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>
66 static DEFINE_MUTEX(cgroup_mutex);
69 * Generate an array of cgroup subsystem pointers. At boot time, this is
70 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
71 * registered after that. The mutable section of this array is protected by
74 #define SUBSYS(_x) &_x ## _subsys,
75 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
76 #include <linux/cgroup_subsys.h>
79 #define MAX_CGROUP_ROOT_NAMELEN 64
82 * A cgroupfs_root represents the root of a cgroup hierarchy,
83 * and may be associated with a superblock to form an active
86 struct cgroupfs_root {
87 struct super_block *sb;
90 * The bitmask of subsystems intended to be attached to this
93 unsigned long subsys_bits;
95 /* Unique id for this hierarchy. */
98 /* The bitmask of subsystems currently attached to this hierarchy */
99 unsigned long actual_subsys_bits;
101 /* A list running through the attached subsystems */
102 struct list_head subsys_list;
104 /* The root cgroup for this hierarchy */
105 struct cgroup top_cgroup;
107 /* Tracks how many cgroups are currently defined in hierarchy.*/
108 int number_of_cgroups;
110 /* A list running through the active hierarchies */
111 struct list_head root_list;
113 /* Hierarchy-specific flags */
116 /* The path to use for release notifications. */
117 char release_agent_path[PATH_MAX];
119 /* The name for this hierarchy - may be empty */
120 char name[MAX_CGROUP_ROOT_NAMELEN];
124 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
125 * subsystems that are otherwise unattached - it never has more than a
126 * single cgroup, and all tasks are part of that cgroup.
128 static struct cgroupfs_root rootnode;
131 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
132 * cgroup_subsys->use_id != 0.
134 #define CSS_ID_MAX (65535)
137 * The css to which this ID points. This pointer is set to valid value
138 * after cgroup is populated. If cgroup is removed, this will be NULL.
139 * This pointer is expected to be RCU-safe because destroy()
140 * is called after synchronize_rcu(). But for safe use, css_is_removed()
141 * css_tryget() should be used for avoiding race.
143 struct cgroup_subsys_state __rcu *css;
149 * Depth in hierarchy which this ID belongs to.
151 unsigned short depth;
153 * ID is freed by RCU. (and lookup routine is RCU safe.)
155 struct rcu_head rcu_head;
157 * Hierarchy of CSS ID belongs to.
159 unsigned short stack[0]; /* Array of Length (depth+1) */
163 * cgroup_event represents events which userspace want to receive.
165 struct cgroup_event {
167 * Cgroup which the event belongs to.
171 * Control file which the event associated.
175 * eventfd to signal userspace about the event.
177 struct eventfd_ctx *eventfd;
179 * Each of these stored in a list by the cgroup.
181 struct list_head list;
183 * All fields below needed to unregister event when
184 * userspace closes eventfd.
187 wait_queue_head_t *wqh;
189 struct work_struct remove;
192 /* The list of hierarchy roots */
194 static LIST_HEAD(roots);
195 static int root_count;
197 static DEFINE_IDA(hierarchy_ida);
198 static int next_hierarchy_id;
199 static DEFINE_SPINLOCK(hierarchy_id_lock);
201 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
202 #define dummytop (&rootnode.top_cgroup)
204 /* This flag indicates whether tasks in the fork and exit paths should
205 * check for fork/exit handlers to call. This avoids us having to do
206 * extra work in the fork/exit path if none of the subsystems need to
209 static int need_forkexit_callback __read_mostly;
211 #ifdef CONFIG_PROVE_LOCKING
212 int cgroup_lock_is_held(void)
214 return lockdep_is_held(&cgroup_mutex);
216 #else /* #ifdef CONFIG_PROVE_LOCKING */
217 int cgroup_lock_is_held(void)
219 return mutex_is_locked(&cgroup_mutex);
221 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
223 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
225 /* convenient tests for these bits */
226 inline int cgroup_is_removed(const struct cgroup *cgrp)
228 return test_bit(CGRP_REMOVED, &cgrp->flags);
231 /* bits in struct cgroupfs_root flags field */
233 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
236 static int cgroup_is_releasable(const struct cgroup *cgrp)
239 (1 << CGRP_RELEASABLE) |
240 (1 << CGRP_NOTIFY_ON_RELEASE);
241 return (cgrp->flags & bits) == bits;
244 static int notify_on_release(const struct cgroup *cgrp)
246 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
249 static int clone_children(const struct cgroup *cgrp)
251 return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
255 * for_each_subsys() allows you to iterate on each subsystem attached to
256 * an active hierarchy
258 #define for_each_subsys(_root, _ss) \
259 list_for_each_entry(_ss, &_root->subsys_list, sibling)
261 /* for_each_active_root() allows you to iterate across the active hierarchies */
262 #define for_each_active_root(_root) \
263 list_for_each_entry(_root, &roots, root_list)
265 /* the list of cgroups eligible for automatic release. Protected by
266 * release_list_lock */
267 static LIST_HEAD(release_list);
268 static DEFINE_RAW_SPINLOCK(release_list_lock);
269 static void cgroup_release_agent(struct work_struct *work);
270 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
271 static void check_for_release(struct cgroup *cgrp);
273 /* Link structure for associating css_set objects with cgroups */
274 struct cg_cgroup_link {
276 * List running through cg_cgroup_links associated with a
277 * cgroup, anchored on cgroup->css_sets
279 struct list_head cgrp_link_list;
282 * List running through cg_cgroup_links pointing at a
283 * single css_set object, anchored on css_set->cg_links
285 struct list_head cg_link_list;
289 /* The default css_set - used by init and its children prior to any
290 * hierarchies being mounted. It contains a pointer to the root state
291 * for each subsystem. Also used to anchor the list of css_sets. Not
292 * reference-counted, to improve performance when child cgroups
293 * haven't been created.
296 static struct css_set init_css_set;
297 static struct cg_cgroup_link init_css_set_link;
299 static int cgroup_init_idr(struct cgroup_subsys *ss,
300 struct cgroup_subsys_state *css);
302 /* css_set_lock protects the list of css_set objects, and the
303 * chain of tasks off each css_set. Nests outside task->alloc_lock
304 * due to cgroup_iter_start() */
305 static DEFINE_RWLOCK(css_set_lock);
306 static int css_set_count;
309 * hash table for cgroup groups. This improves the performance to find
310 * an existing css_set. This hash doesn't (currently) take into
311 * account cgroups in empty hierarchies.
313 #define CSS_SET_HASH_BITS 7
314 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
315 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
317 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
321 unsigned long tmp = 0UL;
323 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
324 tmp += (unsigned long)css[i];
325 tmp = (tmp >> 16) ^ tmp;
327 index = hash_long(tmp, CSS_SET_HASH_BITS);
329 return &css_set_table[index];
332 /* We don't maintain the lists running through each css_set to its
333 * task until after the first call to cgroup_iter_start(). This
334 * reduces the fork()/exit() overhead for people who have cgroups
335 * compiled into their kernel but not actually in use */
336 static int use_task_css_set_links __read_mostly;
338 static void __put_css_set(struct css_set *cg, int taskexit)
340 struct cg_cgroup_link *link;
341 struct cg_cgroup_link *saved_link;
343 * Ensure that the refcount doesn't hit zero while any readers
344 * can see it. Similar to atomic_dec_and_lock(), but for an
347 if (atomic_add_unless(&cg->refcount, -1, 1))
349 write_lock(&css_set_lock);
350 if (!atomic_dec_and_test(&cg->refcount)) {
351 write_unlock(&css_set_lock);
355 /* This css_set is dead. unlink it and release cgroup refcounts */
356 hlist_del(&cg->hlist);
359 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
361 struct cgroup *cgrp = link->cgrp;
362 list_del(&link->cg_link_list);
363 list_del(&link->cgrp_link_list);
366 * We may not be holding cgroup_mutex, and if cgrp->count is
367 * dropped to 0 the cgroup can be destroyed at any time, hence
368 * rcu_read_lock is used to keep it alive.
371 if (atomic_dec_and_test(&cgrp->count) &&
372 notify_on_release(cgrp)) {
374 set_bit(CGRP_RELEASABLE, &cgrp->flags);
375 check_for_release(cgrp);
382 write_unlock(&css_set_lock);
383 kfree_rcu(cg, rcu_head);
387 * refcounted get/put for css_set objects
389 static inline void get_css_set(struct css_set *cg)
391 atomic_inc(&cg->refcount);
394 static inline void put_css_set(struct css_set *cg)
396 __put_css_set(cg, 0);
399 static inline void put_css_set_taskexit(struct css_set *cg)
401 __put_css_set(cg, 1);
405 * compare_css_sets - helper function for find_existing_css_set().
406 * @cg: candidate css_set being tested
407 * @old_cg: existing css_set for a task
408 * @new_cgrp: cgroup that's being entered by the task
409 * @template: desired set of css pointers in css_set (pre-calculated)
411 * Returns true if "cg" matches "old_cg" except for the hierarchy
412 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
414 static bool compare_css_sets(struct css_set *cg,
415 struct css_set *old_cg,
416 struct cgroup *new_cgrp,
417 struct cgroup_subsys_state *template[])
419 struct list_head *l1, *l2;
421 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
422 /* Not all subsystems matched */
427 * Compare cgroup pointers in order to distinguish between
428 * different cgroups in heirarchies with no subsystems. We
429 * could get by with just this check alone (and skip the
430 * memcmp above) but on most setups the memcmp check will
431 * avoid the need for this more expensive check on almost all
436 l2 = &old_cg->cg_links;
438 struct cg_cgroup_link *cgl1, *cgl2;
439 struct cgroup *cg1, *cg2;
443 /* See if we reached the end - both lists are equal length. */
444 if (l1 == &cg->cg_links) {
445 BUG_ON(l2 != &old_cg->cg_links);
448 BUG_ON(l2 == &old_cg->cg_links);
450 /* Locate the cgroups associated with these links. */
451 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
452 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
455 /* Hierarchies should be linked in the same order. */
456 BUG_ON(cg1->root != cg2->root);
459 * If this hierarchy is the hierarchy of the cgroup
460 * that's changing, then we need to check that this
461 * css_set points to the new cgroup; if it's any other
462 * hierarchy, then this css_set should point to the
463 * same cgroup as the old css_set.
465 if (cg1->root == new_cgrp->root) {
477 * find_existing_css_set() is a helper for
478 * find_css_set(), and checks to see whether an existing
479 * css_set is suitable.
481 * oldcg: the cgroup group that we're using before the cgroup
484 * cgrp: the cgroup that we're moving into
486 * template: location in which to build the desired set of subsystem
487 * state objects for the new cgroup group
489 static struct css_set *find_existing_css_set(
490 struct css_set *oldcg,
492 struct cgroup_subsys_state *template[])
495 struct cgroupfs_root *root = cgrp->root;
496 struct hlist_head *hhead;
497 struct hlist_node *node;
501 * Build the set of subsystem state objects that we want to see in the
502 * new css_set. while subsystems can change globally, the entries here
503 * won't change, so no need for locking.
505 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
506 if (root->subsys_bits & (1UL << i)) {
507 /* Subsystem is in this hierarchy. So we want
508 * the subsystem state from the new
510 template[i] = cgrp->subsys[i];
512 /* Subsystem is not in this hierarchy, so we
513 * don't want to change the subsystem state */
514 template[i] = oldcg->subsys[i];
518 hhead = css_set_hash(template);
519 hlist_for_each_entry(cg, node, hhead, hlist) {
520 if (!compare_css_sets(cg, oldcg, cgrp, template))
523 /* This css_set matches what we need */
527 /* No existing cgroup group matched */
531 static void free_cg_links(struct list_head *tmp)
533 struct cg_cgroup_link *link;
534 struct cg_cgroup_link *saved_link;
536 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
537 list_del(&link->cgrp_link_list);
543 * allocate_cg_links() allocates "count" cg_cgroup_link structures
544 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
545 * success or a negative error
547 static int allocate_cg_links(int count, struct list_head *tmp)
549 struct cg_cgroup_link *link;
552 for (i = 0; i < count; i++) {
553 link = kmalloc(sizeof(*link), GFP_KERNEL);
558 list_add(&link->cgrp_link_list, tmp);
564 * link_css_set - a helper function to link a css_set to a cgroup
565 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
566 * @cg: the css_set to be linked
567 * @cgrp: the destination cgroup
569 static void link_css_set(struct list_head *tmp_cg_links,
570 struct css_set *cg, struct cgroup *cgrp)
572 struct cg_cgroup_link *link;
574 BUG_ON(list_empty(tmp_cg_links));
575 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
579 atomic_inc(&cgrp->count);
580 list_move(&link->cgrp_link_list, &cgrp->css_sets);
582 * Always add links to the tail of the list so that the list
583 * is sorted by order of hierarchy creation
585 list_add_tail(&link->cg_link_list, &cg->cg_links);
589 * find_css_set() takes an existing cgroup group and a
590 * cgroup object, and returns a css_set object that's
591 * equivalent to the old group, but with the given cgroup
592 * substituted into the appropriate hierarchy. Must be called with
595 static struct css_set *find_css_set(
596 struct css_set *oldcg, struct cgroup *cgrp)
599 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
601 struct list_head tmp_cg_links;
603 struct hlist_head *hhead;
604 struct cg_cgroup_link *link;
606 /* First see if we already have a cgroup group that matches
608 read_lock(&css_set_lock);
609 res = find_existing_css_set(oldcg, cgrp, template);
612 read_unlock(&css_set_lock);
617 res = kmalloc(sizeof(*res), GFP_KERNEL);
621 /* Allocate all the cg_cgroup_link objects that we'll need */
622 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
627 atomic_set(&res->refcount, 1);
628 INIT_LIST_HEAD(&res->cg_links);
629 INIT_LIST_HEAD(&res->tasks);
630 INIT_HLIST_NODE(&res->hlist);
632 /* Copy the set of subsystem state objects generated in
633 * find_existing_css_set() */
634 memcpy(res->subsys, template, sizeof(res->subsys));
636 write_lock(&css_set_lock);
637 /* Add reference counts and links from the new css_set. */
638 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
639 struct cgroup *c = link->cgrp;
640 if (c->root == cgrp->root)
642 link_css_set(&tmp_cg_links, res, c);
645 BUG_ON(!list_empty(&tmp_cg_links));
649 /* Add this cgroup group to the hash table */
650 hhead = css_set_hash(res->subsys);
651 hlist_add_head(&res->hlist, hhead);
653 write_unlock(&css_set_lock);
659 * Return the cgroup for "task" from the given hierarchy. Must be
660 * called with cgroup_mutex held.
662 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
663 struct cgroupfs_root *root)
666 struct cgroup *res = NULL;
668 BUG_ON(!mutex_is_locked(&cgroup_mutex));
669 read_lock(&css_set_lock);
671 * No need to lock the task - since we hold cgroup_mutex the
672 * task can't change groups, so the only thing that can happen
673 * is that it exits and its css is set back to init_css_set.
676 if (css == &init_css_set) {
677 res = &root->top_cgroup;
679 struct cg_cgroup_link *link;
680 list_for_each_entry(link, &css->cg_links, cg_link_list) {
681 struct cgroup *c = link->cgrp;
682 if (c->root == root) {
688 read_unlock(&css_set_lock);
694 * There is one global cgroup mutex. We also require taking
695 * task_lock() when dereferencing a task's cgroup subsys pointers.
696 * See "The task_lock() exception", at the end of this comment.
698 * A task must hold cgroup_mutex to modify cgroups.
