2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Copyright notices from the original cpuset code:
8 * --------------------------------------------------
9 * Copyright (C) 2003 BULL SA.
10 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
12 * Portions derived from Patrick Mochel's sysfs code.
13 * sysfs is Copyright (c) 2001-3 Patrick Mochel
15 * 2003-10-10 Written by Simon Derr.
16 * 2003-10-22 Updates by Stephen Hemminger.
17 * 2004 May-July Rework by Paul Jackson.
18 * ---------------------------------------------------
20 * This file is subject to the terms and conditions of the GNU General Public
21 * License. See the file COPYING in the main directory of the Linux
22 * distribution for more details.
25 #include <linux/cgroup.h>
26 #include <linux/errno.h>
28 #include <linux/kernel.h>
29 #include <linux/list.h>
31 #include <linux/mutex.h>
32 #include <linux/mount.h>
33 #include <linux/pagemap.h>
34 #include <linux/proc_fs.h>
35 #include <linux/rcupdate.h>
36 #include <linux/sched.h>
37 #include <linux/backing-dev.h>
38 #include <linux/seq_file.h>
39 #include <linux/slab.h>
40 #include <linux/magic.h>
41 #include <linux/spinlock.h>
42 #include <linux/string.h>
43 #include <linux/sort.h>
44 #include <linux/kmod.h>
45 #include <linux/delayacct.h>
46 #include <linux/cgroupstats.h>
48 #include <asm/atomic.h>
50 static DEFINE_MUTEX(cgroup_mutex
);
52 /* Generate an array of cgroup subsystem pointers */
53 #define SUBSYS(_x) &_x ## _subsys,
55 static struct cgroup_subsys
*subsys
[] = {
56 #include <linux/cgroup_subsys.h>
60 * A cgroupfs_root represents the root of a cgroup hierarchy,
61 * and may be associated with a superblock to form an active
64 struct cgroupfs_root
{
65 struct super_block
*sb
;
68 * The bitmask of subsystems intended to be attached to this
71 unsigned long subsys_bits
;
73 /* The bitmask of subsystems currently attached to this hierarchy */
74 unsigned long actual_subsys_bits
;
76 /* A list running through the attached subsystems */
77 struct list_head subsys_list
;
79 /* The root cgroup for this hierarchy */
80 struct cgroup top_cgroup
;
82 /* Tracks how many cgroups are currently defined in hierarchy.*/
83 int number_of_cgroups
;
85 /* A list running through the mounted hierarchies */
86 struct list_head root_list
;
88 /* Hierarchy-specific flags */
91 /* The path to use for release notifications. No locking
92 * between setting and use - so if userspace updates this
93 * while child cgroups exist, you could miss a
94 * notification. We ensure that it's always a valid
95 * NUL-terminated string */
96 char release_agent_path
[PATH_MAX
];
101 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
102 * subsystems that are otherwise unattached - it never has more than a
103 * single cgroup, and all tasks are part of that cgroup.
105 static struct cgroupfs_root rootnode
;
107 /* The list of hierarchy roots */
109 static LIST_HEAD(roots
);
110 static int root_count
;
112 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
113 #define dummytop (&rootnode.top_cgroup)
115 /* This flag indicates whether tasks in the fork and exit paths should
116 <<<<<<< HEAD:kernel/cgroup.c
117 * take callback_mutex and check for fork/exit handlers to call. This
118 * avoids us having to do extra work in the fork/exit path if none of the
119 * subsystems need to be called.
121 * check for fork/exit handlers to call. This avoids us having to do
122 * extra work in the fork/exit path if none of the subsystems need to
124 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
126 static int need_forkexit_callback
;
128 /* bits in struct cgroup flags field */
130 /* Control Group is dead */
132 /* Control Group has previously had a child cgroup or a task,
133 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
135 /* Control Group requires release notifications to userspace */
136 CGRP_NOTIFY_ON_RELEASE
,
139 /* convenient tests for these bits */
140 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
142 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
145 /* bits in struct cgroupfs_root flags field */
147 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
150 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
153 (1 << CGRP_RELEASABLE
) |
154 (1 << CGRP_NOTIFY_ON_RELEASE
);
155 return (cgrp
->flags
& bits
) == bits
;
158 static int notify_on_release(const struct cgroup
*cgrp
)
160 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
164 * for_each_subsys() allows you to iterate on each subsystem attached to
165 * an active hierarchy
167 #define for_each_subsys(_root, _ss) \
168 list_for_each_entry(_ss, &_root->subsys_list, sibling)
170 /* for_each_root() allows you to iterate across the active hierarchies */
171 #define for_each_root(_root) \
172 list_for_each_entry(_root, &roots, root_list)
174 /* the list of cgroups eligible for automatic release. Protected by
175 * release_list_lock */
176 static LIST_HEAD(release_list
);
177 static DEFINE_SPINLOCK(release_list_lock
);
178 static void cgroup_release_agent(struct work_struct
*work
);
179 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
180 static void check_for_release(struct cgroup
*cgrp
);
182 /* Link structure for associating css_set objects with cgroups */
183 struct cg_cgroup_link
{
185 * List running through cg_cgroup_links associated with a
186 * cgroup, anchored on cgroup->css_sets
188 struct list_head cgrp_link_list
;
190 * List running through cg_cgroup_links pointing at a
191 * single css_set object, anchored on css_set->cg_links
193 struct list_head cg_link_list
;
197 /* The default css_set - used by init and its children prior to any
198 * hierarchies being mounted. It contains a pointer to the root state
199 * for each subsystem. Also used to anchor the list of css_sets. Not
200 * reference-counted, to improve performance when child cgroups
201 * haven't been created.
204 static struct css_set init_css_set
;
205 static struct cg_cgroup_link init_css_set_link
;
207 /* css_set_lock protects the list of css_set objects, and the
208 * chain of tasks off each css_set. Nests outside task->alloc_lock
209 * due to cgroup_iter_start() */
210 static DEFINE_RWLOCK(css_set_lock
);
211 static int css_set_count
;
213 /* We don't maintain the lists running through each css_set to its
214 * task until after the first call to cgroup_iter_start(). This
215 * reduces the fork()/exit() overhead for people who have cgroups
216 * compiled into their kernel but not actually in use */
217 static int use_task_css_set_links
;
219 /* When we create or destroy a css_set, the operation simply
220 * takes/releases a reference count on all the cgroups referenced
221 * by subsystems in this css_set. This can end up multiple-counting
222 * some cgroups, but that's OK - the ref-count is just a
223 * busy/not-busy indicator; ensuring that we only count each cgroup
224 * once would require taking a global lock to ensure that no
225 * subsystems moved between hierarchies while we were doing so.
227 * Possible TODO: decide at boot time based on the number of
228 * registered subsystems and the number of CPUs or NUMA nodes whether
229 * it's better for performance to ref-count every subsystem, or to
230 * take a global lock and only add one ref count to each hierarchy.
234 * unlink a css_set from the list and free it
236 static void unlink_css_set(struct css_set
*cg
)
238 write_lock(&css_set_lock
);
241 while (!list_empty(&cg
->cg_links
)) {
242 struct cg_cgroup_link
*link
;
243 link
= list_entry(cg
->cg_links
.next
,
244 struct cg_cgroup_link
, cg_link_list
);
245 list_del(&link
->cg_link_list
);
246 list_del(&link
->cgrp_link_list
);
249 write_unlock(&css_set_lock
);
252 static void __release_css_set(struct kref
*k
, int taskexit
)
255 struct css_set
*cg
= container_of(k
, struct css_set
, ref
);
260 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
261 struct cgroup
*cgrp
= cg
->subsys
[i
]->cgroup
;
262 if (atomic_dec_and_test(&cgrp
->count
) &&
263 notify_on_release(cgrp
)) {
265 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
266 check_for_release(cgrp
);
273 static void release_css_set(struct kref
*k
)
275 __release_css_set(k
, 0);
278 static void release_css_set_taskexit(struct kref
*k
)
280 __release_css_set(k
, 1);
284 * refcounted get/put for css_set objects
286 static inline void get_css_set(struct css_set
*cg
)
291 static inline void put_css_set(struct css_set
*cg
)
293 kref_put(&cg
->ref
, release_css_set
);
296 static inline void put_css_set_taskexit(struct css_set
*cg
)
298 kref_put(&cg
->ref
, release_css_set_taskexit
);
302 * find_existing_css_set() is a helper for
303 * find_css_set(), and checks to see whether an existing
304 * css_set is suitable. This currently walks a linked-list for
305 * simplicity; a later patch will use a hash table for better
308 * oldcg: the cgroup group that we're using before the cgroup
311 * cgrp: the cgroup that we're moving into
313 * template: location in which to build the desired set of subsystem
314 * state objects for the new cgroup group
316 <<<<<<< HEAD
:kernel
/cgroup
.c
319 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
320 static struct css_set
*find_existing_css_set(
321 struct css_set
*oldcg
,
323 struct cgroup_subsys_state
*template[])
326 struct cgroupfs_root
*root
= cgrp
->root
;
327 struct list_head
*l
= &init_css_set
.list
;
329 /* Built the set of subsystem state objects that we want to
330 * see in the new css_set */
331 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
332 <<<<<<< HEAD
:kernel
/cgroup
.c
333 if (root
->subsys_bits
& (1ull << i
)) {
335 if (root
->subsys_bits
& (1UL << i
)) {
336 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
337 /* Subsystem is in this hierarchy. So we want
338 * the subsystem state from the new
340 template[i
] = cgrp
->subsys
[i
];
342 /* Subsystem is not in this hierarchy, so we
343 * don't want to change the subsystem state */
344 template[i
] = oldcg
->subsys
[i
];
348 /* Look through existing cgroup groups to find one to reuse */
351 list_entry(l
, struct css_set
, list
);
353 if (!memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
354 /* All subsystems matched */
357 /* Try the next cgroup group */
359 } while (l
!= &init_css_set
.list
);
361 /* No existing cgroup group matched */
366 * allocate_cg_links() allocates "count" cg_cgroup_link structures
367 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
368 * success or a negative error
370 <<<<<<< HEAD
:kernel
/cgroup
.c
373 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
374 static int allocate_cg_links(int count
, struct list_head
*tmp
)
376 struct cg_cgroup_link
*link
;
379 for (i
= 0; i
< count
; i
++) {
380 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
382 while (!list_empty(tmp
)) {
383 link
= list_entry(tmp
->next
,
384 struct cg_cgroup_link
,
386 list_del(&link
->cgrp_link_list
);
391 list_add(&link
->cgrp_link_list
, tmp
);
396 static void free_cg_links(struct list_head
*tmp
)
398 while (!list_empty(tmp
)) {
399 struct cg_cgroup_link
*link
;
400 link
= list_entry(tmp
->next
,
401 struct cg_cgroup_link
,
403 list_del(&link
->cgrp_link_list
);
409 * find_css_set() takes an existing cgroup group and a
410 * cgroup object, and returns a css_set object that's
411 * equivalent to the old group, but with the given cgroup
412 * substituted into the appropriate hierarchy. Must be called with
415 <<<<<<< HEAD
:kernel
/cgroup
.c
418 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
419 static struct css_set
*find_css_set(
420 struct css_set
*oldcg
, struct cgroup
*cgrp
)
423 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
426 struct list_head tmp_cg_links
;
427 struct cg_cgroup_link
*link
;
429 /* First see if we already have a cgroup group that matches
431 write_lock(&css_set_lock
);
432 res
= find_existing_css_set(oldcg
, cgrp
, template);
435 write_unlock(&css_set_lock
);
440 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
444 /* Allocate all the cg_cgroup_link objects that we'll need */
445 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
450 kref_init(&res
->ref
);
451 INIT_LIST_HEAD(&res
->cg_links
);
452 INIT_LIST_HEAD(&res
->tasks
);
454 /* Copy the set of subsystem state objects generated in
455 * find_existing_css_set() */
456 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
458 write_lock(&css_set_lock
);
459 /* Add reference counts and links from the new css_set. */
460 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
461 struct cgroup
*cgrp
= res
->subsys
[i
]->cgroup
;
462 struct cgroup_subsys
*ss
= subsys
[i
];
463 atomic_inc(&cgrp
->count
);
465 * We want to add a link once per cgroup, so we
466 * only do it for the first subsystem in each
469 if (ss
->root
->subsys_list
.next
== &ss
->sibling
) {
470 BUG_ON(list_empty(&tmp_cg_links
));
471 link
= list_entry(tmp_cg_links
.next
,
472 struct cg_cgroup_link
,
474 list_del(&link
->cgrp_link_list
);
475 list_add(&link
->cgrp_link_list
, &cgrp
->css_sets
);
477 list_add(&link
->cg_link_list
, &res
->cg_links
);
480 if (list_empty(&rootnode
.subsys_list
)) {
481 link
= list_entry(tmp_cg_links
.next
,
482 struct cg_cgroup_link
,
484 list_del(&link
->cgrp_link_list
);
485 list_add(&link
->cgrp_link_list
, &dummytop
->css_sets
);
487 list_add(&link
->cg_link_list
, &res
->cg_links
);
490 BUG_ON(!list_empty(&tmp_cg_links
));
492 /* Link this cgroup group into the list */
493 list_add(&res
->list
, &init_css_set
.list
);
495 <<<<<<< HEAD
:kernel
/cgroup
.c
496 INIT_LIST_HEAD(&res
->tasks
);
498 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
499 write_unlock(&css_set_lock
);
505 * There is one global cgroup mutex. We also require taking
506 * task_lock() when dereferencing a task's cgroup subsys pointers.
