2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Copyright notices from the original cpuset code:
8 * --------------------------------------------------
9 * Copyright (C) 2003 BULL SA.
10 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
12 * Portions derived from Patrick Mochel's sysfs code.
13 * sysfs is Copyright (c) 2001-3 Patrick Mochel
15 * 2003-10-10 Written by Simon Derr.
16 * 2003-10-22 Updates by Stephen Hemminger.
17 * 2004 May-July Rework by Paul Jackson.
18 * ---------------------------------------------------
20 * This file is subject to the terms and conditions of the GNU General Public
21 * License. See the file COPYING in the main directory of the Linux
22 * distribution for more details.
25 #include <linux/cgroup.h>
26 #include <linux/errno.h>
28 #include <linux/kernel.h>
29 #include <linux/list.h>
31 #include <linux/mutex.h>
32 #include <linux/mount.h>
33 #include <linux/pagemap.h>
34 #include <linux/proc_fs.h>
35 #include <linux/rcupdate.h>
36 #include <linux/sched.h>
37 #include <linux/backing-dev.h>
38 #include <linux/seq_file.h>
39 #include <linux/slab.h>
40 #include <linux/magic.h>
41 #include <linux/spinlock.h>
42 #include <linux/string.h>
43 #include <linux/sort.h>
44 #include <linux/kmod.h>
45 #include <linux/delayacct.h>
46 #include <linux/cgroupstats.h>
48 #include <asm/atomic.h>
50 static DEFINE_MUTEX(cgroup_mutex
);
52 /* Generate an array of cgroup subsystem pointers */
53 #define SUBSYS(_x) &_x ## _subsys,
55 static struct cgroup_subsys
*subsys
[] = {
56 #include <linux/cgroup_subsys.h>
60 * A cgroupfs_root represents the root of a cgroup hierarchy,
61 * and may be associated with a superblock to form an active
64 struct cgroupfs_root
{
65 struct super_block
*sb
;
68 * The bitmask of subsystems intended to be attached to this
71 unsigned long subsys_bits
;
73 /* The bitmask of subsystems currently attached to this hierarchy */
74 unsigned long actual_subsys_bits
;
76 /* A list running through the attached subsystems */
77 struct list_head subsys_list
;
79 /* The root cgroup for this hierarchy */
80 struct cgroup top_cgroup
;
82 /* Tracks how many cgroups are currently defined in hierarchy.*/
83 int number_of_cgroups
;
85 /* A list running through the mounted hierarchies */
86 struct list_head root_list
;
88 /* Hierarchy-specific flags */
91 /* The path to use for release notifications. No locking
92 * between setting and use - so if userspace updates this
93 * while child cgroups exist, you could miss a
94 * notification. We ensure that it's always a valid
95 * NUL-terminated string */
96 char release_agent_path
[PATH_MAX
];
101 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
102 * subsystems that are otherwise unattached - it never has more than a
103 * single cgroup, and all tasks are part of that cgroup.
105 static struct cgroupfs_root rootnode
;
107 /* The list of hierarchy roots */
109 static LIST_HEAD(roots
);
110 static int root_count
;
112 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
113 #define dummytop (&rootnode.top_cgroup)
115 /* This flag indicates whether tasks in the fork and exit paths should
116 * take callback_mutex and check for fork/exit handlers to call. This
117 * avoids us having to do extra work in the fork/exit path if none of the
118 * subsystems need to be called.
120 static int need_forkexit_callback
;
122 /* bits in struct cgroup flags field */
124 /* Control Group is dead */
126 /* Control Group has previously had a child cgroup or a task,
127 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
129 /* Control Group requires release notifications to userspace */
130 CGRP_NOTIFY_ON_RELEASE
,
133 /* convenient tests for these bits */
134 inline int cgroup_is_removed(const struct cgroup
*cgrp
)
136 return test_bit(CGRP_REMOVED
, &cgrp
->flags
);
139 /* bits in struct cgroupfs_root flags field */
141 ROOT_NOPREFIX
, /* mounted subsystems have no named prefix */
144 inline int cgroup_is_releasable(const struct cgroup
*cgrp
)
147 (1 << CGRP_RELEASABLE
) |
148 (1 << CGRP_NOTIFY_ON_RELEASE
);
149 return (cgrp
->flags
& bits
) == bits
;
152 inline int notify_on_release(const struct cgroup
*cgrp
)
154 return test_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
158 * for_each_subsys() allows you to iterate on each subsystem attached to
159 * an active hierarchy
161 #define for_each_subsys(_root, _ss) \
162 list_for_each_entry(_ss, &_root->subsys_list, sibling)
164 /* for_each_root() allows you to iterate across the active hierarchies */
165 #define for_each_root(_root) \
166 list_for_each_entry(_root, &roots, root_list)
168 /* the list of cgroups eligible for automatic release. Protected by
169 * release_list_lock */
170 static LIST_HEAD(release_list
);
171 static DEFINE_SPINLOCK(release_list_lock
);
172 static void cgroup_release_agent(struct work_struct
*work
);
173 static DECLARE_WORK(release_agent_work
, cgroup_release_agent
);
174 static void check_for_release(struct cgroup
*cgrp
);
176 /* Link structure for associating css_set objects with cgroups */
177 struct cg_cgroup_link
{
179 * List running through cg_cgroup_links associated with a
180 * cgroup, anchored on cgroup->css_sets
182 struct list_head cgrp_link_list
;
184 * List running through cg_cgroup_links pointing at a
185 * single css_set object, anchored on css_set->cg_links
187 struct list_head cg_link_list
;
191 /* The default css_set - used by init and its children prior to any
192 * hierarchies being mounted. It contains a pointer to the root state
193 * for each subsystem. Also used to anchor the list of css_sets. Not
194 * reference-counted, to improve performance when child cgroups
195 * haven't been created.
198 static struct css_set init_css_set
;
199 static struct cg_cgroup_link init_css_set_link
;
201 /* css_set_lock protects the list of css_set objects, and the
202 * chain of tasks off each css_set. Nests outside task->alloc_lock
203 * due to cgroup_iter_start() */
204 static DEFINE_RWLOCK(css_set_lock
);
205 static int css_set_count
;
207 /* We don't maintain the lists running through each css_set to its
208 * task until after the first call to cgroup_iter_start(). This
209 * reduces the fork()/exit() overhead for people who have cgroups
210 * compiled into their kernel but not actually in use */
211 static int use_task_css_set_links
;
213 /* When we create or destroy a css_set, the operation simply
214 * takes/releases a reference count on all the cgroups referenced
215 * by subsystems in this css_set. This can end up multiple-counting
216 * some cgroups, but that's OK - the ref-count is just a
217 * busy/not-busy indicator; ensuring that we only count each cgroup
218 * once would require taking a global lock to ensure that no
219 * subsystems moved between hierarchies while we were doing so.
221 * Possible TODO: decide at boot time based on the number of
222 * registered subsystems and the number of CPUs or NUMA nodes whether
223 * it's better for performance to ref-count every subsystem, or to
224 * take a global lock and only add one ref count to each hierarchy.
