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>
33 #include <linux/init_task.h>
34 #include <linux/kernel.h>
35 #include <linux/list.h>
37 #include <linux/mutex.h>
38 #include <linux/mount.h>
39 #include <linux/pagemap.h>
40 #include <linux/proc_fs.h>
41 #include <linux/rcupdate.h>
42 #include <linux/sched.h>
43 #include <linux/backing-dev.h>
44 #include <linux/seq_file.h>
45 #include <linux/slab.h>
46 #include <linux/magic.h>
47 #include <linux/spinlock.h>
48 #include <linux/string.h>
49 #include <linux/sort.h>
50 #include <linux/kmod.h>
51 #include <linux/module.h>
52 #include <linux/delayacct.h>
53 #include <linux/cgroupstats.h>
54 #include <linux/hashtable.h>
55 #include <linux/namei.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/idr.h>
58 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
59 #include <linux/eventfd.h>
60 #include <linux/poll.h>
61 #include <linux/flex_array.h> /* used in cgroup_attach_task */
62 #include <linux/kthread.h>
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 #ifdef CONFIG_PROVE_RCU
86 DEFINE_MUTEX(cgroup_mutex);
87 EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for task_subsys_state_check() */
89 static DEFINE_MUTEX(cgroup_mutex);
92 static DEFINE_MUTEX(cgroup_root_mutex);
95 * Generate an array of cgroup subsystem pointers. At boot time, this is
96 * populated with the built in subsystems, and modular subsystems are
97 * registered after that. The mutable section of this array is protected by
100 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
101 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
102 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
103 #include <linux/cgroup_subsys.h>
107 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
108 * subsystems that are otherwise unattached - it never has more than a
109 * single cgroup, and all tasks are part of that cgroup.
111 static struct cgroupfs_root rootnode;
114 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
117 struct list_head node;
118 struct dentry *dentry;
123 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
124 * cgroup_subsys->use_id != 0.
126 #define CSS_ID_MAX (65535)
129 * The css to which this ID points. This pointer is set to valid value
130 * after cgroup is populated. If cgroup is removed, this will be NULL.
131 * This pointer is expected to be RCU-safe because destroy()
132 * is called after synchronize_rcu(). But for safe use, css_tryget()
133 * should be used for avoiding race.
135 struct cgroup_subsys_state __rcu *css;
141 * Depth in hierarchy which this ID belongs to.
143 unsigned short depth;
145 * ID is freed by RCU. (and lookup routine is RCU safe.)
147 struct rcu_head rcu_head;
149 * Hierarchy of CSS ID belongs to.
151 unsigned short stack[0]; /* Array of Length (depth+1) */
155 * cgroup_event represents events which userspace want to receive.
157 struct cgroup_event {
159 * Cgroup which the event belongs to.
163 * Control file which the event associated.
167 * eventfd to signal userspace about the event.
169 struct eventfd_ctx *eventfd;
171 * Each of these stored in a list by the cgroup.
173 struct list_head list;
175 * All fields below needed to unregister event when
176 * userspace closes eventfd.
179 wait_queue_head_t *wqh;
181 struct work_struct remove;
184 /* The list of hierarchy roots */
186 static LIST_HEAD(roots);
187 static int root_count;
189 static DEFINE_IDA(hierarchy_ida);
190 static int next_hierarchy_id;
191 static DEFINE_SPINLOCK(hierarchy_id_lock);
193 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
194 #define dummytop (&rootnode.top_cgroup)
196 static struct cgroup_name root_cgroup_name = { .name = "/" };
198 /* This flag indicates whether tasks in the fork and exit paths should
199 * check for fork/exit handlers to call. This avoids us having to do
200 * extra work in the fork/exit path if none of the subsystems need to
203 static int need_forkexit_callback __read_mostly;
205 static int cgroup_destroy_locked(struct cgroup *cgrp);
206 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
207 struct cftype cfts[], bool is_add);
209 static int css_unbias_refcnt(int refcnt)
211 return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
214 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
215 static int css_refcnt(struct cgroup_subsys_state *css)
217 int v = atomic_read(&css->refcnt);
219 return css_unbias_refcnt(v);
222 /* convenient tests for these bits */
223 inline int cgroup_is_removed(const struct cgroup *cgrp)
225 return test_bit(CGRP_REMOVED, &cgrp->flags);
229 * cgroup_is_descendant - test ancestry
230 * @cgrp: the cgroup to be tested
231 * @ancestor: possible ancestor of @cgrp
233 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
234 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
235 * and @ancestor are accessible.
237 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
240 if (cgrp == ancestor)
246 EXPORT_SYMBOL_GPL(cgroup_is_descendant);
248 static int cgroup_is_releasable(const struct cgroup *cgrp)
251 (1 << CGRP_RELEASABLE) |
252 (1 << CGRP_NOTIFY_ON_RELEASE);
253 return (cgrp->flags & bits) == bits;
256 static int notify_on_release(const struct cgroup *cgrp)
258 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
262 * for_each_subsys() allows you to iterate on each subsystem attached to
263 * an active hierarchy
265 #define for_each_subsys(_root, _ss) \
266 list_for_each_entry(_ss, &_root->subsys_list, sibling)
268 /* for_each_active_root() allows you to iterate across the active hierarchies */
269 #define for_each_active_root(_root) \
270 list_for_each_entry(_root, &roots, root_list)
272 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
274 return dentry->d_fsdata;
277 static inline struct cfent *__d_cfe(struct dentry *dentry)
279 return dentry->d_fsdata;
282 static inline struct cftype *__d_cft(struct dentry *dentry)
284 return __d_cfe(dentry)->type;
288 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
289 * @cgrp: the cgroup to be checked for liveness
291 * On success, returns true; the mutex should be later unlocked. On
292 * failure returns false with no lock held.
294 static bool cgroup_lock_live_group(struct cgroup *cgrp)
296 mutex_lock(&cgroup_mutex);
297 if (cgroup_is_removed(cgrp)) {
298 mutex_unlock(&cgroup_mutex);
304 /* the list of cgroups eligible for automatic release. Protected by
305 * release_list_lock */
306 static LIST_HEAD(release_list);
307 static DEFINE_RAW_SPINLOCK(release_list_lock);
308 static void cgroup_release_agent(struct work_struct *work);
309 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
310 static void check_for_release(struct cgroup *cgrp);
312 /* Link structure for associating css_set objects with cgroups */
313 struct cg_cgroup_link {
315 * List running through cg_cgroup_links associated with a
316 * cgroup, anchored on cgroup->css_sets
318 struct list_head cgrp_link_list;
321 * List running through cg_cgroup_links pointing at a
322 * single css_set object, anchored on css_set->cg_links
324 struct list_head cg_link_list;
328 /* The default css_set - used by init and its children prior to any
329 * hierarchies being mounted. It contains a pointer to the root state
330 * for each subsystem. Also used to anchor the list of css_sets. Not
331 * reference-counted, to improve performance when child cgroups
332 * haven't been created.
335 static struct css_set init_css_set;
336 static struct cg_cgroup_link init_css_set_link;
338 static int cgroup_init_idr(struct cgroup_subsys *ss,
339 struct cgroup_subsys_state *css);
341 /* css_set_lock protects the list of css_set objects, and the
342 * chain of tasks off each css_set. Nests outside task->alloc_lock
343 * due to cgroup_iter_start() */
344 static DEFINE_RWLOCK(css_set_lock);
345 static int css_set_count;
348 * hash table for cgroup groups. This improves the performance to find
349 * an existing css_set. This hash doesn't (currently) take into
350 * account cgroups in empty hierarchies.
352 #define CSS_SET_HASH_BITS 7
353 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
355 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
358 unsigned long key = 0UL;
360 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
361 key += (unsigned long)css[i];
362 key = (key >> 16) ^ key;
367 /* We don't maintain the lists running through each css_set to its
368 * task until after the first call to cgroup_iter_start(). This
369 * reduces the fork()/exit() overhead for people who have cgroups
370 * compiled into their kernel but not actually in use */
371 static int use_task_css_set_links __read_mostly;
373 static void __put_css_set(struct css_set *cg, int taskexit)
375 struct cg_cgroup_link *link;
376 struct cg_cgroup_link *saved_link;
378 * Ensure that the refcount doesn't hit zero while any readers
379 * can see it. Similar to atomic_dec_and_lock(), but for an
382 if (atomic_add_unless(&cg->refcount, -1, 1))
384 write_lock(&css_set_lock);
385 if (!atomic_dec_and_test(&cg->refcount)) {
386 write_unlock(&css_set_lock);
390 /* This css_set is dead. unlink it and release cgroup refcounts */
391 hash_del(&cg->hlist);
394 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
396 struct cgroup *cgrp = link->cgrp;
397 list_del(&link->cg_link_list);
398 list_del(&link->cgrp_link_list);
401 * We may not be holding cgroup_mutex, and if cgrp->count is
402 * dropped to 0 the cgroup can be destroyed at any time, hence
403 * rcu_read_lock is used to keep it alive.
406 if (atomic_dec_and_test(&cgrp->count) &&
407 notify_on_release(cgrp)) {
409 set_bit(CGRP_RELEASABLE, &cgrp->flags);
410 check_for_release(cgrp);
417 write_unlock(&css_set_lock);
418 kfree_rcu(cg, rcu_head);
422 * refcounted get/put for css_set objects
424 static inline void get_css_set(struct css_set *cg)
426 atomic_inc(&cg->refcount);
429 static inline void put_css_set(struct css_set *cg)
431 __put_css_set(cg, 0);
434 static inline void put_css_set_taskexit(struct css_set *cg)
436 __put_css_set(cg, 1);
440 * compare_css_sets - helper function for find_existing_css_set().
441 * @cg: candidate css_set being tested
442 * @old_cg: existing css_set for a task
443 * @new_cgrp: cgroup that's being entered by the task
444 * @template: desired set of css pointers in css_set (pre-calculated)
446 * Returns true if "cg" matches "old_cg" except for the hierarchy
447 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
449 static bool compare_css_sets(struct css_set *cg,
450 struct css_set *old_cg,
451 struct cgroup *new_cgrp,
452 struct cgroup_subsys_state *template[])
454 struct list_head *l1, *l2;
456 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
457 /* Not all subsystems matched */
462 * Compare cgroup pointers in order to distinguish between
463 * different cgroups in heirarchies with no subsystems. We
464 * could get by with just this check alone (and skip the
465 * memcmp above) but on most setups the memcmp check will
466 * avoid the need for this more expensive check on almost all
471 l2 = &old_cg->cg_links;
473 struct cg_cgroup_link *cgl1, *cgl2;
474 struct cgroup *cg1, *cg2;
478 /* See if we reached the end - both lists are equal length. */
479 if (l1 == &cg->cg_links) {
480 BUG_ON(l2 != &old_cg->cg_links);
483 BUG_ON(l2 == &old_cg->cg_links);
485 /* Locate the cgroups associated with these links. */
486 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
487 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
490 /* Hierarchies should be linked in the same order. */
491 BUG_ON(cg1->root != cg2->root);
494 * If this hierarchy is the hierarchy of the cgroup
495 * that's changing, then we need to check that this
496 * css_set points to the new cgroup; if it's any other
497 * hierarchy, then this css_set should point to the
498 * same cgroup as the old css_set.
500 if (cg1->root == new_cgrp->root) {
512 * find_existing_css_set() is a helper for
513 * find_css_set(), and checks to see whether an existing
514 * css_set is suitable.
516 * oldcg: the cgroup group that we're using before the cgroup
519 * cgrp: the cgroup that we're moving into
521 * template: location in which to build the desired set of subsystem
522 * state objects for the new cgroup group
524 static struct css_set *find_existing_css_set(
525 struct css_set *oldcg,
527 struct cgroup_subsys_state *template[])
530 struct cgroupfs_root *root = cgrp->root;
535 * Build the set of subsystem state objects that we want to see in the
536 * new css_set. while subsystems can change globally, the entries here
537 * won't change, so no need for locking.
539 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
540 if (root->subsys_mask & (1UL << i)) {
541 /* Subsystem is in this hierarchy. So we want
542 * the subsystem state from the new
544 template[i] = cgrp->subsys[i];
546 /* Subsystem is not in this hierarchy, so we
547 * don't want to change the subsystem state */
548 template[i] = oldcg->subsys[i];
552 key = css_set_hash(template);
553 hash_for_each_possible(css_set_table, cg, hlist, key) {
554 if (!compare_css_sets(cg, oldcg, cgrp, template))
557 /* This css_set matches what we need */
561 /* No existing cgroup group matched */
565 static void free_cg_links(struct list_head *tmp)
567 struct cg_cgroup_link *link;
568 struct cg_cgroup_link *saved_link;
570 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
571 list_del(&link->cgrp_link_list);
577 * allocate_cg_links() allocates "count" cg_cgroup_link structures
578 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
579 * success or a negative error
581 static int allocate_cg_links(int count, struct list_head *tmp)
583 struct cg_cgroup_link *link;
586 for (i = 0; i < count; i++) {
587 link = kmalloc(sizeof(*link), GFP_KERNEL);
592 list_add(&link->cgrp_link_list, tmp);
598 * link_css_set - a helper function to link a css_set to a cgroup
599 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
600 * @cg: the css_set to be linked
601 * @cgrp: the destination cgroup
603 static void link_css_set(struct list_head *tmp_cg_links,
604 struct css_set *cg, struct cgroup *cgrp)
606 struct cg_cgroup_link *link;
608 BUG_ON(list_empty(tmp_cg_links));
609 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
613 atomic_inc(&cgrp->count);
614 list_move(&link->cgrp_link_list, &cgrp->css_sets);
616 * Always add links to the tail of the list so that the list
617 * is sorted by order of hierarchy creation
619 list_add_tail(&link->cg_link_list, &cg->cg_links);
623 * find_css_set() takes an existing cgroup group and a
624 * cgroup object, and returns a css_set object that's
625 * equivalent to the old group, but with the given cgroup
626 * substituted into the appropriate hierarchy. Must be called with
629 static struct css_set *find_css_set(
630 struct css_set *oldcg, struct cgroup *cgrp)
633 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
635 struct list_head tmp_cg_links;
637 struct cg_cgroup_link *link;
640 /* First see if we already have a cgroup group that matches
642 read_lock(&css_set_lock);
643 res = find_existing_css_set(oldcg, cgrp, template);
646 read_unlock(&css_set_lock);
651 res = kmalloc(sizeof(*res), GFP_KERNEL);
655 /* Allocate all the cg_cgroup_link objects that we'll need */
656 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
661 atomic_set(&res->refcount, 1);
662 INIT_LIST_HEAD(&res->cg_links);
663 INIT_LIST_HEAD(&res->tasks);
664 INIT_HLIST_NODE(&res->hlist);
666 /* Copy the set of subsystem state objects generated in
667 * find_existing_css_set() */
668 memcpy(res->subsys, template, sizeof(res->subsys));
670 write_lock(&css_set_lock);
671 /* Add reference counts and links from the new css_set. */
672 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
673 struct cgroup *c = link->cgrp;
674 if (c->root == cgrp->root)
676 link_css_set(&tmp_cg_links, res, c);
679 BUG_ON(!list_empty(&tmp_cg_links));
683 /* Add this cgroup group to the hash table */
684 key = css_set_hash(res->subsys);
685 hash_add(css_set_table, &res->hlist, key);
687 write_unlock(&css_set_lock);
693 * Return the cgroup for "task" from the given hierarchy. Must be
694 * called with cgroup_mutex held.
696 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
697 struct cgroupfs_root *root)
700 struct cgroup *res = NULL;
702 BUG_ON(!mutex_is_locked(&cgroup_mutex));
703 read_lock(&css_set_lock);
705 * No need to lock the task - since we hold cgroup_mutex the
706 * task can't change groups, so the only thing that can happen
707 * is that it exits and its css is set back to init_css_set.
