2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hash.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_proc */
63 #include <linux/kthread.h>
65 #include <linux/atomic.h>
67 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
68 #define CSS_DEACT_BIAS INT_MIN
71 * cgroup_mutex is the master lock. Any modification to cgroup or its
72 * hierarchy must be performed while holding it.
74 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
75 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
76 * release_agent_path and so on. Modifying requires both cgroup_mutex and
77 * cgroup_root_mutex. Readers can acquire either of the two. This is to
78 * break the following locking order cycle.
80 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
81 * B. namespace_sem -> cgroup_mutex
83 * B happens only through cgroup_show_options() and using cgroup_root_mutex
86 static DEFINE_MUTEX(cgroup_mutex);
87 static DEFINE_MUTEX(cgroup_root_mutex);
90 * Generate an array of cgroup subsystem pointers. At boot time, this is
91 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
92 * registered after that. The mutable section of this array is protected by
95 #define SUBSYS(_x) &_x ## _subsys,
96 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
97 #include <linux/cgroup_subsys.h>
100 #define MAX_CGROUP_ROOT_NAMELEN 64
103 * A cgroupfs_root represents the root of a cgroup hierarchy,
104 * and may be associated with a superblock to form an active
107 struct cgroupfs_root {
108 struct super_block *sb;
111 * The bitmask of subsystems intended to be attached to this
114 unsigned long subsys_bits;
116 /* Unique id for this hierarchy. */
119 /* The bitmask of subsystems currently attached to this hierarchy */
120 unsigned long actual_subsys_bits;
122 /* A list running through the attached subsystems */
123 struct list_head subsys_list;
125 /* The root cgroup for this hierarchy */
126 struct cgroup top_cgroup;
128 /* Tracks how many cgroups are currently defined in hierarchy.*/
129 int number_of_cgroups;
131 /* A list running through the active hierarchies */
132 struct list_head root_list;
134 /* All cgroups on this root, cgroup_mutex protected */
135 struct list_head allcg_list;
137 /* Hierarchy-specific flags */
140 /* The path to use for release notifications. */
141 char release_agent_path[PATH_MAX];
143 /* The name for this hierarchy - may be empty */
144 char name[MAX_CGROUP_ROOT_NAMELEN];
148 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
149 * subsystems that are otherwise unattached - it never has more than a
150 * single cgroup, and all tasks are part of that cgroup.
152 static struct cgroupfs_root rootnode;
155 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
158 struct list_head node;
159 struct dentry *dentry;
164 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
165 * cgroup_subsys->use_id != 0.
167 #define CSS_ID_MAX (65535)
170 * The css to which this ID points. This pointer is set to valid value
171 * after cgroup is populated. If cgroup is removed, this will be NULL.
172 * This pointer is expected to be RCU-safe because destroy()
173 * is called after synchronize_rcu(). But for safe use, css_tryget()
174 * should be used for avoiding race.
176 struct cgroup_subsys_state __rcu *css;
182 * Depth in hierarchy which this ID belongs to.
184 unsigned short depth;
186 * ID is freed by RCU. (and lookup routine is RCU safe.)
188 struct rcu_head rcu_head;
190 * Hierarchy of CSS ID belongs to.
192 unsigned short stack[0]; /* Array of Length (depth+1) */
196 * cgroup_event represents events which userspace want to receive.
198 struct cgroup_event {
200 * Cgroup which the event belongs to.
204 * Control file which the event associated.
208 * eventfd to signal userspace about the event.
210 struct eventfd_ctx *eventfd;
212 * Each of these stored in a list by the cgroup.
214 struct list_head list;
216 * All fields below needed to unregister event when
217 * userspace closes eventfd.
220 wait_queue_head_t *wqh;
222 struct work_struct remove;
225 /* The list of hierarchy roots */
227 static LIST_HEAD(roots);
228 static int root_count;
230 static DEFINE_IDA(hierarchy_ida);
231 static int next_hierarchy_id;
232 static DEFINE_SPINLOCK(hierarchy_id_lock);
234 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
235 #define dummytop (&rootnode.top_cgroup)
237 /* This flag indicates whether tasks in the fork and exit paths should
238 * check for fork/exit handlers to call. This avoids us having to do
239 * extra work in the fork/exit path if none of the subsystems need to
242 static int need_forkexit_callback __read_mostly;
244 #ifdef CONFIG_PROVE_LOCKING
245 int cgroup_lock_is_held(void)
247 return lockdep_is_held(&cgroup_mutex);
249 #else /* #ifdef CONFIG_PROVE_LOCKING */
250 int cgroup_lock_is_held(void)
252 return mutex_is_locked(&cgroup_mutex);
254 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
256 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
258 static int css_unbias_refcnt(int refcnt)
260 return refcnt >= 0 ? refcnt : refcnt - CSS_DEACT_BIAS;
263 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
264 static int css_refcnt(struct cgroup_subsys_state *css)
266 int v = atomic_read(&css->refcnt);
268 return css_unbias_refcnt(v);
271 /* convenient tests for these bits */
272 inline int cgroup_is_removed(const struct cgroup *cgrp)
274 return test_bit(CGRP_REMOVED, &cgrp->flags);
277 /* bits in struct cgroupfs_root flags field */
279 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
282 static int cgroup_is_releasable(const struct cgroup *cgrp)
285 (1 << CGRP_RELEASABLE) |
286 (1 << CGRP_NOTIFY_ON_RELEASE);
287 return (cgrp->flags & bits) == bits;
290 static int notify_on_release(const struct cgroup *cgrp)
292 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
295 static int clone_children(const struct cgroup *cgrp)
297 return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
301 * for_each_subsys() allows you to iterate on each subsystem attached to
302 * an active hierarchy
304 #define for_each_subsys(_root, _ss) \
305 list_for_each_entry(_ss, &_root->subsys_list, sibling)
307 /* for_each_active_root() allows you to iterate across the active hierarchies */
308 #define for_each_active_root(_root) \
309 list_for_each_entry(_root, &roots, root_list)
311 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
313 return dentry->d_fsdata;
316 static inline struct cfent *__d_cfe(struct dentry *dentry)
318 return dentry->d_fsdata;
321 static inline struct cftype *__d_cft(struct dentry *dentry)
323 return __d_cfe(dentry)->type;
326 /* the list of cgroups eligible for automatic release. Protected by
327 * release_list_lock */
328 static LIST_HEAD(release_list);
329 static DEFINE_RAW_SPINLOCK(release_list_lock);
330 static void cgroup_release_agent(struct work_struct *work);
331 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
332 static void check_for_release(struct cgroup *cgrp);
334 /* Link structure for associating css_set objects with cgroups */
335 struct cg_cgroup_link {
337 * List running through cg_cgroup_links associated with a
338 * cgroup, anchored on cgroup->css_sets
340 struct list_head cgrp_link_list;
343 * List running through cg_cgroup_links pointing at a
344 * single css_set object, anchored on css_set->cg_links
346 struct list_head cg_link_list;
350 /* The default css_set - used by init and its children prior to any
351 * hierarchies being mounted. It contains a pointer to the root state
352 * for each subsystem. Also used to anchor the list of css_sets. Not
353 * reference-counted, to improve performance when child cgroups
354 * haven't been created.
357 static struct css_set init_css_set;
358 static struct cg_cgroup_link init_css_set_link;
360 static int cgroup_init_idr(struct cgroup_subsys *ss,
361 struct cgroup_subsys_state *css);
363 /* css_set_lock protects the list of css_set objects, and the
364 * chain of tasks off each css_set. Nests outside task->alloc_lock
365 * due to cgroup_iter_start() */
366 static DEFINE_RWLOCK(css_set_lock);
367 static int css_set_count;
370 * hash table for cgroup groups. This improves the performance to find
371 * an existing css_set. This hash doesn't (currently) take into
372 * account cgroups in empty hierarchies.
374 #define CSS_SET_HASH_BITS 7
375 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
376 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
378 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
382 unsigned long tmp = 0UL;
384 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
385 tmp += (unsigned long)css[i];
386 tmp = (tmp >> 16) ^ tmp;
388 index = hash_long(tmp, CSS_SET_HASH_BITS);
390 return &css_set_table[index];
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 hlist_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_head *hhead;
550 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_bits & (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 hhead = css_set_hash(template);
572 hlist_for_each_entry(cg, node, hhead, hlist) {
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 hlist_head *hhead;
657 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 hhead = css_set_hash(res->subsys);
704 hlist_add_head(&res->hlist, hhead);
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 tasks 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'ss 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);
828 static const struct inode_operations cgroup_dir_inode_operations;
829 static const struct file_operations proc_cgroupstats_operations;
831 static struct backing_dev_info cgroup_backing_dev_info = {
833 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
836 static int alloc_css_id(struct cgroup_subsys *ss,
837 struct cgroup *parent, struct cgroup *child);
839 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
841 struct inode *inode = new_inode(sb);
844 inode->i_ino = get_next_ino();
845 inode->i_mode = mode;
846 inode->i_uid = current_fsuid();
847 inode->i_gid = current_fsgid();
848 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
849 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
854 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
856 /* is dentry a directory ? if so, kfree() associated cgroup */
857 if (S_ISDIR(inode->i_mode)) {
858 struct cgroup *cgrp = dentry->d_fsdata;
859 struct cgroup_subsys *ss;
860 BUG_ON(!(cgroup_is_removed(cgrp)));
861 /* It's possible for external users to be holding css
862 * reference counts on a cgroup; css_put() needs to
863 * be able to access the cgroup after decrementing
864 * the reference count in order to know if it needs to
865 * queue the cgroup to be handled by the release
869 mutex_lock(&cgroup_mutex);
871 * Release the subsystem state objects.
873 for_each_subsys(cgrp->root, ss)
876 cgrp->root->number_of_cgroups--;
877 mutex_unlock(&cgroup_mutex);
880 * Drop the active superblock reference that we took when we
883 deactivate_super(cgrp->root->sb);
886 * if we're getting rid of the cgroup, refcount should ensure
887 * that there are no pidlists left.
889 BUG_ON(!list_empty(&cgrp->pidlists));
891 kfree_rcu(cgrp, rcu_head);
893 struct cfent *cfe = __d_cfe(dentry);
894 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
896 WARN_ONCE(!list_empty(&cfe->node) &&
897 cgrp != &cgrp->root->top_cgroup,
898 "cfe still linked for %s\n", cfe->type->name);
904 static int cgroup_delete(const struct dentry *d)
909 static void remove_dir(struct dentry *d)
911 struct dentry *parent = dget(d->d_parent);
914 simple_rmdir(parent->d_inode, d);
918 static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
922 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
923 lockdep_assert_held(&cgroup_mutex);
925 list_for_each_entry(cfe, &cgrp->files, node) {
926 struct dentry *d = cfe->dentry;
928 if (cft && cfe->type != cft)
933 simple_unlink(cgrp->dentry->d_inode, d);
934 list_del_init(&cfe->node);
942 static void cgroup_clear_directory(struct dentry *dir)
944 struct cgroup *cgrp = __d_cgrp(dir);
946 while (!list_empty(&cgrp->files))
947 cgroup_rm_file(cgrp, NULL);
951 * NOTE : the dentry must have been dget()'ed
953 static void cgroup_d_remove_dir(struct dentry *dentry)
955 struct dentry *parent;
957 cgroup_clear_directory(dentry);
959 parent = dentry->d_parent;
960 spin_lock(&parent->d_lock);
961 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
962 list_del_init(&dentry->d_u.d_child);
963 spin_unlock(&dentry->d_lock);
964 spin_unlock(&parent->d_lock);
969 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
970 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
971 * reference to css->refcnt. In general, this refcnt is expected to goes down
974 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
976 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
978 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
980 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
981 wake_up_all(&cgroup_rmdir_waitq);
984 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
989 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
991 cgroup_wakeup_rmdir_waiter(css->cgroup);
996 * Call with cgroup_mutex held. Drops reference counts on modules, including
997 * any duplicate ones that parse_cgroupfs_options took. If this function
998 * returns an error, no reference counts are touched.
1000 static int rebind_subsystems(struct cgroupfs_root *root,
1001 unsigned long final_bits)
1003 unsigned long added_bits, removed_bits;
1004 struct cgroup *cgrp = &root->top_cgroup;
1007 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1008 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1010 removed_bits = root->actual_subsys_bits & ~final_bits;
1011 added_bits = final_bits & ~root->actual_subsys_bits;
1012 /* Check that any added subsystems are currently free */
1013 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1014 unsigned long bit = 1UL << i;
1015 struct cgroup_subsys *ss = subsys[i];
1016 if (!(bit & added_bits))
1019 * Nobody should tell us to do a subsys that doesn't exist:
1020 * parse_cgroupfs_options should catch that case and refcounts
1021 * ensure that subsystems won't disappear once selected.
1024 if (ss->root != &rootnode) {
1025 /* Subsystem isn't free */
1030 /* Currently we don't handle adding/removing subsystems when
1031 * any child cgroups exist. This is theoretically supportable
1032 * but involves complex error handling, so it's being left until
1034 if (root->number_of_cgroups > 1)
1037 /* Process each subsystem */
1038 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1039 struct cgroup_subsys *ss = subsys[i];
1040 unsigned long bit = 1UL << i;
1041 if (bit & added_bits) {
1042 /* We're binding this subsystem to this hierarchy */
1044 BUG_ON(cgrp->subsys[i]);
1045 BUG_ON(!dummytop->subsys[i]);
1046 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1047 cgrp->subsys[i] = dummytop->subsys[i];
1048 cgrp->subsys[i]->cgroup = cgrp;
1049 list_move(&ss->sibling, &root->subsys_list);
1053 /* refcount was already taken, and we're keeping it */
1054 } else if (bit & removed_bits) {
1055 /* We're removing this subsystem */
1057 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1058 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1061 dummytop->subsys[i]->cgroup = dummytop;
1062 cgrp->subsys[i] = NULL;
1063 subsys[i]->root = &rootnode;
1064 list_move(&ss->sibling, &rootnode.subsys_list);
1065 /* subsystem is now free - drop reference on module */
1066 module_put(ss->module);
1067 } else if (bit & final_bits) {
1068 /* Subsystem state should already exist */
1070 BUG_ON(!cgrp->subsys[i]);
1072 * a refcount was taken, but we already had one, so
1073 * drop the extra reference.
