cgrp;
1024 list_move(&ss->sibling, &root->subsys_list);
1028 mutex_unlock(&ss->hierarchy_mutex);
1029 /* refcount was already taken, and we're keeping it */
1030 } else if (bit & removed_bits) {
1031 /* We're removing this subsystem */
1033 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1034 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1035 mutex_lock(&ss->hierarchy_mutex);
1038 dummytop->subsys[i]->cgroup = dummytop;
1039 cgrp->subsys[i] = NULL;
1040 subsys[i]->root = &rootnode;
1041 list_move(&ss->sibling, &rootnode.subsys_list);
1042 mutex_unlock(&ss->hierarchy_mutex);
1043 /* subsystem is now free - drop reference on module */
1044 module_put(ss->module);
1045 } else if (bit & final_bits) {
1046 /* Subsystem state should already exist */
1048 BUG_ON(!cgrp->subsys[i]);
1050 * a refcount was taken, but we already had one, so
1051 * drop the extra reference.
1053 module_put(ss->module);
1054 #ifdef CONFIG_MODULE_UNLOAD
1055 BUG_ON(ss->module && !module_refcount(ss->module));
1058 /* Subsystem state shouldn't exist */
1059 BUG_ON(cgrp->subsys[i]);
1062 root->subsys_bits = root->actual_subsys_bits = final_bits;
1068 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1070 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1071 struct cgroup_subsys *ss;
1073 mutex_lock(&cgroup_root_mutex);
1074 for_each_subsys(root, ss)
1075 seq_printf(seq, ",%s", ss->name);
1076 if (test_bit(ROOT_NOPREFIX, &root->flags))
1077 seq_puts(seq, ",noprefix");
1078 if (strlen(root->release_agent_path))
1079 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1080 if (clone_children(&root->top_cgroup))
1081 seq_puts(seq, ",clone_children");
1082 if (strlen(root->name))
1083 seq_printf(seq, ",name=%s", root->name);
1084 mutex_unlock(&cgroup_root_mutex);
1088 struct cgroup_sb_opts {
1089 unsigned long subsys_bits;
1090 unsigned long flags;
1091 char *release_agent;
1092 bool clone_children;
1094 /* User explicitly requested empty subsystem */
1097 struct cgroupfs_root *new_root;
1102 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1103 * with cgroup_mutex held to protect the subsys[] array. This function takes
1104 * refcounts on subsystems to be used, unless it returns error, in which case
1105 * no refcounts are taken.
1107 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1109 char *token, *o = data;
1110 bool all_ss = false, one_ss = false;
1111 unsigned long mask = (unsigned long)-1;
1113 bool module_pin_failed = false;
1115 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1117 #ifdef CONFIG_CPUSETS
1118 mask = ~(1UL << cpuset_subsys_id);
1121 memset(opts, 0, sizeof(*opts));
1123 while ((token = strsep(&o, ",")) != NULL) {
1126 if (!strcmp(token, "none")) {
1127 /* Explicitly have no subsystems */
1131 if (!strcmp(token, "all")) {
1132 /* Mutually exclusive option 'all' + subsystem name */
1138 if (!strcmp(token, "noprefix")) {
1139 set_bit(ROOT_NOPREFIX, &opts->flags);
1142 if (!strcmp(token, "clone_children")) {
1143 opts->clone_children = true;
1146 if (!strncmp(token, "release_agent=", 14)) {
1147 /* Specifying two release agents is forbidden */
1148 if (opts->release_agent)
1150 opts->release_agent =
1151 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1152 if (!opts->release_agent)
1156 if (!strncmp(token, "name=", 5)) {
1157 const char *name = token + 5;
1158 /* Can't specify an empty name */
1161 /* Must match [\w.-]+ */
1162 for (i = 0; i < strlen(name); i++) {
1166 if ((c == '.') || (c == '-') || (c == '_'))
1170 /* Specifying two names is forbidden */
1173 opts->name = kstrndup(name,
1174 MAX_CGROUP_ROOT_NAMELEN - 1,
1182 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1183 struct cgroup_subsys *ss = subsys[i];
1186 if (strcmp(token, ss->name))
1191 /* Mutually exclusive option 'all' + subsystem name */
1194 set_bit(i, &opts->subsys_bits);
1199 if (i == CGROUP_SUBSYS_COUNT)
1204 * If the 'all' option was specified select all the subsystems,
1205 * otherwise if 'none', 'name=' and a subsystem name options
1206 * were not specified, let's default to 'all'
1208 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1209 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1210 struct cgroup_subsys *ss = subsys[i];
1215 set_bit(i, &opts->subsys_bits);
1219 /* Consistency checks */
1222 * Option noprefix was introduced just for backward compatibility
1223 * with the old cpuset, so we allow noprefix only if mounting just
1224 * the cpuset subsystem.
1226 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1227 (opts->subsys_bits & mask))
1231 /* Can't specify "none" and some subsystems */
1232 if (opts->subsys_bits && opts->none)
1236 * We either have to specify by name or by subsystems. (So all
1237 * empty hierarchies must have a name).
1239 if (!opts->subsys_bits && !opts->name)
1243 * Grab references on all the modules we'll need, so the subsystems
1244 * don't dance around before rebind_subsystems attaches them. This may
1245 * take duplicate reference counts on a subsystem that's already used,
1246 * but rebind_subsystems handles this case.
1248 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1249 unsigned long bit = 1UL << i;
1251 if (!(bit & opts->subsys_bits))
1253 if (!try_module_get(subsys[i]->module)) {
1254 module_pin_failed = true;
1258 if (module_pin_failed) {
1260 * oops, one of the modules was going away. this means that we
1261 * raced with a module_delete call, and to the user this is
1262 * essentially a "subsystem doesn't exist" case.
1264 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1265 /* drop refcounts only on the ones we took */
1266 unsigned long bit = 1UL << i;
1268 if (!(bit & opts->subsys_bits))
1270 module_put(subsys[i]->module);
1278 static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1281 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1282 unsigned long bit = 1UL << i;
1284 if (!(bit & subsys_bits))
1286 module_put(subsys[i]->module);
1290 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1293 struct cgroupfs_root *root = sb->s_fs_info;
1294 struct cgroup *cgrp = &root->top_cgroup;
1295 struct cgroup_sb_opts opts;
1297 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1298 mutex_lock(&cgroup_mutex);
1299 mutex_lock(&cgroup_root_mutex);
1301 /* See what subsystems are wanted */
1302 ret = parse_cgroupfs_options(data, &opts);
1306 /* Don't allow flags or name to change at remount */
1307 if (opts.flags != root->flags ||
1308 (opts.name && strcmp(opts.name, root->name))) {
1310 drop_parsed_module_refcounts(opts.subsys_bits);
1314 ret = rebind_subsystems(root, opts.subsys_bits);
1316 drop_parsed_module_refcounts(opts.subsys_bits);
1320 /* (re)populate subsystem files */
1321 cgroup_populate_dir(cgrp);
1323 if (opts.release_agent)
1324 strcpy(root->release_agent_path, opts.release_agent);
1326 kfree(opts.release_agent);
1328 mutex_unlock(&cgroup_root_mutex);
1329 mutex_unlock(&cgroup_mutex);
1330 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1334 static const struct super_operations cgroup_ops = {
1335 .statfs = simple_statfs,
1336 .drop_inode = generic_delete_inode,
1337 .show_options = cgroup_show_options,
1338 .remount_fs = cgroup_remount,
1341 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1343 INIT_LIST_HEAD(&cgrp->sibling);
1344 INIT_LIST_HEAD(&cgrp->children);
1345 INIT_LIST_HEAD(&cgrp->css_sets);
1346 INIT_LIST_HEAD(&cgrp->release_list);
1347 INIT_LIST_HEAD(&cgrp->pidlists);
1348 mutex_init(&cgrp->pidlist_mutex);
1349 INIT_LIST_HEAD(&cgrp->event_list);
1350 spin_lock_init(&cgrp->event_list_lock);
1353 static void init_cgroup_root(struct cgroupfs_root *root)
1355 struct cgroup *cgrp = &root->top_cgroup;
1356 INIT_LIST_HEAD(&root->subsys_list);
1357 INIT_LIST_HEAD(&root->root_list);
1358 root->number_of_cgroups = 1;
1360 cgrp->top_cgroup = cgrp;
1361 init_cgroup_housekeeping(cgrp);
1364 static bool init_root_id(struct cgroupfs_root *root)
1369 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1371 spin_lock(&hierarchy_id_lock);
1372 /* Try to allocate the next unused ID */
1373 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1374 &root->hierarchy_id);
1376 /* Try again starting from 0 */
1377 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1379 next_hierarchy_id = root->hierarchy_id + 1;
1380 } else if (ret != -EAGAIN) {
1381 /* Can only get here if the 31-bit IDR is full ... */
1384 spin_unlock(&hierarchy_id_lock);
1389 static int cgroup_test_super(struct super_block *sb, void *data)
1391 struct cgroup_sb_opts *opts = data;
1392 struct cgroupfs_root *root = sb->s_fs_info;
1394 /* If we asked for a name then it must match */
1395 if (opts->name && strcmp(opts->name, root->name))
1399 * If we asked for subsystems (or explicitly for no
1400 * subsystems) then they must match
1402 if ((opts->subsys_bits || opts->none)
1403 && (opts->subsys_bits != root->subsys_bits))
1409 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1411 struct cgroupfs_root *root;
1413 if (!opts->subsys_bits && !opts->none)
1416 root = kzalloc(sizeof(*root), GFP_KERNEL);
1418 return ERR_PTR(-ENOMEM);
1420 if (!init_root_id(root)) {
1422 return ERR_PTR(-ENOMEM);
1424 init_cgroup_root(root);
1426 root->subsys_bits = opts->subsys_bits;
1427 root->flags = opts->flags;
1428 if (opts->release_agent)
1429 strcpy(root->release_agent_path, opts->release_agent);
1431 strcpy(root->name, opts->name);
1432 if (opts->clone_children)
1433 set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1437 static void cgroup_drop_root(struct cgroupfs_root *root)
1442 BUG_ON(!root->hierarchy_id);
1443 spin_lock(&hierarchy_id_lock);
1444 ida_remove(&hierarchy_ida, root->hierarchy_id);
1445 spin_unlock(&hierarchy_id_lock);
1449 static int cgroup_set_super(struct super_block *sb, void *data)
1452 struct cgroup_sb_opts *opts = data;
1454 /* If we don't have a new root, we can't set up a new sb */
1455 if (!opts->new_root)
1458 BUG_ON(!opts->subsys_bits && !opts->none);
1460 ret = set_anon_super(sb, NULL);
1464 sb->s_fs_info = opts->new_root;
1465 opts->new_root->sb = sb;
1467 sb->s_blocksize = PAGE_CACHE_SIZE;
1468 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1469 sb->s_magic = CGROUP_SUPER_MAGIC;
1470 sb->s_op = &cgroup_ops;
1475 static int cgroup_get_rootdir(struct super_block *sb)
1477 static const struct dentry_operations cgroup_dops = {
1478 .d_iput = cgroup_diput,
1479 .d_delete = cgroup_delete,
1482 struct inode *inode =
1483 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1488 inode->i_fop = &simple_dir_operations;
1489 inode->i_op = &cgroup_dir_inode_operations;
1490 /* directories start off with i_nlink == 2 (for "." entry) */
1492 sb->s_root = d_make_root(inode);
1495 /* for everything else we want ->d_op set */
1496 sb->s_d_op = &cgroup_dops;
1500 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1501 int flags, const char *unused_dev_name,
1504 struct cgroup_sb_opts opts;
1505 struct cgroupfs_root *root;
1507 struct super_block *sb;
1508 struct cgroupfs_root *new_root;
1509 struct inode *inode;
1511 /* First find the desired set of subsystems */
1512 mutex_lock(&cgroup_mutex);
1513 ret = parse_cgroupfs_options(data, &opts);
1514 mutex_unlock(&cgroup_mutex);
1519 * Allocate a new cgroup root. We may not need it if we're
1520 * reusing an existing hierarchy.
