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);
1191 * Locate or allocate a new css_set for this task,
1192 * based on its final set of cgroups
1194 newcg = find_css_set(cg, cgrp);
1200 if (tsk->flags & PF_EXITING) {
1205 rcu_assign_pointer(tsk->cgroups, newcg);
1208 /* Update the css_set linked lists if we're using them */
1209 write_lock(&css_set_lock);
1210 if (!list_empty(&tsk->cg_list)) {
1211 list_del(&tsk->cg_list);
1212 list_add(&tsk->cg_list, &newcg->tasks);
1214 write_unlock(&css_set_lock);
1216 for_each_subsys(root, ss) {
1218 ss->attach(ss, cgrp, oldcgrp, tsk);
1221 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1228 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with
1229 * cgroup_mutex, may take task_lock of task
1231 static int attach_task_by_pid(struct cgroup *cgrp, char *pidbuf)
1234 struct task_struct *tsk;
1237 if (sscanf(pidbuf, "%d", &pid) != 1)
1242 tsk = find_task_by_pid(pid);
1243 if (!tsk || tsk->flags & PF_EXITING) {
1247 get_task_struct(tsk);
1250 if ((current->euid) && (current->euid != tsk->uid)
1251 && (current->euid != tsk->suid)) {
1252 put_task_struct(tsk);
1257 get_task_struct(tsk);
1260 ret = attach_task(cgrp, tsk);
1261 put_task_struct(tsk);
1265 /* The various types of files and directories in a cgroup file system */
1267 enum cgroup_filetype {
1271 FILE_NOTIFY_ON_RELEASE,
1276 static ssize_t cgroup_write_uint(struct cgroup *cgrp, struct cftype *cft,
1278 const char __user *userbuf,
1279 size_t nbytes, loff_t *unused_ppos)
1288 if (nbytes >= sizeof(buffer))
1290 if (copy_from_user(buffer, userbuf, nbytes))
1293 buffer[nbytes] = 0; /* nul-terminate */
1295 /* strip newline if necessary */
1296 if (nbytes && (buffer[nbytes-1] == '\n'))
1297 buffer[nbytes-1] = 0;
1298 val = simple_strtoull(buffer, &end, 0);
1302 /* Pass to subsystem */
1303 retval = cft->write_uint(cgrp, cft, val);
1309 static ssize_t cgroup_common_file_write(struct cgroup *cgrp,
1312 const char __user *userbuf,
1313 size_t nbytes, loff_t *unused_ppos)
1315 enum cgroup_filetype type = cft->private;
1319 if (nbytes >= PATH_MAX)
1322 /* +1 for nul-terminator */
1323 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1327 if (copy_from_user(buffer, userbuf, nbytes)) {
1331 buffer[nbytes] = 0; /* nul-terminate */
1333 mutex_lock(&cgroup_mutex);
1335 if (cgroup_is_removed(cgrp)) {
1342 retval = attach_task_by_pid(cgrp, buffer);
1344 case FILE_NOTIFY_ON_RELEASE:
1345 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
1346 if (simple_strtoul(buffer, NULL, 10) != 0)
1347 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
1349 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
1351 case FILE_RELEASE_AGENT:
1353 struct cgroupfs_root *root = cgrp->root;
1354 /* Strip trailing newline */
1355 if (nbytes && (buffer[nbytes-1] == '\n')) {
1356 buffer[nbytes-1] = 0;
1359 /* We never write anything other than '\0'
1360 * into the last char of release_agent_path,
1361 * so it always remains a NUL-terminated
1363 strncpy(root->release_agent_path, buffer, nbytes);
1364 root->release_agent_path[nbytes] = 0;
1376 mutex_unlock(&cgroup_mutex);
1382 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
1383 size_t nbytes, loff_t *ppos)
1385 struct cftype *cft = __d_cft(file->f_dentry);
1386 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1391 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
1392 if (cft->write_uint)
1393 return cgroup_write_uint(cgrp, cft, file, buf, nbytes, ppos);
1397 static ssize_t cgroup_read_uint(struct cgroup *cgrp, struct cftype *cft,
1399 char __user *buf, size_t nbytes,
1403 u64 val = cft->read_uint(cgrp, cft);
1404 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
1406 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1409 static ssize_t cgroup_common_file_read(struct cgroup *cgrp,
1413 size_t nbytes, loff_t *ppos)
1415 enum cgroup_filetype type = cft->private;
1420 if (!(page = (char *)__get_free_page(GFP_KERNEL)))
1426 case FILE_RELEASE_AGENT:
1428 struct cgroupfs_root *root;
1430 mutex_lock(&cgroup_mutex);
1432 n = strnlen(root->release_agent_path,
1433 sizeof(root->release_agent_path));
1434 n = min(n, (size_t) PAGE_SIZE);
1435 strncpy(s, root->release_agent_path, n);
1436 mutex_unlock(&cgroup_mutex);
1446 retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1448 free_page((unsigned long)page);
1452 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
1453 size_t nbytes, loff_t *ppos)
1455 struct cftype *cft = __d_cft(file->f_dentry);
1456 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1462 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
1464 return cgroup_read_uint(cgrp, cft, file, buf, nbytes, ppos);
1468 static int cgroup_file_open(struct inode *inode, struct file *file)
1473 err = generic_file_open(inode, file);
1477 cft = __d_cft(file->f_dentry);
1481 err = cft->open(inode, file);
1488 static int cgroup_file_release(struct inode *inode, struct file *file)
1490 struct cftype *cft = __d_cft(file->f_dentry);
1492 return cft->release(inode, file);
1497 * cgroup_rename - Only allow simple rename of directories in place.
