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 * We get a ref to the parent's dentry, and put the ref when
880 * this cgroup is being freed, so it's guaranteed that the
881 * parent won't be destroyed before its children.
883 dput(cgrp->parent->dentry);
886 * Drop the active superblock reference that we took when we
889 deactivate_super(cgrp->root->sb);
892 * if we're getting rid of the cgroup, refcount should ensure
893 * that there are no pidlists left.
895 BUG_ON(!list_empty(&cgrp->pidlists));
897 simple_xattrs_free(&cgrp->xattrs);
899 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
900 kfree(rcu_dereference_raw(cgrp->name));
904 static void cgroup_free_rcu(struct rcu_head *head)
906 struct cgroup *cgrp = container_of(head, struct cgroup, rcu_head);
908 schedule_work(&cgrp->free_work);
911 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
913 /* is dentry a directory ? if so, kfree() associated cgroup */
914 if (S_ISDIR(inode->i_mode)) {
915 struct cgroup *cgrp = dentry->d_fsdata;
917 BUG_ON(!(cgroup_is_removed(cgrp)));
918 call_rcu(&cgrp->rcu_head, cgroup_free_rcu);
920 struct cfent *cfe = __d_cfe(dentry);
921 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
922 struct cftype *cft = cfe->type;
924 WARN_ONCE(!list_empty(&cfe->node) &&
925 cgrp != &cgrp->root->top_cgroup,
926 "cfe still linked for %s\n", cfe->type->name);
928 simple_xattrs_free(&cft->xattrs);
933 static int cgroup_delete(const struct dentry *d)
938 static void remove_dir(struct dentry *d)
940 struct dentry *parent = dget(d->d_parent);
943 simple_rmdir(parent->d_inode, d);
947 static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
951 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
952 lockdep_assert_held(&cgroup_mutex);
955 * If we're doing cleanup due to failure of cgroup_create(),
956 * the corresponding @cfe may not exist.
958 list_for_each_entry(cfe, &cgrp->files, node) {
959 struct dentry *d = cfe->dentry;
961 if (cft && cfe->type != cft)
966 simple_unlink(cgrp->dentry->d_inode, d);
967 list_del_init(&cfe->node);
975 * cgroup_clear_directory - selective removal of base and subsystem files
976 * @dir: directory containing the files
977 * @base_files: true if the base files should be removed
978 * @subsys_mask: mask of the subsystem ids whose files should be removed
980 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
981 unsigned long subsys_mask)
983 struct cgroup *cgrp = __d_cgrp(dir);
984 struct cgroup_subsys *ss;
986 for_each_subsys(cgrp->root, ss) {
987 struct cftype_set *set;
988 if (!test_bit(ss->subsys_id, &subsys_mask))
990 list_for_each_entry(set, &ss->cftsets, node)
991 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
994 while (!list_empty(&cgrp->files))
995 cgroup_rm_file(cgrp, NULL);
1000 * NOTE : the dentry must have been dget()'ed
1002 static void cgroup_d_remove_dir(struct dentry *dentry)
1004 struct dentry *parent;
1005 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1007 cgroup_clear_directory(dentry, true, root->subsys_mask);
1009 parent = dentry->d_parent;
1010 spin_lock(&parent->d_lock);
1011 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1012 list_del_init(&dentry->d_u.d_child);
1013 spin_unlock(&dentry->d_lock);
1014 spin_unlock(&parent->d_lock);
1019 * Call with cgroup_mutex held. Drops reference counts on modules, including
1020 * any duplicate ones that parse_cgroupfs_options took. If this function
1021 * returns an error, no reference counts are touched.
1023 static int rebind_subsystems(struct cgroupfs_root *root,
1024 unsigned long final_subsys_mask)
1026 unsigned long added_mask, removed_mask;
1027 struct cgroup *cgrp = &root->top_cgroup;
1030 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1031 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1033 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
1034 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1035 /* Check that any added subsystems are currently free */
1036 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1037 unsigned long bit = 1UL << i;
1038 struct cgroup_subsys *ss = subsys[i];
1039 if (!(bit & added_mask))
1042 * Nobody should tell us to do a subsys that doesn't exist:
1043 * parse_cgroupfs_options should catch that case and refcounts
1044 * ensure that subsystems won't disappear once selected.
1047 if (ss->root != &rootnode) {
1048 /* Subsystem isn't free */
1053 /* Currently we don't handle adding/removing subsystems when
1054 * any child cgroups exist. This is theoretically supportable
1055 * but involves complex error handling, so it's being left until
1057 if (root->number_of_cgroups > 1)
1060 /* Process each subsystem */
1061 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1062 struct cgroup_subsys *ss = subsys[i];
1063 unsigned long bit = 1UL << i;
1064 if (bit & added_mask) {
1065 /* We're binding this subsystem to this hierarchy */
1067 BUG_ON(cgrp->subsys[i]);
1068 BUG_ON(!dummytop->subsys[i]);
1069 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1070 cgrp->subsys[i] = dummytop->subsys[i];
1071 cgrp->subsys[i]->cgroup = cgrp;
1072 list_move(&ss->sibling, &root->subsys_list);
1074 /* refcount was already taken, and we're keeping it */
1075 } else if (bit & removed_mask) {
1076 /* We're removing this subsystem */
1078 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1079 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1080 dummytop->subsys[i]->cgroup = dummytop;
1081 cgrp->subsys[i] = NULL;
1082 subsys[i]->root = &rootnode;
1083 list_move(&ss->sibling, &rootnode.subsys_list);
1084 /* subsystem is now free - drop reference on module */
1085 module_put(ss->module);
1086 } else if (bit & final_subsys_mask) {
1087 /* Subsystem state should already exist */
1089 BUG_ON(!cgrp->subsys[i]);
1091 * a refcount was taken, but we already had one, so
1092 * drop the extra reference.
1094 module_put(ss->module);
1095 #ifdef CONFIG_MODULE_UNLOAD
1096 BUG_ON(ss->module && !module_refcount(ss->module));
1099 /* Subsystem state shouldn't exist */
1100 BUG_ON(cgrp->subsys[i]);
1103 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1108 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1110 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1111 struct cgroup_subsys *ss;
1113 mutex_lock(&cgroup_root_mutex);
1114 for_each_subsys(root, ss)
1115 seq_printf(seq, ",%s", ss->name);
1116 if (test_bit(ROOT_NOPREFIX, &root->flags))
1117 seq_puts(seq, ",noprefix");
1118 if (test_bit(ROOT_XATTR, &root->flags))
1119 seq_puts(seq, ",xattr");
1120 if (strlen(root->release_agent_path))
1121 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1122 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1123 seq_puts(seq, ",clone_children");
1124 if (strlen(root->name))
1125 seq_printf(seq, ",name=%s", root->name);
1126 mutex_unlock(&cgroup_root_mutex);
1130 struct cgroup_sb_opts {
1131 unsigned long subsys_mask;
1132 unsigned long flags;
1133 char *release_agent;
1134 bool cpuset_clone_children;
1136 /* User explicitly requested empty subsystem */
1139 struct cgroupfs_root *new_root;
1144 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1145 * with cgroup_mutex held to protect the subsys[] array. This function takes
1146 * refcounts on subsystems to be used, unless it returns error, in which case
1147 * no refcounts are taken.
1149 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1151 char *token, *o = data;
1152 bool all_ss = false, one_ss = false;
1153 unsigned long mask = (unsigned long)-1;
1155 bool module_pin_failed = false;
1157 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1159 #ifdef CONFIG_CPUSETS
1160 mask = ~(1UL << cpuset_subsys_id);
1163 memset(opts, 0, sizeof(*opts));
1165 while ((token = strsep(&o, ",")) != NULL) {
1168 if (!strcmp(token, "none")) {
1169 /* Explicitly have no subsystems */
1173 if (!strcmp(token, "all")) {
1174 /* Mutually exclusive option 'all' + subsystem name */
1180 if (!strcmp(token, "noprefix")) {
1181 set_bit(ROOT_NOPREFIX, &opts->flags);
1184 if (!strcmp(token, "clone_children")) {
1185 opts->cpuset_clone_children = true;
1188 if (!strcmp(token, "xattr")) {
1189 set_bit(ROOT_XATTR, &opts->flags);
1192 if (!strncmp(token, "release_agent=", 14)) {
1193 /* Specifying two release agents is forbidden */
1194 if (opts->release_agent)
1196 opts->release_agent =
1197 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1198 if (!opts->release_agent)
1202 if (!strncmp(token, "name=", 5)) {
1203 const char *name = token + 5;
1204 /* Can't specify an empty name */
1207 /* Must match [\w.-]+ */
1208 for (i = 0; i < strlen(name); i++) {
1212 if ((c == '.') || (c == '-') || (c == '_'))
1216 /* Specifying two names is forbidden */
1219 opts->name = kstrndup(name,
1220 MAX_CGROUP_ROOT_NAMELEN - 1,
1228 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1229 struct cgroup_subsys *ss = subsys[i];
1232 if (strcmp(token, ss->name))
1237 /* Mutually exclusive option 'all' + subsystem name */
1240 set_bit(i, &opts->subsys_mask);
1245 if (i == CGROUP_SUBSYS_COUNT)
1250 * If the 'all' option was specified select all the subsystems,
1251 * otherwise if 'none', 'name=' and a subsystem name options
1252 * were not specified, let's default to 'all'
1254 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1255 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1256 struct cgroup_subsys *ss = subsys[i];
1261 set_bit(i, &opts->subsys_mask);
1265 /* Consistency checks */
1268 * Option noprefix was introduced just for backward compatibility
1269 * with the old cpuset, so we allow noprefix only if mounting just
1270 * the cpuset subsystem.
1272 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1273 (opts->subsys_mask & mask))
1277 /* Can't specify "none" and some subsystems */
1278 if (opts->subsys_mask && opts->none)
1282 * We either have to specify by name or by subsystems. (So all
1283 * empty hierarchies must have a name).
1285 if (!opts->subsys_mask && !opts->name)
1289 * Grab references on all the modules we'll need, so the subsystems
1290 * don't dance around before rebind_subsystems attaches them. This may
1291 * take duplicate reference counts on a subsystem that's already used,
1292 * but rebind_subsystems handles this case.
1294 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1295 unsigned long bit = 1UL << i;
1297 if (!(bit & opts->subsys_mask))
1299 if (!try_module_get(subsys[i]->module)) {
1300 module_pin_failed = true;
1304 if (module_pin_failed) {
1306 * oops, one of the modules was going away. this means that we
1307 * raced with a module_delete call, and to the user this is
1308 * essentially a "subsystem doesn't exist" case.
1310 for (i--; i >= 0; i--) {
1311 /* drop refcounts only on the ones we took */
1312 unsigned long bit = 1UL << i;
1314 if (!(bit & opts->subsys_mask))
1316 module_put(subsys[i]->module);
1324 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1327 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1328 unsigned long bit = 1UL << i;
1330 if (!(bit & subsys_mask))
1332 module_put(subsys[i]->module);
1336 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1339 struct cgroupfs_root *root = sb->s_fs_info;
1340 struct cgroup *cgrp = &root->top_cgroup;
1341 struct cgroup_sb_opts opts;
1342 unsigned long added_mask, removed_mask;
1344 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1345 mutex_lock(&cgroup_mutex);
1346 mutex_lock(&cgroup_root_mutex);
1348 /* See what subsystems are wanted */
1349 ret = parse_cgroupfs_options(data, &opts);
1353 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1354 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1355 task_tgid_nr(current), current->comm);
1357 added_mask = opts.subsys_mask & ~root->subsys_mask;
1358 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1360 /* Don't allow flags or name to change at remount */
1361 if (opts.flags != root->flags ||
1362 (opts.name && strcmp(opts.name, root->name))) {
1364 drop_parsed_module_refcounts(opts.subsys_mask);
1369 * Clear out the files of subsystems that should be removed, do
1370 * this before rebind_subsystems, since rebind_subsystems may
1371 * change this hierarchy's subsys_list.
