2 * Generic process-grouping system.
4 * Based originally on the cpuset system, extracted by Paul Menage
5 * Copyright (C) 2006 Google, Inc
7 * Copyright notices from the original cpuset code:
8 * --------------------------------------------------
9 * Copyright (C) 2003 BULL SA.
10 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
12 * Portions derived from Patrick Mochel's sysfs code.
13 * sysfs is Copyright (c) 2001-3 Patrick Mochel
15 * 2003-10-10 Written by Simon Derr.
16 * 2003-10-22 Updates by Stephen Hemminger.
17 * 2004 May-July Rework by Paul Jackson.
18 * ---------------------------------------------------
20 * This file is subject to the terms and conditions of the GNU General Public
21 * License. See the file COPYING in the main directory of the Linux
22 * distribution for more details.
25 #include <linux/cgroup.h>
26 #include <linux/errno.h>
28 #include <linux/kernel.h>
29 #include <linux/list.h>
31 #include <linux/mutex.h>
32 #include <linux/mount.h>
33 #include <linux/pagemap.h>
34 #include <linux/proc_fs.h>
35 #include <linux/rcupdate.h>
36 #include <linux/sched.h>
37 #include <linux/backing-dev.h>
38 #include <linux/seq_file.h>
39 #include <linux/slab.h>
40 #include <linux/magic.h>
41 #include <linux/spinlock.h>
42 #include <linux/string.h>
43 #include <linux/sort.h>
44 #include <linux/kmod.h>
45 #include <linux/delayacct.h>
46 #include <linux/cgroupstats.h>
47 #include <linux/hash.h>
49 #include <asm/atomic.h>
51 static DEFINE_MUTEX(cgroup_mutex);
53 /* Generate an array of cgroup subsystem pointers */
54 #define SUBSYS(_x) &_x ## _subsys,
56 static struct cgroup_subsys *subsys[] = {
57 #include <linux/cgroup_subsys.h>
61 * A cgroupfs_root represents the root of a cgroup hierarchy,
62 * and may be associated with a superblock to form an active
65 struct cgroupfs_root {
66 struct super_block *sb;
69 * The bitmask of subsystems intended to be attached to this
72 unsigned long subsys_bits;
74 /* The bitmask of subsystems currently attached to this hierarchy */
75 unsigned long actual_subsys_bits;
77 /* A list running through the attached subsystems */
78 struct list_head subsys_list;
80 /* The root cgroup for this hierarchy */
81 struct cgroup top_cgroup;
83 /* Tracks how many cgroups are currently defined in hierarchy.*/
84 int number_of_cgroups;
86 /* A list running through the mounted hierarchies */
87 struct list_head root_list;
89 /* Hierarchy-specific flags */
92 /* The path to use for release notifications. No locking
93 * between setting and use - so if userspace updates this
94 * while child cgroups exist, you could miss a
95 * notification. We ensure that it's always a valid
96 * NUL-terminated string */
97 char release_agent_path[PATH_MAX];
102 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
103 * subsystems that are otherwise unattached - it never has more than a
104 * single cgroup, and all tasks are part of that cgroup.
106 static struct cgroupfs_root rootnode;
108 /* The list of hierarchy roots */
110 static LIST_HEAD(roots);
111 static int root_count;
113 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
114 #define dummytop (&rootnode.top_cgroup)
116 /* This flag indicates whether tasks in the fork and exit paths should
117 * check for fork/exit handlers to call. This avoids us having to do
118 * extra work in the fork/exit path if none of the subsystems need to
121 static int need_forkexit_callback;
123 /* convenient tests for these bits */
124 inline int cgroup_is_removed(const struct cgroup *cgrp)
126 return test_bit(CGRP_REMOVED, &cgrp->flags);
129 /* bits in struct cgroupfs_root flags field */
131 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
134 static int cgroup_is_releasable(const struct cgroup *cgrp)
137 (1 << CGRP_RELEASABLE) |
138 (1 << CGRP_NOTIFY_ON_RELEASE);
139 return (cgrp->flags & bits) == bits;
142 static int notify_on_release(const struct cgroup *cgrp)
144 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
148 * for_each_subsys() allows you to iterate on each subsystem attached to
149 * an active hierarchy
151 #define for_each_subsys(_root, _ss) \
152 list_for_each_entry(_ss, &_root->subsys_list, sibling)
154 /* for_each_root() allows you to iterate across the active hierarchies */
155 #define for_each_root(_root) \
156 list_for_each_entry(_root, &roots, root_list)
158 /* the list of cgroups eligible for automatic release. Protected by
159 * release_list_lock */
160 static LIST_HEAD(release_list);
161 static DEFINE_SPINLOCK(release_list_lock);
162 static void cgroup_release_agent(struct work_struct *work);
163 static DECLARE_WORK(release_agent_work, cgroup_release_agent);
164 static void check_for_release(struct cgroup *cgrp);
166 /* Link structure for associating css_set objects with cgroups */
167 struct cg_cgroup_link {
169 * List running through cg_cgroup_links associated with a
170 * cgroup, anchored on cgroup->css_sets
172 struct list_head cgrp_link_list;
174 * List running through cg_cgroup_links pointing at a
175 * single css_set object, anchored on css_set->cg_links
177 struct list_head cg_link_list;
181 /* The default css_set - used by init and its children prior to any
182 * hierarchies being mounted. It contains a pointer to the root state
183 * for each subsystem. Also used to anchor the list of css_sets. Not
184 * reference-counted, to improve performance when child cgroups
185 * haven't been created.
188 static struct css_set init_css_set;
189 static struct cg_cgroup_link init_css_set_link;
191 /* css_set_lock protects the list of css_set objects, and the
192 * chain of tasks off each css_set. Nests outside task->alloc_lock
193 * due to cgroup_iter_start() */
194 static DEFINE_RWLOCK(css_set_lock);
195 static int css_set_count;
197 /* hash table for cgroup groups. This improves the performance to
198 * find an existing css_set */
199 #define CSS_SET_HASH_BITS 7
200 #define CSS_SET_TABLE_SIZE (1 << CSS_SET_HASH_BITS)
201 static struct hlist_head css_set_table[CSS_SET_TABLE_SIZE];
203 static struct hlist_head *css_set_hash(struct cgroup_subsys_state *css[])
207 unsigned long tmp = 0UL;
209 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++)
210 tmp += (unsigned long)css[i];
211 tmp = (tmp >> 16) ^ tmp;
213 index = hash_long(tmp, CSS_SET_HASH_BITS);
215 return &css_set_table[index];
218 /* We don't maintain the lists running through each css_set to its
219 * task until after the first call to cgroup_iter_start(). This
220 * reduces the fork()/exit() overhead for people who have cgroups
221 * compiled into their kernel but not actually in use */
222 static int use_task_css_set_links;
224 /* When we create or destroy a css_set, the operation simply
225 * takes/releases a reference count on all the cgroups referenced
226 * by subsystems in this css_set. This can end up multiple-counting
227 * some cgroups, but that's OK - the ref-count is just a
228 * busy/not-busy indicator; ensuring that we only count each cgroup
229 * once would require taking a global lock to ensure that no
230 * subsystems moved between hierarchies while we were doing so.
232 * Possible TODO: decide at boot time based on the number of
233 * registered subsystems and the number of CPUs or NUMA nodes whether
234 * it's better for performance to ref-count every subsystem, or to
235 * take a global lock and only add one ref count to each hierarchy.
239 * unlink a css_set from the list and free it
241 static void unlink_css_set(struct css_set *cg)
243 write_lock(&css_set_lock);
244 hlist_del(&cg->hlist);
246 while (!list_empty(&cg->cg_links)) {
247 struct cg_cgroup_link *link;
248 link = list_entry(cg->cg_links.next,
249 struct cg_cgroup_link, cg_link_list);
250 list_del(&link->cg_link_list);
251 list_del(&link->cgrp_link_list);
254 write_unlock(&css_set_lock);
257 static void __release_css_set(struct kref *k, int taskexit)
260 struct css_set *cg = container_of(k, struct css_set, ref);
265 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
266 struct cgroup *cgrp = cg->subsys[i]->cgroup;
267 if (atomic_dec_and_test(&cgrp->count) &&
268 notify_on_release(cgrp)) {
270 set_bit(CGRP_RELEASABLE, &cgrp->flags);
271 check_for_release(cgrp);
278 static void release_css_set(struct kref *k)
280 __release_css_set(k, 0);
283 static void release_css_set_taskexit(struct kref *k)
285 __release_css_set(k, 1);
289 * refcounted get/put for css_set objects
291 static inline void get_css_set(struct css_set *cg)
296 static inline void put_css_set(struct css_set *cg)
298 kref_put(&cg->ref, release_css_set);
301 static inline void put_css_set_taskexit(struct css_set *cg)
303 kref_put(&cg->ref, release_css_set_taskexit);
307 * find_existing_css_set() is a helper for
308 * find_css_set(), and checks to see whether an existing
309 * css_set is suitable.
