2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Notifications support
8 * Copyright (C) 2009 Nokia Corporation
9 * Author: Kirill A. Shutemov
11 * Copyright notices from the original cpuset code:
12 * --------------------------------------------------
13 * Copyright (C) 2003 BULL SA.
14 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
16 * Portions derived from Patrick Mochel's sysfs code.
17 * sysfs is Copyright (c) 2001-3 Patrick Mochel
19 * 2003-10-10 Written by Simon Derr.
20 * 2003-10-22 Updates by Stephen Hemminger.
21 * 2004 May-July Rework by Paul Jackson.
22 * ---------------------------------------------------
24 * This file is subject to the terms and conditions of the GNU General Public
25 * License. See the file COPYING in the main directory of the Linux
26 * distribution for more details.
29 #include <linux/cgroup.h>
30 #include <linux/cred.h>
31 #include <linux/ctype.h>
32 #include <linux/errno.h>
34 #include <linux/init_task.h>
35 #include <linux/kernel.h>
36 #include <linux/list.h>
38 #include <linux/mutex.h>
39 #include <linux/mount.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/backing-dev.h>
45 #include <linux/seq_file.h>
46 #include <linux/slab.h>
47 #include <linux/magic.h>
48 #include <linux/spinlock.h>
49 #include <linux/string.h>
50 #include <linux/sort.h>
51 #include <linux/kmod.h>
52 #include <linux/module.h>
53 #include <linux/delayacct.h>
54 #include <linux/cgroupstats.h>
55 #include <linux/hash.h>
56 #include <linux/namei.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/idr.h>
59 #include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
60 #include <linux/eventfd.h>
61 #include <linux/poll.h>
62 #include <linux/flex_array.h> /* used in cgroup_attach_proc */
64 #include <linux/atomic.h>
67 * cgroup_mutex is the master lock. Any modification to cgroup or its
68 * hierarchy must be performed while holding it.
70 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
71 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
72 * release_agent_path and so on. Modifying requires both cgroup_mutex and
73 * cgroup_root_mutex. Readers can acquire either of the two. This is to
74 * break the following locking order cycle.
76 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
77 * B. namespace_sem -> cgroup_mutex
79 * B happens only through cgroup_show_options() and using cgroup_root_mutex
82 static DEFINE_MUTEX(cgroup_mutex
);
83 static DEFINE_MUTEX(cgroup_root_mutex
);
86 * Generate an array of cgroup subsystem pointers. At boot time, this is
87 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
88 * registered after that. The mutable section of this array is protected by
91 #define SUBSYS(_x) &_x ## _subsys,
92 static struct cgroup_subsys
*subsys
[CGROUP_SUBSYS_COUNT
] = {
93 #include <linux/cgroup_subsys.h>
96 #define MAX_CGROUP_ROOT_NAMELEN 64
99 * A cgroupfs_root represents the root of a cgroup hierarchy,
100 * and may be associated with a superblock to form an active
103 struct cgroupfs_root
{
104 struct super_block
*sb
;
107 * The bitmask of subsystems intended to be attached to this
110 unsigned long subsys_bits
;
112 /* Unique id for this hierarchy. */
115 /* The bitmask of subsystems currently attached to this hierarchy */
116 unsigned long actual_subsys_bits
;
118 /* A list running through the attached subsystems */
119 struct list_head subsys_list
;
121 /* The root cgroup for this hierarchy */
122 struct cgroup top_cgroup
;
124 /* Tracks how many cgroups are currently defined in hierarchy.*/
125 int number_of_cgroups
;
127 /* A list running through the active hierarchies */
128 struct list_head root_list
;
130 /* All cgroups on this root, cgroup_mutex protected */
131 struct list_head allcg_list
;
133 /* Hierarchy-specific flags */
136 /* The path to use for release notifications. */
137 char release_agent_path
[PATH_MAX
];
139 /* The name for this hierarchy - may be empty */
140 char name
[MAX_CGROUP_ROOT_NAMELEN
];
144 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
145 * subsystems that are otherwise unattached - it never has more than a
146 * single cgroup, and all tasks are part of that cgroup.
148 static struct cgroupfs_root rootnode
;
151 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
152 * cgroup_subsys->use_id != 0.
154 #define CSS_ID_MAX (65535)
157 * The css to which this ID points. This pointer is set to valid value
158 * after cgroup is populated. If cgroup is removed, this will be NULL.
159 * This pointer is expected to be RCU-safe because destroy()
160 * is called after synchronize_rcu(). But for safe use, css_is_removed()
161 * css_tryget() should be used for avoiding race.
163 struct cgroup_subsys_state __rcu
*css
;
169 * Depth in hierarchy which this ID belongs to.
171 unsigned short depth
;
173 * ID is freed by RCU. (and lookup routine is RCU safe.)
175 struct rcu_head rcu_head
;
177 * Hierarchy of CSS ID belongs to.
179 unsigned short stack
[0]; /* Array of Length (depth+1) */
183 * cgroup_event represents events which userspace want to receive.
185 struct cgroup_event
{
187 * Cgroup which the event belongs to.
191 * Control file which the event associated.
195 * eventfd to signal userspace about the event.
197 struct eventfd_ctx
*eventfd
;
199 * Each of these stored in a list by the cgroup.
201 struct list_head list
;
203 * All fields below needed to unregister event when
204 * userspace closes eventfd.
207 wait_queue_head_t
*wqh
;
209 struct work_struct remove
;
212 /* The list of hierarchy roots */
214 static LIST_HEAD(roots
);
215 static int root_count
;
217 static DEFINE_IDA(hierarchy_ida
);
218 static int next_hierarchy_id
;
219 static DEFINE_SPINLOCK(hierarchy_id_lock
);
221 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
222 #define dummytop (&rootnode.top_cgroup)
224 /* This flag indicates whether tasks in the fork and exit paths should
225 * check for fork/exit handlers to call. This avoids us having to do
226 * extra work in the fork/exit path if none of the subsystems need to
229 static int need_forkexit_callback __read_mostly
;
231 #ifdef CONFIG_PROVE_LOCKING
232 int cgroup_lock_is_held(void)
234 return lockdep_is_held(&cgroup_mutex
);
236 #else /* #ifdef CONFIG_PROVE_LOCKING */
237 int cgroup_lock_is_held(void)
239 return mutex_is_locked(&cgroup_mutex
);
241 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
243 EXPORT_SYMBOL_GPL(cgroup_lock_is_held
);
245 /* convenient tests for these bits */
246 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
248 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
251 /* bits in struct cgroupfs_root flags field */
253 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
256 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
259 (1 << CGRP_RELEASABLE
) |
260 (1 << CGRP_NOTIFY_ON_RELEASE
);
261 return (cgrp
->flags
& bits
) == bits
;
264 static int notify_on_release(const struct cgroup
*cgrp
)
266 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
269 static int clone_children(const struct cgroup
*cgrp
)
271 return test_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
275 * for_each_subsys() allows you to iterate on each subsystem attached to
276 * an active hierarchy
278 #define for_each_subsys(_root, _ss) \
279 list_for_each_entry(_ss, &_root->subsys_list, sibling)
281 /* for_each_active_root() allows you to iterate across the active hierarchies */
282 #define for_each_active_root(_root) \
283 list_for_each_entry(_root, &roots, root_list)
285 /* the list of cgroups eligible for automatic release. Protected by
286 * release_list_lock */
287 static LIST_HEAD(release_list
);
288 static DEFINE_RAW_SPINLOCK(release_list_lock
);
289 static void cgroup_release_agent(struct work_struct
*work
);
290 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
291 static void check_for_release(struct cgroup
*cgrp
);
293 /* Link structure for associating css_set objects with cgroups */
294 struct cg_cgroup_link
{
296 * List running through cg_cgroup_links associated with a
297 * cgroup, anchored on cgroup->css_sets
299 struct list_head cgrp_link_list
;
302 * List running through cg_cgroup_links pointing at a
303 * single css_set object, anchored on css_set->cg_links
305 struct list_head cg_link_list
;
309 /* The default css_set - used by init and its children prior to any
310 * hierarchies being mounted. It contains a pointer to the root state
311 * for each subsystem. Also used to anchor the list of css_sets. Not
312 * reference-counted, to improve performance when child cgroups
313 * haven't been created.
316 static struct css_set init_css_set
;
317 static struct cg_cgroup_link init_css_set_link
;
319 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
320 struct cgroup_subsys_state
*css
);
322 /* css_set_lock protects the list of css_set objects, and the
323 * chain of tasks off each css_set. Nests outside task->alloc_lock
324 * due to cgroup_iter_start() */
325 static DEFINE_RWLOCK(css_set_lock
);
326 static int css_set_count
;
329 * hash table for cgroup groups. This improves the performance to find
330 * an existing css_set. This hash doesn't (currently) take into
331 * account cgroups in empty hierarchies.
333 #define CSS_SET_HASH_BITS 7
334 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
335 static struct hlist_head css_set_table
[CSS_SET_TABLE_SIZE
];
337 static struct hlist_head
*css_set_hash(struct cgroup_subsys_state
*css
[])
341 unsigned long tmp
= 0UL;
343 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
344 tmp
+= (unsigned long)css
[i
];
345 tmp
= (tmp
>> 16) ^ tmp
;
347 index
= hash_long(tmp
, CSS_SET_HASH_BITS
);
349 return &css_set_table
[index
];
352 /* We don't maintain the lists running through each css_set to its
353 * task until after the first call to cgroup_iter_start(). This
354 * reduces the fork()/exit() overhead for people who have cgroups
355 * compiled into their kernel but not actually in use */
356 static int use_task_css_set_links __read_mostly
;
358 static void __put_css_set(struct css_set
*cg
, int taskexit
)
360 struct cg_cgroup_link
*link
;
361 struct cg_cgroup_link
*saved_link
;
363 * Ensure that the refcount doesn't hit zero while any readers
364 * can see it. Similar to atomic_dec_and_lock(), but for an
367 if (atomic_add_unless(&cg
->refcount
, -1, 1))
369 write_lock(&css_set_lock
);
370 if (!atomic_dec_and_test(&cg
->refcount
)) {
371 write_unlock(&css_set_lock
);
375 /* This css_set is dead. unlink it and release cgroup refcounts */
376 hlist_del(&cg
->hlist
);
379 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
381 struct cgroup
*cgrp
= link
->cgrp
;
382 list_del(&link
->cg_link_list
);
383 list_del(&link
->cgrp_link_list
);
384 if (atomic_dec_and_test(&cgrp
->count
) &&
385 notify_on_release(cgrp
)) {
387 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
388 check_for_release(cgrp
);
394 write_unlock(&css_set_lock
);
395 kfree_rcu(cg
, rcu_head
);
399 * refcounted get/put for css_set objects
401 static inline void get_css_set(struct css_set
*cg
)
403 atomic_inc(&cg
->refcount
);
406 static inline void put_css_set(struct css_set
*cg
)
408 __put_css_set(cg
, 0);
411 static inline void put_css_set_taskexit(struct css_set
*cg
)
413 __put_css_set(cg
, 1);
417 * compare_css_sets - helper function for find_existing_css_set().
418 * @cg: candidate css_set being tested
419 * @old_cg: existing css_set for a task
420 * @new_cgrp: cgroup that's being entered by the task
421 * @template: desired set of css pointers in css_set (pre-calculated)
423 * Returns true if "cg" matches "old_cg" except for the hierarchy
424 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
426 static bool compare_css_sets(struct css_set
*cg
,
427 struct css_set
*old_cg
,
428 struct cgroup
*new_cgrp
,
429 struct cgroup_subsys_state
*template[])
431 struct list_head
*l1
, *l2
;
433 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
434 /* Not all subsystems matched */
439 * Compare cgroup pointers in order to distinguish between
440 * different cgroups in heirarchies with no subsystems. We
441 * could get by with just this check alone (and skip the
442 * memcmp above) but on most setups the memcmp check will
443 * avoid the need for this more expensive check on almost all
448 l2
= &old_cg
->cg_links
;
450 struct cg_cgroup_link
*cgl1
, *cgl2
;
451 struct cgroup
*cg1
, *cg2
;
455 /* See if we reached the end - both lists are equal length. */
456 if (l1
== &cg
->cg_links
) {
457 BUG_ON(l2
!= &old_cg
->cg_links
);
460 BUG_ON(l2
== &old_cg
->cg_links
);
462 /* Locate the cgroups associated with these links. */
463 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
464 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
467 /* Hierarchies should be linked in the same order. */
468 BUG_ON(cg1
->root
!= cg2
->root
);
471 * If this hierarchy is the hierarchy of the cgroup
472 * that's changing, then we need to check that this
473 * css_set points to the new cgroup; if it's any other
474 * hierarchy, then this css_set should point to the
475 * same cgroup as the old css_set.
477 if (cg1
->root
== new_cgrp
->root
) {
489 * find_existing_css_set() is a helper for
490 * find_css_set(), and checks to see whether an existing
491 * css_set is suitable.
493 * oldcg: the cgroup group that we're using before the cgroup
496 * cgrp: the cgroup that we're moving into
498 * template: location in which to build the desired set of subsystem
499 * state objects for the new cgroup group
501 static struct css_set
*find_existing_css_set(
502 struct css_set
*oldcg
,
504 struct cgroup_subsys_state
*template[])
507 struct cgroupfs_root
*root
= cgrp
->root
;
508 struct hlist_head
*hhead
;
509 struct hlist_node
*node
;
513 * Build the set of subsystem state objects that we want to see in the
514 * new css_set. while subsystems can change globally, the entries here
515 * won't change, so no need for locking.
517 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
518 if (root
->subsys_bits
& (1UL << i
)) {
519 /* Subsystem is in this hierarchy. So we want
520 * the subsystem state from the new
522 template[i
] = cgrp
->subsys
[i
];
524 /* Subsystem is not in this hierarchy, so we
525 * don't want to change the subsystem state */
526 template[i
] = oldcg
->subsys
[i
];
530 hhead
= css_set_hash(template);
531 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
532 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
535 /* This css_set matches what we need */
539 /* No existing cgroup group matched */
543 static void free_cg_links(struct list_head
*tmp
)
545 struct cg_cgroup_link
*link
;
546 struct cg_cgroup_link
*saved_link
;
548 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
549 list_del(&link
->cgrp_link_list
);
555 * allocate_cg_links() allocates "count" cg_cgroup_link structures
556 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
557 * success or a negative error
559 static int allocate_cg_links(int count
, struct list_head
*tmp
)
561 struct cg_cgroup_link
*link
;
564 for (i
= 0; i
< count
; i
++) {
565 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
570 list_add(&link
->cgrp_link_list
, tmp
);
576 * link_css_set - a helper function to link a css_set to a cgroup
577 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
578 * @cg: the css_set to be linked
579 * @cgrp: the destination cgroup
581 static void link_css_set(struct list_head
*tmp_cg_links
,
582 struct css_set
*cg
, struct cgroup
*cgrp
)
584 struct cg_cgroup_link
*link
;
586 BUG_ON(list_empty(tmp_cg_links
));
587 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
591 atomic_inc(&cgrp
->count
);
592 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
594 * Always add links to the tail of the list so that the list
595 * is sorted by order of hierarchy creation
597 list_add_tail(&link
->cg_link_list
, &cg
->cg_links
);
601 * find_css_set() takes an existing cgroup group and a
602 * cgroup object, and returns a css_set object that's
603 * equivalent to the old group, but with the given cgroup
604 * substituted into the appropriate hierarchy. Must be called with
607 static struct css_set
*find_css_set(
608 struct css_set
*oldcg
, struct cgroup
*cgrp
)
611 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
613 struct list_head tmp_cg_links
;
615 struct hlist_head
*hhead
;
616 struct cg_cgroup_link
*link
;
618 /* First see if we already have a cgroup group that matches
620 read_lock(&css_set_lock
);
621 res
= find_existing_css_set(oldcg
, cgrp
, template);
624 read_unlock(&css_set_lock
);
629 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
633 /* Allocate all the cg_cgroup_link objects that we'll need */
634 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
639 atomic_set(&res
->refcount
, 1);
640 INIT_LIST_HEAD(&res
->cg_links
);
641 INIT_LIST_HEAD(&res
->tasks
);
642 INIT_HLIST_NODE(&res
->hlist
);
644 /* Copy the set of subsystem state objects generated in
645 * find_existing_css_set() */
646 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
648 write_lock(&css_set_lock
);
649 /* Add reference counts and links from the new css_set. */
650 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
651 struct cgroup
*c
= link
->cgrp
;
652 if (c
->root
== cgrp
->root
)
654 link_css_set(&tmp_cg_links
, res
, c
);
657 BUG_ON(!list_empty(&tmp_cg_links
));
661 /* Add this cgroup group to the hash table */
662 hhead
= css_set_hash(res
->subsys
);
663 hlist_add_head(&res
->hlist
, hhead
);
665 write_unlock(&css_set_lock
);
671 * Return the cgroup for "task" from the given hierarchy. Must be
672 * called with cgroup_mutex held.
