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>
67 * cgroup_mutex is the master lock. Any modification to cgroup or its
68 * hierarchy must be performed while holding it.
70 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
71 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
72 * release_agent_path and so on. Modifying requires both cgroup_mutex and
73 * cgroup_root_mutex. Readers can acquire either of the two. This is to
74 * break the following locking order cycle.
76 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
77 * B. namespace_sem -> cgroup_mutex
79 * B happens only through cgroup_show_options() and using cgroup_root_mutex
82 #ifdef CONFIG_PROVE_RCU
83 DEFINE_MUTEX(cgroup_mutex);
84 EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for task_subsys_state_check() */
86 static DEFINE_MUTEX(cgroup_mutex);
89 static DEFINE_MUTEX(cgroup_root_mutex);
92 * Generate an array of cgroup subsystem pointers. At boot time, this is
93 * populated with the built in subsystems, and modular subsystems are
94 * registered after that. The mutable section of this array is protected by
97 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
98 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
99 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
100 #include <linux/cgroup_subsys.h>
104 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
105 * subsystems that are otherwise unattached - it never has more than a
106 * single cgroup, and all tasks are part of that cgroup.
108 static struct cgroupfs_root rootnode;
111 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
114 struct list_head node;
115 struct dentry *dentry;
119 struct simple_xattrs xattrs;
123 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
124 * cgroup_subsys->use_id != 0.
126 #define CSS_ID_MAX (65535)
129 * The css to which this ID points. This pointer is set to valid value
130 * after cgroup is populated. If cgroup is removed, this will be NULL.
131 * This pointer is expected to be RCU-safe because destroy()
132 * is called after synchronize_rcu(). But for safe use, css_tryget()
133 * should be used for avoiding race.
135 struct cgroup_subsys_state __rcu *css;
141 * Depth in hierarchy which this ID belongs to.
143 unsigned short depth;
145 * ID is freed by RCU. (and lookup routine is RCU safe.)
147 struct rcu_head rcu_head;
149 * Hierarchy of CSS ID belongs to.
151 unsigned short stack[0]; /* Array of Length (depth+1) */
155 * cgroup_event represents events which userspace want to receive.
157 struct cgroup_event {
159 * Cgroup which the event belongs to.
163 * Control file which the event associated.
167 * eventfd to signal userspace about the event.
169 struct eventfd_ctx *eventfd;
171 * Each of these stored in a list by the cgroup.
173 struct list_head list;
175 * All fields below needed to unregister event when
176 * userspace closes eventfd.
179 wait_queue_head_t *wqh;
181 struct work_struct remove;
184 /* The list of hierarchy roots */
186 static LIST_HEAD(roots);
187 static int root_count;
190 * Hierarchy ID allocation and mapping. It follows the same exclusion
191 * rules as other root ops - both cgroup_mutex and cgroup_root_mutex for
192 * writes, either for reads.
194 static DEFINE_IDR(cgroup_hierarchy_idr);
196 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
197 #define dummytop (&rootnode.top_cgroup)
199 static struct cgroup_name root_cgroup_name = { .name = "/" };
202 * Assign a monotonically increasing serial number to cgroups. It
203 * guarantees cgroups with bigger numbers are newer than those with smaller
204 * numbers. Also, as cgroups are always appended to the parent's
205 * ->children list, it guarantees that sibling cgroups are always sorted in
206 * the ascending serial number order on the list.
208 static atomic64_t cgroup_serial_nr_cursor = ATOMIC64_INIT(0);
210 /* This flag indicates whether tasks in the fork and exit paths should
211 * check for fork/exit handlers to call. This avoids us having to do
212 * extra work in the fork/exit path if none of the subsystems need to
215 static int need_forkexit_callback __read_mostly;
217 static void cgroup_offline_fn(struct work_struct *work);
218 static int cgroup_destroy_locked(struct cgroup *cgrp);
219 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
220 struct cftype cfts[], bool is_add);
222 /* convenient tests for these bits */
223 static inline bool cgroup_is_dead(const struct cgroup *cgrp)
225 return test_bit(CGRP_DEAD, &cgrp->flags);
229 * cgroup_is_descendant - test ancestry
230 * @cgrp: the cgroup to be tested
231 * @ancestor: possible ancestor of @cgrp
233 * Test whether @cgrp is a descendant of @ancestor. It also returns %true
234 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp
235 * and @ancestor are accessible.
237 bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor)
240 if (cgrp == ancestor)
246 EXPORT_SYMBOL_GPL(cgroup_is_descendant);
248 static int cgroup_is_releasable(const struct cgroup *cgrp)
251 (1 << CGRP_RELEASABLE) |
252 (1 << CGRP_NOTIFY_ON_RELEASE);
253 return (cgrp->flags & bits) == bits;
256 static int notify_on_release(const struct cgroup *cgrp)
258 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
262 * for_each_subsys() allows you to iterate on each subsystem attached to
263 * an active hierarchy
265 #define for_each_subsys(_root, _ss) \
266 list_for_each_entry(_ss, &_root->subsys_list, sibling)
268 /* for_each_active_root() allows you to iterate across the active hierarchies */
269 #define for_each_active_root(_root) \
270 list_for_each_entry(_root, &roots, root_list)
272 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
274 return dentry->d_fsdata;
277 static inline struct cfent *__d_cfe(struct dentry *dentry)
279 return dentry->d_fsdata;
282 static inline struct cftype *__d_cft(struct dentry *dentry)
284 return __d_cfe(dentry)->type;
288 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
289 * @cgrp: the cgroup to be checked for liveness
291 * On success, returns true; the mutex should be later unlocked. On
292 * failure returns false with no lock held.
294 static bool cgroup_lock_live_group(struct cgroup *cgrp)
296 mutex_lock(&cgroup_mutex);
297 if (cgroup_is_dead(cgrp)) {
298 mutex_unlock(&cgroup_mutex);
304 /* the list of cgroups eligible for automatic release. Protected by
305 * release_list_lock */
306 static LIST_HEAD(release_list);
307 static DEFINE_RAW_SPINLOCK(release_list_lock);
308 static void cgroup_release_agent(struct work_struct *work);
309 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
310 static void check_for_release(struct cgroup *cgrp);
313 * A cgroup can be associated with multiple css_sets as different tasks may
314 * belong to different cgroups on different hierarchies. In the other
315 * direction, a css_set is naturally associated with multiple cgroups.
316 * This M:N relationship is represented by the following link structure
317 * which exists for each association and allows traversing the associations
320 struct cgrp_cset_link {
321 /* the cgroup and css_set this link associates */
323 struct css_set *cset;
325 /* list of cgrp_cset_links anchored at cgrp->cset_links */
326 struct list_head cset_link;
328 /* list of cgrp_cset_links anchored at css_set->cgrp_links */
329 struct list_head cgrp_link;
332 /* The default css_set - used by init and its children prior to any
333 * hierarchies being mounted. It contains a pointer to the root state
334 * for each subsystem. Also used to anchor the list of css_sets. Not
335 * reference-counted, to improve performance when child cgroups
336 * haven't been created.
339 static struct css_set init_css_set;
340 static struct cgrp_cset_link init_cgrp_cset_link;
342 static int cgroup_init_idr(struct cgroup_subsys *ss,
343 struct cgroup_subsys_state *css);
345 /* css_set_lock protects the list of css_set objects, and the
346 * chain of tasks off each css_set. Nests outside task->alloc_lock
347 * due to cgroup_iter_start() */
348 static DEFINE_RWLOCK(css_set_lock);
349 static int css_set_count;
352 * hash table for cgroup groups. This improves the performance to find
353 * an existing css_set. This hash doesn't (currently) take into
354 * account cgroups in empty hierarchies.
356 #define CSS_SET_HASH_BITS 7
357 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
359 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
362 unsigned long key = 0UL;
364 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
365 key += (unsigned long)css[i];
366 key = (key >> 16) ^ key;
371 /* We don't maintain the lists running through each css_set to its
372 * task until after the first call to cgroup_iter_start(). This
373 * reduces the fork()/exit() overhead for people who have cgroups
374 * compiled into their kernel but not actually in use */
375 static int use_task_css_set_links __read_mostly;
377 static void __put_css_set(struct css_set *cset, int taskexit)
379 struct cgrp_cset_link *link, *tmp_link;
382 * Ensure that the refcount doesn't hit zero while any readers
383 * can see it. Similar to atomic_dec_and_lock(), but for an
386 if (atomic_add_unless(&cset->refcount, -1, 1))
388 write_lock(&css_set_lock);
389 if (!atomic_dec_and_test(&cset->refcount)) {
390 write_unlock(&css_set_lock);
394 /* This css_set is dead. unlink it and release cgroup refcounts */
395 hash_del(&cset->hlist);
398 list_for_each_entry_safe(link, tmp_link, &cset->cgrp_links, cgrp_link) {
399 struct cgroup *cgrp = link->cgrp;
401 list_del(&link->cset_link);
402 list_del(&link->cgrp_link);
404 /* @cgrp can't go away while we're holding css_set_lock */
405 if (list_empty(&cgrp->cset_links) && notify_on_release(cgrp)) {
407 set_bit(CGRP_RELEASABLE, &cgrp->flags);
408 check_for_release(cgrp);
414 write_unlock(&css_set_lock);
415 kfree_rcu(cset, rcu_head);
419 * refcounted get/put for css_set objects
421 static inline void get_css_set(struct css_set *cset)
423 atomic_inc(&cset->refcount);
426 static inline void put_css_set(struct css_set *cset)
428 __put_css_set(cset, 0);
431 static inline void put_css_set_taskexit(struct css_set *cset)
433 __put_css_set(cset, 1);
437 * compare_css_sets - helper function for find_existing_css_set().
438 * @cset: candidate css_set being tested
439 * @old_cset: existing css_set for a task
440 * @new_cgrp: cgroup that's being entered by the task
441 * @template: desired set of css pointers in css_set (pre-calculated)
443 * Returns true if "cg" matches "old_cg" except for the hierarchy
444 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
446 static bool compare_css_sets(struct css_set *cset,
447 struct css_set *old_cset,
448 struct cgroup *new_cgrp,
449 struct cgroup_subsys_state *template[])
451 struct list_head *l1, *l2;
453 if (memcmp(template, cset->subsys, sizeof(cset->subsys))) {
454 /* Not all subsystems matched */
459 * Compare cgroup pointers in order to distinguish between
460 * different cgroups in heirarchies with no subsystems. We
461 * could get by with just this check alone (and skip the
462 * memcmp above) but on most setups the memcmp check will
463 * avoid the need for this more expensive check on almost all
467 l1 = &cset->cgrp_links;
468 l2 = &old_cset->cgrp_links;
470 struct cgrp_cset_link *link1, *link2;
471 struct cgroup *cgrp1, *cgrp2;
475 /* See if we reached the end - both lists are equal length. */
476 if (l1 == &cset->cgrp_links) {
477 BUG_ON(l2 != &old_cset->cgrp_links);
480 BUG_ON(l2 == &old_cset->cgrp_links);
482 /* Locate the cgroups associated with these links. */
483 link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
484 link2 = list_entry(l2, struct cgrp_cset_link, cgrp_link);
487 /* Hierarchies should be linked in the same order. */
488 BUG_ON(cgrp1->root != cgrp2->root);
491 * If this hierarchy is the hierarchy of the cgroup
492 * that's changing, then we need to check that this
493 * css_set points to the new cgroup; if it's any other
494 * hierarchy, then this css_set should point to the
495 * same cgroup as the old css_set.
497 if (cgrp1->root == new_cgrp->root) {
498 if (cgrp1 != new_cgrp)
509 * find_existing_css_set() is a helper for
510 * find_css_set(), and checks to see whether an existing
511 * css_set is suitable.
513 * oldcg: the cgroup group that we're using before the cgroup
516 * cgrp: the cgroup that we're moving into
518 * template: location in which to build the desired set of subsystem
519 * state objects for the new cgroup group
521 static struct css_set *find_existing_css_set(struct css_set *old_cset,
523 struct cgroup_subsys_state *template[])
526 struct cgroupfs_root *root = cgrp->root;
527 struct css_set *cset;
531 * Build the set of subsystem state objects that we want to see in the
532 * new css_set. while subsystems can change globally, the entries here
533 * won't change, so no need for locking.
535 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
536 if (root->subsys_mask & (1UL << i)) {
537 /* Subsystem is in this hierarchy. So we want
538 * the subsystem state from the new
540 template[i] = cgrp->subsys[i];
542 /* Subsystem is not in this hierarchy, so we
543 * don't want to change the subsystem state */
544 template[i] = old_cset->subsys[i];
548 key = css_set_hash(template);
549 hash_for_each_possible(css_set_table, cset, hlist, key) {
550 if (!compare_css_sets(cset, old_cset, cgrp, template))
553 /* This css_set matches what we need */
557 /* No existing cgroup group matched */
561 static void free_cgrp_cset_links(struct list_head *links_to_free)
563 struct cgrp_cset_link *link, *tmp_link;
565 list_for_each_entry_safe(link, tmp_link, links_to_free, cset_link) {
566 list_del(&link->cset_link);
572 * allocate_cgrp_cset_links - allocate cgrp_cset_links
573 * @count: the number of links to allocate
574 * @tmp_links: list_head the allocated links are put on
576 * Allocate @count cgrp_cset_link structures and chain them on @tmp_links
577 * through ->cset_link. Returns 0 on success or -errno.
579 static int allocate_cgrp_cset_links(int count, struct list_head *tmp_links)
581 struct cgrp_cset_link *link;
584 INIT_LIST_HEAD(tmp_links);
586 for (i = 0; i < count; i++) {
587 link = kzalloc(sizeof(*link), GFP_KERNEL);
589 free_cgrp_cset_links(tmp_links);
592 list_add(&link->cset_link, tmp_links);
598 * link_css_set - a helper function to link a css_set to a cgroup
599 * @tmp_links: cgrp_cset_link objects allocated by allocate_cgrp_cset_links()
600 * @cset: the css_set to be linked
601 * @cgrp: the destination cgroup
603 static void link_css_set(struct list_head *tmp_links, struct css_set *cset,
606 struct cgrp_cset_link *link;
608 BUG_ON(list_empty(tmp_links));
609 link = list_first_entry(tmp_links, struct cgrp_cset_link, cset_link);
612 list_move(&link->cset_link, &cgrp->cset_links);
614 * Always add links to the tail of the list so that the list
615 * is sorted by order of hierarchy creation
617 list_add_tail(&link->cgrp_link, &cset->cgrp_links);
621 * find_css_set() takes an existing cgroup group and a
622 * cgroup object, and returns a css_set object that's
623 * equivalent to the old group, but with the given cgroup
624 * substituted into the appropriate hierarchy. Must be called with
627 static struct css_set *find_css_set(struct css_set *old_cset,
630 struct css_set *cset;
631 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
632 struct list_head tmp_links;
633 struct cgrp_cset_link *link;
636 /* First see if we already have a cgroup group that matches
638 read_lock(&css_set_lock);
639 cset = find_existing_css_set(old_cset, cgrp, template);
642 read_unlock(&css_set_lock);
647 cset = kzalloc(sizeof(*cset), GFP_KERNEL);
651 /* Allocate all the cgrp_cset_link objects that we'll need */
652 if (allocate_cgrp_cset_links(root_count, &tmp_links) < 0) {
657 atomic_set(&cset->refcount, 1);
658 INIT_LIST_HEAD(&cset->cgrp_links);
659 INIT_LIST_HEAD(&cset->tasks);
660 INIT_HLIST_NODE(&cset->hlist);
662 /* Copy the set of subsystem state objects generated in
663 * find_existing_css_set() */
664 memcpy(cset->subsys, template, sizeof(cset->subsys));
666 write_lock(&css_set_lock);
667 /* Add reference counts and links from the new css_set. */
668 list_for_each_entry(link, &old_cset->cgrp_links, cgrp_link) {
669 struct cgroup *c = link->cgrp;
671 if (c->root == cgrp->root)
673 link_css_set(&tmp_links, cset, c);
676 BUG_ON(!list_empty(&tmp_links));
680 /* Add this cgroup group to the hash table */
681 key = css_set_hash(cset->subsys);
682 hash_add(css_set_table, &cset->hlist, key);
684 write_unlock(&css_set_lock);
690 * Return the cgroup for "task" from the given hierarchy. Must be
691 * called with cgroup_mutex held.
693 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
694 struct cgroupfs_root *root)
696 struct css_set *cset;
697 struct cgroup *res = NULL;
699 BUG_ON(!mutex_is_locked(&cgroup_mutex));
700 read_lock(&css_set_lock);
702 * No need to lock the task - since we hold cgroup_mutex the
703 * task can't change groups, so the only thing that can happen
704 * is that it exits and its css is set back to init_css_set.
706 cset = task->cgroups;
707 if (cset == &init_css_set) {
708 res = &root->top_cgroup;
710 struct cgrp_cset_link *link;
712 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
713 struct cgroup *c = link->cgrp;
715 if (c->root == root) {
721 read_unlock(&css_set_lock);
727 * There is one global cgroup mutex. We also require taking
728 * task_lock() when dereferencing a task's cgroup subsys pointers.
729 * See "The task_lock() exception", at the end of this comment.