1522 new_root = cgroup_root_from_opts(&opts);
1523 if (IS_ERR(new_root)) {
1524 ret = PTR_ERR(new_root);
1527 opts.new_root = new_root;
1529 /* Locate an existing or new sb for this hierarchy */
1530 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1533 cgroup_drop_root(opts.new_root);
1537 root = sb->s_fs_info;
1539 if (root == opts.new_root) {
1540 /* We used the new root structure, so this is a new hierarchy */
1541 struct list_head tmp_cg_links;
1542 struct cgroup *root_cgrp = &root->top_cgroup;
1543 struct cgroupfs_root *existing_root;
1544 const struct cred *cred;
1547 BUG_ON(sb->s_root != NULL);
1549 ret = cgroup_get_rootdir(sb);
1551 goto drop_new_super;
1552 inode = sb->s_root->d_inode;
1554 mutex_lock(&inode->i_mutex);
1555 mutex_lock(&cgroup_mutex);
1556 mutex_lock(&cgroup_root_mutex);
1558 /* Check for name clashes with existing mounts */
1560 if (strlen(root->name))
1561 for_each_active_root(existing_root)
1562 if (!strcmp(existing_root->name, root->name))
1566 * We're accessing css_set_count without locking
1567 * css_set_lock here, but that's OK - it can only be
1568 * increased by someone holding cgroup_lock, and
1569 * that's us. The worst that can happen is that we
1570 * have some link structures left over
1572 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1576 ret = rebind_subsystems(root, root->subsys_bits);
1577 if (ret == -EBUSY) {
1578 free_cg_links(&tmp_cg_links);
1582 * There must be no failure case after here, since rebinding
1583 * takes care of subsystems' refcounts, which are explicitly
1584 * dropped in the failure exit path.
1587 /* EBUSY should be the only error here */
1590 list_add(&root->root_list, &roots);
1593 sb->s_root->d_fsdata = root_cgrp;
1594 root->top_cgroup.dentry = sb->s_root;
1596 /* Link the top cgroup in this hierarchy into all
1597 * the css_set objects */
1598 write_lock(&css_set_lock);
1599 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1600 struct hlist_head *hhead = &css_set_table[i];
1601 struct hlist_node *node;
1604 hlist_for_each_entry(cg, node, hhead, hlist)
1605 link_css_set(&tmp_cg_links, cg, root_cgrp);
1607 write_unlock(&css_set_lock);
1609 free_cg_links(&tmp_cg_links);
1611 BUG_ON(!list_empty(&root_cgrp->sibling));
1612 BUG_ON(!list_empty(&root_cgrp->children));
1613 BUG_ON(root->number_of_cgroups != 1);
1615 cred = override_creds(&init_cred);
1616 cgroup_populate_dir(root_cgrp);
1618 mutex_unlock(&cgroup_root_mutex);
1619 mutex_unlock(&cgroup_mutex);
1620 mutex_unlock(&inode->i_mutex);
1623 * We re-used an existing hierarchy - the new root (if
1624 * any) is not needed
1626 cgroup_drop_root(opts.new_root);
1627 /* no subsys rebinding, so refcounts don't change */
1628 drop_parsed_module_refcounts(opts.subsys_bits);
1631 kfree(opts.release_agent);
1633 return dget(sb->s_root);
1636 mutex_unlock(&cgroup_root_mutex);
1637 mutex_unlock(&cgroup_mutex);
1638 mutex_unlock(&inode->i_mutex);
1640 deactivate_locked_super(sb);
1642 drop_parsed_module_refcounts(opts.subsys_bits);
1644 kfree(opts.release_agent);
1646 return ERR_PTR(ret);
1649 static void cgroup_kill_sb(struct super_block *sb) {
1650 struct cgroupfs_root *root = sb->s_fs_info;
1651 struct cgroup *cgrp = &root->top_cgroup;
1653 struct cg_cgroup_link *link;
1654 struct cg_cgroup_link *saved_link;
1658 BUG_ON(root->number_of_cgroups != 1);
1659 BUG_ON(!list_empty(&cgrp->children));
1660 BUG_ON(!list_empty(&cgrp->sibling));
1662 mutex_lock(&cgroup_mutex);
1663 mutex_lock(&cgroup_root_mutex);
1665 /* Rebind all subsystems back to the default hierarchy */
1666 ret = rebind_subsystems(root, 0);
1667 /* Shouldn't be able to fail ... */
1671 * Release all the links from css_sets to this hierarchy's
1674 write_lock(&css_set_lock);
1676 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1678 list_del(&link->cg_link_list);
1679 list_del(&link->cgrp_link_list);
1682 write_unlock(&css_set_lock);
1684 if (!list_empty(&root->root_list)) {
1685 list_del(&root->root_list);
1689 mutex_unlock(&cgroup_root_mutex);
1690 mutex_unlock(&cgroup_mutex);
1692 kill_litter_super(sb);
1693 cgroup_drop_root(root);
1696 static struct file_system_type cgroup_fs_type = {
1698 .mount = cgroup_mount,
1699 .kill_sb = cgroup_kill_sb,
1702 static struct kobject *cgroup_kobj;
1704 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1706 return dentry->d_fsdata;
1709 static inline struct cftype *__d_cft(struct dentry *dentry)
1711 return dentry->d_fsdata;
1715 * cgroup_path - generate the path of a cgroup
1716 * @cgrp: the cgroup in question
1717 * @buf: the buffer to write the path into
1718 * @buflen: the length of the buffer
1720 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1721 * reference. Writes path of cgroup into buf. Returns 0 on success,
1724 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1727 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1728 cgroup_lock_is_held());
1730 if (!dentry || cgrp == dummytop) {
1732 * Inactive subsystems have no dentry for their root
1739 start = buf + buflen;
1743 int len = dentry->d_name.len;
1745 if ((start -= len) < buf)
1746 return -ENAMETOOLONG;
1747 memcpy(start, dentry->d_name.name, len);
1748 cgrp = cgrp->parent;
1752 dentry = rcu_dereference_check(cgrp->dentry,
1753 cgroup_lock_is_held());
1757 return -ENAMETOOLONG;
1760 memmove(buf, start, buf + buflen - start);
1763 EXPORT_SYMBOL_GPL(cgroup_path);
1766 * Control Group taskset
1768 struct task_and_cgroup {
1769 struct task_struct *task;
1770 struct cgroup *cgrp;
1774 struct cgroup_taskset {
1775 struct task_and_cgroup single;
1776 struct flex_array *tc_array;
1779 struct cgroup *cur_cgrp;
1783 * cgroup_taskset_first - reset taskset and return the first task
1784 * @tset: taskset of interest
1786 * @tset iteration is initialized and the first task is returned.
1788 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1790 if (tset->tc_array) {
1792 return cgroup_taskset_next(tset);
1794 tset->cur_cgrp = tset->single.cgrp;
1795 return tset->single.task;
1798 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1801 * cgroup_taskset_next - iterate to the next task in taskset
1802 * @tset: taskset of interest
1804 * Return the next task in @tset. Iteration must have been initialized
1805 * with cgroup_taskset_first().
1807 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1809 struct task_and_cgroup *tc;
1811 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1814 tc = flex_array_get(tset->tc_array, tset->idx++);
1815 tset->cur_cgrp = tc->cgrp;
1818 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1821 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1822 * @tset: taskset of interest
1824 * Return the cgroup for the current (last returned) task of @tset. This
1825 * function must be preceded by either cgroup_taskset_first() or
1826 * cgroup_taskset_next().
1828 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1830 return tset->cur_cgrp;
1832 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1835 * cgroup_taskset_size - return the number of tasks in taskset
1836 * @tset: taskset of interest
1838 int cgroup_taskset_size(struct cgroup_taskset *tset)
1840 return tset->tc_array ? tset->tc_array_len : 1;
1842 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1846 * cgroup_task_migrate - move a task from one cgroup to another.
1848 * 'guarantee' is set if the caller promises that a new css_set for the task
1849 * will already exist. If not set, this function might sleep, and can fail with
1850 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1852 static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1853 struct task_struct *tsk, struct css_set *newcg)
1855 struct css_set *oldcg;
1858 * We are synchronized through threadgroup_lock() against PF_EXITING
1859 * setting such that we can't race against cgroup_exit() changing the
1860 * css_set to init_css_set and dropping the old one.
1862 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1863 oldcg = tsk->cgroups;
1866 rcu_assign_pointer(tsk->cgroups, newcg);
1869 /* Update the css_set linked lists if we're using them */
1870 write_lock(&css_set_lock);
1871 if (!list_empty(&tsk->cg_list))
1872 list_move(&tsk->cg_list, &newcg->tasks);
1873 write_unlock(&css_set_lock);
1876 * We just gained a reference on oldcg by taking it from the task. As
1877 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1878 * it here; it will be freed under RCU.
1882 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1886 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1887 * @cgrp: the cgroup the task is attaching to
1888 * @tsk: the task to be attached
1890 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1893 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1896 struct cgroup_subsys *ss, *failed_ss = NULL;
1897 struct cgroup *oldcgrp;
1898 struct cgroupfs_root *root = cgrp->root;
1899 struct cgroup_taskset tset = { };
1900 struct css_set *newcg;
1903 /* @tsk either already exited or can't exit until the end */
1904 if (tsk->flags & PF_EXITING)
1907 /* Nothing to do if the task is already in that cgroup */
1908 oldcgrp = task_cgroup_from_root(tsk, root);
1909 if (cgrp == oldcgrp)
1912 tset.single.task = tsk;
1913 tset.single.cgrp = oldcgrp;
1915 for_each_subsys(root, ss) {
1916 if (ss->can_attach) {
1917 retval = ss->can_attach(cgrp, &tset);
1920 * Remember on which subsystem the can_attach()
1921 * failed, so that we only call cancel_attach()
1922 * against the subsystems whose can_attach()
1923 * succeeded. (See below)
1931 newcg = find_css_set(tsk->cgroups, cgrp);
1942 cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1944 for_each_subsys(root, ss) {
1946 ss->attach(cgrp, &tset);
1948 set_bit(CGRP_RELEASABLE, &cgrp->flags);
1949 /* put_css_set will not destroy cg until after an RCU grace period */
1953 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1954 * is no longer empty.
1956 cgroup_wakeup_rmdir_waiter(cgrp);
1959 for_each_subsys(root, ss) {
1960 if (ss == failed_ss)
1962 * This subsystem was the one that failed the
1963 * can_attach() check earlier, so we don't need
1964 * to call cancel_attach() against it or any
1965 * remaining subsystems.
1968 if (ss->cancel_attach)
1969 ss->cancel_attach(cgrp, &tset);
1976 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1977 * @from: attach to all cgroups of a given task
1978 * @tsk: the task to be attached
1980 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
1982 struct cgroupfs_root *root;
1986 for_each_active_root(root) {
1987 struct cgroup *from_cg = task_cgroup_from_root(from, root);
1989 retval = cgroup_attach_task(from_cg, tsk);
1997 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2000 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2001 * @cgrp: the cgroup to attach to
2002 * @leader: the threadgroup leader task_struct of the group to be attached
2004 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2005 * task_lock of each thread in leader's threadgroup individually in turn.
2007 static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2009 int retval, i, group_size;
2010 struct cgroup_subsys *ss, *failed_ss = NULL;
2011 /* guaranteed to be initialized later, but the compiler needs this */
2012 struct cgroupfs_root *root = cgrp->root;
2013 /* threadgroup list cursor and array */
2014 struct task_struct *tsk;
2015 struct task_and_cgroup *tc;
2016 struct flex_array *group;
2017 struct cgroup_taskset tset = { };
2020 * step 0: in order to do expensive, possibly blocking operations for
2021 * every thread, we cannot iterate the thread group list, since it needs
2022 * rcu or tasklist locked. instead, build an array of all threads in the
2023 * group - group_rwsem prevents new threads from appearing, and if
2024 * threads exit, this will just be an over-estimate.
2026 group_size = get_nr_threads(leader);
2027 /* flex_array supports very large thread-groups better than kmalloc. */
2028 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2031 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2032 retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2034 goto out_free_group_list;
2039 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2040 * already PF_EXITING could be freed from underneath us unless we
2041 * take an rcu_read_lock.
2045 struct task_and_cgroup ent;
2047 /* @tsk either already exited or can't exit until the end */
2048 if (tsk->flags & PF_EXITING)
2051 /* as per above, nr_threads may decrease, but not increase. */
2052 BUG_ON(i >= group_size);
2054 ent.cgrp = task_cgroup_from_root(tsk, root);
2055 /* nothing to do if this task is already in the cgroup */
2056 if (ent.cgrp == cgrp)
2059 * saying GFP_ATOMIC has no effect here because we did prealloc
2060 * earlier, but it's good form to communicate our expectations.
2062 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2063 BUG_ON(retval != 0);
2065 } while_each_thread(leader, tsk);
2067 /* remember the number of threads in the array for later. */
2069 tset.tc_array = group;
2070 tset.tc_array_len = group_size;
2072 /* methods shouldn't be called if no task is actually migrating */
2075 goto out_free_group_list;
2078 * step 1: check that we can legitimately attach to the cgroup.
