2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hashtable.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_task */
63 #include <linux/kthread.h>
65 #include <linux/atomic.h>
67 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
68 #define CSS_DEACT_BIAS INT_MIN
71 * cgroup_mutex is the master lock. Any modification to cgroup or its
72 * hierarchy must be performed while holding it.
74 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
75 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
76 * release_agent_path and so on. Modifying requires both cgroup_mutex and
77 * cgroup_root_mutex. Readers can acquire either of the two. This is to
78 * break the following locking order cycle.
80 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
81 * B. namespace_sem -> cgroup_mutex
83 * B happens only through cgroup_show_options() and using cgroup_root_mutex
86 #ifdef CONFIG_PROVE_RCU
87 DEFINE_MUTEX(cgroup_mutex);
88 EXPORT_SYMBOL_GPL(cgroup_mutex); /* only for task_subsys_state_check() */
90 static DEFINE_MUTEX(cgroup_mutex);
93 static DEFINE_MUTEX(cgroup_root_mutex);
96 * Generate an array of cgroup subsystem pointers. At boot time, this is
97 * populated with the built in subsystems, and modular subsystems are
98 * registered after that. The mutable section of this array is protected by
101 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
102 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
103 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
104 #include <linux/cgroup_subsys.h>
107 #define MAX_CGROUP_ROOT_NAMELEN 64
110 * A cgroupfs_root represents the root of a cgroup hierarchy,
111 * and may be associated with a superblock to form an active
114 struct cgroupfs_root {
115 struct super_block *sb;
118 * The bitmask of subsystems intended to be attached to this
121 unsigned long subsys_mask;
123 /* Unique id for this hierarchy. */
126 /* The bitmask of subsystems currently attached to this hierarchy */
127 unsigned long actual_subsys_mask;
129 /* A list running through the attached subsystems */
130 struct list_head subsys_list;
132 /* The root cgroup for this hierarchy */
133 struct cgroup top_cgroup;
135 /* Tracks how many cgroups are currently defined in hierarchy.*/
136 int number_of_cgroups;
138 /* A list running through the active hierarchies */
139 struct list_head root_list;
141 /* All cgroups on this root, cgroup_mutex protected */
142 struct list_head allcg_list;
144 /* Hierarchy-specific flags */
147 /* IDs for cgroups in this hierarchy */
148 struct ida cgroup_ida;
150 /* The path to use for release notifications. */
151 char release_agent_path[PATH_MAX];
153 /* The name for this hierarchy - may be empty */
154 char name[MAX_CGROUP_ROOT_NAMELEN];
158 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
159 * subsystems that are otherwise unattached - it never has more than a
160 * single cgroup, and all tasks are part of that cgroup.
162 static struct cgroupfs_root rootnode;
165 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
168 struct list_head node;
169 struct dentry *dentry;
174 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
175 * cgroup_subsys->use_id != 0.
177 #define CSS_ID_MAX (65535)
180 * The css to which this ID points. This pointer is set to valid value
181 * after cgroup is populated. If cgroup is removed, this will be NULL.
182 * This pointer is expected to be RCU-safe because destroy()
183 * is called after synchronize_rcu(). But for safe use, css_tryget()
184 * should be used for avoiding race.
186 struct cgroup_subsys_state __rcu *css;
192 * Depth in hierarchy which this ID belongs to.
194 unsigned short depth;
196 * ID is freed by RCU. (and lookup routine is RCU safe.)
198 struct rcu_head rcu_head;
200 * Hierarchy of CSS ID belongs to.
202 unsigned short stack[0]; /* Array of Length (depth+1) */
206 * cgroup_event represents events which userspace want to receive.
208 struct cgroup_event {
210 * Cgroup which the event belongs to.
214 * Control file which the event associated.
218 * eventfd to signal userspace about the event.
220 struct eventfd_ctx *eventfd;
222 * Each of these stored in a list by the cgroup.
224 struct list_head list;
226 * All fields below needed to unregister event when
227 * userspace closes eventfd.
230 wait_queue_head_t *wqh;
232 struct work_struct remove;
235 /* The list of hierarchy roots */
237 static LIST_HEAD(roots);
238 static int root_count;
240 static DEFINE_IDA(hierarchy_ida);
241 static int next_hierarchy_id;
242 static DEFINE_SPINLOCK(hierarchy_id_lock);
244 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
245 #define dummytop (&rootnode.top_cgroup)
247 static struct cgroup_name root_cgroup_name = { .name = "/" };
249 /* This flag indicates whether tasks in the fork and exit paths should
250 * check for fork/exit handlers to call. This avoids us having to do
251 * extra work in the fork/exit path if none of the subsystems need to
254 static int need_forkexit_callback __read_mostly;
256 static int cgroup_destroy_locked(struct cgroup *cgrp);
257 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
258 struct cftype cfts[], bool is_add);
260 static int css_unbias_refcnt(int refcnt)
262 return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
265 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
266 static int css_refcnt(struct cgroup_subsys_state *css)
268 int v = atomic_read(&css->refcnt);
270 return css_unbias_refcnt(v);
273 /* convenient tests for these bits */
274 inline int cgroup_is_removed(const struct cgroup *cgrp)
276 return test_bit(CGRP_REMOVED, &cgrp->flags);
279 /* bits in struct cgroupfs_root flags field */
281 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
282 ROOT_XATTR, /* supports extended attributes */
285 static int cgroup_is_releasable(const struct cgroup *cgrp)
288 (1 << CGRP_RELEASABLE) |
289 (1 << CGRP_NOTIFY_ON_RELEASE);
290 return (cgrp->flags & bits) == bits;
293 static int notify_on_release(const struct cgroup *cgrp)
295 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
299 * for_each_subsys() allows you to iterate on each subsystem attached to
300 * an active hierarchy
302 #define for_each_subsys(_root, _ss) \
303 list_for_each_entry(_ss, &_root->subsys_list, sibling)
305 /* for_each_active_root() allows you to iterate across the active hierarchies */
306 #define for_each_active_root(_root) \
307 list_for_each_entry(_root, &roots, root_list)
309 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
311 return dentry->d_fsdata;
314 static inline struct cfent *__d_cfe(struct dentry *dentry)
316 return dentry->d_fsdata;
319 static inline struct cftype *__d_cft(struct dentry *dentry)
321 return __d_cfe(dentry)->type;
325 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
326 * @cgrp: the cgroup to be checked for liveness
328 * On success, returns true; the mutex should be later unlocked. On
329 * failure returns false with no lock held.
331 static bool cgroup_lock_live_group(struct cgroup *cgrp)
333 mutex_lock(&cgroup_mutex);
334 if (cgroup_is_removed(cgrp)) {
335 mutex_unlock(&cgroup_mutex);
341 /* the list of cgroups eligible for automatic release. Protected by
342 * release_list_lock */
343 static LIST_HEAD(release_list);
344 static DEFINE_RAW_SPINLOCK(release_list_lock);
345 static void cgroup_release_agent(struct work_struct *work);
346 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
347 static void check_for_release(struct cgroup *cgrp);
349 /* Link structure for associating css_set objects with cgroups */
350 struct cg_cgroup_link {
352 * List running through cg_cgroup_links associated with a
353 * cgroup, anchored on cgroup->css_sets
355 struct list_head cgrp_link_list;
358 * List running through cg_cgroup_links pointing at a
359 * single css_set object, anchored on css_set->cg_links
361 struct list_head cg_link_list;
365 /* The default css_set - used by init and its children prior to any
366 * hierarchies being mounted. It contains a pointer to the root state
367 * for each subsystem. Also used to anchor the list of css_sets. Not
368 * reference-counted, to improve performance when child cgroups
369 * haven't been created.
372 static struct css_set init_css_set;
373 static struct cg_cgroup_link init_css_set_link;
375 static int cgroup_init_idr(struct cgroup_subsys *ss,
376 struct cgroup_subsys_state *css);
378 /* css_set_lock protects the list of css_set objects, and the
379 * chain of tasks off each css_set. Nests outside task->alloc_lock
380 * due to cgroup_iter_start() */
381 static DEFINE_RWLOCK(css_set_lock);
382 static int css_set_count;
385 * hash table for cgroup groups. This improves the performance to find
386 * an existing css_set. This hash doesn't (currently) take into
387 * account cgroups in empty hierarchies.
389 #define CSS_SET_HASH_BITS 7
390 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
392 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
395 unsigned long key = 0UL;
397 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
398 key += (unsigned long)css[i];
399 key = (key >> 16) ^ key;
404 /* We don't maintain the lists running through each css_set to its
405 * task until after the first call to cgroup_iter_start(). This
406 * reduces the fork()/exit() overhead for people who have cgroups
407 * compiled into their kernel but not actually in use */
408 static int use_task_css_set_links __read_mostly;
410 static void __put_css_set(struct css_set *cg, int taskexit)
412 struct cg_cgroup_link *link;
413 struct cg_cgroup_link *saved_link;
415 * Ensure that the refcount doesn't hit zero while any readers
416 * can see it. Similar to atomic_dec_and_lock(), but for an
419 if (atomic_add_unless(&cg->refcount, -1, 1))
421 write_lock(&css_set_lock);
422 if (!atomic_dec_and_test(&cg->refcount)) {
423 write_unlock(&css_set_lock);
427 /* This css_set is dead. unlink it and release cgroup refcounts */
428 hash_del(&cg->hlist);
431 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
433 struct cgroup *cgrp = link->cgrp;
434 list_del(&link->cg_link_list);
435 list_del(&link->cgrp_link_list);
438 * We may not be holding cgroup_mutex, and if cgrp->count is
439 * dropped to 0 the cgroup can be destroyed at any time, hence
440 * rcu_read_lock is used to keep it alive.
443 if (atomic_dec_and_test(&cgrp->count) &&
444 notify_on_release(cgrp)) {
446 set_bit(CGRP_RELEASABLE, &cgrp->flags);
447 check_for_release(cgrp);
454 write_unlock(&css_set_lock);
455 kfree_rcu(cg, rcu_head);
459 * refcounted get/put for css_set objects
461 static inline void get_css_set(struct css_set *cg)
463 atomic_inc(&cg->refcount);
466 static inline void put_css_set(struct css_set *cg)
468 __put_css_set(cg, 0);
471 static inline void put_css_set_taskexit(struct css_set *cg)
473 __put_css_set(cg, 1);
477 * compare_css_sets - helper function for find_existing_css_set().
478 * @cg: candidate css_set being tested
479 * @old_cg: existing css_set for a task
480 * @new_cgrp: cgroup that's being entered by the task
481 * @template: desired set of css pointers in css_set (pre-calculated)
483 * Returns true if "cg" matches "old_cg" except for the hierarchy
484 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
486 static bool compare_css_sets(struct css_set *cg,
487 struct css_set *old_cg,
488 struct cgroup *new_cgrp,
489 struct cgroup_subsys_state *template[])
491 struct list_head *l1, *l2;
493 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
494 /* Not all subsystems matched */
499 * Compare cgroup pointers in order to distinguish between
500 * different cgroups in heirarchies with no subsystems. We
501 * could get by with just this check alone (and skip the
502 * memcmp above) but on most setups the memcmp check will
503 * avoid the need for this more expensive check on almost all
508 l2 = &old_cg->cg_links;
510 struct cg_cgroup_link *cgl1, *cgl2;
511 struct cgroup *cg1, *cg2;
515 /* See if we reached the end - both lists are equal length. */
516 if (l1 == &cg->cg_links) {
517 BUG_ON(l2 != &old_cg->cg_links);
520 BUG_ON(l2 == &old_cg->cg_links);
522 /* Locate the cgroups associated with these links. */
523 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
524 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
527 /* Hierarchies should be linked in the same order. */
528 BUG_ON(cg1->root != cg2->root);
531 * If this hierarchy is the hierarchy of the cgroup
532 * that's changing, then we need to check that this
533 * css_set points to the new cgroup; if it's any other
534 * hierarchy, then this css_set should point to the
535 * same cgroup as the old css_set.
537 if (cg1->root == new_cgrp->root) {
549 * find_existing_css_set() is a helper for
550 * find_css_set(), and checks to see whether an existing
551 * css_set is suitable.
553 * oldcg: the cgroup group that we're using before the cgroup
556 * cgrp: the cgroup that we're moving into
558 * template: location in which to build the desired set of subsystem
559 * state objects for the new cgroup group
561 static struct css_set *find_existing_css_set(
562 struct css_set *oldcg,
564 struct cgroup_subsys_state *template[])
567 struct cgroupfs_root *root = cgrp->root;
572 * Build the set of subsystem state objects that we want to see in the
573 * new css_set. while subsystems can change globally, the entries here
574 * won't change, so no need for locking.
576 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
577 if (root->subsys_mask & (1UL << i)) {
578 /* Subsystem is in this hierarchy. So we want
579 * the subsystem state from the new
581 template[i] = cgrp->subsys[i];
583 /* Subsystem is not in this hierarchy, so we
584 * don't want to change the subsystem state */
585 template[i] = oldcg->subsys[i];
589 key = css_set_hash(template);
590 hash_for_each_possible(css_set_table, cg, hlist, key) {
591 if (!compare_css_sets(cg, oldcg, cgrp, template))
594 /* This css_set matches what we need */
598 /* No existing cgroup group matched */
602 static void free_cg_links(struct list_head *tmp)
604 struct cg_cgroup_link *link;
605 struct cg_cgroup_link *saved_link;
607 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
608 list_del(&link->cgrp_link_list);
614 * allocate_cg_links() allocates "count" cg_cgroup_link structures
615 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
616 * success or a negative error
618 static int allocate_cg_links(int count, struct list_head *tmp)
620 struct cg_cgroup_link *link;
623 for (i = 0; i < count; i++) {
624 link = kmalloc(sizeof(*link), GFP_KERNEL);
629 list_add(&link->cgrp_link_list, tmp);
635 * link_css_set - a helper function to link a css_set to a cgroup
636 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
637 * @cg: the css_set to be linked
638 * @cgrp: the destination cgroup
640 static void link_css_set(struct list_head *tmp_cg_links,
641 struct css_set *cg, struct cgroup *cgrp)
643 struct cg_cgroup_link *link;
645 BUG_ON(list_empty(tmp_cg_links));
646 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
650 atomic_inc(&cgrp->count);
651 list_move(&link->cgrp_link_list, &cgrp->css_sets);
653 * Always add links to the tail of the list so that the list
654 * is sorted by order of hierarchy creation
656 list_add_tail(&link->cg_link_list, &cg->cg_links);
660 * find_css_set() takes an existing cgroup group and a
661 * cgroup object, and returns a css_set object that's
662 * equivalent to the old group, but with the given cgroup
663 * substituted into the appropriate hierarchy. Must be called with
666 static struct css_set *find_css_set(
667 struct css_set *oldcg, struct cgroup *cgrp)
670 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
672 struct list_head tmp_cg_links;
674 struct cg_cgroup_link *link;
677 /* First see if we already have a cgroup group that matches
679 read_lock(&css_set_lock);
680 res = find_existing_css_set(oldcg, cgrp, template);
683 read_unlock(&css_set_lock);
688 res = kmalloc(sizeof(*res), GFP_KERNEL);
692 /* Allocate all the cg_cgroup_link objects that we'll need */
693 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
698 atomic_set(&res->refcount, 1);
699 INIT_LIST_HEAD(&res->cg_links);
700 INIT_LIST_HEAD(&res->tasks);
701 INIT_HLIST_NODE(&res->hlist);
703 /* Copy the set of subsystem state objects generated in
704 * find_existing_css_set() */
705 memcpy(res->subsys, template, sizeof(res->subsys));
707 write_lock(&css_set_lock);
708 /* Add reference counts and links from the new css_set. */
709 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
710 struct cgroup *c = link->cgrp;
711 if (c->root == cgrp->root)
713 link_css_set(&tmp_cg_links, res, c);
716 BUG_ON(!list_empty(&tmp_cg_links));
720 /* Add this cgroup group to the hash table */
721 key = css_set_hash(res->subsys);
722 hash_add(css_set_table, &res->hlist, key);
724 write_unlock(&css_set_lock);
730 * Return the cgroup for "task" from the given hierarchy. Must be
731 * called with cgroup_mutex held.
