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 * cgroup destruction makes heavy use of work items and there can be a lot
96 * of concurrent destructions. Use a separate workqueue so that cgroup
97 * destruction work items don't end up filling up max_active of system_wq
98 * which may lead to deadlock.
100 static struct workqueue_struct *cgroup_destroy_wq;
103 * Generate an array of cgroup subsystem pointers. At boot time, this is
104 * populated with the built in subsystems, and modular subsystems are
105 * registered after that. The mutable section of this array is protected by
108 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
109 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
110 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
111 #include <linux/cgroup_subsys.h>
115 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
116 * subsystems that are otherwise unattached - it never has more than a
117 * single cgroup, and all tasks are part of that cgroup.
119 static struct cgroupfs_root rootnode;
122 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
125 struct list_head node;
126 struct dentry *dentry;
130 struct simple_xattrs xattrs;
134 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
135 * cgroup_subsys->use_id != 0.
137 #define CSS_ID_MAX (65535)
140 * The css to which this ID points. This pointer is set to valid value
141 * after cgroup is populated. If cgroup is removed, this will be NULL.
142 * This pointer is expected to be RCU-safe because destroy()
143 * is called after synchronize_rcu(). But for safe use, css_tryget()
144 * should be used for avoiding race.
146 struct cgroup_subsys_state __rcu *css;
152 * Depth in hierarchy which this ID belongs to.
154 unsigned short depth;
156 * ID is freed by RCU. (and lookup routine is RCU safe.)
158 struct rcu_head rcu_head;
160 * Hierarchy of CSS ID belongs to.
162 unsigned short stack[0]; /* Array of Length (depth+1) */
166 * cgroup_event represents events which userspace want to receive.
168 struct cgroup_event {
170 * Cgroup which the event belongs to.
174 * Control file which the event associated.
178 * eventfd to signal userspace about the event.
180 struct eventfd_ctx *eventfd;
182 * Each of these stored in a list by the cgroup.
184 struct list_head list;
186 * All fields below needed to unregister event when
187 * userspace closes eventfd.
190 wait_queue_head_t *wqh;
192 struct work_struct remove;
195 /* The list of hierarchy roots */
197 static LIST_HEAD(roots);
198 static int root_count;
200 static DEFINE_IDA(hierarchy_ida);
201 static int next_hierarchy_id;
202 static DEFINE_SPINLOCK(hierarchy_id_lock);
204 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
205 #define dummytop (&rootnode.top_cgroup)
207 static struct cgroup_name root_cgroup_name = { .name = "/" };
209 /* This flag indicates whether tasks in the fork and exit paths should
210 * check for fork/exit handlers to call. This avoids us having to do
211 * extra work in the fork/exit path if none of the subsystems need to
214 static int need_forkexit_callback __read_mostly;
216 static int cgroup_destroy_locked(struct cgroup *cgrp);
217 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
218 struct cftype cfts[], bool is_add);
220 static int css_unbias_refcnt(int refcnt)
222 return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
225 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
226 static int css_refcnt(struct cgroup_subsys_state *css)
228 int v = atomic_read(&css->refcnt);
230 return css_unbias_refcnt(v);
233 /* convenient tests for these bits */
234 inline int cgroup_is_removed(const struct cgroup *cgrp)
236 return test_bit(CGRP_REMOVED, &cgrp->flags);
240 * cgroup_is_descendant - test ancestry
241 * @cgrp: the cgroup to be tested
242 * @ancestor: possible ancestor of @cgrp
244 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
245 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
246 * and @ancestor are accessible.
248 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
251 if (cgrp == ancestor)
257 EXPORT_SYMBOL_GPL(cgroup_is_descendant);
259 static int cgroup_is_releasable(const struct cgroup *cgrp)
262 (1 << CGRP_RELEASABLE) |
263 (1 << CGRP_NOTIFY_ON_RELEASE);
264 return (cgrp->flags & bits) == bits;
267 static int notify_on_release(const struct cgroup *cgrp)
269 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
273 * for_each_subsys() allows you to iterate on each subsystem attached to
274 * an active hierarchy
276 #define for_each_subsys(_root, _ss) \
277 list_for_each_entry(_ss, &_root->subsys_list, sibling)
279 /* for_each_active_root() allows you to iterate across the active hierarchies */
280 #define for_each_active_root(_root) \
281 list_for_each_entry(_root, &roots, root_list)
283 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
285 return dentry->d_fsdata;
288 static inline struct cfent *__d_cfe(struct dentry *dentry)
290 return dentry->d_fsdata;
293 static inline struct cftype *__d_cft(struct dentry *dentry)
295 return __d_cfe(dentry)->type;
299 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
300 * @cgrp: the cgroup to be checked for liveness
302 * On success, returns true; the mutex should be later unlocked. On
303 * failure returns false with no lock held.
305 static bool cgroup_lock_live_group(struct cgroup *cgrp)
307 mutex_lock(&cgroup_mutex);
308 if (cgroup_is_removed(cgrp)) {
309 mutex_unlock(&cgroup_mutex);
315 /* the list of cgroups eligible for automatic release. Protected by
316 * release_list_lock */
317 static LIST_HEAD(release_list);
318 static DEFINE_RAW_SPINLOCK(release_list_lock);
319 static void cgroup_release_agent(struct work_struct *work);
320 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
321 static void check_for_release(struct cgroup *cgrp);
323 /* Link structure for associating css_set objects with cgroups */
324 struct cg_cgroup_link {
326 * List running through cg_cgroup_links associated with a
327 * cgroup, anchored on cgroup->css_sets
329 struct list_head cgrp_link_list;
332 * List running through cg_cgroup_links pointing at a
333 * single css_set object, anchored on css_set->cg_links
335 struct list_head cg_link_list;
339 /* The default css_set - used by init and its children prior to any
340 * hierarchies being mounted. It contains a pointer to the root state
341 * for each subsystem. Also used to anchor the list of css_sets. Not
342 * reference-counted, to improve performance when child cgroups
343 * haven't been created.
346 static struct css_set init_css_set;
347 static struct cg_cgroup_link init_css_set_link;
349 static int cgroup_init_idr(struct cgroup_subsys *ss,
350 struct cgroup_subsys_state *css);
352 /* css_set_lock protects the list of css_set objects, and the
353 * chain of tasks off each css_set. Nests outside task->alloc_lock
354 * due to cgroup_iter_start() */
355 static DEFINE_RWLOCK(css_set_lock);
356 static int css_set_count;
359 * hash table for cgroup groups. This improves the performance to find
360 * an existing css_set. This hash doesn't (currently) take into
361 * account cgroups in empty hierarchies.
363 #define CSS_SET_HASH_BITS 7
364 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
366 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
369 unsigned long key = 0UL;
371 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
372 key += (unsigned long)css[i];
373 key = (key >> 16) ^ key;
378 /* We don't maintain the lists running through each css_set to its
379 * task until after the first call to cgroup_iter_start(). This
380 * reduces the fork()/exit() overhead for people who have cgroups
381 * compiled into their kernel but not actually in use */
382 static int use_task_css_set_links __read_mostly;
384 static void __put_css_set(struct css_set *cg, int taskexit)
386 struct cg_cgroup_link *link;
387 struct cg_cgroup_link *saved_link;
389 * Ensure that the refcount doesn't hit zero while any readers
390 * can see it. Similar to atomic_dec_and_lock(), but for an
393 if (atomic_add_unless(&cg->refcount, -1, 1))
395 write_lock(&css_set_lock);
396 if (!atomic_dec_and_test(&cg->refcount)) {
397 write_unlock(&css_set_lock);
401 /* This css_set is dead. unlink it and release cgroup refcounts */
402 hash_del(&cg->hlist);
405 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
407 struct cgroup *cgrp = link->cgrp;
408 list_del(&link->cg_link_list);
409 list_del(&link->cgrp_link_list);
412 * We may not be holding cgroup_mutex, and if cgrp->count is
413 * dropped to 0 the cgroup can be destroyed at any time, hence
414 * rcu_read_lock is used to keep it alive.
417 if (atomic_dec_and_test(&cgrp->count) &&
418 notify_on_release(cgrp)) {
420 set_bit(CGRP_RELEASABLE, &cgrp->flags);
421 check_for_release(cgrp);
428 write_unlock(&css_set_lock);
429 kfree_rcu(cg, rcu_head);
433 * refcounted get/put for css_set objects
435 static inline void get_css_set(struct css_set *cg)
437 atomic_inc(&cg->refcount);
440 static inline void put_css_set(struct css_set *cg)
442 __put_css_set(cg, 0);
445 static inline void put_css_set_taskexit(struct css_set *cg)
447 __put_css_set(cg, 1);
451 * compare_css_sets - helper function for find_existing_css_set().
452 * @cg: candidate css_set being tested
453 * @old_cg: existing css_set for a task
454 * @new_cgrp: cgroup that's being entered by the task
455 * @template: desired set of css pointers in css_set (pre-calculated)
457 * Returns true if "cg" matches "old_cg" except for the hierarchy
458 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
460 static bool compare_css_sets(struct css_set *cg,
461 struct css_set *old_cg,
462 struct cgroup *new_cgrp,
463 struct cgroup_subsys_state *template[])
465 struct list_head *l1, *l2;
467 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
468 /* Not all subsystems matched */
473 * Compare cgroup pointers in order to distinguish between
474 * different cgroups in heirarchies with no subsystems. We
475 * could get by with just this check alone (and skip the
476 * memcmp above) but on most setups the memcmp check will
477 * avoid the need for this more expensive check on almost all
482 l2 = &old_cg->cg_links;
484 struct cg_cgroup_link *cgl1, *cgl2;
485 struct cgroup *cg1, *cg2;
489 /* See if we reached the end - both lists are equal length. */
490 if (l1 == &cg->cg_links) {
491 BUG_ON(l2 != &old_cg->cg_links);
494 BUG_ON(l2 == &old_cg->cg_links);
496 /* Locate the cgroups associated with these links. */
497 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
498 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
501 /* Hierarchies should be linked in the same order. */
502 BUG_ON(cg1->root != cg2->root);
505 * If this hierarchy is the hierarchy of the cgroup
506 * that's changing, then we need to check that this
507 * css_set points to the new cgroup; if it's any other
508 * hierarchy, then this css_set should point to the
509 * same cgroup as the old css_set.
511 if (cg1->root == new_cgrp->root) {
523 * find_existing_css_set() is a helper for
524 * find_css_set(), and checks to see whether an existing
525 * css_set is suitable.
527 * oldcg: the cgroup group that we're using before the cgroup
530 * cgrp: the cgroup that we're moving into
532 * template: location in which to build the desired set of subsystem
533 * state objects for the new cgroup group
535 static struct css_set *find_existing_css_set(
536 struct css_set *oldcg,
538 struct cgroup_subsys_state *template[])
541 struct cgroupfs_root *root = cgrp->root;
546 * Build the set of subsystem state objects that we want to see in the
547 * new css_set. while subsystems can change globally, the entries here
548 * won't change, so no need for locking.
550 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
551 if (root->subsys_mask & (1UL << i)) {
552 /* Subsystem is in this hierarchy. So we want
553 * the subsystem state from the new
555 template[i] = cgrp->subsys[i];
557 /* Subsystem is not in this hierarchy, so we
558 * don't want to change the subsystem state */
559 template[i] = oldcg->subsys[i];
563 key = css_set_hash(template);
564 hash_for_each_possible(css_set_table, cg, hlist, key) {
565 if (!compare_css_sets(cg, oldcg, cgrp, template))
568 /* This css_set matches what we need */
572 /* No existing cgroup group matched */
576 static void free_cg_links(struct list_head *tmp)
578 struct cg_cgroup_link *link;
579 struct cg_cgroup_link *saved_link;
581 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
582 list_del(&link->cgrp_link_list);
588 * allocate_cg_links() allocates "count" cg_cgroup_link structures
589 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
590 * success or a negative error
592 static int allocate_cg_links(int count, struct list_head *tmp)
594 struct cg_cgroup_link *link;
597 for (i = 0; i < count; i++) {
598 link = kmalloc(sizeof(*link), GFP_KERNEL);
603 list_add(&link->cgrp_link_list, tmp);
609 * link_css_set - a helper function to link a css_set to a cgroup
610 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
611 * @cg: the css_set to be linked
612 * @cgrp: the destination cgroup
614 static void link_css_set(struct list_head *tmp_cg_links,
615 struct css_set *cg, struct cgroup *cgrp)
617 struct cg_cgroup_link *link;
619 BUG_ON(list_empty(tmp_cg_links));
620 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
624 atomic_inc(&cgrp->count);
625 list_move(&link->cgrp_link_list, &cgrp->css_sets);
627 * Always add links to the tail of the list so that the list
628 * is sorted by order of hierarchy creation
630 list_add_tail(&link->cg_link_list, &cg->cg_links);
634 * find_css_set() takes an existing cgroup group and a
635 * cgroup object, and returns a css_set object that's
636 * equivalent to the old group, but with the given cgroup
637 * substituted into the appropriate hierarchy. Must be called with
640 static struct css_set *find_css_set(
641 struct css_set *oldcg, struct cgroup *cgrp)
644 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
646 struct list_head tmp_cg_links;
648 struct cg_cgroup_link *link;
651 /* First see if we already have a cgroup group that matches
653 read_lock(&css_set_lock);
654 res = find_existing_css_set(oldcg, cgrp, template);
657 read_unlock(&css_set_lock);
662 res = kmalloc(sizeof(*res), GFP_KERNEL);
666 /* Allocate all the cg_cgroup_link objects that we'll need */
667 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
672 atomic_set(&res->refcount, 1);
673 INIT_LIST_HEAD(&res->cg_links);
674 INIT_LIST_HEAD(&res->tasks);
675 INIT_HLIST_NODE(&res->hlist);
677 /* Copy the set of subsystem state objects generated in
678 * find_existing_css_set() */
679 memcpy(res->subsys, template, sizeof(res->subsys));
681 write_lock(&css_set_lock);
682 /* Add reference counts and links from the new css_set. */
683 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
684 struct cgroup *c = link->cgrp;
685 if (c->root == cgrp->root)
687 link_css_set(&tmp_cg_links, res, c);
690 BUG_ON(!list_empty(&tmp_cg_links));
694 /* Add this cgroup group to the hash table */
695 key = css_set_hash(res->subsys);
696 hash_add(css_set_table, &res->hlist, key);
698 write_unlock(&css_set_lock);
704 * Return the cgroup for "task" from the given hierarchy. Must be
705 * called with cgroup_mutex held.
707 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
708 struct cgroupfs_root *root)
711 struct cgroup *res = NULL;
713 BUG_ON(!mutex_is_locked(&cgroup_mutex));
714 read_lock(&css_set_lock);
716 * No need to lock the task - since we hold cgroup_mutex the
717 * task can't change groups, so the only thing that can happen
718 * is that it exits and its css is set back to init_css_set.
721 if (css == &init_css_set) {
722 res = &root->top_cgroup;
724 struct cg_cgroup_link *link;
725 list_for_each_entry(link, &css->cg_links, cg_link_list) {
726 struct cgroup *c = link->cgrp;
727 if (c->root == root) {
733 read_unlock(&css_set_lock);
739 * There is one global cgroup mutex. We also require taking
740 * task_lock() when dereferencing a task's cgroup subsys pointers.
741 * See "The task_lock() exception", at the end of this comment.
743 * A task must hold cgroup_mutex to modify cgroups.
745 * Any task can increment and decrement the count field without lock.
