aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/cgroup.c
diff options
context:
space:
mode:
Diffstat (limited to 'kernel/cgroup.c')
-rw-r--r--kernel/cgroup.c2805
1 files changed, 2805 insertions, 0 deletions
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
new file mode 100644
index 000000000000..5987dccdb2a0
--- /dev/null
+++ b/kernel/cgroup.c
@@ -0,0 +1,2805 @@
1/*
2 * kernel/cgroup.c
3 *
4 * Generic process-grouping system.
5 *
6 * Based originally on the cpuset system, extracted by Paul Menage
7 * Copyright (C) 2006 Google, Inc
8 *
9 * Copyright notices from the original cpuset code:
10 * --------------------------------------------------
11 * Copyright (C) 2003 BULL SA.
12 * Copyright (C) 2004-2006 Silicon Graphics, Inc.
13 *
14 * Portions derived from Patrick Mochel's sysfs code.
15 * sysfs is Copyright (c) 2001-3 Patrick Mochel
16 *
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 * ---------------------------------------------------
21 *
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.
25 */
26
27#include <linux/cgroup.h>
28#include <linux/errno.h>
29#include <linux/fs.h>
30#include <linux/kernel.h>
31#include <linux/list.h>
32#include <linux/mm.h>
33#include <linux/mutex.h>
34#include <linux/mount.h>
35#include <linux/pagemap.h>
36#include <linux/proc_fs.h>
37#include <linux/rcupdate.h>
38#include <linux/sched.h>
39#include <linux/backing-dev.h>
40#include <linux/seq_file.h>
41#include <linux/slab.h>
42#include <linux/magic.h>
43#include <linux/spinlock.h>
44#include <linux/string.h>
45#include <linux/sort.h>
46#include <linux/kmod.h>
47#include <linux/delayacct.h>
48#include <linux/cgroupstats.h>
49
50#include <asm/atomic.h>
51
52static DEFINE_MUTEX(cgroup_mutex);
53
54/* Generate an array of cgroup subsystem pointers */
55#define SUBSYS(_x) &_x ## _subsys,
56
57static struct cgroup_subsys *subsys[] = {
58#include <linux/cgroup_subsys.h>
59};
60
61/*
62 * A cgroupfs_root represents the root of a cgroup hierarchy,
63 * and may be associated with a superblock to form an active
64 * hierarchy
65 */
66struct cgroupfs_root {
67 struct super_block *sb;
68
69 /*
70 * The bitmask of subsystems intended to be attached to this
71 * hierarchy
72 */
73 unsigned long subsys_bits;
74
75 /* The bitmask of subsystems currently attached to this hierarchy */
76 unsigned long actual_subsys_bits;
77
78 /* A list running through the attached subsystems */
79 struct list_head subsys_list;
80
81 /* The root cgroup for this hierarchy */
82 struct cgroup top_cgroup;
83
84 /* Tracks how many cgroups are currently defined in hierarchy.*/
85 int number_of_cgroups;
86
87 /* A list running through the mounted hierarchies */
88 struct list_head root_list;
89
90 /* Hierarchy-specific flags */
91 unsigned long flags;
92
93 /* The path to use for release notifications. No locking
94 * between setting and use - so if userspace updates this
95 * while child cgroups exist, you could miss a
96 * notification. We ensure that it's always a valid
97 * NUL-terminated string */
98 char release_agent_path[PATH_MAX];
99};
100
101
102/*
103 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
104 * subsystems that are otherwise unattached - it never has more than a
105 * single cgroup, and all tasks are part of that cgroup.
106 */
107static struct cgroupfs_root rootnode;
108
109/* The list of hierarchy roots */
110
111static LIST_HEAD(roots);
112static int root_count;
113
114/* dummytop is a shorthand for the dummy hierarchy's top cgroup */
115#define dummytop (&rootnode.top_cgroup)
116
117/* This flag indicates whether tasks in the fork and exit paths should
118 * take callback_mutex and check for fork/exit handlers to call. This
119 * avoids us having to do extra work in the fork/exit path if none of the
120 * subsystems need to be called.
121 */
122static int need_forkexit_callback;
123
124/* bits in struct cgroup flags field */
125enum {
126 /* Control Group is dead */
127 CGRP_REMOVED,
128 /* Control Group has previously had a child cgroup or a task,
129 * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) */
130 CGRP_RELEASABLE,
131 /* Control Group requires release notifications to userspace */
132 CGRP_NOTIFY_ON_RELEASE,
133};
134
135/* convenient tests for these bits */
136inline int cgroup_is_removed(const struct cgroup *cgrp)
137{
138 return test_bit(CGRP_REMOVED, &cgrp->flags);
139}
140
141/* bits in struct cgroupfs_root flags field */
142enum {
143 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
144};
145
146inline int cgroup_is_releasable(const struct cgroup *cgrp)
147{
148 const int bits =
149 (1 << CGRP_RELEASABLE) |
150 (1 << CGRP_NOTIFY_ON_RELEASE);
151 return (cgrp->flags & bits) == bits;
152}
153
154inline int notify_on_release(const struct cgroup *cgrp)
155{
156 return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
157}
158
159/*
160 * for_each_subsys() allows you to iterate on each subsystem attached to
161 * an active hierarchy
162 */
163#define for_each_subsys(_root, _ss) \
164list_for_each_entry(_ss, &_root->subsys_list, sibling)
165
166/* for_each_root() allows you to iterate across the active hierarchies */
167#define for_each_root(_root) \
168list_for_each_entry(_root, &roots, root_list)
169
170/* the list of cgroups eligible for automatic release. Protected by
171 * release_list_lock */
172static LIST_HEAD(release_list);
173static DEFINE_SPINLOCK(release_list_lock);
174static void cgroup_release_agent(struct work_struct *work);
175static DECLARE_WORK(release_agent_work, cgroup_release_agent);
176static void check_for_release(struct cgroup *cgrp);
177
178/* Link structure for associating css_set objects with cgroups */
179struct cg_cgroup_link {
180 /*
181 * List running through cg_cgroup_links associated with a
182 * cgroup, anchored on cgroup->css_sets
183 */
184 struct list_head cgrp_link_list;
185 /*
186 * List running through cg_cgroup_links pointing at a
187 * single css_set object, anchored on css_set->cg_links
188 */
189 struct list_head cg_link_list;
190 struct css_set *cg;
191};
192
193/* The default css_set - used by init and its children prior to any
194 * hierarchies being mounted. It contains a pointer to the root state
195 * for each subsystem. Also used to anchor the list of css_sets. Not
196 * reference-counted, to improve performance when child cgroups
197 * haven't been created.
198 */
199
200static struct css_set init_css_set;
201static struct cg_cgroup_link init_css_set_link;
202
203/* css_set_lock protects the list of css_set objects, and the
204 * chain of tasks off each css_set. Nests outside task->alloc_lock
205 * due to cgroup_iter_start() */
206static DEFINE_RWLOCK(css_set_lock);
207static int css_set_count;
208
209/* We don't maintain the lists running through each css_set to its
210 * task until after the first call to cgroup_iter_start(). This
211 * reduces the fork()/exit() overhead for people who have cgroups
212 * compiled into their kernel but not actually in use */
213static int use_task_css_set_links;
214
215/* When we create or destroy a css_set, the operation simply
216 * takes/releases a reference count on all the cgroups referenced
217 * by subsystems in this css_set. This can end up multiple-counting
218 * some cgroups, but that's OK - the ref-count is just a
219 * busy/not-busy indicator; ensuring that we only count each cgroup
220 * once would require taking a global lock to ensure that no
221 * subsystems moved between hierarchies while we were doing so.
222 *
223 * Possible TODO: decide at boot time based on the number of
224 * registered subsystems and the number of CPUs or NUMA nodes whether
225 * it's better for performance to ref-count every subsystem, or to
226 * take a global lock and only add one ref count to each hierarchy.
227 */
228
229/*
230 * unlink a css_set from the list and free it
231 */
232static void unlink_css_set(struct css_set *cg)
233{
234 write_lock(&css_set_lock);
235 list_del(&cg->list);
236 css_set_count--;
237 while (!list_empty(&cg->cg_links)) {
238 struct cg_cgroup_link *link;
239 link = list_entry(cg->cg_links.next,
240 struct cg_cgroup_link, cg_link_list);
241 list_del(&link->cg_link_list);
242 list_del(&link->cgrp_link_list);
243 kfree(link);
244 }
245 write_unlock(&css_set_lock);
246}
247
248static void __release_css_set(struct kref *k, int taskexit)
249{
250 int i;
251 struct css_set *cg = container_of(k, struct css_set, ref);
252
253 unlink_css_set(cg);
254
255 rcu_read_lock();
256 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
257 struct cgroup *cgrp = cg->subsys[i]->cgroup;
258 if (atomic_dec_and_test(&cgrp->count) &&
259 notify_on_release(cgrp)) {
260 if (taskexit)
261 set_bit(CGRP_RELEASABLE, &cgrp->flags);
262 check_for_release(cgrp);
263 }
264 }
265 rcu_read_unlock();
266 kfree(cg);
267}
268
269static void release_css_set(struct kref *k)
270{
271 __release_css_set(k, 0);
272}
273
274static void release_css_set_taskexit(struct kref *k)
275{
276 __release_css_set(k, 1);
277}
278
279/*
280 * refcounted get/put for css_set objects
281 */
282static inline void get_css_set(struct css_set *cg)
283{
284 kref_get(&cg->ref);
285}
286
287static inline void put_css_set(struct css_set *cg)
288{
289 kref_put(&cg->ref, release_css_set);
290}
291
292static inline void put_css_set_taskexit(struct css_set *cg)
293{
294 kref_put(&cg->ref, release_css_set_taskexit);
295}
296
297/*
298 * find_existing_css_set() is a helper for
299 * find_css_set(), and checks to see whether an existing
300 * css_set is suitable. This currently walks a linked-list for
301 * simplicity; a later patch will use a hash table for better
302 * performance
303 *
304 * oldcg: the cgroup group that we're using before the cgroup
305 * transition
306 *
307 * cgrp: the cgroup that we're moving into
308 *
309 * template: location in which to build the desired set of subsystem
310 * state objects for the new cgroup group
311 */
312
313static struct css_set *find_existing_css_set(
314 struct css_set *oldcg,
315 struct cgroup *cgrp,
316 struct cgroup_subsys_state *template[])
317{
318 int i;
319 struct cgroupfs_root *root = cgrp->root;
320 struct list_head *l = &init_css_set.list;
321
322 /* Built the set of subsystem state objects that we want to
323 * see in the new css_set */
324 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
325 if (root->subsys_bits & (1ull << i)) {
326 /* Subsystem is in this hierarchy. So we want
327 * the subsystem state from the new
328 * cgroup */
329 template[i] = cgrp->subsys[i];
330 } else {
331 /* Subsystem is not in this hierarchy, so we
332 * don't want to change the subsystem state */
333 template[i] = oldcg->subsys[i];
334 }
335 }
336
337 /* Look through existing cgroup groups to find one to reuse */
338 do {
339 struct css_set *cg =
340 list_entry(l, struct css_set, list);
341
342 if (!memcmp(template, cg->subsys, sizeof(cg->subsys))) {
343 /* All subsystems matched */
344 return cg;
345 }
346 /* Try the next cgroup group */
347 l = l->next;
348 } while (l != &init_css_set.list);
349
350 /* No existing cgroup group matched */
351 return NULL;
352}
353
354/*
