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 <asm/atomic.h>
66 static DEFINE_MUTEX(cgroup_mutex
);
69 * Generate an array of cgroup subsystem pointers. At boot time, this is
70 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are
71 * registered after that. The mutable section of this array is protected by
74 #define SUBSYS(_x) &_x ## _subsys,
75 static struct cgroup_subsys
*subsys
[CGROUP_SUBSYS_COUNT
] = {
76 #include <linux/cgroup_subsys.h>
79 #define MAX_CGROUP_ROOT_NAMELEN 64
82 * A cgroupfs_root represents the root of a cgroup hierarchy,
83 * and may be associated with a superblock to form an active
86 struct cgroupfs_root
{
87 struct super_block
*sb
;
90 * The bitmask of subsystems intended to be attached to this
93 unsigned long subsys_bits
;
95 /* Unique id for this hierarchy. */
98 /* The bitmask of subsystems currently attached to this hierarchy */
99 unsigned long actual_subsys_bits
;
101 /* A list running through the attached subsystems */
102 struct list_head subsys_list
;
104 /* The root cgroup for this hierarchy */
105 struct cgroup top_cgroup
;
107 /* Tracks how many cgroups are currently defined in hierarchy.*/
108 int number_of_cgroups
;
110 /* A list running through the active hierarchies */
111 struct list_head root_list
;
113 /* Hierarchy-specific flags */
116 /* The path to use for release notifications. */
117 char release_agent_path
[PATH_MAX
];
119 /* The name for this hierarchy - may be empty */
120 char name
[MAX_CGROUP_ROOT_NAMELEN
];
124 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
125 * subsystems that are otherwise unattached - it never has more than a
126 * single cgroup, and all tasks are part of that cgroup.
128 static struct cgroupfs_root rootnode
;
131 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
132 * cgroup_subsys->use_id != 0.
134 #define CSS_ID_MAX (65535)
137 * The css to which this ID points. This pointer is set to valid value
138 * after cgroup is populated. If cgroup is removed, this will be NULL.
139 * This pointer is expected to be RCU-safe because destroy()
140 * is called after synchronize_rcu(). But for safe use, css_is_removed()
141 * css_tryget() should be used for avoiding race.
143 struct cgroup_subsys_state __rcu
*css
;
149 * Depth in hierarchy which this ID belongs to.
151 unsigned short depth
;
153 * ID is freed by RCU. (and lookup routine is RCU safe.)
155 struct rcu_head rcu_head
;
157 * Hierarchy of CSS ID belongs to.
159 unsigned short stack
[0]; /* Array of Length (depth+1) */
163 * cgroup_event represents events which userspace want to receive.
165 struct cgroup_event
{
167 * Cgroup which the event belongs to.
171 * Control file which the event associated.
175 * eventfd to signal userspace about the event.
177 struct eventfd_ctx
*eventfd
;
179 * Each of these stored in a list by the cgroup.
181 struct list_head list
;
183 * All fields below needed to unregister event when
184 * userspace closes eventfd.
187 wait_queue_head_t
*wqh
;
189 struct work_struct remove
;
192 /* The list of hierarchy roots */
194 static LIST_HEAD(roots
);
195 static int root_count
;
197 static DEFINE_IDA(hierarchy_ida
);
198 static int next_hierarchy_id
;
199 static DEFINE_SPINLOCK(hierarchy_id_lock
);
201 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
202 #define dummytop (&rootnode.top_cgroup)
204 /* This flag indicates whether tasks in the fork and exit paths should
205 * check for fork/exit handlers to call. This avoids us having to do
206 * extra work in the fork/exit path if none of the subsystems need to
209 static int need_forkexit_callback __read_mostly
;
211 #ifdef CONFIG_PROVE_LOCKING
212 int cgroup_lock_is_held(void)
214 return lockdep_is_held(&cgroup_mutex
);
216 #else /* #ifdef CONFIG_PROVE_LOCKING */
217 int cgroup_lock_is_held(void)
219 return mutex_is_locked(&cgroup_mutex
);
221 #endif /* #else #ifdef CONFIG_PROVE_LOCKING */
223 EXPORT_SYMBOL_GPL(cgroup_lock_is_held
);
225 /* convenient tests for these bits */
226 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
228 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
231 /* bits in struct cgroupfs_root flags field */
233 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
236 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
239 (1 << CGRP_RELEASABLE
) |
240 (1 << CGRP_NOTIFY_ON_RELEASE
);
241 return (cgrp
->flags
& bits
) == bits
;
244 static int notify_on_release(const struct cgroup
*cgrp
)
246 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
249 static int clone_children(const struct cgroup
*cgrp
)
251 return test_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
255 * for_each_subsys() allows you to iterate on each subsystem attached to
256 * an active hierarchy
258 #define for_each_subsys(_root, _ss) \
259 list_for_each_entry(_ss, &_root->subsys_list, sibling)
261 /* for_each_active_root() allows you to iterate across the active hierarchies */
262 #define for_each_active_root(_root) \
263 list_for_each_entry(_root, &roots, root_list)
265 /* the list of cgroups eligible for automatic release. Protected by
266 * release_list_lock */
267 static LIST_HEAD(release_list
);
268 static DEFINE_SPINLOCK(release_list_lock
);
269 static void cgroup_release_agent(struct work_struct
*work
);
270 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
271 static void check_for_release(struct cgroup
*cgrp
);
273 /* Link structure for associating css_set objects with cgroups */
274 struct cg_cgroup_link
{
276 * List running through cg_cgroup_links associated with a
277 * cgroup, anchored on cgroup->css_sets
279 struct list_head cgrp_link_list
;
282 * List running through cg_cgroup_links pointing at a
283 * single css_set object, anchored on css_set->cg_links
285 struct list_head cg_link_list
;
289 /* The default css_set - used by init and its children prior to any
290 * hierarchies being mounted. It contains a pointer to the root state
291 * for each subsystem. Also used to anchor the list of css_sets. Not
292 * reference-counted, to improve performance when child cgroups
293 * haven't been created.
296 static struct css_set init_css_set
;
297 static struct cg_cgroup_link init_css_set_link
;
299 static int cgroup_init_idr(struct cgroup_subsys
*ss
,
300 struct cgroup_subsys_state
*css
);
302 /* css_set_lock protects the list of css_set objects, and the
303 * chain of tasks off each css_set. Nests outside task->alloc_lock
304 * due to cgroup_iter_start() */
305 static DEFINE_RWLOCK(css_set_lock
);
306 static int css_set_count
;
309 * hash table for cgroup groups. This improves the performance to find
310 * an existing css_set. This hash doesn't (currently) take into
311 * account cgroups in empty hierarchies.
313 #define CSS_SET_HASH_BITS 7
314 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
315 static struct hlist_head css_set_table
[CSS_SET_TABLE_SIZE
];
317 static struct hlist_head
*css_set_hash(struct cgroup_subsys_state
*css
[])
321 unsigned long tmp
= 0UL;
323 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
324 tmp
+= (unsigned long)css
[i
];
325 tmp
= (tmp
>> 16) ^ tmp
;
327 index
= hash_long(tmp
, CSS_SET_HASH_BITS
);
329 return &css_set_table
[index
];
332 /* We don't maintain the lists running through each css_set to its
333 * task until after the first call to cgroup_iter_start(). This
334 * reduces the fork()/exit() overhead for people who have cgroups
335 * compiled into their kernel but not actually in use */
336 static int use_task_css_set_links __read_mostly
;
338 static void __put_css_set(struct css_set
*cg
, int taskexit
)
340 struct cg_cgroup_link
*link
;
341 struct cg_cgroup_link
*saved_link
;
343 * Ensure that the refcount doesn't hit zero while any readers
344 * can see it. Similar to atomic_dec_and_lock(), but for an
347 if (atomic_add_unless(&cg
->refcount
, -1, 1))
349 write_lock(&css_set_lock
);
350 if (!atomic_dec_and_test(&cg
->refcount
)) {
351 write_unlock(&css_set_lock
);
355 /* This css_set is dead. unlink it and release cgroup refcounts */
356 hlist_del(&cg
->hlist
);
359 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
361 struct cgroup
*cgrp
= link
->cgrp
;
362 list_del(&link
->cg_link_list
);
363 list_del(&link
->cgrp_link_list
);
364 if (atomic_dec_and_test(&cgrp
->count
) &&
365 notify_on_release(cgrp
)) {
367 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
368 check_for_release(cgrp
);
374 write_unlock(&css_set_lock
);
375 kfree_rcu(cg
, rcu_head
);
379 * refcounted get/put for css_set objects
381 static inline void get_css_set(struct css_set
*cg
)
383 atomic_inc(&cg
->refcount
);
386 static inline void put_css_set(struct css_set
*cg
)
388 __put_css_set(cg
, 0);
391 static inline void put_css_set_taskexit(struct css_set
*cg
)
393 __put_css_set(cg
, 1);
397 * compare_css_sets - helper function for find_existing_css_set().
398 * @cg: candidate css_set being tested
399 * @old_cg: existing css_set for a task
400 * @new_cgrp: cgroup that's being entered by the task
401 * @template: desired set of css pointers in css_set (pre-calculated)
403 * Returns true if "cg" matches "old_cg" except for the hierarchy
404 * which "new_cgrp" belongs to, for which it should match "new_cgrp".
406 static bool compare_css_sets(struct css_set
*cg
,
407 struct css_set
*old_cg
,
408 struct cgroup
*new_cgrp
,
409 struct cgroup_subsys_state
*template[])
411 struct list_head
*l1
, *l2
;
413 if (memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
414 /* Not all subsystems matched */
419 * Compare cgroup pointers in order to distinguish between
420 * different cgroups in heirarchies with no subsystems. We
421 * could get by with just this check alone (and skip the
422 * memcmp above) but on most setups the memcmp check will
423 * avoid the need for this more expensive check on almost all
428 l2
= &old_cg
->cg_links
;
430 struct cg_cgroup_link
*cgl1
, *cgl2
;
431 struct cgroup
*cg1
, *cg2
;
435 /* See if we reached the end - both lists are equal length. */
436 if (l1
== &cg
->cg_links
) {
437 BUG_ON(l2
!= &old_cg
->cg_links
);
440 BUG_ON(l2
== &old_cg
->cg_links
);
442 /* Locate the cgroups associated with these links. */
443 cgl1
= list_entry(l1
, struct cg_cgroup_link
, cg_link_list
);
444 cgl2
= list_entry(l2
, struct cg_cgroup_link
, cg_link_list
);
447 /* Hierarchies should be linked in the same order. */
448 BUG_ON(cg1
->root
!= cg2
->root
);
451 * If this hierarchy is the hierarchy of the cgroup
452 * that's changing, then we need to check that this
453 * css_set points to the new cgroup; if it's any other
454 * hierarchy, then this css_set should point to the
455 * same cgroup as the old css_set.
457 if (cg1
->root
== new_cgrp
->root
) {
469 * find_existing_css_set() is a helper for
470 * find_css_set(), and checks to see whether an existing
471 * css_set is suitable.
473 * oldcg: the cgroup group that we're using before the cgroup
476 * cgrp: the cgroup that we're moving into
478 * template: location in which to build the desired set of subsystem
479 * state objects for the new cgroup group
481 static struct css_set
*find_existing_css_set(
482 struct css_set
*oldcg
,
484 struct cgroup_subsys_state
*template[])
487 struct cgroupfs_root
*root
= cgrp
->root
;
488 struct hlist_head
*hhead
;
489 struct hlist_node
*node
;
493 * Build the set of subsystem state objects that we want to see in the
494 * new css_set. while subsystems can change globally, the entries here
495 * won't change, so no need for locking.
497 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
498 if (root
->subsys_bits
& (1UL << i
)) {
499 /* Subsystem is in this hierarchy. So we want
500 * the subsystem state from the new
502 template[i
] = cgrp
->subsys
[i
];
504 /* Subsystem is not in this hierarchy, so we
505 * don't want to change the subsystem state */
506 template[i
] = oldcg
->subsys
[i
];
510 hhead
= css_set_hash(template);
511 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
512 if (!compare_css_sets(cg
, oldcg
, cgrp
, template))
515 /* This css_set matches what we need */
519 /* No existing cgroup group matched */
523 static void free_cg_links(struct list_head
*tmp
)
525 struct cg_cgroup_link
*link
;
526 struct cg_cgroup_link
*saved_link
;
528 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
529 list_del(&link
->cgrp_link_list
);
535 * allocate_cg_links() allocates "count" cg_cgroup_link structures
536 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
537 * success or a negative error
539 static int allocate_cg_links(int count
, struct list_head
*tmp
)
541 struct cg_cgroup_link
*link
;
544 for (i
= 0; i
< count
; i
++) {
545 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
550 list_add(&link
->cgrp_link_list
, tmp
);
556 * link_css_set - a helper function to link a css_set to a cgroup
557 * @tmp_cg_links: cg_cgroup_link objects allocated by allocate_cg_links()
558 * @cg: the css_set to be linked
559 * @cgrp: the destination cgroup
561 static void link_css_set(struct list_head
*tmp_cg_links
,
562 struct css_set
*cg
, struct cgroup
*cgrp
)
564 struct cg_cgroup_link
*link
;
566 BUG_ON(list_empty(tmp_cg_links
));
567 link
= list_first_entry(tmp_cg_links
, struct cg_cgroup_link
,
571 atomic_inc(&cgrp
->count
);
572 list_move(&link
->cgrp_link_list
, &cgrp
->css_sets
);
574 * Always add links to the tail of the list so that the list
575 * is sorted by order of hierarchy creation
577 list_add_tail(&link
->cg_link_list
, &cg
->cg_links
);
581 * find_css_set() takes an existing cgroup group and a
582 * cgroup object, and returns a css_set object that's
583 * equivalent to the old group, but with the given cgroup
584 * substituted into the appropriate hierarchy. Must be called with
587 static struct css_set
*find_css_set(
588 struct css_set
*oldcg
, struct cgroup
*cgrp
)
591 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
593 struct list_head tmp_cg_links
;
595 struct hlist_head
*hhead
;
596 struct cg_cgroup_link
*link
;
598 /* First see if we already have a cgroup group that matches
600 read_lock(&css_set_lock
);
601 res
= find_existing_css_set(oldcg
, cgrp
, template);
604 read_unlock(&css_set_lock
);
609 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
613 /* Allocate all the cg_cgroup_link objects that we'll need */
614 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
619 atomic_set(&res
->refcount
, 1);
620 INIT_LIST_HEAD(&res
->cg_links
);
621 INIT_LIST_HEAD(&res
->tasks
);
622 INIT_HLIST_NODE(&res
->hlist
);
624 /* Copy the set of subsystem state objects generated in
625 * find_existing_css_set() */
626 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
628 write_lock(&css_set_lock
);
629 /* Add reference counts and links from the new css_set. */
630 list_for_each_entry(link
, &oldcg
->cg_links
, cg_link_list
) {
631 struct cgroup
*c
= link
->cgrp
;
632 if (c
->root
== cgrp
->root
)
634 link_css_set(&tmp_cg_links
, res
, c
);
637 BUG_ON(!list_empty(&tmp_cg_links
));
641 /* Add this cgroup group to the hash table */
642 hhead
= css_set_hash(res
->subsys
);
643 hlist_add_head(&res
->hlist
, hhead
);
645 write_unlock(&css_set_lock
);
651 * Return the cgroup for "task" from the given hierarchy. Must be
652 * called with cgroup_mutex held.
654 static struct cgroup
*task_cgroup_from_root(struct task_struct
*task
,
655 struct cgroupfs_root
*root
)
658 struct cgroup
*res
= NULL
;
660 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
661 read_lock(&css_set_lock
);
663 * No need to lock the task - since we hold cgroup_mutex the
664 * task can't change groups, so the only thing that can happen
665 * is that it exits and its css is set back to init_css_set.
668 if (css
== &init_css_set
) {
669 res
= &root
->top_cgroup
;
671 struct cg_cgroup_link
*link
;
672 list_for_each_entry(link
, &css
->cg_links
, cg_link_list
) {
673 struct cgroup
*c
= link
->cgrp
;
674 if (c
->root
== root
) {
680 read_unlock(&css_set_lock
);
686 * There is one global cgroup mutex. We also require taking
687 * task_lock() when dereferencing a task's cgroup subsys pointers.
688 * See "The task_lock() exception", at the end of this comment.
690 * A task must hold cgroup_mutex to modify cgroups.
692 * Any task can increment and decrement the count field without lock.
693 * So in general, code holding cgroup_mutex can't rely on the count
694 * field not changing. However, if the count goes to zero, then only
695 * cgroup_attach_task() can increment it again. Because a count of zero
696 * means that no tasks are currently attached, therefore there is no
697 * way a task attached to that cgroup can fork (the other way to
698 * increment the count). So code holding cgroup_mutex can safely
699 * assume that if the count is zero, it will stay zero. Similarly, if
700 * a task holds cgroup_mutex on a cgroup with zero count, it
701 * knows that the cgroup won't be removed, as cgroup_rmdir()
704 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
705 * (usually) take cgroup_mutex. These are the two most performance
706 * critical pieces of code here. The exception occurs on cgroup_exit(),
707 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
708 * is taken, and if the cgroup count is zero, a usermode call made
709 * to the release agent with the name of the cgroup (path relative to
710 * the root of cgroup file system) as the argument.
712 * A cgroup can only be deleted if both its 'count' of using tasks
713 * is zero, and its list of 'children' cgroups is empty. Since all
714 * tasks in the system use _some_ cgroup, and since there is always at
715 * least one task in the system (init, pid == 1), therefore, top_cgroup
716 * always has either children cgroups and/or using tasks. So we don't
717 * need a special hack to ensure that top_cgroup cannot be deleted.
