4 * Generic process-grouping system.
6 * Based originally on the cpuset system, extracted by Paul Menage
7 * Copyright (C) 2006 Google, Inc
9 * Copyright notices from the original cpuset code:
10 * --------------------------------------------------
11 * Copyright (C) 2003 BULL SA.
12 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
14 * Portions derived from Patrick Mochel's sysfs code.
15 * sysfs is Copyright (c) 2001-3 Patrick Mochel
17 * 2003-10-10 Written by Simon Derr.
18 * 2003-10-22 Updates by Stephen Hemminger.
19 * 2004 May-July Rework by Paul Jackson.
20 * ---------------------------------------------------
22 * This file is subject to the terms and conditions of the GNU General Public
23 * License. See the file COPYING in the main directory of the Linux
24 * distribution for more details.
27 #include <linux/cgroup.h>
28 #include <linux/errno.h>
30 #include <linux/kernel.h>
31 #include <linux/list.h>
33 #include <linux/mutex.h>
34 #include <linux/mount.h>
35 #include <linux/pagemap.h>
36 #include <linux/rcupdate.h>
37 #include <linux/sched.h>
38 #include <linux/seq_file.h>
39 #include <linux/slab.h>
40 #include <linux/magic.h>
41 #include <linux/spinlock.h>
42 #include <linux/string.h>
43 #include <linux/sort.h>
44 #include <asm/atomic.h>
46 /* Generate an array of cgroup subsystem pointers */
47 #define SUBSYS(_x) &_x ## _subsys,
49 static struct cgroup_subsys *subsys[] = {
50 #include <linux/cgroup_subsys.h>
54 * A cgroupfs_root represents the root of a cgroup hierarchy,
55 * and may be associated with a superblock to form an active
58 struct cgroupfs_root {
59 struct super_block *sb;
62 * The bitmask of subsystems intended to be attached to this
65 unsigned long subsys_bits;
67 /* The bitmask of subsystems currently attached to this hierarchy */
68 unsigned long actual_subsys_bits;
70 /* A list running through the attached subsystems */
71 struct list_head subsys_list;
73 /* The root cgroup for this hierarchy */
74 struct cgroup top_cgroup;
76 /* Tracks how many cgroups are currently defined in hierarchy.*/
77 int number_of_cgroups;
79 /* A list running through the mounted hierarchies */
80 struct list_head root_list;
82 /* Hierarchy-specific flags */
88 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
89 * subsystems that are otherwise unattached - it never has more than a
90 * single cgroup, and all tasks are part of that cgroup.
92 static struct cgroupfs_root rootnode;
94 /* The list of hierarchy roots */
96 static LIST_HEAD(roots);
98 /* dummytop is a shorthand for the dummy hierarchy's top cgroup */
99 #define dummytop (&rootnode.top_cgroup)
101 /* This flag indicates whether tasks in the fork and exit paths should
102 * take callback_mutex and check for fork/exit handlers to call. This
103 * avoids us having to do extra work in the fork/exit path if none of the
104 * subsystems need to be called.
106 static int need_forkexit_callback;
108 /* bits in struct cgroup flags field */
113 /* convenient tests for these bits */
114 inline int cgroup_is_removed(const struct cgroup *cont)
116 return test_bit(CONT_REMOVED, &cont->flags);
119 /* bits in struct cgroupfs_root flags field */
121 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
125 * for_each_subsys() allows you to iterate on each subsystem attached to
126 * an active hierarchy
128 #define for_each_subsys(_root, _ss) \
129 list_for_each_entry(_ss, &_root->subsys_list, sibling)
131 /* for_each_root() allows you to iterate across the active hierarchies */
132 #define for_each_root(_root) \
133 list_for_each_entry(_root, &roots, root_list)
136 * There is one global cgroup mutex. We also require taking
137 * task_lock() when dereferencing a task's cgroup subsys pointers.
138 * See "The task_lock() exception", at the end of this comment.
140 * A task must hold cgroup_mutex to modify cgroups.
142 * Any task can increment and decrement the count field without lock.
143 * So in general, code holding cgroup_mutex can't rely on the count
144 * field not changing. However, if the count goes to zero, then only
145 * attach_task() can increment it again. Because a count of zero
146 * means that no tasks are currently attached, therefore there is no
147 * way a task attached to that cgroup can fork (the other way to
148 * increment the count). So code holding cgroup_mutex can safely
149 * assume that if the count is zero, it will stay zero. Similarly, if
150 * a task holds cgroup_mutex on a cgroup with zero count, it
151 * knows that the cgroup won't be removed, as cgroup_rmdir()
154 * The cgroup_common_file_write handler for operations that modify
155 * the cgroup hierarchy holds cgroup_mutex across the entire operation,
156 * single threading all such cgroup modifications across the system.