710 if (css == &init_css_set) {
711 res = &root->top_cgroup;
713 struct cg_cgroup_link *link;
714 list_for_each_entry(link, &css->cg_links, cg_link_list) {
715 struct cgroup *c = link->cgrp;
716 if (c->root == root) {
722 read_unlock(&css_set_lock);
728 * There is one global cgroup mutex. We also require taking
729 * task_lock() when dereferencing a task's cgroup subsys pointers.
730 * See "The task_lock() exception", at the end of this comment.
732 * A task must hold cgroup_mutex to modify cgroups.
734 * Any task can increment and decrement the count field without lock.
735 * So in general, code holding cgroup_mutex can't rely on the count
736 * field not changing. However, if the count goes to zero, then only
737 * cgroup_attach_task() can increment it again. Because a count of zero
738 * means that no tasks are currently attached, therefore there is no
739 * way a task attached to that cgroup can fork (the other way to
740 * increment the count). So code holding cgroup_mutex can safely
741 * assume that if the count is zero, it will stay zero. Similarly, if
742 * a task holds cgroup_mutex on a cgroup with zero count, it
743 * knows that the cgroup won't be removed, as cgroup_rmdir()
746 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
747 * (usually) take cgroup_mutex. These are the two most performance
748 * critical pieces of code here. The exception occurs on cgroup_exit(),
749 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
750 * is taken, and if the cgroup count is zero, a usermode call made
751 * to the release agent with the name of the cgroup (path relative to
752 * the root of cgroup file system) as the argument.
754 * A cgroup can only be deleted if both its 'count' of using tasks
755 * is zero, and its list of 'children' cgroups is empty. Since all
756 * tasks in the system use _some_ cgroup, and since there is always at
757 * least one task in the system (init, pid == 1), therefore, top_cgroup
758 * always has either children cgroups and/or using tasks. So we don't
759 * need a special hack to ensure that top_cgroup cannot be deleted.
761 * The task_lock() exception
763 * The need for this exception arises from the action of
764 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
765 * another. It does so using cgroup_mutex, however there are
766 * several performance critical places that need to reference
767 * task->cgroup without the expense of grabbing a system global
768 * mutex. Therefore except as noted below, when dereferencing or, as
769 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
770 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
771 * the task_struct routinely used for such matters.
773 * P.S. One more locking exception. RCU is used to guard the
774 * update of a tasks cgroup pointer by cgroup_attach_task()
778 * A couple of forward declarations required, due to cyclic reference loop:
779 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
780 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
784 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
785 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
786 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
787 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
788 unsigned long subsys_mask);
789 static const struct inode_operations cgroup_dir_inode_operations;
790 static const struct file_operations proc_cgroupstats_operations;
792 static struct backing_dev_info cgroup_backing_dev_info = {
794 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
797 static int alloc_css_id(struct cgroup_subsys *ss,
798 struct cgroup *parent, struct cgroup *child);
800 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
802 struct inode *inode = new_inode(sb);
805 inode->i_ino = get_next_ino();
806 inode->i_mode = mode;
807 inode->i_uid = current_fsuid();
808 inode->i_gid = current_fsgid();
809 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
810 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
815 static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
817 struct cgroup_name *name;
819 name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL);
822 strcpy(name->name, dentry->d_name.name);
826 static void cgroup_free_fn(struct work_struct *work)
828 struct cgroup *cgrp = container_of(work, struct cgroup, free_work);
829 struct cgroup_subsys *ss;
831 mutex_lock(&cgroup_mutex);
833 * Release the subsystem state objects.
835 for_each_subsys(cgrp->root, ss)
838 cgrp->root->number_of_cgroups--;
839 mutex_unlock(&cgroup_mutex);
842 * We get a ref to the parent's dentry, and put the ref when
843 * this cgroup is being freed, so it's guaranteed that the
844 * parent won't be destroyed before its children.
846 dput(cgrp->parent->dentry);
849 * Drop the active superblock reference that we took when we
852 deactivate_super(cgrp->root->sb);
855 * if we're getting rid of the cgroup, refcount should ensure
856 * that there are no pidlists left.
858 BUG_ON(!list_empty(&cgrp->pidlists));
860 simple_xattrs_free(&cgrp->xattrs);
862 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
863 kfree(rcu_dereference_raw(cgrp->name));
867 static void cgroup_free_rcu(struct rcu_head *head)
869 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
871 schedule_work(&cgrp->free_work);
874 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
876 /* is dentry a directory ? if so, kfree() associated cgroup */
877 if (S_ISDIR(inode->i_mode)) {
878 struct cgroup *cgrp = dentry->d_fsdata;
880 BUG_ON(!(cgroup_is_removed(cgrp)));
881 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
883 struct cfent *cfe = __d_cfe(dentry);
884 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
885 struct cftype *cft = cfe->type;
887 WARN_ONCE(!list_empty(&cfe->node) &&
888 cgrp != &cgrp->root->top_cgroup,
889 "cfe still linked for %s\n", cfe->type->name);
891 simple_xattrs_free(&cft->xattrs);
896 static int cgroup_delete(const struct dentry *d)
901 static void remove_dir(struct dentry *d)
903 struct dentry *parent = dget(d->d_parent);
906 simple_rmdir(parent->d_inode, d);
910 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
914 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
915 lockdep_assert_held(&cgroup_mutex);
918 * If we're doing cleanup due to failure of cgroup_create(),
919 * the corresponding @cfe may not exist.
921 list_for_each_entry(cfe, &cgrp->files, node) {
922 struct dentry *d = cfe->dentry;
924 if (cft && cfe->type != cft)
929 simple_unlink(cgrp->dentry->d_inode, d);
930 list_del_init(&cfe->node);
938 * cgroup_clear_directory - selective removal of base and subsystem files
939 * @dir: directory containing the files
940 * @base_files: true if the base files should be removed
941 * @subsys_mask: mask of the subsystem ids whose files should be removed
943 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
944 unsigned long subsys_mask)
946 struct cgroup *cgrp = __d_cgrp(dir);
947 struct cgroup_subsys *ss;
949 for_each_subsys(cgrp->root, ss) {
950 struct cftype_set *set;
951 if (!test_bit(ss->subsys_id, &subsys_mask))
953 list_for_each_entry(set, &ss->cftsets, node)
954 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
957 while (!list_empty(&cgrp->files))
958 cgroup_rm_file(cgrp, NULL);
963 * NOTE : the dentry must have been dget()'ed
965 static void cgroup_d_remove_dir(struct dentry *dentry)
967 struct dentry *parent;
968 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
970 cgroup_clear_directory(dentry, true, root->subsys_mask);
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 * Call with cgroup_mutex held. Drops reference counts on modules, including
983 * any duplicate ones that parse_cgroupfs_options took. If this function
984 * returns an error, no reference counts are touched.
986 static int rebind_subsystems(struct cgroupfs_root *root,
987 unsigned long final_subsys_mask)
989 unsigned long added_mask, removed_mask;
990 struct cgroup *cgrp = &root->top_cgroup;
993 BUG_ON(!mutex_is_locked(&cgroup_mutex));
994 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
996 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
997 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
998 /* Check that any added subsystems are currently free */
999 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1000 unsigned long bit = 1UL << i;
1001 struct cgroup_subsys *ss = subsys[i];
1002 if (!(bit & added_mask))
1005 * Nobody should tell us to do a subsys that doesn't exist:
1006 * parse_cgroupfs_options should catch that case and refcounts
1007 * ensure that subsystems won't disappear once selected.
1010 if (ss->root != &rootnode) {
1011 /* Subsystem isn't free */
1016 /* Currently we don't handle adding/removing subsystems when
1017 * any child cgroups exist. This is theoretically supportable
1018 * but involves complex error handling, so it's being left until
1020 if (root->number_of_cgroups > 1)
1023 /* Process each subsystem */
1024 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1025 struct cgroup_subsys *ss = subsys[i];
1026 unsigned long bit = 1UL << i;
1027 if (bit & added_mask) {
1028 /* We're binding this subsystem to this hierarchy */
1030 BUG_ON(cgrp->subsys[i]);
1031 BUG_ON(!dummytop->subsys[i]);
1032 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1033 cgrp->subsys[i] = dummytop->subsys[i];
1034 cgrp->subsys[i]->cgroup = cgrp;
1035 list_move(&ss->sibling, &root->subsys_list);
1039 /* refcount was already taken, and we're keeping it */
1040 } else if (bit & removed_mask) {
1041 /* We're removing this subsystem */
1043 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1044 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1047 dummytop->subsys[i]->cgroup = dummytop;
1048 cgrp->subsys[i] = NULL;
1049 subsys[i]->root = &rootnode;
1050 list_move(&ss->sibling, &rootnode.subsys_list);
1051 /* subsystem is now free - drop reference on module */
1052 module_put(ss->module);
1053 } else if (bit & final_subsys_mask) {
1054 /* Subsystem state should already exist */
1056 BUG_ON(!cgrp->subsys[i]);
1058 * a refcount was taken, but we already had one, so
1059 * drop the extra reference.
1061 module_put(ss->module);
1062 #ifdef CONFIG_MODULE_UNLOAD
1063 BUG_ON(ss->module && !module_refcount(ss->module));
1066 /* Subsystem state shouldn't exist */
1067 BUG_ON(cgrp->subsys[i]);
1070 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1075 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1077 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1078 struct cgroup_subsys *ss;
1080 mutex_lock(&cgroup_root_mutex);
1081 for_each_subsys(root, ss)
1082 seq_printf(seq, ",%s", ss->name);
1083 if (root->flags & CGRP_ROOT_NOPREFIX)
1084 seq_puts(seq, ",noprefix");
1085 if (root->flags & CGRP_ROOT_XATTR)
1086 seq_puts(seq, ",xattr");
1087 if (strlen(root->release_agent_path))
1088 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1089 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1090 seq_puts(seq, ",clone_children");
1091 if (strlen(root->name))
1092 seq_printf(seq, ",name=%s", root->name);
1093 mutex_unlock(&cgroup_root_mutex);
1097 struct cgroup_sb_opts {
1098 unsigned long subsys_mask;
1099 unsigned long flags;
1100 char *release_agent;
1101 bool cpuset_clone_children;
1103 /* User explicitly requested empty subsystem */
1106 struct cgroupfs_root *new_root;
1111 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1112 * with cgroup_mutex held to protect the subsys[] array. This function takes
1113 * refcounts on subsystems to be used, unless it returns error, in which case
1114 * no refcounts are taken.
1116 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1118 char *token, *o = data;
1119 bool all_ss = false, one_ss = false;
1120 unsigned long mask = (unsigned long)-1;
1122 bool module_pin_failed = false;
1124 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1126 #ifdef CONFIG_CPUSETS
1127 mask = ~(1UL << cpuset_subsys_id);
1130 memset(opts, 0, sizeof(*opts));
1132 while ((token = strsep(&o, ",")) != NULL) {
1135 if (!strcmp(token, "none")) {
1136 /* Explicitly have no subsystems */
1140 if (!strcmp(token, "all")) {
1141 /* Mutually exclusive option 'all' + subsystem name */
1147 if (!strcmp(token, "noprefix")) {
1148 opts->flags |= CGRP_ROOT_NOPREFIX;
1151 if (!strcmp(token, "clone_children")) {
1152 opts->cpuset_clone_children = true;
1155 if (!strcmp(token, "xattr")) {
1156 opts->flags |= CGRP_ROOT_XATTR;
1159 if (!strncmp(token, "release_agent=", 14)) {
1160 /* Specifying two release agents is forbidden */
1161 if (opts->release_agent)
1163 opts->release_agent =
1164 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1165 if (!opts->release_agent)
1169 if (!strncmp(token, "name=", 5)) {
1170 const char *name = token + 5;
1171 /* Can't specify an empty name */
1174 /* Must match [\w.-]+ */
1175 for (i = 0; i < strlen(name); i++) {
1179 if ((c == '.') || (c == '-') || (c == '_'))
1183 /* Specifying two names is forbidden */
1186 opts->name = kstrndup(name,
1187 MAX_CGROUP_ROOT_NAMELEN - 1,
1195 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1196 struct cgroup_subsys *ss = subsys[i];
1199 if (strcmp(token, ss->name))
1204 /* Mutually exclusive option 'all' + subsystem name */
1207 set_bit(i, &opts->subsys_mask);
1212 if (i == CGROUP_SUBSYS_COUNT)
1217 * If the 'all' option was specified select all the subsystems,
1218 * otherwise if 'none', 'name=' and a subsystem name options
1219 * were not specified, let's default to 'all'
1221 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1222 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1223 struct cgroup_subsys *ss = subsys[i];
1228 set_bit(i, &opts->subsys_mask);
1232 /* Consistency checks */
1235 * Option noprefix was introduced just for backward compatibility
1236 * with the old cpuset, so we allow noprefix only if mounting just
1237 * the cpuset subsystem.
1239 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1243 /* Can't specify "none" and some subsystems */
1244 if (opts->subsys_mask && opts->none)
1248 * We either have to specify by name or by subsystems. (So all
1249 * empty hierarchies must have a name).
1251 if (!opts->subsys_mask && !opts->name)
1255 * Grab references on all the modules we'll need, so the subsystems
1256 * don't dance around before rebind_subsystems attaches them. This may
1257 * take duplicate reference counts on a subsystem that's already used,
1258 * but rebind_subsystems handles this case.
1260 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1261 unsigned long bit = 1UL << i;
1263 if (!(bit & opts->subsys_mask))
1265 if (!try_module_get(subsys[i]->module)) {
1266 module_pin_failed = true;
1270 if (module_pin_failed) {
1272 * oops, one of the modules was going away. this means that we
1273 * raced with a module_delete call, and to the user this is
1274 * essentially a "subsystem doesn't exist" case.
1276 for (i--; i >= 0; i--) {
1277 /* drop refcounts only on the ones we took */
1278 unsigned long bit = 1UL << i;
1280 if (!(bit & opts->subsys_mask))
1282 module_put(subsys[i]->module);
1290 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1293 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1294 unsigned long bit = 1UL << i;
1296 if (!(bit & subsys_mask))
1298 module_put(subsys[i]->module);
1302 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1305 struct cgroupfs_root *root = sb->s_fs_info;
1306 struct cgroup *cgrp = &root->top_cgroup;
1307 struct cgroup_sb_opts opts;
1308 unsigned long added_mask, removed_mask;
1310 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1311 mutex_lock(&cgroup_mutex);
1312 mutex_lock(&cgroup_root_mutex);
1314 /* See what subsystems are wanted */
1315 ret = parse_cgroupfs_options(data, &opts);
1319 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1320 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1321 task_tgid_nr(current), current->comm);
1323 added_mask = opts.subsys_mask & ~root->subsys_mask;
1324 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1326 /* Don't allow flags or name to change at remount */
1327 if (opts.flags != root->flags ||
1328 (opts.name && strcmp(opts.name, root->name))) {
1330 drop_parsed_module_refcounts(opts.subsys_mask);
1335 * Clear out the files of subsystems that should be removed, do
1336 * this before rebind_subsystems, since rebind_subsystems may
1337 * change this hierarchy's subsys_list.
