4 * Generic process-grouping system.
6 * Based originally on the cpuset system, extracted by Paul Menage
7 * Copyright (C) 2006 Google, Inc
9 * Copyright notices from the original cpuset code:
10 * --------------------------------------------------
11 * Copyright (C) 2003 BULL SA.
12 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
14 * Portions derived from Patrick Mochel's sysfs code.
15 * sysfs is Copyright (c) 2001-3 Patrick Mochel
17 * 2003-10-10 Written by Simon Derr.
18 * 2003-10-22 Updates by Stephen Hemminger.
19 * 2004 May-July Rework by Paul Jackson.
20 * ---------------------------------------------------
22 * This file is subject to the terms and conditions of the GNU General Public
23 * License. See the file COPYING in the main directory of the Linux
24 * distribution for more details.
27 #include <linux/cgroup.h>
28 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/list.h>
33 #include <linux/mutex.h>
34 #include <linux/mount.h>
35 #include <linux/pagemap.h>
36 #include <linux/rcupdate.h>
37 #include <linux/sched.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>
44 #include <asm/atomic.h>
46 /* Generate an array of cgroup subsystem pointers */
47 #define SUBSYS(_x) &_x ## _subsys,
49 static struct cgroup_subsys
*subsys
[] = {
50 #include <linux/cgroup_subsys.h>
54 * A cgroupfs_root represents the root of a cgroup hierarchy,
55 * and may be associated with a superblock to form an active
58 struct cgroupfs_root
{
59 struct super_block
*sb
;
62 * The bitmask of subsystems intended to be attached to this
65 unsigned long subsys_bits
;
67 /* The bitmask of subsystems currently attached to this hierarchy */
68 unsigned long actual_subsys_bits
;
70 /* A list running through the attached subsystems */
71 struct list_head subsys_list
;
73 /* The root cgroup for this hierarchy */
74 struct cgroup top_cgroup
;
76 /* Tracks how many cgroups are currently defined in hierarchy.*/
77 int number_of_cgroups
;
79 /* A list running through the mounted hierarchies */
80 struct list_head root_list
;
82 /* Hierarchy-specific flags */
88 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
89 * subsystems that are otherwise unattached - it never has more than a
90 * single cgroup, and all tasks are part of that cgroup.
92 static struct cgroupfs_root rootnode
;
94 /* The list of hierarchy roots */
96 static LIST_HEAD(roots
);
98 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
99 #define dummytop (&rootnode.top_cgroup)
101 /* This flag indicates whether tasks in the fork and exit paths should
102 * take callback_mutex and check for fork/exit handlers to call. This
103 * avoids us having to do extra work in the fork/exit path if none of the
104 * subsystems need to be called.
106 static int need_forkexit_callback
;
108 /* bits in struct cgroup flags field */
113 /* convenient tests for these bits */
114 inline int cgroup_is_removed(const struct cgroup
*cont
)
116 return test_bit(CONT_REMOVED
, &cont
->flags
);
119 /* bits in struct cgroupfs_root flags field */
121 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
125 * for_each_subsys() allows you to iterate on each subsystem attached to
126 * an active hierarchy
128 #define for_each_subsys(_root, _ss) \
129 list_for_each_entry(_ss, &_root->subsys_list, sibling)
131 /* for_each_root() allows you to iterate across the active hierarchies */
132 #define for_each_root(_root) \
133 list_for_each_entry(_root, &roots, root_list)
136 * There is one global cgroup mutex. We also require taking
137 * task_lock() when dereferencing a task's cgroup subsys pointers.
138 * See "The task_lock() exception", at the end of this comment.
140 * A task must hold cgroup_mutex to modify cgroups.
142 * Any task can increment and decrement the count field without lock.