719 * The task_lock() exception
721 * The need for this exception arises from the action of
722 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
723 * another. It does so using cgroup_mutex, however there are
724 * several performance critical places that need to reference
725 * task->cgroup without the expense of grabbing a system global
726 * mutex. Therefore except as noted below, when dereferencing or, as
727 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
728 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
729 * the task_struct routinely used for such matters.
731 * P.S. One more locking exception. RCU is used to guard the
732 * update of a tasks cgroup pointer by cgroup_attach_task()
736 * cgroup_lock - lock out any changes to cgroup structures
739 void cgroup_lock(void)
741 mutex_lock(&cgroup_mutex
);
743 EXPORT_SYMBOL_GPL(cgroup_lock
);
746 * cgroup_unlock - release lock on cgroup changes
748 * Undo the lock taken in a previous cgroup_lock() call.
750 void cgroup_unlock(void)
752 mutex_unlock(&cgroup_mutex
);
754 EXPORT_SYMBOL_GPL(cgroup_unlock
);
757 * A couple of forward declarations required, due to cyclic reference loop:
758 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
759 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
763 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
);
764 static struct dentry
*cgroup_lookup(struct inode
*, struct dentry
*, struct nameidata
*);
765 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
766 static int cgroup_populate_dir(struct cgroup
*cgrp
);
767 static const struct inode_operations cgroup_dir_inode_operations
;
768 static const struct file_operations proc_cgroupstats_operations
;
770 static struct backing_dev_info cgroup_backing_dev_info
= {
772 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
775 static int alloc_css_id(struct cgroup_subsys
*ss
,
776 struct cgroup
*parent
, struct cgroup
*child
);
778 static struct inode
*cgroup_new_inode(mode_t mode
, struct super_block
*sb
)
780 struct inode
*inode
= new_inode(sb
);
783 inode
->i_ino
= get_next_ino();
784 inode
->i_mode
= mode
;
785 inode
->i_uid
= current_fsuid();
786 inode
->i_gid
= current_fsgid();
787 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
788 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
794 * Call subsys's pre_destroy handler.
795 * This is called before css refcnt check.
797 static int cgroup_call_pre_destroy(struct cgroup
*cgrp
)
799 struct cgroup_subsys
*ss
;
802 for_each_subsys(cgrp
->root
, ss
)
803 if (ss
->pre_destroy
) {
804 ret
= ss
->pre_destroy(ss
, cgrp
);
812 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
814 /* is dentry a directory ? if so, kfree() associated cgroup */
815 if (S_ISDIR(inode
->i_mode
)) {
816 struct cgroup
*cgrp
= dentry
->d_fsdata
;
817 struct cgroup_subsys
*ss
;
818 BUG_ON(!(cgroup_is_removed(cgrp
)));
819 /* It's possible for external users to be holding css
820 * reference counts on a cgroup; css_put() needs to
821 * be able to access the cgroup after decrementing
822 * the reference count in order to know if it needs to
823 * queue the cgroup to be handled by the release
827 mutex_lock(&cgroup_mutex
);
829 * Release the subsystem state objects.
831 for_each_subsys(cgrp
->root
, ss
)
832 ss
->destroy(ss
, cgrp
);
834 cgrp
->root
->number_of_cgroups
--;
835 mutex_unlock(&cgroup_mutex
);
838 * Drop the active superblock reference that we took when we
841 deactivate_super(cgrp
->root
->sb
);
844 * if we're getting rid of the cgroup, refcount should ensure
845 * that there are no pidlists left.
847 BUG_ON(!list_empty(&cgrp
->pidlists
));
849 kfree_rcu(cgrp
, rcu_head
);
854 static int cgroup_delete(const struct dentry
*d
)
859 static void remove_dir(struct dentry
*d
)
861 struct dentry
*parent
= dget(d
->d_parent
);
864 simple_rmdir(parent
->d_inode
, d
);
868 static void cgroup_clear_directory(struct dentry
*dentry
)
870 struct list_head
*node
;
872 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
873 spin_lock(&dentry
->d_lock
);
874 node
= dentry
->d_subdirs
.next
;
875 while (node
!= &dentry
->d_subdirs
) {
876 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
878 spin_lock_nested(&d
->d_lock
, DENTRY_D_LOCK_NESTED
);
881 /* This should never be called on a cgroup
882 * directory with child cgroups */
883 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
885 spin_unlock(&d
->d_lock
);
886 spin_unlock(&dentry
->d_lock
);
888 simple_unlink(dentry
->d_inode
, d
);
890 spin_lock(&dentry
->d_lock
);
892 spin_unlock(&d
->d_lock
);
893 node
= dentry
->d_subdirs
.next
;
895 spin_unlock(&dentry
->d_lock
);
899 * NOTE : the dentry must have been dget()'ed
901 static void cgroup_d_remove_dir(struct dentry
*dentry
)
903 struct dentry
*parent
;
905 cgroup_clear_directory(dentry
);
907 parent
= dentry
->d_parent
;
908 spin_lock(&parent
->d_lock
);
909 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
910 list_del_init(&dentry
->d_u
.d_child
);
911 spin_unlock(&dentry
->d_lock
);
912 spin_unlock(&parent
->d_lock
);
917 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
918 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
919 * reference to css->refcnt. In general, this refcnt is expected to goes down
922 * CGRP_WAIT_ON_RMDIR flag is set under cgroup's inode->i_mutex;
924 DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq
);
926 static void cgroup_wakeup_rmdir_waiter(struct cgroup
*cgrp
)
928 if (unlikely(test_and_clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
)))
929 wake_up_all(&cgroup_rmdir_waitq
);
932 void cgroup_exclude_rmdir(struct cgroup_subsys_state
*css
)
937 void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state
*css
)
939 cgroup_wakeup_rmdir_waiter(css
->cgroup
);
944 * Call with cgroup_mutex held. Drops reference counts on modules, including
945 * any duplicate ones that parse_cgroupfs_options took. If this function
946 * returns an error, no reference counts are touched.
948 static int rebind_subsystems(struct cgroupfs_root
*root
,
949 unsigned long final_bits
)
951 unsigned long added_bits
, removed_bits
;
952 struct cgroup
*cgrp
= &root
->top_cgroup
;
955 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
957 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
958 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
959 /* Check that any added subsystems are currently free */
960 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
961 unsigned long bit
= 1UL << i
;
962 struct cgroup_subsys
*ss
= subsys
[i
];
963 if (!(bit
& added_bits
))
966 * Nobody should tell us to do a subsys that doesn't exist:
967 * parse_cgroupfs_options should catch that case and refcounts
968 * ensure that subsystems won't disappear once selected.
971 if (ss
->root
!= &rootnode
) {
972 /* Subsystem isn't free */
977 /* Currently we don't handle adding/removing subsystems when
978 * any child cgroups exist. This is theoretically supportable
979 * but involves complex error handling, so it's being left until
981 if (root
->number_of_cgroups
> 1)
984 /* Process each subsystem */
985 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
986 struct cgroup_subsys
*ss
= subsys
[i
];
987 unsigned long bit
= 1UL << i
;
988 if (bit
& added_bits
) {
989 /* We're binding this subsystem to this hierarchy */
991 BUG_ON(cgrp
->subsys
[i
]);
992 BUG_ON(!dummytop
->subsys
[i
]);
993 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
994 mutex_lock(&ss
->hierarchy_mutex
);
995 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
996 cgrp
->subsys
[i
]->cgroup
= cgrp
;
997 list_move(&ss
->sibling
, &root
->subsys_list
);
1001 mutex_unlock(&ss
->hierarchy_mutex
);
1002 /* refcount was already taken, and we're keeping it */
1003 } else if (bit
& removed_bits
) {
1004 /* We're removing this subsystem */
1006 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
1007 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
1008 mutex_lock(&ss
->hierarchy_mutex
);
1010 ss
->bind(ss
, dummytop
);
1011 dummytop
->subsys
[i
]->cgroup
= dummytop
;
1012 cgrp
->subsys
[i
] = NULL
;
1013 subsys
[i
]->root
= &rootnode
;
1014 list_move(&ss
->sibling
, &rootnode
.subsys_list
);
1015 mutex_unlock(&ss
->hierarchy_mutex
);
1016 /* subsystem is now free - drop reference on module */
1017 module_put(ss
->module
);
1018 } else if (bit
& final_bits
) {
1019 /* Subsystem state should already exist */
1021 BUG_ON(!cgrp
->subsys
[i
]);
1023 * a refcount was taken, but we already had one, so
1024 * drop the extra reference.
1026 module_put(ss
->module
);
1027 #ifdef CONFIG_MODULE_UNLOAD
1028 BUG_ON(ss
->module
&& !module_refcount(ss
->module
));
1031 /* Subsystem state shouldn't exist */
1032 BUG_ON(cgrp
->subsys
[i
]);
1035 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
1041 static int cgroup_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
1043 struct cgroupfs_root
*root
= vfs
->mnt_sb
->s_fs_info
;
1044 struct cgroup_subsys
*ss
;
1046 mutex_lock(&cgroup_mutex
);
1047 for_each_subsys(root
, ss
)
1048 seq_printf(seq
, ",%s", ss
->name
);
1049 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
1050 seq_puts(seq
, ",noprefix");
1051 if (strlen(root
->release_agent_path
))
1052 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
1053 if (clone_children(&root
->top_cgroup
))
1054 seq_puts(seq
, ",clone_children");
1055 if (strlen(root
->name
))
1056 seq_printf(seq
, ",name=%s", root
->name
);
1057 mutex_unlock(&cgroup_mutex
);
1061 struct cgroup_sb_opts
{
1062 unsigned long subsys_bits
;
1063 unsigned long flags
;
1064 char *release_agent
;
1065 bool clone_children
;
1067 /* User explicitly requested empty subsystem */
1070 struct cgroupfs_root
*new_root
;
1075 * Convert a hierarchy specifier into a bitmask of subsystems and flags. Call
1076 * with cgroup_mutex held to protect the subsys[] array. This function takes
1077 * refcounts on subsystems to be used, unless it returns error, in which case
1078 * no refcounts are taken.
1080 static int parse_cgroupfs_options(char *data
, struct cgroup_sb_opts
*opts
)
1082 char *token
, *o
= data
;
1083 bool all_ss
= false, one_ss
= false;
1084 unsigned long mask
= (unsigned long)-1;
1086 bool module_pin_failed
= false;
1088 BUG_ON(!mutex_is_locked(&cgroup_mutex
));
1090 #ifdef CONFIG_CPUSETS
1091 mask
= ~(1UL << cpuset_subsys_id
);
1094 memset(opts
, 0, sizeof(*opts
));
1096 while ((token
= strsep(&o
, ",")) != NULL
) {
1099 if (!strcmp(token
, "none")) {
1100 /* Explicitly have no subsystems */
1104 if (!strcmp(token
, "all")) {
1105 /* Mutually exclusive option 'all' + subsystem name */
1111 if (!strcmp(token
, "noprefix")) {
1112 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
1115 if (!strcmp(token
, "clone_children")) {
1116 opts
->clone_children
= true;
1119 if (!strncmp(token
, "release_agent=", 14)) {
1120 /* Specifying two release agents is forbidden */
1121 if (opts
->release_agent
)
1123 opts
->release_agent
=
1124 kstrndup(token
+ 14, PATH_MAX
- 1, GFP_KERNEL
);
1125 if (!opts
->release_agent
)
1129 if (!strncmp(token
, "name=", 5)) {
1130 const char *name
= token
+ 5;
1131 /* Can't specify an empty name */
1134 /* Must match [\w.-]+ */
1135 for (i
= 0; i
< strlen(name
); i
++) {
1139 if ((c
== '.') || (c
== '-') || (c
== '_'))
1143 /* Specifying two names is forbidden */
1146 opts
->name
= kstrndup(name
,
1147 MAX_CGROUP_ROOT_NAMELEN
- 1,
1155 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1156 struct cgroup_subsys
*ss
= subsys
[i
];
1159 if (strcmp(token
, ss
->name
))
1164 /* Mutually exclusive option 'all' + subsystem name */
1167 set_bit(i
, &opts
->subsys_bits
);
1172 if (i
== CGROUP_SUBSYS_COUNT
)
1177 * If the 'all' option was specified select all the subsystems,
1178 * otherwise 'all, 'none' and a subsystem name options were not
1179 * specified, let's default to 'all'
1181 if (all_ss
|| (!all_ss
&& !one_ss
&& !opts
->none
)) {
1182 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1183 struct cgroup_subsys
*ss
= subsys
[i
];
1188 set_bit(i
, &opts
->subsys_bits
);
1192 /* Consistency checks */
1195 * Option noprefix was introduced just for backward compatibility
1196 * with the old cpuset, so we allow noprefix only if mounting just
1197 * the cpuset subsystem.
1199 if (test_bit(ROOT_NOPREFIX
, &opts
->flags
) &&
1200 (opts
->subsys_bits
& mask
))
1204 /* Can't specify "none" and some subsystems */
1205 if (opts
->subsys_bits
&& opts
->none
)
1209 * We either have to specify by name or by subsystems. (So all
1210 * empty hierarchies must have a name).
1212 if (!opts
->subsys_bits
&& !opts
->name
)
1216 * Grab references on all the modules we'll need, so the subsystems
1217 * don't dance around before rebind_subsystems attaches them. This may
1218 * take duplicate reference counts on a subsystem that's already used,
1219 * but rebind_subsystems handles this case.
1221 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1222 unsigned long bit
= 1UL << i
;
1224 if (!(bit
& opts
->subsys_bits
))
1226 if (!try_module_get(subsys
[i
]->module
)) {
1227 module_pin_failed
= true;
1231 if (module_pin_failed
) {
1233 * oops, one of the modules was going away. this means that we
1234 * raced with a module_delete call, and to the user this is
1235 * essentially a "subsystem doesn't exist" case.