733 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
734 struct cgroupfs_root *root)
737 struct cgroup *res = NULL;
739 BUG_ON(!mutex_is_locked(&cgroup_mutex));
740 read_lock(&css_set_lock);
742 * No need to lock the task - since we hold cgroup_mutex the
743 * task can't change groups, so the only thing that can happen
744 * is that it exits and its css is set back to init_css_set.
747 if (css == &init_css_set) {
748 res = &root->top_cgroup;
750 struct cg_cgroup_link *link;
751 list_for_each_entry(link, &css->cg_links, cg_link_list) {
752 struct cgroup *c = link->cgrp;
753 if (c->root == root) {
759 read_unlock(&css_set_lock);
765 * There is one global cgroup mutex. We also require taking
766 * task_lock() when dereferencing a task's cgroup subsys pointers.
767 * See "The task_lock() exception", at the end of this comment.
769 * A task must hold cgroup_mutex to modify cgroups.
771 * Any task can increment and decrement the count field without lock.
772 * So in general, code holding cgroup_mutex can't rely on the count
773 * field not changing. However, if the count goes to zero, then only
774 * cgroup_attach_task() can increment it again. Because a count of zero
775 * means that no tasks are currently attached, therefore there is no
776 * way a task attached to that cgroup can fork (the other way to
777 * increment the count). So code holding cgroup_mutex can safely
778 * assume that if the count is zero, it will stay zero. Similarly, if
779 * a task holds cgroup_mutex on a cgroup with zero count, it
780 * knows that the cgroup won't be removed, as cgroup_rmdir()
783 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
784 * (usually) take cgroup_mutex. These are the two most performance
785 * critical pieces of code here. The exception occurs on cgroup_exit(),
786 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
787 * is taken, and if the cgroup count is zero, a usermode call made
788 * to the release agent with the name of the cgroup (path relative to
789 * the root of cgroup file system) as the argument.
791 * A cgroup can only be deleted if both its 'count' of using tasks
792 * is zero, and its list of 'children' cgroups is empty. Since all
793 * tasks in the system use _some_ cgroup, and since there is always at
794 * least one task in the system (init, pid == 1), therefore, top_cgroup
795 * always has either children cgroups and/or using tasks. So we don't
796 * need a special hack to ensure that top_cgroup cannot be deleted.
798 * The task_lock() exception
800 * The need for this exception arises from the action of
801 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
802 * another. It does so using cgroup_mutex, however there are
803 * several performance critical places that need to reference
804 * task->cgroup without the expense of grabbing a system global
805 * mutex. Therefore except as noted below, when dereferencing or, as
806 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
807 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
808 * the task_struct routinely used for such matters.
810 * P.S. One more locking exception. RCU is used to guard the
811 * update of a tasks cgroup pointer by cgroup_attach_task()
815 * A couple of forward declarations required, due to cyclic reference loop:
816 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
817 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
821 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
822 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
823 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
824 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
825 unsigned long subsys_mask);
826 static const struct inode_operations cgroup_dir_inode_operations;
827 static const struct file_operations proc_cgroupstats_operations;
829 static struct backing_dev_info cgroup_backing_dev_info = {
831 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
834 static int alloc_css_id(struct cgroup_subsys *ss,
835 struct cgroup *parent, struct cgroup *child);
837 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
839 struct inode *inode = new_inode(sb);
842 inode->i_ino = get_next_ino();
843 inode->i_mode = mode;
844 inode->i_uid = current_fsuid();
845 inode->i_gid = current_fsgid();
846 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
847 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
852 static struct cgroup_name *cgroup_alloc_name(struct dentry *dentry)
854 struct cgroup_name *name;
856 name = kmalloc(sizeof(*name) + dentry->d_name.len + 1, GFP_KERNEL);
859 strcpy(name->name, dentry->d_name.name);
863 static void cgroup_free_fn(struct work_struct *work)
865 struct cgroup *cgrp = container_of(work, struct cgroup, free_work);
866 struct cgroup_subsys *ss;
868 mutex_lock(&cgroup_mutex);
870 * Release the subsystem state objects.
872 for_each_subsys(cgrp->root, ss)
875 cgrp->root->number_of_cgroups--;
876 mutex_unlock(&cgroup_mutex);
879 * Drop the active superblock reference that we took when we
882 deactivate_super(cgrp->root->sb);
885 * if we're getting rid of the cgroup, refcount should ensure
886 * that there are no pidlists left.
888 BUG_ON(!list_empty(&cgrp->pidlists));
890 simple_xattrs_free(&cgrp->xattrs);
892 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
893 kfree(rcu_dereference_raw(cgrp->name));
897 static void cgroup_free_rcu(struct rcu_head *head)
899 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
901 schedule_work(&cgrp->free_work);
904 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
906 /* is dentry a directory ? if so, kfree() associated cgroup */
907 if (S_ISDIR(inode->i_mode)) {
908 struct cgroup *cgrp = dentry->d_fsdata;
910 BUG_ON(!(cgroup_is_removed(cgrp)));
911 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
913 struct cfent *cfe = __d_cfe(dentry);
914 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
915 struct cftype *cft = cfe->type;
917 WARN_ONCE(!list_empty(&cfe->node) &&
918 cgrp != &cgrp->root->top_cgroup,
919 "cfe still linked for %s\n", cfe->type->name);
921 simple_xattrs_free(&cft->xattrs);
926 static int cgroup_delete(const struct dentry *d)
931 static void remove_dir(struct dentry *d)
933 struct dentry *parent = dget(d->d_parent);
936 simple_rmdir(parent->d_inode, d);
940 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
944 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
945 lockdep_assert_held(&cgroup_mutex);
948 * If we're doing cleanup due to failure of cgroup_create(),
949 * the corresponding @cfe may not exist.
951 list_for_each_entry(cfe, &cgrp->files, node) {
952 struct dentry *d = cfe->dentry;
954 if (cft && cfe->type != cft)
959 simple_unlink(cgrp->dentry->d_inode, d);
960 list_del_init(&cfe->node);
968 * cgroup_clear_directory - selective removal of base and subsystem files
969 * @dir: directory containing the files
970 * @base_files: true if the base files should be removed
971 * @subsys_mask: mask of the subsystem ids whose files should be removed
973 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
974 unsigned long subsys_mask)
976 struct cgroup *cgrp = __d_cgrp(dir);
977 struct cgroup_subsys *ss;
979 for_each_subsys(cgrp->root, ss) {
980 struct cftype_set *set;
981 if (!test_bit(ss->subsys_id, &subsys_mask))
983 list_for_each_entry(set, &ss->cftsets, node)
984 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
987 while (!list_empty(&cgrp->files))
988 cgroup_rm_file(cgrp, NULL);
993 * NOTE : the dentry must have been dget()'ed
995 static void cgroup_d_remove_dir(struct dentry *dentry)
997 struct dentry *parent;
998 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1000 cgroup_clear_directory(dentry, true, root->subsys_mask);
1002 parent = dentry->d_parent;
1003 spin_lock(&parent->d_lock);
1004 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1005 list_del_init(&dentry->d_u.d_child);
1006 spin_unlock(&dentry->d_lock);
1007 spin_unlock(&parent->d_lock);
1012 * Call with cgroup_mutex held. Drops reference counts on modules, including
1013 * any duplicate ones that parse_cgroupfs_options took. If this function
1014 * returns an error, no reference counts are touched.
1016 static int rebind_subsystems(struct cgroupfs_root *root,
1017 unsigned long final_subsys_mask)
1019 unsigned long added_mask, removed_mask;
1020 struct cgroup *cgrp = &root->top_cgroup;
1023 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1024 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1026 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
1027 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1028 /* Check that any added subsystems are currently free */
1029 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1030 unsigned long bit = 1UL << i;
1031 struct cgroup_subsys *ss = subsys[i];
1032 if (!(bit & added_mask))
1035 * Nobody should tell us to do a subsys that doesn't exist:
1036 * parse_cgroupfs_options should catch that case and refcounts
1037 * ensure that subsystems won't disappear once selected.
1040 if (ss->root != &rootnode) {
1041 /* Subsystem isn't free */
1046 /* Currently we don't handle adding/removing subsystems when
1047 * any child cgroups exist. This is theoretically supportable
1048 * but involves complex error handling, so it's being left until
1050 if (root->number_of_cgroups > 1)
1053 /* Process each subsystem */
1054 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1055 struct cgroup_subsys *ss = subsys[i];
1056 unsigned long bit = 1UL << i;
1057 if (bit & added_mask) {
1058 /* We're binding this subsystem to this hierarchy */
1060 BUG_ON(cgrp->subsys[i]);
1061 BUG_ON(!dummytop->subsys[i]);
1062 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1063 cgrp->subsys[i] = dummytop->subsys[i];
1064 cgrp->subsys[i]->cgroup = cgrp;
1065 list_move(&ss->sibling, &root->subsys_list);
1067 /* refcount was already taken, and we're keeping it */
1068 } else if (bit & removed_mask) {
1069 /* We're removing this subsystem */
1071 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1072 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1073 dummytop->subsys[i]->cgroup = dummytop;
1074 cgrp->subsys[i] = NULL;
1075 subsys[i]->root = &rootnode;
1076 list_move(&ss->sibling, &rootnode.subsys_list);
1077 /* subsystem is now free - drop reference on module */
1078 module_put(ss->module);
1079 } else if (bit & final_subsys_mask) {
1080 /* Subsystem state should already exist */
1082 BUG_ON(!cgrp->subsys[i]);
1084 * a refcount was taken, but we already had one, so
1085 * drop the extra reference.
1087 module_put(ss->module);
1088 #ifdef CONFIG_MODULE_UNLOAD
1089 BUG_ON(ss->module && !module_refcount(ss->module));
1092 /* Subsystem state shouldn't exist */
1093 BUG_ON(cgrp->subsys[i]);
1096 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1101 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1103 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1104 struct cgroup_subsys *ss;
1106 mutex_lock(&cgroup_root_mutex);
1107 for_each_subsys(root, ss)
1108 seq_printf(seq, ",%s", ss->name);
1109 if (test_bit(ROOT_NOPREFIX, &root->flags))
1110 seq_puts(seq, ",noprefix");
1111 if (test_bit(ROOT_XATTR, &root->flags))
1112 seq_puts(seq, ",xattr");
1113 if (strlen(root->release_agent_path))
1114 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1115 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1116 seq_puts(seq, ",clone_children");
1117 if (strlen(root->name))
1118 seq_printf(seq, ",name=%s", root->name);
1119 mutex_unlock(&cgroup_root_mutex);
1123 struct cgroup_sb_opts {
1124 unsigned long subsys_mask;
1125 unsigned long flags;
1126 char *release_agent;
1127 bool cpuset_clone_children;
1129 /* User explicitly requested empty subsystem */
1132 struct cgroupfs_root *new_root;
1137 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1138 * with cgroup_mutex held to protect the subsys[] array. This function takes
1139 * refcounts on subsystems to be used, unless it returns error, in which case
1140 * no refcounts are taken.
1142 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1144 char *token, *o = data;
1145 bool all_ss = false, one_ss = false;
1146 unsigned long mask = (unsigned long)-1;
1148 bool module_pin_failed = false;
1150 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1152 #ifdef CONFIG_CPUSETS
1153 mask = ~(1UL << cpuset_subsys_id);
1156 memset(opts, 0, sizeof(*opts));
1158 while ((token = strsep(&o, ",")) != NULL) {
1161 if (!strcmp(token, "none")) {
1162 /* Explicitly have no subsystems */
1166 if (!strcmp(token, "all")) {
1167 /* Mutually exclusive option 'all' + subsystem name */
1173 if (!strcmp(token, "noprefix")) {
1174 set_bit(ROOT_NOPREFIX, &opts->flags);
1177 if (!strcmp(token, "clone_children")) {
1178 opts->cpuset_clone_children = true;
1181 if (!strcmp(token, "xattr")) {
1182 set_bit(ROOT_XATTR, &opts->flags);
1185 if (!strncmp(token, "release_agent=", 14)) {
1186 /* Specifying two release agents is forbidden */
1187 if (opts->release_agent)
1189 opts->release_agent =
1190 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1191 if (!opts->release_agent)
1195 if (!strncmp(token, "name=", 5)) {
1196 const char *name = token + 5;
1197 /* Can't specify an empty name */
1200 /* Must match [\w.-]+ */
1201 for (i = 0; i < strlen(name); i++) {
1205 if ((c == '.') || (c == '-') || (c == '_'))
1209 /* Specifying two names is forbidden */
1212 opts->name = kstrndup(name,
1213 MAX_CGROUP_ROOT_NAMELEN - 1,
1221 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1222 struct cgroup_subsys *ss = subsys[i];
1225 if (strcmp(token, ss->name))
1230 /* Mutually exclusive option 'all' + subsystem name */
1233 set_bit(i, &opts->subsys_mask);
1238 if (i == CGROUP_SUBSYS_COUNT)
1243 * If the 'all' option was specified select all the subsystems,
1244 * otherwise if 'none', 'name=' and a subsystem name options
1245 * were not specified, let's default to 'all'
1247 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1248 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1249 struct cgroup_subsys *ss = subsys[i];
1254 set_bit(i, &opts->subsys_mask);
1258 /* Consistency checks */
1261 * Option noprefix was introduced just for backward compatibility
1262 * with the old cpuset, so we allow noprefix only if mounting just
1263 * the cpuset subsystem.
1265 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1266 (opts->subsys_mask & mask))
1270 /* Can't specify "none" and some subsystems */
1271 if (opts->subsys_mask && opts->none)
1275 * We either have to specify by name or by subsystems. (So all
1276 * empty hierarchies must have a name).
1278 if (!opts->subsys_mask && !opts->name)
1282 * Grab references on all the modules we'll need, so the subsystems
1283 * don't dance around before rebind_subsystems attaches them. This may
1284 * take duplicate reference counts on a subsystem that's already used,
1285 * but rebind_subsystems handles this case.
1287 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1288 unsigned long bit = 1UL << i;
1290 if (!(bit & opts->subsys_mask))
1292 if (!try_module_get(subsys[i]->module)) {
1293 module_pin_failed = true;
1297 if (module_pin_failed) {
1299 * oops, one of the modules was going away. this means that we
1300 * raced with a module_delete call, and to the user this is
1301 * essentially a "subsystem doesn't exist" case.
1303 for (i--; i >= 0; i--) {
1304 /* drop refcounts only on the ones we took */
1305 unsigned long bit = 1UL << i;
1307 if (!(bit & opts->subsys_mask))
1309 module_put(subsys[i]->module);
1317 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1320 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1321 unsigned long bit = 1UL << i;
1323 if (!(bit & subsys_mask))
1325 module_put(subsys[i]->module);
1329 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1332 struct cgroupfs_root *root = sb->s_fs_info;
1333 struct cgroup *cgrp = &root->top_cgroup;
1334 struct cgroup_sb_opts opts;
1335 unsigned long added_mask, removed_mask;
1337 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1338 mutex_lock(&cgroup_mutex);
1339 mutex_lock(&cgroup_root_mutex);
1341 /* See what subsystems are wanted */
1342 ret = parse_cgroupfs_options(data, &opts);
1346 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1347 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1348 task_tgid_nr(current), current->comm);
1350 added_mask = opts.subsys_mask & ~root->subsys_mask;
1351 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1353 /* Don't allow flags or name to change at remount */
1354 if (opts.flags != root->flags ||
1355 (opts.name && strcmp(opts.name, root->name))) {
1357 drop_parsed_module_refcounts(opts.subsys_mask);
1362 * Clear out the files of subsystems that should be removed, do
1363 * this before rebind_subsystems, since rebind_subsystems may
1364 * change this hierarchy's subsys_list.
