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 * check for fork/exit handlers to call. This avoids us having to do
117 * extra work in the fork/exit path if none of the subsystems need to
120 static int need_forkexit_callback
;
122 /* bits in struct cgroup flags field */
124 /* Control Group is dead */
126 /* Control Group has previously had a child cgroup or a task,
127 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
129 /* Control Group requires release notifications to userspace */
130 CGRP_NOTIFY_ON_RELEASE
,
133 /* convenient tests for these bits */
134 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
136 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
139 /* bits in struct cgroupfs_root flags field */
141 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
144 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
147 (1 << CGRP_RELEASABLE
) |
148 (1 << CGRP_NOTIFY_ON_RELEASE
);
149 return (cgrp
->flags
& bits
) == bits
;
152 static int notify_on_release(const struct cgroup
*cgrp
)
154 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
158 * for_each_subsys() allows you to iterate on each subsystem attached to
159 * an active hierarchy
161 #define for_each_subsys(_root, _ss) \
162 list_for_each_entry(_ss, &_root->subsys_list, sibling)
164 /* for_each_root() allows you to iterate across the active hierarchies */
165 #define for_each_root(_root) \
166 list_for_each_entry(_root, &roots, root_list)
168 /* the list of cgroups eligible for automatic release. Protected by
169 * release_list_lock */
170 static LIST_HEAD(release_list
);
171 static DEFINE_SPINLOCK(release_list_lock
);
172 static void cgroup_release_agent(struct work_struct
*work
);
173 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
174 static void check_for_release(struct cgroup
*cgrp
);
176 /* Link structure for associating css_set objects with cgroups */
177 struct cg_cgroup_link
{
179 * List running through cg_cgroup_links associated with a
180 * cgroup, anchored on cgroup->css_sets
182 struct list_head cgrp_link_list
;
184 * List running through cg_cgroup_links pointing at a
185 * single css_set object, anchored on css_set->cg_links
187 struct list_head cg_link_list
;
191 /* The default css_set - used by init and its children prior to any
192 * hierarchies being mounted. It contains a pointer to the root state
193 * for each subsystem. Also used to anchor the list of css_sets. Not
194 * reference-counted, to improve performance when child cgroups
195 * haven't been created.
198 static struct css_set init_css_set
;
199 static struct cg_cgroup_link init_css_set_link
;
201 /* css_set_lock protects the list of css_set objects, and the
202 * chain of tasks off each css_set. Nests outside task->alloc_lock
203 * due to cgroup_iter_start() */
204 static DEFINE_RWLOCK(css_set_lock
);
205 static int css_set_count
;
207 /* We don't maintain the lists running through each css_set to its
208 * task until after the first call to cgroup_iter_start(). This
209 * reduces the fork()/exit() overhead for people who have cgroups
210 * compiled into their kernel but not actually in use */
211 static int use_task_css_set_links
;
213 /* When we create or destroy a css_set, the operation simply
214 * takes/releases a reference count on all the cgroups referenced
215 * by subsystems in this css_set. This can end up multiple-counting
216 * some cgroups, but that's OK - the ref-count is just a
217 * busy/not-busy indicator; ensuring that we only count each cgroup
218 * once would require taking a global lock to ensure that no
219 * subsystems moved between hierarchies while we were doing so.
221 * Possible TODO: decide at boot time based on the number of
222 * registered subsystems and the number of CPUs or NUMA nodes whether
223 * it's better for performance to ref-count every subsystem, or to
224 * take a global lock and only add one ref count to each hierarchy.
228 * unlink a css_set from the list and free it
230 static void unlink_css_set(struct css_set
*cg
)
232 write_lock(&css_set_lock
);
235 while (!list_empty(&cg
->cg_links
)) {
236 struct cg_cgroup_link
*link
;
237 link
= list_entry(cg
->cg_links
.next
,
238 struct cg_cgroup_link
, cg_link_list
);
239 list_del(&link
->cg_link_list
);
240 list_del(&link
->cgrp_link_list
);
243 write_unlock(&css_set_lock
);
246 static void __release_css_set(struct kref
*k
, int taskexit
)
249 struct css_set
*cg
= container_of(k
, struct css_set
, ref
);
254 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
255 struct cgroup
*cgrp
= cg
->subsys
[i
]->cgroup
;
256 if (atomic_dec_and_test(&cgrp
->count
) &&
257 notify_on_release(cgrp
)) {
259 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
260 check_for_release(cgrp
);
267 static void release_css_set(struct kref
*k
)
269 __release_css_set(k
, 0);
272 static void release_css_set_taskexit(struct kref
*k
)
274 __release_css_set(k
, 1);
278 * refcounted get/put for css_set objects
280 static inline void get_css_set(struct css_set
*cg
)
285 static inline void put_css_set(struct css_set
*cg
)
287 kref_put(&cg
->ref
, release_css_set
);
290 static inline void put_css_set_taskexit(struct css_set
*cg
)
292 kref_put(&cg
->ref
, release_css_set_taskexit
);
296 * find_existing_css_set() is a helper for
297 * find_css_set(), and checks to see whether an existing
298 * css_set is suitable. This currently walks a linked-list for
299 * simplicity; a later patch will use a hash table for better
302 * oldcg: the cgroup group that we're using before the cgroup
305 * cgrp: the cgroup that we're moving into
307 * template: location in which to build the desired set of subsystem
308 * state objects for the new cgroup group
310 static struct css_set
*find_existing_css_set(
311 struct css_set
*oldcg
,
313 struct cgroup_subsys_state
*template[])
316 struct cgroupfs_root
*root
= cgrp
->root
;
317 struct list_head
*l
= &init_css_set
.list
;
319 /* Built the set of subsystem state objects that we want to
320 * see in the new css_set */
321 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
322 if (root
->subsys_bits
& (1UL << i
)) {
323 /* Subsystem is in this hierarchy. So we want
324 * the subsystem state from the new
326 template[i
] = cgrp
->subsys
[i
];
328 /* Subsystem is not in this hierarchy, so we
329 * don't want to change the subsystem state */
330 template[i
] = oldcg
->subsys
[i
];
334 /* Look through existing cgroup groups to find one to reuse */
337 list_entry(l
, struct css_set
, list
);
339 if (!memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
340 /* All subsystems matched */
343 /* Try the next cgroup group */
345 } while (l
!= &init_css_set
.list
);
347 /* No existing cgroup group matched */
352 * allocate_cg_links() allocates "count" cg_cgroup_link structures
353 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
354 * success or a negative error
356 static int allocate_cg_links(int count
, struct list_head
*tmp
)
358 struct cg_cgroup_link
*link
;
361 for (i
= 0; i
< count
; i
++) {
362 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
364 while (!list_empty(tmp
)) {
365 link
= list_entry(tmp
->next
,
366 struct cg_cgroup_link
,
368 list_del(&link
->cgrp_link_list
);
373 list_add(&link
->cgrp_link_list
, tmp
);
378 static void free_cg_links(struct list_head
*tmp
)
380 while (!list_empty(tmp
)) {
381 struct cg_cgroup_link
*link
;
382 link
= list_entry(tmp
->next
,
383 struct cg_cgroup_link
,
385 list_del(&link
->cgrp_link_list
);
391 * find_css_set() takes an existing cgroup group and a
392 * cgroup object, and returns a css_set object that's
393 * equivalent to the old group, but with the given cgroup
394 * substituted into the appropriate hierarchy. Must be called with
397 static struct css_set
*find_css_set(
398 struct css_set
*oldcg
, struct cgroup
*cgrp
)
401 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
404 struct list_head tmp_cg_links
;
405 struct cg_cgroup_link
*link
;
407 /* First see if we already have a cgroup group that matches
409 write_lock(&css_set_lock
);
410 res
= find_existing_css_set(oldcg
, cgrp
, template);
413 write_unlock(&css_set_lock
);
418 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
422 /* Allocate all the cg_cgroup_link objects that we'll need */
423 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
428 kref_init(&res
->ref
);
429 INIT_LIST_HEAD(&res
->cg_links
);
430 INIT_LIST_HEAD(&res
->tasks
);
432 /* Copy the set of subsystem state objects generated in
433 * find_existing_css_set() */
434 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
436 write_lock(&css_set_lock
);
437 /* Add reference counts and links from the new css_set. */
438 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
439 struct cgroup
*cgrp
= res
->subsys
[i
]->cgroup
;
440 struct cgroup_subsys
*ss
= subsys
[i
];
441 atomic_inc(&cgrp
->count
);
443 * We want to add a link once per cgroup, so we
444 * only do it for the first subsystem in each
447 if (ss
->root
->subsys_list
.next
== &ss
->sibling
) {
448 BUG_ON(list_empty(&tmp_cg_links
));
449 link
= list_entry(tmp_cg_links
.next
,
450 struct cg_cgroup_link
,
452 list_del(&link
->cgrp_link_list
);
453 list_add(&link
->cgrp_link_list
, &cgrp
->css_sets
);
455 list_add(&link
->cg_link_list
, &res
->cg_links
);
458 if (list_empty(&rootnode
.subsys_list
)) {
459 link
= list_entry(tmp_cg_links
.next
,
460 struct cg_cgroup_link
,
462 list_del(&link
->cgrp_link_list
);
463 list_add(&link
->cgrp_link_list
, &dummytop
->css_sets
);
465 list_add(&link
->cg_link_list
, &res
->cg_links
);
468 BUG_ON(!list_empty(&tmp_cg_links
));
470 /* Link this cgroup group into the list */
471 list_add(&res
->list
, &init_css_set
.list
);
473 write_unlock(&css_set_lock
);
479 * There is one global cgroup mutex. We also require taking
480 * task_lock() when dereferencing a task's cgroup subsys pointers.