1339 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1341 ret = rebind_subsystems(root, opts.subsys_mask);
1343 /* rebind_subsystems failed, re-populate the removed files */
1344 cgroup_populate_dir(cgrp, false, removed_mask);
1345 drop_parsed_module_refcounts(opts.subsys_mask);
1349 /* re-populate subsystem files */
1350 cgroup_populate_dir(cgrp, false, added_mask);
1352 if (opts.release_agent)
1353 strcpy(root->release_agent_path, opts.release_agent);
1355 kfree(opts.release_agent);
1357 mutex_unlock(&cgroup_root_mutex);
1358 mutex_unlock(&cgroup_mutex);
1359 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1363 static const struct super_operations cgroup_ops = {
1364 .statfs = simple_statfs,
1365 .drop_inode = generic_delete_inode,
1366 .show_options = cgroup_show_options,
1367 .remount_fs = cgroup_remount,
1370 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1372 INIT_LIST_HEAD(&cgrp->sibling);
1373 INIT_LIST_HEAD(&cgrp->children);
1374 INIT_LIST_HEAD(&cgrp->files);
1375 INIT_LIST_HEAD(&cgrp->css_sets);
1376 INIT_LIST_HEAD(&cgrp->allcg_node);
1377 INIT_LIST_HEAD(&cgrp->release_list);
1378 INIT_LIST_HEAD(&cgrp->pidlists);
1379 INIT_WORK(&cgrp->free_work, cgroup_free_fn);
1380 mutex_init(&cgrp->pidlist_mutex);
1381 INIT_LIST_HEAD(&cgrp->event_list);
1382 spin_lock_init(&cgrp->event_list_lock);
1383 simple_xattrs_init(&cgrp->xattrs);
1386 static void init_cgroup_root(struct cgroupfs_root *root)
1388 struct cgroup *cgrp = &root->top_cgroup;
1390 INIT_LIST_HEAD(&root->subsys_list);
1391 INIT_LIST_HEAD(&root->root_list);
1392 INIT_LIST_HEAD(&root->allcg_list);
1393 root->number_of_cgroups = 1;
1395 cgrp->name = &root_cgroup_name;
1396 cgrp->top_cgroup = cgrp;
1397 init_cgroup_housekeeping(cgrp);
1398 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1401 static bool init_root_id(struct cgroupfs_root *root)
1406 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1408 spin_lock(&hierarchy_id_lock);
1409 /* Try to allocate the next unused ID */
1410 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1411 &root->hierarchy_id);
1413 /* Try again starting from 0 */
1414 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1416 next_hierarchy_id = root->hierarchy_id + 1;
1417 } else if (ret != -EAGAIN) {
1418 /* Can only get here if the 31-bit IDR is full ... */
1421 spin_unlock(&hierarchy_id_lock);
1426 static int cgroup_test_super(struct super_block *sb, void *data)
1428 struct cgroup_sb_opts *opts = data;
1429 struct cgroupfs_root *root = sb->s_fs_info;
1431 /* If we asked for a name then it must match */
1432 if (opts->name && strcmp(opts->name, root->name))
1436 * If we asked for subsystems (or explicitly for no
1437 * subsystems) then they must match
1439 if ((opts->subsys_mask || opts->none)
1440 && (opts->subsys_mask != root->subsys_mask))
1446 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1448 struct cgroupfs_root *root;
1450 if (!opts->subsys_mask && !opts->none)
1453 root = kzalloc(sizeof(*root), GFP_KERNEL);
1455 return ERR_PTR(-ENOMEM);
1457 if (!init_root_id(root)) {
1459 return ERR_PTR(-ENOMEM);
1461 init_cgroup_root(root);
1463 root->subsys_mask = opts->subsys_mask;
1464 root->flags = opts->flags;
1465 ida_init(&root->cgroup_ida);
1466 if (opts->release_agent)
1467 strcpy(root->release_agent_path, opts->release_agent);
1469 strcpy(root->name, opts->name);
1470 if (opts->cpuset_clone_children)
1471 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1475 static void cgroup_drop_root(struct cgroupfs_root *root)
1480 BUG_ON(!root->hierarchy_id);
1481 spin_lock(&hierarchy_id_lock);
1482 ida_remove(&hierarchy_ida, root->hierarchy_id);
1483 spin_unlock(&hierarchy_id_lock);
1484 ida_destroy(&root->cgroup_ida);
1488 static int cgroup_set_super(struct super_block *sb, void *data)
1491 struct cgroup_sb_opts *opts = data;
1493 /* If we don't have a new root, we can't set up a new sb */
1494 if (!opts->new_root)
1497 BUG_ON(!opts->subsys_mask && !opts->none);
1499 ret = set_anon_super(sb, NULL);
1503 sb->s_fs_info = opts->new_root;
1504 opts->new_root->sb = sb;
1506 sb->s_blocksize = PAGE_CACHE_SIZE;
1507 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1508 sb->s_magic = CGROUP_SUPER_MAGIC;
1509 sb->s_op = &cgroup_ops;
1514 static int cgroup_get_rootdir(struct super_block *sb)
1516 static const struct dentry_operations cgroup_dops = {
1517 .d_iput = cgroup_diput,
1518 .d_delete = cgroup_delete,
1521 struct inode *inode =
1522 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1527 inode->i_fop = &simple_dir_operations;
1528 inode->i_op = &cgroup_dir_inode_operations;
1529 /* directories start off with i_nlink == 2 (for "." entry) */
1531 sb->s_root = d_make_root(inode);
1534 /* for everything else we want ->d_op set */
1535 sb->s_d_op = &cgroup_dops;
1539 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1540 int flags, const char *unused_dev_name,
1543 struct cgroup_sb_opts opts;
1544 struct cgroupfs_root *root;
1546 struct super_block *sb;
1547 struct cgroupfs_root *new_root;
1548 struct inode *inode;
1550 /* First find the desired set of subsystems */
1551 mutex_lock(&cgroup_mutex);
1552 ret = parse_cgroupfs_options(data, &opts);
1553 mutex_unlock(&cgroup_mutex);
1558 * Allocate a new cgroup root. We may not need it if we're
1559 * reusing an existing hierarchy.
1561 new_root = cgroup_root_from_opts(&opts);
1562 if (IS_ERR(new_root)) {
1563 ret = PTR_ERR(new_root);
1566 opts.new_root = new_root;
1568 /* Locate an existing or new sb for this hierarchy */
1569 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1572 cgroup_drop_root(opts.new_root);
1576 root = sb->s_fs_info;
1578 if (root == opts.new_root) {
1579 /* We used the new root structure, so this is a new hierarchy */
1580 struct list_head tmp_cg_links;
1581 struct cgroup *root_cgrp = &root->top_cgroup;
1582 struct cgroupfs_root *existing_root;
1583 const struct cred *cred;
1587 BUG_ON(sb->s_root != NULL);
1589 ret = cgroup_get_rootdir(sb);
1591 goto drop_new_super;
1592 inode = sb->s_root->d_inode;
1594 mutex_lock(&inode->i_mutex);
1595 mutex_lock(&cgroup_mutex);
1596 mutex_lock(&cgroup_root_mutex);
1598 /* Check for name clashes with existing mounts */
1600 if (strlen(root->name))
1601 for_each_active_root(existing_root)
1602 if (!strcmp(existing_root->name, root->name))
1606 * We're accessing css_set_count without locking
1607 * css_set_lock here, but that's OK - it can only be
1608 * increased by someone holding cgroup_lock, and
1609 * that's us. The worst that can happen is that we
1610 * have some link structures left over
1612 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1616 ret = rebind_subsystems(root, root->subsys_mask);
1617 if (ret == -EBUSY) {
1618 free_cg_links(&tmp_cg_links);
1622 * There must be no failure case after here, since rebinding
1623 * takes care of subsystems' refcounts, which are explicitly
1624 * dropped in the failure exit path.
1627 /* EBUSY should be the only error here */
1630 list_add(&root->root_list, &roots);
1633 sb->s_root->d_fsdata = root_cgrp;
1634 root->top_cgroup.dentry = sb->s_root;
1636 /* Link the top cgroup in this hierarchy into all
1637 * the css_set objects */
1638 write_lock(&css_set_lock);
1639 hash_for_each(css_set_table, i, cg, hlist)
1640 link_css_set(&tmp_cg_links, cg, root_cgrp);
1641 write_unlock(&css_set_lock);
1643 free_cg_links(&tmp_cg_links);
1645 BUG_ON(!list_empty(&root_cgrp->children));
1646 BUG_ON(root->number_of_cgroups != 1);
1648 cred = override_creds(&init_cred);
1649 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1651 mutex_unlock(&cgroup_root_mutex);
1652 mutex_unlock(&cgroup_mutex);
1653 mutex_unlock(&inode->i_mutex);
1656 * We re-used an existing hierarchy - the new root (if
1657 * any) is not needed
1659 cgroup_drop_root(opts.new_root);
1660 /* no subsys rebinding, so refcounts don't change */
1661 drop_parsed_module_refcounts(opts.subsys_mask);
1664 kfree(opts.release_agent);
1666 return dget(sb->s_root);
1669 mutex_unlock(&cgroup_root_mutex);
1670 mutex_unlock(&cgroup_mutex);
1671 mutex_unlock(&inode->i_mutex);
1673 deactivate_locked_super(sb);
1675 drop_parsed_module_refcounts(opts.subsys_mask);
1677 kfree(opts.release_agent);
1679 return ERR_PTR(ret);
1682 static void cgroup_kill_sb(struct super_block *sb) {
1683 struct cgroupfs_root *root = sb->s_fs_info;
1684 struct cgroup *cgrp = &root->top_cgroup;
1686 struct cg_cgroup_link *link;
1687 struct cg_cgroup_link *saved_link;
1691 BUG_ON(root->number_of_cgroups != 1);
1692 BUG_ON(!list_empty(&cgrp->children));
1694 mutex_lock(&cgroup_mutex);
1695 mutex_lock(&cgroup_root_mutex);
1697 /* Rebind all subsystems back to the default hierarchy */
1698 ret = rebind_subsystems(root, 0);
1699 /* Shouldn't be able to fail ... */
1703 * Release all the links from css_sets to this hierarchy's
1706 write_lock(&css_set_lock);
1708 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1710 list_del(&link->cg_link_list);
1711 list_del(&link->cgrp_link_list);
1714 write_unlock(&css_set_lock);
1716 if (!list_empty(&root->root_list)) {
1717 list_del(&root->root_list);
1721 mutex_unlock(&cgroup_root_mutex);
1722 mutex_unlock(&cgroup_mutex);
1724 simple_xattrs_free(&cgrp->xattrs);
1726 kill_litter_super(sb);
1727 cgroup_drop_root(root);
1730 static struct file_system_type cgroup_fs_type = {
1732 .mount = cgroup_mount,
1733 .kill_sb = cgroup_kill_sb,
1736 static struct kobject *cgroup_kobj;
1739 * cgroup_path - generate the path of a cgroup
1740 * @cgrp: the cgroup in question
1741 * @buf: the buffer to write the path into
1742 * @buflen: the length of the buffer
1744 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1746 * We can't generate cgroup path using dentry->d_name, as accessing
1747 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1748 * inode's i_mutex, while on the other hand cgroup_path() can be called
1749 * with some irq-safe spinlocks held.
1751 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1753 int ret = -ENAMETOOLONG;
1756 if (!cgrp->parent) {
1757 if (strlcpy(buf, "/", buflen) >= buflen)
1758 return -ENAMETOOLONG;
1762 start = buf + buflen - 1;
1767 const char *name = cgroup_name(cgrp);
1771 if ((start -= len) < buf)
1773 memcpy(start, name, len);
1779 cgrp = cgrp->parent;
1780 } while (cgrp->parent);
1782 memmove(buf, start, buf + buflen - start);
1787 EXPORT_SYMBOL_GPL(cgroup_path);
1790 * Control Group taskset
1792 struct task_and_cgroup {
1793 struct task_struct *task;
1794 struct cgroup *cgrp;
1798 struct cgroup_taskset {
1799 struct task_and_cgroup single;
1800 struct flex_array *tc_array;
1803 struct cgroup *cur_cgrp;
1807 * cgroup_taskset_first - reset taskset and return the first task
1808 * @tset: taskset of interest
1810 * @tset iteration is initialized and the first task is returned.
1812 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1814 if (tset->tc_array) {
1816 return cgroup_taskset_next(tset);
1818 tset->cur_cgrp = tset->single.cgrp;
1819 return tset->single.task;
1822 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1825 * cgroup_taskset_next - iterate to the next task in taskset
1826 * @tset: taskset of interest
1828 * Return the next task in @tset. Iteration must have been initialized
1829 * with cgroup_taskset_first().
1831 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1833 struct task_and_cgroup *tc;
1835 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1838 tc = flex_array_get(tset->tc_array, tset->idx++);
1839 tset->cur_cgrp = tc->cgrp;
1842 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1845 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1846 * @tset: taskset of interest
1848 * Return the cgroup for the current (last returned) task of @tset. This
1849 * function must be preceded by either cgroup_taskset_first() or
1850 * cgroup_taskset_next().
1852 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1854 return tset->cur_cgrp;
1856 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1859 * cgroup_taskset_size - return the number of tasks in taskset
1860 * @tset: taskset of interest
1862 int cgroup_taskset_size(struct cgroup_taskset *tset)
1864 return tset->tc_array ? tset->tc_array_len : 1;
1866 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1870 * cgroup_task_migrate - move a task from one cgroup to another.
1872 * Must be called with cgroup_mutex and threadgroup locked.
1874 static void cgroup_task_migrate(struct cgroup *oldcgrp,
1875 struct task_struct *tsk, struct css_set *newcg)
1877 struct css_set *oldcg;
1880 * We are synchronized through threadgroup_lock() against PF_EXITING
1881 * setting such that we can't race against cgroup_exit() changing the
1882 * css_set to init_css_set and dropping the old one.
1884 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1885 oldcg = tsk->cgroups;
1888 rcu_assign_pointer(tsk->cgroups, newcg);
1891 /* Update the css_set linked lists if we're using them */
1892 write_lock(&css_set_lock);
1893 if (!list_empty(&tsk->cg_list))
1894 list_move(&tsk->cg_list, &newcg->tasks);
1895 write_unlock(&css_set_lock);
1898 * We just gained a reference on oldcg by taking it from the task. As
1899 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1900 * it here; it will be freed under RCU.
1902 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1907 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1908 * @cgrp: the cgroup to attach to
1909 * @tsk: the task or the leader of the threadgroup to be attached
1910 * @threadgroup: attach the whole threadgroup?
1912 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1913 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1915 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
1918 int retval, i, group_size;
1919 struct cgroup_subsys *ss, *failed_ss = NULL;
1920 struct cgroupfs_root *root = cgrp->root;
1921 /* threadgroup list cursor and array */
1922 struct task_struct *leader = tsk;
1923 struct task_and_cgroup *tc;
1924 struct flex_array *group;
1925 struct cgroup_taskset tset = { };
1928 * step 0: in order to do expensive, possibly blocking operations for
1929 * every thread, we cannot iterate the thread group list, since it needs
1930 * rcu or tasklist locked. instead, build an array of all threads in the
1931 * group - group_rwsem prevents new threads from appearing, and if
1932 * threads exit, this will just be an over-estimate.
1935 group_size = get_nr_threads(tsk);
1938 /* flex_array supports very large thread-groups better than kmalloc. */
1939 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
1942 /* pre-allocate to guarantee space while iterating in rcu read-side. */
1943 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
1945 goto out_free_group_list;
1949 * Prevent freeing of tasks while we take a snapshot. Tasks that are
1950 * already PF_EXITING could be freed from underneath us unless we
1951 * take an rcu_read_lock.
1955 struct task_and_cgroup ent;
1957 /* @tsk either already exited or can't exit until the end */
1958 if (tsk->flags & PF_EXITING)
1961 /* as per above, nr_threads may decrease, but not increase. */
1962 BUG_ON(i >= group_size);
1964 ent.cgrp = task_cgroup_from_root(tsk, root);
1965 /* nothing to do if this task is already in the cgroup */
1966 if (ent.cgrp == cgrp)
1969 * saying GFP_ATOMIC has no effect here because we did prealloc
1970 * earlier, but it's good form to communicate our expectations.
