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/smp_lock.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/idr.h>
58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
59 #include <linux/eventfd.h>
60 #include <linux/poll.h>
61 #include <linux/capability.h>
63 #include <asm/atomic.h>
65 static DEFINE_MUTEX(cgroup_mutex);
68 * Generate an array of cgroup subsystem pointers. At boot time, this is
69 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
70 * registered after that. The mutable section of this array is protected by
73 #define SUBSYS(_x) &_x ## _subsys,
74 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
75 #include <linux/cgroup_subsys.h>
78 #define MAX_CGROUP_ROOT_NAMELEN 64
81 * A cgroupfs_root represents the root of a cgroup hierarchy,
82 * and may be associated with a superblock to form an active
85 struct cgroupfs_root {
86 struct super_block *sb;
89 * The bitmask of subsystems intended to be attached to this
92 unsigned long subsys_bits;
94 /* Unique id for this hierarchy. */
97 /* The bitmask of subsystems currently attached to this hierarchy */
98 unsigned long actual_subsys_bits;
100 /* A list running through the attached subsystems */
101 struct list_head subsys_list;
103 /* The root cgroup for this hierarchy */
104 struct cgroup top_cgroup;
106 /* Tracks how many cgroups are currently defined in hierarchy.*/
107 int number_of_cgroups;
109 /* A list running through the active hierarchies */
110 struct list_head root_list;
112 /* Hierarchy-specific flags */
115 /* The path to use for release notifications. */
116 char release_agent_path[PATH_MAX];
118 /* The name for this hierarchy - may be empty */
119 char name[MAX_CGROUP_ROOT_NAMELEN];
123 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
124 * subsystems that are otherwise unattached - it never has more than a
125 * single cgroup, and all tasks are part of that cgroup.
127 static struct cgroupfs_root rootnode;
130 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
131 * cgroup_subsys->use_id != 0.
133 #define CSS_ID_MAX (65535)
136 * The css to which this ID points. This pointer is set to valid value
137 * after cgroup is populated. If cgroup is removed, this will be NULL.
138 * This pointer is expected to be RCU-safe because destroy()
139 * is called after synchronize_rcu(). But for safe use, css_is_removed()
140 * css_tryget() should be used for avoiding race.
142 struct cgroup_subsys_state *css;
148 * Depth in hierarchy which this ID belongs to.
150 unsigned short depth;
152 * ID is freed by RCU. (and lookup routine is RCU safe.)
154 struct rcu_head rcu_head;
156 * Hierarchy of CSS ID belongs to.
158 unsigned short stack[0]; /* Array of Length (depth+1) */
162 * cgroup_event represents events which userspace want to recieve.
164 struct cgroup_event {
166 * Cgroup which the event belongs to.
170 * Control file which the event associated.
174 * eventfd to signal userspace about the event.
176 struct eventfd_ctx *eventfd;
178 * Each of these stored in a list by the cgroup.
180 struct list_head list;
182 * All fields below needed to unregister event when
183 * userspace closes eventfd.
186 wait_queue_head_t *wqh;
188 struct work_struct remove;
191 /* The list of hierarchy roots */
193 static LIST_HEAD(roots);
194 static int root_count;
196 static DEFINE_IDA(hierarchy_ida);
197 static int next_hierarchy_id;
198 static DEFINE_SPINLOCK(hierarchy_id_lock);
200 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
201 #define dummytop (&rootnode.top_cgroup)
203 /* This flag indicates whether tasks in the fork and exit paths should
204 * check for fork/exit handlers to call. This avoids us having to do
205 * extra work in the fork/exit path if none of the subsystems need to
208 static int need_forkexit_callback __read_mostly;
210 #ifdef CONFIG_PROVE_LOCKING
211 int cgroup_lock_is_held(void)
213 return lockdep_is_held(&cgroup_mutex);
215 #else /* #ifdef CONFIG_PROVE_LOCKING */
216 int cgroup_lock_is_held(void)
218 return mutex_is_locked(&cgroup_mutex);
220 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
222 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
224 /* convenient tests for these bits */
225 inline int cgroup_is_removed(const struct cgroup *cgrp)
227 return test_bit(CGRP_REMOVED, &cgrp->flags);
230 /* bits in struct cgroupfs_root flags field */
232 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
235 static int cgroup_is_releasable(const struct cgroup *cgrp)
238 (1 << CGRP_RELEASABLE) |
239 (1 << CGRP_NOTIFY_ON_RELEASE);
240 return (cgrp->flags & bits) == bits;
243 static int notify_on_release(const struct cgroup *cgrp)
245 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
249 * for_each_subsys() allows you to iterate on each subsystem attached to
250 * an active hierarchy
252 #define for_each_subsys(_root, _ss) \
253 list_for_each_entry(_ss, &_root->subsys_list, sibling)
255 /* for_each_active_root() allows you to iterate across the active hierarchies */
256 #define for_each_active_root(_root) \
257 list_for_each_entry(_root, &roots, root_list)
259 /* the list of cgroups eligible for automatic release. Protected by
260 * release_list_lock */
261 static LIST_HEAD(release_list);
262 static DEFINE_SPINLOCK(release_list_lock);
263 static void cgroup_release_agent(struct work_struct *work);
264 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
265 static void check_for_release(struct cgroup *cgrp);
268 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
269 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
270 * reference to css->refcnt. In general, this refcnt is expected to goes down
273 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
275 DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
277 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
279 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
280 wake_up_all(&cgroup_rmdir_waitq);
283 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
288 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
290 cgroup_wakeup_rmdir_waiter(css->cgroup);
294 /* Link structure for associating css_set objects with cgroups */
295 struct cg_cgroup_link {
297 * List running through cg_cgroup_links associated with a
298 * cgroup, anchored on cgroup->css_sets
300 struct list_head cgrp_link_list;
303 * List running through cg_cgroup_links pointing at a
304 * single css_set object, anchored on css_set->cg_links
306 struct list_head cg_link_list;
310 /* The default css_set - used by init and its children prior to any
311 * hierarchies being mounted. It contains a pointer to the root state
312 * for each subsystem. Also used to anchor the list of css_sets. Not
313 * reference-counted, to improve performance when child cgroups
314 * haven't been created.
317 static struct css_set init_css_set;
318 static struct cg_cgroup_link init_css_set_link;
320 static int cgroup_init_idr(struct cgroup_subsys *ss,
321 struct cgroup_subsys_state *css);
323 /* css_set_lock protects the list of css_set objects, and the
324 * chain of tasks off each css_set. Nests outside task->alloc_lock
325 * due to cgroup_iter_start() */
326 static DEFINE_RWLOCK(css_set_lock);
327 static int css_set_count;
330 * hash table for cgroup groups. This improves the performance to find
331 * an existing css_set. This hash doesn't (currently) take into
332 * account cgroups in empty hierarchies.
334 #define CSS_SET_HASH_BITS 7
335 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
336 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
338 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
342 unsigned long tmp = 0UL;
344 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
345 tmp += (unsigned long)css[i];
346 tmp = (tmp >> 16) ^ tmp;
348 index = hash_long(tmp, CSS_SET_HASH_BITS);
350 return &css_set_table[index];
353 static void free_css_set_work(struct work_struct *work)
355 struct css_set *cg = container_of(work, struct css_set, work);
356 struct cg_cgroup_link *link;
357 struct cg_cgroup_link *saved_link;
359 write_lock(&css_set_lock);
360 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
362 struct cgroup *cgrp = link->cgrp;
363 list_del(&link->cg_link_list);
364 list_del(&link->cgrp_link_list);
365 if (atomic_dec_and_test(&cgrp->count)) {
366 check_for_release(cgrp);
367 cgroup_wakeup_rmdir_waiter(cgrp);
371 write_unlock(&css_set_lock);
376 static void free_css_set_rcu(struct rcu_head *obj)
378 struct css_set *cg = container_of(obj, struct css_set, rcu_head);
380 INIT_WORK(&cg->work, free_css_set_work);
381 schedule_work(&cg->work);
384 /* We don't maintain the lists running through each css_set to its
385 * task until after the first call to cgroup_iter_start(). This
386 * reduces the fork()/exit() overhead for people who have cgroups
387 * compiled into their kernel but not actually in use */
388 static int use_task_css_set_links __read_mostly;
391 * refcounted get/put for css_set objects
393 static inline void get_css_set(struct css_set *cg)
395 atomic_inc(&cg->refcount);
398 static void put_css_set(struct css_set *cg)
401 * Ensure that the refcount doesn't hit zero while any readers
402 * can see it. Similar to atomic_dec_and_lock(), but for an
405 if (atomic_add_unless(&cg->refcount, -1, 1))
407 write_lock(&css_set_lock);
408 if (!atomic_dec_and_test(&cg->refcount)) {
409 write_unlock(&css_set_lock);
413 hlist_del(&cg->hlist);
416 write_unlock(&css_set_lock);
417 call_rcu(&cg->rcu_head, free_css_set_rcu);
421 * compare_css_sets - helper function for find_existing_css_set().
422 * @cg: candidate css_set being tested
423 * @old_cg: existing css_set for a task
424 * @new_cgrp: cgroup that's being entered by the task
425 * @template: desired set of css pointers in css_set (pre-calculated)
427 * Returns true if "cg" matches "old_cg" except for the hierarchy
428 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
430 static bool compare_css_sets(struct css_set *cg,
431 struct css_set *old_cg,
432 struct cgroup *new_cgrp,
433 struct cgroup_subsys_state *template[])
435 struct list_head *l1, *l2;
437 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
438 /* Not all subsystems matched */
443 * Compare cgroup pointers in order to distinguish between
444 * different cgroups in heirarchies with no subsystems. We
445 * could get by with just this check alone (and skip the
446 * memcmp above) but on most setups the memcmp check will
447 * avoid the need for this more expensive check on almost all
452 l2 = &old_cg->cg_links;
454 struct cg_cgroup_link *cgl1, *cgl2;
455 struct cgroup *cg1, *cg2;
459 /* See if we reached the end - both lists are equal length. */
460 if (l1 == &cg->cg_links) {
461 BUG_ON(l2 != &old_cg->cg_links);
464 BUG_ON(l2 == &old_cg->cg_links);
466 /* Locate the cgroups associated with these links. */
467 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
468 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
471 /* Hierarchies should be linked in the same order. */
472 BUG_ON(cg1->root != cg2->root);
475 * If this hierarchy is the hierarchy of the cgroup
476 * that's changing, then we need to check that this
477 * css_set points to the new cgroup; if it's any other
478 * hierarchy, then this css_set should point to the
479 * same cgroup as the old css_set.
481 if (cg1->root == new_cgrp->root) {
493 * find_existing_css_set() is a helper for
494 * find_css_set(), and checks to see whether an existing
495 * css_set is suitable.
497 * oldcg: the cgroup group that we're using before the cgroup
500 * cgrp: the cgroup that we're moving into
502 * template: location in which to build the desired set of subsystem
503 * state objects for the new cgroup group
505 static struct css_set *find_existing_css_set(
506 struct css_set *oldcg,
508 struct cgroup_subsys_state *template[])
511 struct cgroupfs_root *root = cgrp->root;
512 struct hlist_head *hhead;
513 struct hlist_node *node;
517 * Build the set of subsystem state objects that we want to see in the
518 * new css_set. while subsystems can change globally, the entries here
519 * won't change, so no need for locking.
521 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
522 if (root->subsys_bits & (1UL << i)) {
523 /* Subsystem is in this hierarchy. So we want
524 * the subsystem state from the new
526 template[i] = cgrp->subsys[i];
528 /* Subsystem is not in this hierarchy, so we
529 * don't want to change the subsystem state */
530 template[i] = oldcg->subsys[i];
534 hhead = css_set_hash(template);
535 hlist_for_each_entry(cg, node, hhead, hlist) {
536 if (!compare_css_sets(cg, oldcg, cgrp, template))
539 /* This css_set matches what we need */
543 /* No existing cgroup group matched */
547 static void free_cg_links(struct list_head *tmp)
549 struct cg_cgroup_link *link;
550 struct cg_cgroup_link *saved_link;
552 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
553 list_del(&link->cgrp_link_list);
559 * allocate_cg_links() allocates "count" cg_cgroup_link structures
560 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
561 * success or a negative error
563 static int allocate_cg_links(int count, struct list_head *tmp)
565 struct cg_cgroup_link *link;
568 for (i = 0; i < count; i++) {
569 link = kmalloc(sizeof(*link), GFP_KERNEL);
574 list_add(&link->cgrp_link_list, tmp);
580 * link_css_set - a helper function to link a css_set to a cgroup
581 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
582 * @cg: the css_set to be linked
583 * @cgrp: the destination cgroup
585 static void link_css_set(struct list_head *tmp_cg_links,
586 struct css_set *cg, struct cgroup *cgrp)
588 struct cg_cgroup_link *link;
590 BUG_ON(list_empty(tmp_cg_links));
591 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
595 atomic_inc(&cgrp->count);
596 list_move(&link->cgrp_link_list, &cgrp->css_sets);
598 * Always add links to the tail of the list so that the list
599 * is sorted by order of hierarchy creation
601 list_add_tail(&link->cg_link_list, &cg->cg_links);
605 * find_css_set() takes an existing cgroup group and a
606 * cgroup object, and returns a css_set object that's
607 * equivalent to the old group, but with the given cgroup
608 * substituted into the appropriate hierarchy. Must be called with
611 static struct css_set *find_css_set(
612 struct css_set *oldcg, struct cgroup *cgrp)
615 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
617 struct list_head tmp_cg_links;
619 struct hlist_head *hhead;
620 struct cg_cgroup_link *link;
622 /* First see if we already have a cgroup group that matches
624 read_lock(&css_set_lock);
625 res = find_existing_css_set(oldcg, cgrp, template);
628 read_unlock(&css_set_lock);
633 res = kmalloc(sizeof(*res), GFP_KERNEL);
637 /* Allocate all the cg_cgroup_link objects that we'll need */
638 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
643 atomic_set(&res->refcount, 1);
644 INIT_LIST_HEAD(&res->cg_links);
645 INIT_LIST_HEAD(&res->tasks);
646 INIT_HLIST_NODE(&res->hlist);
648 /* Copy the set of subsystem state objects generated in
649 * find_existing_css_set() */
650 memcpy(res->subsys, template, sizeof(res->subsys));
652 write_lock(&css_set_lock);
653 /* Add reference counts and links from the new css_set. */
654 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
655 struct cgroup *c = link->cgrp;
656 if (c->root == cgrp->root)
658 link_css_set(&tmp_cg_links, res, c);
661 BUG_ON(!list_empty(&tmp_cg_links));
665 /* Add this cgroup group to the hash table */
666 hhead = css_set_hash(res->subsys);
667 hlist_add_head(&res->hlist, hhead);
669 write_unlock(&css_set_lock);
675 * Return the cgroup for "task" from the given hierarchy. Must be
676 * called with cgroup_mutex held.
