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 * check for fork/exit handlers to call. This avoids us having to do
117 * extra work in the fork/exit path if none of the subsystems need to
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 static 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 static 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
310 static struct css_set *find_existing_css_set(
311 struct css_set *oldcg,
313 struct cgroup_subsys_state *template[])
316 struct cgroupfs_root *root = cgrp->root;
317 struct list_head *l = &init_css_set.list;
319 /* Built the set of subsystem state objects that we want to
320 * see in the new css_set */
321 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
322 if (root->subsys_bits & (1ull << i)) {
323 /* Subsystem is in this hierarchy. So we want
324 * the subsystem state from the new
326 template[i] = cgrp->subsys[i];
328 /* Subsystem is not in this hierarchy, so we
329 * don't want to change the subsystem state */
330 template[i] = oldcg->subsys[i];
334 /* Look through existing cgroup groups to find one to reuse */
337 list_entry(l, struct css_set, list);
339 if (!memcmp(template, cg->subsys, sizeof(cg->subsys))) {
340 /* All subsystems matched */
343 /* Try the next cgroup group */
345 } while (l != &init_css_set.list);
347 /* No existing cgroup group matched */
352 * allocate_cg_links() allocates "count" cg_cgroup_link structures
353 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
354 * success or a negative error
356 static int allocate_cg_links(int count, struct list_head *tmp)
358 struct cg_cgroup_link *link;
361 for (i = 0; i < count; i++) {
362 link = kmalloc(sizeof(*link), GFP_KERNEL);
364 while (!list_empty(tmp)) {
365 link = list_entry(tmp->next,
366 struct cg_cgroup_link,
368 list_del(&link->cgrp_link_list);
373 list_add(&link->cgrp_link_list, tmp);
378 static void free_cg_links(struct list_head *tmp)
380 while (!list_empty(tmp)) {
381 struct cg_cgroup_link *link;
382 link = list_entry(tmp->next,
383 struct cg_cgroup_link,
385 list_del(&link->cgrp_link_list);
391 * find_css_set() takes an existing cgroup group and a
392 * cgroup object, and returns a css_set object that's
393 * equivalent to the old group, but with the given cgroup
394 * substituted into the appropriate hierarchy. Must be called with
397 static struct css_set *find_css_set(
398 struct css_set *oldcg, struct cgroup *cgrp)
401 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
404 struct list_head tmp_cg_links;
405 struct cg_cgroup_link *link;
407 /* First see if we already have a cgroup group that matches
409 write_lock(&css_set_lock);
410 res = find_existing_css_set(oldcg, cgrp, template);
413 write_unlock(&css_set_lock);
418 res = kmalloc(sizeof(*res), GFP_KERNEL);
422 /* Allocate all the cg_cgroup_link objects that we'll need */
423 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
428 kref_init(&res->ref);
429 INIT_LIST_HEAD(&res->cg_links);
430 INIT_LIST_HEAD(&res->tasks);
432 /* Copy the set of subsystem state objects generated in
433 * find_existing_css_set() */
434 memcpy(res->subsys, template, sizeof(res->subsys));
436 write_lock(&css_set_lock);
437 /* Add reference counts and links from the new css_set. */
438 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
439 struct cgroup *cgrp = res->subsys[i]->cgroup;
440 struct cgroup_subsys *ss = subsys[i];
441 atomic_inc(&cgrp->count);
443 * We want to add a link once per cgroup, so we
444 * only do it for the first subsystem in each
447 if (ss->root->subsys_list.next == &ss->sibling) {
448 BUG_ON(list_empty(&tmp_cg_links));
449 link = list_entry(tmp_cg_links.next,
450 struct cg_cgroup_link,
452 list_del(&link->cgrp_link_list);
453 list_add(&link->cgrp_link_list, &cgrp->css_sets);
455 list_add(&link->cg_link_list, &res->cg_links);
458 if (list_empty(&rootnode.subsys_list)) {
459 link = list_entry(tmp_cg_links.next,
460 struct cg_cgroup_link,
462 list_del(&link->cgrp_link_list);
463 list_add(&link->cgrp_link_list, &dummytop->css_sets);
465 list_add(&link->cg_link_list, &res->cg_links);
468 BUG_ON(!list_empty(&tmp_cg_links));
470 /* Link this cgroup group into the list */
471 list_add(&res->list, &init_css_set.list);
473 INIT_LIST_HEAD(&res->tasks);
474 write_unlock(&css_set_lock);
480 * There is one global cgroup mutex. We also require taking
481 * task_lock() when dereferencing a task's cgroup subsys pointers.
482 * See "The task_lock() exception", at the end of this comment.
484 * A task must hold cgroup_mutex to modify cgroups.
486 * Any task can increment and decrement the count field without lock.
487 * So in general, code holding cgroup_mutex can't rely on the count
488 * field not changing. However, if the count goes to zero, then only
489 * cgroup_attach_task() can increment it again. Because a count of zero
490 * means that no tasks are currently attached, therefore there is no
491 * way a task attached to that cgroup can fork (the other way to
492 * increment the count). So code holding cgroup_mutex can safely
493 * assume that if the count is zero, it will stay zero. Similarly, if
494 * a task holds cgroup_mutex on a cgroup with zero count, it
495 * knows that the cgroup won't be removed, as cgroup_rmdir()
498 * The cgroup_common_file_write handler for operations that modify
499 * the cgroup hierarchy holds cgroup_mutex across the entire operation,
500 * single threading all such cgroup modifications across the system.
502 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
503 * (usually) take cgroup_mutex. These are the two most performance
504 * critical pieces of code here. The exception occurs on cgroup_exit(),
505 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
506 * is taken, and if the cgroup count is zero, a usermode call made
507 * to the release agent with the name of the cgroup (path relative to
508 * the root of cgroup file system) as the argument.
510 * A cgroup can only be deleted if both its 'count' of using tasks
511 * is zero, and its list of 'children' cgroups is empty. Since all
512 * tasks in the system use _some_ cgroup, and since there is always at
513 * least one task in the system (init, pid == 1), therefore, top_cgroup
514 * always has either children cgroups and/or using tasks. So we don't
515 * need a special hack to ensure that top_cgroup cannot be deleted.
517 * The task_lock() exception
519 * The need for this exception arises from the action of
520 * cgroup_attach_task(), which overwrites one tasks cgroup pointer with
521 * another. It does so using cgroup_mutex, however there are
522 * several performance critical places that need to reference
523 * task->cgroup without the expense of grabbing a system global
524 * mutex. Therefore except as noted below, when dereferencing or, as
525 * in cgroup_attach_task(), modifying a task'ss cgroup pointer we use
526 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
527 * the task_struct routinely used for such matters.
529 * P.S. One more locking exception. RCU is used to guard the
530 * update of a tasks cgroup pointer by cgroup_attach_task()
534 * cgroup_lock - lock out any changes to cgroup structures
537 void cgroup_lock(void)
539 mutex_lock(&cgroup_mutex);
543 * cgroup_unlock - release lock on cgroup changes
545 * Undo the lock taken in a previous cgroup_lock() call.
547 void cgroup_unlock(void)
549 mutex_unlock(&cgroup_mutex);
553 * A couple of forward declarations required, due to cyclic reference loop:
554 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
555 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
559 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
560 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
561 static int cgroup_populate_dir(struct cgroup *cgrp);
562 static struct inode_operations cgroup_dir_inode_operations;
563 static struct file_operations proc_cgroupstats_operations;
565 static struct backing_dev_info cgroup_backing_dev_info = {
566 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
569 static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
571 struct inode *inode = new_inode(sb);
574 inode->i_mode = mode;
575 inode->i_uid = current->fsuid;
576 inode->i_gid = current->fsgid;
578 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
579 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
585 * Call subsys's pre_destroy handler.
586 * This is called before css refcnt check.
588 static void cgroup_call_pre_destroy(struct cgroup *cgrp)
590 struct cgroup_subsys *ss;
591 for_each_subsys(cgrp->root, ss)
592 if (ss->pre_destroy && cgrp->subsys[ss->subsys_id])
593 ss->pre_destroy(ss, cgrp);
597 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
599 /* is dentry a directory ? if so, kfree() associated cgroup */
600 if (S_ISDIR(inode->i_mode)) {
601 struct cgroup *cgrp = dentry->d_fsdata;
602 struct cgroup_subsys *ss;
603 BUG_ON(!(cgroup_is_removed(cgrp)));
604 /* It's possible for external users to be holding css
605 * reference counts on a cgroup; css_put() needs to
606 * be able to access the cgroup after decrementing
607 * the reference count in order to know if it needs to
608 * queue the cgroup to be handled by the release
612 mutex_lock(&cgroup_mutex);
614 * Release the subsystem state objects.
