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 * cgroup_task_migrate - move a task from one cgroup to another.
1763 * 'guarantee' is set if the caller promises that a new css_set for the task
1764 * will already exist. If not set, this function might sleep, and can fail with
1765 * -ENOMEM. Otherwise, it can only fail with -ESRCH.
1767 static int cgroup_task_migrate(struct cgroup
*cgrp
, struct cgroup
*oldcgrp
,
1768 struct task_struct
*tsk
, bool guarantee
)
1770 struct css_set
*oldcg
;
1771 struct css_set
*newcg
;
1774 * get old css_set. we need to take task_lock and refcount it, because
1775 * an exiting task can change its css_set to init_css_set and drop its
1776 * old one without taking cgroup_mutex.
1779 oldcg
= tsk
->cgroups
;
1783 /* locate or allocate a new css_set for this task. */
1785 /* we know the css_set we want already exists. */
1786 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
1787 read_lock(&css_set_lock
);
1788 newcg
= find_existing_css_set(oldcg
, cgrp
, template);
1791 read_unlock(&css_set_lock
);
1794 /* find_css_set will give us newcg already referenced. */
1795 newcg
= find_css_set(oldcg
, cgrp
);
1803 /* if PF_EXITING is set, the tsk->cgroups pointer is no longer safe. */
1805 if (tsk
->flags
& PF_EXITING
) {
1810 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1813 /* Update the css_set linked lists if we're using them */
1814 write_lock(&css_set_lock
);
1815 if (!list_empty(&tsk
->cg_list
))
1816 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1817 write_unlock(&css_set_lock
);
1820 * We just gained a reference on oldcg by taking it from the task. As
1821 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1822 * it here; it will be freed under RCU.
1826 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1831 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1832 * @cgrp: the cgroup the task is attaching to
1833 * @tsk: the task to be attached
1835 * Call holding cgroup_mutex. May take task_lock of
1836 * the task 'tsk' during call.
1838 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1841 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1842 struct cgroup
*oldcgrp
;
1843 struct cgroupfs_root
*root
= cgrp
->root
;
1845 /* Nothing to do if the task is already in that cgroup */
1846 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1847 if (cgrp
== oldcgrp
)
1850 for_each_subsys(root
, ss
) {
1851 if (ss
->can_attach
) {
1852 retval
= ss
->can_attach(ss
, cgrp
, tsk
);
1855 * Remember on which subsystem the can_attach()
1856 * failed, so that we only call cancel_attach()
1857 * against the subsystems whose can_attach()
1858 * succeeded. (See below)
1864 if (ss
->can_attach_task
) {
1865 retval
= ss
->can_attach_task(cgrp
, tsk
);
1873 retval
= cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, false);
1877 for_each_subsys(root
, ss
) {
1879 ss
->pre_attach(cgrp
);
1880 if (ss
->attach_task
)
1881 ss
->attach_task(cgrp
, tsk
);
1883 ss
->attach(ss
, cgrp
, oldcgrp
, tsk
);
1889 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1890 * is no longer empty.
1892 cgroup_wakeup_rmdir_waiter(cgrp
);
1895 for_each_subsys(root
, ss
) {
1896 if (ss
== failed_ss
)
1898 * This subsystem was the one that failed the
1899 * can_attach() check earlier, so we don't need
1900 * to call cancel_attach() against it or any
1901 * remaining subsystems.
1904 if (ss
->cancel_attach
)
1905 ss
->cancel_attach(ss
, cgrp
, tsk
);
1912 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1913 * @from: attach to all cgroups of a given task
1914 * @tsk: the task to be attached
1916 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
1918 struct cgroupfs_root
*root
;
1922 for_each_active_root(root
) {
1923 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
1925 retval
= cgroup_attach_task(from_cg
, tsk
);
1933 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
1936 * cgroup_attach_proc works in two stages, the first of which prefetches all
1937 * new css_sets needed (to make sure we have enough memory before committing
1938 * to the move) and stores them in a list of entries of the following type.
1939 * TODO: possible optimization: use css_set->rcu_head for chaining instead
1941 struct cg_list_entry
{
1943 struct list_head links
;
1946 static bool css_set_check_fetched(struct cgroup
*cgrp
,
1947 struct task_struct
*tsk
, struct css_set
*cg
,
1948 struct list_head
*newcg_list
)
1950 struct css_set
*newcg
;
1951 struct cg_list_entry
*cg_entry
;
1952 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
1954 read_lock(&css_set_lock
);
1955 newcg
= find_existing_css_set(cg
, cgrp
, template);
1958 read_unlock(&css_set_lock
);
1960 /* doesn't exist at all? */
1963 /* see if it's already in the list */
1964 list_for_each_entry(cg_entry
, newcg_list
, links
) {
1965 if (cg_entry
->cg
== newcg
) {
1977 * Find the new css_set and store it in the list in preparation for moving the
1978 * given task to the given cgroup. Returns 0 or -ENOMEM.
1980 static int css_set_prefetch(struct cgroup
*cgrp
, struct css_set
*cg
,
1981 struct list_head
*newcg_list
)
1983 struct css_set
*newcg
;
1984 struct cg_list_entry
*cg_entry
;
1986 /* ensure a new css_set will exist for this thread */
1987 newcg
= find_css_set(cg
, cgrp
);
1990 /* add it to the list */
1991 cg_entry
= kmalloc(sizeof(struct cg_list_entry
), GFP_KERNEL
);
1996 cg_entry
->cg
= newcg
;
1997 list_add(&cg_entry
->links
, newcg_list
);
2002 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2003 * @cgrp: the cgroup to attach to
2004 * @leader: the threadgroup leader task_struct of the group to be attached
2006 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2007 * task_lock of each thread in leader's threadgroup individually in turn.
2009 int cgroup_attach_proc(struct cgroup
*cgrp
, struct task_struct
*leader
)
2011 int retval
, i
, group_size
;
2012 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
2013 bool cancel_failed_ss
= false;
2014 /* guaranteed to be initialized later, but the compiler needs this */
2015 struct cgroup
*oldcgrp
= NULL
;
2016 struct css_set
*oldcg
;
2017 struct cgroupfs_root
*root
= cgrp
->root
;
2018 /* threadgroup list cursor and array */
2019 struct task_struct
*tsk
;
2020 struct flex_array
*group
;
2022 * we need to make sure we have css_sets for all the tasks we're
2023 * going to move -before- we actually start moving them, so that in
2024 * case we get an ENOMEM we can bail out before making any changes.
2026 struct list_head newcg_list
;
2027 struct cg_list_entry
*cg_entry
, *temp_nobe
;
2030 * step 0: in order to do expensive, possibly blocking operations for
2031 * every thread, we cannot iterate the thread group list, since it needs
2032 * rcu or tasklist locked. instead, build an array of all threads in the
2033 * group - group_rwsem prevents new threads from appearing, and if
2034 * threads exit, this will just be an over-estimate.
2036 group_size
= get_nr_threads(leader
);
2037 /* flex_array supports very large thread-groups better than kmalloc. */
2038 group
= flex_array_alloc(sizeof(struct task_struct
*), group_size
,
2042 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2043 retval
= flex_array_prealloc(group
, 0, group_size
- 1, GFP_KERNEL
);
2045 goto out_free_group_list
;
2047 /* prevent changes to the threadgroup list while we take a snapshot. */
2048 read_lock(&tasklist_lock
);
2049 if (!thread_group_leader(leader
)) {
2051 * a race with de_thread from another thread's exec() may strip
2052 * us of our leadership, making while_each_thread unsafe to use
2053 * on this task. if this happens, there is no choice but to
2054 * throw this task away and try again (from cgroup_procs_write);
2055 * this is "double-double-toil-and-trouble-check locking".
2057 read_unlock(&tasklist_lock
);
2059 goto out_free_group_list
;
2061 /* take a reference on each task in the group to go in the array. */
2065 /* as per above, nr_threads may decrease, but not increase. */
2066 BUG_ON(i
>= group_size
);
2067 get_task_struct(tsk
);
2069 * saying GFP_ATOMIC has no effect here because we did prealloc
2070 * earlier, but it's good form to communicate our expectations.
2072 retval
= flex_array_put_ptr(group
, i
, tsk
, GFP_ATOMIC
);
2073 BUG_ON(retval
!= 0);
2075 } while_each_thread(leader
, tsk
);
2076 /* remember the number of threads in the array for later. */
2078 read_unlock(&tasklist_lock
);
2081 * step 1: check that we can legitimately attach to the cgroup.
2083 for_each_subsys(root
, ss
) {
2084 if (ss
->can_attach
) {
2085 retval
= ss
->can_attach(ss
, cgrp
, leader
);
2088 goto out_cancel_attach
;
2091 /* a callback to be run on every thread in the threadgroup. */
2092 if (ss
->can_attach_task
) {
2093 /* run on each task in the threadgroup. */
2094 for (i
= 0; i
< group_size
; i
++) {
2095 tsk
= flex_array_get_ptr(group
, i
);
2096 retval
= ss
->can_attach_task(cgrp
, tsk
);
2099 cancel_failed_ss
= true;
2100 goto out_cancel_attach
;
2107 * step 2: make sure css_sets exist for all threads to be migrated.
2108 * we use find_css_set, which allocates a new one if necessary.
2110 INIT_LIST_HEAD(&newcg_list
);
2111 for (i
= 0; i
< group_size
; i
++) {
2112 tsk
= flex_array_get_ptr(group
, i
);
2113 /* nothing to do if this task is already in the cgroup */
2114 oldcgrp
= task_cgroup_from_root(tsk
, root
);
2115 if (cgrp
== oldcgrp
)
2117 /* get old css_set pointer */
2119 if (tsk
->flags
& PF_EXITING
) {
2120 /* ignore this task if it's going away */
2124 oldcg
= tsk
->cgroups
;
2127 /* see if the new one for us is already in the list? */
2128 if (css_set_check_fetched(cgrp
, tsk
, oldcg
, &newcg_list
)) {
2129 /* was already there, nothing to do. */
2132 /* we don't already have it. get new one. */
2133 retval
= css_set_prefetch(cgrp
, oldcg
, &newcg_list
);
2136 goto out_list_teardown
;
2141 * step 3: now that we're guaranteed success wrt the css_sets, proceed
2142 * to move all tasks to the new cgroup, calling ss->attach_task for each
2143 * one along the way. there are no failure cases after here, so this is
2146 for_each_subsys(root
, ss
) {
2148 ss
->pre_attach(cgrp
);
2150 for (i
= 0; i
< group_size
; i
++) {
2151 tsk
= flex_array_get_ptr(group
, i
);
2152 /* leave current thread as it is if it's already there */
2153 oldcgrp
= task_cgroup_from_root(tsk
, root
);
2154 if (cgrp
== oldcgrp
)
2156 /* if the thread is PF_EXITING, it can just get skipped. */
2157 retval
= cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, true);
2159 /* attach each task to each subsystem */
2160 for_each_subsys(root
, ss
) {
2161 if (ss
->attach_task
)
2162 ss
->attach_task(cgrp
, tsk
);
2165 BUG_ON(retval
!= -ESRCH
);
2168 /* nothing is sensitive to fork() after this point. */
2171 * step 4: do expensive, non-thread-specific subsystem callbacks.
