2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hash.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_proc */
64 #include <linux/atomic.h>
67 * cgroup_mutex is the master lock. Any modification to cgroup or its
68 * hierarchy must be performed while holding it.
70 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
71 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
72 * release_agent_path and so on. Modifying requires both cgroup_mutex and
73 * cgroup_root_mutex. Readers can acquire either of the two. This is to
74 * break the following locking order cycle.
76 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
77 * B. namespace_sem -> cgroup_mutex
79 * B happens only through cgroup_show_options() and using cgroup_root_mutex
82 static DEFINE_MUTEX(cgroup_mutex);
83 static DEFINE_MUTEX(cgroup_root_mutex);
86 * Generate an array of cgroup subsystem pointers. At boot time, this is
87 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
88 * registered after that. The mutable section of this array is protected by
91 #define SUBSYS(_x) &_x ## _subsys,
92 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
93 #include <linux/cgroup_subsys.h>
96 #define MAX_CGROUP_ROOT_NAMELEN 64
99 * A cgroupfs_root represents the root of a cgroup hierarchy,
100 * and may be associated with a superblock to form an active
103 struct cgroupfs_root {
104 struct super_block *sb;
107 * The bitmask of subsystems intended to be attached to this
110 unsigned long subsys_bits;
112 /* Unique id for this hierarchy. */
115 /* The bitmask of subsystems currently attached to this hierarchy */
116 unsigned long actual_subsys_bits;
118 /* A list running through the attached subsystems */
119 struct list_head subsys_list;
121 /* The root cgroup for this hierarchy */
122 struct cgroup top_cgroup;
124 /* Tracks how many cgroups are currently defined in hierarchy.*/
125 int number_of_cgroups;
127 /* A list running through the active hierarchies */
128 struct list_head root_list;
130 /* Hierarchy-specific flags */
133 /* The path to use for release notifications. */
134 char release_agent_path[PATH_MAX];
136 /* The name for this hierarchy - may be empty */
137 char name[MAX_CGROUP_ROOT_NAMELEN];
141 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
142 * subsystems that are otherwise unattached - it never has more than a
143 * single cgroup, and all tasks are part of that cgroup.
145 static struct cgroupfs_root rootnode;
148 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
149 * cgroup_subsys->use_id != 0.
151 #define CSS_ID_MAX (65535)
154 * The css to which this ID points. This pointer is set to valid value
155 * after cgroup is populated. If cgroup is removed, this will be NULL.
156 * This pointer is expected to be RCU-safe because destroy()
157 * is called after synchronize_rcu(). But for safe use, css_is_removed()
158 * css_tryget() should be used for avoiding race.
160 struct cgroup_subsys_state __rcu *css;
166 * Depth in hierarchy which this ID belongs to.
168 unsigned short depth;
170 * ID is freed by RCU. (and lookup routine is RCU safe.)
172 struct rcu_head rcu_head;
174 * Hierarchy of CSS ID belongs to.
176 unsigned short stack[0]; /* Array of Length (depth+1) */
180 * cgroup_event represents events which userspace want to receive.
182 struct cgroup_event {
184 * Cgroup which the event belongs to.
188 * Control file which the event associated.
192 * eventfd to signal userspace about the event.
194 struct eventfd_ctx *eventfd;
196 * Each of these stored in a list by the cgroup.
198 struct list_head list;
200 * All fields below needed to unregister event when
201 * userspace closes eventfd.
204 wait_queue_head_t *wqh;
206 struct work_struct remove;
209 /* The list of hierarchy roots */
211 static LIST_HEAD(roots);
212 static int root_count;
214 static DEFINE_IDA(hierarchy_ida);
215 static int next_hierarchy_id;
216 static DEFINE_SPINLOCK(hierarchy_id_lock);
218 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
219 #define dummytop (&rootnode.top_cgroup)
221 /* This flag indicates whether tasks in the fork and exit paths should
222 * check for fork/exit handlers to call. This avoids us having to do
223 * extra work in the fork/exit path if none of the subsystems need to
226 static int need_forkexit_callback __read_mostly;
228 #ifdef CONFIG_PROVE_LOCKING
229 int cgroup_lock_is_held(void)
231 return lockdep_is_held(&cgroup_mutex);
233 #else /* #ifdef CONFIG_PROVE_LOCKING */
234 int cgroup_lock_is_held(void)
236 return mutex_is_locked(&cgroup_mutex);
238 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
240 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
242 /* convenient tests for these bits */
243 inline int cgroup_is_removed(const struct cgroup *cgrp)
245 return test_bit(CGRP_REMOVED, &cgrp->flags);
248 /* bits in struct cgroupfs_root flags field */
250 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
253 static int cgroup_is_releasable(const struct cgroup *cgrp)
256 (1 << CGRP_RELEASABLE) |
257 (1 << CGRP_NOTIFY_ON_RELEASE);
258 return (cgrp->flags & bits) == bits;
261 static int notify_on_release(const struct cgroup *cgrp)
263 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
266 static int clone_children(const struct cgroup *cgrp)
268 return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
272 * for_each_subsys() allows you to iterate on each subsystem attached to
273 * an active hierarchy
275 #define for_each_subsys(_root, _ss) \
276 list_for_each_entry(_ss, &_root->subsys_list, sibling)
278 /* for_each_active_root() allows you to iterate across the active hierarchies */
279 #define for_each_active_root(_root) \
280 list_for_each_entry(_root, &roots, root_list)
282 /* the list of cgroups eligible for automatic release. Protected by
283 * release_list_lock */
284 static LIST_HEAD(release_list);
285 static DEFINE_RAW_SPINLOCK(release_list_lock);
286 static void cgroup_release_agent(struct work_struct *work);
287 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
288 static void check_for_release(struct cgroup *cgrp);
290 /* Link structure for associating css_set objects with cgroups */
291 struct cg_cgroup_link {
293 * List running through cg_cgroup_links associated with a
294 * cgroup, anchored on cgroup->css_sets
296 struct list_head cgrp_link_list;
299 * List running through cg_cgroup_links pointing at a
300 * single css_set object, anchored on css_set->cg_links
302 struct list_head cg_link_list;
306 /* The default css_set - used by init and its children prior to any
307 * hierarchies being mounted. It contains a pointer to the root state
308 * for each subsystem. Also used to anchor the list of css_sets. Not
309 * reference-counted, to improve performance when child cgroups
310 * haven't been created.
313 static struct css_set init_css_set;
314 static struct cg_cgroup_link init_css_set_link;
316 static int cgroup_init_idr(struct cgroup_subsys *ss,
317 struct cgroup_subsys_state *css);
319 /* css_set_lock protects the list of css_set objects, and the
320 * chain of tasks off each css_set. Nests outside task->alloc_lock
321 * due to cgroup_iter_start() */
322 static DEFINE_RWLOCK(css_set_lock);
323 static int css_set_count;
326 * hash table for cgroup groups. This improves the performance to find
327 * an existing css_set. This hash doesn't (currently) take into
328 * account cgroups in empty hierarchies.
330 #define CSS_SET_HASH_BITS 7
331 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
332 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
334 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
338 unsigned long tmp = 0UL;
340 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
341 tmp += (unsigned long)css[i];
342 tmp = (tmp >> 16) ^ tmp;
344 index = hash_long(tmp, CSS_SET_HASH_BITS);
346 return &css_set_table[index];
349 /* We don't maintain the lists running through each css_set to its
350 * task until after the first call to cgroup_iter_start(). This
351 * reduces the fork()/exit() overhead for people who have cgroups
352 * compiled into their kernel but not actually in use */
353 static int use_task_css_set_links __read_mostly;
355 static void __put_css_set(struct css_set *cg, int taskexit)
357 struct cg_cgroup_link *link;
358 struct cg_cgroup_link *saved_link;
360 * Ensure that the refcount doesn't hit zero while any readers
361 * can see it. Similar to atomic_dec_and_lock(), but for an
364 if (atomic_add_unless(&cg->refcount, -1, 1))
366 write_lock(&css_set_lock);
367 if (!atomic_dec_and_test(&cg->refcount)) {
368 write_unlock(&css_set_lock);
372 /* This css_set is dead. unlink it and release cgroup refcounts */
373 hlist_del(&cg->hlist);
376 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
378 struct cgroup *cgrp = link->cgrp;
379 list_del(&link->cg_link_list);
380 list_del(&link->cgrp_link_list);
383 * We may not be holding cgroup_mutex, and if cgrp->count is
384 * dropped to 0 the cgroup can be destroyed at any time, hence
385 * rcu_read_lock is used to keep it alive.
388 if (atomic_dec_and_test(&cgrp->count) &&
389 notify_on_release(cgrp)) {
391 set_bit(CGRP_RELEASABLE, &cgrp->flags);
392 check_for_release(cgrp);
399 write_unlock(&css_set_lock);
400 kfree_rcu(cg, rcu_head);
404 * refcounted get/put for css_set objects
406 static inline void get_css_set(struct css_set *cg)
408 atomic_inc(&cg->refcount);
411 static inline void put_css_set(struct css_set *cg)
413 __put_css_set(cg, 0);
416 static inline void put_css_set_taskexit(struct css_set *cg)
418 __put_css_set(cg, 1);
422 * compare_css_sets - helper function for find_existing_css_set().
423 * @cg: candidate css_set being tested
424 * @old_cg: existing css_set for a task
425 * @new_cgrp: cgroup that's being entered by the task
426 * @template: desired set of css pointers in css_set (pre-calculated)
428 * Returns true if "cg" matches "old_cg" except for the hierarchy
429 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
431 static bool compare_css_sets(struct css_set *cg,
432 struct css_set *old_cg,
433 struct cgroup *new_cgrp,
434 struct cgroup_subsys_state *template[])
436 struct list_head *l1, *l2;
438 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
439 /* Not all subsystems matched */
444 * Compare cgroup pointers in order to distinguish between
445 * different cgroups in heirarchies with no subsystems. We
446 * could get by with just this check alone (and skip the
447 * memcmp above) but on most setups the memcmp check will
448 * avoid the need for this more expensive check on almost all
453 l2 = &old_cg->cg_links;
455 struct cg_cgroup_link *cgl1, *cgl2;
456 struct cgroup *cg1, *cg2;
460 /* See if we reached the end - both lists are equal length. */
461 if (l1 == &cg->cg_links) {
462 BUG_ON(l2 != &old_cg->cg_links);
465 BUG_ON(l2 == &old_cg->cg_links);
467 /* Locate the cgroups associated with these links. */
468 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
469 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
472 /* Hierarchies should be linked in the same order. */
473 BUG_ON(cg1->root != cg2->root);
476 * If this hierarchy is the hierarchy of the cgroup
477 * that's changing, then we need to check that this
478 * css_set points to the new cgroup; if it's any other
479 * hierarchy, then this css_set should point to the
480 * same cgroup as the old css_set.
482 if (cg1->root == new_cgrp->root) {
494 * find_existing_css_set() is a helper for
495 * find_css_set(), and checks to see whether an existing
496 * css_set is suitable.
498 * oldcg: the cgroup group that we're using before the cgroup
501 * cgrp: the cgroup that we're moving into
503 * template: location in which to build the desired set of subsystem
504 * state objects for the new cgroup group
506 static struct css_set *find_existing_css_set(
507 struct css_set *oldcg,
509 struct cgroup_subsys_state *template[])
512 struct cgroupfs_root *root = cgrp->root;
513 struct hlist_head *hhead;
514 struct hlist_node *node;
518 * Build the set of subsystem state objects that we want to see in the
519 * new css_set. while subsystems can change globally, the entries here
520 * won't change, so no need for locking.
522 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
523 if (root->subsys_bits & (1UL << i)) {
524 /* Subsystem is in this hierarchy. So we want
525 * the subsystem state from the new
527 template[i] = cgrp->subsys[i];
529 /* Subsystem is not in this hierarchy, so we
530 * don't want to change the subsystem state */
531 template[i] = oldcg->subsys[i];
535 hhead = css_set_hash(template);
536 hlist_for_each_entry(cg, node, hhead, hlist) {
537 if (!compare_css_sets(cg, oldcg, cgrp, template))
540 /* This css_set matches what we need */
544 /* No existing cgroup group matched */
548 static void free_cg_links(struct list_head *tmp)
550 struct cg_cgroup_link *link;
551 struct cg_cgroup_link *saved_link;
553 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
554 list_del(&link->cgrp_link_list);
560 * allocate_cg_links() allocates "count" cg_cgroup_link structures
561 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
562 * success or a negative error
564 static int allocate_cg_links(int count, struct list_head *tmp)
566 struct cg_cgroup_link *link;
569 for (i = 0; i < count; i++) {
570 link = kmalloc(sizeof(*link), GFP_KERNEL);
575 list_add(&link->cgrp_link_list, tmp);
581 * link_css_set - a helper function to link a css_set to a cgroup
582 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
583 * @cg: the css_set to be linked
584 * @cgrp: the destination cgroup
586 static void link_css_set(struct list_head *tmp_cg_links,
587 struct css_set *cg, struct cgroup *cgrp)
589 struct cg_cgroup_link *link;
591 BUG_ON(list_empty(tmp_cg_links));
592 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
596 atomic_inc(&cgrp->count);
597 list_move(&link->cgrp_link_list, &cgrp->css_sets);
599 * Always add links to the tail of the list so that the list
600 * is sorted by order of hierarchy creation
602 list_add_tail(&link->cg_link_list, &cg->cg_links);
606 * find_css_set() takes an existing cgroup group and a
607 * cgroup object, and returns a css_set object that's
608 * equivalent to the old group, but with the given cgroup
609 * substituted into the appropriate hierarchy. Must be called with
612 static struct css_set *find_css_set(
613 struct css_set *oldcg, struct cgroup *cgrp)
616 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
618 struct list_head tmp_cg_links;
620 struct hlist_head *hhead;
621 struct cg_cgroup_link *link;
623 /* First see if we already have a cgroup group that matches
625 read_lock(&css_set_lock);
626 res = find_existing_css_set(oldcg, cgrp, template);
629 read_unlock(&css_set_lock);
634 res = kmalloc(sizeof(*res), GFP_KERNEL);
638 /* Allocate all the cg_cgroup_link objects that we'll need */
639 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
644 atomic_set(&res->refcount, 1);
645 INIT_LIST_HEAD(&res->cg_links);
646 INIT_LIST_HEAD(&res->tasks);
647 INIT_HLIST_NODE(&res->hlist);
649 /* Copy the set of subsystem state objects generated in
650 * find_existing_css_set() */
651 memcpy(res->subsys, template, sizeof(res->subsys));
653 write_lock(&css_set_lock);
654 /* Add reference counts and links from the new css_set. */
655 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
656 struct cgroup *c = link->cgrp;
657 if (c->root == cgrp->root)
659 link_css_set(&tmp_cg_links, res, c);
662 BUG_ON(!list_empty(&tmp_cg_links));
666 /* Add this cgroup group to the hash table */
667 hhead = css_set_hash(res->subsys);
668 hlist_add_head(&res->hlist, hhead);
670 write_unlock(&css_set_lock);
676 * Return the cgroup for "task" from the given hierarchy. Must be
677 * called with cgroup_mutex held.
