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 /* Hierarchy-specific flags */
133 /* The path to use for release notifications. */
134 char release_agent_path
[PATH_MAX
];
136 /* The name for this hierarchy - may be empty */
137 char name
[MAX_CGROUP_ROOT_NAMELEN
];
141 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
142 * subsystems that are otherwise unattached - it never has more than a
143 * single cgroup, and all tasks are part of that cgroup.
145 static struct cgroupfs_root rootnode
;
148 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
149 * cgroup_subsys->use_id != 0.
151 #define CSS_ID_MAX (65535)
154 * The css to which this ID points. This pointer is set to valid value
155 * after cgroup is populated. If cgroup is removed, this will be NULL.
156 * This pointer is expected to be RCU-safe because destroy()
157 * is called after synchronize_rcu(). But for safe use, css_is_removed()
158 * css_tryget() should be used for avoiding race.
160 struct cgroup_subsys_state __rcu
*css
;
166 * Depth in hierarchy which this ID belongs to.
168 unsigned short depth
;
170 * ID is freed by RCU. (and lookup routine is RCU safe.)
172 struct rcu_head rcu_head
;
174 * Hierarchy of CSS ID belongs to.
176 unsigned short stack
[0]; /* Array of Length (depth+1) */
180 * cgroup_event represents events which userspace want to receive.
182 struct cgroup_event
{
184 * Cgroup which the event belongs to.
188 * Control file which the event associated.
192 * eventfd to signal userspace about the event.
194 struct eventfd_ctx
*eventfd
;
196 * Each of these stored in a list by the cgroup.
198 struct list_head list
;
200 * All fields below needed to unregister event when
201 * userspace closes eventfd.
204 wait_queue_head_t
*wqh
;
206 struct work_struct remove
;
209 /* The list of hierarchy roots */
211 static LIST_HEAD(roots
);
212 static int root_count
;
214 static DEFINE_IDA(hierarchy_ida
);
215 static int next_hierarchy_id
;
216 static DEFINE_SPINLOCK(hierarchy_id_lock
);
218 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
219 #define dummytop (&rootnode.top_cgroup)
221 /* This flag indicates whether tasks in the fork and exit paths should
222 * check for fork/exit handlers to call. This avoids us having to do
223 * extra work in the fork/exit path if none of the subsystems need to
226 static int need_forkexit_callback __read_mostly
;
228 #ifdef CONFIG_PROVE_LOCKING
229 int cgroup_lock_is_held(void)
231 return lockdep_is_held(&cgroup_mutex
);
233 #else /* #ifdef CONFIG_PROVE_LOCKING */
234 int cgroup_lock_is_held(void)
236 return mutex_is_locked(&cgroup_mutex
);
238 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
240 EXPORT_SYMBOL_GPL(cgroup_lock_is_held
);
242 /* convenient tests for these bits */
243 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
245 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
248 /* bits in struct cgroupfs_root flags field */
250 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
253 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
256 (1 << CGRP_RELEASABLE
) |
257 (1 << CGRP_NOTIFY_ON_RELEASE
);
258 return (cgrp
->flags
& bits
) == bits
;
261 static int notify_on_release(const struct cgroup
*cgrp
)
263 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
266 static int clone_children(const struct cgroup
*cgrp
)
268 return test_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
272 * for_each_subsys() allows you to iterate on each subsystem attached to
273 * an active hierarchy
275 #define for_each_subsys(_root, _ss) \
276 list_for_each_entry(_ss, &_root->subsys_list, sibling)
278 /* for_each_active_root() allows you to iterate across the active hierarchies */
279 #define for_each_active_root(_root) \
280 list_for_each_entry(_root, &roots, root_list)
282 /* the list of cgroups eligible for automatic release. Protected by
283 * release_list_lock */
284 static LIST_HEAD(release_list
);
285 static DEFINE_RAW_SPINLOCK(release_list_lock
);
286 static void cgroup_release_agent(struct work_struct
*work
);
287 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
288 static void check_for_release(struct cgroup
*cgrp
);
290 /* Link structure for associating css_set objects with cgroups */
291 struct cg_cgroup_link
{
293 * List running through cg_cgroup_links associated with a
294 * cgroup, anchored on cgroup->css_sets
296 struct list_head cgrp_link_list
;
299 * List running through cg_cgroup_links pointing at a
300 * single css_set object, anchored on css_set->cg_links
302 struct list_head cg_link_list
;
306 /* The default css_set - used by init and its children prior to any
307 * hierarchies being mounted. It contains a pointer to the root state
308 * for each subsystem. Also used to anchor the list of css_sets. Not
309 * reference-counted, to improve performance when child cgroups
310 * haven't been created.
313 static struct css_set init_css_set
;
314 static struct cg_cgroup_link init_css_set_link
;
316 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
317 struct cgroup_subsys_state
*css
);
319 /* css_set_lock protects the list of css_set objects, and the
320 * chain of tasks off each css_set. Nests outside task->alloc_lock
321 * due to cgroup_iter_start() */
322 static DEFINE_RWLOCK(css_set_lock
);
323 static int css_set_count
;
326 * hash table for cgroup groups. This improves the performance to find
327 * an existing css_set. This hash doesn't (currently) take into
328 * account cgroups in empty hierarchies.
330 #define CSS_SET_HASH_BITS 7
331 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
332 static struct hlist_head css_set_table
[CSS_SET_TABLE_SIZE
];
334 static struct hlist_head
*css_set_hash(struct cgroup_subsys_state
*css
[])
338 unsigned long tmp
= 0UL;
340 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
341 tmp
+= (unsigned long)css
[i
];
342 tmp
= (tmp
>> 16) ^ tmp
;
344 index
= hash_long(tmp
, CSS_SET_HASH_BITS
);
346 return &css_set_table
[index
];
349 /* We don't maintain the lists running through each css_set to its
350 * task until after the first call to cgroup_iter_start(). This
351 * reduces the fork()/exit() overhead for people who have cgroups
352 * compiled into their kernel but not actually in use */
353 static int use_task_css_set_links __read_mostly
;
355 static void __put_css_set(struct css_set
*cg
, int taskexit
)
357 struct cg_cgroup_link
*link
;
358 struct cg_cgroup_link
*saved_link
;
360 * Ensure that the refcount doesn't hit zero while any readers
361 * can see it. Similar to atomic_dec_and_lock(), but for an
364 if (atomic_add_unless(&cg
->refcount
, -1, 1))
366 write_lock(&css_set_lock
);
367 if (!atomic_dec_and_test(&cg
->refcount
)) {
368 write_unlock(&css_set_lock
);
372 /* This css_set is dead. unlink it and release cgroup refcounts */
373 hlist_del(&cg
->hlist
);
376 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
378 struct cgroup
*cgrp
= link
->cgrp
;
379 list_del(&link
->cg_link_list
);
380 list_del(&link
->cgrp_link_list
);
381 if (atomic_dec_and_test(&cgrp
->count
) &&
382 notify_on_release(cgrp
)) {
384 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
385 check_for_release(cgrp
);
391 write_unlock(&css_set_lock
);
392 kfree_rcu(cg
, rcu_head
);
396 * refcounted get/put for css_set objects
398 static inline void get_css_set(struct css_set
*cg
)
400 atomic_inc(&cg
->refcount
);
403 static inline void put_css_set(struct css_set
*cg
)
405 __put_css_set(cg
, 0);
408 static inline void put_css_set_taskexit(struct css_set
*cg
)
410 __put_css_set(cg
, 1);
414 * compare_css_sets - helper function for find_existing_css_set().
415 * @cg: candidate css_set being tested
416 * @old_cg: existing css_set for a task
417 * @new_cgrp: cgroup that's being entered by the task
418 * @template: desired set of css pointers in css_set (pre-calculated)
420 * Returns true if "cg" matches "old_cg" except for the hierarchy
421 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
423 static bool compare_css_sets(struct css_set
*cg
,
424 struct css_set
*old_cg
,
425 struct cgroup
*new_cgrp
,
426 struct cgroup_subsys_state
*template[])
428 struct list_head
*l1
, *l2
;
430 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
431 /* Not all subsystems matched */
436 * Compare cgroup pointers in order to distinguish between
437 * different cgroups in heirarchies with no subsystems. We
438 * could get by with just this check alone (and skip the
439 * memcmp above) but on most setups the memcmp check will
440 * avoid the need for this more expensive check on almost all
445 l2
= &old_cg
->cg_links
;
447 struct cg_cgroup_link
*cgl1
, *cgl2
;
448 struct cgroup
*cg1
, *cg2
;
452 /* See if we reached the end - both lists are equal length. */
453 if (l1
== &cg
->cg_links
) {
454 BUG_ON(l2
!= &old_cg
->cg_links
);
457 BUG_ON(l2
== &old_cg
->cg_links
);
459 /* Locate the cgroups associated with these links. */
460 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
461 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
464 /* Hierarchies should be linked in the same order. */
465 BUG_ON(cg1
->root
!= cg2
->root
);
468 * If this hierarchy is the hierarchy of the cgroup
469 * that's changing, then we need to check that this
470 * css_set points to the new cgroup; if it's any other
471 * hierarchy, then this css_set should point to the
472 * same cgroup as the old css_set.
474 if (cg1
->root
== new_cgrp
->root
) {
486 * find_existing_css_set() is a helper for
487 * find_css_set(), and checks to see whether an existing
488 * css_set is suitable.
490 * oldcg: the cgroup group that we're using before the cgroup
493 * cgrp: the cgroup that we're moving into
495 * template: location in which to build the desired set of subsystem
496 * state objects for the new cgroup group
498 static struct css_set
*find_existing_css_set(
499 struct css_set
*oldcg
,
501 struct cgroup_subsys_state
*template[])
504 struct cgroupfs_root
*root
= cgrp
->root
;
505 struct hlist_head
*hhead
;
506 struct hlist_node
*node
;
510 * Build the set of subsystem state objects that we want to see in the
511 * new css_set. while subsystems can change globally, the entries here
512 * won't change, so no need for locking.
514 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
515 if (root
->subsys_bits
& (1UL << i
)) {
516 /* Subsystem is in this hierarchy. So we want
517 * the subsystem state from the new
519 template[i
] = cgrp
->subsys
[i
];
521 /* Subsystem is not in this hierarchy, so we
522 * don't want to change the subsystem state */
523 template[i
] = oldcg
->subsys
[i
];
527 hhead
= css_set_hash(template);
528 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
529 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
532 /* This css_set matches what we need */
536 /* No existing cgroup group matched */
540 static void free_cg_links(struct list_head
*tmp
)
542 struct cg_cgroup_link
*link
;
543 struct cg_cgroup_link
*saved_link
;
545 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
546 list_del(&link
->cgrp_link_list
);
552 * allocate_cg_links() allocates "count" cg_cgroup_link structures
553 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
554 * success or a negative error
556 static int allocate_cg_links(int count
, struct list_head
*tmp
)
558 struct cg_cgroup_link
*link
;
561 for (i
= 0; i
< count
; i
++) {
562 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
567 list_add(&link
->cgrp_link_list
, tmp
);
573 * link_css_set - a helper function to link a css_set to a cgroup
574 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
575 * @cg: the css_set to be linked
576 * @cgrp: the destination cgroup
578 static void link_css_set(struct list_head
*tmp_cg_links
,
579 struct css_set
*cg
, struct cgroup
*cgrp
)
581 struct cg_cgroup_link
*link
;
583 BUG_ON(list_empty(tmp_cg_links
));
584 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
588 atomic_inc(&cgrp
->count
);
589 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
591 * Always add links to the tail of the list so that the list
592 * is sorted by order of hierarchy creation
594 list_add_tail(&link
->cg_link_list
, &cg
->cg_links
);
598 * find_css_set() takes an existing cgroup group and a
599 * cgroup object, and returns a css_set object that's
600 * equivalent to the old group, but with the given cgroup
601 * substituted into the appropriate hierarchy. Must be called with
604 static struct css_set
*find_css_set(
605 struct css_set
*oldcg
, struct cgroup
*cgrp
)
608 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
610 struct list_head tmp_cg_links
;
612 struct hlist_head
*hhead
;
613 struct cg_cgroup_link
*link
;
615 /* First see if we already have a cgroup group that matches
617 read_lock(&css_set_lock
);
618 res
= find_existing_css_set(oldcg
, cgrp
, template);
621 read_unlock(&css_set_lock
);
626 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
630 /* Allocate all the cg_cgroup_link objects that we'll need */
631 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
636 atomic_set(&res
->refcount
, 1);
637 INIT_LIST_HEAD(&res
->cg_links
);
638 INIT_LIST_HEAD(&res
->tasks
);
639 INIT_HLIST_NODE(&res
->hlist
);
641 /* Copy the set of subsystem state objects generated in
642 * find_existing_css_set() */
643 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
645 write_lock(&css_set_lock
);
646 /* Add reference counts and links from the new css_set. */
647 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
648 struct cgroup
*c
= link
->cgrp
;
649 if (c
->root
== cgrp
->root
)
651 link_css_set(&tmp_cg_links
, res
, c
);
654 BUG_ON(!list_empty(&tmp_cg_links
));
658 /* Add this cgroup group to the hash table */
659 hhead
= css_set_hash(res
->subsys
);
660 hlist_add_head(&res
->hlist
, hhead
);
662 write_unlock(&css_set_lock
);
668 * Return the cgroup for "task" from the given hierarchy. Must be
669 * called with cgroup_mutex held.
671 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
672 struct cgroupfs_root
*root
)
675 struct cgroup
*res
= NULL
;
677 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
678 read_lock(&css_set_lock
);
680 * No need to lock the task - since we hold cgroup_mutex the
681 * task can't change groups, so the only thing that can happen
682 * is that it exits and its css is set back to init_css_set.
685 if (css
== &init_css_set
) {
686 res
= &root
->top_cgroup
;
688 struct cg_cgroup_link
*link
;
689 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
690 struct cgroup
*c
= link
->cgrp
;
691 if (c
->root
== root
) {
697 read_unlock(&css_set_lock
);
703 * There is one global cgroup mutex. We also require taking
704 * task_lock() when dereferencing a task's cgroup subsys pointers.
705 * See "The task_lock() exception", at the end of this comment.
707 * A task must hold cgroup_mutex to modify cgroups.
709 * Any task can increment and decrement the count field without lock.
710 * So in general, code holding cgroup_mutex can't rely on the count
711 * field not changing. However, if the count goes to zero, then only
712 * cgroup_attach_task() can increment it again. Because a count of zero
713 * means that no tasks are currently attached, therefore there is no
714 * way a task attached to that cgroup can fork (the other way to
715 * increment the count). So code holding cgroup_mutex can safely
716 * assume that if the count is zero, it will stay zero. Similarly, if
717 * a task holds cgroup_mutex on a cgroup with zero count, it
718 * knows that the cgroup won't be removed, as cgroup_rmdir()
721 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
722 * (usually) take cgroup_mutex. These are the two most performance
723 * critical pieces of code here. The exception occurs on cgroup_exit(),
724 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
725 * is taken, and if the cgroup count is zero, a usermode call made
726 * to the release agent with the name of the cgroup (path relative to
727 * the root of cgroup file system) as the argument.
729 * A cgroup can only be deleted if both its 'count' of using tasks
730 * is zero, and its list of 'children' cgroups is empty. Since all
731 * tasks in the system use _some_ cgroup, and since there is always at
732 * least one task in the system (init, pid == 1), therefore, top_cgroup
733 * always has either children cgroups and/or using tasks. So we don't
734 * need a special hack to ensure that top_cgroup cannot be deleted.
736 * The task_lock() exception
738 * The need for this exception arises from the action of
739 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
740 * another. It does so using cgroup_mutex, however there are
741 * several performance critical places that need to reference
742 * task->cgroup without the expense of grabbing a system global
743 * mutex. Therefore except as noted below, when dereferencing or, as
744 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
745 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
746 * the task_struct routinely used for such matters.
