2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hashtable.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_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 /* IDs for cgroups in this hierarchy */
142 struct ida cgroup_ida;
144 /* The path to use for release notifications. */
145 char release_agent_path[PATH_MAX];
147 /* The name for this hierarchy - may be empty */
148 char name[MAX_CGROUP_ROOT_NAMELEN];
152 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
153 * subsystems that are otherwise unattached - it never has more than a
154 * single cgroup, and all tasks are part of that cgroup.
156 static struct cgroupfs_root rootnode;
159 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
162 struct list_head node;
163 struct dentry *dentry;
168 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
169 * cgroup_subsys->use_id != 0.
171 #define CSS_ID_MAX (65535)
174 * The css to which this ID points. This pointer is set to valid value
175 * after cgroup is populated. If cgroup is removed, this will be NULL.
176 * This pointer is expected to be RCU-safe because destroy()
177 * is called after synchronize_rcu(). But for safe use, css_tryget()
178 * should be used for avoiding race.
180 struct cgroup_subsys_state __rcu *css;
186 * Depth in hierarchy which this ID belongs to.
188 unsigned short depth;
190 * ID is freed by RCU. (and lookup routine is RCU safe.)
192 struct rcu_head rcu_head;
194 * Hierarchy of CSS ID belongs to.
196 unsigned short stack[0]; /* Array of Length (depth+1) */
200 * cgroup_event represents events which userspace want to receive.
202 struct cgroup_event {
204 * Cgroup which the event belongs to.
208 * Control file which the event associated.
212 * eventfd to signal userspace about the event.
214 struct eventfd_ctx *eventfd;
216 * Each of these stored in a list by the cgroup.
218 struct list_head list;
220 * All fields below needed to unregister event when
221 * userspace closes eventfd.
224 wait_queue_head_t *wqh;
226 struct work_struct remove;
229 /* The list of hierarchy roots */
231 static LIST_HEAD(roots);
232 static int root_count;
234 static DEFINE_IDA(hierarchy_ida);
235 static int next_hierarchy_id;
236 static DEFINE_SPINLOCK(hierarchy_id_lock);
238 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
239 #define dummytop (&rootnode.top_cgroup)
241 /* This flag indicates whether tasks in the fork and exit paths should
242 * check for fork/exit handlers to call. This avoids us having to do
243 * extra work in the fork/exit path if none of the subsystems need to
246 static int need_forkexit_callback __read_mostly;
248 static int cgroup_destroy_locked(struct cgroup *cgrp);
249 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
250 struct cftype cfts[], bool is_add);
252 #ifdef CONFIG_PROVE_LOCKING
253 int cgroup_lock_is_held(void)
255 return lockdep_is_held(&cgroup_mutex);
257 #else /* #ifdef CONFIG_PROVE_LOCKING */
258 int cgroup_lock_is_held(void)
260 return mutex_is_locked(&cgroup_mutex);
262 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
264 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
266 static int css_unbias_refcnt(int refcnt)
268 return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
271 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
272 static int css_refcnt(struct cgroup_subsys_state *css)
274 int v = atomic_read(&css->refcnt);
276 return css_unbias_refcnt(v);
279 /* convenient tests for these bits */
280 inline int cgroup_is_removed(const struct cgroup *cgrp)
282 return test_bit(CGRP_REMOVED, &cgrp->flags);
285 /* bits in struct cgroupfs_root flags field */
287 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
288 ROOT_XATTR, /* supports extended attributes */
291 static int cgroup_is_releasable(const struct cgroup *cgrp)
294 (1 << CGRP_RELEASABLE) |
295 (1 << CGRP_NOTIFY_ON_RELEASE);
296 return (cgrp->flags & bits) == bits;
299 static int notify_on_release(const struct cgroup *cgrp)
301 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
305 * for_each_subsys() allows you to iterate on each subsystem attached to
306 * an active hierarchy
308 #define for_each_subsys(_root, _ss) \
309 list_for_each_entry(_ss, &_root->subsys_list, sibling)
311 /* for_each_active_root() allows you to iterate across the active hierarchies */
312 #define for_each_active_root(_root) \
313 list_for_each_entry(_root, &roots, root_list)
315 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
317 return dentry->d_fsdata;
320 static inline struct cfent *__d_cfe(struct dentry *dentry)
322 return dentry->d_fsdata;
325 static inline struct cftype *__d_cft(struct dentry *dentry)
327 return __d_cfe(dentry)->type;
330 /* the list of cgroups eligible for automatic release. Protected by
331 * release_list_lock */
332 static LIST_HEAD(release_list);
333 static DEFINE_RAW_SPINLOCK(release_list_lock);
334 static void cgroup_release_agent(struct work_struct *work);
335 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
336 static void check_for_release(struct cgroup *cgrp);
338 /* Link structure for associating css_set objects with cgroups */
339 struct cg_cgroup_link {
341 * List running through cg_cgroup_links associated with a
342 * cgroup, anchored on cgroup->css_sets
344 struct list_head cgrp_link_list;
347 * List running through cg_cgroup_links pointing at a
348 * single css_set object, anchored on css_set->cg_links
350 struct list_head cg_link_list;
354 /* The default css_set - used by init and its children prior to any
355 * hierarchies being mounted. It contains a pointer to the root state
356 * for each subsystem. Also used to anchor the list of css_sets. Not
357 * reference-counted, to improve performance when child cgroups
358 * haven't been created.
361 static struct css_set init_css_set;
362 static struct cg_cgroup_link init_css_set_link;
364 static int cgroup_init_idr(struct cgroup_subsys *ss,
365 struct cgroup_subsys_state *css);
367 /* css_set_lock protects the list of css_set objects, and the
368 * chain of tasks off each css_set. Nests outside task->alloc_lock
369 * due to cgroup_iter_start() */
370 static DEFINE_RWLOCK(css_set_lock);
371 static int css_set_count;
374 * hash table for cgroup groups. This improves the performance to find
375 * an existing css_set. This hash doesn't (currently) take into
376 * account cgroups in empty hierarchies.
378 #define CSS_SET_HASH_BITS 7
379 static DEFINE_HASHTABLE(css_set_table, CSS_SET_HASH_BITS);
381 static unsigned long css_set_hash(struct cgroup_subsys_state *css[])
384 unsigned long key = 0UL;
386 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
387 key += (unsigned long)css[i];
388 key = (key >> 16) ^ key;
393 /* We don't maintain the lists running through each css_set to its
394 * task until after the first call to cgroup_iter_start(). This
395 * reduces the fork()/exit() overhead for people who have cgroups
396 * compiled into their kernel but not actually in use */
397 static int use_task_css_set_links __read_mostly;
399 static void __put_css_set(struct css_set *cg, int taskexit)
401 struct cg_cgroup_link *link;
402 struct cg_cgroup_link *saved_link;
404 * Ensure that the refcount doesn't hit zero while any readers
405 * can see it. Similar to atomic_dec_and_lock(), but for an
408 if (atomic_add_unless(&cg->refcount, -1, 1))
410 write_lock(&css_set_lock);
411 if (!atomic_dec_and_test(&cg->refcount)) {
412 write_unlock(&css_set_lock);
416 /* This css_set is dead. unlink it and release cgroup refcounts */
417 hash_del(&cg->hlist);
420 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
422 struct cgroup *cgrp = link->cgrp;
423 list_del(&link->cg_link_list);
424 list_del(&link->cgrp_link_list);
425 if (atomic_dec_and_test(&cgrp->count) &&
426 notify_on_release(cgrp)) {
428 set_bit(CGRP_RELEASABLE, &cgrp->flags);
429 check_for_release(cgrp);
435 write_unlock(&css_set_lock);
436 kfree_rcu(cg, rcu_head);
440 * refcounted get/put for css_set objects
442 static inline void get_css_set(struct css_set *cg)
444 atomic_inc(&cg->refcount);
447 static inline void put_css_set(struct css_set *cg)
449 __put_css_set(cg, 0);
452 static inline void put_css_set_taskexit(struct css_set *cg)
454 __put_css_set(cg, 1);
458 * compare_css_sets - helper function for find_existing_css_set().
459 * @cg: candidate css_set being tested
460 * @old_cg: existing css_set for a task
461 * @new_cgrp: cgroup that's being entered by the task
462 * @template: desired set of css pointers in css_set (pre-calculated)
464 * Returns true if "cg" matches "old_cg" except for the hierarchy
465 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
467 static bool compare_css_sets(struct css_set *cg,
468 struct css_set *old_cg,
469 struct cgroup *new_cgrp,
470 struct cgroup_subsys_state *template[])
472 struct list_head *l1, *l2;
474 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
475 /* Not all subsystems matched */
480 * Compare cgroup pointers in order to distinguish between
481 * different cgroups in heirarchies with no subsystems. We
482 * could get by with just this check alone (and skip the
483 * memcmp above) but on most setups the memcmp check will
484 * avoid the need for this more expensive check on almost all
489 l2 = &old_cg->cg_links;
491 struct cg_cgroup_link *cgl1, *cgl2;
492 struct cgroup *cg1, *cg2;
496 /* See if we reached the end - both lists are equal length. */
497 if (l1 == &cg->cg_links) {
498 BUG_ON(l2 != &old_cg->cg_links);
501 BUG_ON(l2 == &old_cg->cg_links);
503 /* Locate the cgroups associated with these links. */
504 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
505 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
508 /* Hierarchies should be linked in the same order. */
509 BUG_ON(cg1->root != cg2->root);
512 * If this hierarchy is the hierarchy of the cgroup
513 * that's changing, then we need to check that this
514 * css_set points to the new cgroup; if it's any other
515 * hierarchy, then this css_set should point to the
516 * same cgroup as the old css_set.
518 if (cg1->root == new_cgrp->root) {
530 * find_existing_css_set() is a helper for
531 * find_css_set(), and checks to see whether an existing
532 * css_set is suitable.
534 * oldcg: the cgroup group that we're using before the cgroup
537 * cgrp: the cgroup that we're moving into
539 * template: location in which to build the desired set of subsystem
540 * state objects for the new cgroup group
542 static struct css_set *find_existing_css_set(
543 struct css_set *oldcg,
545 struct cgroup_subsys_state *template[])
548 struct cgroupfs_root *root = cgrp->root;
549 struct hlist_node *node;
554 * Build the set of subsystem state objects that we want to see in the
555 * new css_set. while subsystems can change globally, the entries here
556 * won't change, so no need for locking.
558 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
559 if (root->subsys_mask & (1UL << i)) {
560 /* Subsystem is in this hierarchy. So we want
561 * the subsystem state from the new
563 template[i] = cgrp->subsys[i];
565 /* Subsystem is not in this hierarchy, so we
566 * don't want to change the subsystem state */
567 template[i] = oldcg->subsys[i];
571 key = css_set_hash(template);
572 hash_for_each_possible(css_set_table, cg, node, hlist, key) {
573 if (!compare_css_sets(cg, oldcg, cgrp, template))
576 /* This css_set matches what we need */
580 /* No existing cgroup group matched */
584 static void free_cg_links(struct list_head *tmp)
586 struct cg_cgroup_link *link;
587 struct cg_cgroup_link *saved_link;
589 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
590 list_del(&link->cgrp_link_list);
596 * allocate_cg_links() allocates "count" cg_cgroup_link structures
597 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
598 * success or a negative error
600 static int allocate_cg_links(int count, struct list_head *tmp)
602 struct cg_cgroup_link *link;
605 for (i = 0; i < count; i++) {
606 link = kmalloc(sizeof(*link), GFP_KERNEL);
611 list_add(&link->cgrp_link_list, tmp);
617 * link_css_set - a helper function to link a css_set to a cgroup
618 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
619 * @cg: the css_set to be linked
620 * @cgrp: the destination cgroup
622 static void link_css_set(struct list_head *tmp_cg_links,
623 struct css_set *cg, struct cgroup *cgrp)
625 struct cg_cgroup_link *link;
627 BUG_ON(list_empty(tmp_cg_links));
628 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
632 atomic_inc(&cgrp->count);
633 list_move(&link->cgrp_link_list, &cgrp->css_sets);
635 * Always add links to the tail of the list so that the list
636 * is sorted by order of hierarchy creation
638 list_add_tail(&link->cg_link_list, &cg->cg_links);
642 * find_css_set() takes an existing cgroup group and a
643 * cgroup object, and returns a css_set object that's
644 * equivalent to the old group, but with the given cgroup
645 * substituted into the appropriate hierarchy. Must be called with
648 static struct css_set *find_css_set(
649 struct css_set *oldcg, struct cgroup *cgrp)
652 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
654 struct list_head tmp_cg_links;
656 struct cg_cgroup_link *link;
659 /* First see if we already have a cgroup group that matches
661 read_lock(&css_set_lock);
662 res = find_existing_css_set(oldcg, cgrp, template);
665 read_unlock(&css_set_lock);
670 res = kmalloc(sizeof(*res), GFP_KERNEL);
674 /* Allocate all the cg_cgroup_link objects that we'll need */
675 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
680 atomic_set(&res->refcount, 1);
681 INIT_LIST_HEAD(&res->cg_links);
682 INIT_LIST_HEAD(&res->tasks);
683 INIT_HLIST_NODE(&res->hlist);
685 /* Copy the set of subsystem state objects generated in
686 * find_existing_css_set() */
687 memcpy(res->subsys, template, sizeof(res->subsys));
689 write_lock(&css_set_lock);
690 /* Add reference counts and links from the new css_set. */
691 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
692 struct cgroup *c = link->cgrp;
693 if (c->root == cgrp->root)
695 link_css_set(&tmp_cg_links, res, c);
698 BUG_ON(!list_empty(&tmp_cg_links));
702 /* Add this cgroup group to the hash table */
703 key = css_set_hash(res->subsys);
704 hash_add(css_set_table, &res->hlist, key);
706 write_unlock(&css_set_lock);
712 * Return the cgroup for "task" from the given hierarchy. Must be
713 * called with cgroup_mutex held.
715 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
716 struct cgroupfs_root *root)
719 struct cgroup *res = NULL;
721 BUG_ON(!mutex_is_locked(&cgroup_mutex));
722 read_lock(&css_set_lock);
724 * No need to lock the task - since we hold cgroup_mutex the
725 * task can't change groups, so the only thing that can happen
726 * is that it exits and its css is set back to init_css_set.
729 if (css == &init_css_set) {
730 res = &root->top_cgroup;
732 struct cg_cgroup_link *link;
733 list_for_each_entry(link, &css->cg_links, cg_link_list) {
734 struct cgroup *c = link->cgrp;
735 if (c->root == root) {
741 read_unlock(&css_set_lock);
747 * There is one global cgroup mutex. We also require taking
748 * task_lock() when dereferencing a task's cgroup subsys pointers.
749 * See "The task_lock() exception", at the end of this comment.
751 * A task must hold cgroup_mutex to modify cgroups.
753 * Any task can increment and decrement the count field without lock.
754 * So in general, code holding cgroup_mutex can't rely on the count
755 * field not changing. However, if the count goes to zero, then only
756 * cgroup_attach_task() can increment it again. Because a count of zero
757 * means that no tasks are currently attached, therefore there is no
758 * way a task attached to that cgroup can fork (the other way to
759 * increment the count). So code holding cgroup_mutex can safely
760 * assume that if the count is zero, it will stay zero. Similarly, if
761 * a task holds cgroup_mutex on a cgroup with zero count, it
762 * knows that the cgroup won't be removed, as cgroup_rmdir()
765 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
766 * (usually) take cgroup_mutex. These are the two most performance
767 * critical pieces of code here. The exception occurs on cgroup_exit(),
768 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
769 * is taken, and if the cgroup count is zero, a usermode call made
770 * to the release agent with the name of the cgroup (path relative to
771 * the root of cgroup file system) as the argument.
773 * A cgroup can only be deleted if both its 'count' of using tasks
774 * is zero, and its list of 'children' cgroups is empty. Since all
775 * tasks in the system use _some_ cgroup, and since there is always at
776 * least one task in the system (init, pid == 1), therefore, top_cgroup
777 * always has either children cgroups and/or using tasks. So we don't
778 * need a special hack to ensure that top_cgroup cannot be deleted.