355 * allocate_cg_links() allocates "count" cg_cgroup_link structures
356 * and chains them on tmp through their cgrp_link_list fields. Returns 0 on
357 * success or a negative error
358 */
359
360static int allocate_cg_links(int count, struct list_head *tmp)
361{
362 struct cg_cgroup_link *link;
363 int i;
364 INIT_LIST_HEAD(tmp);
365 for (i = 0; i < count; i++) {
366 link = kmalloc(sizeof(*link), GFP_KERNEL);
367 if (!link) {
368 while (!list_empty(tmp)) {
369 link = list_entry(tmp->next,
370 struct cg_cgroup_link,
371 cgrp_link_list);
372 list_del(&link->cgrp_link_list);
373 kfree(link);
374 }
375 return -ENOMEM;
376 }
377 list_add(&link->cgrp_link_list, tmp);
378 }
379 return 0;
380}
381
382static void free_cg_links(struct list_head *tmp)
383{
384 while (!list_empty(tmp)) {
385 struct cg_cgroup_link *link;
386 link = list_entry(tmp->next,
387 struct cg_cgroup_link,
388 cgrp_link_list);
389 list_del(&link->cgrp_link_list);
390 kfree(link);
391 }
392}
393
394/*
395 * find_css_set() takes an existing cgroup group and a
396 * cgroup object, and returns a css_set object that's
397 * equivalent to the old group, but with the given cgroup
398 * substituted into the appropriate hierarchy. Must be called with
399 * cgroup_mutex held
400 */
401
402static struct css_set *find_css_set(
403 struct css_set *oldcg, struct cgroup *cgrp)
404{
405 struct css_set *res;
406 struct cgroup_subsys_state *template[CGROUP_SUBSYS_COUNT];
407 int i;
408
409 struct list_head tmp_cg_links;
410 struct cg_cgroup_link *link;
411
412 /* First see if we already have a cgroup group that matches
413 * the desired set */
414 write_lock(&css_set_lock);
415 res = find_existing_css_set(oldcg, cgrp, template);
416 if (res)
417 get_css_set(res);
418 write_unlock(&css_set_lock);
419
420 if (res)
421 return res;
422
423 res = kmalloc(sizeof(*res), GFP_KERNEL);
424 if (!res)
425 return NULL;
426
427 /* Allocate all the cg_cgroup_link objects that we'll need */
428 if (allocate_cg_links(root_count, &tmp_cg_links) < 0) {
429 kfree(res);
430 return NULL;
431 }
432
433 kref_init(&res->ref);
434 INIT_LIST_HEAD(&res->cg_links);
435 INIT_LIST_HEAD(&res->tasks);
436
437 /* Copy the set of subsystem state objects generated in
438 * find_existing_css_set() */
439 memcpy(res->subsys, template, sizeof(res->subsys));
440
441 write_lock(&css_set_lock);
442 /* Add reference counts and links from the new css_set. */
443 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
444 struct cgroup *cgrp = res->subsys[i]->cgroup;
445 struct cgroup_subsys *ss = subsys[i];
446 atomic_inc(&cgrp->count);
447 /*
448 * We want to add a link once per cgroup, so we
449 * only do it for the first subsystem in each
450 * hierarchy
451 */
452 if (ss->root->subsys_list.next == &ss->sibling) {
453 BUG_ON(list_empty(&tmp_cg_links));
454 link = list_entry(tmp_cg_links.next,
455 struct cg_cgroup_link,
456 cgrp_link_list);
457 list_del(&link->cgrp_link_list);
458 list_add(&link->cgrp_link_list, &cgrp->css_sets);
459 link->cg = res;
460 list_add(&link->cg_link_list, &res->cg_links);
461 }
462 }
463 if (list_empty(&rootnode.subsys_list)) {
464 link = list_entry(tmp_cg_links.next,
465 struct cg_cgroup_link,
466 cgrp_link_list);
467 list_del(&link->cgrp_link_list);
468 list_add(&link->cgrp_link_list, &dummytop->css_sets);
469 link->cg = res;
470 list_add(&link->cg_link_list, &res->cg_links);
471 }
472
473 BUG_ON(!list_empty(&tmp_cg_links));
474
475 /* Link this cgroup group into the list */
476 list_add(&res->list, &init_css_set.list);
477 css_set_count++;
478 INIT_LIST_HEAD(&res->tasks);
479 write_unlock(&css_set_lock);
480
481 return res;
482}
483
484/*
485 * There is one global cgroup mutex. We also require taking
486 * task_lock() when dereferencing a task's cgroup subsys pointers.
487 * See "The task_lock() exception", at the end of this comment.
488 *
489 * A task must hold cgroup_mutex to modify cgroups.
490 *
491 * Any task can increment and decrement the count field without lock.
492 * So in general, code holding cgroup_mutex can't rely on the count
493 * field not changing. However, if the count goes to zero, then only
494 * attach_task() can increment it again. Because a count of zero
495 * means that no tasks are currently attached, therefore there is no
496 * way a task attached to that cgroup can fork (the other way to
497 * increment the count). So code holding cgroup_mutex can safely
498 * assume that if the count is zero, it will stay zero. Similarly, if
499 * a task holds cgroup_mutex on a cgroup with zero count, it
500 * knows that the cgroup won't be removed, as cgroup_rmdir()
501 * needs that mutex.
502 *
503 * The cgroup_common_file_write handler for operations that modify
504 * the cgroup hierarchy holds cgroup_mutex across the entire operation,
505 * single threading all such cgroup modifications across the system.
506 *
507 * The fork and exit callbacks cgroup_fork() and cgroup_exit(), don't
508 * (usually) take cgroup_mutex. These are the two most performance
509 * critical pieces of code here. The exception occurs on cgroup_exit(),
510 * when a task in a notify_on_release cgroup exits. Then cgroup_mutex
511 * is taken, and if the cgroup count is zero, a usermode call made
512 * to /sbin/cgroup_release_agent with the name of the cgroup (path
513 * relative to the root of cgroup file system) as the argument.
514 *
515 * A cgroup can only be deleted if both its 'count' of using tasks
516 * is zero, and its list of 'children' cgroups is empty. Since all
517 * tasks in the system use _some_ cgroup, and since there is always at
518 * least one task in the system (init, pid == 1), therefore, top_cgroup
519 * always has either children cgroups and/or using tasks. So we don't
520 * need a special hack to ensure that top_cgroup cannot be deleted.
521 *
522 * The task_lock() exception
523 *
524 * The need for this exception arises from the action of
525 * attach_task(), which overwrites one tasks cgroup pointer with
526 * another. It does so using cgroup_mutexe, however there are
527 * several performance critical places that need to reference
528 * task->cgroup without the expense of grabbing a system global
529 * mutex. Therefore except as noted below, when dereferencing or, as
530 * in attach_task(), modifying a task'ss cgroup pointer we use
531 * task_lock(), which acts on a spinlock (task->alloc_lock) already in
532 * the task_struct routinely used for such matters.
533 *
534 * P.S. One more locking exception. RCU is used to guard the
535 * update of a tasks cgroup pointer by attach_task()
536 */
537
538/**
539 * cgroup_lock - lock out any changes to cgroup structures
540 *
541 */
542
543void cgroup_lock(void)
544{
545 mutex_lock(&cgroup_mutex);
546}
547
548/**
549 * cgroup_unlock - release lock on cgroup changes
550 *
551 * Undo the lock taken in a previous cgroup_lock() call.
552 */
553
554void cgroup_unlock(void)
555{
556 mutex_unlock(&cgroup_mutex);
557}
558
559/*
560 * A couple of forward declarations required, due to cyclic reference loop:
561 * cgroup_mkdir -> cgroup_create -> cgroup_populate_dir ->
562 * cgroup_add_file -> cgroup_create_file -> cgroup_dir_inode_operations
563 * -> cgroup_mkdir.
564 */
565
566static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode);
567static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
568static int cgroup_populate_dir(struct cgroup *cgrp);
569static struct inode_operations cgroup_dir_inode_operations;
570static struct file_operations proc_cgroupstats_operations;
571
572static struct backing_dev_info cgroup_backing_dev_info = {
573 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
574};
575
576static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
577{
578 struct inode *inode = new_inode(sb);
579
580 if (inode) {
581 inode->i_mode = mode;
582 inode->i_uid = current->fsuid;
583 inode->i_gid = current->fsgid;
584 inode->i_blocks = 0;
585 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
586 inode->i_mapping->backing_dev_info = &cgroup_backing_dev_info;
587 }
588 return inode;
589}
590
591static void cgroup_diput(struct dentry *dentry, struct inode *inode)
592{
593 /* is dentry a directory ? if so, kfree() associated cgroup */
594 if (S_ISDIR(inode->i_mode)) {
595 struct cgroup *cgrp = dentry->d_fsdata;
596 BUG_ON(!(cgroup_is_removed(cgrp)));
597 /* It's possible for external users to be holding css
598 * reference counts on a cgroup; css_put() needs to
599 * be able to access the cgroup after decrementing
600 * the reference count in order to know if it needs to
601 * queue the cgroup to be handled by the release
602 * agent */
603 synchronize_rcu();
604 kfree(cgrp);
605 }
606 iput(inode);
607}
608
609static void remove_dir(struct dentry *d)
610{
611 struct dentry *parent = dget(d->d_parent);
612
613 d_delete(d);
614 simple_rmdir(parent->d_inode, d);
615 dput(parent);
616}
617
618static void cgroup_clear_directory(struct dentry *dentry)
619{
620 struct list_head *node;
621
622 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
623 spin_lock(&dcache_lock);
624 node = dentry->d_subdirs.next;
625 while (node != &dentry->d_subdirs) {
626 struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
627 list_del_init(node);
628 if (d->d_inode) {
629 /* This should never be called on a cgroup
630 * directory with child cgroups */
631 BUG_ON(d->d_inode->i_mode & S_IFDIR);
632 d = dget_locked(d);
633 spin_unlock(&dcache_lock);
634 d_delete(d);
635 simple_unlink(dentry->d_inode, d);
636 dput(d);
637 spin_lock(&dcache_lock);
638 }
639 node = dentry->d_subdirs.next;
640 }
641 spin_unlock(&dcache_lock);
642}
643
644/*
645 * NOTE : the dentry must have been dget()'ed
646 */
647static void cgroup_d_remove_dir(struct dentry *dentry)
648{
649 cgroup_clear_directory(dentry);
650
651 spin_lock(&dcache_lock);
652 list_del_init(&dentry->d_u.d_child);
653 spin_unlock(&dcache_lock);
654 remove_dir(dentry);
655}
656
657static int rebind_subsystems(struct cgroupfs_root *root,
658 unsigned long final_bits)
659{
660 unsigned long added_bits, removed_bits;
661 struct cgroup *cgrp = &root->top_cgroup;
662 int i;
663
664 removed_bits = root->actual_subsys_bits & ~final_bits;
665 added_bits = final_bits & ~root->actual_subsys_bits;
666 /* Check that any added subsystems are currently free */
667 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
668 unsigned long long bit = 1ull << i;
669 struct cgroup_subsys *ss = subsys[i];
670 if (!(bit & added_bits))
671 continue;
672 if (ss->root != &rootnode) {
673 /* Subsystem isn't free */
674 return -EBUSY;
675 }
676 }
677
678 /* Currently we don't handle adding/removing subsystems when
679 * any child cgroups exist. This is theoretically supportable
680 * but involves complex error handling, so it's being left until
681 * later */
682 if (!list_empty(&cgrp->children))
683 return -EBUSY;
684
685 /* Process each subsystem */
686 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
687 struct cgroup_subsys *ss = subsys[i];
688 unsigned long bit = 1UL << i;
689 if (bit & added_bits) {
690 /* We're binding this subsystem to this hierarchy */
691 BUG_ON(cgrp->subsys[i]);