1366 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1368 ret = rebind_subsystems(root, opts.subsys_mask);
1370 /* rebind_subsystems failed, re-populate the removed files */
1371 cgroup_populate_dir(cgrp, false, removed_mask);
1372 drop_parsed_module_refcounts(opts.subsys_mask);
1376 /* re-populate subsystem files */
1377 cgroup_populate_dir(cgrp, false, added_mask);
1379 if (opts.release_agent)
1380 strcpy(root->release_agent_path, opts.release_agent);
1382 kfree(opts.release_agent);
1384 mutex_unlock(&cgroup_root_mutex);
1385 mutex_unlock(&cgroup_mutex);
1386 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1390 static const struct super_operations cgroup_ops = {
1391 .statfs = simple_statfs,
1392 .drop_inode = generic_delete_inode,
1393 .show_options = cgroup_show_options,
1394 .remount_fs = cgroup_remount,
1397 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1399 INIT_LIST_HEAD(&cgrp->sibling);
1400 INIT_LIST_HEAD(&cgrp->children);
1401 INIT_LIST_HEAD(&cgrp->files);
1402 INIT_LIST_HEAD(&cgrp->css_sets);
1403 INIT_LIST_HEAD(&cgrp->allcg_node);
1404 INIT_LIST_HEAD(&cgrp->release_list);
1405 INIT_LIST_HEAD(&cgrp->pidlists);
1406 INIT_WORK(&cgrp->free_work, cgroup_free_fn);
1407 mutex_init(&cgrp->pidlist_mutex);
1408 INIT_LIST_HEAD(&cgrp->event_list);
1409 spin_lock_init(&cgrp->event_list_lock);
1410 simple_xattrs_init(&cgrp->xattrs);
1413 static void init_cgroup_root(struct cgroupfs_root *root)
1415 struct cgroup *cgrp = &root->top_cgroup;
1417 INIT_LIST_HEAD(&root->subsys_list);
1418 INIT_LIST_HEAD(&root->root_list);
1419 INIT_LIST_HEAD(&root->allcg_list);
1420 root->number_of_cgroups = 1;
1422 cgrp->name = &root_cgroup_name;
1423 cgrp->top_cgroup = cgrp;
1424 init_cgroup_housekeeping(cgrp);
1425 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1428 static bool init_root_id(struct cgroupfs_root *root)
1433 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1435 spin_lock(&hierarchy_id_lock);
1436 /* Try to allocate the next unused ID */
1437 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1438 &root->hierarchy_id);
1440 /* Try again starting from 0 */
1441 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1443 next_hierarchy_id = root->hierarchy_id + 1;
1444 } else if (ret != -EAGAIN) {
1445 /* Can only get here if the 31-bit IDR is full ... */
1448 spin_unlock(&hierarchy_id_lock);
1453 static int cgroup_test_super(struct super_block *sb, void *data)
1455 struct cgroup_sb_opts *opts = data;
1456 struct cgroupfs_root *root = sb->s_fs_info;
1458 /* If we asked for a name then it must match */
1459 if (opts->name && strcmp(opts->name, root->name))
1463 * If we asked for subsystems (or explicitly for no
1464 * subsystems) then they must match
1466 if ((opts->subsys_mask || opts->none)
1467 && (opts->subsys_mask != root->subsys_mask))
1473 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1475 struct cgroupfs_root *root;
1477 if (!opts->subsys_mask && !opts->none)
1480 root = kzalloc(sizeof(*root), GFP_KERNEL);
1482 return ERR_PTR(-ENOMEM);
1484 if (!init_root_id(root)) {
1486 return ERR_PTR(-ENOMEM);
1488 init_cgroup_root(root);
1490 root->subsys_mask = opts->subsys_mask;
1491 root->flags = opts->flags;
1492 ida_init(&root->cgroup_ida);
1493 if (opts->release_agent)
1494 strcpy(root->release_agent_path, opts->release_agent);
1496 strcpy(root->name, opts->name);
1497 if (opts->cpuset_clone_children)
1498 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1502 static void cgroup_drop_root(struct cgroupfs_root *root)
1507 BUG_ON(!root->hierarchy_id);
1508 spin_lock(&hierarchy_id_lock);
1509 ida_remove(&hierarchy_ida, root->hierarchy_id);
1510 spin_unlock(&hierarchy_id_lock);
1511 ida_destroy(&root->cgroup_ida);
1515 static int cgroup_set_super(struct super_block *sb, void *data)
1518 struct cgroup_sb_opts *opts = data;
1520 /* If we don't have a new root, we can't set up a new sb */
1521 if (!opts->new_root)
1524 BUG_ON(!opts->subsys_mask && !opts->none);
1526 ret = set_anon_super(sb, NULL);
1530 sb->s_fs_info = opts->new_root;
1531 opts->new_root->sb = sb;
1533 sb->s_blocksize = PAGE_CACHE_SIZE;
1534 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1535 sb->s_magic = CGROUP_SUPER_MAGIC;
1536 sb->s_op = &cgroup_ops;
1541 static int cgroup_get_rootdir(struct super_block *sb)
1543 static const struct dentry_operations cgroup_dops = {
1544 .d_iput = cgroup_diput,
1545 .d_delete = cgroup_delete,
1548 struct inode *inode =
1549 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1554 inode->i_fop = &simple_dir_operations;
1555 inode->i_op = &cgroup_dir_inode_operations;
1556 /* directories start off with i_nlink == 2 (for "." entry) */
1558 sb->s_root = d_make_root(inode);
1561 /* for everything else we want ->d_op set */
1562 sb->s_d_op = &cgroup_dops;
1566 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1567 int flags, const char *unused_dev_name,
1570 struct cgroup_sb_opts opts;
1571 struct cgroupfs_root *root;
1573 struct super_block *sb;
1574 struct cgroupfs_root *new_root;
1575 struct inode *inode;
1577 /* First find the desired set of subsystems */
1578 mutex_lock(&cgroup_mutex);
1579 ret = parse_cgroupfs_options(data, &opts);
1580 mutex_unlock(&cgroup_mutex);
1585 * Allocate a new cgroup root. We may not need it if we're
1586 * reusing an existing hierarchy.
1588 new_root = cgroup_root_from_opts(&opts);
1589 if (IS_ERR(new_root)) {
1590 ret = PTR_ERR(new_root);
1593 opts.new_root = new_root;
1595 /* Locate an existing or new sb for this hierarchy */
1596 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1599 cgroup_drop_root(opts.new_root);
1603 root = sb->s_fs_info;
1605 if (root == opts.new_root) {
1606 /* We used the new root structure, so this is a new hierarchy */
1607 struct list_head tmp_cg_links;
1608 struct cgroup *root_cgrp = &root->top_cgroup;
1609 struct cgroupfs_root *existing_root;
1610 const struct cred *cred;
1614 BUG_ON(sb->s_root != NULL);
1616 ret = cgroup_get_rootdir(sb);
1618 goto drop_new_super;
1619 inode = sb->s_root->d_inode;
1621 mutex_lock(&inode->i_mutex);
1622 mutex_lock(&cgroup_mutex);
1623 mutex_lock(&cgroup_root_mutex);
1625 /* Check for name clashes with existing mounts */
1627 if (strlen(root->name))
1628 for_each_active_root(existing_root)
1629 if (!strcmp(existing_root->name, root->name))
1633 * We're accessing css_set_count without locking
1634 * css_set_lock here, but that's OK - it can only be
1635 * increased by someone holding cgroup_lock, and
1636 * that's us. The worst that can happen is that we
1637 * have some link structures left over
1639 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1643 ret = rebind_subsystems(root, root->subsys_mask);
1644 if (ret == -EBUSY) {
1645 free_cg_links(&tmp_cg_links);
1649 * There must be no failure case after here, since rebinding
1650 * takes care of subsystems' refcounts, which are explicitly
1651 * dropped in the failure exit path.
1654 /* EBUSY should be the only error here */
1657 list_add(&root->root_list, &roots);
1660 sb->s_root->d_fsdata = root_cgrp;
1661 root->top_cgroup.dentry = sb->s_root;
1663 /* Link the top cgroup in this hierarchy into all
1664 * the css_set objects */
1665 write_lock(&css_set_lock);
1666 hash_for_each(css_set_table, i, cg, hlist)
1667 link_css_set(&tmp_cg_links, cg, root_cgrp);
1668 write_unlock(&css_set_lock);
1670 free_cg_links(&tmp_cg_links);
1672 BUG_ON(!list_empty(&root_cgrp->children));
1673 BUG_ON(root->number_of_cgroups != 1);
1675 cred = override_creds(&init_cred);
1676 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1678 mutex_unlock(&cgroup_root_mutex);
1679 mutex_unlock(&cgroup_mutex);
1680 mutex_unlock(&inode->i_mutex);
1683 * We re-used an existing hierarchy - the new root (if
1684 * any) is not needed
1686 cgroup_drop_root(opts.new_root);
1687 /* no subsys rebinding, so refcounts don't change */
1688 drop_parsed_module_refcounts(opts.subsys_mask);
1691 kfree(opts.release_agent);
1693 return dget(sb->s_root);
1696 mutex_unlock(&cgroup_root_mutex);
1697 mutex_unlock(&cgroup_mutex);
1698 mutex_unlock(&inode->i_mutex);
1700 deactivate_locked_super(sb);
1702 drop_parsed_module_refcounts(opts.subsys_mask);
1704 kfree(opts.release_agent);
1706 return ERR_PTR(ret);
1709 static void cgroup_kill_sb(struct super_block *sb) {
1710 struct cgroupfs_root *root = sb->s_fs_info;
1711 struct cgroup *cgrp = &root->top_cgroup;
1713 struct cg_cgroup_link *link;
1714 struct cg_cgroup_link *saved_link;
1718 BUG_ON(root->number_of_cgroups != 1);
1719 BUG_ON(!list_empty(&cgrp->children));
1721 mutex_lock(&cgroup_mutex);
1722 mutex_lock(&cgroup_root_mutex);
1724 /* Rebind all subsystems back to the default hierarchy */
1725 ret = rebind_subsystems(root, 0);
1726 /* Shouldn't be able to fail ... */
1730 * Release all the links from css_sets to this hierarchy's
1733 write_lock(&css_set_lock);
1735 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1737 list_del(&link->cg_link_list);
1738 list_del(&link->cgrp_link_list);
1741 write_unlock(&css_set_lock);
1743 if (!list_empty(&root->root_list)) {
1744 list_del(&root->root_list);
1748 mutex_unlock(&cgroup_root_mutex);
1749 mutex_unlock(&cgroup_mutex);
1751 simple_xattrs_free(&cgrp->xattrs);
1753 kill_litter_super(sb);
1754 cgroup_drop_root(root);
1757 static struct file_system_type cgroup_fs_type = {
1759 .mount = cgroup_mount,
1760 .kill_sb = cgroup_kill_sb,
1763 static struct kobject *cgroup_kobj;
1766 * cgroup_path - generate the path of a cgroup
1767 * @cgrp: the cgroup in question
1768 * @buf: the buffer to write the path into
1769 * @buflen: the length of the buffer
1771 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1773 * We can't generate cgroup path using dentry->d_name, as accessing
1774 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1775 * inode's i_mutex, while on the other hand cgroup_path() can be called
1776 * with some irq-safe spinlocks held.
1778 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1780 int ret = -ENAMETOOLONG;
1783 start = buf + buflen - 1;
1788 const char *name = cgroup_name(cgrp);
1792 if ((start -= len) < buf)
1794 memcpy(start, name, len);
1803 cgrp = cgrp->parent;
1806 memmove(buf, start, buf + buflen - start);
1811 EXPORT_SYMBOL_GPL(cgroup_path);
1814 * Control Group taskset
1816 struct task_and_cgroup {
1817 struct task_struct *task;
1818 struct cgroup *cgrp;
1822 struct cgroup_taskset {
1823 struct task_and_cgroup single;
1824 struct flex_array *tc_array;
1827 struct cgroup *cur_cgrp;
1831 * cgroup_taskset_first - reset taskset and return the first task
1832 * @tset: taskset of interest
1834 * @tset iteration is initialized and the first task is returned.
1836 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1838 if (tset->tc_array) {
1840 return cgroup_taskset_next(tset);
1842 tset->cur_cgrp = tset->single.cgrp;
1843 return tset->single.task;
1846 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1849 * cgroup_taskset_next - iterate to the next task in taskset
1850 * @tset: taskset of interest
1852 * Return the next task in @tset. Iteration must have been initialized
1853 * with cgroup_taskset_first().
1855 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1857 struct task_and_cgroup *tc;
1859 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1862 tc = flex_array_get(tset->tc_array, tset->idx++);
1863 tset->cur_cgrp = tc->cgrp;
1866 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1869 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1870 * @tset: taskset of interest
1872 * Return the cgroup for the current (last returned) task of @tset. This
1873 * function must be preceded by either cgroup_taskset_first() or
1874 * cgroup_taskset_next().
1876 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1878 return tset->cur_cgrp;
1880 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1883 * cgroup_taskset_size - return the number of tasks in taskset
1884 * @tset: taskset of interest
1886 int cgroup_taskset_size(struct cgroup_taskset *tset)
1888 return tset->tc_array ? tset->tc_array_len : 1;
1890 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1894 * cgroup_task_migrate - move a task from one cgroup to another.
1896 * Must be called with cgroup_mutex and threadgroup locked.
1898 static void cgroup_task_migrate(struct cgroup *oldcgrp,
1899 struct task_struct *tsk, struct css_set *newcg)
1901 struct css_set *oldcg;
1904 * We are synchronized through threadgroup_lock() against PF_EXITING
1905 * setting such that we can't race against cgroup_exit() changing the
1906 * css_set to init_css_set and dropping the old one.
1908 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1909 oldcg = tsk->cgroups;
1912 rcu_assign_pointer(tsk->cgroups, newcg);
1915 /* Update the css_set linked lists if we're using them */
1916 write_lock(&css_set_lock);
1917 if (!list_empty(&tsk->cg_list))
1918 list_move(&tsk->cg_list, &newcg->tasks);
1919 write_unlock(&css_set_lock);
1922 * We just gained a reference on oldcg by taking it from the task. As
1923 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1924 * it here; it will be freed under RCU.
1926 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1931 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1932 * @cgrp: the cgroup to attach to
1933 * @tsk: the task or the leader of the threadgroup to be attached
1934 * @threadgroup: attach the whole threadgroup?
1936 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1937 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1939 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
1942 int retval, i, group_size;
1943 struct cgroup_subsys *ss, *failed_ss = NULL;
1944 struct cgroupfs_root *root = cgrp->root;
1945 /* threadgroup list cursor and array */
1946 struct task_struct *leader = tsk;
1947 struct task_and_cgroup *tc;
1948 struct flex_array *group;
1949 struct cgroup_taskset tset = { };
1952 * step 0: in order to do expensive, possibly blocking operations for
1953 * every thread, we cannot iterate the thread group list, since it needs
1954 * rcu or tasklist locked. instead, build an array of all threads in the
1955 * group - group_rwsem prevents new threads from appearing, and if
1956 * threads exit, this will just be an over-estimate.
1959 group_size = get_nr_threads(tsk);
1962 /* flex_array supports very large thread-groups better than kmalloc. */
1963 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
1966 /* pre-allocate to guarantee space while iterating in rcu read-side. */
1967 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
1969 goto out_free_group_list;
1973 * Prevent freeing of tasks while we take a snapshot. Tasks that are
1974 * already PF_EXITING could be freed from underneath us unless we
1975 * take an rcu_read_lock.
