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>
47 #include <linux/hash.h>
48 #include <linux/namei.h>
50 #include <asm/atomic.h>
52 static DEFINE_MUTEX(cgroup_mutex
);
54 /* Generate an array of cgroup subsystem pointers */
55 #define SUBSYS(_x) &_x ## _subsys,
57 static struct cgroup_subsys
*subsys
[] = {
58 #include <linux/cgroup_subsys.h>
62 * A cgroupfs_root represents the root of a cgroup hierarchy,
63 * and may be associated with a superblock to form an active
66 struct cgroupfs_root
{
67 struct super_block
*sb
;
70 * The bitmask of subsystems intended to be attached to this
73 unsigned long subsys_bits
;
75 /* The bitmask of subsystems currently attached to this hierarchy */
76 unsigned long actual_subsys_bits
;
78 /* A list running through the attached subsystems */
79 struct list_head subsys_list
;
81 /* The root cgroup for this hierarchy */
82 struct cgroup top_cgroup
;
84 /* Tracks how many cgroups are currently defined in hierarchy.*/
85 int number_of_cgroups
;
87 /* A list running through the mounted hierarchies */
88 struct list_head root_list
;
90 /* Hierarchy-specific flags */
93 /* The path to use for release notifications. */
94 char release_agent_path
[PATH_MAX
];
99 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
100 * subsystems that are otherwise unattached - it never has more than a
101 * single cgroup, and all tasks are part of that cgroup.
103 static struct cgroupfs_root rootnode
;
105 /* The list of hierarchy roots */
107 static LIST_HEAD(roots
);
108 static int root_count
;
110 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
111 #define dummytop (&rootnode.top_cgroup)
113 /* This flag indicates whether tasks in the fork and exit paths should
114 * check for fork/exit handlers to call. This avoids us having to do
115 * extra work in the fork/exit path if none of the subsystems need to
118 static int need_forkexit_callback __read_mostly
;
119 static int need_mm_owner_callback __read_mostly
;
121 /* convenient tests for these bits */
122 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
124 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
127 /* bits in struct cgroupfs_root flags field */
129 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
132 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
135 (1 << CGRP_RELEASABLE
) |
136 (1 << CGRP_NOTIFY_ON_RELEASE
);
137 return (cgrp
->flags
& bits
) == bits
;
140 static int notify_on_release(const struct cgroup
*cgrp
)
142 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
146 * for_each_subsys() allows you to iterate on each subsystem attached to
147 * an active hierarchy
149 #define for_each_subsys(_root, _ss) \
150 list_for_each_entry(_ss, &_root->subsys_list, sibling)
152 /* for_each_root() allows you to iterate across the active hierarchies */
153 #define for_each_root(_root) \
154 list_for_each_entry(_root, &roots, root_list)
156 /* the list of cgroups eligible for automatic release. Protected by
157 * release_list_lock */
158 static LIST_HEAD(release_list
);
159 static DEFINE_SPINLOCK(release_list_lock
);
160 static void cgroup_release_agent(struct work_struct
*work
);
161 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
162 static void check_for_release(struct cgroup
*cgrp
);
164 /* Link structure for associating css_set objects with cgroups */
165 struct cg_cgroup_link
{
167 * List running through cg_cgroup_links associated with a
168 * cgroup, anchored on cgroup->css_sets
170 struct list_head cgrp_link_list
;
172 * List running through cg_cgroup_links pointing at a
173 * single css_set object, anchored on css_set->cg_links
175 struct list_head cg_link_list
;
179 /* The default css_set - used by init and its children prior to any
180 * hierarchies being mounted. It contains a pointer to the root state
181 * for each subsystem. Also used to anchor the list of css_sets. Not
182 * reference-counted, to improve performance when child cgroups
183 * haven't been created.
186 static struct css_set init_css_set
;
187 static struct cg_cgroup_link init_css_set_link
;
189 /* css_set_lock protects the list of css_set objects, and the
190 * chain of tasks off each css_set. Nests outside task->alloc_lock
191 * due to cgroup_iter_start() */
192 static DEFINE_RWLOCK(css_set_lock
);
193 static int css_set_count
;
195 /* hash table for cgroup groups. This improves the performance to
196 * find an existing css_set */
197 #define CSS_SET_HASH_BITS 7
198 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
199 static struct hlist_head css_set_table
[CSS_SET_TABLE_SIZE
];
201 static struct hlist_head
*css_set_hash(struct cgroup_subsys_state
*css
[])
205 unsigned long tmp
= 0UL;
207 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
208 tmp
+= (unsigned long)css
[i
];
209 tmp
= (tmp
>> 16) ^ tmp
;
211 index
= hash_long(tmp
, CSS_SET_HASH_BITS
);
213 return &css_set_table
[index
];
216 /* We don't maintain the lists running through each css_set to its
217 * task until after the first call to cgroup_iter_start(). This
218 * reduces the fork()/exit() overhead for people who have cgroups
219 * compiled into their kernel but not actually in use */
220 static int use_task_css_set_links __read_mostly
;
222 /* When we create or destroy a css_set, the operation simply
223 * takes/releases a reference count on all the cgroups referenced
224 * by subsystems in this css_set. This can end up multiple-counting
225 * some cgroups, but that's OK - the ref-count is just a
226 * busy/not-busy indicator; ensuring that we only count each cgroup
227 * once would require taking a global lock to ensure that no
228 * subsystems moved between hierarchies while we were doing so.
230 * Possible TODO: decide at boot time based on the number of
231 * registered subsystems and the number of CPUs or NUMA nodes whether
232 * it's better for performance to ref-count every subsystem, or to
233 * take a global lock and only add one ref count to each hierarchy.
237 * unlink a css_set from the list and free it
239 static void unlink_css_set(struct css_set
*cg
)
241 struct cg_cgroup_link
*link
;
242 struct cg_cgroup_link
*saved_link
;
244 write_lock(&css_set_lock
);
245 hlist_del(&cg
->hlist
);
248 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
250 list_del(&link
->cg_link_list
);
251 list_del(&link
->cgrp_link_list
);
255 write_unlock(&css_set_lock
);
258 static void __release_css_set(struct kref
*k
, int taskexit
)
261 struct css_set
*cg
= container_of(k
, struct css_set
, ref
);
266 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
267 struct cgroup
*cgrp
= cg
->subsys
[i
]->cgroup
;
268 if (atomic_dec_and_test(&cgrp
->count
) &&
269 notify_on_release(cgrp
)) {
271 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
272 check_for_release(cgrp
);
279 static void release_css_set(struct kref
*k
)
281 __release_css_set(k
, 0);
284 static void release_css_set_taskexit(struct kref
*k
)
286 __release_css_set(k
, 1);
290 * refcounted get/put for css_set objects
292 static inline void get_css_set(struct css_set
*cg
)
297 static inline void put_css_set(struct css_set
*cg
)
299 kref_put(&cg
->ref
, release_css_set
);
302 static inline void put_css_set_taskexit(struct css_set
*cg
)
304 kref_put(&cg
->ref
, release_css_set_taskexit
);
308 * find_existing_css_set() is a helper for
309 * find_css_set(), and checks to see whether an existing
310 * css_set is suitable.
312 * oldcg: the cgroup group that we're using before the cgroup
315 * cgrp: the cgroup that we're moving into
317 * template: location in which to build the desired set of subsystem
318 * state objects for the new cgroup group
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 hlist_head
*hhead
;
328 struct hlist_node
*node
;
331 /* Built the set of subsystem state objects that we want to
332 * see in the new css_set */
333 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
334 if (root
->subsys_bits
& (1UL << i
)) {
335 /* Subsystem is in this hierarchy. So we want
336 * the subsystem state from the new
338 template[i
] = cgrp
->subsys
[i
];
340 /* Subsystem is not in this hierarchy, so we
341 * don't want to change the subsystem state */
342 template[i
] = oldcg
->subsys
[i
];
346 hhead
= css_set_hash(template);
347 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
348 if (!memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
349 /* All subsystems matched */
354 /* No existing cgroup group matched */
359 * allocate_cg_links() allocates "count" cg_cgroup_link structures
360 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
361 * success or a negative error
363 static int allocate_cg_links(int count
, struct list_head
*tmp
)
365 struct cg_cgroup_link
*link
;
366 struct cg_cgroup_link
*saved_link
;
369 for (i
= 0; i
< count
; i
++) {
370 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
372 list_for_each_entry_safe(link
, saved_link
, tmp
,
374 list_del(&link
->cgrp_link_list
);
379 list_add(&link
->cgrp_link_list
, tmp
);
384 static void free_cg_links(struct list_head
*tmp
)
386 struct cg_cgroup_link
*link
;
387 struct cg_cgroup_link
*saved_link
;
389 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
390 list_del(&link
->cgrp_link_list
);
396 * find_css_set() takes an existing cgroup group and a
397 * cgroup object, and returns a css_set object that's
398 * equivalent to the old group, but with the given cgroup
399 * substituted into the appropriate hierarchy. Must be called with
402 static struct css_set
*find_css_set(
403 struct css_set
*oldcg
, struct cgroup
*cgrp
)
406 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
409 struct list_head tmp_cg_links
;
410 struct cg_cgroup_link
*link
;
412 struct hlist_head
*hhead
;
414 /* First see if we already have a cgroup group that matches
416 read_lock(&css_set_lock
);
417 res
= find_existing_css_set(oldcg
, cgrp
, template);
420 read_unlock(&css_set_lock
);
425 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
429 /* Allocate all the cg_cgroup_link objects that we'll need */
430 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
435 kref_init(&res
->ref
);
436 INIT_LIST_HEAD(&res
->cg_links
);
437 INIT_LIST_HEAD(&res
->tasks
);
438 INIT_HLIST_NODE(&res
->hlist
);
440 /* Copy the set of subsystem state objects generated in
441 * find_existing_css_set() */
442 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
444 write_lock(&css_set_lock
);
445 /* Add reference counts and links from the new css_set. */
446 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
447 struct cgroup
*cgrp
= res
->subsys
[i
]->cgroup
;
448 struct cgroup_subsys
*ss
= subsys
[i
];
449 atomic_inc(&cgrp
->count
);
451 * We want to add a link once per cgroup, so we
452 * only do it for the first subsystem in each
455 if (ss
->root
->subsys_list
.next
== &ss
->sibling
) {
456 BUG_ON(list_empty(&tmp_cg_links
));
457 link
= list_entry(tmp_cg_links
.next
,
458 struct cg_cgroup_link
,
460 list_del(&link
->cgrp_link_list
);
461 list_add(&link
->cgrp_link_list
, &cgrp
->css_sets
);
463 list_add(&link
->cg_link_list
, &res
->cg_links
);
466 if (list_empty(&rootnode
.subsys_list
)) {
467 link
= list_entry(tmp_cg_links
.next
,
468 struct cg_cgroup_link
,
470 list_del(&link
->cgrp_link_list
);
471 list_add(&link
->cgrp_link_list
, &dummytop
->css_sets
);
473 list_add(&link
->cg_link_list
, &res
->cg_links
);
476 BUG_ON(!list_empty(&tmp_cg_links
));
480 /* Add this cgroup group to the hash table */
481 hhead
= css_set_hash(res
->subsys
);
482 hlist_add_head(&res
->hlist
, hhead
);
484 write_unlock(&css_set_lock
);
490 * There is one global cgroup mutex. We also require taking
491 * task_lock() when dereferencing a task's cgroup subsys pointers.
