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 recieve.
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 static void free_css_set_rcu(struct rcu_head
*obj
)
331 struct css_set
*cg
= container_of(obj
, struct css_set
, rcu_head
);
335 /* We don't maintain the lists running through each css_set to its
336 * task until after the first call to cgroup_iter_start(). This
337 * reduces the fork()/exit() overhead for people who have cgroups
338 * compiled into their kernel but not actually in use */
339 static int use_task_css_set_links __read_mostly
;
341 static void __put_css_set(struct css_set
*cg
, int taskexit
)
343 struct cg_cgroup_link
*link
;
344 struct cg_cgroup_link
*saved_link
;
346 * Ensure that the refcount doesn't hit zero while any readers
347 * can see it. Similar to atomic_dec_and_lock(), but for an
350 if (atomic_add_unless(&cg
->refcount
, -1, 1))
352 write_lock(&css_set_lock
);
353 if (!atomic_dec_and_test(&cg
->refcount
)) {
354 write_unlock(&css_set_lock
);
358 /* This css_set is dead. unlink it and release cgroup refcounts */
359 hlist_del(&cg
->hlist
);
362 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
364 struct cgroup
*cgrp
= link
->cgrp
;
365 list_del(&link
->cg_link_list
);
366 list_del(&link
->cgrp_link_list
);
367 if (atomic_dec_and_test(&cgrp
->count
) &&
368 notify_on_release(cgrp
)) {
370 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
371 check_for_release(cgrp
);
377 write_unlock(&css_set_lock
);
378 call_rcu(&cg
->rcu_head
, free_css_set_rcu
);
382 * refcounted get/put for css_set objects
384 static inline void get_css_set(struct css_set
*cg
)
386 atomic_inc(&cg
->refcount
);
389 static inline void put_css_set(struct css_set
*cg
)
391 __put_css_set(cg
, 0);
394 static inline void put_css_set_taskexit(struct css_set
*cg
)
396 __put_css_set(cg
, 1);
400 * compare_css_sets - helper function for find_existing_css_set().
401 * @cg: candidate css_set being tested
402 * @old_cg: existing css_set for a task
403 * @new_cgrp: cgroup that's being entered by the task
404 * @template: desired set of css pointers in css_set (pre-calculated)
406 * Returns true if "cg" matches "old_cg" except for the hierarchy
407 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
409 static bool compare_css_sets(struct css_set
*cg
,
410 struct css_set
*old_cg
,
411 struct cgroup
*new_cgrp
,
412 struct cgroup_subsys_state
*template[])
414 struct list_head
*l1
, *l2
;
416 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
417 /* Not all subsystems matched */
422 * Compare cgroup pointers in order to distinguish between
423 * different cgroups in heirarchies with no subsystems. We
424 * could get by with just this check alone (and skip the
425 * memcmp above) but on most setups the memcmp check will
426 * avoid the need for this more expensive check on almost all
431 l2
= &old_cg
->cg_links
;
433 struct cg_cgroup_link
*cgl1
, *cgl2
;
434 struct cgroup
*cg1
, *cg2
;
438 /* See if we reached the end - both lists are equal length. */
439 if (l1
== &cg
->cg_links
) {
440 BUG_ON(l2
!= &old_cg
->cg_links
);
443 BUG_ON(l2
== &old_cg
->cg_links
);
445 /* Locate the cgroups associated with these links. */
446 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
447 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
450 /* Hierarchies should be linked in the same order. */
451 BUG_ON(cg1
->root
!= cg2
->root
);
454 * If this hierarchy is the hierarchy of the cgroup
455 * that's changing, then we need to check that this
456 * css_set points to the new cgroup; if it's any other
457 * hierarchy, then this css_set should point to the
458 * same cgroup as the old css_set.
460 if (cg1
->root
== new_cgrp
->root
) {
472 * find_existing_css_set() is a helper for
473 * find_css_set(), and checks to see whether an existing
474 * css_set is suitable.
476 * oldcg: the cgroup group that we're using before the cgroup
479 * cgrp: the cgroup that we're moving into
481 * template: location in which to build the desired set of subsystem
482 * state objects for the new cgroup group
484 static struct css_set
*find_existing_css_set(
485 struct css_set
*oldcg
,
487 struct cgroup_subsys_state
*template[])
490 struct cgroupfs_root
*root
= cgrp
->root
;
491 struct hlist_head
*hhead
;
492 struct hlist_node
*node
;
496 * Build the set of subsystem state objects that we want to see in the
497 * new css_set. while subsystems can change globally, the entries here
498 * won't change, so no need for locking.
500 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
501 if (root
->subsys_bits
& (1UL << i
)) {
502 /* Subsystem is in this hierarchy. So we want
503 * the subsystem state from the new
505 template[i
] = cgrp
->subsys
[i
];
507 /* Subsystem is not in this hierarchy, so we
508 * don't want to change the subsystem state */
509 template[i
] = oldcg
->subsys
[i
];
513 hhead
= css_set_hash(template);
514 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
515 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
518 /* This css_set matches what we need */
522 /* No existing cgroup group matched */
526 static void free_cg_links(struct list_head
*tmp
)
528 struct cg_cgroup_link
*link
;
529 struct cg_cgroup_link
*saved_link
;
531 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
532 list_del(&link
->cgrp_link_list
);
538 * allocate_cg_links() allocates "count" cg_cgroup_link structures
539 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
540 * success or a negative error
542 static int allocate_cg_links(int count
, struct list_head
*tmp
)
544 struct cg_cgroup_link
*link
;
547 for (i
= 0; i
< count
; i
++) {
548 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
553 list_add(&link
->cgrp_link_list
, tmp
);
559 * link_css_set - a helper function to link a css_set to a cgroup
560 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
561 * @cg: the css_set to be linked
562 * @cgrp: the destination cgroup
564 static void link_css_set(struct list_head
*tmp_cg_links
,
565 struct css_set
*cg
, struct cgroup
*cgrp
)
567 struct cg_cgroup_link
*link
;
569 BUG_ON(list_empty(tmp_cg_links
));
570 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
574 atomic_inc(&cgrp
->count
);
575 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
577 * Always add links to the tail of the list so that the list
578 * is sorted by order of hierarchy creation
580 list_add_tail(&link
->cg_link_list
, &cg
->cg_links
);
584 * find_css_set() takes an existing cgroup group and a
585 * cgroup object, and returns a css_set object that's
586 * equivalent to the old group, but with the given cgroup
587 * substituted into the appropriate hierarchy. Must be called with
590 static struct css_set
*find_css_set(
591 struct css_set
*oldcg
, struct cgroup
*cgrp
)
594 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
596 struct list_head tmp_cg_links
;
598 struct hlist_head
*hhead
;
599 struct cg_cgroup_link
*link
;
601 /* First see if we already have a cgroup group that matches
603 read_lock(&css_set_lock
);
604 res
= find_existing_css_set(oldcg
, cgrp
, template);
607 read_unlock(&css_set_lock
);
612 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
616 /* Allocate all the cg_cgroup_link objects that we'll need */
617 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
622 atomic_set(&res
->refcount
, 1);
623 INIT_LIST_HEAD(&res
->cg_links
);
624 INIT_LIST_HEAD(&res
->tasks
);
625 INIT_HLIST_NODE(&res
->hlist
);
627 /* Copy the set of subsystem state objects generated in
628 * find_existing_css_set() */
629 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
631 write_lock(&css_set_lock
);
632 /* Add reference counts and links from the new css_set. */
633 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
634 struct cgroup
*c
= link
->cgrp
;
635 if (c
->root
== cgrp
->root
)
637 link_css_set(&tmp_cg_links
, res
, c
);
640 BUG_ON(!list_empty(&tmp_cg_links
));
644 /* Add this cgroup group to the hash table */
645 hhead
= css_set_hash(res
->subsys
);
646 hlist_add_head(&res
->hlist
, hhead
);
648 write_unlock(&css_set_lock
);
654 * Return the cgroup for "task" from the given hierarchy. Must be
655 * called with cgroup_mutex held.
657 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
658 struct cgroupfs_root
*root
)
661 struct cgroup
*res
= NULL
;
663 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
664 read_lock(&css_set_lock
);
666 * No need to lock the task - since we hold cgroup_mutex the
667 * task can't change groups, so the only thing that can happen
668 * is that it exits and its css is set back to init_css_set.
671 if (css
== &init_css_set
) {
672 res
= &root
->top_cgroup
;
674 struct cg_cgroup_link
*link
;
675 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
676 struct cgroup
*c
= link
->cgrp
;
677 if (c
->root
== root
) {
683 read_unlock(&css_set_lock
);
689 * There is one global cgroup mutex. We also require taking
690 * task_lock() when dereferencing a task's cgroup subsys pointers.
691 * See "The task_lock() exception", at the end of this comment.
693 * A task must hold cgroup_mutex to modify cgroups.
695 * Any task can increment and decrement the count field without lock.
696 * So in general, code holding cgroup_mutex can't rely on the count
697 * field not changing. However, if the count goes to zero, then only
698 * cgroup_attach_task() can increment it again. Because a count of zero
699 * means that no tasks are currently attached, therefore there is no
700 * way a task attached to that cgroup can fork (the other way to
701 * increment the count). So code holding cgroup_mutex can safely
702 * assume that if the count is zero, it will stay zero. Similarly, if
703 * a task holds cgroup_mutex on a cgroup with zero count, it
704 * knows that the cgroup won't be removed, as cgroup_rmdir()
707 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
708 * (usually) take cgroup_mutex. These are the two most performance
709 * critical pieces of code here. The exception occurs on cgroup_exit(),
710 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
711 * is taken, and if the cgroup count is zero, a usermode call made
712 * to the release agent with the name of the cgroup (path relative to
713 * the root of cgroup file system) as the argument.
715 * A cgroup can only be deleted if both its 'count' of using tasks
716 * is zero, and its list of 'children' cgroups is empty. Since all
717 * tasks in the system use _some_ cgroup, and since there is always at
718 * least one task in the system (init, pid == 1), therefore, top_cgroup
719 * always has either children cgroups and/or using tasks. So we don't
720 * need a special hack to ensure that top_cgroup cannot be deleted.
722 * The task_lock() exception
724 * The need for this exception arises from the action of
725 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
726 * another. It does so using cgroup_mutex, however there are
727 * several performance critical places that need to reference
728 * task->cgroup without the expense of grabbing a system global
729 * mutex. Therefore except as noted below, when dereferencing or, as
730 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
731 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
732 * the task_struct routinely used for such matters.
734 * P.S. One more locking exception. RCU is used to guard the
735 * update of a tasks cgroup pointer by cgroup_attach_task()
739 * cgroup_lock - lock out any changes to cgroup structures
742 void cgroup_lock(void)
744 mutex_lock(&cgroup_mutex
);
746 EXPORT_SYMBOL_GPL(cgroup_lock
);
749 * cgroup_unlock - release lock on cgroup changes
751 * Undo the lock taken in a previous cgroup_lock() call.
753 void cgroup_unlock(void)
755 mutex_unlock(&cgroup_mutex
);
757 EXPORT_SYMBOL_GPL(cgroup_unlock
);
760 * A couple of forward declarations required, due to cyclic reference loop:
761 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
762 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
766 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
);
767 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
768 static int cgroup_populate_dir(struct cgroup
*cgrp
);
769 static const struct inode_operations cgroup_dir_inode_operations
;
770 static const struct file_operations proc_cgroupstats_operations
;
772 static struct backing_dev_info cgroup_backing_dev_info
= {
774 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
777 static int alloc_css_id(struct cgroup_subsys
*ss
,
778 struct cgroup
*parent
, struct cgroup
*child
);
780 static struct inode
*cgroup_new_inode(mode_t mode
, struct super_block
*sb
)
782 struct inode
*inode
= new_inode(sb
);
785 inode
->i_ino
= get_next_ino();
786 inode
->i_mode
= mode
;
787 inode
->i_uid
= current_fsuid();
788 inode
->i_gid
= current_fsgid();
789 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
790 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
796 * Call subsys's pre_destroy handler.
797 * This is called before css refcnt check.
799 static int cgroup_call_pre_destroy(struct cgroup
*cgrp
)
801 struct cgroup_subsys
*ss
;
804 for_each_subsys(cgrp
->root
, ss
)
805 if (ss
->pre_destroy
) {
806 ret
= ss
->pre_destroy(ss
, cgrp
);
814 static void free_cgroup_rcu(struct rcu_head
*obj
)
816 struct cgroup
*cgrp
= container_of(obj
, struct cgroup
, rcu_head
);
821 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
823 /* is dentry a directory ? if so, kfree() associated cgroup */
824 if (S_ISDIR(inode
->i_mode
)) {
825 struct cgroup
*cgrp
= dentry
->d_fsdata
;
826 struct cgroup_subsys
*ss
;
827 BUG_ON(!(cgroup_is_removed(cgrp
)));
828 /* It's possible for external users to be holding css
829 * reference counts on a cgroup; css_put() needs to
830 * be able to access the cgroup after decrementing
831 * the reference count in order to know if it needs to
832 * queue the cgroup to be handled by the release
836 mutex_lock(&cgroup_mutex
);
838 * Release the subsystem state objects.
840 for_each_subsys(cgrp
->root
, ss
)
841 ss
->destroy(ss
, cgrp
);
843 cgrp
->root
->number_of_cgroups
--;
844 mutex_unlock(&cgroup_mutex
);
847 * Drop the active superblock reference that we took when we
850 deactivate_super(cgrp
->root
->sb
);
853 * if we're getting rid of the cgroup, refcount should ensure
854 * that there are no pidlists left.