731 * A task must hold cgroup_mutex to modify cgroups.
733 * Any task can increment and decrement the count field without lock.
734 * So in general, code holding cgroup_mutex can't rely on the count
735 * field not changing. However, if the count goes to zero, then only
736 * cgroup_attach_task() can increment it again. Because a count of zero
737 * means that no tasks are currently attached, therefore there is no
738 * way a task attached to that cgroup can fork (the other way to
739 * increment the count). So code holding cgroup_mutex can safely
740 * assume that if the count is zero, it will stay zero. Similarly, if
741 * a task holds cgroup_mutex on a cgroup with zero count, it
742 * knows that the cgroup won't be removed, as cgroup_rmdir()
745 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
746 * (usually) take cgroup_mutex. These are the two most performance
747 * critical pieces of code here. The exception occurs on cgroup_exit(),
748 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
749 * is taken, and if the cgroup count is zero, a usermode call made
750 * to the release agent with the name of the cgroup (path relative to
751 * the root of cgroup file system) as the argument.
753 * A cgroup can only be deleted if both its 'count' of using tasks
754 * is zero, and its list of 'children' cgroups is empty. Since all
755 * tasks in the system use _some_ cgroup, and since there is always at
756 * least one task in the system (init, pid == 1), therefore, top_cgroup
757 * always has either children cgroups and/or using tasks. So we don't
758 * need a special hack to ensure that top_cgroup cannot be deleted.
760 * The task_lock() exception
762 * The need for this exception arises from the action of
763 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
764 * another. It does so using cgroup_mutex, however there are
765 * several performance critical places that need to reference
766 * task->cgroup without the expense of grabbing a system global
767 * mutex. Therefore except as noted below, when dereferencing or, as
768 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
769 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
770 * the task_struct routinely used for such matters.
772 * P.S. One more locking exception. RCU is used to guard the
773 * update of a tasks cgroup pointer by cgroup_attach_task()
777 * A couple of forward declarations required, due to cyclic reference loop:
778 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
779 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
783 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
784 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
785 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
786 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
787 unsigned long subsys_mask);
788 static const struct inode_operations cgroup_dir_inode_operations;
789 static const struct file_operations proc_cgroupstats_operations;
791 static struct backing_dev_info cgroup_backing_dev_info = {
793 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
796 static int alloc_css_id(struct cgroup_subsys *ss,
797 struct cgroup *parent, struct cgroup *child);
799 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
801 struct inode *inode = new_inode(sb);
804 inode->i_ino = get_next_ino();
805 inode->i_mode = mode;
806 inode->i_uid = current_fsuid();
807 inode->i_gid = current_fsgid();
808 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
809 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
814 static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
816 struct cgroup_name *name;
818 name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL);
821 strcpy(name->name, dentry->d_name.name);
825 static void cgroup_free_fn(struct work_struct *work)
827 struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work);
828 struct cgroup_subsys *ss;
830 mutex_lock(&cgroup_mutex);
832 * Release the subsystem state objects.
834 for_each_subsys(cgrp->root, ss)
837 cgrp->root->number_of_cgroups--;
838 mutex_unlock(&cgroup_mutex);
841 * We get a ref to the parent's dentry, and put the ref when
842 * this cgroup is being freed, so it's guaranteed that the
843 * parent won't be destroyed before its children.
845 dput(cgrp->parent->dentry);
847 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
850 * Drop the active superblock reference that we took when we
851 * created the cgroup. This will free cgrp->root, if we are
852 * holding the last reference to @sb.
854 deactivate_super(cgrp->root->sb);
857 * if we're getting rid of the cgroup, refcount should ensure
858 * that there are no pidlists left.
860 BUG_ON(!list_empty(&cgrp->pidlists));
862 simple_xattrs_free(&cgrp->xattrs);
864 kfree(rcu_dereference_raw(cgrp->name));
868 static void cgroup_free_rcu(struct rcu_head *head)
870 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
872 INIT_WORK(&cgrp->destroy_work, cgroup_free_fn);
873 schedule_work(&cgrp->destroy_work);
876 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
878 /* is dentry a directory ? if so, kfree() associated cgroup */
879 if (S_ISDIR(inode->i_mode)) {
880 struct cgroup *cgrp = dentry->d_fsdata;
882 BUG_ON(!(cgroup_is_dead(cgrp)));
883 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
885 struct cfent *cfe = __d_cfe(dentry);
886 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
888 WARN_ONCE(!list_empty(&cfe->node) &&
889 cgrp != &cgrp->root->top_cgroup,
890 "cfe still linked for %s\n", cfe->type->name);
891 simple_xattrs_free(&cfe->xattrs);
897 static int cgroup_delete(const struct dentry *d)
902 static void remove_dir(struct dentry *d)
904 struct dentry *parent = dget(d->d_parent);
907 simple_rmdir(parent->d_inode, d);
911 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
915 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
916 lockdep_assert_held(&cgroup_mutex);
919 * If we're doing cleanup due to failure of cgroup_create(),
920 * the corresponding @cfe may not exist.
922 list_for_each_entry(cfe, &cgrp->files, node) {
923 struct dentry *d = cfe->dentry;
925 if (cft && cfe->type != cft)
930 simple_unlink(cgrp->dentry->d_inode, d);
931 list_del_init(&cfe->node);
939 * cgroup_clear_directory - selective removal of base and subsystem files
940 * @dir: directory containing the files
941 * @base_files: true if the base files should be removed
942 * @subsys_mask: mask of the subsystem ids whose files should be removed
944 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
945 unsigned long subsys_mask)
947 struct cgroup *cgrp = __d_cgrp(dir);
948 struct cgroup_subsys *ss;
950 for_each_subsys(cgrp->root, ss) {
951 struct cftype_set *set;
952 if (!test_bit(ss->subsys_id, &subsys_mask))
954 list_for_each_entry(set, &ss->cftsets, node)
955 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
958 while (!list_empty(&cgrp->files))
959 cgroup_rm_file(cgrp, NULL);
964 * NOTE : the dentry must have been dget()'ed
966 static void cgroup_d_remove_dir(struct dentry *dentry)
968 struct dentry *parent;
969 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
971 cgroup_clear_directory(dentry, true, root->subsys_mask);
973 parent = dentry->d_parent;
974 spin_lock(&parent->d_lock);
975 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
976 list_del_init(&dentry->d_u.d_child);
977 spin_unlock(&dentry->d_lock);
978 spin_unlock(&parent->d_lock);
983 * Call with cgroup_mutex held. Drops reference counts on modules, including
984 * any duplicate ones that parse_cgroupfs_options took. If this function
985 * returns an error, no reference counts are touched.
987 static int rebind_subsystems(struct cgroupfs_root *root,
988 unsigned long final_subsys_mask)
990 unsigned long added_mask, removed_mask;
991 struct cgroup *cgrp = &root->top_cgroup;
994 BUG_ON(!mutex_is_locked(&cgroup_mutex));
995 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
997 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
998 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
999 /* Check that any added subsystems are currently free */
1000 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1001 unsigned long bit = 1UL << i;
1002 struct cgroup_subsys *ss = subsys[i];
1003 if (!(bit & added_mask))
1006 * Nobody should tell us to do a subsys that doesn't exist:
1007 * parse_cgroupfs_options should catch that case and refcounts
1008 * ensure that subsystems won't disappear once selected.
1011 if (ss->root != &rootnode) {
1012 /* Subsystem isn't free */
1017 /* Currently we don't handle adding/removing subsystems when
1018 * any child cgroups exist. This is theoretically supportable
1019 * but involves complex error handling, so it's being left until
1021 if (root->number_of_cgroups > 1)
1024 /* Process each subsystem */
1025 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1026 struct cgroup_subsys *ss = subsys[i];
1027 unsigned long bit = 1UL << i;
1028 if (bit & added_mask) {
1029 /* We're binding this subsystem to this hierarchy */
1031 BUG_ON(cgrp->subsys[i]);
1032 BUG_ON(!dummytop->subsys[i]);
1033 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1034 cgrp->subsys[i] = dummytop->subsys[i];
1035 cgrp->subsys[i]->cgroup = cgrp;
1036 list_move(&ss->sibling, &root->subsys_list);
1040 /* refcount was already taken, and we're keeping it */
1041 } else if (bit & removed_mask) {
1042 /* We're removing this subsystem */
1044 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1045 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1048 dummytop->subsys[i]->cgroup = dummytop;
1049 cgrp->subsys[i] = NULL;
1050 subsys[i]->root = &rootnode;
1051 list_move(&ss->sibling, &rootnode.subsys_list);
1052 /* subsystem is now free - drop reference on module */
1053 module_put(ss->module);
1054 } else if (bit & final_subsys_mask) {
1055 /* Subsystem state should already exist */
1057 BUG_ON(!cgrp->subsys[i]);
1059 * a refcount was taken, but we already had one, so
1060 * drop the extra reference.
1062 module_put(ss->module);
1063 #ifdef CONFIG_MODULE_UNLOAD
1064 BUG_ON(ss->module && !module_refcount(ss->module));
1067 /* Subsystem state shouldn't exist */
1068 BUG_ON(cgrp->subsys[i]);
1071 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1076 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1078 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1079 struct cgroup_subsys *ss;
1081 mutex_lock(&cgroup_root_mutex);
1082 for_each_subsys(root, ss)
1083 seq_printf(seq, ",%s", ss->name);
1084 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR)
1085 seq_puts(seq, ",sane_behavior");
1086 if (root->flags & CGRP_ROOT_NOPREFIX)
1087 seq_puts(seq, ",noprefix");
1088 if (root->flags & CGRP_ROOT_XATTR)
1089 seq_puts(seq, ",xattr");
1090 if (strlen(root->release_agent_path))
1091 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1092 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1093 seq_puts(seq, ",clone_children");
1094 if (strlen(root->name))
1095 seq_printf(seq, ",name=%s", root->name);
1096 mutex_unlock(&cgroup_root_mutex);
1100 struct cgroup_sb_opts {
1101 unsigned long subsys_mask;
1102 unsigned long flags;
1103 char *release_agent;
1104 bool cpuset_clone_children;
1106 /* User explicitly requested empty subsystem */
1109 struct cgroupfs_root *new_root;
1114 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1115 * with cgroup_mutex held to protect the subsys[] array. This function takes
1116 * refcounts on subsystems to be used, unless it returns error, in which case
1117 * no refcounts are taken.
1119 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1121 char *token, *o = data;
1122 bool all_ss = false, one_ss = false;
1123 unsigned long mask = (unsigned long)-1;
1125 bool module_pin_failed = false;
1127 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1129 #ifdef CONFIG_CPUSETS
1130 mask = ~(1UL << cpuset_subsys_id);
1133 memset(opts, 0, sizeof(*opts));
1135 while ((token = strsep(&o, ",")) != NULL) {
1138 if (!strcmp(token, "none")) {
1139 /* Explicitly have no subsystems */
1143 if (!strcmp(token, "all")) {
1144 /* Mutually exclusive option 'all' + subsystem name */
1150 if (!strcmp(token, "__DEVEL__sane_behavior")) {
1151 opts->flags |= CGRP_ROOT_SANE_BEHAVIOR;
1154 if (!strcmp(token, "noprefix")) {
1155 opts->flags |= CGRP_ROOT_NOPREFIX;
1158 if (!strcmp(token, "clone_children")) {
1159 opts->cpuset_clone_children = true;
1162 if (!strcmp(token, "xattr")) {
1163 opts->flags |= CGRP_ROOT_XATTR;
1166 if (!strncmp(token, "release_agent=", 14)) {
1167 /* Specifying two release agents is forbidden */
1168 if (opts->release_agent)
1170 opts->release_agent =
1171 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1172 if (!opts->release_agent)
1176 if (!strncmp(token, "name=", 5)) {
1177 const char *name = token + 5;
1178 /* Can't specify an empty name */
1181 /* Must match [\w.-]+ */
1182 for (i = 0; i < strlen(name); i++) {
1186 if ((c == '.') || (c == '-') || (c == '_'))
1190 /* Specifying two names is forbidden */
1193 opts->name = kstrndup(name,
1194 MAX_CGROUP_ROOT_NAMELEN - 1,
1202 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1203 struct cgroup_subsys *ss = subsys[i];
1206 if (strcmp(token, ss->name))
1211 /* Mutually exclusive option 'all' + subsystem name */
1214 set_bit(i, &opts->subsys_mask);
1219 if (i == CGROUP_SUBSYS_COUNT)
1224 * If the 'all' option was specified select all the subsystems,
1225 * otherwise if 'none', 'name=' and a subsystem name options
1226 * were not specified, let's default to 'all'
1228 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1229 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1230 struct cgroup_subsys *ss = subsys[i];
1235 set_bit(i, &opts->subsys_mask);
1239 /* Consistency checks */
1241 if (opts->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1242 pr_warning("cgroup: sane_behavior: this is still under development and its behaviors will change, proceed at your own risk\n");
1244 if (opts->flags & CGRP_ROOT_NOPREFIX) {
1245 pr_err("cgroup: sane_behavior: noprefix is not allowed\n");
1249 if (opts->cpuset_clone_children) {
1250 pr_err("cgroup: sane_behavior: clone_children is not allowed\n");
1256 * Option noprefix was introduced just for backward compatibility
1257 * with the old cpuset, so we allow noprefix only if mounting just
1258 * the cpuset subsystem.
1260 if ((opts->flags & CGRP_ROOT_NOPREFIX) && (opts->subsys_mask & mask))
1264 /* Can't specify "none" and some subsystems */
1265 if (opts->subsys_mask && opts->none)
1269 * We either have to specify by name or by subsystems. (So all
1270 * empty hierarchies must have a name).
1272 if (!opts->subsys_mask && !opts->name)
1276 * Grab references on all the modules we'll need, so the subsystems
1277 * don't dance around before rebind_subsystems attaches them. This may
1278 * take duplicate reference counts on a subsystem that's already used,
1279 * but rebind_subsystems handles this case.
1281 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1282 unsigned long bit = 1UL << i;
1284 if (!(bit & opts->subsys_mask))
1286 if (!try_module_get(subsys[i]->module)) {
1287 module_pin_failed = true;
1291 if (module_pin_failed) {
1293 * oops, one of the modules was going away. this means that we
1294 * raced with a module_delete call, and to the user this is
1295 * essentially a "subsystem doesn't exist" case.
1297 for (i--; i >= 0; i--) {
1298 /* drop refcounts only on the ones we took */
1299 unsigned long bit = 1UL << i;
1301 if (!(bit & opts->subsys_mask))
1303 module_put(subsys[i]->module);
1311 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1314 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1315 unsigned long bit = 1UL << i;
1317 if (!(bit & subsys_mask))
1319 module_put(subsys[i]->module);
1323 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1326 struct cgroupfs_root *root = sb->s_fs_info;
1327 struct cgroup *cgrp = &root->top_cgroup;
1328 struct cgroup_sb_opts opts;
1329 unsigned long added_mask, removed_mask;
1331 if (root->flags & CGRP_ROOT_SANE_BEHAVIOR) {
1332 pr_err("cgroup: sane_behavior: remount is not allowed\n");
1336 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1337 mutex_lock(&cgroup_mutex);
1338 mutex_lock(&cgroup_root_mutex);
1340 /* See what subsystems are wanted */
1341 ret = parse_cgroupfs_options(data, &opts);
1345 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1346 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1347 task_tgid_nr(current), current->comm);
1349 added_mask = opts.subsys_mask & ~root->subsys_mask;
1350 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1352 /* Don't allow flags or name to change at remount */
1353 if (opts.flags != root->flags ||
1354 (opts.name && strcmp(opts.name, root->name))) {
1356 drop_parsed_module_refcounts(opts.subsys_mask);
1361 * Clear out the files of subsystems that should be removed, do
1362 * this before rebind_subsystems, since rebind_subsystems may
1363 * change this hierarchy's subsys_list.