481 * See "The task_lock() exception", at the end of this comment.
483 * A task must hold cgroup_mutex to modify cgroups.
485 * Any task can increment and decrement the count field without lock.
486 * So in general, code holding cgroup_mutex can't rely on the count
487 * field not changing. However, if the count goes to zero, then only
488 * cgroup_attach_task() can increment it again. Because a count of zero
489 * means that no tasks are currently attached, therefore there is no
490 * way a task attached to that cgroup can fork (the other way to
491 * increment the count). So code holding cgroup_mutex can safely
492 * assume that if the count is zero, it will stay zero. Similarly, if
493 * a task holds cgroup_mutex on a cgroup with zero count, it
494 * knows that the cgroup won't be removed, as cgroup_rmdir()
497 * The cgroup_common_file_write handler for operations that modify
498 * the cgroup hierarchy holds cgroup_mutex across the entire operation,
499 * single threading all such cgroup modifications across the system.
501 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
502 * (usually) take cgroup_mutex. These are the two most performance
503 * critical pieces of code here. The exception occurs on cgroup_exit(),
504 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
505 * is taken, and if the cgroup count is zero, a usermode call made
506 * to the release agent with the name of the cgroup (path relative to
507 * the root of cgroup file system) as the argument.
509 * A cgroup can only be deleted if both its 'count' of using tasks
510 * is zero, and its list of 'children' cgroups is empty. Since all
511 * tasks in the system use _some_ cgroup, and since there is always at
512 * least one task in the system (init, pid == 1), therefore, top_cgroup
513 * always has either children cgroups and/or using tasks. So we don't
514 * need a special hack to ensure that top_cgroup cannot be deleted.
516 * The task_lock() exception
518 * The need for this exception arises from the action of
519 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
520 * another. It does so using cgroup_mutex, however there are
521 * several performance critical places that need to reference
522 * task->cgroup without the expense of grabbing a system global
523 * mutex. Therefore except as noted below, when dereferencing or, as
524 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
525 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
526 * the task_struct routinely used for such matters.
528 * P.S. One more locking exception. RCU is used to guard the
529 * update of a tasks cgroup pointer by cgroup_attach_task()
533 * cgroup_lock - lock out any changes to cgroup structures
536 void cgroup_lock(void)
538 mutex_lock(&cgroup_mutex
);
542 * cgroup_unlock - release lock on cgroup changes
544 * Undo the lock taken in a previous cgroup_lock() call.
546 void cgroup_unlock(void)
548 mutex_unlock(&cgroup_mutex
);
552 * A couple of forward declarations required, due to cyclic reference loop:
553 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
554 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
558 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
);
559 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
560 static int cgroup_populate_dir(struct cgroup
*cgrp
);
561 static struct inode_operations cgroup_dir_inode_operations
;
562 static struct file_operations proc_cgroupstats_operations
;
564 static struct backing_dev_info cgroup_backing_dev_info
= {
565 .capabilities
= BDI_CAP_NO_ACCT_DIRTY
| BDI_CAP_NO_WRITEBACK
,
568 static struct inode
*cgroup_new_inode(mode_t mode
, struct super_block
*sb
)
570 struct inode
*inode
= new_inode(sb
);
573 inode
->i_mode
= mode
;
574 inode
->i_uid
= current
->fsuid
;
575 inode
->i_gid
= current
->fsgid
;
577 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
578 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
584 * Call subsys's pre_destroy handler.
585 * This is called before css refcnt check.
587 static void cgroup_call_pre_destroy(struct cgroup
*cgrp
)
589 struct cgroup_subsys
*ss
;
590 for_each_subsys(cgrp
->root
, ss
)
591 if (ss
->pre_destroy
&& cgrp
->subsys
[ss
->subsys_id
])
592 ss
->pre_destroy(ss
, cgrp
);
596 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
598 /* is dentry a directory ? if so, kfree() associated cgroup */
599 if (S_ISDIR(inode
->i_mode
)) {
600 struct cgroup
*cgrp
= dentry
->d_fsdata
;
601 struct cgroup_subsys
*ss
;
602 BUG_ON(!(cgroup_is_removed(cgrp
)));
603 /* It's possible for external users to be holding css
604 * reference counts on a cgroup; css_put() needs to
605 * be able to access the cgroup after decrementing
606 * the reference count in order to know if it needs to
607 * queue the cgroup to be handled by the release
611 mutex_lock(&cgroup_mutex
);
613 * Release the subsystem state objects.
615 for_each_subsys(cgrp
->root
, ss
) {
616 if (cgrp
->subsys
[ss
->subsys_id
])
617 ss
->destroy(ss
, cgrp
);
620 cgrp
->root
->number_of_cgroups
--;
621 mutex_unlock(&cgroup_mutex
);
623 /* Drop the active superblock reference that we took when we
624 * created the cgroup */
625 deactivate_super(cgrp
->root
->sb
);
632 static void remove_dir(struct dentry
*d
)
634 struct dentry
*parent
= dget(d
->d_parent
);
637 simple_rmdir(parent
->d_inode
, d
);
641 static void cgroup_clear_directory(struct dentry
*dentry
)
643 struct list_head
*node
;
645 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
646 spin_lock(&dcache_lock
);
647 node
= dentry
->d_subdirs
.next
;
648 while (node
!= &dentry
->d_subdirs
) {
649 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
652 /* This should never be called on a cgroup
653 * directory with child cgroups */
654 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
656 spin_unlock(&dcache_lock
);
658 simple_unlink(dentry
->d_inode
, d
);
660 spin_lock(&dcache_lock
);
662 node
= dentry
->d_subdirs
.next
;
664 spin_unlock(&dcache_lock
);
668 * NOTE : the dentry must have been dget()'ed
670 static void cgroup_d_remove_dir(struct dentry
*dentry
)
672 cgroup_clear_directory(dentry
);
674 spin_lock(&dcache_lock
);
675 list_del_init(&dentry
->d_u
.d_child
);
676 spin_unlock(&dcache_lock
);
680 static int rebind_subsystems(struct cgroupfs_root
*root
,
681 unsigned long final_bits
)
683 unsigned long added_bits
, removed_bits
;
684 struct cgroup
*cgrp
= &root
->top_cgroup
;
687 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
688 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
689 /* Check that any added subsystems are currently free */
690 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
691 unsigned long bit
= 1UL << i
;
692 struct cgroup_subsys
*ss
= subsys
[i
];
693 if (!(bit
& added_bits
))
695 if (ss
->root
!= &rootnode
) {
696 /* Subsystem isn't free */
701 /* Currently we don't handle adding/removing subsystems when
702 * any child cgroups exist. This is theoretically supportable
703 * but involves complex error handling, so it's being left until
705 if (!list_empty(&cgrp
->children
))
708 /* Process each subsystem */
709 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
710 struct cgroup_subsys
*ss
= subsys
[i
];
711 unsigned long bit
= 1UL << i
;
712 if (bit
& added_bits
) {
713 /* We're binding this subsystem to this hierarchy */
714 BUG_ON(cgrp
->subsys
[i
]);
715 BUG_ON(!dummytop
->subsys
[i
]);
716 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
717 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
718 cgrp
->subsys
[i
]->cgroup
= cgrp
;
719 list_add(&ss
->sibling
, &root
->subsys_list
);
720 rcu_assign_pointer(ss
->root
, root
);
724 } else if (bit
& removed_bits
) {
725 /* We're removing this subsystem */
726 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
727 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
729 ss
->bind(ss
, dummytop
);
730 dummytop
->subsys
[i
]->cgroup
= dummytop
;
731 cgrp
->subsys
[i
] = NULL
;
732 rcu_assign_pointer(subsys
[i
]->root
, &rootnode
);
733 list_del(&ss
->sibling
);
734 } else if (bit
& final_bits
) {
735 /* Subsystem state should already exist */
736 BUG_ON(!cgrp
->subsys
[i
]);
738 /* Subsystem state shouldn't exist */
739 BUG_ON(cgrp
->subsys
[i
]);
742 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
748 static int cgroup_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
750 struct cgroupfs_root
*root
= vfs
->mnt_sb
->s_fs_info
;
751 struct cgroup_subsys
*ss
;
753 mutex_lock(&cgroup_mutex
);
754 for_each_subsys(root
, ss
)
755 seq_printf(seq
, ",%s", ss
->name
);