2080 for_each_subsys(root, ss) {
2081 if (ss->can_attach) {
2082 retval = ss->can_attach(cgrp, &tset);
2085 goto out_cancel_attach;
2091 * step 2: make sure css_sets exist for all threads to be migrated.
2092 * we use find_css_set, which allocates a new one if necessary.
2094 for (i = 0; i < group_size; i++) {
2095 tc = flex_array_get(group, i);
2096 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2099 goto out_put_css_set_refs;
2104 * step 3: now that we're guaranteed success wrt the css_sets,
2105 * proceed to move all tasks to the new cgroup. There are no
2106 * failure cases after here, so this is the commit point.
2108 for (i = 0; i < group_size; i++) {
2109 tc = flex_array_get(group, i);
2110 cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2112 /* nothing is sensitive to fork() after this point. */
2115 * step 4: do subsystem attach callbacks.
2117 for_each_subsys(root, ss) {
2119 ss->attach(cgrp, &tset);
2123 * step 5: success! and cleanup
2126 cgroup_wakeup_rmdir_waiter(cgrp);
2128 out_put_css_set_refs:
2130 for (i = 0; i < group_size; i++) {
2131 tc = flex_array_get(group, i);
2134 put_css_set(tc->cg);
2139 for_each_subsys(root, ss) {
2140 if (ss == failed_ss)
2142 if (ss->cancel_attach)
2143 ss->cancel_attach(cgrp, &tset);
2146 out_free_group_list:
2147 flex_array_free(group);
2151 static int cgroup_allow_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
2153 struct cgroup_subsys *ss;
2156 for_each_subsys(cgrp->root, ss) {
2157 if (ss->allow_attach) {
2158 ret = ss->allow_attach(cgrp, tset);
2170 * Find the task_struct of the task to attach by vpid and pass it along to the
2171 * function to attach either it or all tasks in its threadgroup. Will lock
2172 * cgroup_mutex and threadgroup; may take task_lock of task.
2174 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2176 struct task_struct *tsk;
2177 const struct cred *cred = current_cred(), *tcred;
2180 if (!cgroup_lock_live_group(cgrp))
2186 tsk = find_task_by_vpid(pid);
2190 goto out_unlock_cgroup;
2193 * even if we're attaching all tasks in the thread group, we
2194 * only need to check permissions on one of them.
2196 tcred = __task_cred(tsk);
2198 cred->euid != tcred->uid &&
2199 cred->euid != tcred->suid) {
2201 * if the default permission check fails, give each
2202 * cgroup a chance to extend the permission check
2204 struct cgroup_taskset tset = { };
2205 tset.single.task = tsk;
2206 tset.single.cgrp = cgrp;
2207 ret = cgroup_allow_attach(cgrp, &tset);
2218 tsk = tsk->group_leader;
2219 get_task_struct(tsk);
2222 threadgroup_lock(tsk);
2224 if (!thread_group_leader(tsk)) {
2226 * a race with de_thread from another thread's exec()
2227 * may strip us of our leadership, if this happens,
2228 * there is no choice but to throw this task away and
2229 * try again; this is
2230 * "double-double-toil-and-trouble-check locking".
2232 threadgroup_unlock(tsk);
2233 put_task_struct(tsk);
2234 goto retry_find_task;
2236 ret = cgroup_attach_proc(cgrp, tsk);
2238 ret = cgroup_attach_task(cgrp, tsk);
2239 threadgroup_unlock(tsk);
2241 put_task_struct(tsk);
2247 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2249 return attach_task_by_pid(cgrp, pid, false);
2252 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2254 return attach_task_by_pid(cgrp, tgid, true);
2258 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2259 * @cgrp: the cgroup to be checked for liveness
2261 * On success, returns true; the lock should be later released with
2262 * cgroup_unlock(). On failure returns false with no lock held.
2264 bool cgroup_lock_live_group(struct cgroup *cgrp)
2266 mutex_lock(&cgroup_mutex);
2267 if (cgroup_is_removed(cgrp)) {
2268 mutex_unlock(&cgroup_mutex);
2273 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2275 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2278 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2279 if (strlen(buffer) >= PATH_MAX)
2281 if (!cgroup_lock_live_group(cgrp))
2283 mutex_lock(&cgroup_root_mutex);
2284 strcpy(cgrp->root->release_agent_path, buffer);
2285 mutex_unlock(&cgroup_root_mutex);
2290 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2291 struct seq_file *seq)
2293 if (!cgroup_lock_live_group(cgrp))
2295 seq_puts(seq, cgrp->root->release_agent_path);
2296 seq_putc(seq, '\n');
2301 /* A buffer size big enough for numbers or short strings */
2302 #define CGROUP_LOCAL_BUFFER_SIZE 64
2304 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2306 const char __user *userbuf,
2307 size_t nbytes, loff_t *unused_ppos)
2309 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2315 if (nbytes >= sizeof(buffer))
2317 if (copy_from_user(buffer, userbuf, nbytes))
2320 buffer[nbytes] = 0; /* nul-terminate */
2321 if (cft->write_u64) {
2322 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2325 retval = cft->write_u64(cgrp, cft, val);
2327 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2330 retval = cft->write_s64(cgrp, cft, val);
2337 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2339 const char __user *userbuf,
2340 size_t nbytes, loff_t *unused_ppos)
2342 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2344 size_t max_bytes = cft->max_write_len;
2345 char *buffer = local_buffer;
2348 max_bytes = sizeof(local_buffer) - 1;
2349 if (nbytes >= max_bytes)
2351 /* Allocate a dynamic buffer if we need one */
2352 if (nbytes >= sizeof(local_buffer)) {
2353 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2357 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2362 buffer[nbytes] = 0; /* nul-terminate */
2363 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2367 if (buffer != local_buffer)
2372 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2373 size_t nbytes, loff_t *ppos)
2375 struct cftype *cft = __d_cft(file->f_dentry);
2376 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2378 if (cgroup_is_removed(cgrp))
2381 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2382 if (cft->write_u64 || cft->write_s64)
2383 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2384 if (cft->write_string)
2385 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2387 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2388 return ret ? ret : nbytes;
2393 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2395 char __user *buf, size_t nbytes,
2398 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2399 u64 val = cft->read_u64(cgrp, cft);
2400 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2402 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2405 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2407 char __user *buf, size_t nbytes,
2410 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2411 s64 val = cft->read_s64(cgrp, cft);
2412 int len = sprintf(tmp, "%lld\n", (long long) val);
2414 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2417 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2418 size_t nbytes, loff_t *ppos)
2420 struct cftype *cft = __d_cft(file->f_dentry);
2421 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2423 if (cgroup_is_removed(cgrp))
2427 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2429 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2431 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2436 * seqfile ops/methods for returning structured data. Currently just
2437 * supports string->u64 maps, but can be extended in future.
2440 struct cgroup_seqfile_state {
2442 struct cgroup *cgroup;
2445 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2447 struct seq_file *sf = cb->state;
2448 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2451 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2453 struct cgroup_seqfile_state *state = m->private;
2454 struct cftype *cft = state->cft;
2455 if (cft->read_map) {
2456 struct cgroup_map_cb cb = {
2457 .fill = cgroup_map_add,
2460 return cft->read_map(state->cgroup, cft, &cb);
2462 return cft->read_seq_string(state->cgroup, cft, m);
2465 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2467 struct seq_file *seq = file->private_data;
2468 kfree(seq->private);
2469 return single_release(inode, file);
2472 static const struct file_operations cgroup_seqfile_operations = {
2474 .write = cgroup_file_write,
2475 .llseek = seq_lseek,
2476 .release = cgroup_seqfile_release,
2479 static int cgroup_file_open(struct inode *inode, struct file *file)
2484 err = generic_file_open(inode, file);
2487 cft = __d_cft(file->f_dentry);
2489 if (cft->read_map || cft->read_seq_string) {
2490 struct cgroup_seqfile_state *state =
2491 kzalloc(sizeof(*state), GFP_USER);
2495 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2496 file->f_op = &cgroup_seqfile_operations;
2497 err = single_open(file, cgroup_seqfile_show, state);
2500 } else if (cft->open)
2501 err = cft->open(inode, file);
2508 static int cgroup_file_release(struct inode *inode, struct file *file)
2510 struct cftype *cft = __d_cft(file->f_dentry);
2512 return cft->release(inode, file);
2517 * cgroup_rename - Only allow simple rename of directories in place.
2519 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2520 struct inode *new_dir, struct dentry *new_dentry)
2522 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2524 if (new_dentry->d_inode)
2526 if (old_dir != new_dir)
2528 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2531 static const struct file_operations cgroup_file_operations = {
2532 .read = cgroup_file_read,
2533 .write = cgroup_file_write,
2534 .llseek = generic_file_llseek,
2535 .open = cgroup_file_open,
2536 .release = cgroup_file_release,
2539 static const struct inode_operations cgroup_dir_inode_operations = {
2540 .lookup = cgroup_lookup,
2541 .mkdir = cgroup_mkdir,
2542 .rmdir = cgroup_rmdir,
2543 .rename = cgroup_rename,
2546 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
2548 if (dentry->d_name.len > NAME_MAX)
2549 return ERR_PTR(-ENAMETOOLONG);
2550 d_add(dentry, NULL);
2555 * Check if a file is a control file
2557 static inline struct cftype *__file_cft(struct file *file)
2559 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2560 return ERR_PTR(-EINVAL);
2561 return __d_cft(file->f_dentry);
2564 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2565 struct super_block *sb)
2567 struct inode *inode;
2571 if (dentry->d_inode)
2574 inode = cgroup_new_inode(mode, sb);
2578 if (S_ISDIR(mode)) {
2579 inode->i_op = &cgroup_dir_inode_operations;
2580 inode->i_fop = &simple_dir_operations;
2582 /* start off with i_nlink == 2 (for "." entry) */
2585 /* start with the directory inode held, so that we can
2586 * populate it without racing with another mkdir */
2587 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2588 } else if (S_ISREG(mode)) {
2590 inode->i_fop = &cgroup_file_operations;
2592 d_instantiate(dentry, inode);
2593 dget(dentry); /* Extra count - pin the dentry in core */
2598 * cgroup_create_dir - create a directory for an object.
2599 * @cgrp: the cgroup we create the directory for. It must have a valid
2600 * ->parent field. And we are going to fill its ->dentry field.
2601 * @dentry: dentry of the new cgroup
2602 * @mode: mode to set on new directory.
2604 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2607 struct dentry *parent;
2610 parent = cgrp->parent->dentry;
2611 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2613 dentry->d_fsdata = cgrp;
2614 inc_nlink(parent->d_inode);
2615 rcu_assign_pointer(cgrp->dentry, dentry);
2624 * cgroup_file_mode - deduce file mode of a control file
2625 * @cft: the control file in question
2627 * returns cft->mode if ->mode is not 0
2628 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2629 * returns S_IRUGO if it has only a read handler
2630 * returns S_IWUSR if it has only a write hander
2632 static umode_t cgroup_file_mode(const struct cftype *cft)
2639 if (cft->read || cft->read_u64 || cft->read_s64 ||
2640 cft->read_map || cft->read_seq_string)
2643 if (cft->write || cft->write_u64 || cft->write_s64 ||
2644 cft->write_string || cft->trigger)
2650 int cgroup_add_file(struct cgroup *cgrp,
2651 struct cgroup_subsys *subsys,
2652 const struct cftype *cft)
2654 struct dentry *dir = cgrp->dentry;
2655 struct dentry *dentry;
2659 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2660 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2661 strcpy(name, subsys->name);
2664 strcat(name, cft->name);
2665 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2666 dentry = lookup_one_len(name, dir, strlen(name));
2667 if (!IS_ERR(dentry)) {
2668 mode = cgroup_file_mode(cft);
2669 error = cgroup_create_file(dentry, mode | S_IFREG,
2672 dentry->d_fsdata = (void *)cft;
2675 error = PTR_ERR(dentry);
2678 EXPORT_SYMBOL_GPL(cgroup_add_file);
2680 int cgroup_add_files(struct cgroup *cgrp,
2681 struct cgroup_subsys *subsys,
2682 const struct cftype cft[],
2686 for (i = 0; i < count; i++) {
2687 err = cgroup_add_file(cgrp, subsys, &cft[i]);
2693 EXPORT_SYMBOL_GPL(cgroup_add_files);
2696 * cgroup_task_count - count the number of tasks in a cgroup.