492 * See "The task_lock() exception", at the end of this comment.
494 * A task must hold cgroup_mutex to modify cgroups.
496 * Any task can increment and decrement the count field without lock.
497 * So in general, code holding cgroup_mutex can't rely on the count
498 * field not changing. However, if the count goes to zero, then only
499 * cgroup_attach_task() can increment it again. Because a count of zero
500 * means that no tasks are currently attached, therefore there is no
501 * way a task attached to that cgroup can fork (the other way to
502 * increment the count). So code holding cgroup_mutex can safely
503 * assume that if the count is zero, it will stay zero. Similarly, if
504 * a task holds cgroup_mutex on a cgroup with zero count, it
505 * knows that the cgroup won't be removed, as cgroup_rmdir()
508 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
509 * (usually) take cgroup_mutex. These are the two most performance
510 * critical pieces of code here. The exception occurs on cgroup_exit(),
511 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
512 * is taken, and if the cgroup count is zero, a usermode call made
513 * to the release agent with the name of the cgroup (path relative to
514 * the root of cgroup file system) as the argument.
516 * A cgroup can only be deleted if both its 'count' of using tasks
517 * is zero, and its list of 'children' cgroups is empty. Since all
518 * tasks in the system use _some_ cgroup, and since there is always at
519 * least one task in the system (init, pid == 1), therefore, top_cgroup
520 * always has either children cgroups and/or using tasks. So we don't
521 * need a special hack to ensure that top_cgroup cannot be deleted.
523 * The task_lock() exception
525 * The need for this exception arises from the action of
526 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
527 * another. It does so using cgroup_mutex, however there are
528 * several performance critical places that need to reference
529 * task->cgroup without the expense of grabbing a system global
530 * mutex. Therefore except as noted below, when dereferencing or, as
531 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
532 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
533 * the task_struct routinely used for such matters.
535 * P.S. One more locking exception. RCU is used to guard the
536 * update of a tasks cgroup pointer by cgroup_attach_task()
540 * cgroup_lock - lock out any changes to cgroup structures
543 void cgroup_lock(void)
545 mutex_lock(&cgroup_mutex
);
549 * cgroup_unlock - release lock on cgroup changes
551 * Undo the lock taken in a previous cgroup_lock() call.
553 void cgroup_unlock(void)
555 mutex_unlock(&cgroup_mutex
);
559 * A couple of forward declarations required, due to cyclic reference loop:
560 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
561 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
565 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
);
566 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
567 static int cgroup_populate_dir(struct cgroup
*cgrp
);
568 static struct inode_operations cgroup_dir_inode_operations
;
569 static struct file_operations proc_cgroupstats_operations
;
571 static struct backing_dev_info cgroup_backing_dev_info
= {
572 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
575 static struct inode
*cgroup_new_inode(mode_t mode
, struct super_block
*sb
)
577 struct inode
*inode
= new_inode(sb
);
580 inode
->i_mode
= mode
;
581 inode
->i_uid
= current
->fsuid
;
582 inode
->i_gid
= current
->fsgid
;
584 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
585 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
591 * Call subsys's pre_destroy handler.
592 * This is called before css refcnt check.
594 static void cgroup_call_pre_destroy(struct cgroup
*cgrp
)
596 struct cgroup_subsys
*ss
;
597 for_each_subsys(cgrp
->root
, ss
)
598 if (ss
->pre_destroy
&& cgrp
->subsys
[ss
->subsys_id
])
599 ss
->pre_destroy(ss
, cgrp
);
603 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
605 /* is dentry a directory ? if so, kfree() associated cgroup */
606 if (S_ISDIR(inode
->i_mode
)) {
607 struct cgroup
*cgrp
= dentry
->d_fsdata
;
608 struct cgroup_subsys
*ss
;
609 BUG_ON(!(cgroup_is_removed(cgrp
)));
610 /* It's possible for external users to be holding css
611 * reference counts on a cgroup; css_put() needs to
612 * be able to access the cgroup after decrementing
613 * the reference count in order to know if it needs to
614 * queue the cgroup to be handled by the release
618 mutex_lock(&cgroup_mutex
);
620 * Release the subsystem state objects.
622 for_each_subsys(cgrp
->root
, ss
) {
623 if (cgrp
->subsys
[ss
->subsys_id
])
624 ss
->destroy(ss
, cgrp
);
627 cgrp
->root
->number_of_cgroups
--;
628 mutex_unlock(&cgroup_mutex
);
630 /* Drop the active superblock reference that we took when we
631 * created the cgroup */
632 deactivate_super(cgrp
->root
->sb
);
639 static void remove_dir(struct dentry
*d
)
641 struct dentry
*parent
= dget(d
->d_parent
);
644 simple_rmdir(parent
->d_inode
, d
);
648 static void cgroup_clear_directory(struct dentry
*dentry
)
650 struct list_head
*node
;
652 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
653 spin_lock(&dcache_lock
);
654 node
= dentry
->d_subdirs
.next
;
655 while (node
!= &dentry
->d_subdirs
) {
656 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
659 /* This should never be called on a cgroup
660 * directory with child cgroups */
661 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
663 spin_unlock(&dcache_lock
);
665 simple_unlink(dentry
->d_inode
, d
);
667 spin_lock(&dcache_lock
);
669 node
= dentry
->d_subdirs
.next
;
671 spin_unlock(&dcache_lock
);
675 * NOTE : the dentry must have been dget()'ed
677 static void cgroup_d_remove_dir(struct dentry
*dentry
)
679 cgroup_clear_directory(dentry
);
681 spin_lock(&dcache_lock
);
682 list_del_init(&dentry
->d_u
.d_child
);
683 spin_unlock(&dcache_lock
);
687 static int rebind_subsystems(struct cgroupfs_root
*root
,
688 unsigned long final_bits
)
690 unsigned long added_bits
, removed_bits
;
691 struct cgroup
*cgrp
= &root
->top_cgroup
;
694 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
695 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
696 /* Check that any added subsystems are currently free */
697 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
698 unsigned long bit
= 1UL << i
;
699 struct cgroup_subsys
*ss
= subsys
[i
];
700 if (!(bit
& added_bits
))
702 if (ss
->root
!= &rootnode
) {
703 /* Subsystem isn't free */
708 /* Currently we don't handle adding/removing subsystems when
709 * any child cgroups exist. This is theoretically supportable
710 * but involves complex error handling, so it's being left until
712 if (!list_empty(&cgrp
->children
))
715 /* Process each subsystem */
716 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
717 struct cgroup_subsys
*ss
= subsys
[i
];
718 unsigned long bit
= 1UL << i
;
719 if (bit
& added_bits
) {
720 /* We're binding this subsystem to this hierarchy */
721 BUG_ON(cgrp
->subsys
[i
]);
722 BUG_ON(!dummytop
->subsys
[i
]);
723 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
724 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
725 cgrp
->subsys
[i
]->cgroup
= cgrp
;
726 list_add(&ss
->sibling
, &root
->subsys_list
);
727 rcu_assign_pointer(ss
->root
, root
);
731 } else if (bit
& removed_bits
) {
732 /* We're removing this subsystem */
733 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
734 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
736 ss
->bind(ss
, dummytop
);
737 dummytop
->subsys
[i
]->cgroup
= dummytop
;
738 cgrp
->subsys
[i
] = NULL
;
739 rcu_assign_pointer(subsys
[i
]->root
, &rootnode
);
740 list_del(&ss
->sibling
);
741 } else if (bit
& final_bits
) {
742 /* Subsystem state should already exist */
743 BUG_ON(!cgrp
->subsys
[i
]);
745 /* Subsystem state shouldn't exist */
746 BUG_ON(cgrp
->subsys
[i
]);
749 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
755 static int cgroup_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
757 struct cgroupfs_root
*root
= vfs
->mnt_sb
->s_fs_info
;
758 struct cgroup_subsys
*ss
;
760 mutex_lock(&cgroup_mutex
);
761 for_each_subsys(root
, ss
)
762 seq_printf(seq
, ",%s", ss
->name
);
763 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
764 seq_puts(seq
, ",noprefix");
765 if (strlen(root
->release_agent_path
))
766 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
767 mutex_unlock(&cgroup_mutex