780 * The task_lock() exception
782 * The need for this exception arises from the action of
783 * cgroup_attach_task(), which overwrites one task's cgroup pointer with
784 * another. It does so using cgroup_mutex, however there are
785 * several performance critical places that need to reference
786 * task->cgroup without the expense of grabbing a system global
787 * mutex. Therefore except as noted below, when dereferencing or, as
788 * in cgroup_attach_task(), modifying a task's cgroup pointer we use
789 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
790 * the task_struct routinely used for such matters.
792 * P.S. One more locking exception. RCU is used to guard the
793 * update of a tasks cgroup pointer by cgroup_attach_task()
797 * cgroup_lock - lock out any changes to cgroup structures
800 void cgroup_lock(void)
802 mutex_lock(&cgroup_mutex);
804 EXPORT_SYMBOL_GPL(cgroup_lock);
807 * cgroup_unlock - release lock on cgroup changes
809 * Undo the lock taken in a previous cgroup_lock() call.
811 void cgroup_unlock(void)
813 mutex_unlock(&cgroup_mutex);
815 EXPORT_SYMBOL_GPL(cgroup_unlock);
818 * A couple of forward declarations required, due to cyclic reference loop:
819 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
820 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
824 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
825 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
826 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
827 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
828 unsigned long subsys_mask);
829 static const struct inode_operations cgroup_dir_inode_operations;
830 static const struct file_operations proc_cgroupstats_operations;
832 static struct backing_dev_info cgroup_backing_dev_info = {
834 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
837 static int alloc_css_id(struct cgroup_subsys *ss,
838 struct cgroup *parent, struct cgroup *child);
840 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
842 struct inode *inode = new_inode(sb);
845 inode->i_ino = get_next_ino();
846 inode->i_mode = mode;
847 inode->i_uid = current_fsuid();
848 inode->i_gid = current_fsgid();
849 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
850 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
855 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
857 /* is dentry a directory ? if so, kfree() associated cgroup */
858 if (S_ISDIR(inode->i_mode)) {
859 struct cgroup *cgrp = dentry->d_fsdata;
860 struct cgroup_subsys *ss;
861 BUG_ON(!(cgroup_is_removed(cgrp)));
862 /* It's possible for external users to be holding css
863 * reference counts on a cgroup; css_put() needs to
864 * be able to access the cgroup after decrementing
865 * the reference count in order to know if it needs to
866 * queue the cgroup to be handled by the release
870 mutex_lock(&cgroup_mutex);
872 * Release the subsystem state objects.
874 for_each_subsys(cgrp->root, ss)
877 cgrp->root->number_of_cgroups--;
878 mutex_unlock(&cgroup_mutex);
881 * Drop the active superblock reference that we took when we
884 deactivate_super(cgrp->root->sb);
887 * if we're getting rid of the cgroup, refcount should ensure
888 * that there are no pidlists left.
890 BUG_ON(!list_empty(&cgrp->pidlists));
892 simple_xattrs_free(&cgrp->xattrs);
894 ida_simple_remove(&cgrp->root->cgroup_ida, cgrp->id);
895 kfree_rcu(cgrp, rcu_head);
897 struct cfent *cfe = __d_cfe(dentry);
898 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
899 struct cftype *cft = cfe->type;
901 WARN_ONCE(!list_empty(&cfe->node) &&
902 cgrp != &cgrp->root->top_cgroup,
903 "cfe still linked for %s\n", cfe->type->name);
905 simple_xattrs_free(&cft->xattrs);
910 static int cgroup_delete(const struct dentry *d)
915 static void remove_dir(struct dentry *d)
917 struct dentry *parent = dget(d->d_parent);
920 simple_rmdir(parent->d_inode, d);
924 static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
928 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
929 lockdep_assert_held(&cgroup_mutex);
931 list_for_each_entry(cfe, &cgrp->files, node) {
932 struct dentry *d = cfe->dentry;
934 if (cft && cfe->type != cft)
939 simple_unlink(cgrp->dentry->d_inode, d);
940 list_del_init(&cfe->node);
949 * cgroup_clear_directory - selective removal of base and subsystem files
950 * @dir: directory containing the files
951 * @base_files: true if the base files should be removed
952 * @subsys_mask: mask of the subsystem ids whose files should be removed
954 static void cgroup_clear_directory(struct dentry *dir, bool base_files,
955 unsigned long subsys_mask)
957 struct cgroup *cgrp = __d_cgrp(dir);
958 struct cgroup_subsys *ss;
960 for_each_subsys(cgrp->root, ss) {
961 struct cftype_set *set;
962 if (!test_bit(ss->subsys_id, &subsys_mask))
964 list_for_each_entry(set, &ss->cftsets, node)
965 cgroup_addrm_files(cgrp, NULL, set->cfts, false);
968 while (!list_empty(&cgrp->files))
969 cgroup_rm_file(cgrp, NULL);
974 * NOTE : the dentry must have been dget()'ed
976 static void cgroup_d_remove_dir(struct dentry *dentry)
978 struct dentry *parent;
979 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
981 cgroup_clear_directory(dentry, true, root->subsys_mask);
983 parent = dentry->d_parent;
984 spin_lock(&parent->d_lock);
985 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
986 list_del_init(&dentry->d_u.d_child);
987 spin_unlock(&dentry->d_lock);
988 spin_unlock(&parent->d_lock);
993 * Call with cgroup_mutex held. Drops reference counts on modules, including
994 * any duplicate ones that parse_cgroupfs_options took. If this function
995 * returns an error, no reference counts are touched.
997 static int rebind_subsystems(struct cgroupfs_root *root,
998 unsigned long final_subsys_mask)
1000 unsigned long added_mask, removed_mask;
1001 struct cgroup *cgrp = &root->top_cgroup;
1004 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1005 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1007 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
1008 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1009 /* Check that any added subsystems are currently free */
1010 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1011 unsigned long bit = 1UL << i;
1012 struct cgroup_subsys *ss = subsys[i];
1013 if (!(bit & added_mask))
1016 * Nobody should tell us to do a subsys that doesn't exist:
1017 * parse_cgroupfs_options should catch that case and refcounts
1018 * ensure that subsystems won't disappear once selected.
1021 if (ss->root != &rootnode) {
1022 /* Subsystem isn't free */
1027 /* Currently we don't handle adding/removing subsystems when
1028 * any child cgroups exist. This is theoretically supportable
1029 * but involves complex error handling, so it's being left until
1031 if (root->number_of_cgroups > 1)
1034 /* Process each subsystem */
1035 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1036 struct cgroup_subsys *ss = subsys[i];
1037 unsigned long bit = 1UL << i;
1038 if (bit & added_mask) {
1039 /* We're binding this subsystem to this hierarchy */
1041 BUG_ON(cgrp->subsys[i]);
1042 BUG_ON(!dummytop->subsys[i]);
1043 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1044 cgrp->subsys[i] = dummytop->subsys[i];
1045 cgrp->subsys[i]->cgroup = cgrp;
1046 list_move(&ss->sibling, &root->subsys_list);
1050 /* refcount was already taken, and we're keeping it */
1051 } else if (bit & removed_mask) {
1052 /* We're removing this subsystem */
1054 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1055 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1058 dummytop->subsys[i]->cgroup = dummytop;
1059 cgrp->subsys[i] = NULL;
1060 subsys[i]->root = &rootnode;
1061 list_move(&ss->sibling, &rootnode.subsys_list);
1062 /* subsystem is now free - drop reference on module */
1063 module_put(ss->module);
1064 } else if (bit & final_subsys_mask) {
1065 /* Subsystem state should already exist */
1067 BUG_ON(!cgrp->subsys[i]);
1069 * a refcount was taken, but we already had one, so
1070 * drop the extra reference.
1072 module_put(ss->module);
1073 #ifdef CONFIG_MODULE_UNLOAD
1074 BUG_ON(ss->module && !module_refcount(ss->module));
1077 /* Subsystem state shouldn't exist */
1078 BUG_ON(cgrp->subsys[i]);
1081 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1087 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1089 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1090 struct cgroup_subsys *ss;
1092 mutex_lock(&cgroup_root_mutex);
1093 for_each_subsys(root, ss)
1094 seq_printf(seq, ",%s", ss->name);
1095 if (test_bit(ROOT_NOPREFIX, &root->flags))
1096 seq_puts(seq, ",noprefix");
1097 if (test_bit(ROOT_XATTR, &root->flags))
1098 seq_puts(seq, ",xattr");
1099 if (strlen(root->release_agent_path))
1100 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1101 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags))
1102 seq_puts(seq, ",clone_children");
1103 if (strlen(root->name))
1104 seq_printf(seq, ",name=%s", root->name);
1105 mutex_unlock(&cgroup_root_mutex);
1109 struct cgroup_sb_opts {
1110 unsigned long subsys_mask;
1111 unsigned long flags;
1112 char *release_agent;
1113 bool cpuset_clone_children;
1115 /* User explicitly requested empty subsystem */
1118 struct cgroupfs_root *new_root;
1123 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1124 * with cgroup_mutex held to protect the subsys[] array. This function takes
1125 * refcounts on subsystems to be used, unless it returns error, in which case
1126 * no refcounts are taken.
1128 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1130 char *token, *o = data;
1131 bool all_ss = false, one_ss = false;
1132 unsigned long mask = (unsigned long)-1;
1134 bool module_pin_failed = false;
1136 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1138 #ifdef CONFIG_CPUSETS
1139 mask = ~(1UL << cpuset_subsys_id);
1142 memset(opts, 0, sizeof(*opts));
1144 while ((token = strsep(&o, ",")) != NULL) {
1147 if (!strcmp(token, "none")) {
1148 /* Explicitly have no subsystems */
1152 if (!strcmp(token, "all")) {
1153 /* Mutually exclusive option 'all' + subsystem name */
1159 if (!strcmp(token, "noprefix")) {
1160 set_bit(ROOT_NOPREFIX, &opts->flags);
1163 if (!strcmp(token, "clone_children")) {
1164 opts->cpuset_clone_children = true;
1167 if (!strcmp(token, "xattr")) {
1168 set_bit(ROOT_XATTR, &opts->flags);
1171 if (!strncmp(token, "release_agent=", 14)) {
1172 /* Specifying two release agents is forbidden */
1173 if (opts->release_agent)
1175 opts->release_agent =
1176 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1177 if (!opts->release_agent)
1181 if (!strncmp(token, "name=", 5)) {
1182 const char *name = token + 5;
1183 /* Can't specify an empty name */
1186 /* Must match [\w.-]+ */
1187 for (i = 0; i < strlen(name); i++) {
1191 if ((c == '.') || (c == '-') || (c == '_'))
1195 /* Specifying two names is forbidden */
1198 opts->name = kstrndup(name,
1199 MAX_CGROUP_ROOT_NAMELEN - 1,
1207 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1208 struct cgroup_subsys *ss = subsys[i];
1211 if (strcmp(token, ss->name))
1216 /* Mutually exclusive option 'all' + subsystem name */
1219 set_bit(i, &opts->subsys_mask);
1224 if (i == CGROUP_SUBSYS_COUNT)
1229 * If the 'all' option was specified select all the subsystems,
1230 * otherwise if 'none', 'name=' and a subsystem name options
1231 * were not specified, let's default to 'all'
1233 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1234 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1235 struct cgroup_subsys *ss = subsys[i];
1240 set_bit(i, &opts->subsys_mask);
1244 /* Consistency checks */
1247 * Option noprefix was introduced just for backward compatibility
1248 * with the old cpuset, so we allow noprefix only if mounting just
1249 * the cpuset subsystem.
1251 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1252 (opts->subsys_mask & mask))
1256 /* Can't specify "none" and some subsystems */
1257 if (opts->subsys_mask && opts->none)
1261 * We either have to specify by name or by subsystems. (So all
1262 * empty hierarchies must have a name).
1264 if (!opts->subsys_mask && !opts->name)
1268 * Grab references on all the modules we'll need, so the subsystems
1269 * don't dance around before rebind_subsystems attaches them. This may
1270 * take duplicate reference counts on a subsystem that's already used,
1271 * but rebind_subsystems handles this case.
1273 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1274 unsigned long bit = 1UL << i;
1276 if (!(bit & opts->subsys_mask))
1278 if (!try_module_get(subsys[i]->module)) {
1279 module_pin_failed = true;
1283 if (module_pin_failed) {
1285 * oops, one of the modules was going away. this means that we
1286 * raced with a module_delete call, and to the user this is
1287 * essentially a "subsystem doesn't exist" case.
1289 for (i--; i >= 0; i--) {
1290 /* drop refcounts only on the ones we took */
1291 unsigned long bit = 1UL << i;
1293 if (!(bit & opts->subsys_mask))
1295 module_put(subsys[i]->module);
1303 static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1306 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1307 unsigned long bit = 1UL << i;
1309 if (!(bit & subsys_mask))
1311 module_put(subsys[i]->module);
1315 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1318 struct cgroupfs_root *root = sb->s_fs_info;
1319 struct cgroup *cgrp = &root->top_cgroup;
1320 struct cgroup_sb_opts opts;
1321 unsigned long added_mask, removed_mask;
1323 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1324 mutex_lock(&cgroup_mutex);
1325 mutex_lock(&cgroup_root_mutex);
1327 /* See what subsystems are wanted */
1328 ret = parse_cgroupfs_options(data, &opts);
1332 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1333 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1334 task_tgid_nr(current), current->comm);
1336 added_mask = opts.subsys_mask & ~root->subsys_mask;
1337 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1339 /* Don't allow flags or name to change at remount */
1340 if (opts.flags != root->flags ||
1341 (opts.name && strcmp(opts.name, root->name))) {
1343 drop_parsed_module_refcounts(opts.subsys_mask);
1348 * Clear out the files of subsystems that should be removed, do
1349 * this before rebind_subsystems, since rebind_subsystems may
1350 * change this hierarchy's subsys_list.