1499 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
1500 struct inode *new_dir, struct dentry *new_dentry)
1502 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1504 if (new_dentry->d_inode)
1506 if (old_dir != new_dir)
1508 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
1511 static struct file_operations cgroup_file_operations = {
1512 .read = cgroup_file_read,
1513 .write = cgroup_file_write,
1514 .llseek = generic_file_llseek,
1515 .open = cgroup_file_open,
1516 .release = cgroup_file_release,
1519 static struct inode_operations cgroup_dir_inode_operations = {
1520 .lookup = simple_lookup,
1521 .mkdir = cgroup_mkdir,
1522 .rmdir = cgroup_rmdir,
1523 .rename = cgroup_rename,
1526 static int cgroup_create_file(struct dentry *dentry, int mode,
1527 struct super_block *sb)
1529 static struct dentry_operations cgroup_dops = {
1530 .d_iput = cgroup_diput,
1533 struct inode *inode;
1537 if (dentry->d_inode)
1540 inode = cgroup_new_inode(mode, sb);
1544 if (S_ISDIR(mode)) {
1545 inode->i_op = &cgroup_dir_inode_operations;
1546 inode->i_fop = &simple_dir_operations;
1548 /* start off with i_nlink == 2 (for "." entry) */
1551 /* start with the directory inode held, so that we can
1552 * populate it without racing with another mkdir */
1553 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1554 } else if (S_ISREG(mode)) {
1556 inode->i_fop = &cgroup_file_operations;
1558 dentry->d_op = &cgroup_dops;
1559 d_instantiate(dentry, inode);
1560 dget(dentry); /* Extra count - pin the dentry in core */
1565 * cgroup_create_dir - create a directory for an object.
1566 * cgrp: the cgroup we create the directory for.
1567 * It must have a valid ->parent field
1568 * And we are going to fill its ->dentry field.
1569 * dentry: dentry of the new cgroup
1570 * mode: mode to set on new directory.
1572 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
1575 struct dentry *parent;
1578 parent = cgrp->parent->dentry;
1579 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
1581 dentry->d_fsdata = cgrp;
1582 inc_nlink(parent->d_inode);
1583 cgrp->dentry = dentry;
1591 int cgroup_add_file(struct cgroup *cgrp,
1592 struct cgroup_subsys *subsys,
1593 const struct cftype *cft)
1595 struct dentry *dir = cgrp->dentry;
1596 struct dentry *dentry;
1599 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
1600 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
1601 strcpy(name, subsys->name);
1604 strcat(name, cft->name);
1605 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
1606 dentry = lookup_one_len(name, dir, strlen(name));
1607 if (!IS_ERR(dentry)) {
1608 error = cgroup_create_file(dentry, 0644 | S_IFREG,
1611 dentry->d_fsdata = (void *)cft;
1614 error = PTR_ERR(dentry);
1618 int cgroup_add_files(struct cgroup *cgrp,
1619 struct cgroup_subsys *subsys,
1620 const struct cftype cft[],
1624 for (i = 0; i < count; i++) {
1625 err = cgroup_add_file(cgrp, subsys, &cft[i]);
1632 /* Count the number of tasks in a cgroup. */
1634 int cgroup_task_count(const struct cgroup *cgrp)
1637 struct list_head *l;
1639 read_lock(&css_set_lock);
1640 l = cgrp->css_sets.next;
1641 while (l != &cgrp->css_sets) {
1642 struct cg_cgroup_link *link =
1643 list_entry(l, struct cg_cgroup_link, cgrp_link_list);
1644 count += atomic_read(&link->cg->ref.refcount);
1647 read_unlock(&css_set_lock);
1652 * Advance a list_head iterator. The iterator should be positioned at
1653 * the start of a css_set
1655 static void cgroup_advance_iter(struct cgroup *cgrp,
1656 struct cgroup_iter *it)
1658 struct list_head *l = it->cg_link;
1659 struct cg_cgroup_link *link;
1662 /* Advance to the next non-empty css_set */
1665 if (l == &cgrp->css_sets) {
1669 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
1671 } while (list_empty(&cg->tasks));
1673 it->task = cg->tasks.next;
1676 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
1679 * The first time anyone tries to iterate across a cgroup,
1680 * we need to enable the list linking each css_set to its
1681 * tasks, and fix up all existing tasks.