507 * See "The task_lock() exception", at the end of this comment.
509 * A task must hold cgroup_mutex to modify cgroups.
511 * Any task can increment and decrement the count field without lock.
512 * So in general, code holding cgroup_mutex can't rely on the count
513 * field not changing. However, if the count goes to zero, then only
514 * cgroup_attach_task() can increment it again. Because a count of zero
515 * means that no tasks are currently attached, therefore there is no
516 * way a task attached to that cgroup can fork (the other way to
517 * increment the count). So code holding cgroup_mutex can safely
518 * assume that if the count is zero, it will stay zero. Similarly, if
519 * a task holds cgroup_mutex on a cgroup with zero count, it
520 * knows that the cgroup won't be removed, as cgroup_rmdir()
523 * The cgroup_common_file_write handler for operations that modify
524 * the cgroup hierarchy holds cgroup_mutex across the entire operation,
525 * single threading all such cgroup modifications across the system.
527 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
528 * (usually) take cgroup_mutex. These are the two most performance
529 * critical pieces of code here. The exception occurs on cgroup_exit(),
530 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
531 * is taken, and if the cgroup count is zero, a usermode call made
532 <<<<<<< HEAD:kernel/cgroup.c
533 * to /sbin/cgroup_release_agent with the name of the cgroup (path
534 * relative to the root of cgroup file system) as the argument.
536 * to the release agent with the name of the cgroup (path relative to
537 * the root of cgroup file system) as the argument.
538 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
540 * A cgroup can only be deleted if both its 'count' of using tasks
541 * is zero, and its list of 'children' cgroups is empty. Since all
542 * tasks in the system use _some_ cgroup, and since there is always at
543 * least one task in the system (init, pid == 1), therefore, top_cgroup
544 * always has either children cgroups and/or using tasks. So we don't
545 * need a special hack to ensure that top_cgroup cannot be deleted.
547 * The task_lock() exception
549 * The need for this exception arises from the action of
550 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
551 <<<<<<< HEAD:kernel/cgroup.c
552 * another. It does so using cgroup_mutexe, however there are
554 * another. It does so using cgroup_mutex, however there are
555 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
556 * several performance critical places that need to reference
557 * task->cgroup without the expense of grabbing a system global
558 * mutex. Therefore except as noted below, when dereferencing or, as
559 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
560 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
561 * the task_struct routinely used for such matters.
563 * P.S. One more locking exception. RCU is used to guard the
564 * update of a tasks cgroup pointer by cgroup_attach_task()
568 * cgroup_lock - lock out any changes to cgroup structures
571 <<<<<<< HEAD
:kernel
/cgroup
.c
574 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
575 void cgroup_lock(void)
577 mutex_lock(&cgroup_mutex
);
581 * cgroup_unlock - release lock on cgroup changes
583 * Undo the lock taken in a previous cgroup_lock() call.
585 <<<<<<< HEAD
:kernel
/cgroup
.c
588 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
589 void cgroup_unlock(void)
591 mutex_unlock(&cgroup_mutex
);
595 * A couple of forward declarations required, due to cyclic reference loop:
596 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
597 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
601 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
);
602 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
603 static int cgroup_populate_dir(struct cgroup
*cgrp
);
604 static struct inode_operations cgroup_dir_inode_operations
;
605 static struct file_operations proc_cgroupstats_operations
;
607 static struct backing_dev_info cgroup_backing_dev_info
= {
608 .capabilities
= BDI_CAP_NO_ACCT_DIRTY
| BDI_CAP_NO_WRITEBACK
,
611 static struct inode
*cgroup_new_inode(mode_t mode
, struct super_block
*sb
)
613 struct inode
*inode
= new_inode(sb
);
616 inode
->i_mode
= mode
;
617 inode
->i_uid
= current
->fsuid
;
618 inode
->i_gid
= current
->fsgid
;
620 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
621 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
627 * Call subsys's pre_destroy handler.
628 * This is called before css refcnt check.
630 <<<<<<< HEAD
:kernel
/cgroup
.c
633 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
634 static void cgroup_call_pre_destroy(struct cgroup
*cgrp
)
636 struct cgroup_subsys
*ss
;
637 for_each_subsys(cgrp
->root
, ss
)
638 if (ss
->pre_destroy
&& cgrp
->subsys
[ss
->subsys_id
])
639 ss
->pre_destroy(ss
, cgrp
);
643 <<<<<<< HEAD
:kernel
/cgroup
.c
646 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
647 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
649 /* is dentry a directory ? if so, kfree() associated cgroup */
650 if (S_ISDIR(inode
->i_mode
)) {
651 struct cgroup
*cgrp
= dentry
->d_fsdata
;
652 struct cgroup_subsys
*ss
;
653 BUG_ON(!(cgroup_is_removed(cgrp
)));
654 /* It's possible for external users to be holding css
655 * reference counts on a cgroup; css_put() needs to
656 * be able to access the cgroup after decrementing
657 * the reference count in order to know if it needs to
658 * queue the cgroup to be handled by the release
662 mutex_lock(&cgroup_mutex
);
664 * Release the subsystem state objects.
