2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hash.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_proc */
63 #include <linux/kthread.h>
65 #include <linux/atomic.h>
67 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
68 #define CSS_DEACT_BIAS INT_MIN
71 * cgroup_mutex is the master lock. Any modification to cgroup or its
72 * hierarchy must be performed while holding it.
74 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
75 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
76 * release_agent_path and so on. Modifying requires both cgroup_mutex and
77 * cgroup_root_mutex. Readers can acquire either of the two. This is to
78 * break the following locking order cycle.
80 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
81 * B. namespace_sem -> cgroup_mutex
83 * B happens only through cgroup_show_options() and using cgroup_root_mutex
86 static DEFINE_MUTEX(cgroup_mutex
);
87 static DEFINE_MUTEX(cgroup_root_mutex
);
90 * Generate an array of cgroup subsystem pointers. At boot time, this is
91 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
92 * registered after that. The mutable section of this array is protected by
95 #define SUBSYS(_x) &_x ## _subsys,
96 static struct cgroup_subsys
*subsys
[CGROUP_SUBSYS_COUNT
] = {
97 #include <linux/cgroup_subsys.h>
100 #define MAX_CGROUP_ROOT_NAMELEN 64
103 * A cgroupfs_root represents the root of a cgroup hierarchy,
104 * and may be associated with a superblock to form an active
107 struct cgroupfs_root
{
108 struct super_block
*sb
;
111 * The bitmask of subsystems intended to be attached to this
114 unsigned long subsys_bits
;
116 /* Unique id for this hierarchy. */
119 /* The bitmask of subsystems currently attached to this hierarchy */
120 unsigned long actual_subsys_bits
;
122 /* A list running through the attached subsystems */
123 struct list_head subsys_list
;
125 /* The root cgroup for this hierarchy */
126 struct cgroup top_cgroup
;
128 /* Tracks how many cgroups are currently defined in hierarchy.*/
129 int number_of_cgroups
;
131 /* A list running through the active hierarchies */
132 struct list_head root_list
;
134 /* All cgroups on this root, cgroup_mutex protected */
135 struct list_head allcg_list
;
137 /* Hierarchy-specific flags */
140 /* The path to use for release notifications. */
141 char release_agent_path
[PATH_MAX
];
143 /* The name for this hierarchy - may be empty */
144 char name
[MAX_CGROUP_ROOT_NAMELEN
];
148 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
149 * subsystems that are otherwise unattached - it never has more than a
150 * single cgroup, and all tasks are part of that cgroup.
152 static struct cgroupfs_root rootnode
;
155 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
158 struct list_head node
;
159 struct dentry
*dentry
;
164 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
165 * cgroup_subsys->use_id != 0.
167 #define CSS_ID_MAX (65535)
170 * The css to which this ID points. This pointer is set to valid value
171 * after cgroup is populated. If cgroup is removed, this will be NULL.
172 * This pointer is expected to be RCU-safe because destroy()
173 * is called after synchronize_rcu(). But for safe use, css_is_removed()
174 * css_tryget() should be used for avoiding race.
176 struct cgroup_subsys_state __rcu
*css
;
182 * Depth in hierarchy which this ID belongs to.
184 unsigned short depth
;
186 * ID is freed by RCU. (and lookup routine is RCU safe.)
188 struct rcu_head rcu_head
;
190 * Hierarchy of CSS ID belongs to.
192 unsigned short stack
[0]; /* Array of Length (depth+1) */
196 * cgroup_event represents events which userspace want to receive.
198 struct cgroup_event
{
200 * Cgroup which the event belongs to.
204 * Control file which the event associated.
208 * eventfd to signal userspace about the event.
210 struct eventfd_ctx
*eventfd
;
212 * Each of these stored in a list by the cgroup.
214 struct list_head list
;
216 * All fields below needed to unregister event when
217 * userspace closes eventfd.
220 wait_queue_head_t
*wqh
;
222 struct work_struct remove
;
225 /* The list of hierarchy roots */
227 static LIST_HEAD(roots
);
228 static int root_count
;
230 static DEFINE_IDA(hierarchy_ida
);
231 static int next_hierarchy_id
;
232 static DEFINE_SPINLOCK(hierarchy_id_lock
);
234 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
235 #define dummytop (&rootnode.top_cgroup)
237 /* This flag indicates whether tasks in the fork and exit paths should
238 * check for fork/exit handlers to call. This avoids us having to do
239 * extra work in the fork/exit path if none of the subsystems need to
242 static int need_forkexit_callback __read_mostly
;
244 #ifdef CONFIG_PROVE_LOCKING
245 int cgroup_lock_is_held(void)
247 return lockdep_is_held(&cgroup_mutex
);
249 #else /* #ifdef CONFIG_PROVE_LOCKING */
250 int cgroup_lock_is_held(void)
252 return mutex_is_locked(&cgroup_mutex
);
254 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
256 EXPORT_SYMBOL_GPL(cgroup_lock_is_held
);
258 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
259 static int css_refcnt(struct cgroup_subsys_state
*css
)
261 int v
= atomic_read(&css
->refcnt
);
263 return v
>= 0 ? v
: v
- CSS_DEACT_BIAS
;
266 /* convenient tests for these bits */
267 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
269 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
272 /* bits in struct cgroupfs_root flags field */
274 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
277 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
280 (1 << CGRP_RELEASABLE
) |
281 (1 << CGRP_NOTIFY_ON_RELEASE
);
282 return (cgrp
->flags
& bits
) == bits
;
285 static int notify_on_release(const struct cgroup
*cgrp
)
287 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
290 static int clone_children(const struct cgroup
*cgrp
)
292 return test_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
296 * for_each_subsys() allows you to iterate on each subsystem attached to
297 * an active hierarchy
299 #define for_each_subsys(_root, _ss) \
300 list_for_each_entry(_ss, &_root->subsys_list, sibling)
302 /* for_each_active_root() allows you to iterate across the active hierarchies */
303 #define for_each_active_root(_root) \
304 list_for_each_entry(_root, &roots, root_list)
306 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
308 return dentry
->d_fsdata
;
311 static inline struct cfent
*__d_cfe(struct dentry
*dentry
)
313 return dentry
->d_fsdata
;
316 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
318 return __d_cfe(dentry
)->type
;
321 /* the list of cgroups eligible for automatic release. Protected by
322 * release_list_lock */
323 static LIST_HEAD(release_list
);
324 static DEFINE_RAW_SPINLOCK(release_list_lock
);
325 static void cgroup_release_agent(struct work_struct
*work
);
326 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
327 static void check_for_release(struct cgroup
*cgrp
);
329 /* Link structure for associating css_set objects with cgroups */
330 struct cg_cgroup_link
{
332 * List running through cg_cgroup_links associated with a
333 * cgroup, anchored on cgroup->css_sets
335 struct list_head cgrp_link_list
;
338 * List running through cg_cgroup_links pointing at a
339 * single css_set object, anchored on css_set->cg_links
341 struct list_head cg_link_list
;
345 /* The default css_set - used by init and its children prior to any
346 * hierarchies being mounted. It contains a pointer to the root state
347 * for each subsystem. Also used to anchor the list of css_sets. Not
348 * reference-counted, to improve performance when child cgroups
349 * haven't been created.
352 static struct css_set init_css_set
;
353 static struct cg_cgroup_link init_css_set_link
;
355 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
356 struct cgroup_subsys_state
*css
);
358 /* css_set_lock protects the list of css_set objects, and the
359 * chain of tasks off each css_set. Nests outside task->alloc_lock
360 * due to cgroup_iter_start() */
361 static DEFINE_RWLOCK(css_set_lock
);
362 static int css_set_count
;
365 * hash table for cgroup groups. This improves the performance to find
366 * an existing css_set. This hash doesn't (currently) take into
367 * account cgroups in empty hierarchies.
369 #define CSS_SET_HASH_BITS 7
370 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
371 static struct hlist_head css_set_table
[CSS_SET_TABLE_SIZE
];
373 static struct hlist_head
*css_set_hash(struct cgroup_subsys_state
*css
[])
377 unsigned long tmp
= 0UL;
379 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
380 tmp
+= (unsigned long)css
[i
];
381 tmp
= (tmp
>> 16) ^ tmp
;
383 index
= hash_long(tmp
, CSS_SET_HASH_BITS
);
385 return &css_set_table
[index
];
388 /* We don't maintain the lists running through each css_set to its
389 * task until after the first call to cgroup_iter_start(). This
390 * reduces the fork()/exit() overhead for people who have cgroups
391 * compiled into their kernel but not actually in use */
392 static int use_task_css_set_links __read_mostly
;
394 static void __put_css_set(struct css_set
*cg
, int taskexit
)
396 struct cg_cgroup_link
*link
;
397 struct cg_cgroup_link
*saved_link
;
399 * Ensure that the refcount doesn't hit zero while any readers
400 * can see it. Similar to atomic_dec_and_lock(), but for an
403 if (atomic_add_unless(&cg
->refcount
, -1, 1))
405 write_lock(&css_set_lock
);
406 if (!atomic_dec_and_test(&cg
->refcount
)) {
407 write_unlock(&css_set_lock
);
411 /* This css_set is dead. unlink it and release cgroup refcounts */
412 hlist_del(&cg
->hlist
);
415 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
417 struct cgroup
*cgrp
= link
->cgrp
;
418 list_del(&link
->cg_link_list
);
419 list_del(&link
->cgrp_link_list
);
420 if (atomic_dec_and_test(&cgrp
->count
) &&
421 notify_on_release(cgrp
)) {
423 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
424 check_for_release(cgrp
);
430 write_unlock(&css_set_lock
);
431 kfree_rcu(cg
, rcu_head
);
435 * refcounted get/put for css_set objects
437 static inline void get_css_set(struct css_set
*cg
)
439 atomic_inc(&cg
->refcount
);
442 static inline void put_css_set(struct css_set
*cg
)
444 __put_css_set(cg
, 0);
447 static inline void put_css_set_taskexit(struct css_set
*cg
)
449 __put_css_set(cg
, 1);
453 * compare_css_sets - helper function for find_existing_css_set().
454 * @cg: candidate css_set being tested
455 * @old_cg: existing css_set for a task
456 * @new_cgrp: cgroup that's being entered by the task
457 * @template: desired set of css pointers in css_set (pre-calculated)
459 * Returns true if "cg" matches "old_cg" except for the hierarchy
460 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
462 static bool compare_css_sets(struct css_set
*cg
,
463 struct css_set
*old_cg
,
464 struct cgroup
*new_cgrp
,
465 struct cgroup_subsys_state
*template[])
467 struct list_head
*l1
, *l2
;
469 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
470 /* Not all subsystems matched */
475 * Compare cgroup pointers in order to distinguish between
476 * different cgroups in heirarchies with no subsystems. We
477 * could get by with just this check alone (and skip the
478 * memcmp above) but on most setups the memcmp check will
479 * avoid the need for this more expensive check on almost all
484 l2
= &old_cg
->cg_links
;
486 struct cg_cgroup_link
*cgl1
, *cgl2
;
487 struct cgroup
*cg1
, *cg2
;
491 /* See if we reached the end - both lists are equal length. */
492 if (l1
== &cg
->cg_links
) {
493 BUG_ON(l2
!= &old_cg
->cg_links
);
496 BUG_ON(l2
== &old_cg
->cg_links
);
498 /* Locate the cgroups associated with these links. */
499 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
500 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
503 /* Hierarchies should be linked in the same order. */
504 BUG_ON(cg1
->root
!= cg2
->root
);
507 * If this hierarchy is the hierarchy of the cgroup
508 * that's changing, then we need to check that this
509 * css_set points to the new cgroup; if it's any other
510 * hierarchy, then this css_set should point to the
511 * same cgroup as the old css_set.
513 if (cg1
->root
== new_cgrp
->root
) {
525 * find_existing_css_set() is a helper for
526 * find_css_set(), and checks to see whether an existing
527 * css_set is suitable.
529 * oldcg: the cgroup group that we're using before the cgroup
532 * cgrp: the cgroup that we're moving into
534 * template: location in which to build the desired set of subsystem
535 * state objects for the new cgroup group
537 static struct css_set
*find_existing_css_set(
538 struct css_set
*oldcg
,
540 struct cgroup_subsys_state
*template[])
543 struct cgroupfs_root
*root
= cgrp
->root
;
544 struct hlist_head
*hhead
;
545 struct hlist_node
*node
;
549 * Build the set of subsystem state objects that we want to see in the
550 * new css_set. while subsystems can change globally, the entries here
551 * won't change, so no need for locking.
553 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
554 if (root
->subsys_bits
& (1UL << i
)) {
555 /* Subsystem is in this hierarchy. So we want
556 * the subsystem state from the new
558 template[i
] = cgrp
->subsys
[i
];
560 /* Subsystem is not in this hierarchy, so we
561 * don't want to change the subsystem state */
562 template[i
] = oldcg
->subsys
[i
];
566 hhead
= css_set_hash(template);
567 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
568 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
571 /* This css_set matches what we need */
575 /* No existing cgroup group matched */
579 static void free_cg_links(struct list_head
*tmp
)
581 struct cg_cgroup_link
*link
;
582 struct cg_cgroup_link
*saved_link
;
584 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
585 list_del(&link
->cgrp_link_list
);
591 * allocate_cg_links() allocates "count" cg_cgroup_link structures
592 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
593 * success or a negative error
595 static int allocate_cg_links(int count
, struct list_head
*tmp
)
597 struct cg_cgroup_link
*link
;
600 for (i
= 0; i
< count
; i
++) {
601 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
606 list_add(&link
->cgrp_link_list
, tmp
);
612 * link_css_set - a helper function to link a css_set to a cgroup
613 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
614 * @cg: the css_set to be linked
615 * @cgrp: the destination cgroup
617 static void link_css_set(struct list_head
*tmp_cg_links
,
618 struct css_set
*cg
, struct cgroup
*cgrp
)
620 struct cg_cgroup_link
*link
;
622 BUG_ON(list_empty(tmp_cg_links
));
623 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
627 atomic_inc(&cgrp
->count
);
628 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
630 * Always add links to the tail of the list so that the list
631 * is sorted by order of hierarchy creation
633 list_add_tail(&link
->cg_link_list
, &cg
->cg_links
);
637 * find_css_set() takes an existing cgroup group and a
638 * cgroup object, and returns a css_set object that's
639 * equivalent to the old group, but with the given cgroup
640 * substituted into the appropriate hierarchy. Must be called with
643 static struct css_set
*find_css_set(
644 struct css_set
*oldcg
, struct cgroup
*cgrp
)
647 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
649 struct list_head tmp_cg_links
;
651 struct hlist_head
*hhead
;
652 struct cg_cgroup_link
*link
;
654 /* First see if we already have a cgroup group that matches
656 read_lock(&css_set_lock
);
657 res
= find_existing_css_set(oldcg
, cgrp
, template);
660 read_unlock(&css_set_lock
);
665 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
669 /* Allocate all the cg_cgroup_link objects that we'll need */
670 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
675 atomic_set(&res
->refcount
, 1);
676 INIT_LIST_HEAD(&res
->cg_links
);
677 INIT_LIST_HEAD(&res
->tasks
);
678 INIT_HLIST_NODE(&res
->hlist
);
680 /* Copy the set of subsystem state objects generated in
681 * find_existing_css_set() */
682 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
684 write_lock(&css_set_lock
);
685 /* Add reference counts and links from the new css_set. */
686 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
687 struct cgroup
*c
= link
->cgrp
;
688 if (c
->root
== cgrp
->root
)
690 link_css_set(&tmp_cg_links
, res
, c
);
693 BUG_ON(!list_empty(&tmp_cg_links
));
697 /* Add this cgroup group to the hash table */
698 hhead
= css_set_hash(res
->subsys
);
699 hlist_add_head(&res
->hlist
, hhead
);
701 write_unlock(&css_set_lock
);
707 * Return the cgroup for "task" from the given hierarchy. Must be
708 * called with cgroup_mutex held.
710 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
711 struct cgroupfs_root
*root
)
714 struct cgroup
*res
= NULL
;
716 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
717 read_lock(&css_set_lock
);
719 * No need to lock the task - since we hold cgroup_mutex the
720 * task can't change groups, so the only thing that can happen
721 * is that it exits and its css is set back to init_css_set.
