2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Copyright notices from the original cpuset code:
8 * --------------------------------------------------
9 * Copyright (C) 2003 BULL SA.
10 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
12 * Portions derived from Patrick Mochel's sysfs code.
13 * sysfs is Copyright (c) 2001-3 Patrick Mochel
15 * 2003-10-10 Written by Simon Derr.
16 * 2003-10-22 Updates by Stephen Hemminger.
17 * 2004 May-July Rework by Paul Jackson.
18 * ---------------------------------------------------
20 * This file is subject to the terms and conditions of the GNU General Public
21 * License. See the file COPYING in the main directory of the Linux
22 * distribution for more details.
25 #include <linux/cgroup.h>
26 #include <linux/errno.h>
28 #include <linux/kernel.h>
29 #include <linux/list.h>
31 #include <linux/mutex.h>
32 #include <linux/mount.h>
33 #include <linux/pagemap.h>
34 #include <linux/proc_fs.h>
35 #include <linux/rcupdate.h>
36 #include <linux/sched.h>
37 #include <linux/backing-dev.h>
38 #include <linux/seq_file.h>
39 #include <linux/slab.h>
40 #include <linux/magic.h>
41 #include <linux/spinlock.h>
42 #include <linux/string.h>
43 #include <linux/sort.h>
44 #include <linux/kmod.h>
45 #include <linux/delayacct.h>
46 #include <linux/cgroupstats.h>
47 #include <linux/hash.h>
48 #include <linux/namei.h>
50 #include <asm/atomic.h>
52 static DEFINE_MUTEX(cgroup_mutex);
54 /* Generate an array of cgroup subsystem pointers */
55 #define SUBSYS(_x) &_x ## _subsys,
57 static struct cgroup_subsys *subsys[] = {
58 #include <linux/cgroup_subsys.h>
62 * A cgroupfs_root represents the root of a cgroup hierarchy,
63 * and may be associated with a superblock to form an active
66 struct cgroupfs_root {
67 struct super_block *sb;
70 * The bitmask of subsystems intended to be attached to this
73 unsigned long subsys_bits;
75 /* The bitmask of subsystems currently attached to this hierarchy */
76 unsigned long actual_subsys_bits;
78 /* A list running through the attached subsystems */
79 struct list_head subsys_list;
81 /* The root cgroup for this hierarchy */
82 struct cgroup top_cgroup;
84 /* Tracks how many cgroups are currently defined in hierarchy.*/
85 int number_of_cgroups;
87 /* A list running through the active hierarchies */
88 struct list_head root_list;
90 /* Hierarchy-specific flags */
93 /* The path to use for release notifications. */
94 char release_agent_path[PATH_MAX];
98 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
99 * subsystems that are otherwise unattached - it never has more than a
100 * single cgroup, and all tasks are part of that cgroup.
102 static struct cgroupfs_root rootnode;
105 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
106 * cgroup_subsys->use_id != 0.
108 #define CSS_ID_MAX (65535)
111 * The css to which this ID points. This pointer is set to valid value
112 * after cgroup is populated. If cgroup is removed, this will be NULL.
113 * This pointer is expected to be RCU-safe because destroy()
114 * is called after synchronize_rcu(). But for safe use, css_is_removed()
115 * css_tryget() should be used for avoiding race.
117 struct cgroup_subsys_state *css;
123 * Depth in hierarchy which this ID belongs to.
125 unsigned short depth;
127 * ID is freed by RCU. (and lookup routine is RCU safe.)
129 struct rcu_head rcu_head;
131 * Hierarchy of CSS ID belongs to.
133 unsigned short stack[0]; /* Array of Length (depth+1) */
137 /* The list of hierarchy roots */
139 static LIST_HEAD(roots);
140 static int root_count;
142 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
143 #define dummytop (&rootnode.top_cgroup)
145 /* This flag indicates whether tasks in the fork and exit paths should
146 * check for fork/exit handlers to call. This avoids us having to do
147 * extra work in the fork/exit path if none of the subsystems need to
150 static int need_forkexit_callback __read_mostly;
152 /* convenient tests for these bits */
153 inline int cgroup_is_removed(const struct cgroup *cgrp)
155 return test_bit(CGRP_REMOVED, &cgrp->flags);
158 /* bits in struct cgroupfs_root flags field */
160 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
163 static int cgroup_is_releasable(const struct cgroup *cgrp)
166 (1 << CGRP_RELEASABLE) |
167 (1 << CGRP_NOTIFY_ON_RELEASE);
168 return (cgrp->flags & bits) == bits;
171 static int notify_on_release(const struct cgroup *cgrp)
173 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
177 * for_each_subsys() allows you to iterate on each subsystem attached to
178 * an active hierarchy
180 #define for_each_subsys(_root, _ss) \
181 list_for_each_entry(_ss, &_root->subsys_list, sibling)
183 /* for_each_active_root() allows you to iterate across the active hierarchies */
184 #define for_each_active_root(_root) \
185 list_for_each_entry(_root, &roots, root_list)
187 /* the list of cgroups eligible for automatic release. Protected by
188 * release_list_lock */
189 static LIST_HEAD(release_list);
190 static DEFINE_SPINLOCK(release_list_lock);
191 static void cgroup_release_agent(struct work_struct *work);
192 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
193 static void check_for_release(struct cgroup *cgrp);
195 /* Link structure for associating css_set objects with cgroups */
196 struct cg_cgroup_link {
198 * List running through cg_cgroup_links associated with a
199 * cgroup, anchored on cgroup->css_sets
201 struct list_head cgrp_link_list;
203 * List running through cg_cgroup_links pointing at a
204 * single css_set object, anchored on css_set->cg_links
206 struct list_head cg_link_list;
210 /* The default css_set - used by init and its children prior to any
211 * hierarchies being mounted. It contains a pointer to the root state
212 * for each subsystem. Also used to anchor the list of css_sets. Not
213 * reference-counted, to improve performance when child cgroups
214 * haven't been created.
217 static struct css_set init_css_set;
218 static struct cg_cgroup_link init_css_set_link;
220 static int cgroup_subsys_init_idr(struct cgroup_subsys *ss);
222 /* css_set_lock protects the list of css_set objects, and the
223 * chain of tasks off each css_set. Nests outside task->alloc_lock
224 * due to cgroup_iter_start() */
225 static DEFINE_RWLOCK(css_set_lock);
226 static int css_set_count;
228 /* hash table for cgroup groups. This improves the performance to
229 * find an existing css_set */
230 #define CSS_SET_HASH_BITS 7
231 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
232 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
234 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
238 unsigned long tmp = 0UL;
240 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
241 tmp += (unsigned long)css[i];
242 tmp = (tmp >> 16) ^ tmp;
244 index = hash_long(tmp, CSS_SET_HASH_BITS);
246 return &css_set_table[index];
249 /* We don't maintain the lists running through each css_set to its
250 * task until after the first call to cgroup_iter_start(). This
251 * reduces the fork()/exit() overhead for people who have cgroups
252 * compiled into their kernel but not actually in use */
253 static int use_task_css_set_links __read_mostly;
255 /* When we create or destroy a css_set, the operation simply
256 * takes/releases a reference count on all the cgroups referenced
257 * by subsystems in this css_set. This can end up multiple-counting
258 * some cgroups, but that's OK - the ref-count is just a
259 * busy/not-busy indicator; ensuring that we only count each cgroup
260 * once would require taking a global lock to ensure that no
261 * subsystems moved between hierarchies while we were doing so.
263 * Possible TODO: decide at boot time based on the number of
264 * registered subsystems and the number of CPUs or NUMA nodes whether
265 * it's better for performance to ref-count every subsystem, or to
266 * take a global lock and only add one ref count to each hierarchy.
270 * unlink a css_set from the list and free it
272 static void unlink_css_set(struct css_set *cg)
274 struct cg_cgroup_link *link;
275 struct cg_cgroup_link *saved_link;
277 hlist_del(&cg->hlist);
280 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
282 list_del(&link->cg_link_list);
283 list_del(&link->cgrp_link_list);
288 static void __put_css_set(struct css_set *cg, int taskexit)
292 * Ensure that the refcount doesn't hit zero while any readers
293 * can see it. Similar to atomic_dec_and_lock(), but for an
296 if (atomic_add_unless(&cg->refcount, -1, 1))
298 write_lock(&css_set_lock);
299 if (!atomic_dec_and_test(&cg->refcount)) {
300 write_unlock(&css_set_lock);
304 write_unlock(&css_set_lock);
307 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
308 struct cgroup *cgrp = rcu_dereference(cg->subsys[i]->cgroup);
309 if (atomic_dec_and_test(&cgrp->count) &&
310 notify_on_release(cgrp)) {
312 set_bit(CGRP_RELEASABLE, &cgrp->flags);
313 check_for_release(cgrp);
321 * refcounted get/put for css_set objects
323 static inline void get_css_set(struct css_set *cg)
325 atomic_inc(&cg->refcount);
328 static inline void put_css_set(struct css_set *cg)
330 __put_css_set(cg, 0);
333 static inline void put_css_set_taskexit(struct css_set *cg)
335 __put_css_set(cg, 1);
339 * find_existing_css_set() is a helper for
340 * find_css_set(), and checks to see whether an existing
341 * css_set is suitable.