666 for_each_subsys(cgrp
->root
, ss
) {
667 if (cgrp
->subsys
[ss
->subsys_id
])
668 ss
->destroy(ss
, cgrp
);
671 cgrp
->root
->number_of_cgroups
--;
672 mutex_unlock(&cgroup_mutex
);
674 /* Drop the active superblock reference that we took when we
675 * created the cgroup */
676 deactivate_super(cgrp
->root
->sb
);
683 static void remove_dir(struct dentry
*d
)
685 struct dentry
*parent
= dget(d
->d_parent
);
688 simple_rmdir(parent
->d_inode
, d
);
692 static void cgroup_clear_directory(struct dentry
*dentry
)
694 struct list_head
*node
;
696 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
697 spin_lock(&dcache_lock
);
698 node
= dentry
->d_subdirs
.next
;
699 while (node
!= &dentry
->d_subdirs
) {
700 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
703 /* This should never be called on a cgroup
704 * directory with child cgroups */
705 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
707 spin_unlock(&dcache_lock
);
709 simple_unlink(dentry
->d_inode
, d
);
711 spin_lock(&dcache_lock
);
713 node
= dentry
->d_subdirs
.next
;
715 spin_unlock(&dcache_lock
);
719 * NOTE : the dentry must have been dget()'ed
721 static void cgroup_d_remove_dir(struct dentry
*dentry
)
723 cgroup_clear_directory(dentry
);
725 spin_lock(&dcache_lock
);
726 list_del_init(&dentry
->d_u
.d_child
);
727 spin_unlock(&dcache_lock
);
731 static int rebind_subsystems(struct cgroupfs_root
*root
,
732 unsigned long final_bits
)
734 unsigned long added_bits
, removed_bits
;
735 struct cgroup
*cgrp
= &root
->top_cgroup
;
738 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
739 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
740 /* Check that any added subsystems are currently free */
741 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
742 <<<<<<< HEAD
:kernel
/cgroup
.c
743 unsigned long long bit
= 1ull << i
;
745 unsigned long bit
= 1UL << i
;
746 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
747 struct cgroup_subsys
*ss
= subsys
[i
];
748 if (!(bit
& added_bits
))
750 if (ss
->root
!= &rootnode
) {
751 /* Subsystem isn't free */
756 /* Currently we don't handle adding/removing subsystems when
757 * any child cgroups exist. This is theoretically supportable
758 * but involves complex error handling, so it's being left until
760 if (!list_empty(&cgrp
->children
))
763 /* Process each subsystem */
764 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
765 struct cgroup_subsys
*ss
= subsys
[i
];
766 unsigned long bit
= 1UL << i
;
767 if (bit
& added_bits
) {
768 /* We're binding this subsystem to this hierarchy */
769 BUG_ON(cgrp
->subsys
[i
]);
770 BUG_ON(!dummytop
->subsys
[i
]);
771 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
772 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
773 cgrp
->subsys
[i
]->cgroup
= cgrp
;
774 list_add(&ss
->sibling
, &root
->subsys_list
);
775 rcu_assign_pointer(ss
->root
, root
);
779 } else if (bit
& removed_bits
) {
780 /* We're removing this subsystem */
781 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
782 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
784 ss
->bind(ss
, dummytop
);
785 dummytop
->subsys
[i
]->cgroup
= dummytop
;
786 cgrp
->subsys
[i
] = NULL
;
787 rcu_assign_pointer(subsys
[i
]->root
, &rootnode
);
788 list_del(&ss
->sibling
);
789 } else if (bit
& final_bits
) {
790 /* Subsystem state should already exist */
791 BUG_ON(!cgrp
->subsys
[i
]);
793 /* Subsystem state shouldn't exist */
794 BUG_ON(cgrp
->subsys
[i
]);
797 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
803 static int cgroup_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
805 struct cgroupfs_root
*root
= vfs
->mnt_sb
->s_fs_info
;
806 struct cgroup_subsys
*ss
;
808 mutex_lock(&cgroup_mutex
);
809 for_each_subsys(root
, ss
)
810 seq_printf(seq
, ",%s", ss
->name
);
811 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
812 seq_puts(seq
, ",noprefix");
813 if (strlen(root
->release_agent_path
))
814 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
815 mutex_unlock(&cgroup_mutex
);
819 struct cgroup_sb_opts
{
820 unsigned long subsys_bits
;
825 /* Convert a hierarchy specifier into a bitmask of subsystems and
827 static int parse_cgroupfs_options(char *data
,
828 struct cgroup_sb_opts
*opts
)
830 char *token
, *o
= data
?: "all";
832 opts
->subsys_bits
= 0;
834 opts
->release_agent
= NULL
;
836 while ((token
= strsep(&o
, ",")) != NULL
) {
839 if (!strcmp(token
, "all")) {
840 opts
->subsys_bits
= (1 << CGROUP_SUBSYS_COUNT
) - 1;
841 } else if (!strcmp(token
, "noprefix")) {
842 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
843 } else if (!strncmp(token
, "release_agent=", 14)) {
844 /* Specifying two release agents is forbidden */
845 if (opts
->release_agent
)
847 opts
->release_agent
= kzalloc(PATH_MAX
, GFP_KERNEL
);
848 if (!opts
->release_agent
)
850 strncpy(opts
->release_agent
, token
+ 14, PATH_MAX
- 1);
851 opts
->release_agent
[PATH_MAX
- 1] = 0;
853 struct cgroup_subsys
*ss
;
855 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
857 if (!strcmp(token
, ss
->name
)) {
858 set_bit(i
, &opts
->subsys_bits
);
862 if (i
== CGROUP_SUBSYS_COUNT
)
867 /* We can't have an empty hierarchy */
868 if (!opts
->subsys_bits
)
874 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
877 struct cgroupfs_root
*root
= sb
->s_fs_info
;
878 struct cgroup
*cgrp
= &root
->top_cgroup
;
879 struct cgroup_sb_opts opts
;
881 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
882 mutex_lock(&cgroup_mutex
);
884 /* See what subsystems are wanted */
885 ret
= parse_cgroupfs_options(data
, &opts
);
889 /* Don't allow flags to change at remount */
890 if (opts
.flags
!= root
->flags
) {
895 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
897 /* (re)populate subsystem files */
899 cgroup_populate_dir(cgrp
);
901 if (opts
.release_agent
)
902 strcpy(root
->release_agent_path
, opts
.release_agent
);
904 if (opts
.release_agent
)
905 kfree(opts
.release_agent
);
906 mutex_unlock(&cgroup_mutex
);
907 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
911 static struct super_operations cgroup_ops
= {
912 .statfs
= simple_statfs
,
913 .drop_inode
= generic_delete_inode
,
914 .show_options
= cgroup_show_options
,
915 .remount_fs
= cgroup_remount
,
918 static void init_cgroup_root(struct cgroupfs_root
*root
)
920 struct cgroup
*cgrp
= &root
->top_cgroup
;
921 INIT_LIST_HEAD(&root
->subsys_list
);
922 INIT_LIST_HEAD(&root
->root_list
);
923 root
->number_of_cgroups
= 1;
925 cgrp
->top_cgroup
= cgrp
;
926 INIT_LIST_HEAD(&cgrp
->sibling
);
927 INIT_LIST_HEAD(&cgrp
->children
);
928 INIT_LIST_HEAD(&cgrp
->css_sets
);
929 INIT_LIST_HEAD(&cgrp
->release_list
);
932 static int cgroup_test_super(struct super_block
*sb
, void *data
)
934 struct cgroupfs_root
*new = data
;
935 struct cgroupfs_root
*root
= sb
->s_fs_info
;
937 /* First check subsystems */
938 if (new->subsys_bits
!= root
->subsys_bits
)
941 /* Next check flags */
942 if (new->flags
!= root
->flags
)
948 static int cgroup_set_super(struct super_block
*sb
, void *data
)
951 struct cgroupfs_root
*root
= data
;
953 ret
= set_anon_super(sb
, NULL
);
957 sb
->s_fs_info
= root
;
960 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
961 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
962 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
963 sb
->s_op
= &cgroup_ops
;
968 static int cgroup_get_rootdir(struct super_block
*sb
)
970 struct inode
*inode
=
971 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
972 struct dentry
*dentry
;
977 <<<<<<< HEAD
:kernel
/cgroup
.c
978 inode
->i_op
= &simple_dir_inode_operations
;
980 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
981 inode
->i_fop
= &simple_dir_operations
;
982 inode
->i_op
= &cgroup_dir_inode_operations
;
983 /* directories start off with i_nlink == 2 (for "." entry) */
985 dentry
= d_alloc_root(inode
);
994 static int cgroup_get_sb(struct file_system_type
*fs_type
,
995 int flags
, const char *unused_dev_name
,
996 void *data
, struct vfsmount
*mnt
)
998 struct cgroup_sb_opts opts
;
1000 struct super_block
*sb
;
1001 struct cgroupfs_root
*root
;
1002 struct list_head tmp_cg_links
, *l
;
1003 INIT_LIST_HEAD(&tmp_cg_links
);
1005 /* First find the desired set of subsystems */
1006 ret
= parse_cgroupfs_options(data
, &opts
);
1008 if (opts
.release_agent
)
1009 kfree(opts
.release_agent
);
1013 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
1014 <<<<<<< HEAD
:kernel
/cgroup
.c
1018 if (opts
.release_agent
)
1019 kfree(opts
.release_agent
);
1020 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
1022 <<<<<<< HEAD
:kernel
/cgroup
.c
1025 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
1027 init_cgroup_root(root
);
1028 root
->subsys_bits
= opts
.subsys_bits
;
1029 root
->flags
= opts
.flags
;
1030 if (opts
.release_agent
) {
1031 strcpy(root
->release_agent_path
, opts
.release_agent
);
1032 kfree(opts
.release_agent
);
1035 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, root
);
1042 if (sb
->s_fs_info
!= root
) {
1043 /* Reusing an existing superblock */
1044 BUG_ON(sb
->s_root
== NULL
);
1048 /* New superblock */
1049 struct cgroup
*cgrp
= &root
->top_cgroup
;
1050 struct inode
*inode
;
1052 BUG_ON(sb
->s_root
!= NULL
);
1054 ret
= cgroup_get_rootdir(sb
);
1056 goto drop_new_super
;
1057 inode
= sb
->s_root
->d_inode
;
1059 mutex_lock(&inode
->i_mutex
);
1060 mutex_lock(&cgroup_mutex
);
1063 * We're accessing css_set_count without locking
1064 * css_set_lock here, but that's OK - it can only be
1065 * increased by someone holding cgroup_lock, and
1066 * that's us. The worst that can happen is that we
1067 * have some link structures left over
1069 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1071 mutex_unlock(&cgroup_mutex
);
1072 mutex_unlock(&inode
->i_mutex
);
1073 goto drop_new_super
;
1076 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1077 if (ret
== -EBUSY
) {
1078 mutex_unlock(&cgroup_mutex
);
1079 mutex_unlock(&inode
->i_mutex
);
1080 goto drop_new_super
;
1083 /* EBUSY should be the only error here */
1086 list_add(&root
->root_list
, &roots
);
1089 sb
->s_root
->d_fsdata
= &root
->top_cgroup
;
1090 root
->top_cgroup
.dentry
= sb
->s_root
;
1092 /* Link the top cgroup in this hierarchy into all
1093 * the css_set objects */
1094 write_lock(&css_set_lock
);
1095 l
= &init_css_set
.list
;
1098 struct cg_cgroup_link
*link
;
1099 cg
= list_entry(l
, struct css_set
, list
);
1100 BUG_ON(list_empty(&tmp_cg_links
));
1101 link
= list_entry(tmp_cg_links
.next
,
1102 struct cg_cgroup_link
,
1104 list_del(&link
->cgrp_link_list
);
1106 list_add(&link
->cgrp_link_list
,
1107 &root
->top_cgroup
.css_sets
);
1108 list_add(&link
->cg_link_list
, &cg
->cg_links
);
1110 } while (l
!= &init_css_set
.list
);
1111 write_unlock(&css_set_lock
);
1113 free_cg_links(&tmp_cg_links
);
1115 BUG_ON(!list_empty(&cgrp
->sibling
));
1116 BUG_ON(!list_empty(&cgrp
->children
));
1117 BUG_ON(root
->number_of_cgroups
!= 1);
1119 cgroup_populate_dir(cgrp
);
1120 mutex_unlock(&inode
->i_mutex
);
1121 mutex_unlock(&cgroup_mutex
);
1124 return simple_set_mnt(mnt
, sb
);
1127 up_write(&sb
->s_umount
);
1128 deactivate_super(sb
);
1129 free_cg_links(&tmp_cg_links
);
1133 static void cgroup_kill_sb(struct super_block
*sb
) {
1134 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1135 struct cgroup
*cgrp
= &root
->top_cgroup
;
1140 BUG_ON(root
->number_of_cgroups
!= 1);
1141 BUG_ON(!list_empty(&cgrp
->children
));
1142 BUG_ON(!list_empty(&cgrp
->sibling
));
1144 mutex_lock(&cgroup_mutex
);
1146 /* Rebind all subsystems back to the default hierarchy */
1147 ret
= rebind_subsystems(root
, 0);
1148 /* Shouldn't be able to fail ... */
1152 * Release all the links from css_sets to this hierarchy's
1155 write_lock(&css_set_lock
);
1156 while (!list_empty(&cgrp
->css_sets
)) {
1157 struct cg_cgroup_link
*link
;
1158 link
= list_entry(cgrp
->css_sets
.next
,
1159 struct cg_cgroup_link
, cgrp_link_list
);
1160 list_del(&link
->cg_link_list
);
1161 list_del(&link
->cgrp_link_list
);
1164 write_unlock(&css_set_lock
);
1166 if (!list_empty(&root
->root_list
)) {
1167 list_del(&root
->root_list
);
1170 mutex_unlock(&cgroup_mutex
);
1173 kill_litter_super(sb
);
1176 static struct file_system_type cgroup_fs_type
= {
1178 .get_sb
= cgroup_get_sb
,
1179 .kill_sb
= cgroup_kill_sb
,
1182 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1184 return dentry
->d_fsdata
;
1187 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1189 return dentry
->d_fsdata
;
1192 <<<<<<< HEAD
:kernel
/cgroup
.c
1194 * Called with cgroup_mutex held. Writes path of cgroup into buf.