724 if (css
== &init_css_set
) {
725 res
= &root
->top_cgroup
;
727 struct cg_cgroup_link
*link
;
728 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
729 struct cgroup
*c
= link
->cgrp
;
730 if (c
->root
== root
) {
736 read_unlock(&css_set_lock
);
742 * There is one global cgroup mutex. We also require taking
743 * task_lock() when dereferencing a task's cgroup subsys pointers.
744 * See "The task_lock() exception", at the end of this comment.
746 * A task must hold cgroup_mutex to modify cgroups.
748 * Any task can increment and decrement the count field without lock.
749 * So in general, code holding cgroup_mutex can't rely on the count
750 * field not changing. However, if the count goes to zero, then only
751 * cgroup_attach_task() can increment it again. Because a count of zero
752 * means that no tasks are currently attached, therefore there is no
753 * way a task attached to that cgroup can fork (the other way to
754 * increment the count). So code holding cgroup_mutex can safely
755 * assume that if the count is zero, it will stay zero. Similarly, if
756 * a task holds cgroup_mutex on a cgroup with zero count, it
757 * knows that the cgroup won't be removed, as cgroup_rmdir()
760 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
761 * (usually) take cgroup_mutex. These are the two most performance
762 * critical pieces of code here. The exception occurs on cgroup_exit(),
763 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
764 * is taken, and if the cgroup count is zero, a usermode call made
765 * to the release agent with the name of the cgroup (path relative to
766 * the root of cgroup file system) as the argument.
768 * A cgroup can only be deleted if both its 'count' of using tasks
769 * is zero, and its list of 'children' cgroups is empty. Since all
770 * tasks in the system use _some_ cgroup, and since there is always at
771 * least one task in the system (init, pid == 1), therefore, top_cgroup
772 * always has either children cgroups and/or using tasks. So we don't
773 * need a special hack to ensure that top_cgroup cannot be deleted.
775 * The task_lock() exception
777 * The need for this exception arises from the action of
778 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
779 * another. It does so using cgroup_mutex, however there are
780 * several performance critical places that need to reference
781 * task->cgroup without the expense of grabbing a system global
782 * mutex. Therefore except as noted below, when dereferencing or, as
783 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
784 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
785 * the task_struct routinely used for such matters.
787 * P.S. One more locking exception. RCU is used to guard the
788 * update of a tasks cgroup pointer by cgroup_attach_task()
792 * cgroup_lock - lock out any changes to cgroup structures
795 void cgroup_lock(void)
797 mutex_lock(&cgroup_mutex
);
799 EXPORT_SYMBOL_GPL(cgroup_lock
);
802 * cgroup_unlock - release lock on cgroup changes
804 * Undo the lock taken in a previous cgroup_lock() call.
806 void cgroup_unlock(void)
808 mutex_unlock(&cgroup_mutex
);
810 EXPORT_SYMBOL_GPL(cgroup_unlock
);
813 * A couple of forward declarations required, due to cyclic reference loop:
814 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
815 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
819 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
820 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
821 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
822 static int cgroup_populate_dir(struct cgroup
*cgrp
);
823 static const struct inode_operations cgroup_dir_inode_operations
;
824 static const struct file_operations proc_cgroupstats_operations
;
826 static struct backing_dev_info cgroup_backing_dev_info
= {
828 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
831 static int alloc_css_id(struct cgroup_subsys
*ss
,
832 struct cgroup
*parent
, struct cgroup
*child
);
834 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
836 struct inode
*inode
= new_inode(sb
);
839 inode
->i_ino
= get_next_ino();
840 inode
->i_mode
= mode
;
841 inode
->i_uid
= current_fsuid();
842 inode
->i_gid
= current_fsgid();
843 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
844 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
850 * Call subsys's pre_destroy handler.
851 * This is called before css refcnt check.
853 static int cgroup_call_pre_destroy(struct cgroup
*cgrp
)
855 struct cgroup_subsys
*ss
;
858 for_each_subsys(cgrp
->root
, ss
) {
859 if (!ss
->pre_destroy
)
862 ret
= ss
->pre_destroy(cgrp
);
864 /* ->pre_destroy() failure is being deprecated */
865 WARN_ON_ONCE(!ss
->__DEPRECATED_clear_css_refs
);
873 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
875 /* is dentry a directory ? if so, kfree() associated cgroup */
876 if (S_ISDIR(inode
->i_mode
)) {
877 struct cgroup
*cgrp
= dentry
->d_fsdata
;
878 struct cgroup_subsys
*ss
;
879 BUG_ON(!(cgroup_is_removed(cgrp
)));
880 /* It's possible for external users to be holding css
881 * reference counts on a cgroup; css_put() needs to
882 * be able to access the cgroup after decrementing
883 * the reference count in order to know if it needs to
884 * queue the cgroup to be handled by the release
888 mutex_lock(&cgroup_mutex
);
890 * Release the subsystem state objects.
892 for_each_subsys(cgrp
->root
, ss
)
895 cgrp
->root
->number_of_cgroups
--;
896 mutex_unlock(&cgroup_mutex
);
899 * We want to drop the active superblock reference from the
900 * cgroup creation after all the dentry refs are gone -
901 * kill_sb gets mighty unhappy otherwise. Mark
902 * dentry->d_fsdata with cgroup_diput() to tell
903 * cgroup_d_release() to call deactivate_super().
905 dentry
->d_fsdata
= cgroup_diput
;
908 * if we're getting rid of the cgroup, refcount should ensure
909 * that there are no pidlists left.
911 BUG_ON(!list_empty(&cgrp
->pidlists
));
913 kfree_rcu(cgrp
, rcu_head
);
915 struct cfent
*cfe
= __d_cfe(dentry
);
916 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
918 WARN_ONCE(!list_empty(&cfe
->node
) &&
919 cgrp
!= &cgrp
->root
->top_cgroup
,
920 "cfe still linked for %s\n", cfe
->type
->name
);
926 static int cgroup_delete(const struct dentry
*d
)
931 static void cgroup_d_release(struct dentry
*dentry
)
933 /* did cgroup_diput() tell me to deactivate super? */
934 if (dentry
->d_fsdata
== cgroup_diput
)
935 deactivate_super(dentry
->d_sb
);
938 static void remove_dir(struct dentry
*d
)
940 struct dentry
*parent
= dget(d
->d_parent
);
943 simple_rmdir(parent
->d_inode
, d
);
947 static int cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
951 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
952 lockdep_assert_held(&cgroup_mutex
);
954 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
955 struct dentry
*d
= cfe
->dentry
;
957 if (cft
&& cfe
->type
!= cft
)
962 simple_unlink(cgrp
->dentry
->d_inode
, d
);
963 list_del_init(&cfe
->node
);
971 static void cgroup_clear_directory(struct dentry
*dir
)
973 struct cgroup
*cgrp
= __d_cgrp(dir
);
975 while (!list_empty(&cgrp
->files
))
976 cgroup_rm_file(cgrp
, NULL
);
980 * NOTE : the dentry must have been dget()'ed
982 static void cgroup_d_remove_dir(struct dentry
*dentry
)
984 struct dentry
*parent
;
986 cgroup_clear_directory(dentry
);
988 parent
= dentry
->d_parent
;
989 spin_lock(&parent
->d_lock
);
990 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
991 list_del_init(&dentry
->d_u
.d_child
);
992 spin_unlock(&dentry
->d_lock
);
993 spin_unlock(&parent
->d_lock
);
998 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
999 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
1000 * reference to css->refcnt. In general, this refcnt is expected to goes down
1003 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
1005 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq
);
1007 static void cgroup_wakeup_rmdir_waiter(struct cgroup
*cgrp
)
1009 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
)))
1010 wake_up_all(&cgroup_rmdir_waitq
);
1013 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
)
1018 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
)
1020 cgroup_wakeup_rmdir_waiter(css
->cgroup
);
1025 * Call with cgroup_mutex held. Drops reference counts on modules, including
1026 * any duplicate ones that parse_cgroupfs_options took. If this function
1027 * returns an error, no reference counts are touched.
1029 static int rebind_subsystems(struct cgroupfs_root
*root
,
1030 unsigned long final_bits
)
1032 unsigned long added_bits
, removed_bits
;
1033 struct cgroup
*cgrp
= &root
->top_cgroup
;
1036 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1037 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1039 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
1040 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
1041 /* Check that any added subsystems are currently free */
1042 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1043 unsigned long bit
= 1UL << i
;
1044 struct cgroup_subsys
*ss
= subsys
[i
];
1045 if (!(bit
& added_bits
))
1048 * Nobody should tell us to do a subsys that doesn't exist:
1049 * parse_cgroupfs_options should catch that case and refcounts
1050 * ensure that subsystems won't disappear once selected.
1053 if (ss
->root
!= &rootnode
) {
1054 /* Subsystem isn't free */
1059 /* Currently we don't handle adding/removing subsystems when
1060 * any child cgroups exist. This is theoretically supportable
1061 * but involves complex error handling, so it's being left until
1063 if (root
->number_of_cgroups
> 1)
1066 /* Process each subsystem */
1067 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1068 struct cgroup_subsys
*ss
= subsys
[i
];
1069 unsigned long bit
= 1UL << i
;
1070 if (bit
& added_bits
) {
1071 /* We're binding this subsystem to this hierarchy */
1073 BUG_ON(cgrp
->subsys
[i
]);
1074 BUG_ON(!dummytop
->subsys
[i
]);
1075 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
1076 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
1077 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1078 list_move(&ss
->sibling
, &root
->subsys_list
);
1082 /* refcount was already taken, and we're keeping it */
1083 } else if (bit
& removed_bits
) {
1084 /* We're removing this subsystem */
1086 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1087 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1090 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1091 cgrp
->subsys
[i
] = NULL
;
1092 subsys
[i
]->root
= &rootnode
;
1093 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1094 /* subsystem is now free - drop reference on module */
1095 module_put(ss
->module
);
1096 } else if (bit
& final_bits
) {
1097 /* Subsystem state should already exist */
1099 BUG_ON(!cgrp
->subsys
[i
]);
1101 * a refcount was taken, but we already had one, so
1102 * drop the extra reference.
1104 module_put(ss
->module
);
1105 #ifdef CONFIG_MODULE_UNLOAD
1106 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1109 /* Subsystem state shouldn't exist */
1110 BUG_ON(cgrp
->subsys
[i
]);
1113 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
1119 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1121 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1122 struct cgroup_subsys
*ss
;
1124 mutex_lock(&cgroup_root_mutex
);
1125 for_each_subsys(root
, ss
)
1126 seq_printf(seq
, ",%s", ss
->name
);
1127 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1128 seq_puts(seq
, ",noprefix");
1129 if (strlen(root
->release_agent_path
))
1130 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1131 if (clone_children(&root
->top_cgroup
))
1132 seq_puts(seq
, ",clone_children");
1133 if (strlen(root
->name
))
1134 seq_printf(seq
, ",name=%s", root
->name
);
1135 mutex_unlock(&cgroup_root_mutex
);
1139 struct cgroup_sb_opts
{
1140 unsigned long subsys_bits
;
1141 unsigned long flags
;
1142 char *release_agent
;
1143 bool clone_children
;
1145 /* User explicitly requested empty subsystem */
1148 struct cgroupfs_root
*new_root
;
1153 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1154 * with cgroup_mutex held to protect the subsys[] array. This function takes
1155 * refcounts on subsystems to be used, unless it returns error, in which case
1156 * no refcounts are taken.
1158 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1160 char *token
, *o
= data
;
1161 bool all_ss
= false, one_ss
= false;
1162 unsigned long mask
= (unsigned long)-1;
1164 bool module_pin_failed
= false;
1166 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1168 #ifdef CONFIG_CPUSETS
1169 mask
= ~(1UL << cpuset_subsys_id
);
1172 memset(opts
, 0, sizeof(*opts
));
1174 while ((token
= strsep(&o
, ",")) != NULL
) {
1177 if (!strcmp(token
, "none")) {
1178 /* Explicitly have no subsystems */
1182 if (!strcmp(token
, "all")) {
1183 /* Mutually exclusive option 'all' + subsystem name */
1189 if (!strcmp(token
, "noprefix")) {
1190 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1193 if (!strcmp(token
, "clone_children")) {
1194 opts
->clone_children
= true;
1197 if (!strncmp(token
, "release_agent=", 14)) {
1198 /* Specifying two release agents is forbidden */
1199 if (opts
->release_agent
)
1201 opts
->release_agent
=
1202 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1203 if (!opts
->release_agent
)
1207 if (!strncmp(token
, "name=", 5)) {
1208 const char *name
= token
+ 5;
1209 /* Can't specify an empty name */
1212 /* Must match [\w.-]+ */
1213 for (i
= 0; i
< strlen(name
); i
++) {
1217 if ((c
== '.') || (c
== '-') || (c
== '_'))
1221 /* Specifying two names is forbidden */
1224 opts
->name
= kstrndup(name
,
1225 MAX_CGROUP_ROOT_NAMELEN
- 1,
1233 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1234 struct cgroup_subsys
*ss
= subsys
[i
];
1237 if (strcmp(token
, ss
->name
))
1242 /* Mutually exclusive option 'all' + subsystem name */
1245 set_bit(i
, &opts
->subsys_bits
);
1250 if (i
== CGROUP_SUBSYS_COUNT
)
1255 * If the 'all' option was specified select all the subsystems,
1256 * otherwise if 'none', 'name=' and a subsystem name options
1257 * were not specified, let's default to 'all'
1259 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
)) {
1260 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1261 struct cgroup_subsys
*ss
= subsys
[i
];
1266 set_bit(i
, &opts
->subsys_bits
);
1270 /* Consistency checks */
1273 * Option noprefix was introduced just for backward compatibility
1274 * with the old cpuset, so we allow noprefix only if mounting just
1275 * the cpuset subsystem.
1277 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1278 (opts
->subsys_bits
& mask
))
1282 /* Can't specify "none" and some subsystems */
1283 if (opts
->subsys_bits
&& opts
->none
)
1287 * We either have to specify by name or by subsystems. (So all
1288 * empty hierarchies must have a name).
1290 if (!opts
->subsys_bits
&& !opts
->name
)
1294 * Grab references on all the modules we'll need, so the subsystems
1295 * don't dance around before rebind_subsystems attaches them. This may
1296 * take duplicate reference counts on a subsystem that's already used,
1297 * but rebind_subsystems handles this case.
1299 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1300 unsigned long bit
= 1UL << i
;
1302 if (!(bit
& opts
->subsys_bits
))
1304 if (!try_module_get(subsys
[i
]->module
)) {
1305 module_pin_failed
= true;
1309 if (module_pin_failed
) {
1311 * oops, one of the modules was going away. this means that we
1312 * raced with a module_delete call, and to the user this is
1313 * essentially a "subsystem doesn't exist" case.