2172 * TODO: if ever a subsystem needs to know the oldcgrp for each task
2173 * being moved, this call will need to be reworked to communicate that.
2175 for_each_subsys(root
, ss
) {
2177 ss
->attach(ss
, cgrp
, oldcgrp
, leader
);
2181 * step 5: success! and cleanup
2184 cgroup_wakeup_rmdir_waiter(cgrp
);
2187 /* clean up the list of prefetched css_sets. */
2188 list_for_each_entry_safe(cg_entry
, temp_nobe
, &newcg_list
, links
) {
2189 list_del(&cg_entry
->links
);
2190 put_css_set(cg_entry
->cg
);
2194 /* same deal as in cgroup_attach_task */
2196 for_each_subsys(root
, ss
) {
2197 if (ss
== failed_ss
) {
2198 if (cancel_failed_ss
&& ss
->cancel_attach
)
2199 ss
->cancel_attach(ss
, cgrp
, leader
);
2202 if (ss
->cancel_attach
)
2203 ss
->cancel_attach(ss
, cgrp
, leader
);
2206 /* clean up the array of referenced threads in the group. */
2207 for (i
= 0; i
< group_size
; i
++) {
2208 tsk
= flex_array_get_ptr(group
, i
);
2209 put_task_struct(tsk
);
2211 out_free_group_list
:
2212 flex_array_free(group
);
2217 * Find the task_struct of the task to attach by vpid and pass it along to the
2218 * function to attach either it or all tasks in its threadgroup. Will take
2219 * cgroup_mutex; may take task_lock of task.
2221 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2223 struct task_struct
*tsk
;
2224 const struct cred
*cred
= current_cred(), *tcred
;
2227 if (!cgroup_lock_live_group(cgrp
))
2232 tsk
= find_task_by_vpid(pid
);
2240 * RCU protects this access, since tsk was found in the
2241 * tid map. a race with de_thread may cause group_leader
2242 * to stop being the leader, but cgroup_attach_proc will
2245 tsk
= tsk
->group_leader
;
2246 } else if (tsk
->flags
& PF_EXITING
) {
2247 /* optimization for the single-task-only case */
2253 * even if we're attaching all tasks in the thread group, we
2254 * only need to check permissions on one of them.
2256 tcred
= __task_cred(tsk
);
2258 cred
->euid
!= tcred
->uid
&&
2259 cred
->euid
!= tcred
->suid
) {
2264 get_task_struct(tsk
);
2268 tsk
= current
->group_leader
;
2271 get_task_struct(tsk
);
2275 threadgroup_lock(tsk
);
2276 ret
= cgroup_attach_proc(cgrp
, tsk
);
2277 threadgroup_unlock(tsk
);
2279 ret
= cgroup_attach_task(cgrp
, tsk
);
2281 put_task_struct(tsk
);
2286 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2288 return attach_task_by_pid(cgrp
, pid
, false);
2291 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2296 * attach_proc fails with -EAGAIN if threadgroup leadership
2297 * changes in the middle of the operation, in which case we need
2298 * to find the task_struct for the new leader and start over.
2300 ret
= attach_task_by_pid(cgrp
, tgid
, true);
2301 } while (ret
== -EAGAIN
);
2306 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2307 * @cgrp: the cgroup to be checked for liveness
2309 * On success, returns true; the lock should be later released with
2310 * cgroup_unlock(). On failure returns false with no lock held.
2312 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
2314 mutex_lock(&cgroup_mutex
);
2315 if (cgroup_is_removed(cgrp
)) {
2316 mutex_unlock(&cgroup_mutex
);
2321 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
2323 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2326 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2327 if (strlen(buffer
) >= PATH_MAX
)
2329 if (!cgroup_lock_live_group(cgrp
))
2331 mutex_lock(&cgroup_root_mutex
);
2332 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2333 mutex_unlock(&cgroup_root_mutex
);
2338 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2339 struct seq_file
*seq
)
2341 if (!cgroup_lock_live_group(cgrp
))
2343 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2344 seq_putc(seq
, '\n');
2349 /* A buffer size big enough for numbers or short strings */
2350 #define CGROUP_LOCAL_BUFFER_SIZE 64
2352 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2354 const char __user
*userbuf
,
2355 size_t nbytes
, loff_t
*unused_ppos
)
2357 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2363 if (nbytes
>= sizeof(buffer
))
2365 if (copy_from_user(buffer
, userbuf
, nbytes
))
2368 buffer
[nbytes
] = 0; /* nul-terminate */
2369 if (cft
->write_u64
) {
2370 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2373 retval
= cft
->write_u64(cgrp
, cft
, val
);
2375 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2378 retval
= cft
->write_s64(cgrp
, cft
, val
);
2385 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2387 const char __user
*userbuf
,
2388 size_t nbytes
, loff_t
*unused_ppos
)
2390 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2392 size_t max_bytes
= cft
->max_write_len
;
2393 char *buffer
= local_buffer
;
2396 max_bytes
= sizeof(local_buffer
) - 1;
2397 if (nbytes
>= max_bytes
)
2399 /* Allocate a dynamic buffer if we need one */
2400 if (nbytes
>= sizeof(local_buffer
)) {
2401 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2405 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2410 buffer
[nbytes
] = 0; /* nul-terminate */
2411 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2415 if (buffer
!= local_buffer
)
2420 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2421 size_t nbytes
, loff_t
*ppos
)
2423 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2424 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2426 if (cgroup_is_removed(cgrp
))
2429 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2430 if (cft
->write_u64
|| cft
->write_s64
)
2431 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2432 if (cft
->write_string
)
2433 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2435 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2436 return ret
? ret
: nbytes
;
2441 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2443 char __user
*buf
, size_t nbytes
,
2446 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2447 u64 val
= cft
->read_u64(cgrp
, cft
);
2448 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2450 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2453 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2455 char __user
*buf
, size_t nbytes
,
2458 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2459 s64 val
= cft
->read_s64(cgrp
, cft
);
2460 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2462 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2465 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2466 size_t nbytes
, loff_t
*ppos
)
2468 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2469 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2471 if (cgroup_is_removed(cgrp
))
2475 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2477 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2479 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2484 * seqfile ops/methods for returning structured data. Currently just
2485 * supports string->u64 maps, but can be extended in future.
2488 struct cgroup_seqfile_state
{
2490 struct cgroup
*cgroup
;
2493 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2495 struct seq_file
*sf
= cb
->state
;
2496 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2499 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2501 struct cgroup_seqfile_state
*state
= m
->private;
2502 struct cftype
*cft
= state
->cft
;
2503 if (cft
->read_map
) {
2504 struct cgroup_map_cb cb
= {
2505 .fill
= cgroup_map_add
,
2508 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2510 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2513 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2515 struct seq_file
*seq
= file
->private_data
;
2516 kfree(seq
->private);
2517 return single_release(inode
, file
);
2520 static const struct file_operations cgroup_seqfile_operations
= {
2522 .write
= cgroup_file_write
,
2523 .llseek
= seq_lseek
,
2524 .release
= cgroup_seqfile_release
,
2527 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2532 err
= generic_file_open(inode
, file
);
2535 cft
= __d_cft(file
->f_dentry
);
2537 if (cft
->read_map
|| cft
->read_seq_string
) {
2538 struct cgroup_seqfile_state
*state
=
2539 kzalloc(sizeof(*state
), GFP_USER
);
2543 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2544 file
->f_op
= &cgroup_seqfile_operations
;
2545 err
= single_open(file
, cgroup_seqfile_show
, state
);
2548 } else if (cft
->open
)
2549 err
= cft
->open(inode
, file
);
2556 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2558 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2560 return cft
->release(inode
, file
);
2565 * cgroup_rename - Only allow simple rename of directories in place.
2567 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2568 struct inode
*new_dir
, struct dentry
*new_dentry
)
2570 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2572 if (new_dentry
->d_inode
)
2574 if (old_dir
!= new_dir
)
2576 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2579 static const struct file_operations cgroup_file_operations
= {
2580 .read
= cgroup_file_read
,
2581 .write
= cgroup_file_write
,
2582 .llseek
= generic_file_llseek
,
2583 .open
= cgroup_file_open
,
2584 .release
= cgroup_file_release
,
2587 static const struct inode_operations cgroup_dir_inode_operations
= {
2588 .lookup
= cgroup_lookup
,
2589 .mkdir
= cgroup_mkdir
,
2590 .rmdir
= cgroup_rmdir
,
2591 .rename
= cgroup_rename
,
2594 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
2596 if (dentry
->d_name
.len
> NAME_MAX
)
2597 return ERR_PTR(-ENAMETOOLONG
);
2598 d_add(dentry
, NULL
);
2603 * Check if a file is a control file
2605 static inline struct cftype
*__file_cft(struct file
*file
)
2607 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2608 return ERR_PTR(-EINVAL
);
2609 return __d_cft(file
->f_dentry
);
2612 static int cgroup_create_file(struct dentry
*dentry
, mode_t mode
,
2613 struct super_block
*sb
)
2615 struct inode
*inode
;
2619 if (dentry
->d_inode
)
2622 inode
= cgroup_new_inode(mode
, sb
);
2626 if (S_ISDIR(mode
)) {
2627 inode
->i_op
= &cgroup_dir_inode_operations
;
2628 inode
->i_fop
= &simple_dir_operations
;
2630 /* start off with i_nlink == 2 (for "." entry) */
2633 /* start with the directory inode held, so that we can
2634 * populate it without racing with another mkdir */
2635 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2636 } else if (S_ISREG(mode
)) {
2638 inode
->i_fop
= &cgroup_file_operations
;
2640 d_instantiate(dentry
, inode
);
2641 dget(dentry
); /* Extra count - pin the dentry in core */
2646 * cgroup_create_dir - create a directory for an object.
2647 * @cgrp: the cgroup we create the directory for. It must have a valid
2648 * ->parent field. And we are going to fill its ->dentry field.