143 * So in general, code holding cgroup_mutex can't rely on the count
144 * field not changing. However, if the count goes to zero, then only
145 * attach_task() can increment it again. Because a count of zero
146 * means that no tasks are currently attached, therefore there is no
147 * way a task attached to that cgroup can fork (the other way to
148 * increment the count). So code holding cgroup_mutex can safely
149 * assume that if the count is zero, it will stay zero. Similarly, if
150 * a task holds cgroup_mutex on a cgroup with zero count, it
151 * knows that the cgroup won't be removed, as cgroup_rmdir()
154 * The cgroup_common_file_write handler for operations that modify
155 * the cgroup hierarchy holds cgroup_mutex across the entire operation,
156 * single threading all such cgroup modifications across the system.
158 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
159 * (usually) take cgroup_mutex. These are the two most performance
160 * critical pieces of code here. The exception occurs on cgroup_exit(),
161 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
162 * is taken, and if the cgroup count is zero, a usermode call made
163 * to /sbin/cgroup_release_agent with the name of the cgroup (path
164 * relative to the root of cgroup file system) as the argument.
166 * A cgroup can only be deleted if both its 'count' of using tasks
167 * is zero, and its list of 'children' cgroups is empty. Since all
168 * tasks in the system use _some_ cgroup, and since there is always at
169 * least one task in the system (init, pid == 1), therefore, top_cgroup
170 * always has either children cgroups and/or using tasks. So we don't
171 * need a special hack to ensure that top_cgroup cannot be deleted.
173 * The task_lock() exception
175 * The need for this exception arises from the action of
176 * attach_task(), which overwrites one tasks cgroup pointer with
177 * another. It does so using cgroup_mutexe, however there are
178 * several performance critical places that need to reference
179 * task->cgroup without the expense of grabbing a system global
180 * mutex. Therefore except as noted below, when dereferencing or, as
181 * in attach_task(), modifying a task'ss cgroup pointer we use
182 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
183 * the task_struct routinely used for such matters.
185 * P.S. One more locking exception. RCU is used to guard the
186 * update of a tasks cgroup pointer by attach_task()
189 static DEFINE_MUTEX(cgroup_mutex
);
192 * cgroup_lock - lock out any changes to cgroup structures
196 void cgroup_lock(void)
198 mutex_lock(&cgroup_mutex
);
202 * cgroup_unlock - release lock on cgroup changes
204 * Undo the lock taken in a previous cgroup_lock() call.
207 void cgroup_unlock(void)
209 mutex_unlock(&cgroup_mutex
);
213 * A couple of forward declarations required, due to cyclic reference loop:
214 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
215 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
219 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
);
220 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
);
221 static int cgroup_populate_dir(struct cgroup
*cont
);
222 static struct inode_operations cgroup_dir_inode_operations
;
224 static struct inode
*cgroup_new_inode(mode_t mode
, struct super_block
*sb
)
226 struct inode
*inode
= new_inode(sb
);
227 static struct backing_dev_info cgroup_backing_dev_info
= {
228 .capabilities
= BDI_CAP_NO_ACCT_DIRTY
| BDI_CAP_NO_WRITEBACK
,
232 inode
->i_mode
= mode
;
233 inode
->i_uid
= current
->fsuid
;
234 inode
->i_gid
= current
->fsgid
;
236 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
237 inode
->i_mapping
->backing_dev_info
= &cgroup_backing_dev_info
;
242 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