1315 for (i
--; i
>= CGROUP_BUILTIN_SUBSYS_COUNT
; i
--) {
1316 /* drop refcounts only on the ones we took */
1317 unsigned long bit
= 1UL << i
;
1319 if (!(bit
& opts
->subsys_bits
))
1321 module_put(subsys
[i
]->module
);
1329 static void drop_parsed_module_refcounts(unsigned long subsys_bits
)
1332 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1333 unsigned long bit
= 1UL << i
;
1335 if (!(bit
& subsys_bits
))
1337 module_put(subsys
[i
]->module
);
1341 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1344 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1345 struct cgroup
*cgrp
= &root
->top_cgroup
;
1346 struct cgroup_sb_opts opts
;
1348 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1349 mutex_lock(&cgroup_mutex
);
1350 mutex_lock(&cgroup_root_mutex
);
1352 /* See what subsystems are wanted */
1353 ret
= parse_cgroupfs_options(data
, &opts
);
1357 /* See feature-removal-schedule.txt */
1358 if (opts
.subsys_bits
!= root
->actual_subsys_bits
|| opts
.release_agent
)
1359 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1360 task_tgid_nr(current
), current
->comm
);
1362 /* Don't allow flags or name to change at remount */
1363 if (opts
.flags
!= root
->flags
||
1364 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1366 drop_parsed_module_refcounts(opts
.subsys_bits
);
1370 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
1372 drop_parsed_module_refcounts(opts
.subsys_bits
);
1376 /* clear out any existing files and repopulate subsystem files */
1377 cgroup_clear_directory(cgrp
->dentry
);
1378 cgroup_populate_dir(cgrp
);
1380 if (opts
.release_agent
)
1381 strcpy(root
->release_agent_path
, opts
.release_agent
);
1383 kfree(opts
.release_agent
);
1385 mutex_unlock(&cgroup_root_mutex
);
1386 mutex_unlock(&cgroup_mutex
);
1387 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1391 static const struct super_operations cgroup_ops
= {
1392 .statfs
= simple_statfs
,
1393 .drop_inode
= generic_delete_inode
,
1394 .show_options
= cgroup_show_options
,
1395 .remount_fs
= cgroup_remount
,
1398 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1400 INIT_LIST_HEAD(&cgrp
->sibling
);
1401 INIT_LIST_HEAD(&cgrp
->children
);
1402 INIT_LIST_HEAD(&cgrp
->files
);
1403 INIT_LIST_HEAD(&cgrp
->css_sets
);
1404 INIT_LIST_HEAD(&cgrp
->release_list
);
1405 INIT_LIST_HEAD(&cgrp
->pidlists
);
1406 mutex_init(&cgrp
->pidlist_mutex
);
1407 INIT_LIST_HEAD(&cgrp
->event_list
);
1408 spin_lock_init(&cgrp
->event_list_lock
);
1411 static void init_cgroup_root(struct cgroupfs_root
*root
)
1413 struct cgroup
*cgrp
= &root
->top_cgroup
;
1415 INIT_LIST_HEAD(&root
->subsys_list
);
1416 INIT_LIST_HEAD(&root
->root_list
);
1417 INIT_LIST_HEAD(&root
->allcg_list
);
1418 root
->number_of_cgroups
= 1;
1420 cgrp
->top_cgroup
= cgrp
;
1421 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
1422 init_cgroup_housekeeping(cgrp
);
1425 static bool init_root_id(struct cgroupfs_root
*root
)
1430 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1432 spin_lock(&hierarchy_id_lock
);
1433 /* Try to allocate the next unused ID */
1434 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1435 &root
->hierarchy_id
);
1437 /* Try again starting from 0 */
1438 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1440 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1441 } else if (ret
!= -EAGAIN
) {
1442 /* Can only get here if the 31-bit IDR is full ... */
1445 spin_unlock(&hierarchy_id_lock
);
1450 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1452 struct cgroup_sb_opts
*opts
= data
;
1453 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1455 /* If we asked for a name then it must match */
1456 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1460 * If we asked for subsystems (or explicitly for no
1461 * subsystems) then they must match
1463 if ((opts
->subsys_bits
|| opts
->none
)
1464 && (opts
->subsys_bits
!= root
->subsys_bits
))
1470 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1472 struct cgroupfs_root
*root
;
1474 if (!opts
->subsys_bits
&& !opts
->none
)
1477 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1479 return ERR_PTR(-ENOMEM
);
1481 if (!init_root_id(root
)) {
1483 return ERR_PTR(-ENOMEM
);
1485 init_cgroup_root(root
);
1487 root
->subsys_bits
= opts
->subsys_bits
;
1488 root
->flags
= opts
->flags
;
1489 if (opts
->release_agent
)
1490 strcpy(root
->release_agent_path
, opts
->release_agent
);
1492 strcpy(root
->name
, opts
->name
);
1493 if (opts
->clone_children
)
1494 set_bit(CGRP_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1498 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1503 BUG_ON(!root
->hierarchy_id
);
1504 spin_lock(&hierarchy_id_lock
);
1505 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1506 spin_unlock(&hierarchy_id_lock
);
1510 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1513 struct cgroup_sb_opts
*opts
= data
;
1515 /* If we don't have a new root, we can't set up a new sb */
1516 if (!opts
->new_root
)
1519 BUG_ON(!opts
->subsys_bits
&& !opts
->none
);
1521 ret
= set_anon_super(sb
, NULL
);
1525 sb
->s_fs_info
= opts
->new_root
;
1526 opts
->new_root
->sb
= sb
;
1528 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1529 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1530 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1531 sb
->s_op
= &cgroup_ops
;
1536 static int cgroup_get_rootdir(struct super_block
*sb
)
1538 static const struct dentry_operations cgroup_dops
= {
1539 .d_iput
= cgroup_diput
,
1540 .d_delete
= cgroup_delete
,
1541 .d_release
= cgroup_d_release
,
1544 struct inode
*inode
=
1545 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1550 inode
->i_fop
= &simple_dir_operations
;
1551 inode
->i_op
= &cgroup_dir_inode_operations
;
1552 /* directories start off with i_nlink == 2 (for "." entry) */
1554 sb
->s_root
= d_make_root(inode
);
1557 /* for everything else we want ->d_op set */
1558 sb
->s_d_op
= &cgroup_dops
;
1562 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1563 int flags
, const char *unused_dev_name
,
1566 struct cgroup_sb_opts opts
;
1567 struct cgroupfs_root
*root
;
1569 struct super_block
*sb
;
1570 struct cgroupfs_root
*new_root
;
1571 struct inode
*inode
;
1573 /* First find the desired set of subsystems */
1574 mutex_lock(&cgroup_mutex
);
1575 ret
= parse_cgroupfs_options(data
, &opts
);
1576 mutex_unlock(&cgroup_mutex
);
1581 * Allocate a new cgroup root. We may not need it if we're
1582 * reusing an existing hierarchy.
1584 new_root
= cgroup_root_from_opts(&opts
);
1585 if (IS_ERR(new_root
)) {
1586 ret
= PTR_ERR(new_root
);
1589 opts
.new_root
= new_root
;
1591 /* Locate an existing or new sb for this hierarchy */
1592 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, &opts
);
1595 cgroup_drop_root(opts
.new_root
);
1599 root
= sb
->s_fs_info
;
1601 if (root
== opts
.new_root
) {
1602 /* We used the new root structure, so this is a new hierarchy */
1603 struct list_head tmp_cg_links
;
1604 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1605 struct cgroupfs_root
*existing_root
;
1606 const struct cred
*cred
;
1609 BUG_ON(sb
->s_root
!= NULL
);
1611 ret
= cgroup_get_rootdir(sb
);
1613 goto drop_new_super
;
1614 inode
= sb
->s_root
->d_inode
;
1616 mutex_lock(&inode
->i_mutex
);
1617 mutex_lock(&cgroup_mutex
);
1618 mutex_lock(&cgroup_root_mutex
);
1620 /* Check for name clashes with existing mounts */
1622 if (strlen(root
->name
))
1623 for_each_active_root(existing_root
)
1624 if (!strcmp(existing_root
->name
, root
->name
))
1628 * We're accessing css_set_count without locking
1629 * css_set_lock here, but that's OK - it can only be
1630 * increased by someone holding cgroup_lock, and
1631 * that's us. The worst that can happen is that we
1632 * have some link structures left over
1634 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1638 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1639 if (ret
== -EBUSY
) {
1640 free_cg_links(&tmp_cg_links
);
1644 * There must be no failure case after here, since rebinding
1645 * takes care of subsystems' refcounts, which are explicitly
1646 * dropped in the failure exit path.
1649 /* EBUSY should be the only error here */
1652 list_add(&root
->root_list
, &roots
);
1655 sb
->s_root
->d_fsdata
= root_cgrp
;
1656 root
->top_cgroup
.dentry
= sb
->s_root
;
1658 /* Link the top cgroup in this hierarchy into all
1659 * the css_set objects */
1660 write_lock(&css_set_lock
);
1661 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1662 struct hlist_head
*hhead
= &css_set_table
[i
];
1663 struct hlist_node
*node
;
1666 hlist_for_each_entry(cg
, node
, hhead
, hlist
)
1667 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1669 write_unlock(&css_set_lock
);
1671 free_cg_links(&tmp_cg_links
);
1673 BUG_ON(!list_empty(&root_cgrp
->sibling
));
1674 BUG_ON(!list_empty(&root_cgrp
->children
));
1675 BUG_ON(root
->number_of_cgroups
!= 1);
1677 cred
= override_creds(&init_cred
);
1678 cgroup_populate_dir(root_cgrp
);
1680 mutex_unlock(&cgroup_root_mutex
);
1681 mutex_unlock(&cgroup_mutex
);
1682 mutex_unlock(&inode
->i_mutex
);
1685 * We re-used an existing hierarchy - the new root (if
1686 * any) is not needed
1688 cgroup_drop_root(opts
.new_root
);
1689 /* no subsys rebinding, so refcounts don't change */
1690 drop_parsed_module_refcounts(opts
.subsys_bits
);
1693 kfree(opts
.release_agent
);
1695 return dget(sb
->s_root
);
1698 mutex_unlock(&cgroup_root_mutex
);
1699 mutex_unlock(&cgroup_mutex
);
1700 mutex_unlock(&inode
->i_mutex
);
1702 deactivate_locked_super(sb
);
1704 drop_parsed_module_refcounts(opts
.subsys_bits
);
1706 kfree(opts
.release_agent
);
1708 return ERR_PTR(ret
);
1711 static void cgroup_kill_sb(struct super_block
*sb
) {
1712 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1713 struct cgroup
*cgrp
= &root
->top_cgroup
;
1715 struct cg_cgroup_link
*link
;
1716 struct cg_cgroup_link
*saved_link
;
1720 BUG_ON(root
->number_of_cgroups
!= 1);
1721 BUG_ON(!list_empty(&cgrp
->children
));
1722 BUG_ON(!list_empty(&cgrp
->sibling
));
1724 mutex_lock(&cgroup_mutex
);
1725 mutex_lock(&cgroup_root_mutex
);
1727 /* Rebind all subsystems back to the default hierarchy */
1728 ret
= rebind_subsystems(root
, 0);
1729 /* Shouldn't be able to fail ... */
1733 * Release all the links from css_sets to this hierarchy's
1736 write_lock(&css_set_lock
);
1738 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1740 list_del(&link
->cg_link_list
);
1741 list_del(&link
->cgrp_link_list
);
1744 write_unlock(&css_set_lock
);
1746 if (!list_empty(&root
->root_list
)) {
1747 list_del(&root
->root_list
);
1751 mutex_unlock(&cgroup_root_mutex
);
1752 mutex_unlock(&cgroup_mutex
);
1754 kill_litter_super(sb
);
1755 cgroup_drop_root(root
);
1758 static struct file_system_type cgroup_fs_type
= {
1760 .mount
= cgroup_mount
,
1761 .kill_sb
= cgroup_kill_sb
,
1764 static struct kobject
*cgroup_kobj
;
1767 * cgroup_path - generate the path of a cgroup
1768 * @cgrp: the cgroup in question
1769 * @buf: the buffer to write the path into
1770 * @buflen: the length of the buffer
1772 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1773 * reference. Writes path of cgroup into buf. Returns 0 on success,
1776 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1779 struct dentry
*dentry
= rcu_dereference_check(cgrp
->dentry
,
1780 cgroup_lock_is_held());
1782 if (!dentry
|| cgrp
== dummytop
) {
1784 * Inactive subsystems have no dentry for their root
1791 start
= buf
+ buflen
;
1795 int len
= dentry
->d_name
.len
;
1797 if ((start
-= len
) < buf
)
1798 return -ENAMETOOLONG
;
1799 memcpy(start
, dentry
->d_name
.name
, len
);
1800 cgrp
= cgrp
->parent
;
1804 dentry
= rcu_dereference_check(cgrp
->dentry
,
1805 cgroup_lock_is_held());
1809 return -ENAMETOOLONG
;
1812 memmove(buf
, start
, buf
+ buflen
- start
);
1815 EXPORT_SYMBOL_GPL(cgroup_path
);
1818 * Control Group taskset
1820 struct task_and_cgroup
{
1821 struct task_struct
*task
;
1822 struct cgroup
*cgrp
;
1826 struct cgroup_taskset
{
1827 struct task_and_cgroup single
;
1828 struct flex_array
*tc_array
;
1831 struct cgroup
*cur_cgrp
;
1835 * cgroup_taskset_first - reset taskset and return the first task
1836 * @tset: taskset of interest
1838 * @tset iteration is initialized and the first task is returned.
1840 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1842 if (tset
->tc_array
) {
1844 return cgroup_taskset_next(tset
);
1846 tset
->cur_cgrp
= tset
->single
.cgrp
;
1847 return tset
->single
.task
;
1850 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1853 * cgroup_taskset_next - iterate to the next task in taskset
1854 * @tset: taskset of interest
1856 * Return the next task in @tset. Iteration must have been initialized
1857 * with cgroup_taskset_first().
1859 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1861 struct task_and_cgroup
*tc
;
1863 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1866 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1867 tset
->cur_cgrp
= tc
->cgrp
;
1870 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1873 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1874 * @tset: taskset of interest
1876 * Return the cgroup for the current (last returned) task of @tset. This
1877 * function must be preceded by either cgroup_taskset_first() or
1878 * cgroup_taskset_next().
1880 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1882 return tset
->cur_cgrp
;
1884 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1887 * cgroup_taskset_size - return the number of tasks in taskset
1888 * @tset: taskset of interest
1890 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1892 return tset
->tc_array
? tset
->tc_array_len
: 1;
1894 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1898 * cgroup_task_migrate - move a task from one cgroup to another.
1900 * 'guarantee' is set if the caller promises that a new css_set for the task
1901 * will already exist. If not set, this function might sleep, and can fail with
1902 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1904 static void cgroup_task_migrate(struct cgroup
*cgrp
, struct cgroup
*oldcgrp
,
1905 struct task_struct
*tsk
, struct css_set
*newcg
)
1907 struct css_set
*oldcg
;
1910 * We are synchronized through threadgroup_lock() against PF_EXITING
1911 * setting such that we can't race against cgroup_exit() changing the
1912 * css_set to init_css_set and dropping the old one.
1914 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1915 oldcg
= tsk
->cgroups
;
1918 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1921 /* Update the css_set linked lists if we're using them */
1922 write_lock(&css_set_lock
);
1923 if (!list_empty(&tsk
->cg_list
))
1924 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1925 write_unlock(&css_set_lock
);
1928 * We just gained a reference on oldcg by taking it from the task. As
1929 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1930 * it here; it will be freed under RCU.
1934 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1938 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1939 * @cgrp: the cgroup the task is attaching to
1940 * @tsk: the task to be attached
1942 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1945 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1948 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1949 struct cgroup
*oldcgrp
;
1950 struct cgroupfs_root
*root
= cgrp
->root
;
1951 struct cgroup_taskset tset
= { };
1952 struct css_set
*newcg
;
1954 /* @tsk either already exited or can't exit until the end */
1955 if (tsk
->flags
& PF_EXITING
)
1958 /* Nothing to do if the task is already in that cgroup */
1959 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1960 if (cgrp
== oldcgrp
)
1963 tset
.single
.task
= tsk
;
1964 tset
.single
.cgrp
= oldcgrp
;
1966 for_each_subsys(root
, ss
) {
1967 if (ss
->can_attach
) {
1968 retval
= ss
->can_attach(cgrp
, &tset
);
1971 * Remember on which subsystem the can_attach()
1972 * failed, so that we only call cancel_attach()
1973 * against the subsystems whose can_attach()
1974 * succeeded. (See below)
1982 newcg
= find_css_set(tsk
->cgroups
, cgrp
);
1988 cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, newcg
);
1990 for_each_subsys(root
, ss
) {
1992 ss
->attach(cgrp
, &tset
);
1998 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1999 * is no longer empty.
2001 cgroup_wakeup_rmdir_waiter(cgrp
);
2004 for_each_subsys(root
, ss
) {
2005 if (ss
== failed_ss
)
2007 * This subsystem was the one that failed the
2008 * can_attach() check earlier, so we don't need
2009 * to call cancel_attach() against it or any
2010 * remaining subsystems.
2013 if (ss
->cancel_attach
)
2014 ss
->cancel_attach(cgrp
, &tset
);
2021 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2022 * @from: attach to all cgroups of a given task
2023 * @tsk: the task to be attached
2025 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2027 struct cgroupfs_root
*root
;
2031 for_each_active_root(root
) {
2032 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
2034 retval
= cgroup_attach_task(from_cg
, tsk
);
2042 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2045 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2046 * @cgrp: the cgroup to attach to
2047 * @leader: the threadgroup leader task_struct of the group to be attached
2049 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2050 * task_lock of each thread in leader's threadgroup individually in turn.