2697 * @cgrp: the cgroup in question
2699 * Return the number of tasks in the cgroup.
2701 int cgroup_task_count(const struct cgroup *cgrp)
2704 struct cg_cgroup_link *link;
2706 read_lock(&css_set_lock);
2707 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2708 count += atomic_read(&link->cg->refcount);
2710 read_unlock(&css_set_lock);
2715 * Advance a list_head iterator. The iterator should be positioned at
2716 * the start of a css_set
2718 static void cgroup_advance_iter(struct cgroup *cgrp,
2719 struct cgroup_iter *it)
2721 struct list_head *l = it->cg_link;
2722 struct cg_cgroup_link *link;
2725 /* Advance to the next non-empty css_set */
2728 if (l == &cgrp->css_sets) {
2732 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2734 } while (list_empty(&cg->tasks));
2736 it->task = cg->tasks.next;
2740 * To reduce the fork() overhead for systems that are not actually
2741 * using their cgroups capability, we don't maintain the lists running
2742 * through each css_set to its tasks until we see the list actually
2743 * used - in other words after the first call to cgroup_iter_start().
2745 static void cgroup_enable_task_cg_lists(void)
2747 struct task_struct *p, *g;
2748 write_lock(&css_set_lock);
2749 use_task_css_set_links = 1;
2751 * We need tasklist_lock because RCU is not safe against
2752 * while_each_thread(). Besides, a forking task that has passed
2753 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2754 * is not guaranteed to have its child immediately visible in the
2755 * tasklist if we walk through it with RCU.
2757 read_lock(&tasklist_lock);
2758 do_each_thread(g, p) {
2761 * We should check if the process is exiting, otherwise
2762 * it will race with cgroup_exit() in that the list
2763 * entry won't be deleted though the process has exited.
2765 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2766 list_add(&p->cg_list, &p->cgroups->tasks);
2768 } while_each_thread(g, p);
2769 read_unlock(&tasklist_lock);
2770 write_unlock(&css_set_lock);
2773 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2774 __acquires(css_set_lock)
2777 * The first time anyone tries to iterate across a cgroup,
2778 * we need to enable the list linking each css_set to its
2779 * tasks, and fix up all existing tasks.
2781 if (!use_task_css_set_links)
2782 cgroup_enable_task_cg_lists();
2784 read_lock(&css_set_lock);
2785 it->cg_link = &cgrp->css_sets;
2786 cgroup_advance_iter(cgrp, it);
2789 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2790 struct cgroup_iter *it)
2792 struct task_struct *res;
2793 struct list_head *l = it->task;
2794 struct cg_cgroup_link *link;
2796 /* If the iterator cg is NULL, we have no tasks */
2799 res = list_entry(l, struct task_struct, cg_list);
2800 /* Advance iterator to find next entry */
2802 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2803 if (l == &link->cg->tasks) {
2804 /* We reached the end of this task list - move on to
2805 * the next cg_cgroup_link */
2806 cgroup_advance_iter(cgrp, it);
2813 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2814 __releases(css_set_lock)
2816 read_unlock(&css_set_lock);
2819 static inline int started_after_time(struct task_struct *t1,
2820 struct timespec *time,
2821 struct task_struct *t2)
2823 int start_diff = timespec_compare(&t1->start_time, time);
2824 if (start_diff > 0) {
2826 } else if (start_diff < 0) {
2830 * Arbitrarily, if two processes started at the same
2831 * time, we'll say that the lower pointer value
2832 * started first. Note that t2 may have exited by now
2833 * so this may not be a valid pointer any longer, but
2834 * that's fine - it still serves to distinguish
2835 * between two tasks started (effectively) simultaneously.
2842 * This function is a callback from heap_insert() and is used to order
2844 * In this case we order the heap in descending task start time.
2846 static inline int started_after(void *p1, void *p2)
2848 struct task_struct *t1 = p1;
2849 struct task_struct *t2 = p2;
2850 return started_after_time(t1, &t2->start_time, t2);
2854 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2855 * @scan: struct cgroup_scanner containing arguments for the scan
2857 * Arguments include pointers to callback functions test_task() and
2859 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2860 * and if it returns true, call process_task() for it also.
2861 * The test_task pointer may be NULL, meaning always true (select all tasks).
2862 * Effectively duplicates cgroup_iter_{start,next,end}()
2863 * but does not lock css_set_lock for the call to process_task().
2864 * The struct cgroup_scanner may be embedded in any structure of the caller's
2866 * It is guaranteed that process_task() will act on every task that
2867 * is a member of the cgroup for the duration of this call. This
2868 * function may or may not call process_task() for tasks that exit
2869 * or move to a different cgroup during the call, or are forked or
2870 * move into the cgroup during the call.
2872 * Note that test_task() may be called with locks held, and may in some
2873 * situations be called multiple times for the same task, so it should
2875 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2876 * pre-allocated and will be used for heap operations (and its "gt" member will
2877 * be overwritten), else a temporary heap will be used (allocation of which
2878 * may cause this function to fail).
2880 int cgroup_scan_tasks(struct cgroup_scanner *scan)
2883 struct cgroup_iter it;
2884 struct task_struct *p, *dropped;
2885 /* Never dereference latest_task, since it's not refcounted */
2886 struct task_struct *latest_task = NULL;
2887 struct ptr_heap tmp_heap;
2888 struct ptr_heap *heap;
2889 struct timespec latest_time = { 0, 0 };
2892 /* The caller supplied our heap and pre-allocated its memory */
2894 heap->gt = &started_after;
2896 /* We need to allocate our own heap memory */
2898 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2900 /* cannot allocate the heap */
2906 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2907 * to determine which are of interest, and using the scanner's
2908 * "process_task" callback to process any of them that need an update.
2909 * Since we don't want to hold any locks during the task updates,
2910 * gather tasks to be processed in a heap structure.
2911 * The heap is sorted by descending task start time.
2912 * If the statically-sized heap fills up, we overflow tasks that
2913 * started later, and in future iterations only consider tasks that
2914 * started after the latest task in the previous pass. This
2915 * guarantees forward progress and that we don't miss any tasks.
2918 cgroup_iter_start(scan->cg, &it);
2919 while ((p = cgroup_iter_next(scan->cg, &it))) {
2921 * Only affect tasks that qualify per the caller's callback,
2922 * if he provided one
2924 if (scan->test_task && !scan->test_task(p, scan))
2927 * Only process tasks that started after the last task
2930 if (!started_after_time(p, &latest_time, latest_task))
2932 dropped = heap_insert(heap, p);
2933 if (dropped == NULL) {
2935 * The new task was inserted; the heap wasn't
2939 } else if (dropped != p) {
2941 * The new task was inserted, and pushed out a
2945 put_task_struct(dropped);
2948 * Else the new task was newer than anything already in
2949 * the heap and wasn't inserted
2952 cgroup_iter_end(scan->cg, &it);
2955 for (i = 0; i < heap->size; i++) {
2956 struct task_struct *q = heap->ptrs[i];
2958 latest_time = q->start_time;
2961 /* Process the task per the caller's callback */
2962 scan->process_task(q, scan);
2966 * If we had to process any tasks at all, scan again
2967 * in case some of them were in the middle of forking
2968 * children that didn't get processed.
2969 * Not the most efficient way to do it, but it avoids
2970 * having to take callback_mutex in the fork path
2974 if (heap == &tmp_heap)
2975 heap_free(&tmp_heap);
2980 * Stuff for reading the 'tasks'/'procs' files.
2982 * Reading this file can return large amounts of data if a cgroup has
2983 * *lots* of attached tasks. So it may need several calls to read(),
2984 * but we cannot guarantee that the information we produce is correct
2985 * unless we produce it entirely atomically.
2989 /* which pidlist file are we talking about? */
2990 enum cgroup_filetype {
2996 * A pidlist is a list of pids that virtually represents the contents of one
2997 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
2998 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3001 struct cgroup_pidlist {
3003 * used to find which pidlist is wanted. doesn't change as long as
3004 * this particular list stays in the list.
3006 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3009 /* how many elements the above list has */
3011 /* how many files are using the current array */
3013 /* each of these stored in a list by its cgroup */
3014 struct list_head links;
3015 /* pointer to the cgroup we belong to, for list removal purposes */
3016 struct cgroup *owner;
3017 /* protects the other fields */
3018 struct rw_semaphore mutex;
3022 * The following two functions "fix" the issue where there are more pids
3023 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3024 * TODO: replace with a kernel-wide solution to this problem
3026 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3027 static void *pidlist_allocate(int count)
3029 if (PIDLIST_TOO_LARGE(count))
3030 return vmalloc(count * sizeof(pid_t));
3032 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3034 static void pidlist_free(void *p)
3036 if (is_vmalloc_addr(p))
3041 static void *pidlist_resize(void *p, int newcount)
3044 /* note: if new alloc fails, old p will still be valid either way */
3045 if (is_vmalloc_addr(p)) {
3046 newlist = vmalloc(newcount * sizeof(pid_t));
3049 memcpy(newlist, p, newcount * sizeof(pid_t));
3052 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3058 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3059 * If the new stripped list is sufficiently smaller and there's enough memory
3060 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3061 * number of unique elements.
3063 /* is the size difference enough that we should re-allocate the array? */
3064 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3065 static int pidlist_uniq(pid_t **p, int length)
3072 * we presume the 0th element is unique, so i starts at 1. trivial
3073 * edge cases first; no work needs to be done for either
3075 if (length == 0 || length == 1)
3077 /* src and dest walk down the list; dest counts unique elements */
3078 for (src = 1; src < length; src++) {
3079 /* find next unique element */
3080 while (list[src] == list[src-1]) {
3085 /* dest always points to where the next unique element goes */
3086 list[dest] = list[src];
3091 * if the length difference is large enough, we want to allocate a
3092 * smaller buffer to save memory. if this fails due to out of memory,
3093 * we'll just stay with what we've got.
3095 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3096 newlist = pidlist_resize(list, dest);
3103 static int cmppid(const void *a, const void *b)
3105 return *(pid_t *)a - *(pid_t *)b;
3109 * find the appropriate pidlist for our purpose (given procs vs tasks)
3110 * returns with the lock on that pidlist already held, and takes care
3111 * of the use count, or returns NULL with no locks held if we're out of
3114 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3115 enum cgroup_filetype type)
3117 struct cgroup_pidlist *l;
3118 /* don't need task_nsproxy() if we're looking at ourself */
3119 struct pid_namespace *ns = current->nsproxy->pid_ns;
3122 * We can't drop the pidlist_mutex before taking the l->mutex in case
3123 * the last ref-holder is trying to remove l from the list at the same
3124 * time. Holding the pidlist_mutex precludes somebody taking whichever
3125 * list we find out from under us - compare release_pid_array().
3127 mutex_lock(&cgrp->pidlist_mutex);
3128 list_for_each_entry(l, &cgrp->pidlists, links) {
3129 if (l->key.type == type && l->key.ns == ns) {
3130 /* make sure l doesn't vanish out from under us */
3131 down_write(&l->mutex);
3132 mutex_unlock(&cgrp->pidlist_mutex);
3136 /* entry not found; create a new one */
3137 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3139 mutex_unlock(&cgrp->pidlist_mutex);
3142 init_rwsem(&l->mutex);
3143 down_write(&l->mutex);
3145 l->key.ns = get_pid_ns(ns);
3146 l->use_count = 0; /* don't increment here */
3149 list_add(&l->links, &cgrp->pidlists);
3150 mutex_unlock(&cgrp->pidlist_mutex);
3155 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3157 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3158 struct cgroup_pidlist **lp)
3162 int pid, n = 0; /* used for populating the array */
3163 struct cgroup_iter it;
3164 struct task_struct *tsk;
3165 struct cgroup_pidlist *l;
3168 * If cgroup gets more users after we read count, we won't have
3169 * enough space - tough. This race is indistinguishable to the
3170 * caller from the case that the additional cgroup users didn't
3171 * show up until sometime later on.