1352 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1354 ret = rebind_subsystems(root, opts.subsys_mask);
1356 /* rebind_subsystems failed, re-populate the removed files */
1357 cgroup_populate_dir(cgrp, false, removed_mask);
1358 drop_parsed_module_refcounts(opts.subsys_mask);
1362 /* re-populate subsystem files */
1363 cgroup_populate_dir(cgrp, false, added_mask);
1365 if (opts.release_agent)
1366 strcpy(root->release_agent_path, opts.release_agent);
1368 kfree(opts.release_agent);
1370 mutex_unlock(&cgroup_root_mutex);
1371 mutex_unlock(&cgroup_mutex);
1372 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1376 static const struct super_operations cgroup_ops = {
1377 .statfs = simple_statfs,
1378 .drop_inode = generic_delete_inode,
1379 .show_options = cgroup_show_options,
1380 .remount_fs = cgroup_remount,
1383 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1385 INIT_LIST_HEAD(&cgrp->sibling);
1386 INIT_LIST_HEAD(&cgrp->children);
1387 INIT_LIST_HEAD(&cgrp->files);
1388 INIT_LIST_HEAD(&cgrp->css_sets);
1389 INIT_LIST_HEAD(&cgrp->allcg_node);
1390 INIT_LIST_HEAD(&cgrp->release_list);
1391 INIT_LIST_HEAD(&cgrp->pidlists);
1392 mutex_init(&cgrp->pidlist_mutex);
1393 INIT_LIST_HEAD(&cgrp->event_list);
1394 spin_lock_init(&cgrp->event_list_lock);
1395 simple_xattrs_init(&cgrp->xattrs);
1398 static void init_cgroup_root(struct cgroupfs_root *root)
1400 struct cgroup *cgrp = &root->top_cgroup;
1402 INIT_LIST_HEAD(&root->subsys_list);
1403 INIT_LIST_HEAD(&root->root_list);
1404 INIT_LIST_HEAD(&root->allcg_list);
1405 root->number_of_cgroups = 1;
1407 cgrp->top_cgroup = cgrp;
1408 init_cgroup_housekeeping(cgrp);
1409 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1412 static bool init_root_id(struct cgroupfs_root *root)
1417 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1419 spin_lock(&hierarchy_id_lock);
1420 /* Try to allocate the next unused ID */
1421 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1422 &root->hierarchy_id);
1424 /* Try again starting from 0 */
1425 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1427 next_hierarchy_id = root->hierarchy_id + 1;
1428 } else if (ret != -EAGAIN) {
1429 /* Can only get here if the 31-bit IDR is full ... */
1432 spin_unlock(&hierarchy_id_lock);
1437 static int cgroup_test_super(struct super_block *sb, void *data)
1439 struct cgroup_sb_opts *opts = data;
1440 struct cgroupfs_root *root = sb->s_fs_info;
1442 /* If we asked for a name then it must match */
1443 if (opts->name && strcmp(opts->name, root->name))
1447 * If we asked for subsystems (or explicitly for no
1448 * subsystems) then they must match
1450 if ((opts->subsys_mask || opts->none)
1451 && (opts->subsys_mask != root->subsys_mask))
1457 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1459 struct cgroupfs_root *root;
1461 if (!opts->subsys_mask && !opts->none)
1464 root = kzalloc(sizeof(*root), GFP_KERNEL);
1466 return ERR_PTR(-ENOMEM);
1468 if (!init_root_id(root)) {
1470 return ERR_PTR(-ENOMEM);
1472 init_cgroup_root(root);
1474 root->subsys_mask = opts->subsys_mask;
1475 root->flags = opts->flags;
1476 ida_init(&root->cgroup_ida);
1477 if (opts->release_agent)
1478 strcpy(root->release_agent_path, opts->release_agent);
1480 strcpy(root->name, opts->name);
1481 if (opts->cpuset_clone_children)
1482 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &root->top_cgroup.flags);
1486 static void cgroup_drop_root(struct cgroupfs_root *root)
1491 BUG_ON(!root->hierarchy_id);
1492 spin_lock(&hierarchy_id_lock);
1493 ida_remove(&hierarchy_ida, root->hierarchy_id);
1494 spin_unlock(&hierarchy_id_lock);
1495 ida_destroy(&root->cgroup_ida);
1499 static int cgroup_set_super(struct super_block *sb, void *data)
1502 struct cgroup_sb_opts *opts = data;
1504 /* If we don't have a new root, we can't set up a new sb */
1505 if (!opts->new_root)
1508 BUG_ON(!opts->subsys_mask && !opts->none);
1510 ret = set_anon_super(sb, NULL);
1514 sb->s_fs_info = opts->new_root;
1515 opts->new_root->sb = sb;
1517 sb->s_blocksize = PAGE_CACHE_SIZE;
1518 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1519 sb->s_magic = CGROUP_SUPER_MAGIC;
1520 sb->s_op = &cgroup_ops;
1525 static int cgroup_get_rootdir(struct super_block *sb)
1527 static const struct dentry_operations cgroup_dops = {
1528 .d_iput = cgroup_diput,
1529 .d_delete = cgroup_delete,
1532 struct inode *inode =
1533 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1538 inode->i_fop = &simple_dir_operations;
1539 inode->i_op = &cgroup_dir_inode_operations;
1540 /* directories start off with i_nlink == 2 (for "." entry) */
1542 sb->s_root = d_make_root(inode);
1545 /* for everything else we want ->d_op set */
1546 sb->s_d_op = &cgroup_dops;
1550 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1551 int flags, const char *unused_dev_name,
1554 struct cgroup_sb_opts opts;
1555 struct cgroupfs_root *root;
1557 struct super_block *sb;
1558 struct cgroupfs_root *new_root;
1559 struct inode *inode;
1561 /* First find the desired set of subsystems */
1562 mutex_lock(&cgroup_mutex);
1563 ret = parse_cgroupfs_options(data, &opts);
1564 mutex_unlock(&cgroup_mutex);
1569 * Allocate a new cgroup root. We may not need it if we're
1570 * reusing an existing hierarchy.
1572 new_root = cgroup_root_from_opts(&opts);
1573 if (IS_ERR(new_root)) {
1574 ret = PTR_ERR(new_root);
1577 opts.new_root = new_root;
1579 /* Locate an existing or new sb for this hierarchy */
1580 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1583 cgroup_drop_root(opts.new_root);
1587 root = sb->s_fs_info;
1589 if (root == opts.new_root) {
1590 /* We used the new root structure, so this is a new hierarchy */
1591 struct list_head tmp_cg_links;
1592 struct cgroup *root_cgrp = &root->top_cgroup;
1593 struct cgroupfs_root *existing_root;
1594 const struct cred *cred;
1596 struct hlist_node *node;
1599 BUG_ON(sb->s_root != NULL);
1601 ret = cgroup_get_rootdir(sb);
1603 goto drop_new_super;
1604 inode = sb->s_root->d_inode;
1606 mutex_lock(&inode->i_mutex);
1607 mutex_lock(&cgroup_mutex);
1608 mutex_lock(&cgroup_root_mutex);
1610 /* Check for name clashes with existing mounts */
1612 if (strlen(root->name))
1613 for_each_active_root(existing_root)
1614 if (!strcmp(existing_root->name, root->name))
1618 * We're accessing css_set_count without locking
1619 * css_set_lock here, but that's OK - it can only be
1620 * increased by someone holding cgroup_lock, and
1621 * that's us. The worst that can happen is that we
1622 * have some link structures left over
1624 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1628 ret = rebind_subsystems(root, root->subsys_mask);
1629 if (ret == -EBUSY) {
1630 free_cg_links(&tmp_cg_links);
1634 * There must be no failure case after here, since rebinding
1635 * takes care of subsystems' refcounts, which are explicitly
1636 * dropped in the failure exit path.
1639 /* EBUSY should be the only error here */
1642 list_add(&root->root_list, &roots);
1645 sb->s_root->d_fsdata = root_cgrp;
1646 root->top_cgroup.dentry = sb->s_root;
1648 /* Link the top cgroup in this hierarchy into all
1649 * the css_set objects */
1650 write_lock(&css_set_lock);
1651 hash_for_each(css_set_table, i, node, cg, hlist)
1652 link_css_set(&tmp_cg_links, cg, root_cgrp);
1653 write_unlock(&css_set_lock);
1655 free_cg_links(&tmp_cg_links);
1657 BUG_ON(!list_empty(&root_cgrp->children));
1658 BUG_ON(root->number_of_cgroups != 1);
1660 cred = override_creds(&init_cred);
1661 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1663 mutex_unlock(&cgroup_root_mutex);
1664 mutex_unlock(&cgroup_mutex);
1665 mutex_unlock(&inode->i_mutex);
1668 * We re-used an existing hierarchy - the new root (if
1669 * any) is not needed
1671 cgroup_drop_root(opts.new_root);
1672 /* no subsys rebinding, so refcounts don't change */
1673 drop_parsed_module_refcounts(opts.subsys_mask);
1676 kfree(opts.release_agent);
1678 return dget(sb->s_root);
1681 mutex_unlock(&cgroup_root_mutex);
1682 mutex_unlock(&cgroup_mutex);
1683 mutex_unlock(&inode->i_mutex);
1685 deactivate_locked_super(sb);
1687 drop_parsed_module_refcounts(opts.subsys_mask);
1689 kfree(opts.release_agent);
1691 return ERR_PTR(ret);
1694 static void cgroup_kill_sb(struct super_block *sb) {
1695 struct cgroupfs_root *root = sb->s_fs_info;
1696 struct cgroup *cgrp = &root->top_cgroup;
1698 struct cg_cgroup_link *link;
1699 struct cg_cgroup_link *saved_link;
1703 BUG_ON(root->number_of_cgroups != 1);
1704 BUG_ON(!list_empty(&cgrp->children));
1706 mutex_lock(&cgroup_mutex);
1707 mutex_lock(&cgroup_root_mutex);
1709 /* Rebind all subsystems back to the default hierarchy */
1710 ret = rebind_subsystems(root, 0);
1711 /* Shouldn't be able to fail ... */
1715 * Release all the links from css_sets to this hierarchy's
1718 write_lock(&css_set_lock);
1720 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1722 list_del(&link->cg_link_list);
1723 list_del(&link->cgrp_link_list);
1726 write_unlock(&css_set_lock);
1728 if (!list_empty(&root->root_list)) {
1729 list_del(&root->root_list);
1733 mutex_unlock(&cgroup_root_mutex);
1734 mutex_unlock(&cgroup_mutex);
1736 simple_xattrs_free(&cgrp->xattrs);
1738 kill_litter_super(sb);
1739 cgroup_drop_root(root);
1742 static struct file_system_type cgroup_fs_type = {
1744 .mount = cgroup_mount,
1745 .kill_sb = cgroup_kill_sb,
1748 static struct kobject *cgroup_kobj;
1751 * cgroup_path - generate the path of a cgroup
1752 * @cgrp: the cgroup in question
1753 * @buf: the buffer to write the path into
1754 * @buflen: the length of the buffer
1756 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1757 * reference. Writes path of cgroup into buf. Returns 0 on success,
1760 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1762 struct dentry *dentry = cgrp->dentry;
1765 rcu_lockdep_assert(rcu_read_lock_held() || cgroup_lock_is_held(),
1766 "cgroup_path() called without proper locking");
1768 if (!dentry || cgrp == dummytop) {
1770 * Inactive subsystems have no dentry for their root
1777 start = buf + buflen - 1;
1781 int len = dentry->d_name.len;
1783 if ((start -= len) < buf)
1784 return -ENAMETOOLONG;
1785 memcpy(start, dentry->d_name.name, len);
1786 cgrp = cgrp->parent;
1790 dentry = cgrp->dentry;
1794 return -ENAMETOOLONG;
1797 memmove(buf, start, buf + buflen - start);
1800 EXPORT_SYMBOL_GPL(cgroup_path);
1803 * Control Group taskset
1805 struct task_and_cgroup {
1806 struct task_struct *task;
1807 struct cgroup *cgrp;
1811 struct cgroup_taskset {
1812 struct task_and_cgroup single;
1813 struct flex_array *tc_array;
1816 struct cgroup *cur_cgrp;
1820 * cgroup_taskset_first - reset taskset and return the first task
1821 * @tset: taskset of interest
1823 * @tset iteration is initialized and the first task is returned.
1825 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1827 if (tset->tc_array) {
1829 return cgroup_taskset_next(tset);
1831 tset->cur_cgrp = tset->single.cgrp;
1832 return tset->single.task;
1835 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1838 * cgroup_taskset_next - iterate to the next task in taskset
1839 * @tset: taskset of interest
1841 * Return the next task in @tset. Iteration must have been initialized
1842 * with cgroup_taskset_first().
1844 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1846 struct task_and_cgroup *tc;
1848 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1851 tc = flex_array_get(tset->tc_array, tset->idx++);
1852 tset->cur_cgrp = tc->cgrp;
1855 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1858 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1859 * @tset: taskset of interest
1861 * Return the cgroup for the current (last returned) task of @tset. This
1862 * function must be preceded by either cgroup_taskset_first() or
1863 * cgroup_taskset_next().
1865 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1867 return tset->cur_cgrp;
1869 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1872 * cgroup_taskset_size - return the number of tasks in taskset
1873 * @tset: taskset of interest
1875 int cgroup_taskset_size(struct cgroup_taskset *tset)
1877 return tset->tc_array ? tset->tc_array_len : 1;
1879 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1883 * cgroup_task_migrate - move a task from one cgroup to another.
1885 * Must be called with cgroup_mutex and threadgroup locked.
1887 static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1888 struct task_struct *tsk, struct css_set *newcg)
1890 struct css_set *oldcg;
1893 * We are synchronized through threadgroup_lock() against PF_EXITING
1894 * setting such that we can't race against cgroup_exit() changing the
1895 * css_set to init_css_set and dropping the old one.
1897 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1898 oldcg = tsk->cgroups;
1901 rcu_assign_pointer(tsk->cgroups, newcg);
1904 /* Update the css_set linked lists if we're using them */
1905 write_lock(&css_set_lock);
1906 if (!list_empty(&tsk->cg_list))
1907 list_move(&tsk->cg_list, &newcg->tasks);
1908 write_unlock(&css_set_lock);
1911 * We just gained a reference on oldcg by taking it from the task. As
1912 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1913 * it here; it will be freed under RCU.
1915 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1920 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1921 * @cgrp: the cgroup the task is attaching to
1922 * @tsk: the task to be attached
1924 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1927 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1930 struct cgroup_subsys *ss, *failed_ss = NULL;
1931 struct cgroup *oldcgrp;
1932 struct cgroupfs_root *root = cgrp->root;
1933 struct cgroup_taskset tset = { };
1934 struct css_set *newcg;
1936 /* @tsk either already exited or can't exit until the end */
1937 if (tsk->flags & PF_EXITING)
1940 /* Nothing to do if the task is already in that cgroup */
1941 oldcgrp = task_cgroup_from_root(tsk, root);
1942 if (cgrp == oldcgrp)
1945 tset.single.task = tsk;
1946 tset.single.cgrp = oldcgrp;
1948 for_each_subsys(root, ss) {
1949 if (ss->can_attach) {
1950 retval = ss->can_attach(cgrp, &tset);
1953 * Remember on which subsystem the can_attach()
1954 * failed, so that we only call cancel_attach()
1955 * against the subsystems whose can_attach()
1956 * succeeded. (See below)
1964 newcg = find_css_set(tsk->cgroups, cgrp);
1970 cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1972 for_each_subsys(root, ss) {
1974 ss->attach(cgrp, &tset);
1980 for_each_subsys(root, ss) {
1981 if (ss == failed_ss)
1983 * This subsystem was the one that failed the
1984 * can_attach() check earlier, so we don't need
1985 * to call cancel_attach() against it or any
1986 * remaining subsystems.
1989 if (ss->cancel_attach)
1990 ss->cancel_attach(cgrp, &tset);
1997 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1998 * @from: attach to all cgroups of a given task
1999 * @tsk: the task to be attached
2001 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2003 struct cgroupfs_root *root;
2007 for_each_active_root(root) {
2008 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2010 retval = cgroup_attach_task(from_cg, tsk);
2018 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2021 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2022 * @cgrp: the cgroup to attach to
2023 * @leader: the threadgroup leader task_struct of the group to be attached
2025 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2026 * task_lock of each thread in leader's threadgroup individually in turn.
2028 static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2030 int retval, i, group_size;
2031 struct cgroup_subsys *ss, *failed_ss = NULL;
2032 /* guaranteed to be initialized later, but the compiler needs this */
2033 struct cgroupfs_root *root = cgrp->root;
2034 /* threadgroup list cursor and array */
2035 struct task_struct *tsk;
2036 struct task_and_cgroup *tc;
2037 struct flex_array *group;
2038 struct cgroup_taskset tset = { };
2041 * step 0: in order to do expensive, possibly blocking operations for
2042 * every thread, we cannot iterate the thread group list, since it needs
2043 * rcu or tasklist locked. instead, build an array of all threads in the
2044 * group - group_rwsem prevents new threads from appearing, and if
2045 * threads exit, this will just be an over-estimate.
2047 group_size = get_nr_threads(leader);
2048 /* flex_array supports very large thread-groups better than kmalloc. */
2049 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2052 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2053 retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2055 goto out_free_group_list;
2060 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2061 * already PF_EXITING could be freed from underneath us unless we
2062 * take an rcu_read_lock.
2066 struct task_and_cgroup ent;
2068 /* @tsk either already exited or can't exit until the end */
2069 if (tsk->flags & PF_EXITING)
2072 /* as per above, nr_threads may decrease, but not increase. */
2073 BUG_ON(i >= group_size);
2075 ent.cgrp = task_cgroup_from_root(tsk, root);
2076 /* nothing to do if this task is already in the cgroup */
2077 if (ent.cgrp == cgrp)
2080 * saying GFP_ATOMIC has no effect here because we did prealloc
2081 * earlier, but it's good form to communicate our expectations.
2083 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2084 BUG_ON(retval != 0);
2086 } while_each_thread(leader, tsk);
2088 /* remember the number of threads in the array for later. */
2090 tset.tc_array = group;
2091 tset.tc_array_len = group_size;
2093 /* methods shouldn't be called if no task is actually migrating */
2096 goto out_free_group_list;
2099 * step 1: check that we can legitimately attach to the cgroup.
2101 for_each_subsys(root, ss) {
2102 if (ss->can_attach) {
2103 retval = ss->can_attach(cgrp, &tset);
2106 goto out_cancel_attach;
2112 * step 2: make sure css_sets exist for all threads to be migrated.
2113 * we use find_css_set, which allocates a new one if necessary.
2115 for (i = 0; i < group_size; i++) {
2116 tc = flex_array_get(group, i);
2117 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2120 goto out_put_css_set_refs;
2125 * step 3: now that we're guaranteed success wrt the css_sets,
2126 * proceed to move all tasks to the new cgroup. There are no
2127 * failure cases after here, so this is the commit point.