616 for_each_subsys(cgrp->root, ss) {
617 if (cgrp->subsys[ss->subsys_id])
618 ss->destroy(ss, cgrp);
621 cgrp->root->number_of_cgroups--;
622 mutex_unlock(&cgroup_mutex);
624 /* Drop the active superblock reference that we took when we
625 * created the cgroup */
626 deactivate_super(cgrp->root->sb);
633 static void remove_dir(struct dentry *d)
635 struct dentry *parent = dget(d->d_parent);
638 simple_rmdir(parent->d_inode, d);
642 static void cgroup_clear_directory(struct dentry *dentry)
644 struct list_head *node;
646 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
647 spin_lock(&dcache_lock);
648 node = dentry->d_subdirs.next;
649 while (node != &dentry->d_subdirs) {
650 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
653 /* This should never be called on a cgroup
654 * directory with child cgroups */
655 BUG_ON(d->d_inode->i_mode & S_IFDIR);
657 spin_unlock(&dcache_lock);
659 simple_unlink(dentry->d_inode, d);
661 spin_lock(&dcache_lock);
663 node = dentry->d_subdirs.next;
665 spin_unlock(&dcache_lock);
669 * NOTE : the dentry must have been dget()'ed
671 static void cgroup_d_remove_dir(struct dentry *dentry)
673 cgroup_clear_directory(dentry);
675 spin_lock(&dcache_lock);
676 list_del_init(&dentry->d_u.d_child);
677 spin_unlock(&dcache_lock);
681 static int rebind_subsystems(struct cgroupfs_root *root,
682 unsigned long final_bits)
684 unsigned long added_bits, removed_bits;
685 struct cgroup *cgrp = &root->top_cgroup;
688 removed_bits = root->actual_subsys_bits & ~final_bits;
689 added_bits = final_bits & ~root->actual_subsys_bits;
690 /* Check that any added subsystems are currently free */
691 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
692 unsigned long long bit = 1ull << i;
693 struct cgroup_subsys *ss = subsys[i];
694 if (!(bit & added_bits))
696 if (ss->root != &rootnode) {
697 /* Subsystem isn't free */
702 /* Currently we don't handle adding/removing subsystems when
703 * any child cgroups exist. This is theoretically supportable
704 * but involves complex error handling, so it's being left until
706 if (!list_empty(&cgrp->children))
709 /* Process each subsystem */
710 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
711 struct cgroup_subsys *ss = subsys[i];
712 unsigned long bit = 1UL << i;
713 if (bit & added_bits) {
714 /* We're binding this subsystem to this hierarchy */
715 BUG_ON(cgrp->subsys[i]);
716 BUG_ON(!dummytop->subsys[i]);
717 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
718 cgrp->subsys[i] = dummytop->subsys[i];
719 cgrp->subsys[i]->cgroup = cgrp;
720 list_add(&ss->sibling, &root->subsys_list);
721 rcu_assign_pointer(ss->root, root);
725 } else if (bit & removed_bits) {
726 /* We're removing this subsystem */
727 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
728 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
730 ss->bind(ss, dummytop);
731 dummytop->subsys[i]->cgroup = dummytop;
732 cgrp->subsys[i] = NULL;
733 rcu_assign_pointer(subsys[i]->root, &rootnode);
734 list_del(&ss->sibling);
735 } else if (bit & final_bits) {
736 /* Subsystem state should already exist */
737 BUG_ON(!cgrp->subsys[i]);
739 /* Subsystem state shouldn't exist */
740 BUG_ON(cgrp->subsys[i]);
743 root->subsys_bits = root->actual_subsys_bits = final_bits;
749 static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
751 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
752 struct cgroup_subsys *ss;
754 mutex_lock(&cgroup_mutex);
755 for_each_subsys(root, ss)
756 seq_printf(seq, ",%s", ss->name);
757 if (test_bit(ROOT_NOPREFIX, &root->flags))
758 seq_puts(seq, ",noprefix");
759 if (strlen(root->release_agent_path))
760 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
761 mutex_unlock(&cgroup_mutex);
765 struct cgroup_sb_opts {
766 unsigned long subsys_bits;
771 /* Convert a hierarchy specifier into a bitmask of subsystems and
773 static int parse_cgroupfs_options(char *data,
774 struct cgroup_sb_opts *opts)
776 char *token, *o = data ?: "all";
778 opts->subsys_bits = 0;
780 opts->release_agent = NULL;
782 while ((token = strsep(&o, ",")) != NULL) {
785 if (!strcmp(token, "all")) {
786 opts->subsys_bits = (1 << CGROUP_SUBSYS_COUNT) - 1;
787 } else if (!strcmp(token, "noprefix")) {
788 set_bit(ROOT_NOPREFIX, &opts->flags);
789 } else if (!strncmp(token, "release_agent=", 14)) {
790 /* Specifying two release agents is forbidden */
791 if (opts->release_agent)
793 opts->release_agent = kzalloc(PATH_MAX, GFP_KERNEL);
794 if (!opts->release_agent)
796 strncpy(opts->release_agent, token + 14, PATH_MAX - 1);
797 opts->release_agent[PATH_MAX - 1] = 0;
799 struct cgroup_subsys *ss;
801 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
803 if (!strcmp(token, ss->name)) {
804 set_bit(i, &opts->subsys_bits);
808 if (i == CGROUP_SUBSYS_COUNT)
813 /* We can't have an empty hierarchy */
814 if (!opts->subsys_bits)
820 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
823 struct cgroupfs_root *root = sb->s_fs_info;
824 struct cgroup *cgrp = &root->top_cgroup;
825 struct cgroup_sb_opts opts;
827 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
828 mutex_lock(&cgroup_mutex);
830 /* See what subsystems are wanted */
831 ret = parse_cgroupfs_options(data, &opts);
835 /* Don't allow flags to change at remount */
836 if (opts.flags != root->flags) {
841 ret = rebind_subsystems(root, opts.subsys_bits);
843 /* (re)populate subsystem files */
845 cgroup_populate_dir(cgrp);
847 if (opts.release_agent)
848 strcpy(root->release_agent_path, opts.release_agent);
850 if (opts.release_agent)
851 kfree(opts.release_agent);
852 mutex_unlock(&cgroup_mutex);
853 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
857 static struct super_operations cgroup_ops = {
858 .statfs = simple_statfs,
859 .drop_inode = generic_delete_inode,
860 .show_options = cgroup_show_options,
861 .remount_fs = cgroup_remount,
864 static void init_cgroup_root(struct cgroupfs_root *root)
866 struct cgroup *cgrp = &root->top_cgroup;
867 INIT_LIST_HEAD(&root->subsys_list);
868 INIT_LIST_HEAD(&root->root_list);
869 root->number_of_cgroups = 1;
871 cgrp->top_cgroup = cgrp;
872 INIT_LIST_HEAD(&cgrp->sibling);
873 INIT_LIST_HEAD(&cgrp->children);
874 INIT_LIST_HEAD(&cgrp->css_sets);
875 INIT_LIST_HEAD(&cgrp->release_list);
878 static int cgroup_test_super(struct super_block *sb, void *data)
880 struct cgroupfs_root *new = data;
881 struct cgroupfs_root *root = sb->s_fs_info;
883 /* First check subsystems */
884 if (new->subsys_bits != root->subsys_bits)
887 /* Next check flags */
888 if (new->flags != root->flags)
894 static int cgroup_set_super(struct super_block *sb, void *data)
897 struct cgroupfs_root *root = data;
899 ret = set_anon_super(sb, NULL);
903 sb->s_fs_info = root;
906 sb->s_blocksize = PAGE_CACHE_SIZE;
907 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
908 sb->s_magic = CGROUP_SUPER_MAGIC;
909 sb->s_op = &cgroup_ops;
914 static int cgroup_get_rootdir(struct super_block *sb)
916 struct inode *inode =
917 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
918 struct dentry *dentry;
923 inode->i_op = &simple_dir_inode_operations;
924 inode->i_fop = &simple_dir_operations;
925 inode->i_op = &cgroup_dir_inode_operations;
926 /* directories start off with i_nlink == 2 (for "." entry) */
928 dentry = d_alloc_root(inode);
937 static int cgroup_get_sb(struct file_system_type *fs_type,
938 int flags, const char *unused_dev_name,
939 void *data, struct vfsmount *mnt)
941 struct cgroup_sb_opts opts;
943 struct super_block *sb;
944 struct cgroupfs_root *root;
945 struct list_head tmp_cg_links, *l;
946 INIT_LIST_HEAD(&tmp_cg_links);
948 /* First find the desired set of subsystems */
949 ret = parse_cgroupfs_options(data, &opts);
951 if (opts.release_agent)
952 kfree(opts.release_agent);
956 root = kzalloc(sizeof(*root), GFP_KERNEL);
960 init_cgroup_root(root);
961 root->subsys_bits = opts.subsys_bits;
962 root->flags = opts.flags;