);
771 struct cgroup_sb_opts
{
772 unsigned long subsys_bits
;
777 /* Convert a hierarchy specifier into a bitmask of subsystems and
779 static int parse_cgroupfs_options(char *data
,
780 struct cgroup_sb_opts
*opts
)
782 char *token
, *o
= data
?: "all";
784 opts
->subsys_bits
= 0;
786 opts
->release_agent
= NULL
;
788 while ((token
= strsep(&o
, ",")) != NULL
) {
791 if (!strcmp(token
, "all")) {
792 /* Add all non-disabled subsystems */
794 opts
->subsys_bits
= 0;
795 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
796 struct cgroup_subsys
*ss
= subsys
[i
];
798 opts
->subsys_bits
|= 1ul << i
;
800 } else if (!strcmp(token
, "noprefix")) {
801 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
802 } else if (!strncmp(token
, "release_agent=", 14)) {
803 /* Specifying two release agents is forbidden */
804 if (opts
->release_agent
)
806 opts
->release_agent
= kzalloc(PATH_MAX
, GFP_KERNEL
);
807 if (!opts
->release_agent
)
809 strncpy(opts
->release_agent
, token
+ 14, PATH_MAX
- 1);
810 opts
->release_agent
[PATH_MAX
- 1] = 0;
812 struct cgroup_subsys
*ss
;
814 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
816 if (!strcmp(token
, ss
->name
)) {
818 set_bit(i
, &opts
->subsys_bits
);
822 if (i
== CGROUP_SUBSYS_COUNT
)
827 /* We can't have an empty hierarchy */
828 if (!opts
->subsys_bits
)
834 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
837 struct cgroupfs_root
*root
= sb
->s_fs_info
;
838 struct cgroup
*cgrp
= &root
->top_cgroup
;
839 struct cgroup_sb_opts opts
;
841 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
842 mutex_lock(&cgroup_mutex
);
844 /* See what subsystems are wanted */
845 ret
= parse_cgroupfs_options(data
, &opts
);
849 /* Don't allow flags to change at remount */
850 if (opts
.flags
!= root
->flags
) {
855 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
857 /* (re)populate subsystem files */
859 cgroup_populate_dir(cgrp
);
861 if (opts
.release_agent
)
862 strcpy(root
->release_agent_path
, opts
.release_agent
);
864 if (opts
.release_agent
)
865 kfree(opts
.release_agent
);
866 mutex_unlock(&cgroup_mutex
);
867 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
871 static struct super_operations cgroup_ops
= {
872 .statfs
= simple_statfs
,
873 .drop_inode
= generic_delete_inode
,
874 .show_options
= cgroup_show_options
,
875 .remount_fs
= cgroup_remount
,
878 static void init_cgroup_root(struct cgroupfs_root
*root
)
880 struct cgroup
*cgrp
= &root
->top_cgroup
;
881 INIT_LIST_HEAD(&root
->subsys_list
);
882 INIT_LIST_HEAD(&root
->root_list
);
883 root
->number_of_cgroups
= 1;
885 cgrp
->top_cgroup
= cgrp
;
886 INIT_LIST_HEAD(&cgrp
->sibling
);
887 INIT_LIST_HEAD(&cgrp
->children
);
888 INIT_LIST_HEAD(&cgrp
->css_sets
);
889 INIT_LIST_HEAD(&cgrp
->release_list
);
892 static int cgroup_test_super(struct super_block
*sb
, void *data
)
894 struct cgroupfs_root
*new = data
;
895 struct cgroupfs_root
*root
= sb
->s_fs_info
;
897 /* First check subsystems */
898 if (new->subsys_bits
!= root
->subsys_bits
)
901 /* Next check flags */
902 if (new->flags
!= root
->flags
)
908 static int cgroup_set_super(struct super_block
*sb
, void *data
)
911 struct cgroupfs_root
*root
= data
;
913 ret
= set_anon_super(sb
, NULL
);
917 sb
->s_fs_info
= root
;
920 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
921 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
922 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
923 sb
->s_op
= &cgroup_ops
;
928 static int cgroup_get_rootdir(struct super_block
*sb
)
930 struct inode
*inode
=
931 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
932 struct dentry
*dentry
;
937 inode
->i_fop
= &simple_dir_operations
;
938 inode
->i_op
= &cgroup_dir_inode_operations
;
939 /* directories start off with i_nlink == 2 (for "." entry) */
941 dentry
= d_alloc_root(inode
);
950 static int cgroup_get_sb(struct file_system_type
*fs_type
,
951 int flags
, const char *unused_dev_name
,
952 void *data
, struct vfsmount
*mnt
)
954 struct cgroup_sb_opts opts
;
956 struct super_block
*sb
;
957 struct cgroupfs_root
*root
;
958 struct list_head tmp_cg_links
;
959 INIT_LIST_HEAD(&tmp_cg_links
);
961 /* First find the desired set of subsystems */
962 ret
= parse_cgroupfs_options(data
, &opts
);
964 if (opts
.release_agent
)
965 kfree(opts
.release_agent
);
969 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
971 if (opts
.release_agent
)
972 kfree(opts
.release_agent
);
976 init_cgroup_root(root
);
977 root
->subsys_bits
= opts
.subsys_bits
;
978 root
->flags
= opts
.flags
;
979 if (opts
.release_agent
) {
980 strcpy(root
->release_agent_path
, opts
.release_agent
);
981 kfree(opts
.release_agent
);
984 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, root
);
991 if (sb
->s_fs_info
!= root
) {
992 /* Reusing an existing superblock */
993 BUG_ON(sb
->s_root
== NULL
);
998 struct cgroup
*cgrp
= &root
->top_cgroup
;
1002 BUG_ON(sb
->s_root
!= NULL
);
1004 ret
= cgroup_get_rootdir(sb
);
1006 goto drop_new_super
;
1007 inode
= sb
->s_root
->d_inode
;
1009 mutex_lock(&inode
->i_mutex
);
1010 mutex_lock(&cgroup_mutex
);
1013 * We're accessing css_set_count without locking
1014 * css_set_lock here, but that's OK - it can only be
1015 * increased by someone holding cgroup_lock, and
1016 * that's us. The worst that can happen is that we
1017 * have some link structures left over
1019 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1021 mutex_unlock(&cgroup_mutex
);
1022 mutex_unlock(&inode
->i_mutex
);
1023 goto drop_new_super
;
1026 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1027 if (ret
== -EBUSY
) {
1028 mutex_unlock(&cgroup_mutex
);
1029 mutex_unlock(&inode
->i_mutex
);
1030 goto drop_new_super
;
1033 /* EBUSY should be the only error here */
1036 list_add(&root
->root_list
, &roots
);
1039 sb
->s_root
->d_fsdata
= &root
->top_cgroup
;
1040 root
->top_cgroup
.dentry
= sb
->s_root
;
1042 /* Link the top cgroup in this hierarchy into all
1043 * the css_set objects */
1044 write_lock(&css_set_lock
);
1045 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1046 struct hlist_head
*hhead
= &css_set_table
[i
];
1047 struct hlist_node
*node
;
1050 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
1051 struct cg_cgroup_link
*link
;
1053 BUG_ON(list_empty(&tmp_cg_links
));
1054 link
= list_entry(tmp_cg_links
.next
,
1055 struct cg_cgroup_link
,
1057 list_del(&link
->cgrp_link_list
);
1059 list_add(&link
->cgrp_link_list
,
1060 &root
->top_cgroup
.css_sets
);
1061 list_add(&link
->cg_link_list
, &cg
->cg_links
);
1064 write_unlock(&css_set_lock
);
1066 free_cg_links(&tmp_cg_links
);
1068 BUG_ON(!list_empty(&cgrp
->sibling
));
1069 BUG_ON(!list_empty(&cgrp
->children
));
1070 BUG_ON(root
->number_of_cgroups
!= 1);
1072 cgroup_populate_dir(cgrp
);
1073 mutex_unlock(&inode
->i_mutex
);
1074 mutex_unlock(&cgroup_mutex
);
1077 return simple_set_mnt(mnt
, sb
);
1080 up_write(&sb
->s_umount
);
1081 deactivate_super(sb
);
1082 free_cg_links(&tmp_cg_links
);
1086 static void cgroup_kill_sb(struct super_block
*sb
) {
1087 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1088 struct cgroup
*cgrp
= &root
->top_cgroup
;
1090 struct cg_cgroup_link
*link
;
1091 struct cg_cgroup_link
*saved_link
;
1095 BUG_ON(root
->number_of_cgroups
!= 1);
1096 BUG_ON(!list_empty(&cgrp
->children
));
1097 BUG_ON(!list_empty(&cgrp
->sibling
));
1099 mutex_lock(&cgroup_mutex
);
1101 /* Rebind all subsystems back to the default hierarchy */
1102 ret
= rebind_subsystems(root
, 0);
1103 /* Shouldn't be able to fail ... */
1107 * Release all the links from css_sets to this hierarchy's
1110 write_lock(&css_set_lock
);
1112 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1114 list_del(&link
->cg_link_list
);
1115 list_del(&link
->cgrp_link_list
);
1118 write_unlock(&css_set_lock
);
1120 if (!list_empty(&root
->root_list
)) {
1121 list_del(&root
->root_list
);
1124 mutex_unlock(&cgroup_mutex
);
1127 kill_litter_super(sb
);
1130 static struct file_system_type cgroup_fs_type
= {
1132 .get_sb
= cgroup_get_sb
,
1133 .kill_sb
= cgroup_kill_sb
,
1136 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1138 return dentry
->d_fsdata
;
1141 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1143 return dentry
->d_fsdata
;
1147 * cgroup_path - generate the path of a cgroup
1148 * @cgrp: the cgroup in question
1149 * @buf: the buffer to write the path into
1150 * @buflen: the length of the buffer
1152 * Called with cgroup_mutex held. Writes path of cgroup into buf.
1153 * Returns 0 on success, -errno on error.