678 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
679 struct cgroupfs_root *root)
682 struct cgroup *res = NULL;
684 BUG_ON(!mutex_is_locked(&cgroup_mutex));
685 read_lock(&css_set_lock);
687 * No need to lock the task - since we hold cgroup_mutex the
688 * task can't change groups, so the only thing that can happen
689 * is that it exits and its css is set back to init_css_set.
692 if (css == &init_css_set) {
693 res = &root->top_cgroup;
695 struct cg_cgroup_link *link;
696 list_for_each_entry(link, &css->cg_links, cg_link_list) {
697 struct cgroup *c = link->cgrp;
698 if (c->root == root) {
704 read_unlock(&css_set_lock);
710 * There is one global cgroup mutex. We also require taking
711 * task_lock() when dereferencing a task's cgroup subsys pointers.
712 * See "The task_lock() exception", at the end of this comment.
714 * A task must hold cgroup_mutex to modify cgroups.
716 * Any task can increment and decrement the count field without lock.
717 * So in general, code holding cgroup_mutex can't rely on the count
718 * field not changing. However, if the count goes to zero, then only
719 * cgroup_attach_task() can increment it again. Because a count of zero
720 * means that no tasks are currently attached, therefore there is no
721 * way a task attached to that cgroup can fork (the other way to
722 * increment the count). So code holding cgroup_mutex can safely
723 * assume that if the count is zero, it will stay zero. Similarly, if
724 * a task holds cgroup_mutex on a cgroup with zero count, it
725 * knows that the cgroup won't be removed, as cgroup_rmdir()
728 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
729 * (usually) take cgroup_mutex. These are the two most performance
730 * critical pieces of code here. The exception occurs on cgroup_exit(),
731 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
732 * is taken, and if the cgroup count is zero, a usermode call made
733 * to the release agent with the name of the cgroup (path relative to
734 * the root of cgroup file system) as the argument.
736 * A cgroup can only be deleted if both its 'count' of using tasks
737 * is zero, and its list of 'children' cgroups is empty. Since all
738 * tasks in the system use _some_ cgroup, and since there is always at
739 * least one task in the system (init, pid == 1), therefore, top_cgroup
740 * always has either children cgroups and/or using tasks. So we don't
741 * need a special hack to ensure that top_cgroup cannot be deleted.
743 * The task_lock() exception
745 * The need for this exception arises from the action of
746 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
747 * another. It does so using cgroup_mutex, however there are
748 * several performance critical places that need to reference
749 * task->cgroups without the expense of grabbing a system global
750 * mutex. Therefore except as noted below, when dereferencing or, as
751 * in cgroup_attach_task(), modifying a task's cgroups pointer we use
752 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
753 * the task_struct routinely used for such matters.
755 * P.S. One more locking exception. RCU is used to guard the
756 * update of a tasks cgroup pointer by cgroup_attach_task()
760 * cgroup_lock - lock out any changes to cgroup structures
763 void cgroup_lock(void)
765 mutex_lock(&cgroup_mutex);
767 EXPORT_SYMBOL_GPL(cgroup_lock);
770 * cgroup_unlock - release lock on cgroup changes
772 * Undo the lock taken in a previous cgroup_lock() call.
774 void cgroup_unlock(void)
776 mutex_unlock(&cgroup_mutex);
778 EXPORT_SYMBOL_GPL(cgroup_unlock);
781 * A couple of forward declarations required, due to cyclic reference loop:
782 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
783 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
787 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
788 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
789 static int cgroup_populate_dir(struct cgroup *cgrp);
790 static const struct inode_operations cgroup_dir_inode_operations;
791 static const struct file_operations proc_cgroupstats_operations;
793 static struct backing_dev_info cgroup_backing_dev_info = {
795 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
798 static int alloc_css_id(struct cgroup_subsys *ss,
799 struct cgroup *parent, struct cgroup *child);
801 static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
803 struct inode *inode = new_inode(sb);
806 inode->i_mode = mode;
807 inode->i_uid = current_fsuid();
808 inode->i_gid = current_fsgid();
809 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
810 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
816 * Call subsys's pre_destroy handler.
817 * This is called before css refcnt check.
819 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
821 struct cgroup_subsys *ss;
824 for_each_subsys(cgrp->root, ss)
825 if (ss->pre_destroy) {
826 ret = ss->pre_destroy(ss, cgrp);
834 static void free_cgroup_rcu(struct rcu_head *obj)
836 struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head);
841 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
843 /* is dentry a directory ? if so, kfree() associated cgroup */
844 if (S_ISDIR(inode->i_mode)) {
845 struct cgroup *cgrp = dentry->d_fsdata;
846 struct cgroup_subsys *ss;
847 BUG_ON(!(cgroup_is_removed(cgrp)));
848 /* It's possible for external users to be holding css
849 * reference counts on a cgroup; css_put() needs to
850 * be able to access the cgroup after decrementing
851 * the reference count in order to know if it needs to
852 * queue the cgroup to be handled by the release
856 mutex_lock(&cgroup_mutex);
858 * Release the subsystem state objects.
860 for_each_subsys(cgrp->root, ss)
861 ss->destroy(ss, cgrp);
863 cgrp->root->number_of_cgroups--;
864 mutex_unlock(&cgroup_mutex);
867 * Drop the active superblock reference that we took when we
870 deactivate_super(cgrp->root->sb);
873 * if we're getting rid of the cgroup, refcount should ensure
874 * that there are no pidlists left.
876 BUG_ON(!list_empty(&cgrp->pidlists));
878 call_rcu(&cgrp->rcu_head, free_cgroup_rcu);
883 static void remove_dir(struct dentry *d)
885 struct dentry *parent = dget(d->d_parent);
888 simple_rmdir(parent->d_inode, d);
892 static void cgroup_clear_directory(struct dentry *dentry)
894 struct list_head *node;
896 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
897 spin_lock(&dcache_lock);
898 node = dentry->d_subdirs.next;
899 while (node != &dentry->d_subdirs) {
900 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
903 /* This should never be called on a cgroup
904 * directory with child cgroups */
905 BUG_ON(d->d_inode->i_mode & S_IFDIR);
907 spin_unlock(&dcache_lock);
909 simple_unlink(dentry->d_inode, d);
911 spin_lock(&dcache_lock);
913 node = dentry->d_subdirs.next;
915 spin_unlock(&dcache_lock);
919 * NOTE : the dentry must have been dget()'ed
921 static void cgroup_d_remove_dir(struct dentry *dentry)
923 cgroup_clear_directory(dentry);
925 spin_lock(&dcache_lock);
926 list_del_init(&dentry->d_u.d_child);
927 spin_unlock(&dcache_lock);
932 * Call with cgroup_mutex held. Drops reference counts on modules, including
933 * any duplicate ones that parse_cgroupfs_options took. If this function
934 * returns an error, no reference counts are touched.
936 static int rebind_subsystems(struct cgroupfs_root *root,
937 unsigned long final_bits)
939 unsigned long added_bits, removed_bits;
940 struct cgroup *cgrp = &root->top_cgroup;
943 BUG_ON(!mutex_is_locked(&cgroup_mutex));
945 removed_bits = root->actual_subsys_bits & ~final_bits;
946 added_bits = final_bits & ~root->actual_subsys_bits;
947 /* Check that any added subsystems are currently free */
948 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
949 unsigned long bit = 1UL << i;
950 struct cgroup_subsys *ss = subsys[i];
951 if (!(bit & added_bits))
954 * Nobody should tell us to do a subsys that doesn't exist:
955 * parse_cgroupfs_options should catch that case and refcounts
956 * ensure that subsystems won't disappear once selected.
959 if (ss->root != &rootnode) {
960 /* Subsystem isn't free */
965 /* Currently we don't handle adding/removing subsystems when
966 * any child cgroups exist. This is theoretically supportable
967 * but involves complex error handling, so it's being left until
969 if (root->number_of_cgroups > 1)
972 /* Process each subsystem */
973 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
974 struct cgroup_subsys *ss = subsys[i];
975 unsigned long bit = 1UL << i;
976 if (bit & added_bits) {
977 /* We're binding this subsystem to this hierarchy */
979 BUG_ON(cgrp->subsys[i]);
980 BUG_ON(!dummytop->subsys[i]);
981 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
982 mutex_lock(&ss->hierarchy_mutex);
983 cgrp->subsys[i] = dummytop->subsys[i];
984 cgrp->subsys[i]->cgroup = cgrp;
985 list_move(&ss->sibling, &root->subsys_list);
989 mutex_unlock(&ss->hierarchy_mutex);
990 /* refcount was already taken, and we're keeping it */
991 } else if (bit & removed_bits) {
992 /* We're removing this subsystem */
994 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
995 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
996 mutex_lock(&ss->hierarchy_mutex);
998 ss->bind(ss, dummytop);
999 dummytop->subsys[i]->cgroup = dummytop;
1000 cgrp->subsys[i] = NULL;
1001 subsys[i]->root = &rootnode;
1002 list_move(&ss->sibling, &rootnode.subsys_list);
1003 mutex_unlock(&ss->hierarchy_mutex);
1004 /* subsystem is now free - drop reference on module */
1005 module_put(ss->module);
1006 } else if (bit & final_bits) {
1007 /* Subsystem state should already exist */
1009 BUG_ON(!cgrp->subsys[i]);
1011 * a refcount was taken, but we already had one, so
1012 * drop the extra reference.
1014 module_put(ss->module);
1015 #ifdef CONFIG_MODULE_UNLOAD
1016 BUG_ON(ss->module && !module_refcount(ss->module));
1019 /* Subsystem state shouldn't exist */
1020 BUG_ON(cgrp->subsys[i]);
1023 root->subsys_bits = root->actual_subsys_bits = final_bits;
1029 static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
1031 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
1032 struct cgroup_subsys *ss;
1034 mutex_lock(&cgroup_mutex);
1035 for_each_subsys(root, ss)
1036 seq_printf(seq, ",%s", ss->name);
1037 if (test_bit(ROOT_NOPREFIX, &root->flags))
1038 seq_puts(seq, ",noprefix");
1039 if (strlen(root->release_agent_path))
1040 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1041 if (strlen(root->name))
1042 seq_printf(seq, ",name=%s", root->name);
1043 mutex_unlock(&cgroup_mutex);
1047 struct cgroup_sb_opts {
1048 unsigned long subsys_bits;
1049 unsigned long flags;
1050 char *release_agent;
1052 /* User explicitly requested empty subsystem */
1055 struct cgroupfs_root *new_root;
1060 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1061 * with cgroup_mutex held to protect the subsys[] array. This function takes
1062 * refcounts on subsystems to be used, unless it returns error, in which case
1063 * no refcounts are taken.
1065 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1067 char *token, *o = data ?: "all";
1068 unsigned long mask = (unsigned long)-1;
1070 bool module_pin_failed = false;
1072 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1074 #ifdef CONFIG_CPUSETS
1075 mask = ~(1UL << cpuset_subsys_id);
1078 memset(opts, 0, sizeof(*opts));
1080 while ((token = strsep(&o, ",")) != NULL) {
1083 if (!strcmp(token, "all")) {
1084 /* Add all non-disabled subsystems */
1085 opts->subsys_bits = 0;
1086 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1087 struct cgroup_subsys *ss = subsys[i];
1091 opts->subsys_bits |= 1ul << i;
1093 } else if (!strcmp(token, "none")) {
1094 /* Explicitly have no subsystems */
1096 } else if (!strcmp(token, "noprefix")) {
1097 set_bit(ROOT_NOPREFIX, &opts->flags);
1098 } else if (!strncmp(token, "release_agent=", 14)) {
1099 /* Specifying two release agents is forbidden */
1100 if (opts->release_agent)
1102 opts->release_agent =
1103 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1104 if (!opts->release_agent)
1106 } else if (!strncmp(token, "name=", 5)) {
1107 const char *name = token + 5;
1108 /* Can't specify an empty name */
1111 /* Must match [\w.-]+ */
1112 for (i = 0; i < strlen(name); i++) {
1116 if ((c == '.') || (c == '-') || (c == '_'))
1120 /* Specifying two names is forbidden */
1123 opts->name = kstrndup(name,
1124 MAX_CGROUP_ROOT_NAMELEN - 1,
1129 struct cgroup_subsys *ss;
1130 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1134 if (!strcmp(token, ss->name)) {
1136 set_bit(i, &opts->subsys_bits);
1140 if (i == CGROUP_SUBSYS_COUNT)
1145 /* Consistency checks */
1148 * Option noprefix was introduced just for backward compatibility
1149 * with the old cpuset, so we allow noprefix only if mounting just
1150 * the cpuset subsystem.
1152 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1153 (opts->subsys_bits & mask))
1157 /* Can't specify "none" and some subsystems */
1158 if (opts->subsys_bits && opts->none)
1162 * We either have to specify by name or by subsystems. (So all
1163 * empty hierarchies must have a name).
1165 if (!opts->subsys_bits && !opts->name)
1169 * Grab references on all the modules we'll need, so the subsystems
1170 * don't dance around before rebind_subsystems attaches them. This may
1171 * take duplicate reference counts on a subsystem that's already used,
1172 * but rebind_subsystems handles this case.
1174 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1175 unsigned long bit = 1UL << i;
1177 if (!(bit & opts->subsys_bits))
1179 if (!try_module_get(subsys[i]->module)) {
1180 module_pin_failed = true;
1184 if (module_pin_failed) {
1186 * oops, one of the modules was going away. this means that we
1187 * raced with a module_delete call, and to the user this is
1188 * essentially a "subsystem doesn't exist" case.