700 * Any task can increment and decrement the count field without lock.
701 * So in general, code holding cgroup_mutex can't rely on the count
702 * field not changing. However, if the count goes to zero, then only
703 * cgroup_attach_task() can increment it again. Because a count of zero
704 * means that no tasks are currently attached, therefore there is no
705 * way a task attached to that cgroup can fork (the other way to
706 * increment the count). So code holding cgroup_mutex can safely
707 * assume that if the count is zero, it will stay zero. Similarly, if
708 * a task holds cgroup_mutex on a cgroup with zero count, it
709 * knows that the cgroup won't be removed, as cgroup_rmdir()
712 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
713 * (usually) take cgroup_mutex. These are the two most performance
714 * critical pieces of code here. The exception occurs on cgroup_exit(),
715 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
716 * is taken, and if the cgroup count is zero, a usermode call made
717 * to the release agent with the name of the cgroup (path relative to
718 * the root of cgroup file system) as the argument.
720 * A cgroup can only be deleted if both its 'count' of using tasks
721 * is zero, and its list of 'children' cgroups is empty. Since all
722 * tasks in the system use _some_ cgroup, and since there is always at
723 * least one task in the system (init, pid == 1), therefore, top_cgroup
724 * always has either children cgroups and/or using tasks. So we don't
725 * need a special hack to ensure that top_cgroup cannot be deleted.
727 * The task_lock() exception
729 * The need for this exception arises from the action of
730 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
731 * another. It does so using cgroup_mutex, however there are
732 * several performance critical places that need to reference
733 * task->cgroup without the expense of grabbing a system global
734 * mutex. Therefore except as noted below, when dereferencing or, as
735 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
736 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
737 * the task_struct routinely used for such matters.
739 * P.S. One more locking exception. RCU is used to guard the
740 * update of a tasks cgroup pointer by cgroup_attach_task()
744 * cgroup_lock - lock out any changes to cgroup structures
747 void cgroup_lock(void)
749 mutex_lock(&cgroup_mutex);
751 EXPORT_SYMBOL_GPL(cgroup_lock);
754 * cgroup_unlock - release lock on cgroup changes
756 * Undo the lock taken in a previous cgroup_lock() call.
758 void cgroup_unlock(void)
760 mutex_unlock(&cgroup_mutex);
762 EXPORT_SYMBOL_GPL(cgroup_unlock);
765 * A couple of forward declarations required, due to cyclic reference loop:
766 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
767 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
771 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
772 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, struct nameidata *);
773 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
774 static int cgroup_populate_dir(struct cgroup *cgrp);
775 static const struct inode_operations cgroup_dir_inode_operations;
776 static const struct file_operations proc_cgroupstats_operations;
778 static struct backing_dev_info cgroup_backing_dev_info = {
780 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
783 static int alloc_css_id(struct cgroup_subsys *ss,
784 struct cgroup *parent, struct cgroup *child);
786 static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
788 struct inode *inode = new_inode(sb);
791 inode->i_ino = get_next_ino();
792 inode->i_mode = mode;
793 inode->i_uid = current_fsuid();
794 inode->i_gid = current_fsgid();
795 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
796 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
802 * Call subsys's pre_destroy handler.
803 * This is called before css refcnt check.
805 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
807 struct cgroup_subsys *ss;
810 for_each_subsys(cgrp->root, ss)
811 if (ss->pre_destroy) {
812 ret = ss->pre_destroy(ss, cgrp);
820 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
822 /* is dentry a directory ? if so, kfree() associated cgroup */
823 if (S_ISDIR(inode->i_mode)) {
824 struct cgroup *cgrp = dentry->d_fsdata;
825 struct cgroup_subsys *ss;
826 BUG_ON(!(cgroup_is_removed(cgrp)));
827 /* It's possible for external users to be holding css
828 * reference counts on a cgroup; css_put() needs to
829 * be able to access the cgroup after decrementing
830 * the reference count in order to know if it needs to
831 * queue the cgroup to be handled by the release
835 mutex_lock(&cgroup_mutex);
837 * Release the subsystem state objects.
839 for_each_subsys(cgrp->root, ss)
840 ss->destroy(ss, cgrp);
842 cgrp->root->number_of_cgroups--;
843 mutex_unlock(&cgroup_mutex);
846 * Drop the active superblock reference that we took when we
849 deactivate_super(cgrp->root->sb);
852 * if we're getting rid of the cgroup, refcount should ensure
853 * that there are no pidlists left.
855 BUG_ON(!list_empty(&cgrp->pidlists));
857 kfree_rcu(cgrp, rcu_head);
862 static int cgroup_delete(const struct dentry *d)
867 static void remove_dir(struct dentry *d)
869 struct dentry *parent = dget(d->d_parent);
872 simple_rmdir(parent->d_inode, d);
876 static void cgroup_clear_directory(struct dentry *dentry)
878 struct list_head *node;
880 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
881 spin_lock(&dentry->d_lock);
882 node = dentry->d_subdirs.next;
883 while (node != &dentry->d_subdirs) {
884 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
886 spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
889 /* This should never be called on a cgroup
890 * directory with child cgroups */
891 BUG_ON(d->d_inode->i_mode & S_IFDIR);
893 spin_unlock(&d->d_lock);
894 spin_unlock(&dentry->d_lock);
896 simple_unlink(dentry->d_inode, d);
898 spin_lock(&dentry->d_lock);
900 spin_unlock(&d->d_lock);
901 node = dentry->d_subdirs.next;
903 spin_unlock(&dentry->d_lock);
907 * NOTE : the dentry must have been dget()'ed
909 static void cgroup_d_remove_dir(struct dentry *dentry)
911 struct dentry *parent;
913 cgroup_clear_directory(dentry);
915 parent = dentry->d_parent;
916 spin_lock(&parent->d_lock);
917 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
918 list_del_init(&dentry->d_u.d_child);
919 spin_unlock(&dentry->d_lock);
920 spin_unlock(&parent->d_lock);
925 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
926 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
927 * reference to css->refcnt. In general, this refcnt is expected to goes down
930 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
932 DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
934 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
936 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
937 wake_up_all(&cgroup_rmdir_waitq);
940 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
945 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
947 cgroup_wakeup_rmdir_waiter(css->cgroup);
952 * Call with cgroup_mutex held. Drops reference counts on modules, including
953 * any duplicate ones that parse_cgroupfs_options took. If this function
954 * returns an error, no reference counts are touched.
956 static int rebind_subsystems(struct cgroupfs_root *root,
957 unsigned long final_bits)
959 unsigned long added_bits, removed_bits;
960 struct cgroup *cgrp = &root->top_cgroup;
963 BUG_ON(!mutex_is_locked(&cgroup_mutex));
965 removed_bits = root->actual_subsys_bits & ~final_bits;
966 added_bits = final_bits & ~root->actual_subsys_bits;
967 /* Check that any added subsystems are currently free */
968 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
969 unsigned long bit = 1UL << i;
970 struct cgroup_subsys *ss = subsys[i];
971 if (!(bit & added_bits))
974 * Nobody should tell us to do a subsys that doesn't exist:
975 * parse_cgroupfs_options should catch that case and refcounts
976 * ensure that subsystems won't disappear once selected.
979 if (ss->root != &rootnode) {
980 /* Subsystem isn't free */
985 /* Currently we don't handle adding/removing subsystems when
986 * any child cgroups exist. This is theoretically supportable
987 * but involves complex error handling, so it's being left until
989 if (root->number_of_cgroups > 1)
992 /* Process each subsystem */
993 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
994 struct cgroup_subsys *ss = subsys[i];
995 unsigned long bit = 1UL << i;
996 if (bit & added_bits) {
997 /* We're binding this subsystem to this hierarchy */
999 BUG_ON(cgrp->subsys[i]);
1000 BUG_ON(!dummytop->subsys[i]);
1001 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1002 mutex_lock(&ss->hierarchy_mutex);
1003 cgrp->subsys[i] = dummytop->subsys[i];
1004 cgrp->subsys[i]->cgroup = cgrp;
1005 list_move(&ss->sibling, &root->subsys_list);
1009 mutex_unlock(&ss->hierarchy_mutex);
1010 /* refcount was already taken, and we're keeping it */
1011 } else if (bit & removed_bits) {
1012 /* We're removing this subsystem */
1014 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1015 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1016 mutex_lock(&ss->hierarchy_mutex);
1018 ss->bind(ss, dummytop);
1019 dummytop->subsys[i]->cgroup = dummytop;
1020 cgrp->subsys[i] = NULL;
1021 subsys[i]->root = &rootnode;
1022 list_move(&ss->sibling, &rootnode.subsys_list);
1023 mutex_unlock(&ss->hierarchy_mutex);
1024 /* subsystem is now free - drop reference on module */
1025 module_put(ss->module);
1026 } else if (bit & final_bits) {
1027 /* Subsystem state should already exist */
1029 BUG_ON(!cgrp->subsys[i]);
1031 * a refcount was taken, but we already had one, so
1032 * drop the extra reference.
1034 module_put(ss->module);
1035 #ifdef CONFIG_MODULE_UNLOAD
1036 BUG_ON(ss->module && !module_refcount(ss->module));
1039 /* Subsystem state shouldn't exist */
1040 BUG_ON(cgrp->subsys[i]);
1043 root->subsys_bits = root->actual_subsys_bits = final_bits;
1049 static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
1051 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
1052 struct cgroup_subsys *ss;
1054 mutex_lock(&cgroup_mutex);
1055 for_each_subsys(root, ss)
1056 seq_printf(seq, ",%s", ss->name);
1057 if (test_bit(ROOT_NOPREFIX, &root->flags))
1058 seq_puts(seq, ",noprefix");
1059 if (strlen(root->release_agent_path))
1060 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1061 if (clone_children(&root->top_cgroup))
1062 seq_puts(seq, ",clone_children");
1063 if (strlen(root->name))
1064 seq_printf(seq, ",name=%s", root->name);
1065 mutex_unlock(&cgroup_mutex);
1069 struct cgroup_sb_opts {
1070 unsigned long subsys_bits;
1071 unsigned long flags;
1072 char *release_agent;
1073 bool clone_children;
1075 /* User explicitly requested empty subsystem */
1078 struct cgroupfs_root *new_root;
1083 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1084 * with cgroup_mutex held to protect the subsys[] array. This function takes
1085 * refcounts on subsystems to be used, unless it returns error, in which case
1086 * no refcounts are taken.
1088 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1090 char *token, *o = data;
1091 bool all_ss = false, one_ss = false;
1092 unsigned long mask = (unsigned long)-1;
1094 bool module_pin_failed = false;
1096 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1098 #ifdef CONFIG_CPUSETS
1099 mask = ~(1UL << cpuset_subsys_id);
1102 memset(opts, 0, sizeof(*opts));
1104 while ((token = strsep(&o, ",")) != NULL) {
1107 if (!strcmp(token, "none")) {
1108 /* Explicitly have no subsystems */
1112 if (!strcmp(token, "all")) {
1113 /* Mutually exclusive option 'all' + subsystem name */
1119 if (!strcmp(token, "noprefix")) {
1120 set_bit(ROOT_NOPREFIX, &opts->flags);
1123 if (!strcmp(token, "clone_children")) {
1124 opts->clone_children = true;
1127 if (!strncmp(token, "release_agent=", 14)) {
1128 /* Specifying two release agents is forbidden */
1129 if (opts->release_agent)
1131 opts->release_agent =
1132 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1133 if (!opts->release_agent)
1137 if (!strncmp(token, "name=", 5)) {
1138 const char *name = token + 5;
1139 /* Can't specify an empty name */
1142 /* Must match [\w.-]+ */
1143 for (i = 0; i < strlen(name); i++) {
1147 if ((c == '.') || (c == '-') || (c == '_'))
1151 /* Specifying two names is forbidden */
1154 opts->name = kstrndup(name,
1155 MAX_CGROUP_ROOT_NAMELEN - 1,
1163 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1164 struct cgroup_subsys *ss = subsys[i];
1167 if (strcmp(token, ss->name))
1172 /* Mutually exclusive option 'all' + subsystem name */
1175 set_bit(i, &opts->subsys_bits);
1180 if (i == CGROUP_SUBSYS_COUNT)
1185 * If the 'all' option was specified select all the subsystems,
1186 * otherwise if 'none', 'name=' and a subsystem name options
1187 * were not specified, let's default to 'all'
1189 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1190 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1191 struct cgroup_subsys *ss = subsys[i];
1196 set_bit(i, &opts->subsys_bits);
1200 /* Consistency checks */
1203 * Option noprefix was introduced just for backward compatibility
1204 * with the old cpuset, so we allow noprefix only if mounting just
1205 * the cpuset subsystem.
1207 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1208 (opts->subsys_bits & mask))
1212 /* Can't specify "none" and some subsystems */
1213 if (opts->subsys_bits && opts->none)
1217 * We either have to specify by name or by subsystems. (So all
1218 * empty hierarchies must have a name).
1220 if (!opts->subsys_bits && !opts->name)
1224 * Grab references on all the modules we'll need, so the subsystems
1225 * don't dance around before rebind_subsystems attaches them. This may
1226 * take duplicate reference counts on a subsystem that's already used,
1227 * but rebind_subsystems handles this case.
1229 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1230 unsigned long bit = 1UL << i;
1232 if (!(bit & opts->subsys_bits))
1234 if (!try_module_get(subsys[i]->module)) {
1235 module_pin_failed = true;
1239 if (module_pin_failed) {
1241 * oops, one of the modules was going away. this means that we
1242 * raced with a module_delete call, and to the user this is
1243 * essentially a "subsystem doesn't exist" case.
