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 if 'none', 'name=' and a subsystem name options
1180 * were not specified, let's default to 'all'
1182 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
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.
1806 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1812 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1813 * @cgrp: the cgroup the task is attaching to
1814 * @tsk: the task to be attached
1816 * Call holding cgroup_mutex. May take task_lock of
1817 * the task 'tsk' during call.
1819 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1822 struct cgroup_subsys *ss, *failed_ss = NULL;
1823 struct cgroup *oldcgrp;
1824 struct cgroupfs_root *root = cgrp->root;
1827 /* Nothing to do if the task is already in that cgroup */
1828 oldcgrp = task_cgroup_from_root(tsk, root);
1829 if (cgrp == oldcgrp)
1832 for_each_subsys(root, ss) {
1833 if (ss->can_attach) {
1834 retval = ss->can_attach(ss, cgrp, tsk);
1837 * Remember on which subsystem the can_attach()
1838 * failed, so that we only call cancel_attach()
1839 * against the subsystems whose can_attach()
1840 * succeeded. (See below)
1846 if (ss->can_attach_task) {
1847 retval = ss->can_attach_task(cgrp, tsk);
1860 retval = cgroup_task_migrate(cgrp, oldcgrp, tsk, false);
1864 for_each_subsys(root, ss) {
1866 ss->pre_attach(cgrp);
1867 if (ss->attach_task)
1868 ss->attach_task(cgrp, tsk);
1870 ss->attach(ss, cgrp, oldcgrp, tsk);
1872 set_bit(CGRP_RELEASABLE, &cgrp->flags);
1873 /* put_css_set will not destroy cg until after an RCU grace period */
1877 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1878 * is no longer empty.
1880 cgroup_wakeup_rmdir_waiter(cgrp);
1883 for_each_subsys(root, ss) {
1884 if (ss == failed_ss)
1886 * This subsystem was the one that failed the
1887 * can_attach() check earlier, so we don't need
1888 * to call cancel_attach() against it or any
1889 * remaining subsystems.
1892 if (ss->cancel_attach)
1893 ss->cancel_attach(ss, cgrp, tsk);
1900 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1901 * @from: attach to all cgroups of a given task
1902 * @tsk: the task to be attached
1904 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
1906 struct cgroupfs_root *root;
1910 for_each_active_root(root) {
1911 struct cgroup *from_cg = task_cgroup_from_root(from, root);
1913 retval = cgroup_attach_task(from_cg, tsk);
1921 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
1924 * cgroup_attach_proc works in two stages, the first of which prefetches all
1925 * new css_sets needed (to make sure we have enough memory before committing
1926 * to the move) and stores them in a list of entries of the following type.
1927 * TODO: possible optimization: use css_set->rcu_head for chaining instead
1929 struct cg_list_entry {
1931 struct list_head links;
1934 static bool css_set_check_fetched(struct cgroup *cgrp,
1935 struct task_struct *tsk, struct css_set *cg,
1936 struct list_head *newcg_list)
1938 struct css_set *newcg;
1939 struct cg_list_entry *cg_entry;
1940 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
1942 read_lock(&css_set_lock);
1943 newcg = find_existing_css_set(cg, cgrp, template);
1946 read_unlock(&css_set_lock);
1948 /* doesn't exist at all? */
1951 /* see if it's already in the list */
1952 list_for_each_entry(cg_entry, newcg_list, links) {
1953 if (cg_entry->cg == newcg) {
1965 * Find the new css_set and store it in the list in preparation for moving the
1966 * given task to the given cgroup. Returns 0 or -ENOMEM.
1968 static int css_set_prefetch(struct cgroup *cgrp, struct css_set *cg,
1969 struct list_head *newcg_list)
1971 struct css_set *newcg;
1972 struct cg_list_entry *cg_entry;
1974 /* ensure a new css_set will exist for this thread */
1975 newcg = find_css_set(cg, cgrp);
1978 /* add it to the list */
1979 cg_entry = kmalloc(sizeof(struct cg_list_entry), GFP_KERNEL);
1984 cg_entry->cg = newcg;
1985 list_add(&cg_entry->links, newcg_list);
1990 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
1991 * @cgrp: the cgroup to attach to
1992 * @leader: the threadgroup leader task_struct of the group to be attached
1994 * Call holding cgroup_mutex and the threadgroup_fork_lock of the leader. Will
1995 * take task_lock of each thread in leader's threadgroup individually in turn.
1997 int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
1999 int retval, i, group_size;
2000 struct cgroup_subsys *ss, *failed_ss = NULL;
2001 bool cancel_failed_ss = false;
2002 /* guaranteed to be initialized later, but the compiler needs this */
2003 struct cgroup *oldcgrp = NULL;
2004 struct css_set *oldcg;
2005 struct cgroupfs_root *root = cgrp->root;
2006 /* threadgroup list cursor and array */
2007 struct task_struct *tsk;
2008 struct flex_array *group;
2010 * we need to make sure we have css_sets for all the tasks we're
2011 * going to move -before- we actually start moving them, so that in
2012 * case we get an ENOMEM we can bail out before making any changes.
2014 struct list_head newcg_list;
2015 struct cg_list_entry *cg_entry, *temp_nobe;
2018 * step 0: in order to do expensive, possibly blocking operations for
2019 * every thread, we cannot iterate the thread group list, since it needs
2020 * rcu or tasklist locked. instead, build an array of all threads in the
2021 * group - threadgroup_fork_lock prevents new threads from appearing,
2022 * and if threads exit, this will just be an over-estimate.
2024 group_size = get_nr_threads(leader);
2025 /* flex_array supports very large thread-groups better than kmalloc. */
2026 group = flex_array_alloc(sizeof(struct task_struct *), group_size,
2030 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2031 retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2033 goto out_free_group_list;
2035 /* prevent changes to the threadgroup list while we take a snapshot. */
2037 if (!thread_group_leader(leader)) {
2039 * a race with de_thread from another thread's exec() may strip
2040 * us of our leadership, making while_each_thread unsafe to use
2041 * on this task. if this happens, there is no choice but to
2042 * throw this task away and try again (from cgroup_procs_write);
2043 * this is "double-double-toil-and-trouble-check locking".
2047 goto out_free_group_list;
2049 /* take a reference on each task in the group to go in the array. */
2053 /* as per above, nr_threads may decrease, but not increase. */
2054 BUG_ON(i >= group_size);
2055 get_task_struct(tsk);
2057 * saying GFP_ATOMIC has no effect here because we did prealloc
2058 * earlier, but it's good form to communicate our expectations.
2060 retval = flex_array_put_ptr(group, i, tsk, GFP_ATOMIC);
2061 BUG_ON(retval != 0);
2063 } while_each_thread(leader, tsk);
2064 /* remember the number of threads in the array for later. */
2069 * step 1: check that we can legitimately attach to the cgroup.
2071 for_each_subsys(root, ss) {
2072 if (ss->can_attach) {
2073 retval = ss->can_attach(ss, cgrp, leader);
2076 goto out_cancel_attach;
2079 /* a callback to be run on every thread in the threadgroup. */
2080 if (ss->can_attach_task) {
2081 /* run on each task in the threadgroup. */
2082 for (i = 0; i < group_size; i++) {
2083 tsk = flex_array_get_ptr(group, i);
2084 retval = ss->can_attach_task(cgrp, tsk);
2087 cancel_failed_ss = true;
2088 goto out_cancel_attach;
2095 * step 2: make sure css_sets exist for all threads to be migrated.
2096 * we use find_css_set, which allocates a new one if necessary.
2098 INIT_LIST_HEAD(&newcg_list);
2099 for (i = 0; i < group_size; i++) {
2100 tsk = flex_array_get_ptr(group, i);
2101 /* nothing to do if this task is already in the cgroup */
2102 oldcgrp = task_cgroup_from_root(tsk, root);
2103 if (cgrp == oldcgrp)
2105 /* get old css_set pointer */
2107 oldcg = tsk->cgroups;
2110 /* see if the new one for us is already in the list? */
2111 if (css_set_check_fetched(cgrp, tsk, oldcg, &newcg_list)) {
2112 /* was already there, nothing to do. */
2115 /* we don't already have it. get new one. */
2116 retval = css_set_prefetch(cgrp, oldcg, &newcg_list);
2119 goto out_list_teardown;
2124 * step 3: now that we're guaranteed success wrt the css_sets, proceed
2125 * to move all tasks to the new cgroup, calling ss->attach_task for each
2126 * one along the way. there are no failure cases after here, so this is
2129 for_each_subsys(root, ss) {
2131 ss->pre_attach(cgrp);
2133 for (i = 0; i < group_size; i++) {
2134 tsk = flex_array_get_ptr(group, i);
2135 /* leave current thread as it is if it's already there */
2136 oldcgrp = task_cgroup_from_root(tsk, root);
2137 if (cgrp == oldcgrp)
2139 /* attach each task to each subsystem */
2140 for_each_subsys(root, ss) {
2141 if (ss->attach_task)
2142 ss->attach_task(cgrp, tsk);
2144 /* if the thread is PF_EXITING, it can just get skipped. */
2145 retval = cgroup_task_migrate(cgrp, oldcgrp, tsk, true);
2146 BUG_ON(retval != 0 && retval != -ESRCH);
2148 /* nothing is sensitive to fork() after this point. */
2151 * step 4: do expensive, non-thread-specific subsystem callbacks.
2152 * TODO: if ever a subsystem needs to know the oldcgrp for each task
2153 * being moved, this call will need to be reworked to communicate that.
2155 for_each_subsys(root, ss) {
2157 ss->attach(ss, cgrp, oldcgrp, leader);
2161 * step 5: success! and cleanup
2164 cgroup_wakeup_rmdir_waiter(cgrp);
2167 /* clean up the list of prefetched css_sets. */
2168 list_for_each_entry_safe(cg_entry, temp_nobe, &newcg_list, links) {
2169 list_del(&cg_entry->links);
2170 put_css_set(cg_entry->cg);
2174 /* same deal as in cgroup_attach_task */
2176 for_each_subsys(root, ss) {
2177 if (ss == failed_ss) {
2178 if (cancel_failed_ss && ss->cancel_attach)
2179 ss->cancel_attach(ss, cgrp, leader);
2182 if (ss->cancel_attach)
2183 ss->cancel_attach(ss, cgrp, leader);
2186 /* clean up the array of referenced threads in the group. */
2187 for (i = 0; i < group_size; i++) {
2188 tsk = flex_array_get_ptr(group, i);
2189 put_task_struct(tsk);
2191 out_free_group_list:
2192 flex_array_free(group);
2196 static int cgroup_allow_attach(struct cgroup *cgrp, struct task_struct *tsk)
2198 struct cgroup_subsys *ss;
2201 for_each_subsys(cgrp->root, ss) {
2202 if (ss->allow_attach) {
2203 ret = ss->allow_attach(cgrp, tsk);
2215 * Find the task_struct of the task to attach by vpid and pass it along to the
2216 * function to attach either it or all tasks in its threadgroup. Will take
2217 * cgroup_mutex; may take task_lock of task.
