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/hash.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_proc */
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 static DEFINE_MUTEX(cgroup_mutex);
87 static DEFINE_MUTEX(cgroup_root_mutex);
90 * Generate an array of cgroup subsystem pointers. At boot time, this is
91 * populated with the built in subsystems, and modular subsystems are
92 * registered after that. The mutable section of this array is protected by
95 #define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
96 #define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
97 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
98 #include <linux/cgroup_subsys.h>
101 #define MAX_CGROUP_ROOT_NAMELEN 64
104 * A cgroupfs_root represents the root of a cgroup hierarchy,
105 * and may be associated with a superblock to form an active
108 struct cgroupfs_root {
109 struct super_block *sb;
112 * The bitmask of subsystems intended to be attached to this
115 unsigned long subsys_mask;
117 /* Unique id for this hierarchy. */
120 /* The bitmask of subsystems currently attached to this hierarchy */
121 unsigned long actual_subsys_mask;
123 /* A list running through the attached subsystems */
124 struct list_head subsys_list;
126 /* The root cgroup for this hierarchy */
127 struct cgroup top_cgroup;
129 /* Tracks how many cgroups are currently defined in hierarchy.*/
130 int number_of_cgroups;
132 /* A list running through the active hierarchies */
133 struct list_head root_list;
135 /* All cgroups on this root, cgroup_mutex protected */
136 struct list_head allcg_list;
138 /* Hierarchy-specific flags */
141 /* The path to use for release notifications. */
142 char release_agent_path[PATH_MAX];
144 /* The name for this hierarchy - may be empty */
145 char name[MAX_CGROUP_ROOT_NAMELEN];
149 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
150 * subsystems that are otherwise unattached - it never has more than a
151 * single cgroup, and all tasks are part of that cgroup.
153 static struct cgroupfs_root rootnode;
156 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
159 struct list_head node;
160 struct dentry *dentry;
165 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
166 * cgroup_subsys->use_id != 0.
168 #define CSS_ID_MAX (65535)
171 * The css to which this ID points. This pointer is set to valid value
172 * after cgroup is populated. If cgroup is removed, this will be NULL.
173 * This pointer is expected to be RCU-safe because destroy()
174 * is called after synchronize_rcu(). But for safe use, css_tryget()
175 * should be used for avoiding race.
177 struct cgroup_subsys_state __rcu *css;
183 * Depth in hierarchy which this ID belongs to.
185 unsigned short depth;
187 * ID is freed by RCU. (and lookup routine is RCU safe.)
189 struct rcu_head rcu_head;
191 * Hierarchy of CSS ID belongs to.
193 unsigned short stack[0]; /* Array of Length (depth+1) */
197 * cgroup_event represents events which userspace want to receive.
199 struct cgroup_event {
201 * Cgroup which the event belongs to.
205 * Control file which the event associated.
209 * eventfd to signal userspace about the event.
211 struct eventfd_ctx *eventfd;
213 * Each of these stored in a list by the cgroup.
215 struct list_head list;
217 * All fields below needed to unregister event when
218 * userspace closes eventfd.
221 wait_queue_head_t *wqh;
223 struct work_struct remove;
226 /* The list of hierarchy roots */
228 static LIST_HEAD(roots);
229 static int root_count;
231 static DEFINE_IDA(hierarchy_ida);
232 static int next_hierarchy_id;
233 static DEFINE_SPINLOCK(hierarchy_id_lock);
235 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
236 #define dummytop (&rootnode.top_cgroup)
238 /* This flag indicates whether tasks in the fork and exit paths should
239 * check for fork/exit handlers to call. This avoids us having to do
240 * extra work in the fork/exit path if none of the subsystems need to
243 static int need_forkexit_callback __read_mostly;
245 #ifdef CONFIG_PROVE_LOCKING
246 int cgroup_lock_is_held(void)
248 return lockdep_is_held(&cgroup_mutex);
250 #else /* #ifdef CONFIG_PROVE_LOCKING */
251 int cgroup_lock_is_held(void)
253 return mutex_is_locked(&cgroup_mutex);
255 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
257 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
259 static int css_unbias_refcnt(int refcnt)
261 return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
264 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
265 static int css_refcnt(struct cgroup_subsys_state *css)
267 int v = atomic_read(&css->refcnt);
269 return css_unbias_refcnt(v);
272 /* convenient tests for these bits */
273 inline int cgroup_is_removed(const struct cgroup *cgrp)
275 return test_bit(CGRP_REMOVED, &cgrp->flags);
278 /* bits in struct cgroupfs_root flags field */
280 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
281 ROOT_XATTR, /* supports extended attributes */
284 static int cgroup_is_releasable(const struct cgroup *cgrp)
287 (1 << CGRP_RELEASABLE) |
288 (1 << CGRP_NOTIFY_ON_RELEASE);
289 return (cgrp->flags & bits) == bits;
292 static int notify_on_release(const struct cgroup *cgrp)
294 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
297 static int clone_children(const struct cgroup *cgrp)
299 return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
303 * for_each_subsys() allows you to iterate on each subsystem attached to
304 * an active hierarchy
306 #define for_each_subsys(_root, _ss) \
307 list_for_each_entry(_ss, &_root->subsys_list, sibling)
309 /* for_each_active_root() allows you to iterate across the active hierarchies */
310 #define for_each_active_root(_root) \
311 list_for_each_entry(_root, &roots, root_list)
313 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
315 return dentry->d_fsdata;
318 static inline struct cfent *__d_cfe(struct dentry *dentry)
320 return dentry->d_fsdata;
323 static inline struct cftype *__d_cft(struct dentry *dentry)
325 return __d_cfe(dentry)->type;
328 /* the list of cgroups eligible for automatic release. Protected by
329 * release_list_lock */
330 static LIST_HEAD(release_list);
331 static DEFINE_RAW_SPINLOCK(release_list_lock);
332 static void cgroup_release_agent(struct work_struct *work);
333 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
334 static void check_for_release(struct cgroup *cgrp);
336 /* Link structure for associating css_set objects with cgroups */
337 struct cg_cgroup_link {
339 * List running through cg_cgroup_links associated with a
340 * cgroup, anchored on cgroup->css_sets
342 struct list_head cgrp_link_list;
345 * List running through cg_cgroup_links pointing at a
346 * single css_set object, anchored on css_set->cg_links
348 struct list_head cg_link_list;
352 /* The default css_set - used by init and its children prior to any
353 * hierarchies being mounted. It contains a pointer to the root state
354 * for each subsystem. Also used to anchor the list of css_sets. Not
355 * reference-counted, to improve performance when child cgroups
356 * haven't been created.
359 static struct css_set init_css_set;
360 static struct cg_cgroup_link init_css_set_link;
362 static int cgroup_init_idr(struct cgroup_subsys *ss,
363 struct cgroup_subsys_state *css);
365 /* css_set_lock protects the list of css_set objects, and the
366 * chain of tasks off each css_set. Nests outside task->alloc_lock
367 * due to cgroup_iter_start() */
368 static DEFINE_RWLOCK(css_set_lock);
369 static int css_set_count;
372 * hash table for cgroup groups. This improves the performance to find
373 * an existing css_set. This hash doesn't (currently) take into
374 * account cgroups in empty hierarchies.
376 #define CSS_SET_HASH_BITS 7
377 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
378 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
380 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
384 unsigned long tmp = 0UL;
386 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
387 tmp += (unsigned long)css[i];
388 tmp = (tmp >> 16) ^ tmp;
390 index = hash_long(tmp, CSS_SET_HASH_BITS);
392 return &css_set_table[index];
395 /* We don't maintain the lists running through each css_set to its
396 * task until after the first call to cgroup_iter_start(). This
397 * reduces the fork()/exit() overhead for people who have cgroups
398 * compiled into their kernel but not actually in use */
399 static int use_task_css_set_links __read_mostly;
401 static void __put_css_set(struct css_set *cg, int taskexit)
403 struct cg_cgroup_link *link;
404 struct cg_cgroup_link *saved_link;
406 * Ensure that the refcount doesn't hit zero while any readers
407 * can see it. Similar to atomic_dec_and_lock(), but for an
410 if (atomic_add_unless(&cg->refcount, -1, 1))
412 write_lock(&css_set_lock);
413 if (!atomic_dec_and_test(&cg->refcount)) {
414 write_unlock(&css_set_lock);
418 /* This css_set is dead. unlink it and release cgroup refcounts */
419 hlist_del(&cg->hlist);
422 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
424 struct cgroup *cgrp = link->cgrp;
425 list_del(&link->cg_link_list);
426 list_del(&link->cgrp_link_list);
427 if (atomic_dec_and_test(&cgrp->count) &&
428 notify_on_release(cgrp)) {
430 set_bit(CGRP_RELEASABLE, &cgrp->flags);
431 check_for_release(cgrp);
437 write_unlock(&css_set_lock);
438 kfree_rcu(cg, rcu_head);
442 * refcounted get/put for css_set objects
444 static inline void get_css_set(struct css_set *cg)
446 atomic_inc(&cg->refcount);
449 static inline void put_css_set(struct css_set *cg)
451 __put_css_set(cg, 0);
454 static inline void put_css_set_taskexit(struct css_set *cg)
456 __put_css_set(cg, 1);
460 * compare_css_sets - helper function for find_existing_css_set().
461 * @cg: candidate css_set being tested
462 * @old_cg: existing css_set for a task
463 * @new_cgrp: cgroup that's being entered by the task
464 * @template: desired set of css pointers in css_set (pre-calculated)
466 * Returns true if "cg" matches "old_cg" except for the hierarchy
467 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
469 static bool compare_css_sets(struct css_set *cg,
470 struct css_set *old_cg,
471 struct cgroup *new_cgrp,
472 struct cgroup_subsys_state *template[])
474 struct list_head *l1, *l2;
476 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
477 /* Not all subsystems matched */
482 * Compare cgroup pointers in order to distinguish between
483 * different cgroups in heirarchies with no subsystems. We
484 * could get by with just this check alone (and skip the
485 * memcmp above) but on most setups the memcmp check will
486 * avoid the need for this more expensive check on almost all
491 l2 = &old_cg->cg_links;
493 struct cg_cgroup_link *cgl1, *cgl2;
494 struct cgroup *cg1, *cg2;
498 /* See if we reached the end - both lists are equal length. */
499 if (l1 == &cg->cg_links) {
500 BUG_ON(l2 != &old_cg->cg_links);
503 BUG_ON(l2 == &old_cg->cg_links);
505 /* Locate the cgroups associated with these links. */
506 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
507 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
510 /* Hierarchies should be linked in the same order. */
511 BUG_ON(cg1->root != cg2->root);
514 * If this hierarchy is the hierarchy of the cgroup
515 * that's changing, then we need to check that this
516 * css_set points to the new cgroup; if it's any other
517 * hierarchy, then this css_set should point to the
518 * same cgroup as the old css_set.
520 if (cg1->root == new_cgrp->root) {
532 * find_existing_css_set() is a helper for
533 * find_css_set(), and checks to see whether an existing
534 * css_set is suitable.
536 * oldcg: the cgroup group that we're using before the cgroup
539 * cgrp: the cgroup that we're moving into
541 * template: location in which to build the desired set of subsystem
542 * state objects for the new cgroup group
544 static struct css_set *find_existing_css_set(
545 struct css_set *oldcg,
547 struct cgroup_subsys_state *template[])
550 struct cgroupfs_root *root = cgrp->root;
551 struct hlist_head *hhead;
552 struct hlist_node *node;
556 * Build the set of subsystem state objects that we want to see in the
557 * new css_set. while subsystems can change globally, the entries here
558 * won't change, so no need for locking.
560 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
561 if (root->subsys_mask & (1UL << i)) {
562 /* Subsystem is in this hierarchy. So we want
563 * the subsystem state from the new
565 template[i] = cgrp->subsys[i];
567 /* Subsystem is not in this hierarchy, so we
568 * don't want to change the subsystem state */
569 template[i] = oldcg->subsys[i];
573 hhead = css_set_hash(template);
574 hlist_for_each_entry(cg, node, hhead, hlist) {
575 if (!compare_css_sets(cg, oldcg, cgrp, template))
578 /* This css_set matches what we need */
582 /* No existing cgroup group matched */
586 static void free_cg_links(struct list_head *tmp)
588 struct cg_cgroup_link *link;
589 struct cg_cgroup_link *saved_link;
591 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
592 list_del(&link->cgrp_link_list);
598 * allocate_cg_links() allocates "count" cg_cgroup_link structures
599 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
600 * success or a negative error
602 static int allocate_cg_links(int count, struct list_head *tmp)
604 struct cg_cgroup_link *link;
607 for (i = 0; i < count; i++) {
608 link = kmalloc(sizeof(*link), GFP_KERNEL);
613 list_add(&link->cgrp_link_list, tmp);
619 * link_css_set - a helper function to link a css_set to a cgroup
620 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
621 * @cg: the css_set to be linked
622 * @cgrp: the destination cgroup
624 static void link_css_set(struct list_head *tmp_cg_links,
625 struct css_set *cg, struct cgroup *cgrp)
627 struct cg_cgroup_link *link;
629 BUG_ON(list_empty(tmp_cg_links));
630 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
634 atomic_inc(&cgrp->count);
635 list_move(&link->cgrp_link_list, &cgrp->css_sets);
637 * Always add links to the tail of the list so that the list
638 * is sorted by order of hierarchy creation
640 list_add_tail(&link->cg_link_list, &cg->cg_links);
644 * find_css_set() takes an existing cgroup group and a
645 * cgroup object, and returns a css_set object that's
646 * equivalent to the old group, but with the given cgroup
647 * substituted into the appropriate hierarchy. Must be called with
650 static struct css_set *find_css_set(
651 struct css_set *oldcg, struct cgroup *cgrp)
654 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
656 struct list_head tmp_cg_links;
658 struct hlist_head *hhead;
659 struct cg_cgroup_link *link;
661 /* First see if we already have a cgroup group that matches
663 read_lock(&css_set_lock);
664 res = find_existing_css_set(oldcg, cgrp, template);
667 read_unlock(&css_set_lock);
672 res = kmalloc(sizeof(*res), GFP_KERNEL);
676 /* Allocate all the cg_cgroup_link objects that we'll need */
677 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
682 atomic_set(&res->refcount, 1);
683 INIT_LIST_HEAD(&res->cg_links);
684 INIT_LIST_HEAD(&res->tasks);
685 INIT_HLIST_NODE(&res->hlist);
687 /* Copy the set of subsystem state objects generated in
688 * find_existing_css_set() */
689 memcpy(res->subsys, template, sizeof(res->subsys));
691 write_lock(&css_set_lock);
692 /* Add reference counts and links from the new css_set. */
693 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
694 struct cgroup *c = link->cgrp;
695 if (c->root == cgrp->root)
697 link_css_set(&tmp_cg_links, res, c);
700 BUG_ON(!list_empty(&tmp_cg_links));
704 /* Add this cgroup group to the hash table */
705 hhead = css_set_hash(res->subsys);
706 hlist_add_head(&res->hlist, hhead);
708 write_unlock(&css_set_lock);
714 * Return the cgroup for "task" from the given hierarchy. Must be
715 * called with cgroup_mutex held.
717 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
718 struct cgroupfs_root *root)
721 struct cgroup *res = NULL;
723 BUG_ON(!mutex_is_locked(&cgroup_mutex));
724 read_lock(&css_set_lock);
726 * No need to lock the task - since we hold cgroup_mutex the
727 * task can't change groups, so the only thing that can happen
728 * is that it exits and its css is set back to init_css_set.
731 if (css == &init_css_set) {
732 res = &root->top_cgroup;
734 struct cg_cgroup_link *link;
735 list_for_each_entry(link, &css->cg_links, cg_link_list) {
736 struct cgroup *c = link->cgrp;
737 if (c->root == root) {
743 read_unlock(&css_set_lock);
749 * There is one global cgroup mutex. We also require taking
750 * task_lock() when dereferencing a task's cgroup subsys pointers.