1075 module_put(ss->module);
1076 #ifdef CONFIG_MODULE_UNLOAD
1077 BUG_ON(ss->module && !module_refcount(ss->module));
1080 /* Subsystem state shouldn't exist */
1081 BUG_ON(cgrp->subsys[i]);
1084 root->subsys_bits = root->actual_subsys_bits = final_bits;
1090 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1092 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1093 struct cgroup_subsys *ss;
1095 mutex_lock(&cgroup_root_mutex);
1096 for_each_subsys(root, ss)
1097 seq_printf(seq, ",%s", ss->name);
1098 if (test_bit(ROOT_NOPREFIX, &root->flags))
1099 seq_puts(seq, ",noprefix");
1100 if (strlen(root->release_agent_path))
1101 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1102 if (clone_children(&root->top_cgroup))
1103 seq_puts(seq, ",clone_children");
1104 if (strlen(root->name))
1105 seq_printf(seq, ",name=%s", root->name);
1106 mutex_unlock(&cgroup_root_mutex);
1110 struct cgroup_sb_opts {
1111 unsigned long subsys_bits;
1112 unsigned long flags;
1113 char *release_agent;
1114 bool clone_children;
1116 /* User explicitly requested empty subsystem */
1119 struct cgroupfs_root *new_root;
1124 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1125 * with cgroup_mutex held to protect the subsys[] array. This function takes
1126 * refcounts on subsystems to be used, unless it returns error, in which case
1127 * no refcounts are taken.
1129 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1131 char *token, *o = data;
1132 bool all_ss = false, one_ss = false;
1133 unsigned long mask = (unsigned long)-1;
1135 bool module_pin_failed = false;
1137 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1139 #ifdef CONFIG_CPUSETS
1140 mask = ~(1UL << cpuset_subsys_id);
1143 memset(opts, 0, sizeof(*opts));
1145 while ((token = strsep(&o, ",")) != NULL) {
1148 if (!strcmp(token, "none")) {
1149 /* Explicitly have no subsystems */
1153 if (!strcmp(token, "all")) {
1154 /* Mutually exclusive option 'all' + subsystem name */
1160 if (!strcmp(token, "noprefix")) {
1161 set_bit(ROOT_NOPREFIX, &opts->flags);
1164 if (!strcmp(token, "clone_children")) {
1165 opts->clone_children = true;
1168 if (!strncmp(token, "release_agent=", 14)) {
1169 /* Specifying two release agents is forbidden */
1170 if (opts->release_agent)
1172 opts->release_agent =
1173 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1174 if (!opts->release_agent)
1178 if (!strncmp(token, "name=", 5)) {
1179 const char *name = token + 5;
1180 /* Can't specify an empty name */
1183 /* Must match [\w.-]+ */
1184 for (i = 0; i < strlen(name); i++) {
1188 if ((c == '.') || (c == '-') || (c == '_'))
1192 /* Specifying two names is forbidden */
1195 opts->name = kstrndup(name,
1196 MAX_CGROUP_ROOT_NAMELEN - 1,
1204 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1205 struct cgroup_subsys *ss = subsys[i];
1208 if (strcmp(token, ss->name))
1213 /* Mutually exclusive option 'all' + subsystem name */
1216 set_bit(i, &opts->subsys_bits);
1221 if (i == CGROUP_SUBSYS_COUNT)
1226 * If the 'all' option was specified select all the subsystems,
1227 * otherwise if 'none', 'name=' and a subsystem name options
1228 * were not specified, let's default to 'all'
1230 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1231 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1232 struct cgroup_subsys *ss = subsys[i];
1237 set_bit(i, &opts->subsys_bits);
1241 /* Consistency checks */
1244 * Option noprefix was introduced just for backward compatibility
1245 * with the old cpuset, so we allow noprefix only if mounting just
1246 * the cpuset subsystem.
1248 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1249 (opts->subsys_bits & mask))
1253 /* Can't specify "none" and some subsystems */
1254 if (opts->subsys_bits && opts->none)
1258 * We either have to specify by name or by subsystems. (So all
1259 * empty hierarchies must have a name).
1261 if (!opts->subsys_bits && !opts->name)
1265 * Grab references on all the modules we'll need, so the subsystems
1266 * don't dance around before rebind_subsystems attaches them. This may
1267 * take duplicate reference counts on a subsystem that's already used,
1268 * but rebind_subsystems handles this case.
1270 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1271 unsigned long bit = 1UL << i;
1273 if (!(bit & opts->subsys_bits))
1275 if (!try_module_get(subsys[i]->module)) {
1276 module_pin_failed = true;
1280 if (module_pin_failed) {
1282 * oops, one of the modules was going away. this means that we
1283 * raced with a module_delete call, and to the user this is
1284 * essentially a "subsystem doesn't exist" case.
1286 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1287 /* drop refcounts only on the ones we took */
1288 unsigned long bit = 1UL << i;
1290 if (!(bit & opts->subsys_bits))
1292 module_put(subsys[i]->module);
1300 static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1303 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1304 unsigned long bit = 1UL << i;
1306 if (!(bit & subsys_bits))
1308 module_put(subsys[i]->module);
1312 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1315 struct cgroupfs_root *root = sb->s_fs_info;
1316 struct cgroup *cgrp = &root->top_cgroup;
1317 struct cgroup_sb_opts opts;
1319 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1320 mutex_lock(&cgroup_mutex);
1321 mutex_lock(&cgroup_root_mutex);
1323 /* See what subsystems are wanted */
1324 ret = parse_cgroupfs_options(data, &opts);
1328 /* See feature-removal-schedule.txt */
1329 if (opts.subsys_bits != root->actual_subsys_bits || opts.release_agent)
1330 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1331 task_tgid_nr(current), current->comm);
1333 /* Don't allow flags or name to change at remount */
1334 if (opts.flags != root->flags ||
1335 (opts.name && strcmp(opts.name, root->name))) {
1337 drop_parsed_module_refcounts(opts.subsys_bits);
1341 ret = rebind_subsystems(root, opts.subsys_bits);
1343 drop_parsed_module_refcounts(opts.subsys_bits);
1347 /* clear out any existing files and repopulate subsystem files */
1348 cgroup_clear_directory(cgrp->dentry);
1349 cgroup_populate_dir(cgrp);
1351 if (opts.release_agent)
1352 strcpy(root->release_agent_path, opts.release_agent);
1354 kfree(opts.release_agent);
1356 mutex_unlock(&cgroup_root_mutex);
1357 mutex_unlock(&cgroup_mutex);
1358 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1362 static const struct super_operations cgroup_ops = {
1363 .statfs = simple_statfs,
1364 .drop_inode = generic_delete_inode,
1365 .show_options = cgroup_show_options,
1366 .remount_fs = cgroup_remount,
1369 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1371 INIT_LIST_HEAD(&cgrp->sibling);
1372 INIT_LIST_HEAD(&cgrp->children);
1373 INIT_LIST_HEAD(&cgrp->files);
1374 INIT_LIST_HEAD(&cgrp->css_sets);
1375 INIT_LIST_HEAD(&cgrp->release_list);
1376 INIT_LIST_HEAD(&cgrp->pidlists);
1377 mutex_init(&cgrp->pidlist_mutex);
1378 INIT_LIST_HEAD(&cgrp->event_list);
1379 spin_lock_init(&cgrp->event_list_lock);
1382 static void init_cgroup_root(struct cgroupfs_root *root)
1384 struct cgroup *cgrp = &root->top_cgroup;
1386 INIT_LIST_HEAD(&root->subsys_list);
1387 INIT_LIST_HEAD(&root->root_list);
1388 INIT_LIST_HEAD(&root->allcg_list);
1389 root->number_of_cgroups = 1;
1391 cgrp->top_cgroup = cgrp;
1392 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1393 init_cgroup_housekeeping(cgrp);
1396 static bool init_root_id(struct cgroupfs_root *root)
1401 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1403 spin_lock(&hierarchy_id_lock);
1404 /* Try to allocate the next unused ID */
1405 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1406 &root->hierarchy_id);
1408 /* Try again starting from 0 */
1409 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1411 next_hierarchy_id = root->hierarchy_id + 1;
1412 } else if (ret != -EAGAIN) {
1413 /* Can only get here if the 31-bit IDR is full ... */
1416 spin_unlock(&hierarchy_id_lock);
1421 static int cgroup_test_super(struct super_block *sb, void *data)
1423 struct cgroup_sb_opts *opts = data;
1424 struct cgroupfs_root *root = sb->s_fs_info;
1426 /* If we asked for a name then it must match */
1427 if (opts->name && strcmp(opts->name, root->name))
1431 * If we asked for subsystems (or explicitly for no
1432 * subsystems) then they must match
1434 if ((opts->subsys_bits || opts->none)
1435 && (opts->subsys_bits != root->subsys_bits))
1441 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1443 struct cgroupfs_root *root;
1445 if (!opts->subsys_bits && !opts->none)
1448 root = kzalloc(sizeof(*root), GFP_KERNEL);
1450 return ERR_PTR(-ENOMEM);
1452 if (!init_root_id(root)) {
1454 return ERR_PTR(-ENOMEM);
1456 init_cgroup_root(root);
1458 root->subsys_bits = opts->subsys_bits;
1459 root->flags = opts->flags;
1460 if (opts->release_agent)
1461 strcpy(root->release_agent_path, opts->release_agent);
1463 strcpy(root->name, opts->name);
1464 if (opts->clone_children)
1465 set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1469 static void cgroup_drop_root(struct cgroupfs_root *root)
1474 BUG_ON(!root->hierarchy_id);
1475 spin_lock(&hierarchy_id_lock);
1476 ida_remove(&hierarchy_ida, root->hierarchy_id);
1477 spin_unlock(&hierarchy_id_lock);
1481 static int cgroup_set_super(struct super_block *sb, void *data)
1484 struct cgroup_sb_opts *opts = data;
1486 /* If we don't have a new root, we can't set up a new sb */
1487 if (!opts->new_root)
1490 BUG_ON(!opts->subsys_bits && !opts->none);
1492 ret = set_anon_super(sb, NULL);
1496 sb->s_fs_info = opts->new_root;
1497 opts->new_root->sb = sb;
1499 sb->s_blocksize = PAGE_CACHE_SIZE;
1500 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1501 sb->s_magic = CGROUP_SUPER_MAGIC;
1502 sb->s_op = &cgroup_ops;
1507 static int cgroup_get_rootdir(struct super_block *sb)
1509 static const struct dentry_operations cgroup_dops = {
1510 .d_iput = cgroup_diput,
1511 .d_delete = cgroup_delete,
1514 struct inode *inode =
1515 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1520 inode->i_fop = &simple_dir_operations;
1521 inode->i_op = &cgroup_dir_inode_operations;
1522 /* directories start off with i_nlink == 2 (for "." entry) */
1524 sb->s_root = d_make_root(inode);
1527 /* for everything else we want ->d_op set */
1528 sb->s_d_op = &cgroup_dops;
1532 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1533 int flags, const char *unused_dev_name,
1536 struct cgroup_sb_opts opts;
1537 struct cgroupfs_root *root;
1539 struct super_block *sb;
1540 struct cgroupfs_root *new_root;
1541 struct inode *inode;
1543 /* First find the desired set of subsystems */
1544 mutex_lock(&cgroup_mutex);
1545 ret = parse_cgroupfs_options(data, &opts);
1546 mutex_unlock(&cgroup_mutex);
1551 * Allocate a new cgroup root. We may not need it if we're
1552 * reusing an existing hierarchy.
1554 new_root = cgroup_root_from_opts(&opts);
1555 if (IS_ERR(new_root)) {
1556 ret = PTR_ERR(new_root);
1559 opts.new_root = new_root;
1561 /* Locate an existing or new sb for this hierarchy */
1562 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, 0, &opts);
1565 cgroup_drop_root(opts.new_root);
1569 root = sb->s_fs_info;
1571 if (root == opts.new_root) {
1572 /* We used the new root structure, so this is a new hierarchy */
1573 struct list_head tmp_cg_links;
1574 struct cgroup *root_cgrp = &root->top_cgroup;
1575 struct cgroupfs_root *existing_root;
1576 const struct cred *cred;
1579 BUG_ON(sb->s_root != NULL);
1581 ret = cgroup_get_rootdir(sb);
1583 goto drop_new_super;
1584 inode = sb->s_root->d_inode;
1586 mutex_lock(&inode->i_mutex);
1587 mutex_lock(&cgroup_mutex);
1588 mutex_lock(&cgroup_root_mutex);
1590 /* Check for name clashes with existing mounts */
1592 if (strlen(root->name))
1593 for_each_active_root(existing_root)
1594 if (!strcmp(existing_root->name, root->name))
1598 * We're accessing css_set_count without locking
1599 * css_set_lock here, but that's OK - it can only be
1600 * increased by someone holding cgroup_lock, and
1601 * that's us. The worst that can happen is that we
1602 * have some link structures left over
1604 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1608 ret = rebind_subsystems(root, root->subsys_bits);
1609 if (ret == -EBUSY) {
1610 free_cg_links(&tmp_cg_links);
1614 * There must be no failure case after here, since rebinding
1615 * takes care of subsystems' refcounts, which are explicitly
1616 * dropped in the failure exit path.