1365 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1367 ret = rebind_subsystems(root, opts.subsys_mask);
1369 /* rebind_subsystems failed, re-populate the removed files */
1370 cgroup_populate_dir(cgrp, false, removed_mask);
1371 drop_parsed_module_refcounts(opts.subsys_mask);
1375 /* re-populate subsystem files */
1376 cgroup_populate_dir(cgrp, false, added_mask);
1378 if (opts.release_agent)
1379 strcpy(root->release_agent_path, opts.release_agent);
1381 kfree(opts.release_agent);
1383 mutex_unlock(&cgroup_root_mutex);
1384 mutex_unlock(&cgroup_mutex);
1385 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1389 static const struct super_operations cgroup_ops = {
1390 .statfs = simple_statfs,
1391 .drop_inode = generic_delete_inode,
1392 .show_options = cgroup_show_options,
1393 .remount_fs = cgroup_remount,
1396 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1398 INIT_LIST_HEAD(&cgrp->sibling);
1399 INIT_LIST_HEAD(&cgrp->children);
1400 INIT_LIST_HEAD(&cgrp->files);
1401 INIT_LIST_HEAD(&cgrp->cset_links);
1402 INIT_LIST_HEAD(&cgrp->release_list);
1403 INIT_LIST_HEAD(&cgrp->pidlists);
1404 mutex_init(&cgrp->pidlist_mutex);
1405 INIT_LIST_HEAD(&cgrp->event_list);
1406 spin_lock_init(&cgrp->event_list_lock);
1407 simple_xattrs_init(&cgrp->xattrs);
1410 static void init_cgroup_root(struct cgroupfs_root *root)
1412 struct cgroup *cgrp = &root->top_cgroup;
1414 INIT_LIST_HEAD(&root->subsys_list);
1415 INIT_LIST_HEAD(&root->root_list);
1416 root->number_of_cgroups = 1;
1418 cgrp->name = &root_cgroup_name;
1419 init_cgroup_housekeeping(cgrp);
1422 static int cgroup_init_root_id(struct cgroupfs_root *root)
1426 lockdep_assert_held(&cgroup_mutex);
1427 lockdep_assert_held(&cgroup_root_mutex);
1429 id = idr_alloc_cyclic(&cgroup_hierarchy_idr, root, 2, 0, GFP_KERNEL);
1433 root->hierarchy_id = id;
1437 static void cgroup_exit_root_id(struct cgroupfs_root *root)
1439 lockdep_assert_held(&cgroup_mutex);
1440 lockdep_assert_held(&cgroup_root_mutex);
1442 if (root->hierarchy_id) {
1443 idr_remove(&cgroup_hierarchy_idr, root->hierarchy_id);
1444 root->hierarchy_id = 0;
1448 static int cgroup_test_super(struct super_block *sb, void *data)
1450 struct cgroup_sb_opts *opts = data;
1451 struct cgroupfs_root *root = sb->s_fs_info;
1453 /* If we asked for a name then it must match */
1454 if (opts->name && strcmp(opts->name, root->name))
1458 * If we asked for subsystems (or explicitly for no
1459 * subsystems) then they must match
1461 if ((opts->subsys_mask || opts->none)
1462 && (opts->subsys_mask != root->subsys_mask))
1468 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1470 struct cgroupfs_root *root;
1472 if (!opts->subsys_mask && !opts->none)
1475 root = kzalloc(sizeof(*root), GFP_KERNEL);
1477 return ERR_PTR(-ENOMEM);
1479 init_cgroup_root(root);
1481 root->subsys_mask = opts->subsys_mask;
1482 root->flags = opts->flags;
1483 ida_init(&root->cgroup_ida);
1484 if (opts->release_agent)
1485 strcpy(root->release_agent_path, opts->release_agent);
1487 strcpy(root->name, opts->name);
1488 if (opts->cpuset_clone_children)
1489 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1493 static void cgroup_free_root(struct cgroupfs_root *root)
1496 /* hierarhcy ID shoulid already have been released */
1497 WARN_ON_ONCE(root->hierarchy_id);
1499 ida_destroy(&root->cgroup_ida);
1504 static int cgroup_set_super(struct super_block *sb, void *data)
1507 struct cgroup_sb_opts *opts = data;
1509 /* If we don't have a new root, we can't set up a new sb */
1510 if (!opts->new_root)
1513 BUG_ON(!opts->subsys_mask && !opts->none);
1515 ret = set_anon_super(sb, NULL);
1519 sb->s_fs_info = opts->new_root;
1520 opts->new_root->sb = sb;
1522 sb->s_blocksize = PAGE_CACHE_SIZE;
1523 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1524 sb->s_magic = CGROUP_SUPER_MAGIC;
1525 sb->s_op = &cgroup_ops;
1530 static int cgroup_get_rootdir(struct super_block *sb)
1532 static const struct dentry_operations cgroup_dops = {
1533 .d_iput = cgroup_diput,
1534 .d_delete = cgroup_delete,
1537 struct inode *inode =
1538 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1543 inode->i_fop = &simple_dir_operations;
1544 inode->i_op = &cgroup_dir_inode_operations;
1545 /* directories start off with i_nlink == 2 (for "." entry) */
1547 sb->s_root = d_make_root(inode);
1550 /* for everything else we want ->d_op set */
1551 sb->s_d_op = &cgroup_dops;
1555 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1556 int flags, const char *unused_dev_name,
1559 struct cgroup_sb_opts opts;
1560 struct cgroupfs_root *root;
1562 struct super_block *sb;
1563 struct cgroupfs_root *new_root;
1564 struct inode *inode;
1566 /* First find the desired set of subsystems */
1567 mutex_lock(&cgroup_mutex);
1568 ret = parse_cgroupfs_options(data, &opts);
1569 mutex_unlock(&cgroup_mutex);
1574 * Allocate a new cgroup root. We may not need it if we're
1575 * reusing an existing hierarchy.
1577 new_root = cgroup_root_from_opts(&opts);
1578 if (IS_ERR(new_root)) {
1579 ret = PTR_ERR(new_root);
1582 opts.new_root = new_root;
1584 /* Locate an existing or new sb for this hierarchy */
1585 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1588 cgroup_free_root(opts.new_root);
1592 root = sb->s_fs_info;
1594 if (root == opts.new_root) {
1595 /* We used the new root structure, so this is a new hierarchy */
1596 struct list_head tmp_links;
1597 struct cgroup *root_cgrp = &root->top_cgroup;
1598 struct cgroupfs_root *existing_root;
1599 const struct cred *cred;
1601 struct css_set *cset;
1603 BUG_ON(sb->s_root != NULL);
1605 ret = cgroup_get_rootdir(sb);
1607 goto drop_new_super;
1608 inode = sb->s_root->d_inode;
1610 mutex_lock(&inode->i_mutex);
1611 mutex_lock(&cgroup_mutex);
1612 mutex_lock(&cgroup_root_mutex);
1614 /* Check for name clashes with existing mounts */
1616 if (strlen(root->name))
1617 for_each_active_root(existing_root)
1618 if (!strcmp(existing_root->name, root->name))
1622 * We're accessing css_set_count without locking
1623 * css_set_lock here, but that's OK - it can only be
1624 * increased by someone holding cgroup_lock, and
1625 * that's us. The worst that can happen is that we
1626 * have some link structures left over
1628 ret = allocate_cgrp_cset_links(css_set_count, &tmp_links);
1632 ret = cgroup_init_root_id(root);
1636 ret = rebind_subsystems(root, root->subsys_mask);
1637 if (ret == -EBUSY) {
1638 free_cgrp_cset_links(&tmp_links);
1642 * There must be no failure case after here, since rebinding
1643 * takes care of subsystems' refcounts, which are explicitly
1644 * dropped in the failure exit path.
1647 /* EBUSY should be the only error here */
1650 list_add(&root->root_list, &roots);
1653 sb->s_root->d_fsdata = root_cgrp;
1654 root->top_cgroup.dentry = sb->s_root;
1656 /* Link the top cgroup in this hierarchy into all
1657 * the css_set objects */
1658 write_lock(&css_set_lock);
1659 hash_for_each(css_set_table, i, cset, hlist)
1660 link_css_set(&tmp_links, cset, root_cgrp);
1661 write_unlock(&css_set_lock);
1663 free_cgrp_cset_links(&tmp_links);
1665 BUG_ON(!list_empty(&root_cgrp->children));
1666 BUG_ON(root->number_of_cgroups != 1);
1668 cred = override_creds(&init_cred);
1669 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1671 mutex_unlock(&cgroup_root_mutex);
1672 mutex_unlock(&cgroup_mutex);
1673 mutex_unlock(&inode->i_mutex);
1676 * We re-used an existing hierarchy - the new root (if
1677 * any) is not needed
1679 cgroup_free_root(opts.new_root);
1681 if (root->flags != opts.flags) {
1682 if ((root->flags | opts.flags) & CGRP_ROOT_SANE_BEHAVIOR) {
1683 pr_err("cgroup: sane_behavior: new mount options should match the existing superblock\n");
1685 goto drop_new_super;
1687 pr_warning("cgroup: new mount options do not match the existing superblock, will be ignored\n");
1691 /* no subsys rebinding, so refcounts don't change */
1692 drop_parsed_module_refcounts(opts.subsys_mask);
1695 kfree(opts.release_agent);
1697 return dget(sb->s_root);
1700 cgroup_exit_root_id(root);
1701 mutex_unlock(&cgroup_root_mutex);
1702 mutex_unlock(&cgroup_mutex);
1703 mutex_unlock(&inode->i_mutex);
1705 deactivate_locked_super(sb);
1707 drop_parsed_module_refcounts(opts.subsys_mask);
1709 kfree(opts.release_agent);
1711 return ERR_PTR(ret);
1714 static void cgroup_kill_sb(struct super_block *sb) {
1715 struct cgroupfs_root *root = sb->s_fs_info;
1716 struct cgroup *cgrp = &root->top_cgroup;
1717 struct cgrp_cset_link *link, *tmp_link;
1722 BUG_ON(root->number_of_cgroups != 1);
1723 BUG_ON(!list_empty(&cgrp->children));
1725 mutex_lock(&cgroup_mutex);
1726 mutex_lock(&cgroup_root_mutex);
1728 /* Rebind all subsystems back to the default hierarchy */
1729 ret = rebind_subsystems(root, 0);
1730 /* Shouldn't be able to fail ... */
1734 * Release all the links from cset_links to this hierarchy's
1737 write_lock(&css_set_lock);
1739 list_for_each_entry_safe(link, tmp_link, &cgrp->cset_links, cset_link) {
1740 list_del(&link->cset_link);
1741 list_del(&link->cgrp_link);
1744 write_unlock(&css_set_lock);
1746 if (!list_empty(&root->root_list)) {
1747 list_del(&root->root_list);
1751 cgroup_exit_root_id(root);
1753 mutex_unlock(&cgroup_root_mutex);
1754 mutex_unlock(&cgroup_mutex);
1756 simple_xattrs_free(&cgrp->xattrs);
1758 kill_litter_super(sb);
1759 cgroup_free_root(root);
1762 static struct file_system_type cgroup_fs_type = {
1764 .mount = cgroup_mount,
1765 .kill_sb = cgroup_kill_sb,
1768 static struct kobject *cgroup_kobj;
1771 * cgroup_path - generate the path of a cgroup
1772 * @cgrp: the cgroup in question
1773 * @buf: the buffer to write the path into
1774 * @buflen: the length of the buffer
1776 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1778 * We can't generate cgroup path using dentry->d_name, as accessing
1779 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1780 * inode's i_mutex, while on the other hand cgroup_path() can be called
1781 * with some irq-safe spinlocks held.
1783 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1785 int ret = -ENAMETOOLONG;
1788 if (!cgrp->parent) {
1789 if (strlcpy(buf, "/", buflen) >= buflen)
1790 return -ENAMETOOLONG;
1794 start = buf + buflen - 1;
1799 const char *name = cgroup_name(cgrp);
1803 if ((start -= len) < buf)
1805 memcpy(start, name, len);
1811 cgrp = cgrp->parent;
1812 } while (cgrp->parent);
1814 memmove(buf, start, buf + buflen - start);
1819 EXPORT_SYMBOL_GPL(cgroup_path);
1822 * task_cgroup_path_from_hierarchy - cgroup path of a task on a hierarchy
1823 * @task: target task
1824 * @hierarchy_id: the hierarchy to look up @task's cgroup from
1825 * @buf: the buffer to write the path into
1826 * @buflen: the length of the buffer
1828 * Determine @task's cgroup on the hierarchy specified by @hierarchy_id and
1829 * copy its path into @buf. This function grabs cgroup_mutex and shouldn't
1830 * be used inside locks used by cgroup controller callbacks.
1832 int task_cgroup_path_from_hierarchy(struct task_struct *task, int hierarchy_id,
1833 char *buf, size_t buflen)
1835 struct cgroupfs_root *root;
1836 struct cgroup *cgrp = NULL;
1839 mutex_lock(&cgroup_mutex);
1841 root = idr_find(&cgroup_hierarchy_idr, hierarchy_id);
1843 cgrp = task_cgroup_from_root(task, root);
1844 ret = cgroup_path(cgrp, buf, buflen);
1847 mutex_unlock(&cgroup_mutex);
1851 EXPORT_SYMBOL_GPL(task_cgroup_path_from_hierarchy);
1854 * Control Group taskset
1856 struct task_and_cgroup {
1857 struct task_struct *task;
1858 struct cgroup *cgrp;
1862 struct cgroup_taskset {
1863 struct task_and_cgroup single;
1864 struct flex_array *tc_array;
1867 struct cgroup *cur_cgrp;
1871 * cgroup_taskset_first - reset taskset and return the first task
1872 * @tset: taskset of interest
1874 * @tset iteration is initialized and the first task is returned.
1876 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1878 if (tset->tc_array) {
1880 return cgroup_taskset_next(tset);
1882 tset->cur_cgrp = tset->single.cgrp;
1883 return tset->single.task;
1886 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1889 * cgroup_taskset_next - iterate to the next task in taskset
1890 * @tset: taskset of interest
1892 * Return the next task in @tset. Iteration must have been initialized
1893 * with cgroup_taskset_first().
1895 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1897 struct task_and_cgroup *tc;
1899 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1902 tc = flex_array_get(tset->tc_array, tset->idx++);
1903 tset->cur_cgrp = tc->cgrp;
1906 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1909 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1910 * @tset: taskset of interest
1912 * Return the cgroup for the current (last returned) task of @tset. This
1913 * function must be preceded by either cgroup_taskset_first() or
1914 * cgroup_taskset_next().
1916 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1918 return tset->cur_cgrp;
1920 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1923 * cgroup_taskset_size - return the number of tasks in taskset
1924 * @tset: taskset of interest
1926 int cgroup_taskset_size(struct cgroup_taskset *tset)
1928 return tset->tc_array ? tset->tc_array_len : 1;
1930 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1934 * cgroup_task_migrate - move a task from one cgroup to another.
1936 * Must be called with cgroup_mutex and threadgroup locked.
1938 static void cgroup_task_migrate(struct cgroup *old_cgrp,
1939 struct task_struct *tsk,
1940 struct css_set *new_cset)
1942 struct css_set *old_cset;
1945 * We are synchronized through threadgroup_lock() against PF_EXITING
1946 * setting such that we can't race against cgroup_exit() changing the
1947 * css_set to init_css_set and dropping the old one.
1949 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1950 old_cset = tsk->cgroups;
1953 rcu_assign_pointer(tsk->cgroups, new_cset);
1956 /* Update the css_set linked lists if we're using them */
1957 write_lock(&css_set_lock);
1958 if (!list_empty(&tsk->cg_list))
1959 list_move(&tsk->cg_list, &new_cset->tasks);
1960 write_unlock(&css_set_lock);
1963 * We just gained a reference on old_cset by taking it from the
1964 * task. As trading it for new_cset is protected by cgroup_mutex,
1965 * we're safe to drop it here; it will be freed under RCU.
1967 set_bit(CGRP_RELEASABLE, &old_cgrp->flags);
1968 put_css_set(old_cset);
1972 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1973 * @cgrp: the cgroup to attach to
1974 * @tsk: the task or the leader of the threadgroup to be attached
1975 * @threadgroup: attach the whole threadgroup?
1977 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1978 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1980 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
1983 int retval, i, group_size;
1984 struct cgroup_subsys *ss, *failed_ss = NULL;
1985 struct cgroupfs_root *root = cgrp->root;
1986 /* threadgroup list cursor and array */
1987 struct task_struct *leader = tsk;
1988 struct task_and_cgroup *tc;
1989 struct flex_array *group;
1990 struct cgroup_taskset tset = { };
1993 * step 0: in order to do expensive, possibly blocking operations for
1994 * every thread, we cannot iterate the thread group list, since it needs
1995 * rcu or tasklist locked. instead, build an array of all threads in the
1996 * group - group_rwsem prevents new threads from appearing, and if
1997 * threads exit, this will just be an over-estimate.
2000 group_size = get_nr_threads(tsk);
2003 /* flex_array supports very large thread-groups better than kmalloc. */
2004 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2007 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2008 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
2010 goto out_free_group_list;
2014 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2015 * already PF_EXITING could be freed from underneath us unless we
2016 * take an rcu_read_lock.
2020 struct task_and_cgroup ent;
2022 /* @tsk either already exited or can't exit until the end */
2023 if (tsk->flags & PF_EXITING)
2026 /* as per above, nr_threads may decrease, but not increase. */
2027 BUG_ON(i >= group_size);
2029 ent.cgrp = task_cgroup_from_root(tsk, root);
2030 /* nothing to do if this task is already in the cgroup */
2031 if (ent.cgrp == cgrp)
2034 * saying GFP_ATOMIC has no effect here because we did prealloc
2035 * earlier, but it's good form to communicate our expectations.
2037 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2038 BUG_ON(retval != 0);
2043 } while_each_thread(leader, tsk);
2045 /* remember the number of threads in the array for later. */
2047 tset.tc_array = group;
2048 tset.tc_array_len = group_size;
2050 /* methods shouldn't be called if no task is actually migrating */
2053 goto out_free_group_list;
2056 * step 1: check that we can legitimately attach to the cgroup.
2058 for_each_subsys(root, ss) {
2059 if (ss->can_attach) {
2060 retval = ss->can_attach(cgrp, &tset);
2063 goto out_cancel_attach;
2069 * step 2: make sure css_sets exist for all threads to be migrated.