751 * See "The task_lock() exception", at the end of this comment.
753 * A task must hold cgroup_mutex to modify cgroups.
755 * Any task can increment and decrement the count field without lock.
756 * So in general, code holding cgroup_mutex can't rely on the count
757 * field not changing. However, if the count goes to zero, then only
758 * cgroup_attach_task() can increment it again. Because a count of zero
759 * means that no tasks are currently attached, therefore there is no
760 * way a task attached to that cgroup can fork (the other way to
761 * increment the count). So code holding cgroup_mutex can safely
762 * assume that if the count is zero, it will stay zero. Similarly, if
763 * a task holds cgroup_mutex on a cgroup with zero count, it
764 * knows that the cgroup won't be removed, as cgroup_rmdir()
767 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
768 * (usually) take cgroup_mutex. These are the two most performance
769 * critical pieces of code here. The exception occurs on cgroup_exit(),
770 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
771 * is taken, and if the cgroup count is zero, a usermode call made
772 * to the release agent with the name of the cgroup (path relative to
773 * the root of cgroup file system) as the argument.
775 * A cgroup can only be deleted if both its 'count' of using tasks
776 * is zero, and its list of 'children' cgroups is empty. Since all
777 * tasks in the system use _some_ cgroup, and since there is always at
778 * least one task in the system (init, pid == 1), therefore, top_cgroup
779 * always has either children cgroups and/or using tasks. So we don't
780 * need a special hack to ensure that top_cgroup cannot be deleted.
782 * The task_lock() exception
784 * The need for this exception arises from the action of
785 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
786 * another. It does so using cgroup_mutex, however there are
787 * several performance critical places that need to reference
788 * task->cgroup without the expense of grabbing a system global
789 * mutex. Therefore except as noted below, when dereferencing or, as
790 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
791 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
792 * the task_struct routinely used for such matters.
794 * P.S. One more locking exception. RCU is used to guard the
795 * update of a tasks cgroup pointer by cgroup_attach_task()
799 * cgroup_lock - lock out any changes to cgroup structures
802 void cgroup_lock(void)
804 mutex_lock(&cgroup_mutex);
806 EXPORT_SYMBOL_GPL(cgroup_lock);
809 * cgroup_unlock - release lock on cgroup changes
811 * Undo the lock taken in a previous cgroup_lock() call.
813 void cgroup_unlock(void)
815 mutex_unlock(&cgroup_mutex);
817 EXPORT_SYMBOL_GPL(cgroup_unlock);
820 * A couple of forward declarations required, due to cyclic reference loop:
821 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
822 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
826 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
827 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
828 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
829 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
830 unsigned long subsys_mask);
831 static const struct inode_operations cgroup_dir_inode_operations;
832 static const struct file_operations proc_cgroupstats_operations;
834 static struct backing_dev_info cgroup_backing_dev_info = {
836 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
839 static int alloc_css_id(struct cgroup_subsys *ss,
840 struct cgroup *parent, struct cgroup *child);
842 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
844 struct inode *inode = new_inode(sb);
847 inode->i_ino = get_next_ino();
848 inode->i_mode = mode;
849 inode->i_uid = current_fsuid();
850 inode->i_gid = current_fsgid();
851 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
852 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
857 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
859 /* is dentry a directory ? if so, kfree() associated cgroup */
860 if (S_ISDIR(inode->i_mode)) {
861 struct cgroup *cgrp = dentry->d_fsdata;
862 struct cgroup_subsys *ss;
863 BUG_ON(!(cgroup_is_removed(cgrp)));
864 /* It's possible for external users to be holding css
865 * reference counts on a cgroup; css_put() needs to
866 * be able to access the cgroup after decrementing
867 * the reference count in order to know if it needs to
868 * queue the cgroup to be handled by the release
872 mutex_lock(&cgroup_mutex);
874 * Release the subsystem state objects.
876 for_each_subsys(cgrp->root, ss)
879 cgrp->root->number_of_cgroups--;
880 mutex_unlock(&cgroup_mutex);
883 * Drop the active superblock reference that we took when we
886 deactivate_super(cgrp->root->sb);
889 * if we're getting rid of the cgroup, refcount should ensure
890 * that there are no pidlists left.
892 BUG_ON(!list_empty(&cgrp->pidlists));
894 simple_xattrs_free(&cgrp->xattrs);
896 kfree_rcu(cgrp, rcu_head);
898 struct cfent *cfe = __d_cfe(dentry);
899 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
900 struct cftype *cft = cfe->type;
902 WARN_ONCE(!list_empty(&cfe->node) &&
903 cgrp != &cgrp->root->top_cgroup,
904 "cfe still linked for %s\n", cfe->type->name);
906 simple_xattrs_free(&cft->xattrs);
911 static int cgroup_delete(const struct dentry *d)
916 static void remove_dir(struct dentry *d)
918 struct dentry *parent = dget(d->d_parent);
921 simple_rmdir(parent->d_inode, d);
925 static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
929 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
930 lockdep_assert_held(&cgroup_mutex);
932 list_for_each_entry(cfe, &cgrp->files, node) {
933 struct dentry *d = cfe->dentry;
935 if (cft && cfe->type != cft)
940 simple_unlink(cgrp->dentry->d_inode, d);
941 list_del_init(&cfe->node);
950 * cgroup_clear_directory - selective removal of base and subsystem files
951 * @dir: directory containing the files
952 * @base_files: true if the base files should be removed
953 * @subsys_mask: mask of the subsystem ids whose files should be removed
955 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
956 unsigned long subsys_mask)
958 struct cgroup *cgrp = __d_cgrp(dir);
959 struct cgroup_subsys *ss;
961 for_each_subsys(cgrp->root, ss) {
962 struct cftype_set *set;
963 if (!test_bit(ss->subsys_id, &subsys_mask))
965 list_for_each_entry(set, &ss->cftsets, node)
966 cgroup_rm_file(cgrp, set->cfts);
969 while (!list_empty(&cgrp->files))
970 cgroup_rm_file(cgrp, NULL);
975 * NOTE : the dentry must have been dget()'ed
977 static void cgroup_d_remove_dir(struct dentry *dentry)
979 struct dentry *parent;
980 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
982 cgroup_clear_directory(dentry, true, root->subsys_mask);
984 parent = dentry->d_parent;
985 spin_lock(&parent->d_lock);
986 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
987 list_del_init(&dentry->d_u.d_child);
988 spin_unlock(&dentry->d_lock);
989 spin_unlock(&parent->d_lock);
994 * Call with cgroup_mutex held. Drops reference counts on modules, including
995 * any duplicate ones that parse_cgroupfs_options took. If this function
996 * returns an error, no reference counts are touched.
998 static int rebind_subsystems(struct cgroupfs_root *root,
999 unsigned long final_subsys_mask)
1001 unsigned long added_mask, removed_mask;
1002 struct cgroup *cgrp = &root->top_cgroup;
1005 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1006 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1008 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
1009 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1010 /* Check that any added subsystems are currently free */
1011 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1012 unsigned long bit = 1UL << i;
1013 struct cgroup_subsys *ss = subsys[i];
1014 if (!(bit & added_mask))
1017 * Nobody should tell us to do a subsys that doesn't exist:
1018 * parse_cgroupfs_options should catch that case and refcounts
1019 * ensure that subsystems won't disappear once selected.
1022 if (ss->root != &rootnode) {
1023 /* Subsystem isn't free */
1028 /* Currently we don't handle adding/removing subsystems when
1029 * any child cgroups exist. This is theoretically supportable
1030 * but involves complex error handling, so it's being left until
1032 if (root->number_of_cgroups > 1)
1035 /* Process each subsystem */
1036 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1037 struct cgroup_subsys *ss = subsys[i];
1038 unsigned long bit = 1UL << i;
1039 if (bit & added_mask) {
1040 /* We're binding this subsystem to this hierarchy */
1042 BUG_ON(cgrp->subsys[i]);
1043 BUG_ON(!dummytop->subsys[i]);
1044 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1045 cgrp->subsys[i] = dummytop->subsys[i];
1046 cgrp->subsys[i]->cgroup = cgrp;
1047 list_move(&ss->sibling, &root->subsys_list);
1051 /* refcount was already taken, and we're keeping it */
1052 } else if (bit & removed_mask) {
1053 /* We're removing this subsystem */
1055 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1056 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1059 dummytop->subsys[i]->cgroup = dummytop;
1060 cgrp->subsys[i] = NULL;
1061 subsys[i]->root = &rootnode;
1062 list_move(&ss->sibling, &rootnode.subsys_list);
1063 /* subsystem is now free - drop reference on module */
1064 module_put(ss->module);
1065 } else if (bit & final_subsys_mask) {
1066 /* Subsystem state should already exist */
1068 BUG_ON(!cgrp->subsys[i]);
1070 * a refcount was taken, but we already had one, so
1071 * drop the extra reference.
1073 module_put(ss->module);
1074 #ifdef CONFIG_MODULE_UNLOAD
1075 BUG_ON(ss->module && !module_refcount(ss->module));
1078 /* Subsystem state shouldn't exist */
1079 BUG_ON(cgrp->subsys[i]);
1082 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1088 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1090 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1091 struct cgroup_subsys *ss;
1093 mutex_lock(&cgroup_root_mutex);
1094 for_each_subsys(root, ss)
1095 seq_printf(seq, ",%s", ss->name);
1096 if (test_bit(ROOT_NOPREFIX, &root->flags))
1097 seq_puts(seq, ",noprefix");
1098 if (test_bit(ROOT_XATTR, &root->flags))
1099 seq_puts(seq, ",xattr");
1100 if (strlen(root->release_agent_path))
1101 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1102 if (clone_children(&root->top_cgroup))
1103 seq_puts(seq, ",clone_children");
1104 if (strlen(root->name))
1105 seq_printf(seq, ",name=%s", root->name);
1106 mutex_unlock(&cgroup_root_mutex);
1110 struct cgroup_sb_opts {
1111 unsigned long subsys_mask;
1112 unsigned long flags;
1113 char *release_agent;
1114 bool clone_children;
1116 /* User explicitly requested empty subsystem */
1119 struct cgroupfs_root *new_root;
1124 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1125 * with cgroup_mutex held to protect the subsys[] array. This function takes
1126 * refcounts on subsystems to be used, unless it returns error, in which case
1127 * no refcounts are taken.
1129 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1131 char *token, *o = data;
1132 bool all_ss = false, one_ss = false;
1133 unsigned long mask = (unsigned long)-1;
1135 bool module_pin_failed = false;
1137 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1139 #ifdef CONFIG_CPUSETS
1140 mask = ~(1UL << cpuset_subsys_id);
1143 memset(opts, 0, sizeof(*opts));
1145 while ((token = strsep(&o, ",")) != NULL) {
1148 if (!strcmp(token, "none")) {
1149 /* Explicitly have no subsystems */
1153 if (!strcmp(token, "all")) {
1154 /* Mutually exclusive option 'all' + subsystem name */
1160 if (!strcmp(token, "noprefix")) {
1161 set_bit(ROOT_NOPREFIX, &opts->flags);
1164 if (!strcmp(token, "clone_children")) {
1165 opts->clone_children = true;
1168 if (!strcmp(token, "xattr")) {
1169 set_bit(ROOT_XATTR, &opts->flags);
1172 if (!strncmp(token, "release_agent=", 14)) {
1173 /* Specifying two release agents is forbidden */
1174 if (opts->release_agent)
1176 opts->release_agent =
1177 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1178 if (!opts->release_agent)
1182 if (!strncmp(token, "name=", 5)) {
1183 const char *name = token + 5;
1184 /* Can't specify an empty name */
1187 /* Must match [\w.-]+ */
1188 for (i = 0; i < strlen(name); i++) {
1192 if ((c == '.') || (c == '-') || (c == '_'))
1196 /* Specifying two names is forbidden */
1199 opts->name = kstrndup(name,
1200 MAX_CGROUP_ROOT_NAMELEN - 1,
1208 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1209 struct cgroup_subsys *ss = subsys[i];
1212 if (strcmp(token, ss->name))
1217 /* Mutually exclusive option 'all' + subsystem name */
1220 set_bit(i, &opts->subsys_mask);
1225 if (i == CGROUP_SUBSYS_COUNT)
1230 * If the 'all' option was specified select all the subsystems,
1231 * otherwise if 'none', 'name=' and a subsystem name options
1232 * were not specified, let's default to 'all'
1234 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1235 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1236 struct cgroup_subsys *ss = subsys[i];
1241 set_bit(i, &opts->subsys_mask);
1245 /* Consistency checks */
1248 * Option noprefix was introduced just for backward compatibility
1249 * with the old cpuset, so we allow noprefix only if mounting just
1250 * the cpuset subsystem.
1252 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1253 (opts->subsys_mask & mask))
1257 /* Can't specify "none" and some subsystems */
1258 if (opts->subsys_mask && opts->none)
1262 * We either have to specify by name or by subsystems. (So all
1263 * empty hierarchies must have a name).
1265 if (!opts->subsys_mask && !opts->name)
1269 * Grab references on all the modules we'll need, so the subsystems
1270 * don't dance around before rebind_subsystems attaches them. This may
1271 * take duplicate reference counts on a subsystem that's already used,
1272 * but rebind_subsystems handles this case.
1274 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1275 unsigned long bit = 1UL << i;
1277 if (!(bit & opts->subsys_mask))
1279 if (!try_module_get(subsys[i]->module)) {
1280 module_pin_failed = true;
1284 if (module_pin_failed) {
1286 * oops, one of the modules was going away. this means that we
1287 * raced with a module_delete call, and to the user this is
1288 * essentially a "subsystem doesn't exist" case.
