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 static int css_unbias_refcnt(int refcnt
)
260 return refcnt
>= 0 ? refcnt
: refcnt
- CSS_DEACT_BIAS
;
263 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
264 static int css_refcnt(struct cgroup_subsys_state
*css
)
266 int v
= atomic_read(&css
->refcnt
);
268 return css_unbias_refcnt(v
);
271 /* convenient tests for these bits */
272 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
274 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
277 /* bits in struct cgroupfs_root flags field */
279 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
282 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
285 (1 << CGRP_RELEASABLE
) |
286 (1 << CGRP_NOTIFY_ON_RELEASE
);
287 return (cgrp
->flags
& bits
) == bits
;
290 static int notify_on_release(const struct cgroup
*cgrp
)
292 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
295 static int clone_children(const struct cgroup
*cgrp
)
297 return test_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
301 * for_each_subsys() allows you to iterate on each subsystem attached to
302 * an active hierarchy
304 #define for_each_subsys(_root, _ss) \
305 list_for_each_entry(_ss, &_root->subsys_list, sibling)
307 /* for_each_active_root() allows you to iterate across the active hierarchies */
308 #define for_each_active_root(_root) \
309 list_for_each_entry(_root, &roots, root_list)
311 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
313 return dentry
->d_fsdata
;
316 static inline struct cfent
*__d_cfe(struct dentry
*dentry
)
318 return dentry
->d_fsdata
;
321 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
323 return __d_cfe(dentry
)->type
;
326 /* the list of cgroups eligible for automatic release. Protected by
327 * release_list_lock */
328 static LIST_HEAD(release_list
);
329 static DEFINE_RAW_SPINLOCK(release_list_lock
);
330 static void cgroup_release_agent(struct work_struct
*work
);
331 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
332 static void check_for_release(struct cgroup
*cgrp
);
334 /* Link structure for associating css_set objects with cgroups */
335 struct cg_cgroup_link
{
337 * List running through cg_cgroup_links associated with a
338 * cgroup, anchored on cgroup->css_sets
340 struct list_head cgrp_link_list
;
343 * List running through cg_cgroup_links pointing at a
344 * single css_set object, anchored on css_set->cg_links
346 struct list_head cg_link_list
;
350 /* The default css_set - used by init and its children prior to any
351 * hierarchies being mounted. It contains a pointer to the root state
352 * for each subsystem. Also used to anchor the list of css_sets. Not
353 * reference-counted, to improve performance when child cgroups
354 * haven't been created.
357 static struct css_set init_css_set
;
358 static struct cg_cgroup_link init_css_set_link
;
360 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
361 struct cgroup_subsys_state
*css
);
363 /* css_set_lock protects the list of css_set objects, and the
364 * chain of tasks off each css_set. Nests outside task->alloc_lock
365 * due to cgroup_iter_start() */
366 static DEFINE_RWLOCK(css_set_lock
);
367 static int css_set_count
;
370 * hash table for cgroup groups. This improves the performance to find
371 * an existing css_set. This hash doesn't (currently) take into
372 * account cgroups in empty hierarchies.
374 #define CSS_SET_HASH_BITS 7
375 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
376 static struct hlist_head css_set_table
[CSS_SET_TABLE_SIZE
];
378 static struct hlist_head
*css_set_hash(struct cgroup_subsys_state
*css
[])
382 unsigned long tmp
= 0UL;
384 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
385 tmp
+= (unsigned long)css
[i
];
386 tmp
= (tmp
>> 16) ^ tmp
;
388 index
= hash_long(tmp
, CSS_SET_HASH_BITS
);
390 return &css_set_table
[index
];
393 /* We don't maintain the lists running through each css_set to its
394 * task until after the first call to cgroup_iter_start(). This
395 * reduces the fork()/exit() overhead for people who have cgroups
396 * compiled into their kernel but not actually in use */
397 static int use_task_css_set_links __read_mostly
;
399 static void __put_css_set(struct css_set
*cg
, int taskexit
)
401 struct cg_cgroup_link
*link
;
402 struct cg_cgroup_link
*saved_link
;
404 * Ensure that the refcount doesn't hit zero while any readers
405 * can see it. Similar to atomic_dec_and_lock(), but for an
408 if (atomic_add_unless(&cg
->refcount
, -1, 1))
410 write_lock(&css_set_lock
);
411 if (!atomic_dec_and_test(&cg
->refcount
)) {
412 write_unlock(&css_set_lock
);
416 /* This css_set is dead. unlink it and release cgroup refcounts */
417 hlist_del(&cg
->hlist
);
420 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
422 struct cgroup
*cgrp
= link
->cgrp
;
423 list_del(&link
->cg_link_list
);
424 list_del(&link
->cgrp_link_list
);
425 if (atomic_dec_and_test(&cgrp
->count
) &&
426 notify_on_release(cgrp
)) {
428 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
429 check_for_release(cgrp
);
435 write_unlock(&css_set_lock
);
436 kfree_rcu(cg
, rcu_head
);
440 * refcounted get/put for css_set objects
442 static inline void get_css_set(struct css_set
*cg
)
444 atomic_inc(&cg
->refcount
);
447 static inline void put_css_set(struct css_set
*cg
)
449 __put_css_set(cg
, 0);
452 static inline void put_css_set_taskexit(struct css_set
*cg
)
454 __put_css_set(cg
, 1);
458 * compare_css_sets - helper function for find_existing_css_set().
459 * @cg: candidate css_set being tested
460 * @old_cg: existing css_set for a task
461 * @new_cgrp: cgroup that's being entered by the task
462 * @template: desired set of css pointers in css_set (pre-calculated)
464 * Returns true if "cg" matches "old_cg" except for the hierarchy
465 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
467 static bool compare_css_sets(struct css_set
*cg
,
468 struct css_set
*old_cg
,
469 struct cgroup
*new_cgrp
,
470 struct cgroup_subsys_state
*template[])
472 struct list_head
*l1
, *l2
;
474 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
475 /* Not all subsystems matched */
480 * Compare cgroup pointers in order to distinguish between
481 * different cgroups in heirarchies with no subsystems. We
482 * could get by with just this check alone (and skip the
483 * memcmp above) but on most setups the memcmp check will
484 * avoid the need for this more expensive check on almost all
489 l2
= &old_cg
->cg_links
;
491 struct cg_cgroup_link
*cgl1
, *cgl2
;
492 struct cgroup
*cg1
, *cg2
;
496 /* See if we reached the end - both lists are equal length. */
497 if (l1
== &cg
->cg_links
) {
498 BUG_ON(l2
!= &old_cg
->cg_links
);
501 BUG_ON(l2
== &old_cg
->cg_links
);
503 /* Locate the cgroups associated with these links. */
504 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
505 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
508 /* Hierarchies should be linked in the same order. */
509 BUG_ON(cg1
->root
!= cg2
->root
);
512 * If this hierarchy is the hierarchy of the cgroup
513 * that's changing, then we need to check that this
514 * css_set points to the new cgroup; if it's any other
515 * hierarchy, then this css_set should point to the
516 * same cgroup as the old css_set.
518 if (cg1
->root
== new_cgrp
->root
) {
530 * find_existing_css_set() is a helper for
531 * find_css_set(), and checks to see whether an existing
532 * css_set is suitable.
534 * oldcg: the cgroup group that we're using before the cgroup
537 * cgrp: the cgroup that we're moving into
539 * template: location in which to build the desired set of subsystem
540 * state objects for the new cgroup group
542 static struct css_set
*find_existing_css_set(
543 struct css_set
*oldcg
,
545 struct cgroup_subsys_state
*template[])
548 struct cgroupfs_root
*root
= cgrp
->root
;
549 struct hlist_head
*hhead
;
550 struct hlist_node
*node
;
554 * Build the set of subsystem state objects that we want to see in the
555 * new css_set. while subsystems can change globally, the entries here
556 * won't change, so no need for locking.
558 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
559 if (root
->subsys_bits
& (1UL << i
)) {
560 /* Subsystem is in this hierarchy. So we want
561 * the subsystem state from the new
563 template[i
] = cgrp
->subsys
[i
];
565 /* Subsystem is not in this hierarchy, so we
566 * don't want to change the subsystem state */
567 template[i
] = oldcg
->subsys
[i
];
571 hhead
= css_set_hash(template);
572 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
573 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
576 /* This css_set matches what we need */
580 /* No existing cgroup group matched */
584 static void free_cg_links(struct list_head
*tmp
)
586 struct cg_cgroup_link
*link
;
587 struct cg_cgroup_link
*saved_link
;
589 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
590 list_del(&link
->cgrp_link_list
);
596 * allocate_cg_links() allocates "count" cg_cgroup_link structures
597 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
598 * success or a negative error
600 static int allocate_cg_links(int count
, struct list_head
*tmp
)
602 struct cg_cgroup_link
*link
;
605 for (i
= 0; i
< count
; i
++) {
606 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
611 list_add(&link
->cgrp_link_list
, tmp
);
617 * link_css_set - a helper function to link a css_set to a cgroup
618 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
619 * @cg: the css_set to be linked
620 * @cgrp: the destination cgroup
622 static void link_css_set(struct list_head
*tmp_cg_links
,
623 struct css_set
*cg
, struct cgroup
*cgrp
)
625 struct cg_cgroup_link
*link
;
627 BUG_ON(list_empty(tmp_cg_links
));
628 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
632 atomic_inc(&cgrp
->count
);
633 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
635 * Always add links to the tail of the list so that the list
636 * is sorted by order of hierarchy creation
638 list_add_tail(&link
->cg_link_list
, &cg
->cg_links
);
642 * find_css_set() takes an existing cgroup group and a
643 * cgroup object, and returns a css_set object that's
644 * equivalent to the old group, but with the given cgroup
645 * substituted into the appropriate hierarchy. Must be called with
648 static struct css_set
*find_css_set(
649 struct css_set
*oldcg
, struct cgroup
*cgrp
)
652 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
654 struct list_head tmp_cg_links
;
656 struct hlist_head
*hhead
;
657 struct cg_cgroup_link
*link
;
659 /* First see if we already have a cgroup group that matches
661 read_lock(&css_set_lock
);
662 res
= find_existing_css_set(oldcg
, cgrp
, template);
665 read_unlock(&css_set_lock
);
670 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
674 /* Allocate all the cg_cgroup_link objects that we'll need */
675 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
680 atomic_set(&res
->refcount
, 1);
681 INIT_LIST_HEAD(&res
->cg_links
);
682 INIT_LIST_HEAD(&res
->tasks
);
683 INIT_HLIST_NODE(&res
->hlist
);
685 /* Copy the set of subsystem state objects generated in
686 * find_existing_css_set() */
687 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
689 write_lock(&css_set_lock
);
690 /* Add reference counts and links from the new css_set. */
691 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
692 struct cgroup
*c
= link
->cgrp
;
693 if (c
->root
== cgrp
->root
)
695 link_css_set(&tmp_cg_links
, res
, c
);
698 BUG_ON(!list_empty(&tmp_cg_links
));
702 /* Add this cgroup group to the hash table */
703 hhead
= css_set_hash(res
->subsys
);
704 hlist_add_head(&res
->hlist
, hhead
);
706 write_unlock(&css_set_lock
);
712 * Return the cgroup for "task" from the given hierarchy. Must be
713 * called with cgroup_mutex held.
715 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
716 struct cgroupfs_root
*root
)
719 struct cgroup
*res
= NULL
;
721 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
722 read_lock(&css_set_lock
);
724 * No need to lock the task - since we hold cgroup_mutex the
725 * task can't change groups, so the only thing that can happen
726 * is that it exits and its css is set back to init_css_set.
729 if (css
== &init_css_set
) {
730 res
= &root
->top_cgroup
;
732 struct cg_cgroup_link
*link
;
733 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
734 struct cgroup
*c
= link
->cgrp
;
735 if (c
->root
== root
) {
741 read_unlock(&css_set_lock
);
747 * There is one global cgroup mutex. We also require taking
748 * task_lock() when dereferencing a task's cgroup subsys pointers.
749 * See "The task_lock() exception", at the end of this comment.
751 * A task must hold cgroup_mutex to modify cgroups.
753 * Any task can increment and decrement the count field without lock.
754 * So in general, code holding cgroup_mutex can't rely on the count
755 * field not changing. However, if the count goes to zero, then only
756 * cgroup_attach_task() can increment it again. Because a count of zero
757 * means that no tasks are currently attached, therefore there is no
758 * way a task attached to that cgroup can fork (the other way to
759 * increment the count). So code holding cgroup_mutex can safely
760 * assume that if the count is zero, it will stay zero. Similarly, if
761 * a task holds cgroup_mutex on a cgroup with zero count, it
762 * knows that the cgroup won't be removed, as cgroup_rmdir()
765 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
766 * (usually) take cgroup_mutex. These are the two most performance
767 * critical pieces of code here. The exception occurs on cgroup_exit(),
768 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
769 * is taken, and if the cgroup count is zero, a usermode call made
770 * to the release agent with the name of the cgroup (path relative to
771 * the root of cgroup file system) as the argument.
773 * A cgroup can only be deleted if both its 'count' of using tasks
774 * is zero, and its list of 'children' cgroups is empty. Since all
775 * tasks in the system use _some_ cgroup, and since there is always at
776 * least one task in the system (init, pid == 1), therefore, top_cgroup
777 * always has either children cgroups and/or using tasks. So we don't
778 * need a special hack to ensure that top_cgroup cannot be deleted.
780 * The task_lock() exception
782 * The need for this exception arises from the action of
783 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
784 * another. It does so using cgroup_mutex, however there are
785 * several performance critical places that need to reference
786 * task->cgroup without the expense of grabbing a system global
787 * mutex. Therefore except as noted below, when dereferencing or, as
788 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
789 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
790 * the task_struct routinely used for such matters.
792 * P.S. One more locking exception. RCU is used to guard the
793 * update of a tasks cgroup pointer by cgroup_attach_task()
797 * cgroup_lock - lock out any changes to cgroup structures
800 void cgroup_lock(void)
802 mutex_lock(&cgroup_mutex
);
804 EXPORT_SYMBOL_GPL(cgroup_lock
);
807 * cgroup_unlock - release lock on cgroup changes
809 * Undo the lock taken in a previous cgroup_lock() call.
811 void cgroup_unlock(void)
813 mutex_unlock(&cgroup_mutex
);
815 EXPORT_SYMBOL_GPL(cgroup_unlock
);
818 * A couple of forward declarations required, due to cyclic reference loop:
819 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
820 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
824 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
825 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
826 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
827 static int cgroup_populate_dir(struct cgroup
*cgrp
);
828 static const struct inode_operations cgroup_dir_inode_operations
;
829 static const struct file_operations proc_cgroupstats_operations
;
831 static struct backing_dev_info cgroup_backing_dev_info
= {
833 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
836 static int alloc_css_id(struct cgroup_subsys
*ss
,
837 struct cgroup
*parent
, struct cgroup
*child
);
839 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
841 struct inode
*inode
= new_inode(sb
);
844 inode
->i_ino
= get_next_ino();
845 inode
->i_mode
= mode
;
846 inode
->i_uid
= current_fsuid();
847 inode
->i_gid
= current_fsgid();
848 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
849 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
855 * Call subsys's pre_destroy handler.
856 * This is called before css refcnt check.
858 static int cgroup_call_pre_destroy(struct cgroup
*cgrp
)
860 struct cgroup_subsys
*ss
;
863 for_each_subsys(cgrp
->root
, ss
) {
864 if (!ss
->pre_destroy
)
867 ret
= ss
->pre_destroy(cgrp
);
869 /* ->pre_destroy() failure is being deprecated */
870 WARN_ON_ONCE(!ss
->__DEPRECATED_clear_css_refs
);
878 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
880 /* is dentry a directory ? if so, kfree() associated cgroup */
881 if (S_ISDIR(inode
->i_mode
)) {
882 struct cgroup
*cgrp
= dentry
->d_fsdata
;
883 struct cgroup_subsys
*ss
;
884 BUG_ON(!(cgroup_is_removed(cgrp
)));
885 /* It's possible for external users to be holding css
886 * reference counts on a cgroup; css_put() needs to
887 * be able to access the cgroup after decrementing
888 * the reference count in order to know if it needs to
889 * queue the cgroup to be handled by the release
893 mutex_lock(&cgroup_mutex
);
895 * Release the subsystem state objects.
