2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hash.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_proc */
64 #include <linux/atomic.h>
67 * cgroup_mutex is the master lock. Any modification to cgroup or its
68 * hierarchy must be performed while holding it.
70 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
71 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
72 * release_agent_path and so on. Modifying requires both cgroup_mutex and
73 * cgroup_root_mutex. Readers can acquire either of the two. This is to
74 * break the following locking order cycle.
76 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
77 * B. namespace_sem -> cgroup_mutex
79 * B happens only through cgroup_show_options() and using cgroup_root_mutex
82 static DEFINE_MUTEX(cgroup_mutex
);
83 static DEFINE_MUTEX(cgroup_root_mutex
);
86 * Generate an array of cgroup subsystem pointers. At boot time, this is
87 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
88 * registered after that. The mutable section of this array is protected by
91 #define SUBSYS(_x) &_x ## _subsys,
92 static struct cgroup_subsys
*subsys
[CGROUP_SUBSYS_COUNT
] = {
93 #include <linux/cgroup_subsys.h>
96 #define MAX_CGROUP_ROOT_NAMELEN 64
99 * A cgroupfs_root represents the root of a cgroup hierarchy,
100 * and may be associated with a superblock to form an active
103 struct cgroupfs_root
{
104 struct super_block
*sb
;
107 * The bitmask of subsystems intended to be attached to this
110 unsigned long subsys_bits
;
112 /* Unique id for this hierarchy. */
115 /* The bitmask of subsystems currently attached to this hierarchy */
116 unsigned long actual_subsys_bits
;
118 /* A list running through the attached subsystems */
119 struct list_head subsys_list
;
121 /* The root cgroup for this hierarchy */
122 struct cgroup top_cgroup
;
124 /* Tracks how many cgroups are currently defined in hierarchy.*/
125 int number_of_cgroups
;
127 /* A list running through the active hierarchies */
128 struct list_head root_list
;
130 /* Hierarchy-specific flags */
133 /* The path to use for release notifications. */
134 char release_agent_path
[PATH_MAX
];
136 /* The name for this hierarchy - may be empty */
137 char name
[MAX_CGROUP_ROOT_NAMELEN
];
141 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
142 * subsystems that are otherwise unattached - it never has more than a
143 * single cgroup, and all tasks are part of that cgroup.
145 static struct cgroupfs_root rootnode
;
148 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
149 * cgroup_subsys->use_id != 0.
151 #define CSS_ID_MAX (65535)
154 * The css to which this ID points. This pointer is set to valid value
155 * after cgroup is populated. If cgroup is removed, this will be NULL.
156 * This pointer is expected to be RCU-safe because destroy()
157 * is called after synchronize_rcu(). But for safe use, css_is_removed()
158 * css_tryget() should be used for avoiding race.
160 struct cgroup_subsys_state __rcu
*css
;
166 * Depth in hierarchy which this ID belongs to.
168 unsigned short depth
;
170 * ID is freed by RCU. (and lookup routine is RCU safe.)
172 struct rcu_head rcu_head
;
174 * Hierarchy of CSS ID belongs to.
176 unsigned short stack
[0]; /* Array of Length (depth+1) */
180 * cgroup_event represents events which userspace want to receive.
182 struct cgroup_event
{
184 * Cgroup which the event belongs to.
188 * Control file which the event associated.
192 * eventfd to signal userspace about the event.
194 struct eventfd_ctx
*eventfd
;
196 * Each of these stored in a list by the cgroup.
198 struct list_head list
;
200 * All fields below needed to unregister event when
201 * userspace closes eventfd.
204 wait_queue_head_t
*wqh
;
206 struct work_struct remove
;
209 /* The list of hierarchy roots */
211 static LIST_HEAD(roots
);
212 static int root_count
;
214 static DEFINE_IDA(hierarchy_ida
);
215 static int next_hierarchy_id
;
216 static DEFINE_SPINLOCK(hierarchy_id_lock
);
218 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
219 #define dummytop (&rootnode.top_cgroup)
221 /* This flag indicates whether tasks in the fork and exit paths should
222 * check for fork/exit handlers to call. This avoids us having to do
223 * extra work in the fork/exit path if none of the subsystems need to
226 static int need_forkexit_callback __read_mostly
;
228 #ifdef CONFIG_PROVE_LOCKING
229 int cgroup_lock_is_held(void)
231 return lockdep_is_held(&cgroup_mutex
);
233 #else /* #ifdef CONFIG_PROVE_LOCKING */
234 int cgroup_lock_is_held(void)
236 return mutex_is_locked(&cgroup_mutex
);
238 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
240 EXPORT_SYMBOL_GPL(cgroup_lock_is_held
);
242 /* convenient tests for these bits */
243 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
245 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
248 /* bits in struct cgroupfs_root flags field */
250 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
253 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
256 (1 << CGRP_RELEASABLE
) |
257 (1 << CGRP_NOTIFY_ON_RELEASE
);
258 return (cgrp
->flags
& bits
) == bits
;
261 static int notify_on_release(const struct cgroup
*cgrp
)
263 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
266 static int clone_children(const struct cgroup
*cgrp
)
268 return test_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
272 * for_each_subsys() allows you to iterate on each subsystem attached to
273 * an active hierarchy
275 #define for_each_subsys(_root, _ss) \
276 list_for_each_entry(_ss, &_root->subsys_list, sibling)
278 /* for_each_active_root() allows you to iterate across the active hierarchies */
279 #define for_each_active_root(_root) \
280 list_for_each_entry(_root, &roots, root_list)
282 /* the list of cgroups eligible for automatic release. Protected by
283 * release_list_lock */
284 static LIST_HEAD(release_list
);
285 static DEFINE_RAW_SPINLOCK(release_list_lock
);
286 static void cgroup_release_agent(struct work_struct
*work
);
287 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
288 static void check_for_release(struct cgroup
*cgrp
);
290 /* Link structure for associating css_set objects with cgroups */
291 struct cg_cgroup_link
{
293 * List running through cg_cgroup_links associated with a
294 * cgroup, anchored on cgroup->css_sets
296 struct list_head cgrp_link_list
;
299 * List running through cg_cgroup_links pointing at a
300 * single css_set object, anchored on css_set->cg_links
302 struct list_head cg_link_list
;
306 /* The default css_set - used by init and its children prior to any
307 * hierarchies being mounted. It contains a pointer to the root state
308 * for each subsystem. Also used to anchor the list of css_sets. Not
309 * reference-counted, to improve performance when child cgroups
310 * haven't been created.
313 static struct css_set init_css_set
;
314 static struct cg_cgroup_link init_css_set_link
;
316 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
317 struct cgroup_subsys_state
*css
);
319 /* css_set_lock protects the list of css_set objects, and the
320 * chain of tasks off each css_set. Nests outside task->alloc_lock
321 * due to cgroup_iter_start() */
322 static DEFINE_RWLOCK(css_set_lock
);
323 static int css_set_count
;
326 * hash table for cgroup groups. This improves the performance to find
327 * an existing css_set. This hash doesn't (currently) take into
328 * account cgroups in empty hierarchies.
330 #define CSS_SET_HASH_BITS 7
331 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
332 static struct hlist_head css_set_table
[CSS_SET_TABLE_SIZE
];
334 static struct hlist_head
*css_set_hash(struct cgroup_subsys_state
*css
[])
338 unsigned long tmp
= 0UL;
340 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
341 tmp
+= (unsigned long)css
[i
];
342 tmp
= (tmp
>> 16) ^ tmp
;
344 index
= hash_long(tmp
, CSS_SET_HASH_BITS
);
346 return &css_set_table
[index
];
349 /* We don't maintain the lists running through each css_set to its
350 * task until after the first call to cgroup_iter_start(). This
351 * reduces the fork()/exit() overhead for people who have cgroups
352 * compiled into their kernel but not actually in use */
353 static int use_task_css_set_links __read_mostly
;
355 static void __put_css_set(struct css_set
*cg
, int taskexit
)
357 struct cg_cgroup_link
*link
;
358 struct cg_cgroup_link
*saved_link
;
360 * Ensure that the refcount doesn't hit zero while any readers
361 * can see it. Similar to atomic_dec_and_lock(), but for an
364 if (atomic_add_unless(&cg
->refcount
, -1, 1))
366 write_lock(&css_set_lock
);
367 if (!atomic_dec_and_test(&cg
->refcount
)) {
368 write_unlock(&css_set_lock
);
372 /* This css_set is dead. unlink it and release cgroup refcounts */
373 hlist_del(&cg
->hlist
);
376 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
378 struct cgroup
*cgrp
= link
->cgrp
;
379 list_del(&link
->cg_link_list
);
380 list_del(&link
->cgrp_link_list
);
381 if (atomic_dec_and_test(&cgrp
->count
) &&
382 notify_on_release(cgrp
)) {
384 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
385 check_for_release(cgrp
);
391 write_unlock(&css_set_lock
);
392 kfree_rcu(cg
, rcu_head
);
396 * refcounted get/put for css_set objects
398 static inline void get_css_set(struct css_set
*cg
)
400 atomic_inc(&cg
->refcount
);
403 static inline void put_css_set(struct css_set
*cg
)
405 __put_css_set(cg
, 0);
408 static inline void put_css_set_taskexit(struct css_set
*cg
)
410 __put_css_set(cg
, 1);
414 * compare_css_sets - helper function for find_existing_css_set().
415 * @cg: candidate css_set being tested
416 * @old_cg: existing css_set for a task
417 * @new_cgrp: cgroup that's being entered by the task
418 * @template: desired set of css pointers in css_set (pre-calculated)
420 * Returns true if "cg" matches "old_cg" except for the hierarchy
421 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
423 static bool compare_css_sets(struct css_set
*cg
,
424 struct css_set
*old_cg
,
425 struct cgroup
*new_cgrp
,
426 struct cgroup_subsys_state
*template[])
428 struct list_head
*l1
, *l2
;
430 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
431 /* Not all subsystems matched */
436 * Compare cgroup pointers in order to distinguish between
437 * different cgroups in heirarchies with no subsystems. We
438 * could get by with just this check alone (and skip the
439 * memcmp above) but on most setups the memcmp check will
440 * avoid the need for this more expensive check on almost all
445 l2
= &old_cg
->cg_links
;
447 struct cg_cgroup_link
*cgl1
, *cgl2
;
448 struct cgroup
*cg1
, *cg2
;
452 /* See if we reached the end - both lists are equal length. */
453 if (l1
== &cg
->cg_links
) {
454 BUG_ON(l2
!= &old_cg
->cg_links
);
457 BUG_ON(l2
== &old_cg
->cg_links
);
459 /* Locate the cgroups associated with these links. */
460 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
461 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
464 /* Hierarchies should be linked in the same order. */
465 BUG_ON(cg1
->root
!= cg2
->root
);
468 * If this hierarchy is the hierarchy of the cgroup
469 * that's changing, then we need to check that this
470 * css_set points to the new cgroup; if it's any other
471 * hierarchy, then this css_set should point to the
472 * same cgroup as the old css_set.
474 if (cg1
->root
== new_cgrp
->root
) {
486 * find_existing_css_set() is a helper for
487 * find_css_set(), and checks to see whether an existing
488 * css_set is suitable.
490 * oldcg: the cgroup group that we're using before the cgroup
493 * cgrp: the cgroup that we're moving into
495 * template: location in which to build the desired set of subsystem
496 * state objects for the new cgroup group
498 static struct css_set
*find_existing_css_set(
499 struct css_set
*oldcg
,
501 struct cgroup_subsys_state
*template[])
504 struct cgroupfs_root
*root
= cgrp
->root
;
505 struct hlist_head
*hhead
;
506 struct hlist_node
*node
;
510 * Build the set of subsystem state objects that we want to see in the
511 * new css_set. while subsystems can change globally, the entries here
512 * won't change, so no need for locking.
514 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
515 if (root
->subsys_bits
& (1UL << i
)) {
516 /* Subsystem is in this hierarchy. So we want
517 * the subsystem state from the new
519 template[i
] = cgrp
->subsys
[i
];
521 /* Subsystem is not in this hierarchy, so we
522 * don't want to change the subsystem state */
523 template[i
] = oldcg
->subsys
[i
];
527 hhead
= css_set_hash(template);
528 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
529 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
532 /* This css_set matches what we need */
536 /* No existing cgroup group matched */
540 static void free_cg_links(struct list_head
*tmp
)
542 struct cg_cgroup_link
*link
;
543 struct cg_cgroup_link
*saved_link
;
545 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
546 list_del(&link
->cgrp_link_list
);
552 * allocate_cg_links() allocates "count" cg_cgroup_link structures
553 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
554 * success or a negative error
556 static int allocate_cg_links(int count
, struct list_head
*tmp
)
558 struct cg_cgroup_link
*link
;
561 for (i
= 0; i
< count
; i
++) {
562 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
567 list_add(&link
->cgrp_link_list
, tmp
);
573 * link_css_set - a helper function to link a css_set to a cgroup
574 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
575 * @cg: the css_set to be linked
576 * @cgrp: the destination cgroup
578 static void link_css_set(struct list_head
*tmp_cg_links
,
579 struct css_set
*cg
, struct cgroup
*cgrp
)
581 struct cg_cgroup_link
*link
;
583 BUG_ON(list_empty(tmp_cg_links
));
584 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
588 atomic_inc(&cgrp
->count
);
589 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
591 * Always add links to the tail of the list so that the list
592 * is sorted by order of hierarchy creation
594 list_add_tail(&link
->cg_link_list
, &cg
->cg_links
);
598 * find_css_set() takes an existing cgroup group and a
599 * cgroup object, and returns a css_set object that's
600 * equivalent to the old group, but with the given cgroup
601 * substituted into the appropriate hierarchy. Must be called with
604 static struct css_set
*find_css_set(
605 struct css_set
*oldcg
, struct cgroup
*cgrp
)
608 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
610 struct list_head tmp_cg_links
;
612 struct hlist_head
*hhead
;
613 struct cg_cgroup_link
*link
;
615 /* First see if we already have a cgroup group that matches
617 read_lock(&css_set_lock
);
618 res
= find_existing_css_set(oldcg
, cgrp
, template);
621 read_unlock(&css_set_lock
);
626 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
630 /* Allocate all the cg_cgroup_link objects that we'll need */
631 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
636 atomic_set(&res
->refcount
, 1);
637 INIT_LIST_HEAD(&res
->cg_links
);
638 INIT_LIST_HEAD(&res
->tasks
);
639 INIT_HLIST_NODE(&res
->hlist
);
641 /* Copy the set of subsystem state objects generated in
642 * find_existing_css_set() */
643 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
645 write_lock(&css_set_lock
);
646 /* Add reference counts and links from the new css_set. */
647 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
648 struct cgroup
*c
= link
->cgrp
;
649 if (c
->root
== cgrp
->root
)
651 link_css_set(&tmp_cg_links
, res
, c
);
654 BUG_ON(!list_empty(&tmp_cg_links
));
658 /* Add this cgroup group to the hash table */
659 hhead
= css_set_hash(res
->subsys
);
660 hlist_add_head(&res
->hlist
, hhead
);
662 write_unlock(&css_set_lock
);
668 * Return the cgroup for "task" from the given hierarchy. Must be
669 * called with cgroup_mutex held.
671 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
672 struct cgroupfs_root
*root
)
675 struct cgroup
*res
= NULL
;
677 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
678 read_lock(&css_set_lock
);
680 * No need to lock the task - since we hold cgroup_mutex the
681 * task can't change groups, so the only thing that can happen
682 * is that it exits and its css is set back to init_css_set.