1979 struct task_and_cgroup ent;
1981 /* @tsk either already exited or can't exit until the end */
1982 if (tsk->flags & PF_EXITING)
1985 /* as per above, nr_threads may decrease, but not increase. */
1986 BUG_ON(i >= group_size);
1988 ent.cgrp = task_cgroup_from_root(tsk, root);
1989 /* nothing to do if this task is already in the cgroup */
1990 if (ent.cgrp == cgrp)
1993 * saying GFP_ATOMIC has no effect here because we did prealloc
1994 * earlier, but it's good form to communicate our expectations.
1996 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
1997 BUG_ON(retval != 0);
2002 } while_each_thread(leader, tsk);
2004 /* remember the number of threads in the array for later. */
2006 tset.tc_array = group;
2007 tset.tc_array_len = group_size;
2009 /* methods shouldn't be called if no task is actually migrating */
2012 goto out_free_group_list;
2015 * step 1: check that we can legitimately attach to the cgroup.
2017 for_each_subsys(root, ss) {
2018 if (ss->can_attach) {
2019 retval = ss->can_attach(cgrp, &tset);
2022 goto out_cancel_attach;
2028 * step 2: make sure css_sets exist for all threads to be migrated.
2029 * we use find_css_set, which allocates a new one if necessary.
2031 for (i = 0; i < group_size; i++) {
2032 tc = flex_array_get(group, i);
2033 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2036 goto out_put_css_set_refs;
2041 * step 3: now that we're guaranteed success wrt the css_sets,
2042 * proceed to move all tasks to the new cgroup. There are no
2043 * failure cases after here, so this is the commit point.
2045 for (i = 0; i < group_size; i++) {
2046 tc = flex_array_get(group, i);
2047 cgroup_task_migrate(tc->cgrp, tc->task, tc->cg);
2049 /* nothing is sensitive to fork() after this point. */
2052 * step 4: do subsystem attach callbacks.
2054 for_each_subsys(root, ss) {
2056 ss->attach(cgrp, &tset);
2060 * step 5: success! and cleanup
2063 out_put_css_set_refs:
2065 for (i = 0; i < group_size; i++) {
2066 tc = flex_array_get(group, i);
2069 put_css_set(tc->cg);
2074 for_each_subsys(root, ss) {
2075 if (ss == failed_ss)
2077 if (ss->cancel_attach)
2078 ss->cancel_attach(cgrp, &tset);
2081 out_free_group_list:
2082 flex_array_free(group);
2087 * Find the task_struct of the task to attach by vpid and pass it along to the
2088 * function to attach either it or all tasks in its threadgroup. Will lock
2089 * cgroup_mutex and threadgroup; may take task_lock of task.
2091 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2093 struct task_struct *tsk;
2094 const struct cred *cred = current_cred(), *tcred;
2097 if (!cgroup_lock_live_group(cgrp))
2103 tsk = find_task_by_vpid(pid);
2107 goto out_unlock_cgroup;
2110 * even if we're attaching all tasks in the thread group, we
2111 * only need to check permissions on one of them.
2113 tcred = __task_cred(tsk);
2114 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2115 !uid_eq(cred->euid, tcred->uid) &&
2116 !uid_eq(cred->euid, tcred->suid)) {
2119 goto out_unlock_cgroup;
2125 tsk = tsk->group_leader;
2128 * Workqueue threads may acquire PF_THREAD_BOUND and become
2129 * trapped in a cpuset, or RT worker may be born in a cgroup
2130 * with no rt_runtime allocated. Just say no.
2132 if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2135 goto out_unlock_cgroup;
2138 get_task_struct(tsk);
2141 threadgroup_lock(tsk);
2143 if (!thread_group_leader(tsk)) {
2145 * a race with de_thread from another thread's exec()
2146 * may strip us of our leadership, if this happens,
2147 * there is no choice but to throw this task away and
2148 * try again; this is
2149 * "double-double-toil-and-trouble-check locking".
2151 threadgroup_unlock(tsk);
2152 put_task_struct(tsk);
2153 goto retry_find_task;
2157 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2159 threadgroup_unlock(tsk);
2161 put_task_struct(tsk);
2163 mutex_unlock(&cgroup_mutex);
2168 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2169 * @from: attach to all cgroups of a given task
2170 * @tsk: the task to be attached
2172 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2174 struct cgroupfs_root *root;
2177 mutex_lock(&cgroup_mutex);
2178 for_each_active_root(root) {
2179 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2181 retval = cgroup_attach_task(from_cg, tsk, false);
2185 mutex_unlock(&cgroup_mutex);
2189 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2191 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2193 return attach_task_by_pid(cgrp, pid, false);
2196 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2198 return attach_task_by_pid(cgrp, tgid, true);
2201 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2204 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2205 if (strlen(buffer) >= PATH_MAX)
2207 if (!cgroup_lock_live_group(cgrp))
2209 mutex_lock(&cgroup_root_mutex);
2210 strcpy(cgrp->root->release_agent_path, buffer);
2211 mutex_unlock(&cgroup_root_mutex);
2212 mutex_unlock(&cgroup_mutex);
2216 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2217 struct seq_file *seq)
2219 if (!cgroup_lock_live_group(cgrp))
2221 seq_puts(seq, cgrp->root->release_agent_path);
2222 seq_putc(seq, '\n');
2223 mutex_unlock(&cgroup_mutex);
2227 /* A buffer size big enough for numbers or short strings */
2228 #define CGROUP_LOCAL_BUFFER_SIZE 64
2230 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2232 const char __user *userbuf,
2233 size_t nbytes, loff_t *unused_ppos)
2235 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2241 if (nbytes >= sizeof(buffer))
2243 if (copy_from_user(buffer, userbuf, nbytes))
2246 buffer[nbytes] = 0; /* nul-terminate */
2247 if (cft->write_u64) {
2248 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2251 retval = cft->write_u64(cgrp, cft, val);
2253 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2256 retval = cft->write_s64(cgrp, cft, val);
2263 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2265 const char __user *userbuf,
2266 size_t nbytes, loff_t *unused_ppos)
2268 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2270 size_t max_bytes = cft->max_write_len;
2271 char *buffer = local_buffer;
2274 max_bytes = sizeof(local_buffer) - 1;
2275 if (nbytes >= max_bytes)
2277 /* Allocate a dynamic buffer if we need one */
2278 if (nbytes >= sizeof(local_buffer)) {
2279 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2283 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2288 buffer[nbytes] = 0; /* nul-terminate */
2289 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2293 if (buffer != local_buffer)
2298 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2299 size_t nbytes, loff_t *ppos)
2301 struct cftype *cft = __d_cft(file->f_dentry);
2302 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2304 if (cgroup_is_removed(cgrp))
2307 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2308 if (cft->write_u64 || cft->write_s64)
2309 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2310 if (cft->write_string)
2311 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2313 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2314 return ret ? ret : nbytes;
2319 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2321 char __user *buf, size_t nbytes,
2324 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2325 u64 val = cft->read_u64(cgrp, cft);
2326 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2328 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2331 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2333 char __user *buf, size_t nbytes,
2336 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2337 s64 val = cft->read_s64(cgrp, cft);
2338 int len = sprintf(tmp, "%lld\n", (long long) val);
2340 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2343 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2344 size_t nbytes, loff_t *ppos)
2346 struct cftype *cft = __d_cft(file->f_dentry);
2347 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2349 if (cgroup_is_removed(cgrp))
2353 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2355 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2357 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2362 * seqfile ops/methods for returning structured data. Currently just
2363 * supports string->u64 maps, but can be extended in future.
2366 struct cgroup_seqfile_state {
2368 struct cgroup *cgroup;
2371 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2373 struct seq_file *sf = cb->state;
2374 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2377 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2379 struct cgroup_seqfile_state *state = m->private;
2380 struct cftype *cft = state->cft;
2381 if (cft->read_map) {
2382 struct cgroup_map_cb cb = {
2383 .fill = cgroup_map_add,
2386 return cft->read_map(state->cgroup, cft, &cb);
2388 return cft->read_seq_string(state->cgroup, cft, m);
2391 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2393 struct seq_file *seq = file->private_data;
2394 kfree(seq->private);
2395 return single_release(inode, file);
2398 static const struct file_operations cgroup_seqfile_operations = {
2400 .write = cgroup_file_write,
2401 .llseek = seq_lseek,
2402 .release = cgroup_seqfile_release,
2405 static int cgroup_file_open(struct inode *inode, struct file *file)
2410 err = generic_file_open(inode, file);
2413 cft = __d_cft(file->f_dentry);
2415 if (cft->read_map || cft->read_seq_string) {
2416 struct cgroup_seqfile_state *state =
2417 kzalloc(sizeof(*state), GFP_USER);
2421 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2422 file->f_op = &cgroup_seqfile_operations;
2423 err = single_open(file, cgroup_seqfile_show, state);
2426 } else if (cft->open)
2427 err = cft->open(inode, file);
2434 static int cgroup_file_release(struct inode *inode, struct file *file)
2436 struct cftype *cft = __d_cft(file->f_dentry);
2438 return cft->release(inode, file);
2443 * cgroup_rename - Only allow simple rename of directories in place.
2445 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2446 struct inode *new_dir, struct dentry *new_dentry)
2449 struct cgroup_name *name, *old_name;
2450 struct cgroup *cgrp;
2453 * It's convinient to use parent dir's i_mutex to protected
2456 lockdep_assert_held(&old_dir->i_mutex);
2458 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2460 if (new_dentry->d_inode)
2462 if (old_dir != new_dir)
2465 cgrp = __d_cgrp(old_dentry);
2467 name = cgroup_alloc_name(new_dentry);
2471 ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2477 old_name = cgrp->name;
2478 rcu_assign_pointer(cgrp->name, name);
2480 kfree_rcu(old_name, rcu_head);
2484 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2486 if (S_ISDIR(dentry->d_inode->i_mode))
2487 return &__d_cgrp(dentry)->xattrs;
2489 return &__d_cft(dentry)->xattrs;
2492 static inline int xattr_enabled(struct dentry *dentry)
2494 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2495 return test_bit(ROOT_XATTR, &root->flags);
2498 static bool is_valid_xattr(const char *name)
2500 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2501 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2506 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2507 const void *val, size_t size, int flags)
2509 if (!xattr_enabled(dentry))
2511 if (!is_valid_xattr(name))
2513 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2516 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2518 if (!xattr_enabled(dentry))
2520 if (!is_valid_xattr(name))
2522 return simple_xattr_remove(__d_xattrs(dentry), name);
2525 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2526 void *buf, size_t size)
2528 if (!xattr_enabled(dentry))
2530 if (!is_valid_xattr(name))
2532 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2535 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2537 if (!xattr_enabled(dentry))
2539 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2542 static const struct file_operations cgroup_file_operations = {
2543 .read = cgroup_file_read,
2544 .write = cgroup_file_write,
2545 .llseek = generic_file_llseek,
2546 .open = cgroup_file_open,
2547 .release = cgroup_file_release,
2550 static const struct inode_operations cgroup_file_inode_operations = {
2551 .setxattr = cgroup_setxattr,
2552 .getxattr = cgroup_getxattr,
2553 .listxattr = cgroup_listxattr,
2554 .removexattr = cgroup_removexattr,
2557 static const struct inode_operations cgroup_dir_inode_operations = {
2558 .lookup = cgroup_lookup,
2559 .mkdir = cgroup_mkdir,
2560 .rmdir = cgroup_rmdir,
2561 .rename = cgroup_rename,
2562 .setxattr = cgroup_setxattr,
2563 .getxattr = cgroup_getxattr,
2564 .listxattr = cgroup_listxattr,
2565 .removexattr = cgroup_removexattr,
2568 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2570 if (dentry->d_name.len > NAME_MAX)
2571 return ERR_PTR(-ENAMETOOLONG);
2572 d_add(dentry, NULL);
2577 * Check if a file is a control file
2579 static inline struct cftype *__file_cft(struct file *file)
2581 if (file_inode(file)->i_fop != &cgroup_file_operations)
2582 return ERR_PTR(-EINVAL);
2583 return __d_cft(file->f_dentry);
2586 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2587 struct super_block *sb)
2589 struct inode *inode;
2593 if (dentry->d_inode)
2596 inode = cgroup_new_inode(mode, sb);
2600 if (S_ISDIR(mode)) {
2601 inode->i_op = &cgroup_dir_inode_operations;
2602 inode->i_fop = &simple_dir_operations;
2604 /* start off with i_nlink == 2 (for "." entry) */
2606 inc_nlink(dentry->d_parent->d_inode);
2609 * Control reaches here with cgroup_mutex held.
2610 * @inode->i_mutex should nest outside cgroup_mutex but we
2611 * want to populate it immediately without releasing
2612 * cgroup_mutex. As @inode isn't visible to anyone else
2613 * yet, trylock will always succeed without affecting
2616 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2617 } else if (S_ISREG(mode)) {
2619 inode->i_fop = &cgroup_file_operations;
2620 inode->i_op = &cgroup_file_inode_operations;
2622 d_instantiate(dentry, inode);
2623 dget(dentry); /* Extra count - pin the dentry in core */
2628 * cgroup_file_mode - deduce file mode of a control file
2629 * @cft: the control file in question
2631 * returns cft->mode if ->mode is not 0
2632 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2633 * returns S_IRUGO if it has only a read handler
2634 * returns S_IWUSR if it has only a write hander
2636 static umode_t cgroup_file_mode(const struct cftype *cft)
2643 if (cft->read || cft->read_u64 || cft->read_s64 ||
2644 cft->read_map || cft->read_seq_string)
2647 if (cft->write || cft->write_u64 || cft->write_s64 ||
2648 cft->write_string || cft->trigger)
2654 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2657 struct dentry *dir = cgrp->dentry;
2658 struct cgroup *parent = __d_cgrp(dir);
2659 struct dentry *dentry;
2663 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2665 simple_xattrs_init(&cft->xattrs);
2667 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2668 strcpy(name, subsys->name);
2671 strcat(name, cft->name);
2673 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2675 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2679 dentry = lookup_one_len(name, dir, strlen(name));
2680 if (IS_ERR(dentry)) {
2681 error = PTR_ERR(dentry);
2685 mode = cgroup_file_mode(cft);
2686 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2688 cfe->type = (void *)cft;
2689 cfe->dentry = dentry;
2690 dentry->d_fsdata = cfe;
2691 list_add_tail(&cfe->node, &parent->files);
2700 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2701 struct cftype cfts[], bool is_add)
2706 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2707 /* does cft->flags tell us to skip this file on @cgrp? */
2708 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2710 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2714 err = cgroup_add_file(cgrp, subsys, cft);
2716 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2720 cgroup_rm_file(cgrp, cft);
2726 static DEFINE_MUTEX(cgroup_cft_mutex);
2728 static void cgroup_cfts_prepare(void)
2729 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2732 * Thanks to the entanglement with vfs inode locking, we can't walk
2733 * the existing cgroups under cgroup_mutex and create files.
2734 * Instead, we increment reference on all cgroups and build list of
2735 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2736 * exclusive access to the field.
2738 mutex_lock(&cgroup_cft_mutex);
2739 mutex_lock(&cgroup_mutex);
2742 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2743 struct cftype *cfts, bool is_add)
2744 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2747 struct cgroup *cgrp, *n;
2749 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2750 if (cfts && ss->root != &rootnode) {
2751 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2753 list_add_tail(&cgrp->cft_q_node, &pending);
2757 mutex_unlock(&cgroup_mutex);
2760 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2761 * files for all cgroups which were created before.
2763 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2764 struct inode *inode = cgrp->dentry->d_inode;
2766 mutex_lock(&inode->i_mutex);
2767 mutex_lock(&cgroup_mutex);
2768 if (!cgroup_is_removed(cgrp))
2769 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2770 mutex_unlock(&cgroup_mutex);
2771 mutex_unlock(&inode->i_mutex);
2773 list_del_init(&cgrp->cft_q_node);
2777 mutex_unlock(&cgroup_cft_mutex);
2781 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2782 * @ss: target cgroup subsystem
2783 * @cfts: zero-length name terminated array of cftypes
2785 * Register @cfts to @ss. Files described by @cfts are created for all
2786 * existing cgroups to which @ss is attached and all future cgroups will
2787 * have them too. This function can be called anytime whether @ss is
2790 * Returns 0 on successful registration, -errno on failure. Note that this
2791 * function currently returns 0 as long as @cfts registration is successful
2792 * even if some file creation attempts on existing cgroups fail.