2219 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2221 struct task_struct *tsk;
2222 const struct cred *cred = current_cred(), *tcred;
2225 if (!cgroup_lock_live_group(cgrp))
2230 tsk = find_task_by_vpid(pid);
2238 * RCU protects this access, since tsk was found in the
2239 * tid map. a race with de_thread may cause group_leader
2240 * to stop being the leader, but cgroup_attach_proc will
2243 tsk = tsk->group_leader;
2244 } else if (tsk->flags & PF_EXITING) {
2245 /* optimization for the single-task-only case */
2252 * even if we're attaching all tasks in the thread group, we
2253 * only need to check permissions on one of them.
2255 tcred = __task_cred(tsk);
2257 cred->euid != tcred->uid &&
2258 cred->euid != tcred->suid) {
2260 * if the default permission check fails, give each
2261 * cgroup a chance to extend the permission check
2263 ret = cgroup_allow_attach(cgrp, tsk);
2270 get_task_struct(tsk);
2274 tsk = current->group_leader;
2277 get_task_struct(tsk);
2281 threadgroup_fork_write_lock(tsk);
2282 ret = cgroup_attach_proc(cgrp, tsk);
2283 threadgroup_fork_write_unlock(tsk);
2285 ret = cgroup_attach_task(cgrp, tsk);
2287 put_task_struct(tsk);
2292 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2294 return attach_task_by_pid(cgrp, pid, false);
2297 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2302 * attach_proc fails with -EAGAIN if threadgroup leadership
2303 * changes in the middle of the operation, in which case we need
2304 * to find the task_struct for the new leader and start over.
2306 ret = attach_task_by_pid(cgrp, tgid, true);
2307 } while (ret == -EAGAIN);
2312 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2313 * @cgrp: the cgroup to be checked for liveness
2315 * On success, returns true; the lock should be later released with
2316 * cgroup_unlock(). On failure returns false with no lock held.
2318 bool cgroup_lock_live_group(struct cgroup *cgrp)
2320 mutex_lock(&cgroup_mutex);
2321 if (cgroup_is_removed(cgrp)) {
2322 mutex_unlock(&cgroup_mutex);
2327 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2329 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2332 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2333 if (strlen(buffer) >= PATH_MAX)
2335 if (!cgroup_lock_live_group(cgrp))
2337 strcpy(cgrp->root->release_agent_path, buffer);
2342 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2343 struct seq_file *seq)
2345 if (!cgroup_lock_live_group(cgrp))
2347 seq_puts(seq, cgrp->root->release_agent_path);
2348 seq_putc(seq, '\n');
2353 /* A buffer size big enough for numbers or short strings */
2354 #define CGROUP_LOCAL_BUFFER_SIZE 64
2356 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2358 const char __user *userbuf,
2359 size_t nbytes, loff_t *unused_ppos)
2361 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2367 if (nbytes >= sizeof(buffer))
2369 if (copy_from_user(buffer, userbuf, nbytes))
2372 buffer[nbytes] = 0; /* nul-terminate */
2373 if (cft->write_u64) {
2374 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2377 retval = cft->write_u64(cgrp, cft, val);
2379 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2382 retval = cft->write_s64(cgrp, cft, val);
2389 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2391 const char __user *userbuf,
2392 size_t nbytes, loff_t *unused_ppos)
2394 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2396 size_t max_bytes = cft->max_write_len;
2397 char *buffer = local_buffer;
2400 max_bytes = sizeof(local_buffer) - 1;
2401 if (nbytes >= max_bytes)
2403 /* Allocate a dynamic buffer if we need one */
2404 if (nbytes >= sizeof(local_buffer)) {
2405 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2409 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2414 buffer[nbytes] = 0; /* nul-terminate */
2415 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2419 if (buffer != local_buffer)
2424 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2425 size_t nbytes, loff_t *ppos)
2427 struct cftype *cft = __d_cft(file->f_dentry);
2428 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2430 if (cgroup_is_removed(cgrp))
2433 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2434 if (cft->write_u64 || cft->write_s64)
2435 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2436 if (cft->write_string)
2437 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2439 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2440 return ret ? ret : nbytes;
2445 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2447 char __user *buf, size_t nbytes,
2450 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2451 u64 val = cft->read_u64(cgrp, cft);
2452 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2454 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2457 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2459 char __user *buf, size_t nbytes,
2462 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2463 s64 val = cft->read_s64(cgrp, cft);
2464 int len = sprintf(tmp, "%lld\n", (long long) val);
2466 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2469 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2470 size_t nbytes, loff_t *ppos)
2472 struct cftype *cft = __d_cft(file->f_dentry);
2473 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2475 if (cgroup_is_removed(cgrp))
2479 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2481 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2483 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2488 * seqfile ops/methods for returning structured data. Currently just
2489 * supports string->u64 maps, but can be extended in future.
2492 struct cgroup_seqfile_state {
2494 struct cgroup *cgroup;
2497 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2499 struct seq_file *sf = cb->state;
2500 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2503 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2505 struct cgroup_seqfile_state *state = m->private;
2506 struct cftype *cft = state->cft;
2507 if (cft->read_map) {
2508 struct cgroup_map_cb cb = {
2509 .fill = cgroup_map_add,
2512 return cft->read_map(state->cgroup, cft, &cb);
2514 return cft->read_seq_string(state->cgroup, cft, m);
2517 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2519 struct seq_file *seq = file->private_data;
2520 kfree(seq->private);
2521 return single_release(inode, file);
2524 static const struct file_operations cgroup_seqfile_operations = {
2526 .write = cgroup_file_write,
2527 .llseek = seq_lseek,
2528 .release = cgroup_seqfile_release,
2531 static int cgroup_file_open(struct inode *inode, struct file *file)
2536 err = generic_file_open(inode, file);
2539 cft = __d_cft(file->f_dentry);
2541 if (cft->read_map || cft->read_seq_string) {
2542 struct cgroup_seqfile_state *state =
2543 kzalloc(sizeof(*state), GFP_USER);
2547 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2548 file->f_op = &cgroup_seqfile_operations;
2549 err = single_open(file, cgroup_seqfile_show, state);
2552 } else if (cft->open)
2553 err = cft->open(inode, file);
2560 static int cgroup_file_release(struct inode *inode, struct file *file)
2562 struct cftype *cft = __d_cft(file->f_dentry);
2564 return cft->release(inode, file);
2569 * cgroup_rename - Only allow simple rename of directories in place.
2571 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2572 struct inode *new_dir, struct dentry *new_dentry)
2574 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2576 if (new_dentry->d_inode)
2578 if (old_dir != new_dir)
2580 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2583 static const struct file_operations cgroup_file_operations = {
2584 .read = cgroup_file_read,
2585 .write = cgroup_file_write,
2586 .llseek = generic_file_llseek,
2587 .open = cgroup_file_open,
2588 .release = cgroup_file_release,
2591 static const struct inode_operations cgroup_dir_inode_operations = {
2592 .lookup = cgroup_lookup,
2593 .mkdir = cgroup_mkdir,
2594 .rmdir = cgroup_rmdir,
2595 .rename = cgroup_rename,
2598 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
2600 if (dentry->d_name.len > NAME_MAX)
2601 return ERR_PTR(-ENAMETOOLONG);
2602 d_add(dentry, NULL);
2607 * Check if a file is a control file
2609 static inline struct cftype *__file_cft(struct file *file)
2611 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2612 return ERR_PTR(-EINVAL);
2613 return __d_cft(file->f_dentry);
2616 static int cgroup_create_file(struct dentry *dentry, mode_t mode,
2617 struct super_block *sb)
2619 struct inode *inode;
2623 if (dentry->d_inode)
2626 inode = cgroup_new_inode(mode, sb);
2630 if (S_ISDIR(mode)) {
2631 inode->i_op = &cgroup_dir_inode_operations;
2632 inode->i_fop = &simple_dir_operations;
2634 /* start off with i_nlink == 2 (for "." entry) */
2637 /* start with the directory inode held, so that we can
2638 * populate it without racing with another mkdir */
2639 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2640 } else if (S_ISREG(mode)) {
2642 inode->i_fop = &cgroup_file_operations;
2644 d_instantiate(dentry, inode);
2645 dget(dentry); /* Extra count - pin the dentry in core */
2650 * cgroup_create_dir - create a directory for an object.
2651 * @cgrp: the cgroup we create the directory for. It must have a valid
2652 * ->parent field. And we are going to fill its ->dentry field.
2653 * @dentry: dentry of the new cgroup
2654 * @mode: mode to set on new directory.
2656 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2659 struct dentry *parent;
2662 parent = cgrp->parent->dentry;
2663 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2665 dentry->d_fsdata = cgrp;
2666 inc_nlink(parent->d_inode);
2667 rcu_assign_pointer(cgrp->dentry, dentry);
2674 * cgroup_file_mode - deduce file mode of a control file
2675 * @cft: the control file in question
2677 * returns cft->mode if ->mode is not 0
2678 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2679 * returns S_IRUGO if it has only a read handler
2680 * returns S_IWUSR if it has only a write hander
2682 static mode_t cgroup_file_mode(const struct cftype *cft)
2689 if (cft->read || cft->read_u64 || cft->read_s64 ||
2690 cft->read_map || cft->read_seq_string)
2693 if (cft->write || cft->write_u64 || cft->write_s64 ||
2694 cft->write_string || cft->trigger)
2700 int cgroup_add_file(struct cgroup *cgrp,
2701 struct cgroup_subsys *subsys,
2702 const struct cftype *cft)
2704 struct dentry *dir = cgrp->dentry;
2705 struct dentry *dentry;
2709 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2710 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2711 strcpy(name, subsys->name);
2714 strcat(name, cft->name);
2715 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2716 dentry = lookup_one_len(name, dir, strlen(name));
2717 if (!IS_ERR(dentry)) {
2718 mode = cgroup_file_mode(cft);
2719 error = cgroup_create_file(dentry, mode | S_IFREG,
2722 dentry->d_fsdata = (void *)cft;
2725 error = PTR_ERR(dentry);
2728 EXPORT_SYMBOL_GPL(cgroup_add_file);
2730 int cgroup_add_files(struct cgroup *cgrp,
2731 struct cgroup_subsys *subsys,
2732 const struct cftype cft[],
2736 for (i = 0; i < count; i++) {
2737 err = cgroup_add_file(cgrp, subsys, &cft[i]);
2743 EXPORT_SYMBOL_GPL(cgroup_add_files);
2746 * cgroup_task_count - count the number of tasks in a cgroup.
