2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hash.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_proc */
64 #include <linux/atomic.h>
66 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
67 #define CSS_DEACT_BIAS INT_MIN
70 * cgroup_mutex is the master lock. Any modification to cgroup or its
71 * hierarchy must be performed while holding it.
73 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
74 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
75 * release_agent_path and so on. Modifying requires both cgroup_mutex and
76 * cgroup_root_mutex. Readers can acquire either of the two. This is to
77 * break the following locking order cycle.
79 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
80 * B. namespace_sem -> cgroup_mutex
82 * B happens only through cgroup_show_options() and using cgroup_root_mutex
85 static DEFINE_MUTEX(cgroup_mutex);
86 static DEFINE_MUTEX(cgroup_root_mutex);
89 * Generate an array of cgroup subsystem pointers. At boot time, this is
90 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
91 * registered after that. The mutable section of this array is protected by
94 #define SUBSYS(_x) &_x ## _subsys,
95 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
96 #include <linux/cgroup_subsys.h>
99 #define MAX_CGROUP_ROOT_NAMELEN 64
102 * A cgroupfs_root represents the root of a cgroup hierarchy,
103 * and may be associated with a superblock to form an active
106 struct cgroupfs_root {
107 struct super_block *sb;
110 * The bitmask of subsystems intended to be attached to this
113 unsigned long subsys_bits;
115 /* Unique id for this hierarchy. */
118 /* The bitmask of subsystems currently attached to this hierarchy */
119 unsigned long actual_subsys_bits;
121 /* A list running through the attached subsystems */
122 struct list_head subsys_list;
124 /* The root cgroup for this hierarchy */
125 struct cgroup top_cgroup;
127 /* Tracks how many cgroups are currently defined in hierarchy.*/
128 int number_of_cgroups;
130 /* A list running through the active hierarchies */
131 struct list_head root_list;
133 /* All cgroups on this root, cgroup_mutex protected */
134 struct list_head allcg_list;
136 /* Hierarchy-specific flags */
139 /* The path to use for release notifications. */
140 char release_agent_path[PATH_MAX];
142 /* The name for this hierarchy - may be empty */
143 char name[MAX_CGROUP_ROOT_NAMELEN];
147 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
148 * subsystems that are otherwise unattached - it never has more than a
149 * single cgroup, and all tasks are part of that cgroup.
151 static struct cgroupfs_root rootnode;
154 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
157 struct list_head node;
158 struct dentry *dentry;
163 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
164 * cgroup_subsys->use_id != 0.
166 #define CSS_ID_MAX (65535)
169 * The css to which this ID points. This pointer is set to valid value
170 * after cgroup is populated. If cgroup is removed, this will be NULL.
171 * This pointer is expected to be RCU-safe because destroy()
172 * is called after synchronize_rcu(). But for safe use, css_is_removed()
173 * css_tryget() should be used for avoiding race.
175 struct cgroup_subsys_state __rcu *css;
181 * Depth in hierarchy which this ID belongs to.
183 unsigned short depth;
185 * ID is freed by RCU. (and lookup routine is RCU safe.)
187 struct rcu_head rcu_head;
189 * Hierarchy of CSS ID belongs to.
191 unsigned short stack[0]; /* Array of Length (depth+1) */
195 * cgroup_event represents events which userspace want to receive.
197 struct cgroup_event {
199 * Cgroup which the event belongs to.
203 * Control file which the event associated.
207 * eventfd to signal userspace about the event.
209 struct eventfd_ctx *eventfd;
211 * Each of these stored in a list by the cgroup.
213 struct list_head list;
215 * All fields below needed to unregister event when
216 * userspace closes eventfd.
219 wait_queue_head_t *wqh;
221 struct work_struct remove;
224 /* The list of hierarchy roots */
226 static LIST_HEAD(roots);
227 static int root_count;
229 static DEFINE_IDA(hierarchy_ida);
230 static int next_hierarchy_id;
231 static DEFINE_SPINLOCK(hierarchy_id_lock);
233 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
234 #define dummytop (&rootnode.top_cgroup)
236 /* This flag indicates whether tasks in the fork and exit paths should
237 * check for fork/exit handlers to call. This avoids us having to do
238 * extra work in the fork/exit path if none of the subsystems need to
241 static int need_forkexit_callback __read_mostly;
243 #ifdef CONFIG_PROVE_LOCKING
244 int cgroup_lock_is_held(void)
246 return lockdep_is_held(&cgroup_mutex);
248 #else /* #ifdef CONFIG_PROVE_LOCKING */
249 int cgroup_lock_is_held(void)
251 return mutex_is_locked(&cgroup_mutex);
253 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
255 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
257 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
258 static int css_refcnt(struct cgroup_subsys_state *css)
260 int v = atomic_read(&css->refcnt);
262 return v >= 0 ? v : v - CSS_DEACT_BIAS;
265 /* convenient tests for these bits */
266 inline int cgroup_is_removed(const struct cgroup *cgrp)
268 return test_bit(CGRP_REMOVED, &cgrp->flags);
271 /* bits in struct cgroupfs_root flags field */
273 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
276 static int cgroup_is_releasable(const struct cgroup *cgrp)
279 (1 << CGRP_RELEASABLE) |
280 (1 << CGRP_NOTIFY_ON_RELEASE);
281 return (cgrp->flags & bits) == bits;
284 static int notify_on_release(const struct cgroup *cgrp)
286 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
289 static int clone_children(const struct cgroup *cgrp)
291 return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
295 * for_each_subsys() allows you to iterate on each subsystem attached to
296 * an active hierarchy
298 #define for_each_subsys(_root, _ss) \
299 list_for_each_entry(_ss, &_root->subsys_list, sibling)
301 /* for_each_active_root() allows you to iterate across the active hierarchies */
302 #define for_each_active_root(_root) \
303 list_for_each_entry(_root, &roots, root_list)
305 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
307 return dentry->d_fsdata;
310 static inline struct cfent *__d_cfe(struct dentry *dentry)
312 return dentry->d_fsdata;
315 static inline struct cftype *__d_cft(struct dentry *dentry)
317 return __d_cfe(dentry)->type;
320 /* the list of cgroups eligible for automatic release. Protected by
321 * release_list_lock */
322 static LIST_HEAD(release_list);
323 static DEFINE_RAW_SPINLOCK(release_list_lock);
324 static void cgroup_release_agent(struct work_struct *work);
325 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
326 static void check_for_release(struct cgroup *cgrp);
328 /* Link structure for associating css_set objects with cgroups */
329 struct cg_cgroup_link {
331 * List running through cg_cgroup_links associated with a
332 * cgroup, anchored on cgroup->css_sets
334 struct list_head cgrp_link_list;
337 * List running through cg_cgroup_links pointing at a
338 * single css_set object, anchored on css_set->cg_links
340 struct list_head cg_link_list;
344 /* The default css_set - used by init and its children prior to any
345 * hierarchies being mounted. It contains a pointer to the root state
346 * for each subsystem. Also used to anchor the list of css_sets. Not
347 * reference-counted, to improve performance when child cgroups
348 * haven't been created.
351 static struct css_set init_css_set;
352 static struct cg_cgroup_link init_css_set_link;
354 static int cgroup_init_idr(struct cgroup_subsys *ss,
355 struct cgroup_subsys_state *css);
357 /* css_set_lock protects the list of css_set objects, and the
358 * chain of tasks off each css_set. Nests outside task->alloc_lock
359 * due to cgroup_iter_start() */
360 static DEFINE_RWLOCK(css_set_lock);
361 static int css_set_count;
364 * hash table for cgroup groups. This improves the performance to find
365 * an existing css_set. This hash doesn't (currently) take into
366 * account cgroups in empty hierarchies.
368 #define CSS_SET_HASH_BITS 7
369 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
370 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
372 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
376 unsigned long tmp = 0UL;
378 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
379 tmp += (unsigned long)css[i];
380 tmp = (tmp >> 16) ^ tmp;
382 index = hash_long(tmp, CSS_SET_HASH_BITS);
384 return &css_set_table[index];
387 /* We don't maintain the lists running through each css_set to its
388 * task until after the first call to cgroup_iter_start(). This
389 * reduces the fork()/exit() overhead for people who have cgroups
390 * compiled into their kernel but not actually in use */
391 static int use_task_css_set_links __read_mostly;
393 static void __put_css_set(struct css_set *cg, int taskexit)
395 struct cg_cgroup_link *link;
396 struct cg_cgroup_link *saved_link;
398 * Ensure that the refcount doesn't hit zero while any readers
399 * can see it. Similar to atomic_dec_and_lock(), but for an
402 if (atomic_add_unless(&cg->refcount, -1, 1))
404 write_lock(&css_set_lock);
405 if (!atomic_dec_and_test(&cg->refcount)) {
406 write_unlock(&css_set_lock);
410 /* This css_set is dead. unlink it and release cgroup refcounts */
411 hlist_del(&cg->hlist);
414 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
416 struct cgroup *cgrp = link->cgrp;
417 list_del(&link->cg_link_list);
418 list_del(&link->cgrp_link_list);
419 if (atomic_dec_and_test(&cgrp->count) &&
420 notify_on_release(cgrp)) {
422 set_bit(CGRP_RELEASABLE, &cgrp->flags);
423 check_for_release(cgrp);
429 write_unlock(&css_set_lock);
430 kfree_rcu(cg, rcu_head);
434 * refcounted get/put for css_set objects
436 static inline void get_css_set(struct css_set *cg)
438 atomic_inc(&cg->refcount);
441 static inline void put_css_set(struct css_set *cg)
443 __put_css_set(cg, 0);
446 static inline void put_css_set_taskexit(struct css_set *cg)
448 __put_css_set(cg, 1);
452 * compare_css_sets - helper function for find_existing_css_set().
453 * @cg: candidate css_set being tested
454 * @old_cg: existing css_set for a task
455 * @new_cgrp: cgroup that's being entered by the task
456 * @template: desired set of css pointers in css_set (pre-calculated)
458 * Returns true if "cg" matches "old_cg" except for the hierarchy
459 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
461 static bool compare_css_sets(struct css_set *cg,
462 struct css_set *old_cg,
463 struct cgroup *new_cgrp,
464 struct cgroup_subsys_state *template[])
466 struct list_head *l1, *l2;
468 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
469 /* Not all subsystems matched */
474 * Compare cgroup pointers in order to distinguish between
475 * different cgroups in heirarchies with no subsystems. We
476 * could get by with just this check alone (and skip the
477 * memcmp above) but on most setups the memcmp check will
478 * avoid the need for this more expensive check on almost all
483 l2 = &old_cg->cg_links;
485 struct cg_cgroup_link *cgl1, *cgl2;
486 struct cgroup *cg1, *cg2;
490 /* See if we reached the end - both lists are equal length. */
491 if (l1 == &cg->cg_links) {
492 BUG_ON(l2 != &old_cg->cg_links);
495 BUG_ON(l2 == &old_cg->cg_links);
497 /* Locate the cgroups associated with these links. */
498 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
499 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
502 /* Hierarchies should be linked in the same order. */
503 BUG_ON(cg1->root != cg2->root);
506 * If this hierarchy is the hierarchy of the cgroup
507 * that's changing, then we need to check that this
508 * css_set points to the new cgroup; if it's any other
509 * hierarchy, then this css_set should point to the
510 * same cgroup as the old css_set.
512 if (cg1->root == new_cgrp->root) {
524 * find_existing_css_set() is a helper for
525 * find_css_set(), and checks to see whether an existing
526 * css_set is suitable.
528 * oldcg: the cgroup group that we're using before the cgroup
531 * cgrp: the cgroup that we're moving into
533 * template: location in which to build the desired set of subsystem
534 * state objects for the new cgroup group
536 static struct css_set *find_existing_css_set(
537 struct css_set *oldcg,
539 struct cgroup_subsys_state *template[])
542 struct cgroupfs_root *root = cgrp->root;
543 struct hlist_head *hhead;
544 struct hlist_node *node;
548 * Build the set of subsystem state objects that we want to see in the
549 * new css_set. while subsystems can change globally, the entries here
550 * won't change, so no need for locking.
552 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
553 if (root->subsys_bits & (1UL << i)) {
554 /* Subsystem is in this hierarchy. So we want
555 * the subsystem state from the new
557 template[i] = cgrp->subsys[i];
559 /* Subsystem is not in this hierarchy, so we
560 * don't want to change the subsystem state */
561 template[i] = oldcg->subsys[i];
565 hhead = css_set_hash(template);
566 hlist_for_each_entry(cg, node, hhead, hlist) {
567 if (!compare_css_sets(cg, oldcg, cgrp, template))
570 /* This css_set matches what we need */
574 /* No existing cgroup group matched */
578 static void free_cg_links(struct list_head *tmp)
580 struct cg_cgroup_link *link;
581 struct cg_cgroup_link *saved_link;
583 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
584 list_del(&link->cgrp_link_list);
590 * allocate_cg_links() allocates "count" cg_cgroup_link structures
591 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
592 * success or a negative error
594 static int allocate_cg_links(int count, struct list_head *tmp)
596 struct cg_cgroup_link *link;
599 for (i = 0; i < count; i++) {
600 link = kmalloc(sizeof(*link), GFP_KERNEL);
605 list_add(&link->cgrp_link_list, tmp);
611 * link_css_set - a helper function to link a css_set to a cgroup
612 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
613 * @cg: the css_set to be linked
614 * @cgrp: the destination cgroup
616 static void link_css_set(struct list_head *tmp_cg_links,
617 struct css_set *cg, struct cgroup *cgrp)
619 struct cg_cgroup_link *link;
621 BUG_ON(list_empty(tmp_cg_links));
622 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
626 atomic_inc(&cgrp->count);
627 list_move(&link->cgrp_link_list, &cgrp->css_sets);
629 * Always add links to the tail of the list so that the list
630 * is sorted by order of hierarchy creation
632 list_add_tail(&link->cg_link_list, &cg->cg_links);
636 * find_css_set() takes an existing cgroup group and a
637 * cgroup object, and returns a css_set object that's
638 * equivalent to the old group, but with the given cgroup
639 * substituted into the appropriate hierarchy. Must be called with
642 static struct css_set *find_css_set(
643 struct css_set *oldcg, struct cgroup *cgrp)
646 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
648 struct list_head tmp_cg_links;
650 struct hlist_head *hhead;
651 struct cg_cgroup_link *link;
653 /* First see if we already have a cgroup group that matches
655 read_lock(&css_set_lock);
656 res = find_existing_css_set(oldcg, cgrp, template);
659 read_unlock(&css_set_lock);
664 res = kmalloc(sizeof(*res), GFP_KERNEL);
668 /* Allocate all the cg_cgroup_link objects that we'll need */
669 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
674 atomic_set(&res->refcount, 1);
675 INIT_LIST_HEAD(&res->cg_links);
676 INIT_LIST_HEAD(&res->tasks);
677 INIT_HLIST_NODE(&res->hlist);
679 /* Copy the set of subsystem state objects generated in
680 * find_existing_css_set() */
681 memcpy(res->subsys, template, sizeof(res->subsys));
683 write_lock(&css_set_lock);
684 /* Add reference counts and links from the new css_set. */
685 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
686 struct cgroup *c = link->cgrp;
687 if (c->root == cgrp->root)
689 link_css_set(&tmp_cg_links, res, c);
692 BUG_ON(!list_empty(&tmp_cg_links));
696 /* Add this cgroup group to the hash table */
697 hhead = css_set_hash(res->subsys);
698 hlist_add_head(&res->hlist, hhead);
700 write_unlock(&css_set_lock);
706 * Return the cgroup for "task" from the given hierarchy. Must be
707 * called with cgroup_mutex held.
709 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
710 struct cgroupfs_root *root)
713 struct cgroup *res = NULL;
715 BUG_ON(!mutex_is_locked(&cgroup_mutex));
716 read_lock(&css_set_lock);
718 * No need to lock the task - since we hold cgroup_mutex the
719 * task can't change groups, so the only thing that can happen
720 * is that it exits and its css is set back to init_css_set.