746 * So in general, code holding cgroup_mutex can't rely on the count
747 * field not changing. However, if the count goes to zero, then only
748 * cgroup_attach_task() can increment it again. Because a count of zero
749 * means that no tasks are currently attached, therefore there is no
750 * way a task attached to that cgroup can fork (the other way to
751 * increment the count). So code holding cgroup_mutex can safely
752 * assume that if the count is zero, it will stay zero. Similarly, if
753 * a task holds cgroup_mutex on a cgroup with zero count, it
754 * knows that the cgroup won't be removed, as cgroup_rmdir()
757 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
758 * (usually) take cgroup_mutex. These are the two most performance
759 * critical pieces of code here. The exception occurs on cgroup_exit(),
760 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
761 * is taken, and if the cgroup count is zero, a usermode call made
762 * to the release agent with the name of the cgroup (path relative to
763 * the root of cgroup file system) as the argument.
765 * A cgroup can only be deleted if both its 'count' of using tasks
766 * is zero, and its list of 'children' cgroups is empty. Since all
767 * tasks in the system use _some_ cgroup, and since there is always at
768 * least one task in the system (init, pid == 1), therefore, top_cgroup
769 * always has either children cgroups and/or using tasks. So we don't
770 * need a special hack to ensure that top_cgroup cannot be deleted.
772 * The task_lock() exception
774 * The need for this exception arises from the action of
775 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
776 * another. It does so using cgroup_mutex, however there are
777 * several performance critical places that need to reference
778 * task->cgroup without the expense of grabbing a system global
779 * mutex. Therefore except as noted below, when dereferencing or, as
780 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
781 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
782 * the task_struct routinely used for such matters.
784 * P.S. One more locking exception. RCU is used to guard the
785 * update of a tasks cgroup pointer by cgroup_attach_task()
789 * A couple of forward declarations required, due to cyclic reference loop:
790 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
791 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
795 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
796 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
797 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
798 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
799 unsigned long subsys_mask);
800 static const struct inode_operations cgroup_dir_inode_operations;
801 static const struct file_operations proc_cgroupstats_operations;
803 static struct backing_dev_info cgroup_backing_dev_info = {
805 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
808 static int alloc_css_id(struct cgroup_subsys *ss,
809 struct cgroup *parent, struct cgroup *child);
811 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
813 struct inode *inode = new_inode(sb);
816 inode->i_ino = get_next_ino();
817 inode->i_mode = mode;
818 inode->i_uid = current_fsuid();
819 inode->i_gid = current_fsgid();
820 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
821 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
826 static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
828 struct cgroup_name *name;
830 name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL);
833 strcpy(name->name, dentry->d_name.name);
837 static void cgroup_free_fn(struct work_struct *work)
839 struct cgroup *cgrp = container_of(work, struct cgroup, free_work);
840 struct cgroup_subsys *ss;
842 mutex_lock(&cgroup_mutex);
844 * Release the subsystem state objects.
846 for_each_subsys(cgrp->root, ss)
849 cgrp->root->number_of_cgroups--;
850 mutex_unlock(&cgroup_mutex);
853 * We get a ref to the parent's dentry, and put the ref when
854 * this cgroup is being freed, so it's guaranteed that the
855 * parent won't be destroyed before its children.
857 dput(cgrp->parent->dentry);
859 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
862 * Drop the active superblock reference that we took when we
863 * created the cgroup. This will free cgrp->root, if we are
864 * holding the last reference to @sb.
866 deactivate_super(cgrp->root->sb);
869 * if we're getting rid of the cgroup, refcount should ensure
870 * that there are no pidlists left.
872 BUG_ON(!list_empty(&cgrp->pidlists));
874 simple_xattrs_free(&cgrp->xattrs);
876 kfree(rcu_dereference_raw(cgrp->name));
880 static void cgroup_free_rcu(struct rcu_head *head)
882 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
884 queue_work(cgroup_destroy_wq, &cgrp->free_work);
887 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
889 /* is dentry a directory ? if so, kfree() associated cgroup */
890 if (S_ISDIR(inode->i_mode)) {
891 struct cgroup *cgrp = dentry->d_fsdata;
893 BUG_ON(!(cgroup_is_removed(cgrp)));
894 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
896 struct cfent *cfe = __d_cfe(dentry);
897 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
899 WARN_ONCE(!list_empty(&cfe->node) &&
900 cgrp != &cgrp->root->top_cgroup,
901 "cfe still linked for %s\n", cfe->type->name);
902 simple_xattrs_free(&cfe->xattrs);
908 static int cgroup_delete(const struct dentry *d)
913 static void remove_dir(struct dentry *d)
915 struct dentry *parent = dget(d->d_parent);
918 simple_rmdir(parent->d_inode, d);
922 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
926 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
927 lockdep_assert_held(&cgroup_mutex);
930 * If we're doing cleanup due to failure of cgroup_create(),
931 * the corresponding @cfe may not exist.
933 list_for_each_entry(cfe, &cgrp->files, node) {
934 struct dentry *d = cfe->dentry;
936 if (cft && cfe->type != cft)
941 simple_unlink(cgrp->dentry->d_inode, d);
942 list_del_init(&cfe->node);
950 * cgroup_clear_directory - selective removal of base and subsystem files
951 * @dir: directory containing the files
952 * @base_files: true if the base files should be removed
953 * @subsys_mask: mask of the subsystem ids whose files should be removed
955 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
956 unsigned long subsys_mask)
958 struct cgroup *cgrp = __d_cgrp(dir);
959 struct cgroup_subsys *ss;
961 for_each_subsys(cgrp->root, ss) {
962 struct cftype_set *set;
963 if (!test_bit(ss->subsys_id, &subsys_mask))
965 list_for_each_entry(set, &ss->cftsets, node)
966 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
969 while (!list_empty(&cgrp->files))
970 cgroup_rm_file(cgrp, NULL);
975 * NOTE : the dentry must have been dget()'ed
977 static void cgroup_d_remove_dir(struct dentry *dentry)
979 struct dentry *parent;
980 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
982 cgroup_clear_directory(dentry, true, root->subsys_mask);
984 parent = dentry->d_parent;
985 spin_lock(&parent->d_lock);
986 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
987 list_del_init(&dentry->d_child);
988 spin_unlock(&dentry->d_lock);
989 spin_unlock(&parent->d_lock);
994 * Call with cgroup_mutex held. Drops reference counts on modules, including
995 * any duplicate ones that parse_cgroupfs_options took. If this function
996 * returns an error, no reference counts are touched.
998 static int rebind_subsystems(struct cgroupfs_root *root,
999 unsigned long final_subsys_mask)
1001 unsigned long added_mask, removed_mask;
1002 struct cgroup *cgrp = &root->top_cgroup;
1005 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1006 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1008 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
1009 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1010 /* Check that any added subsystems are currently free */
1011 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1012 unsigned long bit = 1UL << i;
1013 struct cgroup_subsys *ss = subsys[i];
1014 if (!(bit & added_mask))
1017 * Nobody should tell us to do a subsys that doesn't exist:
1018 * parse_cgroupfs_options should catch that case and refcounts
1019 * ensure that subsystems won't disappear once selected.
1022 if (ss->root != &rootnode) {
1023 /* Subsystem isn't free */
1028 /* Currently we don't handle adding/removing subsystems when
1029 * any child cgroups exist. This is theoretically supportable
1030 * but involves complex error handling, so it's being left until
1032 if (root->number_of_cgroups > 1)
1035 /* Process each subsystem */
1036 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1037 struct cgroup_subsys *ss = subsys[i];
1038 unsigned long bit = 1UL << i;
1039 if (bit & added_mask) {
1040 /* We're binding this subsystem to this hierarchy */
1042 BUG_ON(cgrp->subsys[i]);
1043 BUG_ON(!dummytop->subsys[i]);
1044 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1045 cgrp->subsys[i] = dummytop->subsys[i];
1046 cgrp->subsys[i]->cgroup = cgrp;
1047 list_move(&ss->sibling, &root->subsys_list);
1051 /* refcount was already taken, and we're keeping it */
1052 } else if (bit & removed_mask) {
1053 /* We're removing this subsystem */
1055 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1056 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1059 dummytop->subsys[i]->cgroup = dummytop;
1060 cgrp->subsys[i] = NULL;
1061 subsys[i]->root = &rootnode;
1062 list_move(&ss->sibling, &rootnode.subsys_list);
1063 /* subsystem is now free - drop reference on module */
1064 module_put(ss->module);
1065 } else if (bit & final_subsys_mask) {
1066 /* Subsystem state should already exist */
1068 BUG_ON(!cgrp->subsys[i]);
1070 * a refcount was taken, but we already had one, so
1071 * drop the extra reference.
1073 module_put(ss->module);
1074 #ifdef CONFIG_MODULE_UNLOAD
1075 BUG_ON(ss->module && !module_refcount(ss->module));
1078 /* Subsystem state shouldn't exist */
1079 BUG_ON(cgrp->subsys[i]);
1082 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1087 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1089 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1090 struct cgroup_subsys *ss;
1092 mutex_lock(&cgroup_root_mutex);
1093 for_each_subsys(root, ss)
1094 seq_printf(seq, ",%s", ss->name);
1095 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR)
1096 seq_puts(seq, ",sane_behavior");
1097 if (root->flags & CGRP_ROOT_NOPREFIX)
1098 seq_puts(seq, ",noprefix");
1099 if (root->flags & CGRP_ROOT_XATTR)
1100 seq_puts(seq, ",xattr");
1101 if (strlen(root->release_agent_path))
1102 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1103 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1104 seq_puts(seq, ",clone_children");
1105 if (strlen(root->name))
1106 seq_printf(seq, ",name=%s", root->name);
1107 mutex_unlock(&cgroup_root_mutex);
1111 struct cgroup_sb_opts {
1112 unsigned long subsys_mask;
1113 unsigned long flags;
1114 char *release_agent;
1115 bool cpuset_clone_children;
1117 /* User explicitly requested empty subsystem */
1120 struct cgroupfs_root *new_root;
1125 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1126 * with cgroup_mutex held to protect the subsys[] array. This function takes
1127 * refcounts on subsystems to be used, unless it returns error, in which case
1128 * no refcounts are taken.
1130 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1132 char *token, *o = data;
1133 bool all_ss = false, one_ss = false;
1134 unsigned long mask = (unsigned long)-1;
1136 bool module_pin_failed = false;
1138 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1140 #ifdef CONFIG_CPUSETS
1141 mask = ~(1UL << cpuset_subsys_id);
1144 memset(opts, 0, sizeof(*opts));
1146 while ((token = strsep(&o, ",")) != NULL) {
1149 if (!strcmp(token, "none")) {
1150 /* Explicitly have no subsystems */
1154 if (!strcmp(token, "all")) {
1155 /* Mutually exclusive option 'all' + subsystem name */
1161 if (!strcmp(token, "__DEVEL__sane_behavior")) {
1162 opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
1165 if (!strcmp(token, "noprefix")) {
1166 opts->flags |= CGRP_ROOT_NOPREFIX;
1169 if (!strcmp(token, "clone_children")) {
1170 opts->cpuset_clone_children = true;
1173 if (!strcmp(token, "xattr")) {
1174 opts->flags |= CGRP_ROOT_XATTR;
1177 if (!strncmp(token, "release_agent=", 14)) {
1178 /* Specifying two release agents is forbidden */
1179 if (opts->release_agent)
1181 opts->release_agent =
1182 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1183 if (!opts->release_agent)
1187 if (!strncmp(token, "name=", 5)) {
1188 const char *name = token + 5;
1189 /* Can't specify an empty name */
1192 /* Must match [\w.-]+ */
1193 for (i = 0; i < strlen(name); i++) {
1197 if ((c == '.') || (c == '-') || (c == '_'))
1201 /* Specifying two names is forbidden */
1204 opts->name = kstrndup(name,
1205 MAX_CGROUP_ROOT_NAMELEN - 1,
1213 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1214 struct cgroup_subsys *ss = subsys[i];
1217 if (strcmp(token, ss->name))
1222 /* Mutually exclusive option 'all' + subsystem name */
1225 set_bit(i, &opts->subsys_mask);
1230 if (i == CGROUP_SUBSYS_COUNT)
1235 * If the 'all' option was specified select all the subsystems,
1236 * otherwise if 'none', 'name=' and a subsystem name options
1237 * were not specified, let's default to 'all'
1239 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1240 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1241 struct cgroup_subsys *ss = subsys[i];
1246 set_bit(i, &opts->subsys_mask);
1250 /* Consistency checks */
1252 if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1253 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1255 if (opts->flags & CGRP_ROOT_NOPREFIX) {
1256 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1260 if (opts->cpuset_clone_children) {
1261 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1267 * Option noprefix was introduced just for backward compatibility
1268 * with the old cpuset, so we allow noprefix only if mounting just
1269 * the cpuset subsystem.
1271 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1275 /* Can't specify "none" and some subsystems */
1276 if (opts->subsys_mask && opts->none)
1280 * We either have to specify by name or by subsystems. (So all
1281 * empty hierarchies must have a name).
1283 if (!opts->subsys_mask && !opts->name)
1287 * Grab references on all the modules we'll need, so the subsystems
1288 * don't dance around before rebind_subsystems attaches them. This may
1289 * take duplicate reference counts on a subsystem that's already used,
1290 * but rebind_subsystems handles this case.
1292 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1293 unsigned long bit = 1UL << i;
1295 if (!(bit & opts->subsys_mask))
1297 if (!try_module_get(subsys[i]->module)) {
1298 module_pin_failed = true;
1302 if (module_pin_failed) {
1304 * oops, one of the modules was going away. this means that we
1305 * raced with a module_delete call, and to the user this is
1306 * essentially a "subsystem doesn't exist" case.
1308 for (i--; i >= 0; i--) {
1309 /* drop refcounts only on the ones we took */
1310 unsigned long bit = 1UL << i;
1312 if (!(bit & opts->subsys_mask))
1314 module_put(subsys[i]->module);
1322 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1325 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1326 unsigned long bit = 1UL << i;
1328 if (!(bit & subsys_mask))
1330 module_put(subsys[i]->module);
1334 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1337 struct cgroupfs_root *root = sb->s_fs_info;
1338 struct cgroup *cgrp = &root->top_cgroup;
1339 struct cgroup_sb_opts opts;
1340 unsigned long added_mask, removed_mask;
1342 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1343 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1347 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1348 mutex_lock(&cgroup_mutex);
1349 mutex_lock(&cgroup_root_mutex);
1351 /* See what subsystems are wanted */
1352 ret = parse_cgroupfs_options(data, &opts);
1356 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1357 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1358 task_tgid_nr(current), current->comm);
1360 added_mask = opts.subsys_mask & ~root->subsys_mask;
1361 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1363 /* Don't allow flags or name to change at remount */
1364 if (opts.flags != root->flags ||
1365 (opts.name && strcmp(opts.name, root->name))) {
1367 drop_parsed_module_refcounts(opts.subsys_mask);
1372 * Clear out the files of subsystems that should be removed, do
1373 * this before rebind_subsystems, since rebind_subsystems may
1374 * change this hierarchy's subsys_list.