2649 * @dentry: dentry of the new cgroup
2650 * @mode: mode to set on new directory.
2652 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
2655 struct dentry
*parent
;
2658 parent
= cgrp
->parent
->dentry
;
2659 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
2661 dentry
->d_fsdata
= cgrp
;
2662 inc_nlink(parent
->d_inode
);
2663 rcu_assign_pointer(cgrp
->dentry
, dentry
);
2672 * cgroup_file_mode - deduce file mode of a control file
2673 * @cft: the control file in question
2675 * returns cft->mode if ->mode is not 0
2676 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2677 * returns S_IRUGO if it has only a read handler
2678 * returns S_IWUSR if it has only a write hander
2680 static mode_t
cgroup_file_mode(const struct cftype
*cft
)
2687 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2688 cft
->read_map
|| cft
->read_seq_string
)
2691 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2692 cft
->write_string
|| cft
->trigger
)
2698 int cgroup_add_file(struct cgroup
*cgrp
,
2699 struct cgroup_subsys
*subsys
,
2700 const struct cftype
*cft
)
2702 struct dentry
*dir
= cgrp
->dentry
;
2703 struct dentry
*dentry
;
2707 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2708 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2709 strcpy(name
, subsys
->name
);
2712 strcat(name
, cft
->name
);
2713 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2714 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2715 if (!IS_ERR(dentry
)) {
2716 mode
= cgroup_file_mode(cft
);
2717 error
= cgroup_create_file(dentry
, mode
| S_IFREG
,
2720 dentry
->d_fsdata
= (void *)cft
;
2723 error
= PTR_ERR(dentry
);
2726 EXPORT_SYMBOL_GPL(cgroup_add_file
);
2728 int cgroup_add_files(struct cgroup
*cgrp
,
2729 struct cgroup_subsys
*subsys
,
2730 const struct cftype cft
[],
2734 for (i
= 0; i
< count
; i
++) {
2735 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
2741 EXPORT_SYMBOL_GPL(cgroup_add_files
);
2744 * cgroup_task_count - count the number of tasks in a cgroup.
2745 * @cgrp: the cgroup in question
2747 * Return the number of tasks in the cgroup.
2749 int cgroup_task_count(const struct cgroup
*cgrp
)
2752 struct cg_cgroup_link
*link
;
2754 read_lock(&css_set_lock
);
2755 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2756 count
+= atomic_read(&link
->cg
->refcount
);
2758 read_unlock(&css_set_lock
);
2763 * Advance a list_head iterator. The iterator should be positioned at
2764 * the start of a css_set
2766 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2767 struct cgroup_iter
*it
)
2769 struct list_head
*l
= it
->cg_link
;
2770 struct cg_cgroup_link
*link
;
2773 /* Advance to the next non-empty css_set */
2776 if (l
== &cgrp
->css_sets
) {
2780 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2782 } while (list_empty(&cg
->tasks
));
2784 it
->task
= cg
->tasks
.next
;
2788 * To reduce the fork() overhead for systems that are not actually
2789 * using their cgroups capability, we don't maintain the lists running
2790 * through each css_set to its tasks until we see the list actually
2791 * used - in other words after the first call to cgroup_iter_start().
2793 * The tasklist_lock is not held here, as do_each_thread() and
2794 * while_each_thread() are protected by RCU.
2796 static void cgroup_enable_task_cg_lists(void)
2798 struct task_struct
*p
, *g
;
2799 write_lock(&css_set_lock
);
2800 use_task_css_set_links
= 1;
2801 do_each_thread(g
, p
) {
2804 * We should check if the process is exiting, otherwise
2805 * it will race with cgroup_exit() in that the list
2806 * entry won't be deleted though the process has exited.
2808 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2809 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2811 } while_each_thread(g
, p
);
2812 write_unlock(&css_set_lock
);
2815 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2818 * The first time anyone tries to iterate across a cgroup,
2819 * we need to enable the list linking each css_set to its
2820 * tasks, and fix up all existing tasks.
2822 if (!use_task_css_set_links
)
2823 cgroup_enable_task_cg_lists();
2825 read_lock(&css_set_lock
);
2826 it
->cg_link
= &cgrp
->css_sets
;
2827 cgroup_advance_iter(cgrp
, it
);
2830 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
2831 struct cgroup_iter
*it
)
2833 struct task_struct
*res
;
2834 struct list_head
*l
= it
->task
;
2835 struct cg_cgroup_link
*link
;
2837 /* If the iterator cg is NULL, we have no tasks */
2840 res
= list_entry(l
, struct task_struct
, cg_list
);
2841 /* Advance iterator to find next entry */
2843 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
2844 if (l
== &link
->cg
->tasks
) {
2845 /* We reached the end of this task list - move on to
2846 * the next cg_cgroup_link */
2847 cgroup_advance_iter(cgrp
, it
);
2854 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2856 read_unlock(&css_set_lock
);
2859 static inline int started_after_time(struct task_struct
*t1
,
2860 struct timespec
*time
,
2861 struct task_struct
*t2
)
2863 int start_diff
= timespec_compare(&t1
->start_time
, time
);
2864 if (start_diff
> 0) {
2866 } else if (start_diff
< 0) {
2870 * Arbitrarily, if two processes started at the same
2871 * time, we'll say that the lower pointer value
2872 * started first. Note that t2 may have exited by now
2873 * so this may not be a valid pointer any longer, but
2874 * that's fine - it still serves to distinguish
2875 * between two tasks started (effectively) simultaneously.
2882 * This function is a callback from heap_insert() and is used to order
2884 * In this case we order the heap in descending task start time.
2886 static inline int started_after(void *p1
, void *p2
)
2888 struct task_struct
*t1
= p1
;
2889 struct task_struct
*t2
= p2
;
2890 return started_after_time(t1
, &t2
->start_time
, t2
);
2894 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2895 * @scan: struct cgroup_scanner containing arguments for the scan
2897 * Arguments include pointers to callback functions test_task() and
2899 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2900 * and if it returns true, call process_task() for it also.
2901 * The test_task pointer may be NULL, meaning always true (select all tasks).
2902 * Effectively duplicates cgroup_iter_{start,next,end}()
2903 * but does not lock css_set_lock for the call to process_task().
2904 * The struct cgroup_scanner may be embedded in any structure of the caller's
2906 * It is guaranteed that process_task() will act on every task that
2907 * is a member of the cgroup for the duration of this call. This
2908 * function may or may not call process_task() for tasks that exit
2909 * or move to a different cgroup during the call, or are forked or
2910 * move into the cgroup during the call.
2912 * Note that test_task() may be called with locks held, and may in some
2913 * situations be called multiple times for the same task, so it should
2915 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2916 * pre-allocated and will be used for heap operations (and its "gt" member will
2917 * be overwritten), else a temporary heap will be used (allocation of which
2918 * may cause this function to fail).
2920 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
2923 struct cgroup_iter it
;
2924 struct task_struct
*p
, *dropped
;
2925 /* Never dereference latest_task, since it's not refcounted */
2926 struct task_struct
*latest_task
= NULL
;
2927 struct ptr_heap tmp_heap
;
2928 struct ptr_heap
*heap
;
2929 struct timespec latest_time
= { 0, 0 };
2932 /* The caller supplied our heap and pre-allocated its memory */
2934 heap
->gt
= &started_after
;
2936 /* We need to allocate our own heap memory */
2938 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
2940 /* cannot allocate the heap */
2946 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2947 * to determine which are of interest, and using the scanner's
2948 * "process_task" callback to process any of them that need an update.
2949 * Since we don't want to hold any locks during the task updates,
2950 * gather tasks to be processed in a heap structure.
2951 * The heap is sorted by descending task start time.
2952 * If the statically-sized heap fills up, we overflow tasks that
2953 * started later, and in future iterations only consider tasks that
2954 * started after the latest task in the previous pass. This
2955 * guarantees forward progress and that we don't miss any tasks.
2958 cgroup_iter_start(scan
->cg
, &it
);
2959 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
2961 * Only affect tasks that qualify per the caller's callback,
2962 * if he provided one
2964 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
2967 * Only process tasks that started after the last task
2970 if (!started_after_time(p
, &latest_time
, latest_task
))
2972 dropped
= heap_insert(heap
, p
);
2973 if (dropped
== NULL
) {
2975 * The new task was inserted; the heap wasn't
2979 } else if (dropped
!= p
) {
2981 * The new task was inserted, and pushed out a
2985 put_task_struct(dropped
);
2988 * Else the new task was newer than anything already in
2989 * the heap and wasn't inserted
2992 cgroup_iter_end(scan
->cg
, &it
);
2995 for (i
= 0; i
< heap
->size
; i
++) {
2996 struct task_struct
*q
= heap
->ptrs
[i
];
2998 latest_time
= q
->start_time
;
3001 /* Process the task per the caller's callback */
3002 scan
->process_task(q
, scan
);
3006 * If we had to process any tasks at all, scan again
3007 * in case some of them were in the middle of forking
3008 * children that didn't get processed.
3009 * Not the most efficient way to do it, but it avoids
3010 * having to take callback_mutex in the fork path
3014 if (heap
== &tmp_heap
)
3015 heap_free(&tmp_heap
);
3020 * Stuff for reading the 'tasks'/'procs' files.
3022 * Reading this file can return large amounts of data if a cgroup has
3023 * *lots* of attached tasks. So it may need several calls to read(),
3024 * but we cannot guarantee that the information we produce is correct
3025 * unless we produce it entirely atomically.
3030 * The following two functions "fix" the issue where there are more pids
3031 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3032 * TODO: replace with a kernel-wide solution to this problem
3034 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3035 static void *pidlist_allocate(int count
)
3037 if (PIDLIST_TOO_LARGE(count
))
3038 return vmalloc(count
* sizeof(pid_t
));
3040 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3042 static void pidlist_free(void *p
)
3044 if (is_vmalloc_addr(p
))
3049 static void *pidlist_resize(void *p
, int newcount
)
3052 /* note: if new alloc fails, old p will still be valid either way */
3053 if (is_vmalloc_addr(p
)) {
3054 newlist
= vmalloc(newcount
* sizeof(pid_t
));
3057 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
3060 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
3066 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3067 * If the new stripped list is sufficiently smaller and there's enough memory
3068 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3069 * number of unique elements.