748 * P.S. One more locking exception. RCU is used to guard the
749 * update of a tasks cgroup pointer by cgroup_attach_task()
753 * cgroup_lock - lock out any changes to cgroup structures
756 void cgroup_lock(void)
758 mutex_lock(&cgroup_mutex
);
760 EXPORT_SYMBOL_GPL(cgroup_lock
);
763 * cgroup_unlock - release lock on cgroup changes
765 * Undo the lock taken in a previous cgroup_lock() call.
767 void cgroup_unlock(void)
769 mutex_unlock(&cgroup_mutex
);
771 EXPORT_SYMBOL_GPL(cgroup_unlock
);
774 * A couple of forward declarations required, due to cyclic reference loop:
775 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
776 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
780 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
);
781 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
782 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
783 static int cgroup_populate_dir(struct cgroup
*cgrp
);
784 static const struct inode_operations cgroup_dir_inode_operations
;
785 static const struct file_operations proc_cgroupstats_operations
;
787 static struct backing_dev_info cgroup_backing_dev_info
= {
789 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
792 static int alloc_css_id(struct cgroup_subsys
*ss
,
793 struct cgroup
*parent
, struct cgroup
*child
);
795 static struct inode
*cgroup_new_inode(mode_t mode
, struct super_block
*sb
)
797 struct inode
*inode
= new_inode(sb
);
800 inode
->i_ino
= get_next_ino();
801 inode
->i_mode
= mode
;
802 inode
->i_uid
= current_fsuid();
803 inode
->i_gid
= current_fsgid();
804 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
805 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
811 * Call subsys's pre_destroy handler.
812 * This is called before css refcnt check.
814 static int cgroup_call_pre_destroy(struct cgroup
*cgrp
)
816 struct cgroup_subsys
*ss
;
819 for_each_subsys(cgrp
->root
, ss
)
820 if (ss
->pre_destroy
) {
821 ret
= ss
->pre_destroy(ss
, cgrp
);
829 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
831 /* is dentry a directory ? if so, kfree() associated cgroup */
832 if (S_ISDIR(inode
->i_mode
)) {
833 struct cgroup
*cgrp
= dentry
->d_fsdata
;
834 struct cgroup_subsys
*ss
;
835 BUG_ON(!(cgroup_is_removed(cgrp
)));
836 /* It's possible for external users to be holding css
837 * reference counts on a cgroup; css_put() needs to
838 * be able to access the cgroup after decrementing
839 * the reference count in order to know if it needs to
840 * queue the cgroup to be handled by the release
844 mutex_lock(&cgroup_mutex
);
846 * Release the subsystem state objects.
848 for_each_subsys(cgrp
->root
, ss
)
849 ss
->destroy(ss
, cgrp
);
851 cgrp
->root
->number_of_cgroups
--;
852 mutex_unlock(&cgroup_mutex
);
855 * Drop the active superblock reference that we took when we
858 deactivate_super(cgrp
->root
->sb
);
861 * if we're getting rid of the cgroup, refcount should ensure
862 * that there are no pidlists left.
864 BUG_ON(!list_empty(&cgrp
->pidlists
));
866 kfree_rcu(cgrp
, rcu_head
);
871 static int cgroup_delete(const struct dentry
*d
)
876 static void remove_dir(struct dentry
*d
)
878 struct dentry
*parent
= dget(d
->d_parent
);
881 simple_rmdir(parent
->d_inode
, d
);
885 static void cgroup_clear_directory(struct dentry
*dentry
)
887 struct list_head
*node
;
889 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
890 spin_lock(&dentry
->d_lock
);
891 node
= dentry
->d_subdirs
.next
;
892 while (node
!= &dentry
->d_subdirs
) {
893 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
895 spin_lock_nested(&d
->d_lock
, DENTRY_D_LOCK_NESTED
);
898 /* This should never be called on a cgroup
899 * directory with child cgroups */
900 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
902 spin_unlock(&d
->d_lock
);
903 spin_unlock(&dentry
->d_lock
);
905 simple_unlink(dentry
->d_inode
, d
);
907 spin_lock(&dentry
->d_lock
);
909 spin_unlock(&d
->d_lock
);
910 node
= dentry
->d_subdirs
.next
;
912 spin_unlock(&dentry
->d_lock
);
916 * NOTE : the dentry must have been dget()'ed
918 static void cgroup_d_remove_dir(struct dentry
*dentry
)
920 struct dentry
*parent
;
922 cgroup_clear_directory(dentry
);
924 parent
= dentry
->d_parent
;
925 spin_lock(&parent
->d_lock
);
926 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
927 list_del_init(&dentry
->d_u
.d_child
);
928 spin_unlock(&dentry
->d_lock
);
929 spin_unlock(&parent
->d_lock
);
934 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
935 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
936 * reference to css->refcnt. In general, this refcnt is expected to goes down
939 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
941 DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq
);
943 static void cgroup_wakeup_rmdir_waiter(struct cgroup
*cgrp
)
945 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
)))
946 wake_up_all(&cgroup_rmdir_waitq
);
949 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
)
954 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
)
956 cgroup_wakeup_rmdir_waiter(css
->cgroup
);
961 * Call with cgroup_mutex held. Drops reference counts on modules, including
962 * any duplicate ones that parse_cgroupfs_options took. If this function
963 * returns an error, no reference counts are touched.
965 static int rebind_subsystems(struct cgroupfs_root
*root
,
966 unsigned long final_bits
)
968 unsigned long added_bits
, removed_bits
;
969 struct cgroup
*cgrp
= &root
->top_cgroup
;
972 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
973 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
975 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
976 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
977 /* Check that any added subsystems are currently free */
978 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
979 unsigned long bit
= 1UL << i
;
980 struct cgroup_subsys
*ss
= subsys
[i
];
981 if (!(bit
& added_bits
))
984 * Nobody should tell us to do a subsys that doesn't exist:
985 * parse_cgroupfs_options should catch that case and refcounts
986 * ensure that subsystems won't disappear once selected.
989 if (ss
->root
!= &rootnode
) {
990 /* Subsystem isn't free */
995 /* Currently we don't handle adding/removing subsystems when
996 * any child cgroups exist. This is theoretically supportable
997 * but involves complex error handling, so it's being left until
999 if (root
->number_of_cgroups
> 1)
1002 /* Process each subsystem */
1003 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1004 struct cgroup_subsys
*ss
= subsys
[i
];
1005 unsigned long bit
= 1UL << i
;
1006 if (bit
& added_bits
) {
1007 /* We're binding this subsystem to this hierarchy */
1009 BUG_ON(cgrp
->subsys
[i
]);
1010 BUG_ON(!dummytop
->subsys
[i
]);
1011 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
1012 mutex_lock(&ss
->hierarchy_mutex
);
1013 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
1014 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1015 list_move(&ss
->sibling
, &root
->subsys_list
);
1019 mutex_unlock(&ss
->hierarchy_mutex
);
1020 /* refcount was already taken, and we're keeping it */
1021 } else if (bit
& removed_bits
) {
1022 /* We're removing this subsystem */
1024 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1025 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1026 mutex_lock(&ss
->hierarchy_mutex
);
1028 ss
->bind(ss
, dummytop
);
1029 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1030 cgrp
->subsys
[i
] = NULL
;
1031 subsys
[i
]->root
= &rootnode
;
1032 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1033 mutex_unlock(&ss
->hierarchy_mutex
);
1034 /* subsystem is now free - drop reference on module */
1035 module_put(ss
->module
);
1036 } else if (bit
& final_bits
) {
1037 /* Subsystem state should already exist */
1039 BUG_ON(!cgrp
->subsys
[i
]);
1041 * a refcount was taken, but we already had one, so
1042 * drop the extra reference.
1044 module_put(ss
->module
);
1045 #ifdef CONFIG_MODULE_UNLOAD
1046 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1049 /* Subsystem state shouldn't exist */
1050 BUG_ON(cgrp
->subsys
[i
]);
1053 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
1059 static int cgroup_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
1061 struct cgroupfs_root
*root
= vfs
->mnt_sb
->s_fs_info
;
1062 struct cgroup_subsys
*ss
;
1064 mutex_lock(&cgroup_root_mutex
);
1065 for_each_subsys(root
, ss
)
1066 seq_printf(seq
, ",%s", ss
->name
);
1067 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1068 seq_puts(seq
, ",noprefix");
1069 if (strlen(root
->release_agent_path
))
1070 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1071 if (clone_children(&root
->top_cgroup
))
1072 seq_puts(seq
, ",clone_children");
1073 if (strlen(root
->name
))
1074 seq_printf(seq
, ",name=%s", root
->name
);
1075 mutex_unlock(&cgroup_root_mutex
);
1079 struct cgroup_sb_opts
{
1080 unsigned long subsys_bits
;
1081 unsigned long flags
;
1082 char *release_agent
;
1083 bool clone_children
;
1085 /* User explicitly requested empty subsystem */
1088 struct cgroupfs_root
*new_root
;
1093 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1094 * with cgroup_mutex held to protect the subsys[] array. This function takes
1095 * refcounts on subsystems to be used, unless it returns error, in which case
1096 * no refcounts are taken.
1098 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1100 char *token
, *o
= data
;
1101 bool all_ss
= false, one_ss
= false;
1102 unsigned long mask
= (unsigned long)-1;
1104 bool module_pin_failed
= false;
1106 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1108 #ifdef CONFIG_CPUSETS
1109 mask
= ~(1UL << cpuset_subsys_id
);
1112 memset(opts
, 0, sizeof(*opts
));
1114 while ((token
= strsep(&o
, ",")) != NULL
) {
1117 if (!strcmp(token
, "none")) {
1118 /* Explicitly have no subsystems */
1122 if (!strcmp(token
, "all")) {
1123 /* Mutually exclusive option 'all' + subsystem name */
1129 if (!strcmp(token
, "noprefix")) {
1130 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1133 if (!strcmp(token
, "clone_children")) {
1134 opts
->clone_children
= true;
1137 if (!strncmp(token
, "release_agent=", 14)) {
1138 /* Specifying two release agents is forbidden */
1139 if (opts
->release_agent
)
1141 opts
->release_agent
=
1142 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1143 if (!opts
->release_agent
)
1147 if (!strncmp(token
, "name=", 5)) {
1148 const char *name
= token
+ 5;
1149 /* Can't specify an empty name */
1152 /* Must match [\w.-]+ */
1153 for (i
= 0; i
< strlen(name
); i
++) {
1157 if ((c
== '.') || (c
== '-') || (c
== '_'))
1161 /* Specifying two names is forbidden */
1164 opts
->name
= kstrndup(name
,
1165 MAX_CGROUP_ROOT_NAMELEN
- 1,
1173 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1174 struct cgroup_subsys
*ss
= subsys
[i
];
1177 if (strcmp(token
, ss
->name
))
1182 /* Mutually exclusive option 'all' + subsystem name */
1185 set_bit(i
, &opts
->subsys_bits
);
1190 if (i
== CGROUP_SUBSYS_COUNT
)
1195 * If the 'all' option was specified select all the subsystems,
1196 * otherwise 'all, 'none' and a subsystem name options were not
1197 * specified, let's default to 'all'
1199 if (all_ss
|| (!all_ss
&& !one_ss
&& !opts
->none
)) {
1200 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1201 struct cgroup_subsys
*ss
= subsys
[i
];
1206 set_bit(i
, &opts
->subsys_bits
);
1210 /* Consistency checks */
1213 * Option noprefix was introduced just for backward compatibility
1214 * with the old cpuset, so we allow noprefix only if mounting just
1215 * the cpuset subsystem.
1217 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1218 (opts
->subsys_bits
& mask
))
1222 /* Can't specify "none" and some subsystems */
1223 if (opts
->subsys_bits
&& opts
->none
)
1227 * We either have to specify by name or by subsystems. (So all
1228 * empty hierarchies must have a name).
1230 if (!opts
->subsys_bits
&& !opts
->name
)
1234 * Grab references on all the modules we'll need, so the subsystems
1235 * don't dance around before rebind_subsystems attaches them. This may
1236 * take duplicate reference counts on a subsystem that's already used,
1237 * but rebind_subsystems handles this case.
1239 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1240 unsigned long bit
= 1UL << i
;
1242 if (!(bit
& opts
->subsys_bits
))
1244 if (!try_module_get(subsys
[i
]->module
)) {
1245 module_pin_failed
= true;
1249 if (module_pin_failed
) {
1251 * oops, one of the modules was going away. this means that we
1252 * raced with a module_delete call, and to the user this is
1253 * essentially a "subsystem doesn't exist" case.