1373 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1375 ret = rebind_subsystems(root, opts.subsys_mask);
1377 /* rebind_subsystems failed, re-populate the removed files */
1378 cgroup_populate_dir(cgrp, false, removed_mask);
1379 drop_parsed_module_refcounts(opts.subsys_mask);
1383 /* re-populate subsystem files */
1384 cgroup_populate_dir(cgrp, false, added_mask);
1386 if (opts.release_agent)
1387 strcpy(root->release_agent_path, opts.release_agent);
1389 kfree(opts.release_agent);
1391 mutex_unlock(&cgroup_root_mutex);
1392 mutex_unlock(&cgroup_mutex);
1393 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1397 static const struct super_operations cgroup_ops = {
1398 .statfs = simple_statfs,
1399 .drop_inode = generic_delete_inode,
1400 .show_options = cgroup_show_options,
1401 .remount_fs = cgroup_remount,
1404 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1406 INIT_LIST_HEAD(&cgrp->sibling);
1407 INIT_LIST_HEAD(&cgrp->children);
1408 INIT_LIST_HEAD(&cgrp->files);
1409 INIT_LIST_HEAD(&cgrp->css_sets);
1410 INIT_LIST_HEAD(&cgrp->allcg_node);
1411 INIT_LIST_HEAD(&cgrp->release_list);
1412 INIT_LIST_HEAD(&cgrp->pidlists);
1413 INIT_WORK(&cgrp->free_work, cgroup_free_fn);
1414 mutex_init(&cgrp->pidlist_mutex);
1415 INIT_LIST_HEAD(&cgrp->event_list);
1416 spin_lock_init(&cgrp->event_list_lock);
1417 simple_xattrs_init(&cgrp->xattrs);
1420 static void init_cgroup_root(struct cgroupfs_root *root)
1422 struct cgroup *cgrp = &root->top_cgroup;
1424 INIT_LIST_HEAD(&root->subsys_list);
1425 INIT_LIST_HEAD(&root->root_list);
1426 INIT_LIST_HEAD(&root->allcg_list);
1427 root->number_of_cgroups = 1;
1429 cgrp->name = &root_cgroup_name;
1430 cgrp->top_cgroup = cgrp;
1431 init_cgroup_housekeeping(cgrp);
1432 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1435 static bool init_root_id(struct cgroupfs_root *root)
1440 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1442 spin_lock(&hierarchy_id_lock);
1443 /* Try to allocate the next unused ID */
1444 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1445 &root->hierarchy_id);
1447 /* Try again starting from 0 */
1448 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1450 next_hierarchy_id = root->hierarchy_id + 1;
1451 } else if (ret != -EAGAIN) {
1452 /* Can only get here if the 31-bit IDR is full ... */
1455 spin_unlock(&hierarchy_id_lock);
1460 static int cgroup_test_super(struct super_block *sb, void *data)
1462 struct cgroup_sb_opts *opts = data;
1463 struct cgroupfs_root *root = sb->s_fs_info;
1465 /* If we asked for a name then it must match */
1466 if (opts->name && strcmp(opts->name, root->name))
1470 * If we asked for subsystems (or explicitly for no
1471 * subsystems) then they must match
1473 if ((opts->subsys_mask || opts->none)
1474 && (opts->subsys_mask != root->subsys_mask))
1480 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1482 struct cgroupfs_root *root;
1484 if (!opts->subsys_mask && !opts->none)
1487 root = kzalloc(sizeof(*root), GFP_KERNEL);
1489 return ERR_PTR(-ENOMEM);
1491 if (!init_root_id(root)) {
1493 return ERR_PTR(-ENOMEM);
1495 init_cgroup_root(root);
1497 root->subsys_mask = opts->subsys_mask;
1498 root->flags = opts->flags;
1499 ida_init(&root->cgroup_ida);
1500 if (opts->release_agent)
1501 strcpy(root->release_agent_path, opts->release_agent);
1503 strcpy(root->name, opts->name);
1504 if (opts->cpuset_clone_children)
1505 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1509 static void cgroup_drop_root(struct cgroupfs_root *root)
1514 BUG_ON(!root->hierarchy_id);
1515 spin_lock(&hierarchy_id_lock);
1516 ida_remove(&hierarchy_ida, root->hierarchy_id);
1517 spin_unlock(&hierarchy_id_lock);
1518 ida_destroy(&root->cgroup_ida);
1522 static int cgroup_set_super(struct super_block *sb, void *data)
1525 struct cgroup_sb_opts *opts = data;
1527 /* If we don't have a new root, we can't set up a new sb */
1528 if (!opts->new_root)
1531 BUG_ON(!opts->subsys_mask && !opts->none);
1533 ret = set_anon_super(sb, NULL);
1537 sb->s_fs_info = opts->new_root;
1538 opts->new_root->sb = sb;
1540 sb->s_blocksize = PAGE_CACHE_SIZE;
1541 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1542 sb->s_magic = CGROUP_SUPER_MAGIC;
1543 sb->s_op = &cgroup_ops;
1548 static int cgroup_get_rootdir(struct super_block *sb)
1550 static const struct dentry_operations cgroup_dops = {
1551 .d_iput = cgroup_diput,
1552 .d_delete = cgroup_delete,
1555 struct inode *inode =
1556 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1561 inode->i_fop = &simple_dir_operations;
1562 inode->i_op = &cgroup_dir_inode_operations;
1563 /* directories start off with i_nlink == 2 (for "." entry) */
1565 sb->s_root = d_make_root(inode);
1568 /* for everything else we want ->d_op set */
1569 sb->s_d_op = &cgroup_dops;
1573 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1574 int flags, const char *unused_dev_name,
1577 struct cgroup_sb_opts opts;
1578 struct cgroupfs_root *root;
1580 struct super_block *sb;
1581 struct cgroupfs_root *new_root;
1582 struct inode *inode;
1584 /* First find the desired set of subsystems */
1585 mutex_lock(&cgroup_mutex);
1586 ret = parse_cgroupfs_options(data, &opts);
1587 mutex_unlock(&cgroup_mutex);
1592 * Allocate a new cgroup root. We may not need it if we're
1593 * reusing an existing hierarchy.
1595 new_root = cgroup_root_from_opts(&opts);
1596 if (IS_ERR(new_root)) {
1597 ret = PTR_ERR(new_root);
1600 opts.new_root = new_root;
1602 /* Locate an existing or new sb for this hierarchy */
1603 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1606 cgroup_drop_root(opts.new_root);
1610 root = sb->s_fs_info;
1612 if (root == opts.new_root) {
1613 /* We used the new root structure, so this is a new hierarchy */
1614 struct list_head tmp_cg_links;
1615 struct cgroup *root_cgrp = &root->top_cgroup;
1616 struct cgroupfs_root *existing_root;
1617 const struct cred *cred;
1621 BUG_ON(sb->s_root != NULL);
1623 ret = cgroup_get_rootdir(sb);
1625 goto drop_new_super;
1626 inode = sb->s_root->d_inode;
1628 mutex_lock(&inode->i_mutex);
1629 mutex_lock(&cgroup_mutex);
1630 mutex_lock(&cgroup_root_mutex);
1632 /* Check for name clashes with existing mounts */
1634 if (strlen(root->name))
1635 for_each_active_root(existing_root)
1636 if (!strcmp(existing_root->name, root->name))
1640 * We're accessing css_set_count without locking
1641 * css_set_lock here, but that's OK - it can only be
1642 * increased by someone holding cgroup_lock, and
1643 * that's us. The worst that can happen is that we
1644 * have some link structures left over
1646 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1650 ret = rebind_subsystems(root, root->subsys_mask);
1651 if (ret == -EBUSY) {
1652 free_cg_links(&tmp_cg_links);
1656 * There must be no failure case after here, since rebinding
1657 * takes care of subsystems' refcounts, which are explicitly
1658 * dropped in the failure exit path.
1661 /* EBUSY should be the only error here */
1664 list_add(&root->root_list, &roots);
1667 sb->s_root->d_fsdata = root_cgrp;
1668 root->top_cgroup.dentry = sb->s_root;
1670 /* Link the top cgroup in this hierarchy into all
1671 * the css_set objects */
1672 write_lock(&css_set_lock);
1673 hash_for_each(css_set_table, i, cg, hlist)
1674 link_css_set(&tmp_cg_links, cg, root_cgrp);
1675 write_unlock(&css_set_lock);
1677 free_cg_links(&tmp_cg_links);
1679 BUG_ON(!list_empty(&root_cgrp->children));
1680 BUG_ON(root->number_of_cgroups != 1);
1682 cred = override_creds(&init_cred);
1683 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1685 mutex_unlock(&cgroup_root_mutex);
1686 mutex_unlock(&cgroup_mutex);
1687 mutex_unlock(&inode->i_mutex);
1690 * We re-used an existing hierarchy - the new root (if
1691 * any) is not needed
1693 cgroup_drop_root(opts.new_root);
1694 /* no subsys rebinding, so refcounts don't change */
1695 drop_parsed_module_refcounts(opts.subsys_mask);
1698 kfree(opts.release_agent);
1700 return dget(sb->s_root);
1703 mutex_unlock(&cgroup_root_mutex);
1704 mutex_unlock(&cgroup_mutex);
1705 mutex_unlock(&inode->i_mutex);
1707 deactivate_locked_super(sb);
1709 drop_parsed_module_refcounts(opts.subsys_mask);
1711 kfree(opts.release_agent);
1713 return ERR_PTR(ret);
1716 static void cgroup_kill_sb(struct super_block *sb) {
1717 struct cgroupfs_root *root = sb->s_fs_info;
1718 struct cgroup *cgrp = &root->top_cgroup;
1720 struct cg_cgroup_link *link;
1721 struct cg_cgroup_link *saved_link;
1725 BUG_ON(root->number_of_cgroups != 1);
1726 BUG_ON(!list_empty(&cgrp->children));
1728 mutex_lock(&cgroup_mutex);
1729 mutex_lock(&cgroup_root_mutex);
1731 /* Rebind all subsystems back to the default hierarchy */
1732 ret = rebind_subsystems(root, 0);
1733 /* Shouldn't be able to fail ... */
1737 * Release all the links from css_sets to this hierarchy's
1740 write_lock(&css_set_lock);
1742 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1744 list_del(&link->cg_link_list);
1745 list_del(&link->cgrp_link_list);
1748 write_unlock(&css_set_lock);
1750 if (!list_empty(&root->root_list)) {
1751 list_del(&root->root_list);
1755 mutex_unlock(&cgroup_root_mutex);
1756 mutex_unlock(&cgroup_mutex);
1758 simple_xattrs_free(&cgrp->xattrs);
1760 kill_litter_super(sb);
1761 cgroup_drop_root(root);
1764 static struct file_system_type cgroup_fs_type = {
1766 .mount = cgroup_mount,
1767 .kill_sb = cgroup_kill_sb,
1770 static struct kobject *cgroup_kobj;
1773 * cgroup_path - generate the path of a cgroup
1774 * @cgrp: the cgroup in question
1775 * @buf: the buffer to write the path into
1776 * @buflen: the length of the buffer
1778 * Writes path of cgroup into buf. Returns 0 on success, -errno on error.
1780 * We can't generate cgroup path using dentry->d_name, as accessing
1781 * dentry->name must be protected by irq-unsafe dentry->d_lock or parent
1782 * inode's i_mutex, while on the other hand cgroup_path() can be called
1783 * with some irq-safe spinlocks held.
1785 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1787 int ret = -ENAMETOOLONG;
1790 start = buf + buflen - 1;
1795 const char *name = cgroup_name(cgrp);
1799 if ((start -= len) < buf)
1801 memcpy(start, name, len);
1810 cgrp = cgrp->parent;
1813 memmove(buf, start, buf + buflen - start);
1818 EXPORT_SYMBOL_GPL(cgroup_path);
1821 * Control Group taskset
1823 struct task_and_cgroup {
1824 struct task_struct *task;
1825 struct cgroup *cgrp;
1829 struct cgroup_taskset {
1830 struct task_and_cgroup single;
1831 struct flex_array *tc_array;
1834 struct cgroup *cur_cgrp;
1838 * cgroup_taskset_first - reset taskset and return the first task
1839 * @tset: taskset of interest
1841 * @tset iteration is initialized and the first task is returned.
1843 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1845 if (tset->tc_array) {
1847 return cgroup_taskset_next(tset);
1849 tset->cur_cgrp = tset->single.cgrp;
1850 return tset->single.task;
1853 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1856 * cgroup_taskset_next - iterate to the next task in taskset
1857 * @tset: taskset of interest
1859 * Return the next task in @tset. Iteration must have been initialized
1860 * with cgroup_taskset_first().
1862 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1864 struct task_and_cgroup *tc;
1866 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1869 tc = flex_array_get(tset->tc_array, tset->idx++);
1870 tset->cur_cgrp = tc->cgrp;
1873 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1876 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1877 * @tset: taskset of interest
1879 * Return the cgroup for the current (last returned) task of @tset. This
1880 * function must be preceded by either cgroup_taskset_first() or
1881 * cgroup_taskset_next().
1883 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1885 return tset->cur_cgrp;
1887 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1890 * cgroup_taskset_size - return the number of tasks in taskset
1891 * @tset: taskset of interest
1893 int cgroup_taskset_size(struct cgroup_taskset *tset)
1895 return tset->tc_array ? tset->tc_array_len : 1;
1897 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1901 * cgroup_task_migrate - move a task from one cgroup to another.
1903 * Must be called with cgroup_mutex and threadgroup locked.
1905 static void cgroup_task_migrate(struct cgroup *oldcgrp,
1906 struct task_struct *tsk, struct css_set *newcg)
1908 struct css_set *oldcg;
1911 * We are synchronized through threadgroup_lock() against PF_EXITING
1912 * setting such that we can't race against cgroup_exit() changing the
1913 * css_set to init_css_set and dropping the old one.
1915 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1916 oldcg = tsk->cgroups;
1919 rcu_assign_pointer(tsk->cgroups, newcg);
1922 /* Update the css_set linked lists if we're using them */
1923 write_lock(&css_set_lock);
1924 if (!list_empty(&tsk->cg_list))
1925 list_move(&tsk->cg_list, &newcg->tasks);
1926 write_unlock(&css_set_lock);
1929 * We just gained a reference on oldcg by taking it from the task. As
1930 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1931 * it here; it will be freed under RCU.
1933 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1938 * cgroup_attach_task - attach a task or a whole threadgroup to a cgroup
1939 * @cgrp: the cgroup to attach to
1940 * @tsk: the task or the leader of the threadgroup to be attached
1941 * @threadgroup: attach the whole threadgroup?
1943 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1944 * task_lock of @tsk or each thread in the threadgroup individually in turn.