158 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
159 * (usually) take cgroup_mutex. These are the two most performance
160 * critical pieces of code here. The exception occurs on cgroup_exit(),
161 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
162 * is taken, and if the cgroup count is zero, a usermode call made
163 * to /sbin/cgroup_release_agent with the name of the cgroup (path
164 * relative to the root of cgroup file system) as the argument.
166 * A cgroup can only be deleted if both its 'count' of using tasks
167 * is zero, and its list of 'children' cgroups is empty. Since all
168 * tasks in the system use _some_ cgroup, and since there is always at
169 * least one task in the system (init, pid == 1), therefore, top_cgroup
170 * always has either children cgroups and/or using tasks. So we don't
171 * need a special hack to ensure that top_cgroup cannot be deleted.
173 * The task_lock() exception
175 * The need for this exception arises from the action of
176 * attach_task(), which overwrites one tasks cgroup pointer with
177 * another. It does so using cgroup_mutexe, however there are
178 * several performance critical places that need to reference
179 * task->cgroup without the expense of grabbing a system global
180 * mutex. Therefore except as noted below, when dereferencing or, as
181 * in attach_task(), modifying a task'ss cgroup pointer we use
182 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
183 * the task_struct routinely used for such matters.
185 * P.S. One more locking exception. RCU is used to guard the
186 * update of a tasks cgroup pointer by attach_task()
189 static DEFINE_MUTEX(cgroup_mutex);
192 * cgroup_lock - lock out any changes to cgroup structures
196 void cgroup_lock(void)
198 mutex_lock(&cgroup_mutex);
202 * cgroup_unlock - release lock on cgroup changes
204 * Undo the lock taken in a previous cgroup_lock() call.
207 void cgroup_unlock(void)
209 mutex_unlock(&cgroup_mutex);
213 * A couple of forward declarations required, due to cyclic reference loop:
214 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
215 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
219 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
220 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
221 static int cgroup_populate_dir(struct cgroup *cont);
222 static struct inode_operations cgroup_dir_inode_operations;
224 static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
226 struct inode *inode = new_inode(sb);
227 static struct backing_dev_info cgroup_backing_dev_info = {
228 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
232 inode->i_mode = mode;
233 inode->i_uid = current->fsuid;
234 inode->i_gid = current->fsgid;
236 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
237 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
242 static void cgroup_diput(struct dentry *dentry, struct inode *inode)
244 /* is dentry a directory ? if so, kfree() associated cgroup */
245 if (S_ISDIR(inode->i_mode)) {
246 struct cgroup *cont = dentry->d_fsdata;
247 BUG_ON(!(cgroup_is_removed(cont)));
253 static void remove_dir(struct dentry *d)
255 struct dentry *parent = dget(d->d_parent);
258 simple_rmdir(parent->d_inode, d);
262 static void cgroup_clear_directory(struct dentry *dentry)
264 struct list_head *node;
266 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
267 spin_lock(&dcache_lock);
268 node = dentry->d_subdirs.next;
269 while (node != &dentry->d_subdirs) {
270 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
273 /* This should never be called on a cgroup
274 * directory with child cgroups */
275 BUG_ON(d->d_inode->i_mode & S_IFDIR);
277 spin_unlock(&dcache_lock);
279 simple_unlink(dentry->d_inode, d);
281 spin_lock(&dcache_lock);
283 node = dentry->d_subdirs.next;
285 spin_unlock(&dcache_lock);
289 * NOTE : the dentry must have been dget()'ed
291 static void cgroup_d_remove_dir(struct dentry *dentry)
293 cgroup_clear_directory(dentry);
295 spin_lock(&dcache_lock);
296 list_del_init(&dentry->d_u.d_child);
297 spin_unlock(&dcache_lock);
301 static int rebind_subsystems(struct cgroupfs_root *root,
302 unsigned long final_bits)
304 unsigned long added_bits, removed_bits;
305 struct cgroup *cont = &root->top_cgroup;
308 removed_bits = root->actual_subsys_bits & ~final_bits;
309 added_bits = final_bits & ~root->actual_subsys_bits;