1255 for (i
--; i
>= CGROUP_BUILTIN_SUBSYS_COUNT
; i
--) {
1256 /* drop refcounts only on the ones we took */
1257 unsigned long bit
= 1UL << i
;
1259 if (!(bit
& opts
->subsys_bits
))
1261 module_put(subsys
[i
]->module
);
1269 static void drop_parsed_module_refcounts(unsigned long subsys_bits
)
1272 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1273 unsigned long bit
= 1UL << i
;
1275 if (!(bit
& subsys_bits
))
1277 module_put(subsys
[i
]->module
);
1281 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1284 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1285 struct cgroup
*cgrp
= &root
->top_cgroup
;
1286 struct cgroup_sb_opts opts
;
1288 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1289 mutex_lock(&cgroup_mutex
);
1290 mutex_lock(&cgroup_root_mutex
);
1292 /* See what subsystems are wanted */
1293 ret
= parse_cgroupfs_options(data
, &opts
);
1297 /* Don't allow flags or name to change at remount */
1298 if (opts
.flags
!= root
->flags
||
1299 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1301 drop_parsed_module_refcounts(opts
.subsys_bits
);
1305 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
1307 drop_parsed_module_refcounts(opts
.subsys_bits
);
1311 /* (re)populate subsystem files */
1312 cgroup_populate_dir(cgrp
);
1314 if (opts
.release_agent
)
1315 strcpy(root
->release_agent_path
, opts
.release_agent
);
1317 kfree(opts
.release_agent
);
1319 mutex_unlock(&cgroup_root_mutex
);
1320 mutex_unlock(&cgroup_mutex
);
1321 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1325 static const struct super_operations cgroup_ops
= {
1326 .statfs
= simple_statfs
,
1327 .drop_inode
= generic_delete_inode
,
1328 .show_options
= cgroup_show_options
,
1329 .remount_fs
= cgroup_remount
,
1332 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1334 INIT_LIST_HEAD(&cgrp
->sibling
);
1335 INIT_LIST_HEAD(&cgrp
->children
);
1336 INIT_LIST_HEAD(&cgrp
->css_sets
);
1337 INIT_LIST_HEAD(&cgrp
->release_list
);
1338 INIT_LIST_HEAD(&cgrp
->pidlists
);
1339 mutex_init(&cgrp
->pidlist_mutex
);
1340 INIT_LIST_HEAD(&cgrp
->event_list
);
1341 spin_lock_init(&cgrp
->event_list_lock
);
1344 static void init_cgroup_root(struct cgroupfs_root
*root
)
1346 struct cgroup
*cgrp
= &root
->top_cgroup
;
1347 INIT_LIST_HEAD(&root
->subsys_list
);
1348 INIT_LIST_HEAD(&root
->root_list
);
1349 root
->number_of_cgroups
= 1;
1351 cgrp
->top_cgroup
= cgrp
;
1352 init_cgroup_housekeeping(cgrp
);
1355 static bool init_root_id(struct cgroupfs_root
*root
)
1360 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1362 spin_lock(&hierarchy_id_lock
);
1363 /* Try to allocate the next unused ID */
1364 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1365 &root
->hierarchy_id
);
1367 /* Try again starting from 0 */
1368 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1370 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1371 } else if (ret
!= -EAGAIN
) {
1372 /* Can only get here if the 31-bit IDR is full ... */
1375 spin_unlock(&hierarchy_id_lock
);
1380 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1382 struct cgroup_sb_opts
*opts
= data
;
1383 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1385 /* If we asked for a name then it must match */
1386 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1390 * If we asked for subsystems (or explicitly for no
1391 * subsystems) then they must match
1393 if ((opts
->subsys_bits
|| opts
->none
)
1394 && (opts
->subsys_bits
!= root
->subsys_bits
))
1400 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1402 struct cgroupfs_root
*root
;
1404 if (!opts
->subsys_bits
&& !opts
->none
)
1407 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1409 return ERR_PTR(-ENOMEM
);
1411 if (!init_root_id(root
)) {
1413 return ERR_PTR(-ENOMEM
);
1415 init_cgroup_root(root
);
1417 root
->subsys_bits
= opts
->subsys_bits
;
1418 root
->flags
= opts
->flags
;
1419 if (opts
->release_agent
)
1420 strcpy(root
->release_agent_path
, opts
->release_agent
);
1422 strcpy(root
->name
, opts
->name
);
1423 if (opts
->clone_children
)
1424 set_bit(CGRP_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1428 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1433 BUG_ON(!root
->hierarchy_id
);
1434 spin_lock(&hierarchy_id_lock
);
1435 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1436 spin_unlock(&hierarchy_id_lock
);
1440 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1443 struct cgroup_sb_opts
*opts
= data
;
1445 /* If we don't have a new root, we can't set up a new sb */
1446 if (!opts
->new_root
)
1449 BUG_ON(!opts
->subsys_bits
&& !opts
->none
);
1451 ret
= set_anon_super(sb
, NULL
);
1455 sb
->s_fs_info
= opts
->new_root
;
1456 opts
->new_root
->sb
= sb
;
1458 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1459 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1460 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1461 sb
->s_op
= &cgroup_ops
;
1466 static int cgroup_get_rootdir(struct super_block
*sb
)
1468 static const struct dentry_operations cgroup_dops
= {
1469 .d_iput
= cgroup_diput
,
1470 .d_delete
= cgroup_delete
,
1473 struct inode
*inode
=
1474 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1475 struct dentry
*dentry
;
1480 inode
->i_fop
= &simple_dir_operations
;
1481 inode
->i_op
= &cgroup_dir_inode_operations
;
1482 /* directories start off with i_nlink == 2 (for "." entry) */
1484 dentry
= d_alloc_root(inode
);
1489 sb
->s_root
= dentry
;
1490 /* for everything else we want ->d_op set */
1491 sb
->s_d_op
= &cgroup_dops
;
1495 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1496 int flags
, const char *unused_dev_name
,
1499 struct cgroup_sb_opts opts
;
1500 struct cgroupfs_root
*root
;
1502 struct super_block
*sb
;
1503 struct cgroupfs_root
*new_root
;
1504 struct inode
*inode
;
1506 /* First find the desired set of subsystems */
1507 mutex_lock(&cgroup_mutex
);
1508 ret
= parse_cgroupfs_options(data
, &opts
);
1509 mutex_unlock(&cgroup_mutex
);
1514 * Allocate a new cgroup root. We may not need it if we're
1515 * reusing an existing hierarchy.
1517 new_root
= cgroup_root_from_opts(&opts
);
1518 if (IS_ERR(new_root
)) {
1519 ret
= PTR_ERR(new_root
);
1522 opts
.new_root
= new_root
;
1524 /* Locate an existing or new sb for this hierarchy */
1525 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, &opts
);
1528 cgroup_drop_root(opts
.new_root
);
1532 root
= sb
->s_fs_info
;
1534 if (root
== opts
.new_root
) {
1535 /* We used the new root structure, so this is a new hierarchy */
1536 struct list_head tmp_cg_links
;
1537 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1538 struct cgroupfs_root
*existing_root
;
1539 const struct cred
*cred
;
1542 BUG_ON(sb
->s_root
!= NULL
);
1544 ret
= cgroup_get_rootdir(sb
);
1546 goto drop_new_super
;
1547 inode
= sb
->s_root
->d_inode
;
1549 mutex_lock(&inode
->i_mutex
);
1550 mutex_lock(&cgroup_mutex
);
1551 mutex_lock(&cgroup_root_mutex
);
1553 /* Check for name clashes with existing mounts */
1555 if (strlen(root
->name
))
1556 for_each_active_root(existing_root
)
1557 if (!strcmp(existing_root
->name
, root
->name
))
1561 * We're accessing css_set_count without locking
1562 * css_set_lock here, but that's OK - it can only be
1563 * increased by someone holding cgroup_lock, and
1564 * that's us. The worst that can happen is that we
1565 * have some link structures left over
1567 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1571 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1572 if (ret
== -EBUSY
) {
1573 free_cg_links(&tmp_cg_links
);
1577 * There must be no failure case after here, since rebinding
1578 * takes care of subsystems' refcounts, which are explicitly
1579 * dropped in the failure exit path.
1582 /* EBUSY should be the only error here */
1585 list_add(&root
->root_list
, &roots
);
1588 sb
->s_root
->d_fsdata
= root_cgrp
;
1589 root
->top_cgroup
.dentry
= sb
->s_root
;
1591 /* Link the top cgroup in this hierarchy into all
1592 * the css_set objects */
1593 write_lock(&css_set_lock
);
1594 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1595 struct hlist_head
*hhead
= &css_set_table
[i
];
1596 struct hlist_node
*node
;
1599 hlist_for_each_entry(cg
, node
, hhead
, hlist
)
1600 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1602 write_unlock(&css_set_lock
);
1604 free_cg_links(&tmp_cg_links
);
1606 BUG_ON(!list_empty(&root_cgrp
->sibling
));
1607 BUG_ON(!list_empty(&root_cgrp
->children
));
1608 BUG_ON(root
->number_of_cgroups
!= 1);
1610 cred
= override_creds(&init_cred
);
1611 cgroup_populate_dir(root_cgrp
);
1613 mutex_unlock(&cgroup_root_mutex
);
1614 mutex_unlock(&cgroup_mutex
);
1615 mutex_unlock(&inode
->i_mutex
);
1618 * We re-used an existing hierarchy - the new root (if
1619 * any) is not needed
1621 cgroup_drop_root(opts
.new_root
);
1622 /* no subsys rebinding, so refcounts don't change */
1623 drop_parsed_module_refcounts(opts
.subsys_bits
);
1626 kfree(opts
.release_agent
);
1628 return dget(sb
->s_root
);
1631 mutex_unlock(&cgroup_root_mutex
);
1632 mutex_unlock(&cgroup_mutex
);
1633 mutex_unlock(&inode
->i_mutex
);
1635 deactivate_locked_super(sb
);
1637 drop_parsed_module_refcounts(opts
.subsys_bits
);
1639 kfree(opts
.release_agent
);
1641 return ERR_PTR(ret
);
1644 static void cgroup_kill_sb(struct super_block
*sb
) {
1645 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1646 struct cgroup
*cgrp
= &root
->top_cgroup
;
1648 struct cg_cgroup_link
*link
;
1649 struct cg_cgroup_link
*saved_link
;
1653 BUG_ON(root
->number_of_cgroups
!= 1);
1654 BUG_ON(!list_empty(&cgrp
->children
));
1655 BUG_ON(!list_empty(&cgrp
->sibling
));
1657 mutex_lock(&cgroup_mutex
);
1658 mutex_lock(&cgroup_root_mutex
);
1660 /* Rebind all subsystems back to the default hierarchy */
1661 ret
= rebind_subsystems(root
, 0);
1662 /* Shouldn't be able to fail ... */
1666 * Release all the links from css_sets to this hierarchy's
1669 write_lock(&css_set_lock
);
1671 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1673 list_del(&link
->cg_link_list
);
1674 list_del(&link
->cgrp_link_list
);
1677 write_unlock(&css_set_lock
);
1679 if (!list_empty(&root
->root_list
)) {
1680 list_del(&root
->root_list
);
1684 mutex_unlock(&cgroup_root_mutex
);
1685 mutex_unlock(&cgroup_mutex
);
1687 kill_litter_super(sb
);
1688 cgroup_drop_root(root
);
1691 static struct file_system_type cgroup_fs_type
= {
1693 .mount
= cgroup_mount
,
1694 .kill_sb
= cgroup_kill_sb
,
1697 static struct kobject
*cgroup_kobj
;
1699 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1701 return dentry
->d_fsdata
;
1704 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1706 return dentry
->d_fsdata
;
1710 * cgroup_path - generate the path of a cgroup
1711 * @cgrp: the cgroup in question
1712 * @buf: the buffer to write the path into
1713 * @buflen: the length of the buffer
1715 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1716 * reference. Writes path of cgroup into buf. Returns 0 on success,
1719 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1722 struct dentry
*dentry
= rcu_dereference_check(cgrp
->dentry
,
1723 cgroup_lock_is_held());
1725 if (!dentry
|| cgrp
== dummytop
) {
1727 * Inactive subsystems have no dentry for their root
1734 start
= buf
+ buflen
;
1738 int len
= dentry
->d_name
.len
;
1740 if ((start
-= len
) < buf
)
1741 return -ENAMETOOLONG
;
1742 memcpy(start
, dentry
->d_name
.name
, len
);
1743 cgrp
= cgrp
->parent
;
1747 dentry
= rcu_dereference_check(cgrp
->dentry
,
1748 cgroup_lock_is_held());
1752 return -ENAMETOOLONG
;
1755 memmove(buf
, start
, buf
+ buflen
- start
);
1758 EXPORT_SYMBOL_GPL(cgroup_path
);
1761 * Control Group taskset
1763 struct task_and_cgroup
{
1764 struct task_struct
*task
;
1765 struct cgroup
*cgrp
;
1768 struct cgroup_taskset
{
1769 struct task_and_cgroup single
;
1770 struct flex_array
*tc_array
;
1773 struct cgroup
*cur_cgrp
;
1777 * cgroup_taskset_first - reset taskset and return the first task
1778 * @tset: taskset of interest
1780 * @tset iteration is initialized and the first task is returned.
1782 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1784 if (tset
->tc_array
) {
1786 return cgroup_taskset_next(tset
);
1788 tset
->cur_cgrp
= tset
->single
.cgrp
;
1789 return tset
->single
.task
;
1792 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1795 * cgroup_taskset_next - iterate to the next task in taskset
1796 * @tset: taskset of interest
1798 * Return the next task in @tset. Iteration must have been initialized
1799 * with cgroup_taskset_first().
1801 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1803 struct task_and_cgroup
*tc
;
1805 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1808 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1809 tset
->cur_cgrp
= tc
->cgrp
;
1812 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1815 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1816 * @tset: taskset of interest
1818 * Return the cgroup for the current (last returned) task of @tset. This
1819 * function must be preceded by either cgroup_taskset_first() or
1820 * cgroup_taskset_next().
1822 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1824 return tset
->cur_cgrp
;
1826 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1829 * cgroup_taskset_size - return the number of tasks in taskset
1830 * @tset: taskset of interest
1832 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1834 return tset
->tc_array
? tset
->tc_array_len
: 1;
1836 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1840 * cgroup_task_migrate - move a task from one cgroup to another.
1842 * 'guarantee' is set if the caller promises that a new css_set for the task
1843 * will already exist. If not set, this function might sleep, and can fail with
1844 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1846 static int cgroup_task_migrate(struct cgroup
*cgrp
, struct cgroup
*oldcgrp
,
1847 struct task_struct
*tsk
, bool guarantee
)
1849 struct css_set
*oldcg
;
1850 struct css_set
*newcg
;
1853 * get old css_set. We are synchronized through threadgroup_lock()
1854 * against PF_EXITING setting such that we can't race against
1855 * cgroup_exit() changing the css_set to init_css_set and dropping the
1858 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1859 oldcg
= tsk
->cgroups
;
1862 /* locate or allocate a new css_set for this task. */
1864 /* we know the css_set we want already exists. */
1865 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
1866 read_lock(&css_set_lock
);
1867 newcg
= find_existing_css_set(oldcg
, cgrp
, template);
1870 read_unlock(&css_set_lock
);
1873 /* find_css_set will give us newcg already referenced. */
1874 newcg
= find_css_set(oldcg
, cgrp
);
1883 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1886 /* Update the css_set linked lists if we're using them */
1887 write_lock(&css_set_lock
);
1888 if (!list_empty(&tsk
->cg_list
))
1889 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1890 write_unlock(&css_set_lock
);
1893 * We just gained a reference on oldcg by taking it from the task. As
1894 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1895 * it here; it will be freed under RCU.
1899 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1904 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1905 * @cgrp: the cgroup the task is attaching to
1906 * @tsk: the task to be attached
1908 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1911 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1914 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1915 struct cgroup
*oldcgrp
;
1916 struct cgroupfs_root
*root
= cgrp
->root
;
1917 struct cgroup_taskset tset
= { };
1919 /* @tsk either already exited or can't exit until the end */
1920 if (tsk
->flags
& PF_EXITING
)
1923 /* Nothing to do if the task is already in that cgroup */
1924 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1925 if (cgrp
== oldcgrp
)
1928 tset
.single
.task
= tsk
;
1929 tset
.single
.cgrp
= oldcgrp
;
1931 for_each_subsys(root
, ss
) {
1932 if (ss
->can_attach
) {
1933 retval
= ss
->can_attach(ss
, cgrp
, &tset
);
1936 * Remember on which subsystem the can_attach()
1937 * failed, so that we only call cancel_attach()
1938 * against the subsystems whose can_attach()
1939 * succeeded. (See below)
1947 retval
= cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, false);
1951 for_each_subsys(root
, ss
) {
1953 ss
->attach(ss
, cgrp
, &tset
);
1959 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1960 * is no longer empty.
1962 cgroup_wakeup_rmdir_waiter(cgrp
);
1965 for_each_subsys(root
, ss
) {
1966 if (ss
== failed_ss
)
1968 * This subsystem was the one that failed the
1969 * can_attach() check earlier, so we don't need
1970 * to call cancel_attach() against it or any
1971 * remaining subsystems.
1974 if (ss
->cancel_attach
)
1975 ss
->cancel_attach(ss
, cgrp
, &tset
);
1982 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1983 * @from: attach to all cgroups of a given task
1984 * @tsk: the task to be attached
1986 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
1988 struct cgroupfs_root
*root
;
1992 for_each_active_root(root
) {
1993 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
1995 retval
= cgroup_attach_task(from_cg
, tsk
);
2003 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2006 * cgroup_attach_proc works in two stages, the first of which prefetches all
2007 * new css_sets needed (to make sure we have enough memory before committing
2008 * to the move) and stores them in a list of entries of the following type.
2009 * TODO: possible optimization: use css_set->rcu_head for chaining instead
2011 struct cg_list_entry
{
2013 struct list_head links
;
2016 static bool css_set_check_fetched(struct cgroup
*cgrp
,
2017 struct task_struct
*tsk
, struct css_set
*cg
,
2018 struct list_head
*newcg_list
)
2020 struct css_set
*newcg
;
2021 struct cg_list_entry
*cg_entry
;
2022 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
2024 read_lock(&css_set_lock
);
2025 newcg
= find_existing_css_set(cg
, cgrp
, template);
2026 read_unlock(&css_set_lock
);
2028 /* doesn't exist at all? */
2031 /* see if it's already in the list */
2032 list_for_each_entry(cg_entry
, newcg_list
, links
)
2033 if (cg_entry
->cg
== newcg
)
2041 * Find the new css_set and store it in the list in preparation for moving the
2042 * given task to the given cgroup. Returns 0 or -ENOMEM.