343 * oldcg: the cgroup group that we're using before the cgroup
346 * cgrp: the cgroup that we're moving into
348 * template: location in which to build the desired set of subsystem
349 * state objects for the new cgroup group
351 static struct css_set *find_existing_css_set(
352 struct css_set *oldcg,
354 struct cgroup_subsys_state *template[])
357 struct cgroupfs_root *root = cgrp->root;
358 struct hlist_head *hhead;
359 struct hlist_node *node;
362 /* Built the set of subsystem state objects that we want to
363 * see in the new css_set */
364 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
365 if (root->subsys_bits & (1UL << i)) {
366 /* Subsystem is in this hierarchy. So we want
367 * the subsystem state from the new
369 template[i] = cgrp->subsys[i];
371 /* Subsystem is not in this hierarchy, so we
372 * don't want to change the subsystem state */
373 template[i] = oldcg->subsys[i];
377 hhead = css_set_hash(template);
378 hlist_for_each_entry(cg, node, hhead, hlist) {
379 if (!memcmp(template, cg->subsys, sizeof(cg->subsys))) {
380 /* All subsystems matched */
385 /* No existing cgroup group matched */
389 static void free_cg_links(struct list_head *tmp)
391 struct cg_cgroup_link *link;
392 struct cg_cgroup_link *saved_link;
394 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
395 list_del(&link->cgrp_link_list);
401 * allocate_cg_links() allocates "count" cg_cgroup_link structures
402 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
403 * success or a negative error
405 static int allocate_cg_links(int count, struct list_head *tmp)
407 struct cg_cgroup_link *link;
410 for (i = 0; i < count; i++) {
411 link = kmalloc(sizeof(*link), GFP_KERNEL);
416 list_add(&link->cgrp_link_list, tmp);
422 * link_css_set - a helper function to link a css_set to a cgroup
423 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
424 * @cg: the css_set to be linked
425 * @cgrp: the destination cgroup
427 static void link_css_set(struct list_head *tmp_cg_links,
428 struct css_set *cg, struct cgroup *cgrp)
430 struct cg_cgroup_link *link;
432 BUG_ON(list_empty(tmp_cg_links));
433 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
436 list_move(&link->cgrp_link_list, &cgrp->css_sets);
437 list_add(&link->cg_link_list, &cg->cg_links);
441 * find_css_set() takes an existing cgroup group and a
442 * cgroup object, and returns a css_set object that's
443 * equivalent to the old group, but with the given cgroup
444 * substituted into the appropriate hierarchy. Must be called with
447 static struct css_set *find_css_set(
448 struct css_set *oldcg, struct cgroup *cgrp)
451 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
454 struct list_head tmp_cg_links;
456 struct hlist_head *hhead;
458 /* First see if we already have a cgroup group that matches
460 read_lock(&css_set_lock);
461 res = find_existing_css_set(oldcg, cgrp, template);
464 read_unlock(&css_set_lock);
469 res = kmalloc(sizeof(*res), GFP_KERNEL);
473 /* Allocate all the cg_cgroup_link objects that we'll need */
474 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
479 atomic_set(&res->refcount, 1);
480 INIT_LIST_HEAD(&res->cg_links);
481 INIT_LIST_HEAD(&res->tasks);
482 INIT_HLIST_NODE(&res->hlist);
484 /* Copy the set of subsystem state objects generated in
485 * find_existing_css_set() */
486 memcpy(res->subsys, template, sizeof(res->subsys));
488 write_lock(&css_set_lock);
489 /* Add reference counts and links from the new css_set. */
490 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
491 struct cgroup *cgrp = res->subsys[i]->cgroup;
492 struct cgroup_subsys *ss = subsys[i];
493 atomic_inc(&cgrp->count);
495 * We want to add a link once per cgroup, so we
496 * only do it for the first subsystem in each
499 if (ss->root->subsys_list.next == &ss->sibling)
500 link_css_set(&tmp_cg_links, res, cgrp);
502 if (list_empty(&rootnode.subsys_list))
503 link_css_set(&tmp_cg_links, res, dummytop);
505 BUG_ON(!list_empty(&tmp_cg_links));
509 /* Add this cgroup group to the hash table */
510 hhead = css_set_hash(res->subsys);
511 hlist_add_head(&res->hlist, hhead);
513 write_unlock(&css_set_lock);
519 * There is one global cgroup mutex. We also require taking
520 * task_lock() when dereferencing a task's cgroup subsys pointers.
521 * See "The task_lock() exception", at the end of this comment.
523 * A task must hold cgroup_mutex to modify cgroups.
525 * Any task can increment and decrement the count field without lock.
526 * So in general, code holding cgroup_mutex can't rely on the count
527 * field not changing. However, if the count goes to zero, then only
528 * cgroup_attach_task() can increment it again. Because a count of zero
529 * means that no tasks are currently attached, therefore there is no
530 * way a task attached to that cgroup can fork (the other way to
531 * increment the count). So code holding cgroup_mutex can safely
532 * assume that if the count is zero, it will stay zero. Similarly, if
533 * a task holds cgroup_mutex on a cgroup with zero count, it
534 * knows that the cgroup won't be removed, as cgroup_rmdir()
537 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
538 * (usually) take cgroup_mutex. These are the two most performance
539 * critical pieces of code here. The exception occurs on cgroup_exit(),
540 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
541 * is taken, and if the cgroup count is zero, a usermode call made
542 * to the release agent with the name of the cgroup (path relative to
543 * the root of cgroup file system) as the argument.
545 * A cgroup can only be deleted if both its 'count' of using tasks
546 * is zero, and its list of 'children' cgroups is empty. Since all
547 * tasks in the system use _some_ cgroup, and since there is always at
548 * least one task in the system (init, pid == 1), therefore, top_cgroup
549 * always has either children cgroups and/or using tasks. So we don't
550 * need a special hack to ensure that top_cgroup cannot be deleted.
552 * The task_lock() exception
554 * The need for this exception arises from the action of
555 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
556 * another. It does so using cgroup_mutex, however there are
557 * several performance critical places that need to reference
558 * task->cgroup without the expense of grabbing a system global
559 * mutex. Therefore except as noted below, when dereferencing or, as
560 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
561 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
562 * the task_struct routinely used for such matters.
564 * P.S. One more locking exception. RCU is used to guard the
565 * update of a tasks cgroup pointer by cgroup_attach_task()
569 * cgroup_lock - lock out any changes to cgroup structures
572 void cgroup_lock(void)
574 mutex_lock(&cgroup_mutex);
578 * cgroup_unlock - release lock on cgroup changes
580 * Undo the lock taken in a previous cgroup_lock() call.
582 void cgroup_unlock(void)
584 mutex_unlock(&cgroup_mutex);
588 * A couple of forward declarations required, due to cyclic reference loop:
589 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
590 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
594 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
595 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
596 static int cgroup_populate_dir(struct cgroup *cgrp);
597 static struct inode_operations cgroup_dir_inode_operations;
598 static struct file_operations proc_cgroupstats_operations;
600 static struct backing_dev_info cgroup_backing_dev_info = {
601 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
604 static int alloc_css_id(struct cgroup_subsys *ss,
605 struct cgroup *parent, struct cgroup *child);
607 static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
609 struct inode *inode = new_inode(sb);
612 inode->i_mode = mode;
613 inode->i_uid = current_fsuid();
614 inode->i_gid = current_fsgid();
615 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
616 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
622 * Call subsys's pre_destroy handler.
623 * This is called before css refcnt check.
625 static void cgroup_call_pre_destroy(struct cgroup *cgrp)
627 struct cgroup_subsys *ss;
628 for_each_subsys(cgrp->root, ss)
630 ss->pre_destroy(ss, cgrp);
634 static void free_cgroup_rcu(struct rcu_head *obj)
636 struct cgroup *cgrp = container_of(obj, struct cgroup, rcu_head);
641 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
643 /* is dentry a directory ? if so, kfree() associated cgroup */
644 if (S_ISDIR(inode->i_mode)) {
645 struct cgroup *cgrp = dentry->d_fsdata;
646 struct cgroup_subsys *ss;
647 BUG_ON(!(cgroup_is_removed(cgrp)));
648 /* It's possible for external users to be holding css
649 * reference counts on a cgroup; css_put() needs to
650 * be able to access the cgroup after decrementing
651 * the reference count in order to know if it needs to
652 * queue the cgroup to be handled by the release
656 mutex_lock(&cgroup_mutex);
658 * Release the subsystem state objects.
660 for_each_subsys(cgrp->root, ss)
661 ss->destroy(ss, cgrp);
663 cgrp->root->number_of_cgroups--;
664 mutex_unlock(&cgroup_mutex);
667 * Drop the active superblock reference that we took when we
670 deactivate_super(cgrp->root->sb);
672 call_rcu(&cgrp->rcu_head, free_cgroup_rcu);
677 static void remove_dir(struct dentry *d)
679 struct dentry *parent = dget(d->d_parent);
682 simple_rmdir(parent->d_inode, d);
686 static void cgroup_clear_directory(struct dentry *dentry)
688 struct list_head *node;
690 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
691 spin_lock(&dcache_lock);
692 node = dentry->d_subdirs.next;
693 while (node != &dentry->d_subdirs) {
694 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
697 /* This should never be called on a cgroup
698 * directory with child cgroups */
699 BUG_ON(d->d_inode->i_mode & S_IFDIR);
701 spin_unlock(&dcache_lock);
703 simple_unlink(dentry->d_inode, d);
705 spin_lock(&dcache_lock);
707 node = dentry->d_subdirs.next;
709 spin_unlock(&dcache_lock);
713 * NOTE : the dentry must have been dget()'ed
715 static void cgroup_d_remove_dir(struct dentry *dentry)
717 cgroup_clear_directory(dentry);
719 spin_lock(&dcache_lock);
720 list_del_init(&dentry->d_u.d_child);
721 spin_unlock(&dcache_lock);
725 static int rebind_subsystems(struct cgroupfs_root *root,
726 unsigned long final_bits)
728 unsigned long added_bits, removed_bits;
729 struct cgroup *cgrp = &root->top_cgroup;
732 removed_bits = root->actual_subsys_bits & ~final_bits;
733 added_bits = final_bits & ~root->actual_subsys_bits;
734 /* Check that any added subsystems are currently free */
735 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
736 unsigned long bit = 1UL << i;
737 struct cgroup_subsys *ss = subsys[i];
738 if (!(bit & added_bits))
740 if (ss->root != &rootnode) {
741 /* Subsystem isn't free */
746 /* Currently we don't handle adding/removing subsystems when
747 * any child cgroups exist. This is theoretically supportable
748 * but involves complex error handling, so it's being left until
750 if (root->number_of_cgroups > 1)
