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 free_cgroup_rcu(struct rcu_head
*obj
)
811 struct cgroup
*cgrp
= container_of(obj
, struct cgroup
, rcu_head
);
816 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
818 /* is dentry a directory ? if so, kfree() associated cgroup */
819 if (S_ISDIR(inode
->i_mode
)) {
820 struct cgroup
*cgrp
= dentry
->d_fsdata
;
821 struct cgroup_subsys
*ss
;
822 BUG_ON(!(cgroup_is_removed(cgrp
)));
823 /* It's possible for external users to be holding css
824 * reference counts on a cgroup; css_put() needs to
825 * be able to access the cgroup after decrementing
826 * the reference count in order to know if it needs to
827 * queue the cgroup to be handled by the release
831 mutex_lock(&cgroup_mutex
);
833 * Release the subsystem state objects.
835 for_each_subsys(cgrp
->root
, ss
)
836 ss
->destroy(ss
, cgrp
);
838 cgrp
->root
->number_of_cgroups
--;
839 mutex_unlock(&cgroup_mutex
);
842 * Drop the active superblock reference that we took when we
845 deactivate_super(cgrp
->root
->sb
);
848 * if we're getting rid of the cgroup, refcount should ensure
849 * that there are no pidlists left.
851 BUG_ON(!list_empty(&cgrp
->pidlists
));
853 call_rcu(&cgrp
->rcu_head
, free_cgroup_rcu
);
858 static int cgroup_delete(const struct dentry
*d
)
863 static void remove_dir(struct dentry
*d
)
865 struct dentry
*parent
= dget(d
->d_parent
);
868 simple_rmdir(parent
->d_inode
, d
);
872 static void cgroup_clear_directory(struct dentry
*dentry
)
874 struct list_head
*node
;
876 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
877 spin_lock(&dentry
->d_lock
);
878 node
= dentry
->d_subdirs
.next
;
879 while (node
!= &dentry
->d_subdirs
) {
880 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
882 spin_lock_nested(&d
->d_lock
, DENTRY_D_LOCK_NESTED
);
885 /* This should never be called on a cgroup
886 * directory with child cgroups */
887 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
889 spin_unlock(&d
->d_lock
);
890 spin_unlock(&dentry
->d_lock
);
892 simple_unlink(dentry
->d_inode
, d
);
894 spin_lock(&dentry
->d_lock
);
896 spin_unlock(&d
->d_lock
);
897 node
= dentry
->d_subdirs
.next
;
899 spin_unlock(&dentry
->d_lock
);
903 * NOTE : the dentry must have been dget()'ed
905 static void cgroup_d_remove_dir(struct dentry
*dentry
)
907 struct dentry
*parent
;
909 cgroup_clear_directory(dentry
);
911 parent
= dentry
->d_parent
;
912 spin_lock(&parent
->d_lock
);
913 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
914 list_del_init(&dentry
->d_u
.d_child
);
915 spin_unlock(&dentry
->d_lock
);
916 spin_unlock(&parent
->d_lock
);
921 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
922 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
923 * reference to css->refcnt. In general, this refcnt is expected to goes down
926 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
928 DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq
);
930 static void cgroup_wakeup_rmdir_waiter(struct cgroup
*cgrp
)
932 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
)))
933 wake_up_all(&cgroup_rmdir_waitq
);
936 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
)
941 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
)
943 cgroup_wakeup_rmdir_waiter(css
->cgroup
);
948 * Call with cgroup_mutex held. Drops reference counts on modules, including
949 * any duplicate ones that parse_cgroupfs_options took. If this function
950 * returns an error, no reference counts are touched.
952 static int rebind_subsystems(struct cgroupfs_root
*root
,
953 unsigned long final_bits
)
955 unsigned long added_bits
, removed_bits
;
956 struct cgroup
*cgrp
= &root
->top_cgroup
;
959 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
961 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
962 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
963 /* Check that any added subsystems are currently free */
964 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
965 unsigned long bit
= 1UL << i
;
966 struct cgroup_subsys
*ss
= subsys
[i
];
967 if (!(bit
& added_bits
))
970 * Nobody should tell us to do a subsys that doesn't exist:
971 * parse_cgroupfs_options should catch that case and refcounts
972 * ensure that subsystems won't disappear once selected.
975 if (ss
->root
!= &rootnode
) {
976 /* Subsystem isn't free */
981 /* Currently we don't handle adding/removing subsystems when
982 * any child cgroups exist. This is theoretically supportable
983 * but involves complex error handling, so it's being left until
985 if (root
->number_of_cgroups
> 1)
988 /* Process each subsystem */
989 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
990 struct cgroup_subsys
*ss
= subsys
[i
];
991 unsigned long bit
= 1UL << i
;
992 if (bit
& added_bits
) {
993 /* We're binding this subsystem to this hierarchy */
995 BUG_ON(cgrp
->subsys
[i
]);
996 BUG_ON(!dummytop
->subsys
[i
]);
997 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
998 mutex_lock(&ss
->hierarchy_mutex
);
999 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
1000 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1001 list_move(&ss
->sibling
, &root
->subsys_list
);
1005 mutex_unlock(&ss
->hierarchy_mutex
);
1006 /* refcount was already taken, and we're keeping it */
1007 } else if (bit
& removed_bits
) {
1008 /* We're removing this subsystem */
1010 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1011 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1012 mutex_lock(&ss
->hierarchy_mutex
);
1014 ss
->bind(ss
, dummytop
);
1015 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1016 cgrp
->subsys
[i
] = NULL
;
1017 subsys
[i
]->root
= &rootnode
;
1018 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1019 mutex_unlock(&ss
->hierarchy_mutex
);
1020 /* subsystem is now free - drop reference on module */
1021 module_put(ss
->module
);
1022 } else if (bit
& final_bits
) {
1023 /* Subsystem state should already exist */
1025 BUG_ON(!cgrp
->subsys
[i
]);
1027 * a refcount was taken, but we already had one, so
1028 * drop the extra reference.
1030 module_put(ss
->module
);
1031 #ifdef CONFIG_MODULE_UNLOAD
1032 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1035 /* Subsystem state shouldn't exist */
1036 BUG_ON(cgrp
->subsys
[i
]);
1039 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
1045 static int cgroup_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
1047 struct cgroupfs_root
*root
= vfs
->mnt_sb
->s_fs_info
;
1048 struct cgroup_subsys
*ss
;
1050 mutex_lock(&cgroup_mutex
);
1051 for_each_subsys(root
, ss
)
1052 seq_printf(seq
, ",%s", ss
->name
);
1053 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1054 seq_puts(seq
, ",noprefix");
1055 if (strlen(root
->release_agent_path
))
1056 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1057 if (clone_children(&root
->top_cgroup
))
1058 seq_puts(seq
, ",clone_children");
1059 if (strlen(root
->name
))
1060 seq_printf(seq
, ",name=%s", root
->name
);
1061 mutex_unlock(&cgroup_mutex
);
1065 struct cgroup_sb_opts
{
1066 unsigned long subsys_bits
;
1067 unsigned long flags
;
1068 char *release_agent
;
1069 bool clone_children
;
1071 /* User explicitly requested empty subsystem */
1074 struct cgroupfs_root
*new_root
;
1079 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1080 * with cgroup_mutex held to protect the subsys[] array. This function takes
1081 * refcounts on subsystems to be used, unless it returns error, in which case
1082 * no refcounts are taken.
1084 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1086 char *token
, *o
= data
;
1087 bool all_ss
= false, one_ss
= false;
1088 unsigned long mask
= (unsigned long)-1;
1090 bool module_pin_failed
= false;
1092 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1094 #ifdef CONFIG_CPUSETS
1095 mask
= ~(1UL << cpuset_subsys_id
);
1098 memset(opts
, 0, sizeof(*opts
));
1100 while ((token
= strsep(&o
, ",")) != NULL
) {
1103 if (!strcmp(token
, "none")) {
1104 /* Explicitly have no subsystems */
1108 if (!strcmp(token
, "all")) {
1109 /* Mutually exclusive option 'all' + subsystem name */
1115 if (!strcmp(token
, "noprefix")) {
1116 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1119 if (!strcmp(token
, "clone_children")) {
1120 opts
->clone_children
= true;
1123 if (!strncmp(token
, "release_agent=", 14)) {
1124 /* Specifying two release agents is forbidden */
1125 if (opts
->release_agent
)
1127 opts
->release_agent
=
1128 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1129 if (!opts
->release_agent
)
1133 if (!strncmp(token
, "name=", 5)) {
1134 const char *name
= token
+ 5;
1135 /* Can't specify an empty name */
1138 /* Must match [\w.-]+ */
1139 for (i
= 0; i
< strlen(name
); i
++) {
1143 if ((c
== '.') || (c
== '-') || (c
== '_'))
1147 /* Specifying two names is forbidden */
1150 opts
->name
= kstrndup(name
,
1151 MAX_CGROUP_ROOT_NAMELEN
- 1,
1159 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1160 struct cgroup_subsys
*ss
= subsys
[i
];
1163 if (strcmp(token
, ss
->name
))
1168 /* Mutually exclusive option 'all' + subsystem name */
1171 set_bit(i
, &opts
->subsys_bits
);
1176 if (i
== CGROUP_SUBSYS_COUNT
)
1181 * If the 'all' option was specified select all the subsystems,
1182 * otherwise 'all, 'none' and a subsystem name options were not
1183 * specified, let's default to 'all'
1185 if (all_ss
|| (!all_ss
&& !one_ss
&& !opts
->none
)) {
1186 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1187 struct cgroup_subsys
*ss
= subsys
[i
];
1192 set_bit(i
, &opts
->subsys_bits
);
1196 /* Consistency checks */
1199 * Option noprefix was introduced just for backward compatibility
1200 * with the old cpuset, so we allow noprefix only if mounting just
1201 * the cpuset subsystem.
1203 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1204 (opts
->subsys_bits
& mask
))
1208 /* Can't specify "none" and some subsystems */
1209 if (opts
->subsys_bits
&& opts
->none
)
1213 * We either have to specify by name or by subsystems. (So all
1214 * empty hierarchies must have a name).
1216 if (!opts
->subsys_bits
&& !opts
->name
)
1220 * Grab references on all the modules we'll need, so the subsystems
1221 * don't dance around before rebind_subsystems attaches them. This may
1222 * take duplicate reference counts on a subsystem that's already used,
1223 * but rebind_subsystems handles this case.
1225 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1226 unsigned long bit
= 1UL << i
;
1228 if (!(bit
& opts
->subsys_bits
))
1230 if (!try_module_get(subsys
[i
]->module
)) {
1231 module_pin_failed
= true;
1235 if (module_pin_failed
) {
1237 * oops, one of the modules was going away. this means that we
1238 * raced with a module_delete call, and to the user this is
1239 * essentially a "subsystem doesn't exist" case.