2044 static int css_set_prefetch(struct cgroup
*cgrp
, struct css_set
*cg
,
2045 struct list_head
*newcg_list
)
2047 struct css_set
*newcg
;
2048 struct cg_list_entry
*cg_entry
;
2050 /* ensure a new css_set will exist for this thread */
2051 newcg
= find_css_set(cg
, cgrp
);
2054 /* add it to the list */
2055 cg_entry
= kmalloc(sizeof(struct cg_list_entry
), GFP_KERNEL
);
2060 cg_entry
->cg
= newcg
;
2061 list_add(&cg_entry
->links
, newcg_list
);
2066 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2067 * @cgrp: the cgroup to attach to
2068 * @leader: the threadgroup leader task_struct of the group to be attached
2070 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2071 * task_lock of each thread in leader's threadgroup individually in turn.
2073 int cgroup_attach_proc(struct cgroup
*cgrp
, struct task_struct
*leader
)
2075 int retval
, i
, group_size
, nr_migrating_tasks
;
2076 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
2077 /* guaranteed to be initialized later, but the compiler needs this */
2078 struct css_set
*oldcg
;
2079 struct cgroupfs_root
*root
= cgrp
->root
;
2080 /* threadgroup list cursor and array */
2081 struct task_struct
*tsk
;
2082 struct task_and_cgroup
*tc
;
2083 struct flex_array
*group
;
2084 struct cgroup_taskset tset
= { };
2086 * we need to make sure we have css_sets for all the tasks we're
2087 * going to move -before- we actually start moving them, so that in
2088 * case we get an ENOMEM we can bail out before making any changes.
2090 struct list_head newcg_list
;
2091 struct cg_list_entry
*cg_entry
, *temp_nobe
;
2094 * step 0: in order to do expensive, possibly blocking operations for
2095 * every thread, we cannot iterate the thread group list, since it needs
2096 * rcu or tasklist locked. instead, build an array of all threads in the
2097 * group - group_rwsem prevents new threads from appearing, and if
2098 * threads exit, this will just be an over-estimate.
2100 group_size
= get_nr_threads(leader
);
2101 /* flex_array supports very large thread-groups better than kmalloc. */
2102 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2105 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2106 retval
= flex_array_prealloc(group
, 0, group_size
- 1, GFP_KERNEL
);
2108 goto out_free_group_list
;
2110 /* prevent changes to the threadgroup list while we take a snapshot. */
2111 read_lock(&tasklist_lock
);
2112 if (!thread_group_leader(leader
)) {
2114 * a race with de_thread from another thread's exec() may strip
2115 * us of our leadership, making while_each_thread unsafe to use
2116 * on this task. if this happens, there is no choice but to
2117 * throw this task away and try again (from cgroup_procs_write);
2118 * this is "double-double-toil-and-trouble-check locking".
2120 read_unlock(&tasklist_lock
);
2122 goto out_free_group_list
;
2124 /* take a reference on each task in the group to go in the array. */
2126 i
= nr_migrating_tasks
= 0;
2128 struct task_and_cgroup ent
;
2130 /* @tsk either already exited or can't exit until the end */
2131 if (tsk
->flags
& PF_EXITING
)
2134 /* as per above, nr_threads may decrease, but not increase. */
2135 BUG_ON(i
>= group_size
);
2136 get_task_struct(tsk
);
2138 * saying GFP_ATOMIC has no effect here because we did prealloc
2139 * earlier, but it's good form to communicate our expectations.
2142 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2143 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2144 BUG_ON(retval
!= 0);
2146 if (ent
.cgrp
!= cgrp
)
2147 nr_migrating_tasks
++;
2148 } while_each_thread(leader
, tsk
);
2149 /* remember the number of threads in the array for later. */
2151 tset
.tc_array
= group
;
2152 tset
.tc_array_len
= group_size
;
2153 read_unlock(&tasklist_lock
);
2155 /* methods shouldn't be called if no task is actually migrating */
2157 if (!nr_migrating_tasks
)
2161 * step 1: check that we can legitimately attach to the cgroup.
2163 for_each_subsys(root
, ss
) {
2164 if (ss
->can_attach
) {
2165 retval
= ss
->can_attach(ss
, cgrp
, &tset
);
2168 goto out_cancel_attach
;
2174 * step 2: make sure css_sets exist for all threads to be migrated.
2175 * we use find_css_set, which allocates a new one if necessary.
2177 INIT_LIST_HEAD(&newcg_list
);
2178 for (i
= 0; i
< group_size
; i
++) {
2179 tc
= flex_array_get(group
, i
);
2180 /* nothing to do if this task is already in the cgroup */
2181 if (tc
->cgrp
== cgrp
)
2184 * get old css_set pointer. threadgroup is locked so this is
2185 * safe against concurrent cgroup_exit() changing this to
2188 oldcg
= tc
->task
->cgroups
;
2190 /* see if the new one for us is already in the list? */
2191 if (css_set_check_fetched(cgrp
, tc
->task
, oldcg
, &newcg_list
)) {
2192 /* was already there, nothing to do. */
2195 /* we don't already have it. get new one. */
2196 retval
= css_set_prefetch(cgrp
, oldcg
, &newcg_list
);
2199 goto out_list_teardown
;
2204 * step 3: now that we're guaranteed success wrt the css_sets,
2205 * proceed to move all tasks to the new cgroup. There are no
2206 * failure cases after here, so this is the commit point.
2208 for (i
= 0; i
< group_size
; i
++) {
2209 tc
= flex_array_get(group
, i
);
2210 /* leave current thread as it is if it's already there */
2211 if (tc
->cgrp
== cgrp
)
2213 retval
= cgroup_task_migrate(cgrp
, tc
->cgrp
, tc
->task
, true);
2216 /* nothing is sensitive to fork() after this point. */
2219 * step 4: do subsystem attach callbacks.
2221 for_each_subsys(root
, ss
) {
2223 ss
->attach(ss
, cgrp
, &tset
);
2227 * step 5: success! and cleanup
2230 cgroup_wakeup_rmdir_waiter(cgrp
);
2233 /* clean up the list of prefetched css_sets. */
2234 list_for_each_entry_safe(cg_entry
, temp_nobe
, &newcg_list
, links
) {
2235 list_del(&cg_entry
->links
);
2236 put_css_set(cg_entry
->cg
);
2240 /* same deal as in cgroup_attach_task */
2242 for_each_subsys(root
, ss
) {
2243 if (ss
== failed_ss
)
2245 if (ss
->cancel_attach
)
2246 ss
->cancel_attach(ss
, cgrp
, &tset
);
2250 /* clean up the array of referenced threads in the group. */
2251 for (i
= 0; i
< group_size
; i
++) {
2252 tc
= flex_array_get(group
, i
);
2253 put_task_struct(tc
->task
);
2255 out_free_group_list
:
2256 flex_array_free(group
);
2261 * Find the task_struct of the task to attach by vpid and pass it along to the
2262 * function to attach either it or all tasks in its threadgroup. Will lock
2263 * cgroup_mutex and threadgroup; may take task_lock of task.
2265 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2267 struct task_struct
*tsk
;
2268 const struct cred
*cred
= current_cred(), *tcred
;
2271 if (!cgroup_lock_live_group(cgrp
))
2276 tsk
= find_task_by_vpid(pid
);
2284 * RCU protects this access, since tsk was found in the
2285 * tid map. a race with de_thread may cause group_leader
2286 * to stop being the leader, but cgroup_attach_proc will
2289 tsk
= tsk
->group_leader
;
2292 * even if we're attaching all tasks in the thread group, we
2293 * only need to check permissions on one of them.
2295 tcred
= __task_cred(tsk
);
2297 cred
->euid
!= tcred
->uid
&&
2298 cred
->euid
!= tcred
->suid
) {
2303 get_task_struct(tsk
);
2307 tsk
= current
->group_leader
;
2310 get_task_struct(tsk
);
2313 threadgroup_lock(tsk
);
2316 ret
= cgroup_attach_proc(cgrp
, tsk
);
2318 ret
= cgroup_attach_task(cgrp
, tsk
);
2320 threadgroup_unlock(tsk
);
2322 put_task_struct(tsk
);
2327 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2329 return attach_task_by_pid(cgrp
, pid
, false);
2332 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2337 * attach_proc fails with -EAGAIN if threadgroup leadership
2338 * changes in the middle of the operation, in which case we need
2339 * to find the task_struct for the new leader and start over.
2341 ret
= attach_task_by_pid(cgrp
, tgid
, true);
2342 } while (ret
== -EAGAIN
);
2347 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2348 * @cgrp: the cgroup to be checked for liveness
2350 * On success, returns true; the lock should be later released with
2351 * cgroup_unlock(). On failure returns false with no lock held.
2353 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
2355 mutex_lock(&cgroup_mutex
);
2356 if (cgroup_is_removed(cgrp
)) {
2357 mutex_unlock(&cgroup_mutex
);
2362 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
2364 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2367 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2368 if (strlen(buffer
) >= PATH_MAX
)
2370 if (!cgroup_lock_live_group(cgrp
))
2372 mutex_lock(&cgroup_root_mutex
);
2373 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2374 mutex_unlock(&cgroup_root_mutex
);
2379 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2380 struct seq_file
*seq
)
2382 if (!cgroup_lock_live_group(cgrp
))
2384 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2385 seq_putc(seq
, '\n');
2390 /* A buffer size big enough for numbers or short strings */
2391 #define CGROUP_LOCAL_BUFFER_SIZE 64
2393 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2395 const char __user
*userbuf
,
2396 size_t nbytes
, loff_t
*unused_ppos
)
2398 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2404 if (nbytes
>= sizeof(buffer
))
2406 if (copy_from_user(buffer
, userbuf
, nbytes
))
2409 buffer
[nbytes
] = 0; /* nul-terminate */
2410 if (cft
->write_u64
) {
2411 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2414 retval
= cft
->write_u64(cgrp
, cft
, val
);
2416 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2419 retval
= cft
->write_s64(cgrp
, cft
, val
);
2426 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2428 const char __user
*userbuf
,
2429 size_t nbytes
, loff_t
*unused_ppos
)
2431 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2433 size_t max_bytes
= cft
->max_write_len
;
2434 char *buffer
= local_buffer
;
2437 max_bytes
= sizeof(local_buffer
) - 1;
2438 if (nbytes
>= max_bytes
)
2440 /* Allocate a dynamic buffer if we need one */
2441 if (nbytes
>= sizeof(local_buffer
)) {
2442 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2446 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2451 buffer
[nbytes
] = 0; /* nul-terminate */
2452 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2456 if (buffer
!= local_buffer
)
2461 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2462 size_t nbytes
, loff_t
*ppos
)
2464 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2465 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2467 if (cgroup_is_removed(cgrp
))
2470 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2471 if (cft
->write_u64
|| cft
->write_s64
)
2472 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2473 if (cft
->write_string
)
2474 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2476 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2477 return ret
? ret
: nbytes
;
2482 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2484 char __user
*buf
, size_t nbytes
,
2487 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2488 u64 val
= cft
->read_u64(cgrp
, cft
);
2489 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2491 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2494 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2496 char __user
*buf
, size_t nbytes
,
2499 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2500 s64 val
= cft
->read_s64(cgrp
, cft
);
2501 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2503 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2506 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2507 size_t nbytes
, loff_t
*ppos
)
2509 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2510 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2512 if (cgroup_is_removed(cgrp
))
2516 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2518 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2520 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2525 * seqfile ops/methods for returning structured data. Currently just
2526 * supports string->u64 maps, but can be extended in future.
2529 struct cgroup_seqfile_state
{
2531 struct cgroup
*cgroup
;
2534 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2536 struct seq_file
*sf
= cb
->state
;
2537 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2540 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2542 struct cgroup_seqfile_state
*state
= m
->private;
2543 struct cftype
*cft
= state
->cft
;
2544 if (cft
->read_map
) {
2545 struct cgroup_map_cb cb
= {
2546 .fill
= cgroup_map_add
,
2549 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2551 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2554 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2556 struct seq_file
*seq
= file
->private_data
;
2557 kfree(seq
->private);
2558 return single_release(inode
, file
);
2561 static const struct file_operations cgroup_seqfile_operations
= {
2563 .write
= cgroup_file_write
,
2564 .llseek
= seq_lseek
,
2565 .release
= cgroup_seqfile_release
,
2568 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2573 err
= generic_file_open(inode
, file
);
2576 cft
= __d_cft(file
->f_dentry
);
2578 if (cft
->read_map
|| cft
->read_seq_string
) {
2579 struct cgroup_seqfile_state
*state
=
2580 kzalloc(sizeof(*state
), GFP_USER
);
2584 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2585 file
->f_op
= &cgroup_seqfile_operations
;
2586 err
= single_open(file
, cgroup_seqfile_show
, state
);
2589 } else if (cft
->open
)
2590 err
= cft
->open(inode
, file
);
2597 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2599 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2601 return cft
->release(inode
, file
);
2606 * cgroup_rename - Only allow simple rename of directories in place.
2608 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2609 struct inode
*new_dir
, struct dentry
*new_dentry
)
2611 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2613 if (new_dentry
->d_inode
)
2615 if (old_dir
!= new_dir
)
2617 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2620 static const struct file_operations cgroup_file_operations
= {
2621 .read
= cgroup_file_read
,
2622 .write
= cgroup_file_write
,
2623 .llseek
= generic_file_llseek
,
2624 .open
= cgroup_file_open
,
2625 .release
= cgroup_file_release
,
2628 static const struct inode_operations cgroup_dir_inode_operations
= {
2629 .lookup
= cgroup_lookup
,
2630 .mkdir
= cgroup_mkdir
,
2631 .rmdir
= cgroup_rmdir
,
2632 .rename
= cgroup_rename
,
2635 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
2637 if (dentry
->d_name
.len
> NAME_MAX
)
2638 return ERR_PTR(-ENAMETOOLONG
);
2639 d_add(dentry
, NULL
);
2644 * Check if a file is a control file
2646 static inline struct cftype
*__file_cft(struct file
*file
)
2648 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2649 return ERR_PTR(-EINVAL
);
2650 return __d_cft(file
->f_dentry
);
2653 static int cgroup_create_file(struct dentry
*dentry
, mode_t mode
,
2654 struct super_block
*sb
)
2656 struct inode
*inode
;
2660 if (dentry
->d_inode
)
2663 inode
= cgroup_new_inode(mode
, sb
);
2667 if (S_ISDIR(mode
)) {
2668 inode
->i_op
= &cgroup_dir_inode_operations
;
2669 inode
->i_fop
= &simple_dir_operations
;
2671 /* start off with i_nlink == 2 (for "." entry) */
2674 /* start with the directory inode held, so that we can
2675 * populate it without racing with another mkdir */
2676 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2677 } else if (S_ISREG(mode
)) {
2679 inode
->i_fop
= &cgroup_file_operations
;
2681 d_instantiate(dentry
, inode
);
2682 dget(dentry
); /* Extra count - pin the dentry in core */
2687 * cgroup_create_dir - create a directory for an object.
2688 * @cgrp: the cgroup we create the directory for. It must have a valid
2689 * ->parent field. And we are going to fill its ->dentry field.
2690 * @dentry: dentry of the new cgroup
2691 * @mode: mode to set on new directory.