674 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
675 struct cgroupfs_root
*root
)
678 struct cgroup
*res
= NULL
;
680 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
681 read_lock(&css_set_lock
);
683 * No need to lock the task - since we hold cgroup_mutex the
684 * task can't change groups, so the only thing that can happen
685 * is that it exits and its css is set back to init_css_set.
688 if (css
== &init_css_set
) {
689 res
= &root
->top_cgroup
;
691 struct cg_cgroup_link
*link
;
692 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
693 struct cgroup
*c
= link
->cgrp
;
694 if (c
->root
== root
) {
700 read_unlock(&css_set_lock
);
706 * There is one global cgroup mutex. We also require taking
707 * task_lock() when dereferencing a task's cgroup subsys pointers.
708 * See "The task_lock() exception", at the end of this comment.
710 * A task must hold cgroup_mutex to modify cgroups.
712 * Any task can increment and decrement the count field without lock.
713 * So in general, code holding cgroup_mutex can't rely on the count
714 * field not changing. However, if the count goes to zero, then only
715 * cgroup_attach_task() can increment it again. Because a count of zero
716 * means that no tasks are currently attached, therefore there is no
717 * way a task attached to that cgroup can fork (the other way to
718 * increment the count). So code holding cgroup_mutex can safely
719 * assume that if the count is zero, it will stay zero. Similarly, if
720 * a task holds cgroup_mutex on a cgroup with zero count, it
721 * knows that the cgroup won't be removed, as cgroup_rmdir()
724 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
725 * (usually) take cgroup_mutex. These are the two most performance
726 * critical pieces of code here. The exception occurs on cgroup_exit(),
727 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
728 * is taken, and if the cgroup count is zero, a usermode call made
729 * to the release agent with the name of the cgroup (path relative to
730 * the root of cgroup file system) as the argument.
732 * A cgroup can only be deleted if both its 'count' of using tasks
733 * is zero, and its list of 'children' cgroups is empty. Since all
734 * tasks in the system use _some_ cgroup, and since there is always at
735 * least one task in the system (init, pid == 1), therefore, top_cgroup
736 * always has either children cgroups and/or using tasks. So we don't
737 * need a special hack to ensure that top_cgroup cannot be deleted.
739 * The task_lock() exception
741 * The need for this exception arises from the action of
742 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
743 * another. It does so using cgroup_mutex, however there are
744 * several performance critical places that need to reference
745 * task->cgroup without the expense of grabbing a system global
746 * mutex. Therefore except as noted below, when dereferencing or, as
747 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
748 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
749 * the task_struct routinely used for such matters.
751 * P.S. One more locking exception. RCU is used to guard the
752 * update of a tasks cgroup pointer by cgroup_attach_task()
756 * cgroup_lock - lock out any changes to cgroup structures
759 void cgroup_lock(void)
761 mutex_lock(&cgroup_mutex
);
763 EXPORT_SYMBOL_GPL(cgroup_lock
);
766 * cgroup_unlock - release lock on cgroup changes
768 * Undo the lock taken in a previous cgroup_lock() call.
770 void cgroup_unlock(void)
772 mutex_unlock(&cgroup_mutex
);
774 EXPORT_SYMBOL_GPL(cgroup_unlock
);
777 * A couple of forward declarations required, due to cyclic reference loop:
778 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
779 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
783 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
784 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
785 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
786 static int cgroup_populate_dir(struct cgroup
*cgrp
);
787 static const struct inode_operations cgroup_dir_inode_operations
;
788 static const struct file_operations proc_cgroupstats_operations
;
790 static struct backing_dev_info cgroup_backing_dev_info
= {
792 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
795 static int alloc_css_id(struct cgroup_subsys
*ss
,
796 struct cgroup
*parent
, struct cgroup
*child
);
798 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
800 struct inode
*inode
= new_inode(sb
);
803 inode
->i_ino
= get_next_ino();
804 inode
->i_mode
= mode
;
805 inode
->i_uid
= current_fsuid();
806 inode
->i_gid
= current_fsgid();
807 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
808 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
814 * Call subsys's pre_destroy handler.
815 * This is called before css refcnt check.
817 static int cgroup_call_pre_destroy(struct cgroup
*cgrp
)
819 struct cgroup_subsys
*ss
;
822 for_each_subsys(cgrp
->root
, ss
)
823 if (ss
->pre_destroy
) {
824 ret
= ss
->pre_destroy(cgrp
);
832 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
834 /* is dentry a directory ? if so, kfree() associated cgroup */
835 if (S_ISDIR(inode
->i_mode
)) {
836 struct cgroup
*cgrp
= dentry
->d_fsdata
;
837 struct cgroup_subsys
*ss
;
838 BUG_ON(!(cgroup_is_removed(cgrp
)));
839 /* It's possible for external users to be holding css
840 * reference counts on a cgroup; css_put() needs to
841 * be able to access the cgroup after decrementing
842 * the reference count in order to know if it needs to
843 * queue the cgroup to be handled by the release
847 mutex_lock(&cgroup_mutex
);
849 * Release the subsystem state objects.
851 for_each_subsys(cgrp
->root
, ss
)
854 cgrp
->root
->number_of_cgroups
--;
855 mutex_unlock(&cgroup_mutex
);
858 * Drop the active superblock reference that we took when we
861 deactivate_super(cgrp
->root
->sb
);
864 * if we're getting rid of the cgroup, refcount should ensure
865 * that there are no pidlists left.
867 BUG_ON(!list_empty(&cgrp
->pidlists
));
869 kfree_rcu(cgrp
, rcu_head
);
874 static int cgroup_delete(const struct dentry
*d
)
879 static void remove_dir(struct dentry
*d
)
881 struct dentry
*parent
= dget(d
->d_parent
);
884 simple_rmdir(parent
->d_inode
, d
);
888 static void cgroup_clear_directory(struct dentry
*dentry
)
890 struct list_head
*node
;
892 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
893 spin_lock(&dentry
->d_lock
);
894 node
= dentry
->d_subdirs
.next
;
895 while (node
!= &dentry
->d_subdirs
) {
896 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
898 spin_lock_nested(&d
->d_lock
, DENTRY_D_LOCK_NESTED
);
901 /* This should never be called on a cgroup
902 * directory with child cgroups */
903 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
905 spin_unlock(&d
->d_lock
);
906 spin_unlock(&dentry
->d_lock
);
908 simple_unlink(dentry
->d_inode
, d
);
910 spin_lock(&dentry
->d_lock
);
912 spin_unlock(&d
->d_lock
);
913 node
= dentry
->d_subdirs
.next
;
915 spin_unlock(&dentry
->d_lock
);
919 * NOTE : the dentry must have been dget()'ed
921 static void cgroup_d_remove_dir(struct dentry
*dentry
)
923 struct dentry
*parent
;
925 cgroup_clear_directory(dentry
);
927 parent
= dentry
->d_parent
;
928 spin_lock(&parent
->d_lock
);
929 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
930 list_del_init(&dentry
->d_u
.d_child
);
931 spin_unlock(&dentry
->d_lock
);
932 spin_unlock(&parent
->d_lock
);
937 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
938 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
939 * reference to css->refcnt. In general, this refcnt is expected to goes down
942 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
944 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq
);
946 static void cgroup_wakeup_rmdir_waiter(struct cgroup
*cgrp
)
948 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
)))
949 wake_up_all(&cgroup_rmdir_waitq
);
952 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
)
957 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
)
959 cgroup_wakeup_rmdir_waiter(css
->cgroup
);
964 * Call with cgroup_mutex held. Drops reference counts on modules, including
965 * any duplicate ones that parse_cgroupfs_options took. If this function
966 * returns an error, no reference counts are touched.
968 static int rebind_subsystems(struct cgroupfs_root
*root
,
969 unsigned long final_bits
)
971 unsigned long added_bits
, removed_bits
;
972 struct cgroup
*cgrp
= &root
->top_cgroup
;
975 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
976 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
978 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
979 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
980 /* Check that any added subsystems are currently free */
981 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
982 unsigned long bit
= 1UL << i
;
983 struct cgroup_subsys
*ss
= subsys
[i
];
984 if (!(bit
& added_bits
))
987 * Nobody should tell us to do a subsys that doesn't exist:
988 * parse_cgroupfs_options should catch that case and refcounts
989 * ensure that subsystems won't disappear once selected.
992 if (ss
->root
!= &rootnode
) {
993 /* Subsystem isn't free */
998 /* Currently we don't handle adding/removing subsystems when
999 * any child cgroups exist. This is theoretically supportable
1000 * but involves complex error handling, so it's being left until
1002 if (root
->number_of_cgroups
> 1)
1005 /* Process each subsystem */
1006 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1007 struct cgroup_subsys
*ss
= subsys
[i
];
1008 unsigned long bit
= 1UL << i
;
1009 if (bit
& added_bits
) {
1010 /* We're binding this subsystem to this hierarchy */
1012 BUG_ON(cgrp
->subsys
[i
]);
1013 BUG_ON(!dummytop
->subsys
[i
]);
1014 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
1015 mutex_lock(&ss
->hierarchy_mutex
);
1016 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
1017 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1018 list_move(&ss
->sibling
, &root
->subsys_list
);
1022 mutex_unlock(&ss
->hierarchy_mutex
);
1023 /* refcount was already taken, and we're keeping it */
1024 } else if (bit
& removed_bits
) {
1025 /* We're removing this subsystem */
1027 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1028 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1029 mutex_lock(&ss
->hierarchy_mutex
);
1032 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1033 cgrp
->subsys
[i
] = NULL
;
1034 subsys
[i
]->root
= &rootnode
;
1035 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1036 mutex_unlock(&ss
->hierarchy_mutex
);
1037 /* subsystem is now free - drop reference on module */
1038 module_put(ss
->module
);
1039 } else if (bit
& final_bits
) {
1040 /* Subsystem state should already exist */
1042 BUG_ON(!cgrp
->subsys
[i
]);
1044 * a refcount was taken, but we already had one, so
1045 * drop the extra reference.
1047 module_put(ss
->module
);
1048 #ifdef CONFIG_MODULE_UNLOAD
1049 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1052 /* Subsystem state shouldn't exist */
1053 BUG_ON(cgrp
->subsys
[i
]);
1056 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
1062 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1064 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1065 struct cgroup_subsys
*ss
;
1067 mutex_lock(&cgroup_root_mutex
);
1068 for_each_subsys(root
, ss
)
1069 seq_printf(seq
, ",%s", ss
->name
);
1070 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1071 seq_puts(seq
, ",noprefix");
1072 if (strlen(root
->release_agent_path
))
1073 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1074 if (clone_children(&root
->top_cgroup
))
1075 seq_puts(seq
, ",clone_children");
1076 if (strlen(root
->name
))
1077 seq_printf(seq
, ",name=%s", root
->name
);
1078 mutex_unlock(&cgroup_root_mutex
);
1082 struct cgroup_sb_opts
{
1083 unsigned long subsys_bits
;
1084 unsigned long flags
;
1085 char *release_agent
;
1086 bool clone_children
;
1088 /* User explicitly requested empty subsystem */
1091 struct cgroupfs_root
*new_root
;
1096 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1097 * with cgroup_mutex held to protect the subsys[] array. This function takes
1098 * refcounts on subsystems to be used, unless it returns error, in which case
1099 * no refcounts are taken.
1101 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1103 char *token
, *o
= data
;
1104 bool all_ss
= false, one_ss
= false;
1105 unsigned long mask
= (unsigned long)-1;
1107 bool module_pin_failed
= false;
1109 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1111 #ifdef CONFIG_CPUSETS
1112 mask
= ~(1UL << cpuset_subsys_id
);
1115 memset(opts
, 0, sizeof(*opts
));
1117 while ((token
= strsep(&o
, ",")) != NULL
) {
1120 if (!strcmp(token
, "none")) {
1121 /* Explicitly have no subsystems */
1125 if (!strcmp(token
, "all")) {
1126 /* Mutually exclusive option 'all' + subsystem name */
1132 if (!strcmp(token
, "noprefix")) {
1133 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1136 if (!strcmp(token
, "clone_children")) {
1137 opts
->clone_children
= true;
1140 if (!strncmp(token
, "release_agent=", 14)) {
1141 /* Specifying two release agents is forbidden */
1142 if (opts
->release_agent
)
1144 opts
->release_agent
=
1145 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1146 if (!opts
->release_agent
)
1150 if (!strncmp(token
, "name=", 5)) {
1151 const char *name
= token
+ 5;
1152 /* Can't specify an empty name */
1155 /* Must match [\w.-]+ */
1156 for (i
= 0; i
< strlen(name
); i
++) {
1160 if ((c
== '.') || (c
== '-') || (c
== '_'))
1164 /* Specifying two names is forbidden */
1167 opts
->name
= kstrndup(name
,
1168 MAX_CGROUP_ROOT_NAMELEN
- 1,
1176 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1177 struct cgroup_subsys
*ss
= subsys
[i
];
1180 if (strcmp(token
, ss
->name
))
1185 /* Mutually exclusive option 'all' + subsystem name */
1188 set_bit(i
, &opts
->subsys_bits
);
1193 if (i
== CGROUP_SUBSYS_COUNT
)
1198 * If the 'all' option was specified select all the subsystems,
1199 * otherwise if 'none', 'name=' and a subsystem name options
1200 * were not specified, let's default to 'all'
1202 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
)) {
1203 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1204 struct cgroup_subsys
*ss
= subsys
[i
];
1209 set_bit(i
, &opts
->subsys_bits
);
1213 /* Consistency checks */
1216 * Option noprefix was introduced just for backward compatibility
1217 * with the old cpuset, so we allow noprefix only if mounting just
1218 * the cpuset subsystem.
1220 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1221 (opts
->subsys_bits
& mask
))
1225 /* Can't specify "none" and some subsystems */
1226 if (opts
->subsys_bits
&& opts
->none
)
1230 * We either have to specify by name or by subsystems. (So all
1231 * empty hierarchies must have a name).
1233 if (!opts
->subsys_bits
&& !opts
->name
)
1237 * Grab references on all the modules we'll need, so the subsystems
1238 * don't dance around before rebind_subsystems attaches them. This may
1239 * take duplicate reference counts on a subsystem that's already used,
1240 * but rebind_subsystems handles this case.
1242 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1243 unsigned long bit
= 1UL << i
;
1245 if (!(bit
& opts
->subsys_bits
))
1247 if (!try_module_get(subsys
[i
]->module
)) {
1248 module_pin_failed
= true;
1252 if (module_pin_failed
) {
1254 * oops, one of the modules was going away. this means that we
1255 * raced with a module_delete call, and to the user this is
1256 * essentially a "subsystem doesn't exist" case.