856 BUG_ON(!list_empty(&cgrp
->pidlists
));
858 call_rcu(&cgrp
->rcu_head
, free_cgroup_rcu
);
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(&dcache_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
);
883 /* This should never be called on a cgroup
884 * directory with child cgroups */
885 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
887 spin_unlock(&dcache_lock
);
889 simple_unlink(dentry
->d_inode
, d
);
891 spin_lock(&dcache_lock
);
893 node
= dentry
->d_subdirs
.next
;
895 spin_unlock(&dcache_lock
);
899 * NOTE : the dentry must have been dget()'ed
901 static void cgroup_d_remove_dir(struct dentry
*dentry
)
903 cgroup_clear_directory(dentry
);
905 spin_lock(&dcache_lock
);
906 list_del_init(&dentry
->d_u
.d_child
);
907 spin_unlock(&dcache_lock
);
912 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
913 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
914 * reference to css->refcnt. In general, this refcnt is expected to goes down
917 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
919 DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq
);
921 static void cgroup_wakeup_rmdir_waiter(struct cgroup
*cgrp
)
923 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
)))
924 wake_up_all(&cgroup_rmdir_waitq
);
927 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
)
932 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
)
934 cgroup_wakeup_rmdir_waiter(css
->cgroup
);
939 * Call with cgroup_mutex held. Drops reference counts on modules, including
940 * any duplicate ones that parse_cgroupfs_options took. If this function
941 * returns an error, no reference counts are touched.
943 static int rebind_subsystems(struct cgroupfs_root
*root
,
944 unsigned long final_bits
)
946 unsigned long added_bits
, removed_bits
;
947 struct cgroup
*cgrp
= &root
->top_cgroup
;
950 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
952 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
953 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
954 /* Check that any added subsystems are currently free */
955 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
956 unsigned long bit
= 1UL << i
;
957 struct cgroup_subsys
*ss
= subsys
[i
];
958 if (!(bit
& added_bits
))
961 * Nobody should tell us to do a subsys that doesn't exist:
962 * parse_cgroupfs_options should catch that case and refcounts
963 * ensure that subsystems won't disappear once selected.
966 if (ss
->root
!= &rootnode
) {
967 /* Subsystem isn't free */
972 /* Currently we don't handle adding/removing subsystems when
973 * any child cgroups exist. This is theoretically supportable
974 * but involves complex error handling, so it's being left until
976 if (root
->number_of_cgroups
> 1)
979 /* Process each subsystem */
980 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
981 struct cgroup_subsys
*ss
= subsys
[i
];
982 unsigned long bit
= 1UL << i
;
983 if (bit
& added_bits
) {
984 /* We're binding this subsystem to this hierarchy */
986 BUG_ON(cgrp
->subsys
[i
]);
987 BUG_ON(!dummytop
->subsys
[i
]);
988 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
989 mutex_lock(&ss
->hierarchy_mutex
);
990 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
991 cgrp
->subsys
[i
]->cgroup
= cgrp
;
992 list_move(&ss
->sibling
, &root
->subsys_list
);
996 mutex_unlock(&ss
->hierarchy_mutex
);
997 /* refcount was already taken, and we're keeping it */
998 } else if (bit
& removed_bits
) {
999 /* We're removing this subsystem */
1001 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1002 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1003 mutex_lock(&ss
->hierarchy_mutex
);
1005 ss
->bind(ss
, dummytop
);
1006 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1007 cgrp
->subsys
[i
] = NULL
;
1008 subsys
[i
]->root
= &rootnode
;
1009 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1010 mutex_unlock(&ss
->hierarchy_mutex
);
1011 /* subsystem is now free - drop reference on module */
1012 module_put(ss
->module
);
1013 } else if (bit
& final_bits
) {
1014 /* Subsystem state should already exist */
1016 BUG_ON(!cgrp
->subsys
[i
]);
1018 * a refcount was taken, but we already had one, so
1019 * drop the extra reference.
1021 module_put(ss
->module
);
1022 #ifdef CONFIG_MODULE_UNLOAD
1023 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1026 /* Subsystem state shouldn't exist */
1027 BUG_ON(cgrp
->subsys
[i
]);
1030 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
1036 static int cgroup_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
1038 struct cgroupfs_root
*root
= vfs
->mnt_sb
->s_fs_info
;
1039 struct cgroup_subsys
*ss
;
1041 mutex_lock(&cgroup_mutex
);
1042 for_each_subsys(root
, ss
)
1043 seq_printf(seq
, ",%s", ss
->name
);
1044 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1045 seq_puts(seq
, ",noprefix");
1046 if (strlen(root
->release_agent_path
))
1047 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1048 if (clone_children(&root
->top_cgroup
))
1049 seq_puts(seq
, ",clone_children");
1050 if (strlen(root
->name
))
1051 seq_printf(seq
, ",name=%s", root
->name
);
1052 mutex_unlock(&cgroup_mutex
);
1056 struct cgroup_sb_opts
{
1057 unsigned long subsys_bits
;
1058 unsigned long flags
;
1059 char *release_agent
;
1060 bool clone_children
;
1062 /* User explicitly requested empty subsystem */
1065 struct cgroupfs_root
*new_root
;
1070 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1071 * with cgroup_mutex held to protect the subsys[] array. This function takes
1072 * refcounts on subsystems to be used, unless it returns error, in which case
1073 * no refcounts are taken.
1075 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1077 char *token
, *o
= data
;
1078 bool all_ss
= false, one_ss
= false;
1079 unsigned long mask
= (unsigned long)-1;
1081 bool module_pin_failed
= false;
1083 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1085 #ifdef CONFIG_CPUSETS
1086 mask
= ~(1UL << cpuset_subsys_id
);
1089 memset(opts
, 0, sizeof(*opts
));
1091 while ((token
= strsep(&o
, ",")) != NULL
) {
1094 if (!strcmp(token
, "none")) {
1095 /* Explicitly have no subsystems */
1099 if (!strcmp(token
, "all")) {
1100 /* Mutually exclusive option 'all' + subsystem name */
1106 if (!strcmp(token
, "noprefix")) {
1107 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1110 if (!strcmp(token
, "clone_children")) {
1111 opts
->clone_children
= true;
1114 if (!strncmp(token
, "release_agent=", 14)) {
1115 /* Specifying two release agents is forbidden */
1116 if (opts
->release_agent
)
1118 opts
->release_agent
=
1119 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1120 if (!opts
->release_agent
)
1124 if (!strncmp(token
, "name=", 5)) {
1125 const char *name
= token
+ 5;
1126 /* Can't specify an empty name */
1129 /* Must match [\w.-]+ */
1130 for (i
= 0; i
< strlen(name
); i
++) {
1134 if ((c
== '.') || (c
== '-') || (c
== '_'))
1138 /* Specifying two names is forbidden */
1141 opts
->name
= kstrndup(name
,
1142 MAX_CGROUP_ROOT_NAMELEN
- 1,
1150 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1151 struct cgroup_subsys
*ss
= subsys
[i
];
1154 if (strcmp(token
, ss
->name
))
1159 /* Mutually exclusive option 'all' + subsystem name */
1162 set_bit(i
, &opts
->subsys_bits
);
1167 if (i
== CGROUP_SUBSYS_COUNT
)
1172 * If the 'all' option was specified select all the subsystems,
1173 * otherwise 'all, 'none' and a subsystem name options were not
1174 * specified, let's default to 'all'
1176 if (all_ss
|| (!all_ss
&& !one_ss
&& !opts
->none
)) {
1177 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1178 struct cgroup_subsys
*ss
= subsys
[i
];
1183 set_bit(i
, &opts
->subsys_bits
);
1187 /* Consistency checks */
1190 * Option noprefix was introduced just for backward compatibility
1191 * with the old cpuset, so we allow noprefix only if mounting just
1192 * the cpuset subsystem.
1194 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1195 (opts
->subsys_bits
& mask
))
1199 /* Can't specify "none" and some subsystems */
1200 if (opts
->subsys_bits
&& opts
->none
)
1204 * We either have to specify by name or by subsystems. (So all
1205 * empty hierarchies must have a name).
1207 if (!opts
->subsys_bits
&& !opts
->name
)
1211 * Grab references on all the modules we'll need, so the subsystems
1212 * don't dance around before rebind_subsystems attaches them. This may
1213 * take duplicate reference counts on a subsystem that's already used,
1214 * but rebind_subsystems handles this case.
1216 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1217 unsigned long bit
= 1UL << i
;
1219 if (!(bit
& opts
->subsys_bits
))
1221 if (!try_module_get(subsys
[i
]->module
)) {
1222 module_pin_failed
= true;
1226 if (module_pin_failed
) {
1228 * oops, one of the modules was going away. this means that we
1229 * raced with a module_delete call, and to the user this is
1230 * essentially a "subsystem doesn't exist" case.
1232 for (i
--; i
>= CGROUP_BUILTIN_SUBSYS_COUNT
; i
--) {
1233 /* drop refcounts only on the ones we took */
1234 unsigned long bit
= 1UL << i
;
1236 if (!(bit
& opts
->subsys_bits
))
1238 module_put(subsys
[i
]->module
);
1246 static void drop_parsed_module_refcounts(unsigned long subsys_bits
)
1249 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1250 unsigned long bit
= 1UL << i
;
1252 if (!(bit
& subsys_bits
))
1254 module_put(subsys
[i
]->module
);
1258 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1261 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1262 struct cgroup
*cgrp
= &root
->top_cgroup
;
1263 struct cgroup_sb_opts opts
;
1265 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1266 mutex_lock(&cgroup_mutex
);
1268 /* See what subsystems are wanted */
1269 ret
= parse_cgroupfs_options(data
, &opts
);
1273 /* Don't allow flags or name to change at remount */
1274 if (opts
.flags
!= root
->flags
||
1275 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1277 drop_parsed_module_refcounts(opts
.subsys_bits
);
1281 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
1283 drop_parsed_module_refcounts(opts
.subsys_bits
);
1287 /* (re)populate subsystem files */
1288 cgroup_populate_dir(cgrp
);
1290 if (opts
.release_agent
)
1291 strcpy(root
->release_agent_path
, opts
.release_agent
);
1293 kfree(opts
.release_agent
);
1295 mutex_unlock(&cgroup_mutex
);
1296 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1300 static const struct super_operations cgroup_ops
= {
1301 .statfs
= simple_statfs
,
1302 .drop_inode
= generic_delete_inode
,
1303 .show_options
= cgroup_show_options
,
1304 .remount_fs
= cgroup_remount
,
1307 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1309 INIT_LIST_HEAD(&cgrp
->sibling
);
1310 INIT_LIST_HEAD(&cgrp
->children
);
1311 INIT_LIST_HEAD(&cgrp
->css_sets
);
1312 INIT_LIST_HEAD(&cgrp
->release_list
);
1313 INIT_LIST_HEAD(&cgrp
->pidlists
);
1314 mutex_init(&cgrp
->pidlist_mutex
);
1315 INIT_LIST_HEAD(&cgrp
->event_list
);
1316 spin_lock_init(&cgrp
->event_list_lock
);
1319 static void init_cgroup_root(struct cgroupfs_root
*root
)
1321 struct cgroup
*cgrp
= &root
->top_cgroup
;
1322 INIT_LIST_HEAD(&root
->subsys_list
);
1323 INIT_LIST_HEAD(&root
->root_list
);
1324 root
->number_of_cgroups
= 1;
1326 cgrp
->top_cgroup
= cgrp
;
1327 init_cgroup_housekeeping(cgrp
);
1330 static bool init_root_id(struct cgroupfs_root
*root
)
1335 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1337 spin_lock(&hierarchy_id_lock
);
1338 /* Try to allocate the next unused ID */
1339 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1340 &root
->hierarchy_id
);
1342 /* Try again starting from 0 */
1343 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1345 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1346 } else if (ret
!= -EAGAIN
) {
1347 /* Can only get here if the 31-bit IDR is full ... */
1350 spin_unlock(&hierarchy_id_lock
);
1355 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1357 struct cgroup_sb_opts
*opts
= data
;
1358 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1360 /* If we asked for a name then it must match */
1361 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1365 * If we asked for subsystems (or explicitly for no
1366 * subsystems) then they must match
1368 if ((opts
->subsys_bits
|| opts
->none
)
1369 && (opts
->subsys_bits
!= root
->subsys_bits
))
1375 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1377 struct cgroupfs_root
*root
;
1379 if (!opts
->subsys_bits
&& !opts
->none
)
1382 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1384 return ERR_PTR(-ENOMEM
);
1386 if (!init_root_id(root
)) {
1388 return ERR_PTR(-ENOMEM
);
1390 init_cgroup_root(root
);
1392 root
->subsys_bits
= opts
->subsys_bits
;
1393 root
->flags
= opts
->flags
;
1394 if (opts
->release_agent
)
1395 strcpy(root
->release_agent_path
, opts
->release_agent
);
1397 strcpy(root
->name
, opts
->name
);
1398 if (opts
->clone_children
)
1399 set_bit(CGRP_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1403 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1408 BUG_ON(!root
->hierarchy_id
);
1409 spin_lock(&hierarchy_id_lock
);
1410 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1411 spin_unlock(&hierarchy_id_lock
);
1415 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1418 struct cgroup_sb_opts
*opts
= data
;
1420 /* If we don't have a new root, we can't set up a new sb */
1421 if (!opts
->new_root
)
1424 BUG_ON(!opts
->subsys_bits
&& !opts
->none
);
1426 ret
= set_anon_super(sb
, NULL
);
1430 sb
->s_fs_info
= opts
->new_root
;
1431 opts
->new_root
->sb
= sb
;
1433 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1434 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1435 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1436 sb
->s_op
= &cgroup_ops
;
1441 static int cgroup_get_rootdir(struct super_block
*sb
)
1443 struct inode
*inode
=
1444 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1445 struct dentry
*dentry
;
1450 inode
->i_fop
= &simple_dir_operations
;
1451 inode
->i_op
= &cgroup_dir_inode_operations
;
1452 /* directories start off with i_nlink == 2 (for "." entry) */
1454 dentry
= d_alloc_root(inode
);
1459 sb
->s_root
= dentry
;
1463 static int cgroup_get_sb(struct file_system_type
*fs_type
,
1464 int flags
, const char *unused_dev_name
,
1465 void *data
, struct vfsmount
*mnt
)
1467 struct cgroup_sb_opts opts
;
1468 struct cgroupfs_root
*root
;
1470 struct super_block
*sb
;
1471 struct cgroupfs_root
*new_root
;
1473 /* First find the desired set of subsystems */
1474 mutex_lock(&cgroup_mutex
);
1475 ret
= parse_cgroupfs_options(data
, &opts
);
1476 mutex_unlock(&cgroup_mutex
);
1481 * Allocate a new cgroup root. We may not need it if we're
1482 * reusing an existing hierarchy.