228 * unlink a css_set from the list and free it
230 static void unlink_css_set(struct css_set
*cg
)
232 write_lock(&css_set_lock
);
235 while (!list_empty(&cg
->cg_links
)) {
236 struct cg_cgroup_link
*link
;
237 link
= list_entry(cg
->cg_links
.next
,
238 struct cg_cgroup_link
, cg_link_list
);
239 list_del(&link
->cg_link_list
);
240 list_del(&link
->cgrp_link_list
);
243 write_unlock(&css_set_lock
);
246 static void __release_css_set(struct kref
*k
, int taskexit
)
249 struct css_set
*cg
= container_of(k
, struct css_set
, ref
);
254 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
255 struct cgroup
*cgrp
= cg
->subsys
[i
]->cgroup
;
256 if (atomic_dec_and_test(&cgrp
->count
) &&
257 notify_on_release(cgrp
)) {
259 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
260 check_for_release(cgrp
);
267 static void release_css_set(struct kref
*k
)
269 __release_css_set(k
, 0);
272 static void release_css_set_taskexit(struct kref
*k
)
274 __release_css_set(k
, 1);
278 * refcounted get/put for css_set objects
280 static inline void get_css_set(struct css_set
*cg
)
285 static inline void put_css_set(struct css_set
*cg
)
287 kref_put(&cg
->ref
, release_css_set
);
290 static inline void put_css_set_taskexit(struct css_set
*cg
)
292 kref_put(&cg
->ref
, release_css_set_taskexit
);
296 * find_existing_css_set() is a helper for
297 * find_css_set(), and checks to see whether an existing
298 * css_set is suitable. This currently walks a linked-list for
299 * simplicity; a later patch will use a hash table for better
302 * oldcg: the cgroup group that we're using before the cgroup
305 * cgrp: the cgroup that we're moving into
307 * template: location in which to build the desired set of subsystem
308 * state objects for the new cgroup group
311 static struct css_set
*find_existing_css_set(
312 struct css_set
*oldcg
,
314 struct cgroup_subsys_state
*template[])
317 struct cgroupfs_root
*root
= cgrp
->root
;
318 struct list_head
*l
= &init_css_set
.list
;
320 /* Built the set of subsystem state objects that we want to
321 * see in the new css_set */
322 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
323 if (root
->subsys_bits
& (1ull << i
)) {
324 /* Subsystem is in this hierarchy. So we want
325 * the subsystem state from the new
327 template[i
] = cgrp
->subsys
[i
];
329 /* Subsystem is not in this hierarchy, so we
330 * don't want to change the subsystem state */
331 template[i
] = oldcg
->subsys
[i
];
335 /* Look through existing cgroup groups to find one to reuse */
338 list_entry(l
, struct css_set
, list
);
340 if (!memcmp(template, cg
->subsys
, sizeof(cg
->subsys
))) {
341 /* All subsystems matched */
344 /* Try the next cgroup group */
346 } while (l
!= &init_css_set
.list
);
348 /* No existing cgroup group matched */
353 * allocate_cg_links() allocates "count" cg_cgroup_link structures
354 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
355 * success or a negative error
358 static int allocate_cg_links(int count
, struct list_head
*tmp
)
360 struct cg_cgroup_link
*link
;
363 for (i
= 0; i
< count
; i
++) {
364 link
= kmalloc(sizeof(*link
), GFP_KERNEL
);
366 while (!list_empty(tmp
)) {
367 link
= list_entry(tmp
->next
,
368 struct cg_cgroup_link
,
370 list_del(&link
->cgrp_link_list
);
375 list_add(&link
->cgrp_link_list
, tmp
);
380 static void free_cg_links(struct list_head
*tmp
)
382 while (!list_empty(tmp
)) {
383 struct cg_cgroup_link
*link
;
384 link
= list_entry(tmp
->next
,
385 struct cg_cgroup_link
,
387 list_del(&link
->cgrp_link_list
);
393 * find_css_set() takes an existing cgroup group and a
394 * cgroup object, and returns a css_set object that's
395 * equivalent to the old group, but with the given cgroup
396 * substituted into the appropriate hierarchy. Must be called with
400 static struct css_set
*find_css_set(
401 struct css_set
*oldcg
, struct cgroup
*cgrp
)
404 struct cgroup_subsys_state
*template[CGROUP_SUBSYS_COUNT
];
407 struct list_head tmp_cg_links
;
408 struct cg_cgroup_link
*link
;
410 /* First see if we already have a cgroup group that matches
412 write_lock(&css_set_lock
);
413 res
= find_existing_css_set(oldcg
, cgrp
, template);
416 write_unlock(&css_set_lock
);
421 res
= kmalloc(sizeof(*res
), GFP_KERNEL
);
425 /* Allocate all the cg_cgroup_link objects that we'll need */
426 if (allocate_cg_links(root_count
, &tmp_cg_links
) < 0) {
431 kref_init(&res
->ref
);
432 INIT_LIST_HEAD(&res
->cg_links
);
433 INIT_LIST_HEAD(&res
->tasks
);
435 /* Copy the set of subsystem state objects generated in
436 * find_existing_css_set() */
437 memcpy(res
->subsys
, template, sizeof(res
->subsys
));
439 write_lock(&css_set_lock
);
440 /* Add reference counts and links from the new css_set. */
441 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
442 struct cgroup
*cgrp
= res
->subsys
[i
]->cgroup
;
443 struct cgroup_subsys
*ss
= subsys
[i
];
444 atomic_inc(&cgrp
->count
);
446 * We want to add a link once per cgroup, so we
447 * only do it for the first subsystem in each
450 if (ss
->root
->subsys_list
.next
== &ss
->sibling
) {
451 BUG_ON(list_empty(&tmp_cg_links
));
452 link
= list_entry(tmp_cg_links
.next
,
453 struct cg_cgroup_link
,
455 list_del(&link
->cgrp_link_list
);
456 list_add(&link
->cgrp_link_list
, &cgrp
->css_sets
);
458 list_add(&link
->cg_link_list
, &res
->cg_links
);
461 if (list_empty(&rootnode
.subsys_list
)) {
462 link
= list_entry(tmp_cg_links
.next
,
463 struct cg_cgroup_link
,
465 list_del(&link
->cgrp_link_list
);
466 list_add(&link
->cgrp_link_list
, &dummytop
->css_sets
);
468 list_add(&link
->cg_link_list
, &res
->cg_links
);
471 BUG_ON(!list_empty(&tmp_cg_links
));
473 /* Link this cgroup group into the list */
474 list_add(&res
->list
, &init_css_set
.list
);
476 INIT_LIST_HEAD(&res
->tasks
);
477 write_unlock(&css_set_lock
);
483 * There is one global cgroup mutex. We also require taking
484 * task_lock() when dereferencing a task's cgroup subsys pointers.
485 * See "The task_lock() exception", at the end of this comment.
487 * A task must hold cgroup_mutex to modify cgroups.
489 * Any task can increment and decrement the count field without lock.
490 * So in general, code holding cgroup_mutex can't rely on the count
491 * field not changing. However, if the count goes to zero, then only
492 * attach_task() can increment it again. Because a count of zero
493 * means that no tasks are currently attached, therefore there is no
494 * way a task attached to that cgroup can fork (the other way to
495 * increment the count). So code holding cgroup_mutex can safely
496 * assume that if the count is zero, it will stay zero. Similarly, if
497 * a task holds cgroup_mutex on a cgroup with zero count, it
498 * knows that the cgroup won't be removed, as cgroup_rmdir()
501 * The cgroup_common_file_write handler for operations that modify
502 * the cgroup hierarchy holds cgroup_mutex across the entire operation,
503 * single threading all such cgroup modifications across the system.
505 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
506 * (usually) take cgroup_mutex. These are the two most performance
507 * critical pieces of code here. The exception occurs on cgroup_exit(),
508 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
509 * is taken, and if the cgroup count is zero, a usermode call made
510 * to /sbin/cgroup_release_agent with the name of the cgroup (path
511 * relative to the root of cgroup file system) as the argument.
513 * A cgroup can only be deleted if both its 'count' of using tasks
514 * is zero, and its list of 'children' cgroups is empty. Since all
515 * tasks in the system use _some_ cgroup, and since there is always at
516 * least one task in the system (init, pid == 1), therefore, top_cgroup
517 * always has either children cgroups and/or using tasks. So we don't
518 * need a special hack to ensure that top_cgroup cannot be deleted.
520 * The task_lock() exception
522 * The need for this exception arises from the action of
523 * attach_task(), which overwrites one tasks cgroup pointer with
524 * another. It does so using cgroup_mutexe, however there are
525 * several performance critical places that need to reference
526 * task->cgroup without the expense of grabbing a system global
527 * mutex. Therefore except as noted below, when dereferencing or, as
528 * in attach_task(), modifying a task'ss cgroup pointer we use
529 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
530 * the task_struct routinely used for such matters.
532 * P.S. One more locking exception. RCU is used to guard the
533 * update of a tasks cgroup pointer by attach_task()
537 * cgroup_lock - lock out any changes to cgroup structures
541 void cgroup_lock(void)
543 mutex_lock(&cgroup_mutex
);
547 * cgroup_unlock - release lock on cgroup changes
549 * 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_DIRTY
| BDI_CAP_NO_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
;
589 static void cgroup_diput(struct dentry
*dentry
, struct inode
*inode
)
591 /* is dentry a directory ? if so, kfree() associated cgroup */
592 if (S_ISDIR(inode
->i_mode
)) {
593 struct cgroup
*cgrp
= dentry
->d_fsdata
;
594 BUG_ON(!(cgroup_is_removed(cgrp
)));
595 /* It's possible for external users to be holding css
596 * reference counts on a cgroup; css_put() needs to
597 * be able to access the cgroup after decrementing
598 * the reference count in order to know if it needs to
599 * queue the cgroup to be handled by the release
607 static void remove_dir(struct dentry
*d
)
609 struct dentry
*parent
= dget(d
->d_parent
);
612 simple_rmdir(parent
->d_inode
, d
);
616 static void cgroup_clear_directory(struct dentry
*dentry
)
618 struct list_head
*node
;
620 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
621 spin_lock(&dcache_lock
);
622 node
= dentry
->d_subdirs
.next
;
623 while (node
!= &dentry
->d_subdirs
) {
624 struct dentry
*d
= list_entry(node
, struct dentry
, d_u
.d_child
);
627 /* This should never be called on a cgroup
628 * directory with child cgroups */
629 BUG_ON(d
->d_inode
->i_mode
& S_IFDIR
);
631 spin_unlock(&dcache_lock
);
633 simple_unlink(dentry
->d_inode
, d
);
635 spin_lock(&dcache_lock
);
637 node
= dentry
->d_subdirs
.next
;
639 spin_unlock(&dcache_lock
);
643 * NOTE : the dentry must have been dget()'ed
645 static void cgroup_d_remove_dir(struct dentry
*dentry
)
647 cgroup_clear_directory(dentry
);
649 spin_lock(&dcache_lock
);
650 list_del_init(&dentry
->d_u
.d_child
);
651 spin_unlock(&dcache_lock
);
