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/ctype.h>
31 #include <linux/errno.h>
33 #include <linux/kernel.h>
34 #include <linux/list.h>
36 #include <linux/mutex.h>
37 #include <linux/mount.h>
38 #include <linux/pagemap.h>
39 #include <linux/proc_fs.h>
40 #include <linux/rcupdate.h>
41 #include <linux/sched.h>
42 #include <linux/backing-dev.h>
43 #include <linux/seq_file.h>
44 #include <linux/slab.h>
45 #include <linux/magic.h>
46 #include <linux/spinlock.h>
47 #include <linux/string.h>
48 #include <linux/sort.h>
49 #include <linux/kmod.h>
50 #include <linux/module.h>
51 #include <linux/delayacct.h>
52 #include <linux/cgroupstats.h>
53 #include <linux/hash.h>
54 #include <linux/namei.h>
55 #include <linux/pid_namespace.h>
56 #include <linux/idr.h>
57 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
58 #include <linux/eventfd.h>
59 #include <linux/poll.h>
61 #include <asm/atomic.h>
63 static DEFINE_MUTEX(cgroup_mutex
);
66 * Generate an array of cgroup subsystem pointers. At boot time, this is
67 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
68 * registered after that. The mutable section of this array is protected by
71 #define SUBSYS(_x) &_x ## _subsys,
72 static struct cgroup_subsys
*subsys
[CGROUP_SUBSYS_COUNT
] = {
73 #include <linux/cgroup_subsys.h>
76 #define MAX_CGROUP_ROOT_NAMELEN 64
79 * A cgroupfs_root represents the root of a cgroup hierarchy,
80 * and may be associated with a superblock to form an active
83 struct cgroupfs_root
{
84 struct super_block
*sb
;
87 * The bitmask of subsystems intended to be attached to this
90 unsigned long subsys_bits
;
92 /* Unique id for this hierarchy. */
95 /* The bitmask of subsystems currently attached to this hierarchy */
96 unsigned long actual_subsys_bits
;
98 /* A list running through the attached subsystems */
99 struct list_head subsys_list
;
101 /* The root cgroup for this hierarchy */
102 struct cgroup top_cgroup
;
104 /* Tracks how many cgroups are currently defined in hierarchy.*/
105 int number_of_cgroups
;
107 /* A list running through the active hierarchies */
108 struct list_head root_list
;
110 /* Hierarchy-specific flags */
113 /* The path to use for release notifications. */
114 char release_agent_path
[PATH_MAX
];
116 /* The name for this hierarchy - may be empty */
117 char name
[MAX_CGROUP_ROOT_NAMELEN
];
121 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
122 * subsystems that are otherwise unattached - it never has more than a
123 * single cgroup, and all tasks are part of that cgroup.
125 static struct cgroupfs_root rootnode
;
128 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
129 * cgroup_subsys->use_id != 0.
131 #define CSS_ID_MAX (65535)
134 * The css to which this ID points. This pointer is set to valid value
135 * after cgroup is populated. If cgroup is removed, this will be NULL.
136 * This pointer is expected to be RCU-safe because destroy()
137 * is called after synchronize_rcu(). But for safe use, css_is_removed()
138 * css_tryget() should be used for avoiding race.
140 struct cgroup_subsys_state __rcu
*css
;
146 * Depth in hierarchy which this ID belongs to.
148 unsigned short depth
;
150 * ID is freed by RCU. (and lookup routine is RCU safe.)
152 struct rcu_head rcu_head
;
154 * Hierarchy of CSS ID belongs to.
156 unsigned short stack
[0]; /* Array of Length (depth+1) */
160 * cgroup_event represents events which userspace want to receive.
162 struct cgroup_event
{
164 * Cgroup which the event belongs to.
168 * Control file which the event associated.
172 * eventfd to signal userspace about the event.
174 struct eventfd_ctx
*eventfd
;
176 * Each of these stored in a list by the cgroup.
178 struct list_head list
;
180 * All fields below needed to unregister event when
181 * userspace closes eventfd.
184 wait_queue_head_t
*wqh
;
186 struct work_struct remove
;
189 /* The list of hierarchy roots */
191 static LIST_HEAD(roots
);
192 static int root_count
;
194 static DEFINE_IDA(hierarchy_ida
);
195 static int next_hierarchy_id
;
196 static DEFINE_SPINLOCK(hierarchy_id_lock
);
198 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
199 #define dummytop (&rootnode.top_cgroup)
201 /* This flag indicates whether tasks in the fork and exit paths should
202 * check for fork/exit handlers to call. This avoids us having to do
203 * extra work in the fork/exit path if none of the subsystems need to
206 static int need_forkexit_callback __read_mostly
;
208 #ifdef CONFIG_PROVE_LOCKING
209 int cgroup_lock_is_held(void)
211 return lockdep_is_held(&cgroup_mutex
);
213 #else /* #ifdef CONFIG_PROVE_LOCKING */
214 int cgroup_lock_is_held(void)
216 return mutex_is_locked(&cgroup_mutex
);
218 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
220 EXPORT_SYMBOL_GPL(cgroup_lock_is_held
);
222 /* convenient tests for these bits */
223 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
225 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
228 /* bits in struct cgroupfs_root flags field */
230 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
233 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
236 (1 << CGRP_RELEASABLE
) |
237 (1 << CGRP_NOTIFY_ON_RELEASE
);
238 return (cgrp
->flags
& bits
) == bits
;
241 static int notify_on_release(const struct cgroup
*cgrp
)
243 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
246 static int clone_children(const struct cgroup
*cgrp
)
248 return test_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
252 * for_each_subsys() allows you to iterate on each subsystem attached to
253 * an active hierarchy
255 #define for_each_subsys(_root, _ss) \
256 list_for_each_entry(_ss, &_root->subsys_list, sibling)
258 /* for_each_active_root() allows you to iterate across the active hierarchies */
259 #define for_each_active_root(_root) \
260 list_for_each_entry(_root, &roots, root_list)
262 /* the list of cgroups eligible for automatic release. Protected by
263 * release_list_lock */
264 static LIST_HEAD(release_list
);
265 static DEFINE_SPINLOCK(release_list_lock
);
266 static void cgroup_release_agent(struct work_struct
*work
);
267 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
268 static void check_for_release(struct cgroup
*cgrp
);
270 /* Link structure for associating css_set objects with cgroups */
271 struct cg_cgroup_link
{
273 * List running through cg_cgroup_links associated with a
274 * cgroup, anchored on cgroup->css_sets
276 struct list_head cgrp_link_list
;
279 * List running through cg_cgroup_links pointing at a
280 * single css_set object, anchored on css_set->cg_links
282 struct list_head cg_link_list
;
286 /* The default css_set - used by init and its children prior to any
287 * hierarchies being mounted. It contains a pointer to the root state
288 * for each subsystem. Also used to anchor the list of css_sets. Not
289 * reference-counted, to improve performance when child cgroups
290 * haven't been created.
293 static struct css_set init_css_set
;
294 static struct cg_cgroup_link init_css_set_link
;
296 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
297 struct cgroup_subsys_state
*css
);
299 /* css_set_lock protects the list of css_set objects, and the
300 * chain of tasks off each css_set. Nests outside task->alloc_lock
301 * due to cgroup_iter_start() */
302 static DEFINE_RWLOCK(css_set_lock
);
303 static int css_set_count
;
306 * hash table for cgroup groups. This improves the performance to find
307 * an existing css_set. This hash doesn't (currently) take into
308 * account cgroups in empty hierarchies.
310 #define CSS_SET_HASH_BITS 7
311 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
312 static struct hlist_head css_set_table
[CSS_SET_TABLE_SIZE
];
314 static struct hlist_head
*css_set_hash(struct cgroup_subsys_state
*css
[])
318 unsigned long tmp
= 0UL;
320 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
321 tmp
+= (unsigned long)css
[i
];
322 tmp
= (tmp
>> 16) ^ tmp
;
324 index
= hash_long(tmp
, CSS_SET_HASH_BITS
);
326 return &css_set_table
[index
];
329 /* We don't maintain the lists running through each css_set to its
330 * task until after the first call to cgroup_iter_start(). This
331 * reduces the fork()/exit() overhead for people who have cgroups
332 * compiled into their kernel but not actually in use */
333 static int use_task_css_set_links __read_mostly
;
335 static void __put_css_set(struct css_set
*cg
, int taskexit
)
337 struct cg_cgroup_link
*link
;
338 struct cg_cgroup_link
*saved_link
;
340 * Ensure that the refcount doesn't hit zero while any readers
341 * can see it. Similar to atomic_dec_and_lock(), but for an
344 if (atomic_add_unless(&cg
->refcount
, -1, 1))
346 write_lock(&css_set_lock
);
347 if (!atomic_dec_and_test(&cg
->refcount
)) {
348 write_unlock(&css_set_lock
);
352 /* This css_set is dead. unlink it and release cgroup refcounts */
353 hlist_del(&cg
->hlist
);
356 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
358 struct cgroup
*cgrp
= link
->cgrp
;
359 list_del(&link
->cg_link_list
);
360 list_del(&link
->cgrp_link_list
);
361 if (atomic_dec_and_test(&cgrp
->count
) &&
362 notify_on_release(cgrp
)) {
364 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
365 check_for_release(cgrp
);
371 write_unlock(&css_set_lock
);
372 kfree_rcu(cg
, rcu_head
);
376 * refcounted get/put for css_set objects
378 static inline void get_css_set(struct css_set
*cg
)
380 atomic_inc(&cg
->refcount
);
383 static inline void put_css_set(struct css_set
*cg
)
385 __put_css_set(cg
, 0);
388 static inline void put_css_set_taskexit(struct css_set
*cg
)
390 __put_css_set(cg
, 1);
394 * compare_css_sets - helper function for find_existing_css_set().
395 * @cg: candidate css_set being tested
396 * @old_cg: existing css_set for a task
397 * @new_cgrp: cgroup that's being entered by the task
398 * @template: desired set of css pointers in css_set (pre-calculated)
400 * Returns true if "cg" matches "old_cg" except for the hierarchy
401 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
403 static bool compare_css_sets(struct css_set
*cg
,
404 struct css_set
*old_cg
,
405 struct cgroup
*new_cgrp
,
406 struct cgroup_subsys_state
*template[])
408 struct list_head
*l1
, *l2
;
410 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
411 /* Not all subsystems matched */
416 * Compare cgroup pointers in order to distinguish between
417 * different cgroups in heirarchies with no subsystems. We
418 * could get by with just this check alone (and skip the
419 * memcmp above) but on most setups the memcmp check will
420 * avoid the need for this more expensive check on almost all
425 l2
= &old_cg
->cg_links
;
427 struct cg_cgroup_link
*cgl1
, *cgl2
;
428 struct cgroup
*cg1
, *cg2
;
432 /* See if we reached the end - both lists are equal length. */
433 if (l1
== &cg
->cg_links
) {
434 BUG_ON(l2
!= &old_cg
->cg_links
);
437 BUG_ON(l2
== &old_cg
->cg_links
);
439 /* Locate the cgroups associated with these links. */
440 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
441 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
444 /* Hierarchies should be linked in the same order. */
445 BUG_ON(cg1
->root
!= cg2
->root
);
448 * If this hierarchy is the hierarchy of the cgroup
449 * that's changing, then we need to check that this
450 * css_set points to the new cgroup; if it's any other
451 * hierarchy, then this css_set should point to the
452 * same cgroup as the old css_set.
454 if (cg1
->root
== new_cgrp
->root
) {
466 * find_existing_css_set() is a helper for
467 * find_css_set(), and checks to see whether an existing
468 * css_set is suitable.
470 * oldcg: the cgroup group that we're using before the cgroup
473 * cgrp: the cgroup that we're moving into
475 * template: location in which to build the desired set of subsystem
476 * state objects for the new cgroup group
478 static struct css_set
*find_existing_css_set(
479 struct css_set
*oldcg
,
481 struct cgroup_subsys_state
*template[])
484 struct cgroupfs_root
*root
= cgrp
->root
;
485 struct hlist_head
*hhead
;
486 struct hlist_node
*node
;
490 * Build the set of subsystem state objects that we want to see in the
491 * new css_set. while subsystems can change globally, the entries here
492 * won't change, so no need for locking.
494 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
495 if (root
->subsys_bits
& (1UL << i
)) {
496 /* Subsystem is in this hierarchy. So we want
497 * the subsystem state from the new
499 template[i
] = cgrp
->subsys
[i
];
501 /* Subsystem is not in this hierarchy, so we
502 * don't want to change the subsystem state */
503 template[i
] = oldcg
->subsys
[i
];
507 hhead
= css_set_hash(template);
508 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
509 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
512 /* This css_set matches what we need */
516 /* No existing cgroup group matched */
520 static void free_cg_links(struct list_head
*tmp
)
522 struct cg_cgroup_link
*link
;
523 struct cg_cgroup_link
*saved_link
;
525 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
526 list_del(&link
->cgrp_link_list
);
532 * allocate_cg_links() allocates "count" cg_cgroup_link structures
533 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
534 * success or a negative error
536 static int allocate_cg_links(int count
, struct list_head
*tmp
)
538 struct cg_cgroup_link
*link
;
541 for (i
= 0; i
< count
; i
++) {
542 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
547 list_add(&link
->cgrp_link_list
, tmp
);
553 * link_css_set - a helper function to link a css_set to a cgroup
554 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
555 * @cg: the css_set to be linked
556 * @cgrp: the destination cgroup
558 static void link_css_set(struct list_head
*tmp_cg_links
,
559 struct css_set
*cg
, struct cgroup
*cgrp
)
561 struct cg_cgroup_link
*link
;
563 BUG_ON(list_empty(tmp_cg_links
));
564 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
568 atomic_inc(&cgrp
->count
);
569 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
571 * Always add links to the tail of the list so that the list
572 * is sorted by order of hierarchy creation
574 list_add_tail(&link
->cg_link_list
, &cg
->cg_links
);
578 * find_css_set() takes an existing cgroup group and a
579 * cgroup object, and returns a css_set object that's
580 * equivalent to the old group, but with the given cgroup
581 * substituted into the appropriate hierarchy. Must be called with
584 static struct css_set
*find_css_set(
585 struct css_set
*oldcg
, struct cgroup
*cgrp
)
588 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
590 struct list_head tmp_cg_links
;
592 struct hlist_head
*hhead
;
593 struct cg_cgroup_link
*link
;
595 /* First see if we already have a cgroup group that matches
597 read_lock(&css_set_lock
);
598 res
= find_existing_css_set(oldcg
, cgrp
, template);
601 read_unlock(&css_set_lock
);
606 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
610 /* Allocate all the cg_cgroup_link objects that we'll need */
611 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
616 atomic_set(&res
->refcount
, 1);
617 INIT_LIST_HEAD(&res
->cg_links
);
618 INIT_LIST_HEAD(&res
->tasks
);
619 INIT_HLIST_NODE(&res
->hlist
);
621 /* Copy the set of subsystem state objects generated in
622 * find_existing_css_set() */
623 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
625 write_lock(&css_set_lock
);
626 /* Add reference counts and links from the new css_set. */
627 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
628 struct cgroup
*c
= link
->cgrp
;
629 if (c
->root
== cgrp
->root
)
631 link_css_set(&tmp_cg_links
, res
, c
);
634 BUG_ON(!list_empty(&tmp_cg_links
));
638 /* Add this cgroup group to the hash table */
639 hhead
= css_set_hash(res
->subsys
);
640 hlist_add_head(&res
->hlist
, hhead
);
642 write_unlock(&css_set_lock
);
648 * Return the cgroup for "task" from the given hierarchy. Must be
649 * called with cgroup_mutex held.
651 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
652 struct cgroupfs_root
*root
)
655 struct cgroup
*res
= NULL
;
657 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
658 read_lock(&css_set_lock
);
660 * No need to lock the task - since we hold cgroup_mutex the
661 * task can't change groups, so the only thing that can happen
662 * is that it exits and its css is set back to init_css_set.
665 if (css
== &init_css_set
) {
666 res
= &root
->top_cgroup
;
668 struct cg_cgroup_link
*link
;
669 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
670 struct cgroup
*c
= link
->cgrp
;
671 if (c
->root
== root
) {
677 read_unlock(&css_set_lock
);
683 * There is one global cgroup mutex. We also require taking
684 * task_lock() when dereferencing a task's cgroup subsys pointers.