1683 if (!use_task_css_set_links) {
1684 struct task_struct *p, *g;
1685 write_lock(&css_set_lock);
1686 use_task_css_set_links = 1;
1687 do_each_thread(g, p) {
1689 if (list_empty(&p->cg_list))
1690 list_add(&p->cg_list, &p->cgroups->tasks);
1692 } while_each_thread(g, p);
1693 write_unlock(&css_set_lock);
1695 read_lock(&css_set_lock);
1696 it->cg_link = &cgrp->css_sets;
1697 cgroup_advance_iter(cgrp, it);
1700 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
1701 struct cgroup_iter *it)
1703 struct task_struct *res;
1704 struct list_head *l = it->task;
1706 /* If the iterator cg is NULL, we have no tasks */
1709 res = list_entry(l, struct task_struct, cg_list);
1710 /* Advance iterator to find next entry */
1712 if (l == &res->cgroups->tasks) {
1713 /* We reached the end of this task list - move on to
1714 * the next cg_cgroup_link */
1715 cgroup_advance_iter(cgrp, it);
1722 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
1724 read_unlock(&css_set_lock);
1728 * Stuff for reading the 'tasks' file.
1730 * Reading this file can return large amounts of data if a cgroup has
1731 * *lots* of attached tasks. So it may need several calls to read(),
1732 * but we cannot guarantee that the information we produce is correct
1733 * unless we produce it entirely atomically.
1735 * Upon tasks file open(), a struct ctr_struct is allocated, that
1736 * will have a pointer to an array (also allocated here). The struct
1737 * ctr_struct * is stored in file->private_data. Its resources will
1738 * be freed by release() when the file is closed. The array is used
1739 * to sprintf the PIDs and then used by read().
1747 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
1748 * 'cgrp'. Return actual number of pids loaded. No need to
1749 * task_lock(p) when reading out p->cgroup, since we're in an RCU
1750 * read section, so the css_set can't go away, and is
1751 * immutable after creation.
1753 static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cgrp)
1756 struct cgroup_iter it;
1757 struct task_struct *tsk;
1758 cgroup_iter_start(cgrp, &it);
1759 while ((tsk = cgroup_iter_next(cgrp, &it))) {
1760 if (unlikely(n == npids))
1762 pidarray[n++] = task_pid_nr(tsk);
1764 cgroup_iter_end(cgrp, &it);
1769 * Build and fill cgroupstats so that taskstats can export it to user
1772 * @stats: cgroupstats to fill information into
1773 * @dentry: A dentry entry belonging to the cgroup for which stats have
1776 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
1779 struct cgroup *cgrp;
1780 struct cgroup_iter it;
1781 struct task_struct *tsk;
1783 * Validate dentry by checking the superblock operations
1785 if (dentry->d_sb->s_op != &cgroup_ops)
1789 cgrp = dentry->d_fsdata;
1792 cgroup_iter_start(cgrp, &it);
1793 while ((tsk = cgroup_iter_next(cgrp, &it))) {
1794 switch (tsk->state) {
1796 stats->nr_running++;
1798 case TASK_INTERRUPTIBLE:
1799 stats->nr_sleeping++;
1801 case TASK_UNINTERRUPTIBLE:
1802 stats->nr_uninterruptible++;
1805 stats->nr_stopped++;
1808 if (delayacct_is_task_waiting_on_io(tsk))
1809 stats->nr_io_wait++;
1813 cgroup_iter_end(cgrp, &it);
1820 static int cmppid(const void *a, const void *b)
1822 return *(pid_t *)a - *(pid_t *)b;
1826 * Convert array 'a' of 'npids' pid_t's to a string of newline separated
1827 * decimal pids in 'buf'. Don't write more than 'sz' chars, but return
1828 * count 'cnt' of how many chars would be written if buf were large enough.
1830 static int pid_array_to_buf(char *buf, int sz, pid_t *a, int npids)
1835 for (i = 0; i < npids; i++)
1836 cnt += snprintf(buf + cnt, max(sz - cnt, 0), "%d\n", a[i]);
1841 * Handle an open on 'tasks' file. Prepare a buffer listing the
1842 * process id's of tasks currently attached to the cgroup being opened.
1844 * Does not require any specific cgroup mutexes, and does not take any.
1846 static int cgroup_tasks_open(struct inode *unused, struct file *file)
1848 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1849 struct ctr_struct *ctr;
1854 if (!(file->f_mode & FMODE_READ))
1857 ctr = kmalloc(sizeof(*ctr), GFP_KERNEL);
1862 * If cgroup gets more users after we read count, we won't have
1863 * enough space - tough. This race is indistinguishable to the
1864 * caller from the case that the additional cgroup users didn't
1865 * show up until sometime later on.