310 /* Check that any added subsystems are currently free */
311 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
312 unsigned long long bit = 1ull << i;
313 struct cgroup_subsys *ss = subsys[i];
314 if (!(bit & added_bits))
316 if (ss->root != &rootnode) {
317 /* Subsystem isn't free */
322 /* Currently we don't handle adding/removing subsystems when
323 * any child cgroups exist. This is theoretically supportable
324 * but involves complex error handling, so it's being left until
326 if (!list_empty(&cont->children))
329 /* Process each subsystem */
330 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
331 struct cgroup_subsys *ss = subsys[i];
332 unsigned long bit = 1UL << i;
333 if (bit & added_bits) {
334 /* We're binding this subsystem to this hierarchy */
335 BUG_ON(cont->subsys[i]);
336 BUG_ON(!dummytop->subsys[i]);
337 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
338 cont->subsys[i] = dummytop->subsys[i];
339 cont->subsys[i]->cgroup = cont;
340 list_add(&ss->sibling, &root->subsys_list);
341 rcu_assign_pointer(ss->root, root);
345 } else if (bit & removed_bits) {
346 /* We're removing this subsystem */
347 BUG_ON(cont->subsys[i] != dummytop->subsys[i]);
348 BUG_ON(cont->subsys[i]->cgroup != cont);
350 ss->bind(ss, dummytop);
351 dummytop->subsys[i]->cgroup = dummytop;
352 cont->subsys[i] = NULL;
353 rcu_assign_pointer(subsys[i]->root, &rootnode);
354 list_del(&ss->sibling);
355 } else if (bit & final_bits) {
356 /* Subsystem state should already exist */
357 BUG_ON(!cont->subsys[i]);
359 /* Subsystem state shouldn't exist */
360 BUG_ON(cont->subsys[i]);
363 root->subsys_bits = root->actual_subsys_bits = final_bits;
369 static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
371 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
372 struct cgroup_subsys *ss;
374 mutex_lock(&cgroup_mutex);
375 for_each_subsys(root, ss)
376 seq_printf(seq, ",%s", ss->name);
377 if (test_bit(ROOT_NOPREFIX, &root->flags))
378 seq_puts(seq, ",noprefix");
379 mutex_unlock(&cgroup_mutex);
383 struct cgroup_sb_opts {
384 unsigned long subsys_bits;
388 /* Convert a hierarchy specifier into a bitmask of subsystems and
390 static int parse_cgroupfs_options(char *data,
391 struct cgroup_sb_opts *opts)
393 char *token, *o = data ?: "all";
395 opts->subsys_bits = 0;
398 while ((token = strsep(&o, ",")) != NULL) {
401 if (!strcmp(token, "all")) {
402 opts->subsys_bits = (1 << CGROUP_SUBSYS_COUNT) - 1;
403 } else if (!strcmp(token, "noprefix")) {
404 set_bit(ROOT_NOPREFIX, &opts->flags);
406 struct cgroup_subsys *ss;
408 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
410 if (!strcmp(token, ss->name)) {
411 set_bit(i, &opts->subsys_bits);
415 if (i == CGROUP_SUBSYS_COUNT)
420 /* We can't have an empty hierarchy */
421 if (!opts->subsys_bits)
427 static int cgroup_remount(struct super_block *sb, int *flags, char *data)
430 struct cgroupfs_root *root = sb->s_fs_info;
431 struct cgroup *cont = &root->top_cgroup;
432 struct cgroup_sb_opts opts;
434 mutex_lock(&cont->dentry->d_inode->i_mutex);
435 mutex_lock(&cgroup_mutex);
437 /* See what subsystems are wanted */
438 ret = parse_cgroupfs_options(data, &opts);
442 /* Don't allow flags to change at remount */
443 if (opts.flags != root->flags) {
448 ret = rebind_subsystems(root, opts.subsys_bits);
450 /* (re)populate subsystem files */
452 cgroup_populate_dir(cont);
455 mutex_unlock(&cgroup_mutex);
456 mutex_unlock(&cont->dentry->d_inode->i_mutex);
460 static struct super_operations cgroup_ops = {
461 .statfs = simple_statfs,
462 .drop_inode = generic_delete_inode,
463 .show_options = cgroup_show_options,
464 .remount_fs = cgroup_remount,
467 static void init_cgroup_root(struct cgroupfs_root *root)
469 struct cgroup *cont = &root->top_cgroup;
470 INIT_LIST_HEAD(&root->subsys_list);
471 INIT_LIST_HEAD(&root->root_list);
472 root->number_of_cgroups = 1;
474 cont->top_cgroup = cont;
475 INIT_LIST_HEAD(&cont->sibling);
476 INIT_LIST_HEAD(&cont->children);
479 static int cgroup_test_super(struct super_block *sb, void *data)
481 struct cgroupfs_root *new = data;
482 struct cgroupfs_root *root = sb->s_fs_info;
484 /* First check subsystems */
485 if (new->subsys_bits != root->subsys_bits)
488 /* Next check flags */
489 if (new->flags != root->flags)
495 static int cgroup_set_super(struct super_block *sb, void *data)
498 struct cgroupfs_root *root = data;