1946 static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
1949 int retval, i, group_size;
1950 struct cgroup_subsys *ss, *failed_ss = NULL;
1951 struct cgroupfs_root *root = cgrp->root;
1952 /* threadgroup list cursor and array */
1953 struct task_struct *leader = tsk;
1954 struct task_and_cgroup *tc;
1955 struct flex_array *group;
1956 struct cgroup_taskset tset = { };
1959 * step 0: in order to do expensive, possibly blocking operations for
1960 * every thread, we cannot iterate the thread group list, since it needs
1961 * rcu or tasklist locked. instead, build an array of all threads in the
1962 * group - group_rwsem prevents new threads from appearing, and if
1963 * threads exit, this will just be an over-estimate.
1966 group_size = get_nr_threads(tsk);
1969 /* flex_array supports very large thread-groups better than kmalloc. */
1970 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
1973 /* pre-allocate to guarantee space while iterating in rcu read-side. */
1974 retval = flex_array_prealloc(group, 0, group_size, GFP_KERNEL);
1976 goto out_free_group_list;
1980 * Prevent freeing of tasks while we take a snapshot. Tasks that are
1981 * already PF_EXITING could be freed from underneath us unless we
1982 * take an rcu_read_lock.
1986 struct task_and_cgroup ent;
1988 /* @tsk either already exited or can't exit until the end */
1989 if (tsk->flags & PF_EXITING)
1992 /* as per above, nr_threads may decrease, but not increase. */
1993 BUG_ON(i >= group_size);
1995 ent.cgrp = task_cgroup_from_root(tsk, root);
1996 /* nothing to do if this task is already in the cgroup */
1997 if (ent.cgrp == cgrp)
2000 * saying GFP_ATOMIC has no effect here because we did prealloc
2001 * earlier, but it's good form to communicate our expectations.
2003 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2004 BUG_ON(retval != 0);
2009 } while_each_thread(leader, tsk);
2011 /* remember the number of threads in the array for later. */
2013 tset.tc_array = group;
2014 tset.tc_array_len = group_size;
2016 /* methods shouldn't be called if no task is actually migrating */
2019 goto out_free_group_list;
2022 * step 1: check that we can legitimately attach to the cgroup.
2024 for_each_subsys(root, ss) {
2025 if (ss->can_attach) {
2026 retval = ss->can_attach(cgrp, &tset);
2029 goto out_cancel_attach;
2035 * step 2: make sure css_sets exist for all threads to be migrated.
2036 * we use find_css_set, which allocates a new one if necessary.
2038 for (i = 0; i < group_size; i++) {
2039 tc = flex_array_get(group, i);
2040 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2043 goto out_put_css_set_refs;
2048 * step 3: now that we're guaranteed success wrt the css_sets,
2049 * proceed to move all tasks to the new cgroup. There are no
2050 * failure cases after here, so this is the commit point.
2052 for (i = 0; i < group_size; i++) {
2053 tc = flex_array_get(group, i);
2054 cgroup_task_migrate(tc->cgrp, tc->task, tc->cg);
2056 /* nothing is sensitive to fork() after this point. */
2059 * step 4: do subsystem attach callbacks.
2061 for_each_subsys(root, ss) {
2063 ss->attach(cgrp, &tset);
2067 * step 5: success! and cleanup
2070 out_put_css_set_refs:
2072 for (i = 0; i < group_size; i++) {
2073 tc = flex_array_get(group, i);
2076 put_css_set(tc->cg);
2081 for_each_subsys(root, ss) {
2082 if (ss == failed_ss)
2084 if (ss->cancel_attach)
2085 ss->cancel_attach(cgrp, &tset);
2088 out_free_group_list:
2089 flex_array_free(group);
2094 * Find the task_struct of the task to attach by vpid and pass it along to the
2095 * function to attach either it or all tasks in its threadgroup. Will lock
2096 * cgroup_mutex and threadgroup; may take task_lock of task.
2098 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2100 struct task_struct *tsk;
2101 const struct cred *cred = current_cred(), *tcred;
2104 if (!cgroup_lock_live_group(cgrp))
2110 tsk = find_task_by_vpid(pid);
2114 goto out_unlock_cgroup;
2117 * even if we're attaching all tasks in the thread group, we
2118 * only need to check permissions on one of them.
2120 tcred = __task_cred(tsk);
2121 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2122 !uid_eq(cred->euid, tcred->uid) &&
2123 !uid_eq(cred->euid, tcred->suid)) {
2126 goto out_unlock_cgroup;
2132 tsk = tsk->group_leader;
2135 * Workqueue threads may acquire PF_THREAD_BOUND and become
2136 * trapped in a cpuset, or RT worker may be born in a cgroup
2137 * with no rt_runtime allocated. Just say no.
2139 if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2142 goto out_unlock_cgroup;
2145 get_task_struct(tsk);
2148 threadgroup_lock(tsk);
2150 if (!thread_group_leader(tsk)) {
2152 * a race with de_thread from another thread's exec()
2153 * may strip us of our leadership, if this happens,
2154 * there is no choice but to throw this task away and
2155 * try again; this is
2156 * "double-double-toil-and-trouble-check locking".
2158 threadgroup_unlock(tsk);
2159 put_task_struct(tsk);
2160 goto retry_find_task;
2164 ret = cgroup_attach_task(cgrp, tsk, threadgroup);
2166 threadgroup_unlock(tsk);
2168 put_task_struct(tsk);
2170 mutex_unlock(&cgroup_mutex);
2175 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2176 * @from: attach to all cgroups of a given task
2177 * @tsk: the task to be attached
2179 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2181 struct cgroupfs_root *root;
2184 mutex_lock(&cgroup_mutex);
2185 for_each_active_root(root) {
2186 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2188 retval = cgroup_attach_task(from_cg, tsk, false);
2192 mutex_unlock(&cgroup_mutex);
2196 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2198 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2200 return attach_task_by_pid(cgrp, pid, false);
2203 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2205 return attach_task_by_pid(cgrp, tgid, true);
2208 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2211 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2212 if (strlen(buffer) >= PATH_MAX)
2214 if (!cgroup_lock_live_group(cgrp))
2216 mutex_lock(&cgroup_root_mutex);
2217 strcpy(cgrp->root->release_agent_path, buffer);
2218 mutex_unlock(&cgroup_root_mutex);
2219 mutex_unlock(&cgroup_mutex);
2223 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2224 struct seq_file *seq)
2226 if (!cgroup_lock_live_group(cgrp))
2228 seq_puts(seq, cgrp->root->release_agent_path);
2229 seq_putc(seq, '\n');
2230 mutex_unlock(&cgroup_mutex);
2234 /* A buffer size big enough for numbers or short strings */
2235 #define CGROUP_LOCAL_BUFFER_SIZE 64
2237 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2239 const char __user *userbuf,
2240 size_t nbytes, loff_t *unused_ppos)
2242 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2248 if (nbytes >= sizeof(buffer))
2250 if (copy_from_user(buffer, userbuf, nbytes))
2253 buffer[nbytes] = 0; /* nul-terminate */
2254 if (cft->write_u64) {
2255 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2258 retval = cft->write_u64(cgrp, cft, val);
2260 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2263 retval = cft->write_s64(cgrp, cft, val);
2270 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2272 const char __user *userbuf,
2273 size_t nbytes, loff_t *unused_ppos)
2275 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2277 size_t max_bytes = cft->max_write_len;
2278 char *buffer = local_buffer;
2281 max_bytes = sizeof(local_buffer) - 1;
2282 if (nbytes >= max_bytes)
2284 /* Allocate a dynamic buffer if we need one */
2285 if (nbytes >= sizeof(local_buffer)) {
2286 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2290 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2295 buffer[nbytes] = 0; /* nul-terminate */
2296 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2300 if (buffer != local_buffer)
2305 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2306 size_t nbytes, loff_t *ppos)
2308 struct cftype *cft = __d_cft(file->f_dentry);
2309 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2311 if (cgroup_is_removed(cgrp))
2314 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2315 if (cft->write_u64 || cft->write_s64)
2316 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2317 if (cft->write_string)
2318 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2320 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2321 return ret ? ret : nbytes;
2326 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2328 char __user *buf, size_t nbytes,
2331 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2332 u64 val = cft->read_u64(cgrp, cft);
2333 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2335 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2338 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2340 char __user *buf, size_t nbytes,
2343 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2344 s64 val = cft->read_s64(cgrp, cft);
2345 int len = sprintf(tmp, "%lld\n", (long long) val);
2347 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2350 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2351 size_t nbytes, loff_t *ppos)
2353 struct cftype *cft = __d_cft(file->f_dentry);
2354 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2356 if (cgroup_is_removed(cgrp))
2360 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2362 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2364 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2369 * seqfile ops/methods for returning structured data. Currently just
2370 * supports string->u64 maps, but can be extended in future.
2373 struct cgroup_seqfile_state {
2375 struct cgroup *cgroup;
2378 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2380 struct seq_file *sf = cb->state;
2381 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2384 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2386 struct cgroup_seqfile_state *state = m->private;
2387 struct cftype *cft = state->cft;
2388 if (cft->read_map) {
2389 struct cgroup_map_cb cb = {
2390 .fill = cgroup_map_add,
2393 return cft->read_map(state->cgroup, cft, &cb);
2395 return cft->read_seq_string(state->cgroup, cft, m);
2398 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2400 struct seq_file *seq = file->private_data;
2401 kfree(seq->private);
2402 return single_release(inode, file);
2405 static const struct file_operations cgroup_seqfile_operations = {
2407 .write = cgroup_file_write,
2408 .llseek = seq_lseek,
2409 .release = cgroup_seqfile_release,
2412 static int cgroup_file_open(struct inode *inode, struct file *file)
2417 err = generic_file_open(inode, file);
2420 cft = __d_cft(file->f_dentry);
2422 if (cft->read_map || cft->read_seq_string) {
2423 struct cgroup_seqfile_state *state =
2424 kzalloc(sizeof(*state), GFP_USER);
2428 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2429 file->f_op = &cgroup_seqfile_operations;
2430 err = single_open(file, cgroup_seqfile_show, state);
2433 } else if (cft->open)
2434 err = cft->open(inode, file);
2441 static int cgroup_file_release(struct inode *inode, struct file *file)
2443 struct cftype *cft = __d_cft(file->f_dentry);
2445 return cft->release(inode, file);
2450 * cgroup_rename - Only allow simple rename of directories in place.
2452 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2453 struct inode *new_dir, struct dentry *new_dentry)
2456 struct cgroup_name *name, *old_name;
2457 struct cgroup *cgrp;
2460 * It's convinient to use parent dir's i_mutex to protected
2463 lockdep_assert_held(&old_dir->i_mutex);
2465 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2467 if (new_dentry->d_inode)
2469 if (old_dir != new_dir)
2472 cgrp = __d_cgrp(old_dentry);
2474 name = cgroup_alloc_name(new_dentry);
2478 ret = simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2484 old_name = cgrp->name;
2485 rcu_assign_pointer(cgrp->name, name);
2487 kfree_rcu(old_name, rcu_head);
2491 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2493 if (S_ISDIR(dentry->d_inode->i_mode))
2494 return &__d_cgrp(dentry)->xattrs;
2496 return &__d_cft(dentry)->xattrs;
2499 static inline int xattr_enabled(struct dentry *dentry)
2501 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2502 return test_bit(ROOT_XATTR, &root->flags);
2505 static bool is_valid_xattr(const char *name)
2507 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2508 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2513 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2514 const void *val, size_t size, int flags)
2516 if (!xattr_enabled(dentry))
2518 if (!is_valid_xattr(name))
2520 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2523 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2525 if (!xattr_enabled(dentry))
2527 if (!is_valid_xattr(name))
2529 return simple_xattr_remove(__d_xattrs(dentry), name);
2532 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2533 void *buf, size_t size)
2535 if (!xattr_enabled(dentry))
2537 if (!is_valid_xattr(name))
2539 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2542 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2544 if (!xattr_enabled(dentry))
2546 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2549 static const struct file_operations cgroup_file_operations = {
2550 .read = cgroup_file_read,
2551 .write = cgroup_file_write,
2552 .llseek = generic_file_llseek,
2553 .open = cgroup_file_open,
2554 .release = cgroup_file_release,
2557 static const struct inode_operations cgroup_file_inode_operations = {
2558 .setxattr = cgroup_setxattr,
2559 .getxattr = cgroup_getxattr,
2560 .listxattr = cgroup_listxattr,
2561 .removexattr = cgroup_removexattr,
2564 static const struct inode_operations cgroup_dir_inode_operations = {
2565 .lookup = cgroup_lookup,
2566 .mkdir = cgroup_mkdir,
2567 .rmdir = cgroup_rmdir,
2568 .rename = cgroup_rename,
2569 .setxattr = cgroup_setxattr,
2570 .getxattr = cgroup_getxattr,
2571 .listxattr = cgroup_listxattr,
2572 .removexattr = cgroup_removexattr,
2575 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2577 if (dentry->d_name.len > NAME_MAX)
2578 return ERR_PTR(-ENAMETOOLONG);
2579 d_add(dentry, NULL);
2584 * Check if a file is a control file
2586 static inline struct cftype *__file_cft(struct file *file)
2588 if (file_inode(file)->i_fop != &cgroup_file_operations)
2589 return ERR_PTR(-EINVAL);
2590 return __d_cft(file->f_dentry);
2593 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2594 struct super_block *sb)
2596 struct inode *inode;
2600 if (dentry->d_inode)
2603 inode = cgroup_new_inode(mode, sb);
2607 if (S_ISDIR(mode)) {
2608 inode->i_op = &cgroup_dir_inode_operations;
2609 inode->i_fop = &simple_dir_operations;
2611 /* start off with i_nlink == 2 (for "." entry) */
2613 inc_nlink(dentry->d_parent->d_inode);
2616 * Control reaches here with cgroup_mutex held.