3071 /* is the size difference enough that we should re-allocate the array? */
3072 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3073 static int pidlist_uniq(pid_t
**p
, int length
)
3080 * we presume the 0th element is unique, so i starts at 1. trivial
3081 * edge cases first; no work needs to be done for either
3083 if (length
== 0 || length
== 1)
3085 /* src and dest walk down the list; dest counts unique elements */
3086 for (src
= 1; src
< length
; src
++) {
3087 /* find next unique element */
3088 while (list
[src
] == list
[src
-1]) {
3093 /* dest always points to where the next unique element goes */
3094 list
[dest
] = list
[src
];
3099 * if the length difference is large enough, we want to allocate a
3100 * smaller buffer to save memory. if this fails due to out of memory,
3101 * we'll just stay with what we've got.
3103 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
3104 newlist
= pidlist_resize(list
, dest
);
3111 static int cmppid(const void *a
, const void *b
)
3113 return *(pid_t
*)a
- *(pid_t
*)b
;
3117 * find the appropriate pidlist for our purpose (given procs vs tasks)
3118 * returns with the lock on that pidlist already held, and takes care
3119 * of the use count, or returns NULL with no locks held if we're out of
3122 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3123 enum cgroup_filetype type
)
3125 struct cgroup_pidlist
*l
;
3126 /* don't need task_nsproxy() if we're looking at ourself */
3127 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
3130 * We can't drop the pidlist_mutex before taking the l->mutex in case
3131 * the last ref-holder is trying to remove l from the list at the same
3132 * time. Holding the pidlist_mutex precludes somebody taking whichever
3133 * list we find out from under us - compare release_pid_array().
3135 mutex_lock(&cgrp
->pidlist_mutex
);
3136 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3137 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3138 /* make sure l doesn't vanish out from under us */
3139 down_write(&l
->mutex
);
3140 mutex_unlock(&cgrp
->pidlist_mutex
);
3144 /* entry not found; create a new one */
3145 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3147 mutex_unlock(&cgrp
->pidlist_mutex
);
3150 init_rwsem(&l
->mutex
);
3151 down_write(&l
->mutex
);
3153 l
->key
.ns
= get_pid_ns(ns
);
3154 l
->use_count
= 0; /* don't increment here */
3157 list_add(&l
->links
, &cgrp
->pidlists
);
3158 mutex_unlock(&cgrp
->pidlist_mutex
);
3163 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3165 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3166 struct cgroup_pidlist
**lp
)
3170 int pid
, n
= 0; /* used for populating the array */
3171 struct cgroup_iter it
;
3172 struct task_struct
*tsk
;
3173 struct cgroup_pidlist
*l
;
3176 * If cgroup gets more users after we read count, we won't have
3177 * enough space - tough. This race is indistinguishable to the
3178 * caller from the case that the additional cgroup users didn't
3179 * show up until sometime later on.
3181 length
= cgroup_task_count(cgrp
);
3182 array
= pidlist_allocate(length
);
3185 /* now, populate the array */
3186 cgroup_iter_start(cgrp
, &it
);
3187 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3188 if (unlikely(n
== length
))
3190 /* get tgid or pid for procs or tasks file respectively */
3191 if (type
== CGROUP_FILE_PROCS
)
3192 pid
= task_tgid_vnr(tsk
);
3194 pid
= task_pid_vnr(tsk
);
3195 if (pid
> 0) /* make sure to only use valid results */
3198 cgroup_iter_end(cgrp
, &it
);
3200 /* now sort & (if procs) strip out duplicates */
3201 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3202 if (type
== CGROUP_FILE_PROCS
)
3203 length
= pidlist_uniq(&array
, length
);
3204 l
= cgroup_pidlist_find(cgrp
, type
);
3206 pidlist_free(array
);
3209 /* store array, freeing old if necessary - lock already held */
3210 pidlist_free(l
->list
);
3214 up_write(&l
->mutex
);
3220 * cgroupstats_build - build and fill cgroupstats
3221 * @stats: cgroupstats to fill information into
3222 * @dentry: A dentry entry belonging to the cgroup for which stats have
3225 * Build and fill cgroupstats so that taskstats can export it to user
3228 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3231 struct cgroup
*cgrp
;
3232 struct cgroup_iter it
;
3233 struct task_struct
*tsk
;
3236 * Validate dentry by checking the superblock operations,
3237 * and make sure it's a directory.
3239 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3240 !S_ISDIR(dentry
->d_inode
->i_mode
))
3244 cgrp
= dentry
->d_fsdata
;
3246 cgroup_iter_start(cgrp
, &it
);
3247 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3248 switch (tsk
->state
) {
3250 stats
->nr_running
++;
3252 case TASK_INTERRUPTIBLE
:
3253 stats
->nr_sleeping
++;
3255 case TASK_UNINTERRUPTIBLE
:
3256 stats
->nr_uninterruptible
++;
3259 stats
->nr_stopped
++;
3262 if (delayacct_is_task_waiting_on_io(tsk
))
3263 stats
->nr_io_wait
++;
3267 cgroup_iter_end(cgrp
, &it
);
3275 * seq_file methods for the tasks/procs files. The seq_file position is the
3276 * next pid to display; the seq_file iterator is a pointer to the pid
3277 * in the cgroup->l->list array.
3280 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3283 * Initially we receive a position value that corresponds to
3284 * one more than the last pid shown (or 0 on the first call or
3285 * after a seek to the start). Use a binary-search to find the
3286 * next pid to display, if any
3288 struct cgroup_pidlist
*l
= s
->private;
3289 int index
= 0, pid
= *pos
;
3292 down_read(&l
->mutex
);
3294 int end
= l
->length
;
3296 while (index
< end
) {
3297 int mid
= (index
+ end
) / 2;
3298 if (l
->list
[mid
] == pid
) {
3301 } else if (l
->list
[mid
] <= pid
)
3307 /* If we're off the end of the array, we're done */
3308 if (index
>= l
->length
)
3310 /* Update the abstract position to be the actual pid that we found */
3311 iter
= l
->list
+ index
;
3316 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3318 struct cgroup_pidlist
*l
= s
->private;
3322 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3324 struct cgroup_pidlist
*l
= s
->private;
3326 pid_t
*end
= l
->list
+ l
->length
;
3328 * Advance to the next pid in the array. If this goes off the
3340 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3342 return seq_printf(s
, "%d\n", *(int *)v
);
3346 * seq_operations functions for iterating on pidlists through seq_file -
3347 * independent of whether it's tasks or procs
3349 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3350 .start
= cgroup_pidlist_start
,
3351 .stop
= cgroup_pidlist_stop
,
3352 .next
= cgroup_pidlist_next
,
3353 .show
= cgroup_pidlist_show
,
3356 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3359 * the case where we're the last user of this particular pidlist will
3360 * have us remove it from the cgroup's list, which entails taking the
3361 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3362 * pidlist_mutex, we have to take pidlist_mutex first.
3364 mutex_lock(&l
->owner
->pidlist_mutex
);
3365 down_write(&l
->mutex
);
3366 BUG_ON(!l
->use_count
);
3367 if (!--l
->use_count
) {
3368 /* we're the last user if refcount is 0; remove and free */
3369 list_del(&l
->links
);
3370 mutex_unlock(&l
->owner
->pidlist_mutex
);
3371 pidlist_free(l
->list
);
3372 put_pid_ns(l
->key
.ns
);
3373 up_write(&l
->mutex
);
3377 mutex_unlock(&l
->owner
->pidlist_mutex
);
3378 up_write(&l
->mutex
);
3381 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3383 struct cgroup_pidlist
*l
;
3384 if (!(file
->f_mode
& FMODE_READ
))
3387 * the seq_file will only be initialized if the file was opened for
3388 * reading; hence we check if it's not null only in that case.
3390 l
= ((struct seq_file
*)file
->private_data
)->private;
3391 cgroup_release_pid_array(l
);
3392 return seq_release(inode
, file
);
3395 static const struct file_operations cgroup_pidlist_operations
= {
3397 .llseek
= seq_lseek
,
3398 .write
= cgroup_file_write
,
3399 .release
= cgroup_pidlist_release
,
3403 * The following functions handle opens on a file that displays a pidlist
3404 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3407 /* helper function for the two below it */
3408 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3410 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3411 struct cgroup_pidlist
*l
;
3414 /* Nothing to do for write-only files */
3415 if (!(file
->f_mode
& FMODE_READ
))
3418 /* have the array populated */
3419 retval
= pidlist_array_load(cgrp
, type
, &l
);
3422 /* configure file information */
3423 file
->f_op
= &cgroup_pidlist_operations
;
3425 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3427 cgroup_release_pid_array(l
);
3430 ((struct seq_file
*)file
->private_data
)->private = l
;
3433 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3435 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3437 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3439 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3442 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3445 return notify_on_release(cgrp
);
3448 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3452 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3454 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3456 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3461 * Unregister event and free resources.
3463 * Gets called from workqueue.
3465 static void cgroup_event_remove(struct work_struct
*work
)
3467 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3469 struct cgroup
*cgrp
= event
->cgrp
;
3471 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3473 eventfd_ctx_put(event
->eventfd
);
3479 * Gets called on POLLHUP on eventfd when user closes it.
3481 * Called with wqh->lock held and interrupts disabled.
3483 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3484 int sync
, void *key
)
3486 struct cgroup_event
*event
= container_of(wait
,
3487 struct cgroup_event
, wait
);
3488 struct cgroup
*cgrp
= event
->cgrp
;
3489 unsigned long flags
= (unsigned long)key
;
3491 if (flags
& POLLHUP
) {
3492 __remove_wait_queue(event
->wqh
, &event
->wait
);
3493 spin_lock(&cgrp
->event_list_lock
);
3494 list_del(&event
->list
);
3495 spin_unlock(&cgrp
->event_list_lock
);
3497 * We are in atomic context, but cgroup_event_remove() may
3498 * sleep, so we have to call it in workqueue.
3500 schedule_work(&event
->remove
);
3506 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3507 wait_queue_head_t
*wqh
, poll_table
*pt
)
3509 struct cgroup_event
*event
= container_of(pt
,
3510 struct cgroup_event
, pt
);
3513 add_wait_queue(wqh
, &event
->wait
);
3517 * Parse input and register new cgroup event handler.
3519 * Input must be in format '<event_fd> <control_fd> <args>'.
3520 * Interpretation of args is defined by control file implementation.