753 /* Process each subsystem */
754 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
755 struct cgroup_subsys *ss = subsys[i];
756 unsigned long bit = 1UL << i;
757 if (bit & added_bits) {
758 /* We're binding this subsystem to this hierarchy */
759 BUG_ON(cgrp->subsys[i]);
760 BUG_ON(!dummytop->subsys[i]);
761 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
762 mutex_lock(&ss->hierarchy_mutex);
763 cgrp->subsys[i] = dummytop->subsys[i];
764 cgrp->subsys[i]->cgroup = cgrp;
765 list_move(&ss->sibling, &root->subsys_list);
769 mutex_unlock(&ss->hierarchy_mutex);
770 } else if (bit & removed_bits) {
771 /* We're removing this subsystem */
772 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
773 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
774 mutex_lock(&ss->hierarchy_mutex);
776 ss->bind(ss, dummytop);
777 dummytop->subsys[i]->cgroup = dummytop;
778 cgrp->subsys[i] = NULL;
779 subsys[i]->root = &rootnode;
780 list_move(&ss->sibling, &rootnode.subsys_list);
781 mutex_unlock(&ss->hierarchy_mutex);
782 } else if (bit & final_bits) {
783 /* Subsystem state should already exist */
784 BUG_ON(!cgrp->subsys[i]);
786 /* Subsystem state shouldn't exist */
787 BUG_ON(cgrp->subsys[i]);
790 root->subsys_bits = root->actual_subsys_bits = final_bits;
796 static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
798 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
799 struct cgroup_subsys *ss;
801 mutex_lock(&cgroup_mutex);
802 for_each_subsys(root, ss)
803 seq_printf(seq, ",%s", ss->name);
804 if (test_bit(ROOT_NOPREFIX, &root->flags))
805 seq_puts(seq, ",noprefix");
806 if (strlen(root->release_agent_path))
807 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
808 mutex_unlock(&cgroup_mutex);
812 struct cgroup_sb_opts {
813 unsigned long subsys_bits;
818 /* Convert a hierarchy specifier into a bitmask of subsystems and
820 static int parse_cgroupfs_options(char *data,
821 struct cgroup_sb_opts *opts)
823 char *token, *o = data ?: "all";
825 opts->subsys_bits = 0;
827 opts->release_agent = NULL;
829 while ((token = strsep(&o, ",")) != NULL) {
832 if (!strcmp(token, "all")) {
833 /* Add all non-disabled subsystems */
835 opts->subsys_bits = 0;
836 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
837 struct cgroup_subsys *ss = subsys[i];
839 opts->subsys_bits |= 1ul << i;
841 } else if (!strcmp(token, "noprefix")) {
842 set_bit(ROOT_NOPREFIX, &opts->flags);
843 } else if (!strncmp(token, "release_agent=", 14)) {
844 /* Specifying two release agents is forbidden */
845 if (opts->release_agent)
847 opts->release_agent = kzalloc(PATH_MAX, GFP_KERNEL);
848 if (!opts->release_agent)
850 strncpy(opts->release_agent, token + 14, PATH_MAX - 1);
851 opts->release_agent[PATH_MAX - 1] = 0;
853 struct cgroup_subsys *ss;
855 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
857 if (!strcmp(token, ss->name)) {
859 set_bit(i, &opts->subsys_bits);
863 if (i == CGROUP_SUBSYS_COUNT)
868 /* We can't have an empty hierarchy */
869 if (!opts->subsys_bits)
875 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
878 struct cgroupfs_root *root = sb->s_fs_info;
879 struct cgroup *cgrp = &root->top_cgroup;
880 struct cgroup_sb_opts opts;
882 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
883 mutex_lock(&cgroup_mutex);
885 /* See what subsystems are wanted */
886 ret = parse_cgroupfs_options(data, &opts);
890 /* Don't allow flags to change at remount */
891 if (opts.flags != root->flags) {
896 ret = rebind_subsystems(root, opts.subsys_bits);
898 /* (re)populate subsystem files */
900 cgroup_populate_dir(cgrp);
902 if (opts.release_agent)
903 strcpy(root->release_agent_path, opts.release_agent);
905 if (opts.release_agent)
906 kfree(opts.release_agent);
907 mutex_unlock(&cgroup_mutex);
908 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
912 static struct super_operations cgroup_ops = {
913 .statfs = simple_statfs,
914 .drop_inode = generic_delete_inode,
915 .show_options = cgroup_show_options,
916 .remount_fs = cgroup_remount,
919 static void init_cgroup_housekeeping(struct cgroup *cgrp)
921 INIT_LIST_HEAD(&cgrp->sibling);
922 INIT_LIST_HEAD(&cgrp->children);
923 INIT_LIST_HEAD(&cgrp->css_sets);
924 INIT_LIST_HEAD(&cgrp->release_list);
925 init_rwsem(&cgrp->pids_mutex);
927 static void init_cgroup_root(struct cgroupfs_root *root)
929 struct cgroup *cgrp = &root->top_cgroup;
930 INIT_LIST_HEAD(&root->subsys_list);
931 INIT_LIST_HEAD(&root->root_list);
932 root->number_of_cgroups = 1;
934 cgrp->top_cgroup = cgrp;
935 init_cgroup_housekeeping(cgrp);
938 static int cgroup_test_super(struct super_block *sb, void *data)
940 struct cgroupfs_root *new = data;
941 struct cgroupfs_root *root = sb->s_fs_info;
943 /* First check subsystems */
944 if (new->subsys_bits != root->subsys_bits)
947 /* Next check flags */
948 if (new->flags != root->flags)
954 static int cgroup_set_super(struct super_block *sb, void *data)
957 struct cgroupfs_root *root = data;
959 ret = set_anon_super(sb, NULL);
963 sb->s_fs_info = root;
966 sb->s_blocksize = PAGE_CACHE_SIZE;
967 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
968 sb->s_magic = CGROUP_SUPER_MAGIC;
969 sb->s_op = &cgroup_ops;
974 static int cgroup_get_rootdir(struct super_block *sb)
976 struct inode *inode =
977 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
978 struct dentry *dentry;
983 inode->i_fop = &simple_dir_operations;
984 inode->i_op = &cgroup_dir_inode_operations;
985 /* directories start off with i_nlink == 2 (for "." entry) */
987 dentry = d_alloc_root(inode);
996 static int cgroup_get_sb(struct file_system_type *fs_type,
997 int flags, const char *unused_dev_name,
998 void *data, struct vfsmount *mnt)
1000 struct cgroup_sb_opts opts;
1002 struct super_block *sb;
1003 struct cgroupfs_root *root;
1004 struct list_head tmp_cg_links;
1006 /* First find the desired set of subsystems */
1007 ret = parse_cgroupfs_options(data, &opts);
1009 if (opts.release_agent)
1010 kfree(opts.release_agent);
1014 root = kzalloc(sizeof(*root), GFP_KERNEL);
1016 if (opts.release_agent)
1017 kfree(opts.release_agent);
1021 init_cgroup_root(root);
1022 root->subsys_bits = opts.subsys_bits;
1023 root->flags = opts.flags;
1024 if (opts.release_agent) {
1025 strcpy(root->release_agent_path, opts.release_agent);
1026 kfree(opts.release_agent);
1029 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, root);
1036 if (sb->s_fs_info != root) {
1037 /* Reusing an existing superblock */
1038 BUG_ON(sb->s_root == NULL);
1042 /* New superblock */
1043 struct cgroup *root_cgrp = &root->top_cgroup;
1044 struct inode *inode;
1047 BUG_ON(sb->s_root != NULL);
1049 ret = cgroup_get_rootdir(sb);
1051 goto drop_new_super;
1052 inode = sb->s_root->d_inode;
1054 mutex_lock(&inode->i_mutex);
1055 mutex_lock(&cgroup_mutex);
1058 * We're accessing css_set_count without locking
1059 * css_set_lock here, but that's OK - it can only be
1060 * increased by someone holding cgroup_lock, and
1061 * that's us. The worst that can happen is that we
1062 * have some link structures left over
1064 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1066 mutex_unlock(&cgroup_mutex);
1067 mutex_unlock(&inode->i_mutex);
1068 goto drop_new_super;
1071 ret = rebind_subsystems(root, root->subsys_bits);
1072 if (ret == -EBUSY) {
1073 mutex_unlock(&cgroup_mutex);
1074 mutex_unlock(&inode->i_mutex);
1078 /* EBUSY should be the only error here */
1081 list_add(&root->root_list, &roots);
1084 sb->s_root->d_fsdata = root_cgrp;
1085 root->top_cgroup.dentry = sb->s_root;
1087 /* Link the top cgroup in this hierarchy into all
1088 * the css_set objects */
1089 write_lock(&css_set_lock);
1090 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1091 struct hlist_head *hhead = &css_set_table[i];
1092 struct hlist_node *node;
1095 hlist_for_each_entry(cg, node, hhead, hlist)
1096 link_css_set(&tmp_cg_links, cg, root_cgrp);
1098 write_unlock(&css_set_lock);
1100 free_cg_links(&tmp_cg_links);
1102 BUG_ON(!list_empty(&root_cgrp->sibling));
1103 BUG_ON(!list_empty(&root_cgrp->children));
1104 BUG_ON(root->number_of_cgroups != 1);
1106 cgroup_populate_dir(root_cgrp);
1107 mutex_unlock(&inode->i_mutex);
1108 mutex_unlock(&cgroup_mutex);
1111 simple_set_mnt(mnt, sb);
1115 free_cg_links(&tmp_cg_links);
1117 up_write(&sb->s_umount);
1118 deactivate_super(sb);
1122 static void cgroup_kill_sb(struct super_block *sb) {
1123 struct cgroupfs_root *root = sb->s_fs_info;
1124 struct cgroup *cgrp = &root->top_cgroup;
1126 struct cg_cgroup_link *link;
1127 struct cg_cgroup_link *saved_link;
1131 BUG_ON(root->number_of_cgroups != 1);
1132 BUG_ON(!list_empty(&cgrp->children));
1133 BUG_ON(!list_empty(&cgrp->sibling));
1135 mutex_lock(&cgroup_mutex);
1137 /* Rebind all subsystems back to the default hierarchy */
1138 ret = rebind_subsystems(root, 0);
1139 /* Shouldn't be able to fail ... */
1143 * Release all the links from css_sets to this hierarchy's
1146 write_lock(&css_set_lock);
1148 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1150 list_del(&link->cg_link_list);
1151 list_del(&link->cgrp_link_list);
1154 write_unlock(&css_set_lock);
1156 if (!list_empty(&root->root_list)) {
1157 list_del(&root->root_list);
1161 mutex_unlock(&cgroup_mutex);
1163 kill_litter_super(sb);
1167 static struct file_system_type cgroup_fs_type = {
1169 .get_sb = cgroup_get_sb,
1170 .kill_sb = cgroup_kill_sb,
1173 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1175 return dentry->d_fsdata;
1178 static inline struct cftype *__d_cft(struct dentry *dentry)
1180 return dentry->d_fsdata;
1184 * cgroup_path - generate the path of a cgroup
1185 * @cgrp: the cgroup in question
1186 * @buf: the buffer to write the path into
1187 * @buflen: the length of the buffer
1189 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1190 * reference. Writes path of cgroup into buf. Returns 0 on success,
1193 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1196 struct dentry *dentry = rcu_dereference(cgrp->dentry);
1198 if (!dentry || cgrp == dummytop) {
1200 * Inactive subsystems have no dentry for their root
1207 start = buf + buflen;
1211 int len = dentry->d_name.len;
1212 if ((start -= len) < buf)
1213 return -ENAMETOOLONG;
1214 memcpy(start, cgrp->dentry->d_name.name, len);
1215 cgrp = cgrp->parent;
1218 dentry = rcu_dereference(cgrp->dentry);
1222 return -ENAMETOOLONG;
1225 memmove(buf, start, buf + buflen - start);
1230 * Return the first subsystem attached to a cgroup's hierarchy, and
1234 static void get_first_subsys(const struct cgroup *cgrp,
1235 struct cgroup_subsys_state **css, int *subsys_id)
1237 const struct cgroupfs_root *root = cgrp->root;
1238 const struct cgroup_subsys *test_ss;
1239 BUG_ON(list_empty(&root->subsys_list));
1240 test_ss = list_entry(root->subsys_list.next,
1241 struct cgroup_subsys, sibling);
1243 *css = cgrp->subsys[test_ss->subsys_id];
1247 *subsys_id = test_ss->subsys_id;
1251 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1252 * @cgrp: the cgroup the task is attaching to
1253 * @tsk: the task to be attached
1255 * Call holding cgroup_mutex. May take task_lock of
1256 * the task 'tsk' during call.
