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>
49 #include <asm/atomic.h>
51 static DEFINE_MUTEX(cgroup_mutex
);
53 /* Generate an array of cgroup subsystem pointers */
54 #define SUBSYS(_x) &_x ## _subsys,
56 static struct cgroup_subsys
*subsys
[] = {
57 #include <linux/cgroup_subsys.h>
61 * A cgroupfs_root represents the root of a cgroup hierarchy,
62 * and may be associated with a superblock to form an active
65 struct cgroupfs_root
{
66 struct super_block
*sb
;
69 * The bitmask of subsystems intended to be attached to this
72 unsigned long subsys_bits
;
74 /* The bitmask of subsystems currently attached to this hierarchy */
75 unsigned long actual_subsys_bits
;
77 /* A list running through the attached subsystems */
78 struct list_head subsys_list
;
80 /* The root cgroup for this hierarchy */
81 struct cgroup top_cgroup
;
83 /* Tracks how many cgroups are currently defined in hierarchy.*/
84 int number_of_cgroups
;
86 /* A list running through the mounted hierarchies */
87 struct list_head root_list
;
89 /* Hierarchy-specific flags */
92 /* The path to use for release notifications. */
93 char release_agent_path
[PATH_MAX
];
98 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
99 * subsystems that are otherwise unattached - it never has more than a
100 * single cgroup, and all tasks are part of that cgroup.
102 static struct cgroupfs_root rootnode
;
104 /* The list of hierarchy roots */
106 static LIST_HEAD(roots
);
107 static int root_count
;
109 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
110 #define dummytop (&rootnode.top_cgroup)
112 /* This flag indicates whether tasks in the fork and exit paths should
113 * check for fork/exit handlers to call. This avoids us having to do
114 * extra work in the fork/exit path if none of the subsystems need to
117 static int need_forkexit_callback __read_mostly
;
118 static int need_mm_owner_callback __read_mostly
;
120 /* convenient tests for these bits */
121 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
123 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
126 /* bits in struct cgroupfs_root flags field */
128 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
131 static int cgroup_is_releasable(const struct cgroup
*cgrp
)
134 (1 << CGRP_RELEASABLE
) |
135 (1 << CGRP_NOTIFY_ON_RELEASE
);
136 return (cgrp
->flags
& bits
) == bits
;
139 static int notify_on_release(const struct cgroup
*cgrp
)
141 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
145 * for_each_subsys() allows you to iterate on each subsystem attached to
146 * an active hierarchy
148 #define for_each_subsys(_root, _ss) \
149 list_for_each_entry(_ss, &_root->subsys_list, sibling)
151 /* for_each_root() allows you to iterate across the active hierarchies */
152 #define for_each_root(_root) \
153 list_for_each_entry(_root, &roots, root_list)
155 /* the list of cgroups eligible for automatic release. Protected by
156 * release_list_lock */
157 static LIST_HEAD(release_list
);
158 static DEFINE_SPINLOCK(release_list_lock
);
159 static void cgroup_release_agent(struct work_struct
*work
);
160 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
161 static void check_for_release(struct cgroup
*cgrp
);
163 /* Link structure for associating css_set objects with cgroups */
164 struct cg_cgroup_link
{
166 * List running through cg_cgroup_links associated with a
167 * cgroup, anchored on cgroup->css_sets
169 struct list_head cgrp_link_list
;
171 * List running through cg_cgroup_links pointing at a
172 * single css_set object, anchored on css_set->cg_links
174 struct list_head cg_link_list
;
178 /* The default css_set - used by init and its children prior to any
179 * hierarchies being mounted. It contains a pointer to the root state
180 * for each subsystem. Also used to anchor the list of css_sets. Not
181 * reference-counted, to improve performance when child cgroups
182 * haven't been created.
185 static struct css_set init_css_set
;
186 static struct cg_cgroup_link init_css_set_link
;
188 /* css_set_lock protects the list of css_set objects, and the
189 * chain of tasks off each css_set. Nests outside task->alloc_lock
190 * due to cgroup_iter_start() */
191 static DEFINE_RWLOCK(css_set_lock
);
192 static int css_set_count
;
194 /* hash table for cgroup groups. This improves the performance to
195 * find an existing css_set */
196 #define CSS_SET_HASH_BITS 7
197 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
198 static struct hlist_head css_set_table
[CSS_SET_TABLE_SIZE
];
200 static struct hlist_head
*css_set_hash(struct cgroup_subsys_state
*css
[])
204 unsigned long tmp
= 0UL;
206 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++)
207 tmp
+= (unsigned long)css
[i
];
208 tmp
= (tmp
>> 16) ^ tmp
;
210 index
= hash_long(tmp
, CSS_SET_HASH_BITS
);
212 return &css_set_table
[index
];
215 /* We don't maintain the lists running through each css_set to its
216 * task until after the first call to cgroup_iter_start(). This
217 * reduces the fork()/exit() overhead for people who have cgroups
218 * compiled into their kernel but not actually in use */
219 static int use_task_css_set_links __read_mostly
;
221 /* When we create or destroy a css_set, the operation simply
222 * takes/releases a reference count on all the cgroups referenced
223 * by subsystems in this css_set. This can end up multiple-counting
224 * some cgroups, but that's OK - the ref-count is just a
225 * busy/not-busy indicator; ensuring that we only count each cgroup
226 * once would require taking a global lock to ensure that no
227 * subsystems moved between hierarchies while we were doing so.
229 * Possible TODO: decide at boot time based on the number of
230 * registered subsystems and the number of CPUs or NUMA nodes whether
231 * it's better for performance to ref-count every subsystem, or to
232 * take a global lock and only add one ref count to each hierarchy.
236 * unlink a css_set from the list and free it
238 static void unlink_css_set(struct css_set
*cg
)
240 struct cg_cgroup_link
*link
;
241 struct cg_cgroup_link
*saved_link
;
243 write_lock(&css_set_lock
);
244 hlist_del(&cg
->hlist
);
247 list_for_each_entry_safe(link
, saved_link
, &cg
->cg_links
,
249 list_del(&link
->cg_link_list
);
250 list_del(&link
->cgrp_link_list
);
254 write_unlock(&css_set_lock
);
257 static void __release_css_set(struct kref
*k
, int taskexit
)
260 struct css_set
*cg
= container_of(k
, struct css_set
, ref
);
265 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
266 struct cgroup
*cgrp
= cg
->subsys
[i
]->cgroup
;
267 if (atomic_dec_and_test(&cgrp
->count
) &&
268 notify_on_release(cgrp
)) {
270 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
271 check_for_release(cgrp
);
278 static void release_css_set(struct kref
*k
)
280 __release_css_set(k
, 0);
283 static void release_css_set_taskexit(struct kref
*k
)
285 __release_css_set(k
, 1);
289 * refcounted get/put for css_set objects
291 static inline void get_css_set(struct css_set
*cg
)
296 static inline void put_css_set(struct css_set
*cg
)
298 kref_put(&cg
->ref
, release_css_set
);
301 static inline void put_css_set_taskexit(struct css_set
*cg
)
303 kref_put(&cg
->ref
, release_css_set_taskexit
);
307 * find_existing_css_set() is a helper for
308 * find_css_set(), and checks to see whether an existing
309 * css_set is suitable.
311 * oldcg: the cgroup group that we're using before the cgroup
314 * cgrp: the cgroup that we're moving into
316 * template: location in which to build the desired set of subsystem
317 * state objects for the new cgroup group
319 static struct css_set
*find_existing_css_set(
320 struct css_set
*oldcg
,
322 struct cgroup_subsys_state
*template[])
325 struct cgroupfs_root
*root
= cgrp
->root
;
326 struct hlist_head
*hhead
;
327 struct hlist_node
*node
;
330 /* Built the set of subsystem state objects that we want to
331 * see in the new css_set */
332 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
333 if (root
->subsys_bits
& (1UL << i
)) {
334 /* Subsystem is in this hierarchy. So we want
335 * the subsystem state from the new
337 template[i
] = cgrp
->subsys
[i
];
339 /* Subsystem is not in this hierarchy, so we
340 * don't want to change the subsystem state */
341 template[i
] = oldcg
->subsys
[i
];
345 hhead
= css_set_hash(template);
346 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
347 if (!memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
348 /* All subsystems matched */
353 /* No existing cgroup group matched */
358 * allocate_cg_links() allocates "count" cg_cgroup_link structures
359 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
360 * success or a negative error
362 static int allocate_cg_links(int count
, struct list_head
*tmp
)
364 struct cg_cgroup_link
*link
;
365 struct cg_cgroup_link
*saved_link
;
368 for (i
= 0; i
< count
; i
++) {
369 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
371 list_for_each_entry_safe(link
, saved_link
, tmp
,
373 list_del(&link
->cgrp_link_list
);
378 list_add(&link
->cgrp_link_list
, tmp
);
383 static void free_cg_links(struct list_head
*tmp
)
385 struct cg_cgroup_link
*link
;
386 struct cg_cgroup_link
*saved_link
;
388 list_for_each_entry_safe(link
, saved_link
, tmp
, cgrp_link_list
) {
389 list_del(&link
->cgrp_link_list
);
395 * find_css_set() takes an existing cgroup group and a
396 * cgroup object, and returns a css_set object that's
397 * equivalent to the old group, but with the given cgroup
398 * substituted into the appropriate hierarchy. Must be called with
401 static struct css_set
*find_css_set(
402 struct css_set
*oldcg
, struct cgroup
*cgrp
)
405 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
408 struct list_head tmp_cg_links
;
409 struct cg_cgroup_link
*link
;
411 struct hlist_head
*hhead
;
413 /* First see if we already have a cgroup group that matches
415 read_lock(&css_set_lock
);
416 res
= find_existing_css_set(oldcg
, cgrp
, template);
419 read_unlock(&css_set_lock
);
424 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
428 /* Allocate all the cg_cgroup_link objects that we'll need */
429 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
434 kref_init(&res
->ref
);
435 INIT_LIST_HEAD(&res
->cg_links
);
436 INIT_LIST_HEAD(&res
->tasks
);
437 INIT_HLIST_NODE(&res
->hlist
);
439 /* Copy the set of subsystem state objects generated in
440 * find_existing_css_set() */
441 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
443 write_lock(&css_set_lock
);
444 /* Add reference counts and links from the new css_set. */
445 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
446 struct cgroup
*cgrp
= res
->subsys
[i
]->cgroup
;
447 struct cgroup_subsys
*ss
= subsys
[i
];
448 atomic_inc(&cgrp
->count
);
450 * We want to add a link once per cgroup, so we
451 * only do it for the first subsystem in each
454 if (ss
->root
->subsys_list
.next
== &ss
->sibling
) {
455 BUG_ON(list_empty(&tmp_cg_links
));
456 link
= list_entry(tmp_cg_links
.next
,
457 struct cg_cgroup_link
,
459 list_del(&link
->cgrp_link_list
);
460 list_add(&link
->cgrp_link_list
, &cgrp
->css_sets
);
462 list_add(&link
->cg_link_list
, &res
->cg_links
);
465 if (list_empty(&rootnode
.subsys_list
)) {
466 link
= list_entry(tmp_cg_links
.next
,
467 struct cg_cgroup_link
,
469 list_del(&link
->cgrp_link_list
);
470 list_add(&link
->cgrp_link_list
, &dummytop
->css_sets
);
472 list_add(&link
->cg_link_list
, &res
->cg_links
);
475 BUG_ON(!list_empty(&tmp_cg_links
));
479 /* Add this cgroup group to the hash table */
480 hhead
= css_set_hash(res
->subsys
);
481 hlist_add_head(&res
->hlist
, hhead
);
483 write_unlock(&css_set_lock
);
489 * There is one global cgroup mutex. We also require taking
490 * task_lock() when dereferencing a task's cgroup subsys pointers.
