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>
66 /* css deactivation bias, makes css->refcnt negative to deny new trygets */
67 #define CSS_DEACT_BIAS INT_MIN
70 * cgroup_mutex is the master lock. Any modification to cgroup or its
71 * hierarchy must be performed while holding it.
73 * cgroup_root_mutex nests inside cgroup_mutex and should be held to modify
74 * cgroupfs_root of any cgroup hierarchy - subsys list, flags,
75 * release_agent_path and so on. Modifying requires both cgroup_mutex and
76 * cgroup_root_mutex. Readers can acquire either of the two. This is to
77 * break the following locking order cycle.
79 * A. cgroup_mutex -> cred_guard_mutex -> s_type->i_mutex_key -> namespace_sem
80 * B. namespace_sem -> cgroup_mutex
82 * B happens only through cgroup_show_options() and using cgroup_root_mutex
85 static DEFINE_MUTEX(cgroup_mutex
);
86 static DEFINE_MUTEX(cgroup_root_mutex
);
89 * Generate an array of cgroup subsystem pointers. At boot time, this is
90 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
91 * registered after that. The mutable section of this array is protected by
94 #define SUBSYS(_x) &_x ## _subsys,
95 static struct cgroup_subsys
*subsys
[CGROUP_SUBSYS_COUNT
] = {
96 #include <linux/cgroup_subsys.h>
99 #define MAX_CGROUP_ROOT_NAMELEN 64
102 * A cgroupfs_root represents the root of a cgroup hierarchy,
103 * and may be associated with a superblock to form an active
106 struct cgroupfs_root
{
107 struct super_block
*sb
;
110 * The bitmask of subsystems intended to be attached to this
113 unsigned long subsys_bits
;
115 /* Unique id for this hierarchy. */
118 /* The bitmask of subsystems currently attached to this hierarchy */
119 unsigned long actual_subsys_bits
;
121 /* A list running through the attached subsystems */
122 struct list_head subsys_list
;
124 /* The root cgroup for this hierarchy */
125 struct cgroup top_cgroup
;
127 /* Tracks how many cgroups are currently defined in hierarchy.*/
128 int number_of_cgroups
;
130 /* A list running through the active hierarchies */
131 struct list_head root_list
;
133 /* All cgroups on this root, cgroup_mutex protected */
134 struct list_head allcg_list
;
136 /* Hierarchy-specific flags */
139 /* The path to use for release notifications. */
140 char release_agent_path
[PATH_MAX
];
142 /* The name for this hierarchy - may be empty */
143 char name
[MAX_CGROUP_ROOT_NAMELEN
];
147 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
148 * subsystems that are otherwise unattached - it never has more than a
149 * single cgroup, and all tasks are part of that cgroup.
151 static struct cgroupfs_root rootnode
;
154 * cgroupfs file entry, pointed to from leaf dentry->d_fsdata.
157 struct list_head node
;
158 struct dentry
*dentry
;
163 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
164 * cgroup_subsys->use_id != 0.
166 #define CSS_ID_MAX (65535)
169 * The css to which this ID points. This pointer is set to valid value
170 * after cgroup is populated. If cgroup is removed, this will be NULL.
171 * This pointer is expected to be RCU-safe because destroy()
172 * is called after synchronize_rcu(). But for safe use, css_is_removed()
173 * css_tryget() should be used for avoiding race.
175 struct cgroup_subsys_state __rcu
*css
;
181 * Depth in hierarchy which this ID belongs to.
183 unsigned short depth
;
185 * ID is freed by RCU. (and lookup routine is RCU safe.)
187 struct rcu_head rcu_head
;
189 * Hierarchy of CSS ID belongs to.
191 unsigned short stack
[0]; /* Array of Length (depth+1) */
195 * cgroup_event represents events which userspace want to receive.
197 struct cgroup_event
{
199 * Cgroup which the event belongs to.
203 * Control file which the event associated.
207 * eventfd to signal userspace about the event.
209 struct eventfd_ctx
*eventfd
;
211 * Each of these stored in a list by the cgroup.
213 struct list_head list
;
215 * All fields below needed to unregister event when
216 * userspace closes eventfd.
219 wait_queue_head_t
*wqh
;
221 struct work_struct remove
;
224 /* The list of hierarchy roots */
226 static LIST_HEAD(roots
);
227 static int root_count
;
229 static DEFINE_IDA(hierarchy_ida
);
230 static int next_hierarchy_id
;
231 static DEFINE_SPINLOCK(hierarchy_id_lock
);
233 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
234 #define dummytop (&rootnode.top_cgroup)
236 /* This flag indicates whether tasks in the fork and exit paths should
237 * check for fork/exit handlers to call. This avoids us having to do
238 * extra work in the fork/exit path if none of the subsystems need to
241 static int need_forkexit_callback __read_mostly
;
243 #ifdef CONFIG_PROVE_LOCKING
244 int cgroup_lock_is_held(void)
246 return lockdep_is_held(&cgroup_mutex
);
248 #else /* #ifdef CONFIG_PROVE_LOCKING */
249 int cgroup_lock_is_held(void)
251 return mutex_is_locked(&cgroup_mutex
);
253 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
255 EXPORT_SYMBOL_GPL(cgroup_lock_is_held
);
257 /* the current nr of refs, always >= 0 whether @css is deactivated or not */
258 static int css_refcnt(struct cgroup_subsys_state
*css
)
260 int v
= atomic_read(&css
->refcnt
);
262 return v
>= 0 ? v
: v
- CSS_DEACT_BIAS
;
265 /* convenient tests for these bits */
266 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
268 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
271 /* bits in struct cgroupfs_root flags field */
273 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
276 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
279 (1 << CGRP_RELEASABLE
) |
280 (1 << CGRP_NOTIFY_ON_RELEASE
);
281 return (cgrp
->flags
& bits
) == bits
;
284 static int notify_on_release(const struct cgroup
*cgrp
)
286 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
289 static int clone_children(const struct cgroup
*cgrp
)
291 return test_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
295 * for_each_subsys() allows you to iterate on each subsystem attached to
296 * an active hierarchy
298 #define for_each_subsys(_root, _ss) \
299 list_for_each_entry(_ss, &_root->subsys_list, sibling)
301 /* for_each_active_root() allows you to iterate across the active hierarchies */
302 #define for_each_active_root(_root) \
303 list_for_each_entry(_root, &roots, root_list)
305 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
307 return dentry
->d_fsdata
;
310 static inline struct cfent
*__d_cfe(struct dentry
*dentry
)
312 return dentry
->d_fsdata
;
315 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
317 return __d_cfe(dentry
)->type
;
320 /* the list of cgroups eligible for automatic release. Protected by
321 * release_list_lock */
322 static LIST_HEAD(release_list
);
323 static DEFINE_RAW_SPINLOCK(release_list_lock
);
324 static void cgroup_release_agent(struct work_struct
*work
);
325 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
326 static void check_for_release(struct cgroup
*cgrp
);
328 /* Link structure for associating css_set objects with cgroups */
329 struct cg_cgroup_link
{
331 * List running through cg_cgroup_links associated with a
332 * cgroup, anchored on cgroup->css_sets
334 struct list_head cgrp_link_list
;
337 * List running through cg_cgroup_links pointing at a
338 * single css_set object, anchored on css_set->cg_links
340 struct list_head cg_link_list
;
344 /* The default css_set - used by init and its children prior to any
345 * hierarchies being mounted. It contains a pointer to the root state
346 * for each subsystem. Also used to anchor the list of css_sets. Not
347 * reference-counted, to improve performance when child cgroups
348 * haven't been created.
351 static struct css_set init_css_set
;
352 static struct cg_cgroup_link init_css_set_link
;
354 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
355 struct cgroup_subsys_state
*css
);
357 /* css_set_lock protects the list of css_set objects, and the
358 * chain of tasks off each css_set. Nests outside task->alloc_lock
359 * due to cgroup_iter_start() */
360 static DEFINE_RWLOCK(css_set_lock
);
361 static int css_set_count
;
364 * hash table for cgroup groups. This improves the performance to find
365 * an existing css_set. This hash doesn't (currently) take into
366 * account cgroups in empty hierarchies.
368 #define CSS_SET_HASH_BITS 7
369 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
370 static struct hlist_head css_set_table
[CSS_SET_TABLE_SIZE
];
372 static struct hlist_head
*css_set_hash(struct cgroup_subsys_state
*css
[])
376 unsigned long tmp
= 0UL;
378 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
379 tmp
+= (unsigned long)css
[i
];
380 tmp
= (tmp
>> 16) ^ tmp
;
382 index
= hash_long(tmp
, CSS_SET_HASH_BITS
);
384 return &css_set_table
[index
];
387 /* We don't maintain the lists running through each css_set to its
388 * task until after the first call to cgroup_iter_start(). This
389 * reduces the fork()/exit() overhead for people who have cgroups
390 * compiled into their kernel but not actually in use */
391 static int use_task_css_set_links __read_mostly
;
393 static void __put_css_set(struct css_set
*cg
, int taskexit
)
395 struct cg_cgroup_link
*link
;
396 struct cg_cgroup_link
*saved_link
;
398 * Ensure that the refcount doesn't hit zero while any readers
399 * can see it. Similar to atomic_dec_and_lock(), but for an
402 if (atomic_add_unless(&cg
->refcount
, -1, 1))
404 write_lock(&css_set_lock
);
405 if (!atomic_dec_and_test(&cg
->refcount
)) {
406 write_unlock(&css_set_lock
);
410 /* This css_set is dead. unlink it and release cgroup refcounts */
411 hlist_del(&cg
->hlist
);
414 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
416 struct cgroup
*cgrp
= link
->cgrp
;
417 list_del(&link
->cg_link_list
);
418 list_del(&link
->cgrp_link_list
);
419 if (atomic_dec_and_test(&cgrp
->count
) &&
420 notify_on_release(cgrp
)) {
422 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
423 check_for_release(cgrp
);
429 write_unlock(&css_set_lock
);
430 kfree_rcu(cg
, rcu_head
);
434 * refcounted get/put for css_set objects
436 static inline void get_css_set(struct css_set
*cg
)
438 atomic_inc(&cg
->refcount
);
441 static inline void put_css_set(struct css_set
*cg
)
443 __put_css_set(cg
, 0);
446 static inline void put_css_set_taskexit(struct css_set
*cg
)
448 __put_css_set(cg
, 1);
452 * compare_css_sets - helper function for find_existing_css_set().
453 * @cg: candidate css_set being tested
454 * @old_cg: existing css_set for a task
455 * @new_cgrp: cgroup that's being entered by the task
456 * @template: desired set of css pointers in css_set (pre-calculated)
458 * Returns true if "cg" matches "old_cg" except for the hierarchy
459 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
461 static bool compare_css_sets(struct css_set
*cg
,
462 struct css_set
*old_cg
,
463 struct cgroup
*new_cgrp
,
464 struct cgroup_subsys_state
*template[])
466 struct list_head
*l1
, *l2
;
468 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
469 /* Not all subsystems matched */
474 * Compare cgroup pointers in order to distinguish between
475 * different cgroups in heirarchies with no subsystems. We
476 * could get by with just this check alone (and skip the
477 * memcmp above) but on most setups the memcmp check will
478 * avoid the need for this more expensive check on almost all
483 l2
= &old_cg
->cg_links
;
485 struct cg_cgroup_link
*cgl1
, *cgl2
;
486 struct cgroup
*cg1
, *cg2
;
490 /* See if we reached the end - both lists are equal length. */
491 if (l1
== &cg
->cg_links
) {
492 BUG_ON(l2
!= &old_cg
->cg_links
);
495 BUG_ON(l2
== &old_cg
->cg_links
);
497 /* Locate the cgroups associated with these links. */
498 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
499 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
502 /* Hierarchies should be linked in the same order. */
503 BUG_ON(cg1
->root
!= cg2
->root
);
506 * If this hierarchy is the hierarchy of the cgroup
507 * that's changing, then we need to check that this
508 * css_set points to the new cgroup; if it's any other
509 * hierarchy, then this css_set should point to the
510 * same cgroup as the old css_set.
512 if (cg1
->root
== new_cgrp
->root
) {
524 * find_existing_css_set() is a helper for
525 * find_css_set(), and checks to see whether an existing
526 * css_set is suitable.
528 * oldcg: the cgroup group that we're using before the cgroup
531 * cgrp: the cgroup that we're moving into
533 * template: location in which to build the desired set of subsystem
534 * state objects for the new cgroup group
536 static struct css_set
*find_existing_css_set(
537 struct css_set
*oldcg
,
539 struct cgroup_subsys_state
*template[])
542 struct cgroupfs_root
*root
= cgrp
->root
;
543 struct hlist_head
*hhead
;
544 struct hlist_node
*node
;
548 * Build the set of subsystem state objects that we want to see in the
549 * new css_set. while subsystems can change globally, the entries here
550 * won't change, so no need for locking.
552 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
553 if (root
->subsys_bits
& (1UL << i
)) {
554 /* Subsystem is in this hierarchy. So we want
555 * the subsystem state from the new
557 template[i
] = cgrp
->subsys
[i
];
559 /* Subsystem is not in this hierarchy, so we
560 * don't want to change the subsystem state */
561 template[i
] = oldcg
->subsys
[i
];
565 hhead
= css_set_hash(template);
566 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
567 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
570 /* This css_set matches what we need */
574 /* No existing cgroup group matched */
578 static void free_cg_links(struct list_head
*tmp
)
580 struct cg_cgroup_link
*link
;
581 struct cg_cgroup_link
*saved_link
;
583 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
584 list_del(&link
->cgrp_link_list
);
590 * allocate_cg_links() allocates "count" cg_cgroup_link structures
591 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
592 * success or a negative error
594 static int allocate_cg_links(int count
, struct list_head
*tmp
)
596 struct cg_cgroup_link
*link
;
599 for (i
= 0; i
< count
; i
++) {
600 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
605 list_add(&link
->cgrp_link_list
, tmp
);
611 * link_css_set - a helper function to link a css_set to a cgroup
612 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
613 * @cg: the css_set to be linked
614 * @cgrp: the destination cgroup
616 static void link_css_set(struct list_head
*tmp_cg_links
,
617 struct css_set
*cg
, struct cgroup
*cgrp
)
619 struct cg_cgroup_link
*link
;
621 BUG_ON(list_empty(tmp_cg_links
));
622 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
626 atomic_inc(&cgrp
->count
);
627 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
629 * Always add links to the tail of the list so that the list
630 * is sorted by order of hierarchy creation
632 list_add_tail(&link
->cg_link_list
, &cg
->cg_links
);
636 * find_css_set() takes an existing cgroup group and a
637 * cgroup object, and returns a css_set object that's
638 * equivalent to the old group, but with the given cgroup
639 * substituted into the appropriate hierarchy. Must be called with
642 static struct css_set
*find_css_set(
643 struct css_set
*oldcg
, struct cgroup
*cgrp
)
646 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
648 struct list_head tmp_cg_links
;
650 struct hlist_head
*hhead
;
651 struct cg_cgroup_link
*link
;
653 /* First see if we already have a cgroup group that matches
655 read_lock(&css_set_lock
);
656 res
= find_existing_css_set(oldcg
, cgrp
, template);
659 read_unlock(&css_set_lock
);
664 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
668 /* Allocate all the cg_cgroup_link objects that we'll need */
669 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
674 atomic_set(&res
->refcount
, 1);
675 INIT_LIST_HEAD(&res
->cg_links
);
676 INIT_LIST_HEAD(&res
->tasks
);
677 INIT_HLIST_NODE(&res
->hlist
);
679 /* Copy the set of subsystem state objects generated in
680 * find_existing_css_set() */
681 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
683 write_lock(&css_set_lock
);
684 /* Add reference counts and links from the new css_set. */
685 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
686 struct cgroup
*c
= link
->cgrp
;
687 if (c
->root
== cgrp
->root
)
689 link_css_set(&tmp_cg_links
, res
, c
);
692 BUG_ON(!list_empty(&tmp_cg_links
));
696 /* Add this cgroup group to the hash table */
697 hhead
= css_set_hash(res
->subsys
);
698 hlist_add_head(&res
->hlist
, hhead
);
700 write_unlock(&css_set_lock
);
706 * Return the cgroup for "task" from the given hierarchy. Must be
707 * called with cgroup_mutex held.
709 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
710 struct cgroupfs_root
*root
)
713 struct cgroup
*res
= NULL
;
715 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
716 read_lock(&css_set_lock
);
718 * No need to lock the task - since we hold cgroup_mutex the
719 * task can't change groups, so the only thing that can happen
720 * is that it exits and its css is set back to init_css_set.
723 if (css
== &init_css_set
) {
724 res
= &root
->top_cgroup
;
726 struct cg_cgroup_link
*link
;
727 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
728 struct cgroup
*c
= link
->cgrp
;
729 if (c
->root
== root
) {
735 read_unlock(&css_set_lock
);
741 * There is one global cgroup mutex. We also require taking
742 * task_lock() when dereferencing a task's cgroup subsys pointers.
743 * See "The task_lock() exception", at the end of this comment.
745 * A task must hold cgroup_mutex to modify cgroups.