685 if (css
== &init_css_set
) {
686 res
= &root
->top_cgroup
;
688 struct cg_cgroup_link
*link
;
689 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
690 struct cgroup
*c
= link
->cgrp
;
691 if (c
->root
== root
) {
697 read_unlock(&css_set_lock
);
703 * There is one global cgroup mutex. We also require taking
704 * task_lock() when dereferencing a task's cgroup subsys pointers.
705 * See "The task_lock() exception", at the end of this comment.
707 * A task must hold cgroup_mutex to modify cgroups.
709 * Any task can increment and decrement the count field without lock.
710 * So in general, code holding cgroup_mutex can't rely on the count
711 * field not changing. However, if the count goes to zero, then only
712 * cgroup_attach_task() can increment it again. Because a count of zero
713 * means that no tasks are currently attached, therefore there is no
714 * way a task attached to that cgroup can fork (the other way to
715 * increment the count). So code holding cgroup_mutex can safely
716 * assume that if the count is zero, it will stay zero. Similarly, if
717 * a task holds cgroup_mutex on a cgroup with zero count, it
718 * knows that the cgroup won't be removed, as cgroup_rmdir()
721 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
722 * (usually) take cgroup_mutex. These are the two most performance
723 * critical pieces of code here. The exception occurs on cgroup_exit(),
724 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
725 * is taken, and if the cgroup count is zero, a usermode call made
726 * to the release agent with the name of the cgroup (path relative to
727 * the root of cgroup file system) as the argument.
729 * A cgroup can only be deleted if both its 'count' of using tasks
730 * is zero, and its list of 'children' cgroups is empty. Since all
731 * tasks in the system use _some_ cgroup, and since there is always at
732 * least one task in the system (init, pid == 1), therefore, top_cgroup
733 * always has either children cgroups and/or using tasks. So we don't
734 * need a special hack to ensure that top_cgroup cannot be deleted.
736 * The task_lock() exception
738 * The need for this exception arises from the action of
739 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
740 * another. It does so using cgroup_mutex, however there are
741 * several performance critical places that need to reference
742 * task->cgroup without the expense of grabbing a system global
743 * mutex. Therefore except as noted below, when dereferencing or, as
744 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
745 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
746 * the task_struct routinely used for such matters.
748 * P.S. One more locking exception. RCU is used to guard the
749 * update of a tasks cgroup pointer by cgroup_attach_task()
753 * cgroup_lock - lock out any changes to cgroup structures
756 void cgroup_lock(void)
758 mutex_lock(&cgroup_mutex
);
760 EXPORT_SYMBOL_GPL(cgroup_lock
);
763 * cgroup_unlock - release lock on cgroup changes
765 * Undo the lock taken in a previous cgroup_lock() call.
767 void cgroup_unlock(void)
769 mutex_unlock(&cgroup_mutex
);
771 EXPORT_SYMBOL_GPL(cgroup_unlock
);
774 * A couple of forward declarations required, due to cyclic reference loop:
775 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
776 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
780 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
781 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
782 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
783 static int cgroup_populate_dir(struct cgroup
*cgrp
);
784 static const struct inode_operations cgroup_dir_inode_operations
;
785 static const struct file_operations proc_cgroupstats_operations
;
787 static struct backing_dev_info cgroup_backing_dev_info
= {
789 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
792 static int alloc_css_id(struct cgroup_subsys
*ss
,
793 struct cgroup
*parent
, struct cgroup
*child
);
795 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
797 struct inode
*inode
= new_inode(sb
);
800 inode
->i_ino
= get_next_ino();
801 inode
->i_mode
= mode
;
802 inode
->i_uid
= current_fsuid();
803 inode
->i_gid
= current_fsgid();
804 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
805 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
811 * Call subsys's pre_destroy handler.
812 * This is called before css refcnt check.
814 static int cgroup_call_pre_destroy(struct cgroup
*cgrp
)
816 struct cgroup_subsys
*ss
;
819 for_each_subsys(cgrp
->root
, ss
)
820 if (ss
->pre_destroy
) {
821 ret
= ss
->pre_destroy(cgrp
);
829 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
831 /* is dentry a directory ? if so, kfree() associated cgroup */
832 if (S_ISDIR(inode
->i_mode
)) {
833 struct cgroup
*cgrp
= dentry
->d_fsdata
;
834 struct cgroup_subsys
*ss
;
835 BUG_ON(!(cgroup_is_removed(cgrp
)));
836 /* It's possible for external users to be holding css
837 * reference counts on a cgroup; css_put() needs to
838 * be able to access the cgroup after decrementing
839 * the reference count in order to know if it needs to
840 * queue the cgroup to be handled by the release
844 mutex_lock(&cgroup_mutex
);
846 * Release the subsystem state objects.
848 for_each_subsys(cgrp
->root
, ss
)
851 cgrp
->root
->number_of_cgroups
--;
852 mutex_unlock(&cgroup_mutex
);
855 * Drop the active superblock reference that we took when we
858 deactivate_super(cgrp
->root
->sb
);
861 * if we're getting rid of the cgroup, refcount should ensure
862 * that there are no pidlists left.
864 BUG_ON(!list_empty(&cgrp
->pidlists
));
866 kfree_rcu(cgrp
, rcu_head
);
871 static int cgroup_delete(const struct dentry
*d
)
876 static void remove_dir(struct dentry
*d
)
878 struct dentry
*parent
= dget(d
->d_parent
);
881 simple_rmdir(parent
->d_inode
, d
);
885 static void cgroup_clear_directory(struct dentry
*dentry
)
887 struct list_head
*node
;
889 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
890 spin_lock(&dentry
->d_lock
);
891 node
= dentry
->d_subdirs
.next
;
892 while (node
!= &dentry
->d_subdirs
) {
893 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
895 spin_lock_nested(&d
->d_lock
, DENTRY_D_LOCK_NESTED
);
898 /* This should never be called on a cgroup
899 * directory with child cgroups */
900 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
902 spin_unlock(&d
->d_lock
);
903 spin_unlock(&dentry
->d_lock
);
905 simple_unlink(dentry
->d_inode
, d
);
907 spin_lock(&dentry
->d_lock
);
909 spin_unlock(&d
->d_lock
);
910 node
= dentry
->d_subdirs
.next
;
912 spin_unlock(&dentry
->d_lock
);
916 * NOTE : the dentry must have been dget()'ed
918 static void cgroup_d_remove_dir(struct dentry
*dentry
)
920 struct dentry
*parent
;
922 cgroup_clear_directory(dentry
);
924 parent
= dentry
->d_parent
;
925 spin_lock(&parent
->d_lock
);
926 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
927 list_del_init(&dentry
->d_u
.d_child
);
928 spin_unlock(&dentry
->d_lock
);
929 spin_unlock(&parent
->d_lock
);
934 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
935 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
936 * reference to css->refcnt. In general, this refcnt is expected to goes down
939 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
941 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq
);
943 static void cgroup_wakeup_rmdir_waiter(struct cgroup
*cgrp
)
945 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
)))
946 wake_up_all(&cgroup_rmdir_waitq
);
949 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
)
954 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
)
956 cgroup_wakeup_rmdir_waiter(css
->cgroup
);
961 * Call with cgroup_mutex held. Drops reference counts on modules, including
962 * any duplicate ones that parse_cgroupfs_options took. If this function
963 * returns an error, no reference counts are touched.
965 static int rebind_subsystems(struct cgroupfs_root
*root
,
966 unsigned long final_bits
)
968 unsigned long added_bits
, removed_bits
;
969 struct cgroup
*cgrp
= &root
->top_cgroup
;
972 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
973 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
975 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
976 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
977 /* Check that any added subsystems are currently free */
978 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
979 unsigned long bit
= 1UL << i
;
980 struct cgroup_subsys
*ss
= subsys
[i
];
981 if (!(bit
& added_bits
))
984 * Nobody should tell us to do a subsys that doesn't exist:
985 * parse_cgroupfs_options should catch that case and refcounts
986 * ensure that subsystems won't disappear once selected.
989 if (ss
->root
!= &rootnode
) {
990 /* Subsystem isn't free */
995 /* Currently we don't handle adding/removing subsystems when
996 * any child cgroups exist. This is theoretically supportable
997 * but involves complex error handling, so it's being left until
999 if (root
->number_of_cgroups
> 1)
1002 /* Process each subsystem */
1003 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1004 struct cgroup_subsys
*ss
= subsys
[i
];
1005 unsigned long bit
= 1UL << i
;
1006 if (bit
& added_bits
) {
1007 /* We're binding this subsystem to this hierarchy */
1009 BUG_ON(cgrp
->subsys
[i
]);
1010 BUG_ON(!dummytop
->subsys
[i
]);
1011 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
1012 mutex_lock(&ss
->hierarchy_mutex
);
1013 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
1014 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1015 list_move(&ss
->sibling
, &root
->subsys_list
);
1019 mutex_unlock(&ss
->hierarchy_mutex
);
1020 /* refcount was already taken, and we're keeping it */
1021 } else if (bit
& removed_bits
) {
1022 /* We're removing this subsystem */
1024 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1025 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1026 mutex_lock(&ss
->hierarchy_mutex
);
1029 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1030 cgrp
->subsys
[i
] = NULL
;
1031 subsys
[i
]->root
= &rootnode
;
1032 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1033 mutex_unlock(&ss
->hierarchy_mutex
);
1034 /* subsystem is now free - drop reference on module */
1035 module_put(ss
->module
);
1036 } else if (bit
& final_bits
) {
1037 /* Subsystem state should already exist */
1039 BUG_ON(!cgrp
->subsys
[i
]);
1041 * a refcount was taken, but we already had one, so
1042 * drop the extra reference.
1044 module_put(ss
->module
);
1045 #ifdef CONFIG_MODULE_UNLOAD
1046 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1049 /* Subsystem state shouldn't exist */
1050 BUG_ON(cgrp
->subsys
[i
]);
1053 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
1059 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1061 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1062 struct cgroup_subsys
*ss
;
1064 mutex_lock(&cgroup_root_mutex
);
1065 for_each_subsys(root
, ss
)
1066 seq_printf(seq
, ",%s", ss
->name
);
1067 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1068 seq_puts(seq
, ",noprefix");
1069 if (strlen(root
->release_agent_path
))
1070 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1071 if (clone_children(&root
->top_cgroup
))
1072 seq_puts(seq
, ",clone_children");
1073 if (strlen(root
->name
))
1074 seq_printf(seq
, ",name=%s", root
->name
);
1075 mutex_unlock(&cgroup_root_mutex
);
1079 struct cgroup_sb_opts
{
1080 unsigned long subsys_bits
;
1081 unsigned long flags
;
1082 char *release_agent
;
1083 bool clone_children
;
1085 /* User explicitly requested empty subsystem */
1088 struct cgroupfs_root
*new_root
;
1093 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1094 * with cgroup_mutex held to protect the subsys[] array. This function takes
1095 * refcounts on subsystems to be used, unless it returns error, in which case
1096 * no refcounts are taken.
1098 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1100 char *token
, *o
= data
;
1101 bool all_ss
= false, one_ss
= false;
1102 unsigned long mask
= (unsigned long)-1;
1104 bool module_pin_failed
= false;
1106 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1108 #ifdef CONFIG_CPUSETS
1109 mask
= ~(1UL << cpuset_subsys_id
);
1112 memset(opts
, 0, sizeof(*opts
));
1114 while ((token
= strsep(&o
, ",")) != NULL
) {
1117 if (!strcmp(token
, "none")) {
1118 /* Explicitly have no subsystems */
1122 if (!strcmp(token
, "all")) {
1123 /* Mutually exclusive option 'all' + subsystem name */
1129 if (!strcmp(token
, "noprefix")) {
1130 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1133 if (!strcmp(token
, "clone_children")) {
1134 opts
->clone_children
= true;
1137 if (!strncmp(token
, "release_agent=", 14)) {
1138 /* Specifying two release agents is forbidden */
1139 if (opts
->release_agent
)
1141 opts
->release_agent
=
1142 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1143 if (!opts
->release_agent
)
1147 if (!strncmp(token
, "name=", 5)) {
1148 const char *name
= token
+ 5;
1149 /* Can't specify an empty name */
1152 /* Must match [\w.-]+ */
1153 for (i
= 0; i
< strlen(name
); i
++) {
1157 if ((c
== '.') || (c
== '-') || (c
== '_'))
1161 /* Specifying two names is forbidden */
1164 opts
->name
= kstrndup(name
,
1165 MAX_CGROUP_ROOT_NAMELEN
- 1,
1173 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1174 struct cgroup_subsys
*ss
= subsys
[i
];
1177 if (strcmp(token
, ss
->name
))
1182 /* Mutually exclusive option 'all' + subsystem name */
1185 set_bit(i
, &opts
->subsys_bits
);
1190 if (i
== CGROUP_SUBSYS_COUNT
)
1195 * If the 'all' option was specified select all the subsystems,
1196 * otherwise if 'none', 'name=' and a subsystem name options
1197 * were not specified, let's default to 'all'
1199 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
)) {
1200 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1201 struct cgroup_subsys
*ss
= subsys
[i
];
1206 set_bit(i
, &opts
->subsys_bits
);
1210 /* Consistency checks */
1213 * Option noprefix was introduced just for backward compatibility
1214 * with the old cpuset, so we allow noprefix only if mounting just
1215 * the cpuset subsystem.
1217 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1218 (opts
->subsys_bits
& mask
))
1222 /* Can't specify "none" and some subsystems */
1223 if (opts
->subsys_bits
&& opts
->none
)
1227 * We either have to specify by name or by subsystems. (So all
1228 * empty hierarchies must have a name).
1230 if (!opts
->subsys_bits
&& !opts
->name
)
1234 * Grab references on all the modules we'll need, so the subsystems
1235 * don't dance around before rebind_subsystems attaches them. This may
1236 * take duplicate reference counts on a subsystem that's already used,
1237 * but rebind_subsystems handles this case.
1239 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1240 unsigned long bit
= 1UL << i
;
1242 if (!(bit
& opts
->subsys_bits
))
1244 if (!try_module_get(subsys
[i
]->module
)) {
1245 module_pin_failed
= true;
1249 if (module_pin_failed
) {
1251 * oops, one of the modules was going away. this means that we
1252 * raced with a module_delete call, and to the user this is
1253 * essentially a "subsystem doesn't exist" case.