1619 /* EBUSY should be the only error here */
1622 list_add(&root->root_list, &roots);
1625 sb->s_root->d_fsdata = root_cgrp;
1626 root->top_cgroup.dentry = sb->s_root;
1628 /* Link the top cgroup in this hierarchy into all
1629 * the css_set objects */
1630 write_lock(&css_set_lock);
1631 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1632 struct hlist_head *hhead = &css_set_table[i];
1633 struct hlist_node *node;
1636 hlist_for_each_entry(cg, node, hhead, hlist)
1637 link_css_set(&tmp_cg_links, cg, root_cgrp);
1639 write_unlock(&css_set_lock);
1641 free_cg_links(&tmp_cg_links);
1643 BUG_ON(!list_empty(&root_cgrp->sibling));
1644 BUG_ON(!list_empty(&root_cgrp->children));
1645 BUG_ON(root->number_of_cgroups != 1);
1647 cred = override_creds(&init_cred);
1648 cgroup_populate_dir(root_cgrp);
1650 mutex_unlock(&cgroup_root_mutex);
1651 mutex_unlock(&cgroup_mutex);
1652 mutex_unlock(&inode->i_mutex);
1655 * We re-used an existing hierarchy - the new root (if
1656 * any) is not needed
1658 cgroup_drop_root(opts.new_root);
1659 /* no subsys rebinding, so refcounts don't change */
1660 drop_parsed_module_refcounts(opts.subsys_bits);
1663 kfree(opts.release_agent);
1665 return dget(sb->s_root);
1668 mutex_unlock(&cgroup_root_mutex);
1669 mutex_unlock(&cgroup_mutex);
1670 mutex_unlock(&inode->i_mutex);
1672 deactivate_locked_super(sb);
1674 drop_parsed_module_refcounts(opts.subsys_bits);
1676 kfree(opts.release_agent);
1678 return ERR_PTR(ret);
1681 static void cgroup_kill_sb(struct super_block *sb) {
1682 struct cgroupfs_root *root = sb->s_fs_info;
1683 struct cgroup *cgrp = &root->top_cgroup;
1685 struct cg_cgroup_link *link;
1686 struct cg_cgroup_link *saved_link;
1690 BUG_ON(root->number_of_cgroups != 1);
1691 BUG_ON(!list_empty(&cgrp->children));
1692 BUG_ON(!list_empty(&cgrp->sibling));
1694 mutex_lock(&cgroup_mutex);
1695 mutex_lock(&cgroup_root_mutex);
1697 /* Rebind all subsystems back to the default hierarchy */
1698 ret = rebind_subsystems(root, 0);
1699 /* Shouldn't be able to fail ... */
1703 * Release all the links from css_sets to this hierarchy's
1706 write_lock(&css_set_lock);
1708 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1710 list_del(&link->cg_link_list);
1711 list_del(&link->cgrp_link_list);
1714 write_unlock(&css_set_lock);
1716 if (!list_empty(&root->root_list)) {
1717 list_del(&root->root_list);
1721 mutex_unlock(&cgroup_root_mutex);
1722 mutex_unlock(&cgroup_mutex);
1724 kill_litter_super(sb);
1725 cgroup_drop_root(root);
1728 static struct file_system_type cgroup_fs_type = {
1730 .mount = cgroup_mount,
1731 .kill_sb = cgroup_kill_sb,
1734 static struct kobject *cgroup_kobj;
1737 * cgroup_path - generate the path of a cgroup
1738 * @cgrp: the cgroup in question
1739 * @buf: the buffer to write the path into
1740 * @buflen: the length of the buffer
1742 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1743 * reference. Writes path of cgroup into buf. Returns 0 on success,
1746 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1749 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1750 cgroup_lock_is_held());
1752 if (!dentry || cgrp == dummytop) {
1754 * Inactive subsystems have no dentry for their root
1761 start = buf + buflen;
1765 int len = dentry->d_name.len;
1767 if ((start -= len) < buf)
1768 return -ENAMETOOLONG;
1769 memcpy(start, dentry->d_name.name, len);
1770 cgrp = cgrp->parent;
1774 dentry = rcu_dereference_check(cgrp->dentry,
1775 cgroup_lock_is_held());
1779 return -ENAMETOOLONG;
1782 memmove(buf, start, buf + buflen - start);
1785 EXPORT_SYMBOL_GPL(cgroup_path);
1788 * Control Group taskset
1790 struct task_and_cgroup {
1791 struct task_struct *task;
1792 struct cgroup *cgrp;
1796 struct cgroup_taskset {
1797 struct task_and_cgroup single;
1798 struct flex_array *tc_array;
1801 struct cgroup *cur_cgrp;
1805 * cgroup_taskset_first - reset taskset and return the first task
1806 * @tset: taskset of interest
1808 * @tset iteration is initialized and the first task is returned.
1810 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1812 if (tset->tc_array) {
1814 return cgroup_taskset_next(tset);
1816 tset->cur_cgrp = tset->single.cgrp;
1817 return tset->single.task;
1820 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1823 * cgroup_taskset_next - iterate to the next task in taskset
1824 * @tset: taskset of interest
1826 * Return the next task in @tset. Iteration must have been initialized
1827 * with cgroup_taskset_first().
1829 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1831 struct task_and_cgroup *tc;
1833 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1836 tc = flex_array_get(tset->tc_array, tset->idx++);
1837 tset->cur_cgrp = tc->cgrp;
1840 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1843 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1844 * @tset: taskset of interest
1846 * Return the cgroup for the current (last returned) task of @tset. This
1847 * function must be preceded by either cgroup_taskset_first() or
1848 * cgroup_taskset_next().
1850 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1852 return tset->cur_cgrp;
1854 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1857 * cgroup_taskset_size - return the number of tasks in taskset
1858 * @tset: taskset of interest
1860 int cgroup_taskset_size(struct cgroup_taskset *tset)
1862 return tset->tc_array ? tset->tc_array_len : 1;
1864 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1868 * cgroup_task_migrate - move a task from one cgroup to another.
1870 * 'guarantee' is set if the caller promises that a new css_set for the task
1871 * will already exist. If not set, this function might sleep, and can fail with
1872 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1874 static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1875 struct task_struct *tsk, struct css_set *newcg)
1877 struct css_set *oldcg;
1880 * We are synchronized through threadgroup_lock() against PF_EXITING
1881 * setting such that we can't race against cgroup_exit() changing the
1882 * css_set to init_css_set and dropping the old one.
1884 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1885 oldcg = tsk->cgroups;
1888 rcu_assign_pointer(tsk->cgroups, newcg);
1891 /* Update the css_set linked lists if we're using them */
1892 write_lock(&css_set_lock);
1893 if (!list_empty(&tsk->cg_list))
1894 list_move(&tsk->cg_list, &newcg->tasks);
1895 write_unlock(&css_set_lock);
1898 * We just gained a reference on oldcg by taking it from the task. As
1899 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1900 * it here; it will be freed under RCU.
1904 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1908 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1909 * @cgrp: the cgroup the task is attaching to
1910 * @tsk: the task to be attached
1912 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1915 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1918 struct cgroup_subsys *ss, *failed_ss = NULL;
1919 struct cgroup *oldcgrp;
1920 struct cgroupfs_root *root = cgrp->root;
1921 struct cgroup_taskset tset = { };
1922 struct css_set *newcg;
1924 /* @tsk either already exited or can't exit until the end */
1925 if (tsk->flags & PF_EXITING)
1928 /* Nothing to do if the task is already in that cgroup */
1929 oldcgrp = task_cgroup_from_root(tsk, root);
1930 if (cgrp == oldcgrp)
1933 tset.single.task = tsk;
1934 tset.single.cgrp = oldcgrp;
1936 for_each_subsys(root, ss) {
1937 if (ss->can_attach) {
1938 retval = ss->can_attach(cgrp, &tset);
1941 * Remember on which subsystem the can_attach()
1942 * failed, so that we only call cancel_attach()
1943 * against the subsystems whose can_attach()
1944 * succeeded. (See below)
1952 newcg = find_css_set(tsk->cgroups, cgrp);
1958 cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1960 for_each_subsys(root, ss) {
1962 ss->attach(cgrp, &tset);
1968 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1969 * is no longer empty.
1971 cgroup_wakeup_rmdir_waiter(cgrp);
1974 for_each_subsys(root, ss) {
1975 if (ss == failed_ss)
1977 * This subsystem was the one that failed the
1978 * can_attach() check earlier, so we don't need
1979 * to call cancel_attach() against it or any
1980 * remaining subsystems.
1983 if (ss->cancel_attach)
1984 ss->cancel_attach(cgrp, &tset);
1991 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1992 * @from: attach to all cgroups of a given task
1993 * @tsk: the task to be attached
1995 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
1997 struct cgroupfs_root *root;
2001 for_each_active_root(root) {
2002 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2004 retval = cgroup_attach_task(from_cg, tsk);
2012 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2015 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2016 * @cgrp: the cgroup to attach to
2017 * @leader: the threadgroup leader task_struct of the group to be attached
2019 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2020 * task_lock of each thread in leader's threadgroup individually in turn.
2022 static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2024 int retval, i, group_size;
2025 struct cgroup_subsys *ss, *failed_ss = NULL;
2026 /* guaranteed to be initialized later, but the compiler needs this */
2027 struct cgroupfs_root *root = cgrp->root;
2028 /* threadgroup list cursor and array */
2029 struct task_struct *tsk;
2030 struct task_and_cgroup *tc;
2031 struct flex_array *group;
2032 struct cgroup_taskset tset = { };
2035 * step 0: in order to do expensive, possibly blocking operations for
2036 * every thread, we cannot iterate the thread group list, since it needs
2037 * rcu or tasklist locked. instead, build an array of all threads in the
2038 * group - group_rwsem prevents new threads from appearing, and if
2039 * threads exit, this will just be an over-estimate.
2041 group_size = get_nr_threads(leader);
2042 /* flex_array supports very large thread-groups better than kmalloc. */
2043 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2046 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2047 retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2049 goto out_free_group_list;
2054 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2055 * already PF_EXITING could be freed from underneath us unless we
2056 * take an rcu_read_lock.
2060 struct task_and_cgroup ent;
2062 /* @tsk either already exited or can't exit until the end */
2063 if (tsk->flags & PF_EXITING)
2066 /* as per above, nr_threads may decrease, but not increase. */
2067 BUG_ON(i >= group_size);
2069 ent.cgrp = task_cgroup_from_root(tsk, root);
2070 /* nothing to do if this task is already in the cgroup */
2071 if (ent.cgrp == cgrp)
2074 * saying GFP_ATOMIC has no effect here because we did prealloc
2075 * earlier, but it's good form to communicate our expectations.
2077 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2078 BUG_ON(retval != 0);
2080 } while_each_thread(leader, tsk);
2082 /* remember the number of threads in the array for later. */
2084 tset.tc_array = group;
2085 tset.tc_array_len = group_size;
2087 /* methods shouldn't be called if no task is actually migrating */
2090 goto out_free_group_list;
2093 * step 1: check that we can legitimately attach to the cgroup.
2095 for_each_subsys(root, ss) {
2096 if (ss->can_attach) {
2097 retval = ss->can_attach(cgrp, &tset);
2100 goto out_cancel_attach;
2106 * step 2: make sure css_sets exist for all threads to be migrated.
2107 * we use find_css_set, which allocates a new one if necessary.
2109 for (i = 0; i < group_size; i++) {
2110 tc = flex_array_get(group, i);
2111 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2114 goto out_put_css_set_refs;
2119 * step 3: now that we're guaranteed success wrt the css_sets,
2120 * proceed to move all tasks to the new cgroup. There are no
2121 * failure cases after here, so this is the commit point.
2123 for (i = 0; i < group_size; i++) {
2124 tc = flex_array_get(group, i);
2125 cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2127 /* nothing is sensitive to fork() after this point. */
2130 * step 4: do subsystem attach callbacks.
2132 for_each_subsys(root, ss) {
2134 ss->attach(cgrp, &tset);
2138 * step 5: success! and cleanup
2141 cgroup_wakeup_rmdir_waiter(cgrp);
2143 out_put_css_set_refs:
2145 for (i = 0; i < group_size; i++) {
2146 tc = flex_array_get(group, i);
2149 put_css_set(tc->cg);
2154 for_each_subsys(root, ss) {
2155 if (ss == failed_ss)
2157 if (ss->cancel_attach)
2158 ss->cancel_attach(cgrp, &tset);
2161 out_free_group_list:
2162 flex_array_free(group);
2167 * Find the task_struct of the task to attach by vpid and pass it along to the
2168 * function to attach either it or all tasks in its threadgroup. Will lock
2169 * cgroup_mutex and threadgroup; may take task_lock of task.
2171 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2173 struct task_struct *tsk;
2174 const struct cred *cred = current_cred(), *tcred;
2177 if (!cgroup_lock_live_group(cgrp))
2183 tsk = find_task_by_vpid(pid);
2187 goto out_unlock_cgroup;
2190 * even if we're attaching all tasks in the thread group, we
2191 * only need to check permissions on one of them.
2193 tcred = __task_cred(tsk);
2194 if (!uid_eq(cred->euid, GLOBAL_ROOT_UID) &&
2195 !uid_eq(cred->euid, tcred->uid) &&
2196 !uid_eq(cred->euid, tcred->suid)) {
2199 goto out_unlock_cgroup;
2205 tsk = tsk->group_leader;
2208 * Workqueue threads may acquire PF_THREAD_BOUND and become
2209 * trapped in a cpuset, or RT worker may be born in a cgroup
2210 * with no rt_runtime allocated. Just say no.
2212 if (tsk == kthreadd_task || (tsk->flags & PF_THREAD_BOUND)) {
2215 goto out_unlock_cgroup;
2218 get_task_struct(tsk);
2221 threadgroup_lock(tsk);
2223 if (!thread_group_leader(tsk)) {
2225 * a race with de_thread from another thread's exec()
2226 * may strip us of our leadership, if this happens,
2227 * there is no choice but to throw this task away and
2228 * try again; this is
2229 * "double-double-toil-and-trouble-check locking".
2231 threadgroup_unlock(tsk);
2232 put_task_struct(tsk);
2233 goto retry_find_task;
2235 ret = cgroup_attach_proc(cgrp, tsk);
2237 ret = cgroup_attach_task(cgrp, tsk);
2238 threadgroup_unlock(tsk);
2240 put_task_struct(tsk);
2246 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2248 return attach_task_by_pid(cgrp, pid, false);
2251 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2253 return attach_task_by_pid(cgrp, tgid, true);
2257 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2258 * @cgrp: the cgroup to be checked for liveness
2260 * On success, returns true; the lock should be later released with
2261 * cgroup_unlock(). On failure returns false with no lock held.