1867 npids = cgroup_task_count(cgrp);
1869 pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL);
1873 npids = pid_array_load(pidarray, npids, cgrp);
1874 sort(pidarray, npids, sizeof(pid_t), cmppid, NULL);
1876 /* Call pid_array_to_buf() twice, first just to get bufsz */
1877 ctr->bufsz = pid_array_to_buf(&c, sizeof(c), pidarray, npids) + 1;
1878 ctr->buf = kmalloc(ctr->bufsz, GFP_KERNEL);
1881 ctr->bufsz = pid_array_to_buf(ctr->buf, ctr->bufsz, pidarray, npids);
1888 file->private_data = ctr;
1899 static ssize_t cgroup_tasks_read(struct cgroup *cgrp,
1901 struct file *file, char __user *buf,
1902 size_t nbytes, loff_t *ppos)
1904 struct ctr_struct *ctr = file->private_data;
1906 return simple_read_from_buffer(buf, nbytes, ppos, ctr->buf, ctr->bufsz);
1909 static int cgroup_tasks_release(struct inode *unused_inode,
1912 struct ctr_struct *ctr;
1914 if (file->f_mode & FMODE_READ) {
1915 ctr = file->private_data;
1922 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
1925 return notify_on_release(cgrp);
1928 static u64 cgroup_read_releasable(struct cgroup *cgrp, struct cftype *cft)
1930 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
1934 * for the common functions, 'private' gives the type of file
1936 static struct cftype files[] = {
1939 .open = cgroup_tasks_open,
1940 .read = cgroup_tasks_read,
1941 .write = cgroup_common_file_write,
1942 .release = cgroup_tasks_release,
1943 .private = FILE_TASKLIST,
1947 .name = "notify_on_release",
1948 .read_uint = cgroup_read_notify_on_release,
1949 .write = cgroup_common_file_write,
1950 .private = FILE_NOTIFY_ON_RELEASE,
1954 .name = "releasable",
1955 .read_uint = cgroup_read_releasable,
1956 .private = FILE_RELEASABLE,
1960 static struct cftype cft_release_agent = {
1961 .name = "release_agent",
1962 .read = cgroup_common_file_read,
1963 .write = cgroup_common_file_write,
1964 .private = FILE_RELEASE_AGENT,
1967 static int cgroup_populate_dir(struct cgroup *cgrp)
1970 struct cgroup_subsys *ss;
1972 /* First clear out any existing files */
1973 cgroup_clear_directory(cgrp->dentry);
1975 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
1979 if (cgrp == cgrp->top_cgroup) {
1980 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
1984 for_each_subsys(cgrp->root, ss) {
1985 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
1992 static void init_cgroup_css(struct cgroup_subsys_state *css,
1993 struct cgroup_subsys *ss,
1994 struct cgroup *cgrp)
1997 atomic_set(&css->refcnt, 0);
1999 if (cgrp == dummytop)
2000 set_bit(CSS_ROOT, &css->flags);
2001 BUG_ON(cgrp->subsys[ss->subsys_id]);
2002 cgrp->subsys[ss->subsys_id] = css;
2006 * cgroup_create - create a cgroup
2007 * parent: cgroup that will be parent of the new cgroup.
2008 * name: name of the new cgroup. Will be strcpy'ed.
2009 * mode: mode to set on new inode
2011 * Must be called with the mutex on the parent inode held
2014 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
2017 struct cgroup *cgrp;
2018 struct cgroupfs_root *root = parent->root;
2020 struct cgroup_subsys *ss;
2021 struct super_block *sb = root->sb;
2023 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
2027 /* Grab a reference on the superblock so the hierarchy doesn't
2028 * get deleted on unmount if there are child cgroups. This
2029 * can be done outside cgroup_mutex, since the sb can't
2030 * disappear while someone has an open control file on the
2032 atomic_inc(&sb->s_active);
2034 mutex_lock(&cgroup_mutex);
2037 INIT_LIST_HEAD(&cgrp->sibling);
2038 INIT_LIST_HEAD(&cgrp->children);
2039 INIT_LIST_HEAD(&cgrp->css_sets);
2040 INIT_LIST_HEAD(&cgrp->release_list);
2042 cgrp->parent = parent;
2043 cgrp->root = parent->root;
2044 cgrp->top_cgroup = parent->top_cgroup;
2046 for_each_subsys(root, ss) {
2047 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
2052 init_cgroup_css(css, ss, cgrp);
2055 list_add(&cgrp->sibling, &cgrp->parent->children);
2056 root->number_of_cgroups++;
2058 err = cgroup_create_dir(cgrp, dentry, mode);
2062 /* The cgroup directory was pre-locked for us */
2063 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
2065 err = cgroup_populate_dir(cgrp);
2066 /* If err < 0, we have a half-filled directory - oh well ;) */
2068 mutex_unlock(&cgroup_mutex);
2069 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
2075 list_del(&cgrp->sibling);
2076 root->number_of_cgroups--;
2080 for_each_subsys(root, ss) {