3522 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3525 struct cgroup_event
*event
= NULL
;
3526 unsigned int efd
, cfd
;
3527 struct file
*efile
= NULL
;
3528 struct file
*cfile
= NULL
;
3532 efd
= simple_strtoul(buffer
, &endp
, 10);
3537 cfd
= simple_strtoul(buffer
, &endp
, 10);
3538 if ((*endp
!= ' ') && (*endp
!= '\0'))
3542 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3546 INIT_LIST_HEAD(&event
->list
);
3547 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3548 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3549 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3551 efile
= eventfd_fget(efd
);
3552 if (IS_ERR(efile
)) {
3553 ret
= PTR_ERR(efile
);
3557 event
->eventfd
= eventfd_ctx_fileget(efile
);
3558 if (IS_ERR(event
->eventfd
)) {
3559 ret
= PTR_ERR(event
->eventfd
);
3569 /* the process need read permission on control file */
3570 /* AV: shouldn't we check that it's been opened for read instead? */
3571 ret
= inode_permission(cfile
->f_path
.dentry
->d_inode
, MAY_READ
);
3575 event
->cft
= __file_cft(cfile
);
3576 if (IS_ERR(event
->cft
)) {
3577 ret
= PTR_ERR(event
->cft
);
3581 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3586 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3587 event
->eventfd
, buffer
);
3591 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3592 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3598 * Events should be removed after rmdir of cgroup directory, but before
3599 * destroying subsystem state objects. Let's take reference to cgroup
3600 * directory dentry to do that.
3604 spin_lock(&cgrp
->event_list_lock
);
3605 list_add(&event
->list
, &cgrp
->event_list
);
3606 spin_unlock(&cgrp
->event_list_lock
);
3617 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3618 eventfd_ctx_put(event
->eventfd
);
3620 if (!IS_ERR_OR_NULL(efile
))
3628 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3631 return clone_children(cgrp
);
3634 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3639 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3641 clear_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3646 * for the common functions, 'private' gives the type of file
3648 /* for hysterical raisins, we can't put this on the older files */
3649 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3650 static struct cftype files
[] = {
3653 .open
= cgroup_tasks_open
,
3654 .write_u64
= cgroup_tasks_write
,
3655 .release
= cgroup_pidlist_release
,
3656 .mode
= S_IRUGO
| S_IWUSR
,
3659 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3660 .open
= cgroup_procs_open
,
3661 .write_u64
= cgroup_procs_write
,
3662 .release
= cgroup_pidlist_release
,
3663 .mode
= S_IRUGO
| S_IWUSR
,
3666 .name
= "notify_on_release",
3667 .read_u64
= cgroup_read_notify_on_release
,
3668 .write_u64
= cgroup_write_notify_on_release
,
3671 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3672 .write_string
= cgroup_write_event_control
,
3676 .name
= "cgroup.clone_children",
3677 .read_u64
= cgroup_clone_children_read
,
3678 .write_u64
= cgroup_clone_children_write
,
3682 static struct cftype cft_release_agent
= {
3683 .name
= "release_agent",
3684 .read_seq_string
= cgroup_release_agent_show
,
3685 .write_string
= cgroup_release_agent_write
,
3686 .max_write_len
= PATH_MAX
,
3689 static int cgroup_populate_dir(struct cgroup
*cgrp
)
3692 struct cgroup_subsys
*ss
;
3694 /* First clear out any existing files */
3695 cgroup_clear_directory(cgrp
->dentry
);
3697 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
3701 if (cgrp
== cgrp
->top_cgroup
) {
3702 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
3706 for_each_subsys(cgrp
->root
, ss
) {
3707 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
3710 /* This cgroup is ready now */
3711 for_each_subsys(cgrp
->root
, ss
) {
3712 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3714 * Update id->css pointer and make this css visible from
3715 * CSS ID functions. This pointer will be dereferened
3716 * from RCU-read-side without locks.
3719 rcu_assign_pointer(css
->id
->css
, css
);
3725 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
3726 struct cgroup_subsys
*ss
,
3727 struct cgroup
*cgrp
)
3730 atomic_set(&css
->refcnt
, 1);
3733 if (cgrp
== dummytop
)
3734 set_bit(CSS_ROOT
, &css
->flags
);
3735 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
3736 cgrp
->subsys
[ss
->subsys_id
] = css
;
3739 static void cgroup_lock_hierarchy(struct cgroupfs_root
*root
)
3741 /* We need to take each hierarchy_mutex in a consistent order */
3745 * No worry about a race with rebind_subsystems that might mess up the
3746 * locking order, since both parties are under cgroup_mutex.
3748 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3749 struct cgroup_subsys
*ss
= subsys
[i
];
3752 if (ss
->root
== root
)
3753 mutex_lock(&ss
->hierarchy_mutex
);
3757 static void cgroup_unlock_hierarchy(struct cgroupfs_root
*root
)
3761 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3762 struct cgroup_subsys
*ss
= subsys
[i
];
3765 if (ss
->root
== root
)
3766 mutex_unlock(&ss
->hierarchy_mutex
);
3771 * cgroup_create - create a cgroup
3772 * @parent: cgroup that will be parent of the new cgroup
3773 * @dentry: dentry of the new cgroup
3774 * @mode: mode to set on new inode
3776 * Must be called with the mutex on the parent inode held
3778 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
3781 struct cgroup
*cgrp
;
3782 struct cgroupfs_root
*root
= parent
->root
;
3784 struct cgroup_subsys
*ss
;
3785 struct super_block
*sb
= root
->sb
;
3787 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3791 /* Grab a reference on the superblock so the hierarchy doesn't
3792 * get deleted on unmount if there are child cgroups. This
3793 * can be done outside cgroup_mutex, since the sb can't
3794 * disappear while someone has an open control file on the
3796 atomic_inc(&sb
->s_active
);
3798 mutex_lock(&cgroup_mutex
);
3800 init_cgroup_housekeeping(cgrp
);
3802 cgrp
->parent
= parent
;
3803 cgrp
->root
= parent
->root
;
3804 cgrp
->top_cgroup
= parent
->top_cgroup
;
3806 if (notify_on_release(parent
))
3807 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3809 if (clone_children(parent
))
3810 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3812 for_each_subsys(root
, ss
) {
3813 struct cgroup_subsys_state
*css
= ss
->create(ss
, cgrp
);
3819 init_cgroup_css(css
, ss
, cgrp
);
3821 err
= alloc_css_id(ss
, parent
, cgrp
);
3825 /* At error, ->destroy() callback has to free assigned ID. */
3826 if (clone_children(parent
) && ss
->post_clone
)
3827 ss
->post_clone(ss
, cgrp
);
3830 cgroup_lock_hierarchy(root
);
3831 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
3832 cgroup_unlock_hierarchy(root
);
3833 root
->number_of_cgroups
++;
3835 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
3839 /* The cgroup directory was pre-locked for us */
3840 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
3842 err
= cgroup_populate_dir(cgrp
);
3843 /* If err < 0, we have a half-filled directory - oh well ;) */
3845 mutex_unlock(&cgroup_mutex
);
3846 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
3852 cgroup_lock_hierarchy(root
);
3853 list_del(&cgrp
->sibling
);
3854 cgroup_unlock_hierarchy(root
);
3855 root
->number_of_cgroups
--;
3859 for_each_subsys(root
, ss
) {
3860 if (cgrp
->subsys
[ss
->subsys_id
])
3861 ss
->destroy(ss
, cgrp
);
3864 mutex_unlock(&cgroup_mutex
);
3866 /* Release the reference count that we took on the superblock */
3867 deactivate_super(sb
);
3873 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3875 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
3877 /* the vfs holds inode->i_mutex already */
3878 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
3881 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
3883 /* Check the reference count on each subsystem. Since we
3884 * already established that there are no tasks in the
3885 * cgroup, if the css refcount is also 1, then there should
3886 * be no outstanding references, so the subsystem is safe to
3887 * destroy. We scan across all subsystems rather than using
3888 * the per-hierarchy linked list of mounted subsystems since
3889 * we can be called via check_for_release() with no
3890 * synchronization other than RCU, and the subsystem linked
3891 * list isn't RCU-safe */
3894 * We won't need to lock the subsys array, because the subsystems
3895 * we're concerned about aren't going anywhere since our cgroup root
3896 * has a reference on them.
3898 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3899 struct cgroup_subsys
*ss
= subsys
[i
];
3900 struct cgroup_subsys_state
*css
;
3901 /* Skip subsystems not present or not in this hierarchy */
3902 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
3904 css
= cgrp
->subsys
[ss
->subsys_id
];
3905 /* When called from check_for_release() it's possible
3906 * that by this point the cgroup has been removed
3907 * and the css deleted. But a false-positive doesn't
3908 * matter, since it can only happen if the cgroup
3909 * has been deleted and hence no longer needs the
3910 * release agent to be called anyway. */
3911 if (css
&& (atomic_read(&css
->refcnt
) > 1))
3918 * Atomically mark all (or else none) of the cgroup's CSS objects as
3919 * CSS_REMOVED. Return true on success, or false if the cgroup has
3920 * busy subsystems. Call with cgroup_mutex held
3923 static int cgroup_clear_css_refs(struct cgroup
*cgrp
)
3925 struct cgroup_subsys
*ss
;
3926 unsigned long flags
;
3927 bool failed
= false;
3928 local_irq_save(flags
);
3929 for_each_subsys(cgrp
->root
, ss
) {
3930 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3933 /* We can only remove a CSS with a refcnt==1 */
3934 refcnt
= atomic_read(&css
->refcnt
);
3941 * Drop the refcnt to 0 while we check other
3942 * subsystems. This will cause any racing
3943 * css_tryget() to spin until we set the
3944 * CSS_REMOVED bits or abort
3946 if (atomic_cmpxchg(&css
->refcnt
, refcnt
, 0) == refcnt
)
3952 for_each_subsys(cgrp
->root
, ss
) {
3953 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3956 * Restore old refcnt if we previously managed
3957 * to clear it from 1 to 0
3959 if (!atomic_read(&css
->refcnt
))
3960 atomic_set(&css
->refcnt
, 1);
3962 /* Commit the fact that the CSS is removed */
3963 set_bit(CSS_REMOVED
, &css
->flags
);
3966 local_irq_restore(flags
);
3970 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
3972 struct cgroup
*cgrp
= dentry
->d_fsdata
;
3974 struct cgroup
*parent
;
3976 struct cgroup_event
*event
, *tmp
;
3979 /* the vfs holds both inode->i_mutex already */
3981 mutex_lock(&cgroup_mutex
);
3982 if (atomic_read(&cgrp
->count
) != 0) {
3983 mutex_unlock(&cgroup_mutex
);
3986 if (!list_empty(&cgrp
->children
)) {
3987 mutex_unlock(&cgroup_mutex
);
3990 mutex_unlock(&cgroup_mutex
);
3993 * In general, subsystem has no css->refcnt after pre_destroy(). But
3994 * in racy cases, subsystem may have to get css->refcnt after
3995 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3996 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3997 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3998 * and subsystem's reference count handling. Please see css_get/put
3999 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4001 set_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4004 * Call pre_destroy handlers of subsys. Notify subsystems
4005 * that rmdir() request comes.