1245 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1246 /* drop refcounts only on the ones we took */
1247 unsigned long bit = 1UL << i;
1249 if (!(bit & opts->subsys_bits))
1251 module_put(subsys[i]->module);
1259 static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1262 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1263 unsigned long bit = 1UL << i;
1265 if (!(bit & subsys_bits))
1267 module_put(subsys[i]->module);
1271 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1274 struct cgroupfs_root *root = sb->s_fs_info;
1275 struct cgroup *cgrp = &root->top_cgroup;
1276 struct cgroup_sb_opts opts;
1278 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1279 mutex_lock(&cgroup_mutex);
1281 /* See what subsystems are wanted */
1282 ret = parse_cgroupfs_options(data, &opts);
1286 /* Don't allow flags or name to change at remount */
1287 if (opts.flags != root->flags ||
1288 (opts.name && strcmp(opts.name, root->name))) {
1290 drop_parsed_module_refcounts(opts.subsys_bits);
1294 ret = rebind_subsystems(root, opts.subsys_bits);
1296 drop_parsed_module_refcounts(opts.subsys_bits);
1300 /* (re)populate subsystem files */
1301 cgroup_populate_dir(cgrp);
1303 if (opts.release_agent)
1304 strcpy(root->release_agent_path, opts.release_agent);
1306 kfree(opts.release_agent);
1308 mutex_unlock(&cgroup_mutex);
1309 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1313 static const struct super_operations cgroup_ops = {
1314 .statfs = simple_statfs,
1315 .drop_inode = generic_delete_inode,
1316 .show_options = cgroup_show_options,
1317 .remount_fs = cgroup_remount,
1320 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1322 INIT_LIST_HEAD(&cgrp->sibling);
1323 INIT_LIST_HEAD(&cgrp->children);
1324 INIT_LIST_HEAD(&cgrp->css_sets);
1325 INIT_LIST_HEAD(&cgrp->release_list);
1326 INIT_LIST_HEAD(&cgrp->pidlists);
1327 mutex_init(&cgrp->pidlist_mutex);
1328 INIT_LIST_HEAD(&cgrp->event_list);
1329 spin_lock_init(&cgrp->event_list_lock);
1332 static void init_cgroup_root(struct cgroupfs_root *root)
1334 struct cgroup *cgrp = &root->top_cgroup;
1335 INIT_LIST_HEAD(&root->subsys_list);
1336 INIT_LIST_HEAD(&root->root_list);
1337 root->number_of_cgroups = 1;
1339 cgrp->top_cgroup = cgrp;
1340 init_cgroup_housekeeping(cgrp);
1343 static bool init_root_id(struct cgroupfs_root *root)
1348 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1350 spin_lock(&hierarchy_id_lock);
1351 /* Try to allocate the next unused ID */
1352 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1353 &root->hierarchy_id);
1355 /* Try again starting from 0 */
1356 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1358 next_hierarchy_id = root->hierarchy_id + 1;
1359 } else if (ret != -EAGAIN) {
1360 /* Can only get here if the 31-bit IDR is full ... */
1363 spin_unlock(&hierarchy_id_lock);
1368 static int cgroup_test_super(struct super_block *sb, void *data)
1370 struct cgroup_sb_opts *opts = data;
1371 struct cgroupfs_root *root = sb->s_fs_info;
1373 /* If we asked for a name then it must match */
1374 if (opts->name && strcmp(opts->name, root->name))
1378 * If we asked for subsystems (or explicitly for no
1379 * subsystems) then they must match
1381 if ((opts->subsys_bits || opts->none)
1382 && (opts->subsys_bits != root->subsys_bits))
1388 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1390 struct cgroupfs_root *root;
1392 if (!opts->subsys_bits && !opts->none)
1395 root = kzalloc(sizeof(*root), GFP_KERNEL);
1397 return ERR_PTR(-ENOMEM);
1399 if (!init_root_id(root)) {
1401 return ERR_PTR(-ENOMEM);
1403 init_cgroup_root(root);
1405 root->subsys_bits = opts->subsys_bits;
1406 root->flags = opts->flags;
1407 if (opts->release_agent)
1408 strcpy(root->release_agent_path, opts->release_agent);
1410 strcpy(root->name, opts->name);
1411 if (opts->clone_children)
1412 set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1416 static void cgroup_drop_root(struct cgroupfs_root *root)
1421 BUG_ON(!root->hierarchy_id);
1422 spin_lock(&hierarchy_id_lock);
1423 ida_remove(&hierarchy_ida, root->hierarchy_id);
1424 spin_unlock(&hierarchy_id_lock);
1428 static int cgroup_set_super(struct super_block *sb, void *data)
1431 struct cgroup_sb_opts *opts = data;
1433 /* If we don't have a new root, we can't set up a new sb */
1434 if (!opts->new_root)
1437 BUG_ON(!opts->subsys_bits && !opts->none);
1439 ret = set_anon_super(sb, NULL);
1443 sb->s_fs_info = opts->new_root;
1444 opts->new_root->sb = sb;
1446 sb->s_blocksize = PAGE_CACHE_SIZE;
1447 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1448 sb->s_magic = CGROUP_SUPER_MAGIC;
1449 sb->s_op = &cgroup_ops;
1454 static int cgroup_get_rootdir(struct super_block *sb)
1456 static const struct dentry_operations cgroup_dops = {
1457 .d_iput = cgroup_diput,
1458 .d_delete = cgroup_delete,
1461 struct inode *inode =
1462 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1463 struct dentry *dentry;
1468 inode->i_fop = &simple_dir_operations;
1469 inode->i_op = &cgroup_dir_inode_operations;
1470 /* directories start off with i_nlink == 2 (for "." entry) */
1472 dentry = d_alloc_root(inode);
1477 sb->s_root = dentry;
1478 /* for everything else we want ->d_op set */
1479 sb->s_d_op = &cgroup_dops;
1483 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1484 int flags, const char *unused_dev_name,
1487 struct cgroup_sb_opts opts;
1488 struct cgroupfs_root *root;
1490 struct super_block *sb;
1491 struct cgroupfs_root *new_root;
1493 /* First find the desired set of subsystems */
1494 mutex_lock(&cgroup_mutex);
1495 ret = parse_cgroupfs_options(data, &opts);
1496 mutex_unlock(&cgroup_mutex);
1501 * Allocate a new cgroup root. We may not need it if we're
1502 * reusing an existing hierarchy.
1504 new_root = cgroup_root_from_opts(&opts);
1505 if (IS_ERR(new_root)) {
1506 ret = PTR_ERR(new_root);
1509 opts.new_root = new_root;
1511 /* Locate an existing or new sb for this hierarchy */
1512 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1515 cgroup_drop_root(opts.new_root);
1519 root = sb->s_fs_info;
1521 if (root == opts.new_root) {
1522 /* We used the new root structure, so this is a new hierarchy */
1523 struct list_head tmp_cg_links;
1524 struct cgroup *root_cgrp = &root->top_cgroup;
1525 struct inode *inode;
1526 struct cgroupfs_root *existing_root;
1527 const struct cred *cred;
1530 BUG_ON(sb->s_root != NULL);
1532 ret = cgroup_get_rootdir(sb);
1534 goto drop_new_super;
1535 inode = sb->s_root->d_inode;
1537 mutex_lock(&inode->i_mutex);
1538 mutex_lock(&cgroup_mutex);
1540 if (strlen(root->name)) {
1541 /* Check for name clashes with existing mounts */
1542 for_each_active_root(existing_root) {
1543 if (!strcmp(existing_root->name, root->name)) {
1545 mutex_unlock(&cgroup_mutex);
1546 mutex_unlock(&inode->i_mutex);
1547 goto drop_new_super;
1553 * We're accessing css_set_count without locking
1554 * css_set_lock here, but that's OK - it can only be
1555 * increased by someone holding cgroup_lock, and
1556 * that's us. The worst that can happen is that we
1557 * have some link structures left over
1559 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1561 mutex_unlock(&cgroup_mutex);
1562 mutex_unlock(&inode->i_mutex);
1563 goto drop_new_super;
1566 ret = rebind_subsystems(root, root->subsys_bits);
1567 if (ret == -EBUSY) {
1568 mutex_unlock(&cgroup_mutex);
1569 mutex_unlock(&inode->i_mutex);
1570 free_cg_links(&tmp_cg_links);
1571 goto drop_new_super;
1574 * There must be no failure case after here, since rebinding
1575 * takes care of subsystems' refcounts, which are explicitly
1576 * dropped in the failure exit path.
1579 /* EBUSY should be the only error here */
1582 list_add(&root->root_list, &roots);
1585 sb->s_root->d_fsdata = root_cgrp;
1586 root->top_cgroup.dentry = sb->s_root;
1588 /* Link the top cgroup in this hierarchy into all
1589 * the css_set objects */
1590 write_lock(&css_set_lock);
1591 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1592 struct hlist_head *hhead = &css_set_table[i];
1593 struct hlist_node *node;
1596 hlist_for_each_entry(cg, node, hhead, hlist)
1597 link_css_set(&tmp_cg_links, cg, root_cgrp);
1599 write_unlock(&css_set_lock);
1601 free_cg_links(&tmp_cg_links);
1603 BUG_ON(!list_empty(&root_cgrp->sibling));
1604 BUG_ON(!list_empty(&root_cgrp->children));
1605 BUG_ON(root->number_of_cgroups != 1);
1607 cred = override_creds(&init_cred);
1608 cgroup_populate_dir(root_cgrp);
1610 mutex_unlock(&cgroup_mutex);
1611 mutex_unlock(&inode->i_mutex);
1614 * We re-used an existing hierarchy - the new root (if
1615 * any) is not needed
1617 cgroup_drop_root(opts.new_root);
1618 /* no subsys rebinding, so refcounts don't change */
1619 drop_parsed_module_refcounts(opts.subsys_bits);
1622 kfree(opts.release_agent);
1624 return dget(sb->s_root);
1627 deactivate_locked_super(sb);
1629 drop_parsed_module_refcounts(opts.subsys_bits);
1631 kfree(opts.release_agent);
1633 return ERR_PTR(ret);
1636 static void cgroup_kill_sb(struct super_block *sb) {
1637 struct cgroupfs_root *root = sb->s_fs_info;
1638 struct cgroup *cgrp = &root->top_cgroup;
1640 struct cg_cgroup_link *link;
1641 struct cg_cgroup_link *saved_link;
1645 BUG_ON(root->number_of_cgroups != 1);
1646 BUG_ON(!list_empty(&cgrp->children));
1647 BUG_ON(!list_empty(&cgrp->sibling));
1649 mutex_lock(&cgroup_mutex);
1651 /* Rebind all subsystems back to the default hierarchy */
1652 ret = rebind_subsystems(root, 0);
1653 /* Shouldn't be able to fail ... */
1657 * Release all the links from css_sets to this hierarchy's
1660 write_lock(&css_set_lock);
1662 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1664 list_del(&link->cg_link_list);
1665 list_del(&link->cgrp_link_list);
1668 write_unlock(&css_set_lock);
1670 if (!list_empty(&root->root_list)) {
1671 list_del(&root->root_list);
1675 mutex_unlock(&cgroup_mutex);
1677 kill_litter_super(sb);
1678 cgroup_drop_root(root);
1681 static struct file_system_type cgroup_fs_type = {
1683 .mount = cgroup_mount,
1684 .kill_sb = cgroup_kill_sb,
1687 static struct kobject *cgroup_kobj;
1689 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1691 return dentry->d_fsdata;
1694 static inline struct cftype *__d_cft(struct dentry *dentry)
1696 return dentry->d_fsdata;
1700 * cgroup_path - generate the path of a cgroup
1701 * @cgrp: the cgroup in question
1702 * @buf: the buffer to write the path into
1703 * @buflen: the length of the buffer
1705 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1706 * reference. Writes path of cgroup into buf. Returns 0 on success,
1709 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1712 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1713 cgroup_lock_is_held());
1715 if (!dentry || cgrp == dummytop) {
1717 * Inactive subsystems have no dentry for their root
1724 start = buf + buflen;
1728 int len = dentry->d_name.len;
1730 if ((start -= len) < buf)
1731 return -ENAMETOOLONG;
1732 memcpy(start, dentry->d_name.name, len);
1733 cgrp = cgrp->parent;
1737 dentry = rcu_dereference_check(cgrp->dentry,
1738 cgroup_lock_is_held());
1742 return -ENAMETOOLONG;
1745 memmove(buf, start, buf + buflen - start);
1748 EXPORT_SYMBOL_GPL(cgroup_path);
1751 * cgroup_task_migrate - move a task from one cgroup to another.
1753 * 'guarantee' is set if the caller promises that a new css_set for the task
1754 * will already exist. If not set, this function might sleep, and can fail with
1755 * -ENOMEM. Otherwise, it can only fail with -ESRCH.
1757 static int cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1758 struct task_struct *tsk, bool guarantee)
1760 struct css_set *oldcg;
1761 struct css_set *newcg;
1764 * get old css_set. we need to take task_lock and refcount it, because
1765 * an exiting task can change its css_set to init_css_set and drop its
1766 * old one without taking cgroup_mutex.
1769 oldcg = tsk->cgroups;
1773 /* locate or allocate a new css_set for this task. */
1775 /* we know the css_set we want already exists. */
1776 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
1777 read_lock(&css_set_lock);
1778 newcg = find_existing_css_set(oldcg, cgrp, template);
1781 read_unlock(&css_set_lock);
1784 /* find_css_set will give us newcg already referenced. */
1785 newcg = find_css_set(oldcg, cgrp);
1793 /* if PF_EXITING is set, the tsk->cgroups pointer is no longer safe. */
1795 if (tsk->flags & PF_EXITING) {
1800 rcu_assign_pointer(tsk->cgroups, newcg);
1803 /* Update the css_set linked lists if we're using them */
1804 write_lock(&css_set_lock);
1805 if (!list_empty(&tsk->cg_list))
1806 list_move(&tsk->cg_list, &newcg->tasks);
1807 write_unlock(&css_set_lock);
1810 * We just gained a reference on oldcg by taking it from the task. As
1811 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1812 * it here; it will be freed under RCU.
1814 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1820 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1821 * @cgrp: the cgroup the task is attaching to
1822 * @tsk: the task to be attached
1824 * Call holding cgroup_mutex. May take task_lock of
1825 * the task 'tsk' during call.
1827 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1830 struct cgroup_subsys *ss, *failed_ss = NULL;
1831 struct cgroup *oldcgrp;
1832 struct cgroupfs_root *root = cgrp->root;
1834 /* Nothing to do if the task is already in that cgroup */
1835 oldcgrp = task_cgroup_from_root(tsk, root);
1836 if (cgrp == oldcgrp)
1839 for_each_subsys(root, ss) {
1840 if (ss->can_attach) {
1841 retval = ss->can_attach(ss, cgrp, tsk);
1844 * Remember on which subsystem the can_attach()
1845 * failed, so that we only call cancel_attach()
1846 * against the subsystems whose can_attach()
1847 * succeeded. (See below)
1853 if (ss->can_attach_task) {
1854 retval = ss->can_attach_task(cgrp, tsk);
1862 retval = cgroup_task_migrate(cgrp, oldcgrp, tsk, false);
1866 for_each_subsys(root, ss) {
1868 ss->pre_attach(cgrp);
1869 if (ss->attach_task)
1870 ss->attach_task(cgrp, tsk);
1872 ss->attach(ss, cgrp, oldcgrp, tsk);
1878 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1879 * is no longer empty.
1881 cgroup_wakeup_rmdir_waiter(cgrp);
1884 for_each_subsys(root, ss) {
1885 if (ss == failed_ss)
1887 * This subsystem was the one that failed the
1888 * can_attach() check earlier, so we don't need
1889 * to call cancel_attach() against it or any
1890 * remaining subsystems.
1893 if (ss->cancel_attach)
1894 ss->cancel_attach(ss, cgrp, tsk);
1901 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1902 * @from: attach to all cgroups of a given task
1903 * @tsk: the task to be attached
1905 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
1907 struct cgroupfs_root *root;
1911 for_each_active_root(root) {
1912 struct cgroup *from_cg = task_cgroup_from_root(from, root);
1914 retval = cgroup_attach_task(from_cg, tsk);
1922 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
1925 * cgroup_attach_proc works in two stages, the first of which prefetches all
1926 * new css_sets needed (to make sure we have enough memory before committing
1927 * to the move) and stores them in a list of entries of the following type.
1928 * TODO: possible optimization: use css_set->rcu_head for chaining instead
1930 struct cg_list_entry {
1932 struct list_head links;
1935 static bool css_set_check_fetched(struct cgroup *cgrp,
1936 struct task_struct *tsk, struct css_set *cg,
1937 struct list_head *newcg_list)
1939 struct css_set *newcg;
1940 struct cg_list_entry *cg_entry;
1941 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
1943 read_lock(&css_set_lock);
1944 newcg = find_existing_css_set(cg, cgrp, template);
1947 read_unlock(&css_set_lock);
1949 /* doesn't exist at all? */
1952 /* see if it's already in the list */
1953 list_for_each_entry(cg_entry, newcg_list, links) {
1954 if (cg_entry->cg == newcg) {
1966 * Find the new css_set and store it in the list in preparation for moving the
1967 * given task to the given cgroup. Returns 0 or -ENOMEM.