897 for_each_subsys(cgrp
->root
, ss
)
900 cgrp
->root
->number_of_cgroups
--;
901 mutex_unlock(&cgroup_mutex
);
904 * Drop the active superblock reference that we took when we
907 deactivate_super(cgrp
->root
->sb
);
910 * if we're getting rid of the cgroup, refcount should ensure
911 * that there are no pidlists left.
913 BUG_ON(!list_empty(&cgrp
->pidlists
));
915 kfree_rcu(cgrp
, rcu_head
);
917 struct cfent
*cfe
= __d_cfe(dentry
);
918 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
920 WARN_ONCE(!list_empty(&cfe
->node
) &&
921 cgrp
!= &cgrp
->root
->top_cgroup
,
922 "cfe still linked for %s\n", cfe
->type
->name
);
928 static int cgroup_delete(const struct dentry
*d
)
933 static void remove_dir(struct dentry
*d
)
935 struct dentry
*parent
= dget(d
->d_parent
);
938 simple_rmdir(parent
->d_inode
, d
);
942 static int cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
946 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
947 lockdep_assert_held(&cgroup_mutex
);
949 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
950 struct dentry
*d
= cfe
->dentry
;
952 if (cft
&& cfe
->type
!= cft
)
957 simple_unlink(d
->d_inode
, d
);
958 list_del_init(&cfe
->node
);
966 static void cgroup_clear_directory(struct dentry
*dir
)
968 struct cgroup
*cgrp
= __d_cgrp(dir
);
970 while (!list_empty(&cgrp
->files
))
971 cgroup_rm_file(cgrp
, NULL
);
975 * NOTE : the dentry must have been dget()'ed
977 static void cgroup_d_remove_dir(struct dentry
*dentry
)
979 struct dentry
*parent
;
981 cgroup_clear_directory(dentry
);
983 parent
= dentry
->d_parent
;
984 spin_lock(&parent
->d_lock
);
985 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
986 list_del_init(&dentry
->d_u
.d_child
);
987 spin_unlock(&dentry
->d_lock
);
988 spin_unlock(&parent
->d_lock
);
993 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
994 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
995 * reference to css->refcnt. In general, this refcnt is expected to goes down
998 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
1000 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq
);
1002 static void cgroup_wakeup_rmdir_waiter(struct cgroup
*cgrp
)
1004 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
)))
1005 wake_up_all(&cgroup_rmdir_waitq
);
1008 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
)
1013 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
)
1015 cgroup_wakeup_rmdir_waiter(css
->cgroup
);
1020 * Call with cgroup_mutex held. Drops reference counts on modules, including
1021 * any duplicate ones that parse_cgroupfs_options took. If this function
1022 * returns an error, no reference counts are touched.
1024 static int rebind_subsystems(struct cgroupfs_root
*root
,
1025 unsigned long final_bits
)
1027 unsigned long added_bits
, removed_bits
;
1028 struct cgroup
*cgrp
= &root
->top_cgroup
;
1031 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1032 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1034 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
1035 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
1036 /* Check that any added subsystems are currently free */
1037 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1038 unsigned long bit
= 1UL << i
;
1039 struct cgroup_subsys
*ss
= subsys
[i
];
1040 if (!(bit
& added_bits
))
1043 * Nobody should tell us to do a subsys that doesn't exist:
1044 * parse_cgroupfs_options should catch that case and refcounts
1045 * ensure that subsystems won't disappear once selected.
1048 if (ss
->root
!= &rootnode
) {
1049 /* Subsystem isn't free */
1054 /* Currently we don't handle adding/removing subsystems when
1055 * any child cgroups exist. This is theoretically supportable
1056 * but involves complex error handling, so it's being left until
1058 if (root
->number_of_cgroups
> 1)
1061 /* Process each subsystem */
1062 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1063 struct cgroup_subsys
*ss
= subsys
[i
];
1064 unsigned long bit
= 1UL << i
;
1065 if (bit
& added_bits
) {
1066 /* We're binding this subsystem to this hierarchy */
1068 BUG_ON(cgrp
->subsys
[i
]);
1069 BUG_ON(!dummytop
->subsys
[i
]);
1070 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
1071 mutex_lock(&ss
->hierarchy_mutex
);
1072 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
1073 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1074 list_move(&ss
->sibling
, &root
->subsys_list
);
1078 mutex_unlock(&ss
->hierarchy_mutex
);
1079 /* refcount was already taken, and we're keeping it */
1080 } else if (bit
& removed_bits
) {
1081 /* We're removing this subsystem */
1083 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1084 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1085 mutex_lock(&ss
->hierarchy_mutex
);
1088 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1089 cgrp
->subsys
[i
] = NULL
;
1090 subsys
[i
]->root
= &rootnode
;
1091 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1092 mutex_unlock(&ss
->hierarchy_mutex
);
1093 /* subsystem is now free - drop reference on module */
1094 module_put(ss
->module
);
1095 } else if (bit
& final_bits
) {
1096 /* Subsystem state should already exist */
1098 BUG_ON(!cgrp
->subsys
[i
]);
1100 * a refcount was taken, but we already had one, so
1101 * drop the extra reference.
1103 module_put(ss
->module
);
1104 #ifdef CONFIG_MODULE_UNLOAD
1105 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1108 /* Subsystem state shouldn't exist */
1109 BUG_ON(cgrp
->subsys
[i
]);
1112 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
1118 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1120 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1121 struct cgroup_subsys
*ss
;
1123 mutex_lock(&cgroup_root_mutex
);
1124 for_each_subsys(root
, ss
)
1125 seq_printf(seq
, ",%s", ss
->name
);
1126 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1127 seq_puts(seq
, ",noprefix");
1128 if (strlen(root
->release_agent_path
))
1129 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1130 if (clone_children(&root
->top_cgroup
))
1131 seq_puts(seq
, ",clone_children");
1132 if (strlen(root
->name
))
1133 seq_printf(seq
, ",name=%s", root
->name
);
1134 mutex_unlock(&cgroup_root_mutex
);
1138 struct cgroup_sb_opts
{
1139 unsigned long subsys_bits
;
1140 unsigned long flags
;
1141 char *release_agent
;
1142 bool clone_children
;
1144 /* User explicitly requested empty subsystem */
1147 struct cgroupfs_root
*new_root
;
1152 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1153 * with cgroup_mutex held to protect the subsys[] array. This function takes
1154 * refcounts on subsystems to be used, unless it returns error, in which case
1155 * no refcounts are taken.
1157 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1159 char *token
, *o
= data
;
1160 bool all_ss
= false, one_ss
= false;
1161 unsigned long mask
= (unsigned long)-1;
1163 bool module_pin_failed
= false;
1165 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1167 #ifdef CONFIG_CPUSETS
1168 mask
= ~(1UL << cpuset_subsys_id
);
1171 memset(opts
, 0, sizeof(*opts
));
1173 while ((token
= strsep(&o
, ",")) != NULL
) {
1176 if (!strcmp(token
, "none")) {
1177 /* Explicitly have no subsystems */
1181 if (!strcmp(token
, "all")) {
1182 /* Mutually exclusive option 'all' + subsystem name */
1188 if (!strcmp(token
, "noprefix")) {
1189 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1192 if (!strcmp(token
, "clone_children")) {
1193 opts
->clone_children
= true;
1196 if (!strncmp(token
, "release_agent=", 14)) {
1197 /* Specifying two release agents is forbidden */
1198 if (opts
->release_agent
)
1200 opts
->release_agent
=
1201 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1202 if (!opts
->release_agent
)
1206 if (!strncmp(token
, "name=", 5)) {
1207 const char *name
= token
+ 5;
1208 /* Can't specify an empty name */
1211 /* Must match [\w.-]+ */
1212 for (i
= 0; i
< strlen(name
); i
++) {
1216 if ((c
== '.') || (c
== '-') || (c
== '_'))
1220 /* Specifying two names is forbidden */
1223 opts
->name
= kstrndup(name
,
1224 MAX_CGROUP_ROOT_NAMELEN
- 1,
1232 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1233 struct cgroup_subsys
*ss
= subsys
[i
];
1236 if (strcmp(token
, ss
->name
))
1241 /* Mutually exclusive option 'all' + subsystem name */
1244 set_bit(i
, &opts
->subsys_bits
);
1249 if (i
== CGROUP_SUBSYS_COUNT
)
1254 * If the 'all' option was specified select all the subsystems,
1255 * otherwise if 'none', 'name=' and a subsystem name options
1256 * were not specified, let's default to 'all'
1258 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
)) {
1259 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1260 struct cgroup_subsys
*ss
= subsys
[i
];
1265 set_bit(i
, &opts
->subsys_bits
);
1269 /* Consistency checks */
1272 * Option noprefix was introduced just for backward compatibility
1273 * with the old cpuset, so we allow noprefix only if mounting just
1274 * the cpuset subsystem.
1276 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1277 (opts
->subsys_bits
& mask
))
1281 /* Can't specify "none" and some subsystems */
1282 if (opts
->subsys_bits
&& opts
->none
)
1286 * We either have to specify by name or by subsystems. (So all
1287 * empty hierarchies must have a name).
1289 if (!opts
->subsys_bits
&& !opts
->name
)
1293 * Grab references on all the modules we'll need, so the subsystems
1294 * don't dance around before rebind_subsystems attaches them. This may
1295 * take duplicate reference counts on a subsystem that's already used,
1296 * but rebind_subsystems handles this case.
1298 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1299 unsigned long bit
= 1UL << i
;
1301 if (!(bit
& opts
->subsys_bits
))
1303 if (!try_module_get(subsys
[i
]->module
)) {
1304 module_pin_failed
= true;
1308 if (module_pin_failed
) {
1310 * oops, one of the modules was going away. this means that we
1311 * raced with a module_delete call, and to the user this is
1312 * essentially a "subsystem doesn't exist" case.
1314 for (i
--; i
>= CGROUP_BUILTIN_SUBSYS_COUNT
; i
--) {
1315 /* drop refcounts only on the ones we took */
1316 unsigned long bit
= 1UL << i
;
1318 if (!(bit
& opts
->subsys_bits
))
1320 module_put(subsys
[i
]->module
);
1328 static void drop_parsed_module_refcounts(unsigned long subsys_bits
)
1331 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1332 unsigned long bit
= 1UL << i
;
1334 if (!(bit
& subsys_bits
))
1336 module_put(subsys
[i
]->module
);
1340 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1343 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1344 struct cgroup
*cgrp
= &root
->top_cgroup
;
1345 struct cgroup_sb_opts opts
;
1347 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1348 mutex_lock(&cgroup_mutex
);
1349 mutex_lock(&cgroup_root_mutex
);
1351 /* See what subsystems are wanted */
1352 ret
= parse_cgroupfs_options(data
, &opts
);
1356 /* See feature-removal-schedule.txt */
1357 if (opts
.subsys_bits
!= root
->actual_subsys_bits
|| opts
.release_agent
)
1358 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1359 task_tgid_nr(current
), current
->comm
);
1361 /* Don't allow flags or name to change at remount */
1362 if (opts
.flags
!= root
->flags
||
1363 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1365 drop_parsed_module_refcounts(opts
.subsys_bits
);
1369 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
1371 drop_parsed_module_refcounts(opts
.subsys_bits
);
1375 /* clear out any existing files and repopulate subsystem files */
1376 cgroup_clear_directory(cgrp
->dentry
);
1377 cgroup_populate_dir(cgrp
);
1379 if (opts
.release_agent
)
1380 strcpy(root
->release_agent_path
, opts
.release_agent
);
1382 kfree(opts
.release_agent
);
1384 mutex_unlock(&cgroup_root_mutex
);
1385 mutex_unlock(&cgroup_mutex
);
1386 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1390 static const struct super_operations cgroup_ops
= {
1391 .statfs
= simple_statfs
,
1392 .drop_inode
= generic_delete_inode
,
1393 .show_options
= cgroup_show_options
,
1394 .remount_fs
= cgroup_remount
,
1397 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1399 INIT_LIST_HEAD(&cgrp
->sibling
);
1400 INIT_LIST_HEAD(&cgrp
->children
);
1401 INIT_LIST_HEAD(&cgrp
->files
);
1402 INIT_LIST_HEAD(&cgrp
->css_sets
);
1403 INIT_LIST_HEAD(&cgrp
->release_list
);
1404 INIT_LIST_HEAD(&cgrp
->pidlists
);
1405 mutex_init(&cgrp
->pidlist_mutex
);
1406 INIT_LIST_HEAD(&cgrp
->event_list
);
1407 spin_lock_init(&cgrp
->event_list_lock
);
1410 static void init_cgroup_root(struct cgroupfs_root
*root
)
1412 struct cgroup
*cgrp
= &root
->top_cgroup
;
1414 INIT_LIST_HEAD(&root
->subsys_list
);
1415 INIT_LIST_HEAD(&root
->root_list
);
1416 INIT_LIST_HEAD(&root
->allcg_list
);
1417 root
->number_of_cgroups
= 1;
1419 cgrp
->top_cgroup
= cgrp
;
1420 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
1421 init_cgroup_housekeeping(cgrp
);
1424 static bool init_root_id(struct cgroupfs_root
*root
)
1429 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1431 spin_lock(&hierarchy_id_lock
);
1432 /* Try to allocate the next unused ID */
1433 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1434 &root
->hierarchy_id
);
1436 /* Try again starting from 0 */
1437 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1439 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1440 } else if (ret
!= -EAGAIN
) {
1441 /* Can only get here if the 31-bit IDR is full ... */
1444 spin_unlock(&hierarchy_id_lock
);
1449 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1451 struct cgroup_sb_opts
*opts
= data
;
1452 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1454 /* If we asked for a name then it must match */
1455 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1459 * If we asked for subsystems (or explicitly for no
1460 * subsystems) then they must match
1462 if ((opts
->subsys_bits
|| opts
->none
)
1463 && (opts
->subsys_bits
!= root
->subsys_bits
))
1469 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1471 struct cgroupfs_root
*root
;
1473 if (!opts
->subsys_bits
&& !opts
->none
)
1476 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1478 return ERR_PTR(-ENOMEM
);
1480 if (!init_root_id(root
)) {
1482 return ERR_PTR(-ENOMEM
);
1484 init_cgroup_root(root
);
1486 root
->subsys_bits
= opts
->subsys_bits
;
1487 root
->flags
= opts
->flags
;
1488 if (opts
->release_agent
)
1489 strcpy(root
->release_agent_path
, opts
->release_agent
);
1491 strcpy(root
->name
, opts
->name
);
1492 if (opts
->clone_children
)
1493 set_bit(CGRP_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1497 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1502 BUG_ON(!root
->hierarchy_id
);
1503 spin_lock(&hierarchy_id_lock
);
1504 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1505 spin_unlock(&hierarchy_id_lock
);
1509 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1512 struct cgroup_sb_opts
*opts
= data
;
1514 /* If we don't have a new root, we can't set up a new sb */
1515 if (!opts
->new_root
)
1518 BUG_ON(!opts
->subsys_bits
&& !opts
->none
);
1520 ret
= set_anon_super(sb
, NULL
);
1524 sb
->s_fs_info
= opts
->new_root
;
1525 opts
->new_root
->sb
= sb
;
1527 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1528 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1529 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1530 sb
->s_op
= &cgroup_ops
;
1535 static int cgroup_get_rootdir(struct super_block
*sb
)
1537 static const struct dentry_operations cgroup_dops
= {
1538 .d_iput
= cgroup_diput
,
1539 .d_delete
= cgroup_delete
,
1542 struct inode
*inode
=
1543 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1548 inode
->i_fop
= &simple_dir_operations
;
1549 inode
->i_op
= &cgroup_dir_inode_operations
;
1550 /* directories start off with i_nlink == 2 (for "." entry) */
1552 sb
->s_root
= d_make_root(inode
);
1555 /* for everything else we want ->d_op set */
1556 sb
->s_d_op
= &cgroup_dops
;
1560 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1561 int flags
, const char *unused_dev_name
,
1564 struct cgroup_sb_opts opts
;
1565 struct cgroupfs_root
*root
;
1567 struct super_block
*sb
;
1568 struct cgroupfs_root
*new_root
;
1569 struct inode
*inode
;
1571 /* First find the desired set of subsystems */
1572 mutex_lock(&cgroup_mutex
);
1573 ret
= parse_cgroupfs_options(data
, &opts
);
1574 mutex_unlock(&cgroup_mutex
);
1579 * Allocate a new cgroup root. We may not need it if we're
1580 * reusing an existing hierarchy.