1197 * cgroup_path - generate the path of a cgroup
1198 * @cgrp: the cgroup in question
1199 * @buf: the buffer to write the path into
1200 * @buflen: the length of the buffer
1202 * Called with cgroup_mutex held. Writes path of cgroup into buf.
1203 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
1204 * Returns 0 on success, -errno on error.
1206 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1210 if (cgrp
== dummytop
) {
1212 * Inactive subsystems have no dentry for their root
1219 start
= buf
+ buflen
;
1223 int len
= cgrp
->dentry
->d_name
.len
;
1224 if ((start
-= len
) < buf
)
1225 return -ENAMETOOLONG
;
1226 memcpy(start
, cgrp
->dentry
->d_name
.name
, len
);
1227 cgrp
= cgrp
->parent
;
1233 return -ENAMETOOLONG
;
1236 memmove(buf
, start
, buf
+ buflen
- start
);
1241 * Return the first subsystem attached to a cgroup's hierarchy, and
1245 static void get_first_subsys(const struct cgroup
*cgrp
,
1246 struct cgroup_subsys_state
**css
, int *subsys_id
)
1248 const struct cgroupfs_root
*root
= cgrp
->root
;
1249 const struct cgroup_subsys
*test_ss
;
1250 BUG_ON(list_empty(&root
->subsys_list
));
1251 test_ss
= list_entry(root
->subsys_list
.next
,
1252 struct cgroup_subsys
, sibling
);
1254 *css
= cgrp
->subsys
[test_ss
->subsys_id
];
1258 *subsys_id
= test_ss
->subsys_id
;
1261 <<<<<<< HEAD
:kernel
/cgroup
.c
1263 * Attach task 'tsk' to cgroup 'cgrp'
1266 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1267 * @cgrp: the cgroup the task is attaching to
1268 * @tsk: the task to be attached
1269 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
1271 <<<<<<< HEAD:kernel/cgroup.c
1272 * Call holding cgroup_mutex. May take task_lock of
1273 * the task 'pid' during call.
1275 * Call holding cgroup_mutex. May take task_lock of
1276 * the task 'tsk' during call.
1277 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
1279 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1282 struct cgroup_subsys
*ss
;
1283 struct cgroup
*oldcgrp
;
1284 struct css_set
*cg
= tsk
->cgroups
;
1285 struct css_set
*newcg
;
1286 struct cgroupfs_root
*root
= cgrp
->root
;
1289 get_first_subsys(cgrp
, NULL
, &subsys_id
);
1291 /* Nothing to do if the task is already in that cgroup */
1292 oldcgrp
= task_cgroup(tsk
, subsys_id
);
1293 if (cgrp
== oldcgrp
)
1296 for_each_subsys(root
, ss
) {
1297 if (ss
->can_attach
) {
1298 retval
= ss
->can_attach(ss
, cgrp
, tsk
);
1305 * Locate or allocate a new css_set for this task,
1306 * based on its final set of cgroups
1308 newcg
= find_css_set(cg
, cgrp
);
1313 if (tsk
->flags
& PF_EXITING
) {
1318 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1321 /* Update the css_set linked lists if we're using them */
1322 write_lock(&css_set_lock
);
1323 if (!list_empty(&tsk
->cg_list
)) {
1324 list_del(&tsk
->cg_list
);
1325 list_add(&tsk
->cg_list
, &newcg
->tasks
);
1327 write_unlock(&css_set_lock
);
1329 for_each_subsys(root
, ss
) {
1331 ss
->attach(ss
, cgrp
, oldcgrp
, tsk
);
1333 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1340 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with
1341 * cgroup_mutex, may take task_lock of task
1343 static int attach_task_by_pid(struct cgroup
*cgrp
, char *pidbuf
)
1346 struct task_struct
*tsk
;
1349 if (sscanf(pidbuf
, "%d", &pid
) != 1)
1354 tsk
= find_task_by_vpid(pid
);
1355 if (!tsk
|| tsk
->flags
& PF_EXITING
) {
1359 get_task_struct(tsk
);
1362 if ((current
->euid
) && (current
->euid
!= tsk
->uid
)
1363 && (current
->euid
!= tsk
->suid
)) {
1364 put_task_struct(tsk
);
1369 get_task_struct(tsk
);
1372 ret
= cgroup_attach_task(cgrp
, tsk
);
1373 put_task_struct(tsk
);
1377 /* The various types of files and directories in a cgroup file system */
1378 <<<<<<< HEAD
:kernel
/cgroup
.c
1381 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
1382 enum cgroup_filetype
{
1386 FILE_NOTIFY_ON_RELEASE
,
1391 static ssize_t
cgroup_write_uint(struct cgroup
*cgrp
, struct cftype
*cft
,
1393 const char __user
*userbuf
,
1394 size_t nbytes
, loff_t
*unused_ppos
)
1403 if (nbytes
>= sizeof(buffer
))
1405 if (copy_from_user(buffer
, userbuf
, nbytes
))
1408 buffer
[nbytes
] = 0; /* nul-terminate */
1410 /* strip newline if necessary */
1411 if (nbytes
&& (buffer
[nbytes
-1] == '\n'))
1412 buffer
[nbytes
-1] = 0;
1413 val
= simple_strtoull(buffer
, &end
, 0);
1417 /* Pass to subsystem */
1418 retval
= cft
->write_uint(cgrp
, cft
, val
);
1424 static ssize_t
cgroup_common_file_write(struct cgroup
*cgrp
,
1427 const char __user
*userbuf
,
1428 size_t nbytes
, loff_t
*unused_ppos
)
1430 enum cgroup_filetype type
= cft
->private;
1434 if (nbytes
>= PATH_MAX
)
1437 /* +1 for nul-terminator */
1438 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
1442 if (copy_from_user(buffer
, userbuf
, nbytes
)) {
1446 buffer
[nbytes
] = 0; /* nul-terminate */
1447 strstrip(buffer
); /* strip -just- trailing whitespace */
1449 mutex_lock(&cgroup_mutex
);
1452 * This was already checked for in cgroup_file_write(), but
1453 * check again now we're holding cgroup_mutex.
1455 if (cgroup_is_removed(cgrp
)) {
1462 retval
= attach_task_by_pid(cgrp
, buffer
);
1464 case FILE_NOTIFY_ON_RELEASE
:
1465 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
1466 if (simple_strtoul(buffer
, NULL
, 10) != 0)
1467 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
1469 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
1471 case FILE_RELEASE_AGENT
:
1472 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
1473 strcpy(cgrp
->root
->release_agent_path
, buffer
);
1483 mutex_unlock(&cgroup_mutex
);
1489 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
1490 size_t nbytes
, loff_t
*ppos
)
1492 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1493 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
1495 if (!cft
|| cgroup_is_removed(cgrp
))
1498 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1499 if (cft
->write_uint
)
1500 return cgroup_write_uint(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1504 static ssize_t
cgroup_read_uint(struct cgroup
*cgrp
, struct cftype
*cft
,
1506 char __user
*buf
, size_t nbytes
,
1510 u64 val
= cft
->read_uint(cgrp
, cft
);
1511 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
1513 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
1516 static ssize_t
cgroup_common_file_read(struct cgroup
*cgrp
,
1520 size_t nbytes
, loff_t
*ppos
)
1522 enum cgroup_filetype type
= cft
->private;
1527 if (!(page
= (char *)__get_free_page(GFP_KERNEL
)))
1533 case FILE_RELEASE_AGENT
:
1535 struct cgroupfs_root
*root
;
1537 mutex_lock(&cgroup_mutex
);
1539 n
= strnlen(root
->release_agent_path
,
1540 sizeof(root
->release_agent_path
));
1541 n
= min(n
, (size_t) PAGE_SIZE
);
1542 strncpy(s
, root
->release_agent_path
, n
);
1543 mutex_unlock(&cgroup_mutex
);
1553 retval
= simple_read_from_buffer(buf
, nbytes
, ppos
, page
, s
- page
);
1555 free_page((unsigned long)page
);
1559 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
1560 size_t nbytes
, loff_t
*ppos
)
1562 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1563 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
1565 if (!cft
|| cgroup_is_removed(cgrp
))
1569 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1571 return cgroup_read_uint(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1575 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
1580 err
= generic_file_open(inode
, file
);
1584 cft
= __d_cft(file
->f_dentry
);
1588 err
= cft
->open(inode
, file
);
1595 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
1597 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1599 return cft
->release(inode
, file
);
1604 * cgroup_rename - Only allow simple rename of directories in place.
1606 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1607 struct inode
*new_dir
, struct dentry
*new_dentry
)
1609 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
1611 if (new_dentry
->d_inode
)
1613 if (old_dir
!= new_dir
)
1615 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
1618 static struct file_operations cgroup_file_operations
= {
1619 .read
= cgroup_file_read
,
1620 .write
= cgroup_file_write
,
1621 .llseek
= generic_file_llseek
,
1622 .open
= cgroup_file_open
,
1623 .release
= cgroup_file_release
,
1626 static struct inode_operations cgroup_dir_inode_operations
= {
1627 .lookup
= simple_lookup
,
1628 .mkdir
= cgroup_mkdir
,
1629 .rmdir
= cgroup_rmdir
,
1630 .rename
= cgroup_rename
,
1633 static int cgroup_create_file(struct dentry
*dentry
, int mode
,
1634 struct super_block
*sb
)
1636 static struct dentry_operations cgroup_dops
= {
1637 .d_iput
= cgroup_diput
,
1640 struct inode
*inode
;
1644 if (dentry
->d_inode
)
1647 inode
= cgroup_new_inode(mode
, sb
);
1651 if (S_ISDIR(mode
)) {
1652 inode
->i_op
= &cgroup_dir_inode_operations
;
1653 inode
->i_fop
= &simple_dir_operations
;
1655 /* start off with i_nlink == 2 (for "." entry) */
1658 /* start with the directory inode held, so that we can
1659 * populate it without racing with another mkdir */
1660 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
1661 } else if (S_ISREG(mode
)) {
1663 inode
->i_fop
= &cgroup_file_operations
;
1665 dentry
->d_op
= &cgroup_dops
;
1666 d_instantiate(dentry
, inode
);
1667 dget(dentry
); /* Extra count - pin the dentry in core */
1672 <<<<<<< HEAD:kernel/cgroup.c
1673 * cgroup_create_dir - create a directory for an object.