2129 for (i = 0; i < group_size; i++) {
2130 tc = flex_array_get(group, i);
2131 cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2133 /* nothing is sensitive to fork() after this point. */
2136 * step 4: do subsystem attach callbacks.
2138 for_each_subsys(root, ss) {
2140 ss->attach(cgrp, &tset);
2144 * step 5: success! and cleanup
2148 out_put_css_set_refs:
2150 for (i = 0; i < group_size; i++) {
2151 tc = flex_array_get(group, i);
2154 put_css_set(tc->cg);
2159 for_each_subsys(root, ss) {
2160 if (ss == failed_ss)
2162 if (ss->cancel_attach)
2163 ss->cancel_attach(cgrp, &tset);
2166 out_free_group_list:
2167 flex_array_free(group);
2172 * Find the task_struct of the task to attach by vpid and pass it along to the
2173 * function to attach either it or all tasks in its threadgroup. Will lock
2174 * cgroup_mutex and threadgroup; may take task_lock of task.
2176 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2178 struct task_struct *tsk;
2179 const struct cred *cred = current_cred(), *tcred;
2182 if (!cgroup_lock_live_group(cgrp))
2188 tsk = find_task_by_vpid(pid);
2192 goto out_unlock_cgroup;
2195 * even if we're attaching all tasks in the thread group, we
2196 * only need to check permissions on one of them.
2198 tcred = __task_cred(tsk);
2199 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2200 !uid_eq(cred->euid, tcred->uid) &&
2201 !uid_eq(cred->euid, tcred->suid)) {
2204 goto out_unlock_cgroup;
2210 tsk = tsk->group_leader;
2213 * Workqueue threads may acquire PF_THREAD_BOUND and become
2214 * trapped in a cpuset, or RT worker may be born in a cgroup
2215 * with no rt_runtime allocated. Just say no.
2217 if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2220 goto out_unlock_cgroup;
2223 get_task_struct(tsk);
2226 threadgroup_lock(tsk);
2228 if (!thread_group_leader(tsk)) {
2230 * a race with de_thread from another thread's exec()
2231 * may strip us of our leadership, if this happens,
2232 * there is no choice but to throw this task away and
2233 * try again; this is
2234 * "double-double-toil-and-trouble-check locking".
2236 threadgroup_unlock(tsk);
2237 put_task_struct(tsk);
2238 goto retry_find_task;
2240 ret = cgroup_attach_proc(cgrp, tsk);
2242 ret = cgroup_attach_task(cgrp, tsk);
2243 threadgroup_unlock(tsk);
2245 put_task_struct(tsk);
2251 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2253 return attach_task_by_pid(cgrp, pid, false);
2256 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2258 return attach_task_by_pid(cgrp, tgid, true);
2262 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2263 * @cgrp: the cgroup to be checked for liveness
2265 * On success, returns true; the lock should be later released with
2266 * cgroup_unlock(). On failure returns false with no lock held.
2268 bool cgroup_lock_live_group(struct cgroup *cgrp)
2270 mutex_lock(&cgroup_mutex);
2271 if (cgroup_is_removed(cgrp)) {
2272 mutex_unlock(&cgroup_mutex);
2277 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2279 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2282 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2283 if (strlen(buffer) >= PATH_MAX)
2285 if (!cgroup_lock_live_group(cgrp))
2287 mutex_lock(&cgroup_root_mutex);
2288 strcpy(cgrp->root->release_agent_path, buffer);
2289 mutex_unlock(&cgroup_root_mutex);
2294 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2295 struct seq_file *seq)
2297 if (!cgroup_lock_live_group(cgrp))
2299 seq_puts(seq, cgrp->root->release_agent_path);
2300 seq_putc(seq, '\n');
2305 /* A buffer size big enough for numbers or short strings */
2306 #define CGROUP_LOCAL_BUFFER_SIZE 64
2308 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2310 const char __user *userbuf,
2311 size_t nbytes, loff_t *unused_ppos)
2313 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2319 if (nbytes >= sizeof(buffer))
2321 if (copy_from_user(buffer, userbuf, nbytes))
2324 buffer[nbytes] = 0; /* nul-terminate */
2325 if (cft->write_u64) {
2326 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2329 retval = cft->write_u64(cgrp, cft, val);
2331 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2334 retval = cft->write_s64(cgrp, cft, val);
2341 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2343 const char __user *userbuf,
2344 size_t nbytes, loff_t *unused_ppos)
2346 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2348 size_t max_bytes = cft->max_write_len;
2349 char *buffer = local_buffer;
2352 max_bytes = sizeof(local_buffer) - 1;
2353 if (nbytes >= max_bytes)
2355 /* Allocate a dynamic buffer if we need one */
2356 if (nbytes >= sizeof(local_buffer)) {
2357 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2361 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2366 buffer[nbytes] = 0; /* nul-terminate */
2367 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2371 if (buffer != local_buffer)
2376 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2377 size_t nbytes, loff_t *ppos)
2379 struct cftype *cft = __d_cft(file->f_dentry);
2380 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2382 if (cgroup_is_removed(cgrp))
2385 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2386 if (cft->write_u64 || cft->write_s64)
2387 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2388 if (cft->write_string)
2389 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2391 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2392 return ret ? ret : nbytes;
2397 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2399 char __user *buf, size_t nbytes,
2402 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2403 u64 val = cft->read_u64(cgrp, cft);
2404 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2406 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2409 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2411 char __user *buf, size_t nbytes,
2414 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2415 s64 val = cft->read_s64(cgrp, cft);
2416 int len = sprintf(tmp, "%lld\n", (long long) val);
2418 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2421 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2422 size_t nbytes, loff_t *ppos)
2424 struct cftype *cft = __d_cft(file->f_dentry);
2425 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2427 if (cgroup_is_removed(cgrp))
2431 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2433 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2435 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2440 * seqfile ops/methods for returning structured data. Currently just
2441 * supports string->u64 maps, but can be extended in future.
2444 struct cgroup_seqfile_state {
2446 struct cgroup *cgroup;
2449 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2451 struct seq_file *sf = cb->state;
2452 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2455 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2457 struct cgroup_seqfile_state *state = m->private;
2458 struct cftype *cft = state->cft;
2459 if (cft->read_map) {
2460 struct cgroup_map_cb cb = {
2461 .fill = cgroup_map_add,
2464 return cft->read_map(state->cgroup, cft, &cb);
2466 return cft->read_seq_string(state->cgroup, cft, m);
2469 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2471 struct seq_file *seq = file->private_data;
2472 kfree(seq->private);
2473 return single_release(inode, file);
2476 static const struct file_operations cgroup_seqfile_operations = {
2478 .write = cgroup_file_write,
2479 .llseek = seq_lseek,
2480 .release = cgroup_seqfile_release,
2483 static int cgroup_file_open(struct inode *inode, struct file *file)
2488 err = generic_file_open(inode, file);
2491 cft = __d_cft(file->f_dentry);
2493 if (cft->read_map || cft->read_seq_string) {
2494 struct cgroup_seqfile_state *state =
2495 kzalloc(sizeof(*state), GFP_USER);
2499 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2500 file->f_op = &cgroup_seqfile_operations;
2501 err = single_open(file, cgroup_seqfile_show, state);
2504 } else if (cft->open)
2505 err = cft->open(inode, file);
2512 static int cgroup_file_release(struct inode *inode, struct file *file)
2514 struct cftype *cft = __d_cft(file->f_dentry);
2516 return cft->release(inode, file);
2521 * cgroup_rename - Only allow simple rename of directories in place.
2523 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2524 struct inode *new_dir, struct dentry *new_dentry)
2526 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2528 if (new_dentry->d_inode)
2530 if (old_dir != new_dir)
2532 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2535 static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2537 if (S_ISDIR(dentry->d_inode->i_mode))
2538 return &__d_cgrp(dentry)->xattrs;
2540 return &__d_cft(dentry)->xattrs;
2543 static inline int xattr_enabled(struct dentry *dentry)
2545 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2546 return test_bit(ROOT_XATTR, &root->flags);
2549 static bool is_valid_xattr(const char *name)
2551 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2552 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2557 static int cgroup_setxattr(struct dentry *dentry, const char *name,
2558 const void *val, size_t size, int flags)
2560 if (!xattr_enabled(dentry))
2562 if (!is_valid_xattr(name))
2564 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2567 static int cgroup_removexattr(struct dentry *dentry, const char *name)
2569 if (!xattr_enabled(dentry))
2571 if (!is_valid_xattr(name))
2573 return simple_xattr_remove(__d_xattrs(dentry), name);
2576 static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2577 void *buf, size_t size)
2579 if (!xattr_enabled(dentry))
2581 if (!is_valid_xattr(name))
2583 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2586 static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2588 if (!xattr_enabled(dentry))
2590 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2593 static const struct file_operations cgroup_file_operations = {
2594 .read = cgroup_file_read,
2595 .write = cgroup_file_write,
2596 .llseek = generic_file_llseek,
2597 .open = cgroup_file_open,
2598 .release = cgroup_file_release,
2601 static const struct inode_operations cgroup_file_inode_operations = {
2602 .setxattr = cgroup_setxattr,
2603 .getxattr = cgroup_getxattr,
2604 .listxattr = cgroup_listxattr,
2605 .removexattr = cgroup_removexattr,
2608 static const struct inode_operations cgroup_dir_inode_operations = {
2609 .lookup = cgroup_lookup,
2610 .mkdir = cgroup_mkdir,
2611 .rmdir = cgroup_rmdir,
2612 .rename = cgroup_rename,
2613 .setxattr = cgroup_setxattr,
2614 .getxattr = cgroup_getxattr,
2615 .listxattr = cgroup_listxattr,
2616 .removexattr = cgroup_removexattr,
2619 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2621 if (dentry->d_name.len > NAME_MAX)
2622 return ERR_PTR(-ENAMETOOLONG);
2623 d_add(dentry, NULL);
2628 * Check if a file is a control file
2630 static inline struct cftype *__file_cft(struct file *file)
2632 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2633 return ERR_PTR(-EINVAL);
2634 return __d_cft(file->f_dentry);
2637 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2638 struct super_block *sb)
2640 struct inode *inode;
2644 if (dentry->d_inode)
2647 inode = cgroup_new_inode(mode, sb);
2651 if (S_ISDIR(mode)) {
2652 inode->i_op = &cgroup_dir_inode_operations;
2653 inode->i_fop = &simple_dir_operations;
2655 /* start off with i_nlink == 2 (for "." entry) */
2657 inc_nlink(dentry->d_parent->d_inode);
2660 * Control reaches here with cgroup_mutex held.
2661 * @inode->i_mutex should nest outside cgroup_mutex but we
2662 * want to populate it immediately without releasing
2663 * cgroup_mutex. As @inode isn't visible to anyone else
2664 * yet, trylock will always succeed without affecting
2667 WARN_ON_ONCE(!mutex_trylock(&inode->i_mutex));
2668 } else if (S_ISREG(mode)) {
2670 inode->i_fop = &cgroup_file_operations;
2671 inode->i_op = &cgroup_file_inode_operations;
2673 d_instantiate(dentry, inode);
2674 dget(dentry); /* Extra count - pin the dentry in core */
2679 * cgroup_file_mode - deduce file mode of a control file
2680 * @cft: the control file in question
2682 * returns cft->mode if ->mode is not 0
2683 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2684 * returns S_IRUGO if it has only a read handler
2685 * returns S_IWUSR if it has only a write hander
2687 static umode_t cgroup_file_mode(const struct cftype *cft)
2694 if (cft->read || cft->read_u64 || cft->read_s64 ||
2695 cft->read_map || cft->read_seq_string)
2698 if (cft->write || cft->write_u64 || cft->write_s64 ||
2699 cft->write_string || cft->trigger)
2705 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2708 struct dentry *dir = cgrp->dentry;
2709 struct cgroup *parent = __d_cgrp(dir);
2710 struct dentry *dentry;
2714 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2716 simple_xattrs_init(&cft->xattrs);
2718 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2719 strcpy(name, subsys->name);
2722 strcat(name, cft->name);
2724 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2726 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2730 dentry = lookup_one_len(name, dir, strlen(name));
2731 if (IS_ERR(dentry)) {
2732 error = PTR_ERR(dentry);
2736 mode = cgroup_file_mode(cft);
2737 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2739 cfe->type = (void *)cft;
2740 cfe->dentry = dentry;
2741 dentry->d_fsdata = cfe;
2742 list_add_tail(&cfe->node, &parent->files);
2751 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2752 struct cftype cfts[], bool is_add)
2757 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2758 /* does cft->flags tell us to skip this file on @cgrp? */
2759 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2761 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2765 err = cgroup_add_file(cgrp, subsys, cft);
2767 err = cgroup_rm_file(cgrp, cft);
2769 pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2770 is_add ? "add" : "remove", cft->name, err);
2777 static DEFINE_MUTEX(cgroup_cft_mutex);
2779 static void cgroup_cfts_prepare(void)
2780 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2783 * Thanks to the entanglement with vfs inode locking, we can't walk
2784 * the existing cgroups under cgroup_mutex and create files.
2785 * Instead, we increment reference on all cgroups and build list of
2786 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2787 * exclusive access to the field.
2789 mutex_lock(&cgroup_cft_mutex);
2790 mutex_lock(&cgroup_mutex);
2793 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2794 struct cftype *cfts, bool is_add)
2795 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2798 struct cgroup *cgrp, *n;
2800 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2801 if (cfts && ss->root != &rootnode) {
2802 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2804 list_add_tail(&cgrp->cft_q_node, &pending);
2808 mutex_unlock(&cgroup_mutex);
2811 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2812 * files for all cgroups which were created before.
2814 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2815 struct inode *inode = cgrp->dentry->d_inode;
2817 mutex_lock(&inode->i_mutex);
2818 mutex_lock(&cgroup_mutex);
2819 if (!cgroup_is_removed(cgrp))
2820 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2821 mutex_unlock(&cgroup_mutex);
2822 mutex_unlock(&inode->i_mutex);
2824 list_del_init(&cgrp->cft_q_node);
2828 mutex_unlock(&cgroup_cft_mutex);
2832 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2833 * @ss: target cgroup subsystem
2834 * @cfts: zero-length name terminated array of cftypes
2836 * Register @cfts to @ss. Files described by @cfts are created for all
2837 * existing cgroups to which @ss is attached and all future cgroups will
2838 * have them too. This function can be called anytime whether @ss is
2841 * Returns 0 on successful registration, -errno on failure. Note that this
2842 * function currently returns 0 as long as @cfts registration is successful
2843 * even if some file creation attempts on existing cgroups fail.
2845 int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2847 struct cftype_set *set;
2849 set = kzalloc(sizeof(*set), GFP_KERNEL);
2853 cgroup_cfts_prepare();
2855 list_add_tail(&set->node, &ss->cftsets);
2856 cgroup_cfts_commit(ss, cfts, true);
2860 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2863 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2864 * @ss: target cgroup subsystem
2865 * @cfts: zero-length name terminated array of cftypes
2867 * Unregister @cfts from @ss. Files described by @cfts are removed from
2868 * all existing cgroups to which @ss is attached and all future cgroups
2869 * won't have them either. This function can be called anytime whether @ss
2870 * is attached or not.
2872 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2873 * registered with @ss.
2875 int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2877 struct cftype_set *set;
2879 cgroup_cfts_prepare();
2881 list_for_each_entry(set, &ss->cftsets, node) {
2882 if (set->cfts == cfts) {
2883 list_del_init(&set->node);
2884 cgroup_cfts_commit(ss, cfts, false);
2889 cgroup_cfts_commit(ss, NULL, false);
2894 * cgroup_task_count - count the number of tasks in a cgroup.
2895 * @cgrp: the cgroup in question
2897 * Return the number of tasks in the cgroup.
2899 int cgroup_task_count(const struct cgroup *cgrp)
2902 struct cg_cgroup_link *link;
2904 read_lock(&css_set_lock);
2905 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2906 count += atomic_read(&link->cg->refcount);
2908 read_unlock(&css_set_lock);
2913 * Advance a list_head iterator. The iterator should be positioned at
2914 * the start of a css_set
2916 static void cgroup_advance_iter(struct cgroup *cgrp,
2917 struct cgroup_iter *it)
2919 struct list_head *l = it->cg_link;
2920 struct cg_cgroup_link *link;
2923 /* Advance to the next non-empty css_set */
2926 if (l == &cgrp->css_sets) {
2930 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2932 } while (list_empty(&cg->tasks));
2934 it->task = cg->tasks.next;
2938 * To reduce the fork() overhead for systems that are not actually
2939 * using their cgroups capability, we don't maintain the lists running
2940 * through each css_set to its tasks until we see the list actually
2941 * used - in other words after the first call to cgroup_iter_start().