1376 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1378 ret = rebind_subsystems(root, opts.subsys_mask);
1380 /* rebind_subsystems failed, re-populate the removed files */
1381 cgroup_populate_dir(cgrp, false, removed_mask);
1382 drop_parsed_module_refcounts(opts.subsys_mask);
1386 /* re-populate subsystem files */
1387 cgroup_populate_dir(cgrp, false, added_mask);
1389 if (opts.release_agent)
1390 strcpy(root->release_agent_path, opts.release_agent);
1392 kfree(opts.release_agent);
1394 mutex_unlock(&cgroup_root_mutex);
1395 mutex_unlock(&cgroup_mutex);
1396 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1400 static const struct super_operations cgroup_ops = {
1401 .statfs = simple_statfs,
1402 .drop_inode = generic_delete_inode,
1403 .show_options = cgroup_show_options,
1404 .remount_fs = cgroup_remount,
1407 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1409 INIT_LIST_HEAD(&cgrp->sibling);
1410 INIT_LIST_HEAD(&cgrp->children);
1411 INIT_LIST_HEAD(&cgrp->files);
1412 INIT_LIST_HEAD(&cgrp->css_sets);
1413 INIT_LIST_HEAD(&cgrp->allcg_node);
1414 INIT_LIST_HEAD(&cgrp->release_list);
1415 INIT_LIST_HEAD(&cgrp->pidlists);
1416 INIT_WORK(&cgrp->free_work, cgroup_free_fn);
1417 mutex_init(&cgrp->pidlist_mutex);
1418 INIT_LIST_HEAD(&cgrp->event_list);
1419 spin_lock_init(&cgrp->event_list_lock);
1420 simple_xattrs_init(&cgrp->xattrs);
1423 static void init_cgroup_root(struct cgroupfs_root *root)
1425 struct cgroup *cgrp = &root->top_cgroup;
1427 INIT_LIST_HEAD(&root->subsys_list);
1428 INIT_LIST_HEAD(&root->root_list);
1429 INIT_LIST_HEAD(&root->allcg_list);
1430 root->number_of_cgroups = 1;
1432 cgrp->name = &root_cgroup_name;
1433 init_cgroup_housekeeping(cgrp);
1434 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1437 static bool init_root_id(struct cgroupfs_root *root)
1442 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1444 spin_lock(&hierarchy_id_lock);
1445 /* Try to allocate the next unused ID */
1446 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1447 &root->hierarchy_id);
1449 /* Try again starting from 0 */
1450 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1452 next_hierarchy_id = root->hierarchy_id + 1;
1453 } else if (ret != -EAGAIN) {
1454 /* Can only get here if the 31-bit IDR is full ... */
1457 spin_unlock(&hierarchy_id_lock);
1462 static int cgroup_test_super(struct super_block *sb, void *data)
1464 struct cgroup_sb_opts *opts = data;
1465 struct cgroupfs_root *root = sb->s_fs_info;
1467 /* If we asked for a name then it must match */
1468 if (opts->name && strcmp(opts->name, root->name))
1472 * If we asked for subsystems (or explicitly for no
1473 * subsystems) then they must match
1475 if ((opts->subsys_mask || opts->none)
1476 && (opts->subsys_mask != root->subsys_mask))
1482 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1484 struct cgroupfs_root *root;
1486 if (!opts->subsys_mask && !opts->none)
1489 root = kzalloc(sizeof(*root), GFP_KERNEL);
1491 return ERR_PTR(-ENOMEM);
1493 if (!init_root_id(root)) {
1495 return ERR_PTR(-ENOMEM);
1497 init_cgroup_root(root);
1499 root->subsys_mask = opts->subsys_mask;
1500 root->flags = opts->flags;
1501 ida_init(&root->cgroup_ida);
1502 if (opts->release_agent)
1503 strcpy(root->release_agent_path, opts->release_agent);
1505 strcpy(root->name, opts->name);
1506 if (opts->cpuset_clone_children)
1507 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1511 static void cgroup_drop_root(struct cgroupfs_root *root)
1516 BUG_ON(!root->hierarchy_id);
1517 spin_lock(&hierarchy_id_lock);
1518 ida_remove(&hierarchy_ida, root->hierarchy_id);
1519 spin_unlock(&hierarchy_id_lock);
1520 ida_destroy(&root->cgroup_ida);
1524 static int cgroup_set_super(struct super_block *sb, void *data)
1527 struct cgroup_sb_opts *opts = data;
1529 /* If we don't have a new root, we can't set up a new sb */
1530 if (!opts->new_root)
1533 BUG_ON(!opts->subsys_mask && !opts->none);
1535 ret = set_anon_super(sb, NULL);
1539 sb->s_fs_info = opts->new_root;
1540 opts->new_root->sb = sb;
1542 sb->s_blocksize = PAGE_CACHE_SIZE;
1543 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1544 sb->s_magic = CGROUP_SUPER_MAGIC;
1545 sb->s_op = &cgroup_ops;
1550 static int cgroup_get_rootdir(struct super_block *sb)
1552 static const struct dentry_operations cgroup_dops = {
1553 .d_iput = cgroup_diput,
1554 .d_delete = cgroup_delete,
1557 struct inode *inode =
1558 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1563 inode->i_fop = &simple_dir_operations;
1564 inode->i_op = &cgroup_dir_inode_operations;
1565 /* directories start off with i_nlink == 2 (for "." entry) */
1567 sb->s_root = d_make_root(inode);
1570 /* for everything else we want ->d_op set */
1571 sb->s_d_op = &cgroup_dops;
1575 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1576 int flags, const char *unused_dev_name,
1579 struct cgroup_sb_opts opts;
1580 struct cgroupfs_root *root;
1582 struct super_block *sb;
1583 struct cgroupfs_root *new_root;
1584 struct inode *inode;
1586 /* First find the desired set of subsystems */
1587 mutex_lock(&cgroup_mutex);
1588 ret = parse_cgroupfs_options(data, &opts);
1589 mutex_unlock(&cgroup_mutex);
1594 * Allocate a new cgroup root. We may not need it if we're
1595 * reusing an existing hierarchy.
1597 new_root = cgroup_root_from_opts(&opts);
1598 if (IS_ERR(new_root)) {
1599 ret = PTR_ERR(new_root);
1602 opts.new_root = new_root;
1604 /* Locate an existing or new sb for this hierarchy */
1605 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1608 cgroup_drop_root(opts.new_root);
1612 root = sb->s_fs_info;
1614 if (root == opts.new_root) {
1615 /* We used the new root structure, so this is a new hierarchy */
1616 struct list_head tmp_cg_links;
1617 struct cgroup *root_cgrp = &root->top_cgroup;
1618 struct cgroupfs_root *existing_root;
1619 const struct cred *cred;
1623 BUG_ON(sb->s_root != NULL);
1625 ret = cgroup_get_rootdir(sb);
1627 goto drop_new_super;
1628 inode = sb->s_root->d_inode;
1630 mutex_lock(&inode->i_mutex);
1631 mutex_lock(&cgroup_mutex);
1632 mutex_lock(&cgroup_root_mutex);
1634 /* Check for name clashes with existing mounts */
1636 if (strlen(root->name))
1637 for_each_active_root(existing_root)
1638 if (!strcmp(existing_root->name, root->name))
1642 * We're accessing css_set_count without locking
1643 * css_set_lock here, but that's OK - it can only be
1644 * increased by someone holding cgroup_lock, and
1645 * that's us. The worst that can happen is that we
1646 * have some link structures left over
1648 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1652 ret = rebind_subsystems(root, root->subsys_mask);
1653 if (ret == -EBUSY) {
1654 free_cg_links(&tmp_cg_links);
1658 * There must be no failure case after here, since rebinding
1659 * takes care of subsystems' refcounts, which are explicitly
1660 * dropped in the failure exit path.
1663 /* EBUSY should be the only error here */
1666 list_add(&root->root_list, &roots);
1669 sb->s_root->d_fsdata = root_cgrp;
1670 root->top_cgroup.dentry = sb->s_root;
1672 /* Link the top cgroup in this hierarchy into all
1673 * the css_set objects */
1674 write_lock(&css_set_lock);
1675 hash_for_each(css_set_table, i, cg, hlist)
1676 link_css_set(&tmp_cg_links, cg, root_cgrp);
1677 write_unlock(&css_set_lock);
1679 free_cg_links(&tmp_cg_links);
1681 BUG_ON(!list_empty(&root_cgrp->children));
1682 BUG_ON(root->number_of_cgroups != 1);
1684 cred = override_creds(&init_cred);
1685 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1687 mutex_unlock(&cgroup_root_mutex);
1688 mutex_unlock(&cgroup_mutex);
1689 mutex_unlock(&inode->i_mutex);
1692 * We re-used an existing hierarchy - the new root (if
1693 * any) is not needed
1695 cgroup_drop_root(opts.new_root);
1697 if (root->flags != opts.flags) {
1698 if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) {
1699 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1701 goto drop_new_super;
1703 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1707 /* no subsys rebinding, so refcounts don't change */
1708 drop_parsed_module_refcounts(opts.subsys_mask);
1711 kfree(opts.release_agent);
1713 return dget(sb->s_root);
1716 mutex_unlock(&cgroup_root_mutex);
1717 mutex_unlock(&cgroup_mutex);
1718 mutex_unlock(&inode->i_mutex);
1720 deactivate_locked_super(sb);
1722 drop_parsed_module_refcounts(opts.subsys_mask);
1724 kfree(opts.release_agent);
1726 return ERR_PTR(ret);
1729 static void cgroup_kill_sb(struct super_block *sb) {
1730 struct cgroupfs_root *root = sb->s_fs_info;
1731 struct cgroup *cgrp = &root->top_cgroup;
1733 struct cg_cgroup_link *link;
1734 struct cg_cgroup_link *saved_link;
1738 BUG_ON(root->number_of_cgroups != 1);
1739 BUG_ON(!list_empty(&cgrp->children));
1741 mutex_lock(&cgroup_mutex);
1742 mutex_lock(&cgroup_root_mutex);
1744 /* Rebind all subsystems back to the default hierarchy */
1745 ret = rebind_subsystems(root, 0);
1746 /* Shouldn't be able to fail ... */
1750 * Release all the links from css_sets to this hierarchy's
1753 write_lock(&css_set_lock);
1755 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1757 list_del(&link->cg_link_list);
1758 list_del(&link->cgrp_link_list);
1761 write_unlock(&css_set_lock);
1763 if (!list_empty(&root->root_list)) {
1764 list_del(&root->root_list);
1768 mutex_unlock(&cgroup_root_mutex);
1769 mutex_unlock(&cgroup_mutex);
1771 simple_xattrs_free(&cgrp->xattrs);
1773 kill_litter_super(sb);
1774 cgroup_drop_root(root);
1777 static struct file_system_type cgroup_fs_type = {
1779 .mount = cgroup_mount,
1780 .kill_sb = cgroup_kill_sb,
1783 static struct kobject *cgroup_kobj;
1786 * cgroup_path - generate the path of a cgroup
1787 * @cgrp: the cgroup in question
1788 * @buf: the buffer to write the path into
1789 * @buflen: the length of the buffer
1791 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1793 * We can't generate cgroup path using dentry->d_name, as accessing
1794 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1795 * inode's i_mutex, while on the other hand cgroup_path() can be called
1796 * with some irq-safe spinlocks held.
1798 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1800 int ret = -ENAMETOOLONG;
1803 if (!cgrp->parent) {
1804 if (strlcpy(buf, "/", buflen) >= buflen)
1805 return -ENAMETOOLONG;
1809 start = buf + buflen - 1;
1814 const char *name = cgroup_name(cgrp);
1818 if ((start -= len) < buf)
1820 memcpy(start, name, len);
1826 cgrp = cgrp->parent;
1827 } while (cgrp->parent);
1829 memmove(buf, start, buf + buflen - start);
1834 EXPORT_SYMBOL_GPL(cgroup_path);
1837 * Control Group taskset
1839 struct task_and_cgroup {
1840 struct task_struct *task;
1841 struct cgroup *cgrp;
1845 struct cgroup_taskset {
1846 struct task_and_cgroup single;
1847 struct flex_array *tc_array;
1850 struct cgroup *cur_cgrp;
1854 * cgroup_taskset_first - reset taskset and return the first task
1855 * @tset: taskset of interest
1857 * @tset iteration is initialized and the first task is returned.
1859 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1861 if (tset->tc_array) {
1863 return cgroup_taskset_next(tset);
1865 tset->cur_cgrp = tset->single.cgrp;
1866 return tset->single.task;
1869 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1872 * cgroup_taskset_next - iterate to the next task in taskset
1873 * @tset: taskset of interest
1875 * Return the next task in @tset. Iteration must have been initialized
1876 * with cgroup_taskset_first().
1878 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1880 struct task_and_cgroup *tc;
1882 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1885 tc = flex_array_get(tset->tc_array, tset->idx++);
1886 tset->cur_cgrp = tc->cgrp;
1889 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1892 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1893 * @tset: taskset of interest
1895 * Return the cgroup for the current (last returned) task of @tset. This
1896 * function must be preceded by either cgroup_taskset_first() or
1897 * cgroup_taskset_next().
1899 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1901 return tset->cur_cgrp;
1903 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1906 * cgroup_taskset_size - return the number of tasks in taskset
1907 * @tset: taskset of interest
1909 int cgroup_taskset_size(struct cgroup_taskset *tset)
1911 return tset->tc_array ? tset->tc_array_len : 1;
1913 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1917 * cgroup_task_migrate - move a task from one cgroup to another.
1919 * Must be called with cgroup_mutex and threadgroup locked.
1921 static void cgroup_task_migrate(struct cgroup *oldcgrp,
1922 struct task_struct *tsk, struct css_set *newcg)
1924 struct css_set *oldcg;
1927 * We are synchronized through threadgroup_lock() against PF_EXITING
1928 * setting such that we can't race against cgroup_exit() changing the
1929 * css_set to init_css_set and dropping the old one.
1931 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1932 oldcg = tsk->cgroups;
1935 rcu_assign_pointer(tsk->cgroups, newcg);
1938 /* Update the css_set linked lists if we're using them */
1939 write_lock(&css_set_lock);
1940 if (!list_empty(&tsk->cg_list))
1941 list_move(&tsk->cg_list, &newcg->tasks);
1942 write_unlock(&css_set_lock);
1945 * We just gained a reference on oldcg by taking it from the task. As
1946 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1947 * it here; it will be freed under RCU.
1949 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1954 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1955 * @cgrp: the cgroup to attach to
1956 * @tsk: the task or the leader of the threadgroup to be attached
1957 * @threadgroup: attach the whole threadgroup?
1959 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1960 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1962 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
1965 int retval, i, group_size;
1966 struct cgroup_subsys *ss, *failed_ss = NULL;
1967 struct cgroupfs_root *root = cgrp->root;
1968 /* threadgroup list cursor and array */
1969 struct task_struct *leader = tsk;
1970 struct task_and_cgroup *tc;
1971 struct flex_array *group;
1972 struct cgroup_taskset tset = { };
1975 * step 0: in order to do expensive, possibly blocking operations for
1976 * every thread, we cannot iterate the thread group list, since it needs
1977 * rcu or tasklist locked. instead, build an array of all threads in the
1978 * group - group_rwsem prevents new threads from appearing, and if
1979 * threads exit, this will just be an over-estimate.
1982 group_size = get_nr_threads(tsk);
1985 /* flex_array supports very large thread-groups better than kmalloc. */
1986 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
1989 /* pre-allocate to guarantee space while iterating in rcu read-side. */
1990 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
1992 goto out_free_group_list;
1996 * Prevent freeing of tasks while we take a snapshot. Tasks that are
1997 * already PF_EXITING could be freed from underneath us unless we
1998 * take an rcu_read_lock.
2002 struct task_and_cgroup ent;
2004 /* @tsk either already exited or can't exit until the end */
2005 if (tsk->flags & PF_EXITING)
2008 /* as per above, nr_threads may decrease, but not increase. */
2009 BUG_ON(i >= group_size);
2011 ent.cgrp = task_cgroup_from_root(tsk, root);
2012 /* nothing to do if this task is already in the cgroup */
2013 if (ent.cgrp == cgrp)
2016 * saying GFP_ATOMIC has no effect here because we did prealloc
2017 * earlier, but it's good form to communicate our expectations.
2019 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2020 BUG_ON(retval != 0);
2025 } while_each_thread(leader, tsk);
2027 /* remember the number of threads in the array for later. */
2029 tset.tc_array = group;
2030 tset.tc_array_len = group_size;
2032 /* methods shouldn't be called if no task is actually migrating */
2035 goto out_free_group_list;
2038 * step 1: check that we can legitimately attach to the cgroup.
2040 for_each_subsys(root, ss) {
2041 if (ss->can_attach) {
2042 retval = ss->can_attach(cgrp, &tset);
2045 goto out_cancel_attach;
2051 * step 2: make sure css_sets exist for all threads to be migrated.