4007 ret
= cgroup_call_pre_destroy(cgrp
);
4009 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4013 mutex_lock(&cgroup_mutex
);
4014 parent
= cgrp
->parent
;
4015 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
)) {
4016 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4017 mutex_unlock(&cgroup_mutex
);
4020 prepare_to_wait(&cgroup_rmdir_waitq
, &wait
, TASK_INTERRUPTIBLE
);
4021 if (!cgroup_clear_css_refs(cgrp
)) {
4022 mutex_unlock(&cgroup_mutex
);
4024 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4025 * prepare_to_wait(), we need to check this flag.
4027 if (test_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
))
4029 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4030 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4031 if (signal_pending(current
))
4035 /* NO css_tryget() can success after here. */
4036 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4037 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4039 raw_spin_lock(&release_list_lock
);
4040 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
4041 if (!list_empty(&cgrp
->release_list
))
4042 list_del_init(&cgrp
->release_list
);
4043 raw_spin_unlock(&release_list_lock
);
4045 cgroup_lock_hierarchy(cgrp
->root
);
4046 /* delete this cgroup from parent->children */
4047 list_del_init(&cgrp
->sibling
);
4048 cgroup_unlock_hierarchy(cgrp
->root
);
4050 d
= dget(cgrp
->dentry
);
4052 cgroup_d_remove_dir(d
);
4055 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4056 check_for_release(parent
);
4059 * Unregister events and notify userspace.
4060 * Notify userspace about cgroup removing only after rmdir of cgroup
4061 * directory to avoid race between userspace and kernelspace
4063 spin_lock(&cgrp
->event_list_lock
);
4064 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4065 list_del(&event
->list
);
4066 remove_wait_queue(event
->wqh
, &event
->wait
);
4067 eventfd_signal(event
->eventfd
, 1);
4068 schedule_work(&event
->remove
);
4070 spin_unlock(&cgrp
->event_list_lock
);
4072 mutex_unlock(&cgroup_mutex
);
4076 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4078 struct cgroup_subsys_state
*css
;
4080 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4082 /* Create the top cgroup state for this subsystem */
4083 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4084 ss
->root
= &rootnode
;
4085 css
= ss
->create(ss
, dummytop
);
4086 /* We don't handle early failures gracefully */
4087 BUG_ON(IS_ERR(css
));
4088 init_cgroup_css(css
, ss
, dummytop
);
4090 /* Update the init_css_set to contain a subsys
4091 * pointer to this state - since the subsystem is
4092 * newly registered, all tasks and hence the
4093 * init_css_set is in the subsystem's top cgroup. */
4094 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
4096 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4098 /* At system boot, before all subsystems have been
4099 * registered, no tasks have been forked, so we don't
4100 * need to invoke fork callbacks here. */
4101 BUG_ON(!list_empty(&init_task
.tasks
));
4103 mutex_init(&ss
->hierarchy_mutex
);
4104 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4107 /* this function shouldn't be used with modular subsystems, since they
4108 * need to register a subsys_id, among other things */
4113 * cgroup_load_subsys: load and register a modular subsystem at runtime
4114 * @ss: the subsystem to load
4116 * This function should be called in a modular subsystem's initcall. If the
4117 * subsystem is built as a module, it will be assigned a new subsys_id and set
4118 * up for use. If the subsystem is built-in anyway, work is delegated to the
4119 * simpler cgroup_init_subsys.
4121 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4124 struct cgroup_subsys_state
*css
;
4126 /* check name and function validity */
4127 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4128 ss
->create
== NULL
|| ss
->destroy
== NULL
)
4132 * we don't support callbacks in modular subsystems. this check is
4133 * before the ss->module check for consistency; a subsystem that could
4134 * be a module should still have no callbacks even if the user isn't
4135 * compiling it as one.
4137 if (ss
->fork
|| ss
->exit
)
4141 * an optionally modular subsystem is built-in: we want to do nothing,
4142 * since cgroup_init_subsys will have already taken care of it.
4144 if (ss
->module
== NULL
) {
4145 /* a few sanity checks */
4146 BUG_ON(ss
->subsys_id
>= CGROUP_BUILTIN_SUBSYS_COUNT
);
4147 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4152 * need to register a subsys id before anything else - for example,
4153 * init_cgroup_css needs it.
4155 mutex_lock(&cgroup_mutex
);
4156 /* find the first empty slot in the array */
4157 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4158 if (subsys
[i
] == NULL
)
4161 if (i
== CGROUP_SUBSYS_COUNT
) {
4162 /* maximum number of subsystems already registered! */
4163 mutex_unlock(&cgroup_mutex
);
4166 /* assign ourselves the subsys_id */
4171 * no ss->create seems to need anything important in the ss struct, so
4172 * this can happen first (i.e. before the rootnode attachment).
4174 css
= ss
->create(ss
, dummytop
);
4176 /* failure case - need to deassign the subsys[] slot. */
4178 mutex_unlock(&cgroup_mutex
);
4179 return PTR_ERR(css
);
4182 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4183 ss
->root
= &rootnode
;
4185 /* our new subsystem will be attached to the dummy hierarchy. */
4186 init_cgroup_css(css
, ss
, dummytop
);
4187 /* init_idr must be after init_cgroup_css because it sets css->id. */
4189 int ret
= cgroup_init_idr(ss
, css
);
4191 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4192 ss
->destroy(ss
, dummytop
);
4194 mutex_unlock(&cgroup_mutex
);
4200 * Now we need to entangle the css into the existing css_sets. unlike
4201 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4202 * will need a new pointer to it; done by iterating the css_set_table.
4203 * furthermore, modifying the existing css_sets will corrupt the hash
4204 * table state, so each changed css_set will need its hash recomputed.
4205 * this is all done under the css_set_lock.
4207 write_lock(&css_set_lock
);
4208 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
4210 struct hlist_node
*node
, *tmp
;
4211 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
4213 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
4214 /* skip entries that we already rehashed */
4215 if (cg
->subsys
[ss
->subsys_id
])
4217 /* remove existing entry */
4218 hlist_del(&cg
->hlist
);
4220 cg
->subsys
[ss
->subsys_id
] = css
;
4221 /* recompute hash and restore entry */
4222 new_bucket
= css_set_hash(cg
->subsys
);
4223 hlist_add_head(&cg
->hlist
, new_bucket
);
4226 write_unlock(&css_set_lock
);
4228 mutex_init(&ss
->hierarchy_mutex
);
4229 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4233 mutex_unlock(&cgroup_mutex
);
4236 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4239 * cgroup_unload_subsys: unload a modular subsystem
4240 * @ss: the subsystem to unload
4242 * This function should be called in a modular subsystem's exitcall. When this
4243 * function is invoked, the refcount on the subsystem's module will be 0, so
4244 * the subsystem will not be attached to any hierarchy.
4246 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4248 struct cg_cgroup_link
*link
;
4249 struct hlist_head
*hhead
;
4251 BUG_ON(ss
->module
== NULL
);
4254 * we shouldn't be called if the subsystem is in use, and the use of
4255 * try_module_get in parse_cgroupfs_options should ensure that it
4256 * doesn't start being used while we're killing it off.
4258 BUG_ON(ss
->root
!= &rootnode
);
4260 mutex_lock(&cgroup_mutex
);
4261 /* deassign the subsys_id */
4262 BUG_ON(ss
->subsys_id
< CGROUP_BUILTIN_SUBSYS_COUNT
);
4263 subsys
[ss
->subsys_id
] = NULL
;
4265 /* remove subsystem from rootnode's list of subsystems */
4266 list_del_init(&ss
->sibling
);
4269 * disentangle the css from all css_sets attached to the dummytop. as
4270 * in loading, we need to pay our respects to the hashtable gods.
4272 write_lock(&css_set_lock
);
4273 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4274 struct css_set
*cg
= link
->cg
;
4276 hlist_del(&cg
->hlist
);
4277 BUG_ON(!cg
->subsys
[ss
->subsys_id
]);
4278 cg
->subsys
[ss
->subsys_id
] = NULL
;
4279 hhead
= css_set_hash(cg
->subsys
);
4280 hlist_add_head(&cg
->hlist
, hhead
);
4282 write_unlock(&css_set_lock
);
4285 * remove subsystem's css from the dummytop and free it - need to free
4286 * before marking as null because ss->destroy needs the cgrp->subsys
4287 * pointer to find their state. note that this also takes care of
4288 * freeing the css_id.
4290 ss
->destroy(ss
, dummytop
);
4291 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4293 mutex_unlock(&cgroup_mutex
);
4295 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4298 * cgroup_init_early - cgroup initialization at system boot
4300 * Initialize cgroups at system boot, and initialize any
4301 * subsystems that request early init.
4303 int __init
cgroup_init_early(void)
4306 atomic_set(&init_css_set
.refcount
, 1);
4307 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4308 INIT_LIST_HEAD(&init_css_set
.tasks
);
4309 INIT_HLIST_NODE(&init_css_set
.hlist
);
4311 init_cgroup_root(&rootnode
);
4313 init_task
.cgroups
= &init_css_set
;
4315 init_css_set_link
.cg
= &init_css_set
;
4316 init_css_set_link
.cgrp
= dummytop
;
4317 list_add(&init_css_set_link
.cgrp_link_list
,
4318 &rootnode
.top_cgroup
.css_sets
);
4319 list_add(&init_css_set_link
.cg_link_list
,
4320 &init_css_set
.cg_links
);
4322 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
4323 INIT_HLIST_HEAD(&css_set_table
[i
]);
4325 /* at bootup time, we don't worry about modular subsystems */
4326 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4327 struct cgroup_subsys
*ss
= subsys
[i
];
4330 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4331 BUG_ON(!ss
->create
);
4332 BUG_ON(!ss
->destroy
);
4333 if (ss
->subsys_id
!= i
) {
4334 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4335 ss
->name
, ss
->subsys_id
);
4340 cgroup_init_subsys(ss
);
4346 * cgroup_init - cgroup initialization
4348 * Register cgroup filesystem and /proc file, and initialize
4349 * any subsystems that didn't request early init.
