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 */
64 #include <linux/atomic.h>
67 * cgroup_mutex is the master lock. Any modification to cgroup or its
68 * hierarchy must be performed while holding it.
70 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
71 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
72 * release_agent_path and so on. Modifying requires both cgroup_mutex and
73 * cgroup_root_mutex. Readers can acquire either of the two. This is to
74 * break the following locking order cycle.
76 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
77 * B. namespace_sem -> cgroup_mutex
79 * B happens only through cgroup_show_options() and using cgroup_root_mutex
82 static DEFINE_MUTEX(cgroup_mutex);
83 static DEFINE_MUTEX(cgroup_root_mutex);
86 * Generate an array of cgroup subsystem pointers. At boot time, this is
87 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
88 * registered after that. The mutable section of this array is protected by
91 #define SUBSYS(_x) &_x ## _subsys,
92 static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
93 #include <linux/cgroup_subsys.h>
96 #define MAX_CGROUP_ROOT_NAMELEN 64
99 * A cgroupfs_root represents the root of a cgroup hierarchy,
100 * and may be associated with a superblock to form an active
103 struct cgroupfs_root {
104 struct super_block *sb;
107 * The bitmask of subsystems intended to be attached to this
110 unsigned long subsys_bits;
112 /* Unique id for this hierarchy. */
115 /* The bitmask of subsystems currently attached to this hierarchy */
116 unsigned long actual_subsys_bits;
118 /* A list running through the attached subsystems */
119 struct list_head subsys_list;
121 /* The root cgroup for this hierarchy */
122 struct cgroup top_cgroup;
124 /* Tracks how many cgroups are currently defined in hierarchy.*/
125 int number_of_cgroups;
127 /* A list running through the active hierarchies */
128 struct list_head root_list;
130 /* All cgroups on this root, cgroup_mutex protected */
131 struct list_head allcg_list;
133 /* Hierarchy-specific flags */
136 /* The path to use for release notifications. */
137 char release_agent_path[PATH_MAX];
139 /* The name for this hierarchy - may be empty */
140 char name[MAX_CGROUP_ROOT_NAMELEN];
144 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
145 * subsystems that are otherwise unattached - it never has more than a
146 * single cgroup, and all tasks are part of that cgroup.
148 static struct cgroupfs_root rootnode;
151 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
154 struct list_head node;
155 struct dentry *dentry;
160 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
161 * cgroup_subsys->use_id != 0.
163 #define CSS_ID_MAX (65535)
166 * The css to which this ID points. This pointer is set to valid value
167 * after cgroup is populated. If cgroup is removed, this will be NULL.
168 * This pointer is expected to be RCU-safe because destroy()
169 * is called after synchronize_rcu(). But for safe use, css_is_removed()
170 * css_tryget() should be used for avoiding race.
172 struct cgroup_subsys_state __rcu *css;
178 * Depth in hierarchy which this ID belongs to.
180 unsigned short depth;
182 * ID is freed by RCU. (and lookup routine is RCU safe.)
184 struct rcu_head rcu_head;
186 * Hierarchy of CSS ID belongs to.
188 unsigned short stack[0]; /* Array of Length (depth+1) */
192 * cgroup_event represents events which userspace want to receive.
194 struct cgroup_event {
196 * Cgroup which the event belongs to.
200 * Control file which the event associated.
204 * eventfd to signal userspace about the event.
206 struct eventfd_ctx *eventfd;
208 * Each of these stored in a list by the cgroup.
210 struct list_head list;
212 * All fields below needed to unregister event when
213 * userspace closes eventfd.
216 wait_queue_head_t *wqh;
218 struct work_struct remove;
221 /* The list of hierarchy roots */
223 static LIST_HEAD(roots);
224 static int root_count;
226 static DEFINE_IDA(hierarchy_ida);
227 static int next_hierarchy_id;
228 static DEFINE_SPINLOCK(hierarchy_id_lock);
230 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
231 #define dummytop (&rootnode.top_cgroup)
233 /* This flag indicates whether tasks in the fork and exit paths should
234 * check for fork/exit handlers to call. This avoids us having to do
235 * extra work in the fork/exit path if none of the subsystems need to
238 static int need_forkexit_callback __read_mostly;
240 #ifdef CONFIG_PROVE_LOCKING
241 int cgroup_lock_is_held(void)
243 return lockdep_is_held(&cgroup_mutex);
245 #else /* #ifdef CONFIG_PROVE_LOCKING */
246 int cgroup_lock_is_held(void)
248 return mutex_is_locked(&cgroup_mutex);
250 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
252 EXPORT_SYMBOL_GPL(cgroup_lock_is_held);
254 /* convenient tests for these bits */
255 inline int cgroup_is_removed(const struct cgroup *cgrp)
257 return test_bit(CGRP_REMOVED, &cgrp->flags);
260 /* bits in struct cgroupfs_root flags field */
262 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
265 static int cgroup_is_releasable(const struct cgroup *cgrp)
268 (1 << CGRP_RELEASABLE) |
269 (1 << CGRP_NOTIFY_ON_RELEASE);
270 return (cgrp->flags & bits) == bits;
273 static int notify_on_release(const struct cgroup *cgrp)
275 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
278 static int clone_children(const struct cgroup *cgrp)
280 return test_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
284 * for_each_subsys() allows you to iterate on each subsystem attached to
285 * an active hierarchy
287 #define for_each_subsys(_root, _ss) \
288 list_for_each_entry(_ss, &_root->subsys_list, sibling)
290 /* for_each_active_root() allows you to iterate across the active hierarchies */
291 #define for_each_active_root(_root) \
292 list_for_each_entry(_root, &roots, root_list)
294 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
296 return dentry->d_fsdata;
299 static inline struct cfent *__d_cfe(struct dentry *dentry)
301 return dentry->d_fsdata;
304 static inline struct cftype *__d_cft(struct dentry *dentry)
306 return __d_cfe(dentry)->type;
309 /* the list of cgroups eligible for automatic release. Protected by
310 * release_list_lock */
311 static LIST_HEAD(release_list);
312 static DEFINE_RAW_SPINLOCK(release_list_lock);
313 static void cgroup_release_agent(struct work_struct *work);
314 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
315 static void check_for_release(struct cgroup *cgrp);
317 /* Link structure for associating css_set objects with cgroups */
318 struct cg_cgroup_link {
320 * List running through cg_cgroup_links associated with a
321 * cgroup, anchored on cgroup->css_sets
323 struct list_head cgrp_link_list;
326 * List running through cg_cgroup_links pointing at a
327 * single css_set object, anchored on css_set->cg_links
329 struct list_head cg_link_list;
333 /* The default css_set - used by init and its children prior to any
334 * hierarchies being mounted. It contains a pointer to the root state
335 * for each subsystem. Also used to anchor the list of css_sets. Not
336 * reference-counted, to improve performance when child cgroups
337 * haven't been created.
340 static struct css_set init_css_set;
341 static struct cg_cgroup_link init_css_set_link;
343 static int cgroup_init_idr(struct cgroup_subsys *ss,
344 struct cgroup_subsys_state *css);
346 /* css_set_lock protects the list of css_set objects, and the
347 * chain of tasks off each css_set. Nests outside task->alloc_lock
348 * due to cgroup_iter_start() */
349 static DEFINE_RWLOCK(css_set_lock);
350 static int css_set_count;
353 * hash table for cgroup groups. This improves the performance to find
354 * an existing css_set. This hash doesn't (currently) take into
355 * account cgroups in empty hierarchies.
357 #define CSS_SET_HASH_BITS 7
358 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
359 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
361 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
365 unsigned long tmp = 0UL;
367 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
368 tmp += (unsigned long)css[i];
369 tmp = (tmp >> 16) ^ tmp;
371 index = hash_long(tmp, CSS_SET_HASH_BITS);
373 return &css_set_table[index];
376 /* We don't maintain the lists running through each css_set to its
377 * task until after the first call to cgroup_iter_start(). This
378 * reduces the fork()/exit() overhead for people who have cgroups
379 * compiled into their kernel but not actually in use */
380 static int use_task_css_set_links __read_mostly;
382 static void __put_css_set(struct css_set *cg, int taskexit)
384 struct cg_cgroup_link *link;
385 struct cg_cgroup_link *saved_link;
387 * Ensure that the refcount doesn't hit zero while any readers
388 * can see it. Similar to atomic_dec_and_lock(), but for an
391 if (atomic_add_unless(&cg->refcount, -1, 1))
393 write_lock(&css_set_lock);
394 if (!atomic_dec_and_test(&cg->refcount)) {
395 write_unlock(&css_set_lock);
399 /* This css_set is dead. unlink it and release cgroup refcounts */
400 hlist_del(&cg->hlist);
403 list_for_each_entry_safe(link, saved_link, &cg->cg_links,
405 struct cgroup *cgrp = link->cgrp;
406 list_del(&link->cg_link_list);
407 list_del(&link->cgrp_link_list);
408 if (atomic_dec_and_test(&cgrp->count) &&
409 notify_on_release(cgrp)) {
411 set_bit(CGRP_RELEASABLE, &cgrp->flags);
412 check_for_release(cgrp);
418 write_unlock(&css_set_lock);
419 kfree_rcu(cg, rcu_head);
423 * refcounted get/put for css_set objects
425 static inline void get_css_set(struct css_set *cg)
427 atomic_inc(&cg->refcount);
430 static inline void put_css_set(struct css_set *cg)
432 __put_css_set(cg, 0);
435 static inline void put_css_set_taskexit(struct css_set *cg)
437 __put_css_set(cg, 1);
441 * compare_css_sets - helper function for find_existing_css_set().
442 * @cg: candidate css_set being tested
443 * @old_cg: existing css_set for a task
444 * @new_cgrp: cgroup that's being entered by the task
445 * @template: desired set of css pointers in css_set (pre-calculated)
447 * Returns true if "cg" matches "old_cg" except for the hierarchy
448 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
450 static bool compare_css_sets(struct css_set *cg,
451 struct css_set *old_cg,
452 struct cgroup *new_cgrp,
453 struct cgroup_subsys_state *template[])
455 struct list_head *l1, *l2;
457 if (memcmp(template, cg->subsys, sizeof(cg->subsys))) {
458 /* Not all subsystems matched */
463 * Compare cgroup pointers in order to distinguish between
464 * different cgroups in heirarchies with no subsystems. We
465 * could get by with just this check alone (and skip the
466 * memcmp above) but on most setups the memcmp check will
467 * avoid the need for this more expensive check on almost all
472 l2 = &old_cg->cg_links;
474 struct cg_cgroup_link *cgl1, *cgl2;
475 struct cgroup *cg1, *cg2;
479 /* See if we reached the end - both lists are equal length. */
480 if (l1 == &cg->cg_links) {
481 BUG_ON(l2 != &old_cg->cg_links);
484 BUG_ON(l2 == &old_cg->cg_links);
486 /* Locate the cgroups associated with these links. */
487 cgl1 = list_entry(l1, struct cg_cgroup_link, cg_link_list);
488 cgl2 = list_entry(l2, struct cg_cgroup_link, cg_link_list);
491 /* Hierarchies should be linked in the same order. */
492 BUG_ON(cg1->root != cg2->root);
495 * If this hierarchy is the hierarchy of the cgroup
496 * that's changing, then we need to check that this
497 * css_set points to the new cgroup; if it's any other
498 * hierarchy, then this css_set should point to the
499 * same cgroup as the old css_set.
501 if (cg1->root == new_cgrp->root) {
513 * find_existing_css_set() is a helper for
514 * find_css_set(), and checks to see whether an existing
515 * css_set is suitable.
517 * oldcg: the cgroup group that we're using before the cgroup
520 * cgrp: the cgroup that we're moving into
522 * template: location in which to build the desired set of subsystem
523 * state objects for the new cgroup group
525 static struct css_set *find_existing_css_set(
526 struct css_set *oldcg,
528 struct cgroup_subsys_state *template[])
531 struct cgroupfs_root *root = cgrp->root;
532 struct hlist_head *hhead;
533 struct hlist_node *node;
537 * Build the set of subsystem state objects that we want to see in the
538 * new css_set. while subsystems can change globally, the entries here
539 * won't change, so no need for locking.
541 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
542 if (root->subsys_bits & (1UL << i)) {
543 /* Subsystem is in this hierarchy. So we want
544 * the subsystem state from the new
546 template[i] = cgrp->subsys[i];
548 /* Subsystem is not in this hierarchy, so we
549 * don't want to change the subsystem state */
550 template[i] = oldcg->subsys[i];
554 hhead = css_set_hash(template);
555 hlist_for_each_entry(cg, node, hhead, hlist) {
556 if (!compare_css_sets(cg, oldcg, cgrp, template))
559 /* This css_set matches what we need */
563 /* No existing cgroup group matched */
567 static void free_cg_links(struct list_head *tmp)
569 struct cg_cgroup_link *link;
570 struct cg_cgroup_link *saved_link;
572 list_for_each_entry_safe(link, saved_link, tmp, cgrp_link_list) {
573 list_del(&link->cgrp_link_list);
579 * allocate_cg_links() allocates "count" cg_cgroup_link structures
580 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
581 * success or a negative error
583 static int allocate_cg_links(int count, struct list_head *tmp)
585 struct cg_cgroup_link *link;
588 for (i = 0; i < count; i++) {
589 link = kmalloc(sizeof(*link), GFP_KERNEL);
594 list_add(&link->cgrp_link_list, tmp);
600 * link_css_set - a helper function to link a css_set to a cgroup
601 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
602 * @cg: the css_set to be linked
603 * @cgrp: the destination cgroup
605 static void link_css_set(struct list_head *tmp_cg_links,
606 struct css_set *cg, struct cgroup *cgrp)
608 struct cg_cgroup_link *link;
610 BUG_ON(list_empty(tmp_cg_links));
611 link = list_first_entry(tmp_cg_links, struct cg_cgroup_link,
615 atomic_inc(&cgrp->count);
616 list_move(&link->cgrp_link_list, &cgrp->css_sets);
618 * Always add links to the tail of the list so that the list
619 * is sorted by order of hierarchy creation
621 list_add_tail(&link->cg_link_list, &cg->cg_links);
625 * find_css_set() takes an existing cgroup group and a
626 * cgroup object, and returns a css_set object that's
627 * equivalent to the old group, but with the given cgroup
628 * substituted into the appropriate hierarchy. Must be called with
631 static struct css_set *find_css_set(
632 struct css_set *oldcg, struct cgroup *cgrp)
635 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
637 struct list_head tmp_cg_links;
639 struct hlist_head *hhead;
640 struct cg_cgroup_link *link;
642 /* First see if we already have a cgroup group that matches
644 read_lock(&css_set_lock);
645 res = find_existing_css_set(oldcg, cgrp, template);
648 read_unlock(&css_set_lock);
653 res = kmalloc(sizeof(*res), GFP_KERNEL);
657 /* Allocate all the cg_cgroup_link objects that we'll need */
658 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
663 atomic_set(&res->refcount, 1);
664 INIT_LIST_HEAD(&res->cg_links);
665 INIT_LIST_HEAD(&res->tasks);
666 INIT_HLIST_NODE(&res->hlist);
668 /* Copy the set of subsystem state objects generated in
669 * find_existing_css_set() */
670 memcpy(res->subsys, template, sizeof(res->subsys));
672 write_lock(&css_set_lock);
673 /* Add reference counts and links from the new css_set. */
674 list_for_each_entry(link, &oldcg->cg_links, cg_link_list) {
675 struct cgroup *c = link->cgrp;
676 if (c->root == cgrp->root)
678 link_css_set(&tmp_cg_links, res, c);
681 BUG_ON(!list_empty(&tmp_cg_links));
685 /* Add this cgroup group to the hash table */
686 hhead = css_set_hash(res->subsys);
687 hlist_add_head(&res->hlist, hhead);
689 write_unlock(&css_set_lock);
695 * Return the cgroup for "task" from the given hierarchy. Must be
696 * called with cgroup_mutex held.
698 static struct cgroup *task_cgroup_from_root(struct task_struct *task,
699 struct cgroupfs_root *root)
702 struct cgroup *res = NULL;
704 BUG_ON(!mutex_is_locked(&cgroup_mutex));
705 read_lock(&css_set_lock);
707 * No need to lock the task - since we hold cgroup_mutex the
708 * task can't change groups, so the only thing that can happen
709 * is that it exits and its css is set back to init_css_set.
712 if (css == &init_css_set) {
713 res = &root->top_cgroup;
715 struct cg_cgroup_link *link;
716 list_for_each_entry(link, &css->cg_links, cg_link_list) {
717 struct cgroup *c = link->cgrp;
718 if (c->root == root) {
724 read_unlock(&css_set_lock);
730 * There is one global cgroup mutex. We also require taking
731 * task_lock() when dereferencing a task's cgroup subsys pointers.