1969 static int css_set_prefetch(struct cgroup *cgrp, struct css_set *cg,
1970 struct list_head *newcg_list)
1972 struct css_set *newcg;
1973 struct cg_list_entry *cg_entry;
1975 /* ensure a new css_set will exist for this thread */
1976 newcg = find_css_set(cg, cgrp);
1979 /* add it to the list */
1980 cg_entry = kmalloc(sizeof(struct cg_list_entry), GFP_KERNEL);
1985 cg_entry->cg = newcg;
1986 list_add(&cg_entry->links, newcg_list);
1991 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
1992 * @cgrp: the cgroup to attach to
1993 * @leader: the threadgroup leader task_struct of the group to be attached
1995 * Call holding cgroup_mutex and the threadgroup_fork_lock of the leader. Will
1996 * take task_lock of each thread in leader's threadgroup individually in turn.
1998 int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2000 int retval, i, group_size;
2001 struct cgroup_subsys *ss, *failed_ss = NULL;
2002 bool cancel_failed_ss = false;
2003 /* guaranteed to be initialized later, but the compiler needs this */
2004 struct cgroup *oldcgrp = NULL;
2005 struct css_set *oldcg;
2006 struct cgroupfs_root *root = cgrp->root;
2007 /* threadgroup list cursor and array */
2008 struct task_struct *tsk;
2009 struct flex_array *group;
2011 * we need to make sure we have css_sets for all the tasks we're
2012 * going to move -before- we actually start moving them, so that in
2013 * case we get an ENOMEM we can bail out before making any changes.
2015 struct list_head newcg_list;
2016 struct cg_list_entry *cg_entry, *temp_nobe;
2019 * step 0: in order to do expensive, possibly blocking operations for
2020 * every thread, we cannot iterate the thread group list, since it needs
2021 * rcu or tasklist locked. instead, build an array of all threads in the
2022 * group - threadgroup_fork_lock prevents new threads from appearing,
2023 * and if threads exit, this will just be an over-estimate.
2025 group_size = get_nr_threads(leader);
2026 /* flex_array supports very large thread-groups better than kmalloc. */
2027 group = flex_array_alloc(sizeof(struct task_struct *), group_size,
2031 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2032 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
2034 goto out_free_group_list;
2036 /* prevent changes to the threadgroup list while we take a snapshot. */
2037 read_lock(&tasklist_lock);
2038 if (!thread_group_leader(leader)) {
2040 * a race with de_thread from another thread's exec() may strip
2041 * us of our leadership, making while_each_thread unsafe to use
2042 * on this task. if this happens, there is no choice but to
2043 * throw this task away and try again (from cgroup_procs_write);
2044 * this is "double-double-toil-and-trouble-check locking".
2046 read_unlock(&tasklist_lock);
2048 goto out_free_group_list;
2050 /* take a reference on each task in the group to go in the array. */
2054 /* as per above, nr_threads may decrease, but not increase. */
2055 BUG_ON(i >= group_size);
2056 get_task_struct(tsk);
2058 * saying GFP_ATOMIC has no effect here because we did prealloc
2059 * earlier, but it's good form to communicate our expectations.
2061 retval = flex_array_put_ptr(group, i, tsk, GFP_ATOMIC);
2062 BUG_ON(retval != 0);
2064 } while_each_thread(leader, tsk);
2065 /* remember the number of threads in the array for later. */
2067 read_unlock(&tasklist_lock);
2070 * step 1: check that we can legitimately attach to the cgroup.
2072 for_each_subsys(root, ss) {
2073 if (ss->can_attach) {
2074 retval = ss->can_attach(ss, cgrp, leader);
2077 goto out_cancel_attach;
2080 /* a callback to be run on every thread in the threadgroup. */
2081 if (ss->can_attach_task) {
2082 /* run on each task in the threadgroup. */
2083 for (i = 0; i < group_size; i++) {
2084 tsk = flex_array_get_ptr(group, i);
2085 retval = ss->can_attach_task(cgrp, tsk);
2088 cancel_failed_ss = true;
2089 goto out_cancel_attach;
2096 * step 2: make sure css_sets exist for all threads to be migrated.
2097 * we use find_css_set, which allocates a new one if necessary.
2099 INIT_LIST_HEAD(&newcg_list);
2100 for (i = 0; i < group_size; i++) {
2101 tsk = flex_array_get_ptr(group, i);
2102 /* nothing to do if this task is already in the cgroup */
2103 oldcgrp = task_cgroup_from_root(tsk, root);
2104 if (cgrp == oldcgrp)
2106 /* get old css_set pointer */
2108 oldcg = tsk->cgroups;
2111 /* see if the new one for us is already in the list? */
2112 if (css_set_check_fetched(cgrp, tsk, oldcg, &newcg_list)) {
2113 /* was already there, nothing to do. */
2116 /* we don't already have it. get new one. */
2117 retval = css_set_prefetch(cgrp, oldcg, &newcg_list);
2120 goto out_list_teardown;
2125 * step 3: now that we're guaranteed success wrt the css_sets, proceed
2126 * to move all tasks to the new cgroup, calling ss->attach_task for each
2127 * one along the way. there are no failure cases after here, so this is
2130 for_each_subsys(root, ss) {
2132 ss->pre_attach(cgrp);
2134 for (i = 0; i < group_size; i++) {
2135 tsk = flex_array_get_ptr(group, i);
2136 /* leave current thread as it is if it's already there */
2137 oldcgrp = task_cgroup_from_root(tsk, root);
2138 if (cgrp == oldcgrp)
2140 /* if the thread is PF_EXITING, it can just get skipped. */
2141 retval = cgroup_task_migrate(cgrp, oldcgrp, tsk, true);
2143 /* attach each task to each subsystem */
2144 for_each_subsys(root, ss) {
2145 if (ss->attach_task)
2146 ss->attach_task(cgrp, tsk);
2149 BUG_ON(retval != -ESRCH);
2152 /* nothing is sensitive to fork() after this point. */
2155 * step 4: do expensive, non-thread-specific subsystem callbacks.
2156 * TODO: if ever a subsystem needs to know the oldcgrp for each task
2157 * being moved, this call will need to be reworked to communicate that.
2159 for_each_subsys(root, ss) {
2161 ss->attach(ss, cgrp, oldcgrp, leader);
2165 * step 5: success! and cleanup
2168 cgroup_wakeup_rmdir_waiter(cgrp);
2171 /* clean up the list of prefetched css_sets. */
2172 list_for_each_entry_safe(cg_entry, temp_nobe, &newcg_list, links) {
2173 list_del(&cg_entry->links);
2174 put_css_set(cg_entry->cg);
2178 /* same deal as in cgroup_attach_task */
2180 for_each_subsys(root, ss) {
2181 if (ss == failed_ss) {
2182 if (cancel_failed_ss && ss->cancel_attach)
2183 ss->cancel_attach(ss, cgrp, leader);
2186 if (ss->cancel_attach)
2187 ss->cancel_attach(ss, cgrp, leader);
2190 /* clean up the array of referenced threads in the group. */
2191 for (i = 0; i < group_size; i++) {
2192 tsk = flex_array_get_ptr(group, i);
2193 put_task_struct(tsk);
2195 out_free_group_list:
2196 flex_array_free(group);
2201 * Find the task_struct of the task to attach by vpid and pass it along to the
2202 * function to attach either it or all tasks in its threadgroup. Will take
2203 * cgroup_mutex; may take task_lock of task.
2205 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2207 struct task_struct *tsk;
2208 const struct cred *cred = current_cred(), *tcred;
2211 if (!cgroup_lock_live_group(cgrp))
2216 tsk = find_task_by_vpid(pid);
2224 * RCU protects this access, since tsk was found in the
2225 * tid map. a race with de_thread may cause group_leader
2226 * to stop being the leader, but cgroup_attach_proc will
2229 tsk = tsk->group_leader;
2230 } else if (tsk->flags & PF_EXITING) {
2231 /* optimization for the single-task-only case */
2238 * even if we're attaching all tasks in the thread group, we
2239 * only need to check permissions on one of them.
2241 tcred = __task_cred(tsk);
2243 cred->euid != tcred->uid &&
2244 cred->euid != tcred->suid) {
2249 get_task_struct(tsk);
2253 tsk = current->group_leader;
2256 get_task_struct(tsk);
2260 threadgroup_fork_write_lock(tsk);
2261 ret = cgroup_attach_proc(cgrp, tsk);
2262 threadgroup_fork_write_unlock(tsk);
2264 ret = cgroup_attach_task(cgrp, tsk);
2266 put_task_struct(tsk);
2271 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2273 return attach_task_by_pid(cgrp, pid, false);
2276 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2281 * attach_proc fails with -EAGAIN if threadgroup leadership
2282 * changes in the middle of the operation, in which case we need
2283 * to find the task_struct for the new leader and start over.
2285 ret = attach_task_by_pid(cgrp, tgid, true);
2286 } while (ret == -EAGAIN);
2291 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2292 * @cgrp: the cgroup to be checked for liveness
2294 * On success, returns true; the lock should be later released with
2295 * cgroup_unlock(). On failure returns false with no lock held.
2297 bool cgroup_lock_live_group(struct cgroup *cgrp)
2299 mutex_lock(&cgroup_mutex);
2300 if (cgroup_is_removed(cgrp)) {
2301 mutex_unlock(&cgroup_mutex);
2306 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2308 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2311 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2312 if (strlen(buffer) >= PATH_MAX)
2314 if (!cgroup_lock_live_group(cgrp))
2316 strcpy(cgrp->root->release_agent_path, buffer);
2321 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2322 struct seq_file *seq)
2324 if (!cgroup_lock_live_group(cgrp))
2326 seq_puts(seq, cgrp->root->release_agent_path);
2327 seq_putc(seq, '\n');
2332 /* A buffer size big enough for numbers or short strings */
2333 #define CGROUP_LOCAL_BUFFER_SIZE 64
2335 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2337 const char __user *userbuf,
2338 size_t nbytes, loff_t *unused_ppos)
2340 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2346 if (nbytes >= sizeof(buffer))
2348 if (copy_from_user(buffer, userbuf, nbytes))
2351 buffer[nbytes] = 0; /* nul-terminate */
2352 if (cft->write_u64) {
2353 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2356 retval = cft->write_u64(cgrp, cft, val);
2358 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2361 retval = cft->write_s64(cgrp, cft, val);
2368 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2370 const char __user *userbuf,
2371 size_t nbytes, loff_t *unused_ppos)
2373 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2375 size_t max_bytes = cft->max_write_len;
2376 char *buffer = local_buffer;
2379 max_bytes = sizeof(local_buffer) - 1;
2380 if (nbytes >= max_bytes)
2382 /* Allocate a dynamic buffer if we need one */
2383 if (nbytes >= sizeof(local_buffer)) {
2384 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2388 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2393 buffer[nbytes] = 0; /* nul-terminate */
2394 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2398 if (buffer != local_buffer)
2403 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2404 size_t nbytes, loff_t *ppos)
2406 struct cftype *cft = __d_cft(file->f_dentry);
2407 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2409 if (cgroup_is_removed(cgrp))
2412 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2413 if (cft->write_u64 || cft->write_s64)
2414 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2415 if (cft->write_string)
2416 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2418 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2419 return ret ? ret : nbytes;
2424 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2426 char __user *buf, size_t nbytes,
2429 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2430 u64 val = cft->read_u64(cgrp, cft);
2431 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2433 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2436 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2438 char __user *buf, size_t nbytes,
2441 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2442 s64 val = cft->read_s64(cgrp, cft);
2443 int len = sprintf(tmp, "%lld\n", (long long) val);
2445 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2448 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2449 size_t nbytes, loff_t *ppos)
2451 struct cftype *cft = __d_cft(file->f_dentry);
2452 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2454 if (cgroup_is_removed(cgrp))
2458 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2460 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2462 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2467 * seqfile ops/methods for returning structured data. Currently just
2468 * supports string->u64 maps, but can be extended in future.
2471 struct cgroup_seqfile_state {
2473 struct cgroup *cgroup;
2476 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2478 struct seq_file *sf = cb->state;
2479 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2482 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2484 struct cgroup_seqfile_state *state = m->private;
2485 struct cftype *cft = state->cft;
2486 if (cft->read_map) {
2487 struct cgroup_map_cb cb = {
2488 .fill = cgroup_map_add,
2491 return cft->read_map(state->cgroup, cft, &cb);
2493 return cft->read_seq_string(state->cgroup, cft, m);
2496 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2498 struct seq_file *seq = file->private_data;
2499 kfree(seq->private);
2500 return single_release(inode, file);
2503 static const struct file_operations cgroup_seqfile_operations = {
2505 .write = cgroup_file_write,
2506 .llseek = seq_lseek,
2507 .release = cgroup_seqfile_release,
2510 static int cgroup_file_open(struct inode *inode, struct file *file)
2515 err = generic_file_open(inode, file);
2518 cft = __d_cft(file->f_dentry);
2520 if (cft->read_map || cft->read_seq_string) {
2521 struct cgroup_seqfile_state *state =
2522 kzalloc(sizeof(*state), GFP_USER);
2526 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2527 file->f_op = &cgroup_seqfile_operations;
2528 err = single_open(file, cgroup_seqfile_show, state);
2531 } else if (cft->open)
2532 err = cft->open(inode, file);
2539 static int cgroup_file_release(struct inode *inode, struct file *file)
2541 struct cftype *cft = __d_cft(file->f_dentry);
2543 return cft->release(inode, file);
2548 * cgroup_rename - Only allow simple rename of directories in place.
2550 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2551 struct inode *new_dir, struct dentry *new_dentry)
2553 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2555 if (new_dentry->d_inode)
2557 if (old_dir != new_dir)
2559 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2562 static const struct file_operations cgroup_file_operations = {
2563 .read = cgroup_file_read,
2564 .write = cgroup_file_write,
2565 .llseek = generic_file_llseek,
2566 .open = cgroup_file_open,
2567 .release = cgroup_file_release,
2570 static const struct inode_operations cgroup_dir_inode_operations = {
2571 .lookup = cgroup_lookup,
2572 .mkdir = cgroup_mkdir,
2573 .rmdir = cgroup_rmdir,
2574 .rename = cgroup_rename,
2577 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
2579 if (dentry->d_name.len > NAME_MAX)
2580 return ERR_PTR(-ENAMETOOLONG);
2581 d_add(dentry, NULL);
2586 * Check if a file is a control file
2588 static inline struct cftype *__file_cft(struct file *file)
2590 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2591 return ERR_PTR(-EINVAL);
2592 return __d_cft(file->f_dentry);
2595 static int cgroup_create_file(struct dentry *dentry, mode_t mode,
2596 struct super_block *sb)
2598 struct inode *inode;
2602 if (dentry->d_inode)
2605 inode = cgroup_new_inode(mode, sb);
2609 if (S_ISDIR(mode)) {
2610 inode->i_op = &cgroup_dir_inode_operations;
2611 inode->i_fop = &simple_dir_operations;
2613 /* start off with i_nlink == 2 (for "." entry) */
2616 /* start with the directory inode held, so that we can
2617 * populate it without racing with another mkdir */
2618 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2619 } else if (S_ISREG(mode)) {
2621 inode->i_fop = &cgroup_file_operations;
2623 d_instantiate(dentry, inode);
2624 dget(dentry); /* Extra count - pin the dentry in core */
2629 * cgroup_create_dir - create a directory for an object.