756 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
757 seq_puts(seq
, ",noprefix");
758 if (strlen(root
->release_agent_path
))
759 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
760 mutex_unlock(&cgroup_mutex
);
764 struct cgroup_sb_opts
{
765 unsigned long subsys_bits
;
770 /* Convert a hierarchy specifier into a bitmask of subsystems and
772 static int parse_cgroupfs_options(char *data
,
773 struct cgroup_sb_opts
*opts
)
775 char *token
, *o
= data
?: "all";
777 opts
->subsys_bits
= 0;
779 opts
->release_agent
= NULL
;
781 while ((token
= strsep(&o
, ",")) != NULL
) {
784 if (!strcmp(token
, "all")) {
785 opts
->subsys_bits
= (1 << CGROUP_SUBSYS_COUNT
) - 1;
786 } else if (!strcmp(token
, "noprefix")) {
787 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
788 } else if (!strncmp(token
, "release_agent=", 14)) {
789 /* Specifying two release agents is forbidden */
790 if (opts
->release_agent
)
792 opts
->release_agent
= kzalloc(PATH_MAX
, GFP_KERNEL
);
793 if (!opts
->release_agent
)
795 strncpy(opts
->release_agent
, token
+ 14, PATH_MAX
- 1);
796 opts
->release_agent
[PATH_MAX
- 1] = 0;
798 struct cgroup_subsys
*ss
;
800 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
802 if (!strcmp(token
, ss
->name
)) {
803 set_bit(i
, &opts
->subsys_bits
);
807 if (i
== CGROUP_SUBSYS_COUNT
)
812 /* We can't have an empty hierarchy */
813 if (!opts
->subsys_bits
)
819 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
822 struct cgroupfs_root
*root
= sb
->s_fs_info
;
823 struct cgroup
*cgrp
= &root
->top_cgroup
;
824 struct cgroup_sb_opts opts
;
826 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
827 mutex_lock(&cgroup_mutex
);
829 /* See what subsystems are wanted */
830 ret
= parse_cgroupfs_options(data
, &opts
);
834 /* Don't allow flags to change at remount */
835 if (opts
.flags
!= root
->flags
) {
840 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
842 /* (re)populate subsystem files */
844 cgroup_populate_dir(cgrp
);
846 if (opts
.release_agent
)
847 strcpy(root
->release_agent_path
, opts
.release_agent
);
849 if (opts
.release_agent
)
850 kfree(opts
.release_agent
);
851 mutex_unlock(&cgroup_mutex
);
852 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
856 static struct super_operations cgroup_ops
= {
857 .statfs
= simple_statfs
,
858 .drop_inode
= generic_delete_inode
,
859 .show_options
= cgroup_show_options
,
860 .remount_fs
= cgroup_remount
,
863 static void init_cgroup_root(struct cgroupfs_root
*root
)
865 struct cgroup
*cgrp
= &root
->top_cgroup
;
866 INIT_LIST_HEAD(&root
->subsys_list
);
867 INIT_LIST_HEAD(&root
->root_list
);
868 root
->number_of_cgroups
= 1;
870 cgrp
->top_cgroup
= cgrp
;
871 INIT_LIST_HEAD(&cgrp
->sibling
);
872 INIT_LIST_HEAD(&cgrp
->children
);
873 INIT_LIST_HEAD(&cgrp
->css_sets
);
874 INIT_LIST_HEAD(&cgrp
->release_list
);
877 static int cgroup_test_super(struct super_block
*sb
, void *data
)
879 struct cgroupfs_root
*new = data
;
880 struct cgroupfs_root
*root
= sb
->s_fs_info
;
882 /* First check subsystems */
883 if (new->subsys_bits
!= root
->subsys_bits
)
886 /* Next check flags */
887 if (new->flags
!= root
->flags
)
893 static int cgroup_set_super(struct super_block
*sb
, void *data
)
896 struct cgroupfs_root
*root
= data
;
898 ret
= set_anon_super(sb
, NULL
);
902 sb
->s_fs_info
= root
;
905 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
906 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
907 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
908 sb
->s_op
= &cgroup_ops
;
913 static int cgroup_get_rootdir(struct super_block
*sb
)
915 struct inode
*inode
=
916 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
917 struct dentry
*dentry
;
922 inode
->i_op
= &simple_dir_inode_operations
;
923 inode
->i_fop
= &simple_dir_operations
;
924 inode
->i_op
= &cgroup_dir_inode_operations
;
925 /* directories start off with i_nlink == 2 (for "." entry) */
927 dentry
= d_alloc_root(inode
);
936 static int cgroup_get_sb(struct file_system_type
*fs_type
,
937 int flags
, const char *unused_dev_name
,
938 void *data
, struct vfsmount
*mnt
)
940 struct cgroup_sb_opts opts
;
942 struct super_block
*sb
;
943 struct cgroupfs_root
*root
;
944 struct list_head tmp_cg_links
, *l
;
945 INIT_LIST_HEAD(&tmp_cg_links
);
947 /* First find the desired set of subsystems */
948 ret
= parse_cgroupfs_options(data
, &opts
);
950 if (opts
.release_agent
)
951 kfree(opts
.release_agent
);
955 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
957 if (opts
.release_agent
)
958 kfree(opts
.release_agent
);
962 init_cgroup_root(root
);
963 root
->subsys_bits
= opts
.subsys_bits
;
964 root
->flags
= opts
.flags
;
965 if (opts
.release_agent
) {
966 strcpy(root
->release_agent_path
, opts
.release_agent
);
967 kfree(opts
.release_agent
);
970 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, root
);
977 if (sb
->s_fs_info
!= root
) {
978 /* Reusing an existing superblock */
979 BUG_ON(sb
->s_root
== NULL
);
984 struct cgroup
*cgrp
= &root
->top_cgroup
;
987 BUG_ON(sb
->s_root
!= NULL
);
989 ret
= cgroup_get_rootdir(sb
);
992 inode
= sb
->s_root
->d_inode
;
994 mutex_lock(&inode
->i_mutex
);
995 mutex_lock(&cgroup_mutex
);
998 * We're accessing css_set_count without locking
999 * css_set_lock here, but that's OK - it can only be
1000 * increased by someone holding cgroup_lock, and
1001 * that's us. The worst that can happen is that we
1002 * have some link structures left over
1004 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1006 mutex_unlock(&cgroup_mutex
);
1007 mutex_unlock(&inode
->i_mutex
);
1008 goto drop_new_super
;
1011 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1012 if (ret
== -EBUSY
) {
1013 mutex_unlock(&cgroup_mutex
);
1014 mutex_unlock(&inode
->i_mutex
);
1015 goto drop_new_super
;
1018 /* EBUSY should be the only error here */
1021 list_add(&root
->root_list
, &roots
);
1024 sb
->s_root
->d_fsdata
= &root
->top_cgroup
;
1025 root
->top_cgroup
.dentry
= sb
->s_root
;
1027 /* Link the top cgroup in this hierarchy into all
1028 * the css_set objects */
1029 write_lock(&css_set_lock
);
1030 l
= &init_css_set
.list
;
1033 struct cg_cgroup_link
*link
;
1034 cg
= list_entry(l
, struct css_set
, list
);
1035 BUG_ON(list_empty(&tmp_cg_links
));
1036 link
= list_entry(tmp_cg_links
.next
,
1037 struct cg_cgroup_link
,
1039 list_del(&link
->cgrp_link_list
);
1041 list_add(&link
->cgrp_link_list
,
1042 &root
->top_cgroup
.css_sets
);
1043 list_add(&link
->cg_link_list
, &cg
->cg_links
);
1045 } while (l
!= &init_css_set
.list
);
1046 write_unlock(&css_set_lock
);
1048 free_cg_links(&tmp_cg_links
);
1050 BUG_ON(!list_empty(&cgrp
->sibling
));
1051 BUG_ON(!list_empty(&cgrp
->children
));
1052 BUG_ON(root
->number_of_cgroups
!= 1);
1054 cgroup_populate_dir(cgrp
);
1055 mutex_unlock(&inode
->i_mutex
);
1056 mutex_unlock(&cgroup_mutex
);
1059 return simple_set_mnt(mnt
, sb
);
1062 up_write(&sb
->s_umount
);
1063 deactivate_super(sb
);
1064 free_cg_links(&tmp_cg_links
);
1068 static void cgroup_kill_sb(struct super_block
*sb
) {
1069 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1070 struct cgroup
*cgrp
= &root
->top_cgroup
;
1075 BUG_ON(root
->number_of_cgroups
!= 1);
1076 BUG_ON(!list_empty(&cgrp
->children
));
1077 BUG_ON(!list_empty(&cgrp
->sibling
));
1079 mutex_lock(&cgroup_mutex
);
1081 /* Rebind all subsystems back to the default hierarchy */
1082 ret
= rebind_subsystems(root
, 0);
1083 /* Shouldn't be able to fail ... */
1087 * Release all the links from css_sets to this hierarchy's
1090 write_lock(&css_set_lock
);
1091 while (!list_empty(&cgrp
->css_sets
)) {
1092 struct cg_cgroup_link
*link
;
1093 link
= list_entry(cgrp
->css_sets
.next
,
1094 struct cg_cgroup_link
, cgrp_link_list
);
1095 list_del(&link
->cg_link_list
);
1096 list_del(&link
->cgrp_link_list
);
1099 write_unlock(&css_set_lock
);
1101 if (!list_empty(&root
->root_list
)) {
1102 list_del(&root
->root_list
);
1105 mutex_unlock(&cgroup_mutex
);
1108 kill_litter_super(sb
);
1111 static struct file_system_type cgroup_fs_type
= {
1113 .get_sb
= cgroup_get_sb
,
1114 .kill_sb
= cgroup_kill_sb
,
1117 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1119 return dentry
->d_fsdata
;
1122 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1124 return dentry
->d_fsdata
;
1128 * cgroup_path - generate the path of a cgroup
1129 * @cgrp: the cgroup in question
1130 * @buf: the buffer to write the path into
1131 * @buflen: the length of the buffer
1133 * Called with cgroup_mutex held. Writes path of cgroup into buf.
1134 * Returns 0 on success, -errno on error.