685 * See "The task_lock() exception", at the end of this comment.
687 * A task must hold cgroup_mutex to modify cgroups.
689 * Any task can increment and decrement the count field without lock.
690 * So in general, code holding cgroup_mutex can't rely on the count
691 * field not changing. However, if the count goes to zero, then only
692 * cgroup_attach_task() can increment it again. Because a count of zero
693 * means that no tasks are currently attached, therefore there is no
694 * way a task attached to that cgroup can fork (the other way to
695 * increment the count). So code holding cgroup_mutex can safely
696 * assume that if the count is zero, it will stay zero. Similarly, if
697 * a task holds cgroup_mutex on a cgroup with zero count, it
698 * knows that the cgroup won't be removed, as cgroup_rmdir()
701 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
702 * (usually) take cgroup_mutex. These are the two most performance
703 * critical pieces of code here. The exception occurs on cgroup_exit(),
704 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
705 * is taken, and if the cgroup count is zero, a usermode call made
706 * to the release agent with the name of the cgroup (path relative to
707 * the root of cgroup file system) as the argument.
709 * A cgroup can only be deleted if both its 'count' of using tasks
710 * is zero, and its list of 'children' cgroups is empty. Since all
711 * tasks in the system use _some_ cgroup, and since there is always at
712 * least one task in the system (init, pid == 1), therefore, top_cgroup
713 * always has either children cgroups and/or using tasks. So we don't
714 * need a special hack to ensure that top_cgroup cannot be deleted.
716 * The task_lock() exception
718 * The need for this exception arises from the action of
719 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
720 * another. It does so using cgroup_mutex, however there are
721 * several performance critical places that need to reference
722 * task->cgroup without the expense of grabbing a system global
723 * mutex. Therefore except as noted below, when dereferencing or, as
724 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
725 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
726 * the task_struct routinely used for such matters.
728 * P.S. One more locking exception. RCU is used to guard the
729 * update of a tasks cgroup pointer by cgroup_attach_task()
733 * cgroup_lock - lock out any changes to cgroup structures
736 void cgroup_lock(void)
738 mutex_lock(&cgroup_mutex
);
740 EXPORT_SYMBOL_GPL(cgroup_lock
);
743 * cgroup_unlock - release lock on cgroup changes
745 * Undo the lock taken in a previous cgroup_lock() call.
747 void cgroup_unlock(void)
749 mutex_unlock(&cgroup_mutex
);
751 EXPORT_SYMBOL_GPL(cgroup_unlock
);
754 * A couple of forward declarations required, due to cyclic reference loop:
755 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
756 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
760 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
);
761 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
762 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
763 static int cgroup_populate_dir(struct cgroup
*cgrp
);
764 static const struct inode_operations cgroup_dir_inode_operations
;
765 static const struct file_operations proc_cgroupstats_operations
;
767 static struct backing_dev_info cgroup_backing_dev_info
= {
769 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
772 static int alloc_css_id(struct cgroup_subsys
*ss
,
773 struct cgroup
*parent
, struct cgroup
*child
);
775 static struct inode
*cgroup_new_inode(mode_t mode
, struct super_block
*sb
)
777 struct inode
*inode
= new_inode(sb
);
780 inode
->i_ino
= get_next_ino();
781 inode
->i_mode
= mode
;
782 inode
->i_uid
= current_fsuid();
783 inode
->i_gid
= current_fsgid();
784 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
785 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
791 * Call subsys's pre_destroy handler.
792 * This is called before css refcnt check.
794 static int cgroup_call_pre_destroy(struct cgroup
*cgrp
)
796 struct cgroup_subsys
*ss
;
799 for_each_subsys(cgrp
->root
, ss
)
800 if (ss
->pre_destroy
) {
801 ret
= ss
->pre_destroy(ss
, cgrp
);
809 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
811 /* is dentry a directory ? if so, kfree() associated cgroup */
812 if (S_ISDIR(inode
->i_mode
)) {
813 struct cgroup
*cgrp
= dentry
->d_fsdata
;
814 struct cgroup_subsys
*ss
;
815 BUG_ON(!(cgroup_is_removed(cgrp
)));
816 /* It's possible for external users to be holding css
817 * reference counts on a cgroup; css_put() needs to
818 * be able to access the cgroup after decrementing
819 * the reference count in order to know if it needs to
820 * queue the cgroup to be handled by the release
824 mutex_lock(&cgroup_mutex
);
826 * Release the subsystem state objects.
828 for_each_subsys(cgrp
->root
, ss
)
829 ss
->destroy(ss
, cgrp
);
831 cgrp
->root
->number_of_cgroups
--;
832 mutex_unlock(&cgroup_mutex
);
835 * Drop the active superblock reference that we took when we
838 deactivate_super(cgrp
->root
->sb
);
841 * if we're getting rid of the cgroup, refcount should ensure
842 * that there are no pidlists left.
844 BUG_ON(!list_empty(&cgrp
->pidlists
));
846 kfree_rcu(cgrp
, rcu_head
);
851 static int cgroup_delete(const struct dentry
*d
)
856 static void remove_dir(struct dentry
*d
)
858 struct dentry
*parent
= dget(d
->d_parent
);
861 simple_rmdir(parent
->d_inode
, d
);
865 static void cgroup_clear_directory(struct dentry
*dentry
)
867 struct list_head
*node
;
869 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
870 spin_lock(&dentry
->d_lock
);
871 node
= dentry
->d_subdirs
.next
;
872 while (node
!= &dentry
->d_subdirs
) {
873 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
875 spin_lock_nested(&d
->d_lock
, DENTRY_D_LOCK_NESTED
);
878 /* This should never be called on a cgroup
879 * directory with child cgroups */
880 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
882 spin_unlock(&d
->d_lock
);
883 spin_unlock(&dentry
->d_lock
);
885 simple_unlink(dentry
->d_inode
, d
);
887 spin_lock(&dentry
->d_lock
);
889 spin_unlock(&d
->d_lock
);
890 node
= dentry
->d_subdirs
.next
;
892 spin_unlock(&dentry
->d_lock
);
896 * NOTE : the dentry must have been dget()'ed
898 static void cgroup_d_remove_dir(struct dentry
*dentry
)
900 struct dentry
*parent
;
902 cgroup_clear_directory(dentry
);
904 parent
= dentry
->d_parent
;
905 spin_lock(&parent
->d_lock
);
906 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
907 list_del_init(&dentry
->d_u
.d_child
);
908 spin_unlock(&dentry
->d_lock
);
909 spin_unlock(&parent
->d_lock
);
914 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
915 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
916 * reference to css->refcnt. In general, this refcnt is expected to goes down
919 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
921 DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq
);
923 static void cgroup_wakeup_rmdir_waiter(struct cgroup
*cgrp
)
925 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
)))
926 wake_up_all(&cgroup_rmdir_waitq
);
929 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
)
934 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
)
936 cgroup_wakeup_rmdir_waiter(css
->cgroup
);
941 * Call with cgroup_mutex held. Drops reference counts on modules, including
942 * any duplicate ones that parse_cgroupfs_options took. If this function
943 * returns an error, no reference counts are touched.
945 static int rebind_subsystems(struct cgroupfs_root
*root
,
946 unsigned long final_bits
)
948 unsigned long added_bits
, removed_bits
;
949 struct cgroup
*cgrp
= &root
->top_cgroup
;
952 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
954 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
955 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
956 /* Check that any added subsystems are currently free */
957 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
958 unsigned long bit
= 1UL << i
;
959 struct cgroup_subsys
*ss
= subsys
[i
];
960 if (!(bit
& added_bits
))
963 * Nobody should tell us to do a subsys that doesn't exist:
964 * parse_cgroupfs_options should catch that case and refcounts
965 * ensure that subsystems won't disappear once selected.
968 if (ss
->root
!= &rootnode
) {
969 /* Subsystem isn't free */
974 /* Currently we don't handle adding/removing subsystems when
975 * any child cgroups exist. This is theoretically supportable
976 * but involves complex error handling, so it's being left until
978 if (root
->number_of_cgroups
> 1)
981 /* Process each subsystem */
982 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
983 struct cgroup_subsys
*ss
= subsys
[i
];
984 unsigned long bit
= 1UL << i
;
985 if (bit
& added_bits
) {
986 /* We're binding this subsystem to this hierarchy */
988 BUG_ON(cgrp
->subsys
[i
]);
989 BUG_ON(!dummytop
->subsys
[i
]);
990 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
991 mutex_lock(&ss
->hierarchy_mutex
);
992 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
993 cgrp
->subsys
[i
]->cgroup
= cgrp
;
994 list_move(&ss
->sibling
, &root
->subsys_list
);
998 mutex_unlock(&ss
->hierarchy_mutex
);
999 /* refcount was already taken, and we're keeping it */
1000 } else if (bit
& removed_bits
) {
1001 /* We're removing this subsystem */
1003 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1004 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1005 mutex_lock(&ss
->hierarchy_mutex
);
1007 ss
->bind(ss
, dummytop
);
1008 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1009 cgrp
->subsys
[i
] = NULL
;
1010 subsys
[i
]->root
= &rootnode
;
1011 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1012 mutex_unlock(&ss
->hierarchy_mutex
);
1013 /* subsystem is now free - drop reference on module */
1014 module_put(ss
->module
);
1015 } else if (bit
& final_bits
) {
1016 /* Subsystem state should already exist */
1018 BUG_ON(!cgrp
->subsys
[i
]);
1020 * a refcount was taken, but we already had one, so
1021 * drop the extra reference.
1023 module_put(ss
->module
);
1024 #ifdef CONFIG_MODULE_UNLOAD
1025 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1028 /* Subsystem state shouldn't exist */
1029 BUG_ON(cgrp
->subsys
[i
]);
1032 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
1038 static int cgroup_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
1040 struct cgroupfs_root
*root
= vfs
->mnt_sb
->s_fs_info
;
1041 struct cgroup_subsys
*ss
;
1043 mutex_lock(&cgroup_mutex
);
1044 for_each_subsys(root
, ss
)
1045 seq_printf(seq
, ",%s", ss
->name
);
1046 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1047 seq_puts(seq
, ",noprefix");
1048 if (strlen(root
->release_agent_path
))
1049 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1050 if (clone_children(&root
->top_cgroup
))
1051 seq_puts(seq
, ",clone_children");
1052 if (strlen(root
->name
))
1053 seq_printf(seq
, ",name=%s", root
->name
);
1054 mutex_unlock(&cgroup_mutex
);
1058 struct cgroup_sb_opts
{
1059 unsigned long subsys_bits
;
1060 unsigned long flags
;
1061 char *release_agent
;
1062 bool clone_children
;
1064 /* User explicitly requested empty subsystem */
1067 struct cgroupfs_root
*new_root
;
1072 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1073 * with cgroup_mutex held to protect the subsys[] array. This function takes
1074 * refcounts on subsystems to be used, unless it returns error, in which case
1075 * no refcounts are taken.
1077 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1079 char *token
, *o
= data
;
1080 bool all_ss
= false, one_ss
= false;
1081 unsigned long mask
= (unsigned long)-1;
1083 bool module_pin_failed
= false;
1085 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1087 #ifdef CONFIG_CPUSETS
1088 mask
= ~(1UL << cpuset_subsys_id
);
1091 memset(opts
, 0, sizeof(*opts
));
1093 while ((token
= strsep(&o
, ",")) != NULL
) {
1096 if (!strcmp(token
, "none")) {
1097 /* Explicitly have no subsystems */
1101 if (!strcmp(token
, "all")) {
1102 /* Mutually exclusive option 'all' + subsystem name */
1108 if (!strcmp(token
, "noprefix")) {
1109 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1112 if (!strcmp(token
, "clone_children")) {
1113 opts
->clone_children
= true;
1116 if (!strncmp(token
, "release_agent=", 14)) {
1117 /* Specifying two release agents is forbidden */
1118 if (opts
->release_agent
)
1120 opts
->release_agent
=
1121 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1122 if (!opts
->release_agent
)
1126 if (!strncmp(token
, "name=", 5)) {
1127 const char *name
= token
+ 5;
1128 /* Can't specify an empty name */
1131 /* Must match [\w.-]+ */
1132 for (i
= 0; i
< strlen(name
); i
++) {
1136 if ((c
== '.') || (c
== '-') || (c
== '_'))
1140 /* Specifying two names is forbidden */
1143 opts
->name
= kstrndup(name
,
1144 MAX_CGROUP_ROOT_NAMELEN
- 1,
1152 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1153 struct cgroup_subsys
*ss
= subsys
[i
];
1156 if (strcmp(token
, ss
->name
))
1161 /* Mutually exclusive option 'all' + subsystem name */
1164 set_bit(i
, &opts
->subsys_bits
);
1169 if (i
== CGROUP_SUBSYS_COUNT
)
1174 * If the 'all' option was specified select all the subsystems,
1175 * otherwise 'all, 'none' and a subsystem name options were not
1176 * specified, let's default to 'all'
1178 if (all_ss
|| (!all_ss
&& !one_ss
&& !opts
->none
)) {
1179 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1180 struct cgroup_subsys
*ss
= subsys
[i
];
1185 set_bit(i
, &opts
->subsys_bits
);
1189 /* Consistency checks */
1192 * Option noprefix was introduced just for backward compatibility
1193 * with the old cpuset, so we allow noprefix only if mounting just
1194 * the cpuset subsystem.
1196 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1197 (opts
->subsys_bits
& mask
))
1201 /* Can't specify "none" and some subsystems */
1202 if (opts
->subsys_bits
&& opts
->none
)
1206 * We either have to specify by name or by subsystems. (So all
1207 * empty hierarchies must have a name).
1209 if (!opts
->subsys_bits
&& !opts
->name
)
1213 * Grab references on all the modules we'll need, so the subsystems
1214 * don't dance around before rebind_subsystems attaches them. This may
1215 * take duplicate reference counts on a subsystem that's already used,
1216 * but rebind_subsystems handles this case.
1218 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1219 unsigned long bit
= 1UL << i
;
1221 if (!(bit
& opts
->subsys_bits
))
1223 if (!try_module_get(subsys
[i
]->module
)) {
1224 module_pin_failed
= true;
1228 if (module_pin_failed
) {
1230 * oops, one of the modules was going away. this means that we
1231 * raced with a module_delete call, and to the user this is
1232 * essentially a "subsystem doesn't exist" case.
