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 */
63 #include <linux/kthread.h>
65 #include <linux/atomic.h>
67 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
68 #define CSS_DEACT_BIAS INT_MIN
71 * cgroup_mutex is the master lock. Any modification to cgroup or its
72 * hierarchy must be performed while holding it.
74 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
75 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
76 * release_agent_path and so on. Modifying requires both cgroup_mutex and
77 * cgroup_root_mutex. Readers can acquire either of the two. This is to
78 * break the following locking order cycle.
80 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
81 * B. namespace_sem -> cgroup_mutex
83 * B happens only through cgroup_show_options() and using cgroup_root_mutex
86 static DEFINE_MUTEX(cgroup_mutex);
87 static DEFINE_MUTEX(cgroup_root_mutex);
90 * Generate an array of cgroup subsystem pointers. At boot time, this is
91 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
92 * registered after that. The mutable section of this array is protected by
95 #define SUBSYS(_x) &_x ## _subsys,
96 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
97 #include <linux/cgroup_subsys.h>
100 #define MAX_CGROUP_ROOT_NAMELEN 64
103 * A cgroupfs_root represents the root of a cgroup hierarchy,
104 * and may be associated with a superblock to form an active
107 struct cgroupfs_root {
108 struct super_block *sb;
111 * The bitmask of subsystems intended to be attached to this
114 unsigned long subsys_bits;
116 /* Unique id for this hierarchy. */
119 /* The bitmask of subsystems currently attached to this hierarchy */
120 unsigned long actual_subsys_bits;
122 /* A list running through the attached subsystems */
123 struct list_head subsys_list;
125 /* The root cgroup for this hierarchy */
126 struct cgroup top_cgroup;
128 /* Tracks how many cgroups are currently defined in hierarchy.*/
129 int number_of_cgroups;
131 /* A list running through the active hierarchies */
132 struct list_head root_list;
134 /* All cgroups on this root, cgroup_mutex protected */
135 struct list_head allcg_list;
137 /* Hierarchy-specific flags */
140 /* The path to use for release notifications. */
141 char release_agent_path[PATH_MAX];
143 /* The name for this hierarchy - may be empty */
144 char name[MAX_CGROUP_ROOT_NAMELEN];
148 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
149 * subsystems that are otherwise unattached - it never has more than a
150 * single cgroup, and all tasks are part of that cgroup.
152 static struct cgroupfs_root rootnode;
155 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
158 struct list_head node;
159 struct dentry *dentry;
164 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
165 * cgroup_subsys->use_id != 0.
167 #define CSS_ID_MAX (65535)
170 * The css to which this ID points. This pointer is set to valid value
171 * after cgroup is populated. If cgroup is removed, this will be NULL.
172 * This pointer is expected to be RCU-safe because destroy()
173 * is called after synchronize_rcu(). But for safe use, css_is_removed()
174 * css_tryget() should be used for avoiding race.
176 struct cgroup_subsys_state __rcu *css;
182 * Depth in hierarchy which this ID belongs to.
184 unsigned short depth;
186 * ID is freed by RCU. (and lookup routine is RCU safe.)
188 struct rcu_head rcu_head;
190 * Hierarchy of CSS ID belongs to.
192 unsigned short stack[0]; /* Array of Length (depth+1) */
196 * cgroup_event represents events which userspace want to receive.
198 struct cgroup_event {
200 * Cgroup which the event belongs to.
204 * Control file which the event associated.
208 * eventfd to signal userspace about the event.
210 struct eventfd_ctx *eventfd;
212 * Each of these stored in a list by the cgroup.
214 struct list_head list;
216 * All fields below needed to unregister event when
217 * userspace closes eventfd.
220 wait_queue_head_t *wqh;
222 struct work_struct remove;
225 /* The list of hierarchy roots */
227 static LIST_HEAD(roots);
228 static int root_count;
230 static DEFINE_IDA(hierarchy_ida);
231 static int next_hierarchy_id;
232 static DEFINE_SPINLOCK(hierarchy_id_lock);
234 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
235 #define dummytop (&rootnode.top_cgroup)
237 /* This flag indicates whether tasks in the fork and exit paths should
238 * check for fork/exit handlers to call. This avoids us having to do
239 * extra work in the fork/exit path if none of the subsystems need to
242 static int need_forkexit_callback __read_mostly;
244 #ifdef CONFIG_PROVE_LOCKING
245 int cgroup_lock_is_held(void)
247 return lockdep_is_held(&cgroup_mutex);
249 #else /* #ifdef CONFIG_PROVE_LOCKING */
250 int cgroup_lock_is_held(void)
252 return mutex_is_locked(&cgroup_mutex);
254 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
256 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
258 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
259 static int css_refcnt(struct cgroup_subsys_state *css)
261 int v = atomic_read(&css->refcnt);
263 return v >= 0 ? v : v - CSS_DEACT_BIAS;
266 /* convenient tests for these bits */
267 inline int cgroup_is_removed(const struct cgroup *cgrp)
269 return test_bit(CGRP_REMOVED, &cgrp->flags);
272 /* bits in struct cgroupfs_root flags field */
274 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
277 static int cgroup_is_releasable(const struct cgroup *cgrp)
280 (1 << CGRP_RELEASABLE) |
281 (1 << CGRP_NOTIFY_ON_RELEASE);
282 return (cgrp->flags & bits) == bits;
285 static int notify_on_release(const struct cgroup *cgrp)
287 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
290 static int clone_children(const struct cgroup *cgrp)
292 return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
296 * for_each_subsys() allows you to iterate on each subsystem attached to
297 * an active hierarchy
299 #define for_each_subsys(_root, _ss) \
300 list_for_each_entry(_ss, &_root->subsys_list, sibling)
302 /* for_each_active_root() allows you to iterate across the active hierarchies */
303 #define for_each_active_root(_root) \
304 list_for_each_entry(_root, &roots, root_list)
306 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
308 return dentry->d_fsdata;
311 static inline struct cfent *__d_cfe(struct dentry *dentry)
313 return dentry->d_fsdata;
316 static inline struct cftype *__d_cft(struct dentry *dentry)
318 return __d_cfe(dentry)->type;
321 /* the list of cgroups eligible for automatic release. Protected by
322 * release_list_lock */
323 static LIST_HEAD(release_list);
324 static DEFINE_RAW_SPINLOCK(release_list_lock);
325 static void cgroup_release_agent(struct work_struct *work);
326 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
327 static void check_for_release(struct cgroup *cgrp);
329 /* Link structure for associating css_set objects with cgroups */
330 struct cg_cgroup_link {
332 * List running through cg_cgroup_links associated with a
333 * cgroup, anchored on cgroup->css_sets
335 struct list_head cgrp_link_list;
338 * List running through cg_cgroup_links pointing at a
339 * single css_set object, anchored on css_set->cg_links
341 struct list_head cg_link_list;
345 /* The default css_set - used by init and its children prior to any
346 * hierarchies being mounted. It contains a pointer to the root state
347 * for each subsystem. Also used to anchor the list of css_sets. Not
348 * reference-counted, to improve performance when child cgroups
349 * haven't been created.
352 static struct css_set init_css_set;
353 static struct cg_cgroup_link init_css_set_link;
355 static int cgroup_init_idr(struct cgroup_subsys *ss,
356 struct cgroup_subsys_state *css);
358 /* css_set_lock protects the list of css_set objects, and the
359 * chain of tasks off each css_set. Nests outside task->alloc_lock
360 * due to cgroup_iter_start() */
361 static DEFINE_RWLOCK(css_set_lock);
362 static int css_set_count;
365 * hash table for cgroup groups. This improves the performance to find
366 * an existing css_set. This hash doesn't (currently) take into
367 * account cgroups in empty hierarchies.
369 #define CSS_SET_HASH_BITS 7
370 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
371 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
373 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
377 unsigned long tmp = 0UL;
379 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
380 tmp += (unsigned long)css[i];
381 tmp = (tmp >> 16) ^ tmp;
383 index = hash_long(tmp, CSS_SET_HASH_BITS);
385 return &css_set_table[index];
388 /* We don't maintain the lists running through each css_set to its
389 * task until after the first call to cgroup_iter_start(). This
390 * reduces the fork()/exit() overhead for people who have cgroups
391 * compiled into their kernel but not actually in use */
392 static int use_task_css_set_links __read_mostly;
394 static void __put_css_set(struct css_set *cg, int taskexit)
396 struct cg_cgroup_link *link;
397 struct cg_cgroup_link *saved_link;
399 * Ensure that the refcount doesn't hit zero while any readers
400 * can see it. Similar to atomic_dec_and_lock(), but for an
403 if (atomic_add_unless(&cg->refcount, -1, 1))
405 write_lock(&css_set_lock);
406 if (!atomic_dec_and_test(&cg->refcount)) {
407 write_unlock(&css_set_lock);
411 /* This css_set is dead. unlink it and release cgroup refcounts */
412 hlist_del(&cg->hlist);
415 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
417 struct cgroup *cgrp = link->cgrp;
418 list_del(&link->cg_link_list);
419 list_del(&link->cgrp_link_list);
420 if (atomic_dec_and_test(&cgrp->count) &&
421 notify_on_release(cgrp)) {
423 set_bit(CGRP_RELEASABLE, &cgrp->flags);
424 check_for_release(cgrp);
430 write_unlock(&css_set_lock);
431 kfree_rcu(cg, rcu_head);
435 * refcounted get/put for css_set objects
437 static inline void get_css_set(struct css_set *cg)
439 atomic_inc(&cg->refcount);
442 static inline void put_css_set(struct css_set *cg)
444 __put_css_set(cg, 0);
447 static inline void put_css_set_taskexit(struct css_set *cg)
449 __put_css_set(cg, 1);
453 * compare_css_sets - helper function for find_existing_css_set().
454 * @cg: candidate css_set being tested
455 * @old_cg: existing css_set for a task
456 * @new_cgrp: cgroup that's being entered by the task
457 * @template: desired set of css pointers in css_set (pre-calculated)
459 * Returns true if "cg" matches "old_cg" except for the hierarchy
460 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
462 static bool compare_css_sets(struct css_set *cg,
463 struct css_set *old_cg,
464 struct cgroup *new_cgrp,
465 struct cgroup_subsys_state *template[])
467 struct list_head *l1, *l2;
469 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
470 /* Not all subsystems matched */
475 * Compare cgroup pointers in order to distinguish between
476 * different cgroups in heirarchies with no subsystems. We
477 * could get by with just this check alone (and skip the
478 * memcmp above) but on most setups the memcmp check will
479 * avoid the need for this more expensive check on almost all
484 l2 = &old_cg->cg_links;
486 struct cg_cgroup_link *cgl1, *cgl2;
487 struct cgroup *cg1, *cg2;
491 /* See if we reached the end - both lists are equal length. */
492 if (l1 == &cg->cg_links) {
493 BUG_ON(l2 != &old_cg->cg_links);
496 BUG_ON(l2 == &old_cg->cg_links);
498 /* Locate the cgroups associated with these links. */
499 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
500 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
503 /* Hierarchies should be linked in the same order. */
504 BUG_ON(cg1->root != cg2->root);
507 * If this hierarchy is the hierarchy of the cgroup
508 * that's changing, then we need to check that this
509 * css_set points to the new cgroup; if it's any other
510 * hierarchy, then this css_set should point to the
511 * same cgroup as the old css_set.
513 if (cg1->root == new_cgrp->root) {
525 * find_existing_css_set() is a helper for
526 * find_css_set(), and checks to see whether an existing
527 * css_set is suitable.
529 * oldcg: the cgroup group that we're using before the cgroup
532 * cgrp: the cgroup that we're moving into
534 * template: location in which to build the desired set of subsystem
535 * state objects for the new cgroup group
537 static struct css_set *find_existing_css_set(
538 struct css_set *oldcg,
540 struct cgroup_subsys_state *template[])
543 struct cgroupfs_root *root = cgrp->root;
544 struct hlist_head *hhead;
545 struct hlist_node *node;
549 * Build the set of subsystem state objects that we want to see in the
550 * new css_set. while subsystems can change globally, the entries here
551 * won't change, so no need for locking.
553 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
554 if (root->subsys_bits & (1UL << i)) {
555 /* Subsystem is in this hierarchy. So we want
556 * the subsystem state from the new
558 template[i] = cgrp->subsys[i];
560 /* Subsystem is not in this hierarchy, so we
561 * don't want to change the subsystem state */
562 template[i] = oldcg->subsys[i];
566 hhead = css_set_hash(template);
567 hlist_for_each_entry(cg, node, hhead, hlist) {
568 if (!compare_css_sets(cg, oldcg, cgrp, template))
571 /* This css_set matches what we need */
575 /* No existing cgroup group matched */
579 static void free_cg_links(struct list_head *tmp)
581 struct cg_cgroup_link *link;
582 struct cg_cgroup_link *saved_link;
584 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
585 list_del(&link->cgrp_link_list);
591 * allocate_cg_links() allocates "count" cg_cgroup_link structures
592 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
593 * success or a negative error
595 static int allocate_cg_links(int count, struct list_head *tmp)
597 struct cg_cgroup_link *link;
600 for (i = 0; i < count; i++) {
601 link = kmalloc(sizeof(*link), GFP_KERNEL);
606 list_add(&link->cgrp_link_list, tmp);
612 * link_css_set - a helper function to link a css_set to a cgroup
613 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
614 * @cg: the css_set to be linked
615 * @cgrp: the destination cgroup
617 static void link_css_set(struct list_head *tmp_cg_links,
618 struct css_set *cg, struct cgroup *cgrp)
620 struct cg_cgroup_link *link;
622 BUG_ON(list_empty(tmp_cg_links));
623 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
627 atomic_inc(&cgrp->count);
628 list_move(&link->cgrp_link_list, &cgrp->css_sets);
630 * Always add links to the tail of the list so that the list
631 * is sorted by order of hierarchy creation
633 list_add_tail(&link->cg_link_list, &cg->cg_links);
637 * find_css_set() takes an existing cgroup group and a
638 * cgroup object, and returns a css_set object that's
639 * equivalent to the old group, but with the given cgroup
640 * substituted into the appropriate hierarchy. Must be called with
643 static struct css_set *find_css_set(
644 struct css_set *oldcg, struct cgroup *cgrp)
647 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
649 struct list_head tmp_cg_links;
651 struct hlist_head *hhead;
652 struct cg_cgroup_link *link;
654 /* First see if we already have a cgroup group that matches
656 read_lock(&css_set_lock);
657 res = find_existing_css_set(oldcg, cgrp, template);
660 read_unlock(&css_set_lock);
665 res = kmalloc(sizeof(*res), GFP_KERNEL);
669 /* Allocate all the cg_cgroup_link objects that we'll need */
670 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
675 atomic_set(&res->refcount, 1);
676 INIT_LIST_HEAD(&res->cg_links);
677 INIT_LIST_HEAD(&res->tasks);
678 INIT_HLIST_NODE(&res->hlist);
680 /* Copy the set of subsystem state objects generated in
681 * find_existing_css_set() */
682 memcpy(res->subsys, template, sizeof(res->subsys));
684 write_lock(&css_set_lock);
685 /* Add reference counts and links from the new css_set. */
686 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
687 struct cgroup *c = link->cgrp;
688 if (c->root == cgrp->root)
690 link_css_set(&tmp_cg_links, res, c);
693 BUG_ON(!list_empty(&tmp_cg_links));
697 /* Add this cgroup group to the hash table */
698 hhead = css_set_hash(res->subsys);
699 hlist_add_head(&res->hlist, hhead);
701 write_unlock(&css_set_lock);
707 * Return the cgroup for "task" from the given hierarchy. Must be
708 * called with cgroup_mutex held.
710 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
711 struct cgroupfs_root *root)
714 struct cgroup *res = NULL;
716 BUG_ON(!mutex_is_locked(&cgroup_mutex));
717 read_lock(&css_set_lock);
719 * No need to lock the task - since we hold cgroup_mutex the
720 * task can't change groups, so the only thing that can happen
721 * is that it exits and its css is set back to init_css_set.