1136 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1140 if (cgrp
== dummytop
) {
1142 * Inactive subsystems have no dentry for their root
1149 start
= buf
+ buflen
;
1153 int len
= cgrp
->dentry
->d_name
.len
;
1154 if ((start
-= len
) < buf
)
1155 return -ENAMETOOLONG
;
1156 memcpy(start
, cgrp
->dentry
->d_name
.name
, len
);
1157 cgrp
= cgrp
->parent
;
1163 return -ENAMETOOLONG
;
1166 memmove(buf
, start
, buf
+ buflen
- start
);
1171 * Return the first subsystem attached to a cgroup's hierarchy, and
1175 static void get_first_subsys(const struct cgroup
*cgrp
,
1176 struct cgroup_subsys_state
**css
, int *subsys_id
)
1178 const struct cgroupfs_root
*root
= cgrp
->root
;
1179 const struct cgroup_subsys
*test_ss
;
1180 BUG_ON(list_empty(&root
->subsys_list
));
1181 test_ss
= list_entry(root
->subsys_list
.next
,
1182 struct cgroup_subsys
, sibling
);
1184 *css
= cgrp
->subsys
[test_ss
->subsys_id
];
1188 *subsys_id
= test_ss
->subsys_id
;
1192 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1193 * @cgrp: the cgroup the task is attaching to
1194 * @tsk: the task to be attached
1196 * Call holding cgroup_mutex. May take task_lock of
1197 * the task 'tsk' during call.
1199 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1202 struct cgroup_subsys
*ss
;
1203 struct cgroup
*oldcgrp
;
1204 struct css_set
*cg
= tsk
->cgroups
;
1205 struct css_set
*newcg
;
1206 struct cgroupfs_root
*root
= cgrp
->root
;
1209 get_first_subsys(cgrp
, NULL
, &subsys_id
);
1211 /* Nothing to do if the task is already in that cgroup */
1212 oldcgrp
= task_cgroup(tsk
, subsys_id
);
1213 if (cgrp
== oldcgrp
)
1216 for_each_subsys(root
, ss
) {
1217 if (ss
->can_attach
) {
1218 retval
= ss
->can_attach(ss
, cgrp
, tsk
);
1225 * Locate or allocate a new css_set for this task,
1226 * based on its final set of cgroups
1228 newcg
= find_css_set(cg
, cgrp
);
1233 if (tsk
->flags
& PF_EXITING
) {
1238 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1241 /* Update the css_set linked lists if we're using them */
1242 write_lock(&css_set_lock
);
1243 if (!list_empty(&tsk
->cg_list
)) {
1244 list_del(&tsk
->cg_list
);
1245 list_add(&tsk
->cg_list
, &newcg
->tasks
);
1247 write_unlock(&css_set_lock
);
1249 for_each_subsys(root
, ss
) {
1251 ss
->attach(ss
, cgrp
, oldcgrp
, tsk
);
1253 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1260 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with
1261 * cgroup_mutex, may take task_lock of task
1263 static int attach_task_by_pid(struct cgroup
*cgrp
, char *pidbuf
)
1266 struct task_struct
*tsk
;
1269 if (sscanf(pidbuf
, "%d", &pid
) != 1)
1274 tsk
= find_task_by_vpid(pid
);
1275 if (!tsk
|| tsk
->flags
& PF_EXITING
) {
1279 get_task_struct(tsk
);
1282 if ((current
->euid
) && (current
->euid
!= tsk
->uid
)
1283 && (current
->euid
!= tsk
->suid
)) {
1284 put_task_struct(tsk
);
1289 get_task_struct(tsk
);
1292 ret
= cgroup_attach_task(cgrp
, tsk
);
1293 put_task_struct(tsk
);
1297 /* The various types of files and directories in a cgroup file system */
1298 enum cgroup_filetype
{
1302 FILE_NOTIFY_ON_RELEASE
,
1307 static ssize_t
cgroup_write_uint(struct cgroup
*cgrp
, struct cftype
*cft
,
1309 const char __user
*userbuf
,
1310 size_t nbytes
, loff_t
*unused_ppos
)
1319 if (nbytes
>= sizeof(buffer
))
1321 if (copy_from_user(buffer
, userbuf
, nbytes
))
1324 buffer
[nbytes
] = 0; /* nul-terminate */
1326 /* strip newline if necessary */
1327 if (nbytes
&& (buffer
[nbytes
-1] == '\n'))
1328 buffer
[nbytes
-1] = 0;
1329 val
= simple_strtoull(buffer
, &end
, 0);
1333 /* Pass to subsystem */
1334 retval
= cft
->write_uint(cgrp
, cft
, val
);
1340 static ssize_t
cgroup_common_file_write(struct cgroup
*cgrp
,
1343 const char __user
*userbuf
,
1344 size_t nbytes
, loff_t
*unused_ppos
)
1346 enum cgroup_filetype type
= cft
->private;
1350 if (nbytes
>= PATH_MAX
)
1353 /* +1 for nul-terminator */
1354 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
1358 if (copy_from_user(buffer
, userbuf
, nbytes
)) {
1362 buffer
[nbytes
] = 0; /* nul-terminate */
1363 strstrip(buffer
); /* strip -just- trailing whitespace */
1365 mutex_lock(&cgroup_mutex
);
1368 * This was already checked for in cgroup_file_write(), but
1369 * check again now we're holding cgroup_mutex.
1371 if (cgroup_is_removed(cgrp
)) {
1378 retval
= attach_task_by_pid(cgrp
, buffer
);
1380 case FILE_NOTIFY_ON_RELEASE
:
1381 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
1382 if (simple_strtoul(buffer
, NULL
, 10) != 0)
1383 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
1385 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
1387 case FILE_RELEASE_AGENT
:
1388 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
1389 strcpy(cgrp
->root
->release_agent_path
, buffer
);
1399 mutex_unlock(&cgroup_mutex
);
1405 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
1406 size_t nbytes
, loff_t
*ppos
)
1408 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1409 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
1411 if (!cft
|| cgroup_is_removed(cgrp
))
1414 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1415 if (cft
->write_uint
)
1416 return cgroup_write_uint(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1420 static ssize_t
cgroup_read_uint(struct cgroup
*cgrp
, struct cftype
*cft
,
1422 char __user
*buf
, size_t nbytes
,
1426 u64 val
= cft
->read_uint(cgrp
, cft
);
1427 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
1429 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
1432 static ssize_t
cgroup_common_file_read(struct cgroup
*cgrp
,
1436 size_t nbytes
, loff_t
*ppos
)
1438 enum cgroup_filetype type
= cft
->private;
1443 if (!(page
= (char *)__get_free_page(GFP_KERNEL
)))
1449 case FILE_RELEASE_AGENT
:
1451 struct cgroupfs_root
*root
;
1453 mutex_lock(&cgroup_mutex
);
1455 n
= strnlen(root
->release_agent_path
,
1456 sizeof(root
->release_agent_path
));
1457 n
= min(n
, (size_t) PAGE_SIZE
);
1458 strncpy(s
, root
->release_agent_path
, n
);
1459 mutex_unlock(&cgroup_mutex
);
1469 retval
= simple_read_from_buffer(buf
, nbytes
, ppos
, page
, s
- page
);
1471 free_page((unsigned long)page
);
1475 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
1476 size_t nbytes
, loff_t
*ppos
)
1478 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1479 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
1481 if (!cft
|| cgroup_is_removed(cgrp
))
1485 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1487 return cgroup_read_uint(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1491 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
1496 err
= generic_file_open(inode
, file
);
1500 cft
= __d_cft(file
->f_dentry
);
1504 err
= cft
->open(inode
, file
);
1511 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
1513 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1515 return cft
->release(inode
, file
);
1520 * cgroup_rename - Only allow simple rename of directories in place.
1522 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1523 struct inode
*new_dir
, struct dentry
*new_dentry
)
1525 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
1527 if (new_dentry
->d_inode
)
1529 if (old_dir
!= new_dir
)
1531 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
1534 static struct file_operations cgroup_file_operations
= {
1535 .read
= cgroup_file_read
,
1536 .write
= cgroup_file_write
,
1537 .llseek
= generic_file_llseek
,
1538 .open
= cgroup_file_open
,
1539 .release
= cgroup_file_release
,
1542 static struct inode_operations cgroup_dir_inode_operations
= {
1543 .lookup
= simple_lookup
,
1544 .mkdir
= cgroup_mkdir
,
1545 .rmdir
= cgroup_rmdir
,
1546 .rename
= cgroup_rename
,
1549 static int cgroup_create_file(struct dentry
*dentry
, int mode
,
1550 struct super_block
*sb
)
1552 static struct dentry_operations cgroup_dops
= {
1553 .d_iput
= cgroup_diput
,
1556 struct inode
*inode
;
1560 if (dentry
->d_inode
)
1563 inode
= cgroup_new_inode(mode
, sb
);
1567 if (S_ISDIR(mode
)) {
1568 inode
->i_op
= &cgroup_dir_inode_operations
;
1569 inode
->i_fop
= &simple_dir_operations
;
1571 /* start off with i_nlink == 2 (for "." entry) */
1574 /* start with the directory inode held, so that we can
1575 * populate it without racing with another mkdir */
1576 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
1577 } else if (S_ISREG(mode
)) {
1579 inode
->i_fop
= &cgroup_file_operations
;
1581 dentry
->d_op
= &cgroup_dops
;
1582 d_instantiate(dentry
, inode
);
1583 dget(dentry
); /* Extra count - pin the dentry in core */
1588 * cgroup_create_dir - create a directory for an object.
1589 * @cgrp: the cgroup we create the directory for. It must have a valid
1590 * ->parent field. And we are going to fill its ->dentry field.
1591 * @dentry: dentry of the new cgroup
1592 * @mode: mode to set on new directory.