1234 for (i
--; i
>= CGROUP_BUILTIN_SUBSYS_COUNT
; i
--) {
1235 /* drop refcounts only on the ones we took */
1236 unsigned long bit
= 1UL << i
;
1238 if (!(bit
& opts
->subsys_bits
))
1240 module_put(subsys
[i
]->module
);
1248 static void drop_parsed_module_refcounts(unsigned long subsys_bits
)
1251 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1252 unsigned long bit
= 1UL << i
;
1254 if (!(bit
& subsys_bits
))
1256 module_put(subsys
[i
]->module
);
1260 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1263 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1264 struct cgroup
*cgrp
= &root
->top_cgroup
;
1265 struct cgroup_sb_opts opts
;
1267 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1268 mutex_lock(&cgroup_mutex
);
1270 /* See what subsystems are wanted */
1271 ret
= parse_cgroupfs_options(data
, &opts
);
1275 /* Don't allow flags or name to change at remount */
1276 if (opts
.flags
!= root
->flags
||
1277 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1279 drop_parsed_module_refcounts(opts
.subsys_bits
);
1283 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
1285 drop_parsed_module_refcounts(opts
.subsys_bits
);
1289 /* (re)populate subsystem files */
1290 cgroup_populate_dir(cgrp
);
1292 if (opts
.release_agent
)
1293 strcpy(root
->release_agent_path
, opts
.release_agent
);
1295 kfree(opts
.release_agent
);
1297 mutex_unlock(&cgroup_mutex
);
1298 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1302 static const struct super_operations cgroup_ops
= {
1303 .statfs
= simple_statfs
,
1304 .drop_inode
= generic_delete_inode
,
1305 .show_options
= cgroup_show_options
,
1306 .remount_fs
= cgroup_remount
,
1309 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1311 INIT_LIST_HEAD(&cgrp
->sibling
);
1312 INIT_LIST_HEAD(&cgrp
->children
);
1313 INIT_LIST_HEAD(&cgrp
->css_sets
);
1314 INIT_LIST_HEAD(&cgrp
->release_list
);
1315 INIT_LIST_HEAD(&cgrp
->pidlists
);
1316 mutex_init(&cgrp
->pidlist_mutex
);
1317 INIT_LIST_HEAD(&cgrp
->event_list
);
1318 spin_lock_init(&cgrp
->event_list_lock
);
1321 static void init_cgroup_root(struct cgroupfs_root
*root
)
1323 struct cgroup
*cgrp
= &root
->top_cgroup
;
1324 INIT_LIST_HEAD(&root
->subsys_list
);
1325 INIT_LIST_HEAD(&root
->root_list
);
1326 root
->number_of_cgroups
= 1;
1328 cgrp
->top_cgroup
= cgrp
;
1329 init_cgroup_housekeeping(cgrp
);
1332 static bool init_root_id(struct cgroupfs_root
*root
)
1337 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1339 spin_lock(&hierarchy_id_lock
);
1340 /* Try to allocate the next unused ID */
1341 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1342 &root
->hierarchy_id
);
1344 /* Try again starting from 0 */
1345 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1347 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1348 } else if (ret
!= -EAGAIN
) {
1349 /* Can only get here if the 31-bit IDR is full ... */
1352 spin_unlock(&hierarchy_id_lock
);
1357 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1359 struct cgroup_sb_opts
*opts
= data
;
1360 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1362 /* If we asked for a name then it must match */
1363 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1367 * If we asked for subsystems (or explicitly for no
1368 * subsystems) then they must match
1370 if ((opts
->subsys_bits
|| opts
->none
)
1371 && (opts
->subsys_bits
!= root
->subsys_bits
))
1377 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1379 struct cgroupfs_root
*root
;
1381 if (!opts
->subsys_bits
&& !opts
->none
)
1384 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1386 return ERR_PTR(-ENOMEM
);
1388 if (!init_root_id(root
)) {
1390 return ERR_PTR(-ENOMEM
);
1392 init_cgroup_root(root
);
1394 root
->subsys_bits
= opts
->subsys_bits
;
1395 root
->flags
= opts
->flags
;
1396 if (opts
->release_agent
)
1397 strcpy(root
->release_agent_path
, opts
->release_agent
);
1399 strcpy(root
->name
, opts
->name
);
1400 if (opts
->clone_children
)
1401 set_bit(CGRP_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1405 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1410 BUG_ON(!root
->hierarchy_id
);
1411 spin_lock(&hierarchy_id_lock
);
1412 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1413 spin_unlock(&hierarchy_id_lock
);
1417 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1420 struct cgroup_sb_opts
*opts
= data
;
1422 /* If we don't have a new root, we can't set up a new sb */
1423 if (!opts
->new_root
)
1426 BUG_ON(!opts
->subsys_bits
&& !opts
->none
);
1428 ret
= set_anon_super(sb
, NULL
);
1432 sb
->s_fs_info
= opts
->new_root
;
1433 opts
->new_root
->sb
= sb
;
1435 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1436 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1437 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1438 sb
->s_op
= &cgroup_ops
;
1443 static int cgroup_get_rootdir(struct super_block
*sb
)
1445 static const struct dentry_operations cgroup_dops
= {
1446 .d_iput
= cgroup_diput
,
1447 .d_delete
= cgroup_delete
,
1450 struct inode
*inode
=
1451 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1452 struct dentry
*dentry
;
1457 inode
->i_fop
= &simple_dir_operations
;
1458 inode
->i_op
= &cgroup_dir_inode_operations
;
1459 /* directories start off with i_nlink == 2 (for "." entry) */
1461 dentry
= d_alloc_root(inode
);
1466 sb
->s_root
= dentry
;
1467 /* for everything else we want ->d_op set */
1468 sb
->s_d_op
= &cgroup_dops
;
1472 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1473 int flags
, const char *unused_dev_name
,
1476 struct cgroup_sb_opts opts
;
1477 struct cgroupfs_root
*root
;
1479 struct super_block
*sb
;
1480 struct cgroupfs_root
*new_root
;
1482 /* First find the desired set of subsystems */
1483 mutex_lock(&cgroup_mutex
);
1484 ret
= parse_cgroupfs_options(data
, &opts
);
1485 mutex_unlock(&cgroup_mutex
);
1490 * Allocate a new cgroup root. We may not need it if we're
1491 * reusing an existing hierarchy.
1493 new_root
= cgroup_root_from_opts(&opts
);
1494 if (IS_ERR(new_root
)) {
1495 ret
= PTR_ERR(new_root
);
1498 opts
.new_root
= new_root
;
1500 /* Locate an existing or new sb for this hierarchy */
1501 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, &opts
);
1504 cgroup_drop_root(opts
.new_root
);
1508 root
= sb
->s_fs_info
;
1510 if (root
== opts
.new_root
) {
1511 /* We used the new root structure, so this is a new hierarchy */
1512 struct list_head tmp_cg_links
;
1513 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1514 struct inode
*inode
;
1515 struct cgroupfs_root
*existing_root
;
1518 BUG_ON(sb
->s_root
!= NULL
);
1520 ret
= cgroup_get_rootdir(sb
);
1522 goto drop_new_super
;
1523 inode
= sb
->s_root
->d_inode
;
1525 mutex_lock(&inode
->i_mutex
);
1526 mutex_lock(&cgroup_mutex
);
1528 if (strlen(root
->name
)) {
1529 /* Check for name clashes with existing mounts */
1530 for_each_active_root(existing_root
) {
1531 if (!strcmp(existing_root
->name
, root
->name
)) {
1533 mutex_unlock(&cgroup_mutex
);
1534 mutex_unlock(&inode
->i_mutex
);
1535 goto drop_new_super
;
1541 * We're accessing css_set_count without locking
1542 * css_set_lock here, but that's OK - it can only be
1543 * increased by someone holding cgroup_lock, and
1544 * that's us. The worst that can happen is that we
1545 * have some link structures left over
1547 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1549 mutex_unlock(&cgroup_mutex
);
1550 mutex_unlock(&inode
->i_mutex
);
1551 goto drop_new_super
;
1554 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1555 if (ret
== -EBUSY
) {
1556 mutex_unlock(&cgroup_mutex
);
1557 mutex_unlock(&inode
->i_mutex
);
1558 free_cg_links(&tmp_cg_links
);
1559 goto drop_new_super
;
1562 * There must be no failure case after here, since rebinding
1563 * takes care of subsystems' refcounts, which are explicitly
1564 * dropped in the failure exit path.
1567 /* EBUSY should be the only error here */
1570 list_add(&root
->root_list
, &roots
);
1573 sb
->s_root
->d_fsdata
= root_cgrp
;
1574 root
->top_cgroup
.dentry
= sb
->s_root
;
1576 /* Link the top cgroup in this hierarchy into all
1577 * the css_set objects */
1578 write_lock(&css_set_lock
);
1579 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1580 struct hlist_head
*hhead
= &css_set_table
[i
];
1581 struct hlist_node
*node
;
1584 hlist_for_each_entry(cg
, node
, hhead
, hlist
)
1585 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1587 write_unlock(&css_set_lock
);
1589 free_cg_links(&tmp_cg_links
);
1591 BUG_ON(!list_empty(&root_cgrp
->sibling
));
1592 BUG_ON(!list_empty(&root_cgrp
->children
));
1593 BUG_ON(root
->number_of_cgroups
!= 1);
1595 cgroup_populate_dir(root_cgrp
);
1596 mutex_unlock(&cgroup_mutex
);
1597 mutex_unlock(&inode
->i_mutex
);
1600 * We re-used an existing hierarchy - the new root (if
1601 * any) is not needed
1603 cgroup_drop_root(opts
.new_root
);
1604 /* no subsys rebinding, so refcounts don't change */
1605 drop_parsed_module_refcounts(opts
.subsys_bits
);
1608 kfree(opts
.release_agent
);
1610 return dget(sb
->s_root
);
1613 deactivate_locked_super(sb
);
1615 drop_parsed_module_refcounts(opts
.subsys_bits
);
1617 kfree(opts
.release_agent
);
1619 return ERR_PTR(ret
);
1622 static void cgroup_kill_sb(struct super_block
*sb
) {
1623 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1624 struct cgroup
*cgrp
= &root
->top_cgroup
;
1626 struct cg_cgroup_link
*link
;
1627 struct cg_cgroup_link
*saved_link
;
1631 BUG_ON(root
->number_of_cgroups
!= 1);
1632 BUG_ON(!list_empty(&cgrp
->children
));
1633 BUG_ON(!list_empty(&cgrp
->sibling
));
1635 mutex_lock(&cgroup_mutex
);
1637 /* Rebind all subsystems back to the default hierarchy */
1638 ret
= rebind_subsystems(root
, 0);
1639 /* Shouldn't be able to fail ... */
1643 * Release all the links from css_sets to this hierarchy's
1646 write_lock(&css_set_lock
);
1648 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1650 list_del(&link
->cg_link_list
);
1651 list_del(&link
->cgrp_link_list
);
1654 write_unlock(&css_set_lock
);
1656 if (!list_empty(&root
->root_list
)) {
1657 list_del(&root
->root_list
);
1661 mutex_unlock(&cgroup_mutex
);
1663 kill_litter_super(sb
);
1664 cgroup_drop_root(root
);
1667 static struct file_system_type cgroup_fs_type
= {
1669 .mount
= cgroup_mount
,
1670 .kill_sb
= cgroup_kill_sb
,
1673 static struct kobject
*cgroup_kobj
;
1675 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1677 return dentry
->d_fsdata
;
1680 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1682 return dentry
->d_fsdata
;
1686 * cgroup_path - generate the path of a cgroup
1687 * @cgrp: the cgroup in question
1688 * @buf: the buffer to write the path into
1689 * @buflen: the length of the buffer
1691 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1692 * reference. Writes path of cgroup into buf. Returns 0 on success,
1695 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1698 struct dentry
*dentry
= rcu_dereference_check(cgrp
->dentry
,
1699 rcu_read_lock_held() ||
1700 cgroup_lock_is_held());
1702 if (!dentry
|| cgrp
== dummytop
) {
1704 * Inactive subsystems have no dentry for their root
1711 start
= buf
+ buflen
;
1715 int len
= dentry
->d_name
.len
;
1717 if ((start
-= len
) < buf
)
1718 return -ENAMETOOLONG
;
1719 memcpy(start
, dentry
->d_name
.name
, len
);
1720 cgrp
= cgrp
->parent
;
1724 dentry
= rcu_dereference_check(cgrp
->dentry
,
1725 rcu_read_lock_held() ||
1726 cgroup_lock_is_held());
1730 return -ENAMETOOLONG
;
1733 memmove(buf
, start
, buf
+ buflen
- start
);
1736 EXPORT_SYMBOL_GPL(cgroup_path
);
1739 * cgroup_task_migrate - move a task from one cgroup to another.
1741 * 'guarantee' is set if the caller promises that a new css_set for the task
1742 * will already exist. If not set, this function might sleep, and can fail with
1743 * -ENOMEM. Otherwise, it can only fail with -ESRCH.
1745 static int cgroup_task_migrate(struct cgroup
*cgrp
, struct cgroup
*oldcgrp
,
1746 struct task_struct
*tsk
, bool guarantee
)
1748 struct css_set
*oldcg
;
1749 struct css_set
*newcg
;
1752 * get old css_set. we need to take task_lock and refcount it, because
1753 * an exiting task can change its css_set to init_css_set and drop its
1754 * old one without taking cgroup_mutex.
1757 oldcg
= tsk
->cgroups
;
1761 /* locate or allocate a new css_set for this task. */
1763 /* we know the css_set we want already exists. */
1764 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
1765 read_lock(&css_set_lock
);
1766 newcg
= find_existing_css_set(oldcg
, cgrp
, template);
1769 read_unlock(&css_set_lock
);
1772 /* find_css_set will give us newcg already referenced. */
1773 newcg
= find_css_set(oldcg
, cgrp
);
1781 /* if PF_EXITING is set, the tsk->cgroups pointer is no longer safe. */
1783 if (tsk
->flags
& PF_EXITING
) {
1788 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1791 /* Update the css_set linked lists if we're using them */
1792 write_lock(&css_set_lock
);
1793 if (!list_empty(&tsk
->cg_list
))
1794 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1795 write_unlock(&css_set_lock
);
1798 * We just gained a reference on oldcg by taking it from the task. As
1799 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1800 * it here; it will be freed under RCU.
1804 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1809 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1810 * @cgrp: the cgroup the task is attaching to
1811 * @tsk: the task to be attached
1813 * Call holding cgroup_mutex. May take task_lock of
1814 * the task 'tsk' during call.
1816 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1819 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1820 struct cgroup
*oldcgrp
;
1821 struct cgroupfs_root
*root
= cgrp
->root
;
1823 /* Nothing to do if the task is already in that cgroup */
1824 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1825 if (cgrp
== oldcgrp
)
1828 for_each_subsys(root
, ss
) {
1829 if (ss
->can_attach
) {
1830 retval
= ss
->can_attach(ss
, cgrp
, tsk
);
1833 * Remember on which subsystem the can_attach()
1834 * failed, so that we only call cancel_attach()
1835 * against the subsystems whose can_attach()
1836 * succeeded. (See below)
1842 if (ss
->can_attach_task
) {
1843 retval
= ss
->can_attach_task(cgrp
, tsk
);
1851 retval
= cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, false);
1855 for_each_subsys(root
, ss
) {
1857 ss
->pre_attach(cgrp
);
1858 if (ss
->attach_task
)
1859 ss
->attach_task(cgrp
, tsk
);
1861 ss
->attach(ss
, cgrp
, oldcgrp
, tsk
);
1867 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1868 * is no longer empty.
1870 cgroup_wakeup_rmdir_waiter(cgrp
);
1873 for_each_subsys(root
, ss
) {
1874 if (ss
== failed_ss
)
1876 * This subsystem was the one that failed the
1877 * can_attach() check earlier, so we don't need
1878 * to call cancel_attach() against it or any
1879 * remaining subsystems.
1882 if (ss
->cancel_attach
)
1883 ss
->cancel_attach(ss
, cgrp
, tsk
);
1890 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1891 * @from: attach to all cgroups of a given task
1892 * @tsk: the task to be attached
1894 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
1896 struct cgroupfs_root
*root
;
1900 for_each_active_root(root
) {
1901 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
1903 retval
= cgroup_attach_task(from_cg
, tsk
);
1911 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
1914 * cgroup_attach_proc works in two stages, the first of which prefetches all
1915 * new css_sets needed (to make sure we have enough memory before committing
1916 * to the move) and stores them in a list of entries of the following type.
1917 * TODO: possible optimization: use css_set->rcu_head for chaining instead
1919 struct cg_list_entry
{
1921 struct list_head links
;
1924 static bool css_set_check_fetched(struct cgroup
*cgrp
,
1925 struct task_struct
*tsk
, struct css_set
*cg
,
1926 struct list_head
*newcg_list
)
1928 struct css_set
*newcg
;
1929 struct cg_list_entry
*cg_entry
;
1930 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
1932 read_lock(&css_set_lock
);
1933 newcg
= find_existing_css_set(cg
, cgrp
, template);
1936 read_unlock(&css_set_lock
);
1938 /* doesn't exist at all? */
1941 /* see if it's already in the list */
1942 list_for_each_entry(cg_entry
, newcg_list
, links
) {
1943 if (cg_entry
->cg
== newcg
) {
1955 * Find the new css_set and store it in the list in preparation for moving the
1956 * given task to the given cgroup. Returns 0 or -ENOMEM.