1237 for (i
--; i
>= CGROUP_BUILTIN_SUBSYS_COUNT
; i
--) {
1238 /* drop refcounts only on the ones we took */
1239 unsigned long bit
= 1UL << i
;
1241 if (!(bit
& opts
->subsys_bits
))
1243 module_put(subsys
[i
]->module
);
1251 static void drop_parsed_module_refcounts(unsigned long subsys_bits
)
1254 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1255 unsigned long bit
= 1UL << i
;
1257 if (!(bit
& subsys_bits
))
1259 module_put(subsys
[i
]->module
);
1263 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
1266 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1267 struct cgroup
*cgrp
= &root
->top_cgroup
;
1268 struct cgroup_sb_opts opts
;
1270 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
1271 mutex_lock(&cgroup_mutex
);
1273 /* See what subsystems are wanted */
1274 ret
= parse_cgroupfs_options(data
, &opts
);
1278 /* Don't allow flags or name to change at remount */
1279 if (opts
.flags
!= root
->flags
||
1280 (opts
.name
&& strcmp(opts
.name
, root
->name
))) {
1282 drop_parsed_module_refcounts(opts
.subsys_bits
);
1286 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
1288 drop_parsed_module_refcounts(opts
.subsys_bits
);
1292 /* (re)populate subsystem files */
1293 cgroup_populate_dir(cgrp
);
1295 if (opts
.release_agent
)
1296 strcpy(root
->release_agent_path
, opts
.release_agent
);
1298 kfree(opts
.release_agent
);
1300 mutex_unlock(&cgroup_mutex
);
1301 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
1305 static const struct super_operations cgroup_ops
= {
1306 .statfs
= simple_statfs
,
1307 .drop_inode
= generic_delete_inode
,
1308 .show_options
= cgroup_show_options
,
1309 .remount_fs
= cgroup_remount
,
1312 static void init_cgroup_housekeeping(struct cgroup
*cgrp
)
1314 INIT_LIST_HEAD(&cgrp
->sibling
);
1315 INIT_LIST_HEAD(&cgrp
->children
);
1316 INIT_LIST_HEAD(&cgrp
->css_sets
);
1317 INIT_LIST_HEAD(&cgrp
->release_list
);
1318 INIT_LIST_HEAD(&cgrp
->pidlists
);
1319 mutex_init(&cgrp
->pidlist_mutex
);
1320 INIT_LIST_HEAD(&cgrp
->event_list
);
1321 spin_lock_init(&cgrp
->event_list_lock
);
1324 static void init_cgroup_root(struct cgroupfs_root
*root
)
1326 struct cgroup
*cgrp
= &root
->top_cgroup
;
1327 INIT_LIST_HEAD(&root
->subsys_list
);
1328 INIT_LIST_HEAD(&root
->root_list
);
1329 root
->number_of_cgroups
= 1;
1331 cgrp
->top_cgroup
= cgrp
;
1332 init_cgroup_housekeeping(cgrp
);
1335 static bool init_root_id(struct cgroupfs_root
*root
)
1340 if (!ida_pre_get(&hierarchy_ida
, GFP_KERNEL
))
1342 spin_lock(&hierarchy_id_lock
);
1343 /* Try to allocate the next unused ID */
1344 ret
= ida_get_new_above(&hierarchy_ida
, next_hierarchy_id
,
1345 &root
->hierarchy_id
);
1347 /* Try again starting from 0 */
1348 ret
= ida_get_new(&hierarchy_ida
, &root
->hierarchy_id
);
1350 next_hierarchy_id
= root
->hierarchy_id
+ 1;
1351 } else if (ret
!= -EAGAIN
) {
1352 /* Can only get here if the 31-bit IDR is full ... */
1355 spin_unlock(&hierarchy_id_lock
);
1360 static int cgroup_test_super(struct super_block
*sb
, void *data
)
1362 struct cgroup_sb_opts
*opts
= data
;
1363 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1365 /* If we asked for a name then it must match */
1366 if (opts
->name
&& strcmp(opts
->name
, root
->name
))
1370 * If we asked for subsystems (or explicitly for no
1371 * subsystems) then they must match
1373 if ((opts
->subsys_bits
|| opts
->none
)
1374 && (opts
->subsys_bits
!= root
->subsys_bits
))
1380 static struct cgroupfs_root
*cgroup_root_from_opts(struct cgroup_sb_opts
*opts
)
1382 struct cgroupfs_root
*root
;
1384 if (!opts
->subsys_bits
&& !opts
->none
)
1387 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1389 return ERR_PTR(-ENOMEM
);
1391 if (!init_root_id(root
)) {
1393 return ERR_PTR(-ENOMEM
);
1395 init_cgroup_root(root
);
1397 root
->subsys_bits
= opts
->subsys_bits
;
1398 root
->flags
= opts
->flags
;
1399 if (opts
->release_agent
)
1400 strcpy(root
->release_agent_path
, opts
->release_agent
);
1402 strcpy(root
->name
, opts
->name
);
1403 if (opts
->clone_children
)
1404 set_bit(CGRP_CLONE_CHILDREN
, &root
->top_cgroup
.flags
);
1408 static void cgroup_drop_root(struct cgroupfs_root
*root
)
1413 BUG_ON(!root
->hierarchy_id
);
1414 spin_lock(&hierarchy_id_lock
);
1415 ida_remove(&hierarchy_ida
, root
->hierarchy_id
);
1416 spin_unlock(&hierarchy_id_lock
);
1420 static int cgroup_set_super(struct super_block
*sb
, void *data
)
1423 struct cgroup_sb_opts
*opts
= data
;
1425 /* If we don't have a new root, we can't set up a new sb */
1426 if (!opts
->new_root
)
1429 BUG_ON(!opts
->subsys_bits
&& !opts
->none
);
1431 ret
= set_anon_super(sb
, NULL
);
1435 sb
->s_fs_info
= opts
->new_root
;
1436 opts
->new_root
->sb
= sb
;
1438 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
1439 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
1440 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
1441 sb
->s_op
= &cgroup_ops
;
1446 static int cgroup_get_rootdir(struct super_block
*sb
)
1448 static const struct dentry_operations cgroup_dops
= {
1449 .d_iput
= cgroup_diput
,
1450 .d_delete
= cgroup_delete
,
1453 struct inode
*inode
=
1454 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
1455 struct dentry
*dentry
;
1460 inode
->i_fop
= &simple_dir_operations
;
1461 inode
->i_op
= &cgroup_dir_inode_operations
;
1462 /* directories start off with i_nlink == 2 (for "." entry) */
1464 dentry
= d_alloc_root(inode
);
1469 sb
->s_root
= dentry
;
1470 /* for everything else we want ->d_op set */
1471 sb
->s_d_op
= &cgroup_dops
;
1475 static struct dentry
*cgroup_mount(struct file_system_type
*fs_type
,
1476 int flags
, const char *unused_dev_name
,
1479 struct cgroup_sb_opts opts
;
1480 struct cgroupfs_root
*root
;
1482 struct super_block
*sb
;
1483 struct cgroupfs_root
*new_root
;
1485 /* First find the desired set of subsystems */
1486 mutex_lock(&cgroup_mutex
);
1487 ret
= parse_cgroupfs_options(data
, &opts
);
1488 mutex_unlock(&cgroup_mutex
);
1493 * Allocate a new cgroup root. We may not need it if we're
1494 * reusing an existing hierarchy.
1496 new_root
= cgroup_root_from_opts(&opts
);
1497 if (IS_ERR(new_root
)) {
1498 ret
= PTR_ERR(new_root
);
1501 opts
.new_root
= new_root
;
1503 /* Locate an existing or new sb for this hierarchy */
1504 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, &opts
);
1507 cgroup_drop_root(opts
.new_root
);
1511 root
= sb
->s_fs_info
;
1513 if (root
== opts
.new_root
) {
1514 /* We used the new root structure, so this is a new hierarchy */
1515 struct list_head tmp_cg_links
;
1516 struct cgroup
*root_cgrp
= &root
->top_cgroup
;
1517 struct inode
*inode
;
1518 struct cgroupfs_root
*existing_root
;
1519 const struct cred
*cred
;
1522 BUG_ON(sb
->s_root
!= NULL
);
1524 ret
= cgroup_get_rootdir(sb
);
1526 goto drop_new_super
;
1527 inode
= sb
->s_root
->d_inode
;
1529 mutex_lock(&inode
->i_mutex
);
1530 mutex_lock(&cgroup_mutex
);
1532 if (strlen(root
->name
)) {
1533 /* Check for name clashes with existing mounts */
1534 for_each_active_root(existing_root
) {
1535 if (!strcmp(existing_root
->name
, root
->name
)) {
1537 mutex_unlock(&cgroup_mutex
);
1538 mutex_unlock(&inode
->i_mutex
);
1539 goto drop_new_super
;
1545 * We're accessing css_set_count without locking
1546 * css_set_lock here, but that's OK - it can only be
1547 * increased by someone holding cgroup_lock, and
1548 * that's us. The worst that can happen is that we
1549 * have some link structures left over
1551 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1553 mutex_unlock(&cgroup_mutex
);
1554 mutex_unlock(&inode
->i_mutex
);
1555 goto drop_new_super
;
1558 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1559 if (ret
== -EBUSY
) {
1560 mutex_unlock(&cgroup_mutex
);
1561 mutex_unlock(&inode
->i_mutex
);
1562 free_cg_links(&tmp_cg_links
);
1563 goto drop_new_super
;
1566 * There must be no failure case after here, since rebinding
1567 * takes care of subsystems' refcounts, which are explicitly
1568 * dropped in the failure exit path.
1571 /* EBUSY should be the only error here */
1574 list_add(&root
->root_list
, &roots
);
1577 sb
->s_root
->d_fsdata
= root_cgrp
;
1578 root
->top_cgroup
.dentry
= sb
->s_root
;
1580 /* Link the top cgroup in this hierarchy into all
1581 * the css_set objects */
1582 write_lock(&css_set_lock
);
1583 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1584 struct hlist_head
*hhead
= &css_set_table
[i
];
1585 struct hlist_node
*node
;
1588 hlist_for_each_entry(cg
, node
, hhead
, hlist
)
1589 link_css_set(&tmp_cg_links
, cg
, root_cgrp
);
1591 write_unlock(&css_set_lock
);
1593 free_cg_links(&tmp_cg_links
);
1595 BUG_ON(!list_empty(&root_cgrp
->sibling
));
1596 BUG_ON(!list_empty(&root_cgrp
->children
));
1597 BUG_ON(root
->number_of_cgroups
!= 1);
1599 cred
= override_creds(&init_cred
);
1600 cgroup_populate_dir(root_cgrp
);
1602 mutex_unlock(&cgroup_mutex
);
1603 mutex_unlock(&inode
->i_mutex
);
1606 * We re-used an existing hierarchy - the new root (if
1607 * any) is not needed
1609 cgroup_drop_root(opts
.new_root
);
1610 /* no subsys rebinding, so refcounts don't change */
1611 drop_parsed_module_refcounts(opts
.subsys_bits
);
1614 kfree(opts
.release_agent
);
1616 return dget(sb
->s_root
);
1619 deactivate_locked_super(sb
);
1621 drop_parsed_module_refcounts(opts
.subsys_bits
);
1623 kfree(opts
.release_agent
);
1625 return ERR_PTR(ret
);
1628 static void cgroup_kill_sb(struct super_block
*sb
) {
1629 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1630 struct cgroup
*cgrp
= &root
->top_cgroup
;
1632 struct cg_cgroup_link
*link
;
1633 struct cg_cgroup_link
*saved_link
;
1637 BUG_ON(root
->number_of_cgroups
!= 1);
1638 BUG_ON(!list_empty(&cgrp
->children
));
1639 BUG_ON(!list_empty(&cgrp
->sibling
));
1641 mutex_lock(&cgroup_mutex
);
1643 /* Rebind all subsystems back to the default hierarchy */
1644 ret
= rebind_subsystems(root
, 0);
1645 /* Shouldn't be able to fail ... */
1649 * Release all the links from css_sets to this hierarchy's
1652 write_lock(&css_set_lock
);
1654 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1656 list_del(&link
->cg_link_list
);
1657 list_del(&link
->cgrp_link_list
);
1660 write_unlock(&css_set_lock
);
1662 if (!list_empty(&root
->root_list
)) {
1663 list_del(&root
->root_list
);
1667 mutex_unlock(&cgroup_mutex
);
1669 kill_litter_super(sb
);
1670 cgroup_drop_root(root
);
1673 static struct file_system_type cgroup_fs_type
= {
1675 .mount
= cgroup_mount
,
1676 .kill_sb
= cgroup_kill_sb
,
1679 static struct kobject
*cgroup_kobj
;
1681 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1683 return dentry
->d_fsdata
;
1686 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1688 return dentry
->d_fsdata
;
1692 * cgroup_path - generate the path of a cgroup
1693 * @cgrp: the cgroup in question
1694 * @buf: the buffer to write the path into
1695 * @buflen: the length of the buffer
1697 * Called with cgroup_mutex held or else with an RCU-protected cgroup
1698 * reference. Writes path of cgroup into buf. Returns 0 on success,
1701 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1704 struct dentry
*dentry
= rcu_dereference_check(cgrp
->dentry
,
1705 rcu_read_lock_held() ||
1706 cgroup_lock_is_held());
1708 if (!dentry
|| cgrp
== dummytop
) {
1710 * Inactive subsystems have no dentry for their root
1717 start
= buf
+ buflen
;
1721 int len
= dentry
->d_name
.len
;
1723 if ((start
-= len
) < buf
)
1724 return -ENAMETOOLONG
;
1725 memcpy(start
, dentry
->d_name
.name
, len
);
1726 cgrp
= cgrp
->parent
;
1730 dentry
= rcu_dereference_check(cgrp
->dentry
,
1731 rcu_read_lock_held() ||
1732 cgroup_lock_is_held());
1736 return -ENAMETOOLONG
;
1739 memmove(buf
, start
, buf
+ buflen
- start
);
1742 EXPORT_SYMBOL_GPL(cgroup_path
);
1745 * cgroup_task_migrate - move a task from one cgroup to another.
1747 * 'guarantee' is set if the caller promises that a new css_set for the task
1748 * will already exist. If not set, this function might sleep, and can fail with
1749 * -ENOMEM. Otherwise, it can only fail with -ESRCH.
1751 static int cgroup_task_migrate(struct cgroup
*cgrp
, struct cgroup
*oldcgrp
,
1752 struct task_struct
*tsk
, bool guarantee
)
1754 struct css_set
*oldcg
;
1755 struct css_set
*newcg
;
1758 * get old css_set. we need to take task_lock and refcount it, because
1759 * an exiting task can change its css_set to init_css_set and drop its
1760 * old one without taking cgroup_mutex.
1763 oldcg
= tsk
->cgroups
;
1767 /* locate or allocate a new css_set for this task. */
1769 /* we know the css_set we want already exists. */
1770 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
1771 read_lock(&css_set_lock
);
1772 newcg
= find_existing_css_set(oldcg
, cgrp
, template);
1775 read_unlock(&css_set_lock
);
1778 /* find_css_set will give us newcg already referenced. */
1779 newcg
= find_css_set(oldcg
, cgrp
);
1787 /* if PF_EXITING is set, the tsk->cgroups pointer is no longer safe. */
1789 if (tsk
->flags
& PF_EXITING
) {
1794 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1797 /* Update the css_set linked lists if we're using them */
1798 write_lock(&css_set_lock
);
1799 if (!list_empty(&tsk
->cg_list
))
1800 list_move(&tsk
->cg_list
, &newcg
->tasks
);
1801 write_unlock(&css_set_lock
);
1804 * We just gained a reference on oldcg by taking it from the task. As
1805 * trading it for newcg is protected by cgroup_mutex, we're safe to drop
1806 * it here; it will be freed under RCU.
1810 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1815 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1816 * @cgrp: the cgroup the task is attaching to
1817 * @tsk: the task to be attached
1819 * Call holding cgroup_mutex. May take task_lock of
1820 * the task 'tsk' during call.
1822 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1825 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1826 struct cgroup
*oldcgrp
;
1827 struct cgroupfs_root
*root
= cgrp
->root
;
1829 /* Nothing to do if the task is already in that cgroup */
1830 oldcgrp
= task_cgroup_from_root(tsk
, root
);
1831 if (cgrp
== oldcgrp
)
1834 for_each_subsys(root
, ss
) {
1835 if (ss
->can_attach
) {
1836 retval
= ss
->can_attach(ss
, cgrp
, tsk
);
1839 * Remember on which subsystem the can_attach()
1840 * failed, so that we only call cancel_attach()
1841 * against the subsystems whose can_attach()
1842 * succeeded. (See below)
1848 if (ss
->can_attach_task
) {
1849 retval
= ss
->can_attach_task(cgrp
, tsk
);
1857 retval
= cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, false);
1861 for_each_subsys(root
, ss
) {
1863 ss
->pre_attach(cgrp
);
1864 if (ss
->attach_task
)
1865 ss
->attach_task(cgrp
, tsk
);
1867 ss
->attach(ss
, cgrp
, oldcgrp
, tsk
);
1873 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1874 * is no longer empty.
1876 cgroup_wakeup_rmdir_waiter(cgrp
);
1879 for_each_subsys(root
, ss
) {
1880 if (ss
== failed_ss
)
1882 * This subsystem was the one that failed the
1883 * can_attach() check earlier, so we don't need
1884 * to call cancel_attach() against it or any
1885 * remaining subsystems.
1888 if (ss
->cancel_attach
)
1889 ss
->cancel_attach(ss
, cgrp
, tsk
);
1896 * cgroup_attach_task_all - attach task 'tsk' to all cgroups of task 'from'
1897 * @from: attach to all cgroups of a given task
1898 * @tsk: the task to be attached
1900 int cgroup_attach_task_all(struct task_struct
*from
, struct task_struct
*tsk
)
1902 struct cgroupfs_root
*root
;
1906 for_each_active_root(root
) {
1907 struct cgroup
*from_cg
= task_cgroup_from_root(from
, root
);
1909 retval
= cgroup_attach_task(from_cg
, tsk
);
1917 EXPORT_SYMBOL_GPL(cgroup_attach_task_all
);
1920 * cgroup_attach_proc works in two stages, the first of which prefetches all
1921 * new css_sets needed (to make sure we have enough memory before committing
1922 * to the move) and stores them in a list of entries of the following type.
1923 * TODO: possible optimization: use css_set->rcu_head for chaining instead
1925 struct cg_list_entry
{
1927 struct list_head links
;
1930 static bool css_set_check_fetched(struct cgroup
*cgrp
,
1931 struct task_struct
*tsk
, struct css_set
*cg
,
1932 struct list_head
*newcg_list
)
1934 struct css_set
*newcg
;
1935 struct cg_list_entry
*cg_entry
;
1936 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
1938 read_lock(&css_set_lock
);
1939 newcg
= find_existing_css_set(cg
, cgrp
, template);
1942 read_unlock(&css_set_lock
);
1944 /* doesn't exist at all? */
1947 /* see if it's already in the list */
1948 list_for_each_entry(cg_entry
, newcg_list
, links
) {
1949 if (cg_entry
->cg
== newcg
) {
1961 * Find the new css_set and store it in the list in preparation for moving the
1962 * given task to the given cgroup. Returns 0 or -ENOMEM.
1964 static int css_set_prefetch(struct cgroup
*cgrp
, struct css_set
*cg
,
1965 struct list_head
*newcg_list
)
1967 struct css_set
*newcg
;
1968 struct cg_list_entry
*cg_entry
;
1970 /* ensure a new css_set will exist for this thread */
1971 newcg
= find_css_set(cg
, cgrp
);
1974 /* add it to the list */
1975 cg_entry
= kmalloc(sizeof(struct cg_list_entry
), GFP_KERNEL
);
1980 cg_entry
->cg
= newcg
;
1981 list_add(&cg_entry
->links
, newcg_list
);
1986 * cgroup_attach_proc - attach all threads in a threadgroup to a cgroup
1987 * @cgrp: the cgroup to attach to
1988 * @leader: the threadgroup leader task_struct of the group to be attached
1990 * Call holding cgroup_mutex and the threadgroup_fork_lock of the leader. Will
1991 * take task_lock of each thread in leader's threadgroup individually in turn.
1993 int cgroup_attach_proc(struct cgroup
*cgrp
, struct task_struct
*leader
)
1995 int retval
, i
, group_size
;
1996 struct cgroup_subsys
*ss
, *failed_ss
= NULL
;
1997 bool cancel_failed_ss
= false;
1998 /* guaranteed to be initialized later, but the compiler needs this */
1999 struct cgroup
*oldcgrp
= NULL
;
2000 struct css_set
*oldcg
;
2001 struct cgroupfs_root
*root
= cgrp
->root
;
2002 /* threadgroup list cursor and array */
2003 struct task_struct
*tsk
;
2004 struct flex_array
*group
;
2006 * we need to make sure we have css_sets for all the tasks we're
2007 * going to move -before- we actually start moving them, so that in
2008 * case we get an ENOMEM we can bail out before making any changes.