1255 for (i
--; i
>= CGROUP_BUILTIN_SUBSYS_COUNT
; i
--) {
1256 /* drop refcounts only on the ones we took */
1257 unsigned long bit
= 1UL << i
;
1259 if (!(bit
& opts
->subsys_bits
))
1261 module_put(subsys
[i
]->module
);
1269 static void drop_parsed_module_refcounts(unsigned long subsys_bits
)
1272 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1273 unsigned long bit
= 1UL << i
;
1275 if (!(bit
& subsys_bits
))
1277 module_put(subsys
[i
]->module
);
1281 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1284 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1285 struct cgroup
*cgrp
= &root
->top_cgroup
;
1286 struct cgroup_sb_opts opts
;
1288 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1289 mutex_lock(&cgroup_mutex
);
1290 mutex_lock(&cgroup_root_mutex
);
1292 /* See what subsystems are wanted */
1293 ret
= parse_cgroupfs_options(data
, &opts
);
1297 /* Don't allow flags or name to change at remount */
1298 if (opts
.flags
!= root
->flags
||
1299 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1301 drop_parsed_module_refcounts(opts
.subsys_bits
);
1305 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
1307 drop_parsed_module_refcounts(opts
.subsys_bits
);
1311 /* (re)populate subsystem files */
1312 cgroup_populate_dir(cgrp
);
1314 if (opts
.release_agent
)
1315 strcpy(root
->release_agent_path
, opts
.release_agent
);
1317 kfree(opts
.release_agent
);
1319 mutex_unlock(&cgroup_root_mutex
);
1320 mutex_unlock(&cgroup_mutex
);
1321 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1325 static const struct super_operations cgroup_ops
= {
1326 .statfs
= simple_statfs
,
1327 .drop_inode
= generic_delete_inode
,
1328 .show_options
= cgroup_show_options
,
1329 .remount_fs
= cgroup_remount
,
1332 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1334 INIT_LIST_HEAD(&cgrp
->sibling
);
1335 INIT_LIST_HEAD(&cgrp
->children
);
1336 INIT_LIST_HEAD(&cgrp
->css_sets
);
1337 INIT_LIST_HEAD(&cgrp
->release_list
);
1338 INIT_LIST_HEAD(&cgrp
->pidlists
);
1339 mutex_init(&cgrp
->pidlist_mutex
);
1340 INIT_LIST_HEAD(&cgrp
->event_list
);
1341 spin_lock_init(&cgrp
->event_list_lock
);
1344 static void init_cgroup_root(struct cgroupfs_root
*root
)
1346 struct cgroup
*cgrp
= &root
->top_cgroup
;
1347 INIT_LIST_HEAD(&root
->subsys_list
);
1348 INIT_LIST_HEAD(&root
->root_list
);
1349 root
->number_of_cgroups
= 1;
1351 cgrp
->top_cgroup
= cgrp
;
1352 init_cgroup_housekeeping(cgrp
);
1355 static bool init_root_id(struct cgroupfs_root
*root
)
1360 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1362 spin_lock(&hierarchy_id_lock
);
1363 /* Try to allocate the next unused ID */
1364 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1365 &root
->hierarchy_id
);
1367 /* Try again starting from 0 */
1368 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1370 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1371 } else if (ret
!= -EAGAIN
) {
1372 /* Can only get here if the 31-bit IDR is full ... */
1375 spin_unlock(&hierarchy_id_lock
);
1380 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1382 struct cgroup_sb_opts
*opts
= data
;
1383 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1385 /* If we asked for a name then it must match */
1386 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1390 * If we asked for subsystems (or explicitly for no
1391 * subsystems) then they must match
1393 if ((opts
->subsys_bits
|| opts
->none
)
1394 && (opts
->subsys_bits
!= root
->subsys_bits
))
1400 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1402 struct cgroupfs_root
*root
;
1404 if (!opts
->subsys_bits
&& !opts
->none
)
1407 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1409 return ERR_PTR(-ENOMEM
);
1411 if (!init_root_id(root
)) {
1413 return ERR_PTR(-ENOMEM
);
1415 init_cgroup_root(root
);
1417 root
->subsys_bits
= opts
->subsys_bits
;
1418 root
->flags
= opts
->flags
;
1419 if (opts
->release_agent
)
1420 strcpy(root
->release_agent_path
, opts
->release_agent
);
1422 strcpy(root
->name
, opts
->name
);
1423 if (opts
->clone_children
)
1424 set_bit(CGRP_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1428 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1433 BUG_ON(!root
->hierarchy_id
);
1434 spin_lock(&hierarchy_id_lock
);
1435 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1436 spin_unlock(&hierarchy_id_lock
);
1440 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1443 struct cgroup_sb_opts
*opts
= data
;
1445 /* If we don't have a new root, we can't set up a new sb */
1446 if (!opts
->new_root
)
1449 BUG_ON(!opts
->subsys_bits
&& !opts
->none
);
1451 ret
= set_anon_super(sb
, NULL
);
1455 sb
->s_fs_info
= opts
->new_root
;
1456 opts
->new_root
->sb
= sb
;
1458 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1459 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1460 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1461 sb
->s_op
= &cgroup_ops
;
1466 static int cgroup_get_rootdir(struct super_block
*sb
)
1468 static const struct dentry_operations cgroup_dops
= {
1469 .d_iput
= cgroup_diput
,
1470 .d_delete
= cgroup_delete
,
1473 struct inode
*inode
=
1474 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1475 struct dentry
*dentry
;
1480 inode
->i_fop
= &simple_dir_operations
;
1481 inode
->i_op
= &cgroup_dir_inode_operations
;
1482 /* directories start off with i_nlink == 2 (for "." entry) */
1484 dentry
= d_alloc_root(inode
);
1489 sb
->s_root
= dentry
;
1490 /* for everything else we want ->d_op set */
1491 sb
->s_d_op
= &cgroup_dops
;
1495 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1496 int flags
, const char *unused_dev_name
,
1499 struct cgroup_sb_opts opts
;
1500 struct cgroupfs_root
*root
;
1502 struct super_block
*sb
;
1503 struct cgroupfs_root
*new_root
;
1504 struct inode
*inode
;
1506 /* First find the desired set of subsystems */
1507 mutex_lock(&cgroup_mutex
);
1508 ret
= parse_cgroupfs_options(data
, &opts
);
1509 mutex_unlock(&cgroup_mutex
);
1514 * Allocate a new cgroup root. We may not need it if we're
1515 * reusing an existing hierarchy.
1517 new_root
= cgroup_root_from_opts(&opts
);
1518 if (IS_ERR(new_root
)) {
1519 ret
= PTR_ERR(new_root
);
1522 opts
.new_root
= new_root
;
1524 /* Locate an existing or new sb for this hierarchy */
1525 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, &opts
);
1528 cgroup_drop_root(opts
.new_root
);
1532 root
= sb
->s_fs_info
;
1534 if (root
== opts
.new_root
) {
1535 /* We used the new root structure, so this is a new hierarchy */
1536 struct list_head tmp_cg_links
;
1537 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1538 struct cgroupfs_root
*existing_root
;
1539 const struct cred
*cred
;
1542 BUG_ON(sb
->s_root
!= NULL
);
1544 ret
= cgroup_get_rootdir(sb
);
1546 goto drop_new_super
;
1547 inode
= sb
->s_root
->d_inode
;
1549 mutex_lock(&inode
->i_mutex
);
1550 mutex_lock(&cgroup_mutex
);
1551 mutex_lock(&cgroup_root_mutex
);
1553 /* Check for name clashes with existing mounts */
1555 if (strlen(root
->name
))
1556 for_each_active_root(existing_root
)
1557 if (!strcmp(existing_root
->name
, root
->name
))
1561 * We're accessing css_set_count without locking
1562 * css_set_lock here, but that's OK - it can only be
1563 * increased by someone holding cgroup_lock, and
1564 * that's us. The worst that can happen is that we
1565 * have some link structures left over
1567 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1571 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1572 if (ret
== -EBUSY
) {
1573 free_cg_links(&tmp_cg_links
);
1577 * There must be no failure case after here, since rebinding
1578 * takes care of subsystems' refcounts, which are explicitly
1579 * dropped in the failure exit path.
1582 /* EBUSY should be the only error here */
1585 list_add(&root
->root_list
, &roots
);
1588 sb
->s_root
->d_fsdata
= root_cgrp
;
1589 root
->top_cgroup
.dentry
= sb
->s_root
;
1591 /* Link the top cgroup in this hierarchy into all
1592 * the css_set objects */
1593 write_lock(&css_set_lock
);
1594 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1595 struct hlist_head
*hhead
= &css_set_table
[i
];
1596 struct hlist_node
*node
;
1599 hlist_for_each_entry(cg
, node
, hhead
, hlist
)
1600 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1602 write_unlock(&css_set_lock
);
1604 free_cg_links(&tmp_cg_links
);
1606 BUG_ON(!list_empty(&root_cgrp
->sibling
));
1607 BUG_ON(!list_empty(&root_cgrp
->children
));
1608 BUG_ON(root
->number_of_cgroups
!= 1);
1610 cred
= override_creds(&init_cred
);
1611 cgroup_populate_dir(root_cgrp
);
1613 mutex_unlock(&cgroup_root_mutex
);
1614 mutex_unlock(&cgroup_mutex
);
1615 mutex_unlock(&inode
->i_mutex
);
1618 * We re-used an existing hierarchy - the new root (if
1619 * any) is not needed
1621 cgroup_drop_root(opts
.new_root
);
1622 /* no subsys rebinding, so refcounts don't change */
1623 drop_parsed_module_refcounts(opts
.subsys_bits
);
1626 kfree(opts
.release_agent
);
1628 return dget(sb
->s_root
);
1631 mutex_unlock(&cgroup_root_mutex
);
1632 mutex_unlock(&cgroup_mutex
);
1633 mutex_unlock(&inode
->i_mutex
);
1635 deactivate_locked_super(sb
);
1637 drop_parsed_module_refcounts(opts
.subsys_bits
);
1639 kfree(opts
.release_agent
);
1641 return ERR_PTR(ret
);
1644 static void cgroup_kill_sb(struct super_block
*sb
) {
1645 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1646 struct cgroup
*cgrp
= &root
->top_cgroup
;
1648 struct cg_cgroup_link
*link
;
1649 struct cg_cgroup_link
*saved_link
;
1653 BUG_ON(root
->number_of_cgroups
!= 1);
1654 BUG_ON(!list_empty(&cgrp
->children
));
1655 BUG_ON(!list_empty(&cgrp
->sibling
));
1657 mutex_lock(&cgroup_mutex
);
1658 mutex_lock(&cgroup_root_mutex
);
1660 /* Rebind all subsystems back to the default hierarchy */
1661 ret
= rebind_subsystems(root
, 0);
1662 /* Shouldn't be able to fail ... */
1666 * Release all the links from css_sets to this hierarchy's
1669 write_lock(&css_set_lock
);
1671 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1673 list_del(&link
->cg_link_list
);
1674 list_del(&link
->cgrp_link_list
);
1677 write_unlock(&css_set_lock
);
1679 if (!list_empty(&root
->root_list
)) {
1680 list_del(&root
->root_list
);
1684 mutex_unlock(&cgroup_root_mutex
);
1685 mutex_unlock(&cgroup_mutex
);
1687 kill_litter_super(sb
);
1688 cgroup_drop_root(root
);
1691 static struct file_system_type cgroup_fs_type
= {
1693 .mount
= cgroup_mount
,
1694 .kill_sb
= cgroup_kill_sb
,
1697 static struct kobject
*cgroup_kobj
;
1699 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1701 return dentry
->d_fsdata
;
1704 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1706 return dentry
->d_fsdata
;
1710 * cgroup_path - generate the path of a cgroup
1711 * @cgrp: the cgroup in question
1712 * @buf: the buffer to write the path into
1713 * @buflen: the length of the buffer
1715 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1716 * reference. Writes path of cgroup into buf. Returns 0 on success,
1719 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1722 struct dentry
*dentry
= rcu_dereference_check(cgrp
->dentry
,
1723 cgroup_lock_is_held());
1725 if (!dentry
|| cgrp
== dummytop
) {
1727 * Inactive subsystems have no dentry for their root
1734 start
= buf
+ buflen
;
1738 int len
= dentry
->d_name
.len
;
1740 if ((start
-= len
) < buf
)
1741 return -ENAMETOOLONG
;
1742 memcpy(start
, dentry
->d_name
.name
, len
);
1743 cgrp
= cgrp
->parent
;
1747 dentry
= rcu_dereference_check(cgrp
->dentry
,
1748 cgroup_lock_is_held());
1752 return -ENAMETOOLONG
;
1755 memmove(buf
, start
, buf
+ buflen
- start
);
1758 EXPORT_SYMBOL_GPL(cgroup_path
);
1761 * Control Group taskset
1763 struct task_and_cgroup
{
1764 struct task_struct
*task
;
1765 struct cgroup
*cgrp
;
1769 struct cgroup_taskset
{
1770 struct task_and_cgroup single
;
1771 struct flex_array
*tc_array
;
1774 struct cgroup
*cur_cgrp
;
1778 * cgroup_taskset_first - reset taskset and return the first task
1779 * @tset: taskset of interest
1781 * @tset iteration is initialized and the first task is returned.
1783 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1785 if (tset
->tc_array
) {
1787 return cgroup_taskset_next(tset
);
1789 tset
->cur_cgrp
= tset
->single
.cgrp
;
1790 return tset
->single
.task
;
1793 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1796 * cgroup_taskset_next - iterate to the next task in taskset
1797 * @tset: taskset of interest
1799 * Return the next task in @tset. Iteration must have been initialized
1800 * with cgroup_taskset_first().
1802 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1804 struct task_and_cgroup
*tc
;
1806 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1809 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1810 tset
->cur_cgrp
= tc
->cgrp
;
1813 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1816 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1817 * @tset: taskset of interest
1819 * Return the cgroup for the current (last returned) task of @tset. This
1820 * function must be preceded by either cgroup_taskset_first() or
1821 * cgroup_taskset_next().
1823 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1825 return tset
->cur_cgrp
;
1827 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1830 * cgroup_taskset_size - return the number of tasks in taskset
1831 * @tset: taskset of interest
1833 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1835 return tset
->tc_array
? tset
->tc_array_len
: 1;
1837 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1841 * cgroup_task_migrate - move a task from one cgroup to another.
1843 * 'guarantee' is set if the caller promises that a new css_set for the task
1844 * will already exist. If not set, this function might sleep, and can fail with
1845 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1847 static void cgroup_task_migrate(struct cgroup
*cgrp
, struct cgroup
*oldcgrp
,
1848 struct task_struct
*tsk
, struct css_set
*newcg
)
1850 struct css_set
*oldcg
;
1853 * We are synchronized through threadgroup_lock() against PF_EXITING
1854 * setting such that we can't race against cgroup_exit() changing the
1855 * css_set to init_css_set and dropping the old one.
1857 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1858 oldcg
= tsk
->cgroups
;
1861 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1864 /* Update the css_set linked lists if we're using them */
1865 write_lock(&css_set_lock
);
1866 if (!list_empty(&tsk
->cg_list
))
1867 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1868 write_unlock(&css_set_lock
);
1871 * We just gained a reference on oldcg by taking it from the task. As
1872 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1873 * it here; it will be freed under RCU.
1877 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1881 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1882 * @cgrp: the cgroup the task is attaching to
1883 * @tsk: the task to be attached
1885 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1888 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1891 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1892 struct cgroup
*oldcgrp
;
1893 struct cgroupfs_root
*root
= cgrp
->root
;
1894 struct cgroup_taskset tset
= { };
1895 struct css_set
*newcg
;
1897 /* @tsk either already exited or can't exit until the end */
1898 if (tsk
->flags
& PF_EXITING
)
1901 /* Nothing to do if the task is already in that cgroup */
1902 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1903 if (cgrp
== oldcgrp
)
1906 tset
.single
.task
= tsk
;
1907 tset
.single
.cgrp
= oldcgrp
;
1909 for_each_subsys(root
, ss
) {
1910 if (ss
->can_attach
) {
1911 retval
= ss
->can_attach(cgrp
, &tset
);
1914 * Remember on which subsystem the can_attach()
1915 * failed, so that we only call cancel_attach()
1916 * against the subsystems whose can_attach()
1917 * succeeded. (See below)
1925 newcg
= find_css_set(tsk
->cgroups
, cgrp
);
1931 cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, newcg
);
1933 for_each_subsys(root
, ss
) {
1935 ss
->attach(cgrp
, &tset
);
1941 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1942 * is no longer empty.
1944 cgroup_wakeup_rmdir_waiter(cgrp
);
1947 for_each_subsys(root
, ss
) {
1948 if (ss
== failed_ss
)
1950 * This subsystem was the one that failed the
1951 * can_attach() check earlier, so we don't need
1952 * to call cancel_attach() against it or any
1953 * remaining subsystems.