655 static int rebind_subsystems(struct cgroupfs_root
*root
,
656 unsigned long final_bits
)
658 unsigned long added_bits
, removed_bits
;
659 struct cgroup
*cgrp
= &root
->top_cgroup
;
662 removed_bits
= root
->actual_subsys_bits
& ~final_bits
;
663 added_bits
= final_bits
& ~root
->actual_subsys_bits
;
664 /* Check that any added subsystems are currently free */
665 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
666 unsigned long long bit
= 1ull << i
;
667 struct cgroup_subsys
*ss
= subsys
[i
];
668 if (!(bit
& added_bits
))
670 if (ss
->root
!= &rootnode
) {
671 /* Subsystem isn't free */
676 /* Currently we don't handle adding/removing subsystems when
677 * any child cgroups exist. This is theoretically supportable
678 * but involves complex error handling, so it's being left until
680 if (!list_empty(&cgrp
->children
))
683 /* Process each subsystem */
684 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
685 struct cgroup_subsys
*ss
= subsys
[i
];
686 unsigned long bit
= 1UL << i
;
687 if (bit
& added_bits
) {
688 /* We're binding this subsystem to this hierarchy */
689 BUG_ON(cgrp
->subsys
[i
]);
690 BUG_ON(!dummytop
->subsys
[i
]);
691 BUG_ON(dummytop
->subsys
[i
]->cgroup
!= dummytop
);
692 cgrp
->subsys
[i
] = dummytop
->subsys
[i
];
693 cgrp
->subsys
[i
]->cgroup
= cgrp
;
694 list_add(&ss
->sibling
, &root
->subsys_list
);
695 rcu_assign_pointer(ss
->root
, root
);
699 } else if (bit
& removed_bits
) {
700 /* We're removing this subsystem */
701 BUG_ON(cgrp
->subsys
[i
] != dummytop
->subsys
[i
]);
702 BUG_ON(cgrp
->subsys
[i
]->cgroup
!= cgrp
);
704 ss
->bind(ss
, dummytop
);
705 dummytop
->subsys
[i
]->cgroup
= dummytop
;
706 cgrp
->subsys
[i
] = NULL
;
707 rcu_assign_pointer(subsys
[i
]->root
, &rootnode
);
708 list_del(&ss
->sibling
);
709 } else if (bit
& final_bits
) {
710 /* Subsystem state should already exist */
711 BUG_ON(!cgrp
->subsys
[i
]);
713 /* Subsystem state shouldn't exist */
714 BUG_ON(cgrp
->subsys
[i
]);
717 root
->subsys_bits
= root
->actual_subsys_bits
= final_bits
;
723 static int cgroup_show_options(struct seq_file
*seq
, struct vfsmount
*vfs
)
725 struct cgroupfs_root
*root
= vfs
->mnt_sb
->s_fs_info
;
726 struct cgroup_subsys
*ss
;
728 mutex_lock(&cgroup_mutex
);
729 for_each_subsys(root
, ss
)
730 seq_printf(seq
, ",%s", ss
->name
);
731 if (test_bit(ROOT_NOPREFIX
, &root
->flags
))
732 seq_puts(seq
, ",noprefix");
733 if (strlen(root
->release_agent_path
))
734 seq_printf(seq
, ",release_agent=%s", root
->release_agent_path
);
735 mutex_unlock(&cgroup_mutex
);
739 struct cgroup_sb_opts
{
740 unsigned long subsys_bits
;
745 /* Convert a hierarchy specifier into a bitmask of subsystems and
747 static int parse_cgroupfs_options(char *data
,
748 struct cgroup_sb_opts
*opts
)
750 char *token
, *o
= data
?: "all";
752 opts
->subsys_bits
= 0;
754 opts
->release_agent
= NULL
;
756 while ((token
= strsep(&o
, ",")) != NULL
) {
759 if (!strcmp(token
, "all")) {
760 opts
->subsys_bits
= (1 << CGROUP_SUBSYS_COUNT
) - 1;
761 } else if (!strcmp(token
, "noprefix")) {
762 set_bit(ROOT_NOPREFIX
, &opts
->flags
);
763 } else if (!strncmp(token
, "release_agent=", 14)) {
764 /* Specifying two release agents is forbidden */
765 if (opts
->release_agent
)
767 opts
->release_agent
= kzalloc(PATH_MAX
, GFP_KERNEL
);
768 if (!opts
->release_agent
)
770 strncpy(opts
->release_agent
, token
+ 14, PATH_MAX
- 1);
771 opts
->release_agent
[PATH_MAX
- 1] = 0;
773 struct cgroup_subsys
*ss
;
775 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
777 if (!strcmp(token
, ss
->name
)) {
778 set_bit(i
, &opts
->subsys_bits
);
782 if (i
== CGROUP_SUBSYS_COUNT
)
787 /* We can't have an empty hierarchy */
788 if (!opts
->subsys_bits
)
794 static int cgroup_remount(struct super_block
*sb
, int *flags
, char *data
)
797 struct cgroupfs_root
*root
= sb
->s_fs_info
;
798 struct cgroup
*cgrp
= &root
->top_cgroup
;
799 struct cgroup_sb_opts opts
;
801 mutex_lock(&cgrp
->dentry
->d_inode
->i_mutex
);
802 mutex_lock(&cgroup_mutex
);
804 /* See what subsystems are wanted */
805 ret
= parse_cgroupfs_options(data
, &opts
);
809 /* Don't allow flags to change at remount */
810 if (opts
.flags
!= root
->flags
) {
815 ret
= rebind_subsystems(root
, opts
.subsys_bits
);
817 /* (re)populate subsystem files */
819 cgroup_populate_dir(cgrp
);
821 if (opts
.release_agent
)
822 strcpy(root
->release_agent_path
, opts
.release_agent
);
824 if (opts
.release_agent
)
825 kfree(opts
.release_agent
);
826 mutex_unlock(&cgroup_mutex
);
827 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
831 static struct super_operations cgroup_ops
= {
832 .statfs
= simple_statfs
,
833 .drop_inode
= generic_delete_inode
,
834 .show_options
= cgroup_show_options
,
835 .remount_fs
= cgroup_remount
,
838 static void init_cgroup_root(struct cgroupfs_root
*root
)
840 struct cgroup
*cgrp
= &root
->top_cgroup
;
841 INIT_LIST_HEAD(&root
->subsys_list
);
842 INIT_LIST_HEAD(&root
->root_list
);
843 root
->number_of_cgroups
= 1;
845 cgrp
->top_cgroup
= cgrp
;
846 INIT_LIST_HEAD(&cgrp
->sibling
);
847 INIT_LIST_HEAD(&cgrp
->children
);
848 INIT_LIST_HEAD(&cgrp
->css_sets
);
849 INIT_LIST_HEAD(&cgrp
->release_list
);
852 static int cgroup_test_super(struct super_block
*sb
, void *data
)
854 struct cgroupfs_root
*new = data
;
855 struct cgroupfs_root
*root
= sb
->s_fs_info
;
857 /* First check subsystems */
858 if (new->subsys_bits
!= root
->subsys_bits
)
861 /* Next check flags */
862 if (new->flags
!= root
->flags
)
868 static int cgroup_set_super(struct super_block
*sb
, void *data
)
871 struct cgroupfs_root
*root
= data
;
873 ret
= set_anon_super(sb
, NULL
);
877 sb
->s_fs_info
= root
;
880 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
881 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
882 sb
->s_magic
= CGROUP_SUPER_MAGIC
;
883 sb
->s_op
= &cgroup_ops
;
888 static int cgroup_get_rootdir(struct super_block
*sb
)
890 struct inode
*inode
=
891 cgroup_new_inode(S_IFDIR
| S_IRUGO
| S_IXUGO
| S_IWUSR
, sb
);
892 struct dentry
*dentry
;
897 inode
->i_op
= &simple_dir_inode_operations
;
898 inode
->i_fop
= &simple_dir_operations
;
899 inode
->i_op
= &cgroup_dir_inode_operations
;
900 /* directories start off with i_nlink == 2 (for "." entry) */
902 dentry
= d_alloc_root(inode
);
911 static int cgroup_get_sb(struct file_system_type
*fs_type
,
912 int flags
, const char *unused_dev_name
,
913 void *data
, struct vfsmount
*mnt
)
915 struct cgroup_sb_opts opts
;
917 struct super_block
*sb
;
918 struct cgroupfs_root
*root
;
919 struct list_head tmp_cg_links
, *l
;
920 INIT_LIST_HEAD(&tmp_cg_links
);
922 /* First find the desired set of subsystems */
923 ret
= parse_cgroupfs_options(data
, &opts
);
925 if (opts
.release_agent
)
926 kfree(opts
.release_agent
);
930 root
= kzalloc(sizeof(*root
), GFP_KERNEL
);
934 init_cgroup_root(root
);
935 root
->subsys_bits
= opts
.subsys_bits
;
936 root
->flags
= opts
.flags
;
937 if (opts
.release_agent
) {
938 strcpy(root
->release_agent_path
, opts
.release_agent
);
939 kfree(opts
.release_agent
);
942 sb
= sget(fs_type
, cgroup_test_super
, cgroup_set_super
, root
);
949 if (sb
->s_fs_info
!= root
) {
950 /* Reusing an existing superblock */
951 BUG_ON(sb
->s_root
== NULL
);
956 struct cgroup
*cgrp
= &root
->top_cgroup
;
959 BUG_ON(sb
->s_root
!= NULL
);
961 ret
= cgroup_get_rootdir(sb
);
964 inode
= sb
->s_root
->d_inode
;
966 mutex_lock(&inode
->i_mutex
);
967 mutex_lock(&cgroup_mutex
);
970 * We're accessing css_set_count without locking
971 * css_set_lock here, but that's OK - it can only be
972 * increased by someone holding cgroup_lock, and
973 * that's us. The worst that can happen is that we
974 * have some link structures left over
976 ret
= allocate_cg_links(css_set_count
, &tmp_cg_links
);
978 mutex_unlock(&cgroup_mutex
);
979 mutex_unlock(&inode
->i_mutex
);
983 ret
= rebind_subsystems(root
, root
->subsys_bits
);
985 mutex_unlock(&cgroup_mutex
);
986 mutex_unlock(&inode
->i_mutex
);
990 /* EBUSY should be the only error here */
993 list_add(&root
->root_list
, &roots
);
996 sb
->s_root
->d_fsdata
= &root
->top_cgroup
;
997 root
->top_cgroup
.dentry
= sb
->s_root
;
999 /* Link the top cgroup in this hierarchy into all
1000 * the css_set objects */
1001 write_lock(&css_set_lock
);
1002 l
= &init_css_set
.list
;
1005 struct cg_cgroup_link
*link
;
1006 cg
= list_entry(l
, struct css_set
, list
);
1007 BUG_ON(list_empty(&tmp_cg_links
));
1008 link
= list_entry(tmp_cg_links
.next
,
1009 struct cg_cgroup_link
,
1011 list_del(&link
->cgrp_link_list
);
1013 list_add(&link
->cgrp_link_list
,
1014 &root
->top_cgroup
.css_sets
);
1015 list_add(&link
->cg_link_list
, &cg
->cg_links
);
1017 } while (l
!= &init_css_set
.list
);
1018 write_unlock(&css_set_lock
);
1020 free_cg_links(&tmp_cg_links
);
1022 BUG_ON(!list_empty(&cgrp
->sibling
));
1023 BUG_ON(!list_empty(&cgrp
->children
));
1024 BUG_ON(root
->number_of_cgroups
!= 1);
1026 cgroup_populate_dir(cgrp
);
1027 mutex_unlock(&inode
->i_mutex
);
1028 mutex_unlock(&cgroup_mutex
);
1031 return simple_set_mnt(mnt
, sb
);
1034 up_write(&sb
->s_umount
);
1035 deactivate_super(sb
);
1036 free_cg_links(&tmp_cg_links
);
1040 static void cgroup_kill_sb(struct super_block
*sb
) {
1041 struct cgroupfs_root
*root
= sb
->s_fs_info
;
1042 struct cgroup
*cgrp
= &root
->top_cgroup
;
1047 BUG_ON(root
->number_of_cgroups
!= 1);
1048 BUG_ON(!list_empty(&cgrp
->children
));
1049 BUG_ON(!list_empty(&cgrp
->sibling
));
1051 mutex_lock(&cgroup_mutex
);
1053 /* Rebind all subsystems back to the default hierarchy */
1054 ret
= rebind_subsystems(root
, 0);
1055 /* Shouldn't be able to fail ... */
1059 * Release all the links from css_sets to this hierarchy's
1062 write_lock(&css_set_lock
);
1063 while (!list_empty(&cgrp
->css_sets
)) {
1064 struct cg_cgroup_link
*link
;
1065 link
= list_entry(cgrp
->css_sets
.next
,
1066 struct cg_cgroup_link
, cgrp_link_list
);
1067 list_del(&link
->cg_link_list
);
1068 list_del(&link
->cgrp_link_list
);
1071 write_unlock(&css_set_lock
);
1073 if (!list_empty(&root
->root_list
)) {
1074 list_del(&root
->root_list
);
1077 mutex_unlock(&cgroup_mutex
);
1080 kill_litter_super(sb
);
1083 static struct file_system_type cgroup_fs_type
= {
1085 .get_sb
= cgroup_get_sb
,
1086 .kill_sb
= cgroup_kill_sb
,
1089 static inline struct cgroup
*__d_cgrp(struct dentry
*dentry
)
1091 return dentry
->d_fsdata
;
1094 static inline struct cftype
*__d_cft(struct dentry
*dentry
)
1096 return dentry
->d_fsdata
;
1100 * Called with cgroup_mutex held. Writes path of cgroup into buf.