692 BUG_ON(!dummytop->subsys[i]);
693 BUG_ON(dummytop->subsys[i]->cgroup != dummytop);
694 cgrp->subsys[i] = dummytop->subsys[i];
695 cgrp->subsys[i]->cgroup = cgrp;
696 list_add(&ss->sibling, &root->subsys_list);
697 rcu_assign_pointer(ss->root, root);
698 if (ss->bind)
699 ss->bind(ss, cgrp);
700
701 } else if (bit & removed_bits) {
702 /* We're removing this subsystem */
703 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
704 BUG_ON(cgrp->subsys[i]->cgroup != cgrp);
705 if (ss->bind)
706 ss->bind(ss, dummytop);
707 dummytop->subsys[i]->cgroup = dummytop;
708 cgrp->subsys[i] = NULL;
709 rcu_assign_pointer(subsys[i]->root, &rootnode);
710 list_del(&ss->sibling);
711 } else if (bit & final_bits) {
712 /* Subsystem state should already exist */
713 BUG_ON(!cgrp->subsys[i]);
714 } else {
715 /* Subsystem state shouldn't exist */
716 BUG_ON(cgrp->subsys[i]);
717 }
718 }
719 root->subsys_bits = root->actual_subsys_bits = final_bits;
720 synchronize_rcu();
721
722 return 0;
723}
724
725static int cgroup_show_options(struct seq_file *seq, struct vfsmount *vfs)
726{
727 struct cgroupfs_root *root = vfs->mnt_sb->s_fs_info;
728 struct cgroup_subsys *ss;
729
730 mutex_lock(&cgroup_mutex);
731 for_each_subsys(root, ss)
732 seq_printf(seq, ",%s", ss->name);
733 if (test_bit(ROOT_NOPREFIX, &root->flags))
734 seq_puts(seq, ",noprefix");
735 if (strlen(root->release_agent_path))
736 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
737 mutex_unlock(&cgroup_mutex);
738 return 0;
739}
740
741struct cgroup_sb_opts {
742 unsigned long subsys_bits;
743 unsigned long flags;
744 char *release_agent;
745};
746
747/* Convert a hierarchy specifier into a bitmask of subsystems and
748 * flags. */
749static int parse_cgroupfs_options(char *data,
750 struct cgroup_sb_opts *opts)
751{
752 char *token, *o = data ?: "all";
753
754 opts->subsys_bits = 0;
755 opts->flags = 0;
756 opts->release_agent = NULL;
757
758 while ((token = strsep(&o, ",")) != NULL) {
759 if (!*token)
760 return -EINVAL;
761 if (!strcmp(token, "all")) {
762 opts->subsys_bits = (1 << CGROUP_SUBSYS_COUNT) - 1;
763 } else if (!strcmp(token, "noprefix")) {
764 set_bit(ROOT_NOPREFIX, &opts->flags);
765 } else if (!strncmp(token, "release_agent=", 14)) {
766 /* Specifying two release agents is forbidden */
767 if (opts->release_agent)
768 return -EINVAL;
769 opts->release_agent = kzalloc(PATH_MAX, GFP_KERNEL);
770 if (!opts->release_agent)
771 return -ENOMEM;
772 strncpy(opts->release_agent, token + 14, PATH_MAX - 1);
773 opts->release_agent[PATH_MAX - 1] = 0;
774 } else {
775 struct cgroup_subsys *ss;
776 int i;
777 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
778 ss = subsys[i];
779 if (!strcmp(token, ss->name)) {
780 set_bit(i, &opts->subsys_bits);
781 break;
782 }
783 }
784 if (i == CGROUP_SUBSYS_COUNT)
785 return -ENOENT;
786 }
787 }
788
789 /* We can't have an empty hierarchy */
790 if (!opts->subsys_bits)
791 return -EINVAL;
792
793 return 0;
794}
795
796static int cgroup_remount(struct super_block *sb, int *flags, char *data)
797{
798 int ret = 0;
799 struct cgroupfs_root *root = sb->s_fs_info;
800 struct cgroup *cgrp = &root->top_cgroup;
801 struct cgroup_sb_opts opts;
802
803 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
804 mutex_lock(&cgroup_mutex);
805
806 /* See what subsystems are wanted */
807 ret = parse_cgroupfs_options(data, &opts);
808 if (ret)
809 goto out_unlock;
810
811 /* Don't allow flags to change at remount */
812 if (opts.flags != root->flags) {
813 ret = -EINVAL;
814 goto out_unlock;
815 }
816
817 ret = rebind_subsystems(root, opts.subsys_bits);
818
819 /* (re)populate subsystem files */
820 if (!ret)
821 cgroup_populate_dir(cgrp);
822
823 if (opts.release_agent)
824 strcpy(root->release_agent_path, opts.release_agent);
825 out_unlock:
826 if (opts.release_agent)
827 kfree(opts.release_agent);
828 mutex_unlock(&cgroup_mutex);
829 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
830 return ret;
831}
832
833static struct super_operations cgroup_ops = {
834 .statfs = simple_statfs,
835 .drop_inode = generic_delete_inode,
836 .show_options = cgroup_show_options,
837 .remount_fs = cgroup_remount,
838};
839
840static void init_cgroup_root(struct cgroupfs_root *root)
841{
842 struct cgroup *cgrp = &root->top_cgroup;
843 INIT_LIST_HEAD(&root->subsys_list);
844 INIT_LIST_HEAD(&root->root_list);
845 root->number_of_cgroups = 1;
846 cgrp->root = root;
847 cgrp->top_cgroup = cgrp;
848 INIT_LIST_HEAD(&cgrp->sibling);
849 INIT_LIST_HEAD(&cgrp->children);
850 INIT_LIST_HEAD(&cgrp->css_sets);
851 INIT_LIST_HEAD(&cgrp->release_list);
852}
853
854static int cgroup_test_super(struct super_block *sb, void *data)
855{
856 struct cgroupfs_root *new = data;
857 struct cgroupfs_root *root = sb->s_fs_info;
858
859 /* First check subsystems */
860 if (new->subsys_bits != root->subsys_bits)
861 return 0;
862
863 /* Next check flags */
864 if (new->flags != root->flags)
865 return 0;
866
867 return 1;
868}
869
870static int cgroup_set_super(struct super_block *sb, void *data)
871{
872 int ret;
873 struct cgroupfs_root *root = data;
874
875 ret = set_anon_super(sb, NULL);
876 if (ret)
877 return ret;
878
879 sb->s_fs_info = root;
880 root->sb = sb;
881
882 sb->s_blocksize = PAGE_CACHE_SIZE;
883 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
884 sb->s_magic = CGROUP_SUPER_MAGIC;
885 sb->s_op = &cgroup_ops;
886
887 return 0;
888}
889
890static int cgroup_get_rootdir(struct super_block *sb)
891{
892 struct inode *inode =
893 cgroup_new_inode(S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR, sb);
894 struct dentry *dentry;
895
896 if (!inode)
897 return -ENOMEM;
898
899 inode->i_op = &simple_dir_inode_operations;
900 inode->i_fop = &simple_dir_operations;
901 inode->i_op = &cgroup_dir_inode_operations;
902 /* directories start off with i_nlink == 2 (for "." entry) */
903 inc_nlink(inode);
904 dentry = d_alloc_root(inode);
905 if (!dentry) {
906 iput(inode);
907 return -ENOMEM;
908 }
909 sb->s_root = dentry;
910 return 0;
911}
912
913static int cgroup_get_sb(struct file_system_type *fs_type,
914 int flags, const char *unused_dev_name,
915 void *data, struct vfsmount *mnt)
916{
917 struct cgroup_sb_opts opts;
918 int ret = 0;
919 struct super_block *sb;
920 struct cgroupfs_root *root;
921 struct list_head tmp_cg_links, *l;
922 INIT_LIST_HEAD(&tmp_cg_links);
923
924 /* First find the desired set of subsystems */
925 ret = parse_cgroupfs_options(data, &opts);
926 if (ret) {
927 if (opts.release_agent)
928 kfree(opts.release_agent);
929 return ret;
930 }
931
932 root = kzalloc(sizeof(*root), GFP_KERNEL);
933 if (!root)
934 return -ENOMEM;
935
936 init_cgroup_root(root);
937 root->subsys_bits = opts.subsys_bits;
938 root->flags = opts.flags;
939 if (opts.release_agent) {
940 strcpy(root->release_agent_path, opts.release_agent);
941 kfree(opts.release_agent);
942 }
943
944 sb = sget(fs_type, cgroup_test_super, cgroup_set_super, root);
945
946 if (IS_ERR(sb)) {
947 kfree(root);
948 return PTR_ERR(sb);
949 }
950
951 if (sb->s_fs_info != root) {
952 /* Reusing an existing superblock */
953 BUG_ON(sb->s_root == NULL);
954 kfree(root);
955 root = NULL;
956 } else {
957 /* New superblock */
958 struct cgroup *cgrp = &root->top_cgroup;
959 struct inode *inode;
960
961 BUG_ON(sb->s_root != NULL);
962
963 ret = cgroup_get_rootdir(sb);
964 if (ret)
965 goto drop_new_super;
966 inode = sb->s_root->d_inode;
967
968 mutex_lock(&inode->i_mutex);
969 mutex_lock(&cgroup_mutex);
970
971 /*
972 * We're accessing css_set_count without locking
973 * css_set_lock here, but that's OK - it can only be
974 * increased by someone holding cgroup_lock, and
975 * that's us. The worst that can happen is that we
976 * have some link structures left over
977 */
978 ret = allocate_cg_links(css_set_count, &tmp_cg_links);
979 if (ret) {
980 mutex_unlock(&cgroup_mutex);
981 mutex_unlock(&inode->i_mutex);
982 goto drop_new_super;
983 }
984
985 ret = rebind_subsystems(root, root->subsys_bits);
986 if (ret == -EBUSY) {
987 mutex_unlock(&cgroup_mutex);
988 mutex_unlock(&inode->i_mutex);
989 goto drop_new_super;
990 }
991
992 /* EBUSY should be the only error here */
993 BUG_ON(ret);
994
995 list_add(&root->root_list, &roots);
996 root_count++;
997
998 sb->s_root->d_fsdata = &root->top_cgroup;
999 root->top_cgroup.dentry = sb->s_root;
1000
1001 /* Link the top cgroup in this hierarchy into all
1002 * the css_set objects */
1003 write_lock(&css_set_lock);
1004 l = &init_css_set.list;
1005 do {
1006 struct css_set *cg;
1007 struct cg_cgroup_link *link;
1008 cg = list_entry(l, struct css_set, list);
1009 BUG_ON(list_empty(&tmp_cg_links));
1010 link = list_entry(tmp_cg_links.next,
1011 struct cg_cgroup_link,
1012 cgrp_link_list);
1013 list_del(&link->cgrp_link_list);
1014 link->cg = cg;
1015 list_add(&link->cgrp_link_list,
1016 &root->top_cgroup.css_sets);
1017 list_add(&link->cg_link_list, &cg->cg_links);
1018 l = l->next;
1019 } while (l != &init_css_set.list);
1020 write_unlock(&css_set_lock);
1021
1022 free_cg_links(&tmp_cg_links);
1023
1024 BUG_ON(!list_empty(&cgrp->sibling));
1025 BUG_ON(!list_empty(&cgrp->children));
1026 BUG_ON(root->number_of_cgroups != 1);
1027
1028 cgroup_populate_dir(cgrp);
1029 mutex_unlock(&inode->i_mutex);
1030 mutex_unlock(&cgroup_mutex);
1031 }
1032
1033 return simple_set_mnt(mnt, sb);
1034
1035 drop_new_super:
1036 up_write(&sb->s_umount);
1037 deactivate_super(sb);
1038 free_cg_links(&tmp_cg_links);
1039 return ret;
1040}
1041
1042static void cgroup_kill_sb(struct super_block *sb) {
1043 struct cgroupfs_root *root = sb->s_fs_info;
1044 struct cgroup *cgrp = &root->top_cgroup;
1045 int ret;
1046
1047 BUG_ON(!root);
1048
1049 BUG_ON(root->number_of_cgroups != 1);
1050 BUG_ON(!list_empty(&cgrp->children));
1051 BUG_ON(!list_empty(&cgrp->sibling));
1052
1053 mutex_lock(&cgroup_mutex);
1054
1055 /* Rebind all subsystems back to the default hierarchy */
1056 ret = rebind_subsystems(root, 0);
1057 /* Shouldn't be able to fail ... */
1058 BUG_ON(ret);
1059
1060 /*
1061 * Release all the links from css_sets to this hierarchy's
1062 * root cgroup
1063 */
1064 write_lock(&css_set_lock);
1065 while (!list_empty(&cgrp->css_sets)) {
1066 struct cg_cgroup_link *link;
1067 link = list_entry(cgrp->css_sets.next,
1068 struct cg_cgroup_link, cgrp_link_list);
1069 list_del(&link->cg_link_list);
1070 list_del(&link->cgrp_link_list);
1071 kfree(link);
1072 }
1073 write_unlock(&css_set_lock);
1074
1075 if (!list_empty(&root->root_list)) {
1076 list_del(&root->root_list);
1077 root_count--;
1078 }
1079 mutex_unlock(&cgroup_mutex);
1080
1081 kfree(root);
1082 kill_litter_super(sb);
1083}
1084
1085static struct file_system_type cgroup_fs_type = {
1086 .name = "cgroup",
1087 .get_sb = cgroup_get_sb,
1088 .kill_sb = cgroup_kill_sb,
1089};
1090
1091static inline struct cgroup *__d_cgrp(struct dentry *dentry)
1092{
1093 return dentry->d_fsdata;
1094}
1095
1096static inline struct cftype *__d_cft(struct dentry *dentry)
1097{
1098 return dentry->d_fsdata;
1099}
1100
1101/*
1102 * Called with cgroup_mutex held. Writes path of cgroup into buf.