1956 if (ss
->cancel_attach
)
1957 ss
->cancel_attach(cgrp
, &tset
);
1964 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1965 * @from: attach to all cgroups of a given task
1966 * @tsk: the task to be attached
1968 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
1970 struct cgroupfs_root
*root
;
1974 for_each_active_root(root
) {
1975 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
1977 retval
= cgroup_attach_task(from_cg
, tsk
);
1985 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
1988 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
1989 * @cgrp: the cgroup to attach to
1990 * @leader: the threadgroup leader task_struct of the group to be attached
1992 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
1993 * task_lock of each thread in leader's threadgroup individually in turn.
1995 static int cgroup_attach_proc(struct cgroup
*cgrp
, struct task_struct
*leader
)
1997 int retval
, i
, group_size
;
1998 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1999 /* guaranteed to be initialized later, but the compiler needs this */
2000 struct cgroupfs_root
*root
= cgrp
->root
;
2001 /* threadgroup list cursor and array */
2002 struct task_struct
*tsk
;
2003 struct task_and_cgroup
*tc
;
2004 struct flex_array
*group
;
2005 struct cgroup_taskset tset
= { };
2008 * step 0: in order to do expensive, possibly blocking operations for
2009 * every thread, we cannot iterate the thread group list, since it needs
2010 * rcu or tasklist locked. instead, build an array of all threads in the
2011 * group - group_rwsem prevents new threads from appearing, and if
2012 * threads exit, this will just be an over-estimate.
2014 group_size
= get_nr_threads(leader
);
2015 /* flex_array supports very large thread-groups better than kmalloc. */
2016 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2019 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2020 retval
= flex_array_prealloc(group
, 0, group_size
- 1, GFP_KERNEL
);
2022 goto out_free_group_list
;
2027 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2028 * already PF_EXITING could be freed from underneath us unless we
2029 * take an rcu_read_lock.
2033 struct task_and_cgroup ent
;
2035 /* @tsk either already exited or can't exit until the end */
2036 if (tsk
->flags
& PF_EXITING
)
2039 /* as per above, nr_threads may decrease, but not increase. */
2040 BUG_ON(i
>= group_size
);
2042 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2043 /* nothing to do if this task is already in the cgroup */
2044 if (ent
.cgrp
== cgrp
)
2047 * saying GFP_ATOMIC has no effect here because we did prealloc
2048 * earlier, but it's good form to communicate our expectations.
2050 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2051 BUG_ON(retval
!= 0);
2053 } while_each_thread(leader
, tsk
);
2055 /* remember the number of threads in the array for later. */
2057 tset
.tc_array
= group
;
2058 tset
.tc_array_len
= group_size
;
2060 /* methods shouldn't be called if no task is actually migrating */
2063 goto out_free_group_list
;
2066 * step 1: check that we can legitimately attach to the cgroup.
2068 for_each_subsys(root
, ss
) {
2069 if (ss
->can_attach
) {
2070 retval
= ss
->can_attach(cgrp
, &tset
);
2073 goto out_cancel_attach
;
2079 * step 2: make sure css_sets exist for all threads to be migrated.
2080 * we use find_css_set, which allocates a new one if necessary.
2082 for (i
= 0; i
< group_size
; i
++) {
2083 tc
= flex_array_get(group
, i
);
2084 tc
->cg
= find_css_set(tc
->task
->cgroups
, cgrp
);
2087 goto out_put_css_set_refs
;
2092 * step 3: now that we're guaranteed success wrt the css_sets,
2093 * proceed to move all tasks to the new cgroup. There are no
2094 * failure cases after here, so this is the commit point.
2096 for (i
= 0; i
< group_size
; i
++) {
2097 tc
= flex_array_get(group
, i
);
2098 cgroup_task_migrate(cgrp
, tc
->cgrp
, tc
->task
, tc
->cg
);
2100 /* nothing is sensitive to fork() after this point. */
2103 * step 4: do subsystem attach callbacks.
2105 for_each_subsys(root
, ss
) {
2107 ss
->attach(cgrp
, &tset
);
2111 * step 5: success! and cleanup
2114 cgroup_wakeup_rmdir_waiter(cgrp
);
2116 out_put_css_set_refs
:
2118 for (i
= 0; i
< group_size
; i
++) {
2119 tc
= flex_array_get(group
, i
);
2122 put_css_set(tc
->cg
);
2127 for_each_subsys(root
, ss
) {
2128 if (ss
== failed_ss
)
2130 if (ss
->cancel_attach
)
2131 ss
->cancel_attach(cgrp
, &tset
);
2134 out_free_group_list
:
2135 flex_array_free(group
);
2140 * Find the task_struct of the task to attach by vpid and pass it along to the
2141 * function to attach either it or all tasks in its threadgroup. Will lock
2142 * cgroup_mutex and threadgroup; may take task_lock of task.
2144 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2146 struct task_struct
*tsk
;
2147 const struct cred
*cred
= current_cred(), *tcred
;
2150 if (!cgroup_lock_live_group(cgrp
))
2156 tsk
= find_task_by_vpid(pid
);
2160 goto out_unlock_cgroup
;
2163 * even if we're attaching all tasks in the thread group, we
2164 * only need to check permissions on one of them.
2166 tcred
= __task_cred(tsk
);
2168 cred
->euid
!= tcred
->uid
&&
2169 cred
->euid
!= tcred
->suid
) {
2172 goto out_unlock_cgroup
;
2178 tsk
= tsk
->group_leader
;
2179 get_task_struct(tsk
);
2182 threadgroup_lock(tsk
);
2184 if (!thread_group_leader(tsk
)) {
2186 * a race with de_thread from another thread's exec()
2187 * may strip us of our leadership, if this happens,
2188 * there is no choice but to throw this task away and
2189 * try again; this is
2190 * "double-double-toil-and-trouble-check locking".
2192 threadgroup_unlock(tsk
);
2193 put_task_struct(tsk
);
2194 goto retry_find_task
;
2196 ret
= cgroup_attach_proc(cgrp
, tsk
);
2198 ret
= cgroup_attach_task(cgrp
, tsk
);
2199 threadgroup_unlock(tsk
);
2201 put_task_struct(tsk
);
2207 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2209 return attach_task_by_pid(cgrp
, pid
, false);
2212 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2214 return attach_task_by_pid(cgrp
, tgid
, true);
2218 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2219 * @cgrp: the cgroup to be checked for liveness
2221 * On success, returns true; the lock should be later released with
2222 * cgroup_unlock(). On failure returns false with no lock held.
2224 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
2226 mutex_lock(&cgroup_mutex
);
2227 if (cgroup_is_removed(cgrp
)) {
2228 mutex_unlock(&cgroup_mutex
);
2233 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
2235 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2238 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2239 if (strlen(buffer
) >= PATH_MAX
)
2241 if (!cgroup_lock_live_group(cgrp
))
2243 mutex_lock(&cgroup_root_mutex
);
2244 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2245 mutex_unlock(&cgroup_root_mutex
);
2250 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2251 struct seq_file
*seq
)
2253 if (!cgroup_lock_live_group(cgrp
))
2255 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2256 seq_putc(seq
, '\n');
2261 /* A buffer size big enough for numbers or short strings */
2262 #define CGROUP_LOCAL_BUFFER_SIZE 64
2264 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2266 const char __user
*userbuf
,
2267 size_t nbytes
, loff_t
*unused_ppos
)
2269 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2275 if (nbytes
>= sizeof(buffer
))
2277 if (copy_from_user(buffer
, userbuf
, nbytes
))
2280 buffer
[nbytes
] = 0; /* nul-terminate */
2281 if (cft
->write_u64
) {
2282 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2285 retval
= cft
->write_u64(cgrp
, cft
, val
);
2287 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2290 retval
= cft
->write_s64(cgrp
, cft
, val
);
2297 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2299 const char __user
*userbuf
,
2300 size_t nbytes
, loff_t
*unused_ppos
)
2302 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2304 size_t max_bytes
= cft
->max_write_len
;
2305 char *buffer
= local_buffer
;
2308 max_bytes
= sizeof(local_buffer
) - 1;
2309 if (nbytes
>= max_bytes
)
2311 /* Allocate a dynamic buffer if we need one */
2312 if (nbytes
>= sizeof(local_buffer
)) {
2313 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2317 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2322 buffer
[nbytes
] = 0; /* nul-terminate */
2323 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2327 if (buffer
!= local_buffer
)
2332 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2333 size_t nbytes
, loff_t
*ppos
)
2335 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2336 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2338 if (cgroup_is_removed(cgrp
))
2341 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2342 if (cft
->write_u64
|| cft
->write_s64
)
2343 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2344 if (cft
->write_string
)
2345 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2347 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2348 return ret
? ret
: nbytes
;
2353 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2355 char __user
*buf
, size_t nbytes
,
2358 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2359 u64 val
= cft
->read_u64(cgrp
, cft
);
2360 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2362 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2365 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2367 char __user
*buf
, size_t nbytes
,
2370 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2371 s64 val
= cft
->read_s64(cgrp
, cft
);
2372 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2374 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2377 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2378 size_t nbytes
, loff_t
*ppos
)
2380 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2381 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2383 if (cgroup_is_removed(cgrp
))
2387 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2389 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2391 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2396 * seqfile ops/methods for returning structured data. Currently just
2397 * supports string->u64 maps, but can be extended in future.
2400 struct cgroup_seqfile_state
{
2402 struct cgroup
*cgroup
;
2405 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2407 struct seq_file
*sf
= cb
->state
;
2408 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2411 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2413 struct cgroup_seqfile_state
*state
= m
->private;
2414 struct cftype
*cft
= state
->cft
;
2415 if (cft
->read_map
) {
2416 struct cgroup_map_cb cb
= {
2417 .fill
= cgroup_map_add
,
2420 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2422 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2425 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2427 struct seq_file
*seq
= file
->private_data
;
2428 kfree(seq
->private);
2429 return single_release(inode
, file
);
2432 static const struct file_operations cgroup_seqfile_operations
= {
2434 .write
= cgroup_file_write
,
2435 .llseek
= seq_lseek
,
2436 .release
= cgroup_seqfile_release
,
2439 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2444 err
= generic_file_open(inode
, file
);
2447 cft
= __d_cft(file
->f_dentry
);
2449 if (cft
->read_map
|| cft
->read_seq_string
) {
2450 struct cgroup_seqfile_state
*state
=
2451 kzalloc(sizeof(*state
), GFP_USER
);
2455 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2456 file
->f_op
= &cgroup_seqfile_operations
;
2457 err
= single_open(file
, cgroup_seqfile_show
, state
);
2460 } else if (cft
->open
)
2461 err
= cft
->open(inode
, file
);
2468 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2470 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2472 return cft
->release(inode
, file
);
2477 * cgroup_rename - Only allow simple rename of directories in place.
2479 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2480 struct inode
*new_dir
, struct dentry
*new_dentry
)
2482 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2484 if (new_dentry
->d_inode
)
2486 if (old_dir
!= new_dir
)
2488 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2491 static const struct file_operations cgroup_file_operations
= {
2492 .read
= cgroup_file_read
,
2493 .write
= cgroup_file_write
,
2494 .llseek
= generic_file_llseek
,
2495 .open
= cgroup_file_open
,
2496 .release
= cgroup_file_release
,
2499 static const struct inode_operations cgroup_dir_inode_operations
= {
2500 .lookup
= cgroup_lookup
,
2501 .mkdir
= cgroup_mkdir
,
2502 .rmdir
= cgroup_rmdir
,
2503 .rename
= cgroup_rename
,
2506 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
2508 if (dentry
->d_name
.len
> NAME_MAX
)
2509 return ERR_PTR(-ENAMETOOLONG
);
2510 d_add(dentry
, NULL
);
2515 * Check if a file is a control file
2517 static inline struct cftype
*__file_cft(struct file
*file
)
2519 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2520 return ERR_PTR(-EINVAL
);
2521 return __d_cft(file
->f_dentry
);
2524 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2525 struct super_block
*sb
)
2527 struct inode
*inode
;
2531 if (dentry
->d_inode
)
2534 inode
= cgroup_new_inode(mode
, sb
);
2538 if (S_ISDIR(mode
)) {
2539 inode
->i_op
= &cgroup_dir_inode_operations
;
2540 inode
->i_fop
= &simple_dir_operations
;
2542 /* start off with i_nlink == 2 (for "." entry) */
2545 /* start with the directory inode held, so that we can
2546 * populate it without racing with another mkdir */
2547 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2548 } else if (S_ISREG(mode
)) {
2550 inode
->i_fop
= &cgroup_file_operations
;
2552 d_instantiate(dentry
, inode
);
2553 dget(dentry
); /* Extra count - pin the dentry in core */
2558 * cgroup_create_dir - create a directory for an object.
2559 * @cgrp: the cgroup we create the directory for. It must have a valid
2560 * ->parent field. And we are going to fill its ->dentry field.
2561 * @dentry: dentry of the new cgroup
2562 * @mode: mode to set on new directory.
2564 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
2567 struct dentry
*parent
;
2570 parent
= cgrp
->parent
->dentry
;
2571 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
2573 dentry
->d_fsdata
= cgrp
;
2574 inc_nlink(parent
->d_inode
);
2575 rcu_assign_pointer(cgrp
->dentry
, dentry
);
2584 * cgroup_file_mode - deduce file mode of a control file
2585 * @cft: the control file in question
2587 * returns cft->mode if ->mode is not 0
2588 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2589 * returns S_IRUGO if it has only a read handler
2590 * returns S_IWUSR if it has only a write hander
2592 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2599 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2600 cft
->read_map
|| cft
->read_seq_string
)
2603 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2604 cft
->write_string
|| cft
->trigger
)
2610 int cgroup_add_file(struct cgroup
*cgrp
,
2611 struct cgroup_subsys
*subsys
,
2612 const struct cftype
*cft
)
2614 struct dentry
*dir
= cgrp
->dentry
;
2615 struct dentry
*dentry
;
2619 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2620 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2621 strcpy(name
, subsys
->name
);
2624 strcat(name
, cft
->name
);
2625 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2626 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2627 if (!IS_ERR(dentry
)) {
2628 mode
= cgroup_file_mode(cft
);
2629 error
= cgroup_create_file(dentry
, mode
| S_IFREG
,
2632 dentry
->d_fsdata
= (void *)cft
;
2635 error
= PTR_ERR(dentry
);
2638 EXPORT_SYMBOL_GPL(cgroup_add_file
);
2640 int cgroup_add_files(struct cgroup
*cgrp
,
2641 struct cgroup_subsys
*subsys
,
2642 const struct cftype cft
[],
2646 for (i
= 0; i
< count
; i
++) {
2647 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
2653 EXPORT_SYMBOL_GPL(cgroup_add_files
);
2656 * cgroup_task_count - count the number of tasks in a cgroup.
2657 * @cgrp: the cgroup in question
2659 * Return the number of tasks in the cgroup.
2661 int cgroup_task_count(const struct cgroup
*cgrp
)
2664 struct cg_cgroup_link
*link
;
2666 read_lock(&css_set_lock
);
2667 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2668 count
+= atomic_read(&link
->cg
->refcount
);
2670 read_unlock(&css_set_lock
);
2675 * Advance a list_head iterator. The iterator should be positioned at
2676 * the start of a css_set
2678 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2679 struct cgroup_iter
*it
)
2681 struct list_head
*l
= it
->cg_link
;
2682 struct cg_cgroup_link
*link
;
2685 /* Advance to the next non-empty css_set */
2688 if (l
== &cgrp
->css_sets
) {
2692 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2694 } while (list_empty(&cg
->tasks
));
2696 it
->task
= cg
->tasks
.next
;
2700 * To reduce the fork() overhead for systems that are not actually
2701 * using their cgroups capability, we don't maintain the lists running
2702 * through each css_set to its tasks until we see the list actually
2703 * used - in other words after the first call to cgroup_iter_start().