2052 static int cgroup_attach_proc(struct cgroup
*cgrp
, struct task_struct
*leader
)
2054 int retval
, i
, group_size
;
2055 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
2056 /* guaranteed to be initialized later, but the compiler needs this */
2057 struct cgroupfs_root
*root
= cgrp
->root
;
2058 /* threadgroup list cursor and array */
2059 struct task_struct
*tsk
;
2060 struct task_and_cgroup
*tc
;
2061 struct flex_array
*group
;
2062 struct cgroup_taskset tset
= { };
2065 * step 0: in order to do expensive, possibly blocking operations for
2066 * every thread, we cannot iterate the thread group list, since it needs
2067 * rcu or tasklist locked. instead, build an array of all threads in the
2068 * group - group_rwsem prevents new threads from appearing, and if
2069 * threads exit, this will just be an over-estimate.
2071 group_size
= get_nr_threads(leader
);
2072 /* flex_array supports very large thread-groups better than kmalloc. */
2073 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2076 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2077 retval
= flex_array_prealloc(group
, 0, group_size
- 1, GFP_KERNEL
);
2079 goto out_free_group_list
;
2084 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2085 * already PF_EXITING could be freed from underneath us unless we
2086 * take an rcu_read_lock.
2090 struct task_and_cgroup ent
;
2092 /* @tsk either already exited or can't exit until the end */
2093 if (tsk
->flags
& PF_EXITING
)
2096 /* as per above, nr_threads may decrease, but not increase. */
2097 BUG_ON(i
>= group_size
);
2099 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2100 /* nothing to do if this task is already in the cgroup */
2101 if (ent
.cgrp
== cgrp
)
2104 * saying GFP_ATOMIC has no effect here because we did prealloc
2105 * earlier, but it's good form to communicate our expectations.
2107 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2108 BUG_ON(retval
!= 0);
2110 } while_each_thread(leader
, tsk
);
2112 /* remember the number of threads in the array for later. */
2114 tset
.tc_array
= group
;
2115 tset
.tc_array_len
= group_size
;
2117 /* methods shouldn't be called if no task is actually migrating */
2120 goto out_free_group_list
;
2123 * step 1: check that we can legitimately attach to the cgroup.
2125 for_each_subsys(root
, ss
) {
2126 if (ss
->can_attach
) {
2127 retval
= ss
->can_attach(cgrp
, &tset
);
2130 goto out_cancel_attach
;
2136 * step 2: make sure css_sets exist for all threads to be migrated.
2137 * we use find_css_set, which allocates a new one if necessary.
2139 for (i
= 0; i
< group_size
; i
++) {
2140 tc
= flex_array_get(group
, i
);
2141 tc
->cg
= find_css_set(tc
->task
->cgroups
, cgrp
);
2144 goto out_put_css_set_refs
;
2149 * step 3: now that we're guaranteed success wrt the css_sets,
2150 * proceed to move all tasks to the new cgroup. There are no
2151 * failure cases after here, so this is the commit point.
2153 for (i
= 0; i
< group_size
; i
++) {
2154 tc
= flex_array_get(group
, i
);
2155 cgroup_task_migrate(cgrp
, tc
->cgrp
, tc
->task
, tc
->cg
);
2157 /* nothing is sensitive to fork() after this point. */
2160 * step 4: do subsystem attach callbacks.
2162 for_each_subsys(root
, ss
) {
2164 ss
->attach(cgrp
, &tset
);
2168 * step 5: success! and cleanup
2171 cgroup_wakeup_rmdir_waiter(cgrp
);
2173 out_put_css_set_refs
:
2175 for (i
= 0; i
< group_size
; i
++) {
2176 tc
= flex_array_get(group
, i
);
2179 put_css_set(tc
->cg
);
2184 for_each_subsys(root
, ss
) {
2185 if (ss
== failed_ss
)
2187 if (ss
->cancel_attach
)
2188 ss
->cancel_attach(cgrp
, &tset
);
2191 out_free_group_list
:
2192 flex_array_free(group
);
2197 * Find the task_struct of the task to attach by vpid and pass it along to the
2198 * function to attach either it or all tasks in its threadgroup. Will lock
2199 * cgroup_mutex and threadgroup; may take task_lock of task.
2201 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2203 struct task_struct
*tsk
;
2204 const struct cred
*cred
= current_cred(), *tcred
;
2207 if (!cgroup_lock_live_group(cgrp
))
2213 tsk
= find_task_by_vpid(pid
);
2217 goto out_unlock_cgroup
;
2220 * even if we're attaching all tasks in the thread group, we
2221 * only need to check permissions on one of them.
2223 tcred
= __task_cred(tsk
);
2224 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2225 !uid_eq(cred
->euid
, tcred
->uid
) &&
2226 !uid_eq(cred
->euid
, tcred
->suid
)) {
2229 goto out_unlock_cgroup
;
2235 tsk
= tsk
->group_leader
;
2238 * Workqueue threads may acquire PF_THREAD_BOUND and become
2239 * trapped in a cpuset, or RT worker may be born in a cgroup
2240 * with no rt_runtime allocated. Just say no.
2242 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_THREAD_BOUND
)) {
2245 goto out_unlock_cgroup
;
2248 get_task_struct(tsk
);
2251 threadgroup_lock(tsk
);
2253 if (!thread_group_leader(tsk
)) {
2255 * a race with de_thread from another thread's exec()
2256 * may strip us of our leadership, if this happens,
2257 * there is no choice but to throw this task away and
2258 * try again; this is
2259 * "double-double-toil-and-trouble-check locking".
2261 threadgroup_unlock(tsk
);
2262 put_task_struct(tsk
);
2263 goto retry_find_task
;
2265 ret
= cgroup_attach_proc(cgrp
, tsk
);
2267 ret
= cgroup_attach_task(cgrp
, tsk
);
2268 threadgroup_unlock(tsk
);
2270 put_task_struct(tsk
);
2276 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2278 return attach_task_by_pid(cgrp
, pid
, false);
2281 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2283 return attach_task_by_pid(cgrp
, tgid
, true);
2287 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2288 * @cgrp: the cgroup to be checked for liveness
2290 * On success, returns true; the lock should be later released with
2291 * cgroup_unlock(). On failure returns false with no lock held.
2293 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
2295 mutex_lock(&cgroup_mutex
);
2296 if (cgroup_is_removed(cgrp
)) {
2297 mutex_unlock(&cgroup_mutex
);
2302 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
2304 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2307 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2308 if (strlen(buffer
) >= PATH_MAX
)
2310 if (!cgroup_lock_live_group(cgrp
))
2312 mutex_lock(&cgroup_root_mutex
);
2313 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2314 mutex_unlock(&cgroup_root_mutex
);
2319 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2320 struct seq_file
*seq
)
2322 if (!cgroup_lock_live_group(cgrp
))
2324 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2325 seq_putc(seq
, '\n');
2330 /* A buffer size big enough for numbers or short strings */
2331 #define CGROUP_LOCAL_BUFFER_SIZE 64
2333 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2335 const char __user
*userbuf
,
2336 size_t nbytes
, loff_t
*unused_ppos
)
2338 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2344 if (nbytes
>= sizeof(buffer
))
2346 if (copy_from_user(buffer
, userbuf
, nbytes
))
2349 buffer
[nbytes
] = 0; /* nul-terminate */
2350 if (cft
->write_u64
) {
2351 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2354 retval
= cft
->write_u64(cgrp
, cft
, val
);
2356 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2359 retval
= cft
->write_s64(cgrp
, cft
, val
);
2366 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2368 const char __user
*userbuf
,
2369 size_t nbytes
, loff_t
*unused_ppos
)
2371 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2373 size_t max_bytes
= cft
->max_write_len
;
2374 char *buffer
= local_buffer
;
2377 max_bytes
= sizeof(local_buffer
) - 1;
2378 if (nbytes
>= max_bytes
)
2380 /* Allocate a dynamic buffer if we need one */
2381 if (nbytes
>= sizeof(local_buffer
)) {
2382 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2386 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2391 buffer
[nbytes
] = 0; /* nul-terminate */
2392 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2396 if (buffer
!= local_buffer
)
2401 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2402 size_t nbytes
, loff_t
*ppos
)
2404 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2405 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2407 if (cgroup_is_removed(cgrp
))
2410 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2411 if (cft
->write_u64
|| cft
->write_s64
)
2412 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2413 if (cft
->write_string
)
2414 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2416 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2417 return ret
? ret
: nbytes
;
2422 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2424 char __user
*buf
, size_t nbytes
,
2427 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2428 u64 val
= cft
->read_u64(cgrp
, cft
);
2429 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2431 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2434 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2436 char __user
*buf
, size_t nbytes
,
2439 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2440 s64 val
= cft
->read_s64(cgrp
, cft
);
2441 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2443 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2446 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2447 size_t nbytes
, loff_t
*ppos
)
2449 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2450 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2452 if (cgroup_is_removed(cgrp
))
2456 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2458 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2460 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2465 * seqfile ops/methods for returning structured data. Currently just
2466 * supports string->u64 maps, but can be extended in future.
2469 struct cgroup_seqfile_state
{
2471 struct cgroup
*cgroup
;
2474 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2476 struct seq_file
*sf
= cb
->state
;
2477 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2480 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2482 struct cgroup_seqfile_state
*state
= m
->private;
2483 struct cftype
*cft
= state
->cft
;
2484 if (cft
->read_map
) {
2485 struct cgroup_map_cb cb
= {
2486 .fill
= cgroup_map_add
,
2489 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2491 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2494 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2496 struct seq_file
*seq
= file
->private_data
;
2497 kfree(seq
->private);
2498 return single_release(inode
, file
);
2501 static const struct file_operations cgroup_seqfile_operations
= {
2503 .write
= cgroup_file_write
,
2504 .llseek
= seq_lseek
,
2505 .release
= cgroup_seqfile_release
,
2508 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2513 err
= generic_file_open(inode
, file
);
2516 cft
= __d_cft(file
->f_dentry
);
2518 if (cft
->read_map
|| cft
->read_seq_string
) {
2519 struct cgroup_seqfile_state
*state
=
2520 kzalloc(sizeof(*state
), GFP_USER
);
2524 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2525 file
->f_op
= &cgroup_seqfile_operations
;
2526 err
= single_open(file
, cgroup_seqfile_show
, state
);
2529 } else if (cft
->open
)
2530 err
= cft
->open(inode
, file
);
2537 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2539 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2541 return cft
->release(inode
, file
);
2546 * cgroup_rename - Only allow simple rename of directories in place.
2548 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2549 struct inode
*new_dir
, struct dentry
*new_dentry
)
2551 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2553 if (new_dentry
->d_inode
)
2555 if (old_dir
!= new_dir
)
2557 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2560 static const struct file_operations cgroup_file_operations
= {
2561 .read
= cgroup_file_read
,
2562 .write
= cgroup_file_write
,
2563 .llseek
= generic_file_llseek
,
2564 .open
= cgroup_file_open
,
2565 .release
= cgroup_file_release
,
2568 static const struct inode_operations cgroup_dir_inode_operations
= {
2569 .lookup
= cgroup_lookup
,
2570 .mkdir
= cgroup_mkdir
,
2571 .rmdir
= cgroup_rmdir
,
2572 .rename
= cgroup_rename
,
2575 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
2577 if (dentry
->d_name
.len
> NAME_MAX
)
2578 return ERR_PTR(-ENAMETOOLONG
);
2579 d_add(dentry
, NULL
);
2584 * Check if a file is a control file
2586 static inline struct cftype
*__file_cft(struct file
*file
)
2588 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2589 return ERR_PTR(-EINVAL
);
2590 return __d_cft(file
->f_dentry
);
2593 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2594 struct super_block
*sb
)
2596 struct inode
*inode
;
2600 if (dentry
->d_inode
)
2603 inode
= cgroup_new_inode(mode
, sb
);
2607 if (S_ISDIR(mode
)) {
2608 inode
->i_op
= &cgroup_dir_inode_operations
;
2609 inode
->i_fop
= &simple_dir_operations
;
2611 /* start off with i_nlink == 2 (for "." entry) */
2614 /* start with the directory inode held, so that we can
2615 * populate it without racing with another mkdir */
2616 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2617 } else if (S_ISREG(mode
)) {
2619 inode
->i_fop
= &cgroup_file_operations
;
2621 d_instantiate(dentry
, inode
);
2622 dget(dentry
); /* Extra count - pin the dentry in core */
2627 * cgroup_create_dir - create a directory for an object.
2628 * @cgrp: the cgroup we create the directory for. It must have a valid
2629 * ->parent field. And we are going to fill its ->dentry field.
2630 * @dentry: dentry of the new cgroup
2631 * @mode: mode to set on new directory.
2633 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
2636 struct dentry
*parent
;
2639 parent
= cgrp
->parent
->dentry
;
2640 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
2642 dentry
->d_fsdata
= cgrp
;
2643 inc_nlink(parent
->d_inode
);
2644 rcu_assign_pointer(cgrp
->dentry
, dentry
);
2653 * cgroup_file_mode - deduce file mode of a control file
2654 * @cft: the control file in question
2656 * returns cft->mode if ->mode is not 0
2657 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2658 * returns S_IRUGO if it has only a read handler
2659 * returns S_IWUSR if it has only a write hander
2661 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2668 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2669 cft
->read_map
|| cft
->read_seq_string
)
2672 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2673 cft
->write_string
|| cft
->trigger
)
2679 static int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2680 const struct cftype
*cft
)
2682 struct dentry
*dir
= cgrp
->dentry
;
2683 struct cgroup
*parent
= __d_cgrp(dir
);
2684 struct dentry
*dentry
;
2688 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2690 /* does @cft->flags tell us to skip creation on @cgrp? */
2691 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2693 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2696 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2697 strcpy(name
, subsys
->name
);
2700 strcat(name
, cft
->name
);
2702 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2704 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2708 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2709 if (IS_ERR(dentry
)) {
2710 error
= PTR_ERR(dentry
);
2714 mode
= cgroup_file_mode(cft
);
2715 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2717 cfe
->type
= (void *)cft
;
2718 cfe
->dentry
= dentry
;
2719 dentry
->d_fsdata
= cfe
;
2720 list_add_tail(&cfe
->node
, &parent
->files
);
2729 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2730 const struct cftype cfts
[], bool is_add
)
2732 const struct cftype
*cft
;
2735 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2737 err
= cgroup_add_file(cgrp
, subsys
, cft
);
2739 err
= cgroup_rm_file(cgrp
, cft
);
2741 pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2742 is_add
? "add" : "remove", cft
->name
, err
);
2749 static DEFINE_MUTEX(cgroup_cft_mutex
);
2751 static void cgroup_cfts_prepare(void)
2752 __acquires(&cgroup_cft_mutex
) __acquires(&cgroup_mutex
)
2755 * Thanks to the entanglement with vfs inode locking, we can't walk
2756 * the existing cgroups under cgroup_mutex and create files.
2757 * Instead, we increment reference on all cgroups and build list of
2758 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2759 * exclusive access to the field.
2761 mutex_lock(&cgroup_cft_mutex
);
2762 mutex_lock(&cgroup_mutex
);
2765 static void cgroup_cfts_commit(struct cgroup_subsys
*ss
,
2766 const struct cftype
*cfts
, bool is_add
)
2767 __releases(&cgroup_mutex
) __releases(&cgroup_cft_mutex
)
2770 struct cgroup
*cgrp
, *n
;
2772 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2773 if (cfts
&& ss
->root
!= &rootnode
) {
2774 list_for_each_entry(cgrp
, &ss
->root
->allcg_list
, allcg_node
) {
2776 list_add_tail(&cgrp
->cft_q_node
, &pending
);
2780 mutex_unlock(&cgroup_mutex
);
2783 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2784 * files for all cgroups which were created before.
2786 list_for_each_entry_safe(cgrp
, n
, &pending
, cft_q_node
) {
2787 struct inode
*inode
= cgrp
->dentry
->d_inode
;
2789 mutex_lock(&inode
->i_mutex
);
2790 mutex_lock(&cgroup_mutex
);
2791 if (!cgroup_is_removed(cgrp
))
2792 cgroup_addrm_files(cgrp
, ss
, cfts
, is_add
);
2793 mutex_unlock(&cgroup_mutex
);
2794 mutex_unlock(&inode
->i_mutex
);
2796 list_del_init(&cgrp
->cft_q_node
);
2800 mutex_unlock(&cgroup_cft_mutex
);
2804 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2805 * @ss: target cgroup subsystem
2806 * @cfts: zero-length name terminated array of cftypes
2808 * Register @cfts to @ss. Files described by @cfts are created for all
2809 * existing cgroups to which @ss is attached and all future cgroups will
2810 * have them too. This function can be called anytime whether @ss is
2813 * Returns 0 on successful registration, -errno on failure. Note that this
2814 * function currently returns 0 as long as @cfts registration is successful
2815 * even if some file creation attempts on existing cgroups fail.