1258 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1261 struct cgroup_subsys *ss;
1262 struct cgroup *oldcgrp;
1264 struct css_set *newcg;
1265 struct cgroupfs_root *root = cgrp->root;
1268 get_first_subsys(cgrp, NULL, &subsys_id);
1270 /* Nothing to do if the task is already in that cgroup */
1271 oldcgrp = task_cgroup(tsk, subsys_id);
1272 if (cgrp == oldcgrp)
1275 for_each_subsys(root, ss) {
1276 if (ss->can_attach) {
1277 retval = ss->can_attach(ss, cgrp, tsk);
1288 * Locate or allocate a new css_set for this task,
1289 * based on its final set of cgroups
1291 newcg = find_css_set(cg, cgrp);
1297 if (tsk->flags & PF_EXITING) {
1302 rcu_assign_pointer(tsk->cgroups, newcg);
1305 /* Update the css_set linked lists if we're using them */
1306 write_lock(&css_set_lock);
1307 if (!list_empty(&tsk->cg_list)) {
1308 list_del(&tsk->cg_list);
1309 list_add(&tsk->cg_list, &newcg->tasks);
1311 write_unlock(&css_set_lock);
1313 for_each_subsys(root, ss) {
1315 ss->attach(ss, cgrp, oldcgrp, tsk);
1317 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1324 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1325 * held. May take task_lock of task
1327 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid)
1329 struct task_struct *tsk;
1330 const struct cred *cred = current_cred(), *tcred;
1335 tsk = find_task_by_vpid(pid);
1336 if (!tsk || tsk->flags & PF_EXITING) {
1341 tcred = __task_cred(tsk);
1343 cred->euid != tcred->uid &&
1344 cred->euid != tcred->suid) {
1348 get_task_struct(tsk);
1352 get_task_struct(tsk);
1355 ret = cgroup_attach_task(cgrp, tsk);
1356 put_task_struct(tsk);
1360 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
1363 if (!cgroup_lock_live_group(cgrp))
1365 ret = attach_task_by_pid(cgrp, pid);
1370 /* The various types of files and directories in a cgroup file system */
1371 enum cgroup_filetype {
1375 FILE_NOTIFY_ON_RELEASE,
1380 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1381 * @cgrp: the cgroup to be checked for liveness
1383 * On success, returns true; the lock should be later released with
1384 * cgroup_unlock(). On failure returns false with no lock held.
1386 bool cgroup_lock_live_group(struct cgroup *cgrp)
1388 mutex_lock(&cgroup_mutex);
1389 if (cgroup_is_removed(cgrp)) {
1390 mutex_unlock(&cgroup_mutex);
1396 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
1399 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1400 if (!cgroup_lock_live_group(cgrp))
1402 strcpy(cgrp->root->release_agent_path, buffer);
1407 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
1408 struct seq_file *seq)
1410 if (!cgroup_lock_live_group(cgrp))
1412 seq_puts(seq, cgrp->root->release_agent_path);
1413 seq_putc(seq, '\n');
1418 /* A buffer size big enough for numbers or short strings */
1419 #define CGROUP_LOCAL_BUFFER_SIZE 64
1421 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
1423 const char __user *userbuf,
1424 size_t nbytes, loff_t *unused_ppos)
1426 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
1432 if (nbytes >= sizeof(buffer))
1434 if (copy_from_user(buffer, userbuf, nbytes))
1437 buffer[nbytes] = 0; /* nul-terminate */
1439 if (cft->write_u64) {
1440 u64 val = simple_strtoull(buffer, &end, 0);
1443 retval = cft->write_u64(cgrp, cft, val);
1445 s64 val = simple_strtoll(buffer, &end, 0);
1448 retval = cft->write_s64(cgrp, cft, val);
1455 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
1457 const char __user *userbuf,
1458 size_t nbytes, loff_t *unused_ppos)
1460 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
1462 size_t max_bytes = cft->max_write_len;
1463 char *buffer = local_buffer;
1466 max_bytes = sizeof(local_buffer) - 1;
1467 if (nbytes >= max_bytes)
1469 /* Allocate a dynamic buffer if we need one */
1470 if (nbytes >= sizeof(local_buffer)) {
1471 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1475 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
1480 buffer[nbytes] = 0; /* nul-terminate */
1482 retval = cft->write_string(cgrp, cft, buffer);
1486 if (buffer != local_buffer)
1491 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
1492 size_t nbytes, loff_t *ppos)
1494 struct cftype *cft = __d_cft(file->f_dentry);
1495 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1497 if (cgroup_is_removed(cgrp))
1500 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
1501 if (cft->write_u64 || cft->write_s64)
1502 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
1503 if (cft->write_string)
1504 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
1506 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
1507 return ret ? ret : nbytes;
1512 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
1514 char __user *buf, size_t nbytes,
1517 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
1518 u64 val = cft->read_u64(cgrp, cft);
1519 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
1521 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1524 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
1526 char __user *buf, size_t nbytes,
1529 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
1530 s64 val = cft->read_s64(cgrp, cft);
1531 int len = sprintf(tmp, "%lld\n", (long long) val);
1533 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1536 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
1537 size_t nbytes, loff_t *ppos)
1539 struct cftype *cft = __d_cft(file->f_dentry);
1540 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1542 if (cgroup_is_removed(cgrp))
1546 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
1548 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
1550 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
1555 * seqfile ops/methods for returning structured data. Currently just
1556 * supports string->u64 maps, but can be extended in future.
1559 struct cgroup_seqfile_state {
1561 struct cgroup *cgroup;
1564 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
1566 struct seq_file *sf = cb->state;
1567 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
1570 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
1572 struct cgroup_seqfile_state *state = m->private;
1573 struct cftype *cft = state->cft;
1574 if (cft->read_map) {
1575 struct cgroup_map_cb cb = {
1576 .fill = cgroup_map_add,
1579 return cft->read_map(state->cgroup, cft, &cb);
1581 return cft->read_seq_string(state->cgroup, cft, m);
1584 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
1586 struct seq_file *seq = file->private_data;
1587 kfree(seq->private);
1588 return single_release(inode, file);
1591 static struct file_operations cgroup_seqfile_operations = {
1593 .write = cgroup_file_write,
1594 .llseek = seq_lseek,
1595 .release = cgroup_seqfile_release,
1598 static int cgroup_file_open(struct inode *inode, struct file *file)
1603 err = generic_file_open(inode, file);
1606 cft = __d_cft(file->f_dentry);
1608 if (cft->read_map || cft->read_seq_string) {
1609 struct cgroup_seqfile_state *state =
1610 kzalloc(sizeof(*state), GFP_USER);
1614 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
1615 file->f_op = &cgroup_seqfile_operations;
1616 err = single_open(file, cgroup_seqfile_show, state);
1619 } else if (cft->open)
1620 err = cft->open(inode, file);
1627 static int cgroup_file_release(struct inode *inode, struct file *file)
1629 struct cftype *cft = __d_cft(file->f_dentry);
1631 return cft->release(inode, file);
1636 * cgroup_rename - Only allow simple rename of directories in place.
1638 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
1639 struct inode *new_dir, struct dentry *new_dentry)
1641 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1643 if (new_dentry->d_inode)
1645 if (old_dir != new_dir)
1647 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
1650 static struct file_operations cgroup_file_operations = {
1651 .read = cgroup_file_read,
1652 .write = cgroup_file_write,
1653 .llseek = generic_file_llseek,
1654 .open = cgroup_file_open,
1655 .release = cgroup_file_release,
1658 static struct inode_operations cgroup_dir_inode_operations = {
1659 .lookup = simple_lookup,
1660 .mkdir = cgroup_mkdir,
1661 .rmdir = cgroup_rmdir,
1662 .rename = cgroup_rename,
1665 static int cgroup_create_file(struct dentry *dentry, int mode,
1666 struct super_block *sb)
1668 static const struct dentry_operations cgroup_dops = {
1669 .d_iput = cgroup_diput,
1672 struct inode *inode;
1676 if (dentry->d_inode)
1679 inode = cgroup_new_inode(mode, sb);
1683 if (S_ISDIR(mode)) {
1684 inode->i_op = &cgroup_dir_inode_operations;
1685 inode->i_fop = &simple_dir_operations;
1687 /* start off with i_nlink == 2 (for "." entry) */
1690 /* start with the directory inode held, so that we can
1691 * populate it without racing with another mkdir */
1692 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1693 } else if (S_ISREG(mode)) {
1695 inode->i_fop = &cgroup_file_operations;
1697 dentry->d_op = &cgroup_dops;
1698 d_instantiate(dentry, inode);
1699 dget(dentry); /* Extra count - pin the dentry in core */
1704 * cgroup_create_dir - create a directory for an object.
1705 * @cgrp: the cgroup we create the directory for. It must have a valid
1706 * ->parent field. And we are going to fill its ->dentry field.
1707 * @dentry: dentry of the new cgroup
1708 * @mode: mode to set on new directory.
1710 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
1713 struct dentry *parent;
1716 parent = cgrp->parent->dentry;
1717 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
1719 dentry->d_fsdata = cgrp;
1720 inc_nlink(parent->d_inode);
1721 rcu_assign_pointer(cgrp->dentry, dentry);
1729 int cgroup_add_file(struct cgroup *cgrp,
1730 struct cgroup_subsys *subsys,
1731 const struct cftype *cft)
1733 struct dentry *dir = cgrp->dentry;
1734 struct dentry *dentry;
1737 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
1738 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
1739 strcpy(name, subsys->name);
1742 strcat(name, cft->name);
1743 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
1744 dentry = lookup_one_len(name, dir, strlen(name));
1745 if (!IS_ERR(dentry)) {
1746 error = cgroup_create_file(dentry, 0644 | S_IFREG,
1749 dentry->d_fsdata = (void *)cft;
1752 error = PTR_ERR(dentry);
1756 int cgroup_add_files(struct cgroup *cgrp,
1757 struct cgroup_subsys *subsys,
1758 const struct cftype cft[],
1762 for (i = 0; i < count; i++) {
1763 err = cgroup_add_file(cgrp, subsys, &cft[i]);
1771 * cgroup_task_count - count the number of tasks in a cgroup.
1772 * @cgrp: the cgroup in question
1774 * Return the number of tasks in the cgroup.
1776 int cgroup_task_count(const struct cgroup *cgrp)
1779 struct cg_cgroup_link *link;
1781 read_lock(&css_set_lock);
1782 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
1783 count += atomic_read(&link->cg->refcount);
1785 read_unlock(&css_set_lock);
1790 * Advance a list_head iterator. The iterator should be positioned at
1791 * the start of a css_set
1793 static void cgroup_advance_iter(struct cgroup *cgrp,
1794 struct cgroup_iter *it)
1796 struct list_head *l = it->cg_link;
1797 struct cg_cgroup_link *link;
1800 /* Advance to the next non-empty css_set */
1803 if (l == &cgrp->css_sets) {
1807 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
1809 } while (list_empty(&cg->tasks));
1811 it->task = cg->tasks.next;
1815 * To reduce the fork() overhead for systems that are not actually
1816 * using their cgroups capability, we don't maintain the lists running
1817 * through each css_set to its tasks until we see the list actually
1818 * used - in other words after the first call to cgroup_iter_start().