2630 * @cgrp: the cgroup we create the directory for. It must have a valid
2631 * ->parent field. And we are going to fill its ->dentry field.
2632 * @dentry: dentry of the new cgroup
2633 * @mode: mode to set on new directory.
2635 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2638 struct dentry *parent;
2641 parent = cgrp->parent->dentry;
2642 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2644 dentry->d_fsdata = cgrp;
2645 inc_nlink(parent->d_inode);
2646 rcu_assign_pointer(cgrp->dentry, dentry);
2653 * cgroup_file_mode - deduce file mode of a control file
2654 * @cft: the control file in question
2656 * returns cft->mode if ->mode is not 0
2657 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2658 * returns S_IRUGO if it has only a read handler
2659 * returns S_IWUSR if it has only a write hander
2661 static mode_t cgroup_file_mode(const struct cftype *cft)
2668 if (cft->read || cft->read_u64 || cft->read_s64 ||
2669 cft->read_map || cft->read_seq_string)
2672 if (cft->write || cft->write_u64 || cft->write_s64 ||
2673 cft->write_string || cft->trigger)
2679 int cgroup_add_file(struct cgroup *cgrp,
2680 struct cgroup_subsys *subsys,
2681 const struct cftype *cft)
2683 struct dentry *dir = cgrp->dentry;
2684 struct dentry *dentry;
2688 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2689 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2690 strcpy(name, subsys->name);
2693 strcat(name, cft->name);
2694 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2695 dentry = lookup_one_len(name, dir, strlen(name));
2696 if (!IS_ERR(dentry)) {
2697 mode = cgroup_file_mode(cft);
2698 error = cgroup_create_file(dentry, mode | S_IFREG,
2701 dentry->d_fsdata = (void *)cft;
2704 error = PTR_ERR(dentry);
2707 EXPORT_SYMBOL_GPL(cgroup_add_file);
2709 int cgroup_add_files(struct cgroup *cgrp,
2710 struct cgroup_subsys *subsys,
2711 const struct cftype cft[],
2715 for (i = 0; i < count; i++) {
2716 err = cgroup_add_file(cgrp, subsys, &cft[i]);
2722 EXPORT_SYMBOL_GPL(cgroup_add_files);
2725 * cgroup_task_count - count the number of tasks in a cgroup.
2726 * @cgrp: the cgroup in question
2728 * Return the number of tasks in the cgroup.
2730 int cgroup_task_count(const struct cgroup *cgrp)
2733 struct cg_cgroup_link *link;
2735 read_lock(&css_set_lock);
2736 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2737 count += atomic_read(&link->cg->refcount);
2739 read_unlock(&css_set_lock);
2744 * Advance a list_head iterator. The iterator should be positioned at
2745 * the start of a css_set
2747 static void cgroup_advance_iter(struct cgroup *cgrp,
2748 struct cgroup_iter *it)
2750 struct list_head *l = it->cg_link;
2751 struct cg_cgroup_link *link;
2754 /* Advance to the next non-empty css_set */
2757 if (l == &cgrp->css_sets) {
2761 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2763 } while (list_empty(&cg->tasks));
2765 it->task = cg->tasks.next;
2769 * To reduce the fork() overhead for systems that are not actually
2770 * using their cgroups capability, we don't maintain the lists running
2771 * through each css_set to its tasks until we see the list actually
2772 * used - in other words after the first call to cgroup_iter_start().
2774 * The tasklist_lock is not held here, as do_each_thread() and
2775 * while_each_thread() are protected by RCU.
2777 static void cgroup_enable_task_cg_lists(void)
2779 struct task_struct *p, *g;
2780 write_lock(&css_set_lock);
2781 use_task_css_set_links = 1;
2782 do_each_thread(g, p) {
2785 * We should check if the process is exiting, otherwise
2786 * it will race with cgroup_exit() in that the list
2787 * entry won't be deleted though the process has exited.
2789 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2790 list_add(&p->cg_list, &p->cgroups->tasks);
2792 } while_each_thread(g, p);
2793 write_unlock(&css_set_lock);
2796 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2799 * The first time anyone tries to iterate across a cgroup,
2800 * we need to enable the list linking each css_set to its
2801 * tasks, and fix up all existing tasks.
2803 if (!use_task_css_set_links)
2804 cgroup_enable_task_cg_lists();
2806 read_lock(&css_set_lock);
2807 it->cg_link = &cgrp->css_sets;
2808 cgroup_advance_iter(cgrp, it);
2811 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2812 struct cgroup_iter *it)
2814 struct task_struct *res;
2815 struct list_head *l = it->task;
2816 struct cg_cgroup_link *link;
2818 /* If the iterator cg is NULL, we have no tasks */
2821 res = list_entry(l, struct task_struct, cg_list);
2822 /* Advance iterator to find next entry */
2824 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2825 if (l == &link->cg->tasks) {
2826 /* We reached the end of this task list - move on to
2827 * the next cg_cgroup_link */
2828 cgroup_advance_iter(cgrp, it);
2835 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2837 read_unlock(&css_set_lock);
2840 static inline int started_after_time(struct task_struct *t1,
2841 struct timespec *time,
2842 struct task_struct *t2)
2844 int start_diff = timespec_compare(&t1->start_time, time);
2845 if (start_diff > 0) {
2847 } else if (start_diff < 0) {
2851 * Arbitrarily, if two processes started at the same
2852 * time, we'll say that the lower pointer value
2853 * started first. Note that t2 may have exited by now
2854 * so this may not be a valid pointer any longer, but
2855 * that's fine - it still serves to distinguish
2856 * between two tasks started (effectively) simultaneously.
2863 * This function is a callback from heap_insert() and is used to order
2865 * In this case we order the heap in descending task start time.
2867 static inline int started_after(void *p1, void *p2)
2869 struct task_struct *t1 = p1;
2870 struct task_struct *t2 = p2;
2871 return started_after_time(t1, &t2->start_time, t2);
2875 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2876 * @scan: struct cgroup_scanner containing arguments for the scan
2878 * Arguments include pointers to callback functions test_task() and
2880 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2881 * and if it returns true, call process_task() for it also.
2882 * The test_task pointer may be NULL, meaning always true (select all tasks).
2883 * Effectively duplicates cgroup_iter_{start,next,end}()
2884 * but does not lock css_set_lock for the call to process_task().
2885 * The struct cgroup_scanner may be embedded in any structure of the caller's
2887 * It is guaranteed that process_task() will act on every task that
2888 * is a member of the cgroup for the duration of this call. This
2889 * function may or may not call process_task() for tasks that exit
2890 * or move to a different cgroup during the call, or are forked or
2891 * move into the cgroup during the call.
2893 * Note that test_task() may be called with locks held, and may in some
2894 * situations be called multiple times for the same task, so it should
2896 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2897 * pre-allocated and will be used for heap operations (and its "gt" member will
2898 * be overwritten), else a temporary heap will be used (allocation of which
2899 * may cause this function to fail).
2901 int cgroup_scan_tasks(struct cgroup_scanner *scan)
2904 struct cgroup_iter it;
2905 struct task_struct *p, *dropped;
2906 /* Never dereference latest_task, since it's not refcounted */
2907 struct task_struct *latest_task = NULL;
2908 struct ptr_heap tmp_heap;
2909 struct ptr_heap *heap;
2910 struct timespec latest_time = { 0, 0 };
2913 /* The caller supplied our heap and pre-allocated its memory */
2915 heap->gt = &started_after;
2917 /* We need to allocate our own heap memory */
2919 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
2921 /* cannot allocate the heap */
2927 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2928 * to determine which are of interest, and using the scanner's
2929 * "process_task" callback to process any of them that need an update.
2930 * Since we don't want to hold any locks during the task updates,
2931 * gather tasks to be processed in a heap structure.
2932 * The heap is sorted by descending task start time.
2933 * If the statically-sized heap fills up, we overflow tasks that
2934 * started later, and in future iterations only consider tasks that
2935 * started after the latest task in the previous pass. This
2936 * guarantees forward progress and that we don't miss any tasks.
2939 cgroup_iter_start(scan->cg, &it);
2940 while ((p = cgroup_iter_next(scan->cg, &it))) {
2942 * Only affect tasks that qualify per the caller's callback,
2943 * if he provided one
2945 if (scan->test_task && !scan->test_task(p, scan))
2948 * Only process tasks that started after the last task
2951 if (!started_after_time(p, &latest_time, latest_task))
2953 dropped = heap_insert(heap, p);
2954 if (dropped == NULL) {
2956 * The new task was inserted; the heap wasn't
2960 } else if (dropped != p) {
2962 * The new task was inserted, and pushed out a
2966 put_task_struct(dropped);
2969 * Else the new task was newer than anything already in
2970 * the heap and wasn't inserted
2973 cgroup_iter_end(scan->cg, &it);
2976 for (i = 0; i < heap->size; i++) {
2977 struct task_struct *q = heap->ptrs[i];
2979 latest_time = q->start_time;
2982 /* Process the task per the caller's callback */
2983 scan->process_task(q, scan);
2987 * If we had to process any tasks at all, scan again
2988 * in case some of them were in the middle of forking
2989 * children that didn't get processed.
2990 * Not the most efficient way to do it, but it avoids
2991 * having to take callback_mutex in the fork path
2995 if (heap == &tmp_heap)
2996 heap_free(&tmp_heap);
3001 * Stuff for reading the 'tasks'/'procs' files.
3003 * Reading this file can return large amounts of data if a cgroup has
3004 * *lots* of attached tasks. So it may need several calls to read(),
3005 * but we cannot guarantee that the information we produce is correct
3006 * unless we produce it entirely atomically.
3011 * The following two functions "fix" the issue where there are more pids
3012 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3013 * TODO: replace with a kernel-wide solution to this problem
3015 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3016 static void *pidlist_allocate(int count)
3018 if (PIDLIST_TOO_LARGE(count))
3019 return vmalloc(count * sizeof(pid_t));
3021 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3023 static void pidlist_free(void *p)
3025 if (is_vmalloc_addr(p))
3030 static void *pidlist_resize(void *p, int newcount)
3033 /* note: if new alloc fails, old p will still be valid either way */
3034 if (is_vmalloc_addr(p)) {
3035 newlist = vmalloc(newcount * sizeof(pid_t));
3038 memcpy(newlist, p, newcount * sizeof(pid_t));
3041 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3047 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3048 * If the new stripped list is sufficiently smaller and there's enough memory
3049 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3050 * number of unique elements.
3052 /* is the size difference enough that we should re-allocate the array? */
3053 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3054 static int pidlist_uniq(pid_t **p, int length)
3061 * we presume the 0th element is unique, so i starts at 1. trivial
3062 * edge cases first; no work needs to be done for either
3064 if (length == 0 || length == 1)
3066 /* src and dest walk down the list; dest counts unique elements */
3067 for (src = 1; src < length; src++) {
3068 /* find next unique element */
3069 while (list[src] == list[src-1]) {
3074 /* dest always points to where the next unique element goes */
3075 list[dest] = list[src];
3080 * if the length difference is large enough, we want to allocate a
3081 * smaller buffer to save memory. if this fails due to out of memory,
3082 * we'll just stay with what we've got.
3084 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3085 newlist = pidlist_resize(list, dest);
3092 static int cmppid(const void *a, const void *b)
3094 return *(pid_t *)a - *(pid_t *)b;
3098 * find the appropriate pidlist for our purpose (given procs vs tasks)
3099 * returns with the lock on that pidlist already held, and takes care
3100 * of the use count, or returns NULL with no locks held if we're out of
3103 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3104 enum cgroup_filetype type)
3106 struct cgroup_pidlist *l;
3107 /* don't need task_nsproxy() if we're looking at ourself */
3108 struct pid_namespace *ns = current->nsproxy->pid_ns;
3111 * We can't drop the pidlist_mutex before taking the l->mutex in case
3112 * the last ref-holder is trying to remove l from the list at the same
3113 * time. Holding the pidlist_mutex precludes somebody taking whichever
3114 * list we find out from under us - compare release_pid_array().
3116 mutex_lock(&cgrp->pidlist_mutex);
3117 list_for_each_entry(l, &cgrp->pidlists, links) {
3118 if (l->key.type == type && l->key.ns == ns) {
3119 /* make sure l doesn't vanish out from under us */
3120 down_write(&l->mutex);
3121 mutex_unlock(&cgrp->pidlist_mutex);
3125 /* entry not found; create a new one */
3126 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3128 mutex_unlock(&cgrp->pidlist_mutex);
3131 init_rwsem(&l->mutex);
3132 down_write(&l->mutex);
3134 l->key.ns = get_pid_ns(ns);
3135 l->use_count = 0; /* don't increment here */
3138 list_add(&l->links, &cgrp->pidlists);
3139 mutex_unlock(&cgrp->pidlist_mutex);
3144 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3146 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3147 struct cgroup_pidlist **lp)
3151 int pid, n = 0; /* used for populating the array */
3152 struct cgroup_iter it;
3153 struct task_struct *tsk;
3154 struct cgroup_pidlist *l;
3157 * If cgroup gets more users after we read count, we won't have
3158 * enough space - tough. This race is indistinguishable to the
3159 * caller from the case that the additional cgroup users didn't
3160 * show up until sometime later on.
3162 length = cgroup_task_count(cgrp);
3163 array = pidlist_allocate(length);
3166 /* now, populate the array */
3167 cgroup_iter_start(cgrp, &it);
3168 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3169 if (unlikely(n == length))
3171 /* get tgid or pid for procs or tasks file respectively */
3172 if (type == CGROUP_FILE_PROCS)
3173 pid = task_tgid_vnr(tsk);
3175 pid = task_pid_vnr(tsk);
3176 if (pid > 0) /* make sure to only use valid results */
3179 cgroup_iter_end(cgrp, &it);
3181 /* now sort & (if procs) strip out duplicates */
3182 sort(array, length, sizeof(pid_t), cmppid, NULL);
3183 if (type == CGROUP_FILE_PROCS)
3184 length = pidlist_uniq(&array, length);
3185 l = cgroup_pidlist_find(cgrp, type);
3187 pidlist_free(array);
3190 /* store array, freeing old if necessary - lock already held */
3191 pidlist_free(l->list);
3195 up_write(&l->mutex);
3201 * cgroupstats_build - build and fill cgroupstats
3202 * @stats: cgroupstats to fill information into
3203 * @dentry: A dentry entry belonging to the cgroup for which stats have
3206 * Build and fill cgroupstats so that taskstats can export it to user
3209 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3212 struct cgroup *cgrp;
3213 struct cgroup_iter it;
3214 struct task_struct *tsk;
3217 * Validate dentry by checking the superblock operations,
3218 * and make sure it's a directory.
3220 if (dentry->d_sb->s_op != &cgroup_ops ||
3221 !S_ISDIR(dentry->d_inode->i_mode))
3225 cgrp = dentry->d_fsdata;
3227 cgroup_iter_start(cgrp, &it);
3228 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3229 switch (tsk->state) {
3231 stats->nr_running++;
3233 case TASK_INTERRUPTIBLE:
3234 stats->nr_sleeping++;
3236 case TASK_UNINTERRUPTIBLE:
3237 stats->nr_uninterruptible++;
3240 stats->nr_stopped++;
3243 if (delayacct_is_task_waiting_on_io(tsk))
3244 stats->nr_io_wait++;
3248 cgroup_iter_end(cgrp, &it);
3256 * seq_file methods for the tasks/procs files. The seq_file position is the
3257 * next pid to display; the seq_file iterator is a pointer to the pid
3258 * in the cgroup->l->list array.