2705 static void cgroup_enable_task_cg_lists(void)
2707 struct task_struct
*p
, *g
;
2708 write_lock(&css_set_lock
);
2709 use_task_css_set_links
= 1;
2711 * We need tasklist_lock because RCU is not safe against
2712 * while_each_thread(). Besides, a forking task that has passed
2713 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2714 * is not guaranteed to have its child immediately visible in the
2715 * tasklist if we walk through it with RCU.
2717 read_lock(&tasklist_lock
);
2718 do_each_thread(g
, p
) {
2721 * We should check if the process is exiting, otherwise
2722 * it will race with cgroup_exit() in that the list
2723 * entry won't be deleted though the process has exited.
2725 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2726 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2728 } while_each_thread(g
, p
);
2729 read_unlock(&tasklist_lock
);
2730 write_unlock(&css_set_lock
);
2733 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2734 __acquires(css_set_lock
)
2737 * The first time anyone tries to iterate across a cgroup,
2738 * we need to enable the list linking each css_set to its
2739 * tasks, and fix up all existing tasks.
2741 if (!use_task_css_set_links
)
2742 cgroup_enable_task_cg_lists();
2744 read_lock(&css_set_lock
);
2745 it
->cg_link
= &cgrp
->css_sets
;
2746 cgroup_advance_iter(cgrp
, it
);
2749 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
2750 struct cgroup_iter
*it
)
2752 struct task_struct
*res
;
2753 struct list_head
*l
= it
->task
;
2754 struct cg_cgroup_link
*link
;
2756 /* If the iterator cg is NULL, we have no tasks */
2759 res
= list_entry(l
, struct task_struct
, cg_list
);
2760 /* Advance iterator to find next entry */
2762 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
2763 if (l
== &link
->cg
->tasks
) {
2764 /* We reached the end of this task list - move on to
2765 * the next cg_cgroup_link */
2766 cgroup_advance_iter(cgrp
, it
);
2773 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2774 __releases(css_set_lock
)
2776 read_unlock(&css_set_lock
);
2779 static inline int started_after_time(struct task_struct
*t1
,
2780 struct timespec
*time
,
2781 struct task_struct
*t2
)
2783 int start_diff
= timespec_compare(&t1
->start_time
, time
);
2784 if (start_diff
> 0) {
2786 } else if (start_diff
< 0) {
2790 * Arbitrarily, if two processes started at the same
2791 * time, we'll say that the lower pointer value
2792 * started first. Note that t2 may have exited by now
2793 * so this may not be a valid pointer any longer, but
2794 * that's fine - it still serves to distinguish
2795 * between two tasks started (effectively) simultaneously.
2802 * This function is a callback from heap_insert() and is used to order
2804 * In this case we order the heap in descending task start time.
2806 static inline int started_after(void *p1
, void *p2
)
2808 struct task_struct
*t1
= p1
;
2809 struct task_struct
*t2
= p2
;
2810 return started_after_time(t1
, &t2
->start_time
, t2
);
2814 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2815 * @scan: struct cgroup_scanner containing arguments for the scan
2817 * Arguments include pointers to callback functions test_task() and
2819 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2820 * and if it returns true, call process_task() for it also.
2821 * The test_task pointer may be NULL, meaning always true (select all tasks).
2822 * Effectively duplicates cgroup_iter_{start,next,end}()
2823 * but does not lock css_set_lock for the call to process_task().
2824 * The struct cgroup_scanner may be embedded in any structure of the caller's
2826 * It is guaranteed that process_task() will act on every task that
2827 * is a member of the cgroup for the duration of this call. This
2828 * function may or may not call process_task() for tasks that exit
2829 * or move to a different cgroup during the call, or are forked or
2830 * move into the cgroup during the call.
2832 * Note that test_task() may be called with locks held, and may in some
2833 * situations be called multiple times for the same task, so it should
2835 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2836 * pre-allocated and will be used for heap operations (and its "gt" member will
2837 * be overwritten), else a temporary heap will be used (allocation of which
2838 * may cause this function to fail).
2840 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
2843 struct cgroup_iter it
;
2844 struct task_struct
*p
, *dropped
;
2845 /* Never dereference latest_task, since it's not refcounted */
2846 struct task_struct
*latest_task
= NULL
;
2847 struct ptr_heap tmp_heap
;
2848 struct ptr_heap
*heap
;
2849 struct timespec latest_time
= { 0, 0 };
2852 /* The caller supplied our heap and pre-allocated its memory */
2854 heap
->gt
= &started_after
;
2856 /* We need to allocate our own heap memory */
2858 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
2860 /* cannot allocate the heap */
2866 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2867 * to determine which are of interest, and using the scanner's
2868 * "process_task" callback to process any of them that need an update.
2869 * Since we don't want to hold any locks during the task updates,
2870 * gather tasks to be processed in a heap structure.
2871 * The heap is sorted by descending task start time.
2872 * If the statically-sized heap fills up, we overflow tasks that
2873 * started later, and in future iterations only consider tasks that
2874 * started after the latest task in the previous pass. This
2875 * guarantees forward progress and that we don't miss any tasks.
2878 cgroup_iter_start(scan
->cg
, &it
);
2879 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
2881 * Only affect tasks that qualify per the caller's callback,
2882 * if he provided one
2884 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
2887 * Only process tasks that started after the last task
2890 if (!started_after_time(p
, &latest_time
, latest_task
))
2892 dropped
= heap_insert(heap
, p
);
2893 if (dropped
== NULL
) {
2895 * The new task was inserted; the heap wasn't
2899 } else if (dropped
!= p
) {
2901 * The new task was inserted, and pushed out a
2905 put_task_struct(dropped
);
2908 * Else the new task was newer than anything already in
2909 * the heap and wasn't inserted
2912 cgroup_iter_end(scan
->cg
, &it
);
2915 for (i
= 0; i
< heap
->size
; i
++) {
2916 struct task_struct
*q
= heap
->ptrs
[i
];
2918 latest_time
= q
->start_time
;
2921 /* Process the task per the caller's callback */
2922 scan
->process_task(q
, scan
);
2926 * If we had to process any tasks at all, scan again
2927 * in case some of them were in the middle of forking
2928 * children that didn't get processed.
2929 * Not the most efficient way to do it, but it avoids
2930 * having to take callback_mutex in the fork path
2934 if (heap
== &tmp_heap
)
2935 heap_free(&tmp_heap
);
2940 * Stuff for reading the 'tasks'/'procs' files.
2942 * Reading this file can return large amounts of data if a cgroup has
2943 * *lots* of attached tasks. So it may need several calls to read(),
2944 * but we cannot guarantee that the information we produce is correct
2945 * unless we produce it entirely atomically.
2949 /* which pidlist file are we talking about? */
2950 enum cgroup_filetype
{
2956 * A pidlist is a list of pids that virtually represents the contents of one
2957 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
2958 * a pair (one each for procs, tasks) for each pid namespace that's relevant
2961 struct cgroup_pidlist
{
2963 * used to find which pidlist is wanted. doesn't change as long as
2964 * this particular list stays in the list.
2966 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
2969 /* how many elements the above list has */
2971 /* how many files are using the current array */
2973 /* each of these stored in a list by its cgroup */
2974 struct list_head links
;
2975 /* pointer to the cgroup we belong to, for list removal purposes */
2976 struct cgroup
*owner
;
2977 /* protects the other fields */
2978 struct rw_semaphore mutex
;
2982 * The following two functions "fix" the issue where there are more pids
2983 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2984 * TODO: replace with a kernel-wide solution to this problem
2986 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2987 static void *pidlist_allocate(int count
)
2989 if (PIDLIST_TOO_LARGE(count
))
2990 return vmalloc(count
* sizeof(pid_t
));
2992 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
2994 static void pidlist_free(void *p
)
2996 if (is_vmalloc_addr(p
))
3001 static void *pidlist_resize(void *p
, int newcount
)
3004 /* note: if new alloc fails, old p will still be valid either way */
3005 if (is_vmalloc_addr(p
)) {
3006 newlist
= vmalloc(newcount
* sizeof(pid_t
));
3009 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
3012 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
3018 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3019 * If the new stripped list is sufficiently smaller and there's enough memory
3020 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3021 * number of unique elements.
3023 /* is the size difference enough that we should re-allocate the array? */
3024 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3025 static int pidlist_uniq(pid_t
**p
, int length
)
3032 * we presume the 0th element is unique, so i starts at 1. trivial
3033 * edge cases first; no work needs to be done for either
3035 if (length
== 0 || length
== 1)
3037 /* src and dest walk down the list; dest counts unique elements */
3038 for (src
= 1; src
< length
; src
++) {
3039 /* find next unique element */
3040 while (list
[src
] == list
[src
-1]) {
3045 /* dest always points to where the next unique element goes */
3046 list
[dest
] = list
[src
];
3051 * if the length difference is large enough, we want to allocate a
3052 * smaller buffer to save memory. if this fails due to out of memory,
3053 * we'll just stay with what we've got.
3055 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
3056 newlist
= pidlist_resize(list
, dest
);
3063 static int cmppid(const void *a
, const void *b
)
3065 return *(pid_t
*)a
- *(pid_t
*)b
;
3069 * find the appropriate pidlist for our purpose (given procs vs tasks)
3070 * returns with the lock on that pidlist already held, and takes care
3071 * of the use count, or returns NULL with no locks held if we're out of
3074 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3075 enum cgroup_filetype type
)
3077 struct cgroup_pidlist
*l
;
3078 /* don't need task_nsproxy() if we're looking at ourself */
3079 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
3082 * We can't drop the pidlist_mutex before taking the l->mutex in case
3083 * the last ref-holder is trying to remove l from the list at the same
3084 * time. Holding the pidlist_mutex precludes somebody taking whichever
3085 * list we find out from under us - compare release_pid_array().
3087 mutex_lock(&cgrp
->pidlist_mutex
);
3088 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3089 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3090 /* make sure l doesn't vanish out from under us */
3091 down_write(&l
->mutex
);
3092 mutex_unlock(&cgrp
->pidlist_mutex
);
3096 /* entry not found; create a new one */
3097 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3099 mutex_unlock(&cgrp
->pidlist_mutex
);
3102 init_rwsem(&l
->mutex
);
3103 down_write(&l
->mutex
);
3105 l
->key
.ns
= get_pid_ns(ns
);
3106 l
->use_count
= 0; /* don't increment here */
3109 list_add(&l
->links
, &cgrp
->pidlists
);
3110 mutex_unlock(&cgrp
->pidlist_mutex
);
3115 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3117 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3118 struct cgroup_pidlist
**lp
)
3122 int pid
, n
= 0; /* used for populating the array */
3123 struct cgroup_iter it
;
3124 struct task_struct
*tsk
;
3125 struct cgroup_pidlist
*l
;
3128 * If cgroup gets more users after we read count, we won't have
3129 * enough space - tough. This race is indistinguishable to the
3130 * caller from the case that the additional cgroup users didn't
3131 * show up until sometime later on.
3133 length
= cgroup_task_count(cgrp
);
3134 array
= pidlist_allocate(length
);
3137 /* now, populate the array */
3138 cgroup_iter_start(cgrp
, &it
);
3139 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3140 if (unlikely(n
== length
))
3142 /* get tgid or pid for procs or tasks file respectively */
3143 if (type
== CGROUP_FILE_PROCS
)
3144 pid
= task_tgid_vnr(tsk
);
3146 pid
= task_pid_vnr(tsk
);
3147 if (pid
> 0) /* make sure to only use valid results */
3150 cgroup_iter_end(cgrp
, &it
);
3152 /* now sort & (if procs) strip out duplicates */
3153 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3154 if (type
== CGROUP_FILE_PROCS
)
3155 length
= pidlist_uniq(&array
, length
);
3156 l
= cgroup_pidlist_find(cgrp
, type
);
3158 pidlist_free(array
);
3161 /* store array, freeing old if necessary - lock already held */
3162 pidlist_free(l
->list
);
3166 up_write(&l
->mutex
);
3172 * cgroupstats_build - build and fill cgroupstats
3173 * @stats: cgroupstats to fill information into
3174 * @dentry: A dentry entry belonging to the cgroup for which stats have
3177 * Build and fill cgroupstats so that taskstats can export it to user
3180 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3183 struct cgroup
*cgrp
;
3184 struct cgroup_iter it
;
3185 struct task_struct
*tsk
;
3188 * Validate dentry by checking the superblock operations,
3189 * and make sure it's a directory.
3191 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3192 !S_ISDIR(dentry
->d_inode
->i_mode
))
3196 cgrp
= dentry
->d_fsdata
;
3198 cgroup_iter_start(cgrp
, &it
);
3199 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3200 switch (tsk
->state
) {
3202 stats
->nr_running
++;
3204 case TASK_INTERRUPTIBLE
:
3205 stats
->nr_sleeping
++;
3207 case TASK_UNINTERRUPTIBLE
:
3208 stats
->nr_uninterruptible
++;
3211 stats
->nr_stopped
++;
3214 if (delayacct_is_task_waiting_on_io(tsk
))
3215 stats
->nr_io_wait
++;
3219 cgroup_iter_end(cgrp
, &it
);
3227 * seq_file methods for the tasks/procs files. The seq_file position is the
3228 * next pid to display; the seq_file iterator is a pointer to the pid
3229 * in the cgroup->l->list array.
3232 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3235 * Initially we receive a position value that corresponds to
3236 * one more than the last pid shown (or 0 on the first call or
3237 * after a seek to the start). Use a binary-search to find the
3238 * next pid to display, if any
3240 struct cgroup_pidlist
*l
= s
->private;
3241 int index
= 0, pid
= *pos
;
3244 down_read(&l
->mutex
);
3246 int end
= l
->length
;
3248 while (index
< end
) {
3249 int mid
= (index
+ end
) / 2;
3250 if (l
->list
[mid
] == pid
) {
3253 } else if (l
->list
[mid
] <= pid
)
3259 /* If we're off the end of the array, we're done */
3260 if (index
>= l
->length
)
3262 /* Update the abstract position to be the actual pid that we found */
3263 iter
= l
->list
+ index
;
3268 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3270 struct cgroup_pidlist
*l
= s
->private;
3274 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3276 struct cgroup_pidlist
*l
= s
->private;
3278 pid_t
*end
= l
->list
+ l
->length
;
3280 * Advance to the next pid in the array. If this goes off the
3292 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3294 return seq_printf(s
, "%d\n", *(int *)v
);
3298 * seq_operations functions for iterating on pidlists through seq_file -
3299 * independent of whether it's tasks or procs
3301 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3302 .start
= cgroup_pidlist_start
,
3303 .stop
= cgroup_pidlist_stop
,
3304 .next
= cgroup_pidlist_next
,
3305 .show
= cgroup_pidlist_show
,
3308 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3311 * the case where we're the last user of this particular pidlist will
3312 * have us remove it from the cgroup's list, which entails taking the
3313 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3314 * pidlist_mutex, we have to take pidlist_mutex first.