1820 * The tasklist_lock is not held here, as do_each_thread() and
1821 * while_each_thread() are protected by RCU.
1823 static void cgroup_enable_task_cg_lists(void)
1825 struct task_struct *p, *g;
1826 write_lock(&css_set_lock);
1827 use_task_css_set_links = 1;
1828 do_each_thread(g, p) {
1831 * We should check if the process is exiting, otherwise
1832 * it will race with cgroup_exit() in that the list
1833 * entry won't be deleted though the process has exited.
1835 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
1836 list_add(&p->cg_list, &p->cgroups->tasks);
1838 } while_each_thread(g, p);
1839 write_unlock(&css_set_lock);
1842 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
1845 * The first time anyone tries to iterate across a cgroup,
1846 * we need to enable the list linking each css_set to its
1847 * tasks, and fix up all existing tasks.
1849 if (!use_task_css_set_links)
1850 cgroup_enable_task_cg_lists();
1852 read_lock(&css_set_lock);
1853 it->cg_link = &cgrp->css_sets;
1854 cgroup_advance_iter(cgrp, it);
1857 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
1858 struct cgroup_iter *it)
1860 struct task_struct *res;
1861 struct list_head *l = it->task;
1862 struct cg_cgroup_link *link;
1864 /* If the iterator cg is NULL, we have no tasks */
1867 res = list_entry(l, struct task_struct, cg_list);
1868 /* Advance iterator to find next entry */
1870 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
1871 if (l == &link->cg->tasks) {
1872 /* We reached the end of this task list - move on to
1873 * the next cg_cgroup_link */
1874 cgroup_advance_iter(cgrp, it);
1881 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
1883 read_unlock(&css_set_lock);
1886 static inline int started_after_time(struct task_struct *t1,
1887 struct timespec *time,
1888 struct task_struct *t2)
1890 int start_diff = timespec_compare(&t1->start_time, time);
1891 if (start_diff > 0) {
1893 } else if (start_diff < 0) {
1897 * Arbitrarily, if two processes started at the same
1898 * time, we'll say that the lower pointer value
1899 * started first. Note that t2 may have exited by now
1900 * so this may not be a valid pointer any longer, but
1901 * that's fine - it still serves to distinguish
1902 * between two tasks started (effectively) simultaneously.
1909 * This function is a callback from heap_insert() and is used to order
1911 * In this case we order the heap in descending task start time.
1913 static inline int started_after(void *p1, void *p2)
1915 struct task_struct *t1 = p1;
1916 struct task_struct *t2 = p2;
1917 return started_after_time(t1, &t2->start_time, t2);
1921 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
1922 * @scan: struct cgroup_scanner containing arguments for the scan
1924 * Arguments include pointers to callback functions test_task() and
1926 * Iterate through all the tasks in a cgroup, calling test_task() for each,
1927 * and if it returns true, call process_task() for it also.
1928 * The test_task pointer may be NULL, meaning always true (select all tasks).
1929 * Effectively duplicates cgroup_iter_{start,next,end}()
1930 * but does not lock css_set_lock for the call to process_task().
1931 * The struct cgroup_scanner may be embedded in any structure of the caller's
1933 * It is guaranteed that process_task() will act on every task that
1934 * is a member of the cgroup for the duration of this call. This
1935 * function may or may not call process_task() for tasks that exit
1936 * or move to a different cgroup during the call, or are forked or
1937 * move into the cgroup during the call.
1939 * Note that test_task() may be called with locks held, and may in some
1940 * situations be called multiple times for the same task, so it should
1942 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
1943 * pre-allocated and will be used for heap operations (and its "gt" member will
1944 * be overwritten), else a temporary heap will be used (allocation of which
1945 * may cause this function to fail).
1947 int cgroup_scan_tasks(struct cgroup_scanner *scan)
1950 struct cgroup_iter it;
1951 struct task_struct *p, *dropped;
1952 /* Never dereference latest_task, since it's not refcounted */
1953 struct task_struct *latest_task = NULL;
1954 struct ptr_heap tmp_heap;
1955 struct ptr_heap *heap;
1956 struct timespec latest_time = { 0, 0 };
1959 /* The caller supplied our heap and pre-allocated its memory */
1961 heap->gt = &started_after;
1963 /* We need to allocate our own heap memory */
1965 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
1967 /* cannot allocate the heap */
1973 * Scan tasks in the cgroup, using the scanner's "test_task" callback
1974 * to determine which are of interest, and using the scanner's
1975 * "process_task" callback to process any of them that need an update.
1976 * Since we don't want to hold any locks during the task updates,
1977 * gather tasks to be processed in a heap structure.
1978 * The heap is sorted by descending task start time.
1979 * If the statically-sized heap fills up, we overflow tasks that
1980 * started later, and in future iterations only consider tasks that
1981 * started after the latest task in the previous pass. This
1982 * guarantees forward progress and that we don't miss any tasks.
1985 cgroup_iter_start(scan->cg, &it);
1986 while ((p = cgroup_iter_next(scan->cg, &it))) {
1988 * Only affect tasks that qualify per the caller's callback,
1989 * if he provided one
1991 if (scan->test_task && !scan->test_task(p, scan))
1994 * Only process tasks that started after the last task
1997 if (!started_after_time(p, &latest_time, latest_task))
1999 dropped = heap_insert(heap, p);
2000 if (dropped == NULL) {
2002 * The new task was inserted; the heap wasn't
2006 } else if (dropped != p) {
2008 * The new task was inserted, and pushed out a
2012 put_task_struct(dropped);
2015 * Else the new task was newer than anything already in
2016 * the heap and wasn't inserted
2019 cgroup_iter_end(scan->cg, &it);
2022 for (i = 0; i < heap->size; i++) {
2023 struct task_struct *q = heap->ptrs[i];
2025 latest_time = q->start_time;
2028 /* Process the task per the caller's callback */
2029 scan->process_task(q, scan);
2033 * If we had to process any tasks at all, scan again
2034 * in case some of them were in the middle of forking
2035 * children that didn't get processed.
2036 * Not the most efficient way to do it, but it avoids
2037 * having to take callback_mutex in the fork path
2041 if (heap == &tmp_heap)
2042 heap_free(&tmp_heap);
2047 * Stuff for reading the 'tasks' file.
2049 * Reading this file can return large amounts of data if a cgroup has
2050 * *lots* of attached tasks. So it may need several calls to read(),
2051 * but we cannot guarantee that the information we produce is correct
2052 * unless we produce it entirely atomically.
2057 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
2058 * 'cgrp'. Return actual number of pids loaded. No need to
2059 * task_lock(p) when reading out p->cgroup, since we're in an RCU
2060 * read section, so the css_set can't go away, and is
2061 * immutable after creation.
2063 static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cgrp)
2066 struct cgroup_iter it;
2067 struct task_struct *tsk;
2068 cgroup_iter_start(cgrp, &it);
2069 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2070 if (unlikely(n == npids))
2072 pid = task_pid_vnr(tsk);
2074 pidarray[n++] = pid;
2076 cgroup_iter_end(cgrp, &it);
2081 * cgroupstats_build - build and fill cgroupstats
2082 * @stats: cgroupstats to fill information into
2083 * @dentry: A dentry entry belonging to the cgroup for which stats have
2086 * Build and fill cgroupstats so that taskstats can export it to user
2089 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2092 struct cgroup *cgrp;
2093 struct cgroup_iter it;
2094 struct task_struct *tsk;
2097 * Validate dentry by checking the superblock operations,
2098 * and make sure it's a directory.
2100 if (dentry->d_sb->s_op != &cgroup_ops ||
2101 !S_ISDIR(dentry->d_inode->i_mode))
2105 cgrp = dentry->d_fsdata;
2107 cgroup_iter_start(cgrp, &it);
2108 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2109 switch (tsk->state) {
2111 stats->nr_running++;
2113 case TASK_INTERRUPTIBLE:
2114 stats->nr_sleeping++;
2116 case TASK_UNINTERRUPTIBLE:
2117 stats->nr_uninterruptible++;
2120 stats->nr_stopped++;
2123 if (delayacct_is_task_waiting_on_io(tsk))
2124 stats->nr_io_wait++;
2128 cgroup_iter_end(cgrp, &it);
2134 static int cmppid(const void *a, const void *b)
2136 return *(pid_t *)a - *(pid_t *)b;
2141 * seq_file methods for the "tasks" file. The seq_file position is the
2142 * next pid to display; the seq_file iterator is a pointer to the pid
2143 * in the cgroup->tasks_pids array.
2146 static void *cgroup_tasks_start(struct seq_file *s, loff_t *pos)
2149 * Initially we receive a position value that corresponds to
2150 * one more than the last pid shown (or 0 on the first call or
2151 * after a seek to the start). Use a binary-search to find the
2152 * next pid to display, if any
2154 struct cgroup *cgrp = s->private;
2155 int index = 0, pid = *pos;
2158 down_read(&cgrp->pids_mutex);
2160 int end = cgrp->pids_length;
2162 while (index < end) {
2163 int mid = (index + end) / 2;
2164 if (cgrp->tasks_pids[mid] == pid) {
2167 } else if (cgrp->tasks_pids[mid] <= pid)
2173 /* If we're off the end of the array, we're done */
2174 if (index >= cgrp->pids_length)
2176 /* Update the abstract position to be the actual pid that we found */
2177 iter = cgrp->tasks_pids + index;
2182 static void cgroup_tasks_stop(struct seq_file *s, void *v)
2184 struct cgroup *cgrp = s->private;
2185 up_read(&cgrp->pids_mutex);
2188 static void *cgroup_tasks_next(struct seq_file *s, void *v, loff_t *pos)
2190 struct cgroup *cgrp = s->private;
2192 int *end = cgrp->tasks_pids + cgrp->pids_length;
2195 * Advance to the next pid in the array. If this goes off the
2207 static int cgroup_tasks_show(struct seq_file *s, void *v)
2209 return seq_printf(s, "%d\n", *(int *)v);
2212 static struct seq_operations cgroup_tasks_seq_operations = {
2213 .start = cgroup_tasks_start,
2214 .stop = cgroup_tasks_stop,
2215 .next = cgroup_tasks_next,
2216 .show = cgroup_tasks_show,
2219 static void release_cgroup_pid_array(struct cgroup *cgrp)
2221 down_write(&cgrp->pids_mutex);
2222 BUG_ON(!cgrp->pids_use_count);
2223 if (!--cgrp->pids_use_count) {
2224 kfree(cgrp->tasks_pids);
2225 cgrp->tasks_pids = NULL;
2226 cgrp->pids_length = 0;
2228 up_write(&cgrp->pids_mutex);
2231 static int cgroup_tasks_release(struct inode *inode, struct file *file)
2233 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2235 if (!(file->f_mode & FMODE_READ))
2238 release_cgroup_pid_array(cgrp);
2239 return seq_release(inode, file);
2242 static struct file_operations cgroup_tasks_operations = {
2244 .llseek = seq_lseek,
2245 .write = cgroup_file_write,
2246 .release = cgroup_tasks_release,
2250 * Handle an open on 'tasks' file. Prepare an array containing the
2251 * process id's of tasks currently attached to the cgroup being opened.