1972 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
1973 BUG_ON(retval != 0);
1978 } while_each_thread(leader, tsk);
1980 /* remember the number of threads in the array for later. */
1982 tset.tc_array = group;
1983 tset.tc_array_len = group_size;
1985 /* methods shouldn't be called if no task is actually migrating */
1988 goto out_free_group_list;
1991 * step 1: check that we can legitimately attach to the cgroup.
1993 for_each_subsys(root, ss) {
1994 if (ss->can_attach) {
1995 retval = ss->can_attach(cgrp, &tset);
1998 goto out_cancel_attach;
2004 * step 2: make sure css_sets exist for all threads to be migrated.
2005 * we use find_css_set, which allocates a new one if necessary.
2007 for (i = 0; i < group_size; i++) {
2008 tc = flex_array_get(group, i);
2009 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2012 goto out_put_css_set_refs;
2017 * step 3: now that we're guaranteed success wrt the css_sets,
2018 * proceed to move all tasks to the new cgroup. There are no
2019 * failure cases after here, so this is the commit point.
2021 for (i = 0; i < group_size; i++) {
2022 tc = flex_array_get(group, i);
2023 cgroup_task_migrate(tc->cgrp, tc->task, tc->cg);
2025 /* nothing is sensitive to fork() after this point. */
2028 * step 4: do subsystem attach callbacks.
2030 for_each_subsys(root, ss) {
2032 ss->attach(cgrp, &tset);
2036 * step 5: success! and cleanup
2039 out_put_css_set_refs:
2041 for (i = 0; i < group_size; i++) {
2042 tc = flex_array_get(group, i);
2045 put_css_set(tc->cg);
2050 for_each_subsys(root, ss) {
2051 if (ss == failed_ss)
2053 if (ss->cancel_attach)
2054 ss->cancel_attach(cgrp, &tset);
2057 out_free_group_list:
2058 flex_array_free(group);
2063 * Find the task_struct of the task to attach by vpid and pass it along to the
2064 * function to attach either it or all tasks in its threadgroup. Will lock
2065 * cgroup_mutex and threadgroup; may take task_lock of task.
2067 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2069 struct task_struct *tsk;
2070 const struct cred *cred = current_cred(), *tcred;
2073 if (!cgroup_lock_live_group(cgrp))
2079 tsk = find_task_by_vpid(pid);
2083 goto out_unlock_cgroup;
2086 * even if we're attaching all tasks in the thread group, we
2087 * only need to check permissions on one of them.
2089 tcred = __task_cred(tsk);
2090 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2091 !uid_eq(cred->euid, tcred->uid) &&
2092 !uid_eq(cred->euid, tcred->suid)) {
2095 goto out_unlock_cgroup;
2101 tsk = tsk->group_leader;
2104 * Workqueue threads may acquire PF_THREAD_BOUND and become
2105 * trapped in a cpuset, or RT worker may be born in a cgroup
2106 * with no rt_runtime allocated. Just say no.
2108 if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2111 goto out_unlock_cgroup;
2114 get_task_struct(tsk);
2117 threadgroup_lock(tsk);
2119 if (!thread_group_leader(tsk)) {
2121 * a race with de_thread from another thread's exec()
2122 * may strip us of our leadership, if this happens,
2123 * there is no choice but to throw this task away and
2124 * try again; this is
2125 * "double-double-toil-and-trouble-check locking".
2127 threadgroup_unlock(tsk);
2128 put_task_struct(tsk);
2129 goto retry_find_task;
2133 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2135 threadgroup_unlock(tsk);
2137 put_task_struct(tsk);
2139 mutex_unlock(&cgroup_mutex);
2144 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2145 * @from: attach to all cgroups of a given task
2146 * @tsk: the task to be attached
2148 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2150 struct cgroupfs_root *root;
2153 mutex_lock(&cgroup_mutex);
2154 for_each_active_root(root) {
2155 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2157 retval = cgroup_attach_task(from_cg, tsk, false);
2161 mutex_unlock(&cgroup_mutex);
2165 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2167 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2169 return attach_task_by_pid(cgrp, pid, false);
2172 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2174 return attach_task_by_pid(cgrp, tgid, true);
2177 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2180 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2181 if (strlen(buffer) >= PATH_MAX)
2183 if (!cgroup_lock_live_group(cgrp))
2185 mutex_lock(&cgroup_root_mutex);
2186 strcpy(cgrp->root->release_agent_path, buffer);
2187 mutex_unlock(&cgroup_root_mutex);
2188 mutex_unlock(&cgroup_mutex);
2192 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2193 struct seq_file *seq)
2195 if (!cgroup_lock_live_group(cgrp))
2197 seq_puts(seq, cgrp->root->release_agent_path);
2198 seq_putc(seq, '\n');
2199 mutex_unlock(&cgroup_mutex);
2203 /* A buffer size big enough for numbers or short strings */
2204 #define CGROUP_LOCAL_BUFFER_SIZE 64
2206 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2208 const char __user *userbuf,
2209 size_t nbytes, loff_t *unused_ppos)
2211 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2217 if (nbytes >= sizeof(buffer))
2219 if (copy_from_user(buffer, userbuf, nbytes))
2222 buffer[nbytes] = 0; /* nul-terminate */
2223 if (cft->write_u64) {
2224 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2227 retval = cft->write_u64(cgrp, cft, val);
2229 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2232 retval = cft->write_s64(cgrp, cft, val);
2239 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2241 const char __user *userbuf,
2242 size_t nbytes, loff_t *unused_ppos)
2244 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2246 size_t max_bytes = cft->max_write_len;
2247 char *buffer = local_buffer;
2250 max_bytes = sizeof(local_buffer) - 1;
2251 if (nbytes >= max_bytes)
2253 /* Allocate a dynamic buffer if we need one */
2254 if (nbytes >= sizeof(local_buffer)) {
2255 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2259 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2264 buffer[nbytes] = 0; /* nul-terminate */
2265 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2269 if (buffer != local_buffer)
2274 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2275 size_t nbytes, loff_t *ppos)
2277 struct cftype *cft = __d_cft(file->f_dentry);
2278 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2280 if (cgroup_is_removed(cgrp))
2283 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2284 if (cft->write_u64 || cft->write_s64)
2285 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2286 if (cft->write_string)
2287 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2289 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2290 return ret ? ret : nbytes;
2295 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2297 char __user *buf, size_t nbytes,
2300 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2301 u64 val = cft->read_u64(cgrp, cft);
2302 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2304 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2307 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2309 char __user *buf, size_t nbytes,
2312 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2313 s64 val = cft->read_s64(cgrp, cft);
2314 int len = sprintf(tmp, "%lld\n", (long long) val);
2316 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2319 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2320 size_t nbytes, loff_t *ppos)
2322 struct cftype *cft = __d_cft(file->f_dentry);
2323 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2325 if (cgroup_is_removed(cgrp))
2329 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2331 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2333 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2338 * seqfile ops/methods for returning structured data. Currently just
2339 * supports string->u64 maps, but can be extended in future.
2342 struct cgroup_seqfile_state {
2344 struct cgroup *cgroup;
2347 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2349 struct seq_file *sf = cb->state;
2350 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2353 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2355 struct cgroup_seqfile_state *state = m->private;
2356 struct cftype *cft = state->cft;
2357 if (cft->read_map) {
2358 struct cgroup_map_cb cb = {
2359 .fill = cgroup_map_add,
2362 return cft->read_map(state->cgroup, cft, &cb);
2364 return cft->read_seq_string(state->cgroup, cft, m);
2367 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2369 struct seq_file *seq = file->private_data;
2370 kfree(seq->private);
2371 return single_release(inode, file);
2374 static const struct file_operations cgroup_seqfile_operations = {
2376 .write = cgroup_file_write,
2377 .llseek = seq_lseek,
2378 .release = cgroup_seqfile_release,
2381 static int cgroup_file_open(struct inode *inode, struct file *file)
2386 err = generic_file_open(inode, file);
2389 cft = __d_cft(file->f_dentry);
2391 if (cft->read_map || cft->read_seq_string) {
2392 struct cgroup_seqfile_state *state =
2393 kzalloc(sizeof(*state), GFP_USER);
2397 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2398 file->f_op = &cgroup_seqfile_operations;
2399 err = single_open(file, cgroup_seqfile_show, state);
2402 } else if (cft->open)
2403 err = cft->open(inode, file);
2410 static int cgroup_file_release(struct inode *inode, struct file *file)
2412 struct cftype *cft = __d_cft(file->f_dentry);
2414 return cft->release(inode, file);
2419 * cgroup_rename - Only allow simple rename of directories in place.
2421 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2422 struct inode *new_dir, struct dentry *new_dentry)
2425 struct cgroup_name *name, *old_name;
2426 struct cgroup *cgrp;
2429 * It's convinient to use parent dir's i_mutex to protected
2432 lockdep_assert_held(&old_dir->i_mutex);
2434 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2436 if (new_dentry->d_inode)
2438 if (old_dir != new_dir)
2441 cgrp = __d_cgrp(old_dentry);
2443 name = cgroup_alloc_name(new_dentry);
2447 ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2453 old_name = cgrp->name;
2454 rcu_assign_pointer(cgrp->name, name);
2456 kfree_rcu(old_name, rcu_head);
2460 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2462 if (S_ISDIR(dentry->d_inode->i_mode))
2463 return &__d_cgrp(dentry)->xattrs;
2465 return &__d_cft(dentry)->xattrs;
2468 static inline int xattr_enabled(struct dentry *dentry)
2470 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2471 return root->flags & CGRP_ROOT_XATTR;
2474 static bool is_valid_xattr(const char *name)
2476 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2477 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2482 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2483 const void *val, size_t size, int flags)
2485 if (!xattr_enabled(dentry))
2487 if (!is_valid_xattr(name))
2489 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2492 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2494 if (!xattr_enabled(dentry))
2496 if (!is_valid_xattr(name))
2498 return simple_xattr_remove(__d_xattrs(dentry), name);
2501 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2502 void *buf, size_t size)
2504 if (!xattr_enabled(dentry))
2506 if (!is_valid_xattr(name))
2508 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2511 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2513 if (!xattr_enabled(dentry))
2515 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2518 static const struct file_operations cgroup_file_operations = {
2519 .read = cgroup_file_read,
2520 .write = cgroup_file_write,
2521 .llseek = generic_file_llseek,
2522 .open = cgroup_file_open,
2523 .release = cgroup_file_release,
2526 static const struct inode_operations cgroup_file_inode_operations = {
2527 .setxattr = cgroup_setxattr,
2528 .getxattr = cgroup_getxattr,
2529 .listxattr = cgroup_listxattr,
2530 .removexattr = cgroup_removexattr,
2533 static const struct inode_operations cgroup_dir_inode_operations = {
2534 .lookup = cgroup_lookup,
2535 .mkdir = cgroup_mkdir,
2536 .rmdir = cgroup_rmdir,
2537 .rename = cgroup_rename,
2538 .setxattr = cgroup_setxattr,
2539 .getxattr = cgroup_getxattr,
2540 .listxattr = cgroup_listxattr,
2541 .removexattr = cgroup_removexattr,
2544 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2546 if (dentry->d_name.len > NAME_MAX)
2547 return ERR_PTR(-ENAMETOOLONG);
2548 d_add(dentry, NULL);
2553 * Check if a file is a control file
2555 static inline struct cftype *__file_cft(struct file *file)
2557 if (file_inode(file)->i_fop != &cgroup_file_operations)
2558 return ERR_PTR(-EINVAL);
2559 return __d_cft(file->f_dentry);
2562 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2563 struct super_block *sb)
2565 struct inode *inode;
2569 if (dentry->d_inode)
2572 inode = cgroup_new_inode(mode, sb);
2576 if (S_ISDIR(mode)) {
2577 inode->i_op = &cgroup_dir_inode_operations;
2578 inode->i_fop = &simple_dir_operations;
2580 /* start off with i_nlink == 2 (for "." entry) */
2582 inc_nlink(dentry->d_parent->d_inode);
2585 * Control reaches here with cgroup_mutex held.
2586 * @inode->i_mutex should nest outside cgroup_mutex but we
2587 * want to populate it immediately without releasing
2588 * cgroup_mutex. As @inode isn't visible to anyone else
2589 * yet, trylock will always succeed without affecting
2592 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2593 } else if (S_ISREG(mode)) {
2595 inode->i_fop = &cgroup_file_operations;
2596 inode->i_op = &cgroup_file_inode_operations;
2598 d_instantiate(dentry, inode);
2599 dget(dentry); /* Extra count - pin the dentry in core */
2604 * cgroup_file_mode - deduce file mode of a control file
2605 * @cft: the control file in question
2607 * returns cft->mode if ->mode is not 0
2608 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2609 * returns S_IRUGO if it has only a read handler
2610 * returns S_IWUSR if it has only a write hander
2612 static umode_t cgroup_file_mode(const struct cftype *cft)
2619 if (cft->read || cft->read_u64 || cft->read_s64 ||
2620 cft->read_map || cft->read_seq_string)
2623 if (cft->write || cft->write_u64 || cft->write_s64 ||
2624 cft->write_string || cft->trigger)
2630 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2633 struct dentry *dir = cgrp->dentry;
2634 struct cgroup *parent = __d_cgrp(dir);
2635 struct dentry *dentry;
2639 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2641 simple_xattrs_init(&cft->xattrs);
2643 if (subsys && !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
2644 strcpy(name, subsys->name);
2647 strcat(name, cft->name);
2649 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2651 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2655 dentry = lookup_one_len(name, dir, strlen(name));
2656 if (IS_ERR(dentry)) {
2657 error = PTR_ERR(dentry);
2661 mode = cgroup_file_mode(cft);
2662 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2664 cfe->type = (void *)cft;
2665 cfe->dentry = dentry;
2666 dentry->d_fsdata = cfe;
2667 list_add_tail(&cfe->node, &parent->files);
2676 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2677 struct cftype cfts[], bool is_add)
2682 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2683 /* does cft->flags tell us to skip this file on @cgrp? */
2684 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2686 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2690 err = cgroup_add_file(cgrp, subsys, cft);
2692 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2696 cgroup_rm_file(cgrp, cft);
2702 static DEFINE_MUTEX(cgroup_cft_mutex);
2704 static void cgroup_cfts_prepare(void)
2705 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2708 * Thanks to the entanglement with vfs inode locking, we can't walk
2709 * the existing cgroups under cgroup_mutex and create files.
2710 * Instead, we increment reference on all cgroups and build list of
2711 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2712 * exclusive access to the field.
2714 mutex_lock(&cgroup_cft_mutex);
2715 mutex_lock(&cgroup_mutex);
2718 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2719 struct cftype *cfts, bool is_add)
2720 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2723 struct cgroup *cgrp, *n;
2725 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2726 if (cfts && ss->root != &rootnode) {
2727 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2729 list_add_tail(&cgrp->cft_q_node, &pending);
2733 mutex_unlock(&cgroup_mutex);
2736 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2737 * files for all cgroups which were created before.
2739 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2740 struct inode *inode = cgrp->dentry->d_inode;
2742 mutex_lock(&inode->i_mutex);
2743 mutex_lock(&cgroup_mutex);
2744 if (!cgroup_is_removed(cgrp))
2745 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2746 mutex_unlock(&cgroup_mutex);
2747 mutex_unlock(&inode->i_mutex);
2749 list_del_init(&cgrp->cft_q_node);
2753 mutex_unlock(&cgroup_cft_mutex);
2757 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2758 * @ss: target cgroup subsystem
2759 * @cfts: zero-length name terminated array of cftypes
2761 * Register @cfts to @ss. Files described by @cfts are created for all
2762 * existing cgroups to which @ss is attached and all future cgroups will
2763 * have them too. This function can be called anytime whether @ss is
2766 * Returns 0 on successful registration, -errno on failure. Note that this
2767 * function currently returns 0 as long as @cfts registration is successful
2768 * even if some file creation attempts on existing cgroups fail.