1241 for (i
--; i
>= CGROUP_BUILTIN_SUBSYS_COUNT
; i
--) {
1242 /* drop refcounts only on the ones we took */
1243 unsigned long bit
= 1UL << i
;
1245 if (!(bit
& opts
->subsys_bits
))
1247 module_put(subsys
[i
]->module
);
1255 static void drop_parsed_module_refcounts(unsigned long subsys_bits
)
1258 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1259 unsigned long bit
= 1UL << i
;
1261 if (!(bit
& subsys_bits
))
1263 module_put(subsys
[i
]->module
);
1267 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1270 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1271 struct cgroup
*cgrp
= &root
->top_cgroup
;
1272 struct cgroup_sb_opts opts
;
1274 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1275 mutex_lock(&cgroup_mutex
);
1277 /* See what subsystems are wanted */
1278 ret
= parse_cgroupfs_options(data
, &opts
);
1282 /* Don't allow flags or name to change at remount */
1283 if (opts
.flags
!= root
->flags
||
1284 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1286 drop_parsed_module_refcounts(opts
.subsys_bits
);
1290 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
1292 drop_parsed_module_refcounts(opts
.subsys_bits
);
1296 /* (re)populate subsystem files */
1297 cgroup_populate_dir(cgrp
);
1299 if (opts
.release_agent
)
1300 strcpy(root
->release_agent_path
, opts
.release_agent
);
1302 kfree(opts
.release_agent
);
1304 mutex_unlock(&cgroup_mutex
);
1305 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1309 static const struct super_operations cgroup_ops
= {
1310 .statfs
= simple_statfs
,
1311 .drop_inode
= generic_delete_inode
,
1312 .show_options
= cgroup_show_options
,
1313 .remount_fs
= cgroup_remount
,
1316 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1318 INIT_LIST_HEAD(&cgrp
->sibling
);
1319 INIT_LIST_HEAD(&cgrp
->children
);
1320 INIT_LIST_HEAD(&cgrp
->css_sets
);
1321 INIT_LIST_HEAD(&cgrp
->release_list
);
1322 INIT_LIST_HEAD(&cgrp
->pidlists
);
1323 mutex_init(&cgrp
->pidlist_mutex
);
1324 INIT_LIST_HEAD(&cgrp
->event_list
);
1325 spin_lock_init(&cgrp
->event_list_lock
);
1328 static void init_cgroup_root(struct cgroupfs_root
*root
)
1330 struct cgroup
*cgrp
= &root
->top_cgroup
;
1331 INIT_LIST_HEAD(&root
->subsys_list
);
1332 INIT_LIST_HEAD(&root
->root_list
);
1333 root
->number_of_cgroups
= 1;
1335 cgrp
->top_cgroup
= cgrp
;
1336 init_cgroup_housekeeping(cgrp
);
1339 static bool init_root_id(struct cgroupfs_root
*root
)
1344 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1346 spin_lock(&hierarchy_id_lock
);
1347 /* Try to allocate the next unused ID */
1348 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1349 &root
->hierarchy_id
);
1351 /* Try again starting from 0 */
1352 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1354 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1355 } else if (ret
!= -EAGAIN
) {
1356 /* Can only get here if the 31-bit IDR is full ... */
1359 spin_unlock(&hierarchy_id_lock
);
1364 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1366 struct cgroup_sb_opts
*opts
= data
;
1367 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1369 /* If we asked for a name then it must match */
1370 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1374 * If we asked for subsystems (or explicitly for no
1375 * subsystems) then they must match
1377 if ((opts
->subsys_bits
|| opts
->none
)
1378 && (opts
->subsys_bits
!= root
->subsys_bits
))
1384 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1386 struct cgroupfs_root
*root
;
1388 if (!opts
->subsys_bits
&& !opts
->none
)
1391 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1393 return ERR_PTR(-ENOMEM
);
1395 if (!init_root_id(root
)) {
1397 return ERR_PTR(-ENOMEM
);
1399 init_cgroup_root(root
);
1401 root
->subsys_bits
= opts
->subsys_bits
;
1402 root
->flags
= opts
->flags
;
1403 if (opts
->release_agent
)
1404 strcpy(root
->release_agent_path
, opts
->release_agent
);
1406 strcpy(root
->name
, opts
->name
);
1407 if (opts
->clone_children
)
1408 set_bit(CGRP_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1412 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1417 BUG_ON(!root
->hierarchy_id
);
1418 spin_lock(&hierarchy_id_lock
);
1419 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1420 spin_unlock(&hierarchy_id_lock
);
1424 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1427 struct cgroup_sb_opts
*opts
= data
;
1429 /* If we don't have a new root, we can't set up a new sb */
1430 if (!opts
->new_root
)
1433 BUG_ON(!opts
->subsys_bits
&& !opts
->none
);
1435 ret
= set_anon_super(sb
, NULL
);
1439 sb
->s_fs_info
= opts
->new_root
;
1440 opts
->new_root
->sb
= sb
;
1442 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1443 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1444 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1445 sb
->s_op
= &cgroup_ops
;
1450 static int cgroup_get_rootdir(struct super_block
*sb
)
1452 static const struct dentry_operations cgroup_dops
= {
1453 .d_iput
= cgroup_diput
,
1454 .d_delete
= cgroup_delete
,
1457 struct inode
*inode
=
1458 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1459 struct dentry
*dentry
;
1464 inode
->i_fop
= &simple_dir_operations
;
1465 inode
->i_op
= &cgroup_dir_inode_operations
;
1466 /* directories start off with i_nlink == 2 (for "." entry) */
1468 dentry
= d_alloc_root(inode
);
1473 sb
->s_root
= dentry
;
1474 /* for everything else we want ->d_op set */
1475 sb
->s_d_op
= &cgroup_dops
;
1479 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1480 int flags
, const char *unused_dev_name
,
1483 struct cgroup_sb_opts opts
;
1484 struct cgroupfs_root
*root
;
1486 struct super_block
*sb
;
1487 struct cgroupfs_root
*new_root
;
1489 /* First find the desired set of subsystems */
1490 mutex_lock(&cgroup_mutex
);
1491 ret
= parse_cgroupfs_options(data
, &opts
);
1492 mutex_unlock(&cgroup_mutex
);
1497 * Allocate a new cgroup root. We may not need it if we're
1498 * reusing an existing hierarchy.
1500 new_root
= cgroup_root_from_opts(&opts
);
1501 if (IS_ERR(new_root
)) {
1502 ret
= PTR_ERR(new_root
);
1505 opts
.new_root
= new_root
;
1507 /* Locate an existing or new sb for this hierarchy */
1508 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, &opts
);
1511 cgroup_drop_root(opts
.new_root
);
1515 root
= sb
->s_fs_info
;
1517 if (root
== opts
.new_root
) {
1518 /* We used the new root structure, so this is a new hierarchy */
1519 struct list_head tmp_cg_links
;
1520 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1521 struct inode
*inode
;
1522 struct cgroupfs_root
*existing_root
;
1525 BUG_ON(sb
->s_root
!= NULL
);
1527 ret
= cgroup_get_rootdir(sb
);
1529 goto drop_new_super
;
1530 inode
= sb
->s_root
->d_inode
;
1532 mutex_lock(&inode
->i_mutex
);
1533 mutex_lock(&cgroup_mutex
);
1535 if (strlen(root
->name
)) {
1536 /* Check for name clashes with existing mounts */
1537 for_each_active_root(existing_root
) {
1538 if (!strcmp(existing_root
->name
, root
->name
)) {
1540 mutex_unlock(&cgroup_mutex
);
1541 mutex_unlock(&inode
->i_mutex
);
1542 goto drop_new_super
;
1548 * We're accessing css_set_count without locking
1549 * css_set_lock here, but that's OK - it can only be
1550 * increased by someone holding cgroup_lock, and
1551 * that's us. The worst that can happen is that we
1552 * have some link structures left over
1554 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1556 mutex_unlock(&cgroup_mutex
);
1557 mutex_unlock(&inode
->i_mutex
);
1558 goto drop_new_super
;
1561 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1562 if (ret
== -EBUSY
) {
1563 mutex_unlock(&cgroup_mutex
);
1564 mutex_unlock(&inode
->i_mutex
);
1565 free_cg_links(&tmp_cg_links
);
1566 goto drop_new_super
;
1569 * There must be no failure case after here, since rebinding
1570 * takes care of subsystems' refcounts, which are explicitly
1571 * dropped in the failure exit path.
1574 /* EBUSY should be the only error here */
1577 list_add(&root
->root_list
, &roots
);
1580 sb
->s_root
->d_fsdata
= root_cgrp
;
1581 root
->top_cgroup
.dentry
= sb
->s_root
;
1583 /* Link the top cgroup in this hierarchy into all
1584 * the css_set objects */
1585 write_lock(&css_set_lock
);
1586 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1587 struct hlist_head
*hhead
= &css_set_table
[i
];
1588 struct hlist_node
*node
;
1591 hlist_for_each_entry(cg
, node
, hhead
, hlist
)
1592 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1594 write_unlock(&css_set_lock
);
1596 free_cg_links(&tmp_cg_links
);
1598 BUG_ON(!list_empty(&root_cgrp
->sibling
));
1599 BUG_ON(!list_empty(&root_cgrp
->children
));
1600 BUG_ON(root
->number_of_cgroups
!= 1);
1602 cgroup_populate_dir(root_cgrp
);
1603 mutex_unlock(&cgroup_mutex
);
1604 mutex_unlock(&inode
->i_mutex
);
1607 * We re-used an existing hierarchy - the new root (if
1608 * any) is not needed
1610 cgroup_drop_root(opts
.new_root
);
1611 /* no subsys rebinding, so refcounts don't change */
1612 drop_parsed_module_refcounts(opts
.subsys_bits
);
1615 kfree(opts
.release_agent
);
1617 return dget(sb
->s_root
);
1620 deactivate_locked_super(sb
);
1622 drop_parsed_module_refcounts(opts
.subsys_bits
);
1624 kfree(opts
.release_agent
);
1626 return ERR_PTR(ret
);
1629 static void cgroup_kill_sb(struct super_block
*sb
) {
1630 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1631 struct cgroup
*cgrp
= &root
->top_cgroup
;
1633 struct cg_cgroup_link
*link
;
1634 struct cg_cgroup_link
*saved_link
;
1638 BUG_ON(root
->number_of_cgroups
!= 1);
1639 BUG_ON(!list_empty(&cgrp
->children
));
1640 BUG_ON(!list_empty(&cgrp
->sibling
));
1642 mutex_lock(&cgroup_mutex
);
1644 /* Rebind all subsystems back to the default hierarchy */
1645 ret
= rebind_subsystems(root
, 0);
1646 /* Shouldn't be able to fail ... */
1650 * Release all the links from css_sets to this hierarchy's
1653 write_lock(&css_set_lock
);
1655 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1657 list_del(&link
->cg_link_list
);
1658 list_del(&link
->cgrp_link_list
);
1661 write_unlock(&css_set_lock
);
1663 if (!list_empty(&root
->root_list
)) {
1664 list_del(&root
->root_list
);
1668 mutex_unlock(&cgroup_mutex
);
1670 kill_litter_super(sb
);
1671 cgroup_drop_root(root
);
1674 static struct file_system_type cgroup_fs_type
= {
1676 .mount
= cgroup_mount
,
1677 .kill_sb
= cgroup_kill_sb
,
1680 static struct kobject
*cgroup_kobj
;
1682 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1684 return dentry
->d_fsdata
;
1687 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1689 return dentry
->d_fsdata
;
1693 * cgroup_path - generate the path of a cgroup
1694 * @cgrp: the cgroup in question
1695 * @buf: the buffer to write the path into
1696 * @buflen: the length of the buffer
1698 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1699 * reference. Writes path of cgroup into buf. Returns 0 on success,
1702 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1705 struct dentry
*dentry
= rcu_dereference_check(cgrp
->dentry
,
1706 rcu_read_lock_held() ||
1707 cgroup_lock_is_held());
1709 if (!dentry
|| cgrp
== dummytop
) {
1711 * Inactive subsystems have no dentry for their root
1718 start
= buf
+ buflen
;
1722 int len
= dentry
->d_name
.len
;
1724 if ((start
-= len
) < buf
)
1725 return -ENAMETOOLONG
;
1726 memcpy(start
, dentry
->d_name
.name
, len
);
1727 cgrp
= cgrp
->parent
;
1731 dentry
= rcu_dereference_check(cgrp
->dentry
,
1732 rcu_read_lock_held() ||
1733 cgroup_lock_is_held());
1737 return -ENAMETOOLONG
;
1740 memmove(buf
, start
, buf
+ buflen
- start
);
1743 EXPORT_SYMBOL_GPL(cgroup_path
);
1746 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1747 * @cgrp: the cgroup the task is attaching to
1748 * @tsk: the task to be attached
1750 * Call holding cgroup_mutex. May take task_lock of
1751 * the task 'tsk' during call.
1753 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1756 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1757 struct cgroup
*oldcgrp
;
1759 struct css_set
*newcg
;
1760 struct cgroupfs_root
*root
= cgrp
->root
;
1762 /* Nothing to do if the task is already in that cgroup */
1763 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1764 if (cgrp
== oldcgrp
)
1767 for_each_subsys(root
, ss
) {
1768 if (ss
->can_attach
) {
1769 retval
= ss
->can_attach(ss
, cgrp
, tsk
, false);
1772 * Remember on which subsystem the can_attach()
1773 * failed, so that we only call cancel_attach()
1774 * against the subsystems whose can_attach()
1775 * succeeded. (See below)
1788 * Locate or allocate a new css_set for this task,
1789 * based on its final set of cgroups
1791 newcg
= find_css_set(cg
, cgrp
);
1799 if (tsk
->flags
& PF_EXITING
) {
1805 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1808 /* Update the css_set linked lists if we're using them */
1809 write_lock(&css_set_lock
);
1810 if (!list_empty(&tsk
->cg_list
))
1811 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1812 write_unlock(&css_set_lock
);
1814 for_each_subsys(root
, ss
) {
1816 ss
->attach(ss
, cgrp
, oldcgrp
, tsk
, false);
1818 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1823 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1824 * is no longer empty.