723 if (css == &init_css_set) {
724 res = &root->top_cgroup;
726 struct cg_cgroup_link *link;
727 list_for_each_entry(link, &css->cg_links, cg_link_list) {
728 struct cgroup *c = link->cgrp;
729 if (c->root == root) {
735 read_unlock(&css_set_lock);
741 * There is one global cgroup mutex. We also require taking
742 * task_lock() when dereferencing a task's cgroup subsys pointers.
743 * See "The task_lock() exception", at the end of this comment.
745 * A task must hold cgroup_mutex to modify cgroups.
747 * Any task can increment and decrement the count field without lock.
748 * So in general, code holding cgroup_mutex can't rely on the count
749 * field not changing. However, if the count goes to zero, then only
750 * cgroup_attach_task() can increment it again. Because a count of zero
751 * means that no tasks are currently attached, therefore there is no
752 * way a task attached to that cgroup can fork (the other way to
753 * increment the count). So code holding cgroup_mutex can safely
754 * assume that if the count is zero, it will stay zero. Similarly, if
755 * a task holds cgroup_mutex on a cgroup with zero count, it
756 * knows that the cgroup won't be removed, as cgroup_rmdir()
759 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
760 * (usually) take cgroup_mutex. These are the two most performance
761 * critical pieces of code here. The exception occurs on cgroup_exit(),
762 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
763 * is taken, and if the cgroup count is zero, a usermode call made
764 * to the release agent with the name of the cgroup (path relative to
765 * the root of cgroup file system) as the argument.
767 * A cgroup can only be deleted if both its 'count' of using tasks
768 * is zero, and its list of 'children' cgroups is empty. Since all
769 * tasks in the system use _some_ cgroup, and since there is always at
770 * least one task in the system (init, pid == 1), therefore, top_cgroup
771 * always has either children cgroups and/or using tasks. So we don't
772 * need a special hack to ensure that top_cgroup cannot be deleted.
774 * The task_lock() exception
776 * The need for this exception arises from the action of
777 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
778 * another. It does so using cgroup_mutex, however there are
779 * several performance critical places that need to reference
780 * task->cgroup without the expense of grabbing a system global
781 * mutex. Therefore except as noted below, when dereferencing or, as
782 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
783 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
784 * the task_struct routinely used for such matters.
786 * P.S. One more locking exception. RCU is used to guard the
787 * update of a tasks cgroup pointer by cgroup_attach_task()
791 * cgroup_lock - lock out any changes to cgroup structures
794 void cgroup_lock(void)
796 mutex_lock(&cgroup_mutex);
798 EXPORT_SYMBOL_GPL(cgroup_lock);
801 * cgroup_unlock - release lock on cgroup changes
803 * Undo the lock taken in a previous cgroup_lock() call.
805 void cgroup_unlock(void)
807 mutex_unlock(&cgroup_mutex);
809 EXPORT_SYMBOL_GPL(cgroup_unlock);
812 * A couple of forward declarations required, due to cyclic reference loop:
813 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
814 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
818 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
819 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, struct nameidata *);
820 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
821 static int cgroup_populate_dir(struct cgroup *cgrp);
822 static const struct inode_operations cgroup_dir_inode_operations;
823 static const struct file_operations proc_cgroupstats_operations;
825 static struct backing_dev_info cgroup_backing_dev_info = {
827 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
830 static int alloc_css_id(struct cgroup_subsys *ss,
831 struct cgroup *parent, struct cgroup *child);
833 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
835 struct inode *inode = new_inode(sb);
838 inode->i_ino = get_next_ino();
839 inode->i_mode = mode;
840 inode->i_uid = current_fsuid();
841 inode->i_gid = current_fsgid();
842 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
843 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
849 * Call subsys's pre_destroy handler.
850 * This is called before css refcnt check.
852 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
854 struct cgroup_subsys *ss;
857 for_each_subsys(cgrp->root, ss)
858 if (ss->pre_destroy) {
859 ret = ss->pre_destroy(cgrp);
867 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
869 /* is dentry a directory ? if so, kfree() associated cgroup */
870 if (S_ISDIR(inode->i_mode)) {
871 struct cgroup *cgrp = dentry->d_fsdata;
872 struct cgroup_subsys *ss;
873 BUG_ON(!(cgroup_is_removed(cgrp)));
874 /* It's possible for external users to be holding css
875 * reference counts on a cgroup; css_put() needs to
876 * be able to access the cgroup after decrementing
877 * the reference count in order to know if it needs to
878 * queue the cgroup to be handled by the release
882 mutex_lock(&cgroup_mutex);
884 * Release the subsystem state objects.
886 for_each_subsys(cgrp->root, ss)
889 cgrp->root->number_of_cgroups--;
890 mutex_unlock(&cgroup_mutex);
893 * Drop the active superblock reference that we took when we
896 deactivate_super(cgrp->root->sb);
899 * if we're getting rid of the cgroup, refcount should ensure
900 * that there are no pidlists left.
902 BUG_ON(!list_empty(&cgrp->pidlists));
904 kfree_rcu(cgrp, rcu_head);
906 struct cfent *cfe = __d_cfe(dentry);
907 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
909 WARN_ONCE(!list_empty(&cfe->node) &&
910 cgrp != &cgrp->root->top_cgroup,
911 "cfe still linked for %s\n", cfe->type->name);
917 static int cgroup_delete(const struct dentry *d)
922 static void remove_dir(struct dentry *d)
924 struct dentry *parent = dget(d->d_parent);
927 simple_rmdir(parent->d_inode, d);
931 static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
935 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
936 lockdep_assert_held(&cgroup_mutex);
938 list_for_each_entry(cfe, &cgrp->files, node) {
939 struct dentry *d = cfe->dentry;
941 if (cft && cfe->type != cft)
946 simple_unlink(d->d_inode, d);
947 list_del_init(&cfe->node);
955 static void cgroup_clear_directory(struct dentry *dir)
957 struct cgroup *cgrp = __d_cgrp(dir);
959 while (!list_empty(&cgrp->files))
960 cgroup_rm_file(cgrp, NULL);
964 * NOTE : the dentry must have been dget()'ed
966 static void cgroup_d_remove_dir(struct dentry *dentry)
968 struct dentry *parent;
970 cgroup_clear_directory(dentry);
972 parent = dentry->d_parent;
973 spin_lock(&parent->d_lock);
974 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
975 list_del_init(&dentry->d_u.d_child);
976 spin_unlock(&dentry->d_lock);
977 spin_unlock(&parent->d_lock);
982 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
983 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
984 * reference to css->refcnt. In general, this refcnt is expected to goes down
987 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
989 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
991 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
993 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
994 wake_up_all(&cgroup_rmdir_waitq);
997 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
1002 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
1004 cgroup_wakeup_rmdir_waiter(css->cgroup);
1009 * Call with cgroup_mutex held. Drops reference counts on modules, including
1010 * any duplicate ones that parse_cgroupfs_options took. If this function
1011 * returns an error, no reference counts are touched.
1013 static int rebind_subsystems(struct cgroupfs_root *root,
1014 unsigned long final_bits)
1016 unsigned long added_bits, removed_bits;
1017 struct cgroup *cgrp = &root->top_cgroup;
1020 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1021 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1023 removed_bits = root->actual_subsys_bits & ~final_bits;
1024 added_bits = final_bits & ~root->actual_subsys_bits;
1025 /* Check that any added subsystems are currently free */
1026 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1027 unsigned long bit = 1UL << i;
1028 struct cgroup_subsys *ss = subsys[i];
1029 if (!(bit & added_bits))
1032 * Nobody should tell us to do a subsys that doesn't exist:
1033 * parse_cgroupfs_options should catch that case and refcounts
1034 * ensure that subsystems won't disappear once selected.
1037 if (ss->root != &rootnode) {
1038 /* Subsystem isn't free */
1043 /* Currently we don't handle adding/removing subsystems when
1044 * any child cgroups exist. This is theoretically supportable
1045 * but involves complex error handling, so it's being left until
1047 if (root->number_of_cgroups > 1)
1050 /* Process each subsystem */
1051 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1052 struct cgroup_subsys *ss = subsys[i];
1053 unsigned long bit = 1UL << i;
1054 if (bit & added_bits) {
1055 /* We're binding this subsystem to this hierarchy */
1057 BUG_ON(cgrp->subsys[i]);
1058 BUG_ON(!dummytop->subsys[i]);
1059 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1060 mutex_lock(&ss->hierarchy_mutex);
1061 cgrp->subsys[i] = dummytop->subsys[i];
1062 cgrp->subsys[i]->cgroup = cgrp;
1063 list_move(&ss->sibling, &root->subsys_list);
1067 mutex_unlock(&ss->hierarchy_mutex);
1068 /* refcount was already taken, and we're keeping it */
1069 } else if (bit & removed_bits) {
1070 /* We're removing this subsystem */
1072 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1073 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1074 mutex_lock(&ss->hierarchy_mutex);
1077 dummytop->subsys[i]->cgroup = dummytop;
1078 cgrp->subsys[i] = NULL;
1079 subsys[i]->root = &rootnode;
1080 list_move(&ss->sibling, &rootnode.subsys_list);
1081 mutex_unlock(&ss->hierarchy_mutex);
1082 /* subsystem is now free - drop reference on module */
1083 module_put(ss->module);
1084 } else if (bit & final_bits) {
1085 /* Subsystem state should already exist */
1087 BUG_ON(!cgrp->subsys[i]);
1089 * a refcount was taken, but we already had one, so
1090 * drop the extra reference.
1092 module_put(ss->module);
1093 #ifdef CONFIG_MODULE_UNLOAD
1094 BUG_ON(ss->module && !module_refcount(ss->module));
1097 /* Subsystem state shouldn't exist */
1098 BUG_ON(cgrp->subsys[i]);
1101 root->subsys_bits = root->actual_subsys_bits = final_bits;
1107 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1109 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1110 struct cgroup_subsys *ss;
1112 mutex_lock(&cgroup_root_mutex);
1113 for_each_subsys(root, ss)
1114 seq_printf(seq, ",%s", ss->name);
1115 if (test_bit(ROOT_NOPREFIX, &root->flags))
1116 seq_puts(seq, ",noprefix");
1117 if (strlen(root->release_agent_path))
1118 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1119 if (clone_children(&root->top_cgroup))
1120 seq_puts(seq, ",clone_children");
1121 if (strlen(root->name))
1122 seq_printf(seq, ",name=%s", root->name);
1123 mutex_unlock(&cgroup_root_mutex);
1127 struct cgroup_sb_opts {
1128 unsigned long subsys_bits;
1129 unsigned long flags;
1130 char *release_agent;
1131 bool clone_children;
1133 /* User explicitly requested empty subsystem */
1136 struct cgroupfs_root *new_root;
1141 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1142 * with cgroup_mutex held to protect the subsys[] array. This function takes
1143 * refcounts on subsystems to be used, unless it returns error, in which case
1144 * no refcounts are taken.
1146 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1148 char *token, *o = data;
1149 bool all_ss = false, one_ss = false;
1150 unsigned long mask = (unsigned long)-1;
1152 bool module_pin_failed = false;
1154 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1156 #ifdef CONFIG_CPUSETS
1157 mask = ~(1UL << cpuset_subsys_id);
1160 memset(opts, 0, sizeof(*opts));
1162 while ((token = strsep(&o, ",")) != NULL) {
1165 if (!strcmp(token, "none")) {
1166 /* Explicitly have no subsystems */
1170 if (!strcmp(token, "all")) {
1171 /* Mutually exclusive option 'all' + subsystem name */
1177 if (!strcmp(token, "noprefix")) {
1178 set_bit(ROOT_NOPREFIX, &opts->flags);
1181 if (!strcmp(token, "clone_children")) {
1182 opts->clone_children = true;
1185 if (!strncmp(token, "release_agent=", 14)) {
1186 /* Specifying two release agents is forbidden */
1187 if (opts->release_agent)
1189 opts->release_agent =
1190 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1191 if (!opts->release_agent)
1195 if (!strncmp(token, "name=", 5)) {
1196 const char *name = token + 5;
1197 /* Can't specify an empty name */
1200 /* Must match [\w.-]+ */
1201 for (i = 0; i < strlen(name); i++) {
1205 if ((c == '.') || (c == '-') || (c == '_'))
1209 /* Specifying two names is forbidden */
1212 opts->name = kstrndup(name,
1213 MAX_CGROUP_ROOT_NAMELEN - 1,
1221 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1222 struct cgroup_subsys *ss = subsys[i];
1225 if (strcmp(token, ss->name))
1230 /* Mutually exclusive option 'all' + subsystem name */
1233 set_bit(i, &opts->subsys_bits);
1238 if (i == CGROUP_SUBSYS_COUNT)
1243 * If the 'all' option was specified select all the subsystems,
1244 * otherwise if 'none', 'name=' and a subsystem name options
1245 * were not specified, let's default to 'all'
1247 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1248 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1249 struct cgroup_subsys *ss = subsys[i];
1254 set_bit(i, &opts->subsys_bits);
1258 /* Consistency checks */
1261 * Option noprefix was introduced just for backward compatibility
1262 * with the old cpuset, so we allow noprefix only if mounting just
1263 * the cpuset subsystem.
1265 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1266 (opts->subsys_bits & mask))
1270 /* Can't specify "none" and some subsystems */
1271 if (opts->subsys_bits && opts->none)
1275 * We either have to specify by name or by subsystems. (So all
1276 * empty hierarchies must have a name).
1278 if (!opts->subsys_bits && !opts->name)
1282 * Grab references on all the modules we'll need, so the subsystems
1283 * don't dance around before rebind_subsystems attaches them. This may
1284 * take duplicate reference counts on a subsystem that's already used,
1285 * but rebind_subsystems handles this case.
1287 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1288 unsigned long bit = 1UL << i;
1290 if (!(bit & opts->subsys_bits))
1292 if (!try_module_get(subsys[i]->module)) {
1293 module_pin_failed = true;
1297 if (module_pin_failed) {
1299 * oops, one of the modules was going away. this means that we
1300 * raced with a module_delete call, and to the user this is
1301 * essentially a "subsystem doesn't exist" case.