1190 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1191 /* drop refcounts only on the ones we took */
1192 unsigned long bit = 1UL << i;
1194 if (!(bit & opts->subsys_bits))
1196 module_put(subsys[i]->module);
1204 static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1207 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1208 unsigned long bit = 1UL << i;
1210 if (!(bit & subsys_bits))
1212 module_put(subsys[i]->module);
1216 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1219 struct cgroupfs_root *root = sb->s_fs_info;
1220 struct cgroup *cgrp = &root->top_cgroup;
1221 struct cgroup_sb_opts opts;
1224 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1225 mutex_lock(&cgroup_mutex);
1227 /* See what subsystems are wanted */
1228 ret = parse_cgroupfs_options(data, &opts);
1232 /* Don't allow flags or name to change at remount */
1233 if (opts.flags != root->flags ||
1234 (opts.name && strcmp(opts.name, root->name))) {
1236 drop_parsed_module_refcounts(opts.subsys_bits);
1240 ret = rebind_subsystems(root, opts.subsys_bits);
1242 drop_parsed_module_refcounts(opts.subsys_bits);
1246 /* (re)populate subsystem files */
1247 cgroup_populate_dir(cgrp);
1249 if (opts.release_agent)
1250 strcpy(root->release_agent_path, opts.release_agent);
1252 kfree(opts.release_agent);
1254 mutex_unlock(&cgroup_mutex);
1255 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1260 static const struct super_operations cgroup_ops = {
1261 .statfs = simple_statfs,
1262 .drop_inode = generic_delete_inode,
1263 .show_options = cgroup_show_options,
1264 .remount_fs = cgroup_remount,
1267 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1269 INIT_LIST_HEAD(&cgrp->sibling);
1270 INIT_LIST_HEAD(&cgrp->children);
1271 INIT_LIST_HEAD(&cgrp->css_sets);
1272 INIT_LIST_HEAD(&cgrp->release_list);
1273 INIT_LIST_HEAD(&cgrp->pidlists);
1274 mutex_init(&cgrp->pidlist_mutex);
1275 INIT_LIST_HEAD(&cgrp->event_list);
1276 spin_lock_init(&cgrp->event_list_lock);
1279 static void init_cgroup_root(struct cgroupfs_root *root)
1281 struct cgroup *cgrp = &root->top_cgroup;
1282 INIT_LIST_HEAD(&root->subsys_list);
1283 INIT_LIST_HEAD(&root->root_list);
1284 root->number_of_cgroups = 1;
1286 cgrp->top_cgroup = cgrp;
1287 init_cgroup_housekeeping(cgrp);
1290 static bool init_root_id(struct cgroupfs_root *root)
1295 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1297 spin_lock(&hierarchy_id_lock);
1298 /* Try to allocate the next unused ID */
1299 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1300 &root->hierarchy_id);
1302 /* Try again starting from 0 */
1303 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1305 next_hierarchy_id = root->hierarchy_id + 1;
1306 } else if (ret != -EAGAIN) {
1307 /* Can only get here if the 31-bit IDR is full ... */
1310 spin_unlock(&hierarchy_id_lock);
1315 static int cgroup_test_super(struct super_block *sb, void *data)
1317 struct cgroup_sb_opts *opts = data;
1318 struct cgroupfs_root *root = sb->s_fs_info;
1320 /* If we asked for a name then it must match */
1321 if (opts->name && strcmp(opts->name, root->name))
1325 * If we asked for subsystems (or explicitly for no
1326 * subsystems) then they must match
1328 if ((opts->subsys_bits || opts->none)
1329 && (opts->subsys_bits != root->subsys_bits))
1335 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1337 struct cgroupfs_root *root;
1339 if (!opts->subsys_bits && !opts->none)
1342 root = kzalloc(sizeof(*root), GFP_KERNEL);
1344 return ERR_PTR(-ENOMEM);
1346 if (!init_root_id(root)) {
1348 return ERR_PTR(-ENOMEM);
1350 init_cgroup_root(root);
1352 root->subsys_bits = opts->subsys_bits;
1353 root->flags = opts->flags;
1354 if (opts->release_agent)
1355 strcpy(root->release_agent_path, opts->release_agent);
1357 strcpy(root->name, opts->name);
1361 static void cgroup_drop_root(struct cgroupfs_root *root)
1366 BUG_ON(!root->hierarchy_id);
1367 spin_lock(&hierarchy_id_lock);
1368 ida_remove(&hierarchy_ida, root->hierarchy_id);
1369 spin_unlock(&hierarchy_id_lock);
1373 static int cgroup_set_super(struct super_block *sb, void *data)
1376 struct cgroup_sb_opts *opts = data;
1378 /* If we don't have a new root, we can't set up a new sb */
1379 if (!opts->new_root)
1382 BUG_ON(!opts->subsys_bits && !opts->none);
1384 ret = set_anon_super(sb, NULL);
1388 sb->s_fs_info = opts->new_root;
1389 opts->new_root->sb = sb;
1391 sb->s_blocksize = PAGE_CACHE_SIZE;
1392 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1393 sb->s_magic = CGROUP_SUPER_MAGIC;
1394 sb->s_op = &cgroup_ops;
1399 static int cgroup_get_rootdir(struct super_block *sb)
1401 struct inode *inode =
1402 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1403 struct dentry *dentry;
1408 inode->i_fop = &simple_dir_operations;
1409 inode->i_op = &cgroup_dir_inode_operations;
1410 /* directories start off with i_nlink == 2 (for "." entry) */
1412 dentry = d_alloc_root(inode);
1417 sb->s_root = dentry;
1421 static int cgroup_get_sb(struct file_system_type *fs_type,
1422 int flags, const char *unused_dev_name,
1423 void *data, struct vfsmount *mnt)
1425 struct cgroup_sb_opts opts;
1426 struct cgroupfs_root *root;
1428 struct super_block *sb;
1429 struct cgroupfs_root *new_root;
1431 /* First find the desired set of subsystems */
1432 mutex_lock(&cgroup_mutex);
1433 ret = parse_cgroupfs_options(data, &opts);
1434 mutex_unlock(&cgroup_mutex);
1439 * Allocate a new cgroup root. We may not need it if we're
1440 * reusing an existing hierarchy.
1442 new_root = cgroup_root_from_opts(&opts);
1443 if (IS_ERR(new_root)) {
1444 ret = PTR_ERR(new_root);
1447 opts.new_root = new_root;
1449 /* Locate an existing or new sb for this hierarchy */
1450 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1453 cgroup_drop_root(opts.new_root);
1457 root = sb->s_fs_info;
1459 if (root == opts.new_root) {
1460 /* We used the new root structure, so this is a new hierarchy */
1461 struct list_head tmp_cg_links;
1462 struct cgroup *root_cgrp = &root->top_cgroup;
1463 struct inode *inode;
1464 struct cgroupfs_root *existing_root;
1467 BUG_ON(sb->s_root != NULL);
1469 ret = cgroup_get_rootdir(sb);
1471 goto drop_new_super;
1472 inode = sb->s_root->d_inode;
1474 mutex_lock(&inode->i_mutex);
1475 mutex_lock(&cgroup_mutex);
1477 if (strlen(root->name)) {
1478 /* Check for name clashes with existing mounts */
1479 for_each_active_root(existing_root) {
1480 if (!strcmp(existing_root->name, root->name)) {
1482 mutex_unlock(&cgroup_mutex);
1483 mutex_unlock(&inode->i_mutex);
1484 goto drop_new_super;
1490 * We're accessing css_set_count without locking
1491 * css_set_lock here, but that's OK - it can only be
1492 * increased by someone holding cgroup_lock, and
1493 * that's us. The worst that can happen is that we
1494 * have some link structures left over
1496 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1498 mutex_unlock(&cgroup_mutex);
1499 mutex_unlock(&inode->i_mutex);
1500 goto drop_new_super;
1503 ret = rebind_subsystems(root, root->subsys_bits);
1504 if (ret == -EBUSY) {
1505 mutex_unlock(&cgroup_mutex);
1506 mutex_unlock(&inode->i_mutex);
1507 free_cg_links(&tmp_cg_links);
1508 goto drop_new_super;
1511 * There must be no failure case after here, since rebinding
1512 * takes care of subsystems' refcounts, which are explicitly
1513 * dropped in the failure exit path.
1516 /* EBUSY should be the only error here */
1519 list_add(&root->root_list, &roots);
1522 sb->s_root->d_fsdata = root_cgrp;
1523 root->top_cgroup.dentry = sb->s_root;
1525 /* Link the top cgroup in this hierarchy into all
1526 * the css_set objects */
1527 write_lock(&css_set_lock);
1528 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1529 struct hlist_head *hhead = &css_set_table[i];
1530 struct hlist_node *node;
1533 hlist_for_each_entry(cg, node, hhead, hlist)
1534 link_css_set(&tmp_cg_links, cg, root_cgrp);
1536 write_unlock(&css_set_lock);
1538 free_cg_links(&tmp_cg_links);
1540 BUG_ON(!list_empty(&root_cgrp->sibling));
1541 BUG_ON(!list_empty(&root_cgrp->children));
1542 BUG_ON(root->number_of_cgroups != 1);
1544 cgroup_populate_dir(root_cgrp);
1545 mutex_unlock(&cgroup_mutex);
1546 mutex_unlock(&inode->i_mutex);
1549 * We re-used an existing hierarchy - the new root (if
1550 * any) is not needed
1552 cgroup_drop_root(opts.new_root);
1553 /* no subsys rebinding, so refcounts don't change */
1554 drop_parsed_module_refcounts(opts.subsys_bits);
1557 simple_set_mnt(mnt, sb);
1558 kfree(opts.release_agent);
1563 deactivate_locked_super(sb);
1565 drop_parsed_module_refcounts(opts.subsys_bits);
1567 kfree(opts.release_agent);
1573 static void cgroup_kill_sb(struct super_block *sb) {
1574 struct cgroupfs_root *root = sb->s_fs_info;
1575 struct cgroup *cgrp = &root->top_cgroup;
1577 struct cg_cgroup_link *link;
1578 struct cg_cgroup_link *saved_link;
1582 BUG_ON(root->number_of_cgroups != 1);
1583 BUG_ON(!list_empty(&cgrp->children));
1584 BUG_ON(!list_empty(&cgrp->sibling));
1586 mutex_lock(&cgroup_mutex);
1588 /* Rebind all subsystems back to the default hierarchy */
1589 ret = rebind_subsystems(root, 0);
1590 /* Shouldn't be able to fail ... */
1594 * Release all the links from css_sets to this hierarchy's
1597 write_lock(&css_set_lock);
1599 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1601 list_del(&link->cg_link_list);
1602 list_del(&link->cgrp_link_list);
1605 write_unlock(&css_set_lock);
1607 if (!list_empty(&root->root_list)) {
1608 list_del(&root->root_list);
1612 mutex_unlock(&cgroup_mutex);
1614 kill_litter_super(sb);
1615 cgroup_drop_root(root);
1618 static struct file_system_type cgroup_fs_type = {
1620 .get_sb = cgroup_get_sb,
1621 .kill_sb = cgroup_kill_sb,
1624 static struct kobject *cgroup_kobj;
1626 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1628 return dentry->d_fsdata;
1631 static inline struct cftype *__d_cft(struct dentry *dentry)
1633 return dentry->d_fsdata;
1637 * cgroup_path - generate the path of a cgroup
1638 * @cgrp: the cgroup in question
1639 * @buf: the buffer to write the path into
1640 * @buflen: the length of the buffer
1642 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1643 * reference. Writes path of cgroup into buf. Returns 0 on success,
1646 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1649 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1650 rcu_read_lock_held() ||
1651 cgroup_lock_is_held());
1653 if (!dentry || cgrp == dummytop) {
1655 * Inactive subsystems have no dentry for their root
1662 start = buf + buflen;
1666 int len = dentry->d_name.len;
1668 if ((start -= len) < buf)
1669 return -ENAMETOOLONG;
1670 memcpy(start, dentry->d_name.name, len);
1671 cgrp = cgrp->parent;
1675 dentry = rcu_dereference_check(cgrp->dentry,
1676 rcu_read_lock_held() ||
1677 cgroup_lock_is_held());
1681 return -ENAMETOOLONG;
1684 memmove(buf, start, buf + buflen - start);
1687 EXPORT_SYMBOL_GPL(cgroup_path);
1690 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1691 * @cgrp: the cgroup the task is attaching to
1692 * @tsk: the task to be attached
1694 * Call holding cgroup_mutex. May take task_lock of
1695 * the task 'tsk' during call.
1697 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1700 struct cgroup_subsys *ss, *failed_ss = NULL;
1701 struct cgroup *oldcgrp;
1703 struct css_set *newcg;
1704 struct cgroupfs_root *root = cgrp->root;
1706 /* Nothing to do if the task is already in that cgroup */
1707 oldcgrp = task_cgroup_from_root(tsk, root);
1708 if (cgrp == oldcgrp)
1711 for_each_subsys(root, ss) {
1712 if (ss->can_attach) {
1713 retval = ss->can_attach(ss, cgrp, tsk, false);
1716 * Remember on which subsystem the can_attach()
1717 * failed, so that we only call cancel_attach()
1718 * against the subsystems whose can_attach()
1719 * succeeded. (See below)
1724 } else if (!capable(CAP_SYS_ADMIN)) {
1725 const struct cred *cred = current_cred(), *tcred;
1727 /* No can_attach() - check perms generically */
1728 tcred = __task_cred(tsk);
1729 if (cred->euid != tcred->uid &&
1730 cred->euid != tcred->suid) {
1741 * Locate or allocate a new css_set for this task,
1742 * based on its final set of cgroups
1744 newcg = find_css_set(cg, cgrp);
1752 if (tsk->flags & PF_EXITING) {
1758 rcu_assign_pointer(tsk->cgroups, newcg);
1761 /* Update the css_set linked lists if we're using them */
1762 write_lock(&css_set_lock);
1763 if (!list_empty(&tsk->cg_list)) {
1764 list_del(&tsk->cg_list);
1765 list_add(&tsk->cg_list, &newcg->tasks);
1767 write_unlock(&css_set_lock);
1769 for_each_subsys(root, ss) {
1771 ss->attach(ss, cgrp, oldcgrp, tsk, false);
1773 set_bit(CGRP_RELEASABLE, &cgrp->flags);
1774 /* put_css_set will not destroy cg until after an RCU grace period */
1778 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1779 * is no longer empty.
1781 cgroup_wakeup_rmdir_waiter(cgrp);
1784 for_each_subsys(root, ss) {
1785 if (ss == failed_ss)
1787 * This subsystem was the one that failed the
1788 * can_attach() check earlier, so we don't need
1789 * to call cancel_attach() against it or any
1790 * remaining subsystems.