679 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
680 struct cgroupfs_root *root)
683 struct cgroup *res = NULL;
685 BUG_ON(!mutex_is_locked(&cgroup_mutex));
686 read_lock(&css_set_lock);
688 * No need to lock the task - since we hold cgroup_mutex the
689 * task can't change groups, so the only thing that can happen
690 * is that it exits and its css is set back to init_css_set.
693 if (css == &init_css_set) {
694 res = &root->top_cgroup;
696 struct cg_cgroup_link *link;
697 list_for_each_entry(link, &css->cg_links, cg_link_list) {
698 struct cgroup *c = link->cgrp;
699 if (c->root == root) {
705 read_unlock(&css_set_lock);
711 * There is one global cgroup mutex. We also require taking
712 * task_lock() when dereferencing a task's cgroup subsys pointers.
713 * See "The task_lock() exception", at the end of this comment.
715 * A task must hold cgroup_mutex to modify cgroups.
717 * Any task can increment and decrement the count field without lock.
718 * So in general, code holding cgroup_mutex can't rely on the count
719 * field not changing. However, if the count goes to zero, then only
720 * cgroup_attach_task() can increment it again. Because a count of zero
721 * means that no tasks are currently attached, therefore there is no
722 * way a task attached to that cgroup can fork (the other way to
723 * increment the count). So code holding cgroup_mutex can safely
724 * assume that if the count is zero, it will stay zero. Similarly, if
725 * a task holds cgroup_mutex on a cgroup with zero count, it
726 * knows that the cgroup won't be removed, as cgroup_rmdir()
729 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
730 * (usually) take cgroup_mutex. These are the two most performance
731 * critical pieces of code here. The exception occurs on cgroup_exit(),
732 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
733 * is taken, and if the cgroup count is zero, a usermode call made
734 * to the release agent with the name of the cgroup (path relative to
735 * the root of cgroup file system) as the argument.
737 * A cgroup can only be deleted if both its 'count' of using tasks
738 * is zero, and its list of 'children' cgroups is empty. Since all
739 * tasks in the system use _some_ cgroup, and since there is always at
740 * least one task in the system (init, pid == 1), therefore, top_cgroup
741 * always has either children cgroups and/or using tasks. So we don't
742 * need a special hack to ensure that top_cgroup cannot be deleted.
744 * The task_lock() exception
746 * The need for this exception arises from the action of
747 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
748 * another. It does so using cgroup_mutex, however there are
749 * several performance critical places that need to reference
750 * task->cgroup without the expense of grabbing a system global
751 * mutex. Therefore except as noted below, when dereferencing or, as
752 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
753 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
754 * the task_struct routinely used for such matters.
756 * P.S. One more locking exception. RCU is used to guard the
757 * update of a tasks cgroup pointer by cgroup_attach_task()
761 * cgroup_lock - lock out any changes to cgroup structures
764 void cgroup_lock(void)
766 mutex_lock(&cgroup_mutex);
768 EXPORT_SYMBOL_GPL(cgroup_lock);
771 * cgroup_unlock - release lock on cgroup changes
773 * Undo the lock taken in a previous cgroup_lock() call.
775 void cgroup_unlock(void)
777 mutex_unlock(&cgroup_mutex);
779 EXPORT_SYMBOL_GPL(cgroup_unlock);
782 * A couple of forward declarations required, due to cyclic reference loop:
783 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
784 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
788 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
789 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, struct nameidata *);
790 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
791 static int cgroup_populate_dir(struct cgroup *cgrp);
792 static const struct inode_operations cgroup_dir_inode_operations;
793 static const struct file_operations proc_cgroupstats_operations;
795 static struct backing_dev_info cgroup_backing_dev_info = {
797 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
800 static int alloc_css_id(struct cgroup_subsys *ss,
801 struct cgroup *parent, struct cgroup *child);
803 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
805 struct inode *inode = new_inode(sb);
808 inode->i_ino = get_next_ino();
809 inode->i_mode = mode;
810 inode->i_uid = current_fsuid();
811 inode->i_gid = current_fsgid();
812 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
813 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
819 * Call subsys's pre_destroy handler.
820 * This is called before css refcnt check.
822 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
824 struct cgroup_subsys *ss;
827 for_each_subsys(cgrp->root, ss)
828 if (ss->pre_destroy) {
829 ret = ss->pre_destroy(cgrp);
837 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
839 /* is dentry a directory ? if so, kfree() associated cgroup */
840 if (S_ISDIR(inode->i_mode)) {
841 struct cgroup *cgrp = dentry->d_fsdata;
842 struct cgroup_subsys *ss;
843 BUG_ON(!(cgroup_is_removed(cgrp)));
844 /* It's possible for external users to be holding css
845 * reference counts on a cgroup; css_put() needs to
846 * be able to access the cgroup after decrementing
847 * the reference count in order to know if it needs to
848 * queue the cgroup to be handled by the release
852 mutex_lock(&cgroup_mutex);
854 * Release the subsystem state objects.
856 for_each_subsys(cgrp->root, ss)
859 cgrp->root->number_of_cgroups--;
860 mutex_unlock(&cgroup_mutex);
863 * Drop the active superblock reference that we took when we
866 deactivate_super(cgrp->root->sb);
869 * if we're getting rid of the cgroup, refcount should ensure
870 * that there are no pidlists left.
872 BUG_ON(!list_empty(&cgrp->pidlists));
874 kfree_rcu(cgrp, rcu_head);
879 static int cgroup_delete(const struct dentry *d)
884 static void remove_dir(struct dentry *d)
886 struct dentry *parent = dget(d->d_parent);
889 simple_rmdir(parent->d_inode, d);
893 static void cgroup_clear_directory(struct dentry *dentry)
895 struct list_head *node;
897 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
898 spin_lock(&dentry->d_lock);
899 node = dentry->d_subdirs.next;
900 while (node != &dentry->d_subdirs) {
901 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
903 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
906 /* This should never be called on a cgroup
907 * directory with child cgroups */
908 BUG_ON(d->d_inode->i_mode & S_IFDIR);
910 spin_unlock(&d->d_lock);
911 spin_unlock(&dentry->d_lock);
913 simple_unlink(dentry->d_inode, d);
915 spin_lock(&dentry->d_lock);
917 spin_unlock(&d->d_lock);
918 node = dentry->d_subdirs.next;
920 spin_unlock(&dentry->d_lock);
924 * NOTE : the dentry must have been dget()'ed
926 static void cgroup_d_remove_dir(struct dentry *dentry)
928 struct dentry *parent;
930 cgroup_clear_directory(dentry);
932 parent = dentry->d_parent;
933 spin_lock(&parent->d_lock);
934 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
935 list_del_init(&dentry->d_u.d_child);
936 spin_unlock(&dentry->d_lock);
937 spin_unlock(&parent->d_lock);
942 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
943 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
944 * reference to css->refcnt. In general, this refcnt is expected to goes down
947 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
949 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
951 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
953 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
954 wake_up_all(&cgroup_rmdir_waitq);
957 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
962 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
964 cgroup_wakeup_rmdir_waiter(css->cgroup);
969 * Call with cgroup_mutex held. Drops reference counts on modules, including
970 * any duplicate ones that parse_cgroupfs_options took. If this function
971 * returns an error, no reference counts are touched.
973 static int rebind_subsystems(struct cgroupfs_root *root,
974 unsigned long final_bits)
976 unsigned long added_bits, removed_bits;
977 struct cgroup *cgrp = &root->top_cgroup;
980 BUG_ON(!mutex_is_locked(&cgroup_mutex));
981 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
983 removed_bits = root->actual_subsys_bits & ~final_bits;
984 added_bits = final_bits & ~root->actual_subsys_bits;
985 /* Check that any added subsystems are currently free */
986 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
987 unsigned long bit = 1UL << i;
988 struct cgroup_subsys *ss = subsys[i];
989 if (!(bit & added_bits))
992 * Nobody should tell us to do a subsys that doesn't exist:
993 * parse_cgroupfs_options should catch that case and refcounts
994 * ensure that subsystems won't disappear once selected.
997 if (ss->root != &rootnode) {
998 /* Subsystem isn't free */
1003 /* Currently we don't handle adding/removing subsystems when
1004 * any child cgroups exist. This is theoretically supportable
1005 * but involves complex error handling, so it's being left until
1007 if (root->number_of_cgroups > 1)
1010 /* Process each subsystem */
1011 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1012 struct cgroup_subsys *ss = subsys[i];
1013 unsigned long bit = 1UL << i;
1014 if (bit & added_bits) {
1015 /* We're binding this subsystem to this hierarchy */
1017 BUG_ON(cgrp->subsys[i]);
1018 BUG_ON(!dummytop->subsys[i]);
1019 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1020 mutex_lock(&ss->hierarchy_mutex);
1021 cgrp->subsys[i] = dummytop->subsys[i];
1022 cgrp->subsys[i]->cgroup = cgrp;
1023 list_move(&ss->sibling, &root->subsys_list);
1027 mutex_unlock(&ss->hierarchy_mutex);
1028 /* refcount was already taken, and we're keeping it */
1029 } else if (bit & removed_bits) {
1030 /* We're removing this subsystem */
1032 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1033 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1034 mutex_lock(&ss->hierarchy_mutex);
1037 dummytop->subsys[i]->cgroup = dummytop;
1038 cgrp->subsys[i] = NULL;
1039 subsys[i]->root = &rootnode;
1040 list_move(&ss->sibling, &rootnode.subsys_list);
1041 mutex_unlock(&ss->hierarchy_mutex);
1042 /* subsystem is now free - drop reference on module */
1043 module_put(ss->module);
1044 } else if (bit & final_bits) {
1045 /* Subsystem state should already exist */
1047 BUG_ON(!cgrp->subsys[i]);
1049 * a refcount was taken, but we already had one, so
1050 * drop the extra reference.
1052 module_put(ss->module);
1053 #ifdef CONFIG_MODULE_UNLOAD
1054 BUG_ON(ss->module && !module_refcount(ss->module));
1057 /* Subsystem state shouldn't exist */
1058 BUG_ON(cgrp->subsys[i]);
1061 root->subsys_bits = root->actual_subsys_bits = final_bits;
1067 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1069 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1070 struct cgroup_subsys *ss;
1072 mutex_lock(&cgroup_root_mutex);
1073 for_each_subsys(root, ss)
1074 seq_printf(seq, ",%s", ss->name);
1075 if (test_bit(ROOT_NOPREFIX, &root->flags))
1076 seq_puts(seq, ",noprefix");
1077 if (strlen(root->release_agent_path))
1078 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1079 if (clone_children(&root->top_cgroup))
1080 seq_puts(seq, ",clone_children");
1081 if (strlen(root->name))
1082 seq_printf(seq, ",name=%s", root->name);
1083 mutex_unlock(&cgroup_root_mutex);
1087 struct cgroup_sb_opts {
1088 unsigned long subsys_bits;
1089 unsigned long flags;
1090 char *release_agent;
1091 bool clone_children;
1093 /* User explicitly requested empty subsystem */
1096 struct cgroupfs_root *new_root;
1101 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1102 * with cgroup_mutex held to protect the subsys[] array. This function takes
1103 * refcounts on subsystems to be used, unless it returns error, in which case
1104 * no refcounts are taken.
1106 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1108 char *token, *o = data;
1109 bool all_ss = false, one_ss = false;
1110 unsigned long mask = (unsigned long)-1;
1112 bool module_pin_failed = false;
1114 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1116 #ifdef CONFIG_CPUSETS
1117 mask = ~(1UL << cpuset_subsys_id);
1120 memset(opts, 0, sizeof(*opts));
1122 while ((token = strsep(&o, ",")) != NULL) {
1125 if (!strcmp(token, "none")) {
1126 /* Explicitly have no subsystems */
1130 if (!strcmp(token, "all")) {
1131 /* Mutually exclusive option 'all' + subsystem name */
1137 if (!strcmp(token, "noprefix")) {
1138 set_bit(ROOT_NOPREFIX, &opts->flags);
1141 if (!strcmp(token, "clone_children")) {
1142 opts->clone_children = true;
1145 if (!strncmp(token, "release_agent=", 14)) {
1146 /* Specifying two release agents is forbidden */
1147 if (opts->release_agent)
1149 opts->release_agent =
1150 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1151 if (!opts->release_agent)
1155 if (!strncmp(token, "name=", 5)) {
1156 const char *name = token + 5;
1157 /* Can't specify an empty name */
1160 /* Must match [\w.-]+ */
1161 for (i = 0; i < strlen(name); i++) {
1165 if ((c == '.') || (c == '-') || (c == '_'))
1169 /* Specifying two names is forbidden */
1172 opts->name = kstrndup(name,
1173 MAX_CGROUP_ROOT_NAMELEN - 1,
1181 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1182 struct cgroup_subsys *ss = subsys[i];
1185 if (strcmp(token, ss->name))
1190 /* Mutually exclusive option 'all' + subsystem name */
1193 set_bit(i, &opts->subsys_bits);
1198 if (i == CGROUP_SUBSYS_COUNT)
1203 * If the 'all' option was specified select all the subsystems,
1204 * otherwise if 'none', 'name=' and a subsystem name options
1205 * were not specified, let's default to 'all'
1207 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1208 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1209 struct cgroup_subsys *ss = subsys[i];
1214 set_bit(i, &opts->subsys_bits);
1218 /* Consistency checks */
1221 * Option noprefix was introduced just for backward compatibility
1222 * with the old cpuset, so we allow noprefix only if mounting just
1223 * the cpuset subsystem.
1225 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1226 (opts->subsys_bits & mask))
1230 /* Can't specify "none" and some subsystems */
1231 if (opts->subsys_bits && opts->none)
1235 * We either have to specify by name or by subsystems. (So all
1236 * empty hierarchies must have a name).
1238 if (!opts->subsys_bits && !opts->name)
1242 * Grab references on all the modules we'll need, so the subsystems
1243 * don't dance around before rebind_subsystems attaches them. This may
1244 * take duplicate reference counts on a subsystem that's already used,
1245 * but rebind_subsystems handles this case.
1247 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1248 unsigned long bit = 1UL << i;
1250 if (!(bit & opts->subsys_bits))
1252 if (!try_module_get(subsys[i]->module)) {
1253 module_pin_failed = true;
1257 if (module_pin_failed) {
1259 * oops, one of the modules was going away. this means that we
1260 * raced with a module_delete call, and to the user this is
1261 * essentially a "subsystem doesn't exist" case.