2693 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
2696 struct dentry
*parent
;
2699 parent
= cgrp
->parent
->dentry
;
2700 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
2702 dentry
->d_fsdata
= cgrp
;
2703 inc_nlink(parent
->d_inode
);
2704 rcu_assign_pointer(cgrp
->dentry
, dentry
);
2713 * cgroup_file_mode - deduce file mode of a control file
2714 * @cft: the control file in question
2716 * returns cft->mode if ->mode is not 0
2717 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2718 * returns S_IRUGO if it has only a read handler
2719 * returns S_IWUSR if it has only a write hander
2721 static mode_t
cgroup_file_mode(const struct cftype
*cft
)
2728 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2729 cft
->read_map
|| cft
->read_seq_string
)
2732 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2733 cft
->write_string
|| cft
->trigger
)
2739 int cgroup_add_file(struct cgroup
*cgrp
,
2740 struct cgroup_subsys
*subsys
,
2741 const struct cftype
*cft
)
2743 struct dentry
*dir
= cgrp
->dentry
;
2744 struct dentry
*dentry
;
2748 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2749 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2750 strcpy(name
, subsys
->name
);
2753 strcat(name
, cft
->name
);
2754 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2755 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2756 if (!IS_ERR(dentry
)) {
2757 mode
= cgroup_file_mode(cft
);
2758 error
= cgroup_create_file(dentry
, mode
| S_IFREG
,
2761 dentry
->d_fsdata
= (void *)cft
;
2764 error
= PTR_ERR(dentry
);
2767 EXPORT_SYMBOL_GPL(cgroup_add_file
);
2769 int cgroup_add_files(struct cgroup
*cgrp
,
2770 struct cgroup_subsys
*subsys
,
2771 const struct cftype cft
[],
2775 for (i
= 0; i
< count
; i
++) {
2776 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
2782 EXPORT_SYMBOL_GPL(cgroup_add_files
);
2785 * cgroup_task_count - count the number of tasks in a cgroup.
2786 * @cgrp: the cgroup in question
2788 * Return the number of tasks in the cgroup.
2790 int cgroup_task_count(const struct cgroup
*cgrp
)
2793 struct cg_cgroup_link
*link
;
2795 read_lock(&css_set_lock
);
2796 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2797 count
+= atomic_read(&link
->cg
->refcount
);
2799 read_unlock(&css_set_lock
);
2804 * Advance a list_head iterator. The iterator should be positioned at
2805 * the start of a css_set
2807 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2808 struct cgroup_iter
*it
)
2810 struct list_head
*l
= it
->cg_link
;
2811 struct cg_cgroup_link
*link
;
2814 /* Advance to the next non-empty css_set */
2817 if (l
== &cgrp
->css_sets
) {
2821 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2823 } while (list_empty(&cg
->tasks
));
2825 it
->task
= cg
->tasks
.next
;
2829 * To reduce the fork() overhead for systems that are not actually
2830 * using their cgroups capability, we don't maintain the lists running
2831 * through each css_set to its tasks until we see the list actually
2832 * used - in other words after the first call to cgroup_iter_start().
2834 * The tasklist_lock is not held here, as do_each_thread() and
2835 * while_each_thread() are protected by RCU.
2837 static void cgroup_enable_task_cg_lists(void)
2839 struct task_struct
*p
, *g
;
2840 write_lock(&css_set_lock
);
2841 use_task_css_set_links
= 1;
2842 do_each_thread(g
, p
) {
2845 * We should check if the process is exiting, otherwise
2846 * it will race with cgroup_exit() in that the list
2847 * entry won't be deleted though the process has exited.
2849 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2850 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2852 } while_each_thread(g
, p
);
2853 write_unlock(&css_set_lock
);
2856 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2859 * The first time anyone tries to iterate across a cgroup,
2860 * we need to enable the list linking each css_set to its
2861 * tasks, and fix up all existing tasks.
2863 if (!use_task_css_set_links
)
2864 cgroup_enable_task_cg_lists();
2866 read_lock(&css_set_lock
);
2867 it
->cg_link
= &cgrp
->css_sets
;
2868 cgroup_advance_iter(cgrp
, it
);
2871 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
2872 struct cgroup_iter
*it
)
2874 struct task_struct
*res
;
2875 struct list_head
*l
= it
->task
;
2876 struct cg_cgroup_link
*link
;
2878 /* If the iterator cg is NULL, we have no tasks */
2881 res
= list_entry(l
, struct task_struct
, cg_list
);
2882 /* Advance iterator to find next entry */
2884 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
2885 if (l
== &link
->cg
->tasks
) {
2886 /* We reached the end of this task list - move on to
2887 * the next cg_cgroup_link */
2888 cgroup_advance_iter(cgrp
, it
);
2895 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2897 read_unlock(&css_set_lock
);
2900 static inline int started_after_time(struct task_struct
*t1
,
2901 struct timespec
*time
,
2902 struct task_struct
*t2
)
2904 int start_diff
= timespec_compare(&t1
->start_time
, time
);
2905 if (start_diff
> 0) {
2907 } else if (start_diff
< 0) {
2911 * Arbitrarily, if two processes started at the same
2912 * time, we'll say that the lower pointer value
2913 * started first. Note that t2 may have exited by now
2914 * so this may not be a valid pointer any longer, but
2915 * that's fine - it still serves to distinguish
2916 * between two tasks started (effectively) simultaneously.
2923 * This function is a callback from heap_insert() and is used to order
2925 * In this case we order the heap in descending task start time.
2927 static inline int started_after(void *p1
, void *p2
)
2929 struct task_struct
*t1
= p1
;
2930 struct task_struct
*t2
= p2
;
2931 return started_after_time(t1
, &t2
->start_time
, t2
);
2935 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2936 * @scan: struct cgroup_scanner containing arguments for the scan
2938 * Arguments include pointers to callback functions test_task() and
2940 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2941 * and if it returns true, call process_task() for it also.
2942 * The test_task pointer may be NULL, meaning always true (select all tasks).
2943 * Effectively duplicates cgroup_iter_{start,next,end}()
2944 * but does not lock css_set_lock for the call to process_task().
2945 * The struct cgroup_scanner may be embedded in any structure of the caller's
2947 * It is guaranteed that process_task() will act on every task that
2948 * is a member of the cgroup for the duration of this call. This
2949 * function may or may not call process_task() for tasks that exit
2950 * or move to a different cgroup during the call, or are forked or
2951 * move into the cgroup during the call.
2953 * Note that test_task() may be called with locks held, and may in some
2954 * situations be called multiple times for the same task, so it should
2956 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2957 * pre-allocated and will be used for heap operations (and its "gt" member will
2958 * be overwritten), else a temporary heap will be used (allocation of which
2959 * may cause this function to fail).
2961 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
2964 struct cgroup_iter it
;
2965 struct task_struct
*p
, *dropped
;
2966 /* Never dereference latest_task, since it's not refcounted */
2967 struct task_struct
*latest_task
= NULL
;
2968 struct ptr_heap tmp_heap
;
2969 struct ptr_heap
*heap
;
2970 struct timespec latest_time
= { 0, 0 };
2973 /* The caller supplied our heap and pre-allocated its memory */
2975 heap
->gt
= &started_after
;
2977 /* We need to allocate our own heap memory */
2979 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
2981 /* cannot allocate the heap */
2987 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2988 * to determine which are of interest, and using the scanner's
2989 * "process_task" callback to process any of them that need an update.
2990 * Since we don't want to hold any locks during the task updates,
2991 * gather tasks to be processed in a heap structure.
2992 * The heap is sorted by descending task start time.
2993 * If the statically-sized heap fills up, we overflow tasks that
2994 * started later, and in future iterations only consider tasks that
2995 * started after the latest task in the previous pass. This
2996 * guarantees forward progress and that we don't miss any tasks.
2999 cgroup_iter_start(scan
->cg
, &it
);
3000 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
3002 * Only affect tasks that qualify per the caller's callback,
3003 * if he provided one
3005 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
3008 * Only process tasks that started after the last task
3011 if (!started_after_time(p
, &latest_time
, latest_task
))
3013 dropped
= heap_insert(heap
, p
);
3014 if (dropped
== NULL
) {
3016 * The new task was inserted; the heap wasn't
3020 } else if (dropped
!= p
) {
3022 * The new task was inserted, and pushed out a
3026 put_task_struct(dropped
);
3029 * Else the new task was newer than anything already in
3030 * the heap and wasn't inserted
3033 cgroup_iter_end(scan
->cg
, &it
);
3036 for (i
= 0; i
< heap
->size
; i
++) {
3037 struct task_struct
*q
= heap
->ptrs
[i
];
3039 latest_time
= q
->start_time
;
3042 /* Process the task per the caller's callback */
3043 scan
->process_task(q
, scan
);
3047 * If we had to process any tasks at all, scan again
3048 * in case some of them were in the middle of forking
3049 * children that didn't get processed.
3050 * Not the most efficient way to do it, but it avoids
3051 * having to take callback_mutex in the fork path
3055 if (heap
== &tmp_heap
)
3056 heap_free(&tmp_heap
);
3061 * Stuff for reading the 'tasks'/'procs' files.
3063 * Reading this file can return large amounts of data if a cgroup has
3064 * *lots* of attached tasks. So it may need several calls to read(),
3065 * but we cannot guarantee that the information we produce is correct
3066 * unless we produce it entirely atomically.
3071 * The following two functions "fix" the issue where there are more pids
3072 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3073 * TODO: replace with a kernel-wide solution to this problem
3075 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3076 static void *pidlist_allocate(int count
)
3078 if (PIDLIST_TOO_LARGE(count
))
3079 return vmalloc(count
* sizeof(pid_t
));
3081 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3083 static void pidlist_free(void *p
)
3085 if (is_vmalloc_addr(p
))
3090 static void *pidlist_resize(void *p
, int newcount
)
3093 /* note: if new alloc fails, old p will still be valid either way */
3094 if (is_vmalloc_addr(p
)) {
3095 newlist
= vmalloc(newcount
* sizeof(pid_t
));
3098 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
3101 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
3107 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3108 * If the new stripped list is sufficiently smaller and there's enough memory
3109 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3110 * number of unique elements.
3112 /* is the size difference enough that we should re-allocate the array? */
3113 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3114 static int pidlist_uniq(pid_t
**p
, int length
)
3121 * we presume the 0th element is unique, so i starts at 1. trivial
3122 * edge cases first; no work needs to be done for either
3124 if (length
== 0 || length
== 1)
3126 /* src and dest walk down the list; dest counts unique elements */
3127 for (src
= 1; src
< length
; src
++) {
3128 /* find next unique element */
3129 while (list
[src
] == list
[src
-1]) {
3134 /* dest always points to where the next unique element goes */
3135 list
[dest
] = list
[src
];
3140 * if the length difference is large enough, we want to allocate a
3141 * smaller buffer to save memory. if this fails due to out of memory,
3142 * we'll just stay with what we've got.
3144 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
3145 newlist
= pidlist_resize(list
, dest
);
3152 static int cmppid(const void *a
, const void *b
)
3154 return *(pid_t
*)a
- *(pid_t
*)b
;
3158 * find the appropriate pidlist for our purpose (given procs vs tasks)
3159 * returns with the lock on that pidlist already held, and takes care
3160 * of the use count, or returns NULL with no locks held if we're out of
3163 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3164 enum cgroup_filetype type
)
3166 struct cgroup_pidlist
*l
;
3167 /* don't need task_nsproxy() if we're looking at ourself */
3168 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
3171 * We can't drop the pidlist_mutex before taking the l->mutex in case
3172 * the last ref-holder is trying to remove l from the list at the same
3173 * time. Holding the pidlist_mutex precludes somebody taking whichever
3174 * list we find out from under us - compare release_pid_array().
3176 mutex_lock(&cgrp
->pidlist_mutex
);
3177 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3178 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3179 /* make sure l doesn't vanish out from under us */
3180 down_write(&l
->mutex
);
3181 mutex_unlock(&cgrp
->pidlist_mutex
);
3185 /* entry not found; create a new one */
3186 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3188 mutex_unlock(&cgrp
->pidlist_mutex
);
3191 init_rwsem(&l
->mutex
);
3192 down_write(&l
->mutex
);
3194 l
->key
.ns
= get_pid_ns(ns
);
3195 l
->use_count
= 0; /* don't increment here */
3198 list_add(&l
->links
, &cgrp
->pidlists
);
3199 mutex_unlock(&cgrp
->pidlist_mutex
);
3204 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3206 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3207 struct cgroup_pidlist
**lp
)
3211 int pid
, n
= 0; /* used for populating the array */
3212 struct cgroup_iter it
;
3213 struct task_struct
*tsk
;
3214 struct cgroup_pidlist
*l
;
3217 * If cgroup gets more users after we read count, we won't have
3218 * enough space - tough. This race is indistinguishable to the
3219 * caller from the case that the additional cgroup users didn't
3220 * show up until sometime later on.
3222 length
= cgroup_task_count(cgrp
);
3223 array
= pidlist_allocate(length
);
3226 /* now, populate the array */
3227 cgroup_iter_start(cgrp
, &it
);
3228 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3229 if (unlikely(n
== length
))
3231 /* get tgid or pid for procs or tasks file respectively */
3232 if (type
== CGROUP_FILE_PROCS
)
3233 pid
= task_tgid_vnr(tsk
);
3235 pid
= task_pid_vnr(tsk
);
3236 if (pid
> 0) /* make sure to only use valid results */
3239 cgroup_iter_end(cgrp
, &it
);
3241 /* now sort & (if procs) strip out duplicates */
3242 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3243 if (type
== CGROUP_FILE_PROCS
)
3244 length
= pidlist_uniq(&array
, length
);
3245 l
= cgroup_pidlist_find(cgrp
, type
);
3247 pidlist_free(array
);
3250 /* store array, freeing old if necessary - lock already held */
3251 pidlist_free(l
->list
);
3255 up_write(&l
->mutex
);
3261 * cgroupstats_build - build and fill cgroupstats
3262 * @stats: cgroupstats to fill information into
3263 * @dentry: A dentry entry belonging to the cgroup for which stats have
3266 * Build and fill cgroupstats so that taskstats can export it to user
3269 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3272 struct cgroup
*cgrp
;
3273 struct cgroup_iter it
;
3274 struct task_struct
*tsk
;
3277 * Validate dentry by checking the superblock operations,
3278 * and make sure it's a directory.
3280 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3281 !S_ISDIR(dentry
->d_inode
->i_mode
))
3285 cgrp
= dentry
->d_fsdata
;
3287 cgroup_iter_start(cgrp
, &it
);
3288 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3289 switch (tsk
->state
) {
3291 stats
->nr_running
++;
3293 case TASK_INTERRUPTIBLE
:
3294 stats
->nr_sleeping
++;
3296 case TASK_UNINTERRUPTIBLE
:
3297 stats
->nr_uninterruptible
++;
3300 stats
->nr_stopped
++;
3303 if (delayacct_is_task_waiting_on_io(tsk
))
3304 stats
->nr_io_wait
++;
3308 cgroup_iter_end(cgrp
, &it
);
3316 * seq_file methods for the tasks/procs files. The seq_file position is the
3317 * next pid to display; the seq_file iterator is a pointer to the pid
3318 * in the cgroup->l->list array.
3321 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3324 * Initially we receive a position value that corresponds to
3325 * one more than the last pid shown (or 0 on the first call or
3326 * after a seek to the start). Use a binary-search to find the
3327 * next pid to display, if any
3329 struct cgroup_pidlist
*l
= s
->private;
3330 int index
= 0, pid
= *pos
;
3333 down_read(&l
->mutex
);
3335 int end
= l
->length
;
3337 while (index
< end
) {
3338 int mid
= (index
+ end
) / 2;
3339 if (l
->list
[mid
] == pid
) {
3342 } else if (l
->list
[mid
] <= pid
)
3348 /* If we're off the end of the array, we're done */
3349 if (index
>= l
->length
)
3351 /* Update the abstract position to be the actual pid that we found */
3352 iter
= l
->list
+ index
;
3357 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3359 struct cgroup_pidlist
*l
= s
->private;
3363 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3365 struct cgroup_pidlist
*l
= s
->private;
3367 pid_t
*end
= l
->list
+ l
->length
;
3369 * Advance to the next pid in the array. If this goes off the
3381 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3383 return seq_printf(s
, "%d\n", *(int *)v
);
3387 * seq_operations functions for iterating on pidlists through seq_file -
3388 * independent of whether it's tasks or procs
3390 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3391 .start
= cgroup_pidlist_start
,
3392 .stop
= cgroup_pidlist_stop
,
3393 .next
= cgroup_pidlist_next
,
3394 .show
= cgroup_pidlist_show
,
3397 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3400 * the case where we're the last user of this particular pidlist will
3401 * have us remove it from the cgroup's list, which entails taking the
3402 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3403 * pidlist_mutex, we have to take pidlist_mutex first.