2770 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2772 struct cftype_set *set;
2774 set = kzalloc(sizeof(*set), GFP_KERNEL);
2778 cgroup_cfts_prepare();
2780 list_add_tail(&set->node, &ss->cftsets);
2781 cgroup_cfts_commit(ss, cfts, true);
2785 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2788 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2789 * @ss: target cgroup subsystem
2790 * @cfts: zero-length name terminated array of cftypes
2792 * Unregister @cfts from @ss. Files described by @cfts are removed from
2793 * all existing cgroups to which @ss is attached and all future cgroups
2794 * won't have them either. This function can be called anytime whether @ss
2795 * is attached or not.
2797 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2798 * registered with @ss.
2800 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2802 struct cftype_set *set;
2804 cgroup_cfts_prepare();
2806 list_for_each_entry(set, &ss->cftsets, node) {
2807 if (set->cfts == cfts) {
2808 list_del_init(&set->node);
2809 cgroup_cfts_commit(ss, cfts, false);
2814 cgroup_cfts_commit(ss, NULL, false);
2819 * cgroup_task_count - count the number of tasks in a cgroup.
2820 * @cgrp: the cgroup in question
2822 * Return the number of tasks in the cgroup.
2824 int cgroup_task_count(const struct cgroup *cgrp)
2827 struct cg_cgroup_link *link;
2829 read_lock(&css_set_lock);
2830 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2831 count += atomic_read(&link->cg->refcount);
2833 read_unlock(&css_set_lock);
2838 * Advance a list_head iterator. The iterator should be positioned at
2839 * the start of a css_set
2841 static void cgroup_advance_iter(struct cgroup *cgrp,
2842 struct cgroup_iter *it)
2844 struct list_head *l = it->cg_link;
2845 struct cg_cgroup_link *link;
2848 /* Advance to the next non-empty css_set */
2851 if (l == &cgrp->css_sets) {
2855 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2857 } while (list_empty(&cg->tasks));
2859 it->task = cg->tasks.next;
2863 * To reduce the fork() overhead for systems that are not actually
2864 * using their cgroups capability, we don't maintain the lists running
2865 * through each css_set to its tasks until we see the list actually
2866 * used - in other words after the first call to cgroup_iter_start().
2868 static void cgroup_enable_task_cg_lists(void)
2870 struct task_struct *p, *g;
2871 write_lock(&css_set_lock);
2872 use_task_css_set_links = 1;
2874 * We need tasklist_lock because RCU is not safe against
2875 * while_each_thread(). Besides, a forking task that has passed
2876 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2877 * is not guaranteed to have its child immediately visible in the
2878 * tasklist if we walk through it with RCU.
2880 read_lock(&tasklist_lock);
2881 do_each_thread(g, p) {
2884 * We should check if the process is exiting, otherwise
2885 * it will race with cgroup_exit() in that the list
2886 * entry won't be deleted though the process has exited.
2888 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2889 list_add(&p->cg_list, &p->cgroups->tasks);
2891 } while_each_thread(g, p);
2892 read_unlock(&tasklist_lock);
2893 write_unlock(&css_set_lock);
2897 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2898 * @pos: the current position (%NULL to initiate traversal)
2899 * @cgroup: cgroup whose descendants to walk
2901 * To be used by cgroup_for_each_descendant_pre(). Find the next
2902 * descendant to visit for pre-order traversal of @cgroup's descendants.
2904 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
2905 struct cgroup *cgroup)
2907 struct cgroup *next;
2909 WARN_ON_ONCE(!rcu_read_lock_held());
2911 /* if first iteration, pretend we just visited @cgroup */
2913 if (list_empty(&cgroup->children))
2918 /* visit the first child if exists */
2919 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
2923 /* no child, visit my or the closest ancestor's next sibling */
2925 next = list_entry_rcu(pos->sibling.next, struct cgroup,
2927 if (&next->sibling != &pos->parent->children)
2931 } while (pos != cgroup);
2935 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
2938 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
2939 * @pos: cgroup of interest
2941 * Return the rightmost descendant of @pos. If there's no descendant,
2942 * @pos is returned. This can be used during pre-order traversal to skip
2945 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
2947 struct cgroup *last, *tmp;
2949 WARN_ON_ONCE(!rcu_read_lock_held());
2953 /* ->prev isn't RCU safe, walk ->next till the end */
2955 list_for_each_entry_rcu(tmp, &last->children, sibling)
2961 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
2963 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
2965 struct cgroup *last;
2969 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
2977 * cgroup_next_descendant_post - find the next descendant for post-order walk
2978 * @pos: the current position (%NULL to initiate traversal)
2979 * @cgroup: cgroup whose descendants to walk
2981 * To be used by cgroup_for_each_descendant_post(). Find the next
2982 * descendant to visit for post-order traversal of @cgroup's descendants.
2984 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
2985 struct cgroup *cgroup)
2987 struct cgroup *next;
2989 WARN_ON_ONCE(!rcu_read_lock_held());
2991 /* if first iteration, visit the leftmost descendant */
2993 next = cgroup_leftmost_descendant(cgroup);
2994 return next != cgroup ? next : NULL;
2997 /* if there's an unvisited sibling, visit its leftmost descendant */
2998 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
2999 if (&next->sibling != &pos->parent->children)
3000 return cgroup_leftmost_descendant(next);
3002 /* no sibling left, visit parent */
3004 return next != cgroup ? next : NULL;
3006 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3008 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3009 __acquires(css_set_lock)
3012 * The first time anyone tries to iterate across a cgroup,
3013 * we need to enable the list linking each css_set to its
3014 * tasks, and fix up all existing tasks.
3016 if (!use_task_css_set_links)
3017 cgroup_enable_task_cg_lists();
3019 read_lock(&css_set_lock);
3020 it->cg_link = &cgrp->css_sets;
3021 cgroup_advance_iter(cgrp, it);
3024 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3025 struct cgroup_iter *it)
3027 struct task_struct *res;
3028 struct list_head *l = it->task;
3029 struct cg_cgroup_link *link;
3031 /* If the iterator cg is NULL, we have no tasks */
3034 res = list_entry(l, struct task_struct, cg_list);
3035 /* Advance iterator to find next entry */
3037 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
3038 if (l == &link->cg->tasks) {
3039 /* We reached the end of this task list - move on to
3040 * the next cg_cgroup_link */
3041 cgroup_advance_iter(cgrp, it);
3048 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3049 __releases(css_set_lock)
3051 read_unlock(&css_set_lock);
3054 static inline int started_after_time(struct task_struct *t1,
3055 struct timespec *time,
3056 struct task_struct *t2)
3058 int start_diff = timespec_compare(&t1->start_time, time);
3059 if (start_diff > 0) {
3061 } else if (start_diff < 0) {
3065 * Arbitrarily, if two processes started at the same
3066 * time, we'll say that the lower pointer value
3067 * started first. Note that t2 may have exited by now
3068 * so this may not be a valid pointer any longer, but
3069 * that's fine - it still serves to distinguish
3070 * between two tasks started (effectively) simultaneously.
3077 * This function is a callback from heap_insert() and is used to order
3079 * In this case we order the heap in descending task start time.
3081 static inline int started_after(void *p1, void *p2)
3083 struct task_struct *t1 = p1;
3084 struct task_struct *t2 = p2;
3085 return started_after_time(t1, &t2->start_time, t2);
3089 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3090 * @scan: struct cgroup_scanner containing arguments for the scan
3092 * Arguments include pointers to callback functions test_task() and
3094 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3095 * and if it returns true, call process_task() for it also.
3096 * The test_task pointer may be NULL, meaning always true (select all tasks).
3097 * Effectively duplicates cgroup_iter_{start,next,end}()
3098 * but does not lock css_set_lock for the call to process_task().
3099 * The struct cgroup_scanner may be embedded in any structure of the caller's
3101 * It is guaranteed that process_task() will act on every task that
3102 * is a member of the cgroup for the duration of this call. This
3103 * function may or may not call process_task() for tasks that exit
3104 * or move to a different cgroup during the call, or are forked or
3105 * move into the cgroup during the call.
3107 * Note that test_task() may be called with locks held, and may in some
3108 * situations be called multiple times for the same task, so it should
3110 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3111 * pre-allocated and will be used for heap operations (and its "gt" member will
3112 * be overwritten), else a temporary heap will be used (allocation of which
3113 * may cause this function to fail).
3115 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3118 struct cgroup_iter it;
3119 struct task_struct *p, *dropped;
3120 /* Never dereference latest_task, since it's not refcounted */
3121 struct task_struct *latest_task = NULL;
3122 struct ptr_heap tmp_heap;
3123 struct ptr_heap *heap;
3124 struct timespec latest_time = { 0, 0 };
3127 /* The caller supplied our heap and pre-allocated its memory */
3129 heap->gt = &started_after;
3131 /* We need to allocate our own heap memory */
3133 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3135 /* cannot allocate the heap */
3141 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3142 * to determine which are of interest, and using the scanner's
3143 * "process_task" callback to process any of them that need an update.
3144 * Since we don't want to hold any locks during the task updates,
3145 * gather tasks to be processed in a heap structure.
3146 * The heap is sorted by descending task start time.
3147 * If the statically-sized heap fills up, we overflow tasks that
3148 * started later, and in future iterations only consider tasks that
3149 * started after the latest task in the previous pass. This
3150 * guarantees forward progress and that we don't miss any tasks.
3153 cgroup_iter_start(scan->cg, &it);
3154 while ((p = cgroup_iter_next(scan->cg, &it))) {
3156 * Only affect tasks that qualify per the caller's callback,
3157 * if he provided one
3159 if (scan->test_task && !scan->test_task(p, scan))
3162 * Only process tasks that started after the last task
3165 if (!started_after_time(p, &latest_time, latest_task))
3167 dropped = heap_insert(heap, p);
3168 if (dropped == NULL) {
3170 * The new task was inserted; the heap wasn't
3174 } else if (dropped != p) {
3176 * The new task was inserted, and pushed out a
3180 put_task_struct(dropped);
3183 * Else the new task was newer than anything already in
3184 * the heap and wasn't inserted
3187 cgroup_iter_end(scan->cg, &it);
3190 for (i = 0; i < heap->size; i++) {
3191 struct task_struct *q = heap->ptrs[i];
3193 latest_time = q->start_time;
3196 /* Process the task per the caller's callback */
3197 scan->process_task(q, scan);
3201 * If we had to process any tasks at all, scan again
3202 * in case some of them were in the middle of forking
3203 * children that didn't get processed.
3204 * Not the most efficient way to do it, but it avoids
3205 * having to take callback_mutex in the fork path
3209 if (heap == &tmp_heap)
3210 heap_free(&tmp_heap);
3214 static void cgroup_transfer_one_task(struct task_struct *task,
3215 struct cgroup_scanner *scan)
3217 struct cgroup *new_cgroup = scan->data;
3219 mutex_lock(&cgroup_mutex);
3220 cgroup_attach_task(new_cgroup, task, false);
3221 mutex_unlock(&cgroup_mutex);
3225 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3226 * @to: cgroup to which the tasks will be moved
3227 * @from: cgroup in which the tasks currently reside
3229 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3231 struct cgroup_scanner scan;
3234 scan.test_task = NULL; /* select all tasks in cgroup */
3235 scan.process_task = cgroup_transfer_one_task;
3239 return cgroup_scan_tasks(&scan);
3243 * Stuff for reading the 'tasks'/'procs' files.
3245 * Reading this file can return large amounts of data if a cgroup has
3246 * *lots* of attached tasks. So it may need several calls to read(),
3247 * but we cannot guarantee that the information we produce is correct
3248 * unless we produce it entirely atomically.
3252 /* which pidlist file are we talking about? */
3253 enum cgroup_filetype {
3259 * A pidlist is a list of pids that virtually represents the contents of one
3260 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3261 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3264 struct cgroup_pidlist {
3266 * used to find which pidlist is wanted. doesn't change as long as
3267 * this particular list stays in the list.
3269 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3272 /* how many elements the above list has */
3274 /* how many files are using the current array */
3276 /* each of these stored in a list by its cgroup */
3277 struct list_head links;
3278 /* pointer to the cgroup we belong to, for list removal purposes */
3279 struct cgroup *owner;
3280 /* protects the other fields */
3281 struct rw_semaphore mutex;
3285 * The following two functions "fix" the issue where there are more pids
3286 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3287 * TODO: replace with a kernel-wide solution to this problem
3289 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3290 static void *pidlist_allocate(int count)
3292 if (PIDLIST_TOO_LARGE(count))
3293 return vmalloc(count * sizeof(pid_t));
3295 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3297 static void pidlist_free(void *p)
3299 if (is_vmalloc_addr(p))
3306 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3307 * Returns the number of unique elements.
3309 static int pidlist_uniq(pid_t *list, int length)
3314 * we presume the 0th element is unique, so i starts at 1. trivial
3315 * edge cases first; no work needs to be done for either
3317 if (length == 0 || length == 1)
3319 /* src and dest walk down the list; dest counts unique elements */
3320 for (src = 1; src < length; src++) {
3321 /* find next unique element */
3322 while (list[src] == list[src-1]) {
3327 /* dest always points to where the next unique element goes */
3328 list[dest] = list[src];
3335 static int cmppid(const void *a, const void *b)
3337 return *(pid_t *)a - *(pid_t *)b;
3341 * find the appropriate pidlist for our purpose (given procs vs tasks)
3342 * returns with the lock on that pidlist already held, and takes care
3343 * of the use count, or returns NULL with no locks held if we're out of
3346 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3347 enum cgroup_filetype type)
3349 struct cgroup_pidlist *l;
3350 /* don't need task_nsproxy() if we're looking at ourself */
3351 struct pid_namespace *ns = task_active_pid_ns(current);
3354 * We can't drop the pidlist_mutex before taking the l->mutex in case
3355 * the last ref-holder is trying to remove l from the list at the same
3356 * time. Holding the pidlist_mutex precludes somebody taking whichever
3357 * list we find out from under us - compare release_pid_array().
3359 mutex_lock(&cgrp->pidlist_mutex);
3360 list_for_each_entry(l, &cgrp->pidlists, links) {
3361 if (l->key.type == type && l->key.ns == ns) {
3362 /* make sure l doesn't vanish out from under us */
3363 down_write(&l->mutex);
3364 mutex_unlock(&cgrp->pidlist_mutex);
3368 /* entry not found; create a new one */
3369 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3371 mutex_unlock(&cgrp->pidlist_mutex);
3374 init_rwsem(&l->mutex);
3375 down_write(&l->mutex);
3377 l->key.ns = get_pid_ns(ns);
3378 l->use_count = 0; /* don't increment here */
3381 list_add(&l->links, &cgrp->pidlists);
3382 mutex_unlock(&cgrp->pidlist_mutex);
3387 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3389 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3390 struct cgroup_pidlist **lp)
3394 int pid, n = 0; /* used for populating the array */
3395 struct cgroup_iter it;
3396 struct task_struct *tsk;
3397 struct cgroup_pidlist *l;
3400 * If cgroup gets more users after we read count, we won't have
3401 * enough space - tough. This race is indistinguishable to the
3402 * caller from the case that the additional cgroup users didn't
3403 * show up until sometime later on.