1826 cgroup_wakeup_rmdir_waiter(cgrp
);
1829 for_each_subsys(root
, ss
) {
1830 if (ss
== failed_ss
)
1832 * This subsystem was the one that failed the
1833 * can_attach() check earlier, so we don't need
1834 * to call cancel_attach() against it or any
1835 * remaining subsystems.
1838 if (ss
->cancel_attach
)
1839 ss
->cancel_attach(ss
, cgrp
, tsk
, false);
1846 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1847 * @from: attach to all cgroups of a given task
1848 * @tsk: the task to be attached
1850 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
1852 struct cgroupfs_root
*root
;
1856 for_each_active_root(root
) {
1857 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
1859 retval
= cgroup_attach_task(from_cg
, tsk
);
1867 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
1870 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1871 * held. May take task_lock of task
1873 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
)
1875 struct task_struct
*tsk
;
1876 const struct cred
*cred
= current_cred(), *tcred
;
1881 tsk
= find_task_by_vpid(pid
);
1882 if (!tsk
|| tsk
->flags
& PF_EXITING
) {
1887 tcred
= __task_cred(tsk
);
1889 cred
->euid
!= tcred
->uid
&&
1890 cred
->euid
!= tcred
->suid
) {
1894 get_task_struct(tsk
);
1898 get_task_struct(tsk
);
1901 ret
= cgroup_attach_task(cgrp
, tsk
);
1902 put_task_struct(tsk
);
1906 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
1909 if (!cgroup_lock_live_group(cgrp
))
1911 ret
= attach_task_by_pid(cgrp
, pid
);
1917 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1918 * @cgrp: the cgroup to be checked for liveness
1920 * On success, returns true; the lock should be later released with
1921 * cgroup_unlock(). On failure returns false with no lock held.
1923 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
1925 mutex_lock(&cgroup_mutex
);
1926 if (cgroup_is_removed(cgrp
)) {
1927 mutex_unlock(&cgroup_mutex
);
1932 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
1934 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
1937 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
1938 if (strlen(buffer
) >= PATH_MAX
)
1940 if (!cgroup_lock_live_group(cgrp
))
1942 strcpy(cgrp
->root
->release_agent_path
, buffer
);
1947 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
1948 struct seq_file
*seq
)
1950 if (!cgroup_lock_live_group(cgrp
))
1952 seq_puts(seq
, cgrp
->root
->release_agent_path
);
1953 seq_putc(seq
, '\n');
1958 /* A buffer size big enough for numbers or short strings */
1959 #define CGROUP_LOCAL_BUFFER_SIZE 64
1961 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
1963 const char __user
*userbuf
,
1964 size_t nbytes
, loff_t
*unused_ppos
)
1966 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
1972 if (nbytes
>= sizeof(buffer
))
1974 if (copy_from_user(buffer
, userbuf
, nbytes
))
1977 buffer
[nbytes
] = 0; /* nul-terminate */
1978 if (cft
->write_u64
) {
1979 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
1982 retval
= cft
->write_u64(cgrp
, cft
, val
);
1984 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
1987 retval
= cft
->write_s64(cgrp
, cft
, val
);
1994 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
1996 const char __user
*userbuf
,
1997 size_t nbytes
, loff_t
*unused_ppos
)
1999 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2001 size_t max_bytes
= cft
->max_write_len
;
2002 char *buffer
= local_buffer
;
2005 max_bytes
= sizeof(local_buffer
) - 1;
2006 if (nbytes
>= max_bytes
)
2008 /* Allocate a dynamic buffer if we need one */
2009 if (nbytes
>= sizeof(local_buffer
)) {
2010 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2014 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2019 buffer
[nbytes
] = 0; /* nul-terminate */
2020 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2024 if (buffer
!= local_buffer
)
2029 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2030 size_t nbytes
, loff_t
*ppos
)
2032 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2033 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2035 if (cgroup_is_removed(cgrp
))
2038 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2039 if (cft
->write_u64
|| cft
->write_s64
)
2040 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2041 if (cft
->write_string
)
2042 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2044 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2045 return ret
? ret
: nbytes
;
2050 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2052 char __user
*buf
, size_t nbytes
,
2055 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2056 u64 val
= cft
->read_u64(cgrp
, cft
);
2057 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2059 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2062 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2064 char __user
*buf
, size_t nbytes
,
2067 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2068 s64 val
= cft
->read_s64(cgrp
, cft
);
2069 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2071 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2074 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2075 size_t nbytes
, loff_t
*ppos
)
2077 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2078 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2080 if (cgroup_is_removed(cgrp
))
2084 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2086 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2088 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2093 * seqfile ops/methods for returning structured data. Currently just
2094 * supports string->u64 maps, but can be extended in future.
2097 struct cgroup_seqfile_state
{
2099 struct cgroup
*cgroup
;
2102 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2104 struct seq_file
*sf
= cb
->state
;
2105 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2108 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2110 struct cgroup_seqfile_state
*state
= m
->private;
2111 struct cftype
*cft
= state
->cft
;
2112 if (cft
->read_map
) {
2113 struct cgroup_map_cb cb
= {
2114 .fill
= cgroup_map_add
,
2117 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2119 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2122 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2124 struct seq_file
*seq
= file
->private_data
;
2125 kfree(seq
->private);
2126 return single_release(inode
, file
);
2129 static const struct file_operations cgroup_seqfile_operations
= {
2131 .write
= cgroup_file_write
,
2132 .llseek
= seq_lseek
,
2133 .release
= cgroup_seqfile_release
,
2136 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2141 err
= generic_file_open(inode
, file
);
2144 cft
= __d_cft(file
->f_dentry
);
2146 if (cft
->read_map
|| cft
->read_seq_string
) {
2147 struct cgroup_seqfile_state
*state
=
2148 kzalloc(sizeof(*state
), GFP_USER
);
2152 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2153 file
->f_op
= &cgroup_seqfile_operations
;
2154 err
= single_open(file
, cgroup_seqfile_show
, state
);
2157 } else if (cft
->open
)
2158 err
= cft
->open(inode
, file
);
2165 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2167 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2169 return cft
->release(inode
, file
);
2174 * cgroup_rename - Only allow simple rename of directories in place.
2176 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2177 struct inode
*new_dir
, struct dentry
*new_dentry
)
2179 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2181 if (new_dentry
->d_inode
)
2183 if (old_dir
!= new_dir
)
2185 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2188 static const struct file_operations cgroup_file_operations
= {
2189 .read
= cgroup_file_read
,
2190 .write
= cgroup_file_write
,
2191 .llseek
= generic_file_llseek
,
2192 .open
= cgroup_file_open
,
2193 .release
= cgroup_file_release
,
2196 static const struct inode_operations cgroup_dir_inode_operations
= {
2197 .lookup
= cgroup_lookup
,
2198 .mkdir
= cgroup_mkdir
,
2199 .rmdir
= cgroup_rmdir
,
2200 .rename
= cgroup_rename
,
2203 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
2205 if (dentry
->d_name
.len
> NAME_MAX
)
2206 return ERR_PTR(-ENAMETOOLONG
);
2207 d_add(dentry
, NULL
);
2212 * Check if a file is a control file
2214 static inline struct cftype
*__file_cft(struct file
*file
)
2216 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2217 return ERR_PTR(-EINVAL
);
2218 return __d_cft(file
->f_dentry
);
2221 static int cgroup_create_file(struct dentry
*dentry
, mode_t mode
,
2222 struct super_block
*sb
)
2224 struct inode
*inode
;
2228 if (dentry
->d_inode
)
2231 inode
= cgroup_new_inode(mode
, sb
);
2235 if (S_ISDIR(mode
)) {
2236 inode
->i_op
= &cgroup_dir_inode_operations
;
2237 inode
->i_fop
= &simple_dir_operations
;
2239 /* start off with i_nlink == 2 (for "." entry) */
2242 /* start with the directory inode held, so that we can
2243 * populate it without racing with another mkdir */
2244 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2245 } else if (S_ISREG(mode
)) {
2247 inode
->i_fop
= &cgroup_file_operations
;
2249 d_instantiate(dentry
, inode
);
2250 dget(dentry
); /* Extra count - pin the dentry in core */
2255 * cgroup_create_dir - create a directory for an object.
2256 * @cgrp: the cgroup we create the directory for. It must have a valid
2257 * ->parent field. And we are going to fill its ->dentry field.
2258 * @dentry: dentry of the new cgroup
2259 * @mode: mode to set on new directory.
2261 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
2264 struct dentry
*parent
;
2267 parent
= cgrp
->parent
->dentry
;
2268 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
2270 dentry
->d_fsdata
= cgrp
;
2271 inc_nlink(parent
->d_inode
);
2272 rcu_assign_pointer(cgrp
->dentry
, dentry
);
2281 * cgroup_file_mode - deduce file mode of a control file
2282 * @cft: the control file in question
2284 * returns cft->mode if ->mode is not 0
2285 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2286 * returns S_IRUGO if it has only a read handler
2287 * returns S_IWUSR if it has only a write hander
2289 static mode_t
cgroup_file_mode(const struct cftype
*cft
)
2296 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2297 cft
->read_map
|| cft
->read_seq_string
)
2300 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2301 cft
->write_string
|| cft
->trigger
)
2307 int cgroup_add_file(struct cgroup
*cgrp
,
2308 struct cgroup_subsys
*subsys
,
2309 const struct cftype
*cft
)
2311 struct dentry
*dir
= cgrp
->dentry
;
2312 struct dentry
*dentry
;
2316 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2317 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2318 strcpy(name
, subsys
->name
);
2321 strcat(name
, cft
->name
);
2322 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2323 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2324 if (!IS_ERR(dentry
)) {
2325 mode
= cgroup_file_mode(cft
);
2326 error
= cgroup_create_file(dentry
, mode
| S_IFREG
,
2329 dentry
->d_fsdata
= (void *)cft
;
2332 error
= PTR_ERR(dentry
);
2335 EXPORT_SYMBOL_GPL(cgroup_add_file
);
2337 int cgroup_add_files(struct cgroup
*cgrp
,
2338 struct cgroup_subsys
*subsys
,
2339 const struct cftype cft
[],
2343 for (i
= 0; i
< count
; i
++) {
2344 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
2350 EXPORT_SYMBOL_GPL(cgroup_add_files
);
2353 * cgroup_task_count - count the number of tasks in a cgroup.
2354 * @cgrp: the cgroup in question
2356 * Return the number of tasks in the cgroup.
2358 int cgroup_task_count(const struct cgroup
*cgrp
)
2361 struct cg_cgroup_link
*link
;
2363 read_lock(&css_set_lock
);
2364 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2365 count
+= atomic_read(&link
->cg
->refcount
);
2367 read_unlock(&css_set_lock
);
2372 * Advance a list_head iterator. The iterator should be positioned at
2373 * the start of a css_set
2375 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2376 struct cgroup_iter
*it
)
2378 struct list_head
*l
= it
->cg_link
;
2379 struct cg_cgroup_link
*link
;
2382 /* Advance to the next non-empty css_set */
2385 if (l
== &cgrp
->css_sets
) {
2389 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2391 } while (list_empty(&cg
->tasks
));
2393 it
->task
= cg
->tasks
.next
;
2397 * To reduce the fork() overhead for systems that are not actually
2398 * using their cgroups capability, we don't maintain the lists running
2399 * through each css_set to its tasks until we see the list actually
2400 * used - in other words after the first call to cgroup_iter_start().
2402 * The tasklist_lock is not held here, as do_each_thread() and
2403 * while_each_thread() are protected by RCU.
2405 static void cgroup_enable_task_cg_lists(void)
2407 struct task_struct
*p
, *g
;
2408 write_lock(&css_set_lock
);
2409 use_task_css_set_links
= 1;
2410 do_each_thread(g
, p
) {
2413 * We should check if the process is exiting, otherwise
2414 * it will race with cgroup_exit() in that the list
2415 * entry won't be deleted though the process has exited.
2417 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2418 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2420 } while_each_thread(g
, p
);
2421 write_unlock(&css_set_lock
);
2424 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2427 * The first time anyone tries to iterate across a cgroup,
2428 * we need to enable the list linking each css_set to its
2429 * tasks, and fix up all existing tasks.