311 * oldcg: the cgroup group that we're using before the cgroup
314 * cgrp: the cgroup that we're moving into
316 * template: location in which to build the desired set of subsystem
317 * state objects for the new cgroup group
319 static struct css_set *find_existing_css_set(
320 struct css_set *oldcg,
322 struct cgroup_subsys_state *template[])
325 struct cgroupfs_root *root = cgrp->root;
326 struct hlist_head *hhead;
327 struct hlist_node *node;
330 /* Built the set of subsystem state objects that we want to
331 * see in the new css_set */
332 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
333 if (root->subsys_bits & (1UL << i)) {
334 /* Subsystem is in this hierarchy. So we want
335 * the subsystem state from the new
337 template[i] = cgrp->subsys[i];
339 /* Subsystem is not in this hierarchy, so we
340 * don't want to change the subsystem state */
341 template[i] = oldcg->subsys[i];
345 hhead = css_set_hash(template);
346 hlist_for_each_entry(cg, node, hhead, hlist) {
347 if (!memcmp(template, cg->subsys, sizeof(cg->subsys))) {
348 /* All subsystems matched */
353 /* No existing cgroup group matched */
358 * allocate_cg_links() allocates "count" cg_cgroup_link structures
359 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
360 * success or a negative error
362 static int allocate_cg_links(int count, struct list_head *tmp)
364 struct cg_cgroup_link *link;
367 for (i = 0; i < count; i++) {
368 link = kmalloc(sizeof(*link), GFP_KERNEL);
370 while (!list_empty(tmp)) {
371 link = list_entry(tmp->next,
372 struct cg_cgroup_link,
374 list_del(&link->cgrp_link_list);
379 list_add(&link->cgrp_link_list, tmp);
384 static void free_cg_links(struct list_head *tmp)
386 while (!list_empty(tmp)) {
387 struct cg_cgroup_link *link;
388 link = list_entry(tmp->next,
389 struct cg_cgroup_link,
391 list_del(&link->cgrp_link_list);
397 * find_css_set() takes an existing cgroup group and a
398 * cgroup object, and returns a css_set object that's
399 * equivalent to the old group, but with the given cgroup
400 * substituted into the appropriate hierarchy. Must be called with
403 static struct css_set *find_css_set(
404 struct css_set *oldcg, struct cgroup *cgrp)
407 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
410 struct list_head tmp_cg_links;
411 struct cg_cgroup_link *link;
413 struct hlist_head *hhead;
415 /* First see if we already have a cgroup group that matches
417 write_lock(&css_set_lock);
418 res = find_existing_css_set(oldcg, cgrp, template);
421 write_unlock(&css_set_lock);
426 res = kmalloc(sizeof(*res), GFP_KERNEL);
430 /* Allocate all the cg_cgroup_link objects that we'll need */
431 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
436 kref_init(&res->ref);
437 INIT_LIST_HEAD(&res->cg_links);
438 INIT_LIST_HEAD(&res->tasks);
439 INIT_HLIST_NODE(&res->hlist);
441 /* Copy the set of subsystem state objects generated in
442 * find_existing_css_set() */
443 memcpy(res->subsys, template, sizeof(res->subsys));
445 write_lock(&css_set_lock);
446 /* Add reference counts and links from the new css_set. */
447 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
448 struct cgroup *cgrp = res->subsys[i]->cgroup;
449 struct cgroup_subsys *ss = subsys[i];
450 atomic_inc(&cgrp->count);
452 * We want to add a link once per cgroup, so we
453 * only do it for the first subsystem in each
456 if (ss->root->subsys_list.next == &ss->sibling) {
457 BUG_ON(list_empty(&tmp_cg_links));
458 link = list_entry(tmp_cg_links.next,
459 struct cg_cgroup_link,
461 list_del(&link->cgrp_link_list);
462 list_add(&link->cgrp_link_list, &cgrp->css_sets);
464 list_add(&link->cg_link_list, &res->cg_links);
467 if (list_empty(&rootnode.subsys_list)) {
468 link = list_entry(tmp_cg_links.next,
469 struct cg_cgroup_link,
471 list_del(&link->cgrp_link_list);
472 list_add(&link->cgrp_link_list, &dummytop->css_sets);
474 list_add(&link->cg_link_list, &res->cg_links);
477 BUG_ON(!list_empty(&tmp_cg_links));
481 /* Add this cgroup group to the hash table */
482 hhead = css_set_hash(res->subsys);
483 hlist_add_head(&res->hlist, hhead);
485 write_unlock(&css_set_lock);
491 * There is one global cgroup mutex. We also require taking
492 * task_lock() when dereferencing a task's cgroup subsys pointers.
493 * See "The task_lock() exception", at the end of this comment.
495 * A task must hold cgroup_mutex to modify cgroups.
497 * Any task can increment and decrement the count field without lock.
498 * So in general, code holding cgroup_mutex can't rely on the count
499 * field not changing. However, if the count goes to zero, then only
500 * cgroup_attach_task() can increment it again. Because a count of zero
501 * means that no tasks are currently attached, therefore there is no
502 * way a task attached to that cgroup can fork (the other way to
503 * increment the count). So code holding cgroup_mutex can safely
504 * assume that if the count is zero, it will stay zero. Similarly, if
505 * a task holds cgroup_mutex on a cgroup with zero count, it
506 * knows that the cgroup won't be removed, as cgroup_rmdir()
509 * The cgroup_common_file_write handler for operations that modify
510 * the cgroup hierarchy holds cgroup_mutex across the entire operation,
511 * single threading all such cgroup modifications across the system.
513 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
514 * (usually) take cgroup_mutex. These are the two most performance
515 * critical pieces of code here. The exception occurs on cgroup_exit(),
516 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
517 * is taken, and if the cgroup count is zero, a usermode call made
518 * to the release agent with the name of the cgroup (path relative to
519 * the root of cgroup file system) as the argument.
521 * A cgroup can only be deleted if both its 'count' of using tasks
522 * is zero, and its list of 'children' cgroups is empty. Since all
523 * tasks in the system use _some_ cgroup, and since there is always at
524 * least one task in the system (init, pid == 1), therefore, top_cgroup
525 * always has either children cgroups and/or using tasks. So we don't
526 * need a special hack to ensure that top_cgroup cannot be deleted.
528 * The task_lock() exception
530 * The need for this exception arises from the action of
531 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
532 * another. It does so using cgroup_mutex, however there are
533 * several performance critical places that need to reference
534 * task->cgroup without the expense of grabbing a system global
535 * mutex. Therefore except as noted below, when dereferencing or, as
536 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
537 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
538 * the task_struct routinely used for such matters.
540 * P.S. One more locking exception. RCU is used to guard the
541 * update of a tasks cgroup pointer by cgroup_attach_task()
545 * cgroup_lock - lock out any changes to cgroup structures
548 void cgroup_lock(void)
550 mutex_lock(&cgroup_mutex);
554 * cgroup_unlock - release lock on cgroup changes
556 * Undo the lock taken in a previous cgroup_lock() call.
558 void cgroup_unlock(void)
560 mutex_unlock(&cgroup_mutex);
564 * A couple of forward declarations required, due to cyclic reference loop:
565 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
566 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
570 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
571 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
572 static int cgroup_populate_dir(struct cgroup *cgrp);
573 static struct inode_operations cgroup_dir_inode_operations;
574 static struct file_operations proc_cgroupstats_operations;
576 static struct backing_dev_info cgroup_backing_dev_info = {
577 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
580 static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
582 struct inode *inode = new_inode(sb);
585 inode->i_mode = mode;
586 inode->i_uid = current->fsuid;
587 inode->i_gid = current->fsgid;
589 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
590 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
596 * Call subsys's pre_destroy handler.
597 * This is called before css refcnt check.
599 static void cgroup_call_pre_destroy(struct cgroup *cgrp)
601 struct cgroup_subsys *ss;
602 for_each_subsys(cgrp->root, ss)
603 if (ss->pre_destroy && cgrp->subsys[ss->subsys_id])
604 ss->pre_destroy(ss, cgrp);
608 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
610 /* is dentry a directory ? if so, kfree() associated cgroup */
611 if (S_ISDIR(inode->i_mode)) {
612 struct cgroup *cgrp = dentry->d_fsdata;
613 struct cgroup_subsys *ss;
614 BUG_ON(!(cgroup_is_removed(cgrp)));
615 /* It's possible for external users to be holding css
616 * reference counts on a cgroup; css_put() needs to
617 * be able to access the cgroup after decrementing
618 * the reference count in order to know if it needs to
619 * queue the cgroup to be handled by the release
623 mutex_lock(&cgroup_mutex);
625 * Release the subsystem state objects.
