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/ctype.h>
31 #include <linux/errno.h>
33 #include <linux/kernel.h>
34 #include <linux/list.h>
36 #include <linux/mutex.h>
37 #include <linux/mount.h>
38 #include <linux/pagemap.h>
39 #include <linux/proc_fs.h>
40 #include <linux/rcupdate.h>
41 #include <linux/sched.h>
42 #include <linux/backing-dev.h>
43 #include <linux/seq_file.h>
44 #include <linux/slab.h>
45 #include <linux/magic.h>
46 #include <linux/spinlock.h>
47 #include <linux/string.h>
48 #include <linux/sort.h>
49 #include <linux/kmod.h>
50 #include <linux/module.h>
51 #include <linux/delayacct.h>
52 #include <linux/cgroupstats.h>
53 #include <linux/hash.h>
54 #include <linux/namei.h>
55 #include <linux/pid_namespace.h>
56 #include <linux/idr.h>
57 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
58 #include <linux/eventfd.h>
59 #include <linux/poll.h>
60 #include <linux/flex_array.h> /* used in cgroup_attach_proc */
62 #include <asm/atomic.h>
64 static DEFINE_MUTEX(cgroup_mutex);
67 * Generate an array of cgroup subsystem pointers. At boot time, this is
68 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
69 * registered after that. The mutable section of this array is protected by
72 #define SUBSYS(_x) &_x ## _subsys,
73 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
74 #include <linux/cgroup_subsys.h>
77 #define MAX_CGROUP_ROOT_NAMELEN 64
80 * A cgroupfs_root represents the root of a cgroup hierarchy,
81 * and may be associated with a superblock to form an active
84 struct cgroupfs_root {
85 struct super_block *sb;
88 * The bitmask of subsystems intended to be attached to this
91 unsigned long subsys_bits;
93 /* Unique id for this hierarchy. */
96 /* The bitmask of subsystems currently attached to this hierarchy */
97 unsigned long actual_subsys_bits;
99 /* A list running through the attached subsystems */
100 struct list_head subsys_list;
102 /* The root cgroup for this hierarchy */
103 struct cgroup top_cgroup;
105 /* Tracks how many cgroups are currently defined in hierarchy.*/
106 int number_of_cgroups;
108 /* A list running through the active hierarchies */
109 struct list_head root_list;
111 /* Hierarchy-specific flags */
114 /* The path to use for release notifications. */
115 char release_agent_path[PATH_MAX];
117 /* The name for this hierarchy - may be empty */
118 char name[MAX_CGROUP_ROOT_NAMELEN];
122 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
123 * subsystems that are otherwise unattached - it never has more than a
124 * single cgroup, and all tasks are part of that cgroup.
126 static struct cgroupfs_root rootnode;
129 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
130 * cgroup_subsys->use_id != 0.
132 #define CSS_ID_MAX (65535)
135 * The css to which this ID points. This pointer is set to valid value
136 * after cgroup is populated. If cgroup is removed, this will be NULL.
137 * This pointer is expected to be RCU-safe because destroy()
138 * is called after synchronize_rcu(). But for safe use, css_is_removed()
139 * css_tryget() should be used for avoiding race.
141 struct cgroup_subsys_state __rcu *css;
147 * Depth in hierarchy which this ID belongs to.
149 unsigned short depth;
151 * ID is freed by RCU. (and lookup routine is RCU safe.)
153 struct rcu_head rcu_head;
155 * Hierarchy of CSS ID belongs to.
157 unsigned short stack[0]; /* Array of Length (depth+1) */
161 * cgroup_event represents events which userspace want to receive.
163 struct cgroup_event {
165 * Cgroup which the event belongs to.
169 * Control file which the event associated.
173 * eventfd to signal userspace about the event.
175 struct eventfd_ctx *eventfd;
177 * Each of these stored in a list by the cgroup.
179 struct list_head list;
181 * All fields below needed to unregister event when
182 * userspace closes eventfd.
185 wait_queue_head_t *wqh;
187 struct work_struct remove;
190 /* The list of hierarchy roots */
192 static LIST_HEAD(roots);
193 static int root_count;
195 static DEFINE_IDA(hierarchy_ida);
196 static int next_hierarchy_id;
197 static DEFINE_SPINLOCK(hierarchy_id_lock);
199 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
200 #define dummytop (&rootnode.top_cgroup)
202 /* This flag indicates whether tasks in the fork and exit paths should
203 * check for fork/exit handlers to call. This avoids us having to do
204 * extra work in the fork/exit path if none of the subsystems need to
207 static int need_forkexit_callback __read_mostly;
209 #ifdef CONFIG_PROVE_LOCKING
210 int cgroup_lock_is_held(void)
212 return lockdep_is_held(&cgroup_mutex);
214 #else /* #ifdef CONFIG_PROVE_LOCKING */
215 int cgroup_lock_is_held(void)
217 return mutex_is_locked(&cgroup_mutex);
219 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
221 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
223 /* convenient tests for these bits */
224 inline int cgroup_is_removed(const struct cgroup *cgrp)
226 return test_bit(CGRP_REMOVED, &cgrp->flags);
229 /* bits in struct cgroupfs_root flags field */
231 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
234 static int cgroup_is_releasable(const struct cgroup *cgrp)
237 (1 << CGRP_RELEASABLE) |
238 (1 << CGRP_NOTIFY_ON_RELEASE);
239 return (cgrp->flags & bits) == bits;
242 static int notify_on_release(const struct cgroup *cgrp)
244 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
247 static int clone_children(const struct cgroup *cgrp)
249 return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
253 * for_each_subsys() allows you to iterate on each subsystem attached to
254 * an active hierarchy
256 #define for_each_subsys(_root, _ss) \
257 list_for_each_entry(_ss, &_root->subsys_list, sibling)
259 /* for_each_active_root() allows you to iterate across the active hierarchies */
260 #define for_each_active_root(_root) \
261 list_for_each_entry(_root, &roots, root_list)
263 /* the list of cgroups eligible for automatic release. Protected by
264 * release_list_lock */
265 static LIST_HEAD(release_list);
266 static DEFINE_SPINLOCK(release_list_lock);
267 static void cgroup_release_agent(struct work_struct *work);
268 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
269 static void check_for_release(struct cgroup *cgrp);
272 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
273 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
274 * reference to css->refcnt. In general, this refcnt is expected to goes down
277 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
279 DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
281 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
283 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
284 wake_up_all(&cgroup_rmdir_waitq);
287 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
292 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
294 cgroup_wakeup_rmdir_waiter(css->cgroup);
298 /* Link structure for associating css_set objects with cgroups */
299 struct cg_cgroup_link {
301 * List running through cg_cgroup_links associated with a
302 * cgroup, anchored on cgroup->css_sets
304 struct list_head cgrp_link_list;
307 * List running through cg_cgroup_links pointing at a
308 * single css_set object, anchored on css_set->cg_links
310 struct list_head cg_link_list;
314 /* The default css_set - used by init and its children prior to any
315 * hierarchies being mounted. It contains a pointer to the root state
316 * for each subsystem. Also used to anchor the list of css_sets. Not
317 * reference-counted, to improve performance when child cgroups
318 * haven't been created.
321 static struct css_set init_css_set;
322 static struct cg_cgroup_link init_css_set_link;
324 static int cgroup_init_idr(struct cgroup_subsys *ss,
325 struct cgroup_subsys_state *css);
327 /* css_set_lock protects the list of css_set objects, and the
328 * chain of tasks off each css_set. Nests outside task->alloc_lock
329 * due to cgroup_iter_start() */
330 static DEFINE_RWLOCK(css_set_lock);
331 static int css_set_count;
334 * hash table for cgroup groups. This improves the performance to find
335 * an existing css_set. This hash doesn't (currently) take into
336 * account cgroups in empty hierarchies.
338 #define CSS_SET_HASH_BITS 7
339 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
340 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
342 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
346 unsigned long tmp = 0UL;
348 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
349 tmp += (unsigned long)css[i];
350 tmp = (tmp >> 16) ^ tmp;
352 index = hash_long(tmp, CSS_SET_HASH_BITS);
354 return &css_set_table[index];
357 static void free_css_set_work(struct work_struct *work)
359 struct css_set *cg = container_of(work, struct css_set, work);
360 struct cg_cgroup_link *link;
361 struct cg_cgroup_link *saved_link;
363 write_lock(&css_set_lock);
364 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
366 struct cgroup *cgrp = link->cgrp;
367 list_del(&link->cg_link_list);
368 list_del(&link->cgrp_link_list);
369 if (atomic_dec_and_test(&cgrp->count)) {
370 check_for_release(cgrp);
371 cgroup_wakeup_rmdir_waiter(cgrp);
375 write_unlock(&css_set_lock);
380 static void free_css_set_rcu(struct rcu_head *obj)
382 struct css_set *cg = container_of(obj, struct css_set, rcu_head);
384 INIT_WORK(&cg->work, free_css_set_work);
385 schedule_work(&cg->work);
388 /* We don't maintain the lists running through each css_set to its
389 * task until after the first call to cgroup_iter_start(). This
390 * reduces the fork()/exit() overhead for people who have cgroups
391 * compiled into their kernel but not actually in use */
392 static int use_task_css_set_links __read_mostly;
395 * refcounted get/put for css_set objects
397 static inline void get_css_set(struct css_set *cg)
399 atomic_inc(&cg->refcount);
402 static void put_css_set(struct css_set *cg)
405 * Ensure that the refcount doesn't hit zero while any readers
406 * can see it. Similar to atomic_dec_and_lock(), but for an
409 if (atomic_add_unless(&cg->refcount, -1, 1))
411 write_lock(&css_set_lock);
412 if (!atomic_dec_and_test(&cg->refcount)) {
413 write_unlock(&css_set_lock);
417 hlist_del(&cg->hlist);
420 write_unlock(&css_set_lock);
421 call_rcu(&cg->rcu_head, free_css_set_rcu);
425 * compare_css_sets - helper function for find_existing_css_set().
426 * @cg: candidate css_set being tested
427 * @old_cg: existing css_set for a task
428 * @new_cgrp: cgroup that's being entered by the task
429 * @template: desired set of css pointers in css_set (pre-calculated)
431 * Returns true if "cg" matches "old_cg" except for the hierarchy
432 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
434 static bool compare_css_sets(struct css_set *cg,
435 struct css_set *old_cg,
436 struct cgroup *new_cgrp,
437 struct cgroup_subsys_state *template[])
439 struct list_head *l1, *l2;
441 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
442 /* Not all subsystems matched */
447 * Compare cgroup pointers in order to distinguish between
448 * different cgroups in heirarchies with no subsystems. We
449 * could get by with just this check alone (and skip the
450 * memcmp above) but on most setups the memcmp check will
451 * avoid the need for this more expensive check on almost all
456 l2 = &old_cg->cg_links;
458 struct cg_cgroup_link *cgl1, *cgl2;
459 struct cgroup *cg1, *cg2;
463 /* See if we reached the end - both lists are equal length. */
464 if (l1 == &cg->cg_links) {
465 BUG_ON(l2 != &old_cg->cg_links);
468 BUG_ON(l2 == &old_cg->cg_links);
470 /* Locate the cgroups associated with these links. */
471 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
472 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
475 /* Hierarchies should be linked in the same order. */
476 BUG_ON(cg1->root != cg2->root);
479 * If this hierarchy is the hierarchy of the cgroup
480 * that's changing, then we need to check that this
481 * css_set points to the new cgroup; if it's any other
482 * hierarchy, then this css_set should point to the
483 * same cgroup as the old css_set.
485 if (cg1->root == new_cgrp->root) {
497 * find_existing_css_set() is a helper for
498 * find_css_set(), and checks to see whether an existing
499 * css_set is suitable.
501 * oldcg: the cgroup group that we're using before the cgroup
504 * cgrp: the cgroup that we're moving into
506 * template: location in which to build the desired set of subsystem
507 * state objects for the new cgroup group
509 static struct css_set *find_existing_css_set(
510 struct css_set *oldcg,
512 struct cgroup_subsys_state *template[])
515 struct cgroupfs_root *root = cgrp->root;
516 struct hlist_head *hhead;
517 struct hlist_node *node;
521 * Build the set of subsystem state objects that we want to see in the
522 * new css_set. while subsystems can change globally, the entries here
523 * won't change, so no need for locking.
525 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
526 if (root->subsys_bits & (1UL << i)) {
527 /* Subsystem is in this hierarchy. So we want
528 * the subsystem state from the new
530 template[i] = cgrp->subsys[i];
532 /* Subsystem is not in this hierarchy, so we
533 * don't want to change the subsystem state */
534 template[i] = oldcg->subsys[i];
538 hhead = css_set_hash(template);
539 hlist_for_each_entry(cg, node, hhead, hlist) {
540 if (!compare_css_sets(cg, oldcg, cgrp, template))
543 /* This css_set matches what we need */
547 /* No existing cgroup group matched */
551 static void free_cg_links(struct list_head *tmp)
553 struct cg_cgroup_link *link;
554 struct cg_cgroup_link *saved_link;
556 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
557 list_del(&link->cgrp_link_list);
563 * allocate_cg_links() allocates "count" cg_cgroup_link structures
564 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
565 * success or a negative error
567 static int allocate_cg_links(int count, struct list_head *tmp)
569 struct cg_cgroup_link *link;
572 for (i = 0; i < count; i++) {
573 link = kmalloc(sizeof(*link), GFP_KERNEL);
578 list_add(&link->cgrp_link_list, tmp);
584 * link_css_set - a helper function to link a css_set to a cgroup
585 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
586 * @cg: the css_set to be linked
587 * @cgrp: the destination cgroup
589 static void link_css_set(struct list_head *tmp_cg_links,
590 struct css_set *cg, struct cgroup *cgrp)
592 struct cg_cgroup_link *link;
594 BUG_ON(list_empty(tmp_cg_links));
595 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
599 atomic_inc(&cgrp->count);
600 list_move(&link->cgrp_link_list, &cgrp->css_sets);
602 * Always add links to the tail of the list so that the list
603 * is sorted by order of hierarchy creation
605 list_add_tail(&link->cg_link_list, &cg->cg_links);
609 * find_css_set() takes an existing cgroup group and a
610 * cgroup object, and returns a css_set object that's
611 * equivalent to the old group, but with the given cgroup
612 * substituted into the appropriate hierarchy. Must be called with
615 static struct css_set *find_css_set(
616 struct css_set *oldcg, struct cgroup *cgrp)
619 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
621 struct list_head tmp_cg_links;
623 struct hlist_head *hhead;
624 struct cg_cgroup_link *link;
626 /* First see if we already have a cgroup group that matches
628 read_lock(&css_set_lock);
629 res = find_existing_css_set(oldcg, cgrp, template);
632 read_unlock(&css_set_lock);
637 res = kmalloc(sizeof(*res), GFP_KERNEL);
641 /* Allocate all the cg_cgroup_link objects that we'll need */
642 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
647 atomic_set(&res->refcount, 1);
648 INIT_LIST_HEAD(&res->cg_links);
649 INIT_LIST_HEAD(&res->tasks);
650 INIT_HLIST_NODE(&res->hlist);
652 /* Copy the set of subsystem state objects generated in
653 * find_existing_css_set() */
654 memcpy(res->subsys, template, sizeof(res->subsys));
656 write_lock(&css_set_lock);
657 /* Add reference counts and links from the new css_set. */
658 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
659 struct cgroup *c = link->cgrp;
660 if (c->root == cgrp->root)
662 link_css_set(&tmp_cg_links, res, c);
665 BUG_ON(!list_empty(&tmp_cg_links));
669 /* Add this cgroup group to the hash table */
670 hhead = css_set_hash(res->subsys);
671 hlist_add_head(&res->hlist, hhead);
673 write_unlock(&css_set_lock);
679 * Return the cgroup for "task" from the given hierarchy. Must be
680 * called with cgroup_mutex held.