500 ret = set_anon_super(sb, NULL);
504 sb->s_fs_info = root;
507 sb->s_blocksize = PAGE_CACHE_SIZE;
508 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
509 sb->s_magic = CGROUP_SUPER_MAGIC;
510 sb->s_op = &cgroup_ops;
515 static int cgroup_get_rootdir(struct super_block *sb)
517 struct inode *inode =
518 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
519 struct dentry *dentry;
524 inode->i_op = &simple_dir_inode_operations;
525 inode->i_fop = &simple_dir_operations;
526 inode->i_op = &cgroup_dir_inode_operations;
527 /* directories start off with i_nlink == 2 (for "." entry) */
529 dentry = d_alloc_root(inode);
538 static int cgroup_get_sb(struct file_system_type *fs_type,
539 int flags, const char *unused_dev_name,
540 void *data, struct vfsmount *mnt)
542 struct cgroup_sb_opts opts;
544 struct super_block *sb;
545 struct cgroupfs_root *root;
547 /* First find the desired set of subsystems */
548 ret = parse_cgroupfs_options(data, &opts);
552 root = kzalloc(sizeof(*root), GFP_KERNEL);
556 init_cgroup_root(root);
557 root->subsys_bits = opts.subsys_bits;
558 root->flags = opts.flags;
560 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, root);
567 if (sb->s_fs_info != root) {
568 /* Reusing an existing superblock */
569 BUG_ON(sb->s_root == NULL);
574 struct cgroup *cont = &root->top_cgroup;
576 BUG_ON(sb->s_root != NULL);
578 ret = cgroup_get_rootdir(sb);
582 mutex_lock(&cgroup_mutex);
584 ret = rebind_subsystems(root, root->subsys_bits);
586 mutex_unlock(&cgroup_mutex);
590 /* EBUSY should be the only error here */
593 list_add(&root->root_list, &roots);
595 sb->s_root->d_fsdata = &root->top_cgroup;
596 root->top_cgroup.dentry = sb->s_root;
598 BUG_ON(!list_empty(&cont->sibling));
599 BUG_ON(!list_empty(&cont->children));
600 BUG_ON(root->number_of_cgroups != 1);
603 * I believe that it's safe to nest i_mutex inside
604 * cgroup_mutex in this case, since no-one else can
605 * be accessing this directory yet. But we still need
606 * to teach lockdep that this is the case - currently
607 * a cgroupfs remount triggers a lockdep warning
609 mutex_lock(&cont->dentry->d_inode->i_mutex);
610 cgroup_populate_dir(cont);
611 mutex_unlock(&cont->dentry->d_inode->i_mutex);
612 mutex_unlock(&cgroup_mutex);
615 return simple_set_mnt(mnt, sb);
618 up_write(&sb->s_umount);
619 deactivate_super(sb);
623 static void cgroup_kill_sb(struct super_block *sb) {
624 struct cgroupfs_root *root = sb->s_fs_info;
625 struct cgroup *cont = &root->top_cgroup;
630 BUG_ON(root->number_of_cgroups != 1);
631 BUG_ON(!list_empty(&cont->children));
632 BUG_ON(!list_empty(&cont->sibling));
634 mutex_lock(&cgroup_mutex);
636 /* Rebind all subsystems back to the default hierarchy */
637 ret = rebind_subsystems(root, 0);
638 /* Shouldn't be able to fail ... */
641 if (!list_empty(&root->root_list))
642 list_del(&root->root_list);
643 mutex_unlock(&cgroup_mutex);
646 kill_litter_super(sb);
649 static struct file_system_type cgroup_fs_type = {
651 .get_sb = cgroup_get_sb,
652 .kill_sb = cgroup_kill_sb,
655 static inline struct cgroup *__d_cont(struct dentry *dentry)
657 return dentry->d_fsdata;
660 static inline struct cftype *__d_cft(struct dentry *dentry)
662 return dentry->d_fsdata;
666 * Called with cgroup_mutex held. Writes path of cgroup into buf.
667 * Returns 0 on success, -errno on error.
669 int cgroup_path(const struct cgroup *cont, char *buf, int buflen)
673 if (cont == dummytop) {
675 * Inactive subsystems have no dentry for their root
682 start = buf + buflen;
686 int len = cont->dentry->d_name.len;
687 if ((start -= len) < buf)
688 return -ENAMETOOLONG;
689 memcpy(start, cont->dentry->d_name.name, len);
696 return -ENAMETOOLONG;
699 memmove(buf, start, buf + buflen - start);
704 * Return the first subsystem attached to a cgroup's hierarchy, and
708 static void get_first_subsys(const struct cgroup *cont,
709 struct cgroup_subsys_state **css, int *subsys_id)
711 const struct cgroupfs_root *root = cont->root;
712 const struct cgroup_subsys *test_ss;
713 BUG_ON(list_empty(&root->subsys_list));
714 test_ss = list_entry(root->subsys_list.next,
715 struct cgroup_subsys, sibling);
717 *css = cont->subsys[test_ss->subsys_id];
721 *subsys_id = test_ss->subsys_id;
725 * Attach task 'tsk' to cgroup 'cont'
727 * Call holding cgroup_mutex. May take task_lock of
728 * the task 'pid' during call.