2817 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, const struct cftype
*cfts
)
2819 struct cftype_set
*set
;
2821 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2825 cgroup_cfts_prepare();
2827 list_add_tail(&set
->node
, &ss
->cftsets
);
2828 cgroup_cfts_commit(ss
, cfts
, true);
2832 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2835 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2836 * @ss: target cgroup subsystem
2837 * @cfts: zero-length name terminated array of cftypes
2839 * Unregister @cfts from @ss. Files described by @cfts are removed from
2840 * all existing cgroups to which @ss is attached and all future cgroups
2841 * won't have them either. This function can be called anytime whether @ss
2842 * is attached or not.
2844 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2845 * registered with @ss.
2847 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, const struct cftype
*cfts
)
2849 struct cftype_set
*set
;
2851 cgroup_cfts_prepare();
2853 list_for_each_entry(set
, &ss
->cftsets
, node
) {
2854 if (set
->cfts
== cfts
) {
2855 list_del_init(&set
->node
);
2856 cgroup_cfts_commit(ss
, cfts
, false);
2861 cgroup_cfts_commit(ss
, NULL
, false);
2866 * cgroup_task_count - count the number of tasks in a cgroup.
2867 * @cgrp: the cgroup in question
2869 * Return the number of tasks in the cgroup.
2871 int cgroup_task_count(const struct cgroup
*cgrp
)
2874 struct cg_cgroup_link
*link
;
2876 read_lock(&css_set_lock
);
2877 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2878 count
+= atomic_read(&link
->cg
->refcount
);
2880 read_unlock(&css_set_lock
);
2885 * Advance a list_head iterator. The iterator should be positioned at
2886 * the start of a css_set
2888 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2889 struct cgroup_iter
*it
)
2891 struct list_head
*l
= it
->cg_link
;
2892 struct cg_cgroup_link
*link
;
2895 /* Advance to the next non-empty css_set */
2898 if (l
== &cgrp
->css_sets
) {
2902 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2904 } while (list_empty(&cg
->tasks
));
2906 it
->task
= cg
->tasks
.next
;
2910 * To reduce the fork() overhead for systems that are not actually
2911 * using their cgroups capability, we don't maintain the lists running
2912 * through each css_set to its tasks until we see the list actually
2913 * used - in other words after the first call to cgroup_iter_start().
2915 static void cgroup_enable_task_cg_lists(void)
2917 struct task_struct
*p
, *g
;
2918 write_lock(&css_set_lock
);
2919 use_task_css_set_links
= 1;
2921 * We need tasklist_lock because RCU is not safe against
2922 * while_each_thread(). Besides, a forking task that has passed
2923 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2924 * is not guaranteed to have its child immediately visible in the
2925 * tasklist if we walk through it with RCU.
2927 read_lock(&tasklist_lock
);
2928 do_each_thread(g
, p
) {
2931 * We should check if the process is exiting, otherwise
2932 * it will race with cgroup_exit() in that the list
2933 * entry won't be deleted though the process has exited.
2935 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2936 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2938 } while_each_thread(g
, p
);
2939 read_unlock(&tasklist_lock
);
2940 write_unlock(&css_set_lock
);
2943 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2944 __acquires(css_set_lock
)
2947 * The first time anyone tries to iterate across a cgroup,
2948 * we need to enable the list linking each css_set to its
2949 * tasks, and fix up all existing tasks.
2951 if (!use_task_css_set_links
)
2952 cgroup_enable_task_cg_lists();
2954 read_lock(&css_set_lock
);
2955 it
->cg_link
= &cgrp
->css_sets
;
2956 cgroup_advance_iter(cgrp
, it
);
2959 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
2960 struct cgroup_iter
*it
)
2962 struct task_struct
*res
;
2963 struct list_head
*l
= it
->task
;
2964 struct cg_cgroup_link
*link
;
2966 /* If the iterator cg is NULL, we have no tasks */
2969 res
= list_entry(l
, struct task_struct
, cg_list
);
2970 /* Advance iterator to find next entry */
2972 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
2973 if (l
== &link
->cg
->tasks
) {
2974 /* We reached the end of this task list - move on to
2975 * the next cg_cgroup_link */
2976 cgroup_advance_iter(cgrp
, it
);
2983 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2984 __releases(css_set_lock
)
2986 read_unlock(&css_set_lock
);
2989 static inline int started_after_time(struct task_struct
*t1
,
2990 struct timespec
*time
,
2991 struct task_struct
*t2
)
2993 int start_diff
= timespec_compare(&t1
->start_time
, time
);
2994 if (start_diff
> 0) {
2996 } else if (start_diff
< 0) {
3000 * Arbitrarily, if two processes started at the same
3001 * time, we'll say that the lower pointer value
3002 * started first. Note that t2 may have exited by now
3003 * so this may not be a valid pointer any longer, but
3004 * that's fine - it still serves to distinguish
3005 * between two tasks started (effectively) simultaneously.
3012 * This function is a callback from heap_insert() and is used to order
3014 * In this case we order the heap in descending task start time.
3016 static inline int started_after(void *p1
, void *p2
)
3018 struct task_struct
*t1
= p1
;
3019 struct task_struct
*t2
= p2
;
3020 return started_after_time(t1
, &t2
->start_time
, t2
);
3024 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3025 * @scan: struct cgroup_scanner containing arguments for the scan
3027 * Arguments include pointers to callback functions test_task() and
3029 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3030 * and if it returns true, call process_task() for it also.
3031 * The test_task pointer may be NULL, meaning always true (select all tasks).
3032 * Effectively duplicates cgroup_iter_{start,next,end}()
3033 * but does not lock css_set_lock for the call to process_task().
3034 * The struct cgroup_scanner may be embedded in any structure of the caller's
3036 * It is guaranteed that process_task() will act on every task that
3037 * is a member of the cgroup for the duration of this call. This
3038 * function may or may not call process_task() for tasks that exit
3039 * or move to a different cgroup during the call, or are forked or
3040 * move into the cgroup during the call.
3042 * Note that test_task() may be called with locks held, and may in some
3043 * situations be called multiple times for the same task, so it should
3045 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3046 * pre-allocated and will be used for heap operations (and its "gt" member will
3047 * be overwritten), else a temporary heap will be used (allocation of which
3048 * may cause this function to fail).
3050 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
3053 struct cgroup_iter it
;
3054 struct task_struct
*p
, *dropped
;
3055 /* Never dereference latest_task, since it's not refcounted */
3056 struct task_struct
*latest_task
= NULL
;
3057 struct ptr_heap tmp_heap
;
3058 struct ptr_heap
*heap
;
3059 struct timespec latest_time
= { 0, 0 };
3062 /* The caller supplied our heap and pre-allocated its memory */
3064 heap
->gt
= &started_after
;
3066 /* We need to allocate our own heap memory */
3068 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3070 /* cannot allocate the heap */
3076 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3077 * to determine which are of interest, and using the scanner's
3078 * "process_task" callback to process any of them that need an update.
3079 * Since we don't want to hold any locks during the task updates,
3080 * gather tasks to be processed in a heap structure.
3081 * The heap is sorted by descending task start time.
3082 * If the statically-sized heap fills up, we overflow tasks that
3083 * started later, and in future iterations only consider tasks that
3084 * started after the latest task in the previous pass. This
3085 * guarantees forward progress and that we don't miss any tasks.
3088 cgroup_iter_start(scan
->cg
, &it
);
3089 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
3091 * Only affect tasks that qualify per the caller's callback,
3092 * if he provided one
3094 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
3097 * Only process tasks that started after the last task
3100 if (!started_after_time(p
, &latest_time
, latest_task
))
3102 dropped
= heap_insert(heap
, p
);
3103 if (dropped
== NULL
) {
3105 * The new task was inserted; the heap wasn't
3109 } else if (dropped
!= p
) {
3111 * The new task was inserted, and pushed out a
3115 put_task_struct(dropped
);
3118 * Else the new task was newer than anything already in
3119 * the heap and wasn't inserted
3122 cgroup_iter_end(scan
->cg
, &it
);
3125 for (i
= 0; i
< heap
->size
; i
++) {
3126 struct task_struct
*q
= heap
->ptrs
[i
];
3128 latest_time
= q
->start_time
;
3131 /* Process the task per the caller's callback */
3132 scan
->process_task(q
, scan
);
3136 * If we had to process any tasks at all, scan again
3137 * in case some of them were in the middle of forking
3138 * children that didn't get processed.
3139 * Not the most efficient way to do it, but it avoids
3140 * having to take callback_mutex in the fork path
3144 if (heap
== &tmp_heap
)
3145 heap_free(&tmp_heap
);
3150 * Stuff for reading the 'tasks'/'procs' files.
3152 * Reading this file can return large amounts of data if a cgroup has
3153 * *lots* of attached tasks. So it may need several calls to read(),
3154 * but we cannot guarantee that the information we produce is correct
3155 * unless we produce it entirely atomically.
3159 /* which pidlist file are we talking about? */
3160 enum cgroup_filetype
{
3166 * A pidlist is a list of pids that virtually represents the contents of one
3167 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3168 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3171 struct cgroup_pidlist
{
3173 * used to find which pidlist is wanted. doesn't change as long as
3174 * this particular list stays in the list.
3176 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3179 /* how many elements the above list has */
3181 /* how many files are using the current array */
3183 /* each of these stored in a list by its cgroup */
3184 struct list_head links
;
3185 /* pointer to the cgroup we belong to, for list removal purposes */
3186 struct cgroup
*owner
;
3187 /* protects the other fields */
3188 struct rw_semaphore mutex
;
3192 * The following two functions "fix" the issue where there are more pids
3193 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3194 * TODO: replace with a kernel-wide solution to this problem
3196 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3197 static void *pidlist_allocate(int count
)
3199 if (PIDLIST_TOO_LARGE(count
))
3200 return vmalloc(count
* sizeof(pid_t
));
3202 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3204 static void pidlist_free(void *p
)
3206 if (is_vmalloc_addr(p
))
3211 static void *pidlist_resize(void *p
, int newcount
)
3214 /* note: if new alloc fails, old p will still be valid either way */
3215 if (is_vmalloc_addr(p
)) {
3216 newlist
= vmalloc(newcount
* sizeof(pid_t
));
3219 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
3222 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
3228 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3229 * If the new stripped list is sufficiently smaller and there's enough memory
3230 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3231 * number of unique elements.
3233 /* is the size difference enough that we should re-allocate the array? */
3234 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3235 static int pidlist_uniq(pid_t
**p
, int length
)
3242 * we presume the 0th element is unique, so i starts at 1. trivial
3243 * edge cases first; no work needs to be done for either
3245 if (length
== 0 || length
== 1)
3247 /* src and dest walk down the list; dest counts unique elements */
3248 for (src
= 1; src
< length
; src
++) {
3249 /* find next unique element */
3250 while (list
[src
] == list
[src
-1]) {
3255 /* dest always points to where the next unique element goes */
3256 list
[dest
] = list
[src
];
3261 * if the length difference is large enough, we want to allocate a
3262 * smaller buffer to save memory. if this fails due to out of memory,
3263 * we'll just stay with what we've got.
3265 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
3266 newlist
= pidlist_resize(list
, dest
);
3273 static int cmppid(const void *a
, const void *b
)
3275 return *(pid_t
*)a
- *(pid_t
*)b
;
3279 * find the appropriate pidlist for our purpose (given procs vs tasks)
3280 * returns with the lock on that pidlist already held, and takes care
3281 * of the use count, or returns NULL with no locks held if we're out of
3284 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3285 enum cgroup_filetype type
)
3287 struct cgroup_pidlist
*l
;
3288 /* don't need task_nsproxy() if we're looking at ourself */
3289 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
3292 * We can't drop the pidlist_mutex before taking the l->mutex in case
3293 * the last ref-holder is trying to remove l from the list at the same
3294 * time. Holding the pidlist_mutex precludes somebody taking whichever
3295 * list we find out from under us - compare release_pid_array().
3297 mutex_lock(&cgrp
->pidlist_mutex
);
3298 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3299 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3300 /* make sure l doesn't vanish out from under us */
3301 down_write(&l
->mutex
);
3302 mutex_unlock(&cgrp
->pidlist_mutex
);
3306 /* entry not found; create a new one */
3307 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3309 mutex_unlock(&cgrp
->pidlist_mutex
);
3312 init_rwsem(&l
->mutex
);
3313 down_write(&l
->mutex
);
3315 l
->key
.ns
= get_pid_ns(ns
);
3316 l
->use_count
= 0; /* don't increment here */
3319 list_add(&l
->links
, &cgrp
->pidlists
);
3320 mutex_unlock(&cgrp
->pidlist_mutex
);
3325 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3327 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3328 struct cgroup_pidlist
**lp
)
3332 int pid
, n
= 0; /* used for populating the array */
3333 struct cgroup_iter it
;
3334 struct task_struct
*tsk
;
3335 struct cgroup_pidlist
*l
;
3338 * If cgroup gets more users after we read count, we won't have
3339 * enough space - tough. This race is indistinguishable to the
3340 * caller from the case that the additional cgroup users didn't
3341 * show up until sometime later on.
3343 length
= cgroup_task_count(cgrp
);
3344 array
= pidlist_allocate(length
);
3347 /* now, populate the array */
3348 cgroup_iter_start(cgrp
, &it
);
3349 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3350 if (unlikely(n
== length
))
3352 /* get tgid or pid for procs or tasks file respectively */
3353 if (type
== CGROUP_FILE_PROCS
)
3354 pid
= task_tgid_vnr(tsk
);
3356 pid
= task_pid_vnr(tsk
);
3357 if (pid
> 0) /* make sure to only use valid results */
3360 cgroup_iter_end(cgrp
, &it
);
3362 /* now sort & (if procs) strip out duplicates */
3363 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3364 if (type
== CGROUP_FILE_PROCS
)
3365 length
= pidlist_uniq(&array
, length
);
3366 l
= cgroup_pidlist_find(cgrp
, type
);
3368 pidlist_free(array
);
3371 /* store array, freeing old if necessary - lock already held */
3372 pidlist_free(l
->list
);
3376 up_write(&l
->mutex
);
3382 * cgroupstats_build - build and fill cgroupstats
3383 * @stats: cgroupstats to fill information into
3384 * @dentry: A dentry entry belonging to the cgroup for which stats have
3387 * Build and fill cgroupstats so that taskstats can export it to user
3390 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3393 struct cgroup
*cgrp
;
3394 struct cgroup_iter it
;
3395 struct task_struct
*tsk
;
3398 * Validate dentry by checking the superblock operations,
3399 * and make sure it's a directory.
3401 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3402 !S_ISDIR(dentry
->d_inode
->i_mode
))
3406 cgrp
= dentry
->d_fsdata
;
3408 cgroup_iter_start(cgrp
, &it
);
3409 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3410 switch (tsk
->state
) {
3412 stats
->nr_running
++;
3414 case TASK_INTERRUPTIBLE
:
3415 stats
->nr_sleeping
++;
3417 case TASK_UNINTERRUPTIBLE
:
3418 stats
->nr_uninterruptible
++;
3421 stats
->nr_stopped
++;
3424 if (delayacct_is_task_waiting_on_io(tsk
))
3425 stats
->nr_io_wait
++;
3429 cgroup_iter_end(cgrp
, &it
);
3437 * seq_file methods for the tasks/procs files. The seq_file position is the
3438 * next pid to display; the seq_file iterator is a pointer to the pid
3439 * in the cgroup->l->list array.
3442 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3445 * Initially we receive a position value that corresponds to
3446 * one more than the last pid shown (or 0 on the first call or
3447 * after a seek to the start). Use a binary-search to find the
3448 * next pid to display, if any
3450 struct cgroup_pidlist
*l
= s
->private;
3451 int index
= 0, pid
= *pos
;
3454 down_read(&l
->mutex
);
3456 int end
= l
->length
;
3458 while (index
< end
) {
3459 int mid
= (index
+ end
) / 2;
3460 if (l
->list
[mid
] == pid
) {
3463 } else if (l
->list
[mid
] <= pid
)
3469 /* If we're off the end of the array, we're done */
3470 if (index
>= l
->length
)
3472 /* Update the abstract position to be the actual pid that we found */
3473 iter
= l
->list
+ index
;
3478 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3480 struct cgroup_pidlist
*l
= s
->private;
3484 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3486 struct cgroup_pidlist
*l
= s
->private;
3488 pid_t
*end
= l
->list
+ l
->length
;
3490 * Advance to the next pid in the array. If this goes off the
3502 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3504 return seq_printf(s
, "%d\n", *(int *)v
);
3508 * seq_operations functions for iterating on pidlists through seq_file -
3509 * independent of whether it's tasks or procs
3511 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3512 .start
= cgroup_pidlist_start
,
3513 .stop
= cgroup_pidlist_stop
,
3514 .next
= cgroup_pidlist_next
,
3515 .show
= cgroup_pidlist_show
,
3518 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3521 * the case where we're the last user of this particular pidlist will
3522 * have us remove it from the cgroup's list, which entails taking the
3523 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3524 * pidlist_mutex, we have to take pidlist_mutex first.