724 if (css == &init_css_set) {
725 res = &root->top_cgroup;
727 struct cg_cgroup_link *link;
728 list_for_each_entry(link, &css->cg_links, cg_link_list) {
729 struct cgroup *c = link->cgrp;
730 if (c->root == root) {
736 read_unlock(&css_set_lock);
742 * There is one global cgroup mutex. We also require taking
743 * task_lock() when dereferencing a task's cgroup subsys pointers.
744 * See "The task_lock() exception", at the end of this comment.
746 * A task must hold cgroup_mutex to modify cgroups.
748 * Any task can increment and decrement the count field without lock.
749 * So in general, code holding cgroup_mutex can't rely on the count
750 * field not changing. However, if the count goes to zero, then only
751 * cgroup_attach_task() can increment it again. Because a count of zero
752 * means that no tasks are currently attached, therefore there is no
753 * way a task attached to that cgroup can fork (the other way to
754 * increment the count). So code holding cgroup_mutex can safely
755 * assume that if the count is zero, it will stay zero. Similarly, if
756 * a task holds cgroup_mutex on a cgroup with zero count, it
757 * knows that the cgroup won't be removed, as cgroup_rmdir()
760 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
761 * (usually) take cgroup_mutex. These are the two most performance
762 * critical pieces of code here. The exception occurs on cgroup_exit(),
763 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
764 * is taken, and if the cgroup count is zero, a usermode call made
765 * to the release agent with the name of the cgroup (path relative to
766 * the root of cgroup file system) as the argument.
768 * A cgroup can only be deleted if both its 'count' of using tasks
769 * is zero, and its list of 'children' cgroups is empty. Since all
770 * tasks in the system use _some_ cgroup, and since there is always at
771 * least one task in the system (init, pid == 1), therefore, top_cgroup
772 * always has either children cgroups and/or using tasks. So we don't
773 * need a special hack to ensure that top_cgroup cannot be deleted.
775 * The task_lock() exception
777 * The need for this exception arises from the action of
778 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
779 * another. It does so using cgroup_mutex, however there are
780 * several performance critical places that need to reference
781 * task->cgroup without the expense of grabbing a system global
782 * mutex. Therefore except as noted below, when dereferencing or, as
783 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
784 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
785 * the task_struct routinely used for such matters.
787 * P.S. One more locking exception. RCU is used to guard the
788 * update of a tasks cgroup pointer by cgroup_attach_task()
792 * cgroup_lock - lock out any changes to cgroup structures
795 void cgroup_lock(void)
797 mutex_lock(&cgroup_mutex);
799 EXPORT_SYMBOL_GPL(cgroup_lock);
802 * cgroup_unlock - release lock on cgroup changes
804 * Undo the lock taken in a previous cgroup_lock() call.
806 void cgroup_unlock(void)
808 mutex_unlock(&cgroup_mutex);
810 EXPORT_SYMBOL_GPL(cgroup_unlock);
813 * A couple of forward declarations required, due to cyclic reference loop:
814 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
815 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
819 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
820 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, struct nameidata *);
821 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
822 static int cgroup_populate_dir(struct cgroup *cgrp);
823 static const struct inode_operations cgroup_dir_inode_operations;
824 static const struct file_operations proc_cgroupstats_operations;
826 static struct backing_dev_info cgroup_backing_dev_info = {
828 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
831 static int alloc_css_id(struct cgroup_subsys *ss,
832 struct cgroup *parent, struct cgroup *child);
834 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
836 struct inode *inode = new_inode(sb);
839 inode->i_ino = get_next_ino();
840 inode->i_mode = mode;
841 inode->i_uid = current_fsuid();
842 inode->i_gid = current_fsgid();
843 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
844 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
850 * Call subsys's pre_destroy handler.
851 * This is called before css refcnt check.
853 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
855 struct cgroup_subsys *ss;
858 for_each_subsys(cgrp->root, ss) {
859 if (!ss->pre_destroy)
862 ret = ss->pre_destroy(cgrp);
864 /* ->pre_destroy() failure is being deprecated */
865 WARN_ON_ONCE(!ss->__DEPRECATED_clear_css_refs);
873 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
875 /* is dentry a directory ? if so, kfree() associated cgroup */
876 if (S_ISDIR(inode->i_mode)) {
877 struct cgroup *cgrp = dentry->d_fsdata;
878 struct cgroup_subsys *ss;
879 BUG_ON(!(cgroup_is_removed(cgrp)));
880 /* It's possible for external users to be holding css
881 * reference counts on a cgroup; css_put() needs to
882 * be able to access the cgroup after decrementing
883 * the reference count in order to know if it needs to
884 * queue the cgroup to be handled by the release
888 mutex_lock(&cgroup_mutex);
890 * Release the subsystem state objects.
892 for_each_subsys(cgrp->root, ss)
895 cgrp->root->number_of_cgroups--;
896 mutex_unlock(&cgroup_mutex);
899 * We want to drop the active superblock reference from the
900 * cgroup creation after all the dentry refs are gone -
901 * kill_sb gets mighty unhappy otherwise. Mark
902 * dentry->d_fsdata with cgroup_diput() to tell
903 * cgroup_d_release() to call deactivate_super().
905 dentry->d_fsdata = cgroup_diput;
908 * if we're getting rid of the cgroup, refcount should ensure
909 * that there are no pidlists left.
911 BUG_ON(!list_empty(&cgrp->pidlists));
913 kfree_rcu(cgrp, rcu_head);
915 struct cfent *cfe = __d_cfe(dentry);
916 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
918 WARN_ONCE(!list_empty(&cfe->node) &&
919 cgrp != &cgrp->root->top_cgroup,
920 "cfe still linked for %s\n", cfe->type->name);
926 static int cgroup_delete(const struct dentry *d)
931 static void cgroup_d_release(struct dentry *dentry)
933 /* did cgroup_diput() tell me to deactivate super? */
934 if (dentry->d_fsdata == cgroup_diput)
935 deactivate_super(dentry->d_sb);
938 static void remove_dir(struct dentry *d)
940 struct dentry *parent = dget(d->d_parent);
943 simple_rmdir(parent->d_inode, d);
947 static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
951 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
952 lockdep_assert_held(&cgroup_mutex);
954 list_for_each_entry(cfe, &cgrp->files, node) {
955 struct dentry *d = cfe->dentry;
957 if (cft && cfe->type != cft)
962 simple_unlink(d->d_inode, d);
963 list_del_init(&cfe->node);
971 static void cgroup_clear_directory(struct dentry *dir)
973 struct cgroup *cgrp = __d_cgrp(dir);
975 while (!list_empty(&cgrp->files))
976 cgroup_rm_file(cgrp, NULL);
980 * NOTE : the dentry must have been dget()'ed
982 static void cgroup_d_remove_dir(struct dentry *dentry)
984 struct dentry *parent;
986 cgroup_clear_directory(dentry);
988 parent = dentry->d_parent;
989 spin_lock(&parent->d_lock);
990 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
991 list_del_init(&dentry->d_u.d_child);
992 spin_unlock(&dentry->d_lock);
993 spin_unlock(&parent->d_lock);
998 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
999 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
1000 * reference to css->refcnt. In general, this refcnt is expected to goes down
1003 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
1005 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
1007 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
1009 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
1010 wake_up_all(&cgroup_rmdir_waitq);
1013 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
1018 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
1020 cgroup_wakeup_rmdir_waiter(css->cgroup);
1025 * Call with cgroup_mutex held. Drops reference counts on modules, including
1026 * any duplicate ones that parse_cgroupfs_options took. If this function
1027 * returns an error, no reference counts are touched.
1029 static int rebind_subsystems(struct cgroupfs_root *root,
1030 unsigned long final_bits)
1032 unsigned long added_bits, removed_bits;
1033 struct cgroup *cgrp = &root->top_cgroup;
1036 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1037 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1039 removed_bits = root->actual_subsys_bits & ~final_bits;
1040 added_bits = final_bits & ~root->actual_subsys_bits;
1041 /* Check that any added subsystems are currently free */
1042 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1043 unsigned long bit = 1UL << i;
1044 struct cgroup_subsys *ss = subsys[i];
1045 if (!(bit & added_bits))
1048 * Nobody should tell us to do a subsys that doesn't exist:
1049 * parse_cgroupfs_options should catch that case and refcounts
1050 * ensure that subsystems won't disappear once selected.
1053 if (ss->root != &rootnode) {
1054 /* Subsystem isn't free */
1059 /* Currently we don't handle adding/removing subsystems when
1060 * any child cgroups exist. This is theoretically supportable
1061 * but involves complex error handling, so it's being left until
1063 if (root->number_of_cgroups > 1)
1066 /* Process each subsystem */
1067 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1068 struct cgroup_subsys *ss = subsys[i];
1069 unsigned long bit = 1UL << i;
1070 if (bit & added_bits) {
1071 /* We're binding this subsystem to this hierarchy */
1073 BUG_ON(cgrp->subsys[i]);
1074 BUG_ON(!dummytop->subsys[i]);
1075 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1076 mutex_lock(&ss->hierarchy_mutex);
1077 cgrp->subsys[i] = dummytop->subsys[i];
1078 cgrp->subsys[i]->cgroup = cgrp;
1079 list_move(&ss->sibling, &root->subsys_list);
1083 mutex_unlock(&ss->hierarchy_mutex);
1084 /* refcount was already taken, and we're keeping it */
1085 } else if (bit & removed_bits) {
1086 /* We're removing this subsystem */
1088 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1089 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1090 mutex_lock(&ss->hierarchy_mutex);
1093 dummytop->subsys[i]->cgroup = dummytop;
1094 cgrp->subsys[i] = NULL;
1095 subsys[i]->root = &rootnode;
1096 list_move(&ss->sibling, &rootnode.subsys_list);
1097 mutex_unlock(&ss->hierarchy_mutex);
1098 /* subsystem is now free - drop reference on module */
1099 module_put(ss->module);
1100 } else if (bit & final_bits) {
1101 /* Subsystem state should already exist */
1103 BUG_ON(!cgrp->subsys[i]);
1105 * a refcount was taken, but we already had one, so
1106 * drop the extra reference.
1108 module_put(ss->module);
1109 #ifdef CONFIG_MODULE_UNLOAD
1110 BUG_ON(ss->module && !module_refcount(ss->module));
1113 /* Subsystem state shouldn't exist */
1114 BUG_ON(cgrp->subsys[i]);
1117 root->subsys_bits = root->actual_subsys_bits = final_bits;
1123 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1125 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1126 struct cgroup_subsys *ss;
1128 mutex_lock(&cgroup_root_mutex);
1129 for_each_subsys(root, ss)
1130 seq_printf(seq, ",%s", ss->name);
1131 if (test_bit(ROOT_NOPREFIX, &root->flags))
1132 seq_puts(seq, ",noprefix");
1133 if (strlen(root->release_agent_path))
1134 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1135 if (clone_children(&root->top_cgroup))
1136 seq_puts(seq, ",clone_children");
1137 if (strlen(root->name))
1138 seq_printf(seq, ",name=%s", root->name);
1139 mutex_unlock(&cgroup_root_mutex);
1143 struct cgroup_sb_opts {
1144 unsigned long subsys_bits;
1145 unsigned long flags;
1146 char *release_agent;
1147 bool clone_children;
1149 /* User explicitly requested empty subsystem */
1152 struct cgroupfs_root *new_root;
1157 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1158 * with cgroup_mutex held to protect the subsys[] array. This function takes
1159 * refcounts on subsystems to be used, unless it returns error, in which case
1160 * no refcounts are taken.
1162 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1164 char *token, *o = data;
1165 bool all_ss = false, one_ss = false;
1166 unsigned long mask = (unsigned long)-1;
1168 bool module_pin_failed = false;
1170 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1172 #ifdef CONFIG_CPUSETS
1173 mask = ~(1UL << cpuset_subsys_id);
1176 memset(opts, 0, sizeof(*opts));
1178 while ((token = strsep(&o, ",")) != NULL) {
1181 if (!strcmp(token, "none")) {
1182 /* Explicitly have no subsystems */
1186 if (!strcmp(token, "all")) {
1187 /* Mutually exclusive option 'all' + subsystem name */
1193 if (!strcmp(token, "noprefix")) {
1194 set_bit(ROOT_NOPREFIX, &opts->flags);
1197 if (!strcmp(token, "clone_children")) {
1198 opts->clone_children = true;
1201 if (!strncmp(token, "release_agent=", 14)) {
1202 /* Specifying two release agents is forbidden */
1203 if (opts->release_agent)
1205 opts->release_agent =
1206 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1207 if (!opts->release_agent)
1211 if (!strncmp(token, "name=", 5)) {
1212 const char *name = token + 5;
1213 /* Can't specify an empty name */
1216 /* Must match [\w.-]+ */
1217 for (i = 0; i < strlen(name); i++) {
1221 if ((c == '.') || (c == '-') || (c == '_'))
1225 /* Specifying two names is forbidden */
1228 opts->name = kstrndup(name,
1229 MAX_CGROUP_ROOT_NAMELEN - 1,
1237 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1238 struct cgroup_subsys *ss = subsys[i];
1241 if (strcmp(token, ss->name))
1246 /* Mutually exclusive option 'all' + subsystem name */
1249 set_bit(i, &opts->subsys_bits);
1254 if (i == CGROUP_SUBSYS_COUNT)
1259 * If the 'all' option was specified select all the subsystems,
1260 * otherwise if 'none', 'name=' and a subsystem name options
1261 * were not specified, let's default to 'all'
1263 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1264 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1265 struct cgroup_subsys *ss = subsys[i];
1270 set_bit(i, &opts->subsys_bits);
1274 /* Consistency checks */
1277 * Option noprefix was introduced just for backward compatibility
1278 * with the old cpuset, so we allow noprefix only if mounting just
1279 * the cpuset subsystem.
1281 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1282 (opts->subsys_bits & mask))
1286 /* Can't specify "none" and some subsystems */
1287 if (opts->subsys_bits && opts->none)
1291 * We either have to specify by name or by subsystems. (So all
1292 * empty hierarchies must have a name).
1294 if (!opts->subsys_bits && !opts->name)
1298 * Grab references on all the modules we'll need, so the subsystems
1299 * don't dance around before rebind_subsystems attaches them. This may
1300 * take duplicate reference counts on a subsystem that's already used,
1301 * but rebind_subsystems handles this case.
1303 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1304 unsigned long bit = 1UL << i;
1306 if (!(bit & opts->subsys_bits))
1308 if (!try_module_get(subsys[i]->module)) {
1309 module_pin_failed = true;
1313 if (module_pin_failed) {
1315 * oops, one of the modules was going away. this means that we
1316 * raced with a module_delete call, and to the user this is
1317 * essentially a "subsystem doesn't exist" case.