730 static int attach_task(struct cgroup *cont, struct task_struct *tsk)
733 struct cgroup_subsys *ss;
734 struct cgroup *oldcont;
735 struct css_set *cg = &tsk->cgroups;
736 struct cgroupfs_root *root = cont->root;
740 get_first_subsys(cont, NULL, &subsys_id);
742 /* Nothing to do if the task is already in that cgroup */
743 oldcont = task_cgroup(tsk, subsys_id);
747 for_each_subsys(root, ss) {
748 if (ss->can_attach) {
749 retval = ss->can_attach(ss, cont, tsk);
757 if (tsk->flags & PF_EXITING) {
761 /* Update the css_set pointers for the subsystems in this
763 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
764 if (root->subsys_bits & (1ull << i)) {
765 /* Subsystem is in this hierarchy. So we want
766 * the subsystem state from the new
767 * cgroup. Transfer the refcount from the
769 atomic_inc(&cont->count);
770 atomic_dec(&cg->subsys[i]->cgroup->count);
771 rcu_assign_pointer(cg->subsys[i], cont->subsys[i]);
776 for_each_subsys(root, ss) {
778 ss->attach(ss, cont, oldcont, tsk);
787 * Attach task with pid 'pid' to cgroup 'cont'. Call with
788 * cgroup_mutex, may take task_lock of task
790 static int attach_task_by_pid(struct cgroup *cont, char *pidbuf)
793 struct task_struct *tsk;
796 if (sscanf(pidbuf, "%d", &pid) != 1)
801 tsk = find_task_by_pid(pid);
802 if (!tsk || tsk->flags & PF_EXITING) {
806 get_task_struct(tsk);
809 if ((current->euid) && (current->euid != tsk->uid)
810 && (current->euid != tsk->suid)) {
811 put_task_struct(tsk);
816 get_task_struct(tsk);
819 ret = attach_task(cont, tsk);
820 put_task_struct(tsk);
824 /* The various types of files and directories in a cgroup file system */
826 enum cgroup_filetype {
832 static ssize_t cgroup_common_file_write(struct cgroup *cont,
835 const char __user *userbuf,
836 size_t nbytes, loff_t *unused_ppos)
838 enum cgroup_filetype type = cft->private;
842 if (nbytes >= PATH_MAX)
845 /* +1 for nul-terminator */
846 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
850 if (copy_from_user(buffer, userbuf, nbytes)) {
854 buffer[nbytes] = 0; /* nul-terminate */
856 mutex_lock(&cgroup_mutex);
858 if (cgroup_is_removed(cont)) {
865 retval = attach_task_by_pid(cont, buffer);
875 mutex_unlock(&cgroup_mutex);
881 static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
882 size_t nbytes, loff_t *ppos)
884 struct cftype *cft = __d_cft(file->f_dentry);
885 struct cgroup *cont = __d_cont(file->f_dentry->d_parent);
892 return cft->write(cont, cft, file, buf, nbytes, ppos);
895 static ssize_t cgroup_read_uint(struct cgroup *cont, struct cftype *cft,
897 char __user *buf, size_t nbytes,
901 u64 val = cft->read_uint(cont, cft);
902 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
904 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
907 static ssize_t cgroup_file_read(struct file *file, char __user *buf,
908 size_t nbytes, loff_t *ppos)
910 struct cftype *cft = __d_cft(file->f_dentry);
911 struct cgroup *cont = __d_cont(file->f_dentry->d_parent);
917 return cft->read(cont, cft, file, buf, nbytes, ppos);
919 return cgroup_read_uint(cont, cft, file, buf, nbytes, ppos);
923 static int cgroup_file_open(struct inode *inode, struct file *file)
928 err = generic_file_open(inode, file);
932 cft = __d_cft(file->f_dentry);
936 err = cft->open(inode, file);
943 static int cgroup_file_release(struct inode *inode, struct file *file)
945 struct cftype *cft = __d_cft(file->f_dentry);
947 return cft->release(inode, file);
952 * cgroup_rename - Only allow simple rename of directories in place.
954 static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
955 struct inode *new_dir, struct dentry *new_dentry)
957 if (!S_ISDIR(old_dentry->d_inode->i_mode))
959 if (new_dentry->d_inode)
961 if (old_dir != new_dir)
963 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
966 static struct file_operations cgroup_file_operations = {
967 .read = cgroup_file_read,
968 .write = cgroup_file_write,
969 .llseek = generic_file_llseek,
970 .open = cgroup_file_open,
971 .release = cgroup_file_release,
974 static struct inode_operations cgroup_dir_inode_operations = {
975 .lookup = simple_lookup,
976 .mkdir = cgroup_mkdir,
977 .rmdir = cgroup_rmdir,
978 .rename = cgroup_rename,
981 static int cgroup_create_file(struct dentry *dentry, int mode,
982 struct super_block *sb)
984 static struct dentry_operations cgroup_dops = {
985 .d_iput = cgroup_diput,
995 inode = cgroup_new_inode(mode, sb);
1000 inode->i_op = &cgroup_dir_inode_operations;
1001 inode->i_fop = &simple_dir_operations;
1003 /* start off with i_nlink == 2 (for "." entry) */
1006 /* start with the directory inode held, so that we can
1007 * populate it without racing with another mkdir */
1008 mutex_lock(&inode->i_mutex);
1009 } else if (S_ISREG(mode)) {
1011 inode->i_fop = &cgroup_file_operations;
1013 dentry->d_op = &cgroup_dops;
1014 d_instantiate(dentry, inode);
1015 dget(dentry); /* Extra count - pin the dentry in core */
1020 * cgroup_create_dir - create a directory for an object.