747 * Any task can increment and decrement the count field without lock.
748 * So in general, code holding cgroup_mutex can't rely on the count
749 * field not changing. However, if the count goes to zero, then only
750 * cgroup_attach_task() can increment it again. Because a count of zero
751 * means that no tasks are currently attached, therefore there is no
752 * way a task attached to that cgroup can fork (the other way to
753 * increment the count). So code holding cgroup_mutex can safely
754 * assume that if the count is zero, it will stay zero. Similarly, if
755 * a task holds cgroup_mutex on a cgroup with zero count, it
756 * knows that the cgroup won't be removed, as cgroup_rmdir()
759 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
760 * (usually) take cgroup_mutex. These are the two most performance
761 * critical pieces of code here. The exception occurs on cgroup_exit(),
762 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
763 * is taken, and if the cgroup count is zero, a usermode call made
764 * to the release agent with the name of the cgroup (path relative to
765 * the root of cgroup file system) as the argument.
767 * A cgroup can only be deleted if both its 'count' of using tasks
768 * is zero, and its list of 'children' cgroups is empty. Since all
769 * tasks in the system use _some_ cgroup, and since there is always at
770 * least one task in the system (init, pid == 1), therefore, top_cgroup
771 * always has either children cgroups and/or using tasks. So we don't
772 * need a special hack to ensure that top_cgroup cannot be deleted.
774 * The task_lock() exception
776 * The need for this exception arises from the action of
777 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
778 * another. It does so using cgroup_mutex, however there are
779 * several performance critical places that need to reference
780 * task->cgroup without the expense of grabbing a system global
781 * mutex. Therefore except as noted below, when dereferencing or, as
782 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
783 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
784 * the task_struct routinely used for such matters.
786 * P.S. One more locking exception. RCU is used to guard the
787 * update of a tasks cgroup pointer by cgroup_attach_task()
791 * cgroup_lock - lock out any changes to cgroup structures
794 void cgroup_lock(void)
796 mutex_lock(&cgroup_mutex
);
798 EXPORT_SYMBOL_GPL(cgroup_lock
);
801 * cgroup_unlock - release lock on cgroup changes
803 * Undo the lock taken in a previous cgroup_lock() call.
805 void cgroup_unlock(void)
807 mutex_unlock(&cgroup_mutex
);
809 EXPORT_SYMBOL_GPL(cgroup_unlock
);
812 * A couple of forward declarations required, due to cyclic reference loop:
813 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
814 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
818 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
);
819 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
820 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
821 static int cgroup_populate_dir(struct cgroup
*cgrp
);
822 static const struct inode_operations cgroup_dir_inode_operations
;
823 static const struct file_operations proc_cgroupstats_operations
;
825 static struct backing_dev_info cgroup_backing_dev_info
= {
827 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
830 static int alloc_css_id(struct cgroup_subsys
*ss
,
831 struct cgroup
*parent
, struct cgroup
*child
);
833 static struct inode
*cgroup_new_inode(umode_t mode
, struct super_block
*sb
)
835 struct inode
*inode
= new_inode(sb
);
838 inode
->i_ino
= get_next_ino();
839 inode
->i_mode
= mode
;
840 inode
->i_uid
= current_fsuid();
841 inode
->i_gid
= current_fsgid();
842 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
843 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
849 * Call subsys's pre_destroy handler.
850 * This is called before css refcnt check.
852 static int cgroup_call_pre_destroy(struct cgroup
*cgrp
)
854 struct cgroup_subsys
*ss
;
857 for_each_subsys(cgrp
->root
, ss
) {
858 if (!ss
->pre_destroy
)
861 ret
= ss
->pre_destroy(cgrp
);
863 /* ->pre_destroy() failure is being deprecated */
864 WARN_ON_ONCE(!ss
->__DEPRECATED_clear_css_refs
);
872 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
874 /* is dentry a directory ? if so, kfree() associated cgroup */
875 if (S_ISDIR(inode
->i_mode
)) {
876 struct cgroup
*cgrp
= dentry
->d_fsdata
;
877 struct cgroup_subsys
*ss
;
878 BUG_ON(!(cgroup_is_removed(cgrp
)));
879 /* It's possible for external users to be holding css
880 * reference counts on a cgroup; css_put() needs to
881 * be able to access the cgroup after decrementing
882 * the reference count in order to know if it needs to
883 * queue the cgroup to be handled by the release
887 mutex_lock(&cgroup_mutex
);
889 * Release the subsystem state objects.
891 for_each_subsys(cgrp
->root
, ss
)
894 cgrp
->root
->number_of_cgroups
--;
895 mutex_unlock(&cgroup_mutex
);
898 * Drop the active superblock reference that we took when we
901 deactivate_super(cgrp
->root
->sb
);
904 * if we're getting rid of the cgroup, refcount should ensure
905 * that there are no pidlists left.
907 BUG_ON(!list_empty(&cgrp
->pidlists
));
909 kfree_rcu(cgrp
, rcu_head
);
911 struct cfent
*cfe
= __d_cfe(dentry
);
912 struct cgroup
*cgrp
= dentry
->d_parent
->d_fsdata
;
914 WARN_ONCE(!list_empty(&cfe
->node
) &&
915 cgrp
!= &cgrp
->root
->top_cgroup
,
916 "cfe still linked for %s\n", cfe
->type
->name
);
922 static int cgroup_delete(const struct dentry
*d
)
927 static void remove_dir(struct dentry
*d
)
929 struct dentry
*parent
= dget(d
->d_parent
);
932 simple_rmdir(parent
->d_inode
, d
);
936 static int cgroup_rm_file(struct cgroup
*cgrp
, const struct cftype
*cft
)
940 lockdep_assert_held(&cgrp
->dentry
->d_inode
->i_mutex
);
941 lockdep_assert_held(&cgroup_mutex
);
943 list_for_each_entry(cfe
, &cgrp
->files
, node
) {
944 struct dentry
*d
= cfe
->dentry
;
946 if (cft
&& cfe
->type
!= cft
)
951 simple_unlink(d
->d_inode
, d
);
952 list_del_init(&cfe
->node
);
960 static void cgroup_clear_directory(struct dentry
*dir
)
962 struct cgroup
*cgrp
= __d_cgrp(dir
);
964 while (!list_empty(&cgrp
->files
))
965 cgroup_rm_file(cgrp
, NULL
);
969 * NOTE : the dentry must have been dget()'ed
971 static void cgroup_d_remove_dir(struct dentry
*dentry
)
973 struct dentry
*parent
;
975 cgroup_clear_directory(dentry
);
977 parent
= dentry
->d_parent
;
978 spin_lock(&parent
->d_lock
);
979 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
980 list_del_init(&dentry
->d_u
.d_child
);
981 spin_unlock(&dentry
->d_lock
);
982 spin_unlock(&parent
->d_lock
);
987 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
988 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
989 * reference to css->refcnt. In general, this refcnt is expected to goes down
992 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
994 static DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq
);
996 static void cgroup_wakeup_rmdir_waiter(struct cgroup
*cgrp
)
998 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
)))
999 wake_up_all(&cgroup_rmdir_waitq
);
1002 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
)
1007 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
)
1009 cgroup_wakeup_rmdir_waiter(css
->cgroup
);
1014 * Call with cgroup_mutex held. Drops reference counts on modules, including
1015 * any duplicate ones that parse_cgroupfs_options took. If this function
1016 * returns an error, no reference counts are touched.
1018 static int rebind_subsystems(struct cgroupfs_root
*root
,
1019 unsigned long final_bits
)
1021 unsigned long added_bits
, removed_bits
;
1022 struct cgroup
*cgrp
= &root
->top_cgroup
;
1025 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1026 BUG_ON(!mutex_is_locked(&cgroup_root_mutex
));
1028 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
1029 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
1030 /* Check that any added subsystems are currently free */
1031 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1032 unsigned long bit
= 1UL << i
;
1033 struct cgroup_subsys
*ss
= subsys
[i
];
1034 if (!(bit
& added_bits
))
1037 * Nobody should tell us to do a subsys that doesn't exist:
1038 * parse_cgroupfs_options should catch that case and refcounts
1039 * ensure that subsystems won't disappear once selected.
1042 if (ss
->root
!= &rootnode
) {
1043 /* Subsystem isn't free */
1048 /* Currently we don't handle adding/removing subsystems when
1049 * any child cgroups exist. This is theoretically supportable
1050 * but involves complex error handling, so it's being left until
1052 if (root
->number_of_cgroups
> 1)
1055 /* Process each subsystem */
1056 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1057 struct cgroup_subsys
*ss
= subsys
[i
];
1058 unsigned long bit
= 1UL << i
;
1059 if (bit
& added_bits
) {
1060 /* We're binding this subsystem to this hierarchy */
1062 BUG_ON(cgrp
->subsys
[i
]);
1063 BUG_ON(!dummytop
->subsys
[i
]);
1064 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
1065 mutex_lock(&ss
->hierarchy_mutex
);
1066 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
1067 cgrp
->subsys
[i
]->cgroup
= cgrp
;
1068 list_move(&ss
->sibling
, &root
->subsys_list
);
1072 mutex_unlock(&ss
->hierarchy_mutex
);
1073 /* refcount was already taken, and we're keeping it */
1074 } else if (bit
& removed_bits
) {
1075 /* We're removing this subsystem */
1077 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1078 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1079 mutex_lock(&ss
->hierarchy_mutex
);
1082 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1083 cgrp
->subsys
[i
] = NULL
;
1084 subsys
[i
]->root
= &rootnode
;
1085 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1086 mutex_unlock(&ss
->hierarchy_mutex
);
1087 /* subsystem is now free - drop reference on module */
1088 module_put(ss
->module
);
1089 } else if (bit
& final_bits
) {
1090 /* Subsystem state should already exist */
1092 BUG_ON(!cgrp
->subsys
[i
]);
1094 * a refcount was taken, but we already had one, so
1095 * drop the extra reference.
1097 module_put(ss
->module
);
1098 #ifdef CONFIG_MODULE_UNLOAD
1099 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1102 /* Subsystem state shouldn't exist */
1103 BUG_ON(cgrp
->subsys
[i
]);
1106 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
1112 static int cgroup_show_options(struct seq_file
*seq
, struct dentry
*dentry
)
1114 struct cgroupfs_root
*root
= dentry
->d_sb
->s_fs_info
;
1115 struct cgroup_subsys
*ss
;
1117 mutex_lock(&cgroup_root_mutex
);
1118 for_each_subsys(root
, ss
)
1119 seq_printf(seq
, ",%s", ss
->name
);
1120 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1121 seq_puts(seq
, ",noprefix");
1122 if (strlen(root
->release_agent_path
))
1123 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1124 if (clone_children(&root
->top_cgroup
))
1125 seq_puts(seq
, ",clone_children");
1126 if (strlen(root
->name
))
1127 seq_printf(seq
, ",name=%s", root
->name
);
1128 mutex_unlock(&cgroup_root_mutex
);
1132 struct cgroup_sb_opts
{
1133 unsigned long subsys_bits
;
1134 unsigned long flags
;
1135 char *release_agent
;
1136 bool clone_children
;
1138 /* User explicitly requested empty subsystem */
1141 struct cgroupfs_root
*new_root
;
1146 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1147 * with cgroup_mutex held to protect the subsys[] array. This function takes
1148 * refcounts on subsystems to be used, unless it returns error, in which case
1149 * no refcounts are taken.
1151 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1153 char *token
, *o
= data
;
1154 bool all_ss
= false, one_ss
= false;
1155 unsigned long mask
= (unsigned long)-1;
1157 bool module_pin_failed
= false;
1159 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1161 #ifdef CONFIG_CPUSETS
1162 mask
= ~(1UL << cpuset_subsys_id
);
1165 memset(opts
, 0, sizeof(*opts
));
1167 while ((token
= strsep(&o
, ",")) != NULL
) {
1170 if (!strcmp(token
, "none")) {
1171 /* Explicitly have no subsystems */
1175 if (!strcmp(token
, "all")) {
1176 /* Mutually exclusive option 'all' + subsystem name */
1182 if (!strcmp(token
, "noprefix")) {
1183 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1186 if (!strcmp(token
, "clone_children")) {
1187 opts
->clone_children
= true;
1190 if (!strncmp(token
, "release_agent=", 14)) {
1191 /* Specifying two release agents is forbidden */
1192 if (opts
->release_agent
)
1194 opts
->release_agent
=
1195 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1196 if (!opts
->release_agent
)
1200 if (!strncmp(token
, "name=", 5)) {
1201 const char *name
= token
+ 5;
1202 /* Can't specify an empty name */
1205 /* Must match [\w.-]+ */
1206 for (i
= 0; i
< strlen(name
); i
++) {
1210 if ((c
== '.') || (c
== '-') || (c
== '_'))
1214 /* Specifying two names is forbidden */
1217 opts
->name
= kstrndup(name
,
1218 MAX_CGROUP_ROOT_NAMELEN
- 1,
1226 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1227 struct cgroup_subsys
*ss
= subsys
[i
];
1230 if (strcmp(token
, ss
->name
))
1235 /* Mutually exclusive option 'all' + subsystem name */
1238 set_bit(i
, &opts
->subsys_bits
);
1243 if (i
== CGROUP_SUBSYS_COUNT
)
1248 * If the 'all' option was specified select all the subsystems,
1249 * otherwise if 'none', 'name=' and a subsystem name options
1250 * were not specified, let's default to 'all'
1252 if (all_ss
|| (!one_ss
&& !opts
->none
&& !opts
->name
)) {
1253 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1254 struct cgroup_subsys
*ss
= subsys
[i
];
1259 set_bit(i
, &opts
->subsys_bits
);
1263 /* Consistency checks */
1266 * Option noprefix was introduced just for backward compatibility
1267 * with the old cpuset, so we allow noprefix only if mounting just
1268 * the cpuset subsystem.
1270 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1271 (opts
->subsys_bits
& mask
))
1275 /* Can't specify "none" and some subsystems */
1276 if (opts
->subsys_bits
&& opts
->none
)
1280 * We either have to specify by name or by subsystems. (So all
1281 * empty hierarchies must have a name).
1283 if (!opts
->subsys_bits
&& !opts
->name
)
1287 * Grab references on all the modules we'll need, so the subsystems
1288 * don't dance around before rebind_subsystems attaches them. This may
1289 * take duplicate reference counts on a subsystem that's already used,
1290 * but rebind_subsystems handles this case.
1292 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1293 unsigned long bit
= 1UL << i
;
1295 if (!(bit
& opts
->subsys_bits
))
1297 if (!try_module_get(subsys
[i
]->module
)) {
1298 module_pin_failed
= true;
1302 if (module_pin_failed
) {
1304 * oops, one of the modules was going away. this means that we
1305 * raced with a module_delete call, and to the user this is
1306 * essentially a "subsystem doesn't exist" case.