491 * See "The task_lock() exception", at the end of this comment.
493 * A task must hold cgroup_mutex to modify cgroups.
495 * Any task can increment and decrement the count field without lock.
496 * So in general, code holding cgroup_mutex can't rely on the count
497 * field not changing. However, if the count goes to zero, then only
498 * cgroup_attach_task() can increment it again. Because a count of zero
499 * means that no tasks are currently attached, therefore there is no
500 * way a task attached to that cgroup can fork (the other way to
501 * increment the count). So code holding cgroup_mutex can safely
502 * assume that if the count is zero, it will stay zero. Similarly, if
503 * a task holds cgroup_mutex on a cgroup with zero count, it
504 * knows that the cgroup won't be removed, as cgroup_rmdir()
507 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
508 * (usually) take cgroup_mutex. These are the two most performance
509 * critical pieces of code here. The exception occurs on cgroup_exit(),
510 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
511 * is taken, and if the cgroup count is zero, a usermode call made
512 * to the release agent with the name of the cgroup (path relative to
513 * the root of cgroup file system) as the argument.
515 * A cgroup can only be deleted if both its 'count' of using tasks
516 * is zero, and its list of 'children' cgroups is empty. Since all
517 * tasks in the system use _some_ cgroup, and since there is always at
518 * least one task in the system (init, pid == 1), therefore, top_cgroup
519 * always has either children cgroups and/or using tasks. So we don't
520 * need a special hack to ensure that top_cgroup cannot be deleted.
522 * The task_lock() exception
524 * The need for this exception arises from the action of
525 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
526 * another. It does so using cgroup_mutex, however there are
527 * several performance critical places that need to reference
528 * task->cgroup without the expense of grabbing a system global
529 * mutex. Therefore except as noted below, when dereferencing or, as
530 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
531 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
532 * the task_struct routinely used for such matters.
534 * P.S. One more locking exception. RCU is used to guard the
535 * update of a tasks cgroup pointer by cgroup_attach_task()
539 * cgroup_lock - lock out any changes to cgroup structures
542 void cgroup_lock(void)
544 mutex_lock(&cgroup_mutex
);
548 * cgroup_unlock - release lock on cgroup changes
550 * Undo the lock taken in a previous cgroup_lock() call.
552 void cgroup_unlock(void)
554 mutex_unlock(&cgroup_mutex
);
558 * A couple of forward declarations required, due to cyclic reference loop:
559 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
560 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
564 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
);
565 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
566 static int cgroup_populate_dir(struct cgroup
*cgrp
);
567 static struct inode_operations cgroup_dir_inode_operations
;
568 static struct file_operations proc_cgroupstats_operations
;
570 static struct backing_dev_info cgroup_backing_dev_info
= {
571 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
574 static struct inode
*cgroup_new_inode(mode_t mode
, struct super_block
*sb
)
576 struct inode
*inode
= new_inode(sb
);
579 inode
->i_mode
= mode
;
580 inode
->i_uid
= current
->fsuid
;
581 inode
->i_gid
= current
->fsgid
;
583 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
584 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
590 * Call subsys's pre_destroy handler.
591 * This is called before css refcnt check.
593 static void cgroup_call_pre_destroy(struct cgroup
*cgrp
)
595 struct cgroup_subsys
*ss
;
596 for_each_subsys(cgrp
->root
, ss
)
597 if (ss
->pre_destroy
&& cgrp
->subsys
[ss
->subsys_id
])
598 ss
->pre_destroy(ss
, cgrp
);
602 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
604 /* is dentry a directory ? if so, kfree() associated cgroup */
605 if (S_ISDIR(inode
->i_mode
)) {
606 struct cgroup
*cgrp
= dentry
->d_fsdata
;
607 struct cgroup_subsys
*ss
;
608 BUG_ON(!(cgroup_is_removed(cgrp
)));
609 /* It's possible for external users to be holding css
610 * reference counts on a cgroup; css_put() needs to
611 * be able to access the cgroup after decrementing
612 * the reference count in order to know if it needs to
613 * queue the cgroup to be handled by the release
617 mutex_lock(&cgroup_mutex
);
619 * Release the subsystem state objects.
621 for_each_subsys(cgrp
->root
, ss
) {
622 if (cgrp
->subsys
[ss
->subsys_id
])
623 ss
->destroy(ss
, cgrp
);
626 cgrp
->root
->number_of_cgroups
--;
627 mutex_unlock(&cgroup_mutex
);
629 /* Drop the active superblock reference that we took when we
630 * created the cgroup */
631 deactivate_super(cgrp
->root
->sb
);
638 static void remove_dir(struct dentry
*d
)
640 struct dentry
*parent
= dget(d
->d_parent
);
643 simple_rmdir(parent
->d_inode
, d
);
647 static void cgroup_clear_directory(struct dentry
*dentry
)
649 struct list_head
*node
;
651 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
652 spin_lock(&dcache_lock
);
653 node
= dentry
->d_subdirs
.next
;
654 while (node
!= &dentry
->d_subdirs
) {
655 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
658 /* This should never be called on a cgroup
659 * directory with child cgroups */
660 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
662 spin_unlock(&dcache_lock
);
664 simple_unlink(dentry
->d_inode
, d
);
666 spin_lock(&dcache_lock
);
668 node
= dentry
->d_subdirs
.next
;
670 spin_unlock(&dcache_lock
);
674 * NOTE : the dentry must have been dget()'ed
676 static void cgroup_d_remove_dir(struct dentry
*dentry
)
678 cgroup_clear_directory(dentry
);
680 spin_lock(&dcache_lock
);
681 list_del_init(&dentry
->d_u
.d_child
);
682 spin_unlock(&dcache_lock
);
686 static int rebind_subsystems(struct cgroupfs_root
*root
,
687 unsigned long final_bits
)
689 unsigned long added_bits
, removed_bits
;
690 struct cgroup
*cgrp
= &root
->top_cgroup
;
693 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
694 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
695 /* Check that any added subsystems are currently free */
696 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
697 unsigned long bit
= 1UL << i
;
698 struct cgroup_subsys
*ss
= subsys
[i
];
699 if (!(bit
& added_bits
))
701 if (ss
->root
!= &rootnode
) {
702 /* Subsystem isn't free */
707 /* Currently we don't handle adding/removing subsystems when
708 * any child cgroups exist. This is theoretically supportable
709 * but involves complex error handling, so it's being left until
711 if (!list_empty(&cgrp
->children
))
714 /* Process each subsystem */
715 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
716 struct cgroup_subsys
*ss
= subsys
[i
];
717 unsigned long bit
= 1UL << i
;
718 if (bit
& added_bits
) {
719 /* We're binding this subsystem to this hierarchy */
720 BUG_ON(cgrp
->subsys
[i
]);
721 BUG_ON(!dummytop
->subsys
[i
]);
722 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
723 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
724 cgrp
->subsys
[i
]->cgroup
= cgrp
;
725 list_add(&ss
->sibling
, &root
->subsys_list
);
726 rcu_assign_pointer(ss
->root
, root
);
730 } else if (bit
& removed_bits
) {
731 /* We're removing this subsystem */
732 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
733 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
735 ss
->bind(ss
, dummytop
);
736 dummytop
->subsys
[i
]->cgroup
= dummytop
;
737 cgrp
->subsys
[i
] = NULL
;
738 rcu_assign_pointer(subsys
[i
]->root
, &rootnode
);
739 list_del(&ss
->sibling
);
740 } else if (bit
& final_bits
) {
741 /* Subsystem state should already exist */
742 BUG_ON(!cgrp
->subsys
[i
]);
744 /* Subsystem state shouldn't exist */
745 BUG_ON(cgrp
->subsys
[i
]);
748 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
754 static int cgroup_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
756 struct cgroupfs_root
*root
= vfs
->mnt_sb
->s_fs_info
;
757 struct cgroup_subsys
*ss
;
759 mutex_lock(&cgroup_mutex
);
760 for_each_subsys(root
, ss
)
761 seq_printf(seq
, ",%s", ss
->name
);
762 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
763 seq_puts(seq
, ",noprefix");
764 if (strlen(root
->release_agent_path
))
765 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
766 mutex_unlock(&cgroup_mutex
);
770 struct cgroup_sb_opts
{
771 unsigned long subsys_bits
;
776 /* Convert a hierarchy specifier into a bitmask of subsystems and
778 static int parse_cgroupfs_options(char *data
,
779 struct cgroup_sb_opts
*opts
)
781 char *token
, *o
= data
?: "all";
783 opts
->subsys_bits
= 0;
785 opts
->release_agent
= NULL
;
787 while ((token
= strsep(&o
, ",")) != NULL
) {
790 if (!strcmp(token
, "all")) {
791 /* Add all non-disabled subsystems */
793 opts
->subsys_bits
= 0;
794 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
795 struct cgroup_subsys
*ss
= subsys
[i
];
797 opts
->subsys_bits
|= 1ul << i