1021 * cont: the cgroup we create the directory for.
1022 * It must have a valid ->parent field
1023 * And we are going to fill its ->dentry field.
1024 * dentry: dentry of the new container
1025 * mode: mode to set on new directory.
1027 static int cgroup_create_dir(struct cgroup *cont, struct dentry *dentry,
1030 struct dentry *parent;
1033 parent = cont->parent->dentry;
1034 error = cgroup_create_file(dentry, S_IFDIR | mode, cont->root->sb);
1036 dentry->d_fsdata = cont;
1037 inc_nlink(parent->d_inode);
1038 cont->dentry = dentry;
1046 int cgroup_add_file(struct cgroup *cont,
1047 struct cgroup_subsys *subsys,
1048 const struct cftype *cft)
1050 struct dentry *dir = cont->dentry;
1051 struct dentry *dentry;
1054 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
1055 if (subsys && !test_bit(ROOT_NOPREFIX, &cont->root->flags)) {
1056 strcpy(name, subsys->name);
1059 strcat(name, cft->name);
1060 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
1061 dentry = lookup_one_len(name, dir, strlen(name));
1062 if (!IS_ERR(dentry)) {
1063 error = cgroup_create_file(dentry, 0644 | S_IFREG,
1066 dentry->d_fsdata = (void *)cft;
1069 error = PTR_ERR(dentry);
1073 int cgroup_add_files(struct cgroup *cont,
1074 struct cgroup_subsys *subsys,
1075 const struct cftype cft[],
1079 for (i = 0; i < count; i++) {
1080 err = cgroup_add_file(cont, subsys, &cft[i]);
1087 /* Count the number of tasks in a cgroup. Could be made more
1088 * time-efficient but less space-efficient with more linked lists
1089 * running through each cgroup and the css_set structures that
1090 * referenced it. Must be called with tasklist_lock held for read or
1091 * write or in an rcu critical section.
1093 int __cgroup_task_count(const struct cgroup *cont)
1096 struct task_struct *g, *p;
1097 struct cgroup_subsys_state *css;
1100 get_first_subsys(cont, &css, &subsys_id);
1101 do_each_thread(g, p) {
1102 if (task_subsys_state(p, subsys_id) == css)
1104 } while_each_thread(g, p);
1109 * Stuff for reading the 'tasks' file.
1111 * Reading this file can return large amounts of data if a cgroup has
1112 * *lots* of attached tasks. So it may need several calls to read(),
1113 * but we cannot guarantee that the information we produce is correct
1114 * unless we produce it entirely atomically.
1116 * Upon tasks file open(), a struct ctr_struct is allocated, that
1117 * will have a pointer to an array (also allocated here). The struct
1118 * ctr_struct * is stored in file->private_data. Its resources will
1119 * be freed by release() when the file is closed. The array is used
1120 * to sprintf the PIDs and then used by read().
1128 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
1129 * 'cont'. Return actual number of pids loaded. No need to
1130 * task_lock(p) when reading out p->cgroup, since we're in an RCU
1131 * read section, so the css_set can't go away, and is
1132 * immutable after creation.
1134 static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cont)
1137 struct task_struct *g, *p;
1138 struct cgroup_subsys_state *css;
1141 get_first_subsys(cont, &css, &subsys_id);
1143 do_each_thread(g, p) {
1144 if (task_subsys_state(p, subsys_id) == css) {
1145 pidarray[n++] = pid_nr(task_pid(p));
1146 if (unlikely(n == npids))
1149 } while_each_thread(g, p);
1156 static int cmppid(const void *a, const void *b)
1158 return *(pid_t *)a - *(pid_t *)b;
1162 * Convert array 'a' of 'npids' pid_t's to a string of newline separated
1163 * decimal pids in 'buf'. Don't write more than 'sz' chars, but return
1164 * count 'cnt' of how many chars would be written if buf were large enough.
1166 static int pid_array_to_buf(char *buf, int sz, pid_t *a, int npids)
1171 for (i = 0; i < npids; i++)
1172 cnt += snprintf(buf + cnt, max(sz - cnt, 0), "%d\n", a[i]);
1177 * Handle an open on 'tasks' file. Prepare a buffer listing the
1178 * process id's of tasks currently attached to the cgroup being opened.