1101 * Returns 0 on success, -errno on error.
1103 int cgroup_path(const struct cgroup
*cgrp
, char *buf
, int buflen
)
1107 if (cgrp
== dummytop
) {
1109 * Inactive subsystems have no dentry for their root
1116 start
= buf
+ buflen
;
1120 int len
= cgrp
->dentry
->d_name
.len
;
1121 if ((start
-= len
) < buf
)
1122 return -ENAMETOOLONG
;
1123 memcpy(start
, cgrp
->dentry
->d_name
.name
, len
);
1124 cgrp
= cgrp
->parent
;
1130 return -ENAMETOOLONG
;
1133 memmove(buf
, start
, buf
+ buflen
- start
);
1138 * Return the first subsystem attached to a cgroup's hierarchy, and
1142 static void get_first_subsys(const struct cgroup
*cgrp
,
1143 struct cgroup_subsys_state
**css
, int *subsys_id
)
1145 const struct cgroupfs_root
*root
= cgrp
->root
;
1146 const struct cgroup_subsys
*test_ss
;
1147 BUG_ON(list_empty(&root
->subsys_list
));
1148 test_ss
= list_entry(root
->subsys_list
.next
,
1149 struct cgroup_subsys
, sibling
);
1151 *css
= cgrp
->subsys
[test_ss
->subsys_id
];
1155 *subsys_id
= test_ss
->subsys_id
;
1159 * Attach task 'tsk' to cgroup 'cgrp'
1161 * Call holding cgroup_mutex. May take task_lock of
1162 * the task 'pid' during call.
1164 static int attach_task(struct cgroup
*cgrp
, struct task_struct
*tsk
)
1167 struct cgroup_subsys
*ss
;
1168 struct cgroup
*oldcgrp
;
1169 struct css_set
*cg
= tsk
->cgroups
;
1170 struct css_set
*newcg
;
1171 struct cgroupfs_root
*root
= cgrp
->root
;
1174 get_first_subsys(cgrp
, NULL
, &subsys_id
);
1176 /* Nothing to do if the task is already in that cgroup */
1177 oldcgrp
= task_cgroup(tsk
, subsys_id
);
1178 if (cgrp
== oldcgrp
)
1181 for_each_subsys(root
, ss
) {
1182 if (ss
->can_attach
) {
1183 retval
= ss
->can_attach(ss
, cgrp
, tsk
);
1190 * Locate or allocate a new css_set for this task,
1191 * based on its final set of cgroups
1193 newcg
= find_css_set(cg
, cgrp
);
1198 if (tsk
->flags
& PF_EXITING
) {
1203 rcu_assign_pointer(tsk
->cgroups
, newcg
);
1206 /* Update the css_set linked lists if we're using them */
1207 write_lock(&css_set_lock
);
1208 if (!list_empty(&tsk
->cg_list
)) {
1209 list_del(&tsk
->cg_list
);
1210 list_add(&tsk
->cg_list
, &newcg
->tasks
);
1212 write_unlock(&css_set_lock
);
1214 for_each_subsys(root
, ss
) {
1216 ss
->attach(ss
, cgrp
, oldcgrp
, tsk
);
1218 set_bit(CGRP_RELEASABLE
, &oldcgrp
->flags
);
1225 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with
1226 * cgroup_mutex, may take task_lock of task
1228 static int attach_task_by_pid(struct cgroup
*cgrp
, char *pidbuf
)
1231 struct task_struct
*tsk
;
1234 if (sscanf(pidbuf
, "%d", &pid
) != 1)
1239 tsk
= find_task_by_pid(pid
);
1240 if (!tsk
|| tsk
->flags
& PF_EXITING
) {
1244 get_task_struct(tsk
);
1247 if ((current
->euid
) && (current
->euid
!= tsk
->uid
)
1248 && (current
->euid
!= tsk
->suid
)) {
1249 put_task_struct(tsk
);
1254 get_task_struct(tsk
);
1257 ret
= attach_task(cgrp
, tsk
);
1258 put_task_struct(tsk
);
1262 /* The various types of files and directories in a cgroup file system */
1264 enum cgroup_filetype
{
1268 FILE_NOTIFY_ON_RELEASE
,
1273 static ssize_t
cgroup_write_uint(struct cgroup
*cgrp
, struct cftype
*cft
,
1275 const char __user
*userbuf
,
1276 size_t nbytes
, loff_t
*unused_ppos
)
1285 if (nbytes
>= sizeof(buffer
))
1287 if (copy_from_user(buffer
, userbuf
, nbytes
))
1290 buffer
[nbytes
] = 0; /* nul-terminate */
1292 /* strip newline if necessary */
1293 if (nbytes
&& (buffer
[nbytes
-1] == '\n'))
1294 buffer
[nbytes
-1] = 0;
1295 val
= simple_strtoull(buffer
, &end
, 0);
1299 /* Pass to subsystem */
1300 retval
= cft
->write_uint(cgrp
, cft
, val
);
1306 static ssize_t
cgroup_common_file_write(struct cgroup
*cgrp
,
1309 const char __user
*userbuf
,
1310 size_t nbytes
, loff_t
*unused_ppos
)
1312 enum cgroup_filetype type
= cft
->private;
1316 if (nbytes
>= PATH_MAX
)
1319 /* +1 for nul-terminator */
1320 buffer
= kmalloc(nbytes
+ 1, GFP_KERNEL
);
1324 if (copy_from_user(buffer
, userbuf
, nbytes
)) {
1328 buffer
[nbytes
] = 0; /* nul-terminate */
1329 strstrip(buffer
); /* strip -just- trailing whitespace */
1331 mutex_lock(&cgroup_mutex
);
1333 if (cgroup_is_removed(cgrp
)) {
1340 retval
= attach_task_by_pid(cgrp
, buffer
);
1342 case FILE_NOTIFY_ON_RELEASE
:
1343 clear_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
1344 if (simple_strtoul(buffer
, NULL
, 10) != 0)
1345 set_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
1347 clear_bit(CGRP_NOTIFY_ON_RELEASE
, &cgrp
->flags
);
1349 case FILE_RELEASE_AGENT
:
1350 BUILD_BUG_ON(sizeof(cgrp
->root
->release_agent_path
) < PATH_MAX
);
1351 strcpy(cgrp
->root
->release_agent_path
, buffer
);
1361 mutex_unlock(&cgroup_mutex
);
1367 static ssize_t
cgroup_file_write(struct file
*file
, const char __user
*buf
,
1368 size_t nbytes
, loff_t
*ppos
)
1370 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1371 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
1376 return cft
->write(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1377 if (cft
->write_uint
)
1378 return cgroup_write_uint(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1382 static ssize_t
cgroup_read_uint(struct cgroup
*cgrp
, struct cftype
*cft
,
1384 char __user
*buf
, size_t nbytes
,
1388 u64 val
= cft
->read_uint(cgrp
, cft
);
1389 int len
= sprintf(tmp
, "%llu\n", (unsigned long long) val
);
1391 return simple_read_from_buffer(buf
, nbytes
, ppos
, tmp
, len
);
1394 static ssize_t
cgroup_common_file_read(struct cgroup
*cgrp
,
1398 size_t nbytes
, loff_t
*ppos
)
1400 enum cgroup_filetype type
= cft
->private;
1405 if (!(page
= (char *)__get_free_page(GFP_KERNEL
)))
1411 case FILE_RELEASE_AGENT
:
1413 struct cgroupfs_root
*root
;
1415 mutex_lock(&cgroup_mutex
);
1417 n
= strnlen(root
->release_agent_path
,
1418 sizeof(root
->release_agent_path
));
1419 n
= min(n
, (size_t) PAGE_SIZE
);
1420 strncpy(s
, root
->release_agent_path
, n
);
1421 mutex_unlock(&cgroup_mutex
);
1431 retval
= simple_read_from_buffer(buf
, nbytes
, ppos
, page
, s
- page
);
1433 free_page((unsigned long)page
);
1437 static ssize_t
cgroup_file_read(struct file
*file
, char __user
*buf
,
1438 size_t nbytes
, loff_t
*ppos
)
1440 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1441 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
1447 return cft
->read(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1449 return cgroup_read_uint(cgrp
, cft
, file
, buf
, nbytes
, ppos
);
1453 static int cgroup_file_open(struct inode
*inode
, struct file
*file
)
1458 err
= generic_file_open(inode
, file
);
1462 cft
= __d_cft(file
->f_dentry
);
1466 err
= cft
->open(inode
, file
);
1473 static int cgroup_file_release(struct inode
*inode
, struct file
*file
)
1475 struct cftype
*cft
= __d_cft(file
->f_dentry
);
1477 return cft
->release(inode
, file
);
1482 * cgroup_rename - Only allow simple rename of directories in place.