1263 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1264 /* drop refcounts only on the ones we took */
1265 unsigned long bit = 1UL << i;
1267 if (!(bit & opts->subsys_bits))
1269 module_put(subsys[i]->module);
1277 static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1280 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1281 unsigned long bit = 1UL << i;
1283 if (!(bit & subsys_bits))
1285 module_put(subsys[i]->module);
1289 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1292 struct cgroupfs_root *root = sb->s_fs_info;
1293 struct cgroup *cgrp = &root->top_cgroup;
1294 struct cgroup_sb_opts opts;
1296 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1297 mutex_lock(&cgroup_mutex);
1298 mutex_lock(&cgroup_root_mutex);
1300 /* See what subsystems are wanted */
1301 ret = parse_cgroupfs_options(data, &opts);
1305 /* Don't allow flags or name to change at remount */
1306 if (opts.flags != root->flags ||
1307 (opts.name && strcmp(opts.name, root->name))) {
1309 drop_parsed_module_refcounts(opts.subsys_bits);
1313 ret = rebind_subsystems(root, opts.subsys_bits);
1315 drop_parsed_module_refcounts(opts.subsys_bits);
1319 /* (re)populate subsystem files */
1320 cgroup_populate_dir(cgrp);
1322 if (opts.release_agent)
1323 strcpy(root->release_agent_path, opts.release_agent);
1325 kfree(opts.release_agent);
1327 mutex_unlock(&cgroup_root_mutex);
1328 mutex_unlock(&cgroup_mutex);
1329 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1333 static const struct super_operations cgroup_ops = {
1334 .statfs = simple_statfs,
1335 .drop_inode = generic_delete_inode,
1336 .show_options = cgroup_show_options,
1337 .remount_fs = cgroup_remount,
1340 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1342 INIT_LIST_HEAD(&cgrp->sibling);
1343 INIT_LIST_HEAD(&cgrp->children);
1344 INIT_LIST_HEAD(&cgrp->css_sets);
1345 INIT_LIST_HEAD(&cgrp->release_list);
1346 INIT_LIST_HEAD(&cgrp->pidlists);
1347 mutex_init(&cgrp->pidlist_mutex);
1348 INIT_LIST_HEAD(&cgrp->event_list);
1349 spin_lock_init(&cgrp->event_list_lock);
1352 static void init_cgroup_root(struct cgroupfs_root *root)
1354 struct cgroup *cgrp = &root->top_cgroup;
1355 INIT_LIST_HEAD(&root->subsys_list);
1356 INIT_LIST_HEAD(&root->root_list);
1357 root->number_of_cgroups = 1;
1359 cgrp->top_cgroup = cgrp;
1360 init_cgroup_housekeeping(cgrp);
1363 static bool init_root_id(struct cgroupfs_root *root)
1368 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1370 spin_lock(&hierarchy_id_lock);
1371 /* Try to allocate the next unused ID */
1372 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1373 &root->hierarchy_id);
1375 /* Try again starting from 0 */
1376 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1378 next_hierarchy_id = root->hierarchy_id + 1;
1379 } else if (ret != -EAGAIN) {
1380 /* Can only get here if the 31-bit IDR is full ... */
1383 spin_unlock(&hierarchy_id_lock);
1388 static int cgroup_test_super(struct super_block *sb, void *data)
1390 struct cgroup_sb_opts *opts = data;
1391 struct cgroupfs_root *root = sb->s_fs_info;
1393 /* If we asked for a name then it must match */
1394 if (opts->name && strcmp(opts->name, root->name))
1398 * If we asked for subsystems (or explicitly for no
1399 * subsystems) then they must match
1401 if ((opts->subsys_bits || opts->none)
1402 && (opts->subsys_bits != root->subsys_bits))
1408 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1410 struct cgroupfs_root *root;
1412 if (!opts->subsys_bits && !opts->none)
1415 root = kzalloc(sizeof(*root), GFP_KERNEL);
1417 return ERR_PTR(-ENOMEM);
1419 if (!init_root_id(root)) {
1421 return ERR_PTR(-ENOMEM);
1423 init_cgroup_root(root);
1425 root->subsys_bits = opts->subsys_bits;
1426 root->flags = opts->flags;
1427 if (opts->release_agent)
1428 strcpy(root->release_agent_path, opts->release_agent);
1430 strcpy(root->name, opts->name);
1431 if (opts->clone_children)
1432 set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1436 static void cgroup_drop_root(struct cgroupfs_root *root)
1441 BUG_ON(!root->hierarchy_id);
1442 spin_lock(&hierarchy_id_lock);
1443 ida_remove(&hierarchy_ida, root->hierarchy_id);
1444 spin_unlock(&hierarchy_id_lock);
1448 static int cgroup_set_super(struct super_block *sb, void *data)
1451 struct cgroup_sb_opts *opts = data;
1453 /* If we don't have a new root, we can't set up a new sb */
1454 if (!opts->new_root)
1457 BUG_ON(!opts->subsys_bits && !opts->none);
1459 ret = set_anon_super(sb, NULL);
1463 sb->s_fs_info = opts->new_root;
1464 opts->new_root->sb = sb;
1466 sb->s_blocksize = PAGE_CACHE_SIZE;
1467 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1468 sb->s_magic = CGROUP_SUPER_MAGIC;
1469 sb->s_op = &cgroup_ops;
1474 static int cgroup_get_rootdir(struct super_block *sb)
1476 static const struct dentry_operations cgroup_dops = {
1477 .d_iput = cgroup_diput,
1478 .d_delete = cgroup_delete,
1481 struct inode *inode =
1482 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1487 inode->i_fop = &simple_dir_operations;
1488 inode->i_op = &cgroup_dir_inode_operations;
1489 /* directories start off with i_nlink == 2 (for "." entry) */
1491 sb->s_root = d_make_root(inode);
1494 /* for everything else we want ->d_op set */
1495 sb->s_d_op = &cgroup_dops;
1499 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1500 int flags, const char *unused_dev_name,
1503 struct cgroup_sb_opts opts;
1504 struct cgroupfs_root *root;
1506 struct super_block *sb;
1507 struct cgroupfs_root *new_root;
1508 struct inode *inode;
1510 /* First find the desired set of subsystems */
1511 mutex_lock(&cgroup_mutex);
1512 ret = parse_cgroupfs_options(data, &opts);
1513 mutex_unlock(&cgroup_mutex);
1518 * Allocate a new cgroup root. We may not need it if we're
1519 * reusing an existing hierarchy.
1521 new_root = cgroup_root_from_opts(&opts);
1522 if (IS_ERR(new_root)) {
1523 ret = PTR_ERR(new_root);
1526 opts.new_root = new_root;
1528 /* Locate an existing or new sb for this hierarchy */
1529 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1532 cgroup_drop_root(opts.new_root);
1536 root = sb->s_fs_info;
1538 if (root == opts.new_root) {
1539 /* We used the new root structure, so this is a new hierarchy */
1540 struct list_head tmp_cg_links;
1541 struct cgroup *root_cgrp = &root->top_cgroup;
1542 struct cgroupfs_root *existing_root;
1543 const struct cred *cred;
1546 BUG_ON(sb->s_root != NULL);
1548 ret = cgroup_get_rootdir(sb);
1550 goto drop_new_super;
1551 inode = sb->s_root->d_inode;
1553 mutex_lock(&inode->i_mutex);
1554 mutex_lock(&cgroup_mutex);
1555 mutex_lock(&cgroup_root_mutex);
1557 /* Check for name clashes with existing mounts */
1559 if (strlen(root->name))
1560 for_each_active_root(existing_root)
1561 if (!strcmp(existing_root->name, root->name))
1565 * We're accessing css_set_count without locking
1566 * css_set_lock here, but that's OK - it can only be
1567 * increased by someone holding cgroup_lock, and
1568 * that's us. The worst that can happen is that we
1569 * have some link structures left over
1571 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1575 ret = rebind_subsystems(root, root->subsys_bits);
1576 if (ret == -EBUSY) {
1577 free_cg_links(&tmp_cg_links);
1581 * There must be no failure case after here, since rebinding
1582 * takes care of subsystems' refcounts, which are explicitly
1583 * dropped in the failure exit path.
1586 /* EBUSY should be the only error here */
1589 list_add(&root->root_list, &roots);
1592 sb->s_root->d_fsdata = root_cgrp;
1593 root->top_cgroup.dentry = sb->s_root;
1595 /* Link the top cgroup in this hierarchy into all
1596 * the css_set objects */
1597 write_lock(&css_set_lock);
1598 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1599 struct hlist_head *hhead = &css_set_table[i];
1600 struct hlist_node *node;
1603 hlist_for_each_entry(cg, node, hhead, hlist)
1604 link_css_set(&tmp_cg_links, cg, root_cgrp);
1606 write_unlock(&css_set_lock);
1608 free_cg_links(&tmp_cg_links);
1610 BUG_ON(!list_empty(&root_cgrp->sibling));
1611 BUG_ON(!list_empty(&root_cgrp->children));
1612 BUG_ON(root->number_of_cgroups != 1);
1614 cred = override_creds(&init_cred);
1615 cgroup_populate_dir(root_cgrp);
1617 mutex_unlock(&cgroup_root_mutex);
1618 mutex_unlock(&cgroup_mutex);
1619 mutex_unlock(&inode->i_mutex);
1622 * We re-used an existing hierarchy - the new root (if
1623 * any) is not needed
1625 cgroup_drop_root(opts.new_root);
1626 /* no subsys rebinding, so refcounts don't change */
1627 drop_parsed_module_refcounts(opts.subsys_bits);
1630 kfree(opts.release_agent);
1632 return dget(sb->s_root);
1635 mutex_unlock(&cgroup_root_mutex);
1636 mutex_unlock(&cgroup_mutex);
1637 mutex_unlock(&inode->i_mutex);
1639 deactivate_locked_super(sb);
1641 drop_parsed_module_refcounts(opts.subsys_bits);
1643 kfree(opts.release_agent);
1645 return ERR_PTR(ret);
1648 static void cgroup_kill_sb(struct super_block *sb) {
1649 struct cgroupfs_root *root = sb->s_fs_info;
1650 struct cgroup *cgrp = &root->top_cgroup;
1652 struct cg_cgroup_link *link;
1653 struct cg_cgroup_link *saved_link;
1657 BUG_ON(root->number_of_cgroups != 1);
1658 BUG_ON(!list_empty(&cgrp->children));
1659 BUG_ON(!list_empty(&cgrp->sibling));
1661 mutex_lock(&cgroup_mutex);
1662 mutex_lock(&cgroup_root_mutex);
1664 /* Rebind all subsystems back to the default hierarchy */
1665 ret = rebind_subsystems(root, 0);
1666 /* Shouldn't be able to fail ... */
1670 * Release all the links from css_sets to this hierarchy's
1673 write_lock(&css_set_lock);
1675 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1677 list_del(&link->cg_link_list);
1678 list_del(&link->cgrp_link_list);
1681 write_unlock(&css_set_lock);
1683 if (!list_empty(&root->root_list)) {
1684 list_del(&root->root_list);
1688 mutex_unlock(&cgroup_root_mutex);
1689 mutex_unlock(&cgroup_mutex);
1691 kill_litter_super(sb);
1692 cgroup_drop_root(root);
1695 static struct file_system_type cgroup_fs_type = {
1697 .mount = cgroup_mount,
1698 .kill_sb = cgroup_kill_sb,
1701 static struct kobject *cgroup_kobj;
1703 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1705 return dentry->d_fsdata;
1708 static inline struct cftype *__d_cft(struct dentry *dentry)
1710 return dentry->d_fsdata;
1714 * cgroup_path - generate the path of a cgroup
1715 * @cgrp: the cgroup in question
1716 * @buf: the buffer to write the path into
1717 * @buflen: the length of the buffer
1719 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1720 * reference. Writes path of cgroup into buf. Returns 0 on success,
1723 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1726 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1727 cgroup_lock_is_held());
1729 if (!dentry || cgrp == dummytop) {
1731 * Inactive subsystems have no dentry for their root
1738 start = buf + buflen;
1742 int len = dentry->d_name.len;
1744 if ((start -= len) < buf)
1745 return -ENAMETOOLONG;
1746 memcpy(start, dentry->d_name.name, len);
1747 cgrp = cgrp->parent;
1751 dentry = rcu_dereference_check(cgrp->dentry,
1752 cgroup_lock_is_held());
1756 return -ENAMETOOLONG;
1759 memmove(buf, start, buf + buflen - start);
1762 EXPORT_SYMBOL_GPL(cgroup_path);
1765 * Control Group taskset
1767 struct task_and_cgroup {
1768 struct task_struct *task;
1769 struct cgroup *cgrp;
1773 struct cgroup_taskset {
1774 struct task_and_cgroup single;
1775 struct flex_array *tc_array;
1778 struct cgroup *cur_cgrp;
1782 * cgroup_taskset_first - reset taskset and return the first task
1783 * @tset: taskset of interest
1785 * @tset iteration is initialized and the first task is returned.
1787 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1789 if (tset->tc_array) {
1791 return cgroup_taskset_next(tset);
1793 tset->cur_cgrp = tset->single.cgrp;
1794 return tset->single.task;
1797 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1800 * cgroup_taskset_next - iterate to the next task in taskset
1801 * @tset: taskset of interest
1803 * Return the next task in @tset. Iteration must have been initialized
1804 * with cgroup_taskset_first().
1806 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1808 struct task_and_cgroup *tc;
1810 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1813 tc = flex_array_get(tset->tc_array, tset->idx++);
1814 tset->cur_cgrp = tc->cgrp;
1817 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1820 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1821 * @tset: taskset of interest
1823 * Return the cgroup for the current (last returned) task of @tset. This
1824 * function must be preceded by either cgroup_taskset_first() or
1825 * cgroup_taskset_next().
1827 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1829 return tset->cur_cgrp;
1831 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1834 * cgroup_taskset_size - return the number of tasks in taskset
1835 * @tset: taskset of interest
1837 int cgroup_taskset_size(struct cgroup_taskset *tset)
1839 return tset->tc_array ? tset->tc_array_len : 1;
1841 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1845 * cgroup_task_migrate - move a task from one cgroup to another.
1847 * 'guarantee' is set if the caller promises that a new css_set for the task
1848 * will already exist. If not set, this function might sleep, and can fail with
1849 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1851 static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1852 struct task_struct *tsk, struct css_set *newcg)
1854 struct css_set *oldcg;
1857 * We are synchronized through threadgroup_lock() against PF_EXITING
1858 * setting such that we can't race against cgroup_exit() changing the
1859 * css_set to init_css_set and dropping the old one.
1861 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1862 oldcg = tsk->cgroups;
1865 rcu_assign_pointer(tsk->cgroups, newcg);
1868 /* Update the css_set linked lists if we're using them */
1869 write_lock(&css_set_lock);
1870 if (!list_empty(&tsk->cg_list))
1871 list_move(&tsk->cg_list, &newcg->tasks);
1872 write_unlock(&css_set_lock);
1875 * We just gained a reference on oldcg by taking it from the task. As
1876 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1877 * it here; it will be freed under RCU.
1879 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1884 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1885 * @cgrp: the cgroup the task is attaching to
1886 * @tsk: the task to be attached
1888 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1891 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1894 struct cgroup_subsys *ss, *failed_ss = NULL;
1895 struct cgroup *oldcgrp;
1896 struct cgroupfs_root *root = cgrp->root;
1897 struct cgroup_taskset tset = { };
1898 struct css_set *newcg;
1900 /* @tsk either already exited or can't exit until the end */
1901 if (tsk->flags & PF_EXITING)
1904 /* Nothing to do if the task is already in that cgroup */
1905 oldcgrp = task_cgroup_from_root(tsk, root);
1906 if (cgrp == oldcgrp)
1909 tset.single.task = tsk;
1910 tset.single.cgrp = oldcgrp;
1912 for_each_subsys(root, ss) {
1913 if (ss->can_attach) {
1914 retval = ss->can_attach(cgrp, &tset);
1917 * Remember on which subsystem the can_attach()
1918 * failed, so that we only call cancel_attach()
1919 * against the subsystems whose can_attach()
1920 * succeeded. (See below)
1928 newcg = find_css_set(tsk->cgroups, cgrp);
1934 cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1936 for_each_subsys(root, ss) {
1938 ss->attach(cgrp, &tset);
1944 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1945 * is no longer empty.