1180 * Does not require any specific cgroup mutexes, and does not take any.
1182 static int cgroup_tasks_open(struct inode *unused, struct file *file)
1184 struct cgroup *cont = __d_cont(file->f_dentry->d_parent);
1185 struct ctr_struct *ctr;
1190 if (!(file->f_mode & FMODE_READ))
1193 ctr = kmalloc(sizeof(*ctr), GFP_KERNEL);
1198 * If cgroup gets more users after we read count, we won't have
1199 * enough space - tough. This race is indistinguishable to the
1200 * caller from the case that the additional cgroup users didn't
1201 * show up until sometime later on.
1203 npids = cgroup_task_count(cont);
1205 pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL);
1209 npids = pid_array_load(pidarray, npids, cont);
1210 sort(pidarray, npids, sizeof(pid_t), cmppid, NULL);
1212 /* Call pid_array_to_buf() twice, first just to get bufsz */
1213 ctr->bufsz = pid_array_to_buf(&c, sizeof(c), pidarray, npids) + 1;
1214 ctr->buf = kmalloc(ctr->bufsz, GFP_KERNEL);
1217 ctr->bufsz = pid_array_to_buf(ctr->buf, ctr->bufsz, pidarray, npids);
1224 file->private_data = ctr;
1235 static ssize_t cgroup_tasks_read(struct cgroup *cont,
1237 struct file *file, char __user *buf,
1238 size_t nbytes, loff_t *ppos)
1240 struct ctr_struct *ctr = file->private_data;
1242 return simple_read_from_buffer(buf, nbytes, ppos, ctr->buf, ctr->bufsz);
1245 static int cgroup_tasks_release(struct inode *unused_inode,
1248 struct ctr_struct *ctr;
1250 if (file->f_mode & FMODE_READ) {
1251 ctr = file->private_data;
1259 * for the common functions, 'private' gives the type of file
1261 static struct cftype cft_tasks = {
1263 .open = cgroup_tasks_open,
1264 .read = cgroup_tasks_read,
1265 .write = cgroup_common_file_write,
1266 .release = cgroup_tasks_release,
1267 .private = FILE_TASKLIST,
1270 static int cgroup_populate_dir(struct cgroup *cont)
1273 struct cgroup_subsys *ss;
1275 /* First clear out any existing files */
1276 cgroup_clear_directory(cont->dentry);
1278 err = cgroup_add_file(cont, NULL, &cft_tasks);
1282 for_each_subsys(cont->root, ss) {
1283 if (ss->populate && (err = ss->populate(ss, cont)) < 0)
1290 static void init_cgroup_css(struct cgroup_subsys_state *css,
1291 struct cgroup_subsys *ss,
1292 struct cgroup *cont)
1295 atomic_set(&css->refcnt, 0);
1297 if (cont == dummytop)
1298 set_bit(CSS_ROOT, &css->flags);
1299 BUG_ON(cont->subsys[ss->subsys_id]);
1300 cont->subsys[ss->subsys_id] = css;
1304 * cgroup_create - create a cgroup
1305 * parent: cgroup that will be parent of the new cgroup.
1306 * name: name of the new cgroup. Will be strcpy'ed.
1307 * mode: mode to set on new inode
1309 * Must be called with the mutex on the parent inode held
1312 static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
1315 struct cgroup *cont;
1316 struct cgroupfs_root *root = parent->root;
1318 struct cgroup_subsys *ss;
1319 struct super_block *sb = root->sb;
1321 cont = kzalloc(sizeof(*cont), GFP_KERNEL);
1325 /* Grab a reference on the superblock so the hierarchy doesn't
1326 * get deleted on unmount if there are child cgroups. This
1327 * can be done outside cgroup_mutex, since the sb can't
1328 * disappear while someone has an open control file on the
1330 atomic_inc(&sb->s_active);
1332 mutex_lock(&cgroup_mutex);
1335 INIT_LIST_HEAD(&cont->sibling);
1336 INIT_LIST_HEAD(&cont->children);
1338 cont->parent = parent;
1339 cont->root = parent->root;
1340 cont->top_cgroup = parent->top_cgroup;
1342 for_each_subsys(root, ss) {
1343 struct cgroup_subsys_state *css = ss->create(ss, cont);
1348 init_cgroup_css(css, ss, cont);
1351 list_add(&cont->sibling, &cont->parent->children);
1352 root->number_of_cgroups++;
1354 err = cgroup_create_dir(cont, dentry, mode);
1358 /* The cgroup directory was pre-locked for us */
1359 BUG_ON(!mutex_is_locked(&cont->dentry->d_inode->i_mutex));
1361 err = cgroup_populate_dir(cont);
1362 /* If err < 0, we have a half-filled directory - oh well ;) */
1364 mutex_unlock(&cgroup_mutex);
1365 mutex_unlock(&cont->dentry->d_inode->i_mutex);
1371 list_del(&cont->sibling);
1372 root->number_of_cgroups--;