1258 for (i
--; i
>= CGROUP_BUILTIN_SUBSYS_COUNT
; i
--) {
1259 /* drop refcounts only on the ones we took */
1260 unsigned long bit
= 1UL << i
;
1262 if (!(bit
& opts
->subsys_bits
))
1264 module_put(subsys
[i
]->module
);
1272 static void drop_parsed_module_refcounts(unsigned long subsys_bits
)
1275 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1276 unsigned long bit
= 1UL << i
;
1278 if (!(bit
& subsys_bits
))
1280 module_put(subsys
[i
]->module
);
1284 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1287 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1288 struct cgroup
*cgrp
= &root
->top_cgroup
;
1289 struct cgroup_sb_opts opts
;
1291 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1292 mutex_lock(&cgroup_mutex
);
1293 mutex_lock(&cgroup_root_mutex
);
1295 /* See what subsystems are wanted */
1296 ret
= parse_cgroupfs_options(data
, &opts
);
1300 /* See feature-removal-schedule.txt */
1301 if (opts
.subsys_bits
!= root
->actual_subsys_bits
|| opts
.release_agent
)
1302 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1303 task_tgid_nr(current
), current
->comm
);
1305 /* Don't allow flags or name to change at remount */
1306 if (opts
.flags
!= root
->flags
||
1307 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1309 drop_parsed_module_refcounts(opts
.subsys_bits
);
1313 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
1315 drop_parsed_module_refcounts(opts
.subsys_bits
);
1319 /* clear out any existing files and repopulate subsystem files */
1320 cgroup_clear_directory(cgrp
->dentry
);
1321 cgroup_populate_dir(cgrp
);
1323 if (opts
.release_agent
)
1324 strcpy(root
->release_agent_path
, opts
.release_agent
);
1326 kfree(opts
.release_agent
);
1328 mutex_unlock(&cgroup_root_mutex
);
1329 mutex_unlock(&cgroup_mutex
);
1330 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1334 static const struct super_operations cgroup_ops
= {
1335 .statfs
= simple_statfs
,
1336 .drop_inode
= generic_delete_inode
,
1337 .show_options
= cgroup_show_options
,
1338 .remount_fs
= cgroup_remount
,
1341 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1343 INIT_LIST_HEAD(&cgrp
->sibling
);
1344 INIT_LIST_HEAD(&cgrp
->children
);
1345 INIT_LIST_HEAD(&cgrp
->css_sets
);
1346 INIT_LIST_HEAD(&cgrp
->release_list
);
1347 INIT_LIST_HEAD(&cgrp
->pidlists
);
1348 mutex_init(&cgrp
->pidlist_mutex
);
1349 INIT_LIST_HEAD(&cgrp
->event_list
);
1350 spin_lock_init(&cgrp
->event_list_lock
);
1353 static void init_cgroup_root(struct cgroupfs_root
*root
)
1355 struct cgroup
*cgrp
= &root
->top_cgroup
;
1357 INIT_LIST_HEAD(&root
->subsys_list
);
1358 INIT_LIST_HEAD(&root
->root_list
);
1359 INIT_LIST_HEAD(&root
->allcg_list
);
1360 root
->number_of_cgroups
= 1;
1362 cgrp
->top_cgroup
= cgrp
;
1363 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
1364 init_cgroup_housekeeping(cgrp
);
1367 static bool init_root_id(struct cgroupfs_root
*root
)
1372 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1374 spin_lock(&hierarchy_id_lock
);
1375 /* Try to allocate the next unused ID */
1376 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1377 &root
->hierarchy_id
);
1379 /* Try again starting from 0 */
1380 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1382 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1383 } else if (ret
!= -EAGAIN
) {
1384 /* Can only get here if the 31-bit IDR is full ... */
1387 spin_unlock(&hierarchy_id_lock
);
1392 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1394 struct cgroup_sb_opts
*opts
= data
;
1395 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1397 /* If we asked for a name then it must match */
1398 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1402 * If we asked for subsystems (or explicitly for no
1403 * subsystems) then they must match
1405 if ((opts
->subsys_bits
|| opts
->none
)
1406 && (opts
->subsys_bits
!= root
->subsys_bits
))
1412 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1414 struct cgroupfs_root
*root
;
1416 if (!opts
->subsys_bits
&& !opts
->none
)
1419 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1421 return ERR_PTR(-ENOMEM
);
1423 if (!init_root_id(root
)) {
1425 return ERR_PTR(-ENOMEM
);
1427 init_cgroup_root(root
);
1429 root
->subsys_bits
= opts
->subsys_bits
;
1430 root
->flags
= opts
->flags
;
1431 if (opts
->release_agent
)
1432 strcpy(root
->release_agent_path
, opts
->release_agent
);
1434 strcpy(root
->name
, opts
->name
);
1435 if (opts
->clone_children
)
1436 set_bit(CGRP_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1440 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1445 BUG_ON(!root
->hierarchy_id
);
1446 spin_lock(&hierarchy_id_lock
);
1447 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1448 spin_unlock(&hierarchy_id_lock
);
1452 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1455 struct cgroup_sb_opts
*opts
= data
;
1457 /* If we don't have a new root, we can't set up a new sb */
1458 if (!opts
->new_root
)
1461 BUG_ON(!opts
->subsys_bits
&& !opts
->none
);
1463 ret
= set_anon_super(sb
, NULL
);
1467 sb
->s_fs_info
= opts
->new_root
;
1468 opts
->new_root
->sb
= sb
;
1470 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1471 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1472 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1473 sb
->s_op
= &cgroup_ops
;
1478 static int cgroup_get_rootdir(struct super_block
*sb
)
1480 static const struct dentry_operations cgroup_dops
= {
1481 .d_iput
= cgroup_diput
,
1482 .d_delete
= cgroup_delete
,
1485 struct inode
*inode
=
1486 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1491 inode
->i_fop
= &simple_dir_operations
;
1492 inode
->i_op
= &cgroup_dir_inode_operations
;
1493 /* directories start off with i_nlink == 2 (for "." entry) */
1495 sb
->s_root
= d_make_root(inode
);
1498 /* for everything else we want ->d_op set */
1499 sb
->s_d_op
= &cgroup_dops
;
1503 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1504 int flags
, const char *unused_dev_name
,
1507 struct cgroup_sb_opts opts
;
1508 struct cgroupfs_root
*root
;
1510 struct super_block
*sb
;
1511 struct cgroupfs_root
*new_root
;
1512 struct inode
*inode
;
1514 /* First find the desired set of subsystems */
1515 mutex_lock(&cgroup_mutex
);
1516 ret
= parse_cgroupfs_options(data
, &opts
);
1517 mutex_unlock(&cgroup_mutex
);
1522 * Allocate a new cgroup root. We may not need it if we're
1523 * reusing an existing hierarchy.
1525 new_root
= cgroup_root_from_opts(&opts
);
1526 if (IS_ERR(new_root
)) {
1527 ret
= PTR_ERR(new_root
);
1530 opts
.new_root
= new_root
;
1532 /* Locate an existing or new sb for this hierarchy */
1533 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, &opts
);
1536 cgroup_drop_root(opts
.new_root
);
1540 root
= sb
->s_fs_info
;
1542 if (root
== opts
.new_root
) {
1543 /* We used the new root structure, so this is a new hierarchy */
1544 struct list_head tmp_cg_links
;
1545 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1546 struct cgroupfs_root
*existing_root
;
1547 const struct cred
*cred
;
1550 BUG_ON(sb
->s_root
!= NULL
);
1552 ret
= cgroup_get_rootdir(sb
);
1554 goto drop_new_super
;
1555 inode
= sb
->s_root
->d_inode
;
1557 mutex_lock(&inode
->i_mutex
);
1558 mutex_lock(&cgroup_mutex
);
1559 mutex_lock(&cgroup_root_mutex
);
1561 /* Check for name clashes with existing mounts */
1563 if (strlen(root
->name
))
1564 for_each_active_root(existing_root
)
1565 if (!strcmp(existing_root
->name
, root
->name
))
1569 * We're accessing css_set_count without locking
1570 * css_set_lock here, but that's OK - it can only be
1571 * increased by someone holding cgroup_lock, and
1572 * that's us. The worst that can happen is that we
1573 * have some link structures left over
1575 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1579 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1580 if (ret
== -EBUSY
) {
1581 free_cg_links(&tmp_cg_links
);
1585 * There must be no failure case after here, since rebinding
1586 * takes care of subsystems' refcounts, which are explicitly
1587 * dropped in the failure exit path.
1590 /* EBUSY should be the only error here */
1593 list_add(&root
->root_list
, &roots
);
1596 sb
->s_root
->d_fsdata
= root_cgrp
;
1597 root
->top_cgroup
.dentry
= sb
->s_root
;
1599 /* Link the top cgroup in this hierarchy into all
1600 * the css_set objects */
1601 write_lock(&css_set_lock
);
1602 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1603 struct hlist_head
*hhead
= &css_set_table
[i
];
1604 struct hlist_node
*node
;
1607 hlist_for_each_entry(cg
, node
, hhead
, hlist
)
1608 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1610 write_unlock(&css_set_lock
);
1612 free_cg_links(&tmp_cg_links
);
1614 BUG_ON(!list_empty(&root_cgrp
->sibling
));
1615 BUG_ON(!list_empty(&root_cgrp
->children
));
1616 BUG_ON(root
->number_of_cgroups
!= 1);
1618 cred
= override_creds(&init_cred
);
1619 cgroup_populate_dir(root_cgrp
);
1621 mutex_unlock(&cgroup_root_mutex
);
1622 mutex_unlock(&cgroup_mutex
);
1623 mutex_unlock(&inode
->i_mutex
);
1626 * We re-used an existing hierarchy - the new root (if
1627 * any) is not needed
1629 cgroup_drop_root(opts
.new_root
);
1630 /* no subsys rebinding, so refcounts don't change */
1631 drop_parsed_module_refcounts(opts
.subsys_bits
);
1634 kfree(opts
.release_agent
);
1636 return dget(sb
->s_root
);
1639 mutex_unlock(&cgroup_root_mutex
);
1640 mutex_unlock(&cgroup_mutex
);
1641 mutex_unlock(&inode
->i_mutex
);
1643 deactivate_locked_super(sb
);
1645 drop_parsed_module_refcounts(opts
.subsys_bits
);
1647 kfree(opts
.release_agent
);
1649 return ERR_PTR(ret
);
1652 static void cgroup_kill_sb(struct super_block
*sb
) {
1653 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1654 struct cgroup
*cgrp
= &root
->top_cgroup
;
1656 struct cg_cgroup_link
*link
;
1657 struct cg_cgroup_link
*saved_link
;
1661 BUG_ON(root
->number_of_cgroups
!= 1);
1662 BUG_ON(!list_empty(&cgrp
->children
));
1663 BUG_ON(!list_empty(&cgrp
->sibling
));
1665 mutex_lock(&cgroup_mutex
);
1666 mutex_lock(&cgroup_root_mutex
);
1668 /* Rebind all subsystems back to the default hierarchy */
1669 ret
= rebind_subsystems(root
, 0);
1670 /* Shouldn't be able to fail ... */
1674 * Release all the links from css_sets to this hierarchy's
1677 write_lock(&css_set_lock
);
1679 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1681 list_del(&link
->cg_link_list
);
1682 list_del(&link
->cgrp_link_list
);
1685 write_unlock(&css_set_lock
);
1687 if (!list_empty(&root
->root_list
)) {
1688 list_del(&root
->root_list
);
1692 mutex_unlock(&cgroup_root_mutex
);
1693 mutex_unlock(&cgroup_mutex
);
1695 kill_litter_super(sb
);
1696 cgroup_drop_root(root
);
1699 static struct file_system_type cgroup_fs_type
= {
1701 .mount
= cgroup_mount
,
1702 .kill_sb
= cgroup_kill_sb
,
1705 static struct kobject
*cgroup_kobj
;
1707 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1709 return dentry
->d_fsdata
;
1712 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1714 return dentry
->d_fsdata
;
1718 * cgroup_path - generate the path of a cgroup
1719 * @cgrp: the cgroup in question
1720 * @buf: the buffer to write the path into
1721 * @buflen: the length of the buffer
1723 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1724 * reference. Writes path of cgroup into buf. Returns 0 on success,
1727 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1730 struct dentry
*dentry
= rcu_dereference_check(cgrp
->dentry
,
1731 cgroup_lock_is_held());
1733 if (!dentry
|| cgrp
== dummytop
) {
1735 * Inactive subsystems have no dentry for their root
1742 start
= buf
+ buflen
;
1746 int len
= dentry
->d_name
.len
;
1748 if ((start
-= len
) < buf
)
1749 return -ENAMETOOLONG
;
1750 memcpy(start
, dentry
->d_name
.name
, len
);
1751 cgrp
= cgrp
->parent
;
1755 dentry
= rcu_dereference_check(cgrp
->dentry
,
1756 cgroup_lock_is_held());
1760 return -ENAMETOOLONG
;
1763 memmove(buf
, start
, buf
+ buflen
- start
);
1766 EXPORT_SYMBOL_GPL(cgroup_path
);
1769 * Control Group taskset
1771 struct task_and_cgroup
{
1772 struct task_struct
*task
;
1773 struct cgroup
*cgrp
;
1777 struct cgroup_taskset
{
1778 struct task_and_cgroup single
;
1779 struct flex_array
*tc_array
;
1782 struct cgroup
*cur_cgrp
;
1786 * cgroup_taskset_first - reset taskset and return the first task
1787 * @tset: taskset of interest
1789 * @tset iteration is initialized and the first task is returned.
1791 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1793 if (tset
->tc_array
) {
1795 return cgroup_taskset_next(tset
);
1797 tset
->cur_cgrp
= tset
->single
.cgrp
;
1798 return tset
->single
.task
;
1801 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1804 * cgroup_taskset_next - iterate to the next task in taskset
1805 * @tset: taskset of interest
1807 * Return the next task in @tset. Iteration must have been initialized
1808 * with cgroup_taskset_first().
1810 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1812 struct task_and_cgroup
*tc
;
1814 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1817 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1818 tset
->cur_cgrp
= tc
->cgrp
;
1821 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1824 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1825 * @tset: taskset of interest
1827 * Return the cgroup for the current (last returned) task of @tset. This
1828 * function must be preceded by either cgroup_taskset_first() or
1829 * cgroup_taskset_next().
1831 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1833 return tset
->cur_cgrp
;
1835 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1838 * cgroup_taskset_size - return the number of tasks in taskset
1839 * @tset: taskset of interest
1841 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1843 return tset
->tc_array
? tset
->tc_array_len
: 1;
1845 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1849 * cgroup_task_migrate - move a task from one cgroup to another.
1851 * 'guarantee' is set if the caller promises that a new css_set for the task
1852 * will already exist. If not set, this function might sleep, and can fail with
1853 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1855 static void cgroup_task_migrate(struct cgroup
*cgrp
, struct cgroup
*oldcgrp
,
1856 struct task_struct
*tsk
, struct css_set
*newcg
)
1858 struct css_set
*oldcg
;
1861 * We are synchronized through threadgroup_lock() against PF_EXITING
1862 * setting such that we can't race against cgroup_exit() changing the
1863 * css_set to init_css_set and dropping the old one.
1865 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1866 oldcg
= tsk
->cgroups
;
1869 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1872 /* Update the css_set linked lists if we're using them */
1873 write_lock(&css_set_lock
);
1874 if (!list_empty(&tsk
->cg_list
))
1875 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1876 write_unlock(&css_set_lock
);
1879 * We just gained a reference on oldcg by taking it from the task. As
1880 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1881 * it here; it will be freed under RCU.
1885 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1889 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1890 * @cgrp: the cgroup the task is attaching to
1891 * @tsk: the task to be attached
1893 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1896 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1899 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1900 struct cgroup
*oldcgrp
;
1901 struct cgroupfs_root
*root
= cgrp
->root
;
1902 struct cgroup_taskset tset
= { };
1903 struct css_set
*newcg
;
1905 /* @tsk either already exited or can't exit until the end */
1906 if (tsk
->flags
& PF_EXITING
)
1909 /* Nothing to do if the task is already in that cgroup */
1910 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1911 if (cgrp
== oldcgrp
)
1914 tset
.single
.task
= tsk
;
1915 tset
.single
.cgrp
= oldcgrp
;
1917 for_each_subsys(root
, ss
) {
1918 if (ss
->can_attach
) {
1919 retval
= ss
->can_attach(cgrp
, &tset
);
1922 * Remember on which subsystem the can_attach()
1923 * failed, so that we only call cancel_attach()
1924 * against the subsystems whose can_attach()
1925 * succeeded. (See below)
1933 newcg
= find_css_set(tsk
->cgroups
, cgrp
);
1939 cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, newcg
);
1941 for_each_subsys(root
, ss
) {
1943 ss
->attach(cgrp
, &tset
);
1949 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1950 * is no longer empty.
1952 cgroup_wakeup_rmdir_waiter(cgrp
);
1955 for_each_subsys(root
, ss
) {
1956 if (ss
== failed_ss
)
1958 * This subsystem was the one that failed the
1959 * can_attach() check earlier, so we don't need
1960 * to call cancel_attach() against it or any
1961 * remaining subsystems.