1947 cgroup_wakeup_rmdir_waiter(cgrp);
1950 for_each_subsys(root, ss) {
1951 if (ss == failed_ss)
1953 * This subsystem was the one that failed the
1954 * can_attach() check earlier, so we don't need
1955 * to call cancel_attach() against it or any
1956 * remaining subsystems.
1959 if (ss->cancel_attach)
1960 ss->cancel_attach(cgrp, &tset);
1967 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1968 * @from: attach to all cgroups of a given task
1969 * @tsk: the task to be attached
1971 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
1973 struct cgroupfs_root *root;
1977 for_each_active_root(root) {
1978 struct cgroup *from_cg = task_cgroup_from_root(from, root);
1980 retval = cgroup_attach_task(from_cg, tsk);
1988 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
1991 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
1992 * @cgrp: the cgroup to attach to
1993 * @leader: the threadgroup leader task_struct of the group to be attached
1995 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1996 * task_lock of each thread in leader's threadgroup individually in turn.
1998 static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2000 int retval, i, group_size;
2001 struct cgroup_subsys *ss, *failed_ss = NULL;
2002 /* guaranteed to be initialized later, but the compiler needs this */
2003 struct cgroupfs_root *root = cgrp->root;
2004 /* threadgroup list cursor and array */
2005 struct task_struct *tsk;
2006 struct task_and_cgroup *tc;
2007 struct flex_array *group;
2008 struct cgroup_taskset tset = { };
2011 * step 0: in order to do expensive, possibly blocking operations for
2012 * every thread, we cannot iterate the thread group list, since it needs
2013 * rcu or tasklist locked. instead, build an array of all threads in the
2014 * group - group_rwsem prevents new threads from appearing, and if
2015 * threads exit, this will just be an over-estimate.
2017 group_size = get_nr_threads(leader);
2018 /* flex_array supports very large thread-groups better than kmalloc. */
2019 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2022 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2023 retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2025 goto out_free_group_list;
2030 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2031 * already PF_EXITING could be freed from underneath us unless we
2032 * take an rcu_read_lock.
2036 struct task_and_cgroup ent;
2038 /* @tsk either already exited or can't exit until the end */
2039 if (tsk->flags & PF_EXITING)
2042 /* as per above, nr_threads may decrease, but not increase. */
2043 BUG_ON(i >= group_size);
2045 ent.cgrp = task_cgroup_from_root(tsk, root);
2046 /* nothing to do if this task is already in the cgroup */
2047 if (ent.cgrp == cgrp)
2050 * saying GFP_ATOMIC has no effect here because we did prealloc
2051 * earlier, but it's good form to communicate our expectations.
2053 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2054 BUG_ON(retval != 0);
2056 } while_each_thread(leader, tsk);
2058 /* remember the number of threads in the array for later. */
2060 tset.tc_array = group;
2061 tset.tc_array_len = group_size;
2063 /* methods shouldn't be called if no task is actually migrating */
2066 goto out_free_group_list;
2069 * step 1: check that we can legitimately attach to the cgroup.
2071 for_each_subsys(root, ss) {
2072 if (ss->can_attach) {
2073 retval = ss->can_attach(cgrp, &tset);
2076 goto out_cancel_attach;
2082 * step 2: make sure css_sets exist for all threads to be migrated.
2083 * we use find_css_set, which allocates a new one if necessary.
2085 for (i = 0; i < group_size; i++) {
2086 tc = flex_array_get(group, i);
2087 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2090 goto out_put_css_set_refs;
2095 * step 3: now that we're guaranteed success wrt the css_sets,
2096 * proceed to move all tasks to the new cgroup. There are no
2097 * failure cases after here, so this is the commit point.
2099 for (i = 0; i < group_size; i++) {
2100 tc = flex_array_get(group, i);
2101 cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2103 /* nothing is sensitive to fork() after this point. */
2106 * step 4: do subsystem attach callbacks.
2108 for_each_subsys(root, ss) {
2110 ss->attach(cgrp, &tset);
2114 * step 5: success! and cleanup
2117 cgroup_wakeup_rmdir_waiter(cgrp);
2119 out_put_css_set_refs:
2121 for (i = 0; i < group_size; i++) {
2122 tc = flex_array_get(group, i);
2125 put_css_set(tc->cg);
2130 for_each_subsys(root, ss) {
2131 if (ss == failed_ss)
2133 if (ss->cancel_attach)
2134 ss->cancel_attach(cgrp, &tset);
2137 out_free_group_list:
2138 flex_array_free(group);
2143 * Find the task_struct of the task to attach by vpid and pass it along to the
2144 * function to attach either it or all tasks in its threadgroup. Will lock
2145 * cgroup_mutex and threadgroup; may take task_lock of task.
2147 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2149 struct task_struct *tsk;
2150 const struct cred *cred = current_cred(), *tcred;
2153 if (!cgroup_lock_live_group(cgrp))
2159 tsk = find_task_by_vpid(pid);
2163 goto out_unlock_cgroup;
2166 * even if we're attaching all tasks in the thread group, we
2167 * only need to check permissions on one of them.
2169 tcred = __task_cred(tsk);
2171 cred->euid != tcred->uid &&
2172 cred->euid != tcred->suid) {
2175 goto out_unlock_cgroup;
2181 tsk = tsk->group_leader;
2182 get_task_struct(tsk);
2185 threadgroup_lock(tsk);
2187 if (!thread_group_leader(tsk)) {
2189 * a race with de_thread from another thread's exec()
2190 * may strip us of our leadership, if this happens,
2191 * there is no choice but to throw this task away and
2192 * try again; this is
2193 * "double-double-toil-and-trouble-check locking".
2195 threadgroup_unlock(tsk);
2196 put_task_struct(tsk);
2197 goto retry_find_task;
2199 ret = cgroup_attach_proc(cgrp, tsk);
2201 ret = cgroup_attach_task(cgrp, tsk);
2202 threadgroup_unlock(tsk);
2204 put_task_struct(tsk);
2210 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2212 return attach_task_by_pid(cgrp, pid, false);
2215 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2217 return attach_task_by_pid(cgrp, tgid, true);
2221 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2222 * @cgrp: the cgroup to be checked for liveness
2224 * On success, returns true; the lock should be later released with
2225 * cgroup_unlock(). On failure returns false with no lock held.
2227 bool cgroup_lock_live_group(struct cgroup *cgrp)
2229 mutex_lock(&cgroup_mutex);
2230 if (cgroup_is_removed(cgrp)) {
2231 mutex_unlock(&cgroup_mutex);
2236 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2238 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2241 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2242 if (strlen(buffer) >= PATH_MAX)
2244 if (!cgroup_lock_live_group(cgrp))
2246 mutex_lock(&cgroup_root_mutex);
2247 strcpy(cgrp->root->release_agent_path, buffer);
2248 mutex_unlock(&cgroup_root_mutex);
2253 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2254 struct seq_file *seq)
2256 if (!cgroup_lock_live_group(cgrp))
2258 seq_puts(seq, cgrp->root->release_agent_path);
2259 seq_putc(seq, '\n');
2264 /* A buffer size big enough for numbers or short strings */
2265 #define CGROUP_LOCAL_BUFFER_SIZE 64
2267 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2269 const char __user *userbuf,
2270 size_t nbytes, loff_t *unused_ppos)
2272 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2278 if (nbytes >= sizeof(buffer))
2280 if (copy_from_user(buffer, userbuf, nbytes))
2283 buffer[nbytes] = 0; /* nul-terminate */
2284 if (cft->write_u64) {
2285 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2288 retval = cft->write_u64(cgrp, cft, val);
2290 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2293 retval = cft->write_s64(cgrp, cft, val);
2300 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2302 const char __user *userbuf,
2303 size_t nbytes, loff_t *unused_ppos)
2305 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2307 size_t max_bytes = cft->max_write_len;
2308 char *buffer = local_buffer;
2311 max_bytes = sizeof(local_buffer) - 1;
2312 if (nbytes >= max_bytes)
2314 /* Allocate a dynamic buffer if we need one */
2315 if (nbytes >= sizeof(local_buffer)) {
2316 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2320 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2325 buffer[nbytes] = 0; /* nul-terminate */
2326 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2330 if (buffer != local_buffer)
2335 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2336 size_t nbytes, loff_t *ppos)
2338 struct cftype *cft = __d_cft(file->f_dentry);
2339 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2341 if (cgroup_is_removed(cgrp))
2344 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2345 if (cft->write_u64 || cft->write_s64)
2346 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2347 if (cft->write_string)
2348 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2350 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2351 return ret ? ret : nbytes;
2356 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2358 char __user *buf, size_t nbytes,
2361 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2362 u64 val = cft->read_u64(cgrp, cft);
2363 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2365 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2368 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2370 char __user *buf, size_t nbytes,
2373 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2374 s64 val = cft->read_s64(cgrp, cft);
2375 int len = sprintf(tmp, "%lld\n", (long long) val);
2377 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2380 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2381 size_t nbytes, loff_t *ppos)
2383 struct cftype *cft = __d_cft(file->f_dentry);
2384 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2386 if (cgroup_is_removed(cgrp))
2390 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2392 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2394 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2399 * seqfile ops/methods for returning structured data. Currently just
2400 * supports string->u64 maps, but can be extended in future.
2403 struct cgroup_seqfile_state {
2405 struct cgroup *cgroup;
2408 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2410 struct seq_file *sf = cb->state;
2411 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2414 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2416 struct cgroup_seqfile_state *state = m->private;
2417 struct cftype *cft = state->cft;
2418 if (cft->read_map) {
2419 struct cgroup_map_cb cb = {
2420 .fill = cgroup_map_add,
2423 return cft->read_map(state->cgroup, cft, &cb);
2425 return cft->read_seq_string(state->cgroup, cft, m);
2428 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2430 struct seq_file *seq = file->private_data;
2431 kfree(seq->private);
2432 return single_release(inode, file);
2435 static const struct file_operations cgroup_seqfile_operations = {
2437 .write = cgroup_file_write,
2438 .llseek = seq_lseek,
2439 .release = cgroup_seqfile_release,
2442 static int cgroup_file_open(struct inode *inode, struct file *file)
2447 err = generic_file_open(inode, file);
2450 cft = __d_cft(file->f_dentry);
2452 if (cft->read_map || cft->read_seq_string) {
2453 struct cgroup_seqfile_state *state =
2454 kzalloc(sizeof(*state), GFP_USER);
2458 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2459 file->f_op = &cgroup_seqfile_operations;
2460 err = single_open(file, cgroup_seqfile_show, state);
2463 } else if (cft->open)
2464 err = cft->open(inode, file);
2471 static int cgroup_file_release(struct inode *inode, struct file *file)
2473 struct cftype *cft = __d_cft(file->f_dentry);
2475 return cft->release(inode, file);
2480 * cgroup_rename - Only allow simple rename of directories in place.
2482 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2483 struct inode *new_dir, struct dentry *new_dentry)
2485 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2487 if (new_dentry->d_inode)
2489 if (old_dir != new_dir)
2491 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2494 static const struct file_operations cgroup_file_operations = {
2495 .read = cgroup_file_read,
2496 .write = cgroup_file_write,
2497 .llseek = generic_file_llseek,
2498 .open = cgroup_file_open,
2499 .release = cgroup_file_release,
2502 static const struct inode_operations cgroup_dir_inode_operations = {
2503 .lookup = cgroup_lookup,
2504 .mkdir = cgroup_mkdir,
2505 .rmdir = cgroup_rmdir,
2506 .rename = cgroup_rename,
2509 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
2511 if (dentry->d_name.len > NAME_MAX)
2512 return ERR_PTR(-ENAMETOOLONG);
2513 d_add(dentry, NULL);
2518 * Check if a file is a control file
2520 static inline struct cftype *__file_cft(struct file *file)
2522 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2523 return ERR_PTR(-EINVAL);
2524 return __d_cft(file->f_dentry);
2527 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2528 struct super_block *sb)
2530 struct inode *inode;
2534 if (dentry->d_inode)
2537 inode = cgroup_new_inode(mode, sb);
2541 if (S_ISDIR(mode)) {
2542 inode->i_op = &cgroup_dir_inode_operations;
2543 inode->i_fop = &simple_dir_operations;
2545 /* start off with i_nlink == 2 (for "." entry) */
2548 /* start with the directory inode held, so that we can
2549 * populate it without racing with another mkdir */
2550 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2551 } else if (S_ISREG(mode)) {
2553 inode->i_fop = &cgroup_file_operations;
2555 d_instantiate(dentry, inode);
2556 dget(dentry); /* Extra count - pin the dentry in core */
2561 * cgroup_create_dir - create a directory for an object.
2562 * @cgrp: the cgroup we create the directory for. It must have a valid
2563 * ->parent field. And we are going to fill its ->dentry field.
2564 * @dentry: dentry of the new cgroup
2565 * @mode: mode to set on new directory.
2567 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2570 struct dentry *parent;
2573 parent = cgrp->parent->dentry;
2574 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2576 dentry->d_fsdata = cgrp;
2577 inc_nlink(parent->d_inode);
2578 rcu_assign_pointer(cgrp->dentry, dentry);
2585 * cgroup_file_mode - deduce file mode of a control file
2586 * @cft: the control file in question
2588 * returns cft->mode if ->mode is not 0
2589 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2590 * returns S_IRUGO if it has only a read handler
2591 * returns S_IWUSR if it has only a write hander
2593 static umode_t cgroup_file_mode(const struct cftype *cft)
2600 if (cft->read || cft->read_u64 || cft->read_s64 ||
2601 cft->read_map || cft->read_seq_string)
2604 if (cft->write || cft->write_u64 || cft->write_s64 ||
2605 cft->write_string || cft->trigger)
2611 int cgroup_add_file(struct cgroup *cgrp,
2612 struct cgroup_subsys *subsys,
2613 const struct cftype *cft)
2615 struct dentry *dir = cgrp->dentry;
2616 struct dentry *dentry;
2620 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2621 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2622 strcpy(name, subsys->name);
2625 strcat(name, cft->name);
2626 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2627 dentry = lookup_one_len(name, dir, strlen(name));
2628 if (!IS_ERR(dentry)) {
2629 mode = cgroup_file_mode(cft);
2630 error = cgroup_create_file(dentry, mode | S_IFREG,
2633 dentry->d_fsdata = (void *)cft;
2636 error = PTR_ERR(dentry);
2639 EXPORT_SYMBOL_GPL(cgroup_add_file);
2641 int cgroup_add_files(struct cgroup *cgrp,
2642 struct cgroup_subsys *subsys,
2643 const struct cftype cft[],
2647 for (i = 0; i < count; i++) {
2648 err = cgroup_add_file(cgrp, subsys, &cft[i]);
2654 EXPORT_SYMBOL_GPL(cgroup_add_files);
2657 * cgroup_task_count - count the number of tasks in a cgroup.
2658 * @cgrp: the cgroup in question
2660 * Return the number of tasks in the cgroup.