4351 int __init
cgroup_init(void)
4355 struct hlist_head
*hhead
;
4357 err
= bdi_init(&cgroup_backing_dev_info
);
4361 /* at bootup time, we don't worry about modular subsystems */
4362 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4363 struct cgroup_subsys
*ss
= subsys
[i
];
4364 if (!ss
->early_init
)
4365 cgroup_init_subsys(ss
);
4367 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4370 /* Add init_css_set to the hash table */
4371 hhead
= css_set_hash(init_css_set
.subsys
);
4372 hlist_add_head(&init_css_set
.hlist
, hhead
);
4373 BUG_ON(!init_root_id(&rootnode
));
4375 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4381 err
= register_filesystem(&cgroup_fs_type
);
4383 kobject_put(cgroup_kobj
);
4387 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4391 bdi_destroy(&cgroup_backing_dev_info
);
4397 * proc_cgroup_show()
4398 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4399 * - Used for /proc/<pid>/cgroup.
4400 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4401 * doesn't really matter if tsk->cgroup changes after we read it,
4402 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4403 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4404 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4405 * cgroup to top_cgroup.
4408 /* TODO: Use a proper seq_file iterator */
4409 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4412 struct task_struct
*tsk
;
4415 struct cgroupfs_root
*root
;
4418 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4424 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4430 mutex_lock(&cgroup_mutex
);
4432 for_each_active_root(root
) {
4433 struct cgroup_subsys
*ss
;
4434 struct cgroup
*cgrp
;
4437 seq_printf(m
, "%d:", root
->hierarchy_id
);
4438 for_each_subsys(root
, ss
)
4439 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4440 if (strlen(root
->name
))
4441 seq_printf(m
, "%sname=%s", count
? "," : "",
4444 cgrp
= task_cgroup_from_root(tsk
, root
);
4445 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4453 mutex_unlock(&cgroup_mutex
);
4454 put_task_struct(tsk
);
4461 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4463 struct pid
*pid
= PROC_I(inode
)->pid
;
4464 return single_open(file
, proc_cgroup_show
, pid
);
4467 const struct file_operations proc_cgroup_operations
= {
4468 .open
= cgroup_open
,
4470 .llseek
= seq_lseek
,
4471 .release
= single_release
,
4474 /* Display information about each subsystem and each hierarchy */
4475 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4479 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4481 * ideally we don't want subsystems moving around while we do this.
4482 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4483 * subsys/hierarchy state.
4485 mutex_lock(&cgroup_mutex
);
4486 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4487 struct cgroup_subsys
*ss
= subsys
[i
];
4490 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4491 ss
->name
, ss
->root
->hierarchy_id
,
4492 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4494 mutex_unlock(&cgroup_mutex
);
4498 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4500 return single_open(file
, proc_cgroupstats_show
, NULL
);
4503 static const struct file_operations proc_cgroupstats_operations
= {
4504 .open
= cgroupstats_open
,
4506 .llseek
= seq_lseek
,
4507 .release
= single_release
,
4511 * cgroup_fork - attach newly forked task to its parents cgroup.
4512 * @child: pointer to task_struct of forking parent process.
4514 * Description: A task inherits its parent's cgroup at fork().
4516 * A pointer to the shared css_set was automatically copied in
4517 * fork.c by dup_task_struct(). However, we ignore that copy, since
4518 * it was not made under the protection of RCU or cgroup_mutex, so
4519 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4520 * have already changed current->cgroups, allowing the previously
4521 * referenced cgroup group to be removed and freed.
4523 * At the point that cgroup_fork() is called, 'current' is the parent
4524 * task, and the passed argument 'child' points to the child task.
4526 void cgroup_fork(struct task_struct
*child
)
4529 child
->cgroups
= current
->cgroups
;
4530 get_css_set(child
->cgroups
);
4531 task_unlock(current
);
4532 INIT_LIST_HEAD(&child
->cg_list
);
4536 * cgroup_fork_callbacks - run fork callbacks
4537 * @child: the new task
4539 * Called on a new task very soon before adding it to the
4540 * tasklist. No need to take any locks since no-one can
4541 * be operating on this task.
4543 void cgroup_fork_callbacks(struct task_struct
*child
)
4545 if (need_forkexit_callback
) {
4548 * forkexit callbacks are only supported for builtin
4549 * subsystems, and the builtin section of the subsys array is
4550 * immutable, so we don't need to lock the subsys array here.
4552 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4553 struct cgroup_subsys
*ss
= subsys
[i
];
4555 ss
->fork(ss
, child
);
4561 * cgroup_post_fork - called on a new task after adding it to the task list
4562 * @child: the task in question
4564 * Adds the task to the list running through its css_set if necessary.
4565 * Has to be after the task is visible on the task list in case we race
4566 * with the first call to cgroup_iter_start() - to guarantee that the
4567 * new task ends up on its list.
4569 void cgroup_post_fork(struct task_struct
*child
)
4571 if (use_task_css_set_links
) {
4572 write_lock(&css_set_lock
);
4574 if (list_empty(&child
->cg_list
))
4575 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4577 write_unlock(&css_set_lock
);
4581 * cgroup_exit - detach cgroup from exiting task
4582 * @tsk: pointer to task_struct of exiting process
4583 * @run_callback: run exit callbacks?
4585 * Description: Detach cgroup from @tsk and release it.
4587 * Note that cgroups marked notify_on_release force every task in
4588 * them to take the global cgroup_mutex mutex when exiting.
4589 * This could impact scaling on very large systems. Be reluctant to
4590 * use notify_on_release cgroups where very high task exit scaling
4591 * is required on large systems.
4593 * the_top_cgroup_hack:
4595 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4597 * We call cgroup_exit() while the task is still competent to
4598 * handle notify_on_release(), then leave the task attached to the
4599 * root cgroup in each hierarchy for the remainder of its exit.
4601 * To do this properly, we would increment the reference count on
4602 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4603 * code we would add a second cgroup function call, to drop that
4604 * reference. This would just create an unnecessary hot spot on
4605 * the top_cgroup reference count, to no avail.
4607 * Normally, holding a reference to a cgroup without bumping its
4608 * count is unsafe. The cgroup could go away, or someone could
4609 * attach us to a different cgroup, decrementing the count on
4610 * the first cgroup that we never incremented. But in this case,
4611 * top_cgroup isn't going away, and either task has PF_EXITING set,
4612 * which wards off any cgroup_attach_task() attempts, or task is a failed
4613 * fork, never visible to cgroup_attach_task.
4615 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4621 * Unlink from the css_set task list if necessary.
4622 * Optimistically check cg_list before taking
4625 if (!list_empty(&tsk
->cg_list
)) {
4626 write_lock(&css_set_lock
);
4627 if (!list_empty(&tsk
->cg_list
))
4628 list_del_init(&tsk
->cg_list
);
4629 write_unlock(&css_set_lock
);
4632 /* Reassign the task to the init_css_set. */
4635 tsk
->cgroups
= &init_css_set
;
4637 if (run_callbacks
&& need_forkexit_callback
) {
4639 * modular subsystems can't use callbacks, so no need to lock
4642 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4643 struct cgroup_subsys
*ss
= subsys
[i
];
4645 struct cgroup
*old_cgrp
=
4646 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4647 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4648 ss
->exit(ss
, cgrp
, old_cgrp
, tsk
);
4655 put_css_set_taskexit(cg
);
4659 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4660 * @cgrp: the cgroup in question
4661 * @task: the task in question
4663 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4666 * If we are sending in dummytop, then presumably we are creating
4667 * the top cgroup in the subsystem.
4669 * Called only by the ns (nsproxy) cgroup.
4671 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
4674 struct cgroup
*target
;
4676 if (cgrp
== dummytop
)
4679 target
= task_cgroup_from_root(task
, cgrp
->root
);
4680 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
4681 cgrp
= cgrp
->parent
;
4682 ret
= (cgrp
== target
);
4686 static void check_for_release(struct cgroup
*cgrp
)
4688 /* All of these checks rely on RCU to keep the cgroup
4689 * structure alive */
4690 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
4691 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
4692 /* Control Group is currently removeable. If it's not
4693 * already queued for a userspace notification, queue
4695 int need_schedule_work
= 0;
4696 raw_spin_lock(&release_list_lock
);
4697 if (!cgroup_is_removed(cgrp
) &&
4698 list_empty(&cgrp
->release_list
)) {
4699 list_add(&cgrp
->release_list
, &release_list
);
4700 need_schedule_work
= 1;
4702 raw_spin_unlock(&release_list_lock
);
4703 if (need_schedule_work
)
4704 schedule_work(&release_agent_work
);
4708 /* Caller must verify that the css is not for root cgroup */
4709 void __css_put(struct cgroup_subsys_state
*css
, int count
)
4711 struct cgroup
*cgrp
= css
->cgroup
;
4714 val
= atomic_sub_return(count
, &css
->refcnt
);
4716 if (notify_on_release(cgrp
)) {
4717 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4718 check_for_release(cgrp
);
4720 cgroup_wakeup_rmdir_waiter(cgrp
);
4723 WARN_ON_ONCE(val
< 1);
4725 EXPORT_SYMBOL_GPL(__css_put
);
4728 * Notify userspace when a cgroup is released, by running the
4729 * configured release agent with the name of the cgroup (path
4730 * relative to the root of cgroup file system) as the argument.
4732 * Most likely, this user command will try to rmdir this cgroup.
4734 * This races with the possibility that some other task will be
4735 * attached to this cgroup before it is removed, or that some other
4736 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4737 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4738 * unused, and this cgroup will be reprieved from its death sentence,
4739 * to continue to serve a useful existence. Next time it's released,
4740 * we will get notified again, if it still has 'notify_on_release' set.
4742 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4743 * means only wait until the task is successfully execve()'d. The
4744 * separate release agent task is forked by call_usermodehelper(),
4745 * then control in this thread returns here, without waiting for the
4746 * release agent task. We don't bother to wait because the caller of
4747 * this routine has no use for the exit status of the release agent
4748 * task, so no sense holding our caller up for that.
4750 static void cgroup_release_agent(struct work_struct
*work
)
4752 BUG_ON(work
!= &release_agent_work
);
4753 mutex_lock(&cgroup_mutex
);
4754 raw_spin_lock(&release_list_lock
);
4755 while (!list_empty(&release_list
)) {
4756 char *argv
[3], *envp
[3];
4758 char *pathbuf
= NULL
, *agentbuf
= NULL
;
4759 struct cgroup
*cgrp
= list_entry(release_list
.next
,
4762 list_del_init(&cgrp
->release_list
);
4763 raw_spin_unlock(&release_list_lock
);
4764 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4767 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
4769 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
4774 argv
[i
++] = agentbuf
;
4775 argv
[i
++] = pathbuf
;
4779 /* minimal command environment */
4780 envp
[i
++] = "HOME=/";
4781 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4784 /* Drop the lock while we invoke the usermode helper,
4785 * since the exec could involve hitting disk and hence
4786 * be a slow process */
4787 mutex_unlock(&cgroup_mutex
);
4788 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
4789 mutex_lock(&cgroup_mutex
);
4793 raw_spin_lock(&release_list_lock
);
4795 raw_spin_unlock(&release_list_lock
);
4796 mutex_unlock(&cgroup_mutex
);
4799 static int __init
cgroup_disable(char *str
)
4804 while ((token
= strsep(&str
, ",")) != NULL
) {
4808 * cgroup_disable, being at boot time, can't know about module
4809 * subsystems, so we don't worry about them.