2617 * @inode->i_mutex should nest outside cgroup_mutex but we
2618 * want to populate it immediately without releasing
2619 * cgroup_mutex. As @inode isn't visible to anyone else
2620 * yet, trylock will always succeed without affecting
2623 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2624 } else if (S_ISREG(mode)) {
2626 inode->i_fop = &cgroup_file_operations;
2627 inode->i_op = &cgroup_file_inode_operations;
2629 d_instantiate(dentry, inode);
2630 dget(dentry); /* Extra count - pin the dentry in core */
2635 * cgroup_file_mode - deduce file mode of a control file
2636 * @cft: the control file in question
2638 * returns cft->mode if ->mode is not 0
2639 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2640 * returns S_IRUGO if it has only a read handler
2641 * returns S_IWUSR if it has only a write hander
2643 static umode_t cgroup_file_mode(const struct cftype *cft)
2650 if (cft->read || cft->read_u64 || cft->read_s64 ||
2651 cft->read_map || cft->read_seq_string)
2654 if (cft->write || cft->write_u64 || cft->write_s64 ||
2655 cft->write_string || cft->trigger)
2661 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2664 struct dentry *dir = cgrp->dentry;
2665 struct cgroup *parent = __d_cgrp(dir);
2666 struct dentry *dentry;
2670 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2672 simple_xattrs_init(&cft->xattrs);
2674 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2675 strcpy(name, subsys->name);
2678 strcat(name, cft->name);
2680 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2682 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2686 dentry = lookup_one_len(name, dir, strlen(name));
2687 if (IS_ERR(dentry)) {
2688 error = PTR_ERR(dentry);
2692 mode = cgroup_file_mode(cft);
2693 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2695 cfe->type = (void *)cft;
2696 cfe->dentry = dentry;
2697 dentry->d_fsdata = cfe;
2698 list_add_tail(&cfe->node, &parent->files);
2707 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2708 struct cftype cfts[], bool is_add)
2713 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2714 /* does cft->flags tell us to skip this file on @cgrp? */
2715 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2717 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2721 err = cgroup_add_file(cgrp, subsys, cft);
2723 pr_warn("cgroup_addrm_files: failed to add %s, err=%d\n",
2727 cgroup_rm_file(cgrp, cft);
2733 static DEFINE_MUTEX(cgroup_cft_mutex);
2735 static void cgroup_cfts_prepare(void)
2736 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2739 * Thanks to the entanglement with vfs inode locking, we can't walk
2740 * the existing cgroups under cgroup_mutex and create files.
2741 * Instead, we increment reference on all cgroups and build list of
2742 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2743 * exclusive access to the field.
2745 mutex_lock(&cgroup_cft_mutex);
2746 mutex_lock(&cgroup_mutex);
2749 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2750 struct cftype *cfts, bool is_add)
2751 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2754 struct cgroup *cgrp, *n;
2756 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2757 if (cfts && ss->root != &rootnode) {
2758 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2760 list_add_tail(&cgrp->cft_q_node, &pending);
2764 mutex_unlock(&cgroup_mutex);
2767 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2768 * files for all cgroups which were created before.
2770 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2771 struct inode *inode = cgrp->dentry->d_inode;
2773 mutex_lock(&inode->i_mutex);
2774 mutex_lock(&cgroup_mutex);
2775 if (!cgroup_is_removed(cgrp))
2776 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2777 mutex_unlock(&cgroup_mutex);
2778 mutex_unlock(&inode->i_mutex);
2780 list_del_init(&cgrp->cft_q_node);
2784 mutex_unlock(&cgroup_cft_mutex);
2788 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2789 * @ss: target cgroup subsystem
2790 * @cfts: zero-length name terminated array of cftypes
2792 * Register @cfts to @ss. Files described by @cfts are created for all
2793 * existing cgroups to which @ss is attached and all future cgroups will
2794 * have them too. This function can be called anytime whether @ss is
2797 * Returns 0 on successful registration, -errno on failure. Note that this
2798 * function currently returns 0 as long as @cfts registration is successful
2799 * even if some file creation attempts on existing cgroups fail.
2801 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2803 struct cftype_set *set;
2805 set = kzalloc(sizeof(*set), GFP_KERNEL);
2809 cgroup_cfts_prepare();
2811 list_add_tail(&set->node, &ss->cftsets);
2812 cgroup_cfts_commit(ss, cfts, true);
2816 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2819 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2820 * @ss: target cgroup subsystem
2821 * @cfts: zero-length name terminated array of cftypes
2823 * Unregister @cfts from @ss. Files described by @cfts are removed from
2824 * all existing cgroups to which @ss is attached and all future cgroups
2825 * won't have them either. This function can be called anytime whether @ss
2826 * is attached or not.
2828 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2829 * registered with @ss.
2831 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2833 struct cftype_set *set;
2835 cgroup_cfts_prepare();
2837 list_for_each_entry(set, &ss->cftsets, node) {
2838 if (set->cfts == cfts) {
2839 list_del_init(&set->node);
2840 cgroup_cfts_commit(ss, cfts, false);
2845 cgroup_cfts_commit(ss, NULL, false);
2850 * cgroup_task_count - count the number of tasks in a cgroup.
2851 * @cgrp: the cgroup in question
2853 * Return the number of tasks in the cgroup.
2855 int cgroup_task_count(const struct cgroup *cgrp)
2858 struct cg_cgroup_link *link;
2860 read_lock(&css_set_lock);
2861 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2862 count += atomic_read(&link->cg->refcount);
2864 read_unlock(&css_set_lock);
2869 * Advance a list_head iterator. The iterator should be positioned at
2870 * the start of a css_set
2872 static void cgroup_advance_iter(struct cgroup *cgrp,
2873 struct cgroup_iter *it)
2875 struct list_head *l = it->cg_link;
2876 struct cg_cgroup_link *link;
2879 /* Advance to the next non-empty css_set */
2882 if (l == &cgrp->css_sets) {
2886 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2888 } while (list_empty(&cg->tasks));
2890 it->task = cg->tasks.next;
2894 * To reduce the fork() overhead for systems that are not actually
2895 * using their cgroups capability, we don't maintain the lists running
2896 * through each css_set to its tasks until we see the list actually
2897 * used - in other words after the first call to cgroup_iter_start().
2899 static void cgroup_enable_task_cg_lists(void)
2901 struct task_struct *p, *g;
2902 write_lock(&css_set_lock);
2903 use_task_css_set_links = 1;
2905 * We need tasklist_lock because RCU is not safe against
2906 * while_each_thread(). Besides, a forking task that has passed
2907 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2908 * is not guaranteed to have its child immediately visible in the
2909 * tasklist if we walk through it with RCU.
2911 read_lock(&tasklist_lock);
2912 do_each_thread(g, p) {
2915 * We should check if the process is exiting, otherwise
2916 * it will race with cgroup_exit() in that the list
2917 * entry won't be deleted though the process has exited.
2919 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2920 list_add(&p->cg_list, &p->cgroups->tasks);
2922 } while_each_thread(g, p);
2923 read_unlock(&tasklist_lock);
2924 write_unlock(&css_set_lock);
2928 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2929 * @pos: the current position (%NULL to initiate traversal)
2930 * @cgroup: cgroup whose descendants to walk
2932 * To be used by cgroup_for_each_descendant_pre(). Find the next
2933 * descendant to visit for pre-order traversal of @cgroup's descendants.
2935 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
2936 struct cgroup *cgroup)
2938 struct cgroup *next;
2940 WARN_ON_ONCE(!rcu_read_lock_held());
2942 /* if first iteration, pretend we just visited @cgroup */
2944 if (list_empty(&cgroup->children))
2949 /* visit the first child if exists */
2950 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
2954 /* no child, visit my or the closest ancestor's next sibling */
2956 next = list_entry_rcu(pos->sibling.next, struct cgroup,
2958 if (&next->sibling != &pos->parent->children)
2962 } while (pos != cgroup);
2966 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
2969 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
2970 * @pos: cgroup of interest
2972 * Return the rightmost descendant of @pos. If there's no descendant,
2973 * @pos is returned. This can be used during pre-order traversal to skip
2976 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
2978 struct cgroup *last, *tmp;
2980 WARN_ON_ONCE(!rcu_read_lock_held());
2984 /* ->prev isn't RCU safe, walk ->next till the end */
2986 list_for_each_entry_rcu(tmp, &last->children, sibling)
2992 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
2994 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
2996 struct cgroup *last;
3000 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3008 * cgroup_next_descendant_post - find the next descendant for post-order walk
3009 * @pos: the current position (%NULL to initiate traversal)
3010 * @cgroup: cgroup whose descendants to walk
3012 * To be used by cgroup_for_each_descendant_post(). Find the next
3013 * descendant to visit for post-order traversal of @cgroup's descendants.
3015 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3016 struct cgroup *cgroup)
3018 struct cgroup *next;
3020 WARN_ON_ONCE(!rcu_read_lock_held());
3022 /* if first iteration, visit the leftmost descendant */
3024 next = cgroup_leftmost_descendant(cgroup);
3025 return next != cgroup ? next : NULL;
3028 /* if there's an unvisited sibling, visit its leftmost descendant */
3029 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3030 if (&next->sibling != &pos->parent->children)
3031 return cgroup_leftmost_descendant(next);
3033 /* no sibling left, visit parent */
3035 return next != cgroup ? next : NULL;
3037 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3039 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3040 __acquires(css_set_lock)
3043 * The first time anyone tries to iterate across a cgroup,
3044 * we need to enable the list linking each css_set to its
3045 * tasks, and fix up all existing tasks.
3047 if (!use_task_css_set_links)
3048 cgroup_enable_task_cg_lists();
3050 read_lock(&css_set_lock);
3051 it->cg_link = &cgrp->css_sets;
3052 cgroup_advance_iter(cgrp, it);
3055 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3056 struct cgroup_iter *it)
3058 struct task_struct *res;
3059 struct list_head *l = it->task;
3060 struct cg_cgroup_link *link;
3062 /* If the iterator cg is NULL, we have no tasks */
3065 res = list_entry(l, struct task_struct, cg_list);
3066 /* Advance iterator to find next entry */
3068 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
3069 if (l == &link->cg->tasks) {
3070 /* We reached the end of this task list - move on to
3071 * the next cg_cgroup_link */
3072 cgroup_advance_iter(cgrp, it);
3079 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3080 __releases(css_set_lock)
3082 read_unlock(&css_set_lock);
3085 static inline int started_after_time(struct task_struct *t1,
3086 struct timespec *time,
3087 struct task_struct *t2)
3089 int start_diff = timespec_compare(&t1->start_time, time);
3090 if (start_diff > 0) {
3092 } else if (start_diff < 0) {
3096 * Arbitrarily, if two processes started at the same
3097 * time, we'll say that the lower pointer value
3098 * started first. Note that t2 may have exited by now
3099 * so this may not be a valid pointer any longer, but
3100 * that's fine - it still serves to distinguish
3101 * between two tasks started (effectively) simultaneously.
3108 * This function is a callback from heap_insert() and is used to order
3110 * In this case we order the heap in descending task start time.
3112 static inline int started_after(void *p1, void *p2)
3114 struct task_struct *t1 = p1;
3115 struct task_struct *t2 = p2;
3116 return started_after_time(t1, &t2->start_time, t2);
3120 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3121 * @scan: struct cgroup_scanner containing arguments for the scan
3123 * Arguments include pointers to callback functions test_task() and
3125 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3126 * and if it returns true, call process_task() for it also.
3127 * The test_task pointer may be NULL, meaning always true (select all tasks).
3128 * Effectively duplicates cgroup_iter_{start,next,end}()
3129 * but does not lock css_set_lock for the call to process_task().
3130 * The struct cgroup_scanner may be embedded in any structure of the caller's
3132 * It is guaranteed that process_task() will act on every task that
3133 * is a member of the cgroup for the duration of this call. This
3134 * function may or may not call process_task() for tasks that exit
3135 * or move to a different cgroup during the call, or are forked or
3136 * move into the cgroup during the call.
3138 * Note that test_task() may be called with locks held, and may in some
3139 * situations be called multiple times for the same task, so it should
3141 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3142 * pre-allocated and will be used for heap operations (and its "gt" member will
3143 * be overwritten), else a temporary heap will be used (allocation of which
3144 * may cause this function to fail).
3146 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3149 struct cgroup_iter it;
3150 struct task_struct *p, *dropped;
3151 /* Never dereference latest_task, since it's not refcounted */
3152 struct task_struct *latest_task = NULL;
3153 struct ptr_heap tmp_heap;
3154 struct ptr_heap *heap;
3155 struct timespec latest_time = { 0, 0 };
3158 /* The caller supplied our heap and pre-allocated its memory */
3160 heap->gt = &started_after;
3162 /* We need to allocate our own heap memory */
3164 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3166 /* cannot allocate the heap */
3172 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3173 * to determine which are of interest, and using the scanner's
3174 * "process_task" callback to process any of them that need an update.
3175 * Since we don't want to hold any locks during the task updates,
3176 * gather tasks to be processed in a heap structure.
3177 * The heap is sorted by descending task start time.
3178 * If the statically-sized heap fills up, we overflow tasks that
3179 * started later, and in future iterations only consider tasks that
3180 * started after the latest task in the previous pass. This
3181 * guarantees forward progress and that we don't miss any tasks.