1964 if (ss
->cancel_attach
)
1965 ss
->cancel_attach(cgrp
, &tset
);
1972 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1973 * @from: attach to all cgroups of a given task
1974 * @tsk: the task to be attached
1976 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
1978 struct cgroupfs_root
*root
;
1982 for_each_active_root(root
) {
1983 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
1985 retval
= cgroup_attach_task(from_cg
, tsk
);
1993 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
1996 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
1997 * @cgrp: the cgroup to attach to
1998 * @leader: the threadgroup leader task_struct of the group to be attached
2000 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2001 * task_lock of each thread in leader's threadgroup individually in turn.
2003 static int cgroup_attach_proc(struct cgroup
*cgrp
, struct task_struct
*leader
)
2005 int retval
, i
, group_size
;
2006 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
2007 /* guaranteed to be initialized later, but the compiler needs this */
2008 struct cgroupfs_root
*root
= cgrp
->root
;
2009 /* threadgroup list cursor and array */
2010 struct task_struct
*tsk
;
2011 struct task_and_cgroup
*tc
;
2012 struct flex_array
*group
;
2013 struct cgroup_taskset tset
= { };
2016 * step 0: in order to do expensive, possibly blocking operations for
2017 * every thread, we cannot iterate the thread group list, since it needs
2018 * rcu or tasklist locked. instead, build an array of all threads in the
2019 * group - group_rwsem prevents new threads from appearing, and if
2020 * threads exit, this will just be an over-estimate.
2022 group_size
= get_nr_threads(leader
);
2023 /* flex_array supports very large thread-groups better than kmalloc. */
2024 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2027 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2028 retval
= flex_array_prealloc(group
, 0, group_size
- 1, GFP_KERNEL
);
2030 goto out_free_group_list
;
2035 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2036 * already PF_EXITING could be freed from underneath us unless we
2037 * take an rcu_read_lock.
2041 struct task_and_cgroup ent
;
2043 /* @tsk either already exited or can't exit until the end */
2044 if (tsk
->flags
& PF_EXITING
)
2047 /* as per above, nr_threads may decrease, but not increase. */
2048 BUG_ON(i
>= group_size
);
2050 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2051 /* nothing to do if this task is already in the cgroup */
2052 if (ent
.cgrp
== cgrp
)
2055 * saying GFP_ATOMIC has no effect here because we did prealloc
2056 * earlier, but it's good form to communicate our expectations.
2058 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2059 BUG_ON(retval
!= 0);
2061 } while_each_thread(leader
, tsk
);
2063 /* remember the number of threads in the array for later. */
2065 tset
.tc_array
= group
;
2066 tset
.tc_array_len
= group_size
;
2068 /* methods shouldn't be called if no task is actually migrating */
2071 goto out_free_group_list
;
2074 * step 1: check that we can legitimately attach to the cgroup.
2076 for_each_subsys(root
, ss
) {
2077 if (ss
->can_attach
) {
2078 retval
= ss
->can_attach(cgrp
, &tset
);
2081 goto out_cancel_attach
;
2087 * step 2: make sure css_sets exist for all threads to be migrated.
2088 * we use find_css_set, which allocates a new one if necessary.
2090 for (i
= 0; i
< group_size
; i
++) {
2091 tc
= flex_array_get(group
, i
);
2092 tc
->cg
= find_css_set(tc
->task
->cgroups
, cgrp
);
2095 goto out_put_css_set_refs
;
2100 * step 3: now that we're guaranteed success wrt the css_sets,
2101 * proceed to move all tasks to the new cgroup. There are no
2102 * failure cases after here, so this is the commit point.
2104 for (i
= 0; i
< group_size
; i
++) {
2105 tc
= flex_array_get(group
, i
);
2106 cgroup_task_migrate(cgrp
, tc
->cgrp
, tc
->task
, tc
->cg
);
2108 /* nothing is sensitive to fork() after this point. */
2111 * step 4: do subsystem attach callbacks.
2113 for_each_subsys(root
, ss
) {
2115 ss
->attach(cgrp
, &tset
);
2119 * step 5: success! and cleanup
2122 cgroup_wakeup_rmdir_waiter(cgrp
);
2124 out_put_css_set_refs
:
2126 for (i
= 0; i
< group_size
; i
++) {
2127 tc
= flex_array_get(group
, i
);
2130 put_css_set(tc
->cg
);
2135 for_each_subsys(root
, ss
) {
2136 if (ss
== failed_ss
)
2138 if (ss
->cancel_attach
)
2139 ss
->cancel_attach(cgrp
, &tset
);
2142 out_free_group_list
:
2143 flex_array_free(group
);
2148 * Find the task_struct of the task to attach by vpid and pass it along to the
2149 * function to attach either it or all tasks in its threadgroup. Will lock
2150 * cgroup_mutex and threadgroup; may take task_lock of task.
2152 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2154 struct task_struct
*tsk
;
2155 const struct cred
*cred
= current_cred(), *tcred
;
2158 if (!cgroup_lock_live_group(cgrp
))
2164 tsk
= find_task_by_vpid(pid
);
2168 goto out_unlock_cgroup
;
2171 * even if we're attaching all tasks in the thread group, we
2172 * only need to check permissions on one of them.
2174 tcred
= __task_cred(tsk
);
2176 cred
->euid
!= tcred
->uid
&&
2177 cred
->euid
!= tcred
->suid
) {
2180 goto out_unlock_cgroup
;
2186 tsk
= tsk
->group_leader
;
2187 get_task_struct(tsk
);
2190 threadgroup_lock(tsk
);
2192 if (!thread_group_leader(tsk
)) {
2194 * a race with de_thread from another thread's exec()
2195 * may strip us of our leadership, if this happens,
2196 * there is no choice but to throw this task away and
2197 * try again; this is
2198 * "double-double-toil-and-trouble-check locking".
2200 threadgroup_unlock(tsk
);
2201 put_task_struct(tsk
);
2202 goto retry_find_task
;
2204 ret
= cgroup_attach_proc(cgrp
, tsk
);
2206 ret
= cgroup_attach_task(cgrp
, tsk
);
2207 threadgroup_unlock(tsk
);
2209 put_task_struct(tsk
);
2215 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2217 return attach_task_by_pid(cgrp
, pid
, false);
2220 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2222 return attach_task_by_pid(cgrp
, tgid
, true);
2226 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2227 * @cgrp: the cgroup to be checked for liveness
2229 * On success, returns true; the lock should be later released with
2230 * cgroup_unlock(). On failure returns false with no lock held.
2232 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
2234 mutex_lock(&cgroup_mutex
);
2235 if (cgroup_is_removed(cgrp
)) {
2236 mutex_unlock(&cgroup_mutex
);
2241 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
2243 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2246 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2247 if (strlen(buffer
) >= PATH_MAX
)
2249 if (!cgroup_lock_live_group(cgrp
))
2251 mutex_lock(&cgroup_root_mutex
);
2252 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2253 mutex_unlock(&cgroup_root_mutex
);
2258 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2259 struct seq_file
*seq
)
2261 if (!cgroup_lock_live_group(cgrp
))
2263 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2264 seq_putc(seq
, '\n');
2269 /* A buffer size big enough for numbers or short strings */
2270 #define CGROUP_LOCAL_BUFFER_SIZE 64
2272 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2274 const char __user
*userbuf
,
2275 size_t nbytes
, loff_t
*unused_ppos
)
2277 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2283 if (nbytes
>= sizeof(buffer
))
2285 if (copy_from_user(buffer
, userbuf
, nbytes
))
2288 buffer
[nbytes
] = 0; /* nul-terminate */
2289 if (cft
->write_u64
) {
2290 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2293 retval
= cft
->write_u64(cgrp
, cft
, val
);
2295 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2298 retval
= cft
->write_s64(cgrp
, cft
, val
);
2305 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2307 const char __user
*userbuf
,
2308 size_t nbytes
, loff_t
*unused_ppos
)
2310 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2312 size_t max_bytes
= cft
->max_write_len
;
2313 char *buffer
= local_buffer
;
2316 max_bytes
= sizeof(local_buffer
) - 1;
2317 if (nbytes
>= max_bytes
)
2319 /* Allocate a dynamic buffer if we need one */
2320 if (nbytes
>= sizeof(local_buffer
)) {
2321 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2325 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2330 buffer
[nbytes
] = 0; /* nul-terminate */
2331 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2335 if (buffer
!= local_buffer
)
2340 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2341 size_t nbytes
, loff_t
*ppos
)
2343 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2344 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2346 if (cgroup_is_removed(cgrp
))
2349 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2350 if (cft
->write_u64
|| cft
->write_s64
)
2351 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2352 if (cft
->write_string
)
2353 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2355 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2356 return ret
? ret
: nbytes
;
2361 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2363 char __user
*buf
, size_t nbytes
,
2366 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2367 u64 val
= cft
->read_u64(cgrp
, cft
);
2368 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2370 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2373 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2375 char __user
*buf
, size_t nbytes
,
2378 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2379 s64 val
= cft
->read_s64(cgrp
, cft
);
2380 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2382 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2385 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2386 size_t nbytes
, loff_t
*ppos
)
2388 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2389 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2391 if (cgroup_is_removed(cgrp
))
2395 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2397 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2399 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2404 * seqfile ops/methods for returning structured data. Currently just
2405 * supports string->u64 maps, but can be extended in future.
2408 struct cgroup_seqfile_state
{
2410 struct cgroup
*cgroup
;
2413 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2415 struct seq_file
*sf
= cb
->state
;
2416 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2419 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2421 struct cgroup_seqfile_state
*state
= m
->private;
2422 struct cftype
*cft
= state
->cft
;
2423 if (cft
->read_map
) {
2424 struct cgroup_map_cb cb
= {
2425 .fill
= cgroup_map_add
,
2428 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2430 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2433 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2435 struct seq_file
*seq
= file
->private_data
;
2436 kfree(seq
->private);
2437 return single_release(inode
, file
);
2440 static const struct file_operations cgroup_seqfile_operations
= {
2442 .write
= cgroup_file_write
,
2443 .llseek
= seq_lseek
,
2444 .release
= cgroup_seqfile_release
,
2447 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2452 err
= generic_file_open(inode
, file
);
2455 cft
= __d_cft(file
->f_dentry
);
2457 if (cft
->read_map
|| cft
->read_seq_string
) {
2458 struct cgroup_seqfile_state
*state
=
2459 kzalloc(sizeof(*state
), GFP_USER
);
2463 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2464 file
->f_op
= &cgroup_seqfile_operations
;
2465 err
= single_open(file
, cgroup_seqfile_show
, state
);
2468 } else if (cft
->open
)
2469 err
= cft
->open(inode
, file
);
2476 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2478 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2480 return cft
->release(inode
, file
);
2485 * cgroup_rename - Only allow simple rename of directories in place.
2487 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2488 struct inode
*new_dir
, struct dentry
*new_dentry
)
2490 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2492 if (new_dentry
->d_inode
)
2494 if (old_dir
!= new_dir
)
2496 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2499 static const struct file_operations cgroup_file_operations
= {
2500 .read
= cgroup_file_read
,
2501 .write
= cgroup_file_write
,
2502 .llseek
= generic_file_llseek
,
2503 .open
= cgroup_file_open
,
2504 .release
= cgroup_file_release
,
2507 static const struct inode_operations cgroup_dir_inode_operations
= {
2508 .lookup
= cgroup_lookup
,
2509 .mkdir
= cgroup_mkdir
,
2510 .rmdir
= cgroup_rmdir
,
2511 .rename
= cgroup_rename
,
2514 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
2516 if (dentry
->d_name
.len
> NAME_MAX
)
2517 return ERR_PTR(-ENAMETOOLONG
);
2518 d_add(dentry
, NULL
);
2523 * Check if a file is a control file
2525 static inline struct cftype
*__file_cft(struct file
*file
)
2527 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2528 return ERR_PTR(-EINVAL
);
2529 return __d_cft(file
->f_dentry
);
2532 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2533 struct super_block
*sb
)
2535 struct inode
*inode
;
2539 if (dentry
->d_inode
)
2542 inode
= cgroup_new_inode(mode
, sb
);
2546 if (S_ISDIR(mode
)) {
2547 inode
->i_op
= &cgroup_dir_inode_operations
;
2548 inode
->i_fop
= &simple_dir_operations
;
2550 /* start off with i_nlink == 2 (for "." entry) */
2553 /* start with the directory inode held, so that we can
2554 * populate it without racing with another mkdir */
2555 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2556 } else if (S_ISREG(mode
)) {
2558 inode
->i_fop
= &cgroup_file_operations
;
2560 d_instantiate(dentry
, inode
);
2561 dget(dentry
); /* Extra count - pin the dentry in core */
2566 * cgroup_create_dir - create a directory for an object.
2567 * @cgrp: the cgroup we create the directory for. It must have a valid
2568 * ->parent field. And we are going to fill its ->dentry field.
2569 * @dentry: dentry of the new cgroup
2570 * @mode: mode to set on new directory.
2572 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
2575 struct dentry
*parent
;
2578 parent
= cgrp
->parent
->dentry
;
2579 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
2581 dentry
->d_fsdata
= cgrp
;
2582 inc_nlink(parent
->d_inode
);
2583 rcu_assign_pointer(cgrp
->dentry
, dentry
);
2592 * cgroup_file_mode - deduce file mode of a control file
2593 * @cft: the control file in question
2595 * returns cft->mode if ->mode is not 0
2596 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2597 * returns S_IRUGO if it has only a read handler
2598 * returns S_IWUSR if it has only a write hander
2600 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2607 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2608 cft
->read_map
|| cft
->read_seq_string
)
2611 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2612 cft
->write_string
|| cft
->trigger
)
2618 int cgroup_add_file(struct cgroup
*cgrp
,
2619 struct cgroup_subsys
*subsys
,
2620 const struct cftype
*cft
)
2622 struct dentry
*dir
= cgrp
->dentry
;
2623 struct dentry
*dentry
;
2626 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2628 /* does @cft->flags tell us to skip creation on @cgrp? */
2629 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2631 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2634 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2635 strcpy(name
, subsys
->name
);
2638 strcat(name
, cft
->name
);
2639 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2640 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2641 if (!IS_ERR(dentry
)) {
2642 mode
= cgroup_file_mode(cft
);
2643 error
= cgroup_create_file(dentry
, mode
| S_IFREG
,
2646 dentry
->d_fsdata
= (void *)cft
;
2649 error
= PTR_ERR(dentry
);
2652 EXPORT_SYMBOL_GPL(cgroup_add_file
);
2654 int cgroup_add_files(struct cgroup
*cgrp
,
2655 struct cgroup_subsys
*subsys
,
2656 const struct cftype cft
[],
2660 for (i
= 0; i
< count
; i
++) {
2661 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
2667 EXPORT_SYMBOL_GPL(cgroup_add_files
);
2669 static DEFINE_MUTEX(cgroup_cft_mutex
);
2671 static void cgroup_cfts_prepare(void)
2672 __acquires(&cgroup_cft_mutex
) __acquires(&cgroup_mutex
)
2675 * Thanks to the entanglement with vfs inode locking, we can't walk
2676 * the existing cgroups under cgroup_mutex and create files.
2677 * Instead, we increment reference on all cgroups and build list of
2678 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2679 * exclusive access to the field.
2681 mutex_lock(&cgroup_cft_mutex
);
2682 mutex_lock(&cgroup_mutex
);
2685 static void cgroup_cfts_commit(struct cgroup_subsys
*ss
,
2686 const struct cftype
*cfts
)
2687 __releases(&cgroup_mutex
) __releases(&cgroup_cft_mutex
)
2690 struct cgroup
*cgrp
, *n
;
2693 while (cfts
[count
].name
[0] != '\0')
2696 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2697 if (cfts
&& ss
->root
!= &rootnode
) {
2698 list_for_each_entry(cgrp
, &ss
->root
->allcg_list
, allcg_node
) {
2700 list_add_tail(&cgrp
->cft_q_node
, &pending
);
2704 mutex_unlock(&cgroup_mutex
);
2707 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2708 * files for all cgroups which were created before.