2431 if (!use_task_css_set_links
)
2432 cgroup_enable_task_cg_lists();
2434 read_lock(&css_set_lock
);
2435 it
->cg_link
= &cgrp
->css_sets
;
2436 cgroup_advance_iter(cgrp
, it
);
2439 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
2440 struct cgroup_iter
*it
)
2442 struct task_struct
*res
;
2443 struct list_head
*l
= it
->task
;
2444 struct cg_cgroup_link
*link
;
2446 /* If the iterator cg is NULL, we have no tasks */
2449 res
= list_entry(l
, struct task_struct
, cg_list
);
2450 /* Advance iterator to find next entry */
2452 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
2453 if (l
== &link
->cg
->tasks
) {
2454 /* We reached the end of this task list - move on to
2455 * the next cg_cgroup_link */
2456 cgroup_advance_iter(cgrp
, it
);
2463 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2465 read_unlock(&css_set_lock
);
2468 static inline int started_after_time(struct task_struct
*t1
,
2469 struct timespec
*time
,
2470 struct task_struct
*t2
)
2472 int start_diff
= timespec_compare(&t1
->start_time
, time
);
2473 if (start_diff
> 0) {
2475 } else if (start_diff
< 0) {
2479 * Arbitrarily, if two processes started at the same
2480 * time, we'll say that the lower pointer value
2481 * started first. Note that t2 may have exited by now
2482 * so this may not be a valid pointer any longer, but
2483 * that's fine - it still serves to distinguish
2484 * between two tasks started (effectively) simultaneously.
2491 * This function is a callback from heap_insert() and is used to order
2493 * In this case we order the heap in descending task start time.
2495 static inline int started_after(void *p1
, void *p2
)
2497 struct task_struct
*t1
= p1
;
2498 struct task_struct
*t2
= p2
;
2499 return started_after_time(t1
, &t2
->start_time
, t2
);
2503 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2504 * @scan: struct cgroup_scanner containing arguments for the scan
2506 * Arguments include pointers to callback functions test_task() and
2508 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2509 * and if it returns true, call process_task() for it also.
2510 * The test_task pointer may be NULL, meaning always true (select all tasks).
2511 * Effectively duplicates cgroup_iter_{start,next,end}()
2512 * but does not lock css_set_lock for the call to process_task().
2513 * The struct cgroup_scanner may be embedded in any structure of the caller's
2515 * It is guaranteed that process_task() will act on every task that
2516 * is a member of the cgroup for the duration of this call. This
2517 * function may or may not call process_task() for tasks that exit
2518 * or move to a different cgroup during the call, or are forked or
2519 * move into the cgroup during the call.
2521 * Note that test_task() may be called with locks held, and may in some
2522 * situations be called multiple times for the same task, so it should
2524 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2525 * pre-allocated and will be used for heap operations (and its "gt" member will
2526 * be overwritten), else a temporary heap will be used (allocation of which
2527 * may cause this function to fail).
2529 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
2532 struct cgroup_iter it
;
2533 struct task_struct
*p
, *dropped
;
2534 /* Never dereference latest_task, since it's not refcounted */
2535 struct task_struct
*latest_task
= NULL
;
2536 struct ptr_heap tmp_heap
;
2537 struct ptr_heap
*heap
;
2538 struct timespec latest_time
= { 0, 0 };
2541 /* The caller supplied our heap and pre-allocated its memory */
2543 heap
->gt
= &started_after
;
2545 /* We need to allocate our own heap memory */
2547 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
2549 /* cannot allocate the heap */
2555 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2556 * to determine which are of interest, and using the scanner's
2557 * "process_task" callback to process any of them that need an update.
2558 * Since we don't want to hold any locks during the task updates,
2559 * gather tasks to be processed in a heap structure.
2560 * The heap is sorted by descending task start time.
2561 * If the statically-sized heap fills up, we overflow tasks that
2562 * started later, and in future iterations only consider tasks that
2563 * started after the latest task in the previous pass. This
2564 * guarantees forward progress and that we don't miss any tasks.
2567 cgroup_iter_start(scan
->cg
, &it
);
2568 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
2570 * Only affect tasks that qualify per the caller's callback,
2571 * if he provided one
2573 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
2576 * Only process tasks that started after the last task
2579 if (!started_after_time(p
, &latest_time
, latest_task
))
2581 dropped
= heap_insert(heap
, p
);
2582 if (dropped
== NULL
) {
2584 * The new task was inserted; the heap wasn't
2588 } else if (dropped
!= p
) {
2590 * The new task was inserted, and pushed out a
2594 put_task_struct(dropped
);
2597 * Else the new task was newer than anything already in
2598 * the heap and wasn't inserted
2601 cgroup_iter_end(scan
->cg
, &it
);
2604 for (i
= 0; i
< heap
->size
; i
++) {
2605 struct task_struct
*q
= heap
->ptrs
[i
];
2607 latest_time
= q
->start_time
;
2610 /* Process the task per the caller's callback */
2611 scan
->process_task(q
, scan
);
2615 * If we had to process any tasks at all, scan again
2616 * in case some of them were in the middle of forking
2617 * children that didn't get processed.
2618 * Not the most efficient way to do it, but it avoids
2619 * having to take callback_mutex in the fork path
2623 if (heap
== &tmp_heap
)
2624 heap_free(&tmp_heap
);
2629 * Stuff for reading the 'tasks'/'procs' files.
2631 * Reading this file can return large amounts of data if a cgroup has
2632 * *lots* of attached tasks. So it may need several calls to read(),
2633 * but we cannot guarantee that the information we produce is correct
2634 * unless we produce it entirely atomically.
2639 * The following two functions "fix" the issue where there are more pids
2640 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2641 * TODO: replace with a kernel-wide solution to this problem
2643 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2644 static void *pidlist_allocate(int count
)
2646 if (PIDLIST_TOO_LARGE(count
))
2647 return vmalloc(count
* sizeof(pid_t
));
2649 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
2651 static void pidlist_free(void *p
)
2653 if (is_vmalloc_addr(p
))
2658 static void *pidlist_resize(void *p
, int newcount
)
2661 /* note: if new alloc fails, old p will still be valid either way */
2662 if (is_vmalloc_addr(p
)) {
2663 newlist
= vmalloc(newcount
* sizeof(pid_t
));
2666 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
2669 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
2675 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2676 * If the new stripped list is sufficiently smaller and there's enough memory
2677 * to allocate a new buffer, will let go of the unneeded memory. Returns the
2678 * number of unique elements.
2680 /* is the size difference enough that we should re-allocate the array? */
2681 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
2682 static int pidlist_uniq(pid_t
**p
, int length
)
2689 * we presume the 0th element is unique, so i starts at 1. trivial
2690 * edge cases first; no work needs to be done for either
2692 if (length
== 0 || length
== 1)
2694 /* src and dest walk down the list; dest counts unique elements */
2695 for (src
= 1; src
< length
; src
++) {
2696 /* find next unique element */
2697 while (list
[src
] == list
[src
-1]) {
2702 /* dest always points to where the next unique element goes */
2703 list
[dest
] = list
[src
];
2708 * if the length difference is large enough, we want to allocate a
2709 * smaller buffer to save memory. if this fails due to out of memory,
2710 * we'll just stay with what we've got.
2712 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
2713 newlist
= pidlist_resize(list
, dest
);
2720 static int cmppid(const void *a
, const void *b
)
2722 return *(pid_t
*)a
- *(pid_t
*)b
;
2726 * find the appropriate pidlist for our purpose (given procs vs tasks)
2727 * returns with the lock on that pidlist already held, and takes care
2728 * of the use count, or returns NULL with no locks held if we're out of
2731 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
2732 enum cgroup_filetype type
)
2734 struct cgroup_pidlist
*l
;
2735 /* don't need task_nsproxy() if we're looking at ourself */
2736 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
2739 * We can't drop the pidlist_mutex before taking the l->mutex in case
2740 * the last ref-holder is trying to remove l from the list at the same
2741 * time. Holding the pidlist_mutex precludes somebody taking whichever
2742 * list we find out from under us - compare release_pid_array().
2744 mutex_lock(&cgrp
->pidlist_mutex
);
2745 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
2746 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
2747 /* make sure l doesn't vanish out from under us */
2748 down_write(&l
->mutex
);
2749 mutex_unlock(&cgrp
->pidlist_mutex
);
2753 /* entry not found; create a new one */
2754 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
2756 mutex_unlock(&cgrp
->pidlist_mutex
);
2759 init_rwsem(&l
->mutex
);
2760 down_write(&l
->mutex
);
2762 l
->key
.ns
= get_pid_ns(ns
);
2763 l
->use_count
= 0; /* don't increment here */
2766 list_add(&l
->links
, &cgrp
->pidlists
);
2767 mutex_unlock(&cgrp
->pidlist_mutex
);
2772 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
2774 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
2775 struct cgroup_pidlist
**lp
)
2779 int pid
, n
= 0; /* used for populating the array */
2780 struct cgroup_iter it
;
2781 struct task_struct
*tsk
;
2782 struct cgroup_pidlist
*l
;
2785 * If cgroup gets more users after we read count, we won't have
2786 * enough space - tough. This race is indistinguishable to the
2787 * caller from the case that the additional cgroup users didn't
2788 * show up until sometime later on.
2790 length
= cgroup_task_count(cgrp
);
2791 array
= pidlist_allocate(length
);
2794 /* now, populate the array */
2795 cgroup_iter_start(cgrp
, &it
);
2796 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
2797 if (unlikely(n
== length
))
2799 /* get tgid or pid for procs or tasks file respectively */
2800 if (type
== CGROUP_FILE_PROCS
)
2801 pid
= task_tgid_vnr(tsk
);
2803 pid
= task_pid_vnr(tsk
);
2804 if (pid
> 0) /* make sure to only use valid results */
2807 cgroup_iter_end(cgrp
, &it
);
2809 /* now sort & (if procs) strip out duplicates */
2810 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
2811 if (type
== CGROUP_FILE_PROCS
)
2812 length
= pidlist_uniq(&array
, length
);
2813 l
= cgroup_pidlist_find(cgrp
, type
);
2815 pidlist_free(array
);
2818 /* store array, freeing old if necessary - lock already held */
2819 pidlist_free(l
->list
);
2823 up_write(&l
->mutex
);
2829 * cgroupstats_build - build and fill cgroupstats
2830 * @stats: cgroupstats to fill information into
2831 * @dentry: A dentry entry belonging to the cgroup for which stats have
2834 * Build and fill cgroupstats so that taskstats can export it to user
2837 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
2840 struct cgroup
*cgrp
;
2841 struct cgroup_iter it
;
2842 struct task_struct
*tsk
;
2845 * Validate dentry by checking the superblock operations,
2846 * and make sure it's a directory.
2848 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
2849 !S_ISDIR(dentry
->d_inode
->i_mode
))
2853 cgrp
= dentry
->d_fsdata
;
2855 cgroup_iter_start(cgrp
, &it
);
2856 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
2857 switch (tsk
->state
) {
2859 stats
->nr_running
++;
2861 case TASK_INTERRUPTIBLE
:
2862 stats
->nr_sleeping
++;
2864 case TASK_UNINTERRUPTIBLE
:
2865 stats
->nr_uninterruptible
++;
2868 stats
->nr_stopped
++;
2871 if (delayacct_is_task_waiting_on_io(tsk
))
2872 stats
->nr_io_wait
++;
2876 cgroup_iter_end(cgrp
, &it
);
2884 * seq_file methods for the tasks/procs files. The seq_file position is the
2885 * next pid to display; the seq_file iterator is a pointer to the pid
2886 * in the cgroup->l->list array.
2889 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
2892 * Initially we receive a position value that corresponds to
2893 * one more than the last pid shown (or 0 on the first call or
2894 * after a seek to the start). Use a binary-search to find the
2895 * next pid to display, if any
2897 struct cgroup_pidlist
*l
= s
->private;
2898 int index
= 0, pid
= *pos
;
2901 down_read(&l
->mutex
);
2903 int end
= l
->length
;
2905 while (index
< end
) {
2906 int mid
= (index
+ end
) / 2;
2907 if (l
->list
[mid
] == pid
) {
2910 } else if (l
->list
[mid
] <= pid
)
2916 /* If we're off the end of the array, we're done */
2917 if (index
>= l
->length
)
2919 /* Update the abstract position to be the actual pid that we found */
2920 iter
= l
->list
+ index
;
2925 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
2927 struct cgroup_pidlist
*l
= s
->private;
2931 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
2933 struct cgroup_pidlist
*l
= s
->private;
2935 pid_t
*end
= l
->list
+ l
->length
;
2937 * Advance to the next pid in the array. If this goes off the
2949 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
2951 return seq_printf(s
, "%d\n", *(int *)v
);
2955 * seq_operations functions for iterating on pidlists through seq_file -
2956 * independent of whether it's tasks or procs
2958 static const struct seq_operations cgroup_pidlist_seq_operations
= {
2959 .start
= cgroup_pidlist_start
,
2960 .stop
= cgroup_pidlist_stop
,
2961 .next
= cgroup_pidlist_next
,
2962 .show
= cgroup_pidlist_show
,
2965 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
2968 * the case where we're the last user of this particular pidlist will
2969 * have us remove it from the cgroup's list, which entails taking the
2970 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
2971 * pidlist_mutex, we have to take pidlist_mutex first.