3261 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3264 * Initially we receive a position value that corresponds to
3265 * one more than the last pid shown (or 0 on the first call or
3266 * after a seek to the start). Use a binary-search to find the
3267 * next pid to display, if any
3269 struct cgroup_pidlist *l = s->private;
3270 int index = 0, pid = *pos;
3273 down_read(&l->mutex);
3275 int end = l->length;
3277 while (index < end) {
3278 int mid = (index + end) / 2;
3279 if (l->list[mid] == pid) {
3282 } else if (l->list[mid] <= pid)
3288 /* If we're off the end of the array, we're done */
3289 if (index >= l->length)
3291 /* Update the abstract position to be the actual pid that we found */
3292 iter = l->list + index;
3297 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3299 struct cgroup_pidlist *l = s->private;
3303 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3305 struct cgroup_pidlist *l = s->private;
3307 pid_t *end = l->list + l->length;
3309 * Advance to the next pid in the array. If this goes off the
3321 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3323 return seq_printf(s, "%d\n", *(int *)v);
3327 * seq_operations functions for iterating on pidlists through seq_file -
3328 * independent of whether it's tasks or procs
3330 static const struct seq_operations cgroup_pidlist_seq_operations = {
3331 .start = cgroup_pidlist_start,
3332 .stop = cgroup_pidlist_stop,
3333 .next = cgroup_pidlist_next,
3334 .show = cgroup_pidlist_show,
3337 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3340 * the case where we're the last user of this particular pidlist will
3341 * have us remove it from the cgroup's list, which entails taking the
3342 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3343 * pidlist_mutex, we have to take pidlist_mutex first.
3345 mutex_lock(&l->owner->pidlist_mutex);
3346 down_write(&l->mutex);
3347 BUG_ON(!l->use_count);
3348 if (!--l->use_count) {
3349 /* we're the last user if refcount is 0; remove and free */
3350 list_del(&l->links);
3351 mutex_unlock(&l->owner->pidlist_mutex);
3352 pidlist_free(l->list);
3353 put_pid_ns(l->key.ns);
3354 up_write(&l->mutex);
3358 mutex_unlock(&l->owner->pidlist_mutex);
3359 up_write(&l->mutex);
3362 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3364 struct cgroup_pidlist *l;
3365 if (!(file->f_mode & FMODE_READ))
3368 * the seq_file will only be initialized if the file was opened for
3369 * reading; hence we check if it's not null only in that case.
3371 l = ((struct seq_file *)file->private_data)->private;
3372 cgroup_release_pid_array(l);
3373 return seq_release(inode, file);
3376 static const struct file_operations cgroup_pidlist_operations = {
3378 .llseek = seq_lseek,
3379 .write = cgroup_file_write,
3380 .release = cgroup_pidlist_release,
3384 * The following functions handle opens on a file that displays a pidlist
3385 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3388 /* helper function for the two below it */
3389 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3391 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3392 struct cgroup_pidlist *l;
3395 /* Nothing to do for write-only files */
3396 if (!(file->f_mode & FMODE_READ))
3399 /* have the array populated */
3400 retval = pidlist_array_load(cgrp, type, &l);
3403 /* configure file information */
3404 file->f_op = &cgroup_pidlist_operations;
3406 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3408 cgroup_release_pid_array(l);
3411 ((struct seq_file *)file->private_data)->private = l;
3414 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3416 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3418 static int cgroup_procs_open(struct inode *unused, struct file *file)
3420 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3423 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3426 return notify_on_release(cgrp);
3429 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3433 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3435 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3437 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3442 * Unregister event and free resources.
3444 * Gets called from workqueue.
3446 static void cgroup_event_remove(struct work_struct *work)
3448 struct cgroup_event *event = container_of(work, struct cgroup_event,
3450 struct cgroup *cgrp = event->cgrp;
3452 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3454 eventfd_ctx_put(event->eventfd);
3460 * Gets called on POLLHUP on eventfd when user closes it.
3462 * Called with wqh->lock held and interrupts disabled.
3464 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3465 int sync, void *key)
3467 struct cgroup_event *event = container_of(wait,
3468 struct cgroup_event, wait);
3469 struct cgroup *cgrp = event->cgrp;
3470 unsigned long flags = (unsigned long)key;
3472 if (flags & POLLHUP) {
3473 __remove_wait_queue(event->wqh, &event->wait);
3474 spin_lock(&cgrp->event_list_lock);
3475 list_del(&event->list);
3476 spin_unlock(&cgrp->event_list_lock);
3478 * We are in atomic context, but cgroup_event_remove() may
3479 * sleep, so we have to call it in workqueue.
3481 schedule_work(&event->remove);
3487 static void cgroup_event_ptable_queue_proc(struct file *file,
3488 wait_queue_head_t *wqh, poll_table *pt)
3490 struct cgroup_event *event = container_of(pt,
3491 struct cgroup_event, pt);
3494 add_wait_queue(wqh, &event->wait);
3498 * Parse input and register new cgroup event handler.
3500 * Input must be in format '<event_fd> <control_fd> <args>'.
3501 * Interpretation of args is defined by control file implementation.
3503 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3506 struct cgroup_event *event = NULL;
3507 struct cgroup *cgrp_cfile;
3508 unsigned int efd, cfd;
3509 struct file *efile = NULL;
3510 struct file *cfile = NULL;
3514 efd = simple_strtoul(buffer, &endp, 10);
3519 cfd = simple_strtoul(buffer, &endp, 10);
3520 if ((*endp != ' ') && (*endp != '\0'))
3524 event = kzalloc(sizeof(*event), GFP_KERNEL);
3528 INIT_LIST_HEAD(&event->list);
3529 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3530 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3531 INIT_WORK(&event->remove, cgroup_event_remove);
3533 efile = eventfd_fget(efd);
3534 if (IS_ERR(efile)) {
3535 ret = PTR_ERR(efile);
3539 event->eventfd = eventfd_ctx_fileget(efile);
3540 if (IS_ERR(event->eventfd)) {
3541 ret = PTR_ERR(event->eventfd);
3551 /* the process need read permission on control file */
3552 /* AV: shouldn't we check that it's been opened for read instead? */
3553 ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3557 event->cft = __file_cft(cfile);
3558 if (IS_ERR(event->cft)) {
3559 ret = PTR_ERR(event->cft);
3564 * The file to be monitored must be in the same cgroup as
3565 * cgroup.event_control is.
3567 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
3568 if (cgrp_cfile != cgrp) {
3573 if (!event->cft->register_event || !event->cft->unregister_event) {
3578 ret = event->cft->register_event(cgrp, event->cft,
3579 event->eventfd, buffer);
3583 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3584 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3590 * Events should be removed after rmdir of cgroup directory, but before
3591 * destroying subsystem state objects. Let's take reference to cgroup
3592 * directory dentry to do that.
3596 spin_lock(&cgrp->event_list_lock);
3597 list_add(&event->list, &cgrp->event_list);
3598 spin_unlock(&cgrp->event_list_lock);
3609 if (event && event->eventfd && !IS_ERR(event->eventfd))
3610 eventfd_ctx_put(event->eventfd);
3612 if (!IS_ERR_OR_NULL(efile))
3620 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3623 return clone_children(cgrp);
3626 static int cgroup_clone_children_write(struct cgroup *cgrp,
3631 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3633 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3638 * for the common functions, 'private' gives the type of file
3640 /* for hysterical raisins, we can't put this on the older files */
3641 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3642 static struct cftype files[] = {
3645 .open = cgroup_tasks_open,
3646 .write_u64 = cgroup_tasks_write,
3647 .release = cgroup_pidlist_release,
3648 .mode = S_IRUGO | S_IWUSR,
3651 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3652 .open = cgroup_procs_open,
3653 .write_u64 = cgroup_procs_write,
3654 .release = cgroup_pidlist_release,
3655 .mode = S_IRUGO | S_IWUSR,
3658 .name = "notify_on_release",
3659 .read_u64 = cgroup_read_notify_on_release,
3660 .write_u64 = cgroup_write_notify_on_release,
3663 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3664 .write_string = cgroup_write_event_control,
3668 .name = "cgroup.clone_children",
3669 .read_u64 = cgroup_clone_children_read,
3670 .write_u64 = cgroup_clone_children_write,
3674 static struct cftype cft_release_agent = {
3675 .name = "release_agent",
3676 .read_seq_string = cgroup_release_agent_show,
3677 .write_string = cgroup_release_agent_write,
3678 .max_write_len = PATH_MAX,
3681 static int cgroup_populate_dir(struct cgroup *cgrp)
3684 struct cgroup_subsys *ss;
3686 /* First clear out any existing files */
3687 cgroup_clear_directory(cgrp->dentry);
3689 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
3693 if (cgrp == cgrp->top_cgroup) {
3694 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
3698 for_each_subsys(cgrp->root, ss) {
3699 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
3702 /* This cgroup is ready now */
3703 for_each_subsys(cgrp->root, ss) {
3704 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3706 * Update id->css pointer and make this css visible from
3707 * CSS ID functions. This pointer will be dereferened
3708 * from RCU-read-side without locks.
3711 rcu_assign_pointer(css->id->css, css);
3717 static void init_cgroup_css(struct cgroup_subsys_state *css,
3718 struct cgroup_subsys *ss,
3719 struct cgroup *cgrp)
3722 atomic_set(&css->refcnt, 1);
3725 if (cgrp == dummytop)
3726 set_bit(CSS_ROOT, &css->flags);
3727 BUG_ON(cgrp->subsys[ss->subsys_id]);
3728 cgrp->subsys[ss->subsys_id] = css;
3731 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3733 /* We need to take each hierarchy_mutex in a consistent order */
3737 * No worry about a race with rebind_subsystems that might mess up the
3738 * locking order, since both parties are under cgroup_mutex.
3740 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3741 struct cgroup_subsys *ss = subsys[i];
3744 if (ss->root == root)
3745 mutex_lock(&ss->hierarchy_mutex);
3749 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3753 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3754 struct cgroup_subsys *ss = subsys[i];
3757 if (ss->root == root)
3758 mutex_unlock(&ss->hierarchy_mutex);
3763 * cgroup_create - create a cgroup
3764 * @parent: cgroup that will be parent of the new cgroup
3765 * @dentry: dentry of the new cgroup
3766 * @mode: mode to set on new inode
3768 * Must be called with the mutex on the parent inode held
3770 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3773 struct cgroup *cgrp;
3774 struct cgroupfs_root *root = parent->root;
3776 struct cgroup_subsys *ss;
3777 struct super_block *sb = root->sb;
3779 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3783 /* Grab a reference on the superblock so the hierarchy doesn't
3784 * get deleted on unmount if there are child cgroups. This
3785 * can be done outside cgroup_mutex, since the sb can't
3786 * disappear while someone has an open control file on the
3788 atomic_inc(&sb->s_active);
3790 mutex_lock(&cgroup_mutex);
3792 init_cgroup_housekeeping(cgrp);
3794 cgrp->parent = parent;
3795 cgrp->root = parent->root;
3796 cgrp->top_cgroup = parent->top_cgroup;
3798 if (notify_on_release(parent))
3799 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3801 if (clone_children(parent))
3802 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3804 for_each_subsys(root, ss) {
3805 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
3811 init_cgroup_css(css, ss, cgrp);
3813 err = alloc_css_id(ss, parent, cgrp);
3817 /* At error, ->destroy() callback has to free assigned ID. */
3818 if (clone_children(parent) && ss->post_clone)
3819 ss->post_clone(ss, cgrp);
3822 cgroup_lock_hierarchy(root);
3823 list_add(&cgrp->sibling, &cgrp->parent->children);
3824 cgroup_unlock_hierarchy(root);
3825 root->number_of_cgroups++;
3827 err = cgroup_create_dir(cgrp, dentry, mode);
3831 /* The cgroup directory was pre-locked for us */
3832 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3834 err = cgroup_populate_dir(cgrp);
3835 /* If err < 0, we have a half-filled directory - oh well ;) */
3837 mutex_unlock(&cgroup_mutex);
3838 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3844 cgroup_lock_hierarchy(root);
3845 list_del(&cgrp->sibling);
3846 cgroup_unlock_hierarchy(root);
3847 root->number_of_cgroups--;
3851 for_each_subsys(root, ss) {
3852 if (cgrp->subsys[ss->subsys_id])
3853 ss->destroy(ss, cgrp);
3856 mutex_unlock(&cgroup_mutex);
3858 /* Release the reference count that we took on the superblock */
3859 deactivate_super(sb);
3865 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3867 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3869 /* the vfs holds inode->i_mutex already */
3870 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3873 static int cgroup_has_css_refs(struct cgroup *cgrp)
3875 /* Check the reference count on each subsystem. Since we
3876 * already established that there are no tasks in the
3877 * cgroup, if the css refcount is also 1, then there should
3878 * be no outstanding references, so the subsystem is safe to
3879 * destroy. We scan across all subsystems rather than using
3880 * the per-hierarchy linked list of mounted subsystems since
3881 * we can be called via check_for_release() with no
3882 * synchronization other than RCU, and the subsystem linked
3883 * list isn't RCU-safe */
3886 * We won't need to lock the subsys array, because the subsystems
3887 * we're concerned about aren't going anywhere since our cgroup root
3888 * has a reference on them.
3890 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3891 struct cgroup_subsys *ss = subsys[i];
3892 struct cgroup_subsys_state *css;
3893 /* Skip subsystems not present or not in this hierarchy */
3894 if (ss == NULL || ss->root != cgrp->root)
3896 css = cgrp->subsys[ss->subsys_id];
3897 /* When called from check_for_release() it's possible
3898 * that by this point the cgroup has been removed
3899 * and the css deleted. But a false-positive doesn't
3900 * matter, since it can only happen if the cgroup
3901 * has been deleted and hence no longer needs the
3902 * release agent to be called anyway. */
3903 if (css && (atomic_read(&css->refcnt) > 1))
3910 * Atomically mark all (or else none) of the cgroup's CSS objects as
3911 * CSS_REMOVED. Return true on success, or false if the cgroup has
3912 * busy subsystems. Call with cgroup_mutex held
3915 static int cgroup_clear_css_refs(struct cgroup *cgrp)
3917 struct cgroup_subsys *ss;
3918 unsigned long flags;
3919 bool failed = false;
3920 local_irq_save(flags);
3921 for_each_subsys(cgrp->root, ss) {
3922 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3925 /* We can only remove a CSS with a refcnt==1 */
3926 refcnt = atomic_read(&css->refcnt);
3933 * Drop the refcnt to 0 while we check other
3934 * subsystems. This will cause any racing
3935 * css_tryget() to spin until we set the
3936 * CSS_REMOVED bits or abort
3938 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
3944 for_each_subsys(cgrp->root, ss) {
3945 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3948 * Restore old refcnt if we previously managed
3949 * to clear it from 1 to 0
3951 if (!atomic_read(&css->refcnt))
3952 atomic_set(&css->refcnt, 1);
3954 /* Commit the fact that the CSS is removed */
3955 set_bit(CSS_REMOVED, &css->flags);
3958 local_irq_restore(flags);
3962 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
3964 struct cgroup *cgrp = dentry->d_fsdata;
3966 struct cgroup *parent;
3968 struct cgroup_event *event, *tmp;
3971 /* the vfs holds both inode->i_mutex already */
3973 mutex_lock(&cgroup_mutex);
3974 if (atomic_read(&cgrp->count) != 0) {
3975 mutex_unlock(&cgroup_mutex);
3978 if (!list_empty(&cgrp->children)) {
3979 mutex_unlock(&cgroup_mutex);
3982 mutex_unlock(&cgroup_mutex);
3985 * In general, subsystem has no css->refcnt after pre_destroy(). But
3986 * in racy cases, subsystem may have to get css->refcnt after
3987 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3988 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3989 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3990 * and subsystem's reference count handling. Please see css_get/put
3991 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3993 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
3996 * Call pre_destroy handlers of subsys. Notify subsystems
3997 * that rmdir() request comes.