682 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
683 struct cgroupfs_root *root)
686 struct cgroup *res = NULL;
688 BUG_ON(!mutex_is_locked(&cgroup_mutex));
689 read_lock(&css_set_lock);
691 * No need to lock the task - since we hold cgroup_mutex the
692 * task can't change groups, so the only thing that can happen
693 * is that it exits and its css is set back to init_css_set.
696 if (css == &init_css_set) {
697 res = &root->top_cgroup;
699 struct cg_cgroup_link *link;
700 list_for_each_entry(link, &css->cg_links, cg_link_list) {
701 struct cgroup *c = link->cgrp;
702 if (c->root == root) {
708 read_unlock(&css_set_lock);
714 * There is one global cgroup mutex. We also require taking
715 * task_lock() when dereferencing a task's cgroup subsys pointers.
716 * See "The task_lock() exception", at the end of this comment.
718 * A task must hold cgroup_mutex to modify cgroups.
720 * Any task can increment and decrement the count field without lock.
721 * So in general, code holding cgroup_mutex can't rely on the count
722 * field not changing. However, if the count goes to zero, then only
723 * cgroup_attach_task() can increment it again. Because a count of zero
724 * means that no tasks are currently attached, therefore there is no
725 * way a task attached to that cgroup can fork (the other way to
726 * increment the count). So code holding cgroup_mutex can safely
727 * assume that if the count is zero, it will stay zero. Similarly, if
728 * a task holds cgroup_mutex on a cgroup with zero count, it
729 * knows that the cgroup won't be removed, as cgroup_rmdir()
732 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
733 * (usually) take cgroup_mutex. These are the two most performance
734 * critical pieces of code here. The exception occurs on cgroup_exit(),
735 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
736 * is taken, and if the cgroup count is zero, a usermode call made
737 * to the release agent with the name of the cgroup (path relative to
738 * the root of cgroup file system) as the argument.
740 * A cgroup can only be deleted if both its 'count' of using tasks
741 * is zero, and its list of 'children' cgroups is empty. Since all
742 * tasks in the system use _some_ cgroup, and since there is always at
743 * least one task in the system (init, pid == 1), therefore, top_cgroup
744 * always has either children cgroups and/or using tasks. So we don't
745 * need a special hack to ensure that top_cgroup cannot be deleted.
747 * The task_lock() exception
749 * The need for this exception arises from the action of
750 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
751 * another. It does so using cgroup_mutex, however there are
752 * several performance critical places that need to reference
753 * task->cgroups without the expense of grabbing a system global
754 * mutex. Therefore except as noted below, when dereferencing or, as
755 * in cgroup_attach_task(), modifying a task's cgroups pointer we use
756 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
757 * the task_struct routinely used for such matters.
759 * P.S. One more locking exception. RCU is used to guard the
760 * update of a tasks cgroup pointer by cgroup_attach_task()
764 * cgroup_lock - lock out any changes to cgroup structures
767 void cgroup_lock(void)
769 mutex_lock(&cgroup_mutex);
771 EXPORT_SYMBOL_GPL(cgroup_lock);
774 * cgroup_unlock - release lock on cgroup changes
776 * Undo the lock taken in a previous cgroup_lock() call.
778 void cgroup_unlock(void)
780 mutex_unlock(&cgroup_mutex);
782 EXPORT_SYMBOL_GPL(cgroup_unlock);
785 * A couple of forward declarations required, due to cyclic reference loop:
786 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
787 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
791 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
792 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, struct nameidata *);
793 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
794 static int cgroup_populate_dir(struct cgroup *cgrp);
795 static const struct inode_operations cgroup_dir_inode_operations;
796 static const struct file_operations proc_cgroupstats_operations;
798 static struct backing_dev_info cgroup_backing_dev_info = {
800 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
803 static int alloc_css_id(struct cgroup_subsys *ss,
804 struct cgroup *parent, struct cgroup *child);
806 static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
808 struct inode *inode = new_inode(sb);
811 inode->i_ino = get_next_ino();
812 inode->i_mode = mode;
813 inode->i_uid = current_fsuid();
814 inode->i_gid = current_fsgid();
815 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
816 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
822 * Call subsys's pre_destroy handler.
823 * This is called before css refcnt check.
825 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
827 struct cgroup_subsys *ss;
830 for_each_subsys(cgrp->root, ss)
831 if (ss->pre_destroy) {
832 ret = ss->pre_destroy(ss, cgrp);
840 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
842 /* is dentry a directory ? if so, kfree() associated cgroup */
843 if (S_ISDIR(inode->i_mode)) {
844 struct cgroup *cgrp = dentry->d_fsdata;
845 struct cgroup_subsys *ss;
846 BUG_ON(!(cgroup_is_removed(cgrp)));
847 /* It's possible for external users to be holding css
848 * reference counts on a cgroup; css_put() needs to
849 * be able to access the cgroup after decrementing
850 * the reference count in order to know if it needs to
851 * queue the cgroup to be handled by the release
855 mutex_lock(&cgroup_mutex);
857 * Release the subsystem state objects.
859 for_each_subsys(cgrp->root, ss)
860 ss->destroy(ss, cgrp);
862 cgrp->root->number_of_cgroups--;
863 mutex_unlock(&cgroup_mutex);
866 * Drop the active superblock reference that we took when we
869 deactivate_super(cgrp->root->sb);
872 * if we're getting rid of the cgroup, refcount should ensure
873 * that there are no pidlists left.
875 BUG_ON(!list_empty(&cgrp->pidlists));
877 kfree_rcu(cgrp, rcu_head);
882 static int cgroup_delete(const struct dentry *d)
887 static void remove_dir(struct dentry *d)
889 struct dentry *parent = dget(d->d_parent);
892 simple_rmdir(parent->d_inode, d);
896 static void cgroup_clear_directory(struct dentry *dentry)
898 struct list_head *node;
900 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
901 spin_lock(&dentry->d_lock);
902 node = dentry->d_subdirs.next;
903 while (node != &dentry->d_subdirs) {
904 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
906 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
909 /* This should never be called on a cgroup
910 * directory with child cgroups */
911 BUG_ON(d->d_inode->i_mode & S_IFDIR);
913 spin_unlock(&d->d_lock);
914 spin_unlock(&dentry->d_lock);
916 simple_unlink(dentry->d_inode, d);
918 spin_lock(&dentry->d_lock);
920 spin_unlock(&d->d_lock);
921 node = dentry->d_subdirs.next;
923 spin_unlock(&dentry->d_lock);
927 * NOTE : the dentry must have been dget()'ed
929 static void cgroup_d_remove_dir(struct dentry *dentry)
931 struct dentry *parent;
933 cgroup_clear_directory(dentry);
935 parent = dentry->d_parent;
936 spin_lock(&parent->d_lock);
937 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
938 list_del_init(&dentry->d_u.d_child);
939 spin_unlock(&dentry->d_lock);
940 spin_unlock(&parent->d_lock);
945 * Call with cgroup_mutex held. Drops reference counts on modules, including
946 * any duplicate ones that parse_cgroupfs_options took. If this function
947 * returns an error, no reference counts are touched.
949 static int rebind_subsystems(struct cgroupfs_root *root,
950 unsigned long final_bits)
952 unsigned long added_bits, removed_bits;
953 struct cgroup *cgrp = &root->top_cgroup;
956 BUG_ON(!mutex_is_locked(&cgroup_mutex));
958 removed_bits = root->actual_subsys_bits & ~final_bits;
959 added_bits = final_bits & ~root->actual_subsys_bits;
960 /* Check that any added subsystems are currently free */
961 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
962 unsigned long bit = 1UL << i;
963 struct cgroup_subsys *ss = subsys[i];
964 if (!(bit & added_bits))
967 * Nobody should tell us to do a subsys that doesn't exist:
968 * parse_cgroupfs_options should catch that case and refcounts
969 * ensure that subsystems won't disappear once selected.
972 if (ss->root != &rootnode) {
973 /* Subsystem isn't free */
978 /* Currently we don't handle adding/removing subsystems when
979 * any child cgroups exist. This is theoretically supportable
980 * but involves complex error handling, so it's being left until
982 if (root->number_of_cgroups > 1)
985 /* Process each subsystem */
986 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
987 struct cgroup_subsys *ss = subsys[i];
988 unsigned long bit = 1UL << i;
989 if (bit & added_bits) {
990 /* We're binding this subsystem to this hierarchy */
992 BUG_ON(cgrp->subsys[i]);
993 BUG_ON(!dummytop->subsys[i]);
994 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
995 mutex_lock(&ss->hierarchy_mutex);
996 cgrp->subsys[i] = dummytop->subsys[i];
997 cgrp->subsys[i]->cgroup = cgrp;
998 list_move(&ss->sibling, &root->subsys_list);
1002 mutex_unlock(&ss->hierarchy_mutex);
1003 /* refcount was already taken, and we're keeping it */
1004 } else if (bit & removed_bits) {
1005 /* We're removing this subsystem */
1007 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1008 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1009 mutex_lock(&ss->hierarchy_mutex);
1011 ss->bind(ss, dummytop);
1012 dummytop->subsys[i]->cgroup = dummytop;
1013 cgrp->subsys[i] = NULL;
1014 subsys[i]->root = &rootnode;
1015 list_move(&ss->sibling, &rootnode.subsys_list);
1016 mutex_unlock(&ss->hierarchy_mutex);
1017 /* subsystem is now free - drop reference on module */
1018 module_put(ss->module);
1019 } else if (bit & final_bits) {
1020 /* Subsystem state should already exist */
1022 BUG_ON(!cgrp->subsys[i]);
1024 * a refcount was taken, but we already had one, so
1025 * drop the extra reference.
1027 module_put(ss->module);
1028 #ifdef CONFIG_MODULE_UNLOAD
1029 BUG_ON(ss->module && !module_refcount(ss->module));
1032 /* Subsystem state shouldn't exist */
1033 BUG_ON(cgrp->subsys[i]);
1036 root->subsys_bits = root->actual_subsys_bits = final_bits;
1042 static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
1044 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
1045 struct cgroup_subsys *ss;
1047 mutex_lock(&cgroup_mutex);
1048 for_each_subsys(root, ss)
1049 seq_printf(seq, ",%s", ss->name);
1050 if (test_bit(ROOT_NOPREFIX, &root->flags))
1051 seq_puts(seq, ",noprefix");
1052 if (strlen(root->release_agent_path))
1053 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1054 if (clone_children(&root->top_cgroup))
1055 seq_puts(seq, ",clone_children");
1056 if (strlen(root->name))
1057 seq_printf(seq, ",name=%s", root->name);
1058 mutex_unlock(&cgroup_mutex);
1062 struct cgroup_sb_opts {
1063 unsigned long subsys_bits;
1064 unsigned long flags;
1065 char *release_agent;
1066 bool clone_children;
1068 /* User explicitly requested empty subsystem */
1071 struct cgroupfs_root *new_root;
1076 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1077 * with cgroup_mutex held to protect the subsys[] array. This function takes
1078 * refcounts on subsystems to be used, unless it returns error, in which case
1079 * no refcounts are taken.
1081 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1083 char *token, *o = data;
1084 bool all_ss = false, one_ss = false;
1085 unsigned long mask = (unsigned long)-1;
1087 bool module_pin_failed = false;
1089 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1091 #ifdef CONFIG_CPUSETS
1092 mask = ~(1UL << cpuset_subsys_id);
1095 memset(opts, 0, sizeof(*opts));
1097 while ((token = strsep(&o, ",")) != NULL) {
1100 if (!strcmp(token, "none")) {
1101 /* Explicitly have no subsystems */
1105 if (!strcmp(token, "all")) {
1106 /* Mutually exclusive option 'all' + subsystem name */
1112 if (!strcmp(token, "noprefix")) {
1113 set_bit(ROOT_NOPREFIX, &opts->flags);
1116 if (!strcmp(token, "clone_children")) {
1117 opts->clone_children = true;
1120 if (!strncmp(token, "release_agent=", 14)) {
1121 /* Specifying two release agents is forbidden */
1122 if (opts->release_agent)
1124 opts->release_agent =
1125 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1126 if (!opts->release_agent)
1130 if (!strncmp(token, "name=", 5)) {
1131 const char *name = token + 5;
1132 /* Can't specify an empty name */
1135 /* Must match [\w.-]+ */
1136 for (i = 0; i < strlen(name); i++) {
1140 if ((c == '.') || (c == '-') || (c == '_'))
1144 /* Specifying two names is forbidden */
1147 opts->name = kstrndup(name,
1148 MAX_CGROUP_ROOT_NAMELEN - 1,
1156 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1157 struct cgroup_subsys *ss = subsys[i];
1160 if (strcmp(token, ss->name))
1165 /* Mutually exclusive option 'all' + subsystem name */
1168 set_bit(i, &opts->subsys_bits);
1173 if (i == CGROUP_SUBSYS_COUNT)
1178 * If the 'all' option was specified select all the subsystems,
1179 * otherwise 'all, 'none' and a subsystem name options were not
1180 * specified, let's default to 'all'
1182 if (all_ss || (!all_ss && !one_ss && !opts->none)) {
1183 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1184 struct cgroup_subsys *ss = subsys[i];
1189 set_bit(i, &opts->subsys_bits);
1193 /* Consistency checks */
1196 * Option noprefix was introduced just for backward compatibility
1197 * with the old cpuset, so we allow noprefix only if mounting just
1198 * the cpuset subsystem.
1200 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1201 (opts->subsys_bits & mask))
1205 /* Can't specify "none" and some subsystems */
1206 if (opts->subsys_bits && opts->none)
1210 * We either have to specify by name or by subsystems. (So all
1211 * empty hierarchies must have a name).
1213 if (!opts->subsys_bits && !opts->name)
1217 * Grab references on all the modules we'll need, so the subsystems
1218 * don't dance around before rebind_subsystems attaches them. This may
1219 * take duplicate reference counts on a subsystem that's already used,
1220 * but rebind_subsystems handles this case.
1222 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1223 unsigned long bit = 1UL << i;
1225 if (!(bit & opts->subsys_bits))
1227 if (!try_module_get(subsys[i]->module)) {
1228 module_pin_failed = true;
1232 if (module_pin_failed) {
1234 * oops, one of the modules was going away. this means that we
1235 * raced with a module_delete call, and to the user this is
1236 * essentially a "subsystem doesn't exist" case.