1303 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1304 /* drop refcounts only on the ones we took */
1305 unsigned long bit = 1UL << i;
1307 if (!(bit & opts->subsys_bits))
1309 module_put(subsys[i]->module);
1317 static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1320 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1321 unsigned long bit = 1UL << i;
1323 if (!(bit & subsys_bits))
1325 module_put(subsys[i]->module);
1329 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1332 struct cgroupfs_root *root = sb->s_fs_info;
1333 struct cgroup *cgrp = &root->top_cgroup;
1334 struct cgroup_sb_opts opts;
1336 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1337 mutex_lock(&cgroup_mutex);
1338 mutex_lock(&cgroup_root_mutex);
1340 /* See what subsystems are wanted */
1341 ret = parse_cgroupfs_options(data, &opts);
1345 /* See feature-removal-schedule.txt */
1346 if (opts.subsys_bits != root->actual_subsys_bits || opts.release_agent)
1347 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1348 task_tgid_nr(current), current->comm);
1350 /* Don't allow flags or name to change at remount */
1351 if (opts.flags != root->flags ||
1352 (opts.name && strcmp(opts.name, root->name))) {
1354 drop_parsed_module_refcounts(opts.subsys_bits);
1358 ret = rebind_subsystems(root, opts.subsys_bits);
1360 drop_parsed_module_refcounts(opts.subsys_bits);
1364 /* clear out any existing files and repopulate subsystem files */
1365 cgroup_clear_directory(cgrp->dentry);
1366 cgroup_populate_dir(cgrp);
1368 if (opts.release_agent)
1369 strcpy(root->release_agent_path, opts.release_agent);
1371 kfree(opts.release_agent);
1373 mutex_unlock(&cgroup_root_mutex);
1374 mutex_unlock(&cgroup_mutex);
1375 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1379 static const struct super_operations cgroup_ops = {
1380 .statfs = simple_statfs,
1381 .drop_inode = generic_delete_inode,
1382 .show_options = cgroup_show_options,
1383 .remount_fs = cgroup_remount,
1386 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1388 INIT_LIST_HEAD(&cgrp->sibling);
1389 INIT_LIST_HEAD(&cgrp->children);
1390 INIT_LIST_HEAD(&cgrp->files);
1391 INIT_LIST_HEAD(&cgrp->css_sets);
1392 INIT_LIST_HEAD(&cgrp->release_list);
1393 INIT_LIST_HEAD(&cgrp->pidlists);
1394 mutex_init(&cgrp->pidlist_mutex);
1395 INIT_LIST_HEAD(&cgrp->event_list);
1396 spin_lock_init(&cgrp->event_list_lock);
1399 static void init_cgroup_root(struct cgroupfs_root *root)
1401 struct cgroup *cgrp = &root->top_cgroup;
1403 INIT_LIST_HEAD(&root->subsys_list);
1404 INIT_LIST_HEAD(&root->root_list);
1405 INIT_LIST_HEAD(&root->allcg_list);
1406 root->number_of_cgroups = 1;
1408 cgrp->top_cgroup = cgrp;
1409 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1410 init_cgroup_housekeeping(cgrp);
1413 static bool init_root_id(struct cgroupfs_root *root)
1418 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1420 spin_lock(&hierarchy_id_lock);
1421 /* Try to allocate the next unused ID */
1422 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1423 &root->hierarchy_id);
1425 /* Try again starting from 0 */
1426 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1428 next_hierarchy_id = root->hierarchy_id + 1;
1429 } else if (ret != -EAGAIN) {
1430 /* Can only get here if the 31-bit IDR is full ... */
1433 spin_unlock(&hierarchy_id_lock);
1438 static int cgroup_test_super(struct super_block *sb, void *data)
1440 struct cgroup_sb_opts *opts = data;
1441 struct cgroupfs_root *root = sb->s_fs_info;
1443 /* If we asked for a name then it must match */
1444 if (opts->name && strcmp(opts->name, root->name))
1448 * If we asked for subsystems (or explicitly for no
1449 * subsystems) then they must match
1451 if ((opts->subsys_bits || opts->none)
1452 && (opts->subsys_bits != root->subsys_bits))
1458 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1460 struct cgroupfs_root *root;
1462 if (!opts->subsys_bits && !opts->none)
1465 root = kzalloc(sizeof(*root), GFP_KERNEL);
1467 return ERR_PTR(-ENOMEM);
1469 if (!init_root_id(root)) {
1471 return ERR_PTR(-ENOMEM);
1473 init_cgroup_root(root);
1475 root->subsys_bits = opts->subsys_bits;
1476 root->flags = opts->flags;
1477 if (opts->release_agent)
1478 strcpy(root->release_agent_path, opts->release_agent);
1480 strcpy(root->name, opts->name);
1481 if (opts->clone_children)
1482 set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1486 static void cgroup_drop_root(struct cgroupfs_root *root)
1491 BUG_ON(!root->hierarchy_id);
1492 spin_lock(&hierarchy_id_lock);
1493 ida_remove(&hierarchy_ida, root->hierarchy_id);
1494 spin_unlock(&hierarchy_id_lock);
1498 static int cgroup_set_super(struct super_block *sb, void *data)
1501 struct cgroup_sb_opts *opts = data;
1503 /* If we don't have a new root, we can't set up a new sb */
1504 if (!opts->new_root)
1507 BUG_ON(!opts->subsys_bits && !opts->none);
1509 ret = set_anon_super(sb, NULL);
1513 sb->s_fs_info = opts->new_root;
1514 opts->new_root->sb = sb;
1516 sb->s_blocksize = PAGE_CACHE_SIZE;
1517 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1518 sb->s_magic = CGROUP_SUPER_MAGIC;
1519 sb->s_op = &cgroup_ops;
1524 static int cgroup_get_rootdir(struct super_block *sb)
1526 static const struct dentry_operations cgroup_dops = {
1527 .d_iput = cgroup_diput,
1528 .d_delete = cgroup_delete,
1531 struct inode *inode =
1532 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1537 inode->i_fop = &simple_dir_operations;
1538 inode->i_op = &cgroup_dir_inode_operations;
1539 /* directories start off with i_nlink == 2 (for "." entry) */
1541 sb->s_root = d_make_root(inode);
1544 /* for everything else we want ->d_op set */
1545 sb->s_d_op = &cgroup_dops;
1549 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1550 int flags, const char *unused_dev_name,
1553 struct cgroup_sb_opts opts;
1554 struct cgroupfs_root *root;
1556 struct super_block *sb;
1557 struct cgroupfs_root *new_root;
1558 struct inode *inode;
1560 /* First find the desired set of subsystems */
1561 mutex_lock(&cgroup_mutex);
1562 ret = parse_cgroupfs_options(data, &opts);
1563 mutex_unlock(&cgroup_mutex);
1568 * Allocate a new cgroup root. We may not need it if we're
1569 * reusing an existing hierarchy.
1571 new_root = cgroup_root_from_opts(&opts);
1572 if (IS_ERR(new_root)) {
1573 ret = PTR_ERR(new_root);
1576 opts.new_root = new_root;
1578 /* Locate an existing or new sb for this hierarchy */
1579 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1582 cgroup_drop_root(opts.new_root);
1586 root = sb->s_fs_info;
1588 if (root == opts.new_root) {
1589 /* We used the new root structure, so this is a new hierarchy */
1590 struct list_head tmp_cg_links;
1591 struct cgroup *root_cgrp = &root->top_cgroup;
1592 struct cgroupfs_root *existing_root;
1593 const struct cred *cred;
1596 BUG_ON(sb->s_root != NULL);
1598 ret = cgroup_get_rootdir(sb);
1600 goto drop_new_super;
1601 inode = sb->s_root->d_inode;
1603 mutex_lock(&inode->i_mutex);
1604 mutex_lock(&cgroup_mutex);
1605 mutex_lock(&cgroup_root_mutex);
1607 /* Check for name clashes with existing mounts */
1609 if (strlen(root->name))
1610 for_each_active_root(existing_root)
1611 if (!strcmp(existing_root->name, root->name))
1615 * We're accessing css_set_count without locking
1616 * css_set_lock here, but that's OK - it can only be
1617 * increased by someone holding cgroup_lock, and
1618 * that's us. The worst that can happen is that we
1619 * have some link structures left over
1621 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1625 ret = rebind_subsystems(root, root->subsys_bits);
1626 if (ret == -EBUSY) {
1627 free_cg_links(&tmp_cg_links);
1631 * There must be no failure case after here, since rebinding
1632 * takes care of subsystems' refcounts, which are explicitly
1633 * dropped in the failure exit path.
1636 /* EBUSY should be the only error here */
1639 list_add(&root->root_list, &roots);
1642 sb->s_root->d_fsdata = root_cgrp;
1643 root->top_cgroup.dentry = sb->s_root;
1645 /* Link the top cgroup in this hierarchy into all
1646 * the css_set objects */
1647 write_lock(&css_set_lock);
1648 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1649 struct hlist_head *hhead = &css_set_table[i];
1650 struct hlist_node *node;
1653 hlist_for_each_entry(cg, node, hhead, hlist)
1654 link_css_set(&tmp_cg_links, cg, root_cgrp);
1656 write_unlock(&css_set_lock);
1658 free_cg_links(&tmp_cg_links);
1660 BUG_ON(!list_empty(&root_cgrp->sibling));
1661 BUG_ON(!list_empty(&root_cgrp->children));
1662 BUG_ON(root->number_of_cgroups != 1);
1664 cred = override_creds(&init_cred);
1665 cgroup_populate_dir(root_cgrp);
1667 mutex_unlock(&cgroup_root_mutex);
1668 mutex_unlock(&cgroup_mutex);
1669 mutex_unlock(&inode->i_mutex);
1672 * We re-used an existing hierarchy - the new root (if
1673 * any) is not needed
1675 cgroup_drop_root(opts.new_root);
1676 /* no subsys rebinding, so refcounts don't change */
1677 drop_parsed_module_refcounts(opts.subsys_bits);
1680 kfree(opts.release_agent);
1682 return dget(sb->s_root);
1685 mutex_unlock(&cgroup_root_mutex);
1686 mutex_unlock(&cgroup_mutex);
1687 mutex_unlock(&inode->i_mutex);
1689 deactivate_locked_super(sb);
1691 drop_parsed_module_refcounts(opts.subsys_bits);
1693 kfree(opts.release_agent);
1695 return ERR_PTR(ret);
1698 static void cgroup_kill_sb(struct super_block *sb) {
1699 struct cgroupfs_root *root = sb->s_fs_info;
1700 struct cgroup *cgrp = &root->top_cgroup;
1702 struct cg_cgroup_link *link;
1703 struct cg_cgroup_link *saved_link;
1707 BUG_ON(root->number_of_cgroups != 1);
1708 BUG_ON(!list_empty(&cgrp->children));
1709 BUG_ON(!list_empty(&cgrp->sibling));
1711 mutex_lock(&cgroup_mutex);
1712 mutex_lock(&cgroup_root_mutex);
1714 /* Rebind all subsystems back to the default hierarchy */
1715 ret = rebind_subsystems(root, 0);
1716 /* Shouldn't be able to fail ... */
1720 * Release all the links from css_sets to this hierarchy's
1723 write_lock(&css_set_lock);
1725 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1727 list_del(&link->cg_link_list);
1728 list_del(&link->cgrp_link_list);
1731 write_unlock(&css_set_lock);
1733 if (!list_empty(&root->root_list)) {
1734 list_del(&root->root_list);
1738 mutex_unlock(&cgroup_root_mutex);
1739 mutex_unlock(&cgroup_mutex);
1741 kill_litter_super(sb);
1742 cgroup_drop_root(root);
1745 static struct file_system_type cgroup_fs_type = {
1747 .mount = cgroup_mount,
1748 .kill_sb = cgroup_kill_sb,
1751 static struct kobject *cgroup_kobj;
1754 * cgroup_path - generate the path of a cgroup
1755 * @cgrp: the cgroup in question
1756 * @buf: the buffer to write the path into
1757 * @buflen: the length of the buffer
1759 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1760 * reference. Writes path of cgroup into buf. Returns 0 on success,
1763 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1766 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1767 cgroup_lock_is_held());
1769 if (!dentry || cgrp == dummytop) {
1771 * Inactive subsystems have no dentry for their root
1778 start = buf + buflen;
1782 int len = dentry->d_name.len;
1784 if ((start -= len) < buf)
1785 return -ENAMETOOLONG;
1786 memcpy(start, dentry->d_name.name, len);
1787 cgrp = cgrp->parent;
1791 dentry = rcu_dereference_check(cgrp->dentry,
1792 cgroup_lock_is_held());
1796 return -ENAMETOOLONG;
1799 memmove(buf, start, buf + buflen - start);
1802 EXPORT_SYMBOL_GPL(cgroup_path);
1805 * Control Group taskset
1807 struct task_and_cgroup {
1808 struct task_struct *task;
1809 struct cgroup *cgrp;
1813 struct cgroup_taskset {
1814 struct task_and_cgroup single;
1815 struct flex_array *tc_array;
1818 struct cgroup *cur_cgrp;
1822 * cgroup_taskset_first - reset taskset and return the first task
1823 * @tset: taskset of interest
1825 * @tset iteration is initialized and the first task is returned.
1827 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1829 if (tset->tc_array) {
1831 return cgroup_taskset_next(tset);
1833 tset->cur_cgrp = tset->single.cgrp;
1834 return tset->single.task;
1837 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1840 * cgroup_taskset_next - iterate to the next task in taskset
1841 * @tset: taskset of interest
1843 * Return the next task in @tset. Iteration must have been initialized
1844 * with cgroup_taskset_first().
1846 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1848 struct task_and_cgroup *tc;
1850 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1853 tc = flex_array_get(tset->tc_array, tset->idx++);
1854 tset->cur_cgrp = tc->cgrp;
1857 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1860 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1861 * @tset: taskset of interest
1863 * Return the cgroup for the current (last returned) task of @tset. This
1864 * function must be preceded by either cgroup_taskset_first() or
1865 * cgroup_taskset_next().
1867 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1869 return tset->cur_cgrp;
1871 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1874 * cgroup_taskset_size - return the number of tasks in taskset
1875 * @tset: taskset of interest
1877 int cgroup_taskset_size(struct cgroup_taskset *tset)
1879 return tset->tc_array ? tset->tc_array_len : 1;
1881 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1885 * cgroup_task_migrate - move a task from one cgroup to another.
1887 * 'guarantee' is set if the caller promises that a new css_set for the task
1888 * will already exist. If not set, this function might sleep, and can fail with
1889 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1891 static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1892 struct task_struct *tsk, struct css_set *newcg)
1894 struct css_set *oldcg;
1897 * We are synchronized through threadgroup_lock() against PF_EXITING
1898 * setting such that we can't race against cgroup_exit() changing the
1899 * css_set to init_css_set and dropping the old one.
1901 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1902 oldcg = tsk->cgroups;
1905 rcu_assign_pointer(tsk->cgroups, newcg);
1908 /* Update the css_set linked lists if we're using them */
1909 write_lock(&css_set_lock);
1910 if (!list_empty(&tsk->cg_list))
1911 list_move(&tsk->cg_list, &newcg->tasks);
1912 write_unlock(&css_set_lock);
1915 * We just gained a reference on oldcg by taking it from the task. As
1916 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1917 * it here; it will be freed under RCU.
1921 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1925 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1926 * @cgrp: the cgroup the task is attaching to
1927 * @tsk: the task to be attached
1929 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1932 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1935 struct cgroup_subsys *ss, *failed_ss = NULL;
1936 struct cgroup *oldcgrp;
1937 struct cgroupfs_root *root = cgrp->root;
1938 struct cgroup_taskset tset = { };
1939 struct css_set *newcg;
1941 /* @tsk either already exited or can't exit until the end */
1942 if (tsk->flags & PF_EXITING)
1945 /* Nothing to do if the task is already in that cgroup */
1946 oldcgrp = task_cgroup_from_root(tsk, root);
1947 if (cgrp == oldcgrp)
1950 tset.single.task = tsk;
1951 tset.single.cgrp = oldcgrp;
1953 for_each_subsys(root, ss) {
1954 if (ss->can_attach) {
1955 retval = ss->can_attach(cgrp, &tset);
1958 * Remember on which subsystem the can_attach()
1959 * failed, so that we only call cancel_attach()
1960 * against the subsystems whose can_attach()
1961 * succeeded. (See below)
1969 newcg = find_css_set(tsk->cgroups, cgrp);
1975 cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1977 for_each_subsys(root, ss) {
1979 ss->attach(cgrp, &tset);
1985 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1986 * is no longer empty.
1988 cgroup_wakeup_rmdir_waiter(cgrp);
1991 for_each_subsys(root, ss) {
1992 if (ss == failed_ss)
1994 * This subsystem was the one that failed the
1995 * can_attach() check earlier, so we don't need
1996 * to call cancel_attach() against it or any
1997 * remaining subsystems.
2000 if (ss->cancel_attach)
2001 ss->cancel_attach(cgrp, &tset);
2008 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2009 * @from: attach to all cgroups of a given task
2010 * @tsk: the task to be attached
2012 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2014 struct cgroupfs_root *root;
2018 for_each_active_root(root) {
2019 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2021 retval = cgroup_attach_task(from_cg, tsk);
2029 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2032 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2033 * @cgrp: the cgroup to attach to
2034 * @leader: the threadgroup leader task_struct of the group to be attached
2036 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2037 * task_lock of each thread in leader's threadgroup individually in turn.