1319 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1320 /* drop refcounts only on the ones we took */
1321 unsigned long bit = 1UL << i;
1323 if (!(bit & opts->subsys_bits))
1325 module_put(subsys[i]->module);
1333 static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1336 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1337 unsigned long bit = 1UL << i;
1339 if (!(bit & subsys_bits))
1341 module_put(subsys[i]->module);
1345 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1348 struct cgroupfs_root *root = sb->s_fs_info;
1349 struct cgroup *cgrp = &root->top_cgroup;
1350 struct cgroup_sb_opts opts;
1352 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1353 mutex_lock(&cgroup_mutex);
1354 mutex_lock(&cgroup_root_mutex);
1356 /* See what subsystems are wanted */
1357 ret = parse_cgroupfs_options(data, &opts);
1361 /* See feature-removal-schedule.txt */
1362 if (opts.subsys_bits != root->actual_subsys_bits || opts.release_agent)
1363 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1364 task_tgid_nr(current), current->comm);
1366 /* Don't allow flags or name to change at remount */
1367 if (opts.flags != root->flags ||
1368 (opts.name && strcmp(opts.name, root->name))) {
1370 drop_parsed_module_refcounts(opts.subsys_bits);
1374 ret = rebind_subsystems(root, opts.subsys_bits);
1376 drop_parsed_module_refcounts(opts.subsys_bits);
1380 /* clear out any existing files and repopulate subsystem files */
1381 cgroup_clear_directory(cgrp->dentry);
1382 cgroup_populate_dir(cgrp);
1384 if (opts.release_agent)
1385 strcpy(root->release_agent_path, opts.release_agent);
1387 kfree(opts.release_agent);
1389 mutex_unlock(&cgroup_root_mutex);
1390 mutex_unlock(&cgroup_mutex);
1391 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1395 static const struct super_operations cgroup_ops = {
1396 .statfs = simple_statfs,
1397 .drop_inode = generic_delete_inode,
1398 .show_options = cgroup_show_options,
1399 .remount_fs = cgroup_remount,
1402 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1404 INIT_LIST_HEAD(&cgrp->sibling);
1405 INIT_LIST_HEAD(&cgrp->children);
1406 INIT_LIST_HEAD(&cgrp->files);
1407 INIT_LIST_HEAD(&cgrp->css_sets);
1408 INIT_LIST_HEAD(&cgrp->release_list);
1409 INIT_LIST_HEAD(&cgrp->pidlists);
1410 mutex_init(&cgrp->pidlist_mutex);
1411 INIT_LIST_HEAD(&cgrp->event_list);
1412 spin_lock_init(&cgrp->event_list_lock);
1415 static void init_cgroup_root(struct cgroupfs_root *root)
1417 struct cgroup *cgrp = &root->top_cgroup;
1419 INIT_LIST_HEAD(&root->subsys_list);
1420 INIT_LIST_HEAD(&root->root_list);
1421 INIT_LIST_HEAD(&root->allcg_list);
1422 root->number_of_cgroups = 1;
1424 cgrp->top_cgroup = cgrp;
1425 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1426 init_cgroup_housekeeping(cgrp);
1429 static bool init_root_id(struct cgroupfs_root *root)
1434 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1436 spin_lock(&hierarchy_id_lock);
1437 /* Try to allocate the next unused ID */
1438 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1439 &root->hierarchy_id);
1441 /* Try again starting from 0 */
1442 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1444 next_hierarchy_id = root->hierarchy_id + 1;
1445 } else if (ret != -EAGAIN) {
1446 /* Can only get here if the 31-bit IDR is full ... */
1449 spin_unlock(&hierarchy_id_lock);
1454 static int cgroup_test_super(struct super_block *sb, void *data)
1456 struct cgroup_sb_opts *opts = data;
1457 struct cgroupfs_root *root = sb->s_fs_info;
1459 /* If we asked for a name then it must match */
1460 if (opts->name && strcmp(opts->name, root->name))
1464 * If we asked for subsystems (or explicitly for no
1465 * subsystems) then they must match
1467 if ((opts->subsys_bits || opts->none)
1468 && (opts->subsys_bits != root->subsys_bits))
1474 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1476 struct cgroupfs_root *root;
1478 if (!opts->subsys_bits && !opts->none)
1481 root = kzalloc(sizeof(*root), GFP_KERNEL);
1483 return ERR_PTR(-ENOMEM);
1485 if (!init_root_id(root)) {
1487 return ERR_PTR(-ENOMEM);
1489 init_cgroup_root(root);
1491 root->subsys_bits = opts->subsys_bits;
1492 root->flags = opts->flags;
1493 if (opts->release_agent)
1494 strcpy(root->release_agent_path, opts->release_agent);
1496 strcpy(root->name, opts->name);
1497 if (opts->clone_children)
1498 set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1502 static void cgroup_drop_root(struct cgroupfs_root *root)
1507 BUG_ON(!root->hierarchy_id);
1508 spin_lock(&hierarchy_id_lock);
1509 ida_remove(&hierarchy_ida, root->hierarchy_id);
1510 spin_unlock(&hierarchy_id_lock);
1514 static int cgroup_set_super(struct super_block *sb, void *data)
1517 struct cgroup_sb_opts *opts = data;
1519 /* If we don't have a new root, we can't set up a new sb */
1520 if (!opts->new_root)
1523 BUG_ON(!opts->subsys_bits && !opts->none);
1525 ret = set_anon_super(sb, NULL);
1529 sb->s_fs_info = opts->new_root;
1530 opts->new_root->sb = sb;
1532 sb->s_blocksize = PAGE_CACHE_SIZE;
1533 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1534 sb->s_magic = CGROUP_SUPER_MAGIC;
1535 sb->s_op = &cgroup_ops;
1540 static int cgroup_get_rootdir(struct super_block *sb)
1542 static const struct dentry_operations cgroup_dops = {
1543 .d_iput = cgroup_diput,
1544 .d_delete = cgroup_delete,
1545 .d_release = cgroup_d_release,
1548 struct inode *inode =
1549 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1554 inode->i_fop = &simple_dir_operations;
1555 inode->i_op = &cgroup_dir_inode_operations;
1556 /* directories start off with i_nlink == 2 (for "." entry) */
1558 sb->s_root = d_make_root(inode);
1561 /* for everything else we want ->d_op set */
1562 sb->s_d_op = &cgroup_dops;
1566 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1567 int flags, const char *unused_dev_name,
1570 struct cgroup_sb_opts opts;
1571 struct cgroupfs_root *root;
1573 struct super_block *sb;
1574 struct cgroupfs_root *new_root;
1575 struct inode *inode;
1577 /* First find the desired set of subsystems */
1578 mutex_lock(&cgroup_mutex);
1579 ret = parse_cgroupfs_options(data, &opts);
1580 mutex_unlock(&cgroup_mutex);
1585 * Allocate a new cgroup root. We may not need it if we're
1586 * reusing an existing hierarchy.
1588 new_root = cgroup_root_from_opts(&opts);
1589 if (IS_ERR(new_root)) {
1590 ret = PTR_ERR(new_root);
1593 opts.new_root = new_root;
1595 /* Locate an existing or new sb for this hierarchy */
1596 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1599 cgroup_drop_root(opts.new_root);
1603 root = sb->s_fs_info;
1605 if (root == opts.new_root) {
1606 /* We used the new root structure, so this is a new hierarchy */
1607 struct list_head tmp_cg_links;
1608 struct cgroup *root_cgrp = &root->top_cgroup;
1609 struct cgroupfs_root *existing_root;
1610 const struct cred *cred;
1613 BUG_ON(sb->s_root != NULL);
1615 ret = cgroup_get_rootdir(sb);
1617 goto drop_new_super;
1618 inode = sb->s_root->d_inode;
1620 mutex_lock(&inode->i_mutex);
1621 mutex_lock(&cgroup_mutex);
1622 mutex_lock(&cgroup_root_mutex);
1624 /* Check for name clashes with existing mounts */
1626 if (strlen(root->name))
1627 for_each_active_root(existing_root)
1628 if (!strcmp(existing_root->name, root->name))
1632 * We're accessing css_set_count without locking
1633 * css_set_lock here, but that's OK - it can only be
1634 * increased by someone holding cgroup_lock, and
1635 * that's us. The worst that can happen is that we
1636 * have some link structures left over
1638 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1642 ret = rebind_subsystems(root, root->subsys_bits);
1643 if (ret == -EBUSY) {
1644 free_cg_links(&tmp_cg_links);
1648 * There must be no failure case after here, since rebinding
1649 * takes care of subsystems' refcounts, which are explicitly
1650 * dropped in the failure exit path.
1653 /* EBUSY should be the only error here */
1656 list_add(&root->root_list, &roots);
1659 sb->s_root->d_fsdata = root_cgrp;
1660 root->top_cgroup.dentry = sb->s_root;
1662 /* Link the top cgroup in this hierarchy into all
1663 * the css_set objects */
1664 write_lock(&css_set_lock);
1665 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1666 struct hlist_head *hhead = &css_set_table[i];
1667 struct hlist_node *node;
1670 hlist_for_each_entry(cg, node, hhead, hlist)
1671 link_css_set(&tmp_cg_links, cg, root_cgrp);
1673 write_unlock(&css_set_lock);
1675 free_cg_links(&tmp_cg_links);
1677 BUG_ON(!list_empty(&root_cgrp->sibling));
1678 BUG_ON(!list_empty(&root_cgrp->children));
1679 BUG_ON(root->number_of_cgroups != 1);
1681 cred = override_creds(&init_cred);
1682 cgroup_populate_dir(root_cgrp);
1684 mutex_unlock(&cgroup_root_mutex);
1685 mutex_unlock(&cgroup_mutex);
1686 mutex_unlock(&inode->i_mutex);
1689 * We re-used an existing hierarchy - the new root (if
1690 * any) is not needed
1692 cgroup_drop_root(opts.new_root);
1693 /* no subsys rebinding, so refcounts don't change */
1694 drop_parsed_module_refcounts(opts.subsys_bits);
1697 kfree(opts.release_agent);
1699 return dget(sb->s_root);
1702 mutex_unlock(&cgroup_root_mutex);
1703 mutex_unlock(&cgroup_mutex);
1704 mutex_unlock(&inode->i_mutex);
1706 deactivate_locked_super(sb);
1708 drop_parsed_module_refcounts(opts.subsys_bits);
1710 kfree(opts.release_agent);
1712 return ERR_PTR(ret);
1715 static void cgroup_kill_sb(struct super_block *sb) {
1716 struct cgroupfs_root *root = sb->s_fs_info;
1717 struct cgroup *cgrp = &root->top_cgroup;
1719 struct cg_cgroup_link *link;
1720 struct cg_cgroup_link *saved_link;
1724 BUG_ON(root->number_of_cgroups != 1);
1725 BUG_ON(!list_empty(&cgrp->children));
1726 BUG_ON(!list_empty(&cgrp->sibling));
1728 mutex_lock(&cgroup_mutex);
1729 mutex_lock(&cgroup_root_mutex);
1731 /* Rebind all subsystems back to the default hierarchy */
1732 ret = rebind_subsystems(root, 0);
1733 /* Shouldn't be able to fail ... */
1737 * Release all the links from css_sets to this hierarchy's
1740 write_lock(&css_set_lock);
1742 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1744 list_del(&link->cg_link_list);
1745 list_del(&link->cgrp_link_list);
1748 write_unlock(&css_set_lock);
1750 if (!list_empty(&root->root_list)) {
1751 list_del(&root->root_list);
1755 mutex_unlock(&cgroup_root_mutex);
1756 mutex_unlock(&cgroup_mutex);
1758 kill_litter_super(sb);
1759 cgroup_drop_root(root);
1762 static struct file_system_type cgroup_fs_type = {
1764 .mount = cgroup_mount,
1765 .kill_sb = cgroup_kill_sb,
1768 static struct kobject *cgroup_kobj;
1771 * cgroup_path - generate the path of a cgroup
1772 * @cgrp: the cgroup in question
1773 * @buf: the buffer to write the path into
1774 * @buflen: the length of the buffer
1776 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1777 * reference. Writes path of cgroup into buf. Returns 0 on success,
1780 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1783 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1784 cgroup_lock_is_held());
1786 if (!dentry || cgrp == dummytop) {
1788 * Inactive subsystems have no dentry for their root
1795 start = buf + buflen;
1799 int len = dentry->d_name.len;
1801 if ((start -= len) < buf)
1802 return -ENAMETOOLONG;
1803 memcpy(start, dentry->d_name.name, len);
1804 cgrp = cgrp->parent;
1808 dentry = rcu_dereference_check(cgrp->dentry,
1809 cgroup_lock_is_held());
1813 return -ENAMETOOLONG;
1816 memmove(buf, start, buf + buflen - start);
1819 EXPORT_SYMBOL_GPL(cgroup_path);
1822 * Control Group taskset
1824 struct task_and_cgroup {
1825 struct task_struct *task;
1826 struct cgroup *cgrp;
1830 struct cgroup_taskset {
1831 struct task_and_cgroup single;
1832 struct flex_array *tc_array;
1835 struct cgroup *cur_cgrp;
1839 * cgroup_taskset_first - reset taskset and return the first task
1840 * @tset: taskset of interest
1842 * @tset iteration is initialized and the first task is returned.
1844 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1846 if (tset->tc_array) {
1848 return cgroup_taskset_next(tset);
1850 tset->cur_cgrp = tset->single.cgrp;
1851 return tset->single.task;
1854 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1857 * cgroup_taskset_next - iterate to the next task in taskset
1858 * @tset: taskset of interest
1860 * Return the next task in @tset. Iteration must have been initialized
1861 * with cgroup_taskset_first().
1863 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1865 struct task_and_cgroup *tc;
1867 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1870 tc = flex_array_get(tset->tc_array, tset->idx++);
1871 tset->cur_cgrp = tc->cgrp;
1874 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1877 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1878 * @tset: taskset of interest
1880 * Return the cgroup for the current (last returned) task of @tset. This
1881 * function must be preceded by either cgroup_taskset_first() or
1882 * cgroup_taskset_next().
1884 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1886 return tset->cur_cgrp;
1888 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1891 * cgroup_taskset_size - return the number of tasks in taskset
1892 * @tset: taskset of interest
1894 int cgroup_taskset_size(struct cgroup_taskset *tset)
1896 return tset->tc_array ? tset->tc_array_len : 1;
1898 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1902 * cgroup_task_migrate - move a task from one cgroup to another.
1904 * 'guarantee' is set if the caller promises that a new css_set for the task
1905 * will already exist. If not set, this function might sleep, and can fail with
1906 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1908 static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1909 struct task_struct *tsk, struct css_set *newcg)
1911 struct css_set *oldcg;
1914 * We are synchronized through threadgroup_lock() against PF_EXITING
1915 * setting such that we can't race against cgroup_exit() changing the
1916 * css_set to init_css_set and dropping the old one.
1918 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1919 oldcg = tsk->cgroups;
1922 rcu_assign_pointer(tsk->cgroups, newcg);
1925 /* Update the css_set linked lists if we're using them */
1926 write_lock(&css_set_lock);
1927 if (!list_empty(&tsk->cg_list))
1928 list_move(&tsk->cg_list, &newcg->tasks);
1929 write_unlock(&css_set_lock);
1932 * We just gained a reference on oldcg by taking it from the task. As
1933 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1934 * it here; it will be freed under RCU.
1938 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1942 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1943 * @cgrp: the cgroup the task is attaching to
1944 * @tsk: the task to be attached
1946 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1949 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1952 struct cgroup_subsys *ss, *failed_ss = NULL;
1953 struct cgroup *oldcgrp;
1954 struct cgroupfs_root *root = cgrp->root;
1955 struct cgroup_taskset tset = { };
1956 struct css_set *newcg;
1958 /* @tsk either already exited or can't exit until the end */
1959 if (tsk->flags & PF_EXITING)
1962 /* Nothing to do if the task is already in that cgroup */
1963 oldcgrp = task_cgroup_from_root(tsk, root);
1964 if (cgrp == oldcgrp)
1967 tset.single.task = tsk;
1968 tset.single.cgrp = oldcgrp;
1970 for_each_subsys(root, ss) {
1971 if (ss->can_attach) {
1972 retval = ss->can_attach(cgrp, &tset);
1975 * Remember on which subsystem the can_attach()
1976 * failed, so that we only call cancel_attach()
1977 * against the subsystems whose can_attach()
1978 * succeeded. (See below)
1986 newcg = find_css_set(tsk->cgroups, cgrp);
1992 cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1994 for_each_subsys(root, ss) {
1996 ss->attach(cgrp, &tset);
2002 * wake up rmdir() waiter. the rmdir should fail since the cgroup
2003 * is no longer empty.
2005 cgroup_wakeup_rmdir_waiter(cgrp);
2008 for_each_subsys(root, ss) {
2009 if (ss == failed_ss)
2011 * This subsystem was the one that failed the
2012 * can_attach() check earlier, so we don't need
2013 * to call cancel_attach() against it or any
2014 * remaining subsystems.
2017 if (ss->cancel_attach)
2018 ss->cancel_attach(cgrp, &tset);
2025 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2026 * @from: attach to all cgroups of a given task
2027 * @tsk: the task to be attached
2029 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2031 struct cgroupfs_root *root;
2035 for_each_active_root(root) {
2036 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2038 retval = cgroup_attach_task(from_cg, tsk);
2046 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2049 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2050 * @cgrp: the cgroup to attach to
2051 * @leader: the threadgroup leader task_struct of the group to be attached
2053 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2054 * task_lock of each thread in leader's threadgroup individually in turn.