963 if (opts.release_agent) {
964 strcpy(root->release_agent_path, opts.release_agent);
965 kfree(opts.release_agent);
968 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, root);
975 if (sb->s_fs_info != root) {
976 /* Reusing an existing superblock */
977 BUG_ON(sb->s_root == NULL);
982 struct cgroup *cgrp = &root->top_cgroup;
985 BUG_ON(sb->s_root != NULL);
987 ret = cgroup_get_rootdir(sb);
990 inode = sb->s_root->d_inode;
992 mutex_lock(&inode->i_mutex);
993 mutex_lock(&cgroup_mutex);
996 * We're accessing css_set_count without locking
997 * css_set_lock here, but that's OK - it can only be
998 * increased by someone holding cgroup_lock, and
999 * that's us. The worst that can happen is that we
1000 * have some link structures left over
1002 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
1004 mutex_unlock(&cgroup_mutex);
1005 mutex_unlock(&inode->i_mutex);
1006 goto drop_new_super;
1009 ret = rebind_subsystems(root, root->subsys_bits);
1010 if (ret == -EBUSY) {
1011 mutex_unlock(&cgroup_mutex);
1012 mutex_unlock(&inode->i_mutex);
1013 goto drop_new_super;
1016 /* EBUSY should be the only error here */
1019 list_add(&root->root_list, &roots);
1022 sb->s_root->d_fsdata = &root->top_cgroup;
1023 root->top_cgroup.dentry = sb->s_root;
1025 /* Link the top cgroup in this hierarchy into all
1026 * the css_set objects */
1027 write_lock(&css_set_lock);
1028 l = &init_css_set.list;
1031 struct cg_cgroup_link *link;
1032 cg = list_entry(l, struct css_set, list);
1033 BUG_ON(list_empty(&tmp_cg_links));
1034 link = list_entry(tmp_cg_links.next,
1035 struct cg_cgroup_link,
1037 list_del(&link->cgrp_link_list);
1039 list_add(&link->cgrp_link_list,
1040 &root->top_cgroup.css_sets);
1041 list_add(&link->cg_link_list, &cg->cg_links);
1043 } while (l != &init_css_set.list);
1044 write_unlock(&css_set_lock);
1046 free_cg_links(&tmp_cg_links);
1048 BUG_ON(!list_empty(&cgrp->sibling));
1049 BUG_ON(!list_empty(&cgrp->children));
1050 BUG_ON(root->number_of_cgroups != 1);
1052 cgroup_populate_dir(cgrp);
1053 mutex_unlock(&inode->i_mutex);
1054 mutex_unlock(&cgroup_mutex);
1057 return simple_set_mnt(mnt, sb);
1060 up_write(&sb->s_umount);
1061 deactivate_super(sb);
1062 free_cg_links(&tmp_cg_links);
1066 static void cgroup_kill_sb(struct super_block *sb) {
1067 struct cgroupfs_root *root = sb->s_fs_info;
1068 struct cgroup *cgrp = &root->top_cgroup;
1073 BUG_ON(root->number_of_cgroups != 1);
1074 BUG_ON(!list_empty(&cgrp->children));
1075 BUG_ON(!list_empty(&cgrp->sibling));
1077 mutex_lock(&cgroup_mutex);
1079 /* Rebind all subsystems back to the default hierarchy */
1080 ret = rebind_subsystems(root, 0);
1081 /* Shouldn't be able to fail ... */
1085 * Release all the links from css_sets to this hierarchy's
1088 write_lock(&css_set_lock);
1089 while (!list_empty(&cgrp->css_sets)) {
1090 struct cg_cgroup_link *link;
1091 link = list_entry(cgrp->css_sets.next,
1092 struct cg_cgroup_link, cgrp_link_list);
1093 list_del(&link->cg_link_list);
1094 list_del(&link->cgrp_link_list);
1097 write_unlock(&css_set_lock);
1099 if (!list_empty(&root->root_list)) {
1100 list_del(&root->root_list);
1103 mutex_unlock(&cgroup_mutex);
1106 kill_litter_super(sb);
1109 static struct file_system_type cgroup_fs_type = {
1111 .get_sb = cgroup_get_sb,
1112 .kill_sb = cgroup_kill_sb,
1115 static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1117 return dentry->d_fsdata;
1120 static inline struct cftype *__d_cft(struct dentry *dentry)
1122 return dentry->d_fsdata;
1126 * cgroup_path - generate the path of a cgroup
1127 * @cgrp: the cgroup in question
1128 * @buf: the buffer to write the path into
1129 * @buflen: the length of the buffer
1131 * Called with cgroup_mutex held. Writes path of cgroup into buf.
1132 * Returns 0 on success, -errno on error.
1134 int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1138 if (cgrp == dummytop) {
1140 * Inactive subsystems have no dentry for their root
1147 start = buf + buflen;
1151 int len = cgrp->dentry->d_name.len;
1152 if ((start -= len) < buf)
1153 return -ENAMETOOLONG;
1154 memcpy(start, cgrp->dentry->d_name.name, len);
1155 cgrp = cgrp->parent;
1161 return -ENAMETOOLONG;
1164 memmove(buf, start, buf + buflen - start);
1169 * Return the first subsystem attached to a cgroup's hierarchy, and
1173 static void get_first_subsys(const struct cgroup *cgrp,
1174 struct cgroup_subsys_state **css, int *subsys_id)
1176 const struct cgroupfs_root *root = cgrp->root;
1177 const struct cgroup_subsys *test_ss;
1178 BUG_ON(list_empty(&root->subsys_list));
1179 test_ss = list_entry(root->subsys_list.next,
1180 struct cgroup_subsys, sibling);
1182 *css = cgrp->subsys[test_ss->subsys_id];
1186 *subsys_id = test_ss->subsys_id;
1190 * cgroup_attach_task - attach task 'tsk' to cgroup 'cgrp'
1191 * @cgrp: the cgroup the task is attaching to
1192 * @tsk: the task to be attached
1194 * Call holding cgroup_mutex. May take task_lock of
1195 * the task 'tsk' during call.
1197 int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1200 struct cgroup_subsys *ss;
1201 struct cgroup *oldcgrp;
1202 struct css_set *cg = tsk->cgroups;
1203 struct css_set *newcg;
1204 struct cgroupfs_root *root = cgrp->root;
1207 get_first_subsys(cgrp, NULL, &subsys_id);
1209 /* Nothing to do if the task is already in that cgroup */
1210 oldcgrp = task_cgroup(tsk, subsys_id);
1211 if (cgrp == oldcgrp)
1214 for_each_subsys(root, ss) {
1215 if (ss->can_attach) {
1216 retval = ss->can_attach(ss, cgrp, tsk);
1223 * Locate or allocate a new css_set for this task,
1224 * based on its final set of cgroups
1226 newcg = find_css_set(cg, cgrp);
1231 if (tsk->flags & PF_EXITING) {
1236 rcu_assign_pointer(tsk->cgroups, newcg);
1239 /* Update the css_set linked lists if we're using them */
1240 write_lock(&css_set_lock);
1241 if (!list_empty(&tsk->cg_list)) {
1242 list_del(&tsk->cg_list);
1243 list_add(&tsk->cg_list, &newcg->tasks);
1245 write_unlock(&css_set_lock);
1247 for_each_subsys(root, ss) {
1249 ss->attach(ss, cgrp, oldcgrp, tsk);
1251 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1258 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with
1259 * cgroup_mutex, may take task_lock of task
1261 static int attach_task_by_pid(struct cgroup *cgrp, char *pidbuf)
1264 struct task_struct *tsk;
1267 if (sscanf(pidbuf, "%d", &pid) != 1)
1272 tsk = find_task_by_vpid(pid);
1273 if (!tsk || tsk->flags & PF_EXITING) {
1277 get_task_struct(tsk);
1280 if ((current->euid) && (current->euid != tsk->uid)
1281 && (current->euid != tsk->suid)) {
1282 put_task_struct(tsk);
1287 get_task_struct(tsk);
1290 ret = cgroup_attach_task(cgrp, tsk);
1291 put_task_struct(tsk);
1295 /* The various types of files and directories in a cgroup file system */
1296 enum cgroup_filetype {
1300 FILE_NOTIFY_ON_RELEASE,
1305 static ssize_t cgroup_write_uint(struct cgroup *cgrp, struct cftype *cft,
1307 const char __user *userbuf,
1308 size_t nbytes, loff_t *unused_ppos)
1317 if (nbytes >= sizeof(buffer))
1319 if (copy_from_user(buffer, userbuf, nbytes))
1322 buffer[nbytes] = 0; /* nul-terminate */
1324 /* strip newline if necessary */
1325 if (nbytes && (buffer[nbytes-1] == '\n'))
1326 buffer[nbytes-1] = 0;
1327 val = simple_strtoull(buffer, &end, 0);
1331 /* Pass to subsystem */
1332 retval = cft->write_uint(cgrp, cft, val);
1338 static ssize_t cgroup_common_file_write(struct cgroup *cgrp,
1341 const char __user *userbuf,
1342 size_t nbytes, loff_t *unused_ppos)
1344 enum cgroup_filetype type = cft->private;
1348 if (nbytes >= PATH_MAX)
1351 /* +1 for nul-terminator */
1352 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1356 if (copy_from_user(buffer, userbuf, nbytes)) {
1360 buffer[nbytes] = 0; /* nul-terminate */
1361 strstrip(buffer); /* strip -just- trailing whitespace */
1363 mutex_lock(&cgroup_mutex);
1366 * This was already checked for in cgroup_file_write(), but
1367 * check again now we're holding cgroup_mutex.