2254 static int cgroup_tasks_open(struct inode *unused, struct file *file)
2256 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2261 /* Nothing to do for write-only files */
2262 if (!(file->f_mode & FMODE_READ))
2266 * If cgroup gets more users after we read count, we won't have
2267 * enough space - tough. This race is indistinguishable to the
2268 * caller from the case that the additional cgroup users didn't
2269 * show up until sometime later on.
2271 npids = cgroup_task_count(cgrp);
2272 pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL);
2275 npids = pid_array_load(pidarray, npids, cgrp);
2276 sort(pidarray, npids, sizeof(pid_t), cmppid, NULL);
2279 * Store the array in the cgroup, freeing the old
2280 * array if necessary
2282 down_write(&cgrp->pids_mutex);
2283 kfree(cgrp->tasks_pids);
2284 cgrp->tasks_pids = pidarray;
2285 cgrp->pids_length = npids;
2286 cgrp->pids_use_count++;
2287 up_write(&cgrp->pids_mutex);
2289 file->f_op = &cgroup_tasks_operations;
2291 retval = seq_open(file, &cgroup_tasks_seq_operations);
2293 release_cgroup_pid_array(cgrp);
2296 ((struct seq_file *)file->private_data)->private = cgrp;
2300 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
2303 return notify_on_release(cgrp);
2306 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
2310 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
2312 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2314 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2319 * for the common functions, 'private' gives the type of file
2321 static struct cftype files[] = {
2324 .open = cgroup_tasks_open,
2325 .write_u64 = cgroup_tasks_write,
2326 .release = cgroup_tasks_release,
2327 .private = FILE_TASKLIST,
2331 .name = "notify_on_release",
2332 .read_u64 = cgroup_read_notify_on_release,
2333 .write_u64 = cgroup_write_notify_on_release,
2334 .private = FILE_NOTIFY_ON_RELEASE,
2338 static struct cftype cft_release_agent = {
2339 .name = "release_agent",
2340 .read_seq_string = cgroup_release_agent_show,
2341 .write_string = cgroup_release_agent_write,
2342 .max_write_len = PATH_MAX,
2343 .private = FILE_RELEASE_AGENT,
2346 static int cgroup_populate_dir(struct cgroup *cgrp)
2349 struct cgroup_subsys *ss;
2351 /* First clear out any existing files */
2352 cgroup_clear_directory(cgrp->dentry);
2354 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
2358 if (cgrp == cgrp->top_cgroup) {
2359 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
2363 for_each_subsys(cgrp->root, ss) {
2364 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
2367 /* This cgroup is ready now */
2368 for_each_subsys(cgrp->root, ss) {
2369 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
2371 * Update id->css pointer and make this css visible from
2372 * CSS ID functions. This pointer will be dereferened
2373 * from RCU-read-side without locks.
2376 rcu_assign_pointer(css->id->css, css);
2382 static void init_cgroup_css(struct cgroup_subsys_state *css,
2383 struct cgroup_subsys *ss,
2384 struct cgroup *cgrp)
2387 atomic_set(&css->refcnt, 1);
2390 if (cgrp == dummytop)
2391 set_bit(CSS_ROOT, &css->flags);
2392 BUG_ON(cgrp->subsys[ss->subsys_id]);
2393 cgrp->subsys[ss->subsys_id] = css;
2396 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
2398 /* We need to take each hierarchy_mutex in a consistent order */
2401 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2402 struct cgroup_subsys *ss = subsys[i];
2403 if (ss->root == root)
2404 mutex_lock(&ss->hierarchy_mutex);
2408 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
2412 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2413 struct cgroup_subsys *ss = subsys[i];
2414 if (ss->root == root)
2415 mutex_unlock(&ss->hierarchy_mutex);
2420 * cgroup_create - create a cgroup
2421 * @parent: cgroup that will be parent of the new cgroup
2422 * @dentry: dentry of the new cgroup
2423 * @mode: mode to set on new inode
2425 * Must be called with the mutex on the parent inode held
2427 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
2430 struct cgroup *cgrp;
2431 struct cgroupfs_root *root = parent->root;
2433 struct cgroup_subsys *ss;
2434 struct super_block *sb = root->sb;
2436 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
2440 /* Grab a reference on the superblock so the hierarchy doesn't
2441 * get deleted on unmount if there are child cgroups. This
2442 * can be done outside cgroup_mutex, since the sb can't
2443 * disappear while someone has an open control file on the
2445 atomic_inc(&sb->s_active);
2447 mutex_lock(&cgroup_mutex);
2449 init_cgroup_housekeeping(cgrp);
2451 cgrp->parent = parent;
2452 cgrp->root = parent->root;
2453 cgrp->top_cgroup = parent->top_cgroup;
2455 if (notify_on_release(parent))
2456 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2458 for_each_subsys(root, ss) {
2459 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
2464 init_cgroup_css(css, ss, cgrp);
2466 if (alloc_css_id(ss, parent, cgrp))
2468 /* At error, ->destroy() callback has to free assigned ID. */
2471 cgroup_lock_hierarchy(root);
2472 list_add(&cgrp->sibling, &cgrp->parent->children);
2473 cgroup_unlock_hierarchy(root);
2474 root->number_of_cgroups++;
2476 err = cgroup_create_dir(cgrp, dentry, mode);
2480 /* The cgroup directory was pre-locked for us */
2481 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
2483 err = cgroup_populate_dir(cgrp);
2484 /* If err < 0, we have a half-filled directory - oh well ;) */
2486 mutex_unlock(&cgroup_mutex);
2487 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
2493 cgroup_lock_hierarchy(root);
2494 list_del(&cgrp->sibling);
2495 cgroup_unlock_hierarchy(root);
2496 root->number_of_cgroups--;
2500 for_each_subsys(root, ss) {
2501 if (cgrp->subsys[ss->subsys_id])
2502 ss->destroy(ss, cgrp);
2505 mutex_unlock(&cgroup_mutex);
2507 /* Release the reference count that we took on the superblock */
2508 deactivate_super(sb);
2514 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2516 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
2518 /* the vfs holds inode->i_mutex already */
2519 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
2522 static int cgroup_has_css_refs(struct cgroup *cgrp)
2524 /* Check the reference count on each subsystem. Since we
2525 * already established that there are no tasks in the
2526 * cgroup, if the css refcount is also 1, then there should
2527 * be no outstanding references, so the subsystem is safe to
2528 * destroy. We scan across all subsystems rather than using
2529 * the per-hierarchy linked list of mounted subsystems since
2530 * we can be called via check_for_release() with no
2531 * synchronization other than RCU, and the subsystem linked
2532 * list isn't RCU-safe */
2534 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2535 struct cgroup_subsys *ss = subsys[i];
2536 struct cgroup_subsys_state *css;
2537 /* Skip subsystems not in this hierarchy */
2538 if (ss->root != cgrp->root)
2540 css = cgrp->subsys[ss->subsys_id];
2541 /* When called from check_for_release() it's possible
2542 * that by this point the cgroup has been removed
2543 * and the css deleted. But a false-positive doesn't
2544 * matter, since it can only happen if the cgroup
2545 * has been deleted and hence no longer needs the
2546 * release agent to be called anyway. */
2547 if (css && (atomic_read(&css->refcnt) > 1))
2554 * Atomically mark all (or else none) of the cgroup's CSS objects as
2555 * CSS_REMOVED. Return true on success, or false if the cgroup has
2556 * busy subsystems. Call with cgroup_mutex held
2559 static int cgroup_clear_css_refs(struct cgroup *cgrp)
2561 struct cgroup_subsys *ss;
2562 unsigned long flags;
2563 bool failed = false;
2564 local_irq_save(flags);
2565 for_each_subsys(cgrp->root, ss) {
2566 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
2569 /* We can only remove a CSS with a refcnt==1 */
2570 refcnt = atomic_read(&css->refcnt);
2577 * Drop the refcnt to 0 while we check other
2578 * subsystems. This will cause any racing
2579 * css_tryget() to spin until we set the
2580 * CSS_REMOVED bits or abort
2582 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
2588 for_each_subsys(cgrp->root, ss) {
2589 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
2592 * Restore old refcnt if we previously managed
2593 * to clear it from 1 to 0
2595 if (!atomic_read(&css->refcnt))
2596 atomic_set(&css->refcnt, 1);
2598 /* Commit the fact that the CSS is removed */
2599 set_bit(CSS_REMOVED, &css->flags);
2602 local_irq_restore(flags);
2606 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
2608 struct cgroup *cgrp = dentry->d_fsdata;
2610 struct cgroup *parent;
2612 /* the vfs holds both inode->i_mutex already */
2614 mutex_lock(&cgroup_mutex);
2615 if (atomic_read(&cgrp->count) != 0) {
2616 mutex_unlock(&cgroup_mutex);
2619 if (!list_empty(&cgrp->children)) {
2620 mutex_unlock(&cgroup_mutex);
2623 mutex_unlock(&cgroup_mutex);
2626 * Call pre_destroy handlers of subsys. Notify subsystems
2627 * that rmdir() request comes.
2629 cgroup_call_pre_destroy(cgrp);
2631 mutex_lock(&cgroup_mutex);
2632 parent = cgrp->parent;
2634 if (atomic_read(&cgrp->count)
2635 || !list_empty(&cgrp->children)
2636 || !cgroup_clear_css_refs(cgrp)) {
2637 mutex_unlock(&cgroup_mutex);
2641 spin_lock(&release_list_lock);
2642 set_bit(CGRP_REMOVED, &cgrp->flags);
2643 if (!list_empty(&cgrp->release_list))
2644 list_del(&cgrp->release_list);
2645 spin_unlock(&release_list_lock);
2647 cgroup_lock_hierarchy(cgrp->root);
2648 /* delete this cgroup from parent->children */
2649 list_del(&cgrp->sibling);
2650 cgroup_unlock_hierarchy(cgrp->root);
2652 spin_lock(&cgrp->dentry->d_lock);
2653 d = dget(cgrp->dentry);
2654 spin_unlock(&d->d_lock);
2656 cgroup_d_remove_dir(d);
2659 set_bit(CGRP_RELEASABLE, &parent->flags);
2660 check_for_release(parent);
2662 mutex_unlock(&cgroup_mutex);
2666 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
2668 struct cgroup_subsys_state *css;
2670 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
2672 /* Create the top cgroup state for this subsystem */
2673 list_add(&ss->sibling, &rootnode.subsys_list);
2674 ss->root = &rootnode;
2675 css = ss->create(ss, dummytop);
2676 /* We don't handle early failures gracefully */
2677 BUG_ON(IS_ERR(css));
2678 init_cgroup_css(css, ss, dummytop);
2680 /* Update the init_css_set to contain a subsys
2681 * pointer to this state - since the subsystem is
2682 * newly registered, all tasks and hence the
2683 * init_css_set is in the subsystem's top cgroup. */
2684 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
2686 need_forkexit_callback |= ss->fork || ss->exit;
2688 /* At system boot, before all subsystems have been
2689 * registered, no tasks have been forked, so we don't
2690 * need to invoke fork callbacks here. */
2691 BUG_ON(!list_empty(&init_task.tasks));
2693 mutex_init(&ss->hierarchy_mutex);
2694 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
2699 * cgroup_init_early - cgroup initialization at system boot
2701 * Initialize cgroups at system boot, and initialize any
2702 * subsystems that request early init.