1290 for (i--; i >= 0; i--) {
1291 /* drop refcounts only on the ones we took */
1292 unsigned long bit = 1UL << i;
1294 if (!(bit & opts->subsys_mask))
1296 module_put(subsys[i]->module);
1304 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1307 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1308 unsigned long bit = 1UL << i;
1310 if (!(bit & subsys_mask))
1312 module_put(subsys[i]->module);
1316 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1319 struct cgroupfs_root *root = sb->s_fs_info;
1320 struct cgroup *cgrp = &root->top_cgroup;
1321 struct cgroup_sb_opts opts;
1322 unsigned long added_mask, removed_mask;
1324 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1325 mutex_lock(&cgroup_mutex);
1326 mutex_lock(&cgroup_root_mutex);
1328 /* See what subsystems are wanted */
1329 ret = parse_cgroupfs_options(data, &opts);
1333 /* See feature-removal-schedule.txt */
1334 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1335 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1336 task_tgid_nr(current), current->comm);
1338 added_mask = opts.subsys_mask & ~root->subsys_mask;
1339 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1341 /* Don't allow flags or name to change at remount */
1342 if (opts.flags != root->flags ||
1343 (opts.name && strcmp(opts.name, root->name))) {
1345 drop_parsed_module_refcounts(opts.subsys_mask);
1349 ret = rebind_subsystems(root, opts.subsys_mask);
1351 drop_parsed_module_refcounts(opts.subsys_mask);
1355 /* clear out any existing files and repopulate subsystem files */
1356 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1357 /* re-populate subsystem files */
1358 cgroup_populate_dir(cgrp, false, added_mask);
1360 if (opts.release_agent)
1361 strcpy(root->release_agent_path, opts.release_agent);
1363 kfree(opts.release_agent);
1365 mutex_unlock(&cgroup_root_mutex);
1366 mutex_unlock(&cgroup_mutex);
1367 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1371 static const struct super_operations cgroup_ops = {
1372 .statfs = simple_statfs,
1373 .drop_inode = generic_delete_inode,
1374 .show_options = cgroup_show_options,
1375 .remount_fs = cgroup_remount,
1378 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1380 INIT_LIST_HEAD(&cgrp->sibling);
1381 INIT_LIST_HEAD(&cgrp->children);
1382 INIT_LIST_HEAD(&cgrp->files);
1383 INIT_LIST_HEAD(&cgrp->css_sets);
1384 INIT_LIST_HEAD(&cgrp->allcg_node);
1385 INIT_LIST_HEAD(&cgrp->release_list);
1386 INIT_LIST_HEAD(&cgrp->pidlists);
1387 mutex_init(&cgrp->pidlist_mutex);
1388 INIT_LIST_HEAD(&cgrp->event_list);
1389 spin_lock_init(&cgrp->event_list_lock);
1390 simple_xattrs_init(&cgrp->xattrs);
1393 static void init_cgroup_root(struct cgroupfs_root *root)
1395 struct cgroup *cgrp = &root->top_cgroup;
1397 INIT_LIST_HEAD(&root->subsys_list);
1398 INIT_LIST_HEAD(&root->root_list);
1399 INIT_LIST_HEAD(&root->allcg_list);
1400 root->number_of_cgroups = 1;
1402 cgrp->top_cgroup = cgrp;
1403 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1404 init_cgroup_housekeeping(cgrp);
1407 static bool init_root_id(struct cgroupfs_root *root)
1412 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1414 spin_lock(&hierarchy_id_lock);
1415 /* Try to allocate the next unused ID */
1416 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1417 &root->hierarchy_id);
1419 /* Try again starting from 0 */
1420 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1422 next_hierarchy_id = root->hierarchy_id + 1;
1423 } else if (ret != -EAGAIN) {
1424 /* Can only get here if the 31-bit IDR is full ... */
1427 spin_unlock(&hierarchy_id_lock);
1432 static int cgroup_test_super(struct super_block *sb, void *data)
1434 struct cgroup_sb_opts *opts = data;
1435 struct cgroupfs_root *root = sb->s_fs_info;
1437 /* If we asked for a name then it must match */
1438 if (opts->name && strcmp(opts->name, root->name))
1442 * If we asked for subsystems (or explicitly for no
1443 * subsystems) then they must match
1445 if ((opts->subsys_mask || opts->none)
1446 && (opts->subsys_mask != root->subsys_mask))
1452 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1454 struct cgroupfs_root *root;
1456 if (!opts->subsys_mask && !opts->none)
1459 root = kzalloc(sizeof(*root), GFP_KERNEL);
1461 return ERR_PTR(-ENOMEM);
1463 if (!init_root_id(root)) {
1465 return ERR_PTR(-ENOMEM);
1467 init_cgroup_root(root);
1469 root->subsys_mask = opts->subsys_mask;
1470 root->flags = opts->flags;
1471 if (opts->release_agent)
1472 strcpy(root->release_agent_path, opts->release_agent);
1474 strcpy(root->name, opts->name);
1475 if (opts->clone_children)
1476 set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1480 static void cgroup_drop_root(struct cgroupfs_root *root)
1485 BUG_ON(!root->hierarchy_id);
1486 spin_lock(&hierarchy_id_lock);
1487 ida_remove(&hierarchy_ida, root->hierarchy_id);
1488 spin_unlock(&hierarchy_id_lock);
1492 static int cgroup_set_super(struct super_block *sb, void *data)
1495 struct cgroup_sb_opts *opts = data;
1497 /* If we don't have a new root, we can't set up a new sb */
1498 if (!opts->new_root)
1501 BUG_ON(!opts->subsys_mask && !opts->none);
1503 ret = set_anon_super(sb, NULL);
1507 sb->s_fs_info = opts->new_root;
1508 opts->new_root->sb = sb;
1510 sb->s_blocksize = PAGE_CACHE_SIZE;
1511 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1512 sb->s_magic = CGROUP_SUPER_MAGIC;
1513 sb->s_op = &cgroup_ops;
1518 static int cgroup_get_rootdir(struct super_block *sb)
1520 static const struct dentry_operations cgroup_dops = {
1521 .d_iput = cgroup_diput,
1522 .d_delete = cgroup_delete,
1525 struct inode *inode =
1526 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1531 inode->i_fop = &simple_dir_operations;
1532 inode->i_op = &cgroup_dir_inode_operations;
1533 /* directories start off with i_nlink == 2 (for "." entry) */
1535 sb->s_root = d_make_root(inode);
1538 /* for everything else we want ->d_op set */
1539 sb->s_d_op = &cgroup_dops;
1543 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1544 int flags, const char *unused_dev_name,
1547 struct cgroup_sb_opts opts;
1548 struct cgroupfs_root *root;
1550 struct super_block *sb;
1551 struct cgroupfs_root *new_root;
1552 struct inode *inode;
1554 /* First find the desired set of subsystems */
1555 mutex_lock(&cgroup_mutex);
1556 ret = parse_cgroupfs_options(data, &opts);
1557 mutex_unlock(&cgroup_mutex);
1562 * Allocate a new cgroup root. We may not need it if we're
1563 * reusing an existing hierarchy.
1565 new_root = cgroup_root_from_opts(&opts);
1566 if (IS_ERR(new_root)) {
1567 ret = PTR_ERR(new_root);
1570 opts.new_root = new_root;
1572 /* Locate an existing or new sb for this hierarchy */
1573 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1576 cgroup_drop_root(opts.new_root);
1580 root = sb->s_fs_info;
1582 if (root == opts.new_root) {
1583 /* We used the new root structure, so this is a new hierarchy */
1584 struct list_head tmp_cg_links;
1585 struct cgroup *root_cgrp = &root->top_cgroup;
1586 struct cgroupfs_root *existing_root;
1587 const struct cred *cred;
1590 BUG_ON(sb->s_root != NULL);
1592 ret = cgroup_get_rootdir(sb);
1594 goto drop_new_super;
1595 inode = sb->s_root->d_inode;
1597 mutex_lock(&inode->i_mutex);
1598 mutex_lock(&cgroup_mutex);
1599 mutex_lock(&cgroup_root_mutex);
1601 /* Check for name clashes with existing mounts */
1603 if (strlen(root->name))
1604 for_each_active_root(existing_root)
1605 if (!strcmp(existing_root->name, root->name))
1609 * We're accessing css_set_count without locking
1610 * css_set_lock here, but that's OK - it can only be
1611 * increased by someone holding cgroup_lock, and
1612 * that's us. The worst that can happen is that we
1613 * have some link structures left over
1615 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1619 ret = rebind_subsystems(root, root->subsys_mask);
1620 if (ret == -EBUSY) {
1621 free_cg_links(&tmp_cg_links);
1625 * There must be no failure case after here, since rebinding
1626 * takes care of subsystems' refcounts, which are explicitly
1627 * dropped in the failure exit path.
1630 /* EBUSY should be the only error here */
1633 list_add(&root->root_list, &roots);
1636 sb->s_root->d_fsdata = root_cgrp;
1637 root->top_cgroup.dentry = sb->s_root;
1639 /* Link the top cgroup in this hierarchy into all
1640 * the css_set objects */
1641 write_lock(&css_set_lock);
1642 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1643 struct hlist_head *hhead = &css_set_table[i];
1644 struct hlist_node *node;
1647 hlist_for_each_entry(cg, node, hhead, hlist)
1648 link_css_set(&tmp_cg_links, cg, root_cgrp);
1650 write_unlock(&css_set_lock);
1652 free_cg_links(&tmp_cg_links);
1654 BUG_ON(!list_empty(&root_cgrp->children));
1655 BUG_ON(root->number_of_cgroups != 1);
1657 cred = override_creds(&init_cred);
1658 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1660 mutex_unlock(&cgroup_root_mutex);
1661 mutex_unlock(&cgroup_mutex);
1662 mutex_unlock(&inode->i_mutex);
1665 * We re-used an existing hierarchy - the new root (if
1666 * any) is not needed
1668 cgroup_drop_root(opts.new_root);
1669 /* no subsys rebinding, so refcounts don't change */
1670 drop_parsed_module_refcounts(opts.subsys_mask);
1673 kfree(opts.release_agent);
1675 return dget(sb->s_root);
1678 mutex_unlock(&cgroup_root_mutex);
1679 mutex_unlock(&cgroup_mutex);
1680 mutex_unlock(&inode->i_mutex);
1682 deactivate_locked_super(sb);
1684 drop_parsed_module_refcounts(opts.subsys_mask);
1686 kfree(opts.release_agent);
1688 return ERR_PTR(ret);
1691 static void cgroup_kill_sb(struct super_block *sb) {
1692 struct cgroupfs_root *root = sb->s_fs_info;
1693 struct cgroup *cgrp = &root->top_cgroup;
1695 struct cg_cgroup_link *link;
1696 struct cg_cgroup_link *saved_link;
1700 BUG_ON(root->number_of_cgroups != 1);
1701 BUG_ON(!list_empty(&cgrp->children));
1703 mutex_lock(&cgroup_mutex);
1704 mutex_lock(&cgroup_root_mutex);
1706 /* Rebind all subsystems back to the default hierarchy */
1707 ret = rebind_subsystems(root, 0);
1708 /* Shouldn't be able to fail ... */
1712 * Release all the links from css_sets to this hierarchy's
1715 write_lock(&css_set_lock);
1717 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1719 list_del(&link->cg_link_list);
1720 list_del(&link->cgrp_link_list);
1723 write_unlock(&css_set_lock);
1725 if (!list_empty(&root->root_list)) {
1726 list_del(&root->root_list);
1730 mutex_unlock(&cgroup_root_mutex);
1731 mutex_unlock(&cgroup_mutex);
1733 simple_xattrs_free(&cgrp->xattrs);
1735 kill_litter_super(sb);
1736 cgroup_drop_root(root);
1739 static struct file_system_type cgroup_fs_type = {
1741 .mount = cgroup_mount,
1742 .kill_sb = cgroup_kill_sb,
1745 static struct kobject *cgroup_kobj;
1748 * cgroup_path - generate the path of a cgroup
1749 * @cgrp: the cgroup in question
1750 * @buf: the buffer to write the path into
1751 * @buflen: the length of the buffer
1753 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1754 * reference. Writes path of cgroup into buf. Returns 0 on success,
1757 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1759 struct dentry *dentry = cgrp->dentry;
1762 rcu_lockdep_assert(rcu_read_lock_held() || cgroup_lock_is_held(),
1763 "cgroup_path() called without proper locking");
1765 if (!dentry || cgrp == dummytop) {
1767 * Inactive subsystems have no dentry for their root
1774 start = buf + buflen - 1;
1778 int len = dentry->d_name.len;
1780 if ((start -= len) < buf)
1781 return -ENAMETOOLONG;
1782 memcpy(start, dentry->d_name.name, len);
1783 cgrp = cgrp->parent;
1787 dentry = cgrp->dentry;
1791 return -ENAMETOOLONG;
1794 memmove(buf, start, buf + buflen - start);
1797 EXPORT_SYMBOL_GPL(cgroup_path);
1800 * Control Group taskset
1802 struct task_and_cgroup {
1803 struct task_struct *task;
1804 struct cgroup *cgrp;
1808 struct cgroup_taskset {
1809 struct task_and_cgroup single;
1810 struct flex_array *tc_array;
1813 struct cgroup *cur_cgrp;
1817 * cgroup_taskset_first - reset taskset and return the first task
1818 * @tset: taskset of interest
1820 * @tset iteration is initialized and the first task is returned.
1822 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1824 if (tset->tc_array) {
1826 return cgroup_taskset_next(tset);
1828 tset->cur_cgrp = tset->single.cgrp;
1829 return tset->single.task;
1832 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1835 * cgroup_taskset_next - iterate to the next task in taskset
1836 * @tset: taskset of interest
1838 * Return the next task in @tset. Iteration must have been initialized
1839 * with cgroup_taskset_first().
1841 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1843 struct task_and_cgroup *tc;
1845 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1848 tc = flex_array_get(tset->tc_array, tset->idx++);
1849 tset->cur_cgrp = tc->cgrp;
1852 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1855 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1856 * @tset: taskset of interest
1858 * Return the cgroup for the current (last returned) task of @tset. This
1859 * function must be preceded by either cgroup_taskset_first() or
1860 * cgroup_taskset_next().
1862 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1864 return tset->cur_cgrp;
1866 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1869 * cgroup_taskset_size - return the number of tasks in taskset
1870 * @tset: taskset of interest
1872 int cgroup_taskset_size(struct cgroup_taskset *tset)
1874 return tset->tc_array ? tset->tc_array_len : 1;
1876 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1880 * cgroup_task_migrate - move a task from one cgroup to another.
1882 * 'guarantee' is set if the caller promises that a new css_set for the task
1883 * will already exist. If not set, this function might sleep, and can fail with
1884 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1886 static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1887 struct task_struct *tsk, struct css_set *newcg)
1889 struct css_set *oldcg;
1892 * We are synchronized through threadgroup_lock() against PF_EXITING
1893 * setting such that we can't race against cgroup_exit() changing the
1894 * css_set to init_css_set and dropping the old one.
1896 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1897 oldcg = tsk->cgroups;
1900 rcu_assign_pointer(tsk->cgroups, newcg);
1903 /* Update the css_set linked lists if we're using them */
1904 write_lock(&css_set_lock);
1905 if (!list_empty(&tsk->cg_list))
1906 list_move(&tsk->cg_list, &newcg->tasks);
1907 write_unlock(&css_set_lock);
1910 * We just gained a reference on oldcg by taking it from the task. As
1911 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1912 * it here; it will be freed under RCU.
1914 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1919 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1920 * @cgrp: the cgroup the task is attaching to
1921 * @tsk: the task to be attached
1923 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1926 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1929 struct cgroup_subsys *ss, *failed_ss = NULL;
1930 struct cgroup *oldcgrp;
1931 struct cgroupfs_root *root = cgrp->root;
1932 struct cgroup_taskset tset = { };
1933 struct css_set *newcg;
1935 /* @tsk either already exited or can't exit until the end */
1936 if (tsk->flags & PF_EXITING)
1939 /* Nothing to do if the task is already in that cgroup */
1940 oldcgrp = task_cgroup_from_root(tsk, root);
1941 if (cgrp == oldcgrp)
1944 tset.single.task = tsk;
1945 tset.single.cgrp = oldcgrp;
1947 for_each_subsys(root, ss) {
1948 if (ss->can_attach) {
1949 retval = ss->can_attach(cgrp, &tset);
1952 * Remember on which subsystem the can_attach()
1953 * failed, so that we only call cancel_attach()
1954 * against the subsystems whose can_attach()
1955 * succeeded. (See below)
1963 newcg = find_css_set(tsk->cgroups, cgrp);
1969 cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1971 for_each_subsys(root, ss) {
1973 ss->attach(cgrp, &tset);
1979 for_each_subsys(root, ss) {
1980 if (ss == failed_ss)
1982 * This subsystem was the one that failed the
1983 * can_attach() check earlier, so we don't need
1984 * to call cancel_attach() against it or any
1985 * remaining subsystems.
1988 if (ss->cancel_attach)
1989 ss->cancel_attach(cgrp, &tset);
1996 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1997 * @from: attach to all cgroups of a given task
1998 * @tsk: the task to be attached
2000 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2002 struct cgroupfs_root *root;
2006 for_each_active_root(root) {
2007 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2009 retval = cgroup_attach_task(from_cg, tsk);
2017 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2020 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2021 * @cgrp: the cgroup to attach to
2022 * @leader: the threadgroup leader task_struct of the group to be attached
2024 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2025 * task_lock of each thread in leader's threadgroup individually in turn.
2027 static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2029 int retval, i, group_size;
2030 struct cgroup_subsys *ss, *failed_ss = NULL;
2031 /* guaranteed to be initialized later, but the compiler needs this */
2032 struct cgroupfs_root *root = cgrp->root;
2033 /* threadgroup list cursor and array */
2034 struct task_struct *tsk;
2035 struct task_and_cgroup *tc;
2036 struct flex_array *group;
2037 struct cgroup_taskset tset = { };
2040 * step 0: in order to do expensive, possibly blocking operations for
2041 * every thread, we cannot iterate the thread group list, since it needs
2042 * rcu or tasklist locked. instead, build an array of all threads in the
2043 * group - group_rwsem prevents new threads from appearing, and if
2044 * threads exit, this will just be an over-estimate.