1238 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1239 /* drop refcounts only on the ones we took */
1240 unsigned long bit = 1UL << i;
1242 if (!(bit & opts->subsys_bits))
1244 module_put(subsys[i]->module);
1252 static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1255 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1256 unsigned long bit = 1UL << i;
1258 if (!(bit & subsys_bits))
1260 module_put(subsys[i]->module);
1264 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1267 struct cgroupfs_root *root = sb->s_fs_info;
1268 struct cgroup *cgrp = &root->top_cgroup;
1269 struct cgroup_sb_opts opts;
1271 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1272 mutex_lock(&cgroup_mutex);
1274 /* See what subsystems are wanted */
1275 ret = parse_cgroupfs_options(data, &opts);
1279 /* Don't allow flags or name to change at remount */
1280 if (opts.flags != root->flags ||
1281 (opts.name && strcmp(opts.name, root->name))) {
1283 drop_parsed_module_refcounts(opts.subsys_bits);
1287 ret = rebind_subsystems(root, opts.subsys_bits);
1289 drop_parsed_module_refcounts(opts.subsys_bits);
1293 /* (re)populate subsystem files */
1294 cgroup_populate_dir(cgrp);
1296 if (opts.release_agent)
1297 strcpy(root->release_agent_path, opts.release_agent);
1299 kfree(opts.release_agent);
1301 mutex_unlock(&cgroup_mutex);
1302 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1306 static const struct super_operations cgroup_ops = {
1307 .statfs = simple_statfs,
1308 .drop_inode = generic_delete_inode,
1309 .show_options = cgroup_show_options,
1310 .remount_fs = cgroup_remount,
1313 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1315 INIT_LIST_HEAD(&cgrp->sibling);
1316 INIT_LIST_HEAD(&cgrp->children);
1317 INIT_LIST_HEAD(&cgrp->css_sets);
1318 INIT_LIST_HEAD(&cgrp->release_list);
1319 INIT_LIST_HEAD(&cgrp->pidlists);
1320 mutex_init(&cgrp->pidlist_mutex);
1321 INIT_LIST_HEAD(&cgrp->event_list);
1322 spin_lock_init(&cgrp->event_list_lock);
1325 static void init_cgroup_root(struct cgroupfs_root *root)
1327 struct cgroup *cgrp = &root->top_cgroup;
1328 INIT_LIST_HEAD(&root->subsys_list);
1329 INIT_LIST_HEAD(&root->root_list);
1330 root->number_of_cgroups = 1;
1332 cgrp->top_cgroup = cgrp;
1333 init_cgroup_housekeeping(cgrp);
1336 static bool init_root_id(struct cgroupfs_root *root)
1341 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1343 spin_lock(&hierarchy_id_lock);
1344 /* Try to allocate the next unused ID */
1345 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1346 &root->hierarchy_id);
1348 /* Try again starting from 0 */
1349 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1351 next_hierarchy_id = root->hierarchy_id + 1;
1352 } else if (ret != -EAGAIN) {
1353 /* Can only get here if the 31-bit IDR is full ... */
1356 spin_unlock(&hierarchy_id_lock);
1361 static int cgroup_test_super(struct super_block *sb, void *data)
1363 struct cgroup_sb_opts *opts = data;
1364 struct cgroupfs_root *root = sb->s_fs_info;
1366 /* If we asked for a name then it must match */
1367 if (opts->name && strcmp(opts->name, root->name))
1371 * If we asked for subsystems (or explicitly for no
1372 * subsystems) then they must match
1374 if ((opts->subsys_bits || opts->none)
1375 && (opts->subsys_bits != root->subsys_bits))
1381 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1383 struct cgroupfs_root *root;
1385 if (!opts->subsys_bits && !opts->none)
1388 root = kzalloc(sizeof(*root), GFP_KERNEL);
1390 return ERR_PTR(-ENOMEM);
1392 if (!init_root_id(root)) {
1394 return ERR_PTR(-ENOMEM);
1396 init_cgroup_root(root);
1398 root->subsys_bits = opts->subsys_bits;
1399 root->flags = opts->flags;
1400 if (opts->release_agent)
1401 strcpy(root->release_agent_path, opts->release_agent);
1403 strcpy(root->name, opts->name);
1404 if (opts->clone_children)
1405 set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1409 static void cgroup_drop_root(struct cgroupfs_root *root)
1414 BUG_ON(!root->hierarchy_id);
1415 spin_lock(&hierarchy_id_lock);
1416 ida_remove(&hierarchy_ida, root->hierarchy_id);
1417 spin_unlock(&hierarchy_id_lock);
1421 static int cgroup_set_super(struct super_block *sb, void *data)
1424 struct cgroup_sb_opts *opts = data;
1426 /* If we don't have a new root, we can't set up a new sb */
1427 if (!opts->new_root)
1430 BUG_ON(!opts->subsys_bits && !opts->none);
1432 ret = set_anon_super(sb, NULL);
1436 sb->s_fs_info = opts->new_root;
1437 opts->new_root->sb = sb;
1439 sb->s_blocksize = PAGE_CACHE_SIZE;
1440 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1441 sb->s_magic = CGROUP_SUPER_MAGIC;
1442 sb->s_op = &cgroup_ops;
1447 static int cgroup_get_rootdir(struct super_block *sb)
1449 static const struct dentry_operations cgroup_dops = {
1450 .d_iput = cgroup_diput,
1451 .d_delete = cgroup_delete,
1454 struct inode *inode =
1455 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1456 struct dentry *dentry;
1461 inode->i_fop = &simple_dir_operations;
1462 inode->i_op = &cgroup_dir_inode_operations;
1463 /* directories start off with i_nlink == 2 (for "." entry) */
1465 dentry = d_alloc_root(inode);
1470 sb->s_root = dentry;
1471 /* for everything else we want ->d_op set */
1472 sb->s_d_op = &cgroup_dops;
1476 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1477 int flags, const char *unused_dev_name,
1480 struct cgroup_sb_opts opts;
1481 struct cgroupfs_root *root;
1483 struct super_block *sb;
1484 struct cgroupfs_root *new_root;
1486 /* First find the desired set of subsystems */
1487 mutex_lock(&cgroup_mutex);
1488 ret = parse_cgroupfs_options(data, &opts);
1489 mutex_unlock(&cgroup_mutex);
1494 * Allocate a new cgroup root. We may not need it if we're
1495 * reusing an existing hierarchy.
1497 new_root = cgroup_root_from_opts(&opts);
1498 if (IS_ERR(new_root)) {
1499 ret = PTR_ERR(new_root);
1502 opts.new_root = new_root;
1504 /* Locate an existing or new sb for this hierarchy */
1505 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1508 cgroup_drop_root(opts.new_root);
1512 root = sb->s_fs_info;
1514 if (root == opts.new_root) {
1515 /* We used the new root structure, so this is a new hierarchy */
1516 struct list_head tmp_cg_links;
1517 struct cgroup *root_cgrp = &root->top_cgroup;
1518 struct inode *inode;
1519 struct cgroupfs_root *existing_root;
1522 BUG_ON(sb->s_root != NULL);
1524 ret = cgroup_get_rootdir(sb);
1526 goto drop_new_super;
1527 inode = sb->s_root->d_inode;
1529 mutex_lock(&inode->i_mutex);
1530 mutex_lock(&cgroup_mutex);
1532 if (strlen(root->name)) {
1533 /* Check for name clashes with existing mounts */
1534 for_each_active_root(existing_root) {
1535 if (!strcmp(existing_root->name, root->name)) {
1537 mutex_unlock(&cgroup_mutex);
1538 mutex_unlock(&inode->i_mutex);
1539 goto drop_new_super;
1545 * We're accessing css_set_count without locking
1546 * css_set_lock here, but that's OK - it can only be
1547 * increased by someone holding cgroup_lock, and
1548 * that's us. The worst that can happen is that we
1549 * have some link structures left over
1551 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1553 mutex_unlock(&cgroup_mutex);
1554 mutex_unlock(&inode->i_mutex);
1555 goto drop_new_super;
1558 ret = rebind_subsystems(root, root->subsys_bits);
1559 if (ret == -EBUSY) {
1560 mutex_unlock(&cgroup_mutex);
1561 mutex_unlock(&inode->i_mutex);
1562 free_cg_links(&tmp_cg_links);
1563 goto drop_new_super;
1566 * There must be no failure case after here, since rebinding
1567 * takes care of subsystems' refcounts, which are explicitly
1568 * dropped in the failure exit path.
1571 /* EBUSY should be the only error here */
1574 list_add(&root->root_list, &roots);
1577 sb->s_root->d_fsdata = root_cgrp;
1578 root->top_cgroup.dentry = sb->s_root;
1580 /* Link the top cgroup in this hierarchy into all
1581 * the css_set objects */
1582 write_lock(&css_set_lock);
1583 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1584 struct hlist_head *hhead = &css_set_table[i];
1585 struct hlist_node *node;
1588 hlist_for_each_entry(cg, node, hhead, hlist)
1589 link_css_set(&tmp_cg_links, cg, root_cgrp);
1591 write_unlock(&css_set_lock);
1593 free_cg_links(&tmp_cg_links);
1595 BUG_ON(!list_empty(&root_cgrp->sibling));
1596 BUG_ON(!list_empty(&root_cgrp->children));
1597 BUG_ON(root->number_of_cgroups != 1);
1599 cgroup_populate_dir(root_cgrp);
1600 mutex_unlock(&cgroup_mutex);
1601 mutex_unlock(&inode->i_mutex);
1604 * We re-used an existing hierarchy - the new root (if
1605 * any) is not needed
1607 cgroup_drop_root(opts.new_root);
1608 /* no subsys rebinding, so refcounts don't change */
1609 drop_parsed_module_refcounts(opts.subsys_bits);
1612 kfree(opts.release_agent);
1614 return dget(sb->s_root);
1617 deactivate_locked_super(sb);
1619 drop_parsed_module_refcounts(opts.subsys_bits);
1621 kfree(opts.release_agent);
1623 return ERR_PTR(ret);
1626 static void cgroup_kill_sb(struct super_block *sb) {
1627 struct cgroupfs_root *root = sb->s_fs_info;
1628 struct cgroup *cgrp = &root->top_cgroup;
1630 struct cg_cgroup_link *link;
1631 struct cg_cgroup_link *saved_link;
1635 BUG_ON(root->number_of_cgroups != 1);
1636 BUG_ON(!list_empty(&cgrp->children));
1637 BUG_ON(!list_empty(&cgrp->sibling));
1639 mutex_lock(&cgroup_mutex);
1641 /* Rebind all subsystems back to the default hierarchy */
1642 ret = rebind_subsystems(root, 0);
1643 /* Shouldn't be able to fail ... */
1647 * Release all the links from css_sets to this hierarchy's
1650 write_lock(&css_set_lock);
1652 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1654 list_del(&link->cg_link_list);
1655 list_del(&link->cgrp_link_list);
1658 write_unlock(&css_set_lock);
1660 if (!list_empty(&root->root_list)) {
1661 list_del(&root->root_list);
1665 mutex_unlock(&cgroup_mutex);
1667 kill_litter_super(sb);
1668 cgroup_drop_root(root);
1671 static struct file_system_type cgroup_fs_type = {
1673 .mount = cgroup_mount,
1674 .kill_sb = cgroup_kill_sb,
1677 static struct kobject *cgroup_kobj;
1679 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1681 return dentry->d_fsdata;
1684 static inline struct cftype *__d_cft(struct dentry *dentry)
1686 return dentry->d_fsdata;
1690 * cgroup_path - generate the path of a cgroup
1691 * @cgrp: the cgroup in question
1692 * @buf: the buffer to write the path into
1693 * @buflen: the length of the buffer
1695 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1696 * reference. Writes path of cgroup into buf. Returns 0 on success,
1699 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1702 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1703 rcu_read_lock_held() ||
1704 cgroup_lock_is_held());
1706 if (!dentry || cgrp == dummytop) {
1708 * Inactive subsystems have no dentry for their root
1715 start = buf + buflen;
1719 int len = dentry->d_name.len;
1721 if ((start -= len) < buf)
1722 return -ENAMETOOLONG;
1723 memcpy(start, dentry->d_name.name, len);
1724 cgrp = cgrp->parent;
1728 dentry = rcu_dereference_check(cgrp->dentry,
1729 rcu_read_lock_held() ||
1730 cgroup_lock_is_held());
1734 return -ENAMETOOLONG;
1737 memmove(buf, start, buf + buflen - start);
1740 EXPORT_SYMBOL_GPL(cgroup_path);
1743 * cgroup_task_migrate - move a task from one cgroup to another.
1745 * 'guarantee' is set if the caller promises that a new css_set for the task
1746 * will already exist. If not set, this function might sleep, and can fail with
1747 * -ENOMEM. Otherwise, it can only fail with -ESRCH.
1749 static int cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1750 struct task_struct *tsk, bool guarantee)
1752 struct css_set *oldcg;
1753 struct css_set *newcg;
1756 * get old css_set. we need to take task_lock and refcount it, because
1757 * an exiting task can change its css_set to init_css_set and drop its
1758 * old one without taking cgroup_mutex.
1761 oldcg = tsk->cgroups;
1765 /* locate or allocate a new css_set for this task. */
1767 /* we know the css_set we want already exists. */
1768 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
1769 read_lock(&css_set_lock);
1770 newcg = find_existing_css_set(oldcg, cgrp, template);
1773 read_unlock(&css_set_lock);
1776 /* find_css_set will give us newcg already referenced. */
1777 newcg = find_css_set(oldcg, cgrp);
1785 /* if PF_EXITING is set, the tsk->cgroups pointer is no longer safe. */
1787 if (tsk->flags & PF_EXITING) {
1792 rcu_assign_pointer(tsk->cgroups, newcg);
1795 /* Update the css_set linked lists if we're using them */
1796 write_lock(&css_set_lock);
1797 if (!list_empty(&tsk->cg_list))
1798 list_move(&tsk->cg_list, &newcg->tasks);
1799 write_unlock(&css_set_lock);
1802 * We just gained a reference on oldcg by taking it from the task. As
1803 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1804 * it here; it will be freed under RCU.
1808 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1813 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1814 * @cgrp: the cgroup the task is attaching to
1815 * @tsk: the task to be attached
1817 * Call holding cgroup_mutex. May take task_lock of
1818 * the task 'tsk' during call.
1820 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1823 struct cgroup_subsys *ss, *failed_ss = NULL;
1824 struct cgroup *oldcgrp;
1825 struct cgroupfs_root *root = cgrp->root;
1828 /* Nothing to do if the task is already in that cgroup */
1829 oldcgrp = task_cgroup_from_root(tsk, root);
1830 if (cgrp == oldcgrp)
1833 for_each_subsys(root, ss) {
1834 if (ss->can_attach) {
1835 retval = ss->can_attach(ss, cgrp, tsk);
1838 * Remember on which subsystem the can_attach()
1839 * failed, so that we only call cancel_attach()
1840 * against the subsystems whose can_attach()
1841 * succeeded. (See below)
1847 if (ss->can_attach_task) {
1848 retval = ss->can_attach_task(cgrp, tsk);
1861 retval = cgroup_task_migrate(cgrp, oldcgrp, tsk, false);
1865 for_each_subsys(root, ss) {
1867 ss->pre_attach(cgrp);
1868 if (ss->attach_task)
1869 ss->attach_task(cgrp, tsk);
1871 ss->attach(ss, cgrp, oldcgrp, tsk);
1873 set_bit(CGRP_RELEASABLE, &cgrp->flags);
1874 /* put_css_set will not destroy cg until after an RCU grace period */
1878 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1879 * is no longer empty.
1881 cgroup_wakeup_rmdir_waiter(cgrp);
1884 for_each_subsys(root, ss) {
1885 if (ss == failed_ss)
1887 * This subsystem was the one that failed the
1888 * can_attach() check earlier, so we don't need
1889 * to call cancel_attach() against it or any
1890 * remaining subsystems.
1893 if (ss->cancel_attach)
1894 ss->cancel_attach(ss, cgrp, tsk);
1901 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1902 * @from: attach to all cgroups of a given task
1903 * @tsk: the task to be attached
1905 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
1907 struct cgroupfs_root *root;
1911 for_each_active_root(root) {
1912 struct cgroup *from_cg = task_cgroup_from_root(from, root);
1914 retval = cgroup_attach_task(from_cg, tsk);
1922 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
1925 * cgroup_attach_proc works in two stages, the first of which prefetches all
1926 * new css_sets needed (to make sure we have enough memory before committing
1927 * to the move) and stores them in a list of entries of the following type.
1928 * TODO: possible optimization: use css_set->rcu_head for chaining instead
1930 struct cg_list_entry {
1932 struct list_head links;
1935 static bool css_set_check_fetched(struct cgroup *cgrp,
1936 struct task_struct *tsk, struct css_set *cg,
1937 struct list_head *newcg_list)
1939 struct css_set *newcg;
1940 struct cg_list_entry *cg_entry;
1941 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
1943 read_lock(&css_set_lock);
1944 newcg = find_existing_css_set(cg, cgrp, template);
1947 read_unlock(&css_set_lock);
1949 /* doesn't exist at all? */
1952 /* see if it's already in the list */
1953 list_for_each_entry(cg_entry, newcg_list, links) {
1954 if (cg_entry->cg == newcg) {
1966 * Find the new css_set and store it in the list in preparation for moving the
1967 * given task to the given cgroup. Returns 0 or -ENOMEM.