2973 mutex_lock(&l
->owner
->pidlist_mutex
);
2974 down_write(&l
->mutex
);
2975 BUG_ON(!l
->use_count
);
2976 if (!--l
->use_count
) {
2977 /* we're the last user if refcount is 0; remove and free */
2978 list_del(&l
->links
);
2979 mutex_unlock(&l
->owner
->pidlist_mutex
);
2980 pidlist_free(l
->list
);
2981 put_pid_ns(l
->key
.ns
);
2982 up_write(&l
->mutex
);
2986 mutex_unlock(&l
->owner
->pidlist_mutex
);
2987 up_write(&l
->mutex
);
2990 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
2992 struct cgroup_pidlist
*l
;
2993 if (!(file
->f_mode
& FMODE_READ
))
2996 * the seq_file will only be initialized if the file was opened for
2997 * reading; hence we check if it's not null only in that case.
2999 l
= ((struct seq_file
*)file
->private_data
)->private;
3000 cgroup_release_pid_array(l
);
3001 return seq_release(inode
, file
);
3004 static const struct file_operations cgroup_pidlist_operations
= {
3006 .llseek
= seq_lseek
,
3007 .write
= cgroup_file_write
,
3008 .release
= cgroup_pidlist_release
,
3012 * The following functions handle opens on a file that displays a pidlist
3013 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3016 /* helper function for the two below it */
3017 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3019 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3020 struct cgroup_pidlist
*l
;
3023 /* Nothing to do for write-only files */
3024 if (!(file
->f_mode
& FMODE_READ
))
3027 /* have the array populated */
3028 retval
= pidlist_array_load(cgrp
, type
, &l
);
3031 /* configure file information */
3032 file
->f_op
= &cgroup_pidlist_operations
;
3034 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3036 cgroup_release_pid_array(l
);
3039 ((struct seq_file
*)file
->private_data
)->private = l
;
3042 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3044 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3046 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3048 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3051 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3054 return notify_on_release(cgrp
);
3057 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3061 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3063 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3065 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3070 * Unregister event and free resources.
3072 * Gets called from workqueue.
3074 static void cgroup_event_remove(struct work_struct
*work
)
3076 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3078 struct cgroup
*cgrp
= event
->cgrp
;
3080 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3082 eventfd_ctx_put(event
->eventfd
);
3088 * Gets called on POLLHUP on eventfd when user closes it.
3090 * Called with wqh->lock held and interrupts disabled.
3092 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3093 int sync
, void *key
)
3095 struct cgroup_event
*event
= container_of(wait
,
3096 struct cgroup_event
, wait
);
3097 struct cgroup
*cgrp
= event
->cgrp
;
3098 unsigned long flags
= (unsigned long)key
;
3100 if (flags
& POLLHUP
) {
3101 __remove_wait_queue(event
->wqh
, &event
->wait
);
3102 spin_lock(&cgrp
->event_list_lock
);
3103 list_del(&event
->list
);
3104 spin_unlock(&cgrp
->event_list_lock
);
3106 * We are in atomic context, but cgroup_event_remove() may
3107 * sleep, so we have to call it in workqueue.
3109 schedule_work(&event
->remove
);
3115 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3116 wait_queue_head_t
*wqh
, poll_table
*pt
)
3118 struct cgroup_event
*event
= container_of(pt
,
3119 struct cgroup_event
, pt
);
3122 add_wait_queue(wqh
, &event
->wait
);
3126 * Parse input and register new cgroup event handler.
3128 * Input must be in format '<event_fd> <control_fd> <args>'.
3129 * Interpretation of args is defined by control file implementation.
3131 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3134 struct cgroup_event
*event
= NULL
;
3135 unsigned int efd
, cfd
;
3136 struct file
*efile
= NULL
;
3137 struct file
*cfile
= NULL
;
3141 efd
= simple_strtoul(buffer
, &endp
, 10);
3146 cfd
= simple_strtoul(buffer
, &endp
, 10);
3147 if ((*endp
!= ' ') && (*endp
!= '\0'))
3151 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3155 INIT_LIST_HEAD(&event
->list
);
3156 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3157 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3158 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3160 efile
= eventfd_fget(efd
);
3161 if (IS_ERR(efile
)) {
3162 ret
= PTR_ERR(efile
);
3166 event
->eventfd
= eventfd_ctx_fileget(efile
);
3167 if (IS_ERR(event
->eventfd
)) {
3168 ret
= PTR_ERR(event
->eventfd
);
3178 /* the process need read permission on control file */
3179 ret
= file_permission(cfile
, MAY_READ
);
3183 event
->cft
= __file_cft(cfile
);
3184 if (IS_ERR(event
->cft
)) {
3185 ret
= PTR_ERR(event
->cft
);
3189 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3194 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3195 event
->eventfd
, buffer
);
3199 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3200 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3206 * Events should be removed after rmdir of cgroup directory, but before
3207 * destroying subsystem state objects. Let's take reference to cgroup
3208 * directory dentry to do that.
3212 spin_lock(&cgrp
->event_list_lock
);
3213 list_add(&event
->list
, &cgrp
->event_list
);
3214 spin_unlock(&cgrp
->event_list_lock
);
3225 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3226 eventfd_ctx_put(event
->eventfd
);
3228 if (!IS_ERR_OR_NULL(efile
))
3236 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3239 return clone_children(cgrp
);
3242 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3247 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3249 clear_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3254 * for the common functions, 'private' gives the type of file
3256 /* for hysterical raisins, we can't put this on the older files */
3257 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3258 static struct cftype files
[] = {
3261 .open
= cgroup_tasks_open
,
3262 .write_u64
= cgroup_tasks_write
,
3263 .release
= cgroup_pidlist_release
,
3264 .mode
= S_IRUGO
| S_IWUSR
,
3267 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3268 .open
= cgroup_procs_open
,
3269 /* .write_u64 = cgroup_procs_write, TODO */
3270 .release
= cgroup_pidlist_release
,
3274 .name
= "notify_on_release",
3275 .read_u64
= cgroup_read_notify_on_release
,
3276 .write_u64
= cgroup_write_notify_on_release
,
3279 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3280 .write_string
= cgroup_write_event_control
,
3284 .name
= "cgroup.clone_children",
3285 .read_u64
= cgroup_clone_children_read
,
3286 .write_u64
= cgroup_clone_children_write
,
3290 static struct cftype cft_release_agent
= {
3291 .name
= "release_agent",
3292 .read_seq_string
= cgroup_release_agent_show
,
3293 .write_string
= cgroup_release_agent_write
,
3294 .max_write_len
= PATH_MAX
,
3297 static int cgroup_populate_dir(struct cgroup
*cgrp
)
3300 struct cgroup_subsys
*ss
;
3302 /* First clear out any existing files */
3303 cgroup_clear_directory(cgrp
->dentry
);
3305 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
3309 if (cgrp
== cgrp
->top_cgroup
) {
3310 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
3314 for_each_subsys(cgrp
->root
, ss
) {
3315 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
3318 /* This cgroup is ready now */
3319 for_each_subsys(cgrp
->root
, ss
) {
3320 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3322 * Update id->css pointer and make this css visible from
3323 * CSS ID functions. This pointer will be dereferened
3324 * from RCU-read-side without locks.
3327 rcu_assign_pointer(css
->id
->css
, css
);
3333 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
3334 struct cgroup_subsys
*ss
,
3335 struct cgroup
*cgrp
)
3338 atomic_set(&css
->refcnt
, 1);
3341 if (cgrp
== dummytop
)
3342 set_bit(CSS_ROOT
, &css
->flags
);
3343 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
3344 cgrp
->subsys
[ss
->subsys_id
] = css
;
3347 static void cgroup_lock_hierarchy(struct cgroupfs_root
*root
)
3349 /* We need to take each hierarchy_mutex in a consistent order */
3353 * No worry about a race with rebind_subsystems that might mess up the
3354 * locking order, since both parties are under cgroup_mutex.
3356 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3357 struct cgroup_subsys
*ss
= subsys
[i
];
3360 if (ss
->root
== root
)
3361 mutex_lock(&ss
->hierarchy_mutex
);
3365 static void cgroup_unlock_hierarchy(struct cgroupfs_root
*root
)
3369 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3370 struct cgroup_subsys
*ss
= subsys
[i
];
3373 if (ss
->root
== root
)
3374 mutex_unlock(&ss
->hierarchy_mutex
);
3379 * cgroup_create - create a cgroup
3380 * @parent: cgroup that will be parent of the new cgroup
3381 * @dentry: dentry of the new cgroup
3382 * @mode: mode to set on new inode
3384 * Must be called with the mutex on the parent inode held
3386 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
3389 struct cgroup
*cgrp
;
3390 struct cgroupfs_root
*root
= parent
->root
;
3392 struct cgroup_subsys
*ss
;
3393 struct super_block
*sb
= root
->sb
;
3395 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3399 /* Grab a reference on the superblock so the hierarchy doesn't
3400 * get deleted on unmount if there are child cgroups. This
3401 * can be done outside cgroup_mutex, since the sb can't
3402 * disappear while someone has an open control file on the
3404 atomic_inc(&sb
->s_active
);
3406 mutex_lock(&cgroup_mutex
);
3408 init_cgroup_housekeeping(cgrp
);
3410 cgrp
->parent
= parent
;
3411 cgrp
->root
= parent
->root
;
3412 cgrp
->top_cgroup
= parent
->top_cgroup
;
3414 if (notify_on_release(parent
))
3415 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3417 if (clone_children(parent
))
3418 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3420 for_each_subsys(root
, ss
) {
3421 struct cgroup_subsys_state
*css
= ss
->create(ss
, cgrp
);
3427 init_cgroup_css(css
, ss
, cgrp
);
3429 err
= alloc_css_id(ss
, parent
, cgrp
);
3433 /* At error, ->destroy() callback has to free assigned ID. */
3434 if (clone_children(parent
) && ss
->post_clone
)
3435 ss
->post_clone(ss
, cgrp
);
3438 cgroup_lock_hierarchy(root
);
3439 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
3440 cgroup_unlock_hierarchy(root
);
3441 root
->number_of_cgroups
++;
3443 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
3447 /* The cgroup directory was pre-locked for us */
3448 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
3450 err
= cgroup_populate_dir(cgrp
);
3451 /* If err < 0, we have a half-filled directory - oh well ;) */
3453 mutex_unlock(&cgroup_mutex
);
3454 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
3460 cgroup_lock_hierarchy(root
);
3461 list_del(&cgrp
->sibling
);
3462 cgroup_unlock_hierarchy(root
);
3463 root
->number_of_cgroups
--;
3467 for_each_subsys(root
, ss
) {
3468 if (cgrp
->subsys
[ss
->subsys_id
])
3469 ss
->destroy(ss
, cgrp
);
3472 mutex_unlock(&cgroup_mutex
);
3474 /* Release the reference count that we took on the superblock */
3475 deactivate_super(sb
);
3481 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3483 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
3485 /* the vfs holds inode->i_mutex already */
3486 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
3489 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
3491 /* Check the reference count on each subsystem. Since we
3492 * already established that there are no tasks in the
3493 * cgroup, if the css refcount is also 1, then there should
3494 * be no outstanding references, so the subsystem is safe to
3495 * destroy. We scan across all subsystems rather than using
3496 * the per-hierarchy linked list of mounted subsystems since
3497 * we can be called via check_for_release() with no
3498 * synchronization other than RCU, and the subsystem linked
3499 * list isn't RCU-safe */
3502 * We won't need to lock the subsys array, because the subsystems
3503 * we're concerned about aren't going anywhere since our cgroup root
3504 * has a reference on them.