1582 new_root
= cgroup_root_from_opts(&opts
);
1583 if (IS_ERR(new_root
)) {
1584 ret
= PTR_ERR(new_root
);
1587 opts
.new_root
= new_root
;
1589 /* Locate an existing or new sb for this hierarchy */
1590 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, &opts
);
1593 cgroup_drop_root(opts
.new_root
);
1597 root
= sb
->s_fs_info
;
1599 if (root
== opts
.new_root
) {
1600 /* We used the new root structure, so this is a new hierarchy */
1601 struct list_head tmp_cg_links
;
1602 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1603 struct cgroupfs_root
*existing_root
;
1604 const struct cred
*cred
;
1607 BUG_ON(sb
->s_root
!= NULL
);
1609 ret
= cgroup_get_rootdir(sb
);
1611 goto drop_new_super
;
1612 inode
= sb
->s_root
->d_inode
;
1614 mutex_lock(&inode
->i_mutex
);
1615 mutex_lock(&cgroup_mutex
);
1616 mutex_lock(&cgroup_root_mutex
);
1618 /* Check for name clashes with existing mounts */
1620 if (strlen(root
->name
))
1621 for_each_active_root(existing_root
)
1622 if (!strcmp(existing_root
->name
, root
->name
))
1626 * We're accessing css_set_count without locking
1627 * css_set_lock here, but that's OK - it can only be
1628 * increased by someone holding cgroup_lock, and
1629 * that's us. The worst that can happen is that we
1630 * have some link structures left over
1632 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1636 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1637 if (ret
== -EBUSY
) {
1638 free_cg_links(&tmp_cg_links
);
1642 * There must be no failure case after here, since rebinding
1643 * takes care of subsystems' refcounts, which are explicitly
1644 * dropped in the failure exit path.
1647 /* EBUSY should be the only error here */
1650 list_add(&root
->root_list
, &roots
);
1653 sb
->s_root
->d_fsdata
= root_cgrp
;
1654 root
->top_cgroup
.dentry
= sb
->s_root
;
1656 /* Link the top cgroup in this hierarchy into all
1657 * the css_set objects */
1658 write_lock(&css_set_lock
);
1659 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1660 struct hlist_head
*hhead
= &css_set_table
[i
];
1661 struct hlist_node
*node
;
1664 hlist_for_each_entry(cg
, node
, hhead
, hlist
)
1665 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1667 write_unlock(&css_set_lock
);
1669 free_cg_links(&tmp_cg_links
);
1671 BUG_ON(!list_empty(&root_cgrp
->sibling
));
1672 BUG_ON(!list_empty(&root_cgrp
->children
));
1673 BUG_ON(root
->number_of_cgroups
!= 1);
1675 cred
= override_creds(&init_cred
);
1676 cgroup_populate_dir(root_cgrp
);
1678 mutex_unlock(&cgroup_root_mutex
);
1679 mutex_unlock(&cgroup_mutex
);
1680 mutex_unlock(&inode
->i_mutex
);
1683 * We re-used an existing hierarchy - the new root (if
1684 * any) is not needed
1686 cgroup_drop_root(opts
.new_root
);
1687 /* no subsys rebinding, so refcounts don't change */
1688 drop_parsed_module_refcounts(opts
.subsys_bits
);
1691 kfree(opts
.release_agent
);
1693 return dget(sb
->s_root
);
1696 mutex_unlock(&cgroup_root_mutex
);
1697 mutex_unlock(&cgroup_mutex
);
1698 mutex_unlock(&inode
->i_mutex
);
1700 deactivate_locked_super(sb
);
1702 drop_parsed_module_refcounts(opts
.subsys_bits
);
1704 kfree(opts
.release_agent
);
1706 return ERR_PTR(ret
);
1709 static void cgroup_kill_sb(struct super_block
*sb
) {
1710 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1711 struct cgroup
*cgrp
= &root
->top_cgroup
;
1713 struct cg_cgroup_link
*link
;
1714 struct cg_cgroup_link
*saved_link
;
1718 BUG_ON(root
->number_of_cgroups
!= 1);
1719 BUG_ON(!list_empty(&cgrp
->children
));
1720 BUG_ON(!list_empty(&cgrp
->sibling
));
1722 mutex_lock(&cgroup_mutex
);
1723 mutex_lock(&cgroup_root_mutex
);
1725 /* Rebind all subsystems back to the default hierarchy */
1726 ret
= rebind_subsystems(root
, 0);
1727 /* Shouldn't be able to fail ... */
1731 * Release all the links from css_sets to this hierarchy's
1734 write_lock(&css_set_lock
);
1736 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1738 list_del(&link
->cg_link_list
);
1739 list_del(&link
->cgrp_link_list
);
1742 write_unlock(&css_set_lock
);
1744 if (!list_empty(&root
->root_list
)) {
1745 list_del(&root
->root_list
);
1749 mutex_unlock(&cgroup_root_mutex
);
1750 mutex_unlock(&cgroup_mutex
);
1752 kill_litter_super(sb
);
1753 cgroup_drop_root(root
);
1756 static struct file_system_type cgroup_fs_type
= {
1758 .mount
= cgroup_mount
,
1759 .kill_sb
= cgroup_kill_sb
,
1762 static struct kobject
*cgroup_kobj
;
1765 * cgroup_path - generate the path of a cgroup
1766 * @cgrp: the cgroup in question
1767 * @buf: the buffer to write the path into
1768 * @buflen: the length of the buffer
1770 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1771 * reference. Writes path of cgroup into buf. Returns 0 on success,
1774 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1777 struct dentry
*dentry
= rcu_dereference_check(cgrp
->dentry
,
1778 cgroup_lock_is_held());
1780 if (!dentry
|| cgrp
== dummytop
) {
1782 * Inactive subsystems have no dentry for their root
1789 start
= buf
+ buflen
;
1793 int len
= dentry
->d_name
.len
;
1795 if ((start
-= len
) < buf
)
1796 return -ENAMETOOLONG
;
1797 memcpy(start
, dentry
->d_name
.name
, len
);
1798 cgrp
= cgrp
->parent
;
1802 dentry
= rcu_dereference_check(cgrp
->dentry
,
1803 cgroup_lock_is_held());
1807 return -ENAMETOOLONG
;
1810 memmove(buf
, start
, buf
+ buflen
- start
);
1813 EXPORT_SYMBOL_GPL(cgroup_path
);
1816 * Control Group taskset
1818 struct task_and_cgroup
{
1819 struct task_struct
*task
;
1820 struct cgroup
*cgrp
;
1824 struct cgroup_taskset
{
1825 struct task_and_cgroup single
;
1826 struct flex_array
*tc_array
;
1829 struct cgroup
*cur_cgrp
;
1833 * cgroup_taskset_first - reset taskset and return the first task
1834 * @tset: taskset of interest
1836 * @tset iteration is initialized and the first task is returned.
1838 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1840 if (tset
->tc_array
) {
1842 return cgroup_taskset_next(tset
);
1844 tset
->cur_cgrp
= tset
->single
.cgrp
;
1845 return tset
->single
.task
;
1848 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1851 * cgroup_taskset_next - iterate to the next task in taskset
1852 * @tset: taskset of interest
1854 * Return the next task in @tset. Iteration must have been initialized
1855 * with cgroup_taskset_first().
1857 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1859 struct task_and_cgroup
*tc
;
1861 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1864 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1865 tset
->cur_cgrp
= tc
->cgrp
;
1868 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1871 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1872 * @tset: taskset of interest
1874 * Return the cgroup for the current (last returned) task of @tset. This
1875 * function must be preceded by either cgroup_taskset_first() or
1876 * cgroup_taskset_next().
1878 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1880 return tset
->cur_cgrp
;
1882 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1885 * cgroup_taskset_size - return the number of tasks in taskset
1886 * @tset: taskset of interest
1888 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1890 return tset
->tc_array
? tset
->tc_array_len
: 1;
1892 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1896 * cgroup_task_migrate - move a task from one cgroup to another.
1898 * 'guarantee' is set if the caller promises that a new css_set for the task
1899 * will already exist. If not set, this function might sleep, and can fail with
1900 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1902 static void cgroup_task_migrate(struct cgroup
*cgrp
, struct cgroup
*oldcgrp
,
1903 struct task_struct
*tsk
, struct css_set
*newcg
)
1905 struct css_set
*oldcg
;
1908 * We are synchronized through threadgroup_lock() against PF_EXITING
1909 * setting such that we can't race against cgroup_exit() changing the
1910 * css_set to init_css_set and dropping the old one.
1912 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1913 oldcg
= tsk
->cgroups
;
1916 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1919 /* Update the css_set linked lists if we're using them */
1920 write_lock(&css_set_lock
);
1921 if (!list_empty(&tsk
->cg_list
))
1922 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1923 write_unlock(&css_set_lock
);
1926 * We just gained a reference on oldcg by taking it from the task. As
1927 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1928 * it here; it will be freed under RCU.
1932 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1936 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1937 * @cgrp: the cgroup the task is attaching to
1938 * @tsk: the task to be attached
1940 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1943 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1946 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1947 struct cgroup
*oldcgrp
;
1948 struct cgroupfs_root
*root
= cgrp
->root
;
1949 struct cgroup_taskset tset
= { };
1950 struct css_set
*newcg
;
1952 /* @tsk either already exited or can't exit until the end */
1953 if (tsk
->flags
& PF_EXITING
)
1956 /* Nothing to do if the task is already in that cgroup */
1957 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1958 if (cgrp
== oldcgrp
)
1961 tset
.single
.task
= tsk
;
1962 tset
.single
.cgrp
= oldcgrp
;
1964 for_each_subsys(root
, ss
) {
1965 if (ss
->can_attach
) {
1966 retval
= ss
->can_attach(cgrp
, &tset
);
1969 * Remember on which subsystem the can_attach()
1970 * failed, so that we only call cancel_attach()
1971 * against the subsystems whose can_attach()
1972 * succeeded. (See below)
1980 newcg
= find_css_set(tsk
->cgroups
, cgrp
);
1986 cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, newcg
);
1988 for_each_subsys(root
, ss
) {
1990 ss
->attach(cgrp
, &tset
);
1996 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1997 * is no longer empty.
1999 cgroup_wakeup_rmdir_waiter(cgrp
);
2002 for_each_subsys(root
, ss
) {
2003 if (ss
== failed_ss
)
2005 * This subsystem was the one that failed the
2006 * can_attach() check earlier, so we don't need
2007 * to call cancel_attach() against it or any
2008 * remaining subsystems.
2011 if (ss
->cancel_attach
)
2012 ss
->cancel_attach(cgrp
, &tset
);
2019 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2020 * @from: attach to all cgroups of a given task
2021 * @tsk: the task to be attached
2023 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2025 struct cgroupfs_root
*root
;
2029 for_each_active_root(root
) {
2030 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
2032 retval
= cgroup_attach_task(from_cg
, tsk
);
2040 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2043 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2044 * @cgrp: the cgroup to attach to
2045 * @leader: the threadgroup leader task_struct of the group to be attached
2047 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2048 * task_lock of each thread in leader's threadgroup individually in turn.
2050 static int cgroup_attach_proc(struct cgroup
*cgrp
, struct task_struct
*leader
)
2052 int retval
, i
, group_size
;
2053 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
2054 /* guaranteed to be initialized later, but the compiler needs this */
2055 struct cgroupfs_root
*root
= cgrp
->root
;
2056 /* threadgroup list cursor and array */
2057 struct task_struct
*tsk
;
2058 struct task_and_cgroup
*tc
;
2059 struct flex_array
*group
;
2060 struct cgroup_taskset tset
= { };
2063 * step 0: in order to do expensive, possibly blocking operations for
2064 * every thread, we cannot iterate the thread group list, since it needs
2065 * rcu or tasklist locked. instead, build an array of all threads in the
2066 * group - group_rwsem prevents new threads from appearing, and if
2067 * threads exit, this will just be an over-estimate.
2069 group_size
= get_nr_threads(leader
);
2070 /* flex_array supports very large thread-groups better than kmalloc. */
2071 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2074 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2075 retval
= flex_array_prealloc(group
, 0, group_size
- 1, GFP_KERNEL
);
2077 goto out_free_group_list
;
2082 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2083 * already PF_EXITING could be freed from underneath us unless we
2084 * take an rcu_read_lock.
2088 struct task_and_cgroup ent
;
2090 /* @tsk either already exited or can't exit until the end */
2091 if (tsk
->flags
& PF_EXITING
)
2094 /* as per above, nr_threads may decrease, but not increase. */
2095 BUG_ON(i
>= group_size
);
2097 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2098 /* nothing to do if this task is already in the cgroup */
2099 if (ent
.cgrp
== cgrp
)
2102 * saying GFP_ATOMIC has no effect here because we did prealloc
2103 * earlier, but it's good form to communicate our expectations.
2105 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2106 BUG_ON(retval
!= 0);
2108 } while_each_thread(leader
, tsk
);
2110 /* remember the number of threads in the array for later. */
2112 tset
.tc_array
= group
;
2113 tset
.tc_array_len
= group_size
;
2115 /* methods shouldn't be called if no task is actually migrating */
2118 goto out_free_group_list
;
2121 * step 1: check that we can legitimately attach to the cgroup.
2123 for_each_subsys(root
, ss
) {
2124 if (ss
->can_attach
) {
2125 retval
= ss
->can_attach(cgrp
, &tset
);
2128 goto out_cancel_attach
;
2134 * step 2: make sure css_sets exist for all threads to be migrated.
2135 * we use find_css_set, which allocates a new one if necessary.
2137 for (i
= 0; i
< group_size
; i
++) {
2138 tc
= flex_array_get(group
, i
);
2139 tc
->cg
= find_css_set(tc
->task
->cgroups
, cgrp
);
2142 goto out_put_css_set_refs
;
2147 * step 3: now that we're guaranteed success wrt the css_sets,
2148 * proceed to move all tasks to the new cgroup. There are no
2149 * failure cases after here, so this is the commit point.
2151 for (i
= 0; i
< group_size
; i
++) {
2152 tc
= flex_array_get(group
, i
);
2153 cgroup_task_migrate(cgrp
, tc
->cgrp
, tc
->task
, tc
->cg
);
2155 /* nothing is sensitive to fork() after this point. */
2158 * step 4: do subsystem attach callbacks.
2160 for_each_subsys(root
, ss
) {
2162 ss
->attach(cgrp
, &tset
);
2166 * step 5: success! and cleanup
2169 cgroup_wakeup_rmdir_waiter(cgrp
);
2171 out_put_css_set_refs
:
2173 for (i
= 0; i
< group_size
; i
++) {
2174 tc
= flex_array_get(group
, i
);
2177 put_css_set(tc
->cg
);
2182 for_each_subsys(root
, ss
) {
2183 if (ss
== failed_ss
)
2185 if (ss
->cancel_attach
)
2186 ss
->cancel_attach(cgrp
, &tset
);
2189 out_free_group_list
:
2190 flex_array_free(group
);
2195 * Find the task_struct of the task to attach by vpid and pass it along to the
2196 * function to attach either it or all tasks in its threadgroup. Will lock
2197 * cgroup_mutex and threadgroup; may take task_lock of task.