3173 length = cgroup_task_count(cgrp);
3174 array = pidlist_allocate(length);
3177 /* now, populate the array */
3178 cgroup_iter_start(cgrp, &it);
3179 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3180 if (unlikely(n == length))
3182 /* get tgid or pid for procs or tasks file respectively */
3183 if (type == CGROUP_FILE_PROCS)
3184 pid = task_tgid_vnr(tsk);
3186 pid = task_pid_vnr(tsk);
3187 if (pid > 0) /* make sure to only use valid results */
3190 cgroup_iter_end(cgrp, &it);
3192 /* now sort & (if procs) strip out duplicates */
3193 sort(array, length, sizeof(pid_t), cmppid, NULL);
3194 if (type == CGROUP_FILE_PROCS)
3195 length = pidlist_uniq(&array, length);
3196 l = cgroup_pidlist_find(cgrp, type);
3198 pidlist_free(array);
3201 /* store array, freeing old if necessary - lock already held */
3202 pidlist_free(l->list);
3206 up_write(&l->mutex);
3212 * cgroupstats_build - build and fill cgroupstats
3213 * @stats: cgroupstats to fill information into
3214 * @dentry: A dentry entry belonging to the cgroup for which stats have
3217 * Build and fill cgroupstats so that taskstats can export it to user
3220 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3223 struct cgroup *cgrp;
3224 struct cgroup_iter it;
3225 struct task_struct *tsk;
3228 * Validate dentry by checking the superblock operations,
3229 * and make sure it's a directory.
3231 if (dentry->d_sb->s_op != &cgroup_ops ||
3232 !S_ISDIR(dentry->d_inode->i_mode))
3236 cgrp = dentry->d_fsdata;
3238 cgroup_iter_start(cgrp, &it);
3239 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3240 switch (tsk->state) {
3242 stats->nr_running++;
3244 case TASK_INTERRUPTIBLE:
3245 stats->nr_sleeping++;
3247 case TASK_UNINTERRUPTIBLE:
3248 stats->nr_uninterruptible++;
3251 stats->nr_stopped++;
3254 if (delayacct_is_task_waiting_on_io(tsk))
3255 stats->nr_io_wait++;
3259 cgroup_iter_end(cgrp, &it);
3267 * seq_file methods for the tasks/procs files. The seq_file position is the
3268 * next pid to display; the seq_file iterator is a pointer to the pid
3269 * in the cgroup->l->list array.
3272 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3275 * Initially we receive a position value that corresponds to
3276 * one more than the last pid shown (or 0 on the first call or
3277 * after a seek to the start). Use a binary-search to find the
3278 * next pid to display, if any
3280 struct cgroup_pidlist *l = s->private;
3281 int index = 0, pid = *pos;
3284 down_read(&l->mutex);
3286 int end = l->length;
3288 while (index < end) {
3289 int mid = (index + end) / 2;
3290 if (l->list[mid] == pid) {
3293 } else if (l->list[mid] <= pid)
3299 /* If we're off the end of the array, we're done */
3300 if (index >= l->length)
3302 /* Update the abstract position to be the actual pid that we found */
3303 iter = l->list + index;
3308 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3310 struct cgroup_pidlist *l = s->private;
3314 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3316 struct cgroup_pidlist *l = s->private;
3318 pid_t *end = l->list + l->length;
3320 * Advance to the next pid in the array. If this goes off the
3332 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3334 return seq_printf(s, "%d\n", *(int *)v);
3338 * seq_operations functions for iterating on pidlists through seq_file -
3339 * independent of whether it's tasks or procs
3341 static const struct seq_operations cgroup_pidlist_seq_operations = {
3342 .start = cgroup_pidlist_start,
3343 .stop = cgroup_pidlist_stop,
3344 .next = cgroup_pidlist_next,
3345 .show = cgroup_pidlist_show,
3348 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3351 * the case where we're the last user of this particular pidlist will
3352 * have us remove it from the cgroup's list, which entails taking the
3353 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3354 * pidlist_mutex, we have to take pidlist_mutex first.
3356 mutex_lock(&l->owner->pidlist_mutex);
3357 down_write(&l->mutex);
3358 BUG_ON(!l->use_count);
3359 if (!--l->use_count) {
3360 /* we're the last user if refcount is 0; remove and free */
3361 list_del(&l->links);
3362 mutex_unlock(&l->owner->pidlist_mutex);
3363 pidlist_free(l->list);
3364 put_pid_ns(l->key.ns);
3365 up_write(&l->mutex);
3369 mutex_unlock(&l->owner->pidlist_mutex);
3370 up_write(&l->mutex);
3373 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3375 struct cgroup_pidlist *l;
3376 if (!(file->f_mode & FMODE_READ))
3379 * the seq_file will only be initialized if the file was opened for
3380 * reading; hence we check if it's not null only in that case.
3382 l = ((struct seq_file *)file->private_data)->private;
3383 cgroup_release_pid_array(l);
3384 return seq_release(inode, file);
3387 static const struct file_operations cgroup_pidlist_operations = {
3389 .llseek = seq_lseek,
3390 .write = cgroup_file_write,
3391 .release = cgroup_pidlist_release,
3395 * The following functions handle opens on a file that displays a pidlist
3396 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3399 /* helper function for the two below it */
3400 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3402 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3403 struct cgroup_pidlist *l;
3406 /* Nothing to do for write-only files */
3407 if (!(file->f_mode & FMODE_READ))
3410 /* have the array populated */
3411 retval = pidlist_array_load(cgrp, type, &l);
3414 /* configure file information */
3415 file->f_op = &cgroup_pidlist_operations;
3417 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3419 cgroup_release_pid_array(l);
3422 ((struct seq_file *)file->private_data)->private = l;
3425 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3427 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3429 static int cgroup_procs_open(struct inode *unused, struct file *file)
3431 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3434 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3437 return notify_on_release(cgrp);
3440 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3444 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3446 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3448 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3453 * Unregister event and free resources.
3455 * Gets called from workqueue.
3457 static void cgroup_event_remove(struct work_struct *work)
3459 struct cgroup_event *event = container_of(work, struct cgroup_event,
3461 struct cgroup *cgrp = event->cgrp;
3463 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3465 eventfd_ctx_put(event->eventfd);
3471 * Gets called on POLLHUP on eventfd when user closes it.
3473 * Called with wqh->lock held and interrupts disabled.
3475 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3476 int sync, void *key)
3478 struct cgroup_event *event = container_of(wait,
3479 struct cgroup_event, wait);
3480 struct cgroup *cgrp = event->cgrp;
3481 unsigned long flags = (unsigned long)key;
3483 if (flags & POLLHUP) {
3484 __remove_wait_queue(event->wqh, &event->wait);
3485 spin_lock(&cgrp->event_list_lock);
3486 list_del(&event->list);
3487 spin_unlock(&cgrp->event_list_lock);
3489 * We are in atomic context, but cgroup_event_remove() may
3490 * sleep, so we have to call it in workqueue.
3492 schedule_work(&event->remove);
3498 static void cgroup_event_ptable_queue_proc(struct file *file,
3499 wait_queue_head_t *wqh, poll_table *pt)
3501 struct cgroup_event *event = container_of(pt,
3502 struct cgroup_event, pt);
3505 add_wait_queue(wqh, &event->wait);
3509 * Parse input and register new cgroup event handler.
3511 * Input must be in format '<event_fd> <control_fd> <args>'.
3512 * Interpretation of args is defined by control file implementation.
3514 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3517 struct cgroup_event *event = NULL;
3518 unsigned int efd, cfd;
3519 struct file *efile = NULL;
3520 struct file *cfile = NULL;
3524 efd = simple_strtoul(buffer, &endp, 10);
3529 cfd = simple_strtoul(buffer, &endp, 10);
3530 if ((*endp != ' ') && (*endp != '\0'))
3534 event = kzalloc(sizeof(*event), GFP_KERNEL);
3538 INIT_LIST_HEAD(&event->list);
3539 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3540 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3541 INIT_WORK(&event->remove, cgroup_event_remove);
3543 efile = eventfd_fget(efd);
3544 if (IS_ERR(efile)) {
3545 ret = PTR_ERR(efile);
3549 event->eventfd = eventfd_ctx_fileget(efile);
3550 if (IS_ERR(event->eventfd)) {
3551 ret = PTR_ERR(event->eventfd);
3561 /* the process need read permission on control file */
3562 /* AV: shouldn't we check that it's been opened for read instead? */
3563 ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3567 event->cft = __file_cft(cfile);
3568 if (IS_ERR(event->cft)) {
3569 ret = PTR_ERR(event->cft);
3573 if (!event->cft->register_event || !event->cft->unregister_event) {
3578 ret = event->cft->register_event(cgrp, event->cft,
3579 event->eventfd, buffer);
3583 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3584 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3590 * Events should be removed after rmdir of cgroup directory, but before
3591 * destroying subsystem state objects. Let's take reference to cgroup
3592 * directory dentry to do that.
3596 spin_lock(&cgrp->event_list_lock);
3597 list_add(&event->list, &cgrp->event_list);
3598 spin_unlock(&cgrp->event_list_lock);
3609 if (event && event->eventfd && !IS_ERR(event->eventfd))
3610 eventfd_ctx_put(event->eventfd);
3612 if (!IS_ERR_OR_NULL(efile))
3620 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3623 return clone_children(cgrp);
3626 static int cgroup_clone_children_write(struct cgroup *cgrp,
3631 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3633 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3638 * for the common functions, 'private' gives the type of file
3640 /* for hysterical raisins, we can't put this on the older files */
3641 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3642 static struct cftype files[] = {
3645 .open = cgroup_tasks_open,
3646 .write_u64 = cgroup_tasks_write,
3647 .release = cgroup_pidlist_release,
3648 .mode = S_IRUGO | S_IWUSR,
3651 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3652 .open = cgroup_procs_open,
3653 .write_u64 = cgroup_procs_write,
3654 .release = cgroup_pidlist_release,
3655 .mode = S_IRUGO | S_IWUSR,
3658 .name = "notify_on_release",
3659 .read_u64 = cgroup_read_notify_on_release,
3660 .write_u64 = cgroup_write_notify_on_release,
3663 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3664 .write_string = cgroup_write_event_control,
3668 .name = "cgroup.clone_children",
3669 .read_u64 = cgroup_clone_children_read,
3670 .write_u64 = cgroup_clone_children_write,
3674 static struct cftype cft_release_agent = {
3675 .name = "release_agent",
3676 .read_seq_string = cgroup_release_agent_show,
3677 .write_string = cgroup_release_agent_write,
3678 .max_write_len = PATH_MAX,
3681 static int cgroup_populate_dir(struct cgroup *cgrp)
3684 struct cgroup_subsys *ss;
3686 /* First clear out any existing files */
3687 cgroup_clear_directory(cgrp->dentry);
3689 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
3693 if (cgrp == cgrp->top_cgroup) {
3694 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
3698 for_each_subsys(cgrp->root, ss) {
3699 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
3702 /* This cgroup is ready now */
3703 for_each_subsys(cgrp->root, ss) {
3704 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3706 * Update id->css pointer and make this css visible from
3707 * CSS ID functions. This pointer will be dereferened
3708 * from RCU-read-side without locks.
3711 rcu_assign_pointer(css->id->css, css);
3717 static void init_cgroup_css(struct cgroup_subsys_state *css,
3718 struct cgroup_subsys *ss,
3719 struct cgroup *cgrp)
3722 atomic_set(&css->refcnt, 1);
3725 if (cgrp == dummytop)
3726 set_bit(CSS_ROOT, &css->flags);
3727 BUG_ON(cgrp->subsys[ss->subsys_id]);
3728 cgrp->subsys[ss->subsys_id] = css;
3731 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3733 /* We need to take each hierarchy_mutex in a consistent order */
3737 * No worry about a race with rebind_subsystems that might mess up the
3738 * locking order, since both parties are under cgroup_mutex.