2010 struct list_head newcg_list
;
2011 struct cg_list_entry
*cg_entry
, *temp_nobe
;
2014 * step 0: in order to do expensive, possibly blocking operations for
2015 * every thread, we cannot iterate the thread group list, since it needs
2016 * rcu or tasklist locked. instead, build an array of all threads in the
2017 * group - threadgroup_fork_lock prevents new threads from appearing,
2018 * and if threads exit, this will just be an over-estimate.
2020 group_size
= get_nr_threads(leader
);
2021 /* flex_array supports very large thread-groups better than kmalloc. */
2022 group
= flex_array_alloc(sizeof(struct task_struct
*), group_size
,
2026 /* pre-allocate to guarantee space while iterating in rcu read-side. */
2027 retval
= flex_array_prealloc(group
, 0, group_size
- 1, GFP_KERNEL
);
2029 goto out_free_group_list
;
2031 /* prevent changes to the threadgroup list while we take a snapshot. */
2033 if (!thread_group_leader(leader
)) {
2035 * a race with de_thread from another thread's exec() may strip
2036 * us of our leadership, making while_each_thread unsafe to use
2037 * on this task. if this happens, there is no choice but to
2038 * throw this task away and try again (from cgroup_procs_write);
2039 * this is "double-double-toil-and-trouble-check locking".
2043 goto out_free_group_list
;
2045 /* take a reference on each task in the group to go in the array. */
2049 /* as per above, nr_threads may decrease, but not increase. */
2050 BUG_ON(i
>= group_size
);
2051 get_task_struct(tsk
);
2053 * saying GFP_ATOMIC has no effect here because we did prealloc
2054 * earlier, but it's good form to communicate our expectations.
2056 retval
= flex_array_put_ptr(group
, i
, tsk
, GFP_ATOMIC
);
2057 BUG_ON(retval
!= 0);
2059 } while_each_thread(leader
, tsk
);
2060 /* remember the number of threads in the array for later. */
2065 * step 1: check that we can legitimately attach to the cgroup.
2067 for_each_subsys(root
, ss
) {
2068 if (ss
->can_attach
) {
2069 retval
= ss
->can_attach(ss
, cgrp
, leader
);
2072 goto out_cancel_attach
;
2075 /* a callback to be run on every thread in the threadgroup. */
2076 if (ss
->can_attach_task
) {
2077 /* run on each task in the threadgroup. */
2078 for (i
= 0; i
< group_size
; i
++) {
2079 tsk
= flex_array_get_ptr(group
, i
);
2080 retval
= ss
->can_attach_task(cgrp
, tsk
);
2083 cancel_failed_ss
= true;
2084 goto out_cancel_attach
;
2091 * step 2: make sure css_sets exist for all threads to be migrated.
2092 * we use find_css_set, which allocates a new one if necessary.
2094 INIT_LIST_HEAD(&newcg_list
);
2095 for (i
= 0; i
< group_size
; i
++) {
2096 tsk
= flex_array_get_ptr(group
, i
);
2097 /* nothing to do if this task is already in the cgroup */
2098 oldcgrp
= task_cgroup_from_root(tsk
, root
);
2099 if (cgrp
== oldcgrp
)
2101 /* get old css_set pointer */
2103 if (tsk
->flags
& PF_EXITING
) {
2104 /* ignore this task if it's going away */
2108 oldcg
= tsk
->cgroups
;
2111 /* see if the new one for us is already in the list? */
2112 if (css_set_check_fetched(cgrp
, tsk
, oldcg
, &newcg_list
)) {
2113 /* was already there, nothing to do. */
2116 /* we don't already have it. get new one. */
2117 retval
= css_set_prefetch(cgrp
, oldcg
, &newcg_list
);
2120 goto out_list_teardown
;
2125 * step 3: now that we're guaranteed success wrt the css_sets, proceed
2126 * to move all tasks to the new cgroup, calling ss->attach_task for each
2127 * one along the way. there are no failure cases after here, so this is
2130 for_each_subsys(root
, ss
) {
2132 ss
->pre_attach(cgrp
);
2134 for (i
= 0; i
< group_size
; i
++) {
2135 tsk
= flex_array_get_ptr(group
, i
);
2136 /* leave current thread as it is if it's already there */
2137 oldcgrp
= task_cgroup_from_root(tsk
, root
);
2138 if (cgrp
== oldcgrp
)
2140 /* attach each task to each subsystem */
2141 for_each_subsys(root
, ss
) {
2142 if (ss
->attach_task
)
2143 ss
->attach_task(cgrp
, tsk
);
2145 /* if the thread is PF_EXITING, it can just get skipped. */
2146 retval
= cgroup_task_migrate(cgrp
, oldcgrp
, tsk
, true);
2147 BUG_ON(retval
!= 0 && retval
!= -ESRCH
);
2149 /* nothing is sensitive to fork() after this point. */
2152 * step 4: do expensive, non-thread-specific subsystem callbacks.
2153 * TODO: if ever a subsystem needs to know the oldcgrp for each task
2154 * being moved, this call will need to be reworked to communicate that.
2156 for_each_subsys(root
, ss
) {
2158 ss
->attach(ss
, cgrp
, oldcgrp
, leader
);
2162 * step 5: success! and cleanup
2165 cgroup_wakeup_rmdir_waiter(cgrp
);
2168 /* clean up the list of prefetched css_sets. */
2169 list_for_each_entry_safe(cg_entry
, temp_nobe
, &newcg_list
, links
) {
2170 list_del(&cg_entry
->links
);
2171 put_css_set(cg_entry
->cg
);
2175 /* same deal as in cgroup_attach_task */
2177 for_each_subsys(root
, ss
) {
2178 if (ss
== failed_ss
) {
2179 if (cancel_failed_ss
&& ss
->cancel_attach
)
2180 ss
->cancel_attach(ss
, cgrp
, leader
);
2183 if (ss
->cancel_attach
)
2184 ss
->cancel_attach(ss
, cgrp
, leader
);
2187 /* clean up the array of referenced threads in the group. */
2188 for (i
= 0; i
< group_size
; i
++) {
2189 tsk
= flex_array_get_ptr(group
, i
);
2190 put_task_struct(tsk
);
2192 out_free_group_list
:
2193 flex_array_free(group
);
2198 * Find the task_struct of the task to attach by vpid and pass it along to the
2199 * function to attach either it or all tasks in its threadgroup. Will take
2200 * cgroup_mutex; may take task_lock of task.
2202 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
, bool threadgroup
)
2204 struct task_struct
*tsk
;
2205 const struct cred
*cred
= current_cred(), *tcred
;
2208 if (!cgroup_lock_live_group(cgrp
))
2213 tsk
= find_task_by_vpid(pid
);
2221 * RCU protects this access, since tsk was found in the
2222 * tid map. a race with de_thread may cause group_leader
2223 * to stop being the leader, but cgroup_attach_proc will
2226 tsk
= tsk
->group_leader
;
2227 } else if (tsk
->flags
& PF_EXITING
) {
2228 /* optimization for the single-task-only case */
2235 * even if we're attaching all tasks in the thread group, we
2236 * only need to check permissions on one of them.
2238 tcred
= __task_cred(tsk
);
2240 cred
->euid
!= tcred
->uid
&&
2241 cred
->euid
!= tcred
->suid
) {
2246 get_task_struct(tsk
);
2250 tsk
= current
->group_leader
;
2253 get_task_struct(tsk
);
2257 threadgroup_fork_write_lock(tsk
);
2258 ret
= cgroup_attach_proc(cgrp
, tsk
);
2259 threadgroup_fork_write_unlock(tsk
);
2261 ret
= cgroup_attach_task(cgrp
, tsk
);
2263 put_task_struct(tsk
);
2268 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
2270 return attach_task_by_pid(cgrp
, pid
, false);
2273 static int cgroup_procs_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 tgid
)
2278 * attach_proc fails with -EAGAIN if threadgroup leadership
2279 * changes in the middle of the operation, in which case we need
2280 * to find the task_struct for the new leader and start over.
2282 ret
= attach_task_by_pid(cgrp
, tgid
, true);
2283 } while (ret
== -EAGAIN
);
2288 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
2289 * @cgrp: the cgroup to be checked for liveness
2291 * On success, returns true; the lock should be later released with
2292 * cgroup_unlock(). On failure returns false with no lock held.
2294 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
2296 mutex_lock(&cgroup_mutex
);
2297 if (cgroup_is_removed(cgrp
)) {
2298 mutex_unlock(&cgroup_mutex
);
2303 EXPORT_SYMBOL_GPL(cgroup_lock_live_group
);
2305 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
2308 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
2309 if (strlen(buffer
) >= PATH_MAX
)
2311 if (!cgroup_lock_live_group(cgrp
))
2313 strcpy(cgrp
->root
->release_agent_path
, buffer
);
2318 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
2319 struct seq_file
*seq
)
2321 if (!cgroup_lock_live_group(cgrp
))
2323 seq_puts(seq
, cgrp
->root
->release_agent_path
);
2324 seq_putc(seq
, '\n');
2329 /* A buffer size big enough for numbers or short strings */
2330 #define CGROUP_LOCAL_BUFFER_SIZE 64
2332 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
2334 const char __user
*userbuf
,
2335 size_t nbytes
, loff_t
*unused_ppos
)
2337 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2343 if (nbytes
>= sizeof(buffer
))
2345 if (copy_from_user(buffer
, userbuf
, nbytes
))
2348 buffer
[nbytes
] = 0; /* nul-terminate */
2349 if (cft
->write_u64
) {
2350 u64 val
= simple_strtoull(strstrip(buffer
), &end
, 0);
2353 retval
= cft
->write_u64(cgrp
, cft
, val
);
2355 s64 val
= simple_strtoll(strstrip(buffer
), &end
, 0);
2358 retval
= cft
->write_s64(cgrp
, cft
, val
);
2365 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
2367 const char __user
*userbuf
,
2368 size_t nbytes
, loff_t
*unused_ppos
)
2370 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
2372 size_t max_bytes
= cft
->max_write_len
;
2373 char *buffer
= local_buffer
;
2376 max_bytes
= sizeof(local_buffer
) - 1;
2377 if (nbytes
>= max_bytes
)
2379 /* Allocate a dynamic buffer if we need one */
2380 if (nbytes
>= sizeof(local_buffer
)) {
2381 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
2385 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
)) {
2390 buffer
[nbytes
] = 0; /* nul-terminate */
2391 retval
= cft
->write_string(cgrp
, cft
, strstrip(buffer
));
2395 if (buffer
!= local_buffer
)
2400 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
2401 size_t nbytes
, loff_t
*ppos
)
2403 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2404 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2406 if (cgroup_is_removed(cgrp
))
2409 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2410 if (cft
->write_u64
|| cft
->write_s64
)
2411 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2412 if (cft
->write_string
)
2413 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2415 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
2416 return ret
? ret
: nbytes
;
2421 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
2423 char __user
*buf
, size_t nbytes
,
2426 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2427 u64 val
= cft
->read_u64(cgrp
, cft
);
2428 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
2430 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2433 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
2435 char __user
*buf
, size_t nbytes
,
2438 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
2439 s64 val
= cft
->read_s64(cgrp
, cft
);
2440 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
2442 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
2445 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
2446 size_t nbytes
, loff_t
*ppos
)
2448 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2449 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2451 if (cgroup_is_removed(cgrp
))
2455 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2457 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2459 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
2464 * seqfile ops/methods for returning structured data. Currently just
2465 * supports string->u64 maps, but can be extended in future.
2468 struct cgroup_seqfile_state
{
2470 struct cgroup
*cgroup
;
2473 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
2475 struct seq_file
*sf
= cb
->state
;
2476 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
2479 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
2481 struct cgroup_seqfile_state
*state
= m
->private;
2482 struct cftype
*cft
= state
->cft
;
2483 if (cft
->read_map
) {
2484 struct cgroup_map_cb cb
= {
2485 .fill
= cgroup_map_add
,
2488 return cft
->read_map(state
->cgroup
, cft
, &cb
);
2490 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
2493 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
2495 struct seq_file
*seq
= file
->private_data
;
2496 kfree(seq
->private);
2497 return single_release(inode
, file
);
2500 static const struct file_operations cgroup_seqfile_operations
= {
2502 .write
= cgroup_file_write
,
2503 .llseek
= seq_lseek
,
2504 .release
= cgroup_seqfile_release
,
2507 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
2512 err
= generic_file_open(inode
, file
);
2515 cft
= __d_cft(file
->f_dentry
);
2517 if (cft
->read_map
|| cft
->read_seq_string
) {
2518 struct cgroup_seqfile_state
*state
=
2519 kzalloc(sizeof(*state
), GFP_USER
);
2523 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
2524 file
->f_op
= &cgroup_seqfile_operations
;
2525 err
= single_open(file
, cgroup_seqfile_show
, state
);
2528 } else if (cft
->open
)
2529 err
= cft
->open(inode
, file
);
2536 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
2538 struct cftype
*cft
= __d_cft(file
->f_dentry
);
2540 return cft
->release(inode
, file
);
2545 * cgroup_rename - Only allow simple rename of directories in place.
2547 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
2548 struct inode
*new_dir
, struct dentry
*new_dentry
)
2550 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
2552 if (new_dentry
->d_inode
)
2554 if (old_dir
!= new_dir
)
2556 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
2559 static const struct file_operations cgroup_file_operations
= {
2560 .read
= cgroup_file_read
,
2561 .write
= cgroup_file_write
,
2562 .llseek
= generic_file_llseek
,
2563 .open
= cgroup_file_open
,
2564 .release
= cgroup_file_release
,
2567 static const struct inode_operations cgroup_dir_inode_operations
= {
2568 .lookup
= cgroup_lookup
,
2569 .mkdir
= cgroup_mkdir
,
2570 .rmdir
= cgroup_rmdir
,
2571 .rename
= cgroup_rename
,
2574 static struct dentry
*cgroup_lookup(struct inode
*dir
, struct dentry
*dentry
, struct nameidata
*nd
)
2576 if (dentry
->d_name
.len
> NAME_MAX
)
2577 return ERR_PTR(-ENAMETOOLONG
);
2578 d_add(dentry
, NULL
);
2583 * Check if a file is a control file
2585 static inline struct cftype
*__file_cft(struct file
*file
)
2587 if (file
->f_dentry
->d_inode
->i_fop
!= &cgroup_file_operations
)
2588 return ERR_PTR(-EINVAL
);
2589 return __d_cft(file
->f_dentry
);
2592 static int cgroup_create_file(struct dentry
*dentry
, mode_t mode
,
2593 struct super_block
*sb
)
2595 struct inode
*inode
;
2599 if (dentry
->d_inode
)
2602 inode
= cgroup_new_inode(mode
, sb
);
2606 if (S_ISDIR(mode
)) {
2607 inode
->i_op
= &cgroup_dir_inode_operations
;
2608 inode
->i_fop
= &simple_dir_operations
;
2610 /* start off with i_nlink == 2 (for "." entry) */
2613 /* start with the directory inode held, so that we can
2614 * populate it without racing with another mkdir */
2615 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
2616 } else if (S_ISREG(mode
)) {
2618 inode
->i_fop
= &cgroup_file_operations
;
2620 d_instantiate(dentry
, inode
);
2621 dget(dentry
); /* Extra count - pin the dentry in core */
2626 * cgroup_create_dir - create a directory for an object.
2627 * @cgrp: the cgroup we create the directory for. It must have a valid
2628 * ->parent field. And we are going to fill its ->dentry field.
2629 * @dentry: dentry of the new cgroup
2630 * @mode: mode to set on new directory.
2632 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
2635 struct dentry
*parent
;
2638 parent
= cgrp
->parent
->dentry
;
2639 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
2641 dentry
->d_fsdata
= cgrp
;
2642 inc_nlink(parent
->d_inode
);
2643 rcu_assign_pointer(cgrp
->dentry
, dentry
);
2652 * cgroup_file_mode - deduce file mode of a control file
2653 * @cft: the control file in question
2655 * returns cft->mode if ->mode is not 0
2656 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
2657 * returns S_IRUGO if it has only a read handler
2658 * returns S_IWUSR if it has only a write hander
2660 static mode_t
cgroup_file_mode(const struct cftype
*cft
)
2667 if (cft
->read
|| cft
->read_u64
|| cft
->read_s64
||
2668 cft
->read_map
|| cft
->read_seq_string
)
2671 if (cft
->write
|| cft
->write_u64
|| cft
->write_s64
||
2672 cft
->write_string
|| cft
->trigger
)
2678 int cgroup_add_file(struct cgroup
*cgrp
,
2679 struct cgroup_subsys
*subsys
,
2680 const struct cftype
*cft
)
2682 struct dentry
*dir
= cgrp
->dentry
;
2683 struct dentry
*dentry
;
2687 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
2688 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
2689 strcpy(name
, subsys
->name
);
2692 strcat(name
, cft
->name
);
2693 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
2694 dentry
= lookup_one_len(name
, dir
, strlen(name
));
2695 if (!IS_ERR(dentry
)) {
2696 mode
= cgroup_file_mode(cft
);
2697 error
= cgroup_create_file(dentry
, mode
| S_IFREG
,
2700 dentry
->d_fsdata
= (void *)cft
;
2703 error
= PTR_ERR(dentry
);
2706 EXPORT_SYMBOL_GPL(cgroup_add_file
);
2708 int cgroup_add_files(struct cgroup
*cgrp
,
2709 struct cgroup_subsys
*subsys
,
2710 const struct cftype cft
[],
2714 for (i
= 0; i
< count
; i
++) {
2715 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
2721 EXPORT_SYMBOL_GPL(cgroup_add_files
);
2724 * cgroup_task_count - count the number of tasks in a cgroup.