2747 * @cgrp: the cgroup in question
2749 * Return the number of tasks in the cgroup.
2751 int cgroup_task_count(const struct cgroup *cgrp)
2754 struct cg_cgroup_link *link;
2756 read_lock(&css_set_lock);
2757 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2758 count += atomic_read(&link->cg->refcount);
2760 read_unlock(&css_set_lock);
2765 * Advance a list_head iterator. The iterator should be positioned at
2766 * the start of a css_set
2768 static void cgroup_advance_iter(struct cgroup *cgrp,
2769 struct cgroup_iter *it)
2771 struct list_head *l = it->cg_link;
2772 struct cg_cgroup_link *link;
2775 /* Advance to the next non-empty css_set */
2778 if (l == &cgrp->css_sets) {
2782 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2784 } while (list_empty(&cg->tasks));
2786 it->task = cg->tasks.next;
2790 * To reduce the fork() overhead for systems that are not actually
2791 * using their cgroups capability, we don't maintain the lists running
2792 * through each css_set to its tasks until we see the list actually
2793 * used - in other words after the first call to cgroup_iter_start().
2795 * The tasklist_lock is not held here, as do_each_thread() and
2796 * while_each_thread() are protected by RCU.
2798 static void cgroup_enable_task_cg_lists(void)
2800 struct task_struct *p, *g;
2801 write_lock(&css_set_lock);
2802 use_task_css_set_links = 1;
2803 do_each_thread(g, p) {
2806 * We should check if the process is exiting, otherwise
2807 * it will race with cgroup_exit() in that the list
2808 * entry won't be deleted though the process has exited.
2810 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2811 list_add(&p->cg_list, &p->cgroups->tasks);
2813 } while_each_thread(g, p);
2814 write_unlock(&css_set_lock);
2817 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2820 * The first time anyone tries to iterate across a cgroup,
2821 * we need to enable the list linking each css_set to its
2822 * tasks, and fix up all existing tasks.
2824 if (!use_task_css_set_links)
2825 cgroup_enable_task_cg_lists();
2827 read_lock(&css_set_lock);
2828 it->cg_link = &cgrp->css_sets;
2829 cgroup_advance_iter(cgrp, it);
2832 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2833 struct cgroup_iter *it)
2835 struct task_struct *res;
2836 struct list_head *l = it->task;
2837 struct cg_cgroup_link *link;
2839 /* If the iterator cg is NULL, we have no tasks */
2842 res = list_entry(l, struct task_struct, cg_list);
2843 /* Advance iterator to find next entry */
2845 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2846 if (l == &link->cg->tasks) {
2847 /* We reached the end of this task list - move on to
2848 * the next cg_cgroup_link */
2849 cgroup_advance_iter(cgrp, it);
2856 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2858 read_unlock(&css_set_lock);
2861 static inline int started_after_time(struct task_struct *t1,
2862 struct timespec *time,
2863 struct task_struct *t2)
2865 int start_diff = timespec_compare(&t1->start_time, time);
2866 if (start_diff > 0) {
2868 } else if (start_diff < 0) {
2872 * Arbitrarily, if two processes started at the same
2873 * time, we'll say that the lower pointer value
2874 * started first. Note that t2 may have exited by now
2875 * so this may not be a valid pointer any longer, but
2876 * that's fine - it still serves to distinguish
2877 * between two tasks started (effectively) simultaneously.
2884 * This function is a callback from heap_insert() and is used to order
2886 * In this case we order the heap in descending task start time.
2888 static inline int started_after(void *p1, void *p2)
2890 struct task_struct *t1 = p1;
2891 struct task_struct *t2 = p2;
2892 return started_after_time(t1, &t2->start_time, t2);
2896 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2897 * @scan: struct cgroup_scanner containing arguments for the scan
2899 * Arguments include pointers to callback functions test_task() and
2901 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2902 * and if it returns true, call process_task() for it also.
2903 * The test_task pointer may be NULL, meaning always true (select all tasks).
2904 * Effectively duplicates cgroup_iter_{start,next,end}()
2905 * but does not lock css_set_lock for the call to process_task().
2906 * The struct cgroup_scanner may be embedded in any structure of the caller's
2908 * It is guaranteed that process_task() will act on every task that
2909 * is a member of the cgroup for the duration of this call. This
2910 * function may or may not call process_task() for tasks that exit
2911 * or move to a different cgroup during the call, or are forked or
2912 * move into the cgroup during the call.
2914 * Note that test_task() may be called with locks held, and may in some
2915 * situations be called multiple times for the same task, so it should
2917 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2918 * pre-allocated and will be used for heap operations (and its "gt" member will
2919 * be overwritten), else a temporary heap will be used (allocation of which
2920 * may cause this function to fail).
2922 int cgroup_scan_tasks(struct cgroup_scanner *scan)
2925 struct cgroup_iter it;
2926 struct task_struct *p, *dropped;
2927 /* Never dereference latest_task, since it's not refcounted */
2928 struct task_struct *latest_task = NULL;
2929 struct ptr_heap tmp_heap;
2930 struct ptr_heap *heap;
2931 struct timespec latest_time = { 0, 0 };
2934 /* The caller supplied our heap and pre-allocated its memory */
2936 heap->gt = &started_after;
2938 /* We need to allocate our own heap memory */
2940 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2942 /* cannot allocate the heap */
2948 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2949 * to determine which are of interest, and using the scanner's
2950 * "process_task" callback to process any of them that need an update.
2951 * Since we don't want to hold any locks during the task updates,
2952 * gather tasks to be processed in a heap structure.
2953 * The heap is sorted by descending task start time.
2954 * If the statically-sized heap fills up, we overflow tasks that
2955 * started later, and in future iterations only consider tasks that
2956 * started after the latest task in the previous pass. This
2957 * guarantees forward progress and that we don't miss any tasks.
2960 cgroup_iter_start(scan->cg, &it);
2961 while ((p = cgroup_iter_next(scan->cg, &it))) {
2963 * Only affect tasks that qualify per the caller's callback,
2964 * if he provided one
2966 if (scan->test_task && !scan->test_task(p, scan))
2969 * Only process tasks that started after the last task
2972 if (!started_after_time(p, &latest_time, latest_task))
2974 dropped = heap_insert(heap, p);
2975 if (dropped == NULL) {
2977 * The new task was inserted; the heap wasn't
2981 } else if (dropped != p) {
2983 * The new task was inserted, and pushed out a
2987 put_task_struct(dropped);
2990 * Else the new task was newer than anything already in
2991 * the heap and wasn't inserted
2994 cgroup_iter_end(scan->cg, &it);
2997 for (i = 0; i < heap->size; i++) {
2998 struct task_struct *q = heap->ptrs[i];
3000 latest_time = q->start_time;
3003 /* Process the task per the caller's callback */
3004 scan->process_task(q, scan);
3008 * If we had to process any tasks at all, scan again
3009 * in case some of them were in the middle of forking
3010 * children that didn't get processed.
3011 * Not the most efficient way to do it, but it avoids
3012 * having to take callback_mutex in the fork path
3016 if (heap == &tmp_heap)
3017 heap_free(&tmp_heap);
3022 * Stuff for reading the 'tasks'/'procs' files.
3024 * Reading this file can return large amounts of data if a cgroup has
3025 * *lots* of attached tasks. So it may need several calls to read(),
3026 * but we cannot guarantee that the information we produce is correct
3027 * unless we produce it entirely atomically.
3032 * The following two functions "fix" the issue where there are more pids
3033 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3034 * TODO: replace with a kernel-wide solution to this problem
3036 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3037 static void *pidlist_allocate(int count)
3039 if (PIDLIST_TOO_LARGE(count))
3040 return vmalloc(count * sizeof(pid_t));
3042 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3044 static void pidlist_free(void *p)
3046 if (is_vmalloc_addr(p))
3051 static void *pidlist_resize(void *p, int newcount)
3054 /* note: if new alloc fails, old p will still be valid either way */
3055 if (is_vmalloc_addr(p)) {
3056 newlist = vmalloc(newcount * sizeof(pid_t));
3059 memcpy(newlist, p, newcount * sizeof(pid_t));
3062 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3068 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3069 * If the new stripped list is sufficiently smaller and there's enough memory
3070 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3071 * number of unique elements.
3073 /* is the size difference enough that we should re-allocate the array? */
3074 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3075 static int pidlist_uniq(pid_t **p, int length)
3082 * we presume the 0th element is unique, so i starts at 1. trivial
3083 * edge cases first; no work needs to be done for either
3085 if (length == 0 || length == 1)
3087 /* src and dest walk down the list; dest counts unique elements */
3088 for (src = 1; src < length; src++) {
3089 /* find next unique element */
3090 while (list[src] == list[src-1]) {
3095 /* dest always points to where the next unique element goes */
3096 list[dest] = list[src];
3101 * if the length difference is large enough, we want to allocate a
3102 * smaller buffer to save memory. if this fails due to out of memory,
3103 * we'll just stay with what we've got.