1484 new_root
= cgroup_root_from_opts(&opts
);
1485 if (IS_ERR(new_root
)) {
1486 ret
= PTR_ERR(new_root
);
1489 opts
.new_root
= new_root
;
1491 /* Locate an existing or new sb for this hierarchy */
1492 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, &opts
);
1495 cgroup_drop_root(opts
.new_root
);
1499 root
= sb
->s_fs_info
;
1501 if (root
== opts
.new_root
) {
1502 /* We used the new root structure, so this is a new hierarchy */
1503 struct list_head tmp_cg_links
;
1504 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1505 struct inode
*inode
;
1506 struct cgroupfs_root
*existing_root
;
1509 BUG_ON(sb
->s_root
!= NULL
);
1511 ret
= cgroup_get_rootdir(sb
);
1513 goto drop_new_super
;
1514 inode
= sb
->s_root
->d_inode
;
1516 mutex_lock(&inode
->i_mutex
);
1517 mutex_lock(&cgroup_mutex
);
1519 if (strlen(root
->name
)) {
1520 /* Check for name clashes with existing mounts */
1521 for_each_active_root(existing_root
) {
1522 if (!strcmp(existing_root
->name
, root
->name
)) {
1524 mutex_unlock(&cgroup_mutex
);
1525 mutex_unlock(&inode
->i_mutex
);
1526 goto drop_new_super
;
1532 * We're accessing css_set_count without locking
1533 * css_set_lock here, but that's OK - it can only be
1534 * increased by someone holding cgroup_lock, and
1535 * that's us. The worst that can happen is that we
1536 * have some link structures left over
1538 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1540 mutex_unlock(&cgroup_mutex
);
1541 mutex_unlock(&inode
->i_mutex
);
1542 goto drop_new_super
;
1545 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1546 if (ret
== -EBUSY
) {
1547 mutex_unlock(&cgroup_mutex
);
1548 mutex_unlock(&inode
->i_mutex
);
1549 free_cg_links(&tmp_cg_links
);
1550 goto drop_new_super
;
1553 * There must be no failure case after here, since rebinding
1554 * takes care of subsystems' refcounts, which are explicitly
1555 * dropped in the failure exit path.
1558 /* EBUSY should be the only error here */
1561 list_add(&root
->root_list
, &roots
);
1564 sb
->s_root
->d_fsdata
= root_cgrp
;
1565 root
->top_cgroup
.dentry
= sb
->s_root
;
1567 /* Link the top cgroup in this hierarchy into all
1568 * the css_set objects */
1569 write_lock(&css_set_lock
);
1570 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1571 struct hlist_head
*hhead
= &css_set_table
[i
];
1572 struct hlist_node
*node
;
1575 hlist_for_each_entry(cg
, node
, hhead
, hlist
)
1576 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1578 write_unlock(&css_set_lock
);
1580 free_cg_links(&tmp_cg_links
);
1582 BUG_ON(!list_empty(&root_cgrp
->sibling
));
1583 BUG_ON(!list_empty(&root_cgrp
->children
));
1584 BUG_ON(root
->number_of_cgroups
!= 1);
1586 cgroup_populate_dir(root_cgrp
);
1587 mutex_unlock(&cgroup_mutex
);
1588 mutex_unlock(&inode
->i_mutex
);
1591 * We re-used an existing hierarchy - the new root (if
1592 * any) is not needed
1594 cgroup_drop_root(opts
.new_root
);
1595 /* no subsys rebinding, so refcounts don't change */
1596 drop_parsed_module_refcounts(opts
.subsys_bits
);
1599 simple_set_mnt(mnt
, sb
);
1600 kfree(opts
.release_agent
);
1605 deactivate_locked_super(sb
);
1607 drop_parsed_module_refcounts(opts
.subsys_bits
);
1609 kfree(opts
.release_agent
);
1614 static void cgroup_kill_sb(struct super_block
*sb
) {
1615 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1616 struct cgroup
*cgrp
= &root
->top_cgroup
;
1618 struct cg_cgroup_link
*link
;
1619 struct cg_cgroup_link
*saved_link
;
1623 BUG_ON(root
->number_of_cgroups
!= 1);
1624 BUG_ON(!list_empty(&cgrp
->children
));
1625 BUG_ON(!list_empty(&cgrp
->sibling
));
1627 mutex_lock(&cgroup_mutex
);
1629 /* Rebind all subsystems back to the default hierarchy */
1630 ret
= rebind_subsystems(root
, 0);
1631 /* Shouldn't be able to fail ... */
1635 * Release all the links from css_sets to this hierarchy's
1638 write_lock(&css_set_lock
);
1640 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1642 list_del(&link
->cg_link_list
);
1643 list_del(&link
->cgrp_link_list
);
1646 write_unlock(&css_set_lock
);
1648 if (!list_empty(&root
->root_list
)) {
1649 list_del(&root
->root_list
);
1653 mutex_unlock(&cgroup_mutex
);
1655 kill_litter_super(sb
);
1656 cgroup_drop_root(root
);
1659 static struct file_system_type cgroup_fs_type
= {
1661 .get_sb
= cgroup_get_sb
,
1662 .kill_sb
= cgroup_kill_sb
,
1665 static struct kobject
*cgroup_kobj
;
1667 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1669 return dentry
->d_fsdata
;
1672 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1674 return dentry
->d_fsdata
;
1678 * cgroup_path - generate the path of a cgroup
1679 * @cgrp: the cgroup in question
1680 * @buf: the buffer to write the path into
1681 * @buflen: the length of the buffer
1683 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1684 * reference. Writes path of cgroup into buf. Returns 0 on success,
1687 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1690 struct dentry
*dentry
= rcu_dereference_check(cgrp
->dentry
,
1691 rcu_read_lock_held() ||
1692 cgroup_lock_is_held());
1694 if (!dentry
|| cgrp
== dummytop
) {
1696 * Inactive subsystems have no dentry for their root
1703 start
= buf
+ buflen
;
1707 int len
= dentry
->d_name
.len
;
1709 if ((start
-= len
) < buf
)
1710 return -ENAMETOOLONG
;
1711 memcpy(start
, dentry
->d_name
.name
, len
);
1712 cgrp
= cgrp
->parent
;
1716 dentry
= rcu_dereference_check(cgrp
->dentry
,
1717 rcu_read_lock_held() ||
1718 cgroup_lock_is_held());
1722 return -ENAMETOOLONG
;
1725 memmove(buf
, start
, buf
+ buflen
- start
);
1728 EXPORT_SYMBOL_GPL(cgroup_path
);
1731 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1732 * @cgrp: the cgroup the task is attaching to
1733 * @tsk: the task to be attached
1735 * Call holding cgroup_mutex. May take task_lock of
1736 * the task 'tsk' during call.
1738 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1741 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1742 struct cgroup
*oldcgrp
;
1744 struct css_set
*newcg
;
1745 struct cgroupfs_root
*root
= cgrp
->root
;
1747 /* Nothing to do if the task is already in that cgroup */
1748 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1749 if (cgrp
== oldcgrp
)
1752 for_each_subsys(root
, ss
) {
1753 if (ss
->can_attach
) {
1754 retval
= ss
->can_attach(ss
, cgrp
, tsk
, false);
1757 * Remember on which subsystem the can_attach()
1758 * failed, so that we only call cancel_attach()
1759 * against the subsystems whose can_attach()
1760 * succeeded. (See below)
1773 * Locate or allocate a new css_set for this task,
1774 * based on its final set of cgroups
1776 newcg
= find_css_set(cg
, cgrp
);
1784 if (tsk
->flags
& PF_EXITING
) {
1790 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1793 /* Update the css_set linked lists if we're using them */
1794 write_lock(&css_set_lock
);
1795 if (!list_empty(&tsk
->cg_list
)) {
1796 list_del(&tsk
->cg_list
);
1797 list_add(&tsk
->cg_list
, &newcg
->tasks
);
1799 write_unlock(&css_set_lock
);
1801 for_each_subsys(root
, ss
) {
1803 ss
->attach(ss
, cgrp
, oldcgrp
, tsk
, false);
1805 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1810 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1811 * is no longer empty.
1813 cgroup_wakeup_rmdir_waiter(cgrp
);
1816 for_each_subsys(root
, ss
) {
1817 if (ss
== failed_ss
)
1819 * This subsystem was the one that failed the
1820 * can_attach() check earlier, so we don't need
1821 * to call cancel_attach() against it or any
1822 * remaining subsystems.
1825 if (ss
->cancel_attach
)
1826 ss
->cancel_attach(ss
, cgrp
, tsk
, false);
1833 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1834 * @from: attach to all cgroups of a given task
1835 * @tsk: the task to be attached
1837 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
1839 struct cgroupfs_root
*root
;
1843 for_each_active_root(root
) {
1844 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
1846 retval
= cgroup_attach_task(from_cg
, tsk
);
1854 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
1857 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1858 * held. May take task_lock of task
1860 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
)
1862 struct task_struct
*tsk
;
1863 const struct cred
*cred
= current_cred(), *tcred
;
1868 tsk
= find_task_by_vpid(pid
);
1869 if (!tsk
|| tsk
->flags
& PF_EXITING
) {
1874 tcred
= __task_cred(tsk
);
1876 cred
->euid
!= tcred
->uid
&&
1877 cred
->euid
!= tcred
->suid
) {
1881 get_task_struct(tsk
);
1885 get_task_struct(tsk
);
1888 ret
= cgroup_attach_task(cgrp
, tsk
);
1889 put_task_struct(tsk
);
1893 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
1896 if (!cgroup_lock_live_group(cgrp
))
1898 ret
= attach_task_by_pid(cgrp
, pid
);
1904 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1905 * @cgrp: the cgroup to be checked for liveness
1907 * On success, returns true; the lock should be later released with
1908 * cgroup_unlock(). On failure returns false with no lock held.
1910 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
1912 mutex_lock(&cgroup_mutex
);
1913 if (cgroup_is_removed(cgrp
)) {
1914 mutex_unlock(&cgroup_mutex
);
1919 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
1921 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
1924 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
1925 if (!cgroup_lock_live_group(cgrp
))
1927 strcpy(cgrp
->root
->release_agent_path
, buffer
);
1932 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
1933 struct seq_file
*seq
)
1935 if (!cgroup_lock_live_group(cgrp
))
1937 seq_puts(seq
, cgrp
->root
->release_agent_path
);
1938 seq_putc(seq
, '\n');
1943 /* A buffer size big enough for numbers or short strings */
1944 #define CGROUP_LOCAL_BUFFER_SIZE 64
1946 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
1948 const char __user
*userbuf
,
1949 size_t nbytes
, loff_t
*unused_ppos
)
1951 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
1957 if (nbytes
>= sizeof(buffer
))
1959 if (copy_from_user(buffer
, userbuf
, nbytes
))
1962 buffer
[nbytes
] = 0; /* nul-terminate */
1963 if (cft
->write_u64
) {
1964 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
1967 retval
= cft
->write_u64(cgrp
, cft
, val
);
1969 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
1972 retval
= cft
->write_s64(cgrp
, cft
, val
);
1979 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
1981 const char __user
*userbuf
,
1982 size_t nbytes
, loff_t
*unused_ppos
)
1984 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
1986 size_t max_bytes
= cft
->max_write_len
;
1987 char *buffer
= local_buffer
;
1990 max_bytes
= sizeof(local_buffer
) - 1;
1991 if (nbytes
>= max_bytes
)
1993 /* Allocate a dynamic buffer if we need one */
1994 if (nbytes
>= sizeof(local_buffer
)) {
1995 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
1999 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2004 buffer
[nbytes
] = 0; /* nul-terminate */
2005 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2009 if (buffer
!= local_buffer
)
2014 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2015 size_t nbytes
, loff_t
*ppos
)
2017 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2018 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2020 if (cgroup_is_removed(cgrp
))
2023 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2024 if (cft
->write_u64
|| cft
->write_s64
)
2025 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2026 if (cft
->write_string
)
2027 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2029 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2030 return ret
? ret
: nbytes
;
2035 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2037 char __user
*buf
, size_t nbytes
,
2040 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2041 u64 val
= cft
->read_u64(cgrp
, cft
);
2042 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2044 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2047 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2049 char __user
*buf
, size_t nbytes
,
2052 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2053 s64 val
= cft
->read_s64(cgrp
, cft
);
2054 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2056 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2059 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2060 size_t nbytes
, loff_t
*ppos
)
2062 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2063 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2065 if (cgroup_is_removed(cgrp
))
2069 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2071 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2073 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2078 * seqfile ops/methods for returning structured data. Currently just
2079 * supports string->u64 maps, but can be extended in future.