;
799 } else if (!strcmp(token
, "noprefix")) {
800 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
801 } else if (!strncmp(token
, "release_agent=", 14)) {
802 /* Specifying two release agents is forbidden */
803 if (opts
->release_agent
)
805 opts
->release_agent
= kzalloc(PATH_MAX
, GFP_KERNEL
);
806 if (!opts
->release_agent
)
808 strncpy(opts
->release_agent
, token
+ 14, PATH_MAX
- 1);
809 opts
->release_agent
[PATH_MAX
- 1] = 0;
811 struct cgroup_subsys
*ss
;
813 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
815 if (!strcmp(token
, ss
->name
)) {
817 set_bit(i
, &opts
->subsys_bits
);
821 if (i
== CGROUP_SUBSYS_COUNT
)
826 /* We can't have an empty hierarchy */
827 if (!opts
->subsys_bits
)
833 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
836 struct cgroupfs_root
*root
= sb
->s_fs_info
;
837 struct cgroup
*cgrp
= &root
->top_cgroup
;
838 struct cgroup_sb_opts opts
;
840 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
841 mutex_lock(&cgroup_mutex
);
843 /* See what subsystems are wanted */
844 ret
= parse_cgroupfs_options(data
, &opts
);
848 /* Don't allow flags to change at remount */
849 if (opts
.flags
!= root
->flags
) {
854 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
856 /* (re)populate subsystem files */
858 cgroup_populate_dir(cgrp
);
860 if (opts
.release_agent
)
861 strcpy(root
->release_agent_path
, opts
.release_agent
);
863 if (opts
.release_agent
)
864 kfree(opts
.release_agent
);
865 mutex_unlock(&cgroup_mutex
);
866 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
870 static struct super_operations cgroup_ops
= {
871 .statfs
= simple_statfs
,
872 .drop_inode
= generic_delete_inode
,
873 .show_options
= cgroup_show_options
,
874 .remount_fs
= cgroup_remount
,
877 static void init_cgroup_root(struct cgroupfs_root
*root
)
879 struct cgroup
*cgrp
= &root
->top_cgroup
;
880 INIT_LIST_HEAD(&root
->subsys_list
);
881 INIT_LIST_HEAD(&root
->root_list
);
882 root
->number_of_cgroups
= 1;
884 cgrp
->top_cgroup
= cgrp
;
885 INIT_LIST_HEAD(&cgrp
->sibling
);
886 INIT_LIST_HEAD(&cgrp
->children
);
887 INIT_LIST_HEAD(&cgrp
->css_sets
);
888 INIT_LIST_HEAD(&cgrp
->release_list
);
891 static int cgroup_test_super(struct super_block
*sb
, void *data
)
893 struct cgroupfs_root
*new = data
;
894 struct cgroupfs_root
*root
= sb
->s_fs_info
;
896 /* First check subsystems */
897 if (new->subsys_bits
!= root
->subsys_bits
)
900 /* Next check flags */
901 if (new->flags
!= root
->flags
)
907 static int cgroup_set_super(struct super_block
*sb
, void *data
)
910 struct cgroupfs_root
*root
= data
;
912 ret
= set_anon_super(sb
, NULL
);
916 sb
->s_fs_info
= root
;
919 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
920 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
921 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
922 sb
->s_op
= &cgroup_ops
;
927 static int cgroup_get_rootdir(struct super_block
*sb
)
929 struct inode
*inode
=
930 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
931 struct dentry
*dentry
;
936 inode
->i_fop
= &simple_dir_operations
;
937 inode
->i_op
= &cgroup_dir_inode_operations
;
938 /* directories start off with i_nlink == 2 (for "." entry) */
940 dentry
= d_alloc_root(inode
);
949 static int cgroup_get_sb(struct file_system_type
*fs_type
,
950 int flags
, const char *unused_dev_name
,
951 void *data
, struct vfsmount
*mnt
)
953 struct cgroup_sb_opts opts
;
955 struct super_block
*sb
;
956 struct cgroupfs_root
*root
;
957 struct list_head tmp_cg_links
;
958 INIT_LIST_HEAD(&tmp_cg_links
);
960 /* First find the desired set of subsystems */
961 ret
= parse_cgroupfs_options(data
, &opts
);
963 if (opts
.release_agent
)
964 kfree(opts
.release_agent
);
968 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
970 if (opts
.release_agent
)
971 kfree(opts
.release_agent
);
975 init_cgroup_root(root
);
976 root
->subsys_bits
= opts
.subsys_bits
;
977 root
->flags
= opts
.flags
;
978 if (opts
.release_agent
) {
979 strcpy(root
->release_agent_path
, opts
.release_agent
);
980 kfree(opts
.release_agent
);
983 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, root
);
990 if (sb
->s_fs_info
!= root
) {
991 /* Reusing an existing superblock */
992 BUG_ON(sb
->s_root
== NULL
);
997 struct cgroup
*cgrp
= &root
->top_cgroup
;
1001 BUG_ON(sb
->s_root
!= NULL
);
1003 ret
= cgroup_get_rootdir(sb
);
1005 goto drop_new_super
;
1006 inode
= sb
->s_root
->d_inode
;
1008 mutex_lock(&inode
->i_mutex
);
1009 mutex_lock(&cgroup_mutex
);
1012 * We're accessing css_set_count without locking
1013 * css_set_lock here, but that's OK - it can only be
1014 * increased by someone holding cgroup_lock, and
1015 * that's us. The worst that can happen is that we
1016 * have some link structures left over
1018 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
1020 mutex_unlock(&cgroup_mutex
);
1021 mutex_unlock(&inode
->i_mutex
);
1022 goto drop_new_super
;
1025 ret
= rebind_subsystems(root
, root
->subsys_bits
);
1026 if (ret
== -EBUSY
) {
1027 mutex_unlock(&cgroup_mutex
);
1028 mutex_unlock(&inode
->i_mutex
);
1029 goto drop_new_super
;
1032 /* EBUSY should be the only error here */
1035 list_add(&root
->root_list
, &roots
);
1038 sb
->s_root
->d_fsdata
= &root
->top_cgroup
;
1039 root
->top_cgroup
.dentry
= sb
->s_root
;
1041 /* Link the top cgroup in this hierarchy into all
1042 * the css_set objects */
1043 write_lock(&css_set_lock
);
1044 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++) {
1045 struct hlist_head
*hhead
= &css_set_table
[i
];
1046 struct hlist_node
*node
;
1049 hlist_for_each_entry(cg
, node
, hhead
, hlist
) {
1050 struct cg_cgroup_link
*link
;
1052 BUG_ON(list_empty(&tmp_cg_links
));
1053 link
= list_entry(tmp_cg_links
.next
,
1054 struct cg_cgroup_link
,
1056 list_del(&link
->cgrp_link_list
);
1058 list_add(&link
->cgrp_link_list
,
1059 &root
->top_cgroup
.css_sets
);
1060 list_add(&link
->cg_link_list
, &cg
->cg_links
);
1063 write_unlock(&css_set_lock
);
1065 free_cg_links(&tmp_cg_links
);
1067 BUG_ON(!list_empty(&cgrp
->sibling
));
1068 BUG_ON(!list_empty(&cgrp
->children
));
1069 BUG_ON(root
->number_of_cgroups
!= 1);
1071 cgroup_populate_dir(cgrp
);
1072 mutex_unlock(&inode
->i_mutex
);
1073 mutex_unlock(&cgroup_mutex
);
1076 return simple_set_mnt(mnt
, sb
);
1079 up_write(&sb
->s_umount
);
1080 deactivate_super(sb
);
1081 free_cg_links(&tmp_cg_links
);
1085 static void cgroup_kill_sb(struct super_block
*sb
) {
1086 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1087 struct cgroup
*cgrp
= &root
->top_cgroup
;
1089 struct cg_cgroup_link
*link
;
1090 struct cg_cgroup_link
*saved_link
;
1094 BUG_ON(root
->number_of_cgroups
!= 1);
1095 BUG_ON(!list_empty(&cgrp
->children
));
1096 BUG_ON(!list_empty(&cgrp
->sibling
));
1098 mutex_lock(&cgroup_mutex
);
1100 /* Rebind all subsystems back to the default hierarchy */
1101 ret
= rebind_subsystems(root
, 0);
1102 /* Shouldn't be able to fail ... */
1106 * Release all the links from css_sets to this hierarchy's
1109 write_lock(&css_set_lock
);
1111 list_for_each_entry_safe(link
, saved_link
, &cgrp
->css_sets
,
1113 list_del(&link
->cg_link_list
);
1114 list_del(&link
->cgrp_link_list
);
1117 write_unlock(&css_set_lock
);
1119 if (!list_empty(&root
->root_list
)) {
1120 list_del(&root
->root_list
);
1123 mutex_unlock(&cgroup_mutex
);
1126 kill_litter_super(sb
);
1129 static struct file_system_type cgroup_fs_type
= {
1131 .get_sb
= cgroup_get_sb
,
1132 .kill_sb
= cgroup_kill_sb
,
1135 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1137 return dentry
->d_fsdata
;
1140 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1142 return dentry
->d_fsdata
;
1146 * cgroup_path - generate the path of a cgroup
1147 * @cgrp: the cgroup in question
1148 * @buf: the buffer to write the path into
1149 * @buflen: the length of the buffer
1151 * Called with cgroup_mutex held. Writes path of cgroup into buf.
1152 * Returns 0 on success, -errno on error.
1154 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1158 if (cgrp
== dummytop
) {
1160 * Inactive subsystems have no dentry for their root
1167 start
= buf
+ buflen
;
1171 int len
= cgrp
->dentry
->d_name
.len
;
1172 if ((start
-= len
) < buf
)
1173 return -ENAMETOOLONG
;
1174 memcpy(start
, cgrp
->dentry
->d_name
.name
, len
);
1175 cgrp
= cgrp
->parent
;
1181 return -ENAMETOOLONG
;
1184 memmove(buf
, start
, buf
+ buflen
- start
);
1189 * Return the first subsystem attached to a cgroup's hierarchy, and
1193 static void get_first_subsys(const struct cgroup
*cgrp
,
1194 struct cgroup_subsys_state
**css
, int *subsys_id
)
1196 const struct cgroupfs_root
*root
= cgrp
->root
;
1197 const struct cgroup_subsys
*test_ss
;
1198 BUG_ON(list_empty(&root
->subsys_list
));
1199 test_ss
= list_entry(root
->subsys_list
.next
,
1200 struct cgroup_subsys
, sibling
);
1202 *css
= cgrp
->subsys
[test_ss
->subsys_id
];
1206 *subsys_id
= test_ss
->subsys_id
;
1210 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1211 * @cgrp: the cgroup the task is attaching to
1212 * @tsk: the task to be attached
1214 * Call holding cgroup_mutex. May take task_lock of
1215 * the task 'tsk' during call.