3316 mutex_lock(&l
->owner
->pidlist_mutex
);
3317 down_write(&l
->mutex
);
3318 BUG_ON(!l
->use_count
);
3319 if (!--l
->use_count
) {
3320 /* we're the last user if refcount is 0; remove and free */
3321 list_del(&l
->links
);
3322 mutex_unlock(&l
->owner
->pidlist_mutex
);
3323 pidlist_free(l
->list
);
3324 put_pid_ns(l
->key
.ns
);
3325 up_write(&l
->mutex
);
3329 mutex_unlock(&l
->owner
->pidlist_mutex
);
3330 up_write(&l
->mutex
);
3333 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3335 struct cgroup_pidlist
*l
;
3336 if (!(file
->f_mode
& FMODE_READ
))
3339 * the seq_file will only be initialized if the file was opened for
3340 * reading; hence we check if it's not null only in that case.
3342 l
= ((struct seq_file
*)file
->private_data
)->private;
3343 cgroup_release_pid_array(l
);
3344 return seq_release(inode
, file
);
3347 static const struct file_operations cgroup_pidlist_operations
= {
3349 .llseek
= seq_lseek
,
3350 .write
= cgroup_file_write
,
3351 .release
= cgroup_pidlist_release
,
3355 * The following functions handle opens on a file that displays a pidlist
3356 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3359 /* helper function for the two below it */
3360 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3362 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3363 struct cgroup_pidlist
*l
;
3366 /* Nothing to do for write-only files */
3367 if (!(file
->f_mode
& FMODE_READ
))
3370 /* have the array populated */
3371 retval
= pidlist_array_load(cgrp
, type
, &l
);
3374 /* configure file information */
3375 file
->f_op
= &cgroup_pidlist_operations
;
3377 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3379 cgroup_release_pid_array(l
);
3382 ((struct seq_file
*)file
->private_data
)->private = l
;
3385 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3387 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3389 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3391 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3394 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3397 return notify_on_release(cgrp
);
3400 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3404 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3406 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3408 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3413 * Unregister event and free resources.
3415 * Gets called from workqueue.
3417 static void cgroup_event_remove(struct work_struct
*work
)
3419 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3421 struct cgroup
*cgrp
= event
->cgrp
;
3423 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3425 eventfd_ctx_put(event
->eventfd
);
3431 * Gets called on POLLHUP on eventfd when user closes it.
3433 * Called with wqh->lock held and interrupts disabled.
3435 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3436 int sync
, void *key
)
3438 struct cgroup_event
*event
= container_of(wait
,
3439 struct cgroup_event
, wait
);
3440 struct cgroup
*cgrp
= event
->cgrp
;
3441 unsigned long flags
= (unsigned long)key
;
3443 if (flags
& POLLHUP
) {
3444 __remove_wait_queue(event
->wqh
, &event
->wait
);
3445 spin_lock(&cgrp
->event_list_lock
);
3446 list_del(&event
->list
);
3447 spin_unlock(&cgrp
->event_list_lock
);
3449 * We are in atomic context, but cgroup_event_remove() may
3450 * sleep, so we have to call it in workqueue.
3452 schedule_work(&event
->remove
);
3458 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3459 wait_queue_head_t
*wqh
, poll_table
*pt
)
3461 struct cgroup_event
*event
= container_of(pt
,
3462 struct cgroup_event
, pt
);
3465 add_wait_queue(wqh
, &event
->wait
);
3469 * Parse input and register new cgroup event handler.
3471 * Input must be in format '<event_fd> <control_fd> <args>'.
3472 * Interpretation of args is defined by control file implementation.
3474 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3477 struct cgroup_event
*event
= NULL
;
3478 unsigned int efd
, cfd
;
3479 struct file
*efile
= NULL
;
3480 struct file
*cfile
= NULL
;
3484 efd
= simple_strtoul(buffer
, &endp
, 10);
3489 cfd
= simple_strtoul(buffer
, &endp
, 10);
3490 if ((*endp
!= ' ') && (*endp
!= '\0'))
3494 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3498 INIT_LIST_HEAD(&event
->list
);
3499 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3500 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3501 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3503 efile
= eventfd_fget(efd
);
3504 if (IS_ERR(efile
)) {
3505 ret
= PTR_ERR(efile
);
3509 event
->eventfd
= eventfd_ctx_fileget(efile
);
3510 if (IS_ERR(event
->eventfd
)) {
3511 ret
= PTR_ERR(event
->eventfd
);
3521 /* the process need read permission on control file */
3522 /* AV: shouldn't we check that it's been opened for read instead? */
3523 ret
= inode_permission(cfile
->f_path
.dentry
->d_inode
, MAY_READ
);
3527 event
->cft
= __file_cft(cfile
);
3528 if (IS_ERR(event
->cft
)) {
3529 ret
= PTR_ERR(event
->cft
);
3533 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3538 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3539 event
->eventfd
, buffer
);
3543 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3544 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3550 * Events should be removed after rmdir of cgroup directory, but before
3551 * destroying subsystem state objects. Let's take reference to cgroup
3552 * directory dentry to do that.
3556 spin_lock(&cgrp
->event_list_lock
);
3557 list_add(&event
->list
, &cgrp
->event_list
);
3558 spin_unlock(&cgrp
->event_list_lock
);
3569 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3570 eventfd_ctx_put(event
->eventfd
);
3572 if (!IS_ERR_OR_NULL(efile
))
3580 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3583 return clone_children(cgrp
);
3586 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3591 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3593 clear_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3598 * for the common functions, 'private' gives the type of file
3600 /* for hysterical raisins, we can't put this on the older files */
3601 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3602 static struct cftype files
[] = {
3605 .open
= cgroup_tasks_open
,
3606 .write_u64
= cgroup_tasks_write
,
3607 .release
= cgroup_pidlist_release
,
3608 .mode
= S_IRUGO
| S_IWUSR
,
3611 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3612 .open
= cgroup_procs_open
,
3613 .write_u64
= cgroup_procs_write
,
3614 .release
= cgroup_pidlist_release
,
3615 .mode
= S_IRUGO
| S_IWUSR
,
3618 .name
= "notify_on_release",
3619 .read_u64
= cgroup_read_notify_on_release
,
3620 .write_u64
= cgroup_write_notify_on_release
,
3623 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3624 .write_string
= cgroup_write_event_control
,
3628 .name
= "cgroup.clone_children",
3629 .read_u64
= cgroup_clone_children_read
,
3630 .write_u64
= cgroup_clone_children_write
,
3634 static struct cftype cft_release_agent
= {
3635 .name
= "release_agent",
3636 .read_seq_string
= cgroup_release_agent_show
,
3637 .write_string
= cgroup_release_agent_write
,
3638 .max_write_len
= PATH_MAX
,
3641 static int cgroup_populate_dir(struct cgroup
*cgrp
)
3644 struct cgroup_subsys
*ss
;
3646 /* First clear out any existing files */
3647 cgroup_clear_directory(cgrp
->dentry
);
3649 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
3653 if (cgrp
== cgrp
->top_cgroup
) {
3654 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
3658 for_each_subsys(cgrp
->root
, ss
) {
3659 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
3662 /* This cgroup is ready now */
3663 for_each_subsys(cgrp
->root
, ss
) {
3664 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3666 * Update id->css pointer and make this css visible from
3667 * CSS ID functions. This pointer will be dereferened
3668 * from RCU-read-side without locks.
3671 rcu_assign_pointer(css
->id
->css
, css
);
3677 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
3678 struct cgroup_subsys
*ss
,
3679 struct cgroup
*cgrp
)
3682 atomic_set(&css
->refcnt
, 1);
3685 if (cgrp
== dummytop
)
3686 set_bit(CSS_ROOT
, &css
->flags
);
3687 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
3688 cgrp
->subsys
[ss
->subsys_id
] = css
;
3691 static void cgroup_lock_hierarchy(struct cgroupfs_root
*root
)
3693 /* We need to take each hierarchy_mutex in a consistent order */
3697 * No worry about a race with rebind_subsystems that might mess up the
3698 * locking order, since both parties are under cgroup_mutex.
3700 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3701 struct cgroup_subsys
*ss
= subsys
[i
];
3704 if (ss
->root
== root
)
3705 mutex_lock(&ss
->hierarchy_mutex
);
3709 static void cgroup_unlock_hierarchy(struct cgroupfs_root
*root
)
3713 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3714 struct cgroup_subsys
*ss
= subsys
[i
];
3717 if (ss
->root
== root
)
3718 mutex_unlock(&ss
->hierarchy_mutex
);
3723 * cgroup_create - create a cgroup
3724 * @parent: cgroup that will be parent of the new cgroup
3725 * @dentry: dentry of the new cgroup
3726 * @mode: mode to set on new inode
3728 * Must be called with the mutex on the parent inode held
3730 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
3733 struct cgroup
*cgrp
;
3734 struct cgroupfs_root
*root
= parent
->root
;
3736 struct cgroup_subsys
*ss
;
3737 struct super_block
*sb
= root
->sb
;
3739 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3743 /* Grab a reference on the superblock so the hierarchy doesn't
3744 * get deleted on unmount if there are child cgroups. This
3745 * can be done outside cgroup_mutex, since the sb can't
3746 * disappear while someone has an open control file on the
3748 atomic_inc(&sb
->s_active
);
3750 mutex_lock(&cgroup_mutex
);
3752 init_cgroup_housekeeping(cgrp
);
3754 cgrp
->parent
= parent
;
3755 cgrp
->root
= parent
->root
;
3756 cgrp
->top_cgroup
= parent
->top_cgroup
;
3758 if (notify_on_release(parent
))
3759 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3761 if (clone_children(parent
))
3762 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3764 for_each_subsys(root
, ss
) {
3765 struct cgroup_subsys_state
*css
= ss
->create(cgrp
);
3771 init_cgroup_css(css
, ss
, cgrp
);
3773 err
= alloc_css_id(ss
, parent
, cgrp
);
3777 /* At error, ->destroy() callback has to free assigned ID. */
3778 if (clone_children(parent
) && ss
->post_clone
)
3779 ss
->post_clone(cgrp
);
3782 cgroup_lock_hierarchy(root
);
3783 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
3784 cgroup_unlock_hierarchy(root
);
3785 root
->number_of_cgroups
++;
3787 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
3791 /* The cgroup directory was pre-locked for us */
3792 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
3794 err
= cgroup_populate_dir(cgrp
);
3795 /* If err < 0, we have a half-filled directory - oh well ;) */
3797 mutex_unlock(&cgroup_mutex
);
3798 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
3804 cgroup_lock_hierarchy(root
);
3805 list_del(&cgrp
->sibling
);
3806 cgroup_unlock_hierarchy(root
);
3807 root
->number_of_cgroups
--;
3811 for_each_subsys(root
, ss
) {
3812 if (cgrp
->subsys
[ss
->subsys_id
])
3816 mutex_unlock(&cgroup_mutex
);
3818 /* Release the reference count that we took on the superblock */
3819 deactivate_super(sb
);
3825 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3827 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
3829 /* the vfs holds inode->i_mutex already */
3830 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
3833 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
3835 /* Check the reference count on each subsystem. Since we
3836 * already established that there are no tasks in the
3837 * cgroup, if the css refcount is also 1, then there should
3838 * be no outstanding references, so the subsystem is safe to
3839 * destroy. We scan across all subsystems rather than using
3840 * the per-hierarchy linked list of mounted subsystems since
3841 * we can be called via check_for_release() with no
3842 * synchronization other than RCU, and the subsystem linked
3843 * list isn't RCU-safe */
3846 * We won't need to lock the subsys array, because the subsystems
3847 * we're concerned about aren't going anywhere since our cgroup root
3848 * has a reference on them.
3850 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3851 struct cgroup_subsys
*ss
= subsys
[i
];
3852 struct cgroup_subsys_state
*css
;
3853 /* Skip subsystems not present or not in this hierarchy */
3854 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
3856 css
= cgrp
->subsys
[ss
->subsys_id
];
3857 /* When called from check_for_release() it's possible
3858 * that by this point the cgroup has been removed
3859 * and the css deleted. But a false-positive doesn't
3860 * matter, since it can only happen if the cgroup
3861 * has been deleted and hence no longer needs the
3862 * release agent to be called anyway. */
3863 if (css
&& (atomic_read(&css
->refcnt
) > 1))
3870 * Atomically mark all (or else none) of the cgroup's CSS objects as
3871 * CSS_REMOVED. Return true on success, or false if the cgroup has
3872 * busy subsystems. Call with cgroup_mutex held
3875 static int cgroup_clear_css_refs(struct cgroup
*cgrp
)
3877 struct cgroup_subsys
*ss
;
3878 unsigned long flags
;
3879 bool failed
= false;
3880 local_irq_save(flags
);
3881 for_each_subsys(cgrp
->root
, ss
) {
3882 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3885 /* We can only remove a CSS with a refcnt==1 */
3886 refcnt
= atomic_read(&css
->refcnt
);
3893 * Drop the refcnt to 0 while we check other
3894 * subsystems. This will cause any racing
3895 * css_tryget() to spin until we set the
3896 * CSS_REMOVED bits or abort
3898 if (atomic_cmpxchg(&css
->refcnt
, refcnt
, 0) == refcnt
)
3904 for_each_subsys(cgrp
->root
, ss
) {
3905 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3908 * Restore old refcnt if we previously managed
3909 * to clear it from 1 to 0
3911 if (!atomic_read(&css
->refcnt
))
3912 atomic_set(&css
->refcnt
, 1);
3914 /* Commit the fact that the CSS is removed */
3915 set_bit(CSS_REMOVED
, &css
->flags
);
3918 local_irq_restore(flags
);
3922 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
3924 struct cgroup
*cgrp
= dentry
->d_fsdata
;
3926 struct cgroup
*parent
;
3928 struct cgroup_event
*event
, *tmp
;
3931 /* the vfs holds both inode->i_mutex already */
3933 mutex_lock(&cgroup_mutex
);
3934 if (atomic_read(&cgrp
->count
) != 0) {
3935 mutex_unlock(&cgroup_mutex
);
3938 if (!list_empty(&cgrp
->children
)) {
3939 mutex_unlock(&cgroup_mutex
);
3942 mutex_unlock(&cgroup_mutex
);
3945 * In general, subsystem has no css->refcnt after pre_destroy(). But
3946 * in racy cases, subsystem may have to get css->refcnt after
3947 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3948 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3949 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3950 * and subsystem's reference count handling. Please see css_get/put
3951 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3953 set_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3956 * Call pre_destroy handlers of subsys. Notify subsystems
3957 * that rmdir() request comes.
3959 ret
= cgroup_call_pre_destroy(cgrp
);
3961 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3965 mutex_lock(&cgroup_mutex
);
3966 parent
= cgrp
->parent
;
3967 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
)) {
3968 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3969 mutex_unlock(&cgroup_mutex
);
3972 prepare_to_wait(&cgroup_rmdir_waitq
, &wait
, TASK_INTERRUPTIBLE
);
3973 if (!cgroup_clear_css_refs(cgrp
)) {
3974 mutex_unlock(&cgroup_mutex
);
3976 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3977 * prepare_to_wait(), we need to check this flag.
3979 if (test_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
))
3981 finish_wait(&cgroup_rmdir_waitq
, &wait
);
3982 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3983 if (signal_pending(current
))
3987 /* NO css_tryget() can success after here. */
3988 finish_wait(&cgroup_rmdir_waitq
, &wait
);
3989 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3991 raw_spin_lock(&release_list_lock
);
3992 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
3993 if (!list_empty(&cgrp
->release_list
))
3994 list_del_init(&cgrp
->release_list
);
3995 raw_spin_unlock(&release_list_lock
);
3997 cgroup_lock_hierarchy(cgrp
->root
);
3998 /* delete this cgroup from parent->children */
3999 list_del_init(&cgrp
->sibling
);
4000 cgroup_unlock_hierarchy(cgrp
->root
);
4002 d
= dget(cgrp
->dentry
);
4004 cgroup_d_remove_dir(d
);
4007 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4008 check_for_release(parent
);
4011 * Unregister events and notify userspace.