3506 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3507 struct cgroup_subsys
*ss
= subsys
[i
];
3508 struct cgroup_subsys_state
*css
;
3509 /* Skip subsystems not present or not in this hierarchy */
3510 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
3512 css
= cgrp
->subsys
[ss
->subsys_id
];
3513 /* When called from check_for_release() it's possible
3514 * that by this point the cgroup has been removed
3515 * and the css deleted. But a false-positive doesn't
3516 * matter, since it can only happen if the cgroup
3517 * has been deleted and hence no longer needs the
3518 * release agent to be called anyway. */
3519 if (css
&& (atomic_read(&css
->refcnt
) > 1))
3526 * Atomically mark all (or else none) of the cgroup's CSS objects as
3527 * CSS_REMOVED. Return true on success, or false if the cgroup has
3528 * busy subsystems. Call with cgroup_mutex held
3531 static int cgroup_clear_css_refs(struct cgroup
*cgrp
)
3533 struct cgroup_subsys
*ss
;
3534 unsigned long flags
;
3535 bool failed
= false;
3536 local_irq_save(flags
);
3537 for_each_subsys(cgrp
->root
, ss
) {
3538 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3541 /* We can only remove a CSS with a refcnt==1 */
3542 refcnt
= atomic_read(&css
->refcnt
);
3549 * Drop the refcnt to 0 while we check other
3550 * subsystems. This will cause any racing
3551 * css_tryget() to spin until we set the
3552 * CSS_REMOVED bits or abort
3554 if (atomic_cmpxchg(&css
->refcnt
, refcnt
, 0) == refcnt
)
3560 for_each_subsys(cgrp
->root
, ss
) {
3561 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3564 * Restore old refcnt if we previously managed
3565 * to clear it from 1 to 0
3567 if (!atomic_read(&css
->refcnt
))
3568 atomic_set(&css
->refcnt
, 1);
3570 /* Commit the fact that the CSS is removed */
3571 set_bit(CSS_REMOVED
, &css
->flags
);
3574 local_irq_restore(flags
);
3578 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
3580 struct cgroup
*cgrp
= dentry
->d_fsdata
;
3582 struct cgroup
*parent
;
3584 struct cgroup_event
*event
, *tmp
;
3587 /* the vfs holds both inode->i_mutex already */
3589 mutex_lock(&cgroup_mutex
);
3590 if (atomic_read(&cgrp
->count
) != 0) {
3591 mutex_unlock(&cgroup_mutex
);
3594 if (!list_empty(&cgrp
->children
)) {
3595 mutex_unlock(&cgroup_mutex
);
3598 mutex_unlock(&cgroup_mutex
);
3601 * In general, subsystem has no css->refcnt after pre_destroy(). But
3602 * in racy cases, subsystem may have to get css->refcnt after
3603 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3604 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3605 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3606 * and subsystem's reference count handling. Please see css_get/put
3607 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3609 set_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3612 * Call pre_destroy handlers of subsys. Notify subsystems
3613 * that rmdir() request comes.
3615 ret
= cgroup_call_pre_destroy(cgrp
);
3617 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3621 mutex_lock(&cgroup_mutex
);
3622 parent
= cgrp
->parent
;
3623 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
)) {
3624 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3625 mutex_unlock(&cgroup_mutex
);
3628 prepare_to_wait(&cgroup_rmdir_waitq
, &wait
, TASK_INTERRUPTIBLE
);
3629 if (!cgroup_clear_css_refs(cgrp
)) {
3630 mutex_unlock(&cgroup_mutex
);
3632 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3633 * prepare_to_wait(), we need to check this flag.
3635 if (test_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
))
3637 finish_wait(&cgroup_rmdir_waitq
, &wait
);
3638 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3639 if (signal_pending(current
))
3643 /* NO css_tryget() can success after here. */
3644 finish_wait(&cgroup_rmdir_waitq
, &wait
);
3645 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3647 spin_lock(&release_list_lock
);
3648 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
3649 if (!list_empty(&cgrp
->release_list
))
3650 list_del_init(&cgrp
->release_list
);
3651 spin_unlock(&release_list_lock
);
3653 cgroup_lock_hierarchy(cgrp
->root
);
3654 /* delete this cgroup from parent->children */
3655 list_del_init(&cgrp
->sibling
);
3656 cgroup_unlock_hierarchy(cgrp
->root
);
3658 d
= dget(cgrp
->dentry
);
3660 cgroup_d_remove_dir(d
);
3663 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
3664 check_for_release(parent
);
3667 * Unregister events and notify userspace.
3668 * Notify userspace about cgroup removing only after rmdir of cgroup
3669 * directory to avoid race between userspace and kernelspace
3671 spin_lock(&cgrp
->event_list_lock
);
3672 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
3673 list_del(&event
->list
);
3674 remove_wait_queue(event
->wqh
, &event
->wait
);
3675 eventfd_signal(event
->eventfd
, 1);
3676 schedule_work(&event
->remove
);
3678 spin_unlock(&cgrp
->event_list_lock
);
3680 mutex_unlock(&cgroup_mutex
);
3684 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
3686 struct cgroup_subsys_state
*css
;
3688 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
3690 /* Create the top cgroup state for this subsystem */
3691 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
3692 ss
->root
= &rootnode
;
3693 css
= ss
->create(ss
, dummytop
);
3694 /* We don't handle early failures gracefully */
3695 BUG_ON(IS_ERR(css
));
3696 init_cgroup_css(css
, ss
, dummytop
);
3698 /* Update the init_css_set to contain a subsys
3699 * pointer to this state - since the subsystem is
3700 * newly registered, all tasks and hence the
3701 * init_css_set is in the subsystem's top cgroup. */
3702 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
3704 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
3706 /* At system boot, before all subsystems have been
3707 * registered, no tasks have been forked, so we don't
3708 * need to invoke fork callbacks here. */
3709 BUG_ON(!list_empty(&init_task
.tasks
));
3711 mutex_init(&ss
->hierarchy_mutex
);
3712 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
3715 /* this function shouldn't be used with modular subsystems, since they
3716 * need to register a subsys_id, among other things */
3721 * cgroup_load_subsys: load and register a modular subsystem at runtime
3722 * @ss: the subsystem to load
3724 * This function should be called in a modular subsystem's initcall. If the
3725 * subsystem is built as a module, it will be assigned a new subsys_id and set
3726 * up for use. If the subsystem is built-in anyway, work is delegated to the
3727 * simpler cgroup_init_subsys.
3729 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
3732 struct cgroup_subsys_state
*css
;
3734 /* check name and function validity */
3735 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
3736 ss
->create
== NULL
|| ss
->destroy
== NULL
)
3740 * we don't support callbacks in modular subsystems. this check is
3741 * before the ss->module check for consistency; a subsystem that could
3742 * be a module should still have no callbacks even if the user isn't
3743 * compiling it as one.
3745 if (ss
->fork
|| ss
->exit
)
3749 * an optionally modular subsystem is built-in: we want to do nothing,
3750 * since cgroup_init_subsys will have already taken care of it.
3752 if (ss
->module
== NULL
) {
3753 /* a few sanity checks */
3754 BUG_ON(ss
->subsys_id
>= CGROUP_BUILTIN_SUBSYS_COUNT
);
3755 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
3760 * need to register a subsys id before anything else - for example,
3761 * init_cgroup_css needs it.
3763 mutex_lock(&cgroup_mutex
);
3764 /* find the first empty slot in the array */
3765 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3766 if (subsys
[i
] == NULL
)
3769 if (i
== CGROUP_SUBSYS_COUNT
) {
3770 /* maximum number of subsystems already registered! */
3771 mutex_unlock(&cgroup_mutex
);
3774 /* assign ourselves the subsys_id */
3779 * no ss->create seems to need anything important in the ss struct, so
3780 * this can happen first (i.e. before the rootnode attachment).
3782 css
= ss
->create(ss
, dummytop
);
3784 /* failure case - need to deassign the subsys[] slot. */
3786 mutex_unlock(&cgroup_mutex
);
3787 return PTR_ERR(css
);
3790 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
3791 ss
->root
= &rootnode
;
3793 /* our new subsystem will be attached to the dummy hierarchy. */
3794 init_cgroup_css(css
, ss
, dummytop
);
3795 /* init_idr must be after init_cgroup_css because it sets css->id. */
3797 int ret
= cgroup_init_idr(ss
, css
);
3799 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
3800 ss
->destroy(ss
, dummytop
);
3802 mutex_unlock(&cgroup_mutex
);
3808 * Now we need to entangle the css into the existing css_sets. unlike
3809 * in cgroup_init_subsys, there are now multiple css_sets, so each one
3810 * will need a new pointer to it; done by iterating the css_set_table.
3811 * furthermore, modifying the existing css_sets will corrupt the hash
3812 * table state, so each changed css_set will need its hash recomputed.
3813 * this is all done under the css_set_lock.
3815 write_lock(&css_set_lock
);
3816 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
3818 struct hlist_node
*node
, *tmp
;
3819 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
3821 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
3822 /* skip entries that we already rehashed */
3823 if (cg
->subsys
[ss
->subsys_id
])
3825 /* remove existing entry */
3826 hlist_del(&cg
->hlist
);
3828 cg
->subsys
[ss
->subsys_id
] = css
;
3829 /* recompute hash and restore entry */
3830 new_bucket
= css_set_hash(cg
->subsys
);
3831 hlist_add_head(&cg
->hlist
, new_bucket
);
3834 write_unlock(&css_set_lock
);
3836 mutex_init(&ss
->hierarchy_mutex
);
3837 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
3841 mutex_unlock(&cgroup_mutex
);
3844 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
3847 * cgroup_unload_subsys: unload a modular subsystem
3848 * @ss: the subsystem to unload
3850 * This function should be called in a modular subsystem's exitcall. When this
3851 * function is invoked, the refcount on the subsystem's module will be 0, so
3852 * the subsystem will not be attached to any hierarchy.
3854 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
3856 struct cg_cgroup_link
*link
;
3857 struct hlist_head
*hhead
;
3859 BUG_ON(ss
->module
== NULL
);
3862 * we shouldn't be called if the subsystem is in use, and the use of
3863 * try_module_get in parse_cgroupfs_options should ensure that it
3864 * doesn't start being used while we're killing it off.
3866 BUG_ON(ss
->root
!= &rootnode
);
3868 mutex_lock(&cgroup_mutex
);
3869 /* deassign the subsys_id */
3870 BUG_ON(ss
->subsys_id
< CGROUP_BUILTIN_SUBSYS_COUNT
);
3871 subsys
[ss
->subsys_id
] = NULL
;
3873 /* remove subsystem from rootnode's list of subsystems */
3874 list_del_init(&ss
->sibling
);
3877 * disentangle the css from all css_sets attached to the dummytop. as
3878 * in loading, we need to pay our respects to the hashtable gods.
3880 write_lock(&css_set_lock
);
3881 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
3882 struct css_set
*cg
= link
->cg
;
3884 hlist_del(&cg
->hlist
);
3885 BUG_ON(!cg
->subsys
[ss
->subsys_id
]);
3886 cg
->subsys
[ss
->subsys_id
] = NULL
;
3887 hhead
= css_set_hash(cg
->subsys
);
3888 hlist_add_head(&cg
->hlist
, hhead
);
3890 write_unlock(&css_set_lock
);
3893 * remove subsystem's css from the dummytop and free it - need to free
3894 * before marking as null because ss->destroy needs the cgrp->subsys
3895 * pointer to find their state. note that this also takes care of
3896 * freeing the css_id.
3898 ss
->destroy(ss
, dummytop
);
3899 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
3901 mutex_unlock(&cgroup_mutex
);
3903 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
3906 * cgroup_init_early - cgroup initialization at system boot
3908 * Initialize cgroups at system boot, and initialize any
3909 * subsystems that request early init.
3911 int __init
cgroup_init_early(void)
3914 atomic_set(&init_css_set
.refcount
, 1);
3915 INIT_LIST_HEAD(&init_css_set
.cg_links
);
3916 INIT_LIST_HEAD(&init_css_set
.tasks
);
3917 INIT_HLIST_NODE(&init_css_set
.hlist
);
3919 init_cgroup_root(&rootnode
);
3921 init_task
.cgroups
= &init_css_set
;
3923 init_css_set_link
.cg
= &init_css_set
;
3924 init_css_set_link
.cgrp
= dummytop
;
3925 list_add(&init_css_set_link
.cgrp_link_list
,
3926 &rootnode
.top_cgroup
.css_sets
);
3927 list_add(&init_css_set_link
.cg_link_list
,
3928 &init_css_set
.cg_links
);
3930 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
3931 INIT_HLIST_HEAD(&css_set_table
[i
]);
3933 /* at bootup time, we don't worry about modular subsystems */
3934 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
3935 struct cgroup_subsys
*ss
= subsys
[i
];
3938 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
3939 BUG_ON(!ss
->create
);
3940 BUG_ON(!ss
->destroy
);
3941 if (ss
->subsys_id
!= i
) {
3942 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
3943 ss
->name
, ss
->subsys_id
);
3948 cgroup_init_subsys(ss
);
3954 * cgroup_init - cgroup initialization
3956 * Register cgroup filesystem and /proc file, and initialize
3957 * any subsystems that didn't request early init.