2263 bool cgroup_lock_live_group(struct cgroup *cgrp)
2265 mutex_lock(&cgroup_mutex);
2266 if (cgroup_is_removed(cgrp)) {
2267 mutex_unlock(&cgroup_mutex);
2272 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2274 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2277 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2278 if (strlen(buffer) >= PATH_MAX)
2280 if (!cgroup_lock_live_group(cgrp))
2282 mutex_lock(&cgroup_root_mutex);
2283 strcpy(cgrp->root->release_agent_path, buffer);
2284 mutex_unlock(&cgroup_root_mutex);
2289 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2290 struct seq_file *seq)
2292 if (!cgroup_lock_live_group(cgrp))
2294 seq_puts(seq, cgrp->root->release_agent_path);
2295 seq_putc(seq, '\n');
2300 /* A buffer size big enough for numbers or short strings */
2301 #define CGROUP_LOCAL_BUFFER_SIZE 64
2303 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2305 const char __user *userbuf,
2306 size_t nbytes, loff_t *unused_ppos)
2308 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2314 if (nbytes >= sizeof(buffer))
2316 if (copy_from_user(buffer, userbuf, nbytes))
2319 buffer[nbytes] = 0; /* nul-terminate */
2320 if (cft->write_u64) {
2321 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2324 retval = cft->write_u64(cgrp, cft, val);
2326 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2329 retval = cft->write_s64(cgrp, cft, val);
2336 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2338 const char __user *userbuf,
2339 size_t nbytes, loff_t *unused_ppos)
2341 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2343 size_t max_bytes = cft->max_write_len;
2344 char *buffer = local_buffer;
2347 max_bytes = sizeof(local_buffer) - 1;
2348 if (nbytes >= max_bytes)
2350 /* Allocate a dynamic buffer if we need one */
2351 if (nbytes >= sizeof(local_buffer)) {
2352 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2356 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2361 buffer[nbytes] = 0; /* nul-terminate */
2362 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2366 if (buffer != local_buffer)
2371 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2372 size_t nbytes, loff_t *ppos)
2374 struct cftype *cft = __d_cft(file->f_dentry);
2375 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2377 if (cgroup_is_removed(cgrp))
2380 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2381 if (cft->write_u64 || cft->write_s64)
2382 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2383 if (cft->write_string)
2384 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2386 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2387 return ret ? ret : nbytes;
2392 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2394 char __user *buf, size_t nbytes,
2397 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2398 u64 val = cft->read_u64(cgrp, cft);
2399 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2401 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2404 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2406 char __user *buf, size_t nbytes,
2409 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2410 s64 val = cft->read_s64(cgrp, cft);
2411 int len = sprintf(tmp, "%lld\n", (long long) val);
2413 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2416 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2417 size_t nbytes, loff_t *ppos)
2419 struct cftype *cft = __d_cft(file->f_dentry);
2420 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2422 if (cgroup_is_removed(cgrp))
2426 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2428 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2430 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2435 * seqfile ops/methods for returning structured data. Currently just
2436 * supports string->u64 maps, but can be extended in future.
2439 struct cgroup_seqfile_state {
2441 struct cgroup *cgroup;
2444 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2446 struct seq_file *sf = cb->state;
2447 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2450 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2452 struct cgroup_seqfile_state *state = m->private;
2453 struct cftype *cft = state->cft;
2454 if (cft->read_map) {
2455 struct cgroup_map_cb cb = {
2456 .fill = cgroup_map_add,
2459 return cft->read_map(state->cgroup, cft, &cb);
2461 return cft->read_seq_string(state->cgroup, cft, m);
2464 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2466 struct seq_file *seq = file->private_data;
2467 kfree(seq->private);
2468 return single_release(inode, file);
2471 static const struct file_operations cgroup_seqfile_operations = {
2473 .write = cgroup_file_write,
2474 .llseek = seq_lseek,
2475 .release = cgroup_seqfile_release,
2478 static int cgroup_file_open(struct inode *inode, struct file *file)
2483 err = generic_file_open(inode, file);
2486 cft = __d_cft(file->f_dentry);
2488 if (cft->read_map || cft->read_seq_string) {
2489 struct cgroup_seqfile_state *state =
2490 kzalloc(sizeof(*state), GFP_USER);
2494 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2495 file->f_op = &cgroup_seqfile_operations;
2496 err = single_open(file, cgroup_seqfile_show, state);
2499 } else if (cft->open)
2500 err = cft->open(inode, file);
2507 static int cgroup_file_release(struct inode *inode, struct file *file)
2509 struct cftype *cft = __d_cft(file->f_dentry);
2511 return cft->release(inode, file);
2516 * cgroup_rename - Only allow simple rename of directories in place.
2518 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2519 struct inode *new_dir, struct dentry *new_dentry)
2521 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2523 if (new_dentry->d_inode)
2525 if (old_dir != new_dir)
2527 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2530 static const struct file_operations cgroup_file_operations = {
2531 .read = cgroup_file_read,
2532 .write = cgroup_file_write,
2533 .llseek = generic_file_llseek,
2534 .open = cgroup_file_open,
2535 .release = cgroup_file_release,
2538 static const struct inode_operations cgroup_dir_inode_operations = {
2539 .lookup = cgroup_lookup,
2540 .mkdir = cgroup_mkdir,
2541 .rmdir = cgroup_rmdir,
2542 .rename = cgroup_rename,
2545 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2547 if (dentry->d_name.len > NAME_MAX)
2548 return ERR_PTR(-ENAMETOOLONG);
2549 d_add(dentry, NULL);
2554 * Check if a file is a control file
2556 static inline struct cftype *__file_cft(struct file *file)
2558 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2559 return ERR_PTR(-EINVAL);
2560 return __d_cft(file->f_dentry);
2563 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2564 struct super_block *sb)
2566 struct inode *inode;
2570 if (dentry->d_inode)
2573 inode = cgroup_new_inode(mode, sb);
2577 if (S_ISDIR(mode)) {
2578 inode->i_op = &cgroup_dir_inode_operations;
2579 inode->i_fop = &simple_dir_operations;
2581 /* start off with i_nlink == 2 (for "." entry) */
2584 /* start with the directory inode held, so that we can
2585 * populate it without racing with another mkdir */
2586 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2587 } else if (S_ISREG(mode)) {
2589 inode->i_fop = &cgroup_file_operations;
2591 d_instantiate(dentry, inode);
2592 dget(dentry); /* Extra count - pin the dentry in core */
2597 * cgroup_create_dir - create a directory for an object.
2598 * @cgrp: the cgroup we create the directory for. It must have a valid
2599 * ->parent field. And we are going to fill its ->dentry field.
2600 * @dentry: dentry of the new cgroup
2601 * @mode: mode to set on new directory.
2603 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2606 struct dentry *parent;
2609 parent = cgrp->parent->dentry;
2610 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2612 dentry->d_fsdata = cgrp;
2613 inc_nlink(parent->d_inode);
2614 rcu_assign_pointer(cgrp->dentry, dentry);
2623 * cgroup_file_mode - deduce file mode of a control file
2624 * @cft: the control file in question
2626 * returns cft->mode if ->mode is not 0
2627 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2628 * returns S_IRUGO if it has only a read handler
2629 * returns S_IWUSR if it has only a write hander
2631 static umode_t cgroup_file_mode(const struct cftype *cft)
2638 if (cft->read || cft->read_u64 || cft->read_s64 ||
2639 cft->read_map || cft->read_seq_string)
2642 if (cft->write || cft->write_u64 || cft->write_s64 ||
2643 cft->write_string || cft->trigger)
2649 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2650 const struct cftype *cft)
2652 struct dentry *dir = cgrp->dentry;
2653 struct cgroup *parent = __d_cgrp(dir);
2654 struct dentry *dentry;
2658 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2660 /* does @cft->flags tell us to skip creation on @cgrp? */
2661 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2663 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2666 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2667 strcpy(name, subsys->name);
2670 strcat(name, cft->name);
2672 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2674 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2678 dentry = lookup_one_len(name, dir, strlen(name));
2679 if (IS_ERR(dentry)) {
2680 error = PTR_ERR(dentry);
2684 mode = cgroup_file_mode(cft);
2685 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2687 cfe->type = (void *)cft;
2688 cfe->dentry = dentry;
2689 dentry->d_fsdata = cfe;
2690 list_add_tail(&cfe->node, &parent->files);
2699 static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2700 const struct cftype cfts[], bool is_add)
2702 const struct cftype *cft;
2705 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2707 err = cgroup_add_file(cgrp, subsys, cft);
2709 err = cgroup_rm_file(cgrp, cft);
2711 pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2712 is_add ? "add" : "remove", cft->name, err);
2719 static DEFINE_MUTEX(cgroup_cft_mutex);
2721 static void cgroup_cfts_prepare(void)
2722 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2725 * Thanks to the entanglement with vfs inode locking, we can't walk
2726 * the existing cgroups under cgroup_mutex and create files.
2727 * Instead, we increment reference on all cgroups and build list of
2728 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2729 * exclusive access to the field.
2731 mutex_lock(&cgroup_cft_mutex);
2732 mutex_lock(&cgroup_mutex);
2735 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2736 const struct cftype *cfts, bool is_add)
2737 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2740 struct cgroup *cgrp, *n;
2742 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2743 if (cfts && ss->root != &rootnode) {
2744 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2746 list_add_tail(&cgrp->cft_q_node, &pending);
2750 mutex_unlock(&cgroup_mutex);
2753 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2754 * files for all cgroups which were created before.
2756 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2757 struct inode *inode = cgrp->dentry->d_inode;
2759 mutex_lock(&inode->i_mutex);
2760 mutex_lock(&cgroup_mutex);
2761 if (!cgroup_is_removed(cgrp))
2762 cgroup_addrm_files(cgrp, ss, cfts, is_add);
2763 mutex_unlock(&cgroup_mutex);
2764 mutex_unlock(&inode->i_mutex);
2766 list_del_init(&cgrp->cft_q_node);
2770 mutex_unlock(&cgroup_cft_mutex);
2774 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2775 * @ss: target cgroup subsystem
2776 * @cfts: zero-length name terminated array of cftypes
2778 * Register @cfts to @ss. Files described by @cfts are created for all
2779 * existing cgroups to which @ss is attached and all future cgroups will
2780 * have them too. This function can be called anytime whether @ss is
2783 * Returns 0 on successful registration, -errno on failure. Note that this
2784 * function currently returns 0 as long as @cfts registration is successful
2785 * even if some file creation attempts on existing cgroups fail.
2787 int cgroup_add_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts)
2789 struct cftype_set *set;
2791 set = kzalloc(sizeof(*set), GFP_KERNEL);
2795 cgroup_cfts_prepare();
2797 list_add_tail(&set->node, &ss->cftsets);
2798 cgroup_cfts_commit(ss, cfts, true);
2802 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2805 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2806 * @ss: target cgroup subsystem
2807 * @cfts: zero-length name terminated array of cftypes
2809 * Unregister @cfts from @ss. Files described by @cfts are removed from
2810 * all existing cgroups to which @ss is attached and all future cgroups
2811 * won't have them either. This function can be called anytime whether @ss
2812 * is attached or not.
2814 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2815 * registered with @ss.
2817 int cgroup_rm_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts)
2819 struct cftype_set *set;
2821 cgroup_cfts_prepare();
2823 list_for_each_entry(set, &ss->cftsets, node) {
2824 if (set->cfts == cfts) {
2825 list_del_init(&set->node);
2826 cgroup_cfts_commit(ss, cfts, false);
2831 cgroup_cfts_commit(ss, NULL, false);
2836 * cgroup_task_count - count the number of tasks in a cgroup.
2837 * @cgrp: the cgroup in question
2839 * Return the number of tasks in the cgroup.
2841 int cgroup_task_count(const struct cgroup *cgrp)
2844 struct cg_cgroup_link *link;
2846 read_lock(&css_set_lock);
2847 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2848 count += atomic_read(&link->cg->refcount);
2850 read_unlock(&css_set_lock);
2855 * Advance a list_head iterator. The iterator should be positioned at
2856 * the start of a css_set
2858 static void cgroup_advance_iter(struct cgroup *cgrp,
2859 struct cgroup_iter *it)
2861 struct list_head *l = it->cg_link;
2862 struct cg_cgroup_link *link;
2865 /* Advance to the next non-empty css_set */
2868 if (l == &cgrp->css_sets) {
2872 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2874 } while (list_empty(&cg->tasks));
2876 it->task = cg->tasks.next;
2880 * To reduce the fork() overhead for systems that are not actually
2881 * using their cgroups capability, we don't maintain the lists running
2882 * through each css_set to its tasks until we see the list actually
2883 * used - in other words after the first call to cgroup_iter_start().
2885 static void cgroup_enable_task_cg_lists(void)
2887 struct task_struct *p, *g;
2888 write_lock(&css_set_lock);
2889 use_task_css_set_links = 1;
2891 * We need tasklist_lock because RCU is not safe against
2892 * while_each_thread(). Besides, a forking task that has passed
2893 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2894 * is not guaranteed to have its child immediately visible in the
2895 * tasklist if we walk through it with RCU.
2897 read_lock(&tasklist_lock);
2898 do_each_thread(g, p) {
2901 * We should check if the process is exiting, otherwise
2902 * it will race with cgroup_exit() in that the list
2903 * entry won't be deleted though the process has exited.
2905 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2906 list_add(&p->cg_list, &p->cgroups->tasks);
2908 } while_each_thread(g, p);
2909 read_unlock(&tasklist_lock);
2910 write_unlock(&css_set_lock);
2913 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2914 __acquires(css_set_lock)
2917 * The first time anyone tries to iterate across a cgroup,
2918 * we need to enable the list linking each css_set to its
2919 * tasks, and fix up all existing tasks.
2921 if (!use_task_css_set_links)
2922 cgroup_enable_task_cg_lists();
2924 read_lock(&css_set_lock);
2925 it->cg_link = &cgrp->css_sets;
2926 cgroup_advance_iter(cgrp, it);
2929 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2930 struct cgroup_iter *it)
2932 struct task_struct *res;
2933 struct list_head *l = it->task;
2934 struct cg_cgroup_link *link;
2936 /* If the iterator cg is NULL, we have no tasks */
2939 res = list_entry(l, struct task_struct, cg_list);
2940 /* Advance iterator to find next entry */
2942 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2943 if (l == &link->cg->tasks) {
2944 /* We reached the end of this task list - move on to
2945 * the next cg_cgroup_link */
2946 cgroup_advance_iter(cgrp, it);
2953 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2954 __releases(css_set_lock)
2956 read_unlock(&css_set_lock);
2959 static inline int started_after_time(struct task_struct *t1,
2960 struct timespec *time,
2961 struct task_struct *t2)
2963 int start_diff = timespec_compare(&t1->start_time, time);
2964 if (start_diff > 0) {
2966 } else if (start_diff < 0) {
2970 * Arbitrarily, if two processes started at the same
2971 * time, we'll say that the lower pointer value
2972 * started first. Note that t2 may have exited by now
2973 * so this may not be a valid pointer any longer, but
2974 * that's fine - it still serves to distinguish
2975 * between two tasks started (effectively) simultaneously.
2982 * This function is a callback from heap_insert() and is used to order
2984 * In this case we order the heap in descending task start time.
2986 static inline int started_after(void *p1, void *p2)
2988 struct task_struct *t1 = p1;
2989 struct task_struct *t2 = p2;
2990 return started_after_time(t1, &t2->start_time, t2);
2994 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2995 * @scan: struct cgroup_scanner containing arguments for the scan
2997 * Arguments include pointers to callback functions test_task() and
2999 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3000 * and if it returns true, call process_task() for it also.
3001 * The test_task pointer may be NULL, meaning always true (select all tasks).