3184 cgroup_iter_start(scan->cg, &it);
3185 while ((p = cgroup_iter_next(scan->cg, &it))) {
3187 * Only affect tasks that qualify per the caller's callback,
3188 * if he provided one
3190 if (scan->test_task && !scan->test_task(p, scan))
3193 * Only process tasks that started after the last task
3196 if (!started_after_time(p, &latest_time, latest_task))
3198 dropped = heap_insert(heap, p);
3199 if (dropped == NULL) {
3201 * The new task was inserted; the heap wasn't
3205 } else if (dropped != p) {
3207 * The new task was inserted, and pushed out a
3211 put_task_struct(dropped);
3214 * Else the new task was newer than anything already in
3215 * the heap and wasn't inserted
3218 cgroup_iter_end(scan->cg, &it);
3221 for (i = 0; i < heap->size; i++) {
3222 struct task_struct *q = heap->ptrs[i];
3224 latest_time = q->start_time;
3227 /* Process the task per the caller's callback */
3228 scan->process_task(q, scan);
3232 * If we had to process any tasks at all, scan again
3233 * in case some of them were in the middle of forking
3234 * children that didn't get processed.
3235 * Not the most efficient way to do it, but it avoids
3236 * having to take callback_mutex in the fork path
3240 if (heap == &tmp_heap)
3241 heap_free(&tmp_heap);
3245 static void cgroup_transfer_one_task(struct task_struct *task,
3246 struct cgroup_scanner *scan)
3248 struct cgroup *new_cgroup = scan->data;
3250 mutex_lock(&cgroup_mutex);
3251 cgroup_attach_task(new_cgroup, task, false);
3252 mutex_unlock(&cgroup_mutex);
3256 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
3257 * @to: cgroup to which the tasks will be moved
3258 * @from: cgroup in which the tasks currently reside
3260 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
3262 struct cgroup_scanner scan;
3265 scan.test_task = NULL; /* select all tasks in cgroup */
3266 scan.process_task = cgroup_transfer_one_task;
3270 return cgroup_scan_tasks(&scan);
3274 * Stuff for reading the 'tasks'/'procs' files.
3276 * Reading this file can return large amounts of data if a cgroup has
3277 * *lots* of attached tasks. So it may need several calls to read(),
3278 * but we cannot guarantee that the information we produce is correct
3279 * unless we produce it entirely atomically.
3283 /* which pidlist file are we talking about? */
3284 enum cgroup_filetype {
3290 * A pidlist is a list of pids that virtually represents the contents of one
3291 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3292 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3295 struct cgroup_pidlist {
3297 * used to find which pidlist is wanted. doesn't change as long as
3298 * this particular list stays in the list.
3300 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3303 /* how many elements the above list has */
3305 /* how many files are using the current array */
3307 /* each of these stored in a list by its cgroup */
3308 struct list_head links;
3309 /* pointer to the cgroup we belong to, for list removal purposes */
3310 struct cgroup *owner;
3311 /* protects the other fields */
3312 struct rw_semaphore mutex;
3316 * The following two functions "fix" the issue where there are more pids
3317 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3318 * TODO: replace with a kernel-wide solution to this problem
3320 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3321 static void *pidlist_allocate(int count)
3323 if (PIDLIST_TOO_LARGE(count))
3324 return vmalloc(count * sizeof(pid_t));
3326 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3328 static void pidlist_free(void *p)
3330 if (is_vmalloc_addr(p))
3337 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3338 * Returns the number of unique elements.
3340 static int pidlist_uniq(pid_t *list, int length)
3345 * we presume the 0th element is unique, so i starts at 1. trivial
3346 * edge cases first; no work needs to be done for either
3348 if (length == 0 || length == 1)
3350 /* src and dest walk down the list; dest counts unique elements */
3351 for (src = 1; src < length; src++) {
3352 /* find next unique element */
3353 while (list[src] == list[src-1]) {
3358 /* dest always points to where the next unique element goes */
3359 list[dest] = list[src];
3366 static int cmppid(const void *a, const void *b)
3368 return *(pid_t *)a - *(pid_t *)b;
3372 * find the appropriate pidlist for our purpose (given procs vs tasks)
3373 * returns with the lock on that pidlist already held, and takes care
3374 * of the use count, or returns NULL with no locks held if we're out of
3377 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3378 enum cgroup_filetype type)
3380 struct cgroup_pidlist *l;
3381 /* don't need task_nsproxy() if we're looking at ourself */
3382 struct pid_namespace *ns = task_active_pid_ns(current);
3385 * We can't drop the pidlist_mutex before taking the l->mutex in case
3386 * the last ref-holder is trying to remove l from the list at the same
3387 * time. Holding the pidlist_mutex precludes somebody taking whichever
3388 * list we find out from under us - compare release_pid_array().
3390 mutex_lock(&cgrp->pidlist_mutex);
3391 list_for_each_entry(l, &cgrp->pidlists, links) {
3392 if (l->key.type == type && l->key.ns == ns) {
3393 /* make sure l doesn't vanish out from under us */
3394 down_write(&l->mutex);
3395 mutex_unlock(&cgrp->pidlist_mutex);
3399 /* entry not found; create a new one */
3400 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3402 mutex_unlock(&cgrp->pidlist_mutex);
3405 init_rwsem(&l->mutex);
3406 down_write(&l->mutex);
3408 l->key.ns = get_pid_ns(ns);
3409 l->use_count = 0; /* don't increment here */
3412 list_add(&l->links, &cgrp->pidlists);
3413 mutex_unlock(&cgrp->pidlist_mutex);
3418 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3420 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3421 struct cgroup_pidlist **lp)
3425 int pid, n = 0; /* used for populating the array */
3426 struct cgroup_iter it;
3427 struct task_struct *tsk;
3428 struct cgroup_pidlist *l;
3431 * If cgroup gets more users after we read count, we won't have
3432 * enough space - tough. This race is indistinguishable to the
3433 * caller from the case that the additional cgroup users didn't
3434 * show up until sometime later on.
3436 length = cgroup_task_count(cgrp);
3437 array = pidlist_allocate(length);
3440 /* now, populate the array */
3441 cgroup_iter_start(cgrp, &it);
3442 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3443 if (unlikely(n == length))
3445 /* get tgid or pid for procs or tasks file respectively */
3446 if (type == CGROUP_FILE_PROCS)
3447 pid = task_tgid_vnr(tsk);
3449 pid = task_pid_vnr(tsk);
3450 if (pid > 0) /* make sure to only use valid results */
3453 cgroup_iter_end(cgrp, &it);
3455 /* now sort & (if procs) strip out duplicates */
3456 sort(array, length, sizeof(pid_t), cmppid, NULL);
3457 if (type == CGROUP_FILE_PROCS)
3458 length = pidlist_uniq(array, length);
3459 l = cgroup_pidlist_find(cgrp, type);
3461 pidlist_free(array);
3464 /* store array, freeing old if necessary - lock already held */
3465 pidlist_free(l->list);
3469 up_write(&l->mutex);
3475 * cgroupstats_build - build and fill cgroupstats
3476 * @stats: cgroupstats to fill information into
3477 * @dentry: A dentry entry belonging to the cgroup for which stats have
3480 * Build and fill cgroupstats so that taskstats can export it to user
3483 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3486 struct cgroup *cgrp;
3487 struct cgroup_iter it;
3488 struct task_struct *tsk;
3491 * Validate dentry by checking the superblock operations,
3492 * and make sure it's a directory.
3494 if (dentry->d_sb->s_op != &cgroup_ops ||
3495 !S_ISDIR(dentry->d_inode->i_mode))
3499 cgrp = dentry->d_fsdata;
3501 cgroup_iter_start(cgrp, &it);
3502 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3503 switch (tsk->state) {
3505 stats->nr_running++;
3507 case TASK_INTERRUPTIBLE:
3508 stats->nr_sleeping++;
3510 case TASK_UNINTERRUPTIBLE:
3511 stats->nr_uninterruptible++;
3514 stats->nr_stopped++;
3517 if (delayacct_is_task_waiting_on_io(tsk))
3518 stats->nr_io_wait++;
3522 cgroup_iter_end(cgrp, &it);
3530 * seq_file methods for the tasks/procs files. The seq_file position is the
3531 * next pid to display; the seq_file iterator is a pointer to the pid
3532 * in the cgroup->l->list array.
3535 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3538 * Initially we receive a position value that corresponds to
3539 * one more than the last pid shown (or 0 on the first call or
3540 * after a seek to the start). Use a binary-search to find the
3541 * next pid to display, if any
3543 struct cgroup_pidlist *l = s->private;
3544 int index = 0, pid = *pos;
3547 down_read(&l->mutex);
3549 int end = l->length;
3551 while (index < end) {
3552 int mid = (index + end) / 2;
3553 if (l->list[mid] == pid) {
3556 } else if (l->list[mid] <= pid)
3562 /* If we're off the end of the array, we're done */
3563 if (index >= l->length)
3565 /* Update the abstract position to be the actual pid that we found */
3566 iter = l->list + index;
3571 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3573 struct cgroup_pidlist *l = s->private;
3577 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3579 struct cgroup_pidlist *l = s->private;
3581 pid_t *end = l->list + l->length;
3583 * Advance to the next pid in the array. If this goes off the
3595 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3597 return seq_printf(s, "%d\n", *(int *)v);
3601 * seq_operations functions for iterating on pidlists through seq_file -
3602 * independent of whether it's tasks or procs
3604 static const struct seq_operations cgroup_pidlist_seq_operations = {
3605 .start = cgroup_pidlist_start,
3606 .stop = cgroup_pidlist_stop,
3607 .next = cgroup_pidlist_next,
3608 .show = cgroup_pidlist_show,
3611 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3614 * the case where we're the last user of this particular pidlist will
3615 * have us remove it from the cgroup's list, which entails taking the
3616 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3617 * pidlist_mutex, we have to take pidlist_mutex first.
3619 mutex_lock(&l->owner->pidlist_mutex);
3620 down_write(&l->mutex);
3621 BUG_ON(!l->use_count);
3622 if (!--l->use_count) {
3623 /* we're the last user if refcount is 0; remove and free */
3624 list_del(&l->links);
3625 mutex_unlock(&l->owner->pidlist_mutex);
3626 pidlist_free(l->list);
3627 put_pid_ns(l->key.ns);
3628 up_write(&l->mutex);
3632 mutex_unlock(&l->owner->pidlist_mutex);
3633 up_write(&l->mutex);
3636 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3638 struct cgroup_pidlist *l;
3639 if (!(file->f_mode & FMODE_READ))
3642 * the seq_file will only be initialized if the file was opened for
3643 * reading; hence we check if it's not null only in that case.
3645 l = ((struct seq_file *)file->private_data)->private;
3646 cgroup_release_pid_array(l);
3647 return seq_release(inode, file);
3650 static const struct file_operations cgroup_pidlist_operations = {
3652 .llseek = seq_lseek,
3653 .write = cgroup_file_write,
3654 .release = cgroup_pidlist_release,
3658 * The following functions handle opens on a file that displays a pidlist
3659 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3662 /* helper function for the two below it */
3663 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3665 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3666 struct cgroup_pidlist *l;
3669 /* Nothing to do for write-only files */
3670 if (!(file->f_mode & FMODE_READ))
3673 /* have the array populated */
3674 retval = pidlist_array_load(cgrp, type, &l);
3677 /* configure file information */
3678 file->f_op = &cgroup_pidlist_operations;
3680 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3682 cgroup_release_pid_array(l);
3685 ((struct seq_file *)file->private_data)->private = l;
3688 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3690 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3692 static int cgroup_procs_open(struct inode *unused, struct file *file)
3694 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3697 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3700 return notify_on_release(cgrp);
3703 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3707 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3709 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3711 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3716 * Unregister event and free resources.
3718 * Gets called from workqueue.
3720 static void cgroup_event_remove(struct work_struct *work)
3722 struct cgroup_event *event = container_of(work, struct cgroup_event,
3724 struct cgroup *cgrp = event->cgrp;
3726 remove_wait_queue(event->wqh, &event->wait);
3728 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3730 /* Notify userspace the event is going away. */
3731 eventfd_signal(event->eventfd, 1);
3733 eventfd_ctx_put(event->eventfd);
3739 * Gets called on POLLHUP on eventfd when user closes it.
3741 * Called with wqh->lock held and interrupts disabled.
3743 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3744 int sync, void *key)
3746 struct cgroup_event *event = container_of(wait,
3747 struct cgroup_event, wait);
3748 struct cgroup *cgrp = event->cgrp;
3749 unsigned long flags = (unsigned long)key;
3751 if (flags & POLLHUP) {
3753 * If the event has been detached at cgroup removal, we
3754 * can simply return knowing the other side will cleanup
3757 * We can't race against event freeing since the other
3758 * side will require wqh->lock via remove_wait_queue(),
3761 spin_lock(&cgrp->event_list_lock);
3762 if (!list_empty(&event->list)) {
3763 list_del_init(&event->list);
3765 * We are in atomic context, but cgroup_event_remove()
3766 * may sleep, so we have to call it in workqueue.
3768 schedule_work(&event->remove);
3770 spin_unlock(&cgrp->event_list_lock);
3776 static void cgroup_event_ptable_queue_proc(struct file *file,
3777 wait_queue_head_t *wqh, poll_table *pt)
3779 struct cgroup_event *event = container_of(pt,
3780 struct cgroup_event, pt);
3783 add_wait_queue(wqh, &event->wait);
3787 * Parse input and register new cgroup event handler.
3789 * Input must be in format '<event_fd> <control_fd> <args>'.