3740 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3741 struct cgroup_subsys *ss = subsys[i];
3744 if (ss->root == root)
3745 mutex_lock(&ss->hierarchy_mutex);
3749 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3753 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3754 struct cgroup_subsys *ss = subsys[i];
3757 if (ss->root == root)
3758 mutex_unlock(&ss->hierarchy_mutex);
3763 * cgroup_create - create a cgroup
3764 * @parent: cgroup that will be parent of the new cgroup
3765 * @dentry: dentry of the new cgroup
3766 * @mode: mode to set on new inode
3768 * Must be called with the mutex on the parent inode held
3770 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3773 struct cgroup *cgrp;
3774 struct cgroupfs_root *root = parent->root;
3776 struct cgroup_subsys *ss;
3777 struct super_block *sb = root->sb;
3779 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3783 /* Grab a reference on the superblock so the hierarchy doesn't
3784 * get deleted on unmount if there are child cgroups. This
3785 * can be done outside cgroup_mutex, since the sb can't
3786 * disappear while someone has an open control file on the
3788 atomic_inc(&sb->s_active);
3790 mutex_lock(&cgroup_mutex);
3792 init_cgroup_housekeeping(cgrp);
3794 cgrp->parent = parent;
3795 cgrp->root = parent->root;
3796 cgrp->top_cgroup = parent->top_cgroup;
3798 if (notify_on_release(parent))
3799 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3801 if (clone_children(parent))
3802 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3804 for_each_subsys(root, ss) {
3805 struct cgroup_subsys_state *css = ss->create(cgrp);
3811 init_cgroup_css(css, ss, cgrp);
3813 err = alloc_css_id(ss, parent, cgrp);
3817 /* At error, ->destroy() callback has to free assigned ID. */
3818 if (clone_children(parent) && ss->post_clone)
3819 ss->post_clone(cgrp);
3822 cgroup_lock_hierarchy(root);
3823 list_add(&cgrp->sibling, &cgrp->parent->children);
3824 cgroup_unlock_hierarchy(root);
3825 root->number_of_cgroups++;
3827 err = cgroup_create_dir(cgrp, dentry, mode);
3831 set_bit(CGRP_RELEASABLE, &parent->flags);
3833 /* The cgroup directory was pre-locked for us */
3834 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3836 err = cgroup_populate_dir(cgrp);
3837 /* If err < 0, we have a half-filled directory - oh well ;) */
3839 mutex_unlock(&cgroup_mutex);
3840 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3846 cgroup_lock_hierarchy(root);
3847 list_del(&cgrp->sibling);
3848 cgroup_unlock_hierarchy(root);
3849 root->number_of_cgroups--;
3853 for_each_subsys(root, ss) {
3854 if (cgrp->subsys[ss->subsys_id])
3858 mutex_unlock(&cgroup_mutex);
3860 /* Release the reference count that we took on the superblock */
3861 deactivate_super(sb);
3867 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3869 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3871 /* the vfs holds inode->i_mutex already */
3872 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3875 static int cgroup_has_css_refs(struct cgroup *cgrp)
3877 /* Check the reference count on each subsystem. Since we
3878 * already established that there are no tasks in the
3879 * cgroup, if the css refcount is also 1, then there should
3880 * be no outstanding references, so the subsystem is safe to
3881 * destroy. We scan across all subsystems rather than using
3882 * the per-hierarchy linked list of mounted subsystems since
3883 * we can be called via check_for_release() with no
3884 * synchronization other than RCU, and the subsystem linked
3885 * list isn't RCU-safe */
3888 * We won't need to lock the subsys array, because the subsystems
3889 * we're concerned about aren't going anywhere since our cgroup root
3890 * has a reference on them.
3892 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3893 struct cgroup_subsys *ss = subsys[i];
3894 struct cgroup_subsys_state *css;
3895 /* Skip subsystems not present or not in this hierarchy */
3896 if (ss == NULL || ss->root != cgrp->root)
3898 css = cgrp->subsys[ss->subsys_id];
3899 /* When called from check_for_release() it's possible
3900 * that by this point the cgroup has been removed
3901 * and the css deleted. But a false-positive doesn't
3902 * matter, since it can only happen if the cgroup
3903 * has been deleted and hence no longer needs the
3904 * release agent to be called anyway. */
3905 if (css && (atomic_read(&css->refcnt) > 1))
3912 * Atomically mark all (or else none) of the cgroup's CSS objects as
3913 * CSS_REMOVED. Return true on success, or false if the cgroup has
3914 * busy subsystems. Call with cgroup_mutex held
3917 static int cgroup_clear_css_refs(struct cgroup *cgrp)
3919 struct cgroup_subsys *ss;
3920 unsigned long flags;
3921 bool failed = false;
3922 local_irq_save(flags);
3923 for_each_subsys(cgrp->root, ss) {
3924 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3927 /* We can only remove a CSS with a refcnt==1 */
3928 refcnt = atomic_read(&css->refcnt);
3935 * Drop the refcnt to 0 while we check other
3936 * subsystems. This will cause any racing
3937 * css_tryget() to spin until we set the
3938 * CSS_REMOVED bits or abort
3940 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3946 for_each_subsys(cgrp->root, ss) {
3947 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3950 * Restore old refcnt if we previously managed
3951 * to clear it from 1 to 0
3953 if (!atomic_read(&css->refcnt))
3954 atomic_set(&css->refcnt, 1);
3956 /* Commit the fact that the CSS is removed */
3957 set_bit(CSS_REMOVED, &css->flags);
3960 local_irq_restore(flags);
3964 /* checks if all of the css_sets attached to a cgroup have a refcount of 0.
3965 * Must be called with css_set_lock held */
3966 static int cgroup_css_sets_empty(struct cgroup *cgrp)
3968 struct cg_cgroup_link *link;
3970 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
3971 struct css_set *cg = link->cg;
3972 if (atomic_read(&cg->refcount) > 0)
3979 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3981 struct cgroup *cgrp = dentry->d_fsdata;
3983 struct cgroup *parent;
3985 struct cgroup_event *event, *tmp;
3988 /* the vfs holds both inode->i_mutex already */
3990 mutex_lock(&cgroup_mutex);
3991 if (!cgroup_css_sets_empty(cgrp)) {
3992 mutex_unlock(&cgroup_mutex);
3995 if (!list_empty(&cgrp->children)) {
3996 mutex_unlock(&cgroup_mutex);
3999 mutex_unlock(&cgroup_mutex);
4002 * In general, subsystem has no css->refcnt after pre_destroy(). But
4003 * in racy cases, subsystem may have to get css->refcnt after
4004 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4005 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4006 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4007 * and subsystem's reference count handling. Please see css_get/put
4008 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4010 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4013 * Call pre_destroy handlers of subsys. Notify subsystems
4014 * that rmdir() request comes.
4016 ret = cgroup_call_pre_destroy(cgrp);
4018 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4022 mutex_lock(&cgroup_mutex);
4023 parent = cgrp->parent;
4024 if (!cgroup_css_sets_empty(cgrp) || !list_empty(&cgrp->children)) {
4025 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4026 mutex_unlock(&cgroup_mutex);
4029 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
4030 if (!cgroup_clear_css_refs(cgrp)) {
4031 mutex_unlock(&cgroup_mutex);
4033 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4034 * prepare_to_wait(), we need to check this flag.
4036 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
4038 finish_wait(&cgroup_rmdir_waitq, &wait);
4039 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4040 if (signal_pending(current))
4044 /* NO css_tryget() can success after here. */
4045 finish_wait(&cgroup_rmdir_waitq, &wait);
4046 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4048 raw_spin_lock(&release_list_lock);
4049 set_bit(CGRP_REMOVED, &cgrp->flags);
4050 if (!list_empty(&cgrp->release_list))
4051 list_del_init(&cgrp->release_list);
4052 raw_spin_unlock(&release_list_lock);
4054 cgroup_lock_hierarchy(cgrp->root);
4055 /* delete this cgroup from parent->children */
4056 list_del_init(&cgrp->sibling);
4057 cgroup_unlock_hierarchy(cgrp->root);
4059 d = dget(cgrp->dentry);
4061 cgroup_d_remove_dir(d);
4064 check_for_release(parent);
4067 * Unregister events and notify userspace.
4068 * Notify userspace about cgroup removing only after rmdir of cgroup
4069 * directory to avoid race between userspace and kernelspace
4071 spin_lock(&cgrp->event_list_lock);
4072 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4073 list_del(&event->list);
4074 remove_wait_queue(event->wqh, &event->wait);
4075 eventfd_signal(event->eventfd, 1);
4076 schedule_work(&event->remove);
4078 spin_unlock(&cgrp->event_list_lock);
4080 mutex_unlock(&cgroup_mutex);
4084 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4086 struct cgroup_subsys_state *css;
4088 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4090 /* Create the top cgroup state for this subsystem */
4091 list_add(&ss->sibling, &rootnode.subsys_list);
4092 ss->root = &rootnode;
4093 css = ss->create(dummytop);
4094 /* We don't handle early failures gracefully */
4095 BUG_ON(IS_ERR(css));
4096 init_cgroup_css(css, ss, dummytop);
4098 /* Update the init_css_set to contain a subsys
4099 * pointer to this state - since the subsystem is
4100 * newly registered, all tasks and hence the
4101 * init_css_set is in the subsystem's top cgroup. */
4102 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4104 need_forkexit_callback |= ss->fork || ss->exit;
4106 /* At system boot, before all subsystems have been
4107 * registered, no tasks have been forked, so we don't
4108 * need to invoke fork callbacks here. */
4109 BUG_ON(!list_empty(&init_task.tasks));
4111 mutex_init(&ss->hierarchy_mutex);
4112 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4115 /* this function shouldn't be used with modular subsystems, since they
4116 * need to register a subsys_id, among other things */
4121 * cgroup_load_subsys: load and register a modular subsystem at runtime
4122 * @ss: the subsystem to load
4124 * This function should be called in a modular subsystem's initcall. If the
4125 * subsystem is built as a module, it will be assigned a new subsys_id and set
4126 * up for use. If the subsystem is built-in anyway, work is delegated to the
4127 * simpler cgroup_init_subsys.
4129 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4132 struct cgroup_subsys_state *css;
4134 /* check name and function validity */
4135 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4136 ss->create == NULL || ss->destroy == NULL)
4140 * we don't support callbacks in modular subsystems. this check is
4141 * before the ss->module check for consistency; a subsystem that could
4142 * be a module should still have no callbacks even if the user isn't
4143 * compiling it as one.
4145 if (ss->fork || ss->exit)
4149 * an optionally modular subsystem is built-in: we want to do nothing,
4150 * since cgroup_init_subsys will have already taken care of it.
4152 if (ss->module == NULL) {
4153 /* a few sanity checks */
4154 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4155 BUG_ON(subsys[ss->subsys_id] != ss);
4160 * need to register a subsys id before anything else - for example,
4161 * init_cgroup_css needs it.
4163 mutex_lock(&cgroup_mutex);
4164 /* find the first empty slot in the array */
4165 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4166 if (subsys[i] == NULL)
4169 if (i == CGROUP_SUBSYS_COUNT) {
4170 /* maximum number of subsystems already registered! */
4171 mutex_unlock(&cgroup_mutex);
4174 /* assign ourselves the subsys_id */
4179 * no ss->create seems to need anything important in the ss struct, so
4180 * this can happen first (i.e. before the rootnode attachment).
4182 css = ss->create(dummytop);
4184 /* failure case - need to deassign the subsys[] slot. */
4186 mutex_unlock(&cgroup_mutex);
4187 return PTR_ERR(css);
4190 list_add(&ss->sibling, &rootnode.subsys_list);
4191 ss->root = &rootnode;
4193 /* our new subsystem will be attached to the dummy hierarchy. */
4194 init_cgroup_css(css, ss, dummytop);
4195 /* init_idr must be after init_cgroup_css because it sets css->id. */
4197 int ret = cgroup_init_idr(ss, css);
4199 dummytop->subsys[ss->subsys_id] = NULL;
4200 ss->destroy(dummytop);
4202 mutex_unlock(&cgroup_mutex);
4208 * Now we need to entangle the css into the existing css_sets. unlike
4209 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4210 * will need a new pointer to it; done by iterating the css_set_table.
4211 * furthermore, modifying the existing css_sets will corrupt the hash
4212 * table state, so each changed css_set will need its hash recomputed.
4213 * this is all done under the css_set_lock.
4215 write_lock(&css_set_lock);
4216 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4218 struct hlist_node *node, *tmp;
4219 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4221 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4222 /* skip entries that we already rehashed */
4223 if (cg->subsys[ss->subsys_id])
4225 /* remove existing entry */
4226 hlist_del(&cg->hlist);
4228 cg->subsys[ss->subsys_id] = css;
4229 /* recompute hash and restore entry */
4230 new_bucket = css_set_hash(cg->subsys);
4231 hlist_add_head(&cg->hlist, new_bucket);
4234 write_unlock(&css_set_lock);
4236 mutex_init(&ss->hierarchy_mutex);
4237 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4241 mutex_unlock(&cgroup_mutex);
4244 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4247 * cgroup_unload_subsys: unload a modular subsystem
4248 * @ss: the subsystem to unload
4250 * This function should be called in a modular subsystem's exitcall. When this
4251 * function is invoked, the refcount on the subsystem's module will be 0, so
4252 * the subsystem will not be attached to any hierarchy.
4254 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4256 struct cg_cgroup_link *link;
4257 struct hlist_head *hhead;
4259 BUG_ON(ss->module == NULL);
4262 * we shouldn't be called if the subsystem is in use, and the use of
4263 * try_module_get in parse_cgroupfs_options should ensure that it
4264 * doesn't start being used while we're killing it off.