1594 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
1597 struct dentry
*parent
;
1600 parent
= cgrp
->parent
->dentry
;
1601 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
1603 dentry
->d_fsdata
= cgrp
;
1604 inc_nlink(parent
->d_inode
);
1605 cgrp
->dentry
= dentry
;
1613 int cgroup_add_file(struct cgroup
*cgrp
,
1614 struct cgroup_subsys
*subsys
,
1615 const struct cftype
*cft
)
1617 struct dentry
*dir
= cgrp
->dentry
;
1618 struct dentry
*dentry
;
1621 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
1622 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
1623 strcpy(name
, subsys
->name
);
1626 strcat(name
, cft
->name
);
1627 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
1628 dentry
= lookup_one_len(name
, dir
, strlen(name
));
1629 if (!IS_ERR(dentry
)) {
1630 error
= cgroup_create_file(dentry
, 0644 | S_IFREG
,
1633 dentry
->d_fsdata
= (void *)cft
;
1636 error
= PTR_ERR(dentry
);
1640 int cgroup_add_files(struct cgroup
*cgrp
,
1641 struct cgroup_subsys
*subsys
,
1642 const struct cftype cft
[],
1646 for (i
= 0; i
< count
; i
++) {
1647 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
1655 * cgroup_task_count - count the number of tasks in a cgroup.
1656 * @cgrp: the cgroup in question
1658 * Return the number of tasks in the cgroup.
1660 int cgroup_task_count(const struct cgroup
*cgrp
)
1663 struct list_head
*l
;
1665 read_lock(&css_set_lock
);
1666 l
= cgrp
->css_sets
.next
;
1667 while (l
!= &cgrp
->css_sets
) {
1668 struct cg_cgroup_link
*link
=
1669 list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
1670 count
+= atomic_read(&link
->cg
->ref
.refcount
);
1673 read_unlock(&css_set_lock
);
1678 * Advance a list_head iterator. The iterator should be positioned at
1679 * the start of a css_set
1681 static void cgroup_advance_iter(struct cgroup
*cgrp
,
1682 struct cgroup_iter
*it
)
1684 struct list_head
*l
= it
->cg_link
;
1685 struct cg_cgroup_link
*link
;
1688 /* Advance to the next non-empty css_set */
1691 if (l
== &cgrp
->css_sets
) {
1695 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
1697 } while (list_empty(&cg
->tasks
));
1699 it
->task
= cg
->tasks
.next
;
1703 * To reduce the fork() overhead for systems that are not actually
1704 * using their cgroups capability, we don't maintain the lists running
1705 * through each css_set to its tasks until we see the list actually
1706 * used - in other words after the first call to cgroup_iter_start().
1708 * The tasklist_lock is not held here, as do_each_thread() and
1709 * while_each_thread() are protected by RCU.
1711 void cgroup_enable_task_cg_lists(void)
1713 struct task_struct
*p
, *g
;
1714 write_lock(&css_set_lock
);
1715 use_task_css_set_links
= 1;
1716 do_each_thread(g
, p
) {
1718 if (list_empty(&p
->cg_list
))
1719 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
1721 } while_each_thread(g
, p
);
1722 write_unlock(&css_set_lock
);
1725 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
1728 * The first time anyone tries to iterate across a cgroup,
1729 * we need to enable the list linking each css_set to its
1730 * tasks, and fix up all existing tasks.
1732 if (!use_task_css_set_links
)
1733 cgroup_enable_task_cg_lists();
1735 read_lock(&css_set_lock
);
1736 it
->cg_link
= &cgrp
->css_sets
;
1737 cgroup_advance_iter(cgrp
, it
);
1740 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
1741 struct cgroup_iter
*it
)
1743 struct task_struct
*res
;
1744 struct list_head
*l
= it
->task
;
1746 /* If the iterator cg is NULL, we have no tasks */
1749 res
= list_entry(l
, struct task_struct
, cg_list
);
1750 /* Advance iterator to find next entry */
1752 if (l
== &res
->cgroups
->tasks
) {
1753 /* We reached the end of this task list - move on to
1754 * the next cg_cgroup_link */
1755 cgroup_advance_iter(cgrp
, it
);
1762 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
1764 read_unlock(&css_set_lock
);
1767 static inline int started_after_time(struct task_struct
*t1
,
1768 struct timespec
*time
,
1769 struct task_struct
*t2
)
1771 int start_diff
= timespec_compare(&t1
->start_time
, time
);
1772 if (start_diff
> 0) {
1774 } else if (start_diff
< 0) {
1778 * Arbitrarily, if two processes started at the same
1779 * time, we'll say that the lower pointer value
1780 * started first. Note that t2 may have exited by now
1781 * so this may not be a valid pointer any longer, but
1782 * that's fine - it still serves to distinguish
1783 * between two tasks started (effectively) simultaneously.
1790 * This function is a callback from heap_insert() and is used to order
1792 * In this case we order the heap in descending task start time.
1794 static inline int started_after(void *p1
, void *p2
)
1796 struct task_struct
*t1
= p1
;
1797 struct task_struct
*t2
= p2
;
1798 return started_after_time(t1
, &t2
->start_time
, t2
);
1802 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
1803 * @scan: struct cgroup_scanner containing arguments for the scan
1805 * Arguments include pointers to callback functions test_task() and
1807 * Iterate through all the tasks in a cgroup, calling test_task() for each,
1808 * and if it returns true, call process_task() for it also.
1809 * The test_task pointer may be NULL, meaning always true (select all tasks).
1810 * Effectively duplicates cgroup_iter_{start,next,end}()
1811 * but does not lock css_set_lock for the call to process_task().
1812 * The struct cgroup_scanner may be embedded in any structure of the caller's
1814 * It is guaranteed that process_task() will act on every task that
1815 * is a member of the cgroup for the duration of this call. This
1816 * function may or may not call process_task() for tasks that exit
1817 * or move to a different cgroup during the call, or are forked or
1818 * move into the cgroup during the call.
1820 * Note that test_task() may be called with locks held, and may in some
1821 * situations be called multiple times for the same task, so it should
1823 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
1824 * pre-allocated and will be used for heap operations (and its "gt" member will
1825 * be overwritten), else a temporary heap will be used (allocation of which
1826 * may cause this function to fail).
1828 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
1831 struct cgroup_iter it
;
1832 struct task_struct
*p
, *dropped
;
1833 /* Never dereference latest_task, since it's not refcounted */
1834 struct task_struct
*latest_task
= NULL
;
1835 struct ptr_heap tmp_heap
;
1836 struct ptr_heap
*heap
;
1837 struct timespec latest_time
= { 0, 0 };
1840 /* The caller supplied our heap and pre-allocated its memory */
1842 heap
->gt
= &started_after
;
1844 /* We need to allocate our own heap memory */
1846 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
1848 /* cannot allocate the heap */
1854 * Scan tasks in the cgroup, using the scanner's "test_task" callback
1855 * to determine which are of interest, and using the scanner's
1856 * "process_task" callback to process any of them that need an update.
1857 * Since we don't want to hold any locks during the task updates,
1858 * gather tasks to be processed in a heap structure.
1859 * The heap is sorted by descending task start time.
1860 * If the statically-sized heap fills up, we overflow tasks that
1861 * started later, and in future iterations only consider tasks that
1862 * started after the latest task in the previous pass. This
1863 * guarantees forward progress and that we don't miss any tasks.
1866 cgroup_iter_start(scan
->cg
, &it
);
1867 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
1869 * Only affect tasks that qualify per the caller's callback,
1870 * if he provided one
1872 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
1875 * Only process tasks that started after the last task
1878 if (!started_after_time(p
, &latest_time
, latest_task
))
1880 dropped
= heap_insert(heap
, p
);
1881 if (dropped
== NULL
) {
1883 * The new task was inserted; the heap wasn't
1887 } else if (dropped
!= p
) {
1889 * The new task was inserted, and pushed out a
1893 put_task_struct(dropped
);
1896 * Else the new task was newer than anything already in
1897 * the heap and wasn't inserted
1900 cgroup_iter_end(scan
->cg
, &it
);
1903 for (i
= 0; i
< heap
->size
; i
++) {
1904 struct task_struct
*p
= heap
->ptrs
[i
];
1906 latest_time
= p
->start_time
;
1909 /* Process the task per the caller's callback */
1910 scan
->process_task(p
, scan
);
1914 * If we had to process any tasks at all, scan again
1915 * in case some of them were in the middle of forking
1916 * children that didn't get processed.
1917 * Not the most efficient way to do it, but it avoids
1918 * having to take callback_mutex in the fork path
1922 if (heap
== &tmp_heap
)
1923 heap_free(&tmp_heap
);
1928 * Stuff for reading the 'tasks' file.
1930 * Reading this file can return large amounts of data if a cgroup has
1931 * *lots* of attached tasks. So it may need several calls to read(),
1932 * but we cannot guarantee that the information we produce is correct
1933 * unless we produce it entirely atomically.
1935 * Upon tasks file open(), a struct ctr_struct is allocated, that
1936 * will have a pointer to an array (also allocated here). The struct
1937 * ctr_struct * is stored in file->private_data. Its resources will
1938 * be freed by release() when the file is closed. The array is used
1939 * to sprintf the PIDs and then used by read().
1947 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
1948 * 'cgrp'. Return actual number of pids loaded. No need to
1949 * task_lock(p) when reading out p->cgroup, since we're in an RCU
1950 * read section, so the css_set can't go away, and is
1951 * immutable after creation.