2046 group_size = get_nr_threads(leader);
2047 /* flex_array supports very large thread-groups better than kmalloc. */
2048 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2051 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2052 retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2054 goto out_free_group_list;
2059 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2060 * already PF_EXITING could be freed from underneath us unless we
2061 * take an rcu_read_lock.
2065 struct task_and_cgroup ent;
2067 /* @tsk either already exited or can't exit until the end */
2068 if (tsk->flags & PF_EXITING)
2071 /* as per above, nr_threads may decrease, but not increase. */
2072 BUG_ON(i >= group_size);
2074 ent.cgrp = task_cgroup_from_root(tsk, root);
2075 /* nothing to do if this task is already in the cgroup */
2076 if (ent.cgrp == cgrp)
2079 * saying GFP_ATOMIC has no effect here because we did prealloc
2080 * earlier, but it's good form to communicate our expectations.
2082 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2083 BUG_ON(retval != 0);
2085 } while_each_thread(leader, tsk);
2087 /* remember the number of threads in the array for later. */
2089 tset.tc_array = group;
2090 tset.tc_array_len = group_size;
2092 /* methods shouldn't be called if no task is actually migrating */
2095 goto out_free_group_list;
2098 * step 1: check that we can legitimately attach to the cgroup.
2100 for_each_subsys(root, ss) {
2101 if (ss->can_attach) {
2102 retval = ss->can_attach(cgrp, &tset);
2105 goto out_cancel_attach;
2111 * step 2: make sure css_sets exist for all threads to be migrated.
2112 * we use find_css_set, which allocates a new one if necessary.
2114 for (i = 0; i < group_size; i++) {
2115 tc = flex_array_get(group, i);
2116 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2119 goto out_put_css_set_refs;
2124 * step 3: now that we're guaranteed success wrt the css_sets,
2125 * proceed to move all tasks to the new cgroup. There are no
2126 * failure cases after here, so this is the commit point.
2128 for (i = 0; i < group_size; i++) {
2129 tc = flex_array_get(group, i);
2130 cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2132 /* nothing is sensitive to fork() after this point. */
2135 * step 4: do subsystem attach callbacks.
2137 for_each_subsys(root, ss) {
2139 ss->attach(cgrp, &tset);
2143 * step 5: success! and cleanup
2147 out_put_css_set_refs:
2149 for (i = 0; i < group_size; i++) {
2150 tc = flex_array_get(group, i);
2153 put_css_set(tc->cg);
2158 for_each_subsys(root, ss) {
2159 if (ss == failed_ss)
2161 if (ss->cancel_attach)
2162 ss->cancel_attach(cgrp, &tset);
2165 out_free_group_list:
2166 flex_array_free(group);
2171 * Find the task_struct of the task to attach by vpid and pass it along to the
2172 * function to attach either it or all tasks in its threadgroup. Will lock
2173 * cgroup_mutex and threadgroup; may take task_lock of task.
2175 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2177 struct task_struct *tsk;
2178 const struct cred *cred = current_cred(), *tcred;
2181 if (!cgroup_lock_live_group(cgrp))
2187 tsk = find_task_by_vpid(pid);
2191 goto out_unlock_cgroup;
2194 * even if we're attaching all tasks in the thread group, we
2195 * only need to check permissions on one of them.
2197 tcred = __task_cred(tsk);
2198 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2199 !uid_eq(cred->euid, tcred->uid) &&
2200 !uid_eq(cred->euid, tcred->suid)) {
2203 goto out_unlock_cgroup;
2209 tsk = tsk->group_leader;
2212 * Workqueue threads may acquire PF_THREAD_BOUND and become
2213 * trapped in a cpuset, or RT worker may be born in a cgroup
2214 * with no rt_runtime allocated. Just say no.
2216 if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2219 goto out_unlock_cgroup;
2222 get_task_struct(tsk);
2225 threadgroup_lock(tsk);
2227 if (!thread_group_leader(tsk)) {
2229 * a race with de_thread from another thread's exec()
2230 * may strip us of our leadership, if this happens,
2231 * there is no choice but to throw this task away and
2232 * try again; this is
2233 * "double-double-toil-and-trouble-check locking".
2235 threadgroup_unlock(tsk);
2236 put_task_struct(tsk);
2237 goto retry_find_task;
2239 ret = cgroup_attach_proc(cgrp, tsk);
2241 ret = cgroup_attach_task(cgrp, tsk);
2242 threadgroup_unlock(tsk);
2244 put_task_struct(tsk);
2250 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2252 return attach_task_by_pid(cgrp, pid, false);
2255 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2257 return attach_task_by_pid(cgrp, tgid, true);
2261 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2262 * @cgrp: the cgroup to be checked for liveness
2264 * On success, returns true; the lock should be later released with
2265 * cgroup_unlock(). On failure returns false with no lock held.
2267 bool cgroup_lock_live_group(struct cgroup *cgrp)
2269 mutex_lock(&cgroup_mutex);
2270 if (cgroup_is_removed(cgrp)) {
2271 mutex_unlock(&cgroup_mutex);
2276 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2278 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2281 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2282 if (strlen(buffer) >= PATH_MAX)
2284 if (!cgroup_lock_live_group(cgrp))
2286 mutex_lock(&cgroup_root_mutex);
2287 strcpy(cgrp->root->release_agent_path, buffer);
2288 mutex_unlock(&cgroup_root_mutex);
2293 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2294 struct seq_file *seq)
2296 if (!cgroup_lock_live_group(cgrp))
2298 seq_puts(seq, cgrp->root->release_agent_path);
2299 seq_putc(seq, '\n');
2304 /* A buffer size big enough for numbers or short strings */
2305 #define CGROUP_LOCAL_BUFFER_SIZE 64
2307 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2309 const char __user *userbuf,
2310 size_t nbytes, loff_t *unused_ppos)
2312 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2318 if (nbytes >= sizeof(buffer))
2320 if (copy_from_user(buffer, userbuf, nbytes))
2323 buffer[nbytes] = 0; /* nul-terminate */
2324 if (cft->write_u64) {
2325 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2328 retval = cft->write_u64(cgrp, cft, val);
2330 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2333 retval = cft->write_s64(cgrp, cft, val);
2340 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2342 const char __user *userbuf,
2343 size_t nbytes, loff_t *unused_ppos)
2345 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2347 size_t max_bytes = cft->max_write_len;
2348 char *buffer = local_buffer;
2351 max_bytes = sizeof(local_buffer) - 1;
2352 if (nbytes >= max_bytes)
2354 /* Allocate a dynamic buffer if we need one */
2355 if (nbytes >= sizeof(local_buffer)) {
2356 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2360 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2365 buffer[nbytes] = 0; /* nul-terminate */
2366 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2370 if (buffer != local_buffer)
2375 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2376 size_t nbytes, loff_t *ppos)
2378 struct cftype *cft = __d_cft(file->f_dentry);
2379 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2381 if (cgroup_is_removed(cgrp))
2384 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2385 if (cft->write_u64 || cft->write_s64)
2386 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2387 if (cft->write_string)
2388 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2390 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2391 return ret ? ret : nbytes;
2396 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2398 char __user *buf, size_t nbytes,
2401 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2402 u64 val = cft->read_u64(cgrp, cft);
2403 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2405 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2408 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2410 char __user *buf, size_t nbytes,
2413 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2414 s64 val = cft->read_s64(cgrp, cft);
2415 int len = sprintf(tmp, "%lld\n", (long long) val);
2417 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2420 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2421 size_t nbytes, loff_t *ppos)
2423 struct cftype *cft = __d_cft(file->f_dentry);
2424 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2426 if (cgroup_is_removed(cgrp))
2430 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2432 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2434 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2439 * seqfile ops/methods for returning structured data. Currently just
2440 * supports string->u64 maps, but can be extended in future.
2443 struct cgroup_seqfile_state {
2445 struct cgroup *cgroup;
2448 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2450 struct seq_file *sf = cb->state;
2451 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2454 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2456 struct cgroup_seqfile_state *state = m->private;
2457 struct cftype *cft = state->cft;
2458 if (cft->read_map) {
2459 struct cgroup_map_cb cb = {
2460 .fill = cgroup_map_add,
2463 return cft->read_map(state->cgroup, cft, &cb);
2465 return cft->read_seq_string(state->cgroup, cft, m);
2468 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2470 struct seq_file *seq = file->private_data;
2471 kfree(seq->private);
2472 return single_release(inode, file);
2475 static const struct file_operations cgroup_seqfile_operations = {
2477 .write = cgroup_file_write,
2478 .llseek = seq_lseek,
2479 .release = cgroup_seqfile_release,
2482 static int cgroup_file_open(struct inode *inode, struct file *file)
2487 err = generic_file_open(inode, file);
2490 cft = __d_cft(file->f_dentry);
2492 if (cft->read_map || cft->read_seq_string) {
2493 struct cgroup_seqfile_state *state =
2494 kzalloc(sizeof(*state), GFP_USER);
2498 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2499 file->f_op = &cgroup_seqfile_operations;
2500 err = single_open(file, cgroup_seqfile_show, state);
2503 } else if (cft->open)
2504 err = cft->open(inode, file);
2511 static int cgroup_file_release(struct inode *inode, struct file *file)
2513 struct cftype *cft = __d_cft(file->f_dentry);
2515 return cft->release(inode, file);
2520 * cgroup_rename - Only allow simple rename of directories in place.
2522 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2523 struct inode *new_dir, struct dentry *new_dentry)
2525 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2527 if (new_dentry->d_inode)
2529 if (old_dir != new_dir)
2531 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2534 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2536 if (S_ISDIR(dentry->d_inode->i_mode))
2537 return &__d_cgrp(dentry)->xattrs;
2539 return &__d_cft(dentry)->xattrs;
2542 static inline int xattr_enabled(struct dentry *dentry)
2544 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2545 return test_bit(ROOT_XATTR, &root->flags);
2548 static bool is_valid_xattr(const char *name)
2550 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2551 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2556 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2557 const void *val, size_t size, int flags)
2559 if (!xattr_enabled(dentry))
2561 if (!is_valid_xattr(name))
2563 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2566 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2568 if (!xattr_enabled(dentry))
2570 if (!is_valid_xattr(name))
2572 return simple_xattr_remove(__d_xattrs(dentry), name);
2575 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2576 void *buf, size_t size)
2578 if (!xattr_enabled(dentry))
2580 if (!is_valid_xattr(name))
2582 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2585 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2587 if (!xattr_enabled(dentry))
2589 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2592 static const struct file_operations cgroup_file_operations = {
2593 .read = cgroup_file_read,
2594 .write = cgroup_file_write,
2595 .llseek = generic_file_llseek,
2596 .open = cgroup_file_open,
2597 .release = cgroup_file_release,
2600 static const struct inode_operations cgroup_file_inode_operations = {
2601 .setxattr = cgroup_setxattr,
2602 .getxattr = cgroup_getxattr,
2603 .listxattr = cgroup_listxattr,
2604 .removexattr = cgroup_removexattr,
2607 static const struct inode_operations cgroup_dir_inode_operations = {
2608 .lookup = cgroup_lookup,
2609 .mkdir = cgroup_mkdir,
2610 .rmdir = cgroup_rmdir,
2611 .rename = cgroup_rename,
2612 .setxattr = cgroup_setxattr,
2613 .getxattr = cgroup_getxattr,
2614 .listxattr = cgroup_listxattr,
2615 .removexattr = cgroup_removexattr,
2618 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2620 if (dentry->d_name.len > NAME_MAX)
2621 return ERR_PTR(-ENAMETOOLONG);
2622 d_add(dentry, NULL);
2627 * Check if a file is a control file
2629 static inline struct cftype *__file_cft(struct file *file)
2631 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2632 return ERR_PTR(-EINVAL);
2633 return __d_cft(file->f_dentry);
2636 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2637 struct super_block *sb)
2639 struct inode *inode;
2643 if (dentry->d_inode)
2646 inode = cgroup_new_inode(mode, sb);
2650 if (S_ISDIR(mode)) {
2651 inode->i_op = &cgroup_dir_inode_operations;
2652 inode->i_fop = &simple_dir_operations;
2654 /* start off with i_nlink == 2 (for "." entry) */
2656 inc_nlink(dentry->d_parent->d_inode);
2658 /* start with the directory inode held, so that we can
2659 * populate it without racing with another mkdir */
2660 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2661 } else if (S_ISREG(mode)) {
2663 inode->i_fop = &cgroup_file_operations;
2664 inode->i_op = &cgroup_file_inode_operations;
2666 d_instantiate(dentry, inode);
2667 dget(dentry); /* Extra count - pin the dentry in core */
2672 * cgroup_file_mode - deduce file mode of a control file
2673 * @cft: the control file in question
2675 * returns cft->mode if ->mode is not 0
2676 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2677 * returns S_IRUGO if it has only a read handler
2678 * returns S_IWUSR if it has only a write hander
2680 static umode_t cgroup_file_mode(const struct cftype *cft)
2687 if (cft->read || cft->read_u64 || cft->read_s64 ||
2688 cft->read_map || cft->read_seq_string)
2691 if (cft->write || cft->write_u64 || cft->write_s64 ||
2692 cft->write_string || cft->trigger)
2698 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2701 struct dentry *dir = cgrp->dentry;
2702 struct cgroup *parent = __d_cgrp(dir);
2703 struct dentry *dentry;
2707 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2709 simple_xattrs_init(&cft->xattrs);
2711 /* does @cft->flags tell us to skip creation on @cgrp? */
2712 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2714 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2717 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2718 strcpy(name, subsys->name);
2721 strcat(name, cft->name);
2723 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2725 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2729 dentry = lookup_one_len(name, dir, strlen(name));
2730 if (IS_ERR(dentry)) {
2731 error = PTR_ERR(dentry);
2735 mode = cgroup_file_mode(cft);
2736 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2738 cfe->type = (void *)cft;
2739 cfe->dentry = dentry;
2740 dentry->d_fsdata = cfe;
2741 list_add_tail(&cfe->node, &parent->files);
2750 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2751 struct cftype cfts[], bool is_add)
2756 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2758 err = cgroup_add_file(cgrp, subsys, cft);
2760 err = cgroup_rm_file(cgrp, cft);
2762 pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2763 is_add ? "add" : "remove", cft->name, err);
2770 static DEFINE_MUTEX(cgroup_cft_mutex);
2772 static void cgroup_cfts_prepare(void)
2773 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2776 * Thanks to the entanglement with vfs inode locking, we can't walk
2777 * the existing cgroups under cgroup_mutex and create files.
2778 * Instead, we increment reference on all cgroups and build list of
2779 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2780 * exclusive access to the field.
2782 mutex_lock(&cgroup_cft_mutex);
2783 mutex_lock(&cgroup_mutex);
2786 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2787 struct cftype *cfts, bool is_add)
2788 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2791 struct cgroup *cgrp, *n;
2793 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2794 if (cfts && ss->root != &rootnode) {
2795 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2797 list_add_tail(&cgrp->cft_q_node, &pending);
2801 mutex_unlock(&cgroup_mutex);
2804 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2805 * files for all cgroups which were created before.
2807 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2808 struct inode *inode = cgrp->dentry->d_inode;
2810 mutex_lock(&inode->i_mutex);
2811 mutex_lock(&cgroup_mutex);
2812 if (!cgroup_is_removed(cgrp))
2813 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2814 mutex_unlock(&cgroup_mutex);
2815 mutex_unlock(&inode->i_mutex);
2817 list_del_init(&cgrp->cft_q_node);
2821 mutex_unlock(&cgroup_cft_mutex);
2825 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2826 * @ss: target cgroup subsystem
2827 * @cfts: zero-length name terminated array of cftypes
2829 * Register @cfts to @ss. Files described by @cfts are created for all
2830 * existing cgroups to which @ss is attached and all future cgroups will
2831 * have them too. This function can be called anytime whether @ss is
2834 * Returns 0 on successful registration, -errno on failure. Note that this
2835 * function currently returns 0 as long as @cfts registration is successful
2836 * even if some file creation attempts on existing cgroups fail.