732 * See "The task_lock() exception", at the end of this comment.
734 * A task must hold cgroup_mutex to modify cgroups.
736 * Any task can increment and decrement the count field without lock.
737 * So in general, code holding cgroup_mutex can't rely on the count
738 * field not changing. However, if the count goes to zero, then only
739 * cgroup_attach_task() can increment it again. Because a count of zero
740 * means that no tasks are currently attached, therefore there is no
741 * way a task attached to that cgroup can fork (the other way to
742 * increment the count). So code holding cgroup_mutex can safely
743 * assume that if the count is zero, it will stay zero. Similarly, if
744 * a task holds cgroup_mutex on a cgroup with zero count, it
745 * knows that the cgroup won't be removed, as cgroup_rmdir()
748 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
749 * (usually) take cgroup_mutex. These are the two most performance
750 * critical pieces of code here. The exception occurs on cgroup_exit(),
751 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
752 * is taken, and if the cgroup count is zero, a usermode call made
753 * to the release agent with the name of the cgroup (path relative to
754 * the root of cgroup file system) as the argument.
756 * A cgroup can only be deleted if both its 'count' of using tasks
757 * is zero, and its list of 'children' cgroups is empty. Since all
758 * tasks in the system use _some_ cgroup, and since there is always at
759 * least one task in the system (init, pid == 1), therefore, top_cgroup
760 * always has either children cgroups and/or using tasks. So we don't
761 * need a special hack to ensure that top_cgroup cannot be deleted.
763 * The task_lock() exception
765 * The need for this exception arises from the action of
766 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
767 * another. It does so using cgroup_mutex, however there are
768 * several performance critical places that need to reference
769 * task->cgroup without the expense of grabbing a system global
770 * mutex. Therefore except as noted below, when dereferencing or, as
771 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
772 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
773 * the task_struct routinely used for such matters.
775 * P.S. One more locking exception. RCU is used to guard the
776 * update of a tasks cgroup pointer by cgroup_attach_task()
780 * cgroup_lock - lock out any changes to cgroup structures
783 void cgroup_lock(void)
785 mutex_lock(&cgroup_mutex);
787 EXPORT_SYMBOL_GPL(cgroup_lock);
790 * cgroup_unlock - release lock on cgroup changes
792 * Undo the lock taken in a previous cgroup_lock() call.
794 void cgroup_unlock(void)
796 mutex_unlock(&cgroup_mutex);
798 EXPORT_SYMBOL_GPL(cgroup_unlock);
801 * A couple of forward declarations required, due to cyclic reference loop:
802 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
803 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
807 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
808 static struct dentry *cgroup_lookup(struct inode *, struct dentry *, struct nameidata *);
809 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
810 static int cgroup_populate_dir(struct cgroup *cgrp);
811 static const struct inode_operations cgroup_dir_inode_operations;
812 static const struct file_operations proc_cgroupstats_operations;
814 static struct backing_dev_info cgroup_backing_dev_info = {
816 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
819 static int alloc_css_id(struct cgroup_subsys *ss,
820 struct cgroup *parent, struct cgroup *child);
822 static struct inode *cgroup_new_inode(umode_t mode, struct super_block *sb)
824 struct inode *inode = new_inode(sb);
827 inode->i_ino = get_next_ino();
828 inode->i_mode = mode;
829 inode->i_uid = current_fsuid();
830 inode->i_gid = current_fsgid();
831 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
832 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
838 * Call subsys's pre_destroy handler.
839 * This is called before css refcnt check.
841 static int cgroup_call_pre_destroy(struct cgroup *cgrp)
843 struct cgroup_subsys *ss;
846 for_each_subsys(cgrp->root, ss)
847 if (ss->pre_destroy) {
848 ret = ss->pre_destroy(cgrp);
856 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
858 /* is dentry a directory ? if so, kfree() associated cgroup */
859 if (S_ISDIR(inode->i_mode)) {
860 struct cgroup *cgrp = dentry->d_fsdata;
861 struct cgroup_subsys *ss;
862 BUG_ON(!(cgroup_is_removed(cgrp)));
863 /* It's possible for external users to be holding css
864 * reference counts on a cgroup; css_put() needs to
865 * be able to access the cgroup after decrementing
866 * the reference count in order to know if it needs to
867 * queue the cgroup to be handled by the release
871 mutex_lock(&cgroup_mutex);
873 * Release the subsystem state objects.
875 for_each_subsys(cgrp->root, ss)
878 cgrp->root->number_of_cgroups--;
879 mutex_unlock(&cgroup_mutex);
882 * Drop the active superblock reference that we took when we
885 deactivate_super(cgrp->root->sb);
888 * if we're getting rid of the cgroup, refcount should ensure
889 * that there are no pidlists left.
891 BUG_ON(!list_empty(&cgrp->pidlists));
893 kfree_rcu(cgrp, rcu_head);
895 struct cfent *cfe = __d_cfe(dentry);
896 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
898 WARN_ONCE(!list_empty(&cfe->node) &&
899 cgrp != &cgrp->root->top_cgroup,
900 "cfe still linked for %s\n", cfe->type->name);
906 static int cgroup_delete(const struct dentry *d)
911 static void remove_dir(struct dentry *d)
913 struct dentry *parent = dget(d->d_parent);
916 simple_rmdir(parent->d_inode, d);
920 static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
924 lockdep_assert_held(&cgrp->dentry->d_inode->i_mutex);
925 lockdep_assert_held(&cgroup_mutex);
927 list_for_each_entry(cfe, &cgrp->files, node) {
928 struct dentry *d = cfe->dentry;
930 if (cft && cfe->type != cft)
935 simple_unlink(d->d_inode, d);
936 list_del_init(&cfe->node);
944 static void cgroup_clear_directory(struct dentry *dir)
946 struct cgroup *cgrp = __d_cgrp(dir);
948 while (!list_empty(&cgrp->files))
949 cgroup_rm_file(cgrp, NULL);
953 * NOTE : the dentry must have been dget()'ed
955 static void cgroup_d_remove_dir(struct dentry *dentry)
957 struct dentry *parent;
959 cgroup_clear_directory(dentry);
961 parent = dentry->d_parent;
962 spin_lock(&parent->d_lock);
963 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
964 list_del_init(&dentry->d_u.d_child);
965 spin_unlock(&dentry->d_lock);
966 spin_unlock(&parent->d_lock);
971 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
972 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
973 * reference to css->refcnt. In general, this refcnt is expected to goes down
976 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
978 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
980 static void cgroup_wakeup_rmdir_waiter(struct cgroup *cgrp)
982 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
983 wake_up_all(&cgroup_rmdir_waitq);
986 void cgroup_exclude_rmdir(struct cgroup_subsys_state *css)
991 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
993 cgroup_wakeup_rmdir_waiter(css->cgroup);
998 * Call with cgroup_mutex held. Drops reference counts on modules, including
999 * any duplicate ones that parse_cgroupfs_options took. If this function
1000 * returns an error, no reference counts are touched.
1002 static int rebind_subsystems(struct cgroupfs_root *root,
1003 unsigned long final_bits)
1005 unsigned long added_bits, removed_bits;
1006 struct cgroup *cgrp = &root->top_cgroup;
1009 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1010 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1012 removed_bits = root->actual_subsys_bits & ~final_bits;
1013 added_bits = final_bits & ~root->actual_subsys_bits;
1014 /* Check that any added subsystems are currently free */
1015 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1016 unsigned long bit = 1UL << i;
1017 struct cgroup_subsys *ss = subsys[i];
1018 if (!(bit & added_bits))
1021 * Nobody should tell us to do a subsys that doesn't exist:
1022 * parse_cgroupfs_options should catch that case and refcounts
1023 * ensure that subsystems won't disappear once selected.
1026 if (ss->root != &rootnode) {
1027 /* Subsystem isn't free */
1032 /* Currently we don't handle adding/removing subsystems when
1033 * any child cgroups exist. This is theoretically supportable
1034 * but involves complex error handling, so it's being left until
1036 if (root->number_of_cgroups > 1)
1039 /* Process each subsystem */
1040 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1041 struct cgroup_subsys *ss = subsys[i];
1042 unsigned long bit = 1UL << i;
1043 if (bit & added_bits) {
1044 /* We're binding this subsystem to this hierarchy */
1046 BUG_ON(cgrp->subsys[i]);
1047 BUG_ON(!dummytop->subsys[i]);
1048 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
1049 mutex_lock(&ss->hierarchy_mutex);
1050 cgrp->subsys[i] = dummytop->subsys[i];
1051 cgrp->subsys[i]->cgroup = cgrp;
1052 list_move(&ss->sibling, &root->subsys_list);
1056 mutex_unlock(&ss->hierarchy_mutex);
1057 /* refcount was already taken, and we're keeping it */
1058 } else if (bit & removed_bits) {
1059 /* We're removing this subsystem */
1061 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
1062 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
1063 mutex_lock(&ss->hierarchy_mutex);
1066 dummytop->subsys[i]->cgroup = dummytop;
1067 cgrp->subsys[i] = NULL;
1068 subsys[i]->root = &rootnode;
1069 list_move(&ss->sibling, &rootnode.subsys_list);
1070 mutex_unlock(&ss->hierarchy_mutex);
1071 /* subsystem is now free - drop reference on module */
1072 module_put(ss->module);
1073 } else if (bit & final_bits) {
1074 /* Subsystem state should already exist */
1076 BUG_ON(!cgrp->subsys[i]);
1078 * a refcount was taken, but we already had one, so
1079 * drop the extra reference.
1081 module_put(ss->module);
1082 #ifdef CONFIG_MODULE_UNLOAD
1083 BUG_ON(ss->module && !module_refcount(ss->module));
1086 /* Subsystem state shouldn't exist */
1087 BUG_ON(cgrp->subsys[i]);
1090 root->subsys_bits = root->actual_subsys_bits = final_bits;
1096 static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1098 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
1099 struct cgroup_subsys *ss;
1101 mutex_lock(&cgroup_root_mutex);
1102 for_each_subsys(root, ss)
1103 seq_printf(seq, ",%s", ss->name);
1104 if (test_bit(ROOT_NOPREFIX, &root->flags))
1105 seq_puts(seq, ",noprefix");
1106 if (strlen(root->release_agent_path))
1107 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1108 if (clone_children(&root->top_cgroup))
1109 seq_puts(seq, ",clone_children");
1110 if (strlen(root->name))
1111 seq_printf(seq, ",name=%s", root->name);
1112 mutex_unlock(&cgroup_root_mutex);
1116 struct cgroup_sb_opts {
1117 unsigned long subsys_bits;
1118 unsigned long flags;
1119 char *release_agent;
1120 bool clone_children;
1122 /* User explicitly requested empty subsystem */
1125 struct cgroupfs_root *new_root;
1130 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1131 * with cgroup_mutex held to protect the subsys[] array. This function takes
1132 * refcounts on subsystems to be used, unless it returns error, in which case
1133 * no refcounts are taken.
1135 static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1137 char *token, *o = data;
1138 bool all_ss = false, one_ss = false;
1139 unsigned long mask = (unsigned long)-1;
1141 bool module_pin_failed = false;
1143 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1145 #ifdef CONFIG_CPUSETS
1146 mask = ~(1UL << cpuset_subsys_id);
1149 memset(opts, 0, sizeof(*opts));
1151 while ((token = strsep(&o, ",")) != NULL) {
1154 if (!strcmp(token, "none")) {
1155 /* Explicitly have no subsystems */
1159 if (!strcmp(token, "all")) {
1160 /* Mutually exclusive option 'all' + subsystem name */
1166 if (!strcmp(token, "noprefix")) {
1167 set_bit(ROOT_NOPREFIX, &opts->flags);
1170 if (!strcmp(token, "clone_children")) {
1171 opts->clone_children = true;
1174 if (!strncmp(token, "release_agent=", 14)) {
1175 /* Specifying two release agents is forbidden */
1176 if (opts->release_agent)
1178 opts->release_agent =
1179 kstrndup(token + 14, PATH_MAX - 1, GFP_KERNEL);
1180 if (!opts->release_agent)
1184 if (!strncmp(token, "name=", 5)) {
1185 const char *name = token + 5;
1186 /* Can't specify an empty name */
1189 /* Must match [\w.-]+ */
1190 for (i = 0; i < strlen(name); i++) {
1194 if ((c == '.') || (c == '-') || (c == '_'))
1198 /* Specifying two names is forbidden */
1201 opts->name = kstrndup(name,
1202 MAX_CGROUP_ROOT_NAMELEN - 1,
1210 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1211 struct cgroup_subsys *ss = subsys[i];
1214 if (strcmp(token, ss->name))
1219 /* Mutually exclusive option 'all' + subsystem name */
1222 set_bit(i, &opts->subsys_bits);
1227 if (i == CGROUP_SUBSYS_COUNT)
1232 * If the 'all' option was specified select all the subsystems,
1233 * otherwise if 'none', 'name=' and a subsystem name options
1234 * were not specified, let's default to 'all'
1236 if (all_ss || (!one_ss && !opts->none && !opts->name)) {
1237 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1238 struct cgroup_subsys *ss = subsys[i];
1243 set_bit(i, &opts->subsys_bits);
1247 /* Consistency checks */
1250 * Option noprefix was introduced just for backward compatibility
1251 * with the old cpuset, so we allow noprefix only if mounting just
1252 * the cpuset subsystem.
1254 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1255 (opts->subsys_bits & mask))
1259 /* Can't specify "none" and some subsystems */
1260 if (opts->subsys_bits && opts->none)
1264 * We either have to specify by name or by subsystems. (So all
1265 * empty hierarchies must have a name).
1267 if (!opts->subsys_bits && !opts->name)
1271 * Grab references on all the modules we'll need, so the subsystems
1272 * don't dance around before rebind_subsystems attaches them. This may
1273 * take duplicate reference counts on a subsystem that's already used,
1274 * but rebind_subsystems handles this case.
1276 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1277 unsigned long bit = 1UL << i;
1279 if (!(bit & opts->subsys_bits))
1281 if (!try_module_get(subsys[i]->module)) {
1282 module_pin_failed = true;
1286 if (module_pin_failed) {
1288 * oops, one of the modules was going away. this means that we
1289 * raced with a module_delete call, and to the user this is
1290 * essentially a "subsystem doesn't exist" case.