3959 int __init
cgroup_init(void)
3963 struct hlist_head
*hhead
;
3965 err
= bdi_init(&cgroup_backing_dev_info
);
3969 /* at bootup time, we don't worry about modular subsystems */
3970 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
3971 struct cgroup_subsys
*ss
= subsys
[i
];
3972 if (!ss
->early_init
)
3973 cgroup_init_subsys(ss
);
3975 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
3978 /* Add init_css_set to the hash table */
3979 hhead
= css_set_hash(init_css_set
.subsys
);
3980 hlist_add_head(&init_css_set
.hlist
, hhead
);
3981 BUG_ON(!init_root_id(&rootnode
));
3983 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
3989 err
= register_filesystem(&cgroup_fs_type
);
3991 kobject_put(cgroup_kobj
);
3995 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
3999 bdi_destroy(&cgroup_backing_dev_info
);
4005 * proc_cgroup_show()
4006 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4007 * - Used for /proc/<pid>/cgroup.
4008 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4009 * doesn't really matter if tsk->cgroup changes after we read it,
4010 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4011 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4012 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4013 * cgroup to top_cgroup.
4016 /* TODO: Use a proper seq_file iterator */
4017 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4020 struct task_struct
*tsk
;
4023 struct cgroupfs_root
*root
;
4026 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4032 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4038 mutex_lock(&cgroup_mutex
);
4040 for_each_active_root(root
) {
4041 struct cgroup_subsys
*ss
;
4042 struct cgroup
*cgrp
;
4045 seq_printf(m
, "%d:", root
->hierarchy_id
);
4046 for_each_subsys(root
, ss
)
4047 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4048 if (strlen(root
->name
))
4049 seq_printf(m
, "%sname=%s", count
? "," : "",
4052 cgrp
= task_cgroup_from_root(tsk
, root
);
4053 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4061 mutex_unlock(&cgroup_mutex
);
4062 put_task_struct(tsk
);
4069 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4071 struct pid
*pid
= PROC_I(inode
)->pid
;
4072 return single_open(file
, proc_cgroup_show
, pid
);
4075 const struct file_operations proc_cgroup_operations
= {
4076 .open
= cgroup_open
,
4078 .llseek
= seq_lseek
,
4079 .release
= single_release
,
4082 /* Display information about each subsystem and each hierarchy */
4083 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4087 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4089 * ideally we don't want subsystems moving around while we do this.
4090 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4091 * subsys/hierarchy state.
4093 mutex_lock(&cgroup_mutex
);
4094 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4095 struct cgroup_subsys
*ss
= subsys
[i
];
4098 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4099 ss
->name
, ss
->root
->hierarchy_id
,
4100 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4102 mutex_unlock(&cgroup_mutex
);
4106 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4108 return single_open(file
, proc_cgroupstats_show
, NULL
);
4111 static const struct file_operations proc_cgroupstats_operations
= {
4112 .open
= cgroupstats_open
,
4114 .llseek
= seq_lseek
,
4115 .release
= single_release
,
4119 * cgroup_fork - attach newly forked task to its parents cgroup.
4120 * @child: pointer to task_struct of forking parent process.
4122 * Description: A task inherits its parent's cgroup at fork().
4124 * A pointer to the shared css_set was automatically copied in
4125 * fork.c by dup_task_struct(). However, we ignore that copy, since
4126 * it was not made under the protection of RCU or cgroup_mutex, so
4127 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4128 * have already changed current->cgroups, allowing the previously
4129 * referenced cgroup group to be removed and freed.
4131 * At the point that cgroup_fork() is called, 'current' is the parent
4132 * task, and the passed argument 'child' points to the child task.
4134 void cgroup_fork(struct task_struct
*child
)
4137 child
->cgroups
= current
->cgroups
;
4138 get_css_set(child
->cgroups
);
4139 task_unlock(current
);
4140 INIT_LIST_HEAD(&child
->cg_list
);
4144 * cgroup_fork_callbacks - run fork callbacks
4145 * @child: the new task
4147 * Called on a new task very soon before adding it to the
4148 * tasklist. No need to take any locks since no-one can
4149 * be operating on this task.
4151 void cgroup_fork_callbacks(struct task_struct
*child
)
4153 if (need_forkexit_callback
) {
4156 * forkexit callbacks are only supported for builtin
4157 * subsystems, and the builtin section of the subsys array is
4158 * immutable, so we don't need to lock the subsys array here.
4160 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4161 struct cgroup_subsys
*ss
= subsys
[i
];
4163 ss
->fork(ss
, child
);
4169 * cgroup_post_fork - called on a new task after adding it to the task list
4170 * @child: the task in question
4172 * Adds the task to the list running through its css_set if necessary.
4173 * Has to be after the task is visible on the task list in case we race
4174 * with the first call to cgroup_iter_start() - to guarantee that the
4175 * new task ends up on its list.
4177 void cgroup_post_fork(struct task_struct
*child
)
4179 if (use_task_css_set_links
) {
4180 write_lock(&css_set_lock
);
4182 if (list_empty(&child
->cg_list
))
4183 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4185 write_unlock(&css_set_lock
);
4189 * cgroup_exit - detach cgroup from exiting task
4190 * @tsk: pointer to task_struct of exiting process
4191 * @run_callback: run exit callbacks?
4193 * Description: Detach cgroup from @tsk and release it.
4195 * Note that cgroups marked notify_on_release force every task in
4196 * them to take the global cgroup_mutex mutex when exiting.
4197 * This could impact scaling on very large systems. Be reluctant to
4198 * use notify_on_release cgroups where very high task exit scaling
4199 * is required on large systems.
4201 * the_top_cgroup_hack:
4203 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4205 * We call cgroup_exit() while the task is still competent to
4206 * handle notify_on_release(), then leave the task attached to the
4207 * root cgroup in each hierarchy for the remainder of its exit.
4209 * To do this properly, we would increment the reference count on
4210 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4211 * code we would add a second cgroup function call, to drop that
4212 * reference. This would just create an unnecessary hot spot on
4213 * the top_cgroup reference count, to no avail.
4215 * Normally, holding a reference to a cgroup without bumping its
4216 * count is unsafe. The cgroup could go away, or someone could
4217 * attach us to a different cgroup, decrementing the count on
4218 * the first cgroup that we never incremented. But in this case,
4219 * top_cgroup isn't going away, and either task has PF_EXITING set,
4220 * which wards off any cgroup_attach_task() attempts, or task is a failed
4221 * fork, never visible to cgroup_attach_task.
4223 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4229 * Unlink from the css_set task list if necessary.
4230 * Optimistically check cg_list before taking
4233 if (!list_empty(&tsk
->cg_list
)) {
4234 write_lock(&css_set_lock
);
4235 if (!list_empty(&tsk
->cg_list
))
4236 list_del_init(&tsk
->cg_list
);
4237 write_unlock(&css_set_lock
);
4240 /* Reassign the task to the init_css_set. */
4243 tsk
->cgroups
= &init_css_set
;
4245 if (run_callbacks
&& need_forkexit_callback
) {
4247 * modular subsystems can't use callbacks, so no need to lock
4250 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4251 struct cgroup_subsys
*ss
= subsys
[i
];
4253 struct cgroup
*old_cgrp
=
4254 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4255 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4256 ss
->exit(ss
, cgrp
, old_cgrp
, tsk
);
4263 put_css_set_taskexit(cg
);
4267 * cgroup_clone - clone the cgroup the given subsystem is attached to
4268 * @tsk: the task to be moved
4269 * @subsys: the given subsystem
4270 * @nodename: the name for the new cgroup
4272 * Duplicate the current cgroup in the hierarchy that the given
4273 * subsystem is attached to, and move this task into the new
4276 int cgroup_clone(struct task_struct
*tsk
, struct cgroup_subsys
*subsys
,
4279 struct dentry
*dentry
;
4281 struct cgroup
*parent
, *child
;
4282 struct inode
*inode
;
4284 struct cgroupfs_root
*root
;
4285 struct cgroup_subsys
*ss
;
4287 /* We shouldn't be called by an unregistered subsystem */
4288 BUG_ON(!subsys
->active
);
4290 /* First figure out what hierarchy and cgroup we're dealing
4291 * with, and pin them so we can drop cgroup_mutex */
4292 mutex_lock(&cgroup_mutex
);
4294 root
= subsys
->root
;
4295 if (root
== &rootnode
) {
4296 mutex_unlock(&cgroup_mutex
);
4300 /* Pin the hierarchy */
4301 if (!atomic_inc_not_zero(&root
->sb
->s_active
)) {
4302 /* We race with the final deactivate_super() */
4303 mutex_unlock(&cgroup_mutex
);
4307 /* Keep the cgroup alive */
4309 parent
= task_cgroup(tsk
, subsys
->subsys_id
);
4314 mutex_unlock(&cgroup_mutex
);
4316 /* Now do the VFS work to create a cgroup */
4317 inode
= parent
->dentry
->d_inode
;
4319 /* Hold the parent directory mutex across this operation to
4320 * stop anyone else deleting the new cgroup */
4321 mutex_lock(&inode
->i_mutex
);
4322 dentry
= lookup_one_len(nodename
, parent
->dentry
, strlen(nodename
));
4323 if (IS_ERR(dentry
)) {
4325 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename
,
4327 ret
= PTR_ERR(dentry
);
4331 /* Create the cgroup directory, which also creates the cgroup */
4332 ret
= vfs_mkdir(inode
, dentry
, 0755);
4333 child
= __d_cgrp(dentry
);
4337 "Failed to create cgroup %s: %d\n", nodename
,
4342 /* The cgroup now exists. Retake cgroup_mutex and check
4343 * that we're still in the same state that we thought we
4345 mutex_lock(&cgroup_mutex
);
4346 if ((root
!= subsys
->root
) ||
4347 (parent
!= task_cgroup(tsk
, subsys
->subsys_id
))) {
4348 /* Aargh, we raced ... */
4349 mutex_unlock(&inode
->i_mutex
);
4352 deactivate_super(root
->sb
);
4353 /* The cgroup is still accessible in the VFS, but
4354 * we're not going to try to rmdir() it at this
4357 "Race in cgroup_clone() - leaking cgroup %s\n",
4362 /* do any required auto-setup */
4363 for_each_subsys(root
, ss
) {
4365 ss
->post_clone(ss
, child
);
4368 /* All seems fine. Finish by moving the task into the new cgroup */
4369 ret
= cgroup_attach_task(child
, tsk
);
4370 mutex_unlock(&cgroup_mutex
);
4373 mutex_unlock(&inode
->i_mutex
);
4375 mutex_lock(&cgroup_mutex
);
4377 mutex_unlock(&cgroup_mutex
);
4378 deactivate_super(root
->sb
);
4383 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4384 * @cgrp: the cgroup in question
4385 * @task: the task in question
4387 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4390 * If we are sending in dummytop, then presumably we are creating
4391 * the top cgroup in the subsystem.
4393 * Called only by the ns (nsproxy) cgroup.
4395 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
4398 struct cgroup
*target
;
4400 if (cgrp
== dummytop
)
4403 target
= task_cgroup_from_root(task
, cgrp
->root
);
4404 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
4405 cgrp
= cgrp
->parent
;
4406 ret
= (cgrp
== target
);
4410 static void check_for_release(struct cgroup
*cgrp
)
4412 /* All of these checks rely on RCU to keep the cgroup
4413 * structure alive */
4414 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
4415 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
4416 /* Control Group is currently removeable. If it's not
4417 * already queued for a userspace notification, queue
4419 int need_schedule_work
= 0;
4420 spin_lock(&release_list_lock
);
4421 if (!cgroup_is_removed(cgrp
) &&
4422 list_empty(&cgrp
->release_list
)) {
4423 list_add(&cgrp
->release_list
, &release_list
);
4424 need_schedule_work
= 1;
4426 spin_unlock(&release_list_lock
);
4427 if (need_schedule_work
)
4428 schedule_work(&release_agent_work
);
4432 /* Caller must verify that the css is not for root cgroup */
4433 void __css_put(struct cgroup_subsys_state
*css
, int count
)
4435 struct cgroup
*cgrp
= css
->cgroup
;
4438 val
= atomic_sub_return(count
, &css
->refcnt
);
4440 if (notify_on_release(cgrp
)) {
4441 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4442 check_for_release(cgrp
);
4444 cgroup_wakeup_rmdir_waiter(cgrp
);
4447 WARN_ON_ONCE(val
< 1);
4449 EXPORT_SYMBOL_GPL(__css_put
);
4452 * Notify userspace when a cgroup is released, by running the
4453 * configured release agent with the name of the cgroup (path
4454 * relative to the root of cgroup file system) as the argument.