2082 struct cgroup_seqfile_state
{
2084 struct cgroup
*cgroup
;
2087 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2089 struct seq_file
*sf
= cb
->state
;
2090 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2093 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2095 struct cgroup_seqfile_state
*state
= m
->private;
2096 struct cftype
*cft
= state
->cft
;
2097 if (cft
->read_map
) {
2098 struct cgroup_map_cb cb
= {
2099 .fill
= cgroup_map_add
,
2102 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2104 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2107 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2109 struct seq_file
*seq
= file
->private_data
;
2110 kfree(seq
->private);
2111 return single_release(inode
, file
);
2114 static const struct file_operations cgroup_seqfile_operations
= {
2116 .write
= cgroup_file_write
,
2117 .llseek
= seq_lseek
,
2118 .release
= cgroup_seqfile_release
,
2121 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2126 err
= generic_file_open(inode
, file
);
2129 cft
= __d_cft(file
->f_dentry
);
2131 if (cft
->read_map
|| cft
->read_seq_string
) {
2132 struct cgroup_seqfile_state
*state
=
2133 kzalloc(sizeof(*state
), GFP_USER
);
2137 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2138 file
->f_op
= &cgroup_seqfile_operations
;
2139 err
= single_open(file
, cgroup_seqfile_show
, state
);
2142 } else if (cft
->open
)
2143 err
= cft
->open(inode
, file
);
2150 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2152 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2154 return cft
->release(inode
, file
);
2159 * cgroup_rename - Only allow simple rename of directories in place.
2161 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2162 struct inode
*new_dir
, struct dentry
*new_dentry
)
2164 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2166 if (new_dentry
->d_inode
)
2168 if (old_dir
!= new_dir
)
2170 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2173 static const struct file_operations cgroup_file_operations
= {
2174 .read
= cgroup_file_read
,
2175 .write
= cgroup_file_write
,
2176 .llseek
= generic_file_llseek
,
2177 .open
= cgroup_file_open
,
2178 .release
= cgroup_file_release
,
2181 static const struct inode_operations cgroup_dir_inode_operations
= {
2182 .lookup
= simple_lookup
,
2183 .mkdir
= cgroup_mkdir
,
2184 .rmdir
= cgroup_rmdir
,
2185 .rename
= cgroup_rename
,
2189 * Check if a file is a control file
2191 static inline struct cftype
*__file_cft(struct file
*file
)
2193 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2194 return ERR_PTR(-EINVAL
);
2195 return __d_cft(file
->f_dentry
);
2198 static int cgroup_create_file(struct dentry
*dentry
, mode_t mode
,
2199 struct super_block
*sb
)
2201 static const struct dentry_operations cgroup_dops
= {
2202 .d_iput
= cgroup_diput
,
2205 struct inode
*inode
;
2209 if (dentry
->d_inode
)
2212 inode
= cgroup_new_inode(mode
, sb
);
2216 if (S_ISDIR(mode
)) {
2217 inode
->i_op
= &cgroup_dir_inode_operations
;
2218 inode
->i_fop
= &simple_dir_operations
;
2220 /* start off with i_nlink == 2 (for "." entry) */
2223 /* start with the directory inode held, so that we can
2224 * populate it without racing with another mkdir */
2225 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2226 } else if (S_ISREG(mode
)) {
2228 inode
->i_fop
= &cgroup_file_operations
;
2230 dentry
->d_op
= &cgroup_dops
;
2231 d_instantiate(dentry
, inode
);
2232 dget(dentry
); /* Extra count - pin the dentry in core */
2237 * cgroup_create_dir - create a directory for an object.
2238 * @cgrp: the cgroup we create the directory for. It must have a valid
2239 * ->parent field. And we are going to fill its ->dentry field.
2240 * @dentry: dentry of the new cgroup
2241 * @mode: mode to set on new directory.
2243 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
2246 struct dentry
*parent
;
2249 parent
= cgrp
->parent
->dentry
;
2250 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
2252 dentry
->d_fsdata
= cgrp
;
2253 inc_nlink(parent
->d_inode
);
2254 rcu_assign_pointer(cgrp
->dentry
, dentry
);
2263 * cgroup_file_mode - deduce file mode of a control file
2264 * @cft: the control file in question
2266 * returns cft->mode if ->mode is not 0
2267 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2268 * returns S_IRUGO if it has only a read handler
2269 * returns S_IWUSR if it has only a write hander
2271 static mode_t
cgroup_file_mode(const struct cftype
*cft
)
2278 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2279 cft
->read_map
|| cft
->read_seq_string
)
2282 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2283 cft
->write_string
|| cft
->trigger
)
2289 int cgroup_add_file(struct cgroup
*cgrp
,
2290 struct cgroup_subsys
*subsys
,
2291 const struct cftype
*cft
)
2293 struct dentry
*dir
= cgrp
->dentry
;
2294 struct dentry
*dentry
;
2298 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2299 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2300 strcpy(name
, subsys
->name
);
2303 strcat(name
, cft
->name
);
2304 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2305 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2306 if (!IS_ERR(dentry
)) {
2307 mode
= cgroup_file_mode(cft
);
2308 error
= cgroup_create_file(dentry
, mode
| S_IFREG
,
2311 dentry
->d_fsdata
= (void *)cft
;
2314 error
= PTR_ERR(dentry
);
2317 EXPORT_SYMBOL_GPL(cgroup_add_file
);
2319 int cgroup_add_files(struct cgroup
*cgrp
,
2320 struct cgroup_subsys
*subsys
,
2321 const struct cftype cft
[],
2325 for (i
= 0; i
< count
; i
++) {
2326 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
2332 EXPORT_SYMBOL_GPL(cgroup_add_files
);
2335 * cgroup_task_count - count the number of tasks in a cgroup.
2336 * @cgrp: the cgroup in question
2338 * Return the number of tasks in the cgroup.
2340 int cgroup_task_count(const struct cgroup
*cgrp
)
2343 struct cg_cgroup_link
*link
;
2345 read_lock(&css_set_lock
);
2346 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2347 count
+= atomic_read(&link
->cg
->refcount
);
2349 read_unlock(&css_set_lock
);
2354 * Advance a list_head iterator. The iterator should be positioned at
2355 * the start of a css_set
2357 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2358 struct cgroup_iter
*it
)
2360 struct list_head
*l
= it
->cg_link
;
2361 struct cg_cgroup_link
*link
;
2364 /* Advance to the next non-empty css_set */
2367 if (l
== &cgrp
->css_sets
) {
2371 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2373 } while (list_empty(&cg
->tasks
));
2375 it
->task
= cg
->tasks
.next
;
2379 * To reduce the fork() overhead for systems that are not actually
2380 * using their cgroups capability, we don't maintain the lists running
2381 * through each css_set to its tasks until we see the list actually
2382 * used - in other words after the first call to cgroup_iter_start().
2384 * The tasklist_lock is not held here, as do_each_thread() and
2385 * while_each_thread() are protected by RCU.
2387 static void cgroup_enable_task_cg_lists(void)
2389 struct task_struct
*p
, *g
;
2390 write_lock(&css_set_lock
);
2391 use_task_css_set_links
= 1;
2392 do_each_thread(g
, p
) {
2395 * We should check if the process is exiting, otherwise
2396 * it will race with cgroup_exit() in that the list
2397 * entry won't be deleted though the process has exited.
2399 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2400 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2402 } while_each_thread(g
, p
);
2403 write_unlock(&css_set_lock
);
2406 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2409 * The first time anyone tries to iterate across a cgroup,
2410 * we need to enable the list linking each css_set to its
2411 * tasks, and fix up all existing tasks.
2413 if (!use_task_css_set_links
)
2414 cgroup_enable_task_cg_lists();
2416 read_lock(&css_set_lock
);
2417 it
->cg_link
= &cgrp
->css_sets
;
2418 cgroup_advance_iter(cgrp
, it
);
2421 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
2422 struct cgroup_iter
*it
)
2424 struct task_struct
*res
;
2425 struct list_head
*l
= it
->task
;
2426 struct cg_cgroup_link
*link
;
2428 /* If the iterator cg is NULL, we have no tasks */
2431 res
= list_entry(l
, struct task_struct
, cg_list
);
2432 /* Advance iterator to find next entry */
2434 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
2435 if (l
== &link
->cg
->tasks
) {
2436 /* We reached the end of this task list - move on to
2437 * the next cg_cgroup_link */
2438 cgroup_advance_iter(cgrp
, it
);
2445 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2447 read_unlock(&css_set_lock
);
2450 static inline int started_after_time(struct task_struct
*t1
,
2451 struct timespec
*time
,
2452 struct task_struct
*t2
)
2454 int start_diff
= timespec_compare(&t1
->start_time
, time
);
2455 if (start_diff
> 0) {
2457 } else if (start_diff
< 0) {
2461 * Arbitrarily, if two processes started at the same
2462 * time, we'll say that the lower pointer value
2463 * started first. Note that t2 may have exited by now
2464 * so this may not be a valid pointer any longer, but
2465 * that's fine - it still serves to distinguish
2466 * between two tasks started (effectively) simultaneously.
2473 * This function is a callback from heap_insert() and is used to order
2475 * In this case we order the heap in descending task start time.
2477 static inline int started_after(void *p1
, void *p2
)
2479 struct task_struct
*t1
= p1
;
2480 struct task_struct
*t2
= p2
;
2481 return started_after_time(t1
, &t2
->start_time
, t2
);
2485 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2486 * @scan: struct cgroup_scanner containing arguments for the scan
2488 * Arguments include pointers to callback functions test_task() and
2490 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2491 * and if it returns true, call process_task() for it also.
2492 * The test_task pointer may be NULL, meaning always true (select all tasks).
2493 * Effectively duplicates cgroup_iter_{start,next,end}()
2494 * but does not lock css_set_lock for the call to process_task().
2495 * The struct cgroup_scanner may be embedded in any structure of the caller's
2497 * It is guaranteed that process_task() will act on every task that
2498 * is a member of the cgroup for the duration of this call. This
2499 * function may or may not call process_task() for tasks that exit
2500 * or move to a different cgroup during the call, or are forked or
2501 * move into the cgroup during the call.
2503 * Note that test_task() may be called with locks held, and may in some
2504 * situations be called multiple times for the same task, so it should
2506 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2507 * pre-allocated and will be used for heap operations (and its "gt" member will
2508 * be overwritten), else a temporary heap will be used (allocation of which
2509 * may cause this function to fail).
2511 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
2514 struct cgroup_iter it
;
2515 struct task_struct
*p
, *dropped
;
2516 /* Never dereference latest_task, since it's not refcounted */
2517 struct task_struct
*latest_task
= NULL
;
2518 struct ptr_heap tmp_heap
;
2519 struct ptr_heap
*heap
;
2520 struct timespec latest_time
= { 0, 0 };
2523 /* The caller supplied our heap and pre-allocated its memory */
2525 heap
->gt
= &started_after
;
2527 /* We need to allocate our own heap memory */
2529 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
2531 /* cannot allocate the heap */
2537 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2538 * to determine which are of interest, and using the scanner's
2539 * "process_task" callback to process any of them that need an update.
2540 * Since we don't want to hold any locks during the task updates,
2541 * gather tasks to be processed in a heap structure.
2542 * The heap is sorted by descending task start time.
2543 * If the statically-sized heap fills up, we overflow tasks that
2544 * started later, and in future iterations only consider tasks that
2545 * started after the latest task in the previous pass. This
2546 * guarantees forward progress and that we don't miss any tasks.
2549 cgroup_iter_start(scan
->cg
, &it
);
2550 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
2552 * Only affect tasks that qualify per the caller's callback,
2553 * if he provided one
2555 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
2558 * Only process tasks that started after the last task
2561 if (!started_after_time(p
, &latest_time
, latest_task
))
2563 dropped
= heap_insert(heap
, p
);
2564 if (dropped
== NULL
) {
2566 * The new task was inserted; the heap wasn't
2570 } else if (dropped
!= p
) {
2572 * The new task was inserted, and pushed out a
2576 put_task_struct(dropped
);
2579 * Else the new task was newer than anything already in
2580 * the heap and wasn't inserted
2583 cgroup_iter_end(scan
->cg
, &it
);
2586 for (i
= 0; i
< heap
->size
; i
++) {
2587 struct task_struct
*q
= heap
->ptrs
[i
];
2589 latest_time
= q
->start_time
;
2592 /* Process the task per the caller's callback */
2593 scan
->process_task(q
, scan
);
2597 * If we had to process any tasks at all, scan again
2598 * in case some of them were in the middle of forking
2599 * children that didn't get processed.
2600 * Not the most efficient way to do it, but it avoids
2601 * having to take callback_mutex in the fork path
2605 if (heap
== &tmp_heap
)
2606 heap_free(&tmp_heap
);
2611 * Stuff for reading the 'tasks'/'procs' files.
2613 * Reading this file can return large amounts of data if a cgroup has
2614 * *lots* of attached tasks. So it may need several calls to read(),
2615 * but we cannot guarantee that the information we produce is correct
2616 * unless we produce it entirely atomically.
2621 * The following two functions "fix" the issue where there are more pids
2622 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
2623 * TODO: replace with a kernel-wide solution to this problem
2625 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
2626 static void *pidlist_allocate(int count
)
2628 if (PIDLIST_TOO_LARGE(count
))
2629 return vmalloc(count
* sizeof(pid_t
));
2631 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
2633 static void pidlist_free(void *p
)
2635 if (is_vmalloc_addr(p
))
2640 static void *pidlist_resize(void *p
, int newcount
)
2643 /* note: if new alloc fails, old p will still be valid either way */
2644 if (is_vmalloc_addr(p
)) {
2645 newlist
= vmalloc(newcount
* sizeof(pid_t
));
2648 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
2651 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
2657 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
2658 * If the new stripped list is sufficiently smaller and there's enough memory
2659 * to allocate a new buffer, will let go of the unneeded memory. Returns the
2660 * number of unique elements.
2662 /* is the size difference enough that we should re-allocate the array? */
2663 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
2664 static int pidlist_uniq(pid_t
**p
, int length
)
2671 * we presume the 0th element is unique, so i starts at 1. trivial
2672 * edge cases first; no work needs to be done for either
2674 if (length
== 0 || length
== 1)
2676 /* src and dest walk down the list; dest counts unique elements */
2677 for (src
= 1; src
< length
; src
++) {
2678 /* find next unique element */
2679 while (list
[src
] == list
[src
-1]) {
2684 /* dest always points to where the next unique element goes */
2685 list
[dest
] = list
[src
];
2690 * if the length difference is large enough, we want to allocate a
2691 * smaller buffer to save memory. if this fails due to out of memory,
2692 * we'll just stay with what we've got.