1103 * Returns 0 on success, -errno on error.
1104 */
1105int cgroup_path(const struct cgroup *cgrp, char *buf, int buflen)
1106{
1107 char *start;
1108
1109 if (cgrp == dummytop) {
1110 /*
1111 * Inactive subsystems have no dentry for their root
1112 * cgroup
1113 */
1114 strcpy(buf, "/");
1115 return 0;
1116 }
1117
1118 start = buf + buflen;
1119
1120 *--start = '\0';
1121 for (;;) {
1122 int len = cgrp->dentry->d_name.len;
1123 if ((start -= len) < buf)
1124 return -ENAMETOOLONG;
1125 memcpy(start, cgrp->dentry->d_name.name, len);
1126 cgrp = cgrp->parent;
1127 if (!cgrp)
1128 break;
1129 if (!cgrp->parent)
1130 continue;
1131 if (--start < buf)
1132 return -ENAMETOOLONG;
1133 *start = '/';
1134 }
1135 memmove(buf, start, buf + buflen - start);
1136 return 0;
1137}
1138
1139/*
1140 * Return the first subsystem attached to a cgroup's hierarchy, and
1141 * its subsystem id.
1142 */
1143
1144static void get_first_subsys(const struct cgroup *cgrp,
1145 struct cgroup_subsys_state **css, int *subsys_id)
1146{
1147 const struct cgroupfs_root *root = cgrp->root;
1148 const struct cgroup_subsys *test_ss;
1149 BUG_ON(list_empty(&root->subsys_list));
1150 test_ss = list_entry(root->subsys_list.next,
1151 struct cgroup_subsys, sibling);
1152 if (css) {
1153 *css = cgrp->subsys[test_ss->subsys_id];
1154 BUG_ON(!*css);
1155 }
1156 if (subsys_id)
1157 *subsys_id = test_ss->subsys_id;
1158}
1159
1160/*
1161 * Attach task 'tsk' to cgroup 'cgrp'
1162 *
1163 * Call holding cgroup_mutex. May take task_lock of
1164 * the task 'pid' during call.
1165 */
1166static int attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1167{
1168 int retval = 0;
1169 struct cgroup_subsys *ss;
1170 struct cgroup *oldcgrp;
1171 struct css_set *cg = tsk->cgroups;
1172 struct css_set *newcg;
1173 struct cgroupfs_root *root = cgrp->root;
1174 int subsys_id;
1175
1176 get_first_subsys(cgrp, NULL, &subsys_id);
1177
1178 /* Nothing to do if the task is already in that cgroup */
1179 oldcgrp = task_cgroup(tsk, subsys_id);
1180 if (cgrp == oldcgrp)
1181 return 0;
1182
1183 for_each_subsys(root, ss) {
1184 if (ss->can_attach) {
1185 retval = ss->can_attach(ss, cgrp, tsk);
1186 if (retval) {
1187 return retval;
1188 }
1189 }
1190 }
1191
1192 /*
1193 * Locate or allocate a new css_set for this task,
1194 * based on its final set of cgroups
1195 */
1196 newcg = find_css_set(cg, cgrp);
1197 if (!newcg) {
1198 return -ENOMEM;
1199 }
1200
1201 task_lock(tsk);
1202 if (tsk->flags & PF_EXITING) {
1203 task_unlock(tsk);
1204 put_css_set(newcg);
1205 return -ESRCH;
1206 }
1207 rcu_assign_pointer(tsk->cgroups, newcg);
1208 task_unlock(tsk);
1209
1210 /* Update the css_set linked lists if we're using them */
1211 write_lock(&css_set_lock);
1212 if (!list_empty(&tsk->cg_list)) {
1213 list_del(&tsk->cg_list);
1214 list_add(&tsk->cg_list, &newcg->tasks);
1215 }
1216 write_unlock(&css_set_lock);
1217
1218 for_each_subsys(root, ss) {
1219 if (ss->attach) {
1220 ss->attach(ss, cgrp, oldcgrp, tsk);
1221 }
1222 }
1223 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1224 synchronize_rcu();
1225 put_css_set(cg);
1226 return 0;
1227}
1228
1229/*
1230 * Attach task with pid 'pid' to cgroup 'cgrp'. Call with
1231 * cgroup_mutex, may take task_lock of task
1232 */
1233static int attach_task_by_pid(struct cgroup *cgrp, char *pidbuf)
1234{
1235 pid_t pid;
1236 struct task_struct *tsk;
1237 int ret;
1238
1239 if (sscanf(pidbuf, "%d", &pid) != 1)
1240 return -EIO;
1241
1242 if (pid) {
1243 rcu_read_lock();
1244 tsk = find_task_by_pid(pid);
1245 if (!tsk || tsk->flags & PF_EXITING) {
1246 rcu_read_unlock();
1247 return -ESRCH;
1248 }
1249 get_task_struct(tsk);
1250 rcu_read_unlock();
1251
1252 if ((current->euid) && (current->euid != tsk->uid)
1253 && (current->euid != tsk->suid)) {
1254 put_task_struct(tsk);
1255 return -EACCES;
1256 }
1257 } else {
1258 tsk = current;
1259 get_task_struct(tsk);
1260 }
1261
1262 ret = attach_task(cgrp, tsk);
1263 put_task_struct(tsk);
1264 return ret;
1265}
1266
1267/* The various types of files and directories in a cgroup file system */
1268
1269enum cgroup_filetype {
1270 FILE_ROOT,
1271 FILE_DIR,
1272 FILE_TASKLIST,
1273 FILE_NOTIFY_ON_RELEASE,
1274 FILE_RELEASABLE,
1275 FILE_RELEASE_AGENT,
1276};
1277
1278static ssize_t cgroup_write_uint(struct cgroup *cgrp, struct cftype *cft,
1279 struct file *file,
1280 const char __user *userbuf,
1281 size_t nbytes, loff_t *unused_ppos)
1282{
1283 char buffer[64];
1284 int retval = 0;
1285 u64 val;
1286 char *end;
1287
1288 if (!nbytes)
1289 return -EINVAL;
1290 if (nbytes >= sizeof(buffer))
1291 return -E2BIG;
1292 if (copy_from_user(buffer, userbuf, nbytes))
1293 return -EFAULT;
1294
1295 buffer[nbytes] = 0; /* nul-terminate */
1296
1297 /* strip newline if necessary */
1298 if (nbytes && (buffer[nbytes-1] == '\n'))
1299 buffer[nbytes-1] = 0;
1300 val = simple_strtoull(buffer, &end, 0);
1301 if (*end)
1302 return -EINVAL;
1303
1304 /* Pass to subsystem */
1305 retval = cft->write_uint(cgrp, cft, val);
1306 if (!retval)
1307 retval = nbytes;
1308 return retval;
1309}
1310
1311static ssize_t cgroup_common_file_write(struct cgroup *cgrp,
1312 struct cftype *cft,
1313 struct file *file,
1314 const char __user *userbuf,
1315 size_t nbytes, loff_t *unused_ppos)
1316{
1317 enum cgroup_filetype type = cft->private;
1318 char *buffer;
1319 int retval = 0;
1320
1321 if (nbytes >= PATH_MAX)
1322 return -E2BIG;
1323
1324 /* +1 for nul-terminator */
1325 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1326 if (buffer == NULL)
1327 return -ENOMEM;
1328
1329 if (copy_from_user(buffer, userbuf, nbytes)) {
1330 retval = -EFAULT;
1331 goto out1;
1332 }
1333 buffer[nbytes] = 0; /* nul-terminate */
1334
1335 mutex_lock(&cgroup_mutex);
1336
1337 if (cgroup_is_removed(cgrp)) {
1338 retval = -ENODEV;
1339 goto out2;
1340 }
1341
1342 switch (type) {
1343 case FILE_TASKLIST:
1344 retval = attach_task_by_pid(cgrp, buffer);
1345 break;
1346 case FILE_NOTIFY_ON_RELEASE:
1347 clear_bit(CGRP_RELEASABLE, &cgrp->flags);
1348 if (simple_strtoul(buffer, NULL, 10) != 0)
1349 set_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
1350 else
1351 clear_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
1352 break;
1353 case FILE_RELEASE_AGENT:
1354 {
1355 struct cgroupfs_root *root = cgrp->root;
1356 /* Strip trailing newline */
1357 if (nbytes && (buffer[nbytes-1] == '\n')) {
1358 buffer[nbytes-1] = 0;
1359 }
1360 if (nbytes < sizeof(root->release_agent_path)) {
1361 /* We never write anything other than '\0'
1362 * into the last char of release_agent_path,
1363 * so it always remains a NUL-terminated
1364 * string */
1365 strncpy(root->release_agent_path, buffer, nbytes);
1366 root->release_agent_path[nbytes] = 0;
1367 } else {
1368 retval = -ENOSPC;
1369 }
1370 break;
1371 }
1372 default:
1373 retval = -EINVAL;
1374 goto out2;
1375 }
1376
1377 if (retval == 0)
1378 retval = nbytes;
1379out2:
1380 mutex_unlock(&cgroup_mutex);
1381out1:
1382 kfree(buffer);
1383 return retval;
1384}
1385
1386static ssize_t cgroup_file_write(struct file *file, const char __user *buf,
1387 size_t nbytes, loff_t *ppos)
1388{
1389 struct cftype *cft = __d_cft(file->f_dentry);
1390 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1391
1392 if (!cft)
1393 return -ENODEV;
1394 if (cft->write)
1395 return cft->write(cgrp, cft, file, buf, nbytes, ppos);
1396 if (cft->write_uint)
1397 return cgroup_write_uint(cgrp, cft, file, buf, nbytes, ppos);
1398 return -EINVAL;
1399}
1400
1401static ssize_t cgroup_read_uint(struct cgroup *cgrp, struct cftype *cft,
1402 struct file *file,
1403 char __user *buf, size_t nbytes,
1404 loff_t *ppos)
1405{
1406 char tmp[64];
1407 u64 val = cft->read_uint(cgrp, cft);
1408 int len = sprintf(tmp, "%llu\n", (unsigned long long) val);
1409
1410 return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
1411}
1412
1413static ssize_t cgroup_common_file_read(struct cgroup *cgrp,
1414 struct cftype *cft,
1415 struct file *file,
1416 char __user *buf,
1417 size_t nbytes, loff_t *ppos)
1418{
1419 enum cgroup_filetype type = cft->private;
1420 char *page;
1421 ssize_t retval = 0;
1422 char *s;
1423
1424 if (!(page = (char *)__get_free_page(GFP_KERNEL)))
1425 return -ENOMEM;
1426
1427 s = page;
1428
1429 switch (type) {
1430 case FILE_RELEASE_AGENT:
1431 {
1432 struct cgroupfs_root *root;
1433 size_t n;
1434 mutex_lock(&cgroup_mutex);
1435 root = cgrp->root;
1436 n = strnlen(root->release_agent_path,
1437 sizeof(root->release_agent_path));
1438 n = min(n, (size_t) PAGE_SIZE);
1439 strncpy(s, root->release_agent_path, n);
1440 mutex_unlock(&cgroup_mutex);
1441 s += n;
1442 break;
1443 }
1444 default:
1445 retval = -EINVAL;
1446 goto out;
1447 }
1448 *s++ = '\n';
1449
1450 retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1451out:
1452 free_page((unsigned long)page);
1453 return retval;
1454}
1455
1456static ssize_t cgroup_file_read(struct file *file, char __user *buf,
1457 size_t nbytes, loff_t *ppos)
1458{
1459 struct cftype *cft = __d_cft(file->f_dentry);
1460 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1461
1462 if (!cft)
1463 return -ENODEV;
1464
1465 if (cft->read)
1466 return cft->read(cgrp, cft, file, buf, nbytes, ppos);
1467 if (cft->read_uint)
1468 return cgroup_read_uint(cgrp, cft, file, buf, nbytes, ppos);
1469 return -EINVAL;
1470}
1471
1472static int cgroup_file_open(struct inode *inode, struct file *file)
1473{
1474 int err;
1475 struct cftype *cft;
1476
1477 err = generic_file_open(inode, file);
1478 if (err)
1479 return err;
1480
1481 cft = __d_cft(file->f_dentry);
1482 if (!cft)
1483 return -ENODEV;
1484 if (cft->open)
1485 err = cft->open(inode, file);
1486 else
1487 err = 0;
1488
1489 return err;
1490}
1491
1492static int cgroup_file_release(struct inode *inode, struct file *file)
1493{
1494 struct cftype *cft = __d_cft(file->f_dentry);
1495 if (cft->release)
1496 return cft->release(inode, file);
1497 return 0;
1498}
1499
1500/*
1501 * cgroup_rename - Only allow simple rename of directories in place.