2052 * we use find_css_set, which allocates a new one if necessary.
2054 for (i = 0; i < group_size; i++) {
2055 tc = flex_array_get(group, i);
2056 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2059 goto out_put_css_set_refs;
2064 * step 3: now that we're guaranteed success wrt the css_sets,
2065 * proceed to move all tasks to the new cgroup. There are no
2066 * failure cases after here, so this is the commit point.
2068 for (i = 0; i < group_size; i++) {
2069 tc = flex_array_get(group, i);
2070 cgroup_task_migrate(tc->cgrp, tc->task, tc->cg);
2072 /* nothing is sensitive to fork() after this point. */
2075 * step 4: do subsystem attach callbacks.
2077 for_each_subsys(root, ss) {
2079 ss->attach(cgrp, &tset);
2083 * step 5: success! and cleanup
2086 out_put_css_set_refs:
2088 for (i = 0; i < group_size; i++) {
2089 tc = flex_array_get(group, i);
2092 put_css_set(tc->cg);
2097 for_each_subsys(root, ss) {
2098 if (ss == failed_ss)
2100 if (ss->cancel_attach)
2101 ss->cancel_attach(cgrp, &tset);
2104 out_free_group_list:
2105 flex_array_free(group);
2109 static int cgroup_allow_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
2111 struct cgroup_subsys *ss;
2114 for_each_subsys(cgrp->root, ss) {
2115 if (ss->allow_attach) {
2116 ret = ss->allow_attach(cgrp, tset);
2127 int subsys_cgroup_allow_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
2129 const struct cred *cred = current_cred(), *tcred;
2130 struct task_struct *task;
2132 if (capable(CAP_SYS_NICE))
2135 cgroup_taskset_for_each(task, cgrp, tset) {
2136 tcred = __task_cred(task);
2138 if (current != task && cred->euid != tcred->uid &&
2139 cred->euid != tcred->suid)
2147 * Find the task_struct of the task to attach by vpid and pass it along to the
2148 * function to attach either it or all tasks in its threadgroup. Will lock
2149 * cgroup_mutex and threadgroup; may take task_lock of task.
2151 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2153 struct task_struct *tsk;
2154 const struct cred *cred = current_cred(), *tcred;
2157 if (!cgroup_lock_live_group(cgrp))
2163 tsk = find_task_by_vpid(pid);
2167 goto out_unlock_cgroup;
2170 * even if we're attaching all tasks in the thread group, we
2171 * only need to check permissions on one of them.
2173 tcred = __task_cred(tsk);
2174 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2175 !uid_eq(cred->euid, tcred->uid) &&
2176 !uid_eq(cred->euid, tcred->suid)) {
2178 * if the default permission check fails, give each
2179 * cgroup a chance to extend the permission check
2181 struct cgroup_taskset tset = { };
2182 tset.single.task = tsk;
2183 tset.single.cgrp = cgrp;
2184 ret = cgroup_allow_attach(cgrp, &tset);
2187 goto out_unlock_cgroup;
2194 tsk = tsk->group_leader;
2197 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2198 * trapped in a cpuset, or RT worker may be born in a cgroup
2199 * with no rt_runtime allocated. Just say no.
2201 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
2204 goto out_unlock_cgroup;
2207 get_task_struct(tsk);
2210 threadgroup_lock(tsk);
2212 if (!thread_group_leader(tsk)) {
2214 * a race with de_thread from another thread's exec()
2215 * may strip us of our leadership, if this happens,
2216 * there is no choice but to throw this task away and
2217 * try again; this is
2218 * "double-double-toil-and-trouble-check locking".
2220 threadgroup_unlock(tsk);
2221 put_task_struct(tsk);
2222 goto retry_find_task;
2226 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2228 threadgroup_unlock(tsk);
2230 put_task_struct(tsk);
2232 mutex_unlock(&cgroup_mutex);
2237 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2238 * @from: attach to all cgroups of a given task
2239 * @tsk: the task to be attached
2241 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2243 struct cgroupfs_root *root;
2246 mutex_lock(&cgroup_mutex);
2247 for_each_active_root(root) {
2248 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2250 retval = cgroup_attach_task(from_cg, tsk, false);
2254 mutex_unlock(&cgroup_mutex);
2258 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2260 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2262 return attach_task_by_pid(cgrp, pid, false);
2265 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2267 return attach_task_by_pid(cgrp, tgid, true);
2270 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2273 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2274 if (strlen(buffer) >= PATH_MAX)
2276 if (!cgroup_lock_live_group(cgrp))
2278 mutex_lock(&cgroup_root_mutex);
2279 strcpy(cgrp->root->release_agent_path, buffer);
2280 mutex_unlock(&cgroup_root_mutex);
2281 mutex_unlock(&cgroup_mutex);
2285 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2286 struct seq_file *seq)
2288 if (!cgroup_lock_live_group(cgrp))
2290 seq_puts(seq, cgrp->root->release_agent_path);
2291 seq_putc(seq, '\n');
2292 mutex_unlock(&cgroup_mutex);
2296 static int cgroup_sane_behavior_show(struct cgroup *cgrp, struct cftype *cft,
2297 struct seq_file *seq)
2299 seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp));
2303 /* A buffer size big enough for numbers or short strings */
2304 #define CGROUP_LOCAL_BUFFER_SIZE 64
2306 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2308 const char __user *userbuf,
2309 size_t nbytes, loff_t *unused_ppos)
2311 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2317 if (nbytes >= sizeof(buffer))
2319 if (copy_from_user(buffer, userbuf, nbytes))
2322 buffer[nbytes] = 0; /* nul-terminate */
2323 if (cft->write_u64) {
2324 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2327 retval = cft->write_u64(cgrp, cft, val);
2329 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2332 retval = cft->write_s64(cgrp, cft, val);
2339 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2341 const char __user *userbuf,
2342 size_t nbytes, loff_t *unused_ppos)
2344 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2346 size_t max_bytes = cft->max_write_len;
2347 char *buffer = local_buffer;
2350 max_bytes = sizeof(local_buffer) - 1;
2351 if (nbytes >= max_bytes)
2353 /* Allocate a dynamic buffer if we need one */
2354 if (nbytes >= sizeof(local_buffer)) {
2355 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2359 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2364 buffer[nbytes] = 0; /* nul-terminate */
2365 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2369 if (buffer != local_buffer)
2374 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2375 size_t nbytes, loff_t *ppos)
2377 struct cftype *cft = __d_cft(file->f_dentry);
2378 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2380 if (cgroup_is_removed(cgrp))
2383 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2384 if (cft->write_u64 || cft->write_s64)
2385 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2386 if (cft->write_string)
2387 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2389 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2390 return ret ? ret : nbytes;
2395 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2397 char __user *buf, size_t nbytes,
2400 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2401 u64 val = cft->read_u64(cgrp, cft);
2402 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2404 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2407 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2409 char __user *buf, size_t nbytes,
2412 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2413 s64 val = cft->read_s64(cgrp, cft);
2414 int len = sprintf(tmp, "%lld\n", (long long) val);
2416 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2419 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2420 size_t nbytes, loff_t *ppos)
2422 struct cftype *cft = __d_cft(file->f_dentry);
2423 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2425 if (cgroup_is_removed(cgrp))
2429 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2431 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2433 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2438 * seqfile ops/methods for returning structured data. Currently just
2439 * supports string->u64 maps, but can be extended in future.
2442 struct cgroup_seqfile_state {
2444 struct cgroup *cgroup;
2447 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2449 struct seq_file *sf = cb->state;
2450 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2453 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2455 struct cgroup_seqfile_state *state = m->private;
2456 struct cftype *cft = state->cft;
2457 if (cft->read_map) {
2458 struct cgroup_map_cb cb = {
2459 .fill = cgroup_map_add,
2462 return cft->read_map(state->cgroup, cft, &cb);
2464 return cft->read_seq_string(state->cgroup, cft, m);
2467 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2469 struct seq_file *seq = file->private_data;
2470 kfree(seq->private);
2471 return single_release(inode, file);
2474 static const struct file_operations cgroup_seqfile_operations = {
2476 .write = cgroup_file_write,
2477 .llseek = seq_lseek,
2478 .release = cgroup_seqfile_release,
2481 static int cgroup_file_open(struct inode *inode, struct file *file)
2486 err = generic_file_open(inode, file);
2489 cft = __d_cft(file->f_dentry);
2491 if (cft->read_map || cft->read_seq_string) {
2492 struct cgroup_seqfile_state *state =
2493 kzalloc(sizeof(*state), GFP_USER);
2497 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2498 file->f_op = &cgroup_seqfile_operations;
2499 err = single_open(file, cgroup_seqfile_show, state);
2502 } else if (cft->open)
2503 err = cft->open(inode, file);
2510 static int cgroup_file_release(struct inode *inode, struct file *file)
2512 struct cftype *cft = __d_cft(file->f_dentry);
2514 return cft->release(inode, file);
2519 * cgroup_rename - Only allow simple rename of directories in place.
2521 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2522 struct inode *new_dir, struct dentry *new_dentry)
2525 struct cgroup_name *name, *old_name;
2526 struct cgroup *cgrp;
2529 * It's convinient to use parent dir's i_mutex to protected
2532 lockdep_assert_held(&old_dir->i_mutex);
2534 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2536 if (new_dentry->d_inode)
2538 if (old_dir != new_dir)
2541 cgrp = __d_cgrp(old_dentry);
2543 name = cgroup_alloc_name(new_dentry);
2547 ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2553 old_name = cgrp->name;
2554 rcu_assign_pointer(cgrp->name, name);
2556 kfree_rcu(old_name, rcu_head);
2560 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2562 if (S_ISDIR(dentry->d_inode->i_mode))
2563 return &__d_cgrp(dentry)->xattrs;
2565 return &__d_cfe(dentry)->xattrs;
2568 static inline int xattr_enabled(struct dentry *dentry)
2570 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2571 return root->flags & CGRP_ROOT_XATTR;
2574 static bool is_valid_xattr(const char *name)
2576 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2577 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2582 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2583 const void *val, size_t size, int flags)
2585 if (!xattr_enabled(dentry))
2587 if (!is_valid_xattr(name))
2589 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2592 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2594 if (!xattr_enabled(dentry))
2596 if (!is_valid_xattr(name))
2598 return simple_xattr_remove(__d_xattrs(dentry), name);
2601 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2602 void *buf, size_t size)
2604 if (!xattr_enabled(dentry))
2606 if (!is_valid_xattr(name))
2608 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2611 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2613 if (!xattr_enabled(dentry))
2615 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2618 static const struct file_operations cgroup_file_operations = {
2619 .read = cgroup_file_read,
2620 .write = cgroup_file_write,
2621 .llseek = generic_file_llseek,
2622 .open = cgroup_file_open,
2623 .release = cgroup_file_release,
2626 static const struct inode_operations cgroup_file_inode_operations = {
2627 .setxattr = cgroup_setxattr,
2628 .getxattr = cgroup_getxattr,
2629 .listxattr = cgroup_listxattr,
2630 .removexattr = cgroup_removexattr,
2633 static const struct inode_operations cgroup_dir_inode_operations = {
2634 .lookup = cgroup_lookup,
2635 .mkdir = cgroup_mkdir,
2636 .rmdir = cgroup_rmdir,
2637 .rename = cgroup_rename,
2638 .setxattr = cgroup_setxattr,
2639 .getxattr = cgroup_getxattr,
2640 .listxattr = cgroup_listxattr,
2641 .removexattr = cgroup_removexattr,
2644 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2646 if (dentry->d_name.len > NAME_MAX)
2647 return ERR_PTR(-ENAMETOOLONG);
2648 d_add(dentry, NULL);
2653 * Check if a file is a control file
2655 static inline struct cftype *__file_cft(struct file *file)
2657 if (file_inode(file)->i_fop != &cgroup_file_operations)
2658 return ERR_PTR(-EINVAL);
2659 return __d_cft(file->f_dentry);
2662 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2663 struct super_block *sb)
2665 struct inode *inode;
2669 if (dentry->d_inode)
2672 inode = cgroup_new_inode(mode, sb);
2676 if (S_ISDIR(mode)) {
2677 inode->i_op = &cgroup_dir_inode_operations;
2678 inode->i_fop = &simple_dir_operations;
2680 /* start off with i_nlink == 2 (for "." entry) */
2682 inc_nlink(dentry->d_parent->d_inode);
2685 * Control reaches here with cgroup_mutex held.
2686 * @inode->i_mutex should nest outside cgroup_mutex but we
2687 * want to populate it immediately without releasing
2688 * cgroup_mutex. As @inode isn't visible to anyone else
2689 * yet, trylock will always succeed without affecting
2692 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2693 } else if (S_ISREG(mode)) {
2695 inode->i_fop = &cgroup_file_operations;
2696 inode->i_op = &cgroup_file_inode_operations;
2698 d_instantiate(dentry, inode);
2699 dget(dentry); /* Extra count - pin the dentry in core */
2704 * cgroup_file_mode - deduce file mode of a control file
2705 * @cft: the control file in question
2707 * returns cft->mode if ->mode is not 0
2708 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2709 * returns S_IRUGO if it has only a read handler
2710 * returns S_IWUSR if it has only a write hander
2712 static umode_t cgroup_file_mode(const struct cftype *cft)
2719 if (cft->read || cft->read_u64 || cft->read_s64 ||
2720 cft->read_map || cft->read_seq_string)
2723 if (cft->write || cft->write_u64 || cft->write_s64 ||
2724 cft->write_string || cft->trigger)
2730 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2733 struct dentry *dir = cgrp->dentry;
2734 struct cgroup *parent = __d_cgrp(dir);
2735 struct dentry *dentry;
2739 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2741 if (subsys && !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
2742 strcpy(name, subsys->name);
2745 strcat(name, cft->name);
2747 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2749 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2753 dentry = lookup_one_len(name, dir, strlen(name));
2754 if (IS_ERR(dentry)) {
2755 error = PTR_ERR(dentry);
2759 cfe->type = (void *)cft;
2760 cfe->dentry = dentry;
2761 dentry->d_fsdata = cfe;
2762 simple_xattrs_init(&cfe->xattrs);
2764 mode = cgroup_file_mode(cft);
2765 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2767 list_add_tail(&cfe->node, &parent->files);
2776 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2777 struct cftype cfts[], bool is_add)
2782 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2783 /* does cft->flags tell us to skip this file on @cgrp? */
2784 if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp))
2786 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2788 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2792 err = cgroup_add_file(cgrp, subsys, cft);
2794 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2798 cgroup_rm_file(cgrp, cft);
2804 static DEFINE_MUTEX(cgroup_cft_mutex);
2806 static void cgroup_cfts_prepare(void)
2807 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2810 * Thanks to the entanglement with vfs inode locking, we can't walk
2811 * the existing cgroups under cgroup_mutex and create files.
2812 * Instead, we increment reference on all cgroups and build list of
2813 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2814 * exclusive access to the field.
2816 mutex_lock(&cgroup_cft_mutex);
2817 mutex_lock(&cgroup_mutex);
2820 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2821 struct cftype *cfts, bool is_add)
2822 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2825 struct cgroup *cgrp, *n;
2826 struct super_block *sb = ss->root->sb;
2828 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2829 if (cfts && ss->root != &rootnode &&
2830 atomic_inc_not_zero(&sb->s_active)) {
2831 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2833 list_add_tail(&cgrp->cft_q_node, &pending);
2839 mutex_unlock(&cgroup_mutex);
2842 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2843 * files for all cgroups which were created before.