2039 static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2041 int retval, i, group_size;
2042 struct cgroup_subsys *ss, *failed_ss = NULL;
2043 /* guaranteed to be initialized later, but the compiler needs this */
2044 struct cgroupfs_root *root = cgrp->root;
2045 /* threadgroup list cursor and array */
2046 struct task_struct *tsk;
2047 struct task_and_cgroup *tc;
2048 struct flex_array *group;
2049 struct cgroup_taskset tset = { };
2052 * step 0: in order to do expensive, possibly blocking operations for
2053 * every thread, we cannot iterate the thread group list, since it needs
2054 * rcu or tasklist locked. instead, build an array of all threads in the
2055 * group - group_rwsem prevents new threads from appearing, and if
2056 * threads exit, this will just be an over-estimate.
2058 group_size = get_nr_threads(leader);
2059 /* flex_array supports very large thread-groups better than kmalloc. */
2060 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2063 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2064 retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2066 goto out_free_group_list;
2071 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2072 * already PF_EXITING could be freed from underneath us unless we
2073 * take an rcu_read_lock.
2077 struct task_and_cgroup ent;
2079 /* @tsk either already exited or can't exit until the end */
2080 if (tsk->flags & PF_EXITING)
2083 /* as per above, nr_threads may decrease, but not increase. */
2084 BUG_ON(i >= group_size);
2086 ent.cgrp = task_cgroup_from_root(tsk, root);
2087 /* nothing to do if this task is already in the cgroup */
2088 if (ent.cgrp == cgrp)
2091 * saying GFP_ATOMIC has no effect here because we did prealloc
2092 * earlier, but it's good form to communicate our expectations.
2094 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2095 BUG_ON(retval != 0);
2097 } while_each_thread(leader, tsk);
2099 /* remember the number of threads in the array for later. */
2101 tset.tc_array = group;
2102 tset.tc_array_len = group_size;
2104 /* methods shouldn't be called if no task is actually migrating */
2107 goto out_free_group_list;
2110 * step 1: check that we can legitimately attach to the cgroup.
2112 for_each_subsys(root, ss) {
2113 if (ss->can_attach) {
2114 retval = ss->can_attach(cgrp, &tset);
2117 goto out_cancel_attach;
2123 * step 2: make sure css_sets exist for all threads to be migrated.
2124 * we use find_css_set, which allocates a new one if necessary.
2126 for (i = 0; i < group_size; i++) {
2127 tc = flex_array_get(group, i);
2128 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2131 goto out_put_css_set_refs;
2136 * step 3: now that we're guaranteed success wrt the css_sets,
2137 * proceed to move all tasks to the new cgroup. There are no
2138 * failure cases after here, so this is the commit point.
2140 for (i = 0; i < group_size; i++) {
2141 tc = flex_array_get(group, i);
2142 cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2144 /* nothing is sensitive to fork() after this point. */
2147 * step 4: do subsystem attach callbacks.
2149 for_each_subsys(root, ss) {
2151 ss->attach(cgrp, &tset);
2155 * step 5: success! and cleanup
2158 cgroup_wakeup_rmdir_waiter(cgrp);
2160 out_put_css_set_refs:
2162 for (i = 0; i < group_size; i++) {
2163 tc = flex_array_get(group, i);
2166 put_css_set(tc->cg);
2171 for_each_subsys(root, ss) {
2172 if (ss == failed_ss)
2174 if (ss->cancel_attach)
2175 ss->cancel_attach(cgrp, &tset);
2178 out_free_group_list:
2179 flex_array_free(group);
2184 * Find the task_struct of the task to attach by vpid and pass it along to the
2185 * function to attach either it or all tasks in its threadgroup. Will lock
2186 * cgroup_mutex and threadgroup; may take task_lock of task.
2188 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2190 struct task_struct *tsk;
2191 const struct cred *cred = current_cred(), *tcred;
2194 if (!cgroup_lock_live_group(cgrp))
2200 tsk = find_task_by_vpid(pid);
2204 goto out_unlock_cgroup;
2207 * even if we're attaching all tasks in the thread group, we
2208 * only need to check permissions on one of them.
2210 tcred = __task_cred(tsk);
2212 cred->euid != tcred->uid &&
2213 cred->euid != tcred->suid) {
2216 goto out_unlock_cgroup;
2222 tsk = tsk->group_leader;
2223 get_task_struct(tsk);
2226 threadgroup_lock(tsk);
2228 if (!thread_group_leader(tsk)) {
2230 * a race with de_thread from another thread's exec()
2231 * may strip us of our leadership, if this happens,
2232 * there is no choice but to throw this task away and
2233 * try again; this is
2234 * "double-double-toil-and-trouble-check locking".
2236 threadgroup_unlock(tsk);
2237 put_task_struct(tsk);
2238 goto retry_find_task;
2240 ret = cgroup_attach_proc(cgrp, tsk);
2242 ret = cgroup_attach_task(cgrp, tsk);
2243 threadgroup_unlock(tsk);
2245 put_task_struct(tsk);
2251 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2253 return attach_task_by_pid(cgrp, pid, false);
2256 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2258 return attach_task_by_pid(cgrp, tgid, true);
2262 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2263 * @cgrp: the cgroup to be checked for liveness
2265 * On success, returns true; the lock should be later released with
2266 * cgroup_unlock(). On failure returns false with no lock held.
2268 bool cgroup_lock_live_group(struct cgroup *cgrp)
2270 mutex_lock(&cgroup_mutex);
2271 if (cgroup_is_removed(cgrp)) {
2272 mutex_unlock(&cgroup_mutex);
2277 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2279 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2282 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2283 if (strlen(buffer) >= PATH_MAX)
2285 if (!cgroup_lock_live_group(cgrp))
2287 mutex_lock(&cgroup_root_mutex);
2288 strcpy(cgrp->root->release_agent_path, buffer);
2289 mutex_unlock(&cgroup_root_mutex);
2294 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2295 struct seq_file *seq)
2297 if (!cgroup_lock_live_group(cgrp))
2299 seq_puts(seq, cgrp->root->release_agent_path);
2300 seq_putc(seq, '\n');
2305 /* A buffer size big enough for numbers or short strings */
2306 #define CGROUP_LOCAL_BUFFER_SIZE 64
2308 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2310 const char __user *userbuf,
2311 size_t nbytes, loff_t *unused_ppos)
2313 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2319 if (nbytes >= sizeof(buffer))
2321 if (copy_from_user(buffer, userbuf, nbytes))
2324 buffer[nbytes] = 0; /* nul-terminate */
2325 if (cft->write_u64) {
2326 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2329 retval = cft->write_u64(cgrp, cft, val);
2331 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2334 retval = cft->write_s64(cgrp, cft, val);
2341 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2343 const char __user *userbuf,
2344 size_t nbytes, loff_t *unused_ppos)
2346 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2348 size_t max_bytes = cft->max_write_len;
2349 char *buffer = local_buffer;
2352 max_bytes = sizeof(local_buffer) - 1;
2353 if (nbytes >= max_bytes)
2355 /* Allocate a dynamic buffer if we need one */
2356 if (nbytes >= sizeof(local_buffer)) {
2357 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2361 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2366 buffer[nbytes] = 0; /* nul-terminate */
2367 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2371 if (buffer != local_buffer)
2376 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2377 size_t nbytes, loff_t *ppos)
2379 struct cftype *cft = __d_cft(file->f_dentry);
2380 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2382 if (cgroup_is_removed(cgrp))
2385 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2386 if (cft->write_u64 || cft->write_s64)
2387 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2388 if (cft->write_string)
2389 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2391 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2392 return ret ? ret : nbytes;
2397 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2399 char __user *buf, size_t nbytes,
2402 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2403 u64 val = cft->read_u64(cgrp, cft);
2404 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2406 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2409 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2411 char __user *buf, size_t nbytes,
2414 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2415 s64 val = cft->read_s64(cgrp, cft);
2416 int len = sprintf(tmp, "%lld\n", (long long) val);
2418 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2421 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2422 size_t nbytes, loff_t *ppos)
2424 struct cftype *cft = __d_cft(file->f_dentry);
2425 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2427 if (cgroup_is_removed(cgrp))
2431 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2433 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2435 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2440 * seqfile ops/methods for returning structured data. Currently just
2441 * supports string->u64 maps, but can be extended in future.
2444 struct cgroup_seqfile_state {
2446 struct cgroup *cgroup;
2449 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2451 struct seq_file *sf = cb->state;
2452 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2455 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2457 struct cgroup_seqfile_state *state = m->private;
2458 struct cftype *cft = state->cft;
2459 if (cft->read_map) {
2460 struct cgroup_map_cb cb = {
2461 .fill = cgroup_map_add,
2464 return cft->read_map(state->cgroup, cft, &cb);
2466 return cft->read_seq_string(state->cgroup, cft, m);
2469 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2471 struct seq_file *seq = file->private_data;
2472 kfree(seq->private);
2473 return single_release(inode, file);
2476 static const struct file_operations cgroup_seqfile_operations = {
2478 .write = cgroup_file_write,
2479 .llseek = seq_lseek,
2480 .release = cgroup_seqfile_release,
2483 static int cgroup_file_open(struct inode *inode, struct file *file)
2488 err = generic_file_open(inode, file);
2491 cft = __d_cft(file->f_dentry);
2493 if (cft->read_map || cft->read_seq_string) {
2494 struct cgroup_seqfile_state *state =
2495 kzalloc(sizeof(*state), GFP_USER);
2499 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2500 file->f_op = &cgroup_seqfile_operations;
2501 err = single_open(file, cgroup_seqfile_show, state);
2504 } else if (cft->open)
2505 err = cft->open(inode, file);
2512 static int cgroup_file_release(struct inode *inode, struct file *file)
2514 struct cftype *cft = __d_cft(file->f_dentry);
2516 return cft->release(inode, file);
2521 * cgroup_rename - Only allow simple rename of directories in place.
2523 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2524 struct inode *new_dir, struct dentry *new_dentry)
2526 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2528 if (new_dentry->d_inode)
2530 if (old_dir != new_dir)
2532 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2535 static const struct file_operations cgroup_file_operations = {
2536 .read = cgroup_file_read,
2537 .write = cgroup_file_write,
2538 .llseek = generic_file_llseek,
2539 .open = cgroup_file_open,
2540 .release = cgroup_file_release,
2543 static const struct inode_operations cgroup_dir_inode_operations = {
2544 .lookup = cgroup_lookup,
2545 .mkdir = cgroup_mkdir,
2546 .rmdir = cgroup_rmdir,
2547 .rename = cgroup_rename,
2550 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
2552 if (dentry->d_name.len > NAME_MAX)
2553 return ERR_PTR(-ENAMETOOLONG);
2554 d_add(dentry, NULL);
2559 * Check if a file is a control file
2561 static inline struct cftype *__file_cft(struct file *file)
2563 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2564 return ERR_PTR(-EINVAL);
2565 return __d_cft(file->f_dentry);
2568 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2569 struct super_block *sb)
2571 struct inode *inode;
2575 if (dentry->d_inode)
2578 inode = cgroup_new_inode(mode, sb);
2582 if (S_ISDIR(mode)) {
2583 inode->i_op = &cgroup_dir_inode_operations;
2584 inode->i_fop = &simple_dir_operations;
2586 /* start off with i_nlink == 2 (for "." entry) */
2589 /* start with the directory inode held, so that we can
2590 * populate it without racing with another mkdir */
2591 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2592 } else if (S_ISREG(mode)) {
2594 inode->i_fop = &cgroup_file_operations;
2596 d_instantiate(dentry, inode);
2597 dget(dentry); /* Extra count - pin the dentry in core */
2602 * cgroup_create_dir - create a directory for an object.
2603 * @cgrp: the cgroup we create the directory for. It must have a valid
2604 * ->parent field. And we are going to fill its ->dentry field.
2605 * @dentry: dentry of the new cgroup
2606 * @mode: mode to set on new directory.
2608 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2611 struct dentry *parent;
2614 parent = cgrp->parent->dentry;
2615 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2617 dentry->d_fsdata = cgrp;
2618 inc_nlink(parent->d_inode);
2619 rcu_assign_pointer(cgrp->dentry, dentry);
2628 * cgroup_file_mode - deduce file mode of a control file
2629 * @cft: the control file in question
2631 * returns cft->mode if ->mode is not 0
2632 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2633 * returns S_IRUGO if it has only a read handler
2634 * returns S_IWUSR if it has only a write hander
2636 static umode_t cgroup_file_mode(const struct cftype *cft)
2643 if (cft->read || cft->read_u64 || cft->read_s64 ||
2644 cft->read_map || cft->read_seq_string)
2647 if (cft->write || cft->write_u64 || cft->write_s64 ||
2648 cft->write_string || cft->trigger)
2654 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2655 const struct cftype *cft)
2657 struct dentry *dir = cgrp->dentry;
2658 struct cgroup *parent = __d_cgrp(dir);
2659 struct dentry *dentry;
2663 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2665 /* does @cft->flags tell us to skip creation on @cgrp? */
2666 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2668 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2671 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2672 strcpy(name, subsys->name);
2675 strcat(name, cft->name);
2677 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2679 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2683 dentry = lookup_one_len(name, dir, strlen(name));
2684 if (IS_ERR(dentry)) {
2685 error = PTR_ERR(dentry);
2689 mode = cgroup_file_mode(cft);
2690 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2692 cfe->type = (void *)cft;
2693 cfe->dentry = dentry;
2694 dentry->d_fsdata = cfe;
2695 list_add_tail(&cfe->node, &parent->files);
2704 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2705 const struct cftype cfts[], bool is_add)
2707 const struct cftype *cft;
2710 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2712 err = cgroup_add_file(cgrp, subsys, cft);
2714 err = cgroup_rm_file(cgrp, cft);
2716 pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2717 is_add ? "add" : "remove", cft->name, err);
2724 static DEFINE_MUTEX(cgroup_cft_mutex);
2726 static void cgroup_cfts_prepare(void)
2727 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2730 * Thanks to the entanglement with vfs inode locking, we can't walk
2731 * the existing cgroups under cgroup_mutex and create files.
2732 * Instead, we increment reference on all cgroups and build list of
2733 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2734 * exclusive access to the field.
2736 mutex_lock(&cgroup_cft_mutex);
2737 mutex_lock(&cgroup_mutex);
2740 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2741 const struct cftype *cfts, bool is_add)
2742 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2745 struct cgroup *cgrp, *n;
2747 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2748 if (cfts && ss->root != &rootnode) {
2749 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2751 list_add_tail(&cgrp->cft_q_node, &pending);
2755 mutex_unlock(&cgroup_mutex);
2758 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2759 * files for all cgroups which were created before.
2761 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2762 struct inode *inode = cgrp->dentry->d_inode;
2764 mutex_lock(&inode->i_mutex);
2765 mutex_lock(&cgroup_mutex);
2766 if (!cgroup_is_removed(cgrp))
2767 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2768 mutex_unlock(&cgroup_mutex);
2769 mutex_unlock(&inode->i_mutex);
2771 list_del_init(&cgrp->cft_q_node);
2775 mutex_unlock(&cgroup_cft_mutex);
2779 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2780 * @ss: target cgroup subsystem
2781 * @cfts: zero-length name terminated array of cftypes
2783 * Register @cfts to @ss. Files described by @cfts are created for all
2784 * existing cgroups to which @ss is attached and all future cgroups will
2785 * have them too. This function can be called anytime whether @ss is
2788 * Returns 0 on successful registration, -errno on failure. Note that this
2789 * function currently returns 0 as long as @cfts registration is successful
2790 * even if some file creation attempts on existing cgroups fail.