244 /* is dentry a directory ? if so, kfree() associated cgroup */
245 if (S_ISDIR(inode
->i_mode
)) {
246 struct cgroup
*cont
= dentry
->d_fsdata
;
247 BUG_ON(!(cgroup_is_removed(cont
)));
253 static void remove_dir(struct dentry
*d
)
255 struct dentry
*parent
= dget(d
->d_parent
);
258 simple_rmdir(parent
->d_inode
, d
);
262 static void cgroup_clear_directory(struct dentry
*dentry
)
264 struct list_head
*node
;
266 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
267 spin_lock(&dcache_lock
);
268 node
= dentry
->d_subdirs
.next
;
269 while (node
!= &dentry
->d_subdirs
) {
270 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
273 /* This should never be called on a cgroup
274 * directory with child cgroups */
275 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
277 spin_unlock(&dcache_lock
);
279 simple_unlink(dentry
->d_inode
, d
);
281 spin_lock(&dcache_lock
);
283 node
= dentry
->d_subdirs
.next
;
285 spin_unlock(&dcache_lock
);
289 * NOTE : the dentry must have been dget()'ed
291 static void cgroup_d_remove_dir(struct dentry
*dentry
)
293 cgroup_clear_directory(dentry
);
295 spin_lock(&dcache_lock
);
296 list_del_init(&dentry
->d_u
.d_child
);
297 spin_unlock(&dcache_lock
);
301 static int rebind_subsystems(struct cgroupfs_root
*root
,
302 unsigned long final_bits
)
304 unsigned long added_bits
, removed_bits
;
305 struct cgroup
*cont
= &root
->top_cgroup
;
308 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
309 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
310 /* Check that any added subsystems are currently free */
311 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
312 unsigned long long bit
= 1ull << i
;
313 struct cgroup_subsys
*ss
= subsys
[i
];
314 if (!(bit
& added_bits
))
316 if (ss
->root
!= &rootnode
) {
317 /* Subsystem isn't free */
322 /* Currently we don't handle adding/removing subsystems when
323 * any child cgroups exist. This is theoretically supportable
324 * but involves complex error handling, so it's being left until
326 if (!list_empty(&cont
->children
))
329 /* Process each subsystem */
330 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
331 struct cgroup_subsys
*ss
= subsys
[i
];
332 unsigned long bit
= 1UL << i
;
333 if (bit
& added_bits
) {
334 /* We're binding this subsystem to this hierarchy */
335 BUG_ON(cont
->subsys
[i
]);
336 BUG_ON(!dummytop
->subsys
[i
]);
337 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
338 cont
->subsys
[i
] = dummytop
->subsys
[i
];
339 cont
->subsys
[i
]->cgroup
= cont
;
340 list_add(&ss
->sibling
, &root
->subsys_list
);
341 rcu_assign_pointer(ss
->root
, root
);
345 } else if (bit
& removed_bits
) {
346 /* We're removing this subsystem */
347 BUG_ON(cont
->subsys
[i
] != dummytop
->subsys
[i
]);
348 BUG_ON(cont
->subsys
[i
]->cgroup
!= cont
);
350 ss
->bind(ss
, dummytop
);
351 dummytop
->subsys
[i
]->cgroup
= dummytop
;
352 cont
->subsys
[i
] = NULL
;
353 rcu_assign_pointer(subsys
[i
]->root
, &rootnode
);
354 list_del(&ss
->sibling
);
355 } else if (bit
& final_bits
) {
356 /* Subsystem state should already exist */
357 BUG_ON(!cont
->subsys
[i
]);
359 /* Subsystem state shouldn't exist */
360 BUG_ON(cont
->subsys
[i
]);
363 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
369 static int cgroup_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
371 struct cgroupfs_root
*root
= vfs
->mnt_sb
->s_fs_info
;
372 struct cgroup_subsys
*ss
;
374 mutex_lock(&cgroup_mutex
);
375 for_each_subsys(root
, ss
)
376 seq_printf(seq
, ",%s", ss
->name
);
377 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
378 seq_puts(seq
, ",noprefix");
379 mutex_unlock(&cgroup_mutex
);
383 struct cgroup_sb_opts
{
384 unsigned long subsys_bits
;
388 /* Convert a hierarchy specifier into a bitmask of subsystems and