4266 BUG_ON(ss->root != &rootnode);
4268 mutex_lock(&cgroup_mutex);
4269 /* deassign the subsys_id */
4270 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4271 subsys[ss->subsys_id] = NULL;
4273 /* remove subsystem from rootnode's list of subsystems */
4274 list_del_init(&ss->sibling);
4277 * disentangle the css from all css_sets attached to the dummytop. as
4278 * in loading, we need to pay our respects to the hashtable gods.
4280 write_lock(&css_set_lock);
4281 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4282 struct css_set *cg = link->cg;
4284 hlist_del(&cg->hlist);
4285 BUG_ON(!cg->subsys[ss->subsys_id]);
4286 cg->subsys[ss->subsys_id] = NULL;
4287 hhead = css_set_hash(cg->subsys);
4288 hlist_add_head(&cg->hlist, hhead);
4290 write_unlock(&css_set_lock);
4293 * remove subsystem's css from the dummytop and free it - need to free
4294 * before marking as null because ss->destroy needs the cgrp->subsys
4295 * pointer to find their state. note that this also takes care of
4296 * freeing the css_id.
4298 ss->destroy(dummytop);
4299 dummytop->subsys[ss->subsys_id] = NULL;
4301 mutex_unlock(&cgroup_mutex);
4303 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4306 * cgroup_init_early - cgroup initialization at system boot
4308 * Initialize cgroups at system boot, and initialize any
4309 * subsystems that request early init.
4311 int __init cgroup_init_early(void)
4314 atomic_set(&init_css_set.refcount, 1);
4315 INIT_LIST_HEAD(&init_css_set.cg_links);
4316 INIT_LIST_HEAD(&init_css_set.tasks);
4317 INIT_HLIST_NODE(&init_css_set.hlist);
4319 init_cgroup_root(&rootnode);
4321 init_task.cgroups = &init_css_set;
4323 init_css_set_link.cg = &init_css_set;
4324 init_css_set_link.cgrp = dummytop;
4325 list_add(&init_css_set_link.cgrp_link_list,
4326 &rootnode.top_cgroup.css_sets);
4327 list_add(&init_css_set_link.cg_link_list,
4328 &init_css_set.cg_links);
4330 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4331 INIT_HLIST_HEAD(&css_set_table[i]);
4333 /* at bootup time, we don't worry about modular subsystems */
4334 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4335 struct cgroup_subsys *ss = subsys[i];
4338 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4339 BUG_ON(!ss->create);
4340 BUG_ON(!ss->destroy);
4341 if (ss->subsys_id != i) {
4342 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4343 ss->name, ss->subsys_id);
4348 cgroup_init_subsys(ss);
4354 * cgroup_init - cgroup initialization
4356 * Register cgroup filesystem and /proc file, and initialize
4357 * any subsystems that didn't request early init.
4359 int __init cgroup_init(void)
4363 struct hlist_head *hhead;
4365 err = bdi_init(&cgroup_backing_dev_info);
4369 /* at bootup time, we don't worry about modular subsystems */
4370 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4371 struct cgroup_subsys *ss = subsys[i];
4372 if (!ss->early_init)
4373 cgroup_init_subsys(ss);
4375 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4378 /* Add init_css_set to the hash table */
4379 hhead = css_set_hash(init_css_set.subsys);
4380 hlist_add_head(&init_css_set.hlist, hhead);
4381 BUG_ON(!init_root_id(&rootnode));
4383 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4389 err = register_filesystem(&cgroup_fs_type);
4391 kobject_put(cgroup_kobj);
4395 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4399 bdi_destroy(&cgroup_backing_dev_info);
4405 * proc_cgroup_show()
4406 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4407 * - Used for /proc/<pid>/cgroup.
4408 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4409 * doesn't really matter if tsk->cgroup changes after we read it,
4410 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4411 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4412 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4413 * cgroup to top_cgroup.
4416 /* TODO: Use a proper seq_file iterator */
4417 static int proc_cgroup_show(struct seq_file *m, void *v)
4420 struct task_struct *tsk;
4423 struct cgroupfs_root *root;
4426 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4432 tsk = get_pid_task(pid, PIDTYPE_PID);
4438 mutex_lock(&cgroup_mutex);
4440 for_each_active_root(root) {
4441 struct cgroup_subsys *ss;
4442 struct cgroup *cgrp;
4445 seq_printf(m, "%d:", root->hierarchy_id);
4446 for_each_subsys(root, ss)
4447 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4448 if (strlen(root->name))
4449 seq_printf(m, "%sname=%s", count ? "," : "",
4452 cgrp = task_cgroup_from_root(tsk, root);
4453 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4461 mutex_unlock(&cgroup_mutex);
4462 put_task_struct(tsk);
4469 static int cgroup_open(struct inode *inode, struct file *file)
4471 struct pid *pid = PROC_I(inode)->pid;
4472 return single_open(file, proc_cgroup_show, pid);
4475 const struct file_operations proc_cgroup_operations = {
4476 .open = cgroup_open,
4478 .llseek = seq_lseek,
4479 .release = single_release,
4482 /* Display information about each subsystem and each hierarchy */
4483 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4487 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4489 * ideally we don't want subsystems moving around while we do this.
4490 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4491 * subsys/hierarchy state.
4493 mutex_lock(&cgroup_mutex);
4494 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4495 struct cgroup_subsys *ss = subsys[i];
4498 seq_printf(m, "%s\t%d\t%d\t%d\n",
4499 ss->name, ss->root->hierarchy_id,
4500 ss->root->number_of_cgroups, !ss->disabled);
4502 mutex_unlock(&cgroup_mutex);
4506 static int cgroupstats_open(struct inode *inode, struct file *file)
4508 return single_open(file, proc_cgroupstats_show, NULL);
4511 static const struct file_operations proc_cgroupstats_operations = {
4512 .open = cgroupstats_open,
4514 .llseek = seq_lseek,
4515 .release = single_release,
4519 * cgroup_fork - attach newly forked task to its parents cgroup.
4520 * @child: pointer to task_struct of forking parent process.
4522 * Description: A task inherits its parent's cgroup at fork().
4524 * A pointer to the shared css_set was automatically copied in
4525 * fork.c by dup_task_struct(). However, we ignore that copy, since
4526 * it was not made under the protection of RCU, cgroup_mutex or
4527 * threadgroup_change_begin(), so it might no longer be a valid
4528 * cgroup pointer. cgroup_attach_task() might have already changed
4529 * current->cgroups, allowing the previously referenced cgroup
4530 * group to be removed and freed.
4532 * Outside the pointer validity we also need to process the css_set
4533 * inheritance between threadgoup_change_begin() and
4534 * threadgoup_change_end(), this way there is no leak in any process
4535 * wide migration performed by cgroup_attach_proc() that could otherwise
4536 * miss a thread because it is too early or too late in the fork stage.
4538 * At the point that cgroup_fork() is called, 'current' is the parent
4539 * task, and the passed argument 'child' points to the child task.
4541 void cgroup_fork(struct task_struct *child)
4544 * We don't need to task_lock() current because current->cgroups
4545 * can't be changed concurrently here. The parent obviously hasn't
4546 * exited and called cgroup_exit(), and we are synchronized against
4547 * cgroup migration through threadgroup_change_begin().
4549 child->cgroups = current->cgroups;
4550 get_css_set(child->cgroups);
4551 INIT_LIST_HEAD(&child->cg_list);
4555 * cgroup_fork_callbacks - run fork callbacks
4556 * @child: the new task
4558 * Called on a new task very soon before adding it to the
4559 * tasklist. No need to take any locks since no-one can
4560 * be operating on this task.
4562 void cgroup_fork_callbacks(struct task_struct *child)
4564 if (need_forkexit_callback) {
4567 * forkexit callbacks are only supported for builtin
4568 * subsystems, and the builtin section of the subsys array is
4569 * immutable, so we don't need to lock the subsys array here.
4571 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4572 struct cgroup_subsys *ss = subsys[i];
4580 * cgroup_post_fork - called on a new task after adding it to the task list
4581 * @child: the task in question
4583 * Adds the task to the list running through its css_set if necessary.
4584 * Has to be after the task is visible on the task list in case we race
4585 * with the first call to cgroup_iter_start() - to guarantee that the
4586 * new task ends up on its list.
4588 void cgroup_post_fork(struct task_struct *child)
4591 * use_task_css_set_links is set to 1 before we walk the tasklist
4592 * under the tasklist_lock and we read it here after we added the child
4593 * to the tasklist under the tasklist_lock as well. If the child wasn't
4594 * yet in the tasklist when we walked through it from
4595 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4596 * should be visible now due to the paired locking and barriers implied
4597 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4598 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4601 if (use_task_css_set_links) {
4602 write_lock(&css_set_lock);
4603 if (list_empty(&child->cg_list)) {
4605 * It's safe to use child->cgroups without task_lock()
4606 * here because we are protected through
4607 * threadgroup_change_begin() against concurrent
4608 * css_set change in cgroup_task_migrate(). Also
4609 * the task can't exit at that point until
4610 * wake_up_new_task() is called, so we are protected
4611 * against cgroup_exit() setting child->cgroup to
4614 list_add(&child->cg_list, &child->cgroups->tasks);
4616 write_unlock(&css_set_lock);
4620 * cgroup_exit - detach cgroup from exiting task
4621 * @tsk: pointer to task_struct of exiting process
4622 * @run_callback: run exit callbacks?
4624 * Description: Detach cgroup from @tsk and release it.
4626 * Note that cgroups marked notify_on_release force every task in
4627 * them to take the global cgroup_mutex mutex when exiting.
4628 * This could impact scaling on very large systems. Be reluctant to
4629 * use notify_on_release cgroups where very high task exit scaling
4630 * is required on large systems.
4632 * the_top_cgroup_hack:
4634 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4636 * We call cgroup_exit() while the task is still competent to
4637 * handle notify_on_release(), then leave the task attached to the
4638 * root cgroup in each hierarchy for the remainder of its exit.
4640 * To do this properly, we would increment the reference count on
4641 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4642 * code we would add a second cgroup function call, to drop that
4643 * reference. This would just create an unnecessary hot spot on
4644 * the top_cgroup reference count, to no avail.
4646 * Normally, holding a reference to a cgroup without bumping its
4647 * count is unsafe. The cgroup could go away, or someone could
4648 * attach us to a different cgroup, decrementing the count on
4649 * the first cgroup that we never incremented. But in this case,
4650 * top_cgroup isn't going away, and either task has PF_EXITING set,
4651 * which wards off any cgroup_attach_task() attempts, or task is a failed
4652 * fork, never visible to cgroup_attach_task.
4654 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4660 * Unlink from the css_set task list if necessary.
4661 * Optimistically check cg_list before taking
4664 if (!list_empty(&tsk->cg_list)) {
4665 write_lock(&css_set_lock);
4666 if (!list_empty(&tsk->cg_list))
4667 list_del_init(&tsk->cg_list);
4668 write_unlock(&css_set_lock);
4671 /* Reassign the task to the init_css_set. */
4674 tsk->cgroups = &init_css_set;
4676 if (run_callbacks && need_forkexit_callback) {
4678 * modular subsystems can't use callbacks, so no need to lock
4681 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4682 struct cgroup_subsys *ss = subsys[i];
4684 struct cgroup *old_cgrp =
4685 rcu_dereference_raw(cg->subsys[i])->cgroup;
4686 struct cgroup *cgrp = task_cgroup(tsk, i);
4687 ss->exit(cgrp, old_cgrp, tsk);
4698 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4699 * @cgrp: the cgroup in question
4700 * @task: the task in question
4702 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4705 * If we are sending in dummytop, then presumably we are creating
4706 * the top cgroup in the subsystem.
4708 * Called only by the ns (nsproxy) cgroup.