2694 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
2695 newlist
= pidlist_resize(list
, dest
);
2702 static int cmppid(const void *a
, const void *b
)
2704 return *(pid_t
*)a
- *(pid_t
*)b
;
2708 * find the appropriate pidlist for our purpose (given procs vs tasks)
2709 * returns with the lock on that pidlist already held, and takes care
2710 * of the use count, or returns NULL with no locks held if we're out of
2713 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
2714 enum cgroup_filetype type
)
2716 struct cgroup_pidlist
*l
;
2717 /* don't need task_nsproxy() if we're looking at ourself */
2718 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
2721 * We can't drop the pidlist_mutex before taking the l->mutex in case
2722 * the last ref-holder is trying to remove l from the list at the same
2723 * time. Holding the pidlist_mutex precludes somebody taking whichever
2724 * list we find out from under us - compare release_pid_array().
2726 mutex_lock(&cgrp
->pidlist_mutex
);
2727 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
2728 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
2729 /* make sure l doesn't vanish out from under us */
2730 down_write(&l
->mutex
);
2731 mutex_unlock(&cgrp
->pidlist_mutex
);
2735 /* entry not found; create a new one */
2736 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
2738 mutex_unlock(&cgrp
->pidlist_mutex
);
2741 init_rwsem(&l
->mutex
);
2742 down_write(&l
->mutex
);
2744 l
->key
.ns
= get_pid_ns(ns
);
2745 l
->use_count
= 0; /* don't increment here */
2748 list_add(&l
->links
, &cgrp
->pidlists
);
2749 mutex_unlock(&cgrp
->pidlist_mutex
);
2754 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
2756 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
2757 struct cgroup_pidlist
**lp
)
2761 int pid
, n
= 0; /* used for populating the array */
2762 struct cgroup_iter it
;
2763 struct task_struct
*tsk
;
2764 struct cgroup_pidlist
*l
;
2767 * If cgroup gets more users after we read count, we won't have
2768 * enough space - tough. This race is indistinguishable to the
2769 * caller from the case that the additional cgroup users didn't
2770 * show up until sometime later on.
2772 length
= cgroup_task_count(cgrp
);
2773 array
= pidlist_allocate(length
);
2776 /* now, populate the array */
2777 cgroup_iter_start(cgrp
, &it
);
2778 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
2779 if (unlikely(n
== length
))
2781 /* get tgid or pid for procs or tasks file respectively */
2782 if (type
== CGROUP_FILE_PROCS
)
2783 pid
= task_tgid_vnr(tsk
);
2785 pid
= task_pid_vnr(tsk
);
2786 if (pid
> 0) /* make sure to only use valid results */
2789 cgroup_iter_end(cgrp
, &it
);
2791 /* now sort & (if procs) strip out duplicates */
2792 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
2793 if (type
== CGROUP_FILE_PROCS
)
2794 length
= pidlist_uniq(&array
, length
);
2795 l
= cgroup_pidlist_find(cgrp
, type
);
2797 pidlist_free(array
);
2800 /* store array, freeing old if necessary - lock already held */
2801 pidlist_free(l
->list
);
2805 up_write(&l
->mutex
);
2811 * cgroupstats_build - build and fill cgroupstats
2812 * @stats: cgroupstats to fill information into
2813 * @dentry: A dentry entry belonging to the cgroup for which stats have
2816 * Build and fill cgroupstats so that taskstats can export it to user
2819 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
2822 struct cgroup
*cgrp
;
2823 struct cgroup_iter it
;
2824 struct task_struct
*tsk
;
2827 * Validate dentry by checking the superblock operations,
2828 * and make sure it's a directory.
2830 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
2831 !S_ISDIR(dentry
->d_inode
->i_mode
))
2835 cgrp
= dentry
->d_fsdata
;
2837 cgroup_iter_start(cgrp
, &it
);
2838 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
2839 switch (tsk
->state
) {
2841 stats
->nr_running
++;
2843 case TASK_INTERRUPTIBLE
:
2844 stats
->nr_sleeping
++;
2846 case TASK_UNINTERRUPTIBLE
:
2847 stats
->nr_uninterruptible
++;
2850 stats
->nr_stopped
++;
2853 if (delayacct_is_task_waiting_on_io(tsk
))
2854 stats
->nr_io_wait
++;
2858 cgroup_iter_end(cgrp
, &it
);
2866 * seq_file methods for the tasks/procs files. The seq_file position is the
2867 * next pid to display; the seq_file iterator is a pointer to the pid
2868 * in the cgroup->l->list array.
2871 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
2874 * Initially we receive a position value that corresponds to
2875 * one more than the last pid shown (or 0 on the first call or
2876 * after a seek to the start). Use a binary-search to find the
2877 * next pid to display, if any
2879 struct cgroup_pidlist
*l
= s
->private;
2880 int index
= 0, pid
= *pos
;
2883 down_read(&l
->mutex
);
2885 int end
= l
->length
;
2887 while (index
< end
) {
2888 int mid
= (index
+ end
) / 2;
2889 if (l
->list
[mid
] == pid
) {
2892 } else if (l
->list
[mid
] <= pid
)
2898 /* If we're off the end of the array, we're done */
2899 if (index
>= l
->length
)
2901 /* Update the abstract position to be the actual pid that we found */
2902 iter
= l
->list
+ index
;
2907 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
2909 struct cgroup_pidlist
*l
= s
->private;
2913 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
2915 struct cgroup_pidlist
*l
= s
->private;
2917 pid_t
*end
= l
->list
+ l
->length
;
2919 * Advance to the next pid in the array. If this goes off the
2931 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
2933 return seq_printf(s
, "%d\n", *(int *)v
);
2937 * seq_operations functions for iterating on pidlists through seq_file -
2938 * independent of whether it's tasks or procs
2940 static const struct seq_operations cgroup_pidlist_seq_operations
= {
2941 .start
= cgroup_pidlist_start
,
2942 .stop
= cgroup_pidlist_stop
,
2943 .next
= cgroup_pidlist_next
,
2944 .show
= cgroup_pidlist_show
,
2947 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
2950 * the case where we're the last user of this particular pidlist will
2951 * have us remove it from the cgroup's list, which entails taking the
2952 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
2953 * pidlist_mutex, we have to take pidlist_mutex first.
2955 mutex_lock(&l
->owner
->pidlist_mutex
);
2956 down_write(&l
->mutex
);
2957 BUG_ON(!l
->use_count
);
2958 if (!--l
->use_count
) {
2959 /* we're the last user if refcount is 0; remove and free */
2960 list_del(&l
->links
);
2961 mutex_unlock(&l
->owner
->pidlist_mutex
);
2962 pidlist_free(l
->list
);
2963 put_pid_ns(l
->key
.ns
);
2964 up_write(&l
->mutex
);
2968 mutex_unlock(&l
->owner
->pidlist_mutex
);
2969 up_write(&l
->mutex
);
2972 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
2974 struct cgroup_pidlist
*l
;
2975 if (!(file
->f_mode
& FMODE_READ
))
2978 * the seq_file will only be initialized if the file was opened for
2979 * reading; hence we check if it's not null only in that case.
2981 l
= ((struct seq_file
*)file
->private_data
)->private;
2982 cgroup_release_pid_array(l
);
2983 return seq_release(inode
, file
);
2986 static const struct file_operations cgroup_pidlist_operations
= {
2988 .llseek
= seq_lseek
,
2989 .write
= cgroup_file_write
,
2990 .release
= cgroup_pidlist_release
,
2994 * The following functions handle opens on a file that displays a pidlist
2995 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
2998 /* helper function for the two below it */
2999 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3001 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3002 struct cgroup_pidlist
*l
;
3005 /* Nothing to do for write-only files */
3006 if (!(file
->f_mode
& FMODE_READ
))
3009 /* have the array populated */
3010 retval
= pidlist_array_load(cgrp
, type
, &l
);
3013 /* configure file information */
3014 file
->f_op
= &cgroup_pidlist_operations
;
3016 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3018 cgroup_release_pid_array(l
);
3021 ((struct seq_file
*)file
->private_data
)->private = l
;
3024 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3026 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3028 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3030 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3033 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3036 return notify_on_release(cgrp
);
3039 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3043 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3045 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3047 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3052 * Unregister event and free resources.
3054 * Gets called from workqueue.
3056 static void cgroup_event_remove(struct work_struct
*work
)
3058 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3060 struct cgroup
*cgrp
= event
->cgrp
;
3062 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3064 eventfd_ctx_put(event
->eventfd
);
3070 * Gets called on POLLHUP on eventfd when user closes it.
3072 * Called with wqh->lock held and interrupts disabled.
3074 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3075 int sync
, void *key
)
3077 struct cgroup_event
*event
= container_of(wait
,
3078 struct cgroup_event
, wait
);
3079 struct cgroup
*cgrp
= event
->cgrp
;
3080 unsigned long flags
= (unsigned long)key
;
3082 if (flags
& POLLHUP
) {
3083 __remove_wait_queue(event
->wqh
, &event
->wait
);
3084 spin_lock(&cgrp
->event_list_lock
);
3085 list_del(&event
->list
);
3086 spin_unlock(&cgrp
->event_list_lock
);
3088 * We are in atomic context, but cgroup_event_remove() may
3089 * sleep, so we have to call it in workqueue.
3091 schedule_work(&event
->remove
);
3097 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3098 wait_queue_head_t
*wqh
, poll_table
*pt
)
3100 struct cgroup_event
*event
= container_of(pt
,
3101 struct cgroup_event
, pt
);
3104 add_wait_queue(wqh
, &event
->wait
);
3108 * Parse input and register new cgroup event handler.
3110 * Input must be in format '<event_fd> <control_fd> <args>'.
3111 * Interpretation of args is defined by control file implementation.
3113 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3116 struct cgroup_event
*event
= NULL
;
3117 unsigned int efd
, cfd
;
3118 struct file
*efile
= NULL
;
3119 struct file
*cfile
= NULL
;
3123 efd
= simple_strtoul(buffer
, &endp
, 10);
3128 cfd
= simple_strtoul(buffer
, &endp
, 10);
3129 if ((*endp
!= ' ') && (*endp
!= '\0'))
3133 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3137 INIT_LIST_HEAD(&event
->list
);
3138 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3139 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3140 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3142 efile
= eventfd_fget(efd
);
3143 if (IS_ERR(efile
)) {
3144 ret
= PTR_ERR(efile
);
3148 event
->eventfd
= eventfd_ctx_fileget(efile
);
3149 if (IS_ERR(event
->eventfd
)) {
3150 ret
= PTR_ERR(event
->eventfd
);
3160 /* the process need read permission on control file */
3161 ret
= file_permission(cfile
, MAY_READ
);
3165 event
->cft
= __file_cft(cfile
);
3166 if (IS_ERR(event
->cft
)) {
3167 ret
= PTR_ERR(event
->cft
);
3171 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3176 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3177 event
->eventfd
, buffer
);
3181 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3182 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3188 * Events should be removed after rmdir of cgroup directory, but before
3189 * destroying subsystem state objects. Let's take reference to cgroup
3190 * directory dentry to do that.
3194 spin_lock(&cgrp
->event_list_lock
);
3195 list_add(&event
->list
, &cgrp
->event_list
);
3196 spin_unlock(&cgrp
->event_list_lock
);
3207 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3208 eventfd_ctx_put(event
->eventfd
);
3210 if (!IS_ERR_OR_NULL(efile
))
3218 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3221 return clone_children(cgrp
);
3224 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3229 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3231 clear_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3236 * for the common functions, 'private' gives the type of file
3238 /* for hysterical raisins, we can't put this on the older files */
3239 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3240 static struct cftype files
[] = {
3243 .open
= cgroup_tasks_open
,
3244 .write_u64
= cgroup_tasks_write
,
3245 .release
= cgroup_pidlist_release
,
3246 .mode
= S_IRUGO
| S_IWUSR
,
3249 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3250 .open
= cgroup_procs_open
,
3251 /* .write_u64 = cgroup_procs_write, TODO */
3252 .release
= cgroup_pidlist_release
,
3256 .name
= "notify_on_release",
3257 .read_u64
= cgroup_read_notify_on_release
,
3258 .write_u64
= cgroup_write_notify_on_release
,
3261 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3262 .write_string
= cgroup_write_event_control
,
3266 .name
= "cgroup.clone_children",
3267 .read_u64
= cgroup_clone_children_read
,
3268 .write_u64
= cgroup_clone_children_write
,
3272 static struct cftype cft_release_agent
= {
3273 .name
= "release_agent",
3274 .read_seq_string
= cgroup_release_agent_show
,
3275 .write_string
= cgroup_release_agent_write
,
3276 .max_write_len
= PATH_MAX
,
3279 static int cgroup_populate_dir(struct cgroup
*cgrp
)
3282 struct cgroup_subsys
*ss
;
3284 /* First clear out any existing files */
3285 cgroup_clear_directory(cgrp
->dentry
);
3287 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
3291 if (cgrp
== cgrp
->top_cgroup
) {
3292 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
3296 for_each_subsys(cgrp
->root
, ss
) {
3297 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
3300 /* This cgroup is ready now */
3301 for_each_subsys(cgrp
->root
, ss
) {
3302 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3304 * Update id->css pointer and make this css visible from
3305 * CSS ID functions. This pointer will be dereferened
3306 * from RCU-read-side without locks.
3309 rcu_assign_pointer(css
->id
->css
, css
);
3315 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
3316 struct cgroup_subsys
*ss
,
3317 struct cgroup
*cgrp
)
3320 atomic_set(&css
->refcnt
, 1);
3323 if (cgrp
== dummytop
)
3324 set_bit(CSS_ROOT
, &css
->flags
);
3325 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
3326 cgrp
->subsys
[ss
->subsys_id
] = css
;
3329 static void cgroup_lock_hierarchy(struct cgroupfs_root
*root
)
3331 /* We need to take each hierarchy_mutex in a consistent order */
3335 * No worry about a race with rebind_subsystems that might mess up the
3336 * locking order, since both parties are under cgroup_mutex.