1155 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1159 if (cgrp
== dummytop
) {
1161 * Inactive subsystems have no dentry for their root
1168 start
= buf
+ buflen
;
1172 int len
= cgrp
->dentry
->d_name
.len
;
1173 if ((start
-= len
) < buf
)
1174 return -ENAMETOOLONG
;
1175 memcpy(start
, cgrp
->dentry
->d_name
.name
, len
);
1176 cgrp
= cgrp
->parent
;
1182 return -ENAMETOOLONG
;
1185 memmove(buf
, start
, buf
+ buflen
- start
);
1190 * Return the first subsystem attached to a cgroup's hierarchy, and
1194 static void get_first_subsys(const struct cgroup
*cgrp
,
1195 struct cgroup_subsys_state
**css
, int *subsys_id
)
1197 const struct cgroupfs_root
*root
= cgrp
->root
;
1198 const struct cgroup_subsys
*test_ss
;
1199 BUG_ON(list_empty(&root
->subsys_list
));
1200 test_ss
= list_entry(root
->subsys_list
.next
,
1201 struct cgroup_subsys
, sibling
);
1203 *css
= cgrp
->subsys
[test_ss
->subsys_id
];
1207 *subsys_id
= test_ss
->subsys_id
;
1211 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1212 * @cgrp: the cgroup the task is attaching to
1213 * @tsk: the task to be attached
1215 * Call holding cgroup_mutex. May take task_lock of
1216 * the task 'tsk' during call.
1218 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1221 struct cgroup_subsys
*ss
;
1222 struct cgroup
*oldcgrp
;
1223 struct css_set
*cg
= tsk
->cgroups
;
1224 struct css_set
*newcg
;
1225 struct cgroupfs_root
*root
= cgrp
->root
;
1228 get_first_subsys(cgrp
, NULL
, &subsys_id
);
1230 /* Nothing to do if the task is already in that cgroup */
1231 oldcgrp
= task_cgroup(tsk
, subsys_id
);
1232 if (cgrp
== oldcgrp
)
1235 for_each_subsys(root
, ss
) {
1236 if (ss
->can_attach
) {
1237 retval
= ss
->can_attach(ss
, cgrp
, tsk
);
1244 * Locate or allocate a new css_set for this task,
1245 * based on its final set of cgroups
1247 newcg
= find_css_set(cg
, cgrp
);
1252 if (tsk
->flags
& PF_EXITING
) {
1257 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1260 /* Update the css_set linked lists if we're using them */
1261 write_lock(&css_set_lock
);
1262 if (!list_empty(&tsk
->cg_list
)) {
1263 list_del(&tsk
->cg_list
);
1264 list_add(&tsk
->cg_list
, &newcg
->tasks
);
1266 write_unlock(&css_set_lock
);
1268 for_each_subsys(root
, ss
) {
1270 ss
->attach(ss
, cgrp
, oldcgrp
, tsk
);
1272 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1279 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1280 * held. May take task_lock of task
1282 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
)
1284 struct task_struct
*tsk
;
1289 tsk
= find_task_by_vpid(pid
);
1290 if (!tsk
|| tsk
->flags
& PF_EXITING
) {
1294 get_task_struct(tsk
);
1297 if ((current
->euid
) && (current
->euid
!= tsk
->uid
)
1298 && (current
->euid
!= tsk
->suid
)) {
1299 put_task_struct(tsk
);
1304 get_task_struct(tsk
);
1307 ret
= cgroup_attach_task(cgrp
, tsk
);
1308 put_task_struct(tsk
);
1312 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
1315 if (!cgroup_lock_live_group(cgrp
))
1317 ret
= attach_task_by_pid(cgrp
, pid
);
1322 /* The various types of files and directories in a cgroup file system */
1323 enum cgroup_filetype
{
1327 FILE_NOTIFY_ON_RELEASE
,
1332 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1333 * @cgrp: the cgroup to be checked for liveness
1335 * On success, returns true; the lock should be later released with
1336 * cgroup_unlock(). On failure returns false with no lock held.
1338 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
1340 mutex_lock(&cgroup_mutex
);
1341 if (cgroup_is_removed(cgrp
)) {
1342 mutex_unlock(&cgroup_mutex
);
1348 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
1351 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
1352 if (!cgroup_lock_live_group(cgrp
))
1354 strcpy(cgrp
->root
->release_agent_path
, buffer
);
1359 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
1360 struct seq_file
*seq
)
1362 if (!cgroup_lock_live_group(cgrp
))
1364 seq_puts(seq
, cgrp
->root
->release_agent_path
);
1365 seq_putc(seq
, '\n');
1370 /* A buffer size big enough for numbers or short strings */
1371 #define CGROUP_LOCAL_BUFFER_SIZE 64
1373 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
1375 const char __user
*userbuf
,
1376 size_t nbytes
, loff_t
*unused_ppos
)
1378 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
1384 if (nbytes
>= sizeof(buffer
))
1386 if (copy_from_user(buffer
, userbuf
, nbytes
))
1389 buffer
[nbytes
] = 0; /* nul-terminate */
1391 if (cft
->write_u64
) {
1392 u64 val
= simple_strtoull(buffer
, &end
, 0);
1395 retval
= cft
->write_u64(cgrp
, cft
, val
);
1397 s64 val
= simple_strtoll(buffer
, &end
, 0);
1400 retval
= cft
->write_s64(cgrp
, cft
, val
);
1407 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
1409 const char __user
*userbuf
,
1410 size_t nbytes
, loff_t
*unused_ppos
)
1412 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
1414 size_t max_bytes
= cft
->max_write_len
;
1415 char *buffer
= local_buffer
;
1418 max_bytes
= sizeof(local_buffer
) - 1;
1419 if (nbytes
>= max_bytes
)
1421 /* Allocate a dynamic buffer if we need one */
1422 if (nbytes
>= sizeof(local_buffer
)) {
1423 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
1427 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
))
1430 buffer
[nbytes
] = 0; /* nul-terminate */
1432 retval
= cft
->write_string(cgrp
, cft
, buffer
);
1435 if (buffer
!= local_buffer
)
1440 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
1441 size_t nbytes
, loff_t
*ppos
)
1443 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1444 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
1446 if (!cft
|| cgroup_is_removed(cgrp
))
1449 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1450 if (cft
->write_u64
|| cft
->write_s64
)
1451 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1452 if (cft
->write_string
)
1453 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1455 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
1456 return ret
? ret
: nbytes
;
1461 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
1463 char __user
*buf
, size_t nbytes
,
1466 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
1467 u64 val
= cft
->read_u64(cgrp
, cft
);
1468 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
1470 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
1473 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
1475 char __user
*buf
, size_t nbytes
,
1478 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
1479 s64 val
= cft
->read_s64(cgrp
, cft
);
1480 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
1482 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
1485 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
1486 size_t nbytes
, loff_t
*ppos
)
1488 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1489 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
1491 if (!cft
|| cgroup_is_removed(cgrp
))
1495 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1497 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1499 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1504 * seqfile ops/methods for returning structured data. Currently just
1505 * supports string->u64 maps, but can be extended in future.
1508 struct cgroup_seqfile_state
{
1510 struct cgroup
*cgroup
;
1513 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
1515 struct seq_file
*sf
= cb
->state
;
1516 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
1519 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
1521 struct cgroup_seqfile_state
*state
= m
->private;
1522 struct cftype
*cft
= state
->cft
;
1523 if (cft
->read_map
) {
1524 struct cgroup_map_cb cb
= {
1525 .fill
= cgroup_map_add
,
1528 return cft
->read_map(state
->cgroup
, cft
, &cb
);
1530 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
1533 static int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
1535 struct seq_file
*seq
= file
->private_data
;
1536 kfree(seq
->private);
1537 return single_release(inode
, file
);
1540 static struct file_operations cgroup_seqfile_operations
= {
1542 .write
= cgroup_file_write
,
1543 .llseek
= seq_lseek
,
1544 .release
= cgroup_seqfile_release
,
1547 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
1552 err
= generic_file_open(inode
, file
);
1556 cft
= __d_cft(file
->f_dentry
);
1559 if (cft
->read_map
|| cft
->read_seq_string
) {
1560 struct cgroup_seqfile_state
*state
=
1561 kzalloc(sizeof(*state
), GFP_USER
);
1565 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
1566 file
->f_op
= &cgroup_seqfile_operations
;
1567 err
= single_open(file
, cgroup_seqfile_show
, state
);
1570 } else if (cft
->open
)
1571 err
= cft
->open(inode
, file
);
1578 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
1580 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1582 return cft
->release(inode
, file
);
1587 * cgroup_rename - Only allow simple rename of directories in place.
1589 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1590 struct inode
*new_dir
, struct dentry
*new_dentry
)
1592 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
1594 if (new_dentry
->d_inode
)
1596 if (old_dir
!= new_dir
)
1598 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
1601 static struct file_operations cgroup_file_operations
= {
1602 .read
= cgroup_file_read
,
1603 .write
= cgroup_file_write
,
1604 .llseek
= generic_file_llseek
,
1605 .open
= cgroup_file_open
,
1606 .release
= cgroup_file_release
,
1609 static struct inode_operations cgroup_dir_inode_operations
= {
1610 .lookup
= simple_lookup
,
1611 .mkdir
= cgroup_mkdir
,
1612 .rmdir
= cgroup_rmdir
,
1613 .rename
= cgroup_rename
,
1616 static int cgroup_create_file(struct dentry
*dentry
, int mode
,
1617 struct super_block
*sb
)
1619 static struct dentry_operations cgroup_dops
= {
1620 .d_iput
= cgroup_diput
,
1623 struct inode
*inode
;
1627 if (dentry
->d_inode
)
1630 inode
= cgroup_new_inode(mode
, sb
);
1634 if (S_ISDIR(mode
)) {
1635 inode
->i_op
= &cgroup_dir_inode_operations
;
1636 inode
->i_fop
= &simple_dir_operations
;
1638 /* start off with i_nlink == 2 (for "." entry) */
1641 /* start with the directory inode held, so that we can
1642 * populate it without racing with another mkdir */
1643 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
1644 } else if (S_ISREG(mode
)) {
1646 inode
->i_fop
= &cgroup_file_operations
;
1648 dentry
->d_op
= &cgroup_dops
;
1649 d_instantiate(dentry
, inode
);
1650 dget(dentry
); /* Extra count - pin the dentry in core */
1655 * cgroup_create_dir - create a directory for an object.