627 for_each_subsys(cgrp->root, ss) {
628 if (cgrp->subsys[ss->subsys_id])
629 ss->destroy(ss, cgrp);
632 cgrp->root->number_of_cgroups--;
633 mutex_unlock(&cgroup_mutex);
635 /* Drop the active superblock reference that we took when we
636 * created the cgroup */
637 deactivate_super(cgrp->root->sb);
644 static void remove_dir(struct dentry *d)
646 struct dentry *parent = dget(d->d_parent);
649 simple_rmdir(parent->d_inode, d);
653 static void cgroup_clear_directory(struct dentry *dentry)
655 struct list_head *node;
657 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
658 spin_lock(&dcache_lock);
659 node = dentry->d_subdirs.next;
660 while (node != &dentry->d_subdirs) {
661 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
664 /* This should never be called on a cgroup
665 * directory with child cgroups */
666 BUG_ON(d->d_inode->i_mode & S_IFDIR);
668 spin_unlock(&dcache_lock);
670 simple_unlink(dentry->d_inode, d);
672 spin_lock(&dcache_lock);
674 node = dentry->d_subdirs.next;
676 spin_unlock(&dcache_lock);
680 * NOTE : the dentry must have been dget()'ed
682 static void cgroup_d_remove_dir(struct dentry *dentry)
684 cgroup_clear_directory(dentry);
686 spin_lock(&dcache_lock);
687 list_del_init(&dentry->d_u.d_child);
688 spin_unlock(&dcache_lock);
692 static int rebind_subsystems(struct cgroupfs_root *root,
693 unsigned long final_bits)
695 unsigned long added_bits, removed_bits;
696 struct cgroup *cgrp = &root->top_cgroup;
699 removed_bits = root->actual_subsys_bits & ~final_bits;
700 added_bits = final_bits & ~root->actual_subsys_bits;
701 /* Check that any added subsystems are currently free */
702 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
703 unsigned long bit = 1UL << i;
704 struct cgroup_subsys *ss = subsys[i];
705 if (!(bit & added_bits))
707 if (ss->root != &rootnode) {
708 /* Subsystem isn't free */
713 /* Currently we don't handle adding/removing subsystems when
714 * any child cgroups exist. This is theoretically supportable
715 * but involves complex error handling, so it's being left until
717 if (!list_empty(&cgrp->children))
720 /* Process each subsystem */
721 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
722 struct cgroup_subsys *ss = subsys[i];
723 unsigned long bit = 1UL << i;
724 if (bit & added_bits) {
725 /* We're binding this subsystem to this hierarchy */
726 BUG_ON(cgrp->subsys[i]);
727 BUG_ON(!dummytop->subsys[i]);
728 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
729 cgrp->subsys[i] = dummytop->subsys[i];
730 cgrp->subsys[i]->cgroup = cgrp;
731 list_add(&ss->sibling, &root->subsys_list);
732 rcu_assign_pointer(ss->root, root);
736 } else if (bit & removed_bits) {
737 /* We're removing this subsystem */
738 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
739 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
741 ss->bind(ss, dummytop);
742 dummytop->subsys[i]->cgroup = dummytop;
743 cgrp->subsys[i] = NULL;
744 rcu_assign_pointer(subsys[i]->root, &rootnode);
745 list_del(&ss->sibling);
746 } else if (bit & final_bits) {
747 /* Subsystem state should already exist */
748 BUG_ON(!cgrp->subsys[i]);
750 /* Subsystem state shouldn't exist */
751 BUG_ON(cgrp->subsys[i]);
754 root->subsys_bits = root->actual_subsys_bits = final_bits;
760 static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
762 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
763 struct cgroup_subsys *ss;
765 mutex_lock(&cgroup_mutex);
766 for_each_subsys(root, ss)
767 seq_printf(seq, ",%s", ss->name);
768 if (test_bit(ROOT_NOPREFIX, &root->flags))
769 seq_puts(seq, ",noprefix");
770 if (strlen(root->release_agent_path))
771 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
772 mutex_unlock(&cgroup_mutex);
776 struct cgroup_sb_opts {
777 unsigned long subsys_bits;
782 /* Convert a hierarchy specifier into a bitmask of subsystems and
784 static int parse_cgroupfs_options(char *data,
785 struct cgroup_sb_opts *opts)
787 char *token, *o = data ?: "all";
789 opts->subsys_bits = 0;
791 opts->release_agent = NULL;
793 while ((token = strsep(&o, ",")) != NULL) {
796 if (!strcmp(token, "all")) {
797 /* Add all non-disabled subsystems */
799 opts->subsys_bits = 0;
800 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
801 struct cgroup_subsys *ss = subsys[i];
803 opts->subsys_bits |= 1ul << i;
805 } else if (!strcmp(token, "noprefix")) {
806 set_bit(ROOT_NOPREFIX, &opts->flags);
807 } else if (!strncmp(token, "release_agent=", 14)) {
808 /* Specifying two release agents is forbidden */
809 if (opts->release_agent)
811 opts->release_agent = kzalloc(PATH_MAX, GFP_KERNEL);
812 if (!opts->release_agent)
814 strncpy(opts->release_agent, token + 14, PATH_MAX - 1);
815 opts->release_agent[PATH_MAX - 1] = 0;
817 struct cgroup_subsys *ss;
819 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
821 if (!strcmp(token, ss->name)) {
823 set_bit(i, &opts->subsys_bits);
827 if (i == CGROUP_SUBSYS_COUNT)
832 /* We can't have an empty hierarchy */
833 if (!opts->subsys_bits)
839 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
842 struct cgroupfs_root *root = sb->s_fs_info;
843 struct cgroup *cgrp = &root->top_cgroup;
844 struct cgroup_sb_opts opts;
846 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
847 mutex_lock(&cgroup_mutex);
849 /* See what subsystems are wanted */
850 ret = parse_cgroupfs_options(data, &opts);
854 /* Don't allow flags to change at remount */
855 if (opts.flags != root->flags) {
860 ret = rebind_subsystems(root, opts.subsys_bits);
862 /* (re)populate subsystem files */
864 cgroup_populate_dir(cgrp);
866 if (opts.release_agent)
867 strcpy(root->release_agent_path, opts.release_agent);
869 if (opts.release_agent)
870 kfree(opts.release_agent);
871 mutex_unlock(&cgroup_mutex);
872 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
876 static struct super_operations cgroup_ops = {
877 .statfs = simple_statfs,
878 .drop_inode = generic_delete_inode,
879 .show_options = cgroup_show_options,
880 .remount_fs = cgroup_remount,
883 static void init_cgroup_root(struct cgroupfs_root *root)
885 struct cgroup *cgrp = &root->top_cgroup;
886 INIT_LIST_HEAD(&root->subsys_list);
887 INIT_LIST_HEAD(&root->root_list);
888 root->number_of_cgroups = 1;
890 cgrp->top_cgroup = cgrp;
891 INIT_LIST_HEAD(&cgrp->sibling);
892 INIT_LIST_HEAD(&cgrp->children);
893 INIT_LIST_HEAD(&cgrp->css_sets);
894 INIT_LIST_HEAD(&cgrp->release_list);
897 static int cgroup_test_super(struct super_block *sb, void *data)
899 struct cgroupfs_root *new = data;
900 struct cgroupfs_root *root = sb->s_fs_info;
902 /* First check subsystems */
903 if (new->subsys_bits != root->subsys_bits)
906 /* Next check flags */
907 if (new->flags != root->flags)
913 static int cgroup_set_super(struct super_block *sb, void *data)
916 struct cgroupfs_root *root = data;
918 ret = set_anon_super(sb, NULL);
922 sb->s_fs_info = root;
925 sb->s_blocksize = PAGE_CACHE_SIZE;
926 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
927 sb->s_magic = CGROUP_SUPER_MAGIC;
928 sb->s_op = &cgroup_ops;
933 static int cgroup_get_rootdir(struct super_block *sb)
935 struct inode *inode =
936 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
937 struct dentry *dentry;
942 inode->i_fop = &simple_dir_operations;
943 inode->i_op = &cgroup_dir_inode_operations;
944 /* directories start off with i_nlink == 2 (for "." entry) */
946 dentry = d_alloc_root(inode);
955 static int cgroup_get_sb(struct file_system_type *fs_type,
956 int flags, const char *unused_dev_name,
957 void *data, struct vfsmount *mnt)
959 struct cgroup_sb_opts opts;
961 struct super_block *sb;
962 struct cgroupfs_root *root;
963 struct list_head tmp_cg_links;
964 INIT_LIST_HEAD(&tmp_cg_links);
966 /* First find the desired set of subsystems */
967 ret = parse_cgroupfs_options(data, &opts);
969 if (opts.release_agent)
970 kfree(opts.release_agent);
974 root = kzalloc(sizeof(*root), GFP_KERNEL);
976 if (opts.release_agent)
977 kfree(opts.release_agent);
981 init_cgroup_root(root);
982 root->subsys_bits = opts.subsys_bits;
983 root->flags = opts.flags;
984 if (opts.release_agent) {
985 strcpy(root->release_agent_path, opts.release_agent);
986 kfree(opts.release_agent);
989 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, root);
996 if (sb->s_fs_info != root) {
997 /* Reusing an existing superblock */
998 BUG_ON(sb->s_root == NULL);
1002 /* New superblock */
1003 struct cgroup *cgrp = &root->top_cgroup;
1004 struct inode *inode;
1007 BUG_ON(sb->s_root != NULL);
1009 ret = cgroup_get_rootdir(sb);
1011 goto drop_new_super;
1012 inode = sb->s_root->d_inode;
1014 mutex_lock(&inode->i_mutex);
1015 mutex_lock(&cgroup_mutex);
1018 * We're accessing css_set_count without locking
1019 * css_set_lock here, but that's OK - it can only be
1020 * increased by someone holding cgroup_lock, and
1021 * that's us. The worst that can happen is that we
1022 * have some link structures left over
1024 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1026 mutex_unlock(&cgroup_mutex);
1027 mutex_unlock(&inode->i_mutex);
1028 goto drop_new_super;
1031 ret = rebind_subsystems(root, root->subsys_bits);
1032 if (ret == -EBUSY) {
1033 mutex_unlock(&cgroup_mutex);
1034 mutex_unlock(&inode->i_mutex);
1035 goto drop_new_super;
1038 /* EBUSY should be the only error here */
1041 list_add(&root->root_list, &roots);
1044 sb->s_root->d_fsdata = &root->top_cgroup;
1045 root->top_cgroup.dentry = sb->s_root;
1047 /* Link the top cgroup in this hierarchy into all
1048 * the css_set objects */
1049 write_lock(&css_set_lock);
1050 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) {
1051 struct hlist_head *hhead = &css_set_table[i];
1052 struct hlist_node *node;
1055 hlist_for_each_entry(cg, node, hhead, hlist) {
1056 struct cg_cgroup_link *link;
1058 BUG_ON(list_empty(&tmp_cg_links));
1059 link = list_entry(tmp_cg_links.next,
1060 struct cg_cgroup_link,
1062 list_del(&link->cgrp_link_list);
1064 list_add(&link->cgrp_link_list,
1065 &root->top_cgroup.css_sets);
1066 list_add(&link->cg_link_list, &cg->cg_links);
1069 write_unlock(&css_set_lock);
1071 free_cg_links(&tmp_cg_links);
1073 BUG_ON(!list_empty(&cgrp->sibling));
1074 BUG_ON(!list_empty(&cgrp->children));
1075 BUG_ON(root->number_of_cgroups != 1);
1077 cgroup_populate_dir(cgrp);
1078 mutex_unlock(&inode->i_mutex);
1079 mutex_unlock(&cgroup_mutex);
1082 return simple_set_mnt(mnt, sb);
1085 up_write(&sb->s_umount);
1086 deactivate_super(sb);
1087 free_cg_links(&tmp_cg_links);
1091 static void cgroup_kill_sb(struct super_block *sb) {
1092 struct cgroupfs_root *root = sb->s_fs_info;
1093 struct cgroup *cgrp = &root->top_cgroup;
1098 BUG_ON(root->number_of_cgroups != 1);
1099 BUG_ON(!list_empty(&cgrp->children));
1100 BUG_ON(!list_empty(&cgrp->sibling));
1102 mutex_lock(&cgroup_mutex);
1104 /* Rebind all subsystems back to the default hierarchy */
1105 ret = rebind_subsystems(root, 0);
1106 /* Shouldn't be able to fail ... */
1110 * Release all the links from css_sets to this hierarchy's
1113 write_lock(&css_set_lock);
1114 while (!list_empty(&cgrp->css_sets)) {
1115 struct cg_cgroup_link *link;
1116 link = list_entry(cgrp->css_sets.next,
1117 struct cg_cgroup_link, cgrp_link_list);
1118 list_del(&link->cg_link_list);
1119 list_del(&link->cgrp_link_list);
1122 write_unlock(&css_set_lock);
1124 if (!list_empty(&root->root_list)) {
1125 list_del(&root->root_list);
1128 mutex_unlock(&cgroup_mutex);
1131 kill_litter_super(sb);
1134 static struct file_system_type cgroup_fs_type = {
1136 .get_sb = cgroup_get_sb,
1137 .kill_sb = cgroup_kill_sb,
1140 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1142 return dentry->d_fsdata;
1145 static inline struct cftype *__d_cft(struct dentry *dentry)
1147 return dentry->d_fsdata;
1151 * cgroup_path - generate the path of a cgroup
1152 * @cgrp: the cgroup in question
1153 * @buf: the buffer to write the path into
1154 * @buflen: the length of the buffer
1156 * Called with cgroup_mutex held. Writes path of cgroup into buf.