1217 int cgroup_attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1220 struct cgroup_subsys
*ss
;
1221 struct cgroup
*oldcgrp
;
1222 struct css_set
*cg
= tsk
->cgroups
;
1223 struct css_set
*newcg
;
1224 struct cgroupfs_root
*root
= cgrp
->root
;
1227 get_first_subsys(cgrp
, NULL
, &subsys_id
);
1229 /* Nothing to do if the task is already in that cgroup */
1230 oldcgrp
= task_cgroup(tsk
, subsys_id
);
1231 if (cgrp
== oldcgrp
)
1234 for_each_subsys(root
, ss
) {
1235 if (ss
->can_attach
) {
1236 retval
= ss
->can_attach(ss
, cgrp
, tsk
);
1243 * Locate or allocate a new css_set for this task,
1244 * based on its final set of cgroups
1246 newcg
= find_css_set(cg
, cgrp
);
1251 if (tsk
->flags
& PF_EXITING
) {
1256 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1259 /* Update the css_set linked lists if we're using them */
1260 write_lock(&css_set_lock
);
1261 if (!list_empty(&tsk
->cg_list
)) {
1262 list_del(&tsk
->cg_list
);
1263 list_add(&tsk
->cg_list
, &newcg
->tasks
);
1265 write_unlock(&css_set_lock
);
1267 for_each_subsys(root
, ss
) {
1269 ss
->attach(ss
, cgrp
, oldcgrp
, tsk
);
1271 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1278 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with cgroup_mutex
1279 * held. May take task_lock of task
1281 static int attach_task_by_pid(struct cgroup
*cgrp
, u64 pid
)
1283 struct task_struct
*tsk
;
1288 tsk
= find_task_by_vpid(pid
);
1289 if (!tsk
|| tsk
->flags
& PF_EXITING
) {
1293 get_task_struct(tsk
);
1296 if ((current
->euid
) && (current
->euid
!= tsk
->uid
)
1297 && (current
->euid
!= tsk
->suid
)) {
1298 put_task_struct(tsk
);
1303 get_task_struct(tsk
);
1306 ret
= cgroup_attach_task(cgrp
, tsk
);
1307 put_task_struct(tsk
);
1311 static int cgroup_tasks_write(struct cgroup
*cgrp
, struct cftype
*cft
, u64 pid
)
1314 if (!cgroup_lock_live_group(cgrp
))
1316 ret
= attach_task_by_pid(cgrp
, pid
);
1321 /* The various types of files and directories in a cgroup file system */
1322 enum cgroup_filetype
{
1326 FILE_NOTIFY_ON_RELEASE
,
1331 * cgroup_lock_live_group - take cgroup_mutex and check that cgrp is alive.
1332 * @cgrp: the cgroup to be checked for liveness
1334 * On success, returns true; the lock should be later released with
1335 * cgroup_unlock(). On failure returns false with no lock held.
1337 bool cgroup_lock_live_group(struct cgroup
*cgrp
)
1339 mutex_lock(&cgroup_mutex
);
1340 if (cgroup_is_removed(cgrp
)) {
1341 mutex_unlock(&cgroup_mutex
);
1347 static int cgroup_release_agent_write(struct cgroup
*cgrp
, struct cftype
*cft
,
1350 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
1351 if (!cgroup_lock_live_group(cgrp
))
1353 strcpy(cgrp
->root
->release_agent_path
, buffer
);
1358 static int cgroup_release_agent_show(struct cgroup
*cgrp
, struct cftype
*cft
,
1359 struct seq_file
*seq
)
1361 if (!cgroup_lock_live_group(cgrp
))
1363 seq_puts(seq
, cgrp
->root
->release_agent_path
);
1364 seq_putc(seq
, '\n');
1369 /* A buffer size big enough for numbers or short strings */
1370 #define CGROUP_LOCAL_BUFFER_SIZE 64
1372 static ssize_t
cgroup_write_X64(struct cgroup
*cgrp
, struct cftype
*cft
,
1374 const char __user
*userbuf
,
1375 size_t nbytes
, loff_t
*unused_ppos
)
1377 char buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
1383 if (nbytes
>= sizeof(buffer
))
1385 if (copy_from_user(buffer
, userbuf
, nbytes
))
1388 buffer
[nbytes
] = 0; /* nul-terminate */
1390 if (cft
->write_u64
) {
1391 u64 val
= simple_strtoull(buffer
, &end
, 0);
1394 retval
= cft
->write_u64(cgrp
, cft
, val
);
1396 s64 val
= simple_strtoll(buffer
, &end
, 0);
1399 retval
= cft
->write_s64(cgrp
, cft
, val
);
1406 static ssize_t
cgroup_write_string(struct cgroup
*cgrp
, struct cftype
*cft
,
1408 const char __user
*userbuf
,
1409 size_t nbytes
, loff_t
*unused_ppos
)
1411 char local_buffer
[CGROUP_LOCAL_BUFFER_SIZE
];
1413 size_t max_bytes
= cft
->max_write_len
;
1414 char *buffer
= local_buffer
;
1417 max_bytes
= sizeof(local_buffer
) - 1;
1418 if (nbytes
>= max_bytes
)
1420 /* Allocate a dynamic buffer if we need one */
1421 if (nbytes
>= sizeof(local_buffer
)) {
1422 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
1426 if (nbytes
&& copy_from_user(buffer
, userbuf
, nbytes
))
1429 buffer
[nbytes
] = 0; /* nul-terminate */
1431 retval
= cft
->write_string(cgrp
, cft
, buffer
);
1434 if (buffer
!= local_buffer
)
1439 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
1440 size_t nbytes
, loff_t
*ppos
)
1442 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1443 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
1445 if (!cft
|| cgroup_is_removed(cgrp
))
1448 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1449 if (cft
->write_u64
|| cft
->write_s64
)
1450 return cgroup_write_X64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1451 if (cft
->write_string
)
1452 return cgroup_write_string(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1454 int ret
= cft
->trigger(cgrp
, (unsigned int)cft
->private);
1455 return ret
? ret
: nbytes
;
1460 static ssize_t
cgroup_read_u64(struct cgroup
*cgrp
, struct cftype
*cft
,
1462 char __user
*buf
, size_t nbytes
,
1465 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
1466 u64 val
= cft
->read_u64(cgrp
, cft
);
1467 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
1469 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
1472 static ssize_t
cgroup_read_s64(struct cgroup
*cgrp
, struct cftype
*cft
,
1474 char __user
*buf
, size_t nbytes
,
1477 char tmp
[CGROUP_LOCAL_BUFFER_SIZE
];
1478 s64 val
= cft
->read_s64(cgrp
, cft
);
1479 int len
= sprintf(tmp
, "%lld\n", (long long) val
);
1481 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
1484 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
1485 size_t nbytes
, loff_t
*ppos
)
1487 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1488 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
1490 if (!cft
|| cgroup_is_removed(cgrp
))
1494 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1496 return cgroup_read_u64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1498 return cgroup_read_s64(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1503 * seqfile ops/methods for returning structured data. Currently just
1504 * supports string->u64 maps, but can be extended in future.
1507 struct cgroup_seqfile_state
{
1509 struct cgroup
*cgroup
;
1512 static int cgroup_map_add(struct cgroup_map_cb
*cb
, const char *key
, u64 value
)
1514 struct seq_file
*sf
= cb
->state
;
1515 return seq_printf(sf
, "%s %llu\n", key
, (unsigned long long)value
);
1518 static int cgroup_seqfile_show(struct seq_file
*m
, void *arg
)
1520 struct cgroup_seqfile_state
*state
= m
->private;
1521 struct cftype
*cft
= state
->cft
;
1522 if (cft
->read_map
) {
1523 struct cgroup_map_cb cb
= {
1524 .fill
= cgroup_map_add
,
1527 return cft
->read_map(state
->cgroup
, cft
, &cb
);
1529 return cft
->read_seq_string(state
->cgroup
, cft
, m
);
1532 int cgroup_seqfile_release(struct inode
*inode
, struct file
*file
)
1534 struct seq_file
*seq
= file
->private_data
;
1535 kfree(seq
->private);
1536 return single_release(inode
, file
);
1539 static struct file_operations cgroup_seqfile_operations
= {
1541 .write
= cgroup_file_write
,
1542 .llseek
= seq_lseek
,
1543 .release
= cgroup_seqfile_release
,
1546 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
1551 err
= generic_file_open(inode
, file
);
1555 cft
= __d_cft(file
->f_dentry
);
1558 if (cft
->read_map
|| cft
->read_seq_string
) {
1559 struct cgroup_seqfile_state
*state
=
1560 kzalloc(sizeof(*state
), GFP_USER
);
1564 state
->cgroup
= __d_cgrp(file
->f_dentry
->d_parent
);
1565 file
->f_op
= &cgroup_seqfile_operations
;
1566 err
= single_open(file
, cgroup_seqfile_show
, state
);
1569 } else if (cft
->open
)
1570 err
= cft
->open(inode
, file
);
1577 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
1579 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1581 return cft
->release(inode
, file
);
1586 * cgroup_rename - Only allow simple rename of directories in place.