3105 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3106 newlist = pidlist_resize(list, dest);
3113 static int cmppid(const void *a, const void *b)
3115 return *(pid_t *)a - *(pid_t *)b;
3119 * find the appropriate pidlist for our purpose (given procs vs tasks)
3120 * returns with the lock on that pidlist already held, and takes care
3121 * of the use count, or returns NULL with no locks held if we're out of
3124 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3125 enum cgroup_filetype type)
3127 struct cgroup_pidlist *l;
3128 /* don't need task_nsproxy() if we're looking at ourself */
3129 struct pid_namespace *ns = current->nsproxy->pid_ns;
3132 * We can't drop the pidlist_mutex before taking the l->mutex in case
3133 * the last ref-holder is trying to remove l from the list at the same
3134 * time. Holding the pidlist_mutex precludes somebody taking whichever
3135 * list we find out from under us - compare release_pid_array().
3137 mutex_lock(&cgrp->pidlist_mutex);
3138 list_for_each_entry(l, &cgrp->pidlists, links) {
3139 if (l->key.type == type && l->key.ns == ns) {
3140 /* make sure l doesn't vanish out from under us */
3141 down_write(&l->mutex);
3142 mutex_unlock(&cgrp->pidlist_mutex);
3146 /* entry not found; create a new one */
3147 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3149 mutex_unlock(&cgrp->pidlist_mutex);
3152 init_rwsem(&l->mutex);
3153 down_write(&l->mutex);
3155 l->key.ns = get_pid_ns(ns);
3156 l->use_count = 0; /* don't increment here */
3159 list_add(&l->links, &cgrp->pidlists);
3160 mutex_unlock(&cgrp->pidlist_mutex);
3165 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3167 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3168 struct cgroup_pidlist **lp)
3172 int pid, n = 0; /* used for populating the array */
3173 struct cgroup_iter it;
3174 struct task_struct *tsk;
3175 struct cgroup_pidlist *l;
3178 * If cgroup gets more users after we read count, we won't have
3179 * enough space - tough. This race is indistinguishable to the
3180 * caller from the case that the additional cgroup users didn't
3181 * show up until sometime later on.
3183 length = cgroup_task_count(cgrp);
3184 array = pidlist_allocate(length);
3187 /* now, populate the array */
3188 cgroup_iter_start(cgrp, &it);
3189 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3190 if (unlikely(n == length))
3192 /* get tgid or pid for procs or tasks file respectively */
3193 if (type == CGROUP_FILE_PROCS)
3194 pid = task_tgid_vnr(tsk);
3196 pid = task_pid_vnr(tsk);
3197 if (pid > 0) /* make sure to only use valid results */
3200 cgroup_iter_end(cgrp, &it);
3202 /* now sort & (if procs) strip out duplicates */
3203 sort(array, length, sizeof(pid_t), cmppid, NULL);
3204 if (type == CGROUP_FILE_PROCS)
3205 length = pidlist_uniq(&array, length);
3206 l = cgroup_pidlist_find(cgrp, type);
3208 pidlist_free(array);
3211 /* store array, freeing old if necessary - lock already held */
3212 pidlist_free(l->list);
3216 up_write(&l->mutex);
3222 * cgroupstats_build - build and fill cgroupstats
3223 * @stats: cgroupstats to fill information into
3224 * @dentry: A dentry entry belonging to the cgroup for which stats have
3227 * Build and fill cgroupstats so that taskstats can export it to user
3230 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3233 struct cgroup *cgrp;
3234 struct cgroup_iter it;
3235 struct task_struct *tsk;
3238 * Validate dentry by checking the superblock operations,
3239 * and make sure it's a directory.
3241 if (dentry->d_sb->s_op != &cgroup_ops ||
3242 !S_ISDIR(dentry->d_inode->i_mode))
3246 cgrp = dentry->d_fsdata;
3248 cgroup_iter_start(cgrp, &it);
3249 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3250 switch (tsk->state) {
3252 stats->nr_running++;
3254 case TASK_INTERRUPTIBLE:
3255 stats->nr_sleeping++;
3257 case TASK_UNINTERRUPTIBLE:
3258 stats->nr_uninterruptible++;
3261 stats->nr_stopped++;
3264 if (delayacct_is_task_waiting_on_io(tsk))
3265 stats->nr_io_wait++;
3269 cgroup_iter_end(cgrp, &it);
3277 * seq_file methods for the tasks/procs files. The seq_file position is the
3278 * next pid to display; the seq_file iterator is a pointer to the pid
3279 * in the cgroup->l->list array.
3282 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3285 * Initially we receive a position value that corresponds to
3286 * one more than the last pid shown (or 0 on the first call or
3287 * after a seek to the start). Use a binary-search to find the
3288 * next pid to display, if any
3290 struct cgroup_pidlist *l = s->private;
3291 int index = 0, pid = *pos;
3294 down_read(&l->mutex);
3296 int end = l->length;
3298 while (index < end) {
3299 int mid = (index + end) / 2;
3300 if (l->list[mid] == pid) {
3303 } else if (l->list[mid] <= pid)
3309 /* If we're off the end of the array, we're done */
3310 if (index >= l->length)
3312 /* Update the abstract position to be the actual pid that we found */
3313 iter = l->list + index;
3318 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3320 struct cgroup_pidlist *l = s->private;
3324 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3326 struct cgroup_pidlist *l = s->private;
3328 pid_t *end = l->list + l->length;
3330 * Advance to the next pid in the array. If this goes off the
3342 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3344 return seq_printf(s, "%d\n", *(int *)v);
3348 * seq_operations functions for iterating on pidlists through seq_file -
3349 * independent of whether it's tasks or procs
3351 static const struct seq_operations cgroup_pidlist_seq_operations = {
3352 .start = cgroup_pidlist_start,
3353 .stop = cgroup_pidlist_stop,
3354 .next = cgroup_pidlist_next,
3355 .show = cgroup_pidlist_show,
3358 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3361 * the case where we're the last user of this particular pidlist will
3362 * have us remove it from the cgroup's list, which entails taking the
3363 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3364 * pidlist_mutex, we have to take pidlist_mutex first.
3366 mutex_lock(&l->owner->pidlist_mutex);
3367 down_write(&l->mutex);
3368 BUG_ON(!l->use_count);
3369 if (!--l->use_count) {
3370 /* we're the last user if refcount is 0; remove and free */
3371 list_del(&l->links);
3372 mutex_unlock(&l->owner->pidlist_mutex);
3373 pidlist_free(l->list);
3374 put_pid_ns(l->key.ns);
3375 up_write(&l->mutex);
3379 mutex_unlock(&l->owner->pidlist_mutex);
3380 up_write(&l->mutex);
3383 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3385 struct cgroup_pidlist *l;
3386 if (!(file->f_mode & FMODE_READ))
3389 * the seq_file will only be initialized if the file was opened for
3390 * reading; hence we check if it's not null only in that case.
3392 l = ((struct seq_file *)file->private_data)->private;
3393 cgroup_release_pid_array(l);
3394 return seq_release(inode, file);
3397 static const struct file_operations cgroup_pidlist_operations = {
3399 .llseek = seq_lseek,
3400 .write = cgroup_file_write,
3401 .release = cgroup_pidlist_release,
3405 * The following functions handle opens on a file that displays a pidlist
3406 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3409 /* helper function for the two below it */
3410 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3412 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3413 struct cgroup_pidlist *l;
3416 /* Nothing to do for write-only files */
3417 if (!(file->f_mode & FMODE_READ))
3420 /* have the array populated */
3421 retval = pidlist_array_load(cgrp, type, &l);
3424 /* configure file information */
3425 file->f_op = &cgroup_pidlist_operations;
3427 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3429 cgroup_release_pid_array(l);
3432 ((struct seq_file *)file->private_data)->private = l;
3435 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3437 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3439 static int cgroup_procs_open(struct inode *unused, struct file *file)
3441 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3444 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3447 return notify_on_release(cgrp);
3450 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3454 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3456 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3458 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3463 * Unregister event and free resources.
3465 * Gets called from workqueue.
3467 static void cgroup_event_remove(struct work_struct *work)
3469 struct cgroup_event *event = container_of(work, struct cgroup_event,
3471 struct cgroup *cgrp = event->cgrp;
3473 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3475 eventfd_ctx_put(event->eventfd);
3481 * Gets called on POLLHUP on eventfd when user closes it.
3483 * Called with wqh->lock held and interrupts disabled.
3485 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3486 int sync, void *key)
3488 struct cgroup_event *event = container_of(wait,
3489 struct cgroup_event, wait);
3490 struct cgroup *cgrp = event->cgrp;
3491 unsigned long flags = (unsigned long)key;
3493 if (flags & POLLHUP) {
3494 __remove_wait_queue(event->wqh, &event->wait);
3495 spin_lock(&cgrp->event_list_lock);
3496 list_del(&event->list);
3497 spin_unlock(&cgrp->event_list_lock);
3499 * We are in atomic context, but cgroup_event_remove() may
3500 * sleep, so we have to call it in workqueue.
3502 schedule_work(&event->remove);
3508 static void cgroup_event_ptable_queue_proc(struct file *file,
3509 wait_queue_head_t *wqh, poll_table *pt)
3511 struct cgroup_event *event = container_of(pt,
3512 struct cgroup_event, pt);
3515 add_wait_queue(wqh, &event->wait);
3519 * Parse input and register new cgroup event handler.
3521 * Input must be in format '<event_fd> <control_fd> <args>'.
3522 * Interpretation of args is defined by control file implementation.
3524 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3527 struct cgroup_event *event = NULL;
3528 unsigned int efd, cfd;
3529 struct file *efile = NULL;
3530 struct file *cfile = NULL;
3534 efd = simple_strtoul(buffer, &endp, 10);
3539 cfd = simple_strtoul(buffer, &endp, 10);
3540 if ((*endp != ' ') && (*endp != '\0'))
3544 event = kzalloc(sizeof(*event), GFP_KERNEL);
3548 INIT_LIST_HEAD(&event->list);
3549 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3550 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3551 INIT_WORK(&event->remove, cgroup_event_remove);
3553 efile = eventfd_fget(efd);
3554 if (IS_ERR(efile)) {
3555 ret = PTR_ERR(efile);
3559 event->eventfd = eventfd_ctx_fileget(efile);
3560 if (IS_ERR(event->eventfd)) {
3561 ret = PTR_ERR(event->eventfd);
3571 /* the process need read permission on control file */
3572 ret = file_permission(cfile, MAY_READ);
3576 event->cft = __file_cft(cfile);
3577 if (IS_ERR(event->cft)) {
3578 ret = PTR_ERR(event->cft);
3582 if (!event->cft->register_event || !event->cft->unregister_event) {
3587 ret = event->cft->register_event(cgrp, event->cft,
3588 event->eventfd, buffer);
3592 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3593 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3599 * Events should be removed after rmdir of cgroup directory, but before
3600 * destroying subsystem state objects. Let's take reference to cgroup
3601 * directory dentry to do that.