1376 for_each_subsys(root, ss) {
1377 if (cont->subsys[ss->subsys_id])
1378 ss->destroy(ss, cont);
1381 mutex_unlock(&cgroup_mutex);
1383 /* Release the reference count that we took on the superblock */
1384 deactivate_super(sb);
1390 static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1392 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
1394 /* the vfs holds inode->i_mutex already */
1395 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
1398 static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
1400 struct cgroup *cont = dentry->d_fsdata;
1402 struct cgroup *parent;
1403 struct cgroup_subsys *ss;
1404 struct super_block *sb;
1405 struct cgroupfs_root *root;
1408 /* the vfs holds both inode->i_mutex already */
1410 mutex_lock(&cgroup_mutex);
1411 if (atomic_read(&cont->count) != 0) {
1412 mutex_unlock(&cgroup_mutex);
1415 if (!list_empty(&cont->children)) {
1416 mutex_unlock(&cgroup_mutex);
1420 parent = cont->parent;
1424 /* Check the reference count on each subsystem. Since we
1425 * already established that there are no tasks in the
1426 * cgroup, if the css refcount is also 0, then there should
1427 * be no outstanding references, so the subsystem is safe to
1429 for_each_subsys(root, ss) {
1430 struct cgroup_subsys_state *css;
1431 css = cont->subsys[ss->subsys_id];
1432 if (atomic_read(&css->refcnt)) {
1438 mutex_unlock(&cgroup_mutex);
1442 for_each_subsys(root, ss) {
1443 if (cont->subsys[ss->subsys_id])
1444 ss->destroy(ss, cont);
1447 set_bit(CONT_REMOVED, &cont->flags);
1448 /* delete my sibling from parent->children */
1449 list_del(&cont->sibling);
1450 spin_lock(&cont->dentry->d_lock);
1451 d = dget(cont->dentry);
1452 cont->dentry = NULL;
1453 spin_unlock(&d->d_lock);
1455 cgroup_d_remove_dir(d);
1457 root->number_of_cgroups--;
1459 mutex_unlock(&cgroup_mutex);
1460 /* Drop the active superblock reference that we took when we
1461 * created the cgroup */
1462 deactivate_super(sb);
1466 static void cgroup_init_subsys(struct cgroup_subsys *ss)
1468 struct task_struct *g, *p;
1469 struct cgroup_subsys_state *css;
1470 printk(KERN_ERR "Initializing cgroup subsys %s\n", ss->name);
1472 /* Create the top cgroup state for this subsystem */
1473 ss->root = &rootnode;
1474 css = ss->create(ss, dummytop);
1475 /* We don't handle early failures gracefully */
1476 BUG_ON(IS_ERR(css));
1477 init_cgroup_css(css, ss, dummytop);
1479 /* Update all tasks to contain a subsys pointer to this state
1480 * - since the subsystem is newly registered, all tasks are in
1481 * the subsystem's top cgroup. */
1483 /* If this subsystem requested that it be notified with fork
1484 * events, we should send it one now for every process in the
1487 read_lock(&tasklist_lock);
1488 init_task.cgroups.subsys[ss->subsys_id] = css;
1490 ss->fork(ss, &init_task);
1492 do_each_thread(g, p) {
1493 printk(KERN_INFO "Setting task %p css to %p (%d)\n", css, p, p->pid);
1494 p->cgroups.subsys[ss->subsys_id] = css;
1497 } while_each_thread(g, p);
1498 read_unlock(&tasklist_lock);
1500 need_forkexit_callback |= ss->fork || ss->exit;
1506 * cgroup_init_early - initialize cgroups at system boot, and
1507 * initialize any subsystems that request early init.
1509 int __init cgroup_init_early(void)
1512 init_cgroup_root(&rootnode);
1513 list_add(&rootnode.root_list, &roots);
1515 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1516 struct cgroup_subsys *ss = subsys[i];
1519 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
1520 BUG_ON(!ss->create);
1521 BUG_ON(!ss->destroy);
1522 if (ss->subsys_id != i) {
1523 printk(KERN_ERR "Subsys %s id == %d\n",
1524 ss->name, ss->subsys_id);
1529 cgroup_init_subsys(ss);
1535 * cgroup_init - register cgroup filesystem and /proc file, and
1536 * initialize any subsystems that didn't request early init.
1538 int __init cgroup_init(void)
1543 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1544 struct cgroup_subsys *ss = subsys[i];
1545 if (!ss->early_init)
1546 cgroup_init_subsys(ss);
1549 err = register_filesystem(&cgroup_fs_type);