1958 static int css_set_prefetch(struct cgroup
*cgrp
, struct css_set
*cg
,
1959 struct list_head
*newcg_list
)
1961 struct css_set
*newcg
;
1962 struct cg_list_entry
*cg_entry
;
1964 /* ensure a new css_set will exist for this thread */
1965 newcg
= find_css_set(cg
, cgrp
);
1968 /* add it to the list */
1969 cg_entry
= kmalloc(sizeof(struct cg_list_entry
), GFP_KERNEL
);
1974 cg_entry
->cg
= newcg
;
1975 list_add(&cg_entry
->links
, newcg_list
);
1980 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
1981 * @cgrp: the cgroup to attach to
1982 * @leader: the threadgroup leader task_struct of the group to be attached
1984 * Call holding cgroup_mutex and the threadgroup_fork_lock of the leader. Will
1985 * take task_lock of each thread in leader's threadgroup individually in turn.
1987 int cgroup_attach_proc(struct cgroup
*cgrp
, struct task_struct
*leader
)
1989 int retval
, i
, group_size
;
1990 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1991 bool cancel_failed_ss
= false;
1992 /* guaranteed to be initialized later, but the compiler needs this */
1993 struct cgroup
*oldcgrp
= NULL
;
1994 struct css_set
*oldcg
;
1995 struct cgroupfs_root
*root
= cgrp
->root
;
1996 /* threadgroup list cursor and array */
1997 struct task_struct
*tsk
;
1998 struct task_struct
**group
;
2000 * we need to make sure we have css_sets for all the tasks we're
2001 * going to move -before- we actually start moving them, so that in
2002 * case we get an ENOMEM we can bail out before making any changes.
2004 struct list_head newcg_list
;
2005 struct cg_list_entry
*cg_entry
, *temp_nobe
;
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 - threadgroup_fork_lock prevents new threads from appearing,
2012 * and if threads exit, this will just be an over-estimate.
2014 group_size
= get_nr_threads(leader
);
2015 group
= kmalloc(group_size
* sizeof(*group
), GFP_KERNEL
);
2019 /* prevent changes to the threadgroup list while we take a snapshot. */
2021 if (!thread_group_leader(leader
)) {
2023 * a race with de_thread from another thread's exec() may strip
2024 * us of our leadership, making while_each_thread unsafe to use
2025 * on this task. if this happens, there is no choice but to
2026 * throw this task away and try again (from cgroup_procs_write);
2027 * this is "double-double-toil-and-trouble-check locking".
2031 goto out_free_group_list
;
2033 /* take a reference on each task in the group to go in the array. */
2037 /* as per above, nr_threads may decrease, but not increase. */
2038 BUG_ON(i
>= group_size
);
2039 get_task_struct(tsk
);
2042 } while_each_thread(leader
, tsk
);
2043 /* remember the number of threads in the array for later. */
2048 * step 1: check that we can legitimately attach to the cgroup.
2050 for_each_subsys(root
, ss
) {
2051 if (ss
->can_attach
) {
2052 retval
= ss
->can_attach(ss
, cgrp
, leader
);
2055 goto out_cancel_attach
;
2058 /* a callback to be run on every thread in the threadgroup. */
2059 if (ss
->can_attach_task
) {
2060 /* run on each task in the threadgroup. */
2061 for (i
= 0; i
< group_size
; i
++) {
2062 retval
= ss
->can_attach_task(cgrp
, group
[i
]);
2065 cancel_failed_ss
= true;
2066 goto out_cancel_attach
;
2073 * step 2: make sure css_sets exist for all threads to be migrated.
2074 * we use find_css_set, which allocates a new one if necessary.
2076 INIT_LIST_HEAD(&newcg_list
);
2077 for (i
= 0; i
< group_size
; i
++) {
2079 /* nothing to do if this task is already in the cgroup */
2080 oldcgrp
= task_cgroup_from_root(tsk
, root
);
2081 if (cgrp
== oldcgrp
)
2083 /* get old css_set pointer */
2085 if (tsk
->flags
& PF_EXITING
) {
2086 /* ignore this task if it's going away */
2090 oldcg
= tsk
->cgroups
;
2093 /* see if the new one for us is already in the list? */
2094 if (css_set_check_fetched(cgrp
, tsk
, oldcg
, &newcg_list
)) {
2095 /* was already there, nothing to do. */
2098 /* we don't already have it. get new one. */
2099 retval
= css_set_prefetch(cgrp
, oldcg
, &newcg_list
);
2102 goto out_list_teardown
;
2107 * step 3: now that we're guaranteed success wrt the css_sets, proceed
2108 * to move all tasks to the new cgroup, calling ss->attach_task for each
2109 * one along the way. there are no failure cases after here, so this is
2112 for_each_subsys(root
, ss
) {
2114 ss
->pre_attach(cgrp
);
2116 for (i
= 0; i
< group_size
; i
++) {
2118 /* leave current thread as it is if it's already there */
2119 oldcgrp
= task_cgroup_from_root(tsk
, root
);
2120 if (cgrp
== oldcgrp
)
2122 /* attach each task to each subsystem */
2123 for_each_subsys(root
, ss
) {
2124 if (ss
->attach_task
)
2125 ss
->attach_task(cgrp
, tsk
);
2127 /* if the thread is PF_EXITING, it can just get skipped. */
2128 retval
= cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, true);
2129 BUG_ON(retval
!= 0 && retval
!= -ESRCH
);
2131 /* nothing is sensitive to fork() after this point. */
2134 * step 4: do expensive, non-thread-specific subsystem callbacks.
2135 * TODO: if ever a subsystem needs to know the oldcgrp for each task
2136 * being moved, this call will need to be reworked to communicate that.
2138 for_each_subsys(root
, ss
) {
2140 ss
->attach(ss
, cgrp
, oldcgrp
, leader
);
2144 * step 5: success! and cleanup
2147 cgroup_wakeup_rmdir_waiter(cgrp
);
2150 /* clean up the list of prefetched css_sets. */
2151 list_for_each_entry_safe(cg_entry
, temp_nobe
, &newcg_list
, links
) {
2152 list_del(&cg_entry
->links
);
2153 put_css_set(cg_entry
->cg
);
2157 /* same deal as in cgroup_attach_task */
2159 for_each_subsys(root
, ss
) {
2160 if (ss
== failed_ss
) {
2161 if (cancel_failed_ss
&& ss
->cancel_attach
)
2162 ss
->cancel_attach(ss
, cgrp
, leader
);
2165 if (ss
->cancel_attach
)
2166 ss
->cancel_attach(ss
, cgrp
, leader
);
2169 /* clean up the array of referenced threads in the group. */
2170 for (i
= 0; i
< group_size
; i
++)
2171 put_task_struct(group
[i
]);
2172 out_free_group_list
:
2178 * Find the task_struct of the task to attach by vpid and pass it along to the
2179 * function to attach either it or all tasks in its threadgroup. Will take
2180 * cgroup_mutex; may take task_lock of task.
2182 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2184 struct task_struct
*tsk
;
2185 const struct cred
*cred
= current_cred(), *tcred
;
2188 if (!cgroup_lock_live_group(cgrp
))
2193 tsk
= find_task_by_vpid(pid
);
2201 * RCU protects this access, since tsk was found in the
2202 * tid map. a race with de_thread may cause group_leader
2203 * to stop being the leader, but cgroup_attach_proc will
2206 tsk
= tsk
->group_leader
;
2207 } else if (tsk
->flags
& PF_EXITING
) {
2208 /* optimization for the single-task-only case */
2215 * even if we're attaching all tasks in the thread group, we
2216 * only need to check permissions on one of them.
2218 tcred
= __task_cred(tsk
);
2220 cred
->euid
!= tcred
->uid
&&
2221 cred
->euid
!= tcred
->suid
) {
2226 get_task_struct(tsk
);
2230 tsk
= current
->group_leader
;
2233 get_task_struct(tsk
);
2237 threadgroup_fork_write_lock(tsk
);
2238 ret
= cgroup_attach_proc(cgrp
, tsk
);
2239 threadgroup_fork_write_unlock(tsk
);
2241 ret
= cgroup_attach_task(cgrp
, tsk
);
2243 put_task_struct(tsk
);
2248 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2250 return attach_task_by_pid(cgrp
, pid
, false);
2253 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2258 * attach_proc fails with -EAGAIN if threadgroup leadership
2259 * changes in the middle of the operation, in which case we need
2260 * to find the task_struct for the new leader and start over.
2262 ret
= attach_task_by_pid(cgrp
, tgid
, true);
2263 } while (ret
== -EAGAIN
);
2268 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2269 * @cgrp: the cgroup to be checked for liveness
2271 * On success, returns true; the lock should be later released with
2272 * cgroup_unlock(). On failure returns false with no lock held.
2274 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
2276 mutex_lock(&cgroup_mutex
);
2277 if (cgroup_is_removed(cgrp
)) {
2278 mutex_unlock(&cgroup_mutex
);
2283 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
2285 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2288 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2289 if (strlen(buffer
) >= PATH_MAX
)
2291 if (!cgroup_lock_live_group(cgrp
))
2293 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2298 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2299 struct seq_file
*seq
)
2301 if (!cgroup_lock_live_group(cgrp
))
2303 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2304 seq_putc(seq
, '\n');
2309 /* A buffer size big enough for numbers or short strings */
2310 #define CGROUP_LOCAL_BUFFER_SIZE 64
2312 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2314 const char __user
*userbuf
,
2315 size_t nbytes
, loff_t
*unused_ppos
)
2317 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2323 if (nbytes
>= sizeof(buffer
))
2325 if (copy_from_user(buffer
, userbuf
, nbytes
))
2328 buffer
[nbytes
] = 0; /* nul-terminate */
2329 if (cft
->write_u64
) {
2330 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2333 retval
= cft
->write_u64(cgrp
, cft
, val
);
2335 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2338 retval
= cft
->write_s64(cgrp
, cft
, val
);
2345 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2347 const char __user
*userbuf
,
2348 size_t nbytes
, loff_t
*unused_ppos
)
2350 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2352 size_t max_bytes
= cft
->max_write_len
;
2353 char *buffer
= local_buffer
;
2356 max_bytes
= sizeof(local_buffer
) - 1;
2357 if (nbytes
>= max_bytes
)
2359 /* Allocate a dynamic buffer if we need one */
2360 if (nbytes
>= sizeof(local_buffer
)) {
2361 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2365 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2370 buffer
[nbytes
] = 0; /* nul-terminate */
2371 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2375 if (buffer
!= local_buffer
)
2380 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2381 size_t nbytes
, loff_t
*ppos
)
2383 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2384 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2386 if (cgroup_is_removed(cgrp
))
2389 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2390 if (cft
->write_u64
|| cft
->write_s64
)
2391 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2392 if (cft
->write_string
)
2393 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2395 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2396 return ret
? ret
: nbytes
;
2401 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2403 char __user
*buf
, size_t nbytes
,
2406 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2407 u64 val
= cft
->read_u64(cgrp
, cft
);
2408 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2410 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2413 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2415 char __user
*buf
, size_t nbytes
,
2418 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2419 s64 val
= cft
->read_s64(cgrp
, cft
);
2420 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2422 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2425 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2426 size_t nbytes
, loff_t
*ppos
)
2428 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2429 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2431 if (cgroup_is_removed(cgrp
))
2435 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2437 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2439 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2444 * seqfile ops/methods for returning structured data. Currently just
2445 * supports string->u64 maps, but can be extended in future.
2448 struct cgroup_seqfile_state
{
2450 struct cgroup
*cgroup
;
2453 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2455 struct seq_file
*sf
= cb
->state
;
2456 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2459 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2461 struct cgroup_seqfile_state
*state
= m
->private;
2462 struct cftype
*cft
= state
->cft
;
2463 if (cft
->read_map
) {
2464 struct cgroup_map_cb cb
= {
2465 .fill
= cgroup_map_add
,
2468 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2470 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2473 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2475 struct seq_file
*seq
= file
->private_data
;
2476 kfree(seq
->private);
2477 return single_release(inode
, file
);
2480 static const struct file_operations cgroup_seqfile_operations
= {
2482 .write
= cgroup_file_write
,
2483 .llseek
= seq_lseek
,
2484 .release
= cgroup_seqfile_release
,
2487 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2492 err
= generic_file_open(inode
, file
);
2495 cft
= __d_cft(file
->f_dentry
);
2497 if (cft
->read_map
|| cft
->read_seq_string
) {
2498 struct cgroup_seqfile_state
*state
=
2499 kzalloc(sizeof(*state
), GFP_USER
);
2503 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2504 file
->f_op
= &cgroup_seqfile_operations
;
2505 err
= single_open(file
, cgroup_seqfile_show
, state
);
2508 } else if (cft
->open
)
2509 err
= cft
->open(inode
, file
);
2516 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2518 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2520 return cft
->release(inode
, file
);
2525 * cgroup_rename - Only allow simple rename of directories in place.
2527 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2528 struct inode
*new_dir
, struct dentry
*new_dentry
)
2530 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2532 if (new_dentry
->d_inode
)
2534 if (old_dir
!= new_dir
)
2536 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2539 static const struct file_operations cgroup_file_operations
= {
2540 .read
= cgroup_file_read
,
2541 .write
= cgroup_file_write
,
2542 .llseek
= generic_file_llseek
,
2543 .open
= cgroup_file_open
,
2544 .release
= cgroup_file_release
,
2547 static const struct inode_operations cgroup_dir_inode_operations
= {
2548 .lookup
= cgroup_lookup
,
2549 .mkdir
= cgroup_mkdir
,
2550 .rmdir
= cgroup_rmdir
,
2551 .rename
= cgroup_rename
,
2554 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
2556 if (dentry
->d_name
.len
> NAME_MAX
)
2557 return ERR_PTR(-ENAMETOOLONG
);
2558 d_add(dentry
, NULL
);
2563 * Check if a file is a control file
2565 static inline struct cftype
*__file_cft(struct file
*file
)
2567 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2568 return ERR_PTR(-EINVAL
);
2569 return __d_cft(file
->f_dentry
);
2572 static int cgroup_create_file(struct dentry
*dentry
, mode_t mode
,
2573 struct super_block
*sb
)
2575 struct inode
*inode
;
2579 if (dentry
->d_inode
)
2582 inode
= cgroup_new_inode(mode
, sb
);
2586 if (S_ISDIR(mode
)) {
2587 inode
->i_op
= &cgroup_dir_inode_operations
;
2588 inode
->i_fop
= &simple_dir_operations
;
2590 /* start off with i_nlink == 2 (for "." entry) */
2593 /* start with the directory inode held, so that we can
2594 * populate it without racing with another mkdir */
2595 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2596 } else if (S_ISREG(mode
)) {
2598 inode
->i_fop
= &cgroup_file_operations
;
2600 d_instantiate(dentry
, inode
);
2601 dget(dentry
); /* Extra count - pin the dentry in core */
2606 * cgroup_create_dir - create a directory for an object.
2607 * @cgrp: the cgroup we create the directory for. It must have a valid
2608 * ->parent field. And we are going to fill its ->dentry field.
2609 * @dentry: dentry of the new cgroup
2610 * @mode: mode to set on new directory.
2612 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
2615 struct dentry
*parent
;
2618 parent
= cgrp
->parent
->dentry
;
2619 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
2621 dentry
->d_fsdata
= cgrp
;
2622 inc_nlink(parent
->d_inode
);
2623 rcu_assign_pointer(cgrp
->dentry
, dentry
);
2632 * cgroup_file_mode - deduce file mode of a control file
2633 * @cft: the control file in question
2635 * returns cft->mode if ->mode is not 0
2636 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2637 * returns S_IRUGO if it has only a read handler
2638 * returns S_IWUSR if it has only a write hander
2640 static mode_t
cgroup_file_mode(const struct cftype
*cft
)
2647 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2648 cft
->read_map
|| cft
->read_seq_string
)
2651 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2652 cft
->write_string
|| cft
->trigger
)
2658 int cgroup_add_file(struct cgroup
*cgrp
,
2659 struct cgroup_subsys
*subsys
,
2660 const struct cftype
*cft
)
2662 struct dentry
*dir
= cgrp
->dentry
;
2663 struct dentry
*dentry
;
2667 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2668 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2669 strcpy(name
, subsys
->name
);
2672 strcat(name
, cft
->name
);
2673 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2674 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2675 if (!IS_ERR(dentry
)) {
2676 mode
= cgroup_file_mode(cft
);
2677 error
= cgroup_create_file(dentry
, mode
| S_IFREG
,
2680 dentry
->d_fsdata
= (void *)cft
;
2683 error
= PTR_ERR(dentry
);
2686 EXPORT_SYMBOL_GPL(cgroup_add_file
);
2688 int cgroup_add_files(struct cgroup
*cgrp
,
2689 struct cgroup_subsys
*subsys
,
2690 const struct cftype cft
[],
2694 for (i
= 0; i
< count
; i
++) {
2695 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
2701 EXPORT_SYMBOL_GPL(cgroup_add_files
);
2704 * cgroup_task_count - count the number of tasks in a cgroup.