3338 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3339 struct cgroup_subsys
*ss
= subsys
[i
];
3342 if (ss
->root
== root
)
3343 mutex_lock(&ss
->hierarchy_mutex
);
3347 static void cgroup_unlock_hierarchy(struct cgroupfs_root
*root
)
3351 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3352 struct cgroup_subsys
*ss
= subsys
[i
];
3355 if (ss
->root
== root
)
3356 mutex_unlock(&ss
->hierarchy_mutex
);
3361 * cgroup_create - create a cgroup
3362 * @parent: cgroup that will be parent of the new cgroup
3363 * @dentry: dentry of the new cgroup
3364 * @mode: mode to set on new inode
3366 * Must be called with the mutex on the parent inode held
3368 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
3371 struct cgroup
*cgrp
;
3372 struct cgroupfs_root
*root
= parent
->root
;
3374 struct cgroup_subsys
*ss
;
3375 struct super_block
*sb
= root
->sb
;
3377 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3381 /* Grab a reference on the superblock so the hierarchy doesn't
3382 * get deleted on unmount if there are child cgroups. This
3383 * can be done outside cgroup_mutex, since the sb can't
3384 * disappear while someone has an open control file on the
3386 atomic_inc(&sb
->s_active
);
3388 mutex_lock(&cgroup_mutex
);
3390 init_cgroup_housekeeping(cgrp
);
3392 cgrp
->parent
= parent
;
3393 cgrp
->root
= parent
->root
;
3394 cgrp
->top_cgroup
= parent
->top_cgroup
;
3396 if (notify_on_release(parent
))
3397 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3399 if (clone_children(parent
))
3400 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3402 for_each_subsys(root
, ss
) {
3403 struct cgroup_subsys_state
*css
= ss
->create(ss
, cgrp
);
3409 init_cgroup_css(css
, ss
, cgrp
);
3411 err
= alloc_css_id(ss
, parent
, cgrp
);
3415 /* At error, ->destroy() callback has to free assigned ID. */
3416 if (clone_children(parent
) && ss
->post_clone
)
3417 ss
->post_clone(ss
, cgrp
);
3420 cgroup_lock_hierarchy(root
);
3421 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
3422 cgroup_unlock_hierarchy(root
);
3423 root
->number_of_cgroups
++;
3425 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
3429 /* The cgroup directory was pre-locked for us */
3430 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
3432 err
= cgroup_populate_dir(cgrp
);
3433 /* If err < 0, we have a half-filled directory - oh well ;) */
3435 mutex_unlock(&cgroup_mutex
);
3436 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
3442 cgroup_lock_hierarchy(root
);
3443 list_del(&cgrp
->sibling
);
3444 cgroup_unlock_hierarchy(root
);
3445 root
->number_of_cgroups
--;
3449 for_each_subsys(root
, ss
) {
3450 if (cgrp
->subsys
[ss
->subsys_id
])
3451 ss
->destroy(ss
, cgrp
);
3454 mutex_unlock(&cgroup_mutex
);
3456 /* Release the reference count that we took on the superblock */
3457 deactivate_super(sb
);
3463 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3465 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
3467 /* the vfs holds inode->i_mutex already */
3468 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
3471 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
3473 /* Check the reference count on each subsystem. Since we
3474 * already established that there are no tasks in the
3475 * cgroup, if the css refcount is also 1, then there should
3476 * be no outstanding references, so the subsystem is safe to
3477 * destroy. We scan across all subsystems rather than using
3478 * the per-hierarchy linked list of mounted subsystems since
3479 * we can be called via check_for_release() with no
3480 * synchronization other than RCU, and the subsystem linked
3481 * list isn't RCU-safe */
3484 * We won't need to lock the subsys array, because the subsystems
3485 * we're concerned about aren't going anywhere since our cgroup root
3486 * has a reference on them.
3488 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3489 struct cgroup_subsys
*ss
= subsys
[i
];
3490 struct cgroup_subsys_state
*css
;
3491 /* Skip subsystems not present or not in this hierarchy */
3492 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
3494 css
= cgrp
->subsys
[ss
->subsys_id
];
3495 /* When called from check_for_release() it's possible
3496 * that by this point the cgroup has been removed
3497 * and the css deleted. But a false-positive doesn't
3498 * matter, since it can only happen if the cgroup
3499 * has been deleted and hence no longer needs the
3500 * release agent to be called anyway. */
3501 if (css
&& (atomic_read(&css
->refcnt
) > 1))
3508 * Atomically mark all (or else none) of the cgroup's CSS objects as
3509 * CSS_REMOVED. Return true on success, or false if the cgroup has
3510 * busy subsystems. Call with cgroup_mutex held
3513 static int cgroup_clear_css_refs(struct cgroup
*cgrp
)
3515 struct cgroup_subsys
*ss
;
3516 unsigned long flags
;
3517 bool failed
= false;
3518 local_irq_save(flags
);
3519 for_each_subsys(cgrp
->root
, ss
) {
3520 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3523 /* We can only remove a CSS with a refcnt==1 */
3524 refcnt
= atomic_read(&css
->refcnt
);
3531 * Drop the refcnt to 0 while we check other
3532 * subsystems. This will cause any racing
3533 * css_tryget() to spin until we set the
3534 * CSS_REMOVED bits or abort
3536 if (atomic_cmpxchg(&css
->refcnt
, refcnt
, 0) == refcnt
)
3542 for_each_subsys(cgrp
->root
, ss
) {
3543 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3546 * Restore old refcnt if we previously managed
3547 * to clear it from 1 to 0
3549 if (!atomic_read(&css
->refcnt
))
3550 atomic_set(&css
->refcnt
, 1);
3552 /* Commit the fact that the CSS is removed */
3553 set_bit(CSS_REMOVED
, &css
->flags
);
3556 local_irq_restore(flags
);
3560 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
3562 struct cgroup
*cgrp
= dentry
->d_fsdata
;
3564 struct cgroup
*parent
;
3566 struct cgroup_event
*event
, *tmp
;
3569 /* the vfs holds both inode->i_mutex already */
3571 mutex_lock(&cgroup_mutex
);
3572 if (atomic_read(&cgrp
->count
) != 0) {
3573 mutex_unlock(&cgroup_mutex
);
3576 if (!list_empty(&cgrp
->children
)) {
3577 mutex_unlock(&cgroup_mutex
);
3580 mutex_unlock(&cgroup_mutex
);
3583 * In general, subsystem has no css->refcnt after pre_destroy(). But
3584 * in racy cases, subsystem may have to get css->refcnt after
3585 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3586 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3587 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3588 * and subsystem's reference count handling. Please see css_get/put
3589 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3591 set_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3594 * Call pre_destroy handlers of subsys. Notify subsystems
3595 * that rmdir() request comes.
3597 ret
= cgroup_call_pre_destroy(cgrp
);
3599 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3603 mutex_lock(&cgroup_mutex
);
3604 parent
= cgrp
->parent
;
3605 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
)) {
3606 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3607 mutex_unlock(&cgroup_mutex
);
3610 prepare_to_wait(&cgroup_rmdir_waitq
, &wait
, TASK_INTERRUPTIBLE
);
3611 if (!cgroup_clear_css_refs(cgrp
)) {
3612 mutex_unlock(&cgroup_mutex
);
3614 * Because someone may call cgroup_wakeup_rmdir_waiter() before
3615 * prepare_to_wait(), we need to check this flag.
3617 if (test_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
))
3619 finish_wait(&cgroup_rmdir_waitq
, &wait
);
3620 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3621 if (signal_pending(current
))
3625 /* NO css_tryget() can success after here. */
3626 finish_wait(&cgroup_rmdir_waitq
, &wait
);
3627 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3629 spin_lock(&release_list_lock
);
3630 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
3631 if (!list_empty(&cgrp
->release_list
))
3632 list_del(&cgrp
->release_list
);
3633 spin_unlock(&release_list_lock
);
3635 cgroup_lock_hierarchy(cgrp
->root
);
3636 /* delete this cgroup from parent->children */
3637 list_del(&cgrp
->sibling
);
3638 cgroup_unlock_hierarchy(cgrp
->root
);
3640 spin_lock(&cgrp
->dentry
->d_lock
);
3641 d
= dget(cgrp
->dentry
);
3642 spin_unlock(&d
->d_lock
);
3644 cgroup_d_remove_dir(d
);
3647 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
3648 check_for_release(parent
);
3651 * Unregister events and notify userspace.
3652 * Notify userspace about cgroup removing only after rmdir of cgroup
3653 * directory to avoid race between userspace and kernelspace
3655 spin_lock(&cgrp
->event_list_lock
);
3656 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
3657 list_del(&event
->list
);
3658 remove_wait_queue(event
->wqh
, &event
->wait
);
3659 eventfd_signal(event
->eventfd
, 1);
3660 schedule_work(&event
->remove
);
3662 spin_unlock(&cgrp
->event_list_lock
);
3664 mutex_unlock(&cgroup_mutex
);
3668 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
3670 struct cgroup_subsys_state
*css
;
3672 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
3674 /* Create the top cgroup state for this subsystem */
3675 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
3676 ss
->root
= &rootnode
;
3677 css
= ss
->create(ss
, dummytop
);
3678 /* We don't handle early failures gracefully */
3679 BUG_ON(IS_ERR(css
));
3680 init_cgroup_css(css
, ss
, dummytop
);
3682 /* Update the init_css_set to contain a subsys
3683 * pointer to this state - since the subsystem is
3684 * newly registered, all tasks and hence the
3685 * init_css_set is in the subsystem's top cgroup. */
3686 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
3688 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
3690 /* At system boot, before all subsystems have been
3691 * registered, no tasks have been forked, so we don't
3692 * need to invoke fork callbacks here. */
3693 BUG_ON(!list_empty(&init_task
.tasks
));
3695 mutex_init(&ss
->hierarchy_mutex
);
3696 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
3699 /* this function shouldn't be used with modular subsystems, since they
3700 * need to register a subsys_id, among other things */
3705 * cgroup_load_subsys: load and register a modular subsystem at runtime
3706 * @ss: the subsystem to load
3708 * This function should be called in a modular subsystem's initcall. If the
3709 * subsystem is built as a module, it will be assigned a new subsys_id and set
3710 * up for use. If the subsystem is built-in anyway, work is delegated to the
3711 * simpler cgroup_init_subsys.
3713 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
3716 struct cgroup_subsys_state
*css
;
3718 /* check name and function validity */
3719 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
3720 ss
->create
== NULL
|| ss
->destroy
== NULL
)
3724 * we don't support callbacks in modular subsystems. this check is
3725 * before the ss->module check for consistency; a subsystem that could
3726 * be a module should still have no callbacks even if the user isn't
3727 * compiling it as one.
3729 if (ss
->fork
|| ss
->exit
)
3733 * an optionally modular subsystem is built-in: we want to do nothing,
3734 * since cgroup_init_subsys will have already taken care of it.
3736 if (ss
->module
== NULL
) {
3737 /* a few sanity checks */
3738 BUG_ON(ss
->subsys_id
>= CGROUP_BUILTIN_SUBSYS_COUNT
);
3739 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
3744 * need to register a subsys id before anything else - for example,
3745 * init_cgroup_css needs it.
3747 mutex_lock(&cgroup_mutex
);
3748 /* find the first empty slot in the array */
3749 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3750 if (subsys
[i
] == NULL
)
3753 if (i
== CGROUP_SUBSYS_COUNT
) {
3754 /* maximum number of subsystems already registered! */
3755 mutex_unlock(&cgroup_mutex
);
3758 /* assign ourselves the subsys_id */
3763 * no ss->create seems to need anything important in the ss struct, so
3764 * this can happen first (i.e. before the rootnode attachment).
3766 css
= ss
->create(ss
, dummytop
);
3768 /* failure case - need to deassign the subsys[] slot. */
3770 mutex_unlock(&cgroup_mutex
);
3771 return PTR_ERR(css
);
3774 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
3775 ss
->root
= &rootnode
;
3777 /* our new subsystem will be attached to the dummy hierarchy. */
3778 init_cgroup_css(css
, ss
, dummytop
);
3779 /* init_idr must be after init_cgroup_css because it sets css->id. */
3781 int ret
= cgroup_init_idr(ss
, css
);
3783 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
3784 ss
->destroy(ss
, dummytop
);
3786 mutex_unlock(&cgroup_mutex
);
3792 * Now we need to entangle the css into the existing css_sets. unlike
3793 * in cgroup_init_subsys, there are now multiple css_sets, so each one
3794 * will need a new pointer to it; done by iterating the css_set_table.
3795 * furthermore, modifying the existing css_sets will corrupt the hash
3796 * table state, so each changed css_set will need its hash recomputed.
3797 * this is all done under the css_set_lock.
3799 write_lock(&css_set_lock
);
3800 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
3802 struct hlist_node
*node
, *tmp
;
3803 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
3805 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
3806 /* skip entries that we already rehashed */
3807 if (cg
->subsys
[ss
->subsys_id
])
3809 /* remove existing entry */
3810 hlist_del(&cg
->hlist
);
3812 cg
->subsys
[ss
->subsys_id
] = css
;
3813 /* recompute hash and restore entry */
3814 new_bucket
= css_set_hash(cg
->subsys
);
3815 hlist_add_head(&cg
->hlist
, new_bucket
);
3818 write_unlock(&css_set_lock
);
3820 mutex_init(&ss
->hierarchy_mutex
);
3821 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
3825 mutex_unlock(&cgroup_mutex
);
3828 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
3831 * cgroup_unload_subsys: unload a modular subsystem
3832 * @ss: the subsystem to unload
3834 * This function should be called in a modular subsystem's exitcall. When this
3835 * function is invoked, the refcount on the subsystem's module will be 0, so
3836 * the subsystem will not be attached to any hierarchy.