2081 if (cgrp->subsys[ss->subsys_id])
2082 ss->destroy(ss, cgrp);
2085 mutex_unlock(&cgroup_mutex);
2087 /* Release the reference count that we took on the superblock */
2088 deactivate_super(sb);
2094 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2096 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
2098 /* the vfs holds inode->i_mutex already */
2099 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
2102 static inline int cgroup_has_css_refs(struct cgroup *cgrp)
2104 /* Check the reference count on each subsystem. Since we
2105 * already established that there are no tasks in the
2106 * cgroup, if the css refcount is also 0, then there should
2107 * be no outstanding references, so the subsystem is safe to
2108 * destroy. We scan across all subsystems rather than using
2109 * the per-hierarchy linked list of mounted subsystems since
2110 * we can be called via check_for_release() with no
2111 * synchronization other than RCU, and the subsystem linked
2112 * list isn't RCU-safe */
2114 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2115 struct cgroup_subsys *ss = subsys[i];
2116 struct cgroup_subsys_state *css;
2117 /* Skip subsystems not in this hierarchy */
2118 if (ss->root != cgrp->root)
2120 css = cgrp->subsys[ss->subsys_id];
2121 /* When called from check_for_release() it's possible
2122 * that by this point the cgroup has been removed
2123 * and the css deleted. But a false-positive doesn't
2124 * matter, since it can only happen if the cgroup
2125 * has been deleted and hence no longer needs the
2126 * release agent to be called anyway. */
2127 if (css && atomic_read(&css->refcnt)) {
2134 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
2136 struct cgroup *cgrp = dentry->d_fsdata;
2138 struct cgroup *parent;
2139 struct cgroup_subsys *ss;
2140 struct super_block *sb;
2141 struct cgroupfs_root *root;
2143 /* the vfs holds both inode->i_mutex already */
2145 mutex_lock(&cgroup_mutex);
2146 if (atomic_read(&cgrp->count) != 0) {
2147 mutex_unlock(&cgroup_mutex);
2150 if (!list_empty(&cgrp->children)) {
2151 mutex_unlock(&cgroup_mutex);
2155 parent = cgrp->parent;
2159 if (cgroup_has_css_refs(cgrp)) {
2160 mutex_unlock(&cgroup_mutex);
2164 for_each_subsys(root, ss) {
2165 if (cgrp->subsys[ss->subsys_id])
2166 ss->destroy(ss, cgrp);
2169 spin_lock(&release_list_lock);
2170 set_bit(CGRP_REMOVED, &cgrp->flags);
2171 if (!list_empty(&cgrp->release_list))
2172 list_del(&cgrp->release_list);
2173 spin_unlock(&release_list_lock);
2174 /* delete my sibling from parent->children */
2175 list_del(&cgrp->sibling);
2176 spin_lock(&cgrp->dentry->d_lock);
2177 d = dget(cgrp->dentry);
2178 cgrp->dentry = NULL;
2179 spin_unlock(&d->d_lock);
2181 cgroup_d_remove_dir(d);
2183 root->number_of_cgroups--;
2185 set_bit(CGRP_RELEASABLE, &parent->flags);
2186 check_for_release(parent);
2188 mutex_unlock(&cgroup_mutex);
2189 /* Drop the active superblock reference that we took when we
2190 * created the cgroup */
2191 deactivate_super(sb);
2195 static void cgroup_init_subsys(struct cgroup_subsys *ss)
2197 struct cgroup_subsys_state *css;
2198 struct list_head *l;
2200 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
2202 /* Create the top cgroup state for this subsystem */
2203 ss->root = &rootnode;
2204 css = ss->create(ss, dummytop);
2205 /* We don't handle early failures gracefully */
2206 BUG_ON(IS_ERR(css));
2207 init_cgroup_css(css, ss, dummytop);
2209 /* Update all cgroup groups to contain a subsys
2210 * pointer to this state - since the subsystem is
2211 * newly registered, all tasks and hence all cgroup
2212 * groups are in the subsystem's top cgroup. */
2213 write_lock(&css_set_lock);
2214 l = &init_css_set.list;
2216 struct css_set *cg =
2217 list_entry(l, struct css_set, list);
2218 cg->subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
2220 } while (l != &init_css_set.list);
2221 write_unlock(&css_set_lock);
2223 /* If this subsystem requested that it be notified with fork
2224 * events, we should send it one now for every process in the
2227 struct task_struct *g, *p;
2229 read_lock(&tasklist_lock);
2230 do_each_thread(g, p) {
2232 } while_each_thread(g, p);
2233 read_unlock(&tasklist_lock);
2236 need_forkexit_callback |= ss->fork || ss->exit;
2242 * cgroup_init_early - initialize cgroups at system boot, and
2243 * initialize any subsystems that request early init.