1292 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) {
1293 /* drop refcounts only on the ones we took */
1294 unsigned long bit = 1UL << i;
1296 if (!(bit & opts->subsys_bits))
1298 module_put(subsys[i]->module);
1306 static void drop_parsed_module_refcounts(unsigned long subsys_bits)
1309 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
1310 unsigned long bit = 1UL << i;
1312 if (!(bit & subsys_bits))
1314 module_put(subsys[i]->module);
1318 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1321 struct cgroupfs_root *root = sb->s_fs_info;
1322 struct cgroup *cgrp = &root->top_cgroup;
1323 struct cgroup_sb_opts opts;
1325 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1326 mutex_lock(&cgroup_mutex);
1327 mutex_lock(&cgroup_root_mutex);
1329 /* See what subsystems are wanted */
1330 ret = parse_cgroupfs_options(data, &opts);
1334 /* See feature-removal-schedule.txt */
1335 if (opts.subsys_bits != root->actual_subsys_bits || opts.release_agent)
1336 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1337 task_tgid_nr(current), current->comm);
1339 /* Don't allow flags or name to change at remount */
1340 if (opts.flags != root->flags ||
1341 (opts.name && strcmp(opts.name, root->name))) {
1343 drop_parsed_module_refcounts(opts.subsys_bits);
1347 ret = rebind_subsystems(root, opts.subsys_bits);
1349 drop_parsed_module_refcounts(opts.subsys_bits);
1353 /* clear out any existing files and repopulate subsystem files */
1354 cgroup_clear_directory(cgrp->dentry);
1355 cgroup_populate_dir(cgrp);
1357 if (opts.release_agent)
1358 strcpy(root->release_agent_path, opts.release_agent);
1360 kfree(opts.release_agent);
1362 mutex_unlock(&cgroup_root_mutex);
1363 mutex_unlock(&cgroup_mutex);
1364 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1368 static const struct super_operations cgroup_ops = {
1369 .statfs = simple_statfs,
1370 .drop_inode = generic_delete_inode,
1371 .show_options = cgroup_show_options,
1372 .remount_fs = cgroup_remount,
1375 static void init_cgroup_housekeeping(struct cgroup *cgrp)
1377 INIT_LIST_HEAD(&cgrp->sibling);
1378 INIT_LIST_HEAD(&cgrp->children);
1379 INIT_LIST_HEAD(&cgrp->files);
1380 INIT_LIST_HEAD(&cgrp->css_sets);
1381 INIT_LIST_HEAD(&cgrp->release_list);
1382 INIT_LIST_HEAD(&cgrp->pidlists);
1383 mutex_init(&cgrp->pidlist_mutex);
1384 INIT_LIST_HEAD(&cgrp->event_list);
1385 spin_lock_init(&cgrp->event_list_lock);
1388 static void init_cgroup_root(struct cgroupfs_root *root)
1390 struct cgroup *cgrp = &root->top_cgroup;
1392 INIT_LIST_HEAD(&root->subsys_list);
1393 INIT_LIST_HEAD(&root->root_list);
1394 INIT_LIST_HEAD(&root->allcg_list);
1395 root->number_of_cgroups = 1;
1397 cgrp->top_cgroup = cgrp;
1398 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
1399 init_cgroup_housekeeping(cgrp);
1402 static bool init_root_id(struct cgroupfs_root *root)
1407 if (!ida_pre_get(&hierarchy_ida, GFP_KERNEL))
1409 spin_lock(&hierarchy_id_lock);
1410 /* Try to allocate the next unused ID */
1411 ret = ida_get_new_above(&hierarchy_ida, next_hierarchy_id,
1412 &root->hierarchy_id);
1414 /* Try again starting from 0 */
1415 ret = ida_get_new(&hierarchy_ida, &root->hierarchy_id);
1417 next_hierarchy_id = root->hierarchy_id + 1;
1418 } else if (ret != -EAGAIN) {
1419 /* Can only get here if the 31-bit IDR is full ... */
1422 spin_unlock(&hierarchy_id_lock);
1427 static int cgroup_test_super(struct super_block *sb, void *data)
1429 struct cgroup_sb_opts *opts = data;
1430 struct cgroupfs_root *root = sb->s_fs_info;
1432 /* If we asked for a name then it must match */
1433 if (opts->name && strcmp(opts->name, root->name))
1437 * If we asked for subsystems (or explicitly for no
1438 * subsystems) then they must match
1440 if ((opts->subsys_bits || opts->none)
1441 && (opts->subsys_bits != root->subsys_bits))
1447 static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1449 struct cgroupfs_root *root;
1451 if (!opts->subsys_bits && !opts->none)
1454 root = kzalloc(sizeof(*root), GFP_KERNEL);
1456 return ERR_PTR(-ENOMEM);
1458 if (!init_root_id(root)) {
1460 return ERR_PTR(-ENOMEM);
1462 init_cgroup_root(root);
1464 root->subsys_bits = opts->subsys_bits;
1465 root->flags = opts->flags;
1466 if (opts->release_agent)
1467 strcpy(root->release_agent_path, opts->release_agent);
1469 strcpy(root->name, opts->name);
1470 if (opts->clone_children)
1471 set_bit(CGRP_CLONE_CHILDREN, &root->top_cgroup.flags);
1475 static void cgroup_drop_root(struct cgroupfs_root *root)
1480 BUG_ON(!root->hierarchy_id);
1481 spin_lock(&hierarchy_id_lock);
1482 ida_remove(&hierarchy_ida, root->hierarchy_id);
1483 spin_unlock(&hierarchy_id_lock);
1487 static int cgroup_set_super(struct super_block *sb, void *data)
1490 struct cgroup_sb_opts *opts = data;
1492 /* If we don't have a new root, we can't set up a new sb */
1493 if (!opts->new_root)
1496 BUG_ON(!opts->subsys_bits && !opts->none);
1498 ret = set_anon_super(sb, NULL);
1502 sb->s_fs_info = opts->new_root;
1503 opts->new_root->sb = sb;
1505 sb->s_blocksize = PAGE_CACHE_SIZE;
1506 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1507 sb->s_magic = CGROUP_SUPER_MAGIC;
1508 sb->s_op = &cgroup_ops;
1513 static int cgroup_get_rootdir(struct super_block *sb)
1515 static const struct dentry_operations cgroup_dops = {
1516 .d_iput = cgroup_diput,
1517 .d_delete = cgroup_delete,
1520 struct inode *inode =
1521 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
1526 inode->i_fop = &simple_dir_operations;
1527 inode->i_op = &cgroup_dir_inode_operations;
1528 /* directories start off with i_nlink == 2 (for "." entry) */
1530 sb->s_root = d_make_root(inode);
1533 /* for everything else we want ->d_op set */
1534 sb->s_d_op = &cgroup_dops;
1538 static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1539 int flags, const char *unused_dev_name,
1542 struct cgroup_sb_opts opts;
1543 struct cgroupfs_root *root;
1545 struct super_block *sb;
1546 struct cgroupfs_root *new_root;
1547 struct inode *inode;
1549 /* First find the desired set of subsystems */
1550 mutex_lock(&cgroup_mutex);
1551 ret = parse_cgroupfs_options(data, &opts);
1552 mutex_unlock(&cgroup_mutex);
1557 * Allocate a new cgroup root. We may not need it if we're
1558 * reusing an existing hierarchy.
1560 new_root = cgroup_root_from_opts(&opts);
1561 if (IS_ERR(new_root)) {
1562 ret = PTR_ERR(new_root);
1565 opts.new_root = new_root;
1567 /* Locate an existing or new sb for this hierarchy */
1568 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, &opts);
1571 cgroup_drop_root(opts.new_root);
1575 root = sb->s_fs_info;
1577 if (root == opts.new_root) {
1578 /* We used the new root structure, so this is a new hierarchy */
1579 struct list_head tmp_cg_links;
1580 struct cgroup *root_cgrp = &root->top_cgroup;
1581 struct cgroupfs_root *existing_root;
1582 const struct cred *cred;
1585 BUG_ON(sb->s_root != NULL);
1587 ret = cgroup_get_rootdir(sb);
1589 goto drop_new_super;
1590 inode = sb->s_root->d_inode;
1592 mutex_lock(&inode->i_mutex);
1593 mutex_lock(&cgroup_mutex);
1594 mutex_lock(&cgroup_root_mutex);
1596 /* Check for name clashes with existing mounts */
1598 if (strlen(root->name))
1599 for_each_active_root(existing_root)
1600 if (!strcmp(existing_root->name, root->name))
1604 * We're accessing css_set_count without locking
1605 * css_set_lock here, but that's OK - it can only be
1606 * increased by someone holding cgroup_lock, and
1607 * that's us. The worst that can happen is that we
1608 * have some link structures left over
1610 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1614 ret = rebind_subsystems(root, root->subsys_bits);
1615 if (ret == -EBUSY) {
1616 free_cg_links(&tmp_cg_links);
1620 * There must be no failure case after here, since rebinding
1621 * takes care of subsystems' refcounts, which are explicitly
1622 * dropped in the failure exit path.
1625 /* EBUSY should be the only error here */
1628 list_add(&root->root_list, &roots);
1631 sb->s_root->d_fsdata = root_cgrp;
1632 root->top_cgroup.dentry = sb->s_root;
1634 /* Link the top cgroup in this hierarchy into all
1635 * the css_set objects */
1636 write_lock(&css_set_lock);
1637 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1638 struct hlist_head *hhead = &css_set_table[i];
1639 struct hlist_node *node;
1642 hlist_for_each_entry(cg, node, hhead, hlist)
1643 link_css_set(&tmp_cg_links, cg, root_cgrp);
1645 write_unlock(&css_set_lock);
1647 free_cg_links(&tmp_cg_links);
1649 BUG_ON(!list_empty(&root_cgrp->sibling));
1650 BUG_ON(!list_empty(&root_cgrp->children));
1651 BUG_ON(root->number_of_cgroups != 1);
1653 cred = override_creds(&init_cred);
1654 cgroup_populate_dir(root_cgrp);
1656 mutex_unlock(&cgroup_root_mutex);
1657 mutex_unlock(&cgroup_mutex);
1658 mutex_unlock(&inode->i_mutex);
1661 * We re-used an existing hierarchy - the new root (if
1662 * any) is not needed
1664 cgroup_drop_root(opts.new_root);
1665 /* no subsys rebinding, so refcounts don't change */
1666 drop_parsed_module_refcounts(opts.subsys_bits);
1669 kfree(opts.release_agent);
1671 return dget(sb->s_root);
1674 mutex_unlock(&cgroup_root_mutex);
1675 mutex_unlock(&cgroup_mutex);
1676 mutex_unlock(&inode->i_mutex);
1678 deactivate_locked_super(sb);
1680 drop_parsed_module_refcounts(opts.subsys_bits);
1682 kfree(opts.release_agent);
1684 return ERR_PTR(ret);
1687 static void cgroup_kill_sb(struct super_block *sb) {
1688 struct cgroupfs_root *root = sb->s_fs_info;
1689 struct cgroup *cgrp = &root->top_cgroup;
1691 struct cg_cgroup_link *link;
1692 struct cg_cgroup_link *saved_link;
1696 BUG_ON(root->number_of_cgroups != 1);
1697 BUG_ON(!list_empty(&cgrp->children));
1698 BUG_ON(!list_empty(&cgrp->sibling));
1700 mutex_lock(&cgroup_mutex);
1701 mutex_lock(&cgroup_root_mutex);
1703 /* Rebind all subsystems back to the default hierarchy */
1704 ret = rebind_subsystems(root, 0);
1705 /* Shouldn't be able to fail ... */
1709 * Release all the links from css_sets to this hierarchy's
1712 write_lock(&css_set_lock);
1714 list_for_each_entry_safe(link, saved_link, &cgrp->css_sets,
1716 list_del(&link->cg_link_list);
1717 list_del(&link->cgrp_link_list);
1720 write_unlock(&css_set_lock);
1722 if (!list_empty(&root->root_list)) {
1723 list_del(&root->root_list);
1727 mutex_unlock(&cgroup_root_mutex);
1728 mutex_unlock(&cgroup_mutex);
1730 kill_litter_super(sb);
1731 cgroup_drop_root(root);
1734 static struct file_system_type cgroup_fs_type = {
1736 .mount = cgroup_mount,
1737 .kill_sb = cgroup_kill_sb,
1740 static struct kobject *cgroup_kobj;
1743 * cgroup_path - generate the path of a cgroup
1744 * @cgrp: the cgroup in question
1745 * @buf: the buffer to write the path into
1746 * @buflen: the length of the buffer
1748 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1749 * reference. Writes path of cgroup into buf. Returns 0 on success,
1752 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1755 struct dentry *dentry = rcu_dereference_check(cgrp->dentry,
1756 cgroup_lock_is_held());
1758 if (!dentry || cgrp == dummytop) {
1760 * Inactive subsystems have no dentry for their root
1767 start = buf + buflen;
1771 int len = dentry->d_name.len;
1773 if ((start -= len) < buf)
1774 return -ENAMETOOLONG;
1775 memcpy(start, dentry->d_name.name, len);
1776 cgrp = cgrp->parent;
1780 dentry = rcu_dereference_check(cgrp->dentry,
1781 cgroup_lock_is_held());
1785 return -ENAMETOOLONG;
1788 memmove(buf, start, buf + buflen - start);
1791 EXPORT_SYMBOL_GPL(cgroup_path);
1794 * Control Group taskset
1796 struct task_and_cgroup {
1797 struct task_struct *task;
1798 struct cgroup *cgrp;
1802 struct cgroup_taskset {
1803 struct task_and_cgroup single;
1804 struct flex_array *tc_array;
1807 struct cgroup *cur_cgrp;
1811 * cgroup_taskset_first - reset taskset and return the first task
1812 * @tset: taskset of interest
1814 * @tset iteration is initialized and the first task is returned.
1816 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset)
1818 if (tset->tc_array) {
1820 return cgroup_taskset_next(tset);
1822 tset->cur_cgrp = tset->single.cgrp;
1823 return tset->single.task;
1826 EXPORT_SYMBOL_GPL(cgroup_taskset_first);
1829 * cgroup_taskset_next - iterate to the next task in taskset
1830 * @tset: taskset of interest
1832 * Return the next task in @tset. Iteration must have been initialized
1833 * with cgroup_taskset_first().
1835 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
1837 struct task_and_cgroup *tc;
1839 if (!tset->tc_array || tset->idx >= tset->tc_array_len)
1842 tc = flex_array_get(tset->tc_array, tset->idx++);
1843 tset->cur_cgrp = tc->cgrp;
1846 EXPORT_SYMBOL_GPL(cgroup_taskset_next);
1849 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1850 * @tset: taskset of interest
1852 * Return the cgroup for the current (last returned) task of @tset. This
1853 * function must be preceded by either cgroup_taskset_first() or
1854 * cgroup_taskset_next().
1856 struct cgroup *cgroup_taskset_cur_cgroup(struct cgroup_taskset *tset)
1858 return tset->cur_cgrp;
1860 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup);
1863 * cgroup_taskset_size - return the number of tasks in taskset
1864 * @tset: taskset of interest
1866 int cgroup_taskset_size(struct cgroup_taskset *tset)
1868 return tset->tc_array ? tset->tc_array_len : 1;
1870 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
1874 * cgroup_task_migrate - move a task from one cgroup to another.
1876 * 'guarantee' is set if the caller promises that a new css_set for the task
1877 * will already exist. If not set, this function might sleep, and can fail with
1878 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1880 static void cgroup_task_migrate(struct cgroup *cgrp, struct cgroup *oldcgrp,
1881 struct task_struct *tsk, struct css_set *newcg)
1883 struct css_set *oldcg;
1886 * We are synchronized through threadgroup_lock() against PF_EXITING
1887 * setting such that we can't race against cgroup_exit() changing the
1888 * css_set to init_css_set and dropping the old one.
1890 WARN_ON_ONCE(tsk->flags & PF_EXITING);
1891 oldcg = tsk->cgroups;
1894 rcu_assign_pointer(tsk->cgroups, newcg);
1897 /* Update the css_set linked lists if we're using them */
1898 write_lock(&css_set_lock);
1899 if (!list_empty(&tsk->cg_list))
1900 list_move(&tsk->cg_list, &newcg->tasks);
1901 write_unlock(&css_set_lock);
1904 * We just gained a reference on oldcg by taking it from the task. As
1905 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1906 * it here; it will be freed under RCU.
1910 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1914 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1915 * @cgrp: the cgroup the task is attaching to
1916 * @tsk: the task to be attached
1918 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1921 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1924 struct cgroup_subsys *ss, *failed_ss = NULL;
1925 struct cgroup *oldcgrp;
1926 struct cgroupfs_root *root = cgrp->root;
1927 struct cgroup_taskset tset = { };
1928 struct css_set *newcg;
1930 /* @tsk either already exited or can't exit until the end */
1931 if (tsk->flags & PF_EXITING)
1934 /* Nothing to do if the task is already in that cgroup */
1935 oldcgrp = task_cgroup_from_root(tsk, root);
1936 if (cgrp == oldcgrp)
1939 tset.single.task = tsk;
1940 tset.single.cgrp = oldcgrp;
1942 for_each_subsys(root, ss) {
1943 if (ss->can_attach) {
1944 retval = ss->can_attach(cgrp, &tset);
1947 * Remember on which subsystem the can_attach()
1948 * failed, so that we only call cancel_attach()
1949 * against the subsystems whose can_attach()
1950 * succeeded. (See below)
1958 newcg = find_css_set(tsk->cgroups, cgrp);
1964 cgroup_task_migrate(cgrp, oldcgrp, tsk, newcg);
1966 for_each_subsys(root, ss) {
1968 ss->attach(cgrp, &tset);
1974 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1975 * is no longer empty.
1977 cgroup_wakeup_rmdir_waiter(cgrp);
1980 for_each_subsys(root, ss) {
1981 if (ss == failed_ss)
1983 * This subsystem was the one that failed the
1984 * can_attach() check earlier, so we don't need
1985 * to call cancel_attach() against it or any
1986 * remaining subsystems.
1989 if (ss->cancel_attach)
1990 ss->cancel_attach(cgrp, &tset);
1997 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1998 * @from: attach to all cgroups of a given task
1999 * @tsk: the task to be attached
2001 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *tsk)
2003 struct cgroupfs_root *root;
2007 for_each_active_root(root) {
2008 struct cgroup *from_cg = task_cgroup_from_root(from, root);
2010 retval = cgroup_attach_task(from_cg, tsk);
2018 EXPORT_SYMBOL_GPL(cgroup_attach_task_all);
2021 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2022 * @cgrp: the cgroup to attach to
2023 * @leader: the threadgroup leader task_struct of the group to be attached
2025 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2026 * task_lock of each thread in leader's threadgroup individually in turn.
2028 static int cgroup_attach_proc(struct cgroup *cgrp, struct task_struct *leader)
2030 int retval, i, group_size;
2031 struct cgroup_subsys *ss, *failed_ss = NULL;
2032 /* guaranteed to be initialized later, but the compiler needs this */
2033 struct cgroupfs_root *root = cgrp->root;
2034 /* threadgroup list cursor and array */
2035 struct task_struct *tsk;
2036 struct task_and_cgroup *tc;
2037 struct flex_array *group;
2038 struct cgroup_taskset tset = { };
2041 * step 0: in order to do expensive, possibly blocking operations for
2042 * every thread, we cannot iterate the thread group list, since it needs
2043 * rcu or tasklist locked. instead, build an array of all threads in the
2044 * group - group_rwsem prevents new threads from appearing, and if
2045 * threads exit, this will just be an over-estimate.