3999 ret = cgroup_call_pre_destroy(cgrp);
4001 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4005 mutex_lock(&cgroup_mutex);
4006 parent = cgrp->parent;
4007 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
4008 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4009 mutex_unlock(&cgroup_mutex);
4012 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
4013 if (!cgroup_clear_css_refs(cgrp)) {
4014 mutex_unlock(&cgroup_mutex);
4016 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4017 * prepare_to_wait(), we need to check this flag.
4019 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
4021 finish_wait(&cgroup_rmdir_waitq, &wait);
4022 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4023 if (signal_pending(current))
4027 /* NO css_tryget() can success after here. */
4028 finish_wait(&cgroup_rmdir_waitq, &wait);
4029 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4031 raw_spin_lock(&release_list_lock);
4032 set_bit(CGRP_REMOVED, &cgrp->flags);
4033 if (!list_empty(&cgrp->release_list))
4034 list_del_init(&cgrp->release_list);
4035 raw_spin_unlock(&release_list_lock);
4037 cgroup_lock_hierarchy(cgrp->root);
4038 /* delete this cgroup from parent->children */
4039 list_del_init(&cgrp->sibling);
4040 cgroup_unlock_hierarchy(cgrp->root);
4042 d = dget(cgrp->dentry);
4044 cgroup_d_remove_dir(d);
4047 set_bit(CGRP_RELEASABLE, &parent->flags);
4048 check_for_release(parent);
4051 * Unregister events and notify userspace.
4052 * Notify userspace about cgroup removing only after rmdir of cgroup
4053 * directory to avoid race between userspace and kernelspace
4055 spin_lock(&cgrp->event_list_lock);
4056 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4057 list_del(&event->list);
4058 remove_wait_queue(event->wqh, &event->wait);
4059 eventfd_signal(event->eventfd, 1);
4060 schedule_work(&event->remove);
4062 spin_unlock(&cgrp->event_list_lock);
4064 mutex_unlock(&cgroup_mutex);
4068 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4070 struct cgroup_subsys_state *css;
4072 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4074 /* Create the top cgroup state for this subsystem */
4075 list_add(&ss->sibling, &rootnode.subsys_list);
4076 ss->root = &rootnode;
4077 css = ss->create(ss, dummytop);
4078 /* We don't handle early failures gracefully */
4079 BUG_ON(IS_ERR(css));
4080 init_cgroup_css(css, ss, dummytop);
4082 /* Update the init_css_set to contain a subsys
4083 * pointer to this state - since the subsystem is
4084 * newly registered, all tasks and hence the
4085 * init_css_set is in the subsystem's top cgroup. */
4086 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4088 need_forkexit_callback |= ss->fork || ss->exit;
4090 /* At system boot, before all subsystems have been
4091 * registered, no tasks have been forked, so we don't
4092 * need to invoke fork callbacks here. */
4093 BUG_ON(!list_empty(&init_task.tasks));
4095 mutex_init(&ss->hierarchy_mutex);
4096 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4099 /* this function shouldn't be used with modular subsystems, since they
4100 * need to register a subsys_id, among other things */
4105 * cgroup_load_subsys: load and register a modular subsystem at runtime
4106 * @ss: the subsystem to load
4108 * This function should be called in a modular subsystem's initcall. If the
4109 * subsystem is built as a module, it will be assigned a new subsys_id and set
4110 * up for use. If the subsystem is built-in anyway, work is delegated to the
4111 * simpler cgroup_init_subsys.
4113 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4116 struct cgroup_subsys_state *css;
4118 /* check name and function validity */
4119 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4120 ss->create == NULL || ss->destroy == NULL)
4124 * we don't support callbacks in modular subsystems. this check is
4125 * before the ss->module check for consistency; a subsystem that could
4126 * be a module should still have no callbacks even if the user isn't
4127 * compiling it as one.
4129 if (ss->fork || ss->exit)
4133 * an optionally modular subsystem is built-in: we want to do nothing,
4134 * since cgroup_init_subsys will have already taken care of it.
4136 if (ss->module == NULL) {
4137 /* a few sanity checks */
4138 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4139 BUG_ON(subsys[ss->subsys_id] != ss);
4144 * need to register a subsys id before anything else - for example,
4145 * init_cgroup_css needs it.
4147 mutex_lock(&cgroup_mutex);
4148 /* find the first empty slot in the array */
4149 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4150 if (subsys[i] == NULL)
4153 if (i == CGROUP_SUBSYS_COUNT) {
4154 /* maximum number of subsystems already registered! */
4155 mutex_unlock(&cgroup_mutex);
4158 /* assign ourselves the subsys_id */
4163 * no ss->create seems to need anything important in the ss struct, so
4164 * this can happen first (i.e. before the rootnode attachment).
4166 css = ss->create(ss, dummytop);
4168 /* failure case - need to deassign the subsys[] slot. */
4170 mutex_unlock(&cgroup_mutex);
4171 return PTR_ERR(css);
4174 list_add(&ss->sibling, &rootnode.subsys_list);
4175 ss->root = &rootnode;
4177 /* our new subsystem will be attached to the dummy hierarchy. */
4178 init_cgroup_css(css, ss, dummytop);
4179 /* init_idr must be after init_cgroup_css because it sets css->id. */
4181 int ret = cgroup_init_idr(ss, css);
4183 dummytop->subsys[ss->subsys_id] = NULL;
4184 ss->destroy(ss, dummytop);
4186 mutex_unlock(&cgroup_mutex);
4192 * Now we need to entangle the css into the existing css_sets. unlike
4193 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4194 * will need a new pointer to it; done by iterating the css_set_table.
4195 * furthermore, modifying the existing css_sets will corrupt the hash
4196 * table state, so each changed css_set will need its hash recomputed.
4197 * this is all done under the css_set_lock.
4199 write_lock(&css_set_lock);
4200 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4202 struct hlist_node *node, *tmp;
4203 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4205 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4206 /* skip entries that we already rehashed */
4207 if (cg->subsys[ss->subsys_id])
4209 /* remove existing entry */
4210 hlist_del(&cg->hlist);
4212 cg->subsys[ss->subsys_id] = css;
4213 /* recompute hash and restore entry */
4214 new_bucket = css_set_hash(cg->subsys);
4215 hlist_add_head(&cg->hlist, new_bucket);
4218 write_unlock(&css_set_lock);
4220 mutex_init(&ss->hierarchy_mutex);
4221 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4225 mutex_unlock(&cgroup_mutex);
4228 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4231 * cgroup_unload_subsys: unload a modular subsystem
4232 * @ss: the subsystem to unload
4234 * This function should be called in a modular subsystem's exitcall. When this
4235 * function is invoked, the refcount on the subsystem's module will be 0, so
4236 * the subsystem will not be attached to any hierarchy.
4238 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4240 struct cg_cgroup_link *link;
4241 struct hlist_head *hhead;
4243 BUG_ON(ss->module == NULL);
4246 * we shouldn't be called if the subsystem is in use, and the use of
4247 * try_module_get in parse_cgroupfs_options should ensure that it
4248 * doesn't start being used while we're killing it off.
4250 BUG_ON(ss->root != &rootnode);
4252 mutex_lock(&cgroup_mutex);
4253 /* deassign the subsys_id */
4254 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4255 subsys[ss->subsys_id] = NULL;
4257 /* remove subsystem from rootnode's list of subsystems */
4258 list_del_init(&ss->sibling);
4261 * disentangle the css from all css_sets attached to the dummytop. as
4262 * in loading, we need to pay our respects to the hashtable gods.
4264 write_lock(&css_set_lock);
4265 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4266 struct css_set *cg = link->cg;
4268 hlist_del(&cg->hlist);
4269 BUG_ON(!cg->subsys[ss->subsys_id]);
4270 cg->subsys[ss->subsys_id] = NULL;
4271 hhead = css_set_hash(cg->subsys);
4272 hlist_add_head(&cg->hlist, hhead);
4274 write_unlock(&css_set_lock);
4277 * remove subsystem's css from the dummytop and free it - need to free
4278 * before marking as null because ss->destroy needs the cgrp->subsys
4279 * pointer to find their state. note that this also takes care of
4280 * freeing the css_id.
4282 ss->destroy(ss, dummytop);
4283 dummytop->subsys[ss->subsys_id] = NULL;
4285 mutex_unlock(&cgroup_mutex);
4287 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4290 * cgroup_init_early - cgroup initialization at system boot
4292 * Initialize cgroups at system boot, and initialize any
4293 * subsystems that request early init.
4295 int __init cgroup_init_early(void)
4298 atomic_set(&init_css_set.refcount, 1);
4299 INIT_LIST_HEAD(&init_css_set.cg_links);
4300 INIT_LIST_HEAD(&init_css_set.tasks);
4301 INIT_HLIST_NODE(&init_css_set.hlist);
4303 init_cgroup_root(&rootnode);
4305 init_task.cgroups = &init_css_set;
4307 init_css_set_link.cg = &init_css_set;
4308 init_css_set_link.cgrp = dummytop;
4309 list_add(&init_css_set_link.cgrp_link_list,
4310 &rootnode.top_cgroup.css_sets);
4311 list_add(&init_css_set_link.cg_link_list,
4312 &init_css_set.cg_links);
4314 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4315 INIT_HLIST_HEAD(&css_set_table[i]);
4317 /* at bootup time, we don't worry about modular subsystems */
4318 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4319 struct cgroup_subsys *ss = subsys[i];
4322 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4323 BUG_ON(!ss->create);
4324 BUG_ON(!ss->destroy);
4325 if (ss->subsys_id != i) {
4326 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4327 ss->name, ss->subsys_id);
4332 cgroup_init_subsys(ss);
4338 * cgroup_init - cgroup initialization
4340 * Register cgroup filesystem and /proc file, and initialize
4341 * any subsystems that didn't request early init.
4343 int __init cgroup_init(void)
4347 struct hlist_head *hhead;
4349 err = bdi_init(&cgroup_backing_dev_info);
4353 /* at bootup time, we don't worry about modular subsystems */
4354 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4355 struct cgroup_subsys *ss = subsys[i];
4356 if (!ss->early_init)
4357 cgroup_init_subsys(ss);
4359 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4362 /* Add init_css_set to the hash table */
4363 hhead = css_set_hash(init_css_set.subsys);
4364 hlist_add_head(&init_css_set.hlist, hhead);
4365 BUG_ON(!init_root_id(&rootnode));
4367 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4373 err = register_filesystem(&cgroup_fs_type);
4375 kobject_put(cgroup_kobj);
4379 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4383 bdi_destroy(&cgroup_backing_dev_info);
4389 * proc_cgroup_show()
4390 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4391 * - Used for /proc/<pid>/cgroup.
4392 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4393 * doesn't really matter if tsk->cgroup changes after we read it,
4394 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4395 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4396 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4397 * cgroup to top_cgroup.
4400 /* TODO: Use a proper seq_file iterator */
4401 static int proc_cgroup_show(struct seq_file *m, void *v)
4404 struct task_struct *tsk;
4407 struct cgroupfs_root *root;
4410 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4416 tsk = get_pid_task(pid, PIDTYPE_PID);
4422 mutex_lock(&cgroup_mutex);
4424 for_each_active_root(root) {
4425 struct cgroup_subsys *ss;
4426 struct cgroup *cgrp;
4429 seq_printf(m, "%d:", root->hierarchy_id);
4430 for_each_subsys(root, ss)
4431 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4432 if (strlen(root->name))
4433 seq_printf(m, "%sname=%s", count ? "," : "",
4436 cgrp = task_cgroup_from_root(tsk, root);
4437 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4445 mutex_unlock(&cgroup_mutex);
4446 put_task_struct(tsk);
4453 static int cgroup_open(struct inode *inode, struct file *file)
4455 struct pid *pid = PROC_I(inode)->pid;
4456 return single_open(file, proc_cgroup_show, pid);
4459 const struct file_operations proc_cgroup_operations = {
4460 .open = cgroup_open,
4462 .llseek = seq_lseek,
4463 .release = single_release,
4466 /* Display information about each subsystem and each hierarchy */
4467 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4471 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4473 * ideally we don't want subsystems moving around while we do this.
4474 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4475 * subsys/hierarchy state.
4477 mutex_lock(&cgroup_mutex);
4478 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4479 struct cgroup_subsys *ss = subsys[i];
4482 seq_printf(m, "%s\t%d\t%d\t%d\n",
4483 ss->name, ss->root->hierarchy_id,
4484 ss->root->number_of_cgroups, !ss->disabled);
4486 mutex_unlock(&cgroup_mutex);
4490 static int cgroupstats_open(struct inode *inode, struct file *file)
4492 return single_open(file, proc_cgroupstats_show, NULL);
4495 static const struct file_operations proc_cgroupstats_operations = {
4496 .open = cgroupstats_open,
4498 .llseek = seq_lseek,
4499 .release = single_release,
4503 * cgroup_fork - attach newly forked task to its parents cgroup.
4504 * @child: pointer to task_struct of forking parent process.
4506 * Description: A task inherits its parent's cgroup at fork().
4508 * A pointer to the shared css_set was automatically copied in
4509 * fork.c by dup_task_struct(). However, we ignore that copy, since
4510 * it was not made under the protection of RCU or cgroup_mutex, so
4511 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4512 * have already changed current->cgroups, allowing the previously
4513 * referenced cgroup group to be removed and freed.
4515 * At the point that cgroup_fork() is called, 'current' is the parent
4516 * task, and the passed argument 'child' points to the child task.
4518 void cgroup_fork(struct task_struct *child)
4521 child->cgroups = current->cgroups;
4522 get_css_set(child->cgroups);
4523 task_unlock(current);
4524 INIT_LIST_HEAD(&child->cg_list);
4528 * cgroup_post_fork - called on a new task after adding it to the task list
4529 * @child: the task in question
4531 * Adds the task to the list running through its css_set if necessary and
4532 * call the subsystem fork() callbacks. Has to be after the task is
4533 * visible on the task list in case we race with the first call to
4534 * cgroup_iter_start() - to guarantee that the new task ends up on its
4537 void cgroup_post_fork(struct task_struct *child)
4541 if (use_task_css_set_links) {
4542 write_lock(&css_set_lock);
4544 if (list_empty(&child->cg_list))
4545 list_add(&child->cg_list, &child->cgroups->tasks);
4547 write_unlock(&css_set_lock);
4551 * Call ss->fork(). This must happen after @child is linked on
4552 * css_set; otherwise, @child might change state between ->fork()
4553 * and addition to css_set.
4555 if (need_forkexit_callback) {
4556 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4557 struct cgroup_subsys *ss = subsys[i];
4559 ss->fork(ss, child);
4565 * cgroup_exit - detach cgroup from exiting task
4566 * @tsk: pointer to task_struct of exiting process
4567 * @run_callback: run exit callbacks?
4569 * Description: Detach cgroup from @tsk and release it.
4571 * Note that cgroups marked notify_on_release force every task in
4572 * them to take the global cgroup_mutex mutex when exiting.