1502 */
1503static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
1504 struct inode *new_dir, struct dentry *new_dentry)
1505{
1506 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1507 return -ENOTDIR;
1508 if (new_dentry->d_inode)
1509 return -EEXIST;
1510 if (old_dir != new_dir)
1511 return -EIO;
1512 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
1513}
1514
1515static struct file_operations cgroup_file_operations = {
1516 .read = cgroup_file_read,
1517 .write = cgroup_file_write,
1518 .llseek = generic_file_llseek,
1519 .open = cgroup_file_open,
1520 .release = cgroup_file_release,
1521};
1522
1523static struct inode_operations cgroup_dir_inode_operations = {
1524 .lookup = simple_lookup,
1525 .mkdir = cgroup_mkdir,
1526 .rmdir = cgroup_rmdir,
1527 .rename = cgroup_rename,
1528};
1529
1530static int cgroup_create_file(struct dentry *dentry, int mode,
1531 struct super_block *sb)
1532{
1533 static struct dentry_operations cgroup_dops = {
1534 .d_iput = cgroup_diput,
1535 };
1536
1537 struct inode *inode;
1538
1539 if (!dentry)
1540 return -ENOENT;
1541 if (dentry->d_inode)
1542 return -EEXIST;
1543
1544 inode = cgroup_new_inode(mode, sb);
1545 if (!inode)
1546 return -ENOMEM;
1547
1548 if (S_ISDIR(mode)) {
1549 inode->i_op = &cgroup_dir_inode_operations;
1550 inode->i_fop = &simple_dir_operations;
1551
1552 /* start off with i_nlink == 2 (for "." entry) */
1553 inc_nlink(inode);
1554
1555 /* start with the directory inode held, so that we can
1556 * populate it without racing with another mkdir */
1557 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1558 } else if (S_ISREG(mode)) {
1559 inode->i_size = 0;
1560 inode->i_fop = &cgroup_file_operations;
1561 }
1562 dentry->d_op = &cgroup_dops;
1563 d_instantiate(dentry, inode);
1564 dget(dentry); /* Extra count - pin the dentry in core */
1565 return 0;
1566}
1567
1568/*
1569 * cgroup_create_dir - create a directory for an object.
1570 * cgrp: the cgroup we create the directory for.
1571 * It must have a valid ->parent field
1572 * And we are going to fill its ->dentry field.
1573 * dentry: dentry of the new cgroup
1574 * mode: mode to set on new directory.
1575 */
1576static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
1577 int mode)
1578{
1579 struct dentry *parent;
1580 int error = 0;
1581
1582 parent = cgrp->parent->dentry;
1583 error = cgroup_create_file(dentry, S_IFDIR | mode, cgrp->root->sb);
1584 if (!error) {
1585 dentry->d_fsdata = cgrp;
1586 inc_nlink(parent->d_inode);
1587 cgrp->dentry = dentry;
1588 dget(dentry);
1589 }
1590 dput(dentry);
1591
1592 return error;
1593}
1594
1595int cgroup_add_file(struct cgroup *cgrp,
1596 struct cgroup_subsys *subsys,
1597 const struct cftype *cft)
1598{
1599 struct dentry *dir = cgrp->dentry;
1600 struct dentry *dentry;
1601 int error;
1602
1603 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
1604 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
1605 strcpy(name, subsys->name);
1606 strcat(name, ".");
1607 }
1608 strcat(name, cft->name);
1609 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
1610 dentry = lookup_one_len(name, dir, strlen(name));
1611 if (!IS_ERR(dentry)) {
1612 error = cgroup_create_file(dentry, 0644 | S_IFREG,
1613 cgrp->root->sb);
1614 if (!error)
1615 dentry->d_fsdata = (void *)cft;
1616 dput(dentry);
1617 } else
1618 error = PTR_ERR(dentry);
1619 return error;
1620}
1621
1622int cgroup_add_files(struct cgroup *cgrp,
1623 struct cgroup_subsys *subsys,
1624 const struct cftype cft[],
1625 int count)
1626{
1627 int i, err;
1628 for (i = 0; i < count; i++) {
1629 err = cgroup_add_file(cgrp, subsys, &cft[i]);
1630 if (err)
1631 return err;
1632 }
1633 return 0;
1634}
1635
1636/* Count the number of tasks in a cgroup. */
1637
1638int cgroup_task_count(const struct cgroup *cgrp)
1639{
1640 int count = 0;
1641 struct list_head *l;
1642
1643 read_lock(&css_set_lock);
1644 l = cgrp->css_sets.next;
1645 while (l != &cgrp->css_sets) {
1646 struct cg_cgroup_link *link =
1647 list_entry(l, struct cg_cgroup_link, cgrp_link_list);
1648 count += atomic_read(&link->cg->ref.refcount);
1649 l = l->next;
1650 }
1651 read_unlock(&css_set_lock);
1652 return count;
1653}
1654
1655/*
1656 * Advance a list_head iterator. The iterator should be positioned at
1657 * the start of a css_set
1658 */
1659static void cgroup_advance_iter(struct cgroup *cgrp,
1660 struct cgroup_iter *it)
1661{
1662 struct list_head *l = it->cg_link;
1663 struct cg_cgroup_link *link;
1664 struct css_set *cg;
1665
1666 /* Advance to the next non-empty css_set */
1667 do {
1668 l = l->next;
1669 if (l == &cgrp->css_sets) {
1670 it->cg_link = NULL;
1671 return;
1672 }
1673 link = list_entry(l, struct cg_cgroup_link, cgrp_link_list);
1674 cg = link->cg;
1675 } while (list_empty(&cg->tasks));
1676 it->cg_link = l;
1677 it->task = cg->tasks.next;
1678}
1679
1680void cgroup_iter_start(struct cgroup *cgrp, struct cgroup_iter *it)
1681{
1682 /*
1683 * The first time anyone tries to iterate across a cgroup,
1684 * we need to enable the list linking each css_set to its
1685 * tasks, and fix up all existing tasks.
1686 */
1687 if (!use_task_css_set_links) {
1688 struct task_struct *p, *g;
1689 write_lock(&css_set_lock);
1690 use_task_css_set_links = 1;
1691 do_each_thread(g, p) {
1692 task_lock(p);
1693 if (list_empty(&p->cg_list))
1694 list_add(&p->cg_list, &p->cgroups->tasks);
1695 task_unlock(p);
1696 } while_each_thread(g, p);
1697 write_unlock(&css_set_lock);
1698 }
1699 read_lock(&css_set_lock);
1700 it->cg_link = &cgrp->css_sets;
1701 cgroup_advance_iter(cgrp, it);
1702}
1703
1704struct task_struct *cgroup_iter_next(struct cgroup *cgrp,
1705 struct cgroup_iter *it)
1706{
1707 struct task_struct *res;
1708 struct list_head *l = it->task;
1709
1710 /* If the iterator cg is NULL, we have no tasks */
1711 if (!it->cg_link)
1712 return NULL;
1713 res = list_entry(l, struct task_struct, cg_list);
1714 /* Advance iterator to find next entry */
1715 l = l->next;
1716 if (l == &res->cgroups->tasks) {
1717 /* We reached the end of this task list - move on to
1718 * the next cg_cgroup_link */
1719 cgroup_advance_iter(cgrp, it);
1720 } else {
1721 it->task = l;
1722 }
1723 return res;
1724}
1725
1726void cgroup_iter_end(struct cgroup *cgrp, struct cgroup_iter *it)
1727{
1728 read_unlock(&css_set_lock);
1729}
1730
1731/*
1732 * Stuff for reading the 'tasks' file.
1733 *
1734 * Reading this file can return large amounts of data if a cgroup has
1735 * *lots* of attached tasks. So it may need several calls to read(),
1736 * but we cannot guarantee that the information we produce is correct
1737 * unless we produce it entirely atomically.
1738 *
1739 * Upon tasks file open(), a struct ctr_struct is allocated, that
1740 * will have a pointer to an array (also allocated here). The struct
1741 * ctr_struct * is stored in file->private_data. Its resources will
1742 * be freed by release() when the file is closed. The array is used
1743 * to sprintf the PIDs and then used by read().
1744 */
1745struct ctr_struct {
1746 char *buf;
1747 int bufsz;
1748};
1749
1750/*
1751 * Load into 'pidarray' up to 'npids' of the tasks using cgroup
1752 * 'cgrp'. Return actual number of pids loaded. No need to
1753 * task_lock(p) when reading out p->cgroup, since we're in an RCU
1754 * read section, so the css_set can't go away, and is
1755 * immutable after creation.
1756 */
1757static int pid_array_load(pid_t *pidarray, int npids, struct cgroup *cgrp)
1758{
1759 int n = 0;
1760 struct cgroup_iter it;
1761 struct task_struct *tsk;
1762 cgroup_iter_start(cgrp, &it);
1763 while ((tsk = cgroup_iter_next(cgrp, &it))) {
1764 if (unlikely(n == npids))
1765 break;
1766 pidarray[n++] = task_pid_nr(tsk);
1767 }
1768 cgroup_iter_end(cgrp, &it);
1769 return n;
1770}
1771
1772/**
1773 * Build and fill cgroupstats so that taskstats can export it to user
1774 * space.
1775 *
1776 * @stats: cgroupstats to fill information into
1777 * @dentry: A dentry entry belonging to the cgroup for which stats have
1778 * been requested.