3002 * Effectively duplicates cgroup_iter_{start,next,end}()
3003 * but does not lock css_set_lock for the call to process_task().
3004 * The struct cgroup_scanner may be embedded in any structure of the caller's
3006 * It is guaranteed that process_task() will act on every task that
3007 * is a member of the cgroup for the duration of this call. This
3008 * function may or may not call process_task() for tasks that exit
3009 * or move to a different cgroup during the call, or are forked or
3010 * move into the cgroup during the call.
3012 * Note that test_task() may be called with locks held, and may in some
3013 * situations be called multiple times for the same task, so it should
3015 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3016 * pre-allocated and will be used for heap operations (and its "gt" member will
3017 * be overwritten), else a temporary heap will be used (allocation of which
3018 * may cause this function to fail).
3020 int cgroup_scan_tasks(struct cgroup_scanner *scan)
3023 struct cgroup_iter it;
3024 struct task_struct *p, *dropped;
3025 /* Never dereference latest_task, since it's not refcounted */
3026 struct task_struct *latest_task = NULL;
3027 struct ptr_heap tmp_heap;
3028 struct ptr_heap *heap;
3029 struct timespec latest_time = { 0, 0 };
3032 /* The caller supplied our heap and pre-allocated its memory */
3034 heap->gt = &started_after;
3036 /* We need to allocate our own heap memory */
3038 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3040 /* cannot allocate the heap */
3046 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3047 * to determine which are of interest, and using the scanner's
3048 * "process_task" callback to process any of them that need an update.
3049 * Since we don't want to hold any locks during the task updates,
3050 * gather tasks to be processed in a heap structure.
3051 * The heap is sorted by descending task start time.
3052 * If the statically-sized heap fills up, we overflow tasks that
3053 * started later, and in future iterations only consider tasks that
3054 * started after the latest task in the previous pass. This
3055 * guarantees forward progress and that we don't miss any tasks.
3058 cgroup_iter_start(scan->cg, &it);
3059 while ((p = cgroup_iter_next(scan->cg, &it))) {
3061 * Only affect tasks that qualify per the caller's callback,
3062 * if he provided one
3064 if (scan->test_task && !scan->test_task(p, scan))
3067 * Only process tasks that started after the last task
3070 if (!started_after_time(p, &latest_time, latest_task))
3072 dropped = heap_insert(heap, p);
3073 if (dropped == NULL) {
3075 * The new task was inserted; the heap wasn't
3079 } else if (dropped != p) {
3081 * The new task was inserted, and pushed out a
3085 put_task_struct(dropped);
3088 * Else the new task was newer than anything already in
3089 * the heap and wasn't inserted
3092 cgroup_iter_end(scan->cg, &it);
3095 for (i = 0; i < heap->size; i++) {
3096 struct task_struct *q = heap->ptrs[i];
3098 latest_time = q->start_time;
3101 /* Process the task per the caller's callback */
3102 scan->process_task(q, scan);
3106 * If we had to process any tasks at all, scan again
3107 * in case some of them were in the middle of forking
3108 * children that didn't get processed.
3109 * Not the most efficient way to do it, but it avoids
3110 * having to take callback_mutex in the fork path
3114 if (heap == &tmp_heap)
3115 heap_free(&tmp_heap);
3120 * Stuff for reading the 'tasks'/'procs' files.
3122 * Reading this file can return large amounts of data if a cgroup has
3123 * *lots* of attached tasks. So it may need several calls to read(),
3124 * but we cannot guarantee that the information we produce is correct
3125 * unless we produce it entirely atomically.
3129 /* which pidlist file are we talking about? */
3130 enum cgroup_filetype {
3136 * A pidlist is a list of pids that virtually represents the contents of one
3137 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3138 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3141 struct cgroup_pidlist {
3143 * used to find which pidlist is wanted. doesn't change as long as
3144 * this particular list stays in the list.
3146 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3149 /* how many elements the above list has */
3151 /* how many files are using the current array */
3153 /* each of these stored in a list by its cgroup */
3154 struct list_head links;
3155 /* pointer to the cgroup we belong to, for list removal purposes */
3156 struct cgroup *owner;
3157 /* protects the other fields */
3158 struct rw_semaphore mutex;
3162 * The following two functions "fix" the issue where there are more pids
3163 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3164 * TODO: replace with a kernel-wide solution to this problem
3166 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3167 static void *pidlist_allocate(int count)
3169 if (PIDLIST_TOO_LARGE(count))
3170 return vmalloc(count * sizeof(pid_t));
3172 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3174 static void pidlist_free(void *p)
3176 if (is_vmalloc_addr(p))
3181 static void *pidlist_resize(void *p, int newcount)
3184 /* note: if new alloc fails, old p will still be valid either way */
3185 if (is_vmalloc_addr(p)) {
3186 newlist = vmalloc(newcount * sizeof(pid_t));
3189 memcpy(newlist, p, newcount * sizeof(pid_t));
3192 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3198 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3199 * If the new stripped list is sufficiently smaller and there's enough memory
3200 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3201 * number of unique elements.
3203 /* is the size difference enough that we should re-allocate the array? */
3204 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3205 static int pidlist_uniq(pid_t **p, int length)
3212 * we presume the 0th element is unique, so i starts at 1. trivial
3213 * edge cases first; no work needs to be done for either
3215 if (length == 0 || length == 1)
3217 /* src and dest walk down the list; dest counts unique elements */
3218 for (src = 1; src < length; src++) {
3219 /* find next unique element */
3220 while (list[src] == list[src-1]) {
3225 /* dest always points to where the next unique element goes */
3226 list[dest] = list[src];
3231 * if the length difference is large enough, we want to allocate a
3232 * smaller buffer to save memory. if this fails due to out of memory,
3233 * we'll just stay with what we've got.
3235 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3236 newlist = pidlist_resize(list, dest);
3243 static int cmppid(const void *a, const void *b)
3245 return *(pid_t *)a - *(pid_t *)b;
3249 * find the appropriate pidlist for our purpose (given procs vs tasks)
3250 * returns with the lock on that pidlist already held, and takes care
3251 * of the use count, or returns NULL with no locks held if we're out of
3254 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3255 enum cgroup_filetype type)
3257 struct cgroup_pidlist *l;
3258 /* don't need task_nsproxy() if we're looking at ourself */
3259 struct pid_namespace *ns = current->nsproxy->pid_ns;
3262 * We can't drop the pidlist_mutex before taking the l->mutex in case
3263 * the last ref-holder is trying to remove l from the list at the same
3264 * time. Holding the pidlist_mutex precludes somebody taking whichever
3265 * list we find out from under us - compare release_pid_array().
3267 mutex_lock(&cgrp->pidlist_mutex);
3268 list_for_each_entry(l, &cgrp->pidlists, links) {
3269 if (l->key.type == type && l->key.ns == ns) {
3270 /* make sure l doesn't vanish out from under us */
3271 down_write(&l->mutex);
3272 mutex_unlock(&cgrp->pidlist_mutex);
3276 /* entry not found; create a new one */
3277 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3279 mutex_unlock(&cgrp->pidlist_mutex);
3282 init_rwsem(&l->mutex);
3283 down_write(&l->mutex);
3285 l->key.ns = get_pid_ns(ns);
3286 l->use_count = 0; /* don't increment here */
3289 list_add(&l->links, &cgrp->pidlists);
3290 mutex_unlock(&cgrp->pidlist_mutex);
3295 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3297 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3298 struct cgroup_pidlist **lp)
3302 int pid, n = 0; /* used for populating the array */
3303 struct cgroup_iter it;
3304 struct task_struct *tsk;
3305 struct cgroup_pidlist *l;
3308 * If cgroup gets more users after we read count, we won't have
3309 * enough space - tough. This race is indistinguishable to the
3310 * caller from the case that the additional cgroup users didn't
3311 * show up until sometime later on.
3313 length = cgroup_task_count(cgrp);
3314 array = pidlist_allocate(length);
3317 /* now, populate the array */
3318 cgroup_iter_start(cgrp, &it);
3319 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3320 if (unlikely(n == length))
3322 /* get tgid or pid for procs or tasks file respectively */
3323 if (type == CGROUP_FILE_PROCS)
3324 pid = task_tgid_vnr(tsk);
3326 pid = task_pid_vnr(tsk);
3327 if (pid > 0) /* make sure to only use valid results */
3330 cgroup_iter_end(cgrp, &it);
3332 /* now sort & (if procs) strip out duplicates */
3333 sort(array, length, sizeof(pid_t), cmppid, NULL);
3334 if (type == CGROUP_FILE_PROCS)
3335 length = pidlist_uniq(&array, length);
3336 l = cgroup_pidlist_find(cgrp, type);
3338 pidlist_free(array);
3341 /* store array, freeing old if necessary - lock already held */
3342 pidlist_free(l->list);
3346 up_write(&l->mutex);
3352 * cgroupstats_build - build and fill cgroupstats
3353 * @stats: cgroupstats to fill information into
3354 * @dentry: A dentry entry belonging to the cgroup for which stats have
3357 * Build and fill cgroupstats so that taskstats can export it to user
3360 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3363 struct cgroup *cgrp;
3364 struct cgroup_iter it;
3365 struct task_struct *tsk;
3368 * Validate dentry by checking the superblock operations,
3369 * and make sure it's a directory.
3371 if (dentry->d_sb->s_op != &cgroup_ops ||
3372 !S_ISDIR(dentry->d_inode->i_mode))
3376 cgrp = dentry->d_fsdata;
3378 cgroup_iter_start(cgrp, &it);
3379 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3380 switch (tsk->state) {
3382 stats->nr_running++;
3384 case TASK_INTERRUPTIBLE:
3385 stats->nr_sleeping++;
3387 case TASK_UNINTERRUPTIBLE:
3388 stats->nr_uninterruptible++;
3391 stats->nr_stopped++;
3394 if (delayacct_is_task_waiting_on_io(tsk))
3395 stats->nr_io_wait++;
3399 cgroup_iter_end(cgrp, &it);
3407 * seq_file methods for the tasks/procs files. The seq_file position is the
3408 * next pid to display; the seq_file iterator is a pointer to the pid
3409 * in the cgroup->l->list array.
3412 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3415 * Initially we receive a position value that corresponds to
3416 * one more than the last pid shown (or 0 on the first call or
3417 * after a seek to the start). Use a binary-search to find the
3418 * next pid to display, if any
3420 struct cgroup_pidlist *l = s->private;
3421 int index = 0, pid = *pos;
3424 down_read(&l->mutex);
3426 int end = l->length;
3428 while (index < end) {
3429 int mid = (index + end) / 2;
3430 if (l->list[mid] == pid) {
3433 } else if (l->list[mid] <= pid)
3439 /* If we're off the end of the array, we're done */
3440 if (index >= l->length)
3442 /* Update the abstract position to be the actual pid that we found */
3443 iter = l->list + index;
3448 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3450 struct cgroup_pidlist *l = s->private;
3454 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3456 struct cgroup_pidlist *l = s->private;
3458 pid_t *end = l->list + l->length;
3460 * Advance to the next pid in the array. If this goes off the
3472 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3474 return seq_printf(s, "%d\n", *(int *)v);
3478 * seq_operations functions for iterating on pidlists through seq_file -
3479 * independent of whether it's tasks or procs
3481 static const struct seq_operations cgroup_pidlist_seq_operations = {
3482 .start = cgroup_pidlist_start,
3483 .stop = cgroup_pidlist_stop,
3484 .next = cgroup_pidlist_next,
3485 .show = cgroup_pidlist_show,
3488 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3491 * the case where we're the last user of this particular pidlist will
3492 * have us remove it from the cgroup's list, which entails taking the
3493 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3494 * pidlist_mutex, we have to take pidlist_mutex first.
3496 mutex_lock(&l->owner->pidlist_mutex);
3497 down_write(&l->mutex);
3498 BUG_ON(!l->use_count);
3499 if (!--l->use_count) {
3500 /* we're the last user if refcount is 0; remove and free */
3501 list_del(&l->links);
3502 mutex_unlock(&l->owner->pidlist_mutex);
3503 pidlist_free(l->list);
3504 put_pid_ns(l->key.ns);
3505 up_write(&l->mutex);
3509 mutex_unlock(&l->owner->pidlist_mutex);
3510 up_write(&l->mutex);
3513 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3515 struct cgroup_pidlist *l;
3516 if (!(file->f_mode & FMODE_READ))
3519 * the seq_file will only be initialized if the file was opened for
3520 * reading; hence we check if it's not null only in that case.
3522 l = ((struct seq_file *)file->private_data)->private;
3523 cgroup_release_pid_array(l);
3524 return seq_release(inode, file);
3527 static const struct file_operations cgroup_pidlist_operations = {
3529 .llseek = seq_lseek,
3530 .write = cgroup_file_write,
3531 .release = cgroup_pidlist_release,
3535 * The following functions handle opens on a file that displays a pidlist
3536 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3539 /* helper function for the two below it */
3540 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3542 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3543 struct cgroup_pidlist *l;
3546 /* Nothing to do for write-only files */
3547 if (!(file->f_mode & FMODE_READ))
3550 /* have the array populated */
3551 retval = pidlist_array_load(cgrp, type, &l);
3554 /* configure file information */
3555 file->f_op = &cgroup_pidlist_operations;
3557 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3559 cgroup_release_pid_array(l);
3562 ((struct seq_file *)file->private_data)->private = l;
3565 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3567 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3569 static int cgroup_procs_open(struct inode *unused, struct file *file)
3571 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3574 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3577 return notify_on_release(cgrp);
3580 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3584 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3586 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3588 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3593 * Unregister event and free resources.
3595 * Gets called from workqueue.
3597 static void cgroup_event_remove(struct work_struct *work)
3599 struct cgroup_event *event = container_of(work, struct cgroup_event,
3601 struct cgroup *cgrp = event->cgrp;
3603 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3605 eventfd_ctx_put(event->eventfd);
3611 * Gets called on POLLHUP on eventfd when user closes it.
3613 * Called with wqh->lock held and interrupts disabled.
3615 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3616 int sync, void *key)
3618 struct cgroup_event *event = container_of(wait,
3619 struct cgroup_event, wait);
3620 struct cgroup *cgrp = event->cgrp;
3621 unsigned long flags = (unsigned long)key;
3623 if (flags & POLLHUP) {
3624 __remove_wait_queue(event->wqh, &event->wait);
3625 spin_lock(&cgrp->event_list_lock);
3626 list_del(&event->list);
3627 spin_unlock(&cgrp->event_list_lock);
3629 * We are in atomic context, but cgroup_event_remove() may
3630 * sleep, so we have to call it in workqueue.