2056 static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2058 int retval, i, group_size;
2059 struct cgroup_subsys *ss, *failed_ss = NULL;
2060 /* guaranteed to be initialized later, but the compiler needs this */
2061 struct cgroupfs_root *root = cgrp->root;
2062 /* threadgroup list cursor and array */
2063 struct task_struct *tsk;
2064 struct task_and_cgroup *tc;
2065 struct flex_array *group;
2066 struct cgroup_taskset tset = { };
2069 * step 0: in order to do expensive, possibly blocking operations for
2070 * every thread, we cannot iterate the thread group list, since it needs
2071 * rcu or tasklist locked. instead, build an array of all threads in the
2072 * group - group_rwsem prevents new threads from appearing, and if
2073 * threads exit, this will just be an over-estimate.
2075 group_size = get_nr_threads(leader);
2076 /* flex_array supports very large thread-groups better than kmalloc. */
2077 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2080 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2081 retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2083 goto out_free_group_list;
2088 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2089 * already PF_EXITING could be freed from underneath us unless we
2090 * take an rcu_read_lock.
2094 struct task_and_cgroup ent;
2096 /* @tsk either already exited or can't exit until the end */
2097 if (tsk->flags & PF_EXITING)
2100 /* as per above, nr_threads may decrease, but not increase. */
2101 BUG_ON(i >= group_size);
2103 ent.cgrp = task_cgroup_from_root(tsk, root);
2104 /* nothing to do if this task is already in the cgroup */
2105 if (ent.cgrp == cgrp)
2108 * saying GFP_ATOMIC has no effect here because we did prealloc
2109 * earlier, but it's good form to communicate our expectations.
2111 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2112 BUG_ON(retval != 0);
2114 } while_each_thread(leader, tsk);
2116 /* remember the number of threads in the array for later. */
2118 tset.tc_array = group;
2119 tset.tc_array_len = group_size;
2121 /* methods shouldn't be called if no task is actually migrating */
2124 goto out_free_group_list;
2127 * step 1: check that we can legitimately attach to the cgroup.
2129 for_each_subsys(root, ss) {
2130 if (ss->can_attach) {
2131 retval = ss->can_attach(cgrp, &tset);
2134 goto out_cancel_attach;
2140 * step 2: make sure css_sets exist for all threads to be migrated.
2141 * we use find_css_set, which allocates a new one if necessary.
2143 for (i = 0; i < group_size; i++) {
2144 tc = flex_array_get(group, i);
2145 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2148 goto out_put_css_set_refs;
2153 * step 3: now that we're guaranteed success wrt the css_sets,
2154 * proceed to move all tasks to the new cgroup. There are no
2155 * failure cases after here, so this is the commit point.
2157 for (i = 0; i < group_size; i++) {
2158 tc = flex_array_get(group, i);
2159 cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2161 /* nothing is sensitive to fork() after this point. */
2164 * step 4: do subsystem attach callbacks.
2166 for_each_subsys(root, ss) {
2168 ss->attach(cgrp, &tset);
2172 * step 5: success! and cleanup
2175 cgroup_wakeup_rmdir_waiter(cgrp);
2177 out_put_css_set_refs:
2179 for (i = 0; i < group_size; i++) {
2180 tc = flex_array_get(group, i);
2183 put_css_set(tc->cg);
2188 for_each_subsys(root, ss) {
2189 if (ss == failed_ss)
2191 if (ss->cancel_attach)
2192 ss->cancel_attach(cgrp, &tset);
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 lock
2203 * cgroup_mutex and threadgroup; 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))
2217 tsk = find_task_by_vpid(pid);
2221 goto out_unlock_cgroup;
2224 * even if we're attaching all tasks in the thread group, we
2225 * only need to check permissions on one of them.
2227 tcred = __task_cred(tsk);
2228 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2229 !uid_eq(cred->euid, tcred->uid) &&
2230 !uid_eq(cred->euid, tcred->suid)) {
2233 goto out_unlock_cgroup;
2239 tsk = tsk->group_leader;
2242 * Workqueue threads may acquire PF_THREAD_BOUND and become
2243 * trapped in a cpuset, or RT worker may be born in a cgroup
2244 * with no rt_runtime allocated. Just say no.
2246 if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2249 goto out_unlock_cgroup;
2252 get_task_struct(tsk);
2255 threadgroup_lock(tsk);
2257 if (!thread_group_leader(tsk)) {
2259 * a race with de_thread from another thread's exec()
2260 * may strip us of our leadership, if this happens,
2261 * there is no choice but to throw this task away and
2262 * try again; this is
2263 * "double-double-toil-and-trouble-check locking".
2265 threadgroup_unlock(tsk);
2266 put_task_struct(tsk);
2267 goto retry_find_task;
2269 ret = cgroup_attach_proc(cgrp, tsk);
2271 ret = cgroup_attach_task(cgrp, tsk);
2272 threadgroup_unlock(tsk);
2274 put_task_struct(tsk);
2280 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2282 return attach_task_by_pid(cgrp, pid, false);
2285 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2287 return attach_task_by_pid(cgrp, tgid, true);
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 mutex_lock(&cgroup_root_mutex);
2317 strcpy(cgrp->root->release_agent_path, buffer);
2318 mutex_unlock(&cgroup_root_mutex);
2323 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2324 struct seq_file *seq)
2326 if (!cgroup_lock_live_group(cgrp))
2328 seq_puts(seq, cgrp->root->release_agent_path);
2329 seq_putc(seq, '\n');
2334 /* A buffer size big enough for numbers or short strings */
2335 #define CGROUP_LOCAL_BUFFER_SIZE 64
2337 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2339 const char __user *userbuf,
2340 size_t nbytes, loff_t *unused_ppos)
2342 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2348 if (nbytes >= sizeof(buffer))
2350 if (copy_from_user(buffer, userbuf, nbytes))
2353 buffer[nbytes] = 0; /* nul-terminate */
2354 if (cft->write_u64) {
2355 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2358 retval = cft->write_u64(cgrp, cft, val);
2360 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2363 retval = cft->write_s64(cgrp, cft, val);
2370 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2372 const char __user *userbuf,
2373 size_t nbytes, loff_t *unused_ppos)
2375 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2377 size_t max_bytes = cft->max_write_len;
2378 char *buffer = local_buffer;
2381 max_bytes = sizeof(local_buffer) - 1;
2382 if (nbytes >= max_bytes)
2384 /* Allocate a dynamic buffer if we need one */
2385 if (nbytes >= sizeof(local_buffer)) {
2386 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2390 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2395 buffer[nbytes] = 0; /* nul-terminate */
2396 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2400 if (buffer != local_buffer)
2405 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2406 size_t nbytes, loff_t *ppos)
2408 struct cftype *cft = __d_cft(file->f_dentry);
2409 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2411 if (cgroup_is_removed(cgrp))
2414 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2415 if (cft->write_u64 || cft->write_s64)
2416 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2417 if (cft->write_string)
2418 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2420 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2421 return ret ? ret : nbytes;
2426 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2428 char __user *buf, size_t nbytes,
2431 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2432 u64 val = cft->read_u64(cgrp, cft);
2433 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2435 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2438 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2440 char __user *buf, size_t nbytes,
2443 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2444 s64 val = cft->read_s64(cgrp, cft);
2445 int len = sprintf(tmp, "%lld\n", (long long) val);
2447 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2450 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2451 size_t nbytes, loff_t *ppos)
2453 struct cftype *cft = __d_cft(file->f_dentry);
2454 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2456 if (cgroup_is_removed(cgrp))
2460 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2462 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2464 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2469 * seqfile ops/methods for returning structured data. Currently just
2470 * supports string->u64 maps, but can be extended in future.
2473 struct cgroup_seqfile_state {
2475 struct cgroup *cgroup;
2478 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2480 struct seq_file *sf = cb->state;
2481 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2484 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2486 struct cgroup_seqfile_state *state = m->private;
2487 struct cftype *cft = state->cft;
2488 if (cft->read_map) {
2489 struct cgroup_map_cb cb = {
2490 .fill = cgroup_map_add,
2493 return cft->read_map(state->cgroup, cft, &cb);
2495 return cft->read_seq_string(state->cgroup, cft, m);
2498 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2500 struct seq_file *seq = file->private_data;
2501 kfree(seq->private);
2502 return single_release(inode, file);
2505 static const struct file_operations cgroup_seqfile_operations = {
2507 .write = cgroup_file_write,
2508 .llseek = seq_lseek,
2509 .release = cgroup_seqfile_release,
2512 static int cgroup_file_open(struct inode *inode, struct file *file)
2517 err = generic_file_open(inode, file);
2520 cft = __d_cft(file->f_dentry);
2522 if (cft->read_map || cft->read_seq_string) {
2523 struct cgroup_seqfile_state *state =
2524 kzalloc(sizeof(*state), GFP_USER);
2528 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2529 file->f_op = &cgroup_seqfile_operations;
2530 err = single_open(file, cgroup_seqfile_show, state);
2533 } else if (cft->open)
2534 err = cft->open(inode, file);
2541 static int cgroup_file_release(struct inode *inode, struct file *file)
2543 struct cftype *cft = __d_cft(file->f_dentry);
2545 return cft->release(inode, file);
2550 * cgroup_rename - Only allow simple rename of directories in place.
2552 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2553 struct inode *new_dir, struct dentry *new_dentry)
2555 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2557 if (new_dentry->d_inode)
2559 if (old_dir != new_dir)
2561 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2564 static const struct file_operations cgroup_file_operations = {
2565 .read = cgroup_file_read,
2566 .write = cgroup_file_write,
2567 .llseek = generic_file_llseek,
2568 .open = cgroup_file_open,
2569 .release = cgroup_file_release,
2572 static const struct inode_operations cgroup_dir_inode_operations = {
2573 .lookup = cgroup_lookup,
2574 .mkdir = cgroup_mkdir,
2575 .rmdir = cgroup_rmdir,
2576 .rename = cgroup_rename,
2579 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
2581 if (dentry->d_name.len > NAME_MAX)
2582 return ERR_PTR(-ENAMETOOLONG);
2583 d_add(dentry, NULL);
2588 * Check if a file is a control file
2590 static inline struct cftype *__file_cft(struct file *file)
2592 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2593 return ERR_PTR(-EINVAL);
2594 return __d_cft(file->f_dentry);
2597 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2598 struct super_block *sb)
2600 struct inode *inode;
2604 if (dentry->d_inode)
2607 inode = cgroup_new_inode(mode, sb);
2611 if (S_ISDIR(mode)) {
2612 inode->i_op = &cgroup_dir_inode_operations;
2613 inode->i_fop = &simple_dir_operations;
2615 /* start off with i_nlink == 2 (for "." entry) */
2618 /* start with the directory inode held, so that we can
2619 * populate it without racing with another mkdir */
2620 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2621 } else if (S_ISREG(mode)) {
2623 inode->i_fop = &cgroup_file_operations;
2625 d_instantiate(dentry, inode);
2626 dget(dentry); /* Extra count - pin the dentry in core */
2631 * cgroup_create_dir - create a directory for an object.
2632 * @cgrp: the cgroup we create the directory for. It must have a valid
2633 * ->parent field. And we are going to fill its ->dentry field.
2634 * @dentry: dentry of the new cgroup
2635 * @mode: mode to set on new directory.
2637 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2640 struct dentry *parent;
2643 parent = cgrp->parent->dentry;
2644 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2646 dentry->d_fsdata = cgrp;
2647 inc_nlink(parent->d_inode);
2648 rcu_assign_pointer(cgrp->dentry, dentry);
2657 * cgroup_file_mode - deduce file mode of a control file
2658 * @cft: the control file in question
2660 * returns cft->mode if ->mode is not 0
2661 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2662 * returns S_IRUGO if it has only a read handler
2663 * returns S_IWUSR if it has only a write hander
2665 static umode_t cgroup_file_mode(const struct cftype *cft)
2672 if (cft->read || cft->read_u64 || cft->read_s64 ||
2673 cft->read_map || cft->read_seq_string)
2676 if (cft->write || cft->write_u64 || cft->write_s64 ||
2677 cft->write_string || cft->trigger)
2683 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2684 const struct cftype *cft)
2686 struct dentry *dir = cgrp->dentry;
2687 struct cgroup *parent = __d_cgrp(dir);
2688 struct dentry *dentry;
2692 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2694 /* does @cft->flags tell us to skip creation on @cgrp? */
2695 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2697 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2700 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2701 strcpy(name, subsys->name);
2704 strcat(name, cft->name);
2706 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2708 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2712 dentry = lookup_one_len(name, dir, strlen(name));
2713 if (IS_ERR(dentry)) {
2714 error = PTR_ERR(dentry);
2718 mode = cgroup_file_mode(cft);
2719 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2721 cfe->type = (void *)cft;
2722 cfe->dentry = dentry;
2723 dentry->d_fsdata = cfe;
2724 list_add_tail(&cfe->node, &parent->files);
2733 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2734 const struct cftype cfts[], bool is_add)
2736 const struct cftype *cft;
2739 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2741 err = cgroup_add_file(cgrp, subsys, cft);
2743 err = cgroup_rm_file(cgrp, cft);
2745 pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2746 is_add ? "add" : "remove", cft->name, err);
2753 static DEFINE_MUTEX(cgroup_cft_mutex);
2755 static void cgroup_cfts_prepare(void)
2756 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2759 * Thanks to the entanglement with vfs inode locking, we can't walk
2760 * the existing cgroups under cgroup_mutex and create files.
2761 * Instead, we increment reference on all cgroups and build list of
2762 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2763 * exclusive access to the field.
2765 mutex_lock(&cgroup_cft_mutex);
2766 mutex_lock(&cgroup_mutex);
2769 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2770 const struct cftype *cfts, bool is_add)
2771 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2774 struct cgroup *cgrp, *n;
2776 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2777 if (cfts && ss->root != &rootnode) {
2778 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2780 list_add_tail(&cgrp->cft_q_node, &pending);
2784 mutex_unlock(&cgroup_mutex);
2787 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2788 * files for all cgroups which were created before.
2790 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2791 struct inode *inode = cgrp->dentry->d_inode;
2793 mutex_lock(&inode->i_mutex);
2794 mutex_lock(&cgroup_mutex);
2795 if (!cgroup_is_removed(cgrp))
2796 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2797 mutex_unlock(&cgroup_mutex);
2798 mutex_unlock(&inode->i_mutex);
2800 list_del_init(&cgrp->cft_q_node);
2804 mutex_unlock(&cgroup_cft_mutex);
2808 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2809 * @ss: target cgroup subsystem
2810 * @cfts: zero-length name terminated array of cftypes
2812 * Register @cfts to @ss. Files described by @cfts are created for all
2813 * existing cgroups to which @ss is attached and all future cgroups will
2814 * have them too. This function can be called anytime whether @ss is
2817 * Returns 0 on successful registration, -errno on failure. Note that this
2818 * function currently returns 0 as long as @cfts registration is successful
2819 * even if some file creation attempts on existing cgroups fail.
2821 int cgroup_add_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts)
2823 struct cftype_set *set;
2825 set = kzalloc(sizeof(*set), GFP_KERNEL);
2829 cgroup_cfts_prepare();
2831 list_add_tail(&set->node, &ss->cftsets);
2832 cgroup_cfts_commit(ss, cfts, true);
2836 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2839 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2840 * @ss: target cgroup subsystem
2841 * @cfts: zero-length name terminated array of cftypes
2843 * Unregister @cfts from @ss. Files described by @cfts are removed from
2844 * all existing cgroups to which @ss is attached and all future cgroups
2845 * won't have them either. This function can be called anytime whether @ss
2846 * is attached or not.
2848 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2849 * registered with @ss.
2851 int cgroup_rm_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts)
2853 struct cftype_set *set;
2855 cgroup_cfts_prepare();
2857 list_for_each_entry(set, &ss->cftsets, node) {
2858 if (set->cfts == cfts) {
2859 list_del_init(&set->node);
2860 cgroup_cfts_commit(ss, cfts, false);
2865 cgroup_cfts_commit(ss, NULL, false);
2870 * cgroup_task_count - count the number of tasks in a cgroup.