2199 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2201 struct task_struct
*tsk
;
2202 const struct cred
*cred
= current_cred(), *tcred
;
2205 if (!cgroup_lock_live_group(cgrp
))
2211 tsk
= find_task_by_vpid(pid
);
2215 goto out_unlock_cgroup
;
2218 * even if we're attaching all tasks in the thread group, we
2219 * only need to check permissions on one of them.
2221 tcred
= __task_cred(tsk
);
2222 if (!uid_eq(cred
->euid
, GLOBAL_ROOT_UID
) &&
2223 !uid_eq(cred
->euid
, tcred
->uid
) &&
2224 !uid_eq(cred
->euid
, tcred
->suid
)) {
2227 goto out_unlock_cgroup
;
2233 tsk
= tsk
->group_leader
;
2236 * Workqueue threads may acquire PF_THREAD_BOUND and become
2237 * trapped in a cpuset, or RT worker may be born in a cgroup
2238 * with no rt_runtime allocated. Just say no.
2240 if (tsk
== kthreadd_task
|| (tsk
->flags
& PF_THREAD_BOUND
)) {
2243 goto out_unlock_cgroup
;
2246 get_task_struct(tsk
);
2249 threadgroup_lock(tsk
);
2251 if (!thread_group_leader(tsk
)) {
2253 * a race with de_thread from another thread's exec()
2254 * may strip us of our leadership, if this happens,
2255 * there is no choice but to throw this task away and
2256 * try again; this is
2257 * "double-double-toil-and-trouble-check locking".
2259 threadgroup_unlock(tsk
);
2260 put_task_struct(tsk
);
2261 goto retry_find_task
;
2263 ret
= cgroup_attach_proc(cgrp
, tsk
);
2265 ret
= cgroup_attach_task(cgrp
, tsk
);
2266 threadgroup_unlock(tsk
);
2268 put_task_struct(tsk
);
2274 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2276 return attach_task_by_pid(cgrp
, pid
, false);
2279 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2281 return attach_task_by_pid(cgrp
, tgid
, true);
2285 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2286 * @cgrp: the cgroup to be checked for liveness
2288 * On success, returns true; the lock should be later released with
2289 * cgroup_unlock(). On failure returns false with no lock held.
2291 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
2293 mutex_lock(&cgroup_mutex
);
2294 if (cgroup_is_removed(cgrp
)) {
2295 mutex_unlock(&cgroup_mutex
);
2300 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
2302 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2305 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2306 if (strlen(buffer
) >= PATH_MAX
)
2308 if (!cgroup_lock_live_group(cgrp
))
2310 mutex_lock(&cgroup_root_mutex
);
2311 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2312 mutex_unlock(&cgroup_root_mutex
);
2317 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2318 struct seq_file
*seq
)
2320 if (!cgroup_lock_live_group(cgrp
))
2322 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2323 seq_putc(seq
, '\n');
2328 /* A buffer size big enough for numbers or short strings */
2329 #define CGROUP_LOCAL_BUFFER_SIZE 64
2331 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2333 const char __user
*userbuf
,
2334 size_t nbytes
, loff_t
*unused_ppos
)
2336 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2342 if (nbytes
>= sizeof(buffer
))
2344 if (copy_from_user(buffer
, userbuf
, nbytes
))
2347 buffer
[nbytes
] = 0; /* nul-terminate */
2348 if (cft
->write_u64
) {
2349 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2352 retval
= cft
->write_u64(cgrp
, cft
, val
);
2354 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2357 retval
= cft
->write_s64(cgrp
, cft
, val
);
2364 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2366 const char __user
*userbuf
,
2367 size_t nbytes
, loff_t
*unused_ppos
)
2369 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2371 size_t max_bytes
= cft
->max_write_len
;
2372 char *buffer
= local_buffer
;
2375 max_bytes
= sizeof(local_buffer
) - 1;
2376 if (nbytes
>= max_bytes
)
2378 /* Allocate a dynamic buffer if we need one */
2379 if (nbytes
>= sizeof(local_buffer
)) {
2380 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2384 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2389 buffer
[nbytes
] = 0; /* nul-terminate */
2390 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2394 if (buffer
!= local_buffer
)
2399 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2400 size_t nbytes
, loff_t
*ppos
)
2402 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2403 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2405 if (cgroup_is_removed(cgrp
))
2408 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2409 if (cft
->write_u64
|| cft
->write_s64
)
2410 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2411 if (cft
->write_string
)
2412 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2414 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2415 return ret
? ret
: nbytes
;
2420 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2422 char __user
*buf
, size_t nbytes
,
2425 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2426 u64 val
= cft
->read_u64(cgrp
, cft
);
2427 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2429 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2432 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2434 char __user
*buf
, size_t nbytes
,
2437 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2438 s64 val
= cft
->read_s64(cgrp
, cft
);
2439 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2441 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2444 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2445 size_t nbytes
, loff_t
*ppos
)
2447 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2448 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2450 if (cgroup_is_removed(cgrp
))
2454 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2456 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2458 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2463 * seqfile ops/methods for returning structured data. Currently just
2464 * supports string->u64 maps, but can be extended in future.
2467 struct cgroup_seqfile_state
{
2469 struct cgroup
*cgroup
;
2472 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2474 struct seq_file
*sf
= cb
->state
;
2475 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2478 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2480 struct cgroup_seqfile_state
*state
= m
->private;
2481 struct cftype
*cft
= state
->cft
;
2482 if (cft
->read_map
) {
2483 struct cgroup_map_cb cb
= {
2484 .fill
= cgroup_map_add
,
2487 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2489 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2492 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2494 struct seq_file
*seq
= file
->private_data
;
2495 kfree(seq
->private);
2496 return single_release(inode
, file
);
2499 static const struct file_operations cgroup_seqfile_operations
= {
2501 .write
= cgroup_file_write
,
2502 .llseek
= seq_lseek
,
2503 .release
= cgroup_seqfile_release
,
2506 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2511 err
= generic_file_open(inode
, file
);
2514 cft
= __d_cft(file
->f_dentry
);
2516 if (cft
->read_map
|| cft
->read_seq_string
) {
2517 struct cgroup_seqfile_state
*state
=
2518 kzalloc(sizeof(*state
), GFP_USER
);
2522 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2523 file
->f_op
= &cgroup_seqfile_operations
;
2524 err
= single_open(file
, cgroup_seqfile_show
, state
);
2527 } else if (cft
->open
)
2528 err
= cft
->open(inode
, file
);
2535 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2537 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2539 return cft
->release(inode
, file
);
2544 * cgroup_rename - Only allow simple rename of directories in place.
2546 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2547 struct inode
*new_dir
, struct dentry
*new_dentry
)
2549 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2551 if (new_dentry
->d_inode
)
2553 if (old_dir
!= new_dir
)
2555 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2558 static const struct file_operations cgroup_file_operations
= {
2559 .read
= cgroup_file_read
,
2560 .write
= cgroup_file_write
,
2561 .llseek
= generic_file_llseek
,
2562 .open
= cgroup_file_open
,
2563 .release
= cgroup_file_release
,
2566 static const struct inode_operations cgroup_dir_inode_operations
= {
2567 .lookup
= cgroup_lookup
,
2568 .mkdir
= cgroup_mkdir
,
2569 .rmdir
= cgroup_rmdir
,
2570 .rename
= cgroup_rename
,
2573 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
2575 if (dentry
->d_name
.len
> NAME_MAX
)
2576 return ERR_PTR(-ENAMETOOLONG
);
2577 d_add(dentry
, NULL
);
2582 * Check if a file is a control file
2584 static inline struct cftype
*__file_cft(struct file
*file
)
2586 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2587 return ERR_PTR(-EINVAL
);
2588 return __d_cft(file
->f_dentry
);
2591 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2592 struct super_block
*sb
)
2594 struct inode
*inode
;
2598 if (dentry
->d_inode
)
2601 inode
= cgroup_new_inode(mode
, sb
);
2605 if (S_ISDIR(mode
)) {
2606 inode
->i_op
= &cgroup_dir_inode_operations
;
2607 inode
->i_fop
= &simple_dir_operations
;
2609 /* start off with i_nlink == 2 (for "." entry) */
2612 /* start with the directory inode held, so that we can
2613 * populate it without racing with another mkdir */
2614 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2615 } else if (S_ISREG(mode
)) {
2617 inode
->i_fop
= &cgroup_file_operations
;
2619 d_instantiate(dentry
, inode
);
2620 dget(dentry
); /* Extra count - pin the dentry in core */
2625 * cgroup_create_dir - create a directory for an object.
2626 * @cgrp: the cgroup we create the directory for. It must have a valid
2627 * ->parent field. And we are going to fill its ->dentry field.
2628 * @dentry: dentry of the new cgroup
2629 * @mode: mode to set on new directory.
2631 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
2634 struct dentry
*parent
;
2637 parent
= cgrp
->parent
->dentry
;
2638 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
2640 dentry
->d_fsdata
= cgrp
;
2641 inc_nlink(parent
->d_inode
);
2642 rcu_assign_pointer(cgrp
->dentry
, dentry
);
2651 * cgroup_file_mode - deduce file mode of a control file
2652 * @cft: the control file in question
2654 * returns cft->mode if ->mode is not 0
2655 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2656 * returns S_IRUGO if it has only a read handler
2657 * returns S_IWUSR if it has only a write hander
2659 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2666 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2667 cft
->read_map
|| cft
->read_seq_string
)
2670 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2671 cft
->write_string
|| cft
->trigger
)
2677 static int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2678 const struct cftype
*cft
)
2680 struct dentry
*dir
= cgrp
->dentry
;
2681 struct cgroup
*parent
= __d_cgrp(dir
);
2682 struct dentry
*dentry
;
2686 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2688 /* does @cft->flags tell us to skip creation on @cgrp? */
2689 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2691 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2694 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2695 strcpy(name
, subsys
->name
);
2698 strcat(name
, cft
->name
);
2700 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2702 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2706 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2707 if (IS_ERR(dentry
)) {
2708 error
= PTR_ERR(dentry
);
2712 mode
= cgroup_file_mode(cft
);
2713 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2715 cfe
->type
= (void *)cft
;
2716 cfe
->dentry
= dentry
;
2717 dentry
->d_fsdata
= cfe
;
2718 list_add_tail(&cfe
->node
, &parent
->files
);
2727 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2728 const struct cftype cfts
[], bool is_add
)
2730 const struct cftype
*cft
;
2733 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2735 err
= cgroup_add_file(cgrp
, subsys
, cft
);
2737 err
= cgroup_rm_file(cgrp
, cft
);
2739 pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2740 is_add
? "add" : "remove", cft
->name
, err
);
2747 static DEFINE_MUTEX(cgroup_cft_mutex
);
2749 static void cgroup_cfts_prepare(void)
2750 __acquires(&cgroup_cft_mutex
) __acquires(&cgroup_mutex
)
2753 * Thanks to the entanglement with vfs inode locking, we can't walk
2754 * the existing cgroups under cgroup_mutex and create files.
2755 * Instead, we increment reference on all cgroups and build list of
2756 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2757 * exclusive access to the field.
2759 mutex_lock(&cgroup_cft_mutex
);
2760 mutex_lock(&cgroup_mutex
);
2763 static void cgroup_cfts_commit(struct cgroup_subsys
*ss
,
2764 const struct cftype
*cfts
, bool is_add
)
2765 __releases(&cgroup_mutex
) __releases(&cgroup_cft_mutex
)
2768 struct cgroup
*cgrp
, *n
;
2770 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2771 if (cfts
&& ss
->root
!= &rootnode
) {
2772 list_for_each_entry(cgrp
, &ss
->root
->allcg_list
, allcg_node
) {
2774 list_add_tail(&cgrp
->cft_q_node
, &pending
);
2778 mutex_unlock(&cgroup_mutex
);
2781 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2782 * files for all cgroups which were created before.
2784 list_for_each_entry_safe(cgrp
, n
, &pending
, cft_q_node
) {
2785 struct inode
*inode
= cgrp
->dentry
->d_inode
;
2787 mutex_lock(&inode
->i_mutex
);
2788 mutex_lock(&cgroup_mutex
);
2789 if (!cgroup_is_removed(cgrp
))
2790 cgroup_addrm_files(cgrp
, ss
, cfts
, is_add
);
2791 mutex_unlock(&cgroup_mutex
);
2792 mutex_unlock(&inode
->i_mutex
);
2794 list_del_init(&cgrp
->cft_q_node
);
2798 mutex_unlock(&cgroup_cft_mutex
);
2802 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2803 * @ss: target cgroup subsystem
2804 * @cfts: zero-length name terminated array of cftypes
2806 * Register @cfts to @ss. Files described by @cfts are created for all
2807 * existing cgroups to which @ss is attached and all future cgroups will
2808 * have them too. This function can be called anytime whether @ss is
2811 * Returns 0 on successful registration, -errno on failure. Note that this
2812 * function currently returns 0 as long as @cfts registration is successful
2813 * even if some file creation attempts on existing cgroups fail.
2815 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, const struct cftype
*cfts
)
2817 struct cftype_set
*set
;
2819 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2823 cgroup_cfts_prepare();
2825 list_add_tail(&set
->node
, &ss
->cftsets
);
2826 cgroup_cfts_commit(ss
, cfts
, true);
2830 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2833 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2834 * @ss: target cgroup subsystem
2835 * @cfts: zero-length name terminated array of cftypes
2837 * Unregister @cfts from @ss. Files described by @cfts are removed from
2838 * all existing cgroups to which @ss is attached and all future cgroups
2839 * won't have them either. This function can be called anytime whether @ss
2840 * is attached or not.
2842 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2843 * registered with @ss.
2845 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, const struct cftype
*cfts
)
2847 struct cftype_set
*set
;
2849 cgroup_cfts_prepare();
2851 list_for_each_entry(set
, &ss
->cftsets
, node
) {
2852 if (set
->cfts
== cfts
) {
2853 list_del_init(&set
->node
);
2854 cgroup_cfts_commit(ss
, cfts
, false);
2859 cgroup_cfts_commit(ss
, NULL
, false);
2864 * cgroup_task_count - count the number of tasks in a cgroup.
2865 * @cgrp: the cgroup in question
2867 * Return the number of tasks in the cgroup.
2869 int cgroup_task_count(const struct cgroup
*cgrp
)
2872 struct cg_cgroup_link
*link
;
2874 read_lock(&css_set_lock
);
2875 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2876 count
+= atomic_read(&link
->cg
->refcount
);
2878 read_unlock(&css_set_lock
);
2883 * Advance a list_head iterator. The iterator should be positioned at
2884 * the start of a css_set
2886 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2887 struct cgroup_iter
*it
)
2889 struct list_head
*l
= it
->cg_link
;
2890 struct cg_cgroup_link
*link
;
2893 /* Advance to the next non-empty css_set */
2896 if (l
== &cgrp
->css_sets
) {
2900 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2902 } while (list_empty(&cg
->tasks
));
2904 it
->task
= cg
->tasks
.next
;
2908 * To reduce the fork() overhead for systems that are not actually
2909 * using their cgroups capability, we don't maintain the lists running
2910 * through each css_set to its tasks until we see the list actually
2911 * used - in other words after the first call to cgroup_iter_start().
2913 static void cgroup_enable_task_cg_lists(void)
2915 struct task_struct
*p
, *g
;
2916 write_lock(&css_set_lock
);
2917 use_task_css_set_links
= 1;
2919 * We need tasklist_lock because RCU is not safe against
2920 * while_each_thread(). Besides, a forking task that has passed
2921 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2922 * is not guaranteed to have its child immediately visible in the
2923 * tasklist if we walk through it with RCU.
2925 read_lock(&tasklist_lock
);
2926 do_each_thread(g
, p
) {
2929 * We should check if the process is exiting, otherwise
2930 * it will race with cgroup_exit() in that the list
2931 * entry won't be deleted though the process has exited.