1953 static int pid_array_load(pid_t
*pidarray
, int npids
, struct cgroup
*cgrp
)
1956 struct cgroup_iter it
;
1957 struct task_struct
*tsk
;
1958 cgroup_iter_start(cgrp
, &it
);
1959 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
1960 if (unlikely(n
== npids
))
1962 pidarray
[n
++] = task_pid_vnr(tsk
);
1964 cgroup_iter_end(cgrp
, &it
);
1969 * cgroupstats_build - build and fill cgroupstats
1970 * @stats: cgroupstats to fill information into
1971 * @dentry: A dentry entry belonging to the cgroup for which stats have
1974 * Build and fill cgroupstats so that taskstats can export it to user
1977 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
1980 struct cgroup
*cgrp
;
1981 struct cgroup_iter it
;
1982 struct task_struct
*tsk
;
1984 * Validate dentry by checking the superblock operations
1986 if (dentry
->d_sb
->s_op
!= &cgroup_ops
)
1990 cgrp
= dentry
->d_fsdata
;
1993 cgroup_iter_start(cgrp
, &it
);
1994 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
1995 switch (tsk
->state
) {
1997 stats
->nr_running
++;
1999 case TASK_INTERRUPTIBLE
:
2000 stats
->nr_sleeping
++;
2002 case TASK_UNINTERRUPTIBLE
:
2003 stats
->nr_uninterruptible
++;
2006 stats
->nr_stopped
++;
2009 if (delayacct_is_task_waiting_on_io(tsk
))
2010 stats
->nr_io_wait
++;
2014 cgroup_iter_end(cgrp
, &it
);
2021 static int cmppid(const void *a
, const void *b
)
2023 return *(pid_t
*)a
- *(pid_t
*)b
;
2027 * Convert array 'a' of 'npids' pid_t's to a string of newline separated
2028 * decimal pids in 'buf'. Don't write more than 'sz' chars, but return
2029 * count 'cnt' of how many chars would be written if buf were large enough.
2031 static int pid_array_to_buf(char *buf
, int sz
, pid_t
*a
, int npids
)
2036 for (i
= 0; i
< npids
; i
++)
2037 cnt
+= snprintf(buf
+ cnt
, max(sz
- cnt
, 0), "%d\n", a
[i
]);
2042 * Handle an open on 'tasks' file. Prepare a buffer listing the
2043 * process id's of tasks currently attached to the cgroup being opened.
2045 * Does not require any specific cgroup mutexes, and does not take any.
2047 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
2049 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2050 struct ctr_struct
*ctr
;
2055 if (!(file
->f_mode
& FMODE_READ
))
2058 ctr
= kmalloc(sizeof(*ctr
), GFP_KERNEL
);
2063 * If cgroup gets more users after we read count, we won't have
2064 * enough space - tough. This race is indistinguishable to the
2065 * caller from the case that the additional cgroup users didn't
2066 * show up until sometime later on.
2068 npids
= cgroup_task_count(cgrp
);
2070 pidarray
= kmalloc(npids
* sizeof(pid_t
), GFP_KERNEL
);
2074 npids
= pid_array_load(pidarray
, npids
, cgrp
);
2075 sort(pidarray
, npids
, sizeof(pid_t
), cmppid
, NULL
);
2077 /* Call pid_array_to_buf() twice, first just to get bufsz */
2078 ctr
->bufsz
= pid_array_to_buf(&c
, sizeof(c
), pidarray
, npids
) + 1;
2079 ctr
->buf
= kmalloc(ctr
->bufsz
, GFP_KERNEL
);
2082 ctr
->bufsz
= pid_array_to_buf(ctr
->buf
, ctr
->bufsz
, pidarray
, npids
);
2089 file
->private_data
= ctr
;
2100 static ssize_t
cgroup_tasks_read(struct cgroup
*cgrp
,
2102 struct file
*file
, char __user
*buf
,
2103 size_t nbytes
, loff_t
*ppos
)
2105 struct ctr_struct
*ctr
= file
->private_data
;
2107 return simple_read_from_buffer(buf
, nbytes
, ppos
, ctr
->buf
, ctr
->bufsz
);
2110 static int cgroup_tasks_release(struct inode
*unused_inode
,
2113 struct ctr_struct
*ctr
;
2115 if (file
->f_mode
& FMODE_READ
) {
2116 ctr
= file
->private_data
;
2123 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
2126 return notify_on_release(cgrp
);
2129 static u64
cgroup_read_releasable(struct cgroup
*cgrp
, struct cftype
*cft
)
2131 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
2135 * for the common functions, 'private' gives the type of file
2137 static struct cftype files
[] = {
2140 .open
= cgroup_tasks_open
,
2141 .read
= cgroup_tasks_read
,
2142 .write
= cgroup_common_file_write
,
2143 .release
= cgroup_tasks_release
,
2144 .private = FILE_TASKLIST
,
2148 .name
= "notify_on_release",
2149 .read_uint
= cgroup_read_notify_on_release
,
2150 .write
= cgroup_common_file_write
,
2151 .private = FILE_NOTIFY_ON_RELEASE
,
2155 .name
= "releasable",
2156 .read_uint
= cgroup_read_releasable
,
2157 .private = FILE_RELEASABLE
,
2161 static struct cftype cft_release_agent
= {
2162 .name
= "release_agent",
2163 .read
= cgroup_common_file_read
,
2164 .write
= cgroup_common_file_write
,
2165 .private = FILE_RELEASE_AGENT
,
2168 static int cgroup_populate_dir(struct cgroup
*cgrp
)
2171 struct cgroup_subsys
*ss
;
2173 /* First clear out any existing files */
2174 cgroup_clear_directory(cgrp
->dentry
);
2176 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
2180 if (cgrp
== cgrp
->top_cgroup
) {
2181 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
2185 for_each_subsys(cgrp
->root
, ss
) {
2186 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
2193 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
2194 struct cgroup_subsys
*ss
,
2195 struct cgroup
*cgrp
)
2198 atomic_set(&css
->refcnt
, 0);
2200 if (cgrp
== dummytop
)
2201 set_bit(CSS_ROOT
, &css
->flags
);
2202 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
2203 cgrp
->subsys
[ss
->subsys_id
] = css
;
2207 * cgroup_create - create a cgroup
2208 * @parent: cgroup that will be parent of the new cgroup
2209 * @dentry: dentry of the new cgroup
2210 * @mode: mode to set on new inode
2212 * Must be called with the mutex on the parent inode held
2214 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
2217 struct cgroup
*cgrp
;
2218 struct cgroupfs_root
*root
= parent
->root
;
2220 struct cgroup_subsys
*ss
;
2221 struct super_block
*sb
= root
->sb
;
2223 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
2227 /* Grab a reference on the superblock so the hierarchy doesn't
2228 * get deleted on unmount if there are child cgroups. This
2229 * can be done outside cgroup_mutex, since the sb can't
2230 * disappear while someone has an open control file on the
2232 atomic_inc(&sb
->s_active
);
2234 mutex_lock(&cgroup_mutex
);
2237 INIT_LIST_HEAD(&cgrp
->sibling
);
2238 INIT_LIST_HEAD(&cgrp
->children
);
2239 INIT_LIST_HEAD(&cgrp
->css_sets
);
2240 INIT_LIST_HEAD(&cgrp
->release_list
);
2242 cgrp
->parent
= parent
;
2243 cgrp
->root
= parent
->root
;
2244 cgrp
->top_cgroup
= parent
->top_cgroup
;
2246 for_each_subsys(root
, ss
) {
2247 struct cgroup_subsys_state
*css
= ss
->create(ss
, cgrp
);
2252 init_cgroup_css(css
, ss
, cgrp
);
2255 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
2256 root
->number_of_cgroups
++;
2258 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
2262 /* The cgroup directory was pre-locked for us */
2263 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
2265 err
= cgroup_populate_dir(cgrp
);
2266 /* If err < 0, we have a half-filled directory - oh well ;) */
2268 mutex_unlock(&cgroup_mutex
);
2269 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
2275 list_del(&cgrp
->sibling
);
2276 root
->number_of_cgroups
--;
2280 for_each_subsys(root
, ss
) {
2281 if (cgrp
->subsys
[ss
->subsys_id
])
2282 ss
->destroy(ss
, cgrp
);
2285 mutex_unlock(&cgroup_mutex
);
2287 /* Release the reference count that we took on the superblock */
2288 deactivate_super(sb
);
2294 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2296 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
2298 /* the vfs holds inode->i_mutex already */
2299 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
2302 static inline int cgroup_has_css_refs(struct cgroup
*cgrp
)
2304 /* Check the reference count on each subsystem. Since we
2305 * already established that there are no tasks in the
2306 * cgroup, if the css refcount is also 0, then there should
2307 * be no outstanding references, so the subsystem is safe to
2308 * destroy. We scan across all subsystems rather than using
2309 * the per-hierarchy linked list of mounted subsystems since
2310 * we can be called via check_for_release() with no
2311 * synchronization other than RCU, and the subsystem linked
2312 * list isn't RCU-safe */
2314 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2315 struct cgroup_subsys
*ss
= subsys
[i
];
2316 struct cgroup_subsys_state
*css
;
2317 /* Skip subsystems not in this hierarchy */
2318 if (ss
->root
!= cgrp
->root
)
2320 css
= cgrp
->subsys
[ss
->subsys_id
];
2321 /* When called from check_for_release() it's possible
2322 * that by this point the cgroup has been removed
2323 * and the css deleted. But a false-positive doesn't
2324 * matter, since it can only happen if the cgroup
2325 * has been deleted and hence no longer needs the
2326 * release agent to be called anyway. */
2327 if (css
&& atomic_read(&css
->refcnt
))
2333 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
2335 struct cgroup
*cgrp
= dentry
->d_fsdata
;
2337 struct cgroup
*parent
;
2338 struct super_block
*sb
;
2339 struct cgroupfs_root
*root
;
2341 /* the vfs holds both inode->i_mutex already */
2343 mutex_lock(&cgroup_mutex
);
2344 if (atomic_read(&cgrp
->count
) != 0) {
2345 mutex_unlock(&cgroup_mutex
);
2348 if (!list_empty(&cgrp
->children
)) {
2349 mutex_unlock(&cgroup_mutex
);
2353 parent
= cgrp
->parent
;
2358 * Call pre_destroy handlers of subsys. Notify subsystems
2359 * that rmdir() request comes.