1674 * cgrp: the cgroup we create the directory for.
1675 * It must have a valid ->parent field
1676 * And we are going to fill its ->dentry field.
1677 * dentry: dentry of the new cgroup
1678 * mode: mode to set on new directory.
1680 * cgroup_create_dir - create a directory for an object.
1681 * @cgrp: the cgroup we create the directory for. It must have a valid
1682 * ->parent field. And we are going to fill its ->dentry field.
1683 * @dentry: dentry of the new cgroup
1684 * @mode: mode to set on new directory.
1685 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
1687 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
1690 struct dentry
*parent
;
1693 parent
= cgrp
->parent
->dentry
;
1694 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
1696 dentry
->d_fsdata
= cgrp
;
1697 inc_nlink(parent
->d_inode
);
1698 cgrp
->dentry
= dentry
;
1706 int cgroup_add_file(struct cgroup
*cgrp
,
1707 struct cgroup_subsys
*subsys
,
1708 const struct cftype
*cft
)
1710 struct dentry
*dir
= cgrp
->dentry
;
1711 struct dentry
*dentry
;
1714 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
1715 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
1716 strcpy(name
, subsys
->name
);
1719 strcat(name
, cft
->name
);
1720 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
1721 dentry
= lookup_one_len(name
, dir
, strlen(name
));
1722 if (!IS_ERR(dentry
)) {
1723 error
= cgroup_create_file(dentry
, 0644 | S_IFREG
,
1726 dentry
->d_fsdata
= (void *)cft
;
1729 error
= PTR_ERR(dentry
);
1733 int cgroup_add_files(struct cgroup
*cgrp
,
1734 struct cgroup_subsys
*subsys
,
1735 const struct cftype cft
[],
1739 for (i
= 0; i
< count
; i
++) {
1740 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
1747 <<<<<<< HEAD
:kernel
/cgroup
.c
1748 /* Count the number of tasks in a cgroup. */
1752 * cgroup_task_count - count the number of tasks in a cgroup.
1753 * @cgrp: the cgroup in question
1755 * Return the number of tasks in the cgroup.
1757 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
1758 int cgroup_task_count(const struct cgroup
*cgrp
)
1761 struct list_head
*l
;
1763 read_lock(&css_set_lock
);
1764 l
= cgrp
->css_sets
.next
;
1765 while (l
!= &cgrp
->css_sets
) {
1766 struct cg_cgroup_link
*link
=
1767 list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
1768 count
+= atomic_read(&link
->cg
->ref
.refcount
);
1771 read_unlock(&css_set_lock
);
1776 * Advance a list_head iterator. The iterator should be positioned at
1777 * the start of a css_set
1779 static void cgroup_advance_iter(struct cgroup
*cgrp
,
1780 struct cgroup_iter
*it
)
1782 struct list_head
*l
= it
->cg_link
;
1783 struct cg_cgroup_link
*link
;
1786 /* Advance to the next non-empty css_set */
1789 if (l
== &cgrp
->css_sets
) {
1793 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
1795 } while (list_empty(&cg
->tasks
));
1797 it
->task
= cg
->tasks
.next
;
1801 * To reduce the fork() overhead for systems that are not actually
1802 * using their cgroups capability, we don't maintain the lists running
1803 * through each css_set to its tasks until we see the list actually
1804 * used - in other words after the first call to cgroup_iter_start().
1806 * The tasklist_lock is not held here, as do_each_thread() and
1807 * while_each_thread() are protected by RCU.
1809 void cgroup_enable_task_cg_lists(void)
1811 struct task_struct
*p
, *g
;
1812 write_lock(&css_set_lock
);
1813 use_task_css_set_links
= 1;
1814 do_each_thread(g
, p
) {
1816 if (list_empty(&p
->cg_list
))
1817 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
1819 } while_each_thread(g
, p
);
1820 write_unlock(&css_set_lock
);
1823 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
1826 * The first time anyone tries to iterate across a cgroup,
1827 * we need to enable the list linking each css_set to its
1828 * tasks, and fix up all existing tasks.
1830 if (!use_task_css_set_links
)
1831 cgroup_enable_task_cg_lists();
1833 read_lock(&css_set_lock
);
1834 it
->cg_link
= &cgrp
->css_sets
;
1835 cgroup_advance_iter(cgrp
, it
);
1838 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
1839 struct cgroup_iter
*it
)
1841 struct task_struct
*res
;
1842 struct list_head
*l
= it
->task
;
1844 /* If the iterator cg is NULL, we have no tasks */
1847 res
= list_entry(l
, struct task_struct
, cg_list
);
1848 /* Advance iterator to find next entry */
1850 if (l
== &res
->cgroups
->tasks
) {
1851 /* We reached the end of this task list - move on to
1852 * the next cg_cgroup_link */
1853 cgroup_advance_iter(cgrp
, it
);
1860 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
1862 read_unlock(&css_set_lock
);
1865 static inline int started_after_time(struct task_struct
*t1
,
1866 struct timespec
*time
,
1867 struct task_struct
*t2
)
1869 int start_diff
= timespec_compare(&t1
->start_time
, time
);
1870 if (start_diff
> 0) {
1872 } else if (start_diff
< 0) {
1876 * Arbitrarily, if two processes started at the same
1877 * time, we'll say that the lower pointer value
1878 * started first. Note that t2 may have exited by now
1879 * so this may not be a valid pointer any longer, but
1880 * that's fine - it still serves to distinguish
1881 * between two tasks started (effectively) simultaneously.
1888 * This function is a callback from heap_insert() and is used to order
1890 * In this case we order the heap in descending task start time.
1892 static inline int started_after(void *p1
, void *p2
)
1894 struct task_struct
*t1
= p1
;
1895 struct task_struct
*t2
= p2
;
1896 return started_after_time(t1
, &t2
->start_time
, t2
);
1900 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
1901 * @scan: struct cgroup_scanner containing arguments for the scan
1903 * Arguments include pointers to callback functions test_task() and
1905 * Iterate through all the tasks in a cgroup, calling test_task() for each,
1906 * and if it returns true, call process_task() for it also.
1907 * The test_task pointer may be NULL, meaning always true (select all tasks).
1908 * Effectively duplicates cgroup_iter_{start,next,end}()
1909 * but does not lock css_set_lock for the call to process_task().
1910 * The struct cgroup_scanner may be embedded in any structure of the caller's
1912 * It is guaranteed that process_task() will act on every task that
1913 * is a member of the cgroup for the duration of this call. This
1914 * function may or may not call process_task() for tasks that exit
1915 * or move to a different cgroup during the call, or are forked or
1916 * move into the cgroup during the call.
1918 * Note that test_task() may be called with locks held, and may in some
1919 * situations be called multiple times for the same task, so it should
1921 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
1922 * pre-allocated and will be used for heap operations (and its "gt" member will
1923 * be overwritten), else a temporary heap will be used (allocation of which
1924 * may cause this function to fail).
1926 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
1929 struct cgroup_iter it
;
1930 struct task_struct
*p
, *dropped
;
1931 /* Never dereference latest_task, since it's not refcounted */
1932 struct task_struct
*latest_task
= NULL
;
1933 struct ptr_heap tmp_heap
;
1934 struct ptr_heap
*heap
;
1935 struct timespec latest_time
= { 0, 0 };
1938 /* The caller supplied our heap and pre-allocated its memory */
1940 heap
->gt
= &started_after
;
1942 /* We need to allocate our own heap memory */
1944 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
1946 /* cannot allocate the heap */
1952 * Scan tasks in the cgroup, using the scanner's "test_task" callback
1953 * to determine which are of interest, and using the scanner's
1954 * "process_task" callback to process any of them that need an update.
1955 * Since we don't want to hold any locks during the task updates,
1956 * gather tasks to be processed in a heap structure.
1957 * The heap is sorted by descending task start time.
1958 * If the statically-sized heap fills up, we overflow tasks that
1959 * started later, and in future iterations only consider tasks that
1960 * started after the latest task in the previous pass. This
1961 * guarantees forward progress and that we don't miss any tasks.
1964 cgroup_iter_start(scan
->cg
, &it
);
1965 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
1967 * Only affect tasks that qualify per the caller's callback,
1968 * if he provided one
1970 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
1973 * Only process tasks that started after the last task
1976 if (!started_after_time(p
, &latest_time
, latest_task
))
1978 dropped
= heap_insert(heap
, p
);
1979 if (dropped
== NULL
) {
1981 * The new task was inserted; the heap wasn't
1985 } else if (dropped
!= p
) {
1987 * The new task was inserted, and pushed out a
1991 put_task_struct(dropped
);
1994 * Else the new task was newer than anything already in
1995 * the heap and wasn't inserted
1998 cgroup_iter_end(scan
->cg
, &it
);
2001 for (i
= 0; i
< heap
->size
; i
++) {
2002 struct task_struct
*p
= heap
->ptrs
[i
];
2004 latest_time
= p
->start_time
;
2007 /* Process the task per the caller's callback */
2008 scan
->process_task(p
, scan
);
2012 * If we had to process any tasks at all, scan again
2013 * in case some of them were in the middle of forking
2014 * children that didn't get processed.
2015 * Not the most efficient way to do it, but it avoids
2016 * having to take callback_mutex in the fork path
2020 if (heap
== &tmp_heap
)
2021 heap_free(&tmp_heap
);
2026 * Stuff for reading the 'tasks' file.
2028 * Reading this file can return large amounts of data if a cgroup has
2029 * *lots* of attached tasks. So it may need several calls to read(),
2030 * but we cannot guarantee that the information we produce is correct
2031 * unless we produce it entirely atomically.
2033 * Upon tasks file open(), a struct ctr_struct is allocated, that
2034 * will have a pointer to an array (also allocated here). The struct
2035 * ctr_struct * is stored in file->private_data. Its resources will
2036 * be freed by release() when the file is closed. The array is used
2037 * to sprintf the PIDs and then used by read().