3838 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
3840 struct cg_cgroup_link
*link
;
3841 struct hlist_head
*hhead
;
3843 BUG_ON(ss
->module
== NULL
);
3846 * we shouldn't be called if the subsystem is in use, and the use of
3847 * try_module_get in parse_cgroupfs_options should ensure that it
3848 * doesn't start being used while we're killing it off.
3850 BUG_ON(ss
->root
!= &rootnode
);
3852 mutex_lock(&cgroup_mutex
);
3853 /* deassign the subsys_id */
3854 BUG_ON(ss
->subsys_id
< CGROUP_BUILTIN_SUBSYS_COUNT
);
3855 subsys
[ss
->subsys_id
] = NULL
;
3857 /* remove subsystem from rootnode's list of subsystems */
3858 list_del(&ss
->sibling
);
3861 * disentangle the css from all css_sets attached to the dummytop. as
3862 * in loading, we need to pay our respects to the hashtable gods.
3864 write_lock(&css_set_lock
);
3865 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
3866 struct css_set
*cg
= link
->cg
;
3868 hlist_del(&cg
->hlist
);
3869 BUG_ON(!cg
->subsys
[ss
->subsys_id
]);
3870 cg
->subsys
[ss
->subsys_id
] = NULL
;
3871 hhead
= css_set_hash(cg
->subsys
);
3872 hlist_add_head(&cg
->hlist
, hhead
);
3874 write_unlock(&css_set_lock
);
3877 * remove subsystem's css from the dummytop and free it - need to free
3878 * before marking as null because ss->destroy needs the cgrp->subsys
3879 * pointer to find their state. note that this also takes care of
3880 * freeing the css_id.
3882 ss
->destroy(ss
, dummytop
);
3883 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
3885 mutex_unlock(&cgroup_mutex
);
3887 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
3890 * cgroup_init_early - cgroup initialization at system boot
3892 * Initialize cgroups at system boot, and initialize any
3893 * subsystems that request early init.
3895 int __init
cgroup_init_early(void)
3898 atomic_set(&init_css_set
.refcount
, 1);
3899 INIT_LIST_HEAD(&init_css_set
.cg_links
);
3900 INIT_LIST_HEAD(&init_css_set
.tasks
);
3901 INIT_HLIST_NODE(&init_css_set
.hlist
);
3903 init_cgroup_root(&rootnode
);
3905 init_task
.cgroups
= &init_css_set
;
3907 init_css_set_link
.cg
= &init_css_set
;
3908 init_css_set_link
.cgrp
= dummytop
;
3909 list_add(&init_css_set_link
.cgrp_link_list
,
3910 &rootnode
.top_cgroup
.css_sets
);
3911 list_add(&init_css_set_link
.cg_link_list
,
3912 &init_css_set
.cg_links
);
3914 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
3915 INIT_HLIST_HEAD(&css_set_table
[i
]);
3917 /* at bootup time, we don't worry about modular subsystems */
3918 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
3919 struct cgroup_subsys
*ss
= subsys
[i
];
3922 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
3923 BUG_ON(!ss
->create
);
3924 BUG_ON(!ss
->destroy
);
3925 if (ss
->subsys_id
!= i
) {
3926 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
3927 ss
->name
, ss
->subsys_id
);
3932 cgroup_init_subsys(ss
);
3938 * cgroup_init - cgroup initialization
3940 * Register cgroup filesystem and /proc file, and initialize
3941 * any subsystems that didn't request early init.
3943 int __init
cgroup_init(void)
3947 struct hlist_head
*hhead
;
3949 err
= bdi_init(&cgroup_backing_dev_info
);
3953 /* at bootup time, we don't worry about modular subsystems */
3954 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
3955 struct cgroup_subsys
*ss
= subsys
[i
];
3956 if (!ss
->early_init
)
3957 cgroup_init_subsys(ss
);
3959 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
3962 /* Add init_css_set to the hash table */
3963 hhead
= css_set_hash(init_css_set
.subsys
);
3964 hlist_add_head(&init_css_set
.hlist
, hhead
);
3965 BUG_ON(!init_root_id(&rootnode
));
3967 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
3973 err
= register_filesystem(&cgroup_fs_type
);
3975 kobject_put(cgroup_kobj
);
3979 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
3983 bdi_destroy(&cgroup_backing_dev_info
);
3989 * proc_cgroup_show()
3990 * - Print task's cgroup paths into seq_file, one line for each hierarchy
3991 * - Used for /proc/<pid>/cgroup.
3992 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
3993 * doesn't really matter if tsk->cgroup changes after we read it,
3994 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
3995 * anyway. No need to check that tsk->cgroup != NULL, thanks to
3996 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
3997 * cgroup to top_cgroup.
4000 /* TODO: Use a proper seq_file iterator */
4001 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4004 struct task_struct
*tsk
;
4007 struct cgroupfs_root
*root
;
4010 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4016 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4022 mutex_lock(&cgroup_mutex
);
4024 for_each_active_root(root
) {
4025 struct cgroup_subsys
*ss
;
4026 struct cgroup
*cgrp
;
4029 seq_printf(m
, "%d:", root
->hierarchy_id
);
4030 for_each_subsys(root
, ss
)
4031 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4032 if (strlen(root
->name
))
4033 seq_printf(m
, "%sname=%s", count
? "," : "",
4036 cgrp
= task_cgroup_from_root(tsk
, root
);
4037 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4045 mutex_unlock(&cgroup_mutex
);
4046 put_task_struct(tsk
);
4053 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4055 struct pid
*pid
= PROC_I(inode
)->pid
;
4056 return single_open(file
, proc_cgroup_show
, pid
);
4059 const struct file_operations proc_cgroup_operations
= {
4060 .open
= cgroup_open
,
4062 .llseek
= seq_lseek
,
4063 .release
= single_release
,
4066 /* Display information about each subsystem and each hierarchy */
4067 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4071 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4073 * ideally we don't want subsystems moving around while we do this.
4074 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4075 * subsys/hierarchy state.
4077 mutex_lock(&cgroup_mutex
);
4078 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4079 struct cgroup_subsys
*ss
= subsys
[i
];
4082 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4083 ss
->name
, ss
->root
->hierarchy_id
,
4084 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4086 mutex_unlock(&cgroup_mutex
);
4090 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4092 return single_open(file
, proc_cgroupstats_show
, NULL
);
4095 static const struct file_operations proc_cgroupstats_operations
= {
4096 .open
= cgroupstats_open
,
4098 .llseek
= seq_lseek
,
4099 .release
= single_release
,
4103 * cgroup_fork - attach newly forked task to its parents cgroup.
4104 * @child: pointer to task_struct of forking parent process.
4106 * Description: A task inherits its parent's cgroup at fork().
4108 * A pointer to the shared css_set was automatically copied in
4109 * fork.c by dup_task_struct(). However, we ignore that copy, since
4110 * it was not made under the protection of RCU or cgroup_mutex, so
4111 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4112 * have already changed current->cgroups, allowing the previously
4113 * referenced cgroup group to be removed and freed.
4115 * At the point that cgroup_fork() is called, 'current' is the parent
4116 * task, and the passed argument 'child' points to the child task.
4118 void cgroup_fork(struct task_struct
*child
)
4121 child
->cgroups
= current
->cgroups
;
4122 get_css_set(child
->cgroups
);
4123 task_unlock(current
);
4124 INIT_LIST_HEAD(&child
->cg_list
);
4128 * cgroup_fork_callbacks - run fork callbacks
4129 * @child: the new task
4131 * Called on a new task very soon before adding it to the
4132 * tasklist. No need to take any locks since no-one can
4133 * be operating on this task.
4135 void cgroup_fork_callbacks(struct task_struct
*child
)
4137 if (need_forkexit_callback
) {
4140 * forkexit callbacks are only supported for builtin
4141 * subsystems, and the builtin section of the subsys array is
4142 * immutable, so we don't need to lock the subsys array here.
4144 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4145 struct cgroup_subsys
*ss
= subsys
[i
];
4147 ss
->fork(ss
, child
);
4153 * cgroup_post_fork - called on a new task after adding it to the task list
4154 * @child: the task in question
4156 * Adds the task to the list running through its css_set if necessary.
4157 * Has to be after the task is visible on the task list in case we race
4158 * with the first call to cgroup_iter_start() - to guarantee that the
4159 * new task ends up on its list.
4161 void cgroup_post_fork(struct task_struct
*child
)
4163 if (use_task_css_set_links
) {
4164 write_lock(&css_set_lock
);
4166 if (list_empty(&child
->cg_list
))
4167 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4169 write_unlock(&css_set_lock
);
4173 * cgroup_exit - detach cgroup from exiting task
4174 * @tsk: pointer to task_struct of exiting process
4175 * @run_callback: run exit callbacks?
4177 * Description: Detach cgroup from @tsk and release it.
4179 * Note that cgroups marked notify_on_release force every task in
4180 * them to take the global cgroup_mutex mutex when exiting.
4181 * This could impact scaling on very large systems. Be reluctant to
4182 * use notify_on_release cgroups where very high task exit scaling
4183 * is required on large systems.
4185 * the_top_cgroup_hack:
4187 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4189 * We call cgroup_exit() while the task is still competent to
4190 * handle notify_on_release(), then leave the task attached to the
4191 * root cgroup in each hierarchy for the remainder of its exit.
4193 * To do this properly, we would increment the reference count on
4194 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4195 * code we would add a second cgroup function call, to drop that
4196 * reference. This would just create an unnecessary hot spot on
4197 * the top_cgroup reference count, to no avail.
4199 * Normally, holding a reference to a cgroup without bumping its
4200 * count is unsafe. The cgroup could go away, or someone could
4201 * attach us to a different cgroup, decrementing the count on
4202 * the first cgroup that we never incremented. But in this case,
4203 * top_cgroup isn't going away, and either task has PF_EXITING set,
4204 * which wards off any cgroup_attach_task() attempts, or task is a failed
4205 * fork, never visible to cgroup_attach_task.
4207 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4212 if (run_callbacks
&& need_forkexit_callback
) {
4214 * modular subsystems can't use callbacks, so no need to lock
4217 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4218 struct cgroup_subsys
*ss
= subsys
[i
];
4225 * Unlink from the css_set task list if necessary.
4226 * Optimistically check cg_list before taking
4229 if (!list_empty(&tsk
->cg_list
)) {
4230 write_lock(&css_set_lock
);
4231 if (!list_empty(&tsk
->cg_list
))
4232 list_del(&tsk
->cg_list
);
4233 write_unlock(&css_set_lock
);
4236 /* Reassign the task to the init_css_set. */
4239 tsk
->cgroups
= &init_css_set
;
4242 put_css_set_taskexit(cg
);
4246 * cgroup_clone - clone the cgroup the given subsystem is attached to
4247 * @tsk: the task to be moved
4248 * @subsys: the given subsystem
4249 * @nodename: the name for the new cgroup
4251 * Duplicate the current cgroup in the hierarchy that the given
4252 * subsystem is attached to, and move this task into the new
4255 int cgroup_clone(struct task_struct
*tsk
, struct cgroup_subsys
*subsys
,
4258 struct dentry
*dentry
;
4260 struct cgroup
*parent
, *child
;
4261 struct inode
*inode
;
4263 struct cgroupfs_root
*root
;
4264 struct cgroup_subsys
*ss
;
4266 /* We shouldn't be called by an unregistered subsystem */
4267 BUG_ON(!subsys
->active
);
4269 /* First figure out what hierarchy and cgroup we're dealing
4270 * with, and pin them so we can drop cgroup_mutex */
4271 mutex_lock(&cgroup_mutex
);
4273 root
= subsys
->root
;
4274 if (root
== &rootnode
) {
4275 mutex_unlock(&cgroup_mutex
);
4279 /* Pin the hierarchy */
4280 if (!atomic_inc_not_zero(&root
->sb
->s_active
)) {
4281 /* We race with the final deactivate_super() */
4282 mutex_unlock(&cgroup_mutex
);
4286 /* Keep the cgroup alive */
4288 parent
= task_cgroup(tsk
, subsys
->subsys_id
);
4293 mutex_unlock(&cgroup_mutex
);
4295 /* Now do the VFS work to create a cgroup */
4296 inode
= parent
->dentry
->d_inode
;
4298 /* Hold the parent directory mutex across this operation to
4299 * stop anyone else deleting the new cgroup */
4300 mutex_lock(&inode
->i_mutex
);
4301 dentry
= lookup_one_len(nodename
, parent
->dentry
, strlen(nodename
));
4302 if (IS_ERR(dentry
)) {
4304 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename
,
4306 ret
= PTR_ERR(dentry
);
4310 /* Create the cgroup directory, which also creates the cgroup */
4311 ret
= vfs_mkdir(inode
, dentry
, 0755);
4312 child
= __d_cgrp(dentry
);
4316 "Failed to create cgroup %s: %d\n", nodename
,
4321 /* The cgroup now exists. Retake cgroup_mutex and check
4322 * that we're still in the same state that we thought we
4324 mutex_lock(&cgroup_mutex
);
4325 if ((root
!= subsys
->root
) ||
4326 (parent
!= task_cgroup(tsk
, subsys
->subsys_id
))) {
4327 /* Aargh, we raced ... */
4328 mutex_unlock(&inode
->i_mutex
);
4331 deactivate_super(root
->sb
);
4332 /* The cgroup is still accessible in the VFS, but
4333 * we're not going to try to rmdir() it at this
4336 "Race in cgroup_clone() - leaking cgroup %s\n",
4341 /* do any required auto-setup */
4342 for_each_subsys(root
, ss
) {
4344 ss
->post_clone(ss
, child
);
4347 /* All seems fine. Finish by moving the task into the new cgroup */
4348 ret
= cgroup_attach_task(child
, tsk
);
4349 mutex_unlock(&cgroup_mutex
);
4352 mutex_unlock(&inode
->i_mutex
);
4354 mutex_lock(&cgroup_mutex
);
4356 mutex_unlock(&cgroup_mutex
);
4357 deactivate_super(root
->sb
);
4362 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4363 * @cgrp: the cgroup in question
4364 * @task: the task in question
4366 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4369 * If we are sending in dummytop, then presumably we are creating
4370 * the top cgroup in the subsystem.