1369 if (cgroup_is_removed(cgrp)) {
1376 retval = attach_task_by_pid(cgrp, buffer);
1378 case FILE_NOTIFY_ON_RELEASE:
1379 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
1380 if (simple_strtoul(buffer, NULL, 10) != 0)
1381 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
1383 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
1385 case FILE_RELEASE_AGENT:
1386 BUILD_BUG_ON(sizeof(cgrp->root->release_agent_path) < PATH_MAX);
1387 strcpy(cgrp->root->release_agent_path, buffer);
1397 mutex_unlock(&cgroup_mutex);
1403 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
1404 size_t nbytes, loff_t *ppos)
1406 struct cftype *cft = __d_cft(file->f_dentry);
1407 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1409 if (!cft || cgroup_is_removed(cgrp))
1412 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
1413 if (cft->write_uint)
1414 return cgroup_write_uint(cgrp, cft, file, buf, nbytes, ppos);
1418 static ssize_t cgroup_read_uint(struct cgroup *cgrp, struct cftype *cft,
1420 char __user *buf, size_t nbytes,
1424 u64 val = cft->read_uint(cgrp, cft);
1425 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
1427 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1430 static ssize_t cgroup_common_file_read(struct cgroup *cgrp,
1434 size_t nbytes, loff_t *ppos)
1436 enum cgroup_filetype type = cft->private;
1441 if (!(page = (char *)__get_free_page(GFP_KERNEL)))
1447 case FILE_RELEASE_AGENT:
1449 struct cgroupfs_root *root;
1451 mutex_lock(&cgroup_mutex);
1453 n = strnlen(root->release_agent_path,
1454 sizeof(root->release_agent_path));
1455 n = min(n, (size_t) PAGE_SIZE);
1456 strncpy(s, root->release_agent_path, n);
1457 mutex_unlock(&cgroup_mutex);
1467 retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1469 free_page((unsigned long)page);
1473 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
1474 size_t nbytes, loff_t *ppos)
1476 struct cftype *cft = __d_cft(file->f_dentry);
1477 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1479 if (!cft || cgroup_is_removed(cgrp))
1483 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
1485 return cgroup_read_uint(cgrp, cft, file, buf, nbytes, ppos);
1489 static int cgroup_file_open(struct inode *inode, struct file *file)
1494 err = generic_file_open(inode, file);
1498 cft = __d_cft(file->f_dentry);
1502 err = cft->open(inode, file);
1509 static int cgroup_file_release(struct inode *inode, struct file *file)
1511 struct cftype *cft = __d_cft(file->f_dentry);
1513 return cft->release(inode, file);
1518 * cgroup_rename - Only allow simple rename of directories in place.
1520 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
1521 struct inode *new_dir, struct dentry *new_dentry)
1523 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1525 if (new_dentry->d_inode)
1527 if (old_dir != new_dir)
1529 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
1532 static struct file_operations cgroup_file_operations = {
1533 .read = cgroup_file_read,
1534 .write = cgroup_file_write,
1535 .llseek = generic_file_llseek,
1536 .open = cgroup_file_open,
1537 .release = cgroup_file_release,
1540 static struct inode_operations cgroup_dir_inode_operations = {
1541 .lookup = simple_lookup,
1542 .mkdir = cgroup_mkdir,
1543 .rmdir = cgroup_rmdir,
1544 .rename = cgroup_rename,
1547 static int cgroup_create_file(struct dentry *dentry, int mode,
1548 struct super_block *sb)
1550 static struct dentry_operations cgroup_dops = {
1551 .d_iput = cgroup_diput,
1554 struct inode *inode;
1558 if (dentry->d_inode)
1561 inode = cgroup_new_inode(mode, sb);
1565 if (S_ISDIR(mode)) {
1566 inode->i_op = &cgroup_dir_inode_operations;
1567 inode->i_fop = &simple_dir_operations;
1569 /* start off with i_nlink == 2 (for "." entry) */
1572 /* start with the directory inode held, so that we can
1573 * populate it without racing with another mkdir */
1574 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1575 } else if (S_ISREG(mode)) {
1577 inode->i_fop = &cgroup_file_operations;
1579 dentry->d_op = &cgroup_dops;
1580 d_instantiate(dentry, inode);
1581 dget(dentry); /* Extra count - pin the dentry in core */
1586 * cgroup_create_dir - create a directory for an object.
1587 * @cgrp: the cgroup we create the directory for. It must have a valid
1588 * ->parent field. And we are going to fill its ->dentry field.
1589 * @dentry: dentry of the new cgroup
1590 * @mode: mode to set on new directory.
1592 static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
1595 struct dentry *parent;
1598 parent = cgrp->parent->dentry;
1599 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
1601 dentry->d_fsdata = cgrp;
1602 inc_nlink(parent->d_inode);
1603 cgrp->dentry = dentry;
1611 int cgroup_add_file(struct cgroup *cgrp,
1612 struct cgroup_subsys *subsys,
1613 const struct cftype *cft)
1615 struct dentry *dir = cgrp->dentry;
1616 struct dentry *dentry;
1619 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
1620 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
1621 strcpy(name, subsys->name);
1624 strcat(name, cft->name);
1625 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
1626 dentry = lookup_one_len(name, dir, strlen(name));
1627 if (!IS_ERR(dentry)) {
1628 error = cgroup_create_file(dentry, 0644 | S_IFREG,
1631 dentry->d_fsdata = (void *)cft;
1634 error = PTR_ERR(dentry);
1638 int cgroup_add_files(struct cgroup *cgrp,
1639 struct cgroup_subsys *subsys,
1640 const struct cftype cft[],
1644 for (i = 0; i < count; i++) {
1645 err = cgroup_add_file(cgrp, subsys, &cft[i]);
1653 * cgroup_task_count - count the number of tasks in a cgroup.
1654 * @cgrp: the cgroup in question
1656 * Return the number of tasks in the cgroup.
1658 int cgroup_task_count(const struct cgroup *cgrp)
1661 struct list_head *l;
1663 read_lock(&css_set_lock);
1664 l = cgrp->css_sets.next;
1665 while (l != &cgrp->css_sets) {
1666 struct cg_cgroup_link *link =
1667 list_entry(l, struct cg_cgroup_link, cgrp_link_list);
1668 count += atomic_read(&link->cg->ref.refcount);
1671 read_unlock(&css_set_lock);
1676 * Advance a list_head iterator. The iterator should be positioned at
1677 * the start of a css_set
1679 static void cgroup_advance_iter(struct cgroup *cgrp,
1680 struct cgroup_iter *it)
1682 struct list_head *l = it->cg_link;
1683 struct cg_cgroup_link *link;
1686 /* Advance to the next non-empty css_set */
1689 if (l == &cgrp->css_sets) {
1693 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
1695 } while (list_empty(&cg->tasks));
1697 it->task = cg->tasks.next;
1701 * To reduce the fork() overhead for systems that are not actually
1702 * using their cgroups capability, we don't maintain the lists running
1703 * through each css_set to its tasks until we see the list actually
1704 * used - in other words after the first call to cgroup_iter_start().
1706 * The tasklist_lock is not held here, as do_each_thread() and
1707 * while_each_thread() are protected by RCU.