1484 static int cgroup_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1485 struct inode
*new_dir
, struct dentry
*new_dentry
)
1487 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
))
1489 if (new_dentry
->d_inode
)
1491 if (old_dir
!= new_dir
)
1493 return simple_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
1496 static struct file_operations cgroup_file_operations
= {
1497 .read
= cgroup_file_read
,
1498 .write
= cgroup_file_write
,
1499 .llseek
= generic_file_llseek
,
1500 .open
= cgroup_file_open
,
1501 .release
= cgroup_file_release
,
1504 static struct inode_operations cgroup_dir_inode_operations
= {
1505 .lookup
= simple_lookup
,
1506 .mkdir
= cgroup_mkdir
,
1507 .rmdir
= cgroup_rmdir
,
1508 .rename
= cgroup_rename
,
1511 static int cgroup_create_file(struct dentry
*dentry
, int mode
,
1512 struct super_block
*sb
)
1514 static struct dentry_operations cgroup_dops
= {
1515 .d_iput
= cgroup_diput
,
1518 struct inode
*inode
;
1522 if (dentry
->d_inode
)
1525 inode
= cgroup_new_inode(mode
, sb
);
1529 if (S_ISDIR(mode
)) {
1530 inode
->i_op
= &cgroup_dir_inode_operations
;
1531 inode
->i_fop
= &simple_dir_operations
;
1533 /* start off with i_nlink == 2 (for "." entry) */
1536 /* start with the directory inode held, so that we can
1537 * populate it without racing with another mkdir */
1538 mutex_lock_nested(&inode
->i_mutex
, I_MUTEX_CHILD
);
1539 } else if (S_ISREG(mode
)) {
1541 inode
->i_fop
= &cgroup_file_operations
;
1543 dentry
->d_op
= &cgroup_dops
;
1544 d_instantiate(dentry
, inode
);
1545 dget(dentry
); /* Extra count - pin the dentry in core */
1550 * cgroup_create_dir - create a directory for an object.
1551 * cgrp: the cgroup we create the directory for.
1552 * It must have a valid ->parent field
1553 * And we are going to fill its ->dentry field.
1554 * dentry: dentry of the new cgroup
1555 * mode: mode to set on new directory.
1557 static int cgroup_create_dir(struct cgroup
*cgrp
, struct dentry
*dentry
,
1560 struct dentry
*parent
;
1563 parent
= cgrp
->parent
->dentry
;
1564 error
= cgroup_create_file(dentry
, S_IFDIR
| mode
, cgrp
->root
->sb
);
1566 dentry
->d_fsdata
= cgrp
;
1567 inc_nlink(parent
->d_inode
);
1568 cgrp
->dentry
= dentry
;
1576 int cgroup_add_file(struct cgroup
*cgrp
,
1577 struct cgroup_subsys
*subsys
,
1578 const struct cftype
*cft
)
1580 struct dentry
*dir
= cgrp
->dentry
;
1581 struct dentry
*dentry
;
1584 char name
[MAX_CGROUP_TYPE_NAMELEN
+ MAX_CFTYPE_NAME
+ 2] = { 0 };
1585 if (subsys
&& !test_bit(ROOT_NOPREFIX
, &cgrp
->root
->flags
)) {
1586 strcpy(name
, subsys
->name
);
1589 strcat(name
, cft
->name
);
1590 BUG_ON(!mutex_is_locked(&dir
->d_inode
->i_mutex
));
1591 dentry
= lookup_one_len(name
, dir
, strlen(name
));
1592 if (!IS_ERR(dentry
)) {
1593 error
= cgroup_create_file(dentry
, 0644 | S_IFREG
,
1596 dentry
->d_fsdata
= (void *)cft
;
1599 error
= PTR_ERR(dentry
);
1603 int cgroup_add_files(struct cgroup
*cgrp
,
1604 struct cgroup_subsys
*subsys
,
1605 const struct cftype cft
[],
1609 for (i
= 0; i
< count
; i
++) {
1610 err
= cgroup_add_file(cgrp
, subsys
, &cft
[i
]);
1617 /* Count the number of tasks in a cgroup. */
1619 int cgroup_task_count(const struct cgroup
*cgrp
)
1622 struct list_head
*l
;
1624 read_lock(&css_set_lock
);
1625 l
= cgrp
->css_sets
.next
;
1626 while (l
!= &cgrp
->css_sets
) {
1627 struct cg_cgroup_link
*link
=
1628 list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
1629 count
+= atomic_read(&link
->cg
->ref
.refcount
);
1632 read_unlock(&css_set_lock
);
1637 * Advance a list_head iterator. The iterator should be positioned at
1638 * the start of a css_set
1640 static void cgroup_advance_iter(struct cgroup
*cgrp
,
1641 struct cgroup_iter
*it
)
1643 struct list_head
*l
= it
->cg_link
;
1644 struct cg_cgroup_link
*link
;
1647 /* Advance to the next non-empty css_set */
1650 if (l
== &cgrp
->css_sets
) {
1654 link
= list_entry(l
, struct cg_cgroup_link
, cgrp_link_list
);
1656 } while (list_empty(&cg
->tasks
));
1658 it
->task
= cg
->tasks
.next
;
1661 void cgroup_iter_start(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
1664 * The first time anyone tries to iterate across a cgroup,
1665 * we need to enable the list linking each css_set to its
1666 * tasks, and fix up all existing tasks.
1668 if (!use_task_css_set_links
) {
1669 struct task_struct
*p
, *g
;
1670 write_lock(&css_set_lock
);
1671 use_task_css_set_links
= 1;
1672 do_each_thread(g
, p
) {
1674 if (list_empty(&p
->cg_list
))
1675 list_add(&p
->cg_list
, &p
->cgroups
->tasks
);
1677 } while_each_thread(g
, p
);
1678 write_unlock(&css_set_lock
);
1680 read_lock(&css_set_lock
);
1681 it
->cg_link
= &cgrp
->css_sets
;
1682 cgroup_advance_iter(cgrp
, it
);
1685 struct task_struct
*cgroup_iter_next(struct cgroup
*cgrp
,
1686 struct cgroup_iter
*it
)
1688 struct task_struct
*res
;
1689 struct list_head
*l
= it
->task
;
1691 /* If the iterator cg is NULL, we have no tasks */
1694 res
= list_entry(l
, struct task_struct
, cg_list
);
1695 /* Advance iterator to find next entry */
1697 if (l
== &res
->cgroups
->tasks
) {
1698 /* We reached the end of this task list - move on to
1699 * the next cg_cgroup_link */
1700 cgroup_advance_iter(cgrp
, it
);
1707 void cgroup_iter_end(struct cgroup
*cgrp
, struct cgroup_iter
*it
)
1709 read_unlock(&css_set_lock
);
1713 * Stuff for reading the 'tasks' file.
1715 * Reading this file can return large amounts of data if a cgroup has
1716 * *lots* of attached tasks. So it may need several calls to read(),
1717 * but we cannot guarantee that the information we produce is correct
1718 * unless we produce it entirely atomically.
1720 * Upon tasks file open(), a struct ctr_struct is allocated, that
1721 * will have a pointer to an array (also allocated here). The struct
1722 * ctr_struct * is stored in file->private_data. Its resources will
1723 * be freed by release() when the file is closed. The array is used
1724 * to sprintf the PIDs and then used by read().
1732 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
1733 * 'cgrp'. Return actual number of pids loaded. No need to
1734 * task_lock(p) when reading out p->cgroup, since we're in an RCU
1735 * read section, so the css_set can't go away, and is
1736 * immutable after creation.
1738 static int pid_array_load(pid_t
*pidarray
, int npids
, struct cgroup
*cgrp
)
1741 struct cgroup_iter it
;
1742 struct task_struct
*tsk
;
1743 cgroup_iter_start(cgrp
, &it
);
1744 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
1745 if (unlikely(n
== npids
))
1747 pidarray
[n
++] = task_pid_nr(tsk
);
1749 cgroup_iter_end(cgrp
, &it
);
1754 * Build and fill cgroupstats so that taskstats can export it to user
1757 * @stats: cgroupstats to fill information into
1758 * @dentry: A dentry entry belonging to the cgroup for which stats have
1761 int cgroupstats_build(struct cgroupstats
*stats
, struct dentry
*dentry
)
1764 struct cgroup
*cgrp
;
1765 struct cgroup_iter it
;
1766 struct task_struct
*tsk
;
1768 * Validate dentry by checking the superblock operations
1770 if (dentry
->d_sb
->s_op
!= &cgroup_ops
)
1774 cgrp
= dentry
->d_fsdata
;
1777 cgroup_iter_start(cgrp
, &it
);
1778 while ((tsk
= cgroup_iter_next(cgrp
, &it
))) {
1779 switch (tsk
->state
) {
1781 stats
->nr_running
++;
1783 case TASK_INTERRUPTIBLE
:
1784 stats
->nr_sleeping
++;
1786 case TASK_UNINTERRUPTIBLE
:
1787 stats
->nr_uninterruptible
++;
1790 stats
->nr_stopped
++;
1793 if (delayacct_is_task_waiting_on_io(tsk
))
1794 stats
->nr_io_wait
++;
1798 cgroup_iter_end(cgrp
, &it
);
1805 static int cmppid(const void *a
, const void *b
)
1807 return *(pid_t
*)a
- *(pid_t
*)b
;
1811 * Convert array 'a' of 'npids' pid_t's to a string of newline separated
1812 * decimal pids in 'buf'. Don't write more than 'sz' chars, but return
1813 * count 'cnt' of how many chars would be written if buf were large enough.