2845 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2846 struct inode *inode = cgrp->dentry->d_inode;
2848 mutex_lock(&inode->i_mutex);
2849 mutex_lock(&cgroup_mutex);
2850 if (!cgroup_is_removed(cgrp))
2851 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2852 mutex_unlock(&cgroup_mutex);
2853 mutex_unlock(&inode->i_mutex);
2855 list_del_init(&cgrp->cft_q_node);
2860 deactivate_super(sb);
2862 mutex_unlock(&cgroup_cft_mutex);
2866 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2867 * @ss: target cgroup subsystem
2868 * @cfts: zero-length name terminated array of cftypes
2870 * Register @cfts to @ss. Files described by @cfts are created for all
2871 * existing cgroups to which @ss is attached and all future cgroups will
2872 * have them too. This function can be called anytime whether @ss is
2875 * Returns 0 on successful registration, -errno on failure. Note that this
2876 * function currently returns 0 as long as @cfts registration is successful
2877 * even if some file creation attempts on existing cgroups fail.
2879 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2881 struct cftype_set *set;
2883 set = kzalloc(sizeof(*set), GFP_KERNEL);
2887 cgroup_cfts_prepare();
2889 list_add_tail(&set->node, &ss->cftsets);
2890 cgroup_cfts_commit(ss, cfts, true);
2894 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2897 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2898 * @ss: target cgroup subsystem
2899 * @cfts: zero-length name terminated array of cftypes
2901 * Unregister @cfts from @ss. Files described by @cfts are removed from
2902 * all existing cgroups to which @ss is attached and all future cgroups
2903 * won't have them either. This function can be called anytime whether @ss
2904 * is attached or not.
2906 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2907 * registered with @ss.
2909 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2911 struct cftype_set *set;
2913 cgroup_cfts_prepare();
2915 list_for_each_entry(set, &ss->cftsets, node) {
2916 if (set->cfts == cfts) {
2917 list_del_init(&set->node);
2918 cgroup_cfts_commit(ss, cfts, false);
2923 cgroup_cfts_commit(ss, NULL, false);
2928 * cgroup_task_count - count the number of tasks in a cgroup.
2929 * @cgrp: the cgroup in question
2931 * Return the number of tasks in the cgroup.
2933 int cgroup_task_count(const struct cgroup *cgrp)
2936 struct cg_cgroup_link *link;
2938 read_lock(&css_set_lock);
2939 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2940 count += atomic_read(&link->cg->refcount);
2942 read_unlock(&css_set_lock);
2947 * Advance a list_head iterator. The iterator should be positioned at
2948 * the start of a css_set
2950 static void cgroup_advance_iter(struct cgroup *cgrp,
2951 struct cgroup_iter *it)
2953 struct list_head *l = it->cg_link;
2954 struct cg_cgroup_link *link;
2957 /* Advance to the next non-empty css_set */
2960 if (l == &cgrp->css_sets) {
2964 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2966 } while (list_empty(&cg->tasks));
2968 it->task = cg->tasks.next;
2972 * To reduce the fork() overhead for systems that are not actually
2973 * using their cgroups capability, we don't maintain the lists running
2974 * through each css_set to its tasks until we see the list actually
2975 * used - in other words after the first call to cgroup_iter_start().
2977 static void cgroup_enable_task_cg_lists(void)
2979 struct task_struct *p, *g;
2980 write_lock(&css_set_lock);
2981 use_task_css_set_links = 1;
2983 * We need tasklist_lock because RCU is not safe against
2984 * while_each_thread(). Besides, a forking task that has passed
2985 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2986 * is not guaranteed to have its child immediately visible in the
2987 * tasklist if we walk through it with RCU.
2989 read_lock(&tasklist_lock);
2990 do_each_thread(g, p) {
2993 * We should check if the process is exiting, otherwise
2994 * it will race with cgroup_exit() in that the list
2995 * entry won't be deleted though the process has exited.
2997 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2998 list_add(&p->cg_list, &p->cgroups->tasks);
3000 } while_each_thread(g, p);
3001 read_unlock(&tasklist_lock);
3002 write_unlock(&css_set_lock);
3006 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
3007 * @pos: the current position (%NULL to initiate traversal)
3008 * @cgroup: cgroup whose descendants to walk
3010 * To be used by cgroup_for_each_descendant_pre(). Find the next
3011 * descendant to visit for pre-order traversal of @cgroup's descendants.
3013 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
3014 struct cgroup *cgroup)
3016 struct cgroup *next;
3018 WARN_ON_ONCE(!rcu_read_lock_held());
3020 /* if first iteration, pretend we just visited @cgroup */
3024 /* visit the first child if exists */
3025 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
3029 /* no child, visit my or the closest ancestor's next sibling */
3030 while (pos != cgroup) {
3031 next = list_entry_rcu(pos->sibling.next, struct cgroup,
3033 if (&next->sibling != &pos->parent->children)
3041 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
3044 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
3045 * @pos: cgroup of interest
3047 * Return the rightmost descendant of @pos. If there's no descendant,
3048 * @pos is returned. This can be used during pre-order traversal to skip
3051 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
3053 struct cgroup *last, *tmp;
3055 WARN_ON_ONCE(!rcu_read_lock_held());
3059 /* ->prev isn't RCU safe, walk ->next till the end */
3061 list_for_each_entry_rcu(tmp, &last->children, sibling)
3067 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
3069 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3071 struct cgroup *last;
3075 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3083 * cgroup_next_descendant_post - find the next descendant for post-order walk
3084 * @pos: the current position (%NULL to initiate traversal)
3085 * @cgroup: cgroup whose descendants to walk
3087 * To be used by cgroup_for_each_descendant_post(). Find the next
3088 * descendant to visit for post-order traversal of @cgroup's descendants.
3090 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3091 struct cgroup *cgroup)
3093 struct cgroup *next;
3095 WARN_ON_ONCE(!rcu_read_lock_held());
3097 /* if first iteration, visit the leftmost descendant */
3099 next = cgroup_leftmost_descendant(cgroup);
3100 return next != cgroup ? next : NULL;
3103 /* if there's an unvisited sibling, visit its leftmost descendant */
3104 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3105 if (&next->sibling != &pos->parent->children)
3106 return cgroup_leftmost_descendant(next);
3108 /* no sibling left, visit parent */
3110 return next != cgroup ? next : NULL;
3112 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3114 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3115 __acquires(css_set_lock)
3118 * The first time anyone tries to iterate across a cgroup,
3119 * we need to enable the list linking each css_set to its
3120 * tasks, and fix up all existing tasks.
3122 if (!use_task_css_set_links)
3123 cgroup_enable_task_cg_lists();
3125 read_lock(&css_set_lock);
3126 it->cg_link = &cgrp->css_sets;
3127 cgroup_advance_iter(cgrp, it);
3130 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3131 struct cgroup_iter *it)
3133 struct task_struct *res;
3134 struct list_head *l = it->task;
3135 struct cg_cgroup_link *link;
3137 /* If the iterator cg is NULL, we have no tasks */
3140 res = list_entry(l, struct task_struct, cg_list);
3141 /* Advance iterator to find next entry */
3143 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
3144 if (l == &link->cg->tasks) {
3145 /* We reached the end of this task list - move on to
3146 * the next cg_cgroup_link */
3147 cgroup_advance_iter(cgrp, it);
3154 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3155 __releases(css_set_lock)
3157 read_unlock(&css_set_lock);
3160 static inline int started_after_time(struct task_struct *t1,
3161 struct timespec *time,
3162 struct task_struct *t2)
3164 int start_diff = timespec_compare(&t1->start_time, time);
3165 if (start_diff > 0) {
3167 } else if (start_diff < 0) {
3171 * Arbitrarily, if two processes started at the same
3172 * time, we'll say that the lower pointer value
3173 * started first. Note that t2 may have exited by now
3174 * so this may not be a valid pointer any longer, but
3175 * that's fine - it still serves to distinguish
3176 * between two tasks started (effectively) simultaneously.
3183 * This function is a callback from heap_insert() and is used to order
3185 * In this case we order the heap in descending task start time.
3187 static inline int started_after(void *p1, void *p2)
3189 struct task_struct *t1 = p1;
3190 struct task_struct *t2 = p2;
3191 return started_after_time(t1, &t2->start_time, t2);
3195 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3196 * @scan: struct cgroup_scanner containing arguments for the scan
3198 * Arguments include pointers to callback functions test_task() and
3200 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3201 * and if it returns true, call process_task() for it also.
3202 * The test_task pointer may be NULL, meaning always true (select all tasks).
3203 * Effectively duplicates cgroup_iter_{start,next,end}()
3204 * but does not lock css_set_lock for the call to process_task().
3205 * The struct cgroup_scanner may be embedded in any structure of the caller's
3207 * It is guaranteed that process_task() will act on every task that
3208 * is a member of the cgroup for the duration of this call. This
3209 * function may or may not call process_task() for tasks that exit
3210 * or move to a different cgroup during the call, or are forked or
3211 * move into the cgroup during the call.
3213 * Note that test_task() may be called with locks held, and may in some
3214 * situations be called multiple times for the same task, so it should
3216 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3217 * pre-allocated and will be used for heap operations (and its "gt" member will
3218 * be overwritten), else a temporary heap will be used (allocation of which
3219 * may cause this function to fail).
3221 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3224 struct cgroup_iter it;
3225 struct task_struct *p, *dropped;
3226 /* Never dereference latest_task, since it's not refcounted */
3227 struct task_struct *latest_task = NULL;
3228 struct ptr_heap tmp_heap;
3229 struct ptr_heap *heap;
3230 struct timespec latest_time = { 0, 0 };
3233 /* The caller supplied our heap and pre-allocated its memory */
3235 heap->gt = &started_after;
3237 /* We need to allocate our own heap memory */
3239 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3241 /* cannot allocate the heap */
3247 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3248 * to determine which are of interest, and using the scanner's
3249 * "process_task" callback to process any of them that need an update.
3250 * Since we don't want to hold any locks during the task updates,
3251 * gather tasks to be processed in a heap structure.
3252 * The heap is sorted by descending task start time.
3253 * If the statically-sized heap fills up, we overflow tasks that
3254 * started later, and in future iterations only consider tasks that
3255 * started after the latest task in the previous pass. This
3256 * guarantees forward progress and that we don't miss any tasks.
3259 cgroup_iter_start(scan->cg, &it);
3260 while ((p = cgroup_iter_next(scan->cg, &it))) {
3262 * Only affect tasks that qualify per the caller's callback,
3263 * if he provided one
3265 if (scan->test_task && !scan->test_task(p, scan))
3268 * Only process tasks that started after the last task
3271 if (!started_after_time(p, &latest_time, latest_task))
3273 dropped = heap_insert(heap, p);
3274 if (dropped == NULL) {
3276 * The new task was inserted; the heap wasn't
3280 } else if (dropped != p) {
3282 * The new task was inserted, and pushed out a
3286 put_task_struct(dropped);
3289 * Else the new task was newer than anything already in
3290 * the heap and wasn't inserted
3293 cgroup_iter_end(scan->cg, &it);
3296 for (i = 0; i < heap->size; i++) {
3297 struct task_struct *q = heap->ptrs[i];
3299 latest_time = q->start_time;
3302 /* Process the task per the caller's callback */
3303 scan->process_task(q, scan);
3307 * If we had to process any tasks at all, scan again
3308 * in case some of them were in the middle of forking
3309 * children that didn't get processed.
3310 * Not the most efficient way to do it, but it avoids
3311 * having to take callback_mutex in the fork path
3315 if (heap == &tmp_heap)
3316 heap_free(&tmp_heap);
3320 static void cgroup_transfer_one_task(struct task_struct *task,
3321 struct cgroup_scanner *scan)
3323 struct cgroup *new_cgroup = scan->data;
3325 mutex_lock(&cgroup_mutex);
3326 cgroup_attach_task(new_cgroup, task, false);
3327 mutex_unlock(&cgroup_mutex);
3331 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3332 * @to: cgroup to which the tasks will be moved
3333 * @from: cgroup in which the tasks currently reside
3335 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3337 struct cgroup_scanner scan;
3340 scan.test_task = NULL; /* select all tasks in cgroup */
3341 scan.process_task = cgroup_transfer_one_task;
3345 return cgroup_scan_tasks(&scan);
3349 * Stuff for reading the 'tasks'/'procs' files.
3351 * Reading this file can return large amounts of data if a cgroup has
3352 * *lots* of attached tasks. So it may need several calls to read(),
3353 * but we cannot guarantee that the information we produce is correct
3354 * unless we produce it entirely atomically.
3358 /* which pidlist file are we talking about? */
3359 enum cgroup_filetype {
3365 * A pidlist is a list of pids that virtually represents the contents of one
3366 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3367 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3370 struct cgroup_pidlist {
3372 * used to find which pidlist is wanted. doesn't change as long as
3373 * this particular list stays in the list.
3375 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3378 /* how many elements the above list has */
3380 /* how many files are using the current array */
3382 /* each of these stored in a list by its cgroup */
3383 struct list_head links;
3384 /* pointer to the cgroup we belong to, for list removal purposes */
3385 struct cgroup *owner;
3386 /* protects the other fields */
3387 struct rw_semaphore mutex;
3391 * The following two functions "fix" the issue where there are more pids
3392 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3393 * TODO: replace with a kernel-wide solution to this problem
3395 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3396 static void *pidlist_allocate(int count)
3398 if (PIDLIST_TOO_LARGE(count))
3399 return vmalloc(count * sizeof(pid_t));
3401 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3403 static void pidlist_free(void *p)
3405 if (is_vmalloc_addr(p))
3412 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3413 * Returns the number of unique elements.
3415 static int pidlist_uniq(pid_t *list, int length)
3420 * we presume the 0th element is unique, so i starts at 1. trivial
3421 * edge cases first; no work needs to be done for either
3423 if (length == 0 || length == 1)
3425 /* src and dest walk down the list; dest counts unique elements */
3426 for (src = 1; src < length; src++) {
3427 /* find next unique element */
3428 while (list[src] == list[src-1]) {
3433 /* dest always points to where the next unique element goes */
3434 list[dest] = list[src];
3441 static int cmppid(const void *a, const void *b)
3443 return *(pid_t *)a - *(pid_t *)b;
3447 * find the appropriate pidlist for our purpose (given procs vs tasks)
3448 * returns with the lock on that pidlist already held, and takes care
3449 * of the use count, or returns NULL with no locks held if we're out of
3452 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3453 enum cgroup_filetype type)
3455 struct cgroup_pidlist *l;
3456 /* don't need task_nsproxy() if we're looking at ourself */
3457 struct pid_namespace *ns = task_active_pid_ns(current);
3460 * We can't drop the pidlist_mutex before taking the l->mutex in case
3461 * the last ref-holder is trying to remove l from the list at the same
3462 * time. Holding the pidlist_mutex precludes somebody taking whichever
3463 * list we find out from under us - compare release_pid_array().
3465 mutex_lock(&cgrp->pidlist_mutex);
3466 list_for_each_entry(l, &cgrp->pidlists, links) {
3467 if (l->key.type == type && l->key.ns == ns) {
3468 /* make sure l doesn't vanish out from under us */
3469 down_write(&l->mutex);
3470 mutex_unlock(&cgrp->pidlist_mutex);
3474 /* entry not found; create a new one */
3475 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3477 mutex_unlock(&cgrp->pidlist_mutex);
3480 init_rwsem(&l->mutex);
3481 down_write(&l->mutex);
3483 l->key.ns = get_pid_ns(ns);
3484 l->use_count = 0; /* don't increment here */
3487 list_add(&l->links, &cgrp->pidlists);
3488 mutex_unlock(&cgrp->pidlist_mutex);
3493 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3495 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3496 struct cgroup_pidlist **lp)
3500 int pid, n = 0; /* used for populating the array */
3501 struct cgroup_iter it;
3502 struct task_struct *tsk;
3503 struct cgroup_pidlist *l;
3506 * If cgroup gets more users after we read count, we won't have
3507 * enough space - tough. This race is indistinguishable to the
3508 * caller from the case that the additional cgroup users didn't
3509 * show up until sometime later on.