2705 * @cgrp: the cgroup in question
2707 * Return the number of tasks in the cgroup.
2709 int cgroup_task_count(const struct cgroup
*cgrp
)
2712 struct cg_cgroup_link
*link
;
2714 read_lock(&css_set_lock
);
2715 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2716 count
+= atomic_read(&link
->cg
->refcount
);
2718 read_unlock(&css_set_lock
);
2723 * Advance a list_head iterator. The iterator should be positioned at
2724 * the start of a css_set
2726 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2727 struct cgroup_iter
*it
)
2729 struct list_head
*l
= it
->cg_link
;
2730 struct cg_cgroup_link
*link
;
2733 /* Advance to the next non-empty css_set */
2736 if (l
== &cgrp
->css_sets
) {
2740 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2742 } while (list_empty(&cg
->tasks
));
2744 it
->task
= cg
->tasks
.next
;
2748 * To reduce the fork() overhead for systems that are not actually
2749 * using their cgroups capability, we don't maintain the lists running
2750 * through each css_set to its tasks until we see the list actually
2751 * used - in other words after the first call to cgroup_iter_start().
2753 * The tasklist_lock is not held here, as do_each_thread() and
2754 * while_each_thread() are protected by RCU.
2756 static void cgroup_enable_task_cg_lists(void)
2758 struct task_struct
*p
, *g
;
2759 write_lock(&css_set_lock
);
2760 use_task_css_set_links
= 1;
2761 do_each_thread(g
, p
) {
2764 * We should check if the process is exiting, otherwise
2765 * it will race with cgroup_exit() in that the list
2766 * entry won't be deleted though the process has exited.
2768 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2769 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2771 } while_each_thread(g
, p
);
2772 write_unlock(&css_set_lock
);
2775 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2778 * The first time anyone tries to iterate across a cgroup,
2779 * we need to enable the list linking each css_set to its
2780 * tasks, and fix up all existing tasks.
2782 if (!use_task_css_set_links
)
2783 cgroup_enable_task_cg_lists();
2785 read_lock(&css_set_lock
);
2786 it
->cg_link
= &cgrp
->css_sets
;
2787 cgroup_advance_iter(cgrp
, it
);
2790 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
2791 struct cgroup_iter
*it
)
2793 struct task_struct
*res
;
2794 struct list_head
*l
= it
->task
;
2795 struct cg_cgroup_link
*link
;
2797 /* If the iterator cg is NULL, we have no tasks */
2800 res
= list_entry(l
, struct task_struct
, cg_list
);
2801 /* Advance iterator to find next entry */
2803 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
2804 if (l
== &link
->cg
->tasks
) {
2805 /* We reached the end of this task list - move on to
2806 * the next cg_cgroup_link */
2807 cgroup_advance_iter(cgrp
, it
);
2814 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2816 read_unlock(&css_set_lock
);
2819 static inline int started_after_time(struct task_struct
*t1
,
2820 struct timespec
*time
,
2821 struct task_struct
*t2
)
2823 int start_diff
= timespec_compare(&t1
->start_time
, time
);
2824 if (start_diff
> 0) {
2826 } else if (start_diff
< 0) {
2830 * Arbitrarily, if two processes started at the same
2831 * time, we'll say that the lower pointer value
2832 * started first. Note that t2 may have exited by now
2833 * so this may not be a valid pointer any longer, but
2834 * that's fine - it still serves to distinguish
2835 * between two tasks started (effectively) simultaneously.
2842 * This function is a callback from heap_insert() and is used to order
2844 * In this case we order the heap in descending task start time.
2846 static inline int started_after(void *p1
, void *p2
)
2848 struct task_struct
*t1
= p1
;
2849 struct task_struct
*t2
= p2
;
2850 return started_after_time(t1
, &t2
->start_time
, t2
);
2854 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2855 * @scan: struct cgroup_scanner containing arguments for the scan
2857 * Arguments include pointers to callback functions test_task() and
2859 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2860 * and if it returns true, call process_task() for it also.
2861 * The test_task pointer may be NULL, meaning always true (select all tasks).
2862 * Effectively duplicates cgroup_iter_{start,next,end}()
2863 * but does not lock css_set_lock for the call to process_task().
2864 * The struct cgroup_scanner may be embedded in any structure of the caller's
2866 * It is guaranteed that process_task() will act on every task that
2867 * is a member of the cgroup for the duration of this call. This
2868 * function may or may not call process_task() for tasks that exit
2869 * or move to a different cgroup during the call, or are forked or
2870 * move into the cgroup during the call.
2872 * Note that test_task() may be called with locks held, and may in some
2873 * situations be called multiple times for the same task, so it should
2875 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2876 * pre-allocated and will be used for heap operations (and its "gt" member will
2877 * be overwritten), else a temporary heap will be used (allocation of which
2878 * may cause this function to fail).
2880 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
2883 struct cgroup_iter it
;
2884 struct task_struct
*p
, *dropped
;
2885 /* Never dereference latest_task, since it's not refcounted */
2886 struct task_struct
*latest_task
= NULL
;
2887 struct ptr_heap tmp_heap
;
2888 struct ptr_heap
*heap
;
2889 struct timespec latest_time
= { 0, 0 };
2892 /* The caller supplied our heap and pre-allocated its memory */
2894 heap
->gt
= &started_after
;
2896 /* We need to allocate our own heap memory */
2898 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
2900 /* cannot allocate the heap */
2906 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2907 * to determine which are of interest, and using the scanner's
2908 * "process_task" callback to process any of them that need an update.
2909 * Since we don't want to hold any locks during the task updates,
2910 * gather tasks to be processed in a heap structure.
2911 * The heap is sorted by descending task start time.
2912 * If the statically-sized heap fills up, we overflow tasks that
2913 * started later, and in future iterations only consider tasks that
2914 * started after the latest task in the previous pass. This
2915 * guarantees forward progress and that we don't miss any tasks.
2918 cgroup_iter_start(scan
->cg
, &it
);
2919 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
2921 * Only affect tasks that qualify per the caller's callback,
2922 * if he provided one
2924 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
2927 * Only process tasks that started after the last task
2930 if (!started_after_time(p
, &latest_time
, latest_task
))
2932 dropped
= heap_insert(heap
, p
);
2933 if (dropped
== NULL
) {
2935 * The new task was inserted; the heap wasn't
2939 } else if (dropped
!= p
) {
2941 * The new task was inserted, and pushed out a
2945 put_task_struct(dropped
);
2948 * Else the new task was newer than anything already in
2949 * the heap and wasn't inserted
2952 cgroup_iter_end(scan
->cg
, &it
);
2955 for (i
= 0; i
< heap
->size
; i
++) {
2956 struct task_struct
*q
= heap
->ptrs
[i
];
2958 latest_time
= q
->start_time
;
2961 /* Process the task per the caller's callback */
2962 scan
->process_task(q
, scan
);
2966 * If we had to process any tasks at all, scan again
2967 * in case some of them were in the middle of forking
2968 * children that didn't get processed.
2969 * Not the most efficient way to do it, but it avoids
2970 * having to take callback_mutex in the fork path
2974 if (heap
== &tmp_heap
)
2975 heap_free(&tmp_heap
);
2980 * Stuff for reading the 'tasks'/'procs' files.
2982 * Reading this file can return large amounts of data if a cgroup has
2983 * *lots* of attached tasks. So it may need several calls to read(),
2984 * but we cannot guarantee that the information we produce is correct
2985 * unless we produce it entirely atomically.
2990 * The following two functions "fix" the issue where there are more pids
2991 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2992 * TODO: replace with a kernel-wide solution to this problem
2994 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2995 static void *pidlist_allocate(int count
)
2997 if (PIDLIST_TOO_LARGE(count
))
2998 return vmalloc(count
* sizeof(pid_t
));
3000 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3002 static void pidlist_free(void *p
)
3004 if (is_vmalloc_addr(p
))
3009 static void *pidlist_resize(void *p
, int newcount
)
3012 /* note: if new alloc fails, old p will still be valid either way */
3013 if (is_vmalloc_addr(p
)) {
3014 newlist
= vmalloc(newcount
* sizeof(pid_t
));
3017 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
3020 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
3026 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3027 * If the new stripped list is sufficiently smaller and there's enough memory
3028 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3029 * number of unique elements.
3031 /* is the size difference enough that we should re-allocate the array? */
3032 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3033 static int pidlist_uniq(pid_t
**p
, int length
)
3040 * we presume the 0th element is unique, so i starts at 1. trivial
3041 * edge cases first; no work needs to be done for either
3043 if (length
== 0 || length
== 1)
3045 /* src and dest walk down the list; dest counts unique elements */
3046 for (src
= 1; src
< length
; src
++) {
3047 /* find next unique element */
3048 while (list
[src
] == list
[src
-1]) {
3053 /* dest always points to where the next unique element goes */
3054 list
[dest
] = list
[src
];
3059 * if the length difference is large enough, we want to allocate a
3060 * smaller buffer to save memory. if this fails due to out of memory,
3061 * we'll just stay with what we've got.
3063 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
3064 newlist
= pidlist_resize(list
, dest
);
3071 static int cmppid(const void *a
, const void *b
)
3073 return *(pid_t
*)a
- *(pid_t
*)b
;
3077 * find the appropriate pidlist for our purpose (given procs vs tasks)
3078 * returns with the lock on that pidlist already held, and takes care
3079 * of the use count, or returns NULL with no locks held if we're out of
3082 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3083 enum cgroup_filetype type
)
3085 struct cgroup_pidlist
*l
;
3086 /* don't need task_nsproxy() if we're looking at ourself */
3087 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
3090 * We can't drop the pidlist_mutex before taking the l->mutex in case
3091 * the last ref-holder is trying to remove l from the list at the same
3092 * time. Holding the pidlist_mutex precludes somebody taking whichever
3093 * list we find out from under us - compare release_pid_array().
3095 mutex_lock(&cgrp
->pidlist_mutex
);
3096 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3097 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3098 /* make sure l doesn't vanish out from under us */
3099 down_write(&l
->mutex
);
3100 mutex_unlock(&cgrp
->pidlist_mutex
);
3104 /* entry not found; create a new one */
3105 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3107 mutex_unlock(&cgrp
->pidlist_mutex
);
3110 init_rwsem(&l
->mutex
);
3111 down_write(&l
->mutex
);
3113 l
->key
.ns
= get_pid_ns(ns
);
3114 l
->use_count
= 0; /* don't increment here */
3117 list_add(&l
->links
, &cgrp
->pidlists
);
3118 mutex_unlock(&cgrp
->pidlist_mutex
);
3123 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3125 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3126 struct cgroup_pidlist
**lp
)
3130 int pid
, n
= 0; /* used for populating the array */
3131 struct cgroup_iter it
;
3132 struct task_struct
*tsk
;
3133 struct cgroup_pidlist
*l
;
3136 * If cgroup gets more users after we read count, we won't have
3137 * enough space - tough. This race is indistinguishable to the
3138 * caller from the case that the additional cgroup users didn't
3139 * show up until sometime later on.
3141 length
= cgroup_task_count(cgrp
);
3142 array
= pidlist_allocate(length
);
3145 /* now, populate the array */
3146 cgroup_iter_start(cgrp
, &it
);
3147 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3148 if (unlikely(n
== length
))
3150 /* get tgid or pid for procs or tasks file respectively */
3151 if (type
== CGROUP_FILE_PROCS
)
3152 pid
= task_tgid_vnr(tsk
);
3154 pid
= task_pid_vnr(tsk
);
3155 if (pid
> 0) /* make sure to only use valid results */
3158 cgroup_iter_end(cgrp
, &it
);
3160 /* now sort & (if procs) strip out duplicates */
3161 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3162 if (type
== CGROUP_FILE_PROCS
)
3163 length
= pidlist_uniq(&array
, length
);
3164 l
= cgroup_pidlist_find(cgrp
, type
);
3166 pidlist_free(array
);
3169 /* store array, freeing old if necessary - lock already held */
3170 pidlist_free(l
->list
);
3174 up_write(&l
->mutex
);
3180 * cgroupstats_build - build and fill cgroupstats
3181 * @stats: cgroupstats to fill information into
3182 * @dentry: A dentry entry belonging to the cgroup for which stats have
3185 * Build and fill cgroupstats so that taskstats can export it to user
3188 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3191 struct cgroup
*cgrp
;
3192 struct cgroup_iter it
;
3193 struct task_struct
*tsk
;
3196 * Validate dentry by checking the superblock operations,
3197 * and make sure it's a directory.
3199 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3200 !S_ISDIR(dentry
->d_inode
->i_mode
))
3204 cgrp
= dentry
->d_fsdata
;
3206 cgroup_iter_start(cgrp
, &it
);
3207 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3208 switch (tsk
->state
) {
3210 stats
->nr_running
++;
3212 case TASK_INTERRUPTIBLE
:
3213 stats
->nr_sleeping
++;
3215 case TASK_UNINTERRUPTIBLE
:
3216 stats
->nr_uninterruptible
++;
3219 stats
->nr_stopped
++;
3222 if (delayacct_is_task_waiting_on_io(tsk
))
3223 stats
->nr_io_wait
++;
3227 cgroup_iter_end(cgrp
, &it
);
3235 * seq_file methods for the tasks/procs files. The seq_file position is the
3236 * next pid to display; the seq_file iterator is a pointer to the pid
3237 * in the cgroup->l->list array.
3240 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3243 * Initially we receive a position value that corresponds to
3244 * one more than the last pid shown (or 0 on the first call or
3245 * after a seek to the start). Use a binary-search to find the
3246 * next pid to display, if any
3248 struct cgroup_pidlist
*l
= s
->private;
3249 int index
= 0, pid
= *pos
;
3252 down_read(&l
->mutex
);
3254 int end
= l
->length
;
3256 while (index
< end
) {
3257 int mid
= (index
+ end
) / 2;
3258 if (l
->list
[mid
] == pid
) {
3261 } else if (l
->list
[mid
] <= pid
)
3267 /* If we're off the end of the array, we're done */
3268 if (index
>= l
->length
)
3270 /* Update the abstract position to be the actual pid that we found */
3271 iter
= l
->list
+ index
;
3276 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3278 struct cgroup_pidlist
*l
= s
->private;
3282 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3284 struct cgroup_pidlist
*l
= s
->private;
3286 pid_t
*end
= l
->list
+ l
->length
;
3288 * Advance to the next pid in the array. If this goes off the
3300 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3302 return seq_printf(s
, "%d\n", *(int *)v
);
3306 * seq_operations functions for iterating on pidlists through seq_file -
3307 * independent of whether it's tasks or procs
3309 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3310 .start
= cgroup_pidlist_start
,
3311 .stop
= cgroup_pidlist_stop
,
3312 .next
= cgroup_pidlist_next
,
3313 .show
= cgroup_pidlist_show
,
3316 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3319 * the case where we're the last user of this particular pidlist will
3320 * have us remove it from the cgroup's list, which entails taking the
3321 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3322 * pidlist_mutex, we have to take pidlist_mutex first.
3324 mutex_lock(&l
->owner
->pidlist_mutex
);
3325 down_write(&l
->mutex
);
3326 BUG_ON(!l
->use_count
);
3327 if (!--l
->use_count
) {
3328 /* we're the last user if refcount is 0; remove and free */
3329 list_del(&l
->links
);
3330 mutex_unlock(&l
->owner
->pidlist_mutex
);
3331 pidlist_free(l
->list
);
3332 put_pid_ns(l
->key
.ns
);
3333 up_write(&l
->mutex
);
3337 mutex_unlock(&l
->owner
->pidlist_mutex
);
3338 up_write(&l
->mutex
);
3341 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3343 struct cgroup_pidlist
*l
;
3344 if (!(file
->f_mode
& FMODE_READ
))
3347 * the seq_file will only be initialized if the file was opened for
3348 * reading; hence we check if it's not null only in that case.