4456 * Most likely, this user command will try to rmdir this cgroup.
4458 * This races with the possibility that some other task will be
4459 * attached to this cgroup before it is removed, or that some other
4460 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4461 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4462 * unused, and this cgroup will be reprieved from its death sentence,
4463 * to continue to serve a useful existence. Next time it's released,
4464 * we will get notified again, if it still has 'notify_on_release' set.
4466 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4467 * means only wait until the task is successfully execve()'d. The
4468 * separate release agent task is forked by call_usermodehelper(),
4469 * then control in this thread returns here, without waiting for the
4470 * release agent task. We don't bother to wait because the caller of
4471 * this routine has no use for the exit status of the release agent
4472 * task, so no sense holding our caller up for that.
4474 static void cgroup_release_agent(struct work_struct
*work
)
4476 BUG_ON(work
!= &release_agent_work
);
4477 mutex_lock(&cgroup_mutex
);
4478 spin_lock(&release_list_lock
);
4479 while (!list_empty(&release_list
)) {
4480 char *argv
[3], *envp
[3];
4482 char *pathbuf
= NULL
, *agentbuf
= NULL
;
4483 struct cgroup
*cgrp
= list_entry(release_list
.next
,
4486 list_del_init(&cgrp
->release_list
);
4487 spin_unlock(&release_list_lock
);
4488 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4491 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
4493 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
4498 argv
[i
++] = agentbuf
;
4499 argv
[i
++] = pathbuf
;
4503 /* minimal command environment */
4504 envp
[i
++] = "HOME=/";
4505 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4508 /* Drop the lock while we invoke the usermode helper,
4509 * since the exec could involve hitting disk and hence
4510 * be a slow process */
4511 mutex_unlock(&cgroup_mutex
);
4512 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
4513 mutex_lock(&cgroup_mutex
);
4517 spin_lock(&release_list_lock
);
4519 spin_unlock(&release_list_lock
);
4520 mutex_unlock(&cgroup_mutex
);
4523 static int __init
cgroup_disable(char *str
)
4528 while ((token
= strsep(&str
, ",")) != NULL
) {
4532 * cgroup_disable, being at boot time, can't know about module
4533 * subsystems, so we don't worry about them.
4535 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4536 struct cgroup_subsys
*ss
= subsys
[i
];
4538 if (!strcmp(token
, ss
->name
)) {
4540 printk(KERN_INFO
"Disabling %s control group"
4541 " subsystem\n", ss
->name
);
4548 __setup("cgroup_disable=", cgroup_disable
);
4551 * Functons for CSS ID.
4555 *To get ID other than 0, this should be called when !cgroup_is_removed().
4557 unsigned short css_id(struct cgroup_subsys_state
*css
)
4559 struct css_id
*cssid
;
4562 * This css_id() can return correct value when somone has refcnt
4563 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4564 * it's unchanged until freed.
4566 cssid
= rcu_dereference_check(css
->id
,
4567 rcu_read_lock_held() || atomic_read(&css
->refcnt
));
4573 EXPORT_SYMBOL_GPL(css_id
);
4575 unsigned short css_depth(struct cgroup_subsys_state
*css
)
4577 struct css_id
*cssid
;
4579 cssid
= rcu_dereference_check(css
->id
,
4580 rcu_read_lock_held() || atomic_read(&css
->refcnt
));
4583 return cssid
->depth
;
4586 EXPORT_SYMBOL_GPL(css_depth
);
4589 * css_is_ancestor - test "root" css is an ancestor of "child"
4590 * @child: the css to be tested.
4591 * @root: the css supporsed to be an ancestor of the child.
4593 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4594 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4595 * But, considering usual usage, the csses should be valid objects after test.
4596 * Assuming that the caller will do some action to the child if this returns
4597 * returns true, the caller must take "child";s reference count.
4598 * If "child" is valid object and this returns true, "root" is valid, too.
4601 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
4602 const struct cgroup_subsys_state
*root
)
4604 struct css_id
*child_id
;
4605 struct css_id
*root_id
;
4609 child_id
= rcu_dereference(child
->id
);
4610 root_id
= rcu_dereference(root
->id
);
4613 || (child_id
->depth
< root_id
->depth
)
4614 || (child_id
->stack
[root_id
->depth
] != root_id
->id
))
4620 static void __free_css_id_cb(struct rcu_head
*head
)
4624 id
= container_of(head
, struct css_id
, rcu_head
);
4628 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
4630 struct css_id
*id
= css
->id
;
4631 /* When this is called before css_id initialization, id can be NULL */
4635 BUG_ON(!ss
->use_id
);
4637 rcu_assign_pointer(id
->css
, NULL
);
4638 rcu_assign_pointer(css
->id
, NULL
);
4639 spin_lock(&ss
->id_lock
);
4640 idr_remove(&ss
->idr
, id
->id
);
4641 spin_unlock(&ss
->id_lock
);
4642 call_rcu(&id
->rcu_head
, __free_css_id_cb
);
4644 EXPORT_SYMBOL_GPL(free_css_id
);
4647 * This is called by init or create(). Then, calls to this function are
4648 * always serialized (By cgroup_mutex() at create()).
4651 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
4653 struct css_id
*newid
;
4654 int myid
, error
, size
;
4656 BUG_ON(!ss
->use_id
);
4658 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
4659 newid
= kzalloc(size
, GFP_KERNEL
);
4661 return ERR_PTR(-ENOMEM
);
4663 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
4667 spin_lock(&ss
->id_lock
);
4668 /* Don't use 0. allocates an ID of 1-65535 */
4669 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
4670 spin_unlock(&ss
->id_lock
);
4672 /* Returns error when there are no free spaces for new ID.*/
4677 if (myid
> CSS_ID_MAX
)
4681 newid
->depth
= depth
;
4685 spin_lock(&ss
->id_lock
);
4686 idr_remove(&ss
->idr
, myid
);
4687 spin_unlock(&ss
->id_lock
);
4690 return ERR_PTR(error
);
4694 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
4695 struct cgroup_subsys_state
*rootcss
)
4697 struct css_id
*newid
;
4699 spin_lock_init(&ss
->id_lock
);
4702 newid
= get_new_cssid(ss
, 0);
4704 return PTR_ERR(newid
);
4706 newid
->stack
[0] = newid
->id
;
4707 newid
->css
= rootcss
;
4708 rootcss
->id
= newid
;
4712 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
4713 struct cgroup
*child
)
4715 int subsys_id
, i
, depth
= 0;
4716 struct cgroup_subsys_state
*parent_css
, *child_css
;
4717 struct css_id
*child_id
, *parent_id
;
4719 subsys_id
= ss
->subsys_id
;
4720 parent_css
= parent
->subsys
[subsys_id
];
4721 child_css
= child
->subsys
[subsys_id
];
4722 parent_id
= parent_css
->id
;
4723 depth
= parent_id
->depth
+ 1;
4725 child_id
= get_new_cssid(ss
, depth
);
4726 if (IS_ERR(child_id
))
4727 return PTR_ERR(child_id
);
4729 for (i
= 0; i
< depth
; i
++)
4730 child_id
->stack
[i
] = parent_id
->stack
[i
];
4731 child_id
->stack
[depth
] = child_id
->id
;
4733 * child_id->css pointer will be set after this cgroup is available
4734 * see cgroup_populate_dir()
4736 rcu_assign_pointer(child_css
->id
, child_id
);
4742 * css_lookup - lookup css by id
4743 * @ss: cgroup subsys to be looked into.
4746 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4747 * NULL if not. Should be called under rcu_read_lock()
4749 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
4751 struct css_id
*cssid
= NULL
;
4753 BUG_ON(!ss
->use_id
);
4754 cssid
= idr_find(&ss
->idr
, id
);
4756 if (unlikely(!cssid
))
4759 return rcu_dereference(cssid
->css
);
4761 EXPORT_SYMBOL_GPL(css_lookup
);
4764 * css_get_next - lookup next cgroup under specified hierarchy.
4765 * @ss: pointer to subsystem
4766 * @id: current position of iteration.
4767 * @root: pointer to css. search tree under this.
4768 * @foundid: position of found object.
4770 * Search next css under the specified hierarchy of rootid. Calling under
4771 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
4773 struct cgroup_subsys_state
*
4774 css_get_next(struct cgroup_subsys
*ss
, int id
,
4775 struct cgroup_subsys_state
*root
, int *foundid
)
4777 struct cgroup_subsys_state
*ret
= NULL
;
4780 int rootid
= css_id(root
);
4781 int depth
= css_depth(root
);
4786 BUG_ON(!ss
->use_id
);
4787 /* fill start point for scan */
4791 * scan next entry from bitmap(tree), tmpid is updated after
4794 spin_lock(&ss
->id_lock
);
4795 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
4796 spin_unlock(&ss
->id_lock
);
4800 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
4801 ret
= rcu_dereference(tmp
->css
);
4807 /* continue to scan from next id */
4814 * get corresponding css from file open on cgroupfs directory
4816 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
4818 struct cgroup
*cgrp
;
4819 struct inode
*inode
;
4820 struct cgroup_subsys_state
*css
;
4822 inode
= f
->f_dentry
->d_inode
;
4823 /* check in cgroup filesystem dir */
4824 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
4825 return ERR_PTR(-EBADF
);
4827 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
4828 return ERR_PTR(-EINVAL
);
4831 cgrp
= __d_cgrp(f
->f_dentry
);
4832 css
= cgrp
->subsys
[id
];
4833 return css
? css
: ERR_PTR(-ENOENT
);
4836 #ifdef CONFIG_CGROUP_DEBUG
4837 static struct cgroup_subsys_state
*debug_create(struct cgroup_subsys
*ss
,
4838 struct cgroup
*cont
)
4840 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
4843 return ERR_PTR(-ENOMEM
);
4848 static void debug_destroy(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4850 kfree(cont
->subsys
[debug_subsys_id
]);
4853 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
4855 return atomic_read(&cont
->count
);
4858 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
4860 return cgroup_task_count(cont
);
4863 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
4865 return (u64
)(unsigned long)current
->cgroups
;
4868 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
4874 count
= atomic_read(¤t
->cgroups
->refcount
);
4879 static int current_css_set_cg_links_read(struct cgroup
*cont
,
4881 struct seq_file
*seq
)
4883 struct cg_cgroup_link
*link
;
4886 read_lock(&css_set_lock
);
4888 cg
= rcu_dereference(current
->cgroups
);
4889 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
4890 struct cgroup
*c
= link
->cgrp
;
4894 name
= c
->dentry
->d_name
.name
;
4897 seq_printf(seq
, "Root %d group %s\n",
4898 c
->root
->hierarchy_id
, name
);
4901 read_unlock(&css_set_lock
);
4905 #define MAX_TASKS_SHOWN_PER_CSS 25
4906 static int cgroup_css_links_read(struct cgroup
*cont
,
4908 struct seq_file
*seq
)
4910 struct cg_cgroup_link
*link
;
4912 read_lock(&css_set_lock
);
4913 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
4914 struct css_set
*cg
= link
->cg
;
4915 struct task_struct
*task
;
4917 seq_printf(seq
, "css_set %p\n", cg
);
4918 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
4919 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
4920 seq_puts(seq
, " ...\n");
4923 seq_printf(seq
, " task %d\n",
4924 task_pid_vnr(task
));
4928 read_unlock(&css_set_lock
);
4932 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4934 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4937 static struct cftype debug_files
[] = {
4939 .name
= "cgroup_refcount",
4940 .read_u64
= cgroup_refcount_read
,
4943 .name
= "taskcount",
4944 .read_u64
= debug_taskcount_read
,
4948 .name
= "current_css_set",
4949 .read_u64
= current_css_set_read
,
4953 .name
= "current_css_set_refcount",
4954 .read_u64
= current_css_set_refcount_read
,
4958 .name
= "current_css_set_cg_links",
4959 .read_seq_string
= current_css_set_cg_links_read
,
4963 .name
= "cgroup_css_links",
4964 .read_seq_string
= cgroup_css_links_read
,
4968 .name
= "releasable",
4969 .read_u64
= releasable_read
,
4973 static int debug_populate(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4975 return cgroup_add_files(cont
, ss
, debug_files
,
4976 ARRAY_SIZE(debug_files
));
4979 struct cgroup_subsys debug_subsys
= {
4981 .create
= debug_create
,
4982 .destroy
= debug_destroy
,
4983 .populate
= debug_populate
,
4984 .subsys_id
= debug_subsys_id
,
4986 #endif /* CONFIG_CGROUP_DEBUG */