2070 * we use find_css_set, which allocates a new one if necessary.
2072 for (i = 0; i < group_size; i++) {
2073 tc = flex_array_get(group, i);
2074 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2077 goto out_put_css_set_refs;
2082 * step 3: now that we're guaranteed success wrt the css_sets,
2083 * proceed to move all tasks to the new cgroup. There are no
2084 * failure cases after here, so this is the commit point.
2086 for (i = 0; i < group_size; i++) {
2087 tc = flex_array_get(group, i);
2088 cgroup_task_migrate(tc->cgrp, tc->task, tc->cg);
2090 /* nothing is sensitive to fork() after this point. */
2093 * step 4: do subsystem attach callbacks.
2095 for_each_subsys(root, ss) {
2097 ss->attach(cgrp, &tset);
2101 * step 5: success! and cleanup
2104 out_put_css_set_refs:
2106 for (i = 0; i < group_size; i++) {
2107 tc = flex_array_get(group, i);
2110 put_css_set(tc->cg);
2115 for_each_subsys(root, ss) {
2116 if (ss == failed_ss)
2118 if (ss->cancel_attach)
2119 ss->cancel_attach(cgrp, &tset);
2122 out_free_group_list:
2123 flex_array_free(group);
2128 * Find the task_struct of the task to attach by vpid and pass it along to the
2129 * function to attach either it or all tasks in its threadgroup. Will lock
2130 * cgroup_mutex and threadgroup; may take task_lock of task.
2132 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2134 struct task_struct *tsk;
2135 const struct cred *cred = current_cred(), *tcred;
2138 if (!cgroup_lock_live_group(cgrp))
2144 tsk = find_task_by_vpid(pid);
2148 goto out_unlock_cgroup;
2151 * even if we're attaching all tasks in the thread group, we
2152 * only need to check permissions on one of them.
2154 tcred = __task_cred(tsk);
2155 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2156 !uid_eq(cred->euid, tcred->uid) &&
2157 !uid_eq(cred->euid, tcred->suid)) {
2160 goto out_unlock_cgroup;
2166 tsk = tsk->group_leader;
2169 * Workqueue threads may acquire PF_NO_SETAFFINITY and become
2170 * trapped in a cpuset, or RT worker may be born in a cgroup
2171 * with no rt_runtime allocated. Just say no.
2173 if (tsk == kthreadd_task || (tsk->flags & PF_NO_SETAFFINITY)) {
2176 goto out_unlock_cgroup;
2179 get_task_struct(tsk);
2182 threadgroup_lock(tsk);
2184 if (!thread_group_leader(tsk)) {
2186 * a race with de_thread from another thread's exec()
2187 * may strip us of our leadership, if this happens,
2188 * there is no choice but to throw this task away and
2189 * try again; this is
2190 * "double-double-toil-and-trouble-check locking".
2192 threadgroup_unlock(tsk);
2193 put_task_struct(tsk);
2194 goto retry_find_task;
2198 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2200 threadgroup_unlock(tsk);
2202 put_task_struct(tsk);
2204 mutex_unlock(&cgroup_mutex);
2209 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2210 * @from: attach to all cgroups of a given task
2211 * @tsk: the task to be attached
2213 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2215 struct cgroupfs_root *root;
2218 mutex_lock(&cgroup_mutex);
2219 for_each_active_root(root) {
2220 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2222 retval = cgroup_attach_task(from_cg, tsk, false);
2226 mutex_unlock(&cgroup_mutex);
2230 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2232 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2234 return attach_task_by_pid(cgrp, pid, false);
2237 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2239 return attach_task_by_pid(cgrp, tgid, true);
2242 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2245 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2246 if (strlen(buffer) >= PATH_MAX)
2248 if (!cgroup_lock_live_group(cgrp))
2250 mutex_lock(&cgroup_root_mutex);
2251 strcpy(cgrp->root->release_agent_path, buffer);
2252 mutex_unlock(&cgroup_root_mutex);
2253 mutex_unlock(&cgroup_mutex);
2257 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2258 struct seq_file *seq)
2260 if (!cgroup_lock_live_group(cgrp))
2262 seq_puts(seq, cgrp->root->release_agent_path);
2263 seq_putc(seq, '\n');
2264 mutex_unlock(&cgroup_mutex);
2268 static int cgroup_sane_behavior_show(struct cgroup *cgrp, struct cftype *cft,
2269 struct seq_file *seq)
2271 seq_printf(seq, "%d\n", cgroup_sane_behavior(cgrp));
2275 /* A buffer size big enough for numbers or short strings */
2276 #define CGROUP_LOCAL_BUFFER_SIZE 64
2278 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2280 const char __user *userbuf,
2281 size_t nbytes, loff_t *unused_ppos)
2283 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2289 if (nbytes >= sizeof(buffer))
2291 if (copy_from_user(buffer, userbuf, nbytes))
2294 buffer[nbytes] = 0; /* nul-terminate */
2295 if (cft->write_u64) {
2296 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2299 retval = cft->write_u64(cgrp, cft, val);
2301 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2304 retval = cft->write_s64(cgrp, cft, val);
2311 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2313 const char __user *userbuf,
2314 size_t nbytes, loff_t *unused_ppos)
2316 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2318 size_t max_bytes = cft->max_write_len;
2319 char *buffer = local_buffer;
2322 max_bytes = sizeof(local_buffer) - 1;
2323 if (nbytes >= max_bytes)
2325 /* Allocate a dynamic buffer if we need one */
2326 if (nbytes >= sizeof(local_buffer)) {
2327 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2331 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2336 buffer[nbytes] = 0; /* nul-terminate */
2337 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2341 if (buffer != local_buffer)
2346 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2347 size_t nbytes, loff_t *ppos)
2349 struct cftype *cft = __d_cft(file->f_dentry);
2350 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2352 if (cgroup_is_dead(cgrp))
2355 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2356 if (cft->write_u64 || cft->write_s64)
2357 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2358 if (cft->write_string)
2359 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2361 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2362 return ret ? ret : nbytes;
2367 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2369 char __user *buf, size_t nbytes,
2372 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2373 u64 val = cft->read_u64(cgrp, cft);
2374 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2376 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2379 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2381 char __user *buf, size_t nbytes,
2384 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2385 s64 val = cft->read_s64(cgrp, cft);
2386 int len = sprintf(tmp, "%lld\n", (long long) val);
2388 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2391 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2392 size_t nbytes, loff_t *ppos)
2394 struct cftype *cft = __d_cft(file->f_dentry);
2395 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2397 if (cgroup_is_dead(cgrp))
2401 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2403 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2405 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2410 * seqfile ops/methods for returning structured data. Currently just
2411 * supports string->u64 maps, but can be extended in future.
2414 struct cgroup_seqfile_state {
2416 struct cgroup *cgroup;
2419 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2421 struct seq_file *sf = cb->state;
2422 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2425 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2427 struct cgroup_seqfile_state *state = m->private;
2428 struct cftype *cft = state->cft;
2429 if (cft->read_map) {
2430 struct cgroup_map_cb cb = {
2431 .fill = cgroup_map_add,
2434 return cft->read_map(state->cgroup, cft, &cb);
2436 return cft->read_seq_string(state->cgroup, cft, m);
2439 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2441 struct seq_file *seq = file->private_data;
2442 kfree(seq->private);
2443 return single_release(inode, file);
2446 static const struct file_operations cgroup_seqfile_operations = {
2448 .write = cgroup_file_write,
2449 .llseek = seq_lseek,
2450 .release = cgroup_seqfile_release,
2453 static int cgroup_file_open(struct inode *inode, struct file *file)
2458 err = generic_file_open(inode, file);
2461 cft = __d_cft(file->f_dentry);
2463 if (cft->read_map || cft->read_seq_string) {
2464 struct cgroup_seqfile_state *state;
2466 state = kzalloc(sizeof(*state), GFP_USER);
2471 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2472 file->f_op = &cgroup_seqfile_operations;
2473 err = single_open(file, cgroup_seqfile_show, state);
2476 } else if (cft->open)
2477 err = cft->open(inode, file);
2484 static int cgroup_file_release(struct inode *inode, struct file *file)
2486 struct cftype *cft = __d_cft(file->f_dentry);
2488 return cft->release(inode, file);
2493 * cgroup_rename - Only allow simple rename of directories in place.
2495 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2496 struct inode *new_dir, struct dentry *new_dentry)
2499 struct cgroup_name *name, *old_name;
2500 struct cgroup *cgrp;
2503 * It's convinient to use parent dir's i_mutex to protected
2506 lockdep_assert_held(&old_dir->i_mutex);
2508 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2510 if (new_dentry->d_inode)
2512 if (old_dir != new_dir)
2515 cgrp = __d_cgrp(old_dentry);
2518 * This isn't a proper migration and its usefulness is very
2519 * limited. Disallow if sane_behavior.
2521 if (cgroup_sane_behavior(cgrp))
2524 name = cgroup_alloc_name(new_dentry);
2528 ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2534 old_name = cgrp->name;
2535 rcu_assign_pointer(cgrp->name, name);
2537 kfree_rcu(old_name, rcu_head);
2541 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2543 if (S_ISDIR(dentry->d_inode->i_mode))
2544 return &__d_cgrp(dentry)->xattrs;
2546 return &__d_cfe(dentry)->xattrs;
2549 static inline int xattr_enabled(struct dentry *dentry)
2551 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2552 return root->flags & CGRP_ROOT_XATTR;
2555 static bool is_valid_xattr(const char *name)
2557 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2558 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2563 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2564 const void *val, size_t size, int flags)
2566 if (!xattr_enabled(dentry))
2568 if (!is_valid_xattr(name))
2570 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2573 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2575 if (!xattr_enabled(dentry))
2577 if (!is_valid_xattr(name))
2579 return simple_xattr_remove(__d_xattrs(dentry), name);
2582 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2583 void *buf, size_t size)
2585 if (!xattr_enabled(dentry))
2587 if (!is_valid_xattr(name))
2589 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2592 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2594 if (!xattr_enabled(dentry))
2596 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2599 static const struct file_operations cgroup_file_operations = {
2600 .read = cgroup_file_read,
2601 .write = cgroup_file_write,
2602 .llseek = generic_file_llseek,
2603 .open = cgroup_file_open,
2604 .release = cgroup_file_release,
2607 static const struct inode_operations cgroup_file_inode_operations = {
2608 .setxattr = cgroup_setxattr,
2609 .getxattr = cgroup_getxattr,
2610 .listxattr = cgroup_listxattr,
2611 .removexattr = cgroup_removexattr,
2614 static const struct inode_operations cgroup_dir_inode_operations = {
2615 .lookup = cgroup_lookup,
2616 .mkdir = cgroup_mkdir,
2617 .rmdir = cgroup_rmdir,
2618 .rename = cgroup_rename,
2619 .setxattr = cgroup_setxattr,
2620 .getxattr = cgroup_getxattr,
2621 .listxattr = cgroup_listxattr,
2622 .removexattr = cgroup_removexattr,
2625 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2627 if (dentry->d_name.len > NAME_MAX)
2628 return ERR_PTR(-ENAMETOOLONG);
2629 d_add(dentry, NULL);
2634 * Check if a file is a control file
2636 static inline struct cftype *__file_cft(struct file *file)
2638 if (file_inode(file)->i_fop != &cgroup_file_operations)
2639 return ERR_PTR(-EINVAL);
2640 return __d_cft(file->f_dentry);
2643 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2644 struct super_block *sb)
2646 struct inode *inode;
2650 if (dentry->d_inode)
2653 inode = cgroup_new_inode(mode, sb);
2657 if (S_ISDIR(mode)) {
2658 inode->i_op = &cgroup_dir_inode_operations;
2659 inode->i_fop = &simple_dir_operations;
2661 /* start off with i_nlink == 2 (for "." entry) */
2663 inc_nlink(dentry->d_parent->d_inode);
2666 * Control reaches here with cgroup_mutex held.
2667 * @inode->i_mutex should nest outside cgroup_mutex but we
2668 * want to populate it immediately without releasing
2669 * cgroup_mutex. As @inode isn't visible to anyone else
2670 * yet, trylock will always succeed without affecting
2673 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2674 } else if (S_ISREG(mode)) {
2676 inode->i_fop = &cgroup_file_operations;
2677 inode->i_op = &cgroup_file_inode_operations;
2679 d_instantiate(dentry, inode);
2680 dget(dentry); /* Extra count - pin the dentry in core */
2685 * cgroup_file_mode - deduce file mode of a control file
2686 * @cft: the control file in question
2688 * returns cft->mode if ->mode is not 0
2689 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2690 * returns S_IRUGO if it has only a read handler
2691 * returns S_IWUSR if it has only a write hander
2693 static umode_t cgroup_file_mode(const struct cftype *cft)
2700 if (cft->read || cft->read_u64 || cft->read_s64 ||
2701 cft->read_map || cft->read_seq_string)
2704 if (cft->write || cft->write_u64 || cft->write_s64 ||
2705 cft->write_string || cft->trigger)
2711 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2714 struct dentry *dir = cgrp->dentry;
2715 struct cgroup *parent = __d_cgrp(dir);
2716 struct dentry *dentry;
2720 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2722 if (subsys && !(cgrp->root->flags & CGRP_ROOT_NOPREFIX)) {
2723 strcpy(name, subsys->name);
2726 strcat(name, cft->name);
2728 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2730 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2734 dentry = lookup_one_len(name, dir, strlen(name));
2735 if (IS_ERR(dentry)) {
2736 error = PTR_ERR(dentry);
2740 cfe->type = (void *)cft;
2741 cfe->dentry = dentry;
2742 dentry->d_fsdata = cfe;
2743 simple_xattrs_init(&cfe->xattrs);
2745 mode = cgroup_file_mode(cft);
2746 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2748 list_add_tail(&cfe->node, &parent->files);
2757 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2758 struct cftype cfts[], bool is_add)
2763 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2764 /* does cft->flags tell us to skip this file on @cgrp? */
2765 if ((cft->flags & CFTYPE_INSANE) && cgroup_sane_behavior(cgrp))
2767 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2769 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2773 err = cgroup_add_file(cgrp, subsys, cft);
2775 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2779 cgroup_rm_file(cgrp, cft);
2785 static void cgroup_cfts_prepare(void)
2786 __acquires(&cgroup_mutex)
2789 * Thanks to the entanglement with vfs inode locking, we can't walk
2790 * the existing cgroups under cgroup_mutex and create files.
2791 * Instead, we use cgroup_for_each_descendant_pre() and drop RCU
2792 * read lock before calling cgroup_addrm_files().
2794 mutex_lock(&cgroup_mutex);
2797 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2798 struct cftype *cfts, bool is_add)
2799 __releases(&cgroup_mutex)
2802 struct cgroup *cgrp, *root = &ss->root->top_cgroup;
2803 struct super_block *sb = ss->root->sb;
2804 struct dentry *prev = NULL;
2805 struct inode *inode;
2808 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2809 if (!cfts || ss->root == &rootnode ||
2810 !atomic_inc_not_zero(&sb->s_active)) {
2811 mutex_unlock(&cgroup_mutex);
2816 * All cgroups which are created after we drop cgroup_mutex will
2817 * have the updated set of files, so we only need to update the
2818 * cgroups created before the current @cgroup_serial_nr_cursor.
2820 update_upto = atomic64_read(&cgroup_serial_nr_cursor);
2822 mutex_unlock(&cgroup_mutex);
2824 /* @root always needs to be updated */
2825 inode = root->dentry->d_inode;
2826 mutex_lock(&inode->i_mutex);
2827 mutex_lock(&cgroup_mutex);
2828 cgroup_addrm_files(root, ss, cfts, is_add);
2829 mutex_unlock(&cgroup_mutex);
2830 mutex_unlock(&inode->i_mutex);
2832 /* add/rm files for all cgroups created before */
2834 cgroup_for_each_descendant_pre(cgrp, root) {
2835 if (cgroup_is_dead(cgrp))
2838 inode = cgrp->dentry->d_inode;
2843 prev = cgrp->dentry;
2845 mutex_lock(&inode->i_mutex);
2846 mutex_lock(&cgroup_mutex);
2847 if (cgrp->serial_nr <= update_upto && !cgroup_is_dead(cgrp))
2848 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2849 mutex_unlock(&cgroup_mutex);
2850 mutex_unlock(&inode->i_mutex);
2856 deactivate_super(sb);
2860 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2861 * @ss: target cgroup subsystem
2862 * @cfts: zero-length name terminated array of cftypes
2864 * Register @cfts to @ss. Files described by @cfts are created for all
2865 * existing cgroups to which @ss is attached and all future cgroups will
2866 * have them too. This function can be called anytime whether @ss is
2869 * Returns 0 on successful registration, -errno on failure. Note that this
2870 * function currently returns 0 as long as @cfts registration is successful
2871 * even if some file creation attempts on existing cgroups fail.
2873 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2875 struct cftype_set *set;
2877 set = kzalloc(sizeof(*set), GFP_KERNEL);
2881 cgroup_cfts_prepare();
2883 list_add_tail(&set->node, &ss->cftsets);
2884 cgroup_cfts_commit(ss, cfts, true);
2888 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2891 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2892 * @ss: target cgroup subsystem
2893 * @cfts: zero-length name terminated array of cftypes
2895 * Unregister @cfts from @ss. Files described by @cfts are removed from
2896 * all existing cgroups to which @ss is attached and all future cgroups
2897 * won't have them either. This function can be called anytime whether @ss
2898 * is attached or not.