3790 * Interpretation of args is defined by control file implementation.
3792 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3795 struct cgroup_event *event = NULL;
3796 struct cgroup *cgrp_cfile;
3797 unsigned int efd, cfd;
3798 struct file *efile = NULL;
3799 struct file *cfile = NULL;
3803 efd = simple_strtoul(buffer, &endp, 10);
3808 cfd = simple_strtoul(buffer, &endp, 10);
3809 if ((*endp != ' ') && (*endp != '\0'))
3813 event = kzalloc(sizeof(*event), GFP_KERNEL);
3817 INIT_LIST_HEAD(&event->list);
3818 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3819 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3820 INIT_WORK(&event->remove, cgroup_event_remove);
3822 efile = eventfd_fget(efd);
3823 if (IS_ERR(efile)) {
3824 ret = PTR_ERR(efile);
3828 event->eventfd = eventfd_ctx_fileget(efile);
3829 if (IS_ERR(event->eventfd)) {
3830 ret = PTR_ERR(event->eventfd);
3840 /* the process need read permission on control file */
3841 /* AV: shouldn't we check that it's been opened for read instead? */
3842 ret = inode_permission(file_inode(cfile), MAY_READ);
3846 event->cft = __file_cft(cfile);
3847 if (IS_ERR(event->cft)) {
3848 ret = PTR_ERR(event->cft);
3853 * The file to be monitored must be in the same cgroup as
3854 * cgroup.event_control is.
3856 cgrp_cfile = __d_cgrp(cfile->f_dentry->d_parent);
3857 if (cgrp_cfile != cgrp) {
3862 if (!event->cft->register_event || !event->cft->unregister_event) {
3867 ret = event->cft->register_event(cgrp, event->cft,
3868 event->eventfd, buffer);
3873 * Events should be removed after rmdir of cgroup directory, but before
3874 * destroying subsystem state objects. Let's take reference to cgroup
3875 * directory dentry to do that.
3879 spin_lock(&cgrp->event_list_lock);
3880 list_add(&event->list, &cgrp->event_list);
3881 spin_unlock(&cgrp->event_list_lock);
3892 if (event && event->eventfd && !IS_ERR(event->eventfd))
3893 eventfd_ctx_put(event->eventfd);
3895 if (!IS_ERR_OR_NULL(efile))
3903 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3906 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3909 static int cgroup_clone_children_write(struct cgroup *cgrp,
3914 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3916 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3921 * for the common functions, 'private' gives the type of file
3923 /* for hysterical raisins, we can't put this on the older files */
3924 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3925 static struct cftype files[] = {
3928 .open = cgroup_tasks_open,
3929 .write_u64 = cgroup_tasks_write,
3930 .release = cgroup_pidlist_release,
3931 .mode = S_IRUGO | S_IWUSR,
3934 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3935 .open = cgroup_procs_open,
3936 .write_u64 = cgroup_procs_write,
3937 .release = cgroup_pidlist_release,
3938 .mode = S_IRUGO | S_IWUSR,
3941 .name = "notify_on_release",
3942 .read_u64 = cgroup_read_notify_on_release,
3943 .write_u64 = cgroup_write_notify_on_release,
3946 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3947 .write_string = cgroup_write_event_control,
3951 .name = "cgroup.clone_children",
3952 .read_u64 = cgroup_clone_children_read,
3953 .write_u64 = cgroup_clone_children_write,
3956 .name = "release_agent",
3957 .flags = CFTYPE_ONLY_ON_ROOT,
3958 .read_seq_string = cgroup_release_agent_show,
3959 .write_string = cgroup_release_agent_write,
3960 .max_write_len = PATH_MAX,
3966 * cgroup_populate_dir - selectively creation of files in a directory
3967 * @cgrp: target cgroup
3968 * @base_files: true if the base files should be added
3969 * @subsys_mask: mask of the subsystem ids whose files should be added
3971 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
3972 unsigned long subsys_mask)
3975 struct cgroup_subsys *ss;
3978 err = cgroup_addrm_files(cgrp, NULL, files, true);
3983 /* process cftsets of each subsystem */
3984 for_each_subsys(cgrp->root, ss) {
3985 struct cftype_set *set;
3986 if (!test_bit(ss->subsys_id, &subsys_mask))
3989 list_for_each_entry(set, &ss->cftsets, node)
3990 cgroup_addrm_files(cgrp, ss, set->cfts, true);
3993 /* This cgroup is ready now */
3994 for_each_subsys(cgrp->root, ss) {
3995 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3997 * Update id->css pointer and make this css visible from
3998 * CSS ID functions. This pointer will be dereferened
3999 * from RCU-read-side without locks.
4002 rcu_assign_pointer(css->id->css, css);
4008 static void css_dput_fn(struct work_struct *work)
4010 struct cgroup_subsys_state *css =
4011 container_of(work, struct cgroup_subsys_state, dput_work);
4012 struct dentry *dentry = css->cgroup->dentry;
4013 struct super_block *sb = dentry->d_sb;
4015 atomic_inc(&sb->s_active);
4017 deactivate_super(sb);
4020 static void init_cgroup_css(struct cgroup_subsys_state *css,
4021 struct cgroup_subsys *ss,
4022 struct cgroup *cgrp)
4025 atomic_set(&css->refcnt, 1);
4028 if (cgrp == dummytop)
4029 css->flags |= CSS_ROOT;
4030 BUG_ON(cgrp->subsys[ss->subsys_id]);
4031 cgrp->subsys[ss->subsys_id] = css;
4034 * css holds an extra ref to @cgrp->dentry which is put on the last
4035 * css_put(). dput() requires process context, which css_put() may
4036 * be called without. @css->dput_work will be used to invoke
4037 * dput() asynchronously from css_put().
4039 INIT_WORK(&css->dput_work, css_dput_fn);
4042 /* invoke ->post_create() on a new CSS and mark it online if successful */
4043 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4047 lockdep_assert_held(&cgroup_mutex);
4050 ret = ss->css_online(cgrp);
4052 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4056 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4057 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4058 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4060 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4062 lockdep_assert_held(&cgroup_mutex);
4064 if (!(css->flags & CSS_ONLINE))
4067 if (ss->css_offline)
4068 ss->css_offline(cgrp);
4070 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4074 * cgroup_create - create a cgroup
4075 * @parent: cgroup that will be parent of the new cgroup
4076 * @dentry: dentry of the new cgroup
4077 * @mode: mode to set on new inode
4079 * Must be called with the mutex on the parent inode held
4081 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4084 struct cgroup *cgrp;
4085 struct cgroup_name *name;
4086 struct cgroupfs_root *root = parent->root;
4088 struct cgroup_subsys *ss;
4089 struct super_block *sb = root->sb;
4091 /* allocate the cgroup and its ID, 0 is reserved for the root */
4092 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4096 name = cgroup_alloc_name(dentry);
4099 rcu_assign_pointer(cgrp->name, name);
4101 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4106 * Only live parents can have children. Note that the liveliness
4107 * check isn't strictly necessary because cgroup_mkdir() and
4108 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4109 * anyway so that locking is contained inside cgroup proper and we
4110 * don't get nasty surprises if we ever grow another caller.
4112 if (!cgroup_lock_live_group(parent)) {
4117 /* Grab a reference on the superblock so the hierarchy doesn't
4118 * get deleted on unmount if there are child cgroups. This
4119 * can be done outside cgroup_mutex, since the sb can't
4120 * disappear while someone has an open control file on the
4122 atomic_inc(&sb->s_active);
4124 init_cgroup_housekeeping(cgrp);
4126 dentry->d_fsdata = cgrp;
4127 cgrp->dentry = dentry;
4129 cgrp->parent = parent;
4130 cgrp->root = parent->root;
4131 cgrp->top_cgroup = parent->top_cgroup;
4133 if (notify_on_release(parent))
4134 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4136 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4137 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4139 for_each_subsys(root, ss) {
4140 struct cgroup_subsys_state *css;
4142 css = ss->css_alloc(cgrp);
4147 init_cgroup_css(css, ss, cgrp);
4149 err = alloc_css_id(ss, parent, cgrp);
4156 * Create directory. cgroup_create_file() returns with the new
4157 * directory locked on success so that it can be populated without
4158 * dropping cgroup_mutex.
4160 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4163 lockdep_assert_held(&dentry->d_inode->i_mutex);
4165 /* allocation complete, commit to creation */
4166 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4167 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4168 root->number_of_cgroups++;
4170 /* each css holds a ref to the cgroup's dentry */
4171 for_each_subsys(root, ss)
4174 /* hold a ref to the parent's dentry */
4175 dget(parent->dentry);
4177 /* creation succeeded, notify subsystems */
4178 for_each_subsys(root, ss) {
4179 err = online_css(ss, cgrp);
4183 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4185 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",
4186 current->comm, current->pid, ss->name);
4187 if (!strcmp(ss->name, "memory"))
4188 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4189 ss->warned_broken_hierarchy = true;
4193 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4197 mutex_unlock(&cgroup_mutex);
4198 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4203 for_each_subsys(root, ss) {
4204 if (cgrp->subsys[ss->subsys_id])
4207 mutex_unlock(&cgroup_mutex);
4208 /* Release the reference count that we took on the superblock */
4209 deactivate_super(sb);
4211 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4213 kfree(rcu_dereference_raw(cgrp->name));
4219 cgroup_destroy_locked(cgrp);
4220 mutex_unlock(&cgroup_mutex);
4221 mutex_unlock(&dentry->d_inode->i_mutex);
4225 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4227 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4229 /* the vfs holds inode->i_mutex already */
4230 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4233 static int cgroup_destroy_locked(struct cgroup *cgrp)
4234 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4236 struct dentry *d = cgrp->dentry;
4237 struct cgroup *parent = cgrp->parent;
4238 struct cgroup_event *event, *tmp;
4239 struct cgroup_subsys *ss;
4241 lockdep_assert_held(&d->d_inode->i_mutex);
4242 lockdep_assert_held(&cgroup_mutex);
4244 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children))
4248 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4249 * removed. This makes future css_tryget() and child creation
4250 * attempts fail thus maintaining the removal conditions verified
4253 for_each_subsys(cgrp->root, ss) {
4254 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4256 WARN_ON(atomic_read(&css->refcnt) < 0);
4257 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4259 set_bit(CGRP_REMOVED, &cgrp->flags);
4261 /* tell subsystems to initate destruction */
4262 for_each_subsys(cgrp->root, ss)
4263 offline_css(ss, cgrp);
4266 * Put all the base refs. Each css holds an extra reference to the
4267 * cgroup's dentry and cgroup removal proceeds regardless of css
4268 * refs. On the last put of each css, whenever that may be, the
4269 * extra dentry ref is put so that dentry destruction happens only
4270 * after all css's are released.
4272 for_each_subsys(cgrp->root, ss)
4273 css_put(cgrp->subsys[ss->subsys_id]);
4275 raw_spin_lock(&release_list_lock);
4276 if (!list_empty(&cgrp->release_list))
4277 list_del_init(&cgrp->release_list);
4278 raw_spin_unlock(&release_list_lock);
4280 /* delete this cgroup from parent->children */
4281 list_del_rcu(&cgrp->sibling);
4282 list_del_init(&cgrp->allcg_node);
4285 cgroup_d_remove_dir(d);
4288 set_bit(CGRP_RELEASABLE, &parent->flags);
4289 check_for_release(parent);
4292 * Unregister events and notify userspace.
4293 * Notify userspace about cgroup removing only after rmdir of cgroup
4294 * directory to avoid race between userspace and kernelspace.
4296 spin_lock(&cgrp->event_list_lock);
4297 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4298 list_del_init(&event->list);
4299 schedule_work(&event->remove);
4301 spin_unlock(&cgrp->event_list_lock);
4306 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4310 mutex_lock(&cgroup_mutex);
4311 ret = cgroup_destroy_locked(dentry->d_fsdata);
4312 mutex_unlock(&cgroup_mutex);
4317 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4319 INIT_LIST_HEAD(&ss->cftsets);
4322 * base_cftset is embedded in subsys itself, no need to worry about
4325 if (ss->base_cftypes) {
4326 ss->base_cftset.cfts = ss->base_cftypes;
4327 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4331 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4333 struct cgroup_subsys_state *css;
4335 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4337 mutex_lock(&cgroup_mutex);
4339 /* init base cftset */
4340 cgroup_init_cftsets(ss);
4342 /* Create the top cgroup state for this subsystem */
4343 list_add(&ss->sibling, &rootnode.subsys_list);
4344 ss->root = &rootnode;
4345 css = ss->css_alloc(dummytop);
4346 /* We don't handle early failures gracefully */
4347 BUG_ON(IS_ERR(css));
4348 init_cgroup_css(css, ss, dummytop);
4350 /* Update the init_css_set to contain a subsys
4351 * pointer to this state - since the subsystem is
4352 * newly registered, all tasks and hence the
4353 * init_css_set is in the subsystem's top cgroup. */
4354 init_css_set.subsys[ss->subsys_id] = css;
4356 need_forkexit_callback |= ss->fork || ss->exit;
4358 /* At system boot, before all subsystems have been
4359 * registered, no tasks have been forked, so we don't
4360 * need to invoke fork callbacks here. */
4361 BUG_ON(!list_empty(&init_task.tasks));
4364 BUG_ON(online_css(ss, dummytop));
4366 mutex_unlock(&cgroup_mutex);
4368 /* this function shouldn't be used with modular subsystems, since they
4369 * need to register a subsys_id, among other things */
4374 * cgroup_load_subsys: load and register a modular subsystem at runtime
4375 * @ss: the subsystem to load
4377 * This function should be called in a modular subsystem's initcall. If the
4378 * subsystem is built as a module, it will be assigned a new subsys_id and set
4379 * up for use. If the subsystem is built-in anyway, work is delegated to the
4380 * simpler cgroup_init_subsys.