3405 mutex_lock(&l
->owner
->pidlist_mutex
);
3406 down_write(&l
->mutex
);
3407 BUG_ON(!l
->use_count
);
3408 if (!--l
->use_count
) {
3409 /* we're the last user if refcount is 0; remove and free */
3410 list_del(&l
->links
);
3411 mutex_unlock(&l
->owner
->pidlist_mutex
);
3412 pidlist_free(l
->list
);
3413 put_pid_ns(l
->key
.ns
);
3414 up_write(&l
->mutex
);
3418 mutex_unlock(&l
->owner
->pidlist_mutex
);
3419 up_write(&l
->mutex
);
3422 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3424 struct cgroup_pidlist
*l
;
3425 if (!(file
->f_mode
& FMODE_READ
))
3428 * the seq_file will only be initialized if the file was opened for
3429 * reading; hence we check if it's not null only in that case.
3431 l
= ((struct seq_file
*)file
->private_data
)->private;
3432 cgroup_release_pid_array(l
);
3433 return seq_release(inode
, file
);
3436 static const struct file_operations cgroup_pidlist_operations
= {
3438 .llseek
= seq_lseek
,
3439 .write
= cgroup_file_write
,
3440 .release
= cgroup_pidlist_release
,
3444 * The following functions handle opens on a file that displays a pidlist
3445 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3448 /* helper function for the two below it */
3449 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3451 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3452 struct cgroup_pidlist
*l
;
3455 /* Nothing to do for write-only files */
3456 if (!(file
->f_mode
& FMODE_READ
))
3459 /* have the array populated */
3460 retval
= pidlist_array_load(cgrp
, type
, &l
);
3463 /* configure file information */
3464 file
->f_op
= &cgroup_pidlist_operations
;
3466 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3468 cgroup_release_pid_array(l
);
3471 ((struct seq_file
*)file
->private_data
)->private = l
;
3474 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3476 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3478 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3480 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3483 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3486 return notify_on_release(cgrp
);
3489 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3493 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3495 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3497 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3502 * Unregister event and free resources.
3504 * Gets called from workqueue.
3506 static void cgroup_event_remove(struct work_struct
*work
)
3508 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3510 struct cgroup
*cgrp
= event
->cgrp
;
3512 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3514 eventfd_ctx_put(event
->eventfd
);
3520 * Gets called on POLLHUP on eventfd when user closes it.
3522 * Called with wqh->lock held and interrupts disabled.
3524 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3525 int sync
, void *key
)
3527 struct cgroup_event
*event
= container_of(wait
,
3528 struct cgroup_event
, wait
);
3529 struct cgroup
*cgrp
= event
->cgrp
;
3530 unsigned long flags
= (unsigned long)key
;
3532 if (flags
& POLLHUP
) {
3533 __remove_wait_queue(event
->wqh
, &event
->wait
);
3534 spin_lock(&cgrp
->event_list_lock
);
3535 list_del(&event
->list
);
3536 spin_unlock(&cgrp
->event_list_lock
);
3538 * We are in atomic context, but cgroup_event_remove() may
3539 * sleep, so we have to call it in workqueue.
3541 schedule_work(&event
->remove
);
3547 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3548 wait_queue_head_t
*wqh
, poll_table
*pt
)
3550 struct cgroup_event
*event
= container_of(pt
,
3551 struct cgroup_event
, pt
);
3554 add_wait_queue(wqh
, &event
->wait
);
3558 * Parse input and register new cgroup event handler.
3560 * Input must be in format '<event_fd> <control_fd> <args>'.
3561 * Interpretation of args is defined by control file implementation.
3563 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3566 struct cgroup_event
*event
= NULL
;
3567 unsigned int efd
, cfd
;
3568 struct file
*efile
= NULL
;
3569 struct file
*cfile
= NULL
;
3573 efd
= simple_strtoul(buffer
, &endp
, 10);
3578 cfd
= simple_strtoul(buffer
, &endp
, 10);
3579 if ((*endp
!= ' ') && (*endp
!= '\0'))
3583 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3587 INIT_LIST_HEAD(&event
->list
);
3588 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3589 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3590 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3592 efile
= eventfd_fget(efd
);
3593 if (IS_ERR(efile
)) {
3594 ret
= PTR_ERR(efile
);
3598 event
->eventfd
= eventfd_ctx_fileget(efile
);
3599 if (IS_ERR(event
->eventfd
)) {
3600 ret
= PTR_ERR(event
->eventfd
);
3610 /* the process need read permission on control file */
3611 /* AV: shouldn't we check that it's been opened for read instead? */
3612 ret
= inode_permission(cfile
->f_path
.dentry
->d_inode
, MAY_READ
);
3616 event
->cft
= __file_cft(cfile
);
3617 if (IS_ERR(event
->cft
)) {
3618 ret
= PTR_ERR(event
->cft
);
3622 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3627 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3628 event
->eventfd
, buffer
);
3632 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3633 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3639 * Events should be removed after rmdir of cgroup directory, but before
3640 * destroying subsystem state objects. Let's take reference to cgroup
3641 * directory dentry to do that.
3645 spin_lock(&cgrp
->event_list_lock
);
3646 list_add(&event
->list
, &cgrp
->event_list
);
3647 spin_unlock(&cgrp
->event_list_lock
);
3658 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3659 eventfd_ctx_put(event
->eventfd
);
3661 if (!IS_ERR_OR_NULL(efile
))
3669 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3672 return clone_children(cgrp
);
3675 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3680 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3682 clear_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3687 * for the common functions, 'private' gives the type of file
3689 /* for hysterical raisins, we can't put this on the older files */
3690 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3691 static struct cftype files
[] = {
3694 .open
= cgroup_tasks_open
,
3695 .write_u64
= cgroup_tasks_write
,
3696 .release
= cgroup_pidlist_release
,
3697 .mode
= S_IRUGO
| S_IWUSR
,
3700 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3701 .open
= cgroup_procs_open
,
3702 .write_u64
= cgroup_procs_write
,
3703 .release
= cgroup_pidlist_release
,
3704 .mode
= S_IRUGO
| S_IWUSR
,
3707 .name
= "notify_on_release",
3708 .read_u64
= cgroup_read_notify_on_release
,
3709 .write_u64
= cgroup_write_notify_on_release
,
3712 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3713 .write_string
= cgroup_write_event_control
,
3717 .name
= "cgroup.clone_children",
3718 .read_u64
= cgroup_clone_children_read
,
3719 .write_u64
= cgroup_clone_children_write
,
3723 static struct cftype cft_release_agent
= {
3724 .name
= "release_agent",
3725 .read_seq_string
= cgroup_release_agent_show
,
3726 .write_string
= cgroup_release_agent_write
,
3727 .max_write_len
= PATH_MAX
,
3730 static int cgroup_populate_dir(struct cgroup
*cgrp
)
3733 struct cgroup_subsys
*ss
;
3735 /* First clear out any existing files */
3736 cgroup_clear_directory(cgrp
->dentry
);
3738 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
3742 if (cgrp
== cgrp
->top_cgroup
) {
3743 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
3747 for_each_subsys(cgrp
->root
, ss
) {
3748 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
3751 /* This cgroup is ready now */
3752 for_each_subsys(cgrp
->root
, ss
) {
3753 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3755 * Update id->css pointer and make this css visible from
3756 * CSS ID functions. This pointer will be dereferened
3757 * from RCU-read-side without locks.
3760 rcu_assign_pointer(css
->id
->css
, css
);
3766 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
3767 struct cgroup_subsys
*ss
,
3768 struct cgroup
*cgrp
)
3771 atomic_set(&css
->refcnt
, 1);
3774 if (cgrp
== dummytop
)
3775 set_bit(CSS_ROOT
, &css
->flags
);
3776 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
3777 cgrp
->subsys
[ss
->subsys_id
] = css
;
3780 static void cgroup_lock_hierarchy(struct cgroupfs_root
*root
)
3782 /* We need to take each hierarchy_mutex in a consistent order */
3786 * No worry about a race with rebind_subsystems that might mess up the
3787 * locking order, since both parties are under cgroup_mutex.
3789 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3790 struct cgroup_subsys
*ss
= subsys
[i
];
3793 if (ss
->root
== root
)
3794 mutex_lock(&ss
->hierarchy_mutex
);
3798 static void cgroup_unlock_hierarchy(struct cgroupfs_root
*root
)
3802 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3803 struct cgroup_subsys
*ss
= subsys
[i
];
3806 if (ss
->root
== root
)
3807 mutex_unlock(&ss
->hierarchy_mutex
);
3812 * cgroup_create - create a cgroup
3813 * @parent: cgroup that will be parent of the new cgroup
3814 * @dentry: dentry of the new cgroup
3815 * @mode: mode to set on new inode
3817 * Must be called with the mutex on the parent inode held
3819 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
3822 struct cgroup
*cgrp
;
3823 struct cgroupfs_root
*root
= parent
->root
;
3825 struct cgroup_subsys
*ss
;
3826 struct super_block
*sb
= root
->sb
;
3828 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3832 /* Grab a reference on the superblock so the hierarchy doesn't
3833 * get deleted on unmount if there are child cgroups. This
3834 * can be done outside cgroup_mutex, since the sb can't
3835 * disappear while someone has an open control file on the
3837 atomic_inc(&sb
->s_active
);
3839 mutex_lock(&cgroup_mutex
);
3841 init_cgroup_housekeeping(cgrp
);
3843 cgrp
->parent
= parent
;
3844 cgrp
->root
= parent
->root
;
3845 cgrp
->top_cgroup
= parent
->top_cgroup
;
3847 if (notify_on_release(parent
))
3848 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3850 if (clone_children(parent
))
3851 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3853 for_each_subsys(root
, ss
) {
3854 struct cgroup_subsys_state
*css
= ss
->create(ss
, cgrp
);
3860 init_cgroup_css(css
, ss
, cgrp
);
3862 err
= alloc_css_id(ss
, parent
, cgrp
);
3866 /* At error, ->destroy() callback has to free assigned ID. */
3867 if (clone_children(parent
) && ss
->post_clone
)
3868 ss
->post_clone(ss
, cgrp
);
3871 cgroup_lock_hierarchy(root
);
3872 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
3873 cgroup_unlock_hierarchy(root
);
3874 root
->number_of_cgroups
++;
3876 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
3880 /* The cgroup directory was pre-locked for us */
3881 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
3883 err
= cgroup_populate_dir(cgrp
);
3884 /* If err < 0, we have a half-filled directory - oh well ;) */
3886 mutex_unlock(&cgroup_mutex
);
3887 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
3893 cgroup_lock_hierarchy(root
);
3894 list_del(&cgrp
->sibling
);
3895 cgroup_unlock_hierarchy(root
);
3896 root
->number_of_cgroups
--;
3900 for_each_subsys(root
, ss
) {
3901 if (cgrp
->subsys
[ss
->subsys_id
])
3902 ss
->destroy(ss
, cgrp
);
3905 mutex_unlock(&cgroup_mutex
);
3907 /* Release the reference count that we took on the superblock */
3908 deactivate_super(sb
);
3914 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3916 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
3918 /* the vfs holds inode->i_mutex already */
3919 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
3922 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
3924 /* Check the reference count on each subsystem. Since we
3925 * already established that there are no tasks in the
3926 * cgroup, if the css refcount is also 1, then there should
3927 * be no outstanding references, so the subsystem is safe to
3928 * destroy. We scan across all subsystems rather than using
3929 * the per-hierarchy linked list of mounted subsystems since
3930 * we can be called via check_for_release() with no
3931 * synchronization other than RCU, and the subsystem linked
3932 * list isn't RCU-safe */
3935 * We won't need to lock the subsys array, because the subsystems
3936 * we're concerned about aren't going anywhere since our cgroup root
3937 * has a reference on them.
3939 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3940 struct cgroup_subsys
*ss
= subsys
[i
];
3941 struct cgroup_subsys_state
*css
;
3942 /* Skip subsystems not present or not in this hierarchy */
3943 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
3945 css
= cgrp
->subsys
[ss
->subsys_id
];
3946 /* When called from check_for_release() it's possible
3947 * that by this point the cgroup has been removed
3948 * and the css deleted. But a false-positive doesn't
3949 * matter, since it can only happen if the cgroup
3950 * has been deleted and hence no longer needs the
3951 * release agent to be called anyway. */
3952 if (css
&& (atomic_read(&css
->refcnt
) > 1))
3959 * Atomically mark all (or else none) of the cgroup's CSS objects as
3960 * CSS_REMOVED. Return true on success, or false if the cgroup has
3961 * busy subsystems. Call with cgroup_mutex held
3964 static int cgroup_clear_css_refs(struct cgroup
*cgrp
)
3966 struct cgroup_subsys
*ss
;
3967 unsigned long flags
;
3968 bool failed
= false;
3969 local_irq_save(flags
);
3970 for_each_subsys(cgrp
->root
, ss
) {
3971 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3974 /* We can only remove a CSS with a refcnt==1 */
3975 refcnt
= atomic_read(&css
->refcnt
);
3982 * Drop the refcnt to 0 while we check other
3983 * subsystems. This will cause any racing
3984 * css_tryget() to spin until we set the
3985 * CSS_REMOVED bits or abort
3987 if (atomic_cmpxchg(&css
->refcnt
, refcnt
, 0) == refcnt
)
3993 for_each_subsys(cgrp
->root
, ss
) {
3994 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3997 * Restore old refcnt if we previously managed
3998 * to clear it from 1 to 0
4000 if (!atomic_read(&css
->refcnt
))
4001 atomic_set(&css
->refcnt
, 1);
4003 /* Commit the fact that the CSS is removed */
4004 set_bit(CSS_REMOVED
, &css
->flags
);
4007 local_irq_restore(flags
);
4011 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4013 struct cgroup
*cgrp
= dentry
->d_fsdata
;
4015 struct cgroup
*parent
;
4017 struct cgroup_event
*event
, *tmp
;
4020 /* the vfs holds both inode->i_mutex already */
4022 mutex_lock(&cgroup_mutex
);
4023 if (atomic_read(&cgrp
->count
) != 0) {
4024 mutex_unlock(&cgroup_mutex
);
4027 if (!list_empty(&cgrp
->children
)) {
4028 mutex_unlock(&cgroup_mutex
);
4031 mutex_unlock(&cgroup_mutex
);
4034 * In general, subsystem has no css->refcnt after pre_destroy(). But
4035 * in racy cases, subsystem may have to get css->refcnt after
4036 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4037 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4038 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4039 * and subsystem's reference count handling. Please see css_get/put
4040 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4042 set_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4045 * Call pre_destroy handlers of subsys. Notify subsystems
4046 * that rmdir() request comes.
4048 ret
= cgroup_call_pre_destroy(cgrp
);
4050 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4054 mutex_lock(&cgroup_mutex
);
4055 parent
= cgrp
->parent
;
4056 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
)) {
4057 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4058 mutex_unlock(&cgroup_mutex
);
4061 prepare_to_wait(&cgroup_rmdir_waitq
, &wait
, TASK_INTERRUPTIBLE
);
4062 if (!cgroup_clear_css_refs(cgrp
)) {
4063 mutex_unlock(&cgroup_mutex
);
4065 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4066 * prepare_to_wait(), we need to check this flag.