2047 group_size = get_nr_threads(leader);
2048 /* flex_array supports very large thread-groups better than kmalloc. */
2049 group = flex_array_alloc(sizeof(*tc), group_size, GFP_KERNEL);
2052 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2053 retval = flex_array_prealloc(group, 0, group_size - 1, GFP_KERNEL);
2055 goto out_free_group_list;
2060 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2061 * already PF_EXITING could be freed from underneath us unless we
2062 * take an rcu_read_lock.
2066 struct task_and_cgroup ent;
2068 /* @tsk either already exited or can't exit until the end */
2069 if (tsk->flags & PF_EXITING)
2072 /* as per above, nr_threads may decrease, but not increase. */
2073 BUG_ON(i >= group_size);
2075 ent.cgrp = task_cgroup_from_root(tsk, root);
2076 /* nothing to do if this task is already in the cgroup */
2077 if (ent.cgrp == cgrp)
2080 * saying GFP_ATOMIC has no effect here because we did prealloc
2081 * earlier, but it's good form to communicate our expectations.
2083 retval = flex_array_put(group, i, &ent, GFP_ATOMIC);
2084 BUG_ON(retval != 0);
2086 } while_each_thread(leader, tsk);
2088 /* remember the number of threads in the array for later. */
2090 tset.tc_array = group;
2091 tset.tc_array_len = group_size;
2093 /* methods shouldn't be called if no task is actually migrating */
2096 goto out_free_group_list;
2099 * step 1: check that we can legitimately attach to the cgroup.
2101 for_each_subsys(root, ss) {
2102 if (ss->can_attach) {
2103 retval = ss->can_attach(cgrp, &tset);
2106 goto out_cancel_attach;
2112 * step 2: make sure css_sets exist for all threads to be migrated.
2113 * we use find_css_set, which allocates a new one if necessary.
2115 for (i = 0; i < group_size; i++) {
2116 tc = flex_array_get(group, i);
2117 tc->cg = find_css_set(tc->task->cgroups, cgrp);
2120 goto out_put_css_set_refs;
2125 * step 3: now that we're guaranteed success wrt the css_sets,
2126 * proceed to move all tasks to the new cgroup. There are no
2127 * failure cases after here, so this is the commit point.
2129 for (i = 0; i < group_size; i++) {
2130 tc = flex_array_get(group, i);
2131 cgroup_task_migrate(cgrp, tc->cgrp, tc->task, tc->cg);
2133 /* nothing is sensitive to fork() after this point. */
2136 * step 4: do subsystem attach callbacks.
2138 for_each_subsys(root, ss) {
2140 ss->attach(cgrp, &tset);
2144 * step 5: success! and cleanup
2147 cgroup_wakeup_rmdir_waiter(cgrp);
2149 out_put_css_set_refs:
2151 for (i = 0; i < group_size; i++) {
2152 tc = flex_array_get(group, i);
2155 put_css_set(tc->cg);
2160 for_each_subsys(root, ss) {
2161 if (ss == failed_ss)
2163 if (ss->cancel_attach)
2164 ss->cancel_attach(cgrp, &tset);
2167 out_free_group_list:
2168 flex_array_free(group);
2173 * Find the task_struct of the task to attach by vpid and pass it along to the
2174 * function to attach either it or all tasks in its threadgroup. Will lock
2175 * cgroup_mutex and threadgroup; may take task_lock of task.
2177 static int attach_task_by_pid(struct cgroup *cgrp, u64 pid, bool threadgroup)
2179 struct task_struct *tsk;
2180 const struct cred *cred = current_cred(), *tcred;
2183 if (!cgroup_lock_live_group(cgrp))
2189 tsk = find_task_by_vpid(pid);
2193 goto out_unlock_cgroup;
2196 * even if we're attaching all tasks in the thread group, we
2197 * only need to check permissions on one of them.
2199 tcred = __task_cred(tsk);
2201 cred->euid != tcred->uid &&
2202 cred->euid != tcred->suid) {
2205 goto out_unlock_cgroup;
2211 tsk = tsk->group_leader;
2212 get_task_struct(tsk);
2215 threadgroup_lock(tsk);
2217 if (!thread_group_leader(tsk)) {
2219 * a race with de_thread from another thread's exec()
2220 * may strip us of our leadership, if this happens,
2221 * there is no choice but to throw this task away and
2222 * try again; this is
2223 * "double-double-toil-and-trouble-check locking".
2225 threadgroup_unlock(tsk);
2226 put_task_struct(tsk);
2227 goto retry_find_task;
2229 ret = cgroup_attach_proc(cgrp, tsk);
2231 ret = cgroup_attach_task(cgrp, tsk);
2232 threadgroup_unlock(tsk);
2234 put_task_struct(tsk);
2240 static int cgroup_tasks_write(struct cgroup *cgrp, struct cftype *cft, u64 pid)
2242 return attach_task_by_pid(cgrp, pid, false);
2245 static int cgroup_procs_write(struct cgroup *cgrp, struct cftype *cft, u64 tgid)
2247 return attach_task_by_pid(cgrp, tgid, true);
2251 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2252 * @cgrp: the cgroup to be checked for liveness
2254 * On success, returns true; the lock should be later released with
2255 * cgroup_unlock(). On failure returns false with no lock held.
2257 bool cgroup_lock_live_group(struct cgroup *cgrp)
2259 mutex_lock(&cgroup_mutex);
2260 if (cgroup_is_removed(cgrp)) {
2261 mutex_unlock(&cgroup_mutex);
2266 EXPORT_SYMBOL_GPL(cgroup_lock_live_group);
2268 static int cgroup_release_agent_write(struct cgroup *cgrp, struct cftype *cft,
2271 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
2272 if (strlen(buffer) >= PATH_MAX)
2274 if (!cgroup_lock_live_group(cgrp))
2276 mutex_lock(&cgroup_root_mutex);
2277 strcpy(cgrp->root->release_agent_path, buffer);
2278 mutex_unlock(&cgroup_root_mutex);
2283 static int cgroup_release_agent_show(struct cgroup *cgrp, struct cftype *cft,
2284 struct seq_file *seq)
2286 if (!cgroup_lock_live_group(cgrp))
2288 seq_puts(seq, cgrp->root->release_agent_path);
2289 seq_putc(seq, '\n');
2294 /* A buffer size big enough for numbers or short strings */
2295 #define CGROUP_LOCAL_BUFFER_SIZE 64
2297 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
2299 const char __user *userbuf,
2300 size_t nbytes, loff_t *unused_ppos)
2302 char buffer[CGROUP_LOCAL_BUFFER_SIZE];
2308 if (nbytes >= sizeof(buffer))
2310 if (copy_from_user(buffer, userbuf, nbytes))
2313 buffer[nbytes] = 0; /* nul-terminate */
2314 if (cft->write_u64) {
2315 u64 val = simple_strtoull(strstrip(buffer), &end, 0);
2318 retval = cft->write_u64(cgrp, cft, val);
2320 s64 val = simple_strtoll(strstrip(buffer), &end, 0);
2323 retval = cft->write_s64(cgrp, cft, val);
2330 static ssize_t cgroup_write_string(struct cgroup *cgrp, struct cftype *cft,
2332 const char __user *userbuf,
2333 size_t nbytes, loff_t *unused_ppos)
2335 char local_buffer[CGROUP_LOCAL_BUFFER_SIZE];
2337 size_t max_bytes = cft->max_write_len;
2338 char *buffer = local_buffer;
2341 max_bytes = sizeof(local_buffer) - 1;
2342 if (nbytes >= max_bytes)
2344 /* Allocate a dynamic buffer if we need one */
2345 if (nbytes >= sizeof(local_buffer)) {
2346 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
2350 if (nbytes && copy_from_user(buffer, userbuf, nbytes)) {
2355 buffer[nbytes] = 0; /* nul-terminate */
2356 retval = cft->write_string(cgrp, cft, strstrip(buffer));
2360 if (buffer != local_buffer)
2365 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
2366 size_t nbytes, loff_t *ppos)
2368 struct cftype *cft = __d_cft(file->f_dentry);
2369 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2371 if (cgroup_is_removed(cgrp))
2374 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
2375 if (cft->write_u64 || cft->write_s64)
2376 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
2377 if (cft->write_string)
2378 return cgroup_write_string(cgrp, cft, file, buf, nbytes, ppos);
2380 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
2381 return ret ? ret : nbytes;
2386 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
2388 char __user *buf, size_t nbytes,
2391 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2392 u64 val = cft->read_u64(cgrp, cft);
2393 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
2395 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2398 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
2400 char __user *buf, size_t nbytes,
2403 char tmp[CGROUP_LOCAL_BUFFER_SIZE];
2404 s64 val = cft->read_s64(cgrp, cft);
2405 int len = sprintf(tmp, "%lld\n", (long long) val);
2407 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
2410 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
2411 size_t nbytes, loff_t *ppos)
2413 struct cftype *cft = __d_cft(file->f_dentry);
2414 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2416 if (cgroup_is_removed(cgrp))
2420 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
2422 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
2424 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
2429 * seqfile ops/methods for returning structured data. Currently just
2430 * supports string->u64 maps, but can be extended in future.
2433 struct cgroup_seqfile_state {
2435 struct cgroup *cgroup;
2438 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
2440 struct seq_file *sf = cb->state;
2441 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
2444 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
2446 struct cgroup_seqfile_state *state = m->private;
2447 struct cftype *cft = state->cft;
2448 if (cft->read_map) {
2449 struct cgroup_map_cb cb = {
2450 .fill = cgroup_map_add,
2453 return cft->read_map(state->cgroup, cft, &cb);
2455 return cft->read_seq_string(state->cgroup, cft, m);
2458 static int cgroup_seqfile_release(struct inode *inode, struct file *file)
2460 struct seq_file *seq = file->private_data;
2461 kfree(seq->private);
2462 return single_release(inode, file);
2465 static const struct file_operations cgroup_seqfile_operations = {
2467 .write = cgroup_file_write,
2468 .llseek = seq_lseek,
2469 .release = cgroup_seqfile_release,
2472 static int cgroup_file_open(struct inode *inode, struct file *file)
2477 err = generic_file_open(inode, file);
2480 cft = __d_cft(file->f_dentry);
2482 if (cft->read_map || cft->read_seq_string) {
2483 struct cgroup_seqfile_state *state =
2484 kzalloc(sizeof(*state), GFP_USER);
2488 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
2489 file->f_op = &cgroup_seqfile_operations;
2490 err = single_open(file, cgroup_seqfile_show, state);
2493 } else if (cft->open)
2494 err = cft->open(inode, file);
2501 static int cgroup_file_release(struct inode *inode, struct file *file)
2503 struct cftype *cft = __d_cft(file->f_dentry);
2505 return cft->release(inode, file);
2510 * cgroup_rename - Only allow simple rename of directories in place.
2512 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2513 struct inode *new_dir, struct dentry *new_dentry)
2515 if (!S_ISDIR(old_dentry->d_inode->i_mode))
2517 if (new_dentry->d_inode)
2519 if (old_dir != new_dir)
2521 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2524 static const struct file_operations cgroup_file_operations = {
2525 .read = cgroup_file_read,
2526 .write = cgroup_file_write,
2527 .llseek = generic_file_llseek,
2528 .open = cgroup_file_open,
2529 .release = cgroup_file_release,
2532 static const struct inode_operations cgroup_dir_inode_operations = {
2533 .lookup = cgroup_lookup,
2534 .mkdir = cgroup_mkdir,
2535 .rmdir = cgroup_rmdir,
2536 .rename = cgroup_rename,
2539 static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
2541 if (dentry->d_name.len > NAME_MAX)
2542 return ERR_PTR(-ENAMETOOLONG);
2543 d_add(dentry, NULL);
2548 * Check if a file is a control file
2550 static inline struct cftype *__file_cft(struct file *file)
2552 if (file->f_dentry->d_inode->i_fop != &cgroup_file_operations)
2553 return ERR_PTR(-EINVAL);
2554 return __d_cft(file->f_dentry);
2557 static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2558 struct super_block *sb)
2560 struct inode *inode;
2564 if (dentry->d_inode)
2567 inode = cgroup_new_inode(mode, sb);
2571 if (S_ISDIR(mode)) {
2572 inode->i_op = &cgroup_dir_inode_operations;
2573 inode->i_fop = &simple_dir_operations;
2575 /* start off with i_nlink == 2 (for "." entry) */
2578 /* start with the directory inode held, so that we can
2579 * populate it without racing with another mkdir */
2580 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2581 } else if (S_ISREG(mode)) {
2583 inode->i_fop = &cgroup_file_operations;
2585 d_instantiate(dentry, inode);
2586 dget(dentry); /* Extra count - pin the dentry in core */
2591 * cgroup_create_dir - create a directory for an object.
2592 * @cgrp: the cgroup we create the directory for. It must have a valid
2593 * ->parent field. And we are going to fill its ->dentry field.
2594 * @dentry: dentry of the new cgroup
2595 * @mode: mode to set on new directory.
2597 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
2600 struct dentry *parent;
2603 parent = cgrp->parent->dentry;
2604 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
2606 dentry->d_fsdata = cgrp;
2607 inc_nlink(parent->d_inode);
2608 rcu_assign_pointer(cgrp->dentry, dentry);
2617 * cgroup_file_mode - deduce file mode of a control file
2618 * @cft: the control file in question
2620 * returns cft->mode if ->mode is not 0
2621 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2622 * returns S_IRUGO if it has only a read handler
2623 * returns S_IWUSR if it has only a write hander
2625 static umode_t cgroup_file_mode(const struct cftype *cft)
2632 if (cft->read || cft->read_u64 || cft->read_s64 ||
2633 cft->read_map || cft->read_seq_string)
2636 if (cft->write || cft->write_u64 || cft->write_s64 ||
2637 cft->write_string || cft->trigger)
2643 static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2644 const struct cftype *cft)
2646 struct dentry *dir = cgrp->dentry;
2647 struct cgroup *parent = __d_cgrp(dir);
2648 struct dentry *dentry;
2652 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2654 /* does @cft->flags tell us to skip creation on @cgrp? */
2655 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2657 if ((cft->flags & CFTYPE_ONLY_ON_ROOT) && cgrp->parent)
2660 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
2661 strcpy(name, subsys->name);
2664 strcat(name, cft->name);
2666 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
2668 cfe = kzalloc(sizeof(*cfe), GFP_KERNEL);
2672 dentry = lookup_one_len(name, dir, strlen(name));
2673 if (IS_ERR(dentry)) {
2674 error = PTR_ERR(dentry);
2678 mode = cgroup_file_mode(cft);
2679 error = cgroup_create_file(dentry, mode | S_IFREG, cgrp->root->sb);
2681 cfe->type = (void *)cft;
2682 cfe->dentry = dentry;
2683 dentry->d_fsdata = cfe;
2684 list_add_tail(&cfe->node, &parent->files);
2693 static int cgroup_add_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2694 const struct cftype cfts[])
2696 const struct cftype *cft;
2699 for (cft = cfts; cft->name[0] != '\0'; cft++) {
2700 err = cgroup_add_file(cgrp, subsys, cft);
2702 pr_warning("cgroup_add_files: failed to create %s, err=%d\n",
2710 static DEFINE_MUTEX(cgroup_cft_mutex);
2712 static void cgroup_cfts_prepare(void)
2713 __acquires(&cgroup_cft_mutex) __acquires(&cgroup_mutex)
2716 * Thanks to the entanglement with vfs inode locking, we can't walk
2717 * the existing cgroups under cgroup_mutex and create files.
2718 * Instead, we increment reference on all cgroups and build list of
2719 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2720 * exclusive access to the field.
2722 mutex_lock(&cgroup_cft_mutex);
2723 mutex_lock(&cgroup_mutex);
2726 static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2727 const struct cftype *cfts)
2728 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2731 struct cgroup *cgrp, *n;
2733 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2734 if (cfts && ss->root != &rootnode) {
2735 list_for_each_entry(cgrp, &ss->root->allcg_list, allcg_node) {
2737 list_add_tail(&cgrp->cft_q_node, &pending);
2741 mutex_unlock(&cgroup_mutex);
2744 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2745 * files for all cgroups which were created before.