1969 static int css_set_prefetch(struct cgroup *cgrp, struct css_set *cg,
1970 struct list_head *newcg_list)
1972 struct css_set *newcg;
1973 struct cg_list_entry *cg_entry;
1975 /* ensure a new css_set will exist for this thread */
1976 newcg = find_css_set(cg, cgrp);
1979 /* add it to the list */
1980 cg_entry = kmalloc(sizeof(struct cg_list_entry), GFP_KERNEL);
1985 cg_entry->cg = newcg;
1986 list_add(&cg_entry->links, newcg_list);
1991 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
1992 * @cgrp: the cgroup to attach to
1993 * @leader: the threadgroup leader task_struct of the group to be attached
1995 * Call holding cgroup_mutex and the threadgroup_fork_lock of the leader. Will
1996 * take task_lock of each thread in leader's threadgroup individually in turn.
1998 int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2000 int retval, i, group_size;
2001 struct cgroup_subsys *ss, *failed_ss = NULL;
2002 bool cancel_failed_ss = false;
2003 /* guaranteed to be initialized later, but the compiler needs this */
2004 struct cgroup *oldcgrp = NULL;
2005 struct css_set *oldcg;
2006 struct cgroupfs_root *root = cgrp->root;
2007 /* threadgroup list cursor and array */
2008 struct task_struct *tsk;
2009 struct flex_array *group;
2011 * we need to make sure we have css_sets for all the tasks we're
2012 * going to move -before- we actually start moving them, so that in
2013 * case we get an ENOMEM we can bail out before making any changes.
2015 struct list_head newcg_list;
2016 struct cg_list_entry *cg_entry, *temp_nobe;
2019 * step 0: in order to do expensive, possibly blocking operations for
2020 * every thread, we cannot iterate the thread group list, since it needs
2021 * rcu or tasklist locked. instead, build an array of all threads in the
2022 * group - threadgroup_fork_lock prevents new threads from appearing,
2023 * and if threads exit, this will just be an over-estimate.
2025 group_size = get_nr_threads(leader);
2026 /* flex_array supports very large thread-groups better than kmalloc. */
2027 group = flex_array_alloc(sizeof(struct task_struct *), group_size,
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;
2036 /* prevent changes to the threadgroup list while we take a snapshot. */
2038 if (!thread_group_leader(leader)) {
2040 * a race with de_thread from another thread's exec() may strip
2041 * us of our leadership, making while_each_thread unsafe to use
2042 * on this task. if this happens, there is no choice but to
2043 * throw this task away and try again (from cgroup_procs_write);
2044 * this is "double-double-toil-and-trouble-check locking".
2048 goto out_free_group_list;
2050 /* take a reference on each task in the group to go in the array. */
2054 /* as per above, nr_threads may decrease, but not increase. */
2055 BUG_ON(i >= group_size);
2056 get_task_struct(tsk);
2058 * saying GFP_ATOMIC has no effect here because we did prealloc
2059 * earlier, but it's good form to communicate our expectations.
2061 retval = flex_array_put_ptr(group, i, tsk, GFP_ATOMIC);
2062 BUG_ON(retval != 0);
2064 } while_each_thread(leader, tsk);
2065 /* remember the number of threads in the array for later. */
2070 * step 1: check that we can legitimately attach to the cgroup.
2072 for_each_subsys(root, ss) {
2073 if (ss->can_attach) {
2074 retval = ss->can_attach(ss, cgrp, leader);
2077 goto out_cancel_attach;
2080 /* a callback to be run on every thread in the threadgroup. */
2081 if (ss->can_attach_task) {
2082 /* run on each task in the threadgroup. */
2083 for (i = 0; i < group_size; i++) {
2084 tsk = flex_array_get_ptr(group, i);
2085 retval = ss->can_attach_task(cgrp, tsk);
2088 cancel_failed_ss = true;
2089 goto out_cancel_attach;
2096 * step 2: make sure css_sets exist for all threads to be migrated.
2097 * we use find_css_set, which allocates a new one if necessary.
2099 INIT_LIST_HEAD(&newcg_list);
2100 for (i = 0; i < group_size; i++) {
2101 tsk = flex_array_get_ptr(group, i);
2102 /* nothing to do if this task is already in the cgroup */
2103 oldcgrp = task_cgroup_from_root(tsk, root);
2104 if (cgrp == oldcgrp)
2106 /* get old css_set pointer */
2108 if (tsk->flags & PF_EXITING) {
2109 /* ignore this task if it's going away */
2113 oldcg = tsk->cgroups;
2116 /* see if the new one for us is already in the list? */
2117 if (css_set_check_fetched(cgrp, tsk, oldcg, &newcg_list)) {
2118 /* was already there, nothing to do. */
2121 /* we don't already have it. get new one. */
2122 retval = css_set_prefetch(cgrp, oldcg, &newcg_list);
2125 goto out_list_teardown;
2130 * step 3: now that we're guaranteed success wrt the css_sets, proceed
2131 * to move all tasks to the new cgroup, calling ss->attach_task for each
2132 * one along the way. there are no failure cases after here, so this is
2135 for_each_subsys(root, ss) {
2137 ss->pre_attach(cgrp);
2139 for (i = 0; i < group_size; i++) {
2140 tsk = flex_array_get_ptr(group, i);
2141 /* leave current thread as it is if it's already there */
2142 oldcgrp = task_cgroup_from_root(tsk, root);
2143 if (cgrp == oldcgrp)
2145 /* attach each task to each subsystem */
2146 for_each_subsys(root, ss) {
2147 if (ss->attach_task)
2148 ss->attach_task(cgrp, tsk);
2150 /* if the thread is PF_EXITING, it can just get skipped. */
2151 retval = cgroup_task_migrate(cgrp, oldcgrp, tsk, true);
2152 BUG_ON(retval != 0 && retval != -ESRCH);
2154 /* nothing is sensitive to fork() after this point. */
2157 * step 4: do expensive, non-thread-specific subsystem callbacks.
2158 * TODO: if ever a subsystem needs to know the oldcgrp for each task
2159 * being moved, this call will need to be reworked to communicate that.
2161 for_each_subsys(root, ss) {
2163 ss->attach(ss, cgrp, oldcgrp, leader);
2167 * step 5: success! and cleanup
2170 cgroup_wakeup_rmdir_waiter(cgrp);
2173 /* clean up the list of prefetched css_sets. */
2174 list_for_each_entry_safe(cg_entry, temp_nobe, &newcg_list, links) {
2175 list_del(&cg_entry->links);
2176 put_css_set(cg_entry->cg);
2180 /* same deal as in cgroup_attach_task */
2182 for_each_subsys(root, ss) {
2183 if (ss == failed_ss) {
2184 if (cancel_failed_ss && ss->cancel_attach)
2185 ss->cancel_attach(ss, cgrp, leader);
2188 if (ss->cancel_attach)
2189 ss->cancel_attach(ss, cgrp, leader);
2192 /* clean up the array of referenced threads in the group. */
2193 for (i = 0; i < group_size; i++) {
2194 tsk = flex_array_get_ptr(group, i);
2195 put_task_struct(tsk);
2197 out_free_group_list:
2198 flex_array_free(group);
2202 static int cgroup_allow_attach(struct cgroup *cgrp, struct task_struct *tsk)
2204 struct cgroup_subsys *ss;
2207 for_each_subsys(cgrp->root, ss) {
2208 if (ss->allow_attach) {
2209 ret = ss->allow_attach(cgrp, tsk);
2221 * Find the task_struct of the task to attach by vpid and pass it along to the
2222 * function to attach either it or all tasks in its threadgroup. Will take
2223 * cgroup_mutex; may take task_lock of task.
2225 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2227 struct task_struct *tsk;
2228 const struct cred *cred = current_cred(), *tcred;
2231 if (!cgroup_lock_live_group(cgrp))
2236 tsk = find_task_by_vpid(pid);
2244 * RCU protects this access, since tsk was found in the
2245 * tid map. a race with de_thread may cause group_leader
2246 * to stop being the leader, but cgroup_attach_proc will
2249 tsk = tsk->group_leader;
2250 } else if (tsk->flags & PF_EXITING) {
2251 /* optimization for the single-task-only case */
2258 * even if we're attaching all tasks in the thread group, we
2259 * only need to check permissions on one of them.
2261 tcred = __task_cred(tsk);
2263 cred->euid != tcred->uid &&
2264 cred->euid != tcred->suid) {
2266 * if the default permission check fails, give each
2267 * cgroup a chance to extend the permission check
2269 ret = cgroup_allow_attach(cgrp, tsk);
2276 get_task_struct(tsk);
2280 tsk = current->group_leader;
2283 get_task_struct(tsk);
2287 threadgroup_fork_write_lock(tsk);
2288 ret = cgroup_attach_proc(cgrp, tsk);
2289 threadgroup_fork_write_unlock(tsk);
2291 ret = cgroup_attach_task(cgrp, tsk);
2293 put_task_struct(tsk);
2298 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2300 return attach_task_by_pid(cgrp, pid, false);
2303 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2308 * attach_proc fails with -EAGAIN if threadgroup leadership
2309 * changes in the middle of the operation, in which case we need
2310 * to find the task_struct for the new leader and start over.
2312 ret = attach_task_by_pid(cgrp, tgid, true);
2313 } while (ret == -EAGAIN);
2318 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2319 * @cgrp: the cgroup to be checked for liveness
2321 * On success, returns true; the lock should be later released with
2322 * cgroup_unlock(). On failure returns false with no lock held.
2324 bool cgroup_lock_live_group(struct cgroup *cgrp)
2326 mutex_lock(&cgroup_mutex);
2327 if (cgroup_is_removed(cgrp)) {
2328 mutex_unlock(&cgroup_mutex);
2333 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2335 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2338 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2339 if (strlen(buffer) >= PATH_MAX)
2341 if (!cgroup_lock_live_group(cgrp))
2343 strcpy(cgrp->root->release_agent_path, buffer);
2348 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2349 struct seq_file *seq)
2351 if (!cgroup_lock_live_group(cgrp))
2353 seq_puts(seq, cgrp->root->release_agent_path);
2354 seq_putc(seq, '\n');
2359 /* A buffer size big enough for numbers or short strings */
2360 #define CGROUP_LOCAL_BUFFER_SIZE 64
2362 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2364 const char __user *userbuf,
2365 size_t nbytes, loff_t *unused_ppos)
2367 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2373 if (nbytes >= sizeof(buffer))
2375 if (copy_from_user(buffer, userbuf, nbytes))
2378 buffer[nbytes] = 0; /* nul-terminate */
2379 if (cft->write_u64) {
2380 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2383 retval = cft->write_u64(cgrp, cft, val);
2385 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2388 retval = cft->write_s64(cgrp, cft, val);
2395 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2397 const char __user *userbuf,
2398 size_t nbytes, loff_t *unused_ppos)
2400 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2402 size_t max_bytes = cft->max_write_len;
2403 char *buffer = local_buffer;
2406 max_bytes = sizeof(local_buffer) - 1;
2407 if (nbytes >= max_bytes)
2409 /* Allocate a dynamic buffer if we need one */
2410 if (nbytes >= sizeof(local_buffer)) {
2411 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2415 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2420 buffer[nbytes] = 0; /* nul-terminate */
2421 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2425 if (buffer != local_buffer)
2430 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2431 size_t nbytes, loff_t *ppos)
2433 struct cftype *cft = __d_cft(file->f_dentry);
2434 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2436 if (cgroup_is_removed(cgrp))
2439 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2440 if (cft->write_u64 || cft->write_s64)
2441 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2442 if (cft->write_string)
2443 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2445 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2446 return ret ? ret : nbytes;
2451 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2453 char __user *buf, size_t nbytes,
2456 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2457 u64 val = cft->read_u64(cgrp, cft);
2458 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2460 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2463 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2465 char __user *buf, size_t nbytes,
2468 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2469 s64 val = cft->read_s64(cgrp, cft);
2470 int len = sprintf(tmp, "%lld\n", (long long) val);
2472 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2475 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2476 size_t nbytes, loff_t *ppos)
2478 struct cftype *cft = __d_cft(file->f_dentry);
2479 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2481 if (cgroup_is_removed(cgrp))
2485 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2487 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2489 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2494 * seqfile ops/methods for returning structured data. Currently just
2495 * supports string->u64 maps, but can be extended in future.
2498 struct cgroup_seqfile_state {
2500 struct cgroup *cgroup;
2503 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2505 struct seq_file *sf = cb->state;
2506 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2509 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2511 struct cgroup_seqfile_state *state = m->private;
2512 struct cftype *cft = state->cft;
2513 if (cft->read_map) {
2514 struct cgroup_map_cb cb = {
2515 .fill = cgroup_map_add,
2518 return cft->read_map(state->cgroup, cft, &cb);
2520 return cft->read_seq_string(state->cgroup, cft, m);
2523 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2525 struct seq_file *seq = file->private_data;
2526 kfree(seq->private);
2527 return single_release(inode, file);
2530 static const struct file_operations cgroup_seqfile_operations = {
2532 .write = cgroup_file_write,
2533 .llseek = seq_lseek,
2534 .release = cgroup_seqfile_release,
2537 static int cgroup_file_open(struct inode *inode, struct file *file)
2542 err = generic_file_open(inode, file);
2545 cft = __d_cft(file->f_dentry);
2547 if (cft->read_map || cft->read_seq_string) {
2548 struct cgroup_seqfile_state *state =
2549 kzalloc(sizeof(*state), GFP_USER);
2553 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2554 file->f_op = &cgroup_seqfile_operations;
2555 err = single_open(file, cgroup_seqfile_show, state);
2558 } else if (cft->open)
2559 err = cft->open(inode, file);
2566 static int cgroup_file_release(struct inode *inode, struct file *file)
2568 struct cftype *cft = __d_cft(file->f_dentry);
2570 return cft->release(inode, file);
2575 * cgroup_rename - Only allow simple rename of directories in place.
2577 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2578 struct inode *new_dir, struct dentry *new_dentry)
2580 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2582 if (new_dentry->d_inode)
2584 if (old_dir != new_dir)
2586 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2589 static const struct file_operations cgroup_file_operations = {
2590 .read = cgroup_file_read,
2591 .write = cgroup_file_write,
2592 .llseek = generic_file_llseek,
2593 .open = cgroup_file_open,
2594 .release = cgroup_file_release,
2597 static const struct inode_operations cgroup_dir_inode_operations = {
2598 .lookup = cgroup_lookup,
2599 .mkdir = cgroup_mkdir,
2600 .rmdir = cgroup_rmdir,
2601 .rename = cgroup_rename,
2604 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
2606 if (dentry->d_name.len > NAME_MAX)
2607 return ERR_PTR(-ENAMETOOLONG);
2608 d_add(dentry, NULL);
2613 * Check if a file is a control file
2615 static inline struct cftype *__file_cft(struct file *file)
2617 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2618 return ERR_PTR(-EINVAL);
2619 return __d_cft(file->f_dentry);
2622 static int cgroup_create_file(struct dentry *dentry, mode_t mode,
2623 struct super_block *sb)
2625 struct inode *inode;
2629 if (dentry->d_inode)
2632 inode = cgroup_new_inode(mode, sb);
2636 if (S_ISDIR(mode)) {
2637 inode->i_op = &cgroup_dir_inode_operations;
2638 inode->i_fop = &simple_dir_operations;
2640 /* start off with i_nlink == 2 (for "." entry) */
2643 /* start with the directory inode held, so that we can
2644 * populate it without racing with another mkdir */
2645 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2646 } else if (S_ISREG(mode)) {
2648 inode->i_fop = &cgroup_file_operations;
2650 d_instantiate(dentry, inode);
2651 dget(dentry); /* Extra count - pin the dentry in core */
2656 * cgroup_create_dir - create a directory for an object.