2792 int cgroup_add_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts)
2794 struct cftype_set *set;
2796 set = kzalloc(sizeof(*set), GFP_KERNEL);
2800 cgroup_cfts_prepare();
2802 list_add_tail(&set->node, &ss->cftsets);
2803 cgroup_cfts_commit(ss, cfts, true);
2807 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2810 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2811 * @ss: target cgroup subsystem
2812 * @cfts: zero-length name terminated array of cftypes
2814 * Unregister @cfts from @ss. Files described by @cfts are removed from
2815 * all existing cgroups to which @ss is attached and all future cgroups
2816 * won't have them either. This function can be called anytime whether @ss
2817 * is attached or not.
2819 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2820 * registered with @ss.
2822 int cgroup_rm_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts)
2824 struct cftype_set *set;
2826 cgroup_cfts_prepare();
2828 list_for_each_entry(set, &ss->cftsets, node) {
2829 if (set->cfts == cfts) {
2830 list_del_init(&set->node);
2831 cgroup_cfts_commit(ss, cfts, false);
2836 cgroup_cfts_commit(ss, NULL, false);
2841 * cgroup_task_count - count the number of tasks in a cgroup.
2842 * @cgrp: the cgroup in question
2844 * Return the number of tasks in the cgroup.
2846 int cgroup_task_count(const struct cgroup *cgrp)
2849 struct cg_cgroup_link *link;
2851 read_lock(&css_set_lock);
2852 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2853 count += atomic_read(&link->cg->refcount);
2855 read_unlock(&css_set_lock);
2860 * Advance a list_head iterator. The iterator should be positioned at
2861 * the start of a css_set
2863 static void cgroup_advance_iter(struct cgroup *cgrp,
2864 struct cgroup_iter *it)
2866 struct list_head *l = it->cg_link;
2867 struct cg_cgroup_link *link;
2870 /* Advance to the next non-empty css_set */
2873 if (l == &cgrp->css_sets) {
2877 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2879 } while (list_empty(&cg->tasks));
2881 it->task = cg->tasks.next;
2885 * To reduce the fork() overhead for systems that are not actually
2886 * using their cgroups capability, we don't maintain the lists running
2887 * through each css_set to its tasks until we see the list actually
2888 * used - in other words after the first call to cgroup_iter_start().
2890 static void cgroup_enable_task_cg_lists(void)
2892 struct task_struct *p, *g;
2893 write_lock(&css_set_lock);
2894 use_task_css_set_links = 1;
2896 * We need tasklist_lock because RCU is not safe against
2897 * while_each_thread(). Besides, a forking task that has passed
2898 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2899 * is not guaranteed to have its child immediately visible in the
2900 * tasklist if we walk through it with RCU.
2902 read_lock(&tasklist_lock);
2903 do_each_thread(g, p) {
2906 * We should check if the process is exiting, otherwise
2907 * it will race with cgroup_exit() in that the list
2908 * entry won't be deleted though the process has exited.
2910 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2911 list_add(&p->cg_list, &p->cgroups->tasks);
2913 } while_each_thread(g, p);
2914 read_unlock(&tasklist_lock);
2915 write_unlock(&css_set_lock);
2918 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2919 __acquires(css_set_lock)
2922 * The first time anyone tries to iterate across a cgroup,
2923 * we need to enable the list linking each css_set to its
2924 * tasks, and fix up all existing tasks.
2926 if (!use_task_css_set_links)
2927 cgroup_enable_task_cg_lists();
2929 read_lock(&css_set_lock);
2930 it->cg_link = &cgrp->css_sets;
2931 cgroup_advance_iter(cgrp, it);
2934 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2935 struct cgroup_iter *it)
2937 struct task_struct *res;
2938 struct list_head *l = it->task;
2939 struct cg_cgroup_link *link;
2941 /* If the iterator cg is NULL, we have no tasks */
2944 res = list_entry(l, struct task_struct, cg_list);
2945 /* Advance iterator to find next entry */
2947 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2948 if (l == &link->cg->tasks) {
2949 /* We reached the end of this task list - move on to
2950 * the next cg_cgroup_link */
2951 cgroup_advance_iter(cgrp, it);
2958 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2959 __releases(css_set_lock)
2961 read_unlock(&css_set_lock);
2964 static inline int started_after_time(struct task_struct *t1,
2965 struct timespec *time,
2966 struct task_struct *t2)
2968 int start_diff = timespec_compare(&t1->start_time, time);
2969 if (start_diff > 0) {
2971 } else if (start_diff < 0) {
2975 * Arbitrarily, if two processes started at the same
2976 * time, we'll say that the lower pointer value
2977 * started first. Note that t2 may have exited by now
2978 * so this may not be a valid pointer any longer, but
2979 * that's fine - it still serves to distinguish
2980 * between two tasks started (effectively) simultaneously.
2987 * This function is a callback from heap_insert() and is used to order
2989 * In this case we order the heap in descending task start time.
2991 static inline int started_after(void *p1, void *p2)
2993 struct task_struct *t1 = p1;
2994 struct task_struct *t2 = p2;
2995 return started_after_time(t1, &t2->start_time, t2);
2999 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3000 * @scan: struct cgroup_scanner containing arguments for the scan
3002 * Arguments include pointers to callback functions test_task() and
3004 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3005 * and if it returns true, call process_task() for it also.
3006 * The test_task pointer may be NULL, meaning always true (select all tasks).
3007 * Effectively duplicates cgroup_iter_{start,next,end}()
3008 * but does not lock css_set_lock for the call to process_task().
3009 * The struct cgroup_scanner may be embedded in any structure of the caller's
3011 * It is guaranteed that process_task() will act on every task that
3012 * is a member of the cgroup for the duration of this call. This
3013 * function may or may not call process_task() for tasks that exit
3014 * or move to a different cgroup during the call, or are forked or
3015 * move into the cgroup during the call.
3017 * Note that test_task() may be called with locks held, and may in some
3018 * situations be called multiple times for the same task, so it should
3020 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3021 * pre-allocated and will be used for heap operations (and its "gt" member will
3022 * be overwritten), else a temporary heap will be used (allocation of which
3023 * may cause this function to fail).
3025 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3028 struct cgroup_iter it;
3029 struct task_struct *p, *dropped;
3030 /* Never dereference latest_task, since it's not refcounted */
3031 struct task_struct *latest_task = NULL;
3032 struct ptr_heap tmp_heap;
3033 struct ptr_heap *heap;
3034 struct timespec latest_time = { 0, 0 };
3037 /* The caller supplied our heap and pre-allocated its memory */
3039 heap->gt = &started_after;
3041 /* We need to allocate our own heap memory */
3043 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3045 /* cannot allocate the heap */
3051 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3052 * to determine which are of interest, and using the scanner's
3053 * "process_task" callback to process any of them that need an update.
3054 * Since we don't want to hold any locks during the task updates,
3055 * gather tasks to be processed in a heap structure.
3056 * The heap is sorted by descending task start time.
3057 * If the statically-sized heap fills up, we overflow tasks that
3058 * started later, and in future iterations only consider tasks that
3059 * started after the latest task in the previous pass. This
3060 * guarantees forward progress and that we don't miss any tasks.
3063 cgroup_iter_start(scan->cg, &it);
3064 while ((p = cgroup_iter_next(scan->cg, &it))) {
3066 * Only affect tasks that qualify per the caller's callback,
3067 * if he provided one
3069 if (scan->test_task && !scan->test_task(p, scan))
3072 * Only process tasks that started after the last task
3075 if (!started_after_time(p, &latest_time, latest_task))
3077 dropped = heap_insert(heap, p);
3078 if (dropped == NULL) {
3080 * The new task was inserted; the heap wasn't
3084 } else if (dropped != p) {
3086 * The new task was inserted, and pushed out a
3090 put_task_struct(dropped);
3093 * Else the new task was newer than anything already in
3094 * the heap and wasn't inserted
3097 cgroup_iter_end(scan->cg, &it);
3100 for (i = 0; i < heap->size; i++) {
3101 struct task_struct *q = heap->ptrs[i];
3103 latest_time = q->start_time;
3106 /* Process the task per the caller's callback */
3107 scan->process_task(q, scan);
3111 * If we had to process any tasks at all, scan again
3112 * in case some of them were in the middle of forking
3113 * children that didn't get processed.
3114 * Not the most efficient way to do it, but it avoids
3115 * having to take callback_mutex in the fork path
3119 if (heap == &tmp_heap)
3120 heap_free(&tmp_heap);
3125 * Stuff for reading the 'tasks'/'procs' files.
3127 * Reading this file can return large amounts of data if a cgroup has
3128 * *lots* of attached tasks. So it may need several calls to read(),
3129 * but we cannot guarantee that the information we produce is correct
3130 * unless we produce it entirely atomically.
3134 /* which pidlist file are we talking about? */
3135 enum cgroup_filetype {
3141 * A pidlist is a list of pids that virtually represents the contents of one
3142 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3143 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3146 struct cgroup_pidlist {
3148 * used to find which pidlist is wanted. doesn't change as long as
3149 * this particular list stays in the list.
3151 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3154 /* how many elements the above list has */
3156 /* how many files are using the current array */
3158 /* each of these stored in a list by its cgroup */
3159 struct list_head links;
3160 /* pointer to the cgroup we belong to, for list removal purposes */
3161 struct cgroup *owner;
3162 /* protects the other fields */
3163 struct rw_semaphore mutex;
3167 * The following two functions "fix" the issue where there are more pids
3168 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3169 * TODO: replace with a kernel-wide solution to this problem
3171 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3172 static void *pidlist_allocate(int count)
3174 if (PIDLIST_TOO_LARGE(count))
3175 return vmalloc(count * sizeof(pid_t));
3177 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3179 static void pidlist_free(void *p)
3181 if (is_vmalloc_addr(p))
3186 static void *pidlist_resize(void *p, int newcount)
3189 /* note: if new alloc fails, old p will still be valid either way */
3190 if (is_vmalloc_addr(p)) {
3191 newlist = vmalloc(newcount * sizeof(pid_t));
3194 memcpy(newlist, p, newcount * sizeof(pid_t));
3197 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3203 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3204 * If the new stripped list is sufficiently smaller and there's enough memory
3205 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3206 * number of unique elements.
3208 /* is the size difference enough that we should re-allocate the array? */
3209 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3210 static int pidlist_uniq(pid_t **p, int length)
3217 * we presume the 0th element is unique, so i starts at 1. trivial
3218 * edge cases first; no work needs to be done for either
3220 if (length == 0 || length == 1)
3222 /* src and dest walk down the list; dest counts unique elements */
3223 for (src = 1; src < length; src++) {
3224 /* find next unique element */
3225 while (list[src] == list[src-1]) {
3230 /* dest always points to where the next unique element goes */
3231 list[dest] = list[src];
3236 * if the length difference is large enough, we want to allocate a
3237 * smaller buffer to save memory. if this fails due to out of memory,
3238 * we'll just stay with what we've got.
3240 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3241 newlist = pidlist_resize(list, dest);
3248 static int cmppid(const void *a, const void *b)
3250 return *(pid_t *)a - *(pid_t *)b;
3254 * find the appropriate pidlist for our purpose (given procs vs tasks)
3255 * returns with the lock on that pidlist already held, and takes care
3256 * of the use count, or returns NULL with no locks held if we're out of
3259 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3260 enum cgroup_filetype type)
3262 struct cgroup_pidlist *l;
3263 /* don't need task_nsproxy() if we're looking at ourself */
3264 struct pid_namespace *ns = current->nsproxy->pid_ns;
3267 * We can't drop the pidlist_mutex before taking the l->mutex in case
3268 * the last ref-holder is trying to remove l from the list at the same
3269 * time. Holding the pidlist_mutex precludes somebody taking whichever
3270 * list we find out from under us - compare release_pid_array().
3272 mutex_lock(&cgrp->pidlist_mutex);
3273 list_for_each_entry(l, &cgrp->pidlists, links) {
3274 if (l->key.type == type && l->key.ns == ns) {
3275 /* make sure l doesn't vanish out from under us */
3276 down_write(&l->mutex);
3277 mutex_unlock(&cgrp->pidlist_mutex);
3281 /* entry not found; create a new one */
3282 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3284 mutex_unlock(&cgrp->pidlist_mutex);
3287 init_rwsem(&l->mutex);
3288 down_write(&l->mutex);
3290 l->key.ns = get_pid_ns(ns);
3291 l->use_count = 0; /* don't increment here */
3294 list_add(&l->links, &cgrp->pidlists);
3295 mutex_unlock(&cgrp->pidlist_mutex);
3300 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3302 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3303 struct cgroup_pidlist **lp)
3307 int pid, n = 0; /* used for populating the array */
3308 struct cgroup_iter it;
3309 struct task_struct *tsk;
3310 struct cgroup_pidlist *l;
3313 * If cgroup gets more users after we read count, we won't have
3314 * enough space - tough. This race is indistinguishable to the
3315 * caller from the case that the additional cgroup users didn't
3316 * show up until sometime later on.
3318 length = cgroup_task_count(cgrp);
3319 array = pidlist_allocate(length);
3322 /* now, populate the array */
3323 cgroup_iter_start(cgrp, &it);
3324 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3325 if (unlikely(n == length))
3327 /* get tgid or pid for procs or tasks file respectively */
3328 if (type == CGROUP_FILE_PROCS)
3329 pid = task_tgid_vnr(tsk);
3331 pid = task_pid_vnr(tsk);
3332 if (pid > 0) /* make sure to only use valid results */
3335 cgroup_iter_end(cgrp, &it);
3337 /* now sort & (if procs) strip out duplicates */
3338 sort(array, length, sizeof(pid_t), cmppid, NULL);
3339 if (type == CGROUP_FILE_PROCS)
3340 length = pidlist_uniq(&array, length);
3341 l = cgroup_pidlist_find(cgrp, type);
3343 pidlist_free(array);
3346 /* store array, freeing old if necessary - lock already held */
3347 pidlist_free(l->list);
3351 up_write(&l->mutex);
3357 * cgroupstats_build - build and fill cgroupstats
3358 * @stats: cgroupstats to fill information into
3359 * @dentry: A dentry entry belonging to the cgroup for which stats have
3362 * Build and fill cgroupstats so that taskstats can export it to user
3365 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3368 struct cgroup *cgrp;
3369 struct cgroup_iter it;
3370 struct task_struct *tsk;
3373 * Validate dentry by checking the superblock operations,
3374 * and make sure it's a directory.
3376 if (dentry->d_sb->s_op != &cgroup_ops ||
3377 !S_ISDIR(dentry->d_inode->i_mode))
3381 cgrp = dentry->d_fsdata;
3383 cgroup_iter_start(cgrp, &it);
3384 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3385 switch (tsk->state) {
3387 stats->nr_running++;
3389 case TASK_INTERRUPTIBLE:
3390 stats->nr_sleeping++;
3392 case TASK_UNINTERRUPTIBLE:
3393 stats->nr_uninterruptible++;
3396 stats->nr_stopped++;
3399 if (delayacct_is_task_waiting_on_io(tsk))
3400 stats->nr_io_wait++;
3404 cgroup_iter_end(cgrp, &it);
3412 * seq_file methods for the tasks/procs files. The seq_file position is the
3413 * next pid to display; the seq_file iterator is a pointer to the pid
3414 * in the cgroup->l->list array.