2747 list_for_each_entry_safe(cgrp, n, &pending, cft_q_node) {
2748 struct inode *inode = cgrp->dentry->d_inode;
2750 mutex_lock(&inode->i_mutex);
2751 mutex_lock(&cgroup_mutex);
2752 if (!cgroup_is_removed(cgrp))
2753 cgroup_add_files(cgrp, ss, cfts);
2754 mutex_unlock(&cgroup_mutex);
2755 mutex_unlock(&inode->i_mutex);
2757 list_del_init(&cgrp->cft_q_node);
2761 mutex_unlock(&cgroup_cft_mutex);
2765 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2766 * @ss: target cgroup subsystem
2767 * @cfts: zero-length name terminated array of cftypes
2769 * Register @cfts to @ss. Files described by @cfts are created for all
2770 * existing cgroups to which @ss is attached and all future cgroups will
2771 * have them too. This function can be called anytime whether @ss is
2774 * Returns 0 on successful registration, -errno on failure. Note that this
2775 * function currently returns 0 as long as @cfts registration is successful
2776 * even if some file creation attempts on existing cgroups fail.
2778 int cgroup_add_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts)
2780 struct cftype_set *set;
2782 set = kzalloc(sizeof(*set), GFP_KERNEL);
2786 cgroup_cfts_prepare();
2788 list_add_tail(&set->node, &ss->cftsets);
2789 cgroup_cfts_commit(ss, cfts);
2793 EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2796 * cgroup_task_count - count the number of tasks in a cgroup.
2797 * @cgrp: the cgroup in question
2799 * Return the number of tasks in the cgroup.
2801 int cgroup_task_count(const struct cgroup *cgrp)
2804 struct cg_cgroup_link *link;
2806 read_lock(&css_set_lock);
2807 list_for_each_entry(link, &cgrp->css_sets, cgrp_link_list) {
2808 count += atomic_read(&link->cg->refcount);
2810 read_unlock(&css_set_lock);
2815 * Advance a list_head iterator. The iterator should be positioned at
2816 * the start of a css_set
2818 static void cgroup_advance_iter(struct cgroup *cgrp,
2819 struct cgroup_iter *it)
2821 struct list_head *l = it->cg_link;
2822 struct cg_cgroup_link *link;
2825 /* Advance to the next non-empty css_set */
2828 if (l == &cgrp->css_sets) {
2832 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
2834 } while (list_empty(&cg->tasks));
2836 it->task = cg->tasks.next;
2840 * To reduce the fork() overhead for systems that are not actually
2841 * using their cgroups capability, we don't maintain the lists running
2842 * through each css_set to its tasks until we see the list actually
2843 * used - in other words after the first call to cgroup_iter_start().
2845 static void cgroup_enable_task_cg_lists(void)
2847 struct task_struct *p, *g;
2848 write_lock(&css_set_lock);
2849 use_task_css_set_links = 1;
2851 * We need tasklist_lock because RCU is not safe against
2852 * while_each_thread(). Besides, a forking task that has passed
2853 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2854 * is not guaranteed to have its child immediately visible in the
2855 * tasklist if we walk through it with RCU.
2857 read_lock(&tasklist_lock);
2858 do_each_thread(g, p) {
2861 * We should check if the process is exiting, otherwise
2862 * it will race with cgroup_exit() in that the list
2863 * entry won't be deleted though the process has exited.
2865 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
2866 list_add(&p->cg_list, &p->cgroups->tasks);
2868 } while_each_thread(g, p);
2869 read_unlock(&tasklist_lock);
2870 write_unlock(&css_set_lock);
2873 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
2874 __acquires(css_set_lock)
2877 * The first time anyone tries to iterate across a cgroup,
2878 * we need to enable the list linking each css_set to its
2879 * tasks, and fix up all existing tasks.
2881 if (!use_task_css_set_links)
2882 cgroup_enable_task_cg_lists();
2884 read_lock(&css_set_lock);
2885 it->cg_link = &cgrp->css_sets;
2886 cgroup_advance_iter(cgrp, it);
2889 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
2890 struct cgroup_iter *it)
2892 struct task_struct *res;
2893 struct list_head *l = it->task;
2894 struct cg_cgroup_link *link;
2896 /* If the iterator cg is NULL, we have no tasks */
2899 res = list_entry(l, struct task_struct, cg_list);
2900 /* Advance iterator to find next entry */
2902 link = list_entry(it->cg_link, struct cg_cgroup_link, cgrp_link_list);
2903 if (l == &link->cg->tasks) {
2904 /* We reached the end of this task list - move on to
2905 * the next cg_cgroup_link */
2906 cgroup_advance_iter(cgrp, it);
2913 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
2914 __releases(css_set_lock)
2916 read_unlock(&css_set_lock);
2919 static inline int started_after_time(struct task_struct *t1,
2920 struct timespec *time,
2921 struct task_struct *t2)
2923 int start_diff = timespec_compare(&t1->start_time, time);
2924 if (start_diff > 0) {
2926 } else if (start_diff < 0) {
2930 * Arbitrarily, if two processes started at the same
2931 * time, we'll say that the lower pointer value
2932 * started first. Note that t2 may have exited by now
2933 * so this may not be a valid pointer any longer, but
2934 * that's fine - it still serves to distinguish
2935 * between two tasks started (effectively) simultaneously.
2942 * This function is a callback from heap_insert() and is used to order
2944 * In this case we order the heap in descending task start time.
2946 static inline int started_after(void *p1, void *p2)
2948 struct task_struct *t1 = p1;
2949 struct task_struct *t2 = p2;
2950 return started_after_time(t1, &t2->start_time, t2);
2954 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2955 * @scan: struct cgroup_scanner containing arguments for the scan
2957 * Arguments include pointers to callback functions test_task() and
2959 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2960 * and if it returns true, call process_task() for it also.
2961 * The test_task pointer may be NULL, meaning always true (select all tasks).
2962 * Effectively duplicates cgroup_iter_{start,next,end}()
2963 * but does not lock css_set_lock for the call to process_task().
2964 * The struct cgroup_scanner may be embedded in any structure of the caller's
2966 * It is guaranteed that process_task() will act on every task that
2967 * is a member of the cgroup for the duration of this call. This
2968 * function may or may not call process_task() for tasks that exit
2969 * or move to a different cgroup during the call, or are forked or
2970 * move into the cgroup during the call.
2972 * Note that test_task() may be called with locks held, and may in some
2973 * situations be called multiple times for the same task, so it should
2975 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2976 * pre-allocated and will be used for heap operations (and its "gt" member will
2977 * be overwritten), else a temporary heap will be used (allocation of which
2978 * may cause this function to fail).
2980 int cgroup_scan_tasks(struct cgroup_scanner *scan)
2983 struct cgroup_iter it;
2984 struct task_struct *p, *dropped;
2985 /* Never dereference latest_task, since it's not refcounted */
2986 struct task_struct *latest_task = NULL;
2987 struct ptr_heap tmp_heap;
2988 struct ptr_heap *heap;
2989 struct timespec latest_time = { 0, 0 };
2992 /* The caller supplied our heap and pre-allocated its memory */
2994 heap->gt = &started_after;
2996 /* We need to allocate our own heap memory */
2998 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
3000 /* cannot allocate the heap */
3006 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3007 * to determine which are of interest, and using the scanner's
3008 * "process_task" callback to process any of them that need an update.
3009 * Since we don't want to hold any locks during the task updates,
3010 * gather tasks to be processed in a heap structure.
3011 * The heap is sorted by descending task start time.
3012 * If the statically-sized heap fills up, we overflow tasks that
3013 * started later, and in future iterations only consider tasks that
3014 * started after the latest task in the previous pass. This
3015 * guarantees forward progress and that we don't miss any tasks.
3018 cgroup_iter_start(scan->cg, &it);
3019 while ((p = cgroup_iter_next(scan->cg, &it))) {
3021 * Only affect tasks that qualify per the caller's callback,
3022 * if he provided one
3024 if (scan->test_task && !scan->test_task(p, scan))
3027 * Only process tasks that started after the last task
3030 if (!started_after_time(p, &latest_time, latest_task))
3032 dropped = heap_insert(heap, p);
3033 if (dropped == NULL) {
3035 * The new task was inserted; the heap wasn't
3039 } else if (dropped != p) {
3041 * The new task was inserted, and pushed out a
3045 put_task_struct(dropped);
3048 * Else the new task was newer than anything already in
3049 * the heap and wasn't inserted
3052 cgroup_iter_end(scan->cg, &it);
3055 for (i = 0; i < heap->size; i++) {
3056 struct task_struct *q = heap->ptrs[i];
3058 latest_time = q->start_time;
3061 /* Process the task per the caller's callback */
3062 scan->process_task(q, scan);
3066 * If we had to process any tasks at all, scan again
3067 * in case some of them were in the middle of forking
3068 * children that didn't get processed.
3069 * Not the most efficient way to do it, but it avoids
3070 * having to take callback_mutex in the fork path
3074 if (heap == &tmp_heap)
3075 heap_free(&tmp_heap);
3080 * Stuff for reading the 'tasks'/'procs' files.
3082 * Reading this file can return large amounts of data if a cgroup has
3083 * *lots* of attached tasks. So it may need several calls to read(),
3084 * but we cannot guarantee that the information we produce is correct
3085 * unless we produce it entirely atomically.
3089 /* which pidlist file are we talking about? */
3090 enum cgroup_filetype {
3096 * A pidlist is a list of pids that virtually represents the contents of one
3097 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3098 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3101 struct cgroup_pidlist {
3103 * used to find which pidlist is wanted. doesn't change as long as
3104 * this particular list stays in the list.
3106 struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
3109 /* how many elements the above list has */
3111 /* how many files are using the current array */
3113 /* each of these stored in a list by its cgroup */
3114 struct list_head links;
3115 /* pointer to the cgroup we belong to, for list removal purposes */
3116 struct cgroup *owner;
3117 /* protects the other fields */
3118 struct rw_semaphore mutex;
3122 * The following two functions "fix" the issue where there are more pids
3123 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3124 * TODO: replace with a kernel-wide solution to this problem
3126 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3127 static void *pidlist_allocate(int count)
3129 if (PIDLIST_TOO_LARGE(count))
3130 return vmalloc(count * sizeof(pid_t));
3132 return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
3134 static void pidlist_free(void *p)
3136 if (is_vmalloc_addr(p))
3141 static void *pidlist_resize(void *p, int newcount)
3144 /* note: if new alloc fails, old p will still be valid either way */
3145 if (is_vmalloc_addr(p)) {
3146 newlist = vmalloc(newcount * sizeof(pid_t));
3149 memcpy(newlist, p, newcount * sizeof(pid_t));
3152 newlist = krealloc(p, newcount * sizeof(pid_t), GFP_KERNEL);
3158 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3159 * If the new stripped list is sufficiently smaller and there's enough memory
3160 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3161 * number of unique elements.
3163 /* is the size difference enough that we should re-allocate the array? */
3164 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3165 static int pidlist_uniq(pid_t **p, int length)
3172 * we presume the 0th element is unique, so i starts at 1. trivial
3173 * edge cases first; no work needs to be done for either
3175 if (length == 0 || length == 1)
3177 /* src and dest walk down the list; dest counts unique elements */
3178 for (src = 1; src < length; src++) {
3179 /* find next unique element */
3180 while (list[src] == list[src-1]) {
3185 /* dest always points to where the next unique element goes */
3186 list[dest] = list[src];
3191 * if the length difference is large enough, we want to allocate a
3192 * smaller buffer to save memory. if this fails due to out of memory,
3193 * we'll just stay with what we've got.
3195 if (PIDLIST_REALLOC_DIFFERENCE(length, dest)) {
3196 newlist = pidlist_resize(list, dest);
3203 static int cmppid(const void *a, const void *b)
3205 return *(pid_t *)a - *(pid_t *)b;
3209 * find the appropriate pidlist for our purpose (given procs vs tasks)
3210 * returns with the lock on that pidlist already held, and takes care
3211 * of the use count, or returns NULL with no locks held if we're out of
3214 static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
3215 enum cgroup_filetype type)
3217 struct cgroup_pidlist *l;
3218 /* don't need task_nsproxy() if we're looking at ourself */
3219 struct pid_namespace *ns = current->nsproxy->pid_ns;
3222 * We can't drop the pidlist_mutex before taking the l->mutex in case
3223 * the last ref-holder is trying to remove l from the list at the same
3224 * time. Holding the pidlist_mutex precludes somebody taking whichever
3225 * list we find out from under us - compare release_pid_array().
3227 mutex_lock(&cgrp->pidlist_mutex);
3228 list_for_each_entry(l, &cgrp->pidlists, links) {
3229 if (l->key.type == type && l->key.ns == ns) {
3230 /* make sure l doesn't vanish out from under us */
3231 down_write(&l->mutex);
3232 mutex_unlock(&cgrp->pidlist_mutex);
3236 /* entry not found; create a new one */
3237 l = kmalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
3239 mutex_unlock(&cgrp->pidlist_mutex);
3242 init_rwsem(&l->mutex);
3243 down_write(&l->mutex);
3245 l->key.ns = get_pid_ns(ns);
3246 l->use_count = 0; /* don't increment here */
3249 list_add(&l->links, &cgrp->pidlists);
3250 mutex_unlock(&cgrp->pidlist_mutex);
3255 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3257 static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
3258 struct cgroup_pidlist **lp)
3262 int pid, n = 0; /* used for populating the array */
3263 struct cgroup_iter it;
3264 struct task_struct *tsk;
3265 struct cgroup_pidlist *l;
3268 * If cgroup gets more users after we read count, we won't have
3269 * enough space - tough. This race is indistinguishable to the
3270 * caller from the case that the additional cgroup users didn't
3271 * show up until sometime later on.
3273 length = cgroup_task_count(cgrp);
3274 array = pidlist_allocate(length);
3277 /* now, populate the array */
3278 cgroup_iter_start(cgrp, &it);
3279 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3280 if (unlikely(n == length))
3282 /* get tgid or pid for procs or tasks file respectively */
3283 if (type == CGROUP_FILE_PROCS)
3284 pid = task_tgid_vnr(tsk);
3286 pid = task_pid_vnr(tsk);
3287 if (pid > 0) /* make sure to only use valid results */
3290 cgroup_iter_end(cgrp, &it);
3292 /* now sort & (if procs) strip out duplicates */
3293 sort(array, length, sizeof(pid_t), cmppid, NULL);
3294 if (type == CGROUP_FILE_PROCS)
3295 length = pidlist_uniq(&array, length);
3296 l = cgroup_pidlist_find(cgrp, type);
3298 pidlist_free(array);
3301 /* store array, freeing old if necessary - lock already held */
3302 pidlist_free(l->list);
3306 up_write(&l->mutex);
3312 * cgroupstats_build - build and fill cgroupstats
3313 * @stats: cgroupstats to fill information into
3314 * @dentry: A dentry entry belonging to the cgroup for which stats have
3317 * Build and fill cgroupstats so that taskstats can export it to user
3320 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
3323 struct cgroup *cgrp;
3324 struct cgroup_iter it;
3325 struct task_struct *tsk;
3328 * Validate dentry by checking the superblock operations,
3329 * and make sure it's a directory.
3331 if (dentry->d_sb->s_op != &cgroup_ops ||
3332 !S_ISDIR(dentry->d_inode->i_mode))
3336 cgrp = dentry->d_fsdata;
3338 cgroup_iter_start(cgrp, &it);
3339 while ((tsk = cgroup_iter_next(cgrp, &it))) {
3340 switch (tsk->state) {
3342 stats->nr_running++;
3344 case TASK_INTERRUPTIBLE:
3345 stats->nr_sleeping++;
3347 case TASK_UNINTERRUPTIBLE:
3348 stats->nr_uninterruptible++;
3351 stats->nr_stopped++;
3354 if (delayacct_is_task_waiting_on_io(tsk))
3355 stats->nr_io_wait++;
3359 cgroup_iter_end(cgrp, &it);
3367 * seq_file methods for the tasks/procs files. The seq_file position is the
3368 * next pid to display; the seq_file iterator is a pointer to the pid
3369 * in the cgroup->l->list array.
3372 static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
3375 * Initially we receive a position value that corresponds to
3376 * one more than the last pid shown (or 0 on the first call or
3377 * after a seek to the start). Use a binary-search to find the
3378 * next pid to display, if any
3380 struct cgroup_pidlist *l = s->private;
3381 int index = 0, pid = *pos;
3384 down_read(&l->mutex);
3386 int end = l->length;
3388 while (index < end) {
3389 int mid = (index + end) / 2;
3390 if (l->list[mid] == pid) {
3393 } else if (l->list[mid] <= pid)
3399 /* If we're off the end of the array, we're done */
3400 if (index >= l->length)
3402 /* Update the abstract position to be the actual pid that we found */
3403 iter = l->list + index;
3408 static void cgroup_pidlist_stop(struct seq_file *s, void *v)
3410 struct cgroup_pidlist *l = s->private;
3414 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
3416 struct cgroup_pidlist *l = s->private;
3418 pid_t *end = l->list + l->length;
3420 * Advance to the next pid in the array. If this goes off the
3432 static int cgroup_pidlist_show(struct seq_file *s, void *v)
3434 return seq_printf(s, "%d\n", *(int *)v);
3438 * seq_operations functions for iterating on pidlists through seq_file -
3439 * independent of whether it's tasks or procs
3441 static const struct seq_operations cgroup_pidlist_seq_operations = {
3442 .start = cgroup_pidlist_start,
3443 .stop = cgroup_pidlist_stop,
3444 .next = cgroup_pidlist_next,
3445 .show = cgroup_pidlist_show,
3448 static void cgroup_release_pid_array(struct cgroup_pidlist *l)
3451 * the case where we're the last user of this particular pidlist will
3452 * have us remove it from the cgroup's list, which entails taking the
3453 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3454 * pidlist_mutex, we have to take pidlist_mutex first.