3405 length = cgroup_task_count(cgrp);
3406 array = pidlist_allocate(length);
3409 /* now, populate the array */
3410 cgroup_iter_start(cgrp, &it);
3411 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3412 if (unlikely(n == length))
3414 /* get tgid or pid for procs or tasks file respectively */
3415 if (type == CGROUP_FILE_PROCS)
3416 pid = task_tgid_vnr(tsk);
3418 pid = task_pid_vnr(tsk);
3419 if (pid > 0) /* make sure to only use valid results */
3422 cgroup_iter_end(cgrp, &it);
3424 /* now sort & (if procs) strip out duplicates */
3425 sort(array, length, sizeof(pid_t), cmppid, NULL);
3426 if (type == CGROUP_FILE_PROCS)
3427 length = pidlist_uniq(array, length);
3428 l = cgroup_pidlist_find(cgrp, type);
3430 pidlist_free(array);
3433 /* store array, freeing old if necessary - lock already held */
3434 pidlist_free(l->list);
3438 up_write(&l->mutex);
3444 * cgroupstats_build - build and fill cgroupstats
3445 * @stats: cgroupstats to fill information into
3446 * @dentry: A dentry entry belonging to the cgroup for which stats have
3449 * Build and fill cgroupstats so that taskstats can export it to user
3452 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3455 struct cgroup *cgrp;
3456 struct cgroup_iter it;
3457 struct task_struct *tsk;
3460 * Validate dentry by checking the superblock operations,
3461 * and make sure it's a directory.
3463 if (dentry->d_sb->s_op != &cgroup_ops ||
3464 !S_ISDIR(dentry->d_inode->i_mode))
3468 cgrp = dentry->d_fsdata;
3470 cgroup_iter_start(cgrp, &it);
3471 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3472 switch (tsk->state) {
3474 stats->nr_running++;
3476 case TASK_INTERRUPTIBLE:
3477 stats->nr_sleeping++;
3479 case TASK_UNINTERRUPTIBLE:
3480 stats->nr_uninterruptible++;
3483 stats->nr_stopped++;
3486 if (delayacct_is_task_waiting_on_io(tsk))
3487 stats->nr_io_wait++;
3491 cgroup_iter_end(cgrp, &it);
3499 * seq_file methods for the tasks/procs files. The seq_file position is the
3500 * next pid to display; the seq_file iterator is a pointer to the pid
3501 * in the cgroup->l->list array.
3504 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3507 * Initially we receive a position value that corresponds to
3508 * one more than the last pid shown (or 0 on the first call or
3509 * after a seek to the start). Use a binary-search to find the
3510 * next pid to display, if any
3512 struct cgroup_pidlist *l = s->private;
3513 int index = 0, pid = *pos;
3516 down_read(&l->mutex);
3518 int end = l->length;
3520 while (index < end) {
3521 int mid = (index + end) / 2;
3522 if (l->list[mid] == pid) {
3525 } else if (l->list[mid] <= pid)
3531 /* If we're off the end of the array, we're done */
3532 if (index >= l->length)
3534 /* Update the abstract position to be the actual pid that we found */
3535 iter = l->list + index;
3540 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3542 struct cgroup_pidlist *l = s->private;
3546 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3548 struct cgroup_pidlist *l = s->private;
3550 pid_t *end = l->list + l->length;
3552 * Advance to the next pid in the array. If this goes off the
3564 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3566 return seq_printf(s, "%d\n", *(int *)v);
3570 * seq_operations functions for iterating on pidlists through seq_file -
3571 * independent of whether it's tasks or procs
3573 static const struct seq_operations cgroup_pidlist_seq_operations = {
3574 .start = cgroup_pidlist_start,
3575 .stop = cgroup_pidlist_stop,
3576 .next = cgroup_pidlist_next,
3577 .show = cgroup_pidlist_show,
3580 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3583 * the case where we're the last user of this particular pidlist will
3584 * have us remove it from the cgroup's list, which entails taking the
3585 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3586 * pidlist_mutex, we have to take pidlist_mutex first.
3588 mutex_lock(&l->owner->pidlist_mutex);
3589 down_write(&l->mutex);
3590 BUG_ON(!l->use_count);
3591 if (!--l->use_count) {
3592 /* we're the last user if refcount is 0; remove and free */
3593 list_del(&l->links);
3594 mutex_unlock(&l->owner->pidlist_mutex);
3595 pidlist_free(l->list);
3596 put_pid_ns(l->key.ns);
3597 up_write(&l->mutex);
3601 mutex_unlock(&l->owner->pidlist_mutex);
3602 up_write(&l->mutex);
3605 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3607 struct cgroup_pidlist *l;
3608 if (!(file->f_mode & FMODE_READ))
3611 * the seq_file will only be initialized if the file was opened for
3612 * reading; hence we check if it's not null only in that case.
3614 l = ((struct seq_file *)file->private_data)->private;
3615 cgroup_release_pid_array(l);
3616 return seq_release(inode, file);
3619 static const struct file_operations cgroup_pidlist_operations = {
3621 .llseek = seq_lseek,
3622 .write = cgroup_file_write,
3623 .release = cgroup_pidlist_release,
3627 * The following functions handle opens on a file that displays a pidlist
3628 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3631 /* helper function for the two below it */
3632 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3634 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3635 struct cgroup_pidlist *l;
3638 /* Nothing to do for write-only files */
3639 if (!(file->f_mode & FMODE_READ))
3642 /* have the array populated */
3643 retval = pidlist_array_load(cgrp, type, &l);
3646 /* configure file information */
3647 file->f_op = &cgroup_pidlist_operations;
3649 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3651 cgroup_release_pid_array(l);
3654 ((struct seq_file *)file->private_data)->private = l;
3657 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3659 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3661 static int cgroup_procs_open(struct inode *unused, struct file *file)
3663 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3666 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3669 return notify_on_release(cgrp);
3672 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3676 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3678 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3680 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3685 * Unregister event and free resources.
3687 * Gets called from workqueue.
3689 static void cgroup_event_remove(struct work_struct *work)
3691 struct cgroup_event *event = container_of(work, struct cgroup_event,
3693 struct cgroup *cgrp = event->cgrp;
3695 remove_wait_queue(event->wqh, &event->wait);
3697 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3699 /* Notify userspace the event is going away. */
3700 eventfd_signal(event->eventfd, 1);
3702 eventfd_ctx_put(event->eventfd);
3708 * Gets called on POLLHUP on eventfd when user closes it.
3710 * Called with wqh->lock held and interrupts disabled.
3712 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3713 int sync, void *key)
3715 struct cgroup_event *event = container_of(wait,
3716 struct cgroup_event, wait);
3717 struct cgroup *cgrp = event->cgrp;
3718 unsigned long flags = (unsigned long)key;
3720 if (flags & POLLHUP) {
3722 * If the event has been detached at cgroup removal, we
3723 * can simply return knowing the other side will cleanup
3726 * We can't race against event freeing since the other
3727 * side will require wqh->lock via remove_wait_queue(),
3730 spin_lock(&cgrp->event_list_lock);
3731 if (!list_empty(&event->list)) {
3732 list_del_init(&event->list);
3734 * We are in atomic context, but cgroup_event_remove()
3735 * may sleep, so we have to call it in workqueue.
3737 schedule_work(&event->remove);
3739 spin_unlock(&cgrp->event_list_lock);
3745 static void cgroup_event_ptable_queue_proc(struct file *file,
3746 wait_queue_head_t *wqh, poll_table *pt)
3748 struct cgroup_event *event = container_of(pt,
3749 struct cgroup_event, pt);
3752 add_wait_queue(wqh, &event->wait);
3756 * Parse input and register new cgroup event handler.
3758 * Input must be in format '<event_fd> <control_fd> <args>'.
3759 * Interpretation of args is defined by control file implementation.
3761 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3764 struct cgroup_event *event = NULL;
3765 struct cgroup *cgrp_cfile;
3766 unsigned int efd, cfd;
3767 struct file *efile = NULL;
3768 struct file *cfile = NULL;
3772 efd = simple_strtoul(buffer, &endp, 10);
3777 cfd = simple_strtoul(buffer, &endp, 10);
3778 if ((*endp != ' ') && (*endp != '\0'))
3782 event = kzalloc(sizeof(*event), GFP_KERNEL);
3786 INIT_LIST_HEAD(&event->list);
3787 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3788 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3789 INIT_WORK(&event->remove, cgroup_event_remove);
3791 efile = eventfd_fget(efd);
3792 if (IS_ERR(efile)) {
3793 ret = PTR_ERR(efile);
3797 event->eventfd = eventfd_ctx_fileget(efile);
3798 if (IS_ERR(event->eventfd)) {
3799 ret = PTR_ERR(event->eventfd);
3809 /* the process need read permission on control file */
3810 /* AV: shouldn't we check that it's been opened for read instead? */
3811 ret = inode_permission(file_inode(cfile), MAY_READ);
3815 event->cft = __file_cft(cfile);
3816 if (IS_ERR(event->cft)) {
3817 ret = PTR_ERR(event->cft);
3822 * The file to be monitored must be in the same cgroup as
3823 * cgroup.event_control is.
3825 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
3826 if (cgrp_cfile != cgrp) {
3831 if (!event->cft->register_event || !event->cft->unregister_event) {
3836 ret = event->cft->register_event(cgrp, event->cft,
3837 event->eventfd, buffer);
3842 * Events should be removed after rmdir of cgroup directory, but before
3843 * destroying subsystem state objects. Let's take reference to cgroup
3844 * directory dentry to do that.
3848 spin_lock(&cgrp->event_list_lock);
3849 list_add(&event->list, &cgrp->event_list);
3850 spin_unlock(&cgrp->event_list_lock);
3861 if (event && event->eventfd && !IS_ERR(event->eventfd))
3862 eventfd_ctx_put(event->eventfd);
3864 if (!IS_ERR_OR_NULL(efile))
3872 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3875 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3878 static int cgroup_clone_children_write(struct cgroup *cgrp,
3883 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3885 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3890 * for the common functions, 'private' gives the type of file
3892 /* for hysterical raisins, we can't put this on the older files */
3893 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3894 static struct cftype files[] = {
3897 .open = cgroup_tasks_open,
3898 .write_u64 = cgroup_tasks_write,
3899 .release = cgroup_pidlist_release,
3900 .mode = S_IRUGO | S_IWUSR,
3903 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3904 .open = cgroup_procs_open,
3905 .write_u64 = cgroup_procs_write,
3906 .release = cgroup_pidlist_release,
3907 .mode = S_IRUGO | S_IWUSR,
3910 .name = "notify_on_release",
3911 .read_u64 = cgroup_read_notify_on_release,
3912 .write_u64 = cgroup_write_notify_on_release,
3915 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3916 .write_string = cgroup_write_event_control,
3920 .name = "cgroup.clone_children",
3921 .read_u64 = cgroup_clone_children_read,
3922 .write_u64 = cgroup_clone_children_write,
3925 .name = "release_agent",
3926 .flags = CFTYPE_ONLY_ON_ROOT,
3927 .read_seq_string = cgroup_release_agent_show,
3928 .write_string = cgroup_release_agent_write,
3929 .max_write_len = PATH_MAX,
3935 * cgroup_populate_dir - selectively creation of files in a directory
3936 * @cgrp: target cgroup
3937 * @base_files: true if the base files should be added
3938 * @subsys_mask: mask of the subsystem ids whose files should be added
3940 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
3941 unsigned long subsys_mask)
3944 struct cgroup_subsys *ss;
3947 err = cgroup_addrm_files(cgrp, NULL, files, true);
3952 /* process cftsets of each subsystem */
3953 for_each_subsys(cgrp->root, ss) {
3954 struct cftype_set *set;
3955 if (!test_bit(ss->subsys_id, &subsys_mask))
3958 list_for_each_entry(set, &ss->cftsets, node)
3959 cgroup_addrm_files(cgrp, ss, set->cfts, true);
3962 /* This cgroup is ready now */
3963 for_each_subsys(cgrp->root, ss) {
3964 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3966 * Update id->css pointer and make this css visible from
3967 * CSS ID functions. This pointer will be dereferened
3968 * from RCU-read-side without locks.
3971 rcu_assign_pointer(css->id->css, css);
3977 static void css_dput_fn(struct work_struct *work)
3979 struct cgroup_subsys_state *css =
3980 container_of(work, struct cgroup_subsys_state, dput_work);
3981 struct dentry *dentry = css->cgroup->dentry;
3982 struct super_block *sb = dentry->d_sb;
3984 atomic_inc(&sb->s_active);
3986 deactivate_super(sb);
3989 static void init_cgroup_css(struct cgroup_subsys_state *css,
3990 struct cgroup_subsys *ss,
3991 struct cgroup *cgrp)
3994 atomic_set(&css->refcnt, 1);
3997 if (cgrp == dummytop)
3998 css->flags |= CSS_ROOT;
3999 BUG_ON(cgrp->subsys[ss->subsys_id]);
4000 cgrp->subsys[ss->subsys_id] = css;
4003 * css holds an extra ref to @cgrp->dentry which is put on the last
4004 * css_put(). dput() requires process context, which css_put() may
4005 * be called without. @css->dput_work will be used to invoke
4006 * dput() asynchronously from css_put().
4008 INIT_WORK(&css->dput_work, css_dput_fn);
4011 /* invoke ->post_create() on a new CSS and mark it online if successful */
4012 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4016 lockdep_assert_held(&cgroup_mutex);
4019 ret = ss->css_online(cgrp);
4021 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4025 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4026 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4027 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4029 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4031 lockdep_assert_held(&cgroup_mutex);
4033 if (!(css->flags & CSS_ONLINE))
4036 if (ss->css_offline)
4037 ss->css_offline(cgrp);
4039 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4043 * cgroup_create - create a cgroup
4044 * @parent: cgroup that will be parent of the new cgroup
4045 * @dentry: dentry of the new cgroup
4046 * @mode: mode to set on new inode
4048 * Must be called with the mutex on the parent inode held
4050 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4053 struct cgroup *cgrp;
4054 struct cgroup_name *name;
4055 struct cgroupfs_root *root = parent->root;
4057 struct cgroup_subsys *ss;
4058 struct super_block *sb = root->sb;
4060 /* allocate the cgroup and its ID, 0 is reserved for the root */
4061 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4065 name = cgroup_alloc_name(dentry);
4068 rcu_assign_pointer(cgrp->name, name);
4070 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4075 * Only live parents can have children. Note that the liveliness
4076 * check isn't strictly necessary because cgroup_mkdir() and
4077 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4078 * anyway so that locking is contained inside cgroup proper and we
4079 * don't get nasty surprises if we ever grow another caller.
4081 if (!cgroup_lock_live_group(parent)) {
4086 /* Grab a reference on the superblock so the hierarchy doesn't
4087 * get deleted on unmount if there are child cgroups. This
4088 * can be done outside cgroup_mutex, since the sb can't
4089 * disappear while someone has an open control file on the
4091 atomic_inc(&sb->s_active);
4093 init_cgroup_housekeeping(cgrp);
4095 dentry->d_fsdata = cgrp;
4096 cgrp->dentry = dentry;
4098 cgrp->parent = parent;
4099 cgrp->root = parent->root;
4100 cgrp->top_cgroup = parent->top_cgroup;
4102 if (notify_on_release(parent))
4103 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4105 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4106 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4108 for_each_subsys(root, ss) {
4109 struct cgroup_subsys_state *css;
4111 css = ss->css_alloc(cgrp);
4116 init_cgroup_css(css, ss, cgrp);
4118 err = alloc_css_id(ss, parent, cgrp);
4125 * Create directory. cgroup_create_file() returns with the new
4126 * directory locked on success so that it can be populated without
4127 * dropping cgroup_mutex.