1308 for (i
--; i
>= CGROUP_BUILTIN_SUBSYS_COUNT
; i
--) {
1309 /* drop refcounts only on the ones we took */
1310 unsigned long bit
= 1UL << i
;
1312 if (!(bit
& opts
->subsys_bits
))
1314 module_put(subsys
[i
]->module
);
1322 static void drop_parsed_module_refcounts(unsigned long subsys_bits
)
1325 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1326 unsigned long bit
= 1UL << i
;
1328 if (!(bit
& subsys_bits
))
1330 module_put(subsys
[i
]->module
);
1334 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1337 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1338 struct cgroup
*cgrp
= &root
->top_cgroup
;
1339 struct cgroup_sb_opts opts
;
1341 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1342 mutex_lock(&cgroup_mutex
);
1343 mutex_lock(&cgroup_root_mutex
);
1345 /* See what subsystems are wanted */
1346 ret
= parse_cgroupfs_options(data
, &opts
);
1350 /* See feature-removal-schedule.txt */
1351 if (opts
.subsys_bits
!= root
->actual_subsys_bits
|| opts
.release_agent
)
1352 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1353 task_tgid_nr(current
), current
->comm
);
1355 /* Don't allow flags or name to change at remount */
1356 if (opts
.flags
!= root
->flags
||
1357 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1359 drop_parsed_module_refcounts(opts
.subsys_bits
);
1363 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
1365 drop_parsed_module_refcounts(opts
.subsys_bits
);
1369 /* clear out any existing files and repopulate subsystem files */
1370 cgroup_clear_directory(cgrp
->dentry
);
1371 cgroup_populate_dir(cgrp
);
1373 if (opts
.release_agent
)
1374 strcpy(root
->release_agent_path
, opts
.release_agent
);
1376 kfree(opts
.release_agent
);
1378 mutex_unlock(&cgroup_root_mutex
);
1379 mutex_unlock(&cgroup_mutex
);
1380 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1384 static const struct super_operations cgroup_ops
= {
1385 .statfs
= simple_statfs
,
1386 .drop_inode
= generic_delete_inode
,
1387 .show_options
= cgroup_show_options
,
1388 .remount_fs
= cgroup_remount
,
1391 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1393 INIT_LIST_HEAD(&cgrp
->sibling
);
1394 INIT_LIST_HEAD(&cgrp
->children
);
1395 INIT_LIST_HEAD(&cgrp
->files
);
1396 INIT_LIST_HEAD(&cgrp
->css_sets
);
1397 INIT_LIST_HEAD(&cgrp
->release_list
);
1398 INIT_LIST_HEAD(&cgrp
->pidlists
);
1399 mutex_init(&cgrp
->pidlist_mutex
);
1400 INIT_LIST_HEAD(&cgrp
->event_list
);
1401 spin_lock_init(&cgrp
->event_list_lock
);
1404 static void init_cgroup_root(struct cgroupfs_root
*root
)
1406 struct cgroup
*cgrp
= &root
->top_cgroup
;
1408 INIT_LIST_HEAD(&root
->subsys_list
);
1409 INIT_LIST_HEAD(&root
->root_list
);
1410 INIT_LIST_HEAD(&root
->allcg_list
);
1411 root
->number_of_cgroups
= 1;
1413 cgrp
->top_cgroup
= cgrp
;
1414 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
1415 init_cgroup_housekeeping(cgrp
);
1418 static bool init_root_id(struct cgroupfs_root
*root
)
1423 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1425 spin_lock(&hierarchy_id_lock
);
1426 /* Try to allocate the next unused ID */
1427 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1428 &root
->hierarchy_id
);
1430 /* Try again starting from 0 */
1431 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1433 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1434 } else if (ret
!= -EAGAIN
) {
1435 /* Can only get here if the 31-bit IDR is full ... */
1438 spin_unlock(&hierarchy_id_lock
);
1443 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1445 struct cgroup_sb_opts
*opts
= data
;
1446 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1448 /* If we asked for a name then it must match */
1449 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1453 * If we asked for subsystems (or explicitly for no
1454 * subsystems) then they must match
1456 if ((opts
->subsys_bits
|| opts
->none
)
1457 && (opts
->subsys_bits
!= root
->subsys_bits
))
1463 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1465 struct cgroupfs_root
*root
;
1467 if (!opts
->subsys_bits
&& !opts
->none
)
1470 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1472 return ERR_PTR(-ENOMEM
);
1474 if (!init_root_id(root
)) {
1476 return ERR_PTR(-ENOMEM
);
1478 init_cgroup_root(root
);
1480 root
->subsys_bits
= opts
->subsys_bits
;
1481 root
->flags
= opts
->flags
;
1482 if (opts
->release_agent
)
1483 strcpy(root
->release_agent_path
, opts
->release_agent
);
1485 strcpy(root
->name
, opts
->name
);
1486 if (opts
->clone_children
)
1487 set_bit(CGRP_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1491 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1496 BUG_ON(!root
->hierarchy_id
);
1497 spin_lock(&hierarchy_id_lock
);
1498 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1499 spin_unlock(&hierarchy_id_lock
);
1503 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1506 struct cgroup_sb_opts
*opts
= data
;
1508 /* If we don't have a new root, we can't set up a new sb */
1509 if (!opts
->new_root
)
1512 BUG_ON(!opts
->subsys_bits
&& !opts
->none
);
1514 ret
= set_anon_super(sb
, NULL
);
1518 sb
->s_fs_info
= opts
->new_root
;
1519 opts
->new_root
->sb
= sb
;
1521 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1522 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1523 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1524 sb
->s_op
= &cgroup_ops
;
1529 static int cgroup_get_rootdir(struct super_block
*sb
)
1531 static const struct dentry_operations cgroup_dops
= {
1532 .d_iput
= cgroup_diput
,
1533 .d_delete
= cgroup_delete
,
1536 struct inode
*inode
=
1537 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1542 inode
->i_fop
= &simple_dir_operations
;
1543 inode
->i_op
= &cgroup_dir_inode_operations
;
1544 /* directories start off with i_nlink == 2 (for "." entry) */
1546 sb
->s_root
= d_make_root(inode
);
1549 /* for everything else we want ->d_op set */
1550 sb
->s_d_op
= &cgroup_dops
;
1554 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1555 int flags
, const char *unused_dev_name
,
1558 struct cgroup_sb_opts opts
;
1559 struct cgroupfs_root
*root
;
1561 struct super_block
*sb
;
1562 struct cgroupfs_root
*new_root
;
1563 struct inode
*inode
;
1565 /* First find the desired set of subsystems */
1566 mutex_lock(&cgroup_mutex
);
1567 ret
= parse_cgroupfs_options(data
, &opts
);
1568 mutex_unlock(&cgroup_mutex
);
1573 * Allocate a new cgroup root. We may not need it if we're
1574 * reusing an existing hierarchy.
1576 new_root
= cgroup_root_from_opts(&opts
);
1577 if (IS_ERR(new_root
)) {
1578 ret
= PTR_ERR(new_root
);
1581 opts
.new_root
= new_root
;
1583 /* Locate an existing or new sb for this hierarchy */
1584 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, &opts
);
1587 cgroup_drop_root(opts
.new_root
);
1591 root
= sb
->s_fs_info
;
1593 if (root
== opts
.new_root
) {
1594 /* We used the new root structure, so this is a new hierarchy */
1595 struct list_head tmp_cg_links
;
1596 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1597 struct cgroupfs_root
*existing_root
;
1598 const struct cred
*cred
;
1601 BUG_ON(sb
->s_root
!= NULL
);
1603 ret
= cgroup_get_rootdir(sb
);
1605 goto drop_new_super
;
1606 inode
= sb
->s_root
->d_inode
;
1608 mutex_lock(&inode
->i_mutex
);
1609 mutex_lock(&cgroup_mutex
);
1610 mutex_lock(&cgroup_root_mutex
);
1612 /* Check for name clashes with existing mounts */
1614 if (strlen(root
->name
))
1615 for_each_active_root(existing_root
)
1616 if (!strcmp(existing_root
->name
, root
->name
))
1620 * We're accessing css_set_count without locking
1621 * css_set_lock here, but that's OK - it can only be
1622 * increased by someone holding cgroup_lock, and
1623 * that's us. The worst that can happen is that we
1624 * have some link structures left over
1626 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1630 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1631 if (ret
== -EBUSY
) {
1632 free_cg_links(&tmp_cg_links
);
1636 * There must be no failure case after here, since rebinding
1637 * takes care of subsystems' refcounts, which are explicitly
1638 * dropped in the failure exit path.
1641 /* EBUSY should be the only error here */
1644 list_add(&root
->root_list
, &roots
);
1647 sb
->s_root
->d_fsdata
= root_cgrp
;
1648 root
->top_cgroup
.dentry
= sb
->s_root
;
1650 /* Link the top cgroup in this hierarchy into all
1651 * the css_set objects */
1652 write_lock(&css_set_lock
);
1653 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1654 struct hlist_head
*hhead
= &css_set_table
[i
];
1655 struct hlist_node
*node
;
1658 hlist_for_each_entry(cg
, node
, hhead
, hlist
)
1659 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1661 write_unlock(&css_set_lock
);
1663 free_cg_links(&tmp_cg_links
);
1665 BUG_ON(!list_empty(&root_cgrp
->sibling
));
1666 BUG_ON(!list_empty(&root_cgrp
->children
));
1667 BUG_ON(root
->number_of_cgroups
!= 1);
1669 cred
= override_creds(&init_cred
);
1670 cgroup_populate_dir(root_cgrp
);
1672 mutex_unlock(&cgroup_root_mutex
);
1673 mutex_unlock(&cgroup_mutex
);
1674 mutex_unlock(&inode
->i_mutex
);
1677 * We re-used an existing hierarchy - the new root (if
1678 * any) is not needed
1680 cgroup_drop_root(opts
.new_root
);
1681 /* no subsys rebinding, so refcounts don't change */
1682 drop_parsed_module_refcounts(opts
.subsys_bits
);
1685 kfree(opts
.release_agent
);
1687 return dget(sb
->s_root
);
1690 mutex_unlock(&cgroup_root_mutex
);
1691 mutex_unlock(&cgroup_mutex
);
1692 mutex_unlock(&inode
->i_mutex
);
1694 deactivate_locked_super(sb
);
1696 drop_parsed_module_refcounts(opts
.subsys_bits
);
1698 kfree(opts
.release_agent
);
1700 return ERR_PTR(ret
);
1703 static void cgroup_kill_sb(struct super_block
*sb
) {
1704 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1705 struct cgroup
*cgrp
= &root
->top_cgroup
;
1707 struct cg_cgroup_link
*link
;
1708 struct cg_cgroup_link
*saved_link
;
1712 BUG_ON(root
->number_of_cgroups
!= 1);
1713 BUG_ON(!list_empty(&cgrp
->children
));
1714 BUG_ON(!list_empty(&cgrp
->sibling
));
1716 mutex_lock(&cgroup_mutex
);
1717 mutex_lock(&cgroup_root_mutex
);
1719 /* Rebind all subsystems back to the default hierarchy */
1720 ret
= rebind_subsystems(root
, 0);
1721 /* Shouldn't be able to fail ... */
1725 * Release all the links from css_sets to this hierarchy's
1728 write_lock(&css_set_lock
);
1730 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1732 list_del(&link
->cg_link_list
);
1733 list_del(&link
->cgrp_link_list
);
1736 write_unlock(&css_set_lock
);
1738 if (!list_empty(&root
->root_list
)) {
1739 list_del(&root
->root_list
);
1743 mutex_unlock(&cgroup_root_mutex
);
1744 mutex_unlock(&cgroup_mutex
);
1746 kill_litter_super(sb
);
1747 cgroup_drop_root(root
);
1750 static struct file_system_type cgroup_fs_type
= {
1752 .mount
= cgroup_mount
,
1753 .kill_sb
= cgroup_kill_sb
,
1756 static struct kobject
*cgroup_kobj
;
1759 * cgroup_path - generate the path of a cgroup
1760 * @cgrp: the cgroup in question
1761 * @buf: the buffer to write the path into
1762 * @buflen: the length of the buffer
1764 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1765 * reference. Writes path of cgroup into buf. Returns 0 on success,
1768 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1771 struct dentry
*dentry
= rcu_dereference_check(cgrp
->dentry
,
1772 cgroup_lock_is_held());
1774 if (!dentry
|| cgrp
== dummytop
) {
1776 * Inactive subsystems have no dentry for their root
1783 start
= buf
+ buflen
;
1787 int len
= dentry
->d_name
.len
;
1789 if ((start
-= len
) < buf
)
1790 return -ENAMETOOLONG
;
1791 memcpy(start
, dentry
->d_name
.name
, len
);
1792 cgrp
= cgrp
->parent
;
1796 dentry
= rcu_dereference_check(cgrp
->dentry
,
1797 cgroup_lock_is_held());
1801 return -ENAMETOOLONG
;
1804 memmove(buf
, start
, buf
+ buflen
- start
);
1807 EXPORT_SYMBOL_GPL(cgroup_path
);
1810 * Control Group taskset
1812 struct task_and_cgroup
{
1813 struct task_struct
*task
;
1814 struct cgroup
*cgrp
;
1818 struct cgroup_taskset
{
1819 struct task_and_cgroup single
;
1820 struct flex_array
*tc_array
;
1823 struct cgroup
*cur_cgrp
;
1827 * cgroup_taskset_first - reset taskset and return the first task
1828 * @tset: taskset of interest
1830 * @tset iteration is initialized and the first task is returned.
1832 struct task_struct
*cgroup_taskset_first(struct cgroup_taskset
*tset
)
1834 if (tset
->tc_array
) {
1836 return cgroup_taskset_next(tset
);
1838 tset
->cur_cgrp
= tset
->single
.cgrp
;
1839 return tset
->single
.task
;
1842 EXPORT_SYMBOL_GPL(cgroup_taskset_first
);
1845 * cgroup_taskset_next - iterate to the next task in taskset
1846 * @tset: taskset of interest
1848 * Return the next task in @tset. Iteration must have been initialized
1849 * with cgroup_taskset_first().
1851 struct task_struct
*cgroup_taskset_next(struct cgroup_taskset
*tset
)
1853 struct task_and_cgroup
*tc
;
1855 if (!tset
->tc_array
|| tset
->idx
>= tset
->tc_array_len
)
1858 tc
= flex_array_get(tset
->tc_array
, tset
->idx
++);
1859 tset
->cur_cgrp
= tc
->cgrp
;
1862 EXPORT_SYMBOL_GPL(cgroup_taskset_next
);
1865 * cgroup_taskset_cur_cgroup - return the matching cgroup for the current task
1866 * @tset: taskset of interest
1868 * Return the cgroup for the current (last returned) task of @tset. This
1869 * function must be preceded by either cgroup_taskset_first() or
1870 * cgroup_taskset_next().
1872 struct cgroup
*cgroup_taskset_cur_cgroup(struct cgroup_taskset
*tset
)
1874 return tset
->cur_cgrp
;
1876 EXPORT_SYMBOL_GPL(cgroup_taskset_cur_cgroup
);
1879 * cgroup_taskset_size - return the number of tasks in taskset
1880 * @tset: taskset of interest
1882 int cgroup_taskset_size(struct cgroup_taskset
*tset
)
1884 return tset
->tc_array
? tset
->tc_array_len
: 1;
1886 EXPORT_SYMBOL_GPL(cgroup_taskset_size
);
1890 * cgroup_task_migrate - move a task from one cgroup to another.
1892 * 'guarantee' is set if the caller promises that a new css_set for the task
1893 * will already exist. If not set, this function might sleep, and can fail with
1894 * -ENOMEM. Must be called with cgroup_mutex and threadgroup locked.
1896 static void cgroup_task_migrate(struct cgroup
*cgrp
, struct cgroup
*oldcgrp
,
1897 struct task_struct
*tsk
, struct css_set
*newcg
)
1899 struct css_set
*oldcg
;
1902 * We are synchronized through threadgroup_lock() against PF_EXITING
1903 * setting such that we can't race against cgroup_exit() changing the
1904 * css_set to init_css_set and dropping the old one.
1906 WARN_ON_ONCE(tsk
->flags
& PF_EXITING
);
1907 oldcg
= tsk
->cgroups
;
1910 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1913 /* Update the css_set linked lists if we're using them */
1914 write_lock(&css_set_lock
);
1915 if (!list_empty(&tsk
->cg_list
))
1916 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1917 write_unlock(&css_set_lock
);
1920 * We just gained a reference on oldcg by taking it from the task. As
1921 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1922 * it here; it will be freed under RCU.
1926 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1930 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1931 * @cgrp: the cgroup the task is attaching to
1932 * @tsk: the task to be attached
1934 * Call with cgroup_mutex and threadgroup locked. May take task_lock of
1937 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1940 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1941 struct cgroup
*oldcgrp
;
1942 struct cgroupfs_root
*root
= cgrp
->root
;
1943 struct cgroup_taskset tset
= { };
1944 struct css_set
*newcg
;
1946 /* @tsk either already exited or can't exit until the end */
1947 if (tsk
->flags
& PF_EXITING
)
1950 /* Nothing to do if the task is already in that cgroup */
1951 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1952 if (cgrp
== oldcgrp
)
1955 tset
.single
.task
= tsk
;
1956 tset
.single
.cgrp
= oldcgrp
;
1958 for_each_subsys(root
, ss
) {
1959 if (ss
->can_attach
) {
1960 retval
= ss
->can_attach(cgrp
, &tset
);
1963 * Remember on which subsystem the can_attach()
1964 * failed, so that we only call cancel_attach()
1965 * against the subsystems whose can_attach()
1966 * succeeded. (See below)
1974 newcg
= find_css_set(tsk
->cgroups
, cgrp
);
1980 cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, newcg
);
1982 for_each_subsys(root
, ss
) {
1984 ss
->attach(cgrp
, &tset
);
1990 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1991 * is no longer empty.
1993 cgroup_wakeup_rmdir_waiter(cgrp
);
1996 for_each_subsys(root
, ss
) {
1997 if (ss
== failed_ss
)
1999 * This subsystem was the one that failed the
2000 * can_attach() check earlier, so we don't need
2001 * to call cancel_attach() against it or any
2002 * remaining subsystems.
2005 if (ss
->cancel_attach
)
2006 ss
->cancel_attach(cgrp
, &tset
);
2013 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
2014 * @from: attach to all cgroups of a given task
2015 * @tsk: the task to be attached
2017 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
2019 struct cgroupfs_root
*root
;
2023 for_each_active_root(root
) {
2024 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
2026 retval
= cgroup_attach_task(from_cg
, tsk
);
2034 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
2037 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
2038 * @cgrp: the cgroup to attach to
2039 * @leader: the threadgroup leader task_struct of the group to be attached
2041 * Call holding cgroup_mutex and the group_rwsem of the leader. Will take
2042 * task_lock of each thread in leader's threadgroup individually in turn.
2044 static int cgroup_attach_proc(struct cgroup
*cgrp
, struct task_struct
*leader
)
2046 int retval
, i
, group_size
;
2047 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
2048 /* guaranteed to be initialized later, but the compiler needs this */
2049 struct cgroupfs_root
*root
= cgrp
->root
;
2050 /* threadgroup list cursor and array */
2051 struct task_struct
*tsk
;
2052 struct task_and_cgroup
*tc
;
2053 struct flex_array
*group
;
2054 struct cgroup_taskset tset
= { };
2057 * step 0: in order to do expensive, possibly blocking operations for
2058 * every thread, we cannot iterate the thread group list, since it needs
2059 * rcu or tasklist locked. instead, build an array of all threads in the
2060 * group - group_rwsem prevents new threads from appearing, and if
2061 * threads exit, this will just be an over-estimate.