2662 int cgroup_task_count(const struct cgroup *cgrp)
2665 struct cg_cgroup_link *link;
2667 read_lock(&css_set_lock);
2668 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2669 count += atomic_read(&link->cg->refcount);
2671 read_unlock(&css_set_lock);
2676 * Advance a list_head iterator. The iterator should be positioned at
2677 * the start of a css_set
2679 static void cgroup_advance_iter(struct cgroup *cgrp,
2680 struct cgroup_iter *it)
2682 struct list_head *l = it->cg_link;
2683 struct cg_cgroup_link *link;
2686 /* Advance to the next non-empty css_set */
2689 if (l == &cgrp->css_sets) {
2693 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2695 } while (list_empty(&cg->tasks));
2697 it->task = cg->tasks.next;
2701 * To reduce the fork() overhead for systems that are not actually
2702 * using their cgroups capability, we don't maintain the lists running
2703 * through each css_set to its tasks until we see the list actually
2704 * used - in other words after the first call to cgroup_iter_start().
2706 static void cgroup_enable_task_cg_lists(void)
2708 struct task_struct *p, *g;
2709 write_lock(&css_set_lock);
2710 use_task_css_set_links = 1;
2712 * We need tasklist_lock because RCU is not safe against
2713 * while_each_thread(). Besides, a forking task that has passed
2714 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2715 * is not guaranteed to have its child immediately visible in the
2716 * tasklist if we walk through it with RCU.
2718 read_lock(&tasklist_lock);
2719 do_each_thread(g, p) {
2722 * We should check if the process is exiting, otherwise
2723 * it will race with cgroup_exit() in that the list
2724 * entry won't be deleted though the process has exited.
2726 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2727 list_add(&p->cg_list, &p->cgroups->tasks);
2729 } while_each_thread(g, p);
2730 read_unlock(&tasklist_lock);
2731 write_unlock(&css_set_lock);
2734 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2735 __acquires(css_set_lock)
2738 * The first time anyone tries to iterate across a cgroup,
2739 * we need to enable the list linking each css_set to its
2740 * tasks, and fix up all existing tasks.
2742 if (!use_task_css_set_links)
2743 cgroup_enable_task_cg_lists();
2745 read_lock(&css_set_lock);
2746 it->cg_link = &cgrp->css_sets;
2747 cgroup_advance_iter(cgrp, it);
2750 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2751 struct cgroup_iter *it)
2753 struct task_struct *res;
2754 struct list_head *l = it->task;
2755 struct cg_cgroup_link *link;
2757 /* If the iterator cg is NULL, we have no tasks */
2760 res = list_entry(l, struct task_struct, cg_list);
2761 /* Advance iterator to find next entry */
2763 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2764 if (l == &link->cg->tasks) {
2765 /* We reached the end of this task list - move on to
2766 * the next cg_cgroup_link */
2767 cgroup_advance_iter(cgrp, it);
2774 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2775 __releases(css_set_lock)
2777 read_unlock(&css_set_lock);
2780 static inline int started_after_time(struct task_struct *t1,
2781 struct timespec *time,
2782 struct task_struct *t2)
2784 int start_diff = timespec_compare(&t1->start_time, time);
2785 if (start_diff > 0) {
2787 } else if (start_diff < 0) {
2791 * Arbitrarily, if two processes started at the same
2792 * time, we'll say that the lower pointer value
2793 * started first. Note that t2 may have exited by now
2794 * so this may not be a valid pointer any longer, but
2795 * that's fine - it still serves to distinguish
2796 * between two tasks started (effectively) simultaneously.
2803 * This function is a callback from heap_insert() and is used to order
2805 * In this case we order the heap in descending task start time.
2807 static inline int started_after(void *p1, void *p2)
2809 struct task_struct *t1 = p1;
2810 struct task_struct *t2 = p2;
2811 return started_after_time(t1, &t2->start_time, t2);
2815 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2816 * @scan: struct cgroup_scanner containing arguments for the scan
2818 * Arguments include pointers to callback functions test_task() and
2820 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2821 * and if it returns true, call process_task() for it also.
2822 * The test_task pointer may be NULL, meaning always true (select all tasks).
2823 * Effectively duplicates cgroup_iter_{start,next,end}()
2824 * but does not lock css_set_lock for the call to process_task().
2825 * The struct cgroup_scanner may be embedded in any structure of the caller's
2827 * It is guaranteed that process_task() will act on every task that
2828 * is a member of the cgroup for the duration of this call. This
2829 * function may or may not call process_task() for tasks that exit
2830 * or move to a different cgroup during the call, or are forked or
2831 * move into the cgroup during the call.
2833 * Note that test_task() may be called with locks held, and may in some
2834 * situations be called multiple times for the same task, so it should
2836 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2837 * pre-allocated and will be used for heap operations (and its "gt" member will
2838 * be overwritten), else a temporary heap will be used (allocation of which
2839 * may cause this function to fail).
2841 int cgroup_scan_tasks(struct cgroup_scanner *scan)
2844 struct cgroup_iter it;
2845 struct task_struct *p, *dropped;
2846 /* Never dereference latest_task, since it's not refcounted */
2847 struct task_struct *latest_task = NULL;
2848 struct ptr_heap tmp_heap;
2849 struct ptr_heap *heap;
2850 struct timespec latest_time = { 0, 0 };
2853 /* The caller supplied our heap and pre-allocated its memory */
2855 heap->gt = &started_after;
2857 /* We need to allocate our own heap memory */
2859 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2861 /* cannot allocate the heap */
2867 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2868 * to determine which are of interest, and using the scanner's
2869 * "process_task" callback to process any of them that need an update.
2870 * Since we don't want to hold any locks during the task updates,
2871 * gather tasks to be processed in a heap structure.
2872 * The heap is sorted by descending task start time.
2873 * If the statically-sized heap fills up, we overflow tasks that
2874 * started later, and in future iterations only consider tasks that
2875 * started after the latest task in the previous pass. This
2876 * guarantees forward progress and that we don't miss any tasks.
2879 cgroup_iter_start(scan->cg, &it);
2880 while ((p = cgroup_iter_next(scan->cg, &it))) {
2882 * Only affect tasks that qualify per the caller's callback,
2883 * if he provided one
2885 if (scan->test_task && !scan->test_task(p, scan))
2888 * Only process tasks that started after the last task
2891 if (!started_after_time(p, &latest_time, latest_task))
2893 dropped = heap_insert(heap, p);
2894 if (dropped == NULL) {
2896 * The new task was inserted; the heap wasn't
2900 } else if (dropped != p) {
2902 * The new task was inserted, and pushed out a
2906 put_task_struct(dropped);
2909 * Else the new task was newer than anything already in
2910 * the heap and wasn't inserted
2913 cgroup_iter_end(scan->cg, &it);
2916 for (i = 0; i < heap->size; i++) {
2917 struct task_struct *q = heap->ptrs[i];
2919 latest_time = q->start_time;
2922 /* Process the task per the caller's callback */
2923 scan->process_task(q, scan);
2927 * If we had to process any tasks at all, scan again
2928 * in case some of them were in the middle of forking
2929 * children that didn't get processed.
2930 * Not the most efficient way to do it, but it avoids
2931 * having to take callback_mutex in the fork path
2935 if (heap == &tmp_heap)
2936 heap_free(&tmp_heap);
2941 * Stuff for reading the 'tasks'/'procs' files.
2943 * Reading this file can return large amounts of data if a cgroup has
2944 * *lots* of attached tasks. So it may need several calls to read(),
2945 * but we cannot guarantee that the information we produce is correct
2946 * unless we produce it entirely atomically.
2950 /* which pidlist file are we talking about? */
2951 enum cgroup_filetype {
2957 * A pidlist is a list of pids that virtually represents the contents of one
2958 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
2959 * a pair (one each for procs, tasks) for each pid namespace that's relevant
2962 struct cgroup_pidlist {
2964 * used to find which pidlist is wanted. doesn't change as long as
2965 * this particular list stays in the list.
2967 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
2970 /* how many elements the above list has */
2972 /* how many files are using the current array */
2974 /* each of these stored in a list by its cgroup */
2975 struct list_head links;
2976 /* pointer to the cgroup we belong to, for list removal purposes */
2977 struct cgroup *owner;
2978 /* protects the other fields */
2979 struct rw_semaphore mutex;
2983 * The following two functions "fix" the issue where there are more pids
2984 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2985 * TODO: replace with a kernel-wide solution to this problem
2987 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2988 static void *pidlist_allocate(int count)
2990 if (PIDLIST_TOO_LARGE(count))
2991 return vmalloc(count * sizeof(pid_t));
2993 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
2995 static void pidlist_free(void *p)
2997 if (is_vmalloc_addr(p))
3002 static void *pidlist_resize(void *p, int newcount)
3005 /* note: if new alloc fails, old p will still be valid either way */
3006 if (is_vmalloc_addr(p)) {
3007 newlist = vmalloc(newcount * sizeof(pid_t));
3010 memcpy(newlist, p, newcount * sizeof(pid_t));
3013 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3019 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3020 * If the new stripped list is sufficiently smaller and there's enough memory
3021 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3022 * number of unique elements.
3024 /* is the size difference enough that we should re-allocate the array? */
3025 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3026 static int pidlist_uniq(pid_t **p, int length)
3033 * we presume the 0th element is unique, so i starts at 1. trivial
3034 * edge cases first; no work needs to be done for either
3036 if (length == 0 || length == 1)
3038 /* src and dest walk down the list; dest counts unique elements */
3039 for (src = 1; src < length; src++) {
3040 /* find next unique element */
3041 while (list[src] == list[src-1]) {
3046 /* dest always points to where the next unique element goes */
3047 list[dest] = list[src];
3052 * if the length difference is large enough, we want to allocate a
3053 * smaller buffer to save memory. if this fails due to out of memory,
3054 * we'll just stay with what we've got.
3056 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3057 newlist = pidlist_resize(list, dest);
3064 static int cmppid(const void *a, const void *b)
3066 return *(pid_t *)a - *(pid_t *)b;
3070 * find the appropriate pidlist for our purpose (given procs vs tasks)
3071 * returns with the lock on that pidlist already held, and takes care
3072 * of the use count, or returns NULL with no locks held if we're out of
3075 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3076 enum cgroup_filetype type)
3078 struct cgroup_pidlist *l;
3079 /* don't need task_nsproxy() if we're looking at ourself */
3080 struct pid_namespace *ns = current->nsproxy->pid_ns;
3083 * We can't drop the pidlist_mutex before taking the l->mutex in case
3084 * the last ref-holder is trying to remove l from the list at the same
3085 * time. Holding the pidlist_mutex precludes somebody taking whichever
3086 * list we find out from under us - compare release_pid_array().
3088 mutex_lock(&cgrp->pidlist_mutex);
3089 list_for_each_entry(l, &cgrp->pidlists, links) {
3090 if (l->key.type == type && l->key.ns == ns) {
3091 /* make sure l doesn't vanish out from under us */
3092 down_write(&l->mutex);
3093 mutex_unlock(&cgrp->pidlist_mutex);
3097 /* entry not found; create a new one */
3098 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3100 mutex_unlock(&cgrp->pidlist_mutex);
3103 init_rwsem(&l->mutex);
3104 down_write(&l->mutex);
3106 l->key.ns = get_pid_ns(ns);
3107 l->use_count = 0; /* don't increment here */
3110 list_add(&l->links, &cgrp->pidlists);
3111 mutex_unlock(&cgrp->pidlist_mutex);
3116 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3118 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3119 struct cgroup_pidlist **lp)
3123 int pid, n = 0; /* used for populating the array */
3124 struct cgroup_iter it;
3125 struct task_struct *tsk;
3126 struct cgroup_pidlist *l;
3129 * If cgroup gets more users after we read count, we won't have
3130 * enough space - tough. This race is indistinguishable to the
3131 * caller from the case that the additional cgroup users didn't
3132 * show up until sometime later on.
3134 length = cgroup_task_count(cgrp);
3135 array = pidlist_allocate(length);
3138 /* now, populate the array */
3139 cgroup_iter_start(cgrp, &it);
3140 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3141 if (unlikely(n == length))
3143 /* get tgid or pid for procs or tasks file respectively */
3144 if (type == CGROUP_FILE_PROCS)
3145 pid = task_tgid_vnr(tsk);
3147 pid = task_pid_vnr(tsk);
3148 if (pid > 0) /* make sure to only use valid results */
3151 cgroup_iter_end(cgrp, &it);
3153 /* now sort & (if procs) strip out duplicates */
3154 sort(array, length, sizeof(pid_t), cmppid, NULL);
3155 if (type == CGROUP_FILE_PROCS)
3156 length = pidlist_uniq(&array, length);
3157 l = cgroup_pidlist_find(cgrp, type);
3159 pidlist_free(array);
3162 /* store array, freeing old if necessary - lock already held */
3163 pidlist_free(l->list);
3167 up_write(&l->mutex);
3173 * cgroupstats_build - build and fill cgroupstats
3174 * @stats: cgroupstats to fill information into
3175 * @dentry: A dentry entry belonging to the cgroup for which stats have
3178 * Build and fill cgroupstats so that taskstats can export it to user
3181 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3184 struct cgroup *cgrp;
3185 struct cgroup_iter it;
3186 struct task_struct *tsk;
3189 * Validate dentry by checking the superblock operations,
3190 * and make sure it's a directory.
3192 if (dentry->d_sb->s_op != &cgroup_ops ||
3193 !S_ISDIR(dentry->d_inode->i_mode))
3197 cgrp = dentry->d_fsdata;
3199 cgroup_iter_start(cgrp, &it);
3200 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3201 switch (tsk->state) {
3203 stats->nr_running++;
3205 case TASK_INTERRUPTIBLE:
3206 stats->nr_sleeping++;
3208 case TASK_UNINTERRUPTIBLE:
3209 stats->nr_uninterruptible++;
3212 stats->nr_stopped++;
3215 if (delayacct_is_task_waiting_on_io(tsk))
3216 stats->nr_io_wait++;
3220 cgroup_iter_end(cgrp, &it);
3228 * seq_file methods for the tasks/procs files. The seq_file position is the
3229 * next pid to display; the seq_file iterator is a pointer to the pid
3230 * in the cgroup->l->list array.
3233 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3236 * Initially we receive a position value that corresponds to
3237 * one more than the last pid shown (or 0 on the first call or
3238 * after a seek to the start). Use a binary-search to find the
3239 * next pid to display, if any
3241 struct cgroup_pidlist *l = s->private;
3242 int index = 0, pid = *pos;
3245 down_read(&l->mutex);
3247 int end = l->length;
3249 while (index < end) {
3250 int mid = (index + end) / 2;
3251 if (l->list[mid] == pid) {
3254 } else if (l->list[mid] <= pid)
3260 /* If we're off the end of the array, we're done */
3261 if (index >= l->length)
3263 /* Update the abstract position to be the actual pid that we found */
3264 iter = l->list + index;
3269 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3271 struct cgroup_pidlist *l = s->private;
3275 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3277 struct cgroup_pidlist *l = s->private;
3279 pid_t *end = l->list + l->length;
3281 * Advance to the next pid in the array. If this goes off the
3293 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3295 return seq_printf(s, "%d\n", *(int *)v);
3299 * seq_operations functions for iterating on pidlists through seq_file -
3300 * independent of whether it's tasks or procs
3302 static const struct seq_operations cgroup_pidlist_seq_operations = {
3303 .start = cgroup_pidlist_start,
3304 .stop = cgroup_pidlist_stop,
3305 .next = cgroup_pidlist_next,
3306 .show = cgroup_pidlist_show,
3309 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3312 * the case where we're the last user of this particular pidlist will
3313 * have us remove it from the cgroup's list, which entails taking the
3314 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3315 * pidlist_mutex, we have to take pidlist_mutex first.