3605 spin_lock(&cgrp->event_list_lock);
3606 list_add(&event->list, &cgrp->event_list);
3607 spin_unlock(&cgrp->event_list_lock);
3618 if (event && event->eventfd && !IS_ERR(event->eventfd))
3619 eventfd_ctx_put(event->eventfd);
3621 if (!IS_ERR_OR_NULL(efile))
3629 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3632 return clone_children(cgrp);
3635 static int cgroup_clone_children_write(struct cgroup *cgrp,
3640 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3642 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3647 * for the common functions, 'private' gives the type of file
3649 /* for hysterical raisins, we can't put this on the older files */
3650 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3651 static struct cftype files[] = {
3654 .open = cgroup_tasks_open,
3655 .write_u64 = cgroup_tasks_write,
3656 .release = cgroup_pidlist_release,
3657 .mode = S_IRUGO | S_IWUSR,
3660 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3661 .open = cgroup_procs_open,
3662 .write_u64 = cgroup_procs_write,
3663 .release = cgroup_pidlist_release,
3664 .mode = S_IRUGO | S_IWUSR,
3667 .name = "notify_on_release",
3668 .read_u64 = cgroup_read_notify_on_release,
3669 .write_u64 = cgroup_write_notify_on_release,
3672 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3673 .write_string = cgroup_write_event_control,
3677 .name = "cgroup.clone_children",
3678 .read_u64 = cgroup_clone_children_read,
3679 .write_u64 = cgroup_clone_children_write,
3683 static struct cftype cft_release_agent = {
3684 .name = "release_agent",
3685 .read_seq_string = cgroup_release_agent_show,
3686 .write_string = cgroup_release_agent_write,
3687 .max_write_len = PATH_MAX,
3690 static int cgroup_populate_dir(struct cgroup *cgrp)
3693 struct cgroup_subsys *ss;
3695 /* First clear out any existing files */
3696 cgroup_clear_directory(cgrp->dentry);
3698 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
3702 if (cgrp == cgrp->top_cgroup) {
3703 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
3707 for_each_subsys(cgrp->root, ss) {
3708 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
3711 /* This cgroup is ready now */
3712 for_each_subsys(cgrp->root, ss) {
3713 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3715 * Update id->css pointer and make this css visible from
3716 * CSS ID functions. This pointer will be dereferened
3717 * from RCU-read-side without locks.
3720 rcu_assign_pointer(css->id->css, css);
3726 static void init_cgroup_css(struct cgroup_subsys_state *css,
3727 struct cgroup_subsys *ss,
3728 struct cgroup *cgrp)
3731 atomic_set(&css->refcnt, 1);
3734 if (cgrp == dummytop)
3735 set_bit(CSS_ROOT, &css->flags);
3736 BUG_ON(cgrp->subsys[ss->subsys_id]);
3737 cgrp->subsys[ss->subsys_id] = css;
3740 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3742 /* We need to take each hierarchy_mutex in a consistent order */
3746 * No worry about a race with rebind_subsystems that might mess up the
3747 * locking order, since both parties are under cgroup_mutex.
3749 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3750 struct cgroup_subsys *ss = subsys[i];
3753 if (ss->root == root)
3754 mutex_lock(&ss->hierarchy_mutex);
3758 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3762 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3763 struct cgroup_subsys *ss = subsys[i];
3766 if (ss->root == root)
3767 mutex_unlock(&ss->hierarchy_mutex);
3772 * cgroup_create - create a cgroup
3773 * @parent: cgroup that will be parent of the new cgroup
3774 * @dentry: dentry of the new cgroup
3775 * @mode: mode to set on new inode
3777 * Must be called with the mutex on the parent inode held
3779 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3782 struct cgroup *cgrp;
3783 struct cgroupfs_root *root = parent->root;
3785 struct cgroup_subsys *ss;
3786 struct super_block *sb = root->sb;
3788 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3792 /* Grab a reference on the superblock so the hierarchy doesn't
3793 * get deleted on unmount if there are child cgroups. This
3794 * can be done outside cgroup_mutex, since the sb can't
3795 * disappear while someone has an open control file on the
3797 atomic_inc(&sb->s_active);
3799 mutex_lock(&cgroup_mutex);
3801 init_cgroup_housekeeping(cgrp);
3803 cgrp->parent = parent;
3804 cgrp->root = parent->root;
3805 cgrp->top_cgroup = parent->top_cgroup;
3807 if (notify_on_release(parent))
3808 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3810 if (clone_children(parent))
3811 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3813 for_each_subsys(root, ss) {
3814 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
3820 init_cgroup_css(css, ss, cgrp);
3822 err = alloc_css_id(ss, parent, cgrp);
3826 /* At error, ->destroy() callback has to free assigned ID. */
3827 if (clone_children(parent) && ss->post_clone)
3828 ss->post_clone(ss, cgrp);
3831 cgroup_lock_hierarchy(root);
3832 list_add(&cgrp->sibling, &cgrp->parent->children);
3833 cgroup_unlock_hierarchy(root);
3834 root->number_of_cgroups++;
3836 err = cgroup_create_dir(cgrp, dentry, mode);
3840 set_bit(CGRP_RELEASABLE, &parent->flags);
3842 /* The cgroup directory was pre-locked for us */
3843 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3845 err = cgroup_populate_dir(cgrp);
3846 /* If err < 0, we have a half-filled directory - oh well ;) */
3848 mutex_unlock(&cgroup_mutex);
3849 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3855 cgroup_lock_hierarchy(root);
3856 list_del(&cgrp->sibling);
3857 cgroup_unlock_hierarchy(root);
3858 root->number_of_cgroups--;
3862 for_each_subsys(root, ss) {
3863 if (cgrp->subsys[ss->subsys_id])
3864 ss->destroy(ss, cgrp);
3867 mutex_unlock(&cgroup_mutex);
3869 /* Release the reference count that we took on the superblock */
3870 deactivate_super(sb);
3876 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3878 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3880 /* the vfs holds inode->i_mutex already */
3881 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3884 static int cgroup_has_css_refs(struct cgroup *cgrp)
3886 /* Check the reference count on each subsystem. Since we
3887 * already established that there are no tasks in the
3888 * cgroup, if the css refcount is also 1, then there should
3889 * be no outstanding references, so the subsystem is safe to
3890 * destroy. We scan across all subsystems rather than using
3891 * the per-hierarchy linked list of mounted subsystems since
3892 * we can be called via check_for_release() with no
3893 * synchronization other than RCU, and the subsystem linked
3894 * list isn't RCU-safe */
3897 * We won't need to lock the subsys array, because the subsystems
3898 * we're concerned about aren't going anywhere since our cgroup root
3899 * has a reference on them.
3901 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3902 struct cgroup_subsys *ss = subsys[i];
3903 struct cgroup_subsys_state *css;
3904 /* Skip subsystems not present or not in this hierarchy */
3905 if (ss == NULL || ss->root != cgrp->root)
3907 css = cgrp->subsys[ss->subsys_id];
3908 /* When called from check_for_release() it's possible
3909 * that by this point the cgroup has been removed
3910 * and the css deleted. But a false-positive doesn't
3911 * matter, since it can only happen if the cgroup
3912 * has been deleted and hence no longer needs the
3913 * release agent to be called anyway. */
3914 if (css && (atomic_read(&css->refcnt) > 1))
3921 * Atomically mark all (or else none) of the cgroup's CSS objects as
3922 * CSS_REMOVED. Return true on success, or false if the cgroup has
3923 * busy subsystems. Call with cgroup_mutex held
3926 static int cgroup_clear_css_refs(struct cgroup *cgrp)
3928 struct cgroup_subsys *ss;
3929 unsigned long flags;
3930 bool failed = false;
3931 local_irq_save(flags);
3932 for_each_subsys(cgrp->root, ss) {
3933 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3936 /* We can only remove a CSS with a refcnt==1 */
3937 refcnt = atomic_read(&css->refcnt);
3944 * Drop the refcnt to 0 while we check other
3945 * subsystems. This will cause any racing
3946 * css_tryget() to spin until we set the
3947 * CSS_REMOVED bits or abort
3949 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3955 for_each_subsys(cgrp->root, ss) {
3956 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3959 * Restore old refcnt if we previously managed
3960 * to clear it from 1 to 0
3962 if (!atomic_read(&css->refcnt))
3963 atomic_set(&css->refcnt, 1);
3965 /* Commit the fact that the CSS is removed */
3966 set_bit(CSS_REMOVED, &css->flags);
3969 local_irq_restore(flags);
3973 /* checks if all of the css_sets attached to a cgroup have a refcount of 0.
3974 * Must be called with css_set_lock held */
3975 static int cgroup_css_sets_empty(struct cgroup *cgrp)
3977 struct cg_cgroup_link *link;
3979 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
3980 struct css_set *cg = link->cg;
3981 if (atomic_read(&cg->refcount) > 0)
3988 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3990 struct cgroup *cgrp = dentry->d_fsdata;
3992 struct cgroup *parent;
3994 struct cgroup_event *event, *tmp;
3997 /* the vfs holds both inode->i_mutex already */
3999 mutex_lock(&cgroup_mutex);
4000 if (!cgroup_css_sets_empty(cgrp)) {
4001 mutex_unlock(&cgroup_mutex);
4004 if (!list_empty(&cgrp->children)) {
4005 mutex_unlock(&cgroup_mutex);
4008 mutex_unlock(&cgroup_mutex);
4011 * In general, subsystem has no css->refcnt after pre_destroy(). But
4012 * in racy cases, subsystem may have to get css->refcnt after
4013 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4014 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4015 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4016 * and subsystem's reference count handling. Please see css_get/put
4017 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4019 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4022 * Call pre_destroy handlers of subsys. Notify subsystems
4023 * that rmdir() request comes.