1815 static int pid_array_to_buf(char *buf
, int sz
, pid_t
*a
, int npids
)
1820 for (i
= 0; i
< npids
; i
++)
1821 cnt
+= snprintf(buf
+ cnt
, max(sz
- cnt
, 0), "%d\n", a
[i
]);
1826 * Handle an open on 'tasks' file. Prepare a buffer listing the
1827 * process id's of tasks currently attached to the cgroup being opened.
1829 * Does not require any specific cgroup mutexes, and does not take any.
1831 static int cgroup_tasks_open(struct inode
*unused
, struct file
*file
)
1833 struct cgroup
*cgrp
= __d_cgrp(file
->f_dentry
->d_parent
);
1834 struct ctr_struct
*ctr
;
1839 if (!(file
->f_mode
& FMODE_READ
))
1842 ctr
= kmalloc(sizeof(*ctr
), GFP_KERNEL
);
1847 * If cgroup gets more users after we read count, we won't have
1848 * enough space - tough. This race is indistinguishable to the
1849 * caller from the case that the additional cgroup users didn't
1850 * show up until sometime later on.
1852 npids
= cgroup_task_count(cgrp
);
1854 pidarray
= kmalloc(npids
* sizeof(pid_t
), GFP_KERNEL
);
1858 npids
= pid_array_load(pidarray
, npids
, cgrp
);
1859 sort(pidarray
, npids
, sizeof(pid_t
), cmppid
, NULL
);
1861 /* Call pid_array_to_buf() twice, first just to get bufsz */
1862 ctr
->bufsz
= pid_array_to_buf(&c
, sizeof(c
), pidarray
, npids
) + 1;
1863 ctr
->buf
= kmalloc(ctr
->bufsz
, GFP_KERNEL
);
1866 ctr
->bufsz
= pid_array_to_buf(ctr
->buf
, ctr
->bufsz
, pidarray
, npids
);
1873 file
->private_data
= ctr
;
1884 static ssize_t
cgroup_tasks_read(struct cgroup
*cgrp
,
1886 struct file
*file
, char __user
*buf
,
1887 size_t nbytes
, loff_t
*ppos
)
1889 struct ctr_struct
*ctr
= file
->private_data
;
1891 return simple_read_from_buffer(buf
, nbytes
, ppos
, ctr
->buf
, ctr
->bufsz
);
1894 static int cgroup_tasks_release(struct inode
*unused_inode
,
1897 struct ctr_struct
*ctr
;
1899 if (file
->f_mode
& FMODE_READ
) {
1900 ctr
= file
->private_data
;
1907 static u64
cgroup_read_notify_on_release(struct cgroup
*cgrp
,
1910 return notify_on_release(cgrp
);
1913 static u64
cgroup_read_releasable(struct cgroup
*cgrp
, struct cftype
*cft
)
1915 return test_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
1919 * for the common functions, 'private' gives the type of file
1921 static struct cftype files
[] = {
1924 .open
= cgroup_tasks_open
,
1925 .read
= cgroup_tasks_read
,
1926 .write
= cgroup_common_file_write
,
1927 .release
= cgroup_tasks_release
,
1928 .private = FILE_TASKLIST
,
1932 .name
= "notify_on_release",
1933 .read_uint
= cgroup_read_notify_on_release
,
1934 .write
= cgroup_common_file_write
,
1935 .private = FILE_NOTIFY_ON_RELEASE
,
1939 .name
= "releasable",
1940 .read_uint
= cgroup_read_releasable
,
1941 .private = FILE_RELEASABLE
,
1945 static struct cftype cft_release_agent
= {
1946 .name
= "release_agent",
1947 .read
= cgroup_common_file_read
,
1948 .write
= cgroup_common_file_write
,
1949 .private = FILE_RELEASE_AGENT
,
1952 static int cgroup_populate_dir(struct cgroup
*cgrp
)
1955 struct cgroup_subsys
*ss
;
1957 /* First clear out any existing files */
1958 cgroup_clear_directory(cgrp
->dentry
);
1960 err
= cgroup_add_files(cgrp
, NULL
, files
, ARRAY_SIZE(files
));
1964 if (cgrp
== cgrp
->top_cgroup
) {
1965 if ((err
= cgroup_add_file(cgrp
, NULL
, &cft_release_agent
)) < 0)
1969 for_each_subsys(cgrp
->root
, ss
) {
1970 if (ss
->populate
&& (err
= ss
->populate(ss
, cgrp
)) < 0)
1977 static void init_cgroup_css(struct cgroup_subsys_state
*css
,
1978 struct cgroup_subsys
*ss
,
1979 struct cgroup
*cgrp
)
1982 atomic_set(&css
->refcnt
, 0);
1984 if (cgrp
== dummytop
)
1985 set_bit(CSS_ROOT
, &css
->flags
);
1986 BUG_ON(cgrp
->subsys
[ss
->subsys_id
]);
1987 cgrp
->subsys
[ss
->subsys_id
] = css
;
1991 * cgroup_create - create a cgroup
1992 * parent: cgroup that will be parent of the new cgroup.
1993 * name: name of the new cgroup. Will be strcpy'ed.
1994 * mode: mode to set on new inode
1996 * Must be called with the mutex on the parent inode held
1999 static long cgroup_create(struct cgroup
*parent
, struct dentry
*dentry
,
2002 struct cgroup
*cgrp
;
2003 struct cgroupfs_root
*root
= parent
->root
;
2005 struct cgroup_subsys
*ss
;
2006 struct super_block
*sb
= root
->sb
;
2008 cgrp
= kzalloc(sizeof(*cgrp
), GFP_KERNEL
);
2012 /* Grab a reference on the superblock so the hierarchy doesn't
2013 * get deleted on unmount if there are child cgroups. This
2014 * can be done outside cgroup_mutex, since the sb can't
2015 * disappear while someone has an open control file on the
2017 atomic_inc(&sb
->s_active
);
2019 mutex_lock(&cgroup_mutex
);
2022 INIT_LIST_HEAD(&cgrp
->sibling
);
2023 INIT_LIST_HEAD(&cgrp
->children
);
2024 INIT_LIST_HEAD(&cgrp
->css_sets
);
2025 INIT_LIST_HEAD(&cgrp
->release_list
);
2027 cgrp
->parent
= parent
;
2028 cgrp
->root
= parent
->root
;
2029 cgrp
->top_cgroup
= parent
->top_cgroup
;
2031 for_each_subsys(root
, ss
) {
2032 struct cgroup_subsys_state
*css
= ss
->create(ss
, cgrp
);
2037 init_cgroup_css(css
, ss
, cgrp
);
2040 list_add(&cgrp
->sibling
, &cgrp
->parent
->children
);
2041 root
->number_of_cgroups
++;
2043 err
= cgroup_create_dir(cgrp
, dentry
, mode
);
2047 /* The cgroup directory was pre-locked for us */
2048 BUG_ON(!mutex_is_locked(&cgrp
->dentry
->d_inode
->i_mutex
));
2050 err
= cgroup_populate_dir(cgrp
);
2051 /* If err < 0, we have a half-filled directory - oh well ;) */
2053 mutex_unlock(&cgroup_mutex
);
2054 mutex_unlock(&cgrp
->dentry
->d_inode
->i_mutex
);
2060 list_del(&cgrp
->sibling
);
2061 root
->number_of_cgroups
--;
2065 for_each_subsys(root
, ss
) {
2066 if (cgrp
->subsys
[ss
->subsys_id
])
2067 ss
->destroy(ss
, cgrp
);
2070 mutex_unlock(&cgroup_mutex
);
2072 /* Release the reference count that we took on the superblock */
2073 deactivate_super(sb
);
2079 static int cgroup_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2081 struct cgroup
*c_parent
= dentry
->d_parent
->d_fsdata
;
2083 /* the vfs holds inode->i_mutex already */
2084 return cgroup_create(c_parent
, dentry
, mode
| S_IFDIR
);
2087 static inline int cgroup_has_css_refs(struct cgroup
*cgrp
)
2089 /* Check the reference count on each subsystem. Since we
2090 * already established that there are no tasks in the
2091 * cgroup, if the css refcount is also 0, then there should
2092 * be no outstanding references, so the subsystem is safe to
2093 * destroy. We scan across all subsystems rather than using
2094 * the per-hierarchy linked list of mounted subsystems since
2095 * we can be called via check_for_release() with no
2096 * synchronization other than RCU, and the subsystem linked
2097 * list isn't RCU-safe */
2099 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2100 struct cgroup_subsys
*ss
= subsys
[i
];
2101 struct cgroup_subsys_state
*css
;
2102 /* Skip subsystems not in this hierarchy */
2103 if (ss
->root
!= cgrp
->root
)
2105 css
= cgrp
->subsys
[ss
->subsys_id
];
2106 /* When called from check_for_release() it's possible
2107 * that by this point the cgroup has been removed
2108 * and the css deleted. But a false-positive doesn't
2109 * matter, since it can only happen if the cgroup
2110 * has been deleted and hence no longer needs the
2111 * release agent to be called anyway. */
2112 if (css
&& atomic_read(&css
->refcnt
))
2118 static int cgroup_rmdir(struct inode
*unused_dir
, struct dentry
*dentry
)
2120 struct cgroup
*cgrp
= dentry
->d_fsdata
;
2122 struct cgroup
*parent
;
2123 struct cgroup_subsys
*ss
;
2124 struct super_block
*sb
;
2125 struct cgroupfs_root
*root
;
2127 /* the vfs holds both inode->i_mutex already */
2129 mutex_lock(&cgroup_mutex
);
2130 if (atomic_read(&cgrp
->count
) != 0) {
2131 mutex_unlock(&cgroup_mutex
);
2134 if (!list_empty(&cgrp
->children
)) {
2135 mutex_unlock(&cgroup_mutex
);
2139 parent
= cgrp
->parent
;
2143 if (cgroup_has_css_refs(cgrp
)) {
2144 mutex_unlock(&cgroup_mutex
);
2148 for_each_subsys(root
, ss
) {
2149 if (cgrp
->subsys
[ss
->subsys_id
])
2150 ss
->destroy(ss
, cgrp
);
2153 spin_lock(&release_list_lock
);
2154 set_bit(CGRP_REMOVED
, &cgrp
->flags
);
2155 if (!list_empty(&cgrp
->release_list
))
2156 list_del(&cgrp
->release_list
);