1588 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1589 struct inode
*new_dir
, struct dentry
*new_dentry
)
1591 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
1593 if (new_dentry
->d_inode
)
1595 if (old_dir
!= new_dir
)
1597 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
1600 static struct file_operations cgroup_file_operations
= {
1601 .read
= cgroup_file_read
,
1602 .write
= cgroup_file_write
,
1603 .llseek
= generic_file_llseek
,
1604 .open
= cgroup_file_open
,
1605 .release
= cgroup_file_release
,
1608 static struct inode_operations cgroup_dir_inode_operations
= {
1609 .lookup
= simple_lookup
,
1610 .mkdir
= cgroup_mkdir
,
1611 .rmdir
= cgroup_rmdir
,
1612 .rename
= cgroup_rename
,
1615 static int cgroup_create_file(struct dentry
*dentry
, int mode
,
1616 struct super_block
*sb
)
1618 static struct dentry_operations cgroup_dops
= {
1619 .d_iput
= cgroup_diput
,
1622 struct inode
*inode
;
1626 if (dentry
->d_inode
)
1629 inode
= cgroup_new_inode(mode
, sb
);
1633 if (S_ISDIR(mode
)) {
1634 inode
->i_op
= &cgroup_dir_inode_operations
;
1635 inode
->i_fop
= &simple_dir_operations
;
1637 /* start off with i_nlink == 2 (for "." entry) */
1640 /* start with the directory inode held, so that we can
1641 * populate it without racing with another mkdir */
1642 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
1643 } else if (S_ISREG(mode
)) {
1645 inode
->i_fop
= &cgroup_file_operations
;
1647 dentry
->d_op
= &cgroup_dops
;
1648 d_instantiate(dentry
, inode
);
1649 dget(dentry
); /* Extra count - pin the dentry in core */
1654 * cgroup_create_dir - create a directory for an object.
1655 * @cgrp: the cgroup we create the directory for. It must have a valid
1656 * ->parent field. And we are going to fill its ->dentry field.
1657 * @dentry: dentry of the new cgroup
1658 * @mode: mode to set on new directory.
1660 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
1663 struct dentry
*parent
;
1666 parent
= cgrp
->parent
->dentry
;
1667 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
1669 dentry
->d_fsdata
= cgrp
;
1670 inc_nlink(parent
->d_inode
);
1671 cgrp
->dentry
= dentry
;
1679 int cgroup_add_file(struct cgroup
*cgrp
,
1680 struct cgroup_subsys
*subsys
,
1681 const struct cftype
*cft
)
1683 struct dentry
*dir
= cgrp
->dentry
;
1684 struct dentry
*dentry
;
1687 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
1688 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
1689 strcpy(name
, subsys
->name
);
1692 strcat(name
, cft
->name
);
1693 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
1694 dentry
= lookup_one_len(name
, dir
, strlen(name
));
1695 if (!IS_ERR(dentry
)) {
1696 error
= cgroup_create_file(dentry
, 0644 | S_IFREG
,
1699 dentry
->d_fsdata
= (void *)cft
;
1702 error
= PTR_ERR(dentry
);
1706 int cgroup_add_files(struct cgroup
*cgrp
,
1707 struct cgroup_subsys
*subsys
,
1708 const struct cftype cft
[],
1712 for (i
= 0; i
< count
; i
++) {
1713 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
1721 * cgroup_task_count - count the number of tasks in a cgroup.
1722 * @cgrp: the cgroup in question
1724 * Return the number of tasks in the cgroup.
1726 int cgroup_task_count(const struct cgroup
*cgrp
)
1729 struct cg_cgroup_link
*link
;
1731 read_lock(&css_set_lock
);
1732 list_for_each_entry(link
, &cgrp
->css_sets
, cgrp_link_list
) {
1733 count
+= atomic_read(&link
->cg
->ref
.refcount
);
1735 read_unlock(&css_set_lock
);
1740 * Advance a list_head iterator. The iterator should be positioned at
1741 * the start of a css_set
1743 static void cgroup_advance_iter(struct cgroup
*cgrp
,
1744 struct cgroup_iter
*it
)
1746 struct list_head
*l
= it
->cg_link
;
1747 struct cg_cgroup_link
*link
;
1750 /* Advance to the next non-empty css_set */
1753 if (l
== &cgrp
->css_sets
) {
1757 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
1759 } while (list_empty(&cg
->tasks
));
1761 it
->task
= cg
->tasks
.next
;
1765 * To reduce the fork() overhead for systems that are not actually
1766 * using their cgroups capability, we don't maintain the lists running
1767 * through each css_set to its tasks until we see the list actually
1768 * used - in other words after the first call to cgroup_iter_start().
1770 * The tasklist_lock is not held here, as do_each_thread() and
1771 * while_each_thread() are protected by RCU.
1773 static void cgroup_enable_task_cg_lists(void)
1775 struct task_struct
*p
, *g
;
1776 write_lock(&css_set_lock
);
1777 use_task_css_set_links
= 1;
1778 do_each_thread(g
, p
) {
1781 * We should check if the process is exiting, otherwise
1782 * it will race with cgroup_exit() in that the list
1783 * entry won't be deleted though the process has exited.
1785 if (!(p
->flags
& PF_EXITING
) && list_empty(&p
->cg_list
))
1786 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
1788 } while_each_thread(g
, p
);
1789 write_unlock(&css_set_lock
);
1792 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
1795 * The first time anyone tries to iterate across a cgroup,
1796 * we need to enable the list linking each css_set to its
1797 * tasks, and fix up all existing tasks.
1799 if (!use_task_css_set_links
)
1800 cgroup_enable_task_cg_lists();
1802 read_lock(&css_set_lock
);
1803 it
->cg_link
= &cgrp
->css_sets
;
1804 cgroup_advance_iter(cgrp
, it
);
1807 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
1808 struct cgroup_iter
*it
)
1810 struct task_struct
*res
;
1811 struct list_head
*l
= it
->task
;
1813 /* If the iterator cg is NULL, we have no tasks */
1816 res
= list_entry(l
, struct task_struct
, cg_list
);
1817 /* Advance iterator to find next entry */
1819 if (l
== &res
->cgroups
->tasks
) {
1820 /* We reached the end of this task list - move on to
1821 * the next cg_cgroup_link */
1822 cgroup_advance_iter(cgrp
, it
);
1829 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
1831 read_unlock(&css_set_lock
);
1834 static inline int started_after_time(struct task_struct
*t1
,
1835 struct timespec
*time
,
1836 struct task_struct
*t2
)
1838 int start_diff
= timespec_compare(&t1
->start_time
, time
);
1839 if (start_diff
> 0) {
1841 } else if (start_diff
< 0) {
1845 * Arbitrarily, if two processes started at the same
1846 * time, we'll say that the lower pointer value
1847 * started first. Note that t2 may have exited by now
1848 * so this may not be a valid pointer any longer, but
1849 * that's fine - it still serves to distinguish
1850 * between two tasks started (effectively) simultaneously.
1857 * This function is a callback from heap_insert() and is used to order
1859 * In this case we order the heap in descending task start time.
1861 static inline int started_after(void *p1
, void *p2
)
1863 struct task_struct
*t1
= p1
;
1864 struct task_struct
*t2
= p2
;
1865 return started_after_time(t1
, &t2
->start_time
, t2
);
1869 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
1870 * @scan: struct cgroup_scanner containing arguments for the scan
1872 * Arguments include pointers to callback functions test_task() and
1874 * Iterate through all the tasks in a cgroup, calling test_task() for each,
1875 * and if it returns true, call process_task() for it also.
1876 * The test_task pointer may be NULL, meaning always true (select all tasks).
1877 * Effectively duplicates cgroup_iter_{start,next,end}()
1878 * but does not lock css_set_lock for the call to process_task().
1879 * The struct cgroup_scanner may be embedded in any structure of the caller's
1881 * It is guaranteed that process_task() will act on every task that
1882 * is a member of the cgroup for the duration of this call. This
1883 * function may or may not call process_task() for tasks that exit
1884 * or move to a different cgroup during the call, or are forked or
1885 * move into the cgroup during the call.
1887 * Note that test_task() may be called with locks held, and may in some
1888 * situations be called multiple times for the same task, so it should
1890 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
1891 * pre-allocated and will be used for heap operations (and its "gt" member will
1892 * be overwritten), else a temporary heap will be used (allocation of which
1893 * may cause this function to fail).
1895 int cgroup_scan_tasks(struct cgroup_scanner
*scan
)
1898 struct cgroup_iter it
;
1899 struct task_struct
*p
, *dropped
;
1900 /* Never dereference latest_task, since it's not refcounted */
1901 struct task_struct
*latest_task
= NULL
;
1902 struct ptr_heap tmp_heap
;
1903 struct ptr_heap
*heap
;
1904 struct timespec latest_time
= { 0, 0 };
1907 /* The caller supplied our heap and pre-allocated its memory */
1909 heap
->gt
= &started_after
;
1911 /* We need to allocate our own heap memory */
1913 retval
= heap_init(heap
, PAGE_SIZE
, GFP_KERNEL
, &started_after
);
1915 /* cannot allocate the heap */
1921 * Scan tasks in the cgroup, using the scanner's "test_task" callback
1922 * to determine which are of interest, and using the scanner's
1923 * "process_task" callback to process any of them that need an update.
1924 * Since we don't want to hold any locks during the task updates,
1925 * gather tasks to be processed in a heap structure.
1926 * The heap is sorted by descending task start time.
1927 * If the statically-sized heap fills up, we overflow tasks that
1928 * started later, and in future iterations only consider tasks that
1929 * started after the latest task in the previous pass. This
1930 * guarantees forward progress and that we don't miss any tasks.
1933 cgroup_iter_start(scan
->cg
, &it
);
1934 while ((p
= cgroup_iter_next(scan
->cg
, &it
))) {
1936 * Only affect tasks that qualify per the caller's callback,
1937 * if he provided one
1939 if (scan
->test_task
&& !scan
->test_task(p
, scan
))
1942 * Only process tasks that started after the last task
1945 if (!started_after_time(p
, &latest_time
, latest_task
))
1947 dropped
= heap_insert(heap
, p
);
1948 if (dropped
== NULL
) {
1950 * The new task was inserted; the heap wasn't
1954 } else if (dropped
!= p
) {
1956 * The new task was inserted, and pushed out a
1960 put_task_struct(dropped
);
1963 * Else the new task was newer than anything already in
1964 * the heap and wasn't inserted
1967 cgroup_iter_end(scan
->cg
, &it
);
1970 for (i
= 0; i
< heap
->size
; i
++) {
1971 struct task_struct
*q
= heap
->ptrs
[i
];
1973 latest_time
= q
->start_time
;
1976 /* Process the task per the caller's callback */
1977 scan
->process_task(q
, scan
);
1981 * If we had to process any tasks at all, scan again
1982 * in case some of them were in the middle of forking
1983 * children that didn't get processed.