3632 schedule_work(&event->remove);
3638 static void cgroup_event_ptable_queue_proc(struct file *file,
3639 wait_queue_head_t *wqh, poll_table *pt)
3641 struct cgroup_event *event = container_of(pt,
3642 struct cgroup_event, pt);
3645 add_wait_queue(wqh, &event->wait);
3649 * Parse input and register new cgroup event handler.
3651 * Input must be in format '<event_fd> <control_fd> <args>'.
3652 * Interpretation of args is defined by control file implementation.
3654 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3657 struct cgroup_event *event = NULL;
3658 unsigned int efd, cfd;
3659 struct file *efile = NULL;
3660 struct file *cfile = NULL;
3664 efd = simple_strtoul(buffer, &endp, 10);
3669 cfd = simple_strtoul(buffer, &endp, 10);
3670 if ((*endp != ' ') && (*endp != '\0'))
3674 event = kzalloc(sizeof(*event), GFP_KERNEL);
3678 INIT_LIST_HEAD(&event->list);
3679 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3680 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3681 INIT_WORK(&event->remove, cgroup_event_remove);
3683 efile = eventfd_fget(efd);
3684 if (IS_ERR(efile)) {
3685 ret = PTR_ERR(efile);
3689 event->eventfd = eventfd_ctx_fileget(efile);
3690 if (IS_ERR(event->eventfd)) {
3691 ret = PTR_ERR(event->eventfd);
3701 /* the process need read permission on control file */
3702 /* AV: shouldn't we check that it's been opened for read instead? */
3703 ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3707 event->cft = __file_cft(cfile);
3708 if (IS_ERR(event->cft)) {
3709 ret = PTR_ERR(event->cft);
3713 if (!event->cft->register_event || !event->cft->unregister_event) {
3718 ret = event->cft->register_event(cgrp, event->cft,
3719 event->eventfd, buffer);
3723 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3724 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3730 * Events should be removed after rmdir of cgroup directory, but before
3731 * destroying subsystem state objects. Let's take reference to cgroup
3732 * directory dentry to do that.
3736 spin_lock(&cgrp->event_list_lock);
3737 list_add(&event->list, &cgrp->event_list);
3738 spin_unlock(&cgrp->event_list_lock);
3749 if (event && event->eventfd && !IS_ERR(event->eventfd))
3750 eventfd_ctx_put(event->eventfd);
3752 if (!IS_ERR_OR_NULL(efile))
3760 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3763 return clone_children(cgrp);
3766 static int cgroup_clone_children_write(struct cgroup *cgrp,
3771 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3773 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3778 * for the common functions, 'private' gives the type of file
3780 /* for hysterical raisins, we can't put this on the older files */
3781 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3782 static struct cftype files[] = {
3785 .open = cgroup_tasks_open,
3786 .write_u64 = cgroup_tasks_write,
3787 .release = cgroup_pidlist_release,
3788 .mode = S_IRUGO | S_IWUSR,
3791 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3792 .open = cgroup_procs_open,
3793 .write_u64 = cgroup_procs_write,
3794 .release = cgroup_pidlist_release,
3795 .mode = S_IRUGO | S_IWUSR,
3798 .name = "notify_on_release",
3799 .read_u64 = cgroup_read_notify_on_release,
3800 .write_u64 = cgroup_write_notify_on_release,
3803 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3804 .write_string = cgroup_write_event_control,
3808 .name = "cgroup.clone_children",
3809 .read_u64 = cgroup_clone_children_read,
3810 .write_u64 = cgroup_clone_children_write,
3813 .name = "release_agent",
3814 .flags = CFTYPE_ONLY_ON_ROOT,
3815 .read_seq_string = cgroup_release_agent_show,
3816 .write_string = cgroup_release_agent_write,
3817 .max_write_len = PATH_MAX,
3822 static int cgroup_populate_dir(struct cgroup *cgrp)
3825 struct cgroup_subsys *ss;
3827 err = cgroup_addrm_files(cgrp, NULL, files, true);
3831 /* process cftsets of each subsystem */
3832 for_each_subsys(cgrp->root, ss) {
3833 struct cftype_set *set;
3835 list_for_each_entry(set, &ss->cftsets, node)
3836 cgroup_addrm_files(cgrp, ss, set->cfts, true);
3839 /* This cgroup is ready now */
3840 for_each_subsys(cgrp->root, ss) {
3841 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3843 * Update id->css pointer and make this css visible from
3844 * CSS ID functions. This pointer will be dereferened
3845 * from RCU-read-side without locks.
3848 rcu_assign_pointer(css->id->css, css);
3854 static void css_dput_fn(struct work_struct *work)
3856 struct cgroup_subsys_state *css =
3857 container_of(work, struct cgroup_subsys_state, dput_work);
3858 struct dentry *dentry = css->cgroup->dentry;
3859 struct super_block *sb = dentry->d_sb;
3861 atomic_inc(&sb->s_active);
3863 deactivate_super(sb);
3866 static void init_cgroup_css(struct cgroup_subsys_state *css,
3867 struct cgroup_subsys *ss,
3868 struct cgroup *cgrp)
3871 atomic_set(&css->refcnt, 1);
3874 if (cgrp == dummytop)
3875 set_bit(CSS_ROOT, &css->flags);
3876 BUG_ON(cgrp->subsys[ss->subsys_id]);
3877 cgrp->subsys[ss->subsys_id] = css;
3880 * css holds an extra ref to @cgrp->dentry which is put on the last
3881 * css_put(). dput() requires process context, which css_put() may
3882 * be called without. @css->dput_work will be used to invoke
3883 * dput() asynchronously from css_put().
3885 INIT_WORK(&css->dput_work, css_dput_fn);
3889 * cgroup_create - create a cgroup
3890 * @parent: cgroup that will be parent of the new cgroup
3891 * @dentry: dentry of the new cgroup
3892 * @mode: mode to set on new inode
3894 * Must be called with the mutex on the parent inode held
3896 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3899 struct cgroup *cgrp;
3900 struct cgroupfs_root *root = parent->root;
3902 struct cgroup_subsys *ss;
3903 struct super_block *sb = root->sb;
3905 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3910 * Only live parents can have children. Note that the liveliness
3911 * check isn't strictly necessary because cgroup_mkdir() and
3912 * cgroup_rmdir() are fully synchronized by i_mutex; however, do it
3913 * anyway so that locking is contained inside cgroup proper and we
3914 * don't get nasty surprises if we ever grow another caller.
3916 if (!cgroup_lock_live_group(parent)) {
3921 /* Grab a reference on the superblock so the hierarchy doesn't
3922 * get deleted on unmount if there are child cgroups. This
3923 * can be done outside cgroup_mutex, since the sb can't
3924 * disappear while someone has an open control file on the
3926 atomic_inc(&sb->s_active);
3928 init_cgroup_housekeeping(cgrp);
3930 cgrp->parent = parent;
3931 cgrp->root = parent->root;
3932 cgrp->top_cgroup = parent->top_cgroup;
3934 if (notify_on_release(parent))
3935 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3937 if (clone_children(parent))
3938 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3940 for_each_subsys(root, ss) {
3941 struct cgroup_subsys_state *css = ss->create(cgrp);
3947 init_cgroup_css(css, ss, cgrp);
3949 err = alloc_css_id(ss, parent, cgrp);
3953 /* At error, ->destroy() callback has to free assigned ID. */
3954 if (clone_children(parent) && ss->post_clone)
3955 ss->post_clone(cgrp);
3958 list_add(&cgrp->sibling, &cgrp->parent->children);
3959 root->number_of_cgroups++;
3961 err = cgroup_create_dir(cgrp, dentry, mode);
3965 /* each css holds a ref to the cgroup's dentry */
3966 for_each_subsys(root, ss)
3969 /* The cgroup directory was pre-locked for us */
3970 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3972 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
3974 err = cgroup_populate_dir(cgrp);
3975 /* If err < 0, we have a half-filled directory - oh well ;) */
3977 mutex_unlock(&cgroup_mutex);
3978 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3984 list_del(&cgrp->sibling);
3985 root->number_of_cgroups--;
3989 for_each_subsys(root, ss) {
3990 if (cgrp->subsys[ss->subsys_id])
3994 mutex_unlock(&cgroup_mutex);
3996 /* Release the reference count that we took on the superblock */
3997 deactivate_super(sb);
4003 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4005 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
4007 /* the vfs holds inode->i_mutex already */
4008 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
4012 * Check the reference count on each subsystem. Since we already
4013 * established that there are no tasks in the cgroup, if the css refcount
4014 * is also 1, then there should be no outstanding references, so the
4015 * subsystem is safe to destroy. We scan across all subsystems rather than
4016 * using the per-hierarchy linked list of mounted subsystems since we can
4017 * be called via check_for_release() with no synchronization other than
4018 * RCU, and the subsystem linked list isn't RCU-safe.
4020 static int cgroup_has_css_refs(struct cgroup *cgrp)
4025 * We won't need to lock the subsys array, because the subsystems
4026 * we're concerned about aren't going anywhere since our cgroup root
4027 * has a reference on them.
4029 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4030 struct cgroup_subsys *ss = subsys[i];
4031 struct cgroup_subsys_state *css;
4033 /* Skip subsystems not present or not in this hierarchy */
4034 if (ss == NULL || ss->root != cgrp->root)
4037 css = cgrp->subsys[ss->subsys_id];
4039 * When called from check_for_release() it's possible
4040 * that by this point the cgroup has been removed
4041 * and the css deleted. But a false-positive doesn't
4042 * matter, since it can only happen if the cgroup
4043 * has been deleted and hence no longer needs the
4044 * release agent to be called anyway.
4046 if (css && css_refcnt(css) > 1)
4052 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4054 struct cgroup *cgrp = dentry->d_fsdata;
4056 struct cgroup *parent;
4058 struct cgroup_event *event, *tmp;
4059 struct cgroup_subsys *ss;
4062 * In general, subsystem has no css->refcnt after pre_destroy(). But
4063 * in racy cases, subsystem may have to get css->refcnt after
4064 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4065 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4066 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4067 * and subsystem's reference count handling. Please see css_get/put
4068 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4070 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4072 /* the vfs holds both inode->i_mutex already */
4073 mutex_lock(&cgroup_mutex);
4074 parent = cgrp->parent;
4075 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
4076 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4077 mutex_unlock(&cgroup_mutex);
4080 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
4083 * Block new css_tryget() by deactivating refcnt and mark @cgrp
4084 * removed. This makes future css_tryget() and child creation
4085 * attempts fail thus maintaining the removal conditions verified
4088 for_each_subsys(cgrp->root, ss) {
4089 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4091 WARN_ON(atomic_read(&css->refcnt) < 0);
4092 atomic_add(CSS_DEACT_BIAS, &css->refcnt);
4094 set_bit(CGRP_REMOVED, &cgrp->flags);
4097 * Tell subsystems to initate destruction. pre_destroy() should be
4098 * called with cgroup_mutex unlocked. See 3fa59dfbc3 ("cgroup: fix
4099 * potential deadlock in pre_destroy") for details.
4101 mutex_unlock(&cgroup_mutex);
4102 for_each_subsys(cgrp->root, ss)
4103 if (ss->pre_destroy)
4104 WARN_ON_ONCE(ss->pre_destroy(cgrp));
4105 mutex_lock(&cgroup_mutex);
4108 * Put all the base refs. Each css holds an extra reference to the
4109 * cgroup's dentry and cgroup removal proceeds regardless of css
4110 * refs. On the last put of each css, whenever that may be, the
4111 * extra dentry ref is put so that dentry destruction happens only
4112 * after all css's are released.
4114 for_each_subsys(cgrp->root, ss)
4115 css_put(cgrp->subsys[ss->subsys_id]);
4117 finish_wait(&cgroup_rmdir_waitq, &wait);
4118 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4120 raw_spin_lock(&release_list_lock);
4121 if (!list_empty(&cgrp->release_list))
4122 list_del_init(&cgrp->release_list);
4123 raw_spin_unlock(&release_list_lock);
4125 /* delete this cgroup from parent->children */
4126 list_del_init(&cgrp->sibling);
4128 list_del_init(&cgrp->allcg_node);
4130 d = dget(cgrp->dentry);
4132 cgroup_d_remove_dir(d);
4135 set_bit(CGRP_RELEASABLE, &parent->flags);
4136 check_for_release(parent);
4139 * Unregister events and notify userspace.
4140 * Notify userspace about cgroup removing only after rmdir of cgroup
4141 * directory to avoid race between userspace and kernelspace
4143 spin_lock(&cgrp->event_list_lock);
4144 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4145 list_del(&event->list);
4146 remove_wait_queue(event->wqh, &event->wait);
4147 eventfd_signal(event->eventfd, 1);
4148 schedule_work(&event->remove);
4150 spin_unlock(&cgrp->event_list_lock);
4152 mutex_unlock(&cgroup_mutex);
4156 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4158 INIT_LIST_HEAD(&ss->cftsets);
4161 * base_cftset is embedded in subsys itself, no need to worry about
4164 if (ss->base_cftypes) {
4165 ss->base_cftset.cfts = ss->base_cftypes;
4166 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4170 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4172 struct cgroup_subsys_state *css;
4174 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4176 /* init base cftset */
4177 cgroup_init_cftsets(ss);
4179 /* Create the top cgroup state for this subsystem */
4180 list_add(&ss->sibling, &rootnode.subsys_list);
4181 ss->root = &rootnode;
4182 css = ss->create(dummytop);
4183 /* We don't handle early failures gracefully */
4184 BUG_ON(IS_ERR(css));
4185 init_cgroup_css(css, ss, dummytop);
4187 /* Update the init_css_set to contain a subsys
4188 * pointer to this state - since the subsystem is
4189 * newly registered, all tasks and hence the
4190 * init_css_set is in the subsystem's top cgroup. */
4191 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4193 need_forkexit_callback |= ss->fork || ss->exit;
4195 /* At system boot, before all subsystems have been
4196 * registered, no tasks have been forked, so we don't
4197 * need to invoke fork callbacks here. */
4198 BUG_ON(!list_empty(&init_task.tasks));
4202 /* this function shouldn't be used with modular subsystems, since they
4203 * need to register a subsys_id, among other things */
4208 * cgroup_load_subsys: load and register a modular subsystem at runtime
4209 * @ss: the subsystem to load
4211 * This function should be called in a modular subsystem's initcall. If the
4212 * subsystem is built as a module, it will be assigned a new subsys_id and set
4213 * up for use. If the subsystem is built-in anyway, work is delegated to the
4214 * simpler cgroup_init_subsys.