2838 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2840 struct cftype_set *set;
2842 set = kzalloc(sizeof(*set), GFP_KERNEL);
2846 cgroup_cfts_prepare();
2848 list_add_tail(&set->node, &ss->cftsets);
2849 cgroup_cfts_commit(ss, cfts, true);
2853 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2856 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2857 * @ss: target cgroup subsystem
2858 * @cfts: zero-length name terminated array of cftypes
2860 * Unregister @cfts from @ss. Files described by @cfts are removed from
2861 * all existing cgroups to which @ss is attached and all future cgroups
2862 * won't have them either. This function can be called anytime whether @ss
2863 * is attached or not.
2865 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2866 * registered with @ss.
2868 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2870 struct cftype_set *set;
2872 cgroup_cfts_prepare();
2874 list_for_each_entry(set, &ss->cftsets, node) {
2875 if (set->cfts == cfts) {
2876 list_del_init(&set->node);
2877 cgroup_cfts_commit(ss, cfts, false);
2882 cgroup_cfts_commit(ss, NULL, false);
2887 * cgroup_task_count - count the number of tasks in a cgroup.
2888 * @cgrp: the cgroup in question
2890 * Return the number of tasks in the cgroup.
2892 int cgroup_task_count(const struct cgroup *cgrp)
2895 struct cg_cgroup_link *link;
2897 read_lock(&css_set_lock);
2898 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2899 count += atomic_read(&link->cg->refcount);
2901 read_unlock(&css_set_lock);
2906 * Advance a list_head iterator. The iterator should be positioned at
2907 * the start of a css_set
2909 static void cgroup_advance_iter(struct cgroup *cgrp,
2910 struct cgroup_iter *it)
2912 struct list_head *l = it->cg_link;
2913 struct cg_cgroup_link *link;
2916 /* Advance to the next non-empty css_set */
2919 if (l == &cgrp->css_sets) {
2923 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2925 } while (list_empty(&cg->tasks));
2927 it->task = cg->tasks.next;
2931 * To reduce the fork() overhead for systems that are not actually
2932 * using their cgroups capability, we don't maintain the lists running
2933 * through each css_set to its tasks until we see the list actually
2934 * used - in other words after the first call to cgroup_iter_start().
2936 static void cgroup_enable_task_cg_lists(void)
2938 struct task_struct *p, *g;
2939 write_lock(&css_set_lock);
2940 use_task_css_set_links = 1;
2942 * We need tasklist_lock because RCU is not safe against
2943 * while_each_thread(). Besides, a forking task that has passed
2944 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2945 * is not guaranteed to have its child immediately visible in the
2946 * tasklist if we walk through it with RCU.
2948 read_lock(&tasklist_lock);
2949 do_each_thread(g, p) {
2952 * We should check if the process is exiting, otherwise
2953 * it will race with cgroup_exit() in that the list
2954 * entry won't be deleted though the process has exited.
2956 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2957 list_add(&p->cg_list, &p->cgroups->tasks);
2959 } while_each_thread(g, p);
2960 read_unlock(&tasklist_lock);
2961 write_unlock(&css_set_lock);
2965 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2966 * @pos: the current position (%NULL to initiate traversal)
2967 * @cgroup: cgroup whose descendants to walk
2969 * To be used by cgroup_for_each_descendant_pre(). Find the next
2970 * descendant to visit for pre-order traversal of @cgroup's descendants.
2972 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
2973 struct cgroup *cgroup)
2975 struct cgroup *next;
2977 WARN_ON_ONCE(!rcu_read_lock_held());
2979 /* if first iteration, pretend we just visited @cgroup */
2981 if (list_empty(&cgroup->children))
2986 /* visit the first child if exists */
2987 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
2991 /* no child, visit my or the closest ancestor's next sibling */
2993 next = list_entry_rcu(pos->sibling.next, struct cgroup,
2995 if (&next->sibling != &pos->parent->children)
2999 } while (pos != cgroup);
3003 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
3005 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3007 struct cgroup *last;
3011 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3019 * cgroup_next_descendant_post - find the next descendant for post-order walk
3020 * @pos: the current position (%NULL to initiate traversal)
3021 * @cgroup: cgroup whose descendants to walk
3023 * To be used by cgroup_for_each_descendant_post(). Find the next
3024 * descendant to visit for post-order traversal of @cgroup's descendants.
3026 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3027 struct cgroup *cgroup)
3029 struct cgroup *next;
3031 WARN_ON_ONCE(!rcu_read_lock_held());
3033 /* if first iteration, visit the leftmost descendant */
3035 next = cgroup_leftmost_descendant(cgroup);
3036 return next != cgroup ? next : NULL;
3039 /* if there's an unvisited sibling, visit its leftmost descendant */
3040 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3041 if (&next->sibling != &pos->parent->children)
3042 return cgroup_leftmost_descendant(next);
3044 /* no sibling left, visit parent */
3046 return next != cgroup ? next : NULL;
3048 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3050 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3051 __acquires(css_set_lock)
3054 * The first time anyone tries to iterate across a cgroup,
3055 * we need to enable the list linking each css_set to its
3056 * tasks, and fix up all existing tasks.
3058 if (!use_task_css_set_links)
3059 cgroup_enable_task_cg_lists();
3061 read_lock(&css_set_lock);
3062 it->cg_link = &cgrp->css_sets;
3063 cgroup_advance_iter(cgrp, it);
3066 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3067 struct cgroup_iter *it)
3069 struct task_struct *res;
3070 struct list_head *l = it->task;
3071 struct cg_cgroup_link *link;
3073 /* If the iterator cg is NULL, we have no tasks */
3076 res = list_entry(l, struct task_struct, cg_list);
3077 /* Advance iterator to find next entry */
3079 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
3080 if (l == &link->cg->tasks) {
3081 /* We reached the end of this task list - move on to
3082 * the next cg_cgroup_link */
3083 cgroup_advance_iter(cgrp, it);
3090 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3091 __releases(css_set_lock)
3093 read_unlock(&css_set_lock);
3096 static inline int started_after_time(struct task_struct *t1,
3097 struct timespec *time,
3098 struct task_struct *t2)
3100 int start_diff = timespec_compare(&t1->start_time, time);
3101 if (start_diff > 0) {
3103 } else if (start_diff < 0) {
3107 * Arbitrarily, if two processes started at the same
3108 * time, we'll say that the lower pointer value
3109 * started first. Note that t2 may have exited by now
3110 * so this may not be a valid pointer any longer, but
3111 * that's fine - it still serves to distinguish
3112 * between two tasks started (effectively) simultaneously.
3119 * This function is a callback from heap_insert() and is used to order
3121 * In this case we order the heap in descending task start time.
3123 static inline int started_after(void *p1, void *p2)
3125 struct task_struct *t1 = p1;
3126 struct task_struct *t2 = p2;
3127 return started_after_time(t1, &t2->start_time, t2);
3131 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3132 * @scan: struct cgroup_scanner containing arguments for the scan
3134 * Arguments include pointers to callback functions test_task() and
3136 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3137 * and if it returns true, call process_task() for it also.
3138 * The test_task pointer may be NULL, meaning always true (select all tasks).
3139 * Effectively duplicates cgroup_iter_{start,next,end}()
3140 * but does not lock css_set_lock for the call to process_task().
3141 * The struct cgroup_scanner may be embedded in any structure of the caller's
3143 * It is guaranteed that process_task() will act on every task that
3144 * is a member of the cgroup for the duration of this call. This
3145 * function may or may not call process_task() for tasks that exit
3146 * or move to a different cgroup during the call, or are forked or
3147 * move into the cgroup during the call.
3149 * Note that test_task() may be called with locks held, and may in some
3150 * situations be called multiple times for the same task, so it should
3152 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3153 * pre-allocated and will be used for heap operations (and its "gt" member will
3154 * be overwritten), else a temporary heap will be used (allocation of which
3155 * may cause this function to fail).
3157 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3160 struct cgroup_iter it;
3161 struct task_struct *p, *dropped;
3162 /* Never dereference latest_task, since it's not refcounted */
3163 struct task_struct *latest_task = NULL;
3164 struct ptr_heap tmp_heap;
3165 struct ptr_heap *heap;
3166 struct timespec latest_time = { 0, 0 };
3169 /* The caller supplied our heap and pre-allocated its memory */
3171 heap->gt = &started_after;
3173 /* We need to allocate our own heap memory */
3175 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3177 /* cannot allocate the heap */
3183 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3184 * to determine which are of interest, and using the scanner's
3185 * "process_task" callback to process any of them that need an update.
3186 * Since we don't want to hold any locks during the task updates,
3187 * gather tasks to be processed in a heap structure.
3188 * The heap is sorted by descending task start time.
3189 * If the statically-sized heap fills up, we overflow tasks that
3190 * started later, and in future iterations only consider tasks that
3191 * started after the latest task in the previous pass. This
3192 * guarantees forward progress and that we don't miss any tasks.
3195 cgroup_iter_start(scan->cg, &it);
3196 while ((p = cgroup_iter_next(scan->cg, &it))) {
3198 * Only affect tasks that qualify per the caller's callback,
3199 * if he provided one
3201 if (scan->test_task && !scan->test_task(p, scan))
3204 * Only process tasks that started after the last task
3207 if (!started_after_time(p, &latest_time, latest_task))
3209 dropped = heap_insert(heap, p);
3210 if (dropped == NULL) {
3212 * The new task was inserted; the heap wasn't
3216 } else if (dropped != p) {
3218 * The new task was inserted, and pushed out a
3222 put_task_struct(dropped);
3225 * Else the new task was newer than anything already in
3226 * the heap and wasn't inserted
3229 cgroup_iter_end(scan->cg, &it);
3232 for (i = 0; i < heap->size; i++) {
3233 struct task_struct *q = heap->ptrs[i];
3235 latest_time = q->start_time;
3238 /* Process the task per the caller's callback */
3239 scan->process_task(q, scan);
3243 * If we had to process any tasks at all, scan again
3244 * in case some of them were in the middle of forking
3245 * children that didn't get processed.
3246 * Not the most efficient way to do it, but it avoids
3247 * having to take callback_mutex in the fork path
3251 if (heap == &tmp_heap)
3252 heap_free(&tmp_heap);
3257 * Stuff for reading the 'tasks'/'procs' files.
3259 * Reading this file can return large amounts of data if a cgroup has
3260 * *lots* of attached tasks. So it may need several calls to read(),
3261 * but we cannot guarantee that the information we produce is correct
3262 * unless we produce it entirely atomically.
3266 /* which pidlist file are we talking about? */
3267 enum cgroup_filetype {
3273 * A pidlist is a list of pids that virtually represents the contents of one
3274 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3275 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3278 struct cgroup_pidlist {
3280 * used to find which pidlist is wanted. doesn't change as long as
3281 * this particular list stays in the list.
3283 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3286 /* how many elements the above list has */
3288 /* how many files are using the current array */
3290 /* each of these stored in a list by its cgroup */
3291 struct list_head links;
3292 /* pointer to the cgroup we belong to, for list removal purposes */
3293 struct cgroup *owner;
3294 /* protects the other fields */
3295 struct rw_semaphore mutex;
3299 * The following two functions "fix" the issue where there are more pids
3300 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3301 * TODO: replace with a kernel-wide solution to this problem
3303 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3304 static void *pidlist_allocate(int count)
3306 if (PIDLIST_TOO_LARGE(count))
3307 return vmalloc(count * sizeof(pid_t));
3309 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3311 static void pidlist_free(void *p)
3313 if (is_vmalloc_addr(p))
3318 static void *pidlist_resize(void *p, int newcount)
3321 /* note: if new alloc fails, old p will still be valid either way */
3322 if (is_vmalloc_addr(p)) {
3323 newlist = vmalloc(newcount * sizeof(pid_t));
3326 memcpy(newlist, p, newcount * sizeof(pid_t));
3329 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3335 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3336 * If the new stripped list is sufficiently smaller and there's enough memory
3337 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3338 * number of unique elements.
3340 /* is the size difference enough that we should re-allocate the array? */
3341 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3342 static int pidlist_uniq(pid_t **p, int length)
3349 * we presume the 0th element is unique, so i starts at 1. trivial
3350 * edge cases first; no work needs to be done for either
3352 if (length == 0 || length == 1)
3354 /* src and dest walk down the list; dest counts unique elements */
3355 for (src = 1; src < length; src++) {
3356 /* find next unique element */
3357 while (list[src] == list[src-1]) {
3362 /* dest always points to where the next unique element goes */
3363 list[dest] = list[src];
3368 * if the length difference is large enough, we want to allocate a
3369 * smaller buffer to save memory. if this fails due to out of memory,
3370 * we'll just stay with what we've got.
3372 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3373 newlist = pidlist_resize(list, dest);
3380 static int cmppid(const void *a, const void *b)
3382 return *(pid_t *)a - *(pid_t *)b;
3386 * find the appropriate pidlist for our purpose (given procs vs tasks)
3387 * returns with the lock on that pidlist already held, and takes care
3388 * of the use count, or returns NULL with no locks held if we're out of
3391 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3392 enum cgroup_filetype type)
3394 struct cgroup_pidlist *l;
3395 /* don't need task_nsproxy() if we're looking at ourself */
3396 struct pid_namespace *ns = current->nsproxy->pid_ns;
3399 * We can't drop the pidlist_mutex before taking the l->mutex in case
3400 * the last ref-holder is trying to remove l from the list at the same
3401 * time. Holding the pidlist_mutex precludes somebody taking whichever
3402 * list we find out from under us - compare release_pid_array().
3404 mutex_lock(&cgrp->pidlist_mutex);
3405 list_for_each_entry(l, &cgrp->pidlists, links) {
3406 if (l->key.type == type && l->key.ns == ns) {
3407 /* make sure l doesn't vanish out from under us */
3408 down_write(&l->mutex);
3409 mutex_unlock(&cgrp->pidlist_mutex);
3413 /* entry not found; create a new one */
3414 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3416 mutex_unlock(&cgrp->pidlist_mutex);
3419 init_rwsem(&l->mutex);
3420 down_write(&l->mutex);
3422 l->key.ns = get_pid_ns(ns);
3423 l->use_count = 0; /* don't increment here */
3426 list_add(&l->links, &cgrp->pidlists);
3427 mutex_unlock(&cgrp->pidlist_mutex);
3432 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3434 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3435 struct cgroup_pidlist **lp)
3439 int pid, n = 0; /* used for populating the array */
3440 struct cgroup_iter it;
3441 struct task_struct *tsk;
3442 struct cgroup_pidlist *l;
3445 * If cgroup gets more users after we read count, we won't have
3446 * enough space - tough. This race is indistinguishable to the
3447 * caller from the case that the additional cgroup users didn't
3448 * show up until sometime later on.
3450 length = cgroup_task_count(cgrp);
3451 array = pidlist_allocate(length);
3454 /* now, populate the array */
3455 cgroup_iter_start(cgrp, &it);
3456 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3457 if (unlikely(n == length))
3459 /* get tgid or pid for procs or tasks file respectively */
3460 if (type == CGROUP_FILE_PROCS)
3461 pid = task_tgid_vnr(tsk);
3463 pid = task_pid_vnr(tsk);
3464 if (pid > 0) /* make sure to only use valid results */
3467 cgroup_iter_end(cgrp, &it);
3469 /* now sort & (if procs) strip out duplicates */
3470 sort(array, length, sizeof(pid_t), cmppid, NULL);
3471 if (type == CGROUP_FILE_PROCS)
3472 length = pidlist_uniq(&array, length);
3473 l = cgroup_pidlist_find(cgrp, type);
3475 pidlist_free(array);
3478 /* store array, freeing old if necessary - lock already held */
3479 pidlist_free(l->list);
3483 up_write(&l->mutex);
3489 * cgroupstats_build - build and fill cgroupstats
3490 * @stats: cgroupstats to fill information into
3491 * @dentry: A dentry entry belonging to the cgroup for which stats have
3494 * Build and fill cgroupstats so that taskstats can export it to user
3497 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3500 struct cgroup *cgrp;
3501 struct cgroup_iter it;
3502 struct task_struct *tsk;
3505 * Validate dentry by checking the superblock operations,
3506 * and make sure it's a directory.