4025 ret = cgroup_call_pre_destroy(cgrp);
4027 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4031 mutex_lock(&cgroup_mutex);
4032 parent = cgrp->parent;
4033 if (!cgroup_css_sets_empty(cgrp) || !list_empty(&cgrp->children)) {
4034 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4035 mutex_unlock(&cgroup_mutex);
4038 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
4039 if (!cgroup_clear_css_refs(cgrp)) {
4040 mutex_unlock(&cgroup_mutex);
4042 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4043 * prepare_to_wait(), we need to check this flag.
4045 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
4047 finish_wait(&cgroup_rmdir_waitq, &wait);
4048 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4049 if (signal_pending(current))
4053 /* NO css_tryget() can success after here. */
4054 finish_wait(&cgroup_rmdir_waitq, &wait);
4055 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4057 spin_lock(&release_list_lock);
4058 set_bit(CGRP_REMOVED, &cgrp->flags);
4059 if (!list_empty(&cgrp->release_list))
4060 list_del_init(&cgrp->release_list);
4061 spin_unlock(&release_list_lock);
4063 cgroup_lock_hierarchy(cgrp->root);
4064 /* delete this cgroup from parent->children */
4065 list_del_init(&cgrp->sibling);
4066 cgroup_unlock_hierarchy(cgrp->root);
4068 d = dget(cgrp->dentry);
4070 cgroup_d_remove_dir(d);
4073 check_for_release(parent);
4076 * Unregister events and notify userspace.
4077 * Notify userspace about cgroup removing only after rmdir of cgroup
4078 * directory to avoid race between userspace and kernelspace
4080 spin_lock(&cgrp->event_list_lock);
4081 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4082 list_del(&event->list);
4083 remove_wait_queue(event->wqh, &event->wait);
4084 eventfd_signal(event->eventfd, 1);
4085 schedule_work(&event->remove);
4087 spin_unlock(&cgrp->event_list_lock);
4089 mutex_unlock(&cgroup_mutex);
4093 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4095 struct cgroup_subsys_state *css;
4097 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4099 /* Create the top cgroup state for this subsystem */
4100 list_add(&ss->sibling, &rootnode.subsys_list);
4101 ss->root = &rootnode;
4102 css = ss->create(ss, dummytop);
4103 /* We don't handle early failures gracefully */
4104 BUG_ON(IS_ERR(css));
4105 init_cgroup_css(css, ss, dummytop);
4107 /* Update the init_css_set to contain a subsys
4108 * pointer to this state - since the subsystem is
4109 * newly registered, all tasks and hence the
4110 * init_css_set is in the subsystem's top cgroup. */
4111 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4113 need_forkexit_callback |= ss->fork || ss->exit;
4115 /* At system boot, before all subsystems have been
4116 * registered, no tasks have been forked, so we don't
4117 * need to invoke fork callbacks here. */
4118 BUG_ON(!list_empty(&init_task.tasks));
4120 mutex_init(&ss->hierarchy_mutex);
4121 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4124 /* this function shouldn't be used with modular subsystems, since they
4125 * need to register a subsys_id, among other things */
4130 * cgroup_load_subsys: load and register a modular subsystem at runtime
4131 * @ss: the subsystem to load
4133 * This function should be called in a modular subsystem's initcall. If the
4134 * subsystem is built as a module, it will be assigned a new subsys_id and set
4135 * up for use. If the subsystem is built-in anyway, work is delegated to the
4136 * simpler cgroup_init_subsys.
4138 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4141 struct cgroup_subsys_state *css;
4143 /* check name and function validity */
4144 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4145 ss->create == NULL || ss->destroy == NULL)
4149 * we don't support callbacks in modular subsystems. this check is
4150 * before the ss->module check for consistency; a subsystem that could
4151 * be a module should still have no callbacks even if the user isn't
4152 * compiling it as one.
4154 if (ss->fork || ss->exit)
4158 * an optionally modular subsystem is built-in: we want to do nothing,
4159 * since cgroup_init_subsys will have already taken care of it.
4161 if (ss->module == NULL) {
4162 /* a few sanity checks */
4163 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4164 BUG_ON(subsys[ss->subsys_id] != ss);
4169 * need to register a subsys id before anything else - for example,
4170 * init_cgroup_css needs it.
4172 mutex_lock(&cgroup_mutex);
4173 /* find the first empty slot in the array */
4174 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4175 if (subsys[i] == NULL)
4178 if (i == CGROUP_SUBSYS_COUNT) {
4179 /* maximum number of subsystems already registered! */
4180 mutex_unlock(&cgroup_mutex);
4183 /* assign ourselves the subsys_id */
4188 * no ss->create seems to need anything important in the ss struct, so
4189 * this can happen first (i.e. before the rootnode attachment).
4191 css = ss->create(ss, dummytop);
4193 /* failure case - need to deassign the subsys[] slot. */
4195 mutex_unlock(&cgroup_mutex);
4196 return PTR_ERR(css);
4199 list_add(&ss->sibling, &rootnode.subsys_list);
4200 ss->root = &rootnode;
4202 /* our new subsystem will be attached to the dummy hierarchy. */
4203 init_cgroup_css(css, ss, dummytop);
4204 /* init_idr must be after init_cgroup_css because it sets css->id. */
4206 int ret = cgroup_init_idr(ss, css);
4208 dummytop->subsys[ss->subsys_id] = NULL;
4209 ss->destroy(ss, dummytop);
4211 mutex_unlock(&cgroup_mutex);
4217 * Now we need to entangle the css into the existing css_sets. unlike
4218 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4219 * will need a new pointer to it; done by iterating the css_set_table.
4220 * furthermore, modifying the existing css_sets will corrupt the hash
4221 * table state, so each changed css_set will need its hash recomputed.
4222 * this is all done under the css_set_lock.
4224 write_lock(&css_set_lock);
4225 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4227 struct hlist_node *node, *tmp;
4228 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4230 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4231 /* skip entries that we already rehashed */
4232 if (cg->subsys[ss->subsys_id])
4234 /* remove existing entry */
4235 hlist_del(&cg->hlist);
4237 cg->subsys[ss->subsys_id] = css;
4238 /* recompute hash and restore entry */
4239 new_bucket = css_set_hash(cg->subsys);
4240 hlist_add_head(&cg->hlist, new_bucket);
4243 write_unlock(&css_set_lock);
4245 mutex_init(&ss->hierarchy_mutex);
4246 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4250 mutex_unlock(&cgroup_mutex);
4253 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4256 * cgroup_unload_subsys: unload a modular subsystem
4257 * @ss: the subsystem to unload
4259 * This function should be called in a modular subsystem's exitcall. When this
4260 * function is invoked, the refcount on the subsystem's module will be 0, so
4261 * the subsystem will not be attached to any hierarchy.
4263 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4265 struct cg_cgroup_link *link;
4266 struct hlist_head *hhead;
4268 BUG_ON(ss->module == NULL);
4271 * we shouldn't be called if the subsystem is in use, and the use of
4272 * try_module_get in parse_cgroupfs_options should ensure that it
4273 * doesn't start being used while we're killing it off.
4275 BUG_ON(ss->root != &rootnode);
4277 mutex_lock(&cgroup_mutex);
4278 /* deassign the subsys_id */
4279 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4280 subsys[ss->subsys_id] = NULL;
4282 /* remove subsystem from rootnode's list of subsystems */
4283 list_del_init(&ss->sibling);
4286 * disentangle the css from all css_sets attached to the dummytop. as
4287 * in loading, we need to pay our respects to the hashtable gods.
4289 write_lock(&css_set_lock);
4290 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4291 struct css_set *cg = link->cg;
4293 hlist_del(&cg->hlist);
4294 BUG_ON(!cg->subsys[ss->subsys_id]);
4295 cg->subsys[ss->subsys_id] = NULL;
4296 hhead = css_set_hash(cg->subsys);
4297 hlist_add_head(&cg->hlist, hhead);
4299 write_unlock(&css_set_lock);
4302 * remove subsystem's css from the dummytop and free it - need to free
4303 * before marking as null because ss->destroy needs the cgrp->subsys
4304 * pointer to find their state. note that this also takes care of
4305 * freeing the css_id.
4307 ss->destroy(ss, dummytop);
4308 dummytop->subsys[ss->subsys_id] = NULL;
4310 mutex_unlock(&cgroup_mutex);
4312 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4315 * cgroup_init_early - cgroup initialization at system boot
4317 * Initialize cgroups at system boot, and initialize any
4318 * subsystems that request early init.
4320 int __init cgroup_init_early(void)
4323 atomic_set(&init_css_set.refcount, 1);
4324 INIT_LIST_HEAD(&init_css_set.cg_links);
4325 INIT_LIST_HEAD(&init_css_set.tasks);
4326 INIT_HLIST_NODE(&init_css_set.hlist);
4328 init_cgroup_root(&rootnode);
4330 init_task.cgroups = &init_css_set;
4332 init_css_set_link.cg = &init_css_set;
4333 init_css_set_link.cgrp = dummytop;
4334 list_add(&init_css_set_link.cgrp_link_list,
4335 &rootnode.top_cgroup.css_sets);
4336 list_add(&init_css_set_link.cg_link_list,
4337 &init_css_set.cg_links);
4339 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4340 INIT_HLIST_HEAD(&css_set_table[i]);
4342 /* at bootup time, we don't worry about modular subsystems */
4343 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4344 struct cgroup_subsys *ss = subsys[i];
4347 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4348 BUG_ON(!ss->create);
4349 BUG_ON(!ss->destroy);
4350 if (ss->subsys_id != i) {
4351 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4352 ss->name, ss->subsys_id);
4357 cgroup_init_subsys(ss);
4363 * cgroup_init - cgroup initialization
4365 * Register cgroup filesystem and /proc file, and initialize
4366 * any subsystems that didn't request early init.
4368 int __init cgroup_init(void)
4372 struct hlist_head *hhead;
4374 err = bdi_init(&cgroup_backing_dev_info);
4378 /* at bootup time, we don't worry about modular subsystems */
4379 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4380 struct cgroup_subsys *ss = subsys[i];
4381 if (!ss->early_init)
4382 cgroup_init_subsys(ss);
4384 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4387 /* Add init_css_set to the hash table */
4388 hhead = css_set_hash(init_css_set.subsys);
4389 hlist_add_head(&init_css_set.hlist, hhead);
4390 BUG_ON(!init_root_id(&rootnode));
4392 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4398 err = register_filesystem(&cgroup_fs_type);
4400 kobject_put(cgroup_kobj);
4404 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4408 bdi_destroy(&cgroup_backing_dev_info);
4414 * proc_cgroup_show()
4415 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4416 * - Used for /proc/<pid>/cgroup.