2063 group_size
= get_nr_threads(leader
);
2064 /* flex_array supports very large thread-groups better than kmalloc. */
2065 group
= flex_array_alloc(sizeof(*tc
), group_size
, GFP_KERNEL
);
2068 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2069 retval
= flex_array_prealloc(group
, 0, group_size
- 1, GFP_KERNEL
);
2071 goto out_free_group_list
;
2076 * Prevent freeing of tasks while we take a snapshot. Tasks that are
2077 * already PF_EXITING could be freed from underneath us unless we
2078 * take an rcu_read_lock.
2082 struct task_and_cgroup ent
;
2084 /* @tsk either already exited or can't exit until the end */
2085 if (tsk
->flags
& PF_EXITING
)
2088 /* as per above, nr_threads may decrease, but not increase. */
2089 BUG_ON(i
>= group_size
);
2091 ent
.cgrp
= task_cgroup_from_root(tsk
, root
);
2092 /* nothing to do if this task is already in the cgroup */
2093 if (ent
.cgrp
== cgrp
)
2096 * saying GFP_ATOMIC has no effect here because we did prealloc
2097 * earlier, but it's good form to communicate our expectations.
2099 retval
= flex_array_put(group
, i
, &ent
, GFP_ATOMIC
);
2100 BUG_ON(retval
!= 0);
2102 } while_each_thread(leader
, tsk
);
2104 /* remember the number of threads in the array for later. */
2106 tset
.tc_array
= group
;
2107 tset
.tc_array_len
= group_size
;
2109 /* methods shouldn't be called if no task is actually migrating */
2112 goto out_free_group_list
;
2115 * step 1: check that we can legitimately attach to the cgroup.
2117 for_each_subsys(root
, ss
) {
2118 if (ss
->can_attach
) {
2119 retval
= ss
->can_attach(cgrp
, &tset
);
2122 goto out_cancel_attach
;
2128 * step 2: make sure css_sets exist for all threads to be migrated.
2129 * we use find_css_set, which allocates a new one if necessary.
2131 for (i
= 0; i
< group_size
; i
++) {
2132 tc
= flex_array_get(group
, i
);
2133 tc
->cg
= find_css_set(tc
->task
->cgroups
, cgrp
);
2136 goto out_put_css_set_refs
;
2141 * step 3: now that we're guaranteed success wrt the css_sets,
2142 * proceed to move all tasks to the new cgroup. There are no
2143 * failure cases after here, so this is the commit point.
2145 for (i
= 0; i
< group_size
; i
++) {
2146 tc
= flex_array_get(group
, i
);
2147 cgroup_task_migrate(cgrp
, tc
->cgrp
, tc
->task
, tc
->cg
);
2149 /* nothing is sensitive to fork() after this point. */
2152 * step 4: do subsystem attach callbacks.
2154 for_each_subsys(root
, ss
) {
2156 ss
->attach(cgrp
, &tset
);
2160 * step 5: success! and cleanup
2163 cgroup_wakeup_rmdir_waiter(cgrp
);
2165 out_put_css_set_refs
:
2167 for (i
= 0; i
< group_size
; i
++) {
2168 tc
= flex_array_get(group
, i
);
2171 put_css_set(tc
->cg
);
2176 for_each_subsys(root
, ss
) {
2177 if (ss
== failed_ss
)
2179 if (ss
->cancel_attach
)
2180 ss
->cancel_attach(cgrp
, &tset
);
2183 out_free_group_list
:
2184 flex_array_free(group
);
2189 * Find the task_struct of the task to attach by vpid and pass it along to the
2190 * function to attach either it or all tasks in its threadgroup. Will lock
2191 * cgroup_mutex and threadgroup; may take task_lock of task.
2193 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2195 struct task_struct
*tsk
;
2196 const struct cred
*cred
= current_cred(), *tcred
;
2199 if (!cgroup_lock_live_group(cgrp
))
2205 tsk
= find_task_by_vpid(pid
);
2209 goto out_unlock_cgroup
;
2212 * even if we're attaching all tasks in the thread group, we
2213 * only need to check permissions on one of them.
2215 tcred
= __task_cred(tsk
);
2217 cred
->euid
!= tcred
->uid
&&
2218 cred
->euid
!= tcred
->suid
) {
2221 goto out_unlock_cgroup
;
2227 tsk
= tsk
->group_leader
;
2228 get_task_struct(tsk
);
2231 threadgroup_lock(tsk
);
2233 if (!thread_group_leader(tsk
)) {
2235 * a race with de_thread from another thread's exec()
2236 * may strip us of our leadership, if this happens,
2237 * there is no choice but to throw this task away and
2238 * try again; this is
2239 * "double-double-toil-and-trouble-check locking".
2241 threadgroup_unlock(tsk
);
2242 put_task_struct(tsk
);
2243 goto retry_find_task
;
2245 ret
= cgroup_attach_proc(cgrp
, tsk
);
2247 ret
= cgroup_attach_task(cgrp
, tsk
);
2248 threadgroup_unlock(tsk
);
2250 put_task_struct(tsk
);
2256 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2258 return attach_task_by_pid(cgrp
, pid
, false);
2261 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2263 return attach_task_by_pid(cgrp
, tgid
, true);
2267 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2268 * @cgrp: the cgroup to be checked for liveness
2270 * On success, returns true; the lock should be later released with
2271 * cgroup_unlock(). On failure returns false with no lock held.
2273 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
2275 mutex_lock(&cgroup_mutex
);
2276 if (cgroup_is_removed(cgrp
)) {
2277 mutex_unlock(&cgroup_mutex
);
2282 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
2284 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2287 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2288 if (strlen(buffer
) >= PATH_MAX
)
2290 if (!cgroup_lock_live_group(cgrp
))
2292 mutex_lock(&cgroup_root_mutex
);
2293 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2294 mutex_unlock(&cgroup_root_mutex
);
2299 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2300 struct seq_file
*seq
)
2302 if (!cgroup_lock_live_group(cgrp
))
2304 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2305 seq_putc(seq
, '\n');
2310 /* A buffer size big enough for numbers or short strings */
2311 #define CGROUP_LOCAL_BUFFER_SIZE 64
2313 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2315 const char __user
*userbuf
,
2316 size_t nbytes
, loff_t
*unused_ppos
)
2318 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2324 if (nbytes
>= sizeof(buffer
))
2326 if (copy_from_user(buffer
, userbuf
, nbytes
))
2329 buffer
[nbytes
] = 0; /* nul-terminate */
2330 if (cft
->write_u64
) {
2331 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2334 retval
= cft
->write_u64(cgrp
, cft
, val
);
2336 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2339 retval
= cft
->write_s64(cgrp
, cft
, val
);
2346 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2348 const char __user
*userbuf
,
2349 size_t nbytes
, loff_t
*unused_ppos
)
2351 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2353 size_t max_bytes
= cft
->max_write_len
;
2354 char *buffer
= local_buffer
;
2357 max_bytes
= sizeof(local_buffer
) - 1;
2358 if (nbytes
>= max_bytes
)
2360 /* Allocate a dynamic buffer if we need one */
2361 if (nbytes
>= sizeof(local_buffer
)) {
2362 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2366 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2371 buffer
[nbytes
] = 0; /* nul-terminate */
2372 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2376 if (buffer
!= local_buffer
)
2381 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2382 size_t nbytes
, loff_t
*ppos
)
2384 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2385 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2387 if (cgroup_is_removed(cgrp
))
2390 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2391 if (cft
->write_u64
|| cft
->write_s64
)
2392 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2393 if (cft
->write_string
)
2394 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2396 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2397 return ret
? ret
: nbytes
;
2402 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2404 char __user
*buf
, size_t nbytes
,
2407 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2408 u64 val
= cft
->read_u64(cgrp
, cft
);
2409 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2411 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2414 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2416 char __user
*buf
, size_t nbytes
,
2419 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2420 s64 val
= cft
->read_s64(cgrp
, cft
);
2421 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2423 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2426 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2427 size_t nbytes
, loff_t
*ppos
)
2429 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2430 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2432 if (cgroup_is_removed(cgrp
))
2436 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2438 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2440 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2445 * seqfile ops/methods for returning structured data. Currently just
2446 * supports string->u64 maps, but can be extended in future.
2449 struct cgroup_seqfile_state
{
2451 struct cgroup
*cgroup
;
2454 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2456 struct seq_file
*sf
= cb
->state
;
2457 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2460 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2462 struct cgroup_seqfile_state
*state
= m
->private;
2463 struct cftype
*cft
= state
->cft
;
2464 if (cft
->read_map
) {
2465 struct cgroup_map_cb cb
= {
2466 .fill
= cgroup_map_add
,
2469 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2471 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2474 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2476 struct seq_file
*seq
= file
->private_data
;
2477 kfree(seq
->private);
2478 return single_release(inode
, file
);
2481 static const struct file_operations cgroup_seqfile_operations
= {
2483 .write
= cgroup_file_write
,
2484 .llseek
= seq_lseek
,
2485 .release
= cgroup_seqfile_release
,
2488 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2493 err
= generic_file_open(inode
, file
);
2496 cft
= __d_cft(file
->f_dentry
);
2498 if (cft
->read_map
|| cft
->read_seq_string
) {
2499 struct cgroup_seqfile_state
*state
=
2500 kzalloc(sizeof(*state
), GFP_USER
);
2504 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2505 file
->f_op
= &cgroup_seqfile_operations
;
2506 err
= single_open(file
, cgroup_seqfile_show
, state
);
2509 } else if (cft
->open
)
2510 err
= cft
->open(inode
, file
);
2517 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2519 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2521 return cft
->release(inode
, file
);
2526 * cgroup_rename - Only allow simple rename of directories in place.
2528 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2529 struct inode
*new_dir
, struct dentry
*new_dentry
)
2531 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2533 if (new_dentry
->d_inode
)
2535 if (old_dir
!= new_dir
)
2537 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2540 static const struct file_operations cgroup_file_operations
= {
2541 .read
= cgroup_file_read
,
2542 .write
= cgroup_file_write
,
2543 .llseek
= generic_file_llseek
,
2544 .open
= cgroup_file_open
,
2545 .release
= cgroup_file_release
,
2548 static const struct inode_operations cgroup_dir_inode_operations
= {
2549 .lookup
= cgroup_lookup
,
2550 .mkdir
= cgroup_mkdir
,
2551 .rmdir
= cgroup_rmdir
,
2552 .rename
= cgroup_rename
,
2555 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
2557 if (dentry
->d_name
.len
> NAME_MAX
)
2558 return ERR_PTR(-ENAMETOOLONG
);
2559 d_add(dentry
, NULL
);
2564 * Check if a file is a control file
2566 static inline struct cftype
*__file_cft(struct file
*file
)
2568 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2569 return ERR_PTR(-EINVAL
);
2570 return __d_cft(file
->f_dentry
);
2573 static int cgroup_create_file(struct dentry
*dentry
, umode_t mode
,
2574 struct super_block
*sb
)
2576 struct inode
*inode
;
2580 if (dentry
->d_inode
)
2583 inode
= cgroup_new_inode(mode
, sb
);
2587 if (S_ISDIR(mode
)) {
2588 inode
->i_op
= &cgroup_dir_inode_operations
;
2589 inode
->i_fop
= &simple_dir_operations
;
2591 /* start off with i_nlink == 2 (for "." entry) */
2594 /* start with the directory inode held, so that we can
2595 * populate it without racing with another mkdir */
2596 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2597 } else if (S_ISREG(mode
)) {
2599 inode
->i_fop
= &cgroup_file_operations
;
2601 d_instantiate(dentry
, inode
);
2602 dget(dentry
); /* Extra count - pin the dentry in core */
2607 * cgroup_create_dir - create a directory for an object.
2608 * @cgrp: the cgroup we create the directory for. It must have a valid
2609 * ->parent field. And we are going to fill its ->dentry field.
2610 * @dentry: dentry of the new cgroup
2611 * @mode: mode to set on new directory.
2613 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
2616 struct dentry
*parent
;
2619 parent
= cgrp
->parent
->dentry
;
2620 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
2622 dentry
->d_fsdata
= cgrp
;
2623 inc_nlink(parent
->d_inode
);
2624 rcu_assign_pointer(cgrp
->dentry
, dentry
);
2633 * cgroup_file_mode - deduce file mode of a control file
2634 * @cft: the control file in question
2636 * returns cft->mode if ->mode is not 0
2637 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2638 * returns S_IRUGO if it has only a read handler
2639 * returns S_IWUSR if it has only a write hander
2641 static umode_t
cgroup_file_mode(const struct cftype
*cft
)
2648 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2649 cft
->read_map
|| cft
->read_seq_string
)
2652 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2653 cft
->write_string
|| cft
->trigger
)
2659 static int cgroup_add_file(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2660 const struct cftype
*cft
)
2662 struct dentry
*dir
= cgrp
->dentry
;
2663 struct cgroup
*parent
= __d_cgrp(dir
);
2664 struct dentry
*dentry
;
2668 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2670 /* does @cft->flags tell us to skip creation on @cgrp? */
2671 if ((cft
->flags
& CFTYPE_NOT_ON_ROOT
) && !cgrp
->parent
)
2673 if ((cft
->flags
& CFTYPE_ONLY_ON_ROOT
) && cgrp
->parent
)
2676 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2677 strcpy(name
, subsys
->name
);
2680 strcat(name
, cft
->name
);
2682 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2684 cfe
= kzalloc(sizeof(*cfe
), GFP_KERNEL
);
2688 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2689 if (IS_ERR(dentry
)) {
2690 error
= PTR_ERR(dentry
);
2694 mode
= cgroup_file_mode(cft
);
2695 error
= cgroup_create_file(dentry
, mode
| S_IFREG
, cgrp
->root
->sb
);
2697 cfe
->type
= (void *)cft
;
2698 cfe
->dentry
= dentry
;
2699 dentry
->d_fsdata
= cfe
;
2700 list_add_tail(&cfe
->node
, &parent
->files
);
2709 static int cgroup_addrm_files(struct cgroup
*cgrp
, struct cgroup_subsys
*subsys
,
2710 const struct cftype cfts
[], bool is_add
)
2712 const struct cftype
*cft
;
2715 for (cft
= cfts
; cft
->name
[0] != '\0'; cft
++) {
2717 err
= cgroup_add_file(cgrp
, subsys
, cft
);
2719 err
= cgroup_rm_file(cgrp
, cft
);
2721 pr_warning("cgroup_addrm_files: failed to %s %s, err=%d\n",
2722 is_add
? "add" : "remove", cft
->name
, err
);
2729 static DEFINE_MUTEX(cgroup_cft_mutex
);
2731 static void cgroup_cfts_prepare(void)
2732 __acquires(&cgroup_cft_mutex
) __acquires(&cgroup_mutex
)
2735 * Thanks to the entanglement with vfs inode locking, we can't walk
2736 * the existing cgroups under cgroup_mutex and create files.
2737 * Instead, we increment reference on all cgroups and build list of
2738 * them using @cgrp->cft_q_node. Grab cgroup_cft_mutex to ensure
2739 * exclusive access to the field.
2741 mutex_lock(&cgroup_cft_mutex
);
2742 mutex_lock(&cgroup_mutex
);
2745 static void cgroup_cfts_commit(struct cgroup_subsys
*ss
,
2746 const struct cftype
*cfts
, bool is_add
)
2747 __releases(&cgroup_mutex
) __releases(&cgroup_cft_mutex
)
2750 struct cgroup
*cgrp
, *n
;
2752 /* %NULL @cfts indicates abort and don't bother if @ss isn't attached */
2753 if (cfts
&& ss
->root
!= &rootnode
) {
2754 list_for_each_entry(cgrp
, &ss
->root
->allcg_list
, allcg_node
) {
2756 list_add_tail(&cgrp
->cft_q_node
, &pending
);
2760 mutex_unlock(&cgroup_mutex
);
2763 * All new cgroups will see @cfts update on @ss->cftsets. Add/rm
2764 * files for all cgroups which were created before.
2766 list_for_each_entry_safe(cgrp
, n
, &pending
, cft_q_node
) {
2767 struct inode
*inode
= cgrp
->dentry
->d_inode
;
2769 mutex_lock(&inode
->i_mutex
);
2770 mutex_lock(&cgroup_mutex
);
2771 if (!cgroup_is_removed(cgrp
))
2772 cgroup_addrm_files(cgrp
, ss
, cfts
, is_add
);
2773 mutex_unlock(&cgroup_mutex
);
2774 mutex_unlock(&inode
->i_mutex
);
2776 list_del_init(&cgrp
->cft_q_node
);
2780 mutex_unlock(&cgroup_cft_mutex
);
2784 * cgroup_add_cftypes - add an array of cftypes to a subsystem
2785 * @ss: target cgroup subsystem
2786 * @cfts: zero-length name terminated array of cftypes
2788 * Register @cfts to @ss. Files described by @cfts are created for all
2789 * existing cgroups to which @ss is attached and all future cgroups will
2790 * have them too. This function can be called anytime whether @ss is
2793 * Returns 0 on successful registration, -errno on failure. Note that this
2794 * function currently returns 0 as long as @cfts registration is successful
2795 * even if some file creation attempts on existing cgroups fail.