2933 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2934 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2936 } while_each_thread(g
, p
);
2937 read_unlock(&tasklist_lock
);
2938 write_unlock(&css_set_lock
);
2941 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2942 __acquires(css_set_lock
)
2945 * The first time anyone tries to iterate across a cgroup,
2946 * we need to enable the list linking each css_set to its
2947 * tasks, and fix up all existing tasks.
2949 if (!use_task_css_set_links
)
2950 cgroup_enable_task_cg_lists();
2952 read_lock(&css_set_lock
);
2953 it
->cg_link
= &cgrp
->css_sets
;
2954 cgroup_advance_iter(cgrp
, it
);
2957 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
2958 struct cgroup_iter
*it
)
2960 struct task_struct
*res
;
2961 struct list_head
*l
= it
->task
;
2962 struct cg_cgroup_link
*link
;
2964 /* If the iterator cg is NULL, we have no tasks */
2967 res
= list_entry(l
, struct task_struct
, cg_list
);
2968 /* Advance iterator to find next entry */
2970 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
2971 if (l
== &link
->cg
->tasks
) {
2972 /* We reached the end of this task list - move on to
2973 * the next cg_cgroup_link */
2974 cgroup_advance_iter(cgrp
, it
);
2981 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2982 __releases(css_set_lock
)
2984 read_unlock(&css_set_lock
);
2987 static inline int started_after_time(struct task_struct
*t1
,
2988 struct timespec
*time
,
2989 struct task_struct
*t2
)
2991 int start_diff
= timespec_compare(&t1
->start_time
, time
);
2992 if (start_diff
> 0) {
2994 } else if (start_diff
< 0) {
2998 * Arbitrarily, if two processes started at the same
2999 * time, we'll say that the lower pointer value
3000 * started first. Note that t2 may have exited by now
3001 * so this may not be a valid pointer any longer, but
3002 * that's fine - it still serves to distinguish
3003 * between two tasks started (effectively) simultaneously.
3010 * This function is a callback from heap_insert() and is used to order
3012 * In this case we order the heap in descending task start time.
3014 static inline int started_after(void *p1
, void *p2
)
3016 struct task_struct
*t1
= p1
;
3017 struct task_struct
*t2
= p2
;
3018 return started_after_time(t1
, &t2
->start_time
, t2
);
3022 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3023 * @scan: struct cgroup_scanner containing arguments for the scan
3025 * Arguments include pointers to callback functions test_task() and
3027 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3028 * and if it returns true, call process_task() for it also.
3029 * The test_task pointer may be NULL, meaning always true (select all tasks).
3030 * Effectively duplicates cgroup_iter_{start,next,end}()
3031 * but does not lock css_set_lock for the call to process_task().
3032 * The struct cgroup_scanner may be embedded in any structure of the caller's
3034 * It is guaranteed that process_task() will act on every task that
3035 * is a member of the cgroup for the duration of this call. This
3036 * function may or may not call process_task() for tasks that exit
3037 * or move to a different cgroup during the call, or are forked or
3038 * move into the cgroup during the call.
3040 * Note that test_task() may be called with locks held, and may in some
3041 * situations be called multiple times for the same task, so it should
3043 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3044 * pre-allocated and will be used for heap operations (and its "gt" member will
3045 * be overwritten), else a temporary heap will be used (allocation of which
3046 * may cause this function to fail).
3048 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
3051 struct cgroup_iter it
;
3052 struct task_struct
*p
, *dropped
;
3053 /* Never dereference latest_task, since it's not refcounted */
3054 struct task_struct
*latest_task
= NULL
;
3055 struct ptr_heap tmp_heap
;
3056 struct ptr_heap
*heap
;
3057 struct timespec latest_time
= { 0, 0 };
3060 /* The caller supplied our heap and pre-allocated its memory */
3062 heap
->gt
= &started_after
;
3064 /* We need to allocate our own heap memory */
3066 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3068 /* cannot allocate the heap */
3074 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3075 * to determine which are of interest, and using the scanner's
3076 * "process_task" callback to process any of them that need an update.
3077 * Since we don't want to hold any locks during the task updates,
3078 * gather tasks to be processed in a heap structure.
3079 * The heap is sorted by descending task start time.
3080 * If the statically-sized heap fills up, we overflow tasks that
3081 * started later, and in future iterations only consider tasks that
3082 * started after the latest task in the previous pass. This
3083 * guarantees forward progress and that we don't miss any tasks.
3086 cgroup_iter_start(scan
->cg
, &it
);
3087 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
3089 * Only affect tasks that qualify per the caller's callback,
3090 * if he provided one
3092 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
3095 * Only process tasks that started after the last task
3098 if (!started_after_time(p
, &latest_time
, latest_task
))
3100 dropped
= heap_insert(heap
, p
);
3101 if (dropped
== NULL
) {
3103 * The new task was inserted; the heap wasn't
3107 } else if (dropped
!= p
) {
3109 * The new task was inserted, and pushed out a
3113 put_task_struct(dropped
);
3116 * Else the new task was newer than anything already in
3117 * the heap and wasn't inserted
3120 cgroup_iter_end(scan
->cg
, &it
);
3123 for (i
= 0; i
< heap
->size
; i
++) {
3124 struct task_struct
*q
= heap
->ptrs
[i
];
3126 latest_time
= q
->start_time
;
3129 /* Process the task per the caller's callback */
3130 scan
->process_task(q
, scan
);
3134 * If we had to process any tasks at all, scan again
3135 * in case some of them were in the middle of forking
3136 * children that didn't get processed.
3137 * Not the most efficient way to do it, but it avoids
3138 * having to take callback_mutex in the fork path
3142 if (heap
== &tmp_heap
)
3143 heap_free(&tmp_heap
);
3148 * Stuff for reading the 'tasks'/'procs' files.
3150 * Reading this file can return large amounts of data if a cgroup has
3151 * *lots* of attached tasks. So it may need several calls to read(),
3152 * but we cannot guarantee that the information we produce is correct
3153 * unless we produce it entirely atomically.
3157 /* which pidlist file are we talking about? */
3158 enum cgroup_filetype
{
3164 * A pidlist is a list of pids that virtually represents the contents of one
3165 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3166 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3169 struct cgroup_pidlist
{
3171 * used to find which pidlist is wanted. doesn't change as long as
3172 * this particular list stays in the list.
3174 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3177 /* how many elements the above list has */
3179 /* how many files are using the current array */
3181 /* each of these stored in a list by its cgroup */
3182 struct list_head links
;
3183 /* pointer to the cgroup we belong to, for list removal purposes */
3184 struct cgroup
*owner
;
3185 /* protects the other fields */
3186 struct rw_semaphore mutex
;
3190 * The following two functions "fix" the issue where there are more pids
3191 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3192 * TODO: replace with a kernel-wide solution to this problem
3194 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3195 static void *pidlist_allocate(int count
)
3197 if (PIDLIST_TOO_LARGE(count
))
3198 return vmalloc(count
* sizeof(pid_t
));
3200 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3202 static void pidlist_free(void *p
)
3204 if (is_vmalloc_addr(p
))
3209 static void *pidlist_resize(void *p
, int newcount
)
3212 /* note: if new alloc fails, old p will still be valid either way */
3213 if (is_vmalloc_addr(p
)) {
3214 newlist
= vmalloc(newcount
* sizeof(pid_t
));
3217 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
3220 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
3226 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3227 * If the new stripped list is sufficiently smaller and there's enough memory
3228 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3229 * number of unique elements.
3231 /* is the size difference enough that we should re-allocate the array? */
3232 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3233 static int pidlist_uniq(pid_t
**p
, int length
)
3240 * we presume the 0th element is unique, so i starts at 1. trivial
3241 * edge cases first; no work needs to be done for either
3243 if (length
== 0 || length
== 1)
3245 /* src and dest walk down the list; dest counts unique elements */
3246 for (src
= 1; src
< length
; src
++) {
3247 /* find next unique element */
3248 while (list
[src
] == list
[src
-1]) {
3253 /* dest always points to where the next unique element goes */
3254 list
[dest
] = list
[src
];
3259 * if the length difference is large enough, we want to allocate a
3260 * smaller buffer to save memory. if this fails due to out of memory,
3261 * we'll just stay with what we've got.
3263 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
3264 newlist
= pidlist_resize(list
, dest
);
3271 static int cmppid(const void *a
, const void *b
)
3273 return *(pid_t
*)a
- *(pid_t
*)b
;
3277 * find the appropriate pidlist for our purpose (given procs vs tasks)
3278 * returns with the lock on that pidlist already held, and takes care
3279 * of the use count, or returns NULL with no locks held if we're out of
3282 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3283 enum cgroup_filetype type
)
3285 struct cgroup_pidlist
*l
;
3286 /* don't need task_nsproxy() if we're looking at ourself */
3287 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
3290 * We can't drop the pidlist_mutex before taking the l->mutex in case
3291 * the last ref-holder is trying to remove l from the list at the same
3292 * time. Holding the pidlist_mutex precludes somebody taking whichever
3293 * list we find out from under us - compare release_pid_array().
3295 mutex_lock(&cgrp
->pidlist_mutex
);
3296 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3297 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3298 /* make sure l doesn't vanish out from under us */
3299 down_write(&l
->mutex
);
3300 mutex_unlock(&cgrp
->pidlist_mutex
);
3304 /* entry not found; create a new one */
3305 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3307 mutex_unlock(&cgrp
->pidlist_mutex
);
3310 init_rwsem(&l
->mutex
);
3311 down_write(&l
->mutex
);
3313 l
->key
.ns
= get_pid_ns(ns
);
3314 l
->use_count
= 0; /* don't increment here */
3317 list_add(&l
->links
, &cgrp
->pidlists
);
3318 mutex_unlock(&cgrp
->pidlist_mutex
);
3323 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3325 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3326 struct cgroup_pidlist
**lp
)
3330 int pid
, n
= 0; /* used for populating the array */
3331 struct cgroup_iter it
;
3332 struct task_struct
*tsk
;
3333 struct cgroup_pidlist
*l
;
3336 * If cgroup gets more users after we read count, we won't have
3337 * enough space - tough. This race is indistinguishable to the
3338 * caller from the case that the additional cgroup users didn't
3339 * show up until sometime later on.
3341 length
= cgroup_task_count(cgrp
);
3342 array
= pidlist_allocate(length
);
3345 /* now, populate the array */
3346 cgroup_iter_start(cgrp
, &it
);
3347 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3348 if (unlikely(n
== length
))
3350 /* get tgid or pid for procs or tasks file respectively */
3351 if (type
== CGROUP_FILE_PROCS
)
3352 pid
= task_tgid_vnr(tsk
);
3354 pid
= task_pid_vnr(tsk
);
3355 if (pid
> 0) /* make sure to only use valid results */
3358 cgroup_iter_end(cgrp
, &it
);
3360 /* now sort & (if procs) strip out duplicates */
3361 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3362 if (type
== CGROUP_FILE_PROCS
)
3363 length
= pidlist_uniq(&array
, length
);
3364 l
= cgroup_pidlist_find(cgrp
, type
);
3366 pidlist_free(array
);
3369 /* store array, freeing old if necessary - lock already held */
3370 pidlist_free(l
->list
);
3374 up_write(&l
->mutex
);
3380 * cgroupstats_build - build and fill cgroupstats
3381 * @stats: cgroupstats to fill information into
3382 * @dentry: A dentry entry belonging to the cgroup for which stats have
3385 * Build and fill cgroupstats so that taskstats can export it to user
3388 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3391 struct cgroup
*cgrp
;
3392 struct cgroup_iter it
;
3393 struct task_struct
*tsk
;
3396 * Validate dentry by checking the superblock operations,
3397 * and make sure it's a directory.
3399 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3400 !S_ISDIR(dentry
->d_inode
->i_mode
))
3404 cgrp
= dentry
->d_fsdata
;
3406 cgroup_iter_start(cgrp
, &it
);
3407 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3408 switch (tsk
->state
) {
3410 stats
->nr_running
++;
3412 case TASK_INTERRUPTIBLE
:
3413 stats
->nr_sleeping
++;
3415 case TASK_UNINTERRUPTIBLE
:
3416 stats
->nr_uninterruptible
++;
3419 stats
->nr_stopped
++;
3422 if (delayacct_is_task_waiting_on_io(tsk
))
3423 stats
->nr_io_wait
++;
3427 cgroup_iter_end(cgrp
, &it
);
3435 * seq_file methods for the tasks/procs files. The seq_file position is the
3436 * next pid to display; the seq_file iterator is a pointer to the pid
3437 * in the cgroup->l->list array.
3440 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3443 * Initially we receive a position value that corresponds to
3444 * one more than the last pid shown (or 0 on the first call or
3445 * after a seek to the start). Use a binary-search to find the
3446 * next pid to display, if any
3448 struct cgroup_pidlist
*l
= s
->private;
3449 int index
= 0, pid
= *pos
;
3452 down_read(&l
->mutex
);
3454 int end
= l
->length
;
3456 while (index
< end
) {
3457 int mid
= (index
+ end
) / 2;
3458 if (l
->list
[mid
] == pid
) {
3461 } else if (l
->list
[mid
] <= pid
)
3467 /* If we're off the end of the array, we're done */
3468 if (index
>= l
->length
)
3470 /* Update the abstract position to be the actual pid that we found */
3471 iter
= l
->list
+ index
;
3476 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3478 struct cgroup_pidlist
*l
= s
->private;
3482 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3484 struct cgroup_pidlist
*l
= s
->private;
3486 pid_t
*end
= l
->list
+ l
->length
;
3488 * Advance to the next pid in the array. If this goes off the
3500 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3502 return seq_printf(s
, "%d\n", *(int *)v
);
3506 * seq_operations functions for iterating on pidlists through seq_file -
3507 * independent of whether it's tasks or procs
3509 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3510 .start
= cgroup_pidlist_start
,
3511 .stop
= cgroup_pidlist_stop
,
3512 .next
= cgroup_pidlist_next
,
3513 .show
= cgroup_pidlist_show
,
3516 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3519 * the case where we're the last user of this particular pidlist will
3520 * have us remove it from the cgroup's list, which entails taking the
3521 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3522 * pidlist_mutex, we have to take pidlist_mutex first.
3524 mutex_lock(&l
->owner
->pidlist_mutex
);
3525 down_write(&l
->mutex
);
3526 BUG_ON(!l
->use_count
);
3527 if (!--l
->use_count
) {
3528 /* we're the last user if refcount is 0; remove and free */
3529 list_del(&l
->links
);
3530 mutex_unlock(&l
->owner
->pidlist_mutex
);
3531 pidlist_free(l
->list
);
3532 put_pid_ns(l
->key
.ns
);
3533 up_write(&l
->mutex
);
3537 mutex_unlock(&l
->owner
->pidlist_mutex
);
3538 up_write(&l
->mutex
);
3541 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3543 struct cgroup_pidlist
*l
;
3544 if (!(file
->f_mode
& FMODE_READ
))
3547 * the seq_file will only be initialized if the file was opened for
3548 * reading; hence we check if it's not null only in that case.
3550 l
= ((struct seq_file
*)file
->private_data
)->private;
3551 cgroup_release_pid_array(l
);
3552 return seq_release(inode
, file
);
3555 static const struct file_operations cgroup_pidlist_operations
= {
3557 .llseek
= seq_lseek
,
3558 .write
= cgroup_file_write
,
3559 .release
= cgroup_pidlist_release
,
3563 * The following functions handle opens on a file that displays a pidlist
3564 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3567 /* helper function for the two below it */
3568 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3570 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3571 struct cgroup_pidlist
*l
;
3574 /* Nothing to do for write-only files */
3575 if (!(file
->f_mode
& FMODE_READ
))
3578 /* have the array populated */
3579 retval
= pidlist_array_load(cgrp
, type
, &l
);
3582 /* configure file information */
3583 file
->f_op
= &cgroup_pidlist_operations
;
3585 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3587 cgroup_release_pid_array(l
);
3590 ((struct seq_file
*)file
->private_data
)->private = l
;
3593 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3595 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3597 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3599 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3602 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3605 return notify_on_release(cgrp
);
3608 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3612 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3614 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3616 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3621 * Unregister event and free resources.