2794 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2796 struct cftype_set *set;
2798 set = kzalloc(sizeof(*set), GFP_KERNEL);
2802 cgroup_cfts_prepare();
2804 list_add_tail(&set->node, &ss->cftsets);
2805 cgroup_cfts_commit(ss, cfts, true);
2809 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2812 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2813 * @ss: target cgroup subsystem
2814 * @cfts: zero-length name terminated array of cftypes
2816 * Unregister @cfts from @ss. Files described by @cfts are removed from
2817 * all existing cgroups to which @ss is attached and all future cgroups
2818 * won't have them either. This function can be called anytime whether @ss
2819 * is attached or not.
2821 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2822 * registered with @ss.
2824 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2826 struct cftype_set *set;
2828 cgroup_cfts_prepare();
2830 list_for_each_entry(set, &ss->cftsets, node) {
2831 if (set->cfts == cfts) {
2832 list_del_init(&set->node);
2833 cgroup_cfts_commit(ss, cfts, false);
2838 cgroup_cfts_commit(ss, NULL, false);
2843 * cgroup_task_count - count the number of tasks in a cgroup.
2844 * @cgrp: the cgroup in question
2846 * Return the number of tasks in the cgroup.
2848 int cgroup_task_count(const struct cgroup *cgrp)
2851 struct cg_cgroup_link *link;
2853 read_lock(&css_set_lock);
2854 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2855 count += atomic_read(&link->cg->refcount);
2857 read_unlock(&css_set_lock);
2862 * Advance a list_head iterator. The iterator should be positioned at
2863 * the start of a css_set
2865 static void cgroup_advance_iter(struct cgroup *cgrp,
2866 struct cgroup_iter *it)
2868 struct list_head *l = it->cg_link;
2869 struct cg_cgroup_link *link;
2872 /* Advance to the next non-empty css_set */
2875 if (l == &cgrp->css_sets) {
2879 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2881 } while (list_empty(&cg->tasks));
2883 it->task = cg->tasks.next;
2887 * To reduce the fork() overhead for systems that are not actually
2888 * using their cgroups capability, we don't maintain the lists running
2889 * through each css_set to its tasks until we see the list actually
2890 * used - in other words after the first call to cgroup_iter_start().
2892 static void cgroup_enable_task_cg_lists(void)
2894 struct task_struct *p, *g;
2895 write_lock(&css_set_lock);
2896 use_task_css_set_links = 1;
2898 * We need tasklist_lock because RCU is not safe against
2899 * while_each_thread(). Besides, a forking task that has passed
2900 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2901 * is not guaranteed to have its child immediately visible in the
2902 * tasklist if we walk through it with RCU.
2904 read_lock(&tasklist_lock);
2905 do_each_thread(g, p) {
2908 * We should check if the process is exiting, otherwise
2909 * it will race with cgroup_exit() in that the list
2910 * entry won't be deleted though the process has exited.
2912 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2913 list_add(&p->cg_list, &p->cgroups->tasks);
2915 } while_each_thread(g, p);
2916 read_unlock(&tasklist_lock);
2917 write_unlock(&css_set_lock);
2921 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2922 * @pos: the current position (%NULL to initiate traversal)
2923 * @cgroup: cgroup whose descendants to walk
2925 * To be used by cgroup_for_each_descendant_pre(). Find the next
2926 * descendant to visit for pre-order traversal of @cgroup's descendants.
2928 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
2929 struct cgroup *cgroup)
2931 struct cgroup *next;
2933 WARN_ON_ONCE(!rcu_read_lock_held());
2935 /* if first iteration, pretend we just visited @cgroup */
2937 if (list_empty(&cgroup->children))
2942 /* visit the first child if exists */
2943 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
2947 /* no child, visit my or the closest ancestor's next sibling */
2949 next = list_entry_rcu(pos->sibling.next, struct cgroup,
2951 if (&next->sibling != &pos->parent->children)
2955 } while (pos != cgroup);
2959 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
2962 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
2963 * @pos: cgroup of interest
2965 * Return the rightmost descendant of @pos. If there's no descendant,
2966 * @pos is returned. This can be used during pre-order traversal to skip
2969 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
2971 struct cgroup *last, *tmp;
2973 WARN_ON_ONCE(!rcu_read_lock_held());
2977 /* ->prev isn't RCU safe, walk ->next till the end */
2979 list_for_each_entry_rcu(tmp, &last->children, sibling)
2985 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
2987 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
2989 struct cgroup *last;
2993 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3001 * cgroup_next_descendant_post - find the next descendant for post-order walk
3002 * @pos: the current position (%NULL to initiate traversal)
3003 * @cgroup: cgroup whose descendants to walk
3005 * To be used by cgroup_for_each_descendant_post(). Find the next
3006 * descendant to visit for post-order traversal of @cgroup's descendants.
3008 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3009 struct cgroup *cgroup)
3011 struct cgroup *next;
3013 WARN_ON_ONCE(!rcu_read_lock_held());
3015 /* if first iteration, visit the leftmost descendant */
3017 next = cgroup_leftmost_descendant(cgroup);
3018 return next != cgroup ? next : NULL;
3021 /* if there's an unvisited sibling, visit its leftmost descendant */
3022 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3023 if (&next->sibling != &pos->parent->children)
3024 return cgroup_leftmost_descendant(next);
3026 /* no sibling left, visit parent */
3028 return next != cgroup ? next : NULL;
3030 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3032 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3033 __acquires(css_set_lock)
3036 * The first time anyone tries to iterate across a cgroup,
3037 * we need to enable the list linking each css_set to its
3038 * tasks, and fix up all existing tasks.
3040 if (!use_task_css_set_links)
3041 cgroup_enable_task_cg_lists();
3043 read_lock(&css_set_lock);
3044 it->cg_link = &cgrp->css_sets;
3045 cgroup_advance_iter(cgrp, it);
3048 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3049 struct cgroup_iter *it)
3051 struct task_struct *res;
3052 struct list_head *l = it->task;
3053 struct cg_cgroup_link *link;
3055 /* If the iterator cg is NULL, we have no tasks */
3058 res = list_entry(l, struct task_struct, cg_list);
3059 /* Advance iterator to find next entry */
3061 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
3062 if (l == &link->cg->tasks) {
3063 /* We reached the end of this task list - move on to
3064 * the next cg_cgroup_link */
3065 cgroup_advance_iter(cgrp, it);
3072 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3073 __releases(css_set_lock)
3075 read_unlock(&css_set_lock);
3078 static inline int started_after_time(struct task_struct *t1,
3079 struct timespec *time,
3080 struct task_struct *t2)
3082 int start_diff = timespec_compare(&t1->start_time, time);
3083 if (start_diff > 0) {
3085 } else if (start_diff < 0) {
3089 * Arbitrarily, if two processes started at the same
3090 * time, we'll say that the lower pointer value
3091 * started first. Note that t2 may have exited by now
3092 * so this may not be a valid pointer any longer, but
3093 * that's fine - it still serves to distinguish
3094 * between two tasks started (effectively) simultaneously.
3101 * This function is a callback from heap_insert() and is used to order
3103 * In this case we order the heap in descending task start time.
3105 static inline int started_after(void *p1, void *p2)
3107 struct task_struct *t1 = p1;
3108 struct task_struct *t2 = p2;
3109 return started_after_time(t1, &t2->start_time, t2);
3113 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3114 * @scan: struct cgroup_scanner containing arguments for the scan
3116 * Arguments include pointers to callback functions test_task() and
3118 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3119 * and if it returns true, call process_task() for it also.
3120 * The test_task pointer may be NULL, meaning always true (select all tasks).
3121 * Effectively duplicates cgroup_iter_{start,next,end}()
3122 * but does not lock css_set_lock for the call to process_task().
3123 * The struct cgroup_scanner may be embedded in any structure of the caller's
3125 * It is guaranteed that process_task() will act on every task that
3126 * is a member of the cgroup for the duration of this call. This
3127 * function may or may not call process_task() for tasks that exit
3128 * or move to a different cgroup during the call, or are forked or
3129 * move into the cgroup during the call.
3131 * Note that test_task() may be called with locks held, and may in some
3132 * situations be called multiple times for the same task, so it should
3134 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3135 * pre-allocated and will be used for heap operations (and its "gt" member will
3136 * be overwritten), else a temporary heap will be used (allocation of which
3137 * may cause this function to fail).
3139 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3142 struct cgroup_iter it;
3143 struct task_struct *p, *dropped;
3144 /* Never dereference latest_task, since it's not refcounted */
3145 struct task_struct *latest_task = NULL;
3146 struct ptr_heap tmp_heap;
3147 struct ptr_heap *heap;
3148 struct timespec latest_time = { 0, 0 };
3151 /* The caller supplied our heap and pre-allocated its memory */
3153 heap->gt = &started_after;
3155 /* We need to allocate our own heap memory */
3157 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3159 /* cannot allocate the heap */
3165 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3166 * to determine which are of interest, and using the scanner's
3167 * "process_task" callback to process any of them that need an update.
3168 * Since we don't want to hold any locks during the task updates,
3169 * gather tasks to be processed in a heap structure.
3170 * The heap is sorted by descending task start time.
3171 * If the statically-sized heap fills up, we overflow tasks that
3172 * started later, and in future iterations only consider tasks that
3173 * started after the latest task in the previous pass. This
3174 * guarantees forward progress and that we don't miss any tasks.
3177 cgroup_iter_start(scan->cg, &it);
3178 while ((p = cgroup_iter_next(scan->cg, &it))) {
3180 * Only affect tasks that qualify per the caller's callback,
3181 * if he provided one
3183 if (scan->test_task && !scan->test_task(p, scan))
3186 * Only process tasks that started after the last task
3189 if (!started_after_time(p, &latest_time, latest_task))
3191 dropped = heap_insert(heap, p);
3192 if (dropped == NULL) {
3194 * The new task was inserted; the heap wasn't
3198 } else if (dropped != p) {
3200 * The new task was inserted, and pushed out a
3204 put_task_struct(dropped);
3207 * Else the new task was newer than anything already in
3208 * the heap and wasn't inserted
3211 cgroup_iter_end(scan->cg, &it);
3214 for (i = 0; i < heap->size; i++) {
3215 struct task_struct *q = heap->ptrs[i];
3217 latest_time = q->start_time;
3220 /* Process the task per the caller's callback */
3221 scan->process_task(q, scan);
3225 * If we had to process any tasks at all, scan again
3226 * in case some of them were in the middle of forking
3227 * children that didn't get processed.
3228 * Not the most efficient way to do it, but it avoids
3229 * having to take callback_mutex in the fork path
3233 if (heap == &tmp_heap)
3234 heap_free(&tmp_heap);
3238 static void cgroup_transfer_one_task(struct task_struct *task,
3239 struct cgroup_scanner *scan)
3241 struct cgroup *new_cgroup = scan->data;
3243 mutex_lock(&cgroup_mutex);
3244 cgroup_attach_task(new_cgroup, task, false);
3245 mutex_unlock(&cgroup_mutex);
3249 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3250 * @to: cgroup to which the tasks will be moved
3251 * @from: cgroup in which the tasks currently reside
3253 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3255 struct cgroup_scanner scan;
3258 scan.test_task = NULL; /* select all tasks in cgroup */
3259 scan.process_task = cgroup_transfer_one_task;
3263 return cgroup_scan_tasks(&scan);
3267 * Stuff for reading the 'tasks'/'procs' files.
3269 * Reading this file can return large amounts of data if a cgroup has
3270 * *lots* of attached tasks. So it may need several calls to read(),
3271 * but we cannot guarantee that the information we produce is correct
3272 * unless we produce it entirely atomically.
3276 /* which pidlist file are we talking about? */
3277 enum cgroup_filetype {
3283 * A pidlist is a list of pids that virtually represents the contents of one
3284 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3285 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3288 struct cgroup_pidlist {
3290 * used to find which pidlist is wanted. doesn't change as long as
3291 * this particular list stays in the list.
3293 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3296 /* how many elements the above list has */
3298 /* how many files are using the current array */
3300 /* each of these stored in a list by its cgroup */
3301 struct list_head links;
3302 /* pointer to the cgroup we belong to, for list removal purposes */
3303 struct cgroup *owner;
3304 /* protects the other fields */
3305 struct rw_semaphore mutex;
3309 * The following two functions "fix" the issue where there are more pids
3310 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3311 * TODO: replace with a kernel-wide solution to this problem
3313 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3314 static void *pidlist_allocate(int count)
3316 if (PIDLIST_TOO_LARGE(count))
3317 return vmalloc(count * sizeof(pid_t));
3319 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3321 static void pidlist_free(void *p)
3323 if (is_vmalloc_addr(p))
3330 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3331 * Returns the number of unique elements.
3333 static int pidlist_uniq(pid_t *list, int length)
3338 * we presume the 0th element is unique, so i starts at 1. trivial
3339 * edge cases first; no work needs to be done for either
3341 if (length == 0 || length == 1)
3343 /* src and dest walk down the list; dest counts unique elements */
3344 for (src = 1; src < length; src++) {
3345 /* find next unique element */
3346 while (list[src] == list[src-1]) {
3351 /* dest always points to where the next unique element goes */
3352 list[dest] = list[src];
3359 static int cmppid(const void *a, const void *b)
3361 return *(pid_t *)a - *(pid_t *)b;
3365 * find the appropriate pidlist for our purpose (given procs vs tasks)
3366 * returns with the lock on that pidlist already held, and takes care
3367 * of the use count, or returns NULL with no locks held if we're out of
3370 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3371 enum cgroup_filetype type)
3373 struct cgroup_pidlist *l;
3374 /* don't need task_nsproxy() if we're looking at ourself */
3375 struct pid_namespace *ns = task_active_pid_ns(current);
3378 * We can't drop the pidlist_mutex before taking the l->mutex in case
3379 * the last ref-holder is trying to remove l from the list at the same
3380 * time. Holding the pidlist_mutex precludes somebody taking whichever
3381 * list we find out from under us - compare release_pid_array().
3383 mutex_lock(&cgrp->pidlist_mutex);
3384 list_for_each_entry(l, &cgrp->pidlists, links) {
3385 if (l->key.type == type && l->key.ns == ns) {
3386 /* make sure l doesn't vanish out from under us */
3387 down_write(&l->mutex);
3388 mutex_unlock(&cgrp->pidlist_mutex);
3392 /* entry not found; create a new one */
3393 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3395 mutex_unlock(&cgrp->pidlist_mutex);
3398 init_rwsem(&l->mutex);
3399 down_write(&l->mutex);
3401 l->key.ns = get_pid_ns(ns);
3402 l->use_count = 0; /* don't increment here */
3405 list_add(&l->links, &cgrp->pidlists);
3406 mutex_unlock(&cgrp->pidlist_mutex);
3411 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3413 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3414 struct cgroup_pidlist **lp)
3418 int pid, n = 0; /* used for populating the array */
3419 struct cgroup_iter it;
3420 struct task_struct *tsk;
3421 struct cgroup_pidlist *l;
3424 * If cgroup gets more users after we read count, we won't have
3425 * enough space - tough. This race is indistinguishable to the
3426 * caller from the case that the additional cgroup users didn't
3427 * show up until sometime later on.