2710 list_for_each_entry_safe(cgrp
, n
, &pending
, cft_q_node
) {
2711 struct inode
*inode
= cgrp
->dentry
->d_inode
;
2713 mutex_lock(&inode
->i_mutex
);
2714 mutex_lock(&cgroup_mutex
);
2715 if (!cgroup_is_removed(cgrp
))
2716 cgroup_add_files(cgrp
, ss
, cfts
, count
);
2717 mutex_unlock(&cgroup_mutex
);
2718 mutex_unlock(&inode
->i_mutex
);
2720 list_del_init(&cgrp
->cft_q_node
);
2724 mutex_unlock(&cgroup_cft_mutex
);
2728 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2729 * @ss: target cgroup subsystem
2730 * @cfts: zero-length name terminated array of cftypes
2732 * Register @cfts to @ss. Files described by @cfts are created for all
2733 * existing cgroups to which @ss is attached and all future cgroups will
2734 * have them too. This function can be called anytime whether @ss is
2737 * Returns 0 on successful registration, -errno on failure. Note that this
2738 * function currently returns 0 as long as @cfts registration is successful
2739 * even if some file creation attempts on existing cgroups fail.
2741 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, const struct cftype
*cfts
)
2743 struct cftype_set
*set
;
2745 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2749 cgroup_cfts_prepare();
2751 list_add_tail(&set
->node
, &ss
->cftsets
);
2752 cgroup_cfts_commit(ss
, cfts
);
2756 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2759 * cgroup_task_count - count the number of tasks in a cgroup.
2760 * @cgrp: the cgroup in question
2762 * Return the number of tasks in the cgroup.
2764 int cgroup_task_count(const struct cgroup
*cgrp
)
2767 struct cg_cgroup_link
*link
;
2769 read_lock(&css_set_lock
);
2770 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2771 count
+= atomic_read(&link
->cg
->refcount
);
2773 read_unlock(&css_set_lock
);
2778 * Advance a list_head iterator. The iterator should be positioned at
2779 * the start of a css_set
2781 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2782 struct cgroup_iter
*it
)
2784 struct list_head
*l
= it
->cg_link
;
2785 struct cg_cgroup_link
*link
;
2788 /* Advance to the next non-empty css_set */
2791 if (l
== &cgrp
->css_sets
) {
2795 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2797 } while (list_empty(&cg
->tasks
));
2799 it
->task
= cg
->tasks
.next
;
2803 * To reduce the fork() overhead for systems that are not actually
2804 * using their cgroups capability, we don't maintain the lists running
2805 * through each css_set to its tasks until we see the list actually
2806 * used - in other words after the first call to cgroup_iter_start().
2808 static void cgroup_enable_task_cg_lists(void)
2810 struct task_struct
*p
, *g
;
2811 write_lock(&css_set_lock
);
2812 use_task_css_set_links
= 1;
2814 * We need tasklist_lock because RCU is not safe against
2815 * while_each_thread(). Besides, a forking task that has passed
2816 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2817 * is not guaranteed to have its child immediately visible in the
2818 * tasklist if we walk through it with RCU.
2820 read_lock(&tasklist_lock
);
2821 do_each_thread(g
, p
) {
2824 * We should check if the process is exiting, otherwise
2825 * it will race with cgroup_exit() in that the list
2826 * entry won't be deleted though the process has exited.
2828 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2829 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2831 } while_each_thread(g
, p
);
2832 read_unlock(&tasklist_lock
);
2833 write_unlock(&css_set_lock
);
2836 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2837 __acquires(css_set_lock
)
2840 * The first time anyone tries to iterate across a cgroup,
2841 * we need to enable the list linking each css_set to its
2842 * tasks, and fix up all existing tasks.
2844 if (!use_task_css_set_links
)
2845 cgroup_enable_task_cg_lists();
2847 read_lock(&css_set_lock
);
2848 it
->cg_link
= &cgrp
->css_sets
;
2849 cgroup_advance_iter(cgrp
, it
);
2852 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
2853 struct cgroup_iter
*it
)
2855 struct task_struct
*res
;
2856 struct list_head
*l
= it
->task
;
2857 struct cg_cgroup_link
*link
;
2859 /* If the iterator cg is NULL, we have no tasks */
2862 res
= list_entry(l
, struct task_struct
, cg_list
);
2863 /* Advance iterator to find next entry */
2865 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
2866 if (l
== &link
->cg
->tasks
) {
2867 /* We reached the end of this task list - move on to
2868 * the next cg_cgroup_link */
2869 cgroup_advance_iter(cgrp
, it
);
2876 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2877 __releases(css_set_lock
)
2879 read_unlock(&css_set_lock
);
2882 static inline int started_after_time(struct task_struct
*t1
,
2883 struct timespec
*time
,
2884 struct task_struct
*t2
)
2886 int start_diff
= timespec_compare(&t1
->start_time
, time
);
2887 if (start_diff
> 0) {
2889 } else if (start_diff
< 0) {
2893 * Arbitrarily, if two processes started at the same
2894 * time, we'll say that the lower pointer value
2895 * started first. Note that t2 may have exited by now
2896 * so this may not be a valid pointer any longer, but
2897 * that's fine - it still serves to distinguish
2898 * between two tasks started (effectively) simultaneously.
2905 * This function is a callback from heap_insert() and is used to order
2907 * In this case we order the heap in descending task start time.
2909 static inline int started_after(void *p1
, void *p2
)
2911 struct task_struct
*t1
= p1
;
2912 struct task_struct
*t2
= p2
;
2913 return started_after_time(t1
, &t2
->start_time
, t2
);
2917 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2918 * @scan: struct cgroup_scanner containing arguments for the scan
2920 * Arguments include pointers to callback functions test_task() and
2922 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2923 * and if it returns true, call process_task() for it also.
2924 * The test_task pointer may be NULL, meaning always true (select all tasks).
2925 * Effectively duplicates cgroup_iter_{start,next,end}()
2926 * but does not lock css_set_lock for the call to process_task().
2927 * The struct cgroup_scanner may be embedded in any structure of the caller's
2929 * It is guaranteed that process_task() will act on every task that
2930 * is a member of the cgroup for the duration of this call. This
2931 * function may or may not call process_task() for tasks that exit
2932 * or move to a different cgroup during the call, or are forked or
2933 * move into the cgroup during the call.
2935 * Note that test_task() may be called with locks held, and may in some
2936 * situations be called multiple times for the same task, so it should
2938 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2939 * pre-allocated and will be used for heap operations (and its "gt" member will
2940 * be overwritten), else a temporary heap will be used (allocation of which
2941 * may cause this function to fail).
2943 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
2946 struct cgroup_iter it
;
2947 struct task_struct
*p
, *dropped
;
2948 /* Never dereference latest_task, since it's not refcounted */
2949 struct task_struct
*latest_task
= NULL
;
2950 struct ptr_heap tmp_heap
;
2951 struct ptr_heap
*heap
;
2952 struct timespec latest_time
= { 0, 0 };
2955 /* The caller supplied our heap and pre-allocated its memory */
2957 heap
->gt
= &started_after
;
2959 /* We need to allocate our own heap memory */
2961 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
2963 /* cannot allocate the heap */
2969 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2970 * to determine which are of interest, and using the scanner's
2971 * "process_task" callback to process any of them that need an update.
2972 * Since we don't want to hold any locks during the task updates,
2973 * gather tasks to be processed in a heap structure.
2974 * The heap is sorted by descending task start time.
2975 * If the statically-sized heap fills up, we overflow tasks that
2976 * started later, and in future iterations only consider tasks that
2977 * started after the latest task in the previous pass. This
2978 * guarantees forward progress and that we don't miss any tasks.
2981 cgroup_iter_start(scan
->cg
, &it
);
2982 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
2984 * Only affect tasks that qualify per the caller's callback,
2985 * if he provided one
2987 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
2990 * Only process tasks that started after the last task
2993 if (!started_after_time(p
, &latest_time
, latest_task
))
2995 dropped
= heap_insert(heap
, p
);
2996 if (dropped
== NULL
) {
2998 * The new task was inserted; the heap wasn't
3002 } else if (dropped
!= p
) {
3004 * The new task was inserted, and pushed out a
3008 put_task_struct(dropped
);
3011 * Else the new task was newer than anything already in
3012 * the heap and wasn't inserted
3015 cgroup_iter_end(scan
->cg
, &it
);
3018 for (i
= 0; i
< heap
->size
; i
++) {
3019 struct task_struct
*q
= heap
->ptrs
[i
];
3021 latest_time
= q
->start_time
;
3024 /* Process the task per the caller's callback */
3025 scan
->process_task(q
, scan
);
3029 * If we had to process any tasks at all, scan again
3030 * in case some of them were in the middle of forking
3031 * children that didn't get processed.
3032 * Not the most efficient way to do it, but it avoids
3033 * having to take callback_mutex in the fork path
3037 if (heap
== &tmp_heap
)
3038 heap_free(&tmp_heap
);
3043 * Stuff for reading the 'tasks'/'procs' files.
3045 * Reading this file can return large amounts of data if a cgroup has
3046 * *lots* of attached tasks. So it may need several calls to read(),
3047 * but we cannot guarantee that the information we produce is correct
3048 * unless we produce it entirely atomically.
3052 /* which pidlist file are we talking about? */
3053 enum cgroup_filetype
{
3059 * A pidlist is a list of pids that virtually represents the contents of one
3060 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3061 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3064 struct cgroup_pidlist
{
3066 * used to find which pidlist is wanted. doesn't change as long as
3067 * this particular list stays in the list.
3069 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3072 /* how many elements the above list has */
3074 /* how many files are using the current array */
3076 /* each of these stored in a list by its cgroup */
3077 struct list_head links
;
3078 /* pointer to the cgroup we belong to, for list removal purposes */
3079 struct cgroup
*owner
;
3080 /* protects the other fields */
3081 struct rw_semaphore mutex
;
3085 * The following two functions "fix" the issue where there are more pids
3086 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3087 * TODO: replace with a kernel-wide solution to this problem
3089 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3090 static void *pidlist_allocate(int count
)
3092 if (PIDLIST_TOO_LARGE(count
))
3093 return vmalloc(count
* sizeof(pid_t
));
3095 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3097 static void pidlist_free(void *p
)
3099 if (is_vmalloc_addr(p
))
3104 static void *pidlist_resize(void *p
, int newcount
)
3107 /* note: if new alloc fails, old p will still be valid either way */
3108 if (is_vmalloc_addr(p
)) {
3109 newlist
= vmalloc(newcount
* sizeof(pid_t
));
3112 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
3115 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
3121 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3122 * If the new stripped list is sufficiently smaller and there's enough memory
3123 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3124 * number of unique elements.
3126 /* is the size difference enough that we should re-allocate the array? */
3127 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3128 static int pidlist_uniq(pid_t
**p
, int length
)
3135 * we presume the 0th element is unique, so i starts at 1. trivial
3136 * edge cases first; no work needs to be done for either
3138 if (length
== 0 || length
== 1)
3140 /* src and dest walk down the list; dest counts unique elements */
3141 for (src
= 1; src
< length
; src
++) {
3142 /* find next unique element */
3143 while (list
[src
] == list
[src
-1]) {
3148 /* dest always points to where the next unique element goes */
3149 list
[dest
] = list
[src
];
3154 * if the length difference is large enough, we want to allocate a
3155 * smaller buffer to save memory. if this fails due to out of memory,
3156 * we'll just stay with what we've got.
3158 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
3159 newlist
= pidlist_resize(list
, dest
);
3166 static int cmppid(const void *a
, const void *b
)
3168 return *(pid_t
*)a
- *(pid_t
*)b
;
3172 * find the appropriate pidlist for our purpose (given procs vs tasks)
3173 * returns with the lock on that pidlist already held, and takes care
3174 * of the use count, or returns NULL with no locks held if we're out of
3177 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3178 enum cgroup_filetype type
)
3180 struct cgroup_pidlist
*l
;
3181 /* don't need task_nsproxy() if we're looking at ourself */
3182 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
3185 * We can't drop the pidlist_mutex before taking the l->mutex in case
3186 * the last ref-holder is trying to remove l from the list at the same
3187 * time. Holding the pidlist_mutex precludes somebody taking whichever
3188 * list we find out from under us - compare release_pid_array().
3190 mutex_lock(&cgrp
->pidlist_mutex
);
3191 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3192 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3193 /* make sure l doesn't vanish out from under us */
3194 down_write(&l
->mutex
);
3195 mutex_unlock(&cgrp
->pidlist_mutex
);
3199 /* entry not found; create a new one */
3200 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3202 mutex_unlock(&cgrp
->pidlist_mutex
);
3205 init_rwsem(&l
->mutex
);
3206 down_write(&l
->mutex
);
3208 l
->key
.ns
= get_pid_ns(ns
);
3209 l
->use_count
= 0; /* don't increment here */
3212 list_add(&l
->links
, &cgrp
->pidlists
);
3213 mutex_unlock(&cgrp
->pidlist_mutex
);
3218 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3220 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3221 struct cgroup_pidlist
**lp
)
3225 int pid
, n
= 0; /* used for populating the array */
3226 struct cgroup_iter it
;
3227 struct task_struct
*tsk
;
3228 struct cgroup_pidlist
*l
;
3231 * If cgroup gets more users after we read count, we won't have
3232 * enough space - tough. This race is indistinguishable to the
3233 * caller from the case that the additional cgroup users didn't
3234 * show up until sometime later on.
3236 length
= cgroup_task_count(cgrp
);
3237 array
= pidlist_allocate(length
);
3240 /* now, populate the array */
3241 cgroup_iter_start(cgrp
, &it
);
3242 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3243 if (unlikely(n
== length
))
3245 /* get tgid or pid for procs or tasks file respectively */
3246 if (type
== CGROUP_FILE_PROCS
)
3247 pid
= task_tgid_vnr(tsk
);
3249 pid
= task_pid_vnr(tsk
);
3250 if (pid
> 0) /* make sure to only use valid results */
3253 cgroup_iter_end(cgrp
, &it
);
3255 /* now sort & (if procs) strip out duplicates */
3256 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3257 if (type
== CGROUP_FILE_PROCS
)
3258 length
= pidlist_uniq(&array
, length
);
3259 l
= cgroup_pidlist_find(cgrp
, type
);
3261 pidlist_free(array
);
3264 /* store array, freeing old if necessary - lock already held */
3265 pidlist_free(l
->list
);
3269 up_write(&l
->mutex
);
3275 * cgroupstats_build - build and fill cgroupstats
3276 * @stats: cgroupstats to fill information into
3277 * @dentry: A dentry entry belonging to the cgroup for which stats have
3280 * Build and fill cgroupstats so that taskstats can export it to user
3283 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3286 struct cgroup
*cgrp
;
3287 struct cgroup_iter it
;
3288 struct task_struct
*tsk
;
3291 * Validate dentry by checking the superblock operations,
3292 * and make sure it's a directory.
3294 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3295 !S_ISDIR(dentry
->d_inode
->i_mode
))
3299 cgrp
= dentry
->d_fsdata
;
3301 cgroup_iter_start(cgrp
, &it
);
3302 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3303 switch (tsk
->state
) {
3305 stats
->nr_running
++;
3307 case TASK_INTERRUPTIBLE
:
3308 stats
->nr_sleeping
++;
3310 case TASK_UNINTERRUPTIBLE
:
3311 stats
->nr_uninterruptible
++;
3314 stats
->nr_stopped
++;
3317 if (delayacct_is_task_waiting_on_io(tsk
))
3318 stats
->nr_io_wait
++;
3322 cgroup_iter_end(cgrp
, &it
);
3330 * seq_file methods for the tasks/procs files. The seq_file position is the
3331 * next pid to display; the seq_file iterator is a pointer to the pid
3332 * in the cgroup->l->list array.