2797 int cgroup_add_cftypes(struct cgroup_subsys
*ss
, const struct cftype
*cfts
)
2799 struct cftype_set
*set
;
2801 set
= kzalloc(sizeof(*set
), GFP_KERNEL
);
2805 cgroup_cfts_prepare();
2807 list_add_tail(&set
->node
, &ss
->cftsets
);
2808 cgroup_cfts_commit(ss
, cfts
, true);
2812 EXPORT_SYMBOL_GPL(cgroup_add_cftypes
);
2815 * cgroup_rm_cftypes - remove an array of cftypes from a subsystem
2816 * @ss: target cgroup subsystem
2817 * @cfts: zero-length name terminated array of cftypes
2819 * Unregister @cfts from @ss. Files described by @cfts are removed from
2820 * all existing cgroups to which @ss is attached and all future cgroups
2821 * won't have them either. This function can be called anytime whether @ss
2822 * is attached or not.
2824 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2825 * registered with @ss.
2827 int cgroup_rm_cftypes(struct cgroup_subsys
*ss
, const struct cftype
*cfts
)
2829 struct cftype_set
*set
;
2831 cgroup_cfts_prepare();
2833 list_for_each_entry(set
, &ss
->cftsets
, node
) {
2834 if (set
->cfts
== cfts
) {
2835 list_del_init(&set
->node
);
2836 cgroup_cfts_commit(ss
, cfts
, false);
2841 cgroup_cfts_commit(ss
, NULL
, false);
2846 * cgroup_task_count - count the number of tasks in a cgroup.
2847 * @cgrp: the cgroup in question
2849 * Return the number of tasks in the cgroup.
2851 int cgroup_task_count(const struct cgroup
*cgrp
)
2854 struct cg_cgroup_link
*link
;
2856 read_lock(&css_set_lock
);
2857 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2858 count
+= atomic_read(&link
->cg
->refcount
);
2860 read_unlock(&css_set_lock
);
2865 * Advance a list_head iterator. The iterator should be positioned at
2866 * the start of a css_set
2868 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2869 struct cgroup_iter
*it
)
2871 struct list_head
*l
= it
->cg_link
;
2872 struct cg_cgroup_link
*link
;
2875 /* Advance to the next non-empty css_set */
2878 if (l
== &cgrp
->css_sets
) {
2882 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2884 } while (list_empty(&cg
->tasks
));
2886 it
->task
= cg
->tasks
.next
;
2890 * To reduce the fork() overhead for systems that are not actually
2891 * using their cgroups capability, we don't maintain the lists running
2892 * through each css_set to its tasks until we see the list actually
2893 * used - in other words after the first call to cgroup_iter_start().
2895 static void cgroup_enable_task_cg_lists(void)
2897 struct task_struct
*p
, *g
;
2898 write_lock(&css_set_lock
);
2899 use_task_css_set_links
= 1;
2901 * We need tasklist_lock because RCU is not safe against
2902 * while_each_thread(). Besides, a forking task that has passed
2903 * cgroup_post_fork() without seeing use_task_css_set_links = 1
2904 * is not guaranteed to have its child immediately visible in the
2905 * tasklist if we walk through it with RCU.
2907 read_lock(&tasklist_lock
);
2908 do_each_thread(g
, p
) {
2911 * We should check if the process is exiting, otherwise
2912 * it will race with cgroup_exit() in that the list
2913 * entry won't be deleted though the process has exited.
2915 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2916 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2918 } while_each_thread(g
, p
);
2919 read_unlock(&tasklist_lock
);
2920 write_unlock(&css_set_lock
);
2923 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2924 __acquires(css_set_lock
)
2927 * The first time anyone tries to iterate across a cgroup,
2928 * we need to enable the list linking each css_set to its
2929 * tasks, and fix up all existing tasks.
2931 if (!use_task_css_set_links
)
2932 cgroup_enable_task_cg_lists();
2934 read_lock(&css_set_lock
);
2935 it
->cg_link
= &cgrp
->css_sets
;
2936 cgroup_advance_iter(cgrp
, it
);
2939 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
2940 struct cgroup_iter
*it
)
2942 struct task_struct
*res
;
2943 struct list_head
*l
= it
->task
;
2944 struct cg_cgroup_link
*link
;
2946 /* If the iterator cg is NULL, we have no tasks */
2949 res
= list_entry(l
, struct task_struct
, cg_list
);
2950 /* Advance iterator to find next entry */
2952 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
2953 if (l
== &link
->cg
->tasks
) {
2954 /* We reached the end of this task list - move on to
2955 * the next cg_cgroup_link */
2956 cgroup_advance_iter(cgrp
, it
);
2963 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2964 __releases(css_set_lock
)
2966 read_unlock(&css_set_lock
);
2969 static inline int started_after_time(struct task_struct
*t1
,
2970 struct timespec
*time
,
2971 struct task_struct
*t2
)
2973 int start_diff
= timespec_compare(&t1
->start_time
, time
);
2974 if (start_diff
> 0) {
2976 } else if (start_diff
< 0) {
2980 * Arbitrarily, if two processes started at the same
2981 * time, we'll say that the lower pointer value
2982 * started first. Note that t2 may have exited by now
2983 * so this may not be a valid pointer any longer, but
2984 * that's fine - it still serves to distinguish
2985 * between two tasks started (effectively) simultaneously.
2992 * This function is a callback from heap_insert() and is used to order
2994 * In this case we order the heap in descending task start time.
2996 static inline int started_after(void *p1
, void *p2
)
2998 struct task_struct
*t1
= p1
;
2999 struct task_struct
*t2
= p2
;
3000 return started_after_time(t1
, &t2
->start_time
, t2
);
3004 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
3005 * @scan: struct cgroup_scanner containing arguments for the scan
3007 * Arguments include pointers to callback functions test_task() and
3009 * Iterate through all the tasks in a cgroup, calling test_task() for each,
3010 * and if it returns true, call process_task() for it also.
3011 * The test_task pointer may be NULL, meaning always true (select all tasks).
3012 * Effectively duplicates cgroup_iter_{start,next,end}()
3013 * but does not lock css_set_lock for the call to process_task().
3014 * The struct cgroup_scanner may be embedded in any structure of the caller's
3016 * It is guaranteed that process_task() will act on every task that
3017 * is a member of the cgroup for the duration of this call. This
3018 * function may or may not call process_task() for tasks that exit
3019 * or move to a different cgroup during the call, or are forked or
3020 * move into the cgroup during the call.
3022 * Note that test_task() may be called with locks held, and may in some
3023 * situations be called multiple times for the same task, so it should
3025 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
3026 * pre-allocated and will be used for heap operations (and its "gt" member will
3027 * be overwritten), else a temporary heap will be used (allocation of which
3028 * may cause this function to fail).
3030 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
3033 struct cgroup_iter it
;
3034 struct task_struct
*p
, *dropped
;
3035 /* Never dereference latest_task, since it's not refcounted */
3036 struct task_struct
*latest_task
= NULL
;
3037 struct ptr_heap tmp_heap
;
3038 struct ptr_heap
*heap
;
3039 struct timespec latest_time
= { 0, 0 };
3042 /* The caller supplied our heap and pre-allocated its memory */
3044 heap
->gt
= &started_after
;
3046 /* We need to allocate our own heap memory */
3048 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
3050 /* cannot allocate the heap */
3056 * Scan tasks in the cgroup, using the scanner's "test_task" callback
3057 * to determine which are of interest, and using the scanner's
3058 * "process_task" callback to process any of them that need an update.
3059 * Since we don't want to hold any locks during the task updates,
3060 * gather tasks to be processed in a heap structure.
3061 * The heap is sorted by descending task start time.
3062 * If the statically-sized heap fills up, we overflow tasks that
3063 * started later, and in future iterations only consider tasks that
3064 * started after the latest task in the previous pass. This
3065 * guarantees forward progress and that we don't miss any tasks.
3068 cgroup_iter_start(scan
->cg
, &it
);
3069 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
3071 * Only affect tasks that qualify per the caller's callback,
3072 * if he provided one
3074 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
3077 * Only process tasks that started after the last task
3080 if (!started_after_time(p
, &latest_time
, latest_task
))
3082 dropped
= heap_insert(heap
, p
);
3083 if (dropped
== NULL
) {
3085 * The new task was inserted; the heap wasn't
3089 } else if (dropped
!= p
) {
3091 * The new task was inserted, and pushed out a
3095 put_task_struct(dropped
);
3098 * Else the new task was newer than anything already in
3099 * the heap and wasn't inserted
3102 cgroup_iter_end(scan
->cg
, &it
);
3105 for (i
= 0; i
< heap
->size
; i
++) {
3106 struct task_struct
*q
= heap
->ptrs
[i
];
3108 latest_time
= q
->start_time
;
3111 /* Process the task per the caller's callback */
3112 scan
->process_task(q
, scan
);
3116 * If we had to process any tasks at all, scan again
3117 * in case some of them were in the middle of forking
3118 * children that didn't get processed.
3119 * Not the most efficient way to do it, but it avoids
3120 * having to take callback_mutex in the fork path
3124 if (heap
== &tmp_heap
)
3125 heap_free(&tmp_heap
);
3130 * Stuff for reading the 'tasks'/'procs' files.
3132 * Reading this file can return large amounts of data if a cgroup has
3133 * *lots* of attached tasks. So it may need several calls to read(),
3134 * but we cannot guarantee that the information we produce is correct
3135 * unless we produce it entirely atomically.
3139 /* which pidlist file are we talking about? */
3140 enum cgroup_filetype
{
3146 * A pidlist is a list of pids that virtually represents the contents of one
3147 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
3148 * a pair (one each for procs, tasks) for each pid namespace that's relevant
3151 struct cgroup_pidlist
{
3153 * used to find which pidlist is wanted. doesn't change as long as
3154 * this particular list stays in the list.
3156 struct { enum cgroup_filetype type
; struct pid_namespace
*ns
; } key
;
3159 /* how many elements the above list has */
3161 /* how many files are using the current array */
3163 /* each of these stored in a list by its cgroup */
3164 struct list_head links
;
3165 /* pointer to the cgroup we belong to, for list removal purposes */
3166 struct cgroup
*owner
;
3167 /* protects the other fields */
3168 struct rw_semaphore mutex
;
3172 * The following two functions "fix" the issue where there are more pids
3173 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3174 * TODO: replace with a kernel-wide solution to this problem
3176 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3177 static void *pidlist_allocate(int count
)
3179 if (PIDLIST_TOO_LARGE(count
))
3180 return vmalloc(count
* sizeof(pid_t
));
3182 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3184 static void pidlist_free(void *p
)
3186 if (is_vmalloc_addr(p
))
3191 static void *pidlist_resize(void *p
, int newcount
)
3194 /* note: if new alloc fails, old p will still be valid either way */
3195 if (is_vmalloc_addr(p
)) {
3196 newlist
= vmalloc(newcount
* sizeof(pid_t
));
3199 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
3202 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
3208 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3209 * If the new stripped list is sufficiently smaller and there's enough memory
3210 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3211 * number of unique elements.
3213 /* is the size difference enough that we should re-allocate the array? */
3214 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3215 static int pidlist_uniq(pid_t
**p
, int length
)
3222 * we presume the 0th element is unique, so i starts at 1. trivial
3223 * edge cases first; no work needs to be done for either
3225 if (length
== 0 || length
== 1)
3227 /* src and dest walk down the list; dest counts unique elements */
3228 for (src
= 1; src
< length
; src
++) {
3229 /* find next unique element */
3230 while (list
[src
] == list
[src
-1]) {
3235 /* dest always points to where the next unique element goes */
3236 list
[dest
] = list
[src
];
3241 * if the length difference is large enough, we want to allocate a
3242 * smaller buffer to save memory. if this fails due to out of memory,
3243 * we'll just stay with what we've got.
3245 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
3246 newlist
= pidlist_resize(list
, dest
);
3253 static int cmppid(const void *a
, const void *b
)
3255 return *(pid_t
*)a
- *(pid_t
*)b
;
3259 * find the appropriate pidlist for our purpose (given procs vs tasks)
3260 * returns with the lock on that pidlist already held, and takes care
3261 * of the use count, or returns NULL with no locks held if we're out of
3264 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3265 enum cgroup_filetype type
)
3267 struct cgroup_pidlist
*l
;
3268 /* don't need task_nsproxy() if we're looking at ourself */
3269 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
3272 * We can't drop the pidlist_mutex before taking the l->mutex in case
3273 * the last ref-holder is trying to remove l from the list at the same
3274 * time. Holding the pidlist_mutex precludes somebody taking whichever
3275 * list we find out from under us - compare release_pid_array().
3277 mutex_lock(&cgrp
->pidlist_mutex
);
3278 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3279 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3280 /* make sure l doesn't vanish out from under us */
3281 down_write(&l
->mutex
);
3282 mutex_unlock(&cgrp
->pidlist_mutex
);
3286 /* entry not found; create a new one */
3287 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3289 mutex_unlock(&cgrp
->pidlist_mutex
);
3292 init_rwsem(&l
->mutex
);
3293 down_write(&l
->mutex
);
3295 l
->key
.ns
= get_pid_ns(ns
);
3296 l
->use_count
= 0; /* don't increment here */
3299 list_add(&l
->links
, &cgrp
->pidlists
);
3300 mutex_unlock(&cgrp
->pidlist_mutex
);
3305 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3307 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3308 struct cgroup_pidlist
**lp
)
3312 int pid
, n
= 0; /* used for populating the array */
3313 struct cgroup_iter it
;
3314 struct task_struct
*tsk
;
3315 struct cgroup_pidlist
*l
;
3318 * If cgroup gets more users after we read count, we won't have
3319 * enough space - tough. This race is indistinguishable to the
3320 * caller from the case that the additional cgroup users didn't
3321 * show up until sometime later on.
3323 length
= cgroup_task_count(cgrp
);
3324 array
= pidlist_allocate(length
);
3327 /* now, populate the array */
3328 cgroup_iter_start(cgrp
, &it
);
3329 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3330 if (unlikely(n
== length
))
3332 /* get tgid or pid for procs or tasks file respectively */
3333 if (type
== CGROUP_FILE_PROCS
)
3334 pid
= task_tgid_vnr(tsk
);
3336 pid
= task_pid_vnr(tsk
);
3337 if (pid
> 0) /* make sure to only use valid results */
3340 cgroup_iter_end(cgrp
, &it
);
3342 /* now sort & (if procs) strip out duplicates */
3343 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3344 if (type
== CGROUP_FILE_PROCS
)
3345 length
= pidlist_uniq(&array
, length
);
3346 l
= cgroup_pidlist_find(cgrp
, type
);
3348 pidlist_free(array
);
3351 /* store array, freeing old if necessary - lock already held */
3352 pidlist_free(l
->list
);
3356 up_write(&l
->mutex
);
3362 * cgroupstats_build - build and fill cgroupstats
3363 * @stats: cgroupstats to fill information into
3364 * @dentry: A dentry entry belonging to the cgroup for which stats have
3367 * Build and fill cgroupstats so that taskstats can export it to user
3370 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3373 struct cgroup
*cgrp
;
3374 struct cgroup_iter it
;
3375 struct task_struct
*tsk
;
3378 * Validate dentry by checking the superblock operations,
3379 * and make sure it's a directory.
3381 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3382 !S_ISDIR(dentry
->d_inode
->i_mode
))
3386 cgrp
= dentry
->d_fsdata
;
3388 cgroup_iter_start(cgrp
, &it
);
3389 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3390 switch (tsk
->state
) {
3392 stats
->nr_running
++;
3394 case TASK_INTERRUPTIBLE
:
3395 stats
->nr_sleeping
++;
3397 case TASK_UNINTERRUPTIBLE
:
3398 stats
->nr_uninterruptible
++;
3401 stats
->nr_stopped
++;
3404 if (delayacct_is_task_waiting_on_io(tsk
))
3405 stats
->nr_io_wait
++;
3409 cgroup_iter_end(cgrp
, &it
);
3417 * seq_file methods for the tasks/procs files. The seq_file position is the
3418 * next pid to display; the seq_file iterator is a pointer to the pid
3419 * in the cgroup->l->list array.