3429 length = cgroup_task_count(cgrp);
3430 array = pidlist_allocate(length);
3433 /* now, populate the array */
3434 cgroup_iter_start(cgrp, &it);
3435 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3436 if (unlikely(n == length))
3438 /* get tgid or pid for procs or tasks file respectively */
3439 if (type == CGROUP_FILE_PROCS)
3440 pid = task_tgid_vnr(tsk);
3442 pid = task_pid_vnr(tsk);
3443 if (pid > 0) /* make sure to only use valid results */
3446 cgroup_iter_end(cgrp, &it);
3448 /* now sort & (if procs) strip out duplicates */
3449 sort(array, length, sizeof(pid_t), cmppid, NULL);
3450 if (type == CGROUP_FILE_PROCS)
3451 length = pidlist_uniq(array, length);
3452 l = cgroup_pidlist_find(cgrp, type);
3454 pidlist_free(array);
3457 /* store array, freeing old if necessary - lock already held */
3458 pidlist_free(l->list);
3462 up_write(&l->mutex);
3468 * cgroupstats_build - build and fill cgroupstats
3469 * @stats: cgroupstats to fill information into
3470 * @dentry: A dentry entry belonging to the cgroup for which stats have
3473 * Build and fill cgroupstats so that taskstats can export it to user
3476 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3479 struct cgroup *cgrp;
3480 struct cgroup_iter it;
3481 struct task_struct *tsk;
3484 * Validate dentry by checking the superblock operations,
3485 * and make sure it's a directory.
3487 if (dentry->d_sb->s_op != &cgroup_ops ||
3488 !S_ISDIR(dentry->d_inode->i_mode))
3492 cgrp = dentry->d_fsdata;
3494 cgroup_iter_start(cgrp, &it);
3495 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3496 switch (tsk->state) {
3498 stats->nr_running++;
3500 case TASK_INTERRUPTIBLE:
3501 stats->nr_sleeping++;
3503 case TASK_UNINTERRUPTIBLE:
3504 stats->nr_uninterruptible++;
3507 stats->nr_stopped++;
3510 if (delayacct_is_task_waiting_on_io(tsk))
3511 stats->nr_io_wait++;
3515 cgroup_iter_end(cgrp, &it);
3523 * seq_file methods for the tasks/procs files. The seq_file position is the
3524 * next pid to display; the seq_file iterator is a pointer to the pid
3525 * in the cgroup->l->list array.
3528 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3531 * Initially we receive a position value that corresponds to
3532 * one more than the last pid shown (or 0 on the first call or
3533 * after a seek to the start). Use a binary-search to find the
3534 * next pid to display, if any
3536 struct cgroup_pidlist *l = s->private;
3537 int index = 0, pid = *pos;
3540 down_read(&l->mutex);
3542 int end = l->length;
3544 while (index < end) {
3545 int mid = (index + end) / 2;
3546 if (l->list[mid] == pid) {
3549 } else if (l->list[mid] <= pid)
3555 /* If we're off the end of the array, we're done */
3556 if (index >= l->length)
3558 /* Update the abstract position to be the actual pid that we found */
3559 iter = l->list + index;
3564 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3566 struct cgroup_pidlist *l = s->private;
3570 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3572 struct cgroup_pidlist *l = s->private;
3574 pid_t *end = l->list + l->length;
3576 * Advance to the next pid in the array. If this goes off the
3588 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3590 return seq_printf(s, "%d\n", *(int *)v);
3594 * seq_operations functions for iterating on pidlists through seq_file -
3595 * independent of whether it's tasks or procs
3597 static const struct seq_operations cgroup_pidlist_seq_operations = {
3598 .start = cgroup_pidlist_start,
3599 .stop = cgroup_pidlist_stop,
3600 .next = cgroup_pidlist_next,
3601 .show = cgroup_pidlist_show,
3604 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3607 * the case where we're the last user of this particular pidlist will
3608 * have us remove it from the cgroup's list, which entails taking the
3609 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3610 * pidlist_mutex, we have to take pidlist_mutex first.
3612 mutex_lock(&l->owner->pidlist_mutex);
3613 down_write(&l->mutex);
3614 BUG_ON(!l->use_count);
3615 if (!--l->use_count) {
3616 /* we're the last user if refcount is 0; remove and free */
3617 list_del(&l->links);
3618 mutex_unlock(&l->owner->pidlist_mutex);
3619 pidlist_free(l->list);
3620 put_pid_ns(l->key.ns);
3621 up_write(&l->mutex);
3625 mutex_unlock(&l->owner->pidlist_mutex);
3626 up_write(&l->mutex);
3629 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3631 struct cgroup_pidlist *l;
3632 if (!(file->f_mode & FMODE_READ))
3635 * the seq_file will only be initialized if the file was opened for
3636 * reading; hence we check if it's not null only in that case.
3638 l = ((struct seq_file *)file->private_data)->private;
3639 cgroup_release_pid_array(l);
3640 return seq_release(inode, file);
3643 static const struct file_operations cgroup_pidlist_operations = {
3645 .llseek = seq_lseek,
3646 .write = cgroup_file_write,
3647 .release = cgroup_pidlist_release,
3651 * The following functions handle opens on a file that displays a pidlist
3652 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3655 /* helper function for the two below it */
3656 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3658 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3659 struct cgroup_pidlist *l;
3662 /* Nothing to do for write-only files */
3663 if (!(file->f_mode & FMODE_READ))
3666 /* have the array populated */
3667 retval = pidlist_array_load(cgrp, type, &l);
3670 /* configure file information */
3671 file->f_op = &cgroup_pidlist_operations;
3673 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3675 cgroup_release_pid_array(l);
3678 ((struct seq_file *)file->private_data)->private = l;
3681 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3683 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3685 static int cgroup_procs_open(struct inode *unused, struct file *file)
3687 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3690 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3693 return notify_on_release(cgrp);
3696 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3700 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3702 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3704 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3709 * Unregister event and free resources.
3711 * Gets called from workqueue.
3713 static void cgroup_event_remove(struct work_struct *work)
3715 struct cgroup_event *event = container_of(work, struct cgroup_event,
3717 struct cgroup *cgrp = event->cgrp;
3719 remove_wait_queue(event->wqh, &event->wait);
3721 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3723 /* Notify userspace the event is going away. */
3724 eventfd_signal(event->eventfd, 1);
3726 eventfd_ctx_put(event->eventfd);
3732 * Gets called on POLLHUP on eventfd when user closes it.
3734 * Called with wqh->lock held and interrupts disabled.
3736 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3737 int sync, void *key)
3739 struct cgroup_event *event = container_of(wait,
3740 struct cgroup_event, wait);
3741 struct cgroup *cgrp = event->cgrp;
3742 unsigned long flags = (unsigned long)key;
3744 if (flags & POLLHUP) {
3746 * If the event has been detached at cgroup removal, we
3747 * can simply return knowing the other side will cleanup
3750 * We can't race against event freeing since the other
3751 * side will require wqh->lock via remove_wait_queue(),
3754 spin_lock(&cgrp->event_list_lock);
3755 if (!list_empty(&event->list)) {
3756 list_del_init(&event->list);
3758 * We are in atomic context, but cgroup_event_remove()
3759 * may sleep, so we have to call it in workqueue.
3761 schedule_work(&event->remove);
3763 spin_unlock(&cgrp->event_list_lock);
3769 static void cgroup_event_ptable_queue_proc(struct file *file,
3770 wait_queue_head_t *wqh, poll_table *pt)
3772 struct cgroup_event *event = container_of(pt,
3773 struct cgroup_event, pt);
3776 add_wait_queue(wqh, &event->wait);
3780 * Parse input and register new cgroup event handler.
3782 * Input must be in format '<event_fd> <control_fd> <args>'.
3783 * Interpretation of args is defined by control file implementation.
3785 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3788 struct cgroup_event *event = NULL;
3789 struct cgroup *cgrp_cfile;
3790 unsigned int efd, cfd;
3791 struct file *efile = NULL;
3792 struct file *cfile = NULL;
3796 efd = simple_strtoul(buffer, &endp, 10);
3801 cfd = simple_strtoul(buffer, &endp, 10);
3802 if ((*endp != ' ') && (*endp != '\0'))
3806 event = kzalloc(sizeof(*event), GFP_KERNEL);
3810 INIT_LIST_HEAD(&event->list);
3811 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3812 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3813 INIT_WORK(&event->remove, cgroup_event_remove);
3815 efile = eventfd_fget(efd);
3816 if (IS_ERR(efile)) {
3817 ret = PTR_ERR(efile);
3821 event->eventfd = eventfd_ctx_fileget(efile);
3822 if (IS_ERR(event->eventfd)) {
3823 ret = PTR_ERR(event->eventfd);
3833 /* the process need read permission on control file */
3834 /* AV: shouldn't we check that it's been opened for read instead? */
3835 ret = inode_permission(file_inode(cfile), MAY_READ);
3839 event->cft = __file_cft(cfile);
3840 if (IS_ERR(event->cft)) {
3841 ret = PTR_ERR(event->cft);
3846 * The file to be monitored must be in the same cgroup as
3847 * cgroup.event_control is.
3849 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
3850 if (cgrp_cfile != cgrp) {
3855 if (!event->cft->register_event || !event->cft->unregister_event) {
3860 ret = event->cft->register_event(cgrp, event->cft,
3861 event->eventfd, buffer);
3866 * Events should be removed after rmdir of cgroup directory, but before
3867 * destroying subsystem state objects. Let's take reference to cgroup
3868 * directory dentry to do that.
3872 spin_lock(&cgrp->event_list_lock);
3873 list_add(&event->list, &cgrp->event_list);
3874 spin_unlock(&cgrp->event_list_lock);
3885 if (event && event->eventfd && !IS_ERR(event->eventfd))
3886 eventfd_ctx_put(event->eventfd);
3888 if (!IS_ERR_OR_NULL(efile))
3896 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3899 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3902 static int cgroup_clone_children_write(struct cgroup *cgrp,
3907 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3909 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3914 * for the common functions, 'private' gives the type of file
3916 /* for hysterical raisins, we can't put this on the older files */
3917 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3918 static struct cftype files[] = {
3921 .open = cgroup_tasks_open,
3922 .write_u64 = cgroup_tasks_write,
3923 .release = cgroup_pidlist_release,
3924 .mode = S_IRUGO | S_IWUSR,
3927 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3928 .open = cgroup_procs_open,
3929 .write_u64 = cgroup_procs_write,
3930 .release = cgroup_pidlist_release,
3931 .mode = S_IRUGO | S_IWUSR,
3934 .name = "notify_on_release",
3935 .read_u64 = cgroup_read_notify_on_release,
3936 .write_u64 = cgroup_write_notify_on_release,
3939 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3940 .write_string = cgroup_write_event_control,
3944 .name = "cgroup.clone_children",
3945 .read_u64 = cgroup_clone_children_read,
3946 .write_u64 = cgroup_clone_children_write,
3949 .name = "release_agent",
3950 .flags = CFTYPE_ONLY_ON_ROOT,
3951 .read_seq_string = cgroup_release_agent_show,
3952 .write_string = cgroup_release_agent_write,
3953 .max_write_len = PATH_MAX,
3959 * cgroup_populate_dir - selectively creation of files in a directory
3960 * @cgrp: target cgroup
3961 * @base_files: true if the base files should be added
3962 * @subsys_mask: mask of the subsystem ids whose files should be added
3964 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
3965 unsigned long subsys_mask)
3968 struct cgroup_subsys *ss;
3971 err = cgroup_addrm_files(cgrp, NULL, files, true);
3976 /* process cftsets of each subsystem */
3977 for_each_subsys(cgrp->root, ss) {
3978 struct cftype_set *set;
3979 if (!test_bit(ss->subsys_id, &subsys_mask))
3982 list_for_each_entry(set, &ss->cftsets, node)
3983 cgroup_addrm_files(cgrp, ss, set->cfts, true);
3986 /* This cgroup is ready now */
3987 for_each_subsys(cgrp->root, ss) {
3988 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3990 * Update id->css pointer and make this css visible from
3991 * CSS ID functions. This pointer will be dereferened
3992 * from RCU-read-side without locks.
3995 rcu_assign_pointer(css->id->css, css);
4001 static void css_dput_fn(struct work_struct *work)
4003 struct cgroup_subsys_state *css =
4004 container_of(work, struct cgroup_subsys_state, dput_work);
4005 struct dentry *dentry = css->cgroup->dentry;
4006 struct super_block *sb = dentry->d_sb;
4008 atomic_inc(&sb->s_active);
4010 deactivate_super(sb);
4013 static void init_cgroup_css(struct cgroup_subsys_state *css,
4014 struct cgroup_subsys *ss,
4015 struct cgroup *cgrp)
4018 atomic_set(&css->refcnt, 1);
4021 if (cgrp == dummytop)
4022 css->flags |= CSS_ROOT;
4023 BUG_ON(cgrp->subsys[ss->subsys_id]);
4024 cgrp->subsys[ss->subsys_id] = css;
4027 * css holds an extra ref to @cgrp->dentry which is put on the last
4028 * css_put(). dput() requires process context, which css_put() may
4029 * be called without. @css->dput_work will be used to invoke
4030 * dput() asynchronously from css_put().
4032 INIT_WORK(&css->dput_work, css_dput_fn);
4035 /* invoke ->post_create() on a new CSS and mark it online if successful */
4036 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4040 lockdep_assert_held(&cgroup_mutex);
4043 ret = ss->css_online(cgrp);
4045 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4049 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4050 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4051 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4053 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4055 lockdep_assert_held(&cgroup_mutex);
4057 if (!(css->flags & CSS_ONLINE))
4060 if (ss->css_offline)
4061 ss->css_offline(cgrp);
4063 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4067 * cgroup_create - create a cgroup
4068 * @parent: cgroup that will be parent of the new cgroup
4069 * @dentry: dentry of the new cgroup
4070 * @mode: mode to set on new inode
4072 * Must be called with the mutex on the parent inode held
4074 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4077 struct cgroup *cgrp;
4078 struct cgroup_name *name;
4079 struct cgroupfs_root *root = parent->root;
4081 struct cgroup_subsys *ss;
4082 struct super_block *sb = root->sb;
4084 /* allocate the cgroup and its ID, 0 is reserved for the root */
4085 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4089 name = cgroup_alloc_name(dentry);
4092 rcu_assign_pointer(cgrp->name, name);
4094 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4099 * Only live parents can have children. Note that the liveliness
4100 * check isn't strictly necessary because cgroup_mkdir() and
4101 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4102 * anyway so that locking is contained inside cgroup proper and we
4103 * don't get nasty surprises if we ever grow another caller.
4105 if (!cgroup_lock_live_group(parent)) {
4110 /* Grab a reference on the superblock so the hierarchy doesn't
4111 * get deleted on unmount if there are child cgroups. This
4112 * can be done outside cgroup_mutex, since the sb can't
4113 * disappear while someone has an open control file on the
4115 atomic_inc(&sb->s_active);
4117 init_cgroup_housekeeping(cgrp);
4119 dentry->d_fsdata = cgrp;
4120 cgrp->dentry = dentry;
4122 cgrp->parent = parent;
4123 cgrp->root = parent->root;
4124 cgrp->top_cgroup = parent->top_cgroup;
4126 if (notify_on_release(parent))
4127 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4129 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4130 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4132 for_each_subsys(root, ss) {
4133 struct cgroup_subsys_state *css;
4135 css = ss->css_alloc(cgrp);
4140 init_cgroup_css(css, ss, cgrp);
4142 err = alloc_css_id(ss, parent, cgrp);
4149 * Create directory. cgroup_create_file() returns with the new
4150 * directory locked on success so that it can be populated without
4151 * dropping cgroup_mutex.