1793 if (ss->cancel_attach)
1794 ss->cancel_attach(ss, cgrp, tsk, false);
1801 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1802 * @from: attach to all cgroups of a given task
1803 * @tsk: the task to be attached
1805 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
1807 struct cgroupfs_root *root;
1811 for_each_active_root(root) {
1812 struct cgroup *from_cg = task_cgroup_from_root(from, root);
1814 retval = cgroup_attach_task(from_cg, tsk);
1822 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
1825 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1826 * held. May take task_lock of task
1828 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid)
1830 struct task_struct *tsk;
1835 tsk = find_task_by_vpid(pid);
1836 if (!tsk || tsk->flags & PF_EXITING) {
1840 get_task_struct(tsk);
1844 get_task_struct(tsk);
1847 ret = cgroup_attach_task(cgrp, tsk);
1848 put_task_struct(tsk);
1852 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
1855 if (!cgroup_lock_live_group(cgrp))
1857 ret = attach_task_by_pid(cgrp, pid);
1863 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1864 * @cgrp: the cgroup to be checked for liveness
1866 * On success, returns true; the lock should be later released with
1867 * cgroup_unlock(). On failure returns false with no lock held.
1869 bool cgroup_lock_live_group(struct cgroup *cgrp)
1871 mutex_lock(&cgroup_mutex);
1872 if (cgroup_is_removed(cgrp)) {
1873 mutex_unlock(&cgroup_mutex);
1878 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
1880 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
1883 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1884 if (!cgroup_lock_live_group(cgrp))
1886 strcpy(cgrp->root->release_agent_path, buffer);
1891 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
1892 struct seq_file *seq)
1894 if (!cgroup_lock_live_group(cgrp))
1896 seq_puts(seq, cgrp->root->release_agent_path);
1897 seq_putc(seq, '\n');
1902 /* A buffer size big enough for numbers or short strings */
1903 #define CGROUP_LOCAL_BUFFER_SIZE 64
1905 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
1907 const char __user *userbuf,
1908 size_t nbytes, loff_t *unused_ppos)
1910 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
1916 if (nbytes >= sizeof(buffer))
1918 if (copy_from_user(buffer, userbuf, nbytes))
1921 buffer[nbytes] = 0; /* nul-terminate */
1922 if (cft->write_u64) {
1923 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
1926 retval = cft->write_u64(cgrp, cft, val);
1928 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
1931 retval = cft->write_s64(cgrp, cft, val);
1938 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
1940 const char __user *userbuf,
1941 size_t nbytes, loff_t *unused_ppos)
1943 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
1945 size_t max_bytes = cft->max_write_len;
1946 char *buffer = local_buffer;
1949 max_bytes = sizeof(local_buffer) - 1;
1950 if (nbytes >= max_bytes)
1952 /* Allocate a dynamic buffer if we need one */
1953 if (nbytes >= sizeof(local_buffer)) {
1954 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1958 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
1963 buffer[nbytes] = 0; /* nul-terminate */
1964 retval = cft->write_string(cgrp, cft, strstrip(buffer));
1968 if (buffer != local_buffer)
1973 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
1974 size_t nbytes, loff_t *ppos)
1976 struct cftype *cft = __d_cft(file->f_dentry);
1977 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1979 if (cgroup_is_removed(cgrp))
1982 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
1983 if (cft->write_u64 || cft->write_s64)
1984 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
1985 if (cft->write_string)
1986 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
1988 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
1989 return ret ? ret : nbytes;
1994 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
1996 char __user *buf, size_t nbytes,
1999 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2000 u64 val = cft->read_u64(cgrp, cft);
2001 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2003 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2006 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2008 char __user *buf, size_t nbytes,
2011 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2012 s64 val = cft->read_s64(cgrp, cft);
2013 int len = sprintf(tmp, "%lld\n", (long long) val);
2015 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2018 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2019 size_t nbytes, loff_t *ppos)
2021 struct cftype *cft = __d_cft(file->f_dentry);
2022 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2024 if (cgroup_is_removed(cgrp))
2028 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2030 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2032 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2037 * seqfile ops/methods for returning structured data. Currently just
2038 * supports string->u64 maps, but can be extended in future.
2041 struct cgroup_seqfile_state {
2043 struct cgroup *cgroup;
2046 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2048 struct seq_file *sf = cb->state;
2049 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2052 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2054 struct cgroup_seqfile_state *state = m->private;
2055 struct cftype *cft = state->cft;
2056 if (cft->read_map) {
2057 struct cgroup_map_cb cb = {
2058 .fill = cgroup_map_add,
2061 return cft->read_map(state->cgroup, cft, &cb);
2063 return cft->read_seq_string(state->cgroup, cft, m);
2066 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2068 struct seq_file *seq = file->private_data;
2069 kfree(seq->private);
2070 return single_release(inode, file);
2073 static const struct file_operations cgroup_seqfile_operations = {
2075 .write = cgroup_file_write,
2076 .llseek = seq_lseek,
2077 .release = cgroup_seqfile_release,
2080 static int cgroup_file_open(struct inode *inode, struct file *file)
2085 err = generic_file_open(inode, file);
2088 cft = __d_cft(file->f_dentry);
2090 if (cft->read_map || cft->read_seq_string) {
2091 struct cgroup_seqfile_state *state =
2092 kzalloc(sizeof(*state), GFP_USER);
2096 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2097 file->f_op = &cgroup_seqfile_operations;
2098 err = single_open(file, cgroup_seqfile_show, state);
2101 } else if (cft->open)
2102 err = cft->open(inode, file);
2109 static int cgroup_file_release(struct inode *inode, struct file *file)
2111 struct cftype *cft = __d_cft(file->f_dentry);
2113 return cft->release(inode, file);
2118 * cgroup_rename - Only allow simple rename of directories in place.
2120 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2121 struct inode *new_dir, struct dentry *new_dentry)
2123 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2125 if (new_dentry->d_inode)
2127 if (old_dir != new_dir)
2129 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2132 static const struct file_operations cgroup_file_operations = {
2133 .read = cgroup_file_read,
2134 .write = cgroup_file_write,
2135 .llseek = generic_file_llseek,
2136 .open = cgroup_file_open,
2137 .release = cgroup_file_release,
2140 static const struct inode_operations cgroup_dir_inode_operations = {
2141 .lookup = simple_lookup,
2142 .mkdir = cgroup_mkdir,
2143 .rmdir = cgroup_rmdir,
2144 .rename = cgroup_rename,
2148 * Check if a file is a control file
2150 static inline struct cftype *__file_cft(struct file *file)
2152 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2153 return ERR_PTR(-EINVAL);
2154 return __d_cft(file->f_dentry);
2157 static int cgroup_create_file(struct dentry *dentry, mode_t mode,
2158 struct super_block *sb)
2160 static const struct dentry_operations cgroup_dops = {
2161 .d_iput = cgroup_diput,
2164 struct inode *inode;
2168 if (dentry->d_inode)
2171 inode = cgroup_new_inode(mode, sb);
2175 if (S_ISDIR(mode)) {
2176 inode->i_op = &cgroup_dir_inode_operations;
2177 inode->i_fop = &simple_dir_operations;
2179 /* start off with i_nlink == 2 (for "." entry) */
2182 /* start with the directory inode held, so that we can
2183 * populate it without racing with another mkdir */
2184 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2185 } else if (S_ISREG(mode)) {
2187 inode->i_fop = &cgroup_file_operations;
2189 dentry->d_op = &cgroup_dops;
2190 d_instantiate(dentry, inode);
2191 dget(dentry); /* Extra count - pin the dentry in core */
2196 * cgroup_create_dir - create a directory for an object.
2197 * @cgrp: the cgroup we create the directory for. It must have a valid
2198 * ->parent field. And we are going to fill its ->dentry field.
2199 * @dentry: dentry of the new cgroup
2200 * @mode: mode to set on new directory.
2202 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2205 struct dentry *parent;
2208 parent = cgrp->parent->dentry;
2209 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2211 dentry->d_fsdata = cgrp;
2212 inc_nlink(parent->d_inode);
2213 rcu_assign_pointer(cgrp->dentry, dentry);
2222 * cgroup_file_mode - deduce file mode of a control file
2223 * @cft: the control file in question
2225 * returns cft->mode if ->mode is not 0
2226 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2227 * returns S_IRUGO if it has only a read handler
2228 * returns S_IWUSR if it has only a write hander
2230 static mode_t cgroup_file_mode(const struct cftype *cft)
2237 if (cft->read || cft->read_u64 || cft->read_s64 ||
2238 cft->read_map || cft->read_seq_string)
2241 if (cft->write || cft->write_u64 || cft->write_s64 ||
2242 cft->write_string || cft->trigger)
2248 int cgroup_add_file(struct cgroup *cgrp,
2249 struct cgroup_subsys *subsys,
2250 const struct cftype *cft)
2252 struct dentry *dir = cgrp->dentry;
2253 struct dentry *dentry;
2257 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2258 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2259 strcpy(name, subsys->name);
2262 strcat(name, cft->name);
2263 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2264 dentry = lookup_one_len(name, dir, strlen(name));
2265 if (!IS_ERR(dentry)) {
2266 mode = cgroup_file_mode(cft);
2267 error = cgroup_create_file(dentry, mode | S_IFREG,
2270 dentry->d_fsdata = (void *)cft;
2273 error = PTR_ERR(dentry);
2276 EXPORT_SYMBOL_GPL(cgroup_add_file);
2278 int cgroup_add_files(struct cgroup *cgrp,
2279 struct cgroup_subsys *subsys,
2280 const struct cftype cft[],
2284 for (i = 0; i < count; i++) {
2285 err = cgroup_add_file(cgrp, subsys, &cft[i]);
2291 EXPORT_SYMBOL_GPL(cgroup_add_files);
2294 * cgroup_task_count - count the number of tasks in a cgroup.
2295 * @cgrp: the cgroup in question
2297 * Return the number of tasks in the cgroup.
2299 int cgroup_task_count(const struct cgroup *cgrp)
2302 struct cg_cgroup_link *link;
2304 read_lock(&css_set_lock);
2305 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2306 count += atomic_read(&link->cg->refcount);
2308 read_unlock(&css_set_lock);
2313 * Advance a list_head iterator. The iterator should be positioned at
2314 * the start of a css_set
2316 static void cgroup_advance_iter(struct cgroup *cgrp,
2317 struct cgroup_iter *it)
2319 struct list_head *l = it->cg_link;
2320 struct cg_cgroup_link *link;
2323 /* Advance to the next non-empty css_set */
2326 if (l == &cgrp->css_sets) {
2330 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2332 } while (list_empty(&cg->tasks));
2334 it->task = cg->tasks.next;
2338 * To reduce the fork() overhead for systems that are not actually
2339 * using their cgroups capability, we don't maintain the lists running
2340 * through each css_set to its tasks until we see the list actually
2341 * used - in other words after the first call to cgroup_iter_start().
2343 * The tasklist_lock is not held here, as do_each_thread() and
2344 * while_each_thread() are protected by RCU.
2346 static void cgroup_enable_task_cg_lists(void)
2348 struct task_struct *p, *g;
2349 write_lock(&css_set_lock);
2350 use_task_css_set_links = 1;
2351 do_each_thread(g, p) {
2354 * We should check if the process is exiting, otherwise
2355 * it will race with cgroup_exit() in that the list
2356 * entry won't be deleted though the process has exited.
2358 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2359 list_add(&p->cg_list, &p->cgroups->tasks);
2361 } while_each_thread(g, p);
2362 write_unlock(&css_set_lock);
2365 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2368 * The first time anyone tries to iterate across a cgroup,
2369 * we need to enable the list linking each css_set to its
2370 * tasks, and fix up all existing tasks.
2372 if (!use_task_css_set_links)
2373 cgroup_enable_task_cg_lists();
2375 read_lock(&css_set_lock);
2376 it->cg_link = &cgrp->css_sets;
2377 cgroup_advance_iter(cgrp, it);
2380 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2381 struct cgroup_iter *it)
2383 struct task_struct *res;
2384 struct list_head *l = it->task;
2385 struct cg_cgroup_link *link;
2387 /* If the iterator cg is NULL, we have no tasks */
2390 res = list_entry(l, struct task_struct, cg_list);
2391 /* Advance iterator to find next entry */
2393 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2394 if (l == &link->cg->tasks) {
2395 /* We reached the end of this task list - move on to
2396 * the next cg_cgroup_link */
2397 cgroup_advance_iter(cgrp, it);
2404 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2406 read_unlock(&css_set_lock);
2409 static inline int started_after_time(struct task_struct *t1,
2410 struct timespec *time,
2411 struct task_struct *t2)
2413 int start_diff = timespec_compare(&t1->start_time, time);
2414 if (start_diff > 0) {
2416 } else if (start_diff < 0) {
2420 * Arbitrarily, if two processes started at the same
2421 * time, we'll say that the lower pointer value
2422 * started first. Note that t2 may have exited by now
2423 * so this may not be a valid pointer any longer, but
2424 * that's fine - it still serves to distinguish
2425 * between two tasks started (effectively) simultaneously.
2432 * This function is a callback from heap_insert() and is used to order
2434 * In this case we order the heap in descending task start time.
2436 static inline int started_after(void *p1, void *p2)
2438 struct task_struct *t1 = p1;
2439 struct task_struct *t2 = p2;
2440 return started_after_time(t1, &t2->start_time, t2);
2444 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2445 * @scan: struct cgroup_scanner containing arguments for the scan
2447 * Arguments include pointers to callback functions test_task() and
2449 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2450 * and if it returns true, call process_task() for it also.
2451 * The test_task pointer may be NULL, meaning always true (select all tasks).
2452 * Effectively duplicates cgroup_iter_{start,next,end}()
2453 * but does not lock css_set_lock for the call to process_task().
2454 * The struct cgroup_scanner may be embedded in any structure of the caller's
2456 * It is guaranteed that process_task() will act on every task that
2457 * is a member of the cgroup for the duration of this call. This
2458 * function may or may not call process_task() for tasks that exit
2459 * or move to a different cgroup during the call, or are forked or
2460 * move into the cgroup during the call.
2462 * Note that test_task() may be called with locks held, and may in some
2463 * situations be called multiple times for the same task, so it should
2465 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2466 * pre-allocated and will be used for heap operations (and its "gt" member will
2467 * be overwritten), else a temporary heap will be used (allocation of which
2468 * may cause this function to fail).
2470 int cgroup_scan_tasks(struct cgroup_scanner *scan)
2473 struct cgroup_iter it;
2474 struct task_struct *p, *dropped;
2475 /* Never dereference latest_task, since it's not refcounted */
2476 struct task_struct *latest_task = NULL;
2477 struct ptr_heap tmp_heap;
2478 struct ptr_heap *heap;
2479 struct timespec latest_time = { 0, 0 };
2482 /* The caller supplied our heap and pre-allocated its memory */
2484 heap->gt = &started_after;
2486 /* We need to allocate our own heap memory */
2488 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2490 /* cannot allocate the heap */
2496 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2497 * to determine which are of interest, and using the scanner's
2498 * "process_task" callback to process any of them that need an update.