3335 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3338 * Initially we receive a position value that corresponds to
3339 * one more than the last pid shown (or 0 on the first call or
3340 * after a seek to the start). Use a binary-search to find the
3341 * next pid to display, if any
3343 struct cgroup_pidlist
*l
= s
->private;
3344 int index
= 0, pid
= *pos
;
3347 down_read(&l
->mutex
);
3349 int end
= l
->length
;
3351 while (index
< end
) {
3352 int mid
= (index
+ end
) / 2;
3353 if (l
->list
[mid
] == pid
) {
3356 } else if (l
->list
[mid
] <= pid
)
3362 /* If we're off the end of the array, we're done */
3363 if (index
>= l
->length
)
3365 /* Update the abstract position to be the actual pid that we found */
3366 iter
= l
->list
+ index
;
3371 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3373 struct cgroup_pidlist
*l
= s
->private;
3377 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3379 struct cgroup_pidlist
*l
= s
->private;
3381 pid_t
*end
= l
->list
+ l
->length
;
3383 * Advance to the next pid in the array. If this goes off the
3395 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3397 return seq_printf(s
, "%d\n", *(int *)v
);
3401 * seq_operations functions for iterating on pidlists through seq_file -
3402 * independent of whether it's tasks or procs
3404 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3405 .start
= cgroup_pidlist_start
,
3406 .stop
= cgroup_pidlist_stop
,
3407 .next
= cgroup_pidlist_next
,
3408 .show
= cgroup_pidlist_show
,
3411 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3414 * the case where we're the last user of this particular pidlist will
3415 * have us remove it from the cgroup's list, which entails taking the
3416 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3417 * pidlist_mutex, we have to take pidlist_mutex first.
3419 mutex_lock(&l
->owner
->pidlist_mutex
);
3420 down_write(&l
->mutex
);
3421 BUG_ON(!l
->use_count
);
3422 if (!--l
->use_count
) {
3423 /* we're the last user if refcount is 0; remove and free */
3424 list_del(&l
->links
);
3425 mutex_unlock(&l
->owner
->pidlist_mutex
);
3426 pidlist_free(l
->list
);
3427 put_pid_ns(l
->key
.ns
);
3428 up_write(&l
->mutex
);
3432 mutex_unlock(&l
->owner
->pidlist_mutex
);
3433 up_write(&l
->mutex
);
3436 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3438 struct cgroup_pidlist
*l
;
3439 if (!(file
->f_mode
& FMODE_READ
))
3442 * the seq_file will only be initialized if the file was opened for
3443 * reading; hence we check if it's not null only in that case.
3445 l
= ((struct seq_file
*)file
->private_data
)->private;
3446 cgroup_release_pid_array(l
);
3447 return seq_release(inode
, file
);
3450 static const struct file_operations cgroup_pidlist_operations
= {
3452 .llseek
= seq_lseek
,
3453 .write
= cgroup_file_write
,
3454 .release
= cgroup_pidlist_release
,
3458 * The following functions handle opens on a file that displays a pidlist
3459 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3462 /* helper function for the two below it */
3463 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3465 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3466 struct cgroup_pidlist
*l
;
3469 /* Nothing to do for write-only files */
3470 if (!(file
->f_mode
& FMODE_READ
))
3473 /* have the array populated */
3474 retval
= pidlist_array_load(cgrp
, type
, &l
);
3477 /* configure file information */
3478 file
->f_op
= &cgroup_pidlist_operations
;
3480 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3482 cgroup_release_pid_array(l
);
3485 ((struct seq_file
*)file
->private_data
)->private = l
;
3488 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3490 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3492 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3494 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3497 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3500 return notify_on_release(cgrp
);
3503 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3507 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3509 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3511 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3516 * Unregister event and free resources.
3518 * Gets called from workqueue.
3520 static void cgroup_event_remove(struct work_struct
*work
)
3522 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3524 struct cgroup
*cgrp
= event
->cgrp
;
3526 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3528 eventfd_ctx_put(event
->eventfd
);
3534 * Gets called on POLLHUP on eventfd when user closes it.
3536 * Called with wqh->lock held and interrupts disabled.
3538 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3539 int sync
, void *key
)
3541 struct cgroup_event
*event
= container_of(wait
,
3542 struct cgroup_event
, wait
);
3543 struct cgroup
*cgrp
= event
->cgrp
;
3544 unsigned long flags
= (unsigned long)key
;
3546 if (flags
& POLLHUP
) {
3547 __remove_wait_queue(event
->wqh
, &event
->wait
);
3548 spin_lock(&cgrp
->event_list_lock
);
3549 list_del(&event
->list
);
3550 spin_unlock(&cgrp
->event_list_lock
);
3552 * We are in atomic context, but cgroup_event_remove() may
3553 * sleep, so we have to call it in workqueue.
3555 schedule_work(&event
->remove
);
3561 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3562 wait_queue_head_t
*wqh
, poll_table
*pt
)
3564 struct cgroup_event
*event
= container_of(pt
,
3565 struct cgroup_event
, pt
);
3568 add_wait_queue(wqh
, &event
->wait
);
3572 * Parse input and register new cgroup event handler.
3574 * Input must be in format '<event_fd> <control_fd> <args>'.
3575 * Interpretation of args is defined by control file implementation.
3577 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3580 struct cgroup_event
*event
= NULL
;
3581 unsigned int efd
, cfd
;
3582 struct file
*efile
= NULL
;
3583 struct file
*cfile
= NULL
;
3587 efd
= simple_strtoul(buffer
, &endp
, 10);
3592 cfd
= simple_strtoul(buffer
, &endp
, 10);
3593 if ((*endp
!= ' ') && (*endp
!= '\0'))
3597 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3601 INIT_LIST_HEAD(&event
->list
);
3602 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3603 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3604 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3606 efile
= eventfd_fget(efd
);
3607 if (IS_ERR(efile
)) {
3608 ret
= PTR_ERR(efile
);
3612 event
->eventfd
= eventfd_ctx_fileget(efile
);
3613 if (IS_ERR(event
->eventfd
)) {
3614 ret
= PTR_ERR(event
->eventfd
);
3624 /* the process need read permission on control file */
3625 /* AV: shouldn't we check that it's been opened for read instead? */
3626 ret
= inode_permission(cfile
->f_path
.dentry
->d_inode
, MAY_READ
);
3630 event
->cft
= __file_cft(cfile
);
3631 if (IS_ERR(event
->cft
)) {
3632 ret
= PTR_ERR(event
->cft
);
3636 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3641 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3642 event
->eventfd
, buffer
);
3646 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3647 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3653 * Events should be removed after rmdir of cgroup directory, but before
3654 * destroying subsystem state objects. Let's take reference to cgroup
3655 * directory dentry to do that.
3659 spin_lock(&cgrp
->event_list_lock
);
3660 list_add(&event
->list
, &cgrp
->event_list
);
3661 spin_unlock(&cgrp
->event_list_lock
);
3672 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3673 eventfd_ctx_put(event
->eventfd
);
3675 if (!IS_ERR_OR_NULL(efile
))
3683 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3686 return clone_children(cgrp
);
3689 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3694 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3696 clear_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3701 * for the common functions, 'private' gives the type of file
3703 /* for hysterical raisins, we can't put this on the older files */
3704 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3705 static struct cftype files
[] = {
3708 .open
= cgroup_tasks_open
,
3709 .write_u64
= cgroup_tasks_write
,
3710 .release
= cgroup_pidlist_release
,
3711 .mode
= S_IRUGO
| S_IWUSR
,
3714 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3715 .open
= cgroup_procs_open
,
3716 .write_u64
= cgroup_procs_write
,
3717 .release
= cgroup_pidlist_release
,
3718 .mode
= S_IRUGO
| S_IWUSR
,
3721 .name
= "notify_on_release",
3722 .read_u64
= cgroup_read_notify_on_release
,
3723 .write_u64
= cgroup_write_notify_on_release
,
3726 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3727 .write_string
= cgroup_write_event_control
,
3731 .name
= "cgroup.clone_children",
3732 .read_u64
= cgroup_clone_children_read
,
3733 .write_u64
= cgroup_clone_children_write
,
3737 static struct cftype cft_release_agent
= {
3738 .name
= "release_agent",
3739 .read_seq_string
= cgroup_release_agent_show
,
3740 .write_string
= cgroup_release_agent_write
,
3741 .max_write_len
= PATH_MAX
,
3744 static int cgroup_populate_dir(struct cgroup
*cgrp
)
3747 struct cgroup_subsys
*ss
;
3749 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
3753 if (cgrp
== cgrp
->top_cgroup
) {
3754 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
3758 /* process cftsets of each subsystem */
3759 for_each_subsys(cgrp
->root
, ss
) {
3760 struct cftype_set
*set
;
3762 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
3765 list_for_each_entry(set
, &ss
->cftsets
, node
) {
3766 const struct cftype
*cft
;
3768 for (cft
= set
->cfts
; cft
->name
[0] != '\0'; cft
++) {
3769 err
= cgroup_add_file(cgrp
, ss
, cft
);
3771 pr_warning("cgroup_populate_dir: failed to create %s, err=%d\n",
3777 /* This cgroup is ready now */
3778 for_each_subsys(cgrp
->root
, ss
) {
3779 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3781 * Update id->css pointer and make this css visible from
3782 * CSS ID functions. This pointer will be dereferened
3783 * from RCU-read-side without locks.
3786 rcu_assign_pointer(css
->id
->css
, css
);
3792 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
3793 struct cgroup_subsys
*ss
,
3794 struct cgroup
*cgrp
)
3797 atomic_set(&css
->refcnt
, 1);
3800 if (cgrp
== dummytop
)
3801 set_bit(CSS_ROOT
, &css
->flags
);
3802 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
3803 cgrp
->subsys
[ss
->subsys_id
] = css
;
3806 static void cgroup_lock_hierarchy(struct cgroupfs_root
*root
)
3808 /* We need to take each hierarchy_mutex in a consistent order */
3812 * No worry about a race with rebind_subsystems that might mess up the
3813 * locking order, since both parties are under cgroup_mutex.
3815 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3816 struct cgroup_subsys
*ss
= subsys
[i
];
3819 if (ss
->root
== root
)
3820 mutex_lock(&ss
->hierarchy_mutex
);
3824 static void cgroup_unlock_hierarchy(struct cgroupfs_root
*root
)
3828 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3829 struct cgroup_subsys
*ss
= subsys
[i
];
3832 if (ss
->root
== root
)
3833 mutex_unlock(&ss
->hierarchy_mutex
);
3838 * cgroup_create - create a cgroup
3839 * @parent: cgroup that will be parent of the new cgroup
3840 * @dentry: dentry of the new cgroup
3841 * @mode: mode to set on new inode
3843 * Must be called with the mutex on the parent inode held
3845 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
3848 struct cgroup
*cgrp
;
3849 struct cgroupfs_root
*root
= parent
->root
;
3851 struct cgroup_subsys
*ss
;
3852 struct super_block
*sb
= root
->sb
;
3854 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3858 /* Grab a reference on the superblock so the hierarchy doesn't
3859 * get deleted on unmount if there are child cgroups. This
3860 * can be done outside cgroup_mutex, since the sb can't
3861 * disappear while someone has an open control file on the
3863 atomic_inc(&sb
->s_active
);
3865 mutex_lock(&cgroup_mutex
);
3867 init_cgroup_housekeeping(cgrp
);
3869 cgrp
->parent
= parent
;
3870 cgrp
->root
= parent
->root
;
3871 cgrp
->top_cgroup
= parent
->top_cgroup
;
3873 if (notify_on_release(parent
))
3874 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3876 if (clone_children(parent
))
3877 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3879 for_each_subsys(root
, ss
) {
3880 struct cgroup_subsys_state
*css
= ss
->create(cgrp
);
3886 init_cgroup_css(css
, ss
, cgrp
);
3888 err
= alloc_css_id(ss
, parent
, cgrp
);
3892 /* At error, ->destroy() callback has to free assigned ID. */
3893 if (clone_children(parent
) && ss
->post_clone
)
3894 ss
->post_clone(cgrp
);
3897 cgroup_lock_hierarchy(root
);
3898 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
3899 cgroup_unlock_hierarchy(root
);
3900 root
->number_of_cgroups
++;
3902 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
3906 /* The cgroup directory was pre-locked for us */
3907 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
3909 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
3911 err
= cgroup_populate_dir(cgrp
);
3912 /* If err < 0, we have a half-filled directory - oh well ;) */
3914 mutex_unlock(&cgroup_mutex
);
3915 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
3921 cgroup_lock_hierarchy(root
);
3922 list_del(&cgrp
->sibling
);
3923 cgroup_unlock_hierarchy(root
);
3924 root
->number_of_cgroups
--;
3928 for_each_subsys(root
, ss
) {
3929 if (cgrp
->subsys
[ss
->subsys_id
])
3933 mutex_unlock(&cgroup_mutex
);
3935 /* Release the reference count that we took on the superblock */
3936 deactivate_super(sb
);
3942 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3944 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
3946 /* the vfs holds inode->i_mutex already */
3947 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
3950 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
3952 /* Check the reference count on each subsystem. Since we
3953 * already established that there are no tasks in the
3954 * cgroup, if the css refcount is also 1, then there should
3955 * be no outstanding references, so the subsystem is safe to
3956 * destroy. We scan across all subsystems rather than using
3957 * the per-hierarchy linked list of mounted subsystems since
3958 * we can be called via check_for_release() with no
3959 * synchronization other than RCU, and the subsystem linked
3960 * list isn't RCU-safe */
3963 * We won't need to lock the subsys array, because the subsystems
3964 * we're concerned about aren't going anywhere since our cgroup root
3965 * has a reference on them.
3967 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3968 struct cgroup_subsys
*ss
= subsys
[i
];
3969 struct cgroup_subsys_state
*css
;
3970 /* Skip subsystems not present or not in this hierarchy */
3971 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
3973 css
= cgrp
->subsys
[ss
->subsys_id
];
3974 /* When called from check_for_release() it's possible
3975 * that by this point the cgroup has been removed
3976 * and the css deleted. But a false-positive doesn't
3977 * matter, since it can only happen if the cgroup
3978 * has been deleted and hence no longer needs the
3979 * release agent to be called anyway. */
3980 if (css
&& (atomic_read(&css
->refcnt
) > 1))
3987 * Atomically mark all (or else none) of the cgroup's CSS objects as
3988 * CSS_REMOVED. Return true on success, or false if the cgroup has
3989 * busy subsystems. Call with cgroup_mutex held
3992 static int cgroup_clear_css_refs(struct cgroup
*cgrp
)
3994 struct cgroup_subsys
*ss
;
3995 unsigned long flags
;
3996 bool failed
= false;
3997 local_irq_save(flags
);
3998 for_each_subsys(cgrp
->root
, ss
) {
3999 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4002 /* We can only remove a CSS with a refcnt==1 */
4003 refcnt
= atomic_read(&css
->refcnt
);
4010 * Drop the refcnt to 0 while we check other
4011 * subsystems. This will cause any racing
4012 * css_tryget() to spin until we set the
4013 * CSS_REMOVED bits or abort
4015 if (atomic_cmpxchg(&css
->refcnt
, refcnt
, 0) == refcnt
)
4021 for_each_subsys(cgrp
->root
, ss
) {
4022 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4025 * Restore old refcnt if we previously managed
4026 * to clear it from 1 to 0
4028 if (!atomic_read(&css
->refcnt
))
4029 atomic_set(&css
->refcnt
, 1);
4031 /* Commit the fact that the CSS is removed */
4032 set_bit(CSS_REMOVED
, &css
->flags
);
4035 local_irq_restore(flags
);
4039 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4041 struct cgroup
*cgrp
= dentry
->d_fsdata
;
4043 struct cgroup
*parent
;
4045 struct cgroup_event
*event
, *tmp
;
4048 /* the vfs holds both inode->i_mutex already */
4050 mutex_lock(&cgroup_mutex
);
4051 if (atomic_read(&cgrp
->count
) != 0) {
4052 mutex_unlock(&cgroup_mutex
);
4055 if (!list_empty(&cgrp
->children
)) {
4056 mutex_unlock(&cgroup_mutex
);
4059 mutex_unlock(&cgroup_mutex
);
4062 * In general, subsystem has no css->refcnt after pre_destroy(). But
4063 * in racy cases, subsystem may have to get css->refcnt after
4064 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4065 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4066 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4067 * and subsystem's reference count handling. Please see css_get/put
4068 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4070 set_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4073 * Call pre_destroy handlers of subsys. Notify subsystems
4074 * that rmdir() request comes.
4076 ret
= cgroup_call_pre_destroy(cgrp
);
4078 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4082 mutex_lock(&cgroup_mutex
);
4083 parent
= cgrp
->parent
;
4084 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
)) {
4085 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4086 mutex_unlock(&cgroup_mutex
);
4089 prepare_to_wait(&cgroup_rmdir_waitq
, &wait
, TASK_INTERRUPTIBLE
);
4090 if (!cgroup_clear_css_refs(cgrp
)) {
4091 mutex_unlock(&cgroup_mutex
);
4093 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4094 * prepare_to_wait(), we need to check this flag.