4153 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4156 lockdep_assert_held(&dentry->d_inode->i_mutex);
4158 /* allocation complete, commit to creation */
4159 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4160 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4161 root->number_of_cgroups++;
4163 /* each css holds a ref to the cgroup's dentry */
4164 for_each_subsys(root, ss)
4167 /* creation succeeded, notify subsystems */
4168 for_each_subsys(root, ss) {
4169 err = online_css(ss, cgrp);
4173 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4175 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",
4176 current->comm, current->pid, ss->name);
4177 if (!strcmp(ss->name, "memory"))
4178 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4179 ss->warned_broken_hierarchy = true;
4183 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4187 mutex_unlock(&cgroup_mutex);
4188 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4193 for_each_subsys(root, ss) {
4194 if (cgrp->subsys[ss->subsys_id])
4197 mutex_unlock(&cgroup_mutex);
4198 /* Release the reference count that we took on the superblock */
4199 deactivate_super(sb);
4201 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4203 kfree(rcu_dereference_raw(cgrp->name));
4209 cgroup_destroy_locked(cgrp);
4210 mutex_unlock(&cgroup_mutex);
4211 mutex_unlock(&dentry->d_inode->i_mutex);
4215 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4217 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4219 /* the vfs holds inode->i_mutex already */
4220 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4223 static int cgroup_destroy_locked(struct cgroup *cgrp)
4224 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4226 struct dentry *d = cgrp->dentry;
4227 struct cgroup *parent = cgrp->parent;
4228 struct cgroup_event *event, *tmp;
4229 struct cgroup_subsys *ss;
4231 lockdep_assert_held(&d->d_inode->i_mutex);
4232 lockdep_assert_held(&cgroup_mutex);
4234 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children))
4238 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4239 * removed. This makes future css_tryget() and child creation
4240 * attempts fail thus maintaining the removal conditions verified
4243 for_each_subsys(cgrp->root, ss) {
4244 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4246 WARN_ON(atomic_read(&css->refcnt) < 0);
4247 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4249 set_bit(CGRP_REMOVED, &cgrp->flags);
4251 /* tell subsystems to initate destruction */
4252 for_each_subsys(cgrp->root, ss)
4253 offline_css(ss, cgrp);
4256 * Put all the base refs. Each css holds an extra reference to the
4257 * cgroup's dentry and cgroup removal proceeds regardless of css
4258 * refs. On the last put of each css, whenever that may be, the
4259 * extra dentry ref is put so that dentry destruction happens only
4260 * after all css's are released.
4262 for_each_subsys(cgrp->root, ss)
4263 css_put(cgrp->subsys[ss->subsys_id]);
4265 raw_spin_lock(&release_list_lock);
4266 if (!list_empty(&cgrp->release_list))
4267 list_del_init(&cgrp->release_list);
4268 raw_spin_unlock(&release_list_lock);
4270 /* delete this cgroup from parent->children */
4271 list_del_rcu(&cgrp->sibling);
4272 list_del_init(&cgrp->allcg_node);
4275 cgroup_d_remove_dir(d);
4278 set_bit(CGRP_RELEASABLE, &parent->flags);
4279 check_for_release(parent);
4282 * Unregister events and notify userspace.
4283 * Notify userspace about cgroup removing only after rmdir of cgroup
4284 * directory to avoid race between userspace and kernelspace.
4286 spin_lock(&cgrp->event_list_lock);
4287 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4288 list_del_init(&event->list);
4289 schedule_work(&event->remove);
4291 spin_unlock(&cgrp->event_list_lock);
4296 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4300 mutex_lock(&cgroup_mutex);
4301 ret = cgroup_destroy_locked(dentry->d_fsdata);
4302 mutex_unlock(&cgroup_mutex);
4307 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4309 INIT_LIST_HEAD(&ss->cftsets);
4312 * base_cftset is embedded in subsys itself, no need to worry about
4315 if (ss->base_cftypes) {
4316 ss->base_cftset.cfts = ss->base_cftypes;
4317 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4321 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4323 struct cgroup_subsys_state *css;
4325 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4327 mutex_lock(&cgroup_mutex);
4329 /* init base cftset */
4330 cgroup_init_cftsets(ss);
4332 /* Create the top cgroup state for this subsystem */
4333 list_add(&ss->sibling, &rootnode.subsys_list);
4334 ss->root = &rootnode;
4335 css = ss->css_alloc(dummytop);
4336 /* We don't handle early failures gracefully */
4337 BUG_ON(IS_ERR(css));
4338 init_cgroup_css(css, ss, dummytop);
4340 /* Update the init_css_set to contain a subsys
4341 * pointer to this state - since the subsystem is
4342 * newly registered, all tasks and hence the
4343 * init_css_set is in the subsystem's top cgroup. */
4344 init_css_set.subsys[ss->subsys_id] = css;
4346 need_forkexit_callback |= ss->fork || ss->exit;
4348 /* At system boot, before all subsystems have been
4349 * registered, no tasks have been forked, so we don't
4350 * need to invoke fork callbacks here. */
4351 BUG_ON(!list_empty(&init_task.tasks));
4354 BUG_ON(online_css(ss, dummytop));
4356 mutex_unlock(&cgroup_mutex);
4358 /* this function shouldn't be used with modular subsystems, since they
4359 * need to register a subsys_id, among other things */
4364 * cgroup_load_subsys: load and register a modular subsystem at runtime
4365 * @ss: the subsystem to load
4367 * This function should be called in a modular subsystem's initcall. If the
4368 * subsystem is built as a module, it will be assigned a new subsys_id and set
4369 * up for use. If the subsystem is built-in anyway, work is delegated to the
4370 * simpler cgroup_init_subsys.
4372 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4374 struct cgroup_subsys_state *css;
4376 struct hlist_node *tmp;
4380 /* check name and function validity */
4381 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4382 ss->css_alloc == NULL || ss->css_free == NULL)
4386 * we don't support callbacks in modular subsystems. this check is
4387 * before the ss->module check for consistency; a subsystem that could
4388 * be a module should still have no callbacks even if the user isn't
4389 * compiling it as one.
4391 if (ss->fork || ss->exit)
4395 * an optionally modular subsystem is built-in: we want to do nothing,
4396 * since cgroup_init_subsys will have already taken care of it.
4398 if (ss->module == NULL) {
4399 /* a sanity check */
4400 BUG_ON(subsys[ss->subsys_id] != ss);
4404 /* init base cftset */
4405 cgroup_init_cftsets(ss);
4407 mutex_lock(&cgroup_mutex);
4408 subsys[ss->subsys_id] = ss;
4411 * no ss->css_alloc seems to need anything important in the ss
4412 * struct, so this can happen first (i.e. before the rootnode
4415 css = ss->css_alloc(dummytop);
4417 /* failure case - need to deassign the subsys[] slot. */
4418 subsys[ss->subsys_id] = NULL;
4419 mutex_unlock(&cgroup_mutex);
4420 return PTR_ERR(css);
4423 list_add(&ss->sibling, &rootnode.subsys_list);
4424 ss->root = &rootnode;
4426 /* our new subsystem will be attached to the dummy hierarchy. */
4427 init_cgroup_css(css, ss, dummytop);
4428 /* init_idr must be after init_cgroup_css because it sets css->id. */
4430 ret = cgroup_init_idr(ss, css);
4436 * Now we need to entangle the css into the existing css_sets. unlike
4437 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4438 * will need a new pointer to it; done by iterating the css_set_table.
4439 * furthermore, modifying the existing css_sets will corrupt the hash
4440 * table state, so each changed css_set will need its hash recomputed.
4441 * this is all done under the css_set_lock.
4443 write_lock(&css_set_lock);
4444 hash_for_each_safe(css_set_table, i, tmp, cg, hlist) {
4445 /* skip entries that we already rehashed */
4446 if (cg->subsys[ss->subsys_id])
4448 /* remove existing entry */
4449 hash_del(&cg->hlist);
4451 cg->subsys[ss->subsys_id] = css;
4452 /* recompute hash and restore entry */
4453 key = css_set_hash(cg->subsys);
4454 hash_add(css_set_table, &cg->hlist, key);
4456 write_unlock(&css_set_lock);
4459 ret = online_css(ss, dummytop);
4464 mutex_unlock(&cgroup_mutex);
4468 mutex_unlock(&cgroup_mutex);
4469 /* @ss can't be mounted here as try_module_get() would fail */
4470 cgroup_unload_subsys(ss);
4473 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4476 * cgroup_unload_subsys: unload a modular subsystem
4477 * @ss: the subsystem to unload
4479 * This function should be called in a modular subsystem's exitcall. When this
4480 * function is invoked, the refcount on the subsystem's module will be 0, so
4481 * the subsystem will not be attached to any hierarchy.
4483 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4485 struct cg_cgroup_link *link;
4487 BUG_ON(ss->module == NULL);
4490 * we shouldn't be called if the subsystem is in use, and the use of
4491 * try_module_get in parse_cgroupfs_options should ensure that it
4492 * doesn't start being used while we're killing it off.
4494 BUG_ON(ss->root != &rootnode);
4496 mutex_lock(&cgroup_mutex);
4498 offline_css(ss, dummytop);
4502 idr_destroy(&ss->idr);
4504 /* deassign the subsys_id */
4505 subsys[ss->subsys_id] = NULL;
4507 /* remove subsystem from rootnode's list of subsystems */
4508 list_del_init(&ss->sibling);
4511 * disentangle the css from all css_sets attached to the dummytop. as
4512 * in loading, we need to pay our respects to the hashtable gods.
4514 write_lock(&css_set_lock);
4515 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4516 struct css_set *cg = link->cg;
4519 hash_del(&cg->hlist);
4520 cg->subsys[ss->subsys_id] = NULL;
4521 key = css_set_hash(cg->subsys);
4522 hash_add(css_set_table, &cg->hlist, key);
4524 write_unlock(&css_set_lock);
4527 * remove subsystem's css from the dummytop and free it - need to
4528 * free before marking as null because ss->css_free needs the
4529 * cgrp->subsys pointer to find their state. note that this also
4530 * takes care of freeing the css_id.
4532 ss->css_free(dummytop);
4533 dummytop->subsys[ss->subsys_id] = NULL;
4535 mutex_unlock(&cgroup_mutex);
4537 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4540 * cgroup_init_early - cgroup initialization at system boot
4542 * Initialize cgroups at system boot, and initialize any
4543 * subsystems that request early init.
4545 int __init cgroup_init_early(void)
4548 atomic_set(&init_css_set.refcount, 1);
4549 INIT_LIST_HEAD(&init_css_set.cg_links);
4550 INIT_LIST_HEAD(&init_css_set.tasks);
4551 INIT_HLIST_NODE(&init_css_set.hlist);
4553 init_cgroup_root(&rootnode);
4555 init_task.cgroups = &init_css_set;
4557 init_css_set_link.cg = &init_css_set;
4558 init_css_set_link.cgrp = dummytop;
4559 list_add(&init_css_set_link.cgrp_link_list,
4560 &rootnode.top_cgroup.css_sets);
4561 list_add(&init_css_set_link.cg_link_list,
4562 &init_css_set.cg_links);
4564 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4565 struct cgroup_subsys *ss = subsys[i];
4567 /* at bootup time, we don't worry about modular subsystems */
4568 if (!ss || ss->module)
4572 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4573 BUG_ON(!ss->css_alloc);
4574 BUG_ON(!ss->css_free);
4575 if (ss->subsys_id != i) {
4576 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4577 ss->name, ss->subsys_id);
4582 cgroup_init_subsys(ss);
4588 * cgroup_init - cgroup initialization
4590 * Register cgroup filesystem and /proc file, and initialize
4591 * any subsystems that didn't request early init.
4593 int __init cgroup_init(void)
4599 err = bdi_init(&cgroup_backing_dev_info);
4603 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4604 struct cgroup_subsys *ss = subsys[i];
4606 /* at bootup time, we don't worry about modular subsystems */
4607 if (!ss || ss->module)
4609 if (!ss->early_init)
4610 cgroup_init_subsys(ss);
4612 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4615 /* Add init_css_set to the hash table */
4616 key = css_set_hash(init_css_set.subsys);
4617 hash_add(css_set_table, &init_css_set.hlist, key);
4618 BUG_ON(!init_root_id(&rootnode));
4620 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4626 err = register_filesystem(&cgroup_fs_type);
4628 kobject_put(cgroup_kobj);
4632 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4636 bdi_destroy(&cgroup_backing_dev_info);
4642 * proc_cgroup_show()
4643 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4644 * - Used for /proc/<pid>/cgroup.
4645 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4646 * doesn't really matter if tsk->cgroup changes after we read it,
4647 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4648 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4649 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4650 * cgroup to top_cgroup.
4653 /* TODO: Use a proper seq_file iterator */
4654 static int proc_cgroup_show(struct seq_file *m, void *v)
4657 struct task_struct *tsk;
4660 struct cgroupfs_root *root;
4663 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4669 tsk = get_pid_task(pid, PIDTYPE_PID);
4675 mutex_lock(&cgroup_mutex);
4677 for_each_active_root(root) {
4678 struct cgroup_subsys *ss;
4679 struct cgroup *cgrp;
4682 seq_printf(m, "%d:", root->hierarchy_id);
4683 for_each_subsys(root, ss)
4684 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4685 if (strlen(root->name))
4686 seq_printf(m, "%sname=%s", count ? "," : "",
4689 cgrp = task_cgroup_from_root(tsk, root);
4690 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4698 mutex_unlock(&cgroup_mutex);
4699 put_task_struct(tsk);
4706 static int cgroup_open(struct inode *inode, struct file *file)
4708 struct pid *pid = PROC_I(inode)->pid;
4709 return single_open(file, proc_cgroup_show, pid);
4712 const struct file_operations proc_cgroup_operations = {
4713 .open = cgroup_open,
4715 .llseek = seq_lseek,
4716 .release = single_release,
4719 /* Display information about each subsystem and each hierarchy */
4720 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4724 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4726 * ideally we don't want subsystems moving around while we do this.
4727 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4728 * subsys/hierarchy state.
4730 mutex_lock(&cgroup_mutex);
4731 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4732 struct cgroup_subsys *ss = subsys[i];
4735 seq_printf(m, "%s\t%d\t%d\t%d\n",
4736 ss->name, ss->root->hierarchy_id,
4737 ss->root->number_of_cgroups, !ss->disabled);
4739 mutex_unlock(&cgroup_mutex);
4743 static int cgroupstats_open(struct inode *inode, struct file *file)
4745 return single_open(file, proc_cgroupstats_show, NULL);
4748 static const struct file_operations proc_cgroupstats_operations = {
4749 .open = cgroupstats_open,
4751 .llseek = seq_lseek,
4752 .release = single_release,
4756 * cgroup_fork - attach newly forked task to its parents cgroup.
4757 * @child: pointer to task_struct of forking parent process.
4759 * Description: A task inherits its parent's cgroup at fork().
4761 * A pointer to the shared css_set was automatically copied in
4762 * fork.c by dup_task_struct(). However, we ignore that copy, since
4763 * it was not made under the protection of RCU or cgroup_mutex, so
4764 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4765 * have already changed current->cgroups, allowing the previously
4766 * referenced cgroup group to be removed and freed.
4768 * At the point that cgroup_fork() is called, 'current' is the parent
4769 * task, and the passed argument 'child' points to the child task.
4771 void cgroup_fork(struct task_struct *child)
4774 child->cgroups = current->cgroups;
4775 get_css_set(child->cgroups);
4776 task_unlock(current);
4777 INIT_LIST_HEAD(&child->cg_list);
4781 * cgroup_post_fork - called on a new task after adding it to the task list
4782 * @child: the task in question
4784 * Adds the task to the list running through its css_set if necessary and
4785 * call the subsystem fork() callbacks. Has to be after the task is
4786 * visible on the task list in case we race with the first call to
4787 * cgroup_iter_start() - to guarantee that the new task ends up on its
4790 void cgroup_post_fork(struct task_struct *child)
4795 * use_task_css_set_links is set to 1 before we walk the tasklist
4796 * under the tasklist_lock and we read it here after we added the child
4797 * to the tasklist under the tasklist_lock as well. If the child wasn't
4798 * yet in the tasklist when we walked through it from
4799 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4800 * should be visible now due to the paired locking and barriers implied
4801 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4802 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4805 if (use_task_css_set_links) {
4806 write_lock(&css_set_lock);
4808 if (list_empty(&child->cg_list))
4809 list_add(&child->cg_list, &child->cgroups->tasks);
4811 write_unlock(&css_set_lock);
4815 * Call ss->fork(). This must happen after @child is linked on
4816 * css_set; otherwise, @child might change state between ->fork()
4817 * and addition to css_set.