3422 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3425 * Initially we receive a position value that corresponds to
3426 * one more than the last pid shown (or 0 on the first call or
3427 * after a seek to the start). Use a binary-search to find the
3428 * next pid to display, if any
3430 struct cgroup_pidlist
*l
= s
->private;
3431 int index
= 0, pid
= *pos
;
3434 down_read(&l
->mutex
);
3436 int end
= l
->length
;
3438 while (index
< end
) {
3439 int mid
= (index
+ end
) / 2;
3440 if (l
->list
[mid
] == pid
) {
3443 } else if (l
->list
[mid
] <= pid
)
3449 /* If we're off the end of the array, we're done */
3450 if (index
>= l
->length
)
3452 /* Update the abstract position to be the actual pid that we found */
3453 iter
= l
->list
+ index
;
3458 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3460 struct cgroup_pidlist
*l
= s
->private;
3464 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3466 struct cgroup_pidlist
*l
= s
->private;
3468 pid_t
*end
= l
->list
+ l
->length
;
3470 * Advance to the next pid in the array. If this goes off the
3482 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3484 return seq_printf(s
, "%d\n", *(int *)v
);
3488 * seq_operations functions for iterating on pidlists through seq_file -
3489 * independent of whether it's tasks or procs
3491 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3492 .start
= cgroup_pidlist_start
,
3493 .stop
= cgroup_pidlist_stop
,
3494 .next
= cgroup_pidlist_next
,
3495 .show
= cgroup_pidlist_show
,
3498 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3501 * the case where we're the last user of this particular pidlist will
3502 * have us remove it from the cgroup's list, which entails taking the
3503 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3504 * pidlist_mutex, we have to take pidlist_mutex first.
3506 mutex_lock(&l
->owner
->pidlist_mutex
);
3507 down_write(&l
->mutex
);
3508 BUG_ON(!l
->use_count
);
3509 if (!--l
->use_count
) {
3510 /* we're the last user if refcount is 0; remove and free */
3511 list_del(&l
->links
);
3512 mutex_unlock(&l
->owner
->pidlist_mutex
);
3513 pidlist_free(l
->list
);
3514 put_pid_ns(l
->key
.ns
);
3515 up_write(&l
->mutex
);
3519 mutex_unlock(&l
->owner
->pidlist_mutex
);
3520 up_write(&l
->mutex
);
3523 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3525 struct cgroup_pidlist
*l
;
3526 if (!(file
->f_mode
& FMODE_READ
))
3529 * the seq_file will only be initialized if the file was opened for
3530 * reading; hence we check if it's not null only in that case.
3532 l
= ((struct seq_file
*)file
->private_data
)->private;
3533 cgroup_release_pid_array(l
);
3534 return seq_release(inode
, file
);
3537 static const struct file_operations cgroup_pidlist_operations
= {
3539 .llseek
= seq_lseek
,
3540 .write
= cgroup_file_write
,
3541 .release
= cgroup_pidlist_release
,
3545 * The following functions handle opens on a file that displays a pidlist
3546 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3549 /* helper function for the two below it */
3550 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3552 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3553 struct cgroup_pidlist
*l
;
3556 /* Nothing to do for write-only files */
3557 if (!(file
->f_mode
& FMODE_READ
))
3560 /* have the array populated */
3561 retval
= pidlist_array_load(cgrp
, type
, &l
);
3564 /* configure file information */
3565 file
->f_op
= &cgroup_pidlist_operations
;
3567 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3569 cgroup_release_pid_array(l
);
3572 ((struct seq_file
*)file
->private_data
)->private = l
;
3575 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3577 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3579 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3581 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3584 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3587 return notify_on_release(cgrp
);
3590 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3594 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3596 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3598 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3603 * Unregister event and free resources.
3605 * Gets called from workqueue.
3607 static void cgroup_event_remove(struct work_struct
*work
)
3609 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3611 struct cgroup
*cgrp
= event
->cgrp
;
3613 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3615 eventfd_ctx_put(event
->eventfd
);
3621 * Gets called on POLLHUP on eventfd when user closes it.
3623 * Called with wqh->lock held and interrupts disabled.
3625 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3626 int sync
, void *key
)
3628 struct cgroup_event
*event
= container_of(wait
,
3629 struct cgroup_event
, wait
);
3630 struct cgroup
*cgrp
= event
->cgrp
;
3631 unsigned long flags
= (unsigned long)key
;
3633 if (flags
& POLLHUP
) {
3634 __remove_wait_queue(event
->wqh
, &event
->wait
);
3635 spin_lock(&cgrp
->event_list_lock
);
3636 list_del(&event
->list
);
3637 spin_unlock(&cgrp
->event_list_lock
);
3639 * We are in atomic context, but cgroup_event_remove() may
3640 * sleep, so we have to call it in workqueue.
3642 schedule_work(&event
->remove
);
3648 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3649 wait_queue_head_t
*wqh
, poll_table
*pt
)
3651 struct cgroup_event
*event
= container_of(pt
,
3652 struct cgroup_event
, pt
);
3655 add_wait_queue(wqh
, &event
->wait
);
3659 * Parse input and register new cgroup event handler.
3661 * Input must be in format '<event_fd> <control_fd> <args>'.
3662 * Interpretation of args is defined by control file implementation.
3664 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3667 struct cgroup_event
*event
= NULL
;
3668 unsigned int efd
, cfd
;
3669 struct file
*efile
= NULL
;
3670 struct file
*cfile
= NULL
;
3674 efd
= simple_strtoul(buffer
, &endp
, 10);
3679 cfd
= simple_strtoul(buffer
, &endp
, 10);
3680 if ((*endp
!= ' ') && (*endp
!= '\0'))
3684 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3688 INIT_LIST_HEAD(&event
->list
);
3689 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3690 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3691 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3693 efile
= eventfd_fget(efd
);
3694 if (IS_ERR(efile
)) {
3695 ret
= PTR_ERR(efile
);
3699 event
->eventfd
= eventfd_ctx_fileget(efile
);
3700 if (IS_ERR(event
->eventfd
)) {
3701 ret
= PTR_ERR(event
->eventfd
);
3711 /* the process need read permission on control file */
3712 /* AV: shouldn't we check that it's been opened for read instead? */
3713 ret
= inode_permission(cfile
->f_path
.dentry
->d_inode
, MAY_READ
);
3717 event
->cft
= __file_cft(cfile
);
3718 if (IS_ERR(event
->cft
)) {
3719 ret
= PTR_ERR(event
->cft
);
3723 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3728 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3729 event
->eventfd
, buffer
);
3733 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3734 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3740 * Events should be removed after rmdir of cgroup directory, but before
3741 * destroying subsystem state objects. Let's take reference to cgroup
3742 * directory dentry to do that.
3746 spin_lock(&cgrp
->event_list_lock
);
3747 list_add(&event
->list
, &cgrp
->event_list
);
3748 spin_unlock(&cgrp
->event_list_lock
);
3759 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3760 eventfd_ctx_put(event
->eventfd
);
3762 if (!IS_ERR_OR_NULL(efile
))
3770 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3773 return clone_children(cgrp
);
3776 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3781 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3783 clear_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3788 * for the common functions, 'private' gives the type of file
3790 /* for hysterical raisins, we can't put this on the older files */
3791 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3792 static struct cftype files
[] = {
3795 .open
= cgroup_tasks_open
,
3796 .write_u64
= cgroup_tasks_write
,
3797 .release
= cgroup_pidlist_release
,
3798 .mode
= S_IRUGO
| S_IWUSR
,
3801 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3802 .open
= cgroup_procs_open
,
3803 .write_u64
= cgroup_procs_write
,
3804 .release
= cgroup_pidlist_release
,
3805 .mode
= S_IRUGO
| S_IWUSR
,
3808 .name
= "notify_on_release",
3809 .read_u64
= cgroup_read_notify_on_release
,
3810 .write_u64
= cgroup_write_notify_on_release
,
3813 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3814 .write_string
= cgroup_write_event_control
,
3818 .name
= "cgroup.clone_children",
3819 .read_u64
= cgroup_clone_children_read
,
3820 .write_u64
= cgroup_clone_children_write
,
3823 .name
= "release_agent",
3824 .flags
= CFTYPE_ONLY_ON_ROOT
,
3825 .read_seq_string
= cgroup_release_agent_show
,
3826 .write_string
= cgroup_release_agent_write
,
3827 .max_write_len
= PATH_MAX
,
3832 static int cgroup_populate_dir(struct cgroup
*cgrp
)
3835 struct cgroup_subsys
*ss
;
3837 err
= cgroup_addrm_files(cgrp
, NULL
, files
, true);
3841 /* process cftsets of each subsystem */
3842 for_each_subsys(cgrp
->root
, ss
) {
3843 struct cftype_set
*set
;
3845 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
3848 list_for_each_entry(set
, &ss
->cftsets
, node
)
3849 cgroup_addrm_files(cgrp
, ss
, set
->cfts
, true);
3852 /* This cgroup is ready now */
3853 for_each_subsys(cgrp
->root
, ss
) {
3854 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3856 * Update id->css pointer and make this css visible from
3857 * CSS ID functions. This pointer will be dereferened
3858 * from RCU-read-side without locks.
3861 rcu_assign_pointer(css
->id
->css
, css
);
3867 static void css_dput_fn(struct work_struct
*work
)
3869 struct cgroup_subsys_state
*css
=
3870 container_of(work
, struct cgroup_subsys_state
, dput_work
);
3872 dput(css
->cgroup
->dentry
);
3875 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
3876 struct cgroup_subsys
*ss
,
3877 struct cgroup
*cgrp
)
3880 atomic_set(&css
->refcnt
, 1);
3883 if (cgrp
== dummytop
)
3884 set_bit(CSS_ROOT
, &css
->flags
);
3885 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
3886 cgrp
->subsys
[ss
->subsys_id
] = css
;
3889 * If !clear_css_refs, css holds an extra ref to @cgrp->dentry
3890 * which is put on the last css_put(). dput() requires process
3891 * context, which css_put() may be called without. @css->dput_work
3892 * will be used to invoke dput() asynchronously from css_put().
3894 INIT_WORK(&css
->dput_work
, css_dput_fn
);
3895 if (ss
->__DEPRECATED_clear_css_refs
)
3896 set_bit(CSS_CLEAR_CSS_REFS
, &css
->flags
);
3899 static void cgroup_lock_hierarchy(struct cgroupfs_root
*root
)
3901 /* We need to take each hierarchy_mutex in a consistent order */
3905 * No worry about a race with rebind_subsystems that might mess up the
3906 * locking order, since both parties are under cgroup_mutex.
3908 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3909 struct cgroup_subsys
*ss
= subsys
[i
];
3912 if (ss
->root
== root
)
3913 mutex_lock(&ss
->hierarchy_mutex
);
3917 static void cgroup_unlock_hierarchy(struct cgroupfs_root
*root
)
3921 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3922 struct cgroup_subsys
*ss
= subsys
[i
];
3925 if (ss
->root
== root
)
3926 mutex_unlock(&ss
->hierarchy_mutex
);
3931 * cgroup_create - create a cgroup
3932 * @parent: cgroup that will be parent of the new cgroup
3933 * @dentry: dentry of the new cgroup
3934 * @mode: mode to set on new inode
3936 * Must be called with the mutex on the parent inode held
3938 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
3941 struct cgroup
*cgrp
;
3942 struct cgroupfs_root
*root
= parent
->root
;
3944 struct cgroup_subsys
*ss
;
3945 struct super_block
*sb
= root
->sb
;
3947 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3951 /* Grab a reference on the superblock so the hierarchy doesn't
3952 * get deleted on unmount if there are child cgroups. This
3953 * can be done outside cgroup_mutex, since the sb can't
3954 * disappear while someone has an open control file on the
3956 atomic_inc(&sb
->s_active
);
3958 mutex_lock(&cgroup_mutex
);
3960 init_cgroup_housekeeping(cgrp
);
3962 cgrp
->parent
= parent
;
3963 cgrp
->root
= parent
->root
;
3964 cgrp
->top_cgroup
= parent
->top_cgroup
;
3966 if (notify_on_release(parent
))
3967 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3969 if (clone_children(parent
))
3970 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3972 for_each_subsys(root
, ss
) {
3973 struct cgroup_subsys_state
*css
= ss
->create(cgrp
);
3979 init_cgroup_css(css
, ss
, cgrp
);
3981 err
= alloc_css_id(ss
, parent
, cgrp
);
3985 /* At error, ->destroy() callback has to free assigned ID. */
3986 if (clone_children(parent
) && ss
->post_clone
)
3987 ss
->post_clone(cgrp
);
3990 cgroup_lock_hierarchy(root
);
3991 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
3992 cgroup_unlock_hierarchy(root
);
3993 root
->number_of_cgroups
++;
3995 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
3999 /* If !clear_css_refs, each css holds a ref to the cgroup's dentry */
4000 for_each_subsys(root
, ss
)
4001 if (!ss
->__DEPRECATED_clear_css_refs
)
4004 /* The cgroup directory was pre-locked for us */
4005 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
4007 list_add_tail(&cgrp
->allcg_node
, &root
->allcg_list
);
4009 err
= cgroup_populate_dir(cgrp
);
4010 /* If err < 0, we have a half-filled directory - oh well ;) */
4012 mutex_unlock(&cgroup_mutex
);
4013 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
4019 cgroup_lock_hierarchy(root
);
4020 list_del(&cgrp
->sibling
);
4021 cgroup_unlock_hierarchy(root
);
4022 root
->number_of_cgroups
--;
4026 for_each_subsys(root
, ss
) {
4027 if (cgrp
->subsys
[ss
->subsys_id
])
4031 mutex_unlock(&cgroup_mutex
);
4033 /* Release the reference count that we took on the superblock */
4034 deactivate_super(sb
);
4040 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4042 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
4044 /* the vfs holds inode->i_mutex already */
4045 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
4049 * Check the reference count on each subsystem. Since we already
4050 * established that there are no tasks in the cgroup, if the css refcount
4051 * is also 1, then there should be no outstanding references, so the
4052 * subsystem is safe to destroy. We scan across all subsystems rather than
4053 * using the per-hierarchy linked list of mounted subsystems since we can
4054 * be called via check_for_release() with no synchronization other than
4055 * RCU, and the subsystem linked list isn't RCU-safe.
4057 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
4062 * We won't need to lock the subsys array, because the subsystems
4063 * we're concerned about aren't going anywhere since our cgroup root
4064 * has a reference on them.
4066 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4067 struct cgroup_subsys
*ss
= subsys
[i
];
4068 struct cgroup_subsys_state
*css
;
4070 /* Skip subsystems not present or not in this hierarchy */
4071 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
4074 css
= cgrp
->subsys
[ss
->subsys_id
];
4076 * When called from check_for_release() it's possible
4077 * that by this point the cgroup has been removed
4078 * and the css deleted. But a false-positive doesn't
4079 * matter, since it can only happen if the cgroup
4080 * has been deleted and hence no longer needs the
4081 * release agent to be called anyway.
4083 if (css
&& css_refcnt(css
) > 1)
4090 * Atomically mark all (or else none) of the cgroup's CSS objects as
4091 * CSS_REMOVED. Return true on success, or false if the cgroup has
4092 * busy subsystems. Call with cgroup_mutex held
4094 * Depending on whether a subsys has __DEPRECATED_clear_css_refs set or
4095 * not, cgroup removal behaves differently.
4097 * If clear is set, css refcnt for the subsystem should be zero before
4098 * cgroup removal can be committed. This is implemented by
4099 * CGRP_WAIT_ON_RMDIR and retry logic around ->pre_destroy(), which may be
4100 * called multiple times until all css refcnts reach zero and is allowed to
4101 * veto removal on any invocation. This behavior is deprecated and will be
4102 * removed as soon as the existing user (memcg) is updated.
4104 * If clear is not set, each css holds an extra reference to the cgroup's
4105 * dentry and cgroup removal proceeds regardless of css refs.
4106 * ->pre_destroy() will be called at least once and is not allowed to fail.
4107 * On the last put of each css, whenever that may be, the extra dentry ref
4108 * is put so that dentry destruction happens only after all css's are
4111 static int cgroup_clear_css_refs(struct cgroup
*cgrp
)
4113 struct cgroup_subsys
*ss
;
4114 unsigned long flags
;
4115 bool failed
= false;
4117 local_irq_save(flags
);
4120 * Block new css_tryget() by deactivating refcnt. If all refcnts
4121 * for subsystems w/ clear_css_refs set were 1 at the moment of
4122 * deactivation, we succeeded.
4124 for_each_subsys(cgrp
->root
, ss
) {
4125 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4127 WARN_ON(atomic_read(&css
->refcnt
) < 0);
4128 atomic_add(CSS_DEACT_BIAS
, &css
->refcnt
);
4130 if (ss
->__DEPRECATED_clear_css_refs
)
4131 failed
|= css_refcnt(css
) != 1;
4135 * If succeeded, set REMOVED and put all the base refs; otherwise,
4136 * restore refcnts to positive values. Either way, all in-progress
4137 * css_tryget() will be released.
4139 for_each_subsys(cgrp
->root
, ss
) {
4140 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
4143 set_bit(CSS_REMOVED
, &css
->flags
);
4146 atomic_sub(CSS_DEACT_BIAS
, &css
->refcnt
);
4150 local_irq_restore(flags
);
4154 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
4156 struct cgroup
*cgrp
= dentry
->d_fsdata
;
4158 struct cgroup
*parent
;
4160 struct cgroup_event
*event
, *tmp
;
4163 /* the vfs holds both inode->i_mutex already */
4165 mutex_lock(&cgroup_mutex
);
4166 if (atomic_read(&cgrp
->count
) != 0) {
4167 mutex_unlock(&cgroup_mutex
);
4170 if (!list_empty(&cgrp
->children
)) {
4171 mutex_unlock(&cgroup_mutex
);
4174 mutex_unlock(&cgroup_mutex
);
4177 * In general, subsystem has no css->refcnt after pre_destroy(). But
4178 * in racy cases, subsystem may have to get css->refcnt after
4179 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
4180 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
4181 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
4182 * and subsystem's reference count handling. Please see css_get/put
4183 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
4185 set_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4188 * Call pre_destroy handlers of subsys. Notify subsystems
4189 * that rmdir() request comes.