2499 * Since we don't want to hold any locks during the task updates,
2500 * gather tasks to be processed in a heap structure.
2501 * The heap is sorted by descending task start time.
2502 * If the statically-sized heap fills up, we overflow tasks that
2503 * started later, and in future iterations only consider tasks that
2504 * started after the latest task in the previous pass. This
2505 * guarantees forward progress and that we don't miss any tasks.
2508 cgroup_iter_start(scan->cg, &it);
2509 while ((p = cgroup_iter_next(scan->cg, &it))) {
2511 * Only affect tasks that qualify per the caller's callback,
2512 * if he provided one
2514 if (scan->test_task && !scan->test_task(p, scan))
2517 * Only process tasks that started after the last task
2520 if (!started_after_time(p, &latest_time, latest_task))
2522 dropped = heap_insert(heap, p);
2523 if (dropped == NULL) {
2525 * The new task was inserted; the heap wasn't
2529 } else if (dropped != p) {
2531 * The new task was inserted, and pushed out a
2535 put_task_struct(dropped);
2538 * Else the new task was newer than anything already in
2539 * the heap and wasn't inserted
2542 cgroup_iter_end(scan->cg, &it);
2545 for (i = 0; i < heap->size; i++) {
2546 struct task_struct *q = heap->ptrs[i];
2548 latest_time = q->start_time;
2551 /* Process the task per the caller's callback */
2552 scan->process_task(q, scan);
2556 * If we had to process any tasks at all, scan again
2557 * in case some of them were in the middle of forking
2558 * children that didn't get processed.
2559 * Not the most efficient way to do it, but it avoids
2560 * having to take callback_mutex in the fork path
2564 if (heap == &tmp_heap)
2565 heap_free(&tmp_heap);
2570 * Stuff for reading the 'tasks'/'procs' files.
2572 * Reading this file can return large amounts of data if a cgroup has
2573 * *lots* of attached tasks. So it may need several calls to read(),
2574 * but we cannot guarantee that the information we produce is correct
2575 * unless we produce it entirely atomically.
2580 * The following two functions "fix" the issue where there are more pids
2581 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2582 * TODO: replace with a kernel-wide solution to this problem
2584 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2585 static void *pidlist_allocate(int count)
2587 if (PIDLIST_TOO_LARGE(count))
2588 return vmalloc(count * sizeof(pid_t));
2590 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
2592 static void pidlist_free(void *p)
2594 if (is_vmalloc_addr(p))
2599 static void *pidlist_resize(void *p, int newcount)
2602 /* note: if new alloc fails, old p will still be valid either way */
2603 if (is_vmalloc_addr(p)) {
2604 newlist = vmalloc(newcount * sizeof(pid_t));
2607 memcpy(newlist, p, newcount * sizeof(pid_t));
2610 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
2616 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2617 * If the new stripped list is sufficiently smaller and there's enough memory
2618 * to allocate a new buffer, will let go of the unneeded memory. Returns the
2619 * number of unique elements.
2621 /* is the size difference enough that we should re-allocate the array? */
2622 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
2623 static int pidlist_uniq(pid_t **p, int length)
2630 * we presume the 0th element is unique, so i starts at 1. trivial
2631 * edge cases first; no work needs to be done for either
2633 if (length == 0 || length == 1)
2635 /* src and dest walk down the list; dest counts unique elements */
2636 for (src = 1; src < length; src++) {
2637 /* find next unique element */
2638 while (list[src] == list[src-1]) {
2643 /* dest always points to where the next unique element goes */
2644 list[dest] = list[src];
2649 * if the length difference is large enough, we want to allocate a
2650 * smaller buffer to save memory. if this fails due to out of memory,
2651 * we'll just stay with what we've got.
2653 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
2654 newlist = pidlist_resize(list, dest);
2661 static int cmppid(const void *a, const void *b)
2663 return *(pid_t *)a - *(pid_t *)b;
2667 * find the appropriate pidlist for our purpose (given procs vs tasks)
2668 * returns with the lock on that pidlist already held, and takes care
2669 * of the use count, or returns NULL with no locks held if we're out of
2672 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
2673 enum cgroup_filetype type)
2675 struct cgroup_pidlist *l;
2676 /* don't need task_nsproxy() if we're looking at ourself */
2677 struct pid_namespace *ns = current->nsproxy->pid_ns;
2680 * We can't drop the pidlist_mutex before taking the l->mutex in case
2681 * the last ref-holder is trying to remove l from the list at the same
2682 * time. Holding the pidlist_mutex precludes somebody taking whichever
2683 * list we find out from under us - compare release_pid_array().
2685 mutex_lock(&cgrp->pidlist_mutex);
2686 list_for_each_entry(l, &cgrp->pidlists, links) {
2687 if (l->key.type == type && l->key.ns == ns) {
2688 /* make sure l doesn't vanish out from under us */
2689 down_write(&l->mutex);
2690 mutex_unlock(&cgrp->pidlist_mutex);
2694 /* entry not found; create a new one */
2695 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
2697 mutex_unlock(&cgrp->pidlist_mutex);
2700 init_rwsem(&l->mutex);
2701 down_write(&l->mutex);
2703 l->key.ns = get_pid_ns(ns);
2704 l->use_count = 0; /* don't increment here */
2707 list_add(&l->links, &cgrp->pidlists);
2708 mutex_unlock(&cgrp->pidlist_mutex);
2713 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
2715 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
2716 struct cgroup_pidlist **lp)
2720 int pid, n = 0; /* used for populating the array */
2721 struct cgroup_iter it;
2722 struct task_struct *tsk;
2723 struct cgroup_pidlist *l;
2726 * If cgroup gets more users after we read count, we won't have
2727 * enough space - tough. This race is indistinguishable to the
2728 * caller from the case that the additional cgroup users didn't
2729 * show up until sometime later on.
2731 length = cgroup_task_count(cgrp);
2732 array = pidlist_allocate(length);
2735 /* now, populate the array */
2736 cgroup_iter_start(cgrp, &it);
2737 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2738 if (unlikely(n == length))
2740 /* get tgid or pid for procs or tasks file respectively */
2741 if (type == CGROUP_FILE_PROCS)
2742 pid = task_tgid_vnr(tsk);
2744 pid = task_pid_vnr(tsk);
2745 if (pid > 0) /* make sure to only use valid results */
2748 cgroup_iter_end(cgrp, &it);
2750 /* now sort & (if procs) strip out duplicates */
2751 sort(array, length, sizeof(pid_t), cmppid, NULL);
2752 if (type == CGROUP_FILE_PROCS)
2753 length = pidlist_uniq(&array, length);
2754 l = cgroup_pidlist_find(cgrp, type);
2756 pidlist_free(array);
2759 /* store array, freeing old if necessary - lock already held */
2760 pidlist_free(l->list);
2764 up_write(&l->mutex);
2770 * cgroupstats_build - build and fill cgroupstats
2771 * @stats: cgroupstats to fill information into
2772 * @dentry: A dentry entry belonging to the cgroup for which stats have
2775 * Build and fill cgroupstats so that taskstats can export it to user
2778 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2781 struct cgroup *cgrp;
2782 struct cgroup_iter it;
2783 struct task_struct *tsk;
2786 * Validate dentry by checking the superblock operations,
2787 * and make sure it's a directory.
2789 if (dentry->d_sb->s_op != &cgroup_ops ||
2790 !S_ISDIR(dentry->d_inode->i_mode))
2794 cgrp = dentry->d_fsdata;
2796 cgroup_iter_start(cgrp, &it);
2797 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2798 switch (tsk->state) {
2800 stats->nr_running++;
2802 case TASK_INTERRUPTIBLE:
2803 stats->nr_sleeping++;
2805 case TASK_UNINTERRUPTIBLE:
2806 stats->nr_uninterruptible++;
2809 stats->nr_stopped++;
2812 if (delayacct_is_task_waiting_on_io(tsk))
2813 stats->nr_io_wait++;
2817 cgroup_iter_end(cgrp, &it);
2825 * seq_file methods for the tasks/procs files. The seq_file position is the
2826 * next pid to display; the seq_file iterator is a pointer to the pid
2827 * in the cgroup->l->list array.
2830 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
2833 * Initially we receive a position value that corresponds to
2834 * one more than the last pid shown (or 0 on the first call or
2835 * after a seek to the start). Use a binary-search to find the
2836 * next pid to display, if any
2838 struct cgroup_pidlist *l = s->private;
2839 int index = 0, pid = *pos;
2842 down_read(&l->mutex);
2844 int end = l->length;
2846 while (index < end) {
2847 int mid = (index + end) / 2;
2848 if (l->list[mid] == pid) {
2851 } else if (l->list[mid] <= pid)
2857 /* If we're off the end of the array, we're done */
2858 if (index >= l->length)
2860 /* Update the abstract position to be the actual pid that we found */
2861 iter = l->list + index;
2866 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
2868 struct cgroup_pidlist *l = s->private;
2872 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
2874 struct cgroup_pidlist *l = s->private;
2876 pid_t *end = l->list + l->length;
2878 * Advance to the next pid in the array. If this goes off the
2890 static int cgroup_pidlist_show(struct seq_file *s, void *v)
2892 return seq_printf(s, "%d\n", *(int *)v);
2896 * seq_operations functions for iterating on pidlists through seq_file -
2897 * independent of whether it's tasks or procs
2899 static const struct seq_operations cgroup_pidlist_seq_operations = {
2900 .start = cgroup_pidlist_start,
2901 .stop = cgroup_pidlist_stop,
2902 .next = cgroup_pidlist_next,
2903 .show = cgroup_pidlist_show,
2906 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
2909 * the case where we're the last user of this particular pidlist will
2910 * have us remove it from the cgroup's list, which entails taking the
2911 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
2912 * pidlist_mutex, we have to take pidlist_mutex first.
2914 mutex_lock(&l->owner->pidlist_mutex);
2915 down_write(&l->mutex);
2916 BUG_ON(!l->use_count);
2917 if (!--l->use_count) {
2918 /* we're the last user if refcount is 0; remove and free */
2919 list_del(&l->links);
2920 mutex_unlock(&l->owner->pidlist_mutex);
2921 pidlist_free(l->list);
2922 put_pid_ns(l->key.ns);
2923 up_write(&l->mutex);
2927 mutex_unlock(&l->owner->pidlist_mutex);
2928 up_write(&l->mutex);
2931 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
2933 struct cgroup_pidlist *l;
2934 if (!(file->f_mode & FMODE_READ))
2937 * the seq_file will only be initialized if the file was opened for
2938 * reading; hence we check if it's not null only in that case.
2940 l = ((struct seq_file *)file->private_data)->private;
2941 cgroup_release_pid_array(l);
2942 return seq_release(inode, file);
2945 static const struct file_operations cgroup_pidlist_operations = {
2947 .llseek = seq_lseek,
2948 .write = cgroup_file_write,
2949 .release = cgroup_pidlist_release,
2953 * The following functions handle opens on a file that displays a pidlist
2954 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
2957 /* helper function for the two below it */
2958 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
2960 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2961 struct cgroup_pidlist *l;
2964 /* Nothing to do for write-only files */
2965 if (!(file->f_mode & FMODE_READ))
2968 /* have the array populated */
2969 retval = pidlist_array_load(cgrp, type, &l);
2972 /* configure file information */
2973 file->f_op = &cgroup_pidlist_operations;
2975 retval = seq_open(file, &cgroup_pidlist_seq_operations);
2977 cgroup_release_pid_array(l);
2980 ((struct seq_file *)file->private_data)->private = l;
2983 static int cgroup_tasks_open(struct inode *unused, struct file *file)
2985 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
2987 static int cgroup_procs_open(struct inode *unused, struct file *file)
2989 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
2992 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
2995 return notify_on_release(cgrp);
2998 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3002 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3004 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3006 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3011 * Unregister event and free resources.
3013 * Gets called from workqueue.
3015 static void cgroup_event_remove(struct work_struct *work)
3017 struct cgroup_event *event = container_of(work, struct cgroup_event,
3019 struct cgroup *cgrp = event->cgrp;
3021 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3023 eventfd_ctx_put(event->eventfd);
3029 * Gets called on POLLHUP on eventfd when user closes it.
3031 * Called with wqh->lock held and interrupts disabled.
3033 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3034 int sync, void *key)
3036 struct cgroup_event *event = container_of(wait,
3037 struct cgroup_event, wait);
3038 struct cgroup *cgrp = event->cgrp;
3039 unsigned long flags = (unsigned long)key;
3041 if (flags & POLLHUP) {
3042 __remove_wait_queue(event->wqh, &event->wait);
3043 spin_lock(&cgrp->event_list_lock);
3044 list_del(&event->list);
3045 spin_unlock(&cgrp->event_list_lock);
3047 * We are in atomic context, but cgroup_event_remove() may
3048 * sleep, so we have to call it in workqueue.
3050 schedule_work(&event->remove);
3056 static void cgroup_event_ptable_queue_proc(struct file *file,
3057 wait_queue_head_t *wqh, poll_table *pt)
3059 struct cgroup_event *event = container_of(pt,
3060 struct cgroup_event, pt);
3063 add_wait_queue(wqh, &event->wait);
3067 * Parse input and register new cgroup event handler.
3069 * Input must be in format '<event_fd> <control_fd> <args>'.
3070 * Interpretation of args is defined by control file implementation.
3072 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3075 struct cgroup_event *event = NULL;
3076 unsigned int efd, cfd;
3077 struct file *efile = NULL;
3078 struct file *cfile = NULL;
3082 efd = simple_strtoul(buffer, &endp, 10);
3087 cfd = simple_strtoul(buffer, &endp, 10);
3088 if ((*endp != ' ') && (*endp != '\0'))
3092 event = kzalloc(sizeof(*event), GFP_KERNEL);
3096 INIT_LIST_HEAD(&event->list);
3097 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3098 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3099 INIT_WORK(&event->remove, cgroup_event_remove);
3101 efile = eventfd_fget(efd);
3102 if (IS_ERR(efile)) {
3103 ret = PTR_ERR(efile);
3107 event->eventfd = eventfd_ctx_fileget(efile);
3108 if (IS_ERR(event->eventfd)) {
3109 ret = PTR_ERR(event->eventfd);
3119 /* the process need read permission on control file */
3120 ret = file_permission(cfile, MAY_READ);
3124 event->cft = __file_cft(cfile);
3125 if (IS_ERR(event->cft)) {
3126 ret = PTR_ERR(event->cft);
3130 if (!event->cft->register_event || !event->cft->unregister_event) {
3135 ret = event->cft->register_event(cgrp, event->cft,
3136 event->eventfd, buffer);
3140 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3141 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3147 * Events should be removed after rmdir of cgroup directory, but before
3148 * destroying subsystem state objects. Let's take reference to cgroup
3149 * directory dentry to do that.