3456 mutex_lock(&l->owner->pidlist_mutex);
3457 down_write(&l->mutex);
3458 BUG_ON(!l->use_count);
3459 if (!--l->use_count) {
3460 /* we're the last user if refcount is 0; remove and free */
3461 list_del(&l->links);
3462 mutex_unlock(&l->owner->pidlist_mutex);
3463 pidlist_free(l->list);
3464 put_pid_ns(l->key.ns);
3465 up_write(&l->mutex);
3469 mutex_unlock(&l->owner->pidlist_mutex);
3470 up_write(&l->mutex);
3473 static int cgroup_pidlist_release(struct inode *inode, struct file *file)
3475 struct cgroup_pidlist *l;
3476 if (!(file->f_mode & FMODE_READ))
3479 * the seq_file will only be initialized if the file was opened for
3480 * reading; hence we check if it's not null only in that case.
3482 l = ((struct seq_file *)file->private_data)->private;
3483 cgroup_release_pid_array(l);
3484 return seq_release(inode, file);
3487 static const struct file_operations cgroup_pidlist_operations = {
3489 .llseek = seq_lseek,
3490 .write = cgroup_file_write,
3491 .release = cgroup_pidlist_release,
3495 * The following functions handle opens on a file that displays a pidlist
3496 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3499 /* helper function for the two below it */
3500 static int cgroup_pidlist_open(struct file *file, enum cgroup_filetype type)
3502 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
3503 struct cgroup_pidlist *l;
3506 /* Nothing to do for write-only files */
3507 if (!(file->f_mode & FMODE_READ))
3510 /* have the array populated */
3511 retval = pidlist_array_load(cgrp, type, &l);
3514 /* configure file information */
3515 file->f_op = &cgroup_pidlist_operations;
3517 retval = seq_open(file, &cgroup_pidlist_seq_operations);
3519 cgroup_release_pid_array(l);
3522 ((struct seq_file *)file->private_data)->private = l;
3525 static int cgroup_tasks_open(struct inode *unused, struct file *file)
3527 return cgroup_pidlist_open(file, CGROUP_FILE_TASKS);
3529 static int cgroup_procs_open(struct inode *unused, struct file *file)
3531 return cgroup_pidlist_open(file, CGROUP_FILE_PROCS);
3534 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
3537 return notify_on_release(cgrp);
3540 static int cgroup_write_notify_on_release(struct cgroup *cgrp,
3544 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
3546 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3548 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3553 * Unregister event and free resources.
3555 * Gets called from workqueue.
3557 static void cgroup_event_remove(struct work_struct *work)
3559 struct cgroup_event *event = container_of(work, struct cgroup_event,
3561 struct cgroup *cgrp = event->cgrp;
3563 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3565 eventfd_ctx_put(event->eventfd);
3571 * Gets called on POLLHUP on eventfd when user closes it.
3573 * Called with wqh->lock held and interrupts disabled.
3575 static int cgroup_event_wake(wait_queue_t *wait, unsigned mode,
3576 int sync, void *key)
3578 struct cgroup_event *event = container_of(wait,
3579 struct cgroup_event, wait);
3580 struct cgroup *cgrp = event->cgrp;
3581 unsigned long flags = (unsigned long)key;
3583 if (flags & POLLHUP) {
3584 __remove_wait_queue(event->wqh, &event->wait);
3585 spin_lock(&cgrp->event_list_lock);
3586 list_del(&event->list);
3587 spin_unlock(&cgrp->event_list_lock);
3589 * We are in atomic context, but cgroup_event_remove() may
3590 * sleep, so we have to call it in workqueue.
3592 schedule_work(&event->remove);
3598 static void cgroup_event_ptable_queue_proc(struct file *file,
3599 wait_queue_head_t *wqh, poll_table *pt)
3601 struct cgroup_event *event = container_of(pt,
3602 struct cgroup_event, pt);
3605 add_wait_queue(wqh, &event->wait);
3609 * Parse input and register new cgroup event handler.
3611 * Input must be in format '<event_fd> <control_fd> <args>'.
3612 * Interpretation of args is defined by control file implementation.
3614 static int cgroup_write_event_control(struct cgroup *cgrp, struct cftype *cft,
3617 struct cgroup_event *event = NULL;
3618 unsigned int efd, cfd;
3619 struct file *efile = NULL;
3620 struct file *cfile = NULL;
3624 efd = simple_strtoul(buffer, &endp, 10);
3629 cfd = simple_strtoul(buffer, &endp, 10);
3630 if ((*endp != ' ') && (*endp != '\0'))
3634 event = kzalloc(sizeof(*event), GFP_KERNEL);
3638 INIT_LIST_HEAD(&event->list);
3639 init_poll_funcptr(&event->pt, cgroup_event_ptable_queue_proc);
3640 init_waitqueue_func_entry(&event->wait, cgroup_event_wake);
3641 INIT_WORK(&event->remove, cgroup_event_remove);
3643 efile = eventfd_fget(efd);
3644 if (IS_ERR(efile)) {
3645 ret = PTR_ERR(efile);
3649 event->eventfd = eventfd_ctx_fileget(efile);
3650 if (IS_ERR(event->eventfd)) {
3651 ret = PTR_ERR(event->eventfd);
3661 /* the process need read permission on control file */
3662 /* AV: shouldn't we check that it's been opened for read instead? */
3663 ret = inode_permission(cfile->f_path.dentry->d_inode, MAY_READ);
3667 event->cft = __file_cft(cfile);
3668 if (IS_ERR(event->cft)) {
3669 ret = PTR_ERR(event->cft);
3673 if (!event->cft->register_event || !event->cft->unregister_event) {
3678 ret = event->cft->register_event(cgrp, event->cft,
3679 event->eventfd, buffer);
3683 if (efile->f_op->poll(efile, &event->pt) & POLLHUP) {
3684 event->cft->unregister_event(cgrp, event->cft, event->eventfd);
3690 * Events should be removed after rmdir of cgroup directory, but before
3691 * destroying subsystem state objects. Let's take reference to cgroup
3692 * directory dentry to do that.
3696 spin_lock(&cgrp->event_list_lock);
3697 list_add(&event->list, &cgrp->event_list);
3698 spin_unlock(&cgrp->event_list_lock);
3709 if (event && event->eventfd && !IS_ERR(event->eventfd))
3710 eventfd_ctx_put(event->eventfd);
3712 if (!IS_ERR_OR_NULL(efile))
3720 static u64 cgroup_clone_children_read(struct cgroup *cgrp,
3723 return clone_children(cgrp);
3726 static int cgroup_clone_children_write(struct cgroup *cgrp,
3731 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3733 clear_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3738 * for the common functions, 'private' gives the type of file
3740 /* for hysterical raisins, we can't put this on the older files */
3741 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3742 static struct cftype files[] = {
3745 .open = cgroup_tasks_open,
3746 .write_u64 = cgroup_tasks_write,
3747 .release = cgroup_pidlist_release,
3748 .mode = S_IRUGO | S_IWUSR,
3751 .name = CGROUP_FILE_GENERIC_PREFIX "procs",
3752 .open = cgroup_procs_open,
3753 .write_u64 = cgroup_procs_write,
3754 .release = cgroup_pidlist_release,
3755 .mode = S_IRUGO | S_IWUSR,
3758 .name = "notify_on_release",
3759 .read_u64 = cgroup_read_notify_on_release,
3760 .write_u64 = cgroup_write_notify_on_release,
3763 .name = CGROUP_FILE_GENERIC_PREFIX "event_control",
3764 .write_string = cgroup_write_event_control,
3768 .name = "cgroup.clone_children",
3769 .read_u64 = cgroup_clone_children_read,
3770 .write_u64 = cgroup_clone_children_write,
3773 .name = "release_agent",
3774 .flags = CFTYPE_ONLY_ON_ROOT,
3775 .read_seq_string = cgroup_release_agent_show,
3776 .write_string = cgroup_release_agent_write,
3777 .max_write_len = PATH_MAX,
3782 static int cgroup_populate_dir(struct cgroup *cgrp)
3785 struct cgroup_subsys *ss;
3787 err = cgroup_add_files(cgrp, NULL, files);
3791 /* process cftsets of each subsystem */
3792 for_each_subsys(cgrp->root, ss) {
3793 struct cftype_set *set;
3795 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
3798 list_for_each_entry(set, &ss->cftsets, node)
3799 cgroup_add_files(cgrp, ss, set->cfts);
3802 /* This cgroup is ready now */
3803 for_each_subsys(cgrp->root, ss) {
3804 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
3806 * Update id->css pointer and make this css visible from
3807 * CSS ID functions. This pointer will be dereferened
3808 * from RCU-read-side without locks.
3811 rcu_assign_pointer(css->id->css, css);
3817 static void init_cgroup_css(struct cgroup_subsys_state *css,
3818 struct cgroup_subsys *ss,
3819 struct cgroup *cgrp)
3822 atomic_set(&css->refcnt, 1);
3825 if (cgrp == dummytop)
3826 set_bit(CSS_ROOT, &css->flags);
3827 BUG_ON(cgrp->subsys[ss->subsys_id]);
3828 cgrp->subsys[ss->subsys_id] = css;
3831 static void cgroup_lock_hierarchy(struct cgroupfs_root *root)
3833 /* We need to take each hierarchy_mutex in a consistent order */
3837 * No worry about a race with rebind_subsystems that might mess up the
3838 * locking order, since both parties are under cgroup_mutex.
3840 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3841 struct cgroup_subsys *ss = subsys[i];
3844 if (ss->root == root)
3845 mutex_lock(&ss->hierarchy_mutex);
3849 static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
3853 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3854 struct cgroup_subsys *ss = subsys[i];
3857 if (ss->root == root)
3858 mutex_unlock(&ss->hierarchy_mutex);
3863 * cgroup_create - create a cgroup
3864 * @parent: cgroup that will be parent of the new cgroup
3865 * @dentry: dentry of the new cgroup
3866 * @mode: mode to set on new inode
3868 * Must be called with the mutex on the parent inode held
3870 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3873 struct cgroup *cgrp;
3874 struct cgroupfs_root *root = parent->root;
3876 struct cgroup_subsys *ss;
3877 struct super_block *sb = root->sb;
3879 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
3883 /* Grab a reference on the superblock so the hierarchy doesn't
3884 * get deleted on unmount if there are child cgroups. This
3885 * can be done outside cgroup_mutex, since the sb can't
3886 * disappear while someone has an open control file on the
3888 atomic_inc(&sb->s_active);
3890 mutex_lock(&cgroup_mutex);
3892 init_cgroup_housekeeping(cgrp);
3894 cgrp->parent = parent;
3895 cgrp->root = parent->root;
3896 cgrp->top_cgroup = parent->top_cgroup;
3898 if (notify_on_release(parent))
3899 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
3901 if (clone_children(parent))
3902 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3904 for_each_subsys(root, ss) {
3905 struct cgroup_subsys_state *css = ss->create(cgrp);
3911 init_cgroup_css(css, ss, cgrp);
3913 err = alloc_css_id(ss, parent, cgrp);
3917 /* At error, ->destroy() callback has to free assigned ID. */
3918 if (clone_children(parent) && ss->post_clone)
3919 ss->post_clone(cgrp);
3922 cgroup_lock_hierarchy(root);
3923 list_add(&cgrp->sibling, &cgrp->parent->children);
3924 cgroup_unlock_hierarchy(root);
3925 root->number_of_cgroups++;
3927 err = cgroup_create_dir(cgrp, dentry, mode);
3931 /* The cgroup directory was pre-locked for us */
3932 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
3934 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
3936 err = cgroup_populate_dir(cgrp);
3937 /* If err < 0, we have a half-filled directory - oh well ;) */
3939 mutex_unlock(&cgroup_mutex);
3940 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
3946 cgroup_lock_hierarchy(root);
3947 list_del(&cgrp->sibling);
3948 cgroup_unlock_hierarchy(root);
3949 root->number_of_cgroups--;
3953 for_each_subsys(root, ss) {
3954 if (cgrp->subsys[ss->subsys_id])
3958 mutex_unlock(&cgroup_mutex);
3960 /* Release the reference count that we took on the superblock */
3961 deactivate_super(sb);
3967 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3969 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
3971 /* the vfs holds inode->i_mutex already */
3972 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
3975 static int cgroup_has_css_refs(struct cgroup *cgrp)
3977 /* Check the reference count on each subsystem. Since we
3978 * already established that there are no tasks in the
3979 * cgroup, if the css refcount is also 1, then there should
3980 * be no outstanding references, so the subsystem is safe to
3981 * destroy. We scan across all subsystems rather than using
3982 * the per-hierarchy linked list of mounted subsystems since
3983 * we can be called via check_for_release() with no
3984 * synchronization other than RCU, and the subsystem linked
3985 * list isn't RCU-safe */
3988 * We won't need to lock the subsys array, because the subsystems
3989 * we're concerned about aren't going anywhere since our cgroup root
3990 * has a reference on them.
3992 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3993 struct cgroup_subsys *ss = subsys[i];
3994 struct cgroup_subsys_state *css;
3995 /* Skip subsystems not present or not in this hierarchy */
3996 if (ss == NULL || ss->root != cgrp->root)
3998 css = cgrp->subsys[ss->subsys_id];
3999 /* When called from check_for_release() it's possible
4000 * that by this point the cgroup has been removed
4001 * and the css deleted. But a false-positive doesn't
4002 * matter, since it can only happen if the cgroup
4003 * has been deleted and hence no longer needs the
4004 * release agent to be called anyway. */
4005 if (css && (atomic_read(&css->refcnt) > 1))
4012 * Atomically mark all (or else none) of the cgroup's CSS objects as
4013 * CSS_REMOVED. Return true on success, or false if the cgroup has
4014 * busy subsystems. Call with cgroup_mutex held
4017 static int cgroup_clear_css_refs(struct cgroup *cgrp)
4019 struct cgroup_subsys *ss;
4020 unsigned long flags;
4021 bool failed = false;
4022 local_irq_save(flags);
4023 for_each_subsys(cgrp->root, ss) {
4024 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4027 /* We can only remove a CSS with a refcnt==1 */
4028 refcnt = atomic_read(&css->refcnt);
4035 * Drop the refcnt to 0 while we check other
4036 * subsystems. This will cause any racing
4037 * css_tryget() to spin until we set the
4038 * CSS_REMOVED bits or abort
4040 if (atomic_cmpxchg(&css->refcnt, refcnt, 0) == refcnt)
4046 for_each_subsys(cgrp->root, ss) {
4047 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
4050 * Restore old refcnt if we previously managed
4051 * to clear it from 1 to 0
4053 if (!atomic_read(&css->refcnt))
4054 atomic_set(&css->refcnt, 1);
4056 /* Commit the fact that the CSS is removed */
4057 set_bit(CSS_REMOVED, &css->flags);
4060 local_irq_restore(flags);
4064 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
4066 struct cgroup *cgrp = dentry->d_fsdata;
4068 struct cgroup *parent;
4070 struct cgroup_event *event, *tmp;
4073 /* the vfs holds both inode->i_mutex already */
4075 mutex_lock(&cgroup_mutex);
4076 if (atomic_read(&cgrp->count) != 0) {
4077 mutex_unlock(&cgroup_mutex);
4080 if (!list_empty(&cgrp->children)) {
4081 mutex_unlock(&cgroup_mutex);
4084 mutex_unlock(&cgroup_mutex);
4087 * In general, subsystem has no css->refcnt after pre_destroy(). But
4088 * in racy cases, subsystem may have to get css->refcnt after
4089 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4090 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4091 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4092 * and subsystem's reference count handling. Please see css_get/put
4093 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4095 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4098 * Call pre_destroy handlers of subsys. Notify subsystems
4099 * that rmdir() request comes.
4101 ret = cgroup_call_pre_destroy(cgrp);
4103 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4107 mutex_lock(&cgroup_mutex);
4108 parent = cgrp->parent;
4109 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
4110 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4111 mutex_unlock(&cgroup_mutex);
4114 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
4115 if (!cgroup_clear_css_refs(cgrp)) {
4116 mutex_unlock(&cgroup_mutex);
4118 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4119 * prepare_to_wait(), we need to check this flag.