4012 * Notify userspace about cgroup removing only after rmdir of cgroup
4013 * directory to avoid race between userspace and kernelspace
4015 spin_lock(&cgrp
->event_list_lock
);
4016 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4017 list_del(&event
->list
);
4018 remove_wait_queue(event
->wqh
, &event
->wait
);
4019 eventfd_signal(event
->eventfd
, 1);
4020 schedule_work(&event
->remove
);
4022 spin_unlock(&cgrp
->event_list_lock
);
4024 mutex_unlock(&cgroup_mutex
);
4028 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4030 struct cgroup_subsys_state
*css
;
4032 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4034 /* Create the top cgroup state for this subsystem */
4035 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4036 ss
->root
= &rootnode
;
4037 css
= ss
->create(dummytop
);
4038 /* We don't handle early failures gracefully */
4039 BUG_ON(IS_ERR(css
));
4040 init_cgroup_css(css
, ss
, dummytop
);
4042 /* Update the init_css_set to contain a subsys
4043 * pointer to this state - since the subsystem is
4044 * newly registered, all tasks and hence the
4045 * init_css_set is in the subsystem's top cgroup. */
4046 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
4048 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4050 /* At system boot, before all subsystems have been
4051 * registered, no tasks have been forked, so we don't
4052 * need to invoke fork callbacks here. */
4053 BUG_ON(!list_empty(&init_task
.tasks
));
4055 mutex_init(&ss
->hierarchy_mutex
);
4056 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4059 /* this function shouldn't be used with modular subsystems, since they
4060 * need to register a subsys_id, among other things */
4065 * cgroup_load_subsys: load and register a modular subsystem at runtime
4066 * @ss: the subsystem to load
4068 * This function should be called in a modular subsystem's initcall. If the
4069 * subsystem is built as a module, it will be assigned a new subsys_id and set
4070 * up for use. If the subsystem is built-in anyway, work is delegated to the
4071 * simpler cgroup_init_subsys.
4073 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4076 struct cgroup_subsys_state
*css
;
4078 /* check name and function validity */
4079 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4080 ss
->create
== NULL
|| ss
->destroy
== NULL
)
4084 * we don't support callbacks in modular subsystems. this check is
4085 * before the ss->module check for consistency; a subsystem that could
4086 * be a module should still have no callbacks even if the user isn't
4087 * compiling it as one.
4089 if (ss
->fork
|| ss
->exit
)
4093 * an optionally modular subsystem is built-in: we want to do nothing,
4094 * since cgroup_init_subsys will have already taken care of it.
4096 if (ss
->module
== NULL
) {
4097 /* a few sanity checks */
4098 BUG_ON(ss
->subsys_id
>= CGROUP_BUILTIN_SUBSYS_COUNT
);
4099 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4104 * need to register a subsys id before anything else - for example,
4105 * init_cgroup_css needs it.
4107 mutex_lock(&cgroup_mutex
);
4108 /* find the first empty slot in the array */
4109 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4110 if (subsys
[i
] == NULL
)
4113 if (i
== CGROUP_SUBSYS_COUNT
) {
4114 /* maximum number of subsystems already registered! */
4115 mutex_unlock(&cgroup_mutex
);
4118 /* assign ourselves the subsys_id */
4123 * no ss->create seems to need anything important in the ss struct, so
4124 * this can happen first (i.e. before the rootnode attachment).
4126 css
= ss
->create(dummytop
);
4128 /* failure case - need to deassign the subsys[] slot. */
4130 mutex_unlock(&cgroup_mutex
);
4131 return PTR_ERR(css
);
4134 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4135 ss
->root
= &rootnode
;
4137 /* our new subsystem will be attached to the dummy hierarchy. */
4138 init_cgroup_css(css
, ss
, dummytop
);
4139 /* init_idr must be after init_cgroup_css because it sets css->id. */
4141 int ret
= cgroup_init_idr(ss
, css
);
4143 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4144 ss
->destroy(dummytop
);
4146 mutex_unlock(&cgroup_mutex
);
4152 * Now we need to entangle the css into the existing css_sets. unlike
4153 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4154 * will need a new pointer to it; done by iterating the css_set_table.
4155 * furthermore, modifying the existing css_sets will corrupt the hash
4156 * table state, so each changed css_set will need its hash recomputed.
4157 * this is all done under the css_set_lock.
4159 write_lock(&css_set_lock
);
4160 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
4162 struct hlist_node
*node
, *tmp
;
4163 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
4165 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
4166 /* skip entries that we already rehashed */
4167 if (cg
->subsys
[ss
->subsys_id
])
4169 /* remove existing entry */
4170 hlist_del(&cg
->hlist
);
4172 cg
->subsys
[ss
->subsys_id
] = css
;
4173 /* recompute hash and restore entry */
4174 new_bucket
= css_set_hash(cg
->subsys
);
4175 hlist_add_head(&cg
->hlist
, new_bucket
);
4178 write_unlock(&css_set_lock
);
4180 mutex_init(&ss
->hierarchy_mutex
);
4181 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4185 mutex_unlock(&cgroup_mutex
);
4188 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4191 * cgroup_unload_subsys: unload a modular subsystem
4192 * @ss: the subsystem to unload
4194 * This function should be called in a modular subsystem's exitcall. When this
4195 * function is invoked, the refcount on the subsystem's module will be 0, so
4196 * the subsystem will not be attached to any hierarchy.
4198 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4200 struct cg_cgroup_link
*link
;
4201 struct hlist_head
*hhead
;
4203 BUG_ON(ss
->module
== NULL
);
4206 * we shouldn't be called if the subsystem is in use, and the use of
4207 * try_module_get in parse_cgroupfs_options should ensure that it
4208 * doesn't start being used while we're killing it off.
4210 BUG_ON(ss
->root
!= &rootnode
);
4212 mutex_lock(&cgroup_mutex
);
4213 /* deassign the subsys_id */
4214 BUG_ON(ss
->subsys_id
< CGROUP_BUILTIN_SUBSYS_COUNT
);
4215 subsys
[ss
->subsys_id
] = NULL
;
4217 /* remove subsystem from rootnode's list of subsystems */
4218 list_del_init(&ss
->sibling
);
4221 * disentangle the css from all css_sets attached to the dummytop. as
4222 * in loading, we need to pay our respects to the hashtable gods.
4224 write_lock(&css_set_lock
);
4225 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4226 struct css_set
*cg
= link
->cg
;
4228 hlist_del(&cg
->hlist
);
4229 BUG_ON(!cg
->subsys
[ss
->subsys_id
]);
4230 cg
->subsys
[ss
->subsys_id
] = NULL
;
4231 hhead
= css_set_hash(cg
->subsys
);
4232 hlist_add_head(&cg
->hlist
, hhead
);
4234 write_unlock(&css_set_lock
);
4237 * remove subsystem's css from the dummytop and free it - need to free
4238 * before marking as null because ss->destroy needs the cgrp->subsys
4239 * pointer to find their state. note that this also takes care of
4240 * freeing the css_id.
4242 ss
->destroy(dummytop
);
4243 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4245 mutex_unlock(&cgroup_mutex
);
4247 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4250 * cgroup_init_early - cgroup initialization at system boot
4252 * Initialize cgroups at system boot, and initialize any
4253 * subsystems that request early init.
4255 int __init
cgroup_init_early(void)
4258 atomic_set(&init_css_set
.refcount
, 1);
4259 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4260 INIT_LIST_HEAD(&init_css_set
.tasks
);
4261 INIT_HLIST_NODE(&init_css_set
.hlist
);
4263 init_cgroup_root(&rootnode
);
4265 init_task
.cgroups
= &init_css_set
;
4267 init_css_set_link
.cg
= &init_css_set
;
4268 init_css_set_link
.cgrp
= dummytop
;
4269 list_add(&init_css_set_link
.cgrp_link_list
,
4270 &rootnode
.top_cgroup
.css_sets
);
4271 list_add(&init_css_set_link
.cg_link_list
,
4272 &init_css_set
.cg_links
);
4274 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
4275 INIT_HLIST_HEAD(&css_set_table
[i
]);
4277 /* at bootup time, we don't worry about modular subsystems */
4278 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4279 struct cgroup_subsys
*ss
= subsys
[i
];
4282 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4283 BUG_ON(!ss
->create
);
4284 BUG_ON(!ss
->destroy
);
4285 if (ss
->subsys_id
!= i
) {
4286 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4287 ss
->name
, ss
->subsys_id
);
4292 cgroup_init_subsys(ss
);
4298 * cgroup_init - cgroup initialization
4300 * Register cgroup filesystem and /proc file, and initialize
4301 * any subsystems that didn't request early init.
4303 int __init
cgroup_init(void)
4307 struct hlist_head
*hhead
;
4309 err
= bdi_init(&cgroup_backing_dev_info
);
4313 /* at bootup time, we don't worry about modular subsystems */
4314 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4315 struct cgroup_subsys
*ss
= subsys
[i
];
4316 if (!ss
->early_init
)
4317 cgroup_init_subsys(ss
);
4319 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4322 /* Add init_css_set to the hash table */
4323 hhead
= css_set_hash(init_css_set
.subsys
);
4324 hlist_add_head(&init_css_set
.hlist
, hhead
);
4325 BUG_ON(!init_root_id(&rootnode
));
4327 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4333 err
= register_filesystem(&cgroup_fs_type
);
4335 kobject_put(cgroup_kobj
);
4339 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4343 bdi_destroy(&cgroup_backing_dev_info
);
4349 * proc_cgroup_show()
4350 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4351 * - Used for /proc/<pid>/cgroup.
4352 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4353 * doesn't really matter if tsk->cgroup changes after we read it,
4354 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4355 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4356 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4357 * cgroup to top_cgroup.
4360 /* TODO: Use a proper seq_file iterator */
4361 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4364 struct task_struct
*tsk
;
4367 struct cgroupfs_root
*root
;
4370 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4376 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4382 mutex_lock(&cgroup_mutex
);
4384 for_each_active_root(root
) {
4385 struct cgroup_subsys
*ss
;
4386 struct cgroup
*cgrp
;
4389 seq_printf(m
, "%d:", root
->hierarchy_id
);
4390 for_each_subsys(root
, ss
)
4391 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4392 if (strlen(root
->name
))
4393 seq_printf(m
, "%sname=%s", count
? "," : "",
4396 cgrp
= task_cgroup_from_root(tsk
, root
);
4397 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4405 mutex_unlock(&cgroup_mutex
);
4406 put_task_struct(tsk
);
4413 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4415 struct pid
*pid
= PROC_I(inode
)->pid
;
4416 return single_open(file
, proc_cgroup_show
, pid
);
4419 const struct file_operations proc_cgroup_operations
= {
4420 .open
= cgroup_open
,
4422 .llseek
= seq_lseek
,
4423 .release
= single_release
,
4426 /* Display information about each subsystem and each hierarchy */
4427 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4431 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4433 * ideally we don't want subsystems moving around while we do this.
4434 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4435 * subsys/hierarchy state.
4437 mutex_lock(&cgroup_mutex
);
4438 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4439 struct cgroup_subsys
*ss
= subsys
[i
];
4442 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4443 ss
->name
, ss
->root
->hierarchy_id
,
4444 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4446 mutex_unlock(&cgroup_mutex
);
4450 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4452 return single_open(file
, proc_cgroupstats_show
, NULL
);
4455 static const struct file_operations proc_cgroupstats_operations
= {
4456 .open
= cgroupstats_open
,
4458 .llseek
= seq_lseek
,
4459 .release
= single_release
,
4463 * cgroup_fork - attach newly forked task to its parents cgroup.
4464 * @child: pointer to task_struct of forking parent process.
4466 * Description: A task inherits its parent's cgroup at fork().
4468 * A pointer to the shared css_set was automatically copied in
4469 * fork.c by dup_task_struct(). However, we ignore that copy, since
4470 * it was not made under the protection of RCU, cgroup_mutex or
4471 * threadgroup_change_begin(), so it might no longer be a valid
4472 * cgroup pointer. cgroup_attach_task() might have already changed
4473 * current->cgroups, allowing the previously referenced cgroup
4474 * group to be removed and freed.
4476 * Outside the pointer validity we also need to process the css_set
4477 * inheritance between threadgoup_change_begin() and
4478 * threadgoup_change_end(), this way there is no leak in any process
4479 * wide migration performed by cgroup_attach_proc() that could otherwise
4480 * miss a thread because it is too early or too late in the fork stage.
4482 * At the point that cgroup_fork() is called, 'current' is the parent
4483 * task, and the passed argument 'child' points to the child task.
4485 void cgroup_fork(struct task_struct
*child
)
4488 * We don't need to task_lock() current because current->cgroups
4489 * can't be changed concurrently here. The parent obviously hasn't
4490 * exited and called cgroup_exit(), and we are synchronized against
4491 * cgroup migration through threadgroup_change_begin().
4493 child
->cgroups
= current
->cgroups
;
4494 get_css_set(child
->cgroups
);
4495 INIT_LIST_HEAD(&child
->cg_list
);
4499 * cgroup_fork_callbacks - run fork callbacks
4500 * @child: the new task
4502 * Called on a new task very soon before adding it to the
4503 * tasklist. No need to take any locks since no-one can
4504 * be operating on this task.
4506 void cgroup_fork_callbacks(struct task_struct
*child
)
4508 if (need_forkexit_callback
) {
4511 * forkexit callbacks are only supported for builtin
4512 * subsystems, and the builtin section of the subsys array is
4513 * immutable, so we don't need to lock the subsys array here.
4515 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4516 struct cgroup_subsys
*ss
= subsys
[i
];
4524 * cgroup_post_fork - called on a new task after adding it to the task list
4525 * @child: the task in question
4527 * Adds the task to the list running through its css_set if necessary.
4528 * Has to be after the task is visible on the task list in case we race
4529 * with the first call to cgroup_iter_start() - to guarantee that the
4530 * new task ends up on its list.
4532 void cgroup_post_fork(struct task_struct
*child
)
4535 * use_task_css_set_links is set to 1 before we walk the tasklist
4536 * under the tasklist_lock and we read it here after we added the child
4537 * to the tasklist under the tasklist_lock as well. If the child wasn't
4538 * yet in the tasklist when we walked through it from
4539 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4540 * should be visible now due to the paired locking and barriers implied
4541 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4542 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4545 if (use_task_css_set_links
) {
4546 write_lock(&css_set_lock
);
4547 if (list_empty(&child
->cg_list
)) {
4549 * It's safe to use child->cgroups without task_lock()
4550 * here because we are protected through
4551 * threadgroup_change_begin() against concurrent
4552 * css_set change in cgroup_task_migrate(). Also
4553 * the task can't exit at that point until
4554 * wake_up_new_task() is called, so we are protected
4555 * against cgroup_exit() setting child->cgroup to
4558 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4560 write_unlock(&css_set_lock
);
4564 * cgroup_exit - detach cgroup from exiting task
4565 * @tsk: pointer to task_struct of exiting process
4566 * @run_callback: run exit callbacks?
4568 * Description: Detach cgroup from @tsk and release it.
4570 * Note that cgroups marked notify_on_release force every task in
4571 * them to take the global cgroup_mutex mutex when exiting.
4572 * This could impact scaling on very large systems. Be reluctant to
4573 * use notify_on_release cgroups where very high task exit scaling
4574 * is required on large systems.
4576 * the_top_cgroup_hack:
4578 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4580 * We call cgroup_exit() while the task is still competent to
4581 * handle notify_on_release(), then leave the task attached to the
4582 * root cgroup in each hierarchy for the remainder of its exit.