4129 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4132 lockdep_assert_held(&dentry->d_inode->i_mutex);
4134 /* allocation complete, commit to creation */
4135 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4136 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4137 root->number_of_cgroups++;
4139 /* each css holds a ref to the cgroup's dentry */
4140 for_each_subsys(root, ss)
4143 /* hold a ref to the parent's dentry */
4144 dget(parent->dentry);
4146 /* creation succeeded, notify subsystems */
4147 for_each_subsys(root, ss) {
4148 err = online_css(ss, cgrp);
4152 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4154 pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4155 current->comm, current->pid, ss->name);
4156 if (!strcmp(ss->name, "memory"))
4157 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4158 ss->warned_broken_hierarchy = true;
4162 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4166 mutex_unlock(&cgroup_mutex);
4167 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4172 for_each_subsys(root, ss) {
4173 if (cgrp->subsys[ss->subsys_id])
4176 mutex_unlock(&cgroup_mutex);
4177 /* Release the reference count that we took on the superblock */
4178 deactivate_super(sb);
4180 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4182 kfree(rcu_dereference_raw(cgrp->name));
4188 cgroup_destroy_locked(cgrp);
4189 mutex_unlock(&cgroup_mutex);
4190 mutex_unlock(&dentry->d_inode->i_mutex);
4194 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4196 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4198 /* the vfs holds inode->i_mutex already */
4199 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4202 static int cgroup_destroy_locked(struct cgroup *cgrp)
4203 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4205 struct dentry *d = cgrp->dentry;
4206 struct cgroup *parent = cgrp->parent;
4207 struct cgroup_event *event, *tmp;
4208 struct cgroup_subsys *ss;
4210 lockdep_assert_held(&d->d_inode->i_mutex);
4211 lockdep_assert_held(&cgroup_mutex);
4213 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children))
4217 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4218 * removed. This makes future css_tryget() and child creation
4219 * attempts fail thus maintaining the removal conditions verified
4222 for_each_subsys(cgrp->root, ss) {
4223 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4225 WARN_ON(atomic_read(&css->refcnt) < 0);
4226 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4228 set_bit(CGRP_REMOVED, &cgrp->flags);
4230 /* tell subsystems to initate destruction */
4231 for_each_subsys(cgrp->root, ss)
4232 offline_css(ss, cgrp);
4235 * Put all the base refs. Each css holds an extra reference to the
4236 * cgroup's dentry and cgroup removal proceeds regardless of css
4237 * refs. On the last put of each css, whenever that may be, the
4238 * extra dentry ref is put so that dentry destruction happens only
4239 * after all css's are released.
4241 for_each_subsys(cgrp->root, ss)
4242 css_put(cgrp->subsys[ss->subsys_id]);
4244 raw_spin_lock(&release_list_lock);
4245 if (!list_empty(&cgrp->release_list))
4246 list_del_init(&cgrp->release_list);
4247 raw_spin_unlock(&release_list_lock);
4249 /* delete this cgroup from parent->children */
4250 list_del_rcu(&cgrp->sibling);
4251 list_del_init(&cgrp->allcg_node);
4254 cgroup_d_remove_dir(d);
4257 set_bit(CGRP_RELEASABLE, &parent->flags);
4258 check_for_release(parent);
4261 * Unregister events and notify userspace.
4262 * Notify userspace about cgroup removing only after rmdir of cgroup
4263 * directory to avoid race between userspace and kernelspace.
4265 spin_lock(&cgrp->event_list_lock);
4266 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4267 list_del_init(&event->list);
4268 schedule_work(&event->remove);
4270 spin_unlock(&cgrp->event_list_lock);
4275 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4279 mutex_lock(&cgroup_mutex);
4280 ret = cgroup_destroy_locked(dentry->d_fsdata);
4281 mutex_unlock(&cgroup_mutex);
4286 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4288 INIT_LIST_HEAD(&ss->cftsets);
4291 * base_cftset is embedded in subsys itself, no need to worry about
4294 if (ss->base_cftypes) {
4295 ss->base_cftset.cfts = ss->base_cftypes;
4296 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4300 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4302 struct cgroup_subsys_state *css;
4304 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4306 mutex_lock(&cgroup_mutex);
4308 /* init base cftset */
4309 cgroup_init_cftsets(ss);
4311 /* Create the top cgroup state for this subsystem */
4312 list_add(&ss->sibling, &rootnode.subsys_list);
4313 ss->root = &rootnode;
4314 css = ss->css_alloc(dummytop);
4315 /* We don't handle early failures gracefully */
4316 BUG_ON(IS_ERR(css));
4317 init_cgroup_css(css, ss, dummytop);
4319 /* Update the init_css_set to contain a subsys
4320 * pointer to this state - since the subsystem is
4321 * newly registered, all tasks and hence the
4322 * init_css_set is in the subsystem's top cgroup. */
4323 init_css_set.subsys[ss->subsys_id] = css;
4325 need_forkexit_callback |= ss->fork || ss->exit;
4327 /* At system boot, before all subsystems have been
4328 * registered, no tasks have been forked, so we don't
4329 * need to invoke fork callbacks here. */
4330 BUG_ON(!list_empty(&init_task.tasks));
4333 BUG_ON(online_css(ss, dummytop));
4335 mutex_unlock(&cgroup_mutex);
4337 /* this function shouldn't be used with modular subsystems, since they
4338 * need to register a subsys_id, among other things */
4343 * cgroup_load_subsys: load and register a modular subsystem at runtime
4344 * @ss: the subsystem to load
4346 * This function should be called in a modular subsystem's initcall. If the
4347 * subsystem is built as a module, it will be assigned a new subsys_id and set
4348 * up for use. If the subsystem is built-in anyway, work is delegated to the
4349 * simpler cgroup_init_subsys.
4351 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4353 struct cgroup_subsys_state *css;
4355 struct hlist_node *tmp;
4359 /* check name and function validity */
4360 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4361 ss->css_alloc == NULL || ss->css_free == NULL)
4365 * we don't support callbacks in modular subsystems. this check is
4366 * before the ss->module check for consistency; a subsystem that could
4367 * be a module should still have no callbacks even if the user isn't
4368 * compiling it as one.
4370 if (ss->fork || ss->exit)
4374 * an optionally modular subsystem is built-in: we want to do nothing,
4375 * since cgroup_init_subsys will have already taken care of it.
4377 if (ss->module == NULL) {
4378 /* a sanity check */
4379 BUG_ON(subsys[ss->subsys_id] != ss);
4383 /* init base cftset */
4384 cgroup_init_cftsets(ss);
4386 mutex_lock(&cgroup_mutex);
4387 subsys[ss->subsys_id] = ss;
4390 * no ss->css_alloc seems to need anything important in the ss
4391 * struct, so this can happen first (i.e. before the rootnode
4394 css = ss->css_alloc(dummytop);
4396 /* failure case - need to deassign the subsys[] slot. */
4397 subsys[ss->subsys_id] = NULL;
4398 mutex_unlock(&cgroup_mutex);
4399 return PTR_ERR(css);
4402 list_add(&ss->sibling, &rootnode.subsys_list);
4403 ss->root = &rootnode;
4405 /* our new subsystem will be attached to the dummy hierarchy. */
4406 init_cgroup_css(css, ss, dummytop);
4407 /* init_idr must be after init_cgroup_css because it sets css->id. */
4409 ret = cgroup_init_idr(ss, css);
4415 * Now we need to entangle the css into the existing css_sets. unlike
4416 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4417 * will need a new pointer to it; done by iterating the css_set_table.
4418 * furthermore, modifying the existing css_sets will corrupt the hash
4419 * table state, so each changed css_set will need its hash recomputed.
4420 * this is all done under the css_set_lock.
4422 write_lock(&css_set_lock);
4423 hash_for_each_safe(css_set_table, i, tmp, cg, hlist) {
4424 /* skip entries that we already rehashed */
4425 if (cg->subsys[ss->subsys_id])
4427 /* remove existing entry */
4428 hash_del(&cg->hlist);
4430 cg->subsys[ss->subsys_id] = css;
4431 /* recompute hash and restore entry */
4432 key = css_set_hash(cg->subsys);
4433 hash_add(css_set_table, &cg->hlist, key);
4435 write_unlock(&css_set_lock);
4438 ret = online_css(ss, dummytop);
4443 mutex_unlock(&cgroup_mutex);
4447 mutex_unlock(&cgroup_mutex);
4448 /* @ss can't be mounted here as try_module_get() would fail */
4449 cgroup_unload_subsys(ss);
4452 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4455 * cgroup_unload_subsys: unload a modular subsystem
4456 * @ss: the subsystem to unload
4458 * This function should be called in a modular subsystem's exitcall. When this
4459 * function is invoked, the refcount on the subsystem's module will be 0, so
4460 * the subsystem will not be attached to any hierarchy.
4462 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4464 struct cg_cgroup_link *link;
4466 BUG_ON(ss->module == NULL);
4469 * we shouldn't be called if the subsystem is in use, and the use of
4470 * try_module_get in parse_cgroupfs_options should ensure that it
4471 * doesn't start being used while we're killing it off.
4473 BUG_ON(ss->root != &rootnode);
4475 mutex_lock(&cgroup_mutex);
4477 offline_css(ss, dummytop);
4481 idr_destroy(&ss->idr);
4483 /* deassign the subsys_id */
4484 subsys[ss->subsys_id] = NULL;
4486 /* remove subsystem from rootnode's list of subsystems */
4487 list_del_init(&ss->sibling);
4490 * disentangle the css from all css_sets attached to the dummytop. as
4491 * in loading, we need to pay our respects to the hashtable gods.
4493 write_lock(&css_set_lock);
4494 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4495 struct css_set *cg = link->cg;
4498 hash_del(&cg->hlist);
4499 cg->subsys[ss->subsys_id] = NULL;
4500 key = css_set_hash(cg->subsys);
4501 hash_add(css_set_table, &cg->hlist, key);
4503 write_unlock(&css_set_lock);
4506 * remove subsystem's css from the dummytop and free it - need to
4507 * free before marking as null because ss->css_free needs the
4508 * cgrp->subsys pointer to find their state. note that this also
4509 * takes care of freeing the css_id.
4511 ss->css_free(dummytop);
4512 dummytop->subsys[ss->subsys_id] = NULL;
4514 mutex_unlock(&cgroup_mutex);
4516 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4519 * cgroup_init_early - cgroup initialization at system boot
4521 * Initialize cgroups at system boot, and initialize any
4522 * subsystems that request early init.
4524 int __init cgroup_init_early(void)
4527 atomic_set(&init_css_set.refcount, 1);
4528 INIT_LIST_HEAD(&init_css_set.cg_links);
4529 INIT_LIST_HEAD(&init_css_set.tasks);
4530 INIT_HLIST_NODE(&init_css_set.hlist);
4532 init_cgroup_root(&rootnode);
4534 init_task.cgroups = &init_css_set;
4536 init_css_set_link.cg = &init_css_set;
4537 init_css_set_link.cgrp = dummytop;
4538 list_add(&init_css_set_link.cgrp_link_list,
4539 &rootnode.top_cgroup.css_sets);
4540 list_add(&init_css_set_link.cg_link_list,
4541 &init_css_set.cg_links);
4543 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4544 struct cgroup_subsys *ss = subsys[i];
4546 /* at bootup time, we don't worry about modular subsystems */
4547 if (!ss || ss->module)
4551 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4552 BUG_ON(!ss->css_alloc);
4553 BUG_ON(!ss->css_free);
4554 if (ss->subsys_id != i) {
4555 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4556 ss->name, ss->subsys_id);
4561 cgroup_init_subsys(ss);
4567 * cgroup_init - cgroup initialization
4569 * Register cgroup filesystem and /proc file, and initialize
4570 * any subsystems that didn't request early init.
4572 int __init cgroup_init(void)
4578 err = bdi_init(&cgroup_backing_dev_info);
4582 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4583 struct cgroup_subsys *ss = subsys[i];
4585 /* at bootup time, we don't worry about modular subsystems */
4586 if (!ss || ss->module)
4588 if (!ss->early_init)
4589 cgroup_init_subsys(ss);
4591 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4594 /* Add init_css_set to the hash table */
4595 key = css_set_hash(init_css_set.subsys);
4596 hash_add(css_set_table, &init_css_set.hlist, key);
4597 BUG_ON(!init_root_id(&rootnode));
4599 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4605 err = register_filesystem(&cgroup_fs_type);
4607 kobject_put(cgroup_kobj);
4611 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4615 bdi_destroy(&cgroup_backing_dev_info);
4621 * proc_cgroup_show()
4622 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4623 * - Used for /proc/<pid>/cgroup.
4624 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4625 * doesn't really matter if tsk->cgroup changes after we read it,
4626 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4627 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4628 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4629 * cgroup to top_cgroup.
4632 /* TODO: Use a proper seq_file iterator */
4633 static int proc_cgroup_show(struct seq_file *m, void *v)
4636 struct task_struct *tsk;
4639 struct cgroupfs_root *root;
4642 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4648 tsk = get_pid_task(pid, PIDTYPE_PID);
4654 mutex_lock(&cgroup_mutex);
4656 for_each_active_root(root) {
4657 struct cgroup_subsys *ss;
4658 struct cgroup *cgrp;
4661 seq_printf(m, "%d:", root->hierarchy_id);
4662 for_each_subsys(root, ss)
4663 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4664 if (strlen(root->name))
4665 seq_printf(m, "%sname=%s", count ? "," : "",
4668 cgrp = task_cgroup_from_root(tsk, root);
4669 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4677 mutex_unlock(&cgroup_mutex);
4678 put_task_struct(tsk);
4685 static int cgroup_open(struct inode *inode, struct file *file)
4687 struct pid *pid = PROC_I(inode)->pid;
4688 return single_open(file, proc_cgroup_show, pid);
4691 const struct file_operations proc_cgroup_operations = {
4692 .open = cgroup_open,
4694 .llseek = seq_lseek,
4695 .release = single_release,
4698 /* Display information about each subsystem and each hierarchy */
4699 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4703 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4705 * ideally we don't want subsystems moving around while we do this.
4706 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4707 * subsys/hierarchy state.
4709 mutex_lock(&cgroup_mutex);
4710 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4711 struct cgroup_subsys *ss = subsys[i];
4714 seq_printf(m, "%s\t%d\t%d\t%d\n",
4715 ss->name, ss->root->hierarchy_id,
4716 ss->root->number_of_cgroups, !ss->disabled);
4718 mutex_unlock(&cgroup_mutex);
4722 static int cgroupstats_open(struct inode *inode, struct file *file)
4724 return single_open(file, proc_cgroupstats_show, NULL);
4727 static const struct file_operations proc_cgroupstats_operations = {
4728 .open = cgroupstats_open,
4730 .llseek = seq_lseek,
4731 .release = single_release,
4735 * cgroup_fork - attach newly forked task to its parents cgroup.
4736 * @child: pointer to task_struct of forking parent process.
4738 * Description: A task inherits its parent's cgroup at fork().
4740 * A pointer to the shared css_set was automatically copied in
4741 * fork.c by dup_task_struct(). However, we ignore that copy, since
4742 * it was not made under the protection of RCU or cgroup_mutex, so
4743 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4744 * have already changed current->cgroups, allowing the previously
4745 * referenced cgroup group to be removed and freed.
4747 * At the point that cgroup_fork() is called, 'current' is the parent
4748 * task, and the passed argument 'child' points to the child task.
4750 void cgroup_fork(struct task_struct *child)
4753 child->cgroups = current->cgroups;
4754 get_css_set(child->cgroups);
4755 task_unlock(current);
4756 INIT_LIST_HEAD(&child->cg_list);
4760 * cgroup_post_fork - called on a new task after adding it to the task list
4761 * @child: the task in question
4763 * Adds the task to the list running through its css_set if necessary and
4764 * call the subsystem fork() callbacks. Has to be after the task is
4765 * visible on the task list in case we race with the first call to
4766 * cgroup_iter_start() - to guarantee that the new task ends up on its
4769 void cgroup_post_fork(struct task_struct *child)
4774 * use_task_css_set_links is set to 1 before we walk the tasklist
4775 * under the tasklist_lock and we read it here after we added the child
4776 * to the tasklist under the tasklist_lock as well. If the child wasn't
4777 * yet in the tasklist when we walked through it from
4778 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4779 * should be visible now due to the paired locking and barriers implied
4780 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4781 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4784 if (use_task_css_set_links) {
4785 write_lock(&css_set_lock);
4787 if (list_empty(&child->cg_list))
4788 list_add(&child->cg_list, &child->cgroups->tasks);
4790 write_unlock(&css_set_lock);
4794 * Call ss->fork(). This must happen after @child is linked on
4795 * css_set; otherwise, @child might change state between ->fork()
4796 * and addition to css_set.