1157 * Returns 0 on success, -errno on error.
1159 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1163 if (cgrp == dummytop) {
1165 * Inactive subsystems have no dentry for their root
1172 start = buf + buflen;
1176 int len = cgrp->dentry->d_name.len;
1177 if ((start -= len) < buf)
1178 return -ENAMETOOLONG;
1179 memcpy(start, cgrp->dentry->d_name.name, len);
1180 cgrp = cgrp->parent;
1186 return -ENAMETOOLONG;
1189 memmove(buf, start, buf + buflen - start);
1194 * Return the first subsystem attached to a cgroup's hierarchy, and
1198 static void get_first_subsys(const struct cgroup *cgrp,
1199 struct cgroup_subsys_state **css, int *subsys_id)
1201 const struct cgroupfs_root *root = cgrp->root;
1202 const struct cgroup_subsys *test_ss;
1203 BUG_ON(list_empty(&root->subsys_list));
1204 test_ss = list_entry(root->subsys_list.next,
1205 struct cgroup_subsys, sibling);
1207 *css = cgrp->subsys[test_ss->subsys_id];
1211 *subsys_id = test_ss->subsys_id;
1215 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1216 * @cgrp: the cgroup the task is attaching to
1217 * @tsk: the task to be attached
1219 * Call holding cgroup_mutex. May take task_lock of
1220 * the task 'tsk' during call.
1222 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1225 struct cgroup_subsys *ss;
1226 struct cgroup *oldcgrp;
1227 struct css_set *cg = tsk->cgroups;
1228 struct css_set *newcg;
1229 struct cgroupfs_root *root = cgrp->root;
1232 get_first_subsys(cgrp, NULL, &subsys_id);
1234 /* Nothing to do if the task is already in that cgroup */
1235 oldcgrp = task_cgroup(tsk, subsys_id);
1236 if (cgrp == oldcgrp)
1239 for_each_subsys(root, ss) {
1240 if (ss->can_attach) {
1241 retval = ss->can_attach(ss, cgrp, tsk);
1248 * Locate or allocate a new css_set for this task,
1249 * based on its final set of cgroups
1251 newcg = find_css_set(cg, cgrp);
1256 if (tsk->flags & PF_EXITING) {
1261 rcu_assign_pointer(tsk->cgroups, newcg);
1264 /* Update the css_set linked lists if we're using them */
1265 write_lock(&css_set_lock);
1266 if (!list_empty(&tsk->cg_list)) {
1267 list_del(&tsk->cg_list);
1268 list_add(&tsk->cg_list, &newcg->tasks);
1270 write_unlock(&css_set_lock);
1272 for_each_subsys(root, ss) {
1274 ss->attach(ss, cgrp, oldcgrp, tsk);
1276 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1283 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with
1284 * cgroup_mutex, may take task_lock of task
1286 static int attach_task_by_pid(struct cgroup *cgrp, char *pidbuf)
1289 struct task_struct *tsk;
1292 if (sscanf(pidbuf, "%d", &pid) != 1)
1297 tsk = find_task_by_vpid(pid);
1298 if (!tsk || tsk->flags & PF_EXITING) {
1302 get_task_struct(tsk);
1305 if ((current->euid) && (current->euid != tsk->uid)
1306 && (current->euid != tsk->suid)) {
1307 put_task_struct(tsk);
1312 get_task_struct(tsk);
1315 ret = cgroup_attach_task(cgrp, tsk);
1316 put_task_struct(tsk);
1320 /* The various types of files and directories in a cgroup file system */
1321 enum cgroup_filetype {
1325 FILE_NOTIFY_ON_RELEASE,
1329 static ssize_t cgroup_write_X64(struct cgroup *cgrp, struct cftype *cft,
1331 const char __user *userbuf,
1332 size_t nbytes, loff_t *unused_ppos)
1340 if (nbytes >= sizeof(buffer))
1342 if (copy_from_user(buffer, userbuf, nbytes))
1345 buffer[nbytes] = 0; /* nul-terminate */
1347 if (cft->write_u64) {
1348 u64 val = simple_strtoull(buffer, &end, 0);
1351 retval = cft->write_u64(cgrp, cft, val);
1353 s64 val = simple_strtoll(buffer, &end, 0);
1356 retval = cft->write_s64(cgrp, cft, val);
1363 static ssize_t cgroup_common_file_write(struct cgroup *cgrp,
1366 const char __user *userbuf,
1367 size_t nbytes, loff_t *unused_ppos)
1369 enum cgroup_filetype type = cft->private;
1373 if (nbytes >= PATH_MAX)
1376 /* +1 for nul-terminator */
1377 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1381 if (copy_from_user(buffer, userbuf, nbytes)) {
1385 buffer[nbytes] = 0; /* nul-terminate */
1386 strstrip(buffer); /* strip -just- trailing whitespace */
1388 mutex_lock(&cgroup_mutex);
1391 * This was already checked for in cgroup_file_write(), but
1392 * check again now we're holding cgroup_mutex.
1394 if (cgroup_is_removed(cgrp)) {
1401 retval = attach_task_by_pid(cgrp, buffer);
1403 case FILE_NOTIFY_ON_RELEASE:
1404 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
1405 if (simple_strtoul(buffer, NULL, 10) != 0)
1406 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
1408 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
1410 case FILE_RELEASE_AGENT:
1411 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1412 strcpy(cgrp->root->release_agent_path, buffer);
1422 mutex_unlock(&cgroup_mutex);
1428 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
1429 size_t nbytes, loff_t *ppos)
1431 struct cftype *cft = __d_cft(file->f_dentry);
1432 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1434 if (!cft || cgroup_is_removed(cgrp))
1437 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
1438 if (cft->write_u64 || cft->write_s64)
1439 return cgroup_write_X64(cgrp, cft, file, buf, nbytes, ppos);
1441 int ret = cft->trigger(cgrp, (unsigned int)cft->private);
1442 return ret ? ret : nbytes;
1447 static ssize_t cgroup_read_u64(struct cgroup *cgrp, struct cftype *cft,
1449 char __user *buf, size_t nbytes,
1453 u64 val = cft->read_u64(cgrp, cft);
1454 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
1456 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1459 static ssize_t cgroup_read_s64(struct cgroup *cgrp, struct cftype *cft,
1461 char __user *buf, size_t nbytes,
1465 s64 val = cft->read_s64(cgrp, cft);
1466 int len = sprintf(tmp, "%lld\n", (long long) val);
1468 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1471 static ssize_t cgroup_common_file_read(struct cgroup *cgrp,
1475 size_t nbytes, loff_t *ppos)
1477 enum cgroup_filetype type = cft->private;
1482 if (!(page = (char *)__get_free_page(GFP_KERNEL)))
1488 case FILE_RELEASE_AGENT:
1490 struct cgroupfs_root *root;
1492 mutex_lock(&cgroup_mutex);
1494 n = strnlen(root->release_agent_path,
1495 sizeof(root->release_agent_path));
1496 n = min(n, (size_t) PAGE_SIZE);
1497 strncpy(s, root->release_agent_path, n);
1498 mutex_unlock(&cgroup_mutex);
1508 retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1510 free_page((unsigned long)page);
1514 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
1515 size_t nbytes, loff_t *ppos)
1517 struct cftype *cft = __d_cft(file->f_dentry);
1518 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1520 if (!cft || cgroup_is_removed(cgrp))
1524 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
1526 return cgroup_read_u64(cgrp, cft, file, buf, nbytes, ppos);
1528 return cgroup_read_s64(cgrp, cft, file, buf, nbytes, ppos);
1533 * seqfile ops/methods for returning structured data. Currently just
1534 * supports string->u64 maps, but can be extended in future.
1537 struct cgroup_seqfile_state {
1539 struct cgroup *cgroup;
1542 static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
1544 struct seq_file *sf = cb->state;
1545 return seq_printf(sf, "%s %llu\n", key, (unsigned long long)value);
1548 static int cgroup_seqfile_show(struct seq_file *m, void *arg)
1550 struct cgroup_seqfile_state *state = m->private;
1551 struct cftype *cft = state->cft;
1552 if (cft->read_map) {
1553 struct cgroup_map_cb cb = {
1554 .fill = cgroup_map_add,
1557 return cft->read_map(state->cgroup, cft, &cb);
1559 return cft->read_seq_string(state->cgroup, cft, m);
1562 int cgroup_seqfile_release(struct inode *inode, struct file *file)
1564 struct seq_file *seq = file->private_data;
1565 kfree(seq->private);
1566 return single_release(inode, file);
1569 static struct file_operations cgroup_seqfile_operations = {
1571 .llseek = seq_lseek,
1572 .release = cgroup_seqfile_release,
1575 static int cgroup_file_open(struct inode *inode, struct file *file)
1580 err = generic_file_open(inode, file);
1584 cft = __d_cft(file->f_dentry);
1587 if (cft->read_map || cft->read_seq_string) {
1588 struct cgroup_seqfile_state *state =
1589 kzalloc(sizeof(*state), GFP_USER);
1593 state->cgroup = __d_cgrp(file->f_dentry->d_parent);
1594 file->f_op = &cgroup_seqfile_operations;
1595 err = single_open(file, cgroup_seqfile_show, state);
1598 } else if (cft->open)
1599 err = cft->open(inode, file);
1606 static int cgroup_file_release(struct inode *inode, struct file *file)
1608 struct cftype *cft = __d_cft(file->f_dentry);
1610 return cft->release(inode, file);
1615 * cgroup_rename - Only allow simple rename of directories in place.