4216 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4219 struct cgroup_subsys_state *css;
4221 /* check name and function validity */
4222 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4223 ss->create == NULL || ss->destroy == NULL)
4227 * we don't support callbacks in modular subsystems. this check is
4228 * before the ss->module check for consistency; a subsystem that could
4229 * be a module should still have no callbacks even if the user isn't
4230 * compiling it as one.
4232 if (ss->fork || ss->exit)
4236 * an optionally modular subsystem is built-in: we want to do nothing,
4237 * since cgroup_init_subsys will have already taken care of it.
4239 if (ss->module == NULL) {
4240 /* a few sanity checks */
4241 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4242 BUG_ON(subsys[ss->subsys_id] != ss);
4246 /* init base cftset */
4247 cgroup_init_cftsets(ss);
4250 * need to register a subsys id before anything else - for example,
4251 * init_cgroup_css needs it.
4253 mutex_lock(&cgroup_mutex);
4254 /* find the first empty slot in the array */
4255 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4256 if (subsys[i] == NULL)
4259 if (i == CGROUP_SUBSYS_COUNT) {
4260 /* maximum number of subsystems already registered! */
4261 mutex_unlock(&cgroup_mutex);
4264 /* assign ourselves the subsys_id */
4269 * no ss->create seems to need anything important in the ss struct, so
4270 * this can happen first (i.e. before the rootnode attachment).
4272 css = ss->create(dummytop);
4274 /* failure case - need to deassign the subsys[] slot. */
4276 mutex_unlock(&cgroup_mutex);
4277 return PTR_ERR(css);
4280 list_add(&ss->sibling, &rootnode.subsys_list);
4281 ss->root = &rootnode;
4283 /* our new subsystem will be attached to the dummy hierarchy. */
4284 init_cgroup_css(css, ss, dummytop);
4285 /* init_idr must be after init_cgroup_css because it sets css->id. */
4287 int ret = cgroup_init_idr(ss, css);
4289 dummytop->subsys[ss->subsys_id] = NULL;
4290 ss->destroy(dummytop);
4292 mutex_unlock(&cgroup_mutex);
4298 * Now we need to entangle the css into the existing css_sets. unlike
4299 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4300 * will need a new pointer to it; done by iterating the css_set_table.
4301 * furthermore, modifying the existing css_sets will corrupt the hash
4302 * table state, so each changed css_set will need its hash recomputed.
4303 * this is all done under the css_set_lock.
4305 write_lock(&css_set_lock);
4306 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4308 struct hlist_node *node, *tmp;
4309 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4311 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4312 /* skip entries that we already rehashed */
4313 if (cg->subsys[ss->subsys_id])
4315 /* remove existing entry */
4316 hlist_del(&cg->hlist);
4318 cg->subsys[ss->subsys_id] = css;
4319 /* recompute hash and restore entry */
4320 new_bucket = css_set_hash(cg->subsys);
4321 hlist_add_head(&cg->hlist, new_bucket);
4324 write_unlock(&css_set_lock);
4329 mutex_unlock(&cgroup_mutex);
4332 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4335 * cgroup_unload_subsys: unload a modular subsystem
4336 * @ss: the subsystem to unload
4338 * This function should be called in a modular subsystem's exitcall. When this
4339 * function is invoked, the refcount on the subsystem's module will be 0, so
4340 * the subsystem will not be attached to any hierarchy.
4342 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4344 struct cg_cgroup_link *link;
4345 struct hlist_head *hhead;
4347 BUG_ON(ss->module == NULL);
4350 * we shouldn't be called if the subsystem is in use, and the use of
4351 * try_module_get in parse_cgroupfs_options should ensure that it
4352 * doesn't start being used while we're killing it off.
4354 BUG_ON(ss->root != &rootnode);
4356 mutex_lock(&cgroup_mutex);
4357 /* deassign the subsys_id */
4358 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4359 subsys[ss->subsys_id] = NULL;
4361 /* remove subsystem from rootnode's list of subsystems */
4362 list_del_init(&ss->sibling);
4365 * disentangle the css from all css_sets attached to the dummytop. as
4366 * in loading, we need to pay our respects to the hashtable gods.
4368 write_lock(&css_set_lock);
4369 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4370 struct css_set *cg = link->cg;
4372 hlist_del(&cg->hlist);
4373 BUG_ON(!cg->subsys[ss->subsys_id]);
4374 cg->subsys[ss->subsys_id] = NULL;
4375 hhead = css_set_hash(cg->subsys);
4376 hlist_add_head(&cg->hlist, hhead);
4378 write_unlock(&css_set_lock);
4381 * remove subsystem's css from the dummytop and free it - need to free
4382 * before marking as null because ss->destroy needs the cgrp->subsys
4383 * pointer to find their state. note that this also takes care of
4384 * freeing the css_id.
4386 ss->destroy(dummytop);
4387 dummytop->subsys[ss->subsys_id] = NULL;
4389 mutex_unlock(&cgroup_mutex);
4391 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4394 * cgroup_init_early - cgroup initialization at system boot
4396 * Initialize cgroups at system boot, and initialize any
4397 * subsystems that request early init.
4399 int __init cgroup_init_early(void)
4402 atomic_set(&init_css_set.refcount, 1);
4403 INIT_LIST_HEAD(&init_css_set.cg_links);
4404 INIT_LIST_HEAD(&init_css_set.tasks);
4405 INIT_HLIST_NODE(&init_css_set.hlist);
4407 init_cgroup_root(&rootnode);
4409 init_task.cgroups = &init_css_set;
4411 init_css_set_link.cg = &init_css_set;
4412 init_css_set_link.cgrp = dummytop;
4413 list_add(&init_css_set_link.cgrp_link_list,
4414 &rootnode.top_cgroup.css_sets);
4415 list_add(&init_css_set_link.cg_link_list,
4416 &init_css_set.cg_links);
4418 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4419 INIT_HLIST_HEAD(&css_set_table[i]);
4421 /* at bootup time, we don't worry about modular subsystems */
4422 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4423 struct cgroup_subsys *ss = subsys[i];
4426 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4427 BUG_ON(!ss->create);
4428 BUG_ON(!ss->destroy);
4429 if (ss->subsys_id != i) {
4430 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4431 ss->name, ss->subsys_id);
4436 cgroup_init_subsys(ss);
4442 * cgroup_init - cgroup initialization
4444 * Register cgroup filesystem and /proc file, and initialize
4445 * any subsystems that didn't request early init.
4447 int __init cgroup_init(void)
4451 struct hlist_head *hhead;
4453 err = bdi_init(&cgroup_backing_dev_info);
4457 /* at bootup time, we don't worry about modular subsystems */
4458 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4459 struct cgroup_subsys *ss = subsys[i];
4460 if (!ss->early_init)
4461 cgroup_init_subsys(ss);
4463 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4466 /* Add init_css_set to the hash table */
4467 hhead = css_set_hash(init_css_set.subsys);
4468 hlist_add_head(&init_css_set.hlist, hhead);
4469 BUG_ON(!init_root_id(&rootnode));
4471 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4477 err = register_filesystem(&cgroup_fs_type);
4479 kobject_put(cgroup_kobj);
4483 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4487 bdi_destroy(&cgroup_backing_dev_info);
4493 * proc_cgroup_show()
4494 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4495 * - Used for /proc/<pid>/cgroup.
4496 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4497 * doesn't really matter if tsk->cgroup changes after we read it,
4498 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4499 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4500 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4501 * cgroup to top_cgroup.
4504 /* TODO: Use a proper seq_file iterator */
4505 static int proc_cgroup_show(struct seq_file *m, void *v)
4508 struct task_struct *tsk;
4511 struct cgroupfs_root *root;
4514 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4520 tsk = get_pid_task(pid, PIDTYPE_PID);
4526 mutex_lock(&cgroup_mutex);
4528 for_each_active_root(root) {
4529 struct cgroup_subsys *ss;
4530 struct cgroup *cgrp;
4533 seq_printf(m, "%d:", root->hierarchy_id);
4534 for_each_subsys(root, ss)
4535 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4536 if (strlen(root->name))
4537 seq_printf(m, "%sname=%s", count ? "," : "",
4540 cgrp = task_cgroup_from_root(tsk, root);
4541 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4549 mutex_unlock(&cgroup_mutex);
4550 put_task_struct(tsk);
4557 static int cgroup_open(struct inode *inode, struct file *file)
4559 struct pid *pid = PROC_I(inode)->pid;
4560 return single_open(file, proc_cgroup_show, pid);
4563 const struct file_operations proc_cgroup_operations = {
4564 .open = cgroup_open,
4566 .llseek = seq_lseek,
4567 .release = single_release,
4570 /* Display information about each subsystem and each hierarchy */
4571 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4575 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4577 * ideally we don't want subsystems moving around while we do this.
4578 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4579 * subsys/hierarchy state.
4581 mutex_lock(&cgroup_mutex);
4582 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4583 struct cgroup_subsys *ss = subsys[i];
4586 seq_printf(m, "%s\t%d\t%d\t%d\n",
4587 ss->name, ss->root->hierarchy_id,
4588 ss->root->number_of_cgroups, !ss->disabled);
4590 mutex_unlock(&cgroup_mutex);
4594 static int cgroupstats_open(struct inode *inode, struct file *file)
4596 return single_open(file, proc_cgroupstats_show, NULL);
4599 static const struct file_operations proc_cgroupstats_operations = {
4600 .open = cgroupstats_open,
4602 .llseek = seq_lseek,
4603 .release = single_release,
4607 * cgroup_fork - attach newly forked task to its parents cgroup.
4608 * @child: pointer to task_struct of forking parent process.
4610 * Description: A task inherits its parent's cgroup at fork().
4612 * A pointer to the shared css_set was automatically copied in
4613 * fork.c by dup_task_struct(). However, we ignore that copy, since
4614 * it was not made under the protection of RCU, cgroup_mutex or
4615 * threadgroup_change_begin(), so it might no longer be a valid
4616 * cgroup pointer. cgroup_attach_task() might have already changed
4617 * current->cgroups, allowing the previously referenced cgroup
4618 * group to be removed and freed.
4620 * Outside the pointer validity we also need to process the css_set
4621 * inheritance between threadgoup_change_begin() and
4622 * threadgoup_change_end(), this way there is no leak in any process
4623 * wide migration performed by cgroup_attach_proc() that could otherwise
4624 * miss a thread because it is too early or too late in the fork stage.
4626 * At the point that cgroup_fork() is called, 'current' is the parent
4627 * task, and the passed argument 'child' points to the child task.
4629 void cgroup_fork(struct task_struct *child)
4632 * We don't need to task_lock() current because current->cgroups
4633 * can't be changed concurrently here. The parent obviously hasn't
4634 * exited and called cgroup_exit(), and we are synchronized against
4635 * cgroup migration through threadgroup_change_begin().
4637 child->cgroups = current->cgroups;
4638 get_css_set(child->cgroups);
4639 INIT_LIST_HEAD(&child->cg_list);
4643 * cgroup_fork_callbacks - run fork callbacks
4644 * @child: the new task
4646 * Called on a new task very soon before adding it to the
4647 * tasklist. No need to take any locks since no-one can
4648 * be operating on this task.
4650 void cgroup_fork_callbacks(struct task_struct *child)
4652 if (need_forkexit_callback) {
4655 * forkexit callbacks are only supported for builtin
4656 * subsystems, and the builtin section of the subsys array is
4657 * immutable, so we don't need to lock the subsys array here.
4659 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4660 struct cgroup_subsys *ss = subsys[i];
4668 * cgroup_post_fork - called on a new task after adding it to the task list
4669 * @child: the task in question
4671 * Adds the task to the list running through its css_set if necessary.
4672 * Has to be after the task is visible on the task list in case we race
4673 * with the first call to cgroup_iter_start() - to guarantee that the
4674 * new task ends up on its list.
4676 void cgroup_post_fork(struct task_struct *child)
4679 * use_task_css_set_links is set to 1 before we walk the tasklist
4680 * under the tasklist_lock and we read it here after we added the child
4681 * to the tasklist under the tasklist_lock as well. If the child wasn't
4682 * yet in the tasklist when we walked through it from
4683 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4684 * should be visible now due to the paired locking and barriers implied
4685 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4686 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4689 if (use_task_css_set_links) {
4690 write_lock(&css_set_lock);
4691 if (list_empty(&child->cg_list)) {
4693 * It's safe to use child->cgroups without task_lock()
4694 * here because we are protected through
4695 * threadgroup_change_begin() against concurrent
4696 * css_set change in cgroup_task_migrate(). Also
4697 * the task can't exit at that point until
4698 * wake_up_new_task() is called, so we are protected
4699 * against cgroup_exit() setting child->cgroup to
4702 list_add(&child->cg_list, &child->cgroups->tasks);
4704 write_unlock(&css_set_lock);
4708 * cgroup_exit - detach cgroup from exiting task
4709 * @tsk: pointer to task_struct of exiting process
4710 * @run_callback: run exit callbacks?
4712 * Description: Detach cgroup from @tsk and release it.
4714 * Note that cgroups marked notify_on_release force every task in
4715 * them to take the global cgroup_mutex mutex when exiting.
4716 * This could impact scaling on very large systems. Be reluctant to
4717 * use notify_on_release cgroups where very high task exit scaling
4718 * is required on large systems.
4720 * the_top_cgroup_hack:
4722 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4724 * We call cgroup_exit() while the task is still competent to
4725 * handle notify_on_release(), then leave the task attached to the
4726 * root cgroup in each hierarchy for the remainder of its exit.
4728 * To do this properly, we would increment the reference count on
4729 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4730 * code we would add a second cgroup function call, to drop that
4731 * reference. This would just create an unnecessary hot spot on
4732 * the top_cgroup reference count, to no avail.
4734 * Normally, holding a reference to a cgroup without bumping its
4735 * count is unsafe. The cgroup could go away, or someone could
4736 * attach us to a different cgroup, decrementing the count on
4737 * the first cgroup that we never incremented. But in this case,
4738 * top_cgroup isn't going away, and either task has PF_EXITING set,
4739 * which wards off any cgroup_attach_task() attempts, or task is a failed
4740 * fork, never visible to cgroup_attach_task.
4742 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4748 * Unlink from the css_set task list if necessary.