2725 * @cgrp: the cgroup in question
2727 * Return the number of tasks in the cgroup.
2729 int cgroup_task_count(const struct cgroup
*cgrp
)
2732 struct cg_cgroup_link
*link
;
2734 read_lock(&css_set_lock
);
2735 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
2736 count
+= atomic_read(&link
->cg
->refcount
);
2738 read_unlock(&css_set_lock
);
2743 * Advance a list_head iterator. The iterator should be positioned at
2744 * the start of a css_set
2746 static void cgroup_advance_iter(struct cgroup
*cgrp
,
2747 struct cgroup_iter
*it
)
2749 struct list_head
*l
= it
->cg_link
;
2750 struct cg_cgroup_link
*link
;
2753 /* Advance to the next non-empty css_set */
2756 if (l
== &cgrp
->css_sets
) {
2760 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
2762 } while (list_empty(&cg
->tasks
));
2764 it
->task
= cg
->tasks
.next
;
2768 * To reduce the fork() overhead for systems that are not actually
2769 * using their cgroups capability, we don't maintain the lists running
2770 * through each css_set to its tasks until we see the list actually
2771 * used - in other words after the first call to cgroup_iter_start().
2773 * The tasklist_lock is not held here, as do_each_thread() and
2774 * while_each_thread() are protected by RCU.
2776 static void cgroup_enable_task_cg_lists(void)
2778 struct task_struct
*p
, *g
;
2779 write_lock(&css_set_lock
);
2780 use_task_css_set_links
= 1;
2781 do_each_thread(g
, p
) {
2784 * We should check if the process is exiting, otherwise
2785 * it will race with cgroup_exit() in that the list
2786 * entry won't be deleted though the process has exited.
2788 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
2789 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
2791 } while_each_thread(g
, p
);
2792 write_unlock(&css_set_lock
);
2795 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2798 * The first time anyone tries to iterate across a cgroup,
2799 * we need to enable the list linking each css_set to its
2800 * tasks, and fix up all existing tasks.
2802 if (!use_task_css_set_links
)
2803 cgroup_enable_task_cg_lists();
2805 read_lock(&css_set_lock
);
2806 it
->cg_link
= &cgrp
->css_sets
;
2807 cgroup_advance_iter(cgrp
, it
);
2810 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
2811 struct cgroup_iter
*it
)
2813 struct task_struct
*res
;
2814 struct list_head
*l
= it
->task
;
2815 struct cg_cgroup_link
*link
;
2817 /* If the iterator cg is NULL, we have no tasks */
2820 res
= list_entry(l
, struct task_struct
, cg_list
);
2821 /* Advance iterator to find next entry */
2823 link
= list_entry(it
->cg_link
, struct cg_cgroup_link
, cgrp_link_list
);
2824 if (l
== &link
->cg
->tasks
) {
2825 /* We reached the end of this task list - move on to
2826 * the next cg_cgroup_link */
2827 cgroup_advance_iter(cgrp
, it
);
2834 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
2836 read_unlock(&css_set_lock
);
2839 static inline int started_after_time(struct task_struct
*t1
,
2840 struct timespec
*time
,
2841 struct task_struct
*t2
)
2843 int start_diff
= timespec_compare(&t1
->start_time
, time
);
2844 if (start_diff
> 0) {
2846 } else if (start_diff
< 0) {
2850 * Arbitrarily, if two processes started at the same
2851 * time, we'll say that the lower pointer value
2852 * started first. Note that t2 may have exited by now
2853 * so this may not be a valid pointer any longer, but
2854 * that's fine - it still serves to distinguish
2855 * between two tasks started (effectively) simultaneously.
2862 * This function is a callback from heap_insert() and is used to order
2864 * In this case we order the heap in descending task start time.
2866 static inline int started_after(void *p1
, void *p2
)
2868 struct task_struct
*t1
= p1
;
2869 struct task_struct
*t2
= p2
;
2870 return started_after_time(t1
, &t2
->start_time
, t2
);
2874 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
2875 * @scan: struct cgroup_scanner containing arguments for the scan
2877 * Arguments include pointers to callback functions test_task() and
2879 * Iterate through all the tasks in a cgroup, calling test_task() for each,
2880 * and if it returns true, call process_task() for it also.
2881 * The test_task pointer may be NULL, meaning always true (select all tasks).
2882 * Effectively duplicates cgroup_iter_{start,next,end}()
2883 * but does not lock css_set_lock for the call to process_task().
2884 * The struct cgroup_scanner may be embedded in any structure of the caller's
2886 * It is guaranteed that process_task() will act on every task that
2887 * is a member of the cgroup for the duration of this call. This
2888 * function may or may not call process_task() for tasks that exit
2889 * or move to a different cgroup during the call, or are forked or
2890 * move into the cgroup during the call.
2892 * Note that test_task() may be called with locks held, and may in some
2893 * situations be called multiple times for the same task, so it should
2895 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
2896 * pre-allocated and will be used for heap operations (and its "gt" member will
2897 * be overwritten), else a temporary heap will be used (allocation of which
2898 * may cause this function to fail).
2900 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
2903 struct cgroup_iter it
;
2904 struct task_struct
*p
, *dropped
;
2905 /* Never dereference latest_task, since it's not refcounted */
2906 struct task_struct
*latest_task
= NULL
;
2907 struct ptr_heap tmp_heap
;
2908 struct ptr_heap
*heap
;
2909 struct timespec latest_time
= { 0, 0 };
2912 /* The caller supplied our heap and pre-allocated its memory */
2914 heap
->gt
= &started_after
;
2916 /* We need to allocate our own heap memory */
2918 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
2920 /* cannot allocate the heap */
2926 * Scan tasks in the cgroup, using the scanner's "test_task" callback
2927 * to determine which are of interest, and using the scanner's
2928 * "process_task" callback to process any of them that need an update.
2929 * Since we don't want to hold any locks during the task updates,
2930 * gather tasks to be processed in a heap structure.
2931 * The heap is sorted by descending task start time.
2932 * If the statically-sized heap fills up, we overflow tasks that
2933 * started later, and in future iterations only consider tasks that
2934 * started after the latest task in the previous pass. This
2935 * guarantees forward progress and that we don't miss any tasks.
2938 cgroup_iter_start(scan
->cg
, &it
);
2939 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
2941 * Only affect tasks that qualify per the caller's callback,
2942 * if he provided one
2944 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
2947 * Only process tasks that started after the last task
2950 if (!started_after_time(p
, &latest_time
, latest_task
))
2952 dropped
= heap_insert(heap
, p
);
2953 if (dropped
== NULL
) {
2955 * The new task was inserted; the heap wasn't
2959 } else if (dropped
!= p
) {
2961 * The new task was inserted, and pushed out a
2965 put_task_struct(dropped
);
2968 * Else the new task was newer than anything already in
2969 * the heap and wasn't inserted
2972 cgroup_iter_end(scan
->cg
, &it
);
2975 for (i
= 0; i
< heap
->size
; i
++) {
2976 struct task_struct
*q
= heap
->ptrs
[i
];
2978 latest_time
= q
->start_time
;
2981 /* Process the task per the caller's callback */
2982 scan
->process_task(q
, scan
);
2986 * If we had to process any tasks at all, scan again
2987 * in case some of them were in the middle of forking
2988 * children that didn't get processed.
2989 * Not the most efficient way to do it, but it avoids
2990 * having to take callback_mutex in the fork path
2994 if (heap
== &tmp_heap
)
2995 heap_free(&tmp_heap
);
3000 * Stuff for reading the 'tasks'/'procs' files.
3002 * Reading this file can return large amounts of data if a cgroup has
3003 * *lots* of attached tasks. So it may need several calls to read(),
3004 * but we cannot guarantee that the information we produce is correct
3005 * unless we produce it entirely atomically.
3010 * The following two functions "fix" the issue where there are more pids
3011 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
3012 * TODO: replace with a kernel-wide solution to this problem
3014 #define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
3015 static void *pidlist_allocate(int count
)
3017 if (PIDLIST_TOO_LARGE(count
))
3018 return vmalloc(count
* sizeof(pid_t
));
3020 return kmalloc(count
* sizeof(pid_t
), GFP_KERNEL
);
3022 static void pidlist_free(void *p
)
3024 if (is_vmalloc_addr(p
))
3029 static void *pidlist_resize(void *p
, int newcount
)
3032 /* note: if new alloc fails, old p will still be valid either way */
3033 if (is_vmalloc_addr(p
)) {
3034 newlist
= vmalloc(newcount
* sizeof(pid_t
));
3037 memcpy(newlist
, p
, newcount
* sizeof(pid_t
));
3040 newlist
= krealloc(p
, newcount
* sizeof(pid_t
), GFP_KERNEL
);
3046 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
3047 * If the new stripped list is sufficiently smaller and there's enough memory
3048 * to allocate a new buffer, will let go of the unneeded memory. Returns the
3049 * number of unique elements.
3051 /* is the size difference enough that we should re-allocate the array? */
3052 #define PIDLIST_REALLOC_DIFFERENCE(old, new) ((old) - PAGE_SIZE >= (new))
3053 static int pidlist_uniq(pid_t
**p
, int length
)
3060 * we presume the 0th element is unique, so i starts at 1. trivial
3061 * edge cases first; no work needs to be done for either
3063 if (length
== 0 || length
== 1)
3065 /* src and dest walk down the list; dest counts unique elements */
3066 for (src
= 1; src
< length
; src
++) {
3067 /* find next unique element */
3068 while (list
[src
] == list
[src
-1]) {
3073 /* dest always points to where the next unique element goes */
3074 list
[dest
] = list
[src
];
3079 * if the length difference is large enough, we want to allocate a
3080 * smaller buffer to save memory. if this fails due to out of memory,
3081 * we'll just stay with what we've got.
3083 if (PIDLIST_REALLOC_DIFFERENCE(length
, dest
)) {
3084 newlist
= pidlist_resize(list
, dest
);
3091 static int cmppid(const void *a
, const void *b
)
3093 return *(pid_t
*)a
- *(pid_t
*)b
;
3097 * find the appropriate pidlist for our purpose (given procs vs tasks)
3098 * returns with the lock on that pidlist already held, and takes care
3099 * of the use count, or returns NULL with no locks held if we're out of
3102 static struct cgroup_pidlist
*cgroup_pidlist_find(struct cgroup
*cgrp
,
3103 enum cgroup_filetype type
)
3105 struct cgroup_pidlist
*l
;
3106 /* don't need task_nsproxy() if we're looking at ourself */
3107 struct pid_namespace
*ns
= current
->nsproxy
->pid_ns
;
3110 * We can't drop the pidlist_mutex before taking the l->mutex in case
3111 * the last ref-holder is trying to remove l from the list at the same
3112 * time. Holding the pidlist_mutex precludes somebody taking whichever
3113 * list we find out from under us - compare release_pid_array().
3115 mutex_lock(&cgrp
->pidlist_mutex
);
3116 list_for_each_entry(l
, &cgrp
->pidlists
, links
) {
3117 if (l
->key
.type
== type
&& l
->key
.ns
== ns
) {
3118 /* make sure l doesn't vanish out from under us */
3119 down_write(&l
->mutex
);
3120 mutex_unlock(&cgrp
->pidlist_mutex
);
3124 /* entry not found; create a new one */
3125 l
= kmalloc(sizeof(struct cgroup_pidlist
), GFP_KERNEL
);
3127 mutex_unlock(&cgrp
->pidlist_mutex
);
3130 init_rwsem(&l
->mutex
);
3131 down_write(&l
->mutex
);
3133 l
->key
.ns
= get_pid_ns(ns
);
3134 l
->use_count
= 0; /* don't increment here */
3137 list_add(&l
->links
, &cgrp
->pidlists
);
3138 mutex_unlock(&cgrp
->pidlist_mutex
);
3143 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
3145 static int pidlist_array_load(struct cgroup
*cgrp
, enum cgroup_filetype type
,
3146 struct cgroup_pidlist
**lp
)
3150 int pid
, n
= 0; /* used for populating the array */
3151 struct cgroup_iter it
;
3152 struct task_struct
*tsk
;
3153 struct cgroup_pidlist
*l
;
3156 * If cgroup gets more users after we read count, we won't have
3157 * enough space - tough. This race is indistinguishable to the
3158 * caller from the case that the additional cgroup users didn't
3159 * show up until sometime later on.
3161 length
= cgroup_task_count(cgrp
);
3162 array
= pidlist_allocate(length
);
3165 /* now, populate the array */
3166 cgroup_iter_start(cgrp
, &it
);
3167 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3168 if (unlikely(n
== length
))
3170 /* get tgid or pid for procs or tasks file respectively */
3171 if (type
== CGROUP_FILE_PROCS
)
3172 pid
= task_tgid_vnr(tsk
);
3174 pid
= task_pid_vnr(tsk
);
3175 if (pid
> 0) /* make sure to only use valid results */
3178 cgroup_iter_end(cgrp
, &it
);
3180 /* now sort & (if procs) strip out duplicates */
3181 sort(array
, length
, sizeof(pid_t
), cmppid
, NULL
);
3182 if (type
== CGROUP_FILE_PROCS
)
3183 length
= pidlist_uniq(&array
, length
);
3184 l
= cgroup_pidlist_find(cgrp
, type
);
3186 pidlist_free(array
);
3189 /* store array, freeing old if necessary - lock already held */
3190 pidlist_free(l
->list
);
3194 up_write(&l
->mutex
);
3200 * cgroupstats_build - build and fill cgroupstats
3201 * @stats: cgroupstats to fill information into
3202 * @dentry: A dentry entry belonging to the cgroup for which stats have
3205 * Build and fill cgroupstats so that taskstats can export it to user
3208 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
3211 struct cgroup
*cgrp
;
3212 struct cgroup_iter it
;
3213 struct task_struct
*tsk
;
3216 * Validate dentry by checking the superblock operations,
3217 * and make sure it's a directory.
3219 if (dentry
->d_sb
->s_op
!= &cgroup_ops
||
3220 !S_ISDIR(dentry
->d_inode
->i_mode
))
3224 cgrp
= dentry
->d_fsdata
;
3226 cgroup_iter_start(cgrp
, &it
);
3227 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
3228 switch (tsk
->state
) {
3230 stats
->nr_running
++;
3232 case TASK_INTERRUPTIBLE
:
3233 stats
->nr_sleeping
++;
3235 case TASK_UNINTERRUPTIBLE
:
3236 stats
->nr_uninterruptible
++;
3239 stats
->nr_stopped
++;
3242 if (delayacct_is_task_waiting_on_io(tsk
))
3243 stats
->nr_io_wait
++;
3247 cgroup_iter_end(cgrp
, &it
);
3255 * seq_file methods for the tasks/procs files. The seq_file position is the
3256 * next pid to display; the seq_file iterator is a pointer to the pid
3257 * in the cgroup->l->list array.
3260 static void *cgroup_pidlist_start(struct seq_file
*s
, loff_t
*pos
)
3263 * Initially we receive a position value that corresponds to
3264 * one more than the last pid shown (or 0 on the first call or
3265 * after a seek to the start). Use a binary-search to find the
3266 * next pid to display, if any
3268 struct cgroup_pidlist
*l
= s
->private;
3269 int index
= 0, pid
= *pos
;
3272 down_read(&l
->mutex
);
3274 int end
= l
->length
;
3276 while (index
< end
) {
3277 int mid
= (index
+ end
) / 2;
3278 if (l
->list
[mid
] == pid
) {
3281 } else if (l
->list
[mid
] <= pid
)
3287 /* If we're off the end of the array, we're done */
3288 if (index
>= l
->length
)
3290 /* Update the abstract position to be the actual pid that we found */
3291 iter
= l
->list
+ index
;
3296 static void cgroup_pidlist_stop(struct seq_file
*s
, void *v
)
3298 struct cgroup_pidlist
*l
= s
->private;
3302 static void *cgroup_pidlist_next(struct seq_file
*s
, void *v
, loff_t
*pos
)
3304 struct cgroup_pidlist
*l
= s
->private;
3306 pid_t
*end
= l
->list
+ l
->length
;
3308 * Advance to the next pid in the array. If this goes off the
3320 static int cgroup_pidlist_show(struct seq_file
*s
, void *v
)
3322 return seq_printf(s
, "%d\n", *(int *)v
);
3326 * seq_operations functions for iterating on pidlists through seq_file -
3327 * independent of whether it's tasks or procs
3329 static const struct seq_operations cgroup_pidlist_seq_operations
= {
3330 .start
= cgroup_pidlist_start
,
3331 .stop
= cgroup_pidlist_stop
,
3332 .next
= cgroup_pidlist_next
,
3333 .show
= cgroup_pidlist_show
,
3336 static void cgroup_release_pid_array(struct cgroup_pidlist
*l
)
3339 * the case where we're the last user of this particular pidlist will
3340 * have us remove it from the cgroup's list, which entails taking the
3341 * mutex. since in pidlist_find the pidlist->lock depends on cgroup->
3342 * pidlist_mutex, we have to take pidlist_mutex first.