1617 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
1618 struct inode *new_dir, struct dentry *new_dentry)
1620 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1622 if (new_dentry->d_inode)
1624 if (old_dir != new_dir)
1626 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
1629 static struct file_operations cgroup_file_operations = {
1630 .read = cgroup_file_read,
1631 .write = cgroup_file_write,
1632 .llseek = generic_file_llseek,
1633 .open = cgroup_file_open,
1634 .release = cgroup_file_release,
1637 static struct inode_operations cgroup_dir_inode_operations = {
1638 .lookup = simple_lookup,
1639 .mkdir = cgroup_mkdir,
1640 .rmdir = cgroup_rmdir,
1641 .rename = cgroup_rename,
1644 static int cgroup_create_file(struct dentry *dentry, int mode,
1645 struct super_block *sb)
1647 static struct dentry_operations cgroup_dops = {
1648 .d_iput = cgroup_diput,
1651 struct inode *inode;
1655 if (dentry->d_inode)
1658 inode = cgroup_new_inode(mode, sb);
1662 if (S_ISDIR(mode)) {
1663 inode->i_op = &cgroup_dir_inode_operations;
1664 inode->i_fop = &simple_dir_operations;
1666 /* start off with i_nlink == 2 (for "." entry) */
1669 /* start with the directory inode held, so that we can
1670 * populate it without racing with another mkdir */
1671 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1672 } else if (S_ISREG(mode)) {
1674 inode->i_fop = &cgroup_file_operations;
1676 dentry->d_op = &cgroup_dops;
1677 d_instantiate(dentry, inode);
1678 dget(dentry); /* Extra count - pin the dentry in core */
1683 * cgroup_create_dir - create a directory for an object.
1684 * @cgrp: the cgroup we create the directory for. It must have a valid
1685 * ->parent field. And we are going to fill its ->dentry field.
1686 * @dentry: dentry of the new cgroup
1687 * @mode: mode to set on new directory.
1689 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
1692 struct dentry *parent;
1695 parent = cgrp->parent->dentry;
1696 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
1698 dentry->d_fsdata = cgrp;
1699 inc_nlink(parent->d_inode);
1700 cgrp->dentry = dentry;
1708 int cgroup_add_file(struct cgroup *cgrp,
1709 struct cgroup_subsys *subsys,
1710 const struct cftype *cft)
1712 struct dentry *dir = cgrp->dentry;
1713 struct dentry *dentry;
1716 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
1717 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
1718 strcpy(name, subsys->name);
1721 strcat(name, cft->name);
1722 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
1723 dentry = lookup_one_len(name, dir, strlen(name));
1724 if (!IS_ERR(dentry)) {
1725 error = cgroup_create_file(dentry, 0644 | S_IFREG,
1728 dentry->d_fsdata = (void *)cft;
1731 error = PTR_ERR(dentry);
1735 int cgroup_add_files(struct cgroup *cgrp,
1736 struct cgroup_subsys *subsys,
1737 const struct cftype cft[],
1741 for (i = 0; i < count; i++) {
1742 err = cgroup_add_file(cgrp, subsys, &cft[i]);
1750 * cgroup_task_count - count the number of tasks in a cgroup.
1751 * @cgrp: the cgroup in question
1753 * Return the number of tasks in the cgroup.
1755 int cgroup_task_count(const struct cgroup *cgrp)
1758 struct list_head *l;
1760 read_lock(&css_set_lock);
1761 l = cgrp->css_sets.next;
1762 while (l != &cgrp->css_sets) {
1763 struct cg_cgroup_link *link =
1764 list_entry(l, struct cg_cgroup_link, cgrp_link_list);
1765 count += atomic_read(&link->cg->ref.refcount);
1768 read_unlock(&css_set_lock);
1773 * Advance a list_head iterator. The iterator should be positioned at
1774 * the start of a css_set
1776 static void cgroup_advance_iter(struct cgroup *cgrp,
1777 struct cgroup_iter *it)
1779 struct list_head *l = it->cg_link;
1780 struct cg_cgroup_link *link;
1783 /* Advance to the next non-empty css_set */
1786 if (l == &cgrp->css_sets) {
1790 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
1792 } while (list_empty(&cg->tasks));
1794 it->task = cg->tasks.next;
1798 * To reduce the fork() overhead for systems that are not actually
1799 * using their cgroups capability, we don't maintain the lists running
1800 * through each css_set to its tasks until we see the list actually
1801 * used - in other words after the first call to cgroup_iter_start().
1803 * The tasklist_lock is not held here, as do_each_thread() and
1804 * while_each_thread() are protected by RCU.
1806 static void cgroup_enable_task_cg_lists(void)
1808 struct task_struct *p, *g;
1809 write_lock(&css_set_lock);
1810 use_task_css_set_links = 1;
1811 do_each_thread(g, p) {
1814 * We should check if the process is exiting, otherwise
1815 * it will race with cgroup_exit() in that the list
1816 * entry won't be deleted though the process has exited.
1818 if (!(p->flags & PF_EXITING) && list_empty(&p->cg_list))
1819 list_add(&p->cg_list, &p->cgroups->tasks);
1821 } while_each_thread(g, p);
1822 write_unlock(&css_set_lock);
1825 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
1828 * The first time anyone tries to iterate across a cgroup,
1829 * we need to enable the list linking each css_set to its
1830 * tasks, and fix up all existing tasks.
1832 if (!use_task_css_set_links)
1833 cgroup_enable_task_cg_lists();
1835 read_lock(&css_set_lock);
1836 it->cg_link = &cgrp->css_sets;
1837 cgroup_advance_iter(cgrp, it);
1840 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
1841 struct cgroup_iter *it)
1843 struct task_struct *res;
1844 struct list_head *l = it->task;
1846 /* If the iterator cg is NULL, we have no tasks */
1849 res = list_entry(l, struct task_struct, cg_list);
1850 /* Advance iterator to find next entry */
1852 if (l == &res->cgroups->tasks) {
1853 /* We reached the end of this task list - move on to
1854 * the next cg_cgroup_link */
1855 cgroup_advance_iter(cgrp, it);
1862 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
1864 read_unlock(&css_set_lock);
1867 static inline int started_after_time(struct task_struct *t1,
1868 struct timespec *time,
1869 struct task_struct *t2)
1871 int start_diff = timespec_compare(&t1->start_time, time);
1872 if (start_diff > 0) {
1874 } else if (start_diff < 0) {
1878 * Arbitrarily, if two processes started at the same
1879 * time, we'll say that the lower pointer value
1880 * started first. Note that t2 may have exited by now
1881 * so this may not be a valid pointer any longer, but
1882 * that's fine - it still serves to distinguish
1883 * between two tasks started (effectively) simultaneously.
1890 * This function is a callback from heap_insert() and is used to order
1892 * In this case we order the heap in descending task start time.
1894 static inline int started_after(void *p1, void *p2)
1896 struct task_struct *t1 = p1;
1897 struct task_struct *t2 = p2;
1898 return started_after_time(t1, &t2->start_time, t2);
1902 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
1903 * @scan: struct cgroup_scanner containing arguments for the scan
1905 * Arguments include pointers to callback functions test_task() and
1907 * Iterate through all the tasks in a cgroup, calling test_task() for each,
1908 * and if it returns true, call process_task() for it also.
1909 * The test_task pointer may be NULL, meaning always true (select all tasks).
1910 * Effectively duplicates cgroup_iter_{start,next,end}()
1911 * but does not lock css_set_lock for the call to process_task().
1912 * The struct cgroup_scanner may be embedded in any structure of the caller's
1914 * It is guaranteed that process_task() will act on every task that
1915 * is a member of the cgroup for the duration of this call. This
1916 * function may or may not call process_task() for tasks that exit
1917 * or move to a different cgroup during the call, or are forked or
1918 * move into the cgroup during the call.
1920 * Note that test_task() may be called with locks held, and may in some
1921 * situations be called multiple times for the same task, so it should
1923 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
1924 * pre-allocated and will be used for heap operations (and its "gt" member will
1925 * be overwritten), else a temporary heap will be used (allocation of which
1926 * may cause this function to fail).
1928 int cgroup_scan_tasks(struct cgroup_scanner *scan)
1931 struct cgroup_iter it;
1932 struct task_struct *p, *dropped;
1933 /* Never dereference latest_task, since it's not refcounted */
1934 struct task_struct *latest_task = NULL;
1935 struct ptr_heap tmp_heap;
1936 struct ptr_heap *heap;
1937 struct timespec latest_time = { 0, 0 };
1940 /* The caller supplied our heap and pre-allocated its memory */
1942 heap->gt = &started_after;
1944 /* We need to allocate our own heap memory */
1946 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
1948 /* cannot allocate the heap */
1954 * Scan tasks in the cgroup, using the scanner's "test_task" callback
1955 * to determine which are of interest, and using the scanner's
1956 * "process_task" callback to process any of them that need an update.
1957 * Since we don't want to hold any locks during the task updates,
1958 * gather tasks to be processed in a heap structure.
1959 * The heap is sorted by descending task start time.