3153 spin_lock(&cgrp->event_list_lock);
3154 list_add(&event->list, &cgrp->event_list);
3155 spin_unlock(&cgrp->event_list_lock);
3166 if (event && event->eventfd && !IS_ERR(event->eventfd))
3167 eventfd_ctx_put(event->eventfd);
3169 if (!IS_ERR_OR_NULL(efile))
3178 * for the common functions, 'private' gives the type of file
3180 /* for hysterical raisins, we can't put this on the older files */
3181 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3182 static struct cftype files[] = {
3185 .open = cgroup_tasks_open,
3186 .write_u64 = cgroup_tasks_write,
3187 .release = cgroup_pidlist_release,
3188 .mode = S_IRUGO | S_IWUSR,
3191 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3192 .open = cgroup_procs_open,
3193 /* .write_u64 = cgroup_procs_write, TODO */
3194 .release = cgroup_pidlist_release,
3198 .name = "notify_on_release",
3199 .read_u64 = cgroup_read_notify_on_release,
3200 .write_u64 = cgroup_write_notify_on_release,
3203 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3204 .write_string = cgroup_write_event_control,
3209 static struct cftype cft_release_agent = {
3210 .name = "release_agent",
3211 .read_seq_string = cgroup_release_agent_show,
3212 .write_string = cgroup_release_agent_write,
3213 .max_write_len = PATH_MAX,
3216 static int cgroup_populate_dir(struct cgroup *cgrp)
3219 struct cgroup_subsys *ss;
3221 /* First clear out any existing files */
3222 cgroup_clear_directory(cgrp->dentry);
3224 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
3228 if (cgrp == cgrp->top_cgroup) {
3229 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
3233 for_each_subsys(cgrp->root, ss) {
3234 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
3237 /* This cgroup is ready now */
3238 for_each_subsys(cgrp->root, ss) {
3239 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3241 * Update id->css pointer and make this css visible from
3242 * CSS ID functions. This pointer will be dereferened
3243 * from RCU-read-side without locks.
3246 rcu_assign_pointer(css->id->css, css);
3252 static void init_cgroup_css(struct cgroup_subsys_state *css,
3253 struct cgroup_subsys *ss,
3254 struct cgroup *cgrp)
3257 atomic_set(&css->refcnt, 1);
3260 if (cgrp == dummytop)
3261 set_bit(CSS_ROOT, &css->flags);
3262 BUG_ON(cgrp->subsys[ss->subsys_id]);
3263 cgrp->subsys[ss->subsys_id] = css;
3266 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3268 /* We need to take each hierarchy_mutex in a consistent order */
3272 * No worry about a race with rebind_subsystems that might mess up the
3273 * locking order, since both parties are under cgroup_mutex.
3275 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3276 struct cgroup_subsys *ss = subsys[i];
3279 if (ss->root == root)
3280 mutex_lock(&ss->hierarchy_mutex);
3284 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3288 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3289 struct cgroup_subsys *ss = subsys[i];
3292 if (ss->root == root)
3293 mutex_unlock(&ss->hierarchy_mutex);
3298 * cgroup_create - create a cgroup
3299 * @parent: cgroup that will be parent of the new cgroup
3300 * @dentry: dentry of the new cgroup
3301 * @mode: mode to set on new inode
3303 * Must be called with the mutex on the parent inode held
3305 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3308 struct cgroup *cgrp;
3309 struct cgroupfs_root *root = parent->root;
3311 struct cgroup_subsys *ss;
3312 struct super_block *sb = root->sb;
3314 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3318 /* Grab a reference on the superblock so the hierarchy doesn't
3319 * get deleted on unmount if there are child cgroups. This
3320 * can be done outside cgroup_mutex, since the sb can't
3321 * disappear while someone has an open control file on the
3323 atomic_inc(&sb->s_active);
3325 mutex_lock(&cgroup_mutex);
3327 init_cgroup_housekeeping(cgrp);
3329 cgrp->parent = parent;
3330 cgrp->root = parent->root;
3331 cgrp->top_cgroup = parent->top_cgroup;
3333 if (notify_on_release(parent))
3334 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3336 for_each_subsys(root, ss) {
3337 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
3343 init_cgroup_css(css, ss, cgrp);
3345 err = alloc_css_id(ss, parent, cgrp);
3349 /* At error, ->destroy() callback has to free assigned ID. */
3352 cgroup_lock_hierarchy(root);
3353 list_add(&cgrp->sibling, &cgrp->parent->children);
3354 cgroup_unlock_hierarchy(root);
3355 root->number_of_cgroups++;
3357 err = cgroup_create_dir(cgrp, dentry, mode);
3361 set_bit(CGRP_RELEASABLE, &parent->flags);
3363 /* The cgroup directory was pre-locked for us */
3364 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3366 err = cgroup_populate_dir(cgrp);
3367 /* If err < 0, we have a half-filled directory - oh well ;) */
3369 mutex_unlock(&cgroup_mutex);
3370 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3376 cgroup_lock_hierarchy(root);
3377 list_del(&cgrp->sibling);
3378 cgroup_unlock_hierarchy(root);
3379 root->number_of_cgroups--;
3383 for_each_subsys(root, ss) {
3384 if (cgrp->subsys[ss->subsys_id])
3385 ss->destroy(ss, cgrp);
3388 mutex_unlock(&cgroup_mutex);
3390 /* Release the reference count that we took on the superblock */
3391 deactivate_super(sb);
3397 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3399 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3401 /* the vfs holds inode->i_mutex already */
3402 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3405 static int cgroup_has_css_refs(struct cgroup *cgrp)
3407 /* Check the reference count on each subsystem. Since we
3408 * already established that there are no tasks in the
3409 * cgroup, if the css refcount is also 1, then there should
3410 * be no outstanding references, so the subsystem is safe to
3411 * destroy. We scan across all subsystems rather than using
3412 * the per-hierarchy linked list of mounted subsystems since
3413 * we can be called via check_for_release() with no
3414 * synchronization other than RCU, and the subsystem linked
3415 * list isn't RCU-safe */
3418 * We won't need to lock the subsys array, because the subsystems
3419 * we're concerned about aren't going anywhere since our cgroup root
3420 * has a reference on them.
3422 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3423 struct cgroup_subsys *ss = subsys[i];
3424 struct cgroup_subsys_state *css;
3425 /* Skip subsystems not present or not in this hierarchy */
3426 if (ss == NULL || ss->root != cgrp->root)
3428 css = cgrp->subsys[ss->subsys_id];
3429 /* When called from check_for_release() it's possible
3430 * that by this point the cgroup has been removed
3431 * and the css deleted. But a false-positive doesn't
3432 * matter, since it can only happen if the cgroup
3433 * has been deleted and hence no longer needs the
3434 * release agent to be called anyway. */
3435 if (css && (atomic_read(&css->refcnt) > 1))
3442 * Atomically mark all (or else none) of the cgroup's CSS objects as
3443 * CSS_REMOVED. Return true on success, or false if the cgroup has
3444 * busy subsystems. Call with cgroup_mutex held
3447 static int cgroup_clear_css_refs(struct cgroup *cgrp)
3449 struct cgroup_subsys *ss;
3450 unsigned long flags;
3451 bool failed = false;
3452 local_irq_save(flags);
3453 for_each_subsys(cgrp->root, ss) {
3454 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3457 /* We can only remove a CSS with a refcnt==1 */
3458 refcnt = atomic_read(&css->refcnt);
3465 * Drop the refcnt to 0 while we check other
3466 * subsystems. This will cause any racing
3467 * css_tryget() to spin until we set the
3468 * CSS_REMOVED bits or abort
3470 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3476 for_each_subsys(cgrp->root, ss) {
3477 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3480 * Restore old refcnt if we previously managed
3481 * to clear it from 1 to 0
3483 if (!atomic_read(&css->refcnt))
3484 atomic_set(&css->refcnt, 1);
3486 /* Commit the fact that the CSS is removed */
3487 set_bit(CSS_REMOVED, &css->flags);
3490 local_irq_restore(flags);
3494 /* checks if all of the css_sets attached to a cgroup have a refcount of 0.
3495 * Must be called with css_set_lock held */
3496 static int cgroup_css_sets_empty(struct cgroup *cgrp)
3498 struct cg_cgroup_link *link;
3500 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
3501 struct css_set *cg = link->cg;
3502 if (atomic_read(&cg->refcount) > 0)
3509 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3511 struct cgroup *cgrp = dentry->d_fsdata;
3513 struct cgroup *parent;
3515 struct cgroup_event *event, *tmp;
3518 /* the vfs holds both inode->i_mutex already */
3520 mutex_lock(&cgroup_mutex);
3521 if (!cgroup_css_sets_empty(cgrp)) {
3522 mutex_unlock(&cgroup_mutex);
3525 if (!list_empty(&cgrp->children)) {
3526 mutex_unlock(&cgroup_mutex);
3529 mutex_unlock(&cgroup_mutex);
3532 * In general, subsystem has no css->refcnt after pre_destroy(). But
3533 * in racy cases, subsystem may have to get css->refcnt after
3534 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3535 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3536 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3537 * and subsystem's reference count handling. Please see css_get/put
3538 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3540 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3543 * Call pre_destroy handlers of subsys. Notify subsystems
3544 * that rmdir() request comes.
3546 ret = cgroup_call_pre_destroy(cgrp);
3548 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3552 mutex_lock(&cgroup_mutex);
3553 parent = cgrp->parent;
3554 if (!cgroup_css_sets_empty(cgrp) || !list_empty(&cgrp->children)) {
3555 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3556 mutex_unlock(&cgroup_mutex);
3559 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
3560 if (!cgroup_clear_css_refs(cgrp)) {
3561 mutex_unlock(&cgroup_mutex);
3563 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3564 * prepare_to_wait(), we need to check this flag.
3566 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
3568 finish_wait(&cgroup_rmdir_waitq, &wait);
3569 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3570 if (signal_pending(current))
3574 /* NO css_tryget() can success after here. */
3575 finish_wait(&cgroup_rmdir_waitq, &wait);
3576 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3578 spin_lock(&release_list_lock);
3579 set_bit(CGRP_REMOVED, &cgrp->flags);
3580 if (!list_empty(&cgrp->release_list))
3581 list_del(&cgrp->release_list);
3582 spin_unlock(&release_list_lock);
3584 cgroup_lock_hierarchy(cgrp->root);
3585 /* delete this cgroup from parent->children */
3586 list_del(&cgrp->sibling);
3587 cgroup_unlock_hierarchy(cgrp->root);
3589 spin_lock(&cgrp->dentry->d_lock);
3590 d = dget(cgrp->dentry);
3591 spin_unlock(&d->d_lock);
3593 cgroup_d_remove_dir(d);
3596 check_for_release(parent);
3599 * Unregister events and notify userspace.
3600 * Notify userspace about cgroup removing only after rmdir of cgroup
3601 * directory to avoid race between userspace and kernelspace
3603 spin_lock(&cgrp->event_list_lock);
3604 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
3605 list_del(&event->list);
3606 remove_wait_queue(event->wqh, &event->wait);
3607 eventfd_signal(event->eventfd, 1);
3608 schedule_work(&event->remove);
3610 spin_unlock(&cgrp->event_list_lock);
3612 mutex_unlock(&cgroup_mutex);
3616 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
3618 struct cgroup_subsys_state *css;
3620 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
3622 /* Create the top cgroup state for this subsystem */
3623 list_add(&ss->sibling, &rootnode.subsys_list);
3624 ss->root = &rootnode;
3625 css = ss->create(ss, dummytop);
3626 /* We don't handle early failures gracefully */
3627 BUG_ON(IS_ERR(css));
3628 init_cgroup_css(css, ss, dummytop);
3630 /* Update the init_css_set to contain a subsys
3631 * pointer to this state - since the subsystem is
3632 * newly registered, all tasks and hence the
3633 * init_css_set is in the subsystem's top cgroup. */
3634 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
3636 need_forkexit_callback |= ss->fork || ss->exit;
3638 /* At system boot, before all subsystems have been
3639 * registered, no tasks have been forked, so we don't
3640 * need to invoke fork callbacks here. */
3641 BUG_ON(!list_empty(&init_task.tasks));
3643 mutex_init(&ss->hierarchy_mutex);
3644 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3647 /* this function shouldn't be used with modular subsystems, since they
3648 * need to register a subsys_id, among other things */
3653 * cgroup_load_subsys: load and register a modular subsystem at runtime
3654 * @ss: the subsystem to load
3656 * This function should be called in a modular subsystem's initcall. If the
3657 * subsystem is built as a module, it will be assigned a new subsys_id and set
3658 * up for use. If the subsystem is built-in anyway, work is delegated to the
3659 * simpler cgroup_init_subsys.
3661 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
3664 struct cgroup_subsys_state *css;
3666 /* check name and function validity */
3667 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
3668 ss->create == NULL || ss->destroy == NULL)
3672 * we don't support callbacks in modular subsystems. this check is
3673 * before the ss->module check for consistency; a subsystem that could
3674 * be a module should still have no callbacks even if the user isn't
3675 * compiling it as one.
3677 if (ss->fork || ss->exit)
3681 * an optionally modular subsystem is built-in: we want to do nothing,
3682 * since cgroup_init_subsys will have already taken care of it.
3684 if (ss->module == NULL) {
3685 /* a few sanity checks */
3686 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
3687 BUG_ON(subsys[ss->subsys_id] != ss);
3692 * need to register a subsys id before anything else - for example,
3693 * init_cgroup_css needs it.
3695 mutex_lock(&cgroup_mutex);
3696 /* find the first empty slot in the array */
3697 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
3698 if (subsys[i] == NULL)
3701 if (i == CGROUP_SUBSYS_COUNT) {
3702 /* maximum number of subsystems already registered! */
3703 mutex_unlock(&cgroup_mutex);
3706 /* assign ourselves the subsys_id */
3711 * no ss->create seems to need anything important in the ss struct, so
3712 * this can happen first (i.e. before the rootnode attachment).