3511 length = cgroup_task_count(cgrp);
3512 array = pidlist_allocate(length);
3515 /* now, populate the array */
3516 cgroup_iter_start(cgrp, &it);
3517 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3518 if (unlikely(n == length))
3520 /* get tgid or pid for procs or tasks file respectively */
3521 if (type == CGROUP_FILE_PROCS)
3522 pid = task_tgid_vnr(tsk);
3524 pid = task_pid_vnr(tsk);
3525 if (pid > 0) /* make sure to only use valid results */
3528 cgroup_iter_end(cgrp, &it);
3530 /* now sort & (if procs) strip out duplicates */
3531 sort(array, length, sizeof(pid_t), cmppid, NULL);
3532 if (type == CGROUP_FILE_PROCS)
3533 length = pidlist_uniq(array, length);
3534 l = cgroup_pidlist_find(cgrp, type);
3536 pidlist_free(array);
3539 /* store array, freeing old if necessary - lock already held */
3540 pidlist_free(l->list);
3544 up_write(&l->mutex);
3550 * cgroupstats_build - build and fill cgroupstats
3551 * @stats: cgroupstats to fill information into
3552 * @dentry: A dentry entry belonging to the cgroup for which stats have
3555 * Build and fill cgroupstats so that taskstats can export it to user
3558 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3561 struct cgroup *cgrp;
3562 struct cgroup_iter it;
3563 struct task_struct *tsk;
3566 * Validate dentry by checking the superblock operations,
3567 * and make sure it's a directory.
3569 if (dentry->d_sb->s_op != &cgroup_ops ||
3570 !S_ISDIR(dentry->d_inode->i_mode))
3574 cgrp = dentry->d_fsdata;
3576 cgroup_iter_start(cgrp, &it);
3577 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3578 switch (tsk->state) {
3580 stats->nr_running++;
3582 case TASK_INTERRUPTIBLE:
3583 stats->nr_sleeping++;
3585 case TASK_UNINTERRUPTIBLE:
3586 stats->nr_uninterruptible++;
3589 stats->nr_stopped++;
3592 if (delayacct_is_task_waiting_on_io(tsk))
3593 stats->nr_io_wait++;
3597 cgroup_iter_end(cgrp, &it);
3605 * seq_file methods for the tasks/procs files. The seq_file position is the
3606 * next pid to display; the seq_file iterator is a pointer to the pid
3607 * in the cgroup->l->list array.
3610 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3613 * Initially we receive a position value that corresponds to
3614 * one more than the last pid shown (or 0 on the first call or
3615 * after a seek to the start). Use a binary-search to find the
3616 * next pid to display, if any
3618 struct cgroup_pidlist *l = s->private;
3619 int index = 0, pid = *pos;
3622 down_read(&l->mutex);
3624 int end = l->length;
3626 while (index < end) {
3627 int mid = (index + end) / 2;
3628 if (l->list[mid] == pid) {
3631 } else if (l->list[mid] <= pid)
3637 /* If we're off the end of the array, we're done */
3638 if (index >= l->length)
3640 /* Update the abstract position to be the actual pid that we found */
3641 iter = l->list + index;
3646 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3648 struct cgroup_pidlist *l = s->private;
3652 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3654 struct cgroup_pidlist *l = s->private;
3656 pid_t *end = l->list + l->length;
3658 * Advance to the next pid in the array. If this goes off the
3670 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3672 return seq_printf(s, "%d\n", *(int *)v);
3676 * seq_operations functions for iterating on pidlists through seq_file -
3677 * independent of whether it's tasks or procs
3679 static const struct seq_operations cgroup_pidlist_seq_operations = {
3680 .start = cgroup_pidlist_start,
3681 .stop = cgroup_pidlist_stop,
3682 .next = cgroup_pidlist_next,
3683 .show = cgroup_pidlist_show,
3686 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3689 * the case where we're the last user of this particular pidlist will
3690 * have us remove it from the cgroup's list, which entails taking the
3691 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3692 * pidlist_mutex, we have to take pidlist_mutex first.
3694 mutex_lock(&l->owner->pidlist_mutex);
3695 down_write(&l->mutex);
3696 BUG_ON(!l->use_count);
3697 if (!--l->use_count) {
3698 /* we're the last user if refcount is 0; remove and free */
3699 list_del(&l->links);
3700 mutex_unlock(&l->owner->pidlist_mutex);
3701 pidlist_free(l->list);
3702 put_pid_ns(l->key.ns);
3703 up_write(&l->mutex);
3707 mutex_unlock(&l->owner->pidlist_mutex);
3708 up_write(&l->mutex);
3711 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3713 struct cgroup_pidlist *l;
3714 if (!(file->f_mode & FMODE_READ))
3717 * the seq_file will only be initialized if the file was opened for
3718 * reading; hence we check if it's not null only in that case.
3720 l = ((struct seq_file *)file->private_data)->private;
3721 cgroup_release_pid_array(l);
3722 return seq_release(inode, file);
3725 static const struct file_operations cgroup_pidlist_operations = {
3727 .llseek = seq_lseek,
3728 .write = cgroup_file_write,
3729 .release = cgroup_pidlist_release,
3733 * The following functions handle opens on a file that displays a pidlist
3734 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3737 /* helper function for the two below it */
3738 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3740 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3741 struct cgroup_pidlist *l;
3744 /* Nothing to do for write-only files */
3745 if (!(file->f_mode & FMODE_READ))
3748 /* have the array populated */
3749 retval = pidlist_array_load(cgrp, type, &l);
3752 /* configure file information */
3753 file->f_op = &cgroup_pidlist_operations;
3755 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3757 cgroup_release_pid_array(l);
3760 ((struct seq_file *)file->private_data)->private = l;
3763 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3765 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3767 static int cgroup_procs_open(struct inode *unused, struct file *file)
3769 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3772 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3775 return notify_on_release(cgrp);
3778 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3782 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3784 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3786 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3791 * When dput() is called asynchronously, if umount has been done and
3792 * then deactivate_super() in cgroup_free_fn() kills the superblock,
3793 * there's a small window that vfs will see the root dentry with non-zero
3794 * refcnt and trigger BUG().
3796 * That's why we hold a reference before dput() and drop it right after.
3798 static void cgroup_dput(struct cgroup *cgrp)
3800 struct super_block *sb = cgrp->root->sb;
3802 atomic_inc(&sb->s_active);
3804 deactivate_super(sb);
3808 * Unregister event and free resources.
3810 * Gets called from workqueue.
3812 static void cgroup_event_remove(struct work_struct *work)
3814 struct cgroup_event *event = container_of(work, struct cgroup_event,
3816 struct cgroup *cgrp = event->cgrp;
3818 remove_wait_queue(event->wqh, &event->wait);
3820 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3822 /* Notify userspace the event is going away. */
3823 eventfd_signal(event->eventfd, 1);
3825 eventfd_ctx_put(event->eventfd);
3831 * Gets called on POLLHUP on eventfd when user closes it.
3833 * Called with wqh->lock held and interrupts disabled.
3835 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3836 int sync, void *key)
3838 struct cgroup_event *event = container_of(wait,
3839 struct cgroup_event, wait);
3840 struct cgroup *cgrp = event->cgrp;
3841 unsigned long flags = (unsigned long)key;
3843 if (flags & POLLHUP) {
3845 * If the event has been detached at cgroup removal, we
3846 * can simply return knowing the other side will cleanup
3849 * We can't race against event freeing since the other
3850 * side will require wqh->lock via remove_wait_queue(),
3853 spin_lock(&cgrp->event_list_lock);
3854 if (!list_empty(&event->list)) {
3855 list_del_init(&event->list);
3857 * We are in atomic context, but cgroup_event_remove()
3858 * may sleep, so we have to call it in workqueue.
3860 schedule_work(&event->remove);
3862 spin_unlock(&cgrp->event_list_lock);
3868 static void cgroup_event_ptable_queue_proc(struct file *file,
3869 wait_queue_head_t *wqh, poll_table *pt)
3871 struct cgroup_event *event = container_of(pt,
3872 struct cgroup_event, pt);
3875 add_wait_queue(wqh, &event->wait);
3879 * Parse input and register new cgroup event handler.
3881 * Input must be in format '<event_fd> <control_fd> <args>'.
3882 * Interpretation of args is defined by control file implementation.
3884 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3887 struct cgroup_event *event = NULL;
3888 struct cgroup *cgrp_cfile;
3889 unsigned int efd, cfd;
3890 struct file *efile = NULL;
3891 struct file *cfile = NULL;
3895 efd = simple_strtoul(buffer, &endp, 10);
3900 cfd = simple_strtoul(buffer, &endp, 10);
3901 if ((*endp != ' ') && (*endp != '\0'))
3905 event = kzalloc(sizeof(*event), GFP_KERNEL);
3909 INIT_LIST_HEAD(&event->list);
3910 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3911 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3912 INIT_WORK(&event->remove, cgroup_event_remove);
3914 efile = eventfd_fget(efd);
3915 if (IS_ERR(efile)) {
3916 ret = PTR_ERR(efile);
3920 event->eventfd = eventfd_ctx_fileget(efile);
3921 if (IS_ERR(event->eventfd)) {
3922 ret = PTR_ERR(event->eventfd);
3932 /* the process need read permission on control file */
3933 /* AV: shouldn't we check that it's been opened for read instead? */
3934 ret = inode_permission(file_inode(cfile), MAY_READ);
3938 event->cft = __file_cft(cfile);
3939 if (IS_ERR(event->cft)) {
3940 ret = PTR_ERR(event->cft);
3945 * The file to be monitored must be in the same cgroup as
3946 * cgroup.event_control is.
3948 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
3949 if (cgrp_cfile != cgrp) {
3954 if (!event->cft->register_event || !event->cft->unregister_event) {
3959 ret = event->cft->register_event(cgrp, event->cft,
3960 event->eventfd, buffer);
3964 efile->f_op->poll(efile, &event->pt);
3967 * Events should be removed after rmdir of cgroup directory, but before
3968 * destroying subsystem state objects. Let's take reference to cgroup
3969 * directory dentry to do that.
3973 spin_lock(&cgrp->event_list_lock);
3974 list_add(&event->list, &cgrp->event_list);
3975 spin_unlock(&cgrp->event_list_lock);
3986 if (event && event->eventfd && !IS_ERR(event->eventfd))
3987 eventfd_ctx_put(event->eventfd);
3989 if (!IS_ERR_OR_NULL(efile))
3997 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
4000 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4003 static int cgroup_clone_children_write(struct cgroup *cgrp,
4008 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4010 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4015 * for the common functions, 'private' gives the type of file
4017 /* for hysterical raisins, we can't put this on the older files */
4018 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
4019 static struct cftype files[] = {
4022 .open = cgroup_tasks_open,
4023 .write_u64 = cgroup_tasks_write,
4024 .release = cgroup_pidlist_release,
4025 .mode = S_IRUGO | S_IWUSR,
4028 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
4029 .open = cgroup_procs_open,
4030 .write_u64 = cgroup_procs_write,
4031 .release = cgroup_pidlist_release,
4032 .mode = S_IRUGO | S_IWUSR,
4035 .name = "notify_on_release",
4036 .read_u64 = cgroup_read_notify_on_release,
4037 .write_u64 = cgroup_write_notify_on_release,
4040 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
4041 .write_string = cgroup_write_event_control,
4045 .name = "cgroup.clone_children",
4046 .flags = CFTYPE_INSANE,
4047 .read_u64 = cgroup_clone_children_read,
4048 .write_u64 = cgroup_clone_children_write,
4051 .name = "cgroup.sane_behavior",
4052 .flags = CFTYPE_ONLY_ON_ROOT,
4053 .read_seq_string = cgroup_sane_behavior_show,
4056 .name = "release_agent",
4057 .flags = CFTYPE_ONLY_ON_ROOT,
4058 .read_seq_string = cgroup_release_agent_show,
4059 .write_string = cgroup_release_agent_write,
4060 .max_write_len = PATH_MAX,
4066 * cgroup_populate_dir - selectively creation of files in a directory
4067 * @cgrp: target cgroup
4068 * @base_files: true if the base files should be added
4069 * @subsys_mask: mask of the subsystem ids whose files should be added
4071 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
4072 unsigned long subsys_mask)
4075 struct cgroup_subsys *ss;
4078 err = cgroup_addrm_files(cgrp, NULL, files, true);
4083 /* process cftsets of each subsystem */
4084 for_each_subsys(cgrp->root, ss) {
4085 struct cftype_set *set;
4086 if (!test_bit(ss->subsys_id, &subsys_mask))
4089 list_for_each_entry(set, &ss->cftsets, node)
4090 cgroup_addrm_files(cgrp, ss, set->cfts, true);
4093 /* This cgroup is ready now */
4094 for_each_subsys(cgrp->root, ss) {
4095 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4097 * Update id->css pointer and make this css visible from
4098 * CSS ID functions. This pointer will be dereferened
4099 * from RCU-read-side without locks.
4102 rcu_assign_pointer(css->id->css, css);
4108 static void css_dput_fn(struct work_struct *work)
4110 struct cgroup_subsys_state *css =
4111 container_of(work, struct cgroup_subsys_state, dput_work);
4113 cgroup_dput(css->cgroup);
4116 static void init_cgroup_css(struct cgroup_subsys_state *css,
4117 struct cgroup_subsys *ss,
4118 struct cgroup *cgrp)
4121 atomic_set(&css->refcnt, 1);
4124 if (cgrp == dummytop)
4125 css->flags |= CSS_ROOT;
4126 BUG_ON(cgrp->subsys[ss->subsys_id]);
4127 cgrp->subsys[ss->subsys_id] = css;
4130 * css holds an extra ref to @cgrp->dentry which is put on the last
4131 * css_put(). dput() requires process context, which css_put() may
4132 * be called without. @css->dput_work will be used to invoke
4133 * dput() asynchronously from css_put().
4135 INIT_WORK(&css->dput_work, css_dput_fn);
4138 /* invoke ->post_create() on a new CSS and mark it online if successful */
4139 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4143 lockdep_assert_held(&cgroup_mutex);
4146 ret = ss->css_online(cgrp);
4148 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4152 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4153 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4154 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4156 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4158 lockdep_assert_held(&cgroup_mutex);
4160 if (!(css->flags & CSS_ONLINE))
4163 if (ss->css_offline)
4164 ss->css_offline(cgrp);
4166 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4170 * cgroup_create - create a cgroup
4171 * @parent: cgroup that will be parent of the new cgroup
4172 * @dentry: dentry of the new cgroup
4173 * @mode: mode to set on new inode
4175 * Must be called with the mutex on the parent inode held
4177 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4180 struct cgroup *cgrp;
4181 struct cgroup_name *name;
4182 struct cgroupfs_root *root = parent->root;
4184 struct cgroup_subsys *ss;
4185 struct super_block *sb = root->sb;
4187 /* allocate the cgroup and its ID, 0 is reserved for the root */
4188 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4192 name = cgroup_alloc_name(dentry);
4195 rcu_assign_pointer(cgrp->name, name);
4197 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4202 * Only live parents can have children. Note that the liveliness
4203 * check isn't strictly necessary because cgroup_mkdir() and
4204 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4205 * anyway so that locking is contained inside cgroup proper and we
4206 * don't get nasty surprises if we ever grow another caller.
4208 if (!cgroup_lock_live_group(parent)) {
4213 /* Grab a reference on the superblock so the hierarchy doesn't
4214 * get deleted on unmount if there are child cgroups. This
4215 * can be done outside cgroup_mutex, since the sb can't
4216 * disappear while someone has an open control file on the
4218 atomic_inc(&sb->s_active);
4220 init_cgroup_housekeeping(cgrp);
4222 dentry->d_fsdata = cgrp;
4223 cgrp->dentry = dentry;
4225 cgrp->parent = parent;
4226 cgrp->root = parent->root;
4228 if (notify_on_release(parent))
4229 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4231 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4232 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4234 for_each_subsys(root, ss) {
4235 struct cgroup_subsys_state *css;
4237 css = ss->css_alloc(cgrp);
4242 init_cgroup_css(css, ss, cgrp);
4244 err = alloc_css_id(ss, parent, cgrp);
4251 * Create directory. cgroup_create_file() returns with the new
4252 * directory locked on success so that it can be populated without
4253 * dropping cgroup_mutex.