1656 * @cgrp: the cgroup we create the directory for. It must have a valid
1657 * ->parent field. And we are going to fill its ->dentry field.
1658 * @dentry: dentry of the new cgroup
1659 * @mode: mode to set on new directory.
1661 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
1664 struct dentry
*parent
;
1667 parent
= cgrp
->parent
->dentry
;
1668 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
1670 dentry
->d_fsdata
= cgrp
;
1671 inc_nlink(parent
->d_inode
);
1672 cgrp
->dentry
= dentry
;
1680 int cgroup_add_file(struct cgroup
*cgrp
,
1681 struct cgroup_subsys
*subsys
,
1682 const struct cftype
*cft
)
1684 struct dentry
*dir
= cgrp
->dentry
;
1685 struct dentry
*dentry
;
1688 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
1689 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
1690 strcpy(name
, subsys
->name
);
1693 strcat(name
, cft
->name
);
1694 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
1695 dentry
= lookup_one_len(name
, dir
, strlen(name
));
1696 if (!IS_ERR(dentry
)) {
1697 error
= cgroup_create_file(dentry
, 0644 | S_IFREG
,
1700 dentry
->d_fsdata
= (void *)cft
;
1703 error
= PTR_ERR(dentry
);
1707 int cgroup_add_files(struct cgroup
*cgrp
,
1708 struct cgroup_subsys
*subsys
,
1709 const struct cftype cft
[],
1713 for (i
= 0; i
< count
; i
++) {
1714 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
1722 * cgroup_task_count - count the number of tasks in a cgroup.
1723 * @cgrp: the cgroup in question
1725 * Return the number of tasks in the cgroup.
1727 int cgroup_task_count(const struct cgroup
*cgrp
)
1730 struct cg_cgroup_link
*link
;
1732 read_lock(&css_set_lock
);
1733 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
1734 count
+= atomic_read(&link
->cg
->ref
.refcount
);
1736 read_unlock(&css_set_lock
);
1741 * Advance a list_head iterator. The iterator should be positioned at
1742 * the start of a css_set
1744 static void cgroup_advance_iter(struct cgroup
*cgrp
,
1745 struct cgroup_iter
*it
)
1747 struct list_head
*l
= it
->cg_link
;
1748 struct cg_cgroup_link
*link
;
1751 /* Advance to the next non-empty css_set */
1754 if (l
== &cgrp
->css_sets
) {
1758 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
1760 } while (list_empty(&cg
->tasks
));
1762 it
->task
= cg
->tasks
.next
;
1766 * To reduce the fork() overhead for systems that are not actually
1767 * using their cgroups capability, we don't maintain the lists running
1768 * through each css_set to its tasks until we see the list actually
1769 * used - in other words after the first call to cgroup_iter_start().
1771 * The tasklist_lock is not held here, as do_each_thread() and
1772 * while_each_thread() are protected by RCU.
1774 static void cgroup_enable_task_cg_lists(void)
1776 struct task_struct
*p
, *g
;
1777 write_lock(&css_set_lock
);
1778 use_task_css_set_links
= 1;
1779 do_each_thread(g
, p
) {
1782 * We should check if the process is exiting, otherwise
1783 * it will race with cgroup_exit() in that the list
1784 * entry won't be deleted though the process has exited.
1786 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
1787 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
1789 } while_each_thread(g
, p
);
1790 write_unlock(&css_set_lock
);
1793 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
1796 * The first time anyone tries to iterate across a cgroup,
1797 * we need to enable the list linking each css_set to its
1798 * tasks, and fix up all existing tasks.
1800 if (!use_task_css_set_links
)
1801 cgroup_enable_task_cg_lists();
1803 read_lock(&css_set_lock
);
1804 it
->cg_link
= &cgrp
->css_sets
;
1805 cgroup_advance_iter(cgrp
, it
);
1808 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
1809 struct cgroup_iter
*it
)
1811 struct task_struct
*res
;
1812 struct list_head
*l
= it
->task
;
1814 /* If the iterator cg is NULL, we have no tasks */
1817 res
= list_entry(l
, struct task_struct
, cg_list
);
1818 /* Advance iterator to find next entry */
1820 if (l
== &res
->cgroups
->tasks
) {
1821 /* We reached the end of this task list - move on to
1822 * the next cg_cgroup_link */
1823 cgroup_advance_iter(cgrp
, it
);
1830 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
1832 read_unlock(&css_set_lock
);
1835 static inline int started_after_time(struct task_struct
*t1
,
1836 struct timespec
*time
,
1837 struct task_struct
*t2
)
1839 int start_diff
= timespec_compare(&t1
->start_time
, time
);
1840 if (start_diff
> 0) {
1842 } else if (start_diff
< 0) {
1846 * Arbitrarily, if two processes started at the same
1847 * time, we'll say that the lower pointer value
1848 * started first. Note that t2 may have exited by now
1849 * so this may not be a valid pointer any longer, but
1850 * that's fine - it still serves to distinguish
1851 * between two tasks started (effectively) simultaneously.
1858 * This function is a callback from heap_insert() and is used to order
1860 * In this case we order the heap in descending task start time.
1862 static inline int started_after(void *p1
, void *p2
)
1864 struct task_struct
*t1
= p1
;
1865 struct task_struct
*t2
= p2
;
1866 return started_after_time(t1
, &t2
->start_time
, t2
);
1870 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
1871 * @scan: struct cgroup_scanner containing arguments for the scan
1873 * Arguments include pointers to callback functions test_task() and
1875 * Iterate through all the tasks in a cgroup, calling test_task() for each,
1876 * and if it returns true, call process_task() for it also.
1877 * The test_task pointer may be NULL, meaning always true (select all tasks).
1878 * Effectively duplicates cgroup_iter_{start,next,end}()
1879 * but does not lock css_set_lock for the call to process_task().
1880 * The struct cgroup_scanner may be embedded in any structure of the caller's
1882 * It is guaranteed that process_task() will act on every task that
1883 * is a member of the cgroup for the duration of this call. This
1884 * function may or may not call process_task() for tasks that exit
1885 * or move to a different cgroup during the call, or are forked or
1886 * move into the cgroup during the call.
1888 * Note that test_task() may be called with locks held, and may in some
1889 * situations be called multiple times for the same task, so it should
1891 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
1892 * pre-allocated and will be used for heap operations (and its "gt" member will
1893 * be overwritten), else a temporary heap will be used (allocation of which
1894 * may cause this function to fail).
1896 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
1899 struct cgroup_iter it
;
1900 struct task_struct
*p
, *dropped
;
1901 /* Never dereference latest_task, since it's not refcounted */
1902 struct task_struct
*latest_task
= NULL
;
1903 struct ptr_heap tmp_heap
;
1904 struct ptr_heap
*heap
;
1905 struct timespec latest_time
= { 0, 0 };
1908 /* The caller supplied our heap and pre-allocated its memory */
1910 heap
->gt
= &started_after
;
1912 /* We need to allocate our own heap memory */
1914 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
1916 /* cannot allocate the heap */
1922 * Scan tasks in the cgroup, using the scanner's "test_task" callback
1923 * to determine which are of interest, and using the scanner's
1924 * "process_task" callback to process any of them that need an update.
1925 * Since we don't want to hold any locks during the task updates,
1926 * gather tasks to be processed in a heap structure.
1927 * The heap is sorted by descending task start time.
1928 * If the statically-sized heap fills up, we overflow tasks that
1929 * started later, and in future iterations only consider tasks that
1930 * started after the latest task in the previous pass. This
1931 * guarantees forward progress and that we don't miss any tasks.
1934 cgroup_iter_start(scan
->cg
, &it
);
1935 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
1937 * Only affect tasks that qualify per the caller's callback,
1938 * if he provided one
1940 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
1943 * Only process tasks that started after the last task
1946 if (!started_after_time(p
, &latest_time
, latest_task
))
1948 dropped
= heap_insert(heap
, p
);
1949 if (dropped
== NULL
) {
1951 * The new task was inserted; the heap wasn't
1955 } else if (dropped
!= p
) {
1957 * The new task was inserted, and pushed out a
1961 put_task_struct(dropped
);
1964 * Else the new task was newer than anything already in
1965 * the heap and wasn't inserted
1968 cgroup_iter_end(scan
->cg
, &it
);
1971 for (i
= 0; i
< heap
->size
; i
++) {
1972 struct task_struct
*q
= heap
->ptrs
[i
];
1974 latest_time
= q
->start_time
;
1977 /* Process the task per the caller's callback */
1978 scan
->process_task(q
, scan
);
1982 * If we had to process any tasks at all, scan again
1983 * in case some of them were in the middle of forking
1984 * children that didn't get processed.
1985 * Not the most efficient way to do it, but it avoids
1986 * having to take callback_mutex in the fork path
1990 if (heap
== &tmp_heap
)
1991 heap_free(&tmp_heap
);
1996 * Stuff for reading the 'tasks' file.
1998 * Reading this file can return large amounts of data if a cgroup has
1999 * *lots* of attached tasks. So it may need several calls to read(),
2000 * but we cannot guarantee that the information we produce is correct
2001 * unless we produce it entirely atomically.
2003 * Upon tasks file open(), a struct ctr_struct is allocated, that
2004 * will have a pointer to an array (also allocated here). The struct
2005 * ctr_struct * is stored in file->private_data. Its resources will
2006 * be freed by release() when the file is closed. The array is used
2007 * to sprintf the PIDs and then used by read().
2015 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
2016 * 'cgrp'. Return actual number of pids loaded. No need to
2017 * task_lock(p) when reading out p->cgroup, since we're in an RCU
2018 * read section, so the css_set can't go away, and is
2019 * immutable after creation.