3317 mutex_lock(&l->owner->pidlist_mutex);
3318 down_write(&l->mutex);
3319 BUG_ON(!l->use_count);
3320 if (!--l->use_count) {
3321 /* we're the last user if refcount is 0; remove and free */
3322 list_del(&l->links);
3323 mutex_unlock(&l->owner->pidlist_mutex);
3324 pidlist_free(l->list);
3325 put_pid_ns(l->key.ns);
3326 up_write(&l->mutex);
3330 mutex_unlock(&l->owner->pidlist_mutex);
3331 up_write(&l->mutex);
3334 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3336 struct cgroup_pidlist *l;
3337 if (!(file->f_mode & FMODE_READ))
3340 * the seq_file will only be initialized if the file was opened for
3341 * reading; hence we check if it's not null only in that case.
3343 l = ((struct seq_file *)file->private_data)->private;
3344 cgroup_release_pid_array(l);
3345 return seq_release(inode, file);
3348 static const struct file_operations cgroup_pidlist_operations = {
3350 .llseek = seq_lseek,
3351 .write = cgroup_file_write,
3352 .release = cgroup_pidlist_release,
3356 * The following functions handle opens on a file that displays a pidlist
3357 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3360 /* helper function for the two below it */
3361 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3363 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3364 struct cgroup_pidlist *l;
3367 /* Nothing to do for write-only files */
3368 if (!(file->f_mode & FMODE_READ))
3371 /* have the array populated */
3372 retval = pidlist_array_load(cgrp, type, &l);
3375 /* configure file information */
3376 file->f_op = &cgroup_pidlist_operations;
3378 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3380 cgroup_release_pid_array(l);
3383 ((struct seq_file *)file->private_data)->private = l;
3386 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3388 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3390 static int cgroup_procs_open(struct inode *unused, struct file *file)
3392 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3395 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3398 return notify_on_release(cgrp);
3401 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3405 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3407 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3409 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3414 * Unregister event and free resources.
3416 * Gets called from workqueue.
3418 static void cgroup_event_remove(struct work_struct *work)
3420 struct cgroup_event *event = container_of(work, struct cgroup_event,
3422 struct cgroup *cgrp = event->cgrp;
3424 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3426 eventfd_ctx_put(event->eventfd);
3432 * Gets called on POLLHUP on eventfd when user closes it.
3434 * Called with wqh->lock held and interrupts disabled.
3436 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3437 int sync, void *key)
3439 struct cgroup_event *event = container_of(wait,
3440 struct cgroup_event, wait);
3441 struct cgroup *cgrp = event->cgrp;
3442 unsigned long flags = (unsigned long)key;
3444 if (flags & POLLHUP) {
3445 __remove_wait_queue(event->wqh, &event->wait);
3446 spin_lock(&cgrp->event_list_lock);
3447 list_del(&event->list);
3448 spin_unlock(&cgrp->event_list_lock);
3450 * We are in atomic context, but cgroup_event_remove() may
3451 * sleep, so we have to call it in workqueue.
3453 schedule_work(&event->remove);
3459 static void cgroup_event_ptable_queue_proc(struct file *file,
3460 wait_queue_head_t *wqh, poll_table *pt)
3462 struct cgroup_event *event = container_of(pt,
3463 struct cgroup_event, pt);
3466 add_wait_queue(wqh, &event->wait);
3470 * Parse input and register new cgroup event handler.
3472 * Input must be in format '<event_fd> <control_fd> <args>'.
3473 * Interpretation of args is defined by control file implementation.
3475 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3478 struct cgroup_event *event = NULL;
3479 unsigned int efd, cfd;
3480 struct file *efile = NULL;
3481 struct file *cfile = NULL;
3485 efd = simple_strtoul(buffer, &endp, 10);
3490 cfd = simple_strtoul(buffer, &endp, 10);
3491 if ((*endp != ' ') && (*endp != '\0'))
3495 event = kzalloc(sizeof(*event), GFP_KERNEL);
3499 INIT_LIST_HEAD(&event->list);
3500 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3501 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3502 INIT_WORK(&event->remove, cgroup_event_remove);
3504 efile = eventfd_fget(efd);
3505 if (IS_ERR(efile)) {
3506 ret = PTR_ERR(efile);
3510 event->eventfd = eventfd_ctx_fileget(efile);
3511 if (IS_ERR(event->eventfd)) {
3512 ret = PTR_ERR(event->eventfd);
3522 /* the process need read permission on control file */
3523 /* AV: shouldn't we check that it's been opened for read instead? */
3524 ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3528 event->cft = __file_cft(cfile);
3529 if (IS_ERR(event->cft)) {
3530 ret = PTR_ERR(event->cft);
3534 if (!event->cft->register_event || !event->cft->unregister_event) {
3539 ret = event->cft->register_event(cgrp, event->cft,
3540 event->eventfd, buffer);
3544 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3545 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3551 * Events should be removed after rmdir of cgroup directory, but before
3552 * destroying subsystem state objects. Let's take reference to cgroup
3553 * directory dentry to do that.
3557 spin_lock(&cgrp->event_list_lock);
3558 list_add(&event->list, &cgrp->event_list);
3559 spin_unlock(&cgrp->event_list_lock);
3570 if (event && event->eventfd && !IS_ERR(event->eventfd))
3571 eventfd_ctx_put(event->eventfd);
3573 if (!IS_ERR_OR_NULL(efile))
3581 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3584 return clone_children(cgrp);
3587 static int cgroup_clone_children_write(struct cgroup *cgrp,
3592 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3594 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3599 * for the common functions, 'private' gives the type of file
3601 /* for hysterical raisins, we can't put this on the older files */
3602 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3603 static struct cftype files[] = {
3606 .open = cgroup_tasks_open,
3607 .write_u64 = cgroup_tasks_write,
3608 .release = cgroup_pidlist_release,
3609 .mode = S_IRUGO | S_IWUSR,
3612 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3613 .open = cgroup_procs_open,
3614 .write_u64 = cgroup_procs_write,
3615 .release = cgroup_pidlist_release,
3616 .mode = S_IRUGO | S_IWUSR,
3619 .name = "notify_on_release",
3620 .read_u64 = cgroup_read_notify_on_release,
3621 .write_u64 = cgroup_write_notify_on_release,
3624 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3625 .write_string = cgroup_write_event_control,
3629 .name = "cgroup.clone_children",
3630 .read_u64 = cgroup_clone_children_read,
3631 .write_u64 = cgroup_clone_children_write,
3635 static struct cftype cft_release_agent = {
3636 .name = "release_agent",
3637 .read_seq_string = cgroup_release_agent_show,
3638 .write_string = cgroup_release_agent_write,
3639 .max_write_len = PATH_MAX,
3642 static int cgroup_populate_dir(struct cgroup *cgrp)
3645 struct cgroup_subsys *ss;
3647 /* First clear out any existing files */
3648 cgroup_clear_directory(cgrp->dentry);
3650 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
3654 if (cgrp == cgrp->top_cgroup) {
3655 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
3659 for_each_subsys(cgrp->root, ss) {
3660 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
3663 /* This cgroup is ready now */
3664 for_each_subsys(cgrp->root, ss) {
3665 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3667 * Update id->css pointer and make this css visible from
3668 * CSS ID functions. This pointer will be dereferened
3669 * from RCU-read-side without locks.
3672 rcu_assign_pointer(css->id->css, css);
3678 static void init_cgroup_css(struct cgroup_subsys_state *css,
3679 struct cgroup_subsys *ss,
3680 struct cgroup *cgrp)
3683 atomic_set(&css->refcnt, 1);
3686 if (cgrp == dummytop)
3687 set_bit(CSS_ROOT, &css->flags);
3688 BUG_ON(cgrp->subsys[ss->subsys_id]);
3689 cgrp->subsys[ss->subsys_id] = css;
3692 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3694 /* We need to take each hierarchy_mutex in a consistent order */
3698 * No worry about a race with rebind_subsystems that might mess up the
3699 * locking order, since both parties are under cgroup_mutex.
3701 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3702 struct cgroup_subsys *ss = subsys[i];
3705 if (ss->root == root)
3706 mutex_lock(&ss->hierarchy_mutex);
3710 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3714 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3715 struct cgroup_subsys *ss = subsys[i];
3718 if (ss->root == root)
3719 mutex_unlock(&ss->hierarchy_mutex);
3724 * cgroup_create - create a cgroup
3725 * @parent: cgroup that will be parent of the new cgroup
3726 * @dentry: dentry of the new cgroup
3727 * @mode: mode to set on new inode
3729 * Must be called with the mutex on the parent inode held
3731 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3734 struct cgroup *cgrp;
3735 struct cgroupfs_root *root = parent->root;
3737 struct cgroup_subsys *ss;
3738 struct super_block *sb = root->sb;
3740 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3744 /* Grab a reference on the superblock so the hierarchy doesn't
3745 * get deleted on unmount if there are child cgroups. This
3746 * can be done outside cgroup_mutex, since the sb can't
3747 * disappear while someone has an open control file on the
3749 atomic_inc(&sb->s_active);
3751 mutex_lock(&cgroup_mutex);
3753 init_cgroup_housekeeping(cgrp);
3755 cgrp->parent = parent;
3756 cgrp->root = parent->root;
3757 cgrp->top_cgroup = parent->top_cgroup;
3759 if (notify_on_release(parent))
3760 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3762 if (clone_children(parent))
3763 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3765 for_each_subsys(root, ss) {
3766 struct cgroup_subsys_state *css = ss->create(cgrp);
3772 init_cgroup_css(css, ss, cgrp);
3774 err = alloc_css_id(ss, parent, cgrp);
3778 /* At error, ->destroy() callback has to free assigned ID. */
3779 if (clone_children(parent) && ss->post_clone)
3780 ss->post_clone(cgrp);
3783 cgroup_lock_hierarchy(root);
3784 list_add(&cgrp->sibling, &cgrp->parent->children);
3785 cgroup_unlock_hierarchy(root);
3786 root->number_of_cgroups++;
3788 err = cgroup_create_dir(cgrp, dentry, mode);
3792 /* The cgroup directory was pre-locked for us */
3793 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3795 err = cgroup_populate_dir(cgrp);
3796 /* If err < 0, we have a half-filled directory - oh well ;) */
3798 mutex_unlock(&cgroup_mutex);
3799 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3805 cgroup_lock_hierarchy(root);
3806 list_del(&cgrp->sibling);
3807 cgroup_unlock_hierarchy(root);
3808 root->number_of_cgroups--;
3812 for_each_subsys(root, ss) {
3813 if (cgrp->subsys[ss->subsys_id])
3817 mutex_unlock(&cgroup_mutex);
3819 /* Release the reference count that we took on the superblock */
3820 deactivate_super(sb);
3826 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3828 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3830 /* the vfs holds inode->i_mutex already */
3831 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3834 static int cgroup_has_css_refs(struct cgroup *cgrp)
3836 /* Check the reference count on each subsystem. Since we
3837 * already established that there are no tasks in the
3838 * cgroup, if the css refcount is also 1, then there should
3839 * be no outstanding references, so the subsystem is safe to
3840 * destroy. We scan across all subsystems rather than using
3841 * the per-hierarchy linked list of mounted subsystems since
3842 * we can be called via check_for_release() with no
3843 * synchronization other than RCU, and the subsystem linked
3844 * list isn't RCU-safe */
3847 * We won't need to lock the subsys array, because the subsystems
3848 * we're concerned about aren't going anywhere since our cgroup root
3849 * has a reference on them.
3851 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3852 struct cgroup_subsys *ss = subsys[i];
3853 struct cgroup_subsys_state *css;
3854 /* Skip subsystems not present or not in this hierarchy */
3855 if (ss == NULL || ss->root != cgrp->root)
3857 css = cgrp->subsys[ss->subsys_id];
3858 /* When called from check_for_release() it's possible
3859 * that by this point the cgroup has been removed
3860 * and the css deleted. But a false-positive doesn't
3861 * matter, since it can only happen if the cgroup
3862 * has been deleted and hence no longer needs the
3863 * release agent to be called anyway. */
3864 if (css && (atomic_read(&css->refcnt) > 1))
3871 * Atomically mark all (or else none) of the cgroup's CSS objects as
3872 * CSS_REMOVED. Return true on success, or false if the cgroup has
3873 * busy subsystems. Call with cgroup_mutex held
3876 static int cgroup_clear_css_refs(struct cgroup *cgrp)
3878 struct cgroup_subsys *ss;
3879 unsigned long flags;
3880 bool failed = false;
3881 local_irq_save(flags);
3882 for_each_subsys(cgrp->root, ss) {
3883 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3886 /* We can only remove a CSS with a refcnt==1 */
3887 refcnt = atomic_read(&css->refcnt);
3894 * Drop the refcnt to 0 while we check other
3895 * subsystems. This will cause any racing
3896 * css_tryget() to spin until we set the
3897 * CSS_REMOVED bits or abort
3899 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3905 for_each_subsys(cgrp->root, ss) {
3906 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3909 * Restore old refcnt if we previously managed
3910 * to clear it from 1 to 0
3912 if (!atomic_read(&css->refcnt))
3913 atomic_set(&css->refcnt, 1);
3915 /* Commit the fact that the CSS is removed */
3916 set_bit(CSS_REMOVED, &css->flags);
3919 local_irq_restore(flags);
3923 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3925 struct cgroup *cgrp = dentry->d_fsdata;
3927 struct cgroup *parent;
3929 struct cgroup_event *event, *tmp;
3932 /* the vfs holds both inode->i_mutex already */
3934 mutex_lock(&cgroup_mutex);
3935 if (atomic_read(&cgrp->count) != 0) {
3936 mutex_unlock(&cgroup_mutex);
3939 if (!list_empty(&cgrp->children)) {
3940 mutex_unlock(&cgroup_mutex);
3943 mutex_unlock(&cgroup_mutex);
3946 * In general, subsystem has no css->refcnt after pre_destroy(). But
3947 * in racy cases, subsystem may have to get css->refcnt after
3948 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3949 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3950 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3951 * and subsystem's reference count handling. Please see css_get/put
3952 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3954 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3957 * Call pre_destroy handlers of subsys. Notify subsystems
3958 * that rmdir() request comes.
3960 ret = cgroup_call_pre_destroy(cgrp);
3962 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3966 mutex_lock(&cgroup_mutex);
3967 parent = cgrp->parent;
3968 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
3969 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3970 mutex_unlock(&cgroup_mutex);
3973 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
3974 if (!cgroup_clear_css_refs(cgrp)) {
3975 mutex_unlock(&cgroup_mutex);
3977 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3978 * prepare_to_wait(), we need to check this flag.