3714 css = ss->create(ss, dummytop);
3716 /* failure case - need to deassign the subsys[] slot. */
3718 mutex_unlock(&cgroup_mutex);
3719 return PTR_ERR(css);
3722 list_add(&ss->sibling, &rootnode.subsys_list);
3723 ss->root = &rootnode;
3725 /* our new subsystem will be attached to the dummy hierarchy. */
3726 init_cgroup_css(css, ss, dummytop);
3727 /* init_idr must be after init_cgroup_css because it sets css->id. */
3729 int ret = cgroup_init_idr(ss, css);
3731 dummytop->subsys[ss->subsys_id] = NULL;
3732 ss->destroy(ss, dummytop);
3734 mutex_unlock(&cgroup_mutex);
3740 * Now we need to entangle the css into the existing css_sets. unlike
3741 * in cgroup_init_subsys, there are now multiple css_sets, so each one
3742 * will need a new pointer to it; done by iterating the css_set_table.
3743 * furthermore, modifying the existing css_sets will corrupt the hash
3744 * table state, so each changed css_set will need its hash recomputed.
3745 * this is all done under the css_set_lock.
3747 write_lock(&css_set_lock);
3748 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
3750 struct hlist_node *node, *tmp;
3751 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
3753 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
3754 /* skip entries that we already rehashed */
3755 if (cg->subsys[ss->subsys_id])
3757 /* remove existing entry */
3758 hlist_del(&cg->hlist);
3760 cg->subsys[ss->subsys_id] = css;
3761 /* recompute hash and restore entry */
3762 new_bucket = css_set_hash(cg->subsys);
3763 hlist_add_head(&cg->hlist, new_bucket);
3766 write_unlock(&css_set_lock);
3768 mutex_init(&ss->hierarchy_mutex);
3769 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
3773 mutex_unlock(&cgroup_mutex);
3776 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
3779 * cgroup_unload_subsys: unload a modular subsystem
3780 * @ss: the subsystem to unload
3782 * This function should be called in a modular subsystem's exitcall. When this
3783 * function is invoked, the refcount on the subsystem's module will be 0, so
3784 * the subsystem will not be attached to any hierarchy.
3786 void cgroup_unload_subsys(struct cgroup_subsys *ss)
3788 struct cg_cgroup_link *link;
3789 struct hlist_head *hhead;
3791 BUG_ON(ss->module == NULL);
3794 * we shouldn't be called if the subsystem is in use, and the use of
3795 * try_module_get in parse_cgroupfs_options should ensure that it
3796 * doesn't start being used while we're killing it off.
3798 BUG_ON(ss->root != &rootnode);
3800 mutex_lock(&cgroup_mutex);
3801 /* deassign the subsys_id */
3802 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
3803 subsys[ss->subsys_id] = NULL;
3805 /* remove subsystem from rootnode's list of subsystems */
3806 list_del(&ss->sibling);
3809 * disentangle the css from all css_sets attached to the dummytop. as
3810 * in loading, we need to pay our respects to the hashtable gods.
3812 write_lock(&css_set_lock);
3813 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
3814 struct css_set *cg = link->cg;
3816 hlist_del(&cg->hlist);
3817 BUG_ON(!cg->subsys[ss->subsys_id]);
3818 cg->subsys[ss->subsys_id] = NULL;
3819 hhead = css_set_hash(cg->subsys);
3820 hlist_add_head(&cg->hlist, hhead);
3822 write_unlock(&css_set_lock);
3825 * remove subsystem's css from the dummytop and free it - need to free
3826 * before marking as null because ss->destroy needs the cgrp->subsys
3827 * pointer to find their state. note that this also takes care of
3828 * freeing the css_id.
3830 ss->destroy(ss, dummytop);
3831 dummytop->subsys[ss->subsys_id] = NULL;
3833 mutex_unlock(&cgroup_mutex);
3835 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
3838 * cgroup_init_early - cgroup initialization at system boot
3840 * Initialize cgroups at system boot, and initialize any
3841 * subsystems that request early init.
3843 int __init cgroup_init_early(void)
3846 atomic_set(&init_css_set.refcount, 1);
3847 INIT_LIST_HEAD(&init_css_set.cg_links);
3848 INIT_LIST_HEAD(&init_css_set.tasks);
3849 INIT_HLIST_NODE(&init_css_set.hlist);
3851 init_cgroup_root(&rootnode);
3853 init_task.cgroups = &init_css_set;
3855 init_css_set_link.cg = &init_css_set;
3856 init_css_set_link.cgrp = dummytop;
3857 list_add(&init_css_set_link.cgrp_link_list,
3858 &rootnode.top_cgroup.css_sets);
3859 list_add(&init_css_set_link.cg_link_list,
3860 &init_css_set.cg_links);
3862 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
3863 INIT_HLIST_HEAD(&css_set_table[i]);
3865 /* at bootup time, we don't worry about modular subsystems */
3866 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
3867 struct cgroup_subsys *ss = subsys[i];
3870 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
3871 BUG_ON(!ss->create);
3872 BUG_ON(!ss->destroy);
3873 if (ss->subsys_id != i) {
3874 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
3875 ss->name, ss->subsys_id);
3880 cgroup_init_subsys(ss);
3886 * cgroup_init - cgroup initialization
3888 * Register cgroup filesystem and /proc file, and initialize
3889 * any subsystems that didn't request early init.
3891 int __init cgroup_init(void)
3895 struct hlist_head *hhead;
3897 err = bdi_init(&cgroup_backing_dev_info);
3901 /* at bootup time, we don't worry about modular subsystems */
3902 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
3903 struct cgroup_subsys *ss = subsys[i];
3904 if (!ss->early_init)
3905 cgroup_init_subsys(ss);
3907 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
3910 /* Add init_css_set to the hash table */
3911 hhead = css_set_hash(init_css_set.subsys);
3912 hlist_add_head(&init_css_set.hlist, hhead);
3913 BUG_ON(!init_root_id(&rootnode));
3915 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
3921 err = register_filesystem(&cgroup_fs_type);
3923 kobject_put(cgroup_kobj);
3927 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
3931 bdi_destroy(&cgroup_backing_dev_info);
3937 * proc_cgroup_show()
3938 * - Print task's cgroup paths into seq_file, one line for each hierarchy
3939 * - Used for /proc/<pid>/cgroup.
3940 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
3941 * doesn't really matter if tsk->cgroup changes after we read it,
3942 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
3943 * anyway. No need to check that tsk->cgroup != NULL, thanks to
3944 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
3945 * cgroup to top_cgroup.
3948 /* TODO: Use a proper seq_file iterator */
3949 static int proc_cgroup_show(struct seq_file *m, void *v)
3952 struct task_struct *tsk;
3955 struct cgroupfs_root *root;
3958 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3964 tsk = get_pid_task(pid, PIDTYPE_PID);
3970 mutex_lock(&cgroup_mutex);
3972 for_each_active_root(root) {
3973 struct cgroup_subsys *ss;
3974 struct cgroup *cgrp;
3977 seq_printf(m, "%d:", root->hierarchy_id);
3978 for_each_subsys(root, ss)
3979 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
3980 if (strlen(root->name))
3981 seq_printf(m, "%sname=%s", count ? "," : "",
3984 cgrp = task_cgroup_from_root(tsk, root);
3985 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
3993 mutex_unlock(&cgroup_mutex);
3994 put_task_struct(tsk);
4001 static int cgroup_open(struct inode *inode, struct file *file)
4003 struct pid *pid = PROC_I(inode)->pid;
4004 return single_open(file, proc_cgroup_show, pid);
4007 const struct file_operations proc_cgroup_operations = {
4008 .open = cgroup_open,
4010 .llseek = seq_lseek,
4011 .release = single_release,
4014 /* Display information about each subsystem and each hierarchy */
4015 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4019 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4021 * ideally we don't want subsystems moving around while we do this.
4022 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4023 * subsys/hierarchy state.
4025 mutex_lock(&cgroup_mutex);
4026 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4027 struct cgroup_subsys *ss = subsys[i];
4030 seq_printf(m, "%s\t%d\t%d\t%d\n",
4031 ss->name, ss->root->hierarchy_id,
4032 ss->root->number_of_cgroups, !ss->disabled);
4034 mutex_unlock(&cgroup_mutex);
4038 static int cgroupstats_open(struct inode *inode, struct file *file)
4040 return single_open(file, proc_cgroupstats_show, NULL);
4043 static const struct file_operations proc_cgroupstats_operations = {
4044 .open = cgroupstats_open,
4046 .llseek = seq_lseek,
4047 .release = single_release,
4051 * cgroup_fork - attach newly forked task to its parents cgroup.
4052 * @child: pointer to task_struct of forking parent process.
4054 * Description: A task inherits its parent's cgroup at fork().
4056 * A pointer to the shared css_set was automatically copied in
4057 * fork.c by dup_task_struct(). However, we ignore that copy, since
4058 * it was not made under the protection of RCU or cgroup_mutex, so
4059 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4060 * have already changed current->cgroups, allowing the previously
4061 * referenced cgroup group to be removed and freed.
4063 * At the point that cgroup_fork() is called, 'current' is the parent
4064 * task, and the passed argument 'child' points to the child task.
4066 void cgroup_fork(struct task_struct *child)
4069 child->cgroups = current->cgroups;
4070 get_css_set(child->cgroups);
4071 task_unlock(current);
4072 INIT_LIST_HEAD(&child->cg_list);
4076 * cgroup_fork_callbacks - run fork callbacks
4077 * @child: the new task
4079 * Called on a new task very soon before adding it to the
4080 * tasklist. No need to take any locks since no-one can
4081 * be operating on this task.
4083 void cgroup_fork_callbacks(struct task_struct *child)
4085 if (need_forkexit_callback) {
4088 * forkexit callbacks are only supported for builtin
4089 * subsystems, and the builtin section of the subsys array is
4090 * immutable, so we don't need to lock the subsys array here.
4092 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4093 struct cgroup_subsys *ss = subsys[i];
4095 ss->fork(ss, child);
4101 * cgroup_post_fork - called on a new task after adding it to the task list
4102 * @child: the task in question
4104 * Adds the task to the list running through its css_set if necessary.
4105 * Has to be after the task is visible on the task list in case we race
4106 * with the first call to cgroup_iter_start() - to guarantee that the
4107 * new task ends up on its list.
4109 void cgroup_post_fork(struct task_struct *child)
4111 if (use_task_css_set_links) {
4112 write_lock(&css_set_lock);
4114 if (list_empty(&child->cg_list))
4115 list_add(&child->cg_list, &child->cgroups->tasks);
4117 write_unlock(&css_set_lock);
4121 * cgroup_exit - detach cgroup from exiting task
4122 * @tsk: pointer to task_struct of exiting process
4123 * @run_callback: run exit callbacks?
4125 * Description: Detach cgroup from @tsk and release it.
4127 * Note that cgroups marked notify_on_release force every task in
4128 * them to take the global cgroup_mutex mutex when exiting.
4129 * This could impact scaling on very large systems. Be reluctant to
4130 * use notify_on_release cgroups where very high task exit scaling
4131 * is required on large systems.
4133 * the_top_cgroup_hack:
4135 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4137 * We call cgroup_exit() while the task is still competent to
4138 * handle notify_on_release(), then leave the task attached to the
4139 * root cgroup in each hierarchy for the remainder of its exit.
4141 * To do this properly, we would increment the reference count on
4142 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4143 * code we would add a second cgroup function call, to drop that
4144 * reference. This would just create an unnecessary hot spot on
4145 * the top_cgroup reference count, to no avail.
4147 * Normally, holding a reference to a cgroup without bumping its
4148 * count is unsafe. The cgroup could go away, or someone could
4149 * attach us to a different cgroup, decrementing the count on
4150 * the first cgroup that we never incremented. But in this case,
4151 * top_cgroup isn't going away, and either task has PF_EXITING set,
4152 * which wards off any cgroup_attach_task() attempts, or task is a failed
4153 * fork, never visible to cgroup_attach_task.
4155 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4160 if (run_callbacks && need_forkexit_callback) {
4162 * modular subsystems can't use callbacks, so no need to lock
4165 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4166 struct cgroup_subsys *ss = subsys[i];
4173 * Unlink from the css_set task list if necessary.
4174 * Optimistically check cg_list before taking
4177 if (!list_empty(&tsk->cg_list)) {
4178 write_lock(&css_set_lock);
4179 if (!list_empty(&tsk->cg_list))
4180 list_del_init(&tsk->cg_list);
4181 write_unlock(&css_set_lock);
4184 /* Reassign the task to the init_css_set. */
4187 tsk->cgroups = &init_css_set;
4194 * cgroup_clone - clone the cgroup the given subsystem is attached to
4195 * @tsk: the task to be moved
4196 * @subsys: the given subsystem
4197 * @nodename: the name for the new cgroup
4199 * Duplicate the current cgroup in the hierarchy that the given
4200 * subsystem is attached to, and move this task into the new
4203 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys,
4206 struct dentry *dentry;
4208 struct cgroup *parent, *child;
4209 struct inode *inode;
4211 struct cgroupfs_root *root;
4212 struct cgroup_subsys *ss;
4214 /* We shouldn't be called by an unregistered subsystem */
4215 BUG_ON(!subsys->active);
4217 /* First figure out what hierarchy and cgroup we're dealing
4218 * with, and pin them so we can drop cgroup_mutex */
4219 mutex_lock(&cgroup_mutex);
4221 root = subsys->root;
4222 if (root == &rootnode) {
4223 mutex_unlock(&cgroup_mutex);
4227 /* Pin the hierarchy */
4228 if (!atomic_inc_not_zero(&root->sb->s_active)) {
4229 /* We race with the final deactivate_super() */
4230 mutex_unlock(&cgroup_mutex);
4234 /* Keep the cgroup alive */
4236 parent = task_cgroup(tsk, subsys->subsys_id);
4241 mutex_unlock(&cgroup_mutex);
4243 /* Now do the VFS work to create a cgroup */
4244 inode = parent->dentry->d_inode;
4246 /* Hold the parent directory mutex across this operation to
4247 * stop anyone else deleting the new cgroup */
4248 mutex_lock(&inode->i_mutex);
4249 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
4250 if (IS_ERR(dentry)) {
4252 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
4254 ret = PTR_ERR(dentry);
4258 /* Create the cgroup directory, which also creates the cgroup */
4259 ret = vfs_mkdir(inode, dentry, 0755);
4260 child = __d_cgrp(dentry);
4264 "Failed to create cgroup %s: %d\n", nodename,
4269 /* The cgroup now exists. Retake cgroup_mutex and check
4270 * that we're still in the same state that we thought we
4272 mutex_lock(&cgroup_mutex);
4273 if ((root != subsys->root) ||
4274 (parent != task_cgroup(tsk, subsys->subsys_id))) {
4275 /* Aargh, we raced ... */
4276 mutex_unlock(&inode->i_mutex);
4279 deactivate_super(root->sb);
4280 /* The cgroup is still accessible in the VFS, but
4281 * we're not going to try to rmdir() it at this
4284 "Race in cgroup_clone() - leaking cgroup %s\n",
4289 /* do any required auto-setup */
4290 for_each_subsys(root, ss) {
4292 ss->post_clone(ss, child);
4295 /* All seems fine. Finish by moving the task into the new cgroup */
4296 ret = cgroup_attach_task(child, tsk);
4297 mutex_unlock(&cgroup_mutex);
4300 mutex_unlock(&inode->i_mutex);
4302 mutex_lock(&cgroup_mutex);
4304 mutex_unlock(&cgroup_mutex);
4305 deactivate_super(root->sb);
4310 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4311 * @cgrp: the cgroup in question
4312 * @task: the task in question
4314 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4317 * If we are sending in dummytop, then presumably we are creating
4318 * the top cgroup in the subsystem.