3344 mutex_lock(&l
->owner
->pidlist_mutex
);
3345 down_write(&l
->mutex
);
3346 BUG_ON(!l
->use_count
);
3347 if (!--l
->use_count
) {
3348 /* we're the last user if refcount is 0; remove and free */
3349 list_del(&l
->links
);
3350 mutex_unlock(&l
->owner
->pidlist_mutex
);
3351 pidlist_free(l
->list
);
3352 put_pid_ns(l
->key
.ns
);
3353 up_write(&l
->mutex
);
3357 mutex_unlock(&l
->owner
->pidlist_mutex
);
3358 up_write(&l
->mutex
);
3361 static int cgroup_pidlist_release(struct inode
*inode
, struct file
*file
)
3363 struct cgroup_pidlist
*l
;
3364 if (!(file
->f_mode
& FMODE_READ
))
3367 * the seq_file will only be initialized if the file was opened for
3368 * reading; hence we check if it's not null only in that case.
3370 l
= ((struct seq_file
*)file
->private_data
)->private;
3371 cgroup_release_pid_array(l
);
3372 return seq_release(inode
, file
);
3375 static const struct file_operations cgroup_pidlist_operations
= {
3377 .llseek
= seq_lseek
,
3378 .write
= cgroup_file_write
,
3379 .release
= cgroup_pidlist_release
,
3383 * The following functions handle opens on a file that displays a pidlist
3384 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
3387 /* helper function for the two below it */
3388 static int cgroup_pidlist_open(struct file
*file
, enum cgroup_filetype type
)
3390 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
3391 struct cgroup_pidlist
*l
;
3394 /* Nothing to do for write-only files */
3395 if (!(file
->f_mode
& FMODE_READ
))
3398 /* have the array populated */
3399 retval
= pidlist_array_load(cgrp
, type
, &l
);
3402 /* configure file information */
3403 file
->f_op
= &cgroup_pidlist_operations
;
3405 retval
= seq_open(file
, &cgroup_pidlist_seq_operations
);
3407 cgroup_release_pid_array(l
);
3410 ((struct seq_file
*)file
->private_data
)->private = l
;
3413 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
3415 return cgroup_pidlist_open(file
, CGROUP_FILE_TASKS
);
3417 static int cgroup_procs_open(struct inode
*unused
, struct file
*file
)
3419 return cgroup_pidlist_open(file
, CGROUP_FILE_PROCS
);
3422 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
3425 return notify_on_release(cgrp
);
3428 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
3432 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3434 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3436 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3441 * Unregister event and free resources.
3443 * Gets called from workqueue.
3445 static void cgroup_event_remove(struct work_struct
*work
)
3447 struct cgroup_event
*event
= container_of(work
, struct cgroup_event
,
3449 struct cgroup
*cgrp
= event
->cgrp
;
3451 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3453 eventfd_ctx_put(event
->eventfd
);
3459 * Gets called on POLLHUP on eventfd when user closes it.
3461 * Called with wqh->lock held and interrupts disabled.
3463 static int cgroup_event_wake(wait_queue_t
*wait
, unsigned mode
,
3464 int sync
, void *key
)
3466 struct cgroup_event
*event
= container_of(wait
,
3467 struct cgroup_event
, wait
);
3468 struct cgroup
*cgrp
= event
->cgrp
;
3469 unsigned long flags
= (unsigned long)key
;
3471 if (flags
& POLLHUP
) {
3472 __remove_wait_queue(event
->wqh
, &event
->wait
);
3473 spin_lock(&cgrp
->event_list_lock
);
3474 list_del(&event
->list
);
3475 spin_unlock(&cgrp
->event_list_lock
);
3477 * We are in atomic context, but cgroup_event_remove() may
3478 * sleep, so we have to call it in workqueue.
3480 schedule_work(&event
->remove
);
3486 static void cgroup_event_ptable_queue_proc(struct file
*file
,
3487 wait_queue_head_t
*wqh
, poll_table
*pt
)
3489 struct cgroup_event
*event
= container_of(pt
,
3490 struct cgroup_event
, pt
);
3493 add_wait_queue(wqh
, &event
->wait
);
3497 * Parse input and register new cgroup event handler.
3499 * Input must be in format '<event_fd> <control_fd> <args>'.
3500 * Interpretation of args is defined by control file implementation.
3502 static int cgroup_write_event_control(struct cgroup
*cgrp
, struct cftype
*cft
,
3505 struct cgroup_event
*event
= NULL
;
3506 unsigned int efd
, cfd
;
3507 struct file
*efile
= NULL
;
3508 struct file
*cfile
= NULL
;
3512 efd
= simple_strtoul(buffer
, &endp
, 10);
3517 cfd
= simple_strtoul(buffer
, &endp
, 10);
3518 if ((*endp
!= ' ') && (*endp
!= '\0'))
3522 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3526 INIT_LIST_HEAD(&event
->list
);
3527 init_poll_funcptr(&event
->pt
, cgroup_event_ptable_queue_proc
);
3528 init_waitqueue_func_entry(&event
->wait
, cgroup_event_wake
);
3529 INIT_WORK(&event
->remove
, cgroup_event_remove
);
3531 efile
= eventfd_fget(efd
);
3532 if (IS_ERR(efile
)) {
3533 ret
= PTR_ERR(efile
);
3537 event
->eventfd
= eventfd_ctx_fileget(efile
);
3538 if (IS_ERR(event
->eventfd
)) {
3539 ret
= PTR_ERR(event
->eventfd
);
3549 /* the process need read permission on control file */
3550 ret
= file_permission(cfile
, MAY_READ
);
3554 event
->cft
= __file_cft(cfile
);
3555 if (IS_ERR(event
->cft
)) {
3556 ret
= PTR_ERR(event
->cft
);
3560 if (!event
->cft
->register_event
|| !event
->cft
->unregister_event
) {
3565 ret
= event
->cft
->register_event(cgrp
, event
->cft
,
3566 event
->eventfd
, buffer
);
3570 if (efile
->f_op
->poll(efile
, &event
->pt
) & POLLHUP
) {
3571 event
->cft
->unregister_event(cgrp
, event
->cft
, event
->eventfd
);
3577 * Events should be removed after rmdir of cgroup directory, but before
3578 * destroying subsystem state objects. Let's take reference to cgroup
3579 * directory dentry to do that.
3583 spin_lock(&cgrp
->event_list_lock
);
3584 list_add(&event
->list
, &cgrp
->event_list
);
3585 spin_unlock(&cgrp
->event_list_lock
);
3596 if (event
&& event
->eventfd
&& !IS_ERR(event
->eventfd
))
3597 eventfd_ctx_put(event
->eventfd
);
3599 if (!IS_ERR_OR_NULL(efile
))
3607 static u64
cgroup_clone_children_read(struct cgroup
*cgrp
,
3610 return clone_children(cgrp
);
3613 static int cgroup_clone_children_write(struct cgroup
*cgrp
,
3618 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3620 clear_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3625 * for the common functions, 'private' gives the type of file
3627 /* for hysterical raisins, we can't put this on the older files */
3628 #define CGROUP_FILE_GENERIC_PREFIX "cgroup."
3629 static struct cftype files
[] = {
3632 .open
= cgroup_tasks_open
,
3633 .write_u64
= cgroup_tasks_write
,
3634 .release
= cgroup_pidlist_release
,
3635 .mode
= S_IRUGO
| S_IWUSR
,
3638 .name
= CGROUP_FILE_GENERIC_PREFIX
"procs",
3639 .open
= cgroup_procs_open
,
3640 .write_u64
= cgroup_procs_write
,
3641 .release
= cgroup_pidlist_release
,
3642 .mode
= S_IRUGO
| S_IWUSR
,
3645 .name
= "notify_on_release",
3646 .read_u64
= cgroup_read_notify_on_release
,
3647 .write_u64
= cgroup_write_notify_on_release
,
3650 .name
= CGROUP_FILE_GENERIC_PREFIX
"event_control",
3651 .write_string
= cgroup_write_event_control
,
3655 .name
= "cgroup.clone_children",
3656 .read_u64
= cgroup_clone_children_read
,
3657 .write_u64
= cgroup_clone_children_write
,
3661 static struct cftype cft_release_agent
= {
3662 .name
= "release_agent",
3663 .read_seq_string
= cgroup_release_agent_show
,
3664 .write_string
= cgroup_release_agent_write
,
3665 .max_write_len
= PATH_MAX
,
3668 static int cgroup_populate_dir(struct cgroup
*cgrp
)
3671 struct cgroup_subsys
*ss
;
3673 /* First clear out any existing files */
3674 cgroup_clear_directory(cgrp
->dentry
);
3676 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
3680 if (cgrp
== cgrp
->top_cgroup
) {
3681 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
3685 for_each_subsys(cgrp
->root
, ss
) {
3686 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
3689 /* This cgroup is ready now */
3690 for_each_subsys(cgrp
->root
, ss
) {
3691 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3693 * Update id->css pointer and make this css visible from
3694 * CSS ID functions. This pointer will be dereferened
3695 * from RCU-read-side without locks.
3698 rcu_assign_pointer(css
->id
->css
, css
);
3704 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
3705 struct cgroup_subsys
*ss
,
3706 struct cgroup
*cgrp
)
3709 atomic_set(&css
->refcnt
, 1);
3712 if (cgrp
== dummytop
)
3713 set_bit(CSS_ROOT
, &css
->flags
);
3714 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
3715 cgrp
->subsys
[ss
->subsys_id
] = css
;
3718 static void cgroup_lock_hierarchy(struct cgroupfs_root
*root
)
3720 /* We need to take each hierarchy_mutex in a consistent order */
3724 * No worry about a race with rebind_subsystems that might mess up the
3725 * locking order, since both parties are under cgroup_mutex.
3727 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3728 struct cgroup_subsys
*ss
= subsys
[i
];
3731 if (ss
->root
== root
)
3732 mutex_lock(&ss
->hierarchy_mutex
);
3736 static void cgroup_unlock_hierarchy(struct cgroupfs_root
*root
)
3740 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3741 struct cgroup_subsys
*ss
= subsys
[i
];
3744 if (ss
->root
== root
)
3745 mutex_unlock(&ss
->hierarchy_mutex
);
3750 * cgroup_create - create a cgroup
3751 * @parent: cgroup that will be parent of the new cgroup
3752 * @dentry: dentry of the new cgroup
3753 * @mode: mode to set on new inode
3755 * Must be called with the mutex on the parent inode held
3757 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
3760 struct cgroup
*cgrp
;
3761 struct cgroupfs_root
*root
= parent
->root
;
3763 struct cgroup_subsys
*ss
;
3764 struct super_block
*sb
= root
->sb
;
3766 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
3770 /* Grab a reference on the superblock so the hierarchy doesn't
3771 * get deleted on unmount if there are child cgroups. This
3772 * can be done outside cgroup_mutex, since the sb can't
3773 * disappear while someone has an open control file on the
3775 atomic_inc(&sb
->s_active
);
3777 mutex_lock(&cgroup_mutex
);
3779 init_cgroup_housekeeping(cgrp
);
3781 cgrp
->parent
= parent
;
3782 cgrp
->root
= parent
->root
;
3783 cgrp
->top_cgroup
= parent
->top_cgroup
;
3785 if (notify_on_release(parent
))
3786 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
3788 if (clone_children(parent
))
3789 set_bit(CGRP_CLONE_CHILDREN
, &cgrp
->flags
);
3791 for_each_subsys(root
, ss
) {
3792 struct cgroup_subsys_state
*css
= ss
->create(ss
, cgrp
);
3798 init_cgroup_css(css
, ss
, cgrp
);
3800 err
= alloc_css_id(ss
, parent
, cgrp
);
3804 /* At error, ->destroy() callback has to free assigned ID. */
3805 if (clone_children(parent
) && ss
->post_clone
)
3806 ss
->post_clone(ss
, cgrp
);
3809 cgroup_lock_hierarchy(root
);
3810 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
3811 cgroup_unlock_hierarchy(root
);
3812 root
->number_of_cgroups
++;
3814 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
3818 /* The cgroup directory was pre-locked for us */
3819 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
3821 err
= cgroup_populate_dir(cgrp
);
3822 /* If err < 0, we have a half-filled directory - oh well ;) */
3824 mutex_unlock(&cgroup_mutex
);
3825 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
3831 cgroup_lock_hierarchy(root
);
3832 list_del(&cgrp
->sibling
);
3833 cgroup_unlock_hierarchy(root
);
3834 root
->number_of_cgroups
--;
3838 for_each_subsys(root
, ss
) {
3839 if (cgrp
->subsys
[ss
->subsys_id
])
3840 ss
->destroy(ss
, cgrp
);
3843 mutex_unlock(&cgroup_mutex
);
3845 /* Release the reference count that we took on the superblock */
3846 deactivate_super(sb
);
3852 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3854 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
3856 /* the vfs holds inode->i_mutex already */
3857 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
3860 static int cgroup_has_css_refs(struct cgroup
*cgrp
)
3862 /* Check the reference count on each subsystem. Since we
3863 * already established that there are no tasks in the
3864 * cgroup, if the css refcount is also 1, then there should
3865 * be no outstanding references, so the subsystem is safe to
3866 * destroy. We scan across all subsystems rather than using
3867 * the per-hierarchy linked list of mounted subsystems since
3868 * we can be called via check_for_release() with no
3869 * synchronization other than RCU, and the subsystem linked
3870 * list isn't RCU-safe */
3873 * We won't need to lock the subsys array, because the subsystems
3874 * we're concerned about aren't going anywhere since our cgroup root
3875 * has a reference on them.
3877 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3878 struct cgroup_subsys
*ss
= subsys
[i
];
3879 struct cgroup_subsys_state
*css
;
3880 /* Skip subsystems not present or not in this hierarchy */
3881 if (ss
== NULL
|| ss
->root
!= cgrp
->root
)
3883 css
= cgrp
->subsys
[ss
->subsys_id
];
3884 /* When called from check_for_release() it's possible
3885 * that by this point the cgroup has been removed
3886 * and the css deleted. But a false-positive doesn't
3887 * matter, since it can only happen if the cgroup
3888 * has been deleted and hence no longer needs the
3889 * release agent to be called anyway. */
3890 if (css
&& (atomic_read(&css
->refcnt
) > 1))
3897 * Atomically mark all (or else none) of the cgroup's CSS objects as
3898 * CSS_REMOVED. Return true on success, or false if the cgroup has
3899 * busy subsystems. Call with cgroup_mutex held
3902 static int cgroup_clear_css_refs(struct cgroup
*cgrp
)
3904 struct cgroup_subsys
*ss
;
3905 unsigned long flags
;
3906 bool failed
= false;
3907 local_irq_save(flags
);
3908 for_each_subsys(cgrp
->root
, ss
) {
3909 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3912 /* We can only remove a CSS with a refcnt==1 */
3913 refcnt
= atomic_read(&css
->refcnt
);
3920 * Drop the refcnt to 0 while we check other
3921 * subsystems. This will cause any racing
3922 * css_tryget() to spin until we set the
3923 * CSS_REMOVED bits or abort
3925 if (atomic_cmpxchg(&css
->refcnt
, refcnt
, 0) == refcnt
)
3931 for_each_subsys(cgrp
->root
, ss
) {
3932 struct cgroup_subsys_state
*css
= cgrp
->subsys
[ss
->subsys_id
];
3935 * Restore old refcnt if we previously managed
3936 * to clear it from 1 to 0
3938 if (!atomic_read(&css
->refcnt
))
3939 atomic_set(&css
->refcnt
, 1);
3941 /* Commit the fact that the CSS is removed */
3942 set_bit(CSS_REMOVED
, &css
->flags
);
3945 local_irq_restore(flags
);
3949 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
3951 struct cgroup
*cgrp
= dentry
->d_fsdata
;
3953 struct cgroup
*parent
;
3955 struct cgroup_event
*event
, *tmp
;
3958 /* the vfs holds both inode->i_mutex already */
3960 mutex_lock(&cgroup_mutex
);
3961 if (atomic_read(&cgrp
->count
) != 0) {
3962 mutex_unlock(&cgroup_mutex
);
3965 if (!list_empty(&cgrp
->children
)) {
3966 mutex_unlock(&cgroup_mutex
);
3969 mutex_unlock(&cgroup_mutex
);
3972 * In general, subsystem has no css->refcnt after pre_destroy(). But
3973 * in racy cases, subsystem may have to get css->refcnt after
3974 * pre_destroy() and it makes rmdir return with -EBUSY. This sometimes
3975 * make rmdir return -EBUSY too often. To avoid that, we use waitqueue
3976 * for cgroup's rmdir. CGRP_WAIT_ON_RMDIR is for synchronizing rmdir
3977 * and subsystem's reference count handling. Please see css_get/put
3978 * and css_tryget() and cgroup_wakeup_rmdir_waiter() implementation.
3980 set_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3983 * Call pre_destroy handlers of subsys. Notify subsystems
3984 * that rmdir() request comes.
3986 ret
= cgroup_call_pre_destroy(cgrp
);
3988 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3992 mutex_lock(&cgroup_mutex
);
3993 parent
= cgrp
->parent
;
3994 if (atomic_read(&cgrp
->count
) || !list_empty(&cgrp
->children
)) {
3995 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
3996 mutex_unlock(&cgroup_mutex
);
3999 prepare_to_wait(&cgroup_rmdir_waitq
, &wait
, TASK_INTERRUPTIBLE
);
4000 if (!cgroup_clear_css_refs(cgrp
)) {
4001 mutex_unlock(&cgroup_mutex
);
4003 * Because someone may call cgroup_wakeup_rmdir_waiter() before
4004 * prepare_to_wait(), we need to check this flag.