4068 if (test_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
))
4070 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4071 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4072 if (signal_pending(current
))
4076 /* NO css_tryget() can success after here. */
4077 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4078 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4080 raw_spin_lock(&release_list_lock
);
4081 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
4082 if (!list_empty(&cgrp
->release_list
))
4083 list_del_init(&cgrp
->release_list
);
4084 raw_spin_unlock(&release_list_lock
);
4086 cgroup_lock_hierarchy(cgrp
->root
);
4087 /* delete this cgroup from parent->children */
4088 list_del_init(&cgrp
->sibling
);
4089 cgroup_unlock_hierarchy(cgrp
->root
);
4091 d
= dget(cgrp
->dentry
);
4093 cgroup_d_remove_dir(d
);
4096 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4097 check_for_release(parent
);
4100 * Unregister events and notify userspace.
4101 * Notify userspace about cgroup removing only after rmdir of cgroup
4102 * directory to avoid race between userspace and kernelspace
4104 spin_lock(&cgrp
->event_list_lock
);
4105 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4106 list_del(&event
->list
);
4107 remove_wait_queue(event
->wqh
, &event
->wait
);
4108 eventfd_signal(event
->eventfd
, 1);
4109 schedule_work(&event
->remove
);
4111 spin_unlock(&cgrp
->event_list_lock
);
4113 mutex_unlock(&cgroup_mutex
);
4117 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4119 struct cgroup_subsys_state
*css
;
4121 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4123 /* Create the top cgroup state for this subsystem */
4124 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4125 ss
->root
= &rootnode
;
4126 css
= ss
->create(ss
, dummytop
);
4127 /* We don't handle early failures gracefully */
4128 BUG_ON(IS_ERR(css
));
4129 init_cgroup_css(css
, ss
, dummytop
);
4131 /* Update the init_css_set to contain a subsys
4132 * pointer to this state - since the subsystem is
4133 * newly registered, all tasks and hence the
4134 * init_css_set is in the subsystem's top cgroup. */
4135 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
4137 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4139 /* At system boot, before all subsystems have been
4140 * registered, no tasks have been forked, so we don't
4141 * need to invoke fork callbacks here. */
4142 BUG_ON(!list_empty(&init_task
.tasks
));
4144 mutex_init(&ss
->hierarchy_mutex
);
4145 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4148 /* this function shouldn't be used with modular subsystems, since they
4149 * need to register a subsys_id, among other things */
4154 * cgroup_load_subsys: load and register a modular subsystem at runtime
4155 * @ss: the subsystem to load
4157 * This function should be called in a modular subsystem's initcall. If the
4158 * subsystem is built as a module, it will be assigned a new subsys_id and set
4159 * up for use. If the subsystem is built-in anyway, work is delegated to the
4160 * simpler cgroup_init_subsys.
4162 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4165 struct cgroup_subsys_state
*css
;
4167 /* check name and function validity */
4168 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4169 ss
->create
== NULL
|| ss
->destroy
== NULL
)
4173 * we don't support callbacks in modular subsystems. this check is
4174 * before the ss->module check for consistency; a subsystem that could
4175 * be a module should still have no callbacks even if the user isn't
4176 * compiling it as one.
4178 if (ss
->fork
|| ss
->exit
)
4182 * an optionally modular subsystem is built-in: we want to do nothing,
4183 * since cgroup_init_subsys will have already taken care of it.
4185 if (ss
->module
== NULL
) {
4186 /* a few sanity checks */
4187 BUG_ON(ss
->subsys_id
>= CGROUP_BUILTIN_SUBSYS_COUNT
);
4188 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4193 * need to register a subsys id before anything else - for example,
4194 * init_cgroup_css needs it.
4196 mutex_lock(&cgroup_mutex
);
4197 /* find the first empty slot in the array */
4198 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4199 if (subsys
[i
] == NULL
)
4202 if (i
== CGROUP_SUBSYS_COUNT
) {
4203 /* maximum number of subsystems already registered! */
4204 mutex_unlock(&cgroup_mutex
);
4207 /* assign ourselves the subsys_id */
4212 * no ss->create seems to need anything important in the ss struct, so
4213 * this can happen first (i.e. before the rootnode attachment).
4215 css
= ss
->create(ss
, dummytop
);
4217 /* failure case - need to deassign the subsys[] slot. */
4219 mutex_unlock(&cgroup_mutex
);
4220 return PTR_ERR(css
);
4223 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4224 ss
->root
= &rootnode
;
4226 /* our new subsystem will be attached to the dummy hierarchy. */
4227 init_cgroup_css(css
, ss
, dummytop
);
4228 /* init_idr must be after init_cgroup_css because it sets css->id. */
4230 int ret
= cgroup_init_idr(ss
, css
);
4232 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4233 ss
->destroy(ss
, dummytop
);
4235 mutex_unlock(&cgroup_mutex
);
4241 * Now we need to entangle the css into the existing css_sets. unlike
4242 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4243 * will need a new pointer to it; done by iterating the css_set_table.
4244 * furthermore, modifying the existing css_sets will corrupt the hash
4245 * table state, so each changed css_set will need its hash recomputed.
4246 * this is all done under the css_set_lock.
4248 write_lock(&css_set_lock
);
4249 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
4251 struct hlist_node
*node
, *tmp
;
4252 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
4254 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
4255 /* skip entries that we already rehashed */
4256 if (cg
->subsys
[ss
->subsys_id
])
4258 /* remove existing entry */
4259 hlist_del(&cg
->hlist
);
4261 cg
->subsys
[ss
->subsys_id
] = css
;
4262 /* recompute hash and restore entry */
4263 new_bucket
= css_set_hash(cg
->subsys
);
4264 hlist_add_head(&cg
->hlist
, new_bucket
);
4267 write_unlock(&css_set_lock
);
4269 mutex_init(&ss
->hierarchy_mutex
);
4270 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4274 mutex_unlock(&cgroup_mutex
);
4277 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4280 * cgroup_unload_subsys: unload a modular subsystem
4281 * @ss: the subsystem to unload
4283 * This function should be called in a modular subsystem's exitcall. When this
4284 * function is invoked, the refcount on the subsystem's module will be 0, so
4285 * the subsystem will not be attached to any hierarchy.
4287 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4289 struct cg_cgroup_link
*link
;
4290 struct hlist_head
*hhead
;
4292 BUG_ON(ss
->module
== NULL
);
4295 * we shouldn't be called if the subsystem is in use, and the use of
4296 * try_module_get in parse_cgroupfs_options should ensure that it
4297 * doesn't start being used while we're killing it off.
4299 BUG_ON(ss
->root
!= &rootnode
);
4301 mutex_lock(&cgroup_mutex
);
4302 /* deassign the subsys_id */
4303 BUG_ON(ss
->subsys_id
< CGROUP_BUILTIN_SUBSYS_COUNT
);
4304 subsys
[ss
->subsys_id
] = NULL
;
4306 /* remove subsystem from rootnode's list of subsystems */
4307 list_del_init(&ss
->sibling
);
4310 * disentangle the css from all css_sets attached to the dummytop. as
4311 * in loading, we need to pay our respects to the hashtable gods.
4313 write_lock(&css_set_lock
);
4314 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4315 struct css_set
*cg
= link
->cg
;
4317 hlist_del(&cg
->hlist
);
4318 BUG_ON(!cg
->subsys
[ss
->subsys_id
]);
4319 cg
->subsys
[ss
->subsys_id
] = NULL
;
4320 hhead
= css_set_hash(cg
->subsys
);
4321 hlist_add_head(&cg
->hlist
, hhead
);
4323 write_unlock(&css_set_lock
);
4326 * remove subsystem's css from the dummytop and free it - need to free
4327 * before marking as null because ss->destroy needs the cgrp->subsys
4328 * pointer to find their state. note that this also takes care of
4329 * freeing the css_id.
4331 ss
->destroy(ss
, dummytop
);
4332 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4334 mutex_unlock(&cgroup_mutex
);
4336 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4339 * cgroup_init_early - cgroup initialization at system boot
4341 * Initialize cgroups at system boot, and initialize any
4342 * subsystems that request early init.
4344 int __init
cgroup_init_early(void)
4347 atomic_set(&init_css_set
.refcount
, 1);
4348 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4349 INIT_LIST_HEAD(&init_css_set
.tasks
);
4350 INIT_HLIST_NODE(&init_css_set
.hlist
);
4352 init_cgroup_root(&rootnode
);
4354 init_task
.cgroups
= &init_css_set
;
4356 init_css_set_link
.cg
= &init_css_set
;
4357 init_css_set_link
.cgrp
= dummytop
;
4358 list_add(&init_css_set_link
.cgrp_link_list
,
4359 &rootnode
.top_cgroup
.css_sets
);
4360 list_add(&init_css_set_link
.cg_link_list
,
4361 &init_css_set
.cg_links
);
4363 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
4364 INIT_HLIST_HEAD(&css_set_table
[i
]);
4366 /* at bootup time, we don't worry about modular subsystems */
4367 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4368 struct cgroup_subsys
*ss
= subsys
[i
];
4371 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4372 BUG_ON(!ss
->create
);
4373 BUG_ON(!ss
->destroy
);
4374 if (ss
->subsys_id
!= i
) {
4375 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4376 ss
->name
, ss
->subsys_id
);
4381 cgroup_init_subsys(ss
);
4387 * cgroup_init - cgroup initialization
4389 * Register cgroup filesystem and /proc file, and initialize
4390 * any subsystems that didn't request early init.
4392 int __init
cgroup_init(void)
4396 struct hlist_head
*hhead
;
4398 err
= bdi_init(&cgroup_backing_dev_info
);
4402 /* at bootup time, we don't worry about modular subsystems */
4403 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4404 struct cgroup_subsys
*ss
= subsys
[i
];
4405 if (!ss
->early_init
)
4406 cgroup_init_subsys(ss
);
4408 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4411 /* Add init_css_set to the hash table */
4412 hhead
= css_set_hash(init_css_set
.subsys
);
4413 hlist_add_head(&init_css_set
.hlist
, hhead
);
4414 BUG_ON(!init_root_id(&rootnode
));
4416 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4422 err
= register_filesystem(&cgroup_fs_type
);
4424 kobject_put(cgroup_kobj
);
4428 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4432 bdi_destroy(&cgroup_backing_dev_info
);
4438 * proc_cgroup_show()
4439 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4440 * - Used for /proc/<pid>/cgroup.
4441 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4442 * doesn't really matter if tsk->cgroup changes after we read it,
4443 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4444 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4445 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4446 * cgroup to top_cgroup.
4449 /* TODO: Use a proper seq_file iterator */
4450 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4453 struct task_struct
*tsk
;
4456 struct cgroupfs_root
*root
;
4459 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4465 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4471 mutex_lock(&cgroup_mutex
);
4473 for_each_active_root(root
) {
4474 struct cgroup_subsys
*ss
;
4475 struct cgroup
*cgrp
;
4478 seq_printf(m
, "%d:", root
->hierarchy_id
);
4479 for_each_subsys(root
, ss
)
4480 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4481 if (strlen(root
->name
))
4482 seq_printf(m
, "%sname=%s", count
? "," : "",
4485 cgrp
= task_cgroup_from_root(tsk
, root
);
4486 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4494 mutex_unlock(&cgroup_mutex
);
4495 put_task_struct(tsk
);
4502 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4504 struct pid
*pid
= PROC_I(inode
)->pid
;
4505 return single_open(file
, proc_cgroup_show
, pid
);
4508 const struct file_operations proc_cgroup_operations
= {
4509 .open
= cgroup_open
,
4511 .llseek
= seq_lseek
,
4512 .release
= single_release
,
4515 /* Display information about each subsystem and each hierarchy */
4516 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4520 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4522 * ideally we don't want subsystems moving around while we do this.
4523 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4524 * subsys/hierarchy state.
4526 mutex_lock(&cgroup_mutex
);
4527 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4528 struct cgroup_subsys
*ss
= subsys
[i
];
4531 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4532 ss
->name
, ss
->root
->hierarchy_id
,
4533 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4535 mutex_unlock(&cgroup_mutex
);
4539 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4541 return single_open(file
, proc_cgroupstats_show
, NULL
);
4544 static const struct file_operations proc_cgroupstats_operations
= {
4545 .open
= cgroupstats_open
,
4547 .llseek
= seq_lseek
,
4548 .release
= single_release
,
4552 * cgroup_fork - attach newly forked task to its parents cgroup.
4553 * @child: pointer to task_struct of forking parent process.
4555 * Description: A task inherits its parent's cgroup at fork().
4557 * A pointer to the shared css_set was automatically copied in
4558 * fork.c by dup_task_struct(). However, we ignore that copy, since
4559 * it was not made under the protection of RCU, cgroup_mutex or
4560 * threadgroup_change_begin(), so it might no longer be a valid
4561 * cgroup pointer. cgroup_attach_task() might have already changed
4562 * current->cgroups, allowing the previously referenced cgroup
4563 * group to be removed and freed.
4565 * Outside the pointer validity we also need to process the css_set
4566 * inheritance between threadgoup_change_begin() and
4567 * threadgoup_change_end(), this way there is no leak in any process
4568 * wide migration performed by cgroup_attach_proc() that could otherwise
4569 * miss a thread because it is too early or too late in the fork stage.
4571 * At the point that cgroup_fork() is called, 'current' is the parent
4572 * task, and the passed argument 'child' points to the child task.
4574 void cgroup_fork(struct task_struct
*child
)
4577 * We don't need to task_lock() current because current->cgroups
4578 * can't be changed concurrently here. The parent obviously hasn't
4579 * exited and called cgroup_exit(), and we are synchronized against
4580 * cgroup migration through threadgroup_change_begin().
4582 child
->cgroups
= current
->cgroups
;
4583 get_css_set(child
->cgroups
);
4584 INIT_LIST_HEAD(&child
->cg_list
);
4588 * cgroup_fork_callbacks - run fork callbacks
4589 * @child: the new task
4591 * Called on a new task very soon before adding it to the
4592 * tasklist. No need to take any locks since no-one can
4593 * be operating on this task.
4595 void cgroup_fork_callbacks(struct task_struct
*child
)
4597 if (need_forkexit_callback
) {
4600 * forkexit callbacks are only supported for builtin
4601 * subsystems, and the builtin section of the subsys array is
4602 * immutable, so we don't need to lock the subsys array here.
4604 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4605 struct cgroup_subsys
*ss
= subsys
[i
];
4607 ss
->fork(ss
, child
);
4613 * cgroup_post_fork - called on a new task after adding it to the task list
4614 * @child: the task in question
4616 * Adds the task to the list running through its css_set if necessary.
4617 * Has to be after the task is visible on the task list in case we race
4618 * with the first call to cgroup_iter_start() - to guarantee that the
4619 * new task ends up on its list.
4621 void cgroup_post_fork(struct task_struct
*child
)
4623 if (use_task_css_set_links
) {
4624 write_lock(&css_set_lock
);
4626 if (list_empty(&child
->cg_list
))
4627 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4629 write_unlock(&css_set_lock
);
4633 * cgroup_exit - detach cgroup from exiting task
4634 * @tsk: pointer to task_struct of exiting process
4635 * @run_callback: run exit callbacks?
4637 * Description: Detach cgroup from @tsk and release it.
4639 * Note that cgroups marked notify_on_release force every task in
4640 * them to take the global cgroup_mutex mutex when exiting.
4641 * This could impact scaling on very large systems. Be reluctant to
4642 * use notify_on_release cgroups where very high task exit scaling
4643 * is required on large systems.
4645 * the_top_cgroup_hack:
4647 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4649 * We call cgroup_exit() while the task is still competent to
4650 * handle notify_on_release(), then leave the task attached to the
4651 * root cgroup in each hierarchy for the remainder of its exit.