4710 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4713 struct cgroup *target;
4715 if (cgrp == dummytop)
4718 target = task_cgroup_from_root(task, cgrp->root);
4719 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4720 cgrp = cgrp->parent;
4721 ret = (cgrp == target);
4725 static void check_for_release(struct cgroup *cgrp)
4727 /* All of these checks rely on RCU to keep the cgroup
4728 * structure alive */
4729 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4730 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4731 /* Control Group is currently removeable. If it's not
4732 * already queued for a userspace notification, queue
4734 int need_schedule_work = 0;
4735 raw_spin_lock(&release_list_lock);
4736 if (!cgroup_is_removed(cgrp) &&
4737 list_empty(&cgrp->release_list)) {
4738 list_add(&cgrp->release_list, &release_list);
4739 need_schedule_work = 1;
4741 raw_spin_unlock(&release_list_lock);
4742 if (need_schedule_work)
4743 schedule_work(&release_agent_work);
4747 /* Caller must verify that the css is not for root cgroup */
4748 void __css_get(struct cgroup_subsys_state *css, int count)
4750 atomic_add(count, &css->refcnt);
4751 set_bit(CGRP_RELEASABLE, &css->cgroup->flags);
4753 EXPORT_SYMBOL_GPL(__css_get);
4755 /* Caller must verify that the css is not for root cgroup */
4756 void __css_put(struct cgroup_subsys_state *css, int count)
4758 struct cgroup *cgrp = css->cgroup;
4761 val = atomic_sub_return(count, &css->refcnt);
4763 check_for_release(cgrp);
4764 cgroup_wakeup_rmdir_waiter(cgrp);
4767 WARN_ON_ONCE(val < 1);
4769 EXPORT_SYMBOL_GPL(__css_put);
4772 * Notify userspace when a cgroup is released, by running the
4773 * configured release agent with the name of the cgroup (path
4774 * relative to the root of cgroup file system) as the argument.
4776 * Most likely, this user command will try to rmdir this cgroup.
4778 * This races with the possibility that some other task will be
4779 * attached to this cgroup before it is removed, or that some other
4780 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4781 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4782 * unused, and this cgroup will be reprieved from its death sentence,
4783 * to continue to serve a useful existence. Next time it's released,
4784 * we will get notified again, if it still has 'notify_on_release' set.
4786 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4787 * means only wait until the task is successfully execve()'d. The
4788 * separate release agent task is forked by call_usermodehelper(),
4789 * then control in this thread returns here, without waiting for the
4790 * release agent task. We don't bother to wait because the caller of
4791 * this routine has no use for the exit status of the release agent
4792 * task, so no sense holding our caller up for that.
4794 static void cgroup_release_agent(struct work_struct *work)
4796 BUG_ON(work != &release_agent_work);
4797 mutex_lock(&cgroup_mutex);
4798 raw_spin_lock(&release_list_lock);
4799 while (!list_empty(&release_list)) {
4800 char *argv[3], *envp[3];
4802 char *pathbuf = NULL, *agentbuf = NULL;
4803 struct cgroup *cgrp = list_entry(release_list.next,
4806 list_del_init(&cgrp->release_list);
4807 raw_spin_unlock(&release_list_lock);
4808 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4811 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4813 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4818 argv[i++] = agentbuf;
4819 argv[i++] = pathbuf;
4823 /* minimal command environment */
4824 envp[i++] = "HOME=/";
4825 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4828 /* Drop the lock while we invoke the usermode helper,
4829 * since the exec could involve hitting disk and hence
4830 * be a slow process */
4831 mutex_unlock(&cgroup_mutex);
4832 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4833 mutex_lock(&cgroup_mutex);
4837 raw_spin_lock(&release_list_lock);
4839 raw_spin_unlock(&release_list_lock);
4840 mutex_unlock(&cgroup_mutex);
4843 static int __init cgroup_disable(char *str)
4848 while ((token = strsep(&str, ",")) != NULL) {
4852 * cgroup_disable, being at boot time, can't know about module
4853 * subsystems, so we don't worry about them.
4855 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4856 struct cgroup_subsys *ss = subsys[i];
4858 if (!strcmp(token, ss->name)) {
4860 printk(KERN_INFO "Disabling %s control group"
4861 " subsystem\n", ss->name);
4868 __setup("cgroup_disable=", cgroup_disable);
4871 * Functons for CSS ID.
4875 *To get ID other than 0, this should be called when !cgroup_is_removed().
4877 unsigned short css_id(struct cgroup_subsys_state *css)
4879 struct css_id *cssid;
4882 * This css_id() can return correct value when somone has refcnt
4883 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4884 * it's unchanged until freed.
4886 cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt));
4892 EXPORT_SYMBOL_GPL(css_id);
4894 unsigned short css_depth(struct cgroup_subsys_state *css)
4896 struct css_id *cssid;
4898 cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt));
4901 return cssid->depth;
4904 EXPORT_SYMBOL_GPL(css_depth);
4907 * css_is_ancestor - test "root" css is an ancestor of "child"
4908 * @child: the css to be tested.
4909 * @root: the css supporsed to be an ancestor of the child.
4911 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4912 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4913 * But, considering usual usage, the csses should be valid objects after test.
4914 * Assuming that the caller will do some action to the child if this returns
4915 * returns true, the caller must take "child";s reference count.
4916 * If "child" is valid object and this returns true, "root" is valid, too.
4919 bool css_is_ancestor(struct cgroup_subsys_state *child,
4920 const struct cgroup_subsys_state *root)
4922 struct css_id *child_id;
4923 struct css_id *root_id;
4927 child_id = rcu_dereference(child->id);
4928 root_id = rcu_dereference(root->id);
4931 || (child_id->depth < root_id->depth)
4932 || (child_id->stack[root_id->depth] != root_id->id))
4938 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
4940 struct css_id *id = css->id;
4941 /* When this is called before css_id initialization, id can be NULL */
4945 BUG_ON(!ss->use_id);
4947 rcu_assign_pointer(id->css, NULL);
4948 rcu_assign_pointer(css->id, NULL);
4949 spin_lock(&ss->id_lock);
4950 idr_remove(&ss->idr, id->id);
4951 spin_unlock(&ss->id_lock);
4952 kfree_rcu(id, rcu_head);
4954 EXPORT_SYMBOL_GPL(free_css_id);
4957 * This is called by init or create(). Then, calls to this function are
4958 * always serialized (By cgroup_mutex() at create()).
4961 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
4963 struct css_id *newid;
4964 int myid, error, size;
4966 BUG_ON(!ss->use_id);
4968 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
4969 newid = kzalloc(size, GFP_KERNEL);
4971 return ERR_PTR(-ENOMEM);
4973 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
4977 spin_lock(&ss->id_lock);
4978 /* Don't use 0. allocates an ID of 1-65535 */
4979 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
4980 spin_unlock(&ss->id_lock);
4982 /* Returns error when there are no free spaces for new ID.*/
4987 if (myid > CSS_ID_MAX)
4991 newid->depth = depth;
4995 spin_lock(&ss->id_lock);
4996 idr_remove(&ss->idr, myid);
4997 spin_unlock(&ss->id_lock);
5000 return ERR_PTR(error);
5004 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5005 struct cgroup_subsys_state *rootcss)
5007 struct css_id *newid;
5009 spin_lock_init(&ss->id_lock);
5012 newid = get_new_cssid(ss, 0);
5014 return PTR_ERR(newid);
5016 newid->stack[0] = newid->id;
5017 newid->css = rootcss;
5018 rootcss->id = newid;
5022 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5023 struct cgroup *child)
5025 int subsys_id, i, depth = 0;
5026 struct cgroup_subsys_state *parent_css, *child_css;
5027 struct css_id *child_id, *parent_id;
5029 subsys_id = ss->subsys_id;
5030 parent_css = parent->subsys[subsys_id];
5031 child_css = child->subsys[subsys_id];
5032 parent_id = parent_css->id;
5033 depth = parent_id->depth + 1;
5035 child_id = get_new_cssid(ss, depth);
5036 if (IS_ERR(child_id))
5037 return PTR_ERR(child_id);
5039 for (i = 0; i < depth; i++)
5040 child_id->stack[i] = parent_id->stack[i];
5041 child_id->stack[depth] = child_id->id;
5043 * child_id->css pointer will be set after this cgroup is available
5044 * see cgroup_populate_dir()
5046 rcu_assign_pointer(child_css->id, child_id);
5052 * css_lookup - lookup css by id
5053 * @ss: cgroup subsys to be looked into.
5056 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5057 * NULL if not. Should be called under rcu_read_lock()
5059 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5061 struct css_id *cssid = NULL;
5063 BUG_ON(!ss->use_id);
5064 cssid = idr_find(&ss->idr, id);
5066 if (unlikely(!cssid))
5069 return rcu_dereference(cssid->css);
5071 EXPORT_SYMBOL_GPL(css_lookup);
5074 * css_get_next - lookup next cgroup under specified hierarchy.
5075 * @ss: pointer to subsystem
5076 * @id: current position of iteration.
5077 * @root: pointer to css. search tree under this.
5078 * @foundid: position of found object.
5080 * Search next css under the specified hierarchy of rootid. Calling under
5081 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5083 struct cgroup_subsys_state *
5084 css_get_next(struct cgroup_subsys *ss, int id,
5085 struct cgroup_subsys_state *root, int *foundid)
5087 struct cgroup_subsys_state *ret = NULL;
5090 int rootid = css_id(root);
5091 int depth = css_depth(root);
5096 BUG_ON(!ss->use_id);
5097 WARN_ON_ONCE(!rcu_read_lock_held());
5099 /* fill start point for scan */
5103 * scan next entry from bitmap(tree), tmpid is updated after
5106 tmp = idr_get_next(&ss->idr, &tmpid);
5109 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5110 ret = rcu_dereference(tmp->css);
5116 /* continue to scan from next id */
5123 * get corresponding css from file open on cgroupfs directory
5125 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5127 struct cgroup *cgrp;
5128 struct inode *inode;
5129 struct cgroup_subsys_state *css;
5131 inode = f->f_dentry->d_inode;
5132 /* check in cgroup filesystem dir */
5133 if (inode->i_op != &cgroup_dir_inode_operations)
5134 return ERR_PTR(-EBADF);
5136 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5137 return ERR_PTR(-EINVAL);
5140 cgrp = __d_cgrp(f->f_dentry);
5141 css = cgrp->subsys[id];
5142 return css ? css : ERR_PTR(-ENOENT);
5145 #ifdef CONFIG_CGROUP_DEBUG
5146 static struct cgroup_subsys_state *debug_create(struct cgroup *cont)
5148 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5151 return ERR_PTR(-ENOMEM);
5156 static void debug_destroy(struct cgroup *cont)
5158 kfree(cont->subsys[debug_subsys_id]);
5161 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5163 return atomic_read(&cont->count);
5166 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5168 return cgroup_task_count(cont);
5171 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5173 return (u64)(unsigned long)current->cgroups;
5176 static u64 current_css_set_refcount_read(struct cgroup *cont,
5182 count = atomic_read(¤t->cgroups->refcount);
5187 static int current_css_set_cg_links_read(struct cgroup *cont,
5189 struct seq_file *seq)
5191 struct cg_cgroup_link *link;
5194 read_lock(&css_set_lock);
5196 cg = rcu_dereference(current->cgroups);
5197 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5198 struct cgroup *c = link->cgrp;
5202 name = c->dentry->d_name.name;
5205 seq_printf(seq, "Root %d group %s\n",
5206 c->root->hierarchy_id, name);
5209 read_unlock(&css_set_lock);
5213 #define MAX_TASKS_SHOWN_PER_CSS 25
5214 static int cgroup_css_links_read(struct cgroup *cont,
5216 struct seq_file *seq)
5218 struct cg_cgroup_link *link;
5220 read_lock(&css_set_lock);
5221 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5222 struct css_set *cg = link->cg;
5223 struct task_struct *task;
5225 seq_printf(seq, "css_set %p\n", cg);
5226 list_for_each_entry(task, &cg->tasks, cg_list) {
5227 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5228 seq_puts(seq, " ...\n");
5231 seq_printf(seq, " task %d\n",
5232 task_pid_vnr(task));
5236 read_unlock(&css_set_lock);
5240 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5242 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5245 static struct cftype debug_files[] = {
5247 .name = "cgroup_refcount",
5248 .read_u64 = cgroup_refcount_read,
5251 .name = "taskcount",
5252 .read_u64 = debug_taskcount_read,
5256 .name = "current_css_set",
5257 .read_u64 = current_css_set_read,
5261 .name = "current_css_set_refcount",
5262 .read_u64 = current_css_set_refcount_read,
5266 .name = "current_css_set_cg_links",
5267 .read_seq_string = current_css_set_cg_links_read,
5271 .name = "cgroup_css_links",
5272 .read_seq_string = cgroup_css_links_read,
5276 .name = "releasable",
5277 .read_u64 = releasable_read,
5281 static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
5283 return cgroup_add_files(cont, ss, debug_files,
5284 ARRAY_SIZE(debug_files));
5287 struct cgroup_subsys debug_subsys = {
5289 .create = debug_create,
5290 .destroy = debug_destroy,
5291 .populate = debug_populate,
5292 .subsys_id = debug_subsys_id,
5294 #endif /* CONFIG_CGROUP_DEBUG */