2900 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2901 * registered with @ss.
2903 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2905 struct cftype_set *set;
2907 cgroup_cfts_prepare();
2909 list_for_each_entry(set, &ss->cftsets, node) {
2910 if (set->cfts == cfts) {
2911 list_del(&set->node);
2913 cgroup_cfts_commit(ss, cfts, false);
2918 cgroup_cfts_commit(ss, NULL, false);
2923 * cgroup_task_count - count the number of tasks in a cgroup.
2924 * @cgrp: the cgroup in question
2926 * Return the number of tasks in the cgroup.
2928 int cgroup_task_count(const struct cgroup *cgrp)
2931 struct cgrp_cset_link *link;
2933 read_lock(&css_set_lock);
2934 list_for_each_entry(link, &cgrp->cset_links, cset_link)
2935 count += atomic_read(&link->cset->refcount);
2936 read_unlock(&css_set_lock);
2941 * Advance a list_head iterator. The iterator should be positioned at
2942 * the start of a css_set
2944 static void cgroup_advance_iter(struct cgroup *cgrp, struct cgroup_iter *it)
2946 struct list_head *l = it->cset_link;
2947 struct cgrp_cset_link *link;
2948 struct css_set *cset;
2950 /* Advance to the next non-empty css_set */
2953 if (l == &cgrp->cset_links) {
2954 it->cset_link = NULL;
2957 link = list_entry(l, struct cgrp_cset_link, cset_link);
2959 } while (list_empty(&cset->tasks));
2961 it->task = cset->tasks.next;
2965 * To reduce the fork() overhead for systems that are not actually
2966 * using their cgroups capability, we don't maintain the lists running
2967 * through each css_set to its tasks until we see the list actually
2968 * used - in other words after the first call to cgroup_iter_start().
2970 static void cgroup_enable_task_cg_lists(void)
2972 struct task_struct *p, *g;
2973 write_lock(&css_set_lock);
2974 use_task_css_set_links = 1;
2976 * We need tasklist_lock because RCU is not safe against
2977 * while_each_thread(). Besides, a forking task that has passed
2978 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2979 * is not guaranteed to have its child immediately visible in the
2980 * tasklist if we walk through it with RCU.
2982 read_lock(&tasklist_lock);
2983 do_each_thread(g, p) {
2986 * We should check if the process is exiting, otherwise
2987 * it will race with cgroup_exit() in that the list
2988 * entry won't be deleted though the process has exited.
2990 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2991 list_add(&p->cg_list, &p->cgroups->tasks);
2993 } while_each_thread(g, p);
2994 read_unlock(&tasklist_lock);
2995 write_unlock(&css_set_lock);
2999 * cgroup_next_sibling - find the next sibling of a given cgroup
3000 * @pos: the current cgroup
3002 * This function returns the next sibling of @pos and should be called
3003 * under RCU read lock. The only requirement is that @pos is accessible.
3004 * The next sibling is guaranteed to be returned regardless of @pos's
3007 struct cgroup *cgroup_next_sibling(struct cgroup *pos)
3009 struct cgroup *next;
3011 WARN_ON_ONCE(!rcu_read_lock_held());
3014 * @pos could already have been removed. Once a cgroup is removed,
3015 * its ->sibling.next is no longer updated when its next sibling
3016 * changes. As CGRP_DEAD assertion is serialized and happens
3017 * before the cgroup is taken off the ->sibling list, if we see it
3018 * unasserted, it's guaranteed that the next sibling hasn't
3019 * finished its grace period even if it's already removed, and thus
3020 * safe to dereference from this RCU critical section. If
3021 * ->sibling.next is inaccessible, cgroup_is_dead() is guaranteed
3022 * to be visible as %true here.
3024 if (likely(!cgroup_is_dead(pos))) {
3025 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3026 if (&next->sibling != &pos->parent->children)
3032 * Can't dereference the next pointer. Each cgroup is given a
3033 * monotonically increasing unique serial number and always
3034 * appended to the sibling list, so the next one can be found by
3035 * walking the parent's children until we see a cgroup with higher
3036 * serial number than @pos's.
3038 * While this path can be slow, it's taken only when either the
3039 * current cgroup is removed or iteration and removal race.
3041 list_for_each_entry_rcu(next, &pos->parent->children, sibling)
3042 if (next->serial_nr > pos->serial_nr)
3046 EXPORT_SYMBOL_GPL(cgroup_next_sibling);
3049 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
3050 * @pos: the current position (%NULL to initiate traversal)
3051 * @cgroup: cgroup whose descendants to walk
3053 * To be used by cgroup_for_each_descendant_pre(). Find the next
3054 * descendant to visit for pre-order traversal of @cgroup's descendants.
3056 * While this function requires RCU read locking, it doesn't require the
3057 * whole traversal to be contained in a single RCU critical section. This
3058 * function will return the correct next descendant as long as both @pos
3059 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3061 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
3062 struct cgroup *cgroup)
3064 struct cgroup *next;
3066 WARN_ON_ONCE(!rcu_read_lock_held());
3068 /* if first iteration, pretend we just visited @cgroup */
3072 /* visit the first child if exists */
3073 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
3077 /* no child, visit my or the closest ancestor's next sibling */
3078 while (pos != cgroup) {
3079 next = cgroup_next_sibling(pos);
3087 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
3090 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
3091 * @pos: cgroup of interest
3093 * Return the rightmost descendant of @pos. If there's no descendant,
3094 * @pos is returned. This can be used during pre-order traversal to skip
3097 * While this function requires RCU read locking, it doesn't require the
3098 * whole traversal to be contained in a single RCU critical section. This
3099 * function will return the correct rightmost descendant as long as @pos is
3102 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
3104 struct cgroup *last, *tmp;
3106 WARN_ON_ONCE(!rcu_read_lock_held());
3110 /* ->prev isn't RCU safe, walk ->next till the end */
3112 list_for_each_entry_rcu(tmp, &last->children, sibling)
3118 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
3120 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3122 struct cgroup *last;
3126 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3134 * cgroup_next_descendant_post - find the next descendant for post-order walk
3135 * @pos: the current position (%NULL to initiate traversal)
3136 * @cgroup: cgroup whose descendants to walk
3138 * To be used by cgroup_for_each_descendant_post(). Find the next
3139 * descendant to visit for post-order traversal of @cgroup's descendants.
3141 * While this function requires RCU read locking, it doesn't require the
3142 * whole traversal to be contained in a single RCU critical section. This
3143 * function will return the correct next descendant as long as both @pos
3144 * and @cgroup are accessible and @pos is a descendant of @cgroup.
3146 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3147 struct cgroup *cgroup)
3149 struct cgroup *next;
3151 WARN_ON_ONCE(!rcu_read_lock_held());
3153 /* if first iteration, visit the leftmost descendant */
3155 next = cgroup_leftmost_descendant(cgroup);
3156 return next != cgroup ? next : NULL;
3159 /* if there's an unvisited sibling, visit its leftmost descendant */
3160 next = cgroup_next_sibling(pos);
3162 return cgroup_leftmost_descendant(next);
3164 /* no sibling left, visit parent */
3166 return next != cgroup ? next : NULL;
3168 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3170 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3171 __acquires(css_set_lock)
3174 * The first time anyone tries to iterate across a cgroup,
3175 * we need to enable the list linking each css_set to its
3176 * tasks, and fix up all existing tasks.
3178 if (!use_task_css_set_links)
3179 cgroup_enable_task_cg_lists();
3181 read_lock(&css_set_lock);
3182 it->cset_link = &cgrp->cset_links;
3183 cgroup_advance_iter(cgrp, it);
3186 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3187 struct cgroup_iter *it)
3189 struct task_struct *res;
3190 struct list_head *l = it->task;
3191 struct cgrp_cset_link *link;
3193 /* If the iterator cg is NULL, we have no tasks */
3196 res = list_entry(l, struct task_struct, cg_list);
3197 /* Advance iterator to find next entry */
3199 link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link);
3200 if (l == &link->cset->tasks) {
3201 /* We reached the end of this task list - move on to
3202 * the next cg_cgroup_link */
3203 cgroup_advance_iter(cgrp, it);
3210 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3211 __releases(css_set_lock)
3213 read_unlock(&css_set_lock);
3216 static inline int started_after_time(struct task_struct *t1,
3217 struct timespec *time,
3218 struct task_struct *t2)
3220 int start_diff = timespec_compare(&t1->start_time, time);
3221 if (start_diff > 0) {
3223 } else if (start_diff < 0) {
3227 * Arbitrarily, if two processes started at the same
3228 * time, we'll say that the lower pointer value
3229 * started first. Note that t2 may have exited by now
3230 * so this may not be a valid pointer any longer, but
3231 * that's fine - it still serves to distinguish
3232 * between two tasks started (effectively) simultaneously.
3239 * This function is a callback from heap_insert() and is used to order
3241 * In this case we order the heap in descending task start time.
3243 static inline int started_after(void *p1, void *p2)
3245 struct task_struct *t1 = p1;
3246 struct task_struct *t2 = p2;
3247 return started_after_time(t1, &t2->start_time, t2);
3251 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3252 * @scan: struct cgroup_scanner containing arguments for the scan
3254 * Arguments include pointers to callback functions test_task() and
3256 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3257 * and if it returns true, call process_task() for it also.
3258 * The test_task pointer may be NULL, meaning always true (select all tasks).
3259 * Effectively duplicates cgroup_iter_{start,next,end}()
3260 * but does not lock css_set_lock for the call to process_task().
3261 * The struct cgroup_scanner may be embedded in any structure of the caller's
3263 * It is guaranteed that process_task() will act on every task that
3264 * is a member of the cgroup for the duration of this call. This
3265 * function may or may not call process_task() for tasks that exit
3266 * or move to a different cgroup during the call, or are forked or
3267 * move into the cgroup during the call.
3269 * Note that test_task() may be called with locks held, and may in some
3270 * situations be called multiple times for the same task, so it should
3272 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3273 * pre-allocated and will be used for heap operations (and its "gt" member will
3274 * be overwritten), else a temporary heap will be used (allocation of which
3275 * may cause this function to fail).
3277 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3280 struct cgroup_iter it;
3281 struct task_struct *p, *dropped;
3282 /* Never dereference latest_task, since it's not refcounted */
3283 struct task_struct *latest_task = NULL;
3284 struct ptr_heap tmp_heap;
3285 struct ptr_heap *heap;
3286 struct timespec latest_time = { 0, 0 };
3289 /* The caller supplied our heap and pre-allocated its memory */
3291 heap->gt = &started_after;
3293 /* We need to allocate our own heap memory */
3295 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3297 /* cannot allocate the heap */
3303 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3304 * to determine which are of interest, and using the scanner's
3305 * "process_task" callback to process any of them that need an update.
3306 * Since we don't want to hold any locks during the task updates,
3307 * gather tasks to be processed in a heap structure.
3308 * The heap is sorted by descending task start time.
3309 * If the statically-sized heap fills up, we overflow tasks that
3310 * started later, and in future iterations only consider tasks that
3311 * started after the latest task in the previous pass. This
3312 * guarantees forward progress and that we don't miss any tasks.
3315 cgroup_iter_start(scan->cg, &it);
3316 while ((p = cgroup_iter_next(scan->cg, &it))) {
3318 * Only affect tasks that qualify per the caller's callback,
3319 * if he provided one
3321 if (scan->test_task && !scan->test_task(p, scan))
3324 * Only process tasks that started after the last task
3327 if (!started_after_time(p, &latest_time, latest_task))
3329 dropped = heap_insert(heap, p);
3330 if (dropped == NULL) {
3332 * The new task was inserted; the heap wasn't
3336 } else if (dropped != p) {
3338 * The new task was inserted, and pushed out a
3342 put_task_struct(dropped);
3345 * Else the new task was newer than anything already in
3346 * the heap and wasn't inserted
3349 cgroup_iter_end(scan->cg, &it);
3352 for (i = 0; i < heap->size; i++) {
3353 struct task_struct *q = heap->ptrs[i];
3355 latest_time = q->start_time;
3358 /* Process the task per the caller's callback */
3359 scan->process_task(q, scan);
3363 * If we had to process any tasks at all, scan again
3364 * in case some of them were in the middle of forking
3365 * children that didn't get processed.
3366 * Not the most efficient way to do it, but it avoids
3367 * having to take callback_mutex in the fork path
3371 if (heap == &tmp_heap)
3372 heap_free(&tmp_heap);
3376 static void cgroup_transfer_one_task(struct task_struct *task,
3377 struct cgroup_scanner *scan)
3379 struct cgroup *new_cgroup = scan->data;
3381 mutex_lock(&cgroup_mutex);
3382 cgroup_attach_task(new_cgroup, task, false);
3383 mutex_unlock(&cgroup_mutex);
3387 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3388 * @to: cgroup to which the tasks will be moved
3389 * @from: cgroup in which the tasks currently reside
3391 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3393 struct cgroup_scanner scan;
3396 scan.test_task = NULL; /* select all tasks in cgroup */
3397 scan.process_task = cgroup_transfer_one_task;
3401 return cgroup_scan_tasks(&scan);
3405 * Stuff for reading the 'tasks'/'procs' files.
3407 * Reading this file can return large amounts of data if a cgroup has
3408 * *lots* of attached tasks. So it may need several calls to read(),
3409 * but we cannot guarantee that the information we produce is correct
3410 * unless we produce it entirely atomically.
3414 /* which pidlist file are we talking about? */
3415 enum cgroup_filetype {
3421 * A pidlist is a list of pids that virtually represents the contents of one
3422 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3423 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3426 struct cgroup_pidlist {
3428 * used to find which pidlist is wanted. doesn't change as long as
3429 * this particular list stays in the list.
3431 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3434 /* how many elements the above list has */
3436 /* how many files are using the current array */
3438 /* each of these stored in a list by its cgroup */
3439 struct list_head links;
3440 /* pointer to the cgroup we belong to, for list removal purposes */
3441 struct cgroup *owner;
3442 /* protects the other fields */
3443 struct rw_semaphore mutex;
3447 * The following two functions "fix" the issue where there are more pids
3448 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3449 * TODO: replace with a kernel-wide solution to this problem
3451 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3452 static void *pidlist_allocate(int count)
3454 if (PIDLIST_TOO_LARGE(count))
3455 return vmalloc(count * sizeof(pid_t));
3457 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3459 static void pidlist_free(void *p)
3461 if (is_vmalloc_addr(p))
3468 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3469 * Returns the number of unique elements.
3471 static int pidlist_uniq(pid_t *list, int length)
3476 * we presume the 0th element is unique, so i starts at 1. trivial
3477 * edge cases first; no work needs to be done for either
3479 if (length == 0 || length == 1)
3481 /* src and dest walk down the list; dest counts unique elements */
3482 for (src = 1; src < length; src++) {
3483 /* find next unique element */
3484 while (list[src] == list[src-1]) {
3489 /* dest always points to where the next unique element goes */
3490 list[dest] = list[src];
3497 static int cmppid(const void *a, const void *b)
3499 return *(pid_t *)a - *(pid_t *)b;
3503 * find the appropriate pidlist for our purpose (given procs vs tasks)
3504 * returns with the lock on that pidlist already held, and takes care
3505 * of the use count, or returns NULL with no locks held if we're out of
3508 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3509 enum cgroup_filetype type)
3511 struct cgroup_pidlist *l;
3512 /* don't need task_nsproxy() if we're looking at ourself */
3513 struct pid_namespace *ns = task_active_pid_ns(current);
3516 * We can't drop the pidlist_mutex before taking the l->mutex in case
3517 * the last ref-holder is trying to remove l from the list at the same
3518 * time. Holding the pidlist_mutex precludes somebody taking whichever
3519 * list we find out from under us - compare release_pid_array().
3521 mutex_lock(&cgrp->pidlist_mutex);
3522 list_for_each_entry(l, &cgrp->pidlists, links) {
3523 if (l->key.type == type && l->key.ns == ns) {
3524 /* make sure l doesn't vanish out from under us */
3525 down_write(&l->mutex);
3526 mutex_unlock(&cgrp->pidlist_mutex);
3530 /* entry not found; create a new one */
3531 l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3533 mutex_unlock(&cgrp->pidlist_mutex);
3536 init_rwsem(&l->mutex);
3537 down_write(&l->mutex);
3539 l->key.ns = get_pid_ns(ns);
3541 list_add(&l->links, &cgrp->pidlists);
3542 mutex_unlock(&cgrp->pidlist_mutex);
3547 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3549 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3550 struct cgroup_pidlist **lp)
3554 int pid, n = 0; /* used for populating the array */
3555 struct cgroup_iter it;
3556 struct task_struct *tsk;
3557 struct cgroup_pidlist *l;
3560 * If cgroup gets more users after we read count, we won't have
3561 * enough space - tough. This race is indistinguishable to the
3562 * caller from the case that the additional cgroup users didn't
3563 * show up until sometime later on.