3623 * Gets called from workqueue.
3625 static void cgroup_event_remove(struct work_struct
*work
)
3627 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3629 struct cgroup
*cgrp
= event
->cgrp
;
3631 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3633 eventfd_ctx_put(event
->eventfd
);
3639 * Gets called on POLLHUP on eventfd when user closes it.
3641 * Called with wqh->lock held and interrupts disabled.
3643 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3644 int sync
, void *key
)
3646 struct cgroup_event
*event
= container_of(wait
,
3647 struct cgroup_event
, wait
);
3648 struct cgroup
*cgrp
= event
->cgrp
;
3649 unsigned long flags
= (unsigned long)key
;
3651 if (flags
& POLLHUP
) {
3652 __remove_wait_queue(event
->wqh
, &event
->wait
);
3653 spin_lock(&cgrp
->event_list_lock
);
3654 list_del(&event
->list
);
3655 spin_unlock(&cgrp
->event_list_lock
);
3657 * We are in atomic context, but cgroup_event_remove() may
3658 * sleep, so we have to call it in workqueue.
3660 schedule_work(&event
->remove
);
3666 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3667 wait_queue_head_t
*wqh
, poll_table
*pt
)
3669 struct cgroup_event
*event
= container_of(pt
,
3670 struct cgroup_event
, pt
);
3673 add_wait_queue(wqh
, &event
->wait
);
3677 * Parse input and register new cgroup event handler.
3679 * Input must be in format '<event_fd> <control_fd> <args>'.
3680 * Interpretation of args is defined by control file implementation.
3682 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3685 struct cgroup_event
*event
= NULL
;
3686 unsigned int efd
, cfd
;
3687 struct file
*efile
= NULL
;
3688 struct file
*cfile
= NULL
;
3692 efd
= simple_strtoul(buffer
, &endp
, 10);
3697 cfd
= simple_strtoul(buffer
, &endp
, 10);
3698 if ((*endp
!= ' ') && (*endp
!= '\0'))
3702 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3706 INIT_LIST_HEAD(&event
->list
);
3707 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3708 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3709 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3711 efile
= eventfd_fget(efd
);
3712 if (IS_ERR(efile
)) {
3713 ret
= PTR_ERR(efile
);
3717 event
->eventfd
= eventfd_ctx_fileget(efile
);
3718 if (IS_ERR(event
->eventfd
)) {
3719 ret
= PTR_ERR(event
->eventfd
);
3729 /* the process need read permission on control file */
3730 /* AV: shouldn't we check that it's been opened for read instead? */
3731 ret
= inode_permission(cfile
->f_path
.dentry
->d_inode
, MAY_READ
);
3735 event
->cft
= __file_cft(cfile
);
3736 if (IS_ERR(event
->cft
)) {
3737 ret
= PTR_ERR(event
->cft
);
3741 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3746 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3747 event
->eventfd
, buffer
);
3751 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3752 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3758 * Events should be removed after rmdir of cgroup directory, but before
3759 * destroying subsystem state objects. Let's take reference to cgroup
3760 * directory dentry to do that.
3764 spin_lock(&cgrp
->event_list_lock
);
3765 list_add(&event
->list
, &cgrp
->event_list
);
3766 spin_unlock(&cgrp
->event_list_lock
);
3777 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3778 eventfd_ctx_put(event
->eventfd
);
3780 if (!IS_ERR_OR_NULL(efile
))
3788 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3791 return clone_children(cgrp
);
3794 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3799 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3801 clear_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3806 * for the common functions, 'private' gives the type of file
3808 /* for hysterical raisins, we can't put this on the older files */
3809 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3810 static struct cftype files
[] = {
3813 .open
= cgroup_tasks_open
,
3814 .write_u64
= cgroup_tasks_write
,
3815 .release
= cgroup_pidlist_release
,
3816 .mode
= S_IRUGO
| S_IWUSR
,
3819 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3820 .open
= cgroup_procs_open
,
3821 .write_u64
= cgroup_procs_write
,
3822 .release
= cgroup_pidlist_release
,
3823 .mode
= S_IRUGO
| S_IWUSR
,
3826 .name
= "notify_on_release",
3827 .read_u64
= cgroup_read_notify_on_release
,
3828 .write_u64
= cgroup_write_notify_on_release
,
3831 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3832 .write_string
= cgroup_write_event_control
,
3836 .name
= "cgroup.clone_children",
3837 .read_u64
= cgroup_clone_children_read
,
3838 .write_u64
= cgroup_clone_children_write
,
3841 .name
= "release_agent",
3842 .flags
= CFTYPE_ONLY_ON_ROOT
,
3843 .read_seq_string
= cgroup_release_agent_show
,
3844 .write_string
= cgroup_release_agent_write
,
3845 .max_write_len
= PATH_MAX
,
3850 static int cgroup_populate_dir(struct cgroup
*cgrp
)
3853 struct cgroup_subsys
*ss
;
3855 err
= cgroup_addrm_files(cgrp
, NULL
, files
, true);
3859 /* process cftsets of each subsystem */
3860 for_each_subsys(cgrp
->root
, ss
) {
3861 struct cftype_set
*set
;
3863 list_for_each_entry(set
, &ss
->cftsets
, node
)
3864 cgroup_addrm_files(cgrp
, ss
, set
->cfts
, true);
3867 /* This cgroup is ready now */
3868 for_each_subsys(cgrp
->root
, ss
) {
3869 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3871 * Update id->css pointer and make this css visible from
3872 * CSS ID functions. This pointer will be dereferened
3873 * from RCU-read-side without locks.
3876 rcu_assign_pointer(css
->id
->css
, css
);
3882 static void css_dput_fn(struct work_struct
*work
)
3884 struct cgroup_subsys_state
*css
=
3885 container_of(work
, struct cgroup_subsys_state
, dput_work
);
3886 struct dentry
*dentry
= css
->cgroup
->dentry
;
3887 struct super_block
*sb
= dentry
->d_sb
;
3889 atomic_inc(&sb
->s_active
);
3891 deactivate_super(sb
);
3894 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
3895 struct cgroup_subsys
*ss
,
3896 struct cgroup
*cgrp
)
3899 atomic_set(&css
->refcnt
, 1);
3902 if (cgrp
== dummytop
)
3903 set_bit(CSS_ROOT
, &css
->flags
);
3904 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
3905 cgrp
->subsys
[ss
->subsys_id
] = css
;
3908 * If !clear_css_refs, css holds an extra ref to @cgrp->dentry
3909 * which is put on the last css_put(). dput() requires process
3910 * context, which css_put() may be called without. @css->dput_work
3911 * will be used to invoke dput() asynchronously from css_put().
3913 INIT_WORK(&css
->dput_work
, css_dput_fn
);
3914 if (ss
->__DEPRECATED_clear_css_refs
)
3915 set_bit(CSS_CLEAR_CSS_REFS
, &css
->flags
);
3918 static void cgroup_lock_hierarchy(struct cgroupfs_root
*root
)
3920 /* We need to take each hierarchy_mutex in a consistent order */
3924 * No worry about a race with rebind_subsystems that might mess up the
3925 * locking order, since both parties are under cgroup_mutex.
3927 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3928 struct cgroup_subsys
*ss
= subsys
[i
];
3931 if (ss
->root
== root
)
3932 mutex_lock(&ss
->hierarchy_mutex
);
3936 static void cgroup_unlock_hierarchy(struct cgroupfs_root
*root
)
3940 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3941 struct cgroup_subsys
*ss
= subsys
[i
];
3944 if (ss
->root
== root
)
3945 mutex_unlock(&ss
->hierarchy_mutex
);
3950 * cgroup_create - create a cgroup
3951 * @parent: cgroup that will be parent of the new cgroup
3952 * @dentry: dentry of the new cgroup
3953 * @mode: mode to set on new inode
3955 * Must be called with the mutex on the parent inode held
3957 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
3960 struct cgroup
*cgrp
;
3961 struct cgroupfs_root
*root
= parent
->root
;
3963 struct cgroup_subsys
*ss
;
3964 struct super_block
*sb
= root
->sb
;
3966 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3970 /* Grab a reference on the superblock so the hierarchy doesn't
3971 * get deleted on unmount if there are child cgroups. This
3972 * can be done outside cgroup_mutex, since the sb can't
3973 * disappear while someone has an open control file on the
3975 atomic_inc(&sb
->s_active
);
3977 mutex_lock(&cgroup_mutex
);
3979 init_cgroup_housekeeping(cgrp
);
3981 cgrp
->parent
= parent
;
3982 cgrp
->root
= parent
->root
;
3983 cgrp
->top_cgroup
= parent
->top_cgroup
;
3985 if (notify_on_release(parent
))
3986 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3988 if (clone_children(parent
))
3989 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3991 for_each_subsys(root
, ss
) {
3992 struct cgroup_subsys_state
*css
= ss
->create(cgrp
);
3998 init_cgroup_css(css
, ss
, cgrp
);
4000 err
= alloc_css_id(ss
, parent
, cgrp
);
4004 /* At error, ->destroy() callback has to free assigned ID. */
4005 if (clone_children(parent
) && ss
->post_clone
)
4006 ss
->post_clone(cgrp
);
4009 cgroup_lock_hierarchy(root
);
4010 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
4011 cgroup_unlock_hierarchy(root
);
4012 root
->number_of_cgroups
++;
4014 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
4018 /* If !clear_css_refs, each css holds a ref to the cgroup's dentry */
4019 for_each_subsys(root
, ss
)
4020 if (!ss
->__DEPRECATED_clear_css_refs
)
4023 /* The cgroup directory was pre-locked for us */
4024 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
4026 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
4028 err
= cgroup_populate_dir(cgrp
);
4029 /* If err < 0, we have a half-filled directory - oh well ;) */
4031 mutex_unlock(&cgroup_mutex
);
4032 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4038 cgroup_lock_hierarchy(root
);
4039 list_del(&cgrp
->sibling
);
4040 cgroup_unlock_hierarchy(root
);
4041 root
->number_of_cgroups
--;
4045 for_each_subsys(root
, ss
) {
4046 if (cgrp
->subsys
[ss
->subsys_id
])
4050 mutex_unlock(&cgroup_mutex
);
4052 /* Release the reference count that we took on the superblock */
4053 deactivate_super(sb
);
4059 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4061 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4063 /* the vfs holds inode->i_mutex already */
4064 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4068 * Check the reference count on each subsystem. Since we already
4069 * established that there are no tasks in the cgroup, if the css refcount
4070 * is also 1, then there should be no outstanding references, so the
4071 * subsystem is safe to destroy. We scan across all subsystems rather than
4072 * using the per-hierarchy linked list of mounted subsystems since we can
4073 * be called via check_for_release() with no synchronization other than
4074 * RCU, and the subsystem linked list isn't RCU-safe.
4076 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
4081 * We won't need to lock the subsys array, because the subsystems
4082 * we're concerned about aren't going anywhere since our cgroup root
4083 * has a reference on them.
4085 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4086 struct cgroup_subsys
*ss
= subsys
[i
];
4087 struct cgroup_subsys_state
*css
;
4089 /* Skip subsystems not present or not in this hierarchy */
4090 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
4093 css
= cgrp
->subsys
[ss
->subsys_id
];
4095 * When called from check_for_release() it's possible
4096 * that by this point the cgroup has been removed
4097 * and the css deleted. But a false-positive doesn't
4098 * matter, since it can only happen if the cgroup
4099 * has been deleted and hence no longer needs the
4100 * release agent to be called anyway.
4102 if (css
&& css_refcnt(css
) > 1)
4109 * Atomically mark all (or else none) of the cgroup's CSS objects as
4110 * CSS_REMOVED. Return true on success, or false if the cgroup has
4111 * busy subsystems. Call with cgroup_mutex held
4113 * Depending on whether a subsys has __DEPRECATED_clear_css_refs set or
4114 * not, cgroup removal behaves differently.
4116 * If clear is set, css refcnt for the subsystem should be zero before
4117 * cgroup removal can be committed. This is implemented by
4118 * CGRP_WAIT_ON_RMDIR and retry logic around ->pre_destroy(), which may be
4119 * called multiple times until all css refcnts reach zero and is allowed to
4120 * veto removal on any invocation. This behavior is deprecated and will be
4121 * removed as soon as the existing user (memcg) is updated.
4123 * If clear is not set, each css holds an extra reference to the cgroup's
4124 * dentry and cgroup removal proceeds regardless of css refs.
4125 * ->pre_destroy() will be called at least once and is not allowed to fail.
4126 * On the last put of each css, whenever that may be, the extra dentry ref
4127 * is put so that dentry destruction happens only after all css's are
4130 static int cgroup_clear_css_refs(struct cgroup
*cgrp
)
4132 struct cgroup_subsys
*ss
;
4133 unsigned long flags
;
4134 bool failed
= false;
4136 local_irq_save(flags
);
4139 * Block new css_tryget() by deactivating refcnt. If all refcnts
4140 * for subsystems w/ clear_css_refs set were 1 at the moment of
4141 * deactivation, we succeeded.
4143 for_each_subsys(cgrp
->root
, ss
) {
4144 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4146 WARN_ON(atomic_read(&css
->refcnt
) < 0);
4147 atomic_add(CSS_DEACT_BIAS
, &css
->refcnt
);
4149 if (ss
->__DEPRECATED_clear_css_refs
)
4150 failed
|= css_refcnt(css
) != 1;
4154 * If succeeded, set REMOVED and put all the base refs; otherwise,
4155 * restore refcnts to positive values. Either way, all in-progress
4156 * css_tryget() will be released.
4158 for_each_subsys(cgrp
->root
, ss
) {
4159 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4162 set_bit(CSS_REMOVED
, &css
->flags
);
4165 atomic_sub(CSS_DEACT_BIAS
, &css
->refcnt
);
4169 local_irq_restore(flags
);
4173 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4175 struct cgroup
*cgrp
= dentry
->d_fsdata
;
4177 struct cgroup
*parent
;
4179 struct cgroup_event
*event
, *tmp
;
4182 /* the vfs holds both inode->i_mutex already */
4184 mutex_lock(&cgroup_mutex
);
4185 if (atomic_read(&cgrp
->count
) != 0) {
4186 mutex_unlock(&cgroup_mutex
);
4189 if (!list_empty(&cgrp
->children
)) {
4190 mutex_unlock(&cgroup_mutex
);
4193 mutex_unlock(&cgroup_mutex
);
4196 * In general, subsystem has no css->refcnt after pre_destroy(). But
4197 * in racy cases, subsystem may have to get css->refcnt after
4198 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4199 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4200 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4201 * and subsystem's reference count handling. Please see css_get/put
4202 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4204 set_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4207 * Call pre_destroy handlers of subsys. Notify subsystems
4208 * that rmdir() request comes.
4210 ret
= cgroup_call_pre_destroy(cgrp
);
4212 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4216 mutex_lock(&cgroup_mutex
);
4217 parent
= cgrp
->parent
;
4218 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
)) {
4219 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4220 mutex_unlock(&cgroup_mutex
);
4223 prepare_to_wait(&cgroup_rmdir_waitq
, &wait
, TASK_INTERRUPTIBLE
);
4224 if (!cgroup_clear_css_refs(cgrp
)) {
4225 mutex_unlock(&cgroup_mutex
);
4227 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4228 * prepare_to_wait(), we need to check this flag.
4230 if (test_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
))
4232 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4233 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4234 if (signal_pending(current
))
4238 /* NO css_tryget() can success after here. */
4239 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4240 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4242 raw_spin_lock(&release_list_lock
);
4243 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
4244 if (!list_empty(&cgrp
->release_list
))
4245 list_del_init(&cgrp
->release_list
);
4246 raw_spin_unlock(&release_list_lock
);
4248 cgroup_lock_hierarchy(cgrp
->root
);
4249 /* delete this cgroup from parent->children */
4250 list_del_init(&cgrp
->sibling
);
4251 cgroup_unlock_hierarchy(cgrp
->root
);
4253 list_del_init(&cgrp
->allcg_node
);
4255 d
= dget(cgrp
->dentry
);
4257 cgroup_d_remove_dir(d
);
4260 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4261 check_for_release(parent
);
4264 * Unregister events and notify userspace.