390 static int parse_cgroupfs_options(char *data
,
391 struct cgroup_sb_opts
*opts
)
393 char *token
, *o
= data
?: "all";
395 opts
->subsys_bits
= 0;
398 while ((token
= strsep(&o
, ",")) != NULL
) {
401 if (!strcmp(token
, "all")) {
402 opts
->subsys_bits
= (1 << CGROUP_SUBSYS_COUNT
) - 1;
403 } else if (!strcmp(token
, "noprefix")) {
404 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
406 struct cgroup_subsys
*ss
;
408 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
410 if (!strcmp(token
, ss
->name
)) {
411 set_bit(i
, &opts
->subsys_bits
);
415 if (i
== CGROUP_SUBSYS_COUNT
)
420 /* We can't have an empty hierarchy */
421 if (!opts
->subsys_bits
)
427 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
430 struct cgroupfs_root
*root
= sb
->s_fs_info
;
431 struct cgroup
*cont
= &root
->top_cgroup
;
432 struct cgroup_sb_opts opts
;
434 mutex_lock(&cont
->dentry
->d_inode
->i_mutex
);
435 mutex_lock(&cgroup_mutex
);
437 /* See what subsystems are wanted */
438 ret
= parse_cgroupfs_options(data
, &opts
);
442 /* Don't allow flags to change at remount */
443 if (opts
.flags
!= root
->flags
) {
448 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
450 /* (re)populate subsystem files */
452 cgroup_populate_dir(cont
);
455 mutex_unlock(&cgroup_mutex
);
456 mutex_unlock(&cont
->dentry
->d_inode
->i_mutex
);
460 static struct super_operations cgroup_ops
= {
461 .statfs
= simple_statfs
,
462 .drop_inode
= generic_delete_inode
,
463 .show_options
= cgroup_show_options
,
464 .remount_fs
= cgroup_remount
,
467 static void init_cgroup_root(struct cgroupfs_root
*root
)
469 struct cgroup
*cont
= &root
->top_cgroup
;
470 INIT_LIST_HEAD(&root
->subsys_list
);
471 INIT_LIST_HEAD(&root
->root_list
);
472 root
->number_of_cgroups
= 1;
474 cont
->top_cgroup
= cont
;
475 INIT_LIST_HEAD(&cont
->sibling
);
476 INIT_LIST_HEAD(&cont
->children
);
479 static int cgroup_test_super(struct super_block
*sb
, void *data
)
481 struct cgroupfs_root
*new = data
;
482 struct cgroupfs_root
*root
= sb
->s_fs_info
;
484 /* First check subsystems */
485 if (new->subsys_bits
!= root
->subsys_bits
)
488 /* Next check flags */
489 if (new->flags
!= root
->flags
)
495 static int cgroup_set_super(struct super_block
*sb
, void *data
)
498 struct cgroupfs_root
*root
= data
;
500 ret
= set_anon_super(sb
, NULL
);
504 sb
->s_fs_info
= root
;
507 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
508 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
509 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
510 sb
->s_op
= &cgroup_ops
;
515 static int cgroup_get_rootdir(struct super_block
*sb
)
517 struct inode
*inode
=
518 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
519 struct dentry
*dentry
;
524 inode
->i_op
= &simple_dir_inode_operations
;
525 inode
->i_fop
= &simple_dir_operations
;
526 inode
->i_op
= &cgroup_dir_inode_operations
;
527 /* directories start off with i_nlink == 2 (for "." entry) */
529 dentry
= d_alloc_root(inode
);
538 static int cgroup_get_sb(struct file_system_type
*fs_type
,
539 int flags
, const char *unused_dev_name
,
540 void *data
, struct vfsmount
*mnt
)
542 struct cgroup_sb_opts opts
;
544 struct super_block
*sb
;
545 struct cgroupfs_root
*root
;
547 /* First find the desired set of subsystems */
548 ret
= parse_cgroupfs_options(data
, &opts
);
552 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
556 init_cgroup_root(root
);
557 root
->subsys_bits
= opts
.subsys_bits
;
558 root
->flags
= opts
.flags
;
560 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, root
);
567 if (sb
->s_fs_info
!= root
) {
568 /* Reusing an existing superblock */
569 BUG_ON(sb
->s_root
== NULL
);
574 struct cgroup
*cont
= &root
->top_cgroup
;
576 BUG_ON(sb
->s_root
!= NULL
);
578 ret
= cgroup_get_rootdir(sb
);
582 mutex_lock(&cgroup_mutex