2361 cgroup_call_pre_destroy(cgrp
);
2363 if (cgroup_has_css_refs(cgrp
)) {
2364 mutex_unlock(&cgroup_mutex
);
2368 spin_lock(&release_list_lock
);
2369 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
2370 if (!list_empty(&cgrp
->release_list
))
2371 list_del(&cgrp
->release_list
);
2372 spin_unlock(&release_list_lock
);
2373 /* delete my sibling from parent->children */
2374 list_del(&cgrp
->sibling
);
2375 spin_lock(&cgrp
->dentry
->d_lock
);
2376 d
= dget(cgrp
->dentry
);
2377 cgrp
->dentry
= NULL
;
2378 spin_unlock(&d
->d_lock
);
2380 cgroup_d_remove_dir(d
);
2383 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
2384 check_for_release(parent
);
2386 mutex_unlock(&cgroup_mutex
);
2390 static void cgroup_init_subsys(struct cgroup_subsys
*ss
)
2392 struct cgroup_subsys_state
*css
;
2393 struct list_head
*l
;
2395 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
2397 /* Create the top cgroup state for this subsystem */
2398 ss
->root
= &rootnode
;
2399 css
= ss
->create(ss
, dummytop
);
2400 /* We don't handle early failures gracefully */
2401 BUG_ON(IS_ERR(css
));
2402 init_cgroup_css(css
, ss
, dummytop
);
2404 /* Update all cgroup groups to contain a subsys
2405 * pointer to this state - since the subsystem is
2406 * newly registered, all tasks and hence all cgroup
2407 * groups are in the subsystem's top cgroup. */
2408 write_lock(&css_set_lock
);
2409 l
= &init_css_set
.list
;
2411 struct css_set
*cg
=
2412 list_entry(l
, struct css_set
, list
);
2413 cg
->subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
2415 } while (l
!= &init_css_set
.list
);
2416 write_unlock(&css_set_lock
);
2418 /* If this subsystem requested that it be notified with fork
2419 * events, we should send it one now for every process in the
2422 struct task_struct
*g
, *p
;
2424 read_lock(&tasklist_lock
);
2425 do_each_thread(g
, p
) {
2427 } while_each_thread(g
, p
);
2428 read_unlock(&tasklist_lock
);
2431 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
2437 * cgroup_init_early - cgroup initialization at system boot
2439 * Initialize cgroups at system boot, and initialize any
2440 * subsystems that request early init.
2442 int __init
cgroup_init_early(void)
2445 kref_init(&init_css_set
.ref
);
2446 kref_get(&init_css_set
.ref
);
2447 INIT_LIST_HEAD(&init_css_set
.list
);
2448 INIT_LIST_HEAD(&init_css_set
.cg_links
);
2449 INIT_LIST_HEAD(&init_css_set
.tasks
);
2451 init_cgroup_root(&rootnode
);
2452 list_add(&rootnode
.root_list
, &roots
);
2454 init_task
.cgroups
= &init_css_set
;
2456 init_css_set_link
.cg
= &init_css_set
;
2457 list_add(&init_css_set_link
.cgrp_link_list
,
2458 &rootnode
.top_cgroup
.css_sets
);
2459 list_add(&init_css_set_link
.cg_link_list
,
2460 &init_css_set
.cg_links
);
2462 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2463 struct cgroup_subsys
*ss
= subsys
[i
];
2466 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
2467 BUG_ON(!ss
->create
);
2468 BUG_ON(!ss
->destroy
);
2469 if (ss
->subsys_id
!= i
) {
2470 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
2471 ss
->name
, ss
->subsys_id
);
2476 cgroup_init_subsys(ss
);
2482 * cgroup_init - cgroup initialization
2484 * Register cgroup filesystem and /proc file, and initialize
2485 * any subsystems that didn't request early init.
2487 int __init
cgroup_init(void)
2491 struct proc_dir_entry
*entry
;
2493 err
= bdi_init(&cgroup_backing_dev_info
);
2497 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2498 struct cgroup_subsys
*ss
= subsys
[i
];
2499 if (!ss
->early_init
)
2500 cgroup_init_subsys(ss
);
2503 err
= register_filesystem(&cgroup_fs_type
);
2507 entry
= create_proc_entry("cgroups", 0, NULL
);
2509 entry
->proc_fops
= &proc_cgroupstats_operations
;
2513 bdi_destroy(&cgroup_backing_dev_info
);
2519 * proc_cgroup_show()
2520 * - Print task's cgroup paths into seq_file, one line for each hierarchy
2521 * - Used for /proc/<pid>/cgroup.
2522 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
2523 * doesn't really matter if tsk->cgroup changes after we read it,
2524 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
2525 * anyway. No need to check that tsk->cgroup != NULL, thanks to
2526 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
2527 * cgroup to top_cgroup.
2530 /* TODO: Use a proper seq_file iterator */
2531 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
2534 struct task_struct
*tsk
;
2537 struct cgroupfs_root
*root
;
2540 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
2546 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
2552 mutex_lock(&cgroup_mutex
);
2554 for_each_root(root
) {
2555 struct cgroup_subsys
*ss
;
2556 struct cgroup
*cgrp
;
2560 /* Skip this hierarchy if it has no active subsystems */
2561 if (!root
->actual_subsys_bits
)
2563 for_each_subsys(root
, ss
)
2564 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
2566 get_first_subsys(&root
->top_cgroup
, NULL
, &subsys_id
);
2567 cgrp
= task_cgroup(tsk
, subsys_id
);
2568 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
2576 mutex_unlock(&cgroup_mutex
);
2577 put_task_struct(tsk
);
2584 static int cgroup_open(struct inode
*inode
, struct file
*file
)
2586 struct pid
*pid
= PROC_I(inode
)->pid
;
2587 return single_open(file
, proc_cgroup_show
, pid
);
2590 struct file_operations proc_cgroup_operations
= {
2591 .open
= cgroup_open
,
2593 .llseek
= seq_lseek
,
2594 .release
= single_release
,
2597 /* Display information about each subsystem and each hierarchy */
2598 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
2602 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\n");
2603 mutex_lock(&cgroup_mutex
);
2604 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2605 struct cgroup_subsys
*ss
= subsys
[i
];
2606 seq_printf(m
, "%s\t%lu\t%d\n",
2607 ss
->name
, ss
->root
->subsys_bits
,
2608 ss
->root
->number_of_cgroups
);
2610 mutex_unlock(&cgroup_mutex
);
2614 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
2616 return single_open(file
, proc_cgroupstats_show
, 0);
2619 static struct file_operations proc_cgroupstats_operations
= {
2620 .open
= cgroupstats_open
,
2622 .llseek
= seq_lseek
,
2623 .release
= single_release
,
2627 * cgroup_fork - attach newly forked task to its parents cgroup.
2628 * @child: pointer to task_struct of forking parent process.
2630 * Description: A task inherits its parent's cgroup at fork().
2632 * A pointer to the shared css_set was automatically copied in
2633 * fork.c by dup_task_struct(). However, we ignore that copy, since
2634 * it was not made under the protection of RCU or cgroup_mutex, so
2635 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
2636 * have already changed current->cgroups, allowing the previously
2637 * referenced cgroup group to be removed and freed.
2639 * At the point that cgroup_fork() is called, 'current' is the parent
2640 * task, and the passed argument 'child' points to the child task.
2642 void cgroup_fork(struct task_struct
*child
)
2645 child
->cgroups
= current
->cgroups
;
2646 get_css_set(child
->cgroups
);
2647 task_unlock(current
);
2648 INIT_LIST_HEAD(&child
->cg_list
);
2652 * cgroup_fork_callbacks - run fork callbacks
2653 * @child: the new task
2655 * Called on a new task very soon before adding it to the
2656 * tasklist. No need to take any locks since no-one can
2657 * be operating on this task.
2659 void cgroup_fork_callbacks(struct task_struct
*child
)
2661 if (need_forkexit_callback
) {
2663 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2664 struct cgroup_subsys
*ss
= subsys
[i
];
2666 ss
->fork(ss
, child
);
2672 * cgroup_post_fork - called on a new task after adding it to the task list
2673 * @child: the task in question
2675 * Adds the task to the list running through its css_set if necessary.
2676 * Has to be after the task is visible on the task list in case we race
2677 * with the first call to cgroup_iter_start() - to guarantee that the
2678 * new task ends up on its list.
2680 void cgroup_post_fork(struct task_struct
*child
)
2682 if (use_task_css_set_links
) {
2683 write_lock(&css_set_lock
);
2684 if (list_empty(&child
->cg_list
))
2685 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
2686 write_unlock(&css_set_lock
);
2690 * cgroup_exit - detach cgroup from exiting task
2691 * @tsk: pointer to task_struct of exiting process
2692 * @run_callback: run exit callbacks?