3417 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3420 * Initially we receive a position value that corresponds to
3421 * one more than the last pid shown (or 0 on the first call or
3422 * after a seek to the start). Use a binary-search to find the
3423 * next pid to display, if any
3425 struct cgroup_pidlist *l = s->private;
3426 int index = 0, pid = *pos;
3429 down_read(&l->mutex);
3431 int end = l->length;
3433 while (index < end) {
3434 int mid = (index + end) / 2;
3435 if (l->list[mid] == pid) {
3438 } else if (l->list[mid] <= pid)
3444 /* If we're off the end of the array, we're done */
3445 if (index >= l->length)
3447 /* Update the abstract position to be the actual pid that we found */
3448 iter = l->list + index;
3453 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3455 struct cgroup_pidlist *l = s->private;
3459 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3461 struct cgroup_pidlist *l = s->private;
3463 pid_t *end = l->list + l->length;
3465 * Advance to the next pid in the array. If this goes off the
3477 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3479 return seq_printf(s, "%d\n", *(int *)v);
3483 * seq_operations functions for iterating on pidlists through seq_file -
3484 * independent of whether it's tasks or procs
3486 static const struct seq_operations cgroup_pidlist_seq_operations = {
3487 .start = cgroup_pidlist_start,
3488 .stop = cgroup_pidlist_stop,
3489 .next = cgroup_pidlist_next,
3490 .show = cgroup_pidlist_show,
3493 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3496 * the case where we're the last user of this particular pidlist will
3497 * have us remove it from the cgroup's list, which entails taking the
3498 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3499 * pidlist_mutex, we have to take pidlist_mutex first.
3501 mutex_lock(&l->owner->pidlist_mutex);
3502 down_write(&l->mutex);
3503 BUG_ON(!l->use_count);
3504 if (!--l->use_count) {
3505 /* we're the last user if refcount is 0; remove and free */
3506 list_del(&l->links);
3507 mutex_unlock(&l->owner->pidlist_mutex);
3508 pidlist_free(l->list);
3509 put_pid_ns(l->key.ns);
3510 up_write(&l->mutex);
3514 mutex_unlock(&l->owner->pidlist_mutex);
3515 up_write(&l->mutex);
3518 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3520 struct cgroup_pidlist *l;
3521 if (!(file->f_mode & FMODE_READ))
3524 * the seq_file will only be initialized if the file was opened for
3525 * reading; hence we check if it's not null only in that case.
3527 l = ((struct seq_file *)file->private_data)->private;
3528 cgroup_release_pid_array(l);
3529 return seq_release(inode, file);
3532 static const struct file_operations cgroup_pidlist_operations = {
3534 .llseek = seq_lseek,
3535 .write = cgroup_file_write,
3536 .release = cgroup_pidlist_release,
3540 * The following functions handle opens on a file that displays a pidlist
3541 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3544 /* helper function for the two below it */
3545 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3547 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3548 struct cgroup_pidlist *l;
3551 /* Nothing to do for write-only files */
3552 if (!(file->f_mode & FMODE_READ))
3555 /* have the array populated */
3556 retval = pidlist_array_load(cgrp, type, &l);
3559 /* configure file information */
3560 file->f_op = &cgroup_pidlist_operations;
3562 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3564 cgroup_release_pid_array(l);
3567 ((struct seq_file *)file->private_data)->private = l;
3570 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3572 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3574 static int cgroup_procs_open(struct inode *unused, struct file *file)
3576 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3579 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3582 return notify_on_release(cgrp);
3585 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3589 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3591 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3593 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3598 * Unregister event and free resources.
3600 * Gets called from workqueue.
3602 static void cgroup_event_remove(struct work_struct *work)
3604 struct cgroup_event *event = container_of(work, struct cgroup_event,
3606 struct cgroup *cgrp = event->cgrp;
3608 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3610 eventfd_ctx_put(event->eventfd);
3616 * Gets called on POLLHUP on eventfd when user closes it.
3618 * Called with wqh->lock held and interrupts disabled.
3620 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3621 int sync, void *key)
3623 struct cgroup_event *event = container_of(wait,
3624 struct cgroup_event, wait);
3625 struct cgroup *cgrp = event->cgrp;
3626 unsigned long flags = (unsigned long)key;
3628 if (flags & POLLHUP) {
3629 __remove_wait_queue(event->wqh, &event->wait);
3630 spin_lock(&cgrp->event_list_lock);
3631 list_del(&event->list);
3632 spin_unlock(&cgrp->event_list_lock);
3634 * We are in atomic context, but cgroup_event_remove() may
3635 * sleep, so we have to call it in workqueue.
3637 schedule_work(&event->remove);
3643 static void cgroup_event_ptable_queue_proc(struct file *file,
3644 wait_queue_head_t *wqh, poll_table *pt)
3646 struct cgroup_event *event = container_of(pt,
3647 struct cgroup_event, pt);
3650 add_wait_queue(wqh, &event->wait);
3654 * Parse input and register new cgroup event handler.
3656 * Input must be in format '<event_fd> <control_fd> <args>'.
3657 * Interpretation of args is defined by control file implementation.
3659 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3662 struct cgroup_event *event = NULL;
3663 unsigned int efd, cfd;
3664 struct file *efile = NULL;
3665 struct file *cfile = NULL;
3669 efd = simple_strtoul(buffer, &endp, 10);
3674 cfd = simple_strtoul(buffer, &endp, 10);
3675 if ((*endp != ' ') && (*endp != '\0'))
3679 event = kzalloc(sizeof(*event), GFP_KERNEL);
3683 INIT_LIST_HEAD(&event->list);
3684 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3685 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3686 INIT_WORK(&event->remove, cgroup_event_remove);
3688 efile = eventfd_fget(efd);
3689 if (IS_ERR(efile)) {
3690 ret = PTR_ERR(efile);
3694 event->eventfd = eventfd_ctx_fileget(efile);
3695 if (IS_ERR(event->eventfd)) {
3696 ret = PTR_ERR(event->eventfd);
3706 /* the process need read permission on control file */
3707 /* AV: shouldn't we check that it's been opened for read instead? */
3708 ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3712 event->cft = __file_cft(cfile);
3713 if (IS_ERR(event->cft)) {
3714 ret = PTR_ERR(event->cft);
3718 if (!event->cft->register_event || !event->cft->unregister_event) {
3723 ret = event->cft->register_event(cgrp, event->cft,
3724 event->eventfd, buffer);
3728 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3729 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3735 * Events should be removed after rmdir of cgroup directory, but before
3736 * destroying subsystem state objects. Let's take reference to cgroup
3737 * directory dentry to do that.
3741 spin_lock(&cgrp->event_list_lock);
3742 list_add(&event->list, &cgrp->event_list);
3743 spin_unlock(&cgrp->event_list_lock);
3754 if (event && event->eventfd && !IS_ERR(event->eventfd))
3755 eventfd_ctx_put(event->eventfd);
3757 if (!IS_ERR_OR_NULL(efile))
3765 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3768 return clone_children(cgrp);
3771 static int cgroup_clone_children_write(struct cgroup *cgrp,
3776 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3778 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3783 * for the common functions, 'private' gives the type of file
3785 /* for hysterical raisins, we can't put this on the older files */
3786 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3787 static struct cftype files[] = {
3790 .open = cgroup_tasks_open,
3791 .write_u64 = cgroup_tasks_write,
3792 .release = cgroup_pidlist_release,
3793 .mode = S_IRUGO | S_IWUSR,
3796 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3797 .open = cgroup_procs_open,
3798 .write_u64 = cgroup_procs_write,
3799 .release = cgroup_pidlist_release,
3800 .mode = S_IRUGO | S_IWUSR,
3803 .name = "notify_on_release",
3804 .read_u64 = cgroup_read_notify_on_release,
3805 .write_u64 = cgroup_write_notify_on_release,
3808 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3809 .write_string = cgroup_write_event_control,
3813 .name = "cgroup.clone_children",
3814 .read_u64 = cgroup_clone_children_read,
3815 .write_u64 = cgroup_clone_children_write,
3818 .name = "release_agent",
3819 .flags = CFTYPE_ONLY_ON_ROOT,
3820 .read_seq_string = cgroup_release_agent_show,
3821 .write_string = cgroup_release_agent_write,
3822 .max_write_len = PATH_MAX,
3827 static int cgroup_populate_dir(struct cgroup *cgrp)
3830 struct cgroup_subsys *ss;
3832 err = cgroup_addrm_files(cgrp, NULL, files, true);
3836 /* process cftsets of each subsystem */
3837 for_each_subsys(cgrp->root, ss) {
3838 struct cftype_set *set;
3840 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
3843 list_for_each_entry(set, &ss->cftsets, node)
3844 cgroup_addrm_files(cgrp, ss, set->cfts, true);
3847 /* This cgroup is ready now */
3848 for_each_subsys(cgrp->root, ss) {
3849 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3851 * Update id->css pointer and make this css visible from
3852 * CSS ID functions. This pointer will be dereferened
3853 * from RCU-read-side without locks.
3856 rcu_assign_pointer(css->id->css, css);
3862 static void init_cgroup_css(struct cgroup_subsys_state *css,
3863 struct cgroup_subsys *ss,
3864 struct cgroup *cgrp)
3867 atomic_set(&css->refcnt, 1);
3870 if (cgrp == dummytop)
3871 set_bit(CSS_ROOT, &css->flags);
3872 BUG_ON(cgrp->subsys[ss->subsys_id]);
3873 cgrp->subsys[ss->subsys_id] = css;
3876 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3878 /* We need to take each hierarchy_mutex in a consistent order */
3882 * No worry about a race with rebind_subsystems that might mess up the
3883 * locking order, since both parties are under cgroup_mutex.
3885 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3886 struct cgroup_subsys *ss = subsys[i];
3889 if (ss->root == root)
3890 mutex_lock(&ss->hierarchy_mutex);
3894 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3898 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3899 struct cgroup_subsys *ss = subsys[i];
3902 if (ss->root == root)
3903 mutex_unlock(&ss->hierarchy_mutex);
3908 * cgroup_create - create a cgroup
3909 * @parent: cgroup that will be parent of the new cgroup
3910 * @dentry: dentry of the new cgroup
3911 * @mode: mode to set on new inode
3913 * Must be called with the mutex on the parent inode held
3915 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3918 struct cgroup *cgrp;
3919 struct cgroupfs_root *root = parent->root;
3921 struct cgroup_subsys *ss;
3922 struct super_block *sb = root->sb;
3924 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3928 /* Grab a reference on the superblock so the hierarchy doesn't
3929 * get deleted on unmount if there are child cgroups. This
3930 * can be done outside cgroup_mutex, since the sb can't
3931 * disappear while someone has an open control file on the
3933 atomic_inc(&sb->s_active);
3935 mutex_lock(&cgroup_mutex);
3937 init_cgroup_housekeeping(cgrp);
3939 cgrp->parent = parent;
3940 cgrp->root = parent->root;
3941 cgrp->top_cgroup = parent->top_cgroup;
3943 if (notify_on_release(parent))
3944 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3946 if (clone_children(parent))
3947 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3949 for_each_subsys(root, ss) {
3950 struct cgroup_subsys_state *css = ss->create(cgrp);
3956 init_cgroup_css(css, ss, cgrp);
3958 err = alloc_css_id(ss, parent, cgrp);
3962 /* At error, ->destroy() callback has to free assigned ID. */
3963 if (clone_children(parent) && ss->post_clone)
3964 ss->post_clone(cgrp);
3967 cgroup_lock_hierarchy(root);
3968 list_add(&cgrp->sibling, &cgrp->parent->children);
3969 cgroup_unlock_hierarchy(root);
3970 root->number_of_cgroups++;
3972 err = cgroup_create_dir(cgrp, dentry, mode);
3976 /* The cgroup directory was pre-locked for us */
3977 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3979 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
3981 err = cgroup_populate_dir(cgrp);
3982 /* If err < 0, we have a half-filled directory - oh well ;) */
3984 mutex_unlock(&cgroup_mutex);
3985 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3991 cgroup_lock_hierarchy(root);
3992 list_del(&cgrp->sibling);
3993 cgroup_unlock_hierarchy(root);
3994 root->number_of_cgroups--;
3998 for_each_subsys(root, ss) {
3999 if (cgrp->subsys[ss->subsys_id])
4003 mutex_unlock(&cgroup_mutex);
4005 /* Release the reference count that we took on the superblock */
4006 deactivate_super(sb);
4012 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4014 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4016 /* the vfs holds inode->i_mutex already */
4017 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4021 * Check the reference count on each subsystem. Since we already
4022 * established that there are no tasks in the cgroup, if the css refcount
4023 * is also 1, then there should be no outstanding references, so the
4024 * subsystem is safe to destroy. We scan across all subsystems rather than
4025 * using the per-hierarchy linked list of mounted subsystems since we can
4026 * be called via check_for_release() with no synchronization other than
4027 * RCU, and the subsystem linked list isn't RCU-safe.
4029 static int cgroup_has_css_refs(struct cgroup *cgrp)
4034 * We won't need to lock the subsys array, because the subsystems
4035 * we're concerned about aren't going anywhere since our cgroup root
4036 * has a reference on them.
4038 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4039 struct cgroup_subsys *ss = subsys[i];
4040 struct cgroup_subsys_state *css;
4042 /* Skip subsystems not present or not in this hierarchy */
4043 if (ss == NULL || ss->root != cgrp->root)
4046 css = cgrp->subsys[ss->subsys_id];
4048 * When called from check_for_release() it's possible
4049 * that by this point the cgroup has been removed
4050 * and the css deleted. But a false-positive doesn't
4051 * matter, since it can only happen if the cgroup
4052 * has been deleted and hence no longer needs the
4053 * release agent to be called anyway.
4055 if (css && css_refcnt(css) > 1)
4062 * Atomically mark all (or else none) of the cgroup's CSS objects as
4063 * CSS_REMOVED. Return true on success, or false if the cgroup has
4064 * busy subsystems. Call with cgroup_mutex held
4067 static int cgroup_clear_css_refs(struct cgroup *cgrp)
4069 struct cgroup_subsys *ss;
4070 unsigned long flags;
4071 bool failed = false;
4073 local_irq_save(flags);
4076 * Block new css_tryget() by deactivating refcnt. If all refcnts
4077 * were 1 at the moment of deactivation, we succeeded.
4079 for_each_subsys(cgrp->root, ss) {
4080 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4082 WARN_ON(atomic_read(&css->refcnt) < 0);
4083 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4084 failed |= css_refcnt(css) != 1;
4088 * If succeeded, set REMOVED and put all the base refs; otherwise,
4089 * restore refcnts to positive values. Either way, all in-progress
4090 * css_tryget() will be released.
4092 for_each_subsys(cgrp->root, ss) {
4093 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4096 set_bit(CSS_REMOVED, &css->flags);
4099 atomic_sub(CSS_DEACT_BIAS, &css->refcnt);
4103 local_irq_restore(flags);
4107 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4109 struct cgroup *cgrp = dentry->d_fsdata;
4111 struct cgroup *parent;
4113 struct cgroup_event *event, *tmp;
4116 /* the vfs holds both inode->i_mutex already */
4118 mutex_lock(&cgroup_mutex);
4119 if (atomic_read(&cgrp->count) != 0) {
4120 mutex_unlock(&cgroup_mutex);
4123 if (!list_empty(&cgrp->children)) {
4124 mutex_unlock(&cgroup_mutex);
4127 mutex_unlock(&cgroup_mutex);
4130 * In general, subsystem has no css->refcnt after pre_destroy(). But
4131 * in racy cases, subsystem may have to get css->refcnt after
4132 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4133 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4134 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4135 * and subsystem's reference count handling. Please see css_get/put
4136 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4138 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4141 * Call pre_destroy handlers of subsys. Notify subsystems
4142 * that rmdir() request comes.
4144 ret = cgroup_call_pre_destroy(cgrp);
4146 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4150 mutex_lock(&cgroup_mutex);
4151 parent = cgrp->parent;
4152 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
4153 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4154 mutex_unlock(&cgroup_mutex);
4157 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
4158 if (!cgroup_clear_css_refs(cgrp)) {
4159 mutex_unlock(&cgroup_mutex);
4161 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4162 * prepare_to_wait(), we need to check this flag.