2871 * @cgrp: the cgroup in question
2873 * Return the number of tasks in the cgroup.
2875 int cgroup_task_count(const struct cgroup *cgrp)
2878 struct cg_cgroup_link *link;
2880 read_lock(&css_set_lock);
2881 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2882 count += atomic_read(&link->cg->refcount);
2884 read_unlock(&css_set_lock);
2889 * Advance a list_head iterator. The iterator should be positioned at
2890 * the start of a css_set
2892 static void cgroup_advance_iter(struct cgroup *cgrp,
2893 struct cgroup_iter *it)
2895 struct list_head *l = it->cg_link;
2896 struct cg_cgroup_link *link;
2899 /* Advance to the next non-empty css_set */
2902 if (l == &cgrp->css_sets) {
2906 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2908 } while (list_empty(&cg->tasks));
2910 it->task = cg->tasks.next;
2914 * To reduce the fork() overhead for systems that are not actually
2915 * using their cgroups capability, we don't maintain the lists running
2916 * through each css_set to its tasks until we see the list actually
2917 * used - in other words after the first call to cgroup_iter_start().
2919 static void cgroup_enable_task_cg_lists(void)
2921 struct task_struct *p, *g;
2922 write_lock(&css_set_lock);
2923 use_task_css_set_links = 1;
2925 * We need tasklist_lock because RCU is not safe against
2926 * while_each_thread(). Besides, a forking task that has passed
2927 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2928 * is not guaranteed to have its child immediately visible in the
2929 * tasklist if we walk through it with RCU.
2931 read_lock(&tasklist_lock);
2932 do_each_thread(g, p) {
2935 * We should check if the process is exiting, otherwise
2936 * it will race with cgroup_exit() in that the list
2937 * entry won't be deleted though the process has exited.
2939 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2940 list_add(&p->cg_list, &p->cgroups->tasks);
2942 } while_each_thread(g, p);
2943 read_unlock(&tasklist_lock);
2944 write_unlock(&css_set_lock);
2947 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2948 __acquires(css_set_lock)
2951 * The first time anyone tries to iterate across a cgroup,
2952 * we need to enable the list linking each css_set to its
2953 * tasks, and fix up all existing tasks.
2955 if (!use_task_css_set_links)
2956 cgroup_enable_task_cg_lists();
2958 read_lock(&css_set_lock);
2959 it->cg_link = &cgrp->css_sets;
2960 cgroup_advance_iter(cgrp, it);
2963 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2964 struct cgroup_iter *it)
2966 struct task_struct *res;
2967 struct list_head *l = it->task;
2968 struct cg_cgroup_link *link;
2970 /* If the iterator cg is NULL, we have no tasks */
2973 res = list_entry(l, struct task_struct, cg_list);
2974 /* Advance iterator to find next entry */
2976 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2977 if (l == &link->cg->tasks) {
2978 /* We reached the end of this task list - move on to
2979 * the next cg_cgroup_link */
2980 cgroup_advance_iter(cgrp, it);
2987 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2988 __releases(css_set_lock)
2990 read_unlock(&css_set_lock);
2993 static inline int started_after_time(struct task_struct *t1,
2994 struct timespec *time,
2995 struct task_struct *t2)
2997 int start_diff = timespec_compare(&t1->start_time, time);
2998 if (start_diff > 0) {
3000 } else if (start_diff < 0) {
3004 * Arbitrarily, if two processes started at the same
3005 * time, we'll say that the lower pointer value
3006 * started first. Note that t2 may have exited by now
3007 * so this may not be a valid pointer any longer, but
3008 * that's fine - it still serves to distinguish
3009 * between two tasks started (effectively) simultaneously.
3016 * This function is a callback from heap_insert() and is used to order
3018 * In this case we order the heap in descending task start time.
3020 static inline int started_after(void *p1, void *p2)
3022 struct task_struct *t1 = p1;
3023 struct task_struct *t2 = p2;
3024 return started_after_time(t1, &t2->start_time, t2);
3028 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3029 * @scan: struct cgroup_scanner containing arguments for the scan
3031 * Arguments include pointers to callback functions test_task() and
3033 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3034 * and if it returns true, call process_task() for it also.
3035 * The test_task pointer may be NULL, meaning always true (select all tasks).
3036 * Effectively duplicates cgroup_iter_{start,next,end}()
3037 * but does not lock css_set_lock for the call to process_task().
3038 * The struct cgroup_scanner may be embedded in any structure of the caller's
3040 * It is guaranteed that process_task() will act on every task that
3041 * is a member of the cgroup for the duration of this call. This
3042 * function may or may not call process_task() for tasks that exit
3043 * or move to a different cgroup during the call, or are forked or
3044 * move into the cgroup during the call.
3046 * Note that test_task() may be called with locks held, and may in some
3047 * situations be called multiple times for the same task, so it should
3049 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3050 * pre-allocated and will be used for heap operations (and its "gt" member will
3051 * be overwritten), else a temporary heap will be used (allocation of which
3052 * may cause this function to fail).
3054 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3057 struct cgroup_iter it;
3058 struct task_struct *p, *dropped;
3059 /* Never dereference latest_task, since it's not refcounted */
3060 struct task_struct *latest_task = NULL;
3061 struct ptr_heap tmp_heap;
3062 struct ptr_heap *heap;
3063 struct timespec latest_time = { 0, 0 };
3066 /* The caller supplied our heap and pre-allocated its memory */
3068 heap->gt = &started_after;
3070 /* We need to allocate our own heap memory */
3072 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3074 /* cannot allocate the heap */
3080 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3081 * to determine which are of interest, and using the scanner's
3082 * "process_task" callback to process any of them that need an update.
3083 * Since we don't want to hold any locks during the task updates,
3084 * gather tasks to be processed in a heap structure.
3085 * The heap is sorted by descending task start time.
3086 * If the statically-sized heap fills up, we overflow tasks that
3087 * started later, and in future iterations only consider tasks that
3088 * started after the latest task in the previous pass. This
3089 * guarantees forward progress and that we don't miss any tasks.
3092 cgroup_iter_start(scan->cg, &it);
3093 while ((p = cgroup_iter_next(scan->cg, &it))) {
3095 * Only affect tasks that qualify per the caller's callback,
3096 * if he provided one
3098 if (scan->test_task && !scan->test_task(p, scan))
3101 * Only process tasks that started after the last task
3104 if (!started_after_time(p, &latest_time, latest_task))
3106 dropped = heap_insert(heap, p);
3107 if (dropped == NULL) {
3109 * The new task was inserted; the heap wasn't
3113 } else if (dropped != p) {
3115 * The new task was inserted, and pushed out a
3119 put_task_struct(dropped);
3122 * Else the new task was newer than anything already in
3123 * the heap and wasn't inserted
3126 cgroup_iter_end(scan->cg, &it);
3129 for (i = 0; i < heap->size; i++) {
3130 struct task_struct *q = heap->ptrs[i];
3132 latest_time = q->start_time;
3135 /* Process the task per the caller's callback */
3136 scan->process_task(q, scan);
3140 * If we had to process any tasks at all, scan again
3141 * in case some of them were in the middle of forking
3142 * children that didn't get processed.
3143 * Not the most efficient way to do it, but it avoids
3144 * having to take callback_mutex in the fork path
3148 if (heap == &tmp_heap)
3149 heap_free(&tmp_heap);
3154 * Stuff for reading the 'tasks'/'procs' files.
3156 * Reading this file can return large amounts of data if a cgroup has
3157 * *lots* of attached tasks. So it may need several calls to read(),
3158 * but we cannot guarantee that the information we produce is correct
3159 * unless we produce it entirely atomically.
3163 /* which pidlist file are we talking about? */
3164 enum cgroup_filetype {
3170 * A pidlist is a list of pids that virtually represents the contents of one
3171 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3172 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3175 struct cgroup_pidlist {
3177 * used to find which pidlist is wanted. doesn't change as long as
3178 * this particular list stays in the list.
3180 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3183 /* how many elements the above list has */
3185 /* how many files are using the current array */
3187 /* each of these stored in a list by its cgroup */
3188 struct list_head links;
3189 /* pointer to the cgroup we belong to, for list removal purposes */
3190 struct cgroup *owner;
3191 /* protects the other fields */
3192 struct rw_semaphore mutex;
3196 * The following two functions "fix" the issue where there are more pids
3197 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3198 * TODO: replace with a kernel-wide solution to this problem
3200 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3201 static void *pidlist_allocate(int count)
3203 if (PIDLIST_TOO_LARGE(count))
3204 return vmalloc(count * sizeof(pid_t));
3206 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3208 static void pidlist_free(void *p)
3210 if (is_vmalloc_addr(p))
3215 static void *pidlist_resize(void *p, int newcount)
3218 /* note: if new alloc fails, old p will still be valid either way */
3219 if (is_vmalloc_addr(p)) {
3220 newlist = vmalloc(newcount * sizeof(pid_t));
3223 memcpy(newlist, p, newcount * sizeof(pid_t));
3226 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3232 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3233 * If the new stripped list is sufficiently smaller and there's enough memory
3234 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3235 * number of unique elements.
3237 /* is the size difference enough that we should re-allocate the array? */
3238 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3239 static int pidlist_uniq(pid_t **p, int length)
3246 * we presume the 0th element is unique, so i starts at 1. trivial
3247 * edge cases first; no work needs to be done for either
3249 if (length == 0 || length == 1)
3251 /* src and dest walk down the list; dest counts unique elements */
3252 for (src = 1; src < length; src++) {
3253 /* find next unique element */
3254 while (list[src] == list[src-1]) {
3259 /* dest always points to where the next unique element goes */
3260 list[dest] = list[src];
3265 * if the length difference is large enough, we want to allocate a
3266 * smaller buffer to save memory. if this fails due to out of memory,
3267 * we'll just stay with what we've got.
3269 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3270 newlist = pidlist_resize(list, dest);
3277 static int cmppid(const void *a, const void *b)
3279 return *(pid_t *)a - *(pid_t *)b;
3283 * find the appropriate pidlist for our purpose (given procs vs tasks)
3284 * returns with the lock on that pidlist already held, and takes care
3285 * of the use count, or returns NULL with no locks held if we're out of
3288 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3289 enum cgroup_filetype type)
3291 struct cgroup_pidlist *l;
3292 /* don't need task_nsproxy() if we're looking at ourself */
3293 struct pid_namespace *ns = current->nsproxy->pid_ns;
3296 * We can't drop the pidlist_mutex before taking the l->mutex in case
3297 * the last ref-holder is trying to remove l from the list at the same
3298 * time. Holding the pidlist_mutex precludes somebody taking whichever
3299 * list we find out from under us - compare release_pid_array().
3301 mutex_lock(&cgrp->pidlist_mutex);
3302 list_for_each_entry(l, &cgrp->pidlists, links) {
3303 if (l->key.type == type && l->key.ns == ns) {
3304 /* make sure l doesn't vanish out from under us */
3305 down_write(&l->mutex);
3306 mutex_unlock(&cgrp->pidlist_mutex);
3310 /* entry not found; create a new one */
3311 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3313 mutex_unlock(&cgrp->pidlist_mutex);
3316 init_rwsem(&l->mutex);
3317 down_write(&l->mutex);
3319 l->key.ns = get_pid_ns(ns);
3320 l->use_count = 0; /* don't increment here */
3323 list_add(&l->links, &cgrp->pidlists);
3324 mutex_unlock(&cgrp->pidlist_mutex);
3329 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3331 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3332 struct cgroup_pidlist **lp)
3336 int pid, n = 0; /* used for populating the array */
3337 struct cgroup_iter it;
3338 struct task_struct *tsk;
3339 struct cgroup_pidlist *l;
3342 * If cgroup gets more users after we read count, we won't have
3343 * enough space - tough. This race is indistinguishable to the
3344 * caller from the case that the additional cgroup users didn't
3345 * show up until sometime later on.
3347 length = cgroup_task_count(cgrp);
3348 array = pidlist_allocate(length);
3351 /* now, populate the array */
3352 cgroup_iter_start(cgrp, &it);
3353 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3354 if (unlikely(n == length))
3356 /* get tgid or pid for procs or tasks file respectively */
3357 if (type == CGROUP_FILE_PROCS)
3358 pid = task_tgid_vnr(tsk);
3360 pid = task_pid_vnr(tsk);
3361 if (pid > 0) /* make sure to only use valid results */
3364 cgroup_iter_end(cgrp, &it);
3366 /* now sort & (if procs) strip out duplicates */
3367 sort(array, length, sizeof(pid_t), cmppid, NULL);
3368 if (type == CGROUP_FILE_PROCS)
3369 length = pidlist_uniq(&array, length);
3370 l = cgroup_pidlist_find(cgrp, type);
3372 pidlist_free(array);
3375 /* store array, freeing old if necessary - lock already held */
3376 pidlist_free(l->list);
3380 up_write(&l->mutex);
3386 * cgroupstats_build - build and fill cgroupstats
3387 * @stats: cgroupstats to fill information into
3388 * @dentry: A dentry entry belonging to the cgroup for which stats have
3391 * Build and fill cgroupstats so that taskstats can export it to user
3394 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3397 struct cgroup *cgrp;
3398 struct cgroup_iter it;
3399 struct task_struct *tsk;
3402 * Validate dentry by checking the superblock operations,
3403 * and make sure it's a directory.
3405 if (dentry->d_sb->s_op != &cgroup_ops ||
3406 !S_ISDIR(dentry->d_inode->i_mode))
3410 cgrp = dentry->d_fsdata;
3412 cgroup_iter_start(cgrp, &it);
3413 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3414 switch (tsk->state) {
3416 stats->nr_running++;
3418 case TASK_INTERRUPTIBLE:
3419 stats->nr_sleeping++;
3421 case TASK_UNINTERRUPTIBLE:
3422 stats->nr_uninterruptible++;
3425 stats->nr_stopped++;
3428 if (delayacct_is_task_waiting_on_io(tsk))
3429 stats->nr_io_wait++;
3433 cgroup_iter_end(cgrp, &it);
3441 * seq_file methods for the tasks/procs files. The seq_file position is the
3442 * next pid to display; the seq_file iterator is a pointer to the pid
3443 * in the cgroup->l->list array.
3446 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3449 * Initially we receive a position value that corresponds to
3450 * one more than the last pid shown (or 0 on the first call or
3451 * after a seek to the start). Use a binary-search to find the
3452 * next pid to display, if any
3454 struct cgroup_pidlist *l = s->private;
3455 int index = 0, pid = *pos;
3458 down_read(&l->mutex);
3460 int end = l->length;
3462 while (index < end) {
3463 int mid = (index + end) / 2;
3464 if (l->list[mid] == pid) {
3467 } else if (l->list[mid] <= pid)
3473 /* If we're off the end of the array, we're done */
3474 if (index >= l->length)
3476 /* Update the abstract position to be the actual pid that we found */
3477 iter = l->list + index;
3482 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3484 struct cgroup_pidlist *l = s->private;
3488 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3490 struct cgroup_pidlist *l = s->private;
3492 pid_t *end = l->list + l->length;
3494 * Advance to the next pid in the array. If this goes off the
3506 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3508 return seq_printf(s, "%d\n", *(int *)v);
3512 * seq_operations functions for iterating on pidlists through seq_file -
3513 * independent of whether it's tasks or procs
3515 static const struct seq_operations cgroup_pidlist_seq_operations = {
3516 .start = cgroup_pidlist_start,
3517 .stop = cgroup_pidlist_stop,
3518 .next = cgroup_pidlist_next,
3519 .show = cgroup_pidlist_show,
3522 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3525 * the case where we're the last user of this particular pidlist will
3526 * have us remove it from the cgroup's list, which entails taking the
3527 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3528 * pidlist_mutex, we have to take pidlist_mutex first.