2157 spin_unlock(&release_list_lock
);
2158 /* delete my sibling from parent->children */
2159 list_del(&cgrp
->sibling
);
2160 spin_lock(&cgrp
->dentry
->d_lock
);
2161 d
= dget(cgrp
->dentry
);
2162 cgrp
->dentry
= NULL
;
2163 spin_unlock(&d
->d_lock
);
2165 cgroup_d_remove_dir(d
);
2167 root
->number_of_cgroups
--;
2169 set_bit(CGRP_RELEASABLE
, &parent
->flags
);
2170 check_for_release(parent
);
2172 mutex_unlock(&cgroup_mutex
);
2173 /* Drop the active superblock reference that we took when we
2174 * created the cgroup */
2175 deactivate_super(sb
);
2179 static void cgroup_init_subsys(struct cgroup_subsys
*ss
)
2181 struct cgroup_subsys_state
*css
;
2182 struct list_head
*l
;
2184 printk(KERN_INFO
"Initializing cgroup subsys %s\n", ss
->name
);
2186 /* Create the top cgroup state for this subsystem */
2187 ss
->root
= &rootnode
;
2188 css
= ss
->create(ss
, dummytop
);
2189 /* We don't handle early failures gracefully */
2190 BUG_ON(IS_ERR(css
));
2191 init_cgroup_css(css
, ss
, dummytop
);
2193 /* Update all cgroup groups to contain a subsys
2194 * pointer to this state - since the subsystem is
2195 * newly registered, all tasks and hence all cgroup
2196 * groups are in the subsystem's top cgroup. */
2197 write_lock(&css_set_lock
);
2198 l
= &init_css_set
.list
;
2200 struct css_set
*cg
=
2201 list_entry(l
, struct css_set
, list
);
2202 cg
->subsys
[ss
->subsys_id
] = dummytop
->subsys
[ss
->subsys_id
];
2204 } while (l
!= &init_css_set
.list
);
2205 write_unlock(&css_set_lock
);
2207 /* If this subsystem requested that it be notified with fork
2208 * events, we should send it one now for every process in the
2211 struct task_struct
*g
, *p
;
2213 read_lock(&tasklist_lock
);
2214 do_each_thread(g
, p
) {
2216 } while_each_thread(g
, p
);
2217 read_unlock(&tasklist_lock
);
2220 need_forkexit_callback
|= ss
->fork
|| ss
->exit
;
2226 * cgroup_init_early - initialize cgroups at system boot, and
2227 * initialize any subsystems that request early init.
2229 int __init
cgroup_init_early(void)
2232 kref_init(&init_css_set
.ref
);
2233 kref_get(&init_css_set
.ref
);
2234 INIT_LIST_HEAD(&init_css_set
.list
);
2235 INIT_LIST_HEAD(&init_css_set
.cg_links
);
2236 INIT_LIST_HEAD(&init_css_set
.tasks
);
2238 init_cgroup_root(&rootnode
);
2239 list_add(&rootnode
.root_list
, &roots
);
2241 init_task
.cgroups
= &init_css_set
;
2243 init_css_set_link
.cg
= &init_css_set
;
2244 list_add(&init_css_set_link
.cgrp_link_list
,
2245 &rootnode
.top_cgroup
.css_sets
);
2246 list_add(&init_css_set_link
.cg_link_list
,
2247 &init_css_set
.cg_links
);
2249 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2250 struct cgroup_subsys
*ss
= subsys
[i
];
2253 BUG_ON(strlen(ss
->name
) > MAX_CGROUP_TYPE_NAMELEN
);
2254 BUG_ON(!ss
->create
);
2255 BUG_ON(!ss
->destroy
);
2256 if (ss
->subsys_id
!= i
) {
2257 printk(KERN_ERR
"cgroup: Subsys %s id == %d\n",
2258 ss
->name
, ss
->subsys_id
);
2263 cgroup_init_subsys(ss
);
2269 * cgroup_init - register cgroup filesystem and /proc file, and
2270 * initialize any subsystems that didn't request early init.
2272 int __init
cgroup_init(void)
2276 struct proc_dir_entry
*entry
;
2278 err
= bdi_init(&cgroup_backing_dev_info
);
2282 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2283 struct cgroup_subsys
*ss
= subsys
[i
];
2284 if (!ss
->early_init
)
2285 cgroup_init_subsys(ss
);
2288 err
= register_filesystem(&cgroup_fs_type
);
2292 entry
= create_proc_entry("cgroups", 0, NULL
);
2294 entry
->proc_fops
= &proc_cgroupstats_operations
;
2298 bdi_destroy(&cgroup_backing_dev_info
);
2304 * proc_cgroup_show()
2305 * - Print task's cgroup paths into seq_file, one line for each hierarchy
2306 * - Used for /proc/<pid>/cgroup.
2307 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
2308 * doesn't really matter if tsk->cgroup changes after we read it,
2309 * and we take cgroup_mutex, keeping attach_task() from changing it
2310 * anyway. No need to check that tsk->cgroup != NULL, thanks to
2311 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
2312 * cgroup to top_cgroup.
2315 /* TODO: Use a proper seq_file iterator */
2316 static int proc_cgroup_show(struct seq_file
*m
, void *v
)
2319 struct task_struct
*tsk
;
2322 struct cgroupfs_root
*root
;
2325 buf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
2331 tsk
= get_pid_task(pid
, PIDTYPE_PID
);
2337 mutex_lock(&cgroup_mutex
);
2339 for_each_root(root
) {
2340 struct cgroup_subsys
*ss
;
2341 struct cgroup
*cgrp
;
2345 /* Skip this hierarchy if it has no active subsystems */
2346 if (!root
->actual_subsys_bits
)
2348 for_each_subsys(root
, ss
)
2349 seq_printf(m
, "%s%s", count
++ ? "," : "", ss
->name
);
2351 get_first_subsys(&root
->top_cgroup
, NULL
, &subsys_id
);
2352 cgrp
= task_cgroup(tsk
, subsys_id
);
2353 retval
= cgroup_path(cgrp
, buf
, PAGE_SIZE
);
2361 mutex_unlock(&cgroup_mutex
);
2362 put_task_struct(tsk
);
2369 static int cgroup_open(struct inode
*inode
, struct file
*file
)
2371 struct pid
*pid
= PROC_I(inode
)->pid
;
2372 return single_open(file
, proc_cgroup_show
, pid
);
2375 struct file_operations proc_cgroup_operations
= {
2376 .open
= cgroup_open
,
2378 .llseek
= seq_lseek
,
2379 .release
= single_release
,
2382 /* Display information about each subsystem and each hierarchy */
2383 static int proc_cgroupstats_show(struct seq_file
*m
, void *v
)
2387 seq_puts(m
, "#subsys_name\thierarchy\tnum_cgroups\n");
2388 mutex_lock(&cgroup_mutex
);
2389 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2390 struct cgroup_subsys
*ss
= subsys
[i
];
2391 seq_printf(m
, "%s\t%lu\t%d\n",
2392 ss
->name
, ss
->root
->subsys_bits
,
2393 ss
->root
->number_of_cgroups
);
2395 mutex_unlock(&cgroup_mutex
);
2399 static int cgroupstats_open(struct inode
*inode
, struct file
*file
)
2401 return single_open(file
, proc_cgroupstats_show
, 0);
2404 static struct file_operations proc_cgroupstats_operations
= {
2405 .open
= cgroupstats_open
,
2407 .llseek
= seq_lseek
,
2408 .release
= single_release
,
2412 * cgroup_fork - attach newly forked task to its parents cgroup.
2413 * @tsk: pointer to task_struct of forking parent process.
2415 * Description: A task inherits its parent's cgroup at fork().
2417 * A pointer to the shared css_set was automatically copied in
2418 * fork.c by dup_task_struct(). However, we ignore that copy, since
2419 * it was not made under the protection of RCU or cgroup_mutex, so
2420 * might no longer be a valid cgroup pointer. attach_task() might
2421 * have already changed current->cgroups, allowing the previously
2422 * referenced cgroup group to be removed and freed.
2424 * At the point that cgroup_fork() is called, 'current' is the parent
2425 * task, and the passed argument 'child' points to the child task.
2427 void cgroup_fork(struct task_struct
*child
)
2430 child
->cgroups
= current
->cgroups
;
2431 get_css_set(child
->cgroups
);
2432 task_unlock(current
);
2433 INIT_LIST_HEAD(&child
->cg_list
);
2437 * cgroup_fork_callbacks - called on a new task very soon before
2438 * adding it to the tasklist. No need to take any locks since no-one
2439 * can be operating on this task
2441 void cgroup_fork_callbacks(struct task_struct
*child
)
2443 if (need_forkexit_callback
) {
2445 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2446 struct cgroup_subsys
*ss
= subsys
[i
];
2448 ss
->fork(ss
, child
);
2454 * cgroup_post_fork - called on a new task after adding it to the
2455 * task list. Adds the task to the list running through its css_set
2456 * if necessary. Has to be after the task is visible on the task list
2457 * in case we race with the first call to cgroup_iter_start() - to
2458 * guarantee that the new task ends up on its list. */
2459 void cgroup_post_fork(struct task_struct
*child
)
2461 if (use_task_css_set_links
) {
2462 write_lock(&css_set_lock
);
2463 if (list_empty(&child
->cg_list
))
2464 list_add(&child
->cg_list
, &child
->cgroups
->tasks
);
2465 write_unlock(&css_set_lock
);
2469 * cgroup_exit - detach cgroup from exiting task
2470 * @tsk: pointer to task_struct of exiting process
2472 * Description: Detach cgroup from @tsk and release it.