4265 * Notify userspace about cgroup removing only after rmdir of cgroup
4266 * directory to avoid race between userspace and kernelspace
4268 spin_lock(&cgrp
->event_list_lock
);
4269 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4270 list_del(&event
->list
);
4271 remove_wait_queue(event
->wqh
, &event
->wait
);
4272 eventfd_signal(event
->eventfd
, 1);
4273 schedule_work(&event
->remove
);
4275 spin_unlock(&cgrp
->event_list_lock
);
4277 mutex_unlock(&cgroup_mutex
);
4281 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4283 INIT_LIST_HEAD(&ss
->cftsets
);
4286 * base_cftset is embedded in subsys itself, no need to worry about
4289 if (ss
->base_cftypes
) {
4290 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4291 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4295 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4297 struct cgroup_subsys_state
*css
;
4299 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4301 /* init base cftset */
4302 cgroup_init_cftsets(ss
);
4304 /* Create the top cgroup state for this subsystem */
4305 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4306 ss
->root
= &rootnode
;
4307 css
= ss
->create(dummytop
);
4308 /* We don't handle early failures gracefully */
4309 BUG_ON(IS_ERR(css
));
4310 init_cgroup_css(css
, ss
, dummytop
);
4312 /* Update the init_css_set to contain a subsys
4313 * pointer to this state - since the subsystem is
4314 * newly registered, all tasks and hence the
4315 * init_css_set is in the subsystem's top cgroup. */
4316 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
4318 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4320 /* At system boot, before all subsystems have been
4321 * registered, no tasks have been forked, so we don't
4322 * need to invoke fork callbacks here. */
4323 BUG_ON(!list_empty(&init_task
.tasks
));
4325 mutex_init(&ss
->hierarchy_mutex
);
4326 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4329 /* this function shouldn't be used with modular subsystems, since they
4330 * need to register a subsys_id, among other things */
4335 * cgroup_load_subsys: load and register a modular subsystem at runtime
4336 * @ss: the subsystem to load
4338 * This function should be called in a modular subsystem's initcall. If the
4339 * subsystem is built as a module, it will be assigned a new subsys_id and set
4340 * up for use. If the subsystem is built-in anyway, work is delegated to the
4341 * simpler cgroup_init_subsys.
4343 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4346 struct cgroup_subsys_state
*css
;
4348 /* check name and function validity */
4349 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4350 ss
->create
== NULL
|| ss
->destroy
== NULL
)
4354 * we don't support callbacks in modular subsystems. this check is
4355 * before the ss->module check for consistency; a subsystem that could
4356 * be a module should still have no callbacks even if the user isn't
4357 * compiling it as one.
4359 if (ss
->fork
|| ss
->exit
)
4363 * an optionally modular subsystem is built-in: we want to do nothing,
4364 * since cgroup_init_subsys will have already taken care of it.
4366 if (ss
->module
== NULL
) {
4367 /* a few sanity checks */
4368 BUG_ON(ss
->subsys_id
>= CGROUP_BUILTIN_SUBSYS_COUNT
);
4369 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4373 /* init base cftset */
4374 cgroup_init_cftsets(ss
);
4377 * need to register a subsys id before anything else - for example,
4378 * init_cgroup_css needs it.
4380 mutex_lock(&cgroup_mutex
);
4381 /* find the first empty slot in the array */
4382 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4383 if (subsys
[i
] == NULL
)
4386 if (i
== CGROUP_SUBSYS_COUNT
) {
4387 /* maximum number of subsystems already registered! */
4388 mutex_unlock(&cgroup_mutex
);
4391 /* assign ourselves the subsys_id */
4396 * no ss->create seems to need anything important in the ss struct, so
4397 * this can happen first (i.e. before the rootnode attachment).
4399 css
= ss
->create(dummytop
);
4401 /* failure case - need to deassign the subsys[] slot. */
4403 mutex_unlock(&cgroup_mutex
);
4404 return PTR_ERR(css
);
4407 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4408 ss
->root
= &rootnode
;
4410 /* our new subsystem will be attached to the dummy hierarchy. */
4411 init_cgroup_css(css
, ss
, dummytop
);
4412 /* init_idr must be after init_cgroup_css because it sets css->id. */
4414 int ret
= cgroup_init_idr(ss
, css
);
4416 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4417 ss
->destroy(dummytop
);
4419 mutex_unlock(&cgroup_mutex
);
4425 * Now we need to entangle the css into the existing css_sets. unlike
4426 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4427 * will need a new pointer to it; done by iterating the css_set_table.
4428 * furthermore, modifying the existing css_sets will corrupt the hash
4429 * table state, so each changed css_set will need its hash recomputed.
4430 * this is all done under the css_set_lock.
4432 write_lock(&css_set_lock
);
4433 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
4435 struct hlist_node
*node
, *tmp
;
4436 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
4438 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
4439 /* skip entries that we already rehashed */
4440 if (cg
->subsys
[ss
->subsys_id
])
4442 /* remove existing entry */
4443 hlist_del(&cg
->hlist
);
4445 cg
->subsys
[ss
->subsys_id
] = css
;
4446 /* recompute hash and restore entry */
4447 new_bucket
= css_set_hash(cg
->subsys
);
4448 hlist_add_head(&cg
->hlist
, new_bucket
);
4451 write_unlock(&css_set_lock
);
4453 mutex_init(&ss
->hierarchy_mutex
);
4454 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4458 mutex_unlock(&cgroup_mutex
);
4461 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4464 * cgroup_unload_subsys: unload a modular subsystem
4465 * @ss: the subsystem to unload
4467 * This function should be called in a modular subsystem's exitcall. When this
4468 * function is invoked, the refcount on the subsystem's module will be 0, so
4469 * the subsystem will not be attached to any hierarchy.
4471 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4473 struct cg_cgroup_link
*link
;
4474 struct hlist_head
*hhead
;
4476 BUG_ON(ss
->module
== NULL
);
4479 * we shouldn't be called if the subsystem is in use, and the use of
4480 * try_module_get in parse_cgroupfs_options should ensure that it
4481 * doesn't start being used while we're killing it off.
4483 BUG_ON(ss
->root
!= &rootnode
);
4485 mutex_lock(&cgroup_mutex
);
4486 /* deassign the subsys_id */
4487 BUG_ON(ss
->subsys_id
< CGROUP_BUILTIN_SUBSYS_COUNT
);
4488 subsys
[ss
->subsys_id
] = NULL
;
4490 /* remove subsystem from rootnode's list of subsystems */
4491 list_del_init(&ss
->sibling
);
4494 * disentangle the css from all css_sets attached to the dummytop. as
4495 * in loading, we need to pay our respects to the hashtable gods.
4497 write_lock(&css_set_lock
);
4498 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4499 struct css_set
*cg
= link
->cg
;
4501 hlist_del(&cg
->hlist
);
4502 BUG_ON(!cg
->subsys
[ss
->subsys_id
]);
4503 cg
->subsys
[ss
->subsys_id
] = NULL
;
4504 hhead
= css_set_hash(cg
->subsys
);
4505 hlist_add_head(&cg
->hlist
, hhead
);
4507 write_unlock(&css_set_lock
);
4510 * remove subsystem's css from the dummytop and free it - need to free
4511 * before marking as null because ss->destroy needs the cgrp->subsys
4512 * pointer to find their state. note that this also takes care of
4513 * freeing the css_id.
4515 ss
->destroy(dummytop
);
4516 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4518 mutex_unlock(&cgroup_mutex
);
4520 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4523 * cgroup_init_early - cgroup initialization at system boot
4525 * Initialize cgroups at system boot, and initialize any
4526 * subsystems that request early init.
4528 int __init
cgroup_init_early(void)
4531 atomic_set(&init_css_set
.refcount
, 1);
4532 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4533 INIT_LIST_HEAD(&init_css_set
.tasks
);
4534 INIT_HLIST_NODE(&init_css_set
.hlist
);
4536 init_cgroup_root(&rootnode
);
4538 init_task
.cgroups
= &init_css_set
;
4540 init_css_set_link
.cg
= &init_css_set
;
4541 init_css_set_link
.cgrp
= dummytop
;
4542 list_add(&init_css_set_link
.cgrp_link_list
,
4543 &rootnode
.top_cgroup
.css_sets
);
4544 list_add(&init_css_set_link
.cg_link_list
,
4545 &init_css_set
.cg_links
);
4547 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
4548 INIT_HLIST_HEAD(&css_set_table
[i
]);
4550 /* at bootup time, we don't worry about modular subsystems */
4551 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4552 struct cgroup_subsys
*ss
= subsys
[i
];
4555 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4556 BUG_ON(!ss
->create
);
4557 BUG_ON(!ss
->destroy
);
4558 if (ss
->subsys_id
!= i
) {
4559 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4560 ss
->name
, ss
->subsys_id
);
4565 cgroup_init_subsys(ss
);
4571 * cgroup_init - cgroup initialization
4573 * Register cgroup filesystem and /proc file, and initialize
4574 * any subsystems that didn't request early init.
4576 int __init
cgroup_init(void)
4580 struct hlist_head
*hhead
;
4582 err
= bdi_init(&cgroup_backing_dev_info
);
4586 /* at bootup time, we don't worry about modular subsystems */
4587 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4588 struct cgroup_subsys
*ss
= subsys
[i
];
4589 if (!ss
->early_init
)
4590 cgroup_init_subsys(ss
);
4592 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4595 /* Add init_css_set to the hash table */
4596 hhead
= css_set_hash(init_css_set
.subsys
);
4597 hlist_add_head(&init_css_set
.hlist
, hhead
);
4598 BUG_ON(!init_root_id(&rootnode
));
4600 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4606 err
= register_filesystem(&cgroup_fs_type
);
4608 kobject_put(cgroup_kobj
);
4612 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4616 bdi_destroy(&cgroup_backing_dev_info
);
4622 * proc_cgroup_show()
4623 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4624 * - Used for /proc/<pid>/cgroup.
4625 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4626 * doesn't really matter if tsk->cgroup changes after we read it,
4627 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4628 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4629 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4630 * cgroup to top_cgroup.
4633 /* TODO: Use a proper seq_file iterator */
4634 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4637 struct task_struct
*tsk
;
4640 struct cgroupfs_root
*root
;
4643 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4649 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4655 mutex_lock(&cgroup_mutex
);
4657 for_each_active_root(root
) {
4658 struct cgroup_subsys
*ss
;
4659 struct cgroup
*cgrp
;
4662 seq_printf(m
, "%d:", root
->hierarchy_id
);
4663 for_each_subsys(root
, ss
)
4664 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4665 if (strlen(root
->name
))
4666 seq_printf(m
, "%sname=%s", count
? "," : "",
4669 cgrp
= task_cgroup_from_root(tsk
, root
);
4670 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4678 mutex_unlock(&cgroup_mutex
);
4679 put_task_struct(tsk
);
4686 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4688 struct pid
*pid
= PROC_I(inode
)->pid
;
4689 return single_open(file
, proc_cgroup_show
, pid
);
4692 const struct file_operations proc_cgroup_operations
= {
4693 .open
= cgroup_open
,
4695 .llseek
= seq_lseek
,
4696 .release
= single_release
,
4699 /* Display information about each subsystem and each hierarchy */
4700 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4704 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4706 * ideally we don't want subsystems moving around while we do this.
4707 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4708 * subsys/hierarchy state.
4710 mutex_lock(&cgroup_mutex
);
4711 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4712 struct cgroup_subsys
*ss
= subsys
[i
];
4715 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4716 ss
->name
, ss
->root
->hierarchy_id
,
4717 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4719 mutex_unlock(&cgroup_mutex
);
4723 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4725 return single_open(file
, proc_cgroupstats_show
, NULL
);
4728 static const struct file_operations proc_cgroupstats_operations
= {
4729 .open
= cgroupstats_open
,
4731 .llseek
= seq_lseek
,
4732 .release
= single_release
,
4736 * cgroup_fork - attach newly forked task to its parents cgroup.
4737 * @child: pointer to task_struct of forking parent process.
4739 * Description: A task inherits its parent's cgroup at fork().
4741 * A pointer to the shared css_set was automatically copied in
4742 * fork.c by dup_task_struct(). However, we ignore that copy, since
4743 * it was not made under the protection of RCU, cgroup_mutex or
4744 * threadgroup_change_begin(), so it might no longer be a valid
4745 * cgroup pointer. cgroup_attach_task() might have already changed
4746 * current->cgroups, allowing the previously referenced cgroup
4747 * group to be removed and freed.
4749 * Outside the pointer validity we also need to process the css_set
4750 * inheritance between threadgoup_change_begin() and
4751 * threadgoup_change_end(), this way there is no leak in any process
4752 * wide migration performed by cgroup_attach_proc() that could otherwise
4753 * miss a thread because it is too early or too late in the fork stage.
4755 * At the point that cgroup_fork() is called, 'current' is the parent
4756 * task, and the passed argument 'child' points to the child task.
4758 void cgroup_fork(struct task_struct
*child
)
4761 * We don't need to task_lock() current because current->cgroups
4762 * can't be changed concurrently here. The parent obviously hasn't
4763 * exited and called cgroup_exit(), and we are synchronized against
4764 * cgroup migration through threadgroup_change_begin().
4766 child
->cgroups
= current
->cgroups
;
4767 get_css_set(child
->cgroups
);
4768 INIT_LIST_HEAD(&child
->cg_list
);
4772 * cgroup_fork_callbacks - run fork callbacks
4773 * @child: the new task
4775 * Called on a new task very soon before adding it to the
4776 * tasklist. No need to take any locks since no-one can
4777 * be operating on this task.
4779 void cgroup_fork_callbacks(struct task_struct
*child
)
4781 if (need_forkexit_callback
) {
4784 * forkexit callbacks are only supported for builtin
4785 * subsystems, and the builtin section of the subsys array is
4786 * immutable, so we don't need to lock the subsys array here.
4788 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4789 struct cgroup_subsys
*ss
= subsys
[i
];
4797 * cgroup_post_fork - called on a new task after adding it to the task list
4798 * @child: the task in question
4800 * Adds the task to the list running through its css_set if necessary.
4801 * Has to be after the task is visible on the task list in case we race
4802 * with the first call to cgroup_iter_start() - to guarantee that the
4803 * new task ends up on its list.
4805 void cgroup_post_fork(struct task_struct
*child
)
4808 * use_task_css_set_links is set to 1 before we walk the tasklist
4809 * under the tasklist_lock and we read it here after we added the child
4810 * to the tasklist under the tasklist_lock as well. If the child wasn't
4811 * yet in the tasklist when we walked through it from
4812 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4813 * should be visible now due to the paired locking and barriers implied
4814 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4815 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4818 if (use_task_css_set_links
) {
4819 write_lock(&css_set_lock
);
4820 if (list_empty(&child
->cg_list
)) {
4822 * It's safe to use child->cgroups without task_lock()
4823 * here because we are protected through
4824 * threadgroup_change_begin() against concurrent
4825 * css_set change in cgroup_task_migrate(). Also
4826 * the task can't exit at that point until
4827 * wake_up_new_task() is called, so we are protected
4828 * against cgroup_exit() setting child->cgroup to
4831 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4833 write_unlock(&css_set_lock
);
4837 * cgroup_exit - detach cgroup from exiting task
4838 * @tsk: pointer to task_struct of exiting process
4839 * @run_callback: run exit callbacks?
4841 * Description: Detach cgroup from @tsk and release it.
4843 * Note that cgroups marked notify_on_release force every task in
4844 * them to take the global cgroup_mutex mutex when exiting.
4845 * This could impact scaling on very large systems. Be reluctant to
4846 * use notify_on_release cgroups where very high task exit scaling
4847 * is required on large systems.