2045 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
2046 * 'cgrp'. Return actual number of pids loaded. No need to
2047 * task_lock(p) when reading out p->cgroup, since we're in an RCU
2048 * read section, so the css_set can't go away, and is
2049 * immutable after creation.
2051 static int pid_array_load(pid_t
*pidarray
, int npids
, struct cgroup
*cgrp
)
2054 struct cgroup_iter it
;
2055 struct task_struct
*tsk
;
2056 cgroup_iter_start(cgrp
, &it
);
2057 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
2058 if (unlikely(n
== npids
))
2060 pidarray
[n
++] = task_pid_vnr(tsk
);
2062 cgroup_iter_end(cgrp
, &it
);
2067 <<<<<<< HEAD:kernel/cgroup.c
2068 * Build and fill cgroupstats so that taskstats can export it to user
2072 * cgroupstats_build - build and fill cgroupstats
2073 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2074 * @stats: cgroupstats to fill information into
2075 * @dentry: A dentry entry belonging to the cgroup for which stats have
2077 <<<<<<< HEAD:kernel/cgroup.c
2080 * Build and fill cgroupstats so that taskstats can export it to user
2082 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2084 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
2087 struct cgroup
*cgrp
;
2088 struct cgroup_iter it
;
2089 struct task_struct
*tsk
;
2091 * Validate dentry by checking the superblock operations
2093 if (dentry
->d_sb
->s_op
!= &cgroup_ops
)
2097 cgrp
= dentry
->d_fsdata
;
2100 cgroup_iter_start(cgrp
, &it
);
2101 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
2102 switch (tsk
->state
) {
2104 stats
->nr_running
++;
2106 case TASK_INTERRUPTIBLE
:
2107 stats
->nr_sleeping
++;
2109 case TASK_UNINTERRUPTIBLE
:
2110 stats
->nr_uninterruptible
++;
2113 stats
->nr_stopped
++;
2116 if (delayacct_is_task_waiting_on_io(tsk
))
2117 stats
->nr_io_wait
++;
2121 cgroup_iter_end(cgrp
, &it
);
2128 static int cmppid(const void *a
, const void *b
)
2130 return *(pid_t
*)a
- *(pid_t
*)b
;
2134 * Convert array 'a' of 'npids' pid_t's to a string of newline separated
2135 * decimal pids in 'buf'. Don't write more than 'sz' chars, but return
2136 * count 'cnt' of how many chars would be written if buf were large enough.
2138 static int pid_array_to_buf(char *buf
, int sz
, pid_t
*a
, int npids
)
2143 for (i
= 0; i
< npids
; i
++)
2144 cnt
+= snprintf(buf
+ cnt
, max(sz
- cnt
, 0), "%d\n", a
[i
]);
2149 * Handle an open on 'tasks' file. Prepare a buffer listing the
2150 * process id's of tasks currently attached to the cgroup being opened.
2152 * Does not require any specific cgroup mutexes, and does not take any.
2154 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
2156 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2157 struct ctr_struct
*ctr
;
2162 if (!(file
->f_mode
& FMODE_READ
))
2165 ctr
= kmalloc(sizeof(*ctr
), GFP_KERNEL
);
2170 * If cgroup gets more users after we read count, we won't have
2171 * enough space - tough. This race is indistinguishable to the
2172 * caller from the case that the additional cgroup users didn't
2173 * show up until sometime later on.
2175 npids
= cgroup_task_count(cgrp
);
2177 pidarray
= kmalloc(npids
* sizeof(pid_t
), GFP_KERNEL
);
2181 npids
= pid_array_load(pidarray
, npids
, cgrp
);
2182 sort(pidarray
, npids
, sizeof(pid_t
), cmppid
, NULL
);
2184 /* Call pid_array_to_buf() twice, first just to get bufsz */
2185 ctr
->bufsz
= pid_array_to_buf(&c
, sizeof(c
), pidarray
, npids
) + 1;
2186 ctr
->buf
= kmalloc(ctr
->bufsz
, GFP_KERNEL
);
2189 ctr
->bufsz
= pid_array_to_buf(ctr
->buf
, ctr
->bufsz
, pidarray
, npids
);
2196 file
->private_data
= ctr
;
2207 static ssize_t
cgroup_tasks_read(struct cgroup
*cgrp
,
2209 struct file
*file
, char __user
*buf
,
2210 size_t nbytes
, loff_t
*ppos
)
2212 struct ctr_struct
*ctr
= file
->private_data
;
2214 return simple_read_from_buffer(buf
, nbytes
, ppos
, ctr
->buf
, ctr
->bufsz
);
2217 static int cgroup_tasks_release(struct inode
*unused_inode
,
2220 struct ctr_struct
*ctr
;
2222 if (file
->f_mode
& FMODE_READ
) {
2223 ctr
= file
->private_data
;
2230 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
2233 return notify_on_release(cgrp
);
2236 static u64
cgroup_read_releasable(struct cgroup
*cgrp
, struct cftype
*cft
)
2238 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
2242 * for the common functions, 'private' gives the type of file
2244 static struct cftype files
[] = {
2247 .open
= cgroup_tasks_open
,
2248 .read
= cgroup_tasks_read
,
2249 .write
= cgroup_common_file_write
,
2250 .release
= cgroup_tasks_release
,
2251 .private = FILE_TASKLIST
,
2255 .name
= "notify_on_release",
2256 .read_uint
= cgroup_read_notify_on_release
,
2257 .write
= cgroup_common_file_write
,
2258 .private = FILE_NOTIFY_ON_RELEASE
,
2262 .name
= "releasable",
2263 .read_uint
= cgroup_read_releasable
,
2264 .private = FILE_RELEASABLE
,
2268 static struct cftype cft_release_agent
= {
2269 .name
= "release_agent",
2270 .read
= cgroup_common_file_read
,
2271 .write
= cgroup_common_file_write
,
2272 .private = FILE_RELEASE_AGENT
,
2275 static int cgroup_populate_dir(struct cgroup
*cgrp
)
2278 struct cgroup_subsys
*ss
;
2280 /* First clear out any existing files */
2281 cgroup_clear_directory(cgrp
->dentry
);
2283 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
2287 if (cgrp
== cgrp
->top_cgroup
) {
2288 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
2292 for_each_subsys(cgrp
->root
, ss
) {
2293 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
2300 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
2301 struct cgroup_subsys
*ss
,
2302 struct cgroup
*cgrp
)
2305 atomic_set(&css
->refcnt
, 0);
2307 if (cgrp
== dummytop
)
2308 set_bit(CSS_ROOT
, &css
->flags
);
2309 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
2310 cgrp
->subsys
[ss
->subsys_id
] = css
;
2314 <<<<<<< HEAD:kernel/cgroup.c
2315 * cgroup_create - create a cgroup
2316 * parent: cgroup that will be parent of the new cgroup.
2317 * name: name of the new cgroup. Will be strcpy'ed.
2318 * mode: mode to set on new inode
2320 * cgroup_create - create a cgroup
2321 * @parent: cgroup that will be parent of the new cgroup
2322 * @dentry: dentry of the new cgroup
2323 * @mode: mode to set on new inode
2324 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2326 <<<<<<< HEAD:kernel/cgroup.c
2327 * Must be called with the mutex on the parent inode held
2329 * Must be called with the mutex on the parent inode held
2330 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2332 <<<<<<< HEAD
:kernel
/cgroup
.c
2335 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
2336 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
2339 struct cgroup
*cgrp
;
2340 struct cgroupfs_root
*root
= parent
->root
;
2342 struct cgroup_subsys
*ss
;
2343 struct super_block
*sb
= root
->sb
;
2345 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
2349 /* Grab a reference on the superblock so the hierarchy doesn't
2350 * get deleted on unmount if there are child cgroups. This
2351 * can be done outside cgroup_mutex, since the sb can't
2352 * disappear while someone has an open control file on the
2354 atomic_inc(&sb
->s_active
);
2356 mutex_lock(&cgroup_mutex
);
2358 <<<<<<< HEAD
:kernel
/cgroup
.c
2361 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
2362 INIT_LIST_HEAD(&cgrp
->sibling
);
2363 INIT_LIST_HEAD(&cgrp
->children
);
2364 INIT_LIST_HEAD(&cgrp
->css_sets
);
2365 INIT_LIST_HEAD(&cgrp
->release_list
);
2367 cgrp
->parent
= parent
;
2368 cgrp
->root
= parent
->root
;
2369 cgrp
->top_cgroup
= parent
->top_cgroup
;
2371 <<<<<<< HEAD
:kernel
/cgroup
.c
2373 if (notify_on_release(parent
))
2374 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
2376 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
2377 for_each_subsys(root
, ss
) {
2378 struct cgroup_subsys_state
*css
= ss
->create(ss
, cgrp
);
2383 init_cgroup_css(css
, ss
, cgrp
);
2386 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
2387 root
->number_of_cgroups
++;
2389 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
2393 /* The cgroup directory was pre-locked for us */
2394 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
2396 err
= cgroup_populate_dir(cgrp
);
2397 /* If err < 0, we have a half-filled directory - oh well ;) */
2399 mutex_unlock(&cgroup_mutex
);
2400 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
2406 list_del(&cgrp
->sibling
);
2407 root
->number_of_cgroups
--;
2411 for_each_subsys(root
, ss
) {
2412 if (cgrp
->subsys
[ss
->subsys_id
])
2413 ss
->destroy(ss
, cgrp
);
2416 mutex_unlock(&cgroup_mutex
);
2418 /* Release the reference count that we took on the superblock */
2419 deactivate_super(sb
);
2425 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2427 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
2429 /* the vfs holds inode->i_mutex already */
2430 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
2433 static inline int cgroup_has_css_refs(struct cgroup
*cgrp
)
2435 /* Check the reference count on each subsystem. Since we
2436 * already established that there are no tasks in the
2437 * cgroup, if the css refcount is also 0, then there should
2438 * be no outstanding references, so the subsystem is safe to
2439 * destroy. We scan across all subsystems rather than using
2440 * the per-hierarchy linked list of mounted subsystems since
2441 * we can be called via check_for_release() with no
2442 * synchronization other than RCU, and the subsystem linked
2443 * list isn't RCU-safe */
2445 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2446 struct cgroup_subsys
*ss
= subsys
[i
];
2447 struct cgroup_subsys_state
*css
;
2448 /* Skip subsystems not in this hierarchy */
2449 if (ss
->root
!= cgrp
->root
)
2451 css
= cgrp
->subsys
[ss
->subsys_id
];
2452 /* When called from check_for_release() it's possible
2453 * that by this point the cgroup has been removed
2454 * and the css deleted. But a false-positive doesn't
2455 * matter, since it can only happen if the cgroup
2456 * has been deleted and hence no longer needs the
2457 * release agent to be called anyway. */
2458 if (css
&& atomic_read(&css
->refcnt
))
2464 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
2466 struct cgroup
*cgrp
= dentry
->d_fsdata
;
2468 struct cgroup
*parent
;
2469 struct super_block
*sb
;
2470 struct cgroupfs_root
*root
;
2472 /* the vfs holds both inode->i_mutex already */
2474 mutex_lock(&cgroup_mutex
);
2475 if (atomic_read(&cgrp
->count
) != 0) {
2476 mutex_unlock(&cgroup_mutex
);
2479 if (!list_empty(&cgrp
->children
)) {
2480 mutex_unlock(&cgroup_mutex
);
2484 parent
= cgrp
->parent
;
2487 <<<<<<< HEAD
:kernel
/cgroup
.c
2490 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
2492 <<<<<<< HEAD:kernel/cgroup.c
2493 * Call pre_destroy handlers of subsys
2495 * Call pre_destroy handlers of subsys. Notify subsystems
2496 * that rmdir() request comes.