2657 * @cgrp: the cgroup we create the directory for. It must have a valid
2658 * ->parent field. And we are going to fill its ->dentry field.
2659 * @dentry: dentry of the new cgroup
2660 * @mode: mode to set on new directory.
2662 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2665 struct dentry *parent;
2668 parent = cgrp->parent->dentry;
2669 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2671 dentry->d_fsdata = cgrp;
2672 inc_nlink(parent->d_inode);
2673 rcu_assign_pointer(cgrp->dentry, dentry);
2682 * cgroup_file_mode - deduce file mode of a control file
2683 * @cft: the control file in question
2685 * returns cft->mode if ->mode is not 0
2686 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2687 * returns S_IRUGO if it has only a read handler
2688 * returns S_IWUSR if it has only a write hander
2690 static mode_t cgroup_file_mode(const struct cftype *cft)
2697 if (cft->read || cft->read_u64 || cft->read_s64 ||
2698 cft->read_map || cft->read_seq_string)
2701 if (cft->write || cft->write_u64 || cft->write_s64 ||
2702 cft->write_string || cft->trigger)
2708 int cgroup_add_file(struct cgroup *cgrp,
2709 struct cgroup_subsys *subsys,
2710 const struct cftype *cft)
2712 struct dentry *dir = cgrp->dentry;
2713 struct dentry *dentry;
2717 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2718 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2719 strcpy(name, subsys->name);
2722 strcat(name, cft->name);
2723 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2724 dentry = lookup_one_len(name, dir, strlen(name));
2725 if (!IS_ERR(dentry)) {
2726 mode = cgroup_file_mode(cft);
2727 error = cgroup_create_file(dentry, mode | S_IFREG,
2730 dentry->d_fsdata = (void *)cft;
2733 error = PTR_ERR(dentry);
2736 EXPORT_SYMBOL_GPL(cgroup_add_file);
2738 int cgroup_add_files(struct cgroup *cgrp,
2739 struct cgroup_subsys *subsys,
2740 const struct cftype cft[],
2744 for (i = 0; i < count; i++) {
2745 err = cgroup_add_file(cgrp, subsys, &cft[i]);
2751 EXPORT_SYMBOL_GPL(cgroup_add_files);
2754 * cgroup_task_count - count the number of tasks in a cgroup.
2755 * @cgrp: the cgroup in question
2757 * Return the number of tasks in the cgroup.
2759 int cgroup_task_count(const struct cgroup *cgrp)
2762 struct cg_cgroup_link *link;
2764 read_lock(&css_set_lock);
2765 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2766 count += atomic_read(&link->cg->refcount);
2768 read_unlock(&css_set_lock);
2773 * Advance a list_head iterator. The iterator should be positioned at
2774 * the start of a css_set
2776 static void cgroup_advance_iter(struct cgroup *cgrp,
2777 struct cgroup_iter *it)
2779 struct list_head *l = it->cg_link;
2780 struct cg_cgroup_link *link;
2783 /* Advance to the next non-empty css_set */
2786 if (l == &cgrp->css_sets) {
2790 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2792 } while (list_empty(&cg->tasks));
2794 it->task = cg->tasks.next;
2798 * To reduce the fork() overhead for systems that are not actually
2799 * using their cgroups capability, we don't maintain the lists running
2800 * through each css_set to its tasks until we see the list actually
2801 * used - in other words after the first call to cgroup_iter_start().
2803 * The tasklist_lock is not held here, as do_each_thread() and
2804 * while_each_thread() are protected by RCU.
2806 static void cgroup_enable_task_cg_lists(void)
2808 struct task_struct *p, *g;
2809 write_lock(&css_set_lock);
2810 use_task_css_set_links = 1;
2811 do_each_thread(g, p) {
2814 * We should check if the process is exiting, otherwise
2815 * it will race with cgroup_exit() in that the list
2816 * entry won't be deleted though the process has exited.
2818 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2819 list_add(&p->cg_list, &p->cgroups->tasks);
2821 } while_each_thread(g, p);
2822 write_unlock(&css_set_lock);
2825 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2828 * The first time anyone tries to iterate across a cgroup,
2829 * we need to enable the list linking each css_set to its
2830 * tasks, and fix up all existing tasks.
2832 if (!use_task_css_set_links)
2833 cgroup_enable_task_cg_lists();
2835 read_lock(&css_set_lock);
2836 it->cg_link = &cgrp->css_sets;
2837 cgroup_advance_iter(cgrp, it);
2840 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2841 struct cgroup_iter *it)
2843 struct task_struct *res;
2844 struct list_head *l = it->task;
2845 struct cg_cgroup_link *link;
2847 /* If the iterator cg is NULL, we have no tasks */
2850 res = list_entry(l, struct task_struct, cg_list);
2851 /* Advance iterator to find next entry */
2853 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2854 if (l == &link->cg->tasks) {
2855 /* We reached the end of this task list - move on to
2856 * the next cg_cgroup_link */
2857 cgroup_advance_iter(cgrp, it);
2864 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2866 read_unlock(&css_set_lock);
2869 static inline int started_after_time(struct task_struct *t1,
2870 struct timespec *time,
2871 struct task_struct *t2)
2873 int start_diff = timespec_compare(&t1->start_time, time);
2874 if (start_diff > 0) {
2876 } else if (start_diff < 0) {
2880 * Arbitrarily, if two processes started at the same
2881 * time, we'll say that the lower pointer value
2882 * started first. Note that t2 may have exited by now
2883 * so this may not be a valid pointer any longer, but
2884 * that's fine - it still serves to distinguish
2885 * between two tasks started (effectively) simultaneously.
2892 * This function is a callback from heap_insert() and is used to order
2894 * In this case we order the heap in descending task start time.
2896 static inline int started_after(void *p1, void *p2)
2898 struct task_struct *t1 = p1;
2899 struct task_struct *t2 = p2;
2900 return started_after_time(t1, &t2->start_time, t2);
2904 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2905 * @scan: struct cgroup_scanner containing arguments for the scan
2907 * Arguments include pointers to callback functions test_task() and
2909 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2910 * and if it returns true, call process_task() for it also.
2911 * The test_task pointer may be NULL, meaning always true (select all tasks).
2912 * Effectively duplicates cgroup_iter_{start,next,end}()
2913 * but does not lock css_set_lock for the call to process_task().
2914 * The struct cgroup_scanner may be embedded in any structure of the caller's
2916 * It is guaranteed that process_task() will act on every task that
2917 * is a member of the cgroup for the duration of this call. This
2918 * function may or may not call process_task() for tasks that exit
2919 * or move to a different cgroup during the call, or are forked or
2920 * move into the cgroup during the call.
2922 * Note that test_task() may be called with locks held, and may in some
2923 * situations be called multiple times for the same task, so it should
2925 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2926 * pre-allocated and will be used for heap operations (and its "gt" member will
2927 * be overwritten), else a temporary heap will be used (allocation of which
2928 * may cause this function to fail).
2930 int cgroup_scan_tasks(struct cgroup_scanner *scan)
2933 struct cgroup_iter it;
2934 struct task_struct *p, *dropped;
2935 /* Never dereference latest_task, since it's not refcounted */
2936 struct task_struct *latest_task = NULL;
2937 struct ptr_heap tmp_heap;
2938 struct ptr_heap *heap;
2939 struct timespec latest_time = { 0, 0 };
2942 /* The caller supplied our heap and pre-allocated its memory */
2944 heap->gt = &started_after;
2946 /* We need to allocate our own heap memory */
2948 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2950 /* cannot allocate the heap */
2956 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2957 * to determine which are of interest, and using the scanner's
2958 * "process_task" callback to process any of them that need an update.
2959 * Since we don't want to hold any locks during the task updates,
2960 * gather tasks to be processed in a heap structure.
2961 * The heap is sorted by descending task start time.
2962 * If the statically-sized heap fills up, we overflow tasks that
2963 * started later, and in future iterations only consider tasks that
2964 * started after the latest task in the previous pass. This
2965 * guarantees forward progress and that we don't miss any tasks.
2968 cgroup_iter_start(scan->cg, &it);
2969 while ((p = cgroup_iter_next(scan->cg, &it))) {
2971 * Only affect tasks that qualify per the caller's callback,
2972 * if he provided one
2974 if (scan->test_task && !scan->test_task(p, scan))
2977 * Only process tasks that started after the last task
2980 if (!started_after_time(p, &latest_time, latest_task))
2982 dropped = heap_insert(heap, p);
2983 if (dropped == NULL) {
2985 * The new task was inserted; the heap wasn't
2989 } else if (dropped != p) {
2991 * The new task was inserted, and pushed out a
2995 put_task_struct(dropped);
2998 * Else the new task was newer than anything already in
2999 * the heap and wasn't inserted
3002 cgroup_iter_end(scan->cg, &it);
3005 for (i = 0; i < heap->size; i++) {
3006 struct task_struct *q = heap->ptrs[i];
3008 latest_time = q->start_time;
3011 /* Process the task per the caller's callback */
3012 scan->process_task(q, scan);
3016 * If we had to process any tasks at all, scan again
3017 * in case some of them were in the middle of forking
3018 * children that didn't get processed.
3019 * Not the most efficient way to do it, but it avoids
3020 * having to take callback_mutex in the fork path
3024 if (heap == &tmp_heap)
3025 heap_free(&tmp_heap);
3030 * Stuff for reading the 'tasks'/'procs' files.
3032 * Reading this file can return large amounts of data if a cgroup has
3033 * *lots* of attached tasks. So it may need several calls to read(),
3034 * but we cannot guarantee that the information we produce is correct
3035 * unless we produce it entirely atomically.
3040 * The following two functions "fix" the issue where there are more pids
3041 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3042 * TODO: replace with a kernel-wide solution to this problem
3044 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3045 static void *pidlist_allocate(int count)
3047 if (PIDLIST_TOO_LARGE(count))
3048 return vmalloc(count * sizeof(pid_t));
3050 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3052 static void pidlist_free(void *p)
3054 if (is_vmalloc_addr(p))
3059 static void *pidlist_resize(void *p, int newcount)
3062 /* note: if new alloc fails, old p will still be valid either way */
3063 if (is_vmalloc_addr(p)) {
3064 newlist = vmalloc(newcount * sizeof(pid_t));
3067 memcpy(newlist, p, newcount * sizeof(pid_t));
3070 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3076 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3077 * If the new stripped list is sufficiently smaller and there's enough memory
3078 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3079 * number of unique elements.
3081 /* is the size difference enough that we should re-allocate the array? */
3082 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3083 static int pidlist_uniq(pid_t **p, int length)
3090 * we presume the 0th element is unique, so i starts at 1. trivial
3091 * edge cases first; no work needs to be done for either
3093 if (length == 0 || length == 1)
3095 /* src and dest walk down the list; dest counts unique elements */
3096 for (src = 1; src < length; src++) {
3097 /* find next unique element */
3098 while (list[src] == list[src-1]) {
3103 /* dest always points to where the next unique element goes */
3104 list[dest] = list[src];
3109 * if the length difference is large enough, we want to allocate a
3110 * smaller buffer to save memory. if this fails due to out of memory,
3111 * we'll just stay with what we've got.
3113 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3114 newlist = pidlist_resize(list, dest);
3121 static int cmppid(const void *a, const void *b)
3123 return *(pid_t *)a - *(pid_t *)b;
3127 * find the appropriate pidlist for our purpose (given procs vs tasks)
3128 * returns with the lock on that pidlist already held, and takes care
3129 * of the use count, or returns NULL with no locks held if we're out of
3132 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3133 enum cgroup_filetype type)
3135 struct cgroup_pidlist *l;
3136 /* don't need task_nsproxy() if we're looking at ourself */
3137 struct pid_namespace *ns = current->nsproxy->pid_ns;
3140 * We can't drop the pidlist_mutex before taking the l->mutex in case
3141 * the last ref-holder is trying to remove l from the list at the same
3142 * time. Holding the pidlist_mutex precludes somebody taking whichever
3143 * list we find out from under us - compare release_pid_array().
3145 mutex_lock(&cgrp->pidlist_mutex);
3146 list_for_each_entry(l, &cgrp->pidlists, links) {
3147 if (l->key.type == type && l->key.ns == ns) {
3148 /* make sure l doesn't vanish out from under us */
3149 down_write(&l->mutex);
3150 mutex_unlock(&cgrp->pidlist_mutex);
3154 /* entry not found; create a new one */
3155 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3157 mutex_unlock(&cgrp->pidlist_mutex);
3160 init_rwsem(&l->mutex);
3161 down_write(&l->mutex);
3163 l->key.ns = get_pid_ns(ns);
3164 l->use_count = 0; /* don't increment here */
3167 list_add(&l->links, &cgrp->pidlists);
3168 mutex_unlock(&cgrp->pidlist_mutex);
3173 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3175 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3176 struct cgroup_pidlist **lp)
3180 int pid, n = 0; /* used for populating the array */
3181 struct cgroup_iter it;
3182 struct task_struct *tsk;
3183 struct cgroup_pidlist *l;
3186 * If cgroup gets more users after we read count, we won't have
3187 * enough space - tough. This race is indistinguishable to the
3188 * caller from the case that the additional cgroup users didn't
3189 * show up until sometime later on.
3191 length = cgroup_task_count(cgrp);
3192 array = pidlist_allocate(length);
3195 /* now, populate the array */
3196 cgroup_iter_start(cgrp, &it);
3197 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3198 if (unlikely(n == length))
3200 /* get tgid or pid for procs or tasks file respectively */
3201 if (type == CGROUP_FILE_PROCS)
3202 pid = task_tgid_vnr(tsk);
3204 pid = task_pid_vnr(tsk);
3205 if (pid > 0) /* make sure to only use valid results */
3208 cgroup_iter_end(cgrp, &it);
3210 /* now sort & (if procs) strip out duplicates */
3211 sort(array, length, sizeof(pid_t), cmppid, NULL);
3212 if (type == CGROUP_FILE_PROCS)
3213 length = pidlist_uniq(&array, length);
3214 l = cgroup_pidlist_find(cgrp, type);
3216 pidlist_free(array);
3219 /* store array, freeing old if necessary - lock already held */
3220 pidlist_free(l->list);
3224 up_write(&l->mutex);
3230 * cgroupstats_build - build and fill cgroupstats
3231 * @stats: cgroupstats to fill information into
3232 * @dentry: A dentry entry belonging to the cgroup for which stats have
3235 * Build and fill cgroupstats so that taskstats can export it to user
3238 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3241 struct cgroup *cgrp;
3242 struct cgroup_iter it;
3243 struct task_struct *tsk;
3246 * Validate dentry by checking the superblock operations,
3247 * and make sure it's a directory.
3249 if (dentry->d_sb->s_op != &cgroup_ops ||
3250 !S_ISDIR(dentry->d_inode->i_mode))
3254 cgrp = dentry->d_fsdata;
3256 cgroup_iter_start(cgrp, &it);
3257 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3258 switch (tsk->state) {
3260 stats->nr_running++;
3262 case TASK_INTERRUPTIBLE:
3263 stats->nr_sleeping++;
3265 case TASK_UNINTERRUPTIBLE:
3266 stats->nr_uninterruptible++;
3269 stats->nr_stopped++;
3272 if (delayacct_is_task_waiting_on_io(tsk))
3273 stats->nr_io_wait++;
3277 cgroup_iter_end(cgrp, &it);
3285 * seq_file methods for the tasks/procs files. The seq_file position is the
3286 * next pid to display; the seq_file iterator is a pointer to the pid
3287 * in the cgroup->l->list array.