1984 * Not the most efficient way to do it, but it avoids
1985 * having to take callback_mutex in the fork path
1989 if (heap
== &tmp_heap
)
1990 heap_free(&tmp_heap
);
1995 * Stuff for reading the 'tasks' file.
1997 * Reading this file can return large amounts of data if a cgroup has
1998 * *lots* of attached tasks. So it may need several calls to read(),
1999 * but we cannot guarantee that the information we produce is correct
2000 * unless we produce it entirely atomically.
2002 * Upon tasks file open(), a struct ctr_struct is allocated, that
2003 * will have a pointer to an array (also allocated here). The struct
2004 * ctr_struct * is stored in file->private_data. Its resources will
2005 * be freed by release() when the file is closed. The array is used
2006 * to sprintf the PIDs and then used by read().
2014 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
2015 * 'cgrp'. Return actual number of pids loaded. No need to
2016 * task_lock(p) when reading out p->cgroup, since we're in an RCU
2017 * read section, so the css_set can't go away, and is
2018 * immutable after creation.
2020 static int pid_array_load(pid_t
*pidarray
, int npids
, struct cgroup
*cgrp
)
2023 struct cgroup_iter it
;
2024 struct task_struct
*tsk
;
2025 cgroup_iter_start(cgrp
, &it
);
2026 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
2027 if (unlikely(n
== npids
))
2029 pidarray
[n
++] = task_pid_vnr(tsk
);
2031 cgroup_iter_end(cgrp
, &it
);
2036 * cgroupstats_build - build and fill cgroupstats
2037 * @stats: cgroupstats to fill information into
2038 * @dentry: A dentry entry belonging to the cgroup for which stats have
2041 * Build and fill cgroupstats so that taskstats can export it to user
2044 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
2047 struct cgroup
*cgrp
;
2048 struct cgroup_iter it
;
2049 struct task_struct
*tsk
;
2051 * Validate dentry by checking the superblock operations
2053 if (dentry
->d_sb
->s_op
!= &cgroup_ops
)
2057 cgrp
= dentry
->d_fsdata
;
2060 cgroup_iter_start(cgrp
, &it
);
2061 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
2062 switch (tsk
->state
) {
2064 stats
->nr_running
++;
2066 case TASK_INTERRUPTIBLE
:
2067 stats
->nr_sleeping
++;
2069 case TASK_UNINTERRUPTIBLE
:
2070 stats
->nr_uninterruptible
++;
2073 stats
->nr_stopped
++;
2076 if (delayacct_is_task_waiting_on_io(tsk
))
2077 stats
->nr_io_wait
++;
2081 cgroup_iter_end(cgrp
, &it
);
2088 static int cmppid(const void *a
, const void *b
)
2090 return *(pid_t
*)a
- *(pid_t
*)b
;
2094 * Convert array 'a' of 'npids' pid_t's to a string of newline separated
2095 * decimal pids in 'buf'. Don't write more than 'sz' chars, but return
2096 * count 'cnt' of how many chars would be written if buf were large enough.
2098 static int pid_array_to_buf(char *buf
, int sz
, pid_t
*a
, int npids
)
2103 for (i
= 0; i
< npids
; i
++)
2104 cnt
+= snprintf(buf
+ cnt
, max(sz
- cnt
, 0), "%d\n", a
[i
]);
2109 * Handle an open on 'tasks' file. Prepare a buffer listing the
2110 * process id's of tasks currently attached to the cgroup being opened.
2112 * Does not require any specific cgroup mutexes, and does not take any.
2114 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
2116 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
2117 struct ctr_struct
*ctr
;
2122 if (!(file
->f_mode
& FMODE_READ
))
2125 ctr
= kmalloc(sizeof(*ctr
), GFP_KERNEL
);
2130 * If cgroup gets more users after we read count, we won't have
2131 * enough space - tough. This race is indistinguishable to the
2132 * caller from the case that the additional cgroup users didn't
2133 * show up until sometime later on.
2135 npids
= cgroup_task_count(cgrp
);
2137 pidarray
= kmalloc(npids
* sizeof(pid_t
), GFP_KERNEL
);
2141 npids
= pid_array_load(pidarray
, npids
, cgrp
);
2142 sort(pidarray
, npids
, sizeof(pid_t
), cmppid
, NULL
);
2144 /* Call pid_array_to_buf() twice, first just to get bufsz */
2145 ctr
->bufsz
= pid_array_to_buf(&c
, sizeof(c
), pidarray
, npids
) + 1;
2146 ctr
->buf
= kmalloc(ctr
->bufsz
, GFP_KERNEL
);
2149 ctr
->bufsz
= pid_array_to_buf(ctr
->buf
, ctr
->bufsz
, pidarray
, npids
);
2156 file
->private_data
= ctr
;
2167 static ssize_t
cgroup_tasks_read(struct cgroup
*cgrp
,
2169 struct file
*file
, char __user
*buf
,
2170 size_t nbytes
, loff_t
*ppos
)
2172 struct ctr_struct
*ctr
= file
->private_data
;
2174 return simple_read_from_buffer(buf
, nbytes
, ppos
, ctr
->buf
, ctr
->bufsz
);
2177 static int cgroup_tasks_release(struct inode
*unused_inode
,
2180 struct ctr_struct
*ctr
;
2182 if (file
->f_mode
& FMODE_READ
) {
2183 ctr
= file
->private_data
;
2190 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
2193 return notify_on_release(cgrp
);
2196 static int cgroup_write_notify_on_release(struct cgroup
*cgrp
,
2200 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
2202 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
2204 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
2209 * for the common functions, 'private' gives the type of file
2211 static struct cftype files
[] = {
2214 .open
= cgroup_tasks_open
,
2215 .read
= cgroup_tasks_read
,
2216 .write_u64
= cgroup_tasks_write
,
2217 .release
= cgroup_tasks_release
,
2218 .private = FILE_TASKLIST
,
2222 .name
= "notify_on_release",
2223 .read_u64
= cgroup_read_notify_on_release
,
2224 .write_u64
= cgroup_write_notify_on_release
,
2225 .private = FILE_NOTIFY_ON_RELEASE
,
2229 static struct cftype cft_release_agent
= {
2230 .name
= "release_agent",
2231 .read_seq_string
= cgroup_release_agent_show
,
2232 .write_string
= cgroup_release_agent_write
,
2233 .max_write_len
= PATH_MAX
,
2234 .private = FILE_RELEASE_AGENT
,
2237 static int cgroup_populate_dir(struct cgroup
*cgrp
)
2240 struct cgroup_subsys
*ss
;
2242 /* First clear out any existing files */
2243 cgroup_clear_directory(cgrp
->dentry
);
2245 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
2249 if (cgrp
== cgrp
->top_cgroup
) {
2250 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
2254 for_each_subsys(cgrp
->root
, ss
) {
2255 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
2262 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
2263 struct cgroup_subsys
*ss
,
2264 struct cgroup
*cgrp
)
2267 atomic_set(&css
->refcnt
, 0);
2269 if (cgrp
== dummytop
)
2270 set_bit(CSS_ROOT
, &css
->flags
);
2271 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
2272 cgrp
->subsys
[ss
->subsys_id
] = css
;
2276 * cgroup_create - create a cgroup
2277 * @parent: cgroup that will be parent of the new cgroup
2278 * @dentry: dentry of the new cgroup
2279 * @mode: mode to set on new inode
2281 * Must be called with the mutex on the parent inode held
2283 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
2286 struct cgroup
*cgrp
;
2287 struct cgroupfs_root
*root
= parent
->root
;
2289 struct cgroup_subsys
*ss
;
2290 struct super_block
*sb
= root
->sb
;
2292 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
2296 /* Grab a reference on the superblock so the hierarchy doesn't
2297 * get deleted on unmount if there are child cgroups. This
2298 * can be done outside cgroup_mutex, since the sb can't
2299 * disappear while someone has an open control file on the
2301 atomic_inc(&sb
->s_active
);
2303 mutex_lock(&cgroup_mutex
);
2305 INIT_LIST_HEAD(&cgrp
->sibling
);
2306 INIT_LIST_HEAD(&cgrp
->children
);
2307 INIT_LIST_HEAD(&cgrp
->css_sets
);
2308 INIT_LIST_HEAD(&cgrp
->release_list
);
2310 cgrp
->parent
= parent
;
2311 cgrp
->root
= parent
->root
;
2312 cgrp
->top_cgroup
= parent
->top_cgroup
;
2314 if (notify_on_release(parent
))
2315 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
2317 for_each_subsys(root
, ss
) {
2318 struct cgroup_subsys_state
*css
= ss
->create(ss
, cgrp
);
2323 init_cgroup_css(css
, ss
, cgrp
);
2326 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
2327 root
->number_of_cgroups
++;
2329 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
2333 /* The cgroup directory was pre-locked for us */
2334 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
2336 err
= cgroup_populate_dir(cgrp
);
2337 /* If err < 0, we have a half-filled directory - oh well ;) */
2339 mutex_unlock(&cgroup_mutex
);
2340 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
2346 list_del(&cgrp
->sibling
);
2347 root
->number_of_cgroups
--;
2351 for_each_subsys(root
, ss
) {
2352 if (cgrp
->subsys
[ss
->subsys_id
])
2353 ss
->destroy(ss
, cgrp
);
2356 mutex_unlock(&cgroup_mutex
);
2358 /* Release the reference count that we took on the superblock */
2359 deactivate_super(sb
);
2365 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2367 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
2369 /* the vfs holds inode->i_mutex already */
2370 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
2373 static inline int cgroup_has_css_refs(struct cgroup
*cgrp
)
2375 /* Check the reference count on each subsystem. Since we
2376 * already established that there are no tasks in the
2377 * cgroup, if the css refcount is also 0, then there should
2378 * be no outstanding references, so the subsystem is safe to
2379 * destroy. We scan across all subsystems rather than using
2380 * the per-hierarchy linked list of mounted subsystems since
2381 * we can be called via check_for_release() with no
2382 * synchronization other than RCU, and the subsystem linked
2383 * list isn't RCU-safe */
2385 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2386 struct cgroup_subsys
*ss
= subsys
[i
];
2387 struct cgroup_subsys_state
*css
;
2388 /* Skip subsystems not in this hierarchy */
2389 if (ss
->root
!= cgrp
->root
)
2391 css
= cgrp
->subsys
[ss
->subsys_id
];
2392 /* When called from check_for_release() it's possible
2393 * that by this point the cgroup has been removed
2394 * and the css deleted. But a false-positive doesn't
2395 * matter, since it can only happen if the cgroup
2396 * has been deleted and hence no longer needs the
2397 * release agent to be called anyway. */
2398 if (css
&& atomic_read(&css
->refcnt
))
2404 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
2406 struct cgroup
*cgrp
= dentry
->d_fsdata
;
2408 struct cgroup
*parent
;
2409 struct super_block
*sb
;
2410 struct cgroupfs_root
*root
;
2412 /* the vfs holds both inode->i_mutex already */
2414 mutex_lock(&cgroup_mutex
);
2415 if (atomic_read(&cgrp
->count
) != 0) {
2416 mutex_unlock(&cgroup_mutex
);
2419 if (!list_empty(&cgrp
->children
)) {
2420 mutex_unlock(&cgroup_mutex
);
2424 parent
= cgrp
->parent
;
2429 * Call pre_destroy handlers of subsys. Notify subsystems
2430 * that rmdir() request comes.