2704 int __init cgroup_init_early(void)
2707 atomic_set(&init_css_set.refcount, 1);
2708 INIT_LIST_HEAD(&init_css_set.cg_links);
2709 INIT_LIST_HEAD(&init_css_set.tasks);
2710 INIT_HLIST_NODE(&init_css_set.hlist);
2712 init_cgroup_root(&rootnode);
2714 init_task.cgroups = &init_css_set;
2716 init_css_set_link.cg = &init_css_set;
2717 list_add(&init_css_set_link.cgrp_link_list,
2718 &rootnode.top_cgroup.css_sets);
2719 list_add(&init_css_set_link.cg_link_list,
2720 &init_css_set.cg_links);
2722 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
2723 INIT_HLIST_HEAD(&css_set_table[i]);
2725 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2726 struct cgroup_subsys *ss = subsys[i];
2729 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
2730 BUG_ON(!ss->create);
2731 BUG_ON(!ss->destroy);
2732 if (ss->subsys_id != i) {
2733 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
2734 ss->name, ss->subsys_id);
2739 cgroup_init_subsys(ss);
2745 * cgroup_init - cgroup initialization
2747 * Register cgroup filesystem and /proc file, and initialize
2748 * any subsystems that didn't request early init.
2750 int __init cgroup_init(void)
2754 struct hlist_head *hhead;
2756 err = bdi_init(&cgroup_backing_dev_info);
2760 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2761 struct cgroup_subsys *ss = subsys[i];
2762 if (!ss->early_init)
2763 cgroup_init_subsys(ss);
2765 cgroup_subsys_init_idr(ss);
2768 /* Add init_css_set to the hash table */
2769 hhead = css_set_hash(init_css_set.subsys);
2770 hlist_add_head(&init_css_set.hlist, hhead);
2772 err = register_filesystem(&cgroup_fs_type);
2776 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
2780 bdi_destroy(&cgroup_backing_dev_info);
2786 * proc_cgroup_show()
2787 * - Print task's cgroup paths into seq_file, one line for each hierarchy
2788 * - Used for /proc/<pid>/cgroup.
2789 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
2790 * doesn't really matter if tsk->cgroup changes after we read it,
2791 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
2792 * anyway. No need to check that tsk->cgroup != NULL, thanks to
2793 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
2794 * cgroup to top_cgroup.
2797 /* TODO: Use a proper seq_file iterator */
2798 static int proc_cgroup_show(struct seq_file *m, void *v)
2801 struct task_struct *tsk;
2804 struct cgroupfs_root *root;
2807 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2813 tsk = get_pid_task(pid, PIDTYPE_PID);
2819 mutex_lock(&cgroup_mutex);
2821 for_each_active_root(root) {
2822 struct cgroup_subsys *ss;
2823 struct cgroup *cgrp;
2827 seq_printf(m, "%lu:", root->subsys_bits);
2828 for_each_subsys(root, ss)
2829 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
2831 get_first_subsys(&root->top_cgroup, NULL, &subsys_id);
2832 cgrp = task_cgroup(tsk, subsys_id);
2833 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
2841 mutex_unlock(&cgroup_mutex);
2842 put_task_struct(tsk);
2849 static int cgroup_open(struct inode *inode, struct file *file)
2851 struct pid *pid = PROC_I(inode)->pid;
2852 return single_open(file, proc_cgroup_show, pid);
2855 struct file_operations proc_cgroup_operations = {
2856 .open = cgroup_open,
2858 .llseek = seq_lseek,
2859 .release = single_release,
2862 /* Display information about each subsystem and each hierarchy */
2863 static int proc_cgroupstats_show(struct seq_file *m, void *v)
2867 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
2868 mutex_lock(&cgroup_mutex);
2869 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2870 struct cgroup_subsys *ss = subsys[i];
2871 seq_printf(m, "%s\t%lu\t%d\t%d\n",
2872 ss->name, ss->root->subsys_bits,
2873 ss->root->number_of_cgroups, !ss->disabled);
2875 mutex_unlock(&cgroup_mutex);
2879 static int cgroupstats_open(struct inode *inode, struct file *file)
2881 return single_open(file, proc_cgroupstats_show, NULL);
2884 static struct file_operations proc_cgroupstats_operations = {
2885 .open = cgroupstats_open,
2887 .llseek = seq_lseek,
2888 .release = single_release,
2892 * cgroup_fork - attach newly forked task to its parents cgroup.
2893 * @child: pointer to task_struct of forking parent process.
2895 * Description: A task inherits its parent's cgroup at fork().
2897 * A pointer to the shared css_set was automatically copied in
2898 * fork.c by dup_task_struct(). However, we ignore that copy, since
2899 * it was not made under the protection of RCU or cgroup_mutex, so
2900 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
2901 * have already changed current->cgroups, allowing the previously
2902 * referenced cgroup group to be removed and freed.
2904 * At the point that cgroup_fork() is called, 'current' is the parent
2905 * task, and the passed argument 'child' points to the child task.
2907 void cgroup_fork(struct task_struct *child)
2910 child->cgroups = current->cgroups;
2911 get_css_set(child->cgroups);
2912 task_unlock(current);
2913 INIT_LIST_HEAD(&child->cg_list);
2917 * cgroup_fork_callbacks - run fork callbacks
2918 * @child: the new task
2920 * Called on a new task very soon before adding it to the
2921 * tasklist. No need to take any locks since no-one can
2922 * be operating on this task.
2924 void cgroup_fork_callbacks(struct task_struct *child)
2926 if (need_forkexit_callback) {
2928 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2929 struct cgroup_subsys *ss = subsys[i];
2931 ss->fork(ss, child);
2937 * cgroup_post_fork - called on a new task after adding it to the task list
2938 * @child: the task in question
2940 * Adds the task to the list running through its css_set if necessary.
2941 * Has to be after the task is visible on the task list in case we race
2942 * with the first call to cgroup_iter_start() - to guarantee that the
2943 * new task ends up on its list.
2945 void cgroup_post_fork(struct task_struct *child)
2947 if (use_task_css_set_links) {
2948 write_lock(&css_set_lock);
2950 if (list_empty(&child->cg_list))
2951 list_add(&child->cg_list, &child->cgroups->tasks);
2953 write_unlock(&css_set_lock);
2957 * cgroup_exit - detach cgroup from exiting task
2958 * @tsk: pointer to task_struct of exiting process
2959 * @run_callback: run exit callbacks?
2961 * Description: Detach cgroup from @tsk and release it.
2963 * Note that cgroups marked notify_on_release force every task in
2964 * them to take the global cgroup_mutex mutex when exiting.
2965 * This could impact scaling on very large systems. Be reluctant to
2966 * use notify_on_release cgroups where very high task exit scaling
2967 * is required on large systems.
2969 * the_top_cgroup_hack:
2971 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
2973 * We call cgroup_exit() while the task is still competent to
2974 * handle notify_on_release(), then leave the task attached to the
2975 * root cgroup in each hierarchy for the remainder of its exit.
2977 * To do this properly, we would increment the reference count on
2978 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
2979 * code we would add a second cgroup function call, to drop that
2980 * reference. This would just create an unnecessary hot spot on
2981 * the top_cgroup reference count, to no avail.
2983 * Normally, holding a reference to a cgroup without bumping its
2984 * count is unsafe. The cgroup could go away, or someone could
2985 * attach us to a different cgroup, decrementing the count on
2986 * the first cgroup that we never incremented. But in this case,
2987 * top_cgroup isn't going away, and either task has PF_EXITING set,
2988 * which wards off any cgroup_attach_task() attempts, or task is a failed
2989 * fork, never visible to cgroup_attach_task.
2991 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
2996 if (run_callbacks && need_forkexit_callback) {
2997 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2998 struct cgroup_subsys *ss = subsys[i];
3005 * Unlink from the css_set task list if necessary.
3006 * Optimistically check cg_list before taking
3009 if (!list_empty(&tsk->cg_list)) {
3010 write_lock(&css_set_lock);
3011 if (!list_empty(&tsk->cg_list))
3012 list_del(&tsk->cg_list);
3013 write_unlock(&css_set_lock);
3016 /* Reassign the task to the init_css_set. */
3019 tsk->cgroups = &init_css_set;
3022 put_css_set_taskexit(cg);
3026 * cgroup_clone - clone the cgroup the given subsystem is attached to
3027 * @tsk: the task to be moved
3028 * @subsys: the given subsystem
3029 * @nodename: the name for the new cgroup
3031 * Duplicate the current cgroup in the hierarchy that the given
3032 * subsystem is attached to, and move this task into the new
3035 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys,
3038 struct dentry *dentry;
3040 struct cgroup *parent, *child;
3041 struct inode *inode;
3043 struct cgroupfs_root *root;
3044 struct cgroup_subsys *ss;
3046 /* We shouldn't be called by an unregistered subsystem */
3047 BUG_ON(!subsys->active);
3049 /* First figure out what hierarchy and cgroup we're dealing
3050 * with, and pin them so we can drop cgroup_mutex */
3051 mutex_lock(&cgroup_mutex);
3053 root = subsys->root;
3054 if (root == &rootnode) {
3055 mutex_unlock(&cgroup_mutex);
3059 /* Pin the hierarchy */
3060 if (!atomic_inc_not_zero(&root->sb->s_active)) {
3061 /* We race with the final deactivate_super() */
3062 mutex_unlock(&cgroup_mutex);
3066 /* Keep the cgroup alive */
3068 parent = task_cgroup(tsk, subsys->subsys_id);
3073 mutex_unlock(&cgroup_mutex);
3075 /* Now do the VFS work to create a cgroup */
3076 inode = parent->dentry->d_inode;
3078 /* Hold the parent directory mutex across this operation to
3079 * stop anyone else deleting the new cgroup */
3080 mutex_lock(&inode->i_mutex);
3081 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
3082 if (IS_ERR(dentry)) {
3084 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
3086 ret = PTR_ERR(dentry);
3090 /* Create the cgroup directory, which also creates the cgroup */
3091 ret = vfs_mkdir(inode, dentry, 0755);
3092 child = __d_cgrp(dentry);
3096 "Failed to create cgroup %s: %d\n", nodename,
3101 /* The cgroup now exists. Retake cgroup_mutex and check
3102 * that we're still in the same state that we thought we
3104 mutex_lock(&cgroup_mutex);
3105 if ((root != subsys->root) ||
3106 (parent != task_cgroup(tsk, subsys->subsys_id))) {
3107 /* Aargh, we raced ... */
3108 mutex_unlock(&inode->i_mutex);
3111 deactivate_super(root->sb);
3112 /* The cgroup is still accessible in the VFS, but
3113 * we're not going to try to rmdir() it at this
3116 "Race in cgroup_clone() - leaking cgroup %s\n",
3121 /* do any required auto-setup */
3122 for_each_subsys(root, ss) {
3124 ss->post_clone(ss, child);
3127 /* All seems fine. Finish by moving the task into the new cgroup */
3128 ret = cgroup_attach_task(child, tsk);
3129 mutex_unlock(&cgroup_mutex);
3132 mutex_unlock(&inode->i_mutex);
3134 mutex_lock(&cgroup_mutex);
3136 mutex_unlock(&cgroup_mutex);
3137 deactivate_super(root->sb);
3142 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
3143 * @cgrp: the cgroup in question
3144 * @task: the task in question
3146 * See if @cgrp is a descendant of @task's cgroup in the appropriate
3149 * If we are sending in dummytop, then presumably we are creating
3150 * the top cgroup in the subsystem.