4121 if (test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags))
4123 finish_wait(&cgroup_rmdir_waitq, &wait);
4124 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4125 if (signal_pending(current))
4129 /* NO css_tryget() can success after here. */
4130 finish_wait(&cgroup_rmdir_waitq, &wait);
4131 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
4133 raw_spin_lock(&release_list_lock);
4134 set_bit(CGRP_REMOVED, &cgrp->flags);
4135 if (!list_empty(&cgrp->release_list))
4136 list_del_init(&cgrp->release_list);
4137 raw_spin_unlock(&release_list_lock);
4139 cgroup_lock_hierarchy(cgrp->root);
4140 /* delete this cgroup from parent->children */
4141 list_del_init(&cgrp->sibling);
4142 cgroup_unlock_hierarchy(cgrp->root);
4144 list_del_init(&cgrp->allcg_node);
4146 d = dget(cgrp->dentry);
4148 cgroup_d_remove_dir(d);
4151 set_bit(CGRP_RELEASABLE, &parent->flags);
4152 check_for_release(parent);
4155 * Unregister events and notify userspace.
4156 * Notify userspace about cgroup removing only after rmdir of cgroup
4157 * directory to avoid race between userspace and kernelspace
4159 spin_lock(&cgrp->event_list_lock);
4160 list_for_each_entry_safe(event, tmp, &cgrp->event_list, list) {
4161 list_del(&event->list);
4162 remove_wait_queue(event->wqh, &event->wait);
4163 eventfd_signal(event->eventfd, 1);
4164 schedule_work(&event->remove);
4166 spin_unlock(&cgrp->event_list_lock);
4168 mutex_unlock(&cgroup_mutex);
4172 static void __init_or_module cgroup_init_cftsets(struct cgroup_subsys *ss)
4174 INIT_LIST_HEAD(&ss->cftsets);
4177 * base_cftset is embedded in subsys itself, no need to worry about
4180 if (ss->base_cftypes) {
4181 ss->base_cftset.cfts = ss->base_cftypes;
4182 list_add_tail(&ss->base_cftset.node, &ss->cftsets);
4186 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
4188 struct cgroup_subsys_state *css;
4190 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
4192 /* init base cftset */
4193 cgroup_init_cftsets(ss);
4195 /* Create the top cgroup state for this subsystem */
4196 list_add(&ss->sibling, &rootnode.subsys_list);
4197 ss->root = &rootnode;
4198 css = ss->create(dummytop);
4199 /* We don't handle early failures gracefully */
4200 BUG_ON(IS_ERR(css));
4201 init_cgroup_css(css, ss, dummytop);
4203 /* Update the init_css_set to contain a subsys
4204 * pointer to this state - since the subsystem is
4205 * newly registered, all tasks and hence the
4206 * init_css_set is in the subsystem's top cgroup. */
4207 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
4209 need_forkexit_callback |= ss->fork || ss->exit;
4211 /* At system boot, before all subsystems have been
4212 * registered, no tasks have been forked, so we don't
4213 * need to invoke fork callbacks here. */
4214 BUG_ON(!list_empty(&init_task.tasks));
4216 mutex_init(&ss->hierarchy_mutex);
4217 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4220 /* this function shouldn't be used with modular subsystems, since they
4221 * need to register a subsys_id, among other things */
4226 * cgroup_load_subsys: load and register a modular subsystem at runtime
4227 * @ss: the subsystem to load
4229 * This function should be called in a modular subsystem's initcall. If the
4230 * subsystem is built as a module, it will be assigned a new subsys_id and set
4231 * up for use. If the subsystem is built-in anyway, work is delegated to the
4232 * simpler cgroup_init_subsys.
4234 int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4237 struct cgroup_subsys_state *css;
4239 /* check name and function validity */
4240 if (ss->name == NULL || strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN ||
4241 ss->create == NULL || ss->destroy == NULL)
4245 * we don't support callbacks in modular subsystems. this check is
4246 * before the ss->module check for consistency; a subsystem that could
4247 * be a module should still have no callbacks even if the user isn't
4248 * compiling it as one.
4250 if (ss->fork || ss->exit)
4254 * an optionally modular subsystem is built-in: we want to do nothing,
4255 * since cgroup_init_subsys will have already taken care of it.
4257 if (ss->module == NULL) {
4258 /* a few sanity checks */
4259 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4260 BUG_ON(subsys[ss->subsys_id] != ss);
4264 /* init base cftset */
4265 cgroup_init_cftsets(ss);
4268 * need to register a subsys id before anything else - for example,
4269 * init_cgroup_css needs it.
4271 mutex_lock(&cgroup_mutex);
4272 /* find the first empty slot in the array */
4273 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4274 if (subsys[i] == NULL)
4277 if (i == CGROUP_SUBSYS_COUNT) {
4278 /* maximum number of subsystems already registered! */
4279 mutex_unlock(&cgroup_mutex);
4282 /* assign ourselves the subsys_id */
4287 * no ss->create seems to need anything important in the ss struct, so
4288 * this can happen first (i.e. before the rootnode attachment).
4290 css = ss->create(dummytop);
4292 /* failure case - need to deassign the subsys[] slot. */
4294 mutex_unlock(&cgroup_mutex);
4295 return PTR_ERR(css);
4298 list_add(&ss->sibling, &rootnode.subsys_list);
4299 ss->root = &rootnode;
4301 /* our new subsystem will be attached to the dummy hierarchy. */
4302 init_cgroup_css(css, ss, dummytop);
4303 /* init_idr must be after init_cgroup_css because it sets css->id. */
4305 int ret = cgroup_init_idr(ss, css);
4307 dummytop->subsys[ss->subsys_id] = NULL;
4308 ss->destroy(dummytop);
4310 mutex_unlock(&cgroup_mutex);
4316 * Now we need to entangle the css into the existing css_sets. unlike
4317 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4318 * will need a new pointer to it; done by iterating the css_set_table.
4319 * furthermore, modifying the existing css_sets will corrupt the hash
4320 * table state, so each changed css_set will need its hash recomputed.
4321 * this is all done under the css_set_lock.
4323 write_lock(&css_set_lock);
4324 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
4326 struct hlist_node *node, *tmp;
4327 struct hlist_head *bucket = &css_set_table[i], *new_bucket;
4329 hlist_for_each_entry_safe(cg, node, tmp, bucket, hlist) {
4330 /* skip entries that we already rehashed */
4331 if (cg->subsys[ss->subsys_id])
4333 /* remove existing entry */
4334 hlist_del(&cg->hlist);
4336 cg->subsys[ss->subsys_id] = css;
4337 /* recompute hash and restore entry */
4338 new_bucket = css_set_hash(cg->subsys);
4339 hlist_add_head(&cg->hlist, new_bucket);
4342 write_unlock(&css_set_lock);
4344 mutex_init(&ss->hierarchy_mutex);
4345 lockdep_set_class(&ss->hierarchy_mutex, &ss->subsys_key);
4349 mutex_unlock(&cgroup_mutex);
4352 EXPORT_SYMBOL_GPL(cgroup_load_subsys);
4355 * cgroup_unload_subsys: unload a modular subsystem
4356 * @ss: the subsystem to unload
4358 * This function should be called in a modular subsystem's exitcall. When this
4359 * function is invoked, the refcount on the subsystem's module will be 0, so
4360 * the subsystem will not be attached to any hierarchy.
4362 void cgroup_unload_subsys(struct cgroup_subsys *ss)
4364 struct cg_cgroup_link *link;
4365 struct hlist_head *hhead;
4367 BUG_ON(ss->module == NULL);
4370 * we shouldn't be called if the subsystem is in use, and the use of
4371 * try_module_get in parse_cgroupfs_options should ensure that it
4372 * doesn't start being used while we're killing it off.
4374 BUG_ON(ss->root != &rootnode);
4376 mutex_lock(&cgroup_mutex);
4377 /* deassign the subsys_id */
4378 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4379 subsys[ss->subsys_id] = NULL;
4381 /* remove subsystem from rootnode's list of subsystems */
4382 list_del_init(&ss->sibling);
4385 * disentangle the css from all css_sets attached to the dummytop. as
4386 * in loading, we need to pay our respects to the hashtable gods.
4388 write_lock(&css_set_lock);
4389 list_for_each_entry(link, &dummytop->css_sets, cgrp_link_list) {
4390 struct css_set *cg = link->cg;
4392 hlist_del(&cg->hlist);
4393 BUG_ON(!cg->subsys[ss->subsys_id]);
4394 cg->subsys[ss->subsys_id] = NULL;
4395 hhead = css_set_hash(cg->subsys);
4396 hlist_add_head(&cg->hlist, hhead);
4398 write_unlock(&css_set_lock);
4401 * remove subsystem's css from the dummytop and free it - need to free
4402 * before marking as null because ss->destroy needs the cgrp->subsys
4403 * pointer to find their state. note that this also takes care of
4404 * freeing the css_id.
4406 ss->destroy(dummytop);
4407 dummytop->subsys[ss->subsys_id] = NULL;
4409 mutex_unlock(&cgroup_mutex);
4411 EXPORT_SYMBOL_GPL(cgroup_unload_subsys);
4414 * cgroup_init_early - cgroup initialization at system boot
4416 * Initialize cgroups at system boot, and initialize any
4417 * subsystems that request early init.
4419 int __init cgroup_init_early(void)
4422 atomic_set(&init_css_set.refcount, 1);
4423 INIT_LIST_HEAD(&init_css_set.cg_links);
4424 INIT_LIST_HEAD(&init_css_set.tasks);
4425 INIT_HLIST_NODE(&init_css_set.hlist);
4427 init_cgroup_root(&rootnode);
4429 init_task.cgroups = &init_css_set;
4431 init_css_set_link.cg = &init_css_set;
4432 init_css_set_link.cgrp = dummytop;
4433 list_add(&init_css_set_link.cgrp_link_list,
4434 &rootnode.top_cgroup.css_sets);
4435 list_add(&init_css_set_link.cg_link_list,
4436 &init_css_set.cg_links);
4438 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4439 INIT_HLIST_HEAD(&css_set_table[i]);
4441 /* at bootup time, we don't worry about modular subsystems */
4442 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4443 struct cgroup_subsys *ss = subsys[i];
4446 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4447 BUG_ON(!ss->create);
4448 BUG_ON(!ss->destroy);
4449 if (ss->subsys_id != i) {
4450 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
4451 ss->name, ss->subsys_id);
4456 cgroup_init_subsys(ss);
4462 * cgroup_init - cgroup initialization
4464 * Register cgroup filesystem and /proc file, and initialize
4465 * any subsystems that didn't request early init.
4467 int __init cgroup_init(void)
4471 struct hlist_head *hhead;
4473 err = bdi_init(&cgroup_backing_dev_info);
4477 /* at bootup time, we don't worry about modular subsystems */
4478 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4479 struct cgroup_subsys *ss = subsys[i];
4480 if (!ss->early_init)
4481 cgroup_init_subsys(ss);
4483 cgroup_init_idr(ss, init_css_set.subsys[ss->subsys_id]);
4486 /* Add init_css_set to the hash table */
4487 hhead = css_set_hash(init_css_set.subsys);
4488 hlist_add_head(&init_css_set.hlist, hhead);
4489 BUG_ON(!init_root_id(&rootnode));
4491 cgroup_kobj = kobject_create_and_add("cgroup", fs_kobj);
4497 err = register_filesystem(&cgroup_fs_type);
4499 kobject_put(cgroup_kobj);
4503 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
4507 bdi_destroy(&cgroup_backing_dev_info);
4513 * proc_cgroup_show()
4514 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4515 * - Used for /proc/<pid>/cgroup.
4516 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4517 * doesn't really matter if tsk->cgroup changes after we read it,
4518 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4519 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4520 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4521 * cgroup to top_cgroup.
4524 /* TODO: Use a proper seq_file iterator */
4525 static int proc_cgroup_show(struct seq_file *m, void *v)
4528 struct task_struct *tsk;
4531 struct cgroupfs_root *root;
4534 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4540 tsk = get_pid_task(pid, PIDTYPE_PID);
4546 mutex_lock(&cgroup_mutex);
4548 for_each_active_root(root) {
4549 struct cgroup_subsys *ss;
4550 struct cgroup *cgrp;
4553 seq_printf(m, "%d:", root->hierarchy_id);
4554 for_each_subsys(root, ss)
4555 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
4556 if (strlen(root->name))
4557 seq_printf(m, "%sname=%s", count ? "," : "",
4560 cgrp = task_cgroup_from_root(tsk, root);
4561 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
4569 mutex_unlock(&cgroup_mutex);
4570 put_task_struct(tsk);
4577 static int cgroup_open(struct inode *inode, struct file *file)
4579 struct pid *pid = PROC_I(inode)->pid;
4580 return single_open(file, proc_cgroup_show, pid);
4583 const struct file_operations proc_cgroup_operations = {
4584 .open = cgroup_open,
4586 .llseek = seq_lseek,
4587 .release = single_release,
4590 /* Display information about each subsystem and each hierarchy */
4591 static int proc_cgroupstats_show(struct seq_file *m, void *v)
4595 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4597 * ideally we don't want subsystems moving around while we do this.
4598 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4599 * subsys/hierarchy state.
4601 mutex_lock(&cgroup_mutex);
4602 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4603 struct cgroup_subsys *ss = subsys[i];
4606 seq_printf(m, "%s\t%d\t%d\t%d\n",
4607 ss->name, ss->root->hierarchy_id,
4608 ss->root->number_of_cgroups, !ss->disabled);
4610 mutex_unlock(&cgroup_mutex);
4614 static int cgroupstats_open(struct inode *inode, struct file *file)
4616 return single_open(file, proc_cgroupstats_show, NULL);
4619 static const struct file_operations proc_cgroupstats_operations = {
4620 .open = cgroupstats_open,
4622 .llseek = seq_lseek,
4623 .release = single_release,
4627 * cgroup_fork - attach newly forked task to its parents cgroup.
4628 * @child: pointer to task_struct of forking parent process.
4630 * Description: A task inherits its parent's cgroup at fork().
4632 * A pointer to the shared css_set was automatically copied in
4633 * fork.c by dup_task_struct(). However, we ignore that copy, since
4634 * it was not made under the protection of RCU, cgroup_mutex or
4635 * threadgroup_change_begin(), so it might no longer be a valid
4636 * cgroup pointer. cgroup_attach_task() might have already changed
4637 * current->cgroups, allowing the previously referenced cgroup
4638 * group to be removed and freed.
4640 * Outside the pointer validity we also need to process the css_set
4641 * inheritance between threadgoup_change_begin() and
4642 * threadgoup_change_end(), this way there is no leak in any process
4643 * wide migration performed by cgroup_attach_proc() that could otherwise
4644 * miss a thread because it is too early or too late in the fork stage.
4646 * At the point that cgroup_fork() is called, 'current' is the parent
4647 * task, and the passed argument 'child' points to the child task.
4649 void cgroup_fork(struct task_struct *child)
4652 * We don't need to task_lock() current because current->cgroups
4653 * can't be changed concurrently here. The parent obviously hasn't
4654 * exited and called cgroup_exit(), and we are synchronized against
4655 * cgroup migration through threadgroup_change_begin().
4657 child->cgroups = current->cgroups;
4658 get_css_set(child->cgroups);
4659 INIT_LIST_HEAD(&child->cg_list);
4663 * cgroup_fork_callbacks - run fork callbacks
4664 * @child: the new task
4666 * Called on a new task very soon before adding it to the
4667 * tasklist. No need to take any locks since no-one can
4668 * be operating on this task.
4670 void cgroup_fork_callbacks(struct task_struct *child)
4672 if (need_forkexit_callback) {
4675 * forkexit callbacks are only supported for builtin
4676 * subsystems, and the builtin section of the subsys array is
4677 * immutable, so we don't need to lock the subsys array here.
4679 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4680 struct cgroup_subsys *ss = subsys[i];
4688 * cgroup_post_fork - called on a new task after adding it to the task list
4689 * @child: the task in question
4691 * Adds the task to the list running through its css_set if necessary.
4692 * Has to be after the task is visible on the task list in case we race
4693 * with the first call to cgroup_iter_start() - to guarantee that the
4694 * new task ends up on its list.