3290 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3293 * Initially we receive a position value that corresponds to
3294 * one more than the last pid shown (or 0 on the first call or
3295 * after a seek to the start). Use a binary-search to find the
3296 * next pid to display, if any
3298 struct cgroup_pidlist *l = s->private;
3299 int index = 0, pid = *pos;
3302 down_read(&l->mutex);
3304 int end = l->length;
3306 while (index < end) {
3307 int mid = (index + end) / 2;
3308 if (l->list[mid] == pid) {
3311 } else if (l->list[mid] <= pid)
3317 /* If we're off the end of the array, we're done */
3318 if (index >= l->length)
3320 /* Update the abstract position to be the actual pid that we found */
3321 iter = l->list + index;
3326 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3328 struct cgroup_pidlist *l = s->private;
3332 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3334 struct cgroup_pidlist *l = s->private;
3336 pid_t *end = l->list + l->length;
3338 * Advance to the next pid in the array. If this goes off the
3350 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3352 return seq_printf(s, "%d\n", *(int *)v);
3356 * seq_operations functions for iterating on pidlists through seq_file -
3357 * independent of whether it's tasks or procs
3359 static const struct seq_operations cgroup_pidlist_seq_operations = {
3360 .start = cgroup_pidlist_start,
3361 .stop = cgroup_pidlist_stop,
3362 .next = cgroup_pidlist_next,
3363 .show = cgroup_pidlist_show,
3366 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3369 * the case where we're the last user of this particular pidlist will
3370 * have us remove it from the cgroup's list, which entails taking the
3371 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3372 * pidlist_mutex, we have to take pidlist_mutex first.
3374 mutex_lock(&l->owner->pidlist_mutex);
3375 down_write(&l->mutex);
3376 BUG_ON(!l->use_count);
3377 if (!--l->use_count) {
3378 /* we're the last user if refcount is 0; remove and free */
3379 list_del(&l->links);
3380 mutex_unlock(&l->owner->pidlist_mutex);
3381 pidlist_free(l->list);
3382 put_pid_ns(l->key.ns);
3383 up_write(&l->mutex);
3387 mutex_unlock(&l->owner->pidlist_mutex);
3388 up_write(&l->mutex);
3391 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3393 struct cgroup_pidlist *l;
3394 if (!(file->f_mode & FMODE_READ))
3397 * the seq_file will only be initialized if the file was opened for
3398 * reading; hence we check if it's not null only in that case.
3400 l = ((struct seq_file *)file->private_data)->private;
3401 cgroup_release_pid_array(l);
3402 return seq_release(inode, file);
3405 static const struct file_operations cgroup_pidlist_operations = {
3407 .llseek = seq_lseek,
3408 .write = cgroup_file_write,
3409 .release = cgroup_pidlist_release,
3413 * The following functions handle opens on a file that displays a pidlist
3414 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3417 /* helper function for the two below it */
3418 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3420 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3421 struct cgroup_pidlist *l;
3424 /* Nothing to do for write-only files */
3425 if (!(file->f_mode & FMODE_READ))
3428 /* have the array populated */
3429 retval = pidlist_array_load(cgrp, type, &l);
3432 /* configure file information */
3433 file->f_op = &cgroup_pidlist_operations;
3435 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3437 cgroup_release_pid_array(l);
3440 ((struct seq_file *)file->private_data)->private = l;
3443 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3445 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3447 static int cgroup_procs_open(struct inode *unused, struct file *file)
3449 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3452 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3455 return notify_on_release(cgrp);
3458 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3462 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3464 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3466 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3471 * Unregister event and free resources.
3473 * Gets called from workqueue.
3475 static void cgroup_event_remove(struct work_struct *work)
3477 struct cgroup_event *event = container_of(work, struct cgroup_event,
3479 struct cgroup *cgrp = event->cgrp;
3481 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3483 eventfd_ctx_put(event->eventfd);
3489 * Gets called on POLLHUP on eventfd when user closes it.
3491 * Called with wqh->lock held and interrupts disabled.
3493 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3494 int sync, void *key)
3496 struct cgroup_event *event = container_of(wait,
3497 struct cgroup_event, wait);
3498 struct cgroup *cgrp = event->cgrp;
3499 unsigned long flags = (unsigned long)key;
3501 if (flags & POLLHUP) {
3502 __remove_wait_queue(event->wqh, &event->wait);
3503 spin_lock(&cgrp->event_list_lock);
3504 list_del(&event->list);
3505 spin_unlock(&cgrp->event_list_lock);
3507 * We are in atomic context, but cgroup_event_remove() may
3508 * sleep, so we have to call it in workqueue.
3510 schedule_work(&event->remove);
3516 static void cgroup_event_ptable_queue_proc(struct file *file,
3517 wait_queue_head_t *wqh, poll_table *pt)
3519 struct cgroup_event *event = container_of(pt,
3520 struct cgroup_event, pt);
3523 add_wait_queue(wqh, &event->wait);
3527 * Parse input and register new cgroup event handler.
3529 * Input must be in format '<event_fd> <control_fd> <args>'.
3530 * Interpretation of args is defined by control file implementation.
3532 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3535 struct cgroup_event *event = NULL;
3536 unsigned int efd, cfd;
3537 struct file *efile = NULL;
3538 struct file *cfile = NULL;
3542 efd = simple_strtoul(buffer, &endp, 10);
3547 cfd = simple_strtoul(buffer, &endp, 10);
3548 if ((*endp != ' ') && (*endp != '\0'))
3552 event = kzalloc(sizeof(*event), GFP_KERNEL);
3556 INIT_LIST_HEAD(&event->list);
3557 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3558 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3559 INIT_WORK(&event->remove, cgroup_event_remove);
3561 efile = eventfd_fget(efd);
3562 if (IS_ERR(efile)) {
3563 ret = PTR_ERR(efile);
3567 event->eventfd = eventfd_ctx_fileget(efile);
3568 if (IS_ERR(event->eventfd)) {
3569 ret = PTR_ERR(event->eventfd);
3579 /* the process need read permission on control file */
3580 ret = file_permission(cfile, MAY_READ);
3584 event->cft = __file_cft(cfile);
3585 if (IS_ERR(event->cft)) {
3586 ret = PTR_ERR(event->cft);
3590 if (!event->cft->register_event || !event->cft->unregister_event) {
3595 ret = event->cft->register_event(cgrp, event->cft,
3596 event->eventfd, buffer);
3600 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3601 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3607 * Events should be removed after rmdir of cgroup directory, but before
3608 * destroying subsystem state objects. Let's take reference to cgroup
3609 * directory dentry to do that.
3613 spin_lock(&cgrp->event_list_lock);
3614 list_add(&event->list, &cgrp->event_list);
3615 spin_unlock(&cgrp->event_list_lock);
3626 if (event && event->eventfd && !IS_ERR(event->eventfd))
3627 eventfd_ctx_put(event->eventfd);
3629 if (!IS_ERR_OR_NULL(efile))
3637 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3640 return clone_children(cgrp);
3643 static int cgroup_clone_children_write(struct cgroup *cgrp,
3648 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3650 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3655 * for the common functions, 'private' gives the type of file
3657 /* for hysterical raisins, we can't put this on the older files */
3658 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3659 static struct cftype files[] = {
3662 .open = cgroup_tasks_open,
3663 .write_u64 = cgroup_tasks_write,
3664 .release = cgroup_pidlist_release,
3665 .mode = S_IRUGO | S_IWUSR,
3668 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3669 .open = cgroup_procs_open,
3670 .write_u64 = cgroup_procs_write,
3671 .release = cgroup_pidlist_release,
3672 .mode = S_IRUGO | S_IWUSR,
3675 .name = "notify_on_release",
3676 .read_u64 = cgroup_read_notify_on_release,
3677 .write_u64 = cgroup_write_notify_on_release,
3680 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3681 .write_string = cgroup_write_event_control,
3685 .name = "cgroup.clone_children",
3686 .read_u64 = cgroup_clone_children_read,
3687 .write_u64 = cgroup_clone_children_write,
3691 static struct cftype cft_release_agent = {
3692 .name = "release_agent",
3693 .read_seq_string = cgroup_release_agent_show,
3694 .write_string = cgroup_release_agent_write,
3695 .max_write_len = PATH_MAX,
3698 static int cgroup_populate_dir(struct cgroup *cgrp)
3701 struct cgroup_subsys *ss;
3703 /* First clear out any existing files */
3704 cgroup_clear_directory(cgrp->dentry);
3706 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
3710 if (cgrp == cgrp->top_cgroup) {
3711 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
3715 for_each_subsys(cgrp->root, ss) {
3716 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
3719 /* This cgroup is ready now */
3720 for_each_subsys(cgrp->root, ss) {
3721 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3723 * Update id->css pointer and make this css visible from
3724 * CSS ID functions. This pointer will be dereferened
3725 * from RCU-read-side without locks.
3728 rcu_assign_pointer(css->id->css, css);
3734 static void init_cgroup_css(struct cgroup_subsys_state *css,
3735 struct cgroup_subsys *ss,
3736 struct cgroup *cgrp)
3739 atomic_set(&css->refcnt, 1);
3742 if (cgrp == dummytop)
3743 set_bit(CSS_ROOT, &css->flags);
3744 BUG_ON(cgrp->subsys[ss->subsys_id]);
3745 cgrp->subsys[ss->subsys_id] = css;
3748 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3750 /* We need to take each hierarchy_mutex in a consistent order */
3754 * No worry about a race with rebind_subsystems that might mess up the
3755 * locking order, since both parties are under cgroup_mutex.
3757 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3758 struct cgroup_subsys *ss = subsys[i];
3761 if (ss->root == root)
3762 mutex_lock(&ss->hierarchy_mutex);
3766 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3770 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3771 struct cgroup_subsys *ss = subsys[i];
3774 if (ss->root == root)
3775 mutex_unlock(&ss->hierarchy_mutex);
3780 * cgroup_create - create a cgroup
3781 * @parent: cgroup that will be parent of the new cgroup
3782 * @dentry: dentry of the new cgroup
3783 * @mode: mode to set on new inode
3785 * Must be called with the mutex on the parent inode held
3787 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3790 struct cgroup *cgrp;
3791 struct cgroupfs_root *root = parent->root;
3793 struct cgroup_subsys *ss;
3794 struct super_block *sb = root->sb;
3796 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3800 /* Grab a reference on the superblock so the hierarchy doesn't
3801 * get deleted on unmount if there are child cgroups. This
3802 * can be done outside cgroup_mutex, since the sb can't
3803 * disappear while someone has an open control file on the
3805 atomic_inc(&sb->s_active);
3807 mutex_lock(&cgroup_mutex);
3809 init_cgroup_housekeeping(cgrp);
3811 cgrp->parent = parent;
3812 cgrp->root = parent->root;
3813 cgrp->top_cgroup = parent->top_cgroup;
3815 if (notify_on_release(parent))
3816 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3818 if (clone_children(parent))
3819 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3821 for_each_subsys(root, ss) {
3822 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
3828 init_cgroup_css(css, ss, cgrp);
3830 err = alloc_css_id(ss, parent, cgrp);
3834 /* At error, ->destroy() callback has to free assigned ID. */
3835 if (clone_children(parent) && ss->post_clone)
3836 ss->post_clone(ss, cgrp);
3839 cgroup_lock_hierarchy(root);
3840 list_add(&cgrp->sibling, &cgrp->parent->children);
3841 cgroup_unlock_hierarchy(root);
3842 root->number_of_cgroups++;
3844 err = cgroup_create_dir(cgrp, dentry, mode);
3848 set_bit(CGRP_RELEASABLE, &parent->flags);
3850 /* The cgroup directory was pre-locked for us */
3851 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3853 err = cgroup_populate_dir(cgrp);
3854 /* If err < 0, we have a half-filled directory - oh well ;) */
3856 mutex_unlock(&cgroup_mutex);
3857 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3863 cgroup_lock_hierarchy(root);
3864 list_del(&cgrp->sibling);
3865 cgroup_unlock_hierarchy(root);
3866 root->number_of_cgroups--;
3870 for_each_subsys(root, ss) {
3871 if (cgrp->subsys[ss->subsys_id])
3872 ss->destroy(ss, cgrp);
3875 mutex_unlock(&cgroup_mutex);
3877 /* Release the reference count that we took on the superblock */
3878 deactivate_super(sb);
3884 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3886 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3888 /* the vfs holds inode->i_mutex already */
3889 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3892 static int cgroup_has_css_refs(struct cgroup *cgrp)
3894 /* Check the reference count on each subsystem. Since we
3895 * already established that there are no tasks in the
3896 * cgroup, if the css refcount is also 1, then there should
3897 * be no outstanding references, so the subsystem is safe to
3898 * destroy. We scan across all subsystems rather than using
3899 * the per-hierarchy linked list of mounted subsystems since
3900 * we can be called via check_for_release() with no
3901 * synchronization other than RCU, and the subsystem linked
3902 * list isn't RCU-safe */
3905 * We won't need to lock the subsys array, because the subsystems
3906 * we're concerned about aren't going anywhere since our cgroup root
3907 * has a reference on them.
3909 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3910 struct cgroup_subsys *ss = subsys[i];
3911 struct cgroup_subsys_state *css;
3912 /* Skip subsystems not present or not in this hierarchy */
3913 if (ss == NULL || ss->root != cgrp->root)
3915 css = cgrp->subsys[ss->subsys_id];
3916 /* When called from check_for_release() it's possible
3917 * that by this point the cgroup has been removed
3918 * and the css deleted. But a false-positive doesn't
3919 * matter, since it can only happen if the cgroup
3920 * has been deleted and hence no longer needs the
3921 * release agent to be called anyway. */
3922 if (css && (atomic_read(&css->refcnt) > 1))
3929 * Atomically mark all (or else none) of the cgroup's CSS objects as
3930 * CSS_REMOVED. Return true on success, or false if the cgroup has
3931 * busy subsystems. Call with cgroup_mutex held
3934 static int cgroup_clear_css_refs(struct cgroup *cgrp)
3936 struct cgroup_subsys *ss;
3937 unsigned long flags;
3938 bool failed = false;
3939 local_irq_save(flags);
3940 for_each_subsys(cgrp->root, ss) {
3941 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3944 /* We can only remove a CSS with a refcnt==1 */
3945 refcnt = atomic_read(&css->refcnt);
3952 * Drop the refcnt to 0 while we check other
3953 * subsystems. This will cause any racing
3954 * css_tryget() to spin until we set the
3955 * CSS_REMOVED bits or abort
3957 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3963 for_each_subsys(cgrp->root, ss) {
3964 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3967 * Restore old refcnt if we previously managed
3968 * to clear it from 1 to 0
3970 if (!atomic_read(&css->refcnt))
3971 atomic_set(&css->refcnt, 1);
3973 /* Commit the fact that the CSS is removed */
3974 set_bit(CSS_REMOVED, &css->flags);
3977 local_irq_restore(flags);
3981 /* checks if all of the css_sets attached to a cgroup have a refcount of 0.