2432 cgroup_call_pre_destroy(cgrp
);
2434 if (cgroup_has_css_refs(cgrp
)) {
2435 mutex_unlock(&cgroup_mutex
);
2439 spin_lock(&release_list_lock
);
2440 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
2441 if (!list_empty(&cgrp
->release_list
))
2442 list_del(&cgrp
->release_list
);
2443 spin_unlock(&release_list_lock
);
2444 /* delete my sibling from parent->children */
2445 list_del(&cgrp
->sibling
);
2446 spin_lock(&cgrp
->dentry
->d_lock
);
2447 d
= dget(cgrp
->dentry
);
2448 cgrp
->dentry
= NULL
;
2449 spin_unlock(&d
->d_lock
);
2451 cgroup_d_remove_dir(d
);
2454 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
2455 check_for_release(parent
);
2457 mutex_unlock(&cgroup_mutex
);
2461 static void __init
cgroup_init_subsys(struct cgroup_subsys
*ss
)
2463 struct cgroup_subsys_state
*css
;
2465 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
2467 /* Create the top cgroup state for this subsystem */
2468 ss
->root
= &rootnode
;
2469 css
= ss
->create(ss
, dummytop
);
2470 /* We don't handle early failures gracefully */
2471 BUG_ON(IS_ERR(css
));
2472 init_cgroup_css(css
, ss
, dummytop
);
2474 /* Update the init_css_set to contain a subsys
2475 * pointer to this state - since the subsystem is
2476 * newly registered, all tasks and hence the
2477 * init_css_set is in the subsystem's top cgroup. */
2478 init_css_set
.subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
2480 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
2481 need_mm_owner_callback
|= !!ss
->mm_owner_changed
;
2483 /* At system boot, before all subsystems have been
2484 * registered, no tasks have been forked, so we don't
2485 * need to invoke fork callbacks here. */
2486 BUG_ON(!list_empty(&init_task
.tasks
));
2492 * cgroup_init_early - cgroup initialization at system boot
2494 * Initialize cgroups at system boot, and initialize any
2495 * subsystems that request early init.
2497 int __init
cgroup_init_early(void)
2500 kref_init(&init_css_set
.ref
);
2501 kref_get(&init_css_set
.ref
);
2502 INIT_LIST_HEAD(&init_css_set
.cg_links
);
2503 INIT_LIST_HEAD(&init_css_set
.tasks
);
2504 INIT_HLIST_NODE(&init_css_set
.hlist
);
2506 init_cgroup_root(&rootnode
);
2507 list_add(&rootnode
.root_list
, &roots
);
2509 init_task
.cgroups
= &init_css_set
;
2511 init_css_set_link
.cg
= &init_css_set
;
2512 list_add(&init_css_set_link
.cgrp_link_list
,
2513 &rootnode
.top_cgroup
.css_sets
);
2514 list_add(&init_css_set_link
.cg_link_list
,
2515 &init_css_set
.cg_links
);
2517 for (i
= 0; i
< CSS_SET_TABLE_SIZE
; i
++)
2518 INIT_HLIST_HEAD(&css_set_table
[i
]);
2520 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2521 struct cgroup_subsys
*ss
= subsys
[i
];
2524 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
2525 BUG_ON(!ss
->create
);
2526 BUG_ON(!ss
->destroy
);
2527 if (ss
->subsys_id
!= i
) {
2528 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
2529 ss
->name
, ss
->subsys_id
);
2534 cgroup_init_subsys(ss
);
2540 * cgroup_init - cgroup initialization
2542 * Register cgroup filesystem and /proc file, and initialize
2543 * any subsystems that didn't request early init.
2545 int __init
cgroup_init(void)
2549 struct hlist_head
*hhead
;
2551 err
= bdi_init(&cgroup_backing_dev_info
);
2555 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2556 struct cgroup_subsys
*ss
= subsys
[i
];
2557 if (!ss
->early_init
)
2558 cgroup_init_subsys(ss
);
2561 /* Add init_css_set to the hash table */
2562 hhead
= css_set_hash(init_css_set
.subsys
);
2563 hlist_add_head(&init_css_set
.hlist
, hhead
);
2565 err
= register_filesystem(&cgroup_fs_type
);
2569 proc_create("cgroups", 0, NULL
, &proc_cgroupstats_operations
);
2573 bdi_destroy(&cgroup_backing_dev_info
);
2579 * proc_cgroup_show()
2580 * - Print task's cgroup paths into seq_file, one line for each hierarchy
2581 * - Used for /proc/<pid>/cgroup.
2582 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
2583 * doesn't really matter if tsk->cgroup changes after we read it,
2584 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
2585 * anyway. No need to check that tsk->cgroup != NULL, thanks to
2586 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
2587 * cgroup to top_cgroup.
2590 /* TODO: Use a proper seq_file iterator */
2591 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
2594 struct task_struct
*tsk
;
2597 struct cgroupfs_root
*root
;
2600 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
2606 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
2612 mutex_lock(&cgroup_mutex
);
2614 for_each_root(root
) {
2615 struct cgroup_subsys
*ss
;
2616 struct cgroup
*cgrp
;
2620 /* Skip this hierarchy if it has no active subsystems */
2621 if (!root
->actual_subsys_bits
)
2623 seq_printf(m
, "%lu:", root
->subsys_bits
);
2624 for_each_subsys(root
, ss
)
2625 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
2627 get_first_subsys(&root
->top_cgroup
, NULL
, &subsys_id
);
2628 cgrp
= task_cgroup(tsk
, subsys_id
);
2629 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
2637 mutex_unlock(&cgroup_mutex
);
2638 put_task_struct(tsk
);
2645 static int cgroup_open(struct inode
*inode
, struct file
*file
)
2647 struct pid
*pid
= PROC_I(inode
)->pid
;
2648 return single_open(file
, proc_cgroup_show
, pid
);
2651 struct file_operations proc_cgroup_operations
= {
2652 .open
= cgroup_open
,
2654 .llseek
= seq_lseek
,
2655 .release
= single_release
,
2658 /* Display information about each subsystem and each hierarchy */
2659 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
2663 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
2664 mutex_lock(&cgroup_mutex
);
2665 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2666 struct cgroup_subsys
*ss
= subsys
[i
];
2667 seq_printf(m
, "%s\t%lu\t%d\t%d\n",
2668 ss
->name
, ss
->root
->subsys_bits
,
2669 ss
->root
->number_of_cgroups
, !ss
->disabled
);
2671 mutex_unlock(&cgroup_mutex
);
2675 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
2677 return single_open(file
, proc_cgroupstats_show
, NULL
);
2680 static struct file_operations proc_cgroupstats_operations
= {
2681 .open
= cgroupstats_open
,
2683 .llseek
= seq_lseek
,
2684 .release
= single_release
,
2688 * cgroup_fork - attach newly forked task to its parents cgroup.
2689 * @child: pointer to task_struct of forking parent process.
2691 * Description: A task inherits its parent's cgroup at fork().
2693 * A pointer to the shared css_set was automatically copied in
2694 * fork.c by dup_task_struct(). However, we ignore that copy, since
2695 * it was not made under the protection of RCU or cgroup_mutex, so
2696 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
2697 * have already changed current->cgroups, allowing the previously
2698 * referenced cgroup group to be removed and freed.
2700 * At the point that cgroup_fork() is called, 'current' is the parent
2701 * task, and the passed argument 'child' points to the child task.
2703 void cgroup_fork(struct task_struct
*child
)
2706 child
->cgroups
= current
->cgroups
;
2707 get_css_set(child
->cgroups
);
2708 task_unlock(current
);
2709 INIT_LIST_HEAD(&child
->cg_list
);
2713 * cgroup_fork_callbacks - run fork callbacks
2714 * @child: the new task
2716 * Called on a new task very soon before adding it to the
2717 * tasklist. No need to take any locks since no-one can
2718 * be operating on this task.
2720 void cgroup_fork_callbacks(struct task_struct
*child
)
2722 if (need_forkexit_callback
) {
2724 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2725 struct cgroup_subsys
*ss
= subsys
[i
];
2727 ss
->fork(ss
, child
);
2732 #ifdef CONFIG_MM_OWNER
2734 * cgroup_mm_owner_callbacks - run callbacks when the mm->owner changes
2737 * Called on every change to mm->owner. mm_init_owner() does not
2738 * invoke this routine, since it assigns the mm->owner the first time
2739 * and does not change it.
2741 void cgroup_mm_owner_callbacks(struct task_struct
*old
, struct task_struct
*new)
2743 struct cgroup
*oldcgrp
, *newcgrp
;
2745 if (need_mm_owner_callback
) {
2747 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2748 struct cgroup_subsys
*ss
= subsys
[i
];
2749 oldcgrp
= task_cgroup(old
, ss
->subsys_id
);
2750 newcgrp
= task_cgroup(new, ss
->subsys_id
);
2751 if (oldcgrp
== newcgrp
)
2753 if (ss
->mm_owner_changed
)
2754 ss
->mm_owner_changed(ss
, oldcgrp
, newcgrp
);
2758 #endif /* CONFIG_MM_OWNER */
2761 * cgroup_post_fork - called on a new task after adding it to the task list
2762 * @child: the task in question
2764 * Adds the task to the list running through its css_set if necessary.