1960 * If the statically-sized heap fills up, we overflow tasks that
1961 * started later, and in future iterations only consider tasks that
1962 * started after the latest task in the previous pass. This
1963 * guarantees forward progress and that we don't miss any tasks.
1966 cgroup_iter_start(scan->cg, &it);
1967 while ((p = cgroup_iter_next(scan->cg, &it))) {
1969 * Only affect tasks that qualify per the caller's callback,
1970 * if he provided one
1972 if (scan->test_task && !scan->test_task(p, scan))
1975 * Only process tasks that started after the last task
1978 if (!started_after_time(p, &latest_time, latest_task))
1980 dropped = heap_insert(heap, p);
1981 if (dropped == NULL) {
1983 * The new task was inserted; the heap wasn't
1987 } else if (dropped != p) {
1989 * The new task was inserted, and pushed out a
1993 put_task_struct(dropped);
1996 * Else the new task was newer than anything already in
1997 * the heap and wasn't inserted
2000 cgroup_iter_end(scan->cg, &it);
2003 for (i = 0; i < heap->size; i++) {
2004 struct task_struct *q = heap->ptrs[i];
2006 latest_time = q->start_time;
2009 /* Process the task per the caller's callback */
2010 scan->process_task(q, scan);
2014 * If we had to process any tasks at all, scan again
2015 * in case some of them were in the middle of forking
2016 * children that didn't get processed.
2017 * Not the most efficient way to do it, but it avoids
2018 * having to take callback_mutex in the fork path
2022 if (heap == &tmp_heap)
2023 heap_free(&tmp_heap);
2028 * Stuff for reading the 'tasks' file.
2030 * Reading this file can return large amounts of data if a cgroup has
2031 * *lots* of attached tasks. So it may need several calls to read(),
2032 * but we cannot guarantee that the information we produce is correct
2033 * unless we produce it entirely atomically.
2035 * Upon tasks file open(), a struct ctr_struct is allocated, that
2036 * will have a pointer to an array (also allocated here). The struct
2037 * ctr_struct * is stored in file->private_data. Its resources will
2038 * be freed by release() when the file is closed. The array is used
2039 * to sprintf the PIDs and then used by read().
2047 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
2048 * 'cgrp'. Return actual number of pids loaded. No need to
2049 * task_lock(p) when reading out p->cgroup, since we're in an RCU
2050 * read section, so the css_set can't go away, and is
2051 * immutable after creation.
2053 static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cgrp)
2056 struct cgroup_iter it;
2057 struct task_struct *tsk;
2058 cgroup_iter_start(cgrp, &it);
2059 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2060 if (unlikely(n == npids))
2062 pidarray[n++] = task_pid_vnr(tsk);
2064 cgroup_iter_end(cgrp, &it);
2069 * cgroupstats_build - build and fill cgroupstats
2070 * @stats: cgroupstats to fill information into
2071 * @dentry: A dentry entry belonging to the cgroup for which stats have
2074 * Build and fill cgroupstats so that taskstats can export it to user
2077 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
2080 struct cgroup *cgrp;
2081 struct cgroup_iter it;
2082 struct task_struct *tsk;
2084 * Validate dentry by checking the superblock operations
2086 if (dentry->d_sb->s_op != &cgroup_ops)
2090 cgrp = dentry->d_fsdata;
2093 cgroup_iter_start(cgrp, &it);
2094 while ((tsk = cgroup_iter_next(cgrp, &it))) {
2095 switch (tsk->state) {
2097 stats->nr_running++;
2099 case TASK_INTERRUPTIBLE:
2100 stats->nr_sleeping++;
2102 case TASK_UNINTERRUPTIBLE:
2103 stats->nr_uninterruptible++;
2106 stats->nr_stopped++;
2109 if (delayacct_is_task_waiting_on_io(tsk))
2110 stats->nr_io_wait++;
2114 cgroup_iter_end(cgrp, &it);
2121 static int cmppid(const void *a, const void *b)
2123 return *(pid_t *)a - *(pid_t *)b;
2127 * Convert array 'a' of 'npids' pid_t's to a string of newline separated
2128 * decimal pids in 'buf'. Don't write more than 'sz' chars, but return
2129 * count 'cnt' of how many chars would be written if buf were large enough.
2131 static int pid_array_to_buf(char *buf, int sz, pid_t *a, int npids)
2136 for (i = 0; i < npids; i++)
2137 cnt += snprintf(buf + cnt, max(sz - cnt, 0), "%d\n", a[i]);
2142 * Handle an open on 'tasks' file. Prepare a buffer listing the
2143 * process id's of tasks currently attached to the cgroup being opened.
2145 * Does not require any specific cgroup mutexes, and does not take any.
2147 static int cgroup_tasks_open(struct inode *unused, struct file *file)
2149 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2150 struct ctr_struct *ctr;
2155 if (!(file->f_mode & FMODE_READ))
2158 ctr = kmalloc(sizeof(*ctr), GFP_KERNEL);
2163 * If cgroup gets more users after we read count, we won't have
2164 * enough space - tough. This race is indistinguishable to the
2165 * caller from the case that the additional cgroup users didn't
2166 * show up until sometime later on.
2168 npids = cgroup_task_count(cgrp);
2170 pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL);
2174 npids = pid_array_load(pidarray, npids, cgrp);
2175 sort(pidarray, npids, sizeof(pid_t), cmppid, NULL);
2177 /* Call pid_array_to_buf() twice, first just to get bufsz */
2178 ctr->bufsz = pid_array_to_buf(&c, sizeof(c), pidarray, npids) + 1;
2179 ctr->buf = kmalloc(ctr->bufsz, GFP_KERNEL);
2182 ctr->bufsz = pid_array_to_buf(ctr->buf, ctr->bufsz, pidarray, npids);
2189 file->private_data = ctr;
2200 static ssize_t cgroup_tasks_read(struct cgroup *cgrp,
2202 struct file *file, char __user *buf,
2203 size_t nbytes, loff_t *ppos)
2205 struct ctr_struct *ctr = file->private_data;
2207 return simple_read_from_buffer(buf, nbytes, ppos, ctr->buf, ctr->bufsz);
2210 static int cgroup_tasks_release(struct inode *unused_inode,
2213 struct ctr_struct *ctr;
2215 if (file->f_mode & FMODE_READ) {
2216 ctr = file->private_data;
2223 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
2226 return notify_on_release(cgrp);
2230 * for the common functions, 'private' gives the type of file
2232 static struct cftype files[] = {
2235 .open = cgroup_tasks_open,
2236 .read = cgroup_tasks_read,
2237 .write = cgroup_common_file_write,
2238 .release = cgroup_tasks_release,
2239 .private = FILE_TASKLIST,
2243 .name = "notify_on_release",
2244 .read_u64 = cgroup_read_notify_on_release,
2245 .write = cgroup_common_file_write,
2246 .private = FILE_NOTIFY_ON_RELEASE,
2250 static struct cftype cft_release_agent = {
2251 .name = "release_agent",
2252 .read = cgroup_common_file_read,
2253 .write = cgroup_common_file_write,
2254 .private = FILE_RELEASE_AGENT,
2257 static int cgroup_populate_dir(struct cgroup *cgrp)
2260 struct cgroup_subsys *ss;
2262 /* First clear out any existing files */
2263 cgroup_clear_directory(cgrp->dentry);
2265 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
2269 if (cgrp == cgrp->top_cgroup) {
2270 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
2274 for_each_subsys(cgrp->root, ss) {
2275 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
2282 static void init_cgroup_css(struct cgroup_subsys_state *css,
2283 struct cgroup_subsys *ss,
2284 struct cgroup *cgrp)
2287 atomic_set(&css->refcnt, 0);
2289 if (cgrp == dummytop)
2290 set_bit(CSS_ROOT, &css->flags);
2291 BUG_ON(cgrp->subsys[ss->subsys_id]);
2292 cgrp->subsys[ss->subsys_id] = css;
2296 * cgroup_create - create a cgroup
2297 * @parent: cgroup that will be parent of the new cgroup
2298 * @dentry: dentry of the new cgroup
2299 * @mode: mode to set on new inode
2301 * Must be called with the mutex on the parent inode held
2303 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
2306 struct cgroup *cgrp;
2307 struct cgroupfs_root *root = parent->root;
2309 struct cgroup_subsys *ss;
2310 struct super_block *sb = root->sb;
2312 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
2316 /* Grab a reference on the superblock so the hierarchy doesn't
2317 * get deleted on unmount if there are child cgroups. This
2318 * can be done outside cgroup_mutex, since the sb can't
2319 * disappear while someone has an open control file on the
2321 atomic_inc(&sb->s_active);
2323 mutex_lock(&cgroup_mutex);
2325 INIT_LIST_HEAD(&cgrp->sibling);
2326 INIT_LIST_HEAD(&cgrp->children);
2327 INIT_LIST_HEAD(&cgrp->css_sets);
2328 INIT_LIST_HEAD(&cgrp->release_list);
2330 cgrp->parent = parent;
2331 cgrp->root = parent->root;
2332 cgrp->top_cgroup = parent->top_cgroup;
2334 if (notify_on_release(parent))
2335 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
2337 for_each_subsys(root, ss) {
2338 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
2343 init_cgroup_css(css, ss, cgrp);
2346 list_add(&cgrp->sibling, &cgrp->parent->children);
2347 root->number_of_cgroups++;
2349 err = cgroup_create_dir(cgrp, dentry, mode);
2353 /* The cgroup directory was pre-locked for us */
2354 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
2356 err = cgroup_populate_dir(cgrp);
2357 /* If err < 0, we have a half-filled directory - oh well ;) */
2359 mutex_unlock(&cgroup_mutex);
2360 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
2366 list_del(&cgrp->sibling);
2367 root->number_of_cgroups--;
2371 for_each_subsys(root, ss) {
2372 if (cgrp->subsys[ss->subsys_id])
2373 ss->destroy(ss, cgrp);
2376 mutex_unlock(&cgroup_mutex);
2378 /* Release the reference count that we took on the superblock */
2379 deactivate_super(sb);
2385 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2387 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
2389 /* the vfs holds inode->i_mutex already */
2390 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
2393 static inline int cgroup_has_css_refs(struct cgroup *cgrp)
2395 /* Check the reference count on each subsystem. Since we
2396 * already established that there are no tasks in the
2397 * cgroup, if the css refcount is also 0, then there should
2398 * be no outstanding references, so the subsystem is safe to
2399 * destroy. We scan across all subsystems rather than using
2400 * the per-hierarchy linked list of mounted subsystems since
2401 * we can be called via check_for_release() with no
2402 * synchronization other than RCU, and the subsystem linked
2403 * list isn't RCU-safe */
2405 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2406 struct cgroup_subsys *ss = subsys[i];
2407 struct cgroup_subsys_state *css;
2408 /* Skip subsystems not in this hierarchy */
2409 if (ss->root != cgrp->root)
2411 css = cgrp->subsys[ss->subsys_id];
2412 /* When called from check_for_release() it's possible
2413 * that by this point the cgroup has been removed
2414 * and the css deleted. But a false-positive doesn't
2415 * matter, since it can only happen if the cgroup
2416 * has been deleted and hence no longer needs the
2417 * release agent to be called anyway. */
2418 if (css && atomic_read(&css->refcnt))
2424 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
2426 struct cgroup *cgrp = dentry->d_fsdata;
2428 struct cgroup *parent;
2429 struct super_block *sb;
2430 struct cgroupfs_root *root;
2432 /* the vfs holds both inode->i_mutex already */
2434 mutex_lock(&cgroup_mutex);
2435 if (atomic_read(&cgrp->count) != 0) {
2436 mutex_unlock(&cgroup_mutex);
2439 if (!list_empty(&cgrp->children)) {
2440 mutex_unlock(&cgroup_mutex);
2444 parent = cgrp->parent;
2449 * Call pre_destroy handlers of subsys. Notify subsystems
2450 * that rmdir() request comes.