4164 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
4166 finish_wait(&cgroup_rmdir_waitq, &wait);
4167 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4168 if (signal_pending(current))
4172 /* NO css_tryget() can success after here. */
4173 finish_wait(&cgroup_rmdir_waitq, &wait);
4174 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4176 raw_spin_lock(&release_list_lock);
4177 set_bit(CGRP_REMOVED, &cgrp->flags);
4178 if (!list_empty(&cgrp->release_list))
4179 list_del_init(&cgrp->release_list);
4180 raw_spin_unlock(&release_list_lock);
4182 cgroup_lock_hierarchy(cgrp->root);
4183 /* delete this cgroup from parent->children */
4184 list_del_init(&cgrp->sibling);
4185 cgroup_unlock_hierarchy(cgrp->root);
4187 list_del_init(&cgrp->allcg_node);
4189 d = dget(cgrp->dentry);
4191 cgroup_d_remove_dir(d);
4194 set_bit(CGRP_RELEASABLE, &parent->flags);
4195 check_for_release(parent);
4198 * Unregister events and notify userspace.
4199 * Notify userspace about cgroup removing only after rmdir of cgroup
4200 * directory to avoid race between userspace and kernelspace
4202 spin_lock(&cgrp->event_list_lock);
4203 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4204 list_del(&event->list);
4205 remove_wait_queue(event->wqh, &event->wait);
4206 eventfd_signal(event->eventfd, 1);
4207 schedule_work(&event->remove);
4209 spin_unlock(&cgrp->event_list_lock);
4211 mutex_unlock(&cgroup_mutex);
4215 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4217 INIT_LIST_HEAD(&ss->cftsets);
4220 * base_cftset is embedded in subsys itself, no need to worry about
4223 if (ss->base_cftypes) {
4224 ss->base_cftset.cfts = ss->base_cftypes;
4225 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4229 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4231 struct cgroup_subsys_state *css;
4233 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4235 /* init base cftset */
4236 cgroup_init_cftsets(ss);
4238 /* Create the top cgroup state for this subsystem */
4239 list_add(&ss->sibling, &rootnode.subsys_list);
4240 ss->root = &rootnode;
4241 css = ss->create(dummytop);
4242 /* We don't handle early failures gracefully */
4243 BUG_ON(IS_ERR(css));
4244 init_cgroup_css(css, ss, dummytop);
4246 /* Update the init_css_set to contain a subsys
4247 * pointer to this state - since the subsystem is
4248 * newly registered, all tasks and hence the
4249 * init_css_set is in the subsystem's top cgroup. */
4250 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4252 need_forkexit_callback |= ss->fork || ss->exit;
4254 /* At system boot, before all subsystems have been
4255 * registered, no tasks have been forked, so we don't
4256 * need to invoke fork callbacks here. */
4257 BUG_ON(!list_empty(&init_task.tasks));
4259 mutex_init(&ss->hierarchy_mutex);
4260 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4263 /* this function shouldn't be used with modular subsystems, since they
4264 * need to register a subsys_id, among other things */
4269 * cgroup_load_subsys: load and register a modular subsystem at runtime
4270 * @ss: the subsystem to load
4272 * This function should be called in a modular subsystem's initcall. If the
4273 * subsystem is built as a module, it will be assigned a new subsys_id and set
4274 * up for use. If the subsystem is built-in anyway, work is delegated to the
4275 * simpler cgroup_init_subsys.
4277 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4280 struct cgroup_subsys_state *css;
4282 /* check name and function validity */
4283 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4284 ss->create == NULL || ss->destroy == NULL)
4288 * we don't support callbacks in modular subsystems. this check is
4289 * before the ss->module check for consistency; a subsystem that could
4290 * be a module should still have no callbacks even if the user isn't
4291 * compiling it as one.
4293 if (ss->fork || ss->exit)
4297 * an optionally modular subsystem is built-in: we want to do nothing,
4298 * since cgroup_init_subsys will have already taken care of it.
4300 if (ss->module == NULL) {
4301 /* a few sanity checks */
4302 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4303 BUG_ON(subsys[ss->subsys_id] != ss);
4307 /* init base cftset */
4308 cgroup_init_cftsets(ss);
4311 * need to register a subsys id before anything else - for example,
4312 * init_cgroup_css needs it.
4314 mutex_lock(&cgroup_mutex);
4315 /* find the first empty slot in the array */
4316 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4317 if (subsys[i] == NULL)
4320 if (i == CGROUP_SUBSYS_COUNT) {
4321 /* maximum number of subsystems already registered! */
4322 mutex_unlock(&cgroup_mutex);
4325 /* assign ourselves the subsys_id */
4330 * no ss->create seems to need anything important in the ss struct, so
4331 * this can happen first (i.e. before the rootnode attachment).
4333 css = ss->create(dummytop);
4335 /* failure case - need to deassign the subsys[] slot. */
4337 mutex_unlock(&cgroup_mutex);
4338 return PTR_ERR(css);
4341 list_add(&ss->sibling, &rootnode.subsys_list);
4342 ss->root = &rootnode;
4344 /* our new subsystem will be attached to the dummy hierarchy. */
4345 init_cgroup_css(css, ss, dummytop);
4346 /* init_idr must be after init_cgroup_css because it sets css->id. */
4348 int ret = cgroup_init_idr(ss, css);
4350 dummytop->subsys[ss->subsys_id] = NULL;
4351 ss->destroy(dummytop);
4353 mutex_unlock(&cgroup_mutex);
4359 * Now we need to entangle the css into the existing css_sets. unlike
4360 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4361 * will need a new pointer to it; done by iterating the css_set_table.
4362 * furthermore, modifying the existing css_sets will corrupt the hash
4363 * table state, so each changed css_set will need its hash recomputed.
4364 * this is all done under the css_set_lock.
4366 write_lock(&css_set_lock);
4367 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4369 struct hlist_node *node, *tmp;
4370 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4372 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4373 /* skip entries that we already rehashed */
4374 if (cg->subsys[ss->subsys_id])
4376 /* remove existing entry */
4377 hlist_del(&cg->hlist);
4379 cg->subsys[ss->subsys_id] = css;
4380 /* recompute hash and restore entry */
4381 new_bucket = css_set_hash(cg->subsys);
4382 hlist_add_head(&cg->hlist, new_bucket);
4385 write_unlock(&css_set_lock);
4387 mutex_init(&ss->hierarchy_mutex);
4388 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4392 mutex_unlock(&cgroup_mutex);
4395 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4398 * cgroup_unload_subsys: unload a modular subsystem
4399 * @ss: the subsystem to unload
4401 * This function should be called in a modular subsystem's exitcall. When this
4402 * function is invoked, the refcount on the subsystem's module will be 0, so
4403 * the subsystem will not be attached to any hierarchy.
4405 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4407 struct cg_cgroup_link *link;
4408 struct hlist_head *hhead;
4410 BUG_ON(ss->module == NULL);
4413 * we shouldn't be called if the subsystem is in use, and the use of
4414 * try_module_get in parse_cgroupfs_options should ensure that it
4415 * doesn't start being used while we're killing it off.
4417 BUG_ON(ss->root != &rootnode);
4419 mutex_lock(&cgroup_mutex);
4420 /* deassign the subsys_id */
4421 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4422 subsys[ss->subsys_id] = NULL;
4424 /* remove subsystem from rootnode's list of subsystems */
4425 list_del_init(&ss->sibling);
4428 * disentangle the css from all css_sets attached to the dummytop. as
4429 * in loading, we need to pay our respects to the hashtable gods.
4431 write_lock(&css_set_lock);
4432 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4433 struct css_set *cg = link->cg;
4435 hlist_del(&cg->hlist);
4436 BUG_ON(!cg->subsys[ss->subsys_id]);
4437 cg->subsys[ss->subsys_id] = NULL;
4438 hhead = css_set_hash(cg->subsys);
4439 hlist_add_head(&cg->hlist, hhead);
4441 write_unlock(&css_set_lock);
4444 * remove subsystem's css from the dummytop and free it - need to free
4445 * before marking as null because ss->destroy needs the cgrp->subsys
4446 * pointer to find their state. note that this also takes care of
4447 * freeing the css_id.
4449 ss->destroy(dummytop);
4450 dummytop->subsys[ss->subsys_id] = NULL;
4452 mutex_unlock(&cgroup_mutex);
4454 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4457 * cgroup_init_early - cgroup initialization at system boot
4459 * Initialize cgroups at system boot, and initialize any
4460 * subsystems that request early init.
4462 int __init cgroup_init_early(void)
4465 atomic_set(&init_css_set.refcount, 1);
4466 INIT_LIST_HEAD(&init_css_set.cg_links);
4467 INIT_LIST_HEAD(&init_css_set.tasks);
4468 INIT_HLIST_NODE(&init_css_set.hlist);
4470 init_cgroup_root(&rootnode);
4472 init_task.cgroups = &init_css_set;
4474 init_css_set_link.cg = &init_css_set;
4475 init_css_set_link.cgrp = dummytop;
4476 list_add(&init_css_set_link.cgrp_link_list,
4477 &rootnode.top_cgroup.css_sets);
4478 list_add(&init_css_set_link.cg_link_list,
4479 &init_css_set.cg_links);
4481 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4482 INIT_HLIST_HEAD(&css_set_table[i]);
4484 /* at bootup time, we don't worry about modular subsystems */
4485 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4486 struct cgroup_subsys *ss = subsys[i];
4489 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4490 BUG_ON(!ss->create);
4491 BUG_ON(!ss->destroy);
4492 if (ss->subsys_id != i) {
4493 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4494 ss->name, ss->subsys_id);
4499 cgroup_init_subsys(ss);
4505 * cgroup_init - cgroup initialization
4507 * Register cgroup filesystem and /proc file, and initialize
4508 * any subsystems that didn't request early init.
4510 int __init cgroup_init(void)
4514 struct hlist_head *hhead;
4516 err = bdi_init(&cgroup_backing_dev_info);
4520 /* at bootup time, we don't worry about modular subsystems */
4521 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4522 struct cgroup_subsys *ss = subsys[i];
4523 if (!ss->early_init)
4524 cgroup_init_subsys(ss);
4526 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4529 /* Add init_css_set to the hash table */
4530 hhead = css_set_hash(init_css_set.subsys);
4531 hlist_add_head(&init_css_set.hlist, hhead);
4532 BUG_ON(!init_root_id(&rootnode));
4534 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4540 err = register_filesystem(&cgroup_fs_type);
4542 kobject_put(cgroup_kobj);
4546 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4550 bdi_destroy(&cgroup_backing_dev_info);
4556 * proc_cgroup_show()
4557 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4558 * - Used for /proc/<pid>/cgroup.
4559 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4560 * doesn't really matter if tsk->cgroup changes after we read it,
4561 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4562 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4563 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4564 * cgroup to top_cgroup.
4567 /* TODO: Use a proper seq_file iterator */
4568 static int proc_cgroup_show(struct seq_file *m, void *v)
4571 struct task_struct *tsk;
4574 struct cgroupfs_root *root;
4577 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4583 tsk = get_pid_task(pid, PIDTYPE_PID);
4589 mutex_lock(&cgroup_mutex);
4591 for_each_active_root(root) {
4592 struct cgroup_subsys *ss;
4593 struct cgroup *cgrp;
4596 seq_printf(m, "%d:", root->hierarchy_id);
4597 for_each_subsys(root, ss)
4598 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4599 if (strlen(root->name))
4600 seq_printf(m, "%sname=%s", count ? "," : "",
4603 cgrp = task_cgroup_from_root(tsk, root);
4604 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4612 mutex_unlock(&cgroup_mutex);
4613 put_task_struct(tsk);
4620 static int cgroup_open(struct inode *inode, struct file *file)
4622 struct pid *pid = PROC_I(inode)->pid;
4623 return single_open(file, proc_cgroup_show, pid);
4626 const struct file_operations proc_cgroup_operations = {
4627 .open = cgroup_open,
4629 .llseek = seq_lseek,
4630 .release = single_release,
4633 /* Display information about each subsystem and each hierarchy */
4634 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4638 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4640 * ideally we don't want subsystems moving around while we do this.
4641 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4642 * subsys/hierarchy state.
4644 mutex_lock(&cgroup_mutex);
4645 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4646 struct cgroup_subsys *ss = subsys[i];
4649 seq_printf(m, "%s\t%d\t%d\t%d\n",
4650 ss->name, ss->root->hierarchy_id,
4651 ss->root->number_of_cgroups, !ss->disabled);
4653 mutex_unlock(&cgroup_mutex);
4657 static int cgroupstats_open(struct inode *inode, struct file *file)
4659 return single_open(file, proc_cgroupstats_show, NULL);
4662 static const struct file_operations proc_cgroupstats_operations = {
4663 .open = cgroupstats_open,
4665 .llseek = seq_lseek,
4666 .release = single_release,
4670 * cgroup_fork - attach newly forked task to its parents cgroup.
4671 * @child: pointer to task_struct of forking parent process.
4673 * Description: A task inherits its parent's cgroup at fork().
4675 * A pointer to the shared css_set was automatically copied in
4676 * fork.c by dup_task_struct(). However, we ignore that copy, since
4677 * it was not made under the protection of RCU, cgroup_mutex or
4678 * threadgroup_change_begin(), so it might no longer be a valid
4679 * cgroup pointer. cgroup_attach_task() might have already changed
4680 * current->cgroups, allowing the previously referenced cgroup
4681 * group to be removed and freed.
4683 * Outside the pointer validity we also need to process the css_set
4684 * inheritance between threadgoup_change_begin() and
4685 * threadgoup_change_end(), this way there is no leak in any process
4686 * wide migration performed by cgroup_attach_proc() that could otherwise
4687 * miss a thread because it is too early or too late in the fork stage.
4689 * At the point that cgroup_fork() is called, 'current' is the parent
4690 * task, and the passed argument 'child' points to the child task.
4692 void cgroup_fork(struct task_struct *child)
4695 * We don't need to task_lock() current because current->cgroups
4696 * can't be changed concurrently here. The parent obviously hasn't
4697 * exited and called cgroup_exit(), and we are synchronized against
4698 * cgroup migration through threadgroup_change_begin().
4700 child->cgroups = current->cgroups;
4701 get_css_set(child->cgroups);
4702 INIT_LIST_HEAD(&child->cg_list);
4706 * cgroup_fork_callbacks - run fork callbacks
4707 * @child: the new task
4709 * Called on a new task very soon before adding it to the
4710 * tasklist. No need to take any locks since no-one can
4711 * be operating on this task.
4713 void cgroup_fork_callbacks(struct task_struct *child)
4715 if (need_forkexit_callback) {
4718 * forkexit callbacks are only supported for builtin
4719 * subsystems, and the builtin section of the subsys array is
4720 * immutable, so we don't need to lock the subsys array here.
4722 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4723 struct cgroup_subsys *ss = subsys[i];
4731 * cgroup_post_fork - called on a new task after adding it to the task list
4732 * @child: the task in question
4734 * Adds the task to the list running through its css_set if necessary.
4735 * Has to be after the task is visible on the task list in case we race
4736 * with the first call to cgroup_iter_start() - to guarantee that the
4737 * new task ends up on its list.
4739 void cgroup_post_fork(struct task_struct *child)
4742 * use_task_css_set_links is set to 1 before we walk the tasklist
4743 * under the tasklist_lock and we read it here after we added the child
4744 * to the tasklist under the tasklist_lock as well. If the child wasn't
4745 * yet in the tasklist when we walked through it from
4746 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4747 * should be visible now due to the paired locking and barriers implied
4748 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4749 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4752 if (use_task_css_set_links) {
4753 write_lock(&css_set_lock);
4754 if (list_empty(&child->cg_list)) {
4756 * It's safe to use child->cgroups without task_lock()
4757 * here because we are protected through
4758 * threadgroup_change_begin() against concurrent
4759 * css_set change in cgroup_task_migrate(). Also
4760 * the task can't exit at that point until
4761 * wake_up_new_task() is called, so we are protected
4762 * against cgroup_exit() setting child->cgroup to
4765 list_add(&child->cg_list, &child->cgroups->tasks);
4767 write_unlock(&css_set_lock);
4771 * cgroup_exit - detach cgroup from exiting task
4772 * @tsk: pointer to task_struct of exiting process
4773 * @run_callback: run exit callbacks?