1709 void cgroup_enable_task_cg_lists(void)
1711 struct task_struct *p, *g;
1712 write_lock(&css_set_lock);
1713 use_task_css_set_links = 1;
1714 do_each_thread(g, p) {
1716 if (list_empty(&p->cg_list))
1717 list_add(&p->cg_list, &p->cgroups->tasks);
1719 } while_each_thread(g, p);
1720 write_unlock(&css_set_lock);
1723 void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
1726 * The first time anyone tries to iterate across a cgroup,
1727 * we need to enable the list linking each css_set to its
1728 * tasks, and fix up all existing tasks.
1730 if (!use_task_css_set_links)
1731 cgroup_enable_task_cg_lists();
1733 read_lock(&css_set_lock);
1734 it->cg_link = &cgrp->css_sets;
1735 cgroup_advance_iter(cgrp, it);
1738 struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
1739 struct cgroup_iter *it)
1741 struct task_struct *res;
1742 struct list_head *l = it->task;
1744 /* If the iterator cg is NULL, we have no tasks */
1747 res = list_entry(l, struct task_struct, cg_list);
1748 /* Advance iterator to find next entry */
1750 if (l == &res->cgroups->tasks) {
1751 /* We reached the end of this task list - move on to
1752 * the next cg_cgroup_link */
1753 cgroup_advance_iter(cgrp, it);
1760 void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
1762 read_unlock(&css_set_lock);
1765 static inline int started_after_time(struct task_struct *t1,
1766 struct timespec *time,
1767 struct task_struct *t2)
1769 int start_diff = timespec_compare(&t1->start_time, time);
1770 if (start_diff > 0) {
1772 } else if (start_diff < 0) {
1776 * Arbitrarily, if two processes started at the same
1777 * time, we'll say that the lower pointer value
1778 * started first. Note that t2 may have exited by now
1779 * so this may not be a valid pointer any longer, but
1780 * that's fine - it still serves to distinguish
1781 * between two tasks started (effectively) simultaneously.
1788 * This function is a callback from heap_insert() and is used to order
1790 * In this case we order the heap in descending task start time.
1792 static inline int started_after(void *p1, void *p2)
1794 struct task_struct *t1 = p1;
1795 struct task_struct *t2 = p2;
1796 return started_after_time(t1, &t2->start_time, t2);
1800 * cgroup_scan_tasks - iterate though all the tasks in a cgroup
1801 * @scan: struct cgroup_scanner containing arguments for the scan
1803 * Arguments include pointers to callback functions test_task() and
1805 * Iterate through all the tasks in a cgroup, calling test_task() for each,
1806 * and if it returns true, call process_task() for it also.
1807 * The test_task pointer may be NULL, meaning always true (select all tasks).
1808 * Effectively duplicates cgroup_iter_{start,next,end}()
1809 * but does not lock css_set_lock for the call to process_task().
1810 * The struct cgroup_scanner may be embedded in any structure of the caller's
1812 * It is guaranteed that process_task() will act on every task that
1813 * is a member of the cgroup for the duration of this call. This
1814 * function may or may not call process_task() for tasks that exit
1815 * or move to a different cgroup during the call, or are forked or
1816 * move into the cgroup during the call.
1818 * Note that test_task() may be called with locks held, and may in some
1819 * situations be called multiple times for the same task, so it should
1821 * If the heap pointer in the struct cgroup_scanner is non-NULL, a heap has been
1822 * pre-allocated and will be used for heap operations (and its "gt" member will
1823 * be overwritten), else a temporary heap will be used (allocation of which
1824 * may cause this function to fail).
1826 int cgroup_scan_tasks(struct cgroup_scanner *scan)
1829 struct cgroup_iter it;
1830 struct task_struct *p, *dropped;
1831 /* Never dereference latest_task, since it's not refcounted */
1832 struct task_struct *latest_task = NULL;
1833 struct ptr_heap tmp_heap;
1834 struct ptr_heap *heap;
1835 struct timespec latest_time = { 0, 0 };
1838 /* The caller supplied our heap and pre-allocated its memory */
1840 heap->gt = &started_after;
1842 /* We need to allocate our own heap memory */
1844 retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
1846 /* cannot allocate the heap */
1852 * Scan tasks in the cgroup, using the scanner's "test_task" callback
1853 * to determine which are of interest, and using the scanner's
1854 * "process_task" callback to process any of them that need an update.
1855 * Since we don't want to hold any locks during the task updates,
1856 * gather tasks to be processed in a heap structure.
1857 * The heap is sorted by descending task start time.
1858 * If the statically-sized heap fills up, we overflow tasks that
1859 * started later, and in future iterations only consider tasks that
1860 * started after the latest task in the previous pass. This
1861 * guarantees forward progress and that we don't miss any tasks.
1864 cgroup_iter_start(scan->cg, &it);
1865 while ((p = cgroup_iter_next(scan->cg, &it))) {
1867 * Only affect tasks that qualify per the caller's callback,
1868 * if he provided one
1870 if (scan->test_task && !scan->test_task(p, scan))
1873 * Only process tasks that started after the last task
1876 if (!started_after_time(p, &latest_time, latest_task))
1878 dropped = heap_insert(heap, p);
1879 if (dropped == NULL) {
1881 * The new task was inserted; the heap wasn't
1885 } else if (dropped != p) {
1887 * The new task was inserted, and pushed out a
1891 put_task_struct(dropped);
1894 * Else the new task was newer than anything already in
1895 * the heap and wasn't inserted
1898 cgroup_iter_end(scan->cg, &it);
1901 for (i = 0; i < heap->size; i++) {
1902 struct task_struct *p = heap->ptrs[i];
1904 latest_time = p->start_time;
1907 /* Process the task per the caller's callback */
1908 scan->process_task(p, scan);
1912 * If we had to process any tasks at all, scan again
1913 * in case some of them were in the middle of forking
1914 * children that didn't get processed.
1915 * Not the most efficient way to do it, but it avoids
1916 * having to take callback_mutex in the fork path
1920 if (heap == &tmp_heap)
1921 heap_free(&tmp_heap);
1926 * Stuff for reading the 'tasks' file.
1928 * Reading this file can return large amounts of data if a cgroup has
1929 * *lots* of attached tasks. So it may need several calls to read(),
1930 * but we cannot guarantee that the information we produce is correct
1931 * unless we produce it entirely atomically.
1933 * Upon tasks file open(), a struct ctr_struct is allocated, that
1934 * will have a pointer to an array (also allocated here). The struct
1935 * ctr_struct * is stored in file->private_data. Its resources will
1936 * be freed by release() when the file is closed. The array is used
1937 * to sprintf the PIDs and then used by read().
1945 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
1946 * 'cgrp'. Return actual number of pids loaded. No need to
1947 * task_lock(p) when reading out p->cgroup, since we're in an RCU
1948 * read section, so the css_set can't go away, and is
1949 * immutable after creation.
1951 static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cgrp)
1954 struct cgroup_iter it;
1955 struct task_struct *tsk;
1956 cgroup_iter_start(cgrp, &it);
1957 while ((tsk = cgroup_iter_next(cgrp, &it))) {
1958 if (unlikely(n == npids))
1960 pidarray[n++] = task_pid_vnr(tsk);
1962 cgroup_iter_end(cgrp, &it);
1967 * cgroupstats_build - build and fill cgroupstats
1968 * @stats: cgroupstats to fill information into
1969 * @dentry: A dentry entry belonging to the cgroup for which stats have
1972 * Build and fill cgroupstats so that taskstats can export it to user
1975 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
1978 struct cgroup *cgrp;
1979 struct cgroup_iter it;
1980 struct task_struct *tsk;
1982 * Validate dentry by checking the superblock operations
1984 if (dentry->d_sb->s_op != &cgroup_ops)
1988 cgrp = dentry->d_fsdata;
1991 cgroup_iter_start(cgrp, &it);
1992 while ((tsk = cgroup_iter_next(cgrp, &it))) {
1993 switch (tsk->state) {
1995 stats->nr_running++;
1997 case TASK_INTERRUPTIBLE:
1998 stats->nr_sleeping++;
2000 case TASK_UNINTERRUPTIBLE:
2001 stats->nr_uninterruptible++;
2004 stats->nr_stopped++;
2007 if (delayacct_is_task_waiting_on_io(tsk))
2008 stats->nr_io_wait++;
2012 cgroup_iter_end(cgrp, &it);
2019 static int cmppid(const void *a, const void *b)
2021 return *(pid_t *)a - *(pid_t *)b;
2025 * Convert array 'a' of 'npids' pid_t's to a string of newline separated
2026 * decimal pids in 'buf'. Don't write more than 'sz' chars, but return
2027 * count 'cnt' of how many chars would be written if buf were large enough.