2245 int __init cgroup_init_early(void)
2248 kref_init(&init_css_set.ref);
2249 kref_get(&init_css_set.ref);
2250 INIT_LIST_HEAD(&init_css_set.list);
2251 INIT_LIST_HEAD(&init_css_set.cg_links);
2252 INIT_LIST_HEAD(&init_css_set.tasks);
2254 init_cgroup_root(&rootnode);
2255 list_add(&rootnode.root_list, &roots);
2257 init_task.cgroups = &init_css_set;
2259 init_css_set_link.cg = &init_css_set;
2260 list_add(&init_css_set_link.cgrp_link_list,
2261 &rootnode.top_cgroup.css_sets);
2262 list_add(&init_css_set_link.cg_link_list,
2263 &init_css_set.cg_links);
2265 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2266 struct cgroup_subsys *ss = subsys[i];
2269 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
2270 BUG_ON(!ss->create);
2271 BUG_ON(!ss->destroy);
2272 if (ss->subsys_id != i) {
2273 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
2274 ss->name, ss->subsys_id);
2279 cgroup_init_subsys(ss);
2285 * cgroup_init - register cgroup filesystem and /proc file, and
2286 * initialize any subsystems that didn't request early init.
2288 int __init cgroup_init(void)
2292 struct proc_dir_entry *entry;
2294 err = bdi_init(&cgroup_backing_dev_info);
2298 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2299 struct cgroup_subsys *ss = subsys[i];
2300 if (!ss->early_init)
2301 cgroup_init_subsys(ss);
2304 err = register_filesystem(&cgroup_fs_type);
2308 entry = create_proc_entry("cgroups", 0, NULL);
2310 entry->proc_fops = &proc_cgroupstats_operations;
2314 bdi_destroy(&cgroup_backing_dev_info);
2320 * proc_cgroup_show()
2321 * - Print task's cgroup paths into seq_file, one line for each hierarchy
2322 * - Used for /proc/<pid>/cgroup.
2323 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
2324 * doesn't really matter if tsk->cgroup changes after we read it,
2325 * and we take cgroup_mutex, keeping attach_task() from changing it
2326 * anyway. No need to check that tsk->cgroup != NULL, thanks to
2327 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
2328 * cgroup to top_cgroup.
2331 /* TODO: Use a proper seq_file iterator */
2332 static int proc_cgroup_show(struct seq_file *m, void *v)
2335 struct task_struct *tsk;
2338 struct cgroupfs_root *root;
2341 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2347 tsk = get_pid_task(pid, PIDTYPE_PID);
2353 mutex_lock(&cgroup_mutex);
2355 for_each_root(root) {
2356 struct cgroup_subsys *ss;
2357 struct cgroup *cgrp;
2361 /* Skip this hierarchy if it has no active subsystems */
2362 if (!root->actual_subsys_bits)
2364 for_each_subsys(root, ss)
2365 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
2367 get_first_subsys(&root->top_cgroup, NULL, &subsys_id);
2368 cgrp = task_cgroup(tsk, subsys_id);
2369 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
2377 mutex_unlock(&cgroup_mutex);
2378 put_task_struct(tsk);
2385 static int cgroup_open(struct inode *inode, struct file *file)
2387 struct pid *pid = PROC_I(inode)->pid;
2388 return single_open(file, proc_cgroup_show, pid);
2391 struct file_operations proc_cgroup_operations = {
2392 .open = cgroup_open,
2394 .llseek = seq_lseek,
2395 .release = single_release,
2398 /* Display information about each subsystem and each hierarchy */
2399 static int proc_cgroupstats_show(struct seq_file *m, void *v)
2403 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\n");
2404 mutex_lock(&cgroup_mutex);
2405 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2406 struct cgroup_subsys *ss = subsys[i];
2407 seq_printf(m, "%s\t%lu\t%d\n",
2408 ss->name, ss->root->subsys_bits,
2409 ss->root->number_of_cgroups);
2411 mutex_unlock(&cgroup_mutex);
2415 static int cgroupstats_open(struct inode *inode, struct file *file)
2417 return single_open(file, proc_cgroupstats_show, 0);
2420 static struct file_operations proc_cgroupstats_operations = {
2421 .open = cgroupstats_open,
2423 .llseek = seq_lseek,
2424 .release = single_release,
2428 * cgroup_fork - attach newly forked task to its parents cgroup.
2429 * @tsk: pointer to task_struct of forking parent process.
2431 * Description: A task inherits its parent's cgroup at fork().
2433 * A pointer to the shared css_set was automatically copied in
2434 * fork.c by dup_task_struct(). However, we ignore that copy, since
2435 * it was not made under the protection of RCU or cgroup_mutex, so
2436 * might no longer be a valid cgroup pointer. attach_task() might
2437 * have already changed current->cgroups, allowing the previously
2438 * referenced cgroup group to be removed and freed.
2440 * At the point that cgroup_fork() is called, 'current' is the parent
2441 * task, and the passed argument 'child' points to the child task.