2765 * Has to be after the task is visible on the task list in case we race
2766 * with the first call to cgroup_iter_start() - to guarantee that the
2767 * new task ends up on its list.
2769 void cgroup_post_fork(struct task_struct
*child
)
2771 if (use_task_css_set_links
) {
2772 write_lock(&css_set_lock
);
2773 if (list_empty(&child
->cg_list
))
2774 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
2775 write_unlock(&css_set_lock
);
2779 * cgroup_exit - detach cgroup from exiting task
2780 * @tsk: pointer to task_struct of exiting process
2781 * @run_callback: run exit callbacks?
2783 * Description: Detach cgroup from @tsk and release it.
2785 * Note that cgroups marked notify_on_release force every task in
2786 * them to take the global cgroup_mutex mutex when exiting.
2787 * This could impact scaling on very large systems. Be reluctant to
2788 * use notify_on_release cgroups where very high task exit scaling
2789 * is required on large systems.
2791 * the_top_cgroup_hack:
2793 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
2795 * We call cgroup_exit() while the task is still competent to
2796 * handle notify_on_release(), then leave the task attached to the
2797 * root cgroup in each hierarchy for the remainder of its exit.
2799 * To do this properly, we would increment the reference count on
2800 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
2801 * code we would add a second cgroup function call, to drop that
2802 * reference. This would just create an unnecessary hot spot on
2803 * the top_cgroup reference count, to no avail.
2805 * Normally, holding a reference to a cgroup without bumping its
2806 * count is unsafe. The cgroup could go away, or someone could
2807 * attach us to a different cgroup, decrementing the count on
2808 * the first cgroup that we never incremented. But in this case,
2809 * top_cgroup isn't going away, and either task has PF_EXITING set,
2810 * which wards off any cgroup_attach_task() attempts, or task is a failed
2811 * fork, never visible to cgroup_attach_task.
2813 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
2818 if (run_callbacks
&& need_forkexit_callback
) {
2819 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2820 struct cgroup_subsys
*ss
= subsys
[i
];
2827 * Unlink from the css_set task list if necessary.
2828 * Optimistically check cg_list before taking
2831 if (!list_empty(&tsk
->cg_list
)) {
2832 write_lock(&css_set_lock
);
2833 if (!list_empty(&tsk
->cg_list
))
2834 list_del(&tsk
->cg_list
);
2835 write_unlock(&css_set_lock
);
2838 /* Reassign the task to the init_css_set. */
2841 tsk
->cgroups
= &init_css_set
;
2844 put_css_set_taskexit(cg
);
2848 * cgroup_clone - clone the cgroup the given subsystem is attached to
2849 * @tsk: the task to be moved
2850 * @subsys: the given subsystem
2851 * @nodename: the name for the new cgroup
2853 * Duplicate the current cgroup in the hierarchy that the given
2854 * subsystem is attached to, and move this task into the new
2857 int cgroup_clone(struct task_struct
*tsk
, struct cgroup_subsys
*subsys
,
2860 struct dentry
*dentry
;
2862 struct cgroup
*parent
, *child
;
2863 struct inode
*inode
;
2865 struct cgroupfs_root
*root
;
2866 struct cgroup_subsys
*ss
;
2868 /* We shouldn't be called by an unregistered subsystem */
2869 BUG_ON(!subsys
->active
);
2871 /* First figure out what hierarchy and cgroup we're dealing
2872 * with, and pin them so we can drop cgroup_mutex */
2873 mutex_lock(&cgroup_mutex
);
2875 root
= subsys
->root
;
2876 if (root
== &rootnode
) {
2878 "Not cloning cgroup for unused subsystem %s\n",
2880 mutex_unlock(&cgroup_mutex
);
2884 parent
= task_cgroup(tsk
, subsys
->subsys_id
);
2886 /* Pin the hierarchy */
2887 atomic_inc(&parent
->root
->sb
->s_active
);
2889 /* Keep the cgroup alive */
2891 mutex_unlock(&cgroup_mutex
);
2893 /* Now do the VFS work to create a cgroup */
2894 inode
= parent
->dentry
->d_inode
;
2896 /* Hold the parent directory mutex across this operation to
2897 * stop anyone else deleting the new cgroup */
2898 mutex_lock(&inode
->i_mutex
);
2899 dentry
= lookup_one_len(nodename
, parent
->dentry
, strlen(nodename
));
2900 if (IS_ERR(dentry
)) {
2902 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename
,
2904 ret
= PTR_ERR(dentry
);
2908 /* Create the cgroup directory, which also creates the cgroup */
2909 ret
= vfs_mkdir(inode
, dentry
, S_IFDIR
| 0755);
2910 child
= __d_cgrp(dentry
);
2914 "Failed to create cgroup %s: %d\n", nodename
,
2921 "Couldn't find new cgroup %s\n", nodename
);
2926 /* The cgroup now exists. Retake cgroup_mutex and check
2927 * that we're still in the same state that we thought we
2929 mutex_lock(&cgroup_mutex
);
2930 if ((root
!= subsys
->root
) ||
2931 (parent
!= task_cgroup(tsk
, subsys
->subsys_id
))) {
2932 /* Aargh, we raced ... */
2933 mutex_unlock(&inode
->i_mutex
);
2936 deactivate_super(parent
->root
->sb
);
2937 /* The cgroup is still accessible in the VFS, but
2938 * we're not going to try to rmdir() it at this
2941 "Race in cgroup_clone() - leaking cgroup %s\n",
2946 /* do any required auto-setup */
2947 for_each_subsys(root
, ss
) {
2949 ss
->post_clone(ss
, child
);
2952 /* All seems fine. Finish by moving the task into the new cgroup */
2953 ret
= cgroup_attach_task(child
, tsk
);
2954 mutex_unlock(&cgroup_mutex
);
2957 mutex_unlock(&inode
->i_mutex
);
2959 mutex_lock(&cgroup_mutex
);
2961 mutex_unlock(&cgroup_mutex
);
2962 deactivate_super(parent
->root
->sb
);
2967 * cgroup_is_descendant - see if @cgrp is a descendant of current task's cgrp
2968 * @cgrp: the cgroup in question
2970 * See if @cgrp is a descendant of the current task's cgroup in
2971 * the appropriate hierarchy.
2973 * If we are sending in dummytop, then presumably we are creating
2974 * the top cgroup in the subsystem.
2976 * Called only by the ns (nsproxy) cgroup.
2978 int cgroup_is_descendant(const struct cgroup
*cgrp
)
2981 struct cgroup
*target
;
2984 if (cgrp
== dummytop
)
2987 get_first_subsys(cgrp
, NULL
, &subsys_id
);
2988 target
= task_cgroup(current
, subsys_id
);
2989 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
2990 cgrp
= cgrp
->parent
;
2991 ret
= (cgrp
== target
);
2995 static void check_for_release(struct cgroup
*cgrp
)
2997 /* All of these checks rely on RCU to keep the cgroup
2998 * structure alive */
2999 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
3000 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
3001 /* Control Group is currently removeable. If it's not
3002 * already queued for a userspace notification, queue
3004 int need_schedule_work
= 0;
3005 spin_lock(&release_list_lock
);
3006 if (!cgroup_is_removed(cgrp
) &&
3007 list_empty(&cgrp
->release_list
)) {
3008 list_add(&cgrp
->release_list
, &release_list
);
3009 need_schedule_work
= 1;
3011 spin_unlock(&release_list_lock
);
3012 if (need_schedule_work
)
3013 schedule_work(&release_agent_work
);
3017 void __css_put(struct cgroup_subsys_state
*css
)
3019 struct cgroup
*cgrp
= css
->cgroup
;
3021 if (atomic_dec_and_test(&css
->refcnt
) && notify_on_release(cgrp
)) {
3022 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
3023 check_for_release(cgrp
);
3029 * Notify userspace when a cgroup is released, by running the
3030 * configured release agent with the name of the cgroup (path
3031 * relative to the root of cgroup file system) as the argument.
3033 * Most likely, this user command will try to rmdir this cgroup.
3035 * This races with the possibility that some other task will be
3036 * attached to this cgroup before it is removed, or that some other
3037 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
3038 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
3039 * unused, and this cgroup will be reprieved from its death sentence,
3040 * to continue to serve a useful existence. Next time it's released,
3041 * we will get notified again, if it still has 'notify_on_release' set.
3043 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
3044 * means only wait until the task is successfully execve()'d. The
3045 * separate release agent task is forked by call_usermodehelper(),
3046 * then control in this thread returns here, without waiting for the
3047 * release agent task. We don't bother to wait because the caller of
3048 * this routine has no use for the exit status of the release agent
3049 * task, so no sense holding our caller up for that.
3051 static void cgroup_release_agent(struct work_struct
*work
)
3053 BUG_ON(work
!= &release_agent_work
);
3054 mutex_lock(&cgroup_mutex
);
3055 spin_lock(&release_list_lock
);
3056 while (!list_empty(&release_list
)) {
3057 char *argv
[3], *envp
[3];
3059 char *pathbuf
= NULL
, *agentbuf
= NULL
;
3060 struct cgroup
*cgrp
= list_entry(release_list
.next
,
3063 list_del_init(&cgrp
->release_list
);
3064 spin_unlock(&release_list_lock
);
3065 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3068 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0)
3070 agentbuf
= kstrdup(cgrp
->root
->release_agent_path
, GFP_KERNEL
);
3075 argv
[i
++] = agentbuf
;
3076 argv
[i
++] = pathbuf
;
3080 /* minimal command environment */
3081 envp
[i
++] = "HOME=/";
3082 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
3085 /* Drop the lock while we invoke the usermode helper,
3086 * since the exec could involve hitting disk and hence
3087 * be a slow process */
3088 mutex_unlock(&cgroup_mutex
);
3089 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
3090 mutex_lock(&cgroup_mutex
);
3094 spin_lock(&release_list_lock
);
3096 spin_unlock(&release_list_lock
);
3097 mutex_unlock(&cgroup_mutex
);
3100 static int __init
cgroup_disable(char *str
)
3105 while ((token
= strsep(&str
, ",")) != NULL
) {
3109 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
3110 struct cgroup_subsys
*ss
= subsys
[i
];
3112 if (!strcmp(token
, ss
->name
)) {
3114 printk(KERN_INFO
"Disabling %s control group"
3115 " subsystem\n", ss
->name
);
3122 __setup("cgroup_disable=", cgroup_disable
);