4191 ret
= cgroup_call_pre_destroy(cgrp
);
4193 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4197 mutex_lock(&cgroup_mutex
);
4198 parent
= cgrp
->parent
;
4199 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
)) {
4200 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4201 mutex_unlock(&cgroup_mutex
);
4204 prepare_to_wait(&cgroup_rmdir_waitq
, &wait
, TASK_INTERRUPTIBLE
);
4205 if (!cgroup_clear_css_refs(cgrp
)) {
4206 mutex_unlock(&cgroup_mutex
);
4208 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4209 * prepare_to_wait(), we need to check this flag.
4211 if (test_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
))
4213 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4214 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4215 if (signal_pending(current
))
4219 /* NO css_tryget() can success after here. */
4220 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4221 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4223 raw_spin_lock(&release_list_lock
);
4224 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
4225 if (!list_empty(&cgrp
->release_list
))
4226 list_del_init(&cgrp
->release_list
);
4227 raw_spin_unlock(&release_list_lock
);
4229 cgroup_lock_hierarchy(cgrp
->root
);
4230 /* delete this cgroup from parent->children */
4231 list_del_init(&cgrp
->sibling
);
4232 cgroup_unlock_hierarchy(cgrp
->root
);
4234 list_del_init(&cgrp
->allcg_node
);
4236 d
= dget(cgrp
->dentry
);
4238 cgroup_d_remove_dir(d
);
4241 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4242 check_for_release(parent
);
4245 * Unregister events and notify userspace.
4246 * Notify userspace about cgroup removing only after rmdir of cgroup
4247 * directory to avoid race between userspace and kernelspace
4249 spin_lock(&cgrp
->event_list_lock
);
4250 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4251 list_del(&event
->list
);
4252 remove_wait_queue(event
->wqh
, &event
->wait
);
4253 eventfd_signal(event
->eventfd
, 1);
4254 schedule_work(&event
->remove
);
4256 spin_unlock(&cgrp
->event_list_lock
);
4258 mutex_unlock(&cgroup_mutex
);
4262 static void __init_or_module
cgroup_init_cftsets(struct cgroup_subsys
*ss
)
4264 INIT_LIST_HEAD(&ss
->cftsets
);
4267 * base_cftset is embedded in subsys itself, no need to worry about
4270 if (ss
->base_cftypes
) {
4271 ss
->base_cftset
.cfts
= ss
->base_cftypes
;
4272 list_add_tail(&ss
->base_cftset
.node
, &ss
->cftsets
);
4276 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4278 struct cgroup_subsys_state
*css
;
4280 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4282 /* init base cftset */
4283 cgroup_init_cftsets(ss
);
4285 /* Create the top cgroup state for this subsystem */
4286 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4287 ss
->root
= &rootnode
;
4288 css
= ss
->create(dummytop
);
4289 /* We don't handle early failures gracefully */
4290 BUG_ON(IS_ERR(css
));
4291 init_cgroup_css(css
, ss
, dummytop
);
4293 /* Update the init_css_set to contain a subsys
4294 * pointer to this state - since the subsystem is
4295 * newly registered, all tasks and hence the
4296 * init_css_set is in the subsystem's top cgroup. */
4297 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
4299 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4301 /* At system boot, before all subsystems have been
4302 * registered, no tasks have been forked, so we don't
4303 * need to invoke fork callbacks here. */
4304 BUG_ON(!list_empty(&init_task
.tasks
));
4306 mutex_init(&ss
->hierarchy_mutex
);
4307 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4310 /* this function shouldn't be used with modular subsystems, since they
4311 * need to register a subsys_id, among other things */
4316 * cgroup_load_subsys: load and register a modular subsystem at runtime
4317 * @ss: the subsystem to load
4319 * This function should be called in a modular subsystem's initcall. If the
4320 * subsystem is built as a module, it will be assigned a new subsys_id and set
4321 * up for use. If the subsystem is built-in anyway, work is delegated to the
4322 * simpler cgroup_init_subsys.
4324 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4327 struct cgroup_subsys_state
*css
;
4329 /* check name and function validity */
4330 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4331 ss
->create
== NULL
|| ss
->destroy
== NULL
)
4335 * we don't support callbacks in modular subsystems. this check is
4336 * before the ss->module check for consistency; a subsystem that could
4337 * be a module should still have no callbacks even if the user isn't
4338 * compiling it as one.
4340 if (ss
->fork
|| ss
->exit
)
4344 * an optionally modular subsystem is built-in: we want to do nothing,
4345 * since cgroup_init_subsys will have already taken care of it.
4347 if (ss
->module
== NULL
) {
4348 /* a few sanity checks */
4349 BUG_ON(ss
->subsys_id
>= CGROUP_BUILTIN_SUBSYS_COUNT
);
4350 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4354 /* init base cftset */
4355 cgroup_init_cftsets(ss
);
4358 * need to register a subsys id before anything else - for example,
4359 * init_cgroup_css needs it.
4361 mutex_lock(&cgroup_mutex
);
4362 /* find the first empty slot in the array */
4363 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4364 if (subsys
[i
] == NULL
)
4367 if (i
== CGROUP_SUBSYS_COUNT
) {
4368 /* maximum number of subsystems already registered! */
4369 mutex_unlock(&cgroup_mutex
);
4372 /* assign ourselves the subsys_id */
4377 * no ss->create seems to need anything important in the ss struct, so
4378 * this can happen first (i.e. before the rootnode attachment).
4380 css
= ss
->create(dummytop
);
4382 /* failure case - need to deassign the subsys[] slot. */
4384 mutex_unlock(&cgroup_mutex
);
4385 return PTR_ERR(css
);
4388 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4389 ss
->root
= &rootnode
;
4391 /* our new subsystem will be attached to the dummy hierarchy. */
4392 init_cgroup_css(css
, ss
, dummytop
);
4393 /* init_idr must be after init_cgroup_css because it sets css->id. */
4395 int ret
= cgroup_init_idr(ss
, css
);
4397 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4398 ss
->destroy(dummytop
);
4400 mutex_unlock(&cgroup_mutex
);
4406 * Now we need to entangle the css into the existing css_sets. unlike
4407 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4408 * will need a new pointer to it; done by iterating the css_set_table.
4409 * furthermore, modifying the existing css_sets will corrupt the hash
4410 * table state, so each changed css_set will need its hash recomputed.
4411 * this is all done under the css_set_lock.
4413 write_lock(&css_set_lock
);
4414 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
4416 struct hlist_node
*node
, *tmp
;
4417 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
4419 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
4420 /* skip entries that we already rehashed */
4421 if (cg
->subsys
[ss
->subsys_id
])
4423 /* remove existing entry */
4424 hlist_del(&cg
->hlist
);
4426 cg
->subsys
[ss
->subsys_id
] = css
;
4427 /* recompute hash and restore entry */
4428 new_bucket
= css_set_hash(cg
->subsys
);
4429 hlist_add_head(&cg
->hlist
, new_bucket
);
4432 write_unlock(&css_set_lock
);
4434 mutex_init(&ss
->hierarchy_mutex
);
4435 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4439 mutex_unlock(&cgroup_mutex
);
4442 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4445 * cgroup_unload_subsys: unload a modular subsystem
4446 * @ss: the subsystem to unload
4448 * This function should be called in a modular subsystem's exitcall. When this
4449 * function is invoked, the refcount on the subsystem's module will be 0, so
4450 * the subsystem will not be attached to any hierarchy.
4452 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4454 struct cg_cgroup_link
*link
;
4455 struct hlist_head
*hhead
;
4457 BUG_ON(ss
->module
== NULL
);
4460 * we shouldn't be called if the subsystem is in use, and the use of
4461 * try_module_get in parse_cgroupfs_options should ensure that it
4462 * doesn't start being used while we're killing it off.
4464 BUG_ON(ss
->root
!= &rootnode
);
4466 mutex_lock(&cgroup_mutex
);
4467 /* deassign the subsys_id */
4468 BUG_ON(ss
->subsys_id
< CGROUP_BUILTIN_SUBSYS_COUNT
);
4469 subsys
[ss
->subsys_id
] = NULL
;
4471 /* remove subsystem from rootnode's list of subsystems */
4472 list_del_init(&ss
->sibling
);
4475 * disentangle the css from all css_sets attached to the dummytop. as
4476 * in loading, we need to pay our respects to the hashtable gods.
4478 write_lock(&css_set_lock
);
4479 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4480 struct css_set
*cg
= link
->cg
;
4482 hlist_del(&cg
->hlist
);
4483 BUG_ON(!cg
->subsys
[ss
->subsys_id
]);
4484 cg
->subsys
[ss
->subsys_id
] = NULL
;
4485 hhead
= css_set_hash(cg
->subsys
);
4486 hlist_add_head(&cg
->hlist
, hhead
);
4488 write_unlock(&css_set_lock
);
4491 * remove subsystem's css from the dummytop and free it - need to free
4492 * before marking as null because ss->destroy needs the cgrp->subsys
4493 * pointer to find their state. note that this also takes care of
4494 * freeing the css_id.
4496 ss
->destroy(dummytop
);
4497 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4499 mutex_unlock(&cgroup_mutex
);
4501 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4504 * cgroup_init_early - cgroup initialization at system boot
4506 * Initialize cgroups at system boot, and initialize any
4507 * subsystems that request early init.
4509 int __init
cgroup_init_early(void)
4512 atomic_set(&init_css_set
.refcount
, 1);
4513 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4514 INIT_LIST_HEAD(&init_css_set
.tasks
);
4515 INIT_HLIST_NODE(&init_css_set
.hlist
);
4517 init_cgroup_root(&rootnode
);
4519 init_task
.cgroups
= &init_css_set
;
4521 init_css_set_link
.cg
= &init_css_set
;
4522 init_css_set_link
.cgrp
= dummytop
;
4523 list_add(&init_css_set_link
.cgrp_link_list
,
4524 &rootnode
.top_cgroup
.css_sets
);
4525 list_add(&init_css_set_link
.cg_link_list
,
4526 &init_css_set
.cg_links
);
4528 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
4529 INIT_HLIST_HEAD(&css_set_table
[i
]);
4531 /* at bootup time, we don't worry about modular subsystems */
4532 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4533 struct cgroup_subsys
*ss
= subsys
[i
];
4536 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4537 BUG_ON(!ss
->create
);
4538 BUG_ON(!ss
->destroy
);
4539 if (ss
->subsys_id
!= i
) {
4540 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4541 ss
->name
, ss
->subsys_id
);
4546 cgroup_init_subsys(ss
);
4552 * cgroup_init - cgroup initialization
4554 * Register cgroup filesystem and /proc file, and initialize
4555 * any subsystems that didn't request early init.
4557 int __init
cgroup_init(void)
4561 struct hlist_head
*hhead
;
4563 err
= bdi_init(&cgroup_backing_dev_info
);
4567 /* at bootup time, we don't worry about modular subsystems */
4568 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4569 struct cgroup_subsys
*ss
= subsys
[i
];
4570 if (!ss
->early_init
)
4571 cgroup_init_subsys(ss
);
4573 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4576 /* Add init_css_set to the hash table */
4577 hhead
= css_set_hash(init_css_set
.subsys
);
4578 hlist_add_head(&init_css_set
.hlist
, hhead
);
4579 BUG_ON(!init_root_id(&rootnode
));
4581 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4587 err
= register_filesystem(&cgroup_fs_type
);
4589 kobject_put(cgroup_kobj
);
4593 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4597 bdi_destroy(&cgroup_backing_dev_info
);
4603 * proc_cgroup_show()
4604 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4605 * - Used for /proc/<pid>/cgroup.
4606 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4607 * doesn't really matter if tsk->cgroup changes after we read it,
4608 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4609 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4610 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4611 * cgroup to top_cgroup.
4614 /* TODO: Use a proper seq_file iterator */
4615 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4618 struct task_struct
*tsk
;
4621 struct cgroupfs_root
*root
;
4624 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4630 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4636 mutex_lock(&cgroup_mutex
);
4638 for_each_active_root(root
) {
4639 struct cgroup_subsys
*ss
;
4640 struct cgroup
*cgrp
;
4643 seq_printf(m
, "%d:", root
->hierarchy_id
);
4644 for_each_subsys(root
, ss
)
4645 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4646 if (strlen(root
->name
))
4647 seq_printf(m
, "%sname=%s", count
? "," : "",
4650 cgrp
= task_cgroup_from_root(tsk
, root
);
4651 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4659 mutex_unlock(&cgroup_mutex
);
4660 put_task_struct(tsk
);
4667 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4669 struct pid
*pid
= PROC_I(inode
)->pid
;
4670 return single_open(file
, proc_cgroup_show
, pid
);
4673 const struct file_operations proc_cgroup_operations
= {
4674 .open
= cgroup_open
,
4676 .llseek
= seq_lseek
,
4677 .release
= single_release
,
4680 /* Display information about each subsystem and each hierarchy */
4681 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4685 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4687 * ideally we don't want subsystems moving around while we do this.
4688 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4689 * subsys/hierarchy state.
4691 mutex_lock(&cgroup_mutex
);
4692 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4693 struct cgroup_subsys
*ss
= subsys
[i
];
4696 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4697 ss
->name
, ss
->root
->hierarchy_id
,
4698 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4700 mutex_unlock(&cgroup_mutex
);
4704 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4706 return single_open(file
, proc_cgroupstats_show
, NULL
);
4709 static const struct file_operations proc_cgroupstats_operations
= {
4710 .open
= cgroupstats_open
,
4712 .llseek
= seq_lseek
,
4713 .release
= single_release
,
4717 * cgroup_fork - attach newly forked task to its parents cgroup.
4718 * @child: pointer to task_struct of forking parent process.
4720 * Description: A task inherits its parent's cgroup at fork().
4722 * A pointer to the shared css_set was automatically copied in
4723 * fork.c by dup_task_struct(). However, we ignore that copy, since
4724 * it was not made under the protection of RCU, cgroup_mutex or
4725 * threadgroup_change_begin(), so it might no longer be a valid
4726 * cgroup pointer. cgroup_attach_task() might have already changed
4727 * current->cgroups, allowing the previously referenced cgroup
4728 * group to be removed and freed.
4730 * Outside the pointer validity we also need to process the css_set
4731 * inheritance between threadgoup_change_begin() and
4732 * threadgoup_change_end(), this way there is no leak in any process
4733 * wide migration performed by cgroup_attach_proc() that could otherwise
4734 * miss a thread because it is too early or too late in the fork stage.
4736 * At the point that cgroup_fork() is called, 'current' is the parent
4737 * task, and the passed argument 'child' points to the child task.
4739 void cgroup_fork(struct task_struct
*child
)
4742 * We don't need to task_lock() current because current->cgroups
4743 * can't be changed concurrently here. The parent obviously hasn't
4744 * exited and called cgroup_exit(), and we are synchronized against
4745 * cgroup migration through threadgroup_change_begin().
4747 child
->cgroups
= current
->cgroups
;
4748 get_css_set(child
->cgroups
);
4749 INIT_LIST_HEAD(&child
->cg_list
);
4753 * cgroup_fork_callbacks - run fork callbacks
4754 * @child: the new task
4756 * Called on a new task very soon before adding it to the
4757 * tasklist. No need to take any locks since no-one can
4758 * be operating on this task.
4760 void cgroup_fork_callbacks(struct task_struct
*child
)
4762 if (need_forkexit_callback
) {
4765 * forkexit callbacks are only supported for builtin
4766 * subsystems, and the builtin section of the subsys array is
4767 * immutable, so we don't need to lock the subsys array here.
4769 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4770 struct cgroup_subsys
*ss
= subsys
[i
];
4778 * cgroup_post_fork - called on a new task after adding it to the task list
4779 * @child: the task in question
4781 * Adds the task to the list running through its css_set if necessary.
4782 * Has to be after the task is visible on the task list in case we race
4783 * with the first call to cgroup_iter_start() - to guarantee that the
4784 * new task ends up on its list.
4786 void cgroup_post_fork(struct task_struct
*child
)
4789 * use_task_css_set_links is set to 1 before we walk the tasklist
4790 * under the tasklist_lock and we read it here after we added the child
4791 * to the tasklist under the tasklist_lock as well. If the child wasn't
4792 * yet in the tasklist when we walked through it from
4793 * cgroup_enable_task_cg_lists(), then use_task_css_set_links value
4794 * should be visible now due to the paired locking and barriers implied
4795 * by LOCK/UNLOCK: it is written before the tasklist_lock unlock
4796 * in cgroup_enable_task_cg_lists() and read here after the tasklist_lock
4799 if (use_task_css_set_links
) {
4800 write_lock(&css_set_lock
);
4801 if (list_empty(&child
->cg_list
)) {
4803 * It's safe to use child->cgroups without task_lock()
4804 * here because we are protected through
4805 * threadgroup_change_begin() against concurrent
4806 * css_set change in cgroup_task_migrate(). Also
4807 * the task can't exit at that point until
4808 * wake_up_new_task() is called, so we are protected
4809 * against cgroup_exit() setting child->cgroup to
4812 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4814 write_unlock(&css_set_lock
);
4818 * cgroup_exit - detach cgroup from exiting task
4819 * @tsk: pointer to task_struct of exiting process
4820 * @run_callback: run exit callbacks?