4006 if (test_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
))
4008 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4009 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4010 if (signal_pending(current
))
4014 /* NO css_tryget() can success after here. */
4015 finish_wait(&cgroup_rmdir_waitq
, &wait
);
4016 clear_bit(CGRP_WAIT_ON_RMDIR
, &cgrp
->flags
);
4018 spin_lock(&release_list_lock
);
4019 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
4020 if (!list_empty(&cgrp
->release_list
))
4021 list_del_init(&cgrp
->release_list
);
4022 spin_unlock(&release_list_lock
);
4024 cgroup_lock_hierarchy(cgrp
->root
);
4025 /* delete this cgroup from parent->children */
4026 list_del_init(&cgrp
->sibling
);
4027 cgroup_unlock_hierarchy(cgrp
->root
);
4029 d
= dget(cgrp
->dentry
);
4031 cgroup_d_remove_dir(d
);
4034 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
4035 check_for_release(parent
);
4038 * Unregister events and notify userspace.
4039 * Notify userspace about cgroup removing only after rmdir of cgroup
4040 * directory to avoid race between userspace and kernelspace
4042 spin_lock(&cgrp
->event_list_lock
);
4043 list_for_each_entry_safe(event
, tmp
, &cgrp
->event_list
, list
) {
4044 list_del(&event
->list
);
4045 remove_wait_queue(event
->wqh
, &event
->wait
);
4046 eventfd_signal(event
->eventfd
, 1);
4047 schedule_work(&event
->remove
);
4049 spin_unlock(&cgrp
->event_list_lock
);
4051 mutex_unlock(&cgroup_mutex
);
4055 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
4057 struct cgroup_subsys_state
*css
;
4059 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
4061 /* Create the top cgroup state for this subsystem */
4062 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4063 ss
->root
= &rootnode
;
4064 css
= ss
->create(ss
, dummytop
);
4065 /* We don't handle early failures gracefully */
4066 BUG_ON(IS_ERR(css
));
4067 init_cgroup_css(css
, ss
, dummytop
);
4069 /* Update the init_css_set to contain a subsys
4070 * pointer to this state - since the subsystem is
4071 * newly registered, all tasks and hence the
4072 * init_css_set is in the subsystem's top cgroup. */
4073 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
4075 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
4077 /* At system boot, before all subsystems have been
4078 * registered, no tasks have been forked, so we don't
4079 * need to invoke fork callbacks here. */
4080 BUG_ON(!list_empty(&init_task
.tasks
));
4082 mutex_init(&ss
->hierarchy_mutex
);
4083 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4086 /* this function shouldn't be used with modular subsystems, since they
4087 * need to register a subsys_id, among other things */
4092 * cgroup_load_subsys: load and register a modular subsystem at runtime
4093 * @ss: the subsystem to load
4095 * This function should be called in a modular subsystem's initcall. If the
4096 * subsystem is built as a module, it will be assigned a new subsys_id and set
4097 * up for use. If the subsystem is built-in anyway, work is delegated to the
4098 * simpler cgroup_init_subsys.
4100 int __init_or_module
cgroup_load_subsys(struct cgroup_subsys
*ss
)
4103 struct cgroup_subsys_state
*css
;
4105 /* check name and function validity */
4106 if (ss
->name
== NULL
|| strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
||
4107 ss
->create
== NULL
|| ss
->destroy
== NULL
)
4111 * we don't support callbacks in modular subsystems. this check is
4112 * before the ss->module check for consistency; a subsystem that could
4113 * be a module should still have no callbacks even if the user isn't
4114 * compiling it as one.
4116 if (ss
->fork
|| ss
->exit
)
4120 * an optionally modular subsystem is built-in: we want to do nothing,
4121 * since cgroup_init_subsys will have already taken care of it.
4123 if (ss
->module
== NULL
) {
4124 /* a few sanity checks */
4125 BUG_ON(ss
->subsys_id
>= CGROUP_BUILTIN_SUBSYS_COUNT
);
4126 BUG_ON(subsys
[ss
->subsys_id
] != ss
);
4131 * need to register a subsys id before anything else - for example,
4132 * init_cgroup_css needs it.
4134 mutex_lock(&cgroup_mutex
);
4135 /* find the first empty slot in the array */
4136 for (i
= CGROUP_BUILTIN_SUBSYS_COUNT
; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4137 if (subsys
[i
] == NULL
)
4140 if (i
== CGROUP_SUBSYS_COUNT
) {
4141 /* maximum number of subsystems already registered! */
4142 mutex_unlock(&cgroup_mutex
);
4145 /* assign ourselves the subsys_id */
4150 * no ss->create seems to need anything important in the ss struct, so
4151 * this can happen first (i.e. before the rootnode attachment).
4153 css
= ss
->create(ss
, dummytop
);
4155 /* failure case - need to deassign the subsys[] slot. */
4157 mutex_unlock(&cgroup_mutex
);
4158 return PTR_ERR(css
);
4161 list_add(&ss
->sibling
, &rootnode
.subsys_list
);
4162 ss
->root
= &rootnode
;
4164 /* our new subsystem will be attached to the dummy hierarchy. */
4165 init_cgroup_css(css
, ss
, dummytop
);
4166 /* init_idr must be after init_cgroup_css because it sets css->id. */
4168 int ret
= cgroup_init_idr(ss
, css
);
4170 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4171 ss
->destroy(ss
, dummytop
);
4173 mutex_unlock(&cgroup_mutex
);
4179 * Now we need to entangle the css into the existing css_sets. unlike
4180 * in cgroup_init_subsys, there are now multiple css_sets, so each one
4181 * will need a new pointer to it; done by iterating the css_set_table.
4182 * furthermore, modifying the existing css_sets will corrupt the hash
4183 * table state, so each changed css_set will need its hash recomputed.
4184 * this is all done under the css_set_lock.
4186 write_lock(&css_set_lock
);
4187 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
4189 struct hlist_node
*node
, *tmp
;
4190 struct hlist_head
*bucket
= &css_set_table
[i
], *new_bucket
;
4192 hlist_for_each_entry_safe(cg
, node
, tmp
, bucket
, hlist
) {
4193 /* skip entries that we already rehashed */
4194 if (cg
->subsys
[ss
->subsys_id
])
4196 /* remove existing entry */
4197 hlist_del(&cg
->hlist
);
4199 cg
->subsys
[ss
->subsys_id
] = css
;
4200 /* recompute hash and restore entry */
4201 new_bucket
= css_set_hash(cg
->subsys
);
4202 hlist_add_head(&cg
->hlist
, new_bucket
);
4205 write_unlock(&css_set_lock
);
4207 mutex_init(&ss
->hierarchy_mutex
);
4208 lockdep_set_class(&ss
->hierarchy_mutex
, &ss
->subsys_key
);
4212 mutex_unlock(&cgroup_mutex
);
4215 EXPORT_SYMBOL_GPL(cgroup_load_subsys
);
4218 * cgroup_unload_subsys: unload a modular subsystem
4219 * @ss: the subsystem to unload
4221 * This function should be called in a modular subsystem's exitcall. When this
4222 * function is invoked, the refcount on the subsystem's module will be 0, so
4223 * the subsystem will not be attached to any hierarchy.
4225 void cgroup_unload_subsys(struct cgroup_subsys
*ss
)
4227 struct cg_cgroup_link
*link
;
4228 struct hlist_head
*hhead
;
4230 BUG_ON(ss
->module
== NULL
);
4233 * we shouldn't be called if the subsystem is in use, and the use of
4234 * try_module_get in parse_cgroupfs_options should ensure that it
4235 * doesn't start being used while we're killing it off.
4237 BUG_ON(ss
->root
!= &rootnode
);
4239 mutex_lock(&cgroup_mutex
);
4240 /* deassign the subsys_id */
4241 BUG_ON(ss
->subsys_id
< CGROUP_BUILTIN_SUBSYS_COUNT
);
4242 subsys
[ss
->subsys_id
] = NULL
;
4244 /* remove subsystem from rootnode's list of subsystems */
4245 list_del_init(&ss
->sibling
);
4248 * disentangle the css from all css_sets attached to the dummytop. as
4249 * in loading, we need to pay our respects to the hashtable gods.
4251 write_lock(&css_set_lock
);
4252 list_for_each_entry(link
, &dummytop
->css_sets
, cgrp_link_list
) {
4253 struct css_set
*cg
= link
->cg
;
4255 hlist_del(&cg
->hlist
);
4256 BUG_ON(!cg
->subsys
[ss
->subsys_id
]);
4257 cg
->subsys
[ss
->subsys_id
] = NULL
;
4258 hhead
= css_set_hash(cg
->subsys
);
4259 hlist_add_head(&cg
->hlist
, hhead
);
4261 write_unlock(&css_set_lock
);
4264 * remove subsystem's css from the dummytop and free it - need to free
4265 * before marking as null because ss->destroy needs the cgrp->subsys
4266 * pointer to find their state. note that this also takes care of
4267 * freeing the css_id.
4269 ss
->destroy(ss
, dummytop
);
4270 dummytop
->subsys
[ss
->subsys_id
] = NULL
;
4272 mutex_unlock(&cgroup_mutex
);
4274 EXPORT_SYMBOL_GPL(cgroup_unload_subsys
);
4277 * cgroup_init_early - cgroup initialization at system boot
4279 * Initialize cgroups at system boot, and initialize any
4280 * subsystems that request early init.
4282 int __init
cgroup_init_early(void)
4285 atomic_set(&init_css_set
.refcount
, 1);
4286 INIT_LIST_HEAD(&init_css_set
.cg_links
);
4287 INIT_LIST_HEAD(&init_css_set
.tasks
);
4288 INIT_HLIST_NODE(&init_css_set
.hlist
);
4290 init_cgroup_root(&rootnode
);
4292 init_task
.cgroups
= &init_css_set
;
4294 init_css_set_link
.cg
= &init_css_set
;
4295 init_css_set_link
.cgrp
= dummytop
;
4296 list_add(&init_css_set_link
.cgrp_link_list
,
4297 &rootnode
.top_cgroup
.css_sets
);
4298 list_add(&init_css_set_link
.cg_link_list
,
4299 &init_css_set
.cg_links
);
4301 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
4302 INIT_HLIST_HEAD(&css_set_table
[i
]);
4304 /* at bootup time, we don't worry about modular subsystems */
4305 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4306 struct cgroup_subsys
*ss
= subsys
[i
];
4309 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
4310 BUG_ON(!ss
->create
);
4311 BUG_ON(!ss
->destroy
);
4312 if (ss
->subsys_id
!= i
) {
4313 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
4314 ss
->name
, ss
->subsys_id
);
4319 cgroup_init_subsys(ss
);
4325 * cgroup_init - cgroup initialization
4327 * Register cgroup filesystem and /proc file, and initialize
4328 * any subsystems that didn't request early init.
4330 int __init
cgroup_init(void)
4334 struct hlist_head
*hhead
;
4336 err
= bdi_init(&cgroup_backing_dev_info
);
4340 /* at bootup time, we don't worry about modular subsystems */
4341 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4342 struct cgroup_subsys
*ss
= subsys
[i
];
4343 if (!ss
->early_init
)
4344 cgroup_init_subsys(ss
);
4346 cgroup_init_idr(ss
, init_css_set
.subsys
[ss
->subsys_id
]);
4349 /* Add init_css_set to the hash table */
4350 hhead
= css_set_hash(init_css_set
.subsys
);
4351 hlist_add_head(&init_css_set
.hlist
, hhead
);
4352 BUG_ON(!init_root_id(&rootnode
));
4354 cgroup_kobj
= kobject_create_and_add("cgroup", fs_kobj
);
4360 err
= register_filesystem(&cgroup_fs_type
);
4362 kobject_put(cgroup_kobj
);
4366 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
4370 bdi_destroy(&cgroup_backing_dev_info
);
4376 * proc_cgroup_show()
4377 * - Print task's cgroup paths into seq_file, one line for each hierarchy
4378 * - Used for /proc/<pid>/cgroup.
4379 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
4380 * doesn't really matter if tsk->cgroup changes after we read it,
4381 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
4382 * anyway. No need to check that tsk->cgroup != NULL, thanks to
4383 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
4384 * cgroup to top_cgroup.
4387 /* TODO: Use a proper seq_file iterator */
4388 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
4391 struct task_struct
*tsk
;
4394 struct cgroupfs_root
*root
;
4397 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4403 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
4409 mutex_lock(&cgroup_mutex
);
4411 for_each_active_root(root
) {
4412 struct cgroup_subsys
*ss
;
4413 struct cgroup
*cgrp
;
4416 seq_printf(m
, "%d:", root
->hierarchy_id
);
4417 for_each_subsys(root
, ss
)
4418 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
4419 if (strlen(root
->name
))
4420 seq_printf(m
, "%sname=%s", count
? "," : "",
4423 cgrp
= task_cgroup_from_root(tsk
, root
);
4424 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
4432 mutex_unlock(&cgroup_mutex
);
4433 put_task_struct(tsk
);
4440 static int cgroup_open(struct inode
*inode
, struct file
*file
)
4442 struct pid
*pid
= PROC_I(inode
)->pid
;
4443 return single_open(file
, proc_cgroup_show
, pid
);
4446 const struct file_operations proc_cgroup_operations
= {
4447 .open
= cgroup_open
,
4449 .llseek
= seq_lseek
,
4450 .release
= single_release
,
4453 /* Display information about each subsystem and each hierarchy */
4454 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
4458 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
4460 * ideally we don't want subsystems moving around while we do this.
4461 * cgroup_mutex is also necessary to guarantee an atomic snapshot of
4462 * subsys/hierarchy state.
4464 mutex_lock(&cgroup_mutex
);
4465 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
4466 struct cgroup_subsys
*ss
= subsys
[i
];
4469 seq_printf(m
, "%s\t%d\t%d\t%d\n",
4470 ss
->name
, ss
->root
->hierarchy_id
,
4471 ss
->root
->number_of_cgroups
, !ss
->disabled
);
4473 mutex_unlock(&cgroup_mutex
);
4477 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
4479 return single_open(file
, proc_cgroupstats_show
, NULL
);
4482 static const struct file_operations proc_cgroupstats_operations
= {
4483 .open
= cgroupstats_open
,
4485 .llseek
= seq_lseek
,
4486 .release
= single_release
,
4490 * cgroup_fork - attach newly forked task to its parents cgroup.
4491 * @child: pointer to task_struct of forking parent process.
4493 * Description: A task inherits its parent's cgroup at fork().
4495 * A pointer to the shared css_set was automatically copied in
4496 * fork.c by dup_task_struct(). However, we ignore that copy, since
4497 * it was not made under the protection of RCU or cgroup_mutex, so
4498 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
4499 * have already changed current->cgroups, allowing the previously
4500 * referenced cgroup group to be removed and freed.
4502 * At the point that cgroup_fork() is called, 'current' is the parent
4503 * task, and the passed argument 'child' points to the child task.
4505 void cgroup_fork(struct task_struct
*child
)
4508 child
->cgroups
= current
->cgroups
;
4509 get_css_set(child
->cgroups
);
4510 task_unlock(current
);
4511 INIT_LIST_HEAD(&child
->cg_list
);
4515 * cgroup_fork_callbacks - run fork callbacks
4516 * @child: the new task
4518 * Called on a new task very soon before adding it to the
4519 * tasklist. No need to take any locks since no-one can
4520 * be operating on this task.
4522 void cgroup_fork_callbacks(struct task_struct
*child
)
4524 if (need_forkexit_callback
) {
4527 * forkexit callbacks are only supported for builtin
4528 * subsystems, and the builtin section of the subsys array is
4529 * immutable, so we don't need to lock the subsys array here.
4531 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4532 struct cgroup_subsys
*ss
= subsys
[i
];
4534 ss
->fork(ss
, child
);
4540 * cgroup_post_fork - called on a new task after adding it to the task list
4541 * @child: the task in question
4543 * Adds the task to the list running through its css_set if necessary.
4544 * Has to be after the task is visible on the task list in case we race
4545 * with the first call to cgroup_iter_start() - to guarantee that the
4546 * new task ends up on its list.
4548 void cgroup_post_fork(struct task_struct
*child
)
4550 if (use_task_css_set_links
) {
4551 write_lock(&css_set_lock
);
4553 if (list_empty(&child
->cg_list
))
4554 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
4556 write_unlock(&css_set_lock
);
4560 * cgroup_exit - detach cgroup from exiting task
4561 * @tsk: pointer to task_struct of exiting process
4562 * @run_callback: run exit callbacks?
4564 * Description: Detach cgroup from @tsk and release it.
4566 * Note that cgroups marked notify_on_release force every task in
4567 * them to take the global cgroup_mutex mutex when exiting.
4568 * This could impact scaling on very large systems. Be reluctant to
4569 * use notify_on_release cgroups where very high task exit scaling
4570 * is required on large systems.
4572 * the_top_cgroup_hack:
4574 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
4576 * We call cgroup_exit() while the task is still competent to
4577 * handle notify_on_release(), then leave the task attached to the
4578 * root cgroup in each hierarchy for the remainder of its exit.
4580 * To do this properly, we would increment the reference count on
4581 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
4582 * code we would add a second cgroup function call, to drop that
4583 * reference. This would just create an unnecessary hot spot on
4584 * the top_cgroup reference count, to no avail.
4586 * Normally, holding a reference to a cgroup without bumping its
4587 * count is unsafe. The cgroup could go away, or someone could
4588 * attach us to a different cgroup, decrementing the count on
4589 * the first cgroup that we never incremented. But in this case,
4590 * top_cgroup isn't going away, and either task has PF_EXITING set,
4591 * which wards off any cgroup_attach_task() attempts, or task is a failed
4592 * fork, never visible to cgroup_attach_task.
4594 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
4600 * Unlink from the css_set task list if necessary.