2943 static void cgroup_enable_task_cg_lists(void)
2945 struct task_struct *p, *g;
2946 write_lock(&css_set_lock);
2947 use_task_css_set_links = 1;
2949 * We need tasklist_lock because RCU is not safe against
2950 * while_each_thread(). Besides, a forking task that has passed
2951 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2952 * is not guaranteed to have its child immediately visible in the
2953 * tasklist if we walk through it with RCU.
2955 read_lock(&tasklist_lock);
2956 do_each_thread(g, p) {
2959 * We should check if the process is exiting, otherwise
2960 * it will race with cgroup_exit() in that the list
2961 * entry won't be deleted though the process has exited.
2963 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2964 list_add(&p->cg_list, &p->cgroups->tasks);
2966 } while_each_thread(g, p);
2967 read_unlock(&tasklist_lock);
2968 write_unlock(&css_set_lock);
2972 * cgroup_next_descendant_pre - find the next descendant for pre-order walk
2973 * @pos: the current position (%NULL to initiate traversal)
2974 * @cgroup: cgroup whose descendants to walk
2976 * To be used by cgroup_for_each_descendant_pre(). Find the next
2977 * descendant to visit for pre-order traversal of @cgroup's descendants.
2979 struct cgroup *cgroup_next_descendant_pre(struct cgroup *pos,
2980 struct cgroup *cgroup)
2982 struct cgroup *next;
2984 WARN_ON_ONCE(!rcu_read_lock_held());
2986 /* if first iteration, pretend we just visited @cgroup */
2988 if (list_empty(&cgroup->children))
2993 /* visit the first child if exists */
2994 next = list_first_or_null_rcu(&pos->children, struct cgroup, sibling);
2998 /* no child, visit my or the closest ancestor's next sibling */
3000 next = list_entry_rcu(pos->sibling.next, struct cgroup,
3002 if (&next->sibling != &pos->parent->children)
3006 } while (pos != cgroup);
3010 EXPORT_SYMBOL_GPL(cgroup_next_descendant_pre);
3013 * cgroup_rightmost_descendant - return the rightmost descendant of a cgroup
3014 * @pos: cgroup of interest
3016 * Return the rightmost descendant of @pos. If there's no descendant,
3017 * @pos is returned. This can be used during pre-order traversal to skip
3020 struct cgroup *cgroup_rightmost_descendant(struct cgroup *pos)
3022 struct cgroup *last, *tmp;
3024 WARN_ON_ONCE(!rcu_read_lock_held());
3028 /* ->prev isn't RCU safe, walk ->next till the end */
3030 list_for_each_entry_rcu(tmp, &last->children, sibling)
3036 EXPORT_SYMBOL_GPL(cgroup_rightmost_descendant);
3038 static struct cgroup *cgroup_leftmost_descendant(struct cgroup *pos)
3040 struct cgroup *last;
3044 pos = list_first_or_null_rcu(&pos->children, struct cgroup,
3052 * cgroup_next_descendant_post - find the next descendant for post-order walk
3053 * @pos: the current position (%NULL to initiate traversal)
3054 * @cgroup: cgroup whose descendants to walk
3056 * To be used by cgroup_for_each_descendant_post(). Find the next
3057 * descendant to visit for post-order traversal of @cgroup's descendants.
3059 struct cgroup *cgroup_next_descendant_post(struct cgroup *pos,
3060 struct cgroup *cgroup)
3062 struct cgroup *next;
3064 WARN_ON_ONCE(!rcu_read_lock_held());
3066 /* if first iteration, visit the leftmost descendant */
3068 next = cgroup_leftmost_descendant(cgroup);
3069 return next != cgroup ? next : NULL;
3072 /* if there's an unvisited sibling, visit its leftmost descendant */
3073 next = list_entry_rcu(pos->sibling.next, struct cgroup, sibling);
3074 if (&next->sibling != &pos->parent->children)
3075 return cgroup_leftmost_descendant(next);
3077 /* no sibling left, visit parent */
3079 return next != cgroup ? next : NULL;
3081 EXPORT_SYMBOL_GPL(cgroup_next_descendant_post);
3083 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
3084 __acquires(css_set_lock)
3087 * The first time anyone tries to iterate across a cgroup,
3088 * we need to enable the list linking each css_set to its
3089 * tasks, and fix up all existing tasks.
3091 if (!use_task_css_set_links)
3092 cgroup_enable_task_cg_lists();
3094 read_lock(&css_set_lock);
3095 it->cg_link = &cgrp->css_sets;
3096 cgroup_advance_iter(cgrp, it);
3099 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
3100 struct cgroup_iter *it)
3102 struct task_struct *res;
3103 struct list_head *l = it->task;
3104 struct cg_cgroup_link *link;
3106 /* If the iterator cg is NULL, we have no tasks */
3109 res = list_entry(l, struct task_struct, cg_list);
3110 /* Advance iterator to find next entry */
3112 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
3113 if (l == &link->cg->tasks) {
3114 /* We reached the end of this task list - move on to
3115 * the next cg_cgroup_link */
3116 cgroup_advance_iter(cgrp, it);
3123 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
3124 __releases(css_set_lock)
3126 read_unlock(&css_set_lock);
3129 static inline int started_after_time(struct task_struct *t1,
3130 struct timespec *time,
3131 struct task_struct *t2)
3133 int start_diff = timespec_compare(&t1->start_time, time);
3134 if (start_diff > 0) {
3136 } else if (start_diff < 0) {
3140 * Arbitrarily, if two processes started at the same
3141 * time, we'll say that the lower pointer value
3142 * started first. Note that t2 may have exited by now
3143 * so this may not be a valid pointer any longer, but
3144 * that's fine - it still serves to distinguish
3145 * between two tasks started (effectively) simultaneously.
3152 * This function is a callback from heap_insert() and is used to order
3154 * In this case we order the heap in descending task start time.
3156 static inline int started_after(void *p1, void *p2)
3158 struct task_struct *t1 = p1;
3159 struct task_struct *t2 = p2;
3160 return started_after_time(t1, &t2->start_time, t2);
3164 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3165 * @scan: struct cgroup_scanner containing arguments for the scan
3167 * Arguments include pointers to callback functions test_task() and
3169 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3170 * and if it returns true, call process_task() for it also.
3171 * The test_task pointer may be NULL, meaning always true (select all tasks).
3172 * Effectively duplicates cgroup_iter_{start,next,end}()
3173 * but does not lock css_set_lock for the call to process_task().
3174 * The struct cgroup_scanner may be embedded in any structure of the caller's
3176 * It is guaranteed that process_task() will act on every task that
3177 * is a member of the cgroup for the duration of this call. This
3178 * function may or may not call process_task() for tasks that exit
3179 * or move to a different cgroup during the call, or are forked or
3180 * move into the cgroup during the call.
3182 * Note that test_task() may be called with locks held, and may in some
3183 * situations be called multiple times for the same task, so it should
3185 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3186 * pre-allocated and will be used for heap operations (and its "gt" member will
3187 * be overwritten), else a temporary heap will be used (allocation of which
3188 * may cause this function to fail).
3190 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3193 struct cgroup_iter it;
3194 struct task_struct *p, *dropped;
3195 /* Never dereference latest_task, since it's not refcounted */
3196 struct task_struct *latest_task = NULL;
3197 struct ptr_heap tmp_heap;
3198 struct ptr_heap *heap;
3199 struct timespec latest_time = { 0, 0 };
3202 /* The caller supplied our heap and pre-allocated its memory */
3204 heap->gt = &started_after;
3206 /* We need to allocate our own heap memory */
3208 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3210 /* cannot allocate the heap */
3216 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3217 * to determine which are of interest, and using the scanner's
3218 * "process_task" callback to process any of them that need an update.
3219 * Since we don't want to hold any locks during the task updates,
3220 * gather tasks to be processed in a heap structure.
3221 * The heap is sorted by descending task start time.
3222 * If the statically-sized heap fills up, we overflow tasks that
3223 * started later, and in future iterations only consider tasks that
3224 * started after the latest task in the previous pass. This
3225 * guarantees forward progress and that we don't miss any tasks.
3228 cgroup_iter_start(scan->cg, &it);
3229 while ((p = cgroup_iter_next(scan->cg, &it))) {
3231 * Only affect tasks that qualify per the caller's callback,
3232 * if he provided one
3234 if (scan->test_task && !scan->test_task(p, scan))
3237 * Only process tasks that started after the last task
3240 if (!started_after_time(p, &latest_time, latest_task))
3242 dropped = heap_insert(heap, p);
3243 if (dropped == NULL) {
3245 * The new task was inserted; the heap wasn't
3249 } else if (dropped != p) {
3251 * The new task was inserted, and pushed out a
3255 put_task_struct(dropped);
3258 * Else the new task was newer than anything already in
3259 * the heap and wasn't inserted
3262 cgroup_iter_end(scan->cg, &it);
3265 for (i = 0; i < heap->size; i++) {
3266 struct task_struct *q = heap->ptrs[i];
3268 latest_time = q->start_time;
3271 /* Process the task per the caller's callback */
3272 scan->process_task(q, scan);
3276 * If we had to process any tasks at all, scan again
3277 * in case some of them were in the middle of forking
3278 * children that didn't get processed.
3279 * Not the most efficient way to do it, but it avoids
3280 * having to take callback_mutex in the fork path
3284 if (heap == &tmp_heap)
3285 heap_free(&tmp_heap);
3290 * Stuff for reading the 'tasks'/'procs' files.
3292 * Reading this file can return large amounts of data if a cgroup has
3293 * *lots* of attached tasks. So it may need several calls to read(),
3294 * but we cannot guarantee that the information we produce is correct
3295 * unless we produce it entirely atomically.
3299 /* which pidlist file are we talking about? */
3300 enum cgroup_filetype {
3306 * A pidlist is a list of pids that virtually represents the contents of one
3307 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3308 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3311 struct cgroup_pidlist {
3313 * used to find which pidlist is wanted. doesn't change as long as
3314 * this particular list stays in the list.
3316 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3319 /* how many elements the above list has */
3321 /* how many files are using the current array */
3323 /* each of these stored in a list by its cgroup */
3324 struct list_head links;
3325 /* pointer to the cgroup we belong to, for list removal purposes */
3326 struct cgroup *owner;
3327 /* protects the other fields */
3328 struct rw_semaphore mutex;
3332 * The following two functions "fix" the issue where there are more pids
3333 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3334 * TODO: replace with a kernel-wide solution to this problem
3336 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3337 static void *pidlist_allocate(int count)
3339 if (PIDLIST_TOO_LARGE(count))
3340 return vmalloc(count * sizeof(pid_t));
3342 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3344 static void pidlist_free(void *p)
3346 if (is_vmalloc_addr(p))
3351 static void *pidlist_resize(void *p, int newcount)
3354 /* note: if new alloc fails, old p will still be valid either way */
3355 if (is_vmalloc_addr(p)) {
3356 newlist = vmalloc(newcount * sizeof(pid_t));
3359 memcpy(newlist, p, newcount * sizeof(pid_t));
3362 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3368 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3369 * If the new stripped list is sufficiently smaller and there's enough memory
3370 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3371 * number of unique elements.
3373 /* is the size difference enough that we should re-allocate the array? */
3374 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3375 static int pidlist_uniq(pid_t **p, int length)
3382 * we presume the 0th element is unique, so i starts at 1. trivial
3383 * edge cases first; no work needs to be done for either
3385 if (length == 0 || length == 1)
3387 /* src and dest walk down the list; dest counts unique elements */
3388 for (src = 1; src < length; src++) {
3389 /* find next unique element */
3390 while (list[src] == list[src-1]) {
3395 /* dest always points to where the next unique element goes */
3396 list[dest] = list[src];
3401 * if the length difference is large enough, we want to allocate a
3402 * smaller buffer to save memory. if this fails due to out of memory,
3403 * we'll just stay with what we've got.
3405 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3406 newlist = pidlist_resize(list, dest);
3413 static int cmppid(const void *a, const void *b)
3415 return *(pid_t *)a - *(pid_t *)b;
3419 * find the appropriate pidlist for our purpose (given procs vs tasks)
3420 * returns with the lock on that pidlist already held, and takes care
3421 * of the use count, or returns NULL with no locks held if we're out of
3424 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3425 enum cgroup_filetype type)
3427 struct cgroup_pidlist *l;
3428 /* don't need task_nsproxy() if we're looking at ourself */
3429 struct pid_namespace *ns = task_active_pid_ns(current);
3432 * We can't drop the pidlist_mutex before taking the l->mutex in case
3433 * the last ref-holder is trying to remove l from the list at the same
3434 * time. Holding the pidlist_mutex precludes somebody taking whichever
3435 * list we find out from under us - compare release_pid_array().
3437 mutex_lock(&cgrp->pidlist_mutex);
3438 list_for_each_entry(l, &cgrp->pidlists, links) {
3439 if (l->key.type == type && l->key.ns == ns) {
3440 /* make sure l doesn't vanish out from under us */
3441 down_write(&l->mutex);
3442 mutex_unlock(&cgrp->pidlist_mutex);
3446 /* entry not found; create a new one */
3447 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3449 mutex_unlock(&cgrp->pidlist_mutex);
3452 init_rwsem(&l->mutex);
3453 down_write(&l->mutex);
3455 l->key.ns = get_pid_ns(ns);
3456 l->use_count = 0; /* don't increment here */
3459 list_add(&l->links, &cgrp->pidlists);
3460 mutex_unlock(&cgrp->pidlist_mutex);
3465 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3467 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3468 struct cgroup_pidlist **lp)
3472 int pid, n = 0; /* used for populating the array */
3473 struct cgroup_iter it;
3474 struct task_struct *tsk;
3475 struct cgroup_pidlist *l;
3478 * If cgroup gets more users after we read count, we won't have
3479 * enough space - tough. This race is indistinguishable to the
3480 * caller from the case that the additional cgroup users didn't
3481 * show up until sometime later on.
3483 length = cgroup_task_count(cgrp);
3484 array = pidlist_allocate(length);
3487 /* now, populate the array */
3488 cgroup_iter_start(cgrp, &it);
3489 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3490 if (unlikely(n == length))
3492 /* get tgid or pid for procs or tasks file respectively */
3493 if (type == CGROUP_FILE_PROCS)
3494 pid = task_tgid_vnr(tsk);
3496 pid = task_pid_vnr(tsk);
3497 if (pid > 0) /* make sure to only use valid results */
3500 cgroup_iter_end(cgrp, &it);
3502 /* now sort & (if procs) strip out duplicates */
3503 sort(array, length, sizeof(pid_t), cmppid, NULL);
3504 if (type == CGROUP_FILE_PROCS)
3505 length = pidlist_uniq(&array, length);
3506 l = cgroup_pidlist_find(cgrp, type);
3508 pidlist_free(array);
3511 /* store array, freeing old if necessary - lock already held */
3512 pidlist_free(l->list);
3516 up_write(&l->mutex);
3522 * cgroupstats_build - build and fill cgroupstats
3523 * @stats: cgroupstats to fill information into
3524 * @dentry: A dentry entry belonging to the cgroup for which stats have
3527 * Build and fill cgroupstats so that taskstats can export it to user
3530 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3533 struct cgroup *cgrp;
3534 struct cgroup_iter it;
3535 struct task_struct *tsk;
3538 * Validate dentry by checking the superblock operations,
3539 * and make sure it's a directory.
3541 if (dentry->d_sb->s_op != &cgroup_ops ||
3542 !S_ISDIR(dentry->d_inode->i_mode))
3546 cgrp = dentry->d_fsdata;
3548 cgroup_iter_start(cgrp, &it);
3549 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3550 switch (tsk->state) {
3552 stats->nr_running++;
3554 case TASK_INTERRUPTIBLE:
3555 stats->nr_sleeping++;
3557 case TASK_UNINTERRUPTIBLE:
3558 stats->nr_uninterruptible++;
3561 stats->nr_stopped++;
3564 if (delayacct_is_task_waiting_on_io(tsk))
3565 stats->nr_io_wait++;
3569 cgroup_iter_end(cgrp, &it);
3577 * seq_file methods for the tasks/procs files. The seq_file position is the
3578 * next pid to display; the seq_file iterator is a pointer to the pid
3579 * in the cgroup->l->list array.