4775 * Description: Detach cgroup from @tsk and release it.
4777 * Note that cgroups marked notify_on_release force every task in
4778 * them to take the global cgroup_mutex mutex when exiting.
4779 * This could impact scaling on very large systems. Be reluctant to
4780 * use notify_on_release cgroups where very high task exit scaling
4781 * is required on large systems.
4783 * the_top_cgroup_hack:
4785 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4787 * We call cgroup_exit() while the task is still competent to
4788 * handle notify_on_release(), then leave the task attached to the
4789 * root cgroup in each hierarchy for the remainder of its exit.
4791 * To do this properly, we would increment the reference count on
4792 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4793 * code we would add a second cgroup function call, to drop that
4794 * reference. This would just create an unnecessary hot spot on
4795 * the top_cgroup reference count, to no avail.
4797 * Normally, holding a reference to a cgroup without bumping its
4798 * count is unsafe. The cgroup could go away, or someone could
4799 * attach us to a different cgroup, decrementing the count on
4800 * the first cgroup that we never incremented. But in this case,
4801 * top_cgroup isn't going away, and either task has PF_EXITING set,
4802 * which wards off any cgroup_attach_task() attempts, or task is a failed
4803 * fork, never visible to cgroup_attach_task.
4805 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4811 * Unlink from the css_set task list if necessary.
4812 * Optimistically check cg_list before taking
4815 if (!list_empty(&tsk->cg_list)) {
4816 write_lock(&css_set_lock);
4817 if (!list_empty(&tsk->cg_list))
4818 list_del_init(&tsk->cg_list);
4819 write_unlock(&css_set_lock);
4822 /* Reassign the task to the init_css_set. */
4825 tsk->cgroups = &init_css_set;
4827 if (run_callbacks && need_forkexit_callback) {
4829 * modular subsystems can't use callbacks, so no need to lock
4832 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4833 struct cgroup_subsys *ss = subsys[i];
4835 struct cgroup *old_cgrp =
4836 rcu_dereference_raw(cg->subsys[i])->cgroup;
4837 struct cgroup *cgrp = task_cgroup(tsk, i);
4838 ss->exit(cgrp, old_cgrp, tsk);
4845 put_css_set_taskexit(cg);
4849 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4850 * @cgrp: the cgroup in question
4851 * @task: the task in question
4853 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4856 * If we are sending in dummytop, then presumably we are creating
4857 * the top cgroup in the subsystem.
4859 * Called only by the ns (nsproxy) cgroup.
4861 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4864 struct cgroup *target;
4866 if (cgrp == dummytop)
4869 target = task_cgroup_from_root(task, cgrp->root);
4870 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4871 cgrp = cgrp->parent;
4872 ret = (cgrp == target);
4876 static void check_for_release(struct cgroup *cgrp)
4878 /* All of these checks rely on RCU to keep the cgroup
4879 * structure alive */
4880 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4881 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4882 /* Control Group is currently removeable. If it's not
4883 * already queued for a userspace notification, queue
4885 int need_schedule_work = 0;
4886 raw_spin_lock(&release_list_lock);
4887 if (!cgroup_is_removed(cgrp) &&
4888 list_empty(&cgrp->release_list)) {
4889 list_add(&cgrp->release_list, &release_list);
4890 need_schedule_work = 1;
4892 raw_spin_unlock(&release_list_lock);
4893 if (need_schedule_work)
4894 schedule_work(&release_agent_work);
4898 /* Caller must verify that the css is not for root cgroup */
4899 bool __css_tryget(struct cgroup_subsys_state *css)
4902 int v = css_refcnt(css);
4904 if (atomic_cmpxchg(&css->refcnt, v, v + 1) == v)
4907 } while (!test_bit(CSS_REMOVED, &css->flags));
4911 EXPORT_SYMBOL_GPL(__css_tryget);
4913 /* Caller must verify that the css is not for root cgroup */
4914 void __css_put(struct cgroup_subsys_state *css)
4916 struct cgroup *cgrp = css->cgroup;
4919 atomic_dec(&css->refcnt);
4920 if (css_refcnt(css) == 1) {
4921 if (notify_on_release(cgrp)) {
4922 set_bit(CGRP_RELEASABLE, &cgrp->flags);
4923 check_for_release(cgrp);
4925 cgroup_wakeup_rmdir_waiter(cgrp);
4929 EXPORT_SYMBOL_GPL(__css_put);
4932 * Notify userspace when a cgroup is released, by running the
4933 * configured release agent with the name of the cgroup (path
4934 * relative to the root of cgroup file system) as the argument.
4936 * Most likely, this user command will try to rmdir this cgroup.
4938 * This races with the possibility that some other task will be
4939 * attached to this cgroup before it is removed, or that some other
4940 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4941 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4942 * unused, and this cgroup will be reprieved from its death sentence,
4943 * to continue to serve a useful existence. Next time it's released,
4944 * we will get notified again, if it still has 'notify_on_release' set.
4946 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4947 * means only wait until the task is successfully execve()'d. The
4948 * separate release agent task is forked by call_usermodehelper(),
4949 * then control in this thread returns here, without waiting for the
4950 * release agent task. We don't bother to wait because the caller of
4951 * this routine has no use for the exit status of the release agent
4952 * task, so no sense holding our caller up for that.
4954 static void cgroup_release_agent(struct work_struct *work)
4956 BUG_ON(work != &release_agent_work);
4957 mutex_lock(&cgroup_mutex);
4958 raw_spin_lock(&release_list_lock);
4959 while (!list_empty(&release_list)) {
4960 char *argv[3], *envp[3];
4962 char *pathbuf = NULL, *agentbuf = NULL;
4963 struct cgroup *cgrp = list_entry(release_list.next,
4966 list_del_init(&cgrp->release_list);
4967 raw_spin_unlock(&release_list_lock);
4968 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4971 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4973 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4978 argv[i++] = agentbuf;
4979 argv[i++] = pathbuf;
4983 /* minimal command environment */
4984 envp[i++] = "HOME=/";
4985 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4988 /* Drop the lock while we invoke the usermode helper,
4989 * since the exec could involve hitting disk and hence
4990 * be a slow process */
4991 mutex_unlock(&cgroup_mutex);
4992 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4993 mutex_lock(&cgroup_mutex);
4997 raw_spin_lock(&release_list_lock);
4999 raw_spin_unlock(&release_list_lock);
5000 mutex_unlock(&cgroup_mutex);
5003 static int __init cgroup_disable(char *str)
5008 while ((token = strsep(&str, ",")) != NULL) {
5012 * cgroup_disable, being at boot time, can't know about module
5013 * subsystems, so we don't worry about them.
5015 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
5016 struct cgroup_subsys *ss = subsys[i];
5018 if (!strcmp(token, ss->name)) {
5020 printk(KERN_INFO "Disabling %s control group"
5021 " subsystem\n", ss->name);
5028 __setup("cgroup_disable=", cgroup_disable);
5031 * Functons for CSS ID.
5035 *To get ID other than 0, this should be called when !cgroup_is_removed().
5037 unsigned short css_id(struct cgroup_subsys_state *css)
5039 struct css_id *cssid;
5042 * This css_id() can return correct value when somone has refcnt
5043 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5044 * it's unchanged until freed.
5046 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5052 EXPORT_SYMBOL_GPL(css_id);
5054 unsigned short css_depth(struct cgroup_subsys_state *css)
5056 struct css_id *cssid;
5058 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5061 return cssid->depth;
5064 EXPORT_SYMBOL_GPL(css_depth);
5067 * css_is_ancestor - test "root" css is an ancestor of "child"
5068 * @child: the css to be tested.
5069 * @root: the css supporsed to be an ancestor of the child.
5071 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5072 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
5073 * But, considering usual usage, the csses should be valid objects after test.
5074 * Assuming that the caller will do some action to the child if this returns
5075 * returns true, the caller must take "child";s reference count.
5076 * If "child" is valid object and this returns true, "root" is valid, too.
5079 bool css_is_ancestor(struct cgroup_subsys_state *child,
5080 const struct cgroup_subsys_state *root)
5082 struct css_id *child_id;
5083 struct css_id *root_id;
5087 child_id = rcu_dereference(child->id);
5088 root_id = rcu_dereference(root->id);
5091 || (child_id->depth < root_id->depth)
5092 || (child_id->stack[root_id->depth] != root_id->id))
5098 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5100 struct css_id *id = css->id;
5101 /* When this is called before css_id initialization, id can be NULL */
5105 BUG_ON(!ss->use_id);
5107 rcu_assign_pointer(id->css, NULL);
5108 rcu_assign_pointer(css->id, NULL);
5109 spin_lock(&ss->id_lock);
5110 idr_remove(&ss->idr, id->id);
5111 spin_unlock(&ss->id_lock);
5112 kfree_rcu(id, rcu_head);
5114 EXPORT_SYMBOL_GPL(free_css_id);
5117 * This is called by init or create(). Then, calls to this function are
5118 * always serialized (By cgroup_mutex() at create()).
5121 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5123 struct css_id *newid;
5124 int myid, error, size;
5126 BUG_ON(!ss->use_id);
5128 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5129 newid = kzalloc(size, GFP_KERNEL);
5131 return ERR_PTR(-ENOMEM);
5133 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
5137 spin_lock(&ss->id_lock);
5138 /* Don't use 0. allocates an ID of 1-65535 */
5139 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
5140 spin_unlock(&ss->id_lock);
5142 /* Returns error when there are no free spaces for new ID.*/
5147 if (myid > CSS_ID_MAX)
5151 newid->depth = depth;
5155 spin_lock(&ss->id_lock);
5156 idr_remove(&ss->idr, myid);
5157 spin_unlock(&ss->id_lock);
5160 return ERR_PTR(error);
5164 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5165 struct cgroup_subsys_state *rootcss)
5167 struct css_id *newid;
5169 spin_lock_init(&ss->id_lock);
5172 newid = get_new_cssid(ss, 0);
5174 return PTR_ERR(newid);
5176 newid->stack[0] = newid->id;
5177 newid->css = rootcss;
5178 rootcss->id = newid;
5182 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5183 struct cgroup *child)
5185 int subsys_id, i, depth = 0;
5186 struct cgroup_subsys_state *parent_css, *child_css;
5187 struct css_id *child_id, *parent_id;
5189 subsys_id = ss->subsys_id;
5190 parent_css = parent->subsys[subsys_id];
5191 child_css = child->subsys[subsys_id];
5192 parent_id = parent_css->id;
5193 depth = parent_id->depth + 1;
5195 child_id = get_new_cssid(ss, depth);
5196 if (IS_ERR(child_id))
5197 return PTR_ERR(child_id);
5199 for (i = 0; i < depth; i++)
5200 child_id->stack[i] = parent_id->stack[i];
5201 child_id->stack[depth] = child_id->id;
5203 * child_id->css pointer will be set after this cgroup is available
5204 * see cgroup_populate_dir()
5206 rcu_assign_pointer(child_css->id, child_id);
5212 * css_lookup - lookup css by id
5213 * @ss: cgroup subsys to be looked into.
5216 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5217 * NULL if not. Should be called under rcu_read_lock()
5219 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5221 struct css_id *cssid = NULL;
5223 BUG_ON(!ss->use_id);
5224 cssid = idr_find(&ss->idr, id);
5226 if (unlikely(!cssid))
5229 return rcu_dereference(cssid->css);
5231 EXPORT_SYMBOL_GPL(css_lookup);
5234 * css_get_next - lookup next cgroup under specified hierarchy.
5235 * @ss: pointer to subsystem
5236 * @id: current position of iteration.
5237 * @root: pointer to css. search tree under this.
5238 * @foundid: position of found object.
5240 * Search next css under the specified hierarchy of rootid. Calling under
5241 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5243 struct cgroup_subsys_state *
5244 css_get_next(struct cgroup_subsys *ss, int id,
5245 struct cgroup_subsys_state *root, int *foundid)
5247 struct cgroup_subsys_state *ret = NULL;
5250 int rootid = css_id(root);
5251 int depth = css_depth(root);
5256 BUG_ON(!ss->use_id);
5257 WARN_ON_ONCE(!rcu_read_lock_held());
5259 /* fill start point for scan */
5263 * scan next entry from bitmap(tree), tmpid is updated after
5266 tmp = idr_get_next(&ss->idr, &tmpid);
5269 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5270 ret = rcu_dereference(tmp->css);
5276 /* continue to scan from next id */
5283 * get corresponding css from file open on cgroupfs directory
5285 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5287 struct cgroup *cgrp;
5288 struct inode *inode;
5289 struct cgroup_subsys_state *css;
5291 inode = f->f_dentry->d_inode;
5292 /* check in cgroup filesystem dir */
5293 if (inode->i_op != &cgroup_dir_inode_operations)
5294 return ERR_PTR(-EBADF);
5296 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5297 return ERR_PTR(-EINVAL);
5300 cgrp = __d_cgrp(f->f_dentry);
5301 css = cgrp->subsys[id];
5302 return css ? css : ERR_PTR(-ENOENT);
5305 #ifdef CONFIG_CGROUP_DEBUG
5306 static struct cgroup_subsys_state *debug_create(struct cgroup *cont)
5308 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5311 return ERR_PTR(-ENOMEM);
5316 static void debug_destroy(struct cgroup *cont)
5318 kfree(cont->subsys[debug_subsys_id]);
5321 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5323 return atomic_read(&cont->count);
5326 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5328 return cgroup_task_count(cont);
5331 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5333 return (u64)(unsigned long)current->cgroups;
5336 static u64 current_css_set_refcount_read(struct cgroup *cont,
5342 count = atomic_read(¤t->cgroups->refcount);
5347 static int current_css_set_cg_links_read(struct cgroup *cont,
5349 struct seq_file *seq)
5351 struct cg_cgroup_link *link;
5354 read_lock(&css_set_lock);
5356 cg = rcu_dereference(current->cgroups);
5357 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5358 struct cgroup *c = link->cgrp;
5362 name = c->dentry->d_name.name;
5365 seq_printf(seq, "Root %d group %s\n",
5366 c->root->hierarchy_id, name);
5369 read_unlock(&css_set_lock);
5373 #define MAX_TASKS_SHOWN_PER_CSS 25
5374 static int cgroup_css_links_read(struct cgroup *cont,
5376 struct seq_file *seq)
5378 struct cg_cgroup_link *link;
5380 read_lock(&css_set_lock);
5381 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5382 struct css_set *cg = link->cg;
5383 struct task_struct *task;
5385 seq_printf(seq, "css_set %p\n", cg);
5386 list_for_each_entry(task, &cg->tasks, cg_list) {
5387 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5388 seq_puts(seq, " ...\n");
5391 seq_printf(seq, " task %d\n",
5392 task_pid_vnr(task));
5396 read_unlock(&css_set_lock);
5400 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5402 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5405 static struct cftype debug_files[] = {
5407 .name = "cgroup_refcount",
5408 .read_u64 = cgroup_refcount_read,
5411 .name = "taskcount",
5412 .read_u64 = debug_taskcount_read,
5416 .name = "current_css_set",
5417 .read_u64 = current_css_set_read,
5421 .name = "current_css_set_refcount",
5422 .read_u64 = current_css_set_refcount_read,
5426 .name = "current_css_set_cg_links",
5427 .read_seq_string = current_css_set_cg_links_read,
5431 .name = "cgroup_css_links",
5432 .read_seq_string = cgroup_css_links_read,
5436 .name = "releasable",
5437 .read_u64 = releasable_read,
5443 struct cgroup_subsys debug_subsys = {
5445 .create = debug_create,
5446 .destroy = debug_destroy,
5447 .subsys_id = debug_subsys_id,
5448 .base_cftypes = debug_files,
5450 #endif /* CONFIG_CGROUP_DEBUG */