2021 static int pid_array_load(pid_t
*pidarray
, int npids
, struct cgroup
*cgrp
)
2024 struct cgroup_iter it
;
2025 struct task_struct
*tsk
;
2026 cgroup_iter_start(cgrp
, &it
);
2027 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
2028 if (unlikely(n
== npids
))
2030 pidarray
[n
++] = task_pid_vnr(tsk
);
2032 cgroup_iter_end(cgrp
, &it
);
2037 * cgroupstats_build - build and fill cgroupstats
2038 * @stats: cgroupstats to fill information into
2039 * @dentry: A dentry entry belonging to the cgroup for which stats have
2042 * Build and fill cgroupstats so that taskstats can export it to user
2045 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
2048 struct cgroup
*cgrp
;
2049 struct cgroup_iter it
;
2050 struct task_struct
*tsk
;
2052 * Validate dentry by checking the superblock operations
2054 if (dentry
->d_sb
->s_op
!= &cgroup_ops
)
2058 cgrp
= dentry
->d_fsdata
;
2061 cgroup_iter_start(cgrp
, &it
);
2062 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
2063 switch (tsk
->state
) {
2065 stats
->nr_running
++;
2067 case TASK_INTERRUPTIBLE
:
2068 stats
->nr_sleeping
++;
2070 case TASK_UNINTERRUPTIBLE
:
2071 stats
->nr_uninterruptible
++;
2074 stats
->nr_stopped
++;
2077 if (delayacct_is_task_waiting_on_io(tsk
))
2078 stats
->nr_io_wait
++;
2082 cgroup_iter_end(cgrp
, &it
);
2089 static int cmppid(const void *a
, const void *b
)
2091 return *(pid_t
*)a
- *(pid_t
*)b
;
2095 * Convert array 'a' of 'npids' pid_t's to a string of newline separated
2096 * decimal pids in 'buf'. Don't write more than 'sz' chars, but return
2097 * count 'cnt' of how many chars would be written if buf were large enough.
2099 static int pid_array_to_buf(char *buf
, int sz
, pid_t
*a
, int npids
)
2104 for (i
= 0; i
< npids
; i
++)
2105 cnt
+= snprintf(buf
+ cnt
, max(sz
- cnt
, 0), "%d\n", a
[i
]);
2110 * Handle an open on 'tasks' file. Prepare a buffer listing the
2111 * process id's of tasks currently attached to the cgroup being opened.
2113 * Does not require any specific cgroup mutexes, and does not take any.
2115 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
2117 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2118 struct ctr_struct
*ctr
;
2123 if (!(file
->f_mode
& FMODE_READ
))
2126 ctr
= kmalloc(sizeof(*ctr
), GFP_KERNEL
);
2131 * If cgroup gets more users after we read count, we won't have
2132 * enough space - tough. This race is indistinguishable to the
2133 * caller from the case that the additional cgroup users didn't
2134 * show up until sometime later on.
2136 npids
= cgroup_task_count(cgrp
);
2138 pidarray
= kmalloc(npids
* sizeof(pid_t
), GFP_KERNEL
);
2142 npids
= pid_array_load(pidarray
, npids
, cgrp
);
2143 sort(pidarray
, npids
, sizeof(pid_t
), cmppid
, NULL
);
2145 /* Call pid_array_to_buf() twice, first just to get bufsz */
2146 ctr
->bufsz
= pid_array_to_buf(&c
, sizeof(c
), pidarray
, npids
) + 1;
2147 ctr
->buf
= kmalloc(ctr
->bufsz
, GFP_KERNEL
);
2150 ctr
->bufsz
= pid_array_to_buf(ctr
->buf
, ctr
->bufsz
, pidarray
, npids
);
2157 file
->private_data
= ctr
;
2168 static ssize_t
cgroup_tasks_read(struct cgroup
*cgrp
,
2170 struct file
*file
, char __user
*buf
,
2171 size_t nbytes
, loff_t
*ppos
)
2173 struct ctr_struct
*ctr
= file
->private_data
;
2175 return simple_read_from_buffer(buf
, nbytes
, ppos
, ctr
->buf
, ctr
->bufsz
);
2178 static int cgroup_tasks_release(struct inode
*unused_inode
,
2181 struct ctr_struct
*ctr
;
2183 if (file
->f_mode
& FMODE_READ
) {
2184 ctr
= file
->private_data
;
2191 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
2194 return notify_on_release(cgrp
);
2197 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
2201 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
2203 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
2205 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
2210 * for the common functions, 'private' gives the type of file
2212 static struct cftype files
[] = {
2215 .open
= cgroup_tasks_open
,
2216 .read
= cgroup_tasks_read
,
2217 .write_u64
= cgroup_tasks_write
,
2218 .release
= cgroup_tasks_release
,
2219 .private = FILE_TASKLIST
,
2223 .name
= "notify_on_release",
2224 .read_u64
= cgroup_read_notify_on_release
,
2225 .write_u64
= cgroup_write_notify_on_release
,
2226 .private = FILE_NOTIFY_ON_RELEASE
,
2230 static struct cftype cft_release_agent
= {
2231 .name
= "release_agent",
2232 .read_seq_string
= cgroup_release_agent_show
,
2233 .write_string
= cgroup_release_agent_write
,
2234 .max_write_len
= PATH_MAX
,
2235 .private = FILE_RELEASE_AGENT
,
2238 static int cgroup_populate_dir(struct cgroup
*cgrp
)
2241 struct cgroup_subsys
*ss
;
2243 /* First clear out any existing files */
2244 cgroup_clear_directory(cgrp
->dentry
);
2246 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
2250 if (cgrp
== cgrp
->top_cgroup
) {
2251 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
2255 for_each_subsys(cgrp
->root
, ss
) {
2256 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
2263 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
2264 struct cgroup_subsys
*ss
,
2265 struct cgroup
*cgrp
)
2268 atomic_set(&css
->refcnt
, 0);
2270 if (cgrp
== dummytop
)
2271 set_bit(CSS_ROOT
, &css
->flags
);
2272 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
2273 cgrp
->subsys
[ss
->subsys_id
] = css
;
2277 * cgroup_create - create a cgroup
2278 * @parent: cgroup that will be parent of the new cgroup
2279 * @dentry: dentry of the new cgroup
2280 * @mode: mode to set on new inode
2282 * Must be called with the mutex on the parent inode held
2284 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
2287 struct cgroup
*cgrp
;
2288 struct cgroupfs_root
*root
= parent
->root
;
2290 struct cgroup_subsys
*ss
;
2291 struct super_block
*sb
= root
->sb
;
2293 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
2297 /* Grab a reference on the superblock so the hierarchy doesn't
2298 * get deleted on unmount if there are child cgroups. This
2299 * can be done outside cgroup_mutex, since the sb can't
2300 * disappear while someone has an open control file on the
2302 atomic_inc(&sb
->s_active
);
2304 mutex_lock(&cgroup_mutex
);
2306 INIT_LIST_HEAD(&cgrp
->sibling
);
2307 INIT_LIST_HEAD(&cgrp
->children
);
2308 INIT_LIST_HEAD(&cgrp
->css_sets
);
2309 INIT_LIST_HEAD(&cgrp
->release_list
);
2311 cgrp
->parent
= parent
;
2312 cgrp
->root
= parent
->root
;
2313 cgrp
->top_cgroup
= parent
->top_cgroup
;
2315 if (notify_on_release(parent
))
2316 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
2318 for_each_subsys(root
, ss
) {
2319 struct cgroup_subsys_state
*css
= ss
->create(ss
, cgrp
);
2324 init_cgroup_css(css
, ss
, cgrp
);
2327 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
2328 root
->number_of_cgroups
++;
2330 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
2334 /* The cgroup directory was pre-locked for us */
2335 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
2337 err
= cgroup_populate_dir(cgrp
);
2338 /* If err < 0, we have a half-filled directory - oh well ;) */
2340 mutex_unlock(&cgroup_mutex
);
2341 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
2347 list_del(&cgrp
->sibling
);
2348 root
->number_of_cgroups
--;
2352 for_each_subsys(root
, ss
) {
2353 if (cgrp
->subsys
[ss
->subsys_id
])
2354 ss
->destroy(ss
, cgrp
);
2357 mutex_unlock(&cgroup_mutex
);
2359 /* Release the reference count that we took on the superblock */
2360 deactivate_super(sb
);
2366 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2368 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
2370 /* the vfs holds inode->i_mutex already */
2371 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
2374 static inline int cgroup_has_css_refs(struct cgroup
*cgrp
)
2376 /* Check the reference count on each subsystem. Since we
2377 * already established that there are no tasks in the
2378 * cgroup, if the css refcount is also 0, then there should
2379 * be no outstanding references, so the subsystem is safe to
2380 * destroy. We scan across all subsystems rather than using
2381 * the per-hierarchy linked list of mounted subsystems since
2382 * we can be called via check_for_release() with no
2383 * synchronization other than RCU, and the subsystem linked
2384 * list isn't RCU-safe */
2386 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2387 struct cgroup_subsys
*ss
= subsys
[i
];
2388 struct cgroup_subsys_state
*css
;
2389 /* Skip subsystems not in this hierarchy */
2390 if (ss
->root
!= cgrp
->root
)
2392 css
= cgrp
->subsys
[ss
->subsys_id
];
2393 /* When called from check_for_release() it's possible
2394 * that by this point the cgroup has been removed
2395 * and the css deleted. But a false-positive doesn't
2396 * matter, since it can only happen if the cgroup
2397 * has been deleted and hence no longer needs the
2398 * release agent to be called anyway. */
2399 if (css
&& atomic_read(&css
->refcnt
))
2405 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
2407 struct cgroup
*cgrp
= dentry
->d_fsdata
;
2409 struct cgroup
*parent
;
2410 struct super_block
*sb
;
2411 struct cgroupfs_root
*root
;
2413 /* the vfs holds both inode->i_mutex already */
2415 mutex_lock(&cgroup_mutex
);
2416 if (atomic_read(&cgrp
->count
) != 0) {
2417 mutex_unlock(&cgroup_mutex
);
2420 if (!list_empty(&cgrp
->children
)) {
2421 mutex_unlock(&cgroup_mutex
);
2425 parent
= cgrp
->parent
;
2430 * Call pre_destroy handlers of subsys. Notify subsystems
2431 * that rmdir() request comes.