3152 * Called only by the ns (nsproxy) cgroup.
3154 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
3157 struct cgroup *target;
3160 if (cgrp == dummytop)
3163 get_first_subsys(cgrp, NULL, &subsys_id);
3164 target = task_cgroup(task, subsys_id);
3165 while (cgrp != target && cgrp!= cgrp->top_cgroup)
3166 cgrp = cgrp->parent;
3167 ret = (cgrp == target);
3171 static void check_for_release(struct cgroup *cgrp)
3173 /* All of these checks rely on RCU to keep the cgroup
3174 * structure alive */
3175 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
3176 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
3177 /* Control Group is currently removeable. If it's not
3178 * already queued for a userspace notification, queue
3180 int need_schedule_work = 0;
3181 spin_lock(&release_list_lock);
3182 if (!cgroup_is_removed(cgrp) &&
3183 list_empty(&cgrp->release_list)) {
3184 list_add(&cgrp->release_list, &release_list);
3185 need_schedule_work = 1;
3187 spin_unlock(&release_list_lock);
3188 if (need_schedule_work)
3189 schedule_work(&release_agent_work);
3193 void __css_put(struct cgroup_subsys_state *css)
3195 struct cgroup *cgrp = css->cgroup;
3197 if ((atomic_dec_return(&css->refcnt) == 1) &&
3198 notify_on_release(cgrp)) {
3199 set_bit(CGRP_RELEASABLE, &cgrp->flags);
3200 check_for_release(cgrp);
3206 * Notify userspace when a cgroup is released, by running the
3207 * configured release agent with the name of the cgroup (path
3208 * relative to the root of cgroup file system) as the argument.
3210 * Most likely, this user command will try to rmdir this cgroup.
3212 * This races with the possibility that some other task will be
3213 * attached to this cgroup before it is removed, or that some other
3214 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
3215 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
3216 * unused, and this cgroup will be reprieved from its death sentence,
3217 * to continue to serve a useful existence. Next time it's released,
3218 * we will get notified again, if it still has 'notify_on_release' set.
3220 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
3221 * means only wait until the task is successfully execve()'d. The
3222 * separate release agent task is forked by call_usermodehelper(),
3223 * then control in this thread returns here, without waiting for the
3224 * release agent task. We don't bother to wait because the caller of
3225 * this routine has no use for the exit status of the release agent
3226 * task, so no sense holding our caller up for that.
3228 static void cgroup_release_agent(struct work_struct *work)
3230 BUG_ON(work != &release_agent_work);
3231 mutex_lock(&cgroup_mutex);
3232 spin_lock(&release_list_lock);
3233 while (!list_empty(&release_list)) {
3234 char *argv[3], *envp[3];
3236 char *pathbuf = NULL, *agentbuf = NULL;
3237 struct cgroup *cgrp = list_entry(release_list.next,
3240 list_del_init(&cgrp->release_list);
3241 spin_unlock(&release_list_lock);
3242 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3245 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
3247 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
3252 argv[i++] = agentbuf;
3253 argv[i++] = pathbuf;
3257 /* minimal command environment */
3258 envp[i++] = "HOME=/";
3259 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
3262 /* Drop the lock while we invoke the usermode helper,
3263 * since the exec could involve hitting disk and hence
3264 * be a slow process */
3265 mutex_unlock(&cgroup_mutex);
3266 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
3267 mutex_lock(&cgroup_mutex);
3271 spin_lock(&release_list_lock);
3273 spin_unlock(&release_list_lock);
3274 mutex_unlock(&cgroup_mutex);
3277 static int __init cgroup_disable(char *str)
3282 while ((token = strsep(&str, ",")) != NULL) {
3286 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3287 struct cgroup_subsys *ss = subsys[i];
3289 if (!strcmp(token, ss->name)) {
3291 printk(KERN_INFO "Disabling %s control group"
3292 " subsystem\n", ss->name);
3299 __setup("cgroup_disable=", cgroup_disable);
3302 * Functons for CSS ID.
3306 *To get ID other than 0, this should be called when !cgroup_is_removed().
3308 unsigned short css_id(struct cgroup_subsys_state *css)
3310 struct css_id *cssid = rcu_dereference(css->id);
3317 unsigned short css_depth(struct cgroup_subsys_state *css)
3319 struct css_id *cssid = rcu_dereference(css->id);
3322 return cssid->depth;
3326 bool css_is_ancestor(struct cgroup_subsys_state *child,
3327 struct cgroup_subsys_state *root)
3329 struct css_id *child_id = rcu_dereference(child->id);
3330 struct css_id *root_id = rcu_dereference(root->id);
3332 if (!child_id || !root_id || (child_id->depth < root_id->depth))
3334 return child_id->stack[root_id->depth] == root_id->id;
3337 static void __free_css_id_cb(struct rcu_head *head)
3341 id = container_of(head, struct css_id, rcu_head);
3345 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
3347 struct css_id *id = css->id;
3348 /* When this is called before css_id initialization, id can be NULL */
3352 BUG_ON(!ss->use_id);
3354 rcu_assign_pointer(id->css, NULL);
3355 rcu_assign_pointer(css->id, NULL);
3356 spin_lock(&ss->id_lock);
3357 idr_remove(&ss->idr, id->id);
3358 spin_unlock(&ss->id_lock);
3359 call_rcu(&id->rcu_head, __free_css_id_cb);
3363 * This is called by init or create(). Then, calls to this function are
3364 * always serialized (By cgroup_mutex() at create()).
3367 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
3369 struct css_id *newid;
3370 int myid, error, size;
3372 BUG_ON(!ss->use_id);
3374 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
3375 newid = kzalloc(size, GFP_KERNEL);
3377 return ERR_PTR(-ENOMEM);
3379 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
3383 spin_lock(&ss->id_lock);
3384 /* Don't use 0. allocates an ID of 1-65535 */
3385 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
3386 spin_unlock(&ss->id_lock);
3388 /* Returns error when there are no free spaces for new ID.*/
3393 if (myid > CSS_ID_MAX)
3397 newid->depth = depth;
3401 spin_lock(&ss->id_lock);
3402 idr_remove(&ss->idr, myid);
3403 spin_unlock(&ss->id_lock);
3406 return ERR_PTR(error);
3410 static int __init cgroup_subsys_init_idr(struct cgroup_subsys *ss)
3412 struct css_id *newid;
3413 struct cgroup_subsys_state *rootcss;
3415 spin_lock_init(&ss->id_lock);
3418 rootcss = init_css_set.subsys[ss->subsys_id];
3419 newid = get_new_cssid(ss, 0);
3421 return PTR_ERR(newid);
3423 newid->stack[0] = newid->id;
3424 newid->css = rootcss;
3425 rootcss->id = newid;
3429 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
3430 struct cgroup *child)
3432 int subsys_id, i, depth = 0;
3433 struct cgroup_subsys_state *parent_css, *child_css;
3434 struct css_id *child_id, *parent_id = NULL;
3436 subsys_id = ss->subsys_id;
3437 parent_css = parent->subsys[subsys_id];
3438 child_css = child->subsys[subsys_id];
3439 depth = css_depth(parent_css) + 1;
3440 parent_id = parent_css->id;
3442 child_id = get_new_cssid(ss, depth);
3443 if (IS_ERR(child_id))
3444 return PTR_ERR(child_id);
3446 for (i = 0; i < depth; i++)
3447 child_id->stack[i] = parent_id->stack[i];
3448 child_id->stack[depth] = child_id->id;
3450 * child_id->css pointer will be set after this cgroup is available
3451 * see cgroup_populate_dir()
3453 rcu_assign_pointer(child_css->id, child_id);
3459 * css_lookup - lookup css by id
3460 * @ss: cgroup subsys to be looked into.
3463 * Returns pointer to cgroup_subsys_state if there is valid one with id.
3464 * NULL if not. Should be called under rcu_read_lock()
3466 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
3468 struct css_id *cssid = NULL;
3470 BUG_ON(!ss->use_id);
3471 cssid = idr_find(&ss->idr, id);
3473 if (unlikely(!cssid))
3476 return rcu_dereference(cssid->css);
3480 * css_get_next - lookup next cgroup under specified hierarchy.
3481 * @ss: pointer to subsystem
3482 * @id: current position of iteration.
3483 * @root: pointer to css. search tree under this.
3484 * @foundid: position of found object.
3486 * Search next css under the specified hierarchy of rootid. Calling under
3487 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
3489 struct cgroup_subsys_state *
3490 css_get_next(struct cgroup_subsys *ss, int id,
3491 struct cgroup_subsys_state *root, int *foundid)
3493 struct cgroup_subsys_state *ret = NULL;
3496 int rootid = css_id(root);
3497 int depth = css_depth(root);
3502 BUG_ON(!ss->use_id);
3503 /* fill start point for scan */
3507 * scan next entry from bitmap(tree), tmpid is updated after
3510 spin_lock(&ss->id_lock);
3511 tmp = idr_get_next(&ss->idr, &tmpid);
3512 spin_unlock(&ss->id_lock);
3516 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
3517 ret = rcu_dereference(tmp->css);
3523 /* continue to scan from next id */