4320 * Called only by the ns (nsproxy) cgroup.
4322 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4325 struct cgroup *target;
4327 if (cgrp == dummytop)
4330 target = task_cgroup_from_root(task, cgrp->root);
4331 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4332 cgrp = cgrp->parent;
4333 ret = (cgrp == target);
4337 static void check_for_release(struct cgroup *cgrp)
4339 /* All of these checks rely on RCU to keep the cgroup
4340 * structure alive */
4341 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4342 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4343 /* Control Group is currently removeable. If it's not
4344 * already queued for a userspace notification, queue
4346 int need_schedule_work = 0;
4347 spin_lock(&release_list_lock);
4348 if (!cgroup_is_removed(cgrp) &&
4349 list_empty(&cgrp->release_list)) {
4350 list_add(&cgrp->release_list, &release_list);
4351 need_schedule_work = 1;
4353 spin_unlock(&release_list_lock);
4354 if (need_schedule_work)
4355 schedule_work(&release_agent_work);
4359 /* Caller must verify that the css is not for root cgroup */
4360 void __css_get(struct cgroup_subsys_state *css, int count)
4362 atomic_add(count, &css->refcnt);
4363 set_bit(CGRP_RELEASABLE, &css->cgroup->flags);
4365 EXPORT_SYMBOL_GPL(__css_get);
4367 /* Caller must verify that the css is not for root cgroup */
4368 void __css_put(struct cgroup_subsys_state *css, int count)
4370 struct cgroup *cgrp = css->cgroup;
4373 val = atomic_sub_return(count, &css->refcnt);
4375 check_for_release(cgrp);
4376 cgroup_wakeup_rmdir_waiter(cgrp);
4379 WARN_ON_ONCE(val < 1);
4381 EXPORT_SYMBOL_GPL(__css_put);
4384 * Notify userspace when a cgroup is released, by running the
4385 * configured release agent with the name of the cgroup (path
4386 * relative to the root of cgroup file system) as the argument.
4388 * Most likely, this user command will try to rmdir this cgroup.
4390 * This races with the possibility that some other task will be
4391 * attached to this cgroup before it is removed, or that some other
4392 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4393 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4394 * unused, and this cgroup will be reprieved from its death sentence,
4395 * to continue to serve a useful existence. Next time it's released,
4396 * we will get notified again, if it still has 'notify_on_release' set.
4398 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4399 * means only wait until the task is successfully execve()'d. The
4400 * separate release agent task is forked by call_usermodehelper(),
4401 * then control in this thread returns here, without waiting for the
4402 * release agent task. We don't bother to wait because the caller of
4403 * this routine has no use for the exit status of the release agent
4404 * task, so no sense holding our caller up for that.
4406 static void cgroup_release_agent(struct work_struct *work)
4408 BUG_ON(work != &release_agent_work);
4409 mutex_lock(&cgroup_mutex);
4410 spin_lock(&release_list_lock);
4411 while (!list_empty(&release_list)) {
4412 char *argv[3], *envp[3];
4414 char *pathbuf = NULL, *agentbuf = NULL;
4415 struct cgroup *cgrp = list_entry(release_list.next,
4418 list_del_init(&cgrp->release_list);
4419 spin_unlock(&release_list_lock);
4420 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4423 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4425 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4430 argv[i++] = agentbuf;
4431 argv[i++] = pathbuf;
4435 /* minimal command environment */
4436 envp[i++] = "HOME=/";
4437 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4440 /* Drop the lock while we invoke the usermode helper,
4441 * since the exec could involve hitting disk and hence
4442 * be a slow process */
4443 mutex_unlock(&cgroup_mutex);
4444 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4445 mutex_lock(&cgroup_mutex);
4449 spin_lock(&release_list_lock);
4451 spin_unlock(&release_list_lock);
4452 mutex_unlock(&cgroup_mutex);
4455 static int __init cgroup_disable(char *str)
4460 while ((token = strsep(&str, ",")) != NULL) {
4464 * cgroup_disable, being at boot time, can't know about module
4465 * subsystems, so we don't worry about them.
4467 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4468 struct cgroup_subsys *ss = subsys[i];
4470 if (!strcmp(token, ss->name)) {
4472 printk(KERN_INFO "Disabling %s control group"
4473 " subsystem\n", ss->name);
4480 __setup("cgroup_disable=", cgroup_disable);
4483 * Functons for CSS ID.
4487 *To get ID other than 0, this should be called when !cgroup_is_removed().
4489 unsigned short css_id(struct cgroup_subsys_state *css)
4491 struct css_id *cssid;
4494 * This css_id() can return correct value when somone has refcnt
4495 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4496 * it's unchanged until freed.
4498 cssid = rcu_dereference_check(css->id,
4499 rcu_read_lock_held() || atomic_read(&css->refcnt));
4505 EXPORT_SYMBOL_GPL(css_id);
4507 unsigned short css_depth(struct cgroup_subsys_state *css)
4509 struct css_id *cssid;
4511 cssid = rcu_dereference_check(css->id,
4512 rcu_read_lock_held() || atomic_read(&css->refcnt));
4515 return cssid->depth;
4518 EXPORT_SYMBOL_GPL(css_depth);
4521 * css_is_ancestor - test "root" css is an ancestor of "child"
4522 * @child: the css to be tested.
4523 * @root: the css supporsed to be an ancestor of the child.
4525 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4526 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4527 * But, considering usual usage, the csses should be valid objects after test.
4528 * Assuming that the caller will do some action to the child if this returns
4529 * returns true, the caller must take "child";s reference count.
4530 * If "child" is valid object and this returns true, "root" is valid, too.
4533 bool css_is_ancestor(struct cgroup_subsys_state *child,
4534 const struct cgroup_subsys_state *root)
4536 struct css_id *child_id;
4537 struct css_id *root_id;
4541 child_id = rcu_dereference(child->id);
4542 root_id = rcu_dereference(root->id);
4545 || (child_id->depth < root_id->depth)
4546 || (child_id->stack[root_id->depth] != root_id->id))
4552 static void __free_css_id_cb(struct rcu_head *head)
4556 id = container_of(head, struct css_id, rcu_head);
4560 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
4562 struct css_id *id = css->id;
4563 /* When this is called before css_id initialization, id can be NULL */
4567 BUG_ON(!ss->use_id);
4569 rcu_assign_pointer(id->css, NULL);
4570 rcu_assign_pointer(css->id, NULL);
4571 spin_lock(&ss->id_lock);
4572 idr_remove(&ss->idr, id->id);
4573 spin_unlock(&ss->id_lock);
4574 call_rcu(&id->rcu_head, __free_css_id_cb);
4576 EXPORT_SYMBOL_GPL(free_css_id);
4579 * This is called by init or create(). Then, calls to this function are
4580 * always serialized (By cgroup_mutex() at create()).
4583 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
4585 struct css_id *newid;
4586 int myid, error, size;
4588 BUG_ON(!ss->use_id);
4590 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
4591 newid = kzalloc(size, GFP_KERNEL);
4593 return ERR_PTR(-ENOMEM);
4595 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
4599 spin_lock(&ss->id_lock);
4600 /* Don't use 0. allocates an ID of 1-65535 */
4601 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
4602 spin_unlock(&ss->id_lock);
4604 /* Returns error when there are no free spaces for new ID.*/
4609 if (myid > CSS_ID_MAX)
4613 newid->depth = depth;
4617 spin_lock(&ss->id_lock);
4618 idr_remove(&ss->idr, myid);
4619 spin_unlock(&ss->id_lock);
4622 return ERR_PTR(error);
4626 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
4627 struct cgroup_subsys_state *rootcss)
4629 struct css_id *newid;
4631 spin_lock_init(&ss->id_lock);
4634 newid = get_new_cssid(ss, 0);
4636 return PTR_ERR(newid);
4638 newid->stack[0] = newid->id;
4639 newid->css = rootcss;
4640 rootcss->id = newid;
4644 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
4645 struct cgroup *child)
4647 int subsys_id, i, depth = 0;
4648 struct cgroup_subsys_state *parent_css, *child_css;
4649 struct css_id *child_id, *parent_id;
4651 subsys_id = ss->subsys_id;
4652 parent_css = parent->subsys[subsys_id];
4653 child_css = child->subsys[subsys_id];
4654 parent_id = parent_css->id;
4655 depth = parent_id->depth + 1;
4657 child_id = get_new_cssid(ss, depth);
4658 if (IS_ERR(child_id))
4659 return PTR_ERR(child_id);
4661 for (i = 0; i < depth; i++)
4662 child_id->stack[i] = parent_id->stack[i];
4663 child_id->stack[depth] = child_id->id;
4665 * child_id->css pointer will be set after this cgroup is available
4666 * see cgroup_populate_dir()
4668 rcu_assign_pointer(child_css->id, child_id);
4674 * css_lookup - lookup css by id
4675 * @ss: cgroup subsys to be looked into.
4678 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4679 * NULL if not. Should be called under rcu_read_lock()
4681 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
4683 struct css_id *cssid = NULL;
4685 BUG_ON(!ss->use_id);
4686 cssid = idr_find(&ss->idr, id);
4688 if (unlikely(!cssid))
4691 return rcu_dereference(cssid->css);
4693 EXPORT_SYMBOL_GPL(css_lookup);
4696 * css_get_next - lookup next cgroup under specified hierarchy.
4697 * @ss: pointer to subsystem
4698 * @id: current position of iteration.
4699 * @root: pointer to css. search tree under this.
4700 * @foundid: position of found object.
4702 * Search next css under the specified hierarchy of rootid. Calling under
4703 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
4705 struct cgroup_subsys_state *
4706 css_get_next(struct cgroup_subsys *ss, int id,
4707 struct cgroup_subsys_state *root, int *foundid)
4709 struct cgroup_subsys_state *ret = NULL;
4712 int rootid = css_id(root);
4713 int depth = css_depth(root);
4718 BUG_ON(!ss->use_id);
4719 /* fill start point for scan */
4723 * scan next entry from bitmap(tree), tmpid is updated after
4726 spin_lock(&ss->id_lock);
4727 tmp = idr_get_next(&ss->idr, &tmpid);
4728 spin_unlock(&ss->id_lock);
4732 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
4733 ret = rcu_dereference(tmp->css);
4739 /* continue to scan from next id */
4745 #ifdef CONFIG_CGROUP_DEBUG
4746 static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
4747 struct cgroup *cont)
4749 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
4752 return ERR_PTR(-ENOMEM);
4757 static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
4759 kfree(cont->subsys[debug_subsys_id]);
4762 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
4764 return atomic_read(&cont->count);
4767 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
4769 return cgroup_task_count(cont);
4772 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
4774 return (u64)(unsigned long)current->cgroups;
4777 static u64 current_css_set_refcount_read(struct cgroup *cont,
4783 count = atomic_read(¤t->cgroups->refcount);
4788 static int current_css_set_cg_links_read(struct cgroup *cont,
4790 struct seq_file *seq)
4792 struct cg_cgroup_link *link;
4795 read_lock(&css_set_lock);
4797 cg = rcu_dereference(current->cgroups);
4798 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
4799 struct cgroup *c = link->cgrp;
4803 name = c->dentry->d_name.name;
4806 seq_printf(seq, "Root %d group %s\n",
4807 c->root->hierarchy_id, name);
4810 read_unlock(&css_set_lock);
4814 #define MAX_TASKS_SHOWN_PER_CSS 25
4815 static int cgroup_css_links_read(struct cgroup *cont,
4817 struct seq_file *seq)
4819 struct cg_cgroup_link *link;
4821 read_lock(&css_set_lock);
4822 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
4823 struct css_set *cg = link->cg;
4824 struct task_struct *task;
4826 seq_printf(seq, "css_set %p\n", cg);
4827 list_for_each_entry(task, &cg->tasks, cg_list) {
4828 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
4829 seq_puts(seq, " ...\n");
4832 seq_printf(seq, " task %d\n",
4833 task_pid_vnr(task));
4837 read_unlock(&css_set_lock);
4841 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
4843 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
4846 static struct cftype debug_files[] = {
4848 .name = "cgroup_refcount",
4849 .read_u64 = cgroup_refcount_read,
4852 .name = "taskcount",
4853 .read_u64 = debug_taskcount_read,
4857 .name = "current_css_set",
4858 .read_u64 = current_css_set_read,
4862 .name = "current_css_set_refcount",
4863 .read_u64 = current_css_set_refcount_read,
4867 .name = "current_css_set_cg_links",
4868 .read_seq_string = current_css_set_cg_links_read,
4872 .name = "cgroup_css_links",
4873 .read_seq_string = cgroup_css_links_read,
4877 .name = "releasable",
4878 .read_u64 = releasable_read,
4882 static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
4884 return cgroup_add_files(cont, ss, debug_files,
4885 ARRAY_SIZE(debug_files));
4888 struct cgroup_subsys debug_subsys = {
4890 .create = debug_create,
4891 .destroy = debug_destroy,
4892 .populate = debug_populate,
4893 .subsys_id = debug_subsys_id,
4895 #endif /* CONFIG_CGROUP_DEBUG */