4584 * To do this properly, we would increment the reference count on
4585 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4586 * code we would add a second cgroup function call, to drop that
4587 * reference. This would just create an unnecessary hot spot on
4588 * the top_cgroup reference count, to no avail.
4590 * Normally, holding a reference to a cgroup without bumping its
4591 * count is unsafe. The cgroup could go away, or someone could
4592 * attach us to a different cgroup, decrementing the count on
4593 * the first cgroup that we never incremented. But in this case,
4594 * top_cgroup isn't going away, and either task has PF_EXITING set,
4595 * which wards off any cgroup_attach_task() attempts, or task is a failed
4596 * fork, never visible to cgroup_attach_task.
4598 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4604 * Unlink from the css_set task list if necessary.
4605 * Optimistically check cg_list before taking
4608 if (!list_empty(&tsk
->cg_list
)) {
4609 write_lock(&css_set_lock
);
4610 if (!list_empty(&tsk
->cg_list
))
4611 list_del_init(&tsk
->cg_list
);
4612 write_unlock(&css_set_lock
);
4615 /* Reassign the task to the init_css_set. */
4618 tsk
->cgroups
= &init_css_set
;
4620 if (run_callbacks
&& need_forkexit_callback
) {
4622 * modular subsystems can't use callbacks, so no need to lock
4625 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4626 struct cgroup_subsys
*ss
= subsys
[i
];
4628 struct cgroup
*old_cgrp
=
4629 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4630 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4631 ss
->exit(cgrp
, old_cgrp
, tsk
);
4638 put_css_set_taskexit(cg
);
4642 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4643 * @cgrp: the cgroup in question
4644 * @task: the task in question
4646 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4649 * If we are sending in dummytop, then presumably we are creating
4650 * the top cgroup in the subsystem.
4652 * Called only by the ns (nsproxy) cgroup.
4654 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
4657 struct cgroup
*target
;
4659 if (cgrp
== dummytop
)
4662 target
= task_cgroup_from_root(task
, cgrp
->root
);
4663 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
4664 cgrp
= cgrp
->parent
;
4665 ret
= (cgrp
== target
);
4669 static void check_for_release(struct cgroup
*cgrp
)
4671 /* All of these checks rely on RCU to keep the cgroup
4672 * structure alive */
4673 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
4674 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
4675 /* Control Group is currently removeable. If it's not
4676 * already queued for a userspace notification, queue
4678 int need_schedule_work
= 0;
4679 raw_spin_lock(&release_list_lock
);
4680 if (!cgroup_is_removed(cgrp
) &&
4681 list_empty(&cgrp
->release_list
)) {
4682 list_add(&cgrp
->release_list
, &release_list
);
4683 need_schedule_work
= 1;
4685 raw_spin_unlock(&release_list_lock
);
4686 if (need_schedule_work
)
4687 schedule_work(&release_agent_work
);
4691 /* Caller must verify that the css is not for root cgroup */
4692 void __css_put(struct cgroup_subsys_state
*css
, int count
)
4694 struct cgroup
*cgrp
= css
->cgroup
;
4697 val
= atomic_sub_return(count
, &css
->refcnt
);
4699 if (notify_on_release(cgrp
)) {
4700 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4701 check_for_release(cgrp
);
4703 cgroup_wakeup_rmdir_waiter(cgrp
);
4706 WARN_ON_ONCE(val
< 1);
4708 EXPORT_SYMBOL_GPL(__css_put
);
4711 * Notify userspace when a cgroup is released, by running the
4712 * configured release agent with the name of the cgroup (path
4713 * relative to the root of cgroup file system) as the argument.
4715 * Most likely, this user command will try to rmdir this cgroup.
4717 * This races with the possibility that some other task will be
4718 * attached to this cgroup before it is removed, or that some other
4719 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4720 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4721 * unused, and this cgroup will be reprieved from its death sentence,
4722 * to continue to serve a useful existence. Next time it's released,
4723 * we will get notified again, if it still has 'notify_on_release' set.
4725 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4726 * means only wait until the task is successfully execve()'d. The
4727 * separate release agent task is forked by call_usermodehelper(),
4728 * then control in this thread returns here, without waiting for the
4729 * release agent task. We don't bother to wait because the caller of
4730 * this routine has no use for the exit status of the release agent
4731 * task, so no sense holding our caller up for that.
4733 static void cgroup_release_agent(struct work_struct
*work
)
4735 BUG_ON(work
!= &release_agent_work
);
4736 mutex_lock(&cgroup_mutex
);
4737 raw_spin_lock(&release_list_lock
);
4738 while (!list_empty(&release_list
)) {
4739 char *argv
[3], *envp
[3];
4741 char *pathbuf
= NULL
, *agentbuf
= NULL
;
4742 struct cgroup
*cgrp
= list_entry(release_list
.next
,
4745 list_del_init(&cgrp
->release_list
);
4746 raw_spin_unlock(&release_list_lock
);
4747 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4750 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
4752 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
4757 argv
[i
++] = agentbuf
;
4758 argv
[i
++] = pathbuf
;
4762 /* minimal command environment */
4763 envp
[i
++] = "HOME=/";
4764 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4767 /* Drop the lock while we invoke the usermode helper,
4768 * since the exec could involve hitting disk and hence
4769 * be a slow process */
4770 mutex_unlock(&cgroup_mutex
);
4771 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
4772 mutex_lock(&cgroup_mutex
);
4776 raw_spin_lock(&release_list_lock
);
4778 raw_spin_unlock(&release_list_lock
);
4779 mutex_unlock(&cgroup_mutex
);
4782 static int __init
cgroup_disable(char *str
)
4787 while ((token
= strsep(&str
, ",")) != NULL
) {
4791 * cgroup_disable, being at boot time, can't know about module
4792 * subsystems, so we don't worry about them.
4794 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4795 struct cgroup_subsys
*ss
= subsys
[i
];
4797 if (!strcmp(token
, ss
->name
)) {
4799 printk(KERN_INFO
"Disabling %s control group"
4800 " subsystem\n", ss
->name
);
4807 __setup("cgroup_disable=", cgroup_disable
);
4810 * Functons for CSS ID.
4814 *To get ID other than 0, this should be called when !cgroup_is_removed().
4816 unsigned short css_id(struct cgroup_subsys_state
*css
)
4818 struct css_id
*cssid
;
4821 * This css_id() can return correct value when somone has refcnt
4822 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4823 * it's unchanged until freed.
4825 cssid
= rcu_dereference_check(css
->id
, atomic_read(&css
->refcnt
));
4831 EXPORT_SYMBOL_GPL(css_id
);
4833 unsigned short css_depth(struct cgroup_subsys_state
*css
)
4835 struct css_id
*cssid
;
4837 cssid
= rcu_dereference_check(css
->id
, atomic_read(&css
->refcnt
));
4840 return cssid
->depth
;
4843 EXPORT_SYMBOL_GPL(css_depth
);
4846 * css_is_ancestor - test "root" css is an ancestor of "child"
4847 * @child: the css to be tested.
4848 * @root: the css supporsed to be an ancestor of the child.
4850 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4851 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4852 * But, considering usual usage, the csses should be valid objects after test.
4853 * Assuming that the caller will do some action to the child if this returns
4854 * returns true, the caller must take "child";s reference count.
4855 * If "child" is valid object and this returns true, "root" is valid, too.
4858 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
4859 const struct cgroup_subsys_state
*root
)
4861 struct css_id
*child_id
;
4862 struct css_id
*root_id
;
4866 child_id
= rcu_dereference(child
->id
);
4867 root_id
= rcu_dereference(root
->id
);
4870 || (child_id
->depth
< root_id
->depth
)
4871 || (child_id
->stack
[root_id
->depth
] != root_id
->id
))
4877 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
4879 struct css_id
*id
= css
->id
;
4880 /* When this is called before css_id initialization, id can be NULL */
4884 BUG_ON(!ss
->use_id
);
4886 rcu_assign_pointer(id
->css
, NULL
);
4887 rcu_assign_pointer(css
->id
, NULL
);
4888 write_lock(&ss
->id_lock
);
4889 idr_remove(&ss
->idr
, id
->id
);
4890 write_unlock(&ss
->id_lock
);
4891 kfree_rcu(id
, rcu_head
);
4893 EXPORT_SYMBOL_GPL(free_css_id
);
4896 * This is called by init or create(). Then, calls to this function are
4897 * always serialized (By cgroup_mutex() at create()).
4900 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
4902 struct css_id
*newid
;
4903 int myid
, error
, size
;
4905 BUG_ON(!ss
->use_id
);
4907 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
4908 newid
= kzalloc(size
, GFP_KERNEL
);
4910 return ERR_PTR(-ENOMEM
);
4912 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
4916 write_lock(&ss
->id_lock
);
4917 /* Don't use 0. allocates an ID of 1-65535 */
4918 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
4919 write_unlock(&ss
->id_lock
);
4921 /* Returns error when there are no free spaces for new ID.*/
4926 if (myid
> CSS_ID_MAX
)
4930 newid
->depth
= depth
;
4934 write_lock(&ss
->id_lock
);
4935 idr_remove(&ss
->idr
, myid
);
4936 write_unlock(&ss
->id_lock
);
4939 return ERR_PTR(error
);
4943 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
4944 struct cgroup_subsys_state
*rootcss
)
4946 struct css_id
*newid
;
4948 rwlock_init(&ss
->id_lock
);
4951 newid
= get_new_cssid(ss
, 0);
4953 return PTR_ERR(newid
);
4955 newid
->stack
[0] = newid
->id
;
4956 newid
->css
= rootcss
;
4957 rootcss
->id
= newid
;
4961 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
4962 struct cgroup
*child
)
4964 int subsys_id
, i
, depth
= 0;
4965 struct cgroup_subsys_state
*parent_css
, *child_css
;
4966 struct css_id
*child_id
, *parent_id
;
4968 subsys_id
= ss
->subsys_id
;
4969 parent_css
= parent
->subsys
[subsys_id
];
4970 child_css
= child
->subsys
[subsys_id
];
4971 parent_id
= parent_css
->id
;
4972 depth
= parent_id
->depth
+ 1;
4974 child_id
= get_new_cssid(ss
, depth
);
4975 if (IS_ERR(child_id
))
4976 return PTR_ERR(child_id
);
4978 for (i
= 0; i
< depth
; i
++)
4979 child_id
->stack
[i
] = parent_id
->stack
[i
];
4980 child_id
->stack
[depth
] = child_id
->id
;
4982 * child_id->css pointer will be set after this cgroup is available
4983 * see cgroup_populate_dir()
4985 rcu_assign_pointer(child_css
->id
, child_id
);
4991 * css_lookup - lookup css by id
4992 * @ss: cgroup subsys to be looked into.
4995 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4996 * NULL if not. Should be called under rcu_read_lock()
4998 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5000 struct css_id
*cssid
= NULL
;
5002 BUG_ON(!ss
->use_id
);
5003 cssid
= idr_find(&ss
->idr
, id
);
5005 if (unlikely(!cssid
))
5008 return rcu_dereference(cssid
->css
);
5010 EXPORT_SYMBOL_GPL(css_lookup
);
5013 * css_get_next - lookup next cgroup under specified hierarchy.
5014 * @ss: pointer to subsystem
5015 * @id: current position of iteration.
5016 * @root: pointer to css. search tree under this.
5017 * @foundid: position of found object.
5019 * Search next css under the specified hierarchy of rootid. Calling under
5020 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5022 struct cgroup_subsys_state
*
5023 css_get_next(struct cgroup_subsys
*ss
, int id
,
5024 struct cgroup_subsys_state
*root
, int *foundid
)
5026 struct cgroup_subsys_state
*ret
= NULL
;
5029 int rootid
= css_id(root
);
5030 int depth
= css_depth(root
);
5035 BUG_ON(!ss
->use_id
);
5036 /* fill start point for scan */
5040 * scan next entry from bitmap(tree), tmpid is updated after
5043 read_lock(&ss
->id_lock
);
5044 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
5045 read_unlock(&ss
->id_lock
);
5049 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
5050 ret
= rcu_dereference(tmp
->css
);
5056 /* continue to scan from next id */
5063 * get corresponding css from file open on cgroupfs directory
5065 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5067 struct cgroup
*cgrp
;
5068 struct inode
*inode
;
5069 struct cgroup_subsys_state
*css
;
5071 inode
= f
->f_dentry
->d_inode
;
5072 /* check in cgroup filesystem dir */
5073 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5074 return ERR_PTR(-EBADF
);
5076 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5077 return ERR_PTR(-EINVAL
);
5080 cgrp
= __d_cgrp(f
->f_dentry
);
5081 css
= cgrp
->subsys
[id
];
5082 return css
? css
: ERR_PTR(-ENOENT
);
5085 #ifdef CONFIG_CGROUP_DEBUG
5086 static struct cgroup_subsys_state
*debug_create(struct cgroup
*cont
)
5088 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5091 return ERR_PTR(-ENOMEM
);
5096 static void debug_destroy(struct cgroup
*cont
)
5098 kfree(cont
->subsys
[debug_subsys_id
]);
5101 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5103 return atomic_read(&cont
->count
);
5106 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5108 return cgroup_task_count(cont
);
5111 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5113 return (u64
)(unsigned long)current
->cgroups
;
5116 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5122 count
= atomic_read(¤t
->cgroups
->refcount
);
5127 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5129 struct seq_file
*seq
)
5131 struct cg_cgroup_link
*link
;
5134 read_lock(&css_set_lock
);
5136 cg
= rcu_dereference(current
->cgroups
);
5137 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5138 struct cgroup
*c
= link
->cgrp
;
5142 name
= c
->dentry
->d_name
.name
;
5145 seq_printf(seq
, "Root %d group %s\n",
5146 c
->root
->hierarchy_id
, name
);
5149 read_unlock(&css_set_lock
);
5153 #define MAX_TASKS_SHOWN_PER_CSS 25
5154 static int cgroup_css_links_read(struct cgroup
*cont
,
5156 struct seq_file
*seq
)
5158 struct cg_cgroup_link
*link
;
5160 read_lock(&css_set_lock
);
5161 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5162 struct css_set
*cg
= link
->cg
;
5163 struct task_struct
*task
;
5165 seq_printf(seq
, "css_set %p\n", cg
);
5166 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5167 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5168 seq_puts(seq
, " ...\n");
5171 seq_printf(seq
, " task %d\n",
5172 task_pid_vnr(task
));
5176 read_unlock(&css_set_lock
);
5180 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5182 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5185 static struct cftype debug_files
[] = {
5187 .name
= "cgroup_refcount",
5188 .read_u64
= cgroup_refcount_read
,
5191 .name
= "taskcount",
5192 .read_u64
= debug_taskcount_read
,
5196 .name
= "current_css_set",
5197 .read_u64
= current_css_set_read
,
5201 .name
= "current_css_set_refcount",
5202 .read_u64
= current_css_set_refcount_read
,
5206 .name
= "current_css_set_cg_links",
5207 .read_seq_string
= current_css_set_cg_links_read
,
5211 .name
= "cgroup_css_links",
5212 .read_seq_string
= cgroup_css_links_read
,
5216 .name
= "releasable",
5217 .read_u64
= releasable_read
,
5221 static int debug_populate(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5223 return cgroup_add_files(cont
, ss
, debug_files
,
5224 ARRAY_SIZE(debug_files
));
5227 struct cgroup_subsys debug_subsys
= {
5229 .create
= debug_create
,
5230 .destroy
= debug_destroy
,
5231 .populate
= debug_populate
,
5232 .subsys_id
= debug_subsys_id
,
5234 #endif /* CONFIG_CGROUP_DEBUG */