3526 mutex_lock(&l
->owner
->pidlist_mutex
);
3527 down_write(&l
->mutex
);
3528 BUG_ON(!l
->use_count
);
3529 if (!--l
->use_count
) {
3530 /* we're the last user if refcount is 0; remove and free */
3531 list_del(&l
->links
);
3532 mutex_unlock(&l
->owner
->pidlist_mutex
);
3533 pidlist_free(l
->list
);
3534 put_pid_ns(l
->key
.ns
);
3535 up_write(&l
->mutex
);
3539 mutex_unlock(&l
->owner
->pidlist_mutex
);
3540 up_write(&l
->mutex
);
3543 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3545 struct cgroup_pidlist
*l
;
3546 if (!(file
->f_mode
& FMODE_READ
))
3549 * the seq_file will only be initialized if the file was opened for
3550 * reading; hence we check if it's not null only in that case.
3552 l
= ((struct seq_file
*)file
->private_data
)->private;
3553 cgroup_release_pid_array(l
);
3554 return seq_release(inode
, file
);
3557 static const struct file_operations cgroup_pidlist_operations
= {
3559 .llseek
= seq_lseek
,
3560 .write
= cgroup_file_write
,
3561 .release
= cgroup_pidlist_release
,
3565 * The following functions handle opens on a file that displays a pidlist
3566 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3569 /* helper function for the two below it */
3570 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3572 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3573 struct cgroup_pidlist
*l
;
3576 /* Nothing to do for write-only files */
3577 if (!(file
->f_mode
& FMODE_READ
))
3580 /* have the array populated */
3581 retval
= pidlist_array_load(cgrp
, type
, &l
);
3584 /* configure file information */
3585 file
->f_op
= &cgroup_pidlist_operations
;
3587 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3589 cgroup_release_pid_array(l
);
3592 ((struct seq_file
*)file
->private_data
)->private = l
;
3595 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3597 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3599 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3601 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3604 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3607 return notify_on_release(cgrp
);
3610 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3614 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3616 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3618 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3623 * Unregister event and free resources.
3625 * Gets called from workqueue.
3627 static void cgroup_event_remove(struct work_struct
*work
)
3629 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3631 struct cgroup
*cgrp
= event
->cgrp
;
3633 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3635 eventfd_ctx_put(event
->eventfd
);
3641 * Gets called on POLLHUP on eventfd when user closes it.
3643 * Called with wqh->lock held and interrupts disabled.
3645 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3646 int sync
, void *key
)
3648 struct cgroup_event
*event
= container_of(wait
,
3649 struct cgroup_event
, wait
);
3650 struct cgroup
*cgrp
= event
->cgrp
;
3651 unsigned long flags
= (unsigned long)key
;
3653 if (flags
& POLLHUP
) {
3654 __remove_wait_queue(event
->wqh
, &event
->wait
);
3655 spin_lock(&cgrp
->event_list_lock
);
3656 list_del(&event
->list
);
3657 spin_unlock(&cgrp
->event_list_lock
);
3659 * We are in atomic context, but cgroup_event_remove() may
3660 * sleep, so we have to call it in workqueue.
3662 schedule_work(&event
->remove
);
3668 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3669 wait_queue_head_t
*wqh
, poll_table
*pt
)
3671 struct cgroup_event
*event
= container_of(pt
,
3672 struct cgroup_event
, pt
);
3675 add_wait_queue(wqh
, &event
->wait
);
3679 * Parse input and register new cgroup event handler.
3681 * Input must be in format '<event_fd> <control_fd> <args>'.
3682 * Interpretation of args is defined by control file implementation.
3684 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3687 struct cgroup_event
*event
= NULL
;
3688 unsigned int efd
, cfd
;
3689 struct file
*efile
= NULL
;
3690 struct file
*cfile
= NULL
;
3694 efd
= simple_strtoul(buffer
, &endp
, 10);
3699 cfd
= simple_strtoul(buffer
, &endp
, 10);
3700 if ((*endp
!= ' ') && (*endp
!= '\0'))
3704 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3708 INIT_LIST_HEAD(&event
->list
);
3709 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3710 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3711 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3713 efile
= eventfd_fget(efd
);
3714 if (IS_ERR(efile
)) {
3715 ret
= PTR_ERR(efile
);
3719 event
->eventfd
= eventfd_ctx_fileget(efile
);
3720 if (IS_ERR(event
->eventfd
)) {
3721 ret
= PTR_ERR(event
->eventfd
);
3731 /* the process need read permission on control file */
3732 /* AV: shouldn't we check that it's been opened for read instead? */
3733 ret
= inode_permission(cfile
->f_path
.dentry
->d_inode
, MAY_READ
);
3737 event
->cft
= __file_cft(cfile
);
3738 if (IS_ERR(event
->cft
)) {
3739 ret
= PTR_ERR(event
->cft
);
3743 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3748 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3749 event
->eventfd
, buffer
);
3753 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3754 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3760 * Events should be removed after rmdir of cgroup directory, but before
3761 * destroying subsystem state objects. Let's take reference to cgroup
3762 * directory dentry to do that.
3766 spin_lock(&cgrp
->event_list_lock
);
3767 list_add(&event
->list
, &cgrp
->event_list
);
3768 spin_unlock(&cgrp
->event_list_lock
);
3779 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3780 eventfd_ctx_put(event
->eventfd
);
3782 if (!IS_ERR_OR_NULL(efile
))
3790 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3793 return clone_children(cgrp
);
3796 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3801 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3803 clear_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3808 * for the common functions, 'private' gives the type of file
3810 /* for hysterical raisins, we can't put this on the older files */
3811 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3812 static struct cftype files
[] = {
3815 .open
= cgroup_tasks_open
,
3816 .write_u64
= cgroup_tasks_write
,
3817 .release
= cgroup_pidlist_release
,
3818 .mode
= S_IRUGO
| S_IWUSR
,
3821 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3822 .open
= cgroup_procs_open
,
3823 .write_u64
= cgroup_procs_write
,
3824 .release
= cgroup_pidlist_release
,
3825 .mode
= S_IRUGO
| S_IWUSR
,
3828 .name
= "notify_on_release",
3829 .read_u64
= cgroup_read_notify_on_release
,
3830 .write_u64
= cgroup_write_notify_on_release
,
3833 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3834 .write_string
= cgroup_write_event_control
,
3838 .name
= "cgroup.clone_children",
3839 .read_u64
= cgroup_clone_children_read
,
3840 .write_u64
= cgroup_clone_children_write
,
3843 .name
= "release_agent",
3844 .flags
= CFTYPE_ONLY_ON_ROOT
,
3845 .read_seq_string
= cgroup_release_agent_show
,
3846 .write_string
= cgroup_release_agent_write
,
3847 .max_write_len
= PATH_MAX
,
3852 static int cgroup_populate_dir(struct cgroup
*cgrp
)
3855 struct cgroup_subsys
*ss
;
3857 err
= cgroup_addrm_files(cgrp
, NULL
, files
, true);
3861 /* process cftsets of each subsystem */
3862 for_each_subsys(cgrp
->root
, ss
) {
3863 struct cftype_set
*set
;
3865 list_for_each_entry(set
, &ss
->cftsets
, node
)
3866 cgroup_addrm_files(cgrp
, ss
, set
->cfts
, true);
3869 /* This cgroup is ready now */
3870 for_each_subsys(cgrp
->root
, ss
) {
3871 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3873 * Update id->css pointer and make this css visible from
3874 * CSS ID functions. This pointer will be dereferened
3875 * from RCU-read-side without locks.
3878 rcu_assign_pointer(css
->id
->css
, css
);
3884 static void css_dput_fn(struct work_struct
*work
)
3886 struct cgroup_subsys_state
*css
=
3887 container_of(work
, struct cgroup_subsys_state
, dput_work
);
3889 dput(css
->cgroup
->dentry
);
3892 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
3893 struct cgroup_subsys
*ss
,
3894 struct cgroup
*cgrp
)
3897 atomic_set(&css
->refcnt
, 1);
3900 if (cgrp
== dummytop
)
3901 set_bit(CSS_ROOT
, &css
->flags
);
3902 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
3903 cgrp
->subsys
[ss
->subsys_id
] = css
;
3906 * If !clear_css_refs, css holds an extra ref to @cgrp->dentry
3907 * which is put on the last css_put(). dput() requires process
3908 * context, which css_put() may be called without. @css->dput_work
3909 * will be used to invoke dput() asynchronously from css_put().
3911 INIT_WORK(&css
->dput_work
, css_dput_fn
);
3912 if (ss
->__DEPRECATED_clear_css_refs
)
3913 set_bit(CSS_CLEAR_CSS_REFS
, &css
->flags
);
3917 * cgroup_create - create a cgroup
3918 * @parent: cgroup that will be parent of the new cgroup
3919 * @dentry: dentry of the new cgroup
3920 * @mode: mode to set on new inode
3922 * Must be called with the mutex on the parent inode held
3924 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
3927 struct cgroup
*cgrp
;
3928 struct cgroupfs_root
*root
= parent
->root
;
3930 struct cgroup_subsys
*ss
;
3931 struct super_block
*sb
= root
->sb
;
3933 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3937 /* Grab a reference on the superblock so the hierarchy doesn't
3938 * get deleted on unmount if there are child cgroups. This
3939 * can be done outside cgroup_mutex, since the sb can't
3940 * disappear while someone has an open control file on the
3942 atomic_inc(&sb
->s_active
);
3944 mutex_lock(&cgroup_mutex
);
3946 init_cgroup_housekeeping(cgrp
);
3948 cgrp
->parent
= parent
;
3949 cgrp
->root
= parent
->root
;
3950 cgrp
->top_cgroup
= parent
->top_cgroup
;
3952 if (notify_on_release(parent
))
3953 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3955 if (clone_children(parent
))
3956 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3958 for_each_subsys(root
, ss
) {
3959 struct cgroup_subsys_state
*css
= ss
->create(cgrp
);
3965 init_cgroup_css(css
, ss
, cgrp
);
3967 err
= alloc_css_id(ss
, parent
, cgrp
);
3971 /* At error, ->destroy() callback has to free assigned ID. */
3972 if (clone_children(parent
) && ss
->post_clone
)
3973 ss
->post_clone(cgrp
);
3976 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
3977 root
->number_of_cgroups
++;
3979 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
3983 /* If !clear_css_refs, each css holds a ref to the cgroup's dentry */
3984 for_each_subsys(root
, ss
)
3985 if (!ss
->__DEPRECATED_clear_css_refs
)
3988 /* The cgroup directory was pre-locked for us */
3989 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
3991 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
3993 err
= cgroup_populate_dir(cgrp
);
3994 /* If err < 0, we have a half-filled directory - oh well ;) */
3996 mutex_unlock(&cgroup_mutex
);
3997 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4003 list_del(&cgrp
->sibling
);
4004 root
->number_of_cgroups
--;
4008 for_each_subsys(root
, ss
) {
4009 if (cgrp
->subsys
[ss
->subsys_id
])
4013 mutex_unlock(&cgroup_mutex
);
4015 /* Release the reference count that we took on the superblock */
4016 deactivate_super(sb
);
4022 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4024 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4026 /* the vfs holds inode->i_mutex already */
4027 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4031 * Check the reference count on each subsystem. Since we already
4032 * established that there are no tasks in the cgroup, if the css refcount
4033 * is also 1, then there should be no outstanding references, so the
4034 * subsystem is safe to destroy. We scan across all subsystems rather than
4035 * using the per-hierarchy linked list of mounted subsystems since we can
4036 * be called via check_for_release() with no synchronization other than
4037 * RCU, and the subsystem linked list isn't RCU-safe.
4039 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
4044 * We won't need to lock the subsys array, because the subsystems
4045 * we're concerned about aren't going anywhere since our cgroup root
4046 * has a reference on them.
4048 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4049 struct cgroup_subsys
*ss
= subsys
[i
];
4050 struct cgroup_subsys_state
*css
;
4052 /* Skip subsystems not present or not in this hierarchy */
4053 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
4056 css
= cgrp
->subsys
[ss
->subsys_id
];
4058 * When called from check_for_release() it's possible
4059 * that by this point the cgroup has been removed
4060 * and the css deleted. But a false-positive doesn't
4061 * matter, since it can only happen if the cgroup
4062 * has been deleted and hence no longer needs the
4063 * release agent to be called anyway.
4065 if (css
&& css_refcnt(css
) > 1)
4072 * Atomically mark all (or else none) of the cgroup's CSS objects as
4073 * CSS_REMOVED. Return true on success, or false if the cgroup has
4074 * busy subsystems. Call with cgroup_mutex held
4076 * Depending on whether a subsys has __DEPRECATED_clear_css_refs set or
4077 * not, cgroup removal behaves differently.
4079 * If clear is set, css refcnt for the subsystem should be zero before
4080 * cgroup removal can be committed. This is implemented by
4081 * CGRP_WAIT_ON_RMDIR and retry logic around ->pre_destroy(), which may be
4082 * called multiple times until all css refcnts reach zero and is allowed to
4083 * veto removal on any invocation. This behavior is deprecated and will be
4084 * removed as soon as the existing user (memcg) is updated.
4086 * If clear is not set, each css holds an extra reference to the cgroup's
4087 * dentry and cgroup removal proceeds regardless of css refs.
4088 * ->pre_destroy() will be called at least once and is not allowed to fail.
4089 * On the last put of each css, whenever that may be, the extra dentry ref
4090 * is put so that dentry destruction happens only after all css's are
4093 static int cgroup_clear_css_refs(struct cgroup
*cgrp
)
4095 struct cgroup_subsys
*ss
;
4096 unsigned long flags
;
4097 bool failed
= false;
4099 local_irq_save(flags
);
4102 * Block new css_tryget() by deactivating refcnt. If all refcnts
4103 * for subsystems w/ clear_css_refs set were 1 at the moment of
4104 * deactivation, we succeeded.
4106 for_each_subsys(cgrp
->root
, ss
) {
4107 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4109 WARN_ON(atomic_read(&css
->refcnt
) < 0);
4110 atomic_add(CSS_DEACT_BIAS
, &css
->refcnt
);
4112 if (ss
->__DEPRECATED_clear_css_refs
)
4113 failed
|= css_refcnt(css
) != 1;
4117 * If succeeded, set REMOVED and put all the base refs; otherwise,
4118 * restore refcnts to positive values. Either way, all in-progress
4119 * css_tryget() will be released.
4121 for_each_subsys(cgrp
->root
, ss
) {
4122 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4125 set_bit(CSS_REMOVED
, &css
->flags
);
4128 atomic_sub(CSS_DEACT_BIAS
, &css
->refcnt
);
4132 local_irq_restore(flags
);
4136 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4138 struct cgroup
*cgrp
= dentry
->d_fsdata
;
4140 struct cgroup
*parent
;
4142 struct cgroup_event
*event
, *tmp
;
4145 /* the vfs holds both inode->i_mutex already */
4147 mutex_lock(&cgroup_mutex
);
4148 if (atomic_read(&cgrp
->count
) != 0) {
4149 mutex_unlock(&cgroup_mutex
);
4152 if (!list_empty(&cgrp
->children
)) {
4153 mutex_unlock(&cgroup_mutex
);
4156 mutex_unlock(&cgroup_mutex
);
4159 * In general, subsystem has no css->refcnt after pre_destroy(). But
4160 * in racy cases, subsystem may have to get css->refcnt after
4161 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4162 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4163 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4164 * and subsystem's reference count handling. Please see css_get/put
4165 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4167 set_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4170 * Call pre_destroy handlers of subsys. Notify subsystems
4171 * that rmdir() request comes.