4417 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4418 * doesn't really matter if tsk->cgroup changes after we read it,
4419 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4420 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4421 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4422 * cgroup to top_cgroup.
4425 /* TODO: Use a proper seq_file iterator */
4426 static int proc_cgroup_show(struct seq_file *m, void *v)
4429 struct task_struct *tsk;
4432 struct cgroupfs_root *root;
4435 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4441 tsk = get_pid_task(pid, PIDTYPE_PID);
4447 mutex_lock(&cgroup_mutex);
4449 for_each_active_root(root) {
4450 struct cgroup_subsys *ss;
4451 struct cgroup *cgrp;
4454 seq_printf(m, "%d:", root->hierarchy_id);
4455 for_each_subsys(root, ss)
4456 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4457 if (strlen(root->name))
4458 seq_printf(m, "%sname=%s", count ? "," : "",
4461 cgrp = task_cgroup_from_root(tsk, root);
4462 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4470 mutex_unlock(&cgroup_mutex);
4471 put_task_struct(tsk);
4478 static int cgroup_open(struct inode *inode, struct file *file)
4480 struct pid *pid = PROC_I(inode)->pid;
4481 return single_open(file, proc_cgroup_show, pid);
4484 const struct file_operations proc_cgroup_operations = {
4485 .open = cgroup_open,
4487 .llseek = seq_lseek,
4488 .release = single_release,
4491 /* Display information about each subsystem and each hierarchy */
4492 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4496 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4498 * ideally we don't want subsystems moving around while we do this.
4499 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4500 * subsys/hierarchy state.
4502 mutex_lock(&cgroup_mutex);
4503 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4504 struct cgroup_subsys *ss = subsys[i];
4507 seq_printf(m, "%s\t%d\t%d\t%d\n",
4508 ss->name, ss->root->hierarchy_id,
4509 ss->root->number_of_cgroups, !ss->disabled);
4511 mutex_unlock(&cgroup_mutex);
4515 static int cgroupstats_open(struct inode *inode, struct file *file)
4517 return single_open(file, proc_cgroupstats_show, NULL);
4520 static const struct file_operations proc_cgroupstats_operations = {
4521 .open = cgroupstats_open,
4523 .llseek = seq_lseek,
4524 .release = single_release,
4528 * cgroup_fork - attach newly forked task to its parents cgroup.
4529 * @child: pointer to task_struct of forking parent process.
4531 * Description: A task inherits its parent's cgroup at fork().
4533 * A pointer to the shared css_set was automatically copied in
4534 * fork.c by dup_task_struct(). However, we ignore that copy, since
4535 * it was not made under the protection of RCU or cgroup_mutex, so
4536 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4537 * have already changed current->cgroups, allowing the previously
4538 * referenced cgroup group to be removed and freed.
4540 * At the point that cgroup_fork() is called, 'current' is the parent
4541 * task, and the passed argument 'child' points to the child task.
4543 void cgroup_fork(struct task_struct *child)
4546 child->cgroups = current->cgroups;
4547 get_css_set(child->cgroups);
4548 task_unlock(current);
4549 INIT_LIST_HEAD(&child->cg_list);
4553 * cgroup_fork_callbacks - run fork callbacks
4554 * @child: the new task
4556 * Called on a new task very soon before adding it to the
4557 * tasklist. No need to take any locks since no-one can
4558 * be operating on this task.
4560 void cgroup_fork_callbacks(struct task_struct *child)
4562 if (need_forkexit_callback) {
4565 * forkexit callbacks are only supported for builtin
4566 * subsystems, and the builtin section of the subsys array is
4567 * immutable, so we don't need to lock the subsys array here.
4569 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4570 struct cgroup_subsys *ss = subsys[i];
4572 ss->fork(ss, child);
4578 * cgroup_post_fork - called on a new task after adding it to the task list
4579 * @child: the task in question
4581 * Adds the task to the list running through its css_set if necessary.
4582 * Has to be after the task is visible on the task list in case we race
4583 * with the first call to cgroup_iter_start() - to guarantee that the
4584 * new task ends up on its list.
4586 void cgroup_post_fork(struct task_struct *child)
4588 if (use_task_css_set_links) {
4589 write_lock(&css_set_lock);
4591 if (list_empty(&child->cg_list))
4592 list_add(&child->cg_list, &child->cgroups->tasks);
4594 write_unlock(&css_set_lock);
4598 * cgroup_exit - detach cgroup from exiting task
4599 * @tsk: pointer to task_struct of exiting process
4600 * @run_callback: run exit callbacks?
4602 * Description: Detach cgroup from @tsk and release it.
4604 * Note that cgroups marked notify_on_release force every task in
4605 * them to take the global cgroup_mutex mutex when exiting.
4606 * This could impact scaling on very large systems. Be reluctant to
4607 * use notify_on_release cgroups where very high task exit scaling
4608 * is required on large systems.
4610 * the_top_cgroup_hack:
4612 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4614 * We call cgroup_exit() while the task is still competent to
4615 * handle notify_on_release(), then leave the task attached to the
4616 * root cgroup in each hierarchy for the remainder of its exit.
4618 * To do this properly, we would increment the reference count on
4619 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4620 * code we would add a second cgroup function call, to drop that
4621 * reference. This would just create an unnecessary hot spot on
4622 * the top_cgroup reference count, to no avail.
4624 * Normally, holding a reference to a cgroup without bumping its
4625 * count is unsafe. The cgroup could go away, or someone could
4626 * attach us to a different cgroup, decrementing the count on
4627 * the first cgroup that we never incremented. But in this case,
4628 * top_cgroup isn't going away, and either task has PF_EXITING set,
4629 * which wards off any cgroup_attach_task() attempts, or task is a failed
4630 * fork, never visible to cgroup_attach_task.
4632 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4638 * Unlink from the css_set task list if necessary.
4639 * Optimistically check cg_list before taking
4642 if (!list_empty(&tsk->cg_list)) {
4643 write_lock(&css_set_lock);
4644 if (!list_empty(&tsk->cg_list))
4645 list_del_init(&tsk->cg_list);
4646 write_unlock(&css_set_lock);
4649 /* Reassign the task to the init_css_set. */
4652 tsk->cgroups = &init_css_set;
4654 if (run_callbacks && need_forkexit_callback) {
4656 * modular subsystems can't use callbacks, so no need to lock
4659 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4660 struct cgroup_subsys *ss = subsys[i];
4662 struct cgroup *old_cgrp =
4663 rcu_dereference_raw(cg->subsys[i])->cgroup;
4664 struct cgroup *cgrp = task_cgroup(tsk, i);
4665 ss->exit(ss, cgrp, old_cgrp, tsk);
4676 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4677 * @cgrp: the cgroup in question
4678 * @task: the task in question
4680 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4683 * If we are sending in dummytop, then presumably we are creating
4684 * the top cgroup in the subsystem.
4686 * Called only by the ns (nsproxy) cgroup.
4688 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4691 struct cgroup *target;
4693 if (cgrp == dummytop)
4696 target = task_cgroup_from_root(task, cgrp->root);
4697 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4698 cgrp = cgrp->parent;
4699 ret = (cgrp == target);
4703 static void check_for_release(struct cgroup *cgrp)
4705 /* All of these checks rely on RCU to keep the cgroup
4706 * structure alive */
4707 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4708 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4709 /* Control Group is currently removeable. If it's not
4710 * already queued for a userspace notification, queue
4712 int need_schedule_work = 0;
4713 spin_lock(&release_list_lock);
4714 if (!cgroup_is_removed(cgrp) &&
4715 list_empty(&cgrp->release_list)) {
4716 list_add(&cgrp->release_list, &release_list);
4717 need_schedule_work = 1;
4719 spin_unlock(&release_list_lock);
4720 if (need_schedule_work)
4721 schedule_work(&release_agent_work);
4725 /* Caller must verify that the css is not for root cgroup */
4726 void __css_get(struct cgroup_subsys_state *css, int count)
4728 atomic_add(count, &css->refcnt);
4729 set_bit(CGRP_RELEASABLE, &css->cgroup->flags);
4731 EXPORT_SYMBOL_GPL(__css_get);
4733 /* Caller must verify that the css is not for root cgroup */
4734 void __css_put(struct cgroup_subsys_state *css, int count)
4736 struct cgroup *cgrp = css->cgroup;
4739 val = atomic_sub_return(count, &css->refcnt);
4741 check_for_release(cgrp);
4742 cgroup_wakeup_rmdir_waiter(cgrp);
4745 WARN_ON_ONCE(val < 1);
4747 EXPORT_SYMBOL_GPL(__css_put);
4750 * Notify userspace when a cgroup is released, by running the
4751 * configured release agent with the name of the cgroup (path
4752 * relative to the root of cgroup file system) as the argument.
4754 * Most likely, this user command will try to rmdir this cgroup.
4756 * This races with the possibility that some other task will be
4757 * attached to this cgroup before it is removed, or that some other
4758 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4759 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4760 * unused, and this cgroup will be reprieved from its death sentence,
4761 * to continue to serve a useful existence. Next time it's released,
4762 * we will get notified again, if it still has 'notify_on_release' set.
4764 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4765 * means only wait until the task is successfully execve()'d. The
4766 * separate release agent task is forked by call_usermodehelper(),
4767 * then control in this thread returns here, without waiting for the
4768 * release agent task. We don't bother to wait because the caller of
4769 * this routine has no use for the exit status of the release agent
4770 * task, so no sense holding our caller up for that.