3582 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3585 * Initially we receive a position value that corresponds to
3586 * one more than the last pid shown (or 0 on the first call or
3587 * after a seek to the start). Use a binary-search to find the
3588 * next pid to display, if any
3590 struct cgroup_pidlist *l = s->private;
3591 int index = 0, pid = *pos;
3594 down_read(&l->mutex);
3596 int end = l->length;
3598 while (index < end) {
3599 int mid = (index + end) / 2;
3600 if (l->list[mid] == pid) {
3603 } else if (l->list[mid] <= pid)
3609 /* If we're off the end of the array, we're done */
3610 if (index >= l->length)
3612 /* Update the abstract position to be the actual pid that we found */
3613 iter = l->list + index;
3618 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3620 struct cgroup_pidlist *l = s->private;
3624 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3626 struct cgroup_pidlist *l = s->private;
3628 pid_t *end = l->list + l->length;
3630 * Advance to the next pid in the array. If this goes off the
3642 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3644 return seq_printf(s, "%d\n", *(int *)v);
3648 * seq_operations functions for iterating on pidlists through seq_file -
3649 * independent of whether it's tasks or procs
3651 static const struct seq_operations cgroup_pidlist_seq_operations = {
3652 .start = cgroup_pidlist_start,
3653 .stop = cgroup_pidlist_stop,
3654 .next = cgroup_pidlist_next,
3655 .show = cgroup_pidlist_show,
3658 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3661 * the case where we're the last user of this particular pidlist will
3662 * have us remove it from the cgroup's list, which entails taking the
3663 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3664 * pidlist_mutex, we have to take pidlist_mutex first.
3666 mutex_lock(&l->owner->pidlist_mutex);
3667 down_write(&l->mutex);
3668 BUG_ON(!l->use_count);
3669 if (!--l->use_count) {
3670 /* we're the last user if refcount is 0; remove and free */
3671 list_del(&l->links);
3672 mutex_unlock(&l->owner->pidlist_mutex);
3673 pidlist_free(l->list);
3674 put_pid_ns(l->key.ns);
3675 up_write(&l->mutex);
3679 mutex_unlock(&l->owner->pidlist_mutex);
3680 up_write(&l->mutex);
3683 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3685 struct cgroup_pidlist *l;
3686 if (!(file->f_mode & FMODE_READ))
3689 * the seq_file will only be initialized if the file was opened for
3690 * reading; hence we check if it's not null only in that case.
3692 l = ((struct seq_file *)file->private_data)->private;
3693 cgroup_release_pid_array(l);
3694 return seq_release(inode, file);
3697 static const struct file_operations cgroup_pidlist_operations = {
3699 .llseek = seq_lseek,
3700 .write = cgroup_file_write,
3701 .release = cgroup_pidlist_release,
3705 * The following functions handle opens on a file that displays a pidlist
3706 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3709 /* helper function for the two below it */
3710 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3712 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3713 struct cgroup_pidlist *l;
3716 /* Nothing to do for write-only files */
3717 if (!(file->f_mode & FMODE_READ))
3720 /* have the array populated */
3721 retval = pidlist_array_load(cgrp, type, &l);
3724 /* configure file information */
3725 file->f_op = &cgroup_pidlist_operations;
3727 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3729 cgroup_release_pid_array(l);
3732 ((struct seq_file *)file->private_data)->private = l;
3735 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3737 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3739 static int cgroup_procs_open(struct inode *unused, struct file *file)
3741 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3744 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3747 return notify_on_release(cgrp);
3750 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3754 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3756 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3758 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3763 * Unregister event and free resources.
3765 * Gets called from workqueue.
3767 static void cgroup_event_remove(struct work_struct *work)
3769 struct cgroup_event *event = container_of(work, struct cgroup_event,
3771 struct cgroup *cgrp = event->cgrp;
3773 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3775 eventfd_ctx_put(event->eventfd);
3781 * Gets called on POLLHUP on eventfd when user closes it.
3783 * Called with wqh->lock held and interrupts disabled.
3785 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3786 int sync, void *key)
3788 struct cgroup_event *event = container_of(wait,
3789 struct cgroup_event, wait);
3790 struct cgroup *cgrp = event->cgrp;
3791 unsigned long flags = (unsigned long)key;
3793 if (flags & POLLHUP) {
3794 __remove_wait_queue(event->wqh, &event->wait);
3795 spin_lock(&cgrp->event_list_lock);
3796 list_del_init(&event->list);
3797 spin_unlock(&cgrp->event_list_lock);
3799 * We are in atomic context, but cgroup_event_remove() may
3800 * sleep, so we have to call it in workqueue.
3802 schedule_work(&event->remove);
3808 static void cgroup_event_ptable_queue_proc(struct file *file,
3809 wait_queue_head_t *wqh, poll_table *pt)
3811 struct cgroup_event *event = container_of(pt,
3812 struct cgroup_event, pt);
3815 add_wait_queue(wqh, &event->wait);
3819 * Parse input and register new cgroup event handler.
3821 * Input must be in format '<event_fd> <control_fd> <args>'.
3822 * Interpretation of args is defined by control file implementation.
3824 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3827 struct cgroup_event *event = NULL;
3828 unsigned int efd, cfd;
3829 struct file *efile = NULL;
3830 struct file *cfile = NULL;
3834 efd = simple_strtoul(buffer, &endp, 10);
3839 cfd = simple_strtoul(buffer, &endp, 10);
3840 if ((*endp != ' ') && (*endp != '\0'))
3844 event = kzalloc(sizeof(*event), GFP_KERNEL);
3848 INIT_LIST_HEAD(&event->list);
3849 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3850 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3851 INIT_WORK(&event->remove, cgroup_event_remove);
3853 efile = eventfd_fget(efd);
3854 if (IS_ERR(efile)) {
3855 ret = PTR_ERR(efile);
3859 event->eventfd = eventfd_ctx_fileget(efile);
3860 if (IS_ERR(event->eventfd)) {
3861 ret = PTR_ERR(event->eventfd);
3871 /* the process need read permission on control file */
3872 /* AV: shouldn't we check that it's been opened for read instead? */
3873 ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3877 event->cft = __file_cft(cfile);
3878 if (IS_ERR(event->cft)) {
3879 ret = PTR_ERR(event->cft);
3883 if (!event->cft->register_event || !event->cft->unregister_event) {
3888 ret = event->cft->register_event(cgrp, event->cft,
3889 event->eventfd, buffer);
3893 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3894 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3900 * Events should be removed after rmdir of cgroup directory, but before
3901 * destroying subsystem state objects. Let's take reference to cgroup
3902 * directory dentry to do that.
3906 spin_lock(&cgrp->event_list_lock);
3907 list_add(&event->list, &cgrp->event_list);
3908 spin_unlock(&cgrp->event_list_lock);
3919 if (event && event->eventfd && !IS_ERR(event->eventfd))
3920 eventfd_ctx_put(event->eventfd);
3922 if (!IS_ERR_OR_NULL(efile))
3930 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3933 return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3936 static int cgroup_clone_children_write(struct cgroup *cgrp,
3941 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3943 clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
3948 * for the common functions, 'private' gives the type of file
3950 /* for hysterical raisins, we can't put this on the older files */
3951 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3952 static struct cftype files[] = {
3955 .open = cgroup_tasks_open,
3956 .write_u64 = cgroup_tasks_write,
3957 .release = cgroup_pidlist_release,
3958 .mode = S_IRUGO | S_IWUSR,
3961 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3962 .open = cgroup_procs_open,
3963 .write_u64 = cgroup_procs_write,
3964 .release = cgroup_pidlist_release,
3965 .mode = S_IRUGO | S_IWUSR,
3968 .name = "notify_on_release",
3969 .read_u64 = cgroup_read_notify_on_release,
3970 .write_u64 = cgroup_write_notify_on_release,
3973 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3974 .write_string = cgroup_write_event_control,
3978 .name = "cgroup.clone_children",
3979 .read_u64 = cgroup_clone_children_read,
3980 .write_u64 = cgroup_clone_children_write,
3983 .name = "release_agent",
3984 .flags = CFTYPE_ONLY_ON_ROOT,
3985 .read_seq_string = cgroup_release_agent_show,
3986 .write_string = cgroup_release_agent_write,
3987 .max_write_len = PATH_MAX,
3993 * cgroup_populate_dir - selectively creation of files in a directory
3994 * @cgrp: target cgroup
3995 * @base_files: true if the base files should be added
3996 * @subsys_mask: mask of the subsystem ids whose files should be added
3998 static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
3999 unsigned long subsys_mask)
4002 struct cgroup_subsys *ss;
4005 err = cgroup_addrm_files(cgrp, NULL, files, true);
4010 /* process cftsets of each subsystem */
4011 for_each_subsys(cgrp->root, ss) {
4012 struct cftype_set *set;
4013 if (!test_bit(ss->subsys_id, &subsys_mask))
4016 list_for_each_entry(set, &ss->cftsets, node)
4017 cgroup_addrm_files(cgrp, ss, set->cfts, true);
4020 /* This cgroup is ready now */
4021 for_each_subsys(cgrp->root, ss) {
4022 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4024 * Update id->css pointer and make this css visible from
4025 * CSS ID functions. This pointer will be dereferened
4026 * from RCU-read-side without locks.
4029 rcu_assign_pointer(css->id->css, css);
4035 static void css_dput_fn(struct work_struct *work)
4037 struct cgroup_subsys_state *css =
4038 container_of(work, struct cgroup_subsys_state, dput_work);
4039 struct dentry *dentry = css->cgroup->dentry;
4040 struct super_block *sb = dentry->d_sb;
4042 atomic_inc(&sb->s_active);
4044 deactivate_super(sb);
4047 static void init_cgroup_css(struct cgroup_subsys_state *css,
4048 struct cgroup_subsys *ss,
4049 struct cgroup *cgrp)
4052 atomic_set(&css->refcnt, 1);
4055 if (cgrp == dummytop)
4056 css->flags |= CSS_ROOT;
4057 BUG_ON(cgrp->subsys[ss->subsys_id]);
4058 cgrp->subsys[ss->subsys_id] = css;
4061 * css holds an extra ref to @cgrp->dentry which is put on the last
4062 * css_put(). dput() requires process context, which css_put() may
4063 * be called without. @css->dput_work will be used to invoke
4064 * dput() asynchronously from css_put().
4066 INIT_WORK(&css->dput_work, css_dput_fn);
4069 /* invoke ->post_create() on a new CSS and mark it online if successful */
4070 static int online_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4074 lockdep_assert_held(&cgroup_mutex);
4077 ret = ss->css_online(cgrp);
4079 cgrp->subsys[ss->subsys_id]->flags |= CSS_ONLINE;
4083 /* if the CSS is online, invoke ->pre_destory() on it and mark it offline */
4084 static void offline_css(struct cgroup_subsys *ss, struct cgroup *cgrp)
4085 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4087 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4089 lockdep_assert_held(&cgroup_mutex);
4091 if (!(css->flags & CSS_ONLINE))
4095 * css_offline() should be called with cgroup_mutex unlocked. See
4096 * 3fa59dfbc3 ("cgroup: fix potential deadlock in pre_destroy") for
4097 * details. This temporary unlocking should go away once
4098 * cgroup_mutex is unexported from controllers.
4100 if (ss->css_offline) {
4101 mutex_unlock(&cgroup_mutex);
4102 ss->css_offline(cgrp);
4103 mutex_lock(&cgroup_mutex);
4106 cgrp->subsys[ss->subsys_id]->flags &= ~CSS_ONLINE;
4110 * cgroup_create - create a cgroup
4111 * @parent: cgroup that will be parent of the new cgroup
4112 * @dentry: dentry of the new cgroup
4113 * @mode: mode to set on new inode
4115 * Must be called with the mutex on the parent inode held
4117 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
4120 struct cgroup *cgrp;
4121 struct cgroupfs_root *root = parent->root;
4123 struct cgroup_subsys *ss;
4124 struct super_block *sb = root->sb;
4126 /* allocate the cgroup and its ID, 0 is reserved for the root */
4127 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
4131 cgrp->id = ida_simple_get(&root->cgroup_ida, 1, 0, GFP_KERNEL);
4136 * Only live parents can have children. Note that the liveliness
4137 * check isn't strictly necessary because cgroup_mkdir() and
4138 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
4139 * anyway so that locking is contained inside cgroup proper and we
4140 * don't get nasty surprises if we ever grow another caller.
4142 if (!cgroup_lock_live_group(parent)) {
4147 /* Grab a reference on the superblock so the hierarchy doesn't
4148 * get deleted on unmount if there are child cgroups. This
4149 * can be done outside cgroup_mutex, since the sb can't
4150 * disappear while someone has an open control file on the
4152 atomic_inc(&sb->s_active);
4154 init_cgroup_housekeeping(cgrp);
4156 cgrp->parent = parent;
4157 cgrp->root = parent->root;
4158 cgrp->top_cgroup = parent->top_cgroup;
4160 if (notify_on_release(parent))
4161 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
4163 if (test_bit(CGRP_CPUSET_CLONE_CHILDREN, &parent->flags))
4164 set_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags);
4166 for_each_subsys(root, ss) {
4167 struct cgroup_subsys_state *css;
4169 css = ss->css_alloc(cgrp);
4174 init_cgroup_css(css, ss, cgrp);
4176 err = alloc_css_id(ss, parent, cgrp);
4183 * Create directory. cgroup_create_file() returns with the new
4184 * directory locked on success so that it can be populated without
4185 * dropping cgroup_mutex.
4187 err = cgroup_create_file(dentry, S_IFDIR | mode, sb);
4190 lockdep_assert_held(&dentry->d_inode->i_mutex);
4192 /* allocation complete, commit to creation */
4193 dentry->d_fsdata = cgrp;
4194 cgrp->dentry = dentry;
4195 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
4196 list_add_tail_rcu(&cgrp->sibling, &cgrp->parent->children);
4197 root->number_of_cgroups++;
4199 /* each css holds a ref to the cgroup's dentry */
4200 for_each_subsys(root, ss)
4203 /* creation succeeded, notify subsystems */
4204 for_each_subsys(root, ss) {
4205 err = online_css(ss, cgrp);
4209 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4211 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",
4212 current->comm, current->pid, ss->name);
4213 if (!strcmp(ss->name, "memory"))
4214 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4215 ss->warned_broken_hierarchy = true;
4219 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
4223 mutex_unlock(&cgroup_mutex);
4224 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
4229 for_each_subsys(root, ss) {
4230 if (cgrp->subsys[ss->subsys_id])
4233 mutex_unlock(&cgroup_mutex);
4234 /* Release the reference count that we took on the superblock */
4235 deactivate_super(sb);
4237 ida_simple_remove(&root->cgroup_ida, cgrp->id);
4243 cgroup_destroy_locked(cgrp);
4244 mutex_unlock(&cgroup_mutex);
4245 mutex_unlock(&dentry->d_inode->i_mutex);
4249 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4251 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4253 /* the vfs holds inode->i_mutex already */
4254 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4258 * Check the reference count on each subsystem. Since we already
4259 * established that there are no tasks in the cgroup, if the css refcount
4260 * is also 1, then there should be no outstanding references, so the
4261 * subsystem is safe to destroy. We scan across all subsystems rather than
4262 * using the per-hierarchy linked list of mounted subsystems since we can
4263 * be called via check_for_release() with no synchronization other than
4264 * RCU, and the subsystem linked list isn't RCU-safe.
4266 static int cgroup_has_css_refs(struct cgroup *cgrp)
4271 * We won't need to lock the subsys array, because the subsystems
4272 * we're concerned about aren't going anywhere since our cgroup root
4273 * has a reference on them.
4275 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4276 struct cgroup_subsys *ss = subsys[i];
4277 struct cgroup_subsys_state *css;
4279 /* Skip subsystems not present or not in this hierarchy */
4280 if (ss == NULL || ss->root != cgrp->root)
4283 css = cgrp->subsys[ss->subsys_id];
4285 * When called from check_for_release() it's possible
4286 * that by this point the cgroup has been removed
4287 * and the css deleted. But a false-positive doesn't
4288 * matter, since it can only happen if the cgroup
4289 * has been deleted and hence no longer needs the
4290 * release agent to be called anyway.