3565 length = cgroup_task_count(cgrp);
3566 array = pidlist_allocate(length);
3569 /* now, populate the array */
3570 cgroup_iter_start(cgrp, &it);
3571 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3572 if (unlikely(n == length))
3574 /* get tgid or pid for procs or tasks file respectively */
3575 if (type == CGROUP_FILE_PROCS)
3576 pid = task_tgid_vnr(tsk);
3578 pid = task_pid_vnr(tsk);
3579 if (pid > 0) /* make sure to only use valid results */
3582 cgroup_iter_end(cgrp, &it);
3584 /* now sort & (if procs) strip out duplicates */
3585 sort(array, length, sizeof(pid_t), cmppid, NULL);
3586 if (type == CGROUP_FILE_PROCS)
3587 length = pidlist_uniq(array, length);
3588 l = cgroup_pidlist_find(cgrp, type);
3590 pidlist_free(array);
3593 /* store array, freeing old if necessary - lock already held */
3594 pidlist_free(l->list);
3598 up_write(&l->mutex);
3604 * cgroupstats_build - build and fill cgroupstats
3605 * @stats: cgroupstats to fill information into
3606 * @dentry: A dentry entry belonging to the cgroup for which stats have
3609 * Build and fill cgroupstats so that taskstats can export it to user
3612 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3615 struct cgroup *cgrp;
3616 struct cgroup_iter it;
3617 struct task_struct *tsk;
3620 * Validate dentry by checking the superblock operations,
3621 * and make sure it's a directory.
3623 if (dentry->d_sb->s_op != &cgroup_ops ||
3624 !S_ISDIR(dentry->d_inode->i_mode))
3628 cgrp = dentry->d_fsdata;
3630 cgroup_iter_start(cgrp, &it);
3631 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3632 switch (tsk->state) {
3634 stats->nr_running++;
3636 case TASK_INTERRUPTIBLE:
3637 stats->nr_sleeping++;
3639 case TASK_UNINTERRUPTIBLE:
3640 stats->nr_uninterruptible++;
3643 stats->nr_stopped++;
3646 if (delayacct_is_task_waiting_on_io(tsk))
3647 stats->nr_io_wait++;
3651 cgroup_iter_end(cgrp, &it);
3659 * seq_file methods for the tasks/procs files. The seq_file position is the
3660 * next pid to display; the seq_file iterator is a pointer to the pid
3661 * in the cgroup->l->list array.
3664 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3667 * Initially we receive a position value that corresponds to
3668 * one more than the last pid shown (or 0 on the first call or
3669 * after a seek to the start). Use a binary-search to find the
3670 * next pid to display, if any
3672 struct cgroup_pidlist *l = s->private;
3673 int index = 0, pid = *pos;
3676 down_read(&l->mutex);
3678 int end = l->length;
3680 while (index < end) {
3681 int mid = (index + end) / 2;
3682 if (l->list[mid] == pid) {
3685 } else if (l->list[mid] <= pid)
3691 /* If we're off the end of the array, we're done */
3692 if (index >= l->length)
3694 /* Update the abstract position to be the actual pid that we found */
3695 iter = l->list + index;
3700 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3702 struct cgroup_pidlist *l = s->private;
3706 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3708 struct cgroup_pidlist *l = s->private;
3710 pid_t *end = l->list + l->length;
3712 * Advance to the next pid in the array. If this goes off the
3724 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3726 return seq_printf(s, "%d\n", *(int *)v);
3730 * seq_operations functions for iterating on pidlists through seq_file -
3731 * independent of whether it's tasks or procs
3733 static const struct seq_operations cgroup_pidlist_seq_operations = {
3734 .start = cgroup_pidlist_start,
3735 .stop = cgroup_pidlist_stop,
3736 .next = cgroup_pidlist_next,
3737 .show = cgroup_pidlist_show,
3740 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3743 * the case where we're the last user of this particular pidlist will
3744 * have us remove it from the cgroup's list, which entails taking the
3745 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3746 * pidlist_mutex, we have to take pidlist_mutex first.
3748 mutex_lock(&l->owner->pidlist_mutex);
3749 down_write(&l->mutex);
3750 BUG_ON(!l->use_count);
3751 if (!--l->use_count) {
3752 /* we're the last user if refcount is 0; remove and free */
3753 list_del(&l->links);
3754 mutex_unlock(&l->owner->pidlist_mutex);
3755 pidlist_free(l->list);
3756 put_pid_ns(l->key.ns);
3757 up_write(&l->mutex);
3761 mutex_unlock(&l->owner->pidlist_mutex);
3762 up_write(&l->mutex);
3765 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3767 struct cgroup_pidlist *l;
3768 if (!(file->f_mode & FMODE_READ))
3771 * the seq_file will only be initialized if the file was opened for
3772 * reading; hence we check if it's not null only in that case.
3774 l = ((struct seq_file *)file->private_data)->private;
3775 cgroup_release_pid_array(l);
3776 return seq_release(inode, file);
3779 static const struct file_operations cgroup_pidlist_operations = {
3781 .llseek = seq_lseek,
3782 .write = cgroup_file_write,
3783 .release = cgroup_pidlist_release,
3787 * The following functions handle opens on a file that displays a pidlist
3788 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3791 /* helper function for the two below it */
3792 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3794 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3795 struct cgroup_pidlist *l;
3798 /* Nothing to do for write-only files */
3799 if (!(file->f_mode & FMODE_READ))
3802 /* have the array populated */
3803 retval = pidlist_array_load(cgrp, type, &l);
3806 /* configure file information */
3807 file->f_op = &cgroup_pidlist_operations;
3809 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3811 cgroup_release_pid_array(l);
3814 ((struct seq_file *)file->private_data)->private = l;
3817 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3819 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3821 static int cgroup_procs_open(struct inode *unused, struct file *file)
3823 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3826 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3829 return notify_on_release(cgrp);
3832 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3836 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3838 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3840 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3845 * When dput() is called asynchronously, if umount has been done and
3846 * then deactivate_super() in cgroup_free_fn() kills the superblock,
3847 * there's a small window that vfs will see the root dentry with non-zero
3848 * refcnt and trigger BUG().
3850 * That's why we hold a reference before dput() and drop it right after.
3852 static void cgroup_dput(struct cgroup *cgrp)
3854 struct super_block *sb = cgrp->root->sb;
3856 atomic_inc(&sb->s_active);
3858 deactivate_super(sb);
3862 * Unregister event and free resources.
3864 * Gets called from workqueue.
3866 static void cgroup_event_remove(struct work_struct *work)
3868 struct cgroup_event *event = container_of(work, struct cgroup_event,
3870 struct cgroup *cgrp = event->cgrp;
3872 remove_wait_queue(event->wqh, &event->wait);
3874 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3876 /* Notify userspace the event is going away. */
3877 eventfd_signal(event->eventfd, 1);
3879 eventfd_ctx_put(event->eventfd);
3885 * Gets called on POLLHUP on eventfd when user closes it.
3887 * Called with wqh->lock held and interrupts disabled.
3889 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3890 int sync, void *key)
3892 struct cgroup_event *event = container_of(wait,
3893 struct cgroup_event, wait);
3894 struct cgroup *cgrp = event->cgrp;
3895 unsigned long flags = (unsigned long)key;
3897 if (flags & POLLHUP) {
3899 * If the event has been detached at cgroup removal, we
3900 * can simply return knowing the other side will cleanup
3903 * We can't race against event freeing since the other
3904 * side will require wqh->lock via remove_wait_queue(),
3907 spin_lock(&cgrp->event_list_lock);
3908 if (!list_empty(&event->list)) {
3909 list_del_init(&event->list);
3911 * We are in atomic context, but cgroup_event_remove()
3912 * may sleep, so we have to call it in workqueue.
3914 schedule_work(&event->remove);
3916 spin_unlock(&cgrp->event_list_lock);
3922 static void cgroup_event_ptable_queue_proc(struct file *file,
3923 wait_queue_head_t *wqh, poll_table *pt)
3925 struct cgroup_event *event = container_of(pt,
3926 struct cgroup_event, pt);
3929 add_wait_queue(wqh, &event->wait);
3933 * Parse input and register new cgroup event handler.
3935 * Input must be in format '<event_fd> <control_fd> <args>'.
3936 * Interpretation of args is defined by control file implementation.
3938 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3941 struct cgroup_event *event = NULL;
3942 struct cgroup *cgrp_cfile;
3943 unsigned int efd, cfd;
3944 struct file *efile = NULL;
3945 struct file *cfile = NULL;
3949 efd = simple_strtoul(buffer, &endp, 10);
3954 cfd = simple_strtoul(buffer, &endp, 10);
3955 if ((*endp != ' ') && (*endp != '\0'))
3959 event = kzalloc(sizeof(*event), GFP_KERNEL);
3963 INIT_LIST_HEAD(&event->list);
3964 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3965 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3966 INIT_WORK(&event->remove, cgroup_event_remove);
3968 efile = eventfd_fget(efd);
3969 if (IS_ERR(efile)) {
3970 ret = PTR_ERR(efile);
3974 event->eventfd = eventfd_ctx_fileget(efile);
3975 if (IS_ERR(event->eventfd)) {
3976 ret = PTR_ERR(event->eventfd);
3986 /* the process need read permission on control file */
3987 /* AV: shouldn't we check that it's been opened for read instead? */
3988 ret = inode_permission(file_inode(cfile), MAY_READ);
3992 event->cft = __file_cft(cfile);
3993 if (IS_ERR(event->cft)) {
3994 ret = PTR_ERR(event->cft);
3999 * The file to be monitored must be in the same cgroup as
4000 * cgroup.event_control is.
4002 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
4003 if (cgrp_cfile != cgrp) {
4008 if (!event->cft->register_event || !event->cft->unregister_event) {
4013 ret = event->cft->register_event(cgrp, event->cft,
4014 event->eventfd, buffer);
4018 efile->f_op->poll(efile, &event->pt);
4021 * Events should be removed after rmdir of cgroup directory, but before
4022 * destroying subsystem state objects. Let's take reference to cgroup
4023 * directory dentry to do that.
4027 spin_lock(&cgrp->event_list_lock);
4028 list_add(&event->list, &cgrp->event_list);
4029 spin_unlock(&cgrp->event_list_lock);
4040 if (event && event->eventfd && !IS_ERR(event->eventfd))
4041 eventfd_ctx_put(event->eventfd);
4043 if (!IS_ERR_OR_NULL(efile))
4051 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
4054 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4057 static int cgroup_clone_children_write(struct cgroup *cgrp,
4062 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4064 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4068 static struct cftype cgroup_base_files[] = {
4070 .name = "cgroup.procs",
4071 .open = cgroup_procs_open,
4072 .write_u64 = cgroup_procs_write,
4073 .release = cgroup_pidlist_release,
4074 .mode = S_IRUGO | S_IWUSR,
4077 .name = "cgroup.event_control",
4078 .write_string = cgroup_write_event_control,
4082 .name = "cgroup.clone_children",
4083 .flags = CFTYPE_INSANE,
4084 .read_u64 = cgroup_clone_children_read,
4085 .write_u64 = cgroup_clone_children_write,
4088 .name = "cgroup.sane_behavior",
4089 .flags = CFTYPE_ONLY_ON_ROOT,
4090 .read_seq_string = cgroup_sane_behavior_show,
4094 * Historical crazy stuff. These don't have "cgroup." prefix and
4095 * don't exist if sane_behavior. If you're depending on these, be
4096 * prepared to be burned.
4100 .flags = CFTYPE_INSANE, /* use "procs" instead */
4101 .open = cgroup_tasks_open,
4102 .write_u64 = cgroup_tasks_write,
4103 .release = cgroup_pidlist_release,
4104 .mode = S_IRUGO | S_IWUSR,
4107 .name = "notify_on_release",
4108 .flags = CFTYPE_INSANE,
4109 .read_u64 = cgroup_read_notify_on_release,
4110 .write_u64 = cgroup_write_notify_on_release,
4113 .name = "release_agent",
4114 .flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT,
4115 .read_seq_string = cgroup_release_agent_show,
4116 .write_string = cgroup_release_agent_write,
4117 .max_write_len = PATH_MAX,
4123 * cgroup_populate_dir - selectively creation of files in a directory
4124 * @cgrp: target cgroup
4125 * @base_files: true if the base files should be added
4126 * @subsys_mask: mask of the subsystem ids whose files should be added
4128 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
4129 unsigned long subsys_mask)
4132 struct cgroup_subsys *ss;
4135 err = cgroup_addrm_files(cgrp, NULL, cgroup_base_files, true);
4140 /* process cftsets of each subsystem */
4141 for_each_subsys(cgrp->root, ss) {
4142 struct cftype_set *set;
4143 if (!test_bit(ss->subsys_id, &subsys_mask))
4146 list_for_each_entry(set, &ss->cftsets, node)
4147 cgroup_addrm_files(cgrp, ss, set->cfts, true);
4150 /* This cgroup is ready now */
4151 for_each_subsys(cgrp->root, ss) {
4152 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4154 * Update id->css pointer and make this css visible from
4155 * CSS ID functions. This pointer will be dereferened
4156 * from RCU-read-side without locks.
4159 rcu_assign_pointer(css->id->css, css);
4165 static void css_dput_fn(struct work_struct *work)
4167 struct cgroup_subsys_state *css =
4168 container_of(work, struct cgroup_subsys_state, dput_work);
4170 cgroup_dput(css->cgroup);
4173 static void css_release(struct percpu_ref *ref)
4175 struct cgroup_subsys_state *css =
4176 container_of(ref, struct cgroup_subsys_state, refcnt);
4178 schedule_work(&css->dput_work);
4181 static void init_cgroup_css(struct cgroup_subsys_state *css,
4182 struct cgroup_subsys *ss,
4183 struct cgroup *cgrp)
4188 if (cgrp == dummytop)
4189 css->flags |= CSS_ROOT;
4190 BUG_ON(cgrp->subsys[ss->subsys_id]);
4191 cgrp->subsys[ss->subsys_id] = css;
4194 * css holds an extra ref to @cgrp->dentry which is put on the last
4195 * css_put(). dput() requires process context, which css_put() may
4196 * be called without. @css->dput_work will be used to invoke
4197 * dput() asynchronously from css_put().
4199 INIT_WORK(&css->dput_work, css_dput_fn);
4202 /* invoke ->post_create() on a new CSS and mark it online if successful */
4203 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4207 lockdep_assert_held(&cgroup_mutex);
4210 ret = ss->css_online(cgrp);
4212 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4216 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4217 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4218 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4220 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4222 lockdep_assert_held(&cgroup_mutex);
4224 if (!(css->flags & CSS_ONLINE))
4227 if (ss->css_offline)
4228 ss->css_offline(cgrp);
4230 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4234 * cgroup_create - create a cgroup
4235 * @parent: cgroup that will be parent of the new cgroup
4236 * @dentry: dentry of the new cgroup
4237 * @mode: mode to set on new inode
4239 * Must be called with the mutex on the parent inode held
4241 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4244 struct cgroup *cgrp;
4245 struct cgroup_name *name;
4246 struct cgroupfs_root *root = parent->root;
4248 struct cgroup_subsys *ss;
4249 struct super_block *sb = root->sb;
4251 /* allocate the cgroup and its ID, 0 is reserved for the root */
4252 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4256 name = cgroup_alloc_name(dentry);
4259 rcu_assign_pointer(cgrp->name, name);
4261 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4266 * Only live parents can have children. Note that the liveliness
4267 * check isn't strictly necessary because cgroup_mkdir() and
4268 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4269 * anyway so that locking is contained inside cgroup proper and we
4270 * don't get nasty surprises if we ever grow another caller.
4272 if (!cgroup_lock_live_group(parent)) {
4277 /* Grab a reference on the superblock so the hierarchy doesn't
4278 * get deleted on unmount if there are child cgroups. This
4279 * can be done outside cgroup_mutex, since the sb can't
4280 * disappear while someone has an open control file on the
4282 atomic_inc(&sb->s_active);
4284 init_cgroup_housekeeping(cgrp);
4286 dentry->d_fsdata = cgrp;
4287 cgrp->dentry = dentry;
4289 cgrp->parent = parent;
4290 cgrp->root = parent->root;
4292 if (notify_on_release(parent))
4293 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4295 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4296 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4298 for_each_subsys(root, ss) {
4299 struct cgroup_subsys_state *css;
4301 css = ss->css_alloc(cgrp);
4307 err = percpu_ref_init(&css->refcnt, css_release);
4311 init_cgroup_css(css, ss, cgrp);
4314 err = alloc_css_id(ss, parent, cgrp);
4321 * Create directory. cgroup_create_file() returns with the new
4322 * directory locked on success so that it can be populated without
4323 * dropping cgroup_mutex.