2029 static int pid_array_to_buf(char *buf, int sz, pid_t *a, int npids)
2034 for (i = 0; i < npids; i++)
2035 cnt += snprintf(buf + cnt, max(sz - cnt, 0), "%d\n", a[i]);
2040 * Handle an open on 'tasks' file. Prepare a buffer listing the
2041 * process id's of tasks currently attached to the cgroup being opened.
2043 * Does not require any specific cgroup mutexes, and does not take any.
2045 static int cgroup_tasks_open(struct inode *unused, struct file *file)
2047 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
2048 struct ctr_struct *ctr;
2053 if (!(file->f_mode & FMODE_READ))
2056 ctr = kmalloc(sizeof(*ctr), GFP_KERNEL);
2061 * If cgroup gets more users after we read count, we won't have
2062 * enough space - tough. This race is indistinguishable to the
2063 * caller from the case that the additional cgroup users didn't
2064 * show up until sometime later on.
2066 npids = cgroup_task_count(cgrp);
2068 pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL);
2072 npids = pid_array_load(pidarray, npids, cgrp);
2073 sort(pidarray, npids, sizeof(pid_t), cmppid, NULL);
2075 /* Call pid_array_to_buf() twice, first just to get bufsz */
2076 ctr->bufsz = pid_array_to_buf(&c, sizeof(c), pidarray, npids) + 1;
2077 ctr->buf = kmalloc(ctr->bufsz, GFP_KERNEL);
2080 ctr->bufsz = pid_array_to_buf(ctr->buf, ctr->bufsz, pidarray, npids);
2087 file->private_data = ctr;
2098 static ssize_t cgroup_tasks_read(struct cgroup *cgrp,
2100 struct file *file, char __user *buf,
2101 size_t nbytes, loff_t *ppos)
2103 struct ctr_struct *ctr = file->private_data;
2105 return simple_read_from_buffer(buf, nbytes, ppos, ctr->buf, ctr->bufsz);
2108 static int cgroup_tasks_release(struct inode *unused_inode,
2111 struct ctr_struct *ctr;
2113 if (file->f_mode & FMODE_READ) {
2114 ctr = file->private_data;
2121 static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
2124 return notify_on_release(cgrp);
2127 static u64 cgroup_read_releasable(struct cgroup *cgrp, struct cftype *cft)
2129 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
2133 * for the common functions, 'private' gives the type of file
2135 static struct cftype files[] = {
2138 .open = cgroup_tasks_open,
2139 .read = cgroup_tasks_read,
2140 .write = cgroup_common_file_write,
2141 .release = cgroup_tasks_release,
2142 .private = FILE_TASKLIST,
2146 .name = "notify_on_release",
2147 .read_uint = cgroup_read_notify_on_release,
2148 .write = cgroup_common_file_write,
2149 .private = FILE_NOTIFY_ON_RELEASE,
2153 .name = "releasable",
2154 .read_uint = cgroup_read_releasable,
2155 .private = FILE_RELEASABLE,
2159 static struct cftype cft_release_agent = {
2160 .name = "release_agent",
2161 .read = cgroup_common_file_read,
2162 .write = cgroup_common_file_write,
2163 .private = FILE_RELEASE_AGENT,
2166 static int cgroup_populate_dir(struct cgroup *cgrp)
2169 struct cgroup_subsys *ss;
2171 /* First clear out any existing files */
2172 cgroup_clear_directory(cgrp->dentry);
2174 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
2178 if (cgrp == cgrp->top_cgroup) {
2179 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
2183 for_each_subsys(cgrp->root, ss) {
2184 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
2191 static void init_cgroup_css(struct cgroup_subsys_state *css,
2192 struct cgroup_subsys *ss,
2193 struct cgroup *cgrp)
2196 atomic_set(&css->refcnt, 0);
2198 if (cgrp == dummytop)
2199 set_bit(CSS_ROOT, &css->flags);
2200 BUG_ON(cgrp->subsys[ss->subsys_id]);
2201 cgrp->subsys[ss->subsys_id] = css;
2205 * cgroup_create - create a cgroup
2206 * @parent: cgroup that will be parent of the new cgroup
2207 * @dentry: dentry of the new cgroup
2208 * @mode: mode to set on new inode
2210 * Must be called with the mutex on the parent inode held
2212 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
2215 struct cgroup *cgrp;
2216 struct cgroupfs_root *root = parent->root;
2218 struct cgroup_subsys *ss;
2219 struct super_block *sb = root->sb;
2221 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
2225 /* Grab a reference on the superblock so the hierarchy doesn't
2226 * get deleted on unmount if there are child cgroups. This
2227 * can be done outside cgroup_mutex, since the sb can't
2228 * disappear while someone has an open control file on the
2230 atomic_inc(&sb->s_active);
2232 mutex_lock(&cgroup_mutex);
2235 INIT_LIST_HEAD(&cgrp->sibling);
2236 INIT_LIST_HEAD(&cgrp->children);
2237 INIT_LIST_HEAD(&cgrp->css_sets);
2238 INIT_LIST_HEAD(&cgrp->release_list);
2240 cgrp->parent = parent;
2241 cgrp->root = parent->root;
2242 cgrp->top_cgroup = parent->top_cgroup;
2244 for_each_subsys(root, ss) {
2245 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
2250 init_cgroup_css(css, ss, cgrp);
2253 list_add(&cgrp->sibling, &cgrp->parent->children);
2254 root->number_of_cgroups++;
2256 err = cgroup_create_dir(cgrp, dentry, mode);
2260 /* The cgroup directory was pre-locked for us */
2261 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
2263 err = cgroup_populate_dir(cgrp);
2264 /* If err < 0, we have a half-filled directory - oh well ;) */
2266 mutex_unlock(&cgroup_mutex);
2267 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
2273 list_del(&cgrp->sibling);
2274 root->number_of_cgroups--;
2278 for_each_subsys(root, ss) {
2279 if (cgrp->subsys[ss->subsys_id])
2280 ss->destroy(ss, cgrp);
2283 mutex_unlock(&cgroup_mutex);
2285 /* Release the reference count that we took on the superblock */
2286 deactivate_super(sb);
2292 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2294 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
2296 /* the vfs holds inode->i_mutex already */
2297 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
2300 static inline int cgroup_has_css_refs(struct cgroup *cgrp)
2302 /* Check the reference count on each subsystem. Since we
2303 * already established that there are no tasks in the
2304 * cgroup, if the css refcount is also 0, then there should
2305 * be no outstanding references, so the subsystem is safe to
2306 * destroy. We scan across all subsystems rather than using
2307 * the per-hierarchy linked list of mounted subsystems since
2308 * we can be called via check_for_release() with no
2309 * synchronization other than RCU, and the subsystem linked
2310 * list isn't RCU-safe */
2312 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2313 struct cgroup_subsys *ss = subsys[i];
2314 struct cgroup_subsys_state *css;
2315 /* Skip subsystems not in this hierarchy */
2316 if (ss->root != cgrp->root)
2318 css = cgrp->subsys[ss->subsys_id];
2319 /* When called from check_for_release() it's possible
2320 * that by this point the cgroup has been removed
2321 * and the css deleted. But a false-positive doesn't
2322 * matter, since it can only happen if the cgroup
2323 * has been deleted and hence no longer needs the
2324 * release agent to be called anyway. */
2325 if (css && atomic_read(&css->refcnt))
2331 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
2333 struct cgroup *cgrp = dentry->d_fsdata;
2335 struct cgroup *parent;
2336 struct super_block *sb;
2337 struct cgroupfs_root *root;
2339 /* the vfs holds both inode->i_mutex already */
2341 mutex_lock(&cgroup_mutex);
2342 if (atomic_read(&cgrp->count) != 0) {
2343 mutex_unlock(&cgroup_mutex);
2346 if (!list_empty(&cgrp->children)) {
2347 mutex_unlock(&cgroup_mutex);
2351 parent = cgrp->parent;
2356 * Call pre_destroy handlers of subsys. Notify subsystems
2357 * that rmdir() request comes.