2443 void cgroup_fork(struct task_struct *child)
2446 child->cgroups = current->cgroups;
2447 get_css_set(child->cgroups);
2448 task_unlock(current);
2449 INIT_LIST_HEAD(&child->cg_list);
2453 * cgroup_fork_callbacks - called on a new task very soon before
2454 * adding it to the tasklist. No need to take any locks since no-one
2455 * can be operating on this task
2457 void cgroup_fork_callbacks(struct task_struct *child)
2459 if (need_forkexit_callback) {
2461 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2462 struct cgroup_subsys *ss = subsys[i];
2464 ss->fork(ss, child);
2470 * cgroup_post_fork - called on a new task after adding it to the
2471 * task list. Adds the task to the list running through its css_set
2472 * if necessary. Has to be after the task is visible on the task list
2473 * in case we race with the first call to cgroup_iter_start() - to
2474 * guarantee that the new task ends up on its list. */
2475 void cgroup_post_fork(struct task_struct *child)
2477 if (use_task_css_set_links) {
2478 write_lock(&css_set_lock);
2479 if (list_empty(&child->cg_list))
2480 list_add(&child->cg_list, &child->cgroups->tasks);
2481 write_unlock(&css_set_lock);
2485 * cgroup_exit - detach cgroup from exiting task
2486 * @tsk: pointer to task_struct of exiting process
2488 * Description: Detach cgroup from @tsk and release it.
2490 * Note that cgroups marked notify_on_release force every task in
2491 * them to take the global cgroup_mutex mutex when exiting.
2492 * This could impact scaling on very large systems. Be reluctant to
2493 * use notify_on_release cgroups where very high task exit scaling
2494 * is required on large systems.
2496 * the_top_cgroup_hack:
2498 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
2500 * We call cgroup_exit() while the task is still competent to
2501 * handle notify_on_release(), then leave the task attached to the
2502 * root cgroup in each hierarchy for the remainder of its exit.
2504 * To do this properly, we would increment the reference count on
2505 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
2506 * code we would add a second cgroup function call, to drop that
2507 * reference. This would just create an unnecessary hot spot on
2508 * the top_cgroup reference count, to no avail.
2510 * Normally, holding a reference to a cgroup without bumping its
2511 * count is unsafe. The cgroup could go away, or someone could
2512 * attach us to a different cgroup, decrementing the count on
2513 * the first cgroup that we never incremented. But in this case,
2514 * top_cgroup isn't going away, and either task has PF_EXITING set,
2515 * which wards off any attach_task() attempts, or task is a failed
2516 * fork, never visible to attach_task.
2519 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
2524 if (run_callbacks && need_forkexit_callback) {
2525 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2526 struct cgroup_subsys *ss = subsys[i];
2533 * Unlink from the css_set task list if necessary.
2534 * Optimistically check cg_list before taking
2537 if (!list_empty(&tsk->cg_list)) {
2538 write_lock(&css_set_lock);
2539 if (!list_empty(&tsk->cg_list))
2540 list_del(&tsk->cg_list);
2541 write_unlock(&css_set_lock);
2544 /* Reassign the task to the init_css_set. */
2547 tsk->cgroups = &init_css_set;
2550 put_css_set_taskexit(cg);
2554 * cgroup_clone - duplicate the current cgroup in the hierarchy
2555 * that the given subsystem is attached to, and move this task into
2558 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys)
2560 struct dentry *dentry;
2562 char nodename[MAX_CGROUP_TYPE_NAMELEN];
2563 struct cgroup *parent, *child;
2564 struct inode *inode;
2566 struct cgroupfs_root *root;
2567 struct cgroup_subsys *ss;
2569 /* We shouldn't be called by an unregistered subsystem */
2570 BUG_ON(!subsys->active);
2572 /* First figure out what hierarchy and cgroup we're dealing
2573 * with, and pin them so we can drop cgroup_mutex */
2574 mutex_lock(&cgroup_mutex);
2576 root = subsys->root;
2577 if (root == &rootnode) {
2579 "Not cloning cgroup for unused subsystem %s\n",
2581 mutex_unlock(&cgroup_mutex);
2585 parent = task_cgroup(tsk, subsys->subsys_id);
2587 snprintf(nodename, MAX_CGROUP_TYPE_NAMELEN, "node_%d", tsk->pid);
2589 /* Pin the hierarchy */
2590 atomic_inc(&parent->root->sb->s_active);
2592 /* Keep the cgroup alive */
2594 mutex_unlock(&cgroup_mutex);
2596 /* Now do the VFS work to create a cgroup */
2597 inode = parent->dentry->d_inode;
2599 /* Hold the parent directory mutex across this operation to
2600 * stop anyone else deleting the new cgroup */
2601 mutex_lock(&inode->i_mutex);
2602 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
2603 if (IS_ERR(dentry)) {
2605 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
2607 ret = PTR_ERR(dentry);
2611 /* Create the cgroup directory, which also creates the cgroup */
2612 ret = vfs_mkdir(inode, dentry, S_IFDIR | 0755);
2613 child = __d_cgrp(dentry);
2617 "Failed to create cgroup %s: %d\n", nodename,
2624 "Couldn't find new cgroup %s\n", nodename);
2629 /* The cgroup now exists. Retake cgroup_mutex and check
2630 * that we're still in the same state that we thought we
2632 mutex_lock(&cgroup_mutex);
2633 if ((root != subsys->root) ||
2634 (parent != task_cgroup(tsk, subsys->subsys_id))) {
2635 /* Aargh, we raced ... */
2636 mutex_unlock(&inode->i_mutex);
2639 deactivate_super(parent->root->sb);
2640 /* The cgroup is still accessible in the VFS, but
2641 * we're not going to try to rmdir() it at this
2644 "Race in cgroup_clone() - leaking cgroup %s\n",
2649 /* do any required auto-setup */
2650 for_each_subsys(root, ss) {
2652 ss->post_clone(ss, child);
2655 /* All seems fine. Finish by moving the task into the new cgroup */
2656 ret = attach_task(child, tsk);
2657 mutex_unlock(&cgroup_mutex);
2660 mutex_unlock(&inode->i_mutex);
2662 mutex_lock(&cgroup_mutex);
2664 mutex_unlock(&cgroup_mutex);
2665 deactivate_super(parent->root->sb);
2670 * See if "cgrp" is a descendant of the current task's cgroup in
2671 * the appropriate hierarchy
2673 * If we are sending in dummytop, then presumably we are creating
2674 * the top cgroup in the subsystem.