4325 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4328 lockdep_assert_held(&dentry->d_inode->i_mutex);
4330 cgrp->serial_nr = atomic64_inc_return(&cgroup_serial_nr_cursor);
4332 /* allocation complete, commit to creation */
4333 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4334 root->number_of_cgroups++;
4336 /* each css holds a ref to the cgroup's dentry */
4337 for_each_subsys(root, ss)
4340 /* hold a ref to the parent's dentry */
4341 dget(parent->dentry);
4343 /* creation succeeded, notify subsystems */
4344 for_each_subsys(root, ss) {
4345 err = online_css(ss, cgrp);
4349 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4351 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",
4352 current->comm, current->pid, ss->name);
4353 if (!strcmp(ss->name, "memory"))
4354 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4355 ss->warned_broken_hierarchy = true;
4359 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4363 mutex_unlock(&cgroup_mutex);
4364 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4369 for_each_subsys(root, ss) {
4370 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4373 percpu_ref_cancel_init(&css->refcnt);
4377 mutex_unlock(&cgroup_mutex);
4378 /* Release the reference count that we took on the superblock */
4379 deactivate_super(sb);
4381 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4383 kfree(rcu_dereference_raw(cgrp->name));
4389 cgroup_destroy_locked(cgrp);
4390 mutex_unlock(&cgroup_mutex);
4391 mutex_unlock(&dentry->d_inode->i_mutex);
4395 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4397 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4399 /* the vfs holds inode->i_mutex already */
4400 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4403 static void cgroup_css_killed(struct cgroup *cgrp)
4405 if (!atomic_dec_and_test(&cgrp->css_kill_cnt))
4408 /* percpu ref's of all css's are killed, kick off the next step */
4409 INIT_WORK(&cgrp->destroy_work, cgroup_offline_fn);
4410 schedule_work(&cgrp->destroy_work);
4413 static void css_ref_killed_fn(struct percpu_ref *ref)
4415 struct cgroup_subsys_state *css =
4416 container_of(ref, struct cgroup_subsys_state, refcnt);
4418 cgroup_css_killed(css->cgroup);
4422 * cgroup_destroy_locked - the first stage of cgroup destruction
4423 * @cgrp: cgroup to be destroyed
4425 * css's make use of percpu refcnts whose killing latency shouldn't be
4426 * exposed to userland and are RCU protected. Also, cgroup core needs to
4427 * guarantee that css_tryget() won't succeed by the time ->css_offline() is
4428 * invoked. To satisfy all the requirements, destruction is implemented in
4429 * the following two steps.
4431 * s1. Verify @cgrp can be destroyed and mark it dying. Remove all
4432 * userland visible parts and start killing the percpu refcnts of
4433 * css's. Set up so that the next stage will be kicked off once all
4434 * the percpu refcnts are confirmed to be killed.
4436 * s2. Invoke ->css_offline(), mark the cgroup dead and proceed with the
4437 * rest of destruction. Once all cgroup references are gone, the
4438 * cgroup is RCU-freed.
4440 * This function implements s1. After this step, @cgrp is gone as far as
4441 * the userland is concerned and a new cgroup with the same name may be
4442 * created. As cgroup doesn't care about the names internally, this
4443 * doesn't cause any problem.
4445 static int cgroup_destroy_locked(struct cgroup *cgrp)
4446 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4448 struct dentry *d = cgrp->dentry;
4449 struct cgroup_event *event, *tmp;
4450 struct cgroup_subsys *ss;
4453 lockdep_assert_held(&d->d_inode->i_mutex);
4454 lockdep_assert_held(&cgroup_mutex);
4457 * css_set_lock synchronizes access to ->cset_links and prevents
4458 * @cgrp from being removed while __put_css_set() is in progress.
4460 read_lock(&css_set_lock);
4461 empty = list_empty(&cgrp->cset_links) && list_empty(&cgrp->children);
4462 read_unlock(&css_set_lock);
4467 * Block new css_tryget() by killing css refcnts. cgroup core
4468 * guarantees that, by the time ->css_offline() is invoked, no new
4469 * css reference will be given out via css_tryget(). We can't
4470 * simply call percpu_ref_kill() and proceed to offlining css's
4471 * because percpu_ref_kill() doesn't guarantee that the ref is seen
4472 * as killed on all CPUs on return.
4474 * Use percpu_ref_kill_and_confirm() to get notifications as each
4475 * css is confirmed to be seen as killed on all CPUs. The
4476 * notification callback keeps track of the number of css's to be
4477 * killed and schedules cgroup_offline_fn() to perform the rest of
4478 * destruction once the percpu refs of all css's are confirmed to
4481 atomic_set(&cgrp->css_kill_cnt, 1);
4482 for_each_subsys(cgrp->root, ss) {
4483 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4486 * Killing would put the base ref, but we need to keep it
4487 * alive until after ->css_offline.
4489 percpu_ref_get(&css->refcnt);
4491 atomic_inc(&cgrp->css_kill_cnt);
4492 percpu_ref_kill_and_confirm(&css->refcnt, css_ref_killed_fn);
4494 cgroup_css_killed(cgrp);
4497 * Mark @cgrp dead. This prevents further task migration and child
4498 * creation by disabling cgroup_lock_live_group(). Note that
4499 * CGRP_DEAD assertion is depended upon by cgroup_next_sibling() to
4500 * resume iteration after dropping RCU read lock. See
4501 * cgroup_next_sibling() for details.
4503 set_bit(CGRP_DEAD, &cgrp->flags);
4505 /* CGRP_DEAD is set, remove from ->release_list for the last time */
4506 raw_spin_lock(&release_list_lock);
4507 if (!list_empty(&cgrp->release_list))
4508 list_del_init(&cgrp->release_list);
4509 raw_spin_unlock(&release_list_lock);
4512 * Remove @cgrp directory. The removal puts the base ref but we
4513 * aren't quite done with @cgrp yet, so hold onto it.
4516 cgroup_d_remove_dir(d);
4519 * Unregister events and notify userspace.
4520 * Notify userspace about cgroup removing only after rmdir of cgroup
4521 * directory to avoid race between userspace and kernelspace.
4523 spin_lock(&cgrp->event_list_lock);
4524 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4525 list_del_init(&event->list);
4526 schedule_work(&event->remove);
4528 spin_unlock(&cgrp->event_list_lock);
4534 * cgroup_offline_fn - the second step of cgroup destruction
4535 * @work: cgroup->destroy_free_work
4537 * This function is invoked from a work item for a cgroup which is being
4538 * destroyed after the percpu refcnts of all css's are guaranteed to be
4539 * seen as killed on all CPUs, and performs the rest of destruction. This
4540 * is the second step of destruction described in the comment above
4541 * cgroup_destroy_locked().
4543 static void cgroup_offline_fn(struct work_struct *work)
4545 struct cgroup *cgrp = container_of(work, struct cgroup, destroy_work);
4546 struct cgroup *parent = cgrp->parent;
4547 struct dentry *d = cgrp->dentry;
4548 struct cgroup_subsys *ss;
4550 mutex_lock(&cgroup_mutex);
4553 * css_tryget() is guaranteed to fail now. Tell subsystems to
4554 * initate destruction.
4556 for_each_subsys(cgrp->root, ss)
4557 offline_css(ss, cgrp);
4560 * Put the css refs from cgroup_destroy_locked(). Each css holds
4561 * an extra reference to the cgroup's dentry and cgroup removal
4562 * proceeds regardless of css refs. On the last put of each css,
4563 * whenever that may be, the extra dentry ref is put so that dentry
4564 * destruction happens only after all css's are released.
4566 for_each_subsys(cgrp->root, ss)
4567 css_put(cgrp->subsys[ss->subsys_id]);
4569 /* delete this cgroup from parent->children */
4570 list_del_rcu(&cgrp->sibling);
4574 set_bit(CGRP_RELEASABLE, &parent->flags);
4575 check_for_release(parent);
4577 mutex_unlock(&cgroup_mutex);
4580 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4584 mutex_lock(&cgroup_mutex);
4585 ret = cgroup_destroy_locked(dentry->d_fsdata);
4586 mutex_unlock(&cgroup_mutex);
4591 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4593 INIT_LIST_HEAD(&ss->cftsets);
4596 * base_cftset is embedded in subsys itself, no need to worry about
4599 if (ss->base_cftypes) {
4600 ss->base_cftset.cfts = ss->base_cftypes;
4601 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4605 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4607 struct cgroup_subsys_state *css;
4609 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4611 mutex_lock(&cgroup_mutex);
4613 /* init base cftset */
4614 cgroup_init_cftsets(ss);
4616 /* Create the top cgroup state for this subsystem */
4617 list_add(&ss->sibling, &rootnode.subsys_list);
4618 ss->root = &rootnode;
4619 css = ss->css_alloc(dummytop);
4620 /* We don't handle early failures gracefully */
4621 BUG_ON(IS_ERR(css));
4622 init_cgroup_css(css, ss, dummytop);
4624 /* Update the init_css_set to contain a subsys
4625 * pointer to this state - since the subsystem is
4626 * newly registered, all tasks and hence the
4627 * init_css_set is in the subsystem's top cgroup. */
4628 init_css_set.subsys[ss->subsys_id] = css;
4630 need_forkexit_callback |= ss->fork || ss->exit;
4632 /* At system boot, before all subsystems have been
4633 * registered, no tasks have been forked, so we don't
4634 * need to invoke fork callbacks here. */
4635 BUG_ON(!list_empty(&init_task.tasks));
4637 BUG_ON(online_css(ss, dummytop));
4639 mutex_unlock(&cgroup_mutex);
4641 /* this function shouldn't be used with modular subsystems, since they
4642 * need to register a subsys_id, among other things */
4647 * cgroup_load_subsys: load and register a modular subsystem at runtime
4648 * @ss: the subsystem to load
4650 * This function should be called in a modular subsystem's initcall. If the
4651 * subsystem is built as a module, it will be assigned a new subsys_id and set
4652 * up for use. If the subsystem is built-in anyway, work is delegated to the
4653 * simpler cgroup_init_subsys.
4655 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4657 struct cgroup_subsys_state *css;
4659 struct hlist_node *tmp;
4660 struct css_set *cset;
4663 /* check name and function validity */
4664 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4665 ss->css_alloc == NULL || ss->css_free == NULL)
4669 * we don't support callbacks in modular subsystems. this check is
4670 * before the ss->module check for consistency; a subsystem that could
4671 * be a module should still have no callbacks even if the user isn't
4672 * compiling it as one.
4674 if (ss->fork || ss->exit)
4678 * an optionally modular subsystem is built-in: we want to do nothing,
4679 * since cgroup_init_subsys will have already taken care of it.
4681 if (ss->module == NULL) {
4682 /* a sanity check */
4683 BUG_ON(subsys[ss->subsys_id] != ss);
4687 /* init base cftset */
4688 cgroup_init_cftsets(ss);
4690 mutex_lock(&cgroup_mutex);
4691 subsys[ss->subsys_id] = ss;
4694 * no ss->css_alloc seems to need anything important in the ss
4695 * struct, so this can happen first (i.e. before the rootnode
4698 css = ss->css_alloc(dummytop);
4700 /* failure case - need to deassign the subsys[] slot. */
4701 subsys[ss->subsys_id] = NULL;
4702 mutex_unlock(&cgroup_mutex);
4703 return PTR_ERR(css);
4706 list_add(&ss->sibling, &rootnode.subsys_list);
4707 ss->root = &rootnode;
4709 /* our new subsystem will be attached to the dummy hierarchy. */
4710 init_cgroup_css(css, ss, dummytop);
4711 /* init_idr must be after init_cgroup_css because it sets css->id. */
4713 ret = cgroup_init_idr(ss, css);
4719 * Now we need to entangle the css into the existing css_sets. unlike
4720 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4721 * will need a new pointer to it; done by iterating the css_set_table.
4722 * furthermore, modifying the existing css_sets will corrupt the hash
4723 * table state, so each changed css_set will need its hash recomputed.
4724 * this is all done under the css_set_lock.
4726 write_lock(&css_set_lock);
4727 hash_for_each_safe(css_set_table, i, tmp, cset, hlist) {
4728 /* skip entries that we already rehashed */
4729 if (cset->subsys[ss->subsys_id])
4731 /* remove existing entry */
4732 hash_del(&cset->hlist);
4734 cset->subsys[ss->subsys_id] = css;
4735 /* recompute hash and restore entry */
4736 key = css_set_hash(cset->subsys);
4737 hash_add(css_set_table, &cset->hlist, key);
4739 write_unlock(&css_set_lock);
4741 ret = online_css(ss, dummytop);
4746 mutex_unlock(&cgroup_mutex);
4750 mutex_unlock(&cgroup_mutex);
4751 /* @ss can't be mounted here as try_module_get() would fail */
4752 cgroup_unload_subsys(ss);
4755 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4758 * cgroup_unload_subsys: unload a modular subsystem
4759 * @ss: the subsystem to unload
4761 * This function should be called in a modular subsystem's exitcall. When this
4762 * function is invoked, the refcount on the subsystem's module will be 0, so
4763 * the subsystem will not be attached to any hierarchy.
4765 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4767 struct cgrp_cset_link *link;
4769 BUG_ON(ss->module == NULL);
4772 * we shouldn't be called if the subsystem is in use, and the use of
4773 * try_module_get in parse_cgroupfs_options should ensure that it
4774 * doesn't start being used while we're killing it off.
4776 BUG_ON(ss->root != &rootnode);
4778 mutex_lock(&cgroup_mutex);
4780 offline_css(ss, dummytop);
4783 idr_destroy(&ss->idr);
4785 /* deassign the subsys_id */
4786 subsys[ss->subsys_id] = NULL;
4788 /* remove subsystem from rootnode's list of subsystems */
4789 list_del_init(&ss->sibling);
4792 * disentangle the css from all css_sets attached to the dummytop. as
4793 * in loading, we need to pay our respects to the hashtable gods.
4795 write_lock(&css_set_lock);
4796 list_for_each_entry(link, &dummytop->cset_links, cset_link) {
4797 struct css_set *cset = link->cset;
4800 hash_del(&cset->hlist);
4801 cset->subsys[ss->subsys_id] = NULL;
4802 key = css_set_hash(cset->subsys);
4803 hash_add(css_set_table, &cset->hlist, key);
4805 write_unlock(&css_set_lock);
4808 * remove subsystem's css from the dummytop and free it - need to
4809 * free before marking as null because ss->css_free needs the
4810 * cgrp->subsys pointer to find their state. note that this also
4811 * takes care of freeing the css_id.
4813 ss->css_free(dummytop);
4814 dummytop->subsys[ss->subsys_id] = NULL;
4816 mutex_unlock(&cgroup_mutex);
4818 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4821 * cgroup_init_early - cgroup initialization at system boot
4823 * Initialize cgroups at system boot, and initialize any
4824 * subsystems that request early init.
4826 int __init cgroup_init_early(void)
4829 atomic_set(&init_css_set.refcount, 1);
4830 INIT_LIST_HEAD(&init_css_set.cgrp_links);
4831 INIT_LIST_HEAD(&init_css_set.tasks);
4832 INIT_HLIST_NODE(&init_css_set.hlist);
4834 init_cgroup_root(&rootnode);
4836 init_task.cgroups = &init_css_set;
4838 init_cgrp_cset_link.cset = &init_css_set;
4839 init_cgrp_cset_link.cgrp = dummytop;
4840 list_add(&init_cgrp_cset_link.cset_link, &rootnode.top_cgroup.cset_links);
4841 list_add(&init_cgrp_cset_link.cgrp_link, &init_css_set.cgrp_links);
4843 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4844 struct cgroup_subsys *ss = subsys[i];
4846 /* at bootup time, we don't worry about modular subsystems */
4847 if (!ss || ss->module)
4851 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4852 BUG_ON(!ss->css_alloc);
4853 BUG_ON(!ss->css_free);
4854 if (ss->subsys_id != i) {
4855 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4856 ss->name, ss->subsys_id);
4861 cgroup_init_subsys(ss);
4867 * cgroup_init - cgroup initialization
4869 * Register cgroup filesystem and /proc file, and initialize
4870 * any subsystems that didn't request early init.
4872 int __init cgroup_init(void)
4878 err = bdi_init(&cgroup_backing_dev_info);
4882 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4883 struct cgroup_subsys *ss = subsys[i];
4885 /* at bootup time, we don't worry about modular subsystems */
4886 if (!ss || ss->module)
4888 if (!ss->early_init)
4889 cgroup_init_subsys(ss);
4891 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4894 /* Add init_css_set to the hash table */
4895 key = css_set_hash(init_css_set.subsys);
4896 hash_add(css_set_table, &init_css_set.hlist, key);
4898 /* allocate id for the dummy hierarchy */
4899 mutex_lock(&cgroup_mutex);
4900 mutex_lock(&cgroup_root_mutex);
4902 BUG_ON(cgroup_init_root_id(&rootnode));
4904 mutex_unlock(&cgroup_root_mutex);
4905 mutex_unlock(&cgroup_mutex);
4907 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4913 err = register_filesystem(&cgroup_fs_type);
4915 kobject_put(cgroup_kobj);
4919 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4923 bdi_destroy(&cgroup_backing_dev_info);
4929 * proc_cgroup_show()
4930 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4931 * - Used for /proc/<pid>/cgroup.
4932 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4933 * doesn't really matter if tsk->cgroup changes after we read it,
4934 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4935 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4936 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4937 * cgroup to top_cgroup.