4096 if (test_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
))
4098 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4099 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4100 if (signal_pending(current
))
4104 /* NO css_tryget() can success after here. */
4105 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4106 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4108 raw_spin_lock(&release_list_lock
);
4109 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
4110 if (!list_empty(&cgrp
->release_list
))
4111 list_del_init(&cgrp
->release_list
);
4112 raw_spin_unlock(&release_list_lock
);
4114 cgroup_lock_hierarchy(cgrp
->root
);
4115 /* delete this cgroup from parent->children */
4116 list_del_init(&cgrp
->sibling
);
4117 cgroup_unlock_hierarchy(cgrp
->root
);
4119 list_del_init(&cgrp
->allcg_node
);
4121 d
= dget(cgrp
->dentry
);
4123 cgroup_d_remove_dir(d
);
4126 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4127 check_for_release(parent
);
4130 * Unregister events and notify userspace.
4131 * Notify userspace about cgroup removing only after rmdir of cgroup
4132 * directory to avoid race between userspace and kernelspace
4134 spin_lock(&cgrp
->event_list_lock
);
4135 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4136 list_del(&event
->list
);
4137 remove_wait_queue(event
->wqh
, &event
->wait
);
4138 eventfd_signal(event
->eventfd
, 1);
4139 schedule_work(&event
->remove
);
4141 spin_unlock(&cgrp
->event_list_lock
);
4143 mutex_unlock(&cgroup_mutex
);
4147 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4149 INIT_LIST_HEAD(&ss
->cftsets
);
4152 * base_cftset is embedded in subsys itself, no need to worry about
4155 if (ss
->base_cftypes
) {
4156 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4157 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4161 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4163 struct cgroup_subsys_state
*css
;
4165 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4167 /* init base cftset */
4168 cgroup_init_cftsets(ss
);
4170 /* Create the top cgroup state for this subsystem */
4171 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4172 ss
->root
= &rootnode
;
4173 css
= ss
->create(dummytop
);
4174 /* We don't handle early failures gracefully */
4175 BUG_ON(IS_ERR(css
));
4176 init_cgroup_css(css
, ss
, dummytop
);
4178 /* Update the init_css_set to contain a subsys
4179 * pointer to this state - since the subsystem is
4180 * newly registered, all tasks and hence the
4181 * init_css_set is in the subsystem's top cgroup. */
4182 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
4184 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4186 /* At system boot, before all subsystems have been
4187 * registered, no tasks have been forked, so we don't
4188 * need to invoke fork callbacks here. */
4189 BUG_ON(!list_empty(&init_task
.tasks
));
4191 mutex_init(&ss
->hierarchy_mutex
);
4192 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4195 /* this function shouldn't be used with modular subsystems, since they
4196 * need to register a subsys_id, among other things */
4201 * cgroup_load_subsys: load and register a modular subsystem at runtime
4202 * @ss: the subsystem to load
4204 * This function should be called in a modular subsystem's initcall. If the
4205 * subsystem is built as a module, it will be assigned a new subsys_id and set
4206 * up for use. If the subsystem is built-in anyway, work is delegated to the
4207 * simpler cgroup_init_subsys.
4209 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4212 struct cgroup_subsys_state
*css
;
4214 /* check name and function validity */
4215 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4216 ss
->create
== NULL
|| ss
->destroy
== NULL
)
4220 * we don't support callbacks in modular subsystems. this check is
4221 * before the ss->module check for consistency; a subsystem that could
4222 * be a module should still have no callbacks even if the user isn't
4223 * compiling it as one.
4225 if (ss
->fork
|| ss
->exit
)
4229 * an optionally modular subsystem is built-in: we want to do nothing,
4230 * since cgroup_init_subsys will have already taken care of it.
4232 if (ss
->module
== NULL
) {
4233 /* a few sanity checks */
4234 BUG_ON(ss
->subsys_id
>= CGROUP_BUILTIN_SUBSYS_COUNT
);
4235 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4239 /* init base cftset */
4240 cgroup_init_cftsets(ss
);
4243 * need to register a subsys id before anything else - for example,
4244 * init_cgroup_css needs it.
4246 mutex_lock(&cgroup_mutex
);
4247 /* find the first empty slot in the array */
4248 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4249 if (subsys
[i
] == NULL
)
4252 if (i
== CGROUP_SUBSYS_COUNT
) {
4253 /* maximum number of subsystems already registered! */
4254 mutex_unlock(&cgroup_mutex
);
4257 /* assign ourselves the subsys_id */
4262 * no ss->create seems to need anything important in the ss struct, so
4263 * this can happen first (i.e. before the rootnode attachment).
4265 css
= ss
->create(dummytop
);
4267 /* failure case - need to deassign the subsys[] slot. */
4269 mutex_unlock(&cgroup_mutex
);
4270 return PTR_ERR(css
);
4273 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4274 ss
->root
= &rootnode
;
4276 /* our new subsystem will be attached to the dummy hierarchy. */
4277 init_cgroup_css(css
, ss
, dummytop
);
4278 /* init_idr must be after init_cgroup_css because it sets css->id. */
4280 int ret
= cgroup_init_idr(ss
, css
);
4282 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4283 ss
->destroy(dummytop
);
4285 mutex_unlock(&cgroup_mutex
);
4291 * Now we need to entangle the css into the existing css_sets. unlike
4292 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4293 * will need a new pointer to it; done by iterating the css_set_table.
4294 * furthermore, modifying the existing css_sets will corrupt the hash
4295 * table state, so each changed css_set will need its hash recomputed.
4296 * this is all done under the css_set_lock.
4298 write_lock(&css_set_lock
);
4299 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
4301 struct hlist_node
*node
, *tmp
;
4302 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
4304 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
4305 /* skip entries that we already rehashed */
4306 if (cg
->subsys
[ss
->subsys_id
])
4308 /* remove existing entry */
4309 hlist_del(&cg
->hlist
);
4311 cg
->subsys
[ss
->subsys_id
] = css
;
4312 /* recompute hash and restore entry */
4313 new_bucket
= css_set_hash(cg
->subsys
);
4314 hlist_add_head(&cg
->hlist
, new_bucket
);
4317 write_unlock(&css_set_lock
);
4319 mutex_init(&ss
->hierarchy_mutex
);
4320 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4324 mutex_unlock(&cgroup_mutex
);
4327 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4330 * cgroup_unload_subsys: unload a modular subsystem
4331 * @ss: the subsystem to unload
4333 * This function should be called in a modular subsystem's exitcall. When this
4334 * function is invoked, the refcount on the subsystem's module will be 0, so
4335 * the subsystem will not be attached to any hierarchy.
4337 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4339 struct cg_cgroup_link
*link
;
4340 struct hlist_head
*hhead
;
4342 BUG_ON(ss
->module
== NULL
);
4345 * we shouldn't be called if the subsystem is in use, and the use of
4346 * try_module_get in parse_cgroupfs_options should ensure that it
4347 * doesn't start being used while we're killing it off.
4349 BUG_ON(ss
->root
!= &rootnode
);
4351 mutex_lock(&cgroup_mutex
);
4352 /* deassign the subsys_id */
4353 BUG_ON(ss
->subsys_id
< CGROUP_BUILTIN_SUBSYS_COUNT
);
4354 subsys
[ss
->subsys_id
] = NULL
;
4356 /* remove subsystem from rootnode's list of subsystems */
4357 list_del_init(&ss
->sibling
);
4360 * disentangle the css from all css_sets attached to the dummytop. as
4361 * in loading, we need to pay our respects to the hashtable gods.
4363 write_lock(&css_set_lock
);
4364 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4365 struct css_set
*cg
= link
->cg
;
4367 hlist_del(&cg
->hlist
);
4368 BUG_ON(!cg
->subsys
[ss
->subsys_id
]);
4369 cg
->subsys
[ss
->subsys_id
] = NULL
;
4370 hhead
= css_set_hash(cg
->subsys
);
4371 hlist_add_head(&cg
->hlist
, hhead
);
4373 write_unlock(&css_set_lock
);
4376 * remove subsystem's css from the dummytop and free it - need to free
4377 * before marking as null because ss->destroy needs the cgrp->subsys
4378 * pointer to find their state. note that this also takes care of
4379 * freeing the css_id.
4381 ss
->destroy(dummytop
);
4382 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4384 mutex_unlock(&cgroup_mutex
);
4386 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4389 * cgroup_init_early - cgroup initialization at system boot
4391 * Initialize cgroups at system boot, and initialize any
4392 * subsystems that request early init.
4394 int __init
cgroup_init_early(void)
4397 atomic_set(&init_css_set
.refcount
, 1);
4398 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4399 INIT_LIST_HEAD(&init_css_set
.tasks
);
4400 INIT_HLIST_NODE(&init_css_set
.hlist
);
4402 init_cgroup_root(&rootnode
);
4404 init_task
.cgroups
= &init_css_set
;
4406 init_css_set_link
.cg
= &init_css_set
;
4407 init_css_set_link
.cgrp
= dummytop
;
4408 list_add(&init_css_set_link
.cgrp_link_list
,
4409 &rootnode
.top_cgroup
.css_sets
);
4410 list_add(&init_css_set_link
.cg_link_list
,
4411 &init_css_set
.cg_links
);
4413 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
4414 INIT_HLIST_HEAD(&css_set_table
[i
]);
4416 /* at bootup time, we don't worry about modular subsystems */
4417 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4418 struct cgroup_subsys
*ss
= subsys
[i
];
4421 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4422 BUG_ON(!ss
->create
);
4423 BUG_ON(!ss
->destroy
);
4424 if (ss
->subsys_id
!= i
) {
4425 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4426 ss
->name
, ss
->subsys_id
);
4431 cgroup_init_subsys(ss
);
4437 * cgroup_init - cgroup initialization
4439 * Register cgroup filesystem and /proc file, and initialize
4440 * any subsystems that didn't request early init.
4442 int __init
cgroup_init(void)
4446 struct hlist_head
*hhead
;
4448 err
= bdi_init(&cgroup_backing_dev_info
);
4452 /* at bootup time, we don't worry about modular subsystems */
4453 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4454 struct cgroup_subsys
*ss
= subsys
[i
];
4455 if (!ss
->early_init
)
4456 cgroup_init_subsys(ss
);
4458 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4461 /* Add init_css_set to the hash table */
4462 hhead
= css_set_hash(init_css_set
.subsys
);
4463 hlist_add_head(&init_css_set
.hlist
, hhead
);
4464 BUG_ON(!init_root_id(&rootnode
));
4466 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4472 err
= register_filesystem(&cgroup_fs_type
);
4474 kobject_put(cgroup_kobj
);
4478 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4482 bdi_destroy(&cgroup_backing_dev_info
);
4488 * proc_cgroup_show()
4489 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4490 * - Used for /proc/<pid>/cgroup.
4491 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4492 * doesn't really matter if tsk->cgroup changes after we read it,
4493 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4494 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4495 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4496 * cgroup to top_cgroup.
4499 /* TODO: Use a proper seq_file iterator */
4500 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4503 struct task_struct
*tsk
;
4506 struct cgroupfs_root
*root
;
4509 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4515 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4521 mutex_lock(&cgroup_mutex
);
4523 for_each_active_root(root
) {
4524 struct cgroup_subsys
*ss
;
4525 struct cgroup
*cgrp
;
4528 seq_printf(m
, "%d:", root
->hierarchy_id
);
4529 for_each_subsys(root
, ss
)
4530 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4531 if (strlen(root
->name
))
4532 seq_printf(m
, "%sname=%s", count
? "," : "",
4535 cgrp
= task_cgroup_from_root(tsk
, root
);
4536 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4544 mutex_unlock(&cgroup_mutex
);
4545 put_task_struct(tsk
);
4552 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4554 struct pid
*pid
= PROC_I(inode
)->pid
;
4555 return single_open(file
, proc_cgroup_show
, pid
);
4558 const struct file_operations proc_cgroup_operations
= {
4559 .open
= cgroup_open
,
4561 .llseek
= seq_lseek
,
4562 .release
= single_release
,
4565 /* Display information about each subsystem and each hierarchy */
4566 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4570 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4572 * ideally we don't want subsystems moving around while we do this.
4573 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4574 * subsys/hierarchy state.
4576 mutex_lock(&cgroup_mutex
);
4577 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4578 struct cgroup_subsys
*ss
= subsys
[i
];
4581 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4582 ss
->name
, ss
->root
->hierarchy_id
,
4583 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4585 mutex_unlock(&cgroup_mutex
);
4589 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4591 return single_open(file
, proc_cgroupstats_show
, NULL
);
4594 static const struct file_operations proc_cgroupstats_operations
= {
4595 .open
= cgroupstats_open
,
4597 .llseek
= seq_lseek
,
4598 .release
= single_release
,
4602 * cgroup_fork - attach newly forked task to its parents cgroup.
4603 * @child: pointer to task_struct of forking parent process.
4605 * Description: A task inherits its parent's cgroup at fork().
4607 * A pointer to the shared css_set was automatically copied in
4608 * fork.c by dup_task_struct(). However, we ignore that copy, since
4609 * it was not made under the protection of RCU, cgroup_mutex or
4610 * threadgroup_change_begin(), so it might no longer be a valid
4611 * cgroup pointer. cgroup_attach_task() might have already changed
4612 * current->cgroups, allowing the previously referenced cgroup
4613 * group to be removed and freed.
4615 * Outside the pointer validity we also need to process the css_set
4616 * inheritance between threadgoup_change_begin() and
4617 * threadgoup_change_end(), this way there is no leak in any process
4618 * wide migration performed by cgroup_attach_proc() that could otherwise
4619 * miss a thread because it is too early or too late in the fork stage.
4621 * At the point that cgroup_fork() is called, 'current' is the parent
4622 * task, and the passed argument 'child' points to the child task.
4624 void cgroup_fork(struct task_struct
*child
)
4627 * We don't need to task_lock() current because current->cgroups
4628 * can't be changed concurrently here. The parent obviously hasn't
4629 * exited and called cgroup_exit(), and we are synchronized against
4630 * cgroup migration through threadgroup_change_begin().
4632 child
->cgroups
= current
->cgroups
;
4633 get_css_set(child
->cgroups
);
4634 INIT_LIST_HEAD(&child
->cg_list
);
4638 * cgroup_fork_callbacks - run fork callbacks
4639 * @child: the new task
4641 * Called on a new task very soon before adding it to the
4642 * tasklist. No need to take any locks since no-one can
4643 * be operating on this task.
4645 void cgroup_fork_callbacks(struct task_struct
*child
)
4647 if (need_forkexit_callback
) {
4650 * forkexit callbacks are only supported for builtin
4651 * subsystems, and the builtin section of the subsys array is
4652 * immutable, so we don't need to lock the subsys array here.
4654 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4655 struct cgroup_subsys
*ss
= subsys
[i
];
4663 * cgroup_post_fork - called on a new task after adding it to the task list
4664 * @child: the task in question
4666 * Adds the task to the list running through its css_set if necessary.
4667 * Has to be after the task is visible on the task list in case we race
4668 * with the first call to cgroup_iter_start() - to guarantee that the
4669 * new task ends up on its list.
4671 void cgroup_post_fork(struct task_struct
*child
)
4674 * use_task_css_set_links is set to 1 before we walk the tasklist
4675 * under the tasklist_lock and we read it here after we added the child
4676 * to the tasklist under the tasklist_lock as well. If the child wasn't
4677 * yet in the tasklist when we walked through it from
4678 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4679 * should be visible now due to the paired locking and barriers implied
4680 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4681 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4684 if (use_task_css_set_links
) {
4685 write_lock(&css_set_lock
);
4686 if (list_empty(&child
->cg_list
)) {
4688 * It's safe to use child->cgroups without task_lock()
4689 * here because we are protected through
4690 * threadgroup_change_begin() against concurrent
4691 * css_set change in cgroup_task_migrate(). Also
4692 * the task can't exit at that point until
4693 * wake_up_new_task() is called, so we are protected
4694 * against cgroup_exit() setting child->cgroup to
4697 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4699 write_unlock(&css_set_lock
);
4703 * cgroup_exit - detach cgroup from exiting task
4704 * @tsk: pointer to task_struct of exiting process
4705 * @run_callback: run exit callbacks?