3508 if (dentry->d_sb->s_op != &cgroup_ops ||
3509 !S_ISDIR(dentry->d_inode->i_mode))
3513 cgrp = dentry->d_fsdata;
3515 cgroup_iter_start(cgrp, &it);
3516 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3517 switch (tsk->state) {
3519 stats->nr_running++;
3521 case TASK_INTERRUPTIBLE:
3522 stats->nr_sleeping++;
3524 case TASK_UNINTERRUPTIBLE:
3525 stats->nr_uninterruptible++;
3528 stats->nr_stopped++;
3531 if (delayacct_is_task_waiting_on_io(tsk))
3532 stats->nr_io_wait++;
3536 cgroup_iter_end(cgrp, &it);
3544 * seq_file methods for the tasks/procs files. The seq_file position is the
3545 * next pid to display; the seq_file iterator is a pointer to the pid
3546 * in the cgroup->l->list array.
3549 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3552 * Initially we receive a position value that corresponds to
3553 * one more than the last pid shown (or 0 on the first call or
3554 * after a seek to the start). Use a binary-search to find the
3555 * next pid to display, if any
3557 struct cgroup_pidlist *l = s->private;
3558 int index = 0, pid = *pos;
3561 down_read(&l->mutex);
3563 int end = l->length;
3565 while (index < end) {
3566 int mid = (index + end) / 2;
3567 if (l->list[mid] == pid) {
3570 } else if (l->list[mid] <= pid)
3576 /* If we're off the end of the array, we're done */
3577 if (index >= l->length)
3579 /* Update the abstract position to be the actual pid that we found */
3580 iter = l->list + index;
3585 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3587 struct cgroup_pidlist *l = s->private;
3591 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3593 struct cgroup_pidlist *l = s->private;
3595 pid_t *end = l->list + l->length;
3597 * Advance to the next pid in the array. If this goes off the
3609 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3611 return seq_printf(s, "%d\n", *(int *)v);
3615 * seq_operations functions for iterating on pidlists through seq_file -
3616 * independent of whether it's tasks or procs
3618 static const struct seq_operations cgroup_pidlist_seq_operations = {
3619 .start = cgroup_pidlist_start,
3620 .stop = cgroup_pidlist_stop,
3621 .next = cgroup_pidlist_next,
3622 .show = cgroup_pidlist_show,
3625 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3628 * the case where we're the last user of this particular pidlist will
3629 * have us remove it from the cgroup's list, which entails taking the
3630 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3631 * pidlist_mutex, we have to take pidlist_mutex first.
3633 mutex_lock(&l->owner->pidlist_mutex);
3634 down_write(&l->mutex);
3635 BUG_ON(!l->use_count);
3636 if (!--l->use_count) {
3637 /* we're the last user if refcount is 0; remove and free */
3638 list_del(&l->links);
3639 mutex_unlock(&l->owner->pidlist_mutex);
3640 pidlist_free(l->list);
3641 put_pid_ns(l->key.ns);
3642 up_write(&l->mutex);
3646 mutex_unlock(&l->owner->pidlist_mutex);
3647 up_write(&l->mutex);
3650 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3652 struct cgroup_pidlist *l;
3653 if (!(file->f_mode & FMODE_READ))
3656 * the seq_file will only be initialized if the file was opened for
3657 * reading; hence we check if it's not null only in that case.
3659 l = ((struct seq_file *)file->private_data)->private;
3660 cgroup_release_pid_array(l);
3661 return seq_release(inode, file);
3664 static const struct file_operations cgroup_pidlist_operations = {
3666 .llseek = seq_lseek,
3667 .write = cgroup_file_write,
3668 .release = cgroup_pidlist_release,
3672 * The following functions handle opens on a file that displays a pidlist
3673 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3676 /* helper function for the two below it */
3677 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3679 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3680 struct cgroup_pidlist *l;
3683 /* Nothing to do for write-only files */
3684 if (!(file->f_mode & FMODE_READ))
3687 /* have the array populated */
3688 retval = pidlist_array_load(cgrp, type, &l);
3691 /* configure file information */
3692 file->f_op = &cgroup_pidlist_operations;
3694 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3696 cgroup_release_pid_array(l);
3699 ((struct seq_file *)file->private_data)->private = l;
3702 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3704 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3706 static int cgroup_procs_open(struct inode *unused, struct file *file)
3708 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3711 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3714 return notify_on_release(cgrp);
3717 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3721 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3723 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3725 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3730 * Unregister event and free resources.
3732 * Gets called from workqueue.
3734 static void cgroup_event_remove(struct work_struct *work)
3736 struct cgroup_event *event = container_of(work, struct cgroup_event,
3738 struct cgroup *cgrp = event->cgrp;
3740 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3742 eventfd_ctx_put(event->eventfd);
3748 * Gets called on POLLHUP on eventfd when user closes it.
3750 * Called with wqh->lock held and interrupts disabled.
3752 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3753 int sync, void *key)
3755 struct cgroup_event *event = container_of(wait,
3756 struct cgroup_event, wait);
3757 struct cgroup *cgrp = event->cgrp;
3758 unsigned long flags = (unsigned long)key;
3760 if (flags & POLLHUP) {
3761 __remove_wait_queue(event->wqh, &event->wait);
3762 spin_lock(&cgrp->event_list_lock);
3763 list_del(&event->list);
3764 spin_unlock(&cgrp->event_list_lock);
3766 * We are in atomic context, but cgroup_event_remove() may
3767 * sleep, so we have to call it in workqueue.
3769 schedule_work(&event->remove);
3775 static void cgroup_event_ptable_queue_proc(struct file *file,
3776 wait_queue_head_t *wqh, poll_table *pt)
3778 struct cgroup_event *event = container_of(pt,
3779 struct cgroup_event, pt);
3782 add_wait_queue(wqh, &event->wait);
3786 * Parse input and register new cgroup event handler.
3788 * Input must be in format '<event_fd> <control_fd> <args>'.
3789 * Interpretation of args is defined by control file implementation.
3791 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3794 struct cgroup_event *event = NULL;
3795 unsigned int efd, cfd;
3796 struct file *efile = NULL;
3797 struct file *cfile = NULL;
3801 efd = simple_strtoul(buffer, &endp, 10);
3806 cfd = simple_strtoul(buffer, &endp, 10);
3807 if ((*endp != ' ') && (*endp != '\0'))
3811 event = kzalloc(sizeof(*event), GFP_KERNEL);
3815 INIT_LIST_HEAD(&event->list);
3816 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3817 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3818 INIT_WORK(&event->remove, cgroup_event_remove);
3820 efile = eventfd_fget(efd);
3821 if (IS_ERR(efile)) {
3822 ret = PTR_ERR(efile);
3826 event->eventfd = eventfd_ctx_fileget(efile);
3827 if (IS_ERR(event->eventfd)) {
3828 ret = PTR_ERR(event->eventfd);
3838 /* the process need read permission on control file */
3839 /* AV: shouldn't we check that it's been opened for read instead? */
3840 ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3844 event->cft = __file_cft(cfile);
3845 if (IS_ERR(event->cft)) {
3846 ret = PTR_ERR(event->cft);
3850 if (!event->cft->register_event || !event->cft->unregister_event) {
3855 ret = event->cft->register_event(cgrp, event->cft,
3856 event->eventfd, buffer);
3860 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3861 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3867 * Events should be removed after rmdir of cgroup directory, but before
3868 * destroying subsystem state objects. Let's take reference to cgroup
3869 * directory dentry to do that.
3873 spin_lock(&cgrp->event_list_lock);
3874 list_add(&event->list, &cgrp->event_list);
3875 spin_unlock(&cgrp->event_list_lock);
3886 if (event && event->eventfd && !IS_ERR(event->eventfd))
3887 eventfd_ctx_put(event->eventfd);
3889 if (!IS_ERR_OR_NULL(efile))
3897 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3900 return clone_children(cgrp);
3903 static int cgroup_clone_children_write(struct cgroup *cgrp,
3908 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3910 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3915 * for the common functions, 'private' gives the type of file
3917 /* for hysterical raisins, we can't put this on the older files */
3918 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3919 static struct cftype files[] = {
3922 .open = cgroup_tasks_open,
3923 .write_u64 = cgroup_tasks_write,
3924 .release = cgroup_pidlist_release,
3925 .mode = S_IRUGO | S_IWUSR,
3928 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3929 .open = cgroup_procs_open,
3930 .write_u64 = cgroup_procs_write,
3931 .release = cgroup_pidlist_release,
3932 .mode = S_IRUGO | S_IWUSR,
3935 .name = "notify_on_release",
3936 .read_u64 = cgroup_read_notify_on_release,
3937 .write_u64 = cgroup_write_notify_on_release,
3940 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3941 .write_string = cgroup_write_event_control,
3945 .name = "cgroup.clone_children",
3946 .read_u64 = cgroup_clone_children_read,
3947 .write_u64 = cgroup_clone_children_write,
3950 .name = "release_agent",
3951 .flags = CFTYPE_ONLY_ON_ROOT,
3952 .read_seq_string = cgroup_release_agent_show,
3953 .write_string = cgroup_release_agent_write,
3954 .max_write_len = PATH_MAX,
3960 * cgroup_populate_dir - selectively creation of files in a directory
3961 * @cgrp: target cgroup
3962 * @base_files: true if the base files should be added
3963 * @subsys_mask: mask of the subsystem ids whose files should be added
3965 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
3966 unsigned long subsys_mask)
3969 struct cgroup_subsys *ss;
3972 err = cgroup_addrm_files(cgrp, NULL, files, true);
3977 /* process cftsets of each subsystem */
3978 for_each_subsys(cgrp->root, ss) {
3979 struct cftype_set *set;
3980 if (!test_bit(ss->subsys_id, &subsys_mask))
3983 list_for_each_entry(set, &ss->cftsets, node)
3984 cgroup_addrm_files(cgrp, ss, set->cfts, true);
3987 /* This cgroup is ready now */
3988 for_each_subsys(cgrp->root, ss) {
3989 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3991 * Update id->css pointer and make this css visible from
3992 * CSS ID functions. This pointer will be dereferened
3993 * from RCU-read-side without locks.
3996 rcu_assign_pointer(css->id->css, css);
4002 static void css_dput_fn(struct work_struct *work)
4004 struct cgroup_subsys_state *css =
4005 container_of(work, struct cgroup_subsys_state, dput_work);
4006 struct dentry *dentry = css->cgroup->dentry;
4007 struct super_block *sb = dentry->d_sb;
4009 atomic_inc(&sb->s_active);
4011 deactivate_super(sb);
4014 static void init_cgroup_css(struct cgroup_subsys_state *css,
4015 struct cgroup_subsys *ss,
4016 struct cgroup *cgrp)
4019 atomic_set(&css->refcnt, 1);
4022 if (cgrp == dummytop)
4023 css->flags |= CSS_ROOT;
4024 BUG_ON(cgrp->subsys[ss->subsys_id]);
4025 cgrp->subsys[ss->subsys_id] = css;
4028 * css holds an extra ref to @cgrp->dentry which is put on the last
4029 * css_put(). dput() requires process context, which css_put() may
4030 * be called without. @css->dput_work will be used to invoke
4031 * dput() asynchronously from css_put().
4033 INIT_WORK(&css->dput_work, css_dput_fn);
4037 * cgroup_create - create a cgroup
4038 * @parent: cgroup that will be parent of the new cgroup
4039 * @dentry: dentry of the new cgroup
4040 * @mode: mode to set on new inode
4042 * Must be called with the mutex on the parent inode held
4044 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4047 struct cgroup *cgrp;
4048 struct cgroupfs_root *root = parent->root;
4050 struct cgroup_subsys *ss;
4051 struct super_block *sb = root->sb;
4053 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4058 * Only live parents can have children. Note that the liveliness
4059 * check isn't strictly necessary because cgroup_mkdir() and
4060 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4061 * anyway so that locking is contained inside cgroup proper and we
4062 * don't get nasty surprises if we ever grow another caller.
4064 if (!cgroup_lock_live_group(parent)) {
4069 /* Grab a reference on the superblock so the hierarchy doesn't
4070 * get deleted on unmount if there are child cgroups. This
4071 * can be done outside cgroup_mutex, since the sb can't
4072 * disappear while someone has an open control file on the
4074 atomic_inc(&sb->s_active);
4076 init_cgroup_housekeeping(cgrp);
4078 cgrp->parent = parent;
4079 cgrp->root = parent->root;
4080 cgrp->top_cgroup = parent->top_cgroup;
4082 if (notify_on_release(parent))
4083 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4085 if (clone_children(parent))
4086 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
4088 for_each_subsys(root, ss) {
4089 struct cgroup_subsys_state *css;
4091 css = ss->create(cgrp);
4096 init_cgroup_css(css, ss, cgrp);
4098 err = alloc_css_id(ss, parent, cgrp);
4102 /* At error, ->destroy() callback has to free assigned ID. */
4103 if (clone_children(parent) && ss->post_clone)
4104 ss->post_clone(cgrp);
4106 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4108 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",
4109 current->comm, current->pid, ss->name);
4110 if (!strcmp(ss->name, "memory"))
4111 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4112 ss->warned_broken_hierarchy = true;
4117 * Create directory. cgroup_create_file() returns with the new
4118 * directory locked on success so that it can be populated without
4119 * dropping cgroup_mutex.
4121 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4124 lockdep_assert_held(&dentry->d_inode->i_mutex);
4126 /* allocation complete, commit to creation */
4127 dentry->d_fsdata = cgrp;
4128 cgrp->dentry = dentry;
4129 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4130 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4131 root->number_of_cgroups++;
4133 for_each_subsys(root, ss) {
4134 /* each css holds a ref to the cgroup's dentry */
4137 /* creation succeeded, notify subsystems */
4138 if (ss->post_create)
4139 ss->post_create(cgrp);
4142 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4143 /* If err < 0, we have a half-filled directory - oh well ;) */
4145 mutex_unlock(&cgroup_mutex);
4146 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4151 for_each_subsys(root, ss) {
4152 if (cgrp->subsys[ss->subsys_id])
4155 mutex_unlock(&cgroup_mutex);
4156 /* Release the reference count that we took on the superblock */
4157 deactivate_super(sb);
4163 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4165 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4167 /* the vfs holds inode->i_mutex already */
4168 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4172 * Check the reference count on each subsystem. Since we already
4173 * established that there are no tasks in the cgroup, if the css refcount
4174 * is also 1, then there should be no outstanding references, so the
4175 * subsystem is safe to destroy. We scan across all subsystems rather than
4176 * using the per-hierarchy linked list of mounted subsystems since we can
4177 * be called via check_for_release() with no synchronization other than
4178 * RCU, and the subsystem linked list isn't RCU-safe.
4180 static int cgroup_has_css_refs(struct cgroup *cgrp)
4185 * We won't need to lock the subsys array, because the subsystems
4186 * we're concerned about aren't going anywhere since our cgroup root
4187 * has a reference on them.
4189 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4190 struct cgroup_subsys *ss = subsys[i];
4191 struct cgroup_subsys_state *css;
4193 /* Skip subsystems not present or not in this hierarchy */
4194 if (ss == NULL || ss->root != cgrp->root)
4197 css = cgrp->subsys[ss->subsys_id];
4199 * When called from check_for_release() it's possible
4200 * that by this point the cgroup has been removed
4201 * and the css deleted. But a false-positive doesn't
4202 * matter, since it can only happen if the cgroup
4203 * has been deleted and hence no longer needs the
4204 * release agent to be called anyway.