4798 if (need_forkexit_callback) {
4800 * fork/exit callbacks are supported only for builtin
4801 * subsystems, and the builtin section of the subsys
4802 * array is immutable, so we don't need to lock the
4803 * subsys array here. On the other hand, modular section
4804 * of the array can be freed at module unload, so we
4807 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4808 struct cgroup_subsys *ss = subsys[i];
4817 * cgroup_exit - detach cgroup from exiting task
4818 * @tsk: pointer to task_struct of exiting process
4819 * @run_callback: run exit callbacks?
4821 * Description: Detach cgroup from @tsk and release it.
4823 * Note that cgroups marked notify_on_release force every task in
4824 * them to take the global cgroup_mutex mutex when exiting.
4825 * This could impact scaling on very large systems. Be reluctant to
4826 * use notify_on_release cgroups where very high task exit scaling
4827 * is required on large systems.
4829 * the_top_cgroup_hack:
4831 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4833 * We call cgroup_exit() while the task is still competent to
4834 * handle notify_on_release(), then leave the task attached to the
4835 * root cgroup in each hierarchy for the remainder of its exit.
4837 * To do this properly, we would increment the reference count on
4838 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4839 * code we would add a second cgroup function call, to drop that
4840 * reference. This would just create an unnecessary hot spot on
4841 * the top_cgroup reference count, to no avail.
4843 * Normally, holding a reference to a cgroup without bumping its
4844 * count is unsafe. The cgroup could go away, or someone could
4845 * attach us to a different cgroup, decrementing the count on
4846 * the first cgroup that we never incremented. But in this case,
4847 * top_cgroup isn't going away, and either task has PF_EXITING set,
4848 * which wards off any cgroup_attach_task() attempts, or task is a failed
4849 * fork, never visible to cgroup_attach_task.
4851 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4857 * Unlink from the css_set task list if necessary.
4858 * Optimistically check cg_list before taking
4861 if (!list_empty(&tsk->cg_list)) {
4862 write_lock(&css_set_lock);
4863 if (!list_empty(&tsk->cg_list))
4864 list_del_init(&tsk->cg_list);
4865 write_unlock(&css_set_lock);
4868 /* Reassign the task to the init_css_set. */
4871 tsk->cgroups = &init_css_set;
4873 if (run_callbacks && need_forkexit_callback) {
4875 * fork/exit callbacks are supported only for builtin
4876 * subsystems, see cgroup_post_fork() for details.
4878 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4879 struct cgroup_subsys *ss = subsys[i];
4882 struct cgroup *old_cgrp =
4883 rcu_dereference_raw(cg->subsys[i])->cgroup;
4884 struct cgroup *cgrp = task_cgroup(tsk, i);
4885 ss->exit(cgrp, old_cgrp, tsk);
4891 put_css_set_taskexit(cg);
4894 static void check_for_release(struct cgroup *cgrp)
4896 /* All of these checks rely on RCU to keep the cgroup
4897 * structure alive */
4898 if (cgroup_is_releasable(cgrp) &&
4899 !atomic_read(&cgrp->count) && list_empty(&cgrp->children)) {
4901 * Control Group is currently removeable. If it's not
4902 * already queued for a userspace notification, queue
4905 int need_schedule_work = 0;
4907 raw_spin_lock(&release_list_lock);
4908 if (!cgroup_is_removed(cgrp) &&
4909 list_empty(&cgrp->release_list)) {
4910 list_add(&cgrp->release_list, &release_list);
4911 need_schedule_work = 1;
4913 raw_spin_unlock(&release_list_lock);
4914 if (need_schedule_work)
4915 schedule_work(&release_agent_work);
4919 /* Caller must verify that the css is not for root cgroup */
4920 bool __css_tryget(struct cgroup_subsys_state *css)
4925 v = css_refcnt(css);
4926 t = atomic_cmpxchg(&css->refcnt, v, v + 1);
4934 EXPORT_SYMBOL_GPL(__css_tryget);
4936 /* Caller must verify that the css is not for root cgroup */
4937 void __css_put(struct cgroup_subsys_state *css)
4941 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
4943 schedule_work(&css->dput_work);
4945 EXPORT_SYMBOL_GPL(__css_put);
4948 * Notify userspace when a cgroup is released, by running the
4949 * configured release agent with the name of the cgroup (path
4950 * relative to the root of cgroup file system) as the argument.
4952 * Most likely, this user command will try to rmdir this cgroup.
4954 * This races with the possibility that some other task will be
4955 * attached to this cgroup before it is removed, or that some other
4956 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4957 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4958 * unused, and this cgroup will be reprieved from its death sentence,
4959 * to continue to serve a useful existence. Next time it's released,
4960 * we will get notified again, if it still has 'notify_on_release' set.
4962 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4963 * means only wait until the task is successfully execve()'d. The
4964 * separate release agent task is forked by call_usermodehelper(),
4965 * then control in this thread returns here, without waiting for the
4966 * release agent task. We don't bother to wait because the caller of
4967 * this routine has no use for the exit status of the release agent
4968 * task, so no sense holding our caller up for that.
4970 static void cgroup_release_agent(struct work_struct *work)
4972 BUG_ON(work != &release_agent_work);
4973 mutex_lock(&cgroup_mutex);
4974 raw_spin_lock(&release_list_lock);
4975 while (!list_empty(&release_list)) {
4976 char *argv[3], *envp[3];
4978 char *pathbuf = NULL, *agentbuf = NULL;
4979 struct cgroup *cgrp = list_entry(release_list.next,
4982 list_del_init(&cgrp->release_list);
4983 raw_spin_unlock(&release_list_lock);
4984 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4987 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4989 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4994 argv[i++] = agentbuf;
4995 argv[i++] = pathbuf;
4999 /* minimal command environment */
5000 envp[i++] = "HOME=/";
5001 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5004 /* Drop the lock while we invoke the usermode helper,
5005 * since the exec could involve hitting disk and hence
5006 * be a slow process */
5007 mutex_unlock(&cgroup_mutex);
5008 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5009 mutex_lock(&cgroup_mutex);
5013 raw_spin_lock(&release_list_lock);
5015 raw_spin_unlock(&release_list_lock);
5016 mutex_unlock(&cgroup_mutex);
5019 static int __init cgroup_disable(char *str)
5024 while ((token = strsep(&str, ",")) != NULL) {
5027 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5028 struct cgroup_subsys *ss = subsys[i];
5031 * cgroup_disable, being at boot time, can't
5032 * know about module subsystems, so we don't
5035 if (!ss || ss->module)
5038 if (!strcmp(token, ss->name)) {
5040 printk(KERN_INFO "Disabling %s control group"
5041 " subsystem\n", ss->name);
5048 __setup("cgroup_disable=", cgroup_disable);
5051 * Functons for CSS ID.
5055 *To get ID other than 0, this should be called when !cgroup_is_removed().
5057 unsigned short css_id(struct cgroup_subsys_state *css)
5059 struct css_id *cssid;
5062 * This css_id() can return correct value when somone has refcnt
5063 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5064 * it's unchanged until freed.
5066 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5072 EXPORT_SYMBOL_GPL(css_id);
5074 unsigned short css_depth(struct cgroup_subsys_state *css)
5076 struct css_id *cssid;
5078 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5081 return cssid->depth;
5084 EXPORT_SYMBOL_GPL(css_depth);
5087 * css_is_ancestor - test "root" css is an ancestor of "child"
5088 * @child: the css to be tested.
5089 * @root: the css supporsed to be an ancestor of the child.
5091 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5092 * this function reads css->id, the caller must hold rcu_read_lock().
5093 * But, considering usual usage, the csses should be valid objects after test.
5094 * Assuming that the caller will do some action to the child if this returns
5095 * returns true, the caller must take "child";s reference count.
5096 * If "child" is valid object and this returns true, "root" is valid, too.
5099 bool css_is_ancestor(struct cgroup_subsys_state *child,
5100 const struct cgroup_subsys_state *root)
5102 struct css_id *child_id;
5103 struct css_id *root_id;
5105 child_id = rcu_dereference(child->id);
5108 root_id = rcu_dereference(root->id);
5111 if (child_id->depth < root_id->depth)
5113 if (child_id->stack[root_id->depth] != root_id->id)
5118 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5120 struct css_id *id = css->id;
5121 /* When this is called before css_id initialization, id can be NULL */
5125 BUG_ON(!ss->use_id);
5127 rcu_assign_pointer(id->css, NULL);
5128 rcu_assign_pointer(css->id, NULL);
5129 spin_lock(&ss->id_lock);
5130 idr_remove(&ss->idr, id->id);
5131 spin_unlock(&ss->id_lock);
5132 kfree_rcu(id, rcu_head);
5134 EXPORT_SYMBOL_GPL(free_css_id);
5137 * This is called by init or create(). Then, calls to this function are
5138 * always serialized (By cgroup_mutex() at create()).
5141 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5143 struct css_id *newid;
5146 BUG_ON(!ss->use_id);
5148 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5149 newid = kzalloc(size, GFP_KERNEL);
5151 return ERR_PTR(-ENOMEM);
5153 idr_preload(GFP_KERNEL);
5154 spin_lock(&ss->id_lock);
5155 /* Don't use 0. allocates an ID of 1-65535 */
5156 ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5157 spin_unlock(&ss->id_lock);
5160 /* Returns error when there are no free spaces for new ID.*/
5165 newid->depth = depth;
5169 return ERR_PTR(ret);
5173 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5174 struct cgroup_subsys_state *rootcss)
5176 struct css_id *newid;
5178 spin_lock_init(&ss->id_lock);
5181 newid = get_new_cssid(ss, 0);
5183 return PTR_ERR(newid);
5185 newid->stack[0] = newid->id;
5186 newid->css = rootcss;
5187 rootcss->id = newid;
5191 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5192 struct cgroup *child)
5194 int subsys_id, i, depth = 0;
5195 struct cgroup_subsys_state *parent_css, *child_css;
5196 struct css_id *child_id, *parent_id;
5198 subsys_id = ss->subsys_id;
5199 parent_css = parent->subsys[subsys_id];
5200 child_css = child->subsys[subsys_id];
5201 parent_id = parent_css->id;
5202 depth = parent_id->depth + 1;
5204 child_id = get_new_cssid(ss, depth);
5205 if (IS_ERR(child_id))
5206 return PTR_ERR(child_id);
5208 for (i = 0; i < depth; i++)
5209 child_id->stack[i] = parent_id->stack[i];
5210 child_id->stack[depth] = child_id->id;
5212 * child_id->css pointer will be set after this cgroup is available
5213 * see cgroup_populate_dir()
5215 rcu_assign_pointer(child_css->id, child_id);
5221 * css_lookup - lookup css by id
5222 * @ss: cgroup subsys to be looked into.
5225 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5226 * NULL if not. Should be called under rcu_read_lock()
5228 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5230 struct css_id *cssid = NULL;
5232 BUG_ON(!ss->use_id);
5233 cssid = idr_find(&ss->idr, id);
5235 if (unlikely(!cssid))
5238 return rcu_dereference(cssid->css);
5240 EXPORT_SYMBOL_GPL(css_lookup);
5243 * css_get_next - lookup next cgroup under specified hierarchy.
5244 * @ss: pointer to subsystem
5245 * @id: current position of iteration.
5246 * @root: pointer to css. search tree under this.
5247 * @foundid: position of found object.
5249 * Search next css under the specified hierarchy of rootid. Calling under
5250 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5252 struct cgroup_subsys_state *
5253 css_get_next(struct cgroup_subsys *ss, int id,
5254 struct cgroup_subsys_state *root, int *foundid)
5256 struct cgroup_subsys_state *ret = NULL;
5259 int rootid = css_id(root);
5260 int depth = css_depth(root);
5265 BUG_ON(!ss->use_id);
5266 WARN_ON_ONCE(!rcu_read_lock_held());
5268 /* fill start point for scan */
5272 * scan next entry from bitmap(tree), tmpid is updated after
5275 tmp = idr_get_next(&ss->idr, &tmpid);
5278 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5279 ret = rcu_dereference(tmp->css);
5285 /* continue to scan from next id */
5292 * get corresponding css from file open on cgroupfs directory
5294 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5296 struct cgroup *cgrp;
5297 struct inode *inode;
5298 struct cgroup_subsys_state *css;
5300 inode = file_inode(f);
5301 /* check in cgroup filesystem dir */
5302 if (inode->i_op != &cgroup_dir_inode_operations)
5303 return ERR_PTR(-EBADF);
5305 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5306 return ERR_PTR(-EINVAL);
5309 cgrp = __d_cgrp(f->f_dentry);
5310 css = cgrp->subsys[id];
5311 return css ? css : ERR_PTR(-ENOENT);
5314 #ifdef CONFIG_CGROUP_DEBUG
5315 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5317 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5320 return ERR_PTR(-ENOMEM);
5325 static void debug_css_free(struct cgroup *cont)
5327 kfree(cont->subsys[debug_subsys_id]);
5330 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5332 return atomic_read(&cont->count);
5335 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5337 return cgroup_task_count(cont);
5340 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5342 return (u64)(unsigned long)current->cgroups;
5345 static u64 current_css_set_refcount_read(struct cgroup *cont,
5351 count = atomic_read(¤t->cgroups->refcount);
5356 static int current_css_set_cg_links_read(struct cgroup *cont,
5358 struct seq_file *seq)
5360 struct cg_cgroup_link *link;
5363 read_lock(&css_set_lock);
5365 cg = rcu_dereference(current->cgroups);
5366 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5367 struct cgroup *c = link->cgrp;
5371 name = c->dentry->d_name.name;
5374 seq_printf(seq, "Root %d group %s\n",
5375 c->root->hierarchy_id, name);
5378 read_unlock(&css_set_lock);
5382 #define MAX_TASKS_SHOWN_PER_CSS 25
5383 static int cgroup_css_links_read(struct cgroup *cont,
5385 struct seq_file *seq)
5387 struct cg_cgroup_link *link;
5389 read_lock(&css_set_lock);
5390 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5391 struct css_set *cg = link->cg;
5392 struct task_struct *task;
5394 seq_printf(seq, "css_set %p\n", cg);
5395 list_for_each_entry(task, &cg->tasks, cg_list) {
5396 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5397 seq_puts(seq, " ...\n");
5400 seq_printf(seq, " task %d\n",
5401 task_pid_vnr(task));
5405 read_unlock(&css_set_lock);
5409 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5411 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5414 static struct cftype debug_files[] = {
5416 .name = "cgroup_refcount",
5417 .read_u64 = cgroup_refcount_read,
5420 .name = "taskcount",
5421 .read_u64 = debug_taskcount_read,
5425 .name = "current_css_set",
5426 .read_u64 = current_css_set_read,
5430 .name = "current_css_set_refcount",
5431 .read_u64 = current_css_set_refcount_read,
5435 .name = "current_css_set_cg_links",
5436 .read_seq_string = current_css_set_cg_links_read,
5440 .name = "cgroup_css_links",
5441 .read_seq_string = cgroup_css_links_read,
5445 .name = "releasable",
5446 .read_u64 = releasable_read,
5452 struct cgroup_subsys debug_subsys = {
5454 .css_alloc = debug_css_alloc,
5455 .css_free = debug_css_free,
5456 .subsys_id = debug_subsys_id,
5457 .base_cftypes = debug_files,
5459 #endif /* CONFIG_CGROUP_DEBUG */