2452 cgroup_call_pre_destroy(cgrp);
2454 if (cgroup_has_css_refs(cgrp)) {
2455 mutex_unlock(&cgroup_mutex);
2459 spin_lock(&release_list_lock);
2460 set_bit(CGRP_REMOVED, &cgrp->flags);
2461 if (!list_empty(&cgrp->release_list))
2462 list_del(&cgrp->release_list);
2463 spin_unlock(&release_list_lock);
2464 /* delete my sibling from parent->children */
2465 list_del(&cgrp->sibling);
2466 spin_lock(&cgrp->dentry->d_lock);
2467 d = dget(cgrp->dentry);
2468 cgrp->dentry = NULL;
2469 spin_unlock(&d->d_lock);
2471 cgroup_d_remove_dir(d);
2474 set_bit(CGRP_RELEASABLE, &parent->flags);
2475 check_for_release(parent);
2477 mutex_unlock(&cgroup_mutex);
2481 static void __init cgroup_init_subsys(struct cgroup_subsys *ss)
2483 struct cgroup_subsys_state *css;
2485 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
2487 /* Create the top cgroup state for this subsystem */
2488 ss->root = &rootnode;
2489 css = ss->create(ss, dummytop);
2490 /* We don't handle early failures gracefully */
2491 BUG_ON(IS_ERR(css));
2492 init_cgroup_css(css, ss, dummytop);
2494 /* Update the init_css_set to contain a subsys
2495 * pointer to this state - since the subsystem is
2496 * newly registered, all tasks and hence the
2497 * init_css_set is in the subsystem's top cgroup. */
2498 init_css_set.subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
2500 need_forkexit_callback |= ss->fork || ss->exit;
2502 /* At system boot, before all subsystems have been
2503 * registered, no tasks have been forked, so we don't
2504 * need to invoke fork callbacks here. */
2505 BUG_ON(!list_empty(&init_task.tasks));
2511 * cgroup_init_early - cgroup initialization at system boot
2513 * Initialize cgroups at system boot, and initialize any
2514 * subsystems that request early init.
2516 int __init cgroup_init_early(void)
2519 kref_init(&init_css_set.ref);
2520 kref_get(&init_css_set.ref);
2521 INIT_LIST_HEAD(&init_css_set.cg_links);
2522 INIT_LIST_HEAD(&init_css_set.tasks);
2523 INIT_HLIST_NODE(&init_css_set.hlist);
2525 init_cgroup_root(&rootnode);
2526 list_add(&rootnode.root_list, &roots);
2528 init_task.cgroups = &init_css_set;
2530 init_css_set_link.cg = &init_css_set;
2531 list_add(&init_css_set_link.cgrp_link_list,
2532 &rootnode.top_cgroup.css_sets);
2533 list_add(&init_css_set_link.cg_link_list,
2534 &init_css_set.cg_links);
2536 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
2537 INIT_HLIST_HEAD(&css_set_table[i]);
2539 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2540 struct cgroup_subsys *ss = subsys[i];
2543 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
2544 BUG_ON(!ss->create);
2545 BUG_ON(!ss->destroy);
2546 if (ss->subsys_id != i) {
2547 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
2548 ss->name, ss->subsys_id);
2553 cgroup_init_subsys(ss);
2559 * cgroup_init - cgroup initialization
2561 * Register cgroup filesystem and /proc file, and initialize
2562 * any subsystems that didn't request early init.
2564 int __init cgroup_init(void)
2568 struct hlist_head *hhead;
2570 err = bdi_init(&cgroup_backing_dev_info);
2574 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2575 struct cgroup_subsys *ss = subsys[i];
2576 if (!ss->early_init)
2577 cgroup_init_subsys(ss);
2580 /* Add init_css_set to the hash table */
2581 hhead = css_set_hash(init_css_set.subsys);
2582 hlist_add_head(&init_css_set.hlist, hhead);
2584 err = register_filesystem(&cgroup_fs_type);
2588 proc_create("cgroups", 0, NULL, &proc_cgroupstats_operations);
2592 bdi_destroy(&cgroup_backing_dev_info);
2598 * proc_cgroup_show()
2599 * - Print task's cgroup paths into seq_file, one line for each hierarchy
2600 * - Used for /proc/<pid>/cgroup.
2601 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
2602 * doesn't really matter if tsk->cgroup changes after we read it,
2603 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
2604 * anyway. No need to check that tsk->cgroup != NULL, thanks to
2605 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
2606 * cgroup to top_cgroup.
2609 /* TODO: Use a proper seq_file iterator */
2610 static int proc_cgroup_show(struct seq_file *m, void *v)
2613 struct task_struct *tsk;
2616 struct cgroupfs_root *root;
2619 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2625 tsk = get_pid_task(pid, PIDTYPE_PID);
2631 mutex_lock(&cgroup_mutex);
2633 for_each_root(root) {
2634 struct cgroup_subsys *ss;
2635 struct cgroup *cgrp;
2639 /* Skip this hierarchy if it has no active subsystems */
2640 if (!root->actual_subsys_bits)
2642 seq_printf(m, "%lu:", root->subsys_bits);
2643 for_each_subsys(root, ss)
2644 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
2646 get_first_subsys(&root->top_cgroup, NULL, &subsys_id);
2647 cgrp = task_cgroup(tsk, subsys_id);
2648 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
2656 mutex_unlock(&cgroup_mutex);
2657 put_task_struct(tsk);
2664 static int cgroup_open(struct inode *inode, struct file *file)
2666 struct pid *pid = PROC_I(inode)->pid;
2667 return single_open(file, proc_cgroup_show, pid);
2670 struct file_operations proc_cgroup_operations = {
2671 .open = cgroup_open,
2673 .llseek = seq_lseek,
2674 .release = single_release,
2677 /* Display information about each subsystem and each hierarchy */
2678 static int proc_cgroupstats_show(struct seq_file *m, void *v)
2682 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\tenabled\n");
2683 mutex_lock(&cgroup_mutex);
2684 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2685 struct cgroup_subsys *ss = subsys[i];
2686 seq_printf(m, "%s\t%lu\t%d\t%d\n",
2687 ss->name, ss->root->subsys_bits,
2688 ss->root->number_of_cgroups, !ss->disabled);
2690 mutex_unlock(&cgroup_mutex);
2694 static int cgroupstats_open(struct inode *inode, struct file *file)
2696 return single_open(file, proc_cgroupstats_show, NULL);
2699 static struct file_operations proc_cgroupstats_operations = {
2700 .open = cgroupstats_open,
2702 .llseek = seq_lseek,
2703 .release = single_release,
2707 * cgroup_fork - attach newly forked task to its parents cgroup.
2708 * @child: pointer to task_struct of forking parent process.
2710 * Description: A task inherits its parent's cgroup at fork().
2712 * A pointer to the shared css_set was automatically copied in
2713 * fork.c by dup_task_struct(). However, we ignore that copy, since
2714 * it was not made under the protection of RCU or cgroup_mutex, so
2715 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
2716 * have already changed current->cgroups, allowing the previously
2717 * referenced cgroup group to be removed and freed.
2719 * At the point that cgroup_fork() is called, 'current' is the parent
2720 * task, and the passed argument 'child' points to the child task.
2722 void cgroup_fork(struct task_struct *child)
2725 child->cgroups = current->cgroups;
2726 get_css_set(child->cgroups);
2727 task_unlock(current);
2728 INIT_LIST_HEAD(&child->cg_list);
2732 * cgroup_fork_callbacks - run fork callbacks
2733 * @child: the new task
2735 * Called on a new task very soon before adding it to the
2736 * tasklist. No need to take any locks since no-one can
2737 * be operating on this task.
2739 void cgroup_fork_callbacks(struct task_struct *child)
2741 if (need_forkexit_callback) {
2743 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2744 struct cgroup_subsys *ss = subsys[i];
2746 ss->fork(ss, child);
2752 * cgroup_post_fork - called on a new task after adding it to the task list
2753 * @child: the task in question
2755 * Adds the task to the list running through its css_set if necessary.