3530 mutex_lock(&l->owner->pidlist_mutex);
3531 down_write(&l->mutex);
3532 BUG_ON(!l->use_count);
3533 if (!--l->use_count) {
3534 /* we're the last user if refcount is 0; remove and free */
3535 list_del(&l->links);
3536 mutex_unlock(&l->owner->pidlist_mutex);
3537 pidlist_free(l->list);
3538 put_pid_ns(l->key.ns);
3539 up_write(&l->mutex);
3543 mutex_unlock(&l->owner->pidlist_mutex);
3544 up_write(&l->mutex);
3547 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3549 struct cgroup_pidlist *l;
3550 if (!(file->f_mode & FMODE_READ))
3553 * the seq_file will only be initialized if the file was opened for
3554 * reading; hence we check if it's not null only in that case.
3556 l = ((struct seq_file *)file->private_data)->private;
3557 cgroup_release_pid_array(l);
3558 return seq_release(inode, file);
3561 static const struct file_operations cgroup_pidlist_operations = {
3563 .llseek = seq_lseek,
3564 .write = cgroup_file_write,
3565 .release = cgroup_pidlist_release,
3569 * The following functions handle opens on a file that displays a pidlist
3570 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3573 /* helper function for the two below it */
3574 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3576 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3577 struct cgroup_pidlist *l;
3580 /* Nothing to do for write-only files */
3581 if (!(file->f_mode & FMODE_READ))
3584 /* have the array populated */
3585 retval = pidlist_array_load(cgrp, type, &l);
3588 /* configure file information */
3589 file->f_op = &cgroup_pidlist_operations;
3591 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3593 cgroup_release_pid_array(l);
3596 ((struct seq_file *)file->private_data)->private = l;
3599 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3601 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3603 static int cgroup_procs_open(struct inode *unused, struct file *file)
3605 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3608 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3611 return notify_on_release(cgrp);
3614 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3618 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3620 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3622 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3627 * Unregister event and free resources.
3629 * Gets called from workqueue.
3631 static void cgroup_event_remove(struct work_struct *work)
3633 struct cgroup_event *event = container_of(work, struct cgroup_event,
3635 struct cgroup *cgrp = event->cgrp;
3637 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3639 eventfd_ctx_put(event->eventfd);
3645 * Gets called on POLLHUP on eventfd when user closes it.
3647 * Called with wqh->lock held and interrupts disabled.
3649 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3650 int sync, void *key)
3652 struct cgroup_event *event = container_of(wait,
3653 struct cgroup_event, wait);
3654 struct cgroup *cgrp = event->cgrp;
3655 unsigned long flags = (unsigned long)key;
3657 if (flags & POLLHUP) {
3658 __remove_wait_queue(event->wqh, &event->wait);
3659 spin_lock(&cgrp->event_list_lock);
3660 list_del(&event->list);
3661 spin_unlock(&cgrp->event_list_lock);
3663 * We are in atomic context, but cgroup_event_remove() may
3664 * sleep, so we have to call it in workqueue.
3666 schedule_work(&event->remove);
3672 static void cgroup_event_ptable_queue_proc(struct file *file,
3673 wait_queue_head_t *wqh, poll_table *pt)
3675 struct cgroup_event *event = container_of(pt,
3676 struct cgroup_event, pt);
3679 add_wait_queue(wqh, &event->wait);
3683 * Parse input and register new cgroup event handler.
3685 * Input must be in format '<event_fd> <control_fd> <args>'.
3686 * Interpretation of args is defined by control file implementation.
3688 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3691 struct cgroup_event *event = NULL;
3692 unsigned int efd, cfd;
3693 struct file *efile = NULL;
3694 struct file *cfile = NULL;
3698 efd = simple_strtoul(buffer, &endp, 10);
3703 cfd = simple_strtoul(buffer, &endp, 10);
3704 if ((*endp != ' ') && (*endp != '\0'))
3708 event = kzalloc(sizeof(*event), GFP_KERNEL);
3712 INIT_LIST_HEAD(&event->list);
3713 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3714 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3715 INIT_WORK(&event->remove, cgroup_event_remove);
3717 efile = eventfd_fget(efd);
3718 if (IS_ERR(efile)) {
3719 ret = PTR_ERR(efile);
3723 event->eventfd = eventfd_ctx_fileget(efile);
3724 if (IS_ERR(event->eventfd)) {
3725 ret = PTR_ERR(event->eventfd);
3735 /* the process need read permission on control file */
3736 /* AV: shouldn't we check that it's been opened for read instead? */
3737 ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3741 event->cft = __file_cft(cfile);
3742 if (IS_ERR(event->cft)) {
3743 ret = PTR_ERR(event->cft);
3747 if (!event->cft->register_event || !event->cft->unregister_event) {
3752 ret = event->cft->register_event(cgrp, event->cft,
3753 event->eventfd, buffer);
3757 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3758 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3764 * Events should be removed after rmdir of cgroup directory, but before
3765 * destroying subsystem state objects. Let's take reference to cgroup
3766 * directory dentry to do that.
3770 spin_lock(&cgrp->event_list_lock);
3771 list_add(&event->list, &cgrp->event_list);
3772 spin_unlock(&cgrp->event_list_lock);
3783 if (event && event->eventfd && !IS_ERR(event->eventfd))
3784 eventfd_ctx_put(event->eventfd);
3786 if (!IS_ERR_OR_NULL(efile))
3794 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3797 return clone_children(cgrp);
3800 static int cgroup_clone_children_write(struct cgroup *cgrp,
3805 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3807 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3812 * for the common functions, 'private' gives the type of file
3814 /* for hysterical raisins, we can't put this on the older files */
3815 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3816 static struct cftype files[] = {
3819 .open = cgroup_tasks_open,
3820 .write_u64 = cgroup_tasks_write,
3821 .release = cgroup_pidlist_release,
3822 .mode = S_IRUGO | S_IWUSR,
3825 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3826 .open = cgroup_procs_open,
3827 .write_u64 = cgroup_procs_write,
3828 .release = cgroup_pidlist_release,
3829 .mode = S_IRUGO | S_IWUSR,
3832 .name = "notify_on_release",
3833 .read_u64 = cgroup_read_notify_on_release,
3834 .write_u64 = cgroup_write_notify_on_release,
3837 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3838 .write_string = cgroup_write_event_control,
3842 .name = "cgroup.clone_children",
3843 .read_u64 = cgroup_clone_children_read,
3844 .write_u64 = cgroup_clone_children_write,
3847 .name = "release_agent",
3848 .flags = CFTYPE_ONLY_ON_ROOT,
3849 .read_seq_string = cgroup_release_agent_show,
3850 .write_string = cgroup_release_agent_write,
3851 .max_write_len = PATH_MAX,
3856 static int cgroup_populate_dir(struct cgroup *cgrp)
3859 struct cgroup_subsys *ss;
3861 err = cgroup_addrm_files(cgrp, NULL, files, true);
3865 /* process cftsets of each subsystem */
3866 for_each_subsys(cgrp->root, ss) {
3867 struct cftype_set *set;
3869 list_for_each_entry(set, &ss->cftsets, node)
3870 cgroup_addrm_files(cgrp, ss, set->cfts, true);
3873 /* This cgroup is ready now */
3874 for_each_subsys(cgrp->root, ss) {
3875 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3877 * Update id->css pointer and make this css visible from
3878 * CSS ID functions. This pointer will be dereferened
3879 * from RCU-read-side without locks.
3882 rcu_assign_pointer(css->id->css, css);
3888 static void css_dput_fn(struct work_struct *work)
3890 struct cgroup_subsys_state *css =
3891 container_of(work, struct cgroup_subsys_state, dput_work);
3893 dput(css->cgroup->dentry);
3896 static void init_cgroup_css(struct cgroup_subsys_state *css,
3897 struct cgroup_subsys *ss,
3898 struct cgroup *cgrp)
3901 atomic_set(&css->refcnt, 1);
3904 if (cgrp == dummytop)
3905 set_bit(CSS_ROOT, &css->flags);
3906 BUG_ON(cgrp->subsys[ss->subsys_id]);
3907 cgrp->subsys[ss->subsys_id] = css;
3910 * If !clear_css_refs, css holds an extra ref to @cgrp->dentry
3911 * which is put on the last css_put(). dput() requires process
3912 * context, which css_put() may be called without. @css->dput_work
3913 * will be used to invoke dput() asynchronously from css_put().
3915 INIT_WORK(&css->dput_work, css_dput_fn);
3916 if (ss->__DEPRECATED_clear_css_refs)
3917 set_bit(CSS_CLEAR_CSS_REFS, &css->flags);
3920 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3922 /* We need to take each hierarchy_mutex in a consistent order */
3926 * No worry about a race with rebind_subsystems that might mess up the
3927 * locking order, since both parties are under cgroup_mutex.
3929 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3930 struct cgroup_subsys *ss = subsys[i];
3933 if (ss->root == root)
3934 mutex_lock(&ss->hierarchy_mutex);
3938 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3942 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3943 struct cgroup_subsys *ss = subsys[i];
3946 if (ss->root == root)
3947 mutex_unlock(&ss->hierarchy_mutex);
3952 * cgroup_create - create a cgroup
3953 * @parent: cgroup that will be parent of the new cgroup
3954 * @dentry: dentry of the new cgroup
3955 * @mode: mode to set on new inode
3957 * Must be called with the mutex on the parent inode held
3959 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3962 struct cgroup *cgrp;
3963 struct cgroupfs_root *root = parent->root;
3965 struct cgroup_subsys *ss;
3966 struct super_block *sb = root->sb;
3968 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3972 /* Grab a reference on the superblock so the hierarchy doesn't
3973 * get deleted on unmount if there are child cgroups. This
3974 * can be done outside cgroup_mutex, since the sb can't
3975 * disappear while someone has an open control file on the
3977 atomic_inc(&sb->s_active);
3979 mutex_lock(&cgroup_mutex);
3981 init_cgroup_housekeeping(cgrp);
3983 cgrp->parent = parent;
3984 cgrp->root = parent->root;
3985 cgrp->top_cgroup = parent->top_cgroup;
3987 if (notify_on_release(parent))
3988 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3990 if (clone_children(parent))
3991 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3993 for_each_subsys(root, ss) {
3994 struct cgroup_subsys_state *css = ss->create(cgrp);
4000 init_cgroup_css(css, ss, cgrp);
4002 err = alloc_css_id(ss, parent, cgrp);
4006 /* At error, ->destroy() callback has to free assigned ID. */
4007 if (clone_children(parent) && ss->post_clone)
4008 ss->post_clone(cgrp);
4011 cgroup_lock_hierarchy(root);
4012 list_add(&cgrp->sibling, &cgrp->parent->children);
4013 cgroup_unlock_hierarchy(root);
4014 root->number_of_cgroups++;
4016 err = cgroup_create_dir(cgrp, dentry, mode);
4020 /* If !clear_css_refs, each css holds a ref to the cgroup's dentry */
4021 for_each_subsys(root, ss)
4022 if (!ss->__DEPRECATED_clear_css_refs)
4025 /* The cgroup directory was pre-locked for us */
4026 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
4028 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4030 err = cgroup_populate_dir(cgrp);
4031 /* If err < 0, we have a half-filled directory - oh well ;) */
4033 mutex_unlock(&cgroup_mutex);
4034 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4040 cgroup_lock_hierarchy(root);
4041 list_del(&cgrp->sibling);
4042 cgroup_unlock_hierarchy(root);
4043 root->number_of_cgroups--;
4047 for_each_subsys(root, ss) {
4048 if (cgrp->subsys[ss->subsys_id])
4052 mutex_unlock(&cgroup_mutex);
4054 /* Release the reference count that we took on the superblock */
4055 deactivate_super(sb);
4061 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4063 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4065 /* the vfs holds inode->i_mutex already */
4066 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4070 * Check the reference count on each subsystem. Since we already
4071 * established that there are no tasks in the cgroup, if the css refcount
4072 * is also 1, then there should be no outstanding references, so the
4073 * subsystem is safe to destroy. We scan across all subsystems rather than
4074 * using the per-hierarchy linked list of mounted subsystems since we can
4075 * be called via check_for_release() with no synchronization other than
4076 * RCU, and the subsystem linked list isn't RCU-safe.
4078 static int cgroup_has_css_refs(struct cgroup *cgrp)
4083 * We won't need to lock the subsys array, because the subsystems
4084 * we're concerned about aren't going anywhere since our cgroup root
4085 * has a reference on them.
4087 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4088 struct cgroup_subsys *ss = subsys[i];
4089 struct cgroup_subsys_state *css;
4091 /* Skip subsystems not present or not in this hierarchy */
4092 if (ss == NULL || ss->root != cgrp->root)
4095 css = cgrp->subsys[ss->subsys_id];
4097 * When called from check_for_release() it's possible
4098 * that by this point the cgroup has been removed
4099 * and the css deleted. But a false-positive doesn't
4100 * matter, since it can only happen if the cgroup
4101 * has been deleted and hence no longer needs the
4102 * release agent to be called anyway.
4104 if (css && css_refcnt(css) > 1)
4111 * Atomically mark all (or else none) of the cgroup's CSS objects as
4112 * CSS_REMOVED. Return true on success, or false if the cgroup has
4113 * busy subsystems. Call with cgroup_mutex held
4115 * Depending on whether a subsys has __DEPRECATED_clear_css_refs set or
4116 * not, cgroup removal behaves differently.
4118 * If clear is set, css refcnt for the subsystem should be zero before
4119 * cgroup removal can be committed. This is implemented by
4120 * CGRP_WAIT_ON_RMDIR and retry logic around ->pre_destroy(), which may be
4121 * called multiple times until all css refcnts reach zero and is allowed to
4122 * veto removal on any invocation. This behavior is deprecated and will be
4123 * removed as soon as the existing user (memcg) is updated.
4125 * If clear is not set, each css holds an extra reference to the cgroup's
4126 * dentry and cgroup removal proceeds regardless of css refs.
4127 * ->pre_destroy() will be called at least once and is not allowed to fail.
4128 * On the last put of each css, whenever that may be, the extra dentry ref
4129 * is put so that dentry destruction happens only after all css's are
4132 static int cgroup_clear_css_refs(struct cgroup *cgrp)
4134 struct cgroup_subsys *ss;
4135 unsigned long flags;
4136 bool failed = false;
4138 local_irq_save(flags);
4141 * Block new css_tryget() by deactivating refcnt. If all refcnts
4142 * for subsystems w/ clear_css_refs set were 1 at the moment of
4143 * deactivation, we succeeded.
4145 for_each_subsys(cgrp->root, ss) {
4146 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4148 WARN_ON(atomic_read(&css->refcnt) < 0);
4149 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4151 if (ss->__DEPRECATED_clear_css_refs)
4152 failed |= css_refcnt(css) != 1;
4156 * If succeeded, set REMOVED and put all the base refs; otherwise,
4157 * restore refcnts to positive values. Either way, all in-progress
4158 * css_tryget() will be released.
4160 for_each_subsys(cgrp->root, ss) {
4161 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4164 set_bit(CSS_REMOVED, &css->flags);
4167 atomic_sub(CSS_DEACT_BIAS, &css->refcnt);
4171 local_irq_restore(flags);
4175 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4177 struct cgroup *cgrp = dentry->d_fsdata;
4179 struct cgroup *parent;
4181 struct cgroup_event *event, *tmp;
4184 /* the vfs holds both inode->i_mutex already */
4186 mutex_lock(&cgroup_mutex);
4187 if (atomic_read(&cgrp->count) != 0) {
4188 mutex_unlock(&cgroup_mutex);
4191 if (!list_empty(&cgrp->children)) {
4192 mutex_unlock(&cgroup_mutex);
4195 mutex_unlock(&cgroup_mutex);
4198 * In general, subsystem has no css->refcnt after pre_destroy(). But
4199 * in racy cases, subsystem may have to get css->refcnt after
4200 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4201 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4202 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4203 * and subsystem's reference count handling. Please see css_get/put
4204 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4206 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4209 * Call pre_destroy handlers of subsys. Notify subsystems
4210 * that rmdir() request comes.
4212 ret = cgroup_call_pre_destroy(cgrp);
4214 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4218 mutex_lock(&cgroup_mutex);
4219 parent = cgrp->parent;
4220 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
4221 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4222 mutex_unlock(&cgroup_mutex);
4225 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
4226 if (!cgroup_clear_css_refs(cgrp)) {
4227 mutex_unlock(&cgroup_mutex);
4229 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4230 * prepare_to_wait(), we need to check this flag.