4601 * Optimistically check cg_list before taking
4604 if (!list_empty(&tsk
->cg_list
)) {
4605 write_lock(&css_set_lock
);
4606 if (!list_empty(&tsk
->cg_list
))
4607 list_del_init(&tsk
->cg_list
);
4608 write_unlock(&css_set_lock
);
4611 /* Reassign the task to the init_css_set. */
4614 tsk
->cgroups
= &init_css_set
;
4616 if (run_callbacks
&& need_forkexit_callback
) {
4618 * modular subsystems can't use callbacks, so no need to lock
4621 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4622 struct cgroup_subsys
*ss
= subsys
[i
];
4624 struct cgroup
*old_cgrp
=
4625 rcu_dereference_raw(cg
->subsys
[i
])->cgroup
;
4626 struct cgroup
*cgrp
= task_cgroup(tsk
, i
);
4627 ss
->exit(ss
, cgrp
, old_cgrp
, tsk
);
4634 put_css_set_taskexit(cg
);
4638 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
4639 * @cgrp: the cgroup in question
4640 * @task: the task in question
4642 * See if @cgrp is a descendant of @task's cgroup in the appropriate
4645 * If we are sending in dummytop, then presumably we are creating
4646 * the top cgroup in the subsystem.
4648 * Called only by the ns (nsproxy) cgroup.
4650 int cgroup_is_descendant(const struct cgroup
*cgrp
, struct task_struct
*task
)
4653 struct cgroup
*target
;
4655 if (cgrp
== dummytop
)
4658 target
= task_cgroup_from_root(task
, cgrp
->root
);
4659 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
4660 cgrp
= cgrp
->parent
;
4661 ret
= (cgrp
== target
);
4665 static void check_for_release(struct cgroup
*cgrp
)
4667 /* All of these checks rely on RCU to keep the cgroup
4668 * structure alive */
4669 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
4670 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
4671 /* Control Group is currently removeable. If it's not
4672 * already queued for a userspace notification, queue
4674 int need_schedule_work
= 0;
4675 spin_lock(&release_list_lock
);
4676 if (!cgroup_is_removed(cgrp
) &&
4677 list_empty(&cgrp
->release_list
)) {
4678 list_add(&cgrp
->release_list
, &release_list
);
4679 need_schedule_work
= 1;
4681 spin_unlock(&release_list_lock
);
4682 if (need_schedule_work
)
4683 schedule_work(&release_agent_work
);
4687 /* Caller must verify that the css is not for root cgroup */
4688 void __css_put(struct cgroup_subsys_state
*css
, int count
)
4690 struct cgroup
*cgrp
= css
->cgroup
;
4693 val
= atomic_sub_return(count
, &css
->refcnt
);
4695 if (notify_on_release(cgrp
)) {
4696 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
4697 check_for_release(cgrp
);
4699 cgroup_wakeup_rmdir_waiter(cgrp
);
4702 WARN_ON_ONCE(val
< 1);
4704 EXPORT_SYMBOL_GPL(__css_put
);
4707 * Notify userspace when a cgroup is released, by running the
4708 * configured release agent with the name of the cgroup (path
4709 * relative to the root of cgroup file system) as the argument.
4711 * Most likely, this user command will try to rmdir this cgroup.
4713 * This races with the possibility that some other task will be
4714 * attached to this cgroup before it is removed, or that some other
4715 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
4716 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
4717 * unused, and this cgroup will be reprieved from its death sentence,
4718 * to continue to serve a useful existence. Next time it's released,
4719 * we will get notified again, if it still has 'notify_on_release' set.
4721 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
4722 * means only wait until the task is successfully execve()'d. The
4723 * separate release agent task is forked by call_usermodehelper(),
4724 * then control in this thread returns here, without waiting for the
4725 * release agent task. We don't bother to wait because the caller of
4726 * this routine has no use for the exit status of the release agent
4727 * task, so no sense holding our caller up for that.
4729 static void cgroup_release_agent(struct work_struct
*work
)
4731 BUG_ON(work
!= &release_agent_work
);
4732 mutex_lock(&cgroup_mutex
);
4733 spin_lock(&release_list_lock
);
4734 while (!list_empty(&release_list
)) {
4735 char *argv
[3], *envp
[3];
4737 char *pathbuf
= NULL
, *agentbuf
= NULL
;
4738 struct cgroup
*cgrp
= list_entry(release_list
.next
,
4741 list_del_init(&cgrp
->release_list
);
4742 spin_unlock(&release_list_lock
);
4743 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4746 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
4748 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
4753 argv
[i
++] = agentbuf
;
4754 argv
[i
++] = pathbuf
;
4758 /* minimal command environment */
4759 envp
[i
++] = "HOME=/";
4760 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
4763 /* Drop the lock while we invoke the usermode helper,
4764 * since the exec could involve hitting disk and hence
4765 * be a slow process */
4766 mutex_unlock(&cgroup_mutex
);
4767 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
4768 mutex_lock(&cgroup_mutex
);
4772 spin_lock(&release_list_lock
);
4774 spin_unlock(&release_list_lock
);
4775 mutex_unlock(&cgroup_mutex
);
4778 static int __init
cgroup_disable(char *str
)
4783 while ((token
= strsep(&str
, ",")) != NULL
) {
4787 * cgroup_disable, being at boot time, can't know about module
4788 * subsystems, so we don't worry about them.
4790 for (i
= 0; i
< CGROUP_BUILTIN_SUBSYS_COUNT
; i
++) {
4791 struct cgroup_subsys
*ss
= subsys
[i
];
4793 if (!strcmp(token
, ss
->name
)) {
4795 printk(KERN_INFO
"Disabling %s control group"
4796 " subsystem\n", ss
->name
);
4803 __setup("cgroup_disable=", cgroup_disable
);
4806 * Functons for CSS ID.
4810 *To get ID other than 0, this should be called when !cgroup_is_removed().
4812 unsigned short css_id(struct cgroup_subsys_state
*css
)
4814 struct css_id
*cssid
;
4817 * This css_id() can return correct value when somone has refcnt
4818 * on this or this is under rcu_read_lock(). Once css->id is allocated,
4819 * it's unchanged until freed.
4821 cssid
= rcu_dereference_check(css
->id
,
4822 rcu_read_lock_held() || atomic_read(&css
->refcnt
));
4828 EXPORT_SYMBOL_GPL(css_id
);
4830 unsigned short css_depth(struct cgroup_subsys_state
*css
)
4832 struct css_id
*cssid
;
4834 cssid
= rcu_dereference_check(css
->id
,
4835 rcu_read_lock_held() || atomic_read(&css
->refcnt
));
4838 return cssid
->depth
;
4841 EXPORT_SYMBOL_GPL(css_depth
);
4844 * css_is_ancestor - test "root" css is an ancestor of "child"
4845 * @child: the css to be tested.
4846 * @root: the css supporsed to be an ancestor of the child.
4848 * Returns true if "root" is an ancestor of "child" in its hierarchy. Because
4849 * this function reads css->id, this use rcu_dereference() and rcu_read_lock().
4850 * But, considering usual usage, the csses should be valid objects after test.
4851 * Assuming that the caller will do some action to the child if this returns
4852 * returns true, the caller must take "child";s reference count.
4853 * If "child" is valid object and this returns true, "root" is valid, too.
4856 bool css_is_ancestor(struct cgroup_subsys_state
*child
,
4857 const struct cgroup_subsys_state
*root
)
4859 struct css_id
*child_id
;
4860 struct css_id
*root_id
;
4864 child_id
= rcu_dereference(child
->id
);
4865 root_id
= rcu_dereference(root
->id
);
4868 || (child_id
->depth
< root_id
->depth
)
4869 || (child_id
->stack
[root_id
->depth
] != root_id
->id
))
4875 void free_css_id(struct cgroup_subsys
*ss
, struct cgroup_subsys_state
*css
)
4877 struct css_id
*id
= css
->id
;
4878 /* When this is called before css_id initialization, id can be NULL */
4882 BUG_ON(!ss
->use_id
);
4884 rcu_assign_pointer(id
->css
, NULL
);
4885 rcu_assign_pointer(css
->id
, NULL
);
4886 spin_lock(&ss
->id_lock
);
4887 idr_remove(&ss
->idr
, id
->id
);
4888 spin_unlock(&ss
->id_lock
);
4889 kfree_rcu(id
, rcu_head
);
4891 EXPORT_SYMBOL_GPL(free_css_id
);
4894 * This is called by init or create(). Then, calls to this function are
4895 * always serialized (By cgroup_mutex() at create()).
4898 static struct css_id
*get_new_cssid(struct cgroup_subsys
*ss
, int depth
)
4900 struct css_id
*newid
;
4901 int myid
, error
, size
;
4903 BUG_ON(!ss
->use_id
);
4905 size
= sizeof(*newid
) + sizeof(unsigned short) * (depth
+ 1);
4906 newid
= kzalloc(size
, GFP_KERNEL
);
4908 return ERR_PTR(-ENOMEM
);
4910 if (unlikely(!idr_pre_get(&ss
->idr
, GFP_KERNEL
))) {
4914 spin_lock(&ss
->id_lock
);
4915 /* Don't use 0. allocates an ID of 1-65535 */
4916 error
= idr_get_new_above(&ss
->idr
, newid
, 1, &myid
);
4917 spin_unlock(&ss
->id_lock
);
4919 /* Returns error when there are no free spaces for new ID.*/
4924 if (myid
> CSS_ID_MAX
)
4928 newid
->depth
= depth
;
4932 spin_lock(&ss
->id_lock
);
4933 idr_remove(&ss
->idr
, myid
);
4934 spin_unlock(&ss
->id_lock
);
4937 return ERR_PTR(error
);
4941 static int __init_or_module
cgroup_init_idr(struct cgroup_subsys
*ss
,
4942 struct cgroup_subsys_state
*rootcss
)
4944 struct css_id
*newid
;
4946 spin_lock_init(&ss
->id_lock
);
4949 newid
= get_new_cssid(ss
, 0);
4951 return PTR_ERR(newid
);
4953 newid
->stack
[0] = newid
->id
;
4954 newid
->css
= rootcss
;
4955 rootcss
->id
= newid
;
4959 static int alloc_css_id(struct cgroup_subsys
*ss
, struct cgroup
*parent
,
4960 struct cgroup
*child
)
4962 int subsys_id
, i
, depth
= 0;
4963 struct cgroup_subsys_state
*parent_css
, *child_css
;
4964 struct css_id
*child_id
, *parent_id
;
4966 subsys_id
= ss
->subsys_id
;
4967 parent_css
= parent
->subsys
[subsys_id
];
4968 child_css
= child
->subsys
[subsys_id
];
4969 parent_id
= parent_css
->id
;
4970 depth
= parent_id
->depth
+ 1;
4972 child_id
= get_new_cssid(ss
, depth
);
4973 if (IS_ERR(child_id
))
4974 return PTR_ERR(child_id
);
4976 for (i
= 0; i
< depth
; i
++)
4977 child_id
->stack
[i
] = parent_id
->stack
[i
];
4978 child_id
->stack
[depth
] = child_id
->id
;
4980 * child_id->css pointer will be set after this cgroup is available
4981 * see cgroup_populate_dir()
4983 rcu_assign_pointer(child_css
->id
, child_id
);
4989 * css_lookup - lookup css by id
4990 * @ss: cgroup subsys to be looked into.
4993 * Returns pointer to cgroup_subsys_state if there is valid one with id.
4994 * NULL if not. Should be called under rcu_read_lock()
4996 struct cgroup_subsys_state
*css_lookup(struct cgroup_subsys
*ss
, int id
)
4998 struct css_id
*cssid
= NULL
;
5000 BUG_ON(!ss
->use_id
);
5001 cssid
= idr_find(&ss
->idr
, id
);
5003 if (unlikely(!cssid
))
5006 return rcu_dereference(cssid
->css
);
5008 EXPORT_SYMBOL_GPL(css_lookup
);
5011 * css_get_next - lookup next cgroup under specified hierarchy.
5012 * @ss: pointer to subsystem
5013 * @id: current position of iteration.
5014 * @root: pointer to css. search tree under this.
5015 * @foundid: position of found object.
5017 * Search next css under the specified hierarchy of rootid. Calling under
5018 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
5020 struct cgroup_subsys_state
*
5021 css_get_next(struct cgroup_subsys
*ss
, int id
,
5022 struct cgroup_subsys_state
*root
, int *foundid
)
5024 struct cgroup_subsys_state
*ret
= NULL
;
5027 int rootid
= css_id(root
);
5028 int depth
= css_depth(root
);
5033 BUG_ON(!ss
->use_id
);
5034 /* fill start point for scan */
5038 * scan next entry from bitmap(tree), tmpid is updated after
5041 spin_lock(&ss
->id_lock
);
5042 tmp
= idr_get_next(&ss
->idr
, &tmpid
);
5043 spin_unlock(&ss
->id_lock
);
5047 if (tmp
->depth
>= depth
&& tmp
->stack
[depth
] == rootid
) {
5048 ret
= rcu_dereference(tmp
->css
);
5054 /* continue to scan from next id */
5061 * get corresponding css from file open on cgroupfs directory
5063 struct cgroup_subsys_state
*cgroup_css_from_dir(struct file
*f
, int id
)
5065 struct cgroup
*cgrp
;
5066 struct inode
*inode
;
5067 struct cgroup_subsys_state
*css
;
5069 inode
= f
->f_dentry
->d_inode
;
5070 /* check in cgroup filesystem dir */
5071 if (inode
->i_op
!= &cgroup_dir_inode_operations
)
5072 return ERR_PTR(-EBADF
);
5074 if (id
< 0 || id
>= CGROUP_SUBSYS_COUNT
)
5075 return ERR_PTR(-EINVAL
);
5078 cgrp
= __d_cgrp(f
->f_dentry
);
5079 css
= cgrp
->subsys
[id
];
5080 return css
? css
: ERR_PTR(-ENOENT
);
5083 #ifdef CONFIG_CGROUP_DEBUG
5084 static struct cgroup_subsys_state
*debug_create(struct cgroup_subsys
*ss
,
5085 struct cgroup
*cont
)
5087 struct cgroup_subsys_state
*css
= kzalloc(sizeof(*css
), GFP_KERNEL
);
5090 return ERR_PTR(-ENOMEM
);
5095 static void debug_destroy(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5097 kfree(cont
->subsys
[debug_subsys_id
]);
5100 static u64
cgroup_refcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5102 return atomic_read(&cont
->count
);
5105 static u64
debug_taskcount_read(struct cgroup
*cont
, struct cftype
*cft
)
5107 return cgroup_task_count(cont
);
5110 static u64
current_css_set_read(struct cgroup
*cont
, struct cftype
*cft
)
5112 return (u64
)(unsigned long)current
->cgroups
;
5115 static u64
current_css_set_refcount_read(struct cgroup
*cont
,
5121 count
= atomic_read(¤t
->cgroups
->refcount
);
5126 static int current_css_set_cg_links_read(struct cgroup
*cont
,
5128 struct seq_file
*seq
)
5130 struct cg_cgroup_link
*link
;
5133 read_lock(&css_set_lock
);
5135 cg
= rcu_dereference(current
->cgroups
);
5136 list_for_each_entry(link
, &cg
->cg_links
, cg_link_list
) {
5137 struct cgroup
*c
= link
->cgrp
;
5141 name
= c
->dentry
->d_name
.name
;
5144 seq_printf(seq
, "Root %d group %s\n",
5145 c
->root
->hierarchy_id
, name
);
5148 read_unlock(&css_set_lock
);
5152 #define MAX_TASKS_SHOWN_PER_CSS 25
5153 static int cgroup_css_links_read(struct cgroup
*cont
,
5155 struct seq_file
*seq
)
5157 struct cg_cgroup_link
*link
;
5159 read_lock(&css_set_lock
);
5160 list_for_each_entry(link
, &cont
->css_sets
, cgrp_link_list
) {
5161 struct css_set
*cg
= link
->cg
;
5162 struct task_struct
*task
;
5164 seq_printf(seq
, "css_set %p\n", cg
);
5165 list_for_each_entry(task
, &cg
->tasks
, cg_list
) {
5166 if (count
++ > MAX_TASKS_SHOWN_PER_CSS
) {
5167 seq_puts(seq
, " ...\n");
5170 seq_printf(seq
, " task %d\n",
5171 task_pid_vnr(task
));
5175 read_unlock(&css_set_lock
);
5179 static u64
releasable_read(struct cgroup
*cgrp
, struct cftype
*cft
)
5181 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
5184 static struct cftype debug_files
[] = {
5186 .name
= "cgroup_refcount",
5187 .read_u64
= cgroup_refcount_read
,
5190 .name
= "taskcount",
5191 .read_u64
= debug_taskcount_read
,
5195 .name
= "current_css_set",
5196 .read_u64
= current_css_set_read
,
5200 .name
= "current_css_set_refcount",
5201 .read_u64
= current_css_set_refcount_read
,
5205 .name
= "current_css_set_cg_links",
5206 .read_seq_string
= current_css_set_cg_links_read
,
5210 .name
= "cgroup_css_links",
5211 .read_seq_string
= cgroup_css_links_read
,
5215 .name
= "releasable",
5216 .read_u64
= releasable_read
,
5220 static int debug_populate(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
5222 return cgroup_add_files(cont
, ss
, debug_files
,
5223 ARRAY_SIZE(debug_files
));
5226 struct cgroup_subsys debug_subsys
= {
5228 .create
= debug_create
,
5229 .destroy
= debug_destroy
,
5230 .populate
= debug_populate
,
5231 .subsys_id
= debug_subsys_id
,
5233 #endif /* CONFIG_CGROUP_DEBUG */