4849 * the_top_cgroup_hack:
4851 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4853 * We call cgroup_exit() while the task is still competent to
4854 * handle notify_on_release(), then leave the task attached to the
4855 * root cgroup in each hierarchy for the remainder of its exit.
4857 * To do this properly, we would increment the reference count on
4858 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4859 * code we would add a second cgroup function call, to drop that
4860 * reference. This would just create an unnecessary hot spot on
4861 * the top_cgroup reference count, to no avail.
4863 * Normally, holding a reference to a cgroup without bumping its
4864 * count is unsafe. The cgroup could go away, or someone could
4865 * attach us to a different cgroup, decrementing the count on
4866 * the first cgroup that we never incremented. But in this case,
4867 * top_cgroup isn't going away, and either task has PF_EXITING set,
4868 * which wards off any cgroup_attach_task() attempts, or task is a failed
4869 * fork, never visible to cgroup_attach_task.
4871 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4877 * Unlink from the css_set task list if necessary.
4878 * Optimistically check cg_list before taking
4881 if (!list_empty(&tsk
->cg_list
)) {
4882 write_lock(&css_set_lock
);
4883 if (!list_empty(&tsk
->cg_list
))
4884 list_del_init(&tsk
->cg_list
);
4885 write_unlock(&css_set_lock
);
4888 /* Reassign the task to the init_css_set. */
4891 tsk
->cgroups
= &init_css_set
;
4893 if (run_callbacks
&& need_forkexit_callback
) {
4895 * modular subsystems can't use callbacks, so no need to lock
4898 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4899 struct cgroup_subsys
*ss
= subsys
[i
];
4901 struct cgroup
*old_cgrp
=
4902 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4903 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4904 ss
->exit(cgrp
, old_cgrp
, tsk
);
4911 put_css_set_taskexit(cg
);
4915 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4916 * @cgrp: the cgroup in question
4917 * @task: the task in question
4919 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4922 * If we are sending in dummytop, then presumably we are creating
4923 * the top cgroup in the subsystem.
4925 * Called only by the ns (nsproxy) cgroup.
4927 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
4930 struct cgroup
*target
;
4932 if (cgrp
== dummytop
)
4935 target
= task_cgroup_from_root(task
, cgrp
->root
);
4936 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
4937 cgrp
= cgrp
->parent
;
4938 ret
= (cgrp
== target
);
4942 static void check_for_release(struct cgroup
*cgrp
)
4944 /* All of these checks rely on RCU to keep the cgroup
4945 * structure alive */
4946 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
4947 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
4948 /* Control Group is currently removeable. If it's not
4949 * already queued for a userspace notification, queue
4951 int need_schedule_work
= 0;
4952 raw_spin_lock(&release_list_lock
);
4953 if (!cgroup_is_removed(cgrp
) &&
4954 list_empty(&cgrp
->release_list
)) {
4955 list_add(&cgrp
->release_list
, &release_list
);
4956 need_schedule_work
= 1;
4958 raw_spin_unlock(&release_list_lock
);
4959 if (need_schedule_work
)
4960 schedule_work(&release_agent_work
);
4964 /* Caller must verify that the css is not for root cgroup */
4965 bool __css_tryget(struct cgroup_subsys_state
*css
)
4968 int v
= css_refcnt(css
);
4970 if (atomic_cmpxchg(&css
->refcnt
, v
, v
+ 1) == v
)
4973 } while (!test_bit(CSS_REMOVED
, &css
->flags
));
4977 EXPORT_SYMBOL_GPL(__css_tryget
);
4979 /* Caller must verify that the css is not for root cgroup */
4980 void __css_put(struct cgroup_subsys_state
*css
)
4982 struct cgroup
*cgrp
= css
->cgroup
;
4986 v
= css_unbias_refcnt(atomic_dec_return(&css
->refcnt
));
4990 if (notify_on_release(cgrp
)) {
4991 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4992 check_for_release(cgrp
);
4994 cgroup_wakeup_rmdir_waiter(cgrp
);
4997 if (!test_bit(CSS_CLEAR_CSS_REFS
, &css
->flags
))
4998 schedule_work(&css
->dput_work
);
5003 EXPORT_SYMBOL_GPL(__css_put
);
5006 * Notify userspace when a cgroup is released, by running the
5007 * configured release agent with the name of the cgroup (path
5008 * relative to the root of cgroup file system) as the argument.
5010 * Most likely, this user command will try to rmdir this cgroup.
5012 * This races with the possibility that some other task will be
5013 * attached to this cgroup before it is removed, or that some other
5014 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
5015 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
5016 * unused, and this cgroup will be reprieved from its death sentence,
5017 * to continue to serve a useful existence. Next time it's released,
5018 * we will get notified again, if it still has 'notify_on_release' set.
5020 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5021 * means only wait until the task is successfully execve()'d. The
5022 * separate release agent task is forked by call_usermodehelper(),
5023 * then control in this thread returns here, without waiting for the
5024 * release agent task. We don't bother to wait because the caller of
5025 * this routine has no use for the exit status of the release agent
5026 * task, so no sense holding our caller up for that.
5028 static void cgroup_release_agent(struct work_struct
*work
)
5030 BUG_ON(work
!= &release_agent_work
);
5031 mutex_lock(&cgroup_mutex
);
5032 raw_spin_lock(&release_list_lock
);
5033 while (!list_empty(&release_list
)) {
5034 char *argv
[3], *envp
[3];
5036 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5037 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5040 list_del_init(&cgrp
->release_list
);
5041 raw_spin_unlock(&release_list_lock
);
5042 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5045 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5047 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5052 argv
[i
++] = agentbuf
;
5053 argv
[i
++] = pathbuf
;
5057 /* minimal command environment */
5058 envp
[i
++] = "HOME=/";
5059 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5062 /* Drop the lock while we invoke the usermode helper,
5063 * since the exec could involve hitting disk and hence
5064 * be a slow process */
5065 mutex_unlock(&cgroup_mutex
);
5066 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5067 mutex_lock(&cgroup_mutex
);
5071 raw_spin_lock(&release_list_lock
);
5073 raw_spin_unlock(&release_list_lock
);
5074 mutex_unlock(&cgroup_mutex
);
5077 static int __init
cgroup_disable(char *str
)
5082 while ((token
= strsep(&str
, ",")) != NULL
) {
5086 * cgroup_disable, being at boot time, can't know about module
5087 * subsystems, so we don't worry about them.
5089 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
5090 struct cgroup_subsys
*ss
= subsys
[i
];
5092 if (!strcmp(token
, ss
->name
)) {
5094 printk(KERN_INFO
"Disabling %s control group"
5095 " subsystem\n", ss
->name
);
5102 __setup("cgroup_disable=", cgroup_disable
);
5105 * Functons for CSS ID.
5109 *To get ID other than 0, this should be called when !cgroup_is_removed().
5111 unsigned short css_id(struct cgroup_subsys_state
*css
)
5113 struct css_id
*cssid
;
5116 * This css_id() can return correct value when somone has refcnt
5117 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5118 * it's unchanged until freed.
5120 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5126 EXPORT_SYMBOL_GPL(css_id
);
5128 unsigned short css_depth(struct cgroup_subsys_state
*css
)
5130 struct css_id
*cssid
;
5132 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5135 return cssid
->depth
;
5138 EXPORT_SYMBOL_GPL(css_depth
);
5141 * css_is_ancestor - test "root" css is an ancestor of "child"
5142 * @child: the css to be tested.
5143 * @root: the css supporsed to be an ancestor of the child.
5145 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5146 * this function reads css->id, the caller must hold rcu_read_lock().
5147 * But, considering usual usage, the csses should be valid objects after test.
5148 * Assuming that the caller will do some action to the child if this returns
5149 * returns true, the caller must take "child";s reference count.
5150 * If "child" is valid object and this returns true, "root" is valid, too.
5153 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5154 const struct cgroup_subsys_state
*root
)
5156 struct css_id
*child_id
;
5157 struct css_id
*root_id
;
5159 child_id
= rcu_dereference(child
->id
);
5162 root_id
= rcu_dereference(root
->id
);
5165 if (child_id
->depth
< root_id
->depth
)
5167 if (child_id
->stack
[root_id
->depth
] != root_id
->id
)
5172 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5174 struct css_id
*id
= css
->id
;
5175 /* When this is called before css_id initialization, id can be NULL */
5179 BUG_ON(!ss
->use_id
);
5181 rcu_assign_pointer(id
->css
, NULL
);
5182 rcu_assign_pointer(css
->id
, NULL
);
5183 spin_lock(&ss
->id_lock
);
5184 idr_remove(&ss
->idr
, id
->id
);
5185 spin_unlock(&ss
->id_lock
);
5186 kfree_rcu(id
, rcu_head
);
5188 EXPORT_SYMBOL_GPL(free_css_id
);
5191 * This is called by init or create(). Then, calls to this function are
5192 * always serialized (By cgroup_mutex() at create()).
5195 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5197 struct css_id
*newid
;
5198 int myid
, error
, size
;
5200 BUG_ON(!ss
->use_id
);
5202 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5203 newid
= kzalloc(size
, GFP_KERNEL
);
5205 return ERR_PTR(-ENOMEM
);
5207 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
5211 spin_lock(&ss
->id_lock
);
5212 /* Don't use 0. allocates an ID of 1-65535 */
5213 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
5214 spin_unlock(&ss
->id_lock
);
5216 /* Returns error when there are no free spaces for new ID.*/
5221 if (myid
> CSS_ID_MAX
)
5225 newid
->depth
= depth
;
5229 spin_lock(&ss
->id_lock
);
5230 idr_remove(&ss
->idr
, myid
);
5231 spin_unlock(&ss
->id_lock
);
5234 return ERR_PTR(error
);
5238 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5239 struct cgroup_subsys_state
*rootcss
)
5241 struct css_id
*newid
;
5243 spin_lock_init(&ss
->id_lock
);
5246 newid
= get_new_cssid(ss
, 0);
5248 return PTR_ERR(newid
);
5250 newid
->stack
[0] = newid
->id
;
5251 newid
->css
= rootcss
;
5252 rootcss
->id
= newid
;
5256 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5257 struct cgroup
*child
)
5259 int subsys_id
, i
, depth
= 0;
5260 struct cgroup_subsys_state
*parent_css
, *child_css
;
5261 struct css_id
*child_id
, *parent_id
;
5263 subsys_id
= ss
->subsys_id
;
5264 parent_css
= parent
->subsys
[subsys_id
];
5265 child_css
= child
->subsys
[subsys_id
];
5266 parent_id
= parent_css
->id
;
5267 depth
= parent_id
->depth
+ 1;
5269 child_id
= get_new_cssid(ss
, depth
);
5270 if (IS_ERR(child_id
))
5271 return PTR_ERR(child_id
);
5273 for (i
= 0; i
< depth
; i
++)
5274 child_id
->stack
[i
] = parent_id
->stack
[i
];
5275 child_id
->stack
[depth
] = child_id
->id
;
5277 * child_id->css pointer will be set after this cgroup is available
5278 * see cgroup_populate_dir()
5280 rcu_assign_pointer(child_css
->id
, child_id
);
5286 * css_lookup - lookup css by id
5287 * @ss: cgroup subsys to be looked into.
5290 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5291 * NULL if not. Should be called under rcu_read_lock()
5293 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5295 struct css_id
*cssid
= NULL
;
5297 BUG_ON(!ss
->use_id
);
5298 cssid
= idr_find(&ss
->idr
, id
);
5300 if (unlikely(!cssid
))
5303 return rcu_dereference(cssid
->css
);
5305 EXPORT_SYMBOL_GPL(css_lookup
);
5308 * css_get_next - lookup next cgroup under specified hierarchy.
5309 * @ss: pointer to subsystem
5310 * @id: current position of iteration.
5311 * @root: pointer to css. search tree under this.
5312 * @foundid: position of found object.
5314 * Search next css under the specified hierarchy of rootid. Calling under
5315 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5317 struct cgroup_subsys_state
*
5318 css_get_next(struct cgroup_subsys
*ss
, int id
,
5319 struct cgroup_subsys_state
*root
, int *foundid
)
5321 struct cgroup_subsys_state
*ret
= NULL
;
5324 int rootid
= css_id(root
);
5325 int depth
= css_depth(root
);
5330 BUG_ON(!ss
->use_id
);
5331 WARN_ON_ONCE(!rcu_read_lock_held());
5333 /* fill start point for scan */
5337 * scan next entry from bitmap(tree), tmpid is updated after
5340 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
5343 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
5344 ret
= rcu_dereference(tmp
->css
);
5350 /* continue to scan from next id */
5357 * get corresponding css from file open on cgroupfs directory
5359 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5361 struct cgroup
*cgrp
;
5362 struct inode
*inode
;
5363 struct cgroup_subsys_state
*css
;
5365 inode
= f
->f_dentry
->d_inode
;
5366 /* check in cgroup filesystem dir */
5367 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5368 return ERR_PTR(-EBADF
);
5370 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5371 return ERR_PTR(-EINVAL
);
5374 cgrp
= __d_cgrp(f
->f_dentry
);
5375 css
= cgrp
->subsys
[id
];
5376 return css
? css
: ERR_PTR(-ENOENT
);
5379 #ifdef CONFIG_CGROUP_DEBUG
5380 static struct cgroup_subsys_state
*debug_create(struct cgroup
*cont
)
5382 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5385 return ERR_PTR(-ENOMEM
);
5390 static void debug_destroy(struct cgroup
*cont
)
5392 kfree(cont
->subsys
[debug_subsys_id
]);
5395 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5397 return atomic_read(&cont
->count
);
5400 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5402 return cgroup_task_count(cont
);
5405 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5407 return (u64
)(unsigned long)current
->cgroups
;
5410 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5416 count
= atomic_read(¤t
->cgroups
->refcount
);
5421 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5423 struct seq_file
*seq
)
5425 struct cg_cgroup_link
*link
;
5428 read_lock(&css_set_lock
);
5430 cg
= rcu_dereference(current
->cgroups
);
5431 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5432 struct cgroup
*c
= link
->cgrp
;
5436 name
= c
->dentry
->d_name
.name
;
5439 seq_printf(seq
, "Root %d group %s\n",
5440 c
->root
->hierarchy_id
, name
);
5443 read_unlock(&css_set_lock
);
5447 #define MAX_TASKS_SHOWN_PER_CSS 25
5448 static int cgroup_css_links_read(struct cgroup
*cont
,
5450 struct seq_file
*seq
)
5452 struct cg_cgroup_link
*link
;
5454 read_lock(&css_set_lock
);
5455 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5456 struct css_set
*cg
= link
->cg
;
5457 struct task_struct
*task
;
5459 seq_printf(seq
, "css_set %p\n", cg
);
5460 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5461 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5462 seq_puts(seq
, " ...\n");
5465 seq_printf(seq
, " task %d\n",
5466 task_pid_vnr(task
));
5470 read_unlock(&css_set_lock
);
5474 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5476 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5479 static struct cftype debug_files
[] = {
5481 .name
= "cgroup_refcount",
5482 .read_u64
= cgroup_refcount_read
,
5485 .name
= "taskcount",
5486 .read_u64
= debug_taskcount_read
,
5490 .name
= "current_css_set",
5491 .read_u64
= current_css_set_read
,
5495 .name
= "current_css_set_refcount",
5496 .read_u64
= current_css_set_refcount_read
,
5500 .name
= "current_css_set_cg_links",
5501 .read_seq_string
= current_css_set_cg_links_read
,
5505 .name
= "cgroup_css_links",
5506 .read_seq_string
= cgroup_css_links_read
,
5510 .name
= "releasable",
5511 .read_u64
= releasable_read
,
5517 struct cgroup_subsys debug_subsys
= {
5519 .create
= debug_create
,
5520 .destroy
= debug_destroy
,
5521 .subsys_id
= debug_subsys_id
,
5522 .base_cftypes
= debug_files
,
5524 #endif /* CONFIG_CGROUP_DEBUG */