4696 void cgroup_post_fork(struct task_struct *child)
4699 * use_task_css_set_links is set to 1 before we walk the tasklist
4700 * under the tasklist_lock and we read it here after we added the child
4701 * to the tasklist under the tasklist_lock as well. If the child wasn't
4702 * yet in the tasklist when we walked through it from
4703 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4704 * should be visible now due to the paired locking and barriers implied
4705 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4706 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4709 if (use_task_css_set_links) {
4710 write_lock(&css_set_lock);
4711 if (list_empty(&child->cg_list)) {
4713 * It's safe to use child->cgroups without task_lock()
4714 * here because we are protected through
4715 * threadgroup_change_begin() against concurrent
4716 * css_set change in cgroup_task_migrate(). Also
4717 * the task can't exit at that point until
4718 * wake_up_new_task() is called, so we are protected
4719 * against cgroup_exit() setting child->cgroup to
4722 list_add(&child->cg_list, &child->cgroups->tasks);
4724 write_unlock(&css_set_lock);
4728 * cgroup_exit - detach cgroup from exiting task
4729 * @tsk: pointer to task_struct of exiting process
4730 * @run_callback: run exit callbacks?
4732 * Description: Detach cgroup from @tsk and release it.
4734 * Note that cgroups marked notify_on_release force every task in
4735 * them to take the global cgroup_mutex mutex when exiting.
4736 * This could impact scaling on very large systems. Be reluctant to
4737 * use notify_on_release cgroups where very high task exit scaling
4738 * is required on large systems.
4740 * the_top_cgroup_hack:
4742 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4744 * We call cgroup_exit() while the task is still competent to
4745 * handle notify_on_release(), then leave the task attached to the
4746 * root cgroup in each hierarchy for the remainder of its exit.
4748 * To do this properly, we would increment the reference count on
4749 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4750 * code we would add a second cgroup function call, to drop that
4751 * reference. This would just create an unnecessary hot spot on
4752 * the top_cgroup reference count, to no avail.
4754 * Normally, holding a reference to a cgroup without bumping its
4755 * count is unsafe. The cgroup could go away, or someone could
4756 * attach us to a different cgroup, decrementing the count on
4757 * the first cgroup that we never incremented. But in this case,
4758 * top_cgroup isn't going away, and either task has PF_EXITING set,
4759 * which wards off any cgroup_attach_task() attempts, or task is a failed
4760 * fork, never visible to cgroup_attach_task.
4762 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4768 * Unlink from the css_set task list if necessary.
4769 * Optimistically check cg_list before taking
4772 if (!list_empty(&tsk->cg_list)) {
4773 write_lock(&css_set_lock);
4774 if (!list_empty(&tsk->cg_list))
4775 list_del_init(&tsk->cg_list);
4776 write_unlock(&css_set_lock);
4779 /* Reassign the task to the init_css_set. */
4782 tsk->cgroups = &init_css_set;
4784 if (run_callbacks && need_forkexit_callback) {
4786 * modular subsystems can't use callbacks, so no need to lock
4789 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4790 struct cgroup_subsys *ss = subsys[i];
4792 struct cgroup *old_cgrp =
4793 rcu_dereference_raw(cg->subsys[i])->cgroup;
4794 struct cgroup *cgrp = task_cgroup(tsk, i);
4795 ss->exit(cgrp, old_cgrp, tsk);
4802 put_css_set_taskexit(cg);
4806 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4807 * @cgrp: the cgroup in question
4808 * @task: the task in question
4810 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4813 * If we are sending in dummytop, then presumably we are creating
4814 * the top cgroup in the subsystem.
4816 * Called only by the ns (nsproxy) cgroup.
4818 int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
4821 struct cgroup *target;
4823 if (cgrp == dummytop)
4826 target = task_cgroup_from_root(task, cgrp->root);
4827 while (cgrp != target && cgrp!= cgrp->top_cgroup)
4828 cgrp = cgrp->parent;
4829 ret = (cgrp == target);
4833 static void check_for_release(struct cgroup *cgrp)
4835 /* All of these checks rely on RCU to keep the cgroup
4836 * structure alive */
4837 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
4838 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
4839 /* Control Group is currently removeable. If it's not
4840 * already queued for a userspace notification, queue
4842 int need_schedule_work = 0;
4843 raw_spin_lock(&release_list_lock);
4844 if (!cgroup_is_removed(cgrp) &&
4845 list_empty(&cgrp->release_list)) {
4846 list_add(&cgrp->release_list, &release_list);
4847 need_schedule_work = 1;
4849 raw_spin_unlock(&release_list_lock);
4850 if (need_schedule_work)
4851 schedule_work(&release_agent_work);
4855 /* Caller must verify that the css is not for root cgroup */
4856 void __css_put(struct cgroup_subsys_state *css, int count)
4858 struct cgroup *cgrp = css->cgroup;
4861 val = atomic_sub_return(count, &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 WARN_ON_ONCE(val < 1);
4872 EXPORT_SYMBOL_GPL(__css_put);
4875 * Notify userspace when a cgroup is released, by running the
4876 * configured release agent with the name of the cgroup (path
4877 * relative to the root of cgroup file system) as the argument.
4879 * Most likely, this user command will try to rmdir this cgroup.
4881 * This races with the possibility that some other task will be
4882 * attached to this cgroup before it is removed, or that some other
4883 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4884 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4885 * unused, and this cgroup will be reprieved from its death sentence,
4886 * to continue to serve a useful existence. Next time it's released,
4887 * we will get notified again, if it still has 'notify_on_release' set.
4889 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4890 * means only wait until the task is successfully execve()'d. The
4891 * separate release agent task is forked by call_usermodehelper(),
4892 * then control in this thread returns here, without waiting for the
4893 * release agent task. We don't bother to wait because the caller of
4894 * this routine has no use for the exit status of the release agent
4895 * task, so no sense holding our caller up for that.
4897 static void cgroup_release_agent(struct work_struct *work)
4899 BUG_ON(work != &release_agent_work);
4900 mutex_lock(&cgroup_mutex);
4901 raw_spin_lock(&release_list_lock);
4902 while (!list_empty(&release_list)) {
4903 char *argv[3], *envp[3];
4905 char *pathbuf = NULL, *agentbuf = NULL;
4906 struct cgroup *cgrp = list_entry(release_list.next,
4909 list_del_init(&cgrp->release_list);
4910 raw_spin_unlock(&release_list_lock);
4911 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
4914 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0)
4916 agentbuf = kstrdup(cgrp->root->release_agent_path, GFP_KERNEL);
4921 argv[i++] = agentbuf;
4922 argv[i++] = pathbuf;
4926 /* minimal command environment */
4927 envp[i++] = "HOME=/";
4928 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4931 /* Drop the lock while we invoke the usermode helper,
4932 * since the exec could involve hitting disk and hence
4933 * be a slow process */
4934 mutex_unlock(&cgroup_mutex);
4935 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
4936 mutex_lock(&cgroup_mutex);
4940 raw_spin_lock(&release_list_lock);
4942 raw_spin_unlock(&release_list_lock);
4943 mutex_unlock(&cgroup_mutex);
4946 static int __init cgroup_disable(char *str)
4951 while ((token = strsep(&str, ",")) != NULL) {
4955 * cgroup_disable, being at boot time, can't know about module
4956 * subsystems, so we don't worry about them.
4958 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4959 struct cgroup_subsys *ss = subsys[i];
4961 if (!strcmp(token, ss->name)) {
4963 printk(KERN_INFO "Disabling %s control group"
4964 " subsystem\n", ss->name);
4971 __setup("cgroup_disable=", cgroup_disable);
4974 * Functons for CSS ID.
4978 *To get ID other than 0, this should be called when !cgroup_is_removed().
4980 unsigned short css_id(struct cgroup_subsys_state *css)
4982 struct css_id *cssid;
4985 * This css_id() can return correct value when somone has refcnt
4986 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4987 * it's unchanged until freed.
4989 cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt));
4995 EXPORT_SYMBOL_GPL(css_id);
4997 unsigned short css_depth(struct cgroup_subsys_state *css)
4999 struct css_id *cssid;
5001 cssid = rcu_dereference_check(css->id, atomic_read(&css->refcnt));
5004 return cssid->depth;
5007 EXPORT_SYMBOL_GPL(css_depth);
5010 * css_is_ancestor - test "root" css is an ancestor of "child"
5011 * @child: the css to be tested.
5012 * @root: the css supporsed to be an ancestor of the child.
5014 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5015 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
5016 * But, considering usual usage, the csses should be valid objects after test.
5017 * Assuming that the caller will do some action to the child if this returns
5018 * returns true, the caller must take "child";s reference count.
5019 * If "child" is valid object and this returns true, "root" is valid, too.
5022 bool css_is_ancestor(struct cgroup_subsys_state *child,
5023 const struct cgroup_subsys_state *root)
5025 struct css_id *child_id;
5026 struct css_id *root_id;
5030 child_id = rcu_dereference(child->id);
5031 root_id = rcu_dereference(root->id);
5034 || (child_id->depth < root_id->depth)
5035 || (child_id->stack[root_id->depth] != root_id->id))
5041 void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
5043 struct css_id *id = css->id;
5044 /* When this is called before css_id initialization, id can be NULL */
5048 BUG_ON(!ss->use_id);
5050 rcu_assign_pointer(id->css, NULL);
5051 rcu_assign_pointer(css->id, NULL);
5052 spin_lock(&ss->id_lock);
5053 idr_remove(&ss->idr, id->id);
5054 spin_unlock(&ss->id_lock);
5055 kfree_rcu(id, rcu_head);
5057 EXPORT_SYMBOL_GPL(free_css_id);
5060 * This is called by init or create(). Then, calls to this function are
5061 * always serialized (By cgroup_mutex() at create()).
5064 static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
5066 struct css_id *newid;
5067 int myid, error, size;
5069 BUG_ON(!ss->use_id);
5071 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
5072 newid = kzalloc(size, GFP_KERNEL);
5074 return ERR_PTR(-ENOMEM);
5076 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
5080 spin_lock(&ss->id_lock);
5081 /* Don't use 0. allocates an ID of 1-65535 */
5082 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
5083 spin_unlock(&ss->id_lock);
5085 /* Returns error when there are no free spaces for new ID.*/
5090 if (myid > CSS_ID_MAX)
5094 newid->depth = depth;
5098 spin_lock(&ss->id_lock);
5099 idr_remove(&ss->idr, myid);
5100 spin_unlock(&ss->id_lock);
5103 return ERR_PTR(error);
5107 static int __init_or_module cgroup_init_idr(struct cgroup_subsys *ss,
5108 struct cgroup_subsys_state *rootcss)
5110 struct css_id *newid;
5112 spin_lock_init(&ss->id_lock);
5115 newid = get_new_cssid(ss, 0);
5117 return PTR_ERR(newid);
5119 newid->stack[0] = newid->id;
5120 newid->css = rootcss;
5121 rootcss->id = newid;
5125 static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
5126 struct cgroup *child)
5128 int subsys_id, i, depth = 0;
5129 struct cgroup_subsys_state *parent_css, *child_css;
5130 struct css_id *child_id, *parent_id;
5132 subsys_id = ss->subsys_id;
5133 parent_css = parent->subsys[subsys_id];
5134 child_css = child->subsys[subsys_id];
5135 parent_id = parent_css->id;
5136 depth = parent_id->depth + 1;
5138 child_id = get_new_cssid(ss, depth);
5139 if (IS_ERR(child_id))
5140 return PTR_ERR(child_id);
5142 for (i = 0; i < depth; i++)
5143 child_id->stack[i] = parent_id->stack[i];
5144 child_id->stack[depth] = child_id->id;
5146 * child_id->css pointer will be set after this cgroup is available
5147 * see cgroup_populate_dir()
5149 rcu_assign_pointer(child_css->id, child_id);
5155 * css_lookup - lookup css by id
5156 * @ss: cgroup subsys to be looked into.
5159 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5160 * NULL if not. Should be called under rcu_read_lock()
5162 struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
5164 struct css_id *cssid = NULL;
5166 BUG_ON(!ss->use_id);
5167 cssid = idr_find(&ss->idr, id);
5169 if (unlikely(!cssid))
5172 return rcu_dereference(cssid->css);
5174 EXPORT_SYMBOL_GPL(css_lookup);
5177 * css_get_next - lookup next cgroup under specified hierarchy.
5178 * @ss: pointer to subsystem
5179 * @id: current position of iteration.
5180 * @root: pointer to css. search tree under this.
5181 * @foundid: position of found object.
5183 * Search next css under the specified hierarchy of rootid. Calling under
5184 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5186 struct cgroup_subsys_state *
5187 css_get_next(struct cgroup_subsys *ss, int id,
5188 struct cgroup_subsys_state *root, int *foundid)
5190 struct cgroup_subsys_state *ret = NULL;
5193 int rootid = css_id(root);
5194 int depth = css_depth(root);
5199 BUG_ON(!ss->use_id);
5200 WARN_ON_ONCE(!rcu_read_lock_held());
5202 /* fill start point for scan */
5206 * scan next entry from bitmap(tree), tmpid is updated after
5209 tmp = idr_get_next(&ss->idr, &tmpid);
5212 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
5213 ret = rcu_dereference(tmp->css);
5219 /* continue to scan from next id */
5226 * get corresponding css from file open on cgroupfs directory
5228 struct cgroup_subsys_state *cgroup_css_from_dir(struct file *f, int id)
5230 struct cgroup *cgrp;
5231 struct inode *inode;
5232 struct cgroup_subsys_state *css;
5234 inode = f->f_dentry->d_inode;
5235 /* check in cgroup filesystem dir */
5236 if (inode->i_op != &cgroup_dir_inode_operations)
5237 return ERR_PTR(-EBADF);
5239 if (id < 0 || id >= CGROUP_SUBSYS_COUNT)
5240 return ERR_PTR(-EINVAL);
5243 cgrp = __d_cgrp(f->f_dentry);
5244 css = cgrp->subsys[id];
5245 return css ? css : ERR_PTR(-ENOENT);
5248 #ifdef CONFIG_CGROUP_DEBUG
5249 static struct cgroup_subsys_state *debug_create(struct cgroup *cont)
5251 struct cgroup_subsys_state *css = kzalloc(sizeof(*css), GFP_KERNEL);
5254 return ERR_PTR(-ENOMEM);
5259 static void debug_destroy(struct cgroup *cont)
5261 kfree(cont->subsys[debug_subsys_id]);
5264 static u64 cgroup_refcount_read(struct cgroup *cont, struct cftype *cft)
5266 return atomic_read(&cont->count);
5269 static u64 debug_taskcount_read(struct cgroup *cont, struct cftype *cft)
5271 return cgroup_task_count(cont);
5274 static u64 current_css_set_read(struct cgroup *cont, struct cftype *cft)
5276 return (u64)(unsigned long)current->cgroups;
5279 static u64 current_css_set_refcount_read(struct cgroup *cont,
5285 count = atomic_read(¤t->cgroups->refcount);
5290 static int current_css_set_cg_links_read(struct cgroup *cont,
5292 struct seq_file *seq)
5294 struct cg_cgroup_link *link;
5297 read_lock(&css_set_lock);
5299 cg = rcu_dereference(current->cgroups);
5300 list_for_each_entry(link, &cg->cg_links, cg_link_list) {
5301 struct cgroup *c = link->cgrp;
5305 name = c->dentry->d_name.name;
5308 seq_printf(seq, "Root %d group %s\n",
5309 c->root->hierarchy_id, name);
5312 read_unlock(&css_set_lock);
5316 #define MAX_TASKS_SHOWN_PER_CSS 25
5317 static int cgroup_css_links_read(struct cgroup *cont,
5319 struct seq_file *seq)
5321 struct cg_cgroup_link *link;
5323 read_lock(&css_set_lock);
5324 list_for_each_entry(link, &cont->css_sets, cgrp_link_list) {
5325 struct css_set *cg = link->cg;
5326 struct task_struct *task;
5328 seq_printf(seq, "css_set %p\n", cg);
5329 list_for_each_entry(task, &cg->tasks, cg_list) {
5330 if (count++ > MAX_TASKS_SHOWN_PER_CSS) {
5331 seq_puts(seq, " ...\n");
5334 seq_printf(seq, " task %d\n",
5335 task_pid_vnr(task));
5339 read_unlock(&css_set_lock);
5343 static u64 releasable_read(struct cgroup *cgrp, struct cftype *cft)
5345 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
5348 static struct cftype debug_files[] = {
5350 .name = "cgroup_refcount",
5351 .read_u64 = cgroup_refcount_read,
5354 .name = "taskcount",
5355 .read_u64 = debug_taskcount_read,
5359 .name = "current_css_set",
5360 .read_u64 = current_css_set_read,
5364 .name = "current_css_set_refcount",
5365 .read_u64 = current_css_set_refcount_read,
5369 .name = "current_css_set_cg_links",
5370 .read_seq_string = current_css_set_cg_links_read,
5374 .name = "cgroup_css_links",
5375 .read_seq_string = cgroup_css_links_read,
5379 .name = "releasable",
5380 .read_u64 = releasable_read,
5386 struct cgroup_subsys debug_subsys = {
5388 .create = debug_create,
5389 .destroy = debug_destroy,
5390 .subsys_id = debug_subsys_id,
5391 .base_cftypes = debug_files,
5393 #endif /* CONFIG_CGROUP_DEBUG */