4940 /* TODO: Use a proper seq_file iterator */
4941 int proc_cgroup_show(struct seq_file *m, void *v)
4944 struct task_struct *tsk;
4947 struct cgroupfs_root *root;
4950 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4956 tsk = get_pid_task(pid, PIDTYPE_PID);
4962 mutex_lock(&cgroup_mutex);
4964 for_each_active_root(root) {
4965 struct cgroup_subsys *ss;
4966 struct cgroup *cgrp;
4969 seq_printf(m, "%d:", root->hierarchy_id);
4970 for_each_subsys(root, ss)
4971 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4972 if (strlen(root->name))
4973 seq_printf(m, "%sname=%s", count ? "," : "",
4976 cgrp = task_cgroup_from_root(tsk, root);
4977 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4985 mutex_unlock(&cgroup_mutex);
4986 put_task_struct(tsk);
4993 /* Display information about each subsystem and each hierarchy */
4994 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4998 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
5000 * ideally we don't want subsystems moving around while we do this.
5001 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
5002 * subsys/hierarchy state.
5004 mutex_lock(&cgroup_mutex);
5005 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5006 struct cgroup_subsys *ss = subsys[i];
5009 seq_printf(m, "%s\t%d\t%d\t%d\n",
5010 ss->name, ss->root->hierarchy_id,
5011 ss->root->number_of_cgroups, !ss->disabled);
5013 mutex_unlock(&cgroup_mutex);
5017 static int cgroupstats_open(struct inode *inode, struct file *file)
5019 return single_open(file, proc_cgroupstats_show, NULL);
5022 static const struct file_operations proc_cgroupstats_operations = {
5023 .open = cgroupstats_open,
5025 .llseek = seq_lseek,
5026 .release = single_release,
5030 * cgroup_fork - attach newly forked task to its parents cgroup.
5031 * @child: pointer to task_struct of forking parent process.
5033 * Description: A task inherits its parent's cgroup at fork().
5035 * A pointer to the shared css_set was automatically copied in
5036 * fork.c by dup_task_struct(). However, we ignore that copy, since
5037 * it was not made under the protection of RCU or cgroup_mutex, so
5038 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
5039 * have already changed current->cgroups, allowing the previously
5040 * referenced cgroup group to be removed and freed.
5042 * At the point that cgroup_fork() is called, 'current' is the parent
5043 * task, and the passed argument 'child' points to the child task.
5045 void cgroup_fork(struct task_struct *child)
5048 child->cgroups = current->cgroups;
5049 get_css_set(child->cgroups);
5050 task_unlock(current);
5051 INIT_LIST_HEAD(&child->cg_list);
5055 * cgroup_post_fork - called on a new task after adding it to the task list
5056 * @child: the task in question
5058 * Adds the task to the list running through its css_set if necessary and
5059 * call the subsystem fork() callbacks. Has to be after the task is
5060 * visible on the task list in case we race with the first call to
5061 * cgroup_iter_start() - to guarantee that the new task ends up on its
5064 void cgroup_post_fork(struct task_struct *child)
5069 * use_task_css_set_links is set to 1 before we walk the tasklist
5070 * under the tasklist_lock and we read it here after we added the child
5071 * to the tasklist under the tasklist_lock as well. If the child wasn't
5072 * yet in the tasklist when we walked through it from
5073 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
5074 * should be visible now due to the paired locking and barriers implied
5075 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
5076 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
5079 if (use_task_css_set_links) {
5080 write_lock(&css_set_lock);
5082 if (list_empty(&child->cg_list))
5083 list_add(&child->cg_list, &child->cgroups->tasks);
5085 write_unlock(&css_set_lock);
5089 * Call ss->fork(). This must happen after @child is linked on
5090 * css_set; otherwise, @child might change state between ->fork()
5091 * and addition to css_set.
5093 if (need_forkexit_callback) {
5095 * fork/exit callbacks are supported only for builtin
5096 * subsystems, and the builtin section of the subsys
5097 * array is immutable, so we don't need to lock the
5098 * subsys array here. On the other hand, modular section
5099 * of the array can be freed at module unload, so we
5102 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
5103 struct cgroup_subsys *ss = subsys[i];
5112 * cgroup_exit - detach cgroup from exiting task
5113 * @tsk: pointer to task_struct of exiting process
5114 * @run_callback: run exit callbacks?
5116 * Description: Detach cgroup from @tsk and release it.
5118 * Note that cgroups marked notify_on_release force every task in
5119 * them to take the global cgroup_mutex mutex when exiting.
5120 * This could impact scaling on very large systems. Be reluctant to
5121 * use notify_on_release cgroups where very high task exit scaling
5122 * is required on large systems.
5124 * the_top_cgroup_hack:
5126 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
5128 * We call cgroup_exit() while the task is still competent to
5129 * handle notify_on_release(), then leave the task attached to the
5130 * root cgroup in each hierarchy for the remainder of its exit.
5132 * To do this properly, we would increment the reference count on
5133 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
5134 * code we would add a second cgroup function call, to drop that
5135 * reference. This would just create an unnecessary hot spot on
5136 * the top_cgroup reference count, to no avail.
5138 * Normally, holding a reference to a cgroup without bumping its
5139 * count is unsafe. The cgroup could go away, or someone could
5140 * attach us to a different cgroup, decrementing the count on
5141 * the first cgroup that we never incremented. But in this case,
5142 * top_cgroup isn't going away, and either task has PF_EXITING set,
5143 * which wards off any cgroup_attach_task() attempts, or task is a failed
5144 * fork, never visible to cgroup_attach_task.
5146 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
5148 struct css_set *cset;
5152 * Unlink from the css_set task list if necessary.
5153 * Optimistically check cg_list before taking
5156 if (!list_empty(&tsk->cg_list)) {
5157 write_lock(&css_set_lock);
5158 if (!list_empty(&tsk->cg_list))
5159 list_del_init(&tsk->cg_list);
5160 write_unlock(&css_set_lock);
5163 /* Reassign the task to the init_css_set. */
5165 cset = tsk->cgroups;
5166 tsk->cgroups = &init_css_set;
5168 if (run_callbacks && need_forkexit_callback) {
5170 * fork/exit callbacks are supported only for builtin
5171 * subsystems, see cgroup_post_fork() for details.
5173 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
5174 struct cgroup_subsys *ss = subsys[i];
5177 struct cgroup *old_cgrp =
5178 rcu_dereference_raw(cset->subsys[i])->cgroup;
5179 struct cgroup *cgrp = task_cgroup(tsk, i);
5180 ss->exit(cgrp, old_cgrp, tsk);
5186 put_css_set_taskexit(cset);
5189 static void check_for_release(struct cgroup *cgrp)
5191 if (cgroup_is_releasable(cgrp) &&
5192 list_empty(&cgrp->cset_links) && list_empty(&cgrp->children)) {
5194 * Control Group is currently removeable. If it's not
5195 * already queued for a userspace notification, queue
5198 int need_schedule_work = 0;
5200 raw_spin_lock(&release_list_lock);
5201 if (!cgroup_is_dead(cgrp) &&
5202 list_empty(&cgrp->release_list)) {
5203 list_add(&cgrp->release_list, &release_list);
5204 need_schedule_work = 1;
5206 raw_spin_unlock(&release_list_lock);
5207 if (need_schedule_work)
5208 schedule_work(&release_agent_work);
5213 * Notify userspace when a cgroup is released, by running the
5214 * configured release agent with the name of the cgroup (path
5215 * relative to the root of cgroup file system) as the argument.
5217 * Most likely, this user command will try to rmdir this cgroup.
5219 * This races with the possibility that some other task will be
5220 * attached to this cgroup before it is removed, or that some other
5221 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5222 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5223 * unused, and this cgroup will be reprieved from its death sentence,
5224 * to continue to serve a useful existence. Next time it's released,
5225 * we will get notified again, if it still has 'notify_on_release' set.
5227 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5228 * means only wait until the task is successfully execve()'d. The
5229 * separate release agent task is forked by call_usermodehelper(),
5230 * then control in this thread returns here, without waiting for the
5231 * release agent task. We don't bother to wait because the caller of
5232 * this routine has no use for the exit status of the release agent
5233 * task, so no sense holding our caller up for that.
5235 static void cgroup_release_agent(struct work_struct *work)
5237 BUG_ON(work != &release_agent_work);
5238 mutex_lock(&cgroup_mutex);
5239 raw_spin_lock(&release_list_lock);
5240 while (!list_empty(&release_list)) {
5241 char *argv[3], *envp[3];
5243 char *pathbuf = NULL, *agentbuf = NULL;
5244 struct cgroup *cgrp = list_entry(release_list.next,
5247 list_del_init(&cgrp->release_list);
5248 raw_spin_unlock(&release_list_lock);
5249 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5252 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5254 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5259 argv[i++] = agentbuf;
5260 argv[i++] = pathbuf;
5264 /* minimal command environment */
5265 envp[i++] = "HOME=/";
5266 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5269 /* Drop the lock while we invoke the usermode helper,
5270 * since the exec could involve hitting disk and hence
5271 * be a slow process */
5272 mutex_unlock(&cgroup_mutex);
5273 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5274 mutex_lock(&cgroup_mutex);
5278 raw_spin_lock(&release_list_lock);
5280 raw_spin_unlock(&release_list_lock);
5281 mutex_unlock(&cgroup_mutex);
5284 static int __init cgroup_disable(char *str)
5289 while ((token = strsep(&str, ",")) != NULL) {
5292 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5293 struct cgroup_subsys *ss = subsys[i];
5296 * cgroup_disable, being at boot time, can't
5297 * know about module subsystems, so we don't
5300 if (!ss || ss->module)
5303 if (!strcmp(token, ss->name)) {
5305 printk(KERN_INFO "Disabling %s control group"
5306 " subsystem\n", ss->name);
5313 __setup("cgroup_disable=", cgroup_disable);
5316 * Functons for CSS ID.
5319 /* to get ID other than 0, this should be called when !cgroup_is_dead() */
5320 unsigned short css_id(struct cgroup_subsys_state *css)
5322 struct css_id *cssid;
5325 * This css_id() can return correct value when somone has refcnt
5326 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5327 * it's unchanged until freed.
5329 cssid = rcu_dereference_raw(css->id);
5335 EXPORT_SYMBOL_GPL(css_id);
5338 * css_is_ancestor - test "root" css is an ancestor of "child"
5339 * @child: the css to be tested.
5340 * @root: the css supporsed to be an ancestor of the child.
5342 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5343 * this function reads css->id, the caller must hold rcu_read_lock().
5344 * But, considering usual usage, the csses should be valid objects after test.
5345 * Assuming that the caller will do some action to the child if this returns
5346 * returns true, the caller must take "child";s reference count.
5347 * If "child" is valid object and this returns true, "root" is valid, too.
5350 bool css_is_ancestor(struct cgroup_subsys_state *child,
5351 const struct cgroup_subsys_state *root)
5353 struct css_id *child_id;
5354 struct css_id *root_id;
5356 child_id = rcu_dereference(child->id);
5359 root_id = rcu_dereference(root->id);
5362 if (child_id->depth < root_id->depth)
5364 if (child_id->stack[root_id->depth] != root_id->id)
5369 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5371 struct css_id *id = css->id;
5372 /* When this is called before css_id initialization, id can be NULL */
5376 BUG_ON(!ss->use_id);
5378 rcu_assign_pointer(id->css, NULL);
5379 rcu_assign_pointer(css->id, NULL);
5380 spin_lock(&ss->id_lock);
5381 idr_remove(&ss->idr, id->id);
5382 spin_unlock(&ss->id_lock);
5383 kfree_rcu(id, rcu_head);
5385 EXPORT_SYMBOL_GPL(free_css_id);
5388 * This is called by init or create(). Then, calls to this function are
5389 * always serialized (By cgroup_mutex() at create()).
5392 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5394 struct css_id *newid;
5397 BUG_ON(!ss->use_id);
5399 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5400 newid = kzalloc(size, GFP_KERNEL);
5402 return ERR_PTR(-ENOMEM);
5404 idr_preload(GFP_KERNEL);
5405 spin_lock(&ss->id_lock);
5406 /* Don't use 0. allocates an ID of 1-65535 */
5407 ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5408 spin_unlock(&ss->id_lock);
5411 /* Returns error when there are no free spaces for new ID.*/
5416 newid->depth = depth;
5420 return ERR_PTR(ret);
5424 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5425 struct cgroup_subsys_state *rootcss)
5427 struct css_id *newid;
5429 spin_lock_init(&ss->id_lock);
5432 newid = get_new_cssid(ss, 0);
5434 return PTR_ERR(newid);
5436 newid->stack[0] = newid->id;
5437 newid->css = rootcss;
5438 rootcss->id = newid;
5442 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5443 struct cgroup *child)
5445 int subsys_id, i, depth = 0;
5446 struct cgroup_subsys_state *parent_css, *child_css;
5447 struct css_id *child_id, *parent_id;
5449 subsys_id = ss->subsys_id;
5450 parent_css = parent->subsys[subsys_id];
5451 child_css = child->subsys[subsys_id];
5452 parent_id = parent_css->id;
5453 depth = parent_id->depth + 1;
5455 child_id = get_new_cssid(ss, depth);
5456 if (IS_ERR(child_id))
5457 return PTR_ERR(child_id);
5459 for (i = 0; i < depth; i++)
5460 child_id->stack[i] = parent_id->stack[i];
5461 child_id->stack[depth] = child_id->id;
5463 * child_id->css pointer will be set after this cgroup is available
5464 * see cgroup_populate_dir()
5466 rcu_assign_pointer(child_css->id, child_id);
5472 * css_lookup - lookup css by id
5473 * @ss: cgroup subsys to be looked into.
5476 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5477 * NULL if not. Should be called under rcu_read_lock()
5479 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5481 struct css_id *cssid = NULL;
5483 BUG_ON(!ss->use_id);
5484 cssid = idr_find(&ss->idr, id);
5486 if (unlikely(!cssid))
5489 return rcu_dereference(cssid->css);
5491 EXPORT_SYMBOL_GPL(css_lookup);
5494 * get corresponding css from file open on cgroupfs directory
5496 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5498 struct cgroup *cgrp;
5499 struct inode *inode;
5500 struct cgroup_subsys_state *css;
5502 inode = file_inode(f);
5503 /* check in cgroup filesystem dir */
5504 if (inode->i_op != &cgroup_dir_inode_operations)
5505 return ERR_PTR(-EBADF);
5507 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5508 return ERR_PTR(-EINVAL);
5511 cgrp = __d_cgrp(f->f_dentry);
5512 css = cgrp->subsys[id];
5513 return css ? css : ERR_PTR(-ENOENT);
5516 #ifdef CONFIG_CGROUP_DEBUG
5517 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5519 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5522 return ERR_PTR(-ENOMEM);
5527 static void debug_css_free(struct cgroup *cont)
5529 kfree(cont->subsys[debug_subsys_id]);
5532 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5534 return cgroup_task_count(cont);
5537 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5539 return (u64)(unsigned long)current->cgroups;
5542 static u64 current_css_set_refcount_read(struct cgroup *cont,
5548 count = atomic_read(¤t->cgroups->refcount);
5553 static int current_css_set_cg_links_read(struct cgroup *cont,
5555 struct seq_file *seq)
5557 struct cgrp_cset_link *link;
5558 struct css_set *cset;
5560 read_lock(&css_set_lock);
5562 cset = rcu_dereference(current->cgroups);
5563 list_for_each_entry(link, &cset->cgrp_links, cgrp_link) {
5564 struct cgroup *c = link->cgrp;
5568 name = c->dentry->d_name.name;
5571 seq_printf(seq, "Root %d group %s\n",
5572 c->root->hierarchy_id, name);
5575 read_unlock(&css_set_lock);
5579 #define MAX_TASKS_SHOWN_PER_CSS 25
5580 static int cgroup_css_links_read(struct cgroup *cont,
5582 struct seq_file *seq)
5584 struct cgrp_cset_link *link;
5586 read_lock(&css_set_lock);
5587 list_for_each_entry(link, &cont->cset_links, cset_link) {
5588 struct css_set *cset = link->cset;
5589 struct task_struct *task;
5591 seq_printf(seq, "css_set %p\n", cset);
5592 list_for_each_entry(task, &cset->tasks, cg_list) {
5593 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5594 seq_puts(seq, " ...\n");
5597 seq_printf(seq, " task %d\n",
5598 task_pid_vnr(task));
5602 read_unlock(&css_set_lock);
5606 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5608 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5611 static struct cftype debug_files[] = {
5613 .name = "taskcount",
5614 .read_u64 = debug_taskcount_read,
5618 .name = "current_css_set",
5619 .read_u64 = current_css_set_read,
5623 .name = "current_css_set_refcount",
5624 .read_u64 = current_css_set_refcount_read,
5628 .name = "current_css_set_cg_links",
5629 .read_seq_string = current_css_set_cg_links_read,
5633 .name = "cgroup_css_links",
5634 .read_seq_string = cgroup_css_links_read,
5638 .name = "releasable",
5639 .read_u64 = releasable_read,
5645 struct cgroup_subsys debug_subsys = {
5647 .css_alloc = debug_css_alloc,
5648 .css_free = debug_css_free,
5649 .subsys_id = debug_subsys_id,
5650 .base_cftypes = debug_files,
5652 #endif /* CONFIG_CGROUP_DEBUG */