4206 if (css && css_refcnt(css) > 1)
4212 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4214 struct cgroup *cgrp = dentry->d_fsdata;
4216 struct cgroup *parent;
4218 struct cgroup_event *event, *tmp;
4219 struct cgroup_subsys *ss;
4221 /* the vfs holds both inode->i_mutex already */
4222 mutex_lock(&cgroup_mutex);
4223 parent = cgrp->parent;
4224 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
4225 mutex_unlock(&cgroup_mutex);
4230 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4231 * removed. This makes future css_tryget() and child creation
4232 * attempts fail thus maintaining the removal conditions verified
4235 for_each_subsys(cgrp->root, ss) {
4236 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4238 WARN_ON(atomic_read(&css->refcnt) < 0);
4239 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4241 set_bit(CGRP_REMOVED, &cgrp->flags);
4244 * Tell subsystems to initate destruction. pre_destroy() should be
4245 * called with cgroup_mutex unlocked. See 3fa59dfbc3 ("cgroup: fix
4246 * potential deadlock in pre_destroy") for details.
4248 mutex_unlock(&cgroup_mutex);
4249 for_each_subsys(cgrp->root, ss)
4250 if (ss->pre_destroy)
4251 ss->pre_destroy(cgrp);
4252 mutex_lock(&cgroup_mutex);
4255 * Put all the base refs. Each css holds an extra reference to the
4256 * cgroup's dentry and cgroup removal proceeds regardless of css
4257 * refs. On the last put of each css, whenever that may be, the
4258 * extra dentry ref is put so that dentry destruction happens only
4259 * after all css's are released.
4261 for_each_subsys(cgrp->root, ss)
4262 css_put(cgrp->subsys[ss->subsys_id]);
4264 raw_spin_lock(&release_list_lock);
4265 if (!list_empty(&cgrp->release_list))
4266 list_del_init(&cgrp->release_list);
4267 raw_spin_unlock(&release_list_lock);
4269 /* delete this cgroup from parent->children */
4270 list_del_rcu(&cgrp->sibling);
4272 list_del_init(&cgrp->allcg_node);
4274 d = dget(cgrp->dentry);
4276 cgroup_d_remove_dir(d);
4279 set_bit(CGRP_RELEASABLE, &parent->flags);
4280 check_for_release(parent);
4283 * Unregister events and notify userspace.
4284 * Notify userspace about cgroup removing only after rmdir of cgroup
4285 * directory to avoid race between userspace and kernelspace
4287 spin_lock(&cgrp->event_list_lock);
4288 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4289 list_del(&event->list);
4290 remove_wait_queue(event->wqh, &event->wait);
4291 eventfd_signal(event->eventfd, 1);
4292 schedule_work(&event->remove);
4294 spin_unlock(&cgrp->event_list_lock);
4296 mutex_unlock(&cgroup_mutex);
4300 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4302 INIT_LIST_HEAD(&ss->cftsets);
4305 * base_cftset is embedded in subsys itself, no need to worry about
4308 if (ss->base_cftypes) {
4309 ss->base_cftset.cfts = ss->base_cftypes;
4310 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4314 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4316 struct cgroup_subsys_state *css;
4318 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4320 /* init base cftset */
4321 cgroup_init_cftsets(ss);
4323 /* Create the top cgroup state for this subsystem */
4324 list_add(&ss->sibling, &rootnode.subsys_list);
4325 ss->root = &rootnode;
4326 css = ss->create(dummytop);
4327 /* We don't handle early failures gracefully */
4328 BUG_ON(IS_ERR(css));
4329 init_cgroup_css(css, ss, dummytop);
4331 /* Update the init_css_set to contain a subsys
4332 * pointer to this state - since the subsystem is
4333 * newly registered, all tasks and hence the
4334 * init_css_set is in the subsystem's top cgroup. */
4335 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4337 need_forkexit_callback |= ss->fork || ss->exit;
4339 /* At system boot, before all subsystems have been
4340 * registered, no tasks have been forked, so we don't
4341 * need to invoke fork callbacks here. */
4342 BUG_ON(!list_empty(&init_task.tasks));
4346 if (ss->post_create)
4347 ss->post_create(&ss->root->top_cgroup);
4349 /* this function shouldn't be used with modular subsystems, since they
4350 * need to register a subsys_id, among other things */
4355 * cgroup_load_subsys: load and register a modular subsystem at runtime
4356 * @ss: the subsystem to load
4358 * This function should be called in a modular subsystem's initcall. If the
4359 * subsystem is built as a module, it will be assigned a new subsys_id and set
4360 * up for use. If the subsystem is built-in anyway, work is delegated to the
4361 * simpler cgroup_init_subsys.
4363 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4366 struct cgroup_subsys_state *css;
4368 /* check name and function validity */
4369 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4370 ss->create == NULL || ss->destroy == NULL)
4374 * we don't support callbacks in modular subsystems. this check is
4375 * before the ss->module check for consistency; a subsystem that could
4376 * be a module should still have no callbacks even if the user isn't
4377 * compiling it as one.
4379 if (ss->fork || ss->exit)
4383 * an optionally modular subsystem is built-in: we want to do nothing,
4384 * since cgroup_init_subsys will have already taken care of it.
4386 if (ss->module == NULL) {
4387 /* a sanity check */
4388 BUG_ON(subsys[ss->subsys_id] != ss);
4392 /* init base cftset */
4393 cgroup_init_cftsets(ss);
4395 mutex_lock(&cgroup_mutex);
4396 subsys[ss->subsys_id] = ss;
4399 * no ss->create seems to need anything important in the ss struct, so
4400 * this can happen first (i.e. before the rootnode attachment).
4402 css = ss->create(dummytop);
4404 /* failure case - need to deassign the subsys[] slot. */
4405 subsys[ss->subsys_id] = NULL;
4406 mutex_unlock(&cgroup_mutex);
4407 return PTR_ERR(css);
4410 list_add(&ss->sibling, &rootnode.subsys_list);
4411 ss->root = &rootnode;
4413 /* our new subsystem will be attached to the dummy hierarchy. */
4414 init_cgroup_css(css, ss, dummytop);
4415 /* init_idr must be after init_cgroup_css because it sets css->id. */
4417 int ret = cgroup_init_idr(ss, css);
4419 dummytop->subsys[ss->subsys_id] = NULL;
4420 ss->destroy(dummytop);
4421 subsys[ss->subsys_id] = NULL;
4422 mutex_unlock(&cgroup_mutex);
4428 * Now we need to entangle the css into the existing css_sets. unlike
4429 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4430 * will need a new pointer to it; done by iterating the css_set_table.
4431 * furthermore, modifying the existing css_sets will corrupt the hash
4432 * table state, so each changed css_set will need its hash recomputed.
4433 * this is all done under the css_set_lock.
4435 write_lock(&css_set_lock);
4436 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4438 struct hlist_node *node, *tmp;
4439 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4441 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4442 /* skip entries that we already rehashed */
4443 if (cg->subsys[ss->subsys_id])
4445 /* remove existing entry */
4446 hlist_del(&cg->hlist);
4448 cg->subsys[ss->subsys_id] = css;
4449 /* recompute hash and restore entry */
4450 new_bucket = css_set_hash(cg->subsys);
4451 hlist_add_head(&cg->hlist, new_bucket);
4454 write_unlock(&css_set_lock);
4458 if (ss->post_create)
4459 ss->post_create(&ss->root->top_cgroup);
4462 mutex_unlock(&cgroup_mutex);
4465 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4468 * cgroup_unload_subsys: unload a modular subsystem
4469 * @ss: the subsystem to unload
4471 * This function should be called in a modular subsystem's exitcall. When this
4472 * function is invoked, the refcount on the subsystem's module will be 0, so
4473 * the subsystem will not be attached to any hierarchy.
4475 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4477 struct cg_cgroup_link *link;
4478 struct hlist_head *hhead;
4480 BUG_ON(ss->module == NULL);
4483 * we shouldn't be called if the subsystem is in use, and the use of
4484 * try_module_get in parse_cgroupfs_options should ensure that it
4485 * doesn't start being used while we're killing it off.
4487 BUG_ON(ss->root != &rootnode);
4489 mutex_lock(&cgroup_mutex);
4490 /* deassign the subsys_id */
4491 subsys[ss->subsys_id] = NULL;
4493 /* remove subsystem from rootnode's list of subsystems */
4494 list_del_init(&ss->sibling);
4497 * disentangle the css from all css_sets attached to the dummytop. as
4498 * in loading, we need to pay our respects to the hashtable gods.
4500 write_lock(&css_set_lock);
4501 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4502 struct css_set *cg = link->cg;
4504 hlist_del(&cg->hlist);
4505 BUG_ON(!cg->subsys[ss->subsys_id]);
4506 cg->subsys[ss->subsys_id] = NULL;
4507 hhead = css_set_hash(cg->subsys);
4508 hlist_add_head(&cg->hlist, hhead);
4510 write_unlock(&css_set_lock);
4513 * remove subsystem's css from the dummytop and free it - need to free
4514 * before marking as null because ss->destroy needs the cgrp->subsys
4515 * pointer to find their state. note that this also takes care of
4516 * freeing the css_id.
4518 ss->destroy(dummytop);
4519 dummytop->subsys[ss->subsys_id] = NULL;
4521 mutex_unlock(&cgroup_mutex);
4523 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4526 * cgroup_init_early - cgroup initialization at system boot
4528 * Initialize cgroups at system boot, and initialize any
4529 * subsystems that request early init.
4531 int __init cgroup_init_early(void)
4534 atomic_set(&init_css_set.refcount, 1);
4535 INIT_LIST_HEAD(&init_css_set.cg_links);
4536 INIT_LIST_HEAD(&init_css_set.tasks);
4537 INIT_HLIST_NODE(&init_css_set.hlist);
4539 init_cgroup_root(&rootnode);
4541 init_task.cgroups = &init_css_set;
4543 init_css_set_link.cg = &init_css_set;
4544 init_css_set_link.cgrp = dummytop;
4545 list_add(&init_css_set_link.cgrp_link_list,
4546 &rootnode.top_cgroup.css_sets);
4547 list_add(&init_css_set_link.cg_link_list,
4548 &init_css_set.cg_links);
4550 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4551 INIT_HLIST_HEAD(&css_set_table[i]);
4553 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4554 struct cgroup_subsys *ss = subsys[i];
4556 /* at bootup time, we don't worry about modular subsystems */
4557 if (!ss || ss->module)
4561 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4562 BUG_ON(!ss->create);
4563 BUG_ON(!ss->destroy);
4564 if (ss->subsys_id != i) {
4565 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4566 ss->name, ss->subsys_id);
4571 cgroup_init_subsys(ss);
4577 * cgroup_init - cgroup initialization
4579 * Register cgroup filesystem and /proc file, and initialize
4580 * any subsystems that didn't request early init.
4582 int __init cgroup_init(void)
4586 struct hlist_head *hhead;
4588 err = bdi_init(&cgroup_backing_dev_info);
4592 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4593 struct cgroup_subsys *ss = subsys[i];
4595 /* at bootup time, we don't worry about modular subsystems */
4596 if (!ss || ss->module)
4598 if (!ss->early_init)
4599 cgroup_init_subsys(ss);
4601 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4604 /* Add init_css_set to the hash table */
4605 hhead = css_set_hash(init_css_set.subsys);
4606 hlist_add_head(&init_css_set.hlist, hhead);
4607 BUG_ON(!init_root_id(&rootnode));
4609 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4615 err = register_filesystem(&cgroup_fs_type);
4617 kobject_put(cgroup_kobj);
4621 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4625 bdi_destroy(&cgroup_backing_dev_info);
4631 * proc_cgroup_show()
4632 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4633 * - Used for /proc/<pid>/cgroup.
4634 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4635 * doesn't really matter if tsk->cgroup changes after we read it,
4636 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4637 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4638 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4639 * cgroup to top_cgroup.
4642 /* TODO: Use a proper seq_file iterator */
4643 static int proc_cgroup_show(struct seq_file *m, void *v)
4646 struct task_struct *tsk;
4649 struct cgroupfs_root *root;
4652 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4658 tsk = get_pid_task(pid, PIDTYPE_PID);
4664 mutex_lock(&cgroup_mutex);
4666 for_each_active_root(root) {
4667 struct cgroup_subsys *ss;
4668 struct cgroup *cgrp;
4671 seq_printf(m, "%d:", root->hierarchy_id);
4672 for_each_subsys(root, ss)
4673 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4674 if (strlen(root->name))
4675 seq_printf(m, "%sname=%s", count ? "," : "",
4678 cgrp = task_cgroup_from_root(tsk, root);
4679 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4687 mutex_unlock(&cgroup_mutex);
4688 put_task_struct(tsk);
4695 static int cgroup_open(struct inode *inode, struct file *file)
4697 struct pid *pid = PROC_I(inode)->pid;
4698 return single_open(file, proc_cgroup_show, pid);
4701 const struct file_operations proc_cgroup_operations = {
4702 .open = cgroup_open,
4704 .llseek = seq_lseek,
4705 .release = single_release,
4708 /* Display information about each subsystem and each hierarchy */
4709 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4713 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4715 * ideally we don't want subsystems moving around while we do this.
4716 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4717 * subsys/hierarchy state.
4719 mutex_lock(&cgroup_mutex);
4720 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4721 struct cgroup_subsys *ss = subsys[i];
4724 seq_printf(m, "%s\t%d\t%d\t%d\n",
4725 ss->name, ss->root->hierarchy_id,
4726 ss->root->number_of_cgroups, !ss->disabled);
4728 mutex_unlock(&cgroup_mutex);
4732 static int cgroupstats_open(struct inode *inode, struct file *file)
4734 return single_open(file, proc_cgroupstats_show, NULL);
4737 static const struct file_operations proc_cgroupstats_operations = {
4738 .open = cgroupstats_open,
4740 .llseek = seq_lseek,
4741 .release = single_release,
4745 * cgroup_fork - attach newly forked task to its parents cgroup.
4746 * @child: pointer to task_struct of forking parent process.
4748 * Description: A task inherits its parent's cgroup at fork().
4750 * A pointer to the shared css_set was automatically copied in
4751 * fork.c by dup_task_struct(). However, we ignore that copy, since
4752 * it was not made under the protection of RCU or cgroup_mutex, so
4753 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4754 * have already changed current->cgroups, allowing the previously
4755 * referenced cgroup group to be removed and freed.
4757 * At the point that cgroup_fork() is called, 'current' is the parent
4758 * task, and the passed argument 'child' points to the child task.
4760 void cgroup_fork(struct task_struct *child)
4763 child->cgroups = current->cgroups;
4764 get_css_set(child->cgroups);
4765 task_unlock(current);
4766 INIT_LIST_HEAD(&child->cg_list);
4770 * cgroup_post_fork - called on a new task after adding it to the task list
4771 * @child: the task in question
4773 * Adds the task to the list running through its css_set if necessary and
4774 * call the subsystem fork() callbacks. Has to be after the task is
4775 * visible on the task list in case we race with the first call to
4776 * cgroup_iter_start() - to guarantee that the new task ends up on its
4779 void cgroup_post_fork(struct task_struct *child)
4784 * use_task_css_set_links is set to 1 before we walk the tasklist
4785 * under the tasklist_lock and we read it here after we added the child
4786 * to the tasklist under the tasklist_lock as well. If the child wasn't
4787 * yet in the tasklist when we walked through it from
4788 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4789 * should be visible now due to the paired locking and barriers implied
4790 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4791 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4794 if (use_task_css_set_links) {
4795 write_lock(&css_set_lock);
4797 if (list_empty(&child->cg_list))
4798 list_add(&child->cg_list, &child->cgroups->tasks);
4800 write_unlock(&css_set_lock);
4804 * Call ss->fork(). This must happen after @child is linked on
4805 * css_set; otherwise, @child might change state between ->fork()
4806 * and addition to css_set.
4808 if (need_forkexit_callback) {
4809 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4810 struct cgroup_subsys *ss = subsys[i];
4813 * fork/exit callbacks are supported only for
4814 * builtin subsystems and we don't need further
4815 * synchronization as they never go away.