2359 cgroup_call_pre_destroy(cgrp);
2361 if (cgroup_has_css_refs(cgrp)) {
2362 mutex_unlock(&cgroup_mutex);
2366 spin_lock(&release_list_lock);
2367 set_bit(CGRP_REMOVED, &cgrp->flags);
2368 if (!list_empty(&cgrp->release_list))
2369 list_del(&cgrp->release_list);
2370 spin_unlock(&release_list_lock);
2371 /* delete my sibling from parent->children */
2372 list_del(&cgrp->sibling);
2373 spin_lock(&cgrp->dentry->d_lock);
2374 d = dget(cgrp->dentry);
2375 cgrp->dentry = NULL;
2376 spin_unlock(&d->d_lock);
2378 cgroup_d_remove_dir(d);
2381 set_bit(CGRP_RELEASABLE, &parent->flags);
2382 check_for_release(parent);
2384 mutex_unlock(&cgroup_mutex);
2388 static void cgroup_init_subsys(struct cgroup_subsys *ss)
2390 struct cgroup_subsys_state *css;
2391 struct list_head *l;
2393 printk(KERN_INFO "Initializing cgroup subsys %s\n", ss->name);
2395 /* Create the top cgroup state for this subsystem */
2396 ss->root = &rootnode;
2397 css = ss->create(ss, dummytop);
2398 /* We don't handle early failures gracefully */
2399 BUG_ON(IS_ERR(css));
2400 init_cgroup_css(css, ss, dummytop);
2402 /* Update all cgroup groups to contain a subsys
2403 * pointer to this state - since the subsystem is
2404 * newly registered, all tasks and hence all cgroup
2405 * groups are in the subsystem's top cgroup. */
2406 write_lock(&css_set_lock);
2407 l = &init_css_set.list;
2409 struct css_set *cg =
2410 list_entry(l, struct css_set, list);
2411 cg->subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
2413 } while (l != &init_css_set.list);
2414 write_unlock(&css_set_lock);
2416 /* If this subsystem requested that it be notified with fork
2417 * events, we should send it one now for every process in the
2420 struct task_struct *g, *p;
2422 read_lock(&tasklist_lock);
2423 do_each_thread(g, p) {
2425 } while_each_thread(g, p);
2426 read_unlock(&tasklist_lock);
2429 need_forkexit_callback |= ss->fork || ss->exit;
2435 * cgroup_init_early - cgroup initialization at system boot
2437 * Initialize cgroups at system boot, and initialize any
2438 * subsystems that request early init.
2440 int __init cgroup_init_early(void)
2443 kref_init(&init_css_set.ref);
2444 kref_get(&init_css_set.ref);
2445 INIT_LIST_HEAD(&init_css_set.list);
2446 INIT_LIST_HEAD(&init_css_set.cg_links);
2447 INIT_LIST_HEAD(&init_css_set.tasks);
2449 init_cgroup_root(&rootnode);
2450 list_add(&rootnode.root_list, &roots);
2452 init_task.cgroups = &init_css_set;
2454 init_css_set_link.cg = &init_css_set;
2455 list_add(&init_css_set_link.cgrp_link_list,
2456 &rootnode.top_cgroup.css_sets);
2457 list_add(&init_css_set_link.cg_link_list,
2458 &init_css_set.cg_links);
2460 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2461 struct cgroup_subsys *ss = subsys[i];
2464 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
2465 BUG_ON(!ss->create);
2466 BUG_ON(!ss->destroy);
2467 if (ss->subsys_id != i) {
2468 printk(KERN_ERR "cgroup: Subsys %s id == %d\n",
2469 ss->name, ss->subsys_id);
2474 cgroup_init_subsys(ss);
2480 * cgroup_init - cgroup initialization
2482 * Register cgroup filesystem and /proc file, and initialize
2483 * any subsystems that didn't request early init.
2485 int __init cgroup_init(void)
2489 struct proc_dir_entry *entry;
2491 err = bdi_init(&cgroup_backing_dev_info);
2495 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2496 struct cgroup_subsys *ss = subsys[i];
2497 if (!ss->early_init)
2498 cgroup_init_subsys(ss);
2501 err = register_filesystem(&cgroup_fs_type);
2505 entry = create_proc_entry("cgroups", 0, NULL);
2507 entry->proc_fops = &proc_cgroupstats_operations;
2511 bdi_destroy(&cgroup_backing_dev_info);
2517 * proc_cgroup_show()
2518 * - Print task's cgroup paths into seq_file, one line for each hierarchy
2519 * - Used for /proc/<pid>/cgroup.
2520 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
2521 * doesn't really matter if tsk->cgroup changes after we read it,
2522 * and we take cgroup_mutex, keeping cgroup_attach_task() from changing it
2523 * anyway. No need to check that tsk->cgroup != NULL, thanks to
2524 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
2525 * cgroup to top_cgroup.
2528 /* TODO: Use a proper seq_file iterator */
2529 static int proc_cgroup_show(struct seq_file *m, void *v)
2532 struct task_struct *tsk;
2535 struct cgroupfs_root *root;
2538 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2544 tsk = get_pid_task(pid, PIDTYPE_PID);
2550 mutex_lock(&cgroup_mutex);
2552 for_each_root(root) {
2553 struct cgroup_subsys *ss;
2554 struct cgroup *cgrp;
2558 /* Skip this hierarchy if it has no active subsystems */
2559 if (!root->actual_subsys_bits)
2561 for_each_subsys(root, ss)
2562 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
2564 get_first_subsys(&root->top_cgroup, NULL, &subsys_id);
2565 cgrp = task_cgroup(tsk, subsys_id);
2566 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
2574 mutex_unlock(&cgroup_mutex);
2575 put_task_struct(tsk);
2582 static int cgroup_open(struct inode *inode, struct file *file)
2584 struct pid *pid = PROC_I(inode)->pid;
2585 return single_open(file, proc_cgroup_show, pid);
2588 struct file_operations proc_cgroup_operations = {
2589 .open = cgroup_open,
2591 .llseek = seq_lseek,
2592 .release = single_release,
2595 /* Display information about each subsystem and each hierarchy */
2596 static int proc_cgroupstats_show(struct seq_file *m, void *v)
2600 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\n");
2601 mutex_lock(&cgroup_mutex);
2602 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2603 struct cgroup_subsys *ss = subsys[i];
2604 seq_printf(m, "%s\t%lu\t%d\n",
2605 ss->name, ss->root->subsys_bits,
2606 ss->root->number_of_cgroups);
2608 mutex_unlock(&cgroup_mutex);
2612 static int cgroupstats_open(struct inode *inode, struct file *file)
2614 return single_open(file, proc_cgroupstats_show, 0);
2617 static struct file_operations proc_cgroupstats_operations = {
2618 .open = cgroupstats_open,
2620 .llseek = seq_lseek,
2621 .release = single_release,
2625 * cgroup_fork - attach newly forked task to its parents cgroup.
2626 * @child: pointer to task_struct of forking parent process.
2628 * Description: A task inherits its parent's cgroup at fork().
2630 * A pointer to the shared css_set was automatically copied in
2631 * fork.c by dup_task_struct(). However, we ignore that copy, since
2632 * it was not made under the protection of RCU or cgroup_mutex, so
2633 * might no longer be a valid cgroup pointer. cgroup_attach_task() might
2634 * have already changed current->cgroups, allowing the previously
2635 * referenced cgroup group to be removed and freed.
2637 * At the point that cgroup_fork() is called, 'current' is the parent
2638 * task, and the passed argument 'child' points to the child task.
2640 void cgroup_fork(struct task_struct *child)
2643 child->cgroups = current->cgroups;
2644 get_css_set(child->cgroups);
2645 task_unlock(current);
2646 INIT_LIST_HEAD(&child->cg_list);
2650 * cgroup_fork_callbacks - run fork callbacks
2651 * @child: the new task
2653 * Called on a new task very soon before adding it to the
2654 * tasklist. No need to take any locks since no-one can
2655 * be operating on this task.
2657 void cgroup_fork_callbacks(struct task_struct *child)
2659 if (need_forkexit_callback) {
2661 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2662 struct cgroup_subsys *ss = subsys[i];
2664 ss->fork(ss, child);
2670 * cgroup_post_fork - called on a new task after adding it to the task list
2671 * @child: the task in question
2673 * Adds the task to the list running through its css_set if necessary.
2674 * Has to be after the task is visible on the task list in case we race
2675 * with the first call to cgroup_iter_start() - to guarantee that the
2676 * new task ends up on its list.
2678 void cgroup_post_fork(struct task_struct *child)
2680 if (use_task_css_set_links) {
2681 write_lock(&css_set_lock);
2682 if (list_empty(&child->cg_list))
2683 list_add(&child->cg_list, &child->cgroups->tasks);
2684 write_unlock(&css_set_lock);
2688 * cgroup_exit - detach cgroup from exiting task
2689 * @tsk: pointer to task_struct of exiting process
2690 * @run_callback: run exit callbacks?
2692 * Description: Detach cgroup from @tsk and release it.
2694 * Note that cgroups marked notify_on_release force every task in
2695 * them to take the global cgroup_mutex mutex when exiting.
2696 * This could impact scaling on very large systems. Be reluctant to
2697 * use notify_on_release cgroups where very high task exit scaling
2698 * is required on large systems.
2700 * the_top_cgroup_hack:
2702 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
2704 * We call cgroup_exit() while the task is still competent to
2705 * handle notify_on_release(), then leave the task attached to the
2706 * root cgroup in each hierarchy for the remainder of its exit.