4232 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
4234 finish_wait(&cgroup_rmdir_waitq, &wait);
4235 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4236 if (signal_pending(current))
4240 /* NO css_tryget() can success after here. */
4241 finish_wait(&cgroup_rmdir_waitq, &wait);
4242 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4244 raw_spin_lock(&release_list_lock);
4245 set_bit(CGRP_REMOVED, &cgrp->flags);
4246 if (!list_empty(&cgrp->release_list))
4247 list_del_init(&cgrp->release_list);
4248 raw_spin_unlock(&release_list_lock);
4250 cgroup_lock_hierarchy(cgrp->root);
4251 /* delete this cgroup from parent->children */
4252 list_del_init(&cgrp->sibling);
4253 cgroup_unlock_hierarchy(cgrp->root);
4255 list_del_init(&cgrp->allcg_node);
4257 d = dget(cgrp->dentry);
4259 cgroup_d_remove_dir(d);
4262 set_bit(CGRP_RELEASABLE, &parent->flags);
4263 check_for_release(parent);
4266 * Unregister events and notify userspace.
4267 * Notify userspace about cgroup removing only after rmdir of cgroup
4268 * directory to avoid race between userspace and kernelspace
4270 spin_lock(&cgrp->event_list_lock);
4271 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4272 list_del(&event->list);
4273 remove_wait_queue(event->wqh, &event->wait);
4274 eventfd_signal(event->eventfd, 1);
4275 schedule_work(&event->remove);
4277 spin_unlock(&cgrp->event_list_lock);
4279 mutex_unlock(&cgroup_mutex);
4283 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4285 INIT_LIST_HEAD(&ss->cftsets);
4288 * base_cftset is embedded in subsys itself, no need to worry about
4291 if (ss->base_cftypes) {
4292 ss->base_cftset.cfts = ss->base_cftypes;
4293 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4297 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4299 struct cgroup_subsys_state *css;
4301 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4303 /* init base cftset */
4304 cgroup_init_cftsets(ss);
4306 /* Create the top cgroup state for this subsystem */
4307 list_add(&ss->sibling, &rootnode.subsys_list);
4308 ss->root = &rootnode;
4309 css = ss->create(dummytop);
4310 /* We don't handle early failures gracefully */
4311 BUG_ON(IS_ERR(css));
4312 init_cgroup_css(css, ss, dummytop);
4314 /* Update the init_css_set to contain a subsys
4315 * pointer to this state - since the subsystem is
4316 * newly registered, all tasks and hence the
4317 * init_css_set is in the subsystem's top cgroup. */
4318 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4320 need_forkexit_callback |= ss->fork || ss->exit;
4322 /* At system boot, before all subsystems have been
4323 * registered, no tasks have been forked, so we don't
4324 * need to invoke fork callbacks here. */
4325 BUG_ON(!list_empty(&init_task.tasks));
4327 mutex_init(&ss->hierarchy_mutex);
4328 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4331 /* this function shouldn't be used with modular subsystems, since they
4332 * need to register a subsys_id, among other things */
4337 * cgroup_load_subsys: load and register a modular subsystem at runtime
4338 * @ss: the subsystem to load
4340 * This function should be called in a modular subsystem's initcall. If the
4341 * subsystem is built as a module, it will be assigned a new subsys_id and set
4342 * up for use. If the subsystem is built-in anyway, work is delegated to the
4343 * simpler cgroup_init_subsys.
4345 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4348 struct cgroup_subsys_state *css;
4350 /* check name and function validity */
4351 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4352 ss->create == NULL || ss->destroy == NULL)
4356 * we don't support callbacks in modular subsystems. this check is
4357 * before the ss->module check for consistency; a subsystem that could
4358 * be a module should still have no callbacks even if the user isn't
4359 * compiling it as one.
4361 if (ss->fork || ss->exit)
4365 * an optionally modular subsystem is built-in: we want to do nothing,
4366 * since cgroup_init_subsys will have already taken care of it.
4368 if (ss->module == NULL) {
4369 /* a few sanity checks */
4370 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4371 BUG_ON(subsys[ss->subsys_id] != ss);
4375 /* init base cftset */
4376 cgroup_init_cftsets(ss);
4379 * need to register a subsys id before anything else - for example,
4380 * init_cgroup_css needs it.
4382 mutex_lock(&cgroup_mutex);
4383 /* find the first empty slot in the array */
4384 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4385 if (subsys[i] == NULL)
4388 if (i == CGROUP_SUBSYS_COUNT) {
4389 /* maximum number of subsystems already registered! */
4390 mutex_unlock(&cgroup_mutex);
4393 /* assign ourselves the subsys_id */
4398 * no ss->create seems to need anything important in the ss struct, so
4399 * this can happen first (i.e. before the rootnode attachment).
4401 css = ss->create(dummytop);
4403 /* failure case - need to deassign the subsys[] slot. */
4405 mutex_unlock(&cgroup_mutex);
4406 return PTR_ERR(css);
4409 list_add(&ss->sibling, &rootnode.subsys_list);
4410 ss->root = &rootnode;
4412 /* our new subsystem will be attached to the dummy hierarchy. */
4413 init_cgroup_css(css, ss, dummytop);
4414 /* init_idr must be after init_cgroup_css because it sets css->id. */
4416 int ret = cgroup_init_idr(ss, css);
4418 dummytop->subsys[ss->subsys_id] = NULL;
4419 ss->destroy(dummytop);
4421 mutex_unlock(&cgroup_mutex);
4427 * Now we need to entangle the css into the existing css_sets. unlike
4428 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4429 * will need a new pointer to it; done by iterating the css_set_table.
4430 * furthermore, modifying the existing css_sets will corrupt the hash
4431 * table state, so each changed css_set will need its hash recomputed.
4432 * this is all done under the css_set_lock.
4434 write_lock(&css_set_lock);
4435 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4437 struct hlist_node *node, *tmp;
4438 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4440 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4441 /* skip entries that we already rehashed */
4442 if (cg->subsys[ss->subsys_id])
4444 /* remove existing entry */
4445 hlist_del(&cg->hlist);
4447 cg->subsys[ss->subsys_id] = css;
4448 /* recompute hash and restore entry */
4449 new_bucket = css_set_hash(cg->subsys);
4450 hlist_add_head(&cg->hlist, new_bucket);
4453 write_unlock(&css_set_lock);
4455 mutex_init(&ss->hierarchy_mutex);
4456 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4460 mutex_unlock(&cgroup_mutex);
4463 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4466 * cgroup_unload_subsys: unload a modular subsystem
4467 * @ss: the subsystem to unload
4469 * This function should be called in a modular subsystem's exitcall. When this
4470 * function is invoked, the refcount on the subsystem's module will be 0, so
4471 * the subsystem will not be attached to any hierarchy.
4473 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4475 struct cg_cgroup_link *link;
4476 struct hlist_head *hhead;
4478 BUG_ON(ss->module == NULL);
4481 * we shouldn't be called if the subsystem is in use, and the use of
4482 * try_module_get in parse_cgroupfs_options should ensure that it
4483 * doesn't start being used while we're killing it off.
4485 BUG_ON(ss->root != &rootnode);
4487 mutex_lock(&cgroup_mutex);
4488 /* deassign the subsys_id */
4489 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4490 subsys[ss->subsys_id] = NULL;
4492 /* remove subsystem from rootnode's list of subsystems */
4493 list_del_init(&ss->sibling);
4496 * disentangle the css from all css_sets attached to the dummytop. as
4497 * in loading, we need to pay our respects to the hashtable gods.
4499 write_lock(&css_set_lock);
4500 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4501 struct css_set *cg = link->cg;
4503 hlist_del(&cg->hlist);
4504 BUG_ON(!cg->subsys[ss->subsys_id]);
4505 cg->subsys[ss->subsys_id] = NULL;
4506 hhead = css_set_hash(cg->subsys);
4507 hlist_add_head(&cg->hlist, hhead);
4509 write_unlock(&css_set_lock);
4512 * remove subsystem's css from the dummytop and free it - need to free
4513 * before marking as null because ss->destroy needs the cgrp->subsys
4514 * pointer to find their state. note that this also takes care of
4515 * freeing the css_id.
4517 ss->destroy(dummytop);
4518 dummytop->subsys[ss->subsys_id] = NULL;
4520 mutex_unlock(&cgroup_mutex);
4522 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4525 * cgroup_init_early - cgroup initialization at system boot
4527 * Initialize cgroups at system boot, and initialize any
4528 * subsystems that request early init.
4530 int __init cgroup_init_early(void)
4533 atomic_set(&init_css_set.refcount, 1);
4534 INIT_LIST_HEAD(&init_css_set.cg_links);
4535 INIT_LIST_HEAD(&init_css_set.tasks);
4536 INIT_HLIST_NODE(&init_css_set.hlist);
4538 init_cgroup_root(&rootnode);
4540 init_task.cgroups = &init_css_set;
4542 init_css_set_link.cg = &init_css_set;
4543 init_css_set_link.cgrp = dummytop;
4544 list_add(&init_css_set_link.cgrp_link_list,
4545 &rootnode.top_cgroup.css_sets);
4546 list_add(&init_css_set_link.cg_link_list,
4547 &init_css_set.cg_links);
4549 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4550 INIT_HLIST_HEAD(&css_set_table[i]);
4552 /* at bootup time, we don't worry about modular subsystems */
4553 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4554 struct cgroup_subsys *ss = subsys[i];
4557 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4558 BUG_ON(!ss->create);
4559 BUG_ON(!ss->destroy);
4560 if (ss->subsys_id != i) {
4561 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4562 ss->name, ss->subsys_id);
4567 cgroup_init_subsys(ss);
4573 * cgroup_init - cgroup initialization
4575 * Register cgroup filesystem and /proc file, and initialize
4576 * any subsystems that didn't request early init.
4578 int __init cgroup_init(void)
4582 struct hlist_head *hhead;
4584 err = bdi_init(&cgroup_backing_dev_info);
4588 /* at bootup time, we don't worry about modular subsystems */
4589 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4590 struct cgroup_subsys *ss = subsys[i];
4591 if (!ss->early_init)
4592 cgroup_init_subsys(ss);
4594 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4597 /* Add init_css_set to the hash table */
4598 hhead = css_set_hash(init_css_set.subsys);
4599 hlist_add_head(&init_css_set.hlist, hhead);
4600 BUG_ON(!init_root_id(&rootnode));
4602 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4608 err = register_filesystem(&cgroup_fs_type);
4610 kobject_put(cgroup_kobj);
4614 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4618 bdi_destroy(&cgroup_backing_dev_info);
4624 * proc_cgroup_show()
4625 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4626 * - Used for /proc/<pid>/cgroup.
4627 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4628 * doesn't really matter if tsk->cgroup changes after we read it,
4629 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4630 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4631 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4632 * cgroup to top_cgroup.
4635 /* TODO: Use a proper seq_file iterator */
4636 static int proc_cgroup_show(struct seq_file *m, void *v)
4639 struct task_struct *tsk;
4642 struct cgroupfs_root *root;
4645 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4651 tsk = get_pid_task(pid, PIDTYPE_PID);
4657 mutex_lock(&cgroup_mutex);
4659 for_each_active_root(root) {
4660 struct cgroup_subsys *ss;
4661 struct cgroup *cgrp;
4664 seq_printf(m, "%d:", root->hierarchy_id);
4665 for_each_subsys(root, ss)
4666 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4667 if (strlen(root->name))
4668 seq_printf(m, "%sname=%s", count ? "," : "",
4671 cgrp = task_cgroup_from_root(tsk, root);
4672 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4680 mutex_unlock(&cgroup_mutex);
4681 put_task_struct(tsk);
4688 static int cgroup_open(struct inode *inode, struct file *file)
4690 struct pid *pid = PROC_I(inode)->pid;
4691 return single_open(file, proc_cgroup_show, pid);
4694 const struct file_operations proc_cgroup_operations = {
4695 .open = cgroup_open,
4697 .llseek = seq_lseek,
4698 .release = single_release,
4701 /* Display information about each subsystem and each hierarchy */
4702 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4706 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4708 * ideally we don't want subsystems moving around while we do this.
4709 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4710 * subsys/hierarchy state.
4712 mutex_lock(&cgroup_mutex);
4713 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4714 struct cgroup_subsys *ss = subsys[i];
4717 seq_printf(m, "%s\t%d\t%d\t%d\n",
4718 ss->name, ss->root->hierarchy_id,
4719 ss->root->number_of_cgroups, !ss->disabled);
4721 mutex_unlock(&cgroup_mutex);
4725 static int cgroupstats_open(struct inode *inode, struct file *file)
4727 return single_open(file, proc_cgroupstats_show, NULL);
4730 static const struct file_operations proc_cgroupstats_operations = {
4731 .open = cgroupstats_open,
4733 .llseek = seq_lseek,
4734 .release = single_release,
4738 * cgroup_fork - attach newly forked task to its parents cgroup.
4739 * @child: pointer to task_struct of forking parent process.
4741 * Description: A task inherits its parent's cgroup at fork().
4743 * A pointer to the shared css_set was automatically copied in
4744 * fork.c by dup_task_struct(). However, we ignore that copy, since
4745 * it was not made under the protection of RCU, cgroup_mutex or
4746 * threadgroup_change_begin(), so it might no longer be a valid
4747 * cgroup pointer. cgroup_attach_task() might have already changed
4748 * current->cgroups, allowing the previously referenced cgroup
4749 * group to be removed and freed.
4751 * Outside the pointer validity we also need to process the css_set
4752 * inheritance between threadgoup_change_begin() and
4753 * threadgoup_change_end(), this way there is no leak in any process
4754 * wide migration performed by cgroup_attach_proc() that could otherwise
4755 * miss a thread because it is too early or too late in the fork stage.
4757 * At the point that cgroup_fork() is called, 'current' is the parent
4758 * task, and the passed argument 'child' points to the child task.
4760 void cgroup_fork(struct task_struct *child)
4763 * We don't need to task_lock() current because current->cgroups
4764 * can't be changed concurrently here. The parent obviously hasn't
4765 * exited and called cgroup_exit(), and we are synchronized against
4766 * cgroup migration through threadgroup_change_begin().
4768 child->cgroups = current->cgroups;
4769 get_css_set(child->cgroups);
4770 INIT_LIST_HEAD(&child->cg_list);
4774 * cgroup_fork_callbacks - run fork callbacks
4775 * @child: the new task
4777 * Called on a new task very soon before adding it to the
4778 * tasklist. No need to take any locks since no-one can
4779 * be operating on this task.
4781 void cgroup_fork_callbacks(struct task_struct *child)
4783 if (need_forkexit_callback) {
4786 * forkexit callbacks are only supported for builtin
4787 * subsystems, and the builtin section of the subsys array is
4788 * immutable, so we don't need to lock the subsys array here.
4790 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4791 struct cgroup_subsys *ss = subsys[i];
4799 * cgroup_post_fork - called on a new task after adding it to the task list
4800 * @child: the task in question
4802 * Adds the task to the list running through its css_set if necessary.
4803 * Has to be after the task is visible on the task list in case we race
4804 * with the first call to cgroup_iter_start() - to guarantee that the
4805 * new task ends up on its list.
4807 void cgroup_post_fork(struct task_struct *child)
4810 * use_task_css_set_links is set to 1 before we walk the tasklist
4811 * under the tasklist_lock and we read it here after we added the child
4812 * to the tasklist under the tasklist_lock as well. If the child wasn't
4813 * yet in the tasklist when we walked through it from
4814 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4815 * should be visible now due to the paired locking and barriers implied
4816 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4817 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4820 if (use_task_css_set_links) {
4821 write_lock(&css_set_lock);
4822 if (list_empty(&child->cg_list)) {
4824 * It's safe to use child->cgroups without task_lock()
4825 * here because we are protected through
4826 * threadgroup_change_begin() against concurrent
4827 * css_set change in cgroup_task_migrate(). Also
4828 * the task can't exit at that point until
4829 * wake_up_new_task() is called, so we are protected
4830 * against cgroup_exit() setting child->cgroup to
4833 list_add(&child->cg_list, &child->cgroups->tasks);
4835 write_unlock(&css_set_lock);
4839 * cgroup_exit - detach cgroup from exiting task
4840 * @tsk: pointer to task_struct of exiting process
4841 * @run_callback: run exit callbacks?