4292 if (css && css_refcnt(css) > 1)
4298 static int cgroup_destroy_locked(struct cgroup *cgrp)
4299 __releases(&cgroup_mutex) __acquires(&cgroup_mutex)
4301 struct dentry *d = cgrp->dentry;
4302 struct cgroup *parent = cgrp->parent;
4304 struct cgroup_event *event, *tmp;
4305 struct cgroup_subsys *ss;
4306 LIST_HEAD(tmp_list);
4308 lockdep_assert_held(&d->d_inode->i_mutex);
4309 lockdep_assert_held(&cgroup_mutex);
4311 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children))
4315 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4316 * removed. This makes future css_tryget() and child creation
4317 * attempts fail thus maintaining the removal conditions verified
4320 for_each_subsys(cgrp->root, ss) {
4321 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4323 WARN_ON(atomic_read(&css->refcnt) < 0);
4324 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4326 set_bit(CGRP_REMOVED, &cgrp->flags);
4328 /* tell subsystems to initate destruction */
4329 for_each_subsys(cgrp->root, ss)
4330 offline_css(ss, cgrp);
4333 * Put all the base refs. Each css holds an extra reference to the
4334 * cgroup's dentry and cgroup removal proceeds regardless of css
4335 * refs. On the last put of each css, whenever that may be, the
4336 * extra dentry ref is put so that dentry destruction happens only
4337 * after all css's are released.
4339 for_each_subsys(cgrp->root, ss)
4340 css_put(cgrp->subsys[ss->subsys_id]);
4342 raw_spin_lock(&release_list_lock);
4343 if (!list_empty(&cgrp->release_list))
4344 list_del_init(&cgrp->release_list);
4345 raw_spin_unlock(&release_list_lock);
4347 /* delete this cgroup from parent->children */
4348 list_del_rcu(&cgrp->sibling);
4349 list_del_init(&cgrp->allcg_node);
4352 cgroup_d_remove_dir(d);
4355 set_bit(CGRP_RELEASABLE, &parent->flags);
4356 check_for_release(parent);
4359 * Unregister events and notify userspace.
4360 * Notify userspace about cgroup removing only after rmdir of cgroup
4361 * directory to avoid race between userspace and kernelspace. Use
4362 * a temporary list to avoid a deadlock with cgroup_event_wake(). Since
4363 * cgroup_event_wake() is called with the wait queue head locked,
4364 * remove_wait_queue() cannot be called while holding event_list_lock.
4366 spin_lock(&cgrp->event_list_lock);
4367 list_splice_init(&cgrp->event_list, &tmp_list);
4368 spin_unlock(&cgrp->event_list_lock);
4369 list_for_each_entry_safe(event, tmp, &tmp_list, list) {
4370 list_del_init(&event->list);
4371 remove_wait_queue(event->wqh, &event->wait);
4372 eventfd_signal(event->eventfd, 1);
4373 schedule_work(&event->remove);
4379 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4383 mutex_lock(&cgroup_mutex);
4384 ret = cgroup_destroy_locked(dentry->d_fsdata);
4385 mutex_unlock(&cgroup_mutex);
4390 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4392 INIT_LIST_HEAD(&ss->cftsets);
4395 * base_cftset is embedded in subsys itself, no need to worry about
4398 if (ss->base_cftypes) {
4399 ss->base_cftset.cfts = ss->base_cftypes;
4400 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4404 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4406 struct cgroup_subsys_state *css;
4408 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4410 mutex_lock(&cgroup_mutex);
4412 /* init base cftset */
4413 cgroup_init_cftsets(ss);
4415 /* Create the top cgroup state for this subsystem */
4416 list_add(&ss->sibling, &rootnode.subsys_list);
4417 ss->root = &rootnode;
4418 css = ss->css_alloc(dummytop);
4419 /* We don't handle early failures gracefully */
4420 BUG_ON(IS_ERR(css));
4421 init_cgroup_css(css, ss, dummytop);
4423 /* Update the init_css_set to contain a subsys
4424 * pointer to this state - since the subsystem is
4425 * newly registered, all tasks and hence the
4426 * init_css_set is in the subsystem's top cgroup. */
4427 init_css_set.subsys[ss->subsys_id] = css;
4429 need_forkexit_callback |= ss->fork || ss->exit;
4431 /* At system boot, before all subsystems have been
4432 * registered, no tasks have been forked, so we don't
4433 * need to invoke fork callbacks here. */
4434 BUG_ON(!list_empty(&init_task.tasks));
4437 BUG_ON(online_css(ss, dummytop));
4439 mutex_unlock(&cgroup_mutex);
4441 /* this function shouldn't be used with modular subsystems, since they
4442 * need to register a subsys_id, among other things */
4447 * cgroup_load_subsys: load and register a modular subsystem at runtime
4448 * @ss: the subsystem to load
4450 * This function should be called in a modular subsystem's initcall. If the
4451 * subsystem is built as a module, it will be assigned a new subsys_id and set
4452 * up for use. If the subsystem is built-in anyway, work is delegated to the
4453 * simpler cgroup_init_subsys.
4455 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4457 struct cgroup_subsys_state *css;
4459 struct hlist_node *node, *tmp;
4463 /* check name and function validity */
4464 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4465 ss->css_alloc == NULL || ss->css_free == NULL)
4469 * we don't support callbacks in modular subsystems. this check is
4470 * before the ss->module check for consistency; a subsystem that could
4471 * be a module should still have no callbacks even if the user isn't
4472 * compiling it as one.
4474 if (ss->fork || ss->exit)
4478 * an optionally modular subsystem is built-in: we want to do nothing,
4479 * since cgroup_init_subsys will have already taken care of it.
4481 if (ss->module == NULL) {
4482 /* a sanity check */
4483 BUG_ON(subsys[ss->subsys_id] != ss);
4487 /* init base cftset */
4488 cgroup_init_cftsets(ss);
4490 mutex_lock(&cgroup_mutex);
4491 subsys[ss->subsys_id] = ss;
4494 * no ss->css_alloc seems to need anything important in the ss
4495 * struct, so this can happen first (i.e. before the rootnode
4498 css = ss->css_alloc(dummytop);
4500 /* failure case - need to deassign the subsys[] slot. */
4501 subsys[ss->subsys_id] = NULL;
4502 mutex_unlock(&cgroup_mutex);
4503 return PTR_ERR(css);
4506 list_add(&ss->sibling, &rootnode.subsys_list);
4507 ss->root = &rootnode;
4509 /* our new subsystem will be attached to the dummy hierarchy. */
4510 init_cgroup_css(css, ss, dummytop);
4511 /* init_idr must be after init_cgroup_css because it sets css->id. */
4513 ret = cgroup_init_idr(ss, css);
4519 * Now we need to entangle the css into the existing css_sets. unlike
4520 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4521 * will need a new pointer to it; done by iterating the css_set_table.
4522 * furthermore, modifying the existing css_sets will corrupt the hash
4523 * table state, so each changed css_set will need its hash recomputed.
4524 * this is all done under the css_set_lock.
4526 write_lock(&css_set_lock);
4527 hash_for_each_safe(css_set_table, i, node, tmp, cg, hlist) {
4528 /* skip entries that we already rehashed */
4529 if (cg->subsys[ss->subsys_id])
4531 /* remove existing entry */
4532 hash_del(&cg->hlist);
4534 cg->subsys[ss->subsys_id] = css;
4535 /* recompute hash and restore entry */
4536 key = css_set_hash(cg->subsys);
4537 hash_add(css_set_table, node, key);
4539 write_unlock(&css_set_lock);
4542 ret = online_css(ss, dummytop);
4547 mutex_unlock(&cgroup_mutex);
4551 mutex_unlock(&cgroup_mutex);
4552 /* @ss can't be mounted here as try_module_get() would fail */
4553 cgroup_unload_subsys(ss);
4556 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4559 * cgroup_unload_subsys: unload a modular subsystem
4560 * @ss: the subsystem to unload
4562 * This function should be called in a modular subsystem's exitcall. When this
4563 * function is invoked, the refcount on the subsystem's module will be 0, so
4564 * the subsystem will not be attached to any hierarchy.
4566 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4568 struct cg_cgroup_link *link;
4570 BUG_ON(ss->module == NULL);
4573 * we shouldn't be called if the subsystem is in use, and the use of
4574 * try_module_get in parse_cgroupfs_options should ensure that it
4575 * doesn't start being used while we're killing it off.
4577 BUG_ON(ss->root != &rootnode);
4579 mutex_lock(&cgroup_mutex);
4581 offline_css(ss, dummytop);
4585 idr_remove_all(&ss->idr);
4586 idr_destroy(&ss->idr);
4589 /* deassign the subsys_id */
4590 subsys[ss->subsys_id] = NULL;
4592 /* remove subsystem from rootnode's list of subsystems */
4593 list_del_init(&ss->sibling);
4596 * disentangle the css from all css_sets attached to the dummytop. as
4597 * in loading, we need to pay our respects to the hashtable gods.
4599 write_lock(&css_set_lock);
4600 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4601 struct css_set *cg = link->cg;
4604 hash_del(&cg->hlist);
4605 cg->subsys[ss->subsys_id] = NULL;
4606 key = css_set_hash(cg->subsys);
4607 hash_add(css_set_table, &cg->hlist, key);
4609 write_unlock(&css_set_lock);
4612 * remove subsystem's css from the dummytop and free it - need to
4613 * free before marking as null because ss->css_free needs the
4614 * cgrp->subsys pointer to find their state. note that this also
4615 * takes care of freeing the css_id.
4617 ss->css_free(dummytop);
4618 dummytop->subsys[ss->subsys_id] = NULL;
4620 mutex_unlock(&cgroup_mutex);
4622 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4625 * cgroup_init_early - cgroup initialization at system boot
4627 * Initialize cgroups at system boot, and initialize any
4628 * subsystems that request early init.
4630 int __init cgroup_init_early(void)
4633 atomic_set(&init_css_set.refcount, 1);
4634 INIT_LIST_HEAD(&init_css_set.cg_links);
4635 INIT_LIST_HEAD(&init_css_set.tasks);
4636 INIT_HLIST_NODE(&init_css_set.hlist);
4638 init_cgroup_root(&rootnode);
4640 init_task.cgroups = &init_css_set;
4642 init_css_set_link.cg = &init_css_set;
4643 init_css_set_link.cgrp = dummytop;
4644 list_add(&init_css_set_link.cgrp_link_list,
4645 &rootnode.top_cgroup.css_sets);
4646 list_add(&init_css_set_link.cg_link_list,
4647 &init_css_set.cg_links);
4649 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4650 struct cgroup_subsys *ss = subsys[i];
4652 /* at bootup time, we don't worry about modular subsystems */
4653 if (!ss || ss->module)
4657 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4658 BUG_ON(!ss->css_alloc);
4659 BUG_ON(!ss->css_free);
4660 if (ss->subsys_id != i) {
4661 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4662 ss->name, ss->subsys_id);
4667 cgroup_init_subsys(ss);
4673 * cgroup_init - cgroup initialization
4675 * Register cgroup filesystem and /proc file, and initialize
4676 * any subsystems that didn't request early init.
4678 int __init cgroup_init(void)
4684 err = bdi_init(&cgroup_backing_dev_info);
4688 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4689 struct cgroup_subsys *ss = subsys[i];
4691 /* at bootup time, we don't worry about modular subsystems */
4692 if (!ss || ss->module)
4694 if (!ss->early_init)
4695 cgroup_init_subsys(ss);
4697 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4700 /* Add init_css_set to the hash table */
4701 key = css_set_hash(init_css_set.subsys);
4702 hash_add(css_set_table, &init_css_set.hlist, key);
4703 BUG_ON(!init_root_id(&rootnode));
4705 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4711 err = register_filesystem(&cgroup_fs_type);
4713 kobject_put(cgroup_kobj);
4717 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4721 bdi_destroy(&cgroup_backing_dev_info);
4727 * proc_cgroup_show()
4728 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4729 * - Used for /proc/<pid>/cgroup.
4730 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4731 * doesn't really matter if tsk->cgroup changes after we read it,
4732 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4733 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4734 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4735 * cgroup to top_cgroup.
4738 /* TODO: Use a proper seq_file iterator */
4739 static int proc_cgroup_show(struct seq_file *m, void *v)
4742 struct task_struct *tsk;
4745 struct cgroupfs_root *root;
4748 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4754 tsk = get_pid_task(pid, PIDTYPE_PID);
4760 mutex_lock(&cgroup_mutex);
4762 for_each_active_root(root) {
4763 struct cgroup_subsys *ss;
4764 struct cgroup *cgrp;
4767 seq_printf(m, "%d:", root->hierarchy_id);
4768 for_each_subsys(root, ss)
4769 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4770 if (strlen(root->name))
4771 seq_printf(m, "%sname=%s", count ? "," : "",
4774 cgrp = task_cgroup_from_root(tsk, root);
4775 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4783 mutex_unlock(&cgroup_mutex);
4784 put_task_struct(tsk);
4791 static int cgroup_open(struct inode *inode, struct file *file)
4793 struct pid *pid = PROC_I(inode)->pid;
4794 return single_open(file, proc_cgroup_show, pid);
4797 const struct file_operations proc_cgroup_operations = {
4798 .open = cgroup_open,
4800 .llseek = seq_lseek,
4801 .release = single_release,
4804 /* Display information about each subsystem and each hierarchy */
4805 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4809 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4811 * ideally we don't want subsystems moving around while we do this.
4812 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4813 * subsys/hierarchy state.
4815 mutex_lock(&cgroup_mutex);
4816 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4817 struct cgroup_subsys *ss = subsys[i];
4820 seq_printf(m, "%s\t%d\t%d\t%d\n",
4821 ss->name, ss->root->hierarchy_id,
4822 ss->root->number_of_cgroups, !ss->disabled);
4824 mutex_unlock(&cgroup_mutex);
4828 static int cgroupstats_open(struct inode *inode, struct file *file)
4830 return single_open(file, proc_cgroupstats_show, NULL);
4833 static const struct file_operations proc_cgroupstats_operations = {
4834 .open = cgroupstats_open,
4836 .llseek = seq_lseek,
4837 .release = single_release,
4841 * cgroup_fork - attach newly forked task to its parents cgroup.
4842 * @child: pointer to task_struct of forking parent process.
4844 * Description: A task inherits its parent's cgroup at fork().
4846 * A pointer to the shared css_set was automatically copied in
4847 * fork.c by dup_task_struct(). However, we ignore that copy, since
4848 * it was not made under the protection of RCU or cgroup_mutex, so
4849 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4850 * have already changed current->cgroups, allowing the previously
4851 * referenced cgroup group to be removed and freed.
4853 * At the point that cgroup_fork() is called, 'current' is the parent
4854 * task, and the passed argument 'child' points to the child task.
4856 void cgroup_fork(struct task_struct *child)
4859 child->cgroups = current->cgroups;
4860 get_css_set(child->cgroups);
4861 task_unlock(current);
4862 INIT_LIST_HEAD(&child->cg_list);
4866 * cgroup_post_fork - called on a new task after adding it to the task list
4867 * @child: the task in question
4869 * Adds the task to the list running through its css_set if necessary and
4870 * call the subsystem fork() callbacks. Has to be after the task is
4871 * visible on the task list in case we race with the first call to
4872 * cgroup_iter_start() - to guarantee that the new task ends up on its
4875 void cgroup_post_fork(struct task_struct *child)
4880 * use_task_css_set_links is set to 1 before we walk the tasklist
4881 * under the tasklist_lock and we read it here after we added the child
4882 * to the tasklist under the tasklist_lock as well. If the child wasn't
4883 * yet in the tasklist when we walked through it from
4884 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4885 * should be visible now due to the paired locking and barriers implied
4886 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4887 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4890 if (use_task_css_set_links) {
4891 write_lock(&css_set_lock);
4893 if (list_empty(&child->cg_list))
4894 list_add(&child->cg_list, &child->cgroups->tasks);
4896 write_unlock(&css_set_lock);
4900 * Call ss->fork(). This must happen after @child is linked on
4901 * css_set; otherwise, @child might change state between ->fork()
4902 * and addition to css_set.