2474 * Note that cgroups marked notify_on_release force every task in
2475 * them to take the global cgroup_mutex mutex when exiting.
2476 * This could impact scaling on very large systems. Be reluctant to
2477 * use notify_on_release cgroups where very high task exit scaling
2478 * is required on large systems.
2480 * the_top_cgroup_hack:
2482 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
2484 * We call cgroup_exit() while the task is still competent to
2485 * handle notify_on_release(), then leave the task attached to the
2486 * root cgroup in each hierarchy for the remainder of its exit.
2488 * To do this properly, we would increment the reference count on
2489 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
2490 * code we would add a second cgroup function call, to drop that
2491 * reference. This would just create an unnecessary hot spot on
2492 * the top_cgroup reference count, to no avail.
2494 * Normally, holding a reference to a cgroup without bumping its
2495 * count is unsafe. The cgroup could go away, or someone could
2496 * attach us to a different cgroup, decrementing the count on
2497 * the first cgroup that we never incremented. But in this case,
2498 * top_cgroup isn't going away, and either task has PF_EXITING set,
2499 * which wards off any attach_task() attempts, or task is a failed
2500 * fork, never visible to attach_task.
2503 void cgroup_exit(struct task_struct
*tsk
, int run_callbacks
)
2508 if (run_callbacks
&& need_forkexit_callback
) {
2509 for (i
= 0; i
< CGROUP_SUBSYS_COUNT
; i
++) {
2510 struct cgroup_subsys
*ss
= subsys
[i
];
2517 * Unlink from the css_set task list if necessary.
2518 * Optimistically check cg_list before taking
2521 if (!list_empty(&tsk
->cg_list
)) {
2522 write_lock(&css_set_lock
);
2523 if (!list_empty(&tsk
->cg_list
))
2524 list_del(&tsk
->cg_list
);
2525 write_unlock(&css_set_lock
);
2528 /* Reassign the task to the init_css_set. */
2531 tsk
->cgroups
= &init_css_set
;
2534 put_css_set_taskexit(cg
);
2538 * cgroup_clone - duplicate the current cgroup in the hierarchy
2539 * that the given subsystem is attached to, and move this task into
2542 int cgroup_clone(struct task_struct
*tsk
, struct cgroup_subsys
*subsys
)
2544 struct dentry
*dentry
;
2546 char nodename
[MAX_CGROUP_TYPE_NAMELEN
];
2547 struct cgroup
*parent
, *child
;
2548 struct inode
*inode
;
2550 struct cgroupfs_root
*root
;
2551 struct cgroup_subsys
*ss
;
2553 /* We shouldn't be called by an unregistered subsystem */
2554 BUG_ON(!subsys
->active
);
2556 /* First figure out what hierarchy and cgroup we're dealing
2557 * with, and pin them so we can drop cgroup_mutex */
2558 mutex_lock(&cgroup_mutex
);
2560 root
= subsys
->root
;
2561 if (root
== &rootnode
) {
2563 "Not cloning cgroup for unused subsystem %s\n",
2565 mutex_unlock(&cgroup_mutex
);
2569 parent
= task_cgroup(tsk
, subsys
->subsys_id
);
2571 snprintf(nodename
, MAX_CGROUP_TYPE_NAMELEN
, "node_%d", tsk
->pid
);
2573 /* Pin the hierarchy */
2574 atomic_inc(&parent
->root
->sb
->s_active
);
2576 /* Keep the cgroup alive */
2578 mutex_unlock(&cgroup_mutex
);
2580 /* Now do the VFS work to create a cgroup */
2581 inode
= parent
->dentry
->d_inode
;
2583 /* Hold the parent directory mutex across this operation to
2584 * stop anyone else deleting the new cgroup */
2585 mutex_lock(&inode
->i_mutex
);
2586 dentry
= lookup_one_len(nodename
, parent
->dentry
, strlen(nodename
));
2587 if (IS_ERR(dentry
)) {
2589 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename
,
2591 ret
= PTR_ERR(dentry
);
2595 /* Create the cgroup directory, which also creates the cgroup */
2596 ret
= vfs_mkdir(inode
, dentry
, S_IFDIR
| 0755);
2597 child
= __d_cgrp(dentry
);
2601 "Failed to create cgroup %s: %d\n", nodename
,
2608 "Couldn't find new cgroup %s\n", nodename
);
2613 /* The cgroup now exists. Retake cgroup_mutex and check
2614 * that we're still in the same state that we thought we
2616 mutex_lock(&cgroup_mutex
);
2617 if ((root
!= subsys
->root
) ||
2618 (parent
!= task_cgroup(tsk
, subsys
->subsys_id
))) {
2619 /* Aargh, we raced ... */
2620 mutex_unlock(&inode
->i_mutex
);
2623 deactivate_super(parent
->root
->sb
);
2624 /* The cgroup is still accessible in the VFS, but
2625 * we're not going to try to rmdir() it at this
2628 "Race in cgroup_clone() - leaking cgroup %s\n",
2633 /* do any required auto-setup */
2634 for_each_subsys(root
, ss
) {
2636 ss
->post_clone(ss
, child
);
2639 /* All seems fine. Finish by moving the task into the new cgroup */
2640 ret
= attach_task(child
, tsk
);
2641 mutex_unlock(&cgroup_mutex
);
2644 mutex_unlock(&inode
->i_mutex
);
2646 mutex_lock(&cgroup_mutex
);
2648 mutex_unlock(&cgroup_mutex
);
2649 deactivate_super(parent
->root
->sb
);
2654 * See if "cgrp" is a descendant of the current task's cgroup in
2655 * the appropriate hierarchy
2657 * If we are sending in dummytop, then presumably we are creating
2658 * the top cgroup in the subsystem.
2660 * Called only by the ns (nsproxy) cgroup.
2662 int cgroup_is_descendant(const struct cgroup
*cgrp
)
2665 struct cgroup
*target
;
2668 if (cgrp
== dummytop
)
2671 get_first_subsys(cgrp
, NULL
, &subsys_id
);
2672 target
= task_cgroup(current
, subsys_id
);
2673 while (cgrp
!= target
&& cgrp
!= cgrp
->top_cgroup
)
2674 cgrp
= cgrp
->parent
;
2675 ret
= (cgrp
== target
);
2679 static void check_for_release(struct cgroup
*cgrp
)
2681 /* All of these checks rely on RCU to keep the cgroup
2682 * structure alive */
2683 if (cgroup_is_releasable(cgrp
) && !atomic_read(&cgrp
->count
)
2684 && list_empty(&cgrp
->children
) && !cgroup_has_css_refs(cgrp
)) {
2685 /* Control Group is currently removeable. If it's not
2686 * already queued for a userspace notification, queue
2688 int need_schedule_work
= 0;
2689 spin_lock(&release_list_lock
);
2690 if (!cgroup_is_removed(cgrp
) &&
2691 list_empty(&cgrp
->release_list
)) {
2692 list_add(&cgrp
->release_list
, &release_list
);
2693 need_schedule_work
= 1;
2695 spin_unlock(&release_list_lock
);
2696 if (need_schedule_work
)
2697 schedule_work(&release_agent_work
);
2701 void __css_put(struct cgroup_subsys_state
*css
)
2703 struct cgroup
*cgrp
= css
->cgroup
;
2705 if (atomic_dec_and_test(&css
->refcnt
) && notify_on_release(cgrp
)) {
2706 set_bit(CGRP_RELEASABLE
, &cgrp
->flags
);
2707 check_for_release(cgrp
);
2713 * Notify userspace when a cgroup is released, by running the
2714 * configured release agent with the name of the cgroup (path
2715 * relative to the root of cgroup file system) as the argument.
2717 * Most likely, this user command will try to rmdir this cgroup.
2719 * This races with the possibility that some other task will be
2720 * attached to this cgroup before it is removed, or that some other
2721 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
2722 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
2723 * unused, and this cgroup will be reprieved from its death sentence,
2724 * to continue to serve a useful existence. Next time it's released,
2725 * we will get notified again, if it still has 'notify_on_release' set.
2727 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
2728 * means only wait until the task is successfully execve()'d. The
2729 * separate release agent task is forked by call_usermodehelper(),
2730 * then control in this thread returns here, without waiting for the
2731 * release agent task. We don't bother to wait because the caller of
2732 * this routine has no use for the exit status of the release agent
2733 * task, so no sense holding our caller up for that.
2737 static void cgroup_release_agent(struct work_struct
*work
)
2739 BUG_ON(work
!= &release_agent_work
);
2740 mutex_lock(&cgroup_mutex
);
2741 spin_lock(&release_list_lock
);
2742 while (!list_empty(&release_list
)) {
2743 char *argv
[3], *envp
[3];
2746 struct cgroup
*cgrp
= list_entry(release_list
.next
,
2749 list_del_init(&cgrp
->release_list
);
2750 spin_unlock(&release_list_lock
);
2751 pathbuf
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
2753 spin_lock(&release_list_lock
);
2757 if (cgroup_path(cgrp
, pathbuf
, PAGE_SIZE
) < 0) {
2759 spin_lock(&release_list_lock
);
2764 argv
[i
++] = cgrp
->root
->release_agent_path
;
2765 argv
[i
++] = (char *)pathbuf
;
2769 /* minimal command environment */
2770 envp
[i
++] = "HOME=/";
2771 envp
[i
++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
2774 /* Drop the lock while we invoke the usermode helper,
2775 * since the exec could involve hitting disk and hence
2776 * be a slow process */
2777 mutex_unlock(&cgroup_mutex
);
2778 call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_EXEC
);
2780 mutex_lock(&cgroup_mutex
);
2781 spin_lock(&release_list_lock
);
2783 spin_unlock(&release_list_lock
);
2784 mutex_unlock(&cgroup_mutex
);