4255 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4258 lockdep_assert_held(&dentry->d_inode->i_mutex);
4260 /* allocation complete, commit to creation */
4261 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4262 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4263 root->number_of_cgroups++;
4265 /* each css holds a ref to the cgroup's dentry */
4266 for_each_subsys(root, ss)
4269 /* hold a ref to the parent's dentry */
4270 dget(parent->dentry);
4272 /* creation succeeded, notify subsystems */
4273 for_each_subsys(root, ss) {
4274 err = online_css(ss, cgrp);
4278 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4280 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",
4281 current->comm, current->pid, ss->name);
4282 if (!strcmp(ss->name, "memory"))
4283 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4284 ss->warned_broken_hierarchy = true;
4288 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4292 mutex_unlock(&cgroup_mutex);
4293 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4298 for_each_subsys(root, ss) {
4299 if (cgrp->subsys[ss->subsys_id])
4302 mutex_unlock(&cgroup_mutex);
4303 /* Release the reference count that we took on the superblock */
4304 deactivate_super(sb);
4306 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4308 kfree(rcu_dereference_raw(cgrp->name));
4314 cgroup_destroy_locked(cgrp);
4315 mutex_unlock(&cgroup_mutex);
4316 mutex_unlock(&dentry->d_inode->i_mutex);
4320 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4322 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4324 /* the vfs holds inode->i_mutex already */
4325 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4328 static int cgroup_destroy_locked(struct cgroup *cgrp)
4329 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4331 struct dentry *d = cgrp->dentry;
4332 struct cgroup *parent = cgrp->parent;
4333 struct cgroup_event *event, *tmp;
4334 struct cgroup_subsys *ss;
4336 lockdep_assert_held(&d->d_inode->i_mutex);
4337 lockdep_assert_held(&cgroup_mutex);
4339 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children))
4343 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4344 * removed. This makes future css_tryget() and child creation
4345 * attempts fail thus maintaining the removal conditions verified
4348 for_each_subsys(cgrp->root, ss) {
4349 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4351 WARN_ON(atomic_read(&css->refcnt) < 0);
4352 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4354 set_bit(CGRP_REMOVED, &cgrp->flags);
4356 /* tell subsystems to initate destruction */
4357 for_each_subsys(cgrp->root, ss)
4358 offline_css(ss, cgrp);
4361 * Put all the base refs. Each css holds an extra reference to the
4362 * cgroup's dentry and cgroup removal proceeds regardless of css
4363 * refs. On the last put of each css, whenever that may be, the
4364 * extra dentry ref is put so that dentry destruction happens only
4365 * after all css's are released.
4367 for_each_subsys(cgrp->root, ss)
4368 css_put(cgrp->subsys[ss->subsys_id]);
4370 raw_spin_lock(&release_list_lock);
4371 if (!list_empty(&cgrp->release_list))
4372 list_del_init(&cgrp->release_list);
4373 raw_spin_unlock(&release_list_lock);
4375 /* delete this cgroup from parent->children */
4376 list_del_rcu(&cgrp->sibling);
4377 list_del_init(&cgrp->allcg_node);
4380 cgroup_d_remove_dir(d);
4383 set_bit(CGRP_RELEASABLE, &parent->flags);
4384 check_for_release(parent);
4387 * Unregister events and notify userspace.
4388 * Notify userspace about cgroup removing only after rmdir of cgroup
4389 * directory to avoid race between userspace and kernelspace.
4391 spin_lock(&cgrp->event_list_lock);
4392 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4393 list_del_init(&event->list);
4394 schedule_work(&event->remove);
4396 spin_unlock(&cgrp->event_list_lock);
4401 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4405 mutex_lock(&cgroup_mutex);
4406 ret = cgroup_destroy_locked(dentry->d_fsdata);
4407 mutex_unlock(&cgroup_mutex);
4412 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4414 INIT_LIST_HEAD(&ss->cftsets);
4417 * base_cftset is embedded in subsys itself, no need to worry about
4420 if (ss->base_cftypes) {
4421 ss->base_cftset.cfts = ss->base_cftypes;
4422 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4426 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4428 struct cgroup_subsys_state *css;
4430 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4432 mutex_lock(&cgroup_mutex);
4434 /* init base cftset */
4435 cgroup_init_cftsets(ss);
4437 /* Create the top cgroup state for this subsystem */
4438 list_add(&ss->sibling, &rootnode.subsys_list);
4439 ss->root = &rootnode;
4440 css = ss->css_alloc(dummytop);
4441 /* We don't handle early failures gracefully */
4442 BUG_ON(IS_ERR(css));
4443 init_cgroup_css(css, ss, dummytop);
4445 /* Update the init_css_set to contain a subsys
4446 * pointer to this state - since the subsystem is
4447 * newly registered, all tasks and hence the
4448 * init_css_set is in the subsystem's top cgroup. */
4449 init_css_set.subsys[ss->subsys_id] = css;
4451 need_forkexit_callback |= ss->fork || ss->exit;
4453 /* At system boot, before all subsystems have been
4454 * registered, no tasks have been forked, so we don't
4455 * need to invoke fork callbacks here. */
4456 BUG_ON(!list_empty(&init_task.tasks));
4458 BUG_ON(online_css(ss, dummytop));
4460 mutex_unlock(&cgroup_mutex);
4462 /* this function shouldn't be used with modular subsystems, since they
4463 * need to register a subsys_id, among other things */
4468 * cgroup_load_subsys: load and register a modular subsystem at runtime
4469 * @ss: the subsystem to load
4471 * This function should be called in a modular subsystem's initcall. If the
4472 * subsystem is built as a module, it will be assigned a new subsys_id and set
4473 * up for use. If the subsystem is built-in anyway, work is delegated to the
4474 * simpler cgroup_init_subsys.
4476 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4478 struct cgroup_subsys_state *css;
4480 struct hlist_node *tmp;
4484 /* check name and function validity */
4485 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4486 ss->css_alloc == NULL || ss->css_free == NULL)
4490 * we don't support callbacks in modular subsystems. this check is
4491 * before the ss->module check for consistency; a subsystem that could
4492 * be a module should still have no callbacks even if the user isn't
4493 * compiling it as one.
4495 if (ss->fork || ss->exit)
4499 * an optionally modular subsystem is built-in: we want to do nothing,
4500 * since cgroup_init_subsys will have already taken care of it.
4502 if (ss->module == NULL) {
4503 /* a sanity check */
4504 BUG_ON(subsys[ss->subsys_id] != ss);
4508 /* init base cftset */
4509 cgroup_init_cftsets(ss);
4511 mutex_lock(&cgroup_mutex);
4512 subsys[ss->subsys_id] = ss;
4515 * no ss->css_alloc seems to need anything important in the ss
4516 * struct, so this can happen first (i.e. before the rootnode
4519 css = ss->css_alloc(dummytop);
4521 /* failure case - need to deassign the subsys[] slot. */
4522 subsys[ss->subsys_id] = NULL;
4523 mutex_unlock(&cgroup_mutex);
4524 return PTR_ERR(css);
4527 list_add(&ss->sibling, &rootnode.subsys_list);
4528 ss->root = &rootnode;
4530 /* our new subsystem will be attached to the dummy hierarchy. */
4531 init_cgroup_css(css, ss, dummytop);
4532 /* init_idr must be after init_cgroup_css because it sets css->id. */
4534 ret = cgroup_init_idr(ss, css);
4540 * Now we need to entangle the css into the existing css_sets. unlike
4541 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4542 * will need a new pointer to it; done by iterating the css_set_table.
4543 * furthermore, modifying the existing css_sets will corrupt the hash
4544 * table state, so each changed css_set will need its hash recomputed.
4545 * this is all done under the css_set_lock.
4547 write_lock(&css_set_lock);
4548 hash_for_each_safe(css_set_table, i, tmp, cg, hlist) {
4549 /* skip entries that we already rehashed */
4550 if (cg->subsys[ss->subsys_id])
4552 /* remove existing entry */
4553 hash_del(&cg->hlist);
4555 cg->subsys[ss->subsys_id] = css;
4556 /* recompute hash and restore entry */
4557 key = css_set_hash(cg->subsys);
4558 hash_add(css_set_table, &cg->hlist, key);
4560 write_unlock(&css_set_lock);
4562 ret = online_css(ss, dummytop);
4567 mutex_unlock(&cgroup_mutex);
4571 mutex_unlock(&cgroup_mutex);
4572 /* @ss can't be mounted here as try_module_get() would fail */
4573 cgroup_unload_subsys(ss);
4576 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4579 * cgroup_unload_subsys: unload a modular subsystem
4580 * @ss: the subsystem to unload
4582 * This function should be called in a modular subsystem's exitcall. When this
4583 * function is invoked, the refcount on the subsystem's module will be 0, so
4584 * the subsystem will not be attached to any hierarchy.
4586 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4588 struct cg_cgroup_link *link;
4590 BUG_ON(ss->module == NULL);
4593 * we shouldn't be called if the subsystem is in use, and the use of
4594 * try_module_get in parse_cgroupfs_options should ensure that it
4595 * doesn't start being used while we're killing it off.
4597 BUG_ON(ss->root != &rootnode);
4599 mutex_lock(&cgroup_mutex);
4601 offline_css(ss, dummytop);
4604 idr_destroy(&ss->idr);
4606 /* deassign the subsys_id */
4607 subsys[ss->subsys_id] = NULL;
4609 /* remove subsystem from rootnode's list of subsystems */
4610 list_del_init(&ss->sibling);
4613 * disentangle the css from all css_sets attached to the dummytop. as
4614 * in loading, we need to pay our respects to the hashtable gods.
4616 write_lock(&css_set_lock);
4617 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4618 struct css_set *cg = link->cg;
4621 hash_del(&cg->hlist);
4622 cg->subsys[ss->subsys_id] = NULL;
4623 key = css_set_hash(cg->subsys);
4624 hash_add(css_set_table, &cg->hlist, key);
4626 write_unlock(&css_set_lock);
4629 * remove subsystem's css from the dummytop and free it - need to
4630 * free before marking as null because ss->css_free needs the
4631 * cgrp->subsys pointer to find their state. note that this also
4632 * takes care of freeing the css_id.
4634 ss->css_free(dummytop);
4635 dummytop->subsys[ss->subsys_id] = NULL;
4637 mutex_unlock(&cgroup_mutex);
4639 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4642 * cgroup_init_early - cgroup initialization at system boot
4644 * Initialize cgroups at system boot, and initialize any
4645 * subsystems that request early init.
4647 int __init cgroup_init_early(void)
4650 atomic_set(&init_css_set.refcount, 1);
4651 INIT_LIST_HEAD(&init_css_set.cg_links);
4652 INIT_LIST_HEAD(&init_css_set.tasks);
4653 INIT_HLIST_NODE(&init_css_set.hlist);
4655 init_cgroup_root(&rootnode);
4657 init_task.cgroups = &init_css_set;
4659 init_css_set_link.cg = &init_css_set;
4660 init_css_set_link.cgrp = dummytop;
4661 list_add(&init_css_set_link.cgrp_link_list,
4662 &rootnode.top_cgroup.css_sets);
4663 list_add(&init_css_set_link.cg_link_list,
4664 &init_css_set.cg_links);
4666 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4667 struct cgroup_subsys *ss = subsys[i];
4669 /* at bootup time, we don't worry about modular subsystems */
4670 if (!ss || ss->module)
4674 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4675 BUG_ON(!ss->css_alloc);
4676 BUG_ON(!ss->css_free);
4677 if (ss->subsys_id != i) {
4678 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4679 ss->name, ss->subsys_id);
4684 cgroup_init_subsys(ss);
4690 * cgroup_init - cgroup initialization
4692 * Register cgroup filesystem and /proc file, and initialize
4693 * any subsystems that didn't request early init.
4695 int __init cgroup_init(void)
4701 err = bdi_init(&cgroup_backing_dev_info);
4705 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4706 struct cgroup_subsys *ss = subsys[i];
4708 /* at bootup time, we don't worry about modular subsystems */
4709 if (!ss || ss->module)
4711 if (!ss->early_init)
4712 cgroup_init_subsys(ss);
4714 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4717 /* Add init_css_set to the hash table */
4718 key = css_set_hash(init_css_set.subsys);
4719 hash_add(css_set_table, &init_css_set.hlist, key);
4720 BUG_ON(!init_root_id(&rootnode));
4722 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4728 err = register_filesystem(&cgroup_fs_type);
4730 kobject_put(cgroup_kobj);
4734 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4738 bdi_destroy(&cgroup_backing_dev_info);
4743 static int __init cgroup_wq_init(void)
4746 * There isn't much point in executing destruction path in
4747 * parallel. Good chunk is serialized with cgroup_mutex anyway.
4748 * Use 1 for @max_active.
4750 * We would prefer to do this in cgroup_init() above, but that
4751 * is called before init_workqueues(): so leave this until after.
4753 cgroup_destroy_wq = alloc_workqueue("cgroup_destroy", 0, 1);
4754 BUG_ON(!cgroup_destroy_wq);
4757 core_initcall(cgroup_wq_init);
4760 * proc_cgroup_show()
4761 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4762 * - Used for /proc/<pid>/cgroup.
4763 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4764 * doesn't really matter if tsk->cgroup changes after we read it,
4765 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4766 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4767 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4768 * cgroup to top_cgroup.
4771 /* TODO: Use a proper seq_file iterator */
4772 int proc_cgroup_show(struct seq_file *m, void *v)
4775 struct task_struct *tsk;
4778 struct cgroupfs_root *root;
4781 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4787 tsk = get_pid_task(pid, PIDTYPE_PID);
4793 mutex_lock(&cgroup_mutex);
4795 for_each_active_root(root) {
4796 struct cgroup_subsys *ss;
4797 struct cgroup *cgrp;
4800 seq_printf(m, "%d:", root->hierarchy_id);
4801 for_each_subsys(root, ss)
4802 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4803 if (strlen(root->name))
4804 seq_printf(m, "%sname=%s", count ? "," : "",
4807 cgrp = task_cgroup_from_root(tsk, root);
4808 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4816 mutex_unlock(&cgroup_mutex);
4817 put_task_struct(tsk);
4824 /* Display information about each subsystem and each hierarchy */
4825 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4829 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4831 * ideally we don't want subsystems moving around while we do this.
4832 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4833 * subsys/hierarchy state.
4835 mutex_lock(&cgroup_mutex);
4836 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4837 struct cgroup_subsys *ss = subsys[i];
4840 seq_printf(m, "%s\t%d\t%d\t%d\n",
4841 ss->name, ss->root->hierarchy_id,
4842 ss->root->number_of_cgroups, !ss->disabled);
4844 mutex_unlock(&cgroup_mutex);
4848 static int cgroupstats_open(struct inode *inode, struct file *file)
4850 return single_open(file, proc_cgroupstats_show, NULL);
4853 static const struct file_operations proc_cgroupstats_operations = {
4854 .open = cgroupstats_open,
4856 .llseek = seq_lseek,
4857 .release = single_release,
4861 * cgroup_fork - attach newly forked task to its parents cgroup.
4862 * @child: pointer to task_struct of forking parent process.
4864 * Description: A task inherits its parent's cgroup at fork().
4866 * A pointer to the shared css_set was automatically copied in
4867 * fork.c by dup_task_struct(). However, we ignore that copy, since
4868 * it was not made under the protection of RCU or cgroup_mutex, so
4869 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4870 * have already changed current->cgroups, allowing the previously
4871 * referenced cgroup group to be removed and freed.