4749 * Optimistically check cg_list before taking
4752 if (!list_empty(&tsk->cg_list)) {
4753 write_lock(&css_set_lock);
4754 if (!list_empty(&tsk->cg_list))
4755 list_del_init(&tsk->cg_list);
4756 write_unlock(&css_set_lock);
4759 /* Reassign the task to the init_css_set. */
4762 tsk->cgroups = &init_css_set;
4764 if (run_callbacks && need_forkexit_callback) {
4766 * modular subsystems can't use callbacks, so no need to lock
4769 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4770 struct cgroup_subsys *ss = subsys[i];
4772 struct cgroup *old_cgrp =
4773 rcu_dereference_raw(cg->subsys[i])->cgroup;
4774 struct cgroup *cgrp = task_cgroup(tsk, i);
4775 ss->exit(cgrp, old_cgrp, tsk);
4782 put_css_set_taskexit(cg);
4786 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4787 * @cgrp: the cgroup in question
4788 * @task: the task in question
4790 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4793 * If we are sending in dummytop, then presumably we are creating
4794 * the top cgroup in the subsystem.
4796 * Called only by the ns (nsproxy) cgroup.
4798 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4801 struct cgroup *target;
4803 if (cgrp == dummytop)
4806 target = task_cgroup_from_root(task, cgrp->root);
4807 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4808 cgrp = cgrp->parent;
4809 ret = (cgrp == target);
4813 static void check_for_release(struct cgroup *cgrp)
4815 /* All of these checks rely on RCU to keep the cgroup
4816 * structure alive */
4817 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4818 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4819 /* Control Group is currently removeable. If it's not
4820 * already queued for a userspace notification, queue
4822 int need_schedule_work = 0;
4823 raw_spin_lock(&release_list_lock);
4824 if (!cgroup_is_removed(cgrp) &&
4825 list_empty(&cgrp->release_list)) {
4826 list_add(&cgrp->release_list, &release_list);
4827 need_schedule_work = 1;
4829 raw_spin_unlock(&release_list_lock);
4830 if (need_schedule_work)
4831 schedule_work(&release_agent_work);
4835 /* Caller must verify that the css is not for root cgroup */
4836 bool __css_tryget(struct cgroup_subsys_state *css)
4841 v = css_refcnt(css);
4842 t = atomic_cmpxchg(&css->refcnt, v, v + 1);
4850 EXPORT_SYMBOL_GPL(__css_tryget);
4852 /* Caller must verify that the css is not for root cgroup */
4853 void __css_put(struct cgroup_subsys_state *css)
4855 struct cgroup *cgrp = css->cgroup;
4859 v = css_unbias_refcnt(atomic_dec_return(&css->refcnt));
4863 if (notify_on_release(cgrp)) {
4864 set_bit(CGRP_RELEASABLE, &cgrp->flags);
4865 check_for_release(cgrp);
4867 cgroup_wakeup_rmdir_waiter(cgrp);
4870 schedule_work(&css->dput_work);
4875 EXPORT_SYMBOL_GPL(__css_put);
4878 * Notify userspace when a cgroup is released, by running the
4879 * configured release agent with the name of the cgroup (path
4880 * relative to the root of cgroup file system) as the argument.
4882 * Most likely, this user command will try to rmdir this cgroup.
4884 * This races with the possibility that some other task will be
4885 * attached to this cgroup before it is removed, or that some other
4886 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4887 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4888 * unused, and this cgroup will be reprieved from its death sentence,
4889 * to continue to serve a useful existence. Next time it's released,
4890 * we will get notified again, if it still has 'notify_on_release' set.
4892 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4893 * means only wait until the task is successfully execve()'d. The
4894 * separate release agent task is forked by call_usermodehelper(),
4895 * then control in this thread returns here, without waiting for the
4896 * release agent task. We don't bother to wait because the caller of
4897 * this routine has no use for the exit status of the release agent
4898 * task, so no sense holding our caller up for that.
4900 static void cgroup_release_agent(struct work_struct *work)
4902 BUG_ON(work != &release_agent_work);
4903 mutex_lock(&cgroup_mutex);
4904 raw_spin_lock(&release_list_lock);
4905 while (!list_empty(&release_list)) {
4906 char *argv[3], *envp[3];
4908 char *pathbuf = NULL, *agentbuf = NULL;
4909 struct cgroup *cgrp = list_entry(release_list.next,
4912 list_del_init(&cgrp->release_list);
4913 raw_spin_unlock(&release_list_lock);
4914 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4917 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4919 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4924 argv[i++] = agentbuf;
4925 argv[i++] = pathbuf;
4929 /* minimal command environment */
4930 envp[i++] = "HOME=/";
4931 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4934 /* Drop the lock while we invoke the usermode helper,
4935 * since the exec could involve hitting disk and hence
4936 * be a slow process */
4937 mutex_unlock(&cgroup_mutex);
4938 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4939 mutex_lock(&cgroup_mutex);
4943 raw_spin_lock(&release_list_lock);
4945 raw_spin_unlock(&release_list_lock);
4946 mutex_unlock(&cgroup_mutex);
4949 static int __init cgroup_disable(char *str)
4954 while ((token = strsep(&str, ",")) != NULL) {
4958 * cgroup_disable, being at boot time, can't know about module
4959 * subsystems, so we don't worry about them.
4961 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4962 struct cgroup_subsys *ss = subsys[i];
4964 if (!strcmp(token, ss->name)) {
4966 printk(KERN_INFO "Disabling %s control group"
4967 " subsystem\n", ss->name);
4974 __setup("cgroup_disable=", cgroup_disable);
4977 * Functons for CSS ID.
4981 *To get ID other than 0, this should be called when !cgroup_is_removed().
4983 unsigned short css_id(struct cgroup_subsys_state *css)
4985 struct css_id *cssid;
4988 * This css_id() can return correct value when somone has refcnt
4989 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4990 * it's unchanged until freed.
4992 cssid = rcu_dereference_check(css->id, css_refcnt(css));
4998 EXPORT_SYMBOL_GPL(css_id);
5000 unsigned short css_depth(struct cgroup_subsys_state *css)
5002 struct css_id *cssid;
5004 cssid = rcu_dereference_check(css->id, css_refcnt(css));
5007 return cssid->depth;
5010 EXPORT_SYMBOL_GPL(css_depth);
5013 * css_is_ancestor - test "root" css is an ancestor of "child"
5014 * @child: the css to be tested.
5015 * @root: the css supporsed to be an ancestor of the child.
5017 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5018 * this function reads css->id, the caller must hold rcu_read_lock().
5019 * But, considering usual usage, the csses should be valid objects after test.
5020 * Assuming that the caller will do some action to the child if this returns
5021 * returns true, the caller must take "child";s reference count.
5022 * If "child" is valid object and this returns true, "root" is valid, too.
5025 bool css_is_ancestor(struct cgroup_subsys_state *child,
5026 const struct cgroup_subsys_state *root)
5028 struct css_id *child_id;
5029 struct css_id *root_id;
5031 child_id = rcu_dereference(child->id);
5034 root_id = rcu_dereference(root->id);
5037 if (child_id->depth < root_id->depth)
5039 if (child_id->stack[root_id->depth] != root_id->id)
5044 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5046 struct css_id *id = css->id;
5047 /* When this is called before css_id initialization, id can be NULL */
5051 BUG_ON(!ss->use_id);
5053 rcu_assign_pointer(id->css, NULL);
5054 rcu_assign_pointer(css->id, NULL);
5055 spin_lock(&ss->id_lock);
5056 idr_remove(&ss->idr, id->id);
5057 spin_unlock(&ss->id_lock);
5058 kfree_rcu(id, rcu_head);
5060 EXPORT_SYMBOL_GPL(free_css_id);
5063 * This is called by init or create(). Then, calls to this function are
5064 * always serialized (By cgroup_mutex() at create()).
5067 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5069 struct css_id *newid;
5070 int myid, error, size;
5072 BUG_ON(!ss->use_id);
5074 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5075 newid = kzalloc(size, GFP_KERNEL);
5077 return ERR_PTR(-ENOMEM);
5079 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
5083 spin_lock(&ss->id_lock);
5084 /* Don't use 0. allocates an ID of 1-65535 */
5085 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
5086 spin_unlock(&ss->id_lock);
5088 /* Returns error when there are no free spaces for new ID.*/
5093 if (myid > CSS_ID_MAX)
5097 newid->depth = depth;
5101 spin_lock(&ss->id_lock);
5102 idr_remove(&ss->idr, myid);
5103 spin_unlock(&ss->id_lock);
5106 return ERR_PTR(error);
5110 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5111 struct cgroup_subsys_state *rootcss)
5113 struct css_id *newid;
5115 spin_lock_init(&ss->id_lock);
5118 newid = get_new_cssid(ss, 0);
5120 return PTR_ERR(newid);
5122 newid->stack[0] = newid->id;
5123 newid->css = rootcss;
5124 rootcss->id = newid;
5128 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5129 struct cgroup *child)
5131 int subsys_id, i, depth = 0;
5132 struct cgroup_subsys_state *parent_css, *child_css;
5133 struct css_id *child_id, *parent_id;
5135 subsys_id = ss->subsys_id;
5136 parent_css = parent->subsys[subsys_id];
5137 child_css = child->subsys[subsys_id];
5138 parent_id = parent_css->id;
5139 depth = parent_id->depth + 1;
5141 child_id = get_new_cssid(ss, depth);
5142 if (IS_ERR(child_id))
5143 return PTR_ERR(child_id);
5145 for (i = 0; i < depth; i++)
5146 child_id->stack[i] = parent_id->stack[i];
5147 child_id->stack[depth] = child_id->id;
5149 * child_id->css pointer will be set after this cgroup is available
5150 * see cgroup_populate_dir()
5152 rcu_assign_pointer(child_css->id, child_id);
5158 * css_lookup - lookup css by id
5159 * @ss: cgroup subsys to be looked into.
5162 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5163 * NULL if not. Should be called under rcu_read_lock()
5165 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5167 struct css_id *cssid = NULL;
5169 BUG_ON(!ss->use_id);
5170 cssid = idr_find(&ss->idr, id);
5172 if (unlikely(!cssid))
5175 return rcu_dereference(cssid->css);
5177 EXPORT_SYMBOL_GPL(css_lookup);
5180 * css_get_next - lookup next cgroup under specified hierarchy.
5181 * @ss: pointer to subsystem
5182 * @id: current position of iteration.
5183 * @root: pointer to css. search tree under this.
5184 * @foundid: position of found object.
5186 * Search next css under the specified hierarchy of rootid. Calling under
5187 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5189 struct cgroup_subsys_state *
5190 css_get_next(struct cgroup_subsys *ss, int id,
5191 struct cgroup_subsys_state *root, int *foundid)
5193 struct cgroup_subsys_state *ret = NULL;
5196 int rootid = css_id(root);
5197 int depth = css_depth(root);
5202 BUG_ON(!ss->use_id);
5203 WARN_ON_ONCE(!rcu_read_lock_held());
5205 /* fill start point for scan */
5209 * scan next entry from bitmap(tree), tmpid is updated after
5212 tmp = idr_get_next(&ss->idr, &tmpid);
5215 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5216 ret = rcu_dereference(tmp->css);
5222 /* continue to scan from next id */
5229 * get corresponding css from file open on cgroupfs directory
5231 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5233 struct cgroup *cgrp;
5234 struct inode *inode;
5235 struct cgroup_subsys_state *css;
5237 inode = f->f_dentry->d_inode;
5238 /* check in cgroup filesystem dir */
5239 if (inode->i_op != &cgroup_dir_inode_operations)
5240 return ERR_PTR(-EBADF);
5242 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5243 return ERR_PTR(-EINVAL);
5246 cgrp = __d_cgrp(f->f_dentry);
5247 css = cgrp->subsys[id];
5248 return css ? css : ERR_PTR(-ENOENT);
5251 #ifdef CONFIG_CGROUP_DEBUG
5252 static struct cgroup_subsys_state *debug_create(struct cgroup *cont)
5254 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5257 return ERR_PTR(-ENOMEM);
5262 static void debug_destroy(struct cgroup *cont)
5264 kfree(cont->subsys[debug_subsys_id]);
5267 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5269 return atomic_read(&cont->count);
5272 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5274 return cgroup_task_count(cont);
5277 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5279 return (u64)(unsigned long)current->cgroups;
5282 static u64 current_css_set_refcount_read(struct cgroup *cont,
5288 count = atomic_read(¤t->cgroups->refcount);
5293 static int current_css_set_cg_links_read(struct cgroup *cont,
5295 struct seq_file *seq)
5297 struct cg_cgroup_link *link;
5300 read_lock(&css_set_lock);
5302 cg = rcu_dereference(current->cgroups);
5303 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5304 struct cgroup *c = link->cgrp;
5308 name = c->dentry->d_name.name;
5311 seq_printf(seq, "Root %d group %s\n",
5312 c->root->hierarchy_id, name);
5315 read_unlock(&css_set_lock);
5319 #define MAX_TASKS_SHOWN_PER_CSS 25
5320 static int cgroup_css_links_read(struct cgroup *cont,
5322 struct seq_file *seq)
5324 struct cg_cgroup_link *link;
5326 read_lock(&css_set_lock);
5327 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5328 struct css_set *cg = link->cg;
5329 struct task_struct *task;
5331 seq_printf(seq, "css_set %p\n", cg);
5332 list_for_each_entry(task, &cg->tasks, cg_list) {
5333 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5334 seq_puts(seq, " ...\n");
5337 seq_printf(seq, " task %d\n",
5338 task_pid_vnr(task));
5342 read_unlock(&css_set_lock);
5346 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5348 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5351 static struct cftype debug_files[] = {
5353 .name = "cgroup_refcount",
5354 .read_u64 = cgroup_refcount_read,
5357 .name = "taskcount",
5358 .read_u64 = debug_taskcount_read,
5362 .name = "current_css_set",
5363 .read_u64 = current_css_set_read,
5367 .name = "current_css_set_refcount",
5368 .read_u64 = current_css_set_refcount_read,
5372 .name = "current_css_set_cg_links",
5373 .read_seq_string = current_css_set_cg_links_read,
5377 .name = "cgroup_css_links",
5378 .read_seq_string = cgroup_css_links_read,
5382 .name = "releasable",
5383 .read_u64 = releasable_read,
5389 struct cgroup_subsys debug_subsys = {
5391 .create = debug_create,
5392 .destroy = debug_destroy,
5393 .subsys_id = debug_subsys_id,
5394 .base_cftypes = debug_files,
5396 #endif /* CONFIG_CGROUP_DEBUG */