1779 */
1780int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
1781{
1782 int ret = -EINVAL;
1783 struct cgroup *cgrp;
1784 struct cgroup_iter it;
1785 struct task_struct *tsk;
1786 /*
1787 * Validate dentry by checking the superblock operations
1788 */
1789 if (dentry->d_sb->s_op != &cgroup_ops)
1790 goto err;
1791
1792 ret = 0;
1793 cgrp = dentry->d_fsdata;
1794 rcu_read_lock();
1795
1796 cgroup_iter_start(cgrp, &it);
1797 while ((tsk = cgroup_iter_next(cgrp, &it))) {
1798 switch (tsk->state) {
1799 case TASK_RUNNING:
1800 stats->nr_running++;
1801 break;
1802 case TASK_INTERRUPTIBLE:
1803 stats->nr_sleeping++;
1804 break;
1805 case TASK_UNINTERRUPTIBLE:
1806 stats->nr_uninterruptible++;
1807 break;
1808 case TASK_STOPPED:
1809 stats->nr_stopped++;
1810 break;
1811 default:
1812 if (delayacct_is_task_waiting_on_io(tsk))
1813 stats->nr_io_wait++;
1814 break;
1815 }
1816 }
1817 cgroup_iter_end(cgrp, &it);
1818
1819 rcu_read_unlock();
1820err:
1821 return ret;
1822}
1823
1824static int cmppid(const void *a, const void *b)
1825{
1826 return *(pid_t *)a - *(pid_t *)b;
1827}
1828
1829/*
1830 * Convert array 'a' of 'npids' pid_t's to a string of newline separated
1831 * decimal pids in 'buf'. Don't write more than 'sz' chars, but return
1832 * count 'cnt' of how many chars would be written if buf were large enough.
1833 */
1834static int pid_array_to_buf(char *buf, int sz, pid_t *a, int npids)
1835{
1836 int cnt = 0;
1837 int i;
1838
1839 for (i = 0; i < npids; i++)
1840 cnt += snprintf(buf + cnt, max(sz - cnt, 0), "%d\n", a[i]);
1841 return cnt;
1842}
1843
1844/*
1845 * Handle an open on 'tasks' file. Prepare a buffer listing the
1846 * process id's of tasks currently attached to the cgroup being opened.
1847 *
1848 * Does not require any specific cgroup mutexes, and does not take any.
1849 */
1850static int cgroup_tasks_open(struct inode *unused, struct file *file)
1851{
1852 struct cgroup *cgrp = __d_cgrp(file->f_dentry->d_parent);
1853 struct ctr_struct *ctr;
1854 pid_t *pidarray;
1855 int npids;
1856 char c;
1857
1858 if (!(file->f_mode & FMODE_READ))
1859 return 0;
1860
1861 ctr = kmalloc(sizeof(*ctr), GFP_KERNEL);
1862 if (!ctr)
1863 goto err0;
1864
1865 /*
1866 * If cgroup gets more users after we read count, we won't have
1867 * enough space - tough. This race is indistinguishable to the
1868 * caller from the case that the additional cgroup users didn't
1869 * show up until sometime later on.
1870 */
1871 npids = cgroup_task_count(cgrp);
1872 if (npids) {
1873 pidarray = kmalloc(npids * sizeof(pid_t), GFP_KERNEL);
1874 if (!pidarray)
1875 goto err1;
1876
1877 npids = pid_array_load(pidarray, npids, cgrp);
1878 sort(pidarray, npids, sizeof(pid_t), cmppid, NULL);
1879
1880 /* Call pid_array_to_buf() twice, first just to get bufsz */
1881 ctr->bufsz = pid_array_to_buf(&c, sizeof(c), pidarray, npids) + 1;
1882 ctr->buf = kmalloc(ctr->bufsz, GFP_KERNEL);
1883 if (!ctr->buf)
1884 goto err2;
1885 ctr->bufsz = pid_array_to_buf(ctr->buf, ctr->bufsz, pidarray, npids);
1886
1887 kfree(pidarray);
1888 } else {
1889 ctr->buf = 0;
1890 ctr->bufsz = 0;
1891 }
1892 file->private_data = ctr;
1893 return 0;
1894
1895err2:
1896 kfree(pidarray);
1897err1:
1898 kfree(ctr);
1899err0:
1900 return -ENOMEM;
1901}
1902
1903static ssize_t cgroup_tasks_read(struct cgroup *cgrp,
1904 struct cftype *cft,
1905 struct file *file, char __user *buf,
1906 size_t nbytes, loff_t *ppos)
1907{
1908 struct ctr_struct *ctr = file->private_data;
1909
1910 return simple_read_from_buffer(buf, nbytes, ppos, ctr->buf, ctr->bufsz);
1911}
1912
1913static int cgroup_tasks_release(struct inode *unused_inode,
1914 struct file *file)
1915{
1916 struct ctr_struct *ctr;
1917
1918 if (file->f_mode & FMODE_READ) {
1919 ctr = file->private_data;
1920 kfree(ctr->buf);
1921 kfree(ctr);
1922 }
1923 return 0;
1924}
1925
1926static u64 cgroup_read_notify_on_release(struct cgroup *cgrp,
1927 struct cftype *cft)
1928{
1929 return notify_on_release(cgrp);
1930}
1931
1932static u64 cgroup_read_releasable(struct cgroup *cgrp, struct cftype *cft)
1933{
1934 return test_bit(CGRP_RELEASABLE, &cgrp->flags);
1935}
1936
1937/*
1938 * for the common functions, 'private' gives the type of file
1939 */
1940static struct cftype files[] = {
1941 {
1942 .name = "tasks",
1943 .open = cgroup_tasks_open,
1944 .read = cgroup_tasks_read,
1945 .write = cgroup_common_file_write,
1946 .release = cgroup_tasks_release,
1947 .private = FILE_TASKLIST,
1948 },
1949
1950 {
1951 .name = "notify_on_release",
1952 .read_uint = cgroup_read_notify_on_release,
1953 .write = cgroup_common_file_write,
1954 .private = FILE_NOTIFY_ON_RELEASE,
1955 },
1956
1957 {
1958 .name = "releasable",
1959 .read_uint = cgroup_read_releasable,
1960 .private = FILE_RELEASABLE,
1961 }
1962};
1963
1964static struct cftype cft_release_agent = {
1965 .name = "release_agent",
1966 .read = cgroup_common_file_read,
1967 .write = cgroup_common_file_write,
1968 .private = FILE_RELEASE_AGENT,
1969};
1970
1971static int cgroup_populate_dir(struct cgroup *cgrp)
1972{
1973 int err;
1974 struct cgroup_subsys *ss;
1975
1976 /* First clear out any existing files */
1977 cgroup_clear_directory(cgrp->dentry);
1978
1979 err = cgroup_add_files(cgrp, NULL, files, ARRAY_SIZE(files));
1980 if (err < 0)
1981 return err;
1982
1983 if (cgrp == cgrp->top_cgroup) {
1984 if ((err = cgroup_add_file(cgrp, NULL, &cft_release_agent)) < 0)
1985 return err;
1986 }
1987
1988 for_each_subsys(cgrp->root, ss) {
1989 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
1990 return err;
1991 }
1992
1993 return 0;
1994}
1995
1996static void init_cgroup_css(struct cgroup_subsys_state *css,
1997 struct cgroup_subsys *ss,
1998 struct cgroup *cgrp)
1999{
2000 css->cgroup = cgrp;
2001 atomic_set(&css->refcnt, 0);
2002 css->flags = 0;
2003 if (cgrp == dummytop)
2004 set_bit(CSS_ROOT, &css->flags);
2005 BUG_ON(cgrp->subsys[ss->subsys_id]);
2006 cgrp->subsys[ss->subsys_id] = css;
2007}
2008
2009/*
2010 * cgroup_create - create a cgroup
2011 * parent: cgroup that will be parent of the new cgroup.
2012 * name: name of the new cgroup. Will be strcpy'ed.
2013 * mode: mode to set on new inode
2014 *
2015 * Must be called with the mutex on the parent inode held
2016 */
2017
2018static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
2019 int mode)
2020{
2021 struct cgroup *cgrp;
2022 struct cgroupfs_root *root = parent->root;
2023 int err = 0;
2024 struct cgroup_subsys *ss;
2025 struct super_block *sb = root->sb;
2026
2027 cgrp = kzalloc(sizeof(*cgrp), GFP_KERNEL);
2028 if (!cgrp)
2029 return -ENOMEM;
2030
2031 /* Grab a reference on the superblock so the hierarchy doesn't
2032 * get deleted on unmount if there are child cgroups. This
2033 * can be done outside cgroup_mutex, since the sb can't
2034 * disappear while someone has an open control file on the
2035 * fs */
2036 atomic_inc(&sb->s_active);
2037
2038 mutex_lock(&cgroup_mutex);
2039
2040 cgrp->flags = 0;
2041 INIT_LIST_HEAD(&cgrp->sibling);
2042 INIT_LIST_HEAD(&cgrp->children);
2043 INIT_LIST_HEAD(&cgrp->css_sets);
2044 INIT_LIST_HEAD(&cgrp->release_list);
2045
2046 cgrp->parent = parent;
2047 cgrp->root = parent->root;
2048 cgrp->top_cgroup = parent->top_cgroup;
2049
2050 for_each_subsys(root, ss) {
2051 struct cgroup_subsys_state *css = ss->create(ss, cgrp);
2052 if (IS_ERR(css)) {
2053 err = PTR_ERR(css);
2054 goto err_destroy;
2055 }
2056 init_cgroup_css(css, ss, cgrp);
2057 }
2058
2059 list_add(&cgrp->sibling, &cgrp->parent->children);
2060 root->number_of_cgroups++;
2061
2062 err = cgroup_create_dir(cgrp, dentry, mode);
2063 if (err < 0)
2064 goto err_remove;
2065
2066 /* The cgroup directory was pre-locked for us */
2067 BUG_ON(!mutex_is_locked(&cgrp->dentry->d_inode->i_mutex));
2068
2069 err = cgroup_populate_dir(cgrp);
2070 /* If err < 0, we have a half-filled directory - oh well ;) */
2071
2072 mutex_unlock(&cgroup_mutex);
2073 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
2074
2075 return 0;
2076
2077 err_remove:
2078
2079 list_del(&cgrp->sibling);
2080 root->number_of_cgroups--;
2081
2082 err_destroy:
2083
2084 for_each_subsys(root, ss) {
2085 if (cgrp->subsys[ss->subsys_id])
2086 ss->destroy(ss, cgrp);
2087 }
2088
2089 mutex_unlock(&cgroup_mutex);
2090
2091 /* Release the reference count that we took on the superblock */
2092 deactivate_super(sb);
2093
2094 kfree(cgrp);
2095 return err;
2096}
2097
2098static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2099{
2100 struct cgroup *c_parent = dentry->d_parent->d_fsdata;
2101
2102 /* the vfs holds inode->i_mutex already */
2103 return cgroup_create(c_parent, dentry, mode | S_IFDIR);
2104}
2105
2106static inline int cgroup_has_css_refs(struct cgroup *cgrp)
2107{
2108 /* Check the reference count on each subsystem. Since we
2109 * already established that there are no tasks in the
2110 * cgroup, if the css refcount is also 0, then there should
2111 * be no outstanding references, so the subsystem is safe to
2112 * destroy. We scan across all subsystems rather than using
2113 * the per-hierarchy linked list of mounted subsystems since
2114 * we can be called via check_for_release() with no
2115 * synchronization other than RCU, and the subsystem linked
2116 * list isn't RCU-safe */
2117 int i;
2118 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2119 struct cgroup_subsys *ss = subsys[i];
2120 struct cgroup_subsys_state *css;
2121 /* Skip subsystems not in this hierarchy */
2122 if (ss->root != cgrp->root)
2123 continue;
2124 css = cgrp->subsys[ss->subsys_id];
2125 /* When called from check_for_release() it's possible
2126 * that by this point the cgroup has been removed
2127 * and the css deleted. But a false-positive doesn't
2128 * matter, since it can only happen if the cgroup
2129 * has been deleted and hence no longer needs the
2130 * release agent to be called anyway. */
2131 if (css && atomic_read(&css->refcnt)) {
2132 return 1;
2133 }
2134 }
2135 return 0;
2136}
2137
2138static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
2139{
2140 struct cgroup *cgrp = dentry->d_fsdata;
2141 struct dentry *d;
2142 struct cgroup *parent;
2143 struct cgroup_subsys *ss;
2144 struct super_block *sb;
2145 struct cgroupfs_root *root;
2146
2147 /* the vfs holds both inode->i_mutex already */
2148
2149 mutex_lock(&cgroup_mutex);
2150 if (atomic_read(&cgrp->count) != 0) {
2151 mutex_unlock(&cgroup_mutex);
2152 return -EBUSY;
2153 }
2154 if (!list_empty(&cgrp->children)) {
2155 mutex_unlock(&cgroup_mutex);
2156 return -EBUSY;
2157 }
2158
2159 parent = cgrp->parent;
2160 root = cgrp->root;
2161 sb = root->sb;
2162
2163 if (cgroup_has_css_refs(cgrp)) {
2164 mutex_unlock(&cgroup_mutex);
2165 return -EBUSY;
2166 }
2167
2168 for_each_subsys(root, ss) {
2169 if (cgrp->subsys[ss->subsys_id])
2170 ss->destroy(ss, cgrp);
2171 }
2172
2173 spin_lock(&release_list_lock);
2174 set_bit(CGRP_REMOVED, &cgrp->flags);
2175 if (!list_empty(&cgrp->release_list))
2176 list_del(&cgrp->release_list);
2177 spin_unlock(&release_list_lock);
2178 /* delete my sibling from parent->children */
2179 list_del(&cgrp->sibling);
2180 spin_lock(&cgrp->dentry->d_lock);
2181 d = dget(cgrp->dentry);
2182 cgrp->dentry = NULL;
2183 spin_unlock(&d->d_lock);
2184
2185 cgroup_d_remove_dir(d);
2186 dput(d);
2187 root->number_of_cgroups--;
2188
2189 set_bit(CGRP_RELEASABLE, &parent->flags);
2190 check_for_release(parent);
2191
2192 mutex_unlock(&cgroup_mutex);
2193 /* Drop the active superblock reference that we took when we
2194 * created the cgroup */
2195 deactivate_super(sb);
2196 return 0;
2197}
2198
2199static void cgroup_init_subsys(struct cgroup_subsys *ss)
2200{
2201 struct cgroup_subsys_state *css;
2202 struct list_head *l;
2203 printk(KERN_ERR "Initializing cgroup subsys %s\n", ss->name);
2204
2205 /* Create the top cgroup state for this subsystem */
2206 ss->root = &rootnode;
2207 css = ss->create(ss, dummytop);
2208 /* We don't handle early failures gracefully */
2209 BUG_ON(IS_ERR(css));
2210 init_cgroup_css(css, ss, dummytop);
2211
2212 /* Update all cgroup groups to contain a subsys
2213 * pointer to this state - since the subsystem is
2214 * newly registered, all tasks and hence all cgroup
2215 * groups are in the subsystem's top cgroup. */
2216 write_lock(&css_set_lock);
2217 l = &init_css_set.list;
2218 do {
2219 struct css_set *cg =
2220 list_entry(l, struct css_set, list);
2221 cg->subsys[ss->subsys_id] = dummytop->subsys[ss->subsys_id];
2222 l = l->next;
2223 } while (l != &init_css_set.list);
2224 write_unlock(&css_set_lock);
2225
2226 /* If this subsystem requested that it be notified with fork
2227 * events, we should send it one now for every process in the
2228 * system */
2229 if (ss->fork) {
2230 struct task_struct *g, *p;
2231
2232 read_lock(&tasklist_lock);
2233 do_each_thread(g, p) {
2234 ss->fork(ss, p);
2235 } while_each_thread(g, p);
2236 read_unlock(&tasklist_lock);
2237 }
2238
2239 need_forkexit_callback |= ss->fork || ss->exit;
2240
2241 ss->active = 1;
2242}
2243
2244/**
2245 * cgroup_init_early - initialize cgroups at system boot, and
2246 * initialize any subsystems that request early init.