4173 ret
= cgroup_call_pre_destroy(cgrp
);
4175 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4179 mutex_lock(&cgroup_mutex
);
4180 parent
= cgrp
->parent
;
4181 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
)) {
4182 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4183 mutex_unlock(&cgroup_mutex
);
4186 prepare_to_wait(&cgroup_rmdir_waitq
, &wait
, TASK_INTERRUPTIBLE
);
4187 if (!cgroup_clear_css_refs(cgrp
)) {
4188 mutex_unlock(&cgroup_mutex
);
4190 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4191 * prepare_to_wait(), we need to check this flag.
4193 if (test_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
))
4195 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4196 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4197 if (signal_pending(current
))
4201 /* NO css_tryget() can success after here. */
4202 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4203 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4205 raw_spin_lock(&release_list_lock
);
4206 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
4207 if (!list_empty(&cgrp
->release_list
))
4208 list_del_init(&cgrp
->release_list
);
4209 raw_spin_unlock(&release_list_lock
);
4211 /* delete this cgroup from parent->children */
4212 list_del_init(&cgrp
->sibling
);
4214 list_del_init(&cgrp
->allcg_node
);
4216 d
= dget(cgrp
->dentry
);
4218 cgroup_d_remove_dir(d
);
4221 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4222 check_for_release(parent
);
4225 * Unregister events and notify userspace.
4226 * Notify userspace about cgroup removing only after rmdir of cgroup
4227 * directory to avoid race between userspace and kernelspace
4229 spin_lock(&cgrp
->event_list_lock
);
4230 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4231 list_del(&event
->list
);
4232 remove_wait_queue(event
->wqh
, &event
->wait
);
4233 eventfd_signal(event
->eventfd
, 1);
4234 schedule_work(&event
->remove
);
4236 spin_unlock(&cgrp
->event_list_lock
);
4238 mutex_unlock(&cgroup_mutex
);
4242 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4244 INIT_LIST_HEAD(&ss
->cftsets
);
4247 * base_cftset is embedded in subsys itself, no need to worry about
4250 if (ss
->base_cftypes
) {
4251 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4252 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4256 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4258 struct cgroup_subsys_state
*css
;
4260 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4262 /* init base cftset */
4263 cgroup_init_cftsets(ss
);
4265 /* Create the top cgroup state for this subsystem */
4266 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4267 ss
->root
= &rootnode
;
4268 css
= ss
->create(dummytop
);
4269 /* We don't handle early failures gracefully */
4270 BUG_ON(IS_ERR(css
));
4271 init_cgroup_css(css
, ss
, dummytop
);
4273 /* Update the init_css_set to contain a subsys
4274 * pointer to this state - since the subsystem is
4275 * newly registered, all tasks and hence the
4276 * init_css_set is in the subsystem's top cgroup. */
4277 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
4279 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4281 /* At system boot, before all subsystems have been
4282 * registered, no tasks have been forked, so we don't
4283 * need to invoke fork callbacks here. */
4284 BUG_ON(!list_empty(&init_task
.tasks
));
4288 /* this function shouldn't be used with modular subsystems, since they
4289 * need to register a subsys_id, among other things */
4294 * cgroup_load_subsys: load and register a modular subsystem at runtime
4295 * @ss: the subsystem to load
4297 * This function should be called in a modular subsystem's initcall. If the
4298 * subsystem is built as a module, it will be assigned a new subsys_id and set
4299 * up for use. If the subsystem is built-in anyway, work is delegated to the
4300 * simpler cgroup_init_subsys.
4302 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4305 struct cgroup_subsys_state
*css
;
4307 /* check name and function validity */
4308 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4309 ss
->create
== NULL
|| ss
->destroy
== NULL
)
4313 * we don't support callbacks in modular subsystems. this check is
4314 * before the ss->module check for consistency; a subsystem that could
4315 * be a module should still have no callbacks even if the user isn't
4316 * compiling it as one.
4318 if (ss
->fork
|| ss
->exit
)
4322 * an optionally modular subsystem is built-in: we want to do nothing,
4323 * since cgroup_init_subsys will have already taken care of it.
4325 if (ss
->module
== NULL
) {
4326 /* a few sanity checks */
4327 BUG_ON(ss
->subsys_id
>= CGROUP_BUILTIN_SUBSYS_COUNT
);
4328 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4332 /* init base cftset */
4333 cgroup_init_cftsets(ss
);
4336 * need to register a subsys id before anything else - for example,
4337 * init_cgroup_css needs it.
4339 mutex_lock(&cgroup_mutex
);
4340 /* find the first empty slot in the array */
4341 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4342 if (subsys
[i
] == NULL
)
4345 if (i
== CGROUP_SUBSYS_COUNT
) {
4346 /* maximum number of subsystems already registered! */
4347 mutex_unlock(&cgroup_mutex
);
4350 /* assign ourselves the subsys_id */
4355 * no ss->create seems to need anything important in the ss struct, so
4356 * this can happen first (i.e. before the rootnode attachment).
4358 css
= ss
->create(dummytop
);
4360 /* failure case - need to deassign the subsys[] slot. */
4362 mutex_unlock(&cgroup_mutex
);
4363 return PTR_ERR(css
);
4366 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4367 ss
->root
= &rootnode
;
4369 /* our new subsystem will be attached to the dummy hierarchy. */
4370 init_cgroup_css(css
, ss
, dummytop
);
4371 /* init_idr must be after init_cgroup_css because it sets css->id. */
4373 int ret
= cgroup_init_idr(ss
, css
);
4375 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4376 ss
->destroy(dummytop
);
4378 mutex_unlock(&cgroup_mutex
);
4384 * Now we need to entangle the css into the existing css_sets. unlike
4385 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4386 * will need a new pointer to it; done by iterating the css_set_table.
4387 * furthermore, modifying the existing css_sets will corrupt the hash
4388 * table state, so each changed css_set will need its hash recomputed.
4389 * this is all done under the css_set_lock.
4391 write_lock(&css_set_lock
);
4392 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
4394 struct hlist_node
*node
, *tmp
;
4395 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
4397 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
4398 /* skip entries that we already rehashed */
4399 if (cg
->subsys
[ss
->subsys_id
])
4401 /* remove existing entry */
4402 hlist_del(&cg
->hlist
);
4404 cg
->subsys
[ss
->subsys_id
] = css
;
4405 /* recompute hash and restore entry */
4406 new_bucket
= css_set_hash(cg
->subsys
);
4407 hlist_add_head(&cg
->hlist
, new_bucket
);
4410 write_unlock(&css_set_lock
);
4415 mutex_unlock(&cgroup_mutex
);
4418 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4421 * cgroup_unload_subsys: unload a modular subsystem
4422 * @ss: the subsystem to unload
4424 * This function should be called in a modular subsystem's exitcall. When this
4425 * function is invoked, the refcount on the subsystem's module will be 0, so
4426 * the subsystem will not be attached to any hierarchy.
4428 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4430 struct cg_cgroup_link
*link
;
4431 struct hlist_head
*hhead
;
4433 BUG_ON(ss
->module
== NULL
);
4436 * we shouldn't be called if the subsystem is in use, and the use of
4437 * try_module_get in parse_cgroupfs_options should ensure that it
4438 * doesn't start being used while we're killing it off.
4440 BUG_ON(ss
->root
!= &rootnode
);
4442 mutex_lock(&cgroup_mutex
);
4443 /* deassign the subsys_id */
4444 BUG_ON(ss
->subsys_id
< CGROUP_BUILTIN_SUBSYS_COUNT
);
4445 subsys
[ss
->subsys_id
] = NULL
;
4447 /* remove subsystem from rootnode's list of subsystems */
4448 list_del_init(&ss
->sibling
);
4451 * disentangle the css from all css_sets attached to the dummytop. as
4452 * in loading, we need to pay our respects to the hashtable gods.
4454 write_lock(&css_set_lock
);
4455 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4456 struct css_set
*cg
= link
->cg
;
4458 hlist_del(&cg
->hlist
);
4459 BUG_ON(!cg
->subsys
[ss
->subsys_id
]);
4460 cg
->subsys
[ss
->subsys_id
] = NULL
;
4461 hhead
= css_set_hash(cg
->subsys
);
4462 hlist_add_head(&cg
->hlist
, hhead
);
4464 write_unlock(&css_set_lock
);
4467 * remove subsystem's css from the dummytop and free it - need to free
4468 * before marking as null because ss->destroy needs the cgrp->subsys
4469 * pointer to find their state. note that this also takes care of
4470 * freeing the css_id.
4472 ss
->destroy(dummytop
);
4473 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4475 mutex_unlock(&cgroup_mutex
);
4477 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4480 * cgroup_init_early - cgroup initialization at system boot
4482 * Initialize cgroups at system boot, and initialize any
4483 * subsystems that request early init.
4485 int __init
cgroup_init_early(void)
4488 atomic_set(&init_css_set
.refcount
, 1);
4489 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4490 INIT_LIST_HEAD(&init_css_set
.tasks
);
4491 INIT_HLIST_NODE(&init_css_set
.hlist
);
4493 init_cgroup_root(&rootnode
);
4495 init_task
.cgroups
= &init_css_set
;
4497 init_css_set_link
.cg
= &init_css_set
;
4498 init_css_set_link
.cgrp
= dummytop
;
4499 list_add(&init_css_set_link
.cgrp_link_list
,
4500 &rootnode
.top_cgroup
.css_sets
);
4501 list_add(&init_css_set_link
.cg_link_list
,
4502 &init_css_set
.cg_links
);
4504 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
4505 INIT_HLIST_HEAD(&css_set_table
[i
]);
4507 /* at bootup time, we don't worry about modular subsystems */
4508 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4509 struct cgroup_subsys
*ss
= subsys
[i
];
4512 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4513 BUG_ON(!ss
->create
);
4514 BUG_ON(!ss
->destroy
);
4515 if (ss
->subsys_id
!= i
) {
4516 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4517 ss
->name
, ss
->subsys_id
);
4522 cgroup_init_subsys(ss
);
4528 * cgroup_init - cgroup initialization
4530 * Register cgroup filesystem and /proc file, and initialize
4531 * any subsystems that didn't request early init.
4533 int __init
cgroup_init(void)
4537 struct hlist_head
*hhead
;
4539 err
= bdi_init(&cgroup_backing_dev_info
);
4543 /* at bootup time, we don't worry about modular subsystems */
4544 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4545 struct cgroup_subsys
*ss
= subsys
[i
];
4546 if (!ss
->early_init
)
4547 cgroup_init_subsys(ss
);
4549 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4552 /* Add init_css_set to the hash table */
4553 hhead
= css_set_hash(init_css_set
.subsys
);
4554 hlist_add_head(&init_css_set
.hlist
, hhead
);
4555 BUG_ON(!init_root_id(&rootnode
));
4557 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4563 err
= register_filesystem(&cgroup_fs_type
);
4565 kobject_put(cgroup_kobj
);
4569 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4573 bdi_destroy(&cgroup_backing_dev_info
);
4579 * proc_cgroup_show()
4580 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4581 * - Used for /proc/<pid>/cgroup.
4582 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4583 * doesn't really matter if tsk->cgroup changes after we read it,
4584 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4585 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4586 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4587 * cgroup to top_cgroup.
4590 /* TODO: Use a proper seq_file iterator */
4591 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4594 struct task_struct
*tsk
;
4597 struct cgroupfs_root
*root
;
4600 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4606 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4612 mutex_lock(&cgroup_mutex
);
4614 for_each_active_root(root
) {
4615 struct cgroup_subsys
*ss
;
4616 struct cgroup
*cgrp
;
4619 seq_printf(m
, "%d:", root
->hierarchy_id
);
4620 for_each_subsys(root
, ss
)
4621 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4622 if (strlen(root
->name
))
4623 seq_printf(m
, "%sname=%s", count
? "," : "",
4626 cgrp
= task_cgroup_from_root(tsk
, root
);
4627 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4635 mutex_unlock(&cgroup_mutex
);
4636 put_task_struct(tsk
);
4643 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4645 struct pid
*pid
= PROC_I(inode
)->pid
;
4646 return single_open(file
, proc_cgroup_show
, pid
);
4649 const struct file_operations proc_cgroup_operations
= {
4650 .open
= cgroup_open
,
4652 .llseek
= seq_lseek
,
4653 .release
= single_release
,
4656 /* Display information about each subsystem and each hierarchy */
4657 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4661 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4663 * ideally we don't want subsystems moving around while we do this.
4664 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4665 * subsys/hierarchy state.
4667 mutex_lock(&cgroup_mutex
);
4668 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4669 struct cgroup_subsys
*ss
= subsys
[i
];
4672 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4673 ss
->name
, ss
->root
->hierarchy_id
,
4674 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4676 mutex_unlock(&cgroup_mutex
);
4680 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4682 return single_open(file
, proc_cgroupstats_show
, NULL
);
4685 static const struct file_operations proc_cgroupstats_operations
= {
4686 .open
= cgroupstats_open
,
4688 .llseek
= seq_lseek
,
4689 .release
= single_release
,
4693 * cgroup_fork - attach newly forked task to its parents cgroup.
4694 * @child: pointer to task_struct of forking parent process.
4696 * Description: A task inherits its parent's cgroup at fork().
4698 * A pointer to the shared css_set was automatically copied in
4699 * fork.c by dup_task_struct(). However, we ignore that copy, since
4700 * it was not made under the protection of RCU, cgroup_mutex or
4701 * threadgroup_change_begin(), so it might no longer be a valid
4702 * cgroup pointer. cgroup_attach_task() might have already changed
4703 * current->cgroups, allowing the previously referenced cgroup
4704 * group to be removed and freed.
4706 * Outside the pointer validity we also need to process the css_set
4707 * inheritance between threadgoup_change_begin() and
4708 * threadgoup_change_end(), this way there is no leak in any process
4709 * wide migration performed by cgroup_attach_proc() that could otherwise
4710 * miss a thread because it is too early or too late in the fork stage.
4712 * At the point that cgroup_fork() is called, 'current' is the parent
4713 * task, and the passed argument 'child' points to the child task.
4715 void cgroup_fork(struct task_struct
*child
)
4718 * We don't need to task_lock() current because current->cgroups
4719 * can't be changed concurrently here. The parent obviously hasn't
4720 * exited and called cgroup_exit(), and we are synchronized against
4721 * cgroup migration through threadgroup_change_begin().
4723 child
->cgroups
= current
->cgroups
;
4724 get_css_set(child
->cgroups
);
4725 INIT_LIST_HEAD(&child
->cg_list
);
4729 * cgroup_fork_callbacks - run fork callbacks
4730 * @child: the new task
4732 * Called on a new task very soon before adding it to the
4733 * tasklist. No need to take any locks since no-one can
4734 * be operating on this task.
4736 void cgroup_fork_callbacks(struct task_struct
*child
)
4738 if (need_forkexit_callback
) {
4741 * forkexit callbacks are only supported for builtin
4742 * subsystems, and the builtin section of the subsys array is
4743 * immutable, so we don't need to lock the subsys array here.
4745 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4746 struct cgroup_subsys
*ss
= subsys
[i
];
4754 * cgroup_post_fork - called on a new task after adding it to the task list
4755 * @child: the task in question
4757 * Adds the task to the list running through its css_set if necessary.
4758 * Has to be after the task is visible on the task list in case we race
4759 * with the first call to cgroup_iter_start() - to guarantee that the
4760 * new task ends up on its list.
4762 void cgroup_post_fork(struct task_struct
*child
)
4765 * use_task_css_set_links is set to 1 before we walk the tasklist
4766 * under the tasklist_lock and we read it here after we added the child
4767 * to the tasklist under the tasklist_lock as well. If the child wasn't
4768 * yet in the tasklist when we walked through it from
4769 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4770 * should be visible now due to the paired locking and barriers implied
4771 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4772 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4775 if (use_task_css_set_links
) {
4776 write_lock(&css_set_lock
);
4777 if (list_empty(&child
->cg_list
)) {
4779 * It's safe to use child->cgroups without task_lock()
4780 * here because we are protected through
4781 * threadgroup_change_begin() against concurrent
4782 * css_set change in cgroup_task_migrate(). Also
4783 * the task can't exit at that point until
4784 * wake_up_new_task() is called, so we are protected
4785 * against cgroup_exit() setting child->cgroup to
4788 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4790 write_unlock(&css_set_lock
);
4794 * cgroup_exit - detach cgroup from exiting task
4795 * @tsk: pointer to task_struct of exiting process
4796 * @run_callback: run exit callbacks?