2497 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2499 cgroup_call_pre_destroy(cgrp
);
2500 <<<<<<< HEAD
:kernel
/cgroup
.c
2502 * Notify subsyses that rmdir() request comes.
2505 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
2507 if (cgroup_has_css_refs(cgrp
)) {
2508 mutex_unlock(&cgroup_mutex
);
2512 spin_lock(&release_list_lock
);
2513 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
2514 if (!list_empty(&cgrp
->release_list
))
2515 list_del(&cgrp
->release_list
);
2516 spin_unlock(&release_list_lock
);
2517 /* delete my sibling from parent->children */
2518 list_del(&cgrp
->sibling
);
2519 spin_lock(&cgrp
->dentry
->d_lock
);
2520 d
= dget(cgrp
->dentry
);
2521 cgrp
->dentry
= NULL
;
2522 spin_unlock(&d
->d_lock
);
2524 cgroup_d_remove_dir(d
);
2527 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
2528 check_for_release(parent
);
2530 mutex_unlock(&cgroup_mutex
);
2534 static void cgroup_init_subsys(struct cgroup_subsys
*ss
)
2536 struct cgroup_subsys_state
*css
;
2537 struct list_head
*l
;
2539 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
2541 /* Create the top cgroup state for this subsystem */
2542 ss
->root
= &rootnode
;
2543 css
= ss
->create(ss
, dummytop
);
2544 /* We don't handle early failures gracefully */
2545 BUG_ON(IS_ERR(css
));
2546 init_cgroup_css(css
, ss
, dummytop
);
2548 /* Update all cgroup groups to contain a subsys
2549 * pointer to this state - since the subsystem is
2550 * newly registered, all tasks and hence all cgroup
2551 * groups are in the subsystem's top cgroup. */
2552 write_lock(&css_set_lock
);
2553 l
= &init_css_set
.list
;
2555 struct css_set
*cg
=
2556 list_entry(l
, struct css_set
, list
);
2557 cg
->subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
2559 } while (l
!= &init_css_set
.list
);
2560 write_unlock(&css_set_lock
);
2562 /* If this subsystem requested that it be notified with fork
2563 * events, we should send it one now for every process in the
2566 struct task_struct
*g
, *p
;
2568 read_lock(&tasklist_lock
);
2569 do_each_thread(g
, p
) {
2571 } while_each_thread(g
, p
);
2572 read_unlock(&tasklist_lock
);
2575 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
2581 <<<<<<< HEAD:kernel/cgroup.c
2582 * cgroup_init_early - initialize cgroups at system boot, and
2583 * initialize any subsystems that request early init.
2585 * cgroup_init_early - cgroup initialization at system boot
2587 * Initialize cgroups at system boot, and initialize any
2588 * subsystems that request early init.
2589 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2591 int __init
cgroup_init_early(void)
2594 kref_init(&init_css_set
.ref
);
2595 kref_get(&init_css_set
.ref
);
2596 INIT_LIST_HEAD(&init_css_set
.list
);
2597 INIT_LIST_HEAD(&init_css_set
.cg_links
);
2598 INIT_LIST_HEAD(&init_css_set
.tasks
);
2600 init_cgroup_root(&rootnode
);
2601 list_add(&rootnode
.root_list
, &roots
);
2603 init_task
.cgroups
= &init_css_set
;
2605 init_css_set_link
.cg
= &init_css_set
;
2606 list_add(&init_css_set_link
.cgrp_link_list
,
2607 &rootnode
.top_cgroup
.css_sets
);
2608 list_add(&init_css_set_link
.cg_link_list
,
2609 &init_css_set
.cg_links
);
2611 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2612 struct cgroup_subsys
*ss
= subsys
[i
];
2615 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
2616 BUG_ON(!ss
->create
);
2617 BUG_ON(!ss
->destroy
);
2618 if (ss
->subsys_id
!= i
) {
2619 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
2620 ss
->name
, ss
->subsys_id
);
2625 cgroup_init_subsys(ss
);
2631 <<<<<<< HEAD:kernel/cgroup.c
2632 * cgroup_init - register cgroup filesystem and /proc file, and
2633 * initialize any subsystems that didn't request early init.
2635 * cgroup_init - cgroup initialization
2637 * Register cgroup filesystem and /proc file, and initialize
2638 * any subsystems that didn't request early init.
2639 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2641 int __init
cgroup_init(void)
2645 struct proc_dir_entry
*entry
;
2647 err
= bdi_init(&cgroup_backing_dev_info
);
2651 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2652 struct cgroup_subsys
*ss
= subsys
[i
];
2653 if (!ss
->early_init
)
2654 cgroup_init_subsys(ss
);
2657 err
= register_filesystem(&cgroup_fs_type
);
2661 entry
= create_proc_entry("cgroups", 0, NULL
);
2663 entry
->proc_fops
= &proc_cgroupstats_operations
;
2667 bdi_destroy(&cgroup_backing_dev_info
);
2673 * proc_cgroup_show()
2674 * - Print task's cgroup paths into seq_file, one line for each hierarchy
2675 * - Used for /proc/<pid>/cgroup.
2676 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
2677 * doesn't really matter if tsk->cgroup changes after we read it,
2678 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
2679 * anyway. No need to check that tsk->cgroup != NULL, thanks to
2680 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
2681 * cgroup to top_cgroup.
2684 /* TODO: Use a proper seq_file iterator */
2685 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
2688 struct task_struct
*tsk
;
2691 struct cgroupfs_root
*root
;
2694 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
2700 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
2706 mutex_lock(&cgroup_mutex
);
2708 for_each_root(root
) {
2709 struct cgroup_subsys
*ss
;
2710 struct cgroup
*cgrp
;
2714 /* Skip this hierarchy if it has no active subsystems */
2715 if (!root
->actual_subsys_bits
)
2717 for_each_subsys(root
, ss
)
2718 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
2720 get_first_subsys(&root
->top_cgroup
, NULL
, &subsys_id
);
2721 cgrp
= task_cgroup(tsk
, subsys_id
);
2722 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
2730 mutex_unlock(&cgroup_mutex
);
2731 put_task_struct(tsk
);
2738 static int cgroup_open(struct inode
*inode
, struct file
*file
)
2740 struct pid
*pid
= PROC_I(inode
)->pid
;
2741 return single_open(file
, proc_cgroup_show
, pid
);
2744 struct file_operations proc_cgroup_operations
= {
2745 .open
= cgroup_open
,
2747 .llseek
= seq_lseek
,
2748 .release
= single_release
,
2751 /* Display information about each subsystem and each hierarchy */
2752 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
2756 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\n");
2757 mutex_lock(&cgroup_mutex
);
2758 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2759 struct cgroup_subsys
*ss
= subsys
[i
];
2760 seq_printf(m
, "%s\t%lu\t%d\n",
2761 ss
->name
, ss
->root
->subsys_bits
,
2762 ss
->root
->number_of_cgroups
);
2764 mutex_unlock(&cgroup_mutex
);
2768 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
2770 return single_open(file
, proc_cgroupstats_show
, 0);
2773 static struct file_operations proc_cgroupstats_operations
= {
2774 .open
= cgroupstats_open
,
2776 .llseek
= seq_lseek
,
2777 .release
= single_release
,
2781 * cgroup_fork - attach newly forked task to its parents cgroup.
2782 <<<<<<< HEAD:kernel/cgroup.c
2783 * @tsk: pointer to task_struct of forking parent process.
2785 * @child: pointer to task_struct of forking parent process.
2786 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2788 * Description: A task inherits its parent's cgroup at fork().
2790 * A pointer to the shared css_set was automatically copied in
2791 * fork.c by dup_task_struct(). However, we ignore that copy, since
2792 * it was not made under the protection of RCU or cgroup_mutex, so
2793 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
2794 * have already changed current->cgroups, allowing the previously
2795 * referenced cgroup group to be removed and freed.
2797 * At the point that cgroup_fork() is called, 'current' is the parent
2798 * task, and the passed argument 'child' points to the child task.
2800 void cgroup_fork(struct task_struct
*child
)
2803 child
->cgroups
= current
->cgroups
;
2804 get_css_set(child
->cgroups
);
2805 task_unlock(current
);
2806 INIT_LIST_HEAD(&child
->cg_list
);
2810 <<<<<<< HEAD:kernel/cgroup.c
2811 * cgroup_fork_callbacks - called on a new task very soon before
2812 * adding it to the tasklist. No need to take any locks since no-one
2813 * can be operating on this task
2815 * cgroup_fork_callbacks - run fork callbacks
2816 * @child: the new task
2818 * Called on a new task very soon before adding it to the
2819 * tasklist. No need to take any locks since no-one can
2820 * be operating on this task.
2821 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2823 void cgroup_fork_callbacks(struct task_struct
*child
)
2825 if (need_forkexit_callback
) {
2827 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2828 struct cgroup_subsys
*ss
= subsys
[i
];
2830 ss
->fork(ss
, child
);
2836 <<<<<<< HEAD:kernel/cgroup.c
2837 * cgroup_post_fork - called on a new task after adding it to the
2838 * task list. Adds the task to the list running through its css_set
2839 * if necessary. Has to be after the task is visible on the task list
2840 * in case we race with the first call to cgroup_iter_start() - to
2841 * guarantee that the new task ends up on its list. */
2843 * cgroup_post_fork
- called on a
new task after adding it to the task list
2844 * @child
: the task in question
2846 * Adds the task to the list running through its css_set
if necessary
.
2847 * Has to be after the task is visible on the task list in
case we race
2848 * with the first call to
cgroup_iter_start() - to guarantee that the
2849 * new task ends up on its list
.