3980 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
3982 finish_wait(&cgroup_rmdir_waitq, &wait);
3983 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3984 if (signal_pending(current))
3988 /* NO css_tryget() can success after here. */
3989 finish_wait(&cgroup_rmdir_waitq, &wait);
3990 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3992 raw_spin_lock(&release_list_lock);
3993 set_bit(CGRP_REMOVED, &cgrp->flags);
3994 if (!list_empty(&cgrp->release_list))
3995 list_del_init(&cgrp->release_list);
3996 raw_spin_unlock(&release_list_lock);
3998 cgroup_lock_hierarchy(cgrp->root);
3999 /* delete this cgroup from parent->children */
4000 list_del_init(&cgrp->sibling);
4001 cgroup_unlock_hierarchy(cgrp->root);
4003 d = dget(cgrp->dentry);
4005 cgroup_d_remove_dir(d);
4008 set_bit(CGRP_RELEASABLE, &parent->flags);
4009 check_for_release(parent);
4012 * Unregister events and notify userspace.
4013 * Notify userspace about cgroup removing only after rmdir of cgroup
4014 * directory to avoid race between userspace and kernelspace
4016 spin_lock(&cgrp->event_list_lock);
4017 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4018 list_del(&event->list);
4019 remove_wait_queue(event->wqh, &event->wait);
4020 eventfd_signal(event->eventfd, 1);
4021 schedule_work(&event->remove);
4023 spin_unlock(&cgrp->event_list_lock);
4025 mutex_unlock(&cgroup_mutex);
4029 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4031 struct cgroup_subsys_state *css;
4033 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4035 /* Create the top cgroup state for this subsystem */
4036 list_add(&ss->sibling, &rootnode.subsys_list);
4037 ss->root = &rootnode;
4038 css = ss->create(dummytop);
4039 /* We don't handle early failures gracefully */
4040 BUG_ON(IS_ERR(css));
4041 init_cgroup_css(css, ss, dummytop);
4043 /* Update the init_css_set to contain a subsys
4044 * pointer to this state - since the subsystem is
4045 * newly registered, all tasks and hence the
4046 * init_css_set is in the subsystem's top cgroup. */
4047 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4049 need_forkexit_callback |= ss->fork || ss->exit;
4051 /* At system boot, before all subsystems have been
4052 * registered, no tasks have been forked, so we don't
4053 * need to invoke fork callbacks here. */
4054 BUG_ON(!list_empty(&init_task.tasks));
4056 mutex_init(&ss->hierarchy_mutex);
4057 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4060 /* this function shouldn't be used with modular subsystems, since they
4061 * need to register a subsys_id, among other things */
4066 * cgroup_load_subsys: load and register a modular subsystem at runtime
4067 * @ss: the subsystem to load
4069 * This function should be called in a modular subsystem's initcall. If the
4070 * subsystem is built as a module, it will be assigned a new subsys_id and set
4071 * up for use. If the subsystem is built-in anyway, work is delegated to the
4072 * simpler cgroup_init_subsys.
4074 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4077 struct cgroup_subsys_state *css;
4079 /* check name and function validity */
4080 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4081 ss->create == NULL || ss->destroy == NULL)
4085 * we don't support callbacks in modular subsystems. this check is
4086 * before the ss->module check for consistency; a subsystem that could
4087 * be a module should still have no callbacks even if the user isn't
4088 * compiling it as one.
4090 if (ss->fork || ss->exit)
4094 * an optionally modular subsystem is built-in: we want to do nothing,
4095 * since cgroup_init_subsys will have already taken care of it.
4097 if (ss->module == NULL) {
4098 /* a few sanity checks */
4099 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4100 BUG_ON(subsys[ss->subsys_id] != ss);
4105 * need to register a subsys id before anything else - for example,
4106 * init_cgroup_css needs it.
4108 mutex_lock(&cgroup_mutex);
4109 /* find the first empty slot in the array */
4110 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4111 if (subsys[i] == NULL)
4114 if (i == CGROUP_SUBSYS_COUNT) {
4115 /* maximum number of subsystems already registered! */
4116 mutex_unlock(&cgroup_mutex);
4119 /* assign ourselves the subsys_id */
4124 * no ss->create seems to need anything important in the ss struct, so
4125 * this can happen first (i.e. before the rootnode attachment).
4127 css = ss->create(dummytop);
4129 /* failure case - need to deassign the subsys[] slot. */
4131 mutex_unlock(&cgroup_mutex);
4132 return PTR_ERR(css);
4135 list_add(&ss->sibling, &rootnode.subsys_list);
4136 ss->root = &rootnode;
4138 /* our new subsystem will be attached to the dummy hierarchy. */
4139 init_cgroup_css(css, ss, dummytop);
4140 /* init_idr must be after init_cgroup_css because it sets css->id. */
4142 int ret = cgroup_init_idr(ss, css);
4144 dummytop->subsys[ss->subsys_id] = NULL;
4145 ss->destroy(dummytop);
4147 mutex_unlock(&cgroup_mutex);
4153 * Now we need to entangle the css into the existing css_sets. unlike
4154 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4155 * will need a new pointer to it; done by iterating the css_set_table.
4156 * furthermore, modifying the existing css_sets will corrupt the hash
4157 * table state, so each changed css_set will need its hash recomputed.
4158 * this is all done under the css_set_lock.
4160 write_lock(&css_set_lock);
4161 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4163 struct hlist_node *node, *tmp;
4164 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4166 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4167 /* skip entries that we already rehashed */
4168 if (cg->subsys[ss->subsys_id])
4170 /* remove existing entry */
4171 hlist_del(&cg->hlist);
4173 cg->subsys[ss->subsys_id] = css;
4174 /* recompute hash and restore entry */
4175 new_bucket = css_set_hash(cg->subsys);
4176 hlist_add_head(&cg->hlist, new_bucket);
4179 write_unlock(&css_set_lock);
4181 mutex_init(&ss->hierarchy_mutex);
4182 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4186 mutex_unlock(&cgroup_mutex);
4189 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4192 * cgroup_unload_subsys: unload a modular subsystem
4193 * @ss: the subsystem to unload
4195 * This function should be called in a modular subsystem's exitcall. When this
4196 * function is invoked, the refcount on the subsystem's module will be 0, so
4197 * the subsystem will not be attached to any hierarchy.
4199 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4201 struct cg_cgroup_link *link;
4202 struct hlist_head *hhead;
4204 BUG_ON(ss->module == NULL);
4207 * we shouldn't be called if the subsystem is in use, and the use of
4208 * try_module_get in parse_cgroupfs_options should ensure that it
4209 * doesn't start being used while we're killing it off.
4211 BUG_ON(ss->root != &rootnode);
4213 mutex_lock(&cgroup_mutex);
4214 /* deassign the subsys_id */
4215 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4216 subsys[ss->subsys_id] = NULL;
4218 /* remove subsystem from rootnode's list of subsystems */
4219 list_del_init(&ss->sibling);
4222 * disentangle the css from all css_sets attached to the dummytop. as
4223 * in loading, we need to pay our respects to the hashtable gods.
4225 write_lock(&css_set_lock);
4226 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4227 struct css_set *cg = link->cg;
4229 hlist_del(&cg->hlist);
4230 BUG_ON(!cg->subsys[ss->subsys_id]);
4231 cg->subsys[ss->subsys_id] = NULL;
4232 hhead = css_set_hash(cg->subsys);
4233 hlist_add_head(&cg->hlist, hhead);
4235 write_unlock(&css_set_lock);
4238 * remove subsystem's css from the dummytop and free it - need to free
4239 * before marking as null because ss->destroy needs the cgrp->subsys
4240 * pointer to find their state. note that this also takes care of
4241 * freeing the css_id.
4243 ss->destroy(dummytop);
4244 dummytop->subsys[ss->subsys_id] = NULL;
4246 mutex_unlock(&cgroup_mutex);
4248 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4251 * cgroup_init_early - cgroup initialization at system boot
4253 * Initialize cgroups at system boot, and initialize any
4254 * subsystems that request early init.
4256 int __init cgroup_init_early(void)
4259 atomic_set(&init_css_set.refcount, 1);
4260 INIT_LIST_HEAD(&init_css_set.cg_links);
4261 INIT_LIST_HEAD(&init_css_set.tasks);
4262 INIT_HLIST_NODE(&init_css_set.hlist);
4264 init_cgroup_root(&rootnode);
4266 init_task.cgroups = &init_css_set;
4268 init_css_set_link.cg = &init_css_set;
4269 init_css_set_link.cgrp = dummytop;
4270 list_add(&init_css_set_link.cgrp_link_list,
4271 &rootnode.top_cgroup.css_sets);
4272 list_add(&init_css_set_link.cg_link_list,
4273 &init_css_set.cg_links);
4275 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4276 INIT_HLIST_HEAD(&css_set_table[i]);
4278 /* at bootup time, we don't worry about modular subsystems */
4279 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4280 struct cgroup_subsys *ss = subsys[i];
4283 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4284 BUG_ON(!ss->create);
4285 BUG_ON(!ss->destroy);
4286 if (ss->subsys_id != i) {
4287 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4288 ss->name, ss->subsys_id);
4293 cgroup_init_subsys(ss);
4299 * cgroup_init - cgroup initialization
4301 * Register cgroup filesystem and /proc file, and initialize
4302 * any subsystems that didn't request early init.
4304 int __init cgroup_init(void)
4308 struct hlist_head *hhead;
4310 err = bdi_init(&cgroup_backing_dev_info);
4314 /* at bootup time, we don't worry about modular subsystems */
4315 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4316 struct cgroup_subsys *ss = subsys[i];
4317 if (!ss->early_init)
4318 cgroup_init_subsys(ss);
4320 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4323 /* Add init_css_set to the hash table */
4324 hhead = css_set_hash(init_css_set.subsys);
4325 hlist_add_head(&init_css_set.hlist, hhead);
4326 BUG_ON(!init_root_id(&rootnode));
4328 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4334 err = register_filesystem(&cgroup_fs_type);
4336 kobject_put(cgroup_kobj);
4340 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4344 bdi_destroy(&cgroup_backing_dev_info);
4350 * proc_cgroup_show()
4351 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4352 * - Used for /proc/<pid>/cgroup.
4353 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4354 * doesn't really matter if tsk->cgroup changes after we read it,
4355 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4356 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4357 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4358 * cgroup to top_cgroup.
4361 /* TODO: Use a proper seq_file iterator */
4362 static int proc_cgroup_show(struct seq_file *m, void *v)
4365 struct task_struct *tsk;
4368 struct cgroupfs_root *root;
4371 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4377 tsk = get_pid_task(pid, PIDTYPE_PID);
4383 mutex_lock(&cgroup_mutex);
4385 for_each_active_root(root) {
4386 struct cgroup_subsys *ss;
4387 struct cgroup *cgrp;
4390 seq_printf(m, "%d:", root->hierarchy_id);
4391 for_each_subsys(root, ss)
4392 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4393 if (strlen(root->name))
4394 seq_printf(m, "%sname=%s", count ? "," : "",
4397 cgrp = task_cgroup_from_root(tsk, root);
4398 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4406 mutex_unlock(&cgroup_mutex);
4407 put_task_struct(tsk);
4414 static int cgroup_open(struct inode *inode, struct file *file)
4416 struct pid *pid = PROC_I(inode)->pid;
4417 return single_open(file, proc_cgroup_show, pid);
4420 const struct file_operations proc_cgroup_operations = {
4421 .open = cgroup_open,
4423 .llseek = seq_lseek,
4424 .release = single_release,
4427 /* Display information about each subsystem and each hierarchy */
4428 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4432 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4434 * ideally we don't want subsystems moving around while we do this.
4435 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4436 * subsys/hierarchy state.
4438 mutex_lock(&cgroup_mutex);
4439 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4440 struct cgroup_subsys *ss = subsys[i];
4443 seq_printf(m, "%s\t%d\t%d\t%d\n",
4444 ss->name, ss->root->hierarchy_id,
4445 ss->root->number_of_cgroups, !ss->disabled);
4447 mutex_unlock(&cgroup_mutex);
4451 static int cgroupstats_open(struct inode *inode, struct file *file)
4453 return single_open(file, proc_cgroupstats_show, NULL);
4456 static const struct file_operations proc_cgroupstats_operations = {
4457 .open = cgroupstats_open,
4459 .llseek = seq_lseek,
4460 .release = single_release,
4464 * cgroup_fork - attach newly forked task to its parents cgroup.
4465 * @child: pointer to task_struct of forking parent process.
4467 * Description: A task inherits its parent's cgroup at fork().
4469 * A pointer to the shared css_set was automatically copied in
4470 * fork.c by dup_task_struct(). However, we ignore that copy, since
4471 * it was not made under the protection of RCU or cgroup_mutex, so
4472 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4473 * have already changed current->cgroups, allowing the previously
4474 * referenced cgroup group to be removed and freed.
4476 * At the point that cgroup_fork() is called, 'current' is the parent
4477 * task, and the passed argument 'child' points to the child task.
4479 void cgroup_fork(struct task_struct *child)
4482 child->cgroups = current->cgroups;
4483 get_css_set(child->cgroups);
4484 task_unlock(current);
4485 INIT_LIST_HEAD(&child->cg_list);
4489 * cgroup_fork_callbacks - run fork callbacks
4490 * @child: the new task
4492 * Called on a new task very soon before adding it to the
4493 * tasklist. No need to take any locks since no-one can
4494 * be operating on this task.
4496 void cgroup_fork_callbacks(struct task_struct *child)
4498 if (need_forkexit_callback) {
4501 * forkexit callbacks are only supported for builtin
4502 * subsystems, and the builtin section of the subsys array is
4503 * immutable, so we don't need to lock the subsys array here.
4505 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4506 struct cgroup_subsys *ss = subsys[i];
4514 * cgroup_post_fork - called on a new task after adding it to the task list
4515 * @child: the task in question
4517 * Adds the task to the list running through its css_set if necessary.
4518 * Has to be after the task is visible on the task list in case we race
4519 * with the first call to cgroup_iter_start() - to guarantee that the
4520 * new task ends up on its list.
4522 void cgroup_post_fork(struct task_struct *child)
4525 * use_task_css_set_links is set to 1 before we walk the tasklist
4526 * under the tasklist_lock and we read it here after we added the child
4527 * to the tasklist under the tasklist_lock as well. If the child wasn't
4528 * yet in the tasklist when we walked through it from
4529 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4530 * should be visible now due to the paired locking and barriers implied
4531 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4532 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4535 if (use_task_css_set_links) {
4536 write_lock(&css_set_lock);
4538 if (list_empty(&child->cg_list))
4539 list_add(&child->cg_list, &child->cgroups->tasks);
4541 write_unlock(&css_set_lock);
4545 * cgroup_exit - detach cgroup from exiting task
4546 * @tsk: pointer to task_struct of exiting process
4547 * @run_callback: run exit callbacks?
4549 * Description: Detach cgroup from @tsk and release it.
4551 * Note that cgroups marked notify_on_release force every task in
4552 * them to take the global cgroup_mutex mutex when exiting.
4553 * This could impact scaling on very large systems. Be reluctant to
4554 * use notify_on_release cgroups where very high task exit scaling
4555 * is required on large systems.