4798 * Description: Detach cgroup from @tsk and release it.
4800 * Note that cgroups marked notify_on_release force every task in
4801 * them to take the global cgroup_mutex mutex when exiting.
4802 * This could impact scaling on very large systems. Be reluctant to
4803 * use notify_on_release cgroups where very high task exit scaling
4804 * is required on large systems.
4806 * the_top_cgroup_hack:
4808 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4810 * We call cgroup_exit() while the task is still competent to
4811 * handle notify_on_release(), then leave the task attached to the
4812 * root cgroup in each hierarchy for the remainder of its exit.
4814 * To do this properly, we would increment the reference count on
4815 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4816 * code we would add a second cgroup function call, to drop that
4817 * reference. This would just create an unnecessary hot spot on
4818 * the top_cgroup reference count, to no avail.
4820 * Normally, holding a reference to a cgroup without bumping its
4821 * count is unsafe. The cgroup could go away, or someone could
4822 * attach us to a different cgroup, decrementing the count on
4823 * the first cgroup that we never incremented. But in this case,
4824 * top_cgroup isn't going away, and either task has PF_EXITING set,
4825 * which wards off any cgroup_attach_task() attempts, or task is a failed
4826 * fork, never visible to cgroup_attach_task.
4828 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4834 * Unlink from the css_set task list if necessary.
4835 * Optimistically check cg_list before taking
4838 if (!list_empty(&tsk
->cg_list
)) {
4839 write_lock(&css_set_lock
);
4840 if (!list_empty(&tsk
->cg_list
))
4841 list_del_init(&tsk
->cg_list
);
4842 write_unlock(&css_set_lock
);
4845 /* Reassign the task to the init_css_set. */
4848 tsk
->cgroups
= &init_css_set
;
4850 if (run_callbacks
&& need_forkexit_callback
) {
4852 * modular subsystems can't use callbacks, so no need to lock
4855 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4856 struct cgroup_subsys
*ss
= subsys
[i
];
4858 struct cgroup
*old_cgrp
=
4859 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4860 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4861 ss
->exit(cgrp
, old_cgrp
, tsk
);
4868 put_css_set_taskexit(cg
);
4872 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4873 * @cgrp: the cgroup in question
4874 * @task: the task in question
4876 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4879 * If we are sending in dummytop, then presumably we are creating
4880 * the top cgroup in the subsystem.
4882 * Called only by the ns (nsproxy) cgroup.
4884 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
4887 struct cgroup
*target
;
4889 if (cgrp
== dummytop
)
4892 target
= task_cgroup_from_root(task
, cgrp
->root
);
4893 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
4894 cgrp
= cgrp
->parent
;
4895 ret
= (cgrp
== target
);
4899 static void check_for_release(struct cgroup
*cgrp
)
4901 /* All of these checks rely on RCU to keep the cgroup
4902 * structure alive */
4903 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
4904 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
4905 /* Control Group is currently removeable. If it's not
4906 * already queued for a userspace notification, queue
4908 int need_schedule_work
= 0;
4909 raw_spin_lock(&release_list_lock
);
4910 if (!cgroup_is_removed(cgrp
) &&
4911 list_empty(&cgrp
->release_list
)) {
4912 list_add(&cgrp
->release_list
, &release_list
);
4913 need_schedule_work
= 1;
4915 raw_spin_unlock(&release_list_lock
);
4916 if (need_schedule_work
)
4917 schedule_work(&release_agent_work
);
4921 /* Caller must verify that the css is not for root cgroup */
4922 bool __css_tryget(struct cgroup_subsys_state
*css
)
4925 int v
= css_refcnt(css
);
4927 if (atomic_cmpxchg(&css
->refcnt
, v
, v
+ 1) == v
)
4930 } while (!test_bit(CSS_REMOVED
, &css
->flags
));
4934 EXPORT_SYMBOL_GPL(__css_tryget
);
4936 /* Caller must verify that the css is not for root cgroup */
4937 void __css_put(struct cgroup_subsys_state
*css
)
4939 struct cgroup
*cgrp
= css
->cgroup
;
4942 switch (atomic_dec_return(&css
->refcnt
)) {
4944 if (notify_on_release(cgrp
)) {
4945 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4946 check_for_release(cgrp
);
4948 cgroup_wakeup_rmdir_waiter(cgrp
);
4951 if (!test_bit(CSS_CLEAR_CSS_REFS
, &css
->flags
))
4952 schedule_work(&css
->dput_work
);
4957 EXPORT_SYMBOL_GPL(__css_put
);
4960 * Notify userspace when a cgroup is released, by running the
4961 * configured release agent with the name of the cgroup (path
4962 * relative to the root of cgroup file system) as the argument.
4964 * Most likely, this user command will try to rmdir this cgroup.
4966 * This races with the possibility that some other task will be
4967 * attached to this cgroup before it is removed, or that some other
4968 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4969 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4970 * unused, and this cgroup will be reprieved from its death sentence,
4971 * to continue to serve a useful existence. Next time it's released,
4972 * we will get notified again, if it still has 'notify_on_release' set.
4974 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4975 * means only wait until the task is successfully execve()'d. The
4976 * separate release agent task is forked by call_usermodehelper(),
4977 * then control in this thread returns here, without waiting for the
4978 * release agent task. We don't bother to wait because the caller of
4979 * this routine has no use for the exit status of the release agent
4980 * task, so no sense holding our caller up for that.
4982 static void cgroup_release_agent(struct work_struct
*work
)
4984 BUG_ON(work
!= &release_agent_work
);
4985 mutex_lock(&cgroup_mutex
);
4986 raw_spin_lock(&release_list_lock
);
4987 while (!list_empty(&release_list
)) {
4988 char *argv
[3], *envp
[3];
4990 char *pathbuf
= NULL
, *agentbuf
= NULL
;
4991 struct cgroup
*cgrp
= list_entry(release_list
.next
,
4994 list_del_init(&cgrp
->release_list
);
4995 raw_spin_unlock(&release_list_lock
);
4996 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4999 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5001 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5006 argv
[i
++] = agentbuf
;
5007 argv
[i
++] = pathbuf
;
5011 /* minimal command environment */
5012 envp
[i
++] = "HOME=/";
5013 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5016 /* Drop the lock while we invoke the usermode helper,
5017 * since the exec could involve hitting disk and hence
5018 * be a slow process */
5019 mutex_unlock(&cgroup_mutex
);
5020 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5021 mutex_lock(&cgroup_mutex
);
5025 raw_spin_lock(&release_list_lock
);
5027 raw_spin_unlock(&release_list_lock
);
5028 mutex_unlock(&cgroup_mutex
);
5031 static int __init
cgroup_disable(char *str
)
5036 while ((token
= strsep(&str
, ",")) != NULL
) {
5040 * cgroup_disable, being at boot time, can't know about module
5041 * subsystems, so we don't worry about them.
5043 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
5044 struct cgroup_subsys
*ss
= subsys
[i
];
5046 if (!strcmp(token
, ss
->name
)) {
5048 printk(KERN_INFO
"Disabling %s control group"
5049 " subsystem\n", ss
->name
);
5056 __setup("cgroup_disable=", cgroup_disable
);
5059 * Functons for CSS ID.
5063 *To get ID other than 0, this should be called when !cgroup_is_removed().
5065 unsigned short css_id(struct cgroup_subsys_state
*css
)
5067 struct css_id
*cssid
;
5070 * This css_id() can return correct value when somone has refcnt
5071 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5072 * it's unchanged until freed.
5074 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5080 EXPORT_SYMBOL_GPL(css_id
);
5082 unsigned short css_depth(struct cgroup_subsys_state
*css
)
5084 struct css_id
*cssid
;
5086 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5089 return cssid
->depth
;
5092 EXPORT_SYMBOL_GPL(css_depth
);
5095 * css_is_ancestor - test "root" css is an ancestor of "child"
5096 * @child: the css to be tested.
5097 * @root: the css supporsed to be an ancestor of the child.
5099 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5100 * this function reads css->id, the caller must hold rcu_read_lock().
5101 * But, considering usual usage, the csses should be valid objects after test.
5102 * Assuming that the caller will do some action to the child if this returns
5103 * returns true, the caller must take "child";s reference count.
5104 * If "child" is valid object and this returns true, "root" is valid, too.
5107 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5108 const struct cgroup_subsys_state
*root
)
5110 struct css_id
*child_id
;
5111 struct css_id
*root_id
;
5113 child_id
= rcu_dereference(child
->id
);
5116 root_id
= rcu_dereference(root
->id
);
5119 if (child_id
->depth
< root_id
->depth
)
5121 if (child_id
->stack
[root_id
->depth
] != root_id
->id
)
5126 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5128 struct css_id
*id
= css
->id
;
5129 /* When this is called before css_id initialization, id can be NULL */
5133 BUG_ON(!ss
->use_id
);
5135 rcu_assign_pointer(id
->css
, NULL
);
5136 rcu_assign_pointer(css
->id
, NULL
);
5137 spin_lock(&ss
->id_lock
);
5138 idr_remove(&ss
->idr
, id
->id
);
5139 spin_unlock(&ss
->id_lock
);
5140 kfree_rcu(id
, rcu_head
);
5142 EXPORT_SYMBOL_GPL(free_css_id
);
5145 * This is called by init or create(). Then, calls to this function are
5146 * always serialized (By cgroup_mutex() at create()).
5149 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5151 struct css_id
*newid
;
5152 int myid
, error
, size
;
5154 BUG_ON(!ss
->use_id
);
5156 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5157 newid
= kzalloc(size
, GFP_KERNEL
);
5159 return ERR_PTR(-ENOMEM
);
5161 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
5165 spin_lock(&ss
->id_lock
);
5166 /* Don't use 0. allocates an ID of 1-65535 */
5167 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
5168 spin_unlock(&ss
->id_lock
);
5170 /* Returns error when there are no free spaces for new ID.*/
5175 if (myid
> CSS_ID_MAX
)
5179 newid
->depth
= depth
;
5183 spin_lock(&ss
->id_lock
);
5184 idr_remove(&ss
->idr
, myid
);
5185 spin_unlock(&ss
->id_lock
);
5188 return ERR_PTR(error
);
5192 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5193 struct cgroup_subsys_state
*rootcss
)
5195 struct css_id
*newid
;
5197 spin_lock_init(&ss
->id_lock
);
5200 newid
= get_new_cssid(ss
, 0);
5202 return PTR_ERR(newid
);
5204 newid
->stack
[0] = newid
->id
;
5205 newid
->css
= rootcss
;
5206 rootcss
->id
= newid
;
5210 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5211 struct cgroup
*child
)
5213 int subsys_id
, i
, depth
= 0;
5214 struct cgroup_subsys_state
*parent_css
, *child_css
;
5215 struct css_id
*child_id
, *parent_id
;
5217 subsys_id
= ss
->subsys_id
;
5218 parent_css
= parent
->subsys
[subsys_id
];
5219 child_css
= child
->subsys
[subsys_id
];
5220 parent_id
= parent_css
->id
;
5221 depth
= parent_id
->depth
+ 1;
5223 child_id
= get_new_cssid(ss
, depth
);
5224 if (IS_ERR(child_id
))
5225 return PTR_ERR(child_id
);
5227 for (i
= 0; i
< depth
; i
++)
5228 child_id
->stack
[i
] = parent_id
->stack
[i
];
5229 child_id
->stack
[depth
] = child_id
->id
;
5231 * child_id->css pointer will be set after this cgroup is available
5232 * see cgroup_populate_dir()
5234 rcu_assign_pointer(child_css
->id
, child_id
);
5240 * css_lookup - lookup css by id
5241 * @ss: cgroup subsys to be looked into.
5244 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5245 * NULL if not. Should be called under rcu_read_lock()
5247 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5249 struct css_id
*cssid
= NULL
;
5251 BUG_ON(!ss
->use_id
);
5252 cssid
= idr_find(&ss
->idr
, id
);
5254 if (unlikely(!cssid
))
5257 return rcu_dereference(cssid
->css
);
5259 EXPORT_SYMBOL_GPL(css_lookup
);
5262 * css_get_next - lookup next cgroup under specified hierarchy.
5263 * @ss: pointer to subsystem
5264 * @id: current position of iteration.
5265 * @root: pointer to css. search tree under this.
5266 * @foundid: position of found object.
5268 * Search next css under the specified hierarchy of rootid. Calling under
5269 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5271 struct cgroup_subsys_state
*
5272 css_get_next(struct cgroup_subsys
*ss
, int id
,
5273 struct cgroup_subsys_state
*root
, int *foundid
)
5275 struct cgroup_subsys_state
*ret
= NULL
;
5278 int rootid
= css_id(root
);
5279 int depth
= css_depth(root
);
5284 BUG_ON(!ss
->use_id
);
5285 WARN_ON_ONCE(!rcu_read_lock_held());
5287 /* fill start point for scan */
5291 * scan next entry from bitmap(tree), tmpid is updated after
5294 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
5297 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
5298 ret
= rcu_dereference(tmp
->css
);
5304 /* continue to scan from next id */
5311 * get corresponding css from file open on cgroupfs directory
5313 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5315 struct cgroup
*cgrp
;
5316 struct inode
*inode
;
5317 struct cgroup_subsys_state
*css
;
5319 inode
= f
->f_dentry
->d_inode
;
5320 /* check in cgroup filesystem dir */
5321 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5322 return ERR_PTR(-EBADF
);
5324 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5325 return ERR_PTR(-EINVAL
);
5328 cgrp
= __d_cgrp(f
->f_dentry
);
5329 css
= cgrp
->subsys
[id
];
5330 return css
? css
: ERR_PTR(-ENOENT
);
5333 #ifdef CONFIG_CGROUP_DEBUG
5334 static struct cgroup_subsys_state
*debug_create(struct cgroup
*cont
)
5336 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5339 return ERR_PTR(-ENOMEM
);
5344 static void debug_destroy(struct cgroup
*cont
)
5346 kfree(cont
->subsys
[debug_subsys_id
]);
5349 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5351 return atomic_read(&cont
->count
);
5354 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5356 return cgroup_task_count(cont
);
5359 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5361 return (u64
)(unsigned long)current
->cgroups
;
5364 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5370 count
= atomic_read(¤t
->cgroups
->refcount
);
5375 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5377 struct seq_file
*seq
)
5379 struct cg_cgroup_link
*link
;
5382 read_lock(&css_set_lock
);
5384 cg
= rcu_dereference(current
->cgroups
);
5385 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5386 struct cgroup
*c
= link
->cgrp
;
5390 name
= c
->dentry
->d_name
.name
;
5393 seq_printf(seq
, "Root %d group %s\n",
5394 c
->root
->hierarchy_id
, name
);
5397 read_unlock(&css_set_lock
);
5401 #define MAX_TASKS_SHOWN_PER_CSS 25
5402 static int cgroup_css_links_read(struct cgroup
*cont
,
5404 struct seq_file
*seq
)
5406 struct cg_cgroup_link
*link
;
5408 read_lock(&css_set_lock
);
5409 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5410 struct css_set
*cg
= link
->cg
;
5411 struct task_struct
*task
;
5413 seq_printf(seq
, "css_set %p\n", cg
);
5414 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5415 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5416 seq_puts(seq
, " ...\n");
5419 seq_printf(seq
, " task %d\n",
5420 task_pid_vnr(task
));
5424 read_unlock(&css_set_lock
);
5428 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5430 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5433 static struct cftype debug_files
[] = {
5435 .name
= "cgroup_refcount",
5436 .read_u64
= cgroup_refcount_read
,
5439 .name
= "taskcount",
5440 .read_u64
= debug_taskcount_read
,
5444 .name
= "current_css_set",
5445 .read_u64
= current_css_set_read
,
5449 .name
= "current_css_set_refcount",
5450 .read_u64
= current_css_set_refcount_read
,
5454 .name
= "current_css_set_cg_links",
5455 .read_seq_string
= current_css_set_cg_links_read
,
5459 .name
= "cgroup_css_links",
5460 .read_seq_string
= cgroup_css_links_read
,
5464 .name
= "releasable",
5465 .read_u64
= releasable_read
,
5471 struct cgroup_subsys debug_subsys
= {
5473 .create
= debug_create
,
5474 .destroy
= debug_destroy
,
5475 .subsys_id
= debug_subsys_id
,
5476 .base_cftypes
= debug_files
,
5478 #endif /* CONFIG_CGROUP_DEBUG */