4372 * Called only by the ns (nsproxy) cgroup.
4374 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
4377 struct cgroup
*target
;
4379 if (cgrp
== dummytop
)
4382 target
= task_cgroup_from_root(task
, cgrp
->root
);
4383 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
4384 cgrp
= cgrp
->parent
;
4385 ret
= (cgrp
== target
);
4389 static void check_for_release(struct cgroup
*cgrp
)
4391 /* All of these checks rely on RCU to keep the cgroup
4392 * structure alive */
4393 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
4394 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
4395 /* Control Group is currently removeable. If it's not
4396 * already queued for a userspace notification, queue
4398 int need_schedule_work
= 0;
4399 spin_lock(&release_list_lock
);
4400 if (!cgroup_is_removed(cgrp
) &&
4401 list_empty(&cgrp
->release_list
)) {
4402 list_add(&cgrp
->release_list
, &release_list
);
4403 need_schedule_work
= 1;
4405 spin_unlock(&release_list_lock
);
4406 if (need_schedule_work
)
4407 schedule_work(&release_agent_work
);
4411 /* Caller must verify that the css is not for root cgroup */
4412 void __css_put(struct cgroup_subsys_state
*css
, int count
)
4414 struct cgroup
*cgrp
= css
->cgroup
;
4417 val
= atomic_sub_return(count
, &css
->refcnt
);
4419 if (notify_on_release(cgrp
)) {
4420 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4421 check_for_release(cgrp
);
4423 cgroup_wakeup_rmdir_waiter(cgrp
);
4426 WARN_ON_ONCE(val
< 1);
4428 EXPORT_SYMBOL_GPL(__css_put
);
4431 * Notify userspace when a cgroup is released, by running the
4432 * configured release agent with the name of the cgroup (path
4433 * relative to the root of cgroup file system) as the argument.
4435 * Most likely, this user command will try to rmdir this cgroup.
4437 * This races with the possibility that some other task will be
4438 * attached to this cgroup before it is removed, or that some other
4439 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4440 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4441 * unused, and this cgroup will be reprieved from its death sentence,
4442 * to continue to serve a useful existence. Next time it's released,
4443 * we will get notified again, if it still has 'notify_on_release' set.
4445 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4446 * means only wait until the task is successfully execve()'d. The
4447 * separate release agent task is forked by call_usermodehelper(),
4448 * then control in this thread returns here, without waiting for the
4449 * release agent task. We don't bother to wait because the caller of
4450 * this routine has no use for the exit status of the release agent
4451 * task, so no sense holding our caller up for that.
4453 static void cgroup_release_agent(struct work_struct
*work
)
4455 BUG_ON(work
!= &release_agent_work
);
4456 mutex_lock(&cgroup_mutex
);
4457 spin_lock(&release_list_lock
);
4458 while (!list_empty(&release_list
)) {
4459 char *argv
[3], *envp
[3];
4461 char *pathbuf
= NULL
, *agentbuf
= NULL
;
4462 struct cgroup
*cgrp
= list_entry(release_list
.next
,
4465 list_del_init(&cgrp
->release_list
);
4466 spin_unlock(&release_list_lock
);
4467 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4470 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
4472 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
4477 argv
[i
++] = agentbuf
;
4478 argv
[i
++] = pathbuf
;
4482 /* minimal command environment */
4483 envp
[i
++] = "HOME=/";
4484 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4487 /* Drop the lock while we invoke the usermode helper,
4488 * since the exec could involve hitting disk and hence
4489 * be a slow process */
4490 mutex_unlock(&cgroup_mutex
);
4491 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
4492 mutex_lock(&cgroup_mutex
);
4496 spin_lock(&release_list_lock
);
4498 spin_unlock(&release_list_lock
);
4499 mutex_unlock(&cgroup_mutex
);
4502 static int __init
cgroup_disable(char *str
)
4507 while ((token
= strsep(&str
, ",")) != NULL
) {
4511 * cgroup_disable, being at boot time, can't know about module
4512 * subsystems, so we don't worry about them.
4514 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4515 struct cgroup_subsys
*ss
= subsys
[i
];
4517 if (!strcmp(token
, ss
->name
)) {
4519 printk(KERN_INFO
"Disabling %s control group"
4520 " subsystem\n", ss
->name
);
4527 __setup("cgroup_disable=", cgroup_disable
);
4530 * Functons for CSS ID.
4534 *To get ID other than 0, this should be called when !cgroup_is_removed().
4536 unsigned short css_id(struct cgroup_subsys_state
*css
)
4538 struct css_id
*cssid
;
4541 * This css_id() can return correct value when somone has refcnt
4542 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4543 * it's unchanged until freed.
4545 cssid
= rcu_dereference_check(css
->id
,
4546 rcu_read_lock_held() || atomic_read(&css
->refcnt
));
4552 EXPORT_SYMBOL_GPL(css_id
);
4554 unsigned short css_depth(struct cgroup_subsys_state
*css
)
4556 struct css_id
*cssid
;
4558 cssid
= rcu_dereference_check(css
->id
,
4559 rcu_read_lock_held() || atomic_read(&css
->refcnt
));
4562 return cssid
->depth
;
4565 EXPORT_SYMBOL_GPL(css_depth
);
4568 * css_is_ancestor - test "root" css is an ancestor of "child"
4569 * @child: the css to be tested.
4570 * @root: the css supporsed to be an ancestor of the child.
4572 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4573 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4574 * But, considering usual usage, the csses should be valid objects after test.
4575 * Assuming that the caller will do some action to the child if this returns
4576 * returns true, the caller must take "child";s reference count.
4577 * If "child" is valid object and this returns true, "root" is valid, too.
4580 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
4581 const struct cgroup_subsys_state
*root
)
4583 struct css_id
*child_id
;
4584 struct css_id
*root_id
;
4588 child_id
= rcu_dereference(child
->id
);
4589 root_id
= rcu_dereference(root
->id
);
4592 || (child_id
->depth
< root_id
->depth
)
4593 || (child_id
->stack
[root_id
->depth
] != root_id
->id
))
4599 static void __free_css_id_cb(struct rcu_head
*head
)
4603 id
= container_of(head
, struct css_id
, rcu_head
);
4607 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
4609 struct css_id
*id
= css
->id
;
4610 /* When this is called before css_id initialization, id can be NULL */
4614 BUG_ON(!ss
->use_id
);
4616 rcu_assign_pointer(id
->css
, NULL
);
4617 rcu_assign_pointer(css
->id
, NULL
);
4618 spin_lock(&ss
->id_lock
);
4619 idr_remove(&ss
->idr
, id
->id
);
4620 spin_unlock(&ss
->id_lock
);
4621 call_rcu(&id
->rcu_head
, __free_css_id_cb
);
4623 EXPORT_SYMBOL_GPL(free_css_id
);
4626 * This is called by init or create(). Then, calls to this function are
4627 * always serialized (By cgroup_mutex() at create()).
4630 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
4632 struct css_id
*newid
;
4633 int myid
, error
, size
;
4635 BUG_ON(!ss
->use_id
);
4637 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
4638 newid
= kzalloc(size
, GFP_KERNEL
);
4640 return ERR_PTR(-ENOMEM
);
4642 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
4646 spin_lock(&ss
->id_lock
);
4647 /* Don't use 0. allocates an ID of 1-65535 */
4648 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
4649 spin_unlock(&ss
->id_lock
);
4651 /* Returns error when there are no free spaces for new ID.*/
4656 if (myid
> CSS_ID_MAX
)
4660 newid
->depth
= depth
;
4664 spin_lock(&ss
->id_lock
);
4665 idr_remove(&ss
->idr
, myid
);
4666 spin_unlock(&ss
->id_lock
);
4669 return ERR_PTR(error
);
4673 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
4674 struct cgroup_subsys_state
*rootcss
)
4676 struct css_id
*newid
;
4678 spin_lock_init(&ss
->id_lock
);
4681 newid
= get_new_cssid(ss
, 0);
4683 return PTR_ERR(newid
);
4685 newid
->stack
[0] = newid
->id
;
4686 newid
->css
= rootcss
;
4687 rootcss
->id
= newid
;
4691 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
4692 struct cgroup
*child
)
4694 int subsys_id
, i
, depth
= 0;
4695 struct cgroup_subsys_state
*parent_css
, *child_css
;
4696 struct css_id
*child_id
, *parent_id
;
4698 subsys_id
= ss
->subsys_id
;
4699 parent_css
= parent
->subsys
[subsys_id
];
4700 child_css
= child
->subsys
[subsys_id
];
4701 parent_id
= parent_css
->id
;
4702 depth
= parent_id
->depth
+ 1;
4704 child_id
= get_new_cssid(ss
, depth
);
4705 if (IS_ERR(child_id
))
4706 return PTR_ERR(child_id
);
4708 for (i
= 0; i
< depth
; i
++)
4709 child_id
->stack
[i
] = parent_id
->stack
[i
];
4710 child_id
->stack
[depth
] = child_id
->id
;
4712 * child_id->css pointer will be set after this cgroup is available
4713 * see cgroup_populate_dir()
4715 rcu_assign_pointer(child_css
->id
, child_id
);
4721 * css_lookup - lookup css by id
4722 * @ss: cgroup subsys to be looked into.
4725 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4726 * NULL if not. Should be called under rcu_read_lock()
4728 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
4730 struct css_id
*cssid
= NULL
;
4732 BUG_ON(!ss
->use_id
);
4733 cssid
= idr_find(&ss
->idr
, id
);
4735 if (unlikely(!cssid
))
4738 return rcu_dereference(cssid
->css
);
4740 EXPORT_SYMBOL_GPL(css_lookup
);
4743 * css_get_next - lookup next cgroup under specified hierarchy.
4744 * @ss: pointer to subsystem
4745 * @id: current position of iteration.
4746 * @root: pointer to css. search tree under this.
4747 * @foundid: position of found object.
4749 * Search next css under the specified hierarchy of rootid. Calling under
4750 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
4752 struct cgroup_subsys_state
*
4753 css_get_next(struct cgroup_subsys
*ss
, int id
,
4754 struct cgroup_subsys_state
*root
, int *foundid
)
4756 struct cgroup_subsys_state
*ret
= NULL
;
4759 int rootid
= css_id(root
);
4760 int depth
= css_depth(root
);
4765 BUG_ON(!ss
->use_id
);
4766 /* fill start point for scan */
4770 * scan next entry from bitmap(tree), tmpid is updated after
4773 spin_lock(&ss
->id_lock
);
4774 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
4775 spin_unlock(&ss
->id_lock
);
4779 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
4780 ret
= rcu_dereference(tmp
->css
);
4786 /* continue to scan from next id */
4792 #ifdef CONFIG_CGROUP_DEBUG
4793 static struct cgroup_subsys_state
*debug_create(struct cgroup_subsys
*ss
,
4794 struct cgroup
*cont
)
4796 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
4799 return ERR_PTR(-ENOMEM
);
4804 static void debug_destroy(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4806 kfree(cont
->subsys
[debug_subsys_id
]);
4809 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
4811 return atomic_read(&cont
->count
);
4814 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
4816 return cgroup_task_count(cont
);
4819 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
4821 return (u64
)(unsigned long)current
->cgroups
;
4824 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
4830 count
= atomic_read(¤t
->cgroups
->refcount
);
4835 static int current_css_set_cg_links_read(struct cgroup
*cont
,
4837 struct seq_file
*seq
)
4839 struct cg_cgroup_link
*link
;
4842 read_lock(&css_set_lock
);
4844 cg
= rcu_dereference(current
->cgroups
);
4845 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
4846 struct cgroup
*c
= link
->cgrp
;
4850 name
= c
->dentry
->d_name
.name
;
4853 seq_printf(seq
, "Root %d group %s\n",
4854 c
->root
->hierarchy_id
, name
);
4857 read_unlock(&css_set_lock
);
4861 #define MAX_TASKS_SHOWN_PER_CSS 25
4862 static int cgroup_css_links_read(struct cgroup
*cont
,
4864 struct seq_file
*seq
)
4866 struct cg_cgroup_link
*link
;
4868 read_lock(&css_set_lock
);
4869 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
4870 struct css_set
*cg
= link
->cg
;
4871 struct task_struct
*task
;
4873 seq_printf(seq
, "css_set %p\n", cg
);
4874 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
4875 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
4876 seq_puts(seq
, " ...\n");
4879 seq_printf(seq
, " task %d\n",
4880 task_pid_vnr(task
));
4884 read_unlock(&css_set_lock
);
4888 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4890 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4893 static struct cftype debug_files
[] = {
4895 .name
= "cgroup_refcount",
4896 .read_u64
= cgroup_refcount_read
,
4899 .name
= "taskcount",
4900 .read_u64
= debug_taskcount_read
,
4904 .name
= "current_css_set",
4905 .read_u64
= current_css_set_read
,
4909 .name
= "current_css_set_refcount",
4910 .read_u64
= current_css_set_refcount_read
,
4914 .name
= "current_css_set_cg_links",
4915 .read_seq_string
= current_css_set_cg_links_read
,
4919 .name
= "cgroup_css_links",
4920 .read_seq_string
= cgroup_css_links_read
,
4924 .name
= "releasable",
4925 .read_u64
= releasable_read
,
4929 static int debug_populate(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4931 return cgroup_add_files(cont
, ss
, debug_files
,
4932 ARRAY_SIZE(debug_files
));
4935 struct cgroup_subsys debug_subsys
= {
4937 .create
= debug_create
,
4938 .destroy
= debug_destroy
,
4939 .populate
= debug_populate
,
4940 .subsys_id
= debug_subsys_id
,
4942 #endif /* CONFIG_CGROUP_DEBUG */