);
584 ret
= rebind_subsystems(root
, root
->subsys_bits
);
586 mutex_unlock(&cgroup_mutex
);
590 /* EBUSY should be the only error here */
593 list_add(&root
->root_list
, &roots
);
595 sb
->s_root
->d_fsdata
= &root
->top_cgroup
;
596 root
->top_cgroup
.dentry
= sb
->s_root
;
598 BUG_ON(!list_empty(&cont
->sibling
));
599 BUG_ON(!list_empty(&cont
->children
));
600 BUG_ON(root
->number_of_cgroups
!= 1);
603 * I believe that it's safe to nest i_mutex inside
604 * cgroup_mutex in this case, since no-one else can
605 * be accessing this directory yet. But we still need
606 * to teach lockdep that this is the case - currently
607 * a cgroupfs remount triggers a lockdep warning
609 mutex_lock(&cont
->dentry
->d_inode
->i_mutex
);
610 cgroup_populate_dir(cont
);
611 mutex_unlock(&cont
->dentry
->d_inode
->i_mutex
);
612 mutex_unlock(&cgroup_mutex
);
615 return simple_set_mnt(mnt
, sb
);
618 up_write(&sb
->s_umount
);
619 deactivate_super(sb
);
623 static void cgroup_kill_sb(struct super_block
*sb
) {
624 struct cgroupfs_root
*root
= sb
->s_fs_info
;
625 struct cgroup
*cont
= &root
->top_cgroup
;
630 BUG_ON(root
->number_of_cgroups
!= 1);
631 BUG_ON(!list_empty(&cont
->children
));
632 BUG_ON(!list_empty(&cont
->sibling
));
634 mutex_lock(&cgroup_mutex
);
636 /* Rebind all subsystems back to the default hierarchy */
637 ret
= rebind_subsystems(root
, 0);
638 /* Shouldn't be able to fail ... */
641 if (!list_empty(&root
->root_list
))
642 list_del(&root
->root_list
);
643 mutex_unlock(&cgroup_mutex
);
646 kill_litter_super(sb
);
649 static struct file_system_type cgroup_fs_type
= {
651 .get_sb
= cgroup_get_sb
,
652 .kill_sb
= cgroup_kill_sb
,
655 static inline struct cgroup
*__d_cont(struct dentry
*dentry
)
657 return dentry
->d_fsdata
;
660 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
662 return dentry
->d_fsdata
;
666 * Called with cgroup_mutex held. Writes path of cgroup into buf.
667 * Returns 0 on success, -errno on error.
669 int cgroup_path(const struct cgroup
*cont
, char *buf
, int buflen
)
673 if (cont
== dummytop
) {
675 * Inactive subsystems have no dentry for their root
682 start
= buf
+ buflen
;
686 int len
= cont
->dentry
->d_name
.len
;
687 if ((start
-= len
) < buf
)
688 return -ENAMETOOLONG
;
689 memcpy(start
, cont
->dentry
->d_name
.name
, len
);
696 return -ENAMETOOLONG
;
699 memmove(buf
, start
, buf
+ buflen
- start
);
703 /* The various types of files and directories in a cgroup file system */
705 enum cgroup_filetype
{
711 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
712 size_t nbytes
, loff_t
*ppos
)
714 struct cftype
*cft
= __d_cft(file
->f_dentry
);
715 struct cgroup
*cont
= __d_cont(file
->f_dentry
->d_parent
);
722 return cft
->write(cont
, cft
, file
, buf
, nbytes
, ppos
);
725 static ssize_t
cgroup_read_uint(struct cgroup
*cont
, struct cftype
*cft
,
727 char __user
*buf
, size_t nbytes
,
731 u64 val
= cft
->read_uint(cont
, cft
);
732 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
734 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
737 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
738 size_t nbytes
, loff_t
*ppos
)
740 struct cftype
*cft
= __d_cft(file
->f_dentry
);
741 struct cgroup
*cont
= __d_cont(file
->f_dentry
->d_parent
);
747 return cft
->read(cont
, cft
, file
, buf
, nbytes
, ppos
);
749 return cgroup_read_uint(cont
, cft
, file
, buf
, nbytes
, ppos
);
753 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
758 err
= generic_file_open(inode
, file
);
762 cft
= __d_cft(file
->f_dentry
);
766 err
= cft
->open(inode
, file
);
773 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
775 struct cftype
*cft
= __d_cft(file
->f_dentry
);
777 return cft
->release(inode
, file
);
782 * cgroup_rename - Only allow simple rename of directories in place.