3982 * Must be called with css_set_lock held */
3983 static int cgroup_css_sets_empty(struct cgroup *cgrp)
3985 struct cg_cgroup_link *link;
3987 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
3988 struct css_set *cg = link->cg;
3989 if (atomic_read(&cg->refcount) > 0)
3996 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3998 struct cgroup *cgrp = dentry->d_fsdata;
4000 struct cgroup *parent;
4002 struct cgroup_event *event, *tmp;
4005 /* the vfs holds both inode->i_mutex already */
4007 mutex_lock(&cgroup_mutex);
4008 if (!cgroup_css_sets_empty(cgrp)) {
4009 mutex_unlock(&cgroup_mutex);
4012 if (!list_empty(&cgrp->children)) {
4013 mutex_unlock(&cgroup_mutex);
4016 mutex_unlock(&cgroup_mutex);
4019 * In general, subsystem has no css->refcnt after pre_destroy(). But
4020 * in racy cases, subsystem may have to get css->refcnt after
4021 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4022 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4023 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4024 * and subsystem's reference count handling. Please see css_get/put
4025 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4027 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4030 * Call pre_destroy handlers of subsys. Notify subsystems
4031 * that rmdir() request comes.
4033 ret = cgroup_call_pre_destroy(cgrp);
4035 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4039 mutex_lock(&cgroup_mutex);
4040 parent = cgrp->parent;
4041 if (!cgroup_css_sets_empty(cgrp) || !list_empty(&cgrp->children)) {
4042 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4043 mutex_unlock(&cgroup_mutex);
4046 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
4047 if (!cgroup_clear_css_refs(cgrp)) {
4048 mutex_unlock(&cgroup_mutex);
4050 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4051 * prepare_to_wait(), we need to check this flag.
4053 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
4055 finish_wait(&cgroup_rmdir_waitq, &wait);
4056 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4057 if (signal_pending(current))
4061 /* NO css_tryget() can success after here. */
4062 finish_wait(&cgroup_rmdir_waitq, &wait);
4063 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4065 spin_lock(&release_list_lock);
4066 set_bit(CGRP_REMOVED, &cgrp->flags);
4067 if (!list_empty(&cgrp->release_list))
4068 list_del_init(&cgrp->release_list);
4069 spin_unlock(&release_list_lock);
4071 cgroup_lock_hierarchy(cgrp->root);
4072 /* delete this cgroup from parent->children */
4073 list_del_init(&cgrp->sibling);
4074 cgroup_unlock_hierarchy(cgrp->root);
4076 d = dget(cgrp->dentry);
4078 cgroup_d_remove_dir(d);
4081 check_for_release(parent);
4084 * Unregister events and notify userspace.
4085 * Notify userspace about cgroup removing only after rmdir of cgroup
4086 * directory to avoid race between userspace and kernelspace
4088 spin_lock(&cgrp->event_list_lock);
4089 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4090 list_del(&event->list);
4091 remove_wait_queue(event->wqh, &event->wait);
4092 eventfd_signal(event->eventfd, 1);
4093 schedule_work(&event->remove);
4095 spin_unlock(&cgrp->event_list_lock);
4097 mutex_unlock(&cgroup_mutex);
4101 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4103 struct cgroup_subsys_state *css;
4105 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4107 /* Create the top cgroup state for this subsystem */
4108 list_add(&ss->sibling, &rootnode.subsys_list);
4109 ss->root = &rootnode;
4110 css = ss->create(ss, dummytop);
4111 /* We don't handle early failures gracefully */
4112 BUG_ON(IS_ERR(css));
4113 init_cgroup_css(css, ss, dummytop);
4115 /* Update the init_css_set to contain a subsys
4116 * pointer to this state - since the subsystem is
4117 * newly registered, all tasks and hence the
4118 * init_css_set is in the subsystem's top cgroup. */
4119 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4121 need_forkexit_callback |= ss->fork || ss->exit;
4123 /* At system boot, before all subsystems have been
4124 * registered, no tasks have been forked, so we don't
4125 * need to invoke fork callbacks here. */
4126 BUG_ON(!list_empty(&init_task.tasks));
4128 mutex_init(&ss->hierarchy_mutex);
4129 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4132 /* this function shouldn't be used with modular subsystems, since they
4133 * need to register a subsys_id, among other things */
4138 * cgroup_load_subsys: load and register a modular subsystem at runtime
4139 * @ss: the subsystem to load
4141 * This function should be called in a modular subsystem's initcall. If the
4142 * subsystem is built as a module, it will be assigned a new subsys_id and set
4143 * up for use. If the subsystem is built-in anyway, work is delegated to the
4144 * simpler cgroup_init_subsys.
4146 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4149 struct cgroup_subsys_state *css;
4151 /* check name and function validity */
4152 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4153 ss->create == NULL || ss->destroy == NULL)
4157 * we don't support callbacks in modular subsystems. this check is
4158 * before the ss->module check for consistency; a subsystem that could
4159 * be a module should still have no callbacks even if the user isn't
4160 * compiling it as one.
4162 if (ss->fork || ss->exit)
4166 * an optionally modular subsystem is built-in: we want to do nothing,
4167 * since cgroup_init_subsys will have already taken care of it.
4169 if (ss->module == NULL) {
4170 /* a few sanity checks */
4171 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4172 BUG_ON(subsys[ss->subsys_id] != ss);
4177 * need to register a subsys id before anything else - for example,
4178 * init_cgroup_css needs it.
4180 mutex_lock(&cgroup_mutex);
4181 /* find the first empty slot in the array */
4182 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4183 if (subsys[i] == NULL)
4186 if (i == CGROUP_SUBSYS_COUNT) {
4187 /* maximum number of subsystems already registered! */
4188 mutex_unlock(&cgroup_mutex);
4191 /* assign ourselves the subsys_id */
4196 * no ss->create seems to need anything important in the ss struct, so
4197 * this can happen first (i.e. before the rootnode attachment).
4199 css = ss->create(ss, dummytop);
4201 /* failure case - need to deassign the subsys[] slot. */
4203 mutex_unlock(&cgroup_mutex);
4204 return PTR_ERR(css);
4207 list_add(&ss->sibling, &rootnode.subsys_list);
4208 ss->root = &rootnode;
4210 /* our new subsystem will be attached to the dummy hierarchy. */
4211 init_cgroup_css(css, ss, dummytop);
4212 /* init_idr must be after init_cgroup_css because it sets css->id. */
4214 int ret = cgroup_init_idr(ss, css);
4216 dummytop->subsys[ss->subsys_id] = NULL;
4217 ss->destroy(ss, dummytop);
4219 mutex_unlock(&cgroup_mutex);
4225 * Now we need to entangle the css into the existing css_sets. unlike
4226 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4227 * will need a new pointer to it; done by iterating the css_set_table.
4228 * furthermore, modifying the existing css_sets will corrupt the hash
4229 * table state, so each changed css_set will need its hash recomputed.
4230 * this is all done under the css_set_lock.
4232 write_lock(&css_set_lock);
4233 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4235 struct hlist_node *node, *tmp;
4236 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4238 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4239 /* skip entries that we already rehashed */
4240 if (cg->subsys[ss->subsys_id])
4242 /* remove existing entry */
4243 hlist_del(&cg->hlist);
4245 cg->subsys[ss->subsys_id] = css;
4246 /* recompute hash and restore entry */
4247 new_bucket = css_set_hash(cg->subsys);
4248 hlist_add_head(&cg->hlist, new_bucket);
4251 write_unlock(&css_set_lock);
4253 mutex_init(&ss->hierarchy_mutex);
4254 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4258 mutex_unlock(&cgroup_mutex);
4261 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4264 * cgroup_unload_subsys: unload a modular subsystem
4265 * @ss: the subsystem to unload
4267 * This function should be called in a modular subsystem's exitcall. When this
4268 * function is invoked, the refcount on the subsystem's module will be 0, so
4269 * the subsystem will not be attached to any hierarchy.
4271 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4273 struct cg_cgroup_link *link;
4274 struct hlist_head *hhead;
4276 BUG_ON(ss->module == NULL);
4279 * we shouldn't be called if the subsystem is in use, and the use of
4280 * try_module_get in parse_cgroupfs_options should ensure that it
4281 * doesn't start being used while we're killing it off.
4283 BUG_ON(ss->root != &rootnode);
4285 mutex_lock(&cgroup_mutex);
4286 /* deassign the subsys_id */
4287 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4288 subsys[ss->subsys_id] = NULL;
4290 /* remove subsystem from rootnode's list of subsystems */
4291 list_del_init(&ss->sibling);
4294 * disentangle the css from all css_sets attached to the dummytop. as
4295 * in loading, we need to pay our respects to the hashtable gods.
4297 write_lock(&css_set_lock);
4298 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4299 struct css_set *cg = link->cg;
4301 hlist_del(&cg->hlist);
4302 BUG_ON(!cg->subsys[ss->subsys_id]);
4303 cg->subsys[ss->subsys_id] = NULL;
4304 hhead = css_set_hash(cg->subsys);
4305 hlist_add_head(&cg->hlist, hhead);
4307 write_unlock(&css_set_lock);
4310 * remove subsystem's css from the dummytop and free it - need to free
4311 * before marking as null because ss->destroy needs the cgrp->subsys
4312 * pointer to find their state. note that this also takes care of
4313 * freeing the css_id.
4315 ss->destroy(ss, dummytop);
4316 dummytop->subsys[ss->subsys_id] = NULL;
4318 mutex_unlock(&cgroup_mutex);
4320 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4323 * cgroup_init_early - cgroup initialization at system boot
4325 * Initialize cgroups at system boot, and initialize any
4326 * subsystems that request early init.
4328 int __init cgroup_init_early(void)
4331 atomic_set(&init_css_set.refcount, 1);
4332 INIT_LIST_HEAD(&init_css_set.cg_links);
4333 INIT_LIST_HEAD(&init_css_set.tasks);
4334 INIT_HLIST_NODE(&init_css_set.hlist);
4336 init_cgroup_root(&rootnode);
4338 init_task.cgroups = &init_css_set;
4340 init_css_set_link.cg = &init_css_set;
4341 init_css_set_link.cgrp = dummytop;
4342 list_add(&init_css_set_link.cgrp_link_list,
4343 &rootnode.top_cgroup.css_sets);
4344 list_add(&init_css_set_link.cg_link_list,
4345 &init_css_set.cg_links);
4347 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4348 INIT_HLIST_HEAD(&css_set_table[i]);
4350 /* at bootup time, we don't worry about modular subsystems */
4351 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4352 struct cgroup_subsys *ss = subsys[i];
4355 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4356 BUG_ON(!ss->create);
4357 BUG_ON(!ss->destroy);
4358 if (ss->subsys_id != i) {
4359 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4360 ss->name, ss->subsys_id);
4365 cgroup_init_subsys(ss);
4371 * cgroup_init - cgroup initialization
4373 * Register cgroup filesystem and /proc file, and initialize
4374 * any subsystems that didn't request early init.
4376 int __init cgroup_init(void)
4380 struct hlist_head *hhead;
4382 err = bdi_init(&cgroup_backing_dev_info);
4386 /* at bootup time, we don't worry about modular subsystems */
4387 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4388 struct cgroup_subsys *ss = subsys[i];
4389 if (!ss->early_init)
4390 cgroup_init_subsys(ss);
4392 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4395 /* Add init_css_set to the hash table */
4396 hhead = css_set_hash(init_css_set.subsys);
4397 hlist_add_head(&init_css_set.hlist, hhead);
4398 BUG_ON(!init_root_id(&rootnode));
4400 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4406 err = register_filesystem(&cgroup_fs_type);
4408 kobject_put(cgroup_kobj);
4412 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4416 bdi_destroy(&cgroup_backing_dev_info);
4422 * proc_cgroup_show()
4423 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4424 * - Used for /proc/<pid>/cgroup.
4425 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4426 * doesn't really matter if tsk->cgroup changes after we read it,
4427 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4428 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4429 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4430 * cgroup to top_cgroup.
4433 /* TODO: Use a proper seq_file iterator */
4434 static int proc_cgroup_show(struct seq_file *m, void *v)
4437 struct task_struct *tsk;
4440 struct cgroupfs_root *root;
4443 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4449 tsk = get_pid_task(pid, PIDTYPE_PID);
4455 mutex_lock(&cgroup_mutex);
4457 for_each_active_root(root) {
4458 struct cgroup_subsys *ss;
4459 struct cgroup *cgrp;
4462 seq_printf(m, "%d:", root->hierarchy_id);
4463 for_each_subsys(root, ss)
4464 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4465 if (strlen(root->name))
4466 seq_printf(m, "%sname=%s", count ? "," : "",
4469 cgrp = task_cgroup_from_root(tsk, root);
4470 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4478 mutex_unlock(&cgroup_mutex);
4479 put_task_struct(tsk);
4486 static int cgroup_open(struct inode *inode, struct file *file)
4488 struct pid *pid = PROC_I(inode)->pid;
4489 return single_open(file, proc_cgroup_show, pid);
4492 const struct file_operations proc_cgroup_operations = {
4493 .open = cgroup_open,
4495 .llseek = seq_lseek,
4496 .release = single_release,
4499 /* Display information about each subsystem and each hierarchy */
4500 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4504 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4506 * ideally we don't want subsystems moving around while we do this.
4507 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4508 * subsys/hierarchy state.
4510 mutex_lock(&cgroup_mutex);
4511 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4512 struct cgroup_subsys *ss = subsys[i];
4515 seq_printf(m, "%s\t%d\t%d\t%d\n",
4516 ss->name, ss->root->hierarchy_id,
4517 ss->root->number_of_cgroups, !ss->disabled);
4519 mutex_unlock(&cgroup_mutex);
4523 static int cgroupstats_open(struct inode *inode, struct file *file)
4525 return single_open(file, proc_cgroupstats_show, NULL);
4528 static const struct file_operations proc_cgroupstats_operations = {
4529 .open = cgroupstats_open,
4531 .llseek = seq_lseek,
4532 .release = single_release,
4536 * cgroup_fork - attach newly forked task to its parents cgroup.
4537 * @child: pointer to task_struct of forking parent process.
4539 * Description: A task inherits its parent's cgroup at fork().
4541 * A pointer to the shared css_set was automatically copied in
4542 * fork.c by dup_task_struct(). However, we ignore that copy, since
4543 * it was not made under the protection of RCU or cgroup_mutex, so
4544 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4545 * have already changed current->cgroups, allowing the previously
4546 * referenced cgroup group to be removed and freed.
4548 * At the point that cgroup_fork() is called, 'current' is the parent
4549 * task, and the passed argument 'child' points to the child task.
4551 void cgroup_fork(struct task_struct *child)
4554 child->cgroups = current->cgroups;
4555 get_css_set(child->cgroups);
4556 task_unlock(current);
4557 INIT_LIST_HEAD(&child->cg_list);
4561 * cgroup_fork_callbacks - run fork callbacks
4562 * @child: the new task
4564 * Called on a new task very soon before adding it to the
4565 * tasklist. No need to take any locks since no-one can
4566 * be operating on this task.
4568 void cgroup_fork_callbacks(struct task_struct *child)
4570 if (need_forkexit_callback) {
4573 * forkexit callbacks are only supported for builtin
4574 * subsystems, and the builtin section of the subsys array is
4575 * immutable, so we don't need to lock the subsys array here.
4577 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4578 struct cgroup_subsys *ss = subsys[i];
4580 ss->fork(ss, child);
4586 * cgroup_post_fork - called on a new task after adding it to the task list
4587 * @child: the task in question
4589 * Adds the task to the list running through its css_set if necessary.
4590 * Has to be after the task is visible on the task list in case we race
4591 * with the first call to cgroup_iter_start() - to guarantee that the
4592 * new task ends up on its list.
4594 void cgroup_post_fork(struct task_struct *child)
4596 if (use_task_css_set_links) {
4597 write_lock(&css_set_lock);
4599 if (list_empty(&child->cg_list))
4600 list_add(&child->cg_list, &child->cgroups->tasks);
4602 write_unlock(&css_set_lock);
4606 * cgroup_exit - detach cgroup from exiting task
4607 * @tsk: pointer to task_struct of exiting process
4608 * @run_callback: run exit callbacks?
4610 * Description: Detach cgroup from @tsk and release it.