4653 * To do this properly, we would increment the reference count on
4654 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4655 * code we would add a second cgroup function call, to drop that
4656 * reference. This would just create an unnecessary hot spot on
4657 * the top_cgroup reference count, to no avail.
4659 * Normally, holding a reference to a cgroup without bumping its
4660 * count is unsafe. The cgroup could go away, or someone could
4661 * attach us to a different cgroup, decrementing the count on
4662 * the first cgroup that we never incremented. But in this case,
4663 * top_cgroup isn't going away, and either task has PF_EXITING set,
4664 * which wards off any cgroup_attach_task() attempts, or task is a failed
4665 * fork, never visible to cgroup_attach_task.
4667 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4673 * Unlink from the css_set task list if necessary.
4674 * Optimistically check cg_list before taking
4677 if (!list_empty(&tsk
->cg_list
)) {
4678 write_lock(&css_set_lock
);
4679 if (!list_empty(&tsk
->cg_list
))
4680 list_del_init(&tsk
->cg_list
);
4681 write_unlock(&css_set_lock
);
4684 /* Reassign the task to the init_css_set. */
4687 tsk
->cgroups
= &init_css_set
;
4689 if (run_callbacks
&& need_forkexit_callback
) {
4691 * modular subsystems can't use callbacks, so no need to lock
4694 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4695 struct cgroup_subsys
*ss
= subsys
[i
];
4697 struct cgroup
*old_cgrp
=
4698 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4699 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4700 ss
->exit(ss
, cgrp
, old_cgrp
, tsk
);
4707 put_css_set_taskexit(cg
);
4711 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4712 * @cgrp: the cgroup in question
4713 * @task: the task in question
4715 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4718 * If we are sending in dummytop, then presumably we are creating
4719 * the top cgroup in the subsystem.
4721 * Called only by the ns (nsproxy) cgroup.
4723 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
4726 struct cgroup
*target
;
4728 if (cgrp
== dummytop
)
4731 target
= task_cgroup_from_root(task
, cgrp
->root
);
4732 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
4733 cgrp
= cgrp
->parent
;
4734 ret
= (cgrp
== target
);
4738 static void check_for_release(struct cgroup
*cgrp
)
4740 /* All of these checks rely on RCU to keep the cgroup
4741 * structure alive */
4742 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
4743 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
4744 /* Control Group is currently removeable. If it's not
4745 * already queued for a userspace notification, queue
4747 int need_schedule_work
= 0;
4748 raw_spin_lock(&release_list_lock
);
4749 if (!cgroup_is_removed(cgrp
) &&
4750 list_empty(&cgrp
->release_list
)) {
4751 list_add(&cgrp
->release_list
, &release_list
);
4752 need_schedule_work
= 1;
4754 raw_spin_unlock(&release_list_lock
);
4755 if (need_schedule_work
)
4756 schedule_work(&release_agent_work
);
4760 /* Caller must verify that the css is not for root cgroup */
4761 void __css_put(struct cgroup_subsys_state
*css
, int count
)
4763 struct cgroup
*cgrp
= css
->cgroup
;
4766 val
= atomic_sub_return(count
, &css
->refcnt
);
4768 if (notify_on_release(cgrp
)) {
4769 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4770 check_for_release(cgrp
);
4772 cgroup_wakeup_rmdir_waiter(cgrp
);
4775 WARN_ON_ONCE(val
< 1);
4777 EXPORT_SYMBOL_GPL(__css_put
);
4780 * Notify userspace when a cgroup is released, by running the
4781 * configured release agent with the name of the cgroup (path
4782 * relative to the root of cgroup file system) as the argument.
4784 * Most likely, this user command will try to rmdir this cgroup.
4786 * This races with the possibility that some other task will be
4787 * attached to this cgroup before it is removed, or that some other
4788 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4789 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4790 * unused, and this cgroup will be reprieved from its death sentence,
4791 * to continue to serve a useful existence. Next time it's released,
4792 * we will get notified again, if it still has 'notify_on_release' set.
4794 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4795 * means only wait until the task is successfully execve()'d. The
4796 * separate release agent task is forked by call_usermodehelper(),
4797 * then control in this thread returns here, without waiting for the
4798 * release agent task. We don't bother to wait because the caller of
4799 * this routine has no use for the exit status of the release agent
4800 * task, so no sense holding our caller up for that.
4802 static void cgroup_release_agent(struct work_struct
*work
)
4804 BUG_ON(work
!= &release_agent_work
);
4805 mutex_lock(&cgroup_mutex
);
4806 raw_spin_lock(&release_list_lock
);
4807 while (!list_empty(&release_list
)) {
4808 char *argv
[3], *envp
[3];
4810 char *pathbuf
= NULL
, *agentbuf
= NULL
;
4811 struct cgroup
*cgrp
= list_entry(release_list
.next
,
4814 list_del_init(&cgrp
->release_list
);
4815 raw_spin_unlock(&release_list_lock
);
4816 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4819 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
4821 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
4826 argv
[i
++] = agentbuf
;
4827 argv
[i
++] = pathbuf
;
4831 /* minimal command environment */
4832 envp
[i
++] = "HOME=/";
4833 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4836 /* Drop the lock while we invoke the usermode helper,
4837 * since the exec could involve hitting disk and hence
4838 * be a slow process */
4839 mutex_unlock(&cgroup_mutex
);
4840 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
4841 mutex_lock(&cgroup_mutex
);
4845 raw_spin_lock(&release_list_lock
);
4847 raw_spin_unlock(&release_list_lock
);
4848 mutex_unlock(&cgroup_mutex
);
4851 static int __init
cgroup_disable(char *str
)
4856 while ((token
= strsep(&str
, ",")) != NULL
) {
4860 * cgroup_disable, being at boot time, can't know about module
4861 * subsystems, so we don't worry about them.
4863 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4864 struct cgroup_subsys
*ss
= subsys
[i
];
4866 if (!strcmp(token
, ss
->name
)) {
4868 printk(KERN_INFO
"Disabling %s control group"
4869 " subsystem\n", ss
->name
);
4876 __setup("cgroup_disable=", cgroup_disable
);
4879 * Functons for CSS ID.
4883 *To get ID other than 0, this should be called when !cgroup_is_removed().
4885 unsigned short css_id(struct cgroup_subsys_state
*css
)
4887 struct css_id
*cssid
;
4890 * This css_id() can return correct value when somone has refcnt
4891 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4892 * it's unchanged until freed.
4894 cssid
= rcu_dereference_check(css
->id
, atomic_read(&css
->refcnt
));
4900 EXPORT_SYMBOL_GPL(css_id
);
4902 unsigned short css_depth(struct cgroup_subsys_state
*css
)
4904 struct css_id
*cssid
;
4906 cssid
= rcu_dereference_check(css
->id
, atomic_read(&css
->refcnt
));
4909 return cssid
->depth
;
4912 EXPORT_SYMBOL_GPL(css_depth
);
4915 * css_is_ancestor - test "root" css is an ancestor of "child"
4916 * @child: the css to be tested.
4917 * @root: the css supporsed to be an ancestor of the child.
4919 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4920 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4921 * But, considering usual usage, the csses should be valid objects after test.
4922 * Assuming that the caller will do some action to the child if this returns
4923 * returns true, the caller must take "child";s reference count.
4924 * If "child" is valid object and this returns true, "root" is valid, too.
4927 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
4928 const struct cgroup_subsys_state
*root
)
4930 struct css_id
*child_id
;
4931 struct css_id
*root_id
;
4935 child_id
= rcu_dereference(child
->id
);
4936 root_id
= rcu_dereference(root
->id
);
4939 || (child_id
->depth
< root_id
->depth
)
4940 || (child_id
->stack
[root_id
->depth
] != root_id
->id
))
4946 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
4948 struct css_id
*id
= css
->id
;
4949 /* When this is called before css_id initialization, id can be NULL */
4953 BUG_ON(!ss
->use_id
);
4955 rcu_assign_pointer(id
->css
, NULL
);
4956 rcu_assign_pointer(css
->id
, NULL
);
4957 write_lock(&ss
->id_lock
);
4958 idr_remove(&ss
->idr
, id
->id
);
4959 write_unlock(&ss
->id_lock
);
4960 kfree_rcu(id
, rcu_head
);
4962 EXPORT_SYMBOL_GPL(free_css_id
);
4965 * This is called by init or create(). Then, calls to this function are
4966 * always serialized (By cgroup_mutex() at create()).
4969 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
4971 struct css_id
*newid
;
4972 int myid
, error
, size
;
4974 BUG_ON(!ss
->use_id
);
4976 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
4977 newid
= kzalloc(size
, GFP_KERNEL
);
4979 return ERR_PTR(-ENOMEM
);
4981 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
4985 write_lock(&ss
->id_lock
);
4986 /* Don't use 0. allocates an ID of 1-65535 */
4987 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
4988 write_unlock(&ss
->id_lock
);
4990 /* Returns error when there are no free spaces for new ID.*/
4995 if (myid
> CSS_ID_MAX
)
4999 newid
->depth
= depth
;
5003 write_lock(&ss
->id_lock
);
5004 idr_remove(&ss
->idr
, myid
);
5005 write_unlock(&ss
->id_lock
);
5008 return ERR_PTR(error
);
5012 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5013 struct cgroup_subsys_state
*rootcss
)
5015 struct css_id
*newid
;
5017 rwlock_init(&ss
->id_lock
);
5020 newid
= get_new_cssid(ss
, 0);
5022 return PTR_ERR(newid
);
5024 newid
->stack
[0] = newid
->id
;
5025 newid
->css
= rootcss
;
5026 rootcss
->id
= newid
;
5030 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5031 struct cgroup
*child
)
5033 int subsys_id
, i
, depth
= 0;
5034 struct cgroup_subsys_state
*parent_css
, *child_css
;
5035 struct css_id
*child_id
, *parent_id
;
5037 subsys_id
= ss
->subsys_id
;
5038 parent_css
= parent
->subsys
[subsys_id
];
5039 child_css
= child
->subsys
[subsys_id
];
5040 parent_id
= parent_css
->id
;
5041 depth
= parent_id
->depth
+ 1;
5043 child_id
= get_new_cssid(ss
, depth
);
5044 if (IS_ERR(child_id
))
5045 return PTR_ERR(child_id
);
5047 for (i
= 0; i
< depth
; i
++)
5048 child_id
->stack
[i
] = parent_id
->stack
[i
];
5049 child_id
->stack
[depth
] = child_id
->id
;
5051 * child_id->css pointer will be set after this cgroup is available
5052 * see cgroup_populate_dir()
5054 rcu_assign_pointer(child_css
->id
, child_id
);
5060 * css_lookup - lookup css by id
5061 * @ss: cgroup subsys to be looked into.
5064 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5065 * NULL if not. Should be called under rcu_read_lock()
5067 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5069 struct css_id
*cssid
= NULL
;
5071 BUG_ON(!ss
->use_id
);
5072 cssid
= idr_find(&ss
->idr
, id
);
5074 if (unlikely(!cssid
))
5077 return rcu_dereference(cssid
->css
);
5079 EXPORT_SYMBOL_GPL(css_lookup
);
5082 * css_get_next - lookup next cgroup under specified hierarchy.
5083 * @ss: pointer to subsystem
5084 * @id: current position of iteration.
5085 * @root: pointer to css. search tree under this.
5086 * @foundid: position of found object.
5088 * Search next css under the specified hierarchy of rootid. Calling under
5089 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5091 struct cgroup_subsys_state
*
5092 css_get_next(struct cgroup_subsys
*ss
, int id
,
5093 struct cgroup_subsys_state
*root
, int *foundid
)
5095 struct cgroup_subsys_state
*ret
= NULL
;
5098 int rootid
= css_id(root
);
5099 int depth
= css_depth(root
);
5104 BUG_ON(!ss
->use_id
);
5105 /* fill start point for scan */
5109 * scan next entry from bitmap(tree), tmpid is updated after
5112 read_lock(&ss
->id_lock
);
5113 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
5114 read_unlock(&ss
->id_lock
);
5118 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
5119 ret
= rcu_dereference(tmp
->css
);
5125 /* continue to scan from next id */
5132 * get corresponding css from file open on cgroupfs directory
5134 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5136 struct cgroup
*cgrp
;
5137 struct inode
*inode
;
5138 struct cgroup_subsys_state
*css
;
5140 inode
= f
->f_dentry
->d_inode
;
5141 /* check in cgroup filesystem dir */
5142 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5143 return ERR_PTR(-EBADF
);
5145 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5146 return ERR_PTR(-EINVAL
);
5149 cgrp
= __d_cgrp(f
->f_dentry
);
5150 css
= cgrp
->subsys
[id
];
5151 return css
? css
: ERR_PTR(-ENOENT
);
5154 #ifdef CONFIG_CGROUP_DEBUG
5155 static struct cgroup_subsys_state
*debug_create(struct cgroup_subsys
*ss
,
5156 struct cgroup
*cont
)
5158 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5161 return ERR_PTR(-ENOMEM
);
5166 static void debug_destroy(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5168 kfree(cont
->subsys
[debug_subsys_id
]);
5171 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5173 return atomic_read(&cont
->count
);
5176 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5178 return cgroup_task_count(cont
);
5181 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5183 return (u64
)(unsigned long)current
->cgroups
;
5186 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5192 count
= atomic_read(¤t
->cgroups
->refcount
);
5197 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5199 struct seq_file
*seq
)
5201 struct cg_cgroup_link
*link
;
5204 read_lock(&css_set_lock
);
5206 cg
= rcu_dereference(current
->cgroups
);
5207 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5208 struct cgroup
*c
= link
->cgrp
;
5212 name
= c
->dentry
->d_name
.name
;
5215 seq_printf(seq
, "Root %d group %s\n",
5216 c
->root
->hierarchy_id
, name
);
5219 read_unlock(&css_set_lock
);
5223 #define MAX_TASKS_SHOWN_PER_CSS 25
5224 static int cgroup_css_links_read(struct cgroup
*cont
,
5226 struct seq_file
*seq
)
5228 struct cg_cgroup_link
*link
;
5230 read_lock(&css_set_lock
);
5231 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5232 struct css_set
*cg
= link
->cg
;
5233 struct task_struct
*task
;
5235 seq_printf(seq
, "css_set %p\n", cg
);
5236 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5237 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5238 seq_puts(seq
, " ...\n");
5241 seq_printf(seq
, " task %d\n",
5242 task_pid_vnr(task
));
5246 read_unlock(&css_set_lock
);
5250 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5252 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5255 static struct cftype debug_files
[] = {
5257 .name
= "cgroup_refcount",
5258 .read_u64
= cgroup_refcount_read
,
5261 .name
= "taskcount",
5262 .read_u64
= debug_taskcount_read
,
5266 .name
= "current_css_set",
5267 .read_u64
= current_css_set_read
,
5271 .name
= "current_css_set_refcount",
5272 .read_u64
= current_css_set_refcount_read
,
5276 .name
= "current_css_set_cg_links",
5277 .read_seq_string
= current_css_set_cg_links_read
,
5281 .name
= "cgroup_css_links",
5282 .read_seq_string
= cgroup_css_links_read
,
5286 .name
= "releasable",
5287 .read_u64
= releasable_read
,
5291 static int debug_populate(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5293 return cgroup_add_files(cont
, ss
, debug_files
,
5294 ARRAY_SIZE(debug_files
));
5297 struct cgroup_subsys debug_subsys
= {
5299 .create
= debug_create
,
5300 .destroy
= debug_destroy
,
5301 .populate
= debug_populate
,
5302 .subsys_id
= debug_subsys_id
,
5304 #endif /* CONFIG_CGROUP_DEBUG */