4772 static void cgroup_release_agent(struct work_struct *work)
4774 BUG_ON(work != &release_agent_work);
4775 mutex_lock(&cgroup_mutex);
4776 spin_lock(&release_list_lock);
4777 while (!list_empty(&release_list)) {
4778 char *argv[3], *envp[3];
4780 char *pathbuf = NULL, *agentbuf = NULL;
4781 struct cgroup *cgrp = list_entry(release_list.next,
4784 list_del_init(&cgrp->release_list);
4785 spin_unlock(&release_list_lock);
4786 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4789 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4791 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4796 argv[i++] = agentbuf;
4797 argv[i++] = pathbuf;
4801 /* minimal command environment */
4802 envp[i++] = "HOME=/";
4803 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4806 /* Drop the lock while we invoke the usermode helper,
4807 * since the exec could involve hitting disk and hence
4808 * be a slow process */
4809 mutex_unlock(&cgroup_mutex);
4810 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4811 mutex_lock(&cgroup_mutex);
4815 spin_lock(&release_list_lock);
4817 spin_unlock(&release_list_lock);
4818 mutex_unlock(&cgroup_mutex);
4821 static int __init cgroup_disable(char *str)
4826 while ((token = strsep(&str, ",")) != NULL) {
4830 * cgroup_disable, being at boot time, can't know about module
4831 * subsystems, so we don't worry about them.
4833 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4834 struct cgroup_subsys *ss = subsys[i];
4836 if (!strcmp(token, ss->name)) {
4838 printk(KERN_INFO "Disabling %s control group"
4839 " subsystem\n", ss->name);
4846 __setup("cgroup_disable=", cgroup_disable);
4849 * Functons for CSS ID.
4853 *To get ID other than 0, this should be called when !cgroup_is_removed().
4855 unsigned short css_id(struct cgroup_subsys_state *css)
4857 struct css_id *cssid;
4860 * This css_id() can return correct value when somone has refcnt
4861 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4862 * it's unchanged until freed.
4864 cssid = rcu_dereference_check(css->id,
4865 rcu_read_lock_held() || atomic_read(&css->refcnt));
4871 EXPORT_SYMBOL_GPL(css_id);
4873 unsigned short css_depth(struct cgroup_subsys_state *css)
4875 struct css_id *cssid;
4877 cssid = rcu_dereference_check(css->id,
4878 rcu_read_lock_held() || atomic_read(&css->refcnt));
4881 return cssid->depth;
4884 EXPORT_SYMBOL_GPL(css_depth);
4887 * css_is_ancestor - test "root" css is an ancestor of "child"
4888 * @child: the css to be tested.
4889 * @root: the css supporsed to be an ancestor of the child.
4891 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4892 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4893 * But, considering usual usage, the csses should be valid objects after test.
4894 * Assuming that the caller will do some action to the child if this returns
4895 * returns true, the caller must take "child";s reference count.
4896 * If "child" is valid object and this returns true, "root" is valid, too.
4899 bool css_is_ancestor(struct cgroup_subsys_state *child,
4900 const struct cgroup_subsys_state *root)
4902 struct css_id *child_id;
4903 struct css_id *root_id;
4907 child_id = rcu_dereference(child->id);
4908 root_id = rcu_dereference(root->id);
4911 || (child_id->depth < root_id->depth)
4912 || (child_id->stack[root_id->depth] != root_id->id))
4918 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
4920 struct css_id *id = css->id;
4921 /* When this is called before css_id initialization, id can be NULL */
4925 BUG_ON(!ss->use_id);
4927 rcu_assign_pointer(id->css, NULL);
4928 rcu_assign_pointer(css->id, NULL);
4929 spin_lock(&ss->id_lock);
4930 idr_remove(&ss->idr, id->id);
4931 spin_unlock(&ss->id_lock);
4932 kfree_rcu(id, rcu_head);
4934 EXPORT_SYMBOL_GPL(free_css_id);
4937 * This is called by init or create(). Then, calls to this function are
4938 * always serialized (By cgroup_mutex() at create()).
4941 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
4943 struct css_id *newid;
4944 int myid, error, size;
4946 BUG_ON(!ss->use_id);
4948 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
4949 newid = kzalloc(size, GFP_KERNEL);
4951 return ERR_PTR(-ENOMEM);
4953 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
4957 spin_lock(&ss->id_lock);
4958 /* Don't use 0. allocates an ID of 1-65535 */
4959 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
4960 spin_unlock(&ss->id_lock);
4962 /* Returns error when there are no free spaces for new ID.*/
4967 if (myid > CSS_ID_MAX)
4971 newid->depth = depth;
4975 spin_lock(&ss->id_lock);
4976 idr_remove(&ss->idr, myid);
4977 spin_unlock(&ss->id_lock);
4980 return ERR_PTR(error);
4984 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
4985 struct cgroup_subsys_state *rootcss)
4987 struct css_id *newid;
4989 spin_lock_init(&ss->id_lock);
4992 newid = get_new_cssid(ss, 0);
4994 return PTR_ERR(newid);
4996 newid->stack[0] = newid->id;
4997 newid->css = rootcss;
4998 rootcss->id = newid;
5002 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5003 struct cgroup *child)
5005 int subsys_id, i, depth = 0;
5006 struct cgroup_subsys_state *parent_css, *child_css;
5007 struct css_id *child_id, *parent_id;
5009 subsys_id = ss->subsys_id;
5010 parent_css = parent->subsys[subsys_id];
5011 child_css = child->subsys[subsys_id];
5012 parent_id = parent_css->id;
5013 depth = parent_id->depth + 1;
5015 child_id = get_new_cssid(ss, depth);
5016 if (IS_ERR(child_id))
5017 return PTR_ERR(child_id);
5019 for (i = 0; i < depth; i++)
5020 child_id->stack[i] = parent_id->stack[i];
5021 child_id->stack[depth] = child_id->id;
5023 * child_id->css pointer will be set after this cgroup is available
5024 * see cgroup_populate_dir()
5026 rcu_assign_pointer(child_css->id, child_id);
5032 * css_lookup - lookup css by id
5033 * @ss: cgroup subsys to be looked into.
5036 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5037 * NULL if not. Should be called under rcu_read_lock()
5039 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5041 struct css_id *cssid = NULL;
5043 BUG_ON(!ss->use_id);
5044 cssid = idr_find(&ss->idr, id);
5046 if (unlikely(!cssid))
5049 return rcu_dereference(cssid->css);
5051 EXPORT_SYMBOL_GPL(css_lookup);
5054 * css_get_next - lookup next cgroup under specified hierarchy.
5055 * @ss: pointer to subsystem
5056 * @id: current position of iteration.
5057 * @root: pointer to css. search tree under this.
5058 * @foundid: position of found object.
5060 * Search next css under the specified hierarchy of rootid. Calling under
5061 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5063 struct cgroup_subsys_state *
5064 css_get_next(struct cgroup_subsys *ss, int id,
5065 struct cgroup_subsys_state *root, int *foundid)
5067 struct cgroup_subsys_state *ret = NULL;
5070 int rootid = css_id(root);
5071 int depth = css_depth(root);
5076 BUG_ON(!ss->use_id);
5077 /* fill start point for scan */
5081 * scan next entry from bitmap(tree), tmpid is updated after
5084 spin_lock(&ss->id_lock);
5085 tmp = idr_get_next(&ss->idr, &tmpid);
5086 spin_unlock(&ss->id_lock);
5090 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5091 ret = rcu_dereference(tmp->css);
5097 /* continue to scan from next id */
5104 * get corresponding css from file open on cgroupfs directory
5106 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5108 struct cgroup *cgrp;
5109 struct inode *inode;
5110 struct cgroup_subsys_state *css;
5112 inode = f->f_dentry->d_inode;
5113 /* check in cgroup filesystem dir */
5114 if (inode->i_op != &cgroup_dir_inode_operations)
5115 return ERR_PTR(-EBADF);
5117 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5118 return ERR_PTR(-EINVAL);
5121 cgrp = __d_cgrp(f->f_dentry);
5122 css = cgrp->subsys[id];
5123 return css ? css : ERR_PTR(-ENOENT);
5126 #ifdef CONFIG_CGROUP_DEBUG
5127 static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
5128 struct cgroup *cont)
5130 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5133 return ERR_PTR(-ENOMEM);
5138 static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
5140 kfree(cont->subsys[debug_subsys_id]);
5143 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5145 return atomic_read(&cont->count);
5148 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5150 return cgroup_task_count(cont);
5153 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5155 return (u64)(unsigned long)current->cgroups;
5158 static u64 current_css_set_refcount_read(struct cgroup *cont,
5164 count = atomic_read(¤t->cgroups->refcount);
5169 static int current_css_set_cg_links_read(struct cgroup *cont,
5171 struct seq_file *seq)
5173 struct cg_cgroup_link *link;
5176 read_lock(&css_set_lock);
5178 cg = rcu_dereference(current->cgroups);
5179 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5180 struct cgroup *c = link->cgrp;
5184 name = c->dentry->d_name.name;
5187 seq_printf(seq, "Root %d group %s\n",
5188 c->root->hierarchy_id, name);
5191 read_unlock(&css_set_lock);
5195 #define MAX_TASKS_SHOWN_PER_CSS 25
5196 static int cgroup_css_links_read(struct cgroup *cont,
5198 struct seq_file *seq)
5200 struct cg_cgroup_link *link;
5202 read_lock(&css_set_lock);
5203 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5204 struct css_set *cg = link->cg;
5205 struct task_struct *task;
5207 seq_printf(seq, "css_set %p\n", cg);
5208 list_for_each_entry(task, &cg->tasks, cg_list) {
5209 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5210 seq_puts(seq, " ...\n");
5213 seq_printf(seq, " task %d\n",
5214 task_pid_vnr(task));
5218 read_unlock(&css_set_lock);
5222 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5224 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5227 static struct cftype debug_files[] = {
5229 .name = "cgroup_refcount",
5230 .read_u64 = cgroup_refcount_read,
5233 .name = "taskcount",
5234 .read_u64 = debug_taskcount_read,
5238 .name = "current_css_set",
5239 .read_u64 = current_css_set_read,
5243 .name = "current_css_set_refcount",
5244 .read_u64 = current_css_set_refcount_read,
5248 .name = "current_css_set_cg_links",
5249 .read_seq_string = current_css_set_cg_links_read,
5253 .name = "cgroup_css_links",
5254 .read_seq_string = cgroup_css_links_read,
5258 .name = "releasable",
5259 .read_u64 = releasable_read,
5263 static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
5265 return cgroup_add_files(cont, ss, debug_files,
5266 ARRAY_SIZE(debug_files));
5269 struct cgroup_subsys debug_subsys = {
5271 .create = debug_create,
5272 .destroy = debug_destroy,
5273 .populate = debug_populate,
5274 .subsys_id = debug_subsys_id,
5276 #endif /* CONFIG_CGROUP_DEBUG */