2694 * Description: Detach cgroup from @tsk and release it.
2696 * Note that cgroups marked notify_on_release force every task in
2697 * them to take the global cgroup_mutex mutex when exiting.
2698 * This could impact scaling on very large systems. Be reluctant to
2699 * use notify_on_release cgroups where very high task exit scaling
2700 * is required on large systems.
2702 * the_top_cgroup_hack:
2704 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
2706 * We call cgroup_exit() while the task is still competent to
2707 * handle notify_on_release(), then leave the task attached to the
2708 * root cgroup in each hierarchy for the remainder of its exit.
2710 * To do this properly, we would increment the reference count on
2711 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
2712 * code we would add a second cgroup function call, to drop that
2713 * reference. This would just create an unnecessary hot spot on
2714 * the top_cgroup reference count, to no avail.
2716 * Normally, holding a reference to a cgroup without bumping its
2717 * count is unsafe. The cgroup could go away, or someone could
2718 * attach us to a different cgroup, decrementing the count on
2719 * the first cgroup that we never incremented. But in this case,
2720 * top_cgroup isn't going away, and either task has PF_EXITING set,
2721 * which wards off any cgroup_attach_task() attempts, or task is a failed
2722 * fork, never visible to cgroup_attach_task.
2724 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
2729 if (run_callbacks
&& need_forkexit_callback
) {
2730 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2731 struct cgroup_subsys
*ss
= subsys
[i
];
2738 * Unlink from the css_set task list if necessary.
2739 * Optimistically check cg_list before taking
2742 if (!list_empty(&tsk
->cg_list
)) {
2743 write_lock(&css_set_lock
);
2744 if (!list_empty(&tsk
->cg_list
))
2745 list_del(&tsk
->cg_list
);
2746 write_unlock(&css_set_lock
);
2749 /* Reassign the task to the init_css_set. */
2752 tsk
->cgroups
= &init_css_set
;
2755 put_css_set_taskexit(cg
);
2759 * cgroup_clone - clone the cgroup the given subsystem is attached to
2760 * @tsk: the task to be moved
2761 * @subsys: the given subsystem
2763 * Duplicate the current cgroup in the hierarchy that the given
2764 * subsystem is attached to, and move this task into the new
2767 int cgroup_clone(struct task_struct
*tsk
, struct cgroup_subsys
*subsys
)
2769 struct dentry
*dentry
;
2771 char nodename
[MAX_CGROUP_TYPE_NAMELEN
];
2772 struct cgroup
*parent
, *child
;
2773 struct inode
*inode
;
2775 struct cgroupfs_root
*root
;
2776 struct cgroup_subsys
*ss
;
2778 /* We shouldn't be called by an unregistered subsystem */
2779 BUG_ON(!subsys
->active
);
2781 /* First figure out what hierarchy and cgroup we're dealing
2782 * with, and pin them so we can drop cgroup_mutex */
2783 mutex_lock(&cgroup_mutex
);
2785 root
= subsys
->root
;
2786 if (root
== &rootnode
) {
2788 "Not cloning cgroup for unused subsystem %s\n",
2790 mutex_unlock(&cgroup_mutex
);
2794 parent
= task_cgroup(tsk
, subsys
->subsys_id
);
2796 snprintf(nodename
, MAX_CGROUP_TYPE_NAMELEN
, "node_%d", tsk
->pid
);
2798 /* Pin the hierarchy */
2799 atomic_inc(&parent
->root
->sb
->s_active
);
2801 /* Keep the cgroup alive */
2803 mutex_unlock(&cgroup_mutex
);
2805 /* Now do the VFS work to create a cgroup */
2806 inode
= parent
->dentry
->d_inode
;
2808 /* Hold the parent directory mutex across this operation to
2809 * stop anyone else deleting the new cgroup */
2810 mutex_lock(&inode
->i_mutex
);
2811 dentry
= lookup_one_len(nodename
, parent
->dentry
, strlen(nodename
));
2812 if (IS_ERR(dentry
)) {
2814 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename
,
2816 ret
= PTR_ERR(dentry
);
2820 /* Create the cgroup directory, which also creates the cgroup */
2821 ret
= vfs_mkdir(inode
, dentry
, S_IFDIR
| 0755);
2822 child
= __d_cgrp(dentry
);
2826 "Failed to create cgroup %s: %d\n", nodename
,
2833 "Couldn't find new cgroup %s\n", nodename
);
2838 /* The cgroup now exists. Retake cgroup_mutex and check
2839 * that we're still in the same state that we thought we
2841 mutex_lock(&cgroup_mutex
);
2842 if ((root
!= subsys
->root
) ||
2843 (parent
!= task_cgroup(tsk
, subsys
->subsys_id
))) {
2844 /* Aargh, we raced ... */
2845 mutex_unlock(&inode
->i_mutex
);
2848 deactivate_super(parent
->root
->sb
);
2849 /* The cgroup is still accessible in the VFS, but
2850 * we're not going to try to rmdir() it at this
2853 "Race in cgroup_clone() - leaking cgroup %s\n",
2858 /* do any required auto-setup */
2859 for_each_subsys(root
, ss
) {
2861 ss
->post_clone(ss
, child
);
2864 /* All seems fine. Finish by moving the task into the new cgroup */
2865 ret
= cgroup_attach_task(child
, tsk
);
2866 mutex_unlock(&cgroup_mutex
);
2869 mutex_unlock(&inode
->i_mutex
);
2871 mutex_lock(&cgroup_mutex
);
2873 mutex_unlock(&cgroup_mutex
);
2874 deactivate_super(parent
->root
->sb
);
2879 * cgroup_is_descendant - see if @cgrp is a descendant of current task's cgrp
2880 * @cgrp: the cgroup in question
2882 * See if @cgrp is a descendant of the current task's cgroup in
2883 * the appropriate hierarchy.
2885 * If we are sending in dummytop, then presumably we are creating
2886 * the top cgroup in the subsystem.
2888 * Called only by the ns (nsproxy) cgroup.
2890 int cgroup_is_descendant(const struct cgroup
*cgrp
)
2893 struct cgroup
*target
;
2896 if (cgrp
== dummytop
)
2899 get_first_subsys(cgrp
, NULL
, &subsys_id
);
2900 target
= task_cgroup(current
, subsys_id
);
2901 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
2902 cgrp
= cgrp
->parent
;
2903 ret
= (cgrp
== target
);
2907 static void check_for_release(struct cgroup
*cgrp
)
2909 /* All of these checks rely on RCU to keep the cgroup
2910 * structure alive */
2911 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
2912 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
2913 /* Control Group is currently removeable. If it's not
2914 * already queued for a userspace notification, queue
2916 int need_schedule_work
= 0;
2917 spin_lock(&release_list_lock
);
2918 if (!cgroup_is_removed(cgrp
) &&
2919 list_empty(&cgrp
->release_list
)) {
2920 list_add(&cgrp
->release_list
, &release_list
);
2921 need_schedule_work
= 1;
2923 spin_unlock(&release_list_lock
);
2924 if (need_schedule_work
)
2925 schedule_work(&release_agent_work
);
2929 void __css_put(struct cgroup_subsys_state
*css
)
2931 struct cgroup
*cgrp
= css
->cgroup
;
2933 if (atomic_dec_and_test(&css
->refcnt
) && notify_on_release(cgrp
)) {
2934 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
2935 check_for_release(cgrp
);
2941 * Notify userspace when a cgroup is released, by running the
2942 * configured release agent with the name of the cgroup (path
2943 * relative to the root of cgroup file system) as the argument.
2945 * Most likely, this user command will try to rmdir this cgroup.
2947 * This races with the possibility that some other task will be
2948 * attached to this cgroup before it is removed, or that some other
2949 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
2950 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
2951 * unused, and this cgroup will be reprieved from its death sentence,
2952 * to continue to serve a useful existence. Next time it's released,
2953 * we will get notified again, if it still has 'notify_on_release' set.
2955 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
2956 * means only wait until the task is successfully execve()'d. The
2957 * separate release agent task is forked by call_usermodehelper(),
2958 * then control in this thread returns here, without waiting for the
2959 * release agent task. We don't bother to wait because the caller of
2960 * this routine has no use for the exit status of the release agent
2961 * task, so no sense holding our caller up for that.
2963 static void cgroup_release_agent(struct work_struct
*work
)
2965 BUG_ON(work
!= &release_agent_work
);
2966 mutex_lock(&cgroup_mutex
);
2967 spin_lock(&release_list_lock
);
2968 while (!list_empty(&release_list
)) {
2969 char *argv
[3], *envp
[3];
2972 struct cgroup
*cgrp
= list_entry(release_list
.next
,
2975 list_del_init(&cgrp
->release_list
);
2976 spin_unlock(&release_list_lock
);
2977 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
2979 spin_lock(&release_list_lock
);
2983 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0) {
2985 spin_lock(&release_list_lock
);
2990 argv
[i
++] = cgrp
->root
->release_agent_path
;
2991 argv
[i
++] = (char *)pathbuf
;
2995 /* minimal command environment */
2996 envp
[i
++] = "HOME=/";
2997 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
3000 /* Drop the lock while we invoke the usermode helper,
3001 * since the exec could involve hitting disk and hence
3002 * be a slow process */
3003 mutex_unlock(&cgroup_mutex
);
3004 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
3006 mutex_lock(&cgroup_mutex
);
3007 spin_lock(&release_list_lock
);
3009 spin_unlock(&release_list_lock
);
3010 mutex_unlock(&cgroup_mutex
);