2708 * To do this properly, we would increment the reference count on
2709 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
2710 * code we would add a second cgroup function call, to drop that
2711 * reference. This would just create an unnecessary hot spot on
2712 * the top_cgroup reference count, to no avail.
2714 * Normally, holding a reference to a cgroup without bumping its
2715 * count is unsafe. The cgroup could go away, or someone could
2716 * attach us to a different cgroup, decrementing the count on
2717 * the first cgroup that we never incremented. But in this case,
2718 * top_cgroup isn't going away, and either task has PF_EXITING set,
2719 * which wards off any cgroup_attach_task() attempts, or task is a failed
2720 * fork, never visible to cgroup_attach_task.
2722 void cgroup_exit(struct task_struct *tsk, int run_callbacks)
2727 if (run_callbacks && need_forkexit_callback) {
2728 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2729 struct cgroup_subsys *ss = subsys[i];
2736 * Unlink from the css_set task list if necessary.
2737 * Optimistically check cg_list before taking
2740 if (!list_empty(&tsk->cg_list)) {
2741 write_lock(&css_set_lock);
2742 if (!list_empty(&tsk->cg_list))
2743 list_del(&tsk->cg_list);
2744 write_unlock(&css_set_lock);
2747 /* Reassign the task to the init_css_set. */
2750 tsk->cgroups = &init_css_set;
2753 put_css_set_taskexit(cg);
2757 * cgroup_clone - clone the cgroup the given subsystem is attached to
2758 * @tsk: the task to be moved
2759 * @subsys: the given subsystem
2761 * Duplicate the current cgroup in the hierarchy that the given
2762 * subsystem is attached to, and move this task into the new
2765 int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys)
2767 struct dentry *dentry;
2769 char nodename[MAX_CGROUP_TYPE_NAMELEN];
2770 struct cgroup *parent, *child;
2771 struct inode *inode;
2773 struct cgroupfs_root *root;
2774 struct cgroup_subsys *ss;
2776 /* We shouldn't be called by an unregistered subsystem */
2777 BUG_ON(!subsys->active);
2779 /* First figure out what hierarchy and cgroup we're dealing
2780 * with, and pin them so we can drop cgroup_mutex */
2781 mutex_lock(&cgroup_mutex);
2783 root = subsys->root;
2784 if (root == &rootnode) {
2786 "Not cloning cgroup for unused subsystem %s\n",
2788 mutex_unlock(&cgroup_mutex);
2792 parent = task_cgroup(tsk, subsys->subsys_id);
2794 snprintf(nodename, MAX_CGROUP_TYPE_NAMELEN, "node_%d", tsk->pid);
2796 /* Pin the hierarchy */
2797 atomic_inc(&parent->root->sb->s_active);
2799 /* Keep the cgroup alive */
2801 mutex_unlock(&cgroup_mutex);
2803 /* Now do the VFS work to create a cgroup */
2804 inode = parent->dentry->d_inode;
2806 /* Hold the parent directory mutex across this operation to
2807 * stop anyone else deleting the new cgroup */
2808 mutex_lock(&inode->i_mutex);
2809 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
2810 if (IS_ERR(dentry)) {
2812 "cgroup: Couldn't allocate dentry for %s: %ld\n", nodename,
2814 ret = PTR_ERR(dentry);
2818 /* Create the cgroup directory, which also creates the cgroup */
2819 ret = vfs_mkdir(inode, dentry, S_IFDIR | 0755);
2820 child = __d_cgrp(dentry);
2824 "Failed to create cgroup %s: %d\n", nodename,
2831 "Couldn't find new cgroup %s\n", nodename);
2836 /* The cgroup now exists. Retake cgroup_mutex and check
2837 * that we're still in the same state that we thought we
2839 mutex_lock(&cgroup_mutex);
2840 if ((root != subsys->root) ||
2841 (parent != task_cgroup(tsk, subsys->subsys_id))) {
2842 /* Aargh, we raced ... */
2843 mutex_unlock(&inode->i_mutex);
2846 deactivate_super(parent->root->sb);
2847 /* The cgroup is still accessible in the VFS, but
2848 * we're not going to try to rmdir() it at this
2851 "Race in cgroup_clone() - leaking cgroup %s\n",
2856 /* do any required auto-setup */
2857 for_each_subsys(root, ss) {
2859 ss->post_clone(ss, child);
2862 /* All seems fine. Finish by moving the task into the new cgroup */
2863 ret = cgroup_attach_task(child, tsk);
2864 mutex_unlock(&cgroup_mutex);
2867 mutex_unlock(&inode->i_mutex);
2869 mutex_lock(&cgroup_mutex);
2871 mutex_unlock(&cgroup_mutex);
2872 deactivate_super(parent->root->sb);
2877 * cgroup_is_descendant - see if @cgrp is a descendant of current task's cgrp
2878 * @cgrp: the cgroup in question
2880 * See if @cgrp is a descendant of the current task's cgroup in
2881 * the appropriate hierarchy.
2883 * If we are sending in dummytop, then presumably we are creating
2884 * the top cgroup in the subsystem.
2886 * Called only by the ns (nsproxy) cgroup.
2888 int cgroup_is_descendant(const struct cgroup *cgrp)
2891 struct cgroup *target;
2894 if (cgrp == dummytop)
2897 get_first_subsys(cgrp, NULL, &subsys_id);
2898 target = task_cgroup(current, subsys_id);
2899 while (cgrp != target && cgrp!= cgrp->top_cgroup)
2900 cgrp = cgrp->parent;
2901 ret = (cgrp == target);
2905 static void check_for_release(struct cgroup *cgrp)
2907 /* All of these checks rely on RCU to keep the cgroup
2908 * structure alive */
2909 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
2910 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
2911 /* Control Group is currently removeable. If it's not
2912 * already queued for a userspace notification, queue
2914 int need_schedule_work = 0;
2915 spin_lock(&release_list_lock);
2916 if (!cgroup_is_removed(cgrp) &&
2917 list_empty(&cgrp->release_list)) {
2918 list_add(&cgrp->release_list, &release_list);
2919 need_schedule_work = 1;
2921 spin_unlock(&release_list_lock);
2922 if (need_schedule_work)
2923 schedule_work(&release_agent_work);
2927 void __css_put(struct cgroup_subsys_state *css)
2929 struct cgroup *cgrp = css->cgroup;
2931 if (atomic_dec_and_test(&css->refcnt) && notify_on_release(cgrp)) {
2932 set_bit(CGRP_RELEASABLE, &cgrp->flags);
2933 check_for_release(cgrp);
2939 * Notify userspace when a cgroup is released, by running the
2940 * configured release agent with the name of the cgroup (path
2941 * relative to the root of cgroup file system) as the argument.
2943 * Most likely, this user command will try to rmdir this cgroup.
2945 * This races with the possibility that some other task will be
2946 * attached to this cgroup before it is removed, or that some other
2947 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
2948 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
2949 * unused, and this cgroup will be reprieved from its death sentence,
2950 * to continue to serve a useful existence. Next time it's released,
2951 * we will get notified again, if it still has 'notify_on_release' set.
2953 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
2954 * means only wait until the task is successfully execve()'d. The
2955 * separate release agent task is forked by call_usermodehelper(),
2956 * then control in this thread returns here, without waiting for the
2957 * release agent task. We don't bother to wait because the caller of
2958 * this routine has no use for the exit status of the release agent
2959 * task, so no sense holding our caller up for that.
2961 static void cgroup_release_agent(struct work_struct *work)
2963 BUG_ON(work != &release_agent_work);
2964 mutex_lock(&cgroup_mutex);
2965 spin_lock(&release_list_lock);
2966 while (!list_empty(&release_list)) {
2967 char *argv[3], *envp[3];
2970 struct cgroup *cgrp = list_entry(release_list.next,
2973 list_del_init(&cgrp->release_list);
2974 spin_unlock(&release_list_lock);
2975 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2977 spin_lock(&release_list_lock);
2981 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) {
2983 spin_lock(&release_list_lock);
2988 argv[i++] = cgrp->root->release_agent_path;
2989 argv[i++] = (char *)pathbuf;
2993 /* minimal command environment */
2994 envp[i++] = "HOME=/";
2995 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
2998 /* Drop the lock while we invoke the usermode helper,
2999 * since the exec could involve hitting disk and hence
3000 * be a slow process */
3001 mutex_unlock(&cgroup_mutex);
3002 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
3004 mutex_lock(&cgroup_mutex);
3005 spin_lock(&release_list_lock);
3007 spin_unlock(&release_list_lock);
3008 mutex_unlock(&cgroup_mutex);