2676 * Called only by the ns (nsproxy) cgroup.
2678 int cgroup_is_descendant(const struct cgroup *cgrp)
2681 struct cgroup *target;
2684 if (cgrp == dummytop)
2687 get_first_subsys(cgrp, NULL, &subsys_id);
2688 target = task_cgroup(current, subsys_id);
2689 while (cgrp != target && cgrp!= cgrp->top_cgroup)
2690 cgrp = cgrp->parent;
2691 ret = (cgrp == target);
2695 static void check_for_release(struct cgroup *cgrp)
2697 /* All of these checks rely on RCU to keep the cgroup
2698 * structure alive */
2699 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
2700 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
2701 /* Control Group is currently removeable. If it's not
2702 * already queued for a userspace notification, queue
2704 int need_schedule_work = 0;
2705 spin_lock(&release_list_lock);
2706 if (!cgroup_is_removed(cgrp) &&
2707 list_empty(&cgrp->release_list)) {
2708 list_add(&cgrp->release_list, &release_list);
2709 need_schedule_work = 1;
2711 spin_unlock(&release_list_lock);
2712 if (need_schedule_work)
2713 schedule_work(&release_agent_work);
2717 void __css_put(struct cgroup_subsys_state *css)
2719 struct cgroup *cgrp = css->cgroup;
2721 if (atomic_dec_and_test(&css->refcnt) && notify_on_release(cgrp)) {
2722 set_bit(CGRP_RELEASABLE, &cgrp->flags);
2723 check_for_release(cgrp);
2729 * Notify userspace when a cgroup is released, by running the
2730 * configured release agent with the name of the cgroup (path
2731 * relative to the root of cgroup file system) as the argument.
2733 * Most likely, this user command will try to rmdir this cgroup.
2735 * This races with the possibility that some other task will be
2736 * attached to this cgroup before it is removed, or that some other
2737 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
2738 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
2739 * unused, and this cgroup will be reprieved from its death sentence,
2740 * to continue to serve a useful existence. Next time it's released,
2741 * we will get notified again, if it still has 'notify_on_release' set.
2743 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
2744 * means only wait until the task is successfully execve()'d. The
2745 * separate release agent task is forked by call_usermodehelper(),
2746 * then control in this thread returns here, without waiting for the
2747 * release agent task. We don't bother to wait because the caller of
2748 * this routine has no use for the exit status of the release agent
2749 * task, so no sense holding our caller up for that.
2753 static void cgroup_release_agent(struct work_struct *work)
2755 BUG_ON(work != &release_agent_work);
2756 mutex_lock(&cgroup_mutex);
2757 spin_lock(&release_list_lock);
2758 while (!list_empty(&release_list)) {
2759 char *argv[3], *envp[3];
2762 struct cgroup *cgrp = list_entry(release_list.next,
2765 list_del_init(&cgrp->release_list);
2766 spin_unlock(&release_list_lock);
2767 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2769 spin_lock(&release_list_lock);
2773 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) {
2775 spin_lock(&release_list_lock);
2780 argv[i++] = cgrp->root->release_agent_path;
2781 argv[i++] = (char *)pathbuf;
2785 /* minimal command environment */
2786 envp[i++] = "HOME=/";
2787 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
2790 /* Drop the lock while we invoke the usermode helper,
2791 * since the exec could involve hitting disk and hence
2792 * be a slow process */
2793 mutex_unlock(&cgroup_mutex);
2794 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
2796 mutex_lock(&cgroup_mutex);
2797 spin_lock(&release_list_lock);
2799 spin_unlock(&release_list_lock);
2800 mutex_unlock(&cgroup_mutex);