4873 * At the point that cgroup_fork() is called, 'current' is the parent
4874 * task, and the passed argument 'child' points to the child task.
4876 void cgroup_fork(struct task_struct *child)
4879 child->cgroups = current->cgroups;
4880 get_css_set(child->cgroups);
4881 task_unlock(current);
4882 INIT_LIST_HEAD(&child->cg_list);
4886 * cgroup_post_fork - called on a new task after adding it to the task list
4887 * @child: the task in question
4889 * Adds the task to the list running through its css_set if necessary and
4890 * call the subsystem fork() callbacks. Has to be after the task is
4891 * visible on the task list in case we race with the first call to
4892 * cgroup_iter_start() - to guarantee that the new task ends up on its
4895 void cgroup_post_fork(struct task_struct *child)
4900 * use_task_css_set_links is set to 1 before we walk the tasklist
4901 * under the tasklist_lock and we read it here after we added the child
4902 * to the tasklist under the tasklist_lock as well. If the child wasn't
4903 * yet in the tasklist when we walked through it from
4904 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4905 * should be visible now due to the paired locking and barriers implied
4906 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4907 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4910 if (use_task_css_set_links) {
4911 write_lock(&css_set_lock);
4913 if (list_empty(&child->cg_list))
4914 list_add(&child->cg_list, &child->cgroups->tasks);
4916 write_unlock(&css_set_lock);
4920 * Call ss->fork(). This must happen after @child is linked on
4921 * css_set; otherwise, @child might change state between ->fork()
4922 * and addition to css_set.
4924 if (need_forkexit_callback) {
4926 * fork/exit callbacks are supported only for builtin
4927 * subsystems, and the builtin section of the subsys
4928 * array is immutable, so we don't need to lock the
4929 * subsys array here. On the other hand, modular section
4930 * of the array can be freed at module unload, so we
4933 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4934 struct cgroup_subsys *ss = subsys[i];
4943 * cgroup_exit - detach cgroup from exiting task
4944 * @tsk: pointer to task_struct of exiting process
4945 * @run_callback: run exit callbacks?
4947 * Description: Detach cgroup from @tsk and release it.
4949 * Note that cgroups marked notify_on_release force every task in
4950 * them to take the global cgroup_mutex mutex when exiting.
4951 * This could impact scaling on very large systems. Be reluctant to
4952 * use notify_on_release cgroups where very high task exit scaling
4953 * is required on large systems.
4955 * the_top_cgroup_hack:
4957 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4959 * We call cgroup_exit() while the task is still competent to
4960 * handle notify_on_release(), then leave the task attached to the
4961 * root cgroup in each hierarchy for the remainder of its exit.
4963 * To do this properly, we would increment the reference count on
4964 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4965 * code we would add a second cgroup function call, to drop that
4966 * reference. This would just create an unnecessary hot spot on
4967 * the top_cgroup reference count, to no avail.
4969 * Normally, holding a reference to a cgroup without bumping its
4970 * count is unsafe. The cgroup could go away, or someone could
4971 * attach us to a different cgroup, decrementing the count on
4972 * the first cgroup that we never incremented. But in this case,
4973 * top_cgroup isn't going away, and either task has PF_EXITING set,
4974 * which wards off any cgroup_attach_task() attempts, or task is a failed
4975 * fork, never visible to cgroup_attach_task.
4977 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4983 * Unlink from the css_set task list if necessary.
4984 * Optimistically check cg_list before taking
4987 if (!list_empty(&tsk->cg_list)) {
4988 write_lock(&css_set_lock);
4989 if (!list_empty(&tsk->cg_list))
4990 list_del_init(&tsk->cg_list);
4991 write_unlock(&css_set_lock);
4994 /* Reassign the task to the init_css_set. */
4997 tsk->cgroups = &init_css_set;
4999 if (run_callbacks && need_forkexit_callback) {
5001 * fork/exit callbacks are supported only for builtin
5002 * subsystems, see cgroup_post_fork() for details.
5004 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
5005 struct cgroup_subsys *ss = subsys[i];
5008 struct cgroup *old_cgrp =
5009 rcu_dereference_raw(cg->subsys[i])->cgroup;
5010 struct cgroup *cgrp = task_cgroup(tsk, i);
5011 ss->exit(cgrp, old_cgrp, tsk);
5017 put_css_set_taskexit(cg);
5020 static void check_for_release(struct cgroup *cgrp)
5022 /* All of these checks rely on RCU to keep the cgroup
5023 * structure alive */
5024 if (cgroup_is_releasable(cgrp) &&
5025 !atomic_read(&cgrp->count) && list_empty(&cgrp->children)) {
5027 * Control Group is currently removeable. If it's not
5028 * already queued for a userspace notification, queue
5031 int need_schedule_work = 0;
5033 raw_spin_lock(&release_list_lock);
5034 if (!cgroup_is_removed(cgrp) &&
5035 list_empty(&cgrp->release_list)) {
5036 list_add(&cgrp->release_list, &release_list);
5037 need_schedule_work = 1;
5039 raw_spin_unlock(&release_list_lock);
5040 if (need_schedule_work)
5041 schedule_work(&release_agent_work);
5045 /* Caller must verify that the css is not for root cgroup */
5046 bool __css_tryget(struct cgroup_subsys_state *css)
5051 v = css_refcnt(css);
5052 t = atomic_cmpxchg(&css->refcnt, v, v + 1);
5060 EXPORT_SYMBOL_GPL(__css_tryget);
5062 /* Caller must verify that the css is not for root cgroup */
5063 void __css_put(struct cgroup_subsys_state *css)
5067 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
5069 queue_work(cgroup_destroy_wq, &css->dput_work);
5071 EXPORT_SYMBOL_GPL(__css_put);
5074 * Notify userspace when a cgroup is released, by running the
5075 * configured release agent with the name of the cgroup (path
5076 * relative to the root of cgroup file system) as the argument.
5078 * Most likely, this user command will try to rmdir this cgroup.
5080 * This races with the possibility that some other task will be
5081 * attached to this cgroup before it is removed, or that some other
5082 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5083 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5084 * unused, and this cgroup will be reprieved from its death sentence,
5085 * to continue to serve a useful existence. Next time it's released,
5086 * we will get notified again, if it still has 'notify_on_release' set.
5088 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5089 * means only wait until the task is successfully execve()'d. The
5090 * separate release agent task is forked by call_usermodehelper(),
5091 * then control in this thread returns here, without waiting for the
5092 * release agent task. We don't bother to wait because the caller of
5093 * this routine has no use for the exit status of the release agent
5094 * task, so no sense holding our caller up for that.
5096 static void cgroup_release_agent(struct work_struct *work)
5098 BUG_ON(work != &release_agent_work);
5099 mutex_lock(&cgroup_mutex);
5100 raw_spin_lock(&release_list_lock);
5101 while (!list_empty(&release_list)) {
5102 char *argv[3], *envp[3];
5104 char *pathbuf = NULL, *agentbuf = NULL;
5105 struct cgroup *cgrp = list_entry(release_list.next,
5108 list_del_init(&cgrp->release_list);
5109 raw_spin_unlock(&release_list_lock);
5110 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5113 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5115 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5120 argv[i++] = agentbuf;
5121 argv[i++] = pathbuf;
5125 /* minimal command environment */
5126 envp[i++] = "HOME=/";
5127 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5130 /* Drop the lock while we invoke the usermode helper,
5131 * since the exec could involve hitting disk and hence
5132 * be a slow process */
5133 mutex_unlock(&cgroup_mutex);
5134 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5135 mutex_lock(&cgroup_mutex);
5139 raw_spin_lock(&release_list_lock);
5141 raw_spin_unlock(&release_list_lock);
5142 mutex_unlock(&cgroup_mutex);
5145 static int __init cgroup_disable(char *str)
5150 while ((token = strsep(&str, ",")) != NULL) {
5153 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5154 struct cgroup_subsys *ss = subsys[i];
5157 * cgroup_disable, being at boot time, can't
5158 * know about module subsystems, so we don't
5161 if (!ss || ss->module)
5164 if (!strcmp(token, ss->name)) {
5166 printk(KERN_INFO "Disabling %s control group"
5167 " subsystem\n", ss->name);
5174 __setup("cgroup_disable=", cgroup_disable);
5177 * Functons for CSS ID.
5181 *To get ID other than 0, this should be called when !cgroup_is_removed().
5183 unsigned short css_id(struct cgroup_subsys_state *css)
5185 struct css_id *cssid;
5188 * This css_id() can return correct value when somone has refcnt
5189 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5190 * it's unchanged until freed.
5192 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5198 EXPORT_SYMBOL_GPL(css_id);
5200 unsigned short css_depth(struct cgroup_subsys_state *css)
5202 struct css_id *cssid;
5204 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5207 return cssid->depth;
5210 EXPORT_SYMBOL_GPL(css_depth);
5213 * css_is_ancestor - test "root" css is an ancestor of "child"
5214 * @child: the css to be tested.
5215 * @root: the css supporsed to be an ancestor of the child.
5217 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5218 * this function reads css->id, the caller must hold rcu_read_lock().
5219 * But, considering usual usage, the csses should be valid objects after test.
5220 * Assuming that the caller will do some action to the child if this returns
5221 * returns true, the caller must take "child";s reference count.
5222 * If "child" is valid object and this returns true, "root" is valid, too.
5225 bool css_is_ancestor(struct cgroup_subsys_state *child,
5226 const struct cgroup_subsys_state *root)
5228 struct css_id *child_id;
5229 struct css_id *root_id;
5231 child_id = rcu_dereference(child->id);
5234 root_id = rcu_dereference(root->id);
5237 if (child_id->depth < root_id->depth)
5239 if (child_id->stack[root_id->depth] != root_id->id)
5244 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5246 struct css_id *id = css->id;
5247 /* When this is called before css_id initialization, id can be NULL */
5251 BUG_ON(!ss->use_id);
5253 rcu_assign_pointer(id->css, NULL);
5254 rcu_assign_pointer(css->id, NULL);
5255 spin_lock(&ss->id_lock);
5256 idr_remove(&ss->idr, id->id);
5257 spin_unlock(&ss->id_lock);
5258 kfree_rcu(id, rcu_head);
5260 EXPORT_SYMBOL_GPL(free_css_id);
5263 * This is called by init or create(). Then, calls to this function are
5264 * always serialized (By cgroup_mutex() at create()).
5267 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5269 struct css_id *newid;
5272 BUG_ON(!ss->use_id);
5274 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5275 newid = kzalloc(size, GFP_KERNEL);
5277 return ERR_PTR(-ENOMEM);
5279 idr_preload(GFP_KERNEL);
5280 spin_lock(&ss->id_lock);
5281 /* Don't use 0. allocates an ID of 1-65535 */
5282 ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5283 spin_unlock(&ss->id_lock);
5286 /* Returns error when there are no free spaces for new ID.*/
5291 newid->depth = depth;
5295 return ERR_PTR(ret);
5299 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5300 struct cgroup_subsys_state *rootcss)
5302 struct css_id *newid;
5304 spin_lock_init(&ss->id_lock);
5307 newid = get_new_cssid(ss, 0);
5309 return PTR_ERR(newid);
5311 newid->stack[0] = newid->id;
5312 newid->css = rootcss;
5313 rootcss->id = newid;
5317 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5318 struct cgroup *child)
5320 int subsys_id, i, depth = 0;
5321 struct cgroup_subsys_state *parent_css, *child_css;
5322 struct css_id *child_id, *parent_id;
5324 subsys_id = ss->subsys_id;
5325 parent_css = parent->subsys[subsys_id];
5326 child_css = child->subsys[subsys_id];
5327 parent_id = parent_css->id;
5328 depth = parent_id->depth + 1;
5330 child_id = get_new_cssid(ss, depth);
5331 if (IS_ERR(child_id))
5332 return PTR_ERR(child_id);
5334 for (i = 0; i < depth; i++)
5335 child_id->stack[i] = parent_id->stack[i];
5336 child_id->stack[depth] = child_id->id;
5338 * child_id->css pointer will be set after this cgroup is available
5339 * see cgroup_populate_dir()
5341 rcu_assign_pointer(child_css->id, child_id);
5347 * css_lookup - lookup css by id
5348 * @ss: cgroup subsys to be looked into.
5351 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5352 * NULL if not. Should be called under rcu_read_lock()
5354 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5356 struct css_id *cssid = NULL;
5358 BUG_ON(!ss->use_id);
5359 cssid = idr_find(&ss->idr, id);
5361 if (unlikely(!cssid))
5364 return rcu_dereference(cssid->css);
5366 EXPORT_SYMBOL_GPL(css_lookup);
5369 * get corresponding css from file open on cgroupfs directory
5371 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5373 struct cgroup *cgrp;
5374 struct inode *inode;
5375 struct cgroup_subsys_state *css;
5377 inode = file_inode(f);
5378 /* check in cgroup filesystem dir */
5379 if (inode->i_op != &cgroup_dir_inode_operations)
5380 return ERR_PTR(-EBADF);
5382 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5383 return ERR_PTR(-EINVAL);
5386 cgrp = __d_cgrp(f->f_dentry);
5387 css = cgrp->subsys[id];
5388 return css ? css : ERR_PTR(-ENOENT);
5391 #ifdef CONFIG_CGROUP_DEBUG
5392 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5394 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5397 return ERR_PTR(-ENOMEM);
5402 static void debug_css_free(struct cgroup *cont)
5404 kfree(cont->subsys[debug_subsys_id]);
5407 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5409 return atomic_read(&cont->count);
5412 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5414 return cgroup_task_count(cont);
5417 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5419 return (u64)(unsigned long)current->cgroups;
5422 static u64 current_css_set_refcount_read(struct cgroup *cont,
5428 count = atomic_read(¤t->cgroups->refcount);
5433 static int current_css_set_cg_links_read(struct cgroup *cont,
5435 struct seq_file *seq)
5437 struct cg_cgroup_link *link;
5440 read_lock(&css_set_lock);
5442 cg = rcu_dereference(current->cgroups);
5443 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5444 struct cgroup *c = link->cgrp;
5448 name = c->dentry->d_name.name;
5451 seq_printf(seq, "Root %d group %s\n",
5452 c->root->hierarchy_id, name);
5455 read_unlock(&css_set_lock);
5459 #define MAX_TASKS_SHOWN_PER_CSS 25
5460 static int cgroup_css_links_read(struct cgroup *cont,
5462 struct seq_file *seq)
5464 struct cg_cgroup_link *link;
5466 read_lock(&css_set_lock);
5467 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5468 struct css_set *cg = link->cg;
5469 struct task_struct *task;
5471 seq_printf(seq, "css_set %p\n", cg);
5472 list_for_each_entry(task, &cg->tasks, cg_list) {
5473 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5474 seq_puts(seq, " ...\n");
5477 seq_printf(seq, " task %d\n",
5478 task_pid_vnr(task));
5482 read_unlock(&css_set_lock);
5486 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5488 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5491 static struct cftype debug_files[] = {
5493 .name = "cgroup_refcount",
5494 .read_u64 = cgroup_refcount_read,
5497 .name = "taskcount",
5498 .read_u64 = debug_taskcount_read,
5502 .name = "current_css_set",
5503 .read_u64 = current_css_set_read,
5507 .name = "current_css_set_refcount",
5508 .read_u64 = current_css_set_refcount_read,
5512 .name = "current_css_set_cg_links",
5513 .read_seq_string = current_css_set_cg_links_read,
5517 .name = "cgroup_css_links",
5518 .read_seq_string = cgroup_css_links_read,
5522 .name = "releasable",
5523 .read_u64 = releasable_read,
5529 struct cgroup_subsys debug_subsys = {
5531 .css_alloc = debug_css_alloc,
5532 .css_free = debug_css_free,
5533 .subsys_id = debug_subsys_id,
5534 .base_cftypes = debug_files,
5536 #endif /* CONFIG_CGROUP_DEBUG */