2851 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
2852 void cgroup_post_fork(struct task_struct
*child
)
2854 if (use_task_css_set_links
) {
2855 write_lock(&css_set_lock
);
2856 if (list_empty(&child
->cg_list
))
2857 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
2858 write_unlock(&css_set_lock
);
2862 * cgroup_exit - detach cgroup from exiting task
2863 * @tsk: pointer to task_struct of exiting process
2864 <<<<<<< HEAD:kernel/cgroup.c
2866 * @run_callback: run exit callbacks?
2867 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2869 * Description: Detach cgroup from @tsk and release it.
2871 * Note that cgroups marked notify_on_release force every task in
2872 * them to take the global cgroup_mutex mutex when exiting.
2873 * This could impact scaling on very large systems. Be reluctant to
2874 * use notify_on_release cgroups where very high task exit scaling
2875 * is required on large systems.
2877 * the_top_cgroup_hack:
2879 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
2881 * We call cgroup_exit() while the task is still competent to
2882 * handle notify_on_release(), then leave the task attached to the
2883 * root cgroup in each hierarchy for the remainder of its exit.
2885 * To do this properly, we would increment the reference count on
2886 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
2887 * code we would add a second cgroup function call, to drop that
2888 * reference. This would just create an unnecessary hot spot on
2889 * the top_cgroup reference count, to no avail.
2891 * Normally, holding a reference to a cgroup without bumping its
2892 * count is unsafe. The cgroup could go away, or someone could
2893 * attach us to a different cgroup, decrementing the count on
2894 * the first cgroup that we never incremented. But in this case,
2895 * top_cgroup isn't going away, and either task has PF_EXITING set,
2896 * which wards off any cgroup_attach_task() attempts, or task is a failed
2897 * fork, never visible to cgroup_attach_task.
2898 <<<<<<< HEAD:kernel/cgroup.c
2901 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2903 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
2908 if (run_callbacks
&& need_forkexit_callback
) {
2909 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2910 struct cgroup_subsys
*ss
= subsys
[i
];
2917 * Unlink from the css_set task list if necessary.
2918 * Optimistically check cg_list before taking
2921 if (!list_empty(&tsk
->cg_list
)) {
2922 write_lock(&css_set_lock
);
2923 if (!list_empty(&tsk
->cg_list
))
2924 list_del(&tsk
->cg_list
);
2925 write_unlock(&css_set_lock
);
2928 /* Reassign the task to the init_css_set. */
2931 tsk
->cgroups
= &init_css_set
;
2934 put_css_set_taskexit(cg
);
2938 <<<<<<< HEAD:kernel/cgroup.c
2939 * cgroup_clone - duplicate the current cgroup in the hierarchy
2940 * that the given subsystem is attached to, and move this task into
2943 * cgroup_clone - clone the cgroup the given subsystem is attached to
2944 * @tsk: the task to be moved
2945 * @subsys: the given subsystem
2947 * Duplicate the current cgroup in the hierarchy that the given
2948 * subsystem is attached to, and move this task into the new
2950 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
2952 int cgroup_clone(struct task_struct
*tsk
, struct cgroup_subsys
*subsys
)
2954 struct dentry
*dentry
;
2956 char nodename
[MAX_CGROUP_TYPE_NAMELEN
];
2957 struct cgroup
*parent
, *child
;
2958 struct inode
*inode
;
2960 struct cgroupfs_root
*root
;
2961 struct cgroup_subsys
*ss
;
2963 /* We shouldn't be called by an unregistered subsystem */
2964 BUG_ON(!subsys
->active
);
2966 /* First figure out what hierarchy and cgroup we're dealing
2967 * with, and pin them so we can drop cgroup_mutex */
2968 mutex_lock(&cgroup_mutex
);
2970 root
= subsys
->root
;
2971 if (root
== &rootnode
) {
2973 "Not cloning cgroup for unused subsystem %s\n",
2975 mutex_unlock(&cgroup_mutex
);
2979 parent
= task_cgroup(tsk
, subsys
->subsys_id
);
2981 snprintf(nodename
, MAX_CGROUP_TYPE_NAMELEN
, "node_%d", tsk
->pid
);
2983 /* Pin the hierarchy */
2984 atomic_inc(&parent
->root
->sb
->s_active
);
2986 /* Keep the cgroup alive */
2988 mutex_unlock(&cgroup_mutex
);
2990 /* Now do the VFS work to create a cgroup */
2991 inode
= parent
->dentry
->d_inode
;
2993 /* Hold the parent directory mutex across this operation to
2994 * stop anyone else deleting the new cgroup */
2995 mutex_lock(&inode
->i_mutex
);
2996 dentry
= lookup_one_len(nodename
, parent
->dentry
, strlen(nodename
));
2997 if (IS_ERR(dentry
)) {
2999 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename
,
3001 ret
= PTR_ERR(dentry
);
3005 /* Create the cgroup directory, which also creates the cgroup */
3006 ret
= vfs_mkdir(inode
, dentry
, S_IFDIR
| 0755);
3007 child
= __d_cgrp(dentry
);
3011 "Failed to create cgroup %s: %d\n", nodename
,
3018 "Couldn't find new cgroup %s\n", nodename
);
3023 /* The cgroup now exists. Retake cgroup_mutex and check
3024 * that we're still in the same state that we thought we
3026 mutex_lock(&cgroup_mutex
);
3027 if ((root
!= subsys
->root
) ||
3028 (parent
!= task_cgroup(tsk
, subsys
->subsys_id
))) {
3029 /* Aargh, we raced ... */
3030 mutex_unlock(&inode
->i_mutex
);
3033 deactivate_super(parent
->root
->sb
);
3034 /* The cgroup is still accessible in the VFS, but
3035 * we're not going to try to rmdir() it at this
3038 "Race in cgroup_clone() - leaking cgroup %s\n",
3043 /* do any required auto-setup */
3044 for_each_subsys(root
, ss
) {
3046 ss
->post_clone(ss
, child
);
3049 /* All seems fine. Finish by moving the task into the new cgroup */
3050 ret
= cgroup_attach_task(child
, tsk
);
3051 mutex_unlock(&cgroup_mutex
);
3054 mutex_unlock(&inode
->i_mutex
);
3056 mutex_lock(&cgroup_mutex
);
3058 mutex_unlock(&cgroup_mutex
);
3059 deactivate_super(parent
->root
->sb
);
3063 <<<<<<< HEAD
:kernel
/cgroup
.c
3065 * See if "cgrp" is a descendant of the current task's cgroup in
3066 * the appropriate hierarchy
3069 * cgroup_is_descendant - see if @cgrp is a descendant of current task's cgrp
3070 * @cgrp: the cgroup in question
3072 * See if @cgrp is a descendant of the current task's cgroup in
3073 * the appropriate hierarchy.
3074 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
3076 * If we are sending in dummytop, then presumably we are creating
3077 * the top cgroup in the subsystem.
3079 * Called only by the ns (nsproxy) cgroup.
3081 int cgroup_is_descendant(const struct cgroup
*cgrp
)
3084 struct cgroup
*target
;
3087 if (cgrp
== dummytop
)
3090 get_first_subsys(cgrp
, NULL
, &subsys_id
);
3091 target
= task_cgroup(current
, subsys_id
);
3092 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
3093 cgrp
= cgrp
->parent
;
3094 ret
= (cgrp
== target
);
3098 static void check_for_release(struct cgroup
*cgrp
)
3100 /* All of these checks rely on RCU to keep the cgroup
3101 * structure alive */
3102 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
3103 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
3104 /* Control Group is currently removeable. If it's not
3105 * already queued for a userspace notification, queue
3107 int need_schedule_work
= 0;
3108 spin_lock(&release_list_lock
);
3109 if (!cgroup_is_removed(cgrp
) &&
3110 list_empty(&cgrp
->release_list
)) {
3111 list_add(&cgrp
->release_list
, &release_list
);
3112 need_schedule_work
= 1;
3114 spin_unlock(&release_list_lock
);
3115 if (need_schedule_work
)
3116 schedule_work(&release_agent_work
);
3120 void __css_put(struct cgroup_subsys_state
*css
)
3122 struct cgroup
*cgrp
= css
->cgroup
;
3124 if (atomic_dec_and_test(&css
->refcnt
) && notify_on_release(cgrp
)) {
3125 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3126 check_for_release(cgrp
);
3132 * Notify userspace when a cgroup is released, by running the
3133 * configured release agent with the name of the cgroup (path
3134 * relative to the root of cgroup file system) as the argument.
3136 * Most likely, this user command will try to rmdir this cgroup.
3138 * This races with the possibility that some other task will be
3139 * attached to this cgroup before it is removed, or that some other
3140 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
3141 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
3142 * unused, and this cgroup will be reprieved from its death sentence,
3143 * to continue to serve a useful existence. Next time it's released,
3144 * we will get notified again, if it still has 'notify_on_release' set.
3146 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
3147 * means only wait until the task is successfully execve()'d. The
3148 * separate release agent task is forked by call_usermodehelper(),
3149 * then control in this thread returns here, without waiting for the
3150 * release agent task. We don't bother to wait because the caller of
3151 * this routine has no use for the exit status of the release agent
3152 * task, so no sense holding our caller up for that.
3153 <<<<<<< HEAD:kernel/cgroup.c
3156 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a:kernel/cgroup.c
3158 <<<<<<< HEAD
:kernel
/cgroup
.c
3161 >>>>>>> 264e3e889d86e552b4191d69bb60f4f3b383135a
:kernel
/cgroup
.c
3162 static void cgroup_release_agent(struct work_struct
*work
)
3164 BUG_ON(work
!= &release_agent_work
);
3165 mutex_lock(&cgroup_mutex
);
3166 spin_lock(&release_list_lock
);
3167 while (!list_empty(&release_list
)) {
3168 char *argv
[3], *envp
[3];
3171 struct cgroup
*cgrp
= list_entry(release_list
.next
,
3174 list_del_init(&cgrp
->release_list
);
3175 spin_unlock(&release_list_lock
);
3176 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3178 spin_lock(&release_list_lock
);
3182 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0) {
3184 spin_lock(&release_list_lock
);
3189 argv
[i
++] = cgrp
->root
->release_agent_path
;
3190 argv
[i
++] = (char *)pathbuf
;
3194 /* minimal command environment */
3195 envp
[i
++] = "HOME=/";
3196 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
3199 /* Drop the lock while we invoke the usermode helper,
3200 * since the exec could involve hitting disk and hence
3201 * be a slow process */
3202 mutex_unlock(&cgroup_mutex
);
3203 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
3205 mutex_lock(&cgroup_mutex
);
3206 spin_lock(&release_list_lock
);
3208 spin_unlock(&release_list_lock
);
3209 mutex_unlock(&cgroup_mutex
);