4811 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4812 struct cgroup_subsys
*ss
= subsys
[i
];
4814 if (!strcmp(token
, ss
->name
)) {
4816 printk(KERN_INFO
"Disabling %s control group"
4817 " subsystem\n", ss
->name
);
4824 __setup("cgroup_disable=", cgroup_disable
);
4827 * Functons for CSS ID.
4831 *To get ID other than 0, this should be called when !cgroup_is_removed().
4833 unsigned short css_id(struct cgroup_subsys_state
*css
)
4835 struct css_id
*cssid
;
4838 * This css_id() can return correct value when somone has refcnt
4839 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4840 * it's unchanged until freed.
4842 cssid
= rcu_dereference_check(css
->id
, atomic_read(&css
->refcnt
));
4848 EXPORT_SYMBOL_GPL(css_id
);
4850 unsigned short css_depth(struct cgroup_subsys_state
*css
)
4852 struct css_id
*cssid
;
4854 cssid
= rcu_dereference_check(css
->id
, atomic_read(&css
->refcnt
));
4857 return cssid
->depth
;
4860 EXPORT_SYMBOL_GPL(css_depth
);
4863 * css_is_ancestor - test "root" css is an ancestor of "child"
4864 * @child: the css to be tested.
4865 * @root: the css supporsed to be an ancestor of the child.
4867 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4868 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4869 * But, considering usual usage, the csses should be valid objects after test.
4870 * Assuming that the caller will do some action to the child if this returns
4871 * returns true, the caller must take "child";s reference count.
4872 * If "child" is valid object and this returns true, "root" is valid, too.
4875 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
4876 const struct cgroup_subsys_state
*root
)
4878 struct css_id
*child_id
;
4879 struct css_id
*root_id
;
4883 child_id
= rcu_dereference(child
->id
);
4884 root_id
= rcu_dereference(root
->id
);
4887 || (child_id
->depth
< root_id
->depth
)
4888 || (child_id
->stack
[root_id
->depth
] != root_id
->id
))
4894 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
4896 struct css_id
*id
= css
->id
;
4897 /* When this is called before css_id initialization, id can be NULL */
4901 BUG_ON(!ss
->use_id
);
4903 rcu_assign_pointer(id
->css
, NULL
);
4904 rcu_assign_pointer(css
->id
, NULL
);
4905 write_lock(&ss
->id_lock
);
4906 idr_remove(&ss
->idr
, id
->id
);
4907 write_unlock(&ss
->id_lock
);
4908 kfree_rcu(id
, rcu_head
);
4910 EXPORT_SYMBOL_GPL(free_css_id
);
4913 * This is called by init or create(). Then, calls to this function are
4914 * always serialized (By cgroup_mutex() at create()).
4917 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
4919 struct css_id
*newid
;
4920 int myid
, error
, size
;
4922 BUG_ON(!ss
->use_id
);
4924 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
4925 newid
= kzalloc(size
, GFP_KERNEL
);
4927 return ERR_PTR(-ENOMEM
);
4929 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
4933 write_lock(&ss
->id_lock
);
4934 /* Don't use 0. allocates an ID of 1-65535 */
4935 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
4936 write_unlock(&ss
->id_lock
);
4938 /* Returns error when there are no free spaces for new ID.*/
4943 if (myid
> CSS_ID_MAX
)
4947 newid
->depth
= depth
;
4951 write_lock(&ss
->id_lock
);
4952 idr_remove(&ss
->idr
, myid
);
4953 write_unlock(&ss
->id_lock
);
4956 return ERR_PTR(error
);
4960 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
4961 struct cgroup_subsys_state
*rootcss
)
4963 struct css_id
*newid
;
4965 rwlock_init(&ss
->id_lock
);
4968 newid
= get_new_cssid(ss
, 0);
4970 return PTR_ERR(newid
);
4972 newid
->stack
[0] = newid
->id
;
4973 newid
->css
= rootcss
;
4974 rootcss
->id
= newid
;
4978 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
4979 struct cgroup
*child
)
4981 int subsys_id
, i
, depth
= 0;
4982 struct cgroup_subsys_state
*parent_css
, *child_css
;
4983 struct css_id
*child_id
, *parent_id
;
4985 subsys_id
= ss
->subsys_id
;
4986 parent_css
= parent
->subsys
[subsys_id
];
4987 child_css
= child
->subsys
[subsys_id
];
4988 parent_id
= parent_css
->id
;
4989 depth
= parent_id
->depth
+ 1;
4991 child_id
= get_new_cssid(ss
, depth
);
4992 if (IS_ERR(child_id
))
4993 return PTR_ERR(child_id
);
4995 for (i
= 0; i
< depth
; i
++)
4996 child_id
->stack
[i
] = parent_id
->stack
[i
];
4997 child_id
->stack
[depth
] = child_id
->id
;
4999 * child_id->css pointer will be set after this cgroup is available
5000 * see cgroup_populate_dir()
5002 rcu_assign_pointer(child_css
->id
, child_id
);
5008 * css_lookup - lookup css by id
5009 * @ss: cgroup subsys to be looked into.
5012 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5013 * NULL if not. Should be called under rcu_read_lock()
5015 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5017 struct css_id
*cssid
= NULL
;
5019 BUG_ON(!ss
->use_id
);
5020 cssid
= idr_find(&ss
->idr
, id
);
5022 if (unlikely(!cssid
))
5025 return rcu_dereference(cssid
->css
);
5027 EXPORT_SYMBOL_GPL(css_lookup
);
5030 * css_get_next - lookup next cgroup under specified hierarchy.
5031 * @ss: pointer to subsystem
5032 * @id: current position of iteration.
5033 * @root: pointer to css. search tree under this.
5034 * @foundid: position of found object.
5036 * Search next css under the specified hierarchy of rootid. Calling under
5037 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5039 struct cgroup_subsys_state
*
5040 css_get_next(struct cgroup_subsys
*ss
, int id
,
5041 struct cgroup_subsys_state
*root
, int *foundid
)
5043 struct cgroup_subsys_state
*ret
= NULL
;
5046 int rootid
= css_id(root
);
5047 int depth
= css_depth(root
);
5052 BUG_ON(!ss
->use_id
);
5053 /* fill start point for scan */
5057 * scan next entry from bitmap(tree), tmpid is updated after
5060 read_lock(&ss
->id_lock
);
5061 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
5062 read_unlock(&ss
->id_lock
);
5066 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
5067 ret
= rcu_dereference(tmp
->css
);
5073 /* continue to scan from next id */
5080 * get corresponding css from file open on cgroupfs directory
5082 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5084 struct cgroup
*cgrp
;
5085 struct inode
*inode
;
5086 struct cgroup_subsys_state
*css
;
5088 inode
= f
->f_dentry
->d_inode
;
5089 /* check in cgroup filesystem dir */
5090 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5091 return ERR_PTR(-EBADF
);
5093 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5094 return ERR_PTR(-EINVAL
);
5097 cgrp
= __d_cgrp(f
->f_dentry
);
5098 css
= cgrp
->subsys
[id
];
5099 return css
? css
: ERR_PTR(-ENOENT
);
5102 #ifdef CONFIG_CGROUP_DEBUG
5103 static struct cgroup_subsys_state
*debug_create(struct cgroup_subsys
*ss
,
5104 struct cgroup
*cont
)
5106 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5109 return ERR_PTR(-ENOMEM
);
5114 static void debug_destroy(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5116 kfree(cont
->subsys
[debug_subsys_id
]);
5119 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5121 return atomic_read(&cont
->count
);
5124 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5126 return cgroup_task_count(cont
);
5129 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5131 return (u64
)(unsigned long)current
->cgroups
;
5134 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5140 count
= atomic_read(¤t
->cgroups
->refcount
);
5145 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5147 struct seq_file
*seq
)
5149 struct cg_cgroup_link
*link
;
5152 read_lock(&css_set_lock
);
5154 cg
= rcu_dereference(current
->cgroups
);
5155 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5156 struct cgroup
*c
= link
->cgrp
;
5160 name
= c
->dentry
->d_name
.name
;
5163 seq_printf(seq
, "Root %d group %s\n",
5164 c
->root
->hierarchy_id
, name
);
5167 read_unlock(&css_set_lock
);
5171 #define MAX_TASKS_SHOWN_PER_CSS 25
5172 static int cgroup_css_links_read(struct cgroup
*cont
,
5174 struct seq_file
*seq
)
5176 struct cg_cgroup_link
*link
;
5178 read_lock(&css_set_lock
);
5179 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5180 struct css_set
*cg
= link
->cg
;
5181 struct task_struct
*task
;
5183 seq_printf(seq
, "css_set %p\n", cg
);
5184 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5185 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5186 seq_puts(seq
, " ...\n");
5189 seq_printf(seq
, " task %d\n",
5190 task_pid_vnr(task
));
5194 read_unlock(&css_set_lock
);
5198 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5200 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5203 static struct cftype debug_files
[] = {
5205 .name
= "cgroup_refcount",
5206 .read_u64
= cgroup_refcount_read
,
5209 .name
= "taskcount",
5210 .read_u64
= debug_taskcount_read
,
5214 .name
= "current_css_set",
5215 .read_u64
= current_css_set_read
,
5219 .name
= "current_css_set_refcount",
5220 .read_u64
= current_css_set_refcount_read
,
5224 .name
= "current_css_set_cg_links",
5225 .read_seq_string
= current_css_set_cg_links_read
,
5229 .name
= "cgroup_css_links",
5230 .read_seq_string
= cgroup_css_links_read
,
5234 .name
= "releasable",
5235 .read_u64
= releasable_read
,
5239 static int debug_populate(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5241 return cgroup_add_files(cont
, ss
, debug_files
,
5242 ARRAY_SIZE(debug_files
));
5245 struct cgroup_subsys debug_subsys
= {
5247 .create
= debug_create
,
5248 .destroy
= debug_destroy
,
5249 .populate
= debug_populate
,
5250 .subsys_id
= debug_subsys_id
,
5252 #endif /* CONFIG_CGROUP_DEBUG */