4382 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4384 struct cgroup_subsys_state *css;
4386 struct hlist_node *tmp;
4390 /* check name and function validity */
4391 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4392 ss->css_alloc == NULL || ss->css_free == NULL)
4396 * we don't support callbacks in modular subsystems. this check is
4397 * before the ss->module check for consistency; a subsystem that could
4398 * be a module should still have no callbacks even if the user isn't
4399 * compiling it as one.
4401 if (ss->fork || ss->exit)
4405 * an optionally modular subsystem is built-in: we want to do nothing,
4406 * since cgroup_init_subsys will have already taken care of it.
4408 if (ss->module == NULL) {
4409 /* a sanity check */
4410 BUG_ON(subsys[ss->subsys_id] != ss);
4414 /* init base cftset */
4415 cgroup_init_cftsets(ss);
4417 mutex_lock(&cgroup_mutex);
4418 subsys[ss->subsys_id] = ss;
4421 * no ss->css_alloc seems to need anything important in the ss
4422 * struct, so this can happen first (i.e. before the rootnode
4425 css = ss->css_alloc(dummytop);
4427 /* failure case - need to deassign the subsys[] slot. */
4428 subsys[ss->subsys_id] = NULL;
4429 mutex_unlock(&cgroup_mutex);
4430 return PTR_ERR(css);
4433 list_add(&ss->sibling, &rootnode.subsys_list);
4434 ss->root = &rootnode;
4436 /* our new subsystem will be attached to the dummy hierarchy. */
4437 init_cgroup_css(css, ss, dummytop);
4438 /* init_idr must be after init_cgroup_css because it sets css->id. */
4440 ret = cgroup_init_idr(ss, css);
4446 * Now we need to entangle the css into the existing css_sets. unlike
4447 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4448 * will need a new pointer to it; done by iterating the css_set_table.
4449 * furthermore, modifying the existing css_sets will corrupt the hash
4450 * table state, so each changed css_set will need its hash recomputed.
4451 * this is all done under the css_set_lock.
4453 write_lock(&css_set_lock);
4454 hash_for_each_safe(css_set_table, i, tmp, cg, hlist) {
4455 /* skip entries that we already rehashed */
4456 if (cg->subsys[ss->subsys_id])
4458 /* remove existing entry */
4459 hash_del(&cg->hlist);
4461 cg->subsys[ss->subsys_id] = css;
4462 /* recompute hash and restore entry */
4463 key = css_set_hash(cg->subsys);
4464 hash_add(css_set_table, &cg->hlist, key);
4466 write_unlock(&css_set_lock);
4469 ret = online_css(ss, dummytop);
4474 mutex_unlock(&cgroup_mutex);
4478 mutex_unlock(&cgroup_mutex);
4479 /* @ss can't be mounted here as try_module_get() would fail */
4480 cgroup_unload_subsys(ss);
4483 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4486 * cgroup_unload_subsys: unload a modular subsystem
4487 * @ss: the subsystem to unload
4489 * This function should be called in a modular subsystem's exitcall. When this
4490 * function is invoked, the refcount on the subsystem's module will be 0, so
4491 * the subsystem will not be attached to any hierarchy.
4493 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4495 struct cg_cgroup_link *link;
4497 BUG_ON(ss->module == NULL);
4500 * we shouldn't be called if the subsystem is in use, and the use of
4501 * try_module_get in parse_cgroupfs_options should ensure that it
4502 * doesn't start being used while we're killing it off.
4504 BUG_ON(ss->root != &rootnode);
4506 mutex_lock(&cgroup_mutex);
4508 offline_css(ss, dummytop);
4512 idr_destroy(&ss->idr);
4514 /* deassign the subsys_id */
4515 subsys[ss->subsys_id] = NULL;
4517 /* remove subsystem from rootnode's list of subsystems */
4518 list_del_init(&ss->sibling);
4521 * disentangle the css from all css_sets attached to the dummytop. as
4522 * in loading, we need to pay our respects to the hashtable gods.
4524 write_lock(&css_set_lock);
4525 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4526 struct css_set *cg = link->cg;
4529 hash_del(&cg->hlist);
4530 cg->subsys[ss->subsys_id] = NULL;
4531 key = css_set_hash(cg->subsys);
4532 hash_add(css_set_table, &cg->hlist, key);
4534 write_unlock(&css_set_lock);
4537 * remove subsystem's css from the dummytop and free it - need to
4538 * free before marking as null because ss->css_free needs the
4539 * cgrp->subsys pointer to find their state. note that this also
4540 * takes care of freeing the css_id.
4542 ss->css_free(dummytop);
4543 dummytop->subsys[ss->subsys_id] = NULL;
4545 mutex_unlock(&cgroup_mutex);
4547 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4550 * cgroup_init_early - cgroup initialization at system boot
4552 * Initialize cgroups at system boot, and initialize any
4553 * subsystems that request early init.
4555 int __init cgroup_init_early(void)
4558 atomic_set(&init_css_set.refcount, 1);
4559 INIT_LIST_HEAD(&init_css_set.cg_links);
4560 INIT_LIST_HEAD(&init_css_set.tasks);
4561 INIT_HLIST_NODE(&init_css_set.hlist);
4563 init_cgroup_root(&rootnode);
4565 init_task.cgroups = &init_css_set;
4567 init_css_set_link.cg = &init_css_set;
4568 init_css_set_link.cgrp = dummytop;
4569 list_add(&init_css_set_link.cgrp_link_list,
4570 &rootnode.top_cgroup.css_sets);
4571 list_add(&init_css_set_link.cg_link_list,
4572 &init_css_set.cg_links);
4574 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4575 struct cgroup_subsys *ss = subsys[i];
4577 /* at bootup time, we don't worry about modular subsystems */
4578 if (!ss || ss->module)
4582 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4583 BUG_ON(!ss->css_alloc);
4584 BUG_ON(!ss->css_free);
4585 if (ss->subsys_id != i) {
4586 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4587 ss->name, ss->subsys_id);
4592 cgroup_init_subsys(ss);
4598 * cgroup_init - cgroup initialization
4600 * Register cgroup filesystem and /proc file, and initialize
4601 * any subsystems that didn't request early init.
4603 int __init cgroup_init(void)
4609 err = bdi_init(&cgroup_backing_dev_info);
4613 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4614 struct cgroup_subsys *ss = subsys[i];
4616 /* at bootup time, we don't worry about modular subsystems */
4617 if (!ss || ss->module)
4619 if (!ss->early_init)
4620 cgroup_init_subsys(ss);
4622 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4625 /* Add init_css_set to the hash table */
4626 key = css_set_hash(init_css_set.subsys);
4627 hash_add(css_set_table, &init_css_set.hlist, key);
4628 BUG_ON(!init_root_id(&rootnode));
4630 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4636 err = register_filesystem(&cgroup_fs_type);
4638 kobject_put(cgroup_kobj);
4642 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4646 bdi_destroy(&cgroup_backing_dev_info);
4652 * proc_cgroup_show()
4653 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4654 * - Used for /proc/<pid>/cgroup.
4655 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4656 * doesn't really matter if tsk->cgroup changes after we read it,
4657 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4658 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4659 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4660 * cgroup to top_cgroup.
4663 /* TODO: Use a proper seq_file iterator */
4664 static int proc_cgroup_show(struct seq_file *m, void *v)
4667 struct task_struct *tsk;
4670 struct cgroupfs_root *root;
4673 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4679 tsk = get_pid_task(pid, PIDTYPE_PID);
4685 mutex_lock(&cgroup_mutex);
4687 for_each_active_root(root) {
4688 struct cgroup_subsys *ss;
4689 struct cgroup *cgrp;
4692 seq_printf(m, "%d:", root->hierarchy_id);
4693 for_each_subsys(root, ss)
4694 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4695 if (strlen(root->name))
4696 seq_printf(m, "%sname=%s", count ? "," : "",
4699 cgrp = task_cgroup_from_root(tsk, root);
4700 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4708 mutex_unlock(&cgroup_mutex);
4709 put_task_struct(tsk);
4716 static int cgroup_open(struct inode *inode, struct file *file)
4718 struct pid *pid = PROC_I(inode)->pid;
4719 return single_open(file, proc_cgroup_show, pid);
4722 const struct file_operations proc_cgroup_operations = {
4723 .open = cgroup_open,
4725 .llseek = seq_lseek,
4726 .release = single_release,
4729 /* Display information about each subsystem and each hierarchy */
4730 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4734 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4736 * ideally we don't want subsystems moving around while we do this.
4737 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4738 * subsys/hierarchy state.
4740 mutex_lock(&cgroup_mutex);
4741 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4742 struct cgroup_subsys *ss = subsys[i];
4745 seq_printf(m, "%s\t%d\t%d\t%d\n",
4746 ss->name, ss->root->hierarchy_id,
4747 ss->root->number_of_cgroups, !ss->disabled);
4749 mutex_unlock(&cgroup_mutex);
4753 static int cgroupstats_open(struct inode *inode, struct file *file)
4755 return single_open(file, proc_cgroupstats_show, NULL);
4758 static const struct file_operations proc_cgroupstats_operations = {
4759 .open = cgroupstats_open,
4761 .llseek = seq_lseek,
4762 .release = single_release,
4766 * cgroup_fork - attach newly forked task to its parents cgroup.
4767 * @child: pointer to task_struct of forking parent process.
4769 * Description: A task inherits its parent's cgroup at fork().
4771 * A pointer to the shared css_set was automatically copied in
4772 * fork.c by dup_task_struct(). However, we ignore that copy, since
4773 * it was not made under the protection of RCU or cgroup_mutex, so
4774 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4775 * have already changed current->cgroups, allowing the previously
4776 * referenced cgroup group to be removed and freed.
4778 * At the point that cgroup_fork() is called, 'current' is the parent
4779 * task, and the passed argument 'child' points to the child task.
4781 void cgroup_fork(struct task_struct *child)
4784 child->cgroups = current->cgroups;
4785 get_css_set(child->cgroups);
4786 task_unlock(current);
4787 INIT_LIST_HEAD(&child->cg_list);
4791 * cgroup_post_fork - called on a new task after adding it to the task list
4792 * @child: the task in question
4794 * Adds the task to the list running through its css_set if necessary and
4795 * call the subsystem fork() callbacks. Has to be after the task is
4796 * visible on the task list in case we race with the first call to
4797 * cgroup_iter_start() - to guarantee that the new task ends up on its
4800 void cgroup_post_fork(struct task_struct *child)
4805 * use_task_css_set_links is set to 1 before we walk the tasklist
4806 * under the tasklist_lock and we read it here after we added the child
4807 * to the tasklist under the tasklist_lock as well. If the child wasn't
4808 * yet in the tasklist when we walked through it from
4809 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4810 * should be visible now due to the paired locking and barriers implied
4811 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4812 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4815 if (use_task_css_set_links) {
4816 write_lock(&css_set_lock);
4818 if (list_empty(&child->cg_list))
4819 list_add(&child->cg_list, &child->cgroups->tasks);
4821 write_unlock(&css_set_lock);
4825 * Call ss->fork(). This must happen after @child is linked on
4826 * css_set; otherwise, @child might change state between ->fork()
4827 * and addition to css_set.
4829 if (need_forkexit_callback) {
4831 * fork/exit callbacks are supported only for builtin
4832 * subsystems, and the builtin section of the subsys
4833 * array is immutable, so we don't need to lock the
4834 * subsys array here. On the other hand, modular section
4835 * of the array can be freed at module unload, so we
4838 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4839 struct cgroup_subsys *ss = subsys[i];
4848 * cgroup_exit - detach cgroup from exiting task
4849 * @tsk: pointer to task_struct of exiting process
4850 * @run_callback: run exit callbacks?
4852 * Description: Detach cgroup from @tsk and release it.
4854 * Note that cgroups marked notify_on_release force every task in
4855 * them to take the global cgroup_mutex mutex when exiting.
4856 * This could impact scaling on very large systems. Be reluctant to
4857 * use notify_on_release cgroups where very high task exit scaling
4858 * is required on large systems.
4860 * the_top_cgroup_hack:
4862 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4864 * We call cgroup_exit() while the task is still competent to
4865 * handle notify_on_release(), then leave the task attached to the
4866 * root cgroup in each hierarchy for the remainder of its exit.
4868 * To do this properly, we would increment the reference count on
4869 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4870 * code we would add a second cgroup function call, to drop that
4871 * reference. This would just create an unnecessary hot spot on
4872 * the top_cgroup reference count, to no avail.
4874 * Normally, holding a reference to a cgroup without bumping its
4875 * count is unsafe. The cgroup could go away, or someone could
4876 * attach us to a different cgroup, decrementing the count on
4877 * the first cgroup that we never incremented. But in this case,
4878 * top_cgroup isn't going away, and either task has PF_EXITING set,
4879 * which wards off any cgroup_attach_task() attempts, or task is a failed
4880 * fork, never visible to cgroup_attach_task.
4882 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4888 * Unlink from the css_set task list if necessary.