4819 if (need_forkexit_callback) {
4821 * fork/exit callbacks are supported only for builtin
4822 * subsystems, and the builtin section of the subsys
4823 * array is immutable, so we don't need to lock the
4824 * subsys array here. On the other hand, modular section
4825 * of the array can be freed at module unload, so we
4828 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4829 struct cgroup_subsys *ss = subsys[i];
4838 * cgroup_exit - detach cgroup from exiting task
4839 * @tsk: pointer to task_struct of exiting process
4840 * @run_callback: run exit callbacks?
4842 * Description: Detach cgroup from @tsk and release it.
4844 * Note that cgroups marked notify_on_release force every task in
4845 * them to take the global cgroup_mutex mutex when exiting.
4846 * This could impact scaling on very large systems. Be reluctant to
4847 * use notify_on_release cgroups where very high task exit scaling
4848 * is required on large systems.
4850 * the_top_cgroup_hack:
4852 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4854 * We call cgroup_exit() while the task is still competent to
4855 * handle notify_on_release(), then leave the task attached to the
4856 * root cgroup in each hierarchy for the remainder of its exit.
4858 * To do this properly, we would increment the reference count on
4859 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4860 * code we would add a second cgroup function call, to drop that
4861 * reference. This would just create an unnecessary hot spot on
4862 * the top_cgroup reference count, to no avail.
4864 * Normally, holding a reference to a cgroup without bumping its
4865 * count is unsafe. The cgroup could go away, or someone could
4866 * attach us to a different cgroup, decrementing the count on
4867 * the first cgroup that we never incremented. But in this case,
4868 * top_cgroup isn't going away, and either task has PF_EXITING set,
4869 * which wards off any cgroup_attach_task() attempts, or task is a failed
4870 * fork, never visible to cgroup_attach_task.
4872 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4878 * Unlink from the css_set task list if necessary.
4879 * Optimistically check cg_list before taking
4882 if (!list_empty(&tsk->cg_list)) {
4883 write_lock(&css_set_lock);
4884 if (!list_empty(&tsk->cg_list))
4885 list_del_init(&tsk->cg_list);
4886 write_unlock(&css_set_lock);
4889 /* Reassign the task to the init_css_set. */
4892 tsk->cgroups = &init_css_set;
4894 if (run_callbacks && need_forkexit_callback) {
4896 * fork/exit callbacks are supported only for builtin
4897 * subsystems, see cgroup_post_fork() for details.
4899 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4900 struct cgroup_subsys *ss = subsys[i];
4903 struct cgroup *old_cgrp =
4904 rcu_dereference_raw(cg->subsys[i])->cgroup;
4905 struct cgroup *cgrp = task_cgroup(tsk, i);
4906 ss->exit(cgrp, old_cgrp, tsk);
4912 put_css_set_taskexit(cg);
4915 static void check_for_release(struct cgroup *cgrp)
4917 /* All of these checks rely on RCU to keep the cgroup
4918 * structure alive */
4919 if (cgroup_is_releasable(cgrp) &&
4920 !atomic_read(&cgrp->count) && list_empty(&cgrp->children)) {
4922 * Control Group is currently removeable. If it's not
4923 * already queued for a userspace notification, queue
4926 int need_schedule_work = 0;
4928 raw_spin_lock(&release_list_lock);
4929 if (!cgroup_is_removed(cgrp) &&
4930 list_empty(&cgrp->release_list)) {
4931 list_add(&cgrp->release_list, &release_list);
4932 need_schedule_work = 1;
4934 raw_spin_unlock(&release_list_lock);
4935 if (need_schedule_work)
4936 schedule_work(&release_agent_work);
4940 /* Caller must verify that the css is not for root cgroup */
4941 bool __css_tryget(struct cgroup_subsys_state *css)
4946 v = css_refcnt(css);
4947 t = atomic_cmpxchg(&css->refcnt, v, v + 1);
4955 EXPORT_SYMBOL_GPL(__css_tryget);
4957 /* Caller must verify that the css is not for root cgroup */
4958 void __css_put(struct cgroup_subsys_state *css)
4962 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
4964 schedule_work(&css->dput_work);
4966 EXPORT_SYMBOL_GPL(__css_put);
4969 * Notify userspace when a cgroup is released, by running the
4970 * configured release agent with the name of the cgroup (path
4971 * relative to the root of cgroup file system) as the argument.
4973 * Most likely, this user command will try to rmdir this cgroup.
4975 * This races with the possibility that some other task will be
4976 * attached to this cgroup before it is removed, or that some other
4977 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4978 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4979 * unused, and this cgroup will be reprieved from its death sentence,
4980 * to continue to serve a useful existence. Next time it's released,
4981 * we will get notified again, if it still has 'notify_on_release' set.
4983 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4984 * means only wait until the task is successfully execve()'d. The
4985 * separate release agent task is forked by call_usermodehelper(),
4986 * then control in this thread returns here, without waiting for the
4987 * release agent task. We don't bother to wait because the caller of
4988 * this routine has no use for the exit status of the release agent
4989 * task, so no sense holding our caller up for that.
4991 static void cgroup_release_agent(struct work_struct *work)
4993 BUG_ON(work != &release_agent_work);
4994 mutex_lock(&cgroup_mutex);
4995 raw_spin_lock(&release_list_lock);
4996 while (!list_empty(&release_list)) {
4997 char *argv[3], *envp[3];
4999 char *pathbuf = NULL, *agentbuf = NULL;
5000 struct cgroup *cgrp = list_entry(release_list.next,
5003 list_del_init(&cgrp->release_list);
5004 raw_spin_unlock(&release_list_lock);
5005 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5008 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5010 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5015 argv[i++] = agentbuf;
5016 argv[i++] = pathbuf;
5020 /* minimal command environment */
5021 envp[i++] = "HOME=/";
5022 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5025 /* Drop the lock while we invoke the usermode helper,
5026 * since the exec could involve hitting disk and hence
5027 * be a slow process */
5028 mutex_unlock(&cgroup_mutex);
5029 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5030 mutex_lock(&cgroup_mutex);
5034 raw_spin_lock(&release_list_lock);
5036 raw_spin_unlock(&release_list_lock);
5037 mutex_unlock(&cgroup_mutex);
5040 static int __init cgroup_disable(char *str)
5045 while ((token = strsep(&str, ",")) != NULL) {
5048 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5049 struct cgroup_subsys *ss = subsys[i];
5052 * cgroup_disable, being at boot time, can't
5053 * know about module subsystems, so we don't
5056 if (!ss || ss->module)
5059 if (!strcmp(token, ss->name)) {
5061 printk(KERN_INFO "Disabling %s control group"
5062 " subsystem\n", ss->name);
5069 __setup("cgroup_disable=", cgroup_disable);
5072 * Functons for CSS ID.
5076 *To get ID other than 0, this should be called when !cgroup_is_removed().
5078 unsigned short css_id(struct cgroup_subsys_state *css)
5080 struct css_id *cssid;
5083 * This css_id() can return correct value when somone has refcnt
5084 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5085 * it's unchanged until freed.
5087 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5093 EXPORT_SYMBOL_GPL(css_id);
5095 unsigned short css_depth(struct cgroup_subsys_state *css)
5097 struct css_id *cssid;
5099 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5102 return cssid->depth;
5105 EXPORT_SYMBOL_GPL(css_depth);
5108 * css_is_ancestor - test "root" css is an ancestor of "child"
5109 * @child: the css to be tested.
5110 * @root: the css supporsed to be an ancestor of the child.
5112 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5113 * this function reads css->id, the caller must hold rcu_read_lock().
5114 * But, considering usual usage, the csses should be valid objects after test.
5115 * Assuming that the caller will do some action to the child if this returns
5116 * returns true, the caller must take "child";s reference count.
5117 * If "child" is valid object and this returns true, "root" is valid, too.
5120 bool css_is_ancestor(struct cgroup_subsys_state *child,
5121 const struct cgroup_subsys_state *root)
5123 struct css_id *child_id;
5124 struct css_id *root_id;
5126 child_id = rcu_dereference(child->id);
5129 root_id = rcu_dereference(root->id);
5132 if (child_id->depth < root_id->depth)
5134 if (child_id->stack[root_id->depth] != root_id->id)
5139 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5141 struct css_id *id = css->id;
5142 /* When this is called before css_id initialization, id can be NULL */
5146 BUG_ON(!ss->use_id);
5148 rcu_assign_pointer(id->css, NULL);
5149 rcu_assign_pointer(css->id, NULL);
5150 spin_lock(&ss->id_lock);
5151 idr_remove(&ss->idr, id->id);
5152 spin_unlock(&ss->id_lock);
5153 kfree_rcu(id, rcu_head);
5155 EXPORT_SYMBOL_GPL(free_css_id);
5158 * This is called by init or create(). Then, calls to this function are
5159 * always serialized (By cgroup_mutex() at create()).
5162 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5164 struct css_id *newid;
5167 BUG_ON(!ss->use_id);
5169 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5170 newid = kzalloc(size, GFP_KERNEL);
5172 return ERR_PTR(-ENOMEM);
5174 idr_preload(GFP_KERNEL);
5175 spin_lock(&ss->id_lock);
5176 /* Don't use 0. allocates an ID of 1-65535 */
5177 ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5178 spin_unlock(&ss->id_lock);
5181 /* Returns error when there are no free spaces for new ID.*/
5186 newid->depth = depth;
5190 return ERR_PTR(ret);
5194 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5195 struct cgroup_subsys_state *rootcss)
5197 struct css_id *newid;
5199 spin_lock_init(&ss->id_lock);
5202 newid = get_new_cssid(ss, 0);
5204 return PTR_ERR(newid);
5206 newid->stack[0] = newid->id;
5207 newid->css = rootcss;
5208 rootcss->id = newid;
5212 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5213 struct cgroup *child)
5215 int subsys_id, i, depth = 0;
5216 struct cgroup_subsys_state *parent_css, *child_css;
5217 struct css_id *child_id, *parent_id;
5219 subsys_id = ss->subsys_id;
5220 parent_css = parent->subsys[subsys_id];
5221 child_css = child->subsys[subsys_id];
5222 parent_id = parent_css->id;
5223 depth = parent_id->depth + 1;
5225 child_id = get_new_cssid(ss, depth);
5226 if (IS_ERR(child_id))
5227 return PTR_ERR(child_id);
5229 for (i = 0; i < depth; i++)
5230 child_id->stack[i] = parent_id->stack[i];
5231 child_id->stack[depth] = child_id->id;
5233 * child_id->css pointer will be set after this cgroup is available
5234 * see cgroup_populate_dir()
5236 rcu_assign_pointer(child_css->id, child_id);
5242 * css_lookup - lookup css by id
5243 * @ss: cgroup subsys to be looked into.
5246 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5247 * NULL if not. Should be called under rcu_read_lock()
5249 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5251 struct css_id *cssid = NULL;
5253 BUG_ON(!ss->use_id);
5254 cssid = idr_find(&ss->idr, id);
5256 if (unlikely(!cssid))
5259 return rcu_dereference(cssid->css);
5261 EXPORT_SYMBOL_GPL(css_lookup);
5264 * css_get_next - lookup next cgroup under specified hierarchy.
5265 * @ss: pointer to subsystem
5266 * @id: current position of iteration.
5267 * @root: pointer to css. search tree under this.
5268 * @foundid: position of found object.
5270 * Search next css under the specified hierarchy of rootid. Calling under
5271 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5273 struct cgroup_subsys_state *
5274 css_get_next(struct cgroup_subsys *ss, int id,
5275 struct cgroup_subsys_state *root, int *foundid)
5277 struct cgroup_subsys_state *ret = NULL;
5280 int rootid = css_id(root);
5281 int depth = css_depth(root);
5286 BUG_ON(!ss->use_id);
5287 WARN_ON_ONCE(!rcu_read_lock_held());
5289 /* fill start point for scan */
5293 * scan next entry from bitmap(tree), tmpid is updated after
5296 tmp = idr_get_next(&ss->idr, &tmpid);
5299 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5300 ret = rcu_dereference(tmp->css);
5306 /* continue to scan from next id */
5313 * get corresponding css from file open on cgroupfs directory
5315 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5317 struct cgroup *cgrp;
5318 struct inode *inode;
5319 struct cgroup_subsys_state *css;
5321 inode = file_inode(f);
5322 /* check in cgroup filesystem dir */
5323 if (inode->i_op != &cgroup_dir_inode_operations)
5324 return ERR_PTR(-EBADF);
5326 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5327 return ERR_PTR(-EINVAL);
5330 cgrp = __d_cgrp(f->f_dentry);
5331 css = cgrp->subsys[id];
5332 return css ? css : ERR_PTR(-ENOENT);
5335 #ifdef CONFIG_CGROUP_DEBUG
5336 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5338 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5341 return ERR_PTR(-ENOMEM);
5346 static void debug_css_free(struct cgroup *cont)
5348 kfree(cont->subsys[debug_subsys_id]);
5351 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5353 return atomic_read(&cont->count);
5356 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5358 return cgroup_task_count(cont);
5361 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5363 return (u64)(unsigned long)current->cgroups;
5366 static u64 current_css_set_refcount_read(struct cgroup *cont,
5372 count = atomic_read(¤t->cgroups->refcount);
5377 static int current_css_set_cg_links_read(struct cgroup *cont,
5379 struct seq_file *seq)
5381 struct cg_cgroup_link *link;
5384 read_lock(&css_set_lock);
5386 cg = rcu_dereference(current->cgroups);
5387 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5388 struct cgroup *c = link->cgrp;
5392 name = c->dentry->d_name.name;
5395 seq_printf(seq, "Root %d group %s\n",
5396 c->root->hierarchy_id, name);
5399 read_unlock(&css_set_lock);
5403 #define MAX_TASKS_SHOWN_PER_CSS 25
5404 static int cgroup_css_links_read(struct cgroup *cont,
5406 struct seq_file *seq)
5408 struct cg_cgroup_link *link;
5410 read_lock(&css_set_lock);
5411 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5412 struct css_set *cg = link->cg;
5413 struct task_struct *task;
5415 seq_printf(seq, "css_set %p\n", cg);
5416 list_for_each_entry(task, &cg->tasks, cg_list) {
5417 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5418 seq_puts(seq, " ...\n");
5421 seq_printf(seq, " task %d\n",
5422 task_pid_vnr(task));
5426 read_unlock(&css_set_lock);
5430 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5432 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5435 static struct cftype debug_files[] = {
5437 .name = "cgroup_refcount",
5438 .read_u64 = cgroup_refcount_read,
5441 .name = "taskcount",
5442 .read_u64 = debug_taskcount_read,
5446 .name = "current_css_set",
5447 .read_u64 = current_css_set_read,
5451 .name = "current_css_set_refcount",
5452 .read_u64 = current_css_set_refcount_read,
5456 .name = "current_css_set_cg_links",
5457 .read_seq_string = current_css_set_cg_links_read,
5461 .name = "cgroup_css_links",
5462 .read_seq_string = cgroup_css_links_read,
5466 .name = "releasable",
5467 .read_u64 = releasable_read,
5473 struct cgroup_subsys debug_subsys = {
5475 .css_alloc = debug_css_alloc,
5476 .css_free = debug_css_free,
5477 .subsys_id = debug_subsys_id,
5478 .base_cftypes = debug_files,
5480 #endif /* CONFIG_CGROUP_DEBUG */