784 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
785 struct inode
*new_dir
, struct dentry
*new_dentry
)
787 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
789 if (new_dentry
->d_inode
)
791 if (old_dir
!= new_dir
)
793 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
796 static struct file_operations cgroup_file_operations
= {
797 .read
= cgroup_file_read
,
798 .write
= cgroup_file_write
,
799 .llseek
= generic_file_llseek
,
800 .open
= cgroup_file_open
,
801 .release
= cgroup_file_release
,
804 static struct inode_operations cgroup_dir_inode_operations
= {
805 .lookup
= simple_lookup
,
806 .mkdir
= cgroup_mkdir
,
807 .rmdir
= cgroup_rmdir
,
808 .rename
= cgroup_rename
,
811 static int cgroup_create_file(struct dentry
*dentry
, int mode
,
812 struct super_block
*sb
)
814 static struct dentry_operations cgroup_dops
= {
815 .d_iput
= cgroup_diput
,
825 inode
= cgroup_new_inode(mode
, sb
);
830 inode
->i_op
= &cgroup_dir_inode_operations
;
831 inode
->i_fop
= &simple_dir_operations
;
833 /* start off with i_nlink == 2 (for "." entry) */
836 /* start with the directory inode held, so that we can
837 * populate it without racing with another mkdir */
838 mutex_lock(&inode
->i_mutex
);
839 } else if (S_ISREG(mode
)) {
841 inode
->i_fop
= &cgroup_file_operations
;
843 dentry
->d_op
= &cgroup_dops
;
844 d_instantiate(dentry
, inode
);
845 dget(dentry
); /* Extra count - pin the dentry in core */
850 * cgroup_create_dir - create a directory for an object.
851 * cont: the cgroup we create the directory for.
852 * It must have a valid ->parent field
853 * And we are going to fill its ->dentry field.
854 * dentry: dentry of the new container
855 * mode: mode to set on new directory.
857 static int cgroup_create_dir(struct cgroup
*cont
, struct dentry
*dentry
,
860 struct dentry
*parent
;
863 parent
= cont
->parent
->dentry
;
864 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cont
->root
->sb
);
866 dentry
->d_fsdata
= cont
;
867 inc_nlink(parent
->d_inode
);
868 cont
->dentry
= dentry
;
876 int cgroup_add_file(struct cgroup
*cont
,
877 struct cgroup_subsys
*subsys
,
878 const struct cftype
*cft
)
880 struct dentry
*dir
= cont
->dentry
;
881 struct dentry
*dentry
;
884 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
885 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cont
->root
->flags
)) {
886 strcpy(name
, subsys
->name
);
889 strcat(name
, cft
->name
);
890 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
891 dentry
= lookup_one_len(name
, dir
, strlen(name
));
892 if (!IS_ERR(dentry
)) {
893 error
= cgroup_create_file(dentry
, 0644 | S_IFREG
,
896 dentry
->d_fsdata
= (void *)cft
;
899 error
= PTR_ERR(dentry
);
903 int cgroup_add_files(struct cgroup
*cont
,
904 struct cgroup_subsys
*subsys
,
905 const struct cftype cft
[],
909 for (i
= 0; i
< count
; i
++) {
910 err
= cgroup_add_file(cont
, subsys
, &cft
[i
]);
917 static int cgroup_populate_dir(struct cgroup
*cont
)
920 struct cgroup_subsys
*ss
;
922 /* First clear out any existing files */
923 cgroup_clear_directory(cont
->dentry
);
925 for_each_subsys(cont
->root
, ss
) {
926 if (ss
->populate
&& (err
= ss
->populate(ss
, cont
)) < 0)
933 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
934 struct cgroup_subsys
*ss
,
938 atomic_set(&css
->refcnt
, 0);
940 if (cont
== dummytop
)
941 set_bit(CSS_ROOT
, &css
->flags
);
942 BUG_ON(cont
->subsys
[ss
->subsys_id
]);
943 cont
->subsys
[ss
->subsys_id
] = css
;
947 * cgroup_create - create a cgroup
948 * parent: cgroup that will be parent of the new cgroup.