4707 * Description: Detach cgroup from @tsk and release it.
4709 * Note that cgroups marked notify_on_release force every task in
4710 * them to take the global cgroup_mutex mutex when exiting.
4711 * This could impact scaling on very large systems. Be reluctant to
4712 * use notify_on_release cgroups where very high task exit scaling
4713 * is required on large systems.
4715 * the_top_cgroup_hack:
4717 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4719 * We call cgroup_exit() while the task is still competent to
4720 * handle notify_on_release(), then leave the task attached to the
4721 * root cgroup in each hierarchy for the remainder of its exit.
4723 * To do this properly, we would increment the reference count on
4724 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4725 * code we would add a second cgroup function call, to drop that
4726 * reference. This would just create an unnecessary hot spot on
4727 * the top_cgroup reference count, to no avail.
4729 * Normally, holding a reference to a cgroup without bumping its
4730 * count is unsafe. The cgroup could go away, or someone could
4731 * attach us to a different cgroup, decrementing the count on
4732 * the first cgroup that we never incremented. But in this case,
4733 * top_cgroup isn't going away, and either task has PF_EXITING set,
4734 * which wards off any cgroup_attach_task() attempts, or task is a failed
4735 * fork, never visible to cgroup_attach_task.
4737 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4743 * Unlink from the css_set task list if necessary.
4744 * Optimistically check cg_list before taking
4747 if (!list_empty(&tsk
->cg_list
)) {
4748 write_lock(&css_set_lock
);
4749 if (!list_empty(&tsk
->cg_list
))
4750 list_del_init(&tsk
->cg_list
);
4751 write_unlock(&css_set_lock
);
4754 /* Reassign the task to the init_css_set. */
4757 tsk
->cgroups
= &init_css_set
;
4759 if (run_callbacks
&& need_forkexit_callback
) {
4761 * modular subsystems can't use callbacks, so no need to lock
4764 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4765 struct cgroup_subsys
*ss
= subsys
[i
];
4767 struct cgroup
*old_cgrp
=
4768 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4769 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4770 ss
->exit(cgrp
, old_cgrp
, tsk
);
4777 put_css_set_taskexit(cg
);
4781 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4782 * @cgrp: the cgroup in question
4783 * @task: the task in question
4785 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4788 * If we are sending in dummytop, then presumably we are creating
4789 * the top cgroup in the subsystem.
4791 * Called only by the ns (nsproxy) cgroup.
4793 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
4796 struct cgroup
*target
;
4798 if (cgrp
== dummytop
)
4801 target
= task_cgroup_from_root(task
, cgrp
->root
);
4802 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
4803 cgrp
= cgrp
->parent
;
4804 ret
= (cgrp
== target
);
4808 static void check_for_release(struct cgroup
*cgrp
)
4810 /* All of these checks rely on RCU to keep the cgroup
4811 * structure alive */
4812 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
4813 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
4814 /* Control Group is currently removeable. If it's not
4815 * already queued for a userspace notification, queue
4817 int need_schedule_work
= 0;
4818 raw_spin_lock(&release_list_lock
);
4819 if (!cgroup_is_removed(cgrp
) &&
4820 list_empty(&cgrp
->release_list
)) {
4821 list_add(&cgrp
->release_list
, &release_list
);
4822 need_schedule_work
= 1;
4824 raw_spin_unlock(&release_list_lock
);
4825 if (need_schedule_work
)
4826 schedule_work(&release_agent_work
);
4830 /* Caller must verify that the css is not for root cgroup */
4831 void __css_put(struct cgroup_subsys_state
*css
, int count
)
4833 struct cgroup
*cgrp
= css
->cgroup
;
4836 val
= atomic_sub_return(count
, &css
->refcnt
);
4838 if (notify_on_release(cgrp
)) {
4839 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4840 check_for_release(cgrp
);
4842 cgroup_wakeup_rmdir_waiter(cgrp
);
4845 WARN_ON_ONCE(val
< 1);
4847 EXPORT_SYMBOL_GPL(__css_put
);
4850 * Notify userspace when a cgroup is released, by running the
4851 * configured release agent with the name of the cgroup (path
4852 * relative to the root of cgroup file system) as the argument.
4854 * Most likely, this user command will try to rmdir this cgroup.
4856 * This races with the possibility that some other task will be
4857 * attached to this cgroup before it is removed, or that some other
4858 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4859 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4860 * unused, and this cgroup will be reprieved from its death sentence,
4861 * to continue to serve a useful existence. Next time it's released,
4862 * we will get notified again, if it still has 'notify_on_release' set.
4864 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4865 * means only wait until the task is successfully execve()'d. The
4866 * separate release agent task is forked by call_usermodehelper(),
4867 * then control in this thread returns here, without waiting for the
4868 * release agent task. We don't bother to wait because the caller of
4869 * this routine has no use for the exit status of the release agent
4870 * task, so no sense holding our caller up for that.
4872 static void cgroup_release_agent(struct work_struct
*work
)
4874 BUG_ON(work
!= &release_agent_work
);
4875 mutex_lock(&cgroup_mutex
);
4876 raw_spin_lock(&release_list_lock
);
4877 while (!list_empty(&release_list
)) {
4878 char *argv
[3], *envp
[3];
4880 char *pathbuf
= NULL
, *agentbuf
= NULL
;
4881 struct cgroup
*cgrp
= list_entry(release_list
.next
,
4884 list_del_init(&cgrp
->release_list
);
4885 raw_spin_unlock(&release_list_lock
);
4886 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4889 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
4891 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
4896 argv
[i
++] = agentbuf
;
4897 argv
[i
++] = pathbuf
;
4901 /* minimal command environment */
4902 envp
[i
++] = "HOME=/";
4903 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4906 /* Drop the lock while we invoke the usermode helper,
4907 * since the exec could involve hitting disk and hence
4908 * be a slow process */
4909 mutex_unlock(&cgroup_mutex
);
4910 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
4911 mutex_lock(&cgroup_mutex
);
4915 raw_spin_lock(&release_list_lock
);
4917 raw_spin_unlock(&release_list_lock
);
4918 mutex_unlock(&cgroup_mutex
);
4921 static int __init
cgroup_disable(char *str
)
4926 while ((token
= strsep(&str
, ",")) != NULL
) {
4930 * cgroup_disable, being at boot time, can't know about module
4931 * subsystems, so we don't worry about them.
4933 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4934 struct cgroup_subsys
*ss
= subsys
[i
];
4936 if (!strcmp(token
, ss
->name
)) {
4938 printk(KERN_INFO
"Disabling %s control group"
4939 " subsystem\n", ss
->name
);
4946 __setup("cgroup_disable=", cgroup_disable
);
4949 * Functons for CSS ID.
4953 *To get ID other than 0, this should be called when !cgroup_is_removed().
4955 unsigned short css_id(struct cgroup_subsys_state
*css
)
4957 struct css_id
*cssid
;
4960 * This css_id() can return correct value when somone has refcnt
4961 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4962 * it's unchanged until freed.
4964 cssid
= rcu_dereference_check(css
->id
, atomic_read(&css
->refcnt
));
4970 EXPORT_SYMBOL_GPL(css_id
);
4972 unsigned short css_depth(struct cgroup_subsys_state
*css
)
4974 struct css_id
*cssid
;
4976 cssid
= rcu_dereference_check(css
->id
, atomic_read(&css
->refcnt
));
4979 return cssid
->depth
;
4982 EXPORT_SYMBOL_GPL(css_depth
);
4985 * css_is_ancestor - test "root" css is an ancestor of "child"
4986 * @child: the css to be tested.
4987 * @root: the css supporsed to be an ancestor of the child.
4989 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4990 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4991 * But, considering usual usage, the csses should be valid objects after test.
4992 * Assuming that the caller will do some action to the child if this returns
4993 * returns true, the caller must take "child";s reference count.
4994 * If "child" is valid object and this returns true, "root" is valid, too.
4997 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
4998 const struct cgroup_subsys_state
*root
)
5000 struct css_id
*child_id
;
5001 struct css_id
*root_id
;
5005 child_id
= rcu_dereference(child
->id
);
5006 root_id
= rcu_dereference(root
->id
);
5009 || (child_id
->depth
< root_id
->depth
)
5010 || (child_id
->stack
[root_id
->depth
] != root_id
->id
))
5016 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5018 struct css_id
*id
= css
->id
;
5019 /* When this is called before css_id initialization, id can be NULL */
5023 BUG_ON(!ss
->use_id
);
5025 rcu_assign_pointer(id
->css
, NULL
);
5026 rcu_assign_pointer(css
->id
, NULL
);
5027 spin_lock(&ss
->id_lock
);
5028 idr_remove(&ss
->idr
, id
->id
);
5029 spin_unlock(&ss
->id_lock
);
5030 kfree_rcu(id
, rcu_head
);
5032 EXPORT_SYMBOL_GPL(free_css_id
);
5035 * This is called by init or create(). Then, calls to this function are
5036 * always serialized (By cgroup_mutex() at create()).
5039 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5041 struct css_id
*newid
;
5042 int myid
, error
, size
;
5044 BUG_ON(!ss
->use_id
);
5046 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5047 newid
= kzalloc(size
, GFP_KERNEL
);
5049 return ERR_PTR(-ENOMEM
);
5051 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
5055 spin_lock(&ss
->id_lock
);
5056 /* Don't use 0. allocates an ID of 1-65535 */
5057 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
5058 spin_unlock(&ss
->id_lock
);
5060 /* Returns error when there are no free spaces for new ID.*/
5065 if (myid
> CSS_ID_MAX
)
5069 newid
->depth
= depth
;
5073 spin_lock(&ss
->id_lock
);
5074 idr_remove(&ss
->idr
, myid
);
5075 spin_unlock(&ss
->id_lock
);
5078 return ERR_PTR(error
);
5082 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5083 struct cgroup_subsys_state
*rootcss
)
5085 struct css_id
*newid
;
5087 spin_lock_init(&ss
->id_lock
);
5090 newid
= get_new_cssid(ss
, 0);
5092 return PTR_ERR(newid
);
5094 newid
->stack
[0] = newid
->id
;
5095 newid
->css
= rootcss
;
5096 rootcss
->id
= newid
;
5100 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5101 struct cgroup
*child
)
5103 int subsys_id
, i
, depth
= 0;
5104 struct cgroup_subsys_state
*parent_css
, *child_css
;
5105 struct css_id
*child_id
, *parent_id
;
5107 subsys_id
= ss
->subsys_id
;
5108 parent_css
= parent
->subsys
[subsys_id
];
5109 child_css
= child
->subsys
[subsys_id
];
5110 parent_id
= parent_css
->id
;
5111 depth
= parent_id
->depth
+ 1;
5113 child_id
= get_new_cssid(ss
, depth
);
5114 if (IS_ERR(child_id
))
5115 return PTR_ERR(child_id
);
5117 for (i
= 0; i
< depth
; i
++)
5118 child_id
->stack
[i
] = parent_id
->stack
[i
];
5119 child_id
->stack
[depth
] = child_id
->id
;
5121 * child_id->css pointer will be set after this cgroup is available
5122 * see cgroup_populate_dir()
5124 rcu_assign_pointer(child_css
->id
, child_id
);
5130 * css_lookup - lookup css by id
5131 * @ss: cgroup subsys to be looked into.
5134 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5135 * NULL if not. Should be called under rcu_read_lock()
5137 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5139 struct css_id
*cssid
= NULL
;
5141 BUG_ON(!ss
->use_id
);
5142 cssid
= idr_find(&ss
->idr
, id
);
5144 if (unlikely(!cssid
))
5147 return rcu_dereference(cssid
->css
);
5149 EXPORT_SYMBOL_GPL(css_lookup
);
5152 * css_get_next - lookup next cgroup under specified hierarchy.
5153 * @ss: pointer to subsystem
5154 * @id: current position of iteration.
5155 * @root: pointer to css. search tree under this.
5156 * @foundid: position of found object.
5158 * Search next css under the specified hierarchy of rootid. Calling under
5159 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5161 struct cgroup_subsys_state
*
5162 css_get_next(struct cgroup_subsys
*ss
, int id
,
5163 struct cgroup_subsys_state
*root
, int *foundid
)
5165 struct cgroup_subsys_state
*ret
= NULL
;
5168 int rootid
= css_id(root
);
5169 int depth
= css_depth(root
);
5174 BUG_ON(!ss
->use_id
);
5175 WARN_ON_ONCE(!rcu_read_lock_held());
5177 /* fill start point for scan */
5181 * scan next entry from bitmap(tree), tmpid is updated after
5184 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
5187 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
5188 ret
= rcu_dereference(tmp
->css
);
5194 /* continue to scan from next id */
5201 * get corresponding css from file open on cgroupfs directory
5203 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5205 struct cgroup
*cgrp
;
5206 struct inode
*inode
;
5207 struct cgroup_subsys_state
*css
;
5209 inode
= f
->f_dentry
->d_inode
;
5210 /* check in cgroup filesystem dir */
5211 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5212 return ERR_PTR(-EBADF
);
5214 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5215 return ERR_PTR(-EINVAL
);
5218 cgrp
= __d_cgrp(f
->f_dentry
);
5219 css
= cgrp
->subsys
[id
];
5220 return css
? css
: ERR_PTR(-ENOENT
);
5223 #ifdef CONFIG_CGROUP_DEBUG
5224 static struct cgroup_subsys_state
*debug_create(struct cgroup
*cont
)
5226 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5229 return ERR_PTR(-ENOMEM
);
5234 static void debug_destroy(struct cgroup
*cont
)
5236 kfree(cont
->subsys
[debug_subsys_id
]);
5239 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5241 return atomic_read(&cont
->count
);
5244 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5246 return cgroup_task_count(cont
);
5249 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5251 return (u64
)(unsigned long)current
->cgroups
;
5254 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5260 count
= atomic_read(¤t
->cgroups
->refcount
);
5265 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5267 struct seq_file
*seq
)
5269 struct cg_cgroup_link
*link
;
5272 read_lock(&css_set_lock
);
5274 cg
= rcu_dereference(current
->cgroups
);
5275 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5276 struct cgroup
*c
= link
->cgrp
;
5280 name
= c
->dentry
->d_name
.name
;
5283 seq_printf(seq
, "Root %d group %s\n",
5284 c
->root
->hierarchy_id
, name
);
5287 read_unlock(&css_set_lock
);
5291 #define MAX_TASKS_SHOWN_PER_CSS 25
5292 static int cgroup_css_links_read(struct cgroup
*cont
,
5294 struct seq_file
*seq
)
5296 struct cg_cgroup_link
*link
;
5298 read_lock(&css_set_lock
);
5299 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5300 struct css_set
*cg
= link
->cg
;
5301 struct task_struct
*task
;
5303 seq_printf(seq
, "css_set %p\n", cg
);
5304 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5305 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5306 seq_puts(seq
, " ...\n");
5309 seq_printf(seq
, " task %d\n",
5310 task_pid_vnr(task
));
5314 read_unlock(&css_set_lock
);
5318 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5320 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5323 static struct cftype debug_files
[] = {
5325 .name
= "cgroup_refcount",
5326 .read_u64
= cgroup_refcount_read
,
5329 .name
= "taskcount",
5330 .read_u64
= debug_taskcount_read
,
5334 .name
= "current_css_set",
5335 .read_u64
= current_css_set_read
,
5339 .name
= "current_css_set_refcount",
5340 .read_u64
= current_css_set_refcount_read
,
5344 .name
= "current_css_set_cg_links",
5345 .read_seq_string
= current_css_set_cg_links_read
,
5349 .name
= "cgroup_css_links",
5350 .read_seq_string
= cgroup_css_links_read
,
5354 .name
= "releasable",
5355 .read_u64
= releasable_read
,
5359 static int debug_populate(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5361 return cgroup_add_files(cont
, ss
, debug_files
,
5362 ARRAY_SIZE(debug_files
));
5365 struct cgroup_subsys debug_subsys
= {
5367 .create
= debug_create
,
5368 .destroy
= debug_destroy
,
5369 .populate
= debug_populate
,
5370 .subsys_id
= debug_subsys_id
,
5372 #endif /* CONFIG_CGROUP_DEBUG */