4904 if (need_forkexit_callback) {
4905 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4906 struct cgroup_subsys *ss = subsys[i];
4909 * fork/exit callbacks are supported only for
4910 * builtin subsystems and we don't need further
4911 * synchronization as they never go away.
4913 if (!ss || ss->module)
4923 * cgroup_exit - detach cgroup from exiting task
4924 * @tsk: pointer to task_struct of exiting process
4925 * @run_callback: run exit callbacks?
4927 * Description: Detach cgroup from @tsk and release it.
4929 * Note that cgroups marked notify_on_release force every task in
4930 * them to take the global cgroup_mutex mutex when exiting.
4931 * This could impact scaling on very large systems. Be reluctant to
4932 * use notify_on_release cgroups where very high task exit scaling
4933 * is required on large systems.
4935 * the_top_cgroup_hack:
4937 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4939 * We call cgroup_exit() while the task is still competent to
4940 * handle notify_on_release(), then leave the task attached to the
4941 * root cgroup in each hierarchy for the remainder of its exit.
4943 * To do this properly, we would increment the reference count on
4944 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4945 * code we would add a second cgroup function call, to drop that
4946 * reference. This would just create an unnecessary hot spot on
4947 * the top_cgroup reference count, to no avail.
4949 * Normally, holding a reference to a cgroup without bumping its
4950 * count is unsafe. The cgroup could go away, or someone could
4951 * attach us to a different cgroup, decrementing the count on
4952 * the first cgroup that we never incremented. But in this case,
4953 * top_cgroup isn't going away, and either task has PF_EXITING set,
4954 * which wards off any cgroup_attach_task() attempts, or task is a failed
4955 * fork, never visible to cgroup_attach_task.
4957 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4963 * Unlink from the css_set task list if necessary.
4964 * Optimistically check cg_list before taking
4967 if (!list_empty(&tsk->cg_list)) {
4968 write_lock(&css_set_lock);
4969 if (!list_empty(&tsk->cg_list))
4970 list_del_init(&tsk->cg_list);
4971 write_unlock(&css_set_lock);
4974 /* Reassign the task to the init_css_set. */
4977 tsk->cgroups = &init_css_set;
4979 if (run_callbacks && need_forkexit_callback) {
4980 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4981 struct cgroup_subsys *ss = subsys[i];
4983 /* modular subsystems can't use callbacks */
4984 if (!ss || ss->module)
4988 struct cgroup *old_cgrp =
4989 rcu_dereference_raw(cg->subsys[i])->cgroup;
4990 struct cgroup *cgrp = task_cgroup(tsk, i);
4991 ss->exit(cgrp, old_cgrp, tsk);
4998 put_css_set_taskexit(cg);
5002 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
5003 * @cgrp: the cgroup in question
5004 * @task: the task in question
5006 * See if @cgrp is a descendant of @task's cgroup in the appropriate
5009 * If we are sending in dummytop, then presumably we are creating
5010 * the top cgroup in the subsystem.
5012 * Called only by the ns (nsproxy) cgroup.
5014 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
5017 struct cgroup *target;
5019 if (cgrp == dummytop)
5022 target = task_cgroup_from_root(task, cgrp->root);
5023 while (cgrp != target && cgrp!= cgrp->top_cgroup)
5024 cgrp = cgrp->parent;
5025 ret = (cgrp == target);
5029 static void check_for_release(struct cgroup *cgrp)
5031 /* All of these checks rely on RCU to keep the cgroup
5032 * structure alive */
5033 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
5034 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
5035 /* Control Group is currently removeable. If it's not
5036 * already queued for a userspace notification, queue
5038 int need_schedule_work = 0;
5039 raw_spin_lock(&release_list_lock);
5040 if (!cgroup_is_removed(cgrp) &&
5041 list_empty(&cgrp->release_list)) {
5042 list_add(&cgrp->release_list, &release_list);
5043 need_schedule_work = 1;
5045 raw_spin_unlock(&release_list_lock);
5046 if (need_schedule_work)
5047 schedule_work(&release_agent_work);
5051 /* Caller must verify that the css is not for root cgroup */
5052 bool __css_tryget(struct cgroup_subsys_state *css)
5057 v = css_refcnt(css);
5058 t = atomic_cmpxchg(&css->refcnt, v, v + 1);
5066 EXPORT_SYMBOL_GPL(__css_tryget);
5068 /* Caller must verify that the css is not for root cgroup */
5069 void __css_put(struct cgroup_subsys_state *css)
5071 struct cgroup *cgrp = css->cgroup;
5075 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
5079 if (notify_on_release(cgrp)) {
5080 set_bit(CGRP_RELEASABLE, &cgrp->flags);
5081 check_for_release(cgrp);
5085 schedule_work(&css->dput_work);
5090 EXPORT_SYMBOL_GPL(__css_put);
5093 * Notify userspace when a cgroup is released, by running the
5094 * configured release agent with the name of the cgroup (path
5095 * relative to the root of cgroup file system) as the argument.
5097 * Most likely, this user command will try to rmdir this cgroup.
5099 * This races with the possibility that some other task will be
5100 * attached to this cgroup before it is removed, or that some other
5101 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5102 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5103 * unused, and this cgroup will be reprieved from its death sentence,
5104 * to continue to serve a useful existence. Next time it's released,
5105 * we will get notified again, if it still has 'notify_on_release' set.
5107 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5108 * means only wait until the task is successfully execve()'d. The
5109 * separate release agent task is forked by call_usermodehelper(),
5110 * then control in this thread returns here, without waiting for the
5111 * release agent task. We don't bother to wait because the caller of
5112 * this routine has no use for the exit status of the release agent
5113 * task, so no sense holding our caller up for that.
5115 static void cgroup_release_agent(struct work_struct *work)
5117 BUG_ON(work != &release_agent_work);
5118 mutex_lock(&cgroup_mutex);
5119 raw_spin_lock(&release_list_lock);
5120 while (!list_empty(&release_list)) {
5121 char *argv[3], *envp[3];
5123 char *pathbuf = NULL, *agentbuf = NULL;
5124 struct cgroup *cgrp = list_entry(release_list.next,
5127 list_del_init(&cgrp->release_list);
5128 raw_spin_unlock(&release_list_lock);
5129 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
5132 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
5134 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
5139 argv[i++] = agentbuf;
5140 argv[i++] = pathbuf;
5144 /* minimal command environment */
5145 envp[i++] = "HOME=/";
5146 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5149 /* Drop the lock while we invoke the usermode helper,
5150 * since the exec could involve hitting disk and hence
5151 * be a slow process */
5152 mutex_unlock(&cgroup_mutex);
5153 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
5154 mutex_lock(&cgroup_mutex);
5158 raw_spin_lock(&release_list_lock);
5160 raw_spin_unlock(&release_list_lock);
5161 mutex_unlock(&cgroup_mutex);
5164 static int __init cgroup_disable(char *str)
5169 while ((token = strsep(&str, ",")) != NULL) {
5172 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5173 struct cgroup_subsys *ss = subsys[i];
5176 * cgroup_disable, being at boot time, can't
5177 * know about module subsystems, so we don't
5180 if (!ss || ss->module)
5183 if (!strcmp(token, ss->name)) {
5185 printk(KERN_INFO "Disabling %s control group"
5186 " subsystem\n", ss->name);
5193 __setup("cgroup_disable=", cgroup_disable);
5196 * Functons for CSS ID.
5200 *To get ID other than 0, this should be called when !cgroup_is_removed().
5202 unsigned short css_id(struct cgroup_subsys_state *css)
5204 struct css_id *cssid;
5207 * This css_id() can return correct value when somone has refcnt
5208 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5209 * it's unchanged until freed.
5211 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5217 EXPORT_SYMBOL_GPL(css_id);
5219 unsigned short css_depth(struct cgroup_subsys_state *css)
5221 struct css_id *cssid;
5223 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5226 return cssid->depth;
5229 EXPORT_SYMBOL_GPL(css_depth);
5232 * css_is_ancestor - test "root" css is an ancestor of "child"
5233 * @child: the css to be tested.
5234 * @root: the css supporsed to be an ancestor of the child.
5236 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5237 * this function reads css->id, the caller must hold rcu_read_lock().
5238 * But, considering usual usage, the csses should be valid objects after test.
5239 * Assuming that the caller will do some action to the child if this returns
5240 * returns true, the caller must take "child";s reference count.
5241 * If "child" is valid object and this returns true, "root" is valid, too.
5244 bool css_is_ancestor(struct cgroup_subsys_state *child,
5245 const struct cgroup_subsys_state *root)
5247 struct css_id *child_id;
5248 struct css_id *root_id;
5250 child_id = rcu_dereference(child->id);
5253 root_id = rcu_dereference(root->id);
5256 if (child_id->depth < root_id->depth)
5258 if (child_id->stack[root_id->depth] != root_id->id)
5263 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5265 struct css_id *id = css->id;
5266 /* When this is called before css_id initialization, id can be NULL */
5270 BUG_ON(!ss->use_id);
5272 rcu_assign_pointer(id->css, NULL);
5273 rcu_assign_pointer(css->id, NULL);
5274 spin_lock(&ss->id_lock);
5275 idr_remove(&ss->idr, id->id);
5276 spin_unlock(&ss->id_lock);
5277 kfree_rcu(id, rcu_head);
5279 EXPORT_SYMBOL_GPL(free_css_id);
5282 * This is called by init or create(). Then, calls to this function are
5283 * always serialized (By cgroup_mutex() at create()).
5286 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5288 struct css_id *newid;
5289 int myid, error, size;
5291 BUG_ON(!ss->use_id);
5293 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5294 newid = kzalloc(size, GFP_KERNEL);
5296 return ERR_PTR(-ENOMEM);
5298 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
5302 spin_lock(&ss->id_lock);
5303 /* Don't use 0. allocates an ID of 1-65535 */
5304 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
5305 spin_unlock(&ss->id_lock);
5307 /* Returns error when there are no free spaces for new ID.*/
5312 if (myid > CSS_ID_MAX)
5316 newid->depth = depth;
5320 spin_lock(&ss->id_lock);
5321 idr_remove(&ss->idr, myid);
5322 spin_unlock(&ss->id_lock);
5325 return ERR_PTR(error);
5329 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5330 struct cgroup_subsys_state *rootcss)
5332 struct css_id *newid;
5334 spin_lock_init(&ss->id_lock);
5337 newid = get_new_cssid(ss, 0);
5339 return PTR_ERR(newid);
5341 newid->stack[0] = newid->id;
5342 newid->css = rootcss;
5343 rootcss->id = newid;
5347 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5348 struct cgroup *child)
5350 int subsys_id, i, depth = 0;
5351 struct cgroup_subsys_state *parent_css, *child_css;
5352 struct css_id *child_id, *parent_id;
5354 subsys_id = ss->subsys_id;
5355 parent_css = parent->subsys[subsys_id];
5356 child_css = child->subsys[subsys_id];
5357 parent_id = parent_css->id;
5358 depth = parent_id->depth + 1;
5360 child_id = get_new_cssid(ss, depth);
5361 if (IS_ERR(child_id))
5362 return PTR_ERR(child_id);
5364 for (i = 0; i < depth; i++)
5365 child_id->stack[i] = parent_id->stack[i];
5366 child_id->stack[depth] = child_id->id;
5368 * child_id->css pointer will be set after this cgroup is available
5369 * see cgroup_populate_dir()
5371 rcu_assign_pointer(child_css->id, child_id);
5377 * css_lookup - lookup css by id
5378 * @ss: cgroup subsys to be looked into.
5381 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5382 * NULL if not. Should be called under rcu_read_lock()
5384 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5386 struct css_id *cssid = NULL;
5388 BUG_ON(!ss->use_id);
5389 cssid = idr_find(&ss->idr, id);
5391 if (unlikely(!cssid))
5394 return rcu_dereference(cssid->css);
5396 EXPORT_SYMBOL_GPL(css_lookup);
5399 * css_get_next - lookup next cgroup under specified hierarchy.
5400 * @ss: pointer to subsystem
5401 * @id: current position of iteration.
5402 * @root: pointer to css. search tree under this.
5403 * @foundid: position of found object.
5405 * Search next css under the specified hierarchy of rootid. Calling under
5406 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5408 struct cgroup_subsys_state *
5409 css_get_next(struct cgroup_subsys *ss, int id,
5410 struct cgroup_subsys_state *root, int *foundid)
5412 struct cgroup_subsys_state *ret = NULL;
5415 int rootid = css_id(root);
5416 int depth = css_depth(root);
5421 BUG_ON(!ss->use_id);
5422 WARN_ON_ONCE(!rcu_read_lock_held());
5424 /* fill start point for scan */
5428 * scan next entry from bitmap(tree), tmpid is updated after
5431 tmp = idr_get_next(&ss->idr, &tmpid);
5434 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5435 ret = rcu_dereference(tmp->css);
5441 /* continue to scan from next id */
5448 * get corresponding css from file open on cgroupfs directory
5450 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5452 struct cgroup *cgrp;
5453 struct inode *inode;
5454 struct cgroup_subsys_state *css;
5456 inode = f->f_dentry->d_inode;
5457 /* check in cgroup filesystem dir */
5458 if (inode->i_op != &cgroup_dir_inode_operations)
5459 return ERR_PTR(-EBADF);
5461 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5462 return ERR_PTR(-EINVAL);
5465 cgrp = __d_cgrp(f->f_dentry);
5466 css = cgrp->subsys[id];
5467 return css ? css : ERR_PTR(-ENOENT);
5470 #ifdef CONFIG_CGROUP_DEBUG
5471 static struct cgroup_subsys_state *debug_css_alloc(struct cgroup *cont)
5473 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5476 return ERR_PTR(-ENOMEM);
5481 static void debug_css_free(struct cgroup *cont)
5483 kfree(cont->subsys[debug_subsys_id]);
5486 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5488 return atomic_read(&cont->count);
5491 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5493 return cgroup_task_count(cont);
5496 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5498 return (u64)(unsigned long)current->cgroups;
5501 static u64 current_css_set_refcount_read(struct cgroup *cont,
5507 count = atomic_read(¤t->cgroups->refcount);
5512 static int current_css_set_cg_links_read(struct cgroup *cont,
5514 struct seq_file *seq)
5516 struct cg_cgroup_link *link;
5519 read_lock(&css_set_lock);
5521 cg = rcu_dereference(current->cgroups);
5522 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5523 struct cgroup *c = link->cgrp;
5527 name = c->dentry->d_name.name;
5530 seq_printf(seq, "Root %d group %s\n",
5531 c->root->hierarchy_id, name);
5534 read_unlock(&css_set_lock);
5538 #define MAX_TASKS_SHOWN_PER_CSS 25
5539 static int cgroup_css_links_read(struct cgroup *cont,
5541 struct seq_file *seq)
5543 struct cg_cgroup_link *link;
5545 read_lock(&css_set_lock);
5546 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5547 struct css_set *cg = link->cg;
5548 struct task_struct *task;
5550 seq_printf(seq, "css_set %p\n", cg);
5551 list_for_each_entry(task, &cg->tasks, cg_list) {
5552 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5553 seq_puts(seq, " ...\n");
5556 seq_printf(seq, " task %d\n",
5557 task_pid_vnr(task));
5561 read_unlock(&css_set_lock);
5565 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5567 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5570 static struct cftype debug_files[] = {
5572 .name = "cgroup_refcount",
5573 .read_u64 = cgroup_refcount_read,
5576 .name = "taskcount",
5577 .read_u64 = debug_taskcount_read,
5581 .name = "current_css_set",
5582 .read_u64 = current_css_set_read,
5586 .name = "current_css_set_refcount",
5587 .read_u64 = current_css_set_refcount_read,
5591 .name = "current_css_set_cg_links",
5592 .read_seq_string = current_css_set_cg_links_read,
5596 .name = "cgroup_css_links",
5597 .read_seq_string = cgroup_css_links_read,
5601 .name = "releasable",
5602 .read_u64 = releasable_read,
5608 struct cgroup_subsys debug_subsys = {
5610 .css_alloc = debug_css_alloc,
5611 .css_free = debug_css_free,
5612 .subsys_id = debug_subsys_id,
5613 .base_cftypes = debug_files,
5615 #endif /* CONFIG_CGROUP_DEBUG */