2433 cgroup_call_pre_destroy(cgrp
);
2435 if (cgroup_has_css_refs(cgrp
)) {
2436 mutex_unlock(&cgroup_mutex
);
2440 spin_lock(&release_list_lock
);
2441 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
2442 if (!list_empty(&cgrp
->release_list
))
2443 list_del(&cgrp
->release_list
);
2444 spin_unlock(&release_list_lock
);
2445 /* delete my sibling from parent->children */
2446 list_del(&cgrp
->sibling
);
2447 spin_lock(&cgrp
->dentry
->d_lock
);
2448 d
= dget(cgrp
->dentry
);
2449 cgrp
->dentry
= NULL
;
2450 spin_unlock(&d
->d_lock
);
2452 cgroup_d_remove_dir(d
);
2455 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
2456 check_for_release(parent
);
2458 mutex_unlock(&cgroup_mutex
);
2462 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
2464 struct cgroup_subsys_state
*css
;
2466 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
2468 /* Create the top cgroup state for this subsystem */
2469 ss
->root
= &rootnode
;
2470 css
= ss
->create(ss
, dummytop
);
2471 /* We don't handle early failures gracefully */
2472 BUG_ON(IS_ERR(css
));
2473 init_cgroup_css(css
, ss
, dummytop
);
2475 /* Update the init_css_set to contain a subsys
2476 * pointer to this state - since the subsystem is
2477 * newly registered, all tasks and hence the
2478 * init_css_set is in the subsystem's top cgroup. */
2479 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
2481 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
2482 need_mm_owner_callback
|= !!ss
->mm_owner_changed
;
2484 /* At system boot, before all subsystems have been
2485 * registered, no tasks have been forked, so we don't
2486 * need to invoke fork callbacks here. */
2487 BUG_ON(!list_empty(&init_task
.tasks
));
2493 * cgroup_init_early - cgroup initialization at system boot
2495 * Initialize cgroups at system boot, and initialize any
2496 * subsystems that request early init.
2498 int __init
cgroup_init_early(void)
2501 kref_init(&init_css_set
.ref
);
2502 kref_get(&init_css_set
.ref
);
2503 INIT_LIST_HEAD(&init_css_set
.cg_links
);
2504 INIT_LIST_HEAD(&init_css_set
.tasks
);
2505 INIT_HLIST_NODE(&init_css_set
.hlist
);
2507 init_cgroup_root(&rootnode
);
2508 list_add(&rootnode
.root_list
, &roots
);
2510 init_task
.cgroups
= &init_css_set
;
2512 init_css_set_link
.cg
= &init_css_set
;
2513 list_add(&init_css_set_link
.cgrp_link_list
,
2514 &rootnode
.top_cgroup
.css_sets
);
2515 list_add(&init_css_set_link
.cg_link_list
,
2516 &init_css_set
.cg_links
);
2518 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
2519 INIT_HLIST_HEAD(&css_set_table
[i
]);
2521 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2522 struct cgroup_subsys
*ss
= subsys
[i
];
2525 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
2526 BUG_ON(!ss
->create
);
2527 BUG_ON(!ss
->destroy
);
2528 if (ss
->subsys_id
!= i
) {
2529 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
2530 ss
->name
, ss
->subsys_id
);
2535 cgroup_init_subsys(ss
);
2541 * cgroup_init - cgroup initialization
2543 * Register cgroup filesystem and /proc file, and initialize
2544 * any subsystems that didn't request early init.
2546 int __init
cgroup_init(void)
2550 struct hlist_head
*hhead
;
2552 err
= bdi_init(&cgroup_backing_dev_info
);
2556 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2557 struct cgroup_subsys
*ss
= subsys
[i
];
2558 if (!ss
->early_init
)
2559 cgroup_init_subsys(ss
);
2562 /* Add init_css_set to the hash table */
2563 hhead
= css_set_hash(init_css_set
.subsys
);
2564 hlist_add_head(&init_css_set
.hlist
, hhead
);
2566 err
= register_filesystem(&cgroup_fs_type
);
2570 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
2574 bdi_destroy(&cgroup_backing_dev_info
);
2580 * proc_cgroup_show()
2581 * - Print task's cgroup paths into seq_file, one line for each hierarchy
2582 * - Used for /proc/<pid>/cgroup.
2583 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
2584 * doesn't really matter if tsk->cgroup changes after we read it,
2585 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
2586 * anyway. No need to check that tsk->cgroup != NULL, thanks to
2587 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
2588 * cgroup to top_cgroup.
2591 /* TODO: Use a proper seq_file iterator */
2592 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
2595 struct task_struct
*tsk
;
2598 struct cgroupfs_root
*root
;
2601 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
2607 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
2613 mutex_lock(&cgroup_mutex
);
2615 for_each_root(root
) {
2616 struct cgroup_subsys
*ss
;
2617 struct cgroup
*cgrp
;
2621 /* Skip this hierarchy if it has no active subsystems */
2622 if (!root
->actual_subsys_bits
)
2624 seq_printf(m
, "%lu:", root
->subsys_bits
);
2625 for_each_subsys(root
, ss
)
2626 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
2628 get_first_subsys(&root
->top_cgroup
, NULL
, &subsys_id
);
2629 cgrp
= task_cgroup(tsk
, subsys_id
);
2630 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
2638 mutex_unlock(&cgroup_mutex
);
2639 put_task_struct(tsk
);
2646 static int cgroup_open(struct inode
*inode
, struct file
*file
)
2648 struct pid
*pid
= PROC_I(inode
)->pid
;
2649 return single_open(file
, proc_cgroup_show
, pid
);
2652 struct file_operations proc_cgroup_operations
= {
2653 .open
= cgroup_open
,
2655 .llseek
= seq_lseek
,
2656 .release
= single_release
,
2659 /* Display information about each subsystem and each hierarchy */
2660 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
2664 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
2665 mutex_lock(&cgroup_mutex
);
2666 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2667 struct cgroup_subsys
*ss
= subsys
[i
];
2668 seq_printf(m
, "%s\t%lu\t%d\t%d\n",
2669 ss
->name
, ss
->root
->subsys_bits
,
2670 ss
->root
->number_of_cgroups
, !ss
->disabled
);
2672 mutex_unlock(&cgroup_mutex
);
2676 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
2678 return single_open(file
, proc_cgroupstats_show
, NULL
);
2681 static struct file_operations proc_cgroupstats_operations
= {
2682 .open
= cgroupstats_open
,
2684 .llseek
= seq_lseek
,
2685 .release
= single_release
,
2689 * cgroup_fork - attach newly forked task to its parents cgroup.
2690 * @child: pointer to task_struct of forking parent process.
2692 * Description: A task inherits its parent's cgroup at fork().
2694 * A pointer to the shared css_set was automatically copied in
2695 * fork.c by dup_task_struct(). However, we ignore that copy, since
2696 * it was not made under the protection of RCU or cgroup_mutex, so
2697 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
2698 * have already changed current->cgroups, allowing the previously
2699 * referenced cgroup group to be removed and freed.
2701 * At the point that cgroup_fork() is called, 'current' is the parent
2702 * task, and the passed argument 'child' points to the child task.
2704 void cgroup_fork(struct task_struct
*child
)
2707 child
->cgroups
= current
->cgroups
;
2708 get_css_set(child
->cgroups
);
2709 task_unlock(current
);
2710 INIT_LIST_HEAD(&child
->cg_list
);
2714 * cgroup_fork_callbacks - run fork callbacks
2715 * @child: the new task
2717 * Called on a new task very soon before adding it to the
2718 * tasklist. No need to take any locks since no-one can
2719 * be operating on this task.
2721 void cgroup_fork_callbacks(struct task_struct
*child
)
2723 if (need_forkexit_callback
) {
2725 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2726 struct cgroup_subsys
*ss
= subsys
[i
];
2728 ss
->fork(ss
, child
);
2733 #ifdef CONFIG_MM_OWNER
2735 * cgroup_mm_owner_callbacks - run callbacks when the mm->owner changes
2738 * Called on every change to mm->owner. mm_init_owner() does not
2739 * invoke this routine, since it assigns the mm->owner the first time
2740 * and does not change it.
2742 void cgroup_mm_owner_callbacks(struct task_struct
*old
, struct task_struct
*new)
2744 struct cgroup
*oldcgrp
, *newcgrp
;
2746 if (need_mm_owner_callback
) {
2748 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2749 struct cgroup_subsys
*ss
= subsys
[i
];
2750 oldcgrp
= task_cgroup(old
, ss
->subsys_id
);
2751 newcgrp
= task_cgroup(new, ss
->subsys_id
);
2752 if (oldcgrp
== newcgrp
)
2754 if (ss
->mm_owner_changed
)
2755 ss
->mm_owner_changed(ss
, oldcgrp
, newcgrp
);
2759 #endif /* CONFIG_MM_OWNER */
2762 * cgroup_post_fork - called on a new task after adding it to the task list
2763 * @child: the task in question
2765 * Adds the task to the list running through its css_set if necessary.
2766 * Has to be after the task is visible on the task list in case we race
2767 * with the first call to cgroup_iter_start() - to guarantee that the
2768 * new task ends up on its list.
2770 void cgroup_post_fork(struct task_struct
*child
)
2772 if (use_task_css_set_links
) {
2773 write_lock(&css_set_lock
);
2774 if (list_empty(&child
->cg_list
))
2775 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
2776 write_unlock(&css_set_lock
);
2780 * cgroup_exit - detach cgroup from exiting task
2781 * @tsk: pointer to task_struct of exiting process
2782 * @run_callback: run exit callbacks?