4557 * the_top_cgroup_hack:
4559 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4561 * We call cgroup_exit() while the task is still competent to
4562 * handle notify_on_release(), then leave the task attached to the
4563 * root cgroup in each hierarchy for the remainder of its exit.
4565 * To do this properly, we would increment the reference count on
4566 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4567 * code we would add a second cgroup function call, to drop that
4568 * reference. This would just create an unnecessary hot spot on
4569 * the top_cgroup reference count, to no avail.
4571 * Normally, holding a reference to a cgroup without bumping its
4572 * count is unsafe. The cgroup could go away, or someone could
4573 * attach us to a different cgroup, decrementing the count on
4574 * the first cgroup that we never incremented. But in this case,
4575 * top_cgroup isn't going away, and either task has PF_EXITING set,
4576 * which wards off any cgroup_attach_task() attempts, or task is a failed
4577 * fork, never visible to cgroup_attach_task.
4579 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4585 * Unlink from the css_set task list if necessary.
4586 * Optimistically check cg_list before taking
4589 if (!list_empty(&tsk->cg_list)) {
4590 write_lock(&css_set_lock);
4591 if (!list_empty(&tsk->cg_list))
4592 list_del_init(&tsk->cg_list);
4593 write_unlock(&css_set_lock);
4596 /* Reassign the task to the init_css_set. */
4599 tsk->cgroups = &init_css_set;
4601 if (run_callbacks && need_forkexit_callback) {
4603 * modular subsystems can't use callbacks, so no need to lock
4606 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4607 struct cgroup_subsys *ss = subsys[i];
4609 struct cgroup *old_cgrp =
4610 rcu_dereference_raw(cg->subsys[i])->cgroup;
4611 struct cgroup *cgrp = task_cgroup(tsk, i);
4612 ss->exit(cgrp, old_cgrp, tsk);
4619 put_css_set_taskexit(cg);
4623 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4624 * @cgrp: the cgroup in question
4625 * @task: the task in question
4627 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4630 * If we are sending in dummytop, then presumably we are creating
4631 * the top cgroup in the subsystem.
4633 * Called only by the ns (nsproxy) cgroup.
4635 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4638 struct cgroup *target;
4640 if (cgrp == dummytop)
4643 target = task_cgroup_from_root(task, cgrp->root);
4644 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4645 cgrp = cgrp->parent;
4646 ret = (cgrp == target);
4650 static void check_for_release(struct cgroup *cgrp)
4652 /* All of these checks rely on RCU to keep the cgroup
4653 * structure alive */
4654 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4655 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4656 /* Control Group is currently removeable. If it's not
4657 * already queued for a userspace notification, queue
4659 int need_schedule_work = 0;
4660 raw_spin_lock(&release_list_lock);
4661 if (!cgroup_is_removed(cgrp) &&
4662 list_empty(&cgrp->release_list)) {
4663 list_add(&cgrp->release_list, &release_list);
4664 need_schedule_work = 1;
4666 raw_spin_unlock(&release_list_lock);
4667 if (need_schedule_work)
4668 schedule_work(&release_agent_work);
4672 /* Caller must verify that the css is not for root cgroup */
4673 void __css_put(struct cgroup_subsys_state *css, int count)
4675 struct cgroup *cgrp = css->cgroup;
4678 val = atomic_sub_return(count, &css->refcnt);
4680 if (notify_on_release(cgrp)) {
4681 set_bit(CGRP_RELEASABLE, &cgrp->flags);
4682 check_for_release(cgrp);
4684 cgroup_wakeup_rmdir_waiter(cgrp);
4687 WARN_ON_ONCE(val < 1);
4689 EXPORT_SYMBOL_GPL(__css_put);
4692 * Notify userspace when a cgroup is released, by running the
4693 * configured release agent with the name of the cgroup (path
4694 * relative to the root of cgroup file system) as the argument.
4696 * Most likely, this user command will try to rmdir this cgroup.
4698 * This races with the possibility that some other task will be
4699 * attached to this cgroup before it is removed, or that some other
4700 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4701 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4702 * unused, and this cgroup will be reprieved from its death sentence,
4703 * to continue to serve a useful existence. Next time it's released,
4704 * we will get notified again, if it still has 'notify_on_release' set.
4706 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4707 * means only wait until the task is successfully execve()'d. The
4708 * separate release agent task is forked by call_usermodehelper(),
4709 * then control in this thread returns here, without waiting for the
4710 * release agent task. We don't bother to wait because the caller of
4711 * this routine has no use for the exit status of the release agent
4712 * task, so no sense holding our caller up for that.
4714 static void cgroup_release_agent(struct work_struct *work)
4716 BUG_ON(work != &release_agent_work);
4717 mutex_lock(&cgroup_mutex);
4718 raw_spin_lock(&release_list_lock);
4719 while (!list_empty(&release_list)) {
4720 char *argv[3], *envp[3];
4722 char *pathbuf = NULL, *agentbuf = NULL;
4723 struct cgroup *cgrp = list_entry(release_list.next,
4726 list_del_init(&cgrp->release_list);
4727 raw_spin_unlock(&release_list_lock);
4728 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4731 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4733 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4738 argv[i++] = agentbuf;
4739 argv[i++] = pathbuf;
4743 /* minimal command environment */
4744 envp[i++] = "HOME=/";
4745 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4748 /* Drop the lock while we invoke the usermode helper,
4749 * since the exec could involve hitting disk and hence
4750 * be a slow process */
4751 mutex_unlock(&cgroup_mutex);
4752 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4753 mutex_lock(&cgroup_mutex);
4757 raw_spin_lock(&release_list_lock);
4759 raw_spin_unlock(&release_list_lock);
4760 mutex_unlock(&cgroup_mutex);
4763 static int __init cgroup_disable(char *str)
4768 while ((token = strsep(&str, ",")) != NULL) {
4772 * cgroup_disable, being at boot time, can't know about module
4773 * subsystems, so we don't worry about them.
4775 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4776 struct cgroup_subsys *ss = subsys[i];
4778 if (!strcmp(token, ss->name)) {
4780 printk(KERN_INFO "Disabling %s control group"
4781 " subsystem\n", ss->name);
4788 __setup("cgroup_disable=", cgroup_disable);
4791 * Functons for CSS ID.
4795 *To get ID other than 0, this should be called when !cgroup_is_removed().
4797 unsigned short css_id(struct cgroup_subsys_state *css)
4799 struct css_id *cssid;
4802 * This css_id() can return correct value when somone has refcnt
4803 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4804 * it's unchanged until freed.
4806 cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt));
4812 EXPORT_SYMBOL_GPL(css_id);
4814 unsigned short css_depth(struct cgroup_subsys_state *css)
4816 struct css_id *cssid;
4818 cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt));
4821 return cssid->depth;
4824 EXPORT_SYMBOL_GPL(css_depth);
4827 * css_is_ancestor - test "root" css is an ancestor of "child"
4828 * @child: the css to be tested.
4829 * @root: the css supporsed to be an ancestor of the child.
4831 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4832 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4833 * But, considering usual usage, the csses should be valid objects after test.
4834 * Assuming that the caller will do some action to the child if this returns
4835 * returns true, the caller must take "child";s reference count.
4836 * If "child" is valid object and this returns true, "root" is valid, too.
4839 bool css_is_ancestor(struct cgroup_subsys_state *child,
4840 const struct cgroup_subsys_state *root)
4842 struct css_id *child_id;
4843 struct css_id *root_id;
4847 child_id = rcu_dereference(child->id);
4848 root_id = rcu_dereference(root->id);
4851 || (child_id->depth < root_id->depth)
4852 || (child_id->stack[root_id->depth] != root_id->id))
4858 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
4860 struct css_id *id = css->id;
4861 /* When this is called before css_id initialization, id can be NULL */
4865 BUG_ON(!ss->use_id);
4867 rcu_assign_pointer(id->css, NULL);
4868 rcu_assign_pointer(css->id, NULL);
4869 spin_lock(&ss->id_lock);
4870 idr_remove(&ss->idr, id->id);
4871 spin_unlock(&ss->id_lock);
4872 kfree_rcu(id, rcu_head);
4874 EXPORT_SYMBOL_GPL(free_css_id);
4877 * This is called by init or create(). Then, calls to this function are
4878 * always serialized (By cgroup_mutex() at create()).
4881 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
4883 struct css_id *newid;
4884 int myid, error, size;
4886 BUG_ON(!ss->use_id);
4888 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
4889 newid = kzalloc(size, GFP_KERNEL);
4891 return ERR_PTR(-ENOMEM);
4893 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
4897 spin_lock(&ss->id_lock);
4898 /* Don't use 0. allocates an ID of 1-65535 */
4899 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
4900 spin_unlock(&ss->id_lock);
4902 /* Returns error when there are no free spaces for new ID.*/
4907 if (myid > CSS_ID_MAX)
4911 newid->depth = depth;
4915 spin_lock(&ss->id_lock);
4916 idr_remove(&ss->idr, myid);
4917 spin_unlock(&ss->id_lock);
4920 return ERR_PTR(error);
4924 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
4925 struct cgroup_subsys_state *rootcss)
4927 struct css_id *newid;
4929 spin_lock_init(&ss->id_lock);
4932 newid = get_new_cssid(ss, 0);
4934 return PTR_ERR(newid);
4936 newid->stack[0] = newid->id;
4937 newid->css = rootcss;
4938 rootcss->id = newid;
4942 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
4943 struct cgroup *child)
4945 int subsys_id, i, depth = 0;
4946 struct cgroup_subsys_state *parent_css, *child_css;
4947 struct css_id *child_id, *parent_id;
4949 subsys_id = ss->subsys_id;
4950 parent_css = parent->subsys[subsys_id];
4951 child_css = child->subsys[subsys_id];
4952 parent_id = parent_css->id;
4953 depth = parent_id->depth + 1;
4955 child_id = get_new_cssid(ss, depth);
4956 if (IS_ERR(child_id))
4957 return PTR_ERR(child_id);
4959 for (i = 0; i < depth; i++)
4960 child_id->stack[i] = parent_id->stack[i];
4961 child_id->stack[depth] = child_id->id;
4963 * child_id->css pointer will be set after this cgroup is available
4964 * see cgroup_populate_dir()
4966 rcu_assign_pointer(child_css->id, child_id);
4972 * css_lookup - lookup css by id
4973 * @ss: cgroup subsys to be looked into.
4976 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4977 * NULL if not. Should be called under rcu_read_lock()
4979 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
4981 struct css_id *cssid = NULL;
4983 BUG_ON(!ss->use_id);
4984 cssid = idr_find(&ss->idr, id);
4986 if (unlikely(!cssid))
4989 return rcu_dereference(cssid->css);
4991 EXPORT_SYMBOL_GPL(css_lookup);
4994 * css_get_next - lookup next cgroup under specified hierarchy.
4995 * @ss: pointer to subsystem
4996 * @id: current position of iteration.
4997 * @root: pointer to css. search tree under this.
4998 * @foundid: position of found object.
5000 * Search next css under the specified hierarchy of rootid. Calling under
5001 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5003 struct cgroup_subsys_state *
5004 css_get_next(struct cgroup_subsys *ss, int id,
5005 struct cgroup_subsys_state *root, int *foundid)
5007 struct cgroup_subsys_state *ret = NULL;
5010 int rootid = css_id(root);
5011 int depth = css_depth(root);
5016 BUG_ON(!ss->use_id);
5017 WARN_ON_ONCE(!rcu_read_lock_held());
5019 /* fill start point for scan */
5023 * scan next entry from bitmap(tree), tmpid is updated after
5026 tmp = idr_get_next(&ss->idr, &tmpid);
5029 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5030 ret = rcu_dereference(tmp->css);
5036 /* continue to scan from next id */
5043 * get corresponding css from file open on cgroupfs directory
5045 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5047 struct cgroup *cgrp;
5048 struct inode *inode;
5049 struct cgroup_subsys_state *css;
5051 inode = f->f_dentry->d_inode;
5052 /* check in cgroup filesystem dir */
5053 if (inode->i_op != &cgroup_dir_inode_operations)
5054 return ERR_PTR(-EBADF);
5056 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5057 return ERR_PTR(-EINVAL);
5060 cgrp = __d_cgrp(f->f_dentry);
5061 css = cgrp->subsys[id];
5062 return css ? css : ERR_PTR(-ENOENT);
5065 #ifdef CONFIG_CGROUP_DEBUG
5066 static struct cgroup_subsys_state *debug_create(struct cgroup *cont)
5068 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5071 return ERR_PTR(-ENOMEM);
5076 static void debug_destroy(struct cgroup *cont)
5078 kfree(cont->subsys[debug_subsys_id]);
5081 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5083 return atomic_read(&cont->count);
5086 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5088 return cgroup_task_count(cont);
5091 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5093 return (u64)(unsigned long)current->cgroups;
5096 static u64 current_css_set_refcount_read(struct cgroup *cont,
5102 count = atomic_read(¤t->cgroups->refcount);
5107 static int current_css_set_cg_links_read(struct cgroup *cont,
5109 struct seq_file *seq)
5111 struct cg_cgroup_link *link;
5114 read_lock(&css_set_lock);
5116 cg = rcu_dereference(current->cgroups);
5117 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5118 struct cgroup *c = link->cgrp;
5122 name = c->dentry->d_name.name;
5125 seq_printf(seq, "Root %d group %s\n",
5126 c->root->hierarchy_id, name);
5129 read_unlock(&css_set_lock);
5133 #define MAX_TASKS_SHOWN_PER_CSS 25
5134 static int cgroup_css_links_read(struct cgroup *cont,
5136 struct seq_file *seq)
5138 struct cg_cgroup_link *link;
5140 read_lock(&css_set_lock);
5141 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5142 struct css_set *cg = link->cg;
5143 struct task_struct *task;
5145 seq_printf(seq, "css_set %p\n", cg);
5146 list_for_each_entry(task, &cg->tasks, cg_list) {
5147 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5148 seq_puts(seq, " ...\n");
5151 seq_printf(seq, " task %d\n",
5152 task_pid_vnr(task));
5156 read_unlock(&css_set_lock);
5160 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5162 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5165 static struct cftype debug_files[] = {
5167 .name = "cgroup_refcount",
5168 .read_u64 = cgroup_refcount_read,
5171 .name = "taskcount",
5172 .read_u64 = debug_taskcount_read,
5176 .name = "current_css_set",
5177 .read_u64 = current_css_set_read,
5181 .name = "current_css_set_refcount",
5182 .read_u64 = current_css_set_refcount_read,
5186 .name = "current_css_set_cg_links",
5187 .read_seq_string = current_css_set_cg_links_read,
5191 .name = "cgroup_css_links",
5192 .read_seq_string = cgroup_css_links_read,
5196 .name = "releasable",
5197 .read_u64 = releasable_read,
5201 static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
5203 return cgroup_add_files(cont, ss, debug_files,
5204 ARRAY_SIZE(debug_files));
5207 struct cgroup_subsys debug_subsys = {
5209 .create = debug_create,
5210 .destroy = debug_destroy,
5211 .populate = debug_populate,
5212 .subsys_id = debug_subsys_id,
5214 #endif /* CONFIG_CGROUP_DEBUG */