2247 */
2248int __init cgroup_init_early(void)
2249{
2250 int i;
2251 kref_init(&init_css_set.ref);
2252 kref_get(&init_css_set.ref);
2253 INIT_LIST_HEAD(&init_css_set.list);
2254 INIT_LIST_HEAD(&init_css_set.cg_links);
2255 INIT_LIST_HEAD(&init_css_set.tasks);
2256 css_set_count = 1;
2257 init_cgroup_root(&rootnode);
2258 list_add(&rootnode.root_list, &roots);
2259 root_count = 1;
2260 init_task.cgroups = &init_css_set;
2261
2262 init_css_set_link.cg = &init_css_set;
2263 list_add(&init_css_set_link.cgrp_link_list,
2264 &rootnode.top_cgroup.css_sets);
2265 list_add(&init_css_set_link.cg_link_list,
2266 &init_css_set.cg_links);
2267
2268 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2269 struct cgroup_subsys *ss = subsys[i];
2270
2271 BUG_ON(!ss->name);
2272 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
2273 BUG_ON(!ss->create);
2274 BUG_ON(!ss->destroy);
2275 if (ss->subsys_id != i) {
2276 printk(KERN_ERR "Subsys %s id == %d\n",
2277 ss->name, ss->subsys_id);
2278 BUG();
2279 }
2280
2281 if (ss->early_init)
2282 cgroup_init_subsys(ss);
2283 }
2284 return 0;
2285}
2286
2287/**
2288 * cgroup_init - register cgroup filesystem and /proc file, and
2289 * initialize any subsystems that didn't request early init.
2290 */
2291int __init cgroup_init(void)
2292{
2293 int err;
2294 int i;
2295 struct proc_dir_entry *entry;
2296
2297 err = bdi_init(&cgroup_backing_dev_info);
2298 if (err)
2299 return err;
2300
2301 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2302 struct cgroup_subsys *ss = subsys[i];
2303 if (!ss->early_init)
2304 cgroup_init_subsys(ss);
2305 }
2306
2307 err = register_filesystem(&cgroup_fs_type);
2308 if (err < 0)
2309 goto out;
2310
2311 entry = create_proc_entry("cgroups", 0, NULL);
2312 if (entry)
2313 entry->proc_fops = &proc_cgroupstats_operations;
2314
2315out:
2316 if (err)
2317 bdi_destroy(&cgroup_backing_dev_info);
2318
2319 return err;
2320}
2321
2322/*
2323 * proc_cgroup_show()
2324 * - Print task's cgroup paths into seq_file, one line for each hierarchy
2325 * - Used for /proc/<pid>/cgroup.
2326 * - No need to task_lock(tsk) on this tsk->cgroup reference, as it
2327 * doesn't really matter if tsk->cgroup changes after we read it,
2328 * and we take cgroup_mutex, keeping attach_task() from changing it
2329 * anyway. No need to check that tsk->cgroup != NULL, thanks to
2330 * the_top_cgroup_hack in cgroup_exit(), which sets an exiting tasks
2331 * cgroup to top_cgroup.
2332 */
2333
2334/* TODO: Use a proper seq_file iterator */
2335static int proc_cgroup_show(struct seq_file *m, void *v)
2336{
2337 struct pid *pid;
2338 struct task_struct *tsk;
2339 char *buf;
2340 int retval;
2341 struct cgroupfs_root *root;
2342
2343 retval = -ENOMEM;
2344 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2345 if (!buf)
2346 goto out;
2347
2348 retval = -ESRCH;
2349 pid = m->private;
2350 tsk = get_pid_task(pid, PIDTYPE_PID);
2351 if (!tsk)
2352 goto out_free;
2353
2354 retval = 0;
2355
2356 mutex_lock(&cgroup_mutex);
2357
2358 for_each_root(root) {
2359 struct cgroup_subsys *ss;
2360 struct cgroup *cgrp;
2361 int subsys_id;
2362 int count = 0;
2363
2364 /* Skip this hierarchy if it has no active subsystems */
2365 if (!root->actual_subsys_bits)
2366 continue;
2367 for_each_subsys(root, ss)
2368 seq_printf(m, "%s%s", count++ ? "," : "", ss->name);
2369 seq_putc(m, ':');
2370 get_first_subsys(&root->top_cgroup, NULL, &subsys_id);
2371 cgrp = task_cgroup(tsk, subsys_id);
2372 retval = cgroup_path(cgrp, buf, PAGE_SIZE);
2373 if (retval < 0)
2374 goto out_unlock;
2375 seq_puts(m, buf);
2376 seq_putc(m, '\n');
2377 }
2378
2379out_unlock:
2380 mutex_unlock(&cgroup_mutex);
2381 put_task_struct(tsk);
2382out_free:
2383 kfree(buf);
2384out:
2385 return retval;
2386}
2387
2388static int cgroup_open(struct inode *inode, struct file *file)
2389{
2390 struct pid *pid = PROC_I(inode)->pid;
2391 return single_open(file, proc_cgroup_show, pid);
2392}
2393
2394struct file_operations proc_cgroup_operations = {
2395 .open = cgroup_open,
2396 .read = seq_read,
2397 .llseek = seq_lseek,
2398 .release = single_release,
2399};
2400
2401/* Display information about each subsystem and each hierarchy */
2402static int proc_cgroupstats_show(struct seq_file *m, void *v)
2403{
2404 int i;
2405 struct cgroupfs_root *root;
2406
2407 seq_puts(m, "#subsys_name\thierarchy\tnum_cgroups\n");
2408 mutex_lock(&cgroup_mutex);
2409 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2410 struct cgroup_subsys *ss = subsys[i];
2411 seq_printf(m, "%s\t%lu\t%d\n",
2412 ss->name, ss->root->subsys_bits,
2413 ss->root->number_of_cgroups);
2414 }
2415 mutex_unlock(&cgroup_mutex);
2416 return 0;
2417}
2418
2419static int cgroupstats_open(struct inode *inode, struct file *file)
2420{
2421 return single_open(file, proc_cgroupstats_show, 0);
2422}
2423
2424static struct file_operations proc_cgroupstats_operations = {
2425 .open = cgroupstats_open,
2426 .read = seq_read,
2427 .llseek = seq_lseek,
2428 .release = single_release,
2429};
2430
2431/**
2432 * cgroup_fork - attach newly forked task to its parents cgroup.
2433 * @tsk: pointer to task_struct of forking parent process.
2434 *
2435 * Description: A task inherits its parent's cgroup at fork().
2436 *
2437 * A pointer to the shared css_set was automatically copied in
2438 * fork.c by dup_task_struct(). However, we ignore that copy, since
2439 * it was not made under the protection of RCU or cgroup_mutex, so
2440 * might no longer be a valid cgroup pointer. attach_task() might
2441 * have already changed current->cgroups, allowing the previously
2442 * referenced cgroup group to be removed and freed.
2443 *
2444 * At the point that cgroup_fork() is called, 'current' is the parent
2445 * task, and the passed argument 'child' points to the child task.
2446 */
2447void cgroup_fork(struct task_struct *child)
2448{
2449 task_lock(current);
2450 child->cgroups = current->cgroups;
2451 get_css_set(child->cgroups);
2452 task_unlock(current);
2453 INIT_LIST_HEAD(&child->cg_list);
2454}
2455
2456/**
2457 * cgroup_fork_callbacks - called on a new task very soon before
2458 * adding it to the tasklist. No need to take any locks since no-one
2459 * can be operating on this task
2460 */
2461void cgroup_fork_callbacks(struct task_struct *child)
2462{
2463 if (need_forkexit_callback) {
2464 int i;
2465 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2466 struct cgroup_subsys *ss = subsys[i];
2467 if (ss->fork)
2468 ss->fork(ss, child);
2469 }
2470 }
2471}
2472
2473/**
2474 * cgroup_post_fork - called on a new task after adding it to the
2475 * task list. Adds the task to the list running through its css_set
2476 * if necessary. Has to be after the task is visible on the task list
2477 * in case we race with the first call to cgroup_iter_start() - to
2478 * guarantee that the new task ends up on its list. */
2479void cgroup_post_fork(struct task_struct *child)
2480{
2481 if (use_task_css_set_links) {
2482 write_lock(&css_set_lock);
2483 if (list_empty(&child->cg_list))
2484 list_add(&child->cg_list, &child->cgroups->tasks);
2485 write_unlock(&css_set_lock);
2486 }
2487}
2488/**
2489 * cgroup_exit - detach cgroup from exiting task
2490 * @tsk: pointer to task_struct of exiting process
2491 *
2492 * Description: Detach cgroup from @tsk and release it.