4612 * Note that cgroups marked notify_on_release force every task in
4613 * them to take the global cgroup_mutex mutex when exiting.
4614 * This could impact scaling on very large systems. Be reluctant to
4615 * use notify_on_release cgroups where very high task exit scaling
4616 * is required on large systems.
4618 * the_top_cgroup_hack:
4620 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4622 * We call cgroup_exit() while the task is still competent to
4623 * handle notify_on_release(), then leave the task attached to the
4624 * root cgroup in each hierarchy for the remainder of its exit.
4626 * To do this properly, we would increment the reference count on
4627 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4628 * code we would add a second cgroup function call, to drop that
4629 * reference. This would just create an unnecessary hot spot on
4630 * the top_cgroup reference count, to no avail.
4632 * Normally, holding a reference to a cgroup without bumping its
4633 * count is unsafe. The cgroup could go away, or someone could
4634 * attach us to a different cgroup, decrementing the count on
4635 * the first cgroup that we never incremented. But in this case,
4636 * top_cgroup isn't going away, and either task has PF_EXITING set,
4637 * which wards off any cgroup_attach_task() attempts, or task is a failed
4638 * fork, never visible to cgroup_attach_task.
4640 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4646 * Unlink from the css_set task list if necessary.
4647 * Optimistically check cg_list before taking
4650 if (!list_empty(&tsk->cg_list)) {
4651 write_lock(&css_set_lock);
4652 if (!list_empty(&tsk->cg_list))
4653 list_del_init(&tsk->cg_list);
4654 write_unlock(&css_set_lock);
4657 /* Reassign the task to the init_css_set. */
4660 tsk->cgroups = &init_css_set;
4662 if (run_callbacks && need_forkexit_callback) {
4664 * modular subsystems can't use callbacks, so no need to lock
4667 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4668 struct cgroup_subsys *ss = subsys[i];
4670 struct cgroup *old_cgrp =
4671 rcu_dereference_raw(cg->subsys[i])->cgroup;
4672 struct cgroup *cgrp = task_cgroup(tsk, i);
4673 ss->exit(ss, cgrp, old_cgrp, tsk);
4684 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4685 * @cgrp: the cgroup in question
4686 * @task: the task in question
4688 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4691 * If we are sending in dummytop, then presumably we are creating
4692 * the top cgroup in the subsystem.
4694 * Called only by the ns (nsproxy) cgroup.
4696 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4699 struct cgroup *target;
4701 if (cgrp == dummytop)
4704 target = task_cgroup_from_root(task, cgrp->root);
4705 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4706 cgrp = cgrp->parent;
4707 ret = (cgrp == target);
4711 static void check_for_release(struct cgroup *cgrp)
4713 /* All of these checks rely on RCU to keep the cgroup
4714 * structure alive */
4715 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4716 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4717 /* Control Group is currently removeable. If it's not
4718 * already queued for a userspace notification, queue
4720 int need_schedule_work = 0;
4721 spin_lock(&release_list_lock);
4722 if (!cgroup_is_removed(cgrp) &&
4723 list_empty(&cgrp->release_list)) {
4724 list_add(&cgrp->release_list, &release_list);
4725 need_schedule_work = 1;
4727 spin_unlock(&release_list_lock);
4728 if (need_schedule_work)
4729 schedule_work(&release_agent_work);
4733 /* Caller must verify that the css is not for root cgroup */
4734 void __css_get(struct cgroup_subsys_state *css, int count)
4736 atomic_add(count, &css->refcnt);
4737 set_bit(CGRP_RELEASABLE, &css->cgroup->flags);
4739 EXPORT_SYMBOL_GPL(__css_get);
4741 /* Caller must verify that the css is not for root cgroup */
4742 void __css_put(struct cgroup_subsys_state *css, int count)
4744 struct cgroup *cgrp = css->cgroup;
4747 val = atomic_sub_return(count, &css->refcnt);
4749 check_for_release(cgrp);
4750 cgroup_wakeup_rmdir_waiter(cgrp);
4753 WARN_ON_ONCE(val < 1);
4755 EXPORT_SYMBOL_GPL(__css_put);
4758 * Notify userspace when a cgroup is released, by running the
4759 * configured release agent with the name of the cgroup (path
4760 * relative to the root of cgroup file system) as the argument.
4762 * Most likely, this user command will try to rmdir this cgroup.
4764 * This races with the possibility that some other task will be
4765 * attached to this cgroup before it is removed, or that some other
4766 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4767 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4768 * unused, and this cgroup will be reprieved from its death sentence,
4769 * to continue to serve a useful existence. Next time it's released,
4770 * we will get notified again, if it still has 'notify_on_release' set.
4772 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4773 * means only wait until the task is successfully execve()'d. The
4774 * separate release agent task is forked by call_usermodehelper(),
4775 * then control in this thread returns here, without waiting for the
4776 * release agent task. We don't bother to wait because the caller of
4777 * this routine has no use for the exit status of the release agent
4778 * task, so no sense holding our caller up for that.
4780 static void cgroup_release_agent(struct work_struct *work)
4782 BUG_ON(work != &release_agent_work);
4783 mutex_lock(&cgroup_mutex);
4784 spin_lock(&release_list_lock);
4785 while (!list_empty(&release_list)) {
4786 char *argv[3], *envp[3];
4788 char *pathbuf = NULL, *agentbuf = NULL;
4789 struct cgroup *cgrp = list_entry(release_list.next,
4792 list_del_init(&cgrp->release_list);
4793 spin_unlock(&release_list_lock);
4794 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4797 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4799 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4804 argv[i++] = agentbuf;
4805 argv[i++] = pathbuf;
4809 /* minimal command environment */
4810 envp[i++] = "HOME=/";
4811 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4814 /* Drop the lock while we invoke the usermode helper,
4815 * since the exec could involve hitting disk and hence
4816 * be a slow process */
4817 mutex_unlock(&cgroup_mutex);
4818 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4819 mutex_lock(&cgroup_mutex);
4823 spin_lock(&release_list_lock);
4825 spin_unlock(&release_list_lock);
4826 mutex_unlock(&cgroup_mutex);
4829 static int __init cgroup_disable(char *str)
4834 while ((token = strsep(&str, ",")) != NULL) {
4838 * cgroup_disable, being at boot time, can't know about module
4839 * subsystems, so we don't worry about them.
4841 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4842 struct cgroup_subsys *ss = subsys[i];
4844 if (!strcmp(token, ss->name)) {
4846 printk(KERN_INFO "Disabling %s control group"
4847 " subsystem\n", ss->name);
4854 __setup("cgroup_disable=", cgroup_disable);
4857 * Functons for CSS ID.
4861 *To get ID other than 0, this should be called when !cgroup_is_removed().
4863 unsigned short css_id(struct cgroup_subsys_state *css)
4865 struct css_id *cssid;
4868 * This css_id() can return correct value when somone has refcnt
4869 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4870 * it's unchanged until freed.
4872 cssid = rcu_dereference_check(css->id,
4873 rcu_read_lock_held() || atomic_read(&css->refcnt));
4879 EXPORT_SYMBOL_GPL(css_id);
4881 unsigned short css_depth(struct cgroup_subsys_state *css)
4883 struct css_id *cssid;
4885 cssid = rcu_dereference_check(css->id,
4886 rcu_read_lock_held() || atomic_read(&css->refcnt));
4889 return cssid->depth;
4892 EXPORT_SYMBOL_GPL(css_depth);
4895 * css_is_ancestor - test "root" css is an ancestor of "child"
4896 * @child: the css to be tested.
4897 * @root: the css supporsed to be an ancestor of the child.
4899 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4900 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4901 * But, considering usual usage, the csses should be valid objects after test.
4902 * Assuming that the caller will do some action to the child if this returns
4903 * returns true, the caller must take "child";s reference count.
4904 * If "child" is valid object and this returns true, "root" is valid, too.
4907 bool css_is_ancestor(struct cgroup_subsys_state *child,
4908 const struct cgroup_subsys_state *root)
4910 struct css_id *child_id;
4911 struct css_id *root_id;
4915 child_id = rcu_dereference(child->id);
4916 root_id = rcu_dereference(root->id);
4919 || (child_id->depth < root_id->depth)
4920 || (child_id->stack[root_id->depth] != root_id->id))
4926 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
4928 struct css_id *id = css->id;
4929 /* When this is called before css_id initialization, id can be NULL */
4933 BUG_ON(!ss->use_id);
4935 rcu_assign_pointer(id->css, NULL);
4936 rcu_assign_pointer(css->id, NULL);
4937 spin_lock(&ss->id_lock);
4938 idr_remove(&ss->idr, id->id);
4939 spin_unlock(&ss->id_lock);
4940 kfree_rcu(id, rcu_head);
4942 EXPORT_SYMBOL_GPL(free_css_id);
4945 * This is called by init or create(). Then, calls to this function are
4946 * always serialized (By cgroup_mutex() at create()).
4949 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
4951 struct css_id *newid;
4952 int myid, error, size;
4954 BUG_ON(!ss->use_id);
4956 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
4957 newid = kzalloc(size, GFP_KERNEL);
4959 return ERR_PTR(-ENOMEM);
4961 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
4965 spin_lock(&ss->id_lock);
4966 /* Don't use 0. allocates an ID of 1-65535 */
4967 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
4968 spin_unlock(&ss->id_lock);
4970 /* Returns error when there are no free spaces for new ID.*/
4975 if (myid > CSS_ID_MAX)
4979 newid->depth = depth;
4983 spin_lock(&ss->id_lock);
4984 idr_remove(&ss->idr, myid);
4985 spin_unlock(&ss->id_lock);
4988 return ERR_PTR(error);
4992 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
4993 struct cgroup_subsys_state *rootcss)
4995 struct css_id *newid;
4997 spin_lock_init(&ss->id_lock);
5000 newid = get_new_cssid(ss, 0);
5002 return PTR_ERR(newid);
5004 newid->stack[0] = newid->id;
5005 newid->css = rootcss;
5006 rootcss->id = newid;
5010 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5011 struct cgroup *child)
5013 int subsys_id, i, depth = 0;
5014 struct cgroup_subsys_state *parent_css, *child_css;
5015 struct css_id *child_id, *parent_id;
5017 subsys_id = ss->subsys_id;
5018 parent_css = parent->subsys[subsys_id];
5019 child_css = child->subsys[subsys_id];
5020 parent_id = parent_css->id;
5021 depth = parent_id->depth + 1;
5023 child_id = get_new_cssid(ss, depth);
5024 if (IS_ERR(child_id))
5025 return PTR_ERR(child_id);
5027 for (i = 0; i < depth; i++)
5028 child_id->stack[i] = parent_id->stack[i];
5029 child_id->stack[depth] = child_id->id;
5031 * child_id->css pointer will be set after this cgroup is available
5032 * see cgroup_populate_dir()
5034 rcu_assign_pointer(child_css->id, child_id);
5040 * css_lookup - lookup css by id
5041 * @ss: cgroup subsys to be looked into.
5044 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5045 * NULL if not. Should be called under rcu_read_lock()
5047 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5049 struct css_id *cssid = NULL;
5051 BUG_ON(!ss->use_id);
5052 cssid = idr_find(&ss->idr, id);
5054 if (unlikely(!cssid))
5057 return rcu_dereference(cssid->css);
5059 EXPORT_SYMBOL_GPL(css_lookup);
5062 * css_get_next - lookup next cgroup under specified hierarchy.
5063 * @ss: pointer to subsystem
5064 * @id: current position of iteration.
5065 * @root: pointer to css. search tree under this.
5066 * @foundid: position of found object.
5068 * Search next css under the specified hierarchy of rootid. Calling under
5069 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5071 struct cgroup_subsys_state *
5072 css_get_next(struct cgroup_subsys *ss, int id,
5073 struct cgroup_subsys_state *root, int *foundid)
5075 struct cgroup_subsys_state *ret = NULL;
5078 int rootid = css_id(root);
5079 int depth = css_depth(root);
5084 BUG_ON(!ss->use_id);
5085 /* fill start point for scan */
5089 * scan next entry from bitmap(tree), tmpid is updated after
5092 spin_lock(&ss->id_lock);
5093 tmp = idr_get_next(&ss->idr, &tmpid);
5094 spin_unlock(&ss->id_lock);
5098 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5099 ret = rcu_dereference(tmp->css);
5105 /* continue to scan from next id */
5112 * get corresponding css from file open on cgroupfs directory
5114 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5116 struct cgroup *cgrp;
5117 struct inode *inode;
5118 struct cgroup_subsys_state *css;
5120 inode = f->f_dentry->d_inode;
5121 /* check in cgroup filesystem dir */
5122 if (inode->i_op != &cgroup_dir_inode_operations)
5123 return ERR_PTR(-EBADF);
5125 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5126 return ERR_PTR(-EINVAL);
5129 cgrp = __d_cgrp(f->f_dentry);
5130 css = cgrp->subsys[id];
5131 return css ? css : ERR_PTR(-ENOENT);
5134 #ifdef CONFIG_CGROUP_DEBUG
5135 static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
5136 struct cgroup *cont)
5138 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5141 return ERR_PTR(-ENOMEM);
5146 static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
5148 kfree(cont->subsys[debug_subsys_id]);
5151 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5153 return atomic_read(&cont->count);
5156 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5158 return cgroup_task_count(cont);
5161 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5163 return (u64)(unsigned long)current->cgroups;
5166 static u64 current_css_set_refcount_read(struct cgroup *cont,
5172 count = atomic_read(¤t->cgroups->refcount);
5177 static int current_css_set_cg_links_read(struct cgroup *cont,
5179 struct seq_file *seq)
5181 struct cg_cgroup_link *link;
5184 read_lock(&css_set_lock);
5186 cg = rcu_dereference(current->cgroups);
5187 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5188 struct cgroup *c = link->cgrp;
5192 name = c->dentry->d_name.name;
5195 seq_printf(seq, "Root %d group %s\n",
5196 c->root->hierarchy_id, name);
5199 read_unlock(&css_set_lock);
5203 #define MAX_TASKS_SHOWN_PER_CSS 25
5204 static int cgroup_css_links_read(struct cgroup *cont,
5206 struct seq_file *seq)
5208 struct cg_cgroup_link *link;
5210 read_lock(&css_set_lock);
5211 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5212 struct css_set *cg = link->cg;
5213 struct task_struct *task;
5215 seq_printf(seq, "css_set %p\n", cg);
5216 list_for_each_entry(task, &cg->tasks, cg_list) {
5217 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5218 seq_puts(seq, " ...\n");
5221 seq_printf(seq, " task %d\n",
5222 task_pid_vnr(task));
5226 read_unlock(&css_set_lock);
5230 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5232 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5235 static struct cftype debug_files[] = {
5237 .name = "cgroup_refcount",
5238 .read_u64 = cgroup_refcount_read,
5241 .name = "taskcount",
5242 .read_u64 = debug_taskcount_read,
5246 .name = "current_css_set",
5247 .read_u64 = current_css_set_read,
5251 .name = "current_css_set_refcount",
5252 .read_u64 = current_css_set_refcount_read,
5256 .name = "current_css_set_cg_links",
5257 .read_seq_string = current_css_set_cg_links_read,
5261 .name = "cgroup_css_links",
5262 .read_seq_string = cgroup_css_links_read,
5266 .name = "releasable",
5267 .read_u64 = releasable_read,
5271 static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
5273 return cgroup_add_files(cont, ss, debug_files,
5274 ARRAY_SIZE(debug_files));
5277 struct cgroup_subsys debug_subsys = {
5279 .create = debug_create,
5280 .destroy = debug_destroy,
5281 .populate = debug_populate,
5282 .subsys_id = debug_subsys_id,
5284 #endif /* CONFIG_CGROUP_DEBUG */