3350 l
= ((struct seq_file
*)file
->private_data
)->private;
3351 cgroup_release_pid_array(l
);
3352 return seq_release(inode
, file
);
3355 static const struct file_operations cgroup_pidlist_operations
= {
3357 .llseek
= seq_lseek
,
3358 .write
= cgroup_file_write
,
3359 .release
= cgroup_pidlist_release
,
3363 * The following functions handle opens on a file that displays a pidlist
3364 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3367 /* helper function for the two below it */
3368 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3370 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3371 struct cgroup_pidlist
*l
;
3374 /* Nothing to do for write-only files */
3375 if (!(file
->f_mode
& FMODE_READ
))
3378 /* have the array populated */
3379 retval
= pidlist_array_load(cgrp
, type
, &l
);
3382 /* configure file information */
3383 file
->f_op
= &cgroup_pidlist_operations
;
3385 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3387 cgroup_release_pid_array(l
);
3390 ((struct seq_file
*)file
->private_data
)->private = l
;
3393 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3395 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3397 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3399 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3402 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3405 return notify_on_release(cgrp
);
3408 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3412 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3414 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3416 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3421 * Unregister event and free resources.
3423 * Gets called from workqueue.
3425 static void cgroup_event_remove(struct work_struct
*work
)
3427 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3429 struct cgroup
*cgrp
= event
->cgrp
;
3431 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3433 eventfd_ctx_put(event
->eventfd
);
3439 * Gets called on POLLHUP on eventfd when user closes it.
3441 * Called with wqh->lock held and interrupts disabled.
3443 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3444 int sync
, void *key
)
3446 struct cgroup_event
*event
= container_of(wait
,
3447 struct cgroup_event
, wait
);
3448 struct cgroup
*cgrp
= event
->cgrp
;
3449 unsigned long flags
= (unsigned long)key
;
3451 if (flags
& POLLHUP
) {
3452 __remove_wait_queue(event
->wqh
, &event
->wait
);
3453 spin_lock(&cgrp
->event_list_lock
);
3454 list_del(&event
->list
);
3455 spin_unlock(&cgrp
->event_list_lock
);
3457 * We are in atomic context, but cgroup_event_remove() may
3458 * sleep, so we have to call it in workqueue.
3460 schedule_work(&event
->remove
);
3466 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3467 wait_queue_head_t
*wqh
, poll_table
*pt
)
3469 struct cgroup_event
*event
= container_of(pt
,
3470 struct cgroup_event
, pt
);
3473 add_wait_queue(wqh
, &event
->wait
);
3477 * Parse input and register new cgroup event handler.
3479 * Input must be in format '<event_fd> <control_fd> <args>'.
3480 * Interpretation of args is defined by control file implementation.
3482 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3485 struct cgroup_event
*event
= NULL
;
3486 unsigned int efd
, cfd
;
3487 struct file
*efile
= NULL
;
3488 struct file
*cfile
= NULL
;
3492 efd
= simple_strtoul(buffer
, &endp
, 10);
3497 cfd
= simple_strtoul(buffer
, &endp
, 10);
3498 if ((*endp
!= ' ') && (*endp
!= '\0'))
3502 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3506 INIT_LIST_HEAD(&event
->list
);
3507 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3508 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3509 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3511 efile
= eventfd_fget(efd
);
3512 if (IS_ERR(efile
)) {
3513 ret
= PTR_ERR(efile
);
3517 event
->eventfd
= eventfd_ctx_fileget(efile
);
3518 if (IS_ERR(event
->eventfd
)) {
3519 ret
= PTR_ERR(event
->eventfd
);
3529 /* the process need read permission on control file */
3530 ret
= file_permission(cfile
, MAY_READ
);
3534 event
->cft
= __file_cft(cfile
);
3535 if (IS_ERR(event
->cft
)) {
3536 ret
= PTR_ERR(event
->cft
);
3540 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3545 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3546 event
->eventfd
, buffer
);
3550 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3551 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3557 * Events should be removed after rmdir of cgroup directory, but before
3558 * destroying subsystem state objects. Let's take reference to cgroup
3559 * directory dentry to do that.
3563 spin_lock(&cgrp
->event_list_lock
);
3564 list_add(&event
->list
, &cgrp
->event_list
);
3565 spin_unlock(&cgrp
->event_list_lock
);
3576 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3577 eventfd_ctx_put(event
->eventfd
);
3579 if (!IS_ERR_OR_NULL(efile
))
3587 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3590 return clone_children(cgrp
);
3593 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3598 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3600 clear_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3605 * for the common functions, 'private' gives the type of file
3607 /* for hysterical raisins, we can't put this on the older files */
3608 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3609 static struct cftype files
[] = {
3612 .open
= cgroup_tasks_open
,
3613 .write_u64
= cgroup_tasks_write
,
3614 .release
= cgroup_pidlist_release
,
3615 .mode
= S_IRUGO
| S_IWUSR
,
3618 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3619 .open
= cgroup_procs_open
,
3620 .write_u64
= cgroup_procs_write
,
3621 .release
= cgroup_pidlist_release
,
3622 .mode
= S_IRUGO
| S_IWUSR
,
3625 .name
= "notify_on_release",
3626 .read_u64
= cgroup_read_notify_on_release
,
3627 .write_u64
= cgroup_write_notify_on_release
,
3630 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3631 .write_string
= cgroup_write_event_control
,
3635 .name
= "cgroup.clone_children",
3636 .read_u64
= cgroup_clone_children_read
,
3637 .write_u64
= cgroup_clone_children_write
,
3641 static struct cftype cft_release_agent
= {
3642 .name
= "release_agent",
3643 .read_seq_string
= cgroup_release_agent_show
,
3644 .write_string
= cgroup_release_agent_write
,
3645 .max_write_len
= PATH_MAX
,
3648 static int cgroup_populate_dir(struct cgroup
*cgrp
)
3651 struct cgroup_subsys
*ss
;
3653 /* First clear out any existing files */
3654 cgroup_clear_directory(cgrp
->dentry
);
3656 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
3660 if (cgrp
== cgrp
->top_cgroup
) {
3661 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
3665 for_each_subsys(cgrp
->root
, ss
) {
3666 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
3669 /* This cgroup is ready now */
3670 for_each_subsys(cgrp
->root
, ss
) {
3671 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3673 * Update id->css pointer and make this css visible from
3674 * CSS ID functions. This pointer will be dereferened
3675 * from RCU-read-side without locks.
3678 rcu_assign_pointer(css
->id
->css
, css
);
3684 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
3685 struct cgroup_subsys
*ss
,
3686 struct cgroup
*cgrp
)
3689 atomic_set(&css
->refcnt
, 1);
3692 if (cgrp
== dummytop
)
3693 set_bit(CSS_ROOT
, &css
->flags
);
3694 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
3695 cgrp
->subsys
[ss
->subsys_id
] = css
;
3698 static void cgroup_lock_hierarchy(struct cgroupfs_root
*root
)
3700 /* We need to take each hierarchy_mutex in a consistent order */
3704 * No worry about a race with rebind_subsystems that might mess up the
3705 * locking order, since both parties are under cgroup_mutex.
3707 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3708 struct cgroup_subsys
*ss
= subsys
[i
];
3711 if (ss
->root
== root
)
3712 mutex_lock(&ss
->hierarchy_mutex
);
3716 static void cgroup_unlock_hierarchy(struct cgroupfs_root
*root
)
3720 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3721 struct cgroup_subsys
*ss
= subsys
[i
];
3724 if (ss
->root
== root
)
3725 mutex_unlock(&ss
->hierarchy_mutex
);
3730 * cgroup_create - create a cgroup
3731 * @parent: cgroup that will be parent of the new cgroup
3732 * @dentry: dentry of the new cgroup
3733 * @mode: mode to set on new inode
3735 * Must be called with the mutex on the parent inode held
3737 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
3740 struct cgroup
*cgrp
;
3741 struct cgroupfs_root
*root
= parent
->root
;
3743 struct cgroup_subsys
*ss
;
3744 struct super_block
*sb
= root
->sb
;
3746 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3750 /* Grab a reference on the superblock so the hierarchy doesn't
3751 * get deleted on unmount if there are child cgroups. This
3752 * can be done outside cgroup_mutex, since the sb can't
3753 * disappear while someone has an open control file on the
3755 atomic_inc(&sb
->s_active
);
3757 mutex_lock(&cgroup_mutex
);
3759 init_cgroup_housekeeping(cgrp
);
3761 cgrp
->parent
= parent
;
3762 cgrp
->root
= parent
->root
;
3763 cgrp
->top_cgroup
= parent
->top_cgroup
;
3765 if (notify_on_release(parent
))
3766 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3768 if (clone_children(parent
))
3769 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3771 for_each_subsys(root
, ss
) {
3772 struct cgroup_subsys_state
*css
= ss
->create(ss
, cgrp
);
3778 init_cgroup_css(css
, ss
, cgrp
);
3780 err
= alloc_css_id(ss
, parent
, cgrp
);
3784 /* At error, ->destroy() callback has to free assigned ID. */
3785 if (clone_children(parent
) && ss
->post_clone
)
3786 ss
->post_clone(ss
, cgrp
);
3789 cgroup_lock_hierarchy(root
);
3790 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
3791 cgroup_unlock_hierarchy(root
);
3792 root
->number_of_cgroups
++;
3794 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
3798 /* The cgroup directory was pre-locked for us */
3799 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
3801 err
= cgroup_populate_dir(cgrp
);
3802 /* If err < 0, we have a half-filled directory - oh well ;) */
3804 mutex_unlock(&cgroup_mutex
);
3805 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
3811 cgroup_lock_hierarchy(root
);
3812 list_del(&cgrp
->sibling
);
3813 cgroup_unlock_hierarchy(root
);
3814 root
->number_of_cgroups
--;
3818 for_each_subsys(root
, ss
) {
3819 if (cgrp
->subsys
[ss
->subsys_id
])
3820 ss
->destroy(ss
, cgrp
);
3823 mutex_unlock(&cgroup_mutex
);
3825 /* Release the reference count that we took on the superblock */
3826 deactivate_super(sb
);
3832 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3834 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
3836 /* the vfs holds inode->i_mutex already */
3837 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
3840 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
3842 /* Check the reference count on each subsystem. Since we
3843 * already established that there are no tasks in the
3844 * cgroup, if the css refcount is also 1, then there should
3845 * be no outstanding references, so the subsystem is safe to
3846 * destroy. We scan across all subsystems rather than using
3847 * the per-hierarchy linked list of mounted subsystems since
3848 * we can be called via check_for_release() with no
3849 * synchronization other than RCU, and the subsystem linked
3850 * list isn't RCU-safe */
3853 * We won't need to lock the subsys array, because the subsystems
3854 * we're concerned about aren't going anywhere since our cgroup root
3855 * has a reference on them.
3857 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3858 struct cgroup_subsys
*ss
= subsys
[i
];
3859 struct cgroup_subsys_state
*css
;
3860 /* Skip subsystems not present or not in this hierarchy */
3861 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
3863 css
= cgrp
->subsys
[ss
->subsys_id
];
3864 /* When called from check_for_release() it's possible
3865 * that by this point the cgroup has been removed
3866 * and the css deleted. But a false-positive doesn't
3867 * matter, since it can only happen if the cgroup
3868 * has been deleted and hence no longer needs the
3869 * release agent to be called anyway. */
3870 if (css
&& (atomic_read(&css
->refcnt
) > 1))
3877 * Atomically mark all (or else none) of the cgroup's CSS objects as
3878 * CSS_REMOVED. Return true on success, or false if the cgroup has
3879 * busy subsystems. Call with cgroup_mutex held
3882 static int cgroup_clear_css_refs(struct cgroup
*cgrp
)
3884 struct cgroup_subsys
*ss
;
3885 unsigned long flags
;
3886 bool failed
= false;
3887 local_irq_save(flags
);
3888 for_each_subsys(cgrp
->root
, ss
) {
3889 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3892 /* We can only remove a CSS with a refcnt==1 */
3893 refcnt
= atomic_read(&css
->refcnt
);
3900 * Drop the refcnt to 0 while we check other
3901 * subsystems. This will cause any racing
3902 * css_tryget() to spin until we set the
3903 * CSS_REMOVED bits or abort
3905 if (atomic_cmpxchg(&css
->refcnt
, refcnt
, 0) == refcnt
)
3911 for_each_subsys(cgrp
->root
, ss
) {
3912 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3915 * Restore old refcnt if we previously managed
3916 * to clear it from 1 to 0
3918 if (!atomic_read(&css
->refcnt
))
3919 atomic_set(&css
->refcnt
, 1);
3921 /* Commit the fact that the CSS is removed */
3922 set_bit(CSS_REMOVED
, &css
->flags
);
3925 local_irq_restore(flags
);
3929 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
3931 struct cgroup
*cgrp
= dentry
->d_fsdata
;
3933 struct cgroup
*parent
;
3935 struct cgroup_event
*event
, *tmp
;
3938 /* the vfs holds both inode->i_mutex already */
3940 mutex_lock(&cgroup_mutex
);
3941 if (atomic_read(&cgrp
->count
) != 0) {
3942 mutex_unlock(&cgroup_mutex
);
3945 if (!list_empty(&cgrp
->children
)) {
3946 mutex_unlock(&cgroup_mutex
);
3949 mutex_unlock(&cgroup_mutex
);
3952 * In general, subsystem has no css->refcnt after pre_destroy(). But
3953 * in racy cases, subsystem may have to get css->refcnt after
3954 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3955 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3956 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3957 * and subsystem's reference count handling. Please see css_get/put
3958 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3960 set_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3963 * Call pre_destroy handlers of subsys. Notify subsystems
3964 * that rmdir() request comes.
3966 ret
= cgroup_call_pre_destroy(cgrp
);
3968 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3972 mutex_lock(&cgroup_mutex
);
3973 parent
= cgrp
->parent
;
3974 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
)) {
3975 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3976 mutex_unlock(&cgroup_mutex
);
3979 prepare_to_wait(&cgroup_rmdir_waitq
, &wait
, TASK_INTERRUPTIBLE
);
3980 if (!cgroup_clear_css_refs(cgrp
)) {
3981 mutex_unlock(&cgroup_mutex
);
3983 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3984 * prepare_to_wait(), we need to check this flag.
3986 if (test_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
))
3988 finish_wait(&cgroup_rmdir_waitq
, &wait
);
3989 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3990 if (signal_pending(current
))
3994 /* NO css_tryget() can success after here. */
3995 finish_wait(&cgroup_rmdir_waitq
, &wait
);
3996 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3998 spin_lock(&release_list_lock
);
3999 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
4000 if (!list_empty(&cgrp
->release_list
))
4001 list_del_init(&cgrp
->release_list
);
4002 spin_unlock(&release_list_lock
);
4004 cgroup_lock_hierarchy(cgrp
->root
);
4005 /* delete this cgroup from parent->children */
4006 list_del_init(&cgrp
->sibling
);
4007 cgroup_unlock_hierarchy(cgrp
->root
);
4009 d
= dget(cgrp
->dentry
);
4011 cgroup_d_remove_dir(d
);
4014 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4015 check_for_release(parent
);
4018 * Unregister events and notify userspace.