2784 * Description: Detach cgroup from @tsk and release it.
2786 * Note that cgroups marked notify_on_release force every task in
2787 * them to take the global cgroup_mutex mutex when exiting.
2788 * This could impact scaling on very large systems. Be reluctant to
2789 * use notify_on_release cgroups where very high task exit scaling
2790 * is required on large systems.
2792 * the_top_cgroup_hack:
2794 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
2796 * We call cgroup_exit() while the task is still competent to
2797 * handle notify_on_release(), then leave the task attached to the
2798 * root cgroup in each hierarchy for the remainder of its exit.
2800 * To do this properly, we would increment the reference count on
2801 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
2802 * code we would add a second cgroup function call, to drop that
2803 * reference. This would just create an unnecessary hot spot on
2804 * the top_cgroup reference count, to no avail.
2806 * Normally, holding a reference to a cgroup without bumping its
2807 * count is unsafe. The cgroup could go away, or someone could
2808 * attach us to a different cgroup, decrementing the count on
2809 * the first cgroup that we never incremented. But in this case,
2810 * top_cgroup isn't going away, and either task has PF_EXITING set,
2811 * which wards off any cgroup_attach_task() attempts, or task is a failed
2812 * fork, never visible to cgroup_attach_task.
2814 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
2819 if (run_callbacks
&& need_forkexit_callback
) {
2820 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2821 struct cgroup_subsys
*ss
= subsys
[i
];
2828 * Unlink from the css_set task list if necessary.
2829 * Optimistically check cg_list before taking
2832 if (!list_empty(&tsk
->cg_list
)) {
2833 write_lock(&css_set_lock
);
2834 if (!list_empty(&tsk
->cg_list
))
2835 list_del(&tsk
->cg_list
);
2836 write_unlock(&css_set_lock
);
2839 /* Reassign the task to the init_css_set. */
2842 tsk
->cgroups
= &init_css_set
;
2845 put_css_set_taskexit(cg
);
2849 * cgroup_clone - clone the cgroup the given subsystem is attached to
2850 * @tsk: the task to be moved
2851 * @subsys: the given subsystem
2852 * @nodename: the name for the new cgroup
2854 * Duplicate the current cgroup in the hierarchy that the given
2855 * subsystem is attached to, and move this task into the new
2858 int cgroup_clone(struct task_struct
*tsk
, struct cgroup_subsys
*subsys
,
2861 struct dentry
*dentry
;
2863 struct cgroup
*parent
, *child
;
2864 struct inode
*inode
;
2866 struct cgroupfs_root
*root
;
2867 struct cgroup_subsys
*ss
;
2869 /* We shouldn't be called by an unregistered subsystem */
2870 BUG_ON(!subsys
->active
);
2872 /* First figure out what hierarchy and cgroup we're dealing
2873 * with, and pin them so we can drop cgroup_mutex */
2874 mutex_lock(&cgroup_mutex
);
2876 root
= subsys
->root
;
2877 if (root
== &rootnode
) {
2879 "Not cloning cgroup for unused subsystem %s\n",
2881 mutex_unlock(&cgroup_mutex
);
2885 parent
= task_cgroup(tsk
, subsys
->subsys_id
);
2887 /* Pin the hierarchy */
2888 atomic_inc(&parent
->root
->sb
->s_active
);
2890 /* Keep the cgroup alive */
2892 mutex_unlock(&cgroup_mutex
);
2894 /* Now do the VFS work to create a cgroup */
2895 inode
= parent
->dentry
->d_inode
;
2897 /* Hold the parent directory mutex across this operation to
2898 * stop anyone else deleting the new cgroup */
2899 mutex_lock(&inode
->i_mutex
);
2900 dentry
= lookup_one_len(nodename
, parent
->dentry
, strlen(nodename
));
2901 if (IS_ERR(dentry
)) {
2903 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename
,
2905 ret
= PTR_ERR(dentry
);
2909 /* Create the cgroup directory, which also creates the cgroup */
2910 ret
= vfs_mkdir(inode
, dentry
, S_IFDIR
| 0755);
2911 child
= __d_cgrp(dentry
);
2915 "Failed to create cgroup %s: %d\n", nodename
,
2922 "Couldn't find new cgroup %s\n", nodename
);
2927 /* The cgroup now exists. Retake cgroup_mutex and check
2928 * that we're still in the same state that we thought we
2930 mutex_lock(&cgroup_mutex
);
2931 if ((root
!= subsys
->root
) ||
2932 (parent
!= task_cgroup(tsk
, subsys
->subsys_id
))) {
2933 /* Aargh, we raced ... */
2934 mutex_unlock(&inode
->i_mutex
);
2937 deactivate_super(parent
->root
->sb
);
2938 /* The cgroup is still accessible in the VFS, but
2939 * we're not going to try to rmdir() it at this
2942 "Race in cgroup_clone() - leaking cgroup %s\n",
2947 /* do any required auto-setup */
2948 for_each_subsys(root
, ss
) {
2950 ss
->post_clone(ss
, child
);
2953 /* All seems fine. Finish by moving the task into the new cgroup */
2954 ret
= cgroup_attach_task(child
, tsk
);
2955 mutex_unlock(&cgroup_mutex
);
2958 mutex_unlock(&inode
->i_mutex
);
2960 mutex_lock(&cgroup_mutex
);
2962 mutex_unlock(&cgroup_mutex
);
2963 deactivate_super(parent
->root
->sb
);
2968 * cgroup_is_descendant - see if @cgrp is a descendant of current task's cgrp
2969 * @cgrp: the cgroup in question
2971 * See if @cgrp is a descendant of the current task's cgroup in
2972 * the appropriate hierarchy.
2974 * If we are sending in dummytop, then presumably we are creating
2975 * the top cgroup in the subsystem.
2977 * Called only by the ns (nsproxy) cgroup.
2979 int cgroup_is_descendant(const struct cgroup
*cgrp
)
2982 struct cgroup
*target
;
2985 if (cgrp
== dummytop
)
2988 get_first_subsys(cgrp
, NULL
, &subsys_id
);
2989 target
= task_cgroup(current
, subsys_id
);
2990 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
2991 cgrp
= cgrp
->parent
;
2992 ret
= (cgrp
== target
);
2996 static void check_for_release(struct cgroup
*cgrp
)
2998 /* All of these checks rely on RCU to keep the cgroup
2999 * structure alive */
3000 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
3001 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
3002 /* Control Group is currently removeable. If it's not
3003 * already queued for a userspace notification, queue
3005 int need_schedule_work
= 0;
3006 spin_lock(&release_list_lock
);
3007 if (!cgroup_is_removed(cgrp
) &&
3008 list_empty(&cgrp
->release_list
)) {
3009 list_add(&cgrp
->release_list
, &release_list
);
3010 need_schedule_work
= 1;
3012 spin_unlock(&release_list_lock
);
3013 if (need_schedule_work
)
3014 schedule_work(&release_agent_work
);
3018 void __css_put(struct cgroup_subsys_state
*css
)
3020 struct cgroup
*cgrp
= css
->cgroup
;
3022 if (atomic_dec_and_test(&css
->refcnt
) && notify_on_release(cgrp
)) {
3023 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3024 check_for_release(cgrp
);
3030 * Notify userspace when a cgroup is released, by running the
3031 * configured release agent with the name of the cgroup (path
3032 * relative to the root of cgroup file system) as the argument.
3034 * Most likely, this user command will try to rmdir this cgroup.
3036 * This races with the possibility that some other task will be
3037 * attached to this cgroup before it is removed, or that some other
3038 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
3039 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
3040 * unused, and this cgroup will be reprieved from its death sentence,
3041 * to continue to serve a useful existence. Next time it's released,
3042 * we will get notified again, if it still has 'notify_on_release' set.
3044 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
3045 * means only wait until the task is successfully execve()'d. The
3046 * separate release agent task is forked by call_usermodehelper(),
3047 * then control in this thread returns here, without waiting for the
3048 * release agent task. We don't bother to wait because the caller of
3049 * this routine has no use for the exit status of the release agent
3050 * task, so no sense holding our caller up for that.
3052 static void cgroup_release_agent(struct work_struct
*work
)
3054 BUG_ON(work
!= &release_agent_work
);
3055 mutex_lock(&cgroup_mutex
);
3056 spin_lock(&release_list_lock
);
3057 while (!list_empty(&release_list
)) {
3058 char *argv
[3], *envp
[3];
3060 char *pathbuf
= NULL
, *agentbuf
= NULL
;
3061 struct cgroup
*cgrp
= list_entry(release_list
.next
,
3064 list_del_init(&cgrp
->release_list
);
3065 spin_unlock(&release_list_lock
);
3066 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3069 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
3071 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
3076 argv
[i
++] = agentbuf
;
3077 argv
[i
++] = pathbuf
;
3081 /* minimal command environment */
3082 envp
[i
++] = "HOME=/";
3083 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
3086 /* Drop the lock while we invoke the usermode helper,
3087 * since the exec could involve hitting disk and hence
3088 * be a slow process */
3089 mutex_unlock(&cgroup_mutex
);
3090 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
3091 mutex_lock(&cgroup_mutex
);
3095 spin_lock(&release_list_lock
);
3097 spin_unlock(&release_list_lock
);
3098 mutex_unlock(&cgroup_mutex
);
3101 static int __init
cgroup_disable(char *str
)
3106 while ((token
= strsep(&str
, ",")) != NULL
) {
3110 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3111 struct cgroup_subsys
*ss
= subsys
[i
];
3113 if (!strcmp(token
, ss
->name
)) {
3115 printk(KERN_INFO
"Disabling %s control group"
3116 " subsystem\n", ss
->name
);
3123 __setup("cgroup_disable=", cgroup_disable
);