4573 * This could impact scaling on very large systems. Be reluctant to
4574 * use notify_on_release cgroups where very high task exit scaling
4575 * is required on large systems.
4577 * the_top_cgroup_hack:
4579 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4581 * We call cgroup_exit() while the task is still competent to
4582 * handle notify_on_release(), then leave the task attached to the
4583 * root cgroup in each hierarchy for the remainder of its exit.
4585 * To do this properly, we would increment the reference count on
4586 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4587 * code we would add a second cgroup function call, to drop that
4588 * reference. This would just create an unnecessary hot spot on
4589 * the top_cgroup reference count, to no avail.
4591 * Normally, holding a reference to a cgroup without bumping its
4592 * count is unsafe. The cgroup could go away, or someone could
4593 * attach us to a different cgroup, decrementing the count on
4594 * the first cgroup that we never incremented. But in this case,
4595 * top_cgroup isn't going away, and either task has PF_EXITING set,
4596 * which wards off any cgroup_attach_task() attempts, or task is a failed
4597 * fork, never visible to cgroup_attach_task.
4599 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4605 * Unlink from the css_set task list if necessary.
4606 * Optimistically check cg_list before taking
4609 if (!list_empty(&tsk->cg_list)) {
4610 write_lock(&css_set_lock);
4611 if (!list_empty(&tsk->cg_list))
4612 list_del_init(&tsk->cg_list);
4613 write_unlock(&css_set_lock);
4616 /* Reassign the task to the init_css_set. */
4619 tsk->cgroups = &init_css_set;
4621 if (run_callbacks && need_forkexit_callback) {
4623 * modular subsystems can't use callbacks, so no need to lock
4626 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4627 struct cgroup_subsys *ss = subsys[i];
4629 struct cgroup *old_cgrp =
4630 rcu_dereference_raw(cg->subsys[i])->cgroup;
4631 struct cgroup *cgrp = task_cgroup(tsk, i);
4632 ss->exit(ss, cgrp, old_cgrp, tsk);
4639 put_css_set_taskexit(cg);
4643 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4644 * @cgrp: the cgroup in question
4645 * @task: the task in question
4647 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4650 * If we are sending in dummytop, then presumably we are creating
4651 * the top cgroup in the subsystem.
4653 * Called only by the ns (nsproxy) cgroup.
4655 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4658 struct cgroup *target;
4660 if (cgrp == dummytop)
4663 target = task_cgroup_from_root(task, cgrp->root);
4664 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4665 cgrp = cgrp->parent;
4666 ret = (cgrp == target);
4670 static void check_for_release(struct cgroup *cgrp)
4672 /* All of these checks rely on RCU to keep the cgroup
4673 * structure alive */
4674 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4675 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4676 /* Control Group is currently removeable. If it's not
4677 * already queued for a userspace notification, queue
4679 int need_schedule_work = 0;
4680 raw_spin_lock(&release_list_lock);
4681 if (!cgroup_is_removed(cgrp) &&
4682 list_empty(&cgrp->release_list)) {
4683 list_add(&cgrp->release_list, &release_list);
4684 need_schedule_work = 1;
4686 raw_spin_unlock(&release_list_lock);
4687 if (need_schedule_work)
4688 schedule_work(&release_agent_work);
4692 /* Caller must verify that the css is not for root cgroup */
4693 void __css_put(struct cgroup_subsys_state *css, int count)
4695 struct cgroup *cgrp = css->cgroup;
4698 val = atomic_sub_return(count, &css->refcnt);
4700 if (notify_on_release(cgrp)) {
4701 set_bit(CGRP_RELEASABLE, &cgrp->flags);
4702 check_for_release(cgrp);
4704 cgroup_wakeup_rmdir_waiter(cgrp);
4707 WARN_ON_ONCE(val < 1);
4709 EXPORT_SYMBOL_GPL(__css_put);
4712 * Notify userspace when a cgroup is released, by running the
4713 * configured release agent with the name of the cgroup (path
4714 * relative to the root of cgroup file system) as the argument.
4716 * Most likely, this user command will try to rmdir this cgroup.
4718 * This races with the possibility that some other task will be
4719 * attached to this cgroup before it is removed, or that some other
4720 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4721 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4722 * unused, and this cgroup will be reprieved from its death sentence,
4723 * to continue to serve a useful existence. Next time it's released,
4724 * we will get notified again, if it still has 'notify_on_release' set.
4726 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4727 * means only wait until the task is successfully execve()'d. The
4728 * separate release agent task is forked by call_usermodehelper(),
4729 * then control in this thread returns here, without waiting for the
4730 * release agent task. We don't bother to wait because the caller of
4731 * this routine has no use for the exit status of the release agent
4732 * task, so no sense holding our caller up for that.
4734 static void cgroup_release_agent(struct work_struct *work)
4736 BUG_ON(work != &release_agent_work);
4737 mutex_lock(&cgroup_mutex);
4738 raw_spin_lock(&release_list_lock);
4739 while (!list_empty(&release_list)) {
4740 char *argv[3], *envp[3];
4742 char *pathbuf = NULL, *agentbuf = NULL;
4743 struct cgroup *cgrp = list_entry(release_list.next,
4746 list_del_init(&cgrp->release_list);
4747 raw_spin_unlock(&release_list_lock);
4748 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4751 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4753 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4758 argv[i++] = agentbuf;
4759 argv[i++] = pathbuf;
4763 /* minimal command environment */
4764 envp[i++] = "HOME=/";
4765 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4768 /* Drop the lock while we invoke the usermode helper,
4769 * since the exec could involve hitting disk and hence
4770 * be a slow process */
4771 mutex_unlock(&cgroup_mutex);
4772 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4773 mutex_lock(&cgroup_mutex);
4777 raw_spin_lock(&release_list_lock);
4779 raw_spin_unlock(&release_list_lock);
4780 mutex_unlock(&cgroup_mutex);
4783 static int __init cgroup_disable(char *str)
4788 while ((token = strsep(&str, ",")) != NULL) {
4792 * cgroup_disable, being at boot time, can't know about module
4793 * subsystems, so we don't worry about them.
4795 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4796 struct cgroup_subsys *ss = subsys[i];
4798 if (!strcmp(token, ss->name)) {
4800 printk(KERN_INFO "Disabling %s control group"
4801 " subsystem\n", ss->name);
4808 __setup("cgroup_disable=", cgroup_disable);
4811 * Functons for CSS ID.
4815 *To get ID other than 0, this should be called when !cgroup_is_removed().
4817 unsigned short css_id(struct cgroup_subsys_state *css)
4819 struct css_id *cssid;
4822 * This css_id() can return correct value when somone has refcnt
4823 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4824 * it's unchanged until freed.
4826 cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt));
4832 EXPORT_SYMBOL_GPL(css_id);
4834 unsigned short css_depth(struct cgroup_subsys_state *css)
4836 struct css_id *cssid;
4838 cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt));
4841 return cssid->depth;
4844 EXPORT_SYMBOL_GPL(css_depth);
4847 * css_is_ancestor - test "root" css is an ancestor of "child"
4848 * @child: the css to be tested.
4849 * @root: the css supporsed to be an ancestor of the child.
4851 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4852 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4853 * But, considering usual usage, the csses should be valid objects after test.
4854 * Assuming that the caller will do some action to the child if this returns
4855 * returns true, the caller must take "child";s reference count.
4856 * If "child" is valid object and this returns true, "root" is valid, too.
4859 bool css_is_ancestor(struct cgroup_subsys_state *child,
4860 const struct cgroup_subsys_state *root)
4862 struct css_id *child_id;
4863 struct css_id *root_id;
4867 child_id = rcu_dereference(child->id);
4868 root_id = rcu_dereference(root->id);
4871 || (child_id->depth < root_id->depth)
4872 || (child_id->stack[root_id->depth] != root_id->id))
4878 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
4880 struct css_id *id = css->id;
4881 /* When this is called before css_id initialization, id can be NULL */
4885 BUG_ON(!ss->use_id);
4887 rcu_assign_pointer(id->css, NULL);
4888 rcu_assign_pointer(css->id, NULL);
4889 write_lock(&ss->id_lock);
4890 idr_remove(&ss->idr, id->id);
4891 write_unlock(&ss->id_lock);
4892 kfree_rcu(id, rcu_head);
4894 EXPORT_SYMBOL_GPL(free_css_id);
4897 * This is called by init or create(). Then, calls to this function are
4898 * always serialized (By cgroup_mutex() at create()).
4901 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
4903 struct css_id *newid;
4904 int myid, error, size;
4906 BUG_ON(!ss->use_id);
4908 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
4909 newid = kzalloc(size, GFP_KERNEL);
4911 return ERR_PTR(-ENOMEM);
4913 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
4917 write_lock(&ss->id_lock);
4918 /* Don't use 0. allocates an ID of 1-65535 */
4919 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
4920 write_unlock(&ss->id_lock);
4922 /* Returns error when there are no free spaces for new ID.*/
4927 if (myid > CSS_ID_MAX)
4931 newid->depth = depth;
4935 write_lock(&ss->id_lock);
4936 idr_remove(&ss->idr, myid);
4937 write_unlock(&ss->id_lock);
4940 return ERR_PTR(error);
4944 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
4945 struct cgroup_subsys_state *rootcss)
4947 struct css_id *newid;
4949 rwlock_init(&ss->id_lock);
4952 newid = get_new_cssid(ss, 0);
4954 return PTR_ERR(newid);
4956 newid->stack[0] = newid->id;
4957 newid->css = rootcss;
4958 rootcss->id = newid;
4962 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
4963 struct cgroup *child)
4965 int subsys_id, i, depth = 0;
4966 struct cgroup_subsys_state *parent_css, *child_css;
4967 struct css_id *child_id, *parent_id;
4969 subsys_id = ss->subsys_id;
4970 parent_css = parent->subsys[subsys_id];
4971 child_css = child->subsys[subsys_id];
4972 parent_id = parent_css->id;
4973 depth = parent_id->depth + 1;
4975 child_id = get_new_cssid(ss, depth);
4976 if (IS_ERR(child_id))
4977 return PTR_ERR(child_id);
4979 for (i = 0; i < depth; i++)
4980 child_id->stack[i] = parent_id->stack[i];
4981 child_id->stack[depth] = child_id->id;
4983 * child_id->css pointer will be set after this cgroup is available
4984 * see cgroup_populate_dir()
4986 rcu_assign_pointer(child_css->id, child_id);
4992 * css_lookup - lookup css by id
4993 * @ss: cgroup subsys to be looked into.
4996 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4997 * NULL if not. Should be called under rcu_read_lock()
4999 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5001 struct css_id *cssid = NULL;
5003 BUG_ON(!ss->use_id);
5004 cssid = idr_find(&ss->idr, id);
5006 if (unlikely(!cssid))
5009 return rcu_dereference(cssid->css);
5011 EXPORT_SYMBOL_GPL(css_lookup);
5014 * css_get_next - lookup next cgroup under specified hierarchy.
5015 * @ss: pointer to subsystem
5016 * @id: current position of iteration.
5017 * @root: pointer to css. search tree under this.
5018 * @foundid: position of found object.
5020 * Search next css under the specified hierarchy of rootid. Calling under
5021 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5023 struct cgroup_subsys_state *
5024 css_get_next(struct cgroup_subsys *ss, int id,
5025 struct cgroup_subsys_state *root, int *foundid)
5027 struct cgroup_subsys_state *ret = NULL;
5030 int rootid = css_id(root);
5031 int depth = css_depth(root);
5036 BUG_ON(!ss->use_id);
5037 /* fill start point for scan */
5041 * scan next entry from bitmap(tree), tmpid is updated after
5044 read_lock(&ss->id_lock);
5045 tmp = idr_get_next(&ss->idr, &tmpid);
5046 read_unlock(&ss->id_lock);
5050 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5051 ret = rcu_dereference(tmp->css);
5057 /* continue to scan from next id */
5064 * get corresponding css from file open on cgroupfs directory
5066 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5068 struct cgroup *cgrp;
5069 struct inode *inode;
5070 struct cgroup_subsys_state *css;
5072 inode = f->f_dentry->d_inode;
5073 /* check in cgroup filesystem dir */
5074 if (inode->i_op != &cgroup_dir_inode_operations)
5075 return ERR_PTR(-EBADF);
5077 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5078 return ERR_PTR(-EINVAL);
5081 cgrp = __d_cgrp(f->f_dentry);
5082 css = cgrp->subsys[id];
5083 return css ? css : ERR_PTR(-ENOENT);
5086 #ifdef CONFIG_CGROUP_DEBUG
5087 static struct cgroup_subsys_state *debug_create(struct cgroup_subsys *ss,
5088 struct cgroup *cont)
5090 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5093 return ERR_PTR(-ENOMEM);
5098 static void debug_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
5100 kfree(cont->subsys[debug_subsys_id]);
5103 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5105 return atomic_read(&cont->count);
5108 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5110 return cgroup_task_count(cont);
5113 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5115 return (u64)(unsigned long)current->cgroups;
5118 static u64 current_css_set_refcount_read(struct cgroup *cont,
5124 count = atomic_read(¤t->cgroups->refcount);
5129 static int current_css_set_cg_links_read(struct cgroup *cont,
5131 struct seq_file *seq)
5133 struct cg_cgroup_link *link;
5136 read_lock(&css_set_lock);
5138 cg = rcu_dereference(current->cgroups);
5139 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5140 struct cgroup *c = link->cgrp;
5144 name = c->dentry->d_name.name;
5147 seq_printf(seq, "Root %d group %s\n",
5148 c->root->hierarchy_id, name);
5151 read_unlock(&css_set_lock);
5155 #define MAX_TASKS_SHOWN_PER_CSS 25
5156 static int cgroup_css_links_read(struct cgroup *cont,
5158 struct seq_file *seq)
5160 struct cg_cgroup_link *link;
5162 read_lock(&css_set_lock);
5163 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5164 struct css_set *cg = link->cg;
5165 struct task_struct *task;
5167 seq_printf(seq, "css_set %p\n", cg);
5168 list_for_each_entry(task, &cg->tasks, cg_list) {
5169 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5170 seq_puts(seq, " ...\n");
5173 seq_printf(seq, " task %d\n",
5174 task_pid_vnr(task));
5178 read_unlock(&css_set_lock);
5182 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5184 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5187 static struct cftype debug_files[] = {
5189 .name = "cgroup_refcount",
5190 .read_u64 = cgroup_refcount_read,
5193 .name = "taskcount",
5194 .read_u64 = debug_taskcount_read,
5198 .name = "current_css_set",
5199 .read_u64 = current_css_set_read,
5203 .name = "current_css_set_refcount",
5204 .read_u64 = current_css_set_refcount_read,
5208 .name = "current_css_set_cg_links",
5209 .read_seq_string = current_css_set_cg_links_read,
5213 .name = "cgroup_css_links",
5214 .read_seq_string = cgroup_css_links_read,
5218 .name = "releasable",
5219 .read_u64 = releasable_read,
5223 static int debug_populate(struct cgroup_subsys *ss, struct cgroup *cont)
5225 return cgroup_add_files(cont, ss, debug_files,
5226 ARRAY_SIZE(debug_files));
5229 struct cgroup_subsys debug_subsys = {
5231 .create = debug_create,
5232 .destroy = debug_destroy,
5233 .populate = debug_populate,
5234 .subsys_id = debug_subsys_id,
5236 #endif /* CONFIG_CGROUP_DEBUG */