949 * name: name of the new cgroup. Will be strcpy'ed.
950 * mode: mode to set on new inode
952 * Must be called with the mutex on the parent inode held
955 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
959 struct cgroupfs_root
*root
= parent
->root
;
961 struct cgroup_subsys
*ss
;
962 struct super_block
*sb
= root
->sb
;
964 cont
= kzalloc(sizeof(*cont
), GFP_KERNEL
);
968 /* Grab a reference on the superblock so the hierarchy doesn't
969 * get deleted on unmount if there are child cgroups. This
970 * can be done outside cgroup_mutex, since the sb can't
971 * disappear while someone has an open control file on the
973 atomic_inc(&sb
->s_active
);
975 mutex_lock(&cgroup_mutex
);
978 INIT_LIST_HEAD(&cont
->sibling
);
979 INIT_LIST_HEAD(&cont
->children
);
981 cont
->parent
= parent
;
982 cont
->root
= parent
->root
;
983 cont
->top_cgroup
= parent
->top_cgroup
;
985 for_each_subsys(root
, ss
) {
986 struct cgroup_subsys_state
*css
= ss
->create(ss
, cont
);
991 init_cgroup_css(css
, ss
, cont
);
994 list_add(&cont
->sibling
, &cont
->parent
->children
);
995 root
->number_of_cgroups
++;
997 err
= cgroup_create_dir(cont
, dentry
, mode
);
1001 /* The cgroup directory was pre-locked for us */
1002 BUG_ON(!mutex_is_locked(&cont
->dentry
->d_inode
->i_mutex
));
1004 err
= cgroup_populate_dir(cont
);
1005 /* If err < 0, we have a half-filled directory - oh well ;) */
1007 mutex_unlock(&cgroup_mutex
);
1008 mutex_unlock(&cont
->dentry
->d_inode
->i_mutex
);
1014 list_del(&cont
->sibling
);
1015 root
->number_of_cgroups
--;
1019 for_each_subsys(root
, ss
) {
1020 if (cont
->subsys
[ss
->subsys_id
])
1021 ss
->destroy(ss
, cont
);
1024 mutex_unlock(&cgroup_mutex
);
1026 /* Release the reference count that we took on the superblock */
1027 deactivate_super(sb
);
1033 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
1035 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
1037 /* the vfs holds inode->i_mutex already */
1038 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
1041 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
1043 struct cgroup
*cont
= dentry
->d_fsdata
;
1045 struct cgroup
*parent
;
1046 struct cgroup_subsys
*ss
;
1047 struct super_block
*sb
;
1048 struct cgroupfs_root
*root
;
1051 /* the vfs holds both inode->i_mutex already */
1053 mutex_lock(&cgroup_mutex
);
1054 if (atomic_read(&cont
->count
) != 0) {
1055 mutex_unlock(&cgroup_mutex
);
1058 if (!list_empty(&cont
->children
)) {
1059 mutex_unlock(&cgroup_mutex
);
1063 parent
= cont
->parent
;
1067 /* Check the reference count on each subsystem. Since we
1068 * already established that there are no tasks in the
1069 * cgroup, if the css refcount is also 0, then there should
1070 * be no outstanding references, so the subsystem is safe to
1072 for_each_subsys(root
, ss
) {
1073 struct cgroup_subsys_state
*css
;
1074 css
= cont