4019 * Notify userspace about cgroup removing only after rmdir of cgroup
4020 * directory to avoid race between userspace and kernelspace
4022 spin_lock(&cgrp
->event_list_lock
);
4023 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4024 list_del(&event
->list
);
4025 remove_wait_queue(event
->wqh
, &event
->wait
);
4026 eventfd_signal(event
->eventfd
, 1);
4027 schedule_work(&event
->remove
);
4029 spin_unlock(&cgrp
->event_list_lock
);
4031 mutex_unlock(&cgroup_mutex
);
4035 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4037 struct cgroup_subsys_state
*css
;
4039 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4041 /* Create the top cgroup state for this subsystem */
4042 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4043 ss
->root
= &rootnode
;
4044 css
= ss
->create(ss
, dummytop
);
4045 /* We don't handle early failures gracefully */
4046 BUG_ON(IS_ERR(css
));
4047 init_cgroup_css(css
, ss
, dummytop
);
4049 /* Update the init_css_set to contain a subsys
4050 * pointer to this state - since the subsystem is
4051 * newly registered, all tasks and hence the
4052 * init_css_set is in the subsystem's top cgroup. */
4053 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
4055 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4057 /* At system boot, before all subsystems have been
4058 * registered, no tasks have been forked, so we don't
4059 * need to invoke fork callbacks here. */
4060 BUG_ON(!list_empty(&init_task
.tasks
));
4062 mutex_init(&ss
->hierarchy_mutex
);
4063 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4066 /* this function shouldn't be used with modular subsystems, since they
4067 * need to register a subsys_id, among other things */
4072 * cgroup_load_subsys: load and register a modular subsystem at runtime
4073 * @ss: the subsystem to load
4075 * This function should be called in a modular subsystem's initcall. If the
4076 * subsystem is built as a module, it will be assigned a new subsys_id and set
4077 * up for use. If the subsystem is built-in anyway, work is delegated to the
4078 * simpler cgroup_init_subsys.
4080 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4083 struct cgroup_subsys_state
*css
;
4085 /* check name and function validity */
4086 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4087 ss
->create
== NULL
|| ss
->destroy
== NULL
)
4091 * we don't support callbacks in modular subsystems. this check is
4092 * before the ss->module check for consistency; a subsystem that could
4093 * be a module should still have no callbacks even if the user isn't
4094 * compiling it as one.
4096 if (ss
->fork
|| ss
->exit
)
4100 * an optionally modular subsystem is built-in: we want to do nothing,
4101 * since cgroup_init_subsys will have already taken care of it.
4103 if (ss
->module
== NULL
) {
4104 /* a few sanity checks */
4105 BUG_ON(ss
->subsys_id
>= CGROUP_BUILTIN_SUBSYS_COUNT
);
4106 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4111 * need to register a subsys id before anything else - for example,
4112 * init_cgroup_css needs it.
4114 mutex_lock(&cgroup_mutex
);
4115 /* find the first empty slot in the array */
4116 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4117 if (subsys
[i
] == NULL
)
4120 if (i
== CGROUP_SUBSYS_COUNT
) {
4121 /* maximum number of subsystems already registered! */
4122 mutex_unlock(&cgroup_mutex
);
4125 /* assign ourselves the subsys_id */
4130 * no ss->create seems to need anything important in the ss struct, so
4131 * this can happen first (i.e. before the rootnode attachment).
4133 css
= ss
->create(ss
, dummytop
);
4135 /* failure case - need to deassign the subsys[] slot. */
4137 mutex_unlock(&cgroup_mutex
);
4138 return PTR_ERR(css
);
4141 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4142 ss
->root
= &rootnode
;
4144 /* our new subsystem will be attached to the dummy hierarchy. */
4145 init_cgroup_css(css
, ss
, dummytop
);
4146 /* init_idr must be after init_cgroup_css because it sets css->id. */
4148 int ret
= cgroup_init_idr(ss
, css
);
4150 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4151 ss
->destroy(ss
, dummytop
);
4153 mutex_unlock(&cgroup_mutex
);
4159 * Now we need to entangle the css into the existing css_sets. unlike
4160 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4161 * will need a new pointer to it; done by iterating the css_set_table.
4162 * furthermore, modifying the existing css_sets will corrupt the hash
4163 * table state, so each changed css_set will need its hash recomputed.
4164 * this is all done under the css_set_lock.
4166 write_lock(&css_set_lock
);
4167 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
4169 struct hlist_node
*node
, *tmp
;
4170 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
4172 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
4173 /* skip entries that we already rehashed */
4174 if (cg
->subsys
[ss
->subsys_id
])
4176 /* remove existing entry */
4177 hlist_del(&cg
->hlist
);
4179 cg
->subsys
[ss
->subsys_id
] = css
;
4180 /* recompute hash and restore entry */
4181 new_bucket
= css_set_hash(cg
->subsys
);
4182 hlist_add_head(&cg
->hlist
, new_bucket
);
4185 write_unlock(&css_set_lock
);
4187 mutex_init(&ss
->hierarchy_mutex
);
4188 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4192 mutex_unlock(&cgroup_mutex
);
4195 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4198 * cgroup_unload_subsys: unload a modular subsystem
4199 * @ss: the subsystem to unload
4201 * This function should be called in a modular subsystem's exitcall. When this
4202 * function is invoked, the refcount on the subsystem's module will be 0, so
4203 * the subsystem will not be attached to any hierarchy.
4205 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4207 struct cg_cgroup_link
*link
;
4208 struct hlist_head
*hhead
;
4210 BUG_ON(ss
->module
== NULL
);
4213 * we shouldn't be called if the subsystem is in use, and the use of
4214 * try_module_get in parse_cgroupfs_options should ensure that it
4215 * doesn't start being used while we're killing it off.
4217 BUG_ON(ss
->root
!= &rootnode
);
4219 mutex_lock(&cgroup_mutex
);
4220 /* deassign the subsys_id */
4221 BUG_ON(ss
->subsys_id
< CGROUP_BUILTIN_SUBSYS_COUNT
);
4222 subsys
[ss
->subsys_id
] = NULL
;
4224 /* remove subsystem from rootnode's list of subsystems */
4225 list_del_init(&ss
->sibling
);
4228 * disentangle the css from all css_sets attached to the dummytop. as
4229 * in loading, we need to pay our respects to the hashtable gods.
4231 write_lock(&css_set_lock
);
4232 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4233 struct css_set
*cg
= link
->cg
;
4235 hlist_del(&cg
->hlist
);
4236 BUG_ON(!cg
->subsys
[ss
->subsys_id
]);
4237 cg
->subsys
[ss
->subsys_id
] = NULL
;
4238 hhead
= css_set_hash(cg
->subsys
);
4239 hlist_add_head(&cg
->hlist
, hhead
);
4241 write_unlock(&css_set_lock
);
4244 * remove subsystem's css from the dummytop and free it - need to free
4245 * before marking as null because ss->destroy needs the cgrp->subsys
4246 * pointer to find their state. note that this also takes care of
4247 * freeing the css_id.
4249 ss
->destroy(ss
, dummytop
);
4250 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4252 mutex_unlock(&cgroup_mutex
);
4254 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4257 * cgroup_init_early - cgroup initialization at system boot
4259 * Initialize cgroups at system boot, and initialize any
4260 * subsystems that request early init.
4262 int __init
cgroup_init_early(void)
4265 atomic_set(&init_css_set
.refcount
, 1);
4266 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4267 INIT_LIST_HEAD(&init_css_set
.tasks
);
4268 INIT_HLIST_NODE(&init_css_set
.hlist
);
4270 init_cgroup_root(&rootnode
);
4272 init_task
.cgroups
= &init_css_set
;
4274 init_css_set_link
.cg
= &init_css_set
;
4275 init_css_set_link
.cgrp
= dummytop
;
4276 list_add(&init_css_set_link
.cgrp_link_list
,
4277 &rootnode
.top_cgroup
.css_sets
);
4278 list_add(&init_css_set_link
.cg_link_list
,
4279 &init_css_set
.cg_links
);
4281 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
4282 INIT_HLIST_HEAD(&css_set_table
[i
]);
4284 /* at bootup time, we don't worry about modular subsystems */
4285 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4286 struct cgroup_subsys
*ss
= subsys
[i
];
4289 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4290 BUG_ON(!ss
->create
);
4291 BUG_ON(!ss
->destroy
);
4292 if (ss
->subsys_id
!= i
) {
4293 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4294 ss
->name
, ss
->subsys_id
);
4299 cgroup_init_subsys(ss
);
4305 * cgroup_init - cgroup initialization
4307 * Register cgroup filesystem and /proc file, and initialize
4308 * any subsystems that didn't request early init.
4310 int __init
cgroup_init(void)
4314 struct hlist_head
*hhead
;
4316 err
= bdi_init(&cgroup_backing_dev_info
);
4320 /* at bootup time, we don't worry about modular subsystems */
4321 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4322 struct cgroup_subsys
*ss
= subsys
[i
];
4323 if (!ss
->early_init
)
4324 cgroup_init_subsys(ss
);
4326 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4329 /* Add init_css_set to the hash table */
4330 hhead
= css_set_hash(init_css_set
.subsys
);
4331 hlist_add_head(&init_css_set
.hlist
, hhead
);
4332 BUG_ON(!init_root_id(&rootnode
));
4334 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4340 err
= register_filesystem(&cgroup_fs_type
);
4342 kobject_put(cgroup_kobj
);
4346 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4350 bdi_destroy(&cgroup_backing_dev_info
);
4356 * proc_cgroup_show()
4357 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4358 * - Used for /proc/<pid>/cgroup.
4359 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4360 * doesn't really matter if tsk->cgroup changes after we read it,
4361 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4362 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4363 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4364 * cgroup to top_cgroup.
4367 /* TODO: Use a proper seq_file iterator */
4368 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4371 struct task_struct
*tsk
;
4374 struct cgroupfs_root
*root
;
4377 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4383 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4389 mutex_lock(&cgroup_mutex
);
4391 for_each_active_root(root
) {
4392 struct cgroup_subsys
*ss
;
4393 struct cgroup
*cgrp
;
4396 seq_printf(m
, "%d:", root
->hierarchy_id
);
4397 for_each_subsys(root
, ss
)
4398 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4399 if (strlen(root
->name
))
4400 seq_printf(m
, "%sname=%s", count
? "," : "",
4403 cgrp
= task_cgroup_from_root(tsk
, root
);
4404 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4412 mutex_unlock(&cgroup_mutex
);
4413 put_task_struct(tsk
);
4420 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4422 struct pid
*pid
= PROC_I(inode
)->pid
;
4423 return single_open(file
, proc_cgroup_show
, pid
);
4426 const struct file_operations proc_cgroup_operations
= {
4427 .open
= cgroup_open
,
4429 .llseek
= seq_lseek
,
4430 .release
= single_release
,
4433 /* Display information about each subsystem and each hierarchy */
4434 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4438 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4440 * ideally we don't want subsystems moving around while we do this.
4441 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4442 * subsys/hierarchy state.
4444 mutex_lock(&cgroup_mutex
);
4445 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4446 struct cgroup_subsys
*ss
= subsys
[i
];
4449 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4450 ss
->name
, ss
->root
->hierarchy_id
,
4451 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4453 mutex_unlock(&cgroup_mutex
);
4457 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4459 return single_open(file
, proc_cgroupstats_show
, NULL
);
4462 static const struct file_operations proc_cgroupstats_operations
= {
4463 .open
= cgroupstats_open
,
4465 .llseek
= seq_lseek
,
4466 .release
= single_release
,
4470 * cgroup_fork - attach newly forked task to its parents cgroup.
4471 * @child: pointer to task_struct of forking parent process.
4473 * Description: A task inherits its parent's cgroup at fork().
4475 * A pointer to the shared css_set was automatically copied in
4476 * fork.c by dup_task_struct(). However, we ignore that copy, since
4477 * it was not made under the protection of RCU or cgroup_mutex, so
4478 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4479 * have already changed current->cgroups, allowing the previously
4480 * referenced cgroup group to be removed and freed.
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 child
->cgroups
= current
->cgroups
;
4489 get_css_set(child
->cgroups
);
4490 task_unlock(current
);
4491 INIT_LIST_HEAD(&child
->cg_list
);
4495 * cgroup_fork_callbacks - run fork callbacks
4496 * @child: the new task
4498 * Called on a new task very soon before adding it to the
4499 * tasklist. No need to take any locks since no-one can
4500 * be operating on this task.
4502 void cgroup_fork_callbacks(struct task_struct
*child
)
4504 if (need_forkexit_callback
) {
4507 * forkexit callbacks are only supported for builtin
4508 * subsystems, and the builtin section of the subsys array is
4509 * immutable, so we don't need to lock the subsys array here.
4511 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4512 struct cgroup_subsys
*ss
= subsys
[i
];
4514 ss
->fork(ss
, child
);
4520 * cgroup_post_fork - called on a new task after adding it to the task list
4521 * @child: the task in question
4523 * Adds the task to the list running through its css_set if necessary.
4524 * Has to be after the task is visible on the task list in case we race
4525 * with the first call to cgroup_iter_start() - to guarantee that the
4526 * new task ends up on its list.
4528 void cgroup_post_fork(struct task_struct
*child
)
4530 if (use_task_css_set_links
) {
4531 write_lock(&css_set_lock
);
4533 if (list_empty(&child
->cg_list
))
4534 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4536 write_unlock(&css_set_lock
);
4540 * cgroup_exit - detach cgroup from exiting task
4541 * @tsk: pointer to task_struct of exiting process
4542 * @run_callback: run exit callbacks?
4544 * Description: Detach cgroup from @tsk and release it.
4546 * Note that cgroups marked notify_on_release force every task in
4547 * them to take the global cgroup_mutex mutex when exiting.
4548 * This could impact scaling on very large systems. Be reluctant to
4549 * use notify_on_release cgroups where very high task exit scaling
4550 * is required on large systems.
4552 * the_top_cgroup_hack:
4554 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4556 * We call cgroup_exit() while the task is still competent to
4557 * handle notify_on_release(), then leave the task attached to the
4558 * root cgroup in each hierarchy for the remainder of its exit.
4560 * To do this properly, we would increment the reference count on
4561 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4562 * code we would add a second cgroup function call, to drop that
4563 * reference. This would just create an unnecessary hot spot on
4564 * the top_cgroup reference count, to no avail.
4566 * Normally, holding a reference to a cgroup without bumping its
4567 * count is unsafe. The cgroup could go away, or someone could
4568 * attach us to a different cgroup, decrementing the count on
4569 * the first cgroup that we never incremented. But in this case,
4570 * top_cgroup isn't going away, and either task has PF_EXITING set,
4571 * which wards off any cgroup_attach_task() attempts, or task is a failed
4572 * fork, never visible to cgroup_attach_task.
4574 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4580 * Unlink from the css_set task list if necessary.