2756 * Has to be after the task is visible on the task list in case we race
2757 * with the first call to cgroup_iter_start() - to guarantee that the
2758 * new task ends up on its list.
2760 void cgroup_post_fork(struct task_struct *child)
2762 if (use_task_css_set_links) {
2763 write_lock(&css_set_lock);
2764 if (list_empty(&child->cg_list))
2765 list_add(&child->cg_list, &child->cgroups->tasks);
2766 write_unlock(&css_set_lock);
2770 * cgroup_exit - detach cgroup from exiting task
2771 * @tsk: pointer to task_struct of exiting process
2772 * @run_callback: run exit callbacks?
2774 * Description: Detach cgroup from @tsk and release it.
2776 * Note that cgroups marked notify_on_release force every task in
2777 * them to take the global cgroup_mutex mutex when exiting.
2778 * This could impact scaling on very large systems. Be reluctant to
2779 * use notify_on_release cgroups where very high task exit scaling
2780 * is required on large systems.
2782 * the_top_cgroup_hack:
2784 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
2786 * We call cgroup_exit() while the task is still competent to
2787 * handle notify_on_release(), then leave the task attached to the
2788 * root cgroup in each hierarchy for the remainder of its exit.
2790 * To do this properly, we would increment the reference count on
2791 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
2792 * code we would add a second cgroup function call, to drop that
2793 * reference. This would just create an unnecessary hot spot on
2794 * the top_cgroup reference count, to no avail.
2796 * Normally, holding a reference to a cgroup without bumping its
2797 * count is unsafe. The cgroup could go away, or someone could
2798 * attach us to a different cgroup, decrementing the count on
2799 * the first cgroup that we never incremented. But in this case,
2800 * top_cgroup isn't going away, and either task has PF_EXITING set,
2801 * which wards off any cgroup_attach_task() attempts, or task is a failed
2802 * fork, never visible to cgroup_attach_task.
2804 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
2809 if (run_callbacks && need_forkexit_callback) {
2810 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2811 struct cgroup_subsys *ss = subsys[i];
2818 * Unlink from the css_set task list if necessary.
2819 * Optimistically check cg_list before taking
2822 if (!list_empty(&tsk->cg_list)) {
2823 write_lock(&css_set_lock);
2824 if (!list_empty(&tsk->cg_list))
2825 list_del(&tsk->cg_list);
2826 write_unlock(&css_set_lock);
2829 /* Reassign the task to the init_css_set. */
2832 tsk->cgroups = &init_css_set;
2835 put_css_set_taskexit(cg);
2839 * cgroup_clone - clone the cgroup the given subsystem is attached to
2840 * @tsk: the task to be moved
2841 * @subsys: the given subsystem
2843 * Duplicate the current cgroup in the hierarchy that the given
2844 * subsystem is attached to, and move this task into the new
2847 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys)
2849 struct dentry *dentry;
2851 char nodename[MAX_CGROUP_TYPE_NAMELEN];
2852 struct cgroup *parent, *child;
2853 struct inode *inode;
2855 struct cgroupfs_root *root;
2856 struct cgroup_subsys *ss;
2858 /* We shouldn't be called by an unregistered subsystem */
2859 BUG_ON(!subsys->active);
2861 /* First figure out what hierarchy and cgroup we're dealing
2862 * with, and pin them so we can drop cgroup_mutex */
2863 mutex_lock(&cgroup_mutex);
2865 root = subsys->root;
2866 if (root == &rootnode) {
2868 "Not cloning cgroup for unused subsystem %s\n",
2870 mutex_unlock(&cgroup_mutex);
2874 parent = task_cgroup(tsk, subsys->subsys_id);
2876 snprintf(nodename, MAX_CGROUP_TYPE_NAMELEN, "node_%d", tsk->pid);
2878 /* Pin the hierarchy */
2879 atomic_inc(&parent->root->sb->s_active);
2881 /* Keep the cgroup alive */
2883 mutex_unlock(&cgroup_mutex);
2885 /* Now do the VFS work to create a cgroup */
2886 inode = parent->dentry->d_inode;
2888 /* Hold the parent directory mutex across this operation to
2889 * stop anyone else deleting the new cgroup */
2890 mutex_lock(&inode->i_mutex);
2891 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
2892 if (IS_ERR(dentry)) {
2894 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
2896 ret = PTR_ERR(dentry);
2900 /* Create the cgroup directory, which also creates the cgroup */
2901 ret = vfs_mkdir(inode, dentry, S_IFDIR | 0755);
2902 child = __d_cgrp(dentry);
2906 "Failed to create cgroup %s: %d\n", nodename,
2913 "Couldn't find new cgroup %s\n", nodename);
2918 /* The cgroup now exists. Retake cgroup_mutex and check
2919 * that we're still in the same state that we thought we
2921 mutex_lock(&cgroup_mutex);
2922 if ((root != subsys->root) ||
2923 (parent != task_cgroup(tsk, subsys->subsys_id))) {
2924 /* Aargh, we raced ... */
2925 mutex_unlock(&inode->i_mutex);
2928 deactivate_super(parent->root->sb);
2929 /* The cgroup is still accessible in the VFS, but
2930 * we're not going to try to rmdir() it at this
2933 "Race in cgroup_clone() - leaking cgroup %s\n",
2938 /* do any required auto-setup */
2939 for_each_subsys(root, ss) {
2941 ss->post_clone(ss, child);
2944 /* All seems fine. Finish by moving the task into the new cgroup */
2945 ret = cgroup_attach_task(child, tsk);
2946 mutex_unlock(&cgroup_mutex);
2949 mutex_unlock(&inode->i_mutex);
2951 mutex_lock(&cgroup_mutex);
2953 mutex_unlock(&cgroup_mutex);
2954 deactivate_super(parent->root->sb);
2959 * cgroup_is_descendant - see if @cgrp is a descendant of current task's cgrp
2960 * @cgrp: the cgroup in question
2962 * See if @cgrp is a descendant of the current task's cgroup in
2963 * the appropriate hierarchy.
2965 * If we are sending in dummytop, then presumably we are creating
2966 * the top cgroup in the subsystem.
2968 * Called only by the ns (nsproxy) cgroup.
2970 int cgroup_is_descendant(const struct cgroup *cgrp)
2973 struct cgroup *target;
2976 if (cgrp == dummytop)
2979 get_first_subsys(cgrp, NULL, &subsys_id);
2980 target = task_cgroup(current, subsys_id);
2981 while (cgrp != target && cgrp!= cgrp->top_cgroup)
2982 cgrp = cgrp->parent;
2983 ret = (cgrp == target);
2987 static void check_for_release(struct cgroup *cgrp)
2989 /* All of these checks rely on RCU to keep the cgroup
2990 * structure alive */
2991 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
2992 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
2993 /* Control Group is currently removeable. If it's not
2994 * already queued for a userspace notification, queue
2996 int need_schedule_work = 0;
2997 spin_lock(&release_list_lock);
2998 if (!cgroup_is_removed(cgrp) &&
2999 list_empty(&cgrp->release_list)) {
3000 list_add(&cgrp->release_list, &release_list);
3001 need_schedule_work = 1;
3003 spin_unlock(&release_list_lock);
3004 if (need_schedule_work)
3005 schedule_work(&release_agent_work);
3009 void __css_put(struct cgroup_subsys_state *css)
3011 struct cgroup *cgrp = css->cgroup;
3013 if (atomic_dec_and_test(&css->refcnt) && notify_on_release(cgrp)) {
3014 set_bit(CGRP_RELEASABLE, &cgrp->flags);
3015 check_for_release(cgrp);
3021 * Notify userspace when a cgroup is released, by running the
3022 * configured release agent with the name of the cgroup (path
3023 * relative to the root of cgroup file system) as the argument.
3025 * Most likely, this user command will try to rmdir this cgroup.
3027 * This races with the possibility that some other task will be
3028 * attached to this cgroup before it is removed, or that some other
3029 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
3030 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
3031 * unused, and this cgroup will be reprieved from its death sentence,
3032 * to continue to serve a useful existence. Next time it's released,
3033 * we will get notified again, if it still has 'notify_on_release' set.
3035 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
3036 * means only wait until the task is successfully execve()'d. The
3037 * separate release agent task is forked by call_usermodehelper(),
3038 * then control in this thread returns here, without waiting for the
3039 * release agent task. We don't bother to wait because the caller of
3040 * this routine has no use for the exit status of the release agent
3041 * task, so no sense holding our caller up for that.
3043 static void cgroup_release_agent(struct work_struct *work)
3045 BUG_ON(work != &release_agent_work);
3046 mutex_lock(&cgroup_mutex);
3047 spin_lock(&release_list_lock);
3048 while (!list_empty(&release_list)) {
3049 char *argv[3], *envp[3];
3052 struct cgroup *cgrp = list_entry(release_list.next,
3055 list_del_init(&cgrp->release_list);
3056 spin_unlock(&release_list_lock);
3057 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3059 spin_lock(&release_list_lock);
3063 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) {
3065 spin_lock(&release_list_lock);
3070 argv[i++] = cgrp->root->release_agent_path;
3071 argv[i++] = (char *)pathbuf;
3075 /* minimal command environment */
3076 envp[i++] = "HOME=/";
3077 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
3080 /* Drop the lock while we invoke the usermode helper,
3081 * since the exec could involve hitting disk and hence
3082 * be a slow process */
3083 mutex_unlock(&cgroup_mutex);
3084 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
3086 mutex_lock(&cgroup_mutex);
3087 spin_lock(&release_list_lock);
3089 spin_unlock(&release_list_lock);
3090 mutex_unlock(&cgroup_mutex);
3093 static int __init cgroup_disable(char *str)
3098 while ((token = strsep(&str, ",")) != NULL) {
3102 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
3103 struct cgroup_subsys *ss = subsys[i];
3105 if (!strcmp(token, ss->name)) {
3107 printk(KERN_INFO "Disabling %s control group"
3108 " subsystem\n", ss->name);
3115 __setup("cgroup_disable=", cgroup_disable);