2493 *
2494 * Note that cgroups marked notify_on_release force every task in
2495 * them to take the global cgroup_mutex mutex when exiting.
2496 * This could impact scaling on very large systems. Be reluctant to
2497 * use notify_on_release cgroups where very high task exit scaling
2498 * is required on large systems.
2499 *
2500 * the_top_cgroup_hack:
2501 *
2502 * Set the exiting tasks cgroup to the root cgroup (top_cgroup).
2503 *
2504 * We call cgroup_exit() while the task is still competent to
2505 * handle notify_on_release(), then leave the task attached to the
2506 * root cgroup in each hierarchy for the remainder of its exit.
2507 *
2508 * To do this properly, we would increment the reference count on
2509 * top_cgroup, and near the very end of the kernel/exit.c do_exit()
2510 * code we would add a second cgroup function call, to drop that
2511 * reference. This would just create an unnecessary hot spot on
2512 * the top_cgroup reference count, to no avail.
2513 *
2514 * Normally, holding a reference to a cgroup without bumping its
2515 * count is unsafe. The cgroup could go away, or someone could
2516 * attach us to a different cgroup, decrementing the count on
2517 * the first cgroup that we never incremented. But in this case,
2518 * top_cgroup isn't going away, and either task has PF_EXITING set,
2519 * which wards off any attach_task() attempts, or task is a failed
2520 * fork, never visible to attach_task.
2521 *
2522 */
2523void cgroup_exit(struct task_struct *tsk, int run_callbacks)
2524{
2525 int i;
2526 struct css_set *cg;
2527
2528 if (run_callbacks && need_forkexit_callback) {
2529 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
2530 struct cgroup_subsys *ss = subsys[i];
2531 if (ss->exit)
2532 ss->exit(ss, tsk);
2533 }
2534 }
2535
2536 /*
2537 * Unlink from the css_set task list if necessary.
2538 * Optimistically check cg_list before taking
2539 * css_set_lock
2540 */
2541 if (!list_empty(&tsk->cg_list)) {
2542 write_lock(&css_set_lock);
2543 if (!list_empty(&tsk->cg_list))
2544 list_del(&tsk->cg_list);
2545 write_unlock(&css_set_lock);
2546 }
2547
2548 /* Reassign the task to the init_css_set. */
2549 task_lock(tsk);
2550 cg = tsk->cgroups;
2551 tsk->cgroups = &init_css_set;
2552 task_unlock(tsk);
2553 if (cg)
2554 put_css_set_taskexit(cg);
2555}
2556
2557/**
2558 * cgroup_clone - duplicate the current cgroup in the hierarchy
2559 * that the given subsystem is attached to, and move this task into
2560 * the new child
2561 */
2562int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys)
2563{
2564 struct dentry *dentry;
2565 int ret = 0;
2566 char nodename[MAX_CGROUP_TYPE_NAMELEN];
2567 struct cgroup *parent, *child;
2568 struct inode *inode;
2569 struct css_set *cg;
2570 struct cgroupfs_root *root;
2571 struct cgroup_subsys *ss;
2572
2573 /* We shouldn't be called by an unregistered subsystem */
2574 BUG_ON(!subsys->active);
2575
2576 /* First figure out what hierarchy and cgroup we're dealing
2577 * with, and pin them so we can drop cgroup_mutex */
2578 mutex_lock(&cgroup_mutex);
2579 again:
2580 root = subsys->root;
2581 if (root == &rootnode) {
2582 printk(KERN_INFO
2583 "Not cloning cgroup for unused subsystem %s\n",
2584 subsys->name);
2585 mutex_unlock(&cgroup_mutex);
2586 return 0;
2587 }
2588 cg = tsk->cgroups;
2589 parent = task_cgroup(tsk, subsys->subsys_id);
2590
2591 snprintf(nodename, MAX_CGROUP_TYPE_NAMELEN, "node_%d", tsk->pid);
2592
2593 /* Pin the hierarchy */
2594 atomic_inc(&parent->root->sb->s_active);
2595
2596 /* Keep the cgroup alive */
2597 get_css_set(cg);
2598 mutex_unlock(&cgroup_mutex);
2599
2600 /* Now do the VFS work to create a cgroup */
2601 inode = parent->dentry->d_inode;
2602
2603 /* Hold the parent directory mutex across this operation to
2604 * stop anyone else deleting the new cgroup */
2605 mutex_lock(&inode->i_mutex);
2606 dentry = lookup_one_len(nodename, parent->dentry, strlen(nodename));
2607 if (IS_ERR(dentry)) {
2608 printk(KERN_INFO
2609 "Couldn't allocate dentry for %s: %ld\n", nodename,
2610 PTR_ERR(dentry));
2611 ret = PTR_ERR(dentry);
2612 goto out_release;
2613 }
2614
2615 /* Create the cgroup directory, which also creates the cgroup */
2616 ret = vfs_mkdir(inode, dentry, S_IFDIR | 0755);
2617 child = __d_cgrp(dentry);
2618 dput(dentry);
2619 if (ret) {
2620 printk(KERN_INFO
2621 "Failed to create cgroup %s: %d\n", nodename,
2622 ret);
2623 goto out_release;
2624 }
2625
2626 if (!child) {
2627 printk(KERN_INFO
2628 "Couldn't find new cgroup %s\n", nodename);
2629 ret = -ENOMEM;
2630 goto out_release;
2631 }
2632
2633 /* The cgroup now exists. Retake cgroup_mutex and check
2634 * that we're still in the same state that we thought we
2635 * were. */
2636 mutex_lock(&cgroup_mutex);
2637 if ((root != subsys->root) ||
2638 (parent != task_cgroup(tsk, subsys->subsys_id))) {
2639 /* Aargh, we raced ... */
2640 mutex_unlock(&inode->i_mutex);
2641 put_css_set(cg);
2642
2643 deactivate_super(parent->root->sb);
2644 /* The cgroup is still accessible in the VFS, but
2645 * we're not going to try to rmdir() it at this
2646 * point. */
2647 printk(KERN_INFO
2648 "Race in cgroup_clone() - leaking cgroup %s\n",
2649 nodename);
2650 goto again;
2651 }
2652
2653 /* do any required auto-setup */
2654 for_each_subsys(root, ss) {
2655 if (ss->post_clone)
2656 ss->post_clone(ss, child);
2657 }
2658
2659 /* All seems fine. Finish by moving the task into the new cgroup */
2660 ret = attach_task(child, tsk);
2661 mutex_unlock(&cgroup_mutex);
2662
2663 out_release:
2664 mutex_unlock(&inode->i_mutex);
2665
2666 mutex_lock(&cgroup_mutex);
2667 put_css_set(cg);
2668 mutex_unlock(&cgroup_mutex);
2669 deactivate_super(parent->root->sb);
2670 return ret;
2671}
2672
2673/*
2674 * See if "cgrp" is a descendant of the current task's cgroup in
2675 * the appropriate hierarchy
2676 *
2677 * If we are sending in dummytop, then presumably we are creating
2678 * the top cgroup in the subsystem.
2679 *
2680 * Called only by the ns (nsproxy) cgroup.
2681 */
2682int cgroup_is_descendant(const struct cgroup *cgrp)
2683{
2684 int ret;
2685 struct cgroup *target;
2686 int subsys_id;
2687
2688 if (cgrp == dummytop)
2689 return 1;
2690
2691 get_first_subsys(cgrp, NULL, &subsys_id);
2692 target = task_cgroup(current, subsys_id);
2693 while (cgrp != target && cgrp!= cgrp->top_cgroup)
2694 cgrp = cgrp->parent;
2695 ret = (cgrp == target);
2696 return ret;
2697}
2698
2699static void check_for_release(struct cgroup *cgrp)
2700{
2701 /* All of these checks rely on RCU to keep the cgroup
2702 * structure alive */
2703 if (cgroup_is_releasable(cgrp) && !atomic_read(&cgrp->count)
2704 && list_empty(&cgrp->children) && !cgroup_has_css_refs(cgrp)) {
2705 /* Control Group is currently removeable. If it's not
2706 * already queued for a userspace notification, queue
2707 * it now */
2708 int need_schedule_work = 0;
2709 spin_lock(&release_list_lock);
2710 if (!cgroup_is_removed(cgrp) &&
2711 list_empty(&cgrp->release_list)) {
2712 list_add(&cgrp->release_list, &release_list);
2713 need_schedule_work = 1;
2714 }
2715 spin_unlock(&release_list_lock);
2716 if (need_schedule_work)
2717 schedule_work(&release_agent_work);
2718 }
2719}
2720
2721void __css_put(struct cgroup_subsys_state *css)
2722{
2723 struct cgroup *cgrp = css->cgroup;
2724 rcu_read_lock();
2725 if (atomic_dec_and_test(&css->refcnt) && notify_on_release(cgrp)) {
2726 set_bit(CGRP_RELEASABLE, &cgrp->flags);
2727 check_for_release(cgrp);
2728 }
2729 rcu_read_unlock();
2730}
2731
2732/*
2733 * Notify userspace when a cgroup is released, by running the
2734 * configured release agent with the name of the cgroup (path
2735 * relative to the root of cgroup file system) as the argument.
2736 *
2737 * Most likely, this user command will try to rmdir this cgroup.
2738 *
2739 * This races with the possibility that some other task will be
2740 * attached to this cgroup before it is removed, or that some other
2741 * user task will 'mkdir' a child cgroup of this cgroup. That's ok.
2742 * The presumed 'rmdir' will fail quietly if this cgroup is no longer
2743 * unused, and this cgroup will be reprieved from its death sentence,
2744 * to continue to serve a useful existence. Next time it's released,
2745 * we will get notified again, if it still has 'notify_on_release' set.
2746 *
2747 * The final arg to call_usermodehelper() is UMH_WAIT_EXEC, which
2748 * means only wait until the task is successfully execve()'d. The
2749 * separate release agent task is forked by call_usermodehelper(),
2750 * then control in this thread returns here, without waiting for the
2751 * release agent task. We don't bother to wait because the caller of
2752 * this routine has no use for the exit status of the release agent
2753 * task, so no sense holding our caller up for that.
2754 *
2755 */
2756
2757static void cgroup_release_agent(struct work_struct *work)
2758{
2759 BUG_ON(work != &release_agent_work);
2760 mutex_lock(&cgroup_mutex);
2761 spin_lock(&release_list_lock);
2762 while (!list_empty(&release_list)) {
2763 char *argv[3], *envp[3];
2764 int i;
2765 char *pathbuf;
2766 struct cgroup *cgrp = list_entry(release_list.next,
2767 struct cgroup,
2768 release_list);
2769 list_del_init(&cgrp->release_list);
2770 spin_unlock(&release_list_lock);
2771 pathbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2772 if (!pathbuf) {
2773 spin_lock(&release_list_lock);
2774 continue;
2775 }
2776
2777 if (cgroup_path(cgrp, pathbuf, PAGE_SIZE) < 0) {
2778 kfree(pathbuf);
2779 spin_lock(&release_list_lock);
2780 continue;
2781 }
2782
2783 i = 0;
2784 argv[i++] = cgrp->root->release_agent_path;
2785 argv[i++] = (char *)pathbuf;
2786 argv[i] = NULL;
2787
2788 i = 0;
2789 /* minimal command environment */
2790 envp[i++] = "HOME=/";
2791 envp[i++] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
2792 envp[i] = NULL;
2793
2794 /* Drop the lock while we invoke the usermode helper,
2795 * since the exec could involve hitting disk and hence
2796 * be a slow process */
2797 mutex_unlock(&cgroup_mutex);
2798 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
2799 kfree(pathbuf);
2800 mutex_lock(&cgroup_mutex);
2801 spin_lock(&release_list_lock);
2802 }
2803 spin_unlock(&release_list_lock);
2804 mutex_unlock(&cgroup_mutex);
2805}
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-11 02:12:24 -0500