4581 * Optimistically check cg_list before taking
4584 if (!list_empty(&tsk
->cg_list
)) {
4585 write_lock(&css_set_lock
);
4586 if (!list_empty(&tsk
->cg_list
))
4587 list_del_init(&tsk
->cg_list
);
4588 write_unlock(&css_set_lock
);
4591 /* Reassign the task to the init_css_set. */
4594 tsk
->cgroups
= &init_css_set
;
4596 if (run_callbacks
&& need_forkexit_callback
) {
4598 * modular subsystems can't use callbacks, so no need to lock
4601 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4602 struct cgroup_subsys
*ss
= subsys
[i
];
4604 struct cgroup
*old_cgrp
=
4605 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4606 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4607 ss
->exit(ss
, cgrp
, old_cgrp
, tsk
);
4614 put_css_set_taskexit(cg
);
4618 * cgroup_clone - clone the cgroup the given subsystem is attached to
4619 * @tsk: the task to be moved
4620 * @subsys: the given subsystem
4621 * @nodename: the name for the new cgroup
4623 * Duplicate the current cgroup in the hierarchy that the given
4624 * subsystem is attached to, and move this task into the new
4627 int cgroup_clone(struct task_struct
*tsk
, struct cgroup_subsys
*subsys
,
4630 struct dentry
*dentry
;
4632 struct cgroup
*parent
, *child
;
4633 struct inode
*inode
;
4635 struct cgroupfs_root
*root
;
4636 struct cgroup_subsys
*ss
;
4638 /* We shouldn't be called by an unregistered subsystem */
4639 BUG_ON(!subsys
->active
);
4641 /* First figure out what hierarchy and cgroup we're dealing
4642 * with, and pin them so we can drop cgroup_mutex */
4643 mutex_lock(&cgroup_mutex
);
4645 root
= subsys
->root
;
4646 if (root
== &rootnode
) {
4647 mutex_unlock(&cgroup_mutex
);
4651 /* Pin the hierarchy */
4652 if (!atomic_inc_not_zero(&root
->sb
->s_active
)) {
4653 /* We race with the final deactivate_super() */
4654 mutex_unlock(&cgroup_mutex
);
4658 /* Keep the cgroup alive */
4660 parent
= task_cgroup(tsk
, subsys
->subsys_id
);
4665 mutex_unlock(&cgroup_mutex
);
4667 /* Now do the VFS work to create a cgroup */
4668 inode
= parent
->dentry
->d_inode
;
4670 /* Hold the parent directory mutex across this operation to
4671 * stop anyone else deleting the new cgroup */
4672 mutex_lock(&inode
->i_mutex
);
4673 dentry
= lookup_one_len(nodename
, parent
->dentry
, strlen(nodename
));
4674 if (IS_ERR(dentry
)) {
4676 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename
,
4678 ret
= PTR_ERR(dentry
);
4682 /* Create the cgroup directory, which also creates the cgroup */
4683 ret
= vfs_mkdir(inode
, dentry
, 0755);
4684 child
= __d_cgrp(dentry
);
4688 "Failed to create cgroup %s: %d\n", nodename
,
4693 /* The cgroup now exists. Retake cgroup_mutex and check
4694 * that we're still in the same state that we thought we
4696 mutex_lock(&cgroup_mutex
);
4697 if ((root
!= subsys
->root
) ||
4698 (parent
!= task_cgroup(tsk
, subsys
->subsys_id
))) {
4699 /* Aargh, we raced ... */
4700 mutex_unlock(&inode
->i_mutex
);
4703 deactivate_super(root
->sb
);
4704 /* The cgroup is still accessible in the VFS, but
4705 * we're not going to try to rmdir() it at this
4708 "Race in cgroup_clone() - leaking cgroup %s\n",
4713 /* do any required auto-setup */
4714 for_each_subsys(root
, ss
) {
4716 ss
->post_clone(ss
, child
);
4719 /* All seems fine. Finish by moving the task into the new cgroup */
4720 ret
= cgroup_attach_task(child
, tsk
);
4721 mutex_unlock(&cgroup_mutex
);
4724 mutex_unlock(&inode
->i_mutex
);
4726 mutex_lock(&cgroup_mutex
);
4728 mutex_unlock(&cgroup_mutex
);
4729 deactivate_super(root
->sb
);
4734 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4735 * @cgrp: the cgroup in question
4736 * @task: the task in question
4738 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4741 * If we are sending in dummytop, then presumably we are creating
4742 * the top cgroup in the subsystem.
4744 * Called only by the ns (nsproxy) cgroup.
4746 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
4749 struct cgroup
*target
;
4751 if (cgrp
== dummytop
)
4754 target
= task_cgroup_from_root(task
, cgrp
->root
);
4755 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
4756 cgrp
= cgrp
->parent
;
4757 ret
= (cgrp
== target
);
4761 static void check_for_release(struct cgroup
*cgrp
)
4763 /* All of these checks rely on RCU to keep the cgroup
4764 * structure alive */
4765 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
4766 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
4767 /* Control Group is currently removeable. If it's not
4768 * already queued for a userspace notification, queue
4770 int need_schedule_work
= 0;
4771 spin_lock(&release_list_lock
);
4772 if (!cgroup_is_removed(cgrp
) &&
4773 list_empty(&cgrp
->release_list
)) {
4774 list_add(&cgrp
->release_list
, &release_list
);
4775 need_schedule_work
= 1;
4777 spin_unlock(&release_list_lock
);
4778 if (need_schedule_work
)
4779 schedule_work(&release_agent_work
);
4783 /* Caller must verify that the css is not for root cgroup */
4784 void __css_put(struct cgroup_subsys_state
*css
, int count
)
4786 struct cgroup
*cgrp
= css
->cgroup
;
4789 val
= atomic_sub_return(count
, &css
->refcnt
);
4791 if (notify_on_release(cgrp
)) {
4792 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4793 check_for_release(cgrp
);
4795 cgroup_wakeup_rmdir_waiter(cgrp
);
4798 WARN_ON_ONCE(val
< 1);
4800 EXPORT_SYMBOL_GPL(__css_put
);
4803 * Notify userspace when a cgroup is released, by running the
4804 * configured release agent with the name of the cgroup (path
4805 * relative to the root of cgroup file system) as the argument.
4807 * Most likely, this user command will try to rmdir this cgroup.
4809 * This races with the possibility that some other task will be
4810 * attached to this cgroup before it is removed, or that some other
4811 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4812 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4813 * unused, and this cgroup will be reprieved from its death sentence,
4814 * to continue to serve a useful existence. Next time it's released,
4815 * we will get notified again, if it still has 'notify_on_release' set.
4817 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4818 * means only wait until the task is successfully execve()'d. The
4819 * separate release agent task is forked by call_usermodehelper(),
4820 * then control in this thread returns here, without waiting for the
4821 * release agent task. We don't bother to wait because the caller of
4822 * this routine has no use for the exit status of the release agent
4823 * task, so no sense holding our caller up for that.
4825 static void cgroup_release_agent(struct work_struct
*work
)
4827 BUG_ON(work
!= &release_agent_work
);
4828 mutex_lock(&cgroup_mutex
);
4829 spin_lock(&release_list_lock
);
4830 while (!list_empty(&release_list
)) {
4831 char *argv
[3], *envp
[3];
4833 char *pathbuf
= NULL
, *agentbuf
= NULL
;
4834 struct cgroup
*cgrp
= list_entry(release_list
.next
,
4837 list_del_init(&cgrp
->release_list
);
4838 spin_unlock(&release_list_lock
);
4839 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4842 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
4844 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
4849 argv
[i
++] = agentbuf
;
4850 argv
[i
++] = pathbuf
;
4854 /* minimal command environment */
4855 envp
[i
++] = "HOME=/";
4856 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4859 /* Drop the lock while we invoke the usermode helper,
4860 * since the exec could involve hitting disk and hence
4861 * be a slow process */
4862 mutex_unlock(&cgroup_mutex
);
4863 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
4864 mutex_lock(&cgroup_mutex
);
4868 spin_lock(&release_list_lock
);
4870 spin_unlock(&release_list_lock
);
4871 mutex_unlock(&cgroup_mutex
);
4874 static int __init
cgroup_disable(char *str
)
4879 while ((token
= strsep(&str
, ",")) != NULL
) {
4883 * cgroup_disable, being at boot time, can't know about module
4884 * subsystems, so we don't worry about them.
4886 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4887 struct cgroup_subsys
*ss
= subsys
[i
];
4889 if (!strcmp(token
, ss
->name
)) {
4891 printk(KERN_INFO
"Disabling %s control group"
4892 " subsystem\n", ss
->name
);
4899 __setup("cgroup_disable=", cgroup_disable
);
4902 * Functons for CSS ID.
4906 *To get ID other than 0, this should be called when !cgroup_is_removed().
4908 unsigned short css_id(struct cgroup_subsys_state
*css
)
4910 struct css_id
*cssid
;
4913 * This css_id() can return correct value when somone has refcnt
4914 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4915 * it's unchanged until freed.
4917 cssid
= rcu_dereference_check(css
->id
,
4918 rcu_read_lock_held() || atomic_read(&css
->refcnt
));
4924 EXPORT_SYMBOL_GPL(css_id
);
4926 unsigned short css_depth(struct cgroup_subsys_state
*css
)
4928 struct css_id
*cssid
;
4930 cssid
= rcu_dereference_check(css
->id
,
4931 rcu_read_lock_held() || atomic_read(&css
->refcnt
));
4934 return cssid
->depth
;
4937 EXPORT_SYMBOL_GPL(css_depth
);
4940 * css_is_ancestor - test "root" css is an ancestor of "child"
4941 * @child: the css to be tested.
4942 * @root: the css supporsed to be an ancestor of the child.
4944 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4945 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4946 * But, considering usual usage, the csses should be valid objects after test.
4947 * Assuming that the caller will do some action to the child if this returns
4948 * returns true, the caller must take "child";s reference count.
4949 * If "child" is valid object and this returns true, "root" is valid, too.
4952 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
4953 const struct cgroup_subsys_state
*root
)
4955 struct css_id
*child_id
;
4956 struct css_id
*root_id
;
4960 child_id
= rcu_dereference(child
->id
);
4961 root_id
= rcu_dereference(root
->id
);
4964 || (child_id
->depth
< root_id
->depth
)
4965 || (child_id
->stack
[root_id
->depth
] != root_id
->id
))
4971 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
4973 struct css_id
*id
= css
->id
;
4974 /* When this is called before css_id initialization, id can be NULL */
4978 BUG_ON(!ss
->use_id
);
4980 rcu_assign_pointer(id
->css
, NULL
);
4981 rcu_assign_pointer(css
->id
, NULL
);
4982 spin_lock(&ss
->id_lock
);
4983 idr_remove(&ss
->idr
, id
->id
);
4984 spin_unlock(&ss
->id_lock
);
4985 kfree_rcu(id
, rcu_head
);
4987 EXPORT_SYMBOL_GPL(free_css_id
);
4990 * This is called by init or create(). Then, calls to this function are
4991 * always serialized (By cgroup_mutex() at create()).
4994 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
4996 struct css_id
*newid
;
4997 int myid
, error
, size
;
4999 BUG_ON(!ss
->use_id
);
5001 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5002 newid
= kzalloc(size
, GFP_KERNEL
);
5004 return ERR_PTR(-ENOMEM
);
5006 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
5010 spin_lock(&ss
->id_lock
);
5011 /* Don't use 0. allocates an ID of 1-65535 */
5012 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
5013 spin_unlock(&ss
->id_lock
);
5015 /* Returns error when there are no free spaces for new ID.*/
5020 if (myid
> CSS_ID_MAX
)
5024 newid
->depth
= depth
;
5028 spin_lock(&ss
->id_lock
);
5029 idr_remove(&ss
->idr
, myid
);
5030 spin_unlock(&ss
->id_lock
);
5033 return ERR_PTR(error
);
5037 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5038 struct cgroup_subsys_state
*rootcss
)
5040 struct css_id
*newid
;
5042 spin_lock_init(&ss
->id_lock
);
5045 newid
= get_new_cssid(ss
, 0);
5047 return PTR_ERR(newid
);
5049 newid
->stack
[0] = newid
->id
;
5050 newid
->css
= rootcss
;
5051 rootcss
->id
= newid
;
5055 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5056 struct cgroup
*child
)
5058 int subsys_id
, i
, depth
= 0;
5059 struct cgroup_subsys_state
*parent_css
, *child_css
;
5060 struct css_id
*child_id
, *parent_id
;
5062 subsys_id
= ss
->subsys_id
;
5063 parent_css
= parent
->subsys
[subsys_id
];
5064 child_css
= child
->subsys
[subsys_id
];
5065 parent_id
= parent_css
->id
;
5066 depth
= parent_id
->depth
+ 1;
5068 child_id
= get_new_cssid(ss
, depth
);
5069 if (IS_ERR(child_id
))
5070 return PTR_ERR(child_id
);
5072 for (i
= 0; i
< depth
; i
++)
5073 child_id
->stack
[i
] = parent_id
->stack
[i
];
5074 child_id
->stack
[depth
] = child_id
->id
;
5076 * child_id->css pointer will be set after this cgroup is available
5077 * see cgroup_populate_dir()
5079 rcu_assign_pointer(child_css
->id
, child_id
);
5085 * css_lookup - lookup css by id
5086 * @ss: cgroup subsys to be looked into.
5089 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5090 * NULL if not. Should be called under rcu_read_lock()
5092 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5094 struct css_id
*cssid
= NULL
;
5096 BUG_ON(!ss
->use_id
);
5097 cssid
= idr_find(&ss
->idr
, id
);
5099 if (unlikely(!cssid
))
5102 return rcu_dereference(cssid
->css
);
5104 EXPORT_SYMBOL_GPL(css_lookup
);
5107 * css_get_next - lookup next cgroup under specified hierarchy.
5108 * @ss: pointer to subsystem
5109 * @id: current position of iteration.
5110 * @root: pointer to css. search tree under this.
5111 * @foundid: position of found object.
5113 * Search next css under the specified hierarchy of rootid. Calling under
5114 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5116 struct cgroup_subsys_state
*
5117 css_get_next(struct cgroup_subsys
*ss
, int id
,
5118 struct cgroup_subsys_state
*root
, int *foundid
)
5120 struct cgroup_subsys_state
*ret
= NULL
;
5123 int rootid
= css_id(root
);
5124 int depth
= css_depth(root
);
5129 BUG_ON(!ss
->use_id
);
5130 /* fill start point for scan */
5134 * scan next entry from bitmap(tree), tmpid is updated after
5137 spin_lock(&ss
->id_lock
);
5138 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
5139 spin_unlock(&ss
->id_lock
);
5143 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
5144 ret
= rcu_dereference(tmp
->css
);
5150 /* continue to scan from next id */
5157 * get corresponding css from file open on cgroupfs directory
5159 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5161 struct cgroup
*cgrp
;
5162 struct inode
*inode
;
5163 struct cgroup_subsys_state
*css
;
5165 inode
= f
->f_dentry
->d_inode
;
5166 /* check in cgroup filesystem dir */
5167 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5168 return ERR_PTR(-EBADF
);
5170 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5171 return ERR_PTR(-EINVAL
);
5174 cgrp
= __d_cgrp(f
->f_dentry
);
5175 css
= cgrp
->subsys
[id
];
5176 return css
? css
: ERR_PTR(-ENOENT
);
5179 #ifdef CONFIG_CGROUP_DEBUG
5180 static struct cgroup_subsys_state
*debug_create(struct cgroup_subsys
*ss
,
5181 struct cgroup
*cont
)
5183 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5186 return ERR_PTR(-ENOMEM
);
5191 static void debug_destroy(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5193 kfree(cont
->subsys
[debug_subsys_id
]);
5196 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5198 return atomic_read(&cont
->count
);
5201 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5203 return cgroup_task_count(cont
);
5206 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5208 return (u64
)(unsigned long)current
->cgroups
;
5211 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5217 count
= atomic_read(¤t
->cgroups
->refcount
);
5222 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5224 struct seq_file
*seq
)
5226 struct cg_cgroup_link
*link
;
5229 read_lock(&css_set_lock
);
5231 cg
= rcu_dereference(current
->cgroups
);
5232 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5233 struct cgroup
*c
= link
->cgrp
;
5237 name
= c
->dentry
->d_name
.name
;
5240 seq_printf(seq
, "Root %d group %s\n",
5241 c
->root
->hierarchy_id
, name
);
5244 read_unlock(&css_set_lock
);
5248 #define MAX_TASKS_SHOWN_PER_CSS 25
5249 static int cgroup_css_links_read(struct cgroup
*cont
,
5251 struct seq_file
*seq
)
5253 struct cg_cgroup_link
*link
;
5255 read_lock(&css_set_lock
);
5256 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5257 struct css_set
*cg
= link
->cg
;
5258 struct task_struct
*task
;
5260 seq_printf(seq
, "css_set %p\n", cg
);
5261 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5262 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5263 seq_puts(seq
, " ...\n");
5266 seq_printf(seq
, " task %d\n",
5267 task_pid_vnr(task
));
5271 read_unlock(&css_set_lock
);
5275 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5277 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5280 static struct cftype debug_files
[] = {
5282 .name
= "cgroup_refcount",
5283 .read_u64
= cgroup_refcount_read
,
5286 .name
= "taskcount",
5287 .read_u64
= debug_taskcount_read
,
5291 .name
= "current_css_set",
5292 .read_u64
= current_css_set_read
,
5296 .name
= "current_css_set_refcount",
5297 .read_u64
= current_css_set_refcount_read
,
5301 .name
= "current_css_set_cg_links",
5302 .read_seq_string
= current_css_set_cg_links_read
,
5306 .name
= "cgroup_css_links",
5307 .read_seq_string
= cgroup_css_links_read
,
5311 .name
= "releasable",
5312 .read_u64
= releasable_read
,
5316 static int debug_populate(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5318 return cgroup_add_files(cont
, ss
, debug_files
,
5319 ARRAY_SIZE(debug_files
));
5322 struct cgroup_subsys debug_subsys
= {
5324 .create
= debug_create
,
5325 .destroy
= debug_destroy
,
5326 .populate
= debug_populate
,
5327 .subsys_id
= debug_subsys_id
,
5329 #endif /* CONFIG_CGROUP_DEBUG */