->subsys
[ss
->subsys_id
];
1075 if (atomic_read(&css
->refcnt
)) {
1081 mutex_unlock(&cgroup_mutex
);
1085 for_each_subsys(root
, ss
) {
1086 if (cont
->subsys
[ss
->subsys_id
])
1087 ss
->destroy(ss
, cont
);
1090 set_bit(CONT_REMOVED
, &cont
->flags
);
1091 /* delete my sibling from parent->children */
1092 list_del(&cont
->sibling
);
1093 spin_lock(&cont
->dentry
->d_lock
);
1094 d
= dget(cont
->dentry
);
1095 cont
->dentry
= NULL
;
1096 spin_unlock(&d
->d_lock
);
1098 cgroup_d_remove_dir(d
);
1100 root
->number_of_cgroups
--;
1102 mutex_unlock(&cgroup_mutex
);
1103 /* Drop the active superblock reference that we took when we
1104 * created the cgroup */
1105 deactivate_super(sb
);
1109 static void cgroup_init_subsys(struct cgroup_subsys
*ss
)
1111 struct task_struct
*g
, *p
;
1112 struct cgroup_subsys_state
*css
;
1113 printk(KERN_ERR
"Initializing cgroup subsys %s\n", ss
->name
);
1115 /* Create the top cgroup state for this subsystem */
1116 ss
->root
= &rootnode
;
1117 css
= ss
->create(ss
, dummytop
);
1118 /* We don't handle early failures gracefully */
1119 BUG_ON(IS_ERR(css
));
1120 init_cgroup_css(css
, ss
, dummytop
);
1122 /* Update all tasks to contain a subsys pointer to this state
1123 * - since the subsystem is newly registered, all tasks are in
1124 * the subsystem's top cgroup. */
1126 /* If this subsystem requested that it be notified with fork
1127 * events, we should send it one now for every process in the
1130 read_lock(&tasklist_lock
);
1131 init_task
.cgroups
.subsys
[ss
->subsys_id
] = css
;
1133 ss
->fork(ss
, &init_task
);
1135 do_each_thread(g
, p
) {
1136 printk(KERN_INFO
"Setting task %p css to %p (%d)\n", css
, p
, p
->pid
);
1137 p
->cgroups
.subsys
[ss
->subsys_id
] = css
;
1140 } while_each_thread(g
, p
);
1141 read_unlock(&tasklist_lock
);
1143 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
1149 * cgroup_init_early - initialize cgroups at system boot, and
1150 * initialize any subsystems that request early init.
1152 int __init
cgroup_init_early(void)
1155 init_cgroup_root(&rootnode
);
1156 list_add(&rootnode
.root_list
, &roots
);
1158 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1159 struct cgroup_subsys
*ss
= subsys
[i
];
1162 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
1163 BUG_ON(!ss
->create
);
1164 BUG_ON(!ss
->destroy
);
1165 if (ss
->subsys_id
!= i
) {
1166 printk(KERN_ERR
"Subsys %s id == %d\n",
1167 ss
->name
, ss
->subsys_id
);
1172 cgroup_init_subsys(ss
);
1178 * cgroup_init - register cgroup filesystem and /proc file, and
1179 * initialize any subsystems that didn't request early init.
1181 int __init
cgroup_init(void)
1186 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
1187 struct cgroup_subsys
*ss
= subsys
[i
];
1188 if (!ss
->early_init
)
1189 cgroup_init_subsys(ss
);
1192 err
= register_filesystem(&cgroup_fs_type
);