4817 if (!ss || ss->module)
4827 * cgroup_exit - detach cgroup from exiting task
4828 * @tsk: pointer to task_struct of exiting process
4829 * @run_callback: run exit callbacks?
4831 * Description: Detach cgroup from @tsk and release it.
4833 * Note that cgroups marked notify_on_release force every task in
4834 * them to take the global cgroup_mutex mutex when exiting.
4835 * This could impact scaling on very large systems. Be reluctant to
4836 * use notify_on_release cgroups where very high task exit scaling
4837 * is required on large systems.
4839 * the_top_cgroup_hack:
4841 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4843 * We call cgroup_exit() while the task is still competent to
4844 * handle notify_on_release(), then leave the task attached to the
4845 * root cgroup in each hierarchy for the remainder of its exit.
4847 * To do this properly, we would increment the reference count on
4848 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4849 * code we would add a second cgroup function call, to drop that
4850 * reference. This would just create an unnecessary hot spot on
4851 * the top_cgroup reference count, to no avail.
4853 * Normally, holding a reference to a cgroup without bumping its
4854 * count is unsafe. The cgroup could go away, or someone could
4855 * attach us to a different cgroup, decrementing the count on
4856 * the first cgroup that we never incremented. But in this case,
4857 * top_cgroup isn't going away, and either task has PF_EXITING set,
4858 * which wards off any cgroup_attach_task() attempts, or task is a failed
4859 * fork, never visible to cgroup_attach_task.
4861 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4867 * Unlink from the css_set task list if necessary.
4868 * Optimistically check cg_list before taking
4871 if (!list_empty(&tsk->cg_list)) {
4872 write_lock(&css_set_lock);
4873 if (!list_empty(&tsk->cg_list))
4874 list_del_init(&tsk->cg_list);
4875 write_unlock(&css_set_lock);
4878 /* Reassign the task to the init_css_set. */
4881 tsk->cgroups = &init_css_set;
4883 if (run_callbacks && need_forkexit_callback) {
4884 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4885 struct cgroup_subsys *ss = subsys[i];
4887 /* modular subsystems can't use callbacks */
4888 if (!ss || ss->module)
4892 struct cgroup *old_cgrp =
4893 rcu_dereference_raw(cg->subsys[i])->cgroup;
4894 struct cgroup *cgrp = task_cgroup(tsk, i);
4895 ss->exit(cgrp, old_cgrp, tsk);
4902 put_css_set_taskexit(cg);
4906 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4907 * @cgrp: the cgroup in question
4908 * @task: the task in question
4910 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4913 * If we are sending in dummytop, then presumably we are creating
4914 * the top cgroup in the subsystem.
4916 * Called only by the ns (nsproxy) cgroup.
4918 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4921 struct cgroup *target;
4923 if (cgrp == dummytop)
4926 target = task_cgroup_from_root(task, cgrp->root);
4927 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4928 cgrp = cgrp->parent;
4929 ret = (cgrp == target);
4933 static void check_for_release(struct cgroup *cgrp)
4935 /* All of these checks rely on RCU to keep the cgroup
4936 * structure alive */
4937 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4938 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4939 /* Control Group is currently removeable. If it's not
4940 * already queued for a userspace notification, queue
4942 int need_schedule_work = 0;
4943 raw_spin_lock(&release_list_lock);
4944 if (!cgroup_is_removed(cgrp) &&
4945 list_empty(&cgrp->release_list)) {
4946 list_add(&cgrp->release_list, &release_list);
4947 need_schedule_work = 1;
4949 raw_spin_unlock(&release_list_lock);
4950 if (need_schedule_work)
4951 schedule_work(&release_agent_work);
4955 /* Caller must verify that the css is not for root cgroup */
4956 bool __css_tryget(struct cgroup_subsys_state *css)
4961 v = css_refcnt(css);
4962 t = atomic_cmpxchg(&css->refcnt, v, v + 1);
4970 EXPORT_SYMBOL_GPL(__css_tryget);
4972 /* Caller must verify that the css is not for root cgroup */
4973 void __css_put(struct cgroup_subsys_state *css)
4975 struct cgroup *cgrp = css->cgroup;
4979 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
4983 if (notify_on_release(cgrp)) {
4984 set_bit(CGRP_RELEASABLE, &cgrp->flags);
4985 check_for_release(cgrp);
4989 schedule_work(&css->dput_work);
4994 EXPORT_SYMBOL_GPL(__css_put);
4997 * Notify userspace when a cgroup is released, by running the
4998 * configured release agent with the name of the cgroup (path
4999 * relative to the root of cgroup file system) as the argument.
5001 * Most likely, this user command will try to rmdir this cgroup.
5003 * This races with the possibility that some other task will be
5004 * attached to this cgroup before it is removed, or that some other
5005 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5006 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5007 * unused, and this cgroup will be reprieved from its death sentence,
5008 * to continue to serve a useful existence. Next time it's released,
5009 * we will get notified again, if it still has 'notify_on_release' set.
5011 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5012 * means only wait until the task is successfully execve()'d. The
5013 * separate release agent task is forked by call_usermodehelper(),
5014 * then control in this thread returns here, without waiting for the
5015 * release agent task. We don't bother to wait because the caller of
5016 * this routine has no use for the exit status of the release agent
5017 * task, so no sense holding our caller up for that.
5019 static void cgroup_release_agent(struct work_struct *work)
5021 BUG_ON(work != &release_agent_work);
5022 mutex_lock(&cgroup_mutex);
5023 raw_spin_lock(&release_list_lock);
5024 while (!list_empty(&release_list)) {
5025 char *argv[3], *envp[3];
5027 char *pathbuf = NULL, *agentbuf = NULL;
5028 struct cgroup *cgrp = list_entry(release_list.next,
5031 list_del_init(&cgrp->release_list);
5032 raw_spin_unlock(&release_list_lock);
5033 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5036 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5038 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5043 argv[i++] = agentbuf;
5044 argv[i++] = pathbuf;
5048 /* minimal command environment */
5049 envp[i++] = "HOME=/";
5050 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5053 /* Drop the lock while we invoke the usermode helper,
5054 * since the exec could involve hitting disk and hence
5055 * be a slow process */
5056 mutex_unlock(&cgroup_mutex);
5057 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5058 mutex_lock(&cgroup_mutex);
5062 raw_spin_lock(&release_list_lock);
5064 raw_spin_unlock(&release_list_lock);
5065 mutex_unlock(&cgroup_mutex);
5068 static int __init cgroup_disable(char *str)
5073 while ((token = strsep(&str, ",")) != NULL) {
5076 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5077 struct cgroup_subsys *ss = subsys[i];
5080 * cgroup_disable, being at boot time, can't
5081 * know about module subsystems, so we don't
5084 if (!ss || ss->module)
5087 if (!strcmp(token, ss->name)) {
5089 printk(KERN_INFO "Disabling %s control group"
5090 " subsystem\n", ss->name);
5097 __setup("cgroup_disable=", cgroup_disable);
5100 * Functons for CSS ID.
5104 *To get ID other than 0, this should be called when !cgroup_is_removed().
5106 unsigned short css_id(struct cgroup_subsys_state *css)
5108 struct css_id *cssid;
5111 * This css_id() can return correct value when somone has refcnt
5112 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5113 * it's unchanged until freed.
5115 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5121 EXPORT_SYMBOL_GPL(css_id);
5123 unsigned short css_depth(struct cgroup_subsys_state *css)
5125 struct css_id *cssid;
5127 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5130 return cssid->depth;
5133 EXPORT_SYMBOL_GPL(css_depth);
5136 * css_is_ancestor - test "root" css is an ancestor of "child"
5137 * @child: the css to be tested.
5138 * @root: the css supporsed to be an ancestor of the child.
5140 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5141 * this function reads css->id, the caller must hold rcu_read_lock().
5142 * But, considering usual usage, the csses should be valid objects after test.
5143 * Assuming that the caller will do some action to the child if this returns
5144 * returns true, the caller must take "child";s reference count.
5145 * If "child" is valid object and this returns true, "root" is valid, too.
5148 bool css_is_ancestor(struct cgroup_subsys_state *child,
5149 const struct cgroup_subsys_state *root)
5151 struct css_id *child_id;
5152 struct css_id *root_id;
5154 child_id = rcu_dereference(child->id);
5157 root_id = rcu_dereference(root->id);
5160 if (child_id->depth < root_id->depth)
5162 if (child_id->stack[root_id->depth] != root_id->id)
5167 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5169 struct css_id *id = css->id;
5170 /* When this is called before css_id initialization, id can be NULL */
5174 BUG_ON(!ss->use_id);
5176 rcu_assign_pointer(id->css, NULL);
5177 rcu_assign_pointer(css->id, NULL);
5178 spin_lock(&ss->id_lock);
5179 idr_remove(&ss->idr, id->id);
5180 spin_unlock(&ss->id_lock);
5181 kfree_rcu(id, rcu_head);
5183 EXPORT_SYMBOL_GPL(free_css_id);
5186 * This is called by init or create(). Then, calls to this function are
5187 * always serialized (By cgroup_mutex() at create()).
5190 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5192 struct css_id *newid;
5193 int myid, error, size;
5195 BUG_ON(!ss->use_id);
5197 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5198 newid = kzalloc(size, GFP_KERNEL);
5200 return ERR_PTR(-ENOMEM);
5202 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
5206 spin_lock(&ss->id_lock);
5207 /* Don't use 0. allocates an ID of 1-65535 */
5208 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
5209 spin_unlock(&ss->id_lock);
5211 /* Returns error when there are no free spaces for new ID.*/
5216 if (myid > CSS_ID_MAX)
5220 newid->depth = depth;
5224 spin_lock(&ss->id_lock);
5225 idr_remove(&ss->idr, myid);
5226 spin_unlock(&ss->id_lock);
5229 return ERR_PTR(error);
5233 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5234 struct cgroup_subsys_state *rootcss)
5236 struct css_id *newid;
5238 spin_lock_init(&ss->id_lock);
5241 newid = get_new_cssid(ss, 0);
5243 return PTR_ERR(newid);
5245 newid->stack[0] = newid->id;
5246 newid->css = rootcss;
5247 rootcss->id = newid;
5251 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5252 struct cgroup *child)
5254 int subsys_id, i, depth = 0;
5255 struct cgroup_subsys_state *parent_css, *child_css;
5256 struct css_id *child_id, *parent_id;
5258 subsys_id = ss->subsys_id;
5259 parent_css = parent->subsys[subsys_id];
5260 child_css = child->subsys[subsys_id];
5261 parent_id = parent_css->id;
5262 depth = parent_id->depth + 1;
5264 child_id = get_new_cssid(ss, depth);
5265 if (IS_ERR(child_id))
5266 return PTR_ERR(child_id);
5268 for (i = 0; i < depth; i++)
5269 child_id->stack[i] = parent_id->stack[i];
5270 child_id->stack[depth] = child_id->id;
5272 * child_id->css pointer will be set after this cgroup is available
5273 * see cgroup_populate_dir()
5275 rcu_assign_pointer(child_css->id, child_id);
5281 * css_lookup - lookup css by id
5282 * @ss: cgroup subsys to be looked into.
5285 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5286 * NULL if not. Should be called under rcu_read_lock()
5288 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5290 struct css_id *cssid = NULL;
5292 BUG_ON(!ss->use_id);
5293 cssid = idr_find(&ss->idr, id);
5295 if (unlikely(!cssid))
5298 return rcu_dereference(cssid->css);
5300 EXPORT_SYMBOL_GPL(css_lookup);
5303 * css_get_next - lookup next cgroup under specified hierarchy.
5304 * @ss: pointer to subsystem
5305 * @id: current position of iteration.
5306 * @root: pointer to css. search tree under this.
5307 * @foundid: position of found object.
5309 * Search next css under the specified hierarchy of rootid. Calling under
5310 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5312 struct cgroup_subsys_state *
5313 css_get_next(struct cgroup_subsys *ss, int id,
5314 struct cgroup_subsys_state *root, int *foundid)
5316 struct cgroup_subsys_state *ret = NULL;
5319 int rootid = css_id(root);
5320 int depth = css_depth(root);
5325 BUG_ON(!ss->use_id);
5326 WARN_ON_ONCE(!rcu_read_lock_held());
5328 /* fill start point for scan */
5332 * scan next entry from bitmap(tree), tmpid is updated after
5335 tmp = idr_get_next(&ss->idr, &tmpid);
5338 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5339 ret = rcu_dereference(tmp->css);
5345 /* continue to scan from next id */
5352 * get corresponding css from file open on cgroupfs directory
5354 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5356 struct cgroup *cgrp;
5357 struct inode *inode;
5358 struct cgroup_subsys_state *css;
5360 inode = f->f_dentry->d_inode;
5361 /* check in cgroup filesystem dir */
5362 if (inode->i_op != &cgroup_dir_inode_operations)
5363 return ERR_PTR(-EBADF);
5365 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5366 return ERR_PTR(-EINVAL);
5369 cgrp = __d_cgrp(f->f_dentry);
5370 css = cgrp->subsys[id];
5371 return css ? css : ERR_PTR(-ENOENT);
5374 #ifdef CONFIG_CGROUP_DEBUG
5375 static struct cgroup_subsys_state *debug_create(struct cgroup *cont)
5377 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5380 return ERR_PTR(-ENOMEM);
5385 static void debug_destroy(struct cgroup *cont)
5387 kfree(cont->subsys[debug_subsys_id]);
5390 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5392 return atomic_read(&cont->count);
5395 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5397 return cgroup_task_count(cont);
5400 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5402 return (u64)(unsigned long)current->cgroups;
5405 static u64 current_css_set_refcount_read(struct cgroup *cont,
5411 count = atomic_read(¤t->cgroups->refcount);
5416 static int current_css_set_cg_links_read(struct cgroup *cont,
5418 struct seq_file *seq)
5420 struct cg_cgroup_link *link;
5423 read_lock(&css_set_lock);
5425 cg = rcu_dereference(current->cgroups);
5426 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5427 struct cgroup *c = link->cgrp;
5431 name = c->dentry->d_name.name;
5434 seq_printf(seq, "Root %d group %s\n",
5435 c->root->hierarchy_id, name);
5438 read_unlock(&css_set_lock);
5442 #define MAX_TASKS_SHOWN_PER_CSS 25
5443 static int cgroup_css_links_read(struct cgroup *cont,
5445 struct seq_file *seq)
5447 struct cg_cgroup_link *link;
5449 read_lock(&css_set_lock);
5450 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5451 struct css_set *cg = link->cg;
5452 struct task_struct *task;
5454 seq_printf(seq, "css_set %p\n", cg);
5455 list_for_each_entry(task, &cg->tasks, cg_list) {
5456 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5457 seq_puts(seq, " ...\n");
5460 seq_printf(seq, " task %d\n",
5461 task_pid_vnr(task));
5465 read_unlock(&css_set_lock);
5469 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5471 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5474 static struct cftype debug_files[] = {
5476 .name = "cgroup_refcount",
5477 .read_u64 = cgroup_refcount_read,
5480 .name = "taskcount",
5481 .read_u64 = debug_taskcount_read,
5485 .name = "current_css_set",
5486 .read_u64 = current_css_set_read,
5490 .name = "current_css_set_refcount",
5491 .read_u64 = current_css_set_refcount_read,
5495 .name = "current_css_set_cg_links",
5496 .read_seq_string = current_css_set_cg_links_read,
5500 .name = "cgroup_css_links",
5501 .read_seq_string = cgroup_css_links_read,
5505 .name = "releasable",
5506 .read_u64 = releasable_read,
5512 struct cgroup_subsys debug_subsys = {
5514 .create = debug_create,
5515 .destroy = debug_destroy,
5516 .subsys_id = debug_subsys_id,
5517 .base_cftypes = debug_files,
5519 #endif /* CONFIG_CGROUP_DEBUG */