4843 * Description: Detach cgroup from @tsk and release it.
4845 * Note that cgroups marked notify_on_release force every task in
4846 * them to take the global cgroup_mutex mutex when exiting.
4847 * This could impact scaling on very large systems. Be reluctant to
4848 * use notify_on_release cgroups where very high task exit scaling
4849 * is required on large systems.
4851 * the_top_cgroup_hack:
4853 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4855 * We call cgroup_exit() while the task is still competent to
4856 * handle notify_on_release(), then leave the task attached to the
4857 * root cgroup in each hierarchy for the remainder of its exit.
4859 * To do this properly, we would increment the reference count on
4860 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4861 * code we would add a second cgroup function call, to drop that
4862 * reference. This would just create an unnecessary hot spot on
4863 * the top_cgroup reference count, to no avail.
4865 * Normally, holding a reference to a cgroup without bumping its
4866 * count is unsafe. The cgroup could go away, or someone could
4867 * attach us to a different cgroup, decrementing the count on
4868 * the first cgroup that we never incremented. But in this case,
4869 * top_cgroup isn't going away, and either task has PF_EXITING set,
4870 * which wards off any cgroup_attach_task() attempts, or task is a failed
4871 * fork, never visible to cgroup_attach_task.
4873 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4879 * Unlink from the css_set task list if necessary.
4880 * Optimistically check cg_list before taking
4883 if (!list_empty(&tsk->cg_list)) {
4884 write_lock(&css_set_lock);
4885 if (!list_empty(&tsk->cg_list))
4886 list_del_init(&tsk->cg_list);
4887 write_unlock(&css_set_lock);
4890 /* Reassign the task to the init_css_set. */
4893 tsk->cgroups = &init_css_set;
4895 if (run_callbacks && need_forkexit_callback) {
4897 * modular subsystems can't use callbacks, so no need to lock
4900 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4901 struct cgroup_subsys *ss = subsys[i];
4903 struct cgroup *old_cgrp =
4904 rcu_dereference_raw(cg->subsys[i])->cgroup;
4905 struct cgroup *cgrp = task_cgroup(tsk, i);
4906 ss->exit(cgrp, old_cgrp, tsk);
4913 put_css_set_taskexit(cg);
4917 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4918 * @cgrp: the cgroup in question
4919 * @task: the task in question
4921 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4924 * If we are sending in dummytop, then presumably we are creating
4925 * the top cgroup in the subsystem.
4927 * Called only by the ns (nsproxy) cgroup.
4929 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4932 struct cgroup *target;
4934 if (cgrp == dummytop)
4937 target = task_cgroup_from_root(task, cgrp->root);
4938 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4939 cgrp = cgrp->parent;
4940 ret = (cgrp == target);
4944 static void check_for_release(struct cgroup *cgrp)
4946 /* All of these checks rely on RCU to keep the cgroup
4947 * structure alive */
4948 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4949 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4950 /* Control Group is currently removeable. If it's not
4951 * already queued for a userspace notification, queue
4953 int need_schedule_work = 0;
4954 raw_spin_lock(&release_list_lock);
4955 if (!cgroup_is_removed(cgrp) &&
4956 list_empty(&cgrp->release_list)) {
4957 list_add(&cgrp->release_list, &release_list);
4958 need_schedule_work = 1;
4960 raw_spin_unlock(&release_list_lock);
4961 if (need_schedule_work)
4962 schedule_work(&release_agent_work);
4966 /* Caller must verify that the css is not for root cgroup */
4967 bool __css_tryget(struct cgroup_subsys_state *css)
4970 int v = css_refcnt(css);
4972 if (atomic_cmpxchg(&css->refcnt, v, v + 1) == v)
4975 } while (!test_bit(CSS_REMOVED, &css->flags));
4979 EXPORT_SYMBOL_GPL(__css_tryget);
4981 /* Caller must verify that the css is not for root cgroup */
4982 void __css_put(struct cgroup_subsys_state *css)
4984 struct cgroup *cgrp = css->cgroup;
4987 atomic_dec(&css->refcnt);
4988 switch (css_refcnt(css)) {
4990 if (notify_on_release(cgrp)) {
4991 set_bit(CGRP_RELEASABLE, &cgrp->flags);
4992 check_for_release(cgrp);
4994 cgroup_wakeup_rmdir_waiter(cgrp);
4997 if (!test_bit(CSS_CLEAR_CSS_REFS, &css->flags))
4998 schedule_work(&css->dput_work);
5003 EXPORT_SYMBOL_GPL(__css_put);
5006 * Notify userspace when a cgroup is released, by running the
5007 * configured release agent with the name of the cgroup (path
5008 * relative to the root of cgroup file system) as the argument.
5010 * Most likely, this user command will try to rmdir this cgroup.
5012 * This races with the possibility that some other task will be
5013 * attached to this cgroup before it is removed, or that some other
5014 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5015 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5016 * unused, and this cgroup will be reprieved from its death sentence,
5017 * to continue to serve a useful existence. Next time it's released,
5018 * we will get notified again, if it still has 'notify_on_release' set.
5020 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5021 * means only wait until the task is successfully execve()'d. The
5022 * separate release agent task is forked by call_usermodehelper(),
5023 * then control in this thread returns here, without waiting for the
5024 * release agent task. We don't bother to wait because the caller of
5025 * this routine has no use for the exit status of the release agent
5026 * task, so no sense holding our caller up for that.
5028 static void cgroup_release_agent(struct work_struct *work)
5030 BUG_ON(work != &release_agent_work);
5031 mutex_lock(&cgroup_mutex);
5032 raw_spin_lock(&release_list_lock);
5033 while (!list_empty(&release_list)) {
5034 char *argv[3], *envp[3];
5036 char *pathbuf = NULL, *agentbuf = NULL;
5037 struct cgroup *cgrp = list_entry(release_list.next,
5040 list_del_init(&cgrp->release_list);
5041 raw_spin_unlock(&release_list_lock);
5042 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5045 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5047 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5052 argv[i++] = agentbuf;
5053 argv[i++] = pathbuf;
5057 /* minimal command environment */
5058 envp[i++] = "HOME=/";
5059 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5062 /* Drop the lock while we invoke the usermode helper,
5063 * since the exec could involve hitting disk and hence
5064 * be a slow process */
5065 mutex_unlock(&cgroup_mutex);
5066 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5067 mutex_lock(&cgroup_mutex);
5071 raw_spin_lock(&release_list_lock);
5073 raw_spin_unlock(&release_list_lock);
5074 mutex_unlock(&cgroup_mutex);
5077 static int __init cgroup_disable(char *str)
5082 while ((token = strsep(&str, ",")) != NULL) {
5086 * cgroup_disable, being at boot time, can't know about module
5087 * subsystems, so we don't worry about them.
5089 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
5090 struct cgroup_subsys *ss = subsys[i];
5092 if (!strcmp(token, ss->name)) {
5094 printk(KERN_INFO "Disabling %s control group"
5095 " subsystem\n", ss->name);
5102 __setup("cgroup_disable=", cgroup_disable);
5105 * Functons for CSS ID.
5109 *To get ID other than 0, this should be called when !cgroup_is_removed().
5111 unsigned short css_id(struct cgroup_subsys_state *css)
5113 struct css_id *cssid;
5116 * This css_id() can return correct value when somone has refcnt
5117 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5118 * it's unchanged until freed.
5120 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5126 EXPORT_SYMBOL_GPL(css_id);
5128 unsigned short css_depth(struct cgroup_subsys_state *css)
5130 struct css_id *cssid;
5132 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5135 return cssid->depth;
5138 EXPORT_SYMBOL_GPL(css_depth);
5141 * css_is_ancestor - test "root" css is an ancestor of "child"
5142 * @child: the css to be tested.
5143 * @root: the css supporsed to be an ancestor of the child.
5145 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5146 * this function reads css->id, the caller must hold rcu_read_lock().
5147 * But, considering usual usage, the csses should be valid objects after test.
5148 * Assuming that the caller will do some action to the child if this returns
5149 * returns true, the caller must take "child";s reference count.
5150 * If "child" is valid object and this returns true, "root" is valid, too.
5153 bool css_is_ancestor(struct cgroup_subsys_state *child,
5154 const struct cgroup_subsys_state *root)
5156 struct css_id *child_id;
5157 struct css_id *root_id;
5159 child_id = rcu_dereference(child->id);
5162 root_id = rcu_dereference(root->id);
5165 if (child_id->depth < root_id->depth)
5167 if (child_id->stack[root_id->depth] != root_id->id)
5172 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5174 struct css_id *id = css->id;
5175 /* When this is called before css_id initialization, id can be NULL */
5179 BUG_ON(!ss->use_id);
5181 rcu_assign_pointer(id->css, NULL);
5182 rcu_assign_pointer(css->id, NULL);
5183 spin_lock(&ss->id_lock);
5184 idr_remove(&ss->idr, id->id);
5185 spin_unlock(&ss->id_lock);
5186 kfree_rcu(id, rcu_head);
5188 EXPORT_SYMBOL_GPL(free_css_id);
5191 * This is called by init or create(). Then, calls to this function are
5192 * always serialized (By cgroup_mutex() at create()).
5195 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5197 struct css_id *newid;
5198 int myid, error, size;
5200 BUG_ON(!ss->use_id);
5202 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5203 newid = kzalloc(size, GFP_KERNEL);
5205 return ERR_PTR(-ENOMEM);
5207 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
5211 spin_lock(&ss->id_lock);
5212 /* Don't use 0. allocates an ID of 1-65535 */
5213 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
5214 spin_unlock(&ss->id_lock);
5216 /* Returns error when there are no free spaces for new ID.*/
5221 if (myid > CSS_ID_MAX)
5225 newid->depth = depth;
5229 spin_lock(&ss->id_lock);
5230 idr_remove(&ss->idr, myid);
5231 spin_unlock(&ss->id_lock);
5234 return ERR_PTR(error);
5238 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5239 struct cgroup_subsys_state *rootcss)
5241 struct css_id *newid;
5243 spin_lock_init(&ss->id_lock);
5246 newid = get_new_cssid(ss, 0);
5248 return PTR_ERR(newid);
5250 newid->stack[0] = newid->id;
5251 newid->css = rootcss;
5252 rootcss->id = newid;
5256 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5257 struct cgroup *child)
5259 int subsys_id, i, depth = 0;
5260 struct cgroup_subsys_state *parent_css, *child_css;
5261 struct css_id *child_id, *parent_id;
5263 subsys_id = ss->subsys_id;
5264 parent_css = parent->subsys[subsys_id];
5265 child_css = child->subsys[subsys_id];
5266 parent_id = parent_css->id;
5267 depth = parent_id->depth + 1;
5269 child_id = get_new_cssid(ss, depth);
5270 if (IS_ERR(child_id))
5271 return PTR_ERR(child_id);
5273 for (i = 0; i < depth; i++)
5274 child_id->stack[i] = parent_id->stack[i];
5275 child_id->stack[depth] = child_id->id;
5277 * child_id->css pointer will be set after this cgroup is available
5278 * see cgroup_populate_dir()
5280 rcu_assign_pointer(child_css->id, child_id);
5286 * css_lookup - lookup css by id
5287 * @ss: cgroup subsys to be looked into.
5290 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5291 * NULL if not. Should be called under rcu_read_lock()
5293 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5295 struct css_id *cssid = NULL;
5297 BUG_ON(!ss->use_id);
5298 cssid = idr_find(&ss->idr, id);
5300 if (unlikely(!cssid))
5303 return rcu_dereference(cssid->css);
5305 EXPORT_SYMBOL_GPL(css_lookup);
5308 * css_get_next - lookup next cgroup under specified hierarchy.
5309 * @ss: pointer to subsystem
5310 * @id: current position of iteration.
5311 * @root: pointer to css. search tree under this.
5312 * @foundid: position of found object.
5314 * Search next css under the specified hierarchy of rootid. Calling under
5315 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5317 struct cgroup_subsys_state *
5318 css_get_next(struct cgroup_subsys *ss, int id,
5319 struct cgroup_subsys_state *root, int *foundid)
5321 struct cgroup_subsys_state *ret = NULL;
5324 int rootid = css_id(root);
5325 int depth = css_depth(root);
5330 BUG_ON(!ss->use_id);
5331 WARN_ON_ONCE(!rcu_read_lock_held());
5333 /* fill start point for scan */
5337 * scan next entry from bitmap(tree), tmpid is updated after
5340 tmp = idr_get_next(&ss->idr, &tmpid);
5343 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5344 ret = rcu_dereference(tmp->css);
5350 /* continue to scan from next id */
5357 * get corresponding css from file open on cgroupfs directory
5359 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5361 struct cgroup *cgrp;
5362 struct inode *inode;
5363 struct cgroup_subsys_state *css;
5365 inode = f->f_dentry->d_inode;
5366 /* check in cgroup filesystem dir */
5367 if (inode->i_op != &cgroup_dir_inode_operations)
5368 return ERR_PTR(-EBADF);
5370 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5371 return ERR_PTR(-EINVAL);
5374 cgrp = __d_cgrp(f->f_dentry);
5375 css = cgrp->subsys[id];
5376 return css ? css : ERR_PTR(-ENOENT);
5379 #ifdef CONFIG_CGROUP_DEBUG
5380 static struct cgroup_subsys_state *debug_create(struct cgroup *cont)
5382 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5385 return ERR_PTR(-ENOMEM);
5390 static void debug_destroy(struct cgroup *cont)
5392 kfree(cont->subsys[debug_subsys_id]);
5395 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5397 return atomic_read(&cont->count);
5400 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5402 return cgroup_task_count(cont);
5405 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5407 return (u64)(unsigned long)current->cgroups;
5410 static u64 current_css_set_refcount_read(struct cgroup *cont,
5416 count = atomic_read(¤t->cgroups->refcount);
5421 static int current_css_set_cg_links_read(struct cgroup *cont,
5423 struct seq_file *seq)
5425 struct cg_cgroup_link *link;
5428 read_lock(&css_set_lock);
5430 cg = rcu_dereference(current->cgroups);
5431 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5432 struct cgroup *c = link->cgrp;
5436 name = c->dentry->d_name.name;
5439 seq_printf(seq, "Root %d group %s\n",
5440 c->root->hierarchy_id, name);
5443 read_unlock(&css_set_lock);
5447 #define MAX_TASKS_SHOWN_PER_CSS 25
5448 static int cgroup_css_links_read(struct cgroup *cont,
5450 struct seq_file *seq)
5452 struct cg_cgroup_link *link;
5454 read_lock(&css_set_lock);
5455 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5456 struct css_set *cg = link->cg;
5457 struct task_struct *task;
5459 seq_printf(seq, "css_set %p\n", cg);
5460 list_for_each_entry(task, &cg->tasks, cg_list) {
5461 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5462 seq_puts(seq, " ...\n");
5465 seq_printf(seq, " task %d\n",
5466 task_pid_vnr(task));
5470 read_unlock(&css_set_lock);
5474 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5476 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5479 static struct cftype debug_files[] = {
5481 .name = "cgroup_refcount",
5482 .read_u64 = cgroup_refcount_read,
5485 .name = "taskcount",
5486 .read_u64 = debug_taskcount_read,
5490 .name = "current_css_set",
5491 .read_u64 = current_css_set_read,
5495 .name = "current_css_set_refcount",
5496 .read_u64 = current_css_set_refcount_read,
5500 .name = "current_css_set_cg_links",
5501 .read_seq_string = current_css_set_cg_links_read,
5505 .name = "cgroup_css_links",
5506 .read_seq_string = cgroup_css_links_read,
5510 .name = "releasable",
5511 .read_u64 = releasable_read,
5517 struct cgroup_subsys debug_subsys = {
5519 .create = debug_create,
5520 .destroy = debug_destroy,
5521 .subsys_id = debug_subsys_id,
5522 .base_cftypes = debug_files,
5524 #endif /* CONFIG_CGROUP_DEBUG */