4822 * Description: Detach cgroup from @tsk and release it.
4824 * Note that cgroups marked notify_on_release force every task in
4825 * them to take the global cgroup_mutex mutex when exiting.
4826 * This could impact scaling on very large systems. Be reluctant to
4827 * use notify_on_release cgroups where very high task exit scaling
4828 * is required on large systems.
4830 * the_top_cgroup_hack:
4832 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4834 * We call cgroup_exit() while the task is still competent to
4835 * handle notify_on_release(), then leave the task attached to the
4836 * root cgroup in each hierarchy for the remainder of its exit.
4838 * To do this properly, we would increment the reference count on
4839 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4840 * code we would add a second cgroup function call, to drop that
4841 * reference. This would just create an unnecessary hot spot on
4842 * the top_cgroup reference count, to no avail.
4844 * Normally, holding a reference to a cgroup without bumping its
4845 * count is unsafe. The cgroup could go away, or someone could
4846 * attach us to a different cgroup, decrementing the count on
4847 * the first cgroup that we never incremented. But in this case,
4848 * top_cgroup isn't going away, and either task has PF_EXITING set,
4849 * which wards off any cgroup_attach_task() attempts, or task is a failed
4850 * fork, never visible to cgroup_attach_task.
4852 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4858 * Unlink from the css_set task list if necessary.
4859 * Optimistically check cg_list before taking
4862 if (!list_empty(&tsk
->cg_list
)) {
4863 write_lock(&css_set_lock
);
4864 if (!list_empty(&tsk
->cg_list
))
4865 list_del_init(&tsk
->cg_list
);
4866 write_unlock(&css_set_lock
);
4869 /* Reassign the task to the init_css_set. */
4872 tsk
->cgroups
= &init_css_set
;
4874 if (run_callbacks
&& need_forkexit_callback
) {
4876 * modular subsystems can't use callbacks, so no need to lock
4879 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4880 struct cgroup_subsys
*ss
= subsys
[i
];
4882 struct cgroup
*old_cgrp
=
4883 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4884 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4885 ss
->exit(cgrp
, old_cgrp
, tsk
);
4892 put_css_set_taskexit(cg
);
4896 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4897 * @cgrp: the cgroup in question
4898 * @task: the task in question
4900 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4903 * If we are sending in dummytop, then presumably we are creating
4904 * the top cgroup in the subsystem.
4906 * Called only by the ns (nsproxy) cgroup.
4908 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
4911 struct cgroup
*target
;
4913 if (cgrp
== dummytop
)
4916 target
= task_cgroup_from_root(task
, cgrp
->root
);
4917 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
4918 cgrp
= cgrp
->parent
;
4919 ret
= (cgrp
== target
);
4923 static void check_for_release(struct cgroup
*cgrp
)
4925 /* All of these checks rely on RCU to keep the cgroup
4926 * structure alive */
4927 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
4928 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
4929 /* Control Group is currently removeable. If it's not
4930 * already queued for a userspace notification, queue
4932 int need_schedule_work
= 0;
4933 raw_spin_lock(&release_list_lock
);
4934 if (!cgroup_is_removed(cgrp
) &&
4935 list_empty(&cgrp
->release_list
)) {
4936 list_add(&cgrp
->release_list
, &release_list
);
4937 need_schedule_work
= 1;
4939 raw_spin_unlock(&release_list_lock
);
4940 if (need_schedule_work
)
4941 schedule_work(&release_agent_work
);
4945 /* Caller must verify that the css is not for root cgroup */
4946 bool __css_tryget(struct cgroup_subsys_state
*css
)
4949 int v
= css_refcnt(css
);
4951 if (atomic_cmpxchg(&css
->refcnt
, v
, v
+ 1) == v
)
4954 } while (!test_bit(CSS_REMOVED
, &css
->flags
));
4958 EXPORT_SYMBOL_GPL(__css_tryget
);
4960 /* Caller must verify that the css is not for root cgroup */
4961 void __css_put(struct cgroup_subsys_state
*css
)
4963 struct cgroup
*cgrp
= css
->cgroup
;
4966 atomic_dec(&css
->refcnt
);
4967 switch (css_refcnt(css
)) {
4969 if (notify_on_release(cgrp
)) {
4970 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4971 check_for_release(cgrp
);
4973 cgroup_wakeup_rmdir_waiter(cgrp
);
4976 if (!test_bit(CSS_CLEAR_CSS_REFS
, &css
->flags
))
4977 schedule_work(&css
->dput_work
);
4982 EXPORT_SYMBOL_GPL(__css_put
);
4985 * Notify userspace when a cgroup is released, by running the
4986 * configured release agent with the name of the cgroup (path
4987 * relative to the root of cgroup file system) as the argument.
4989 * Most likely, this user command will try to rmdir this cgroup.
4991 * This races with the possibility that some other task will be
4992 * attached to this cgroup before it is removed, or that some other
4993 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4994 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4995 * unused, and this cgroup will be reprieved from its death sentence,
4996 * to continue to serve a useful existence. Next time it's released,
4997 * we will get notified again, if it still has 'notify_on_release' set.
4999 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
5000 * means only wait until the task is successfully execve()'d. The
5001 * separate release agent task is forked by call_usermodehelper(),
5002 * then control in this thread returns here, without waiting for the
5003 * release agent task. We don't bother to wait because the caller of
5004 * this routine has no use for the exit status of the release agent
5005 * task, so no sense holding our caller up for that.
5007 static void cgroup_release_agent(struct work_struct
*work
)
5009 BUG_ON(work
!= &release_agent_work
);
5010 mutex_lock(&cgroup_mutex
);
5011 raw_spin_lock(&release_list_lock
);
5012 while (!list_empty(&release_list
)) {
5013 char *argv
[3], *envp
[3];
5015 char *pathbuf
= NULL
, *agentbuf
= NULL
;
5016 struct cgroup
*cgrp
= list_entry(release_list
.next
,
5019 list_del_init(&cgrp
->release_list
);
5020 raw_spin_unlock(&release_list_lock
);
5021 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
5024 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
5026 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
5031 argv
[i
++] = agentbuf
;
5032 argv
[i
++] = pathbuf
;
5036 /* minimal command environment */
5037 envp
[i
++] = "HOME=/";
5038 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
5041 /* Drop the lock while we invoke the usermode helper,
5042 * since the exec could involve hitting disk and hence
5043 * be a slow process */
5044 mutex_unlock(&cgroup_mutex
);
5045 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
5046 mutex_lock(&cgroup_mutex
);
5050 raw_spin_lock(&release_list_lock
);
5052 raw_spin_unlock(&release_list_lock
);
5053 mutex_unlock(&cgroup_mutex
);
5056 static int __init
cgroup_disable(char *str
)
5061 while ((token
= strsep(&str
, ",")) != NULL
) {
5065 * cgroup_disable, being at boot time, can't know about module
5066 * subsystems, so we don't worry about them.
5068 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
5069 struct cgroup_subsys
*ss
= subsys
[i
];
5071 if (!strcmp(token
, ss
->name
)) {
5073 printk(KERN_INFO
"Disabling %s control group"
5074 " subsystem\n", ss
->name
);
5081 __setup("cgroup_disable=", cgroup_disable
);
5084 * Functons for CSS ID.
5088 *To get ID other than 0, this should be called when !cgroup_is_removed().
5090 unsigned short css_id(struct cgroup_subsys_state
*css
)
5092 struct css_id
*cssid
;
5095 * This css_id() can return correct value when somone has refcnt
5096 * on this or this is under rcu_read_lock(). Once css->id is allocated,
5097 * it's unchanged until freed.
5099 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5105 EXPORT_SYMBOL_GPL(css_id
);
5107 unsigned short css_depth(struct cgroup_subsys_state
*css
)
5109 struct css_id
*cssid
;
5111 cssid
= rcu_dereference_check(css
->id
, css_refcnt(css
));
5114 return cssid
->depth
;
5117 EXPORT_SYMBOL_GPL(css_depth
);
5120 * css_is_ancestor - test "root" css is an ancestor of "child"
5121 * @child: the css to be tested.
5122 * @root: the css supporsed to be an ancestor of the child.
5124 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
5125 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
5126 * But, considering usual usage, the csses should be valid objects after test.
5127 * Assuming that the caller will do some action to the child if this returns
5128 * returns true, the caller must take "child";s reference count.
5129 * If "child" is valid object and this returns true, "root" is valid, too.
5132 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
5133 const struct cgroup_subsys_state
*root
)
5135 struct css_id
*child_id
;
5136 struct css_id
*root_id
;
5140 child_id
= rcu_dereference(child
->id
);
5141 root_id
= rcu_dereference(root
->id
);
5144 || (child_id
->depth
< root_id
->depth
)
5145 || (child_id
->stack
[root_id
->depth
] != root_id
->id
))
5151 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
5153 struct css_id
*id
= css
->id
;
5154 /* When this is called before css_id initialization, id can be NULL */
5158 BUG_ON(!ss
->use_id
);
5160 rcu_assign_pointer(id
->css
, NULL
);
5161 rcu_assign_pointer(css
->id
, NULL
);
5162 spin_lock(&ss
->id_lock
);
5163 idr_remove(&ss
->idr
, id
->id
);
5164 spin_unlock(&ss
->id_lock
);
5165 kfree_rcu(id
, rcu_head
);
5167 EXPORT_SYMBOL_GPL(free_css_id
);
5170 * This is called by init or create(). Then, calls to this function are
5171 * always serialized (By cgroup_mutex() at create()).
5174 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
5176 struct css_id
*newid
;
5177 int myid
, error
, size
;
5179 BUG_ON(!ss
->use_id
);
5181 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
5182 newid
= kzalloc(size
, GFP_KERNEL
);
5184 return ERR_PTR(-ENOMEM
);
5186 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
5190 spin_lock(&ss
->id_lock
);
5191 /* Don't use 0. allocates an ID of 1-65535 */
5192 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
5193 spin_unlock(&ss
->id_lock
);
5195 /* Returns error when there are no free spaces for new ID.*/
5200 if (myid
> CSS_ID_MAX
)
5204 newid
->depth
= depth
;
5208 spin_lock(&ss
->id_lock
);
5209 idr_remove(&ss
->idr
, myid
);
5210 spin_unlock(&ss
->id_lock
);
5213 return ERR_PTR(error
);
5217 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
5218 struct cgroup_subsys_state
*rootcss
)
5220 struct css_id
*newid
;
5222 spin_lock_init(&ss
->id_lock
);
5225 newid
= get_new_cssid(ss
, 0);
5227 return PTR_ERR(newid
);
5229 newid
->stack
[0] = newid
->id
;
5230 newid
->css
= rootcss
;
5231 rootcss
->id
= newid
;
5235 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
5236 struct cgroup
*child
)
5238 int subsys_id
, i
, depth
= 0;
5239 struct cgroup_subsys_state
*parent_css
, *child_css
;
5240 struct css_id
*child_id
, *parent_id
;
5242 subsys_id
= ss
->subsys_id
;
5243 parent_css
= parent
->subsys
[subsys_id
];
5244 child_css
= child
->subsys
[subsys_id
];
5245 parent_id
= parent_css
->id
;
5246 depth
= parent_id
->depth
+ 1;
5248 child_id
= get_new_cssid(ss
, depth
);
5249 if (IS_ERR(child_id
))
5250 return PTR_ERR(child_id
);
5252 for (i
= 0; i
< depth
; i
++)
5253 child_id
->stack
[i
] = parent_id
->stack
[i
];
5254 child_id
->stack
[depth
] = child_id
->id
;
5256 * child_id->css pointer will be set after this cgroup is available
5257 * see cgroup_populate_dir()
5259 rcu_assign_pointer(child_css
->id
, child_id
);
5265 * css_lookup - lookup css by id
5266 * @ss: cgroup subsys to be looked into.
5269 * Returns pointer to cgroup_subsys_state if there is valid one with id.
5270 * NULL if not. Should be called under rcu_read_lock()
5272 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
5274 struct css_id
*cssid
= NULL
;
5276 BUG_ON(!ss
->use_id
);
5277 cssid
= idr_find(&ss
->idr
, id
);
5279 if (unlikely(!cssid
))
5282 return rcu_dereference(cssid
->css
);
5284 EXPORT_SYMBOL_GPL(css_lookup
);
5287 * css_get_next - lookup next cgroup under specified hierarchy.
5288 * @ss: pointer to subsystem
5289 * @id: current position of iteration.
5290 * @root: pointer to css. search tree under this.
5291 * @foundid: position of found object.
5293 * Search next css under the specified hierarchy of rootid. Calling under
5294 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5296 struct cgroup_subsys_state
*
5297 css_get_next(struct cgroup_subsys
*ss
, int id
,
5298 struct cgroup_subsys_state
*root
, int *foundid
)
5300 struct cgroup_subsys_state
*ret
= NULL
;
5303 int rootid
= css_id(root
);
5304 int depth
= css_depth(root
);
5309 BUG_ON(!ss
->use_id
);
5310 WARN_ON_ONCE(!rcu_read_lock_held());
5312 /* fill start point for scan */
5316 * scan next entry from bitmap(tree), tmpid is updated after
5319 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
5322 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
5323 ret
= rcu_dereference(tmp
->css
);
5329 /* continue to scan from next id */
5336 * get corresponding css from file open on cgroupfs directory
5338 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5340 struct cgroup
*cgrp
;
5341 struct inode
*inode
;
5342 struct cgroup_subsys_state
*css
;
5344 inode
= f
->f_dentry
->d_inode
;
5345 /* check in cgroup filesystem dir */
5346 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5347 return ERR_PTR(-EBADF
);
5349 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5350 return ERR_PTR(-EINVAL
);
5353 cgrp
= __d_cgrp(f
->f_dentry
);
5354 css
= cgrp
->subsys
[id
];
5355 return css
? css
: ERR_PTR(-ENOENT
);
5358 #ifdef CONFIG_CGROUP_DEBUG
5359 static struct cgroup_subsys_state
*debug_create(struct cgroup
*cont
)
5361 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5364 return ERR_PTR(-ENOMEM
);
5369 static void debug_destroy(struct cgroup
*cont
)
5371 kfree(cont
->subsys
[debug_subsys_id
]);
5374 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5376 return atomic_read(&cont
->count
);
5379 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5381 return cgroup_task_count(cont
);
5384 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5386 return (u64
)(unsigned long)current
->cgroups
;
5389 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5395 count
= atomic_read(¤t
->cgroups
->refcount
);
5400 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5402 struct seq_file
*seq
)
5404 struct cg_cgroup_link
*link
;
5407 read_lock(&css_set_lock
);
5409 cg
= rcu_dereference(current
->cgroups
);
5410 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5411 struct cgroup
*c
= link
->cgrp
;
5415 name
= c
->dentry
->d_name
.name
;
5418 seq_printf(seq
, "Root %d group %s\n",
5419 c
->root
->hierarchy_id
, name
);
5422 read_unlock(&css_set_lock
);
5426 #define MAX_TASKS_SHOWN_PER_CSS 25
5427 static int cgroup_css_links_read(struct cgroup
*cont
,
5429 struct seq_file
*seq
)
5431 struct cg_cgroup_link
*link
;
5433 read_lock(&css_set_lock
);
5434 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5435 struct css_set
*cg
= link
->cg
;
5436 struct task_struct
*task
;
5438 seq_printf(seq
, "css_set %p\n", cg
);
5439 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5440 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5441 seq_puts(seq
, " ...\n");
5444 seq_printf(seq
, " task %d\n",
5445 task_pid_vnr(task
));
5449 read_unlock(&css_set_lock
);
5453 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5455 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5458 static struct cftype debug_files
[] = {
5460 .name
= "cgroup_refcount",
5461 .read_u64
= cgroup_refcount_read
,
5464 .name
= "taskcount",
5465 .read_u64
= debug_taskcount_read
,
5469 .name
= "current_css_set",
5470 .read_u64
= current_css_set_read
,
5474 .name
= "current_css_set_refcount",
5475 .read_u64
= current_css_set_refcount_read
,
5479 .name
= "current_css_set_cg_links",
5480 .read_seq_string
= current_css_set_cg_links_read
,
5484 .name
= "cgroup_css_links",
5485 .read_seq_string
= cgroup_css_links_read
,
5489 .name
= "releasable",
5490 .read_u64
= releasable_read
,
5496 struct cgroup_subsys debug_subsys
= {
5498 .create
= debug_create
,
5499 .destroy
= debug_destroy
,
5500 .subsys_id
= debug_subsys_id
,
5501 .base_cftypes
= debug_files
,
5503 #endif /* CONFIG_CGROUP_DEBUG */