4889 * Optimistically check cg_list before taking
4892 if (!list_empty(&tsk->cg_list)) {
4893 write_lock(&css_set_lock);
4894 if (!list_empty(&tsk->cg_list))
4895 list_del_init(&tsk->cg_list);
4896 write_unlock(&css_set_lock);
4899 /* Reassign the task to the init_css_set. */
4902 tsk->cgroups = &init_css_set;
4904 if (run_callbacks && need_forkexit_callback) {
4906 * fork/exit callbacks are supported only for builtin
4907 * subsystems, see cgroup_post_fork() for details.
4909 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4910 struct cgroup_subsys *ss = subsys[i];
4913 struct cgroup *old_cgrp =
4914 rcu_dereference_raw(cg->subsys[i])->cgroup;
4915 struct cgroup *cgrp = task_cgroup(tsk, i);
4916 ss->exit(cgrp, old_cgrp, tsk);
4922 put_css_set_taskexit(cg);
4925 static void check_for_release(struct cgroup *cgrp)
4927 /* All of these checks rely on RCU to keep the cgroup
4928 * structure alive */
4929 if (cgroup_is_releasable(cgrp) &&
4930 !atomic_read(&cgrp->count) && list_empty(&cgrp->children)) {
4932 * Control Group is currently removeable. If it's not
4933 * already queued for a userspace notification, queue
4936 int need_schedule_work = 0;
4938 raw_spin_lock(&release_list_lock);
4939 if (!cgroup_is_removed(cgrp) &&
4940 list_empty(&cgrp->release_list)) {
4941 list_add(&cgrp->release_list, &release_list);
4942 need_schedule_work = 1;
4944 raw_spin_unlock(&release_list_lock);
4945 if (need_schedule_work)
4946 schedule_work(&release_agent_work);
4950 /* Caller must verify that the css is not for root cgroup */
4951 bool __css_tryget(struct cgroup_subsys_state *css)
4956 v = css_refcnt(css);
4957 t = atomic_cmpxchg(&css->refcnt, v, v + 1);
4965 EXPORT_SYMBOL_GPL(__css_tryget);
4967 /* Caller must verify that the css is not for root cgroup */
4968 void __css_put(struct cgroup_subsys_state *css)
4972 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
4974 schedule_work(&css->dput_work);
4976 EXPORT_SYMBOL_GPL(__css_put);
4979 * Notify userspace when a cgroup is released, by running the
4980 * configured release agent with the name of the cgroup (path
4981 * relative to the root of cgroup file system) as the argument.
4983 * Most likely, this user command will try to rmdir this cgroup.
4985 * This races with the possibility that some other task will be
4986 * attached to this cgroup before it is removed, or that some other
4987 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4988 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4989 * unused, and this cgroup will be reprieved from its death sentence,
4990 * to continue to serve a useful existence. Next time it's released,
4991 * we will get notified again, if it still has 'notify_on_release' set.
4993 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4994 * means only wait until the task is successfully execve()'d. The
4995 * separate release agent task is forked by call_usermodehelper(),
4996 * then control in this thread returns here, without waiting for the
4997 * release agent task. We don't bother to wait because the caller of
4998 * this routine has no use for the exit status of the release agent
4999 * task, so no sense holding our caller up for that.
5001 static void cgroup_release_agent(struct work_struct *work)
5003 BUG_ON(work != &release_agent_work);
5004 mutex_lock(&cgroup_mutex);
5005 raw_spin_lock(&release_list_lock);
5006 while (!list_empty(&release_list)) {
5007 char *argv[3], *envp[3];
5009 char *pathbuf = NULL, *agentbuf = NULL;
5010 struct cgroup *cgrp = list_entry(release_list.next,
5013 list_del_init(&cgrp->release_list);
5014 raw_spin_unlock(&release_list_lock);
5015 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5018 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5020 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5025 argv[i++] = agentbuf;
5026 argv[i++] = pathbuf;
5030 /* minimal command environment */
5031 envp[i++] = "HOME=/";
5032 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5035 /* Drop the lock while we invoke the usermode helper,
5036 * since the exec could involve hitting disk and hence
5037 * be a slow process */
5038 mutex_unlock(&cgroup_mutex);
5039 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5040 mutex_lock(&cgroup_mutex);
5044 raw_spin_lock(&release_list_lock);
5046 raw_spin_unlock(&release_list_lock);
5047 mutex_unlock(&cgroup_mutex);
5050 static int __init cgroup_disable(char *str)
5055 while ((token = strsep(&str, ",")) != NULL) {
5058 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5059 struct cgroup_subsys *ss = subsys[i];
5062 * cgroup_disable, being at boot time, can't
5063 * know about module subsystems, so we don't
5066 if (!ss || ss->module)
5069 if (!strcmp(token, ss->name)) {
5071 printk(KERN_INFO "Disabling %s control group"
5072 " subsystem\n", ss->name);
5079 __setup("cgroup_disable=", cgroup_disable);
5082 * Functons for CSS ID.
5086 *To get ID other than 0, this should be called when !cgroup_is_removed().
5088 unsigned short css_id(struct cgroup_subsys_state *css)
5090 struct css_id *cssid;
5093 * This css_id() can return correct value when somone has refcnt
5094 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5095 * it's unchanged until freed.
5097 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5103 EXPORT_SYMBOL_GPL(css_id);
5105 unsigned short css_depth(struct cgroup_subsys_state *css)
5107 struct css_id *cssid;
5109 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5112 return cssid->depth;
5115 EXPORT_SYMBOL_GPL(css_depth);
5118 * css_is_ancestor - test "root" css is an ancestor of "child"
5119 * @child: the css to be tested.
5120 * @root: the css supporsed to be an ancestor of the child.
5122 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5123 * this function reads css->id, the caller must hold rcu_read_lock().
5124 * But, considering usual usage, the csses should be valid objects after test.
5125 * Assuming that the caller will do some action to the child if this returns
5126 * returns true, the caller must take "child";s reference count.
5127 * If "child" is valid object and this returns true, "root" is valid, too.
5130 bool css_is_ancestor(struct cgroup_subsys_state *child,
5131 const struct cgroup_subsys_state *root)
5133 struct css_id *child_id;
5134 struct css_id *root_id;
5136 child_id = rcu_dereference(child->id);
5139 root_id = rcu_dereference(root->id);
5142 if (child_id->depth < root_id->depth)
5144 if (child_id->stack[root_id->depth] != root_id->id)
5149 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5151 struct css_id *id = css->id;
5152 /* When this is called before css_id initialization, id can be NULL */
5156 BUG_ON(!ss->use_id);
5158 rcu_assign_pointer(id->css, NULL);
5159 rcu_assign_pointer(css->id, NULL);
5160 spin_lock(&ss->id_lock);
5161 idr_remove(&ss->idr, id->id);
5162 spin_unlock(&ss->id_lock);
5163 kfree_rcu(id, rcu_head);
5165 EXPORT_SYMBOL_GPL(free_css_id);
5168 * This is called by init or create(). Then, calls to this function are
5169 * always serialized (By cgroup_mutex() at create()).
5172 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5174 struct css_id *newid;
5177 BUG_ON(!ss->use_id);
5179 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5180 newid = kzalloc(size, GFP_KERNEL);
5182 return ERR_PTR(-ENOMEM);
5184 idr_preload(GFP_KERNEL);
5185 spin_lock(&ss->id_lock);
5186 /* Don't use 0. allocates an ID of 1-65535 */
5187 ret = idr_alloc(&ss->idr, newid, 1, CSS_ID_MAX + 1, GFP_NOWAIT);
5188 spin_unlock(&ss->id_lock);
5191 /* Returns error when there are no free spaces for new ID.*/
5196 newid->depth = depth;
5200 return ERR_PTR(ret);
5204 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5205 struct cgroup_subsys_state *rootcss)
5207 struct css_id *newid;
5209 spin_lock_init(&ss->id_lock);
5212 newid = get_new_cssid(ss, 0);
5214 return PTR_ERR(newid);
5216 newid->stack[0] = newid->id;
5217 newid->css = rootcss;
5218 rootcss->id = newid;
5222 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5223 struct cgroup *child)
5225 int subsys_id, i, depth = 0;
5226 struct cgroup_subsys_state *parent_css, *child_css;
5227 struct css_id *child_id, *parent_id;
5229 subsys_id = ss->subsys_id;
5230 parent_css = parent->subsys[subsys_id];
5231 child_css = child->subsys[subsys_id];
5232 parent_id = parent_css->id;
5233 depth = parent_id->depth + 1;
5235 child_id = get_new_cssid(ss, depth);
5236 if (IS_ERR(child_id))
5237 return PTR_ERR(child_id);
5239 for (i = 0; i < depth; i++)
5240 child_id->stack[i] = parent_id->stack[i];
5241 child_id->stack[depth] = child_id->id;
5243 * child_id->css pointer will be set after this cgroup is available
5244 * see cgroup_populate_dir()
5246 rcu_assign_pointer(child_css->id, child_id);
5252 * css_lookup - lookup css by id
5253 * @ss: cgroup subsys to be looked into.
5256 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5257 * NULL if not. Should be called under rcu_read_lock()
5259 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5261 struct css_id *cssid = NULL;
5263 BUG_ON(!ss->use_id);
5264 cssid = idr_find(&ss->idr, id);
5266 if (unlikely(!cssid))
5269 return rcu_dereference(cssid->css);
5271 EXPORT_SYMBOL_GPL(css_lookup);
5274 * css_get_next - lookup next cgroup under specified hierarchy.
5275 * @ss: pointer to subsystem
5276 * @id: current position of iteration.
5277 * @root: pointer to css. search tree under this.
5278 * @foundid: position of found object.
5280 * Search next css under the specified hierarchy of rootid. Calling under
5281 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5283 struct cgroup_subsys_state *
5284 css_get_next(struct cgroup_subsys *ss, int id,
5285 struct cgroup_subsys_state *root, int *foundid)
5287 struct cgroup_subsys_state *ret = NULL;
5290 int rootid = css_id(root);
5291 int depth = css_depth(root);
5296 BUG_ON(!ss->use_id);
5297 WARN_ON_ONCE(!rcu_read_lock_held());
5299 /* fill start point for scan */
5303 * scan next entry from bitmap(tree), tmpid is updated after
5306 tmp = idr_get_next(&ss->idr, &tmpid);
5309 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5310 ret = rcu_dereference(tmp->css);
5316 /* continue to scan from next id */
5323 * get corresponding css from file open on cgroupfs directory
5325 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5327 struct cgroup *cgrp;
5328 struct inode *inode;
5329 struct cgroup_subsys_state *css;
5331 inode = file_inode(f);
5332 /* check in cgroup filesystem dir */
5333 if (inode->i_op != &cgroup_dir_inode_operations)
5334 return ERR_PTR(-EBADF);
5336 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5337 return ERR_PTR(-EINVAL);
5340 cgrp = __d_cgrp(f->f_dentry);
5341 css = cgrp->subsys[id];
5342 return css ? css : ERR_PTR(-ENOENT);
5345 #ifdef CONFIG_CGROUP_DEBUG
5346 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5348 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5351 return ERR_PTR(-ENOMEM);
5356 static void debug_css_free(struct cgroup *cont)
5358 kfree(cont->subsys[debug_subsys_id]);
5361 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5363 return atomic_read(&cont->count);
5366 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5368 return cgroup_task_count(cont);
5371 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5373 return (u64)(unsigned long)current->cgroups;
5376 static u64 current_css_set_refcount_read(struct cgroup *cont,
5382 count = atomic_read(¤t->cgroups->refcount);
5387 static int current_css_set_cg_links_read(struct cgroup *cont,
5389 struct seq_file *seq)
5391 struct cg_cgroup_link *link;
5394 read_lock(&css_set_lock);
5396 cg = rcu_dereference(current->cgroups);
5397 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5398 struct cgroup *c = link->cgrp;
5402 name = c->dentry->d_name.name;
5405 seq_printf(seq, "Root %d group %s\n",
5406 c->root->hierarchy_id, name);
5409 read_unlock(&css_set_lock);
5413 #define MAX_TASKS_SHOWN_PER_CSS 25
5414 static int cgroup_css_links_read(struct cgroup *cont,
5416 struct seq_file *seq)
5418 struct cg_cgroup_link *link;
5420 read_lock(&css_set_lock);
5421 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5422 struct css_set *cg = link->cg;
5423 struct task_struct *task;
5425 seq_printf(seq, "css_set %p\n", cg);
5426 list_for_each_entry(task, &cg->tasks, cg_list) {
5427 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5428 seq_puts(seq, " ...\n");
5431 seq_printf(seq, " task %d\n",
5432 task_pid_vnr(task));
5436 read_unlock(&css_set_lock);
5440 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5442 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5445 static struct cftype debug_files[] = {
5447 .name = "cgroup_refcount",
5448 .read_u64 = cgroup_refcount_read,
5451 .name = "taskcount",
5452 .read_u64 = debug_taskcount_read,
5456 .name = "current_css_set",
5457 .read_u64 = current_css_set_read,
5461 .name = "current_css_set_refcount",
5462 .read_u64 = current_css_set_refcount_read,
5466 .name = "current_css_set_cg_links",
5467 .read_seq_string = current_css_set_cg_links_read,
5471 .name = "cgroup_css_links",
5472 .read_seq_string = cgroup_css_links_read,
5476 .name = "releasable",
5477 .read_u64 = releasable_read,
5483 struct cgroup_subsys debug_subsys = {
5485 .css_alloc = debug_css_alloc,
5486 .css_free = debug_css_free,
5487 .subsys_id = debug_subsys_id,
5488 .base_cftypes = debug_files,
5490 #endif /* CONFIG_CGROUP_DEBUG */